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 static void cma_add_one(struct ib_device *device);
 150 static void cma_remove_one(struct ib_device *device, void *client_data);
 151
 152 static struct ib_client cma_client = {
 153         .name   = "cma",
 154         .add    = cma_add_one,
 155         .remove = cma_remove_one
 156 };
 157
 158 static struct ib_sa_client sa_client;
 159 static struct rdma_addr_client addr_client;
 160 static LIST_HEAD(dev_list);
 161 static LIST_HEAD(listen_any_list);
 162 static DEFINE_MUTEX(lock);
 163 static struct workqueue_struct *cma_wq;
 164 static unsigned int cma_pernet_id;
 165
 166 struct cma_pernet {
 167         struct idr tcp_ps;
 168         struct idr udp_ps;
 169         struct idr ipoib_ps;
 170         struct idr ib_ps;
 171 };
 172
 173 static struct cma_pernet *cma_pernet(struct net *net)
 174 {
 175         return net_generic(net, cma_pernet_id);
 176 }
 177
 178 static struct idr *cma_pernet_idr(struct net *net, enum rdma_ucm_port_space ps)
 179 {
 180         struct cma_pernet *pernet = cma_pernet(net);
 181
 182         switch (ps) {
 183         case RDMA_PS_TCP:
 184                 return &pernet->tcp_ps;
 185         case RDMA_PS_UDP:
 186                 return &pernet->udp_ps;
 187         case RDMA_PS_IPOIB:
 188                 return &pernet->ipoib_ps;
 189         case RDMA_PS_IB:
 190                 return &pernet->ib_ps;
 191         default:
 192                 return NULL;
 193         }
 194 }
 195
 196 struct cma_device {
 197         struct list_head        list;
 198         struct ib_device        *device;
 199         struct completion       comp;
 200         atomic_t                refcount;
 201         struct list_head        id_list;
 202         enum ib_gid_type        *default_gid_type;
 203         u8                      *default_roce_tos;
 204 };
 205
 206 struct rdma_bind_list {
 207         enum rdma_ucm_port_space ps;
 208         struct hlist_head       owners;
 209         unsigned short          port;
 210 };
 211
 212 struct class_port_info_context {
 213         struct ib_class_port_info       *class_port_info;
 214         struct ib_device                *device;
 215         struct completion               done;
 216         struct ib_sa_query              *sa_query;
 217         u8                              port_num;
 218 };
 219
 220 static int cma_ps_alloc(struct net *net, enum rdma_ucm_port_space ps,
 221                         struct rdma_bind_list *bind_list, int snum)
 222 {
 223         struct idr *idr = cma_pernet_idr(net, ps);
 224
 225         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 226 }
 227
 228 static struct rdma_bind_list *cma_ps_find(struct net *net,
 229                                           enum rdma_ucm_port_space ps, int snum)
 230 {
 231         struct idr *idr = cma_pernet_idr(net, ps);
 232
 233         return idr_find(idr, snum);
 234 }
 235
 236 static void cma_ps_remove(struct net *net, enum rdma_ucm_port_space ps,
 237                           int snum)
 238 {
 239         struct idr *idr = cma_pernet_idr(net, ps);
 240
 241         idr_remove(idr, snum);
 242 }
 243
 244 enum {
 245         CMA_OPTION_AFONLY,
 246 };
 247
 248 void cma_ref_dev(struct cma_device *cma_dev)
 249 {
 250         atomic_inc(&cma_dev->refcount);
 251 }
 252
 253 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 254                                              void               *cookie)
 255 {
 256         struct cma_device *cma_dev;
 257         struct cma_device *found_cma_dev = NULL;
 258
 259         mutex_lock(&lock);
 260
 261         list_for_each_entry(cma_dev, &dev_list, list)
 262                 if (filter(cma_dev->device, cookie)) {
 263                         found_cma_dev = cma_dev;
 264                         break;
 265                 }
 266
 267         if (found_cma_dev)
 268                 cma_ref_dev(found_cma_dev);
 269         mutex_unlock(&lock);
 270         return found_cma_dev;
 271 }
 272
 273 int cma_get_default_gid_type(struct cma_device *cma_dev,
 274                              unsigned int port)
 275 {
 276         if (!rdma_is_port_valid(cma_dev->device, port))
 277                 return -EINVAL;
 278
 279         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 280 }
 281
 282 int cma_set_default_gid_type(struct cma_device *cma_dev,
 283                              unsigned int port,
 284                              enum ib_gid_type default_gid_type)
 285 {
 286         unsigned long supported_gids;
 287
 288         if (!rdma_is_port_valid(cma_dev->device, port))
 289                 return -EINVAL;
 290
 291         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 292
 293         if (!(supported_gids & 1 << default_gid_type))
 294                 return -EINVAL;
 295
 296         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 297                 default_gid_type;
 298
 299         return 0;
 300 }
 301
 302 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 303 {
 304         if (!rdma_is_port_valid(cma_dev->device, port))
 305                 return -EINVAL;
 306
 307         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 308 }
 309
 310 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 311                              u8 default_roce_tos)
 312 {
 313         if (!rdma_is_port_valid(cma_dev->device, port))
 314                 return -EINVAL;
 315
 316         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 317                  default_roce_tos;
 318
 319         return 0;
 320 }
 321 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 322 {
 323         return cma_dev->device;
 324 }
 325
 326 /*
 327  * Device removal can occur at anytime, so we need extra handling to
 328  * serialize notifying the user of device removal with other callbacks.
 329  * We do this by disabling removal notification while a callback is in process,
 330  * and reporting it after the callback completes.
 331  */
 332
 333 struct cma_multicast {
 334         struct rdma_id_private *id_priv;
 335         union {
 336                 struct ib_sa_multicast *ib;
 337         } multicast;
 338         struct list_head        list;
 339         void                    *context;
 340         struct sockaddr_storage addr;
 341         struct kref             mcref;
 342         bool                    igmp_joined;
 343         u8                      join_state;
 344 };
 345
 346 struct cma_work {
 347         struct work_struct      work;
 348         struct rdma_id_private  *id;
 349         enum rdma_cm_state      old_state;
 350         enum rdma_cm_state      new_state;
 351         struct rdma_cm_event    event;
 352 };
 353
 354 struct cma_ndev_work {
 355         struct work_struct      work;
 356         struct rdma_id_private  *id;
 357         struct rdma_cm_event    event;
 358 };
 359
 360 struct iboe_mcast_work {
 361         struct work_struct       work;
 362         struct rdma_id_private  *id;
 363         struct cma_multicast    *mc;
 364 };
 365
 366 union cma_ip_addr {
 367         struct in6_addr ip6;
 368         struct {
 369                 __be32 pad[3];
 370                 __be32 addr;
 371         } ip4;
 372 };
 373
 374 struct cma_hdr {
 375         u8 cma_version;
 376         u8 ip_version;  /* IP version: 7:4 */
 377         __be16 port;
 378         union cma_ip_addr src_addr;
 379         union cma_ip_addr dst_addr;
 380 };
 381
 382 #define CMA_VERSION 0x00
 383
 384 struct cma_req_info {
 385         struct ib_device *device;
 386         int port;
 387         union ib_gid local_gid;
 388         __be64 service_id;
 389         u16 pkey;
 390         bool has_gid:1;
 391 };
 392
 393 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 394 {
 395         unsigned long flags;
 396         int ret;
 397
 398         spin_lock_irqsave(&id_priv->lock, flags);
 399         ret = (id_priv->state == comp);
 400         spin_unlock_irqrestore(&id_priv->lock, flags);
 401         return ret;
 402 }
 403
 404 static int cma_comp_exch(struct rdma_id_private *id_priv,
 405                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 406 {
 407         unsigned long flags;
 408         int ret;
 409
 410         spin_lock_irqsave(&id_priv->lock, flags);
 411         if ((ret = (id_priv->state == comp)))
 412                 id_priv->state = exch;
 413         spin_unlock_irqrestore(&id_priv->lock, flags);
 414         return ret;
 415 }
 416
 417 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 418                                    enum rdma_cm_state exch)
 419 {
 420         unsigned long flags;
 421         enum rdma_cm_state old;
 422
 423         spin_lock_irqsave(&id_priv->lock, flags);
 424         old = id_priv->state;
 425         id_priv->state = exch;
 426         spin_unlock_irqrestore(&id_priv->lock, flags);
 427         return old;
 428 }
 429
 430 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 431 {
 432         return hdr->ip_version >> 4;
 433 }
 434
 435 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 436 {
 437         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 438 }
 439
 440 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 441 {
 442         struct in_device *in_dev = NULL;
 443
 444         if (ndev) {
 445                 rtnl_lock();
 446                 in_dev = __in_dev_get_rtnl(ndev);
 447                 if (in_dev) {
 448                         if (join)
 449                                 ip_mc_inc_group(in_dev,
 450                                                 *(__be32 *)(mgid->raw + 12));
 451                         else
 452                                 ip_mc_dec_group(in_dev,
 453                                                 *(__be32 *)(mgid->raw + 12));
 454                 }
 455                 rtnl_unlock();
 456         }
 457         return (in_dev) ? 0 : -ENODEV;
 458 }
 459
 460 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 461                                struct cma_device *cma_dev)
 462 {
 463         cma_ref_dev(cma_dev);
 464         id_priv->cma_dev = cma_dev;
 465         id_priv->gid_type = 0;
 466         id_priv->id.device = cma_dev->device;
 467         id_priv->id.route.addr.dev_addr.transport =
 468                 rdma_node_get_transport(cma_dev->device->node_type);
 469         list_add_tail(&id_priv->list, &cma_dev->id_list);
 470         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 471         rdma_restrack_add(&id_priv->res);
 472 }
 473
 474 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 475                               struct cma_device *cma_dev)
 476 {
 477         _cma_attach_to_dev(id_priv, cma_dev);
 478         id_priv->gid_type =
 479                 cma_dev->default_gid_type[id_priv->id.port_num -
 480                                           rdma_start_port(cma_dev->device)];
 481 }
 482
 483 void cma_deref_dev(struct cma_device *cma_dev)
 484 {
 485         if (atomic_dec_and_test(&cma_dev->refcount))
 486                 complete(&cma_dev->comp);
 487 }
 488
 489 static inline void release_mc(struct kref *kref)
 490 {
 491         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 492
 493         kfree(mc->multicast.ib);
 494         kfree(mc);
 495 }
 496
 497 static void cma_release_dev(struct rdma_id_private *id_priv)
 498 {
 499         mutex_lock(&lock);
 500         list_del(&id_priv->list);
 501         cma_deref_dev(id_priv->cma_dev);
 502         id_priv->cma_dev = NULL;
 503         mutex_unlock(&lock);
 504 }
 505
 506 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 507 {
 508         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 509 }
 510
 511 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 512 {
 513         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 514 }
 515
 516 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 517 {
 518         return id_priv->id.route.addr.src_addr.ss_family;
 519 }
 520
 521 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 522 {
 523         struct ib_sa_mcmember_rec rec;
 524         int ret = 0;
 525
 526         if (id_priv->qkey) {
 527                 if (qkey && id_priv->qkey != qkey)
 528                         return -EINVAL;
 529                 return 0;
 530         }
 531
 532         if (qkey) {
 533                 id_priv->qkey = qkey;
 534                 return 0;
 535         }
 536
 537         switch (id_priv->id.ps) {
 538         case RDMA_PS_UDP:
 539         case RDMA_PS_IB:
 540                 id_priv->qkey = RDMA_UDP_QKEY;
 541                 break;
 542         case RDMA_PS_IPOIB:
 543                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 544                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 545                                              id_priv->id.port_num, &rec.mgid,
 546                                              &rec);
 547                 if (!ret)
 548                         id_priv->qkey = be32_to_cpu(rec.qkey);
 549                 break;
 550         default:
 551                 break;
 552         }
 553         return ret;
 554 }
 555
 556 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 557 {
 558         dev_addr->dev_type = ARPHRD_INFINIBAND;
 559         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 560         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 561 }
 562
 563 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 564 {
 565         int ret;
 566
 567         if (addr->sa_family != AF_IB) {
 568                 ret = rdma_translate_ip(addr, dev_addr);
 569         } else {
 570                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 571                 ret = 0;
 572         }
 573
 574         return ret;
 575 }
 576
 577 static inline int cma_validate_port(struct ib_device *device, u8 port,
 578                                     enum ib_gid_type gid_type,
 579                                     union ib_gid *gid,
 580                                     struct rdma_id_private *id_priv)
 581 {
 582         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 583         int bound_if_index = dev_addr->bound_dev_if;
 584         int dev_type = dev_addr->dev_type;
 585         struct net_device *ndev = NULL;
 586         int ret = -ENODEV;
 587
 588         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 589                 return ret;
 590
 591         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 592                 return ret;
 593
 594         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 595                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 596                 if (!ndev)
 597                         return ret;
 598         } else {
 599                 gid_type = IB_GID_TYPE_IB;
 600         }
 601
 602         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 603                                          ndev, NULL);
 604
 605         if (ndev)
 606                 dev_put(ndev);
 607
 608         return ret;
 609 }
 610
 611 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 612                            struct rdma_id_private *listen_id_priv)
 613 {
 614         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 615         struct cma_device *cma_dev;
 616         union ib_gid gid, iboe_gid, *gidp;
 617         int ret = -ENODEV;
 618         u8 port;
 619
 620         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 621             id_priv->id.ps == RDMA_PS_IPOIB)
 622                 return -EINVAL;
 623
 624         mutex_lock(&lock);
 625         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 626                     &iboe_gid);
 627
 628         memcpy(&gid, dev_addr->src_dev_addr +
 629                rdma_addr_gid_offset(dev_addr), sizeof gid);
 630
 631         if (listen_id_priv) {
 632                 cma_dev = listen_id_priv->cma_dev;
 633                 port = listen_id_priv->id.port_num;
 634                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 635                        &iboe_gid : &gid;
 636
 637                 ret = cma_validate_port(cma_dev->device, port,
 638                                         rdma_protocol_ib(cma_dev->device, port) ?
 639                                         IB_GID_TYPE_IB :
 640                                         listen_id_priv->gid_type, gidp,
 641                                         id_priv);
 642                 if (!ret) {
 643                         id_priv->id.port_num = port;
 644                         goto out;
 645                 }
 646         }
 647
 648         list_for_each_entry(cma_dev, &dev_list, list) {
 649                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 650                         if (listen_id_priv &&
 651                             listen_id_priv->cma_dev == cma_dev &&
 652                             listen_id_priv->id.port_num == port)
 653                                 continue;
 654
 655                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 656                                &iboe_gid : &gid;
 657
 658                         ret = cma_validate_port(cma_dev->device, port,
 659                                                 rdma_protocol_ib(cma_dev->device, port) ?
 660                                                 IB_GID_TYPE_IB :
 661                                                 cma_dev->default_gid_type[port - 1],
 662                                                 gidp, id_priv);
 663                         if (!ret) {
 664                                 id_priv->id.port_num = port;
 665                                 goto out;
 666                         }
 667                 }
 668         }
 669
 670 out:
 671         if (!ret)
 672                 cma_attach_to_dev(id_priv, cma_dev);
 673
 674         mutex_unlock(&lock);
 675         return ret;
 676 }
 677
 678 /*
 679  * Select the source IB device and address to reach the destination IB address.
 680  */
 681 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 682 {
 683         struct cma_device *cma_dev, *cur_dev;
 684         struct sockaddr_ib *addr;
 685         union ib_gid gid, sgid, *dgid;
 686         u16 pkey, index;
 687         u8 p;
 688         enum ib_port_state port_state;
 689         int i;
 690
 691         cma_dev = NULL;
 692         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 693         dgid = (union ib_gid *) &addr->sib_addr;
 694         pkey = ntohs(addr->sib_pkey);
 695
 696         list_for_each_entry(cur_dev, &dev_list, list) {
 697                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 698                         if (!rdma_cap_af_ib(cur_dev->device, p))
 699                                 continue;
 700
 701                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 702                                 continue;
 703
 704                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 705                                 continue;
 706                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 707                                                        &gid, NULL);
 708                              i++) {
 709                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 710                                         cma_dev = cur_dev;
 711                                         sgid = gid;
 712                                         id_priv->id.port_num = p;
 713                                         goto found;
 714                                 }
 715
 716                                 if (!cma_dev && (gid.global.subnet_prefix ==
 717                                     dgid->global.subnet_prefix) &&
 718                                     port_state == IB_PORT_ACTIVE) {
 719                                         cma_dev = cur_dev;
 720                                         sgid = gid;
 721                                         id_priv->id.port_num = p;
 722                                 }
 723                         }
 724                 }
 725         }
 726
 727         if (!cma_dev)
 728                 return -ENODEV;
 729
 730 found:
 731         cma_attach_to_dev(id_priv, cma_dev);
 732         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 733         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 734         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 735         return 0;
 736 }
 737
 738 static void cma_deref_id(struct rdma_id_private *id_priv)
 739 {
 740         if (atomic_dec_and_test(&id_priv->refcount))
 741                 complete(&id_priv->comp);
 742 }
 743
 744 struct rdma_cm_id *__rdma_create_id(struct net *net,
 745                                     rdma_cm_event_handler event_handler,
 746                                     void *context, enum rdma_ucm_port_space ps,
 747                                     enum ib_qp_type qp_type, const char *caller)
 748 {
 749         struct rdma_id_private *id_priv;
 750
 751         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 752         if (!id_priv)
 753                 return ERR_PTR(-ENOMEM);
 754
 755         if (caller)
 756                 id_priv->res.kern_name = caller;
 757         else
 758                 rdma_restrack_set_task(&id_priv->res, current);
 759         id_priv->state = RDMA_CM_IDLE;
 760         id_priv->id.context = context;
 761         id_priv->id.event_handler = event_handler;
 762         id_priv->id.ps = ps;
 763         id_priv->id.qp_type = qp_type;
 764         id_priv->tos_set = false;
 765         spin_lock_init(&id_priv->lock);
 766         mutex_init(&id_priv->qp_mutex);
 767         init_completion(&id_priv->comp);
 768         atomic_set(&id_priv->refcount, 1);
 769         mutex_init(&id_priv->handler_mutex);
 770         INIT_LIST_HEAD(&id_priv->listen_list);
 771         INIT_LIST_HEAD(&id_priv->mc_list);
 772         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 773         id_priv->id.route.addr.dev_addr.net = get_net(net);
 774         id_priv->seq_num &= 0x00ffffff;
 775
 776         return &id_priv->id;
 777 }
 778 EXPORT_SYMBOL(__rdma_create_id);
 779
 780 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 781 {
 782         struct ib_qp_attr qp_attr;
 783         int qp_attr_mask, ret;
 784
 785         qp_attr.qp_state = IB_QPS_INIT;
 786         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 787         if (ret)
 788                 return ret;
 789
 790         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 791         if (ret)
 792                 return ret;
 793
 794         qp_attr.qp_state = IB_QPS_RTR;
 795         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 796         if (ret)
 797                 return ret;
 798
 799         qp_attr.qp_state = IB_QPS_RTS;
 800         qp_attr.sq_psn = 0;
 801         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 802
 803         return ret;
 804 }
 805
 806 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 807 {
 808         struct ib_qp_attr qp_attr;
 809         int qp_attr_mask, ret;
 810
 811         qp_attr.qp_state = IB_QPS_INIT;
 812         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 813         if (ret)
 814                 return ret;
 815
 816         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 817 }
 818
 819 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 820                    struct ib_qp_init_attr *qp_init_attr)
 821 {
 822         struct rdma_id_private *id_priv;
 823         struct ib_qp *qp;
 824         int ret;
 825
 826         id_priv = container_of(id, struct rdma_id_private, id);
 827         if (id->device != pd->device)
 828                 return -EINVAL;
 829
 830         qp_init_attr->port_num = id->port_num;
 831         qp = ib_create_qp(pd, qp_init_attr);
 832         if (IS_ERR(qp))
 833                 return PTR_ERR(qp);
 834
 835         if (id->qp_type == IB_QPT_UD)
 836                 ret = cma_init_ud_qp(id_priv, qp);
 837         else
 838                 ret = cma_init_conn_qp(id_priv, qp);
 839         if (ret)
 840                 goto err;
 841
 842         id->qp = qp;
 843         id_priv->qp_num = qp->qp_num;
 844         id_priv->srq = (qp->srq != NULL);
 845         return 0;
 846 err:
 847         ib_destroy_qp(qp);
 848         return ret;
 849 }
 850 EXPORT_SYMBOL(rdma_create_qp);
 851
 852 void rdma_destroy_qp(struct rdma_cm_id *id)
 853 {
 854         struct rdma_id_private *id_priv;
 855
 856         id_priv = container_of(id, struct rdma_id_private, id);
 857         mutex_lock(&id_priv->qp_mutex);
 858         ib_destroy_qp(id_priv->id.qp);
 859         id_priv->id.qp = NULL;
 860         mutex_unlock(&id_priv->qp_mutex);
 861 }
 862 EXPORT_SYMBOL(rdma_destroy_qp);
 863
 864 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 865                              struct rdma_conn_param *conn_param)
 866 {
 867         struct ib_qp_attr qp_attr;
 868         int qp_attr_mask, ret;
 869         union ib_gid sgid;
 870
 871         mutex_lock(&id_priv->qp_mutex);
 872         if (!id_priv->id.qp) {
 873                 ret = 0;
 874                 goto out;
 875         }
 876
 877         /* Need to update QP attributes from default values. */
 878         qp_attr.qp_state = IB_QPS_INIT;
 879         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 880         if (ret)
 881                 goto out;
 882
 883         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 884         if (ret)
 885                 goto out;
 886
 887         qp_attr.qp_state = IB_QPS_RTR;
 888         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 889         if (ret)
 890                 goto out;
 891
 892         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 893                            rdma_ah_read_grh(&qp_attr.ah_attr)->sgid_index,
 894                            &sgid, NULL);
 895         if (ret)
 896                 goto out;
 897
 898         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 899
 900         if (conn_param)
 901                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 902         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 903 out:
 904         mutex_unlock(&id_priv->qp_mutex);
 905         return ret;
 906 }
 907
 908 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 909                              struct rdma_conn_param *conn_param)
 910 {
 911         struct ib_qp_attr qp_attr;
 912         int qp_attr_mask, ret;
 913
 914         mutex_lock(&id_priv->qp_mutex);
 915         if (!id_priv->id.qp) {
 916                 ret = 0;
 917                 goto out;
 918         }
 919
 920         qp_attr.qp_state = IB_QPS_RTS;
 921         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 922         if (ret)
 923                 goto out;
 924
 925         if (conn_param)
 926                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 927         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 928 out:
 929         mutex_unlock(&id_priv->qp_mutex);
 930         return ret;
 931 }
 932
 933 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 934 {
 935         struct ib_qp_attr qp_attr;
 936         int ret;
 937
 938         mutex_lock(&id_priv->qp_mutex);
 939         if (!id_priv->id.qp) {
 940                 ret = 0;
 941                 goto out;
 942         }
 943
 944         qp_attr.qp_state = IB_QPS_ERR;
 945         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 946 out:
 947         mutex_unlock(&id_priv->qp_mutex);
 948         return ret;
 949 }
 950
 951 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 952                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 953 {
 954         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 955         int ret;
 956         u16 pkey;
 957
 958         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 959                 pkey = 0xffff;
 960         else
 961                 pkey = ib_addr_get_pkey(dev_addr);
 962
 963         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 964                                   pkey, &qp_attr->pkey_index);
 965         if (ret)
 966                 return ret;
 967
 968         qp_attr->port_num = id_priv->id.port_num;
 969         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 970
 971         if (id_priv->id.qp_type == IB_QPT_UD) {
 972                 ret = cma_set_qkey(id_priv, 0);
 973                 if (ret)
 974                         return ret;
 975
 976                 qp_attr->qkey = id_priv->qkey;
 977                 *qp_attr_mask |= IB_QP_QKEY;
 978         } else {
 979                 qp_attr->qp_access_flags = 0;
 980                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
 981         }
 982         return 0;
 983 }
 984
 985 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
 986                        int *qp_attr_mask)
 987 {
 988         struct rdma_id_private *id_priv;
 989         int ret = 0;
 990
 991         id_priv = container_of(id, struct rdma_id_private, id);
 992         if (rdma_cap_ib_cm(id->device, id->port_num)) {
 993                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
 994                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
 995                 else
 996                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
 997                                                  qp_attr_mask);
 998
 999                 if (qp_attr->qp_state == IB_QPS_RTR)
1000                         qp_attr->rq_psn = id_priv->seq_num;
1001         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1002                 if (!id_priv->cm_id.iw) {
1003                         qp_attr->qp_access_flags = 0;
1004                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1005                 } else
1006                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1007                                                  qp_attr_mask);
1008                 qp_attr->port_num = id_priv->id.port_num;
1009                 *qp_attr_mask |= IB_QP_PORT;
1010         } else
1011                 ret = -ENOSYS;
1012
1013         return ret;
1014 }
1015 EXPORT_SYMBOL(rdma_init_qp_attr);
1016
1017 static inline int cma_zero_addr(struct sockaddr *addr)
1018 {
1019         switch (addr->sa_family) {
1020         case AF_INET:
1021                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1022         case AF_INET6:
1023                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1024         case AF_IB:
1025                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1026         default:
1027                 return 0;
1028         }
1029 }
1030
1031 static inline int cma_loopback_addr(struct sockaddr *addr)
1032 {
1033         switch (addr->sa_family) {
1034         case AF_INET:
1035                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1036         case AF_INET6:
1037                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1038         case AF_IB:
1039                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1040         default:
1041                 return 0;
1042         }
1043 }
1044
1045 static inline int cma_any_addr(struct sockaddr *addr)
1046 {
1047         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1048 }
1049
1050 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1051 {
1052         if (src->sa_family != dst->sa_family)
1053                 return -1;
1054
1055         switch (src->sa_family) {
1056         case AF_INET:
1057                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1058                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1059         case AF_INET6:
1060                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1061                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1062         default:
1063                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1064                                    &((struct sockaddr_ib *) dst)->sib_addr);
1065         }
1066 }
1067
1068 static __be16 cma_port(struct sockaddr *addr)
1069 {
1070         struct sockaddr_ib *sib;
1071
1072         switch (addr->sa_family) {
1073         case AF_INET:
1074                 return ((struct sockaddr_in *) addr)->sin_port;
1075         case AF_INET6:
1076                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1077         case AF_IB:
1078                 sib = (struct sockaddr_ib *) addr;
1079                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1080                                     be64_to_cpu(sib->sib_sid_mask)));
1081         default:
1082                 return 0;
1083         }
1084 }
1085
1086 static inline int cma_any_port(struct sockaddr *addr)
1087 {
1088         return !cma_port(addr);
1089 }
1090
1091 static void cma_save_ib_info(struct sockaddr *src_addr,
1092                              struct sockaddr *dst_addr,
1093                              struct rdma_cm_id *listen_id,
1094                              struct sa_path_rec *path)
1095 {
1096         struct sockaddr_ib *listen_ib, *ib;
1097
1098         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1099         if (src_addr) {
1100                 ib = (struct sockaddr_ib *)src_addr;
1101                 ib->sib_family = AF_IB;
1102                 if (path) {
1103                         ib->sib_pkey = path->pkey;
1104                         ib->sib_flowinfo = path->flow_label;
1105                         memcpy(&ib->sib_addr, &path->sgid, 16);
1106                         ib->sib_sid = path->service_id;
1107                         ib->sib_scope_id = 0;
1108                 } else {
1109                         ib->sib_pkey = listen_ib->sib_pkey;
1110                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1111                         ib->sib_addr = listen_ib->sib_addr;
1112                         ib->sib_sid = listen_ib->sib_sid;
1113                         ib->sib_scope_id = listen_ib->sib_scope_id;
1114                 }
1115                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1116         }
1117         if (dst_addr) {
1118                 ib = (struct sockaddr_ib *)dst_addr;
1119                 ib->sib_family = AF_IB;
1120                 if (path) {
1121                         ib->sib_pkey = path->pkey;
1122                         ib->sib_flowinfo = path->flow_label;
1123                         memcpy(&ib->sib_addr, &path->dgid, 16);
1124                 }
1125         }
1126 }
1127
1128 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1129                               struct sockaddr_in *dst_addr,
1130                               struct cma_hdr *hdr,
1131                               __be16 local_port)
1132 {
1133         if (src_addr) {
1134                 *src_addr = (struct sockaddr_in) {
1135                         .sin_family = AF_INET,
1136                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1137                         .sin_port = local_port,
1138                 };
1139         }
1140
1141         if (dst_addr) {
1142                 *dst_addr = (struct sockaddr_in) {
1143                         .sin_family = AF_INET,
1144                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1145                         .sin_port = hdr->port,
1146                 };
1147         }
1148 }
1149
1150 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1151                               struct sockaddr_in6 *dst_addr,
1152                               struct cma_hdr *hdr,
1153                               __be16 local_port)
1154 {
1155         if (src_addr) {
1156                 *src_addr = (struct sockaddr_in6) {
1157                         .sin6_family = AF_INET6,
1158                         .sin6_addr = hdr->dst_addr.ip6,
1159                         .sin6_port = local_port,
1160                 };
1161         }
1162
1163         if (dst_addr) {
1164                 *dst_addr = (struct sockaddr_in6) {
1165                         .sin6_family = AF_INET6,
1166                         .sin6_addr = hdr->src_addr.ip6,
1167                         .sin6_port = hdr->port,
1168                 };
1169         }
1170 }
1171
1172 static u16 cma_port_from_service_id(__be64 service_id)
1173 {
1174         return (u16)be64_to_cpu(service_id);
1175 }
1176
1177 static int cma_save_ip_info(struct sockaddr *src_addr,
1178                             struct sockaddr *dst_addr,
1179                             struct ib_cm_event *ib_event,
1180                             __be64 service_id)
1181 {
1182         struct cma_hdr *hdr;
1183         __be16 port;
1184
1185         hdr = ib_event->private_data;
1186         if (hdr->cma_version != CMA_VERSION)
1187                 return -EINVAL;
1188
1189         port = htons(cma_port_from_service_id(service_id));
1190
1191         switch (cma_get_ip_ver(hdr)) {
1192         case 4:
1193                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1194                                   (struct sockaddr_in *)dst_addr, hdr, port);
1195                 break;
1196         case 6:
1197                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1198                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1199                 break;
1200         default:
1201                 return -EAFNOSUPPORT;
1202         }
1203
1204         return 0;
1205 }
1206
1207 static int cma_save_net_info(struct sockaddr *src_addr,
1208                              struct sockaddr *dst_addr,
1209                              struct rdma_cm_id *listen_id,
1210                              struct ib_cm_event *ib_event,
1211                              sa_family_t sa_family, __be64 service_id)
1212 {
1213         if (sa_family == AF_IB) {
1214                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1215                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1216                                          ib_event->param.req_rcvd.primary_path);
1217                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1218                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1219                 return 0;
1220         }
1221
1222         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1223 }
1224
1225 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1226                              struct cma_req_info *req)
1227 {
1228         const struct ib_cm_req_event_param *req_param =
1229                 &ib_event->param.req_rcvd;
1230         const struct ib_cm_sidr_req_event_param *sidr_param =
1231                 &ib_event->param.sidr_req_rcvd;
1232
1233         switch (ib_event->event) {
1234         case IB_CM_REQ_RECEIVED:
1235                 req->device     = req_param->listen_id->device;
1236                 req->port       = req_param->port;
1237                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1238                        sizeof(req->local_gid));
1239                 req->has_gid    = true;
1240                 req->service_id = req_param->primary_path->service_id;
1241                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1242                 if (req->pkey != req_param->bth_pkey)
1243                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1244                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1245                                             req_param->bth_pkey, req->pkey);
1246                 break;
1247         case IB_CM_SIDR_REQ_RECEIVED:
1248                 req->device     = sidr_param->listen_id->device;
1249                 req->port       = sidr_param->port;
1250                 req->has_gid    = false;
1251                 req->service_id = sidr_param->service_id;
1252                 req->pkey       = sidr_param->pkey;
1253                 if (req->pkey != sidr_param->bth_pkey)
1254                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1255                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1256                                             sidr_param->bth_pkey, req->pkey);
1257                 break;
1258         default:
1259                 return -EINVAL;
1260         }
1261
1262         return 0;
1263 }
1264
1265 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1266                                   const struct sockaddr_in *dst_addr,
1267                                   const struct sockaddr_in *src_addr)
1268 {
1269         __be32 daddr = dst_addr->sin_addr.s_addr,
1270                saddr = src_addr->sin_addr.s_addr;
1271         struct fib_result res;
1272         struct flowi4 fl4;
1273         int err;
1274         bool ret;
1275
1276         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1277             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1278             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1279             ipv4_is_loopback(saddr))
1280                 return false;
1281
1282         memset(&fl4, 0, sizeof(fl4));
1283         fl4.flowi4_iif = net_dev->ifindex;
1284         fl4.daddr = daddr;
1285         fl4.saddr = saddr;
1286
1287         rcu_read_lock();
1288         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1289         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1290         rcu_read_unlock();
1291
1292         return ret;
1293 }
1294
1295 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1296                                   const struct sockaddr_in6 *dst_addr,
1297                                   const struct sockaddr_in6 *src_addr)
1298 {
1299 #if IS_ENABLED(CONFIG_IPV6)
1300         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1301                            IPV6_ADDR_LINKLOCAL;
1302         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1303                                          &src_addr->sin6_addr, net_dev->ifindex,
1304                                          NULL, strict);
1305         bool ret;
1306
1307         if (!rt)
1308                 return false;
1309
1310         ret = rt->rt6i_idev->dev == net_dev;
1311         ip6_rt_put(rt);
1312
1313         return ret;
1314 #else
1315         return false;
1316 #endif
1317 }
1318
1319 static bool validate_net_dev(struct net_device *net_dev,
1320                              const struct sockaddr *daddr,
1321                              const struct sockaddr *saddr)
1322 {
1323         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1324         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1325         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1326         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1327
1328         switch (daddr->sa_family) {
1329         case AF_INET:
1330                 return saddr->sa_family == AF_INET &&
1331                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1332
1333         case AF_INET6:
1334                 return saddr->sa_family == AF_INET6 &&
1335                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1336
1337         default:
1338                 return false;
1339         }
1340 }
1341
1342 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1343                                           const struct cma_req_info *req)
1344 {
1345         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1346         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1347                         *src_addr = (struct sockaddr *)&src_addr_storage;
1348         struct net_device *net_dev;
1349         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1350         int err;
1351
1352         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1353                                req->service_id);
1354         if (err)
1355                 return ERR_PTR(err);
1356
1357         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1358                                            gid, listen_addr);
1359         if (!net_dev)
1360                 return ERR_PTR(-ENODEV);
1361
1362         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1363                 dev_put(net_dev);
1364                 return ERR_PTR(-EHOSTUNREACH);
1365         }
1366
1367         return net_dev;
1368 }
1369
1370 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1371 {
1372         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1373 }
1374
1375 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1376                                    const struct cma_hdr *hdr)
1377 {
1378         struct sockaddr *addr = cma_src_addr(id_priv);
1379         __be32 ip4_addr;
1380         struct in6_addr ip6_addr;
1381
1382         if (cma_any_addr(addr) && !id_priv->afonly)
1383                 return true;
1384
1385         switch (addr->sa_family) {
1386         case AF_INET:
1387                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1388                 if (cma_get_ip_ver(hdr) != 4)
1389                         return false;
1390                 if (!cma_any_addr(addr) &&
1391                     hdr->dst_addr.ip4.addr != ip4_addr)
1392                         return false;
1393                 break;
1394         case AF_INET6:
1395                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1396                 if (cma_get_ip_ver(hdr) != 6)
1397                         return false;
1398                 if (!cma_any_addr(addr) &&
1399                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1400                         return false;
1401                 break;
1402         case AF_IB:
1403                 return true;
1404         default:
1405                 return false;
1406         }
1407
1408         return true;
1409 }
1410
1411 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1412 {
1413         struct ib_device *device = id->device;
1414         const int port_num = id->port_num ?: rdma_start_port(device);
1415
1416         return rdma_protocol_roce(device, port_num);
1417 }
1418
1419 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1420                               const struct net_device *net_dev,
1421                               u8 port_num)
1422 {
1423         const struct rdma_addr *addr = &id->route.addr;
1424
1425         if (!net_dev)
1426                 /* This request is an AF_IB request or a RoCE request */
1427                 return (!id->port_num || id->port_num == port_num) &&
1428                        (addr->src_addr.ss_family == AF_IB ||
1429                         rdma_protocol_roce(id->device, port_num));
1430
1431         return !addr->dev_addr.bound_dev_if ||
1432                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1433                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1434 }
1435
1436 static struct rdma_id_private *cma_find_listener(
1437                 const struct rdma_bind_list *bind_list,
1438                 const struct ib_cm_id *cm_id,
1439                 const struct ib_cm_event *ib_event,
1440                 const struct cma_req_info *req,
1441                 const struct net_device *net_dev)
1442 {
1443         struct rdma_id_private *id_priv, *id_priv_dev;
1444
1445         if (!bind_list)
1446                 return ERR_PTR(-EINVAL);
1447
1448         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1449                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1450                         if (id_priv->id.device == cm_id->device &&
1451                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1452                                 return id_priv;
1453                         list_for_each_entry(id_priv_dev,
1454                                             &id_priv->listen_list,
1455                                             listen_list) {
1456                                 if (id_priv_dev->id.device == cm_id->device &&
1457                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1458                                         return id_priv_dev;
1459                         }
1460                 }
1461         }
1462
1463         return ERR_PTR(-EINVAL);
1464 }
1465
1466 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1467                                                  struct ib_cm_event *ib_event,
1468                                                  struct net_device **net_dev)
1469 {
1470         struct cma_req_info req;
1471         struct rdma_bind_list *bind_list;
1472         struct rdma_id_private *id_priv;
1473         int err;
1474
1475         err = cma_save_req_info(ib_event, &req);
1476         if (err)
1477                 return ERR_PTR(err);
1478
1479         *net_dev = cma_get_net_dev(ib_event, &req);
1480         if (IS_ERR(*net_dev)) {
1481                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1482                         /* Assuming the protocol is AF_IB */
1483                         *net_dev = NULL;
1484                 } else if (rdma_protocol_roce(req.device, req.port)) {
1485                         /* TODO find the net dev matching the request parameters
1486                          * through the RoCE GID table */
1487                         *net_dev = NULL;
1488                 } else {
1489                         return ERR_CAST(*net_dev);
1490                 }
1491         }
1492
1493         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1494                                 rdma_ps_from_service_id(req.service_id),
1495                                 cma_port_from_service_id(req.service_id));
1496         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1497         if (IS_ERR(id_priv) && *net_dev) {
1498                 dev_put(*net_dev);
1499                 *net_dev = NULL;
1500         }
1501
1502         return id_priv;
1503 }
1504
1505 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1506 {
1507         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1508 }
1509
1510 static void cma_cancel_route(struct rdma_id_private *id_priv)
1511 {
1512         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1513                 if (id_priv->query)
1514                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1515         }
1516 }
1517
1518 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1519 {
1520         struct rdma_id_private *dev_id_priv;
1521
1522         /*
1523          * Remove from listen_any_list to prevent added devices from spawning
1524          * additional listen requests.
1525          */
1526         mutex_lock(&lock);
1527         list_del(&id_priv->list);
1528
1529         while (!list_empty(&id_priv->listen_list)) {
1530                 dev_id_priv = list_entry(id_priv->listen_list.next,
1531                                          struct rdma_id_private, listen_list);
1532                 /* sync with device removal to avoid duplicate destruction */
1533                 list_del_init(&dev_id_priv->list);
1534                 list_del(&dev_id_priv->listen_list);
1535                 mutex_unlock(&lock);
1536
1537                 rdma_destroy_id(&dev_id_priv->id);
1538                 mutex_lock(&lock);
1539         }
1540         mutex_unlock(&lock);
1541 }
1542
1543 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1544                                  enum rdma_cm_state state)
1545 {
1546         switch (state) {
1547         case RDMA_CM_ADDR_QUERY:
1548                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1549                 break;
1550         case RDMA_CM_ROUTE_QUERY:
1551                 cma_cancel_route(id_priv);
1552                 break;
1553         case RDMA_CM_LISTEN:
1554                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1555                         cma_cancel_listens(id_priv);
1556                 break;
1557         default:
1558                 break;
1559         }
1560 }
1561
1562 static void cma_release_port(struct rdma_id_private *id_priv)
1563 {
1564         struct rdma_bind_list *bind_list = id_priv->bind_list;
1565         struct net *net = id_priv->id.route.addr.dev_addr.net;
1566
1567         if (!bind_list)
1568                 return;
1569
1570         mutex_lock(&lock);
1571         hlist_del(&id_priv->node);
1572         if (hlist_empty(&bind_list->owners)) {
1573                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1574                 kfree(bind_list);
1575         }
1576         mutex_unlock(&lock);
1577 }
1578
1579 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1580 {
1581         struct cma_multicast *mc;
1582
1583         while (!list_empty(&id_priv->mc_list)) {
1584                 mc = container_of(id_priv->mc_list.next,
1585                                   struct cma_multicast, list);
1586                 list_del(&mc->list);
1587                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1588                                       id_priv->id.port_num)) {
1589                         ib_sa_free_multicast(mc->multicast.ib);
1590                         kfree(mc);
1591                 } else {
1592                         if (mc->igmp_joined) {
1593                                 struct rdma_dev_addr *dev_addr =
1594                                         &id_priv->id.route.addr.dev_addr;
1595                                 struct net_device *ndev = NULL;
1596
1597                                 if (dev_addr->bound_dev_if)
1598                                         ndev = dev_get_by_index(&init_net,
1599                                                                 dev_addr->bound_dev_if);
1600                                 if (ndev) {
1601                                         cma_igmp_send(ndev,
1602                                                       &mc->multicast.ib->rec.mgid,
1603                                                       false);
1604                                         dev_put(ndev);
1605                                 }
1606                         }
1607                         kref_put(&mc->mcref, release_mc);
1608                 }
1609         }
1610 }
1611
1612 void rdma_destroy_id(struct rdma_cm_id *id)
1613 {
1614         struct rdma_id_private *id_priv;
1615         enum rdma_cm_state state;
1616
1617         id_priv = container_of(id, struct rdma_id_private, id);
1618         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1619         cma_cancel_operation(id_priv, state);
1620
1621         /*
1622          * Wait for any active callback to finish.  New callbacks will find
1623          * the id_priv state set to destroying and abort.
1624          */
1625         mutex_lock(&id_priv->handler_mutex);
1626         mutex_unlock(&id_priv->handler_mutex);
1627
1628         if (id_priv->cma_dev) {
1629                 rdma_restrack_del(&id_priv->res);
1630                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1631                         if (id_priv->cm_id.ib)
1632                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1633                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1634                         if (id_priv->cm_id.iw)
1635                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1636                 }
1637                 cma_leave_mc_groups(id_priv);
1638                 cma_release_dev(id_priv);
1639         }
1640
1641         cma_release_port(id_priv);
1642         cma_deref_id(id_priv);
1643         wait_for_completion(&id_priv->comp);
1644
1645         if (id_priv->internal_id)
1646                 cma_deref_id(id_priv->id.context);
1647
1648         kfree(id_priv->id.route.path_rec);
1649         put_net(id_priv->id.route.addr.dev_addr.net);
1650         kfree(id_priv);
1651 }
1652 EXPORT_SYMBOL(rdma_destroy_id);
1653
1654 static int cma_rep_recv(struct rdma_id_private *id_priv)
1655 {
1656         int ret;
1657
1658         ret = cma_modify_qp_rtr(id_priv, NULL);
1659         if (ret)
1660                 goto reject;
1661
1662         ret = cma_modify_qp_rts(id_priv, NULL);
1663         if (ret)
1664                 goto reject;
1665
1666         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1667         if (ret)
1668                 goto reject;
1669
1670         return 0;
1671 reject:
1672         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1673         cma_modify_qp_err(id_priv);
1674         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1675                        NULL, 0, NULL, 0);
1676         return ret;
1677 }
1678
1679 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1680                                    struct ib_cm_rep_event_param *rep_data,
1681                                    void *private_data)
1682 {
1683         event->param.conn.private_data = private_data;
1684         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1685         event->param.conn.responder_resources = rep_data->responder_resources;
1686         event->param.conn.initiator_depth = rep_data->initiator_depth;
1687         event->param.conn.flow_control = rep_data->flow_control;
1688         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1689         event->param.conn.srq = rep_data->srq;
1690         event->param.conn.qp_num = rep_data->remote_qpn;
1691 }
1692
1693 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1694 {
1695         struct rdma_id_private *id_priv = cm_id->context;
1696         struct rdma_cm_event event;
1697         int ret = 0;
1698
1699         mutex_lock(&id_priv->handler_mutex);
1700         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1701              id_priv->state != RDMA_CM_CONNECT) ||
1702             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1703              id_priv->state != RDMA_CM_DISCONNECT))
1704                 goto out;
1705
1706         memset(&event, 0, sizeof event);
1707         switch (ib_event->event) {
1708         case IB_CM_REQ_ERROR:
1709         case IB_CM_REP_ERROR:
1710                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1711                 event.status = -ETIMEDOUT;
1712                 break;
1713         case IB_CM_REP_RECEIVED:
1714                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1715                     (id_priv->id.qp_type != IB_QPT_UD))
1716                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1717                 if (id_priv->id.qp) {
1718                         event.status = cma_rep_recv(id_priv);
1719                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1720                                                      RDMA_CM_EVENT_ESTABLISHED;
1721                 } else {
1722                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1723                 }
1724                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1725                                        ib_event->private_data);
1726                 break;
1727         case IB_CM_RTU_RECEIVED:
1728         case IB_CM_USER_ESTABLISHED:
1729                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1730                 break;
1731         case IB_CM_DREQ_ERROR:
1732                 event.status = -ETIMEDOUT; /* fall through */
1733         case IB_CM_DREQ_RECEIVED:
1734         case IB_CM_DREP_RECEIVED:
1735                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1736                                    RDMA_CM_DISCONNECT))
1737                         goto out;
1738                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1739                 break;
1740         case IB_CM_TIMEWAIT_EXIT:
1741                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1742                 break;
1743         case IB_CM_MRA_RECEIVED:
1744                 /* ignore event */
1745                 goto out;
1746         case IB_CM_REJ_RECEIVED:
1747                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1748                                                                                 ib_event->param.rej_rcvd.reason));
1749                 cma_modify_qp_err(id_priv);
1750                 event.status = ib_event->param.rej_rcvd.reason;
1751                 event.event = RDMA_CM_EVENT_REJECTED;
1752                 event.param.conn.private_data = ib_event->private_data;
1753                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1754                 break;
1755         default:
1756                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1757                        ib_event->event);
1758                 goto out;
1759         }
1760
1761         ret = id_priv->id.event_handler(&id_priv->id, &event);
1762         if (ret) {
1763                 /* Destroy the CM ID by returning a non-zero value. */
1764                 id_priv->cm_id.ib = NULL;
1765                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1766                 mutex_unlock(&id_priv->handler_mutex);
1767                 rdma_destroy_id(&id_priv->id);
1768                 return ret;
1769         }
1770 out:
1771         mutex_unlock(&id_priv->handler_mutex);
1772         return ret;
1773 }
1774
1775 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1776                                                struct ib_cm_event *ib_event,
1777                                                struct net_device *net_dev)
1778 {
1779         struct rdma_id_private *listen_id_priv;
1780         struct rdma_id_private *id_priv;
1781         struct rdma_cm_id *id;
1782         struct rdma_route *rt;
1783         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1784         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1785         const __be64 service_id =
1786                 ib_event->param.req_rcvd.primary_path->service_id;
1787         int ret;
1788
1789         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1790         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1791                             listen_id->event_handler, listen_id->context,
1792                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1793                             listen_id_priv->res.kern_name);
1794         if (IS_ERR(id))
1795                 return NULL;
1796
1797         id_priv = container_of(id, struct rdma_id_private, id);
1798         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1799                               (struct sockaddr *)&id->route.addr.dst_addr,
1800                               listen_id, ib_event, ss_family, service_id))
1801                 goto err;
1802
1803         rt = &id->route;
1804         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1805         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1806                                GFP_KERNEL);
1807         if (!rt->path_rec)
1808                 goto err;
1809
1810         rt->path_rec[0] = *path;
1811         if (rt->num_paths == 2)
1812                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1813
1814         if (net_dev) {
1815                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1816         } else {
1817                 if (!cma_protocol_roce(listen_id) &&
1818                     cma_any_addr(cma_src_addr(id_priv))) {
1819                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1820                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1821                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1822                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1823                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1824                         if (ret)
1825                                 goto err;
1826                 }
1827         }
1828         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1829
1830         id_priv->state = RDMA_CM_CONNECT;
1831         return id_priv;
1832
1833 err:
1834         rdma_destroy_id(id);
1835         return NULL;
1836 }
1837
1838 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1839                                               struct ib_cm_event *ib_event,
1840                                               struct net_device *net_dev)
1841 {
1842         struct rdma_id_private *listen_id_priv;
1843         struct rdma_id_private *id_priv;
1844         struct rdma_cm_id *id;
1845         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1846         struct net *net = listen_id->route.addr.dev_addr.net;
1847         int ret;
1848
1849         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1850         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
1851                               listen_id->ps, IB_QPT_UD,
1852                               listen_id_priv->res.kern_name);
1853         if (IS_ERR(id))
1854                 return NULL;
1855
1856         id_priv = container_of(id, struct rdma_id_private, id);
1857         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1858                               (struct sockaddr *)&id->route.addr.dst_addr,
1859                               listen_id, ib_event, ss_family,
1860                               ib_event->param.sidr_req_rcvd.service_id))
1861                 goto err;
1862
1863         if (net_dev) {
1864                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1865         } else {
1866                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1867                         ret = cma_translate_addr(cma_src_addr(id_priv),
1868                                                  &id->route.addr.dev_addr);
1869                         if (ret)
1870                                 goto err;
1871                 }
1872         }
1873
1874         id_priv->state = RDMA_CM_CONNECT;
1875         return id_priv;
1876 err:
1877         rdma_destroy_id(id);
1878         return NULL;
1879 }
1880
1881 static void cma_set_req_event_data(struct rdma_cm_event *event,
1882                                    struct ib_cm_req_event_param *req_data,
1883                                    void *private_data, int offset)
1884 {
1885         event->param.conn.private_data = private_data + offset;
1886         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1887         event->param.conn.responder_resources = req_data->responder_resources;
1888         event->param.conn.initiator_depth = req_data->initiator_depth;
1889         event->param.conn.flow_control = req_data->flow_control;
1890         event->param.conn.retry_count = req_data->retry_count;
1891         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1892         event->param.conn.srq = req_data->srq;
1893         event->param.conn.qp_num = req_data->remote_qpn;
1894 }
1895
1896 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1897 {
1898         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1899                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1900                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1901                  (id->qp_type == IB_QPT_UD)) ||
1902                 (!id->qp_type));
1903 }
1904
1905 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1906 {
1907         struct rdma_id_private *listen_id, *conn_id = NULL;
1908         struct rdma_cm_event event;
1909         struct net_device *net_dev;
1910         u8 offset;
1911         int ret;
1912
1913         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1914         if (IS_ERR(listen_id))
1915                 return PTR_ERR(listen_id);
1916
1917         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1918                 ret = -EINVAL;
1919                 goto net_dev_put;
1920         }
1921
1922         mutex_lock(&listen_id->handler_mutex);
1923         if (listen_id->state != RDMA_CM_LISTEN) {
1924                 ret = -ECONNABORTED;
1925                 goto err1;
1926         }
1927
1928         memset(&event, 0, sizeof event);
1929         offset = cma_user_data_offset(listen_id);
1930         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1931         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1932                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1933                 event.param.ud.private_data = ib_event->private_data + offset;
1934                 event.param.ud.private_data_len =
1935                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1936         } else {
1937                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1938                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1939                                        ib_event->private_data, offset);
1940         }
1941         if (!conn_id) {
1942                 ret = -ENOMEM;
1943                 goto err1;
1944         }
1945
1946         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1947         ret = cma_acquire_dev(conn_id, listen_id);
1948         if (ret)
1949                 goto err2;
1950
1951         conn_id->cm_id.ib = cm_id;
1952         cm_id->context = conn_id;
1953         cm_id->cm_handler = cma_ib_handler;
1954
1955         /*
1956          * Protect against the user destroying conn_id from another thread
1957          * until we're done accessing it.
1958          */
1959         atomic_inc(&conn_id->refcount);
1960         ret = conn_id->id.event_handler(&conn_id->id, &event);
1961         if (ret)
1962                 goto err3;
1963         /*
1964          * Acquire mutex to prevent user executing rdma_destroy_id()
1965          * while we're accessing the cm_id.
1966          */
1967         mutex_lock(&lock);
1968         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
1969             (conn_id->id.qp_type != IB_QPT_UD))
1970                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1971         mutex_unlock(&lock);
1972         mutex_unlock(&conn_id->handler_mutex);
1973         mutex_unlock(&listen_id->handler_mutex);
1974         cma_deref_id(conn_id);
1975         if (net_dev)
1976                 dev_put(net_dev);
1977         return 0;
1978
1979 err3:
1980         cma_deref_id(conn_id);
1981         /* Destroy the CM ID by returning a non-zero value. */
1982         conn_id->cm_id.ib = NULL;
1983 err2:
1984         cma_exch(conn_id, RDMA_CM_DESTROYING);
1985         mutex_unlock(&conn_id->handler_mutex);
1986 err1:
1987         mutex_unlock(&listen_id->handler_mutex);
1988         if (conn_id)
1989                 rdma_destroy_id(&conn_id->id);
1990
1991 net_dev_put:
1992         if (net_dev)
1993                 dev_put(net_dev);
1994
1995         return ret;
1996 }
1997
1998 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
1999 {
2000         if (addr->sa_family == AF_IB)
2001                 return ((struct sockaddr_ib *) addr)->sib_sid;
2002
2003         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2004 }
2005 EXPORT_SYMBOL(rdma_get_service_id);
2006
2007 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2008                     union ib_gid *dgid)
2009 {
2010         struct rdma_addr *addr = &cm_id->route.addr;
2011
2012         if (!cm_id->device) {
2013                 if (sgid)
2014                         memset(sgid, 0, sizeof(*sgid));
2015                 if (dgid)
2016                         memset(dgid, 0, sizeof(*dgid));
2017                 return;
2018         }
2019
2020         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2021                 if (sgid)
2022                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2023                 if (dgid)
2024                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2025         } else {
2026                 if (sgid)
2027                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2028                 if (dgid)
2029                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2030         }
2031 }
2032 EXPORT_SYMBOL(rdma_read_gids);
2033
2034 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2035 {
2036         struct rdma_id_private *id_priv = iw_id->context;
2037         struct rdma_cm_event event;
2038         int ret = 0;
2039         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2040         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2041
2042         mutex_lock(&id_priv->handler_mutex);
2043         if (id_priv->state != RDMA_CM_CONNECT)
2044                 goto out;
2045
2046         memset(&event, 0, sizeof event);
2047         switch (iw_event->event) {
2048         case IW_CM_EVENT_CLOSE:
2049                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2050                 break;
2051         case IW_CM_EVENT_CONNECT_REPLY:
2052                 memcpy(cma_src_addr(id_priv), laddr,
2053                        rdma_addr_size(laddr));
2054                 memcpy(cma_dst_addr(id_priv), raddr,
2055                        rdma_addr_size(raddr));
2056                 switch (iw_event->status) {
2057                 case 0:
2058                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2059                         event.param.conn.initiator_depth = iw_event->ird;
2060                         event.param.conn.responder_resources = iw_event->ord;
2061                         break;
2062                 case -ECONNRESET:
2063                 case -ECONNREFUSED:
2064                         event.event = RDMA_CM_EVENT_REJECTED;
2065                         break;
2066                 case -ETIMEDOUT:
2067                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2068                         break;
2069                 default:
2070                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2071                         break;
2072                 }
2073                 break;
2074         case IW_CM_EVENT_ESTABLISHED:
2075                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2076                 event.param.conn.initiator_depth = iw_event->ird;
2077                 event.param.conn.responder_resources = iw_event->ord;
2078                 break;
2079         default:
2080                 BUG_ON(1);
2081         }
2082
2083         event.status = iw_event->status;
2084         event.param.conn.private_data = iw_event->private_data;
2085         event.param.conn.private_data_len = iw_event->private_data_len;
2086         ret = id_priv->id.event_handler(&id_priv->id, &event);
2087         if (ret) {
2088                 /* Destroy the CM ID by returning a non-zero value. */
2089                 id_priv->cm_id.iw = NULL;
2090                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2091                 mutex_unlock(&id_priv->handler_mutex);
2092                 rdma_destroy_id(&id_priv->id);
2093                 return ret;
2094         }
2095
2096 out:
2097         mutex_unlock(&id_priv->handler_mutex);
2098         return ret;
2099 }
2100
2101 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2102                                struct iw_cm_event *iw_event)
2103 {
2104         struct rdma_cm_id *new_cm_id;
2105         struct rdma_id_private *listen_id, *conn_id;
2106         struct rdma_cm_event event;
2107         int ret = -ECONNABORTED;
2108         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2109         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2110
2111         listen_id = cm_id->context;
2112
2113         mutex_lock(&listen_id->handler_mutex);
2114         if (listen_id->state != RDMA_CM_LISTEN)
2115                 goto out;
2116
2117         /* Create a new RDMA id for the new IW CM ID */
2118         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2119                                      listen_id->id.event_handler,
2120                                      listen_id->id.context,
2121                                      RDMA_PS_TCP, IB_QPT_RC,
2122                                      listen_id->res.kern_name);
2123         if (IS_ERR(new_cm_id)) {
2124                 ret = -ENOMEM;
2125                 goto out;
2126         }
2127         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2128         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2129         conn_id->state = RDMA_CM_CONNECT;
2130
2131         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2132         if (ret) {
2133                 mutex_unlock(&conn_id->handler_mutex);
2134                 rdma_destroy_id(new_cm_id);
2135                 goto out;
2136         }
2137
2138         ret = cma_acquire_dev(conn_id, listen_id);
2139         if (ret) {
2140                 mutex_unlock(&conn_id->handler_mutex);
2141                 rdma_destroy_id(new_cm_id);
2142                 goto out;
2143         }
2144
2145         conn_id->cm_id.iw = cm_id;
2146         cm_id->context = conn_id;
2147         cm_id->cm_handler = cma_iw_handler;
2148
2149         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2150         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2151
2152         memset(&event, 0, sizeof event);
2153         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2154         event.param.conn.private_data = iw_event->private_data;
2155         event.param.conn.private_data_len = iw_event->private_data_len;
2156         event.param.conn.initiator_depth = iw_event->ird;
2157         event.param.conn.responder_resources = iw_event->ord;
2158
2159         /*
2160          * Protect against the user destroying conn_id from another thread
2161          * until we're done accessing it.
2162          */
2163         atomic_inc(&conn_id->refcount);
2164         ret = conn_id->id.event_handler(&conn_id->id, &event);
2165         if (ret) {
2166                 /* User wants to destroy the CM ID */
2167                 conn_id->cm_id.iw = NULL;
2168                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2169                 mutex_unlock(&conn_id->handler_mutex);
2170                 cma_deref_id(conn_id);
2171                 rdma_destroy_id(&conn_id->id);
2172                 goto out;
2173         }
2174
2175         mutex_unlock(&conn_id->handler_mutex);
2176         cma_deref_id(conn_id);
2177
2178 out:
2179         mutex_unlock(&listen_id->handler_mutex);
2180         return ret;
2181 }
2182
2183 static int cma_ib_listen(struct rdma_id_private *id_priv)
2184 {
2185         struct sockaddr *addr;
2186         struct ib_cm_id *id;
2187         __be64 svc_id;
2188
2189         addr = cma_src_addr(id_priv);
2190         svc_id = rdma_get_service_id(&id_priv->id, addr);
2191         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2192         if (IS_ERR(id))
2193                 return PTR_ERR(id);
2194         id_priv->cm_id.ib = id;
2195
2196         return 0;
2197 }
2198
2199 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2200 {
2201         int ret;
2202         struct iw_cm_id *id;
2203
2204         id = iw_create_cm_id(id_priv->id.device,
2205                              iw_conn_req_handler,
2206                              id_priv);
2207         if (IS_ERR(id))
2208                 return PTR_ERR(id);
2209
2210         id->tos = id_priv->tos;
2211         id_priv->cm_id.iw = id;
2212
2213         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2214                rdma_addr_size(cma_src_addr(id_priv)));
2215
2216         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2217
2218         if (ret) {
2219                 iw_destroy_cm_id(id_priv->cm_id.iw);
2220                 id_priv->cm_id.iw = NULL;
2221         }
2222
2223         return ret;
2224 }
2225
2226 static int cma_listen_handler(struct rdma_cm_id *id,
2227                               struct rdma_cm_event *event)
2228 {
2229         struct rdma_id_private *id_priv = id->context;
2230
2231         id->context = id_priv->id.context;
2232         id->event_handler = id_priv->id.event_handler;
2233         return id_priv->id.event_handler(id, event);
2234 }
2235
2236 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2237                               struct cma_device *cma_dev)
2238 {
2239         struct rdma_id_private *dev_id_priv;
2240         struct rdma_cm_id *id;
2241         struct net *net = id_priv->id.route.addr.dev_addr.net;
2242         int ret;
2243
2244         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2245                 return;
2246
2247         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2248                               id_priv->id.qp_type, id_priv->res.kern_name);
2249         if (IS_ERR(id))
2250                 return;
2251
2252         dev_id_priv = container_of(id, struct rdma_id_private, id);
2253
2254         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2255         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2256                rdma_addr_size(cma_src_addr(id_priv)));
2257
2258         _cma_attach_to_dev(dev_id_priv, cma_dev);
2259         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2260         atomic_inc(&id_priv->refcount);
2261         dev_id_priv->internal_id = 1;
2262         dev_id_priv->afonly = id_priv->afonly;
2263
2264         ret = rdma_listen(id, id_priv->backlog);
2265         if (ret)
2266                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2267                         ret, cma_dev->device->name);
2268 }
2269
2270 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2271 {
2272         struct cma_device *cma_dev;
2273
2274         mutex_lock(&lock);
2275         list_add_tail(&id_priv->list, &listen_any_list);
2276         list_for_each_entry(cma_dev, &dev_list, list)
2277                 cma_listen_on_dev(id_priv, cma_dev);
2278         mutex_unlock(&lock);
2279 }
2280
2281 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2282 {
2283         struct rdma_id_private *id_priv;
2284
2285         id_priv = container_of(id, struct rdma_id_private, id);
2286         id_priv->tos = (u8) tos;
2287         id_priv->tos_set = true;
2288 }
2289 EXPORT_SYMBOL(rdma_set_service_type);
2290
2291 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2292                               void *context)
2293 {
2294         struct cma_work *work = context;
2295         struct rdma_route *route;
2296
2297         route = &work->id->id.route;
2298
2299         if (!status) {
2300                 route->num_paths = 1;
2301                 *route->path_rec = *path_rec;
2302         } else {
2303                 work->old_state = RDMA_CM_ROUTE_QUERY;
2304                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2305                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2306                 work->event.status = status;
2307                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2308                                      status);
2309         }
2310
2311         queue_work(cma_wq, &work->work);
2312 }
2313
2314 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2315                               struct cma_work *work)
2316 {
2317         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2318         struct sa_path_rec path_rec;
2319         ib_sa_comp_mask comp_mask;
2320         struct sockaddr_in6 *sin6;
2321         struct sockaddr_ib *sib;
2322
2323         memset(&path_rec, 0, sizeof path_rec);
2324
2325         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2326                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2327         else
2328                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2329         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2330         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2331         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2332         path_rec.numb_path = 1;
2333         path_rec.reversible = 1;
2334         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2335                                                   cma_dst_addr(id_priv));
2336
2337         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2338                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2339                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2340
2341         switch (cma_family(id_priv)) {
2342         case AF_INET:
2343                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2344                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2345                 break;
2346         case AF_INET6:
2347                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2348                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2349                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2350                 break;
2351         case AF_IB:
2352                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2353                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2354                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2355                 break;
2356         }
2357
2358         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2359                                                id_priv->id.port_num, &path_rec,
2360                                                comp_mask, timeout_ms,
2361                                                GFP_KERNEL, cma_query_handler,
2362                                                work, &id_priv->query);
2363
2364         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2365 }
2366
2367 static void cma_work_handler(struct work_struct *_work)
2368 {
2369         struct cma_work *work = container_of(_work, struct cma_work, work);
2370         struct rdma_id_private *id_priv = work->id;
2371         int destroy = 0;
2372
2373         mutex_lock(&id_priv->handler_mutex);
2374         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2375                 goto out;
2376
2377         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2378                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2379                 destroy = 1;
2380         }
2381 out:
2382         mutex_unlock(&id_priv->handler_mutex);
2383         cma_deref_id(id_priv);
2384         if (destroy)
2385                 rdma_destroy_id(&id_priv->id);
2386         kfree(work);
2387 }
2388
2389 static void cma_ndev_work_handler(struct work_struct *_work)
2390 {
2391         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2392         struct rdma_id_private *id_priv = work->id;
2393         int destroy = 0;
2394
2395         mutex_lock(&id_priv->handler_mutex);
2396         if (id_priv->state == RDMA_CM_DESTROYING ||
2397             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2398                 goto out;
2399
2400         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2401                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2402                 destroy = 1;
2403         }
2404
2405 out:
2406         mutex_unlock(&id_priv->handler_mutex);
2407         cma_deref_id(id_priv);
2408         if (destroy)
2409                 rdma_destroy_id(&id_priv->id);
2410         kfree(work);
2411 }
2412
2413 static void cma_init_resolve_route_work(struct cma_work *work,
2414                                         struct rdma_id_private *id_priv)
2415 {
2416         work->id = id_priv;
2417         INIT_WORK(&work->work, cma_work_handler);
2418         work->old_state = RDMA_CM_ROUTE_QUERY;
2419         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2420         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2421 }
2422
2423 static void cma_init_resolve_addr_work(struct cma_work *work,
2424                                        struct rdma_id_private *id_priv)
2425 {
2426         work->id = id_priv;
2427         INIT_WORK(&work->work, cma_work_handler);
2428         work->old_state = RDMA_CM_ADDR_QUERY;
2429         work->new_state = RDMA_CM_ADDR_RESOLVED;
2430         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2431 }
2432
2433 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2434 {
2435         struct rdma_route *route = &id_priv->id.route;
2436         struct cma_work *work;
2437         int ret;
2438
2439         work = kzalloc(sizeof *work, GFP_KERNEL);
2440         if (!work)
2441                 return -ENOMEM;
2442
2443         cma_init_resolve_route_work(work, id_priv);
2444
2445         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2446         if (!route->path_rec) {
2447                 ret = -ENOMEM;
2448                 goto err1;
2449         }
2450
2451         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2452         if (ret)
2453                 goto err2;
2454
2455         return 0;
2456 err2:
2457         kfree(route->path_rec);
2458         route->path_rec = NULL;
2459 err1:
2460         kfree(work);
2461         return ret;
2462 }
2463
2464 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2465                                            unsigned long supported_gids,
2466                                            enum ib_gid_type default_gid)
2467 {
2468         if ((network_type == RDMA_NETWORK_IPV4 ||
2469              network_type == RDMA_NETWORK_IPV6) &&
2470             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2471                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2472
2473         return default_gid;
2474 }
2475
2476 /*
2477  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2478  * path record type based on GID type.
2479  * It also sets up other L2 fields which includes destination mac address
2480  * netdev ifindex, of the path record.
2481  * It returns the netdev of the bound interface for this path record entry.
2482  */
2483 static struct net_device *
2484 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2485 {
2486         struct rdma_route *route = &id_priv->id.route;
2487         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2488         struct rdma_addr *addr = &route->addr;
2489         unsigned long supported_gids;
2490         struct net_device *ndev;
2491
2492         if (!addr->dev_addr.bound_dev_if)
2493                 return NULL;
2494
2495         ndev = dev_get_by_index(addr->dev_addr.net,
2496                                 addr->dev_addr.bound_dev_if);
2497         if (!ndev)
2498                 return NULL;
2499
2500         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2501                                                     id_priv->id.port_num);
2502         gid_type = cma_route_gid_type(addr->dev_addr.network,
2503                                       supported_gids,
2504                                       id_priv->gid_type);
2505         /* Use the hint from IP Stack to select GID Type */
2506         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2507                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2508         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2509
2510         route->path_rec->roce.route_resolved = true;
2511         sa_path_set_ndev(route->path_rec, addr->dev_addr.net);
2512         sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2513         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2514         return ndev;
2515 }
2516
2517 int rdma_set_ib_path(struct rdma_cm_id *id,
2518                      struct sa_path_rec *path_rec)
2519 {
2520         struct rdma_id_private *id_priv;
2521         struct net_device *ndev;
2522         int ret;
2523
2524         id_priv = container_of(id, struct rdma_id_private, id);
2525         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2526                            RDMA_CM_ROUTE_RESOLVED))
2527                 return -EINVAL;
2528
2529         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2530                                      GFP_KERNEL);
2531         if (!id->route.path_rec) {
2532                 ret = -ENOMEM;
2533                 goto err;
2534         }
2535
2536         if (rdma_protocol_roce(id->device, id->port_num)) {
2537                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2538                 if (!ndev) {
2539                         ret = -ENODEV;
2540                         goto err_free;
2541                 }
2542                 dev_put(ndev);
2543         }
2544
2545         id->route.num_paths = 1;
2546         return 0;
2547
2548 err_free:
2549         kfree(id->route.path_rec);
2550         id->route.path_rec = NULL;
2551 err:
2552         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2553         return ret;
2554 }
2555 EXPORT_SYMBOL(rdma_set_ib_path);
2556
2557 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2558 {
2559         struct cma_work *work;
2560
2561         work = kzalloc(sizeof *work, GFP_KERNEL);
2562         if (!work)
2563                 return -ENOMEM;
2564
2565         cma_init_resolve_route_work(work, id_priv);
2566         queue_work(cma_wq, &work->work);
2567         return 0;
2568 }
2569
2570 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2571 {
2572         int prio;
2573         struct net_device *dev;
2574
2575         prio = rt_tos2priority(tos);
2576         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2577         if (dev->num_tc)
2578                 return netdev_get_prio_tc_map(dev, prio);
2579
2580 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2581         if (is_vlan_dev(ndev))
2582                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2583                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2584 #endif
2585         return 0;
2586 }
2587
2588 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2589 {
2590         struct rdma_route *route = &id_priv->id.route;
2591         struct rdma_addr *addr = &route->addr;
2592         struct cma_work *work;
2593         int ret;
2594         struct net_device *ndev;
2595
2596         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2597                                         rdma_start_port(id_priv->cma_dev->device)];
2598         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2599
2600
2601         work = kzalloc(sizeof *work, GFP_KERNEL);
2602         if (!work)
2603                 return -ENOMEM;
2604
2605         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2606         if (!route->path_rec) {
2607                 ret = -ENOMEM;
2608                 goto err1;
2609         }
2610
2611         route->num_paths = 1;
2612
2613         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2614         if (!ndev) {
2615                 ret = -ENODEV;
2616                 goto err2;
2617         }
2618
2619         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2620                     &route->path_rec->sgid);
2621         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2622                     &route->path_rec->dgid);
2623
2624         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2625                 /* TODO: get the hoplimit from the inet/inet6 device */
2626                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2627         else
2628                 route->path_rec->hop_limit = 1;
2629         route->path_rec->reversible = 1;
2630         route->path_rec->pkey = cpu_to_be16(0xffff);
2631         route->path_rec->mtu_selector = IB_SA_EQ;
2632         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2633         route->path_rec->traffic_class = tos;
2634         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2635         route->path_rec->rate_selector = IB_SA_EQ;
2636         route->path_rec->rate = iboe_get_rate(ndev);
2637         dev_put(ndev);
2638         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2639         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2640         if (!route->path_rec->mtu) {
2641                 ret = -EINVAL;
2642                 goto err2;
2643         }
2644
2645         cma_init_resolve_route_work(work, id_priv);
2646         queue_work(cma_wq, &work->work);
2647
2648         return 0;
2649
2650 err2:
2651         kfree(route->path_rec);
2652         route->path_rec = NULL;
2653 err1:
2654         kfree(work);
2655         return ret;
2656 }
2657
2658 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2659 {
2660         struct rdma_id_private *id_priv;
2661         int ret;
2662
2663         id_priv = container_of(id, struct rdma_id_private, id);
2664         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2665                 return -EINVAL;
2666
2667         atomic_inc(&id_priv->refcount);
2668         if (rdma_cap_ib_sa(id->device, id->port_num))
2669                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2670         else if (rdma_protocol_roce(id->device, id->port_num))
2671                 ret = cma_resolve_iboe_route(id_priv);
2672         else if (rdma_protocol_iwarp(id->device, id->port_num))
2673                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2674         else
2675                 ret = -ENOSYS;
2676
2677         if (ret)
2678                 goto err;
2679
2680         return 0;
2681 err:
2682         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2683         cma_deref_id(id_priv);
2684         return ret;
2685 }
2686 EXPORT_SYMBOL(rdma_resolve_route);
2687
2688 static void cma_set_loopback(struct sockaddr *addr)
2689 {
2690         switch (addr->sa_family) {
2691         case AF_INET:
2692                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2693                 break;
2694         case AF_INET6:
2695                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2696                               0, 0, 0, htonl(1));
2697                 break;
2698         default:
2699                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2700                             0, 0, 0, htonl(1));
2701                 break;
2702         }
2703 }
2704
2705 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2706 {
2707         struct cma_device *cma_dev, *cur_dev;
2708         union ib_gid gid;
2709         enum ib_port_state port_state;
2710         u16 pkey;
2711         int ret;
2712         u8 p;
2713
2714         cma_dev = NULL;
2715         mutex_lock(&lock);
2716         list_for_each_entry(cur_dev, &dev_list, list) {
2717                 if (cma_family(id_priv) == AF_IB &&
2718                     !rdma_cap_ib_cm(cur_dev->device, 1))
2719                         continue;
2720
2721                 if (!cma_dev)
2722                         cma_dev = cur_dev;
2723
2724                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2725                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2726                             port_state == IB_PORT_ACTIVE) {
2727                                 cma_dev = cur_dev;
2728                                 goto port_found;
2729                         }
2730                 }
2731         }
2732
2733         if (!cma_dev) {
2734                 ret = -ENODEV;
2735                 goto out;
2736         }
2737
2738         p = 1;
2739
2740 port_found:
2741         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2742         if (ret)
2743                 goto out;
2744
2745         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2746         if (ret)
2747                 goto out;
2748
2749         id_priv->id.route.addr.dev_addr.dev_type =
2750                 (rdma_protocol_ib(cma_dev->device, p)) ?
2751                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2752
2753         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2754         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2755         id_priv->id.port_num = p;
2756         cma_attach_to_dev(id_priv, cma_dev);
2757         cma_set_loopback(cma_src_addr(id_priv));
2758 out:
2759         mutex_unlock(&lock);
2760         return ret;
2761 }
2762
2763 static void addr_handler(int status, struct sockaddr *src_addr,
2764                          struct rdma_dev_addr *dev_addr, void *context)
2765 {
2766         struct rdma_id_private *id_priv = context;
2767         struct rdma_cm_event event;
2768
2769         memset(&event, 0, sizeof event);
2770         mutex_lock(&id_priv->handler_mutex);
2771         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2772                            RDMA_CM_ADDR_RESOLVED))
2773                 goto out;
2774
2775         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2776         if (!status && !id_priv->cma_dev) {
2777                 status = cma_acquire_dev(id_priv, NULL);
2778                 if (status)
2779                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2780                                              status);
2781         } else {
2782                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2783         }
2784
2785         if (status) {
2786                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2787                                    RDMA_CM_ADDR_BOUND))
2788                         goto out;
2789                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2790                 event.status = status;
2791         } else
2792                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2793
2794         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2795                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2796                 mutex_unlock(&id_priv->handler_mutex);
2797                 cma_deref_id(id_priv);
2798                 rdma_destroy_id(&id_priv->id);
2799                 return;
2800         }
2801 out:
2802         mutex_unlock(&id_priv->handler_mutex);
2803         cma_deref_id(id_priv);
2804 }
2805
2806 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2807 {
2808         struct cma_work *work;
2809         union ib_gid gid;
2810         int ret;
2811
2812         work = kzalloc(sizeof *work, GFP_KERNEL);
2813         if (!work)
2814                 return -ENOMEM;
2815
2816         if (!id_priv->cma_dev) {
2817                 ret = cma_bind_loopback(id_priv);
2818                 if (ret)
2819                         goto err;
2820         }
2821
2822         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2823         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2824
2825         cma_init_resolve_addr_work(work, id_priv);
2826         queue_work(cma_wq, &work->work);
2827         return 0;
2828 err:
2829         kfree(work);
2830         return ret;
2831 }
2832
2833 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2834 {
2835         struct cma_work *work;
2836         int ret;
2837
2838         work = kzalloc(sizeof *work, GFP_KERNEL);
2839         if (!work)
2840                 return -ENOMEM;
2841
2842         if (!id_priv->cma_dev) {
2843                 ret = cma_resolve_ib_dev(id_priv);
2844                 if (ret)
2845                         goto err;
2846         }
2847
2848         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2849                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2850
2851         cma_init_resolve_addr_work(work, id_priv);
2852         queue_work(cma_wq, &work->work);
2853         return 0;
2854 err:
2855         kfree(work);
2856         return ret;
2857 }
2858
2859 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2860                          struct sockaddr *dst_addr)
2861 {
2862         if (!src_addr || !src_addr->sa_family) {
2863                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2864                 src_addr->sa_family = dst_addr->sa_family;
2865                 if (IS_ENABLED(CONFIG_IPV6) &&
2866                     dst_addr->sa_family == AF_INET6) {
2867                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2868                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2869                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2870                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2871                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2872                 } else if (dst_addr->sa_family == AF_IB) {
2873                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2874                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2875                 }
2876         }
2877         return rdma_bind_addr(id, src_addr);
2878 }
2879
2880 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2881                       struct sockaddr *dst_addr, int timeout_ms)
2882 {
2883         struct rdma_id_private *id_priv;
2884         int ret;
2885
2886         id_priv = container_of(id, struct rdma_id_private, id);
2887         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2888         if (id_priv->state == RDMA_CM_IDLE) {
2889                 ret = cma_bind_addr(id, src_addr, dst_addr);
2890                 if (ret) {
2891                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2892                         return ret;
2893                 }
2894         }
2895
2896         if (cma_family(id_priv) != dst_addr->sa_family) {
2897                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2898                 return -EINVAL;
2899         }
2900
2901         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2902                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2903                 return -EINVAL;
2904         }
2905
2906         atomic_inc(&id_priv->refcount);
2907         if (cma_any_addr(dst_addr)) {
2908                 ret = cma_resolve_loopback(id_priv);
2909         } else {
2910                 if (dst_addr->sa_family == AF_IB) {
2911                         ret = cma_resolve_ib_addr(id_priv);
2912                 } else {
2913                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2914                                               dst_addr, &id->route.addr.dev_addr,
2915                                               timeout_ms, addr_handler, id_priv);
2916                 }
2917         }
2918         if (ret)
2919                 goto err;
2920
2921         return 0;
2922 err:
2923         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2924         cma_deref_id(id_priv);
2925         return ret;
2926 }
2927 EXPORT_SYMBOL(rdma_resolve_addr);
2928
2929 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2930 {
2931         struct rdma_id_private *id_priv;
2932         unsigned long flags;
2933         int ret;
2934
2935         id_priv = container_of(id, struct rdma_id_private, id);
2936         spin_lock_irqsave(&id_priv->lock, flags);
2937         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2938                 id_priv->reuseaddr = reuse;
2939                 ret = 0;
2940         } else {
2941                 ret = -EINVAL;
2942         }
2943         spin_unlock_irqrestore(&id_priv->lock, flags);
2944         return ret;
2945 }
2946 EXPORT_SYMBOL(rdma_set_reuseaddr);
2947
2948 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2949 {
2950         struct rdma_id_private *id_priv;
2951         unsigned long flags;
2952         int ret;
2953
2954         id_priv = container_of(id, struct rdma_id_private, id);
2955         spin_lock_irqsave(&id_priv->lock, flags);
2956         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2957                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2958                 id_priv->afonly = afonly;
2959                 ret = 0;
2960         } else {
2961                 ret = -EINVAL;
2962         }
2963         spin_unlock_irqrestore(&id_priv->lock, flags);
2964         return ret;
2965 }
2966 EXPORT_SYMBOL(rdma_set_afonly);
2967
2968 static void cma_bind_port(struct rdma_bind_list *bind_list,
2969                           struct rdma_id_private *id_priv)
2970 {
2971         struct sockaddr *addr;
2972         struct sockaddr_ib *sib;
2973         u64 sid, mask;
2974         __be16 port;
2975
2976         addr = cma_src_addr(id_priv);
2977         port = htons(bind_list->port);
2978
2979         switch (addr->sa_family) {
2980         case AF_INET:
2981                 ((struct sockaddr_in *) addr)->sin_port = port;
2982                 break;
2983         case AF_INET6:
2984                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2985                 break;
2986         case AF_IB:
2987                 sib = (struct sockaddr_ib *) addr;
2988                 sid = be64_to_cpu(sib->sib_sid);
2989                 mask = be64_to_cpu(sib->sib_sid_mask);
2990                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2991                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2992                 break;
2993         }
2994         id_priv->bind_list = bind_list;
2995         hlist_add_head(&id_priv->node, &bind_list->owners);
2996 }
2997
2998 static int cma_alloc_port(enum rdma_ucm_port_space ps,
2999                           struct rdma_id_private *id_priv, unsigned short snum)
3000 {
3001         struct rdma_bind_list *bind_list;
3002         int ret;
3003
3004         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3005         if (!bind_list)
3006                 return -ENOMEM;
3007
3008         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3009                            snum);
3010         if (ret < 0)
3011                 goto err;
3012
3013         bind_list->ps = ps;
3014         bind_list->port = (unsigned short)ret;
3015         cma_bind_port(bind_list, id_priv);
3016         return 0;
3017 err:
3018         kfree(bind_list);
3019         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3020 }
3021
3022 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3023                               struct rdma_id_private *id_priv)
3024 {
3025         struct rdma_id_private *cur_id;
3026         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3027         struct sockaddr  *saddr = cma_src_addr(id_priv);
3028         __be16 dport = cma_port(daddr);
3029
3030         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3031                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3032                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3033                 __be16 cur_dport = cma_port(cur_daddr);
3034
3035                 if (id_priv == cur_id)
3036                         continue;
3037
3038                 /* different dest port -> unique */
3039                 if (!cma_any_port(daddr) &&
3040                     !cma_any_port(cur_daddr) &&
3041                     (dport != cur_dport))
3042                         continue;
3043
3044                 /* different src address -> unique */
3045                 if (!cma_any_addr(saddr) &&
3046                     !cma_any_addr(cur_saddr) &&
3047                     cma_addr_cmp(saddr, cur_saddr))
3048                         continue;
3049
3050                 /* different dst address -> unique */
3051                 if (!cma_any_addr(daddr) &&
3052                     !cma_any_addr(cur_daddr) &&
3053                     cma_addr_cmp(daddr, cur_daddr))
3054                         continue;
3055
3056                 return -EADDRNOTAVAIL;
3057         }
3058         return 0;
3059 }
3060
3061 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3062                               struct rdma_id_private *id_priv)
3063 {
3064         static unsigned int last_used_port;
3065         int low, high, remaining;
3066         unsigned int rover;
3067         struct net *net = id_priv->id.route.addr.dev_addr.net;
3068
3069         inet_get_local_port_range(net, &low, &high);
3070         remaining = (high - low) + 1;
3071         rover = prandom_u32() % remaining + low;
3072 retry:
3073         if (last_used_port != rover) {
3074                 struct rdma_bind_list *bind_list;
3075                 int ret;
3076
3077                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3078
3079                 if (!bind_list) {
3080                         ret = cma_alloc_port(ps, id_priv, rover);
3081                 } else {
3082                         ret = cma_port_is_unique(bind_list, id_priv);
3083                         if (!ret)
3084                                 cma_bind_port(bind_list, id_priv);
3085                 }
3086                 /*
3087                  * Remember previously used port number in order to avoid
3088                  * re-using same port immediately after it is closed.
3089                  */
3090                 if (!ret)
3091                         last_used_port = rover;
3092                 if (ret != -EADDRNOTAVAIL)
3093                         return ret;
3094         }
3095         if (--remaining) {
3096                 rover++;
3097                 if ((rover < low) || (rover > high))
3098                         rover = low;
3099                 goto retry;
3100         }
3101         return -EADDRNOTAVAIL;
3102 }
3103
3104 /*
3105  * Check that the requested port is available.  This is called when trying to
3106  * bind to a specific port, or when trying to listen on a bound port.  In
3107  * the latter case, the provided id_priv may already be on the bind_list, but
3108  * we still need to check that it's okay to start listening.
3109  */
3110 static int cma_check_port(struct rdma_bind_list *bind_list,
3111                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3112 {
3113         struct rdma_id_private *cur_id;
3114         struct sockaddr *addr, *cur_addr;
3115
3116         addr = cma_src_addr(id_priv);
3117         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3118                 if (id_priv == cur_id)
3119                         continue;
3120
3121                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3122                     cur_id->reuseaddr)
3123                         continue;
3124
3125                 cur_addr = cma_src_addr(cur_id);
3126                 if (id_priv->afonly && cur_id->afonly &&
3127                     (addr->sa_family != cur_addr->sa_family))
3128                         continue;
3129
3130                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3131                         return -EADDRNOTAVAIL;
3132
3133                 if (!cma_addr_cmp(addr, cur_addr))
3134                         return -EADDRINUSE;
3135         }
3136         return 0;
3137 }
3138
3139 static int cma_use_port(enum rdma_ucm_port_space ps,
3140                         struct rdma_id_private *id_priv)
3141 {
3142         struct rdma_bind_list *bind_list;
3143         unsigned short snum;
3144         int ret;
3145
3146         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3147         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3148                 return -EACCES;
3149
3150         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3151         if (!bind_list) {
3152                 ret = cma_alloc_port(ps, id_priv, snum);
3153         } else {
3154                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3155                 if (!ret)
3156                         cma_bind_port(bind_list, id_priv);
3157         }
3158         return ret;
3159 }
3160
3161 static int cma_bind_listen(struct rdma_id_private *id_priv)
3162 {
3163         struct rdma_bind_list *bind_list = id_priv->bind_list;
3164         int ret = 0;
3165
3166         mutex_lock(&lock);
3167         if (bind_list->owners.first->next)
3168                 ret = cma_check_port(bind_list, id_priv, 0);
3169         mutex_unlock(&lock);
3170         return ret;
3171 }
3172
3173 static enum rdma_ucm_port_space
3174 cma_select_inet_ps(struct rdma_id_private *id_priv)
3175 {
3176         switch (id_priv->id.ps) {
3177         case RDMA_PS_TCP:
3178         case RDMA_PS_UDP:
3179         case RDMA_PS_IPOIB:
3180         case RDMA_PS_IB:
3181                 return id_priv->id.ps;
3182         default:
3183
3184                 return 0;
3185         }
3186 }
3187
3188 static enum rdma_ucm_port_space
3189 cma_select_ib_ps(struct rdma_id_private *id_priv)
3190 {
3191         enum rdma_ucm_port_space ps = 0;
3192         struct sockaddr_ib *sib;
3193         u64 sid_ps, mask, sid;
3194
3195         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3196         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3197         sid = be64_to_cpu(sib->sib_sid) & mask;
3198
3199         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3200                 sid_ps = RDMA_IB_IP_PS_IB;
3201                 ps = RDMA_PS_IB;
3202         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3203                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3204                 sid_ps = RDMA_IB_IP_PS_TCP;
3205                 ps = RDMA_PS_TCP;
3206         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3207                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3208                 sid_ps = RDMA_IB_IP_PS_UDP;
3209                 ps = RDMA_PS_UDP;
3210         }
3211
3212         if (ps) {
3213                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3214                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3215                                                 be64_to_cpu(sib->sib_sid_mask));
3216         }
3217         return ps;
3218 }
3219
3220 static int cma_get_port(struct rdma_id_private *id_priv)
3221 {
3222         enum rdma_ucm_port_space ps;
3223         int ret;
3224
3225         if (cma_family(id_priv) != AF_IB)
3226                 ps = cma_select_inet_ps(id_priv);
3227         else
3228                 ps = cma_select_ib_ps(id_priv);
3229         if (!ps)
3230                 return -EPROTONOSUPPORT;
3231
3232         mutex_lock(&lock);
3233         if (cma_any_port(cma_src_addr(id_priv)))
3234                 ret = cma_alloc_any_port(ps, id_priv);
3235         else
3236                 ret = cma_use_port(ps, id_priv);
3237         mutex_unlock(&lock);
3238
3239         return ret;
3240 }
3241
3242 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3243                                struct sockaddr *addr)
3244 {
3245 #if IS_ENABLED(CONFIG_IPV6)
3246         struct sockaddr_in6 *sin6;
3247
3248         if (addr->sa_family != AF_INET6)
3249                 return 0;
3250
3251         sin6 = (struct sockaddr_in6 *) addr;
3252
3253         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3254                 return 0;
3255
3256         if (!sin6->sin6_scope_id)
3257                         return -EINVAL;
3258
3259         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3260 #endif
3261         return 0;
3262 }
3263
3264 int rdma_listen(struct rdma_cm_id *id, int backlog)
3265 {
3266         struct rdma_id_private *id_priv;
3267         int ret;
3268
3269         id_priv = container_of(id, struct rdma_id_private, id);
3270         if (id_priv->state == RDMA_CM_IDLE) {
3271                 id->route.addr.src_addr.ss_family = AF_INET;
3272                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3273                 if (ret)
3274                         return ret;
3275         }
3276
3277         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3278                 return -EINVAL;
3279
3280         if (id_priv->reuseaddr) {
3281                 ret = cma_bind_listen(id_priv);
3282                 if (ret)
3283                         goto err;
3284         }
3285
3286         id_priv->backlog = backlog;
3287         if (id->device) {
3288                 if (rdma_cap_ib_cm(id->device, 1)) {
3289                         ret = cma_ib_listen(id_priv);
3290                         if (ret)
3291                                 goto err;
3292                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3293                         ret = cma_iw_listen(id_priv, backlog);
3294                         if (ret)
3295                                 goto err;
3296                 } else {
3297                         ret = -ENOSYS;
3298                         goto err;
3299                 }
3300         } else
3301                 cma_listen_on_all(id_priv);
3302
3303         return 0;
3304 err:
3305         id_priv->backlog = 0;
3306         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3307         return ret;
3308 }
3309 EXPORT_SYMBOL(rdma_listen);
3310
3311 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3312 {
3313         struct rdma_id_private *id_priv;
3314         int ret;
3315         struct sockaddr  *daddr;
3316
3317         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3318             addr->sa_family != AF_IB)
3319                 return -EAFNOSUPPORT;
3320
3321         id_priv = container_of(id, struct rdma_id_private, id);
3322         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3323                 return -EINVAL;
3324
3325         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3326         if (ret)
3327                 goto err1;
3328
3329         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3330         if (!cma_any_addr(addr)) {
3331                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3332                 if (ret)
3333                         goto err1;
3334
3335                 ret = cma_acquire_dev(id_priv, NULL);
3336                 if (ret)
3337                         goto err1;
3338         }
3339
3340         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3341                 if (addr->sa_family == AF_INET)
3342                         id_priv->afonly = 1;
3343 #if IS_ENABLED(CONFIG_IPV6)
3344                 else if (addr->sa_family == AF_INET6) {
3345                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3346
3347                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3348                 }
3349 #endif
3350         }
3351         daddr = cma_dst_addr(id_priv);
3352         daddr->sa_family = addr->sa_family;
3353
3354         ret = cma_get_port(id_priv);
3355         if (ret)
3356                 goto err2;
3357
3358         return 0;
3359 err2:
3360         if (id_priv->cma_dev) {
3361                 rdma_restrack_del(&id_priv->res);
3362                 cma_release_dev(id_priv);
3363         }
3364 err1:
3365         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3366         return ret;
3367 }
3368 EXPORT_SYMBOL(rdma_bind_addr);
3369
3370 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3371 {
3372         struct cma_hdr *cma_hdr;
3373
3374         cma_hdr = hdr;
3375         cma_hdr->cma_version = CMA_VERSION;
3376         if (cma_family(id_priv) == AF_INET) {
3377                 struct sockaddr_in *src4, *dst4;
3378
3379                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3380                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3381
3382                 cma_set_ip_ver(cma_hdr, 4);
3383                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3384                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3385                 cma_hdr->port = src4->sin_port;
3386         } else if (cma_family(id_priv) == AF_INET6) {
3387                 struct sockaddr_in6 *src6, *dst6;
3388
3389                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3390                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3391
3392                 cma_set_ip_ver(cma_hdr, 6);
3393                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3394                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3395                 cma_hdr->port = src6->sin6_port;
3396         }
3397         return 0;
3398 }
3399
3400 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3401                                 struct ib_cm_event *ib_event)
3402 {
3403         struct rdma_id_private *id_priv = cm_id->context;
3404         struct rdma_cm_event event;
3405         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3406         int ret = 0;
3407
3408         mutex_lock(&id_priv->handler_mutex);
3409         if (id_priv->state != RDMA_CM_CONNECT)
3410                 goto out;
3411
3412         memset(&event, 0, sizeof event);
3413         switch (ib_event->event) {
3414         case IB_CM_SIDR_REQ_ERROR:
3415                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3416                 event.status = -ETIMEDOUT;
3417                 break;
3418         case IB_CM_SIDR_REP_RECEIVED:
3419                 event.param.ud.private_data = ib_event->private_data;
3420                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3421                 if (rep->status != IB_SIDR_SUCCESS) {
3422                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3423                         event.status = ib_event->param.sidr_rep_rcvd.status;
3424                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3425                                              event.status);
3426                         break;
3427                 }
3428                 ret = cma_set_qkey(id_priv, rep->qkey);
3429                 if (ret) {
3430                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3431                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3432                         event.status = ret;
3433                         break;
3434                 }
3435                 ib_init_ah_attr_from_path(id_priv->id.device,
3436                                           id_priv->id.port_num,
3437                                           id_priv->id.route.path_rec,
3438                                           &event.param.ud.ah_attr);
3439                 event.param.ud.qp_num = rep->qpn;
3440                 event.param.ud.qkey = rep->qkey;
3441                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3442                 event.status = 0;
3443                 break;
3444         default:
3445                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3446                        ib_event->event);
3447                 goto out;
3448         }
3449
3450         ret = id_priv->id.event_handler(&id_priv->id, &event);
3451         if (ret) {
3452                 /* Destroy the CM ID by returning a non-zero value. */
3453                 id_priv->cm_id.ib = NULL;
3454                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3455                 mutex_unlock(&id_priv->handler_mutex);
3456                 rdma_destroy_id(&id_priv->id);
3457                 return ret;
3458         }
3459 out:
3460         mutex_unlock(&id_priv->handler_mutex);
3461         return ret;
3462 }
3463
3464 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3465                               struct rdma_conn_param *conn_param)
3466 {
3467         struct ib_cm_sidr_req_param req;
3468         struct ib_cm_id *id;
3469         void *private_data;
3470         u8 offset;
3471         int ret;
3472
3473         memset(&req, 0, sizeof req);
3474         offset = cma_user_data_offset(id_priv);
3475         req.private_data_len = offset + conn_param->private_data_len;
3476         if (req.private_data_len < conn_param->private_data_len)
3477                 return -EINVAL;
3478
3479         if (req.private_data_len) {
3480                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3481                 if (!private_data)
3482                         return -ENOMEM;
3483         } else {
3484                 private_data = NULL;
3485         }
3486
3487         if (conn_param->private_data && conn_param->private_data_len)
3488                 memcpy(private_data + offset, conn_param->private_data,
3489                        conn_param->private_data_len);
3490
3491         if (private_data) {
3492                 ret = cma_format_hdr(private_data, id_priv);
3493                 if (ret)
3494                         goto out;
3495                 req.private_data = private_data;
3496         }
3497
3498         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3499                              id_priv);
3500         if (IS_ERR(id)) {
3501                 ret = PTR_ERR(id);
3502                 goto out;
3503         }
3504         id_priv->cm_id.ib = id;
3505
3506         req.path = id_priv->id.route.path_rec;
3507         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3508         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3509         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3510
3511         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3512         if (ret) {
3513                 ib_destroy_cm_id(id_priv->cm_id.ib);
3514                 id_priv->cm_id.ib = NULL;
3515         }
3516 out:
3517         kfree(private_data);
3518         return ret;
3519 }
3520
3521 static int cma_connect_ib(struct rdma_id_private *id_priv,
3522                           struct rdma_conn_param *conn_param)
3523 {
3524         struct ib_cm_req_param req;
3525         struct rdma_route *route;
3526         void *private_data;
3527         struct ib_cm_id *id;
3528         u8 offset;
3529         int ret;
3530
3531         memset(&req, 0, sizeof req);
3532         offset = cma_user_data_offset(id_priv);
3533         req.private_data_len = offset + conn_param->private_data_len;
3534         if (req.private_data_len < conn_param->private_data_len)
3535                 return -EINVAL;
3536
3537         if (req.private_data_len) {
3538                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3539                 if (!private_data)
3540                         return -ENOMEM;
3541         } else {
3542                 private_data = NULL;
3543         }
3544
3545         if (conn_param->private_data && conn_param->private_data_len)
3546                 memcpy(private_data + offset, conn_param->private_data,
3547                        conn_param->private_data_len);
3548
3549         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3550         if (IS_ERR(id)) {
3551                 ret = PTR_ERR(id);
3552                 goto out;
3553         }
3554         id_priv->cm_id.ib = id;
3555
3556         route = &id_priv->id.route;
3557         if (private_data) {
3558                 ret = cma_format_hdr(private_data, id_priv);
3559                 if (ret)
3560                         goto out;
3561                 req.private_data = private_data;
3562         }
3563
3564         req.primary_path = &route->path_rec[0];
3565         if (route->num_paths == 2)
3566                 req.alternate_path = &route->path_rec[1];
3567
3568         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3569         req.qp_num = id_priv->qp_num;
3570         req.qp_type = id_priv->id.qp_type;
3571         req.starting_psn = id_priv->seq_num;
3572         req.responder_resources = conn_param->responder_resources;
3573         req.initiator_depth = conn_param->initiator_depth;
3574         req.flow_control = conn_param->flow_control;
3575         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3576         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3577         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3578         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3579         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3580         req.srq = id_priv->srq ? 1 : 0;
3581
3582         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3583 out:
3584         if (ret && !IS_ERR(id)) {
3585                 ib_destroy_cm_id(id);
3586                 id_priv->cm_id.ib = NULL;
3587         }
3588
3589         kfree(private_data);
3590         return ret;
3591 }
3592
3593 static int cma_connect_iw(struct rdma_id_private *id_priv,
3594                           struct rdma_conn_param *conn_param)
3595 {
3596         struct iw_cm_id *cm_id;
3597         int ret;
3598         struct iw_cm_conn_param iw_param;
3599
3600         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3601         if (IS_ERR(cm_id))
3602                 return PTR_ERR(cm_id);
3603
3604         cm_id->tos = id_priv->tos;
3605         id_priv->cm_id.iw = cm_id;
3606
3607         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3608                rdma_addr_size(cma_src_addr(id_priv)));
3609         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3610                rdma_addr_size(cma_dst_addr(id_priv)));
3611
3612         ret = cma_modify_qp_rtr(id_priv, conn_param);
3613         if (ret)
3614                 goto out;
3615
3616         if (conn_param) {
3617                 iw_param.ord = conn_param->initiator_depth;
3618                 iw_param.ird = conn_param->responder_resources;
3619                 iw_param.private_data = conn_param->private_data;
3620                 iw_param.private_data_len = conn_param->private_data_len;
3621                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3622         } else {
3623                 memset(&iw_param, 0, sizeof iw_param);
3624                 iw_param.qpn = id_priv->qp_num;
3625         }
3626         ret = iw_cm_connect(cm_id, &iw_param);
3627 out:
3628         if (ret) {
3629                 iw_destroy_cm_id(cm_id);
3630                 id_priv->cm_id.iw = NULL;
3631         }
3632         return ret;
3633 }
3634
3635 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3636 {
3637         struct rdma_id_private *id_priv;
3638         int ret;
3639
3640         id_priv = container_of(id, struct rdma_id_private, id);
3641         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3642                 return -EINVAL;
3643
3644         if (!id->qp) {
3645                 id_priv->qp_num = conn_param->qp_num;
3646                 id_priv->srq = conn_param->srq;
3647         }
3648
3649         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3650                 if (id->qp_type == IB_QPT_UD)
3651                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3652                 else
3653                         ret = cma_connect_ib(id_priv, conn_param);
3654         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3655                 ret = cma_connect_iw(id_priv, conn_param);
3656         else
3657                 ret = -ENOSYS;
3658         if (ret)
3659                 goto err;
3660
3661         return 0;
3662 err:
3663         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3664         return ret;
3665 }
3666 EXPORT_SYMBOL(rdma_connect);
3667
3668 static int cma_accept_ib(struct rdma_id_private *id_priv,
3669                          struct rdma_conn_param *conn_param)
3670 {
3671         struct ib_cm_rep_param rep;
3672         int ret;
3673
3674         ret = cma_modify_qp_rtr(id_priv, conn_param);
3675         if (ret)
3676                 goto out;
3677
3678         ret = cma_modify_qp_rts(id_priv, conn_param);
3679         if (ret)
3680                 goto out;
3681
3682         memset(&rep, 0, sizeof rep);
3683         rep.qp_num = id_priv->qp_num;
3684         rep.starting_psn = id_priv->seq_num;
3685         rep.private_data = conn_param->private_data;
3686         rep.private_data_len = conn_param->private_data_len;
3687         rep.responder_resources = conn_param->responder_resources;
3688         rep.initiator_depth = conn_param->initiator_depth;
3689         rep.failover_accepted = 0;
3690         rep.flow_control = conn_param->flow_control;
3691         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3692         rep.srq = id_priv->srq ? 1 : 0;
3693
3694         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3695 out:
3696         return ret;
3697 }
3698
3699 static int cma_accept_iw(struct rdma_id_private *id_priv,
3700                   struct rdma_conn_param *conn_param)
3701 {
3702         struct iw_cm_conn_param iw_param;
3703         int ret;
3704
3705         if (!conn_param)
3706                 return -EINVAL;
3707
3708         ret = cma_modify_qp_rtr(id_priv, conn_param);
3709         if (ret)
3710                 return ret;
3711
3712         iw_param.ord = conn_param->initiator_depth;
3713         iw_param.ird = conn_param->responder_resources;
3714         iw_param.private_data = conn_param->private_data;
3715         iw_param.private_data_len = conn_param->private_data_len;
3716         if (id_priv->id.qp) {
3717                 iw_param.qpn = id_priv->qp_num;
3718         } else
3719                 iw_param.qpn = conn_param->qp_num;
3720
3721         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3722 }
3723
3724 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3725                              enum ib_cm_sidr_status status, u32 qkey,
3726                              const void *private_data, int private_data_len)
3727 {
3728         struct ib_cm_sidr_rep_param rep;
3729         int ret;
3730
3731         memset(&rep, 0, sizeof rep);
3732         rep.status = status;
3733         if (status == IB_SIDR_SUCCESS) {
3734                 ret = cma_set_qkey(id_priv, qkey);
3735                 if (ret)
3736                         return ret;
3737                 rep.qp_num = id_priv->qp_num;
3738                 rep.qkey = id_priv->qkey;
3739         }
3740         rep.private_data = private_data;
3741         rep.private_data_len = private_data_len;
3742
3743         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3744 }
3745
3746 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3747                   const char *caller)
3748 {
3749         struct rdma_id_private *id_priv;
3750         int ret;
3751
3752         id_priv = container_of(id, struct rdma_id_private, id);
3753
3754         if (caller)
3755                 id_priv->res.kern_name = caller;
3756         else
3757                 rdma_restrack_set_task(&id_priv->res, current);
3758
3759         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3760                 return -EINVAL;
3761
3762         if (!id->qp && conn_param) {
3763                 id_priv->qp_num = conn_param->qp_num;
3764                 id_priv->srq = conn_param->srq;
3765         }
3766
3767         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3768                 if (id->qp_type == IB_QPT_UD) {
3769                         if (conn_param)
3770                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3771                                                         conn_param->qkey,
3772                                                         conn_param->private_data,
3773                                                         conn_param->private_data_len);
3774                         else
3775                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3776                                                         0, NULL, 0);
3777                 } else {
3778                         if (conn_param)
3779                                 ret = cma_accept_ib(id_priv, conn_param);
3780                         else
3781                                 ret = cma_rep_recv(id_priv);
3782                 }
3783         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3784                 ret = cma_accept_iw(id_priv, conn_param);
3785         else
3786                 ret = -ENOSYS;
3787
3788         if (ret)
3789                 goto reject;
3790
3791         return 0;
3792 reject:
3793         cma_modify_qp_err(id_priv);
3794         rdma_reject(id, NULL, 0);
3795         return ret;
3796 }
3797 EXPORT_SYMBOL(__rdma_accept);
3798
3799 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3800 {
3801         struct rdma_id_private *id_priv;
3802         int ret;
3803
3804         id_priv = container_of(id, struct rdma_id_private, id);
3805         if (!id_priv->cm_id.ib)
3806                 return -EINVAL;
3807
3808         switch (id->device->node_type) {
3809         case RDMA_NODE_IB_CA:
3810                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3811                 break;
3812         default:
3813                 ret = 0;
3814                 break;
3815         }
3816         return ret;
3817 }
3818 EXPORT_SYMBOL(rdma_notify);
3819
3820 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3821                 u8 private_data_len)
3822 {
3823         struct rdma_id_private *id_priv;
3824         int ret;
3825
3826         id_priv = container_of(id, struct rdma_id_private, id);
3827         if (!id_priv->cm_id.ib)
3828                 return -EINVAL;
3829
3830         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3831                 if (id->qp_type == IB_QPT_UD)
3832                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3833                                                 private_data, private_data_len);
3834                 else
3835                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3836                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3837                                              0, private_data, private_data_len);
3838         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3839                 ret = iw_cm_reject(id_priv->cm_id.iw,
3840                                    private_data, private_data_len);
3841         } else
3842                 ret = -ENOSYS;
3843
3844         return ret;
3845 }
3846 EXPORT_SYMBOL(rdma_reject);
3847
3848 int rdma_disconnect(struct rdma_cm_id *id)
3849 {
3850         struct rdma_id_private *id_priv;
3851         int ret;
3852
3853         id_priv = container_of(id, struct rdma_id_private, id);
3854         if (!id_priv->cm_id.ib)
3855                 return -EINVAL;
3856
3857         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3858                 ret = cma_modify_qp_err(id_priv);
3859                 if (ret)
3860                         goto out;
3861                 /* Initiate or respond to a disconnect. */
3862                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3863                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3864         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3865                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3866         } else
3867                 ret = -EINVAL;
3868
3869 out:
3870         return ret;
3871 }
3872 EXPORT_SYMBOL(rdma_disconnect);
3873
3874 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3875 {
3876         struct rdma_id_private *id_priv;
3877         struct cma_multicast *mc = multicast->context;
3878         struct rdma_cm_event event;
3879         int ret = 0;
3880
3881         id_priv = mc->id_priv;
3882         mutex_lock(&id_priv->handler_mutex);
3883         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3884             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3885                 goto out;
3886
3887         if (!status)
3888                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3889         else
3890                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3891                                      status);
3892         mutex_lock(&id_priv->qp_mutex);
3893         if (!status && id_priv->id.qp) {
3894                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3895                                          be16_to_cpu(multicast->rec.mlid));
3896                 if (status)
3897                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3898                                              status);
3899         }
3900         mutex_unlock(&id_priv->qp_mutex);
3901
3902         memset(&event, 0, sizeof event);
3903         event.status = status;
3904         event.param.ud.private_data = mc->context;
3905         if (!status) {
3906                 struct rdma_dev_addr *dev_addr =
3907                         &id_priv->id.route.addr.dev_addr;
3908                 struct net_device *ndev =
3909                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
3910                 enum ib_gid_type gid_type =
3911                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3912                         rdma_start_port(id_priv->cma_dev->device)];
3913
3914                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3915                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
3916                                                id_priv->id.port_num,
3917                                                &multicast->rec,
3918                                                ndev, gid_type,
3919                                                &event.param.ud.ah_attr);
3920                 if (ret)
3921                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3922
3923                 event.param.ud.qp_num = 0xFFFFFF;
3924                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3925                 if (ndev)
3926                         dev_put(ndev);
3927         } else
3928                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3929
3930         ret = id_priv->id.event_handler(&id_priv->id, &event);
3931         if (ret) {
3932                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3933                 mutex_unlock(&id_priv->handler_mutex);
3934                 rdma_destroy_id(&id_priv->id);
3935                 return 0;
3936         }
3937
3938 out:
3939         mutex_unlock(&id_priv->handler_mutex);
3940         return 0;
3941 }
3942
3943 static void cma_set_mgid(struct rdma_id_private *id_priv,
3944                          struct sockaddr *addr, union ib_gid *mgid)
3945 {
3946         unsigned char mc_map[MAX_ADDR_LEN];
3947         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3948         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3949         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3950
3951         if (cma_any_addr(addr)) {
3952                 memset(mgid, 0, sizeof *mgid);
3953         } else if ((addr->sa_family == AF_INET6) &&
3954                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3955                                                                  0xFF10A01B)) {
3956                 /* IPv6 address is an SA assigned MGID. */
3957                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3958         } else if (addr->sa_family == AF_IB) {
3959                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3960         } else if ((addr->sa_family == AF_INET6)) {
3961                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3962                 if (id_priv->id.ps == RDMA_PS_UDP)
3963                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3964                 *mgid = *(union ib_gid *) (mc_map + 4);
3965         } else {
3966                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3967                 if (id_priv->id.ps == RDMA_PS_UDP)
3968                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3969                 *mgid = *(union ib_gid *) (mc_map + 4);
3970         }
3971 }
3972
3973 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3974                                  struct cma_multicast *mc)
3975 {
3976         struct ib_sa_mcmember_rec rec;
3977         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3978         ib_sa_comp_mask comp_mask;
3979         int ret;
3980
3981         ib_addr_get_mgid(dev_addr, &rec.mgid);
3982         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3983                                      &rec.mgid, &rec);
3984         if (ret)
3985                 return ret;
3986
3987         ret = cma_set_qkey(id_priv, 0);
3988         if (ret)
3989                 return ret;
3990
3991         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3992         rec.qkey = cpu_to_be32(id_priv->qkey);
3993         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3994         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3995         rec.join_state = mc->join_state;
3996
3997         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3998             (!ib_sa_sendonly_fullmem_support(&sa_client,
3999                                              id_priv->id.device,
4000                                              id_priv->id.port_num))) {
4001                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4002                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4003                         id_priv->id.device->name, id_priv->id.port_num);
4004                 return -EOPNOTSUPP;
4005         }
4006
4007         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4008                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4009                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4010                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4011                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4012
4013         if (id_priv->id.ps == RDMA_PS_IPOIB)
4014                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4015                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4016                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4017                              IB_SA_MCMEMBER_REC_MTU |
4018                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4019
4020         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4021                                                 id_priv->id.port_num, &rec,
4022                                                 comp_mask, GFP_KERNEL,
4023                                                 cma_ib_mc_handler, mc);
4024         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4025 }
4026
4027 static void iboe_mcast_work_handler(struct work_struct *work)
4028 {
4029         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4030         struct cma_multicast *mc = mw->mc;
4031         struct ib_sa_multicast *m = mc->multicast.ib;
4032
4033         mc->multicast.ib->context = mc;
4034         cma_ib_mc_handler(0, m);
4035         kref_put(&mc->mcref, release_mc);
4036         kfree(mw);
4037 }
4038
4039 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4040                               enum ib_gid_type gid_type)
4041 {
4042         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4043         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4044
4045         if (cma_any_addr(addr)) {
4046                 memset(mgid, 0, sizeof *mgid);
4047         } else if (addr->sa_family == AF_INET6) {
4048                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4049         } else {
4050                 mgid->raw[0] =
4051                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4052                 mgid->raw[1] =
4053                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4054                 mgid->raw[2] = 0;
4055                 mgid->raw[3] = 0;
4056                 mgid->raw[4] = 0;
4057                 mgid->raw[5] = 0;
4058                 mgid->raw[6] = 0;
4059                 mgid->raw[7] = 0;
4060                 mgid->raw[8] = 0;
4061                 mgid->raw[9] = 0;
4062                 mgid->raw[10] = 0xff;
4063                 mgid->raw[11] = 0xff;
4064                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4065         }
4066 }
4067
4068 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4069                                    struct cma_multicast *mc)
4070 {
4071         struct iboe_mcast_work *work;
4072         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4073         int err = 0;
4074         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4075         struct net_device *ndev = NULL;
4076         enum ib_gid_type gid_type;
4077         bool send_only;
4078
4079         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4080
4081         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4082                 return -EINVAL;
4083
4084         work = kzalloc(sizeof *work, GFP_KERNEL);
4085         if (!work)
4086                 return -ENOMEM;
4087
4088         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4089         if (!mc->multicast.ib) {
4090                 err = -ENOMEM;
4091                 goto out1;
4092         }
4093
4094         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4095                    rdma_start_port(id_priv->cma_dev->device)];
4096         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4097
4098         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4099         if (id_priv->id.ps == RDMA_PS_UDP)
4100                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4101
4102         if (dev_addr->bound_dev_if)
4103                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4104         if (!ndev) {
4105                 err = -ENODEV;
4106                 goto out2;
4107         }
4108         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4109         mc->multicast.ib->rec.hop_limit = 1;
4110         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4111
4112         if (addr->sa_family == AF_INET) {
4113                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4114                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4115                         if (!send_only) {
4116                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4117                                                     true);
4118                                 if (!err)
4119                                         mc->igmp_joined = true;
4120                         }
4121                 }
4122         } else {
4123                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4124                         err = -ENOTSUPP;
4125         }
4126         dev_put(ndev);
4127         if (err || !mc->multicast.ib->rec.mtu) {
4128                 if (!err)
4129                         err = -EINVAL;
4130                 goto out2;
4131         }
4132         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4133                     &mc->multicast.ib->rec.port_gid);
4134         work->id = id_priv;
4135         work->mc = mc;
4136         INIT_WORK(&work->work, iboe_mcast_work_handler);
4137         kref_get(&mc->mcref);
4138         queue_work(cma_wq, &work->work);
4139
4140         return 0;
4141
4142 out2:
4143         kfree(mc->multicast.ib);
4144 out1:
4145         kfree(work);
4146         return err;
4147 }
4148
4149 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4150                         u8 join_state, void *context)
4151 {
4152         struct rdma_id_private *id_priv;
4153         struct cma_multicast *mc;
4154         int ret;
4155
4156         if (!id->device)
4157                 return -EINVAL;
4158
4159         id_priv = container_of(id, struct rdma_id_private, id);
4160         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4161             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4162                 return -EINVAL;
4163
4164         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4165         if (!mc)
4166                 return -ENOMEM;
4167
4168         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4169         mc->context = context;
4170         mc->id_priv = id_priv;
4171         mc->igmp_joined = false;
4172         mc->join_state = join_state;
4173         spin_lock(&id_priv->lock);
4174         list_add(&mc->list, &id_priv->mc_list);
4175         spin_unlock(&id_priv->lock);
4176
4177         if (rdma_protocol_roce(id->device, id->port_num)) {
4178                 kref_init(&mc->mcref);
4179                 ret = cma_iboe_join_multicast(id_priv, mc);
4180         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4181                 ret = cma_join_ib_multicast(id_priv, mc);
4182         else
4183                 ret = -ENOSYS;
4184
4185         if (ret) {
4186                 spin_lock_irq(&id_priv->lock);
4187                 list_del(&mc->list);
4188                 spin_unlock_irq(&id_priv->lock);
4189                 kfree(mc);
4190         }
4191         return ret;
4192 }
4193 EXPORT_SYMBOL(rdma_join_multicast);
4194
4195 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4196 {
4197         struct rdma_id_private *id_priv;
4198         struct cma_multicast *mc;
4199
4200         id_priv = container_of(id, struct rdma_id_private, id);
4201         spin_lock_irq(&id_priv->lock);
4202         list_for_each_entry(mc, &id_priv->mc_list, list) {
4203                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4204                         list_del(&mc->list);
4205                         spin_unlock_irq(&id_priv->lock);
4206
4207                         if (id->qp)
4208                                 ib_detach_mcast(id->qp,
4209                                                 &mc->multicast.ib->rec.mgid,
4210                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4211
4212                         BUG_ON(id_priv->cma_dev->device != id->device);
4213
4214                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4215                                 ib_sa_free_multicast(mc->multicast.ib);
4216                                 kfree(mc);
4217                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4218                                 if (mc->igmp_joined) {
4219                                         struct rdma_dev_addr *dev_addr =
4220                                                 &id->route.addr.dev_addr;
4221                                         struct net_device *ndev = NULL;
4222
4223                                         if (dev_addr->bound_dev_if)
4224                                                 ndev = dev_get_by_index(dev_addr->net,
4225                                                                         dev_addr->bound_dev_if);
4226                                         if (ndev) {
4227                                                 cma_igmp_send(ndev,
4228                                                               &mc->multicast.ib->rec.mgid,
4229                                                               false);
4230                                                 dev_put(ndev);
4231                                         }
4232                                         mc->igmp_joined = false;
4233                                 }
4234                                 kref_put(&mc->mcref, release_mc);
4235                         }
4236                         return;
4237                 }
4238         }
4239         spin_unlock_irq(&id_priv->lock);
4240 }
4241 EXPORT_SYMBOL(rdma_leave_multicast);
4242
4243 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4244 {
4245         struct rdma_dev_addr *dev_addr;
4246         struct cma_ndev_work *work;
4247
4248         dev_addr = &id_priv->id.route.addr.dev_addr;
4249
4250         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4251             (net_eq(dev_net(ndev), dev_addr->net)) &&
4252             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4253                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4254                         ndev->name, &id_priv->id);
4255                 work = kzalloc(sizeof *work, GFP_KERNEL);
4256                 if (!work)
4257                         return -ENOMEM;
4258
4259                 INIT_WORK(&work->work, cma_ndev_work_handler);
4260                 work->id = id_priv;
4261                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4262                 atomic_inc(&id_priv->refcount);
4263                 queue_work(cma_wq, &work->work);
4264         }
4265
4266         return 0;
4267 }
4268
4269 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4270                                void *ptr)
4271 {
4272         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4273         struct cma_device *cma_dev;
4274         struct rdma_id_private *id_priv;
4275         int ret = NOTIFY_DONE;
4276
4277         if (event != NETDEV_BONDING_FAILOVER)
4278                 return NOTIFY_DONE;
4279
4280         if (!netif_is_bond_master(ndev))
4281                 return NOTIFY_DONE;
4282
4283         mutex_lock(&lock);
4284         list_for_each_entry(cma_dev, &dev_list, list)
4285                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4286                         ret = cma_netdev_change(ndev, id_priv);
4287                         if (ret)
4288                                 goto out;
4289                 }
4290
4291 out:
4292         mutex_unlock(&lock);
4293         return ret;
4294 }
4295
4296 static struct notifier_block cma_nb = {
4297         .notifier_call = cma_netdev_callback
4298 };
4299
4300 static void cma_add_one(struct ib_device *device)
4301 {
4302         struct cma_device *cma_dev;
4303         struct rdma_id_private *id_priv;
4304         unsigned int i;
4305         unsigned long supported_gids = 0;
4306
4307         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4308         if (!cma_dev)
4309                 return;
4310
4311         cma_dev->device = device;
4312         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4313                                             sizeof(*cma_dev->default_gid_type),
4314                                             GFP_KERNEL);
4315         if (!cma_dev->default_gid_type)
4316                 goto free_cma_dev;
4317
4318         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4319                                             sizeof(*cma_dev->default_roce_tos),
4320                                             GFP_KERNEL);
4321         if (!cma_dev->default_roce_tos)
4322                 goto free_gid_type;
4323
4324         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4325                 supported_gids = roce_gid_type_mask_support(device, i);
4326                 WARN_ON(!supported_gids);
4327                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4328                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4329                                 CMA_PREFERRED_ROCE_GID_TYPE;
4330                 else
4331                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4332                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4333                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4334         }
4335
4336         init_completion(&cma_dev->comp);
4337         atomic_set(&cma_dev->refcount, 1);
4338         INIT_LIST_HEAD(&cma_dev->id_list);
4339         ib_set_client_data(device, &cma_client, cma_dev);
4340
4341         mutex_lock(&lock);
4342         list_add_tail(&cma_dev->list, &dev_list);
4343         list_for_each_entry(id_priv, &listen_any_list, list)
4344                 cma_listen_on_dev(id_priv, cma_dev);
4345         mutex_unlock(&lock);
4346
4347         return;
4348
4349 free_gid_type:
4350         kfree(cma_dev->default_gid_type);
4351
4352 free_cma_dev:
4353         kfree(cma_dev);
4354
4355         return;
4356 }
4357
4358 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4359 {
4360         struct rdma_cm_event event;
4361         enum rdma_cm_state state;
4362         int ret = 0;
4363
4364         /* Record that we want to remove the device */
4365         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4366         if (state == RDMA_CM_DESTROYING)
4367                 return 0;
4368
4369         cma_cancel_operation(id_priv, state);
4370         mutex_lock(&id_priv->handler_mutex);
4371
4372         /* Check for destruction from another callback. */
4373         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4374                 goto out;
4375
4376         memset(&event, 0, sizeof event);
4377         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4378         ret = id_priv->id.event_handler(&id_priv->id, &event);
4379 out:
4380         mutex_unlock(&id_priv->handler_mutex);
4381         return ret;
4382 }
4383
4384 static void cma_process_remove(struct cma_device *cma_dev)
4385 {
4386         struct rdma_id_private *id_priv;
4387         int ret;
4388
4389         mutex_lock(&lock);
4390         while (!list_empty(&cma_dev->id_list)) {
4391                 id_priv = list_entry(cma_dev->id_list.next,
4392                                      struct rdma_id_private, list);
4393
4394                 list_del(&id_priv->listen_list);
4395                 list_del_init(&id_priv->list);
4396                 atomic_inc(&id_priv->refcount);
4397                 mutex_unlock(&lock);
4398
4399                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4400                 cma_deref_id(id_priv);
4401                 if (ret)
4402                         rdma_destroy_id(&id_priv->id);
4403
4404                 mutex_lock(&lock);
4405         }
4406         mutex_unlock(&lock);
4407
4408         cma_deref_dev(cma_dev);
4409         wait_for_completion(&cma_dev->comp);
4410 }
4411
4412 static void cma_remove_one(struct ib_device *device, void *client_data)
4413 {
4414         struct cma_device *cma_dev = client_data;
4415
4416         if (!cma_dev)
4417                 return;
4418
4419         mutex_lock(&lock);
4420         list_del(&cma_dev->list);
4421         mutex_unlock(&lock);
4422
4423         cma_process_remove(cma_dev);
4424         kfree(cma_dev->default_roce_tos);
4425         kfree(cma_dev->default_gid_type);
4426         kfree(cma_dev);
4427 }
4428
4429 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4430 {
4431         struct nlmsghdr *nlh;
4432         struct rdma_cm_id_stats *id_stats;
4433         struct rdma_id_private *id_priv;
4434         struct rdma_cm_id *id = NULL;
4435         struct cma_device *cma_dev;
4436         int i_dev = 0, i_id = 0;
4437
4438         /*
4439          * We export all of the IDs as a sequence of messages.  Each
4440          * ID gets its own netlink message.
4441          */
4442         mutex_lock(&lock);
4443
4444         list_for_each_entry(cma_dev, &dev_list, list) {
4445                 if (i_dev < cb->args[0]) {
4446                         i_dev++;
4447                         continue;
4448                 }
4449
4450                 i_id = 0;
4451                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4452                         if (i_id < cb->args[1]) {
4453                                 i_id++;
4454                                 continue;
4455                         }
4456
4457                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4458                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4459                                                 RDMA_NL_RDMA_CM_ID_STATS,
4460                                                 NLM_F_MULTI);
4461                         if (!id_stats)
4462                                 goto out;
4463
4464                         memset(id_stats, 0, sizeof *id_stats);
4465                         id = &id_priv->id;
4466                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4467                         id_stats->port_num = id->port_num;
4468                         id_stats->bound_dev_if =
4469                                 id->route.addr.dev_addr.bound_dev_if;
4470
4471                         if (ibnl_put_attr(skb, nlh,
4472                                           rdma_addr_size(cma_src_addr(id_priv)),
4473                                           cma_src_addr(id_priv),
4474                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4475                                 goto out;
4476                         if (ibnl_put_attr(skb, nlh,
4477                                           rdma_addr_size(cma_dst_addr(id_priv)),
4478                                           cma_dst_addr(id_priv),
4479                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4480                                 goto out;
4481
4482                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4483                         id_stats->port_space    = id->ps;
4484                         id_stats->cm_state      = id_priv->state;
4485                         id_stats->qp_num        = id_priv->qp_num;
4486                         id_stats->qp_type       = id->qp_type;
4487
4488                         i_id++;
4489                         nlmsg_end(skb, nlh);
4490                 }
4491
4492                 cb->args[1] = 0;
4493                 i_dev++;
4494         }
4495
4496 out:
4497         mutex_unlock(&lock);
4498         cb->args[0] = i_dev;
4499         cb->args[1] = i_id;
4500
4501         return skb->len;
4502 }
4503
4504 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4505         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4506 };
4507
4508 static int cma_init_net(struct net *net)
4509 {
4510         struct cma_pernet *pernet = cma_pernet(net);
4511
4512         idr_init(&pernet->tcp_ps);
4513         idr_init(&pernet->udp_ps);
4514         idr_init(&pernet->ipoib_ps);
4515         idr_init(&pernet->ib_ps);
4516
4517         return 0;
4518 }
4519
4520 static void cma_exit_net(struct net *net)
4521 {
4522         struct cma_pernet *pernet = cma_pernet(net);
4523
4524         idr_destroy(&pernet->tcp_ps);
4525         idr_destroy(&pernet->udp_ps);
4526         idr_destroy(&pernet->ipoib_ps);
4527         idr_destroy(&pernet->ib_ps);
4528 }
4529
4530 static struct pernet_operations cma_pernet_operations = {
4531         .init = cma_init_net,
4532         .exit = cma_exit_net,
4533         .id = &cma_pernet_id,
4534         .size = sizeof(struct cma_pernet),
4535 };
4536
4537 static int __init cma_init(void)
4538 {
4539         int ret;
4540
4541         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4542         if (!cma_wq)
4543                 return -ENOMEM;
4544
4545         ret = register_pernet_subsys(&cma_pernet_operations);
4546         if (ret)
4547                 goto err_wq;
4548
4549         ib_sa_register_client(&sa_client);
4550         rdma_addr_register_client(&addr_client);
4551         register_netdevice_notifier(&cma_nb);
4552
4553         ret = ib_register_client(&cma_client);
4554         if (ret)
4555                 goto err;
4556
4557         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4558         cma_configfs_init();
4559
4560         return 0;
4561
4562 err:
4563         unregister_netdevice_notifier(&cma_nb);
4564         rdma_addr_unregister_client(&addr_client);
4565         ib_sa_unregister_client(&sa_client);
4566 err_wq:
4567         destroy_workqueue(cma_wq);
4568         return ret;
4569 }
4570
4571 static void __exit cma_cleanup(void)
4572 {
4573         cma_configfs_exit();
4574         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4575         ib_unregister_client(&cma_client);
4576         unregister_netdevice_notifier(&cma_nb);
4577         rdma_addr_unregister_client(&addr_client);
4578         ib_sa_unregister_client(&sa_client);
4579         unregister_pernet_subsys(&cma_pernet_operations);
4580         destroy_workqueue(cma_wq);
4581 }
4582
4583 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4584
4585 module_init(cma_init);
4586 module_exit(cma_cleanup);