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         u8                      join_state;
 370 };
 371
 372 struct cma_work {
 373         struct work_struct      work;
 374         struct rdma_id_private  *id;
 375         enum rdma_cm_state      old_state;
 376         enum rdma_cm_state      new_state;
 377         struct rdma_cm_event    event;
 378 };
 379
 380 struct cma_ndev_work {
 381         struct work_struct      work;
 382         struct rdma_id_private  *id;
 383         struct rdma_cm_event    event;
 384 };
 385
 386 struct iboe_mcast_work {
 387         struct work_struct       work;
 388         struct rdma_id_private  *id;
 389         struct cma_multicast    *mc;
 390 };
 391
 392 union cma_ip_addr {
 393         struct in6_addr ip6;
 394         struct {
 395                 __be32 pad[3];
 396                 __be32 addr;
 397         } ip4;
 398 };
 399
 400 struct cma_hdr {
 401         u8 cma_version;
 402         u8 ip_version;  /* IP version: 7:4 */
 403         __be16 port;
 404         union cma_ip_addr src_addr;
 405         union cma_ip_addr dst_addr;
 406 };
 407
 408 #define CMA_VERSION 0x00
 409
 410 struct cma_req_info {
 411         struct sockaddr_storage listen_addr_storage;
 412         struct sockaddr_storage src_addr_storage;
 413         struct ib_device *device;
 414         union ib_gid local_gid;
 415         __be64 service_id;
 416         int port;
 417         bool has_gid;
 418         u16 pkey;
 419 };
 420
 421 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 422 {
 423         unsigned long flags;
 424         int ret;
 425
 426         spin_lock_irqsave(&id_priv->lock, flags);
 427         ret = (id_priv->state == comp);
 428         spin_unlock_irqrestore(&id_priv->lock, flags);
 429         return ret;
 430 }
 431
 432 static int cma_comp_exch(struct rdma_id_private *id_priv,
 433                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 434 {
 435         unsigned long flags;
 436         int ret;
 437
 438         spin_lock_irqsave(&id_priv->lock, flags);
 439         if ((ret = (id_priv->state == comp)))
 440                 id_priv->state = exch;
 441         spin_unlock_irqrestore(&id_priv->lock, flags);
 442         return ret;
 443 }
 444
 445 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 446                                    enum rdma_cm_state exch)
 447 {
 448         unsigned long flags;
 449         enum rdma_cm_state old;
 450
 451         spin_lock_irqsave(&id_priv->lock, flags);
 452         old = id_priv->state;
 453         id_priv->state = exch;
 454         spin_unlock_irqrestore(&id_priv->lock, flags);
 455         return old;
 456 }
 457
 458 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 459 {
 460         return hdr->ip_version >> 4;
 461 }
 462
 463 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 464 {
 465         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 466 }
 467
 468 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 469 {
 470         struct in_device *in_dev = NULL;
 471
 472         if (ndev) {
 473                 rtnl_lock();
 474                 in_dev = __in_dev_get_rtnl(ndev);
 475                 if (in_dev) {
 476                         if (join)
 477                                 ip_mc_inc_group(in_dev,
 478                                                 *(__be32 *)(mgid->raw + 12));
 479                         else
 480                                 ip_mc_dec_group(in_dev,
 481                                                 *(__be32 *)(mgid->raw + 12));
 482                 }
 483                 rtnl_unlock();
 484         }
 485         return (in_dev) ? 0 : -ENODEV;
 486 }
 487
 488 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 489                                struct cma_device *cma_dev)
 490 {
 491         cma_ref_dev(cma_dev);
 492         id_priv->cma_dev = cma_dev;
 493         id_priv->id.device = cma_dev->device;
 494         id_priv->id.route.addr.dev_addr.transport =
 495                 rdma_node_get_transport(cma_dev->device->node_type);
 496         list_add_tail(&id_priv->list, &cma_dev->id_list);
 497         rdma_restrack_kadd(&id_priv->res);
 498 }
 499
 500 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 501                               struct cma_device *cma_dev)
 502 {
 503         _cma_attach_to_dev(id_priv, cma_dev);
 504         id_priv->gid_type =
 505                 cma_dev->default_gid_type[id_priv->id.port_num -
 506                                           rdma_start_port(cma_dev->device)];
 507 }
 508
 509 void cma_deref_dev(struct cma_device *cma_dev)
 510 {
 511         if (atomic_dec_and_test(&cma_dev->refcount))
 512                 complete(&cma_dev->comp);
 513 }
 514
 515 static inline void release_mc(struct kref *kref)
 516 {
 517         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 518
 519         kfree(mc->multicast.ib);
 520         kfree(mc);
 521 }
 522
 523 static void cma_release_dev(struct rdma_id_private *id_priv)
 524 {
 525         mutex_lock(&lock);
 526         list_del(&id_priv->list);
 527         cma_deref_dev(id_priv->cma_dev);
 528         id_priv->cma_dev = NULL;
 529         mutex_unlock(&lock);
 530 }
 531
 532 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 533 {
 534         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 535 }
 536
 537 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 538 {
 539         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 540 }
 541
 542 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 543 {
 544         return id_priv->id.route.addr.src_addr.ss_family;
 545 }
 546
 547 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 548 {
 549         struct ib_sa_mcmember_rec rec;
 550         int ret = 0;
 551
 552         if (id_priv->qkey) {
 553                 if (qkey && id_priv->qkey != qkey)
 554                         return -EINVAL;
 555                 return 0;
 556         }
 557
 558         if (qkey) {
 559                 id_priv->qkey = qkey;
 560                 return 0;
 561         }
 562
 563         switch (id_priv->id.ps) {
 564         case RDMA_PS_UDP:
 565         case RDMA_PS_IB:
 566                 id_priv->qkey = RDMA_UDP_QKEY;
 567                 break;
 568         case RDMA_PS_IPOIB:
 569                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 570                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 571                                              id_priv->id.port_num, &rec.mgid,
 572                                              &rec);
 573                 if (!ret)
 574                         id_priv->qkey = be32_to_cpu(rec.qkey);
 575                 break;
 576         default:
 577                 break;
 578         }
 579         return ret;
 580 }
 581
 582 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 583 {
 584         dev_addr->dev_type = ARPHRD_INFINIBAND;
 585         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 586         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 587 }
 588
 589 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 590 {
 591         int ret;
 592
 593         if (addr->sa_family != AF_IB) {
 594                 ret = rdma_translate_ip(addr, dev_addr);
 595         } else {
 596                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 597                 ret = 0;
 598         }
 599
 600         return ret;
 601 }
 602
 603 static const struct ib_gid_attr *
 604 cma_validate_port(struct ib_device *device, u8 port,
 605                   enum ib_gid_type gid_type,
 606                   union ib_gid *gid,
 607                   struct rdma_id_private *id_priv)
 608 {
 609         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 610         int bound_if_index = dev_addr->bound_dev_if;
 611         const struct ib_gid_attr *sgid_attr;
 612         int dev_type = dev_addr->dev_type;
 613         struct net_device *ndev = NULL;
 614
 615         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 616                 return ERR_PTR(-ENODEV);
 617
 618         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 619                 return ERR_PTR(-ENODEV);
 620
 621         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 622                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 623                 if (!ndev)
 624                         return ERR_PTR(-ENODEV);
 625         } else {
 626                 gid_type = IB_GID_TYPE_IB;
 627         }
 628
 629         sgid_attr = rdma_find_gid_by_port(device, gid, gid_type, port, ndev);
 630         if (ndev)
 631                 dev_put(ndev);
 632         return sgid_attr;
 633 }
 634
 635 static void cma_bind_sgid_attr(struct rdma_id_private *id_priv,
 636                                const struct ib_gid_attr *sgid_attr)
 637 {
 638         WARN_ON(id_priv->id.route.addr.dev_addr.sgid_attr);
 639         id_priv->id.route.addr.dev_addr.sgid_attr = sgid_attr;
 640 }
 641
 642 /**
 643  * cma_acquire_dev_by_src_ip - Acquire cma device, port, gid attribute
 644  * based on source ip address.
 645  * @id_priv:    cm_id which should be bound to cma device
 646  *
 647  * cma_acquire_dev_by_src_ip() binds cm id to cma device, port and GID attribute
 648  * based on source IP address. It returns 0 on success or error code otherwise.
 649  * It is applicable to active and passive side cm_id.
 650  */
 651 static int cma_acquire_dev_by_src_ip(struct rdma_id_private *id_priv)
 652 {
 653         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 654         const struct ib_gid_attr *sgid_attr;
 655         union ib_gid gid, iboe_gid, *gidp;
 656         struct cma_device *cma_dev;
 657         enum ib_gid_type gid_type;
 658         int ret = -ENODEV;
 659         u8 port;
 660
 661         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 662             id_priv->id.ps == RDMA_PS_IPOIB)
 663                 return -EINVAL;
 664
 665         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 666                     &iboe_gid);
 667
 668         memcpy(&gid, dev_addr->src_dev_addr +
 669                rdma_addr_gid_offset(dev_addr), sizeof(gid));
 670
 671         mutex_lock(&lock);
 672         list_for_each_entry(cma_dev, &dev_list, list) {
 673                 for (port = rdma_start_port(cma_dev->device);
 674                      port <= rdma_end_port(cma_dev->device); port++) {
 675                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 676                                &iboe_gid : &gid;
 677                         gid_type = cma_dev->default_gid_type[port - 1];
 678                         sgid_attr = cma_validate_port(cma_dev->device, port,
 679                                                       gid_type, gidp, id_priv);
 680                         if (!IS_ERR(sgid_attr)) {
 681                                 id_priv->id.port_num = port;
 682                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 683                                 cma_attach_to_dev(id_priv, cma_dev);
 684                                 ret = 0;
 685                                 goto out;
 686                         }
 687                 }
 688         }
 689 out:
 690         mutex_unlock(&lock);
 691         return ret;
 692 }
 693
 694 /**
 695  * cma_ib_acquire_dev - Acquire cma device, port and SGID attribute
 696  * @id_priv:            cm id to bind to cma device
 697  * @listen_id_priv:     listener cm id to match against
 698  * @req:                Pointer to req structure containaining incoming
 699  *                      request information
 700  * cma_ib_acquire_dev() acquires cma device, port and SGID attribute when
 701  * rdma device matches for listen_id and incoming request. It also verifies
 702  * that a GID table entry is present for the source address.
 703  * Returns 0 on success, or returns error code otherwise.
 704  */
 705 static int cma_ib_acquire_dev(struct rdma_id_private *id_priv,
 706                               const struct rdma_id_private *listen_id_priv,
 707                               struct cma_req_info *req)
 708 {
 709         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 710         const struct ib_gid_attr *sgid_attr;
 711         enum ib_gid_type gid_type;
 712         union ib_gid gid;
 713
 714         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 715             id_priv->id.ps == RDMA_PS_IPOIB)
 716                 return -EINVAL;
 717
 718         if (rdma_protocol_roce(req->device, req->port))
 719                 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 720                             &gid);
 721         else
 722                 memcpy(&gid, dev_addr->src_dev_addr +
 723                        rdma_addr_gid_offset(dev_addr), sizeof(gid));
 724
 725         gid_type = listen_id_priv->cma_dev->default_gid_type[req->port - 1];
 726         sgid_attr = cma_validate_port(req->device, req->port,
 727                                       gid_type, &gid, id_priv);
 728         if (IS_ERR(sgid_attr))
 729                 return PTR_ERR(sgid_attr);
 730
 731         id_priv->id.port_num = req->port;
 732         cma_bind_sgid_attr(id_priv, sgid_attr);
 733         /* Need to acquire lock to protect against reader
 734          * of cma_dev->id_list such as cma_netdev_callback() and
 735          * cma_process_remove().
 736          */
 737         mutex_lock(&lock);
 738         cma_attach_to_dev(id_priv, listen_id_priv->cma_dev);
 739         mutex_unlock(&lock);
 740         return 0;
 741 }
 742
 743 static int cma_iw_acquire_dev(struct rdma_id_private *id_priv,
 744                               const struct rdma_id_private *listen_id_priv)
 745 {
 746         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 747         const struct ib_gid_attr *sgid_attr;
 748         struct cma_device *cma_dev;
 749         enum ib_gid_type gid_type;
 750         int ret = -ENODEV;
 751         union ib_gid gid;
 752         u8 port;
 753
 754         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 755             id_priv->id.ps == RDMA_PS_IPOIB)
 756                 return -EINVAL;
 757
 758         memcpy(&gid, dev_addr->src_dev_addr +
 759                rdma_addr_gid_offset(dev_addr), sizeof(gid));
 760
 761         mutex_lock(&lock);
 762
 763         cma_dev = listen_id_priv->cma_dev;
 764         port = listen_id_priv->id.port_num;
 765         gid_type = listen_id_priv->gid_type;
 766         sgid_attr = cma_validate_port(cma_dev->device, port,
 767                                       gid_type, &gid, id_priv);
 768         if (!IS_ERR(sgid_attr)) {
 769                 id_priv->id.port_num = port;
 770                 cma_bind_sgid_attr(id_priv, sgid_attr);
 771                 ret = 0;
 772                 goto out;
 773         }
 774
 775         list_for_each_entry(cma_dev, &dev_list, list) {
 776                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 777                         if (listen_id_priv->cma_dev == cma_dev &&
 778                             listen_id_priv->id.port_num == port)
 779                                 continue;
 780
 781                         gid_type = cma_dev->default_gid_type[port - 1];
 782                         sgid_attr = cma_validate_port(cma_dev->device, port,
 783                                                       gid_type, &gid, id_priv);
 784                         if (!IS_ERR(sgid_attr)) {
 785                                 id_priv->id.port_num = port;
 786                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 787                                 ret = 0;
 788                                 goto out;
 789                         }
 790                 }
 791         }
 792
 793 out:
 794         if (!ret)
 795                 cma_attach_to_dev(id_priv, cma_dev);
 796
 797         mutex_unlock(&lock);
 798         return ret;
 799 }
 800
 801 /*
 802  * Select the source IB device and address to reach the destination IB address.
 803  */
 804 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 805 {
 806         struct cma_device *cma_dev, *cur_dev;
 807         struct sockaddr_ib *addr;
 808         union ib_gid gid, sgid, *dgid;
 809         u16 pkey, index;
 810         u8 p;
 811         enum ib_port_state port_state;
 812         int i;
 813
 814         cma_dev = NULL;
 815         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 816         dgid = (union ib_gid *) &addr->sib_addr;
 817         pkey = ntohs(addr->sib_pkey);
 818
 819         mutex_lock(&lock);
 820         list_for_each_entry(cur_dev, &dev_list, list) {
 821                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 822                         if (!rdma_cap_af_ib(cur_dev->device, p))
 823                                 continue;
 824
 825                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 826                                 continue;
 827
 828                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 829                                 continue;
 830                         for (i = 0; !rdma_query_gid(cur_dev->device,
 831                                                     p, i, &gid);
 832                              i++) {
 833                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 834                                         cma_dev = cur_dev;
 835                                         sgid = gid;
 836                                         id_priv->id.port_num = p;
 837                                         goto found;
 838                                 }
 839
 840                                 if (!cma_dev && (gid.global.subnet_prefix ==
 841                                     dgid->global.subnet_prefix) &&
 842                                     port_state == IB_PORT_ACTIVE) {
 843                                         cma_dev = cur_dev;
 844                                         sgid = gid;
 845                                         id_priv->id.port_num = p;
 846                                         goto found;
 847                                 }
 848                         }
 849                 }
 850         }
 851         mutex_unlock(&lock);
 852         return -ENODEV;
 853
 854 found:
 855         cma_attach_to_dev(id_priv, cma_dev);
 856         mutex_unlock(&lock);
 857         addr = (struct sockaddr_ib *)cma_src_addr(id_priv);
 858         memcpy(&addr->sib_addr, &sgid, sizeof(sgid));
 859         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 860         return 0;
 861 }
 862
 863 static void cma_deref_id(struct rdma_id_private *id_priv)
 864 {
 865         if (atomic_dec_and_test(&id_priv->refcount))
 866                 complete(&id_priv->comp);
 867 }
 868
 869 struct rdma_cm_id *__rdma_create_id(struct net *net,
 870                                     rdma_cm_event_handler event_handler,
 871                                     void *context, enum rdma_ucm_port_space ps,
 872                                     enum ib_qp_type qp_type, const char *caller)
 873 {
 874         struct rdma_id_private *id_priv;
 875
 876         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 877         if (!id_priv)
 878                 return ERR_PTR(-ENOMEM);
 879
 880         rdma_restrack_set_task(&id_priv->res, caller);
 881         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 882         id_priv->state = RDMA_CM_IDLE;
 883         id_priv->id.context = context;
 884         id_priv->id.event_handler = event_handler;
 885         id_priv->id.ps = ps;
 886         id_priv->id.qp_type = qp_type;
 887         id_priv->tos_set = false;
 888         id_priv->gid_type = IB_GID_TYPE_IB;
 889         spin_lock_init(&id_priv->lock);
 890         mutex_init(&id_priv->qp_mutex);
 891         init_completion(&id_priv->comp);
 892         atomic_set(&id_priv->refcount, 1);
 893         mutex_init(&id_priv->handler_mutex);
 894         INIT_LIST_HEAD(&id_priv->listen_list);
 895         INIT_LIST_HEAD(&id_priv->mc_list);
 896         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 897         id_priv->id.route.addr.dev_addr.net = get_net(net);
 898         id_priv->seq_num &= 0x00ffffff;
 899
 900         return &id_priv->id;
 901 }
 902 EXPORT_SYMBOL(__rdma_create_id);
 903
 904 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 905 {
 906         struct ib_qp_attr qp_attr;
 907         int qp_attr_mask, ret;
 908
 909         qp_attr.qp_state = IB_QPS_INIT;
 910         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 911         if (ret)
 912                 return ret;
 913
 914         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 915         if (ret)
 916                 return ret;
 917
 918         qp_attr.qp_state = IB_QPS_RTR;
 919         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 920         if (ret)
 921                 return ret;
 922
 923         qp_attr.qp_state = IB_QPS_RTS;
 924         qp_attr.sq_psn = 0;
 925         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 926
 927         return ret;
 928 }
 929
 930 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 931 {
 932         struct ib_qp_attr qp_attr;
 933         int qp_attr_mask, ret;
 934
 935         qp_attr.qp_state = IB_QPS_INIT;
 936         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 937         if (ret)
 938                 return ret;
 939
 940         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 941 }
 942
 943 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 944                    struct ib_qp_init_attr *qp_init_attr)
 945 {
 946         struct rdma_id_private *id_priv;
 947         struct ib_qp *qp;
 948         int ret;
 949
 950         id_priv = container_of(id, struct rdma_id_private, id);
 951         if (id->device != pd->device)
 952                 return -EINVAL;
 953
 954         qp_init_attr->port_num = id->port_num;
 955         qp = ib_create_qp(pd, qp_init_attr);
 956         if (IS_ERR(qp))
 957                 return PTR_ERR(qp);
 958
 959         if (id->qp_type == IB_QPT_UD)
 960                 ret = cma_init_ud_qp(id_priv, qp);
 961         else
 962                 ret = cma_init_conn_qp(id_priv, qp);
 963         if (ret)
 964                 goto err;
 965
 966         id->qp = qp;
 967         id_priv->qp_num = qp->qp_num;
 968         id_priv->srq = (qp->srq != NULL);
 969         return 0;
 970 err:
 971         ib_destroy_qp(qp);
 972         return ret;
 973 }
 974 EXPORT_SYMBOL(rdma_create_qp);
 975
 976 void rdma_destroy_qp(struct rdma_cm_id *id)
 977 {
 978         struct rdma_id_private *id_priv;
 979
 980         id_priv = container_of(id, struct rdma_id_private, id);
 981         mutex_lock(&id_priv->qp_mutex);
 982         ib_destroy_qp(id_priv->id.qp);
 983         id_priv->id.qp = NULL;
 984         mutex_unlock(&id_priv->qp_mutex);
 985 }
 986 EXPORT_SYMBOL(rdma_destroy_qp);
 987
 988 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 989                              struct rdma_conn_param *conn_param)
 990 {
 991         struct ib_qp_attr qp_attr;
 992         int qp_attr_mask, ret;
 993
 994         mutex_lock(&id_priv->qp_mutex);
 995         if (!id_priv->id.qp) {
 996                 ret = 0;
 997                 goto out;
 998         }
 999
1000         /* Need to update QP attributes from default values. */
1001         qp_attr.qp_state = IB_QPS_INIT;
1002         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1003         if (ret)
1004                 goto out;
1005
1006         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1007         if (ret)
1008                 goto out;
1009
1010         qp_attr.qp_state = IB_QPS_RTR;
1011         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1012         if (ret)
1013                 goto out;
1014
1015         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
1016
1017         if (conn_param)
1018                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
1019         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1020 out:
1021         mutex_unlock(&id_priv->qp_mutex);
1022         return ret;
1023 }
1024
1025 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
1026                              struct rdma_conn_param *conn_param)
1027 {
1028         struct ib_qp_attr qp_attr;
1029         int qp_attr_mask, ret;
1030
1031         mutex_lock(&id_priv->qp_mutex);
1032         if (!id_priv->id.qp) {
1033                 ret = 0;
1034                 goto out;
1035         }
1036
1037         qp_attr.qp_state = IB_QPS_RTS;
1038         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1039         if (ret)
1040                 goto out;
1041
1042         if (conn_param)
1043                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
1044         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1045 out:
1046         mutex_unlock(&id_priv->qp_mutex);
1047         return ret;
1048 }
1049
1050 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
1051 {
1052         struct ib_qp_attr qp_attr;
1053         int ret;
1054
1055         mutex_lock(&id_priv->qp_mutex);
1056         if (!id_priv->id.qp) {
1057                 ret = 0;
1058                 goto out;
1059         }
1060
1061         qp_attr.qp_state = IB_QPS_ERR;
1062         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
1063 out:
1064         mutex_unlock(&id_priv->qp_mutex);
1065         return ret;
1066 }
1067
1068 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
1069                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
1070 {
1071         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1072         int ret;
1073         u16 pkey;
1074
1075         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
1076                 pkey = 0xffff;
1077         else
1078                 pkey = ib_addr_get_pkey(dev_addr);
1079
1080         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
1081                                   pkey, &qp_attr->pkey_index);
1082         if (ret)
1083                 return ret;
1084
1085         qp_attr->port_num = id_priv->id.port_num;
1086         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1087
1088         if (id_priv->id.qp_type == IB_QPT_UD) {
1089                 ret = cma_set_qkey(id_priv, 0);
1090                 if (ret)
1091                         return ret;
1092
1093                 qp_attr->qkey = id_priv->qkey;
1094                 *qp_attr_mask |= IB_QP_QKEY;
1095         } else {
1096                 qp_attr->qp_access_flags = 0;
1097                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1098         }
1099         return 0;
1100 }
1101
1102 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1103                        int *qp_attr_mask)
1104 {
1105         struct rdma_id_private *id_priv;
1106         int ret = 0;
1107
1108         id_priv = container_of(id, struct rdma_id_private, id);
1109         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1110                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1111                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1112                 else
1113                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1114                                                  qp_attr_mask);
1115
1116                 if (qp_attr->qp_state == IB_QPS_RTR)
1117                         qp_attr->rq_psn = id_priv->seq_num;
1118         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1119                 if (!id_priv->cm_id.iw) {
1120                         qp_attr->qp_access_flags = 0;
1121                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1122                 } else
1123                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1124                                                  qp_attr_mask);
1125                 qp_attr->port_num = id_priv->id.port_num;
1126                 *qp_attr_mask |= IB_QP_PORT;
1127         } else
1128                 ret = -ENOSYS;
1129
1130         return ret;
1131 }
1132 EXPORT_SYMBOL(rdma_init_qp_attr);
1133
1134 static inline bool cma_zero_addr(const struct sockaddr *addr)
1135 {
1136         switch (addr->sa_family) {
1137         case AF_INET:
1138                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1139         case AF_INET6:
1140                 return ipv6_addr_any(&((struct sockaddr_in6 *)addr)->sin6_addr);
1141         case AF_IB:
1142                 return ib_addr_any(&((struct sockaddr_ib *)addr)->sib_addr);
1143         default:
1144                 return false;
1145         }
1146 }
1147
1148 static inline bool cma_loopback_addr(const struct sockaddr *addr)
1149 {
1150         switch (addr->sa_family) {
1151         case AF_INET:
1152                 return ipv4_is_loopback(
1153                         ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1154         case AF_INET6:
1155                 return ipv6_addr_loopback(
1156                         &((struct sockaddr_in6 *)addr)->sin6_addr);
1157         case AF_IB:
1158                 return ib_addr_loopback(
1159                         &((struct sockaddr_ib *)addr)->sib_addr);
1160         default:
1161                 return false;
1162         }
1163 }
1164
1165 static inline bool cma_any_addr(const struct sockaddr *addr)
1166 {
1167         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1168 }
1169
1170 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1171 {
1172         if (src->sa_family != dst->sa_family)
1173                 return -1;
1174
1175         switch (src->sa_family) {
1176         case AF_INET:
1177                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1178                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1179         case AF_INET6:
1180                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1181                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1182         default:
1183                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1184                                    &((struct sockaddr_ib *) dst)->sib_addr);
1185         }
1186 }
1187
1188 static __be16 cma_port(const struct sockaddr *addr)
1189 {
1190         struct sockaddr_ib *sib;
1191
1192         switch (addr->sa_family) {
1193         case AF_INET:
1194                 return ((struct sockaddr_in *) addr)->sin_port;
1195         case AF_INET6:
1196                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1197         case AF_IB:
1198                 sib = (struct sockaddr_ib *) addr;
1199                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1200                                     be64_to_cpu(sib->sib_sid_mask)));
1201         default:
1202                 return 0;
1203         }
1204 }
1205
1206 static inline int cma_any_port(const struct sockaddr *addr)
1207 {
1208         return !cma_port(addr);
1209 }
1210
1211 static void cma_save_ib_info(struct sockaddr *src_addr,
1212                              struct sockaddr *dst_addr,
1213                              const struct rdma_cm_id *listen_id,
1214                              const struct sa_path_rec *path)
1215 {
1216         struct sockaddr_ib *listen_ib, *ib;
1217
1218         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1219         if (src_addr) {
1220                 ib = (struct sockaddr_ib *)src_addr;
1221                 ib->sib_family = AF_IB;
1222                 if (path) {
1223                         ib->sib_pkey = path->pkey;
1224                         ib->sib_flowinfo = path->flow_label;
1225                         memcpy(&ib->sib_addr, &path->sgid, 16);
1226                         ib->sib_sid = path->service_id;
1227                         ib->sib_scope_id = 0;
1228                 } else {
1229                         ib->sib_pkey = listen_ib->sib_pkey;
1230                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1231                         ib->sib_addr = listen_ib->sib_addr;
1232                         ib->sib_sid = listen_ib->sib_sid;
1233                         ib->sib_scope_id = listen_ib->sib_scope_id;
1234                 }
1235                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1236         }
1237         if (dst_addr) {
1238                 ib = (struct sockaddr_ib *)dst_addr;
1239                 ib->sib_family = AF_IB;
1240                 if (path) {
1241                         ib->sib_pkey = path->pkey;
1242                         ib->sib_flowinfo = path->flow_label;
1243                         memcpy(&ib->sib_addr, &path->dgid, 16);
1244                 }
1245         }
1246 }
1247
1248 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1249                               struct sockaddr_in *dst_addr,
1250                               struct cma_hdr *hdr,
1251                               __be16 local_port)
1252 {
1253         if (src_addr) {
1254                 *src_addr = (struct sockaddr_in) {
1255                         .sin_family = AF_INET,
1256                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1257                         .sin_port = local_port,
1258                 };
1259         }
1260
1261         if (dst_addr) {
1262                 *dst_addr = (struct sockaddr_in) {
1263                         .sin_family = AF_INET,
1264                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1265                         .sin_port = hdr->port,
1266                 };
1267         }
1268 }
1269
1270 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1271                               struct sockaddr_in6 *dst_addr,
1272                               struct cma_hdr *hdr,
1273                               __be16 local_port)
1274 {
1275         if (src_addr) {
1276                 *src_addr = (struct sockaddr_in6) {
1277                         .sin6_family = AF_INET6,
1278                         .sin6_addr = hdr->dst_addr.ip6,
1279                         .sin6_port = local_port,
1280                 };
1281         }
1282
1283         if (dst_addr) {
1284                 *dst_addr = (struct sockaddr_in6) {
1285                         .sin6_family = AF_INET6,
1286                         .sin6_addr = hdr->src_addr.ip6,
1287                         .sin6_port = hdr->port,
1288                 };
1289         }
1290 }
1291
1292 static u16 cma_port_from_service_id(__be64 service_id)
1293 {
1294         return (u16)be64_to_cpu(service_id);
1295 }
1296
1297 static int cma_save_ip_info(struct sockaddr *src_addr,
1298                             struct sockaddr *dst_addr,
1299                             const struct ib_cm_event *ib_event,
1300                             __be64 service_id)
1301 {
1302         struct cma_hdr *hdr;
1303         __be16 port;
1304
1305         hdr = ib_event->private_data;
1306         if (hdr->cma_version != CMA_VERSION)
1307                 return -EINVAL;
1308
1309         port = htons(cma_port_from_service_id(service_id));
1310
1311         switch (cma_get_ip_ver(hdr)) {
1312         case 4:
1313                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1314                                   (struct sockaddr_in *)dst_addr, hdr, port);
1315                 break;
1316         case 6:
1317                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1318                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1319                 break;
1320         default:
1321                 return -EAFNOSUPPORT;
1322         }
1323
1324         return 0;
1325 }
1326
1327 static int cma_save_net_info(struct sockaddr *src_addr,
1328                              struct sockaddr *dst_addr,
1329                              const struct rdma_cm_id *listen_id,
1330                              const struct ib_cm_event *ib_event,
1331                              sa_family_t sa_family, __be64 service_id)
1332 {
1333         if (sa_family == AF_IB) {
1334                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1335                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1336                                          ib_event->param.req_rcvd.primary_path);
1337                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1338                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1339                 return 0;
1340         }
1341
1342         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1343 }
1344
1345 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1346                              struct cma_req_info *req)
1347 {
1348         const struct ib_cm_req_event_param *req_param =
1349                 &ib_event->param.req_rcvd;
1350         const struct ib_cm_sidr_req_event_param *sidr_param =
1351                 &ib_event->param.sidr_req_rcvd;
1352
1353         switch (ib_event->event) {
1354         case IB_CM_REQ_RECEIVED:
1355                 req->device     = req_param->listen_id->device;
1356                 req->port       = req_param->port;
1357                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1358                        sizeof(req->local_gid));
1359                 req->has_gid    = true;
1360                 req->service_id = req_param->primary_path->service_id;
1361                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1362                 if (req->pkey != req_param->bth_pkey)
1363                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1364                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1365                                             req_param->bth_pkey, req->pkey);
1366                 break;
1367         case IB_CM_SIDR_REQ_RECEIVED:
1368                 req->device     = sidr_param->listen_id->device;
1369                 req->port       = sidr_param->port;
1370                 req->has_gid    = false;
1371                 req->service_id = sidr_param->service_id;
1372                 req->pkey       = sidr_param->pkey;
1373                 if (req->pkey != sidr_param->bth_pkey)
1374                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1375                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1376                                             sidr_param->bth_pkey, req->pkey);
1377                 break;
1378         default:
1379                 return -EINVAL;
1380         }
1381
1382         return 0;
1383 }
1384
1385 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1386                                   const struct sockaddr_in *dst_addr,
1387                                   const struct sockaddr_in *src_addr)
1388 {
1389         __be32 daddr = dst_addr->sin_addr.s_addr,
1390                saddr = src_addr->sin_addr.s_addr;
1391         struct fib_result res;
1392         struct flowi4 fl4;
1393         int err;
1394         bool ret;
1395
1396         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1397             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1398             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1399             ipv4_is_loopback(saddr))
1400                 return false;
1401
1402         memset(&fl4, 0, sizeof(fl4));
1403         fl4.flowi4_iif = net_dev->ifindex;
1404         fl4.daddr = daddr;
1405         fl4.saddr = saddr;
1406
1407         rcu_read_lock();
1408         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1409         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1410         rcu_read_unlock();
1411
1412         return ret;
1413 }
1414
1415 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1416                                   const struct sockaddr_in6 *dst_addr,
1417                                   const struct sockaddr_in6 *src_addr)
1418 {
1419 #if IS_ENABLED(CONFIG_IPV6)
1420         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1421                            IPV6_ADDR_LINKLOCAL;
1422         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1423                                          &src_addr->sin6_addr, net_dev->ifindex,
1424                                          NULL, strict);
1425         bool ret;
1426
1427         if (!rt)
1428                 return false;
1429
1430         ret = rt->rt6i_idev->dev == net_dev;
1431         ip6_rt_put(rt);
1432
1433         return ret;
1434 #else
1435         return false;
1436 #endif
1437 }
1438
1439 static bool validate_net_dev(struct net_device *net_dev,
1440                              const struct sockaddr *daddr,
1441                              const struct sockaddr *saddr)
1442 {
1443         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1444         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1445         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1446         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1447
1448         switch (daddr->sa_family) {
1449         case AF_INET:
1450                 return saddr->sa_family == AF_INET &&
1451                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1452
1453         case AF_INET6:
1454                 return saddr->sa_family == AF_INET6 &&
1455                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1456
1457         default:
1458                 return false;
1459         }
1460 }
1461
1462 static struct net_device *
1463 roce_get_net_dev_by_cm_event(const struct ib_cm_event *ib_event)
1464 {
1465         const struct ib_gid_attr *sgid_attr = NULL;
1466
1467         if (ib_event->event == IB_CM_REQ_RECEIVED)
1468                 sgid_attr = ib_event->param.req_rcvd.ppath_sgid_attr;
1469         else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1470                 sgid_attr = ib_event->param.sidr_req_rcvd.sgid_attr;
1471
1472         if (!sgid_attr)
1473                 return NULL;
1474         dev_hold(sgid_attr->ndev);
1475         return sgid_attr->ndev;
1476 }
1477
1478 static struct net_device *cma_get_net_dev(const struct ib_cm_event *ib_event,
1479                                           struct cma_req_info *req)
1480 {
1481         struct sockaddr *listen_addr =
1482                         (struct sockaddr *)&req->listen_addr_storage;
1483         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1484         struct net_device *net_dev;
1485         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1486         int err;
1487
1488         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1489                                req->service_id);
1490         if (err)
1491                 return ERR_PTR(err);
1492
1493         if (rdma_protocol_roce(req->device, req->port))
1494                 net_dev = roce_get_net_dev_by_cm_event(ib_event);
1495         else
1496                 net_dev = ib_get_net_dev_by_params(req->device, req->port,
1497                                                    req->pkey,
1498                                                    gid, listen_addr);
1499         if (!net_dev)
1500                 return ERR_PTR(-ENODEV);
1501
1502         return net_dev;
1503 }
1504
1505 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1506 {
1507         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1508 }
1509
1510 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1511                                    const struct cma_hdr *hdr)
1512 {
1513         struct sockaddr *addr = cma_src_addr(id_priv);
1514         __be32 ip4_addr;
1515         struct in6_addr ip6_addr;
1516
1517         if (cma_any_addr(addr) && !id_priv->afonly)
1518                 return true;
1519
1520         switch (addr->sa_family) {
1521         case AF_INET:
1522                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1523                 if (cma_get_ip_ver(hdr) != 4)
1524                         return false;
1525                 if (!cma_any_addr(addr) &&
1526                     hdr->dst_addr.ip4.addr != ip4_addr)
1527                         return false;
1528                 break;
1529         case AF_INET6:
1530                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1531                 if (cma_get_ip_ver(hdr) != 6)
1532                         return false;
1533                 if (!cma_any_addr(addr) &&
1534                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1535                         return false;
1536                 break;
1537         case AF_IB:
1538                 return true;
1539         default:
1540                 return false;
1541         }
1542
1543         return true;
1544 }
1545
1546 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1547 {
1548         struct ib_device *device = id->device;
1549         const int port_num = id->port_num ?: rdma_start_port(device);
1550
1551         return rdma_protocol_roce(device, port_num);
1552 }
1553
1554 static bool cma_is_req_ipv6_ll(const struct cma_req_info *req)
1555 {
1556         const struct sockaddr *daddr =
1557                         (const struct sockaddr *)&req->listen_addr_storage;
1558         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1559
1560         /* Returns true if the req is for IPv6 link local */
1561         return (daddr->sa_family == AF_INET6 &&
1562                 (ipv6_addr_type(&daddr6->sin6_addr) & IPV6_ADDR_LINKLOCAL));
1563 }
1564
1565 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1566                               const struct net_device *net_dev,
1567                               const struct cma_req_info *req)
1568 {
1569         const struct rdma_addr *addr = &id->route.addr;
1570
1571         if (!net_dev)
1572                 /* This request is an AF_IB request */
1573                 return (!id->port_num || id->port_num == req->port) &&
1574                        (addr->src_addr.ss_family == AF_IB);
1575
1576         /*
1577          * If the request is not for IPv6 link local, allow matching
1578          * request to any netdevice of the one or multiport rdma device.
1579          */
1580         if (!cma_is_req_ipv6_ll(req))
1581                 return true;
1582         /*
1583          * Net namespaces must match, and if the listner is listening
1584          * on a specific netdevice than netdevice must match as well.
1585          */
1586         if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1587             (!!addr->dev_addr.bound_dev_if ==
1588              (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1589                 return true;
1590         else
1591                 return false;
1592 }
1593
1594 static struct rdma_id_private *cma_find_listener(
1595                 const struct rdma_bind_list *bind_list,
1596                 const struct ib_cm_id *cm_id,
1597                 const struct ib_cm_event *ib_event,
1598                 const struct cma_req_info *req,
1599                 const struct net_device *net_dev)
1600 {
1601         struct rdma_id_private *id_priv, *id_priv_dev;
1602
1603         if (!bind_list)
1604                 return ERR_PTR(-EINVAL);
1605
1606         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1607                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1608                         if (id_priv->id.device == cm_id->device &&
1609                             cma_match_net_dev(&id_priv->id, net_dev, req))
1610                                 return id_priv;
1611                         list_for_each_entry(id_priv_dev,
1612                                             &id_priv->listen_list,
1613                                             listen_list) {
1614                                 if (id_priv_dev->id.device == cm_id->device &&
1615                                     cma_match_net_dev(&id_priv_dev->id,
1616                                                       net_dev, req))
1617                                         return id_priv_dev;
1618                         }
1619                 }
1620         }
1621
1622         return ERR_PTR(-EINVAL);
1623 }
1624
1625 static struct rdma_id_private *
1626 cma_ib_id_from_event(struct ib_cm_id *cm_id,
1627                      const struct ib_cm_event *ib_event,
1628                      struct cma_req_info *req,
1629                      struct net_device **net_dev)
1630 {
1631         struct rdma_bind_list *bind_list;
1632         struct rdma_id_private *id_priv;
1633         int err;
1634
1635         err = cma_save_req_info(ib_event, req);
1636         if (err)
1637                 return ERR_PTR(err);
1638
1639         *net_dev = cma_get_net_dev(ib_event, req);
1640         if (IS_ERR(*net_dev)) {
1641                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1642                         /* Assuming the protocol is AF_IB */
1643                         *net_dev = NULL;
1644                 } else {
1645                         return ERR_CAST(*net_dev);
1646                 }
1647         }
1648
1649         /*
1650          * Net namespace might be getting deleted while route lookup,
1651          * cm_id lookup is in progress. Therefore, perform netdevice
1652          * validation, cm_id lookup under rcu lock.
1653          * RCU lock along with netdevice state check, synchronizes with
1654          * netdevice migrating to different net namespace and also avoids
1655          * case where net namespace doesn't get deleted while lookup is in
1656          * progress.
1657          * If the device state is not IFF_UP, its properties such as ifindex
1658          * and nd_net cannot be trusted to remain valid without rcu lock.
1659          * net/core/dev.c change_net_namespace() ensures to synchronize with
1660          * ongoing operations on net device after device is closed using
1661          * synchronize_net().
1662          */
1663         rcu_read_lock();
1664         if (*net_dev) {
1665                 /*
1666                  * If netdevice is down, it is likely that it is administratively
1667                  * down or it might be migrating to different namespace.
1668                  * In that case avoid further processing, as the net namespace
1669                  * or ifindex may change.
1670                  */
1671                 if (((*net_dev)->flags & IFF_UP) == 0) {
1672                         id_priv = ERR_PTR(-EHOSTUNREACH);
1673                         goto err;
1674                 }
1675
1676                 if (!validate_net_dev(*net_dev,
1677                                  (struct sockaddr *)&req->listen_addr_storage,
1678                                  (struct sockaddr *)&req->src_addr_storage)) {
1679                         id_priv = ERR_PTR(-EHOSTUNREACH);
1680                         goto err;
1681                 }
1682         }
1683
1684         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1685                                 rdma_ps_from_service_id(req->service_id),
1686                                 cma_port_from_service_id(req->service_id));
1687         id_priv = cma_find_listener(bind_list, cm_id, ib_event, req, *net_dev);
1688 err:
1689         rcu_read_unlock();
1690         if (IS_ERR(id_priv) && *net_dev) {
1691                 dev_put(*net_dev);
1692                 *net_dev = NULL;
1693         }
1694         return id_priv;
1695 }
1696
1697 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1698 {
1699         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1700 }
1701
1702 static void cma_cancel_route(struct rdma_id_private *id_priv)
1703 {
1704         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1705                 if (id_priv->query)
1706                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1707         }
1708 }
1709
1710 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1711 {
1712         struct rdma_id_private *dev_id_priv;
1713
1714         /*
1715          * Remove from listen_any_list to prevent added devices from spawning
1716          * additional listen requests.
1717          */
1718         mutex_lock(&lock);
1719         list_del(&id_priv->list);
1720
1721         while (!list_empty(&id_priv->listen_list)) {
1722                 dev_id_priv = list_entry(id_priv->listen_list.next,
1723                                          struct rdma_id_private, listen_list);
1724                 /* sync with device removal to avoid duplicate destruction */
1725                 list_del_init(&dev_id_priv->list);
1726                 list_del(&dev_id_priv->listen_list);
1727                 mutex_unlock(&lock);
1728
1729                 rdma_destroy_id(&dev_id_priv->id);
1730                 mutex_lock(&lock);
1731         }
1732         mutex_unlock(&lock);
1733 }
1734
1735 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1736                                  enum rdma_cm_state state)
1737 {
1738         switch (state) {
1739         case RDMA_CM_ADDR_QUERY:
1740                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1741                 break;
1742         case RDMA_CM_ROUTE_QUERY:
1743                 cma_cancel_route(id_priv);
1744                 break;
1745         case RDMA_CM_LISTEN:
1746                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1747                         cma_cancel_listens(id_priv);
1748                 break;
1749         default:
1750                 break;
1751         }
1752 }
1753
1754 static void cma_release_port(struct rdma_id_private *id_priv)
1755 {
1756         struct rdma_bind_list *bind_list = id_priv->bind_list;
1757         struct net *net = id_priv->id.route.addr.dev_addr.net;
1758
1759         if (!bind_list)
1760                 return;
1761
1762         mutex_lock(&lock);
1763         hlist_del(&id_priv->node);
1764         if (hlist_empty(&bind_list->owners)) {
1765                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1766                 kfree(bind_list);
1767         }
1768         mutex_unlock(&lock);
1769 }
1770
1771 static void cma_leave_roce_mc_group(struct rdma_id_private *id_priv,
1772                                     struct cma_multicast *mc)
1773 {
1774         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1775         struct net_device *ndev = NULL;
1776
1777         if (dev_addr->bound_dev_if)
1778                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
1779         if (ndev) {
1780                 cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid, false);
1781                 dev_put(ndev);
1782         }
1783         kref_put(&mc->mcref, release_mc);
1784 }
1785
1786 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1787 {
1788         struct cma_multicast *mc;
1789
1790         while (!list_empty(&id_priv->mc_list)) {
1791                 mc = container_of(id_priv->mc_list.next,
1792                                   struct cma_multicast, list);
1793                 list_del(&mc->list);
1794                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1795                                       id_priv->id.port_num)) {
1796                         ib_sa_free_multicast(mc->multicast.ib);
1797                         kfree(mc);
1798                 } else {
1799                         cma_leave_roce_mc_group(id_priv, mc);
1800                 }
1801         }
1802 }
1803
1804 void rdma_destroy_id(struct rdma_cm_id *id)
1805 {
1806         struct rdma_id_private *id_priv;
1807         enum rdma_cm_state state;
1808
1809         id_priv = container_of(id, struct rdma_id_private, id);
1810         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1811         cma_cancel_operation(id_priv, state);
1812
1813         /*
1814          * Wait for any active callback to finish.  New callbacks will find
1815          * the id_priv state set to destroying and abort.
1816          */
1817         mutex_lock(&id_priv->handler_mutex);
1818         mutex_unlock(&id_priv->handler_mutex);
1819
1820         rdma_restrack_del(&id_priv->res);
1821         if (id_priv->cma_dev) {
1822                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1823                         if (id_priv->cm_id.ib)
1824                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1825                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1826                         if (id_priv->cm_id.iw)
1827                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1828                 }
1829                 cma_leave_mc_groups(id_priv);
1830                 cma_release_dev(id_priv);
1831         }
1832
1833         cma_release_port(id_priv);
1834         cma_deref_id(id_priv);
1835         wait_for_completion(&id_priv->comp);
1836
1837         if (id_priv->internal_id)
1838                 cma_deref_id(id_priv->id.context);
1839
1840         kfree(id_priv->id.route.path_rec);
1841
1842         if (id_priv->id.route.addr.dev_addr.sgid_attr)
1843                 rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
1844
1845         put_net(id_priv->id.route.addr.dev_addr.net);
1846         kfree(id_priv);
1847 }
1848 EXPORT_SYMBOL(rdma_destroy_id);
1849
1850 static int cma_rep_recv(struct rdma_id_private *id_priv)
1851 {
1852         int ret;
1853
1854         ret = cma_modify_qp_rtr(id_priv, NULL);
1855         if (ret)
1856                 goto reject;
1857
1858         ret = cma_modify_qp_rts(id_priv, NULL);
1859         if (ret)
1860                 goto reject;
1861
1862         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1863         if (ret)
1864                 goto reject;
1865
1866         return 0;
1867 reject:
1868         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1869         cma_modify_qp_err(id_priv);
1870         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1871                        NULL, 0, NULL, 0);
1872         return ret;
1873 }
1874
1875 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1876                                    const struct ib_cm_rep_event_param *rep_data,
1877                                    void *private_data)
1878 {
1879         event->param.conn.private_data = private_data;
1880         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1881         event->param.conn.responder_resources = rep_data->responder_resources;
1882         event->param.conn.initiator_depth = rep_data->initiator_depth;
1883         event->param.conn.flow_control = rep_data->flow_control;
1884         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1885         event->param.conn.srq = rep_data->srq;
1886         event->param.conn.qp_num = rep_data->remote_qpn;
1887 }
1888
1889 static int cma_ib_handler(struct ib_cm_id *cm_id,
1890                           const struct ib_cm_event *ib_event)
1891 {
1892         struct rdma_id_private *id_priv = cm_id->context;
1893         struct rdma_cm_event event = {};
1894         int ret = 0;
1895
1896         mutex_lock(&id_priv->handler_mutex);
1897         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1898              id_priv->state != RDMA_CM_CONNECT) ||
1899             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1900              id_priv->state != RDMA_CM_DISCONNECT))
1901                 goto out;
1902
1903         switch (ib_event->event) {
1904         case IB_CM_REQ_ERROR:
1905         case IB_CM_REP_ERROR:
1906                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1907                 event.status = -ETIMEDOUT;
1908                 break;
1909         case IB_CM_REP_RECEIVED:
1910                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1911                     (id_priv->id.qp_type != IB_QPT_UD))
1912                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1913                 if (id_priv->id.qp) {
1914                         event.status = cma_rep_recv(id_priv);
1915                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1916                                                      RDMA_CM_EVENT_ESTABLISHED;
1917                 } else {
1918                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1919                 }
1920                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1921                                        ib_event->private_data);
1922                 break;
1923         case IB_CM_RTU_RECEIVED:
1924         case IB_CM_USER_ESTABLISHED:
1925                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1926                 break;
1927         case IB_CM_DREQ_ERROR:
1928                 event.status = -ETIMEDOUT; /* fall through */
1929         case IB_CM_DREQ_RECEIVED:
1930         case IB_CM_DREP_RECEIVED:
1931                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1932                                    RDMA_CM_DISCONNECT))
1933                         goto out;
1934                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1935                 break;
1936         case IB_CM_TIMEWAIT_EXIT:
1937                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1938                 break;
1939         case IB_CM_MRA_RECEIVED:
1940                 /* ignore event */
1941                 goto out;
1942         case IB_CM_REJ_RECEIVED:
1943                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1944                                                                                 ib_event->param.rej_rcvd.reason));
1945                 cma_modify_qp_err(id_priv);
1946                 event.status = ib_event->param.rej_rcvd.reason;
1947                 event.event = RDMA_CM_EVENT_REJECTED;
1948                 event.param.conn.private_data = ib_event->private_data;
1949                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1950                 break;
1951         default:
1952                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1953                        ib_event->event);
1954                 goto out;
1955         }
1956
1957         ret = id_priv->id.event_handler(&id_priv->id, &event);
1958         if (ret) {
1959                 /* Destroy the CM ID by returning a non-zero value. */
1960                 id_priv->cm_id.ib = NULL;
1961                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1962                 mutex_unlock(&id_priv->handler_mutex);
1963                 rdma_destroy_id(&id_priv->id);
1964                 return ret;
1965         }
1966 out:
1967         mutex_unlock(&id_priv->handler_mutex);
1968         return ret;
1969 }
1970
1971 static struct rdma_id_private *
1972 cma_ib_new_conn_id(const struct rdma_cm_id *listen_id,
1973                    const struct ib_cm_event *ib_event,
1974                    struct net_device *net_dev)
1975 {
1976         struct rdma_id_private *listen_id_priv;
1977         struct rdma_id_private *id_priv;
1978         struct rdma_cm_id *id;
1979         struct rdma_route *rt;
1980         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1981         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1982         const __be64 service_id =
1983                 ib_event->param.req_rcvd.primary_path->service_id;
1984         int ret;
1985
1986         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1987         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1988                             listen_id->event_handler, listen_id->context,
1989                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1990                             listen_id_priv->res.kern_name);
1991         if (IS_ERR(id))
1992                 return NULL;
1993
1994         id_priv = container_of(id, struct rdma_id_private, id);
1995         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1996                               (struct sockaddr *)&id->route.addr.dst_addr,
1997                               listen_id, ib_event, ss_family, service_id))
1998                 goto err;
1999
2000         rt = &id->route;
2001         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
2002         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
2003                                      GFP_KERNEL);
2004         if (!rt->path_rec)
2005                 goto err;
2006
2007         rt->path_rec[0] = *path;
2008         if (rt->num_paths == 2)
2009                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
2010
2011         if (net_dev) {
2012                 rdma_copy_src_l2_addr(&rt->addr.dev_addr, net_dev);
2013         } else {
2014                 if (!cma_protocol_roce(listen_id) &&
2015                     cma_any_addr(cma_src_addr(id_priv))) {
2016                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
2017                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
2018                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
2019                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
2020                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
2021                         if (ret)
2022                                 goto err;
2023                 }
2024         }
2025         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
2026
2027         id_priv->state = RDMA_CM_CONNECT;
2028         return id_priv;
2029
2030 err:
2031         rdma_destroy_id(id);
2032         return NULL;
2033 }
2034
2035 static struct rdma_id_private *
2036 cma_ib_new_udp_id(const struct rdma_cm_id *listen_id,
2037                   const struct ib_cm_event *ib_event,
2038                   struct net_device *net_dev)
2039 {
2040         const struct rdma_id_private *listen_id_priv;
2041         struct rdma_id_private *id_priv;
2042         struct rdma_cm_id *id;
2043         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
2044         struct net *net = listen_id->route.addr.dev_addr.net;
2045         int ret;
2046
2047         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
2048         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
2049                               listen_id->ps, IB_QPT_UD,
2050                               listen_id_priv->res.kern_name);
2051         if (IS_ERR(id))
2052                 return NULL;
2053
2054         id_priv = container_of(id, struct rdma_id_private, id);
2055         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
2056                               (struct sockaddr *)&id->route.addr.dst_addr,
2057                               listen_id, ib_event, ss_family,
2058                               ib_event->param.sidr_req_rcvd.service_id))
2059                 goto err;
2060
2061         if (net_dev) {
2062                 rdma_copy_src_l2_addr(&id->route.addr.dev_addr, net_dev);
2063         } else {
2064                 if (!cma_any_addr(cma_src_addr(id_priv))) {
2065                         ret = cma_translate_addr(cma_src_addr(id_priv),
2066                                                  &id->route.addr.dev_addr);
2067                         if (ret)
2068                                 goto err;
2069                 }
2070         }
2071
2072         id_priv->state = RDMA_CM_CONNECT;
2073         return id_priv;
2074 err:
2075         rdma_destroy_id(id);
2076         return NULL;
2077 }
2078
2079 static void cma_set_req_event_data(struct rdma_cm_event *event,
2080                                    const struct ib_cm_req_event_param *req_data,
2081                                    void *private_data, int offset)
2082 {
2083         event->param.conn.private_data = private_data + offset;
2084         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
2085         event->param.conn.responder_resources = req_data->responder_resources;
2086         event->param.conn.initiator_depth = req_data->initiator_depth;
2087         event->param.conn.flow_control = req_data->flow_control;
2088         event->param.conn.retry_count = req_data->retry_count;
2089         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
2090         event->param.conn.srq = req_data->srq;
2091         event->param.conn.qp_num = req_data->remote_qpn;
2092 }
2093
2094 static int cma_ib_check_req_qp_type(const struct rdma_cm_id *id,
2095                                     const struct ib_cm_event *ib_event)
2096 {
2097         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
2098                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
2099                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
2100                  (id->qp_type == IB_QPT_UD)) ||
2101                 (!id->qp_type));
2102 }
2103
2104 static int cma_ib_req_handler(struct ib_cm_id *cm_id,
2105                               const struct ib_cm_event *ib_event)
2106 {
2107         struct rdma_id_private *listen_id, *conn_id = NULL;
2108         struct rdma_cm_event event = {};
2109         struct cma_req_info req = {};
2110         struct net_device *net_dev;
2111         u8 offset;
2112         int ret;
2113
2114         listen_id = cma_ib_id_from_event(cm_id, ib_event, &req, &net_dev);
2115         if (IS_ERR(listen_id))
2116                 return PTR_ERR(listen_id);
2117
2118         if (!cma_ib_check_req_qp_type(&listen_id->id, ib_event)) {
2119                 ret = -EINVAL;
2120                 goto net_dev_put;
2121         }
2122
2123         mutex_lock(&listen_id->handler_mutex);
2124         if (listen_id->state != RDMA_CM_LISTEN) {
2125                 ret = -ECONNABORTED;
2126                 goto err1;
2127         }
2128
2129         offset = cma_user_data_offset(listen_id);
2130         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2131         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
2132                 conn_id = cma_ib_new_udp_id(&listen_id->id, ib_event, net_dev);
2133                 event.param.ud.private_data = ib_event->private_data + offset;
2134                 event.param.ud.private_data_len =
2135                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
2136         } else {
2137                 conn_id = cma_ib_new_conn_id(&listen_id->id, ib_event, net_dev);
2138                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
2139                                        ib_event->private_data, offset);
2140         }
2141         if (!conn_id) {
2142                 ret = -ENOMEM;
2143                 goto err1;
2144         }
2145
2146         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2147         ret = cma_ib_acquire_dev(conn_id, listen_id, &req);
2148         if (ret)
2149                 goto err2;
2150
2151         conn_id->cm_id.ib = cm_id;
2152         cm_id->context = conn_id;
2153         cm_id->cm_handler = cma_ib_handler;
2154
2155         /*
2156          * Protect against the user destroying conn_id from another thread
2157          * until we're done accessing it.
2158          */
2159         atomic_inc(&conn_id->refcount);
2160         ret = conn_id->id.event_handler(&conn_id->id, &event);
2161         if (ret)
2162                 goto err3;
2163         /*
2164          * Acquire mutex to prevent user executing rdma_destroy_id()
2165          * while we're accessing the cm_id.
2166          */
2167         mutex_lock(&lock);
2168         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2169             (conn_id->id.qp_type != IB_QPT_UD))
2170                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2171         mutex_unlock(&lock);
2172         mutex_unlock(&conn_id->handler_mutex);
2173         mutex_unlock(&listen_id->handler_mutex);
2174         cma_deref_id(conn_id);
2175         if (net_dev)
2176                 dev_put(net_dev);
2177         return 0;
2178
2179 err3:
2180         cma_deref_id(conn_id);
2181         /* Destroy the CM ID by returning a non-zero value. */
2182         conn_id->cm_id.ib = NULL;
2183 err2:
2184         cma_exch(conn_id, RDMA_CM_DESTROYING);
2185         mutex_unlock(&conn_id->handler_mutex);
2186 err1:
2187         mutex_unlock(&listen_id->handler_mutex);
2188         if (conn_id)
2189                 rdma_destroy_id(&conn_id->id);
2190
2191 net_dev_put:
2192         if (net_dev)
2193                 dev_put(net_dev);
2194
2195         return ret;
2196 }
2197
2198 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2199 {
2200         if (addr->sa_family == AF_IB)
2201                 return ((struct sockaddr_ib *) addr)->sib_sid;
2202
2203         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2204 }
2205 EXPORT_SYMBOL(rdma_get_service_id);
2206
2207 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2208                     union ib_gid *dgid)
2209 {
2210         struct rdma_addr *addr = &cm_id->route.addr;
2211
2212         if (!cm_id->device) {
2213                 if (sgid)
2214                         memset(sgid, 0, sizeof(*sgid));
2215                 if (dgid)
2216                         memset(dgid, 0, sizeof(*dgid));
2217                 return;
2218         }
2219
2220         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2221                 if (sgid)
2222                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2223                 if (dgid)
2224                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2225         } else {
2226                 if (sgid)
2227                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2228                 if (dgid)
2229                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2230         }
2231 }
2232 EXPORT_SYMBOL(rdma_read_gids);
2233
2234 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2235 {
2236         struct rdma_id_private *id_priv = iw_id->context;
2237         struct rdma_cm_event event = {};
2238         int ret = 0;
2239         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2240         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2241
2242         mutex_lock(&id_priv->handler_mutex);
2243         if (id_priv->state != RDMA_CM_CONNECT)
2244                 goto out;
2245
2246         switch (iw_event->event) {
2247         case IW_CM_EVENT_CLOSE:
2248                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2249                 break;
2250         case IW_CM_EVENT_CONNECT_REPLY:
2251                 memcpy(cma_src_addr(id_priv), laddr,
2252                        rdma_addr_size(laddr));
2253                 memcpy(cma_dst_addr(id_priv), raddr,
2254                        rdma_addr_size(raddr));
2255                 switch (iw_event->status) {
2256                 case 0:
2257                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2258                         event.param.conn.initiator_depth = iw_event->ird;
2259                         event.param.conn.responder_resources = iw_event->ord;
2260                         break;
2261                 case -ECONNRESET:
2262                 case -ECONNREFUSED:
2263                         event.event = RDMA_CM_EVENT_REJECTED;
2264                         break;
2265                 case -ETIMEDOUT:
2266                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2267                         break;
2268                 default:
2269                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2270                         break;
2271                 }
2272                 break;
2273         case IW_CM_EVENT_ESTABLISHED:
2274                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2275                 event.param.conn.initiator_depth = iw_event->ird;
2276                 event.param.conn.responder_resources = iw_event->ord;
2277                 break;
2278         default:
2279                 goto out;
2280         }
2281
2282         event.status = iw_event->status;
2283         event.param.conn.private_data = iw_event->private_data;
2284         event.param.conn.private_data_len = iw_event->private_data_len;
2285         ret = id_priv->id.event_handler(&id_priv->id, &event);
2286         if (ret) {
2287                 /* Destroy the CM ID by returning a non-zero value. */
2288                 id_priv->cm_id.iw = NULL;
2289                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2290                 mutex_unlock(&id_priv->handler_mutex);
2291                 rdma_destroy_id(&id_priv->id);
2292                 return ret;
2293         }
2294
2295 out:
2296         mutex_unlock(&id_priv->handler_mutex);
2297         return ret;
2298 }
2299
2300 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2301                                struct iw_cm_event *iw_event)
2302 {
2303         struct rdma_cm_id *new_cm_id;
2304         struct rdma_id_private *listen_id, *conn_id;
2305         struct rdma_cm_event event = {};
2306         int ret = -ECONNABORTED;
2307         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2308         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2309
2310         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2311         event.param.conn.private_data = iw_event->private_data;
2312         event.param.conn.private_data_len = iw_event->private_data_len;
2313         event.param.conn.initiator_depth = iw_event->ird;
2314         event.param.conn.responder_resources = iw_event->ord;
2315
2316         listen_id = cm_id->context;
2317
2318         mutex_lock(&listen_id->handler_mutex);
2319         if (listen_id->state != RDMA_CM_LISTEN)
2320                 goto out;
2321
2322         /* Create a new RDMA id for the new IW CM ID */
2323         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2324                                      listen_id->id.event_handler,
2325                                      listen_id->id.context,
2326                                      RDMA_PS_TCP, IB_QPT_RC,
2327                                      listen_id->res.kern_name);
2328         if (IS_ERR(new_cm_id)) {
2329                 ret = -ENOMEM;
2330                 goto out;
2331         }
2332         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2333         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2334         conn_id->state = RDMA_CM_CONNECT;
2335
2336         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2337         if (ret) {
2338                 mutex_unlock(&conn_id->handler_mutex);
2339                 rdma_destroy_id(new_cm_id);
2340                 goto out;
2341         }
2342
2343         ret = cma_iw_acquire_dev(conn_id, listen_id);
2344         if (ret) {
2345                 mutex_unlock(&conn_id->handler_mutex);
2346                 rdma_destroy_id(new_cm_id);
2347                 goto out;
2348         }
2349
2350         conn_id->cm_id.iw = cm_id;
2351         cm_id->context = conn_id;
2352         cm_id->cm_handler = cma_iw_handler;
2353
2354         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2355         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2356
2357         /*
2358          * Protect against the user destroying conn_id from another thread
2359          * until we're done accessing it.
2360          */
2361         atomic_inc(&conn_id->refcount);
2362         ret = conn_id->id.event_handler(&conn_id->id, &event);
2363         if (ret) {
2364                 /* User wants to destroy the CM ID */
2365                 conn_id->cm_id.iw = NULL;
2366                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2367                 mutex_unlock(&conn_id->handler_mutex);
2368                 cma_deref_id(conn_id);
2369                 rdma_destroy_id(&conn_id->id);
2370                 goto out;
2371         }
2372
2373         mutex_unlock(&conn_id->handler_mutex);
2374         cma_deref_id(conn_id);
2375
2376 out:
2377         mutex_unlock(&listen_id->handler_mutex);
2378         return ret;
2379 }
2380
2381 static int cma_ib_listen(struct rdma_id_private *id_priv)
2382 {
2383         struct sockaddr *addr;
2384         struct ib_cm_id *id;
2385         __be64 svc_id;
2386
2387         addr = cma_src_addr(id_priv);
2388         svc_id = rdma_get_service_id(&id_priv->id, addr);
2389         id = ib_cm_insert_listen(id_priv->id.device,
2390                                  cma_ib_req_handler, svc_id);
2391         if (IS_ERR(id))
2392                 return PTR_ERR(id);
2393         id_priv->cm_id.ib = id;
2394
2395         return 0;
2396 }
2397
2398 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2399 {
2400         int ret;
2401         struct iw_cm_id *id;
2402
2403         id = iw_create_cm_id(id_priv->id.device,
2404                              iw_conn_req_handler,
2405                              id_priv);
2406         if (IS_ERR(id))
2407                 return PTR_ERR(id);
2408
2409         id->tos = id_priv->tos;
2410         id_priv->cm_id.iw = id;
2411
2412         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2413                rdma_addr_size(cma_src_addr(id_priv)));
2414
2415         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2416
2417         if (ret) {
2418                 iw_destroy_cm_id(id_priv->cm_id.iw);
2419                 id_priv->cm_id.iw = NULL;
2420         }
2421
2422         return ret;
2423 }
2424
2425 static int cma_listen_handler(struct rdma_cm_id *id,
2426                               struct rdma_cm_event *event)
2427 {
2428         struct rdma_id_private *id_priv = id->context;
2429
2430         id->context = id_priv->id.context;
2431         id->event_handler = id_priv->id.event_handler;
2432         return id_priv->id.event_handler(id, event);
2433 }
2434
2435 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2436                               struct cma_device *cma_dev)
2437 {
2438         struct rdma_id_private *dev_id_priv;
2439         struct rdma_cm_id *id;
2440         struct net *net = id_priv->id.route.addr.dev_addr.net;
2441         int ret;
2442
2443         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2444                 return;
2445
2446         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2447                               id_priv->id.qp_type, id_priv->res.kern_name);
2448         if (IS_ERR(id))
2449                 return;
2450
2451         dev_id_priv = container_of(id, struct rdma_id_private, id);
2452
2453         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2454         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2455                rdma_addr_size(cma_src_addr(id_priv)));
2456
2457         _cma_attach_to_dev(dev_id_priv, cma_dev);
2458         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2459         atomic_inc(&id_priv->refcount);
2460         dev_id_priv->internal_id = 1;
2461         dev_id_priv->afonly = id_priv->afonly;
2462
2463         ret = rdma_listen(id, id_priv->backlog);
2464         if (ret)
2465                 dev_warn(&cma_dev->device->dev,
2466                          "RDMA CMA: cma_listen_on_dev, error %d\n", ret);
2467 }
2468
2469 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2470 {
2471         struct cma_device *cma_dev;
2472
2473         mutex_lock(&lock);
2474         list_add_tail(&id_priv->list, &listen_any_list);
2475         list_for_each_entry(cma_dev, &dev_list, list)
2476                 cma_listen_on_dev(id_priv, cma_dev);
2477         mutex_unlock(&lock);
2478 }
2479
2480 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2481 {
2482         struct rdma_id_private *id_priv;
2483
2484         id_priv = container_of(id, struct rdma_id_private, id);
2485         id_priv->tos = (u8) tos;
2486         id_priv->tos_set = true;
2487 }
2488 EXPORT_SYMBOL(rdma_set_service_type);
2489
2490 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2491                               void *context)
2492 {
2493         struct cma_work *work = context;
2494         struct rdma_route *route;
2495
2496         route = &work->id->id.route;
2497
2498         if (!status) {
2499                 route->num_paths = 1;
2500                 *route->path_rec = *path_rec;
2501         } else {
2502                 work->old_state = RDMA_CM_ROUTE_QUERY;
2503                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2504                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2505                 work->event.status = status;
2506                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2507                                      status);
2508         }
2509
2510         queue_work(cma_wq, &work->work);
2511 }
2512
2513 static int cma_query_ib_route(struct rdma_id_private *id_priv,
2514                               unsigned long timeout_ms, struct cma_work *work)
2515 {
2516         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2517         struct sa_path_rec path_rec;
2518         ib_sa_comp_mask comp_mask;
2519         struct sockaddr_in6 *sin6;
2520         struct sockaddr_ib *sib;
2521
2522         memset(&path_rec, 0, sizeof path_rec);
2523
2524         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2525                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2526         else
2527                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2528         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2529         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2530         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2531         path_rec.numb_path = 1;
2532         path_rec.reversible = 1;
2533         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2534                                                   cma_dst_addr(id_priv));
2535
2536         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2537                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2538                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2539
2540         switch (cma_family(id_priv)) {
2541         case AF_INET:
2542                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2543                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2544                 break;
2545         case AF_INET6:
2546                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2547                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2548                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2549                 break;
2550         case AF_IB:
2551                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2552                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2553                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2554                 break;
2555         }
2556
2557         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2558                                                id_priv->id.port_num, &path_rec,
2559                                                comp_mask, timeout_ms,
2560                                                GFP_KERNEL, cma_query_handler,
2561                                                work, &id_priv->query);
2562
2563         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2564 }
2565
2566 static void cma_work_handler(struct work_struct *_work)
2567 {
2568         struct cma_work *work = container_of(_work, struct cma_work, work);
2569         struct rdma_id_private *id_priv = work->id;
2570         int destroy = 0;
2571
2572         mutex_lock(&id_priv->handler_mutex);
2573         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2574                 goto out;
2575
2576         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2577                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2578                 destroy = 1;
2579         }
2580 out:
2581         mutex_unlock(&id_priv->handler_mutex);
2582         cma_deref_id(id_priv);
2583         if (destroy)
2584                 rdma_destroy_id(&id_priv->id);
2585         kfree(work);
2586 }
2587
2588 static void cma_ndev_work_handler(struct work_struct *_work)
2589 {
2590         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2591         struct rdma_id_private *id_priv = work->id;
2592         int destroy = 0;
2593
2594         mutex_lock(&id_priv->handler_mutex);
2595         if (id_priv->state == RDMA_CM_DESTROYING ||
2596             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2597                 goto out;
2598
2599         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2600                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2601                 destroy = 1;
2602         }
2603
2604 out:
2605         mutex_unlock(&id_priv->handler_mutex);
2606         cma_deref_id(id_priv);
2607         if (destroy)
2608                 rdma_destroy_id(&id_priv->id);
2609         kfree(work);
2610 }
2611
2612 static void cma_init_resolve_route_work(struct cma_work *work,
2613                                         struct rdma_id_private *id_priv)
2614 {
2615         work->id = id_priv;
2616         INIT_WORK(&work->work, cma_work_handler);
2617         work->old_state = RDMA_CM_ROUTE_QUERY;
2618         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2619         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2620 }
2621
2622 static void cma_init_resolve_addr_work(struct cma_work *work,
2623                                        struct rdma_id_private *id_priv)
2624 {
2625         work->id = id_priv;
2626         INIT_WORK(&work->work, cma_work_handler);
2627         work->old_state = RDMA_CM_ADDR_QUERY;
2628         work->new_state = RDMA_CM_ADDR_RESOLVED;
2629         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2630 }
2631
2632 static int cma_resolve_ib_route(struct rdma_id_private *id_priv,
2633                                 unsigned long timeout_ms)
2634 {
2635         struct rdma_route *route = &id_priv->id.route;
2636         struct cma_work *work;
2637         int ret;
2638
2639         work = kzalloc(sizeof *work, GFP_KERNEL);
2640         if (!work)
2641                 return -ENOMEM;
2642
2643         cma_init_resolve_route_work(work, id_priv);
2644
2645         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2646         if (!route->path_rec) {
2647                 ret = -ENOMEM;
2648                 goto err1;
2649         }
2650
2651         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2652         if (ret)
2653                 goto err2;
2654
2655         return 0;
2656 err2:
2657         kfree(route->path_rec);
2658         route->path_rec = NULL;
2659 err1:
2660         kfree(work);
2661         return ret;
2662 }
2663
2664 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2665                                            unsigned long supported_gids,
2666                                            enum ib_gid_type default_gid)
2667 {
2668         if ((network_type == RDMA_NETWORK_IPV4 ||
2669              network_type == RDMA_NETWORK_IPV6) &&
2670             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2671                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2672
2673         return default_gid;
2674 }
2675
2676 /*
2677  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2678  * path record type based on GID type.
2679  * It also sets up other L2 fields which includes destination mac address
2680  * netdev ifindex, of the path record.
2681  * It returns the netdev of the bound interface for this path record entry.
2682  */
2683 static struct net_device *
2684 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2685 {
2686         struct rdma_route *route = &id_priv->id.route;
2687         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2688         struct rdma_addr *addr = &route->addr;
2689         unsigned long supported_gids;
2690         struct net_device *ndev;
2691
2692         if (!addr->dev_addr.bound_dev_if)
2693                 return NULL;
2694
2695         ndev = dev_get_by_index(addr->dev_addr.net,
2696                                 addr->dev_addr.bound_dev_if);
2697         if (!ndev)
2698                 return NULL;
2699
2700         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2701                                                     id_priv->id.port_num);
2702         gid_type = cma_route_gid_type(addr->dev_addr.network,
2703                                       supported_gids,
2704                                       id_priv->gid_type);
2705         /* Use the hint from IP Stack to select GID Type */
2706         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2707                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2708         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2709
2710         route->path_rec->roce.route_resolved = true;
2711         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2712         return ndev;
2713 }
2714
2715 int rdma_set_ib_path(struct rdma_cm_id *id,
2716                      struct sa_path_rec *path_rec)
2717 {
2718         struct rdma_id_private *id_priv;
2719         struct net_device *ndev;
2720         int ret;
2721
2722         id_priv = container_of(id, struct rdma_id_private, id);
2723         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2724                            RDMA_CM_ROUTE_RESOLVED))
2725                 return -EINVAL;
2726
2727         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2728                                      GFP_KERNEL);
2729         if (!id->route.path_rec) {
2730                 ret = -ENOMEM;
2731                 goto err;
2732         }
2733
2734         if (rdma_protocol_roce(id->device, id->port_num)) {
2735                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2736                 if (!ndev) {
2737                         ret = -ENODEV;
2738                         goto err_free;
2739                 }
2740                 dev_put(ndev);
2741         }
2742
2743         id->route.num_paths = 1;
2744         return 0;
2745
2746 err_free:
2747         kfree(id->route.path_rec);
2748         id->route.path_rec = NULL;
2749 err:
2750         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2751         return ret;
2752 }
2753 EXPORT_SYMBOL(rdma_set_ib_path);
2754
2755 static int cma_resolve_iw_route(struct rdma_id_private *id_priv)
2756 {
2757         struct cma_work *work;
2758
2759         work = kzalloc(sizeof *work, GFP_KERNEL);
2760         if (!work)
2761                 return -ENOMEM;
2762
2763         cma_init_resolve_route_work(work, id_priv);
2764         queue_work(cma_wq, &work->work);
2765         return 0;
2766 }
2767
2768 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2769 {
2770         int prio;
2771         struct net_device *dev;
2772
2773         prio = rt_tos2priority(tos);
2774         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2775         if (dev->num_tc)
2776                 return netdev_get_prio_tc_map(dev, prio);
2777
2778 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2779         if (is_vlan_dev(ndev))
2780                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2781                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2782 #endif
2783         return 0;
2784 }
2785
2786 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2787 {
2788         struct rdma_route *route = &id_priv->id.route;
2789         struct rdma_addr *addr = &route->addr;
2790         struct cma_work *work;
2791         int ret;
2792         struct net_device *ndev;
2793
2794         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2795                                         rdma_start_port(id_priv->cma_dev->device)];
2796         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2797
2798
2799         work = kzalloc(sizeof *work, GFP_KERNEL);
2800         if (!work)
2801                 return -ENOMEM;
2802
2803         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2804         if (!route->path_rec) {
2805                 ret = -ENOMEM;
2806                 goto err1;
2807         }
2808
2809         route->num_paths = 1;
2810
2811         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2812         if (!ndev) {
2813                 ret = -ENODEV;
2814                 goto err2;
2815         }
2816
2817         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2818                     &route->path_rec->sgid);
2819         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2820                     &route->path_rec->dgid);
2821
2822         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2823                 /* TODO: get the hoplimit from the inet/inet6 device */
2824                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2825         else
2826                 route->path_rec->hop_limit = 1;
2827         route->path_rec->reversible = 1;
2828         route->path_rec->pkey = cpu_to_be16(0xffff);
2829         route->path_rec->mtu_selector = IB_SA_EQ;
2830         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2831         route->path_rec->traffic_class = tos;
2832         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2833         route->path_rec->rate_selector = IB_SA_EQ;
2834         route->path_rec->rate = iboe_get_rate(ndev);
2835         dev_put(ndev);
2836         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2837         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2838         if (!route->path_rec->mtu) {
2839                 ret = -EINVAL;
2840                 goto err2;
2841         }
2842
2843         cma_init_resolve_route_work(work, id_priv);
2844         queue_work(cma_wq, &work->work);
2845
2846         return 0;
2847
2848 err2:
2849         kfree(route->path_rec);
2850         route->path_rec = NULL;
2851 err1:
2852         kfree(work);
2853         return ret;
2854 }
2855
2856 int rdma_resolve_route(struct rdma_cm_id *id, unsigned long timeout_ms)
2857 {
2858         struct rdma_id_private *id_priv;
2859         int ret;
2860
2861         id_priv = container_of(id, struct rdma_id_private, id);
2862         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2863                 return -EINVAL;
2864
2865         atomic_inc(&id_priv->refcount);
2866         if (rdma_cap_ib_sa(id->device, id->port_num))
2867                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2868         else if (rdma_protocol_roce(id->device, id->port_num))
2869                 ret = cma_resolve_iboe_route(id_priv);
2870         else if (rdma_protocol_iwarp(id->device, id->port_num))
2871                 ret = cma_resolve_iw_route(id_priv);
2872         else
2873                 ret = -ENOSYS;
2874
2875         if (ret)
2876                 goto err;
2877
2878         return 0;
2879 err:
2880         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2881         cma_deref_id(id_priv);
2882         return ret;
2883 }
2884 EXPORT_SYMBOL(rdma_resolve_route);
2885
2886 static void cma_set_loopback(struct sockaddr *addr)
2887 {
2888         switch (addr->sa_family) {
2889         case AF_INET:
2890                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2891                 break;
2892         case AF_INET6:
2893                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2894                               0, 0, 0, htonl(1));
2895                 break;
2896         default:
2897                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2898                             0, 0, 0, htonl(1));
2899                 break;
2900         }
2901 }
2902
2903 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2904 {
2905         struct cma_device *cma_dev, *cur_dev;
2906         union ib_gid gid;
2907         enum ib_port_state port_state;
2908         u16 pkey;
2909         int ret;
2910         u8 p;
2911
2912         cma_dev = NULL;
2913         mutex_lock(&lock);
2914         list_for_each_entry(cur_dev, &dev_list, list) {
2915                 if (cma_family(id_priv) == AF_IB &&
2916                     !rdma_cap_ib_cm(cur_dev->device, 1))
2917                         continue;
2918
2919                 if (!cma_dev)
2920                         cma_dev = cur_dev;
2921
2922                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2923                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2924                             port_state == IB_PORT_ACTIVE) {
2925                                 cma_dev = cur_dev;
2926                                 goto port_found;
2927                         }
2928                 }
2929         }
2930
2931         if (!cma_dev) {
2932                 ret = -ENODEV;
2933                 goto out;
2934         }
2935
2936         p = 1;
2937
2938 port_found:
2939         ret = rdma_query_gid(cma_dev->device, p, 0, &gid);
2940         if (ret)
2941                 goto out;
2942
2943         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2944         if (ret)
2945                 goto out;
2946
2947         id_priv->id.route.addr.dev_addr.dev_type =
2948                 (rdma_protocol_ib(cma_dev->device, p)) ?
2949                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2950
2951         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2952         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2953         id_priv->id.port_num = p;
2954         cma_attach_to_dev(id_priv, cma_dev);
2955         cma_set_loopback(cma_src_addr(id_priv));
2956 out:
2957         mutex_unlock(&lock);
2958         return ret;
2959 }
2960
2961 static void addr_handler(int status, struct sockaddr *src_addr,
2962                          struct rdma_dev_addr *dev_addr, void *context)
2963 {
2964         struct rdma_id_private *id_priv = context;
2965         struct rdma_cm_event event = {};
2966
2967         mutex_lock(&id_priv->handler_mutex);
2968         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2969                            RDMA_CM_ADDR_RESOLVED))
2970                 goto out;
2971
2972         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2973         if (!status && !id_priv->cma_dev) {
2974                 status = cma_acquire_dev_by_src_ip(id_priv);
2975                 if (status)
2976                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2977                                              status);
2978         } else {
2979                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2980         }
2981
2982         if (status) {
2983                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2984                                    RDMA_CM_ADDR_BOUND))
2985                         goto out;
2986                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2987                 event.status = status;
2988         } else
2989                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2990
2991         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2992                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2993                 mutex_unlock(&id_priv->handler_mutex);
2994                 rdma_destroy_id(&id_priv->id);
2995                 return;
2996         }
2997 out:
2998         mutex_unlock(&id_priv->handler_mutex);
2999 }
3000
3001 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
3002 {
3003         struct cma_work *work;
3004         union ib_gid gid;
3005         int ret;
3006
3007         work = kzalloc(sizeof *work, GFP_KERNEL);
3008         if (!work)
3009                 return -ENOMEM;
3010
3011         if (!id_priv->cma_dev) {
3012                 ret = cma_bind_loopback(id_priv);
3013                 if (ret)
3014                         goto err;
3015         }
3016
3017         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
3018         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
3019
3020         cma_init_resolve_addr_work(work, id_priv);
3021         queue_work(cma_wq, &work->work);
3022         return 0;
3023 err:
3024         kfree(work);
3025         return ret;
3026 }
3027
3028 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
3029 {
3030         struct cma_work *work;
3031         int ret;
3032
3033         work = kzalloc(sizeof *work, GFP_KERNEL);
3034         if (!work)
3035                 return -ENOMEM;
3036
3037         if (!id_priv->cma_dev) {
3038                 ret = cma_resolve_ib_dev(id_priv);
3039                 if (ret)
3040                         goto err;
3041         }
3042
3043         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
3044                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
3045
3046         cma_init_resolve_addr_work(work, id_priv);
3047         queue_work(cma_wq, &work->work);
3048         return 0;
3049 err:
3050         kfree(work);
3051         return ret;
3052 }
3053
3054 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
3055                          const struct sockaddr *dst_addr)
3056 {
3057         if (!src_addr || !src_addr->sa_family) {
3058                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
3059                 src_addr->sa_family = dst_addr->sa_family;
3060                 if (IS_ENABLED(CONFIG_IPV6) &&
3061                     dst_addr->sa_family == AF_INET6) {
3062                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
3063                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
3064                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
3065                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
3066                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
3067                 } else if (dst_addr->sa_family == AF_IB) {
3068                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
3069                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
3070                 }
3071         }
3072         return rdma_bind_addr(id, src_addr);
3073 }
3074
3075 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
3076                       const struct sockaddr *dst_addr, unsigned long timeout_ms)
3077 {
3078         struct rdma_id_private *id_priv;
3079         int ret;
3080
3081         id_priv = container_of(id, struct rdma_id_private, id);
3082         if (id_priv->state == RDMA_CM_IDLE) {
3083                 ret = cma_bind_addr(id, src_addr, dst_addr);
3084                 if (ret)
3085                         return ret;
3086         }
3087
3088         if (cma_family(id_priv) != dst_addr->sa_family)
3089                 return -EINVAL;
3090
3091         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
3092                 return -EINVAL;
3093
3094         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
3095         if (cma_any_addr(dst_addr)) {
3096                 ret = cma_resolve_loopback(id_priv);
3097         } else {
3098                 if (dst_addr->sa_family == AF_IB) {
3099                         ret = cma_resolve_ib_addr(id_priv);
3100                 } else {
3101                         ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
3102                                               &id->route.addr.dev_addr,
3103                                               timeout_ms, addr_handler,
3104                                               false, id_priv);
3105                 }
3106         }
3107         if (ret)
3108                 goto err;
3109
3110         return 0;
3111 err:
3112         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
3113         return ret;
3114 }
3115 EXPORT_SYMBOL(rdma_resolve_addr);
3116
3117 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
3118 {
3119         struct rdma_id_private *id_priv;
3120         unsigned long flags;
3121         int ret;
3122
3123         id_priv = container_of(id, struct rdma_id_private, id);
3124         spin_lock_irqsave(&id_priv->lock, flags);
3125         if (reuse || id_priv->state == RDMA_CM_IDLE) {
3126                 id_priv->reuseaddr = reuse;
3127                 ret = 0;
3128         } else {
3129                 ret = -EINVAL;
3130         }
3131         spin_unlock_irqrestore(&id_priv->lock, flags);
3132         return ret;
3133 }
3134 EXPORT_SYMBOL(rdma_set_reuseaddr);
3135
3136 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3137 {
3138         struct rdma_id_private *id_priv;
3139         unsigned long flags;
3140         int ret;
3141
3142         id_priv = container_of(id, struct rdma_id_private, id);
3143         spin_lock_irqsave(&id_priv->lock, flags);
3144         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3145                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3146                 id_priv->afonly = afonly;
3147                 ret = 0;
3148         } else {
3149                 ret = -EINVAL;
3150         }
3151         spin_unlock_irqrestore(&id_priv->lock, flags);
3152         return ret;
3153 }
3154 EXPORT_SYMBOL(rdma_set_afonly);
3155
3156 static void cma_bind_port(struct rdma_bind_list *bind_list,
3157                           struct rdma_id_private *id_priv)
3158 {
3159         struct sockaddr *addr;
3160         struct sockaddr_ib *sib;
3161         u64 sid, mask;
3162         __be16 port;
3163
3164         addr = cma_src_addr(id_priv);
3165         port = htons(bind_list->port);
3166
3167         switch (addr->sa_family) {
3168         case AF_INET:
3169                 ((struct sockaddr_in *) addr)->sin_port = port;
3170                 break;
3171         case AF_INET6:
3172                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3173                 break;
3174         case AF_IB:
3175                 sib = (struct sockaddr_ib *) addr;
3176                 sid = be64_to_cpu(sib->sib_sid);
3177                 mask = be64_to_cpu(sib->sib_sid_mask);
3178                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3179                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3180                 break;
3181         }
3182         id_priv->bind_list = bind_list;
3183         hlist_add_head(&id_priv->node, &bind_list->owners);
3184 }
3185
3186 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3187                           struct rdma_id_private *id_priv, unsigned short snum)
3188 {
3189         struct rdma_bind_list *bind_list;
3190         int ret;
3191
3192         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3193         if (!bind_list)
3194                 return -ENOMEM;
3195
3196         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3197                            snum);
3198         if (ret < 0)
3199                 goto err;
3200
3201         bind_list->ps = ps;
3202         bind_list->port = (unsigned short)ret;
3203         cma_bind_port(bind_list, id_priv);
3204         return 0;
3205 err:
3206         kfree(bind_list);
3207         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3208 }
3209
3210 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3211                               struct rdma_id_private *id_priv)
3212 {
3213         struct rdma_id_private *cur_id;
3214         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3215         struct sockaddr  *saddr = cma_src_addr(id_priv);
3216         __be16 dport = cma_port(daddr);
3217
3218         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3219                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3220                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3221                 __be16 cur_dport = cma_port(cur_daddr);
3222
3223                 if (id_priv == cur_id)
3224                         continue;
3225
3226                 /* different dest port -> unique */
3227                 if (!cma_any_port(daddr) &&
3228                     !cma_any_port(cur_daddr) &&
3229                     (dport != cur_dport))
3230                         continue;
3231
3232                 /* different src address -> unique */
3233                 if (!cma_any_addr(saddr) &&
3234                     !cma_any_addr(cur_saddr) &&
3235                     cma_addr_cmp(saddr, cur_saddr))
3236                         continue;
3237
3238                 /* different dst address -> unique */
3239                 if (!cma_any_addr(daddr) &&
3240                     !cma_any_addr(cur_daddr) &&
3241                     cma_addr_cmp(daddr, cur_daddr))
3242                         continue;
3243
3244                 return -EADDRNOTAVAIL;
3245         }
3246         return 0;
3247 }
3248
3249 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3250                               struct rdma_id_private *id_priv)
3251 {
3252         static unsigned int last_used_port;
3253         int low, high, remaining;
3254         unsigned int rover;
3255         struct net *net = id_priv->id.route.addr.dev_addr.net;
3256
3257         inet_get_local_port_range(net, &low, &high);
3258         remaining = (high - low) + 1;
3259         rover = prandom_u32() % remaining + low;
3260 retry:
3261         if (last_used_port != rover) {
3262                 struct rdma_bind_list *bind_list;
3263                 int ret;
3264
3265                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3266
3267                 if (!bind_list) {
3268                         ret = cma_alloc_port(ps, id_priv, rover);
3269                 } else {
3270                         ret = cma_port_is_unique(bind_list, id_priv);
3271                         if (!ret)
3272                                 cma_bind_port(bind_list, id_priv);
3273                 }
3274                 /*
3275                  * Remember previously used port number in order to avoid
3276                  * re-using same port immediately after it is closed.
3277                  */
3278                 if (!ret)
3279                         last_used_port = rover;
3280                 if (ret != -EADDRNOTAVAIL)
3281                         return ret;
3282         }
3283         if (--remaining) {
3284                 rover++;
3285                 if ((rover < low) || (rover > high))
3286                         rover = low;
3287                 goto retry;
3288         }
3289         return -EADDRNOTAVAIL;
3290 }
3291
3292 /*
3293  * Check that the requested port is available.  This is called when trying to
3294  * bind to a specific port, or when trying to listen on a bound port.  In
3295  * the latter case, the provided id_priv may already be on the bind_list, but
3296  * we still need to check that it's okay to start listening.
3297  */
3298 static int cma_check_port(struct rdma_bind_list *bind_list,
3299                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3300 {
3301         struct rdma_id_private *cur_id;
3302         struct sockaddr *addr, *cur_addr;
3303
3304         addr = cma_src_addr(id_priv);
3305         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3306                 if (id_priv == cur_id)
3307                         continue;
3308
3309                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3310                     cur_id->reuseaddr)
3311                         continue;
3312
3313                 cur_addr = cma_src_addr(cur_id);
3314                 if (id_priv->afonly && cur_id->afonly &&
3315                     (addr->sa_family != cur_addr->sa_family))
3316                         continue;
3317
3318                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3319                         return -EADDRNOTAVAIL;
3320
3321                 if (!cma_addr_cmp(addr, cur_addr))
3322                         return -EADDRINUSE;
3323         }
3324         return 0;
3325 }
3326
3327 static int cma_use_port(enum rdma_ucm_port_space ps,
3328                         struct rdma_id_private *id_priv)
3329 {
3330         struct rdma_bind_list *bind_list;
3331         unsigned short snum;
3332         int ret;
3333
3334         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3335         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3336                 return -EACCES;
3337
3338         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3339         if (!bind_list) {
3340                 ret = cma_alloc_port(ps, id_priv, snum);
3341         } else {
3342                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3343                 if (!ret)
3344                         cma_bind_port(bind_list, id_priv);
3345         }
3346         return ret;
3347 }
3348
3349 static int cma_bind_listen(struct rdma_id_private *id_priv)
3350 {
3351         struct rdma_bind_list *bind_list = id_priv->bind_list;
3352         int ret = 0;
3353
3354         mutex_lock(&lock);
3355         if (bind_list->owners.first->next)
3356                 ret = cma_check_port(bind_list, id_priv, 0);
3357         mutex_unlock(&lock);
3358         return ret;
3359 }
3360
3361 static enum rdma_ucm_port_space
3362 cma_select_inet_ps(struct rdma_id_private *id_priv)
3363 {
3364         switch (id_priv->id.ps) {
3365         case RDMA_PS_TCP:
3366         case RDMA_PS_UDP:
3367         case RDMA_PS_IPOIB:
3368         case RDMA_PS_IB:
3369                 return id_priv->id.ps;
3370         default:
3371
3372                 return 0;
3373         }
3374 }
3375
3376 static enum rdma_ucm_port_space
3377 cma_select_ib_ps(struct rdma_id_private *id_priv)
3378 {
3379         enum rdma_ucm_port_space ps = 0;
3380         struct sockaddr_ib *sib;
3381         u64 sid_ps, mask, sid;
3382
3383         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3384         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3385         sid = be64_to_cpu(sib->sib_sid) & mask;
3386
3387         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3388                 sid_ps = RDMA_IB_IP_PS_IB;
3389                 ps = RDMA_PS_IB;
3390         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3391                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3392                 sid_ps = RDMA_IB_IP_PS_TCP;
3393                 ps = RDMA_PS_TCP;
3394         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3395                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3396                 sid_ps = RDMA_IB_IP_PS_UDP;
3397                 ps = RDMA_PS_UDP;
3398         }
3399
3400         if (ps) {
3401                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3402                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3403                                                 be64_to_cpu(sib->sib_sid_mask));
3404         }
3405         return ps;
3406 }
3407
3408 static int cma_get_port(struct rdma_id_private *id_priv)
3409 {
3410         enum rdma_ucm_port_space ps;
3411         int ret;
3412
3413         if (cma_family(id_priv) != AF_IB)
3414                 ps = cma_select_inet_ps(id_priv);
3415         else
3416                 ps = cma_select_ib_ps(id_priv);
3417         if (!ps)
3418                 return -EPROTONOSUPPORT;
3419
3420         mutex_lock(&lock);
3421         if (cma_any_port(cma_src_addr(id_priv)))
3422                 ret = cma_alloc_any_port(ps, id_priv);
3423         else
3424                 ret = cma_use_port(ps, id_priv);
3425         mutex_unlock(&lock);
3426
3427         return ret;
3428 }
3429
3430 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3431                                struct sockaddr *addr)
3432 {
3433 #if IS_ENABLED(CONFIG_IPV6)
3434         struct sockaddr_in6 *sin6;
3435
3436         if (addr->sa_family != AF_INET6)
3437                 return 0;
3438
3439         sin6 = (struct sockaddr_in6 *) addr;
3440
3441         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3442                 return 0;
3443
3444         if (!sin6->sin6_scope_id)
3445                         return -EINVAL;
3446
3447         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3448 #endif
3449         return 0;
3450 }
3451
3452 int rdma_listen(struct rdma_cm_id *id, int backlog)
3453 {
3454         struct rdma_id_private *id_priv;
3455         int ret;
3456
3457         id_priv = container_of(id, struct rdma_id_private, id);
3458         if (id_priv->state == RDMA_CM_IDLE) {
3459                 id->route.addr.src_addr.ss_family = AF_INET;
3460                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3461                 if (ret)
3462                         return ret;
3463         }
3464
3465         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3466                 return -EINVAL;
3467
3468         if (id_priv->reuseaddr) {
3469                 ret = cma_bind_listen(id_priv);
3470                 if (ret)
3471                         goto err;
3472         }
3473
3474         id_priv->backlog = backlog;
3475         if (id->device) {
3476                 if (rdma_cap_ib_cm(id->device, 1)) {
3477                         ret = cma_ib_listen(id_priv);
3478                         if (ret)
3479                                 goto err;
3480                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3481                         ret = cma_iw_listen(id_priv, backlog);
3482                         if (ret)
3483                                 goto err;
3484                 } else {
3485                         ret = -ENOSYS;
3486                         goto err;
3487                 }
3488         } else
3489                 cma_listen_on_all(id_priv);
3490
3491         return 0;
3492 err:
3493         id_priv->backlog = 0;
3494         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3495         return ret;
3496 }
3497 EXPORT_SYMBOL(rdma_listen);
3498
3499 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3500 {
3501         struct rdma_id_private *id_priv;
3502         int ret;
3503         struct sockaddr  *daddr;
3504
3505         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3506             addr->sa_family != AF_IB)
3507                 return -EAFNOSUPPORT;
3508
3509         id_priv = container_of(id, struct rdma_id_private, id);
3510         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3511                 return -EINVAL;
3512
3513         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3514         if (ret)
3515                 goto err1;
3516
3517         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3518         if (!cma_any_addr(addr)) {
3519                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3520                 if (ret)
3521                         goto err1;
3522
3523                 ret = cma_acquire_dev_by_src_ip(id_priv);
3524                 if (ret)
3525                         goto err1;
3526         }
3527
3528         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3529                 if (addr->sa_family == AF_INET)
3530                         id_priv->afonly = 1;
3531 #if IS_ENABLED(CONFIG_IPV6)
3532                 else if (addr->sa_family == AF_INET6) {
3533                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3534
3535                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3536                 }
3537 #endif
3538         }
3539         daddr = cma_dst_addr(id_priv);
3540         daddr->sa_family = addr->sa_family;
3541
3542         ret = cma_get_port(id_priv);
3543         if (ret)
3544                 goto err2;
3545
3546         return 0;
3547 err2:
3548         rdma_restrack_del(&id_priv->res);
3549         if (id_priv->cma_dev)
3550                 cma_release_dev(id_priv);
3551 err1:
3552         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3553         return ret;
3554 }
3555 EXPORT_SYMBOL(rdma_bind_addr);
3556
3557 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3558 {
3559         struct cma_hdr *cma_hdr;
3560
3561         cma_hdr = hdr;
3562         cma_hdr->cma_version = CMA_VERSION;
3563         if (cma_family(id_priv) == AF_INET) {
3564                 struct sockaddr_in *src4, *dst4;
3565
3566                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3567                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3568
3569                 cma_set_ip_ver(cma_hdr, 4);
3570                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3571                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3572                 cma_hdr->port = src4->sin_port;
3573         } else if (cma_family(id_priv) == AF_INET6) {
3574                 struct sockaddr_in6 *src6, *dst6;
3575
3576                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3577                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3578
3579                 cma_set_ip_ver(cma_hdr, 6);
3580                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3581                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3582                 cma_hdr->port = src6->sin6_port;
3583         }
3584         return 0;
3585 }
3586
3587 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3588                                 const struct ib_cm_event *ib_event)
3589 {
3590         struct rdma_id_private *id_priv = cm_id->context;
3591         struct rdma_cm_event event = {};
3592         const struct ib_cm_sidr_rep_event_param *rep =
3593                                 &ib_event->param.sidr_rep_rcvd;
3594         int ret = 0;
3595
3596         mutex_lock(&id_priv->handler_mutex);
3597         if (id_priv->state != RDMA_CM_CONNECT)
3598                 goto out;
3599
3600         switch (ib_event->event) {
3601         case IB_CM_SIDR_REQ_ERROR:
3602                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3603                 event.status = -ETIMEDOUT;
3604                 break;
3605         case IB_CM_SIDR_REP_RECEIVED:
3606                 event.param.ud.private_data = ib_event->private_data;
3607                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3608                 if (rep->status != IB_SIDR_SUCCESS) {
3609                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3610                         event.status = ib_event->param.sidr_rep_rcvd.status;
3611                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3612                                              event.status);
3613                         break;
3614                 }
3615                 ret = cma_set_qkey(id_priv, rep->qkey);
3616                 if (ret) {
3617                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3618                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3619                         event.status = ret;
3620                         break;
3621                 }
3622                 ib_init_ah_attr_from_path(id_priv->id.device,
3623                                           id_priv->id.port_num,
3624                                           id_priv->id.route.path_rec,
3625                                           &event.param.ud.ah_attr,
3626                                           rep->sgid_attr);
3627                 event.param.ud.qp_num = rep->qpn;
3628                 event.param.ud.qkey = rep->qkey;
3629                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3630                 event.status = 0;
3631                 break;
3632         default:
3633                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3634                        ib_event->event);
3635                 goto out;
3636         }
3637
3638         ret = id_priv->id.event_handler(&id_priv->id, &event);
3639
3640         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
3641         if (ret) {
3642                 /* Destroy the CM ID by returning a non-zero value. */
3643                 id_priv->cm_id.ib = NULL;
3644                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3645                 mutex_unlock(&id_priv->handler_mutex);
3646                 rdma_destroy_id(&id_priv->id);
3647                 return ret;
3648         }
3649 out:
3650         mutex_unlock(&id_priv->handler_mutex);
3651         return ret;
3652 }
3653
3654 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3655                               struct rdma_conn_param *conn_param)
3656 {
3657         struct ib_cm_sidr_req_param req;
3658         struct ib_cm_id *id;
3659         void *private_data;
3660         u8 offset;
3661         int ret;
3662
3663         memset(&req, 0, sizeof req);
3664         offset = cma_user_data_offset(id_priv);
3665         req.private_data_len = offset + conn_param->private_data_len;
3666         if (req.private_data_len < conn_param->private_data_len)
3667                 return -EINVAL;
3668
3669         if (req.private_data_len) {
3670                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3671                 if (!private_data)
3672                         return -ENOMEM;
3673         } else {
3674                 private_data = NULL;
3675         }
3676
3677         if (conn_param->private_data && conn_param->private_data_len)
3678                 memcpy(private_data + offset, conn_param->private_data,
3679                        conn_param->private_data_len);
3680
3681         if (private_data) {
3682                 ret = cma_format_hdr(private_data, id_priv);
3683                 if (ret)
3684                         goto out;
3685                 req.private_data = private_data;
3686         }
3687
3688         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3689                              id_priv);
3690         if (IS_ERR(id)) {
3691                 ret = PTR_ERR(id);
3692                 goto out;
3693         }
3694         id_priv->cm_id.ib = id;
3695
3696         req.path = id_priv->id.route.path_rec;
3697         req.sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3698         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3699         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3700         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3701
3702         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3703         if (ret) {
3704                 ib_destroy_cm_id(id_priv->cm_id.ib);
3705                 id_priv->cm_id.ib = NULL;
3706         }
3707 out:
3708         kfree(private_data);
3709         return ret;
3710 }
3711
3712 static int cma_connect_ib(struct rdma_id_private *id_priv,
3713                           struct rdma_conn_param *conn_param)
3714 {
3715         struct ib_cm_req_param req;
3716         struct rdma_route *route;
3717         void *private_data;
3718         struct ib_cm_id *id;
3719         u8 offset;
3720         int ret;
3721
3722         memset(&req, 0, sizeof req);
3723         offset = cma_user_data_offset(id_priv);
3724         req.private_data_len = offset + conn_param->private_data_len;
3725         if (req.private_data_len < conn_param->private_data_len)
3726                 return -EINVAL;
3727
3728         if (req.private_data_len) {
3729                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3730                 if (!private_data)
3731                         return -ENOMEM;
3732         } else {
3733                 private_data = NULL;
3734         }
3735
3736         if (conn_param->private_data && conn_param->private_data_len)
3737                 memcpy(private_data + offset, conn_param->private_data,
3738                        conn_param->private_data_len);
3739
3740         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3741         if (IS_ERR(id)) {
3742                 ret = PTR_ERR(id);
3743                 goto out;
3744         }
3745         id_priv->cm_id.ib = id;
3746
3747         route = &id_priv->id.route;
3748         if (private_data) {
3749                 ret = cma_format_hdr(private_data, id_priv);
3750                 if (ret)
3751                         goto out;
3752                 req.private_data = private_data;
3753         }
3754
3755         req.primary_path = &route->path_rec[0];
3756         if (route->num_paths == 2)
3757                 req.alternate_path = &route->path_rec[1];
3758
3759         req.ppath_sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3760         /* Alternate path SGID attribute currently unsupported */
3761         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3762         req.qp_num = id_priv->qp_num;
3763         req.qp_type = id_priv->id.qp_type;
3764         req.starting_psn = id_priv->seq_num;
3765         req.responder_resources = conn_param->responder_resources;
3766         req.initiator_depth = conn_param->initiator_depth;
3767         req.flow_control = conn_param->flow_control;
3768         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3769         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3770         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3771         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3772         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3773         req.srq = id_priv->srq ? 1 : 0;
3774
3775         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3776 out:
3777         if (ret && !IS_ERR(id)) {
3778                 ib_destroy_cm_id(id);
3779                 id_priv->cm_id.ib = NULL;
3780         }
3781
3782         kfree(private_data);
3783         return ret;
3784 }
3785
3786 static int cma_connect_iw(struct rdma_id_private *id_priv,
3787                           struct rdma_conn_param *conn_param)
3788 {
3789         struct iw_cm_id *cm_id;
3790         int ret;
3791         struct iw_cm_conn_param iw_param;
3792
3793         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3794         if (IS_ERR(cm_id))
3795                 return PTR_ERR(cm_id);
3796
3797         cm_id->tos = id_priv->tos;
3798         id_priv->cm_id.iw = cm_id;
3799
3800         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3801                rdma_addr_size(cma_src_addr(id_priv)));
3802         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3803                rdma_addr_size(cma_dst_addr(id_priv)));
3804
3805         ret = cma_modify_qp_rtr(id_priv, conn_param);
3806         if (ret)
3807                 goto out;
3808
3809         if (conn_param) {
3810                 iw_param.ord = conn_param->initiator_depth;
3811                 iw_param.ird = conn_param->responder_resources;
3812                 iw_param.private_data = conn_param->private_data;
3813                 iw_param.private_data_len = conn_param->private_data_len;
3814                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3815         } else {
3816                 memset(&iw_param, 0, sizeof iw_param);
3817                 iw_param.qpn = id_priv->qp_num;
3818         }
3819         ret = iw_cm_connect(cm_id, &iw_param);
3820 out:
3821         if (ret) {
3822                 iw_destroy_cm_id(cm_id);
3823                 id_priv->cm_id.iw = NULL;
3824         }
3825         return ret;
3826 }
3827
3828 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3829 {
3830         struct rdma_id_private *id_priv;
3831         int ret;
3832
3833         id_priv = container_of(id, struct rdma_id_private, id);
3834         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3835                 return -EINVAL;
3836
3837         if (!id->qp) {
3838                 id_priv->qp_num = conn_param->qp_num;
3839                 id_priv->srq = conn_param->srq;
3840         }
3841
3842         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3843                 if (id->qp_type == IB_QPT_UD)
3844                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3845                 else
3846                         ret = cma_connect_ib(id_priv, conn_param);
3847         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3848                 ret = cma_connect_iw(id_priv, conn_param);
3849         else
3850                 ret = -ENOSYS;
3851         if (ret)
3852                 goto err;
3853
3854         return 0;
3855 err:
3856         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3857         return ret;
3858 }
3859 EXPORT_SYMBOL(rdma_connect);
3860
3861 static int cma_accept_ib(struct rdma_id_private *id_priv,
3862                          struct rdma_conn_param *conn_param)
3863 {
3864         struct ib_cm_rep_param rep;
3865         int ret;
3866
3867         ret = cma_modify_qp_rtr(id_priv, conn_param);
3868         if (ret)
3869                 goto out;
3870
3871         ret = cma_modify_qp_rts(id_priv, conn_param);
3872         if (ret)
3873                 goto out;
3874
3875         memset(&rep, 0, sizeof rep);
3876         rep.qp_num = id_priv->qp_num;
3877         rep.starting_psn = id_priv->seq_num;
3878         rep.private_data = conn_param->private_data;
3879         rep.private_data_len = conn_param->private_data_len;
3880         rep.responder_resources = conn_param->responder_resources;
3881         rep.initiator_depth = conn_param->initiator_depth;
3882         rep.failover_accepted = 0;
3883         rep.flow_control = conn_param->flow_control;
3884         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3885         rep.srq = id_priv->srq ? 1 : 0;
3886
3887         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3888 out:
3889         return ret;
3890 }
3891
3892 static int cma_accept_iw(struct rdma_id_private *id_priv,
3893                   struct rdma_conn_param *conn_param)
3894 {
3895         struct iw_cm_conn_param iw_param;
3896         int ret;
3897
3898         if (!conn_param)
3899                 return -EINVAL;
3900
3901         ret = cma_modify_qp_rtr(id_priv, conn_param);
3902         if (ret)
3903                 return ret;
3904
3905         iw_param.ord = conn_param->initiator_depth;
3906         iw_param.ird = conn_param->responder_resources;
3907         iw_param.private_data = conn_param->private_data;
3908         iw_param.private_data_len = conn_param->private_data_len;
3909         if (id_priv->id.qp) {
3910                 iw_param.qpn = id_priv->qp_num;
3911         } else
3912                 iw_param.qpn = conn_param->qp_num;
3913
3914         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3915 }
3916
3917 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3918                              enum ib_cm_sidr_status status, u32 qkey,
3919                              const void *private_data, int private_data_len)
3920 {
3921         struct ib_cm_sidr_rep_param rep;
3922         int ret;
3923
3924         memset(&rep, 0, sizeof rep);
3925         rep.status = status;
3926         if (status == IB_SIDR_SUCCESS) {
3927                 ret = cma_set_qkey(id_priv, qkey);
3928                 if (ret)
3929                         return ret;
3930                 rep.qp_num = id_priv->qp_num;
3931                 rep.qkey = id_priv->qkey;
3932         }
3933         rep.private_data = private_data;
3934         rep.private_data_len = private_data_len;
3935
3936         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3937 }
3938
3939 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3940                   const char *caller)
3941 {
3942         struct rdma_id_private *id_priv;
3943         int ret;
3944
3945         id_priv = container_of(id, struct rdma_id_private, id);
3946
3947         rdma_restrack_set_task(&id_priv->res, caller);
3948
3949         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3950                 return -EINVAL;
3951
3952         if (!id->qp && conn_param) {
3953                 id_priv->qp_num = conn_param->qp_num;
3954                 id_priv->srq = conn_param->srq;
3955         }
3956
3957         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3958                 if (id->qp_type == IB_QPT_UD) {
3959                         if (conn_param)
3960                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3961                                                         conn_param->qkey,
3962                                                         conn_param->private_data,
3963                                                         conn_param->private_data_len);
3964                         else
3965                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3966                                                         0, NULL, 0);
3967                 } else {
3968                         if (conn_param)
3969                                 ret = cma_accept_ib(id_priv, conn_param);
3970                         else
3971                                 ret = cma_rep_recv(id_priv);
3972                 }
3973         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3974                 ret = cma_accept_iw(id_priv, conn_param);
3975         else
3976                 ret = -ENOSYS;
3977
3978         if (ret)
3979                 goto reject;
3980
3981         return 0;
3982 reject:
3983         cma_modify_qp_err(id_priv);
3984         rdma_reject(id, NULL, 0);
3985         return ret;
3986 }
3987 EXPORT_SYMBOL(__rdma_accept);
3988
3989 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3990 {
3991         struct rdma_id_private *id_priv;
3992         int ret;
3993
3994         id_priv = container_of(id, struct rdma_id_private, id);
3995         if (!id_priv->cm_id.ib)
3996                 return -EINVAL;
3997
3998         switch (id->device->node_type) {
3999         case RDMA_NODE_IB_CA:
4000                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
4001                 break;
4002         default:
4003                 ret = 0;
4004                 break;
4005         }
4006         return ret;
4007 }
4008 EXPORT_SYMBOL(rdma_notify);
4009
4010 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
4011                 u8 private_data_len)
4012 {
4013         struct rdma_id_private *id_priv;
4014         int ret;
4015
4016         id_priv = container_of(id, struct rdma_id_private, id);
4017         if (!id_priv->cm_id.ib)
4018                 return -EINVAL;
4019
4020         if (rdma_cap_ib_cm(id->device, id->port_num)) {
4021                 if (id->qp_type == IB_QPT_UD)
4022                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
4023                                                 private_data, private_data_len);
4024                 else
4025                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
4026                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
4027                                              0, private_data, private_data_len);
4028         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
4029                 ret = iw_cm_reject(id_priv->cm_id.iw,
4030                                    private_data, private_data_len);
4031         } else
4032                 ret = -ENOSYS;
4033
4034         return ret;
4035 }
4036 EXPORT_SYMBOL(rdma_reject);
4037
4038 int rdma_disconnect(struct rdma_cm_id *id)
4039 {
4040         struct rdma_id_private *id_priv;
4041         int ret;
4042
4043         id_priv = container_of(id, struct rdma_id_private, id);
4044         if (!id_priv->cm_id.ib)
4045                 return -EINVAL;
4046
4047         if (rdma_cap_ib_cm(id->device, id->port_num)) {
4048                 ret = cma_modify_qp_err(id_priv);
4049                 if (ret)
4050                         goto out;
4051                 /* Initiate or respond to a disconnect. */
4052                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
4053                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
4054         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
4055                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
4056         } else
4057                 ret = -EINVAL;
4058
4059 out:
4060         return ret;
4061 }
4062 EXPORT_SYMBOL(rdma_disconnect);
4063
4064 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
4065 {
4066         struct rdma_id_private *id_priv;
4067         struct cma_multicast *mc = multicast->context;
4068         struct rdma_cm_event event = {};
4069         int ret = 0;
4070
4071         id_priv = mc->id_priv;
4072         mutex_lock(&id_priv->handler_mutex);
4073         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
4074             id_priv->state != RDMA_CM_ADDR_RESOLVED)
4075                 goto out;
4076
4077         if (!status)
4078                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
4079         else
4080                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
4081                                      status);
4082         mutex_lock(&id_priv->qp_mutex);
4083         if (!status && id_priv->id.qp) {
4084                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
4085                                          be16_to_cpu(multicast->rec.mlid));
4086                 if (status)
4087                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
4088                                              status);
4089         }
4090         mutex_unlock(&id_priv->qp_mutex);
4091
4092         event.status = status;
4093         event.param.ud.private_data = mc->context;
4094         if (!status) {
4095                 struct rdma_dev_addr *dev_addr =
4096                         &id_priv->id.route.addr.dev_addr;
4097                 struct net_device *ndev =
4098                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4099                 enum ib_gid_type gid_type =
4100                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4101                         rdma_start_port(id_priv->cma_dev->device)];
4102
4103                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
4104                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
4105                                                id_priv->id.port_num,
4106                                                &multicast->rec,
4107                                                ndev, gid_type,
4108                                                &event.param.ud.ah_attr);
4109                 if (ret)
4110                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4111
4112                 event.param.ud.qp_num = 0xFFFFFF;
4113                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
4114                 if (ndev)
4115                         dev_put(ndev);
4116         } else
4117                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4118
4119         ret = id_priv->id.event_handler(&id_priv->id, &event);
4120
4121         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
4122         if (ret) {
4123                 cma_exch(id_priv, RDMA_CM_DESTROYING);
4124                 mutex_unlock(&id_priv->handler_mutex);
4125                 rdma_destroy_id(&id_priv->id);
4126                 return 0;
4127         }
4128
4129 out:
4130         mutex_unlock(&id_priv->handler_mutex);
4131         return 0;
4132 }
4133
4134 static void cma_set_mgid(struct rdma_id_private *id_priv,
4135                          struct sockaddr *addr, union ib_gid *mgid)
4136 {
4137         unsigned char mc_map[MAX_ADDR_LEN];
4138         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4139         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4140         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4141
4142         if (cma_any_addr(addr)) {
4143                 memset(mgid, 0, sizeof *mgid);
4144         } else if ((addr->sa_family == AF_INET6) &&
4145                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4146                                                                  0xFF10A01B)) {
4147                 /* IPv6 address is an SA assigned MGID. */
4148                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4149         } else if (addr->sa_family == AF_IB) {
4150                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4151         } else if (addr->sa_family == AF_INET6) {
4152                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4153                 if (id_priv->id.ps == RDMA_PS_UDP)
4154                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4155                 *mgid = *(union ib_gid *) (mc_map + 4);
4156         } else {
4157                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4158                 if (id_priv->id.ps == RDMA_PS_UDP)
4159                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4160                 *mgid = *(union ib_gid *) (mc_map + 4);
4161         }
4162 }
4163
4164 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4165                                  struct cma_multicast *mc)
4166 {
4167         struct ib_sa_mcmember_rec rec;
4168         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4169         ib_sa_comp_mask comp_mask;
4170         int ret;
4171
4172         ib_addr_get_mgid(dev_addr, &rec.mgid);
4173         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4174                                      &rec.mgid, &rec);
4175         if (ret)
4176                 return ret;
4177
4178         ret = cma_set_qkey(id_priv, 0);
4179         if (ret)
4180                 return ret;
4181
4182         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4183         rec.qkey = cpu_to_be32(id_priv->qkey);
4184         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4185         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4186         rec.join_state = mc->join_state;
4187
4188         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4189             (!ib_sa_sendonly_fullmem_support(&sa_client,
4190                                              id_priv->id.device,
4191                                              id_priv->id.port_num))) {
4192                 dev_warn(
4193                         &id_priv->id.device->dev,
4194                         "RDMA CM: port %u Unable to multicast join: SM doesn't support Send Only Full Member option\n",
4195                         id_priv->id.port_num);
4196                 return -EOPNOTSUPP;
4197         }
4198
4199         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4200                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4201                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4202                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4203                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4204
4205         if (id_priv->id.ps == RDMA_PS_IPOIB)
4206                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4207                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4208                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4209                              IB_SA_MCMEMBER_REC_MTU |
4210                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4211
4212         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4213                                                 id_priv->id.port_num, &rec,
4214                                                 comp_mask, GFP_KERNEL,
4215                                                 cma_ib_mc_handler, mc);
4216         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4217 }
4218
4219 static void iboe_mcast_work_handler(struct work_struct *work)
4220 {
4221         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4222         struct cma_multicast *mc = mw->mc;
4223         struct ib_sa_multicast *m = mc->multicast.ib;
4224
4225         mc->multicast.ib->context = mc;
4226         cma_ib_mc_handler(0, m);
4227         kref_put(&mc->mcref, release_mc);
4228         kfree(mw);
4229 }
4230
4231 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4232                               enum ib_gid_type gid_type)
4233 {
4234         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4235         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4236
4237         if (cma_any_addr(addr)) {
4238                 memset(mgid, 0, sizeof *mgid);
4239         } else if (addr->sa_family == AF_INET6) {
4240                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4241         } else {
4242                 mgid->raw[0] =
4243                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4244                 mgid->raw[1] =
4245                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4246                 mgid->raw[2] = 0;
4247                 mgid->raw[3] = 0;
4248                 mgid->raw[4] = 0;
4249                 mgid->raw[5] = 0;
4250                 mgid->raw[6] = 0;
4251                 mgid->raw[7] = 0;
4252                 mgid->raw[8] = 0;
4253                 mgid->raw[9] = 0;
4254                 mgid->raw[10] = 0xff;
4255                 mgid->raw[11] = 0xff;
4256                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4257         }
4258 }
4259
4260 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4261                                    struct cma_multicast *mc)
4262 {
4263         struct iboe_mcast_work *work;
4264         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4265         int err = 0;
4266         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4267         struct net_device *ndev = NULL;
4268         enum ib_gid_type gid_type;
4269         bool send_only;
4270
4271         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4272
4273         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4274                 return -EINVAL;
4275
4276         work = kzalloc(sizeof *work, GFP_KERNEL);
4277         if (!work)
4278                 return -ENOMEM;
4279
4280         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4281         if (!mc->multicast.ib) {
4282                 err = -ENOMEM;
4283                 goto out1;
4284         }
4285
4286         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4287                    rdma_start_port(id_priv->cma_dev->device)];
4288         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4289
4290         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4291         if (id_priv->id.ps == RDMA_PS_UDP)
4292                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4293
4294         if (dev_addr->bound_dev_if)
4295                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4296         if (!ndev) {
4297                 err = -ENODEV;
4298                 goto out2;
4299         }
4300         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4301         mc->multicast.ib->rec.hop_limit = 1;
4302         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4303
4304         if (addr->sa_family == AF_INET) {
4305                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4306                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4307                         if (!send_only) {
4308                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4309                                                     true);
4310                         }
4311                 }
4312         } else {
4313                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4314                         err = -ENOTSUPP;
4315         }
4316         dev_put(ndev);
4317         if (err || !mc->multicast.ib->rec.mtu) {
4318                 if (!err)
4319                         err = -EINVAL;
4320                 goto out2;
4321         }
4322         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4323                     &mc->multicast.ib->rec.port_gid);
4324         work->id = id_priv;
4325         work->mc = mc;
4326         INIT_WORK(&work->work, iboe_mcast_work_handler);
4327         kref_get(&mc->mcref);
4328         queue_work(cma_wq, &work->work);
4329
4330         return 0;
4331
4332 out2:
4333         kfree(mc->multicast.ib);
4334 out1:
4335         kfree(work);
4336         return err;
4337 }
4338
4339 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4340                         u8 join_state, void *context)
4341 {
4342         struct rdma_id_private *id_priv;
4343         struct cma_multicast *mc;
4344         int ret;
4345
4346         if (!id->device)
4347                 return -EINVAL;
4348
4349         id_priv = container_of(id, struct rdma_id_private, id);
4350         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4351             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4352                 return -EINVAL;
4353
4354         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4355         if (!mc)
4356                 return -ENOMEM;
4357
4358         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4359         mc->context = context;
4360         mc->id_priv = id_priv;
4361         mc->join_state = join_state;
4362
4363         if (rdma_protocol_roce(id->device, id->port_num)) {
4364                 kref_init(&mc->mcref);
4365                 ret = cma_iboe_join_multicast(id_priv, mc);
4366                 if (ret)
4367                         goto out_err;
4368         } else if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4369                 ret = cma_join_ib_multicast(id_priv, mc);
4370                 if (ret)
4371                         goto out_err;
4372         } else {
4373                 ret = -ENOSYS;
4374                 goto out_err;
4375         }
4376
4377         spin_lock(&id_priv->lock);
4378         list_add(&mc->list, &id_priv->mc_list);
4379         spin_unlock(&id_priv->lock);
4380
4381         return 0;
4382 out_err:
4383         kfree(mc);
4384         return ret;
4385 }
4386 EXPORT_SYMBOL(rdma_join_multicast);
4387
4388 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4389 {
4390         struct rdma_id_private *id_priv;
4391         struct cma_multicast *mc;
4392
4393         id_priv = container_of(id, struct rdma_id_private, id);
4394         spin_lock_irq(&id_priv->lock);
4395         list_for_each_entry(mc, &id_priv->mc_list, list) {
4396                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4397                         list_del(&mc->list);
4398                         spin_unlock_irq(&id_priv->lock);
4399
4400                         if (id->qp)
4401                                 ib_detach_mcast(id->qp,
4402                                                 &mc->multicast.ib->rec.mgid,
4403                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4404
4405                         BUG_ON(id_priv->cma_dev->device != id->device);
4406
4407                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4408                                 ib_sa_free_multicast(mc->multicast.ib);
4409                                 kfree(mc);
4410                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4411                                 cma_leave_roce_mc_group(id_priv, mc);
4412                         }
4413                         return;
4414                 }
4415         }
4416         spin_unlock_irq(&id_priv->lock);
4417 }
4418 EXPORT_SYMBOL(rdma_leave_multicast);
4419
4420 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4421 {
4422         struct rdma_dev_addr *dev_addr;
4423         struct cma_ndev_work *work;
4424
4425         dev_addr = &id_priv->id.route.addr.dev_addr;
4426
4427         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4428             (net_eq(dev_net(ndev), dev_addr->net)) &&
4429             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4430                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4431                         ndev->name, &id_priv->id);
4432                 work = kzalloc(sizeof *work, GFP_KERNEL);
4433                 if (!work)
4434                         return -ENOMEM;
4435
4436                 INIT_WORK(&work->work, cma_ndev_work_handler);
4437                 work->id = id_priv;
4438                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4439                 atomic_inc(&id_priv->refcount);
4440                 queue_work(cma_wq, &work->work);
4441         }
4442
4443         return 0;
4444 }
4445
4446 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4447                                void *ptr)
4448 {
4449         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4450         struct cma_device *cma_dev;
4451         struct rdma_id_private *id_priv;
4452         int ret = NOTIFY_DONE;
4453
4454         if (event != NETDEV_BONDING_FAILOVER)
4455                 return NOTIFY_DONE;
4456
4457         if (!netif_is_bond_master(ndev))
4458                 return NOTIFY_DONE;
4459
4460         mutex_lock(&lock);
4461         list_for_each_entry(cma_dev, &dev_list, list)
4462                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4463                         ret = cma_netdev_change(ndev, id_priv);
4464                         if (ret)
4465                                 goto out;
4466                 }
4467
4468 out:
4469         mutex_unlock(&lock);
4470         return ret;
4471 }
4472
4473 static struct notifier_block cma_nb = {
4474         .notifier_call = cma_netdev_callback
4475 };
4476
4477 static void cma_add_one(struct ib_device *device)
4478 {
4479         struct cma_device *cma_dev;
4480         struct rdma_id_private *id_priv;
4481         unsigned int i;
4482         unsigned long supported_gids = 0;
4483
4484         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4485         if (!cma_dev)
4486                 return;
4487
4488         cma_dev->device = device;
4489         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4490                                             sizeof(*cma_dev->default_gid_type),
4491                                             GFP_KERNEL);
4492         if (!cma_dev->default_gid_type)
4493                 goto free_cma_dev;
4494
4495         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4496                                             sizeof(*cma_dev->default_roce_tos),
4497                                             GFP_KERNEL);
4498         if (!cma_dev->default_roce_tos)
4499                 goto free_gid_type;
4500
4501         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4502                 supported_gids = roce_gid_type_mask_support(device, i);
4503                 WARN_ON(!supported_gids);
4504                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4505                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4506                                 CMA_PREFERRED_ROCE_GID_TYPE;
4507                 else
4508                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4509                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4510                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4511         }
4512
4513         init_completion(&cma_dev->comp);
4514         atomic_set(&cma_dev->refcount, 1);
4515         INIT_LIST_HEAD(&cma_dev->id_list);
4516         ib_set_client_data(device, &cma_client, cma_dev);
4517
4518         mutex_lock(&lock);
4519         list_add_tail(&cma_dev->list, &dev_list);
4520         list_for_each_entry(id_priv, &listen_any_list, list)
4521                 cma_listen_on_dev(id_priv, cma_dev);
4522         mutex_unlock(&lock);
4523
4524         return;
4525
4526 free_gid_type:
4527         kfree(cma_dev->default_gid_type);
4528
4529 free_cma_dev:
4530         kfree(cma_dev);
4531
4532         return;
4533 }
4534
4535 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4536 {
4537         struct rdma_cm_event event = {};
4538         enum rdma_cm_state state;
4539         int ret = 0;
4540
4541         /* Record that we want to remove the device */
4542         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4543         if (state == RDMA_CM_DESTROYING)
4544                 return 0;
4545
4546         cma_cancel_operation(id_priv, state);
4547         mutex_lock(&id_priv->handler_mutex);
4548
4549         /* Check for destruction from another callback. */
4550         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4551                 goto out;
4552
4553         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4554         ret = id_priv->id.event_handler(&id_priv->id, &event);
4555 out:
4556         mutex_unlock(&id_priv->handler_mutex);
4557         return ret;
4558 }
4559
4560 static void cma_process_remove(struct cma_device *cma_dev)
4561 {
4562         struct rdma_id_private *id_priv;
4563         int ret;
4564
4565         mutex_lock(&lock);
4566         while (!list_empty(&cma_dev->id_list)) {
4567                 id_priv = list_entry(cma_dev->id_list.next,
4568                                      struct rdma_id_private, list);
4569
4570                 list_del(&id_priv->listen_list);
4571                 list_del_init(&id_priv->list);
4572                 atomic_inc(&id_priv->refcount);
4573                 mutex_unlock(&lock);
4574
4575                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4576                 cma_deref_id(id_priv);
4577                 if (ret)
4578                         rdma_destroy_id(&id_priv->id);
4579
4580                 mutex_lock(&lock);
4581         }
4582         mutex_unlock(&lock);
4583
4584         cma_deref_dev(cma_dev);
4585         wait_for_completion(&cma_dev->comp);
4586 }
4587
4588 static void cma_remove_one(struct ib_device *device, void *client_data)
4589 {
4590         struct cma_device *cma_dev = client_data;
4591
4592         if (!cma_dev)
4593                 return;
4594
4595         mutex_lock(&lock);
4596         list_del(&cma_dev->list);
4597         mutex_unlock(&lock);
4598
4599         cma_process_remove(cma_dev);
4600         kfree(cma_dev->default_roce_tos);
4601         kfree(cma_dev->default_gid_type);
4602         kfree(cma_dev);
4603 }
4604
4605 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4606 {
4607         struct nlmsghdr *nlh;
4608         struct rdma_cm_id_stats *id_stats;
4609         struct rdma_id_private *id_priv;
4610         struct rdma_cm_id *id = NULL;
4611         struct cma_device *cma_dev;
4612         int i_dev = 0, i_id = 0;
4613
4614         /*
4615          * We export all of the IDs as a sequence of messages.  Each
4616          * ID gets its own netlink message.
4617          */
4618         mutex_lock(&lock);
4619
4620         list_for_each_entry(cma_dev, &dev_list, list) {
4621                 if (i_dev < cb->args[0]) {
4622                         i_dev++;
4623                         continue;
4624                 }
4625
4626                 i_id = 0;
4627                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4628                         if (i_id < cb->args[1]) {
4629                                 i_id++;
4630                                 continue;
4631                         }
4632
4633                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4634                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4635                                                 RDMA_NL_RDMA_CM_ID_STATS,
4636                                                 NLM_F_MULTI);
4637                         if (!id_stats)
4638                                 goto out;
4639
4640                         memset(id_stats, 0, sizeof *id_stats);
4641                         id = &id_priv->id;
4642                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4643                         id_stats->port_num = id->port_num;
4644                         id_stats->bound_dev_if =
4645                                 id->route.addr.dev_addr.bound_dev_if;
4646
4647                         if (ibnl_put_attr(skb, nlh,
4648                                           rdma_addr_size(cma_src_addr(id_priv)),
4649                                           cma_src_addr(id_priv),
4650                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4651                                 goto out;
4652                         if (ibnl_put_attr(skb, nlh,
4653                                           rdma_addr_size(cma_dst_addr(id_priv)),
4654                                           cma_dst_addr(id_priv),
4655                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4656                                 goto out;
4657
4658                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4659                         id_stats->port_space    = id->ps;
4660                         id_stats->cm_state      = id_priv->state;
4661                         id_stats->qp_num        = id_priv->qp_num;
4662                         id_stats->qp_type       = id->qp_type;
4663
4664                         i_id++;
4665                         nlmsg_end(skb, nlh);
4666                 }
4667
4668                 cb->args[1] = 0;
4669                 i_dev++;
4670         }
4671
4672 out:
4673         mutex_unlock(&lock);
4674         cb->args[0] = i_dev;
4675         cb->args[1] = i_id;
4676
4677         return skb->len;
4678 }
4679
4680 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4681         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4682 };
4683
4684 static int cma_init_net(struct net *net)
4685 {
4686         struct cma_pernet *pernet = cma_pernet(net);
4687
4688         idr_init(&pernet->tcp_ps);
4689         idr_init(&pernet->udp_ps);
4690         idr_init(&pernet->ipoib_ps);
4691         idr_init(&pernet->ib_ps);
4692
4693         return 0;
4694 }
4695
4696 static void cma_exit_net(struct net *net)
4697 {
4698         struct cma_pernet *pernet = cma_pernet(net);
4699
4700         idr_destroy(&pernet->tcp_ps);
4701         idr_destroy(&pernet->udp_ps);
4702         idr_destroy(&pernet->ipoib_ps);
4703         idr_destroy(&pernet->ib_ps);
4704 }
4705
4706 static struct pernet_operations cma_pernet_operations = {
4707         .init = cma_init_net,
4708         .exit = cma_exit_net,
4709         .id = &cma_pernet_id,
4710         .size = sizeof(struct cma_pernet),
4711 };
4712
4713 static int __init cma_init(void)
4714 {
4715         int ret;
4716
4717         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4718         if (!cma_wq)
4719                 return -ENOMEM;
4720
4721         ret = register_pernet_subsys(&cma_pernet_operations);
4722         if (ret)
4723                 goto err_wq;
4724
4725         ib_sa_register_client(&sa_client);
4726         register_netdevice_notifier(&cma_nb);
4727
4728         ret = ib_register_client(&cma_client);
4729         if (ret)
4730                 goto err;
4731
4732         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4733         cma_configfs_init();
4734
4735         return 0;
4736
4737 err:
4738         unregister_netdevice_notifier(&cma_nb);
4739         ib_sa_unregister_client(&sa_client);
4740 err_wq:
4741         destroy_workqueue(cma_wq);
4742         return ret;
4743 }
4744
4745 static void __exit cma_cleanup(void)
4746 {
4747         cma_configfs_exit();
4748         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4749         ib_unregister_client(&cma_client);
4750         unregister_netdevice_notifier(&cma_nb);
4751         ib_sa_unregister_client(&sa_client);
4752         unregister_pernet_subsys(&cma_pernet_operations);
4753         destroy_workqueue(cma_wq);
4754 }
4755
4756 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4757
4758 module_init(cma_init);
4759 module_exit(cma_cleanup);