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         if (id_priv->res.kern_name)
 498                 rdma_restrack_kadd(&id_priv->res);
 499         else
 500                 rdma_restrack_uadd(&id_priv->res);
 501 }
 502
 503 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 504                               struct cma_device *cma_dev)
 505 {
 506         _cma_attach_to_dev(id_priv, cma_dev);
 507         id_priv->gid_type =
 508                 cma_dev->default_gid_type[id_priv->id.port_num -
 509                                           rdma_start_port(cma_dev->device)];
 510 }
 511
 512 void cma_deref_dev(struct cma_device *cma_dev)
 513 {
 514         if (atomic_dec_and_test(&cma_dev->refcount))
 515                 complete(&cma_dev->comp);
 516 }
 517
 518 static inline void release_mc(struct kref *kref)
 519 {
 520         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 521
 522         kfree(mc->multicast.ib);
 523         kfree(mc);
 524 }
 525
 526 static void cma_release_dev(struct rdma_id_private *id_priv)
 527 {
 528         mutex_lock(&lock);
 529         list_del(&id_priv->list);
 530         cma_deref_dev(id_priv->cma_dev);
 531         id_priv->cma_dev = NULL;
 532         mutex_unlock(&lock);
 533 }
 534
 535 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 536 {
 537         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 538 }
 539
 540 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 541 {
 542         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 543 }
 544
 545 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 546 {
 547         return id_priv->id.route.addr.src_addr.ss_family;
 548 }
 549
 550 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 551 {
 552         struct ib_sa_mcmember_rec rec;
 553         int ret = 0;
 554
 555         if (id_priv->qkey) {
 556                 if (qkey && id_priv->qkey != qkey)
 557                         return -EINVAL;
 558                 return 0;
 559         }
 560
 561         if (qkey) {
 562                 id_priv->qkey = qkey;
 563                 return 0;
 564         }
 565
 566         switch (id_priv->id.ps) {
 567         case RDMA_PS_UDP:
 568         case RDMA_PS_IB:
 569                 id_priv->qkey = RDMA_UDP_QKEY;
 570                 break;
 571         case RDMA_PS_IPOIB:
 572                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 573                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 574                                              id_priv->id.port_num, &rec.mgid,
 575                                              &rec);
 576                 if (!ret)
 577                         id_priv->qkey = be32_to_cpu(rec.qkey);
 578                 break;
 579         default:
 580                 break;
 581         }
 582         return ret;
 583 }
 584
 585 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 586 {
 587         dev_addr->dev_type = ARPHRD_INFINIBAND;
 588         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 589         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 590 }
 591
 592 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 593 {
 594         int ret;
 595
 596         if (addr->sa_family != AF_IB) {
 597                 ret = rdma_translate_ip(addr, dev_addr);
 598         } else {
 599                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 600                 ret = 0;
 601         }
 602
 603         return ret;
 604 }
 605
 606 static const struct ib_gid_attr *
 607 cma_validate_port(struct ib_device *device, u8 port,
 608                   enum ib_gid_type gid_type,
 609                   union ib_gid *gid,
 610                   struct rdma_id_private *id_priv)
 611 {
 612         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 613         int bound_if_index = dev_addr->bound_dev_if;
 614         const struct ib_gid_attr *sgid_attr;
 615         int dev_type = dev_addr->dev_type;
 616         struct net_device *ndev = NULL;
 617
 618         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 619                 return ERR_PTR(-ENODEV);
 620
 621         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 622                 return ERR_PTR(-ENODEV);
 623
 624         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 625                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 626                 if (!ndev)
 627                         return ERR_PTR(-ENODEV);
 628         } else {
 629                 gid_type = IB_GID_TYPE_IB;
 630         }
 631
 632         sgid_attr = rdma_find_gid_by_port(device, gid, gid_type, port, ndev);
 633         if (ndev)
 634                 dev_put(ndev);
 635         return sgid_attr;
 636 }
 637
 638 static void cma_bind_sgid_attr(struct rdma_id_private *id_priv,
 639                                const struct ib_gid_attr *sgid_attr)
 640 {
 641         WARN_ON(id_priv->id.route.addr.dev_addr.sgid_attr);
 642         id_priv->id.route.addr.dev_addr.sgid_attr = sgid_attr;
 643 }
 644
 645 /**
 646  * cma_acquire_dev_by_src_ip - Acquire cma device, port, gid attribute
 647  * based on source ip address.
 648  * @id_priv:    cm_id which should be bound to cma device
 649  *
 650  * cma_acquire_dev_by_src_ip() binds cm id to cma device, port and GID attribute
 651  * based on source IP address. It returns 0 on success or error code otherwise.
 652  * It is applicable to active and passive side cm_id.
 653  */
 654 static int cma_acquire_dev_by_src_ip(struct rdma_id_private *id_priv)
 655 {
 656         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 657         const struct ib_gid_attr *sgid_attr;
 658         union ib_gid gid, iboe_gid, *gidp;
 659         struct cma_device *cma_dev;
 660         enum ib_gid_type gid_type;
 661         int ret = -ENODEV;
 662         u8 port;
 663
 664         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 665             id_priv->id.ps == RDMA_PS_IPOIB)
 666                 return -EINVAL;
 667
 668         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 669                     &iboe_gid);
 670
 671         memcpy(&gid, dev_addr->src_dev_addr +
 672                rdma_addr_gid_offset(dev_addr), sizeof(gid));
 673
 674         mutex_lock(&lock);
 675         list_for_each_entry(cma_dev, &dev_list, list) {
 676                 for (port = rdma_start_port(cma_dev->device);
 677                      port <= rdma_end_port(cma_dev->device); port++) {
 678                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 679                                &iboe_gid : &gid;
 680                         gid_type = cma_dev->default_gid_type[port - 1];
 681                         sgid_attr = cma_validate_port(cma_dev->device, port,
 682                                                       gid_type, gidp, id_priv);
 683                         if (!IS_ERR(sgid_attr)) {
 684                                 id_priv->id.port_num = port;
 685                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 686                                 cma_attach_to_dev(id_priv, cma_dev);
 687                                 ret = 0;
 688                                 goto out;
 689                         }
 690                 }
 691         }
 692 out:
 693         mutex_unlock(&lock);
 694         return ret;
 695 }
 696
 697 /**
 698  * cma_ib_acquire_dev - Acquire cma device, port and SGID attribute
 699  * @id_priv:            cm id to bind to cma device
 700  * @listen_id_priv:     listener cm id to match against
 701  * @req:                Pointer to req structure containaining incoming
 702  *                      request information
 703  * cma_ib_acquire_dev() acquires cma device, port and SGID attribute when
 704  * rdma device matches for listen_id and incoming request. It also verifies
 705  * that a GID table entry is present for the source address.
 706  * Returns 0 on success, or returns error code otherwise.
 707  */
 708 static int cma_ib_acquire_dev(struct rdma_id_private *id_priv,
 709                               const struct rdma_id_private *listen_id_priv,
 710                               struct cma_req_info *req)
 711 {
 712         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 713         const struct ib_gid_attr *sgid_attr;
 714         enum ib_gid_type gid_type;
 715         union ib_gid gid;
 716
 717         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 718             id_priv->id.ps == RDMA_PS_IPOIB)
 719                 return -EINVAL;
 720
 721         if (rdma_protocol_roce(req->device, req->port))
 722                 rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 723                             &gid);
 724         else
 725                 memcpy(&gid, dev_addr->src_dev_addr +
 726                        rdma_addr_gid_offset(dev_addr), sizeof(gid));
 727
 728         gid_type = listen_id_priv->cma_dev->default_gid_type[req->port - 1];
 729         sgid_attr = cma_validate_port(req->device, req->port,
 730                                       gid_type, &gid, id_priv);
 731         if (IS_ERR(sgid_attr))
 732                 return PTR_ERR(sgid_attr);
 733
 734         id_priv->id.port_num = req->port;
 735         cma_bind_sgid_attr(id_priv, sgid_attr);
 736         /* Need to acquire lock to protect against reader
 737          * of cma_dev->id_list such as cma_netdev_callback() and
 738          * cma_process_remove().
 739          */
 740         mutex_lock(&lock);
 741         cma_attach_to_dev(id_priv, listen_id_priv->cma_dev);
 742         mutex_unlock(&lock);
 743         return 0;
 744 }
 745
 746 static int cma_iw_acquire_dev(struct rdma_id_private *id_priv,
 747                               const struct rdma_id_private *listen_id_priv)
 748 {
 749         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 750         const struct ib_gid_attr *sgid_attr;
 751         struct cma_device *cma_dev;
 752         enum ib_gid_type gid_type;
 753         int ret = -ENODEV;
 754         union ib_gid gid;
 755         u8 port;
 756
 757         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 758             id_priv->id.ps == RDMA_PS_IPOIB)
 759                 return -EINVAL;
 760
 761         memcpy(&gid, dev_addr->src_dev_addr +
 762                rdma_addr_gid_offset(dev_addr), sizeof(gid));
 763
 764         mutex_lock(&lock);
 765
 766         cma_dev = listen_id_priv->cma_dev;
 767         port = listen_id_priv->id.port_num;
 768         gid_type = listen_id_priv->gid_type;
 769         sgid_attr = cma_validate_port(cma_dev->device, port,
 770                                       gid_type, &gid, id_priv);
 771         if (!IS_ERR(sgid_attr)) {
 772                 id_priv->id.port_num = port;
 773                 cma_bind_sgid_attr(id_priv, sgid_attr);
 774                 ret = 0;
 775                 goto out;
 776         }
 777
 778         list_for_each_entry(cma_dev, &dev_list, list) {
 779                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 780                         if (listen_id_priv->cma_dev == cma_dev &&
 781                             listen_id_priv->id.port_num == port)
 782                                 continue;
 783
 784                         gid_type = cma_dev->default_gid_type[port - 1];
 785                         sgid_attr = cma_validate_port(cma_dev->device, port,
 786                                                       gid_type, &gid, id_priv);
 787                         if (!IS_ERR(sgid_attr)) {
 788                                 id_priv->id.port_num = port;
 789                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 790                                 ret = 0;
 791                                 goto out;
 792                         }
 793                 }
 794         }
 795
 796 out:
 797         if (!ret)
 798                 cma_attach_to_dev(id_priv, cma_dev);
 799
 800         mutex_unlock(&lock);
 801         return ret;
 802 }
 803
 804 /*
 805  * Select the source IB device and address to reach the destination IB address.
 806  */
 807 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 808 {
 809         struct cma_device *cma_dev, *cur_dev;
 810         struct sockaddr_ib *addr;
 811         union ib_gid gid, sgid, *dgid;
 812         u16 pkey, index;
 813         u8 p;
 814         enum ib_port_state port_state;
 815         int i;
 816
 817         cma_dev = NULL;
 818         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 819         dgid = (union ib_gid *) &addr->sib_addr;
 820         pkey = ntohs(addr->sib_pkey);
 821
 822         mutex_lock(&lock);
 823         list_for_each_entry(cur_dev, &dev_list, list) {
 824                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 825                         if (!rdma_cap_af_ib(cur_dev->device, p))
 826                                 continue;
 827
 828                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 829                                 continue;
 830
 831                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 832                                 continue;
 833                         for (i = 0; !rdma_query_gid(cur_dev->device,
 834                                                     p, i, &gid);
 835                              i++) {
 836                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 837                                         cma_dev = cur_dev;
 838                                         sgid = gid;
 839                                         id_priv->id.port_num = p;
 840                                         goto found;
 841                                 }
 842
 843                                 if (!cma_dev && (gid.global.subnet_prefix ==
 844                                     dgid->global.subnet_prefix) &&
 845                                     port_state == IB_PORT_ACTIVE) {
 846                                         cma_dev = cur_dev;
 847                                         sgid = gid;
 848                                         id_priv->id.port_num = p;
 849                                         goto found;
 850                                 }
 851                         }
 852                 }
 853         }
 854         mutex_unlock(&lock);
 855         return -ENODEV;
 856
 857 found:
 858         cma_attach_to_dev(id_priv, cma_dev);
 859         mutex_unlock(&lock);
 860         addr = (struct sockaddr_ib *)cma_src_addr(id_priv);
 861         memcpy(&addr->sib_addr, &sgid, sizeof(sgid));
 862         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 863         return 0;
 864 }
 865
 866 static void cma_deref_id(struct rdma_id_private *id_priv)
 867 {
 868         if (atomic_dec_and_test(&id_priv->refcount))
 869                 complete(&id_priv->comp);
 870 }
 871
 872 struct rdma_cm_id *__rdma_create_id(struct net *net,
 873                                     rdma_cm_event_handler event_handler,
 874                                     void *context, enum rdma_ucm_port_space ps,
 875                                     enum ib_qp_type qp_type, const char *caller)
 876 {
 877         struct rdma_id_private *id_priv;
 878
 879         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 880         if (!id_priv)
 881                 return ERR_PTR(-ENOMEM);
 882
 883         rdma_restrack_set_task(&id_priv->res, caller);
 884         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 885         id_priv->state = RDMA_CM_IDLE;
 886         id_priv->id.context = context;
 887         id_priv->id.event_handler = event_handler;
 888         id_priv->id.ps = ps;
 889         id_priv->id.qp_type = qp_type;
 890         id_priv->tos_set = false;
 891         id_priv->gid_type = IB_GID_TYPE_IB;
 892         spin_lock_init(&id_priv->lock);
 893         mutex_init(&id_priv->qp_mutex);
 894         init_completion(&id_priv->comp);
 895         atomic_set(&id_priv->refcount, 1);
 896         mutex_init(&id_priv->handler_mutex);
 897         INIT_LIST_HEAD(&id_priv->listen_list);
 898         INIT_LIST_HEAD(&id_priv->mc_list);
 899         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 900         id_priv->id.route.addr.dev_addr.net = get_net(net);
 901         id_priv->seq_num &= 0x00ffffff;
 902
 903         return &id_priv->id;
 904 }
 905 EXPORT_SYMBOL(__rdma_create_id);
 906
 907 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 908 {
 909         struct ib_qp_attr qp_attr;
 910         int qp_attr_mask, ret;
 911
 912         qp_attr.qp_state = IB_QPS_INIT;
 913         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 914         if (ret)
 915                 return ret;
 916
 917         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 918         if (ret)
 919                 return ret;
 920
 921         qp_attr.qp_state = IB_QPS_RTR;
 922         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 923         if (ret)
 924                 return ret;
 925
 926         qp_attr.qp_state = IB_QPS_RTS;
 927         qp_attr.sq_psn = 0;
 928         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 929
 930         return ret;
 931 }
 932
 933 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 934 {
 935         struct ib_qp_attr qp_attr;
 936         int qp_attr_mask, ret;
 937
 938         qp_attr.qp_state = IB_QPS_INIT;
 939         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 940         if (ret)
 941                 return ret;
 942
 943         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 944 }
 945
 946 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 947                    struct ib_qp_init_attr *qp_init_attr)
 948 {
 949         struct rdma_id_private *id_priv;
 950         struct ib_qp *qp;
 951         int ret;
 952
 953         id_priv = container_of(id, struct rdma_id_private, id);
 954         if (id->device != pd->device)
 955                 return -EINVAL;
 956
 957         qp_init_attr->port_num = id->port_num;
 958         qp = ib_create_qp(pd, qp_init_attr);
 959         if (IS_ERR(qp))
 960                 return PTR_ERR(qp);
 961
 962         if (id->qp_type == IB_QPT_UD)
 963                 ret = cma_init_ud_qp(id_priv, qp);
 964         else
 965                 ret = cma_init_conn_qp(id_priv, qp);
 966         if (ret)
 967                 goto err;
 968
 969         id->qp = qp;
 970         id_priv->qp_num = qp->qp_num;
 971         id_priv->srq = (qp->srq != NULL);
 972         return 0;
 973 err:
 974         ib_destroy_qp(qp);
 975         return ret;
 976 }
 977 EXPORT_SYMBOL(rdma_create_qp);
 978
 979 void rdma_destroy_qp(struct rdma_cm_id *id)
 980 {
 981         struct rdma_id_private *id_priv;
 982
 983         id_priv = container_of(id, struct rdma_id_private, id);
 984         mutex_lock(&id_priv->qp_mutex);
 985         ib_destroy_qp(id_priv->id.qp);
 986         id_priv->id.qp = NULL;
 987         mutex_unlock(&id_priv->qp_mutex);
 988 }
 989 EXPORT_SYMBOL(rdma_destroy_qp);
 990
 991 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 992                              struct rdma_conn_param *conn_param)
 993 {
 994         struct ib_qp_attr qp_attr;
 995         int qp_attr_mask, ret;
 996
 997         mutex_lock(&id_priv->qp_mutex);
 998         if (!id_priv->id.qp) {
 999                 ret = 0;
1000                 goto out;
1001         }
1002
1003         /* Need to update QP attributes from default values. */
1004         qp_attr.qp_state = IB_QPS_INIT;
1005         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1006         if (ret)
1007                 goto out;
1008
1009         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1010         if (ret)
1011                 goto out;
1012
1013         qp_attr.qp_state = IB_QPS_RTR;
1014         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1015         if (ret)
1016                 goto out;
1017
1018         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
1019
1020         if (conn_param)
1021                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
1022         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1023 out:
1024         mutex_unlock(&id_priv->qp_mutex);
1025         return ret;
1026 }
1027
1028 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
1029                              struct rdma_conn_param *conn_param)
1030 {
1031         struct ib_qp_attr qp_attr;
1032         int qp_attr_mask, ret;
1033
1034         mutex_lock(&id_priv->qp_mutex);
1035         if (!id_priv->id.qp) {
1036                 ret = 0;
1037                 goto out;
1038         }
1039
1040         qp_attr.qp_state = IB_QPS_RTS;
1041         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
1042         if (ret)
1043                 goto out;
1044
1045         if (conn_param)
1046                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
1047         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
1048 out:
1049         mutex_unlock(&id_priv->qp_mutex);
1050         return ret;
1051 }
1052
1053 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
1054 {
1055         struct ib_qp_attr qp_attr;
1056         int ret;
1057
1058         mutex_lock(&id_priv->qp_mutex);
1059         if (!id_priv->id.qp) {
1060                 ret = 0;
1061                 goto out;
1062         }
1063
1064         qp_attr.qp_state = IB_QPS_ERR;
1065         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
1066 out:
1067         mutex_unlock(&id_priv->qp_mutex);
1068         return ret;
1069 }
1070
1071 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
1072                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
1073 {
1074         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1075         int ret;
1076         u16 pkey;
1077
1078         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
1079                 pkey = 0xffff;
1080         else
1081                 pkey = ib_addr_get_pkey(dev_addr);
1082
1083         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
1084                                   pkey, &qp_attr->pkey_index);
1085         if (ret)
1086                 return ret;
1087
1088         qp_attr->port_num = id_priv->id.port_num;
1089         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1090
1091         if (id_priv->id.qp_type == IB_QPT_UD) {
1092                 ret = cma_set_qkey(id_priv, 0);
1093                 if (ret)
1094                         return ret;
1095
1096                 qp_attr->qkey = id_priv->qkey;
1097                 *qp_attr_mask |= IB_QP_QKEY;
1098         } else {
1099                 qp_attr->qp_access_flags = 0;
1100                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1101         }
1102         return 0;
1103 }
1104
1105 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1106                        int *qp_attr_mask)
1107 {
1108         struct rdma_id_private *id_priv;
1109         int ret = 0;
1110
1111         id_priv = container_of(id, struct rdma_id_private, id);
1112         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1113                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1114                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1115                 else
1116                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1117                                                  qp_attr_mask);
1118
1119                 if (qp_attr->qp_state == IB_QPS_RTR)
1120                         qp_attr->rq_psn = id_priv->seq_num;
1121         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1122                 if (!id_priv->cm_id.iw) {
1123                         qp_attr->qp_access_flags = 0;
1124                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1125                 } else
1126                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1127                                                  qp_attr_mask);
1128                 qp_attr->port_num = id_priv->id.port_num;
1129                 *qp_attr_mask |= IB_QP_PORT;
1130         } else
1131                 ret = -ENOSYS;
1132
1133         return ret;
1134 }
1135 EXPORT_SYMBOL(rdma_init_qp_attr);
1136
1137 static inline bool cma_zero_addr(const struct sockaddr *addr)
1138 {
1139         switch (addr->sa_family) {
1140         case AF_INET:
1141                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1142         case AF_INET6:
1143                 return ipv6_addr_any(&((struct sockaddr_in6 *)addr)->sin6_addr);
1144         case AF_IB:
1145                 return ib_addr_any(&((struct sockaddr_ib *)addr)->sib_addr);
1146         default:
1147                 return false;
1148         }
1149 }
1150
1151 static inline bool cma_loopback_addr(const struct sockaddr *addr)
1152 {
1153         switch (addr->sa_family) {
1154         case AF_INET:
1155                 return ipv4_is_loopback(
1156                         ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1157         case AF_INET6:
1158                 return ipv6_addr_loopback(
1159                         &((struct sockaddr_in6 *)addr)->sin6_addr);
1160         case AF_IB:
1161                 return ib_addr_loopback(
1162                         &((struct sockaddr_ib *)addr)->sib_addr);
1163         default:
1164                 return false;
1165         }
1166 }
1167
1168 static inline bool cma_any_addr(const struct sockaddr *addr)
1169 {
1170         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1171 }
1172
1173 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1174 {
1175         if (src->sa_family != dst->sa_family)
1176                 return -1;
1177
1178         switch (src->sa_family) {
1179         case AF_INET:
1180                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1181                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1182         case AF_INET6:
1183                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1184                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1185         default:
1186                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1187                                    &((struct sockaddr_ib *) dst)->sib_addr);
1188         }
1189 }
1190
1191 static __be16 cma_port(const struct sockaddr *addr)
1192 {
1193         struct sockaddr_ib *sib;
1194
1195         switch (addr->sa_family) {
1196         case AF_INET:
1197                 return ((struct sockaddr_in *) addr)->sin_port;
1198         case AF_INET6:
1199                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1200         case AF_IB:
1201                 sib = (struct sockaddr_ib *) addr;
1202                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1203                                     be64_to_cpu(sib->sib_sid_mask)));
1204         default:
1205                 return 0;
1206         }
1207 }
1208
1209 static inline int cma_any_port(const struct sockaddr *addr)
1210 {
1211         return !cma_port(addr);
1212 }
1213
1214 static void cma_save_ib_info(struct sockaddr *src_addr,
1215                              struct sockaddr *dst_addr,
1216                              const struct rdma_cm_id *listen_id,
1217                              const struct sa_path_rec *path)
1218 {
1219         struct sockaddr_ib *listen_ib, *ib;
1220
1221         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1222         if (src_addr) {
1223                 ib = (struct sockaddr_ib *)src_addr;
1224                 ib->sib_family = AF_IB;
1225                 if (path) {
1226                         ib->sib_pkey = path->pkey;
1227                         ib->sib_flowinfo = path->flow_label;
1228                         memcpy(&ib->sib_addr, &path->sgid, 16);
1229                         ib->sib_sid = path->service_id;
1230                         ib->sib_scope_id = 0;
1231                 } else {
1232                         ib->sib_pkey = listen_ib->sib_pkey;
1233                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1234                         ib->sib_addr = listen_ib->sib_addr;
1235                         ib->sib_sid = listen_ib->sib_sid;
1236                         ib->sib_scope_id = listen_ib->sib_scope_id;
1237                 }
1238                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1239         }
1240         if (dst_addr) {
1241                 ib = (struct sockaddr_ib *)dst_addr;
1242                 ib->sib_family = AF_IB;
1243                 if (path) {
1244                         ib->sib_pkey = path->pkey;
1245                         ib->sib_flowinfo = path->flow_label;
1246                         memcpy(&ib->sib_addr, &path->dgid, 16);
1247                 }
1248         }
1249 }
1250
1251 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1252                               struct sockaddr_in *dst_addr,
1253                               struct cma_hdr *hdr,
1254                               __be16 local_port)
1255 {
1256         if (src_addr) {
1257                 *src_addr = (struct sockaddr_in) {
1258                         .sin_family = AF_INET,
1259                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1260                         .sin_port = local_port,
1261                 };
1262         }
1263
1264         if (dst_addr) {
1265                 *dst_addr = (struct sockaddr_in) {
1266                         .sin_family = AF_INET,
1267                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1268                         .sin_port = hdr->port,
1269                 };
1270         }
1271 }
1272
1273 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1274                               struct sockaddr_in6 *dst_addr,
1275                               struct cma_hdr *hdr,
1276                               __be16 local_port)
1277 {
1278         if (src_addr) {
1279                 *src_addr = (struct sockaddr_in6) {
1280                         .sin6_family = AF_INET6,
1281                         .sin6_addr = hdr->dst_addr.ip6,
1282                         .sin6_port = local_port,
1283                 };
1284         }
1285
1286         if (dst_addr) {
1287                 *dst_addr = (struct sockaddr_in6) {
1288                         .sin6_family = AF_INET6,
1289                         .sin6_addr = hdr->src_addr.ip6,
1290                         .sin6_port = hdr->port,
1291                 };
1292         }
1293 }
1294
1295 static u16 cma_port_from_service_id(__be64 service_id)
1296 {
1297         return (u16)be64_to_cpu(service_id);
1298 }
1299
1300 static int cma_save_ip_info(struct sockaddr *src_addr,
1301                             struct sockaddr *dst_addr,
1302                             const struct ib_cm_event *ib_event,
1303                             __be64 service_id)
1304 {
1305         struct cma_hdr *hdr;
1306         __be16 port;
1307
1308         hdr = ib_event->private_data;
1309         if (hdr->cma_version != CMA_VERSION)
1310                 return -EINVAL;
1311
1312         port = htons(cma_port_from_service_id(service_id));
1313
1314         switch (cma_get_ip_ver(hdr)) {
1315         case 4:
1316                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1317                                   (struct sockaddr_in *)dst_addr, hdr, port);
1318                 break;
1319         case 6:
1320                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1321                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1322                 break;
1323         default:
1324                 return -EAFNOSUPPORT;
1325         }
1326
1327         return 0;
1328 }
1329
1330 static int cma_save_net_info(struct sockaddr *src_addr,
1331                              struct sockaddr *dst_addr,
1332                              const struct rdma_cm_id *listen_id,
1333                              const struct ib_cm_event *ib_event,
1334                              sa_family_t sa_family, __be64 service_id)
1335 {
1336         if (sa_family == AF_IB) {
1337                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1338                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1339                                          ib_event->param.req_rcvd.primary_path);
1340                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1341                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1342                 return 0;
1343         }
1344
1345         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1346 }
1347
1348 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1349                              struct cma_req_info *req)
1350 {
1351         const struct ib_cm_req_event_param *req_param =
1352                 &ib_event->param.req_rcvd;
1353         const struct ib_cm_sidr_req_event_param *sidr_param =
1354                 &ib_event->param.sidr_req_rcvd;
1355
1356         switch (ib_event->event) {
1357         case IB_CM_REQ_RECEIVED:
1358                 req->device     = req_param->listen_id->device;
1359                 req->port       = req_param->port;
1360                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1361                        sizeof(req->local_gid));
1362                 req->has_gid    = true;
1363                 req->service_id = req_param->primary_path->service_id;
1364                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1365                 if (req->pkey != req_param->bth_pkey)
1366                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1367                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1368                                             req_param->bth_pkey, req->pkey);
1369                 break;
1370         case IB_CM_SIDR_REQ_RECEIVED:
1371                 req->device     = sidr_param->listen_id->device;
1372                 req->port       = sidr_param->port;
1373                 req->has_gid    = false;
1374                 req->service_id = sidr_param->service_id;
1375                 req->pkey       = sidr_param->pkey;
1376                 if (req->pkey != sidr_param->bth_pkey)
1377                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1378                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1379                                             sidr_param->bth_pkey, req->pkey);
1380                 break;
1381         default:
1382                 return -EINVAL;
1383         }
1384
1385         return 0;
1386 }
1387
1388 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1389                                   const struct sockaddr_in *dst_addr,
1390                                   const struct sockaddr_in *src_addr)
1391 {
1392         __be32 daddr = dst_addr->sin_addr.s_addr,
1393                saddr = src_addr->sin_addr.s_addr;
1394         struct fib_result res;
1395         struct flowi4 fl4;
1396         int err;
1397         bool ret;
1398
1399         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1400             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1401             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1402             ipv4_is_loopback(saddr))
1403                 return false;
1404
1405         memset(&fl4, 0, sizeof(fl4));
1406         fl4.flowi4_iif = net_dev->ifindex;
1407         fl4.daddr = daddr;
1408         fl4.saddr = saddr;
1409
1410         rcu_read_lock();
1411         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1412         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1413         rcu_read_unlock();
1414
1415         return ret;
1416 }
1417
1418 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1419                                   const struct sockaddr_in6 *dst_addr,
1420                                   const struct sockaddr_in6 *src_addr)
1421 {
1422 #if IS_ENABLED(CONFIG_IPV6)
1423         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1424                            IPV6_ADDR_LINKLOCAL;
1425         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1426                                          &src_addr->sin6_addr, net_dev->ifindex,
1427                                          NULL, strict);
1428         bool ret;
1429
1430         if (!rt)
1431                 return false;
1432
1433         ret = rt->rt6i_idev->dev == net_dev;
1434         ip6_rt_put(rt);
1435
1436         return ret;
1437 #else
1438         return false;
1439 #endif
1440 }
1441
1442 static bool validate_net_dev(struct net_device *net_dev,
1443                              const struct sockaddr *daddr,
1444                              const struct sockaddr *saddr)
1445 {
1446         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1447         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1448         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1449         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1450
1451         switch (daddr->sa_family) {
1452         case AF_INET:
1453                 return saddr->sa_family == AF_INET &&
1454                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1455
1456         case AF_INET6:
1457                 return saddr->sa_family == AF_INET6 &&
1458                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1459
1460         default:
1461                 return false;
1462         }
1463 }
1464
1465 static struct net_device *
1466 roce_get_net_dev_by_cm_event(const struct ib_cm_event *ib_event)
1467 {
1468         const struct ib_gid_attr *sgid_attr = NULL;
1469
1470         if (ib_event->event == IB_CM_REQ_RECEIVED)
1471                 sgid_attr = ib_event->param.req_rcvd.ppath_sgid_attr;
1472         else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1473                 sgid_attr = ib_event->param.sidr_req_rcvd.sgid_attr;
1474
1475         if (!sgid_attr)
1476                 return NULL;
1477         dev_hold(sgid_attr->ndev);
1478         return sgid_attr->ndev;
1479 }
1480
1481 static struct net_device *cma_get_net_dev(const struct ib_cm_event *ib_event,
1482                                           struct cma_req_info *req)
1483 {
1484         struct sockaddr *listen_addr =
1485                         (struct sockaddr *)&req->listen_addr_storage;
1486         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1487         struct net_device *net_dev;
1488         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1489         int err;
1490
1491         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1492                                req->service_id);
1493         if (err)
1494                 return ERR_PTR(err);
1495
1496         if (rdma_protocol_roce(req->device, req->port))
1497                 net_dev = roce_get_net_dev_by_cm_event(ib_event);
1498         else
1499                 net_dev = ib_get_net_dev_by_params(req->device, req->port,
1500                                                    req->pkey,
1501                                                    gid, listen_addr);
1502         if (!net_dev)
1503                 return ERR_PTR(-ENODEV);
1504
1505         return net_dev;
1506 }
1507
1508 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1509 {
1510         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1511 }
1512
1513 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1514                                    const struct cma_hdr *hdr)
1515 {
1516         struct sockaddr *addr = cma_src_addr(id_priv);
1517         __be32 ip4_addr;
1518         struct in6_addr ip6_addr;
1519
1520         if (cma_any_addr(addr) && !id_priv->afonly)
1521                 return true;
1522
1523         switch (addr->sa_family) {
1524         case AF_INET:
1525                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1526                 if (cma_get_ip_ver(hdr) != 4)
1527                         return false;
1528                 if (!cma_any_addr(addr) &&
1529                     hdr->dst_addr.ip4.addr != ip4_addr)
1530                         return false;
1531                 break;
1532         case AF_INET6:
1533                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1534                 if (cma_get_ip_ver(hdr) != 6)
1535                         return false;
1536                 if (!cma_any_addr(addr) &&
1537                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1538                         return false;
1539                 break;
1540         case AF_IB:
1541                 return true;
1542         default:
1543                 return false;
1544         }
1545
1546         return true;
1547 }
1548
1549 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1550 {
1551         struct ib_device *device = id->device;
1552         const int port_num = id->port_num ?: rdma_start_port(device);
1553
1554         return rdma_protocol_roce(device, port_num);
1555 }
1556
1557 static bool cma_is_req_ipv6_ll(const struct cma_req_info *req)
1558 {
1559         const struct sockaddr *daddr =
1560                         (const struct sockaddr *)&req->listen_addr_storage;
1561         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1562
1563         /* Returns true if the req is for IPv6 link local */
1564         return (daddr->sa_family == AF_INET6 &&
1565                 (ipv6_addr_type(&daddr6->sin6_addr) & IPV6_ADDR_LINKLOCAL));
1566 }
1567
1568 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1569                               const struct net_device *net_dev,
1570                               const struct cma_req_info *req)
1571 {
1572         const struct rdma_addr *addr = &id->route.addr;
1573
1574         if (!net_dev)
1575                 /* This request is an AF_IB request */
1576                 return (!id->port_num || id->port_num == req->port) &&
1577                        (addr->src_addr.ss_family == AF_IB);
1578
1579         /*
1580          * If the request is not for IPv6 link local, allow matching
1581          * request to any netdevice of the one or multiport rdma device.
1582          */
1583         if (!cma_is_req_ipv6_ll(req))
1584                 return true;
1585         /*
1586          * Net namespaces must match, and if the listner is listening
1587          * on a specific netdevice than netdevice must match as well.
1588          */
1589         if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1590             (!!addr->dev_addr.bound_dev_if ==
1591              (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1592                 return true;
1593         else
1594                 return false;
1595 }
1596
1597 static struct rdma_id_private *cma_find_listener(
1598                 const struct rdma_bind_list *bind_list,
1599                 const struct ib_cm_id *cm_id,
1600                 const struct ib_cm_event *ib_event,
1601                 const struct cma_req_info *req,
1602                 const struct net_device *net_dev)
1603 {
1604         struct rdma_id_private *id_priv, *id_priv_dev;
1605
1606         if (!bind_list)
1607                 return ERR_PTR(-EINVAL);
1608
1609         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1610                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1611                         if (id_priv->id.device == cm_id->device &&
1612                             cma_match_net_dev(&id_priv->id, net_dev, req))
1613                                 return id_priv;
1614                         list_for_each_entry(id_priv_dev,
1615                                             &id_priv->listen_list,
1616                                             listen_list) {
1617                                 if (id_priv_dev->id.device == cm_id->device &&
1618                                     cma_match_net_dev(&id_priv_dev->id,
1619                                                       net_dev, req))
1620                                         return id_priv_dev;
1621                         }
1622                 }
1623         }
1624
1625         return ERR_PTR(-EINVAL);
1626 }
1627
1628 static struct rdma_id_private *
1629 cma_ib_id_from_event(struct ib_cm_id *cm_id,
1630                      const struct ib_cm_event *ib_event,
1631                      struct cma_req_info *req,
1632                      struct net_device **net_dev)
1633 {
1634         struct rdma_bind_list *bind_list;
1635         struct rdma_id_private *id_priv;
1636         int err;
1637
1638         err = cma_save_req_info(ib_event, req);
1639         if (err)
1640                 return ERR_PTR(err);
1641
1642         *net_dev = cma_get_net_dev(ib_event, req);
1643         if (IS_ERR(*net_dev)) {
1644                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1645                         /* Assuming the protocol is AF_IB */
1646                         *net_dev = NULL;
1647                 } else {
1648                         return ERR_CAST(*net_dev);
1649                 }
1650         }
1651
1652         /*
1653          * Net namespace might be getting deleted while route lookup,
1654          * cm_id lookup is in progress. Therefore, perform netdevice
1655          * validation, cm_id lookup under rcu lock.
1656          * RCU lock along with netdevice state check, synchronizes with
1657          * netdevice migrating to different net namespace and also avoids
1658          * case where net namespace doesn't get deleted while lookup is in
1659          * progress.
1660          * If the device state is not IFF_UP, its properties such as ifindex
1661          * and nd_net cannot be trusted to remain valid without rcu lock.
1662          * net/core/dev.c change_net_namespace() ensures to synchronize with
1663          * ongoing operations on net device after device is closed using
1664          * synchronize_net().
1665          */
1666         rcu_read_lock();
1667         if (*net_dev) {
1668                 /*
1669                  * If netdevice is down, it is likely that it is administratively
1670                  * down or it might be migrating to different namespace.
1671                  * In that case avoid further processing, as the net namespace
1672                  * or ifindex may change.
1673                  */
1674                 if (((*net_dev)->flags & IFF_UP) == 0) {
1675                         id_priv = ERR_PTR(-EHOSTUNREACH);
1676                         goto err;
1677                 }
1678
1679                 if (!validate_net_dev(*net_dev,
1680                                  (struct sockaddr *)&req->listen_addr_storage,
1681                                  (struct sockaddr *)&req->src_addr_storage)) {
1682                         id_priv = ERR_PTR(-EHOSTUNREACH);
1683                         goto err;
1684                 }
1685         }
1686
1687         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1688                                 rdma_ps_from_service_id(req->service_id),
1689                                 cma_port_from_service_id(req->service_id));
1690         id_priv = cma_find_listener(bind_list, cm_id, ib_event, req, *net_dev);
1691 err:
1692         rcu_read_unlock();
1693         if (IS_ERR(id_priv) && *net_dev) {
1694                 dev_put(*net_dev);
1695                 *net_dev = NULL;
1696         }
1697         return id_priv;
1698 }
1699
1700 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1701 {
1702         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1703 }
1704
1705 static void cma_cancel_route(struct rdma_id_private *id_priv)
1706 {
1707         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1708                 if (id_priv->query)
1709                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1710         }
1711 }
1712
1713 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1714 {
1715         struct rdma_id_private *dev_id_priv;
1716
1717         /*
1718          * Remove from listen_any_list to prevent added devices from spawning
1719          * additional listen requests.
1720          */
1721         mutex_lock(&lock);
1722         list_del(&id_priv->list);
1723
1724         while (!list_empty(&id_priv->listen_list)) {
1725                 dev_id_priv = list_entry(id_priv->listen_list.next,
1726                                          struct rdma_id_private, listen_list);
1727                 /* sync with device removal to avoid duplicate destruction */
1728                 list_del_init(&dev_id_priv->list);
1729                 list_del(&dev_id_priv->listen_list);
1730                 mutex_unlock(&lock);
1731
1732                 rdma_destroy_id(&dev_id_priv->id);
1733                 mutex_lock(&lock);
1734         }
1735         mutex_unlock(&lock);
1736 }
1737
1738 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1739                                  enum rdma_cm_state state)
1740 {
1741         switch (state) {
1742         case RDMA_CM_ADDR_QUERY:
1743                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1744                 break;
1745         case RDMA_CM_ROUTE_QUERY:
1746                 cma_cancel_route(id_priv);
1747                 break;
1748         case RDMA_CM_LISTEN:
1749                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1750                         cma_cancel_listens(id_priv);
1751                 break;
1752         default:
1753                 break;
1754         }
1755 }
1756
1757 static void cma_release_port(struct rdma_id_private *id_priv)
1758 {
1759         struct rdma_bind_list *bind_list = id_priv->bind_list;
1760         struct net *net = id_priv->id.route.addr.dev_addr.net;
1761
1762         if (!bind_list)
1763                 return;
1764
1765         mutex_lock(&lock);
1766         hlist_del(&id_priv->node);
1767         if (hlist_empty(&bind_list->owners)) {
1768                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1769                 kfree(bind_list);
1770         }
1771         mutex_unlock(&lock);
1772 }
1773
1774 static void cma_leave_roce_mc_group(struct rdma_id_private *id_priv,
1775                                     struct cma_multicast *mc)
1776 {
1777         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1778         struct net_device *ndev = NULL;
1779
1780         if (dev_addr->bound_dev_if)
1781                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
1782         if (ndev) {
1783                 cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid, false);
1784                 dev_put(ndev);
1785         }
1786         kref_put(&mc->mcref, release_mc);
1787 }
1788
1789 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1790 {
1791         struct cma_multicast *mc;
1792
1793         while (!list_empty(&id_priv->mc_list)) {
1794                 mc = container_of(id_priv->mc_list.next,
1795                                   struct cma_multicast, list);
1796                 list_del(&mc->list);
1797                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1798                                       id_priv->id.port_num)) {
1799                         ib_sa_free_multicast(mc->multicast.ib);
1800                         kfree(mc);
1801                 } else {
1802                         cma_leave_roce_mc_group(id_priv, mc);
1803                 }
1804         }
1805 }
1806
1807 void rdma_destroy_id(struct rdma_cm_id *id)
1808 {
1809         struct rdma_id_private *id_priv;
1810         enum rdma_cm_state state;
1811
1812         id_priv = container_of(id, struct rdma_id_private, id);
1813         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1814         cma_cancel_operation(id_priv, state);
1815
1816         /*
1817          * Wait for any active callback to finish.  New callbacks will find
1818          * the id_priv state set to destroying and abort.
1819          */
1820         mutex_lock(&id_priv->handler_mutex);
1821         mutex_unlock(&id_priv->handler_mutex);
1822
1823         rdma_restrack_del(&id_priv->res);
1824         if (id_priv->cma_dev) {
1825                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1826                         if (id_priv->cm_id.ib)
1827                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1828                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1829                         if (id_priv->cm_id.iw)
1830                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1831                 }
1832                 cma_leave_mc_groups(id_priv);
1833                 cma_release_dev(id_priv);
1834         }
1835
1836         cma_release_port(id_priv);
1837         cma_deref_id(id_priv);
1838         wait_for_completion(&id_priv->comp);
1839
1840         if (id_priv->internal_id)
1841                 cma_deref_id(id_priv->id.context);
1842
1843         kfree(id_priv->id.route.path_rec);
1844
1845         if (id_priv->id.route.addr.dev_addr.sgid_attr)
1846                 rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
1847
1848         put_net(id_priv->id.route.addr.dev_addr.net);
1849         kfree(id_priv);
1850 }
1851 EXPORT_SYMBOL(rdma_destroy_id);
1852
1853 static int cma_rep_recv(struct rdma_id_private *id_priv)
1854 {
1855         int ret;
1856
1857         ret = cma_modify_qp_rtr(id_priv, NULL);
1858         if (ret)
1859                 goto reject;
1860
1861         ret = cma_modify_qp_rts(id_priv, NULL);
1862         if (ret)
1863                 goto reject;
1864
1865         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1866         if (ret)
1867                 goto reject;
1868
1869         return 0;
1870 reject:
1871         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1872         cma_modify_qp_err(id_priv);
1873         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1874                        NULL, 0, NULL, 0);
1875         return ret;
1876 }
1877
1878 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1879                                    const struct ib_cm_rep_event_param *rep_data,
1880                                    void *private_data)
1881 {
1882         event->param.conn.private_data = private_data;
1883         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1884         event->param.conn.responder_resources = rep_data->responder_resources;
1885         event->param.conn.initiator_depth = rep_data->initiator_depth;
1886         event->param.conn.flow_control = rep_data->flow_control;
1887         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1888         event->param.conn.srq = rep_data->srq;
1889         event->param.conn.qp_num = rep_data->remote_qpn;
1890 }
1891
1892 static int cma_ib_handler(struct ib_cm_id *cm_id,
1893                           const struct ib_cm_event *ib_event)
1894 {
1895         struct rdma_id_private *id_priv = cm_id->context;
1896         struct rdma_cm_event event = {};
1897         int ret = 0;
1898
1899         mutex_lock(&id_priv->handler_mutex);
1900         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1901              id_priv->state != RDMA_CM_CONNECT) ||
1902             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1903              id_priv->state != RDMA_CM_DISCONNECT))
1904                 goto out;
1905
1906         switch (ib_event->event) {
1907         case IB_CM_REQ_ERROR:
1908         case IB_CM_REP_ERROR:
1909                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1910                 event.status = -ETIMEDOUT;
1911                 break;
1912         case IB_CM_REP_RECEIVED:
1913                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1914                     (id_priv->id.qp_type != IB_QPT_UD))
1915                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1916                 if (id_priv->id.qp) {
1917                         event.status = cma_rep_recv(id_priv);
1918                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1919                                                      RDMA_CM_EVENT_ESTABLISHED;
1920                 } else {
1921                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1922                 }
1923                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1924                                        ib_event->private_data);
1925                 break;
1926         case IB_CM_RTU_RECEIVED:
1927         case IB_CM_USER_ESTABLISHED:
1928                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1929                 break;
1930         case IB_CM_DREQ_ERROR:
1931                 event.status = -ETIMEDOUT; /* fall through */
1932         case IB_CM_DREQ_RECEIVED:
1933         case IB_CM_DREP_RECEIVED:
1934                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1935                                    RDMA_CM_DISCONNECT))
1936                         goto out;
1937                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1938                 break;
1939         case IB_CM_TIMEWAIT_EXIT:
1940                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1941                 break;
1942         case IB_CM_MRA_RECEIVED:
1943                 /* ignore event */
1944                 goto out;
1945         case IB_CM_REJ_RECEIVED:
1946                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1947                                                                                 ib_event->param.rej_rcvd.reason));
1948                 cma_modify_qp_err(id_priv);
1949                 event.status = ib_event->param.rej_rcvd.reason;
1950                 event.event = RDMA_CM_EVENT_REJECTED;
1951                 event.param.conn.private_data = ib_event->private_data;
1952                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1953                 break;
1954         default:
1955                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1956                        ib_event->event);
1957                 goto out;
1958         }
1959
1960         ret = id_priv->id.event_handler(&id_priv->id, &event);
1961         if (ret) {
1962                 /* Destroy the CM ID by returning a non-zero value. */
1963                 id_priv->cm_id.ib = NULL;
1964                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1965                 mutex_unlock(&id_priv->handler_mutex);
1966                 rdma_destroy_id(&id_priv->id);
1967                 return ret;
1968         }
1969 out:
1970         mutex_unlock(&id_priv->handler_mutex);
1971         return ret;
1972 }
1973
1974 static struct rdma_id_private *
1975 cma_ib_new_conn_id(const struct rdma_cm_id *listen_id,
1976                    const struct ib_cm_event *ib_event,
1977                    struct net_device *net_dev)
1978 {
1979         struct rdma_id_private *listen_id_priv;
1980         struct rdma_id_private *id_priv;
1981         struct rdma_cm_id *id;
1982         struct rdma_route *rt;
1983         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1984         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1985         const __be64 service_id =
1986                 ib_event->param.req_rcvd.primary_path->service_id;
1987         int ret;
1988
1989         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1990         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1991                             listen_id->event_handler, listen_id->context,
1992                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1993                             listen_id_priv->res.kern_name);
1994         if (IS_ERR(id))
1995                 return NULL;
1996
1997         id_priv = container_of(id, struct rdma_id_private, id);
1998         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1999                               (struct sockaddr *)&id->route.addr.dst_addr,
2000                               listen_id, ib_event, ss_family, service_id))
2001                 goto err;
2002
2003         rt = &id->route;
2004         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
2005         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
2006                                      GFP_KERNEL);
2007         if (!rt->path_rec)
2008                 goto err;
2009
2010         rt->path_rec[0] = *path;
2011         if (rt->num_paths == 2)
2012                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
2013
2014         if (net_dev) {
2015                 rdma_copy_src_l2_addr(&rt->addr.dev_addr, net_dev);
2016         } else {
2017                 if (!cma_protocol_roce(listen_id) &&
2018                     cma_any_addr(cma_src_addr(id_priv))) {
2019                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
2020                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
2021                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
2022                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
2023                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
2024                         if (ret)
2025                                 goto err;
2026                 }
2027         }
2028         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
2029
2030         id_priv->state = RDMA_CM_CONNECT;
2031         return id_priv;
2032
2033 err:
2034         rdma_destroy_id(id);
2035         return NULL;
2036 }
2037
2038 static struct rdma_id_private *
2039 cma_ib_new_udp_id(const struct rdma_cm_id *listen_id,
2040                   const struct ib_cm_event *ib_event,
2041                   struct net_device *net_dev)
2042 {
2043         const struct rdma_id_private *listen_id_priv;
2044         struct rdma_id_private *id_priv;
2045         struct rdma_cm_id *id;
2046         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
2047         struct net *net = listen_id->route.addr.dev_addr.net;
2048         int ret;
2049
2050         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
2051         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
2052                               listen_id->ps, IB_QPT_UD,
2053                               listen_id_priv->res.kern_name);
2054         if (IS_ERR(id))
2055                 return NULL;
2056
2057         id_priv = container_of(id, struct rdma_id_private, id);
2058         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
2059                               (struct sockaddr *)&id->route.addr.dst_addr,
2060                               listen_id, ib_event, ss_family,
2061                               ib_event->param.sidr_req_rcvd.service_id))
2062                 goto err;
2063
2064         if (net_dev) {
2065                 rdma_copy_src_l2_addr(&id->route.addr.dev_addr, net_dev);
2066         } else {
2067                 if (!cma_any_addr(cma_src_addr(id_priv))) {
2068                         ret = cma_translate_addr(cma_src_addr(id_priv),
2069                                                  &id->route.addr.dev_addr);
2070                         if (ret)
2071                                 goto err;
2072                 }
2073         }
2074
2075         id_priv->state = RDMA_CM_CONNECT;
2076         return id_priv;
2077 err:
2078         rdma_destroy_id(id);
2079         return NULL;
2080 }
2081
2082 static void cma_set_req_event_data(struct rdma_cm_event *event,
2083                                    const struct ib_cm_req_event_param *req_data,
2084                                    void *private_data, int offset)
2085 {
2086         event->param.conn.private_data = private_data + offset;
2087         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
2088         event->param.conn.responder_resources = req_data->responder_resources;
2089         event->param.conn.initiator_depth = req_data->initiator_depth;
2090         event->param.conn.flow_control = req_data->flow_control;
2091         event->param.conn.retry_count = req_data->retry_count;
2092         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
2093         event->param.conn.srq = req_data->srq;
2094         event->param.conn.qp_num = req_data->remote_qpn;
2095 }
2096
2097 static int cma_ib_check_req_qp_type(const struct rdma_cm_id *id,
2098                                     const struct ib_cm_event *ib_event)
2099 {
2100         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
2101                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
2102                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
2103                  (id->qp_type == IB_QPT_UD)) ||
2104                 (!id->qp_type));
2105 }
2106
2107 static int cma_ib_req_handler(struct ib_cm_id *cm_id,
2108                               const struct ib_cm_event *ib_event)
2109 {
2110         struct rdma_id_private *listen_id, *conn_id = NULL;
2111         struct rdma_cm_event event = {};
2112         struct cma_req_info req = {};
2113         struct net_device *net_dev;
2114         u8 offset;
2115         int ret;
2116
2117         listen_id = cma_ib_id_from_event(cm_id, ib_event, &req, &net_dev);
2118         if (IS_ERR(listen_id))
2119                 return PTR_ERR(listen_id);
2120
2121         if (!cma_ib_check_req_qp_type(&listen_id->id, ib_event)) {
2122                 ret = -EINVAL;
2123                 goto net_dev_put;
2124         }
2125
2126         mutex_lock(&listen_id->handler_mutex);
2127         if (listen_id->state != RDMA_CM_LISTEN) {
2128                 ret = -ECONNABORTED;
2129                 goto err1;
2130         }
2131
2132         offset = cma_user_data_offset(listen_id);
2133         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2134         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
2135                 conn_id = cma_ib_new_udp_id(&listen_id->id, ib_event, net_dev);
2136                 event.param.ud.private_data = ib_event->private_data + offset;
2137                 event.param.ud.private_data_len =
2138                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
2139         } else {
2140                 conn_id = cma_ib_new_conn_id(&listen_id->id, ib_event, net_dev);
2141                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
2142                                        ib_event->private_data, offset);
2143         }
2144         if (!conn_id) {
2145                 ret = -ENOMEM;
2146                 goto err1;
2147         }
2148
2149         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2150         ret = cma_ib_acquire_dev(conn_id, listen_id, &req);
2151         if (ret)
2152                 goto err2;
2153
2154         conn_id->cm_id.ib = cm_id;
2155         cm_id->context = conn_id;
2156         cm_id->cm_handler = cma_ib_handler;
2157
2158         /*
2159          * Protect against the user destroying conn_id from another thread
2160          * until we're done accessing it.
2161          */
2162         atomic_inc(&conn_id->refcount);
2163         ret = conn_id->id.event_handler(&conn_id->id, &event);
2164         if (ret)
2165                 goto err3;
2166         /*
2167          * Acquire mutex to prevent user executing rdma_destroy_id()
2168          * while we're accessing the cm_id.
2169          */
2170         mutex_lock(&lock);
2171         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2172             (conn_id->id.qp_type != IB_QPT_UD))
2173                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2174         mutex_unlock(&lock);
2175         mutex_unlock(&conn_id->handler_mutex);
2176         mutex_unlock(&listen_id->handler_mutex);
2177         cma_deref_id(conn_id);
2178         if (net_dev)
2179                 dev_put(net_dev);
2180         return 0;
2181
2182 err3:
2183         cma_deref_id(conn_id);
2184         /* Destroy the CM ID by returning a non-zero value. */
2185         conn_id->cm_id.ib = NULL;
2186 err2:
2187         cma_exch(conn_id, RDMA_CM_DESTROYING);
2188         mutex_unlock(&conn_id->handler_mutex);
2189 err1:
2190         mutex_unlock(&listen_id->handler_mutex);
2191         if (conn_id)
2192                 rdma_destroy_id(&conn_id->id);
2193
2194 net_dev_put:
2195         if (net_dev)
2196                 dev_put(net_dev);
2197
2198         return ret;
2199 }
2200
2201 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2202 {
2203         if (addr->sa_family == AF_IB)
2204                 return ((struct sockaddr_ib *) addr)->sib_sid;
2205
2206         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2207 }
2208 EXPORT_SYMBOL(rdma_get_service_id);
2209
2210 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2211                     union ib_gid *dgid)
2212 {
2213         struct rdma_addr *addr = &cm_id->route.addr;
2214
2215         if (!cm_id->device) {
2216                 if (sgid)
2217                         memset(sgid, 0, sizeof(*sgid));
2218                 if (dgid)
2219                         memset(dgid, 0, sizeof(*dgid));
2220                 return;
2221         }
2222
2223         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2224                 if (sgid)
2225                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2226                 if (dgid)
2227                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2228         } else {
2229                 if (sgid)
2230                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2231                 if (dgid)
2232                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2233         }
2234 }
2235 EXPORT_SYMBOL(rdma_read_gids);
2236
2237 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2238 {
2239         struct rdma_id_private *id_priv = iw_id->context;
2240         struct rdma_cm_event event = {};
2241         int ret = 0;
2242         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2243         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2244
2245         mutex_lock(&id_priv->handler_mutex);
2246         if (id_priv->state != RDMA_CM_CONNECT)
2247                 goto out;
2248
2249         switch (iw_event->event) {
2250         case IW_CM_EVENT_CLOSE:
2251                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2252                 break;
2253         case IW_CM_EVENT_CONNECT_REPLY:
2254                 memcpy(cma_src_addr(id_priv), laddr,
2255                        rdma_addr_size(laddr));
2256                 memcpy(cma_dst_addr(id_priv), raddr,
2257                        rdma_addr_size(raddr));
2258                 switch (iw_event->status) {
2259                 case 0:
2260                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2261                         event.param.conn.initiator_depth = iw_event->ird;
2262                         event.param.conn.responder_resources = iw_event->ord;
2263                         break;
2264                 case -ECONNRESET:
2265                 case -ECONNREFUSED:
2266                         event.event = RDMA_CM_EVENT_REJECTED;
2267                         break;
2268                 case -ETIMEDOUT:
2269                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2270                         break;
2271                 default:
2272                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2273                         break;
2274                 }
2275                 break;
2276         case IW_CM_EVENT_ESTABLISHED:
2277                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2278                 event.param.conn.initiator_depth = iw_event->ird;
2279                 event.param.conn.responder_resources = iw_event->ord;
2280                 break;
2281         default:
2282                 goto out;
2283         }
2284
2285         event.status = iw_event->status;
2286         event.param.conn.private_data = iw_event->private_data;
2287         event.param.conn.private_data_len = iw_event->private_data_len;
2288         ret = id_priv->id.event_handler(&id_priv->id, &event);
2289         if (ret) {
2290                 /* Destroy the CM ID by returning a non-zero value. */
2291                 id_priv->cm_id.iw = NULL;
2292                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2293                 mutex_unlock(&id_priv->handler_mutex);
2294                 rdma_destroy_id(&id_priv->id);
2295                 return ret;
2296         }
2297
2298 out:
2299         mutex_unlock(&id_priv->handler_mutex);
2300         return ret;
2301 }
2302
2303 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2304                                struct iw_cm_event *iw_event)
2305 {
2306         struct rdma_cm_id *new_cm_id;
2307         struct rdma_id_private *listen_id, *conn_id;
2308         struct rdma_cm_event event = {};
2309         int ret = -ECONNABORTED;
2310         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2311         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2312
2313         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2314         event.param.conn.private_data = iw_event->private_data;
2315         event.param.conn.private_data_len = iw_event->private_data_len;
2316         event.param.conn.initiator_depth = iw_event->ird;
2317         event.param.conn.responder_resources = iw_event->ord;
2318
2319         listen_id = cm_id->context;
2320
2321         mutex_lock(&listen_id->handler_mutex);
2322         if (listen_id->state != RDMA_CM_LISTEN)
2323                 goto out;
2324
2325         /* Create a new RDMA id for the new IW CM ID */
2326         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2327                                      listen_id->id.event_handler,
2328                                      listen_id->id.context,
2329                                      RDMA_PS_TCP, IB_QPT_RC,
2330                                      listen_id->res.kern_name);
2331         if (IS_ERR(new_cm_id)) {
2332                 ret = -ENOMEM;
2333                 goto out;
2334         }
2335         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2336         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2337         conn_id->state = RDMA_CM_CONNECT;
2338
2339         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2340         if (ret) {
2341                 mutex_unlock(&conn_id->handler_mutex);
2342                 rdma_destroy_id(new_cm_id);
2343                 goto out;
2344         }
2345
2346         ret = cma_iw_acquire_dev(conn_id, listen_id);
2347         if (ret) {
2348                 mutex_unlock(&conn_id->handler_mutex);
2349                 rdma_destroy_id(new_cm_id);
2350                 goto out;
2351         }
2352
2353         conn_id->cm_id.iw = cm_id;
2354         cm_id->context = conn_id;
2355         cm_id->cm_handler = cma_iw_handler;
2356
2357         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2358         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2359
2360         /*
2361          * Protect against the user destroying conn_id from another thread
2362          * until we're done accessing it.
2363          */
2364         atomic_inc(&conn_id->refcount);
2365         ret = conn_id->id.event_handler(&conn_id->id, &event);
2366         if (ret) {
2367                 /* User wants to destroy the CM ID */
2368                 conn_id->cm_id.iw = NULL;
2369                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2370                 mutex_unlock(&conn_id->handler_mutex);
2371                 cma_deref_id(conn_id);
2372                 rdma_destroy_id(&conn_id->id);
2373                 goto out;
2374         }
2375
2376         mutex_unlock(&conn_id->handler_mutex);
2377         cma_deref_id(conn_id);
2378
2379 out:
2380         mutex_unlock(&listen_id->handler_mutex);
2381         return ret;
2382 }
2383
2384 static int cma_ib_listen(struct rdma_id_private *id_priv)
2385 {
2386         struct sockaddr *addr;
2387         struct ib_cm_id *id;
2388         __be64 svc_id;
2389
2390         addr = cma_src_addr(id_priv);
2391         svc_id = rdma_get_service_id(&id_priv->id, addr);
2392         id = ib_cm_insert_listen(id_priv->id.device,
2393                                  cma_ib_req_handler, svc_id);
2394         if (IS_ERR(id))
2395                 return PTR_ERR(id);
2396         id_priv->cm_id.ib = id;
2397
2398         return 0;
2399 }
2400
2401 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2402 {
2403         int ret;
2404         struct iw_cm_id *id;
2405
2406         id = iw_create_cm_id(id_priv->id.device,
2407                              iw_conn_req_handler,
2408                              id_priv);
2409         if (IS_ERR(id))
2410                 return PTR_ERR(id);
2411
2412         id->tos = id_priv->tos;
2413         id_priv->cm_id.iw = id;
2414
2415         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2416                rdma_addr_size(cma_src_addr(id_priv)));
2417
2418         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2419
2420         if (ret) {
2421                 iw_destroy_cm_id(id_priv->cm_id.iw);
2422                 id_priv->cm_id.iw = NULL;
2423         }
2424
2425         return ret;
2426 }
2427
2428 static int cma_listen_handler(struct rdma_cm_id *id,
2429                               struct rdma_cm_event *event)
2430 {
2431         struct rdma_id_private *id_priv = id->context;
2432
2433         id->context = id_priv->id.context;
2434         id->event_handler = id_priv->id.event_handler;
2435         return id_priv->id.event_handler(id, event);
2436 }
2437
2438 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2439                               struct cma_device *cma_dev)
2440 {
2441         struct rdma_id_private *dev_id_priv;
2442         struct rdma_cm_id *id;
2443         struct net *net = id_priv->id.route.addr.dev_addr.net;
2444         int ret;
2445
2446         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2447                 return;
2448
2449         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2450                               id_priv->id.qp_type, id_priv->res.kern_name);
2451         if (IS_ERR(id))
2452                 return;
2453
2454         dev_id_priv = container_of(id, struct rdma_id_private, id);
2455
2456         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2457         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2458                rdma_addr_size(cma_src_addr(id_priv)));
2459
2460         _cma_attach_to_dev(dev_id_priv, cma_dev);
2461         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2462         atomic_inc(&id_priv->refcount);
2463         dev_id_priv->internal_id = 1;
2464         dev_id_priv->afonly = id_priv->afonly;
2465
2466         ret = rdma_listen(id, id_priv->backlog);
2467         if (ret)
2468                 dev_warn(&cma_dev->device->dev,
2469                          "RDMA CMA: cma_listen_on_dev, error %d\n", ret);
2470 }
2471
2472 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2473 {
2474         struct cma_device *cma_dev;
2475
2476         mutex_lock(&lock);
2477         list_add_tail(&id_priv->list, &listen_any_list);
2478         list_for_each_entry(cma_dev, &dev_list, list)
2479                 cma_listen_on_dev(id_priv, cma_dev);
2480         mutex_unlock(&lock);
2481 }
2482
2483 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2484 {
2485         struct rdma_id_private *id_priv;
2486
2487         id_priv = container_of(id, struct rdma_id_private, id);
2488         id_priv->tos = (u8) tos;
2489         id_priv->tos_set = true;
2490 }
2491 EXPORT_SYMBOL(rdma_set_service_type);
2492
2493 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2494                               void *context)
2495 {
2496         struct cma_work *work = context;
2497         struct rdma_route *route;
2498
2499         route = &work->id->id.route;
2500
2501         if (!status) {
2502                 route->num_paths = 1;
2503                 *route->path_rec = *path_rec;
2504         } else {
2505                 work->old_state = RDMA_CM_ROUTE_QUERY;
2506                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2507                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2508                 work->event.status = status;
2509                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2510                                      status);
2511         }
2512
2513         queue_work(cma_wq, &work->work);
2514 }
2515
2516 static int cma_query_ib_route(struct rdma_id_private *id_priv,
2517                               unsigned long timeout_ms, struct cma_work *work)
2518 {
2519         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2520         struct sa_path_rec path_rec;
2521         ib_sa_comp_mask comp_mask;
2522         struct sockaddr_in6 *sin6;
2523         struct sockaddr_ib *sib;
2524
2525         memset(&path_rec, 0, sizeof path_rec);
2526
2527         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2528                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2529         else
2530                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2531         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2532         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2533         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2534         path_rec.numb_path = 1;
2535         path_rec.reversible = 1;
2536         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2537                                                   cma_dst_addr(id_priv));
2538
2539         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2540                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2541                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2542
2543         switch (cma_family(id_priv)) {
2544         case AF_INET:
2545                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2546                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2547                 break;
2548         case AF_INET6:
2549                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2550                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2551                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2552                 break;
2553         case AF_IB:
2554                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2555                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2556                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2557                 break;
2558         }
2559
2560         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2561                                                id_priv->id.port_num, &path_rec,
2562                                                comp_mask, timeout_ms,
2563                                                GFP_KERNEL, cma_query_handler,
2564                                                work, &id_priv->query);
2565
2566         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2567 }
2568
2569 static void cma_work_handler(struct work_struct *_work)
2570 {
2571         struct cma_work *work = container_of(_work, struct cma_work, work);
2572         struct rdma_id_private *id_priv = work->id;
2573         int destroy = 0;
2574
2575         mutex_lock(&id_priv->handler_mutex);
2576         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2577                 goto out;
2578
2579         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2580                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2581                 destroy = 1;
2582         }
2583 out:
2584         mutex_unlock(&id_priv->handler_mutex);
2585         cma_deref_id(id_priv);
2586         if (destroy)
2587                 rdma_destroy_id(&id_priv->id);
2588         kfree(work);
2589 }
2590
2591 static void cma_ndev_work_handler(struct work_struct *_work)
2592 {
2593         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2594         struct rdma_id_private *id_priv = work->id;
2595         int destroy = 0;
2596
2597         mutex_lock(&id_priv->handler_mutex);
2598         if (id_priv->state == RDMA_CM_DESTROYING ||
2599             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2600                 goto out;
2601
2602         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2603                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2604                 destroy = 1;
2605         }
2606
2607 out:
2608         mutex_unlock(&id_priv->handler_mutex);
2609         cma_deref_id(id_priv);
2610         if (destroy)
2611                 rdma_destroy_id(&id_priv->id);
2612         kfree(work);
2613 }
2614
2615 static void cma_init_resolve_route_work(struct cma_work *work,
2616                                         struct rdma_id_private *id_priv)
2617 {
2618         work->id = id_priv;
2619         INIT_WORK(&work->work, cma_work_handler);
2620         work->old_state = RDMA_CM_ROUTE_QUERY;
2621         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2622         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2623 }
2624
2625 static void cma_init_resolve_addr_work(struct cma_work *work,
2626                                        struct rdma_id_private *id_priv)
2627 {
2628         work->id = id_priv;
2629         INIT_WORK(&work->work, cma_work_handler);
2630         work->old_state = RDMA_CM_ADDR_QUERY;
2631         work->new_state = RDMA_CM_ADDR_RESOLVED;
2632         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2633 }
2634
2635 static int cma_resolve_ib_route(struct rdma_id_private *id_priv,
2636                                 unsigned long timeout_ms)
2637 {
2638         struct rdma_route *route = &id_priv->id.route;
2639         struct cma_work *work;
2640         int ret;
2641
2642         work = kzalloc(sizeof *work, GFP_KERNEL);
2643         if (!work)
2644                 return -ENOMEM;
2645
2646         cma_init_resolve_route_work(work, id_priv);
2647
2648         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2649         if (!route->path_rec) {
2650                 ret = -ENOMEM;
2651                 goto err1;
2652         }
2653
2654         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2655         if (ret)
2656                 goto err2;
2657
2658         return 0;
2659 err2:
2660         kfree(route->path_rec);
2661         route->path_rec = NULL;
2662 err1:
2663         kfree(work);
2664         return ret;
2665 }
2666
2667 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2668                                            unsigned long supported_gids,
2669                                            enum ib_gid_type default_gid)
2670 {
2671         if ((network_type == RDMA_NETWORK_IPV4 ||
2672              network_type == RDMA_NETWORK_IPV6) &&
2673             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2674                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2675
2676         return default_gid;
2677 }
2678
2679 /*
2680  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2681  * path record type based on GID type.
2682  * It also sets up other L2 fields which includes destination mac address
2683  * netdev ifindex, of the path record.
2684  * It returns the netdev of the bound interface for this path record entry.
2685  */
2686 static struct net_device *
2687 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2688 {
2689         struct rdma_route *route = &id_priv->id.route;
2690         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2691         struct rdma_addr *addr = &route->addr;
2692         unsigned long supported_gids;
2693         struct net_device *ndev;
2694
2695         if (!addr->dev_addr.bound_dev_if)
2696                 return NULL;
2697
2698         ndev = dev_get_by_index(addr->dev_addr.net,
2699                                 addr->dev_addr.bound_dev_if);
2700         if (!ndev)
2701                 return NULL;
2702
2703         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2704                                                     id_priv->id.port_num);
2705         gid_type = cma_route_gid_type(addr->dev_addr.network,
2706                                       supported_gids,
2707                                       id_priv->gid_type);
2708         /* Use the hint from IP Stack to select GID Type */
2709         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2710                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2711         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2712
2713         route->path_rec->roce.route_resolved = true;
2714         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2715         return ndev;
2716 }
2717
2718 int rdma_set_ib_path(struct rdma_cm_id *id,
2719                      struct sa_path_rec *path_rec)
2720 {
2721         struct rdma_id_private *id_priv;
2722         struct net_device *ndev;
2723         int ret;
2724
2725         id_priv = container_of(id, struct rdma_id_private, id);
2726         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2727                            RDMA_CM_ROUTE_RESOLVED))
2728                 return -EINVAL;
2729
2730         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2731                                      GFP_KERNEL);
2732         if (!id->route.path_rec) {
2733                 ret = -ENOMEM;
2734                 goto err;
2735         }
2736
2737         if (rdma_protocol_roce(id->device, id->port_num)) {
2738                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2739                 if (!ndev) {
2740                         ret = -ENODEV;
2741                         goto err_free;
2742                 }
2743                 dev_put(ndev);
2744         }
2745
2746         id->route.num_paths = 1;
2747         return 0;
2748
2749 err_free:
2750         kfree(id->route.path_rec);
2751         id->route.path_rec = NULL;
2752 err:
2753         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2754         return ret;
2755 }
2756 EXPORT_SYMBOL(rdma_set_ib_path);
2757
2758 static int cma_resolve_iw_route(struct rdma_id_private *id_priv)
2759 {
2760         struct cma_work *work;
2761
2762         work = kzalloc(sizeof *work, GFP_KERNEL);
2763         if (!work)
2764                 return -ENOMEM;
2765
2766         cma_init_resolve_route_work(work, id_priv);
2767         queue_work(cma_wq, &work->work);
2768         return 0;
2769 }
2770
2771 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2772 {
2773         int prio;
2774         struct net_device *dev;
2775
2776         prio = rt_tos2priority(tos);
2777         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2778         if (dev->num_tc)
2779                 return netdev_get_prio_tc_map(dev, prio);
2780
2781 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2782         if (is_vlan_dev(ndev))
2783                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2784                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2785 #endif
2786         return 0;
2787 }
2788
2789 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2790 {
2791         struct rdma_route *route = &id_priv->id.route;
2792         struct rdma_addr *addr = &route->addr;
2793         struct cma_work *work;
2794         int ret;
2795         struct net_device *ndev;
2796
2797         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2798                                         rdma_start_port(id_priv->cma_dev->device)];
2799         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2800
2801
2802         work = kzalloc(sizeof *work, GFP_KERNEL);
2803         if (!work)
2804                 return -ENOMEM;
2805
2806         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2807         if (!route->path_rec) {
2808                 ret = -ENOMEM;
2809                 goto err1;
2810         }
2811
2812         route->num_paths = 1;
2813
2814         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2815         if (!ndev) {
2816                 ret = -ENODEV;
2817                 goto err2;
2818         }
2819
2820         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2821                     &route->path_rec->sgid);
2822         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2823                     &route->path_rec->dgid);
2824
2825         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2826                 /* TODO: get the hoplimit from the inet/inet6 device */
2827                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2828         else
2829                 route->path_rec->hop_limit = 1;
2830         route->path_rec->reversible = 1;
2831         route->path_rec->pkey = cpu_to_be16(0xffff);
2832         route->path_rec->mtu_selector = IB_SA_EQ;
2833         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2834         route->path_rec->traffic_class = tos;
2835         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2836         route->path_rec->rate_selector = IB_SA_EQ;
2837         route->path_rec->rate = iboe_get_rate(ndev);
2838         dev_put(ndev);
2839         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2840         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2841         if (!route->path_rec->mtu) {
2842                 ret = -EINVAL;
2843                 goto err2;
2844         }
2845
2846         cma_init_resolve_route_work(work, id_priv);
2847         queue_work(cma_wq, &work->work);
2848
2849         return 0;
2850
2851 err2:
2852         kfree(route->path_rec);
2853         route->path_rec = NULL;
2854 err1:
2855         kfree(work);
2856         return ret;
2857 }
2858
2859 int rdma_resolve_route(struct rdma_cm_id *id, unsigned long timeout_ms)
2860 {
2861         struct rdma_id_private *id_priv;
2862         int ret;
2863
2864         id_priv = container_of(id, struct rdma_id_private, id);
2865         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2866                 return -EINVAL;
2867
2868         atomic_inc(&id_priv->refcount);
2869         if (rdma_cap_ib_sa(id->device, id->port_num))
2870                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2871         else if (rdma_protocol_roce(id->device, id->port_num))
2872                 ret = cma_resolve_iboe_route(id_priv);
2873         else if (rdma_protocol_iwarp(id->device, id->port_num))
2874                 ret = cma_resolve_iw_route(id_priv);
2875         else
2876                 ret = -ENOSYS;
2877
2878         if (ret)
2879                 goto err;
2880
2881         return 0;
2882 err:
2883         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2884         cma_deref_id(id_priv);
2885         return ret;
2886 }
2887 EXPORT_SYMBOL(rdma_resolve_route);
2888
2889 static void cma_set_loopback(struct sockaddr *addr)
2890 {
2891         switch (addr->sa_family) {
2892         case AF_INET:
2893                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2894                 break;
2895         case AF_INET6:
2896                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2897                               0, 0, 0, htonl(1));
2898                 break;
2899         default:
2900                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2901                             0, 0, 0, htonl(1));
2902                 break;
2903         }
2904 }
2905
2906 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2907 {
2908         struct cma_device *cma_dev, *cur_dev;
2909         union ib_gid gid;
2910         enum ib_port_state port_state;
2911         u16 pkey;
2912         int ret;
2913         u8 p;
2914
2915         cma_dev = NULL;
2916         mutex_lock(&lock);
2917         list_for_each_entry(cur_dev, &dev_list, list) {
2918                 if (cma_family(id_priv) == AF_IB &&
2919                     !rdma_cap_ib_cm(cur_dev->device, 1))
2920                         continue;
2921
2922                 if (!cma_dev)
2923                         cma_dev = cur_dev;
2924
2925                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2926                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2927                             port_state == IB_PORT_ACTIVE) {
2928                                 cma_dev = cur_dev;
2929                                 goto port_found;
2930                         }
2931                 }
2932         }
2933
2934         if (!cma_dev) {
2935                 ret = -ENODEV;
2936                 goto out;
2937         }
2938
2939         p = 1;
2940
2941 port_found:
2942         ret = rdma_query_gid(cma_dev->device, p, 0, &gid);
2943         if (ret)
2944                 goto out;
2945
2946         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2947         if (ret)
2948                 goto out;
2949
2950         id_priv->id.route.addr.dev_addr.dev_type =
2951                 (rdma_protocol_ib(cma_dev->device, p)) ?
2952                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2953
2954         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2955         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2956         id_priv->id.port_num = p;
2957         cma_attach_to_dev(id_priv, cma_dev);
2958         cma_set_loopback(cma_src_addr(id_priv));
2959 out:
2960         mutex_unlock(&lock);
2961         return ret;
2962 }
2963
2964 static void addr_handler(int status, struct sockaddr *src_addr,
2965                          struct rdma_dev_addr *dev_addr, void *context)
2966 {
2967         struct rdma_id_private *id_priv = context;
2968         struct rdma_cm_event event = {};
2969
2970         mutex_lock(&id_priv->handler_mutex);
2971         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2972                            RDMA_CM_ADDR_RESOLVED))
2973                 goto out;
2974
2975         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2976         if (!status && !id_priv->cma_dev) {
2977                 status = cma_acquire_dev_by_src_ip(id_priv);
2978                 if (status)
2979                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2980                                              status);
2981         } else {
2982                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2983         }
2984
2985         if (status) {
2986                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2987                                    RDMA_CM_ADDR_BOUND))
2988                         goto out;
2989                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2990                 event.status = status;
2991         } else
2992                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2993
2994         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2995                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2996                 mutex_unlock(&id_priv->handler_mutex);
2997                 rdma_destroy_id(&id_priv->id);
2998                 return;
2999         }
3000 out:
3001         mutex_unlock(&id_priv->handler_mutex);
3002 }
3003
3004 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
3005 {
3006         struct cma_work *work;
3007         union ib_gid gid;
3008         int ret;
3009
3010         work = kzalloc(sizeof *work, GFP_KERNEL);
3011         if (!work)
3012                 return -ENOMEM;
3013
3014         if (!id_priv->cma_dev) {
3015                 ret = cma_bind_loopback(id_priv);
3016                 if (ret)
3017                         goto err;
3018         }
3019
3020         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
3021         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
3022
3023         cma_init_resolve_addr_work(work, id_priv);
3024         queue_work(cma_wq, &work->work);
3025         return 0;
3026 err:
3027         kfree(work);
3028         return ret;
3029 }
3030
3031 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
3032 {
3033         struct cma_work *work;
3034         int ret;
3035
3036         work = kzalloc(sizeof *work, GFP_KERNEL);
3037         if (!work)
3038                 return -ENOMEM;
3039
3040         if (!id_priv->cma_dev) {
3041                 ret = cma_resolve_ib_dev(id_priv);
3042                 if (ret)
3043                         goto err;
3044         }
3045
3046         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
3047                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
3048
3049         cma_init_resolve_addr_work(work, id_priv);
3050         queue_work(cma_wq, &work->work);
3051         return 0;
3052 err:
3053         kfree(work);
3054         return ret;
3055 }
3056
3057 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
3058                          const struct sockaddr *dst_addr)
3059 {
3060         if (!src_addr || !src_addr->sa_family) {
3061                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
3062                 src_addr->sa_family = dst_addr->sa_family;
3063                 if (IS_ENABLED(CONFIG_IPV6) &&
3064                     dst_addr->sa_family == AF_INET6) {
3065                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
3066                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
3067                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
3068                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
3069                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
3070                 } else if (dst_addr->sa_family == AF_IB) {
3071                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
3072                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
3073                 }
3074         }
3075         return rdma_bind_addr(id, src_addr);
3076 }
3077
3078 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
3079                       const struct sockaddr *dst_addr, unsigned long timeout_ms)
3080 {
3081         struct rdma_id_private *id_priv;
3082         int ret;
3083
3084         id_priv = container_of(id, struct rdma_id_private, id);
3085         if (id_priv->state == RDMA_CM_IDLE) {
3086                 ret = cma_bind_addr(id, src_addr, dst_addr);
3087                 if (ret)
3088                         return ret;
3089         }
3090
3091         if (cma_family(id_priv) != dst_addr->sa_family)
3092                 return -EINVAL;
3093
3094         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
3095                 return -EINVAL;
3096
3097         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
3098         if (cma_any_addr(dst_addr)) {
3099                 ret = cma_resolve_loopback(id_priv);
3100         } else {
3101                 if (dst_addr->sa_family == AF_IB) {
3102                         ret = cma_resolve_ib_addr(id_priv);
3103                 } else {
3104                         ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
3105                                               &id->route.addr.dev_addr,
3106                                               timeout_ms, addr_handler,
3107                                               false, id_priv);
3108                 }
3109         }
3110         if (ret)
3111                 goto err;
3112
3113         return 0;
3114 err:
3115         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
3116         return ret;
3117 }
3118 EXPORT_SYMBOL(rdma_resolve_addr);
3119
3120 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
3121 {
3122         struct rdma_id_private *id_priv;
3123         unsigned long flags;
3124         int ret;
3125
3126         id_priv = container_of(id, struct rdma_id_private, id);
3127         spin_lock_irqsave(&id_priv->lock, flags);
3128         if (reuse || id_priv->state == RDMA_CM_IDLE) {
3129                 id_priv->reuseaddr = reuse;
3130                 ret = 0;
3131         } else {
3132                 ret = -EINVAL;
3133         }
3134         spin_unlock_irqrestore(&id_priv->lock, flags);
3135         return ret;
3136 }
3137 EXPORT_SYMBOL(rdma_set_reuseaddr);
3138
3139 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3140 {
3141         struct rdma_id_private *id_priv;
3142         unsigned long flags;
3143         int ret;
3144
3145         id_priv = container_of(id, struct rdma_id_private, id);
3146         spin_lock_irqsave(&id_priv->lock, flags);
3147         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3148                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3149                 id_priv->afonly = afonly;
3150                 ret = 0;
3151         } else {
3152                 ret = -EINVAL;
3153         }
3154         spin_unlock_irqrestore(&id_priv->lock, flags);
3155         return ret;
3156 }
3157 EXPORT_SYMBOL(rdma_set_afonly);
3158
3159 static void cma_bind_port(struct rdma_bind_list *bind_list,
3160                           struct rdma_id_private *id_priv)
3161 {
3162         struct sockaddr *addr;
3163         struct sockaddr_ib *sib;
3164         u64 sid, mask;
3165         __be16 port;
3166
3167         addr = cma_src_addr(id_priv);
3168         port = htons(bind_list->port);
3169
3170         switch (addr->sa_family) {
3171         case AF_INET:
3172                 ((struct sockaddr_in *) addr)->sin_port = port;
3173                 break;
3174         case AF_INET6:
3175                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3176                 break;
3177         case AF_IB:
3178                 sib = (struct sockaddr_ib *) addr;
3179                 sid = be64_to_cpu(sib->sib_sid);
3180                 mask = be64_to_cpu(sib->sib_sid_mask);
3181                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3182                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3183                 break;
3184         }
3185         id_priv->bind_list = bind_list;
3186         hlist_add_head(&id_priv->node, &bind_list->owners);
3187 }
3188
3189 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3190                           struct rdma_id_private *id_priv, unsigned short snum)
3191 {
3192         struct rdma_bind_list *bind_list;
3193         int ret;
3194
3195         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3196         if (!bind_list)
3197                 return -ENOMEM;
3198
3199         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3200                            snum);
3201         if (ret < 0)
3202                 goto err;
3203
3204         bind_list->ps = ps;
3205         bind_list->port = (unsigned short)ret;
3206         cma_bind_port(bind_list, id_priv);
3207         return 0;
3208 err:
3209         kfree(bind_list);
3210         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3211 }
3212
3213 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3214                               struct rdma_id_private *id_priv)
3215 {
3216         struct rdma_id_private *cur_id;
3217         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3218         struct sockaddr  *saddr = cma_src_addr(id_priv);
3219         __be16 dport = cma_port(daddr);
3220
3221         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3222                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3223                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3224                 __be16 cur_dport = cma_port(cur_daddr);
3225
3226                 if (id_priv == cur_id)
3227                         continue;
3228
3229                 /* different dest port -> unique */
3230                 if (!cma_any_port(daddr) &&
3231                     !cma_any_port(cur_daddr) &&
3232                     (dport != cur_dport))
3233                         continue;
3234
3235                 /* different src address -> unique */
3236                 if (!cma_any_addr(saddr) &&
3237                     !cma_any_addr(cur_saddr) &&
3238                     cma_addr_cmp(saddr, cur_saddr))
3239                         continue;
3240
3241                 /* different dst address -> unique */
3242                 if (!cma_any_addr(daddr) &&
3243                     !cma_any_addr(cur_daddr) &&
3244                     cma_addr_cmp(daddr, cur_daddr))
3245                         continue;
3246
3247                 return -EADDRNOTAVAIL;
3248         }
3249         return 0;
3250 }
3251
3252 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3253                               struct rdma_id_private *id_priv)
3254 {
3255         static unsigned int last_used_port;
3256         int low, high, remaining;
3257         unsigned int rover;
3258         struct net *net = id_priv->id.route.addr.dev_addr.net;
3259
3260         inet_get_local_port_range(net, &low, &high);
3261         remaining = (high - low) + 1;
3262         rover = prandom_u32() % remaining + low;
3263 retry:
3264         if (last_used_port != rover) {
3265                 struct rdma_bind_list *bind_list;
3266                 int ret;
3267
3268                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3269
3270                 if (!bind_list) {
3271                         ret = cma_alloc_port(ps, id_priv, rover);
3272                 } else {
3273                         ret = cma_port_is_unique(bind_list, id_priv);
3274                         if (!ret)
3275                                 cma_bind_port(bind_list, id_priv);
3276                 }
3277                 /*
3278                  * Remember previously used port number in order to avoid
3279                  * re-using same port immediately after it is closed.
3280                  */
3281                 if (!ret)
3282                         last_used_port = rover;
3283                 if (ret != -EADDRNOTAVAIL)
3284                         return ret;
3285         }
3286         if (--remaining) {
3287                 rover++;
3288                 if ((rover < low) || (rover > high))
3289                         rover = low;
3290                 goto retry;
3291         }
3292         return -EADDRNOTAVAIL;
3293 }
3294
3295 /*
3296  * Check that the requested port is available.  This is called when trying to
3297  * bind to a specific port, or when trying to listen on a bound port.  In
3298  * the latter case, the provided id_priv may already be on the bind_list, but
3299  * we still need to check that it's okay to start listening.
3300  */
3301 static int cma_check_port(struct rdma_bind_list *bind_list,
3302                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3303 {
3304         struct rdma_id_private *cur_id;
3305         struct sockaddr *addr, *cur_addr;
3306
3307         addr = cma_src_addr(id_priv);
3308         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3309                 if (id_priv == cur_id)
3310                         continue;
3311
3312                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3313                     cur_id->reuseaddr)
3314                         continue;
3315
3316                 cur_addr = cma_src_addr(cur_id);
3317                 if (id_priv->afonly && cur_id->afonly &&
3318                     (addr->sa_family != cur_addr->sa_family))
3319                         continue;
3320
3321                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3322                         return -EADDRNOTAVAIL;
3323
3324                 if (!cma_addr_cmp(addr, cur_addr))
3325                         return -EADDRINUSE;
3326         }
3327         return 0;
3328 }
3329
3330 static int cma_use_port(enum rdma_ucm_port_space ps,
3331                         struct rdma_id_private *id_priv)
3332 {
3333         struct rdma_bind_list *bind_list;
3334         unsigned short snum;
3335         int ret;
3336
3337         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3338         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3339                 return -EACCES;
3340
3341         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3342         if (!bind_list) {
3343                 ret = cma_alloc_port(ps, id_priv, snum);
3344         } else {
3345                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3346                 if (!ret)
3347                         cma_bind_port(bind_list, id_priv);
3348         }
3349         return ret;
3350 }
3351
3352 static int cma_bind_listen(struct rdma_id_private *id_priv)
3353 {
3354         struct rdma_bind_list *bind_list = id_priv->bind_list;
3355         int ret = 0;
3356
3357         mutex_lock(&lock);
3358         if (bind_list->owners.first->next)
3359                 ret = cma_check_port(bind_list, id_priv, 0);
3360         mutex_unlock(&lock);
3361         return ret;
3362 }
3363
3364 static enum rdma_ucm_port_space
3365 cma_select_inet_ps(struct rdma_id_private *id_priv)
3366 {
3367         switch (id_priv->id.ps) {
3368         case RDMA_PS_TCP:
3369         case RDMA_PS_UDP:
3370         case RDMA_PS_IPOIB:
3371         case RDMA_PS_IB:
3372                 return id_priv->id.ps;
3373         default:
3374
3375                 return 0;
3376         }
3377 }
3378
3379 static enum rdma_ucm_port_space
3380 cma_select_ib_ps(struct rdma_id_private *id_priv)
3381 {
3382         enum rdma_ucm_port_space ps = 0;
3383         struct sockaddr_ib *sib;
3384         u64 sid_ps, mask, sid;
3385
3386         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3387         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3388         sid = be64_to_cpu(sib->sib_sid) & mask;
3389
3390         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3391                 sid_ps = RDMA_IB_IP_PS_IB;
3392                 ps = RDMA_PS_IB;
3393         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3394                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3395                 sid_ps = RDMA_IB_IP_PS_TCP;
3396                 ps = RDMA_PS_TCP;
3397         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3398                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3399                 sid_ps = RDMA_IB_IP_PS_UDP;
3400                 ps = RDMA_PS_UDP;
3401         }
3402
3403         if (ps) {
3404                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3405                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3406                                                 be64_to_cpu(sib->sib_sid_mask));
3407         }
3408         return ps;
3409 }
3410
3411 static int cma_get_port(struct rdma_id_private *id_priv)
3412 {
3413         enum rdma_ucm_port_space ps;
3414         int ret;
3415
3416         if (cma_family(id_priv) != AF_IB)
3417                 ps = cma_select_inet_ps(id_priv);
3418         else
3419                 ps = cma_select_ib_ps(id_priv);
3420         if (!ps)
3421                 return -EPROTONOSUPPORT;
3422
3423         mutex_lock(&lock);
3424         if (cma_any_port(cma_src_addr(id_priv)))
3425                 ret = cma_alloc_any_port(ps, id_priv);
3426         else
3427                 ret = cma_use_port(ps, id_priv);
3428         mutex_unlock(&lock);
3429
3430         return ret;
3431 }
3432
3433 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3434                                struct sockaddr *addr)
3435 {
3436 #if IS_ENABLED(CONFIG_IPV6)
3437         struct sockaddr_in6 *sin6;
3438
3439         if (addr->sa_family != AF_INET6)
3440                 return 0;
3441
3442         sin6 = (struct sockaddr_in6 *) addr;
3443
3444         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3445                 return 0;
3446
3447         if (!sin6->sin6_scope_id)
3448                         return -EINVAL;
3449
3450         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3451 #endif
3452         return 0;
3453 }
3454
3455 int rdma_listen(struct rdma_cm_id *id, int backlog)
3456 {
3457         struct rdma_id_private *id_priv;
3458         int ret;
3459
3460         id_priv = container_of(id, struct rdma_id_private, id);
3461         if (id_priv->state == RDMA_CM_IDLE) {
3462                 id->route.addr.src_addr.ss_family = AF_INET;
3463                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3464                 if (ret)
3465                         return ret;
3466         }
3467
3468         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3469                 return -EINVAL;
3470
3471         if (id_priv->reuseaddr) {
3472                 ret = cma_bind_listen(id_priv);
3473                 if (ret)
3474                         goto err;
3475         }
3476
3477         id_priv->backlog = backlog;
3478         if (id->device) {
3479                 if (rdma_cap_ib_cm(id->device, 1)) {
3480                         ret = cma_ib_listen(id_priv);
3481                         if (ret)
3482                                 goto err;
3483                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3484                         ret = cma_iw_listen(id_priv, backlog);
3485                         if (ret)
3486                                 goto err;
3487                 } else {
3488                         ret = -ENOSYS;
3489                         goto err;
3490                 }
3491         } else
3492                 cma_listen_on_all(id_priv);
3493
3494         return 0;
3495 err:
3496         id_priv->backlog = 0;
3497         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3498         return ret;
3499 }
3500 EXPORT_SYMBOL(rdma_listen);
3501
3502 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3503 {
3504         struct rdma_id_private *id_priv;
3505         int ret;
3506         struct sockaddr  *daddr;
3507
3508         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3509             addr->sa_family != AF_IB)
3510                 return -EAFNOSUPPORT;
3511
3512         id_priv = container_of(id, struct rdma_id_private, id);
3513         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3514                 return -EINVAL;
3515
3516         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3517         if (ret)
3518                 goto err1;
3519
3520         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3521         if (!cma_any_addr(addr)) {
3522                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3523                 if (ret)
3524                         goto err1;
3525
3526                 ret = cma_acquire_dev_by_src_ip(id_priv);
3527                 if (ret)
3528                         goto err1;
3529         }
3530
3531         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3532                 if (addr->sa_family == AF_INET)
3533                         id_priv->afonly = 1;
3534 #if IS_ENABLED(CONFIG_IPV6)
3535                 else if (addr->sa_family == AF_INET6) {
3536                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3537
3538                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3539                 }
3540 #endif
3541         }
3542         daddr = cma_dst_addr(id_priv);
3543         daddr->sa_family = addr->sa_family;
3544
3545         ret = cma_get_port(id_priv);
3546         if (ret)
3547                 goto err2;
3548
3549         return 0;
3550 err2:
3551         rdma_restrack_del(&id_priv->res);
3552         if (id_priv->cma_dev)
3553                 cma_release_dev(id_priv);
3554 err1:
3555         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3556         return ret;
3557 }
3558 EXPORT_SYMBOL(rdma_bind_addr);
3559
3560 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3561 {
3562         struct cma_hdr *cma_hdr;
3563
3564         cma_hdr = hdr;
3565         cma_hdr->cma_version = CMA_VERSION;
3566         if (cma_family(id_priv) == AF_INET) {
3567                 struct sockaddr_in *src4, *dst4;
3568
3569                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3570                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3571
3572                 cma_set_ip_ver(cma_hdr, 4);
3573                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3574                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3575                 cma_hdr->port = src4->sin_port;
3576         } else if (cma_family(id_priv) == AF_INET6) {
3577                 struct sockaddr_in6 *src6, *dst6;
3578
3579                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3580                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3581
3582                 cma_set_ip_ver(cma_hdr, 6);
3583                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3584                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3585                 cma_hdr->port = src6->sin6_port;
3586         }
3587         return 0;
3588 }
3589
3590 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3591                                 const struct ib_cm_event *ib_event)
3592 {
3593         struct rdma_id_private *id_priv = cm_id->context;
3594         struct rdma_cm_event event = {};
3595         const struct ib_cm_sidr_rep_event_param *rep =
3596                                 &ib_event->param.sidr_rep_rcvd;
3597         int ret = 0;
3598
3599         mutex_lock(&id_priv->handler_mutex);
3600         if (id_priv->state != RDMA_CM_CONNECT)
3601                 goto out;
3602
3603         switch (ib_event->event) {
3604         case IB_CM_SIDR_REQ_ERROR:
3605                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3606                 event.status = -ETIMEDOUT;
3607                 break;
3608         case IB_CM_SIDR_REP_RECEIVED:
3609                 event.param.ud.private_data = ib_event->private_data;
3610                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3611                 if (rep->status != IB_SIDR_SUCCESS) {
3612                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3613                         event.status = ib_event->param.sidr_rep_rcvd.status;
3614                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3615                                              event.status);
3616                         break;
3617                 }
3618                 ret = cma_set_qkey(id_priv, rep->qkey);
3619                 if (ret) {
3620                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3621                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3622                         event.status = ret;
3623                         break;
3624                 }
3625                 ib_init_ah_attr_from_path(id_priv->id.device,
3626                                           id_priv->id.port_num,
3627                                           id_priv->id.route.path_rec,
3628                                           &event.param.ud.ah_attr,
3629                                           rep->sgid_attr);
3630                 event.param.ud.qp_num = rep->qpn;
3631                 event.param.ud.qkey = rep->qkey;
3632                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3633                 event.status = 0;
3634                 break;
3635         default:
3636                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3637                        ib_event->event);
3638                 goto out;
3639         }
3640
3641         ret = id_priv->id.event_handler(&id_priv->id, &event);
3642
3643         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
3644         if (ret) {
3645                 /* Destroy the CM ID by returning a non-zero value. */
3646                 id_priv->cm_id.ib = NULL;
3647                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3648                 mutex_unlock(&id_priv->handler_mutex);
3649                 rdma_destroy_id(&id_priv->id);
3650                 return ret;
3651         }
3652 out:
3653         mutex_unlock(&id_priv->handler_mutex);
3654         return ret;
3655 }
3656
3657 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3658                               struct rdma_conn_param *conn_param)
3659 {
3660         struct ib_cm_sidr_req_param req;
3661         struct ib_cm_id *id;
3662         void *private_data;
3663         u8 offset;
3664         int ret;
3665
3666         memset(&req, 0, sizeof req);
3667         offset = cma_user_data_offset(id_priv);
3668         req.private_data_len = offset + conn_param->private_data_len;
3669         if (req.private_data_len < conn_param->private_data_len)
3670                 return -EINVAL;
3671
3672         if (req.private_data_len) {
3673                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3674                 if (!private_data)
3675                         return -ENOMEM;
3676         } else {
3677                 private_data = NULL;
3678         }
3679
3680         if (conn_param->private_data && conn_param->private_data_len)
3681                 memcpy(private_data + offset, conn_param->private_data,
3682                        conn_param->private_data_len);
3683
3684         if (private_data) {
3685                 ret = cma_format_hdr(private_data, id_priv);
3686                 if (ret)
3687                         goto out;
3688                 req.private_data = private_data;
3689         }
3690
3691         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3692                              id_priv);
3693         if (IS_ERR(id)) {
3694                 ret = PTR_ERR(id);
3695                 goto out;
3696         }
3697         id_priv->cm_id.ib = id;
3698
3699         req.path = id_priv->id.route.path_rec;
3700         req.sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3701         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3702         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3703         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3704
3705         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3706         if (ret) {
3707                 ib_destroy_cm_id(id_priv->cm_id.ib);
3708                 id_priv->cm_id.ib = NULL;
3709         }
3710 out:
3711         kfree(private_data);
3712         return ret;
3713 }
3714
3715 static int cma_connect_ib(struct rdma_id_private *id_priv,
3716                           struct rdma_conn_param *conn_param)
3717 {
3718         struct ib_cm_req_param req;
3719         struct rdma_route *route;
3720         void *private_data;
3721         struct ib_cm_id *id;
3722         u8 offset;
3723         int ret;
3724
3725         memset(&req, 0, sizeof req);
3726         offset = cma_user_data_offset(id_priv);
3727         req.private_data_len = offset + conn_param->private_data_len;
3728         if (req.private_data_len < conn_param->private_data_len)
3729                 return -EINVAL;
3730
3731         if (req.private_data_len) {
3732                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3733                 if (!private_data)
3734                         return -ENOMEM;
3735         } else {
3736                 private_data = NULL;
3737         }
3738
3739         if (conn_param->private_data && conn_param->private_data_len)
3740                 memcpy(private_data + offset, conn_param->private_data,
3741                        conn_param->private_data_len);
3742
3743         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3744         if (IS_ERR(id)) {
3745                 ret = PTR_ERR(id);
3746                 goto out;
3747         }
3748         id_priv->cm_id.ib = id;
3749
3750         route = &id_priv->id.route;
3751         if (private_data) {
3752                 ret = cma_format_hdr(private_data, id_priv);
3753                 if (ret)
3754                         goto out;
3755                 req.private_data = private_data;
3756         }
3757
3758         req.primary_path = &route->path_rec[0];
3759         if (route->num_paths == 2)
3760                 req.alternate_path = &route->path_rec[1];
3761
3762         req.ppath_sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3763         /* Alternate path SGID attribute currently unsupported */
3764         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3765         req.qp_num = id_priv->qp_num;
3766         req.qp_type = id_priv->id.qp_type;
3767         req.starting_psn = id_priv->seq_num;
3768         req.responder_resources = conn_param->responder_resources;
3769         req.initiator_depth = conn_param->initiator_depth;
3770         req.flow_control = conn_param->flow_control;
3771         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3772         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3773         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3774         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3775         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3776         req.srq = id_priv->srq ? 1 : 0;
3777
3778         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3779 out:
3780         if (ret && !IS_ERR(id)) {
3781                 ib_destroy_cm_id(id);
3782                 id_priv->cm_id.ib = NULL;
3783         }
3784
3785         kfree(private_data);
3786         return ret;
3787 }
3788
3789 static int cma_connect_iw(struct rdma_id_private *id_priv,
3790                           struct rdma_conn_param *conn_param)
3791 {
3792         struct iw_cm_id *cm_id;
3793         int ret;
3794         struct iw_cm_conn_param iw_param;
3795
3796         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3797         if (IS_ERR(cm_id))
3798                 return PTR_ERR(cm_id);
3799
3800         cm_id->tos = id_priv->tos;
3801         id_priv->cm_id.iw = cm_id;
3802
3803         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3804                rdma_addr_size(cma_src_addr(id_priv)));
3805         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3806                rdma_addr_size(cma_dst_addr(id_priv)));
3807
3808         ret = cma_modify_qp_rtr(id_priv, conn_param);
3809         if (ret)
3810                 goto out;
3811
3812         if (conn_param) {
3813                 iw_param.ord = conn_param->initiator_depth;
3814                 iw_param.ird = conn_param->responder_resources;
3815                 iw_param.private_data = conn_param->private_data;
3816                 iw_param.private_data_len = conn_param->private_data_len;
3817                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3818         } else {
3819                 memset(&iw_param, 0, sizeof iw_param);
3820                 iw_param.qpn = id_priv->qp_num;
3821         }
3822         ret = iw_cm_connect(cm_id, &iw_param);
3823 out:
3824         if (ret) {
3825                 iw_destroy_cm_id(cm_id);
3826                 id_priv->cm_id.iw = NULL;
3827         }
3828         return ret;
3829 }
3830
3831 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3832 {
3833         struct rdma_id_private *id_priv;
3834         int ret;
3835
3836         id_priv = container_of(id, struct rdma_id_private, id);
3837         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3838                 return -EINVAL;
3839
3840         if (!id->qp) {
3841                 id_priv->qp_num = conn_param->qp_num;
3842                 id_priv->srq = conn_param->srq;
3843         }
3844
3845         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3846                 if (id->qp_type == IB_QPT_UD)
3847                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3848                 else
3849                         ret = cma_connect_ib(id_priv, conn_param);
3850         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3851                 ret = cma_connect_iw(id_priv, conn_param);
3852         else
3853                 ret = -ENOSYS;
3854         if (ret)
3855                 goto err;
3856
3857         return 0;
3858 err:
3859         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3860         return ret;
3861 }
3862 EXPORT_SYMBOL(rdma_connect);
3863
3864 static int cma_accept_ib(struct rdma_id_private *id_priv,
3865                          struct rdma_conn_param *conn_param)
3866 {
3867         struct ib_cm_rep_param rep;
3868         int ret;
3869
3870         ret = cma_modify_qp_rtr(id_priv, conn_param);
3871         if (ret)
3872                 goto out;
3873
3874         ret = cma_modify_qp_rts(id_priv, conn_param);
3875         if (ret)
3876                 goto out;
3877
3878         memset(&rep, 0, sizeof rep);
3879         rep.qp_num = id_priv->qp_num;
3880         rep.starting_psn = id_priv->seq_num;
3881         rep.private_data = conn_param->private_data;
3882         rep.private_data_len = conn_param->private_data_len;
3883         rep.responder_resources = conn_param->responder_resources;
3884         rep.initiator_depth = conn_param->initiator_depth;
3885         rep.failover_accepted = 0;
3886         rep.flow_control = conn_param->flow_control;
3887         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3888         rep.srq = id_priv->srq ? 1 : 0;
3889
3890         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3891 out:
3892         return ret;
3893 }
3894
3895 static int cma_accept_iw(struct rdma_id_private *id_priv,
3896                   struct rdma_conn_param *conn_param)
3897 {
3898         struct iw_cm_conn_param iw_param;
3899         int ret;
3900
3901         if (!conn_param)
3902                 return -EINVAL;
3903
3904         ret = cma_modify_qp_rtr(id_priv, conn_param);
3905         if (ret)
3906                 return ret;
3907
3908         iw_param.ord = conn_param->initiator_depth;
3909         iw_param.ird = conn_param->responder_resources;
3910         iw_param.private_data = conn_param->private_data;
3911         iw_param.private_data_len = conn_param->private_data_len;
3912         if (id_priv->id.qp) {
3913                 iw_param.qpn = id_priv->qp_num;
3914         } else
3915                 iw_param.qpn = conn_param->qp_num;
3916
3917         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3918 }
3919
3920 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3921                              enum ib_cm_sidr_status status, u32 qkey,
3922                              const void *private_data, int private_data_len)
3923 {
3924         struct ib_cm_sidr_rep_param rep;
3925         int ret;
3926
3927         memset(&rep, 0, sizeof rep);
3928         rep.status = status;
3929         if (status == IB_SIDR_SUCCESS) {
3930                 ret = cma_set_qkey(id_priv, qkey);
3931                 if (ret)
3932                         return ret;
3933                 rep.qp_num = id_priv->qp_num;
3934                 rep.qkey = id_priv->qkey;
3935         }
3936         rep.private_data = private_data;
3937         rep.private_data_len = private_data_len;
3938
3939         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3940 }
3941
3942 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3943                   const char *caller)
3944 {
3945         struct rdma_id_private *id_priv;
3946         int ret;
3947
3948         id_priv = container_of(id, struct rdma_id_private, id);
3949
3950         rdma_restrack_set_task(&id_priv->res, caller);
3951
3952         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3953                 return -EINVAL;
3954
3955         if (!id->qp && conn_param) {
3956                 id_priv->qp_num = conn_param->qp_num;
3957                 id_priv->srq = conn_param->srq;
3958         }
3959
3960         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3961                 if (id->qp_type == IB_QPT_UD) {
3962                         if (conn_param)
3963                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3964                                                         conn_param->qkey,
3965                                                         conn_param->private_data,
3966                                                         conn_param->private_data_len);
3967                         else
3968                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3969                                                         0, NULL, 0);
3970                 } else {
3971                         if (conn_param)
3972                                 ret = cma_accept_ib(id_priv, conn_param);
3973                         else
3974                                 ret = cma_rep_recv(id_priv);
3975                 }
3976         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3977                 ret = cma_accept_iw(id_priv, conn_param);
3978         else
3979                 ret = -ENOSYS;
3980
3981         if (ret)
3982                 goto reject;
3983
3984         return 0;
3985 reject:
3986         cma_modify_qp_err(id_priv);
3987         rdma_reject(id, NULL, 0);
3988         return ret;
3989 }
3990 EXPORT_SYMBOL(__rdma_accept);
3991
3992 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3993 {
3994         struct rdma_id_private *id_priv;
3995         int ret;
3996
3997         id_priv = container_of(id, struct rdma_id_private, id);
3998         if (!id_priv->cm_id.ib)
3999                 return -EINVAL;
4000
4001         switch (id->device->node_type) {
4002         case RDMA_NODE_IB_CA:
4003                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
4004                 break;
4005         default:
4006                 ret = 0;
4007                 break;
4008         }
4009         return ret;
4010 }
4011 EXPORT_SYMBOL(rdma_notify);
4012
4013 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
4014                 u8 private_data_len)
4015 {
4016         struct rdma_id_private *id_priv;
4017         int ret;
4018
4019         id_priv = container_of(id, struct rdma_id_private, id);
4020         if (!id_priv->cm_id.ib)
4021                 return -EINVAL;
4022
4023         if (rdma_cap_ib_cm(id->device, id->port_num)) {
4024                 if (id->qp_type == IB_QPT_UD)
4025                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
4026                                                 private_data, private_data_len);
4027                 else
4028                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
4029                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
4030                                              0, private_data, private_data_len);
4031         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
4032                 ret = iw_cm_reject(id_priv->cm_id.iw,
4033                                    private_data, private_data_len);
4034         } else
4035                 ret = -ENOSYS;
4036
4037         return ret;
4038 }
4039 EXPORT_SYMBOL(rdma_reject);
4040
4041 int rdma_disconnect(struct rdma_cm_id *id)
4042 {
4043         struct rdma_id_private *id_priv;
4044         int ret;
4045
4046         id_priv = container_of(id, struct rdma_id_private, id);
4047         if (!id_priv->cm_id.ib)
4048                 return -EINVAL;
4049
4050         if (rdma_cap_ib_cm(id->device, id->port_num)) {
4051                 ret = cma_modify_qp_err(id_priv);
4052                 if (ret)
4053                         goto out;
4054                 /* Initiate or respond to a disconnect. */
4055                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
4056                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
4057         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
4058                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
4059         } else
4060                 ret = -EINVAL;
4061
4062 out:
4063         return ret;
4064 }
4065 EXPORT_SYMBOL(rdma_disconnect);
4066
4067 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
4068 {
4069         struct rdma_id_private *id_priv;
4070         struct cma_multicast *mc = multicast->context;
4071         struct rdma_cm_event event = {};
4072         int ret = 0;
4073
4074         id_priv = mc->id_priv;
4075         mutex_lock(&id_priv->handler_mutex);
4076         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
4077             id_priv->state != RDMA_CM_ADDR_RESOLVED)
4078                 goto out;
4079
4080         if (!status)
4081                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
4082         else
4083                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
4084                                      status);
4085         mutex_lock(&id_priv->qp_mutex);
4086         if (!status && id_priv->id.qp) {
4087                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
4088                                          be16_to_cpu(multicast->rec.mlid));
4089                 if (status)
4090                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
4091                                              status);
4092         }
4093         mutex_unlock(&id_priv->qp_mutex);
4094
4095         event.status = status;
4096         event.param.ud.private_data = mc->context;
4097         if (!status) {
4098                 struct rdma_dev_addr *dev_addr =
4099                         &id_priv->id.route.addr.dev_addr;
4100                 struct net_device *ndev =
4101                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4102                 enum ib_gid_type gid_type =
4103                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4104                         rdma_start_port(id_priv->cma_dev->device)];
4105
4106                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
4107                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
4108                                                id_priv->id.port_num,
4109                                                &multicast->rec,
4110                                                ndev, gid_type,
4111                                                &event.param.ud.ah_attr);
4112                 if (ret)
4113                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4114
4115                 event.param.ud.qp_num = 0xFFFFFF;
4116                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
4117                 if (ndev)
4118                         dev_put(ndev);
4119         } else
4120                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4121
4122         ret = id_priv->id.event_handler(&id_priv->id, &event);
4123
4124         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
4125         if (ret) {
4126                 cma_exch(id_priv, RDMA_CM_DESTROYING);
4127                 mutex_unlock(&id_priv->handler_mutex);
4128                 rdma_destroy_id(&id_priv->id);
4129                 return 0;
4130         }
4131
4132 out:
4133         mutex_unlock(&id_priv->handler_mutex);
4134         return 0;
4135 }
4136
4137 static void cma_set_mgid(struct rdma_id_private *id_priv,
4138                          struct sockaddr *addr, union ib_gid *mgid)
4139 {
4140         unsigned char mc_map[MAX_ADDR_LEN];
4141         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4142         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4143         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4144
4145         if (cma_any_addr(addr)) {
4146                 memset(mgid, 0, sizeof *mgid);
4147         } else if ((addr->sa_family == AF_INET6) &&
4148                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4149                                                                  0xFF10A01B)) {
4150                 /* IPv6 address is an SA assigned MGID. */
4151                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4152         } else if (addr->sa_family == AF_IB) {
4153                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4154         } else if (addr->sa_family == AF_INET6) {
4155                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4156                 if (id_priv->id.ps == RDMA_PS_UDP)
4157                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4158                 *mgid = *(union ib_gid *) (mc_map + 4);
4159         } else {
4160                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4161                 if (id_priv->id.ps == RDMA_PS_UDP)
4162                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4163                 *mgid = *(union ib_gid *) (mc_map + 4);
4164         }
4165 }
4166
4167 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4168                                  struct cma_multicast *mc)
4169 {
4170         struct ib_sa_mcmember_rec rec;
4171         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4172         ib_sa_comp_mask comp_mask;
4173         int ret;
4174
4175         ib_addr_get_mgid(dev_addr, &rec.mgid);
4176         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4177                                      &rec.mgid, &rec);
4178         if (ret)
4179                 return ret;
4180
4181         ret = cma_set_qkey(id_priv, 0);
4182         if (ret)
4183                 return ret;
4184
4185         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4186         rec.qkey = cpu_to_be32(id_priv->qkey);
4187         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4188         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4189         rec.join_state = mc->join_state;
4190
4191         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4192             (!ib_sa_sendonly_fullmem_support(&sa_client,
4193                                              id_priv->id.device,
4194                                              id_priv->id.port_num))) {
4195                 dev_warn(
4196                         &id_priv->id.device->dev,
4197                         "RDMA CM: port %u Unable to multicast join: SM doesn't support Send Only Full Member option\n",
4198                         id_priv->id.port_num);
4199                 return -EOPNOTSUPP;
4200         }
4201
4202         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4203                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4204                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4205                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4206                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4207
4208         if (id_priv->id.ps == RDMA_PS_IPOIB)
4209                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4210                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4211                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4212                              IB_SA_MCMEMBER_REC_MTU |
4213                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4214
4215         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4216                                                 id_priv->id.port_num, &rec,
4217                                                 comp_mask, GFP_KERNEL,
4218                                                 cma_ib_mc_handler, mc);
4219         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4220 }
4221
4222 static void iboe_mcast_work_handler(struct work_struct *work)
4223 {
4224         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4225         struct cma_multicast *mc = mw->mc;
4226         struct ib_sa_multicast *m = mc->multicast.ib;
4227
4228         mc->multicast.ib->context = mc;
4229         cma_ib_mc_handler(0, m);
4230         kref_put(&mc->mcref, release_mc);
4231         kfree(mw);
4232 }
4233
4234 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4235                               enum ib_gid_type gid_type)
4236 {
4237         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4238         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4239
4240         if (cma_any_addr(addr)) {
4241                 memset(mgid, 0, sizeof *mgid);
4242         } else if (addr->sa_family == AF_INET6) {
4243                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4244         } else {
4245                 mgid->raw[0] =
4246                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4247                 mgid->raw[1] =
4248                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4249                 mgid->raw[2] = 0;
4250                 mgid->raw[3] = 0;
4251                 mgid->raw[4] = 0;
4252                 mgid->raw[5] = 0;
4253                 mgid->raw[6] = 0;
4254                 mgid->raw[7] = 0;
4255                 mgid->raw[8] = 0;
4256                 mgid->raw[9] = 0;
4257                 mgid->raw[10] = 0xff;
4258                 mgid->raw[11] = 0xff;
4259                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4260         }
4261 }
4262
4263 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4264                                    struct cma_multicast *mc)
4265 {
4266         struct iboe_mcast_work *work;
4267         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4268         int err = 0;
4269         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4270         struct net_device *ndev = NULL;
4271         enum ib_gid_type gid_type;
4272         bool send_only;
4273
4274         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4275
4276         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4277                 return -EINVAL;
4278
4279         work = kzalloc(sizeof *work, GFP_KERNEL);
4280         if (!work)
4281                 return -ENOMEM;
4282
4283         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4284         if (!mc->multicast.ib) {
4285                 err = -ENOMEM;
4286                 goto out1;
4287         }
4288
4289         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4290                    rdma_start_port(id_priv->cma_dev->device)];
4291         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4292
4293         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4294         if (id_priv->id.ps == RDMA_PS_UDP)
4295                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4296
4297         if (dev_addr->bound_dev_if)
4298                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4299         if (!ndev) {
4300                 err = -ENODEV;
4301                 goto out2;
4302         }
4303         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4304         mc->multicast.ib->rec.hop_limit = 1;
4305         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4306
4307         if (addr->sa_family == AF_INET) {
4308                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4309                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4310                         if (!send_only) {
4311                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4312                                                     true);
4313                         }
4314                 }
4315         } else {
4316                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4317                         err = -ENOTSUPP;
4318         }
4319         dev_put(ndev);
4320         if (err || !mc->multicast.ib->rec.mtu) {
4321                 if (!err)
4322                         err = -EINVAL;
4323                 goto out2;
4324         }
4325         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4326                     &mc->multicast.ib->rec.port_gid);
4327         work->id = id_priv;
4328         work->mc = mc;
4329         INIT_WORK(&work->work, iboe_mcast_work_handler);
4330         kref_get(&mc->mcref);
4331         queue_work(cma_wq, &work->work);
4332
4333         return 0;
4334
4335 out2:
4336         kfree(mc->multicast.ib);
4337 out1:
4338         kfree(work);
4339         return err;
4340 }
4341
4342 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4343                         u8 join_state, void *context)
4344 {
4345         struct rdma_id_private *id_priv;
4346         struct cma_multicast *mc;
4347         int ret;
4348
4349         if (!id->device)
4350                 return -EINVAL;
4351
4352         id_priv = container_of(id, struct rdma_id_private, id);
4353         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4354             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4355                 return -EINVAL;
4356
4357         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4358         if (!mc)
4359                 return -ENOMEM;
4360
4361         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4362         mc->context = context;
4363         mc->id_priv = id_priv;
4364         mc->join_state = join_state;
4365
4366         if (rdma_protocol_roce(id->device, id->port_num)) {
4367                 kref_init(&mc->mcref);
4368                 ret = cma_iboe_join_multicast(id_priv, mc);
4369                 if (ret)
4370                         goto out_err;
4371         } else if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4372                 ret = cma_join_ib_multicast(id_priv, mc);
4373                 if (ret)
4374                         goto out_err;
4375         } else {
4376                 ret = -ENOSYS;
4377                 goto out_err;
4378         }
4379
4380         spin_lock(&id_priv->lock);
4381         list_add(&mc->list, &id_priv->mc_list);
4382         spin_unlock(&id_priv->lock);
4383
4384         return 0;
4385 out_err:
4386         kfree(mc);
4387         return ret;
4388 }
4389 EXPORT_SYMBOL(rdma_join_multicast);
4390
4391 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4392 {
4393         struct rdma_id_private *id_priv;
4394         struct cma_multicast *mc;
4395
4396         id_priv = container_of(id, struct rdma_id_private, id);
4397         spin_lock_irq(&id_priv->lock);
4398         list_for_each_entry(mc, &id_priv->mc_list, list) {
4399                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4400                         list_del(&mc->list);
4401                         spin_unlock_irq(&id_priv->lock);
4402
4403                         if (id->qp)
4404                                 ib_detach_mcast(id->qp,
4405                                                 &mc->multicast.ib->rec.mgid,
4406                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4407
4408                         BUG_ON(id_priv->cma_dev->device != id->device);
4409
4410                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4411                                 ib_sa_free_multicast(mc->multicast.ib);
4412                                 kfree(mc);
4413                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4414                                 cma_leave_roce_mc_group(id_priv, mc);
4415                         }
4416                         return;
4417                 }
4418         }
4419         spin_unlock_irq(&id_priv->lock);
4420 }
4421 EXPORT_SYMBOL(rdma_leave_multicast);
4422
4423 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4424 {
4425         struct rdma_dev_addr *dev_addr;
4426         struct cma_ndev_work *work;
4427
4428         dev_addr = &id_priv->id.route.addr.dev_addr;
4429
4430         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4431             (net_eq(dev_net(ndev), dev_addr->net)) &&
4432             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4433                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4434                         ndev->name, &id_priv->id);
4435                 work = kzalloc(sizeof *work, GFP_KERNEL);
4436                 if (!work)
4437                         return -ENOMEM;
4438
4439                 INIT_WORK(&work->work, cma_ndev_work_handler);
4440                 work->id = id_priv;
4441                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4442                 atomic_inc(&id_priv->refcount);
4443                 queue_work(cma_wq, &work->work);
4444         }
4445
4446         return 0;
4447 }
4448
4449 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4450                                void *ptr)
4451 {
4452         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4453         struct cma_device *cma_dev;
4454         struct rdma_id_private *id_priv;
4455         int ret = NOTIFY_DONE;
4456
4457         if (event != NETDEV_BONDING_FAILOVER)
4458                 return NOTIFY_DONE;
4459
4460         if (!netif_is_bond_master(ndev))
4461                 return NOTIFY_DONE;
4462
4463         mutex_lock(&lock);
4464         list_for_each_entry(cma_dev, &dev_list, list)
4465                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4466                         ret = cma_netdev_change(ndev, id_priv);
4467                         if (ret)
4468                                 goto out;
4469                 }
4470
4471 out:
4472         mutex_unlock(&lock);
4473         return ret;
4474 }
4475
4476 static struct notifier_block cma_nb = {
4477         .notifier_call = cma_netdev_callback
4478 };
4479
4480 static void cma_add_one(struct ib_device *device)
4481 {
4482         struct cma_device *cma_dev;
4483         struct rdma_id_private *id_priv;
4484         unsigned int i;
4485         unsigned long supported_gids = 0;
4486
4487         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4488         if (!cma_dev)
4489                 return;
4490
4491         cma_dev->device = device;
4492         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4493                                             sizeof(*cma_dev->default_gid_type),
4494                                             GFP_KERNEL);
4495         if (!cma_dev->default_gid_type)
4496                 goto free_cma_dev;
4497
4498         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4499                                             sizeof(*cma_dev->default_roce_tos),
4500                                             GFP_KERNEL);
4501         if (!cma_dev->default_roce_tos)
4502                 goto free_gid_type;
4503
4504         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4505                 supported_gids = roce_gid_type_mask_support(device, i);
4506                 WARN_ON(!supported_gids);
4507                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4508                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4509                                 CMA_PREFERRED_ROCE_GID_TYPE;
4510                 else
4511                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4512                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4513                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4514         }
4515
4516         init_completion(&cma_dev->comp);
4517         atomic_set(&cma_dev->refcount, 1);
4518         INIT_LIST_HEAD(&cma_dev->id_list);
4519         ib_set_client_data(device, &cma_client, cma_dev);
4520
4521         mutex_lock(&lock);
4522         list_add_tail(&cma_dev->list, &dev_list);
4523         list_for_each_entry(id_priv, &listen_any_list, list)
4524                 cma_listen_on_dev(id_priv, cma_dev);
4525         mutex_unlock(&lock);
4526
4527         return;
4528
4529 free_gid_type:
4530         kfree(cma_dev->default_gid_type);
4531
4532 free_cma_dev:
4533         kfree(cma_dev);
4534
4535         return;
4536 }
4537
4538 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4539 {
4540         struct rdma_cm_event event = {};
4541         enum rdma_cm_state state;
4542         int ret = 0;
4543
4544         /* Record that we want to remove the device */
4545         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4546         if (state == RDMA_CM_DESTROYING)
4547                 return 0;
4548
4549         cma_cancel_operation(id_priv, state);
4550         mutex_lock(&id_priv->handler_mutex);
4551
4552         /* Check for destruction from another callback. */
4553         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4554                 goto out;
4555
4556         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4557         ret = id_priv->id.event_handler(&id_priv->id, &event);
4558 out:
4559         mutex_unlock(&id_priv->handler_mutex);
4560         return ret;
4561 }
4562
4563 static void cma_process_remove(struct cma_device *cma_dev)
4564 {
4565         struct rdma_id_private *id_priv;
4566         int ret;
4567
4568         mutex_lock(&lock);
4569         while (!list_empty(&cma_dev->id_list)) {
4570                 id_priv = list_entry(cma_dev->id_list.next,
4571                                      struct rdma_id_private, list);
4572
4573                 list_del(&id_priv->listen_list);
4574                 list_del_init(&id_priv->list);
4575                 atomic_inc(&id_priv->refcount);
4576                 mutex_unlock(&lock);
4577
4578                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4579                 cma_deref_id(id_priv);
4580                 if (ret)
4581                         rdma_destroy_id(&id_priv->id);
4582
4583                 mutex_lock(&lock);
4584         }
4585         mutex_unlock(&lock);
4586
4587         cma_deref_dev(cma_dev);
4588         wait_for_completion(&cma_dev->comp);
4589 }
4590
4591 static void cma_remove_one(struct ib_device *device, void *client_data)
4592 {
4593         struct cma_device *cma_dev = client_data;
4594
4595         if (!cma_dev)
4596                 return;
4597
4598         mutex_lock(&lock);
4599         list_del(&cma_dev->list);
4600         mutex_unlock(&lock);
4601
4602         cma_process_remove(cma_dev);
4603         kfree(cma_dev->default_roce_tos);
4604         kfree(cma_dev->default_gid_type);
4605         kfree(cma_dev);
4606 }
4607
4608 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4609 {
4610         struct nlmsghdr *nlh;
4611         struct rdma_cm_id_stats *id_stats;
4612         struct rdma_id_private *id_priv;
4613         struct rdma_cm_id *id = NULL;
4614         struct cma_device *cma_dev;
4615         int i_dev = 0, i_id = 0;
4616
4617         /*
4618          * We export all of the IDs as a sequence of messages.  Each
4619          * ID gets its own netlink message.
4620          */
4621         mutex_lock(&lock);
4622
4623         list_for_each_entry(cma_dev, &dev_list, list) {
4624                 if (i_dev < cb->args[0]) {
4625                         i_dev++;
4626                         continue;
4627                 }
4628
4629                 i_id = 0;
4630                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4631                         if (i_id < cb->args[1]) {
4632                                 i_id++;
4633                                 continue;
4634                         }
4635
4636                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4637                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4638                                                 RDMA_NL_RDMA_CM_ID_STATS,
4639                                                 NLM_F_MULTI);
4640                         if (!id_stats)
4641                                 goto out;
4642
4643                         memset(id_stats, 0, sizeof *id_stats);
4644                         id = &id_priv->id;
4645                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4646                         id_stats->port_num = id->port_num;
4647                         id_stats->bound_dev_if =
4648                                 id->route.addr.dev_addr.bound_dev_if;
4649
4650                         if (ibnl_put_attr(skb, nlh,
4651                                           rdma_addr_size(cma_src_addr(id_priv)),
4652                                           cma_src_addr(id_priv),
4653                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4654                                 goto out;
4655                         if (ibnl_put_attr(skb, nlh,
4656                                           rdma_addr_size(cma_dst_addr(id_priv)),
4657                                           cma_dst_addr(id_priv),
4658                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4659                                 goto out;
4660
4661                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4662                         id_stats->port_space    = id->ps;
4663                         id_stats->cm_state      = id_priv->state;
4664                         id_stats->qp_num        = id_priv->qp_num;
4665                         id_stats->qp_type       = id->qp_type;
4666
4667                         i_id++;
4668                         nlmsg_end(skb, nlh);
4669                 }
4670
4671                 cb->args[1] = 0;
4672                 i_dev++;
4673         }
4674
4675 out:
4676         mutex_unlock(&lock);
4677         cb->args[0] = i_dev;
4678         cb->args[1] = i_id;
4679
4680         return skb->len;
4681 }
4682
4683 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4684         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4685 };
4686
4687 static int cma_init_net(struct net *net)
4688 {
4689         struct cma_pernet *pernet = cma_pernet(net);
4690
4691         idr_init(&pernet->tcp_ps);
4692         idr_init(&pernet->udp_ps);
4693         idr_init(&pernet->ipoib_ps);
4694         idr_init(&pernet->ib_ps);
4695
4696         return 0;
4697 }
4698
4699 static void cma_exit_net(struct net *net)
4700 {
4701         struct cma_pernet *pernet = cma_pernet(net);
4702
4703         idr_destroy(&pernet->tcp_ps);
4704         idr_destroy(&pernet->udp_ps);
4705         idr_destroy(&pernet->ipoib_ps);
4706         idr_destroy(&pernet->ib_ps);
4707 }
4708
4709 static struct pernet_operations cma_pernet_operations = {
4710         .init = cma_init_net,
4711         .exit = cma_exit_net,
4712         .id = &cma_pernet_id,
4713         .size = sizeof(struct cma_pernet),
4714 };
4715
4716 static int __init cma_init(void)
4717 {
4718         int ret;
4719
4720         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4721         if (!cma_wq)
4722                 return -ENOMEM;
4723
4724         ret = register_pernet_subsys(&cma_pernet_operations);
4725         if (ret)
4726                 goto err_wq;
4727
4728         ib_sa_register_client(&sa_client);
4729         register_netdevice_notifier(&cma_nb);
4730
4731         ret = ib_register_client(&cma_client);
4732         if (ret)
4733                 goto err;
4734
4735         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4736         cma_configfs_init();
4737
4738         return 0;
4739
4740 err:
4741         unregister_netdevice_notifier(&cma_nb);
4742         ib_sa_unregister_client(&sa_client);
4743 err_wq:
4744         destroy_workqueue(cma_wq);
4745         return ret;
4746 }
4747
4748 static void __exit cma_cleanup(void)
4749 {
4750         cma_configfs_exit();
4751         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4752         ib_unregister_client(&cma_client);
4753         unregister_netdevice_notifier(&cma_nb);
4754         ib_sa_unregister_client(&sa_client);
4755         unregister_pernet_subsys(&cma_pernet_operations);
4756         destroy_workqueue(cma_wq);
4757 }
4758
4759 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4760
4761 module_init(cma_init);
4762 module_exit(cma_cleanup);