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 int 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                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1850                 if (ret)
1851                         goto err;
1852         } else {
1853                 if (!cma_protocol_roce(listen_id) &&
1854                     cma_any_addr(cma_src_addr(id_priv))) {
1855                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1856                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1857                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1858                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1859                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1860                         if (ret)
1861                                 goto err;
1862                 }
1863         }
1864         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1865
1866         id_priv->state = RDMA_CM_CONNECT;
1867         return id_priv;
1868
1869 err:
1870         rdma_destroy_id(id);
1871         return NULL;
1872 }
1873
1874 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1875                                               struct ib_cm_event *ib_event,
1876                                               struct net_device *net_dev)
1877 {
1878         struct rdma_id_private *id_priv;
1879         struct rdma_cm_id *id;
1880         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1881         struct net *net = listen_id->route.addr.dev_addr.net;
1882         int ret;
1883
1884         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1885                             listen_id->ps, IB_QPT_UD);
1886         if (IS_ERR(id))
1887                 return NULL;
1888
1889         id_priv = container_of(id, struct rdma_id_private, id);
1890         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1891                               (struct sockaddr *)&id->route.addr.dst_addr,
1892                               listen_id, ib_event, ss_family,
1893                               ib_event->param.sidr_req_rcvd.service_id))
1894                 goto err;
1895
1896         if (net_dev) {
1897                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1898                 if (ret)
1899                         goto err;
1900         } else {
1901                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1902                         ret = cma_translate_addr(cma_src_addr(id_priv),
1903                                                  &id->route.addr.dev_addr);
1904                         if (ret)
1905                                 goto err;
1906                 }
1907         }
1908
1909         id_priv->state = RDMA_CM_CONNECT;
1910         return id_priv;
1911 err:
1912         rdma_destroy_id(id);
1913         return NULL;
1914 }
1915
1916 static void cma_set_req_event_data(struct rdma_cm_event *event,
1917                                    struct ib_cm_req_event_param *req_data,
1918                                    void *private_data, int offset)
1919 {
1920         event->param.conn.private_data = private_data + offset;
1921         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1922         event->param.conn.responder_resources = req_data->responder_resources;
1923         event->param.conn.initiator_depth = req_data->initiator_depth;
1924         event->param.conn.flow_control = req_data->flow_control;
1925         event->param.conn.retry_count = req_data->retry_count;
1926         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1927         event->param.conn.srq = req_data->srq;
1928         event->param.conn.qp_num = req_data->remote_qpn;
1929 }
1930
1931 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1932 {
1933         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1934                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1935                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1936                  (id->qp_type == IB_QPT_UD)) ||
1937                 (!id->qp_type));
1938 }
1939
1940 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1941 {
1942         struct rdma_id_private *listen_id, *conn_id = NULL;
1943         struct rdma_cm_event event;
1944         struct net_device *net_dev;
1945         int offset, ret;
1946
1947         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1948         if (IS_ERR(listen_id))
1949                 return PTR_ERR(listen_id);
1950
1951         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1952                 ret = -EINVAL;
1953                 goto net_dev_put;
1954         }
1955
1956         mutex_lock(&listen_id->handler_mutex);
1957         if (listen_id->state != RDMA_CM_LISTEN) {
1958                 ret = -ECONNABORTED;
1959                 goto err1;
1960         }
1961
1962         memset(&event, 0, sizeof event);
1963         offset = cma_user_data_offset(listen_id);
1964         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1965         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1966                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1967                 event.param.ud.private_data = ib_event->private_data + offset;
1968                 event.param.ud.private_data_len =
1969                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1970         } else {
1971                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1972                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1973                                        ib_event->private_data, offset);
1974         }
1975         if (!conn_id) {
1976                 ret = -ENOMEM;
1977                 goto err1;
1978         }
1979
1980         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1981         ret = cma_acquire_dev(conn_id, listen_id);
1982         if (ret)
1983                 goto err2;
1984
1985         conn_id->cm_id.ib = cm_id;
1986         cm_id->context = conn_id;
1987         cm_id->cm_handler = cma_ib_handler;
1988
1989         /*
1990          * Protect against the user destroying conn_id from another thread
1991          * until we're done accessing it.
1992          */
1993         atomic_inc(&conn_id->refcount);
1994         ret = conn_id->id.event_handler(&conn_id->id, &event);
1995         if (ret)
1996                 goto err3;
1997         /*
1998          * Acquire mutex to prevent user executing rdma_destroy_id()
1999          * while we're accessing the cm_id.
2000          */
2001         mutex_lock(&lock);
2002         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2003             (conn_id->id.qp_type != IB_QPT_UD))
2004                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2005         mutex_unlock(&lock);
2006         mutex_unlock(&conn_id->handler_mutex);
2007         mutex_unlock(&listen_id->handler_mutex);
2008         cma_deref_id(conn_id);
2009         if (net_dev)
2010                 dev_put(net_dev);
2011         return 0;
2012
2013 err3:
2014         cma_deref_id(conn_id);
2015         /* Destroy the CM ID by returning a non-zero value. */
2016         conn_id->cm_id.ib = NULL;
2017 err2:
2018         cma_exch(conn_id, RDMA_CM_DESTROYING);
2019         mutex_unlock(&conn_id->handler_mutex);
2020 err1:
2021         mutex_unlock(&listen_id->handler_mutex);
2022         if (conn_id)
2023                 rdma_destroy_id(&conn_id->id);
2024
2025 net_dev_put:
2026         if (net_dev)
2027                 dev_put(net_dev);
2028
2029         return ret;
2030 }
2031
2032 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2033 {
2034         if (addr->sa_family == AF_IB)
2035                 return ((struct sockaddr_ib *) addr)->sib_sid;
2036
2037         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2038 }
2039 EXPORT_SYMBOL(rdma_get_service_id);
2040
2041 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2042 {
2043         struct rdma_id_private *id_priv = iw_id->context;
2044         struct rdma_cm_event event;
2045         int ret = 0;
2046         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2047         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2048
2049         mutex_lock(&id_priv->handler_mutex);
2050         if (id_priv->state != RDMA_CM_CONNECT)
2051                 goto out;
2052
2053         memset(&event, 0, sizeof event);
2054         switch (iw_event->event) {
2055         case IW_CM_EVENT_CLOSE:
2056                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2057                 break;
2058         case IW_CM_EVENT_CONNECT_REPLY:
2059                 memcpy(cma_src_addr(id_priv), laddr,
2060                        rdma_addr_size(laddr));
2061                 memcpy(cma_dst_addr(id_priv), raddr,
2062                        rdma_addr_size(raddr));
2063                 switch (iw_event->status) {
2064                 case 0:
2065                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2066                         event.param.conn.initiator_depth = iw_event->ird;
2067                         event.param.conn.responder_resources = iw_event->ord;
2068                         break;
2069                 case -ECONNRESET:
2070                 case -ECONNREFUSED:
2071                         event.event = RDMA_CM_EVENT_REJECTED;
2072                         break;
2073                 case -ETIMEDOUT:
2074                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2075                         break;
2076                 default:
2077                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2078                         break;
2079                 }
2080                 break;
2081         case IW_CM_EVENT_ESTABLISHED:
2082                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2083                 event.param.conn.initiator_depth = iw_event->ird;
2084                 event.param.conn.responder_resources = iw_event->ord;
2085                 break;
2086         default:
2087                 BUG_ON(1);
2088         }
2089
2090         event.status = iw_event->status;
2091         event.param.conn.private_data = iw_event->private_data;
2092         event.param.conn.private_data_len = iw_event->private_data_len;
2093         ret = id_priv->id.event_handler(&id_priv->id, &event);
2094         if (ret) {
2095                 /* Destroy the CM ID by returning a non-zero value. */
2096                 id_priv->cm_id.iw = NULL;
2097                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2098                 mutex_unlock(&id_priv->handler_mutex);
2099                 rdma_destroy_id(&id_priv->id);
2100                 return ret;
2101         }
2102
2103 out:
2104         mutex_unlock(&id_priv->handler_mutex);
2105         return ret;
2106 }
2107
2108 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2109                                struct iw_cm_event *iw_event)
2110 {
2111         struct rdma_cm_id *new_cm_id;
2112         struct rdma_id_private *listen_id, *conn_id;
2113         struct rdma_cm_event event;
2114         int ret = -ECONNABORTED;
2115         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2116         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2117
2118         listen_id = cm_id->context;
2119
2120         mutex_lock(&listen_id->handler_mutex);
2121         if (listen_id->state != RDMA_CM_LISTEN)
2122                 goto out;
2123
2124         /* Create a new RDMA id for the new IW CM ID */
2125         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2126                                    listen_id->id.event_handler,
2127                                    listen_id->id.context,
2128                                    RDMA_PS_TCP, IB_QPT_RC);
2129         if (IS_ERR(new_cm_id)) {
2130                 ret = -ENOMEM;
2131                 goto out;
2132         }
2133         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2134         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2135         conn_id->state = RDMA_CM_CONNECT;
2136
2137         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2138         if (ret) {
2139                 mutex_unlock(&conn_id->handler_mutex);
2140                 rdma_destroy_id(new_cm_id);
2141                 goto out;
2142         }
2143
2144         ret = cma_acquire_dev(conn_id, listen_id);
2145         if (ret) {
2146                 mutex_unlock(&conn_id->handler_mutex);
2147                 rdma_destroy_id(new_cm_id);
2148                 goto out;
2149         }
2150
2151         conn_id->cm_id.iw = cm_id;
2152         cm_id->context = conn_id;
2153         cm_id->cm_handler = cma_iw_handler;
2154
2155         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2156         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2157
2158         memset(&event, 0, sizeof event);
2159         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2160         event.param.conn.private_data = iw_event->private_data;
2161         event.param.conn.private_data_len = iw_event->private_data_len;
2162         event.param.conn.initiator_depth = iw_event->ird;
2163         event.param.conn.responder_resources = iw_event->ord;
2164
2165         /*
2166          * Protect against the user destroying conn_id from another thread
2167          * until we're done accessing it.
2168          */
2169         atomic_inc(&conn_id->refcount);
2170         ret = conn_id->id.event_handler(&conn_id->id, &event);
2171         if (ret) {
2172                 /* User wants to destroy the CM ID */
2173                 conn_id->cm_id.iw = NULL;
2174                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2175                 mutex_unlock(&conn_id->handler_mutex);
2176                 cma_deref_id(conn_id);
2177                 rdma_destroy_id(&conn_id->id);
2178                 goto out;
2179         }
2180
2181         mutex_unlock(&conn_id->handler_mutex);
2182         cma_deref_id(conn_id);
2183
2184 out:
2185         mutex_unlock(&listen_id->handler_mutex);
2186         return ret;
2187 }
2188
2189 static int cma_ib_listen(struct rdma_id_private *id_priv)
2190 {
2191         struct sockaddr *addr;
2192         struct ib_cm_id *id;
2193         __be64 svc_id;
2194
2195         addr = cma_src_addr(id_priv);
2196         svc_id = rdma_get_service_id(&id_priv->id, addr);
2197         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2198         if (IS_ERR(id))
2199                 return PTR_ERR(id);
2200         id_priv->cm_id.ib = id;
2201
2202         return 0;
2203 }
2204
2205 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2206 {
2207         int ret;
2208         struct iw_cm_id *id;
2209
2210         id = iw_create_cm_id(id_priv->id.device,
2211                              iw_conn_req_handler,
2212                              id_priv);
2213         if (IS_ERR(id))
2214                 return PTR_ERR(id);
2215
2216         id->tos = id_priv->tos;
2217         id_priv->cm_id.iw = id;
2218
2219         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2220                rdma_addr_size(cma_src_addr(id_priv)));
2221
2222         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2223
2224         if (ret) {
2225                 iw_destroy_cm_id(id_priv->cm_id.iw);
2226                 id_priv->cm_id.iw = NULL;
2227         }
2228
2229         return ret;
2230 }
2231
2232 static int cma_listen_handler(struct rdma_cm_id *id,
2233                               struct rdma_cm_event *event)
2234 {
2235         struct rdma_id_private *id_priv = id->context;
2236
2237         id->context = id_priv->id.context;
2238         id->event_handler = id_priv->id.event_handler;
2239         return id_priv->id.event_handler(id, event);
2240 }
2241
2242 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2243                               struct cma_device *cma_dev)
2244 {
2245         struct rdma_id_private *dev_id_priv;
2246         struct rdma_cm_id *id;
2247         struct net *net = id_priv->id.route.addr.dev_addr.net;
2248         int ret;
2249
2250         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2251                 return;
2252
2253         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2254                             id_priv->id.qp_type);
2255         if (IS_ERR(id))
2256                 return;
2257
2258         dev_id_priv = container_of(id, struct rdma_id_private, id);
2259
2260         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2261         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2262                rdma_addr_size(cma_src_addr(id_priv)));
2263
2264         _cma_attach_to_dev(dev_id_priv, cma_dev);
2265         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2266         atomic_inc(&id_priv->refcount);
2267         dev_id_priv->internal_id = 1;
2268         dev_id_priv->afonly = id_priv->afonly;
2269
2270         ret = rdma_listen(id, id_priv->backlog);
2271         if (ret)
2272                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2273                         ret, cma_dev->device->name);
2274 }
2275
2276 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2277 {
2278         struct cma_device *cma_dev;
2279
2280         mutex_lock(&lock);
2281         list_add_tail(&id_priv->list, &listen_any_list);
2282         list_for_each_entry(cma_dev, &dev_list, list)
2283                 cma_listen_on_dev(id_priv, cma_dev);
2284         mutex_unlock(&lock);
2285 }
2286
2287 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2288 {
2289         struct rdma_id_private *id_priv;
2290
2291         id_priv = container_of(id, struct rdma_id_private, id);
2292         id_priv->tos = (u8) tos;
2293         id_priv->tos_set = true;
2294 }
2295 EXPORT_SYMBOL(rdma_set_service_type);
2296
2297 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2298                               void *context)
2299 {
2300         struct cma_work *work = context;
2301         struct rdma_route *route;
2302
2303         route = &work->id->id.route;
2304
2305         if (!status) {
2306                 route->num_paths = 1;
2307                 *route->path_rec = *path_rec;
2308         } else {
2309                 work->old_state = RDMA_CM_ROUTE_QUERY;
2310                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2311                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2312                 work->event.status = status;
2313                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2314                                      status);
2315         }
2316
2317         queue_work(cma_wq, &work->work);
2318 }
2319
2320 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2321                               struct cma_work *work)
2322 {
2323         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2324         struct sa_path_rec path_rec;
2325         ib_sa_comp_mask comp_mask;
2326         struct sockaddr_in6 *sin6;
2327         struct sockaddr_ib *sib;
2328
2329         memset(&path_rec, 0, sizeof path_rec);
2330
2331         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2332                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2333         else
2334                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2335         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2336         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2337         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2338         path_rec.numb_path = 1;
2339         path_rec.reversible = 1;
2340         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2341                                                   cma_dst_addr(id_priv));
2342
2343         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2344                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2345                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2346
2347         switch (cma_family(id_priv)) {
2348         case AF_INET:
2349                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2350                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2351                 break;
2352         case AF_INET6:
2353                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2354                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2355                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2356                 break;
2357         case AF_IB:
2358                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2359                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2360                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2361                 break;
2362         }
2363
2364         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2365                                                id_priv->id.port_num, &path_rec,
2366                                                comp_mask, timeout_ms,
2367                                                GFP_KERNEL, cma_query_handler,
2368                                                work, &id_priv->query);
2369
2370         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2371 }
2372
2373 static void cma_work_handler(struct work_struct *_work)
2374 {
2375         struct cma_work *work = container_of(_work, struct cma_work, work);
2376         struct rdma_id_private *id_priv = work->id;
2377         int destroy = 0;
2378
2379         mutex_lock(&id_priv->handler_mutex);
2380         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2381                 goto out;
2382
2383         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2384                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2385                 destroy = 1;
2386         }
2387 out:
2388         mutex_unlock(&id_priv->handler_mutex);
2389         cma_deref_id(id_priv);
2390         if (destroy)
2391                 rdma_destroy_id(&id_priv->id);
2392         kfree(work);
2393 }
2394
2395 static void cma_ndev_work_handler(struct work_struct *_work)
2396 {
2397         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2398         struct rdma_id_private *id_priv = work->id;
2399         int destroy = 0;
2400
2401         mutex_lock(&id_priv->handler_mutex);
2402         if (id_priv->state == RDMA_CM_DESTROYING ||
2403             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2404                 goto out;
2405
2406         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2407                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2408                 destroy = 1;
2409         }
2410
2411 out:
2412         mutex_unlock(&id_priv->handler_mutex);
2413         cma_deref_id(id_priv);
2414         if (destroy)
2415                 rdma_destroy_id(&id_priv->id);
2416         kfree(work);
2417 }
2418
2419 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2420 {
2421         struct rdma_route *route = &id_priv->id.route;
2422         struct cma_work *work;
2423         int ret;
2424
2425         work = kzalloc(sizeof *work, GFP_KERNEL);
2426         if (!work)
2427                 return -ENOMEM;
2428
2429         work->id = id_priv;
2430         INIT_WORK(&work->work, cma_work_handler);
2431         work->old_state = RDMA_CM_ROUTE_QUERY;
2432         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2433         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2434
2435         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2436         if (!route->path_rec) {
2437                 ret = -ENOMEM;
2438                 goto err1;
2439         }
2440
2441         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2442         if (ret)
2443                 goto err2;
2444
2445         return 0;
2446 err2:
2447         kfree(route->path_rec);
2448         route->path_rec = NULL;
2449 err1:
2450         kfree(work);
2451         return ret;
2452 }
2453
2454 int rdma_set_ib_paths(struct rdma_cm_id *id,
2455                       struct sa_path_rec *path_rec, int num_paths)
2456 {
2457         struct rdma_id_private *id_priv;
2458         int ret;
2459
2460         id_priv = container_of(id, struct rdma_id_private, id);
2461         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2462                            RDMA_CM_ROUTE_RESOLVED))
2463                 return -EINVAL;
2464
2465         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2466                                      GFP_KERNEL);
2467         if (!id->route.path_rec) {
2468                 ret = -ENOMEM;
2469                 goto err;
2470         }
2471
2472         id->route.num_paths = num_paths;
2473         return 0;
2474 err:
2475         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2476         return ret;
2477 }
2478 EXPORT_SYMBOL(rdma_set_ib_paths);
2479
2480 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2481 {
2482         struct cma_work *work;
2483
2484         work = kzalloc(sizeof *work, GFP_KERNEL);
2485         if (!work)
2486                 return -ENOMEM;
2487
2488         work->id = id_priv;
2489         INIT_WORK(&work->work, cma_work_handler);
2490         work->old_state = RDMA_CM_ROUTE_QUERY;
2491         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2492         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2493         queue_work(cma_wq, &work->work);
2494         return 0;
2495 }
2496
2497 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2498 {
2499         int prio;
2500         struct net_device *dev;
2501
2502         prio = rt_tos2priority(tos);
2503         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2504         if (dev->num_tc)
2505                 return netdev_get_prio_tc_map(dev, prio);
2506
2507 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2508         if (is_vlan_dev(ndev))
2509                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2510                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2511 #endif
2512         return 0;
2513 }
2514
2515 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2516                                            unsigned long supported_gids,
2517                                            enum ib_gid_type default_gid)
2518 {
2519         if ((network_type == RDMA_NETWORK_IPV4 ||
2520              network_type == RDMA_NETWORK_IPV6) &&
2521             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2522                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2523
2524         return default_gid;
2525 }
2526
2527 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2528 {
2529         struct rdma_route *route = &id_priv->id.route;
2530         struct rdma_addr *addr = &route->addr;
2531         struct cma_work *work;
2532         int ret;
2533         struct net_device *ndev = NULL;
2534         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
2535         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2536                                         rdma_start_port(id_priv->cma_dev->device)];
2537         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2538
2539
2540         work = kzalloc(sizeof *work, GFP_KERNEL);
2541         if (!work)
2542                 return -ENOMEM;
2543
2544         work->id = id_priv;
2545         INIT_WORK(&work->work, cma_work_handler);
2546
2547         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2548         if (!route->path_rec) {
2549                 ret = -ENOMEM;
2550                 goto err1;
2551         }
2552
2553         route->num_paths = 1;
2554
2555         if (addr->dev_addr.bound_dev_if) {
2556                 unsigned long supported_gids;
2557
2558                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2559                 if (!ndev) {
2560                         ret = -ENODEV;
2561                         goto err2;
2562                 }
2563
2564                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2565                                                             id_priv->id.port_num);
2566                 gid_type = cma_route_gid_type(addr->dev_addr.network,
2567                                               supported_gids,
2568                                               id_priv->gid_type);
2569                 route->path_rec->rec_type =
2570                         sa_conv_gid_to_pathrec_type(gid_type);
2571                 sa_path_set_ndev(route->path_rec, &init_net);
2572                 sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2573         }
2574         if (!ndev) {
2575                 ret = -ENODEV;
2576                 goto err2;
2577         }
2578
2579         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2580
2581         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2582                     &route->path_rec->sgid);
2583         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2584                     &route->path_rec->dgid);
2585
2586         /* Use the hint from IP Stack to select GID Type */
2587         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2588                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2589         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2590
2591         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2592                 /* TODO: get the hoplimit from the inet/inet6 device */
2593                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2594         else
2595                 route->path_rec->hop_limit = 1;
2596         route->path_rec->reversible = 1;
2597         route->path_rec->pkey = cpu_to_be16(0xffff);
2598         route->path_rec->mtu_selector = IB_SA_EQ;
2599         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2600         route->path_rec->traffic_class = tos;
2601         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2602         route->path_rec->rate_selector = IB_SA_EQ;
2603         route->path_rec->rate = iboe_get_rate(ndev);
2604         dev_put(ndev);
2605         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2606         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2607         if (!route->path_rec->mtu) {
2608                 ret = -EINVAL;
2609                 goto err2;
2610         }
2611
2612         work->old_state = RDMA_CM_ROUTE_QUERY;
2613         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2614         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2615         work->event.status = 0;
2616
2617         queue_work(cma_wq, &work->work);
2618
2619         return 0;
2620
2621 err2:
2622         kfree(route->path_rec);
2623         route->path_rec = NULL;
2624 err1:
2625         kfree(work);
2626         return ret;
2627 }
2628
2629 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2630 {
2631         struct rdma_id_private *id_priv;
2632         int ret;
2633
2634         id_priv = container_of(id, struct rdma_id_private, id);
2635         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2636                 return -EINVAL;
2637
2638         atomic_inc(&id_priv->refcount);
2639         if (rdma_cap_ib_sa(id->device, id->port_num))
2640                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2641         else if (rdma_protocol_roce(id->device, id->port_num))
2642                 ret = cma_resolve_iboe_route(id_priv);
2643         else if (rdma_protocol_iwarp(id->device, id->port_num))
2644                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2645         else
2646                 ret = -ENOSYS;
2647
2648         if (ret)
2649                 goto err;
2650
2651         return 0;
2652 err:
2653         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2654         cma_deref_id(id_priv);
2655         return ret;
2656 }
2657 EXPORT_SYMBOL(rdma_resolve_route);
2658
2659 static void cma_set_loopback(struct sockaddr *addr)
2660 {
2661         switch (addr->sa_family) {
2662         case AF_INET:
2663                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2664                 break;
2665         case AF_INET6:
2666                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2667                               0, 0, 0, htonl(1));
2668                 break;
2669         default:
2670                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2671                             0, 0, 0, htonl(1));
2672                 break;
2673         }
2674 }
2675
2676 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2677 {
2678         struct cma_device *cma_dev, *cur_dev;
2679         union ib_gid gid;
2680         enum ib_port_state port_state;
2681         u16 pkey;
2682         int ret;
2683         u8 p;
2684
2685         cma_dev = NULL;
2686         mutex_lock(&lock);
2687         list_for_each_entry(cur_dev, &dev_list, list) {
2688                 if (cma_family(id_priv) == AF_IB &&
2689                     !rdma_cap_ib_cm(cur_dev->device, 1))
2690                         continue;
2691
2692                 if (!cma_dev)
2693                         cma_dev = cur_dev;
2694
2695                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2696                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2697                             port_state == IB_PORT_ACTIVE) {
2698                                 cma_dev = cur_dev;
2699                                 goto port_found;
2700                         }
2701                 }
2702         }
2703
2704         if (!cma_dev) {
2705                 ret = -ENODEV;
2706                 goto out;
2707         }
2708
2709         p = 1;
2710
2711 port_found:
2712         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2713         if (ret)
2714                 goto out;
2715
2716         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2717         if (ret)
2718                 goto out;
2719
2720         id_priv->id.route.addr.dev_addr.dev_type =
2721                 (rdma_protocol_ib(cma_dev->device, p)) ?
2722                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2723
2724         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2725         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2726         id_priv->id.port_num = p;
2727         cma_attach_to_dev(id_priv, cma_dev);
2728         cma_set_loopback(cma_src_addr(id_priv));
2729 out:
2730         mutex_unlock(&lock);
2731         return ret;
2732 }
2733
2734 static void addr_handler(int status, struct sockaddr *src_addr,
2735                          struct rdma_dev_addr *dev_addr, void *context)
2736 {
2737         struct rdma_id_private *id_priv = context;
2738         struct rdma_cm_event event;
2739
2740         memset(&event, 0, sizeof event);
2741         mutex_lock(&id_priv->handler_mutex);
2742         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2743                            RDMA_CM_ADDR_RESOLVED))
2744                 goto out;
2745
2746         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2747         if (!status && !id_priv->cma_dev) {
2748                 status = cma_acquire_dev(id_priv, NULL);
2749                 if (status)
2750                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2751                                              status);
2752         } else {
2753                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2754         }
2755
2756         if (status) {
2757                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2758                                    RDMA_CM_ADDR_BOUND))
2759                         goto out;
2760                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2761                 event.status = status;
2762         } else
2763                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2764
2765         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2766                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2767                 mutex_unlock(&id_priv->handler_mutex);
2768                 cma_deref_id(id_priv);
2769                 rdma_destroy_id(&id_priv->id);
2770                 return;
2771         }
2772 out:
2773         mutex_unlock(&id_priv->handler_mutex);
2774         cma_deref_id(id_priv);
2775 }
2776
2777 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2778 {
2779         struct cma_work *work;
2780         union ib_gid gid;
2781         int ret;
2782
2783         work = kzalloc(sizeof *work, GFP_KERNEL);
2784         if (!work)
2785                 return -ENOMEM;
2786
2787         if (!id_priv->cma_dev) {
2788                 ret = cma_bind_loopback(id_priv);
2789                 if (ret)
2790                         goto err;
2791         }
2792
2793         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2794         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2795
2796         work->id = id_priv;
2797         INIT_WORK(&work->work, cma_work_handler);
2798         work->old_state = RDMA_CM_ADDR_QUERY;
2799         work->new_state = RDMA_CM_ADDR_RESOLVED;
2800         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2801         queue_work(cma_wq, &work->work);
2802         return 0;
2803 err:
2804         kfree(work);
2805         return ret;
2806 }
2807
2808 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2809 {
2810         struct cma_work *work;
2811         int ret;
2812
2813         work = kzalloc(sizeof *work, GFP_KERNEL);
2814         if (!work)
2815                 return -ENOMEM;
2816
2817         if (!id_priv->cma_dev) {
2818                 ret = cma_resolve_ib_dev(id_priv);
2819                 if (ret)
2820                         goto err;
2821         }
2822
2823         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2824                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2825
2826         work->id = id_priv;
2827         INIT_WORK(&work->work, cma_work_handler);
2828         work->old_state = RDMA_CM_ADDR_QUERY;
2829         work->new_state = RDMA_CM_ADDR_RESOLVED;
2830         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2831         queue_work(cma_wq, &work->work);
2832         return 0;
2833 err:
2834         kfree(work);
2835         return ret;
2836 }
2837
2838 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2839                          struct sockaddr *dst_addr)
2840 {
2841         if (!src_addr || !src_addr->sa_family) {
2842                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2843                 src_addr->sa_family = dst_addr->sa_family;
2844                 if (IS_ENABLED(CONFIG_IPV6) &&
2845                     dst_addr->sa_family == AF_INET6) {
2846                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2847                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2848                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2849                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2850                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2851                 } else if (dst_addr->sa_family == AF_IB) {
2852                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2853                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2854                 }
2855         }
2856         return rdma_bind_addr(id, src_addr);
2857 }
2858
2859 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2860                       struct sockaddr *dst_addr, int timeout_ms)
2861 {
2862         struct rdma_id_private *id_priv;
2863         int ret;
2864
2865         id_priv = container_of(id, struct rdma_id_private, id);
2866         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2867         if (id_priv->state == RDMA_CM_IDLE) {
2868                 ret = cma_bind_addr(id, src_addr, dst_addr);
2869                 if (ret) {
2870                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2871                         return ret;
2872                 }
2873         }
2874
2875         if (cma_family(id_priv) != dst_addr->sa_family) {
2876                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2877                 return -EINVAL;
2878         }
2879
2880         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2881                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2882                 return -EINVAL;
2883         }
2884
2885         atomic_inc(&id_priv->refcount);
2886         if (cma_any_addr(dst_addr)) {
2887                 ret = cma_resolve_loopback(id_priv);
2888         } else {
2889                 if (dst_addr->sa_family == AF_IB) {
2890                         ret = cma_resolve_ib_addr(id_priv);
2891                 } else {
2892                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2893                                               dst_addr, &id->route.addr.dev_addr,
2894                                               timeout_ms, addr_handler, id_priv);
2895                 }
2896         }
2897         if (ret)
2898                 goto err;
2899
2900         return 0;
2901 err:
2902         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2903         cma_deref_id(id_priv);
2904         return ret;
2905 }
2906 EXPORT_SYMBOL(rdma_resolve_addr);
2907
2908 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2909 {
2910         struct rdma_id_private *id_priv;
2911         unsigned long flags;
2912         int ret;
2913
2914         id_priv = container_of(id, struct rdma_id_private, id);
2915         spin_lock_irqsave(&id_priv->lock, flags);
2916         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2917                 id_priv->reuseaddr = reuse;
2918                 ret = 0;
2919         } else {
2920                 ret = -EINVAL;
2921         }
2922         spin_unlock_irqrestore(&id_priv->lock, flags);
2923         return ret;
2924 }
2925 EXPORT_SYMBOL(rdma_set_reuseaddr);
2926
2927 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2928 {
2929         struct rdma_id_private *id_priv;
2930         unsigned long flags;
2931         int ret;
2932
2933         id_priv = container_of(id, struct rdma_id_private, id);
2934         spin_lock_irqsave(&id_priv->lock, flags);
2935         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2936                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2937                 id_priv->afonly = afonly;
2938                 ret = 0;
2939         } else {
2940                 ret = -EINVAL;
2941         }
2942         spin_unlock_irqrestore(&id_priv->lock, flags);
2943         return ret;
2944 }
2945 EXPORT_SYMBOL(rdma_set_afonly);
2946
2947 static void cma_bind_port(struct rdma_bind_list *bind_list,
2948                           struct rdma_id_private *id_priv)
2949 {
2950         struct sockaddr *addr;
2951         struct sockaddr_ib *sib;
2952         u64 sid, mask;
2953         __be16 port;
2954
2955         addr = cma_src_addr(id_priv);
2956         port = htons(bind_list->port);
2957
2958         switch (addr->sa_family) {
2959         case AF_INET:
2960                 ((struct sockaddr_in *) addr)->sin_port = port;
2961                 break;
2962         case AF_INET6:
2963                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2964                 break;
2965         case AF_IB:
2966                 sib = (struct sockaddr_ib *) addr;
2967                 sid = be64_to_cpu(sib->sib_sid);
2968                 mask = be64_to_cpu(sib->sib_sid_mask);
2969                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2970                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2971                 break;
2972         }
2973         id_priv->bind_list = bind_list;
2974         hlist_add_head(&id_priv->node, &bind_list->owners);
2975 }
2976
2977 static int cma_alloc_port(enum rdma_port_space ps,
2978                           struct rdma_id_private *id_priv, unsigned short snum)
2979 {
2980         struct rdma_bind_list *bind_list;
2981         int ret;
2982
2983         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2984         if (!bind_list)
2985                 return -ENOMEM;
2986
2987         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2988                            snum);
2989         if (ret < 0)
2990                 goto err;
2991
2992         bind_list->ps = ps;
2993         bind_list->port = (unsigned short)ret;
2994         cma_bind_port(bind_list, id_priv);
2995         return 0;
2996 err:
2997         kfree(bind_list);
2998         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2999 }
3000
3001 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3002                               struct rdma_id_private *id_priv)
3003 {
3004         struct rdma_id_private *cur_id;
3005         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3006         struct sockaddr  *saddr = cma_src_addr(id_priv);
3007         __be16 dport = cma_port(daddr);
3008
3009         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3010                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3011                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3012                 __be16 cur_dport = cma_port(cur_daddr);
3013
3014                 if (id_priv == cur_id)
3015                         continue;
3016
3017                 /* different dest port -> unique */
3018                 if (!cma_any_port(cur_daddr) &&
3019                     (dport != cur_dport))
3020                         continue;
3021
3022                 /* different src address -> unique */
3023                 if (!cma_any_addr(saddr) &&
3024                     !cma_any_addr(cur_saddr) &&
3025                     cma_addr_cmp(saddr, cur_saddr))
3026                         continue;
3027
3028                 /* different dst address -> unique */
3029                 if (!cma_any_addr(cur_daddr) &&
3030                     cma_addr_cmp(daddr, cur_daddr))
3031                         continue;
3032
3033                 return -EADDRNOTAVAIL;
3034         }
3035         return 0;
3036 }
3037
3038 static int cma_alloc_any_port(enum rdma_port_space ps,
3039                               struct rdma_id_private *id_priv)
3040 {
3041         static unsigned int last_used_port;
3042         int low, high, remaining;
3043         unsigned int rover;
3044         struct net *net = id_priv->id.route.addr.dev_addr.net;
3045
3046         inet_get_local_port_range(net, &low, &high);
3047         remaining = (high - low) + 1;
3048         rover = prandom_u32() % remaining + low;
3049 retry:
3050         if (last_used_port != rover) {
3051                 struct rdma_bind_list *bind_list;
3052                 int ret;
3053
3054                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3055
3056                 if (!bind_list) {
3057                         ret = cma_alloc_port(ps, id_priv, rover);
3058                 } else {
3059                         ret = cma_port_is_unique(bind_list, id_priv);
3060                         if (!ret)
3061                                 cma_bind_port(bind_list, id_priv);
3062                 }
3063                 /*
3064                  * Remember previously used port number in order to avoid
3065                  * re-using same port immediately after it is closed.
3066                  */
3067                 if (!ret)
3068                         last_used_port = rover;
3069                 if (ret != -EADDRNOTAVAIL)
3070                         return ret;
3071         }
3072         if (--remaining) {
3073                 rover++;
3074                 if ((rover < low) || (rover > high))
3075                         rover = low;
3076                 goto retry;
3077         }
3078         return -EADDRNOTAVAIL;
3079 }
3080
3081 /*
3082  * Check that the requested port is available.  This is called when trying to
3083  * bind to a specific port, or when trying to listen on a bound port.  In
3084  * the latter case, the provided id_priv may already be on the bind_list, but
3085  * we still need to check that it's okay to start listening.
3086  */
3087 static int cma_check_port(struct rdma_bind_list *bind_list,
3088                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3089 {
3090         struct rdma_id_private *cur_id;
3091         struct sockaddr *addr, *cur_addr;
3092
3093         addr = cma_src_addr(id_priv);
3094         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3095                 if (id_priv == cur_id)
3096                         continue;
3097
3098                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3099                     cur_id->reuseaddr)
3100                         continue;
3101
3102                 cur_addr = cma_src_addr(cur_id);
3103                 if (id_priv->afonly && cur_id->afonly &&
3104                     (addr->sa_family != cur_addr->sa_family))
3105                         continue;
3106
3107                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3108                         return -EADDRNOTAVAIL;
3109
3110                 if (!cma_addr_cmp(addr, cur_addr))
3111                         return -EADDRINUSE;
3112         }
3113         return 0;
3114 }
3115
3116 static int cma_use_port(enum rdma_port_space ps,
3117                         struct rdma_id_private *id_priv)
3118 {
3119         struct rdma_bind_list *bind_list;
3120         unsigned short snum;
3121         int ret;
3122
3123         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3124         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3125                 return -EACCES;
3126
3127         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3128         if (!bind_list) {
3129                 ret = cma_alloc_port(ps, id_priv, snum);
3130         } else {
3131                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3132                 if (!ret)
3133                         cma_bind_port(bind_list, id_priv);
3134         }
3135         return ret;
3136 }
3137
3138 static int cma_bind_listen(struct rdma_id_private *id_priv)
3139 {
3140         struct rdma_bind_list *bind_list = id_priv->bind_list;
3141         int ret = 0;
3142
3143         mutex_lock(&lock);
3144         if (bind_list->owners.first->next)
3145                 ret = cma_check_port(bind_list, id_priv, 0);
3146         mutex_unlock(&lock);
3147         return ret;
3148 }
3149
3150 static enum rdma_port_space cma_select_inet_ps(
3151                 struct rdma_id_private *id_priv)
3152 {
3153         switch (id_priv->id.ps) {
3154         case RDMA_PS_TCP:
3155         case RDMA_PS_UDP:
3156         case RDMA_PS_IPOIB:
3157         case RDMA_PS_IB:
3158                 return id_priv->id.ps;
3159         default:
3160
3161                 return 0;
3162         }
3163 }
3164
3165 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3166 {
3167         enum rdma_port_space ps = 0;
3168         struct sockaddr_ib *sib;
3169         u64 sid_ps, mask, sid;
3170
3171         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3172         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3173         sid = be64_to_cpu(sib->sib_sid) & mask;
3174
3175         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3176                 sid_ps = RDMA_IB_IP_PS_IB;
3177                 ps = RDMA_PS_IB;
3178         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3179                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3180                 sid_ps = RDMA_IB_IP_PS_TCP;
3181                 ps = RDMA_PS_TCP;
3182         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3183                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3184                 sid_ps = RDMA_IB_IP_PS_UDP;
3185                 ps = RDMA_PS_UDP;
3186         }
3187
3188         if (ps) {
3189                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3190                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3191                                                 be64_to_cpu(sib->sib_sid_mask));
3192         }
3193         return ps;
3194 }
3195
3196 static int cma_get_port(struct rdma_id_private *id_priv)
3197 {
3198         enum rdma_port_space ps;
3199         int ret;
3200
3201         if (cma_family(id_priv) != AF_IB)
3202                 ps = cma_select_inet_ps(id_priv);
3203         else
3204                 ps = cma_select_ib_ps(id_priv);
3205         if (!ps)
3206                 return -EPROTONOSUPPORT;
3207
3208         mutex_lock(&lock);
3209         if (cma_any_port(cma_src_addr(id_priv)))
3210                 ret = cma_alloc_any_port(ps, id_priv);
3211         else
3212                 ret = cma_use_port(ps, id_priv);
3213         mutex_unlock(&lock);
3214
3215         return ret;
3216 }
3217
3218 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3219                                struct sockaddr *addr)
3220 {
3221 #if IS_ENABLED(CONFIG_IPV6)
3222         struct sockaddr_in6 *sin6;
3223
3224         if (addr->sa_family != AF_INET6)
3225                 return 0;
3226
3227         sin6 = (struct sockaddr_in6 *) addr;
3228
3229         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3230                 return 0;
3231
3232         if (!sin6->sin6_scope_id)
3233                         return -EINVAL;
3234
3235         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3236 #endif
3237         return 0;
3238 }
3239
3240 int rdma_listen(struct rdma_cm_id *id, int backlog)
3241 {
3242         struct rdma_id_private *id_priv;
3243         int ret;
3244
3245         id_priv = container_of(id, struct rdma_id_private, id);
3246         if (id_priv->state == RDMA_CM_IDLE) {
3247                 id->route.addr.src_addr.ss_family = AF_INET;
3248                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3249                 if (ret)
3250                         return ret;
3251         }
3252
3253         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3254                 return -EINVAL;
3255
3256         if (id_priv->reuseaddr) {
3257                 ret = cma_bind_listen(id_priv);
3258                 if (ret)
3259                         goto err;
3260         }
3261
3262         id_priv->backlog = backlog;
3263         if (id->device) {
3264                 if (rdma_cap_ib_cm(id->device, 1)) {
3265                         ret = cma_ib_listen(id_priv);
3266                         if (ret)
3267                                 goto err;
3268                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3269                         ret = cma_iw_listen(id_priv, backlog);
3270                         if (ret)
3271                                 goto err;
3272                 } else {
3273                         ret = -ENOSYS;
3274                         goto err;
3275                 }
3276         } else
3277                 cma_listen_on_all(id_priv);
3278
3279         return 0;
3280 err:
3281         id_priv->backlog = 0;
3282         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3283         return ret;
3284 }
3285 EXPORT_SYMBOL(rdma_listen);
3286
3287 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3288 {
3289         struct rdma_id_private *id_priv;
3290         int ret;
3291         struct sockaddr  *daddr;
3292
3293         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3294             addr->sa_family != AF_IB)
3295                 return -EAFNOSUPPORT;
3296
3297         id_priv = container_of(id, struct rdma_id_private, id);
3298         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3299                 return -EINVAL;
3300
3301         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3302         if (ret)
3303                 goto err1;
3304
3305         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3306         if (!cma_any_addr(addr)) {
3307                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3308                 if (ret)
3309                         goto err1;
3310
3311                 ret = cma_acquire_dev(id_priv, NULL);
3312                 if (ret)
3313                         goto err1;
3314         }
3315
3316         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3317                 if (addr->sa_family == AF_INET)
3318                         id_priv->afonly = 1;
3319 #if IS_ENABLED(CONFIG_IPV6)
3320                 else if (addr->sa_family == AF_INET6) {
3321                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3322
3323                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3324                 }
3325 #endif
3326         }
3327         ret = cma_get_port(id_priv);
3328         if (ret)
3329                 goto err2;
3330
3331         daddr = cma_dst_addr(id_priv);
3332         daddr->sa_family = addr->sa_family;
3333
3334         return 0;
3335 err2:
3336         if (id_priv->cma_dev)
3337                 cma_release_dev(id_priv);
3338 err1:
3339         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3340         return ret;
3341 }
3342 EXPORT_SYMBOL(rdma_bind_addr);
3343
3344 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3345 {
3346         struct cma_hdr *cma_hdr;
3347
3348         cma_hdr = hdr;
3349         cma_hdr->cma_version = CMA_VERSION;
3350         if (cma_family(id_priv) == AF_INET) {
3351                 struct sockaddr_in *src4, *dst4;
3352
3353                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3354                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3355
3356                 cma_set_ip_ver(cma_hdr, 4);
3357                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3358                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3359                 cma_hdr->port = src4->sin_port;
3360         } else if (cma_family(id_priv) == AF_INET6) {
3361                 struct sockaddr_in6 *src6, *dst6;
3362
3363                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3364                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3365
3366                 cma_set_ip_ver(cma_hdr, 6);
3367                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3368                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3369                 cma_hdr->port = src6->sin6_port;
3370         }
3371         return 0;
3372 }
3373
3374 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3375                                 struct ib_cm_event *ib_event)
3376 {
3377         struct rdma_id_private *id_priv = cm_id->context;
3378         struct rdma_cm_event event;
3379         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3380         int ret = 0;
3381
3382         mutex_lock(&id_priv->handler_mutex);
3383         if (id_priv->state != RDMA_CM_CONNECT)
3384                 goto out;
3385
3386         memset(&event, 0, sizeof event);
3387         switch (ib_event->event) {
3388         case IB_CM_SIDR_REQ_ERROR:
3389                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3390                 event.status = -ETIMEDOUT;
3391                 break;
3392         case IB_CM_SIDR_REP_RECEIVED:
3393                 event.param.ud.private_data = ib_event->private_data;
3394                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3395                 if (rep->status != IB_SIDR_SUCCESS) {
3396                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3397                         event.status = ib_event->param.sidr_rep_rcvd.status;
3398                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3399                                              event.status);
3400                         break;
3401                 }
3402                 ret = cma_set_qkey(id_priv, rep->qkey);
3403                 if (ret) {
3404                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3405                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3406                         event.status = ret;
3407                         break;
3408                 }
3409                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3410                                      id_priv->id.route.path_rec,
3411                                      &event.param.ud.ah_attr);
3412                 event.param.ud.qp_num = rep->qpn;
3413                 event.param.ud.qkey = rep->qkey;
3414                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3415                 event.status = 0;
3416                 break;
3417         default:
3418                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3419                        ib_event->event);
3420                 goto out;
3421         }
3422
3423         ret = id_priv->id.event_handler(&id_priv->id, &event);
3424         if (ret) {
3425                 /* Destroy the CM ID by returning a non-zero value. */
3426                 id_priv->cm_id.ib = NULL;
3427                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3428                 mutex_unlock(&id_priv->handler_mutex);
3429                 rdma_destroy_id(&id_priv->id);
3430                 return ret;
3431         }
3432 out:
3433         mutex_unlock(&id_priv->handler_mutex);
3434         return ret;
3435 }
3436
3437 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3438                               struct rdma_conn_param *conn_param)
3439 {
3440         struct ib_cm_sidr_req_param req;
3441         struct ib_cm_id *id;
3442         void *private_data;
3443         int offset, ret;
3444
3445         memset(&req, 0, sizeof req);
3446         offset = cma_user_data_offset(id_priv);
3447         req.private_data_len = offset + conn_param->private_data_len;
3448         if (req.private_data_len < conn_param->private_data_len)
3449                 return -EINVAL;
3450
3451         if (req.private_data_len) {
3452                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3453                 if (!private_data)
3454                         return -ENOMEM;
3455         } else {
3456                 private_data = NULL;
3457         }
3458
3459         if (conn_param->private_data && conn_param->private_data_len)
3460                 memcpy(private_data + offset, conn_param->private_data,
3461                        conn_param->private_data_len);
3462
3463         if (private_data) {
3464                 ret = cma_format_hdr(private_data, id_priv);
3465                 if (ret)
3466                         goto out;
3467                 req.private_data = private_data;
3468         }
3469
3470         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3471                              id_priv);
3472         if (IS_ERR(id)) {
3473                 ret = PTR_ERR(id);
3474                 goto out;
3475         }
3476         id_priv->cm_id.ib = id;
3477
3478         req.path = id_priv->id.route.path_rec;
3479         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3480         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3481         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3482
3483         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3484         if (ret) {
3485                 ib_destroy_cm_id(id_priv->cm_id.ib);
3486                 id_priv->cm_id.ib = NULL;
3487         }
3488 out:
3489         kfree(private_data);
3490         return ret;
3491 }
3492
3493 static int cma_connect_ib(struct rdma_id_private *id_priv,
3494                           struct rdma_conn_param *conn_param)
3495 {
3496         struct ib_cm_req_param req;
3497         struct rdma_route *route;
3498         void *private_data;
3499         struct ib_cm_id *id;
3500         int offset, ret;
3501
3502         memset(&req, 0, sizeof req);
3503         offset = cma_user_data_offset(id_priv);
3504         req.private_data_len = offset + conn_param->private_data_len;
3505         if (req.private_data_len < conn_param->private_data_len)
3506                 return -EINVAL;
3507
3508         if (req.private_data_len) {
3509                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3510                 if (!private_data)
3511                         return -ENOMEM;
3512         } else {
3513                 private_data = NULL;
3514         }
3515
3516         if (conn_param->private_data && conn_param->private_data_len)
3517                 memcpy(private_data + offset, conn_param->private_data,
3518                        conn_param->private_data_len);
3519
3520         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3521         if (IS_ERR(id)) {
3522                 ret = PTR_ERR(id);
3523                 goto out;
3524         }
3525         id_priv->cm_id.ib = id;
3526
3527         route = &id_priv->id.route;
3528         if (private_data) {
3529                 ret = cma_format_hdr(private_data, id_priv);
3530                 if (ret)
3531                         goto out;
3532                 req.private_data = private_data;
3533         }
3534
3535         req.primary_path = &route->path_rec[0];
3536         if (route->num_paths == 2)
3537                 req.alternate_path = &route->path_rec[1];
3538
3539         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3540         req.qp_num = id_priv->qp_num;
3541         req.qp_type = id_priv->id.qp_type;
3542         req.starting_psn = id_priv->seq_num;
3543         req.responder_resources = conn_param->responder_resources;
3544         req.initiator_depth = conn_param->initiator_depth;
3545         req.flow_control = conn_param->flow_control;
3546         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3547         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3548         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3549         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3550         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3551         req.srq = id_priv->srq ? 1 : 0;
3552
3553         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3554 out:
3555         if (ret && !IS_ERR(id)) {
3556                 ib_destroy_cm_id(id);
3557                 id_priv->cm_id.ib = NULL;
3558         }
3559
3560         kfree(private_data);
3561         return ret;
3562 }
3563
3564 static int cma_connect_iw(struct rdma_id_private *id_priv,
3565                           struct rdma_conn_param *conn_param)
3566 {
3567         struct iw_cm_id *cm_id;
3568         int ret;
3569         struct iw_cm_conn_param iw_param;
3570
3571         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3572         if (IS_ERR(cm_id))
3573                 return PTR_ERR(cm_id);
3574
3575         cm_id->tos = id_priv->tos;
3576         id_priv->cm_id.iw = cm_id;
3577
3578         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3579                rdma_addr_size(cma_src_addr(id_priv)));
3580         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3581                rdma_addr_size(cma_dst_addr(id_priv)));
3582
3583         ret = cma_modify_qp_rtr(id_priv, conn_param);
3584         if (ret)
3585                 goto out;
3586
3587         if (conn_param) {
3588                 iw_param.ord = conn_param->initiator_depth;
3589                 iw_param.ird = conn_param->responder_resources;
3590                 iw_param.private_data = conn_param->private_data;
3591                 iw_param.private_data_len = conn_param->private_data_len;
3592                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3593         } else {
3594                 memset(&iw_param, 0, sizeof iw_param);
3595                 iw_param.qpn = id_priv->qp_num;
3596         }
3597         ret = iw_cm_connect(cm_id, &iw_param);
3598 out:
3599         if (ret) {
3600                 iw_destroy_cm_id(cm_id);
3601                 id_priv->cm_id.iw = NULL;
3602         }
3603         return ret;
3604 }
3605
3606 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3607 {
3608         struct rdma_id_private *id_priv;
3609         int ret;
3610
3611         id_priv = container_of(id, struct rdma_id_private, id);
3612         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3613                 return -EINVAL;
3614
3615         if (!id->qp) {
3616                 id_priv->qp_num = conn_param->qp_num;
3617                 id_priv->srq = conn_param->srq;
3618         }
3619
3620         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3621                 if (id->qp_type == IB_QPT_UD)
3622                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3623                 else
3624                         ret = cma_connect_ib(id_priv, conn_param);
3625         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3626                 ret = cma_connect_iw(id_priv, conn_param);
3627         else
3628                 ret = -ENOSYS;
3629         if (ret)
3630                 goto err;
3631
3632         return 0;
3633 err:
3634         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3635         return ret;
3636 }
3637 EXPORT_SYMBOL(rdma_connect);
3638
3639 static int cma_accept_ib(struct rdma_id_private *id_priv,
3640                          struct rdma_conn_param *conn_param)
3641 {
3642         struct ib_cm_rep_param rep;
3643         int ret;
3644
3645         ret = cma_modify_qp_rtr(id_priv, conn_param);
3646         if (ret)
3647                 goto out;
3648
3649         ret = cma_modify_qp_rts(id_priv, conn_param);
3650         if (ret)
3651                 goto out;
3652
3653         memset(&rep, 0, sizeof rep);
3654         rep.qp_num = id_priv->qp_num;
3655         rep.starting_psn = id_priv->seq_num;
3656         rep.private_data = conn_param->private_data;
3657         rep.private_data_len = conn_param->private_data_len;
3658         rep.responder_resources = conn_param->responder_resources;
3659         rep.initiator_depth = conn_param->initiator_depth;
3660         rep.failover_accepted = 0;
3661         rep.flow_control = conn_param->flow_control;
3662         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3663         rep.srq = id_priv->srq ? 1 : 0;
3664
3665         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3666 out:
3667         return ret;
3668 }
3669
3670 static int cma_accept_iw(struct rdma_id_private *id_priv,
3671                   struct rdma_conn_param *conn_param)
3672 {
3673         struct iw_cm_conn_param iw_param;
3674         int ret;
3675
3676         if (!conn_param)
3677                 return -EINVAL;
3678
3679         ret = cma_modify_qp_rtr(id_priv, conn_param);
3680         if (ret)
3681                 return ret;
3682
3683         iw_param.ord = conn_param->initiator_depth;
3684         iw_param.ird = conn_param->responder_resources;
3685         iw_param.private_data = conn_param->private_data;
3686         iw_param.private_data_len = conn_param->private_data_len;
3687         if (id_priv->id.qp) {
3688                 iw_param.qpn = id_priv->qp_num;
3689         } else
3690                 iw_param.qpn = conn_param->qp_num;
3691
3692         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3693 }
3694
3695 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3696                              enum ib_cm_sidr_status status, u32 qkey,
3697                              const void *private_data, int private_data_len)
3698 {
3699         struct ib_cm_sidr_rep_param rep;
3700         int ret;
3701
3702         memset(&rep, 0, sizeof rep);
3703         rep.status = status;
3704         if (status == IB_SIDR_SUCCESS) {
3705                 ret = cma_set_qkey(id_priv, qkey);
3706                 if (ret)
3707                         return ret;
3708                 rep.qp_num = id_priv->qp_num;
3709                 rep.qkey = id_priv->qkey;
3710         }
3711         rep.private_data = private_data;
3712         rep.private_data_len = private_data_len;
3713
3714         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3715 }
3716
3717 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3718 {
3719         struct rdma_id_private *id_priv;
3720         int ret;
3721
3722         id_priv = container_of(id, struct rdma_id_private, id);
3723
3724         id_priv->owner = task_pid_nr(current);
3725
3726         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3727                 return -EINVAL;
3728
3729         if (!id->qp && conn_param) {
3730                 id_priv->qp_num = conn_param->qp_num;
3731                 id_priv->srq = conn_param->srq;
3732         }
3733
3734         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3735                 if (id->qp_type == IB_QPT_UD) {
3736                         if (conn_param)
3737                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3738                                                         conn_param->qkey,
3739                                                         conn_param->private_data,
3740                                                         conn_param->private_data_len);
3741                         else
3742                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3743                                                         0, NULL, 0);
3744                 } else {
3745                         if (conn_param)
3746                                 ret = cma_accept_ib(id_priv, conn_param);
3747                         else
3748                                 ret = cma_rep_recv(id_priv);
3749                 }
3750         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3751                 ret = cma_accept_iw(id_priv, conn_param);
3752         else
3753                 ret = -ENOSYS;
3754
3755         if (ret)
3756                 goto reject;
3757
3758         return 0;
3759 reject:
3760         cma_modify_qp_err(id_priv);
3761         rdma_reject(id, NULL, 0);
3762         return ret;
3763 }
3764 EXPORT_SYMBOL(rdma_accept);
3765
3766 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3767 {
3768         struct rdma_id_private *id_priv;
3769         int ret;
3770
3771         id_priv = container_of(id, struct rdma_id_private, id);
3772         if (!id_priv->cm_id.ib)
3773                 return -EINVAL;
3774
3775         switch (id->device->node_type) {
3776         case RDMA_NODE_IB_CA:
3777                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3778                 break;
3779         default:
3780                 ret = 0;
3781                 break;
3782         }
3783         return ret;
3784 }
3785 EXPORT_SYMBOL(rdma_notify);
3786
3787 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3788                 u8 private_data_len)
3789 {
3790         struct rdma_id_private *id_priv;
3791         int ret;
3792
3793         id_priv = container_of(id, struct rdma_id_private, id);
3794         if (!id_priv->cm_id.ib)
3795                 return -EINVAL;
3796
3797         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3798                 if (id->qp_type == IB_QPT_UD)
3799                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3800                                                 private_data, private_data_len);
3801                 else
3802                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3803                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3804                                              0, private_data, private_data_len);
3805         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3806                 ret = iw_cm_reject(id_priv->cm_id.iw,
3807                                    private_data, private_data_len);
3808         } else
3809                 ret = -ENOSYS;
3810
3811         return ret;
3812 }
3813 EXPORT_SYMBOL(rdma_reject);
3814
3815 int rdma_disconnect(struct rdma_cm_id *id)
3816 {
3817         struct rdma_id_private *id_priv;
3818         int ret;
3819
3820         id_priv = container_of(id, struct rdma_id_private, id);
3821         if (!id_priv->cm_id.ib)
3822                 return -EINVAL;
3823
3824         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3825                 ret = cma_modify_qp_err(id_priv);
3826                 if (ret)
3827                         goto out;
3828                 /* Initiate or respond to a disconnect. */
3829                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3830                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3831         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3832                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3833         } else
3834                 ret = -EINVAL;
3835
3836 out:
3837         return ret;
3838 }
3839 EXPORT_SYMBOL(rdma_disconnect);
3840
3841 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3842 {
3843         struct rdma_id_private *id_priv;
3844         struct cma_multicast *mc = multicast->context;
3845         struct rdma_cm_event event;
3846         int ret = 0;
3847
3848         id_priv = mc->id_priv;
3849         mutex_lock(&id_priv->handler_mutex);
3850         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3851             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3852                 goto out;
3853
3854         if (!status)
3855                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3856         else
3857                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3858                                      status);
3859         mutex_lock(&id_priv->qp_mutex);
3860         if (!status && id_priv->id.qp) {
3861                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3862                                          be16_to_cpu(multicast->rec.mlid));
3863                 if (status)
3864                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3865                                              status);
3866         }
3867         mutex_unlock(&id_priv->qp_mutex);
3868
3869         memset(&event, 0, sizeof event);
3870         event.status = status;
3871         event.param.ud.private_data = mc->context;
3872         if (!status) {
3873                 struct rdma_dev_addr *dev_addr =
3874                         &id_priv->id.route.addr.dev_addr;
3875                 struct net_device *ndev =
3876                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3877                 enum ib_gid_type gid_type =
3878                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3879                         rdma_start_port(id_priv->cma_dev->device)];
3880
3881                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3882                 ib_init_ah_from_mcmember(id_priv->id.device,
3883                                          id_priv->id.port_num, &multicast->rec,
3884                                          ndev, gid_type,
3885                                          &event.param.ud.ah_attr);
3886                 event.param.ud.qp_num = 0xFFFFFF;
3887                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3888                 if (ndev)
3889                         dev_put(ndev);
3890         } else
3891                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3892
3893         ret = id_priv->id.event_handler(&id_priv->id, &event);
3894         if (ret) {
3895                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3896                 mutex_unlock(&id_priv->handler_mutex);
3897                 rdma_destroy_id(&id_priv->id);
3898                 return 0;
3899         }
3900
3901 out:
3902         mutex_unlock(&id_priv->handler_mutex);
3903         return 0;
3904 }
3905
3906 static void cma_set_mgid(struct rdma_id_private *id_priv,
3907                          struct sockaddr *addr, union ib_gid *mgid)
3908 {
3909         unsigned char mc_map[MAX_ADDR_LEN];
3910         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3911         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3912         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3913
3914         if (cma_any_addr(addr)) {
3915                 memset(mgid, 0, sizeof *mgid);
3916         } else if ((addr->sa_family == AF_INET6) &&
3917                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3918                                                                  0xFF10A01B)) {
3919                 /* IPv6 address is an SA assigned MGID. */
3920                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3921         } else if (addr->sa_family == AF_IB) {
3922                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3923         } else if ((addr->sa_family == AF_INET6)) {
3924                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3925                 if (id_priv->id.ps == RDMA_PS_UDP)
3926                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3927                 *mgid = *(union ib_gid *) (mc_map + 4);
3928         } else {
3929                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3930                 if (id_priv->id.ps == RDMA_PS_UDP)
3931                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3932                 *mgid = *(union ib_gid *) (mc_map + 4);
3933         }
3934 }
3935
3936 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3937                                  struct cma_multicast *mc)
3938 {
3939         struct ib_sa_mcmember_rec rec;
3940         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3941         ib_sa_comp_mask comp_mask;
3942         int ret;
3943
3944         ib_addr_get_mgid(dev_addr, &rec.mgid);
3945         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3946                                      &rec.mgid, &rec);
3947         if (ret)
3948                 return ret;
3949
3950         ret = cma_set_qkey(id_priv, 0);
3951         if (ret)
3952                 return ret;
3953
3954         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3955         rec.qkey = cpu_to_be32(id_priv->qkey);
3956         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3957         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3958         rec.join_state = mc->join_state;
3959
3960         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3961             (!ib_sa_sendonly_fullmem_support(&sa_client,
3962                                              id_priv->id.device,
3963                                              id_priv->id.port_num))) {
3964                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3965                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
3966                         id_priv->id.device->name, id_priv->id.port_num);
3967                 return -EOPNOTSUPP;
3968         }
3969
3970         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3971                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3972                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3973                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3974                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3975
3976         if (id_priv->id.ps == RDMA_PS_IPOIB)
3977                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3978                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3979                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3980                              IB_SA_MCMEMBER_REC_MTU |
3981                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3982
3983         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3984                                                 id_priv->id.port_num, &rec,
3985                                                 comp_mask, GFP_KERNEL,
3986                                                 cma_ib_mc_handler, mc);
3987         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3988 }
3989
3990 static void iboe_mcast_work_handler(struct work_struct *work)
3991 {
3992         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3993         struct cma_multicast *mc = mw->mc;
3994         struct ib_sa_multicast *m = mc->multicast.ib;
3995
3996         mc->multicast.ib->context = mc;
3997         cma_ib_mc_handler(0, m);
3998         kref_put(&mc->mcref, release_mc);
3999         kfree(mw);
4000 }
4001
4002 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4003                               enum ib_gid_type gid_type)
4004 {
4005         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4006         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4007
4008         if (cma_any_addr(addr)) {
4009                 memset(mgid, 0, sizeof *mgid);
4010         } else if (addr->sa_family == AF_INET6) {
4011                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4012         } else {
4013                 mgid->raw[0] = (gid_type == IB_GID_TYPE_IB) ? 0xff : 0;
4014                 mgid->raw[1] = (gid_type == IB_GID_TYPE_IB) ? 0x0e : 0;
4015                 mgid->raw[2] = 0;
4016                 mgid->raw[3] = 0;
4017                 mgid->raw[4] = 0;
4018                 mgid->raw[5] = 0;
4019                 mgid->raw[6] = 0;
4020                 mgid->raw[7] = 0;
4021                 mgid->raw[8] = 0;
4022                 mgid->raw[9] = 0;
4023                 mgid->raw[10] = 0xff;
4024                 mgid->raw[11] = 0xff;
4025                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4026         }
4027 }
4028
4029 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4030                                    struct cma_multicast *mc)
4031 {
4032         struct iboe_mcast_work *work;
4033         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4034         int err = 0;
4035         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4036         struct net_device *ndev = NULL;
4037         enum ib_gid_type gid_type;
4038         bool send_only;
4039
4040         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4041
4042         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4043                 return -EINVAL;
4044
4045         work = kzalloc(sizeof *work, GFP_KERNEL);
4046         if (!work)
4047                 return -ENOMEM;
4048
4049         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4050         if (!mc->multicast.ib) {
4051                 err = -ENOMEM;
4052                 goto out1;
4053         }
4054
4055         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4056                    rdma_start_port(id_priv->cma_dev->device)];
4057         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4058
4059         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4060         if (id_priv->id.ps == RDMA_PS_UDP)
4061                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4062
4063         if (dev_addr->bound_dev_if)
4064                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4065         if (!ndev) {
4066                 err = -ENODEV;
4067                 goto out2;
4068         }
4069         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4070         mc->multicast.ib->rec.hop_limit = 1;
4071         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4072
4073         if (addr->sa_family == AF_INET) {
4074                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4075                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4076                         if (!send_only) {
4077                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4078                                                     true);
4079                                 if (!err)
4080                                         mc->igmp_joined = true;
4081                         }
4082                 }
4083         } else {
4084                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4085                         err = -ENOTSUPP;
4086         }
4087         dev_put(ndev);
4088         if (err || !mc->multicast.ib->rec.mtu) {
4089                 if (!err)
4090                         err = -EINVAL;
4091                 goto out2;
4092         }
4093         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4094                     &mc->multicast.ib->rec.port_gid);
4095         work->id = id_priv;
4096         work->mc = mc;
4097         INIT_WORK(&work->work, iboe_mcast_work_handler);
4098         kref_get(&mc->mcref);
4099         queue_work(cma_wq, &work->work);
4100
4101         return 0;
4102
4103 out2:
4104         kfree(mc->multicast.ib);
4105 out1:
4106         kfree(work);
4107         return err;
4108 }
4109
4110 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4111                         u8 join_state, void *context)
4112 {
4113         struct rdma_id_private *id_priv;
4114         struct cma_multicast *mc;
4115         int ret;
4116
4117         id_priv = container_of(id, struct rdma_id_private, id);
4118         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4119             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4120                 return -EINVAL;
4121
4122         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4123         if (!mc)
4124                 return -ENOMEM;
4125
4126         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4127         mc->context = context;
4128         mc->id_priv = id_priv;
4129         mc->igmp_joined = false;
4130         mc->join_state = join_state;
4131         spin_lock(&id_priv->lock);
4132         list_add(&mc->list, &id_priv->mc_list);
4133         spin_unlock(&id_priv->lock);
4134
4135         if (rdma_protocol_roce(id->device, id->port_num)) {
4136                 kref_init(&mc->mcref);
4137                 ret = cma_iboe_join_multicast(id_priv, mc);
4138         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4139                 ret = cma_join_ib_multicast(id_priv, mc);
4140         else
4141                 ret = -ENOSYS;
4142
4143         if (ret) {
4144                 spin_lock_irq(&id_priv->lock);
4145                 list_del(&mc->list);
4146                 spin_unlock_irq(&id_priv->lock);
4147                 kfree(mc);
4148         }
4149         return ret;
4150 }
4151 EXPORT_SYMBOL(rdma_join_multicast);
4152
4153 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4154 {
4155         struct rdma_id_private *id_priv;
4156         struct cma_multicast *mc;
4157
4158         id_priv = container_of(id, struct rdma_id_private, id);
4159         spin_lock_irq(&id_priv->lock);
4160         list_for_each_entry(mc, &id_priv->mc_list, list) {
4161                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4162                         list_del(&mc->list);
4163                         spin_unlock_irq(&id_priv->lock);
4164
4165                         if (id->qp)
4166                                 ib_detach_mcast(id->qp,
4167                                                 &mc->multicast.ib->rec.mgid,
4168                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4169
4170                         BUG_ON(id_priv->cma_dev->device != id->device);
4171
4172                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4173                                 ib_sa_free_multicast(mc->multicast.ib);
4174                                 kfree(mc);
4175                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4176                                 if (mc->igmp_joined) {
4177                                         struct rdma_dev_addr *dev_addr =
4178                                                 &id->route.addr.dev_addr;
4179                                         struct net_device *ndev = NULL;
4180
4181                                         if (dev_addr->bound_dev_if)
4182                                                 ndev = dev_get_by_index(&init_net,
4183                                                                         dev_addr->bound_dev_if);
4184                                         if (ndev) {
4185                                                 cma_igmp_send(ndev,
4186                                                               &mc->multicast.ib->rec.mgid,
4187                                                               false);
4188                                                 dev_put(ndev);
4189                                         }
4190                                         mc->igmp_joined = false;
4191                                 }
4192                                 kref_put(&mc->mcref, release_mc);
4193                         }
4194                         return;
4195                 }
4196         }
4197         spin_unlock_irq(&id_priv->lock);
4198 }
4199 EXPORT_SYMBOL(rdma_leave_multicast);
4200
4201 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4202 {
4203         struct rdma_dev_addr *dev_addr;
4204         struct cma_ndev_work *work;
4205
4206         dev_addr = &id_priv->id.route.addr.dev_addr;
4207
4208         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4209             (net_eq(dev_net(ndev), dev_addr->net)) &&
4210             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4211                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4212                         ndev->name, &id_priv->id);
4213                 work = kzalloc(sizeof *work, GFP_KERNEL);
4214                 if (!work)
4215                         return -ENOMEM;
4216
4217                 INIT_WORK(&work->work, cma_ndev_work_handler);
4218                 work->id = id_priv;
4219                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4220                 atomic_inc(&id_priv->refcount);
4221                 queue_work(cma_wq, &work->work);
4222         }
4223
4224         return 0;
4225 }
4226
4227 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4228                                void *ptr)
4229 {
4230         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4231         struct cma_device *cma_dev;
4232         struct rdma_id_private *id_priv;
4233         int ret = NOTIFY_DONE;
4234
4235         if (event != NETDEV_BONDING_FAILOVER)
4236                 return NOTIFY_DONE;
4237
4238         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4239                 return NOTIFY_DONE;
4240
4241         mutex_lock(&lock);
4242         list_for_each_entry(cma_dev, &dev_list, list)
4243                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4244                         ret = cma_netdev_change(ndev, id_priv);
4245                         if (ret)
4246                                 goto out;
4247                 }
4248
4249 out:
4250         mutex_unlock(&lock);
4251         return ret;
4252 }
4253
4254 static struct notifier_block cma_nb = {
4255         .notifier_call = cma_netdev_callback
4256 };
4257
4258 static void cma_add_one(struct ib_device *device)
4259 {
4260         struct cma_device *cma_dev;
4261         struct rdma_id_private *id_priv;
4262         unsigned int i;
4263         unsigned long supported_gids = 0;
4264
4265         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4266         if (!cma_dev)
4267                 return;
4268
4269         cma_dev->device = device;
4270         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4271                                             sizeof(*cma_dev->default_gid_type),
4272                                             GFP_KERNEL);
4273         if (!cma_dev->default_gid_type)
4274                 goto free_cma_dev;
4275
4276         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4277                                             sizeof(*cma_dev->default_roce_tos),
4278                                             GFP_KERNEL);
4279         if (!cma_dev->default_roce_tos)
4280                 goto free_gid_type;
4281
4282         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4283                 supported_gids = roce_gid_type_mask_support(device, i);
4284                 WARN_ON(!supported_gids);
4285                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4286                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4287                                 CMA_PREFERRED_ROCE_GID_TYPE;
4288                 else
4289                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4290                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4291                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4292         }
4293
4294         init_completion(&cma_dev->comp);
4295         atomic_set(&cma_dev->refcount, 1);
4296         INIT_LIST_HEAD(&cma_dev->id_list);
4297         ib_set_client_data(device, &cma_client, cma_dev);
4298
4299         mutex_lock(&lock);
4300         list_add_tail(&cma_dev->list, &dev_list);
4301         list_for_each_entry(id_priv, &listen_any_list, list)
4302                 cma_listen_on_dev(id_priv, cma_dev);
4303         mutex_unlock(&lock);
4304
4305         return;
4306
4307 free_gid_type:
4308         kfree(cma_dev->default_gid_type);
4309
4310 free_cma_dev:
4311         kfree(cma_dev);
4312
4313         return;
4314 }
4315
4316 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4317 {
4318         struct rdma_cm_event event;
4319         enum rdma_cm_state state;
4320         int ret = 0;
4321
4322         /* Record that we want to remove the device */
4323         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4324         if (state == RDMA_CM_DESTROYING)
4325                 return 0;
4326
4327         cma_cancel_operation(id_priv, state);
4328         mutex_lock(&id_priv->handler_mutex);
4329
4330         /* Check for destruction from another callback. */
4331         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4332                 goto out;
4333
4334         memset(&event, 0, sizeof event);
4335         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4336         ret = id_priv->id.event_handler(&id_priv->id, &event);
4337 out:
4338         mutex_unlock(&id_priv->handler_mutex);
4339         return ret;
4340 }
4341
4342 static void cma_process_remove(struct cma_device *cma_dev)
4343 {
4344         struct rdma_id_private *id_priv;
4345         int ret;
4346
4347         mutex_lock(&lock);
4348         while (!list_empty(&cma_dev->id_list)) {
4349                 id_priv = list_entry(cma_dev->id_list.next,
4350                                      struct rdma_id_private, list);
4351
4352                 list_del(&id_priv->listen_list);
4353                 list_del_init(&id_priv->list);
4354                 atomic_inc(&id_priv->refcount);
4355                 mutex_unlock(&lock);
4356
4357                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4358                 cma_deref_id(id_priv);
4359                 if (ret)
4360                         rdma_destroy_id(&id_priv->id);
4361
4362                 mutex_lock(&lock);
4363         }
4364         mutex_unlock(&lock);
4365
4366         cma_deref_dev(cma_dev);
4367         wait_for_completion(&cma_dev->comp);
4368 }
4369
4370 static void cma_remove_one(struct ib_device *device, void *client_data)
4371 {
4372         struct cma_device *cma_dev = client_data;
4373
4374         if (!cma_dev)
4375                 return;
4376
4377         mutex_lock(&lock);
4378         list_del(&cma_dev->list);
4379         mutex_unlock(&lock);
4380
4381         cma_process_remove(cma_dev);
4382         kfree(cma_dev->default_roce_tos);
4383         kfree(cma_dev->default_gid_type);
4384         kfree(cma_dev);
4385 }
4386
4387 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4388 {
4389         struct nlmsghdr *nlh;
4390         struct rdma_cm_id_stats *id_stats;
4391         struct rdma_id_private *id_priv;
4392         struct rdma_cm_id *id = NULL;
4393         struct cma_device *cma_dev;
4394         int i_dev = 0, i_id = 0;
4395
4396         /*
4397          * We export all of the IDs as a sequence of messages.  Each
4398          * ID gets its own netlink message.
4399          */
4400         mutex_lock(&lock);
4401
4402         list_for_each_entry(cma_dev, &dev_list, list) {
4403                 if (i_dev < cb->args[0]) {
4404                         i_dev++;
4405                         continue;
4406                 }
4407
4408                 i_id = 0;
4409                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4410                         if (i_id < cb->args[1]) {
4411                                 i_id++;
4412                                 continue;
4413                         }
4414
4415                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4416                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4417                                                 RDMA_NL_RDMA_CM_ID_STATS,
4418                                                 NLM_F_MULTI);
4419                         if (!id_stats)
4420                                 goto out;
4421
4422                         memset(id_stats, 0, sizeof *id_stats);
4423                         id = &id_priv->id;
4424                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4425                         id_stats->port_num = id->port_num;
4426                         id_stats->bound_dev_if =
4427                                 id->route.addr.dev_addr.bound_dev_if;
4428
4429                         if (ibnl_put_attr(skb, nlh,
4430                                           rdma_addr_size(cma_src_addr(id_priv)),
4431                                           cma_src_addr(id_priv),
4432                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4433                                 goto out;
4434                         if (ibnl_put_attr(skb, nlh,
4435                                           rdma_addr_size(cma_src_addr(id_priv)),
4436                                           cma_dst_addr(id_priv),
4437                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4438                                 goto out;
4439
4440                         id_stats->pid           = id_priv->owner;
4441                         id_stats->port_space    = id->ps;
4442                         id_stats->cm_state      = id_priv->state;
4443                         id_stats->qp_num        = id_priv->qp_num;
4444                         id_stats->qp_type       = id->qp_type;
4445
4446                         i_id++;
4447                 }
4448
4449                 cb->args[1] = 0;
4450                 i_dev++;
4451         }
4452
4453 out:
4454         mutex_unlock(&lock);
4455         cb->args[0] = i_dev;
4456         cb->args[1] = i_id;
4457
4458         return skb->len;
4459 }
4460
4461 static const struct rdma_nl_cbs cma_cb_table[] = {
4462         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4463 };
4464
4465 static int cma_init_net(struct net *net)
4466 {
4467         struct cma_pernet *pernet = cma_pernet(net);
4468
4469         idr_init(&pernet->tcp_ps);
4470         idr_init(&pernet->udp_ps);
4471         idr_init(&pernet->ipoib_ps);
4472         idr_init(&pernet->ib_ps);
4473
4474         return 0;
4475 }
4476
4477 static void cma_exit_net(struct net *net)
4478 {
4479         struct cma_pernet *pernet = cma_pernet(net);
4480
4481         idr_destroy(&pernet->tcp_ps);
4482         idr_destroy(&pernet->udp_ps);
4483         idr_destroy(&pernet->ipoib_ps);
4484         idr_destroy(&pernet->ib_ps);
4485 }
4486
4487 static struct pernet_operations cma_pernet_operations = {
4488         .init = cma_init_net,
4489         .exit = cma_exit_net,
4490         .id = &cma_pernet_id,
4491         .size = sizeof(struct cma_pernet),
4492 };
4493
4494 static int __init cma_init(void)
4495 {
4496         int ret;
4497
4498         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4499         if (!cma_wq)
4500                 return -ENOMEM;
4501
4502         ret = register_pernet_subsys(&cma_pernet_operations);
4503         if (ret)
4504                 goto err_wq;
4505
4506         ib_sa_register_client(&sa_client);
4507         rdma_addr_register_client(&addr_client);
4508         register_netdevice_notifier(&cma_nb);
4509
4510         ret = ib_register_client(&cma_client);
4511         if (ret)
4512                 goto err;
4513
4514         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4515         cma_configfs_init();
4516
4517         return 0;
4518
4519 err:
4520         unregister_netdevice_notifier(&cma_nb);
4521         rdma_addr_unregister_client(&addr_client);
4522         ib_sa_unregister_client(&sa_client);
4523 err_wq:
4524         destroy_workqueue(cma_wq);
4525         return ret;
4526 }
4527
4528 static void __exit cma_cleanup(void)
4529 {
4530         cma_configfs_exit();
4531         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4532         ib_unregister_client(&cma_client);
4533         unregister_netdevice_notifier(&cma_nb);
4534         rdma_addr_unregister_client(&addr_client);
4535         ib_sa_unregister_client(&sa_client);
4536         unregister_pernet_subsys(&cma_pernet_operations);
4537         destroy_workqueue(cma_wq);
4538 }
4539
4540 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4541
4542 module_init(cma_init);
4543 module_exit(cma_cleanup);