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 #define CMA_PREFERRED_ROCE_GID_TYPE IB_GID_TYPE_ROCE_UDP_ENCAP
  76
  77 static const char * const cma_events[] = {
  78         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  79         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  80         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  81         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  82         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  83         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  84         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  85         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  86         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  87         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  88         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  89         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  90         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  91         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  92         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  93         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  94 };
  95
  96 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  97 {
  98         size_t index = event;
  99
 100         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 101                         cma_events[index] : "unrecognized event";
 102 }
 103 EXPORT_SYMBOL(rdma_event_msg);
 104
 105 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 106                                                 int reason)
 107 {
 108         if (rdma_ib_or_roce(id->device, id->port_num))
 109                 return ibcm_reject_msg(reason);
 110
 111         if (rdma_protocol_iwarp(id->device, id->port_num))
 112                 return iwcm_reject_msg(reason);
 113
 114         WARN_ON_ONCE(1);
 115         return "unrecognized transport";
 116 }
 117 EXPORT_SYMBOL(rdma_reject_msg);
 118
 119 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 120 {
 121         if (rdma_ib_or_roce(id->device, id->port_num))
 122                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 123
 124         if (rdma_protocol_iwarp(id->device, id->port_num))
 125                 return reason == -ECONNREFUSED;
 126
 127         WARN_ON_ONCE(1);
 128         return false;
 129 }
 130 EXPORT_SYMBOL(rdma_is_consumer_reject);
 131
 132 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 133                                       struct rdma_cm_event *ev, u8 *data_len)
 134 {
 135         const void *p;
 136
 137         if (rdma_is_consumer_reject(id, ev->status)) {
 138                 *data_len = ev->param.conn.private_data_len;
 139                 p = ev->param.conn.private_data;
 140         } else {
 141                 *data_len = 0;
 142                 p = NULL;
 143         }
 144         return p;
 145 }
 146 EXPORT_SYMBOL(rdma_consumer_reject_data);
 147
 148 static void cma_add_one(struct ib_device *device);
 149 static void cma_remove_one(struct ib_device *device, void *client_data);
 150
 151 static struct ib_client cma_client = {
 152         .name   = "cma",
 153         .add    = cma_add_one,
 154         .remove = cma_remove_one
 155 };
 156
 157 static struct ib_sa_client sa_client;
 158 static struct rdma_addr_client addr_client;
 159 static LIST_HEAD(dev_list);
 160 static LIST_HEAD(listen_any_list);
 161 static DEFINE_MUTEX(lock);
 162 static struct workqueue_struct *cma_wq;
 163 static unsigned int cma_pernet_id;
 164
 165 struct cma_pernet {
 166         struct idr tcp_ps;
 167         struct idr udp_ps;
 168         struct idr ipoib_ps;
 169         struct idr ib_ps;
 170 };
 171
 172 static struct cma_pernet *cma_pernet(struct net *net)
 173 {
 174         return net_generic(net, cma_pernet_id);
 175 }
 176
 177 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 178 {
 179         struct cma_pernet *pernet = cma_pernet(net);
 180
 181         switch (ps) {
 182         case RDMA_PS_TCP:
 183                 return &pernet->tcp_ps;
 184         case RDMA_PS_UDP:
 185                 return &pernet->udp_ps;
 186         case RDMA_PS_IPOIB:
 187                 return &pernet->ipoib_ps;
 188         case RDMA_PS_IB:
 189                 return &pernet->ib_ps;
 190         default:
 191                 return NULL;
 192         }
 193 }
 194
 195 struct cma_device {
 196         struct list_head        list;
 197         struct ib_device        *device;
 198         struct completion       comp;
 199         atomic_t                refcount;
 200         struct list_head        id_list;
 201         enum ib_gid_type        *default_gid_type;
 202         u8                      *default_roce_tos;
 203 };
 204
 205 struct rdma_bind_list {
 206         enum rdma_port_space    ps;
 207         struct hlist_head       owners;
 208         unsigned short          port;
 209 };
 210
 211 struct class_port_info_context {
 212         struct ib_class_port_info       *class_port_info;
 213         struct ib_device                *device;
 214         struct completion               done;
 215         struct ib_sa_query              *sa_query;
 216         u8                              port_num;
 217 };
 218
 219 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 220                         struct rdma_bind_list *bind_list, int snum)
 221 {
 222         struct idr *idr = cma_pernet_idr(net, ps);
 223
 224         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 225 }
 226
 227 static struct rdma_bind_list *cma_ps_find(struct net *net,
 228                                           enum rdma_port_space ps, int snum)
 229 {
 230         struct idr *idr = cma_pernet_idr(net, ps);
 231
 232         return idr_find(idr, snum);
 233 }
 234
 235 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 236 {
 237         struct idr *idr = cma_pernet_idr(net, ps);
 238
 239         idr_remove(idr, snum);
 240 }
 241
 242 enum {
 243         CMA_OPTION_AFONLY,
 244 };
 245
 246 void cma_ref_dev(struct cma_device *cma_dev)
 247 {
 248         atomic_inc(&cma_dev->refcount);
 249 }
 250
 251 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 252                                              void               *cookie)
 253 {
 254         struct cma_device *cma_dev;
 255         struct cma_device *found_cma_dev = NULL;
 256
 257         mutex_lock(&lock);
 258
 259         list_for_each_entry(cma_dev, &dev_list, list)
 260                 if (filter(cma_dev->device, cookie)) {
 261                         found_cma_dev = cma_dev;
 262                         break;
 263                 }
 264
 265         if (found_cma_dev)
 266                 cma_ref_dev(found_cma_dev);
 267         mutex_unlock(&lock);
 268         return found_cma_dev;
 269 }
 270
 271 int cma_get_default_gid_type(struct cma_device *cma_dev,
 272                              unsigned int port)
 273 {
 274         if (!rdma_is_port_valid(cma_dev->device, port))
 275                 return -EINVAL;
 276
 277         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 278 }
 279
 280 int cma_set_default_gid_type(struct cma_device *cma_dev,
 281                              unsigned int port,
 282                              enum ib_gid_type default_gid_type)
 283 {
 284         unsigned long supported_gids;
 285
 286         if (!rdma_is_port_valid(cma_dev->device, port))
 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 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 301 {
 302         if (!rdma_is_port_valid(cma_dev->device, port))
 303                 return -EINVAL;
 304
 305         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 306 }
 307
 308 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 309                              u8 default_roce_tos)
 310 {
 311         if (!rdma_is_port_valid(cma_dev->device, port))
 312                 return -EINVAL;
 313
 314         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 315                  default_roce_tos;
 316
 317         return 0;
 318 }
 319 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 320 {
 321         return cma_dev->device;
 322 }
 323
 324 /*
 325  * Device removal can occur at anytime, so we need extra handling to
 326  * serialize notifying the user of device removal with other callbacks.
 327  * We do this by disabling removal notification while a callback is in process,
 328  * and reporting it after the callback completes.
 329  */
 330 struct rdma_id_private {
 331         struct rdma_cm_id       id;
 332
 333         struct rdma_bind_list   *bind_list;
 334         struct hlist_node       node;
 335         struct list_head        list; /* listen_any_list or cma_device.list */
 336         struct list_head        listen_list; /* per device listens */
 337         struct cma_device       *cma_dev;
 338         struct list_head        mc_list;
 339
 340         int                     internal_id;
 341         enum rdma_cm_state      state;
 342         spinlock_t              lock;
 343         struct mutex            qp_mutex;
 344
 345         struct completion       comp;
 346         atomic_t                refcount;
 347         struct mutex            handler_mutex;
 348
 349         int                     backlog;
 350         int                     timeout_ms;
 351         struct ib_sa_query      *query;
 352         int                     query_id;
 353         union {
 354                 struct ib_cm_id *ib;
 355                 struct iw_cm_id *iw;
 356         } cm_id;
 357
 358         u32                     seq_num;
 359         u32                     qkey;
 360         u32                     qp_num;
 361         pid_t                   owner;
 362         u32                     options;
 363         u8                      srq;
 364         u8                      tos;
 365         bool                    tos_set;
 366         u8                      reuseaddr;
 367         u8                      afonly;
 368         enum ib_gid_type        gid_type;
 369 };
 370
 371 struct cma_multicast {
 372         struct rdma_id_private *id_priv;
 373         union {
 374                 struct ib_sa_multicast *ib;
 375         } multicast;
 376         struct list_head        list;
 377         void                    *context;
 378         struct sockaddr_storage addr;
 379         struct kref             mcref;
 380         bool                    igmp_joined;
 381         u8                      join_state;
 382 };
 383
 384 struct cma_work {
 385         struct work_struct      work;
 386         struct rdma_id_private  *id;
 387         enum rdma_cm_state      old_state;
 388         enum rdma_cm_state      new_state;
 389         struct rdma_cm_event    event;
 390 };
 391
 392 struct cma_ndev_work {
 393         struct work_struct      work;
 394         struct rdma_id_private  *id;
 395         struct rdma_cm_event    event;
 396 };
 397
 398 struct iboe_mcast_work {
 399         struct work_struct       work;
 400         struct rdma_id_private  *id;
 401         struct cma_multicast    *mc;
 402 };
 403
 404 union cma_ip_addr {
 405         struct in6_addr ip6;
 406         struct {
 407                 __be32 pad[3];
 408                 __be32 addr;
 409         } ip4;
 410 };
 411
 412 struct cma_hdr {
 413         u8 cma_version;
 414         u8 ip_version;  /* IP version: 7:4 */
 415         __be16 port;
 416         union cma_ip_addr src_addr;
 417         union cma_ip_addr dst_addr;
 418 };
 419
 420 #define CMA_VERSION 0x00
 421
 422 struct cma_req_info {
 423         struct ib_device *device;
 424         int port;
 425         union ib_gid local_gid;
 426         __be64 service_id;
 427         u16 pkey;
 428         bool has_gid:1;
 429 };
 430
 431 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 432 {
 433         unsigned long flags;
 434         int ret;
 435
 436         spin_lock_irqsave(&id_priv->lock, flags);
 437         ret = (id_priv->state == comp);
 438         spin_unlock_irqrestore(&id_priv->lock, flags);
 439         return ret;
 440 }
 441
 442 static int cma_comp_exch(struct rdma_id_private *id_priv,
 443                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 444 {
 445         unsigned long flags;
 446         int ret;
 447
 448         spin_lock_irqsave(&id_priv->lock, flags);
 449         if ((ret = (id_priv->state == comp)))
 450                 id_priv->state = exch;
 451         spin_unlock_irqrestore(&id_priv->lock, flags);
 452         return ret;
 453 }
 454
 455 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 456                                    enum rdma_cm_state exch)
 457 {
 458         unsigned long flags;
 459         enum rdma_cm_state old;
 460
 461         spin_lock_irqsave(&id_priv->lock, flags);
 462         old = id_priv->state;
 463         id_priv->state = exch;
 464         spin_unlock_irqrestore(&id_priv->lock, flags);
 465         return old;
 466 }
 467
 468 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 469 {
 470         return hdr->ip_version >> 4;
 471 }
 472
 473 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 474 {
 475         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 476 }
 477
 478 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 479 {
 480         struct in_device *in_dev = NULL;
 481
 482         if (ndev) {
 483                 rtnl_lock();
 484                 in_dev = __in_dev_get_rtnl(ndev);
 485                 if (in_dev) {
 486                         if (join)
 487                                 ip_mc_inc_group(in_dev,
 488                                                 *(__be32 *)(mgid->raw + 12));
 489                         else
 490                                 ip_mc_dec_group(in_dev,
 491                                                 *(__be32 *)(mgid->raw + 12));
 492                 }
 493                 rtnl_unlock();
 494         }
 495         return (in_dev) ? 0 : -ENODEV;
 496 }
 497
 498 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 499                                struct cma_device *cma_dev)
 500 {
 501         cma_ref_dev(cma_dev);
 502         id_priv->cma_dev = cma_dev;
 503         id_priv->gid_type = 0;
 504         id_priv->id.device = cma_dev->device;
 505         id_priv->id.route.addr.dev_addr.transport =
 506                 rdma_node_get_transport(cma_dev->device->node_type);
 507         list_add_tail(&id_priv->list, &cma_dev->id_list);
 508 }
 509
 510 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 511                               struct cma_device *cma_dev)
 512 {
 513         _cma_attach_to_dev(id_priv, cma_dev);
 514         id_priv->gid_type =
 515                 cma_dev->default_gid_type[id_priv->id.port_num -
 516                                           rdma_start_port(cma_dev->device)];
 517 }
 518
 519 void cma_deref_dev(struct cma_device *cma_dev)
 520 {
 521         if (atomic_dec_and_test(&cma_dev->refcount))
 522                 complete(&cma_dev->comp);
 523 }
 524
 525 static inline void release_mc(struct kref *kref)
 526 {
 527         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 528
 529         kfree(mc->multicast.ib);
 530         kfree(mc);
 531 }
 532
 533 static void cma_release_dev(struct rdma_id_private *id_priv)
 534 {
 535         mutex_lock(&lock);
 536         list_del(&id_priv->list);
 537         cma_deref_dev(id_priv->cma_dev);
 538         id_priv->cma_dev = NULL;
 539         mutex_unlock(&lock);
 540 }
 541
 542 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 543 {
 544         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 545 }
 546
 547 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 548 {
 549         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 550 }
 551
 552 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 553 {
 554         return id_priv->id.route.addr.src_addr.ss_family;
 555 }
 556
 557 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 558 {
 559         struct ib_sa_mcmember_rec rec;
 560         int ret = 0;
 561
 562         if (id_priv->qkey) {
 563                 if (qkey && id_priv->qkey != qkey)
 564                         return -EINVAL;
 565                 return 0;
 566         }
 567
 568         if (qkey) {
 569                 id_priv->qkey = qkey;
 570                 return 0;
 571         }
 572
 573         switch (id_priv->id.ps) {
 574         case RDMA_PS_UDP:
 575         case RDMA_PS_IB:
 576                 id_priv->qkey = RDMA_UDP_QKEY;
 577                 break;
 578         case RDMA_PS_IPOIB:
 579                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 580                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 581                                              id_priv->id.port_num, &rec.mgid,
 582                                              &rec);
 583                 if (!ret)
 584                         id_priv->qkey = be32_to_cpu(rec.qkey);
 585                 break;
 586         default:
 587                 break;
 588         }
 589         return ret;
 590 }
 591
 592 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 593 {
 594         dev_addr->dev_type = ARPHRD_INFINIBAND;
 595         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 596         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 597 }
 598
 599 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 600 {
 601         int ret;
 602
 603         if (addr->sa_family != AF_IB) {
 604                 ret = rdma_translate_ip(addr, dev_addr);
 605         } else {
 606                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 607                 ret = 0;
 608         }
 609
 610         return ret;
 611 }
 612
 613 static inline int cma_validate_port(struct ib_device *device, u8 port,
 614                                     enum ib_gid_type gid_type,
 615                                     union ib_gid *gid,
 616                                     struct rdma_id_private *id_priv)
 617 {
 618         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 619         int bound_if_index = dev_addr->bound_dev_if;
 620         int dev_type = dev_addr->dev_type;
 621         struct net_device *ndev = NULL;
 622         int ret = -ENODEV;
 623
 624         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 625                 return ret;
 626
 627         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 628                 return ret;
 629
 630         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 631                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 632                 if (!ndev)
 633                         return ret;
 634         } else {
 635                 gid_type = IB_GID_TYPE_IB;
 636         }
 637
 638         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 639                                          ndev, NULL);
 640
 641         if (ndev)
 642                 dev_put(ndev);
 643
 644         return ret;
 645 }
 646
 647 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 648                            struct rdma_id_private *listen_id_priv)
 649 {
 650         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 651         struct cma_device *cma_dev;
 652         union ib_gid gid, iboe_gid, *gidp;
 653         int ret = -ENODEV;
 654         u8 port;
 655
 656         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 657             id_priv->id.ps == RDMA_PS_IPOIB)
 658                 return -EINVAL;
 659
 660         mutex_lock(&lock);
 661         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 662                     &iboe_gid);
 663
 664         memcpy(&gid, dev_addr->src_dev_addr +
 665                rdma_addr_gid_offset(dev_addr), sizeof gid);
 666
 667         if (listen_id_priv) {
 668                 cma_dev = listen_id_priv->cma_dev;
 669                 port = listen_id_priv->id.port_num;
 670                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 671                        &iboe_gid : &gid;
 672
 673                 ret = cma_validate_port(cma_dev->device, port,
 674                                         rdma_protocol_ib(cma_dev->device, port) ?
 675                                         IB_GID_TYPE_IB :
 676                                         listen_id_priv->gid_type, gidp,
 677                                         id_priv);
 678                 if (!ret) {
 679                         id_priv->id.port_num = port;
 680                         goto out;
 681                 }
 682         }
 683
 684         list_for_each_entry(cma_dev, &dev_list, list) {
 685                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 686                         if (listen_id_priv &&
 687                             listen_id_priv->cma_dev == cma_dev &&
 688                             listen_id_priv->id.port_num == port)
 689                                 continue;
 690
 691                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 692                                &iboe_gid : &gid;
 693
 694                         ret = cma_validate_port(cma_dev->device, port,
 695                                                 rdma_protocol_ib(cma_dev->device, port) ?
 696                                                 IB_GID_TYPE_IB :
 697                                                 cma_dev->default_gid_type[port - 1],
 698                                                 gidp, id_priv);
 699                         if (!ret) {
 700                                 id_priv->id.port_num = port;
 701                                 goto out;
 702                         }
 703                 }
 704         }
 705
 706 out:
 707         if (!ret)
 708                 cma_attach_to_dev(id_priv, cma_dev);
 709
 710         mutex_unlock(&lock);
 711         return ret;
 712 }
 713
 714 /*
 715  * Select the source IB device and address to reach the destination IB address.
 716  */
 717 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 718 {
 719         struct cma_device *cma_dev, *cur_dev;
 720         struct sockaddr_ib *addr;
 721         union ib_gid gid, sgid, *dgid;
 722         u16 pkey, index;
 723         u8 p;
 724         enum ib_port_state port_state;
 725         int i;
 726
 727         cma_dev = NULL;
 728         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 729         dgid = (union ib_gid *) &addr->sib_addr;
 730         pkey = ntohs(addr->sib_pkey);
 731
 732         list_for_each_entry(cur_dev, &dev_list, list) {
 733                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 734                         if (!rdma_cap_af_ib(cur_dev->device, p))
 735                                 continue;
 736
 737                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 738                                 continue;
 739
 740                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 741                                 continue;
 742                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 743                                                        &gid, NULL);
 744                              i++) {
 745                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 746                                         cma_dev = cur_dev;
 747                                         sgid = gid;
 748                                         id_priv->id.port_num = p;
 749                                         goto found;
 750                                 }
 751
 752                                 if (!cma_dev && (gid.global.subnet_prefix ==
 753                                     dgid->global.subnet_prefix) &&
 754                                     port_state == IB_PORT_ACTIVE) {
 755                                         cma_dev = cur_dev;
 756                                         sgid = gid;
 757                                         id_priv->id.port_num = p;
 758                                 }
 759                         }
 760                 }
 761         }
 762
 763         if (!cma_dev)
 764                 return -ENODEV;
 765
 766 found:
 767         cma_attach_to_dev(id_priv, cma_dev);
 768         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 769         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 770         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 771         return 0;
 772 }
 773
 774 static void cma_deref_id(struct rdma_id_private *id_priv)
 775 {
 776         if (atomic_dec_and_test(&id_priv->refcount))
 777                 complete(&id_priv->comp);
 778 }
 779
 780 struct rdma_cm_id *rdma_create_id(struct net *net,
 781                                   rdma_cm_event_handler event_handler,
 782                                   void *context, enum rdma_port_space ps,
 783                                   enum ib_qp_type qp_type)
 784 {
 785         struct rdma_id_private *id_priv;
 786
 787         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 788         if (!id_priv)
 789                 return ERR_PTR(-ENOMEM);
 790
 791         id_priv->owner = task_pid_nr(current);
 792         id_priv->state = RDMA_CM_IDLE;
 793         id_priv->id.context = context;
 794         id_priv->id.event_handler = event_handler;
 795         id_priv->id.ps = ps;
 796         id_priv->id.qp_type = qp_type;
 797         id_priv->tos_set = false;
 798         spin_lock_init(&id_priv->lock);
 799         mutex_init(&id_priv->qp_mutex);
 800         init_completion(&id_priv->comp);
 801         atomic_set(&id_priv->refcount, 1);
 802         mutex_init(&id_priv->handler_mutex);
 803         INIT_LIST_HEAD(&id_priv->listen_list);
 804         INIT_LIST_HEAD(&id_priv->mc_list);
 805         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 806         id_priv->id.route.addr.dev_addr.net = get_net(net);
 807         id_priv->seq_num &= 0x00ffffff;
 808
 809         return &id_priv->id;
 810 }
 811 EXPORT_SYMBOL(rdma_create_id);
 812
 813 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 814 {
 815         struct ib_qp_attr qp_attr;
 816         int qp_attr_mask, ret;
 817
 818         qp_attr.qp_state = IB_QPS_INIT;
 819         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 820         if (ret)
 821                 return ret;
 822
 823         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 824         if (ret)
 825                 return ret;
 826
 827         qp_attr.qp_state = IB_QPS_RTR;
 828         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 829         if (ret)
 830                 return ret;
 831
 832         qp_attr.qp_state = IB_QPS_RTS;
 833         qp_attr.sq_psn = 0;
 834         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 835
 836         return ret;
 837 }
 838
 839 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 840 {
 841         struct ib_qp_attr qp_attr;
 842         int qp_attr_mask, ret;
 843
 844         qp_attr.qp_state = IB_QPS_INIT;
 845         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 846         if (ret)
 847                 return ret;
 848
 849         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 850 }
 851
 852 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 853                    struct ib_qp_init_attr *qp_init_attr)
 854 {
 855         struct rdma_id_private *id_priv;
 856         struct ib_qp *qp;
 857         int ret;
 858
 859         id_priv = container_of(id, struct rdma_id_private, id);
 860         if (id->device != pd->device)
 861                 return -EINVAL;
 862
 863         qp_init_attr->port_num = id->port_num;
 864         qp = ib_create_qp(pd, qp_init_attr);
 865         if (IS_ERR(qp))
 866                 return PTR_ERR(qp);
 867
 868         if (id->qp_type == IB_QPT_UD)
 869                 ret = cma_init_ud_qp(id_priv, qp);
 870         else
 871                 ret = cma_init_conn_qp(id_priv, qp);
 872         if (ret)
 873                 goto err;
 874
 875         id->qp = qp;
 876         id_priv->qp_num = qp->qp_num;
 877         id_priv->srq = (qp->srq != NULL);
 878         return 0;
 879 err:
 880         ib_destroy_qp(qp);
 881         return ret;
 882 }
 883 EXPORT_SYMBOL(rdma_create_qp);
 884
 885 void rdma_destroy_qp(struct rdma_cm_id *id)
 886 {
 887         struct rdma_id_private *id_priv;
 888
 889         id_priv = container_of(id, struct rdma_id_private, id);
 890         mutex_lock(&id_priv->qp_mutex);
 891         ib_destroy_qp(id_priv->id.qp);
 892         id_priv->id.qp = NULL;
 893         mutex_unlock(&id_priv->qp_mutex);
 894 }
 895 EXPORT_SYMBOL(rdma_destroy_qp);
 896
 897 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 898                              struct rdma_conn_param *conn_param)
 899 {
 900         struct ib_qp_attr qp_attr;
 901         int qp_attr_mask, ret;
 902         union ib_gid sgid;
 903
 904         mutex_lock(&id_priv->qp_mutex);
 905         if (!id_priv->id.qp) {
 906                 ret = 0;
 907                 goto out;
 908         }
 909
 910         /* Need to update QP attributes from default values. */
 911         qp_attr.qp_state = IB_QPS_INIT;
 912         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 913         if (ret)
 914                 goto out;
 915
 916         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 917         if (ret)
 918                 goto out;
 919
 920         qp_attr.qp_state = IB_QPS_RTR;
 921         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 922         if (ret)
 923                 goto out;
 924
 925         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 926                            rdma_ah_read_grh(&qp_attr.ah_attr)->sgid_index,
 927                            &sgid, NULL);
 928         if (ret)
 929                 goto out;
 930
 931         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 932
 933         if (conn_param)
 934                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 935         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 936 out:
 937         mutex_unlock(&id_priv->qp_mutex);
 938         return ret;
 939 }
 940
 941 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 942                              struct rdma_conn_param *conn_param)
 943 {
 944         struct ib_qp_attr qp_attr;
 945         int qp_attr_mask, ret;
 946
 947         mutex_lock(&id_priv->qp_mutex);
 948         if (!id_priv->id.qp) {
 949                 ret = 0;
 950                 goto out;
 951         }
 952
 953         qp_attr.qp_state = IB_QPS_RTS;
 954         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 955         if (ret)
 956                 goto out;
 957
 958         if (conn_param)
 959                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 960         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 961 out:
 962         mutex_unlock(&id_priv->qp_mutex);
 963         return ret;
 964 }
 965
 966 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 967 {
 968         struct ib_qp_attr qp_attr;
 969         int ret;
 970
 971         mutex_lock(&id_priv->qp_mutex);
 972         if (!id_priv->id.qp) {
 973                 ret = 0;
 974                 goto out;
 975         }
 976
 977         qp_attr.qp_state = IB_QPS_ERR;
 978         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 979 out:
 980         mutex_unlock(&id_priv->qp_mutex);
 981         return ret;
 982 }
 983
 984 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 985                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 986 {
 987         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 988         int ret;
 989         u16 pkey;
 990
 991         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 992                 pkey = 0xffff;
 993         else
 994                 pkey = ib_addr_get_pkey(dev_addr);
 995
 996         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 997                                   pkey, &qp_attr->pkey_index);
 998         if (ret)
 999                 return ret;
1000
1001         qp_attr->port_num = id_priv->id.port_num;
1002         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1003
1004         if (id_priv->id.qp_type == IB_QPT_UD) {
1005                 ret = cma_set_qkey(id_priv, 0);
1006                 if (ret)
1007                         return ret;
1008
1009                 qp_attr->qkey = id_priv->qkey;
1010                 *qp_attr_mask |= IB_QP_QKEY;
1011         } else {
1012                 qp_attr->qp_access_flags = 0;
1013                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1014         }
1015         return 0;
1016 }
1017
1018 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1019                        int *qp_attr_mask)
1020 {
1021         struct rdma_id_private *id_priv;
1022         int ret = 0;
1023
1024         id_priv = container_of(id, struct rdma_id_private, id);
1025         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1026                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1027                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1028                 else
1029                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1030                                                  qp_attr_mask);
1031
1032                 if (qp_attr->qp_state == IB_QPS_RTR)
1033                         qp_attr->rq_psn = id_priv->seq_num;
1034         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1035                 if (!id_priv->cm_id.iw) {
1036                         qp_attr->qp_access_flags = 0;
1037                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1038                 } else
1039                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1040                                                  qp_attr_mask);
1041                 qp_attr->port_num = id_priv->id.port_num;
1042                 *qp_attr_mask |= IB_QP_PORT;
1043         } else
1044                 ret = -ENOSYS;
1045
1046         return ret;
1047 }
1048 EXPORT_SYMBOL(rdma_init_qp_attr);
1049
1050 static inline int cma_zero_addr(struct sockaddr *addr)
1051 {
1052         switch (addr->sa_family) {
1053         case AF_INET:
1054                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1055         case AF_INET6:
1056                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1057         case AF_IB:
1058                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1059         default:
1060                 return 0;
1061         }
1062 }
1063
1064 static inline int cma_loopback_addr(struct sockaddr *addr)
1065 {
1066         switch (addr->sa_family) {
1067         case AF_INET:
1068                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1069         case AF_INET6:
1070                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1071         case AF_IB:
1072                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1073         default:
1074                 return 0;
1075         }
1076 }
1077
1078 static inline int cma_any_addr(struct sockaddr *addr)
1079 {
1080         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1081 }
1082
1083 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1084 {
1085         if (src->sa_family != dst->sa_family)
1086                 return -1;
1087
1088         switch (src->sa_family) {
1089         case AF_INET:
1090                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1091                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1092         case AF_INET6:
1093                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1094                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1095         default:
1096                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1097                                    &((struct sockaddr_ib *) dst)->sib_addr);
1098         }
1099 }
1100
1101 static __be16 cma_port(struct sockaddr *addr)
1102 {
1103         struct sockaddr_ib *sib;
1104
1105         switch (addr->sa_family) {
1106         case AF_INET:
1107                 return ((struct sockaddr_in *) addr)->sin_port;
1108         case AF_INET6:
1109                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1110         case AF_IB:
1111                 sib = (struct sockaddr_ib *) addr;
1112                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1113                                     be64_to_cpu(sib->sib_sid_mask)));
1114         default:
1115                 return 0;
1116         }
1117 }
1118
1119 static inline int cma_any_port(struct sockaddr *addr)
1120 {
1121         return !cma_port(addr);
1122 }
1123
1124 static void cma_save_ib_info(struct sockaddr *src_addr,
1125                              struct sockaddr *dst_addr,
1126                              struct rdma_cm_id *listen_id,
1127                              struct sa_path_rec *path)
1128 {
1129         struct sockaddr_ib *listen_ib, *ib;
1130
1131         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1132         if (src_addr) {
1133                 ib = (struct sockaddr_ib *)src_addr;
1134                 ib->sib_family = AF_IB;
1135                 if (path) {
1136                         ib->sib_pkey = path->pkey;
1137                         ib->sib_flowinfo = path->flow_label;
1138                         memcpy(&ib->sib_addr, &path->sgid, 16);
1139                         ib->sib_sid = path->service_id;
1140                         ib->sib_scope_id = 0;
1141                 } else {
1142                         ib->sib_pkey = listen_ib->sib_pkey;
1143                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1144                         ib->sib_addr = listen_ib->sib_addr;
1145                         ib->sib_sid = listen_ib->sib_sid;
1146                         ib->sib_scope_id = listen_ib->sib_scope_id;
1147                 }
1148                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1149         }
1150         if (dst_addr) {
1151                 ib = (struct sockaddr_ib *)dst_addr;
1152                 ib->sib_family = AF_IB;
1153                 if (path) {
1154                         ib->sib_pkey = path->pkey;
1155                         ib->sib_flowinfo = path->flow_label;
1156                         memcpy(&ib->sib_addr, &path->dgid, 16);
1157                 }
1158         }
1159 }
1160
1161 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1162                               struct sockaddr_in *dst_addr,
1163                               struct cma_hdr *hdr,
1164                               __be16 local_port)
1165 {
1166         if (src_addr) {
1167                 *src_addr = (struct sockaddr_in) {
1168                         .sin_family = AF_INET,
1169                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1170                         .sin_port = local_port,
1171                 };
1172         }
1173
1174         if (dst_addr) {
1175                 *dst_addr = (struct sockaddr_in) {
1176                         .sin_family = AF_INET,
1177                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1178                         .sin_port = hdr->port,
1179                 };
1180         }
1181 }
1182
1183 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1184                               struct sockaddr_in6 *dst_addr,
1185                               struct cma_hdr *hdr,
1186                               __be16 local_port)
1187 {
1188         if (src_addr) {
1189                 *src_addr = (struct sockaddr_in6) {
1190                         .sin6_family = AF_INET6,
1191                         .sin6_addr = hdr->dst_addr.ip6,
1192                         .sin6_port = local_port,
1193                 };
1194         }
1195
1196         if (dst_addr) {
1197                 *dst_addr = (struct sockaddr_in6) {
1198                         .sin6_family = AF_INET6,
1199                         .sin6_addr = hdr->src_addr.ip6,
1200                         .sin6_port = hdr->port,
1201                 };
1202         }
1203 }
1204
1205 static u16 cma_port_from_service_id(__be64 service_id)
1206 {
1207         return (u16)be64_to_cpu(service_id);
1208 }
1209
1210 static int cma_save_ip_info(struct sockaddr *src_addr,
1211                             struct sockaddr *dst_addr,
1212                             struct ib_cm_event *ib_event,
1213                             __be64 service_id)
1214 {
1215         struct cma_hdr *hdr;
1216         __be16 port;
1217
1218         hdr = ib_event->private_data;
1219         if (hdr->cma_version != CMA_VERSION)
1220                 return -EINVAL;
1221
1222         port = htons(cma_port_from_service_id(service_id));
1223
1224         switch (cma_get_ip_ver(hdr)) {
1225         case 4:
1226                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1227                                   (struct sockaddr_in *)dst_addr, hdr, port);
1228                 break;
1229         case 6:
1230                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1231                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1232                 break;
1233         default:
1234                 return -EAFNOSUPPORT;
1235         }
1236
1237         return 0;
1238 }
1239
1240 static int cma_save_net_info(struct sockaddr *src_addr,
1241                              struct sockaddr *dst_addr,
1242                              struct rdma_cm_id *listen_id,
1243                              struct ib_cm_event *ib_event,
1244                              sa_family_t sa_family, __be64 service_id)
1245 {
1246         if (sa_family == AF_IB) {
1247                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1248                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1249                                          ib_event->param.req_rcvd.primary_path);
1250                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1251                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1252                 return 0;
1253         }
1254
1255         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1256 }
1257
1258 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1259                              struct cma_req_info *req)
1260 {
1261         const struct ib_cm_req_event_param *req_param =
1262                 &ib_event->param.req_rcvd;
1263         const struct ib_cm_sidr_req_event_param *sidr_param =
1264                 &ib_event->param.sidr_req_rcvd;
1265
1266         switch (ib_event->event) {
1267         case IB_CM_REQ_RECEIVED:
1268                 req->device     = req_param->listen_id->device;
1269                 req->port       = req_param->port;
1270                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1271                        sizeof(req->local_gid));
1272                 req->has_gid    = true;
1273                 req->service_id = req_param->primary_path->service_id;
1274                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1275                 if (req->pkey != req_param->bth_pkey)
1276                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1277                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1278                                             req_param->bth_pkey, req->pkey);
1279                 break;
1280         case IB_CM_SIDR_REQ_RECEIVED:
1281                 req->device     = sidr_param->listen_id->device;
1282                 req->port       = sidr_param->port;
1283                 req->has_gid    = false;
1284                 req->service_id = sidr_param->service_id;
1285                 req->pkey       = sidr_param->pkey;
1286                 if (req->pkey != sidr_param->bth_pkey)
1287                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1288                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1289                                             sidr_param->bth_pkey, req->pkey);
1290                 break;
1291         default:
1292                 return -EINVAL;
1293         }
1294
1295         return 0;
1296 }
1297
1298 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1299                                   const struct sockaddr_in *dst_addr,
1300                                   const struct sockaddr_in *src_addr)
1301 {
1302         __be32 daddr = dst_addr->sin_addr.s_addr,
1303                saddr = src_addr->sin_addr.s_addr;
1304         struct fib_result res;
1305         struct flowi4 fl4;
1306         int err;
1307         bool ret;
1308
1309         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1310             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1311             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1312             ipv4_is_loopback(saddr))
1313                 return false;
1314
1315         memset(&fl4, 0, sizeof(fl4));
1316         fl4.flowi4_iif = net_dev->ifindex;
1317         fl4.daddr = daddr;
1318         fl4.saddr = saddr;
1319
1320         rcu_read_lock();
1321         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1322         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1323         rcu_read_unlock();
1324
1325         return ret;
1326 }
1327
1328 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1329                                   const struct sockaddr_in6 *dst_addr,
1330                                   const struct sockaddr_in6 *src_addr)
1331 {
1332 #if IS_ENABLED(CONFIG_IPV6)
1333         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1334                            IPV6_ADDR_LINKLOCAL;
1335         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1336                                          &src_addr->sin6_addr, net_dev->ifindex,
1337                                          NULL, strict);
1338         bool ret;
1339
1340         if (!rt)
1341                 return false;
1342
1343         ret = rt->rt6i_idev->dev == net_dev;
1344         ip6_rt_put(rt);
1345
1346         return ret;
1347 #else
1348         return false;
1349 #endif
1350 }
1351
1352 static bool validate_net_dev(struct net_device *net_dev,
1353                              const struct sockaddr *daddr,
1354                              const struct sockaddr *saddr)
1355 {
1356         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1357         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1358         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1359         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1360
1361         switch (daddr->sa_family) {
1362         case AF_INET:
1363                 return saddr->sa_family == AF_INET &&
1364                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1365
1366         case AF_INET6:
1367                 return saddr->sa_family == AF_INET6 &&
1368                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1369
1370         default:
1371                 return false;
1372         }
1373 }
1374
1375 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1376                                           const struct cma_req_info *req)
1377 {
1378         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1379         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1380                         *src_addr = (struct sockaddr *)&src_addr_storage;
1381         struct net_device *net_dev;
1382         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1383         int err;
1384
1385         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1386                                req->service_id);
1387         if (err)
1388                 return ERR_PTR(err);
1389
1390         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1391                                            gid, listen_addr);
1392         if (!net_dev)
1393                 return ERR_PTR(-ENODEV);
1394
1395         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1396                 dev_put(net_dev);
1397                 return ERR_PTR(-EHOSTUNREACH);
1398         }
1399
1400         return net_dev;
1401 }
1402
1403 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1404 {
1405         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1406 }
1407
1408 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1409                                    const struct cma_hdr *hdr)
1410 {
1411         struct sockaddr *addr = cma_src_addr(id_priv);
1412         __be32 ip4_addr;
1413         struct in6_addr ip6_addr;
1414
1415         if (cma_any_addr(addr) && !id_priv->afonly)
1416                 return true;
1417
1418         switch (addr->sa_family) {
1419         case AF_INET:
1420                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1421                 if (cma_get_ip_ver(hdr) != 4)
1422                         return false;
1423                 if (!cma_any_addr(addr) &&
1424                     hdr->dst_addr.ip4.addr != ip4_addr)
1425                         return false;
1426                 break;
1427         case AF_INET6:
1428                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1429                 if (cma_get_ip_ver(hdr) != 6)
1430                         return false;
1431                 if (!cma_any_addr(addr) &&
1432                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1433                         return false;
1434                 break;
1435         case AF_IB:
1436                 return true;
1437         default:
1438                 return false;
1439         }
1440
1441         return true;
1442 }
1443
1444 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1445 {
1446         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1447         enum rdma_transport_type transport =
1448                 rdma_node_get_transport(device->node_type);
1449
1450         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1451 }
1452
1453 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1454 {
1455         struct ib_device *device = id->device;
1456         const int port_num = id->port_num ?: rdma_start_port(device);
1457
1458         return cma_protocol_roce_dev_port(device, port_num);
1459 }
1460
1461 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1462                               const struct net_device *net_dev,
1463                               u8 port_num)
1464 {
1465         const struct rdma_addr *addr = &id->route.addr;
1466
1467         if (!net_dev)
1468                 /* This request is an AF_IB request or a RoCE request */
1469                 return (!id->port_num || id->port_num == port_num) &&
1470                        (addr->src_addr.ss_family == AF_IB ||
1471                         cma_protocol_roce_dev_port(id->device, port_num));
1472
1473         return !addr->dev_addr.bound_dev_if ||
1474                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1475                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1476 }
1477
1478 static struct rdma_id_private *cma_find_listener(
1479                 const struct rdma_bind_list *bind_list,
1480                 const struct ib_cm_id *cm_id,
1481                 const struct ib_cm_event *ib_event,
1482                 const struct cma_req_info *req,
1483                 const struct net_device *net_dev)
1484 {
1485         struct rdma_id_private *id_priv, *id_priv_dev;
1486
1487         if (!bind_list)
1488                 return ERR_PTR(-EINVAL);
1489
1490         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1491                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1492                         if (id_priv->id.device == cm_id->device &&
1493                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1494                                 return id_priv;
1495                         list_for_each_entry(id_priv_dev,
1496                                             &id_priv->listen_list,
1497                                             listen_list) {
1498                                 if (id_priv_dev->id.device == cm_id->device &&
1499                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1500                                         return id_priv_dev;
1501                         }
1502                 }
1503         }
1504
1505         return ERR_PTR(-EINVAL);
1506 }
1507
1508 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1509                                                  struct ib_cm_event *ib_event,
1510                                                  struct net_device **net_dev)
1511 {
1512         struct cma_req_info req;
1513         struct rdma_bind_list *bind_list;
1514         struct rdma_id_private *id_priv;
1515         int err;
1516
1517         err = cma_save_req_info(ib_event, &req);
1518         if (err)
1519                 return ERR_PTR(err);
1520
1521         *net_dev = cma_get_net_dev(ib_event, &req);
1522         if (IS_ERR(*net_dev)) {
1523                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1524                         /* Assuming the protocol is AF_IB */
1525                         *net_dev = NULL;
1526                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1527                         /* TODO find the net dev matching the request parameters
1528                          * through the RoCE GID table */
1529                         *net_dev = NULL;
1530                 } else {
1531                         return ERR_CAST(*net_dev);
1532                 }
1533         }
1534
1535         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1536                                 rdma_ps_from_service_id(req.service_id),
1537                                 cma_port_from_service_id(req.service_id));
1538         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1539         if (IS_ERR(id_priv) && *net_dev) {
1540                 dev_put(*net_dev);
1541                 *net_dev = NULL;
1542         }
1543
1544         return id_priv;
1545 }
1546
1547 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1548 {
1549         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1550 }
1551
1552 static void cma_cancel_route(struct rdma_id_private *id_priv)
1553 {
1554         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1555                 if (id_priv->query)
1556                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1557         }
1558 }
1559
1560 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1561 {
1562         struct rdma_id_private *dev_id_priv;
1563
1564         /*
1565          * Remove from listen_any_list to prevent added devices from spawning
1566          * additional listen requests.
1567          */
1568         mutex_lock(&lock);
1569         list_del(&id_priv->list);
1570
1571         while (!list_empty(&id_priv->listen_list)) {
1572                 dev_id_priv = list_entry(id_priv->listen_list.next,
1573                                          struct rdma_id_private, listen_list);
1574                 /* sync with device removal to avoid duplicate destruction */
1575                 list_del_init(&dev_id_priv->list);
1576                 list_del(&dev_id_priv->listen_list);
1577                 mutex_unlock(&lock);
1578
1579                 rdma_destroy_id(&dev_id_priv->id);
1580                 mutex_lock(&lock);
1581         }
1582         mutex_unlock(&lock);
1583 }
1584
1585 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1586                                  enum rdma_cm_state state)
1587 {
1588         switch (state) {
1589         case RDMA_CM_ADDR_QUERY:
1590                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1591                 break;
1592         case RDMA_CM_ROUTE_QUERY:
1593                 cma_cancel_route(id_priv);
1594                 break;
1595         case RDMA_CM_LISTEN:
1596                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1597                         cma_cancel_listens(id_priv);
1598                 break;
1599         default:
1600                 break;
1601         }
1602 }
1603
1604 static void cma_release_port(struct rdma_id_private *id_priv)
1605 {
1606         struct rdma_bind_list *bind_list = id_priv->bind_list;
1607         struct net *net = id_priv->id.route.addr.dev_addr.net;
1608
1609         if (!bind_list)
1610                 return;
1611
1612         mutex_lock(&lock);
1613         hlist_del(&id_priv->node);
1614         if (hlist_empty(&bind_list->owners)) {
1615                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1616                 kfree(bind_list);
1617         }
1618         mutex_unlock(&lock);
1619 }
1620
1621 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1622 {
1623         struct cma_multicast *mc;
1624
1625         while (!list_empty(&id_priv->mc_list)) {
1626                 mc = container_of(id_priv->mc_list.next,
1627                                   struct cma_multicast, list);
1628                 list_del(&mc->list);
1629                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1630                                       id_priv->id.port_num)) {
1631                         ib_sa_free_multicast(mc->multicast.ib);
1632                         kfree(mc);
1633                 } else {
1634                         if (mc->igmp_joined) {
1635                                 struct rdma_dev_addr *dev_addr =
1636                                         &id_priv->id.route.addr.dev_addr;
1637                                 struct net_device *ndev = NULL;
1638
1639                                 if (dev_addr->bound_dev_if)
1640                                         ndev = dev_get_by_index(&init_net,
1641                                                                 dev_addr->bound_dev_if);
1642                                 if (ndev) {
1643                                         cma_igmp_send(ndev,
1644                                                       &mc->multicast.ib->rec.mgid,
1645                                                       false);
1646                                         dev_put(ndev);
1647                                 }
1648                         }
1649                         kref_put(&mc->mcref, release_mc);
1650                 }
1651         }
1652 }
1653
1654 void rdma_destroy_id(struct rdma_cm_id *id)
1655 {
1656         struct rdma_id_private *id_priv;
1657         enum rdma_cm_state state;
1658
1659         id_priv = container_of(id, struct rdma_id_private, id);
1660         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1661         cma_cancel_operation(id_priv, state);
1662
1663         /*
1664          * Wait for any active callback to finish.  New callbacks will find
1665          * the id_priv state set to destroying and abort.
1666          */
1667         mutex_lock(&id_priv->handler_mutex);
1668         mutex_unlock(&id_priv->handler_mutex);
1669
1670         if (id_priv->cma_dev) {
1671                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1672                         if (id_priv->cm_id.ib)
1673                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1674                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1675                         if (id_priv->cm_id.iw)
1676                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1677                 }
1678                 cma_leave_mc_groups(id_priv);
1679                 cma_release_dev(id_priv);
1680         }
1681
1682         cma_release_port(id_priv);
1683         cma_deref_id(id_priv);
1684         wait_for_completion(&id_priv->comp);
1685
1686         if (id_priv->internal_id)
1687                 cma_deref_id(id_priv->id.context);
1688
1689         kfree(id_priv->id.route.path_rec);
1690         put_net(id_priv->id.route.addr.dev_addr.net);
1691         kfree(id_priv);
1692 }
1693 EXPORT_SYMBOL(rdma_destroy_id);
1694
1695 static int cma_rep_recv(struct rdma_id_private *id_priv)
1696 {
1697         int ret;
1698
1699         ret = cma_modify_qp_rtr(id_priv, NULL);
1700         if (ret)
1701                 goto reject;
1702
1703         ret = cma_modify_qp_rts(id_priv, NULL);
1704         if (ret)
1705                 goto reject;
1706
1707         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1708         if (ret)
1709                 goto reject;
1710
1711         return 0;
1712 reject:
1713         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1714         cma_modify_qp_err(id_priv);
1715         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1716                        NULL, 0, NULL, 0);
1717         return ret;
1718 }
1719
1720 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1721                                    struct ib_cm_rep_event_param *rep_data,
1722                                    void *private_data)
1723 {
1724         event->param.conn.private_data = private_data;
1725         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1726         event->param.conn.responder_resources = rep_data->responder_resources;
1727         event->param.conn.initiator_depth = rep_data->initiator_depth;
1728         event->param.conn.flow_control = rep_data->flow_control;
1729         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1730         event->param.conn.srq = rep_data->srq;
1731         event->param.conn.qp_num = rep_data->remote_qpn;
1732 }
1733
1734 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1735 {
1736         struct rdma_id_private *id_priv = cm_id->context;
1737         struct rdma_cm_event event;
1738         int ret = 0;
1739
1740         mutex_lock(&id_priv->handler_mutex);
1741         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1742              id_priv->state != RDMA_CM_CONNECT) ||
1743             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1744              id_priv->state != RDMA_CM_DISCONNECT))
1745                 goto out;
1746
1747         memset(&event, 0, sizeof event);
1748         switch (ib_event->event) {
1749         case IB_CM_REQ_ERROR:
1750         case IB_CM_REP_ERROR:
1751                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1752                 event.status = -ETIMEDOUT;
1753                 break;
1754         case IB_CM_REP_RECEIVED:
1755                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1756                     (id_priv->id.qp_type != IB_QPT_UD))
1757                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1758                 if (id_priv->id.qp) {
1759                         event.status = cma_rep_recv(id_priv);
1760                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1761                                                      RDMA_CM_EVENT_ESTABLISHED;
1762                 } else {
1763                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1764                 }
1765                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1766                                        ib_event->private_data);
1767                 break;
1768         case IB_CM_RTU_RECEIVED:
1769         case IB_CM_USER_ESTABLISHED:
1770                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1771                 break;
1772         case IB_CM_DREQ_ERROR:
1773                 event.status = -ETIMEDOUT; /* fall through */
1774         case IB_CM_DREQ_RECEIVED:
1775         case IB_CM_DREP_RECEIVED:
1776                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1777                                    RDMA_CM_DISCONNECT))
1778                         goto out;
1779                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1780                 break;
1781         case IB_CM_TIMEWAIT_EXIT:
1782                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1783                 break;
1784         case IB_CM_MRA_RECEIVED:
1785                 /* ignore event */
1786                 goto out;
1787         case IB_CM_REJ_RECEIVED:
1788                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1789                                                                                 ib_event->param.rej_rcvd.reason));
1790                 cma_modify_qp_err(id_priv);
1791                 event.status = ib_event->param.rej_rcvd.reason;
1792                 event.event = RDMA_CM_EVENT_REJECTED;
1793                 event.param.conn.private_data = ib_event->private_data;
1794                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1795                 break;
1796         default:
1797                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1798                        ib_event->event);
1799                 goto out;
1800         }
1801
1802         ret = id_priv->id.event_handler(&id_priv->id, &event);
1803         if (ret) {
1804                 /* Destroy the CM ID by returning a non-zero value. */
1805                 id_priv->cm_id.ib = NULL;
1806                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1807                 mutex_unlock(&id_priv->handler_mutex);
1808                 rdma_destroy_id(&id_priv->id);
1809                 return ret;
1810         }
1811 out:
1812         mutex_unlock(&id_priv->handler_mutex);
1813         return ret;
1814 }
1815
1816 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1817                                                struct ib_cm_event *ib_event,
1818                                                struct net_device *net_dev)
1819 {
1820         struct rdma_id_private *id_priv;
1821         struct rdma_cm_id *id;
1822         struct rdma_route *rt;
1823         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1824         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1825         const __be64 service_id =
1826                 ib_event->param.req_rcvd.primary_path->service_id;
1827         int ret;
1828
1829         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1830                             listen_id->event_handler, listen_id->context,
1831                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1832         if (IS_ERR(id))
1833                 return NULL;
1834
1835         id_priv = container_of(id, struct rdma_id_private, id);
1836         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1837                               (struct sockaddr *)&id->route.addr.dst_addr,
1838                               listen_id, ib_event, ss_family, service_id))
1839                 goto err;
1840
1841         rt = &id->route;
1842         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1843         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1844                                GFP_KERNEL);
1845         if (!rt->path_rec)
1846                 goto err;
1847
1848         rt->path_rec[0] = *path;
1849         if (rt->num_paths == 2)
1850                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1851
1852         if (net_dev) {
1853                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1854         } else {
1855                 if (!cma_protocol_roce(listen_id) &&
1856                     cma_any_addr(cma_src_addr(id_priv))) {
1857                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1858                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1859                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1860                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1861                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1862                         if (ret)
1863                                 goto err;
1864                 }
1865         }
1866         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1867
1868         id_priv->state = RDMA_CM_CONNECT;
1869         return id_priv;
1870
1871 err:
1872         rdma_destroy_id(id);
1873         return NULL;
1874 }
1875
1876 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1877                                               struct ib_cm_event *ib_event,
1878                                               struct net_device *net_dev)
1879 {
1880         struct rdma_id_private *id_priv;
1881         struct rdma_cm_id *id;
1882         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1883         struct net *net = listen_id->route.addr.dev_addr.net;
1884         int ret;
1885
1886         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1887                             listen_id->ps, IB_QPT_UD);
1888         if (IS_ERR(id))
1889                 return NULL;
1890
1891         id_priv = container_of(id, struct rdma_id_private, id);
1892         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1893                               (struct sockaddr *)&id->route.addr.dst_addr,
1894                               listen_id, ib_event, ss_family,
1895                               ib_event->param.sidr_req_rcvd.service_id))
1896                 goto err;
1897
1898         if (net_dev) {
1899                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1900         } else {
1901                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1902                         ret = cma_translate_addr(cma_src_addr(id_priv),
1903                                                  &id->route.addr.dev_addr);
1904                         if (ret)
1905                                 goto err;
1906                 }
1907         }
1908
1909         id_priv->state = RDMA_CM_CONNECT;
1910         return id_priv;
1911 err:
1912         rdma_destroy_id(id);
1913         return NULL;
1914 }
1915
1916 static void cma_set_req_event_data(struct rdma_cm_event *event,
1917                                    struct ib_cm_req_event_param *req_data,
1918                                    void *private_data, int offset)
1919 {
1920         event->param.conn.private_data = private_data + offset;
1921         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1922         event->param.conn.responder_resources = req_data->responder_resources;
1923         event->param.conn.initiator_depth = req_data->initiator_depth;
1924         event->param.conn.flow_control = req_data->flow_control;
1925         event->param.conn.retry_count = req_data->retry_count;
1926         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1927         event->param.conn.srq = req_data->srq;
1928         event->param.conn.qp_num = req_data->remote_qpn;
1929 }
1930
1931 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1932 {
1933         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1934                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1935                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1936                  (id->qp_type == IB_QPT_UD)) ||
1937                 (!id->qp_type));
1938 }
1939
1940 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1941 {
1942         struct rdma_id_private *listen_id, *conn_id = NULL;
1943         struct rdma_cm_event event;
1944         struct net_device *net_dev;
1945         u8 offset;
1946         int ret;
1947
1948         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1949         if (IS_ERR(listen_id))
1950                 return PTR_ERR(listen_id);
1951
1952         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1953                 ret = -EINVAL;
1954                 goto net_dev_put;
1955         }
1956
1957         mutex_lock(&listen_id->handler_mutex);
1958         if (listen_id->state != RDMA_CM_LISTEN) {
1959                 ret = -ECONNABORTED;
1960                 goto err1;
1961         }
1962
1963         memset(&event, 0, sizeof event);
1964         offset = cma_user_data_offset(listen_id);
1965         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1966         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1967                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1968                 event.param.ud.private_data = ib_event->private_data + offset;
1969                 event.param.ud.private_data_len =
1970                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1971         } else {
1972                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1973                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1974                                        ib_event->private_data, offset);
1975         }
1976         if (!conn_id) {
1977                 ret = -ENOMEM;
1978                 goto err1;
1979         }
1980
1981         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1982         ret = cma_acquire_dev(conn_id, listen_id);
1983         if (ret)
1984                 goto err2;
1985
1986         conn_id->cm_id.ib = cm_id;
1987         cm_id->context = conn_id;
1988         cm_id->cm_handler = cma_ib_handler;
1989
1990         /*
1991          * Protect against the user destroying conn_id from another thread
1992          * until we're done accessing it.
1993          */
1994         atomic_inc(&conn_id->refcount);
1995         ret = conn_id->id.event_handler(&conn_id->id, &event);
1996         if (ret)
1997                 goto err3;
1998         /*
1999          * Acquire mutex to prevent user executing rdma_destroy_id()
2000          * while we're accessing the cm_id.
2001          */
2002         mutex_lock(&lock);
2003         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2004             (conn_id->id.qp_type != IB_QPT_UD))
2005                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2006         mutex_unlock(&lock);
2007         mutex_unlock(&conn_id->handler_mutex);
2008         mutex_unlock(&listen_id->handler_mutex);
2009         cma_deref_id(conn_id);
2010         if (net_dev)
2011                 dev_put(net_dev);
2012         return 0;
2013
2014 err3:
2015         cma_deref_id(conn_id);
2016         /* Destroy the CM ID by returning a non-zero value. */
2017         conn_id->cm_id.ib = NULL;
2018 err2:
2019         cma_exch(conn_id, RDMA_CM_DESTROYING);
2020         mutex_unlock(&conn_id->handler_mutex);
2021 err1:
2022         mutex_unlock(&listen_id->handler_mutex);
2023         if (conn_id)
2024                 rdma_destroy_id(&conn_id->id);
2025
2026 net_dev_put:
2027         if (net_dev)
2028                 dev_put(net_dev);
2029
2030         return ret;
2031 }
2032
2033 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2034 {
2035         if (addr->sa_family == AF_IB)
2036                 return ((struct sockaddr_ib *) addr)->sib_sid;
2037
2038         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2039 }
2040 EXPORT_SYMBOL(rdma_get_service_id);
2041
2042 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2043                     union ib_gid *dgid)
2044 {
2045         struct rdma_addr *addr = &cm_id->route.addr;
2046
2047         if (!cm_id->device) {
2048                 if (sgid)
2049                         memset(sgid, 0, sizeof(*sgid));
2050                 if (dgid)
2051                         memset(dgid, 0, sizeof(*dgid));
2052                 return;
2053         }
2054
2055         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2056                 if (sgid)
2057                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2058                 if (dgid)
2059                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2060         } else {
2061                 if (sgid)
2062                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2063                 if (dgid)
2064                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2065         }
2066 }
2067 EXPORT_SYMBOL(rdma_read_gids);
2068
2069 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2070 {
2071         struct rdma_id_private *id_priv = iw_id->context;
2072         struct rdma_cm_event event;
2073         int ret = 0;
2074         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2075         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2076
2077         mutex_lock(&id_priv->handler_mutex);
2078         if (id_priv->state != RDMA_CM_CONNECT)
2079                 goto out;
2080
2081         memset(&event, 0, sizeof event);
2082         switch (iw_event->event) {
2083         case IW_CM_EVENT_CLOSE:
2084                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2085                 break;
2086         case IW_CM_EVENT_CONNECT_REPLY:
2087                 memcpy(cma_src_addr(id_priv), laddr,
2088                        rdma_addr_size(laddr));
2089                 memcpy(cma_dst_addr(id_priv), raddr,
2090                        rdma_addr_size(raddr));
2091                 switch (iw_event->status) {
2092                 case 0:
2093                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2094                         event.param.conn.initiator_depth = iw_event->ird;
2095                         event.param.conn.responder_resources = iw_event->ord;
2096                         break;
2097                 case -ECONNRESET:
2098                 case -ECONNREFUSED:
2099                         event.event = RDMA_CM_EVENT_REJECTED;
2100                         break;
2101                 case -ETIMEDOUT:
2102                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2103                         break;
2104                 default:
2105                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2106                         break;
2107                 }
2108                 break;
2109         case IW_CM_EVENT_ESTABLISHED:
2110                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2111                 event.param.conn.initiator_depth = iw_event->ird;
2112                 event.param.conn.responder_resources = iw_event->ord;
2113                 break;
2114         default:
2115                 BUG_ON(1);
2116         }
2117
2118         event.status = iw_event->status;
2119         event.param.conn.private_data = iw_event->private_data;
2120         event.param.conn.private_data_len = iw_event->private_data_len;
2121         ret = id_priv->id.event_handler(&id_priv->id, &event);
2122         if (ret) {
2123                 /* Destroy the CM ID by returning a non-zero value. */
2124                 id_priv->cm_id.iw = NULL;
2125                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2126                 mutex_unlock(&id_priv->handler_mutex);
2127                 rdma_destroy_id(&id_priv->id);
2128                 return ret;
2129         }
2130
2131 out:
2132         mutex_unlock(&id_priv->handler_mutex);
2133         return ret;
2134 }
2135
2136 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2137                                struct iw_cm_event *iw_event)
2138 {
2139         struct rdma_cm_id *new_cm_id;
2140         struct rdma_id_private *listen_id, *conn_id;
2141         struct rdma_cm_event event;
2142         int ret = -ECONNABORTED;
2143         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2144         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2145
2146         listen_id = cm_id->context;
2147
2148         mutex_lock(&listen_id->handler_mutex);
2149         if (listen_id->state != RDMA_CM_LISTEN)
2150                 goto out;
2151
2152         /* Create a new RDMA id for the new IW CM ID */
2153         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2154                                    listen_id->id.event_handler,
2155                                    listen_id->id.context,
2156                                    RDMA_PS_TCP, IB_QPT_RC);
2157         if (IS_ERR(new_cm_id)) {
2158                 ret = -ENOMEM;
2159                 goto out;
2160         }
2161         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2162         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2163         conn_id->state = RDMA_CM_CONNECT;
2164
2165         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2166         if (ret) {
2167                 mutex_unlock(&conn_id->handler_mutex);
2168                 rdma_destroy_id(new_cm_id);
2169                 goto out;
2170         }
2171
2172         ret = cma_acquire_dev(conn_id, listen_id);
2173         if (ret) {
2174                 mutex_unlock(&conn_id->handler_mutex);
2175                 rdma_destroy_id(new_cm_id);
2176                 goto out;
2177         }
2178
2179         conn_id->cm_id.iw = cm_id;
2180         cm_id->context = conn_id;
2181         cm_id->cm_handler = cma_iw_handler;
2182
2183         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2184         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2185
2186         memset(&event, 0, sizeof event);
2187         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2188         event.param.conn.private_data = iw_event->private_data;
2189         event.param.conn.private_data_len = iw_event->private_data_len;
2190         event.param.conn.initiator_depth = iw_event->ird;
2191         event.param.conn.responder_resources = iw_event->ord;
2192
2193         /*
2194          * Protect against the user destroying conn_id from another thread
2195          * until we're done accessing it.
2196          */
2197         atomic_inc(&conn_id->refcount);
2198         ret = conn_id->id.event_handler(&conn_id->id, &event);
2199         if (ret) {
2200                 /* User wants to destroy the CM ID */
2201                 conn_id->cm_id.iw = NULL;
2202                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2203                 mutex_unlock(&conn_id->handler_mutex);
2204                 cma_deref_id(conn_id);
2205                 rdma_destroy_id(&conn_id->id);
2206                 goto out;
2207         }
2208
2209         mutex_unlock(&conn_id->handler_mutex);
2210         cma_deref_id(conn_id);
2211
2212 out:
2213         mutex_unlock(&listen_id->handler_mutex);
2214         return ret;
2215 }
2216
2217 static int cma_ib_listen(struct rdma_id_private *id_priv)
2218 {
2219         struct sockaddr *addr;
2220         struct ib_cm_id *id;
2221         __be64 svc_id;
2222
2223         addr = cma_src_addr(id_priv);
2224         svc_id = rdma_get_service_id(&id_priv->id, addr);
2225         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2226         if (IS_ERR(id))
2227                 return PTR_ERR(id);
2228         id_priv->cm_id.ib = id;
2229
2230         return 0;
2231 }
2232
2233 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2234 {
2235         int ret;
2236         struct iw_cm_id *id;
2237
2238         id = iw_create_cm_id(id_priv->id.device,
2239                              iw_conn_req_handler,
2240                              id_priv);
2241         if (IS_ERR(id))
2242                 return PTR_ERR(id);
2243
2244         id->tos = id_priv->tos;
2245         id_priv->cm_id.iw = id;
2246
2247         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2248                rdma_addr_size(cma_src_addr(id_priv)));
2249
2250         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2251
2252         if (ret) {
2253                 iw_destroy_cm_id(id_priv->cm_id.iw);
2254                 id_priv->cm_id.iw = NULL;
2255         }
2256
2257         return ret;
2258 }
2259
2260 static int cma_listen_handler(struct rdma_cm_id *id,
2261                               struct rdma_cm_event *event)
2262 {
2263         struct rdma_id_private *id_priv = id->context;
2264
2265         id->context = id_priv->id.context;
2266         id->event_handler = id_priv->id.event_handler;
2267         return id_priv->id.event_handler(id, event);
2268 }
2269
2270 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2271                               struct cma_device *cma_dev)
2272 {
2273         struct rdma_id_private *dev_id_priv;
2274         struct rdma_cm_id *id;
2275         struct net *net = id_priv->id.route.addr.dev_addr.net;
2276         int ret;
2277
2278         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2279                 return;
2280
2281         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2282                             id_priv->id.qp_type);
2283         if (IS_ERR(id))
2284                 return;
2285
2286         dev_id_priv = container_of(id, struct rdma_id_private, id);
2287
2288         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2289         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2290                rdma_addr_size(cma_src_addr(id_priv)));
2291
2292         _cma_attach_to_dev(dev_id_priv, cma_dev);
2293         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2294         atomic_inc(&id_priv->refcount);
2295         dev_id_priv->internal_id = 1;
2296         dev_id_priv->afonly = id_priv->afonly;
2297
2298         ret = rdma_listen(id, id_priv->backlog);
2299         if (ret)
2300                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2301                         ret, cma_dev->device->name);
2302 }
2303
2304 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2305 {
2306         struct cma_device *cma_dev;
2307
2308         mutex_lock(&lock);
2309         list_add_tail(&id_priv->list, &listen_any_list);
2310         list_for_each_entry(cma_dev, &dev_list, list)
2311                 cma_listen_on_dev(id_priv, cma_dev);
2312         mutex_unlock(&lock);
2313 }
2314
2315 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2316 {
2317         struct rdma_id_private *id_priv;
2318
2319         id_priv = container_of(id, struct rdma_id_private, id);
2320         id_priv->tos = (u8) tos;
2321         id_priv->tos_set = true;
2322 }
2323 EXPORT_SYMBOL(rdma_set_service_type);
2324
2325 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2326                               void *context)
2327 {
2328         struct cma_work *work = context;
2329         struct rdma_route *route;
2330
2331         route = &work->id->id.route;
2332
2333         if (!status) {
2334                 route->num_paths = 1;
2335                 *route->path_rec = *path_rec;
2336         } else {
2337                 work->old_state = RDMA_CM_ROUTE_QUERY;
2338                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2339                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2340                 work->event.status = status;
2341                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2342                                      status);
2343         }
2344
2345         queue_work(cma_wq, &work->work);
2346 }
2347
2348 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2349                               struct cma_work *work)
2350 {
2351         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2352         struct sa_path_rec path_rec;
2353         ib_sa_comp_mask comp_mask;
2354         struct sockaddr_in6 *sin6;
2355         struct sockaddr_ib *sib;
2356
2357         memset(&path_rec, 0, sizeof path_rec);
2358
2359         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2360                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2361         else
2362                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2363         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2364         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2365         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2366         path_rec.numb_path = 1;
2367         path_rec.reversible = 1;
2368         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2369                                                   cma_dst_addr(id_priv));
2370
2371         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2372                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2373                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2374
2375         switch (cma_family(id_priv)) {
2376         case AF_INET:
2377                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2378                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2379                 break;
2380         case AF_INET6:
2381                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2382                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2383                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2384                 break;
2385         case AF_IB:
2386                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2387                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2388                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2389                 break;
2390         }
2391
2392         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2393                                                id_priv->id.port_num, &path_rec,
2394                                                comp_mask, timeout_ms,
2395                                                GFP_KERNEL, cma_query_handler,
2396                                                work, &id_priv->query);
2397
2398         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2399 }
2400
2401 static void cma_work_handler(struct work_struct *_work)
2402 {
2403         struct cma_work *work = container_of(_work, struct cma_work, work);
2404         struct rdma_id_private *id_priv = work->id;
2405         int destroy = 0;
2406
2407         mutex_lock(&id_priv->handler_mutex);
2408         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2409                 goto out;
2410
2411         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2412                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2413                 destroy = 1;
2414         }
2415 out:
2416         mutex_unlock(&id_priv->handler_mutex);
2417         cma_deref_id(id_priv);
2418         if (destroy)
2419                 rdma_destroy_id(&id_priv->id);
2420         kfree(work);
2421 }
2422
2423 static void cma_ndev_work_handler(struct work_struct *_work)
2424 {
2425         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2426         struct rdma_id_private *id_priv = work->id;
2427         int destroy = 0;
2428
2429         mutex_lock(&id_priv->handler_mutex);
2430         if (id_priv->state == RDMA_CM_DESTROYING ||
2431             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2432                 goto out;
2433
2434         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2435                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2436                 destroy = 1;
2437         }
2438
2439 out:
2440         mutex_unlock(&id_priv->handler_mutex);
2441         cma_deref_id(id_priv);
2442         if (destroy)
2443                 rdma_destroy_id(&id_priv->id);
2444         kfree(work);
2445 }
2446
2447 static void cma_init_resolve_route_work(struct cma_work *work,
2448                                         struct rdma_id_private *id_priv)
2449 {
2450         work->id = id_priv;
2451         INIT_WORK(&work->work, cma_work_handler);
2452         work->old_state = RDMA_CM_ROUTE_QUERY;
2453         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2454         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2455 }
2456
2457 static void cma_init_resolve_addr_work(struct cma_work *work,
2458                                        struct rdma_id_private *id_priv)
2459 {
2460         work->id = id_priv;
2461         INIT_WORK(&work->work, cma_work_handler);
2462         work->old_state = RDMA_CM_ADDR_QUERY;
2463         work->new_state = RDMA_CM_ADDR_RESOLVED;
2464         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2465 }
2466
2467 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2468 {
2469         struct rdma_route *route = &id_priv->id.route;
2470         struct cma_work *work;
2471         int ret;
2472
2473         work = kzalloc(sizeof *work, GFP_KERNEL);
2474         if (!work)
2475                 return -ENOMEM;
2476
2477         cma_init_resolve_route_work(work, id_priv);
2478
2479         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2480         if (!route->path_rec) {
2481                 ret = -ENOMEM;
2482                 goto err1;
2483         }
2484
2485         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2486         if (ret)
2487                 goto err2;
2488
2489         return 0;
2490 err2:
2491         kfree(route->path_rec);
2492         route->path_rec = NULL;
2493 err1:
2494         kfree(work);
2495         return ret;
2496 }
2497
2498 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2499                                            unsigned long supported_gids,
2500                                            enum ib_gid_type default_gid)
2501 {
2502         if ((network_type == RDMA_NETWORK_IPV4 ||
2503              network_type == RDMA_NETWORK_IPV6) &&
2504             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2505                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2506
2507         return default_gid;
2508 }
2509
2510 /*
2511  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2512  * path record type based on GID type.
2513  * It also sets up other L2 fields which includes destination mac address
2514  * netdev ifindex, of the path record.
2515  * It returns the netdev of the bound interface for this path record entry.
2516  */
2517 static struct net_device *
2518 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2519 {
2520         struct rdma_route *route = &id_priv->id.route;
2521         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2522         struct rdma_addr *addr = &route->addr;
2523         unsigned long supported_gids;
2524         struct net_device *ndev;
2525
2526         if (!addr->dev_addr.bound_dev_if)
2527                 return NULL;
2528
2529         ndev = dev_get_by_index(addr->dev_addr.net,
2530                                 addr->dev_addr.bound_dev_if);
2531         if (!ndev)
2532                 return NULL;
2533
2534         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2535                                                     id_priv->id.port_num);
2536         gid_type = cma_route_gid_type(addr->dev_addr.network,
2537                                       supported_gids,
2538                                       id_priv->gid_type);
2539         /* Use the hint from IP Stack to select GID Type */
2540         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2541                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2542         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2543
2544         sa_path_set_ndev(route->path_rec, addr->dev_addr.net);
2545         sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2546         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2547         return ndev;
2548 }
2549
2550 int rdma_set_ib_path(struct rdma_cm_id *id,
2551                      struct sa_path_rec *path_rec)
2552 {
2553         struct rdma_id_private *id_priv;
2554         struct net_device *ndev;
2555         int ret;
2556
2557         id_priv = container_of(id, struct rdma_id_private, id);
2558         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2559                            RDMA_CM_ROUTE_RESOLVED))
2560                 return -EINVAL;
2561
2562         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2563                                      GFP_KERNEL);
2564         if (!id->route.path_rec) {
2565                 ret = -ENOMEM;
2566                 goto err;
2567         }
2568
2569         if (rdma_protocol_roce(id->device, id->port_num)) {
2570                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2571                 if (!ndev) {
2572                         ret = -ENODEV;
2573                         goto err_free;
2574                 }
2575                 dev_put(ndev);
2576         }
2577
2578         id->route.num_paths = 1;
2579         return 0;
2580
2581 err_free:
2582         kfree(id->route.path_rec);
2583         id->route.path_rec = NULL;
2584 err:
2585         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2586         return ret;
2587 }
2588 EXPORT_SYMBOL(rdma_set_ib_path);
2589
2590 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2591 {
2592         struct cma_work *work;
2593
2594         work = kzalloc(sizeof *work, GFP_KERNEL);
2595         if (!work)
2596                 return -ENOMEM;
2597
2598         cma_init_resolve_route_work(work, id_priv);
2599         queue_work(cma_wq, &work->work);
2600         return 0;
2601 }
2602
2603 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2604 {
2605         int prio;
2606         struct net_device *dev;
2607
2608         prio = rt_tos2priority(tos);
2609         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2610         if (dev->num_tc)
2611                 return netdev_get_prio_tc_map(dev, prio);
2612
2613 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2614         if (is_vlan_dev(ndev))
2615                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2616                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2617 #endif
2618         return 0;
2619 }
2620
2621 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2622 {
2623         struct rdma_route *route = &id_priv->id.route;
2624         struct rdma_addr *addr = &route->addr;
2625         struct cma_work *work;
2626         int ret;
2627         struct net_device *ndev;
2628
2629         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2630                                         rdma_start_port(id_priv->cma_dev->device)];
2631         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2632
2633
2634         work = kzalloc(sizeof *work, GFP_KERNEL);
2635         if (!work)
2636                 return -ENOMEM;
2637
2638         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2639         if (!route->path_rec) {
2640                 ret = -ENOMEM;
2641                 goto err1;
2642         }
2643
2644         route->num_paths = 1;
2645
2646         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2647         if (!ndev) {
2648                 ret = -ENODEV;
2649                 goto err2;
2650         }
2651
2652         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2653                     &route->path_rec->sgid);
2654         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2655                     &route->path_rec->dgid);
2656
2657         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2658                 /* TODO: get the hoplimit from the inet/inet6 device */
2659                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2660         else
2661                 route->path_rec->hop_limit = 1;
2662         route->path_rec->reversible = 1;
2663         route->path_rec->pkey = cpu_to_be16(0xffff);
2664         route->path_rec->mtu_selector = IB_SA_EQ;
2665         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2666         route->path_rec->traffic_class = tos;
2667         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2668         route->path_rec->rate_selector = IB_SA_EQ;
2669         route->path_rec->rate = iboe_get_rate(ndev);
2670         dev_put(ndev);
2671         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2672         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2673         if (!route->path_rec->mtu) {
2674                 ret = -EINVAL;
2675                 goto err2;
2676         }
2677
2678         cma_init_resolve_route_work(work, id_priv);
2679         queue_work(cma_wq, &work->work);
2680
2681         return 0;
2682
2683 err2:
2684         kfree(route->path_rec);
2685         route->path_rec = NULL;
2686 err1:
2687         kfree(work);
2688         return ret;
2689 }
2690
2691 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2692 {
2693         struct rdma_id_private *id_priv;
2694         int ret;
2695
2696         id_priv = container_of(id, struct rdma_id_private, id);
2697         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2698                 return -EINVAL;
2699
2700         atomic_inc(&id_priv->refcount);
2701         if (rdma_cap_ib_sa(id->device, id->port_num))
2702                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2703         else if (rdma_protocol_roce(id->device, id->port_num))
2704                 ret = cma_resolve_iboe_route(id_priv);
2705         else if (rdma_protocol_iwarp(id->device, id->port_num))
2706                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2707         else
2708                 ret = -ENOSYS;
2709
2710         if (ret)
2711                 goto err;
2712
2713         return 0;
2714 err:
2715         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2716         cma_deref_id(id_priv);
2717         return ret;
2718 }
2719 EXPORT_SYMBOL(rdma_resolve_route);
2720
2721 static void cma_set_loopback(struct sockaddr *addr)
2722 {
2723         switch (addr->sa_family) {
2724         case AF_INET:
2725                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2726                 break;
2727         case AF_INET6:
2728                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2729                               0, 0, 0, htonl(1));
2730                 break;
2731         default:
2732                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2733                             0, 0, 0, htonl(1));
2734                 break;
2735         }
2736 }
2737
2738 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2739 {
2740         struct cma_device *cma_dev, *cur_dev;
2741         union ib_gid gid;
2742         enum ib_port_state port_state;
2743         u16 pkey;
2744         int ret;
2745         u8 p;
2746
2747         cma_dev = NULL;
2748         mutex_lock(&lock);
2749         list_for_each_entry(cur_dev, &dev_list, list) {
2750                 if (cma_family(id_priv) == AF_IB &&
2751                     !rdma_cap_ib_cm(cur_dev->device, 1))
2752                         continue;
2753
2754                 if (!cma_dev)
2755                         cma_dev = cur_dev;
2756
2757                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2758                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2759                             port_state == IB_PORT_ACTIVE) {
2760                                 cma_dev = cur_dev;
2761                                 goto port_found;
2762                         }
2763                 }
2764         }
2765
2766         if (!cma_dev) {
2767                 ret = -ENODEV;
2768                 goto out;
2769         }
2770
2771         p = 1;
2772
2773 port_found:
2774         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2775         if (ret)
2776                 goto out;
2777
2778         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2779         if (ret)
2780                 goto out;
2781
2782         id_priv->id.route.addr.dev_addr.dev_type =
2783                 (rdma_protocol_ib(cma_dev->device, p)) ?
2784                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2785
2786         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2787         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2788         id_priv->id.port_num = p;
2789         cma_attach_to_dev(id_priv, cma_dev);
2790         cma_set_loopback(cma_src_addr(id_priv));
2791 out:
2792         mutex_unlock(&lock);
2793         return ret;
2794 }
2795
2796 static void addr_handler(int status, struct sockaddr *src_addr,
2797                          struct rdma_dev_addr *dev_addr, void *context)
2798 {
2799         struct rdma_id_private *id_priv = context;
2800         struct rdma_cm_event event;
2801
2802         memset(&event, 0, sizeof event);
2803         mutex_lock(&id_priv->handler_mutex);
2804         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2805                            RDMA_CM_ADDR_RESOLVED))
2806                 goto out;
2807
2808         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2809         if (!status && !id_priv->cma_dev) {
2810                 status = cma_acquire_dev(id_priv, NULL);
2811                 if (status)
2812                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2813                                              status);
2814         } else {
2815                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2816         }
2817
2818         if (status) {
2819                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2820                                    RDMA_CM_ADDR_BOUND))
2821                         goto out;
2822                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2823                 event.status = status;
2824         } else
2825                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2826
2827         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2828                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2829                 mutex_unlock(&id_priv->handler_mutex);
2830                 cma_deref_id(id_priv);
2831                 rdma_destroy_id(&id_priv->id);
2832                 return;
2833         }
2834 out:
2835         mutex_unlock(&id_priv->handler_mutex);
2836         cma_deref_id(id_priv);
2837 }
2838
2839 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2840 {
2841         struct cma_work *work;
2842         union ib_gid gid;
2843         int ret;
2844
2845         work = kzalloc(sizeof *work, GFP_KERNEL);
2846         if (!work)
2847                 return -ENOMEM;
2848
2849         if (!id_priv->cma_dev) {
2850                 ret = cma_bind_loopback(id_priv);
2851                 if (ret)
2852                         goto err;
2853         }
2854
2855         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2856         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2857
2858         cma_init_resolve_addr_work(work, id_priv);
2859         queue_work(cma_wq, &work->work);
2860         return 0;
2861 err:
2862         kfree(work);
2863         return ret;
2864 }
2865
2866 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2867 {
2868         struct cma_work *work;
2869         int ret;
2870
2871         work = kzalloc(sizeof *work, GFP_KERNEL);
2872         if (!work)
2873                 return -ENOMEM;
2874
2875         if (!id_priv->cma_dev) {
2876                 ret = cma_resolve_ib_dev(id_priv);
2877                 if (ret)
2878                         goto err;
2879         }
2880
2881         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2882                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2883
2884         cma_init_resolve_addr_work(work, id_priv);
2885         queue_work(cma_wq, &work->work);
2886         return 0;
2887 err:
2888         kfree(work);
2889         return ret;
2890 }
2891
2892 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2893                          struct sockaddr *dst_addr)
2894 {
2895         if (!src_addr || !src_addr->sa_family) {
2896                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2897                 src_addr->sa_family = dst_addr->sa_family;
2898                 if (IS_ENABLED(CONFIG_IPV6) &&
2899                     dst_addr->sa_family == AF_INET6) {
2900                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2901                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2902                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2903                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2904                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2905                 } else if (dst_addr->sa_family == AF_IB) {
2906                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2907                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2908                 }
2909         }
2910         return rdma_bind_addr(id, src_addr);
2911 }
2912
2913 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2914                       struct sockaddr *dst_addr, int timeout_ms)
2915 {
2916         struct rdma_id_private *id_priv;
2917         int ret;
2918
2919         id_priv = container_of(id, struct rdma_id_private, id);
2920         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2921         if (id_priv->state == RDMA_CM_IDLE) {
2922                 ret = cma_bind_addr(id, src_addr, dst_addr);
2923                 if (ret) {
2924                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2925                         return ret;
2926                 }
2927         }
2928
2929         if (cma_family(id_priv) != dst_addr->sa_family) {
2930                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2931                 return -EINVAL;
2932         }
2933
2934         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2935                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2936                 return -EINVAL;
2937         }
2938
2939         atomic_inc(&id_priv->refcount);
2940         if (cma_any_addr(dst_addr)) {
2941                 ret = cma_resolve_loopback(id_priv);
2942         } else {
2943                 if (dst_addr->sa_family == AF_IB) {
2944                         ret = cma_resolve_ib_addr(id_priv);
2945                 } else {
2946                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2947                                               dst_addr, &id->route.addr.dev_addr,
2948                                               timeout_ms, addr_handler, id_priv);
2949                 }
2950         }
2951         if (ret)
2952                 goto err;
2953
2954         return 0;
2955 err:
2956         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2957         cma_deref_id(id_priv);
2958         return ret;
2959 }
2960 EXPORT_SYMBOL(rdma_resolve_addr);
2961
2962 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2963 {
2964         struct rdma_id_private *id_priv;
2965         unsigned long flags;
2966         int ret;
2967
2968         id_priv = container_of(id, struct rdma_id_private, id);
2969         spin_lock_irqsave(&id_priv->lock, flags);
2970         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2971                 id_priv->reuseaddr = reuse;
2972                 ret = 0;
2973         } else {
2974                 ret = -EINVAL;
2975         }
2976         spin_unlock_irqrestore(&id_priv->lock, flags);
2977         return ret;
2978 }
2979 EXPORT_SYMBOL(rdma_set_reuseaddr);
2980
2981 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2982 {
2983         struct rdma_id_private *id_priv;
2984         unsigned long flags;
2985         int ret;
2986
2987         id_priv = container_of(id, struct rdma_id_private, id);
2988         spin_lock_irqsave(&id_priv->lock, flags);
2989         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2990                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2991                 id_priv->afonly = afonly;
2992                 ret = 0;
2993         } else {
2994                 ret = -EINVAL;
2995         }
2996         spin_unlock_irqrestore(&id_priv->lock, flags);
2997         return ret;
2998 }
2999 EXPORT_SYMBOL(rdma_set_afonly);
3000
3001 static void cma_bind_port(struct rdma_bind_list *bind_list,
3002                           struct rdma_id_private *id_priv)
3003 {
3004         struct sockaddr *addr;
3005         struct sockaddr_ib *sib;
3006         u64 sid, mask;
3007         __be16 port;
3008
3009         addr = cma_src_addr(id_priv);
3010         port = htons(bind_list->port);
3011
3012         switch (addr->sa_family) {
3013         case AF_INET:
3014                 ((struct sockaddr_in *) addr)->sin_port = port;
3015                 break;
3016         case AF_INET6:
3017                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3018                 break;
3019         case AF_IB:
3020                 sib = (struct sockaddr_ib *) addr;
3021                 sid = be64_to_cpu(sib->sib_sid);
3022                 mask = be64_to_cpu(sib->sib_sid_mask);
3023                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3024                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3025                 break;
3026         }
3027         id_priv->bind_list = bind_list;
3028         hlist_add_head(&id_priv->node, &bind_list->owners);
3029 }
3030
3031 static int cma_alloc_port(enum rdma_port_space ps,
3032                           struct rdma_id_private *id_priv, unsigned short snum)
3033 {
3034         struct rdma_bind_list *bind_list;
3035         int ret;
3036
3037         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3038         if (!bind_list)
3039                 return -ENOMEM;
3040
3041         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3042                            snum);
3043         if (ret < 0)
3044                 goto err;
3045
3046         bind_list->ps = ps;
3047         bind_list->port = (unsigned short)ret;
3048         cma_bind_port(bind_list, id_priv);
3049         return 0;
3050 err:
3051         kfree(bind_list);
3052         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3053 }
3054
3055 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3056                               struct rdma_id_private *id_priv)
3057 {
3058         struct rdma_id_private *cur_id;
3059         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3060         struct sockaddr  *saddr = cma_src_addr(id_priv);
3061         __be16 dport = cma_port(daddr);
3062
3063         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3064                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3065                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3066                 __be16 cur_dport = cma_port(cur_daddr);
3067
3068                 if (id_priv == cur_id)
3069                         continue;
3070
3071                 /* different dest port -> unique */
3072                 if (!cma_any_port(daddr) &&
3073                     !cma_any_port(cur_daddr) &&
3074                     (dport != cur_dport))
3075                         continue;
3076
3077                 /* different src address -> unique */
3078                 if (!cma_any_addr(saddr) &&
3079                     !cma_any_addr(cur_saddr) &&
3080                     cma_addr_cmp(saddr, cur_saddr))
3081                         continue;
3082
3083                 /* different dst address -> unique */
3084                 if (!cma_any_addr(daddr) &&
3085                     !cma_any_addr(cur_daddr) &&
3086                     cma_addr_cmp(daddr, cur_daddr))
3087                         continue;
3088
3089                 return -EADDRNOTAVAIL;
3090         }
3091         return 0;
3092 }
3093
3094 static int cma_alloc_any_port(enum rdma_port_space ps,
3095                               struct rdma_id_private *id_priv)
3096 {
3097         static unsigned int last_used_port;
3098         int low, high, remaining;
3099         unsigned int rover;
3100         struct net *net = id_priv->id.route.addr.dev_addr.net;
3101
3102         inet_get_local_port_range(net, &low, &high);
3103         remaining = (high - low) + 1;
3104         rover = prandom_u32() % remaining + low;
3105 retry:
3106         if (last_used_port != rover) {
3107                 struct rdma_bind_list *bind_list;
3108                 int ret;
3109
3110                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3111
3112                 if (!bind_list) {
3113                         ret = cma_alloc_port(ps, id_priv, rover);
3114                 } else {
3115                         ret = cma_port_is_unique(bind_list, id_priv);
3116                         if (!ret)
3117                                 cma_bind_port(bind_list, id_priv);
3118                 }
3119                 /*
3120                  * Remember previously used port number in order to avoid
3121                  * re-using same port immediately after it is closed.
3122                  */
3123                 if (!ret)
3124                         last_used_port = rover;
3125                 if (ret != -EADDRNOTAVAIL)
3126                         return ret;
3127         }
3128         if (--remaining) {
3129                 rover++;
3130                 if ((rover < low) || (rover > high))
3131                         rover = low;
3132                 goto retry;
3133         }
3134         return -EADDRNOTAVAIL;
3135 }
3136
3137 /*
3138  * Check that the requested port is available.  This is called when trying to
3139  * bind to a specific port, or when trying to listen on a bound port.  In
3140  * the latter case, the provided id_priv may already be on the bind_list, but
3141  * we still need to check that it's okay to start listening.
3142  */
3143 static int cma_check_port(struct rdma_bind_list *bind_list,
3144                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3145 {
3146         struct rdma_id_private *cur_id;
3147         struct sockaddr *addr, *cur_addr;
3148
3149         addr = cma_src_addr(id_priv);
3150         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3151                 if (id_priv == cur_id)
3152                         continue;
3153
3154                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3155                     cur_id->reuseaddr)
3156                         continue;
3157
3158                 cur_addr = cma_src_addr(cur_id);
3159                 if (id_priv->afonly && cur_id->afonly &&
3160                     (addr->sa_family != cur_addr->sa_family))
3161                         continue;
3162
3163                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3164                         return -EADDRNOTAVAIL;
3165
3166                 if (!cma_addr_cmp(addr, cur_addr))
3167                         return -EADDRINUSE;
3168         }
3169         return 0;
3170 }
3171
3172 static int cma_use_port(enum rdma_port_space ps,
3173                         struct rdma_id_private *id_priv)
3174 {
3175         struct rdma_bind_list *bind_list;
3176         unsigned short snum;
3177         int ret;
3178
3179         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3180         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3181                 return -EACCES;
3182
3183         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3184         if (!bind_list) {
3185                 ret = cma_alloc_port(ps, id_priv, snum);
3186         } else {
3187                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3188                 if (!ret)
3189                         cma_bind_port(bind_list, id_priv);
3190         }
3191         return ret;
3192 }
3193
3194 static int cma_bind_listen(struct rdma_id_private *id_priv)
3195 {
3196         struct rdma_bind_list *bind_list = id_priv->bind_list;
3197         int ret = 0;
3198
3199         mutex_lock(&lock);
3200         if (bind_list->owners.first->next)
3201                 ret = cma_check_port(bind_list, id_priv, 0);
3202         mutex_unlock(&lock);
3203         return ret;
3204 }
3205
3206 static enum rdma_port_space cma_select_inet_ps(
3207                 struct rdma_id_private *id_priv)
3208 {
3209         switch (id_priv->id.ps) {
3210         case RDMA_PS_TCP:
3211         case RDMA_PS_UDP:
3212         case RDMA_PS_IPOIB:
3213         case RDMA_PS_IB:
3214                 return id_priv->id.ps;
3215         default:
3216
3217                 return 0;
3218         }
3219 }
3220
3221 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3222 {
3223         enum rdma_port_space ps = 0;
3224         struct sockaddr_ib *sib;
3225         u64 sid_ps, mask, sid;
3226
3227         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3228         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3229         sid = be64_to_cpu(sib->sib_sid) & mask;
3230
3231         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3232                 sid_ps = RDMA_IB_IP_PS_IB;
3233                 ps = RDMA_PS_IB;
3234         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3235                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3236                 sid_ps = RDMA_IB_IP_PS_TCP;
3237                 ps = RDMA_PS_TCP;
3238         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3239                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3240                 sid_ps = RDMA_IB_IP_PS_UDP;
3241                 ps = RDMA_PS_UDP;
3242         }
3243
3244         if (ps) {
3245                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3246                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3247                                                 be64_to_cpu(sib->sib_sid_mask));
3248         }
3249         return ps;
3250 }
3251
3252 static int cma_get_port(struct rdma_id_private *id_priv)
3253 {
3254         enum rdma_port_space ps;
3255         int ret;
3256
3257         if (cma_family(id_priv) != AF_IB)
3258                 ps = cma_select_inet_ps(id_priv);
3259         else
3260                 ps = cma_select_ib_ps(id_priv);
3261         if (!ps)
3262                 return -EPROTONOSUPPORT;
3263
3264         mutex_lock(&lock);
3265         if (cma_any_port(cma_src_addr(id_priv)))
3266                 ret = cma_alloc_any_port(ps, id_priv);
3267         else
3268                 ret = cma_use_port(ps, id_priv);
3269         mutex_unlock(&lock);
3270
3271         return ret;
3272 }
3273
3274 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3275                                struct sockaddr *addr)
3276 {
3277 #if IS_ENABLED(CONFIG_IPV6)
3278         struct sockaddr_in6 *sin6;
3279
3280         if (addr->sa_family != AF_INET6)
3281                 return 0;
3282
3283         sin6 = (struct sockaddr_in6 *) addr;
3284
3285         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3286                 return 0;
3287
3288         if (!sin6->sin6_scope_id)
3289                         return -EINVAL;
3290
3291         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3292 #endif
3293         return 0;
3294 }
3295
3296 int rdma_listen(struct rdma_cm_id *id, int backlog)
3297 {
3298         struct rdma_id_private *id_priv;
3299         int ret;
3300
3301         id_priv = container_of(id, struct rdma_id_private, id);
3302         if (id_priv->state == RDMA_CM_IDLE) {
3303                 id->route.addr.src_addr.ss_family = AF_INET;
3304                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3305                 if (ret)
3306                         return ret;
3307         }
3308
3309         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3310                 return -EINVAL;
3311
3312         if (id_priv->reuseaddr) {
3313                 ret = cma_bind_listen(id_priv);
3314                 if (ret)
3315                         goto err;
3316         }
3317
3318         id_priv->backlog = backlog;
3319         if (id->device) {
3320                 if (rdma_cap_ib_cm(id->device, 1)) {
3321                         ret = cma_ib_listen(id_priv);
3322                         if (ret)
3323                                 goto err;
3324                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3325                         ret = cma_iw_listen(id_priv, backlog);
3326                         if (ret)
3327                                 goto err;
3328                 } else {
3329                         ret = -ENOSYS;
3330                         goto err;
3331                 }
3332         } else
3333                 cma_listen_on_all(id_priv);
3334
3335         return 0;
3336 err:
3337         id_priv->backlog = 0;
3338         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3339         return ret;
3340 }
3341 EXPORT_SYMBOL(rdma_listen);
3342
3343 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3344 {
3345         struct rdma_id_private *id_priv;
3346         int ret;
3347         struct sockaddr  *daddr;
3348
3349         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3350             addr->sa_family != AF_IB)
3351                 return -EAFNOSUPPORT;
3352
3353         id_priv = container_of(id, struct rdma_id_private, id);
3354         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3355                 return -EINVAL;
3356
3357         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3358         if (ret)
3359                 goto err1;
3360
3361         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3362         if (!cma_any_addr(addr)) {
3363                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3364                 if (ret)
3365                         goto err1;
3366
3367                 ret = cma_acquire_dev(id_priv, NULL);
3368                 if (ret)
3369                         goto err1;
3370         }
3371
3372         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3373                 if (addr->sa_family == AF_INET)
3374                         id_priv->afonly = 1;
3375 #if IS_ENABLED(CONFIG_IPV6)
3376                 else if (addr->sa_family == AF_INET6) {
3377                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3378
3379                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3380                 }
3381 #endif
3382         }
3383         daddr = cma_dst_addr(id_priv);
3384         daddr->sa_family = addr->sa_family;
3385
3386         ret = cma_get_port(id_priv);
3387         if (ret)
3388                 goto err2;
3389
3390         return 0;
3391 err2:
3392         if (id_priv->cma_dev)
3393                 cma_release_dev(id_priv);
3394 err1:
3395         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3396         return ret;
3397 }
3398 EXPORT_SYMBOL(rdma_bind_addr);
3399
3400 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3401 {
3402         struct cma_hdr *cma_hdr;
3403
3404         cma_hdr = hdr;
3405         cma_hdr->cma_version = CMA_VERSION;
3406         if (cma_family(id_priv) == AF_INET) {
3407                 struct sockaddr_in *src4, *dst4;
3408
3409                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3410                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3411
3412                 cma_set_ip_ver(cma_hdr, 4);
3413                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3414                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3415                 cma_hdr->port = src4->sin_port;
3416         } else if (cma_family(id_priv) == AF_INET6) {
3417                 struct sockaddr_in6 *src6, *dst6;
3418
3419                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3420                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3421
3422                 cma_set_ip_ver(cma_hdr, 6);
3423                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3424                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3425                 cma_hdr->port = src6->sin6_port;
3426         }
3427         return 0;
3428 }
3429
3430 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3431                                 struct ib_cm_event *ib_event)
3432 {
3433         struct rdma_id_private *id_priv = cm_id->context;
3434         struct rdma_cm_event event;
3435         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3436         int ret = 0;
3437
3438         mutex_lock(&id_priv->handler_mutex);
3439         if (id_priv->state != RDMA_CM_CONNECT)
3440                 goto out;
3441
3442         memset(&event, 0, sizeof event);
3443         switch (ib_event->event) {
3444         case IB_CM_SIDR_REQ_ERROR:
3445                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3446                 event.status = -ETIMEDOUT;
3447                 break;
3448         case IB_CM_SIDR_REP_RECEIVED:
3449                 event.param.ud.private_data = ib_event->private_data;
3450                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3451                 if (rep->status != IB_SIDR_SUCCESS) {
3452                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3453                         event.status = ib_event->param.sidr_rep_rcvd.status;
3454                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3455                                              event.status);
3456                         break;
3457                 }
3458                 ret = cma_set_qkey(id_priv, rep->qkey);
3459                 if (ret) {
3460                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3461                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3462                         event.status = ret;
3463                         break;
3464                 }
3465                 ib_init_ah_attr_from_path(id_priv->id.device,
3466                                           id_priv->id.port_num,
3467                                           id_priv->id.route.path_rec,
3468                                           &event.param.ud.ah_attr);
3469                 event.param.ud.qp_num = rep->qpn;
3470                 event.param.ud.qkey = rep->qkey;
3471                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3472                 event.status = 0;
3473                 break;
3474         default:
3475                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3476                        ib_event->event);
3477                 goto out;
3478         }
3479
3480         ret = id_priv->id.event_handler(&id_priv->id, &event);
3481         if (ret) {
3482                 /* Destroy the CM ID by returning a non-zero value. */
3483                 id_priv->cm_id.ib = NULL;
3484                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3485                 mutex_unlock(&id_priv->handler_mutex);
3486                 rdma_destroy_id(&id_priv->id);
3487                 return ret;
3488         }
3489 out:
3490         mutex_unlock(&id_priv->handler_mutex);
3491         return ret;
3492 }
3493
3494 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3495                               struct rdma_conn_param *conn_param)
3496 {
3497         struct ib_cm_sidr_req_param req;
3498         struct ib_cm_id *id;
3499         void *private_data;
3500         u8 offset;
3501         int ret;
3502
3503         memset(&req, 0, sizeof req);
3504         offset = cma_user_data_offset(id_priv);
3505         req.private_data_len = offset + conn_param->private_data_len;
3506         if (req.private_data_len < conn_param->private_data_len)
3507                 return -EINVAL;
3508
3509         if (req.private_data_len) {
3510                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3511                 if (!private_data)
3512                         return -ENOMEM;
3513         } else {
3514                 private_data = NULL;
3515         }
3516
3517         if (conn_param->private_data && conn_param->private_data_len)
3518                 memcpy(private_data + offset, conn_param->private_data,
3519                        conn_param->private_data_len);
3520
3521         if (private_data) {
3522                 ret = cma_format_hdr(private_data, id_priv);
3523                 if (ret)
3524                         goto out;
3525                 req.private_data = private_data;
3526         }
3527
3528         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3529                              id_priv);
3530         if (IS_ERR(id)) {
3531                 ret = PTR_ERR(id);
3532                 goto out;
3533         }
3534         id_priv->cm_id.ib = id;
3535
3536         req.path = id_priv->id.route.path_rec;
3537         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3538         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3539         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3540
3541         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3542         if (ret) {
3543                 ib_destroy_cm_id(id_priv->cm_id.ib);
3544                 id_priv->cm_id.ib = NULL;
3545         }
3546 out:
3547         kfree(private_data);
3548         return ret;
3549 }
3550
3551 static int cma_connect_ib(struct rdma_id_private *id_priv,
3552                           struct rdma_conn_param *conn_param)
3553 {
3554         struct ib_cm_req_param req;
3555         struct rdma_route *route;
3556         void *private_data;
3557         struct ib_cm_id *id;
3558         u8 offset;
3559         int ret;
3560
3561         memset(&req, 0, sizeof req);
3562         offset = cma_user_data_offset(id_priv);
3563         req.private_data_len = offset + conn_param->private_data_len;
3564         if (req.private_data_len < conn_param->private_data_len)
3565                 return -EINVAL;
3566
3567         if (req.private_data_len) {
3568                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3569                 if (!private_data)
3570                         return -ENOMEM;
3571         } else {
3572                 private_data = NULL;
3573         }
3574
3575         if (conn_param->private_data && conn_param->private_data_len)
3576                 memcpy(private_data + offset, conn_param->private_data,
3577                        conn_param->private_data_len);
3578
3579         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3580         if (IS_ERR(id)) {
3581                 ret = PTR_ERR(id);
3582                 goto out;
3583         }
3584         id_priv->cm_id.ib = id;
3585
3586         route = &id_priv->id.route;
3587         if (private_data) {
3588                 ret = cma_format_hdr(private_data, id_priv);
3589                 if (ret)
3590                         goto out;
3591                 req.private_data = private_data;
3592         }
3593
3594         req.primary_path = &route->path_rec[0];
3595         if (route->num_paths == 2)
3596                 req.alternate_path = &route->path_rec[1];
3597
3598         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3599         req.qp_num = id_priv->qp_num;
3600         req.qp_type = id_priv->id.qp_type;
3601         req.starting_psn = id_priv->seq_num;
3602         req.responder_resources = conn_param->responder_resources;
3603         req.initiator_depth = conn_param->initiator_depth;
3604         req.flow_control = conn_param->flow_control;
3605         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3606         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3607         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3608         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3609         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3610         req.srq = id_priv->srq ? 1 : 0;
3611
3612         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3613 out:
3614         if (ret && !IS_ERR(id)) {
3615                 ib_destroy_cm_id(id);
3616                 id_priv->cm_id.ib = NULL;
3617         }
3618
3619         kfree(private_data);
3620         return ret;
3621 }
3622
3623 static int cma_connect_iw(struct rdma_id_private *id_priv,
3624                           struct rdma_conn_param *conn_param)
3625 {
3626         struct iw_cm_id *cm_id;
3627         int ret;
3628         struct iw_cm_conn_param iw_param;
3629
3630         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3631         if (IS_ERR(cm_id))
3632                 return PTR_ERR(cm_id);
3633
3634         cm_id->tos = id_priv->tos;
3635         id_priv->cm_id.iw = cm_id;
3636
3637         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3638                rdma_addr_size(cma_src_addr(id_priv)));
3639         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3640                rdma_addr_size(cma_dst_addr(id_priv)));
3641
3642         ret = cma_modify_qp_rtr(id_priv, conn_param);
3643         if (ret)
3644                 goto out;
3645
3646         if (conn_param) {
3647                 iw_param.ord = conn_param->initiator_depth;
3648                 iw_param.ird = conn_param->responder_resources;
3649                 iw_param.private_data = conn_param->private_data;
3650                 iw_param.private_data_len = conn_param->private_data_len;
3651                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3652         } else {
3653                 memset(&iw_param, 0, sizeof iw_param);
3654                 iw_param.qpn = id_priv->qp_num;
3655         }
3656         ret = iw_cm_connect(cm_id, &iw_param);
3657 out:
3658         if (ret) {
3659                 iw_destroy_cm_id(cm_id);
3660                 id_priv->cm_id.iw = NULL;
3661         }
3662         return ret;
3663 }
3664
3665 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3666 {
3667         struct rdma_id_private *id_priv;
3668         int ret;
3669
3670         id_priv = container_of(id, struct rdma_id_private, id);
3671         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3672                 return -EINVAL;
3673
3674         if (!id->qp) {
3675                 id_priv->qp_num = conn_param->qp_num;
3676                 id_priv->srq = conn_param->srq;
3677         }
3678
3679         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3680                 if (id->qp_type == IB_QPT_UD)
3681                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3682                 else
3683                         ret = cma_connect_ib(id_priv, conn_param);
3684         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3685                 ret = cma_connect_iw(id_priv, conn_param);
3686         else
3687                 ret = -ENOSYS;
3688         if (ret)
3689                 goto err;
3690
3691         return 0;
3692 err:
3693         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3694         return ret;
3695 }
3696 EXPORT_SYMBOL(rdma_connect);
3697
3698 static int cma_accept_ib(struct rdma_id_private *id_priv,
3699                          struct rdma_conn_param *conn_param)
3700 {
3701         struct ib_cm_rep_param rep;
3702         int ret;
3703
3704         ret = cma_modify_qp_rtr(id_priv, conn_param);
3705         if (ret)
3706                 goto out;
3707
3708         ret = cma_modify_qp_rts(id_priv, conn_param);
3709         if (ret)
3710                 goto out;
3711
3712         memset(&rep, 0, sizeof rep);
3713         rep.qp_num = id_priv->qp_num;
3714         rep.starting_psn = id_priv->seq_num;
3715         rep.private_data = conn_param->private_data;
3716         rep.private_data_len = conn_param->private_data_len;
3717         rep.responder_resources = conn_param->responder_resources;
3718         rep.initiator_depth = conn_param->initiator_depth;
3719         rep.failover_accepted = 0;
3720         rep.flow_control = conn_param->flow_control;
3721         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3722         rep.srq = id_priv->srq ? 1 : 0;
3723
3724         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3725 out:
3726         return ret;
3727 }
3728
3729 static int cma_accept_iw(struct rdma_id_private *id_priv,
3730                   struct rdma_conn_param *conn_param)
3731 {
3732         struct iw_cm_conn_param iw_param;
3733         int ret;
3734
3735         if (!conn_param)
3736                 return -EINVAL;
3737
3738         ret = cma_modify_qp_rtr(id_priv, conn_param);
3739         if (ret)
3740                 return ret;
3741
3742         iw_param.ord = conn_param->initiator_depth;
3743         iw_param.ird = conn_param->responder_resources;
3744         iw_param.private_data = conn_param->private_data;
3745         iw_param.private_data_len = conn_param->private_data_len;
3746         if (id_priv->id.qp) {
3747                 iw_param.qpn = id_priv->qp_num;
3748         } else
3749                 iw_param.qpn = conn_param->qp_num;
3750
3751         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3752 }
3753
3754 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3755                              enum ib_cm_sidr_status status, u32 qkey,
3756                              const void *private_data, int private_data_len)
3757 {
3758         struct ib_cm_sidr_rep_param rep;
3759         int ret;
3760
3761         memset(&rep, 0, sizeof rep);
3762         rep.status = status;
3763         if (status == IB_SIDR_SUCCESS) {
3764                 ret = cma_set_qkey(id_priv, qkey);
3765                 if (ret)
3766                         return ret;
3767                 rep.qp_num = id_priv->qp_num;
3768                 rep.qkey = id_priv->qkey;
3769         }
3770         rep.private_data = private_data;
3771         rep.private_data_len = private_data_len;
3772
3773         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3774 }
3775
3776 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3777 {
3778         struct rdma_id_private *id_priv;
3779         int ret;
3780
3781         id_priv = container_of(id, struct rdma_id_private, id);
3782
3783         id_priv->owner = task_pid_nr(current);
3784
3785         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3786                 return -EINVAL;
3787
3788         if (!id->qp && conn_param) {
3789                 id_priv->qp_num = conn_param->qp_num;
3790                 id_priv->srq = conn_param->srq;
3791         }
3792
3793         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3794                 if (id->qp_type == IB_QPT_UD) {
3795                         if (conn_param)
3796                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3797                                                         conn_param->qkey,
3798                                                         conn_param->private_data,
3799                                                         conn_param->private_data_len);
3800                         else
3801                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3802                                                         0, NULL, 0);
3803                 } else {
3804                         if (conn_param)
3805                                 ret = cma_accept_ib(id_priv, conn_param);
3806                         else
3807                                 ret = cma_rep_recv(id_priv);
3808                 }
3809         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3810                 ret = cma_accept_iw(id_priv, conn_param);
3811         else
3812                 ret = -ENOSYS;
3813
3814         if (ret)
3815                 goto reject;
3816
3817         return 0;
3818 reject:
3819         cma_modify_qp_err(id_priv);
3820         rdma_reject(id, NULL, 0);
3821         return ret;
3822 }
3823 EXPORT_SYMBOL(rdma_accept);
3824
3825 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3826 {
3827         struct rdma_id_private *id_priv;
3828         int ret;
3829
3830         id_priv = container_of(id, struct rdma_id_private, id);
3831         if (!id_priv->cm_id.ib)
3832                 return -EINVAL;
3833
3834         switch (id->device->node_type) {
3835         case RDMA_NODE_IB_CA:
3836                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3837                 break;
3838         default:
3839                 ret = 0;
3840                 break;
3841         }
3842         return ret;
3843 }
3844 EXPORT_SYMBOL(rdma_notify);
3845
3846 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3847                 u8 private_data_len)
3848 {
3849         struct rdma_id_private *id_priv;
3850         int ret;
3851
3852         id_priv = container_of(id, struct rdma_id_private, id);
3853         if (!id_priv->cm_id.ib)
3854                 return -EINVAL;
3855
3856         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3857                 if (id->qp_type == IB_QPT_UD)
3858                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3859                                                 private_data, private_data_len);
3860                 else
3861                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3862                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3863                                              0, private_data, private_data_len);
3864         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3865                 ret = iw_cm_reject(id_priv->cm_id.iw,
3866                                    private_data, private_data_len);
3867         } else
3868                 ret = -ENOSYS;
3869
3870         return ret;
3871 }
3872 EXPORT_SYMBOL(rdma_reject);
3873
3874 int rdma_disconnect(struct rdma_cm_id *id)
3875 {
3876         struct rdma_id_private *id_priv;
3877         int ret;
3878
3879         id_priv = container_of(id, struct rdma_id_private, id);
3880         if (!id_priv->cm_id.ib)
3881                 return -EINVAL;
3882
3883         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3884                 ret = cma_modify_qp_err(id_priv);
3885                 if (ret)
3886                         goto out;
3887                 /* Initiate or respond to a disconnect. */
3888                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3889                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3890         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3891                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3892         } else
3893                 ret = -EINVAL;
3894
3895 out:
3896         return ret;
3897 }
3898 EXPORT_SYMBOL(rdma_disconnect);
3899
3900 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3901 {
3902         struct rdma_id_private *id_priv;
3903         struct cma_multicast *mc = multicast->context;
3904         struct rdma_cm_event event;
3905         int ret = 0;
3906
3907         id_priv = mc->id_priv;
3908         mutex_lock(&id_priv->handler_mutex);
3909         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3910             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3911                 goto out;
3912
3913         if (!status)
3914                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3915         else
3916                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3917                                      status);
3918         mutex_lock(&id_priv->qp_mutex);
3919         if (!status && id_priv->id.qp) {
3920                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3921                                          be16_to_cpu(multicast->rec.mlid));
3922                 if (status)
3923                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3924                                              status);
3925         }
3926         mutex_unlock(&id_priv->qp_mutex);
3927
3928         memset(&event, 0, sizeof event);
3929         event.status = status;
3930         event.param.ud.private_data = mc->context;
3931         if (!status) {
3932                 struct rdma_dev_addr *dev_addr =
3933                         &id_priv->id.route.addr.dev_addr;
3934                 struct net_device *ndev =
3935                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
3936                 enum ib_gid_type gid_type =
3937                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3938                         rdma_start_port(id_priv->cma_dev->device)];
3939
3940                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3941                 ib_init_ah_from_mcmember(id_priv->id.device,
3942                                          id_priv->id.port_num, &multicast->rec,
3943                                          ndev, gid_type,
3944                                          &event.param.ud.ah_attr);
3945                 event.param.ud.qp_num = 0xFFFFFF;
3946                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3947                 if (ndev)
3948                         dev_put(ndev);
3949         } else
3950                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3951
3952         ret = id_priv->id.event_handler(&id_priv->id, &event);
3953         if (ret) {
3954                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3955                 mutex_unlock(&id_priv->handler_mutex);
3956                 rdma_destroy_id(&id_priv->id);
3957                 return 0;
3958         }
3959
3960 out:
3961         mutex_unlock(&id_priv->handler_mutex);
3962         return 0;
3963 }
3964
3965 static void cma_set_mgid(struct rdma_id_private *id_priv,
3966                          struct sockaddr *addr, union ib_gid *mgid)
3967 {
3968         unsigned char mc_map[MAX_ADDR_LEN];
3969         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3970         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3971         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3972
3973         if (cma_any_addr(addr)) {
3974                 memset(mgid, 0, sizeof *mgid);
3975         } else if ((addr->sa_family == AF_INET6) &&
3976                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3977                                                                  0xFF10A01B)) {
3978                 /* IPv6 address is an SA assigned MGID. */
3979                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3980         } else if (addr->sa_family == AF_IB) {
3981                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3982         } else if ((addr->sa_family == AF_INET6)) {
3983                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3984                 if (id_priv->id.ps == RDMA_PS_UDP)
3985                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3986                 *mgid = *(union ib_gid *) (mc_map + 4);
3987         } else {
3988                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3989                 if (id_priv->id.ps == RDMA_PS_UDP)
3990                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3991                 *mgid = *(union ib_gid *) (mc_map + 4);
3992         }
3993 }
3994
3995 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3996                                  struct cma_multicast *mc)
3997 {
3998         struct ib_sa_mcmember_rec rec;
3999         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4000         ib_sa_comp_mask comp_mask;
4001         int ret;
4002
4003         ib_addr_get_mgid(dev_addr, &rec.mgid);
4004         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4005                                      &rec.mgid, &rec);
4006         if (ret)
4007                 return ret;
4008
4009         ret = cma_set_qkey(id_priv, 0);
4010         if (ret)
4011                 return ret;
4012
4013         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4014         rec.qkey = cpu_to_be32(id_priv->qkey);
4015         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4016         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4017         rec.join_state = mc->join_state;
4018
4019         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4020             (!ib_sa_sendonly_fullmem_support(&sa_client,
4021                                              id_priv->id.device,
4022                                              id_priv->id.port_num))) {
4023                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4024                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4025                         id_priv->id.device->name, id_priv->id.port_num);
4026                 return -EOPNOTSUPP;
4027         }
4028
4029         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4030                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4031                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4032                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4033                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4034
4035         if (id_priv->id.ps == RDMA_PS_IPOIB)
4036                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4037                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4038                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4039                              IB_SA_MCMEMBER_REC_MTU |
4040                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4041
4042         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4043                                                 id_priv->id.port_num, &rec,
4044                                                 comp_mask, GFP_KERNEL,
4045                                                 cma_ib_mc_handler, mc);
4046         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4047 }
4048
4049 static void iboe_mcast_work_handler(struct work_struct *work)
4050 {
4051         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4052         struct cma_multicast *mc = mw->mc;
4053         struct ib_sa_multicast *m = mc->multicast.ib;
4054
4055         mc->multicast.ib->context = mc;
4056         cma_ib_mc_handler(0, m);
4057         kref_put(&mc->mcref, release_mc);
4058         kfree(mw);
4059 }
4060
4061 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4062                               enum ib_gid_type gid_type)
4063 {
4064         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4065         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4066
4067         if (cma_any_addr(addr)) {
4068                 memset(mgid, 0, sizeof *mgid);
4069         } else if (addr->sa_family == AF_INET6) {
4070                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4071         } else {
4072                 mgid->raw[0] =
4073                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4074                 mgid->raw[1] =
4075                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4076                 mgid->raw[2] = 0;
4077                 mgid->raw[3] = 0;
4078                 mgid->raw[4] = 0;
4079                 mgid->raw[5] = 0;
4080                 mgid->raw[6] = 0;
4081                 mgid->raw[7] = 0;
4082                 mgid->raw[8] = 0;
4083                 mgid->raw[9] = 0;
4084                 mgid->raw[10] = 0xff;
4085                 mgid->raw[11] = 0xff;
4086                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4087         }
4088 }
4089
4090 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4091                                    struct cma_multicast *mc)
4092 {
4093         struct iboe_mcast_work *work;
4094         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4095         int err = 0;
4096         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4097         struct net_device *ndev = NULL;
4098         enum ib_gid_type gid_type;
4099         bool send_only;
4100
4101         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4102
4103         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4104                 return -EINVAL;
4105
4106         work = kzalloc(sizeof *work, GFP_KERNEL);
4107         if (!work)
4108                 return -ENOMEM;
4109
4110         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4111         if (!mc->multicast.ib) {
4112                 err = -ENOMEM;
4113                 goto out1;
4114         }
4115
4116         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4117                    rdma_start_port(id_priv->cma_dev->device)];
4118         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4119
4120         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4121         if (id_priv->id.ps == RDMA_PS_UDP)
4122                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4123
4124         if (dev_addr->bound_dev_if)
4125                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4126         if (!ndev) {
4127                 err = -ENODEV;
4128                 goto out2;
4129         }
4130         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4131         mc->multicast.ib->rec.hop_limit = 1;
4132         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4133
4134         if (addr->sa_family == AF_INET) {
4135                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4136                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4137                         if (!send_only) {
4138                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4139                                                     true);
4140                                 if (!err)
4141                                         mc->igmp_joined = true;
4142                         }
4143                 }
4144         } else {
4145                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4146                         err = -ENOTSUPP;
4147         }
4148         dev_put(ndev);
4149         if (err || !mc->multicast.ib->rec.mtu) {
4150                 if (!err)
4151                         err = -EINVAL;
4152                 goto out2;
4153         }
4154         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4155                     &mc->multicast.ib->rec.port_gid);
4156         work->id = id_priv;
4157         work->mc = mc;
4158         INIT_WORK(&work->work, iboe_mcast_work_handler);
4159         kref_get(&mc->mcref);
4160         queue_work(cma_wq, &work->work);
4161
4162         return 0;
4163
4164 out2:
4165         kfree(mc->multicast.ib);
4166 out1:
4167         kfree(work);
4168         return err;
4169 }
4170
4171 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4172                         u8 join_state, void *context)
4173 {
4174         struct rdma_id_private *id_priv;
4175         struct cma_multicast *mc;
4176         int ret;
4177
4178         if (!id->device)
4179                 return -EINVAL;
4180
4181         id_priv = container_of(id, struct rdma_id_private, id);
4182         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4183             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4184                 return -EINVAL;
4185
4186         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4187         if (!mc)
4188                 return -ENOMEM;
4189
4190         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4191         mc->context = context;
4192         mc->id_priv = id_priv;
4193         mc->igmp_joined = false;
4194         mc->join_state = join_state;
4195         spin_lock(&id_priv->lock);
4196         list_add(&mc->list, &id_priv->mc_list);
4197         spin_unlock(&id_priv->lock);
4198
4199         if (rdma_protocol_roce(id->device, id->port_num)) {
4200                 kref_init(&mc->mcref);
4201                 ret = cma_iboe_join_multicast(id_priv, mc);
4202         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4203                 ret = cma_join_ib_multicast(id_priv, mc);
4204         else
4205                 ret = -ENOSYS;
4206
4207         if (ret) {
4208                 spin_lock_irq(&id_priv->lock);
4209                 list_del(&mc->list);
4210                 spin_unlock_irq(&id_priv->lock);
4211                 kfree(mc);
4212         }
4213         return ret;
4214 }
4215 EXPORT_SYMBOL(rdma_join_multicast);
4216
4217 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4218 {
4219         struct rdma_id_private *id_priv;
4220         struct cma_multicast *mc;
4221
4222         id_priv = container_of(id, struct rdma_id_private, id);
4223         spin_lock_irq(&id_priv->lock);
4224         list_for_each_entry(mc, &id_priv->mc_list, list) {
4225                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4226                         list_del(&mc->list);
4227                         spin_unlock_irq(&id_priv->lock);
4228
4229                         if (id->qp)
4230                                 ib_detach_mcast(id->qp,
4231                                                 &mc->multicast.ib->rec.mgid,
4232                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4233
4234                         BUG_ON(id_priv->cma_dev->device != id->device);
4235
4236                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4237                                 ib_sa_free_multicast(mc->multicast.ib);
4238                                 kfree(mc);
4239                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4240                                 if (mc->igmp_joined) {
4241                                         struct rdma_dev_addr *dev_addr =
4242                                                 &id->route.addr.dev_addr;
4243                                         struct net_device *ndev = NULL;
4244
4245                                         if (dev_addr->bound_dev_if)
4246                                                 ndev = dev_get_by_index(dev_addr->net,
4247                                                                         dev_addr->bound_dev_if);
4248                                         if (ndev) {
4249                                                 cma_igmp_send(ndev,
4250                                                               &mc->multicast.ib->rec.mgid,
4251                                                               false);
4252                                                 dev_put(ndev);
4253                                         }
4254                                         mc->igmp_joined = false;
4255                                 }
4256                                 kref_put(&mc->mcref, release_mc);
4257                         }
4258                         return;
4259                 }
4260         }
4261         spin_unlock_irq(&id_priv->lock);
4262 }
4263 EXPORT_SYMBOL(rdma_leave_multicast);
4264
4265 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4266 {
4267         struct rdma_dev_addr *dev_addr;
4268         struct cma_ndev_work *work;
4269
4270         dev_addr = &id_priv->id.route.addr.dev_addr;
4271
4272         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4273             (net_eq(dev_net(ndev), dev_addr->net)) &&
4274             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4275                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4276                         ndev->name, &id_priv->id);
4277                 work = kzalloc(sizeof *work, GFP_KERNEL);
4278                 if (!work)
4279                         return -ENOMEM;
4280
4281                 INIT_WORK(&work->work, cma_ndev_work_handler);
4282                 work->id = id_priv;
4283                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4284                 atomic_inc(&id_priv->refcount);
4285                 queue_work(cma_wq, &work->work);
4286         }
4287
4288         return 0;
4289 }
4290
4291 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4292                                void *ptr)
4293 {
4294         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4295         struct cma_device *cma_dev;
4296         struct rdma_id_private *id_priv;
4297         int ret = NOTIFY_DONE;
4298
4299         if (event != NETDEV_BONDING_FAILOVER)
4300                 return NOTIFY_DONE;
4301
4302         if (!netif_is_bond_master(ndev))
4303                 return NOTIFY_DONE;
4304
4305         mutex_lock(&lock);
4306         list_for_each_entry(cma_dev, &dev_list, list)
4307                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4308                         ret = cma_netdev_change(ndev, id_priv);
4309                         if (ret)
4310                                 goto out;
4311                 }
4312
4313 out:
4314         mutex_unlock(&lock);
4315         return ret;
4316 }
4317
4318 static struct notifier_block cma_nb = {
4319         .notifier_call = cma_netdev_callback
4320 };
4321
4322 static void cma_add_one(struct ib_device *device)
4323 {
4324         struct cma_device *cma_dev;
4325         struct rdma_id_private *id_priv;
4326         unsigned int i;
4327         unsigned long supported_gids = 0;
4328
4329         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4330         if (!cma_dev)
4331                 return;
4332
4333         cma_dev->device = device;
4334         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4335                                             sizeof(*cma_dev->default_gid_type),
4336                                             GFP_KERNEL);
4337         if (!cma_dev->default_gid_type)
4338                 goto free_cma_dev;
4339
4340         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4341                                             sizeof(*cma_dev->default_roce_tos),
4342                                             GFP_KERNEL);
4343         if (!cma_dev->default_roce_tos)
4344                 goto free_gid_type;
4345
4346         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4347                 supported_gids = roce_gid_type_mask_support(device, i);
4348                 WARN_ON(!supported_gids);
4349                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4350                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4351                                 CMA_PREFERRED_ROCE_GID_TYPE;
4352                 else
4353                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4354                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4355                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4356         }
4357
4358         init_completion(&cma_dev->comp);
4359         atomic_set(&cma_dev->refcount, 1);
4360         INIT_LIST_HEAD(&cma_dev->id_list);
4361         ib_set_client_data(device, &cma_client, cma_dev);
4362
4363         mutex_lock(&lock);
4364         list_add_tail(&cma_dev->list, &dev_list);
4365         list_for_each_entry(id_priv, &listen_any_list, list)
4366                 cma_listen_on_dev(id_priv, cma_dev);
4367         mutex_unlock(&lock);
4368
4369         return;
4370
4371 free_gid_type:
4372         kfree(cma_dev->default_gid_type);
4373
4374 free_cma_dev:
4375         kfree(cma_dev);
4376
4377         return;
4378 }
4379
4380 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4381 {
4382         struct rdma_cm_event event;
4383         enum rdma_cm_state state;
4384         int ret = 0;
4385
4386         /* Record that we want to remove the device */
4387         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4388         if (state == RDMA_CM_DESTROYING)
4389                 return 0;
4390
4391         cma_cancel_operation(id_priv, state);
4392         mutex_lock(&id_priv->handler_mutex);
4393
4394         /* Check for destruction from another callback. */
4395         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4396                 goto out;
4397
4398         memset(&event, 0, sizeof event);
4399         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4400         ret = id_priv->id.event_handler(&id_priv->id, &event);
4401 out:
4402         mutex_unlock(&id_priv->handler_mutex);
4403         return ret;
4404 }
4405
4406 static void cma_process_remove(struct cma_device *cma_dev)
4407 {
4408         struct rdma_id_private *id_priv;
4409         int ret;
4410
4411         mutex_lock(&lock);
4412         while (!list_empty(&cma_dev->id_list)) {
4413                 id_priv = list_entry(cma_dev->id_list.next,
4414                                      struct rdma_id_private, list);
4415
4416                 list_del(&id_priv->listen_list);
4417                 list_del_init(&id_priv->list);
4418                 atomic_inc(&id_priv->refcount);
4419                 mutex_unlock(&lock);
4420
4421                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4422                 cma_deref_id(id_priv);
4423                 if (ret)
4424                         rdma_destroy_id(&id_priv->id);
4425
4426                 mutex_lock(&lock);
4427         }
4428         mutex_unlock(&lock);
4429
4430         cma_deref_dev(cma_dev);
4431         wait_for_completion(&cma_dev->comp);
4432 }
4433
4434 static void cma_remove_one(struct ib_device *device, void *client_data)
4435 {
4436         struct cma_device *cma_dev = client_data;
4437
4438         if (!cma_dev)
4439                 return;
4440
4441         mutex_lock(&lock);
4442         list_del(&cma_dev->list);
4443         mutex_unlock(&lock);
4444
4445         cma_process_remove(cma_dev);
4446         kfree(cma_dev->default_roce_tos);
4447         kfree(cma_dev->default_gid_type);
4448         kfree(cma_dev);
4449 }
4450
4451 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4452 {
4453         struct nlmsghdr *nlh;
4454         struct rdma_cm_id_stats *id_stats;
4455         struct rdma_id_private *id_priv;
4456         struct rdma_cm_id *id = NULL;
4457         struct cma_device *cma_dev;
4458         int i_dev = 0, i_id = 0;
4459
4460         /*
4461          * We export all of the IDs as a sequence of messages.  Each
4462          * ID gets its own netlink message.
4463          */
4464         mutex_lock(&lock);
4465
4466         list_for_each_entry(cma_dev, &dev_list, list) {
4467                 if (i_dev < cb->args[0]) {
4468                         i_dev++;
4469                         continue;
4470                 }
4471
4472                 i_id = 0;
4473                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4474                         if (i_id < cb->args[1]) {
4475                                 i_id++;
4476                                 continue;
4477                         }
4478
4479                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4480                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4481                                                 RDMA_NL_RDMA_CM_ID_STATS,
4482                                                 NLM_F_MULTI);
4483                         if (!id_stats)
4484                                 goto out;
4485
4486                         memset(id_stats, 0, sizeof *id_stats);
4487                         id = &id_priv->id;
4488                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4489                         id_stats->port_num = id->port_num;
4490                         id_stats->bound_dev_if =
4491                                 id->route.addr.dev_addr.bound_dev_if;
4492
4493                         if (ibnl_put_attr(skb, nlh,
4494                                           rdma_addr_size(cma_src_addr(id_priv)),
4495                                           cma_src_addr(id_priv),
4496                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4497                                 goto out;
4498                         if (ibnl_put_attr(skb, nlh,
4499                                           rdma_addr_size(cma_dst_addr(id_priv)),
4500                                           cma_dst_addr(id_priv),
4501                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4502                                 goto out;
4503
4504                         id_stats->pid           = id_priv->owner;
4505                         id_stats->port_space    = id->ps;
4506                         id_stats->cm_state      = id_priv->state;
4507                         id_stats->qp_num        = id_priv->qp_num;
4508                         id_stats->qp_type       = id->qp_type;
4509
4510                         i_id++;
4511                         nlmsg_end(skb, nlh);
4512                 }
4513
4514                 cb->args[1] = 0;
4515                 i_dev++;
4516         }
4517
4518 out:
4519         mutex_unlock(&lock);
4520         cb->args[0] = i_dev;
4521         cb->args[1] = i_id;
4522
4523         return skb->len;
4524 }
4525
4526 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4527         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4528 };
4529
4530 static int cma_init_net(struct net *net)
4531 {
4532         struct cma_pernet *pernet = cma_pernet(net);
4533
4534         idr_init(&pernet->tcp_ps);
4535         idr_init(&pernet->udp_ps);
4536         idr_init(&pernet->ipoib_ps);
4537         idr_init(&pernet->ib_ps);
4538
4539         return 0;
4540 }
4541
4542 static void cma_exit_net(struct net *net)
4543 {
4544         struct cma_pernet *pernet = cma_pernet(net);
4545
4546         idr_destroy(&pernet->tcp_ps);
4547         idr_destroy(&pernet->udp_ps);
4548         idr_destroy(&pernet->ipoib_ps);
4549         idr_destroy(&pernet->ib_ps);
4550 }
4551
4552 static struct pernet_operations cma_pernet_operations = {
4553         .init = cma_init_net,
4554         .exit = cma_exit_net,
4555         .id = &cma_pernet_id,
4556         .size = sizeof(struct cma_pernet),
4557 };
4558
4559 static int __init cma_init(void)
4560 {
4561         int ret;
4562
4563         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4564         if (!cma_wq)
4565                 return -ENOMEM;
4566
4567         ret = register_pernet_subsys(&cma_pernet_operations);
4568         if (ret)
4569                 goto err_wq;
4570
4571         ib_sa_register_client(&sa_client);
4572         rdma_addr_register_client(&addr_client);
4573         register_netdevice_notifier(&cma_nb);
4574
4575         ret = ib_register_client(&cma_client);
4576         if (ret)
4577                 goto err;
4578
4579         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4580         cma_configfs_init();
4581
4582         return 0;
4583
4584 err:
4585         unregister_netdevice_notifier(&cma_nb);
4586         rdma_addr_unregister_client(&addr_client);
4587         ib_sa_unregister_client(&sa_client);
4588 err_wq:
4589         destroy_workqueue(cma_wq);
4590         return ret;
4591 }
4592
4593 static void __exit cma_cleanup(void)
4594 {
4595         cma_configfs_exit();
4596         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4597         ib_unregister_client(&cma_client);
4598         unregister_netdevice_notifier(&cma_nb);
4599         rdma_addr_unregister_client(&addr_client);
4600         ib_sa_unregister_client(&sa_client);
4601         unregister_pernet_subsys(&cma_pernet_operations);
4602         destroy_workqueue(cma_wq);
4603 }
4604
4605 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4606
4607 module_init(cma_init);
4608 module_exit(cma_cleanup);