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 #include "cma_priv.h"
  66
  67 MODULE_AUTHOR("Sean Hefty");
  68 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  69 MODULE_LICENSE("Dual BSD/GPL");
  70
  71 #define CMA_CM_RESPONSE_TIMEOUT 20
  72 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  73 #define CMA_MAX_CM_RETRIES 15
  74 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  75 #define CMA_IBOE_PACKET_LIFETIME 18
  76 #define CMA_PREFERRED_ROCE_GID_TYPE IB_GID_TYPE_ROCE_UDP_ENCAP
  77
  78 static const char * const cma_events[] = {
  79         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  80         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  81         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  82         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  83         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  84         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  85         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  86         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  87         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  88         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  89         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  90         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  91         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  92         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  93         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  94         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  95 };
  96
  97 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  98 {
  99         size_t index = event;
 100
 101         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 102                         cma_events[index] : "unrecognized event";
 103 }
 104 EXPORT_SYMBOL(rdma_event_msg);
 105
 106 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 107                                                 int reason)
 108 {
 109         if (rdma_ib_or_roce(id->device, id->port_num))
 110                 return ibcm_reject_msg(reason);
 111
 112         if (rdma_protocol_iwarp(id->device, id->port_num))
 113                 return iwcm_reject_msg(reason);
 114
 115         WARN_ON_ONCE(1);
 116         return "unrecognized transport";
 117 }
 118 EXPORT_SYMBOL(rdma_reject_msg);
 119
 120 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 121 {
 122         if (rdma_ib_or_roce(id->device, id->port_num))
 123                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 124
 125         if (rdma_protocol_iwarp(id->device, id->port_num))
 126                 return reason == -ECONNREFUSED;
 127
 128         WARN_ON_ONCE(1);
 129         return false;
 130 }
 131 EXPORT_SYMBOL(rdma_is_consumer_reject);
 132
 133 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 134                                       struct rdma_cm_event *ev, u8 *data_len)
 135 {
 136         const void *p;
 137
 138         if (rdma_is_consumer_reject(id, ev->status)) {
 139                 *data_len = ev->param.conn.private_data_len;
 140                 p = ev->param.conn.private_data;
 141         } else {
 142                 *data_len = 0;
 143                 p = NULL;
 144         }
 145         return p;
 146 }
 147 EXPORT_SYMBOL(rdma_consumer_reject_data);
 148
 149 /**
 150  * rdma_iw_cm_id() - return the iw_cm_id pointer for this cm_id.
 151  * @id: Communication Identifier
 152  */
 153 struct iw_cm_id *rdma_iw_cm_id(struct rdma_cm_id *id)
 154 {
 155         struct rdma_id_private *id_priv;
 156
 157         id_priv = container_of(id, struct rdma_id_private, id);
 158         if (id->device->node_type == RDMA_NODE_RNIC)
 159                 return id_priv->cm_id.iw;
 160         return NULL;
 161 }
 162 EXPORT_SYMBOL(rdma_iw_cm_id);
 163
 164 /**
 165  * rdma_res_to_id() - return the rdma_cm_id pointer for this restrack.
 166  * @res: rdma resource tracking entry pointer
 167  */
 168 struct rdma_cm_id *rdma_res_to_id(struct rdma_restrack_entry *res)
 169 {
 170         struct rdma_id_private *id_priv =
 171                 container_of(res, struct rdma_id_private, res);
 172
 173         return &id_priv->id;
 174 }
 175 EXPORT_SYMBOL(rdma_res_to_id);
 176
 177 static void cma_add_one(struct ib_device *device);
 178 static void cma_remove_one(struct ib_device *device, void *client_data);
 179
 180 static struct ib_client cma_client = {
 181         .name   = "cma",
 182         .add    = cma_add_one,
 183         .remove = cma_remove_one
 184 };
 185
 186 static struct ib_sa_client sa_client;
 187 static LIST_HEAD(dev_list);
 188 static LIST_HEAD(listen_any_list);
 189 static DEFINE_MUTEX(lock);
 190 static struct workqueue_struct *cma_wq;
 191 static unsigned int cma_pernet_id;
 192
 193 struct cma_pernet {
 194         struct idr tcp_ps;
 195         struct idr udp_ps;
 196         struct idr ipoib_ps;
 197         struct idr ib_ps;
 198 };
 199
 200 static struct cma_pernet *cma_pernet(struct net *net)
 201 {
 202         return net_generic(net, cma_pernet_id);
 203 }
 204
 205 static struct idr *cma_pernet_idr(struct net *net, enum rdma_ucm_port_space ps)
 206 {
 207         struct cma_pernet *pernet = cma_pernet(net);
 208
 209         switch (ps) {
 210         case RDMA_PS_TCP:
 211                 return &pernet->tcp_ps;
 212         case RDMA_PS_UDP:
 213                 return &pernet->udp_ps;
 214         case RDMA_PS_IPOIB:
 215                 return &pernet->ipoib_ps;
 216         case RDMA_PS_IB:
 217                 return &pernet->ib_ps;
 218         default:
 219                 return NULL;
 220         }
 221 }
 222
 223 struct cma_device {
 224         struct list_head        list;
 225         struct ib_device        *device;
 226         struct completion       comp;
 227         atomic_t                refcount;
 228         struct list_head        id_list;
 229         enum ib_gid_type        *default_gid_type;
 230         u8                      *default_roce_tos;
 231 };
 232
 233 struct rdma_bind_list {
 234         enum rdma_ucm_port_space ps;
 235         struct hlist_head       owners;
 236         unsigned short          port;
 237 };
 238
 239 struct class_port_info_context {
 240         struct ib_class_port_info       *class_port_info;
 241         struct ib_device                *device;
 242         struct completion               done;
 243         struct ib_sa_query              *sa_query;
 244         u8                              port_num;
 245 };
 246
 247 static int cma_ps_alloc(struct net *net, enum rdma_ucm_port_space ps,
 248                         struct rdma_bind_list *bind_list, int snum)
 249 {
 250         struct idr *idr = cma_pernet_idr(net, ps);
 251
 252         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 253 }
 254
 255 static struct rdma_bind_list *cma_ps_find(struct net *net,
 256                                           enum rdma_ucm_port_space ps, int snum)
 257 {
 258         struct idr *idr = cma_pernet_idr(net, ps);
 259
 260         return idr_find(idr, snum);
 261 }
 262
 263 static void cma_ps_remove(struct net *net, enum rdma_ucm_port_space ps,
 264                           int snum)
 265 {
 266         struct idr *idr = cma_pernet_idr(net, ps);
 267
 268         idr_remove(idr, snum);
 269 }
 270
 271 enum {
 272         CMA_OPTION_AFONLY,
 273 };
 274
 275 void cma_ref_dev(struct cma_device *cma_dev)
 276 {
 277         atomic_inc(&cma_dev->refcount);
 278 }
 279
 280 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 281                                              void               *cookie)
 282 {
 283         struct cma_device *cma_dev;
 284         struct cma_device *found_cma_dev = NULL;
 285
 286         mutex_lock(&lock);
 287
 288         list_for_each_entry(cma_dev, &dev_list, list)
 289                 if (filter(cma_dev->device, cookie)) {
 290                         found_cma_dev = cma_dev;
 291                         break;
 292                 }
 293
 294         if (found_cma_dev)
 295                 cma_ref_dev(found_cma_dev);
 296         mutex_unlock(&lock);
 297         return found_cma_dev;
 298 }
 299
 300 int cma_get_default_gid_type(struct cma_device *cma_dev,
 301                              unsigned int port)
 302 {
 303         if (!rdma_is_port_valid(cma_dev->device, port))
 304                 return -EINVAL;
 305
 306         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 307 }
 308
 309 int cma_set_default_gid_type(struct cma_device *cma_dev,
 310                              unsigned int port,
 311                              enum ib_gid_type default_gid_type)
 312 {
 313         unsigned long supported_gids;
 314
 315         if (!rdma_is_port_valid(cma_dev->device, port))
 316                 return -EINVAL;
 317
 318         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 319
 320         if (!(supported_gids & 1 << default_gid_type))
 321                 return -EINVAL;
 322
 323         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 324                 default_gid_type;
 325
 326         return 0;
 327 }
 328
 329 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 330 {
 331         if (!rdma_is_port_valid(cma_dev->device, port))
 332                 return -EINVAL;
 333
 334         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 335 }
 336
 337 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 338                              u8 default_roce_tos)
 339 {
 340         if (!rdma_is_port_valid(cma_dev->device, port))
 341                 return -EINVAL;
 342
 343         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 344                  default_roce_tos;
 345
 346         return 0;
 347 }
 348 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 349 {
 350         return cma_dev->device;
 351 }
 352
 353 /*
 354  * Device removal can occur at anytime, so we need extra handling to
 355  * serialize notifying the user of device removal with other callbacks.
 356  * We do this by disabling removal notification while a callback is in process,
 357  * and reporting it after the callback completes.
 358  */
 359
 360 struct cma_multicast {
 361         struct rdma_id_private *id_priv;
 362         union {
 363                 struct ib_sa_multicast *ib;
 364         } multicast;
 365         struct list_head        list;
 366         void                    *context;
 367         struct sockaddr_storage addr;
 368         struct kref             mcref;
 369         bool                    igmp_joined;
 370         u8                      join_state;
 371 };
 372
 373 struct cma_work {
 374         struct work_struct      work;
 375         struct rdma_id_private  *id;
 376         enum rdma_cm_state      old_state;
 377         enum rdma_cm_state      new_state;
 378         struct rdma_cm_event    event;
 379 };
 380
 381 struct cma_ndev_work {
 382         struct work_struct      work;
 383         struct rdma_id_private  *id;
 384         struct rdma_cm_event    event;
 385 };
 386
 387 struct iboe_mcast_work {
 388         struct work_struct       work;
 389         struct rdma_id_private  *id;
 390         struct cma_multicast    *mc;
 391 };
 392
 393 union cma_ip_addr {
 394         struct in6_addr ip6;
 395         struct {
 396                 __be32 pad[3];
 397                 __be32 addr;
 398         } ip4;
 399 };
 400
 401 struct cma_hdr {
 402         u8 cma_version;
 403         u8 ip_version;  /* IP version: 7:4 */
 404         __be16 port;
 405         union cma_ip_addr src_addr;
 406         union cma_ip_addr dst_addr;
 407 };
 408
 409 #define CMA_VERSION 0x00
 410
 411 struct cma_req_info {
 412         struct sockaddr_storage listen_addr_storage;
 413         struct sockaddr_storage src_addr_storage;
 414         struct ib_device *device;
 415         int port;
 416         union ib_gid local_gid;
 417         __be64 service_id;
 418         u16 pkey;
 419         bool has_gid:1;
 420 };
 421
 422 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 423 {
 424         unsigned long flags;
 425         int ret;
 426
 427         spin_lock_irqsave(&id_priv->lock, flags);
 428         ret = (id_priv->state == comp);
 429         spin_unlock_irqrestore(&id_priv->lock, flags);
 430         return ret;
 431 }
 432
 433 static int cma_comp_exch(struct rdma_id_private *id_priv,
 434                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 435 {
 436         unsigned long flags;
 437         int ret;
 438
 439         spin_lock_irqsave(&id_priv->lock, flags);
 440         if ((ret = (id_priv->state == comp)))
 441                 id_priv->state = exch;
 442         spin_unlock_irqrestore(&id_priv->lock, flags);
 443         return ret;
 444 }
 445
 446 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 447                                    enum rdma_cm_state exch)
 448 {
 449         unsigned long flags;
 450         enum rdma_cm_state old;
 451
 452         spin_lock_irqsave(&id_priv->lock, flags);
 453         old = id_priv->state;
 454         id_priv->state = exch;
 455         spin_unlock_irqrestore(&id_priv->lock, flags);
 456         return old;
 457 }
 458
 459 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 460 {
 461         return hdr->ip_version >> 4;
 462 }
 463
 464 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 465 {
 466         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 467 }
 468
 469 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 470 {
 471         struct in_device *in_dev = NULL;
 472
 473         if (ndev) {
 474                 rtnl_lock();
 475                 in_dev = __in_dev_get_rtnl(ndev);
 476                 if (in_dev) {
 477                         if (join)
 478                                 ip_mc_inc_group(in_dev,
 479                                                 *(__be32 *)(mgid->raw + 12));
 480                         else
 481                                 ip_mc_dec_group(in_dev,
 482                                                 *(__be32 *)(mgid->raw + 12));
 483                 }
 484                 rtnl_unlock();
 485         }
 486         return (in_dev) ? 0 : -ENODEV;
 487 }
 488
 489 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 490                                struct cma_device *cma_dev)
 491 {
 492         cma_ref_dev(cma_dev);
 493         id_priv->cma_dev = cma_dev;
 494         id_priv->gid_type = 0;
 495         id_priv->id.device = cma_dev->device;
 496         id_priv->id.route.addr.dev_addr.transport =
 497                 rdma_node_get_transport(cma_dev->device->node_type);
 498         list_add_tail(&id_priv->list, &cma_dev->id_list);
 499         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 500         rdma_restrack_add(&id_priv->res);
 501 }
 502
 503 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 504                               struct cma_device *cma_dev)
 505 {
 506         _cma_attach_to_dev(id_priv, cma_dev);
 507         id_priv->gid_type =
 508                 cma_dev->default_gid_type[id_priv->id.port_num -
 509                                           rdma_start_port(cma_dev->device)];
 510 }
 511
 512 void cma_deref_dev(struct cma_device *cma_dev)
 513 {
 514         if (atomic_dec_and_test(&cma_dev->refcount))
 515                 complete(&cma_dev->comp);
 516 }
 517
 518 static inline void release_mc(struct kref *kref)
 519 {
 520         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 521
 522         kfree(mc->multicast.ib);
 523         kfree(mc);
 524 }
 525
 526 static void cma_release_dev(struct rdma_id_private *id_priv)
 527 {
 528         mutex_lock(&lock);
 529         list_del(&id_priv->list);
 530         cma_deref_dev(id_priv->cma_dev);
 531         id_priv->cma_dev = NULL;
 532         mutex_unlock(&lock);
 533 }
 534
 535 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 536 {
 537         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 538 }
 539
 540 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 541 {
 542         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 543 }
 544
 545 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 546 {
 547         return id_priv->id.route.addr.src_addr.ss_family;
 548 }
 549
 550 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 551 {
 552         struct ib_sa_mcmember_rec rec;
 553         int ret = 0;
 554
 555         if (id_priv->qkey) {
 556                 if (qkey && id_priv->qkey != qkey)
 557                         return -EINVAL;
 558                 return 0;
 559         }
 560
 561         if (qkey) {
 562                 id_priv->qkey = qkey;
 563                 return 0;
 564         }
 565
 566         switch (id_priv->id.ps) {
 567         case RDMA_PS_UDP:
 568         case RDMA_PS_IB:
 569                 id_priv->qkey = RDMA_UDP_QKEY;
 570                 break;
 571         case RDMA_PS_IPOIB:
 572                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 573                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 574                                              id_priv->id.port_num, &rec.mgid,
 575                                              &rec);
 576                 if (!ret)
 577                         id_priv->qkey = be32_to_cpu(rec.qkey);
 578                 break;
 579         default:
 580                 break;
 581         }
 582         return ret;
 583 }
 584
 585 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 586 {
 587         dev_addr->dev_type = ARPHRD_INFINIBAND;
 588         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 589         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 590 }
 591
 592 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 593 {
 594         int ret;
 595
 596         if (addr->sa_family != AF_IB) {
 597                 ret = rdma_translate_ip(addr, dev_addr);
 598         } else {
 599                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 600                 ret = 0;
 601         }
 602
 603         return ret;
 604 }
 605
 606 static inline int cma_validate_port(struct ib_device *device, u8 port,
 607                                     enum ib_gid_type gid_type,
 608                                     union ib_gid *gid,
 609                                     struct rdma_id_private *id_priv)
 610 {
 611         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 612         int bound_if_index = dev_addr->bound_dev_if;
 613         int dev_type = dev_addr->dev_type;
 614         struct net_device *ndev = NULL;
 615         int ret = -ENODEV;
 616
 617         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 618                 return ret;
 619
 620         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 621                 return ret;
 622
 623         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 624                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 625                 if (!ndev)
 626                         return ret;
 627         } else {
 628                 gid_type = IB_GID_TYPE_IB;
 629         }
 630
 631         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 632                                          ndev, NULL);
 633
 634         if (ndev)
 635                 dev_put(ndev);
 636
 637         return ret;
 638 }
 639
 640 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 641                            struct rdma_id_private *listen_id_priv)
 642 {
 643         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 644         struct cma_device *cma_dev;
 645         union ib_gid gid, iboe_gid, *gidp;
 646         int ret = -ENODEV;
 647         u8 port;
 648
 649         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 650             id_priv->id.ps == RDMA_PS_IPOIB)
 651                 return -EINVAL;
 652
 653         mutex_lock(&lock);
 654         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 655                     &iboe_gid);
 656
 657         memcpy(&gid, dev_addr->src_dev_addr +
 658                rdma_addr_gid_offset(dev_addr), sizeof gid);
 659
 660         if (listen_id_priv) {
 661                 cma_dev = listen_id_priv->cma_dev;
 662                 port = listen_id_priv->id.port_num;
 663                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 664                        &iboe_gid : &gid;
 665
 666                 ret = cma_validate_port(cma_dev->device, port,
 667                                         rdma_protocol_ib(cma_dev->device, port) ?
 668                                         IB_GID_TYPE_IB :
 669                                         listen_id_priv->gid_type, gidp,
 670                                         id_priv);
 671                 if (!ret) {
 672                         id_priv->id.port_num = port;
 673                         goto out;
 674                 }
 675         }
 676
 677         list_for_each_entry(cma_dev, &dev_list, list) {
 678                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 679                         if (listen_id_priv &&
 680                             listen_id_priv->cma_dev == cma_dev &&
 681                             listen_id_priv->id.port_num == port)
 682                                 continue;
 683
 684                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 685                                &iboe_gid : &gid;
 686
 687                         ret = cma_validate_port(cma_dev->device, port,
 688                                                 rdma_protocol_ib(cma_dev->device, port) ?
 689                                                 IB_GID_TYPE_IB :
 690                                                 cma_dev->default_gid_type[port - 1],
 691                                                 gidp, id_priv);
 692                         if (!ret) {
 693                                 id_priv->id.port_num = port;
 694                                 goto out;
 695                         }
 696                 }
 697         }
 698
 699 out:
 700         if (!ret)
 701                 cma_attach_to_dev(id_priv, cma_dev);
 702
 703         mutex_unlock(&lock);
 704         return ret;
 705 }
 706
 707 /*
 708  * Select the source IB device and address to reach the destination IB address.
 709  */
 710 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 711 {
 712         struct cma_device *cma_dev, *cur_dev;
 713         struct sockaddr_ib *addr;
 714         union ib_gid gid, sgid, *dgid;
 715         u16 pkey, index;
 716         u8 p;
 717         enum ib_port_state port_state;
 718         int i;
 719
 720         cma_dev = NULL;
 721         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 722         dgid = (union ib_gid *) &addr->sib_addr;
 723         pkey = ntohs(addr->sib_pkey);
 724
 725         list_for_each_entry(cur_dev, &dev_list, list) {
 726                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 727                         if (!rdma_cap_af_ib(cur_dev->device, p))
 728                                 continue;
 729
 730                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 731                                 continue;
 732
 733                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 734                                 continue;
 735                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 736                                                        &gid, NULL);
 737                              i++) {
 738                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 739                                         cma_dev = cur_dev;
 740                                         sgid = gid;
 741                                         id_priv->id.port_num = p;
 742                                         goto found;
 743                                 }
 744
 745                                 if (!cma_dev && (gid.global.subnet_prefix ==
 746                                     dgid->global.subnet_prefix) &&
 747                                     port_state == IB_PORT_ACTIVE) {
 748                                         cma_dev = cur_dev;
 749                                         sgid = gid;
 750                                         id_priv->id.port_num = p;
 751                                 }
 752                         }
 753                 }
 754         }
 755
 756         if (!cma_dev)
 757                 return -ENODEV;
 758
 759 found:
 760         cma_attach_to_dev(id_priv, cma_dev);
 761         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 762         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 763         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 764         return 0;
 765 }
 766
 767 static void cma_deref_id(struct rdma_id_private *id_priv)
 768 {
 769         if (atomic_dec_and_test(&id_priv->refcount))
 770                 complete(&id_priv->comp);
 771 }
 772
 773 struct rdma_cm_id *__rdma_create_id(struct net *net,
 774                                     rdma_cm_event_handler event_handler,
 775                                     void *context, enum rdma_ucm_port_space ps,
 776                                     enum ib_qp_type qp_type, const char *caller)
 777 {
 778         struct rdma_id_private *id_priv;
 779
 780         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 781         if (!id_priv)
 782                 return ERR_PTR(-ENOMEM);
 783
 784         if (caller)
 785                 id_priv->res.kern_name = caller;
 786         else
 787                 rdma_restrack_set_task(&id_priv->res, current);
 788         id_priv->state = RDMA_CM_IDLE;
 789         id_priv->id.context = context;
 790         id_priv->id.event_handler = event_handler;
 791         id_priv->id.ps = ps;
 792         id_priv->id.qp_type = qp_type;
 793         id_priv->tos_set = false;
 794         spin_lock_init(&id_priv->lock);
 795         mutex_init(&id_priv->qp_mutex);
 796         init_completion(&id_priv->comp);
 797         atomic_set(&id_priv->refcount, 1);
 798         mutex_init(&id_priv->handler_mutex);
 799         INIT_LIST_HEAD(&id_priv->listen_list);
 800         INIT_LIST_HEAD(&id_priv->mc_list);
 801         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 802         id_priv->id.route.addr.dev_addr.net = get_net(net);
 803         id_priv->seq_num &= 0x00ffffff;
 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
 899         mutex_lock(&id_priv->qp_mutex);
 900         if (!id_priv->id.qp) {
 901                 ret = 0;
 902                 goto out;
 903         }
 904
 905         /* Need to update QP attributes from default values. */
 906         qp_attr.qp_state = IB_QPS_INIT;
 907         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 908         if (ret)
 909                 goto out;
 910
 911         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 912         if (ret)
 913                 goto out;
 914
 915         qp_attr.qp_state = IB_QPS_RTR;
 916         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 917         if (ret)
 918                 goto out;
 919
 920         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 921
 922         if (conn_param)
 923                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 924         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 925 out:
 926         mutex_unlock(&id_priv->qp_mutex);
 927         return ret;
 928 }
 929
 930 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 931                              struct rdma_conn_param *conn_param)
 932 {
 933         struct ib_qp_attr qp_attr;
 934         int qp_attr_mask, ret;
 935
 936         mutex_lock(&id_priv->qp_mutex);
 937         if (!id_priv->id.qp) {
 938                 ret = 0;
 939                 goto out;
 940         }
 941
 942         qp_attr.qp_state = IB_QPS_RTS;
 943         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 944         if (ret)
 945                 goto out;
 946
 947         if (conn_param)
 948                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 949         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 950 out:
 951         mutex_unlock(&id_priv->qp_mutex);
 952         return ret;
 953 }
 954
 955 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 956 {
 957         struct ib_qp_attr qp_attr;
 958         int ret;
 959
 960         mutex_lock(&id_priv->qp_mutex);
 961         if (!id_priv->id.qp) {
 962                 ret = 0;
 963                 goto out;
 964         }
 965
 966         qp_attr.qp_state = IB_QPS_ERR;
 967         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 968 out:
 969         mutex_unlock(&id_priv->qp_mutex);
 970         return ret;
 971 }
 972
 973 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 974                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 975 {
 976         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 977         int ret;
 978         u16 pkey;
 979
 980         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 981                 pkey = 0xffff;
 982         else
 983                 pkey = ib_addr_get_pkey(dev_addr);
 984
 985         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 986                                   pkey, &qp_attr->pkey_index);
 987         if (ret)
 988                 return ret;
 989
 990         qp_attr->port_num = id_priv->id.port_num;
 991         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 992
 993         if (id_priv->id.qp_type == IB_QPT_UD) {
 994                 ret = cma_set_qkey(id_priv, 0);
 995                 if (ret)
 996                         return ret;
 997
 998                 qp_attr->qkey = id_priv->qkey;
 999                 *qp_attr_mask |= IB_QP_QKEY;
1000         } else {
1001                 qp_attr->qp_access_flags = 0;
1002                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1003         }
1004         return 0;
1005 }
1006
1007 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1008                        int *qp_attr_mask)
1009 {
1010         struct rdma_id_private *id_priv;
1011         int ret = 0;
1012
1013         id_priv = container_of(id, struct rdma_id_private, id);
1014         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1015                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1016                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1017                 else
1018                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1019                                                  qp_attr_mask);
1020
1021                 if (qp_attr->qp_state == IB_QPS_RTR)
1022                         qp_attr->rq_psn = id_priv->seq_num;
1023         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1024                 if (!id_priv->cm_id.iw) {
1025                         qp_attr->qp_access_flags = 0;
1026                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1027                 } else
1028                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1029                                                  qp_attr_mask);
1030                 qp_attr->port_num = id_priv->id.port_num;
1031                 *qp_attr_mask |= IB_QP_PORT;
1032         } else
1033                 ret = -ENOSYS;
1034
1035         return ret;
1036 }
1037 EXPORT_SYMBOL(rdma_init_qp_attr);
1038
1039 static inline int cma_zero_addr(struct sockaddr *addr)
1040 {
1041         switch (addr->sa_family) {
1042         case AF_INET:
1043                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1044         case AF_INET6:
1045                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1046         case AF_IB:
1047                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1048         default:
1049                 return 0;
1050         }
1051 }
1052
1053 static inline int cma_loopback_addr(struct sockaddr *addr)
1054 {
1055         switch (addr->sa_family) {
1056         case AF_INET:
1057                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1058         case AF_INET6:
1059                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1060         case AF_IB:
1061                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1062         default:
1063                 return 0;
1064         }
1065 }
1066
1067 static inline int cma_any_addr(struct sockaddr *addr)
1068 {
1069         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1070 }
1071
1072 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1073 {
1074         if (src->sa_family != dst->sa_family)
1075                 return -1;
1076
1077         switch (src->sa_family) {
1078         case AF_INET:
1079                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1080                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1081         case AF_INET6:
1082                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1083                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1084         default:
1085                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1086                                    &((struct sockaddr_ib *) dst)->sib_addr);
1087         }
1088 }
1089
1090 static __be16 cma_port(struct sockaddr *addr)
1091 {
1092         struct sockaddr_ib *sib;
1093
1094         switch (addr->sa_family) {
1095         case AF_INET:
1096                 return ((struct sockaddr_in *) addr)->sin_port;
1097         case AF_INET6:
1098                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1099         case AF_IB:
1100                 sib = (struct sockaddr_ib *) addr;
1101                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1102                                     be64_to_cpu(sib->sib_sid_mask)));
1103         default:
1104                 return 0;
1105         }
1106 }
1107
1108 static inline int cma_any_port(struct sockaddr *addr)
1109 {
1110         return !cma_port(addr);
1111 }
1112
1113 static void cma_save_ib_info(struct sockaddr *src_addr,
1114                              struct sockaddr *dst_addr,
1115                              struct rdma_cm_id *listen_id,
1116                              struct sa_path_rec *path)
1117 {
1118         struct sockaddr_ib *listen_ib, *ib;
1119
1120         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1121         if (src_addr) {
1122                 ib = (struct sockaddr_ib *)src_addr;
1123                 ib->sib_family = AF_IB;
1124                 if (path) {
1125                         ib->sib_pkey = path->pkey;
1126                         ib->sib_flowinfo = path->flow_label;
1127                         memcpy(&ib->sib_addr, &path->sgid, 16);
1128                         ib->sib_sid = path->service_id;
1129                         ib->sib_scope_id = 0;
1130                 } else {
1131                         ib->sib_pkey = listen_ib->sib_pkey;
1132                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1133                         ib->sib_addr = listen_ib->sib_addr;
1134                         ib->sib_sid = listen_ib->sib_sid;
1135                         ib->sib_scope_id = listen_ib->sib_scope_id;
1136                 }
1137                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1138         }
1139         if (dst_addr) {
1140                 ib = (struct sockaddr_ib *)dst_addr;
1141                 ib->sib_family = AF_IB;
1142                 if (path) {
1143                         ib->sib_pkey = path->pkey;
1144                         ib->sib_flowinfo = path->flow_label;
1145                         memcpy(&ib->sib_addr, &path->dgid, 16);
1146                 }
1147         }
1148 }
1149
1150 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1151                               struct sockaddr_in *dst_addr,
1152                               struct cma_hdr *hdr,
1153                               __be16 local_port)
1154 {
1155         if (src_addr) {
1156                 *src_addr = (struct sockaddr_in) {
1157                         .sin_family = AF_INET,
1158                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1159                         .sin_port = local_port,
1160                 };
1161         }
1162
1163         if (dst_addr) {
1164                 *dst_addr = (struct sockaddr_in) {
1165                         .sin_family = AF_INET,
1166                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1167                         .sin_port = hdr->port,
1168                 };
1169         }
1170 }
1171
1172 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1173                               struct sockaddr_in6 *dst_addr,
1174                               struct cma_hdr *hdr,
1175                               __be16 local_port)
1176 {
1177         if (src_addr) {
1178                 *src_addr = (struct sockaddr_in6) {
1179                         .sin6_family = AF_INET6,
1180                         .sin6_addr = hdr->dst_addr.ip6,
1181                         .sin6_port = local_port,
1182                 };
1183         }
1184
1185         if (dst_addr) {
1186                 *dst_addr = (struct sockaddr_in6) {
1187                         .sin6_family = AF_INET6,
1188                         .sin6_addr = hdr->src_addr.ip6,
1189                         .sin6_port = hdr->port,
1190                 };
1191         }
1192 }
1193
1194 static u16 cma_port_from_service_id(__be64 service_id)
1195 {
1196         return (u16)be64_to_cpu(service_id);
1197 }
1198
1199 static int cma_save_ip_info(struct sockaddr *src_addr,
1200                             struct sockaddr *dst_addr,
1201                             struct ib_cm_event *ib_event,
1202                             __be64 service_id)
1203 {
1204         struct cma_hdr *hdr;
1205         __be16 port;
1206
1207         hdr = ib_event->private_data;
1208         if (hdr->cma_version != CMA_VERSION)
1209                 return -EINVAL;
1210
1211         port = htons(cma_port_from_service_id(service_id));
1212
1213         switch (cma_get_ip_ver(hdr)) {
1214         case 4:
1215                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1216                                   (struct sockaddr_in *)dst_addr, hdr, port);
1217                 break;
1218         case 6:
1219                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1220                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1221                 break;
1222         default:
1223                 return -EAFNOSUPPORT;
1224         }
1225
1226         return 0;
1227 }
1228
1229 static int cma_save_net_info(struct sockaddr *src_addr,
1230                              struct sockaddr *dst_addr,
1231                              struct rdma_cm_id *listen_id,
1232                              struct ib_cm_event *ib_event,
1233                              sa_family_t sa_family, __be64 service_id)
1234 {
1235         if (sa_family == AF_IB) {
1236                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1237                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1238                                          ib_event->param.req_rcvd.primary_path);
1239                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1240                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1241                 return 0;
1242         }
1243
1244         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1245 }
1246
1247 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1248                              struct cma_req_info *req)
1249 {
1250         const struct ib_cm_req_event_param *req_param =
1251                 &ib_event->param.req_rcvd;
1252         const struct ib_cm_sidr_req_event_param *sidr_param =
1253                 &ib_event->param.sidr_req_rcvd;
1254
1255         switch (ib_event->event) {
1256         case IB_CM_REQ_RECEIVED:
1257                 req->device     = req_param->listen_id->device;
1258                 req->port       = req_param->port;
1259                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1260                        sizeof(req->local_gid));
1261                 req->has_gid    = true;
1262                 req->service_id = req_param->primary_path->service_id;
1263                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1264                 if (req->pkey != req_param->bth_pkey)
1265                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1266                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1267                                             req_param->bth_pkey, req->pkey);
1268                 break;
1269         case IB_CM_SIDR_REQ_RECEIVED:
1270                 req->device     = sidr_param->listen_id->device;
1271                 req->port       = sidr_param->port;
1272                 req->has_gid    = false;
1273                 req->service_id = sidr_param->service_id;
1274                 req->pkey       = sidr_param->pkey;
1275                 if (req->pkey != sidr_param->bth_pkey)
1276                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1277                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1278                                             sidr_param->bth_pkey, req->pkey);
1279                 break;
1280         default:
1281                 return -EINVAL;
1282         }
1283
1284         return 0;
1285 }
1286
1287 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1288                                   const struct sockaddr_in *dst_addr,
1289                                   const struct sockaddr_in *src_addr)
1290 {
1291         __be32 daddr = dst_addr->sin_addr.s_addr,
1292                saddr = src_addr->sin_addr.s_addr;
1293         struct fib_result res;
1294         struct flowi4 fl4;
1295         int err;
1296         bool ret;
1297
1298         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1299             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1300             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1301             ipv4_is_loopback(saddr))
1302                 return false;
1303
1304         memset(&fl4, 0, sizeof(fl4));
1305         fl4.flowi4_iif = net_dev->ifindex;
1306         fl4.daddr = daddr;
1307         fl4.saddr = saddr;
1308
1309         rcu_read_lock();
1310         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1311         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1312         rcu_read_unlock();
1313
1314         return ret;
1315 }
1316
1317 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1318                                   const struct sockaddr_in6 *dst_addr,
1319                                   const struct sockaddr_in6 *src_addr)
1320 {
1321 #if IS_ENABLED(CONFIG_IPV6)
1322         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1323                            IPV6_ADDR_LINKLOCAL;
1324         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1325                                          &src_addr->sin6_addr, net_dev->ifindex,
1326                                          NULL, strict);
1327         bool ret;
1328
1329         if (!rt)
1330                 return false;
1331
1332         ret = rt->rt6i_idev->dev == net_dev;
1333         ip6_rt_put(rt);
1334
1335         return ret;
1336 #else
1337         return false;
1338 #endif
1339 }
1340
1341 static bool validate_net_dev(struct net_device *net_dev,
1342                              const struct sockaddr *daddr,
1343                              const struct sockaddr *saddr)
1344 {
1345         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1346         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1347         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1348         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1349
1350         switch (daddr->sa_family) {
1351         case AF_INET:
1352                 return saddr->sa_family == AF_INET &&
1353                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1354
1355         case AF_INET6:
1356                 return saddr->sa_family == AF_INET6 &&
1357                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1358
1359         default:
1360                 return false;
1361         }
1362 }
1363
1364 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1365                                           struct cma_req_info *req)
1366 {
1367         struct sockaddr *listen_addr =
1368                         (struct sockaddr *)&req->listen_addr_storage;
1369         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1370         struct net_device *net_dev;
1371         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1372         int err;
1373
1374         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1375                                req->service_id);
1376         if (err)
1377                 return ERR_PTR(err);
1378
1379         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1380                                            gid, listen_addr);
1381         if (!net_dev)
1382                 return ERR_PTR(-ENODEV);
1383
1384         return net_dev;
1385 }
1386
1387 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1388 {
1389         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1390 }
1391
1392 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1393                                    const struct cma_hdr *hdr)
1394 {
1395         struct sockaddr *addr = cma_src_addr(id_priv);
1396         __be32 ip4_addr;
1397         struct in6_addr ip6_addr;
1398
1399         if (cma_any_addr(addr) && !id_priv->afonly)
1400                 return true;
1401
1402         switch (addr->sa_family) {
1403         case AF_INET:
1404                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1405                 if (cma_get_ip_ver(hdr) != 4)
1406                         return false;
1407                 if (!cma_any_addr(addr) &&
1408                     hdr->dst_addr.ip4.addr != ip4_addr)
1409                         return false;
1410                 break;
1411         case AF_INET6:
1412                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1413                 if (cma_get_ip_ver(hdr) != 6)
1414                         return false;
1415                 if (!cma_any_addr(addr) &&
1416                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1417                         return false;
1418                 break;
1419         case AF_IB:
1420                 return true;
1421         default:
1422                 return false;
1423         }
1424
1425         return true;
1426 }
1427
1428 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1429 {
1430         struct ib_device *device = id->device;
1431         const int port_num = id->port_num ?: rdma_start_port(device);
1432
1433         return rdma_protocol_roce(device, port_num);
1434 }
1435
1436 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1437                               const struct net_device *net_dev,
1438                               u8 port_num)
1439 {
1440         const struct rdma_addr *addr = &id->route.addr;
1441
1442         if (!net_dev)
1443                 /* This request is an AF_IB request or a RoCE request */
1444                 return (!id->port_num || id->port_num == port_num) &&
1445                        (addr->src_addr.ss_family == AF_IB ||
1446                         rdma_protocol_roce(id->device, port_num));
1447
1448         return !addr->dev_addr.bound_dev_if ||
1449                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1450                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1451 }
1452
1453 static struct rdma_id_private *cma_find_listener(
1454                 const struct rdma_bind_list *bind_list,
1455                 const struct ib_cm_id *cm_id,
1456                 const struct ib_cm_event *ib_event,
1457                 const struct cma_req_info *req,
1458                 const struct net_device *net_dev)
1459 {
1460         struct rdma_id_private *id_priv, *id_priv_dev;
1461
1462         if (!bind_list)
1463                 return ERR_PTR(-EINVAL);
1464
1465         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1466                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1467                         if (id_priv->id.device == cm_id->device &&
1468                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1469                                 return id_priv;
1470                         list_for_each_entry(id_priv_dev,
1471                                             &id_priv->listen_list,
1472                                             listen_list) {
1473                                 if (id_priv_dev->id.device == cm_id->device &&
1474                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1475                                         return id_priv_dev;
1476                         }
1477                 }
1478         }
1479
1480         return ERR_PTR(-EINVAL);
1481 }
1482
1483 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1484                                                  struct ib_cm_event *ib_event,
1485                                                  struct net_device **net_dev)
1486 {
1487         struct cma_req_info req;
1488         struct rdma_bind_list *bind_list;
1489         struct rdma_id_private *id_priv;
1490         int err;
1491
1492         err = cma_save_req_info(ib_event, &req);
1493         if (err)
1494                 return ERR_PTR(err);
1495
1496         *net_dev = cma_get_net_dev(ib_event, &req);
1497         if (IS_ERR(*net_dev)) {
1498                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1499                         /* Assuming the protocol is AF_IB */
1500                         *net_dev = NULL;
1501                 } else if (rdma_protocol_roce(req.device, req.port)) {
1502                         /* TODO find the net dev matching the request parameters
1503                          * through the RoCE GID table */
1504                         *net_dev = NULL;
1505                 } else {
1506                         return ERR_CAST(*net_dev);
1507                 }
1508         }
1509
1510         /*
1511          * Net namespace might be getting deleted while route lookup,
1512          * cm_id lookup is in progress. Therefore, perform netdevice
1513          * validation, cm_id lookup under rcu lock.
1514          * RCU lock along with netdevice state check, synchronizes with
1515          * netdevice migrating to different net namespace and also avoids
1516          * case where net namespace doesn't get deleted while lookup is in
1517          * progress.
1518          * If the device state is not IFF_UP, its properties such as ifindex
1519          * and nd_net cannot be trusted to remain valid without rcu lock.
1520          * net/core/dev.c change_net_namespace() ensures to synchronize with
1521          * ongoing operations on net device after device is closed using
1522          * synchronize_net().
1523          */
1524         rcu_read_lock();
1525         if (*net_dev) {
1526                 /*
1527                  * If netdevice is down, it is likely that it is administratively
1528                  * down or it might be migrating to different namespace.
1529                  * In that case avoid further processing, as the net namespace
1530                  * or ifindex may change.
1531                  */
1532                 if (((*net_dev)->flags & IFF_UP) == 0) {
1533                         id_priv = ERR_PTR(-EHOSTUNREACH);
1534                         goto err;
1535                 }
1536
1537                 if (!validate_net_dev(*net_dev,
1538                                  (struct sockaddr *)&req.listen_addr_storage,
1539                                  (struct sockaddr *)&req.src_addr_storage)) {
1540                         id_priv = ERR_PTR(-EHOSTUNREACH);
1541                         goto err;
1542                 }
1543         }
1544
1545         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1546                                 rdma_ps_from_service_id(req.service_id),
1547                                 cma_port_from_service_id(req.service_id));
1548         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1549 err:
1550         rcu_read_unlock();
1551         if (IS_ERR(id_priv) && *net_dev) {
1552                 dev_put(*net_dev);
1553                 *net_dev = NULL;
1554         }
1555         return id_priv;
1556 }
1557
1558 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1559 {
1560         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1561 }
1562
1563 static void cma_cancel_route(struct rdma_id_private *id_priv)
1564 {
1565         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1566                 if (id_priv->query)
1567                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1568         }
1569 }
1570
1571 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1572 {
1573         struct rdma_id_private *dev_id_priv;
1574
1575         /*
1576          * Remove from listen_any_list to prevent added devices from spawning
1577          * additional listen requests.
1578          */
1579         mutex_lock(&lock);
1580         list_del(&id_priv->list);
1581
1582         while (!list_empty(&id_priv->listen_list)) {
1583                 dev_id_priv = list_entry(id_priv->listen_list.next,
1584                                          struct rdma_id_private, listen_list);
1585                 /* sync with device removal to avoid duplicate destruction */
1586                 list_del_init(&dev_id_priv->list);
1587                 list_del(&dev_id_priv->listen_list);
1588                 mutex_unlock(&lock);
1589
1590                 rdma_destroy_id(&dev_id_priv->id);
1591                 mutex_lock(&lock);
1592         }
1593         mutex_unlock(&lock);
1594 }
1595
1596 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1597                                  enum rdma_cm_state state)
1598 {
1599         switch (state) {
1600         case RDMA_CM_ADDR_QUERY:
1601                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1602                 break;
1603         case RDMA_CM_ROUTE_QUERY:
1604                 cma_cancel_route(id_priv);
1605                 break;
1606         case RDMA_CM_LISTEN:
1607                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1608                         cma_cancel_listens(id_priv);
1609                 break;
1610         default:
1611                 break;
1612         }
1613 }
1614
1615 static void cma_release_port(struct rdma_id_private *id_priv)
1616 {
1617         struct rdma_bind_list *bind_list = id_priv->bind_list;
1618         struct net *net = id_priv->id.route.addr.dev_addr.net;
1619
1620         if (!bind_list)
1621                 return;
1622
1623         mutex_lock(&lock);
1624         hlist_del(&id_priv->node);
1625         if (hlist_empty(&bind_list->owners)) {
1626                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1627                 kfree(bind_list);
1628         }
1629         mutex_unlock(&lock);
1630 }
1631
1632 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1633 {
1634         struct cma_multicast *mc;
1635
1636         while (!list_empty(&id_priv->mc_list)) {
1637                 mc = container_of(id_priv->mc_list.next,
1638                                   struct cma_multicast, list);
1639                 list_del(&mc->list);
1640                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1641                                       id_priv->id.port_num)) {
1642                         ib_sa_free_multicast(mc->multicast.ib);
1643                         kfree(mc);
1644                 } else {
1645                         if (mc->igmp_joined) {
1646                                 struct rdma_dev_addr *dev_addr =
1647                                         &id_priv->id.route.addr.dev_addr;
1648                                 struct net_device *ndev = NULL;
1649
1650                                 if (dev_addr->bound_dev_if)
1651                                         ndev = dev_get_by_index(&init_net,
1652                                                                 dev_addr->bound_dev_if);
1653                                 if (ndev) {
1654                                         cma_igmp_send(ndev,
1655                                                       &mc->multicast.ib->rec.mgid,
1656                                                       false);
1657                                         dev_put(ndev);
1658                                 }
1659                         }
1660                         kref_put(&mc->mcref, release_mc);
1661                 }
1662         }
1663 }
1664
1665 void rdma_destroy_id(struct rdma_cm_id *id)
1666 {
1667         struct rdma_id_private *id_priv;
1668         enum rdma_cm_state state;
1669
1670         id_priv = container_of(id, struct rdma_id_private, id);
1671         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1672         cma_cancel_operation(id_priv, state);
1673
1674         /*
1675          * Wait for any active callback to finish.  New callbacks will find
1676          * the id_priv state set to destroying and abort.
1677          */
1678         mutex_lock(&id_priv->handler_mutex);
1679         mutex_unlock(&id_priv->handler_mutex);
1680
1681         if (id_priv->cma_dev) {
1682                 rdma_restrack_del(&id_priv->res);
1683                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1684                         if (id_priv->cm_id.ib)
1685                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1686                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1687                         if (id_priv->cm_id.iw)
1688                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1689                 }
1690                 cma_leave_mc_groups(id_priv);
1691                 cma_release_dev(id_priv);
1692         }
1693
1694         cma_release_port(id_priv);
1695         cma_deref_id(id_priv);
1696         wait_for_completion(&id_priv->comp);
1697
1698         if (id_priv->internal_id)
1699                 cma_deref_id(id_priv->id.context);
1700
1701         kfree(id_priv->id.route.path_rec);
1702         put_net(id_priv->id.route.addr.dev_addr.net);
1703         kfree(id_priv);
1704 }
1705 EXPORT_SYMBOL(rdma_destroy_id);
1706
1707 static int cma_rep_recv(struct rdma_id_private *id_priv)
1708 {
1709         int ret;
1710
1711         ret = cma_modify_qp_rtr(id_priv, NULL);
1712         if (ret)
1713                 goto reject;
1714
1715         ret = cma_modify_qp_rts(id_priv, NULL);
1716         if (ret)
1717                 goto reject;
1718
1719         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1720         if (ret)
1721                 goto reject;
1722
1723         return 0;
1724 reject:
1725         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1726         cma_modify_qp_err(id_priv);
1727         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1728                        NULL, 0, NULL, 0);
1729         return ret;
1730 }
1731
1732 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1733                                    struct ib_cm_rep_event_param *rep_data,
1734                                    void *private_data)
1735 {
1736         event->param.conn.private_data = private_data;
1737         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1738         event->param.conn.responder_resources = rep_data->responder_resources;
1739         event->param.conn.initiator_depth = rep_data->initiator_depth;
1740         event->param.conn.flow_control = rep_data->flow_control;
1741         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1742         event->param.conn.srq = rep_data->srq;
1743         event->param.conn.qp_num = rep_data->remote_qpn;
1744 }
1745
1746 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1747 {
1748         struct rdma_id_private *id_priv = cm_id->context;
1749         struct rdma_cm_event event;
1750         int ret = 0;
1751
1752         mutex_lock(&id_priv->handler_mutex);
1753         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1754              id_priv->state != RDMA_CM_CONNECT) ||
1755             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1756              id_priv->state != RDMA_CM_DISCONNECT))
1757                 goto out;
1758
1759         memset(&event, 0, sizeof event);
1760         switch (ib_event->event) {
1761         case IB_CM_REQ_ERROR:
1762         case IB_CM_REP_ERROR:
1763                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1764                 event.status = -ETIMEDOUT;
1765                 break;
1766         case IB_CM_REP_RECEIVED:
1767                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1768                     (id_priv->id.qp_type != IB_QPT_UD))
1769                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1770                 if (id_priv->id.qp) {
1771                         event.status = cma_rep_recv(id_priv);
1772                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1773                                                      RDMA_CM_EVENT_ESTABLISHED;
1774                 } else {
1775                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1776                 }
1777                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1778                                        ib_event->private_data);
1779                 break;
1780         case IB_CM_RTU_RECEIVED:
1781         case IB_CM_USER_ESTABLISHED:
1782                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1783                 break;
1784         case IB_CM_DREQ_ERROR:
1785                 event.status = -ETIMEDOUT; /* fall through */
1786         case IB_CM_DREQ_RECEIVED:
1787         case IB_CM_DREP_RECEIVED:
1788                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1789                                    RDMA_CM_DISCONNECT))
1790                         goto out;
1791                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1792                 break;
1793         case IB_CM_TIMEWAIT_EXIT:
1794                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1795                 break;
1796         case IB_CM_MRA_RECEIVED:
1797                 /* ignore event */
1798                 goto out;
1799         case IB_CM_REJ_RECEIVED:
1800                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1801                                                                                 ib_event->param.rej_rcvd.reason));
1802                 cma_modify_qp_err(id_priv);
1803                 event.status = ib_event->param.rej_rcvd.reason;
1804                 event.event = RDMA_CM_EVENT_REJECTED;
1805                 event.param.conn.private_data = ib_event->private_data;
1806                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1807                 break;
1808         default:
1809                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1810                        ib_event->event);
1811                 goto out;
1812         }
1813
1814         ret = id_priv->id.event_handler(&id_priv->id, &event);
1815         if (ret) {
1816                 /* Destroy the CM ID by returning a non-zero value. */
1817                 id_priv->cm_id.ib = NULL;
1818                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1819                 mutex_unlock(&id_priv->handler_mutex);
1820                 rdma_destroy_id(&id_priv->id);
1821                 return ret;
1822         }
1823 out:
1824         mutex_unlock(&id_priv->handler_mutex);
1825         return ret;
1826 }
1827
1828 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1829                                                struct ib_cm_event *ib_event,
1830                                                struct net_device *net_dev)
1831 {
1832         struct rdma_id_private *listen_id_priv;
1833         struct rdma_id_private *id_priv;
1834         struct rdma_cm_id *id;
1835         struct rdma_route *rt;
1836         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1837         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1838         const __be64 service_id =
1839                 ib_event->param.req_rcvd.primary_path->service_id;
1840         int ret;
1841
1842         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1843         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1844                             listen_id->event_handler, listen_id->context,
1845                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1846                             listen_id_priv->res.kern_name);
1847         if (IS_ERR(id))
1848                 return NULL;
1849
1850         id_priv = container_of(id, struct rdma_id_private, id);
1851         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1852                               (struct sockaddr *)&id->route.addr.dst_addr,
1853                               listen_id, ib_event, ss_family, service_id))
1854                 goto err;
1855
1856         rt = &id->route;
1857         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1858         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
1859                                      GFP_KERNEL);
1860         if (!rt->path_rec)
1861                 goto err;
1862
1863         rt->path_rec[0] = *path;
1864         if (rt->num_paths == 2)
1865                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1866
1867         if (net_dev) {
1868                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1869         } else {
1870                 if (!cma_protocol_roce(listen_id) &&
1871                     cma_any_addr(cma_src_addr(id_priv))) {
1872                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1873                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1874                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1875                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1876                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1877                         if (ret)
1878                                 goto err;
1879                 }
1880         }
1881         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1882
1883         id_priv->state = RDMA_CM_CONNECT;
1884         return id_priv;
1885
1886 err:
1887         rdma_destroy_id(id);
1888         return NULL;
1889 }
1890
1891 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1892                                               struct ib_cm_event *ib_event,
1893                                               struct net_device *net_dev)
1894 {
1895         struct rdma_id_private *listen_id_priv;
1896         struct rdma_id_private *id_priv;
1897         struct rdma_cm_id *id;
1898         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1899         struct net *net = listen_id->route.addr.dev_addr.net;
1900         int ret;
1901
1902         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1903         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
1904                               listen_id->ps, IB_QPT_UD,
1905                               listen_id_priv->res.kern_name);
1906         if (IS_ERR(id))
1907                 return NULL;
1908
1909         id_priv = container_of(id, struct rdma_id_private, id);
1910         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1911                               (struct sockaddr *)&id->route.addr.dst_addr,
1912                               listen_id, ib_event, ss_family,
1913                               ib_event->param.sidr_req_rcvd.service_id))
1914                 goto err;
1915
1916         if (net_dev) {
1917                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1918         } else {
1919                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1920                         ret = cma_translate_addr(cma_src_addr(id_priv),
1921                                                  &id->route.addr.dev_addr);
1922                         if (ret)
1923                                 goto err;
1924                 }
1925         }
1926
1927         id_priv->state = RDMA_CM_CONNECT;
1928         return id_priv;
1929 err:
1930         rdma_destroy_id(id);
1931         return NULL;
1932 }
1933
1934 static void cma_set_req_event_data(struct rdma_cm_event *event,
1935                                    struct ib_cm_req_event_param *req_data,
1936                                    void *private_data, int offset)
1937 {
1938         event->param.conn.private_data = private_data + offset;
1939         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1940         event->param.conn.responder_resources = req_data->responder_resources;
1941         event->param.conn.initiator_depth = req_data->initiator_depth;
1942         event->param.conn.flow_control = req_data->flow_control;
1943         event->param.conn.retry_count = req_data->retry_count;
1944         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1945         event->param.conn.srq = req_data->srq;
1946         event->param.conn.qp_num = req_data->remote_qpn;
1947 }
1948
1949 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1950 {
1951         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1952                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1953                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1954                  (id->qp_type == IB_QPT_UD)) ||
1955                 (!id->qp_type));
1956 }
1957
1958 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1959 {
1960         struct rdma_id_private *listen_id, *conn_id = NULL;
1961         struct rdma_cm_event event;
1962         struct net_device *net_dev;
1963         u8 offset;
1964         int ret;
1965
1966         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1967         if (IS_ERR(listen_id))
1968                 return PTR_ERR(listen_id);
1969
1970         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1971                 ret = -EINVAL;
1972                 goto net_dev_put;
1973         }
1974
1975         mutex_lock(&listen_id->handler_mutex);
1976         if (listen_id->state != RDMA_CM_LISTEN) {
1977                 ret = -ECONNABORTED;
1978                 goto err1;
1979         }
1980
1981         memset(&event, 0, sizeof event);
1982         offset = cma_user_data_offset(listen_id);
1983         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1984         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1985                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1986                 event.param.ud.private_data = ib_event->private_data + offset;
1987                 event.param.ud.private_data_len =
1988                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1989         } else {
1990                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1991                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1992                                        ib_event->private_data, offset);
1993         }
1994         if (!conn_id) {
1995                 ret = -ENOMEM;
1996                 goto err1;
1997         }
1998
1999         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2000         ret = cma_acquire_dev(conn_id, listen_id);
2001         if (ret)
2002                 goto err2;
2003
2004         conn_id->cm_id.ib = cm_id;
2005         cm_id->context = conn_id;
2006         cm_id->cm_handler = cma_ib_handler;
2007
2008         /*
2009          * Protect against the user destroying conn_id from another thread
2010          * until we're done accessing it.
2011          */
2012         atomic_inc(&conn_id->refcount);
2013         ret = conn_id->id.event_handler(&conn_id->id, &event);
2014         if (ret)
2015                 goto err3;
2016         /*
2017          * Acquire mutex to prevent user executing rdma_destroy_id()
2018          * while we're accessing the cm_id.
2019          */
2020         mutex_lock(&lock);
2021         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2022             (conn_id->id.qp_type != IB_QPT_UD))
2023                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2024         mutex_unlock(&lock);
2025         mutex_unlock(&conn_id->handler_mutex);
2026         mutex_unlock(&listen_id->handler_mutex);
2027         cma_deref_id(conn_id);
2028         if (net_dev)
2029                 dev_put(net_dev);
2030         return 0;
2031
2032 err3:
2033         cma_deref_id(conn_id);
2034         /* Destroy the CM ID by returning a non-zero value. */
2035         conn_id->cm_id.ib = NULL;
2036 err2:
2037         cma_exch(conn_id, RDMA_CM_DESTROYING);
2038         mutex_unlock(&conn_id->handler_mutex);
2039 err1:
2040         mutex_unlock(&listen_id->handler_mutex);
2041         if (conn_id)
2042                 rdma_destroy_id(&conn_id->id);
2043
2044 net_dev_put:
2045         if (net_dev)
2046                 dev_put(net_dev);
2047
2048         return ret;
2049 }
2050
2051 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2052 {
2053         if (addr->sa_family == AF_IB)
2054                 return ((struct sockaddr_ib *) addr)->sib_sid;
2055
2056         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2057 }
2058 EXPORT_SYMBOL(rdma_get_service_id);
2059
2060 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2061                     union ib_gid *dgid)
2062 {
2063         struct rdma_addr *addr = &cm_id->route.addr;
2064
2065         if (!cm_id->device) {
2066                 if (sgid)
2067                         memset(sgid, 0, sizeof(*sgid));
2068                 if (dgid)
2069                         memset(dgid, 0, sizeof(*dgid));
2070                 return;
2071         }
2072
2073         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2074                 if (sgid)
2075                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2076                 if (dgid)
2077                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2078         } else {
2079                 if (sgid)
2080                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2081                 if (dgid)
2082                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2083         }
2084 }
2085 EXPORT_SYMBOL(rdma_read_gids);
2086
2087 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2088 {
2089         struct rdma_id_private *id_priv = iw_id->context;
2090         struct rdma_cm_event event;
2091         int ret = 0;
2092         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2093         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2094
2095         mutex_lock(&id_priv->handler_mutex);
2096         if (id_priv->state != RDMA_CM_CONNECT)
2097                 goto out;
2098
2099         memset(&event, 0, sizeof event);
2100         switch (iw_event->event) {
2101         case IW_CM_EVENT_CLOSE:
2102                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2103                 break;
2104         case IW_CM_EVENT_CONNECT_REPLY:
2105                 memcpy(cma_src_addr(id_priv), laddr,
2106                        rdma_addr_size(laddr));
2107                 memcpy(cma_dst_addr(id_priv), raddr,
2108                        rdma_addr_size(raddr));
2109                 switch (iw_event->status) {
2110                 case 0:
2111                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2112                         event.param.conn.initiator_depth = iw_event->ird;
2113                         event.param.conn.responder_resources = iw_event->ord;
2114                         break;
2115                 case -ECONNRESET:
2116                 case -ECONNREFUSED:
2117                         event.event = RDMA_CM_EVENT_REJECTED;
2118                         break;
2119                 case -ETIMEDOUT:
2120                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2121                         break;
2122                 default:
2123                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2124                         break;
2125                 }
2126                 break;
2127         case IW_CM_EVENT_ESTABLISHED:
2128                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2129                 event.param.conn.initiator_depth = iw_event->ird;
2130                 event.param.conn.responder_resources = iw_event->ord;
2131                 break;
2132         default:
2133                 goto out;
2134         }
2135
2136         event.status = iw_event->status;
2137         event.param.conn.private_data = iw_event->private_data;
2138         event.param.conn.private_data_len = iw_event->private_data_len;
2139         ret = id_priv->id.event_handler(&id_priv->id, &event);
2140         if (ret) {
2141                 /* Destroy the CM ID by returning a non-zero value. */
2142                 id_priv->cm_id.iw = NULL;
2143                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2144                 mutex_unlock(&id_priv->handler_mutex);
2145                 rdma_destroy_id(&id_priv->id);
2146                 return ret;
2147         }
2148
2149 out:
2150         mutex_unlock(&id_priv->handler_mutex);
2151         return ret;
2152 }
2153
2154 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2155                                struct iw_cm_event *iw_event)
2156 {
2157         struct rdma_cm_id *new_cm_id;
2158         struct rdma_id_private *listen_id, *conn_id;
2159         struct rdma_cm_event event;
2160         int ret = -ECONNABORTED;
2161         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2162         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2163
2164         listen_id = cm_id->context;
2165
2166         mutex_lock(&listen_id->handler_mutex);
2167         if (listen_id->state != RDMA_CM_LISTEN)
2168                 goto out;
2169
2170         /* Create a new RDMA id for the new IW CM ID */
2171         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2172                                      listen_id->id.event_handler,
2173                                      listen_id->id.context,
2174                                      RDMA_PS_TCP, IB_QPT_RC,
2175                                      listen_id->res.kern_name);
2176         if (IS_ERR(new_cm_id)) {
2177                 ret = -ENOMEM;
2178                 goto out;
2179         }
2180         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2181         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2182         conn_id->state = RDMA_CM_CONNECT;
2183
2184         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2185         if (ret) {
2186                 mutex_unlock(&conn_id->handler_mutex);
2187                 rdma_destroy_id(new_cm_id);
2188                 goto out;
2189         }
2190
2191         ret = cma_acquire_dev(conn_id, listen_id);
2192         if (ret) {
2193                 mutex_unlock(&conn_id->handler_mutex);
2194                 rdma_destroy_id(new_cm_id);
2195                 goto out;
2196         }
2197
2198         conn_id->cm_id.iw = cm_id;
2199         cm_id->context = conn_id;
2200         cm_id->cm_handler = cma_iw_handler;
2201
2202         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2203         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2204
2205         memset(&event, 0, sizeof event);
2206         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2207         event.param.conn.private_data = iw_event->private_data;
2208         event.param.conn.private_data_len = iw_event->private_data_len;
2209         event.param.conn.initiator_depth = iw_event->ird;
2210         event.param.conn.responder_resources = iw_event->ord;
2211
2212         /*
2213          * Protect against the user destroying conn_id from another thread
2214          * until we're done accessing it.
2215          */
2216         atomic_inc(&conn_id->refcount);
2217         ret = conn_id->id.event_handler(&conn_id->id, &event);
2218         if (ret) {
2219                 /* User wants to destroy the CM ID */
2220                 conn_id->cm_id.iw = NULL;
2221                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2222                 mutex_unlock(&conn_id->handler_mutex);
2223                 cma_deref_id(conn_id);
2224                 rdma_destroy_id(&conn_id->id);
2225                 goto out;
2226         }
2227
2228         mutex_unlock(&conn_id->handler_mutex);
2229         cma_deref_id(conn_id);
2230
2231 out:
2232         mutex_unlock(&listen_id->handler_mutex);
2233         return ret;
2234 }
2235
2236 static int cma_ib_listen(struct rdma_id_private *id_priv)
2237 {
2238         struct sockaddr *addr;
2239         struct ib_cm_id *id;
2240         __be64 svc_id;
2241
2242         addr = cma_src_addr(id_priv);
2243         svc_id = rdma_get_service_id(&id_priv->id, addr);
2244         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2245         if (IS_ERR(id))
2246                 return PTR_ERR(id);
2247         id_priv->cm_id.ib = id;
2248
2249         return 0;
2250 }
2251
2252 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2253 {
2254         int ret;
2255         struct iw_cm_id *id;
2256
2257         id = iw_create_cm_id(id_priv->id.device,
2258                              iw_conn_req_handler,
2259                              id_priv);
2260         if (IS_ERR(id))
2261                 return PTR_ERR(id);
2262
2263         id->tos = id_priv->tos;
2264         id_priv->cm_id.iw = id;
2265
2266         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2267                rdma_addr_size(cma_src_addr(id_priv)));
2268
2269         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2270
2271         if (ret) {
2272                 iw_destroy_cm_id(id_priv->cm_id.iw);
2273                 id_priv->cm_id.iw = NULL;
2274         }
2275
2276         return ret;
2277 }
2278
2279 static int cma_listen_handler(struct rdma_cm_id *id,
2280                               struct rdma_cm_event *event)
2281 {
2282         struct rdma_id_private *id_priv = id->context;
2283
2284         id->context = id_priv->id.context;
2285         id->event_handler = id_priv->id.event_handler;
2286         return id_priv->id.event_handler(id, event);
2287 }
2288
2289 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2290                               struct cma_device *cma_dev)
2291 {
2292         struct rdma_id_private *dev_id_priv;
2293         struct rdma_cm_id *id;
2294         struct net *net = id_priv->id.route.addr.dev_addr.net;
2295         int ret;
2296
2297         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2298                 return;
2299
2300         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2301                               id_priv->id.qp_type, id_priv->res.kern_name);
2302         if (IS_ERR(id))
2303                 return;
2304
2305         dev_id_priv = container_of(id, struct rdma_id_private, id);
2306
2307         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2308         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2309                rdma_addr_size(cma_src_addr(id_priv)));
2310
2311         _cma_attach_to_dev(dev_id_priv, cma_dev);
2312         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2313         atomic_inc(&id_priv->refcount);
2314         dev_id_priv->internal_id = 1;
2315         dev_id_priv->afonly = id_priv->afonly;
2316
2317         ret = rdma_listen(id, id_priv->backlog);
2318         if (ret)
2319                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2320                         ret, cma_dev->device->name);
2321 }
2322
2323 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2324 {
2325         struct cma_device *cma_dev;
2326
2327         mutex_lock(&lock);
2328         list_add_tail(&id_priv->list, &listen_any_list);
2329         list_for_each_entry(cma_dev, &dev_list, list)
2330                 cma_listen_on_dev(id_priv, cma_dev);
2331         mutex_unlock(&lock);
2332 }
2333
2334 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2335 {
2336         struct rdma_id_private *id_priv;
2337
2338         id_priv = container_of(id, struct rdma_id_private, id);
2339         id_priv->tos = (u8) tos;
2340         id_priv->tos_set = true;
2341 }
2342 EXPORT_SYMBOL(rdma_set_service_type);
2343
2344 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2345                               void *context)
2346 {
2347         struct cma_work *work = context;
2348         struct rdma_route *route;
2349
2350         route = &work->id->id.route;
2351
2352         if (!status) {
2353                 route->num_paths = 1;
2354                 *route->path_rec = *path_rec;
2355         } else {
2356                 work->old_state = RDMA_CM_ROUTE_QUERY;
2357                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2358                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2359                 work->event.status = status;
2360                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2361                                      status);
2362         }
2363
2364         queue_work(cma_wq, &work->work);
2365 }
2366
2367 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2368                               struct cma_work *work)
2369 {
2370         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2371         struct sa_path_rec path_rec;
2372         ib_sa_comp_mask comp_mask;
2373         struct sockaddr_in6 *sin6;
2374         struct sockaddr_ib *sib;
2375
2376         memset(&path_rec, 0, sizeof path_rec);
2377
2378         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2379                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2380         else
2381                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2382         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2383         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2384         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2385         path_rec.numb_path = 1;
2386         path_rec.reversible = 1;
2387         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2388                                                   cma_dst_addr(id_priv));
2389
2390         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2391                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2392                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2393
2394         switch (cma_family(id_priv)) {
2395         case AF_INET:
2396                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2397                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2398                 break;
2399         case AF_INET6:
2400                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2401                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2402                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2403                 break;
2404         case AF_IB:
2405                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2406                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2407                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2408                 break;
2409         }
2410
2411         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2412                                                id_priv->id.port_num, &path_rec,
2413                                                comp_mask, timeout_ms,
2414                                                GFP_KERNEL, cma_query_handler,
2415                                                work, &id_priv->query);
2416
2417         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2418 }
2419
2420 static void cma_work_handler(struct work_struct *_work)
2421 {
2422         struct cma_work *work = container_of(_work, struct cma_work, work);
2423         struct rdma_id_private *id_priv = work->id;
2424         int destroy = 0;
2425
2426         mutex_lock(&id_priv->handler_mutex);
2427         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2428                 goto out;
2429
2430         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2431                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2432                 destroy = 1;
2433         }
2434 out:
2435         mutex_unlock(&id_priv->handler_mutex);
2436         cma_deref_id(id_priv);
2437         if (destroy)
2438                 rdma_destroy_id(&id_priv->id);
2439         kfree(work);
2440 }
2441
2442 static void cma_ndev_work_handler(struct work_struct *_work)
2443 {
2444         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2445         struct rdma_id_private *id_priv = work->id;
2446         int destroy = 0;
2447
2448         mutex_lock(&id_priv->handler_mutex);
2449         if (id_priv->state == RDMA_CM_DESTROYING ||
2450             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2451                 goto out;
2452
2453         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2454                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2455                 destroy = 1;
2456         }
2457
2458 out:
2459         mutex_unlock(&id_priv->handler_mutex);
2460         cma_deref_id(id_priv);
2461         if (destroy)
2462                 rdma_destroy_id(&id_priv->id);
2463         kfree(work);
2464 }
2465
2466 static void cma_init_resolve_route_work(struct cma_work *work,
2467                                         struct rdma_id_private *id_priv)
2468 {
2469         work->id = id_priv;
2470         INIT_WORK(&work->work, cma_work_handler);
2471         work->old_state = RDMA_CM_ROUTE_QUERY;
2472         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2473         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2474 }
2475
2476 static void cma_init_resolve_addr_work(struct cma_work *work,
2477                                        struct rdma_id_private *id_priv)
2478 {
2479         work->id = id_priv;
2480         INIT_WORK(&work->work, cma_work_handler);
2481         work->old_state = RDMA_CM_ADDR_QUERY;
2482         work->new_state = RDMA_CM_ADDR_RESOLVED;
2483         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2484 }
2485
2486 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2487 {
2488         struct rdma_route *route = &id_priv->id.route;
2489         struct cma_work *work;
2490         int ret;
2491
2492         work = kzalloc(sizeof *work, GFP_KERNEL);
2493         if (!work)
2494                 return -ENOMEM;
2495
2496         cma_init_resolve_route_work(work, id_priv);
2497
2498         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2499         if (!route->path_rec) {
2500                 ret = -ENOMEM;
2501                 goto err1;
2502         }
2503
2504         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2505         if (ret)
2506                 goto err2;
2507
2508         return 0;
2509 err2:
2510         kfree(route->path_rec);
2511         route->path_rec = NULL;
2512 err1:
2513         kfree(work);
2514         return ret;
2515 }
2516
2517 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2518                                            unsigned long supported_gids,
2519                                            enum ib_gid_type default_gid)
2520 {
2521         if ((network_type == RDMA_NETWORK_IPV4 ||
2522              network_type == RDMA_NETWORK_IPV6) &&
2523             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2524                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2525
2526         return default_gid;
2527 }
2528
2529 /*
2530  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2531  * path record type based on GID type.
2532  * It also sets up other L2 fields which includes destination mac address
2533  * netdev ifindex, of the path record.
2534  * It returns the netdev of the bound interface for this path record entry.
2535  */
2536 static struct net_device *
2537 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2538 {
2539         struct rdma_route *route = &id_priv->id.route;
2540         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2541         struct rdma_addr *addr = &route->addr;
2542         unsigned long supported_gids;
2543         struct net_device *ndev;
2544
2545         if (!addr->dev_addr.bound_dev_if)
2546                 return NULL;
2547
2548         ndev = dev_get_by_index(addr->dev_addr.net,
2549                                 addr->dev_addr.bound_dev_if);
2550         if (!ndev)
2551                 return NULL;
2552
2553         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2554                                                     id_priv->id.port_num);
2555         gid_type = cma_route_gid_type(addr->dev_addr.network,
2556                                       supported_gids,
2557                                       id_priv->gid_type);
2558         /* Use the hint from IP Stack to select GID Type */
2559         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2560                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2561         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2562
2563         route->path_rec->roce.route_resolved = true;
2564         sa_path_set_ndev(route->path_rec, addr->dev_addr.net);
2565         sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2566         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2567         return ndev;
2568 }
2569
2570 int rdma_set_ib_path(struct rdma_cm_id *id,
2571                      struct sa_path_rec *path_rec)
2572 {
2573         struct rdma_id_private *id_priv;
2574         struct net_device *ndev;
2575         int ret;
2576
2577         id_priv = container_of(id, struct rdma_id_private, id);
2578         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2579                            RDMA_CM_ROUTE_RESOLVED))
2580                 return -EINVAL;
2581
2582         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2583                                      GFP_KERNEL);
2584         if (!id->route.path_rec) {
2585                 ret = -ENOMEM;
2586                 goto err;
2587         }
2588
2589         if (rdma_protocol_roce(id->device, id->port_num)) {
2590                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2591                 if (!ndev) {
2592                         ret = -ENODEV;
2593                         goto err_free;
2594                 }
2595                 dev_put(ndev);
2596         }
2597
2598         id->route.num_paths = 1;
2599         return 0;
2600
2601 err_free:
2602         kfree(id->route.path_rec);
2603         id->route.path_rec = NULL;
2604 err:
2605         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2606         return ret;
2607 }
2608 EXPORT_SYMBOL(rdma_set_ib_path);
2609
2610 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2611 {
2612         struct cma_work *work;
2613
2614         work = kzalloc(sizeof *work, GFP_KERNEL);
2615         if (!work)
2616                 return -ENOMEM;
2617
2618         cma_init_resolve_route_work(work, id_priv);
2619         queue_work(cma_wq, &work->work);
2620         return 0;
2621 }
2622
2623 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2624 {
2625         int prio;
2626         struct net_device *dev;
2627
2628         prio = rt_tos2priority(tos);
2629         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2630         if (dev->num_tc)
2631                 return netdev_get_prio_tc_map(dev, prio);
2632
2633 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2634         if (is_vlan_dev(ndev))
2635                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2636                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2637 #endif
2638         return 0;
2639 }
2640
2641 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2642 {
2643         struct rdma_route *route = &id_priv->id.route;
2644         struct rdma_addr *addr = &route->addr;
2645         struct cma_work *work;
2646         int ret;
2647         struct net_device *ndev;
2648
2649         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2650                                         rdma_start_port(id_priv->cma_dev->device)];
2651         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2652
2653
2654         work = kzalloc(sizeof *work, GFP_KERNEL);
2655         if (!work)
2656                 return -ENOMEM;
2657
2658         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2659         if (!route->path_rec) {
2660                 ret = -ENOMEM;
2661                 goto err1;
2662         }
2663
2664         route->num_paths = 1;
2665
2666         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2667         if (!ndev) {
2668                 ret = -ENODEV;
2669                 goto err2;
2670         }
2671
2672         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2673                     &route->path_rec->sgid);
2674         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2675                     &route->path_rec->dgid);
2676
2677         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2678                 /* TODO: get the hoplimit from the inet/inet6 device */
2679                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2680         else
2681                 route->path_rec->hop_limit = 1;
2682         route->path_rec->reversible = 1;
2683         route->path_rec->pkey = cpu_to_be16(0xffff);
2684         route->path_rec->mtu_selector = IB_SA_EQ;
2685         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2686         route->path_rec->traffic_class = tos;
2687         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2688         route->path_rec->rate_selector = IB_SA_EQ;
2689         route->path_rec->rate = iboe_get_rate(ndev);
2690         dev_put(ndev);
2691         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2692         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2693         if (!route->path_rec->mtu) {
2694                 ret = -EINVAL;
2695                 goto err2;
2696         }
2697
2698         cma_init_resolve_route_work(work, id_priv);
2699         queue_work(cma_wq, &work->work);
2700
2701         return 0;
2702
2703 err2:
2704         kfree(route->path_rec);
2705         route->path_rec = NULL;
2706 err1:
2707         kfree(work);
2708         return ret;
2709 }
2710
2711 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2712 {
2713         struct rdma_id_private *id_priv;
2714         int ret;
2715
2716         id_priv = container_of(id, struct rdma_id_private, id);
2717         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2718                 return -EINVAL;
2719
2720         atomic_inc(&id_priv->refcount);
2721         if (rdma_cap_ib_sa(id->device, id->port_num))
2722                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2723         else if (rdma_protocol_roce(id->device, id->port_num))
2724                 ret = cma_resolve_iboe_route(id_priv);
2725         else if (rdma_protocol_iwarp(id->device, id->port_num))
2726                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2727         else
2728                 ret = -ENOSYS;
2729
2730         if (ret)
2731                 goto err;
2732
2733         return 0;
2734 err:
2735         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2736         cma_deref_id(id_priv);
2737         return ret;
2738 }
2739 EXPORT_SYMBOL(rdma_resolve_route);
2740
2741 static void cma_set_loopback(struct sockaddr *addr)
2742 {
2743         switch (addr->sa_family) {
2744         case AF_INET:
2745                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2746                 break;
2747         case AF_INET6:
2748                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2749                               0, 0, 0, htonl(1));
2750                 break;
2751         default:
2752                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2753                             0, 0, 0, htonl(1));
2754                 break;
2755         }
2756 }
2757
2758 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2759 {
2760         struct cma_device *cma_dev, *cur_dev;
2761         union ib_gid gid;
2762         enum ib_port_state port_state;
2763         u16 pkey;
2764         int ret;
2765         u8 p;
2766
2767         cma_dev = NULL;
2768         mutex_lock(&lock);
2769         list_for_each_entry(cur_dev, &dev_list, list) {
2770                 if (cma_family(id_priv) == AF_IB &&
2771                     !rdma_cap_ib_cm(cur_dev->device, 1))
2772                         continue;
2773
2774                 if (!cma_dev)
2775                         cma_dev = cur_dev;
2776
2777                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2778                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2779                             port_state == IB_PORT_ACTIVE) {
2780                                 cma_dev = cur_dev;
2781                                 goto port_found;
2782                         }
2783                 }
2784         }
2785
2786         if (!cma_dev) {
2787                 ret = -ENODEV;
2788                 goto out;
2789         }
2790
2791         p = 1;
2792
2793 port_found:
2794         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2795         if (ret)
2796                 goto out;
2797
2798         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2799         if (ret)
2800                 goto out;
2801
2802         id_priv->id.route.addr.dev_addr.dev_type =
2803                 (rdma_protocol_ib(cma_dev->device, p)) ?
2804                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2805
2806         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2807         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2808         id_priv->id.port_num = p;
2809         cma_attach_to_dev(id_priv, cma_dev);
2810         cma_set_loopback(cma_src_addr(id_priv));
2811 out:
2812         mutex_unlock(&lock);
2813         return ret;
2814 }
2815
2816 static void addr_handler(int status, struct sockaddr *src_addr,
2817                          struct rdma_dev_addr *dev_addr, void *context)
2818 {
2819         struct rdma_id_private *id_priv = context;
2820         struct rdma_cm_event event;
2821
2822         memset(&event, 0, sizeof event);
2823         mutex_lock(&id_priv->handler_mutex);
2824         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2825                            RDMA_CM_ADDR_RESOLVED))
2826                 goto out;
2827
2828         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2829         if (!status && !id_priv->cma_dev) {
2830                 status = cma_acquire_dev(id_priv, NULL);
2831                 if (status)
2832                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2833                                              status);
2834         } else {
2835                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2836         }
2837
2838         if (status) {
2839                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2840                                    RDMA_CM_ADDR_BOUND))
2841                         goto out;
2842                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2843                 event.status = status;
2844         } else
2845                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2846
2847         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2848                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2849                 mutex_unlock(&id_priv->handler_mutex);
2850                 cma_deref_id(id_priv);
2851                 rdma_destroy_id(&id_priv->id);
2852                 return;
2853         }
2854 out:
2855         mutex_unlock(&id_priv->handler_mutex);
2856         cma_deref_id(id_priv);
2857 }
2858
2859 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2860 {
2861         struct cma_work *work;
2862         union ib_gid gid;
2863         int ret;
2864
2865         work = kzalloc(sizeof *work, GFP_KERNEL);
2866         if (!work)
2867                 return -ENOMEM;
2868
2869         if (!id_priv->cma_dev) {
2870                 ret = cma_bind_loopback(id_priv);
2871                 if (ret)
2872                         goto err;
2873         }
2874
2875         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2876         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2877
2878         cma_init_resolve_addr_work(work, id_priv);
2879         queue_work(cma_wq, &work->work);
2880         return 0;
2881 err:
2882         kfree(work);
2883         return ret;
2884 }
2885
2886 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2887 {
2888         struct cma_work *work;
2889         int ret;
2890
2891         work = kzalloc(sizeof *work, GFP_KERNEL);
2892         if (!work)
2893                 return -ENOMEM;
2894
2895         if (!id_priv->cma_dev) {
2896                 ret = cma_resolve_ib_dev(id_priv);
2897                 if (ret)
2898                         goto err;
2899         }
2900
2901         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2902                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2903
2904         cma_init_resolve_addr_work(work, id_priv);
2905         queue_work(cma_wq, &work->work);
2906         return 0;
2907 err:
2908         kfree(work);
2909         return ret;
2910 }
2911
2912 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2913                          struct sockaddr *dst_addr)
2914 {
2915         if (!src_addr || !src_addr->sa_family) {
2916                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2917                 src_addr->sa_family = dst_addr->sa_family;
2918                 if (IS_ENABLED(CONFIG_IPV6) &&
2919                     dst_addr->sa_family == AF_INET6) {
2920                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2921                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2922                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2923                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2924                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2925                 } else if (dst_addr->sa_family == AF_IB) {
2926                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2927                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2928                 }
2929         }
2930         return rdma_bind_addr(id, src_addr);
2931 }
2932
2933 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2934                       struct sockaddr *dst_addr, int timeout_ms)
2935 {
2936         struct rdma_id_private *id_priv;
2937         int ret;
2938
2939         id_priv = container_of(id, struct rdma_id_private, id);
2940         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2941         if (id_priv->state == RDMA_CM_IDLE) {
2942                 ret = cma_bind_addr(id, src_addr, dst_addr);
2943                 if (ret) {
2944                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2945                         return ret;
2946                 }
2947         }
2948
2949         if (cma_family(id_priv) != dst_addr->sa_family) {
2950                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2951                 return -EINVAL;
2952         }
2953
2954         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2955                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2956                 return -EINVAL;
2957         }
2958
2959         atomic_inc(&id_priv->refcount);
2960         if (cma_any_addr(dst_addr)) {
2961                 ret = cma_resolve_loopback(id_priv);
2962         } else {
2963                 if (dst_addr->sa_family == AF_IB) {
2964                         ret = cma_resolve_ib_addr(id_priv);
2965                 } else {
2966                         ret = rdma_resolve_ip(cma_src_addr(id_priv),
2967                                               dst_addr, &id->route.addr.dev_addr,
2968                                               timeout_ms, addr_handler, id_priv);
2969                 }
2970         }
2971         if (ret)
2972                 goto err;
2973
2974         return 0;
2975 err:
2976         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2977         cma_deref_id(id_priv);
2978         return ret;
2979 }
2980 EXPORT_SYMBOL(rdma_resolve_addr);
2981
2982 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2983 {
2984         struct rdma_id_private *id_priv;
2985         unsigned long flags;
2986         int ret;
2987
2988         id_priv = container_of(id, struct rdma_id_private, id);
2989         spin_lock_irqsave(&id_priv->lock, flags);
2990         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2991                 id_priv->reuseaddr = reuse;
2992                 ret = 0;
2993         } else {
2994                 ret = -EINVAL;
2995         }
2996         spin_unlock_irqrestore(&id_priv->lock, flags);
2997         return ret;
2998 }
2999 EXPORT_SYMBOL(rdma_set_reuseaddr);
3000
3001 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3002 {
3003         struct rdma_id_private *id_priv;
3004         unsigned long flags;
3005         int ret;
3006
3007         id_priv = container_of(id, struct rdma_id_private, id);
3008         spin_lock_irqsave(&id_priv->lock, flags);
3009         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3010                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3011                 id_priv->afonly = afonly;
3012                 ret = 0;
3013         } else {
3014                 ret = -EINVAL;
3015         }
3016         spin_unlock_irqrestore(&id_priv->lock, flags);
3017         return ret;
3018 }
3019 EXPORT_SYMBOL(rdma_set_afonly);
3020
3021 static void cma_bind_port(struct rdma_bind_list *bind_list,
3022                           struct rdma_id_private *id_priv)
3023 {
3024         struct sockaddr *addr;
3025         struct sockaddr_ib *sib;
3026         u64 sid, mask;
3027         __be16 port;
3028
3029         addr = cma_src_addr(id_priv);
3030         port = htons(bind_list->port);
3031
3032         switch (addr->sa_family) {
3033         case AF_INET:
3034                 ((struct sockaddr_in *) addr)->sin_port = port;
3035                 break;
3036         case AF_INET6:
3037                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3038                 break;
3039         case AF_IB:
3040                 sib = (struct sockaddr_ib *) addr;
3041                 sid = be64_to_cpu(sib->sib_sid);
3042                 mask = be64_to_cpu(sib->sib_sid_mask);
3043                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3044                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3045                 break;
3046         }
3047         id_priv->bind_list = bind_list;
3048         hlist_add_head(&id_priv->node, &bind_list->owners);
3049 }
3050
3051 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3052                           struct rdma_id_private *id_priv, unsigned short snum)
3053 {
3054         struct rdma_bind_list *bind_list;
3055         int ret;
3056
3057         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3058         if (!bind_list)
3059                 return -ENOMEM;
3060
3061         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3062                            snum);
3063         if (ret < 0)
3064                 goto err;
3065
3066         bind_list->ps = ps;
3067         bind_list->port = (unsigned short)ret;
3068         cma_bind_port(bind_list, id_priv);
3069         return 0;
3070 err:
3071         kfree(bind_list);
3072         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3073 }
3074
3075 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3076                               struct rdma_id_private *id_priv)
3077 {
3078         struct rdma_id_private *cur_id;
3079         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3080         struct sockaddr  *saddr = cma_src_addr(id_priv);
3081         __be16 dport = cma_port(daddr);
3082
3083         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3084                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3085                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3086                 __be16 cur_dport = cma_port(cur_daddr);
3087
3088                 if (id_priv == cur_id)
3089                         continue;
3090
3091                 /* different dest port -> unique */
3092                 if (!cma_any_port(daddr) &&
3093                     !cma_any_port(cur_daddr) &&
3094                     (dport != cur_dport))
3095                         continue;
3096
3097                 /* different src address -> unique */
3098                 if (!cma_any_addr(saddr) &&
3099                     !cma_any_addr(cur_saddr) &&
3100                     cma_addr_cmp(saddr, cur_saddr))
3101                         continue;
3102
3103                 /* different dst address -> unique */
3104                 if (!cma_any_addr(daddr) &&
3105                     !cma_any_addr(cur_daddr) &&
3106                     cma_addr_cmp(daddr, cur_daddr))
3107                         continue;
3108
3109                 return -EADDRNOTAVAIL;
3110         }
3111         return 0;
3112 }
3113
3114 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3115                               struct rdma_id_private *id_priv)
3116 {
3117         static unsigned int last_used_port;
3118         int low, high, remaining;
3119         unsigned int rover;
3120         struct net *net = id_priv->id.route.addr.dev_addr.net;
3121
3122         inet_get_local_port_range(net, &low, &high);
3123         remaining = (high - low) + 1;
3124         rover = prandom_u32() % remaining + low;
3125 retry:
3126         if (last_used_port != rover) {
3127                 struct rdma_bind_list *bind_list;
3128                 int ret;
3129
3130                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3131
3132                 if (!bind_list) {
3133                         ret = cma_alloc_port(ps, id_priv, rover);
3134                 } else {
3135                         ret = cma_port_is_unique(bind_list, id_priv);
3136                         if (!ret)
3137                                 cma_bind_port(bind_list, id_priv);
3138                 }
3139                 /*
3140                  * Remember previously used port number in order to avoid
3141                  * re-using same port immediately after it is closed.
3142                  */
3143                 if (!ret)
3144                         last_used_port = rover;
3145                 if (ret != -EADDRNOTAVAIL)
3146                         return ret;
3147         }
3148         if (--remaining) {
3149                 rover++;
3150                 if ((rover < low) || (rover > high))
3151                         rover = low;
3152                 goto retry;
3153         }
3154         return -EADDRNOTAVAIL;
3155 }
3156
3157 /*
3158  * Check that the requested port is available.  This is called when trying to
3159  * bind to a specific port, or when trying to listen on a bound port.  In
3160  * the latter case, the provided id_priv may already be on the bind_list, but
3161  * we still need to check that it's okay to start listening.
3162  */
3163 static int cma_check_port(struct rdma_bind_list *bind_list,
3164                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3165 {
3166         struct rdma_id_private *cur_id;
3167         struct sockaddr *addr, *cur_addr;
3168
3169         addr = cma_src_addr(id_priv);
3170         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3171                 if (id_priv == cur_id)
3172                         continue;
3173
3174                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3175                     cur_id->reuseaddr)
3176                         continue;
3177
3178                 cur_addr = cma_src_addr(cur_id);
3179                 if (id_priv->afonly && cur_id->afonly &&
3180                     (addr->sa_family != cur_addr->sa_family))
3181                         continue;
3182
3183                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3184                         return -EADDRNOTAVAIL;
3185
3186                 if (!cma_addr_cmp(addr, cur_addr))
3187                         return -EADDRINUSE;
3188         }
3189         return 0;
3190 }
3191
3192 static int cma_use_port(enum rdma_ucm_port_space ps,
3193                         struct rdma_id_private *id_priv)
3194 {
3195         struct rdma_bind_list *bind_list;
3196         unsigned short snum;
3197         int ret;
3198
3199         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3200         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3201                 return -EACCES;
3202
3203         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3204         if (!bind_list) {
3205                 ret = cma_alloc_port(ps, id_priv, snum);
3206         } else {
3207                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3208                 if (!ret)
3209                         cma_bind_port(bind_list, id_priv);
3210         }
3211         return ret;
3212 }
3213
3214 static int cma_bind_listen(struct rdma_id_private *id_priv)
3215 {
3216         struct rdma_bind_list *bind_list = id_priv->bind_list;
3217         int ret = 0;
3218
3219         mutex_lock(&lock);
3220         if (bind_list->owners.first->next)
3221                 ret = cma_check_port(bind_list, id_priv, 0);
3222         mutex_unlock(&lock);
3223         return ret;
3224 }
3225
3226 static enum rdma_ucm_port_space
3227 cma_select_inet_ps(struct rdma_id_private *id_priv)
3228 {
3229         switch (id_priv->id.ps) {
3230         case RDMA_PS_TCP:
3231         case RDMA_PS_UDP:
3232         case RDMA_PS_IPOIB:
3233         case RDMA_PS_IB:
3234                 return id_priv->id.ps;
3235         default:
3236
3237                 return 0;
3238         }
3239 }
3240
3241 static enum rdma_ucm_port_space
3242 cma_select_ib_ps(struct rdma_id_private *id_priv)
3243 {
3244         enum rdma_ucm_port_space ps = 0;
3245         struct sockaddr_ib *sib;
3246         u64 sid_ps, mask, sid;
3247
3248         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3249         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3250         sid = be64_to_cpu(sib->sib_sid) & mask;
3251
3252         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3253                 sid_ps = RDMA_IB_IP_PS_IB;
3254                 ps = RDMA_PS_IB;
3255         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3256                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3257                 sid_ps = RDMA_IB_IP_PS_TCP;
3258                 ps = RDMA_PS_TCP;
3259         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3260                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3261                 sid_ps = RDMA_IB_IP_PS_UDP;
3262                 ps = RDMA_PS_UDP;
3263         }
3264
3265         if (ps) {
3266                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3267                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3268                                                 be64_to_cpu(sib->sib_sid_mask));
3269         }
3270         return ps;
3271 }
3272
3273 static int cma_get_port(struct rdma_id_private *id_priv)
3274 {
3275         enum rdma_ucm_port_space ps;
3276         int ret;
3277
3278         if (cma_family(id_priv) != AF_IB)
3279                 ps = cma_select_inet_ps(id_priv);
3280         else
3281                 ps = cma_select_ib_ps(id_priv);
3282         if (!ps)
3283                 return -EPROTONOSUPPORT;
3284
3285         mutex_lock(&lock);
3286         if (cma_any_port(cma_src_addr(id_priv)))
3287                 ret = cma_alloc_any_port(ps, id_priv);
3288         else
3289                 ret = cma_use_port(ps, id_priv);
3290         mutex_unlock(&lock);
3291
3292         return ret;
3293 }
3294
3295 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3296                                struct sockaddr *addr)
3297 {
3298 #if IS_ENABLED(CONFIG_IPV6)
3299         struct sockaddr_in6 *sin6;
3300
3301         if (addr->sa_family != AF_INET6)
3302                 return 0;
3303
3304         sin6 = (struct sockaddr_in6 *) addr;
3305
3306         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3307                 return 0;
3308
3309         if (!sin6->sin6_scope_id)
3310                         return -EINVAL;
3311
3312         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3313 #endif
3314         return 0;
3315 }
3316
3317 int rdma_listen(struct rdma_cm_id *id, int backlog)
3318 {
3319         struct rdma_id_private *id_priv;
3320         int ret;
3321
3322         id_priv = container_of(id, struct rdma_id_private, id);
3323         if (id_priv->state == RDMA_CM_IDLE) {
3324                 id->route.addr.src_addr.ss_family = AF_INET;
3325                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3326                 if (ret)
3327                         return ret;
3328         }
3329
3330         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3331                 return -EINVAL;
3332
3333         if (id_priv->reuseaddr) {
3334                 ret = cma_bind_listen(id_priv);
3335                 if (ret)
3336                         goto err;
3337         }
3338
3339         id_priv->backlog = backlog;
3340         if (id->device) {
3341                 if (rdma_cap_ib_cm(id->device, 1)) {
3342                         ret = cma_ib_listen(id_priv);
3343                         if (ret)
3344                                 goto err;
3345                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3346                         ret = cma_iw_listen(id_priv, backlog);
3347                         if (ret)
3348                                 goto err;
3349                 } else {
3350                         ret = -ENOSYS;
3351                         goto err;
3352                 }
3353         } else
3354                 cma_listen_on_all(id_priv);
3355
3356         return 0;
3357 err:
3358         id_priv->backlog = 0;
3359         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3360         return ret;
3361 }
3362 EXPORT_SYMBOL(rdma_listen);
3363
3364 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3365 {
3366         struct rdma_id_private *id_priv;
3367         int ret;
3368         struct sockaddr  *daddr;
3369
3370         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3371             addr->sa_family != AF_IB)
3372                 return -EAFNOSUPPORT;
3373
3374         id_priv = container_of(id, struct rdma_id_private, id);
3375         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3376                 return -EINVAL;
3377
3378         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3379         if (ret)
3380                 goto err1;
3381
3382         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3383         if (!cma_any_addr(addr)) {
3384                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3385                 if (ret)
3386                         goto err1;
3387
3388                 ret = cma_acquire_dev(id_priv, NULL);
3389                 if (ret)
3390                         goto err1;
3391         }
3392
3393         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3394                 if (addr->sa_family == AF_INET)
3395                         id_priv->afonly = 1;
3396 #if IS_ENABLED(CONFIG_IPV6)
3397                 else if (addr->sa_family == AF_INET6) {
3398                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3399
3400                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3401                 }
3402 #endif
3403         }
3404         daddr = cma_dst_addr(id_priv);
3405         daddr->sa_family = addr->sa_family;
3406
3407         ret = cma_get_port(id_priv);
3408         if (ret)
3409                 goto err2;
3410
3411         return 0;
3412 err2:
3413         if (id_priv->cma_dev) {
3414                 rdma_restrack_del(&id_priv->res);
3415                 cma_release_dev(id_priv);
3416         }
3417 err1:
3418         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3419         return ret;
3420 }
3421 EXPORT_SYMBOL(rdma_bind_addr);
3422
3423 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3424 {
3425         struct cma_hdr *cma_hdr;
3426
3427         cma_hdr = hdr;
3428         cma_hdr->cma_version = CMA_VERSION;
3429         if (cma_family(id_priv) == AF_INET) {
3430                 struct sockaddr_in *src4, *dst4;
3431
3432                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3433                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3434
3435                 cma_set_ip_ver(cma_hdr, 4);
3436                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3437                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3438                 cma_hdr->port = src4->sin_port;
3439         } else if (cma_family(id_priv) == AF_INET6) {
3440                 struct sockaddr_in6 *src6, *dst6;
3441
3442                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3443                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3444
3445                 cma_set_ip_ver(cma_hdr, 6);
3446                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3447                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3448                 cma_hdr->port = src6->sin6_port;
3449         }
3450         return 0;
3451 }
3452
3453 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3454                                 struct ib_cm_event *ib_event)
3455 {
3456         struct rdma_id_private *id_priv = cm_id->context;
3457         struct rdma_cm_event event;
3458         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3459         int ret = 0;
3460
3461         mutex_lock(&id_priv->handler_mutex);
3462         if (id_priv->state != RDMA_CM_CONNECT)
3463                 goto out;
3464
3465         memset(&event, 0, sizeof event);
3466         switch (ib_event->event) {
3467         case IB_CM_SIDR_REQ_ERROR:
3468                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3469                 event.status = -ETIMEDOUT;
3470                 break;
3471         case IB_CM_SIDR_REP_RECEIVED:
3472                 event.param.ud.private_data = ib_event->private_data;
3473                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3474                 if (rep->status != IB_SIDR_SUCCESS) {
3475                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3476                         event.status = ib_event->param.sidr_rep_rcvd.status;
3477                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3478                                              event.status);
3479                         break;
3480                 }
3481                 ret = cma_set_qkey(id_priv, rep->qkey);
3482                 if (ret) {
3483                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3484                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3485                         event.status = ret;
3486                         break;
3487                 }
3488                 ib_init_ah_attr_from_path(id_priv->id.device,
3489                                           id_priv->id.port_num,
3490                                           id_priv->id.route.path_rec,
3491                                           &event.param.ud.ah_attr);
3492                 event.param.ud.qp_num = rep->qpn;
3493                 event.param.ud.qkey = rep->qkey;
3494                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3495                 event.status = 0;
3496                 break;
3497         default:
3498                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3499                        ib_event->event);
3500                 goto out;
3501         }
3502
3503         ret = id_priv->id.event_handler(&id_priv->id, &event);
3504         if (ret) {
3505                 /* Destroy the CM ID by returning a non-zero value. */
3506                 id_priv->cm_id.ib = NULL;
3507                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3508                 mutex_unlock(&id_priv->handler_mutex);
3509                 rdma_destroy_id(&id_priv->id);
3510                 return ret;
3511         }
3512 out:
3513         mutex_unlock(&id_priv->handler_mutex);
3514         return ret;
3515 }
3516
3517 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3518                               struct rdma_conn_param *conn_param)
3519 {
3520         struct ib_cm_sidr_req_param req;
3521         struct ib_cm_id *id;
3522         void *private_data;
3523         u8 offset;
3524         int ret;
3525
3526         memset(&req, 0, sizeof req);
3527         offset = cma_user_data_offset(id_priv);
3528         req.private_data_len = offset + conn_param->private_data_len;
3529         if (req.private_data_len < conn_param->private_data_len)
3530                 return -EINVAL;
3531
3532         if (req.private_data_len) {
3533                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3534                 if (!private_data)
3535                         return -ENOMEM;
3536         } else {
3537                 private_data = NULL;
3538         }
3539
3540         if (conn_param->private_data && conn_param->private_data_len)
3541                 memcpy(private_data + offset, conn_param->private_data,
3542                        conn_param->private_data_len);
3543
3544         if (private_data) {
3545                 ret = cma_format_hdr(private_data, id_priv);
3546                 if (ret)
3547                         goto out;
3548                 req.private_data = private_data;
3549         }
3550
3551         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3552                              id_priv);
3553         if (IS_ERR(id)) {
3554                 ret = PTR_ERR(id);
3555                 goto out;
3556         }
3557         id_priv->cm_id.ib = id;
3558
3559         req.path = id_priv->id.route.path_rec;
3560         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3561         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3562         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3563
3564         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3565         if (ret) {
3566                 ib_destroy_cm_id(id_priv->cm_id.ib);
3567                 id_priv->cm_id.ib = NULL;
3568         }
3569 out:
3570         kfree(private_data);
3571         return ret;
3572 }
3573
3574 static int cma_connect_ib(struct rdma_id_private *id_priv,
3575                           struct rdma_conn_param *conn_param)
3576 {
3577         struct ib_cm_req_param req;
3578         struct rdma_route *route;
3579         void *private_data;
3580         struct ib_cm_id *id;
3581         u8 offset;
3582         int ret;
3583
3584         memset(&req, 0, sizeof req);
3585         offset = cma_user_data_offset(id_priv);
3586         req.private_data_len = offset + conn_param->private_data_len;
3587         if (req.private_data_len < conn_param->private_data_len)
3588                 return -EINVAL;
3589
3590         if (req.private_data_len) {
3591                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3592                 if (!private_data)
3593                         return -ENOMEM;
3594         } else {
3595                 private_data = NULL;
3596         }
3597
3598         if (conn_param->private_data && conn_param->private_data_len)
3599                 memcpy(private_data + offset, conn_param->private_data,
3600                        conn_param->private_data_len);
3601
3602         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3603         if (IS_ERR(id)) {
3604                 ret = PTR_ERR(id);
3605                 goto out;
3606         }
3607         id_priv->cm_id.ib = id;
3608
3609         route = &id_priv->id.route;
3610         if (private_data) {
3611                 ret = cma_format_hdr(private_data, id_priv);
3612                 if (ret)
3613                         goto out;
3614                 req.private_data = private_data;
3615         }
3616
3617         req.primary_path = &route->path_rec[0];
3618         if (route->num_paths == 2)
3619                 req.alternate_path = &route->path_rec[1];
3620
3621         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3622         req.qp_num = id_priv->qp_num;
3623         req.qp_type = id_priv->id.qp_type;
3624         req.starting_psn = id_priv->seq_num;
3625         req.responder_resources = conn_param->responder_resources;
3626         req.initiator_depth = conn_param->initiator_depth;
3627         req.flow_control = conn_param->flow_control;
3628         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3629         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3630         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3631         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3632         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3633         req.srq = id_priv->srq ? 1 : 0;
3634
3635         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3636 out:
3637         if (ret && !IS_ERR(id)) {
3638                 ib_destroy_cm_id(id);
3639                 id_priv->cm_id.ib = NULL;
3640         }
3641
3642         kfree(private_data);
3643         return ret;
3644 }
3645
3646 static int cma_connect_iw(struct rdma_id_private *id_priv,
3647                           struct rdma_conn_param *conn_param)
3648 {
3649         struct iw_cm_id *cm_id;
3650         int ret;
3651         struct iw_cm_conn_param iw_param;
3652
3653         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3654         if (IS_ERR(cm_id))
3655                 return PTR_ERR(cm_id);
3656
3657         cm_id->tos = id_priv->tos;
3658         id_priv->cm_id.iw = cm_id;
3659
3660         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3661                rdma_addr_size(cma_src_addr(id_priv)));
3662         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3663                rdma_addr_size(cma_dst_addr(id_priv)));
3664
3665         ret = cma_modify_qp_rtr(id_priv, conn_param);
3666         if (ret)
3667                 goto out;
3668
3669         if (conn_param) {
3670                 iw_param.ord = conn_param->initiator_depth;
3671                 iw_param.ird = conn_param->responder_resources;
3672                 iw_param.private_data = conn_param->private_data;
3673                 iw_param.private_data_len = conn_param->private_data_len;
3674                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3675         } else {
3676                 memset(&iw_param, 0, sizeof iw_param);
3677                 iw_param.qpn = id_priv->qp_num;
3678         }
3679         ret = iw_cm_connect(cm_id, &iw_param);
3680 out:
3681         if (ret) {
3682                 iw_destroy_cm_id(cm_id);
3683                 id_priv->cm_id.iw = NULL;
3684         }
3685         return ret;
3686 }
3687
3688 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3689 {
3690         struct rdma_id_private *id_priv;
3691         int ret;
3692
3693         id_priv = container_of(id, struct rdma_id_private, id);
3694         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3695                 return -EINVAL;
3696
3697         if (!id->qp) {
3698                 id_priv->qp_num = conn_param->qp_num;
3699                 id_priv->srq = conn_param->srq;
3700         }
3701
3702         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3703                 if (id->qp_type == IB_QPT_UD)
3704                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3705                 else
3706                         ret = cma_connect_ib(id_priv, conn_param);
3707         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3708                 ret = cma_connect_iw(id_priv, conn_param);
3709         else
3710                 ret = -ENOSYS;
3711         if (ret)
3712                 goto err;
3713
3714         return 0;
3715 err:
3716         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3717         return ret;
3718 }
3719 EXPORT_SYMBOL(rdma_connect);
3720
3721 static int cma_accept_ib(struct rdma_id_private *id_priv,
3722                          struct rdma_conn_param *conn_param)
3723 {
3724         struct ib_cm_rep_param rep;
3725         int ret;
3726
3727         ret = cma_modify_qp_rtr(id_priv, conn_param);
3728         if (ret)
3729                 goto out;
3730
3731         ret = cma_modify_qp_rts(id_priv, conn_param);
3732         if (ret)
3733                 goto out;
3734
3735         memset(&rep, 0, sizeof rep);
3736         rep.qp_num = id_priv->qp_num;
3737         rep.starting_psn = id_priv->seq_num;
3738         rep.private_data = conn_param->private_data;
3739         rep.private_data_len = conn_param->private_data_len;
3740         rep.responder_resources = conn_param->responder_resources;
3741         rep.initiator_depth = conn_param->initiator_depth;
3742         rep.failover_accepted = 0;
3743         rep.flow_control = conn_param->flow_control;
3744         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3745         rep.srq = id_priv->srq ? 1 : 0;
3746
3747         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3748 out:
3749         return ret;
3750 }
3751
3752 static int cma_accept_iw(struct rdma_id_private *id_priv,
3753                   struct rdma_conn_param *conn_param)
3754 {
3755         struct iw_cm_conn_param iw_param;
3756         int ret;
3757
3758         if (!conn_param)
3759                 return -EINVAL;
3760
3761         ret = cma_modify_qp_rtr(id_priv, conn_param);
3762         if (ret)
3763                 return ret;
3764
3765         iw_param.ord = conn_param->initiator_depth;
3766         iw_param.ird = conn_param->responder_resources;
3767         iw_param.private_data = conn_param->private_data;
3768         iw_param.private_data_len = conn_param->private_data_len;
3769         if (id_priv->id.qp) {
3770                 iw_param.qpn = id_priv->qp_num;
3771         } else
3772                 iw_param.qpn = conn_param->qp_num;
3773
3774         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3775 }
3776
3777 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3778                              enum ib_cm_sidr_status status, u32 qkey,
3779                              const void *private_data, int private_data_len)
3780 {
3781         struct ib_cm_sidr_rep_param rep;
3782         int ret;
3783
3784         memset(&rep, 0, sizeof rep);
3785         rep.status = status;
3786         if (status == IB_SIDR_SUCCESS) {
3787                 ret = cma_set_qkey(id_priv, qkey);
3788                 if (ret)
3789                         return ret;
3790                 rep.qp_num = id_priv->qp_num;
3791                 rep.qkey = id_priv->qkey;
3792         }
3793         rep.private_data = private_data;
3794         rep.private_data_len = private_data_len;
3795
3796         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3797 }
3798
3799 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3800                   const char *caller)
3801 {
3802         struct rdma_id_private *id_priv;
3803         int ret;
3804
3805         id_priv = container_of(id, struct rdma_id_private, id);
3806
3807         if (caller)
3808                 id_priv->res.kern_name = caller;
3809         else
3810                 rdma_restrack_set_task(&id_priv->res, current);
3811
3812         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3813                 return -EINVAL;
3814
3815         if (!id->qp && conn_param) {
3816                 id_priv->qp_num = conn_param->qp_num;
3817                 id_priv->srq = conn_param->srq;
3818         }
3819
3820         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3821                 if (id->qp_type == IB_QPT_UD) {
3822                         if (conn_param)
3823                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3824                                                         conn_param->qkey,
3825                                                         conn_param->private_data,
3826                                                         conn_param->private_data_len);
3827                         else
3828                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3829                                                         0, NULL, 0);
3830                 } else {
3831                         if (conn_param)
3832                                 ret = cma_accept_ib(id_priv, conn_param);
3833                         else
3834                                 ret = cma_rep_recv(id_priv);
3835                 }
3836         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3837                 ret = cma_accept_iw(id_priv, conn_param);
3838         else
3839                 ret = -ENOSYS;
3840
3841         if (ret)
3842                 goto reject;
3843
3844         return 0;
3845 reject:
3846         cma_modify_qp_err(id_priv);
3847         rdma_reject(id, NULL, 0);
3848         return ret;
3849 }
3850 EXPORT_SYMBOL(__rdma_accept);
3851
3852 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3853 {
3854         struct rdma_id_private *id_priv;
3855         int ret;
3856
3857         id_priv = container_of(id, struct rdma_id_private, id);
3858         if (!id_priv->cm_id.ib)
3859                 return -EINVAL;
3860
3861         switch (id->device->node_type) {
3862         case RDMA_NODE_IB_CA:
3863                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3864                 break;
3865         default:
3866                 ret = 0;
3867                 break;
3868         }
3869         return ret;
3870 }
3871 EXPORT_SYMBOL(rdma_notify);
3872
3873 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3874                 u8 private_data_len)
3875 {
3876         struct rdma_id_private *id_priv;
3877         int ret;
3878
3879         id_priv = container_of(id, struct rdma_id_private, id);
3880         if (!id_priv->cm_id.ib)
3881                 return -EINVAL;
3882
3883         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3884                 if (id->qp_type == IB_QPT_UD)
3885                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3886                                                 private_data, private_data_len);
3887                 else
3888                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3889                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3890                                              0, private_data, private_data_len);
3891         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3892                 ret = iw_cm_reject(id_priv->cm_id.iw,
3893                                    private_data, private_data_len);
3894         } else
3895                 ret = -ENOSYS;
3896
3897         return ret;
3898 }
3899 EXPORT_SYMBOL(rdma_reject);
3900
3901 int rdma_disconnect(struct rdma_cm_id *id)
3902 {
3903         struct rdma_id_private *id_priv;
3904         int ret;
3905
3906         id_priv = container_of(id, struct rdma_id_private, id);
3907         if (!id_priv->cm_id.ib)
3908                 return -EINVAL;
3909
3910         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3911                 ret = cma_modify_qp_err(id_priv);
3912                 if (ret)
3913                         goto out;
3914                 /* Initiate or respond to a disconnect. */
3915                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3916                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3917         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3918                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3919         } else
3920                 ret = -EINVAL;
3921
3922 out:
3923         return ret;
3924 }
3925 EXPORT_SYMBOL(rdma_disconnect);
3926
3927 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3928 {
3929         struct rdma_id_private *id_priv;
3930         struct cma_multicast *mc = multicast->context;
3931         struct rdma_cm_event event;
3932         int ret = 0;
3933
3934         id_priv = mc->id_priv;
3935         mutex_lock(&id_priv->handler_mutex);
3936         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3937             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3938                 goto out;
3939
3940         if (!status)
3941                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3942         else
3943                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3944                                      status);
3945         mutex_lock(&id_priv->qp_mutex);
3946         if (!status && id_priv->id.qp) {
3947                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3948                                          be16_to_cpu(multicast->rec.mlid));
3949                 if (status)
3950                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3951                                              status);
3952         }
3953         mutex_unlock(&id_priv->qp_mutex);
3954
3955         memset(&event, 0, sizeof event);
3956         event.status = status;
3957         event.param.ud.private_data = mc->context;
3958         if (!status) {
3959                 struct rdma_dev_addr *dev_addr =
3960                         &id_priv->id.route.addr.dev_addr;
3961                 struct net_device *ndev =
3962                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
3963                 enum ib_gid_type gid_type =
3964                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3965                         rdma_start_port(id_priv->cma_dev->device)];
3966
3967                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3968                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
3969                                                id_priv->id.port_num,
3970                                                &multicast->rec,
3971                                                ndev, gid_type,
3972                                                &event.param.ud.ah_attr);
3973                 if (ret)
3974                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3975
3976                 event.param.ud.qp_num = 0xFFFFFF;
3977                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3978                 if (ndev)
3979                         dev_put(ndev);
3980         } else
3981                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3982
3983         ret = id_priv->id.event_handler(&id_priv->id, &event);
3984         if (ret) {
3985                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3986                 mutex_unlock(&id_priv->handler_mutex);
3987                 rdma_destroy_id(&id_priv->id);
3988                 return 0;
3989         }
3990
3991 out:
3992         mutex_unlock(&id_priv->handler_mutex);
3993         return 0;
3994 }
3995
3996 static void cma_set_mgid(struct rdma_id_private *id_priv,
3997                          struct sockaddr *addr, union ib_gid *mgid)
3998 {
3999         unsigned char mc_map[MAX_ADDR_LEN];
4000         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4001         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4002         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4003
4004         if (cma_any_addr(addr)) {
4005                 memset(mgid, 0, sizeof *mgid);
4006         } else if ((addr->sa_family == AF_INET6) &&
4007                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4008                                                                  0xFF10A01B)) {
4009                 /* IPv6 address is an SA assigned MGID. */
4010                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4011         } else if (addr->sa_family == AF_IB) {
4012                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4013         } else if ((addr->sa_family == AF_INET6)) {
4014                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4015                 if (id_priv->id.ps == RDMA_PS_UDP)
4016                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4017                 *mgid = *(union ib_gid *) (mc_map + 4);
4018         } else {
4019                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4020                 if (id_priv->id.ps == RDMA_PS_UDP)
4021                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4022                 *mgid = *(union ib_gid *) (mc_map + 4);
4023         }
4024 }
4025
4026 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4027                                  struct cma_multicast *mc)
4028 {
4029         struct ib_sa_mcmember_rec rec;
4030         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4031         ib_sa_comp_mask comp_mask;
4032         int ret;
4033
4034         ib_addr_get_mgid(dev_addr, &rec.mgid);
4035         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4036                                      &rec.mgid, &rec);
4037         if (ret)
4038                 return ret;
4039
4040         ret = cma_set_qkey(id_priv, 0);
4041         if (ret)
4042                 return ret;
4043
4044         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4045         rec.qkey = cpu_to_be32(id_priv->qkey);
4046         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4047         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4048         rec.join_state = mc->join_state;
4049
4050         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4051             (!ib_sa_sendonly_fullmem_support(&sa_client,
4052                                              id_priv->id.device,
4053                                              id_priv->id.port_num))) {
4054                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4055                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4056                         id_priv->id.device->name, id_priv->id.port_num);
4057                 return -EOPNOTSUPP;
4058         }
4059
4060         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4061                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4062                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4063                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4064                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4065
4066         if (id_priv->id.ps == RDMA_PS_IPOIB)
4067                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4068                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4069                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4070                              IB_SA_MCMEMBER_REC_MTU |
4071                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4072
4073         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4074                                                 id_priv->id.port_num, &rec,
4075                                                 comp_mask, GFP_KERNEL,
4076                                                 cma_ib_mc_handler, mc);
4077         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4078 }
4079
4080 static void iboe_mcast_work_handler(struct work_struct *work)
4081 {
4082         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4083         struct cma_multicast *mc = mw->mc;
4084         struct ib_sa_multicast *m = mc->multicast.ib;
4085
4086         mc->multicast.ib->context = mc;
4087         cma_ib_mc_handler(0, m);
4088         kref_put(&mc->mcref, release_mc);
4089         kfree(mw);
4090 }
4091
4092 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4093                               enum ib_gid_type gid_type)
4094 {
4095         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4096         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4097
4098         if (cma_any_addr(addr)) {
4099                 memset(mgid, 0, sizeof *mgid);
4100         } else if (addr->sa_family == AF_INET6) {
4101                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4102         } else {
4103                 mgid->raw[0] =
4104                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4105                 mgid->raw[1] =
4106                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4107                 mgid->raw[2] = 0;
4108                 mgid->raw[3] = 0;
4109                 mgid->raw[4] = 0;
4110                 mgid->raw[5] = 0;
4111                 mgid->raw[6] = 0;
4112                 mgid->raw[7] = 0;
4113                 mgid->raw[8] = 0;
4114                 mgid->raw[9] = 0;
4115                 mgid->raw[10] = 0xff;
4116                 mgid->raw[11] = 0xff;
4117                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4118         }
4119 }
4120
4121 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4122                                    struct cma_multicast *mc)
4123 {
4124         struct iboe_mcast_work *work;
4125         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4126         int err = 0;
4127         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4128         struct net_device *ndev = NULL;
4129         enum ib_gid_type gid_type;
4130         bool send_only;
4131
4132         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4133
4134         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4135                 return -EINVAL;
4136
4137         work = kzalloc(sizeof *work, GFP_KERNEL);
4138         if (!work)
4139                 return -ENOMEM;
4140
4141         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4142         if (!mc->multicast.ib) {
4143                 err = -ENOMEM;
4144                 goto out1;
4145         }
4146
4147         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4148                    rdma_start_port(id_priv->cma_dev->device)];
4149         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4150
4151         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4152         if (id_priv->id.ps == RDMA_PS_UDP)
4153                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4154
4155         if (dev_addr->bound_dev_if)
4156                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4157         if (!ndev) {
4158                 err = -ENODEV;
4159                 goto out2;
4160         }
4161         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4162         mc->multicast.ib->rec.hop_limit = 1;
4163         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4164
4165         if (addr->sa_family == AF_INET) {
4166                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4167                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4168                         if (!send_only) {
4169                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4170                                                     true);
4171                                 if (!err)
4172                                         mc->igmp_joined = true;
4173                         }
4174                 }
4175         } else {
4176                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4177                         err = -ENOTSUPP;
4178         }
4179         dev_put(ndev);
4180         if (err || !mc->multicast.ib->rec.mtu) {
4181                 if (!err)
4182                         err = -EINVAL;
4183                 goto out2;
4184         }
4185         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4186                     &mc->multicast.ib->rec.port_gid);
4187         work->id = id_priv;
4188         work->mc = mc;
4189         INIT_WORK(&work->work, iboe_mcast_work_handler);
4190         kref_get(&mc->mcref);
4191         queue_work(cma_wq, &work->work);
4192
4193         return 0;
4194
4195 out2:
4196         kfree(mc->multicast.ib);
4197 out1:
4198         kfree(work);
4199         return err;
4200 }
4201
4202 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4203                         u8 join_state, void *context)
4204 {
4205         struct rdma_id_private *id_priv;
4206         struct cma_multicast *mc;
4207         int ret;
4208
4209         if (!id->device)
4210                 return -EINVAL;
4211
4212         id_priv = container_of(id, struct rdma_id_private, id);
4213         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4214             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4215                 return -EINVAL;
4216
4217         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4218         if (!mc)
4219                 return -ENOMEM;
4220
4221         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4222         mc->context = context;
4223         mc->id_priv = id_priv;
4224         mc->igmp_joined = false;
4225         mc->join_state = join_state;
4226         spin_lock(&id_priv->lock);
4227         list_add(&mc->list, &id_priv->mc_list);
4228         spin_unlock(&id_priv->lock);
4229
4230         if (rdma_protocol_roce(id->device, id->port_num)) {
4231                 kref_init(&mc->mcref);
4232                 ret = cma_iboe_join_multicast(id_priv, mc);
4233         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4234                 ret = cma_join_ib_multicast(id_priv, mc);
4235         else
4236                 ret = -ENOSYS;
4237
4238         if (ret) {
4239                 spin_lock_irq(&id_priv->lock);
4240                 list_del(&mc->list);
4241                 spin_unlock_irq(&id_priv->lock);
4242                 kfree(mc);
4243         }
4244         return ret;
4245 }
4246 EXPORT_SYMBOL(rdma_join_multicast);
4247
4248 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4249 {
4250         struct rdma_id_private *id_priv;
4251         struct cma_multicast *mc;
4252
4253         id_priv = container_of(id, struct rdma_id_private, id);
4254         spin_lock_irq(&id_priv->lock);
4255         list_for_each_entry(mc, &id_priv->mc_list, list) {
4256                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4257                         list_del(&mc->list);
4258                         spin_unlock_irq(&id_priv->lock);
4259
4260                         if (id->qp)
4261                                 ib_detach_mcast(id->qp,
4262                                                 &mc->multicast.ib->rec.mgid,
4263                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4264
4265                         BUG_ON(id_priv->cma_dev->device != id->device);
4266
4267                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4268                                 ib_sa_free_multicast(mc->multicast.ib);
4269                                 kfree(mc);
4270                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4271                                 if (mc->igmp_joined) {
4272                                         struct rdma_dev_addr *dev_addr =
4273                                                 &id->route.addr.dev_addr;
4274                                         struct net_device *ndev = NULL;
4275
4276                                         if (dev_addr->bound_dev_if)
4277                                                 ndev = dev_get_by_index(dev_addr->net,
4278                                                                         dev_addr->bound_dev_if);
4279                                         if (ndev) {
4280                                                 cma_igmp_send(ndev,
4281                                                               &mc->multicast.ib->rec.mgid,
4282                                                               false);
4283                                                 dev_put(ndev);
4284                                         }
4285                                         mc->igmp_joined = false;
4286                                 }
4287                                 kref_put(&mc->mcref, release_mc);
4288                         }
4289                         return;
4290                 }
4291         }
4292         spin_unlock_irq(&id_priv->lock);
4293 }
4294 EXPORT_SYMBOL(rdma_leave_multicast);
4295
4296 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4297 {
4298         struct rdma_dev_addr *dev_addr;
4299         struct cma_ndev_work *work;
4300
4301         dev_addr = &id_priv->id.route.addr.dev_addr;
4302
4303         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4304             (net_eq(dev_net(ndev), dev_addr->net)) &&
4305             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4306                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4307                         ndev->name, &id_priv->id);
4308                 work = kzalloc(sizeof *work, GFP_KERNEL);
4309                 if (!work)
4310                         return -ENOMEM;
4311
4312                 INIT_WORK(&work->work, cma_ndev_work_handler);
4313                 work->id = id_priv;
4314                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4315                 atomic_inc(&id_priv->refcount);
4316                 queue_work(cma_wq, &work->work);
4317         }
4318
4319         return 0;
4320 }
4321
4322 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4323                                void *ptr)
4324 {
4325         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4326         struct cma_device *cma_dev;
4327         struct rdma_id_private *id_priv;
4328         int ret = NOTIFY_DONE;
4329
4330         if (event != NETDEV_BONDING_FAILOVER)
4331                 return NOTIFY_DONE;
4332
4333         if (!netif_is_bond_master(ndev))
4334                 return NOTIFY_DONE;
4335
4336         mutex_lock(&lock);
4337         list_for_each_entry(cma_dev, &dev_list, list)
4338                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4339                         ret = cma_netdev_change(ndev, id_priv);
4340                         if (ret)
4341                                 goto out;
4342                 }
4343
4344 out:
4345         mutex_unlock(&lock);
4346         return ret;
4347 }
4348
4349 static struct notifier_block cma_nb = {
4350         .notifier_call = cma_netdev_callback
4351 };
4352
4353 static void cma_add_one(struct ib_device *device)
4354 {
4355         struct cma_device *cma_dev;
4356         struct rdma_id_private *id_priv;
4357         unsigned int i;
4358         unsigned long supported_gids = 0;
4359
4360         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4361         if (!cma_dev)
4362                 return;
4363
4364         cma_dev->device = device;
4365         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4366                                             sizeof(*cma_dev->default_gid_type),
4367                                             GFP_KERNEL);
4368         if (!cma_dev->default_gid_type)
4369                 goto free_cma_dev;
4370
4371         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4372                                             sizeof(*cma_dev->default_roce_tos),
4373                                             GFP_KERNEL);
4374         if (!cma_dev->default_roce_tos)
4375                 goto free_gid_type;
4376
4377         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4378                 supported_gids = roce_gid_type_mask_support(device, i);
4379                 WARN_ON(!supported_gids);
4380                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4381                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4382                                 CMA_PREFERRED_ROCE_GID_TYPE;
4383                 else
4384                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4385                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4386                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4387         }
4388
4389         init_completion(&cma_dev->comp);
4390         atomic_set(&cma_dev->refcount, 1);
4391         INIT_LIST_HEAD(&cma_dev->id_list);
4392         ib_set_client_data(device, &cma_client, cma_dev);
4393
4394         mutex_lock(&lock);
4395         list_add_tail(&cma_dev->list, &dev_list);
4396         list_for_each_entry(id_priv, &listen_any_list, list)
4397                 cma_listen_on_dev(id_priv, cma_dev);
4398         mutex_unlock(&lock);
4399
4400         return;
4401
4402 free_gid_type:
4403         kfree(cma_dev->default_gid_type);
4404
4405 free_cma_dev:
4406         kfree(cma_dev);
4407
4408         return;
4409 }
4410
4411 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4412 {
4413         struct rdma_cm_event event;
4414         enum rdma_cm_state state;
4415         int ret = 0;
4416
4417         /* Record that we want to remove the device */
4418         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4419         if (state == RDMA_CM_DESTROYING)
4420                 return 0;
4421
4422         cma_cancel_operation(id_priv, state);
4423         mutex_lock(&id_priv->handler_mutex);
4424
4425         /* Check for destruction from another callback. */
4426         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4427                 goto out;
4428
4429         memset(&event, 0, sizeof event);
4430         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4431         ret = id_priv->id.event_handler(&id_priv->id, &event);
4432 out:
4433         mutex_unlock(&id_priv->handler_mutex);
4434         return ret;
4435 }
4436
4437 static void cma_process_remove(struct cma_device *cma_dev)
4438 {
4439         struct rdma_id_private *id_priv;
4440         int ret;
4441
4442         mutex_lock(&lock);
4443         while (!list_empty(&cma_dev->id_list)) {
4444                 id_priv = list_entry(cma_dev->id_list.next,
4445                                      struct rdma_id_private, list);
4446
4447                 list_del(&id_priv->listen_list);
4448                 list_del_init(&id_priv->list);
4449                 atomic_inc(&id_priv->refcount);
4450                 mutex_unlock(&lock);
4451
4452                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4453                 cma_deref_id(id_priv);
4454                 if (ret)
4455                         rdma_destroy_id(&id_priv->id);
4456
4457                 mutex_lock(&lock);
4458         }
4459         mutex_unlock(&lock);
4460
4461         cma_deref_dev(cma_dev);
4462         wait_for_completion(&cma_dev->comp);
4463 }
4464
4465 static void cma_remove_one(struct ib_device *device, void *client_data)
4466 {
4467         struct cma_device *cma_dev = client_data;
4468
4469         if (!cma_dev)
4470                 return;
4471
4472         mutex_lock(&lock);
4473         list_del(&cma_dev->list);
4474         mutex_unlock(&lock);
4475
4476         cma_process_remove(cma_dev);
4477         kfree(cma_dev->default_roce_tos);
4478         kfree(cma_dev->default_gid_type);
4479         kfree(cma_dev);
4480 }
4481
4482 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4483 {
4484         struct nlmsghdr *nlh;
4485         struct rdma_cm_id_stats *id_stats;
4486         struct rdma_id_private *id_priv;
4487         struct rdma_cm_id *id = NULL;
4488         struct cma_device *cma_dev;
4489         int i_dev = 0, i_id = 0;
4490
4491         /*
4492          * We export all of the IDs as a sequence of messages.  Each
4493          * ID gets its own netlink message.
4494          */
4495         mutex_lock(&lock);
4496
4497         list_for_each_entry(cma_dev, &dev_list, list) {
4498                 if (i_dev < cb->args[0]) {
4499                         i_dev++;
4500                         continue;
4501                 }
4502
4503                 i_id = 0;
4504                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4505                         if (i_id < cb->args[1]) {
4506                                 i_id++;
4507                                 continue;
4508                         }
4509
4510                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4511                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4512                                                 RDMA_NL_RDMA_CM_ID_STATS,
4513                                                 NLM_F_MULTI);
4514                         if (!id_stats)
4515                                 goto out;
4516
4517                         memset(id_stats, 0, sizeof *id_stats);
4518                         id = &id_priv->id;
4519                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4520                         id_stats->port_num = id->port_num;
4521                         id_stats->bound_dev_if =
4522                                 id->route.addr.dev_addr.bound_dev_if;
4523
4524                         if (ibnl_put_attr(skb, nlh,
4525                                           rdma_addr_size(cma_src_addr(id_priv)),
4526                                           cma_src_addr(id_priv),
4527                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4528                                 goto out;
4529                         if (ibnl_put_attr(skb, nlh,
4530                                           rdma_addr_size(cma_dst_addr(id_priv)),
4531                                           cma_dst_addr(id_priv),
4532                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4533                                 goto out;
4534
4535                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4536                         id_stats->port_space    = id->ps;
4537                         id_stats->cm_state      = id_priv->state;
4538                         id_stats->qp_num        = id_priv->qp_num;
4539                         id_stats->qp_type       = id->qp_type;
4540
4541                         i_id++;
4542                         nlmsg_end(skb, nlh);
4543                 }
4544
4545                 cb->args[1] = 0;
4546                 i_dev++;
4547         }
4548
4549 out:
4550         mutex_unlock(&lock);
4551         cb->args[0] = i_dev;
4552         cb->args[1] = i_id;
4553
4554         return skb->len;
4555 }
4556
4557 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4558         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4559 };
4560
4561 static int cma_init_net(struct net *net)
4562 {
4563         struct cma_pernet *pernet = cma_pernet(net);
4564
4565         idr_init(&pernet->tcp_ps);
4566         idr_init(&pernet->udp_ps);
4567         idr_init(&pernet->ipoib_ps);
4568         idr_init(&pernet->ib_ps);
4569
4570         return 0;
4571 }
4572
4573 static void cma_exit_net(struct net *net)
4574 {
4575         struct cma_pernet *pernet = cma_pernet(net);
4576
4577         idr_destroy(&pernet->tcp_ps);
4578         idr_destroy(&pernet->udp_ps);
4579         idr_destroy(&pernet->ipoib_ps);
4580         idr_destroy(&pernet->ib_ps);
4581 }
4582
4583 static struct pernet_operations cma_pernet_operations = {
4584         .init = cma_init_net,
4585         .exit = cma_exit_net,
4586         .id = &cma_pernet_id,
4587         .size = sizeof(struct cma_pernet),
4588 };
4589
4590 static int __init cma_init(void)
4591 {
4592         int ret;
4593
4594         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4595         if (!cma_wq)
4596                 return -ENOMEM;
4597
4598         ret = register_pernet_subsys(&cma_pernet_operations);
4599         if (ret)
4600                 goto err_wq;
4601
4602         ib_sa_register_client(&sa_client);
4603         register_netdevice_notifier(&cma_nb);
4604
4605         ret = ib_register_client(&cma_client);
4606         if (ret)
4607                 goto err;
4608
4609         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4610         cma_configfs_init();
4611
4612         return 0;
4613
4614 err:
4615         unregister_netdevice_notifier(&cma_nb);
4616         ib_sa_unregister_client(&sa_client);
4617 err_wq:
4618         destroy_workqueue(cma_wq);
4619         return ret;
4620 }
4621
4622 static void __exit cma_cleanup(void)
4623 {
4624         cma_configfs_exit();
4625         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4626         ib_unregister_client(&cma_client);
4627         unregister_netdevice_notifier(&cma_nb);
4628         ib_sa_unregister_client(&sa_client);
4629         unregister_pernet_subsys(&cma_pernet_operations);
4630         destroy_workqueue(cma_wq);
4631 }
4632
4633 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4634
4635 module_init(cma_init);
4636 module_exit(cma_cleanup);