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