drivers/infiniband/core/cma.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005-2006 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/completion.h>
  37 #include <linux/in.h>
  38 #include <linux/in6.h>
  39 #include <linux/mutex.h>
  40 #include <linux/random.h>
  41 #include <linux/igmp.h>
  42 #include <linux/idr.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/slab.h>
  45 #include <linux/module.h>
  46 #include <net/route.h>
  47
  48 #include <net/net_namespace.h>
  49 #include <net/netns/generic.h>
  50 #include <net/tcp.h>
  51 #include <net/ipv6.h>
  52 #include <net/ip_fib.h>
  53 #include <net/ip6_route.h>
  54
  55 #include <rdma/rdma_cm.h>
  56 #include <rdma/rdma_cm_ib.h>
  57 #include <rdma/rdma_netlink.h>
  58 #include <rdma/ib.h>
  59 #include <rdma/ib_cache.h>
  60 #include <rdma/ib_cm.h>
  61 #include <rdma/ib_sa.h>
  62 #include <rdma/iw_cm.h>
  63
  64 #include "core_priv.h"
  65
  66 MODULE_AUTHOR("Sean Hefty");
  67 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  68 MODULE_LICENSE("Dual BSD/GPL");
  69
  70 #define CMA_CM_RESPONSE_TIMEOUT 20
  71 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  72 #define CMA_MAX_CM_RETRIES 15
  73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  74 #define CMA_IBOE_PACKET_LIFETIME 18
  75
  76 static const char * const cma_events[] = {
  77         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  78         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  79         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  80         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  81         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  82         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  83         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  84         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  85         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  86         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  87         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  88         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  89         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  90         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  91         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  92         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  93 };
  94
  95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  96 {
  97         size_t index = event;
  98
  99         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 100                         cma_events[index] : "unrecognized event";
 101 }
 102 EXPORT_SYMBOL(rdma_event_msg);
 103
 104 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 105                                                 int reason)
 106 {
 107         if (rdma_ib_or_roce(id->device, id->port_num))
 108                 return ibcm_reject_msg(reason);
 109
 110         if (rdma_protocol_iwarp(id->device, id->port_num))
 111                 return iwcm_reject_msg(reason);
 112
 113         WARN_ON_ONCE(1);
 114         return "unrecognized transport";
 115 }
 116 EXPORT_SYMBOL(rdma_reject_msg);
 117
 118 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 119 {
 120         if (rdma_ib_or_roce(id->device, id->port_num))
 121                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 122
 123         if (rdma_protocol_iwarp(id->device, id->port_num))
 124                 return reason == -ECONNREFUSED;
 125
 126         WARN_ON_ONCE(1);
 127         return false;
 128 }
 129 EXPORT_SYMBOL(rdma_is_consumer_reject);
 130
 131 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 132                                       struct rdma_cm_event *ev, u8 *data_len)
 133 {
 134         const void *p;
 135
 136         if (rdma_is_consumer_reject(id, ev->status)) {
 137                 *data_len = ev->param.conn.private_data_len;
 138                 p = ev->param.conn.private_data;
 139         } else {
 140                 *data_len = 0;
 141                 p = NULL;
 142         }
 143         return p;
 144 }
 145 EXPORT_SYMBOL(rdma_consumer_reject_data);
 146
 147 static void cma_add_one(struct ib_device *device);
 148 static void cma_remove_one(struct ib_device *device, void *client_data);
 149
 150 static struct ib_client cma_client = {
 151         .name   = "cma",
 152         .add    = cma_add_one,
 153         .remove = cma_remove_one
 154 };
 155
 156 static struct ib_sa_client sa_client;
 157 static struct rdma_addr_client addr_client;
 158 static LIST_HEAD(dev_list);
 159 static LIST_HEAD(listen_any_list);
 160 static DEFINE_MUTEX(lock);
 161 static struct workqueue_struct *cma_wq;
 162 static unsigned int cma_pernet_id;
 163
 164 struct cma_pernet {
 165         struct idr tcp_ps;
 166         struct idr udp_ps;
 167         struct idr ipoib_ps;
 168         struct idr ib_ps;
 169 };
 170
 171 static struct cma_pernet *cma_pernet(struct net *net)
 172 {
 173         return net_generic(net, cma_pernet_id);
 174 }
 175
 176 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 177 {
 178         struct cma_pernet *pernet = cma_pernet(net);
 179
 180         switch (ps) {
 181         case RDMA_PS_TCP:
 182                 return &pernet->tcp_ps;
 183         case RDMA_PS_UDP:
 184                 return &pernet->udp_ps;
 185         case RDMA_PS_IPOIB:
 186                 return &pernet->ipoib_ps;
 187         case RDMA_PS_IB:
 188                 return &pernet->ib_ps;
 189         default:
 190                 return NULL;
 191         }
 192 }
 193
 194 struct cma_device {
 195         struct list_head        list;
 196         struct ib_device        *device;
 197         struct completion       comp;
 198         atomic_t                refcount;
 199         struct list_head        id_list;
 200         enum ib_gid_type        *default_gid_type;
 201         u8                      *default_roce_tos;
 202 };
 203
 204 struct rdma_bind_list {
 205         enum rdma_port_space    ps;
 206         struct hlist_head       owners;
 207         unsigned short          port;
 208 };
 209
 210 struct class_port_info_context {
 211         struct ib_class_port_info       *class_port_info;
 212         struct ib_device                *device;
 213         struct completion               done;
 214         struct ib_sa_query              *sa_query;
 215         u8                              port_num;
 216 };
 217
 218 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 219                         struct rdma_bind_list *bind_list, int snum)
 220 {
 221         struct idr *idr = cma_pernet_idr(net, ps);
 222
 223         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 224 }
 225
 226 static struct rdma_bind_list *cma_ps_find(struct net *net,
 227                                           enum rdma_port_space ps, int snum)
 228 {
 229         struct idr *idr = cma_pernet_idr(net, ps);
 230
 231         return idr_find(idr, snum);
 232 }
 233
 234 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 235 {
 236         struct idr *idr = cma_pernet_idr(net, ps);
 237
 238         idr_remove(idr, snum);
 239 }
 240
 241 enum {
 242         CMA_OPTION_AFONLY,
 243 };
 244
 245 void cma_ref_dev(struct cma_device *cma_dev)
 246 {
 247         atomic_inc(&cma_dev->refcount);
 248 }
 249
 250 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 251                                              void               *cookie)
 252 {
 253         struct cma_device *cma_dev;
 254         struct cma_device *found_cma_dev = NULL;
 255
 256         mutex_lock(&lock);
 257
 258         list_for_each_entry(cma_dev, &dev_list, list)
 259                 if (filter(cma_dev->device, cookie)) {
 260                         found_cma_dev = cma_dev;
 261                         break;
 262                 }
 263
 264         if (found_cma_dev)
 265                 cma_ref_dev(found_cma_dev);
 266         mutex_unlock(&lock);
 267         return found_cma_dev;
 268 }
 269
 270 int cma_get_default_gid_type(struct cma_device *cma_dev,
 271                              unsigned int port)
 272 {
 273         if (!rdma_is_port_valid(cma_dev->device, port))
 274                 return -EINVAL;
 275
 276         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 277 }
 278
 279 int cma_set_default_gid_type(struct cma_device *cma_dev,
 280                              unsigned int port,
 281                              enum ib_gid_type default_gid_type)
 282 {
 283         unsigned long supported_gids;
 284
 285         if (!rdma_is_port_valid(cma_dev->device, port))
 286                 return -EINVAL;
 287
 288         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 289
 290         if (!(supported_gids & 1 << default_gid_type))
 291                 return -EINVAL;
 292
 293         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 294                 default_gid_type;
 295
 296         return 0;
 297 }
 298
 299 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 300 {
 301         if (!rdma_is_port_valid(cma_dev->device, port))
 302                 return -EINVAL;
 303
 304         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 305 }
 306
 307 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 308                              u8 default_roce_tos)
 309 {
 310         if (!rdma_is_port_valid(cma_dev->device, port))
 311                 return -EINVAL;
 312
 313         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 314                  default_roce_tos;
 315
 316         return 0;
 317 }
 318 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 319 {
 320         return cma_dev->device;
 321 }
 322
 323 /*
 324  * Device removal can occur at anytime, so we need extra handling to
 325  * serialize notifying the user of device removal with other callbacks.
 326  * We do this by disabling removal notification while a callback is in process,
 327  * and reporting it after the callback completes.
 328  */
 329 struct rdma_id_private {
 330         struct rdma_cm_id       id;
 331
 332         struct rdma_bind_list   *bind_list;
 333         struct hlist_node       node;
 334         struct list_head        list; /* listen_any_list or cma_device.list */
 335         struct list_head        listen_list; /* per device listens */
 336         struct cma_device       *cma_dev;
 337         struct list_head        mc_list;
 338
 339         int                     internal_id;
 340         enum rdma_cm_state      state;
 341         spinlock_t              lock;
 342         struct mutex            qp_mutex;
 343
 344         struct completion       comp;
 345         atomic_t                refcount;
 346         struct mutex            handler_mutex;
 347
 348         int                     backlog;
 349         int                     timeout_ms;
 350         struct ib_sa_query      *query;
 351         int                     query_id;
 352         union {
 353                 struct ib_cm_id *ib;
 354                 struct iw_cm_id *iw;
 355         } cm_id;
 356
 357         u32                     seq_num;
 358         u32                     qkey;
 359         u32                     qp_num;
 360         pid_t                   owner;
 361         u32                     options;
 362         u8                      srq;
 363         u8                      tos;
 364         bool                    tos_set;
 365         u8                      reuseaddr;
 366         u8                      afonly;
 367         enum ib_gid_type        gid_type;
 368 };
 369
 370 struct cma_multicast {
 371         struct rdma_id_private *id_priv;
 372         union {
 373                 struct ib_sa_multicast *ib;
 374         } multicast;
 375         struct list_head        list;
 376         void                    *context;
 377         struct sockaddr_storage addr;
 378         struct kref             mcref;
 379         bool                    igmp_joined;
 380         u8                      join_state;
 381 };
 382
 383 struct cma_work {
 384         struct work_struct      work;
 385         struct rdma_id_private  *id;
 386         enum rdma_cm_state      old_state;
 387         enum rdma_cm_state      new_state;
 388         struct rdma_cm_event    event;
 389 };
 390
 391 struct cma_ndev_work {
 392         struct work_struct      work;
 393         struct rdma_id_private  *id;
 394         struct rdma_cm_event    event;
 395 };
 396
 397 struct iboe_mcast_work {
 398         struct work_struct       work;
 399         struct rdma_id_private  *id;
 400         struct cma_multicast    *mc;
 401 };
 402
 403 union cma_ip_addr {
 404         struct in6_addr ip6;
 405         struct {
 406                 __be32 pad[3];
 407                 __be32 addr;
 408         } ip4;
 409 };
 410
 411 struct cma_hdr {
 412         u8 cma_version;
 413         u8 ip_version;  /* IP version: 7:4 */
 414         __be16 port;
 415         union cma_ip_addr src_addr;
 416         union cma_ip_addr dst_addr;
 417 };
 418
 419 #define CMA_VERSION 0x00
 420
 421 struct cma_req_info {
 422         struct ib_device *device;
 423         int port;
 424         union ib_gid local_gid;
 425         __be64 service_id;
 426         u16 pkey;
 427         bool has_gid:1;
 428 };
 429
 430 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 431 {
 432         unsigned long flags;
 433         int ret;
 434
 435         spin_lock_irqsave(&id_priv->lock, flags);
 436         ret = (id_priv->state == comp);
 437         spin_unlock_irqrestore(&id_priv->lock, flags);
 438         return ret;
 439 }
 440
 441 static int cma_comp_exch(struct rdma_id_private *id_priv,
 442                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 443 {
 444         unsigned long flags;
 445         int ret;
 446
 447         spin_lock_irqsave(&id_priv->lock, flags);
 448         if ((ret = (id_priv->state == comp)))
 449                 id_priv->state = exch;
 450         spin_unlock_irqrestore(&id_priv->lock, flags);
 451         return ret;
 452 }
 453
 454 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 455                                    enum rdma_cm_state exch)
 456 {
 457         unsigned long flags;
 458         enum rdma_cm_state old;
 459
 460         spin_lock_irqsave(&id_priv->lock, flags);
 461         old = id_priv->state;
 462         id_priv->state = exch;
 463         spin_unlock_irqrestore(&id_priv->lock, flags);
 464         return old;
 465 }
 466
 467 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 468 {
 469         return hdr->ip_version >> 4;
 470 }
 471
 472 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 473 {
 474         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 475 }
 476
 477 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 478 {
 479         struct in_device *in_dev = NULL;
 480
 481         if (ndev) {
 482                 rtnl_lock();
 483                 in_dev = __in_dev_get_rtnl(ndev);
 484                 if (in_dev) {
 485                         if (join)
 486                                 ip_mc_inc_group(in_dev,
 487                                                 *(__be32 *)(mgid->raw + 12));
 488                         else
 489                                 ip_mc_dec_group(in_dev,
 490                                                 *(__be32 *)(mgid->raw + 12));
 491                 }
 492                 rtnl_unlock();
 493         }
 494         return (in_dev) ? 0 : -ENODEV;
 495 }
 496
 497 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 498                                struct cma_device *cma_dev)
 499 {
 500         cma_ref_dev(cma_dev);
 501         id_priv->cma_dev = cma_dev;
 502         id_priv->gid_type = 0;
 503         id_priv->id.device = cma_dev->device;
 504         id_priv->id.route.addr.dev_addr.transport =
 505                 rdma_node_get_transport(cma_dev->device->node_type);
 506         list_add_tail(&id_priv->list, &cma_dev->id_list);
 507 }
 508
 509 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 510                               struct cma_device *cma_dev)
 511 {
 512         _cma_attach_to_dev(id_priv, cma_dev);
 513         id_priv->gid_type =
 514                 cma_dev->default_gid_type[id_priv->id.port_num -
 515                                           rdma_start_port(cma_dev->device)];
 516 }
 517
 518 void cma_deref_dev(struct cma_device *cma_dev)
 519 {
 520         if (atomic_dec_and_test(&cma_dev->refcount))
 521                 complete(&cma_dev->comp);
 522 }
 523
 524 static inline void release_mc(struct kref *kref)
 525 {
 526         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 527
 528         kfree(mc->multicast.ib);
 529         kfree(mc);
 530 }
 531
 532 static void cma_release_dev(struct rdma_id_private *id_priv)
 533 {
 534         mutex_lock(&lock);
 535         list_del(&id_priv->list);
 536         cma_deref_dev(id_priv->cma_dev);
 537         id_priv->cma_dev = NULL;
 538         mutex_unlock(&lock);
 539 }
 540
 541 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 542 {
 543         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 544 }
 545
 546 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 547 {
 548         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 549 }
 550
 551 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 552 {
 553         return id_priv->id.route.addr.src_addr.ss_family;
 554 }
 555
 556 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 557 {
 558         struct ib_sa_mcmember_rec rec;
 559         int ret = 0;
 560
 561         if (id_priv->qkey) {
 562                 if (qkey && id_priv->qkey != qkey)
 563                         return -EINVAL;
 564                 return 0;
 565         }
 566
 567         if (qkey) {
 568                 id_priv->qkey = qkey;
 569                 return 0;
 570         }
 571
 572         switch (id_priv->id.ps) {
 573         case RDMA_PS_UDP:
 574         case RDMA_PS_IB:
 575                 id_priv->qkey = RDMA_UDP_QKEY;
 576                 break;
 577         case RDMA_PS_IPOIB:
 578                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 579                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 580                                              id_priv->id.port_num, &rec.mgid,
 581                                              &rec);
 582                 if (!ret)
 583                         id_priv->qkey = be32_to_cpu(rec.qkey);
 584                 break;
 585         default:
 586                 break;
 587         }
 588         return ret;
 589 }
 590
 591 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 592 {
 593         dev_addr->dev_type = ARPHRD_INFINIBAND;
 594         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 595         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 596 }
 597
 598 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 599 {
 600         int ret;
 601
 602         if (addr->sa_family != AF_IB) {
 603                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 604         } else {
 605                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 606                 ret = 0;
 607         }
 608
 609         return ret;
 610 }
 611
 612 static inline int cma_validate_port(struct ib_device *device, u8 port,
 613                                     enum ib_gid_type gid_type,
 614                                       union ib_gid *gid, int dev_type,
 615                                       int bound_if_index)
 616 {
 617         int ret = -ENODEV;
 618         struct net_device *ndev = NULL;
 619
 620         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 621                 return ret;
 622
 623         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 624                 return ret;
 625
 626         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port))
 627                 ndev = dev_get_by_index(&init_net, bound_if_index);
 628         else
 629                 gid_type = IB_GID_TYPE_IB;
 630
 631
 632         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 633                                          ndev, NULL);
 634
 635         if (ndev)
 636                 dev_put(ndev);
 637
 638         return ret;
 639 }
 640
 641 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 642                            struct rdma_id_private *listen_id_priv)
 643 {
 644         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 645         struct cma_device *cma_dev;
 646         union ib_gid gid, iboe_gid, *gidp;
 647         int ret = -ENODEV;
 648         u8 port;
 649
 650         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 651             id_priv->id.ps == RDMA_PS_IPOIB)
 652                 return -EINVAL;
 653
 654         mutex_lock(&lock);
 655         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 656                     &iboe_gid);
 657
 658         memcpy(&gid, dev_addr->src_dev_addr +
 659                rdma_addr_gid_offset(dev_addr), sizeof gid);
 660
 661         if (listen_id_priv) {
 662                 cma_dev = listen_id_priv->cma_dev;
 663                 port = listen_id_priv->id.port_num;
 664                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 665                        &iboe_gid : &gid;
 666
 667                 ret = cma_validate_port(cma_dev->device, port,
 668                                         rdma_protocol_ib(cma_dev->device, port) ?
 669                                         IB_GID_TYPE_IB :
 670                                         listen_id_priv->gid_type, gidp,
 671                                         dev_addr->dev_type,
 672                                         dev_addr->bound_dev_if);
 673                 if (!ret) {
 674                         id_priv->id.port_num = port;
 675                         goto out;
 676                 }
 677         }
 678
 679         list_for_each_entry(cma_dev, &dev_list, list) {
 680                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 681                         if (listen_id_priv &&
 682                             listen_id_priv->cma_dev == cma_dev &&
 683                             listen_id_priv->id.port_num == port)
 684                                 continue;
 685
 686                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 687                                &iboe_gid : &gid;
 688
 689                         ret = cma_validate_port(cma_dev->device, port,
 690                                                 rdma_protocol_ib(cma_dev->device, port) ?
 691                                                 IB_GID_TYPE_IB :
 692                                                 cma_dev->default_gid_type[port - 1],
 693                                                 gidp, dev_addr->dev_type,
 694                                                 dev_addr->bound_dev_if);
 695                         if (!ret) {
 696                                 id_priv->id.port_num = port;
 697                                 goto out;
 698                         }
 699                 }
 700         }
 701
 702 out:
 703         if (!ret)
 704                 cma_attach_to_dev(id_priv, cma_dev);
 705
 706         mutex_unlock(&lock);
 707         return ret;
 708 }
 709
 710 /*
 711  * Select the source IB device and address to reach the destination IB address.
 712  */
 713 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 714 {
 715         struct cma_device *cma_dev, *cur_dev;
 716         struct sockaddr_ib *addr;
 717         union ib_gid gid, sgid, *dgid;
 718         u16 pkey, index;
 719         u8 p;
 720         enum ib_port_state port_state;
 721         int i;
 722
 723         cma_dev = NULL;
 724         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 725         dgid = (union ib_gid *) &addr->sib_addr;
 726         pkey = ntohs(addr->sib_pkey);
 727
 728         list_for_each_entry(cur_dev, &dev_list, list) {
 729                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 730                         if (!rdma_cap_af_ib(cur_dev->device, p))
 731                                 continue;
 732
 733                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 734                                 continue;
 735
 736                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 737                                 continue;
 738                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 739                                                        &gid, NULL);
 740                              i++) {
 741                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 742                                         cma_dev = cur_dev;
 743                                         sgid = gid;
 744                                         id_priv->id.port_num = p;
 745                                         goto found;
 746                                 }
 747
 748                                 if (!cma_dev && (gid.global.subnet_prefix ==
 749                                     dgid->global.subnet_prefix) &&
 750                                     port_state == IB_PORT_ACTIVE) {
 751                                         cma_dev = cur_dev;
 752                                         sgid = gid;
 753                                         id_priv->id.port_num = p;
 754                                 }
 755                         }
 756                 }
 757         }
 758
 759         if (!cma_dev)
 760                 return -ENODEV;
 761
 762 found:
 763         cma_attach_to_dev(id_priv, cma_dev);
 764         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 765         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 766         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 767         return 0;
 768 }
 769
 770 static void cma_deref_id(struct rdma_id_private *id_priv)
 771 {
 772         if (atomic_dec_and_test(&id_priv->refcount))
 773                 complete(&id_priv->comp);
 774 }
 775
 776 struct rdma_cm_id *rdma_create_id(struct net *net,
 777                                   rdma_cm_event_handler event_handler,
 778                                   void *context, enum rdma_port_space ps,
 779                                   enum ib_qp_type qp_type)
 780 {
 781         struct rdma_id_private *id_priv;
 782
 783         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 784         if (!id_priv)
 785                 return ERR_PTR(-ENOMEM);
 786
 787         id_priv->owner = task_pid_nr(current);
 788         id_priv->state = RDMA_CM_IDLE;
 789         id_priv->id.context = context;
 790         id_priv->id.event_handler = event_handler;
 791         id_priv->id.ps = ps;
 792         id_priv->id.qp_type = qp_type;
 793         id_priv->tos_set = false;
 794         spin_lock_init(&id_priv->lock);
 795         mutex_init(&id_priv->qp_mutex);
 796         init_completion(&id_priv->comp);
 797         atomic_set(&id_priv->refcount, 1);
 798         mutex_init(&id_priv->handler_mutex);
 799         INIT_LIST_HEAD(&id_priv->listen_list);
 800         INIT_LIST_HEAD(&id_priv->mc_list);
 801         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 802         id_priv->id.route.addr.dev_addr.net = get_net(net);
 803
 804         return &id_priv->id;
 805 }
 806 EXPORT_SYMBOL(rdma_create_id);
 807
 808 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 809 {
 810         struct ib_qp_attr qp_attr;
 811         int qp_attr_mask, ret;
 812
 813         qp_attr.qp_state = IB_QPS_INIT;
 814         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 815         if (ret)
 816                 return ret;
 817
 818         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 819         if (ret)
 820                 return ret;
 821
 822         qp_attr.qp_state = IB_QPS_RTR;
 823         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 824         if (ret)
 825                 return ret;
 826
 827         qp_attr.qp_state = IB_QPS_RTS;
 828         qp_attr.sq_psn = 0;
 829         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 830
 831         return ret;
 832 }
 833
 834 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 835 {
 836         struct ib_qp_attr qp_attr;
 837         int qp_attr_mask, ret;
 838
 839         qp_attr.qp_state = IB_QPS_INIT;
 840         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 841         if (ret)
 842                 return ret;
 843
 844         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 845 }
 846
 847 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 848                    struct ib_qp_init_attr *qp_init_attr)
 849 {
 850         struct rdma_id_private *id_priv;
 851         struct ib_qp *qp;
 852         int ret;
 853
 854         id_priv = container_of(id, struct rdma_id_private, id);
 855         if (id->device != pd->device)
 856                 return -EINVAL;
 857
 858         qp_init_attr->port_num = id->port_num;
 859         qp = ib_create_qp(pd, qp_init_attr);
 860         if (IS_ERR(qp))
 861                 return PTR_ERR(qp);
 862
 863         if (id->qp_type == IB_QPT_UD)
 864                 ret = cma_init_ud_qp(id_priv, qp);
 865         else
 866                 ret = cma_init_conn_qp(id_priv, qp);
 867         if (ret)
 868                 goto err;
 869
 870         id->qp = qp;
 871         id_priv->qp_num = qp->qp_num;
 872         id_priv->srq = (qp->srq != NULL);
 873         return 0;
 874 err:
 875         ib_destroy_qp(qp);
 876         return ret;
 877 }
 878 EXPORT_SYMBOL(rdma_create_qp);
 879
 880 void rdma_destroy_qp(struct rdma_cm_id *id)
 881 {
 882         struct rdma_id_private *id_priv;
 883
 884         id_priv = container_of(id, struct rdma_id_private, id);
 885         mutex_lock(&id_priv->qp_mutex);
 886         ib_destroy_qp(id_priv->id.qp);
 887         id_priv->id.qp = NULL;
 888         mutex_unlock(&id_priv->qp_mutex);
 889 }
 890 EXPORT_SYMBOL(rdma_destroy_qp);
 891
 892 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 893                              struct rdma_conn_param *conn_param)
 894 {
 895         struct ib_qp_attr qp_attr;
 896         int qp_attr_mask, ret;
 897         union ib_gid sgid;
 898
 899         mutex_lock(&id_priv->qp_mutex);
 900         if (!id_priv->id.qp) {
 901                 ret = 0;
 902                 goto out;
 903         }
 904
 905         /* Need to update QP attributes from default values. */
 906         qp_attr.qp_state = IB_QPS_INIT;
 907         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 908         if (ret)
 909                 goto out;
 910
 911         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 912         if (ret)
 913                 goto out;
 914
 915         qp_attr.qp_state = IB_QPS_RTR;
 916         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 917         if (ret)
 918                 goto out;
 919
 920         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 921                            rdma_ah_read_grh(&qp_attr.ah_attr)->sgid_index,
 922                            &sgid, NULL);
 923         if (ret)
 924                 goto out;
 925
 926         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 927
 928         if (conn_param)
 929                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 930         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 931 out:
 932         mutex_unlock(&id_priv->qp_mutex);
 933         return ret;
 934 }
 935
 936 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 937                              struct rdma_conn_param *conn_param)
 938 {
 939         struct ib_qp_attr qp_attr;
 940         int qp_attr_mask, ret;
 941
 942         mutex_lock(&id_priv->qp_mutex);
 943         if (!id_priv->id.qp) {
 944                 ret = 0;
 945                 goto out;
 946         }
 947
 948         qp_attr.qp_state = IB_QPS_RTS;
 949         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 950         if (ret)
 951                 goto out;
 952
 953         if (conn_param)
 954                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 955         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 956 out:
 957         mutex_unlock(&id_priv->qp_mutex);
 958         return ret;
 959 }
 960
 961 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 962 {
 963         struct ib_qp_attr qp_attr;
 964         int ret;
 965
 966         mutex_lock(&id_priv->qp_mutex);
 967         if (!id_priv->id.qp) {
 968                 ret = 0;
 969                 goto out;
 970         }
 971
 972         qp_attr.qp_state = IB_QPS_ERR;
 973         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 974 out:
 975         mutex_unlock(&id_priv->qp_mutex);
 976         return ret;
 977 }
 978
 979 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 980                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 981 {
 982         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 983         int ret;
 984         u16 pkey;
 985
 986         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 987                 pkey = 0xffff;
 988         else
 989                 pkey = ib_addr_get_pkey(dev_addr);
 990
 991         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 992                                   pkey, &qp_attr->pkey_index);
 993         if (ret)
 994                 return ret;
 995
 996         qp_attr->port_num = id_priv->id.port_num;
 997         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 998
 999         if (id_priv->id.qp_type == IB_QPT_UD) {
1000                 ret = cma_set_qkey(id_priv, 0);
1001                 if (ret)
1002                         return ret;
1003
1004                 qp_attr->qkey = id_priv->qkey;
1005                 *qp_attr_mask |= IB_QP_QKEY;
1006         } else {
1007                 qp_attr->qp_access_flags = 0;
1008                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1009         }
1010         return 0;
1011 }
1012
1013 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1014                        int *qp_attr_mask)
1015 {
1016         struct rdma_id_private *id_priv;
1017         int ret = 0;
1018
1019         id_priv = container_of(id, struct rdma_id_private, id);
1020         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1021                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1022                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1023                 else
1024                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1025                                                  qp_attr_mask);
1026
1027                 if (qp_attr->qp_state == IB_QPS_RTR)
1028                         qp_attr->rq_psn = id_priv->seq_num;
1029         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1030                 if (!id_priv->cm_id.iw) {
1031                         qp_attr->qp_access_flags = 0;
1032                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1033                 } else
1034                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1035                                                  qp_attr_mask);
1036                 qp_attr->port_num = id_priv->id.port_num;
1037                 *qp_attr_mask |= IB_QP_PORT;
1038         } else
1039                 ret = -ENOSYS;
1040
1041         return ret;
1042 }
1043 EXPORT_SYMBOL(rdma_init_qp_attr);
1044
1045 static inline int cma_zero_addr(struct sockaddr *addr)
1046 {
1047         switch (addr->sa_family) {
1048         case AF_INET:
1049                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1050         case AF_INET6:
1051                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1052         case AF_IB:
1053                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1054         default:
1055                 return 0;
1056         }
1057 }
1058
1059 static inline int cma_loopback_addr(struct sockaddr *addr)
1060 {
1061         switch (addr->sa_family) {
1062         case AF_INET:
1063                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1064         case AF_INET6:
1065                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1066         case AF_IB:
1067                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1068         default:
1069                 return 0;
1070         }
1071 }
1072
1073 static inline int cma_any_addr(struct sockaddr *addr)
1074 {
1075         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1076 }
1077
1078 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1079 {
1080         if (src->sa_family != dst->sa_family)
1081                 return -1;
1082
1083         switch (src->sa_family) {
1084         case AF_INET:
1085                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1086                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1087         case AF_INET6:
1088                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1089                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1090         default:
1091                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1092                                    &((struct sockaddr_ib *) dst)->sib_addr);
1093         }
1094 }
1095
1096 static __be16 cma_port(struct sockaddr *addr)
1097 {
1098         struct sockaddr_ib *sib;
1099
1100         switch (addr->sa_family) {
1101         case AF_INET:
1102                 return ((struct sockaddr_in *) addr)->sin_port;
1103         case AF_INET6:
1104                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1105         case AF_IB:
1106                 sib = (struct sockaddr_ib *) addr;
1107                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1108                                     be64_to_cpu(sib->sib_sid_mask)));
1109         default:
1110                 return 0;
1111         }
1112 }
1113
1114 static inline int cma_any_port(struct sockaddr *addr)
1115 {
1116         return !cma_port(addr);
1117 }
1118
1119 static void cma_save_ib_info(struct sockaddr *src_addr,
1120                              struct sockaddr *dst_addr,
1121                              struct rdma_cm_id *listen_id,
1122                              struct sa_path_rec *path)
1123 {
1124         struct sockaddr_ib *listen_ib, *ib;
1125
1126         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1127         if (src_addr) {
1128                 ib = (struct sockaddr_ib *)src_addr;
1129                 ib->sib_family = AF_IB;
1130                 if (path) {
1131                         ib->sib_pkey = path->pkey;
1132                         ib->sib_flowinfo = path->flow_label;
1133                         memcpy(&ib->sib_addr, &path->sgid, 16);
1134                         ib->sib_sid = path->service_id;
1135                         ib->sib_scope_id = 0;
1136                 } else {
1137                         ib->sib_pkey = listen_ib->sib_pkey;
1138                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1139                         ib->sib_addr = listen_ib->sib_addr;
1140                         ib->sib_sid = listen_ib->sib_sid;
1141                         ib->sib_scope_id = listen_ib->sib_scope_id;
1142                 }
1143                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1144         }
1145         if (dst_addr) {
1146                 ib = (struct sockaddr_ib *)dst_addr;
1147                 ib->sib_family = AF_IB;
1148                 if (path) {
1149                         ib->sib_pkey = path->pkey;
1150                         ib->sib_flowinfo = path->flow_label;
1151                         memcpy(&ib->sib_addr, &path->dgid, 16);
1152                 }
1153         }
1154 }
1155
1156 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1157                               struct sockaddr_in *dst_addr,
1158                               struct cma_hdr *hdr,
1159                               __be16 local_port)
1160 {
1161         if (src_addr) {
1162                 *src_addr = (struct sockaddr_in) {
1163                         .sin_family = AF_INET,
1164                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1165                         .sin_port = local_port,
1166                 };
1167         }
1168
1169         if (dst_addr) {
1170                 *dst_addr = (struct sockaddr_in) {
1171                         .sin_family = AF_INET,
1172                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1173                         .sin_port = hdr->port,
1174                 };
1175         }
1176 }
1177
1178 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1179                               struct sockaddr_in6 *dst_addr,
1180                               struct cma_hdr *hdr,
1181                               __be16 local_port)
1182 {
1183         if (src_addr) {
1184                 *src_addr = (struct sockaddr_in6) {
1185                         .sin6_family = AF_INET6,
1186                         .sin6_addr = hdr->dst_addr.ip6,
1187                         .sin6_port = local_port,
1188                 };
1189         }
1190
1191         if (dst_addr) {
1192                 *dst_addr = (struct sockaddr_in6) {
1193                         .sin6_family = AF_INET6,
1194                         .sin6_addr = hdr->src_addr.ip6,
1195                         .sin6_port = hdr->port,
1196                 };
1197         }
1198 }
1199
1200 static u16 cma_port_from_service_id(__be64 service_id)
1201 {
1202         return (u16)be64_to_cpu(service_id);
1203 }
1204
1205 static int cma_save_ip_info(struct sockaddr *src_addr,
1206                             struct sockaddr *dst_addr,
1207                             struct ib_cm_event *ib_event,
1208                             __be64 service_id)
1209 {
1210         struct cma_hdr *hdr;
1211         __be16 port;
1212
1213         hdr = ib_event->private_data;
1214         if (hdr->cma_version != CMA_VERSION)
1215                 return -EINVAL;
1216
1217         port = htons(cma_port_from_service_id(service_id));
1218
1219         switch (cma_get_ip_ver(hdr)) {
1220         case 4:
1221                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1222                                   (struct sockaddr_in *)dst_addr, hdr, port);
1223                 break;
1224         case 6:
1225                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1226                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1227                 break;
1228         default:
1229                 return -EAFNOSUPPORT;
1230         }
1231
1232         return 0;
1233 }
1234
1235 static int cma_save_net_info(struct sockaddr *src_addr,
1236                              struct sockaddr *dst_addr,
1237                              struct rdma_cm_id *listen_id,
1238                              struct ib_cm_event *ib_event,
1239                              sa_family_t sa_family, __be64 service_id)
1240 {
1241         if (sa_family == AF_IB) {
1242                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1243                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1244                                          ib_event->param.req_rcvd.primary_path);
1245                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1246                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1247                 return 0;
1248         }
1249
1250         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1251 }
1252
1253 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1254                              struct cma_req_info *req)
1255 {
1256         const struct ib_cm_req_event_param *req_param =
1257                 &ib_event->param.req_rcvd;
1258         const struct ib_cm_sidr_req_event_param *sidr_param =
1259                 &ib_event->param.sidr_req_rcvd;
1260
1261         switch (ib_event->event) {
1262         case IB_CM_REQ_RECEIVED:
1263                 req->device     = req_param->listen_id->device;
1264                 req->port       = req_param->port;
1265                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1266                        sizeof(req->local_gid));
1267                 req->has_gid    = true;
1268                 req->service_id = req_param->primary_path->service_id;
1269                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1270                 if (req->pkey != req_param->bth_pkey)
1271                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1272                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1273                                             req_param->bth_pkey, req->pkey);
1274                 break;
1275         case IB_CM_SIDR_REQ_RECEIVED:
1276                 req->device     = sidr_param->listen_id->device;
1277                 req->port       = sidr_param->port;
1278                 req->has_gid    = false;
1279                 req->service_id = sidr_param->service_id;
1280                 req->pkey       = sidr_param->pkey;
1281                 if (req->pkey != sidr_param->bth_pkey)
1282                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1283                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1284                                             sidr_param->bth_pkey, req->pkey);
1285                 break;
1286         default:
1287                 return -EINVAL;
1288         }
1289
1290         return 0;
1291 }
1292
1293 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1294                                   const struct sockaddr_in *dst_addr,
1295                                   const struct sockaddr_in *src_addr)
1296 {
1297         __be32 daddr = dst_addr->sin_addr.s_addr,
1298                saddr = src_addr->sin_addr.s_addr;
1299         struct fib_result res;
1300         struct flowi4 fl4;
1301         int err;
1302         bool ret;
1303
1304         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1305             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1306             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1307             ipv4_is_loopback(saddr))
1308                 return false;
1309
1310         memset(&fl4, 0, sizeof(fl4));
1311         fl4.flowi4_iif = net_dev->ifindex;
1312         fl4.daddr = daddr;
1313         fl4.saddr = saddr;
1314
1315         rcu_read_lock();
1316         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1317         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1318         rcu_read_unlock();
1319
1320         return ret;
1321 }
1322
1323 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1324                                   const struct sockaddr_in6 *dst_addr,
1325                                   const struct sockaddr_in6 *src_addr)
1326 {
1327 #if IS_ENABLED(CONFIG_IPV6)
1328         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1329                            IPV6_ADDR_LINKLOCAL;
1330         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1331                                          &src_addr->sin6_addr, net_dev->ifindex,
1332                                          strict);
1333         bool ret;
1334
1335         if (!rt)
1336                 return false;
1337
1338         ret = rt->rt6i_idev->dev == net_dev;
1339         ip6_rt_put(rt);
1340
1341         return ret;
1342 #else
1343         return false;
1344 #endif
1345 }
1346
1347 static bool validate_net_dev(struct net_device *net_dev,
1348                              const struct sockaddr *daddr,
1349                              const struct sockaddr *saddr)
1350 {
1351         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1352         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1353         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1354         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1355
1356         switch (daddr->sa_family) {
1357         case AF_INET:
1358                 return saddr->sa_family == AF_INET &&
1359                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1360
1361         case AF_INET6:
1362                 return saddr->sa_family == AF_INET6 &&
1363                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1364
1365         default:
1366                 return false;
1367         }
1368 }
1369
1370 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1371                                           const struct cma_req_info *req)
1372 {
1373         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1374         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1375                         *src_addr = (struct sockaddr *)&src_addr_storage;
1376         struct net_device *net_dev;
1377         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1378         int err;
1379
1380         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1381                                req->service_id);
1382         if (err)
1383                 return ERR_PTR(err);
1384
1385         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1386                                            gid, listen_addr);
1387         if (!net_dev)
1388                 return ERR_PTR(-ENODEV);
1389
1390         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1391                 dev_put(net_dev);
1392                 return ERR_PTR(-EHOSTUNREACH);
1393         }
1394
1395         return net_dev;
1396 }
1397
1398 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1399 {
1400         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1401 }
1402
1403 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1404                                    const struct cma_hdr *hdr)
1405 {
1406         struct sockaddr *addr = cma_src_addr(id_priv);
1407         __be32 ip4_addr;
1408         struct in6_addr ip6_addr;
1409
1410         if (cma_any_addr(addr) && !id_priv->afonly)
1411                 return true;
1412
1413         switch (addr->sa_family) {
1414         case AF_INET:
1415                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1416                 if (cma_get_ip_ver(hdr) != 4)
1417                         return false;
1418                 if (!cma_any_addr(addr) &&
1419                     hdr->dst_addr.ip4.addr != ip4_addr)
1420                         return false;
1421                 break;
1422         case AF_INET6:
1423                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1424                 if (cma_get_ip_ver(hdr) != 6)
1425                         return false;
1426                 if (!cma_any_addr(addr) &&
1427                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1428                         return false;
1429                 break;
1430         case AF_IB:
1431                 return true;
1432         default:
1433                 return false;
1434         }
1435
1436         return true;
1437 }
1438
1439 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1440 {
1441         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1442         enum rdma_transport_type transport =
1443                 rdma_node_get_transport(device->node_type);
1444
1445         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1446 }
1447
1448 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1449 {
1450         struct ib_device *device = id->device;
1451         const int port_num = id->port_num ?: rdma_start_port(device);
1452
1453         return cma_protocol_roce_dev_port(device, port_num);
1454 }
1455
1456 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1457                               const struct net_device *net_dev,
1458                               u8 port_num)
1459 {
1460         const struct rdma_addr *addr = &id->route.addr;
1461
1462         if (!net_dev)
1463                 /* This request is an AF_IB request or a RoCE request */
1464                 return (!id->port_num || id->port_num == port_num) &&
1465                        (addr->src_addr.ss_family == AF_IB ||
1466                         cma_protocol_roce_dev_port(id->device, port_num));
1467
1468         return !addr->dev_addr.bound_dev_if ||
1469                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1470                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1471 }
1472
1473 static struct rdma_id_private *cma_find_listener(
1474                 const struct rdma_bind_list *bind_list,
1475                 const struct ib_cm_id *cm_id,
1476                 const struct ib_cm_event *ib_event,
1477                 const struct cma_req_info *req,
1478                 const struct net_device *net_dev)
1479 {
1480         struct rdma_id_private *id_priv, *id_priv_dev;
1481
1482         if (!bind_list)
1483                 return ERR_PTR(-EINVAL);
1484
1485         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1486                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1487                         if (id_priv->id.device == cm_id->device &&
1488                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1489                                 return id_priv;
1490                         list_for_each_entry(id_priv_dev,
1491                                             &id_priv->listen_list,
1492                                             listen_list) {
1493                                 if (id_priv_dev->id.device == cm_id->device &&
1494                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1495                                         return id_priv_dev;
1496                         }
1497                 }
1498         }
1499
1500         return ERR_PTR(-EINVAL);
1501 }
1502
1503 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1504                                                  struct ib_cm_event *ib_event,
1505                                                  struct net_device **net_dev)
1506 {
1507         struct cma_req_info req;
1508         struct rdma_bind_list *bind_list;
1509         struct rdma_id_private *id_priv;
1510         int err;
1511
1512         err = cma_save_req_info(ib_event, &req);
1513         if (err)
1514                 return ERR_PTR(err);
1515
1516         *net_dev = cma_get_net_dev(ib_event, &req);
1517         if (IS_ERR(*net_dev)) {
1518                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1519                         /* Assuming the protocol is AF_IB */
1520                         *net_dev = NULL;
1521                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1522                         /* TODO find the net dev matching the request parameters
1523                          * through the RoCE GID table */
1524                         *net_dev = NULL;
1525                 } else {
1526                         return ERR_CAST(*net_dev);
1527                 }
1528         }
1529
1530         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1531                                 rdma_ps_from_service_id(req.service_id),
1532                                 cma_port_from_service_id(req.service_id));
1533         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1534         if (IS_ERR(id_priv) && *net_dev) {
1535                 dev_put(*net_dev);
1536                 *net_dev = NULL;
1537         }
1538
1539         return id_priv;
1540 }
1541
1542 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1543 {
1544         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1545 }
1546
1547 static void cma_cancel_route(struct rdma_id_private *id_priv)
1548 {
1549         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1550                 if (id_priv->query)
1551                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1552         }
1553 }
1554
1555 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1556 {
1557         struct rdma_id_private *dev_id_priv;
1558
1559         /*
1560          * Remove from listen_any_list to prevent added devices from spawning
1561          * additional listen requests.
1562          */
1563         mutex_lock(&lock);
1564         list_del(&id_priv->list);
1565
1566         while (!list_empty(&id_priv->listen_list)) {
1567                 dev_id_priv = list_entry(id_priv->listen_list.next,
1568                                          struct rdma_id_private, listen_list);
1569                 /* sync with device removal to avoid duplicate destruction */
1570                 list_del_init(&dev_id_priv->list);
1571                 list_del(&dev_id_priv->listen_list);
1572                 mutex_unlock(&lock);
1573
1574                 rdma_destroy_id(&dev_id_priv->id);
1575                 mutex_lock(&lock);
1576         }
1577         mutex_unlock(&lock);
1578 }
1579
1580 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1581                                  enum rdma_cm_state state)
1582 {
1583         switch (state) {
1584         case RDMA_CM_ADDR_QUERY:
1585                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1586                 break;
1587         case RDMA_CM_ROUTE_QUERY:
1588                 cma_cancel_route(id_priv);
1589                 break;
1590         case RDMA_CM_LISTEN:
1591                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1592                         cma_cancel_listens(id_priv);
1593                 break;
1594         default:
1595                 break;
1596         }
1597 }
1598
1599 static void cma_release_port(struct rdma_id_private *id_priv)
1600 {
1601         struct rdma_bind_list *bind_list = id_priv->bind_list;
1602         struct net *net = id_priv->id.route.addr.dev_addr.net;
1603
1604         if (!bind_list)
1605                 return;
1606
1607         mutex_lock(&lock);
1608         hlist_del(&id_priv->node);
1609         if (hlist_empty(&bind_list->owners)) {
1610                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1611                 kfree(bind_list);
1612         }
1613         mutex_unlock(&lock);
1614 }
1615
1616 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1617 {
1618         struct cma_multicast *mc;
1619
1620         while (!list_empty(&id_priv->mc_list)) {
1621                 mc = container_of(id_priv->mc_list.next,
1622                                   struct cma_multicast, list);
1623                 list_del(&mc->list);
1624                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1625                                       id_priv->id.port_num)) {
1626                         ib_sa_free_multicast(mc->multicast.ib);
1627                         kfree(mc);
1628                 } else {
1629                         if (mc->igmp_joined) {
1630                                 struct rdma_dev_addr *dev_addr =
1631                                         &id_priv->id.route.addr.dev_addr;
1632                                 struct net_device *ndev = NULL;
1633
1634                                 if (dev_addr->bound_dev_if)
1635                                         ndev = dev_get_by_index(&init_net,
1636                                                                 dev_addr->bound_dev_if);
1637                                 if (ndev) {
1638                                         cma_igmp_send(ndev,
1639                                                       &mc->multicast.ib->rec.mgid,
1640                                                       false);
1641                                         dev_put(ndev);
1642                                 }
1643                         }
1644                         kref_put(&mc->mcref, release_mc);
1645                 }
1646         }
1647 }
1648
1649 void rdma_destroy_id(struct rdma_cm_id *id)
1650 {
1651         struct rdma_id_private *id_priv;
1652         enum rdma_cm_state state;
1653
1654         id_priv = container_of(id, struct rdma_id_private, id);
1655         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1656         cma_cancel_operation(id_priv, state);
1657
1658         /*
1659          * Wait for any active callback to finish.  New callbacks will find
1660          * the id_priv state set to destroying and abort.
1661          */
1662         mutex_lock(&id_priv->handler_mutex);
1663         mutex_unlock(&id_priv->handler_mutex);
1664
1665         if (id_priv->cma_dev) {
1666                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1667                         if (id_priv->cm_id.ib)
1668                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1669                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1670                         if (id_priv->cm_id.iw)
1671                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1672                 }
1673                 cma_leave_mc_groups(id_priv);
1674                 cma_release_dev(id_priv);
1675         }
1676
1677         cma_release_port(id_priv);
1678         cma_deref_id(id_priv);
1679         wait_for_completion(&id_priv->comp);
1680
1681         if (id_priv->internal_id)
1682                 cma_deref_id(id_priv->id.context);
1683
1684         kfree(id_priv->id.route.path_rec);
1685         put_net(id_priv->id.route.addr.dev_addr.net);
1686         kfree(id_priv);
1687 }
1688 EXPORT_SYMBOL(rdma_destroy_id);
1689
1690 static int cma_rep_recv(struct rdma_id_private *id_priv)
1691 {
1692         int ret;
1693
1694         ret = cma_modify_qp_rtr(id_priv, NULL);
1695         if (ret)
1696                 goto reject;
1697
1698         ret = cma_modify_qp_rts(id_priv, NULL);
1699         if (ret)
1700                 goto reject;
1701
1702         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1703         if (ret)
1704                 goto reject;
1705
1706         return 0;
1707 reject:
1708         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1709         cma_modify_qp_err(id_priv);
1710         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1711                        NULL, 0, NULL, 0);
1712         return ret;
1713 }
1714
1715 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1716                                    struct ib_cm_rep_event_param *rep_data,
1717                                    void *private_data)
1718 {
1719         event->param.conn.private_data = private_data;
1720         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1721         event->param.conn.responder_resources = rep_data->responder_resources;
1722         event->param.conn.initiator_depth = rep_data->initiator_depth;
1723         event->param.conn.flow_control = rep_data->flow_control;
1724         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1725         event->param.conn.srq = rep_data->srq;
1726         event->param.conn.qp_num = rep_data->remote_qpn;
1727 }
1728
1729 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1730 {
1731         struct rdma_id_private *id_priv = cm_id->context;
1732         struct rdma_cm_event event;
1733         int ret = 0;
1734
1735         mutex_lock(&id_priv->handler_mutex);
1736         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1737              id_priv->state != RDMA_CM_CONNECT) ||
1738             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1739              id_priv->state != RDMA_CM_DISCONNECT))
1740                 goto out;
1741
1742         memset(&event, 0, sizeof event);
1743         switch (ib_event->event) {
1744         case IB_CM_REQ_ERROR:
1745         case IB_CM_REP_ERROR:
1746                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1747                 event.status = -ETIMEDOUT;
1748                 break;
1749         case IB_CM_REP_RECEIVED:
1750                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1751                     (id_priv->id.qp_type != IB_QPT_UD))
1752                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1753                 if (id_priv->id.qp) {
1754                         event.status = cma_rep_recv(id_priv);
1755                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1756                                                      RDMA_CM_EVENT_ESTABLISHED;
1757                 } else {
1758                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1759                 }
1760                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1761                                        ib_event->private_data);
1762                 break;
1763         case IB_CM_RTU_RECEIVED:
1764         case IB_CM_USER_ESTABLISHED:
1765                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1766                 break;
1767         case IB_CM_DREQ_ERROR:
1768                 event.status = -ETIMEDOUT; /* fall through */
1769         case IB_CM_DREQ_RECEIVED:
1770         case IB_CM_DREP_RECEIVED:
1771                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1772                                    RDMA_CM_DISCONNECT))
1773                         goto out;
1774                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1775                 break;
1776         case IB_CM_TIMEWAIT_EXIT:
1777                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1778                 break;
1779         case IB_CM_MRA_RECEIVED:
1780                 /* ignore event */
1781                 goto out;
1782         case IB_CM_REJ_RECEIVED:
1783                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1784                                                                                 ib_event->param.rej_rcvd.reason));
1785                 cma_modify_qp_err(id_priv);
1786                 event.status = ib_event->param.rej_rcvd.reason;
1787                 event.event = RDMA_CM_EVENT_REJECTED;
1788                 event.param.conn.private_data = ib_event->private_data;
1789                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1790                 break;
1791         default:
1792                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1793                        ib_event->event);
1794                 goto out;
1795         }
1796
1797         ret = id_priv->id.event_handler(&id_priv->id, &event);
1798         if (ret) {
1799                 /* Destroy the CM ID by returning a non-zero value. */
1800                 id_priv->cm_id.ib = NULL;
1801                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1802                 mutex_unlock(&id_priv->handler_mutex);
1803                 rdma_destroy_id(&id_priv->id);
1804                 return ret;
1805         }
1806 out:
1807         mutex_unlock(&id_priv->handler_mutex);
1808         return ret;
1809 }
1810
1811 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1812                                                struct ib_cm_event *ib_event,
1813                                                struct net_device *net_dev)
1814 {
1815         struct rdma_id_private *id_priv;
1816         struct rdma_cm_id *id;
1817         struct rdma_route *rt;
1818         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1819         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1820         const __be64 service_id =
1821                 ib_event->param.req_rcvd.primary_path->service_id;
1822         int ret;
1823
1824         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1825                             listen_id->event_handler, listen_id->context,
1826                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1827         if (IS_ERR(id))
1828                 return NULL;
1829
1830         id_priv = container_of(id, struct rdma_id_private, id);
1831         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1832                               (struct sockaddr *)&id->route.addr.dst_addr,
1833                               listen_id, ib_event, ss_family, service_id))
1834                 goto err;
1835
1836         rt = &id->route;
1837         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1838         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1839                                GFP_KERNEL);
1840         if (!rt->path_rec)
1841                 goto err;
1842
1843         rt->path_rec[0] = *path;
1844         if (rt->num_paths == 2)
1845                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1846
1847         if (net_dev) {
1848                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1849                 if (ret)
1850                         goto err;
1851         } else {
1852                 if (!cma_protocol_roce(listen_id) &&
1853                     cma_any_addr(cma_src_addr(id_priv))) {
1854                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1855                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1856                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1857                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1858                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1859                         if (ret)
1860                                 goto err;
1861                 }
1862         }
1863         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1864
1865         id_priv->state = RDMA_CM_CONNECT;
1866         return id_priv;
1867
1868 err:
1869         rdma_destroy_id(id);
1870         return NULL;
1871 }
1872
1873 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1874                                               struct ib_cm_event *ib_event,
1875                                               struct net_device *net_dev)
1876 {
1877         struct rdma_id_private *id_priv;
1878         struct rdma_cm_id *id;
1879         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1880         struct net *net = listen_id->route.addr.dev_addr.net;
1881         int ret;
1882
1883         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1884                             listen_id->ps, IB_QPT_UD);
1885         if (IS_ERR(id))
1886                 return NULL;
1887
1888         id_priv = container_of(id, struct rdma_id_private, id);
1889         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1890                               (struct sockaddr *)&id->route.addr.dst_addr,
1891                               listen_id, ib_event, ss_family,
1892                               ib_event->param.sidr_req_rcvd.service_id))
1893                 goto err;
1894
1895         if (net_dev) {
1896                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1897                 if (ret)
1898                         goto err;
1899         } else {
1900                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1901                         ret = cma_translate_addr(cma_src_addr(id_priv),
1902                                                  &id->route.addr.dev_addr);
1903                         if (ret)
1904                                 goto err;
1905                 }
1906         }
1907
1908         id_priv->state = RDMA_CM_CONNECT;
1909         return id_priv;
1910 err:
1911         rdma_destroy_id(id);
1912         return NULL;
1913 }
1914
1915 static void cma_set_req_event_data(struct rdma_cm_event *event,
1916                                    struct ib_cm_req_event_param *req_data,
1917                                    void *private_data, int offset)
1918 {
1919         event->param.conn.private_data = private_data + offset;
1920         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1921         event->param.conn.responder_resources = req_data->responder_resources;
1922         event->param.conn.initiator_depth = req_data->initiator_depth;
1923         event->param.conn.flow_control = req_data->flow_control;
1924         event->param.conn.retry_count = req_data->retry_count;
1925         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1926         event->param.conn.srq = req_data->srq;
1927         event->param.conn.qp_num = req_data->remote_qpn;
1928 }
1929
1930 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1931 {
1932         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1933                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1934                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1935                  (id->qp_type == IB_QPT_UD)) ||
1936                 (!id->qp_type));
1937 }
1938
1939 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1940 {
1941         struct rdma_id_private *listen_id, *conn_id = NULL;
1942         struct rdma_cm_event event;
1943         struct net_device *net_dev;
1944         int offset, ret;
1945
1946         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1947         if (IS_ERR(listen_id))
1948                 return PTR_ERR(listen_id);
1949
1950         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1951                 ret = -EINVAL;
1952                 goto net_dev_put;
1953         }
1954
1955         mutex_lock(&listen_id->handler_mutex);
1956         if (listen_id->state != RDMA_CM_LISTEN) {
1957                 ret = -ECONNABORTED;
1958                 goto err1;
1959         }
1960
1961         memset(&event, 0, sizeof event);
1962         offset = cma_user_data_offset(listen_id);
1963         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1964         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1965                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1966                 event.param.ud.private_data = ib_event->private_data + offset;
1967                 event.param.ud.private_data_len =
1968                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1969         } else {
1970                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1971                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1972                                        ib_event->private_data, offset);
1973         }
1974         if (!conn_id) {
1975                 ret = -ENOMEM;
1976                 goto err1;
1977         }
1978
1979         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1980         ret = cma_acquire_dev(conn_id, listen_id);
1981         if (ret)
1982                 goto err2;
1983
1984         conn_id->cm_id.ib = cm_id;
1985         cm_id->context = conn_id;
1986         cm_id->cm_handler = cma_ib_handler;
1987
1988         /*
1989          * Protect against the user destroying conn_id from another thread
1990          * until we're done accessing it.
1991          */
1992         atomic_inc(&conn_id->refcount);
1993         ret = conn_id->id.event_handler(&conn_id->id, &event);
1994         if (ret)
1995                 goto err3;
1996         /*
1997          * Acquire mutex to prevent user executing rdma_destroy_id()
1998          * while we're accessing the cm_id.
1999          */
2000         mutex_lock(&lock);
2001         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2002             (conn_id->id.qp_type != IB_QPT_UD))
2003                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2004         mutex_unlock(&lock);
2005         mutex_unlock(&conn_id->handler_mutex);
2006         mutex_unlock(&listen_id->handler_mutex);
2007         cma_deref_id(conn_id);
2008         if (net_dev)
2009                 dev_put(net_dev);
2010         return 0;
2011
2012 err3:
2013         cma_deref_id(conn_id);
2014         /* Destroy the CM ID by returning a non-zero value. */
2015         conn_id->cm_id.ib = NULL;
2016 err2:
2017         cma_exch(conn_id, RDMA_CM_DESTROYING);
2018         mutex_unlock(&conn_id->handler_mutex);
2019 err1:
2020         mutex_unlock(&listen_id->handler_mutex);
2021         if (conn_id)
2022                 rdma_destroy_id(&conn_id->id);
2023
2024 net_dev_put:
2025         if (net_dev)
2026                 dev_put(net_dev);
2027
2028         return ret;
2029 }
2030
2031 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2032 {
2033         if (addr->sa_family == AF_IB)
2034                 return ((struct sockaddr_ib *) addr)->sib_sid;
2035
2036         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2037 }
2038 EXPORT_SYMBOL(rdma_get_service_id);
2039
2040 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2041 {
2042         struct rdma_id_private *id_priv = iw_id->context;
2043         struct rdma_cm_event event;
2044         int ret = 0;
2045         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2046         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2047
2048         mutex_lock(&id_priv->handler_mutex);
2049         if (id_priv->state != RDMA_CM_CONNECT)
2050                 goto out;
2051
2052         memset(&event, 0, sizeof event);
2053         switch (iw_event->event) {
2054         case IW_CM_EVENT_CLOSE:
2055                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2056                 break;
2057         case IW_CM_EVENT_CONNECT_REPLY:
2058                 memcpy(cma_src_addr(id_priv), laddr,
2059                        rdma_addr_size(laddr));
2060                 memcpy(cma_dst_addr(id_priv), raddr,
2061                        rdma_addr_size(raddr));
2062                 switch (iw_event->status) {
2063                 case 0:
2064                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2065                         event.param.conn.initiator_depth = iw_event->ird;
2066                         event.param.conn.responder_resources = iw_event->ord;
2067                         break;
2068                 case -ECONNRESET:
2069                 case -ECONNREFUSED:
2070                         event.event = RDMA_CM_EVENT_REJECTED;
2071                         break;
2072                 case -ETIMEDOUT:
2073                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2074                         break;
2075                 default:
2076                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2077                         break;
2078                 }
2079                 break;
2080         case IW_CM_EVENT_ESTABLISHED:
2081                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2082                 event.param.conn.initiator_depth = iw_event->ird;
2083                 event.param.conn.responder_resources = iw_event->ord;
2084                 break;
2085         default:
2086                 BUG_ON(1);
2087         }
2088
2089         event.status = iw_event->status;
2090         event.param.conn.private_data = iw_event->private_data;
2091         event.param.conn.private_data_len = iw_event->private_data_len;
2092         ret = id_priv->id.event_handler(&id_priv->id, &event);
2093         if (ret) {
2094                 /* Destroy the CM ID by returning a non-zero value. */
2095                 id_priv->cm_id.iw = NULL;
2096                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2097                 mutex_unlock(&id_priv->handler_mutex);
2098                 rdma_destroy_id(&id_priv->id);
2099                 return ret;
2100         }
2101
2102 out:
2103         mutex_unlock(&id_priv->handler_mutex);
2104         return ret;
2105 }
2106
2107 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2108                                struct iw_cm_event *iw_event)
2109 {
2110         struct rdma_cm_id *new_cm_id;
2111         struct rdma_id_private *listen_id, *conn_id;
2112         struct rdma_cm_event event;
2113         int ret = -ECONNABORTED;
2114         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2115         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2116
2117         listen_id = cm_id->context;
2118
2119         mutex_lock(&listen_id->handler_mutex);
2120         if (listen_id->state != RDMA_CM_LISTEN)
2121                 goto out;
2122
2123         /* Create a new RDMA id for the new IW CM ID */
2124         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2125                                    listen_id->id.event_handler,
2126                                    listen_id->id.context,
2127                                    RDMA_PS_TCP, IB_QPT_RC);
2128         if (IS_ERR(new_cm_id)) {
2129                 ret = -ENOMEM;
2130                 goto out;
2131         }
2132         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2133         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2134         conn_id->state = RDMA_CM_CONNECT;
2135
2136         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2137         if (ret) {
2138                 mutex_unlock(&conn_id->handler_mutex);
2139                 rdma_destroy_id(new_cm_id);
2140                 goto out;
2141         }
2142
2143         ret = cma_acquire_dev(conn_id, listen_id);
2144         if (ret) {
2145                 mutex_unlock(&conn_id->handler_mutex);
2146                 rdma_destroy_id(new_cm_id);
2147                 goto out;
2148         }
2149
2150         conn_id->cm_id.iw = cm_id;
2151         cm_id->context = conn_id;
2152         cm_id->cm_handler = cma_iw_handler;
2153
2154         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2155         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2156
2157         memset(&event, 0, sizeof event);
2158         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2159         event.param.conn.private_data = iw_event->private_data;
2160         event.param.conn.private_data_len = iw_event->private_data_len;
2161         event.param.conn.initiator_depth = iw_event->ird;
2162         event.param.conn.responder_resources = iw_event->ord;
2163
2164         /*
2165          * Protect against the user destroying conn_id from another thread
2166          * until we're done accessing it.
2167          */
2168         atomic_inc(&conn_id->refcount);
2169         ret = conn_id->id.event_handler(&conn_id->id, &event);
2170         if (ret) {
2171                 /* User wants to destroy the CM ID */
2172                 conn_id->cm_id.iw = NULL;
2173                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2174                 mutex_unlock(&conn_id->handler_mutex);
2175                 cma_deref_id(conn_id);
2176                 rdma_destroy_id(&conn_id->id);
2177                 goto out;
2178         }
2179
2180         mutex_unlock(&conn_id->handler_mutex);
2181         cma_deref_id(conn_id);
2182
2183 out:
2184         mutex_unlock(&listen_id->handler_mutex);
2185         return ret;
2186 }
2187
2188 static int cma_ib_listen(struct rdma_id_private *id_priv)
2189 {
2190         struct sockaddr *addr;
2191         struct ib_cm_id *id;
2192         __be64 svc_id;
2193
2194         addr = cma_src_addr(id_priv);
2195         svc_id = rdma_get_service_id(&id_priv->id, addr);
2196         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2197         if (IS_ERR(id))
2198                 return PTR_ERR(id);
2199         id_priv->cm_id.ib = id;
2200
2201         return 0;
2202 }
2203
2204 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2205 {
2206         int ret;
2207         struct iw_cm_id *id;
2208
2209         id = iw_create_cm_id(id_priv->id.device,
2210                              iw_conn_req_handler,
2211                              id_priv);
2212         if (IS_ERR(id))
2213                 return PTR_ERR(id);
2214
2215         id->tos = id_priv->tos;
2216         id_priv->cm_id.iw = id;
2217
2218         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2219                rdma_addr_size(cma_src_addr(id_priv)));
2220
2221         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2222
2223         if (ret) {
2224                 iw_destroy_cm_id(id_priv->cm_id.iw);
2225                 id_priv->cm_id.iw = NULL;
2226         }
2227
2228         return ret;
2229 }
2230
2231 static int cma_listen_handler(struct rdma_cm_id *id,
2232                               struct rdma_cm_event *event)
2233 {
2234         struct rdma_id_private *id_priv = id->context;
2235
2236         id->context = id_priv->id.context;
2237         id->event_handler = id_priv->id.event_handler;
2238         return id_priv->id.event_handler(id, event);
2239 }
2240
2241 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2242                               struct cma_device *cma_dev)
2243 {
2244         struct rdma_id_private *dev_id_priv;
2245         struct rdma_cm_id *id;
2246         struct net *net = id_priv->id.route.addr.dev_addr.net;
2247         int ret;
2248
2249         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2250                 return;
2251
2252         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2253                             id_priv->id.qp_type);
2254         if (IS_ERR(id))
2255                 return;
2256
2257         dev_id_priv = container_of(id, struct rdma_id_private, id);
2258
2259         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2260         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2261                rdma_addr_size(cma_src_addr(id_priv)));
2262
2263         _cma_attach_to_dev(dev_id_priv, cma_dev);
2264         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2265         atomic_inc(&id_priv->refcount);
2266         dev_id_priv->internal_id = 1;
2267         dev_id_priv->afonly = id_priv->afonly;
2268
2269         ret = rdma_listen(id, id_priv->backlog);
2270         if (ret)
2271                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2272                         ret, cma_dev->device->name);
2273 }
2274
2275 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2276 {
2277         struct cma_device *cma_dev;
2278
2279         mutex_lock(&lock);
2280         list_add_tail(&id_priv->list, &listen_any_list);
2281         list_for_each_entry(cma_dev, &dev_list, list)
2282                 cma_listen_on_dev(id_priv, cma_dev);
2283         mutex_unlock(&lock);
2284 }
2285
2286 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2287 {
2288         struct rdma_id_private *id_priv;
2289
2290         id_priv = container_of(id, struct rdma_id_private, id);
2291         id_priv->tos = (u8) tos;
2292         id_priv->tos_set = true;
2293 }
2294 EXPORT_SYMBOL(rdma_set_service_type);
2295
2296 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2297                               void *context)
2298 {
2299         struct cma_work *work = context;
2300         struct rdma_route *route;
2301
2302         route = &work->id->id.route;
2303
2304         if (!status) {
2305                 route->num_paths = 1;
2306                 *route->path_rec = *path_rec;
2307         } else {
2308                 work->old_state = RDMA_CM_ROUTE_QUERY;
2309                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2310                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2311                 work->event.status = status;
2312                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2313                                      status);
2314         }
2315
2316         queue_work(cma_wq, &work->work);
2317 }
2318
2319 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2320                               struct cma_work *work)
2321 {
2322         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2323         struct sa_path_rec path_rec;
2324         ib_sa_comp_mask comp_mask;
2325         struct sockaddr_in6 *sin6;
2326         struct sockaddr_ib *sib;
2327
2328         memset(&path_rec, 0, sizeof path_rec);
2329
2330         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2331                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2332         else
2333                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2334         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2335         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2336         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2337         path_rec.numb_path = 1;
2338         path_rec.reversible = 1;
2339         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2340                                                   cma_dst_addr(id_priv));
2341
2342         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2343                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2344                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2345
2346         switch (cma_family(id_priv)) {
2347         case AF_INET:
2348                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2349                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2350                 break;
2351         case AF_INET6:
2352                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2353                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2354                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2355                 break;
2356         case AF_IB:
2357                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2358                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2359                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2360                 break;
2361         }
2362
2363         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2364                                                id_priv->id.port_num, &path_rec,
2365                                                comp_mask, timeout_ms,
2366                                                GFP_KERNEL, cma_query_handler,
2367                                                work, &id_priv->query);
2368
2369         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2370 }
2371
2372 static void cma_work_handler(struct work_struct *_work)
2373 {
2374         struct cma_work *work = container_of(_work, struct cma_work, work);
2375         struct rdma_id_private *id_priv = work->id;
2376         int destroy = 0;
2377
2378         mutex_lock(&id_priv->handler_mutex);
2379         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2380                 goto out;
2381
2382         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2383                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2384                 destroy = 1;
2385         }
2386 out:
2387         mutex_unlock(&id_priv->handler_mutex);
2388         cma_deref_id(id_priv);
2389         if (destroy)
2390                 rdma_destroy_id(&id_priv->id);
2391         kfree(work);
2392 }
2393
2394 static void cma_ndev_work_handler(struct work_struct *_work)
2395 {
2396         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2397         struct rdma_id_private *id_priv = work->id;
2398         int destroy = 0;
2399
2400         mutex_lock(&id_priv->handler_mutex);
2401         if (id_priv->state == RDMA_CM_DESTROYING ||
2402             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2403                 goto out;
2404
2405         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2406                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2407                 destroy = 1;
2408         }
2409
2410 out:
2411         mutex_unlock(&id_priv->handler_mutex);
2412         cma_deref_id(id_priv);
2413         if (destroy)
2414                 rdma_destroy_id(&id_priv->id);
2415         kfree(work);
2416 }
2417
2418 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2419 {
2420         struct rdma_route *route = &id_priv->id.route;
2421         struct cma_work *work;
2422         int ret;
2423
2424         work = kzalloc(sizeof *work, GFP_KERNEL);
2425         if (!work)
2426                 return -ENOMEM;
2427
2428         work->id = id_priv;
2429         INIT_WORK(&work->work, cma_work_handler);
2430         work->old_state = RDMA_CM_ROUTE_QUERY;
2431         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2432         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2433
2434         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2435         if (!route->path_rec) {
2436                 ret = -ENOMEM;
2437                 goto err1;
2438         }
2439
2440         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2441         if (ret)
2442                 goto err2;
2443
2444         return 0;
2445 err2:
2446         kfree(route->path_rec);
2447         route->path_rec = NULL;
2448 err1:
2449         kfree(work);
2450         return ret;
2451 }
2452
2453 int rdma_set_ib_paths(struct rdma_cm_id *id,
2454                       struct sa_path_rec *path_rec, int num_paths)
2455 {
2456         struct rdma_id_private *id_priv;
2457         int ret;
2458
2459         id_priv = container_of(id, struct rdma_id_private, id);
2460         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2461                            RDMA_CM_ROUTE_RESOLVED))
2462                 return -EINVAL;
2463
2464         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2465                                      GFP_KERNEL);
2466         if (!id->route.path_rec) {
2467                 ret = -ENOMEM;
2468                 goto err;
2469         }
2470
2471         id->route.num_paths = num_paths;
2472         return 0;
2473 err:
2474         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2475         return ret;
2476 }
2477 EXPORT_SYMBOL(rdma_set_ib_paths);
2478
2479 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2480 {
2481         struct cma_work *work;
2482
2483         work = kzalloc(sizeof *work, GFP_KERNEL);
2484         if (!work)
2485                 return -ENOMEM;
2486
2487         work->id = id_priv;
2488         INIT_WORK(&work->work, cma_work_handler);
2489         work->old_state = RDMA_CM_ROUTE_QUERY;
2490         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2491         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2492         queue_work(cma_wq, &work->work);
2493         return 0;
2494 }
2495
2496 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2497 {
2498         int prio;
2499         struct net_device *dev;
2500
2501         prio = rt_tos2priority(tos);
2502         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2503         if (dev->num_tc)
2504                 return netdev_get_prio_tc_map(dev, prio);
2505
2506 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2507         if (is_vlan_dev(ndev))
2508                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2509                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2510 #endif
2511         return 0;
2512 }
2513
2514 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2515                                            unsigned long supported_gids,
2516                                            enum ib_gid_type default_gid)
2517 {
2518         if ((network_type == RDMA_NETWORK_IPV4 ||
2519              network_type == RDMA_NETWORK_IPV6) &&
2520             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2521                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2522
2523         return default_gid;
2524 }
2525
2526 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2527 {
2528         struct rdma_route *route = &id_priv->id.route;
2529         struct rdma_addr *addr = &route->addr;
2530         struct cma_work *work;
2531         int ret;
2532         struct net_device *ndev = NULL;
2533         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
2534         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2535                                         rdma_start_port(id_priv->cma_dev->device)];
2536         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2537
2538
2539         work = kzalloc(sizeof *work, GFP_KERNEL);
2540         if (!work)
2541                 return -ENOMEM;
2542
2543         work->id = id_priv;
2544         INIT_WORK(&work->work, cma_work_handler);
2545
2546         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2547         if (!route->path_rec) {
2548                 ret = -ENOMEM;
2549                 goto err1;
2550         }
2551
2552         route->num_paths = 1;
2553
2554         if (addr->dev_addr.bound_dev_if) {
2555                 unsigned long supported_gids;
2556
2557                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2558                 if (!ndev) {
2559                         ret = -ENODEV;
2560                         goto err2;
2561                 }
2562
2563                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2564                                                             id_priv->id.port_num);
2565                 gid_type = cma_route_gid_type(addr->dev_addr.network,
2566                                               supported_gids,
2567                                               id_priv->gid_type);
2568                 route->path_rec->rec_type =
2569                         sa_conv_gid_to_pathrec_type(gid_type);
2570                 sa_path_set_ndev(route->path_rec, &init_net);
2571                 sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2572         }
2573         if (!ndev) {
2574                 ret = -ENODEV;
2575                 goto err2;
2576         }
2577
2578         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2579
2580         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2581                     &route->path_rec->sgid);
2582         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2583                     &route->path_rec->dgid);
2584
2585         /* Use the hint from IP Stack to select GID Type */
2586         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2587                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2588         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2589
2590         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2591                 /* TODO: get the hoplimit from the inet/inet6 device */
2592                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2593         else
2594                 route->path_rec->hop_limit = 1;
2595         route->path_rec->reversible = 1;
2596         route->path_rec->pkey = cpu_to_be16(0xffff);
2597         route->path_rec->mtu_selector = IB_SA_EQ;
2598         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2599         route->path_rec->traffic_class = tos;
2600         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2601         route->path_rec->rate_selector = IB_SA_EQ;
2602         route->path_rec->rate = iboe_get_rate(ndev);
2603         dev_put(ndev);
2604         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2605         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2606         if (!route->path_rec->mtu) {
2607                 ret = -EINVAL;
2608                 goto err2;
2609         }
2610
2611         work->old_state = RDMA_CM_ROUTE_QUERY;
2612         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2613         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2614         work->event.status = 0;
2615
2616         queue_work(cma_wq, &work->work);
2617
2618         return 0;
2619
2620 err2:
2621         kfree(route->path_rec);
2622         route->path_rec = NULL;
2623 err1:
2624         kfree(work);
2625         return ret;
2626 }
2627
2628 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2629 {
2630         struct rdma_id_private *id_priv;
2631         int ret;
2632
2633         id_priv = container_of(id, struct rdma_id_private, id);
2634         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2635                 return -EINVAL;
2636
2637         atomic_inc(&id_priv->refcount);
2638         if (rdma_cap_ib_sa(id->device, id->port_num))
2639                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2640         else if (rdma_protocol_roce(id->device, id->port_num))
2641                 ret = cma_resolve_iboe_route(id_priv);
2642         else if (rdma_protocol_iwarp(id->device, id->port_num))
2643                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2644         else
2645                 ret = -ENOSYS;
2646
2647         if (ret)
2648                 goto err;
2649
2650         return 0;
2651 err:
2652         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2653         cma_deref_id(id_priv);
2654         return ret;
2655 }
2656 EXPORT_SYMBOL(rdma_resolve_route);
2657
2658 static void cma_set_loopback(struct sockaddr *addr)
2659 {
2660         switch (addr->sa_family) {
2661         case AF_INET:
2662                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2663                 break;
2664         case AF_INET6:
2665                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2666                               0, 0, 0, htonl(1));
2667                 break;
2668         default:
2669                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2670                             0, 0, 0, htonl(1));
2671                 break;
2672         }
2673 }
2674
2675 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2676 {
2677         struct cma_device *cma_dev, *cur_dev;
2678         union ib_gid gid;
2679         enum ib_port_state port_state;
2680         u16 pkey;
2681         int ret;
2682         u8 p;
2683
2684         cma_dev = NULL;
2685         mutex_lock(&lock);
2686         list_for_each_entry(cur_dev, &dev_list, list) {
2687                 if (cma_family(id_priv) == AF_IB &&
2688                     !rdma_cap_ib_cm(cur_dev->device, 1))
2689                         continue;
2690
2691                 if (!cma_dev)
2692                         cma_dev = cur_dev;
2693
2694                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2695                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2696                             port_state == IB_PORT_ACTIVE) {
2697                                 cma_dev = cur_dev;
2698                                 goto port_found;
2699                         }
2700                 }
2701         }
2702
2703         if (!cma_dev) {
2704                 ret = -ENODEV;
2705                 goto out;
2706         }
2707
2708         p = 1;
2709
2710 port_found:
2711         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2712         if (ret)
2713                 goto out;
2714
2715         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2716         if (ret)
2717                 goto out;
2718
2719         id_priv->id.route.addr.dev_addr.dev_type =
2720                 (rdma_protocol_ib(cma_dev->device, p)) ?
2721                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2722
2723         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2724         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2725         id_priv->id.port_num = p;
2726         cma_attach_to_dev(id_priv, cma_dev);
2727         cma_set_loopback(cma_src_addr(id_priv));
2728 out:
2729         mutex_unlock(&lock);
2730         return ret;
2731 }
2732
2733 static void addr_handler(int status, struct sockaddr *src_addr,
2734                          struct rdma_dev_addr *dev_addr, void *context)
2735 {
2736         struct rdma_id_private *id_priv = context;
2737         struct rdma_cm_event event;
2738
2739         memset(&event, 0, sizeof event);
2740         mutex_lock(&id_priv->handler_mutex);
2741         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2742                            RDMA_CM_ADDR_RESOLVED))
2743                 goto out;
2744
2745         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2746         if (!status && !id_priv->cma_dev) {
2747                 status = cma_acquire_dev(id_priv, NULL);
2748                 if (status)
2749                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2750                                              status);
2751         } else {
2752                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2753         }
2754
2755         if (status) {
2756                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2757                                    RDMA_CM_ADDR_BOUND))
2758                         goto out;
2759                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2760                 event.status = status;
2761         } else
2762                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2763
2764         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2765                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2766                 mutex_unlock(&id_priv->handler_mutex);
2767                 cma_deref_id(id_priv);
2768                 rdma_destroy_id(&id_priv->id);
2769                 return;
2770         }
2771 out:
2772         mutex_unlock(&id_priv->handler_mutex);
2773         cma_deref_id(id_priv);
2774 }
2775
2776 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2777 {
2778         struct cma_work *work;
2779         union ib_gid gid;
2780         int ret;
2781
2782         work = kzalloc(sizeof *work, GFP_KERNEL);
2783         if (!work)
2784                 return -ENOMEM;
2785
2786         if (!id_priv->cma_dev) {
2787                 ret = cma_bind_loopback(id_priv);
2788                 if (ret)
2789                         goto err;
2790         }
2791
2792         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2793         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2794
2795         work->id = id_priv;
2796         INIT_WORK(&work->work, cma_work_handler);
2797         work->old_state = RDMA_CM_ADDR_QUERY;
2798         work->new_state = RDMA_CM_ADDR_RESOLVED;
2799         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2800         queue_work(cma_wq, &work->work);
2801         return 0;
2802 err:
2803         kfree(work);
2804         return ret;
2805 }
2806
2807 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2808 {
2809         struct cma_work *work;
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_resolve_ib_dev(id_priv);
2818                 if (ret)
2819                         goto err;
2820         }
2821
2822         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2823                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2824
2825         work->id = id_priv;
2826         INIT_WORK(&work->work, cma_work_handler);
2827         work->old_state = RDMA_CM_ADDR_QUERY;
2828         work->new_state = RDMA_CM_ADDR_RESOLVED;
2829         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2830         queue_work(cma_wq, &work->work);
2831         return 0;
2832 err:
2833         kfree(work);
2834         return ret;
2835 }
2836
2837 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2838                          struct sockaddr *dst_addr)
2839 {
2840         if (!src_addr || !src_addr->sa_family) {
2841                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2842                 src_addr->sa_family = dst_addr->sa_family;
2843                 if (IS_ENABLED(CONFIG_IPV6) &&
2844                     dst_addr->sa_family == AF_INET6) {
2845                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2846                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2847                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2848                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2849                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2850                 } else if (dst_addr->sa_family == AF_IB) {
2851                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2852                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2853                 }
2854         }
2855         return rdma_bind_addr(id, src_addr);
2856 }
2857
2858 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2859                       struct sockaddr *dst_addr, int timeout_ms)
2860 {
2861         struct rdma_id_private *id_priv;
2862         int ret;
2863
2864         id_priv = container_of(id, struct rdma_id_private, id);
2865         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2866         if (id_priv->state == RDMA_CM_IDLE) {
2867                 ret = cma_bind_addr(id, src_addr, dst_addr);
2868                 if (ret) {
2869                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2870                         return ret;
2871                 }
2872         }
2873
2874         if (cma_family(id_priv) != dst_addr->sa_family) {
2875                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2876                 return -EINVAL;
2877         }
2878
2879         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2880                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2881                 return -EINVAL;
2882         }
2883
2884         atomic_inc(&id_priv->refcount);
2885         if (cma_any_addr(dst_addr)) {
2886                 ret = cma_resolve_loopback(id_priv);
2887         } else {
2888                 if (dst_addr->sa_family == AF_IB) {
2889                         ret = cma_resolve_ib_addr(id_priv);
2890                 } else {
2891                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2892                                               dst_addr, &id->route.addr.dev_addr,
2893                                               timeout_ms, addr_handler, id_priv);
2894                 }
2895         }
2896         if (ret)
2897                 goto err;
2898
2899         return 0;
2900 err:
2901         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2902         cma_deref_id(id_priv);
2903         return ret;
2904 }
2905 EXPORT_SYMBOL(rdma_resolve_addr);
2906
2907 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2908 {
2909         struct rdma_id_private *id_priv;
2910         unsigned long flags;
2911         int ret;
2912
2913         id_priv = container_of(id, struct rdma_id_private, id);
2914         spin_lock_irqsave(&id_priv->lock, flags);
2915         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2916                 id_priv->reuseaddr = reuse;
2917                 ret = 0;
2918         } else {
2919                 ret = -EINVAL;
2920         }
2921         spin_unlock_irqrestore(&id_priv->lock, flags);
2922         return ret;
2923 }
2924 EXPORT_SYMBOL(rdma_set_reuseaddr);
2925
2926 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2927 {
2928         struct rdma_id_private *id_priv;
2929         unsigned long flags;
2930         int ret;
2931
2932         id_priv = container_of(id, struct rdma_id_private, id);
2933         spin_lock_irqsave(&id_priv->lock, flags);
2934         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2935                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2936                 id_priv->afonly = afonly;
2937                 ret = 0;
2938         } else {
2939                 ret = -EINVAL;
2940         }
2941         spin_unlock_irqrestore(&id_priv->lock, flags);
2942         return ret;
2943 }
2944 EXPORT_SYMBOL(rdma_set_afonly);
2945
2946 static void cma_bind_port(struct rdma_bind_list *bind_list,
2947                           struct rdma_id_private *id_priv)
2948 {
2949         struct sockaddr *addr;
2950         struct sockaddr_ib *sib;
2951         u64 sid, mask;
2952         __be16 port;
2953
2954         addr = cma_src_addr(id_priv);
2955         port = htons(bind_list->port);
2956
2957         switch (addr->sa_family) {
2958         case AF_INET:
2959                 ((struct sockaddr_in *) addr)->sin_port = port;
2960                 break;
2961         case AF_INET6:
2962                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2963                 break;
2964         case AF_IB:
2965                 sib = (struct sockaddr_ib *) addr;
2966                 sid = be64_to_cpu(sib->sib_sid);
2967                 mask = be64_to_cpu(sib->sib_sid_mask);
2968                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2969                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2970                 break;
2971         }
2972         id_priv->bind_list = bind_list;
2973         hlist_add_head(&id_priv->node, &bind_list->owners);
2974 }
2975
2976 static int cma_alloc_port(enum rdma_port_space ps,
2977                           struct rdma_id_private *id_priv, unsigned short snum)
2978 {
2979         struct rdma_bind_list *bind_list;
2980         int ret;
2981
2982         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2983         if (!bind_list)
2984                 return -ENOMEM;
2985
2986         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2987                            snum);
2988         if (ret < 0)
2989                 goto err;
2990
2991         bind_list->ps = ps;
2992         bind_list->port = (unsigned short)ret;
2993         cma_bind_port(bind_list, id_priv);
2994         return 0;
2995 err:
2996         kfree(bind_list);
2997         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2998 }
2999
3000 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3001                               struct rdma_id_private *id_priv)
3002 {
3003         struct rdma_id_private *cur_id;
3004         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3005         struct sockaddr  *saddr = cma_src_addr(id_priv);
3006         __be16 dport = cma_port(daddr);
3007
3008         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3009                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3010                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3011                 __be16 cur_dport = cma_port(cur_daddr);
3012
3013                 if (id_priv == cur_id)
3014                         continue;
3015
3016                 /* different dest port -> unique */
3017                 if (!cma_any_port(cur_daddr) &&
3018                     (dport != cur_dport))
3019                         continue;
3020
3021                 /* different src address -> unique */
3022                 if (!cma_any_addr(saddr) &&
3023                     !cma_any_addr(cur_saddr) &&
3024                     cma_addr_cmp(saddr, cur_saddr))
3025                         continue;
3026
3027                 /* different dst address -> unique */
3028                 if (!cma_any_addr(cur_daddr) &&
3029                     cma_addr_cmp(daddr, cur_daddr))
3030                         continue;
3031
3032                 return -EADDRNOTAVAIL;
3033         }
3034         return 0;
3035 }
3036
3037 static int cma_alloc_any_port(enum rdma_port_space ps,
3038                               struct rdma_id_private *id_priv)
3039 {
3040         static unsigned int last_used_port;
3041         int low, high, remaining;
3042         unsigned int rover;
3043         struct net *net = id_priv->id.route.addr.dev_addr.net;
3044
3045         inet_get_local_port_range(net, &low, &high);
3046         remaining = (high - low) + 1;
3047         rover = prandom_u32() % remaining + low;
3048 retry:
3049         if (last_used_port != rover) {
3050                 struct rdma_bind_list *bind_list;
3051                 int ret;
3052
3053                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3054
3055                 if (!bind_list) {
3056                         ret = cma_alloc_port(ps, id_priv, rover);
3057                 } else {
3058                         ret = cma_port_is_unique(bind_list, id_priv);
3059                         if (!ret)
3060                                 cma_bind_port(bind_list, id_priv);
3061                 }
3062                 /*
3063                  * Remember previously used port number in order to avoid
3064                  * re-using same port immediately after it is closed.
3065                  */
3066                 if (!ret)
3067                         last_used_port = rover;
3068                 if (ret != -EADDRNOTAVAIL)
3069                         return ret;
3070         }
3071         if (--remaining) {
3072                 rover++;
3073                 if ((rover < low) || (rover > high))
3074                         rover = low;
3075                 goto retry;
3076         }
3077         return -EADDRNOTAVAIL;
3078 }
3079
3080 /*
3081  * Check that the requested port is available.  This is called when trying to
3082  * bind to a specific port, or when trying to listen on a bound port.  In
3083  * the latter case, the provided id_priv may already be on the bind_list, but
3084  * we still need to check that it's okay to start listening.
3085  */
3086 static int cma_check_port(struct rdma_bind_list *bind_list,
3087                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3088 {
3089         struct rdma_id_private *cur_id;
3090         struct sockaddr *addr, *cur_addr;
3091
3092         addr = cma_src_addr(id_priv);
3093         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3094                 if (id_priv == cur_id)
3095                         continue;
3096
3097                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3098                     cur_id->reuseaddr)
3099                         continue;
3100
3101                 cur_addr = cma_src_addr(cur_id);
3102                 if (id_priv->afonly && cur_id->afonly &&
3103                     (addr->sa_family != cur_addr->sa_family))
3104                         continue;
3105
3106                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3107                         return -EADDRNOTAVAIL;
3108
3109                 if (!cma_addr_cmp(addr, cur_addr))
3110                         return -EADDRINUSE;
3111         }
3112         return 0;
3113 }
3114
3115 static int cma_use_port(enum rdma_port_space ps,
3116                         struct rdma_id_private *id_priv)
3117 {
3118         struct rdma_bind_list *bind_list;
3119         unsigned short snum;
3120         int ret;
3121
3122         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3123         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3124                 return -EACCES;
3125
3126         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3127         if (!bind_list) {
3128                 ret = cma_alloc_port(ps, id_priv, snum);
3129         } else {
3130                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3131                 if (!ret)
3132                         cma_bind_port(bind_list, id_priv);
3133         }
3134         return ret;
3135 }
3136
3137 static int cma_bind_listen(struct rdma_id_private *id_priv)
3138 {
3139         struct rdma_bind_list *bind_list = id_priv->bind_list;
3140         int ret = 0;
3141
3142         mutex_lock(&lock);
3143         if (bind_list->owners.first->next)
3144                 ret = cma_check_port(bind_list, id_priv, 0);
3145         mutex_unlock(&lock);
3146         return ret;
3147 }
3148
3149 static enum rdma_port_space cma_select_inet_ps(
3150                 struct rdma_id_private *id_priv)
3151 {
3152         switch (id_priv->id.ps) {
3153         case RDMA_PS_TCP:
3154         case RDMA_PS_UDP:
3155         case RDMA_PS_IPOIB:
3156         case RDMA_PS_IB:
3157                 return id_priv->id.ps;
3158         default:
3159
3160                 return 0;
3161         }
3162 }
3163
3164 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3165 {
3166         enum rdma_port_space ps = 0;
3167         struct sockaddr_ib *sib;
3168         u64 sid_ps, mask, sid;
3169
3170         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3171         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3172         sid = be64_to_cpu(sib->sib_sid) & mask;
3173
3174         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3175                 sid_ps = RDMA_IB_IP_PS_IB;
3176                 ps = RDMA_PS_IB;
3177         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3178                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3179                 sid_ps = RDMA_IB_IP_PS_TCP;
3180                 ps = RDMA_PS_TCP;
3181         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3182                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3183                 sid_ps = RDMA_IB_IP_PS_UDP;
3184                 ps = RDMA_PS_UDP;
3185         }
3186
3187         if (ps) {
3188                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3189                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3190                                                 be64_to_cpu(sib->sib_sid_mask));
3191         }
3192         return ps;
3193 }
3194
3195 static int cma_get_port(struct rdma_id_private *id_priv)
3196 {
3197         enum rdma_port_space ps;
3198         int ret;
3199
3200         if (cma_family(id_priv) != AF_IB)
3201                 ps = cma_select_inet_ps(id_priv);
3202         else
3203                 ps = cma_select_ib_ps(id_priv);
3204         if (!ps)
3205                 return -EPROTONOSUPPORT;
3206
3207         mutex_lock(&lock);
3208         if (cma_any_port(cma_src_addr(id_priv)))
3209                 ret = cma_alloc_any_port(ps, id_priv);
3210         else
3211                 ret = cma_use_port(ps, id_priv);
3212         mutex_unlock(&lock);
3213
3214         return ret;
3215 }
3216
3217 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3218                                struct sockaddr *addr)
3219 {
3220 #if IS_ENABLED(CONFIG_IPV6)
3221         struct sockaddr_in6 *sin6;
3222
3223         if (addr->sa_family != AF_INET6)
3224                 return 0;
3225
3226         sin6 = (struct sockaddr_in6 *) addr;
3227
3228         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3229                 return 0;
3230
3231         if (!sin6->sin6_scope_id)
3232                         return -EINVAL;
3233
3234         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3235 #endif
3236         return 0;
3237 }
3238
3239 int rdma_listen(struct rdma_cm_id *id, int backlog)
3240 {
3241         struct rdma_id_private *id_priv;
3242         int ret;
3243
3244         id_priv = container_of(id, struct rdma_id_private, id);
3245         if (id_priv->state == RDMA_CM_IDLE) {
3246                 id->route.addr.src_addr.ss_family = AF_INET;
3247                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3248                 if (ret)
3249                         return ret;
3250         }
3251
3252         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3253                 return -EINVAL;
3254
3255         if (id_priv->reuseaddr) {
3256                 ret = cma_bind_listen(id_priv);
3257                 if (ret)
3258                         goto err;
3259         }
3260
3261         id_priv->backlog = backlog;
3262         if (id->device) {
3263                 if (rdma_cap_ib_cm(id->device, 1)) {
3264                         ret = cma_ib_listen(id_priv);
3265                         if (ret)
3266                                 goto err;
3267                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3268                         ret = cma_iw_listen(id_priv, backlog);
3269                         if (ret)
3270                                 goto err;
3271                 } else {
3272                         ret = -ENOSYS;
3273                         goto err;
3274                 }
3275         } else
3276                 cma_listen_on_all(id_priv);
3277
3278         return 0;
3279 err:
3280         id_priv->backlog = 0;
3281         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3282         return ret;
3283 }
3284 EXPORT_SYMBOL(rdma_listen);
3285
3286 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3287 {
3288         struct rdma_id_private *id_priv;
3289         int ret;
3290         struct sockaddr  *daddr;
3291
3292         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3293             addr->sa_family != AF_IB)
3294                 return -EAFNOSUPPORT;
3295
3296         id_priv = container_of(id, struct rdma_id_private, id);
3297         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3298                 return -EINVAL;
3299
3300         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3301         if (ret)
3302                 goto err1;
3303
3304         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3305         if (!cma_any_addr(addr)) {
3306                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3307                 if (ret)
3308                         goto err1;
3309
3310                 ret = cma_acquire_dev(id_priv, NULL);
3311                 if (ret)
3312                         goto err1;
3313         }
3314
3315         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3316                 if (addr->sa_family == AF_INET)
3317                         id_priv->afonly = 1;
3318 #if IS_ENABLED(CONFIG_IPV6)
3319                 else if (addr->sa_family == AF_INET6) {
3320                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3321
3322                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3323                 }
3324 #endif
3325         }
3326         ret = cma_get_port(id_priv);
3327         if (ret)
3328                 goto err2;
3329
3330         daddr = cma_dst_addr(id_priv);
3331         daddr->sa_family = addr->sa_family;
3332
3333         return 0;
3334 err2:
3335         if (id_priv->cma_dev)
3336                 cma_release_dev(id_priv);
3337 err1:
3338         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3339         return ret;
3340 }
3341 EXPORT_SYMBOL(rdma_bind_addr);
3342
3343 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3344 {
3345         struct cma_hdr *cma_hdr;
3346
3347         cma_hdr = hdr;
3348         cma_hdr->cma_version = CMA_VERSION;
3349         if (cma_family(id_priv) == AF_INET) {
3350                 struct sockaddr_in *src4, *dst4;
3351
3352                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3353                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3354
3355                 cma_set_ip_ver(cma_hdr, 4);
3356                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3357                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3358                 cma_hdr->port = src4->sin_port;
3359         } else if (cma_family(id_priv) == AF_INET6) {
3360                 struct sockaddr_in6 *src6, *dst6;
3361
3362                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3363                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3364
3365                 cma_set_ip_ver(cma_hdr, 6);
3366                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3367                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3368                 cma_hdr->port = src6->sin6_port;
3369         }
3370         return 0;
3371 }
3372
3373 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3374                                 struct ib_cm_event *ib_event)
3375 {
3376         struct rdma_id_private *id_priv = cm_id->context;
3377         struct rdma_cm_event event;
3378         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3379         int ret = 0;
3380
3381         mutex_lock(&id_priv->handler_mutex);
3382         if (id_priv->state != RDMA_CM_CONNECT)
3383                 goto out;
3384
3385         memset(&event, 0, sizeof event);
3386         switch (ib_event->event) {
3387         case IB_CM_SIDR_REQ_ERROR:
3388                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3389                 event.status = -ETIMEDOUT;
3390                 break;
3391         case IB_CM_SIDR_REP_RECEIVED:
3392                 event.param.ud.private_data = ib_event->private_data;
3393                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3394                 if (rep->status != IB_SIDR_SUCCESS) {
3395                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3396                         event.status = ib_event->param.sidr_rep_rcvd.status;
3397                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3398                                              event.status);
3399                         break;
3400                 }
3401                 ret = cma_set_qkey(id_priv, rep->qkey);
3402                 if (ret) {
3403                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3404                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3405                         event.status = ret;
3406                         break;
3407                 }
3408                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3409                                      id_priv->id.route.path_rec,
3410                                      &event.param.ud.ah_attr);
3411                 event.param.ud.qp_num = rep->qpn;
3412                 event.param.ud.qkey = rep->qkey;
3413                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3414                 event.status = 0;
3415                 break;
3416         default:
3417                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3418                        ib_event->event);
3419                 goto out;
3420         }
3421
3422         ret = id_priv->id.event_handler(&id_priv->id, &event);
3423         if (ret) {
3424                 /* Destroy the CM ID by returning a non-zero value. */
3425                 id_priv->cm_id.ib = NULL;
3426                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3427                 mutex_unlock(&id_priv->handler_mutex);
3428                 rdma_destroy_id(&id_priv->id);
3429                 return ret;
3430         }
3431 out:
3432         mutex_unlock(&id_priv->handler_mutex);
3433         return ret;
3434 }
3435
3436 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3437                               struct rdma_conn_param *conn_param)
3438 {
3439         struct ib_cm_sidr_req_param req;
3440         struct ib_cm_id *id;
3441         void *private_data;
3442         int offset, ret;
3443
3444         memset(&req, 0, sizeof req);
3445         offset = cma_user_data_offset(id_priv);
3446         req.private_data_len = offset + conn_param->private_data_len;
3447         if (req.private_data_len < conn_param->private_data_len)
3448                 return -EINVAL;
3449
3450         if (req.private_data_len) {
3451                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3452                 if (!private_data)
3453                         return -ENOMEM;
3454         } else {
3455                 private_data = NULL;
3456         }
3457
3458         if (conn_param->private_data && conn_param->private_data_len)
3459                 memcpy(private_data + offset, conn_param->private_data,
3460                        conn_param->private_data_len);
3461
3462         if (private_data) {
3463                 ret = cma_format_hdr(private_data, id_priv);
3464                 if (ret)
3465                         goto out;
3466                 req.private_data = private_data;
3467         }
3468
3469         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3470                              id_priv);
3471         if (IS_ERR(id)) {
3472                 ret = PTR_ERR(id);
3473                 goto out;
3474         }
3475         id_priv->cm_id.ib = id;
3476
3477         req.path = id_priv->id.route.path_rec;
3478         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3479         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3480         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3481
3482         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3483         if (ret) {
3484                 ib_destroy_cm_id(id_priv->cm_id.ib);
3485                 id_priv->cm_id.ib = NULL;
3486         }
3487 out:
3488         kfree(private_data);
3489         return ret;
3490 }
3491
3492 static int cma_connect_ib(struct rdma_id_private *id_priv,
3493                           struct rdma_conn_param *conn_param)
3494 {
3495         struct ib_cm_req_param req;
3496         struct rdma_route *route;
3497         void *private_data;
3498         struct ib_cm_id *id;
3499         int offset, ret;
3500
3501         memset(&req, 0, sizeof req);
3502         offset = cma_user_data_offset(id_priv);
3503         req.private_data_len = offset + conn_param->private_data_len;
3504         if (req.private_data_len < conn_param->private_data_len)
3505                 return -EINVAL;
3506
3507         if (req.private_data_len) {
3508                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3509                 if (!private_data)
3510                         return -ENOMEM;
3511         } else {
3512                 private_data = NULL;
3513         }
3514
3515         if (conn_param->private_data && conn_param->private_data_len)
3516                 memcpy(private_data + offset, conn_param->private_data,
3517                        conn_param->private_data_len);
3518
3519         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3520         if (IS_ERR(id)) {
3521                 ret = PTR_ERR(id);
3522                 goto out;
3523         }
3524         id_priv->cm_id.ib = id;
3525
3526         route = &id_priv->id.route;
3527         if (private_data) {
3528                 ret = cma_format_hdr(private_data, id_priv);
3529                 if (ret)
3530                         goto out;
3531                 req.private_data = private_data;
3532         }
3533
3534         req.primary_path = &route->path_rec[0];
3535         if (route->num_paths == 2)
3536                 req.alternate_path = &route->path_rec[1];
3537
3538         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3539         req.qp_num = id_priv->qp_num;
3540         req.qp_type = id_priv->id.qp_type;
3541         req.starting_psn = id_priv->seq_num;
3542         req.responder_resources = conn_param->responder_resources;
3543         req.initiator_depth = conn_param->initiator_depth;
3544         req.flow_control = conn_param->flow_control;
3545         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3546         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3547         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3548         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3549         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3550         req.srq = id_priv->srq ? 1 : 0;
3551
3552         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3553 out:
3554         if (ret && !IS_ERR(id)) {
3555                 ib_destroy_cm_id(id);
3556                 id_priv->cm_id.ib = NULL;
3557         }
3558
3559         kfree(private_data);
3560         return ret;
3561 }
3562
3563 static int cma_connect_iw(struct rdma_id_private *id_priv,
3564                           struct rdma_conn_param *conn_param)
3565 {
3566         struct iw_cm_id *cm_id;
3567         int ret;
3568         struct iw_cm_conn_param iw_param;
3569
3570         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3571         if (IS_ERR(cm_id))
3572                 return PTR_ERR(cm_id);
3573
3574         cm_id->tos = id_priv->tos;
3575         id_priv->cm_id.iw = cm_id;
3576
3577         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3578                rdma_addr_size(cma_src_addr(id_priv)));
3579         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3580                rdma_addr_size(cma_dst_addr(id_priv)));
3581
3582         ret = cma_modify_qp_rtr(id_priv, conn_param);
3583         if (ret)
3584                 goto out;
3585
3586         if (conn_param) {
3587                 iw_param.ord = conn_param->initiator_depth;
3588                 iw_param.ird = conn_param->responder_resources;
3589                 iw_param.private_data = conn_param->private_data;
3590                 iw_param.private_data_len = conn_param->private_data_len;
3591                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3592         } else {
3593                 memset(&iw_param, 0, sizeof iw_param);
3594                 iw_param.qpn = id_priv->qp_num;
3595         }
3596         ret = iw_cm_connect(cm_id, &iw_param);
3597 out:
3598         if (ret) {
3599                 iw_destroy_cm_id(cm_id);
3600                 id_priv->cm_id.iw = NULL;
3601         }
3602         return ret;
3603 }
3604
3605 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3606 {
3607         struct rdma_id_private *id_priv;
3608         int ret;
3609
3610         id_priv = container_of(id, struct rdma_id_private, id);
3611         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3612                 return -EINVAL;
3613
3614         if (!id->qp) {
3615                 id_priv->qp_num = conn_param->qp_num;
3616                 id_priv->srq = conn_param->srq;
3617         }
3618
3619         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3620                 if (id->qp_type == IB_QPT_UD)
3621                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3622                 else
3623                         ret = cma_connect_ib(id_priv, conn_param);
3624         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3625                 ret = cma_connect_iw(id_priv, conn_param);
3626         else
3627                 ret = -ENOSYS;
3628         if (ret)
3629                 goto err;
3630
3631         return 0;
3632 err:
3633         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3634         return ret;
3635 }
3636 EXPORT_SYMBOL(rdma_connect);
3637
3638 static int cma_accept_ib(struct rdma_id_private *id_priv,
3639                          struct rdma_conn_param *conn_param)
3640 {
3641         struct ib_cm_rep_param rep;
3642         int ret;
3643
3644         ret = cma_modify_qp_rtr(id_priv, conn_param);
3645         if (ret)
3646                 goto out;
3647
3648         ret = cma_modify_qp_rts(id_priv, conn_param);
3649         if (ret)
3650                 goto out;
3651
3652         memset(&rep, 0, sizeof rep);
3653         rep.qp_num = id_priv->qp_num;
3654         rep.starting_psn = id_priv->seq_num;
3655         rep.private_data = conn_param->private_data;
3656         rep.private_data_len = conn_param->private_data_len;
3657         rep.responder_resources = conn_param->responder_resources;
3658         rep.initiator_depth = conn_param->initiator_depth;
3659         rep.failover_accepted = 0;
3660         rep.flow_control = conn_param->flow_control;
3661         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3662         rep.srq = id_priv->srq ? 1 : 0;
3663
3664         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3665 out:
3666         return ret;
3667 }
3668
3669 static int cma_accept_iw(struct rdma_id_private *id_priv,
3670                   struct rdma_conn_param *conn_param)
3671 {
3672         struct iw_cm_conn_param iw_param;
3673         int ret;
3674
3675         if (!conn_param)
3676                 return -EINVAL;
3677
3678         ret = cma_modify_qp_rtr(id_priv, conn_param);
3679         if (ret)
3680                 return ret;
3681
3682         iw_param.ord = conn_param->initiator_depth;
3683         iw_param.ird = conn_param->responder_resources;
3684         iw_param.private_data = conn_param->private_data;
3685         iw_param.private_data_len = conn_param->private_data_len;
3686         if (id_priv->id.qp) {
3687                 iw_param.qpn = id_priv->qp_num;
3688         } else
3689                 iw_param.qpn = conn_param->qp_num;
3690
3691         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3692 }
3693
3694 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3695                              enum ib_cm_sidr_status status, u32 qkey,
3696                              const void *private_data, int private_data_len)
3697 {
3698         struct ib_cm_sidr_rep_param rep;
3699         int ret;
3700
3701         memset(&rep, 0, sizeof rep);
3702         rep.status = status;
3703         if (status == IB_SIDR_SUCCESS) {
3704                 ret = cma_set_qkey(id_priv, qkey);
3705                 if (ret)
3706                         return ret;
3707                 rep.qp_num = id_priv->qp_num;
3708                 rep.qkey = id_priv->qkey;
3709         }
3710         rep.private_data = private_data;
3711         rep.private_data_len = private_data_len;
3712
3713         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3714 }
3715
3716 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3717 {
3718         struct rdma_id_private *id_priv;
3719         int ret;
3720
3721         id_priv = container_of(id, struct rdma_id_private, id);
3722
3723         id_priv->owner = task_pid_nr(current);
3724
3725         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3726                 return -EINVAL;
3727
3728         if (!id->qp && conn_param) {
3729                 id_priv->qp_num = conn_param->qp_num;
3730                 id_priv->srq = conn_param->srq;
3731         }
3732
3733         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3734                 if (id->qp_type == IB_QPT_UD) {
3735                         if (conn_param)
3736                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3737                                                         conn_param->qkey,
3738                                                         conn_param->private_data,
3739                                                         conn_param->private_data_len);
3740                         else
3741                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3742                                                         0, NULL, 0);
3743                 } else {
3744                         if (conn_param)
3745                                 ret = cma_accept_ib(id_priv, conn_param);
3746                         else
3747                                 ret = cma_rep_recv(id_priv);
3748                 }
3749         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3750                 ret = cma_accept_iw(id_priv, conn_param);
3751         else
3752                 ret = -ENOSYS;
3753
3754         if (ret)
3755                 goto reject;
3756
3757         return 0;
3758 reject:
3759         cma_modify_qp_err(id_priv);
3760         rdma_reject(id, NULL, 0);
3761         return ret;
3762 }
3763 EXPORT_SYMBOL(rdma_accept);
3764
3765 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3766 {
3767         struct rdma_id_private *id_priv;
3768         int ret;
3769
3770         id_priv = container_of(id, struct rdma_id_private, id);
3771         if (!id_priv->cm_id.ib)
3772                 return -EINVAL;
3773
3774         switch (id->device->node_type) {
3775         case RDMA_NODE_IB_CA:
3776                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3777                 break;
3778         default:
3779                 ret = 0;
3780                 break;
3781         }
3782         return ret;
3783 }
3784 EXPORT_SYMBOL(rdma_notify);
3785
3786 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3787                 u8 private_data_len)
3788 {
3789         struct rdma_id_private *id_priv;
3790         int ret;
3791
3792         id_priv = container_of(id, struct rdma_id_private, id);
3793         if (!id_priv->cm_id.ib)
3794                 return -EINVAL;
3795
3796         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3797                 if (id->qp_type == IB_QPT_UD)
3798                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3799                                                 private_data, private_data_len);
3800                 else
3801                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3802                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3803                                              0, private_data, private_data_len);
3804         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3805                 ret = iw_cm_reject(id_priv->cm_id.iw,
3806                                    private_data, private_data_len);
3807         } else
3808                 ret = -ENOSYS;
3809
3810         return ret;
3811 }
3812 EXPORT_SYMBOL(rdma_reject);
3813
3814 int rdma_disconnect(struct rdma_cm_id *id)
3815 {
3816         struct rdma_id_private *id_priv;
3817         int ret;
3818
3819         id_priv = container_of(id, struct rdma_id_private, id);
3820         if (!id_priv->cm_id.ib)
3821                 return -EINVAL;
3822
3823         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3824                 ret = cma_modify_qp_err(id_priv);
3825                 if (ret)
3826                         goto out;
3827                 /* Initiate or respond to a disconnect. */
3828                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3829                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3830         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3831                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3832         } else
3833                 ret = -EINVAL;
3834
3835 out:
3836         return ret;
3837 }
3838 EXPORT_SYMBOL(rdma_disconnect);
3839
3840 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3841 {
3842         struct rdma_id_private *id_priv;
3843         struct cma_multicast *mc = multicast->context;
3844         struct rdma_cm_event event;
3845         int ret = 0;
3846
3847         id_priv = mc->id_priv;
3848         mutex_lock(&id_priv->handler_mutex);
3849         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3850             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3851                 goto out;
3852
3853         if (!status)
3854                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3855         else
3856                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3857                                      status);
3858         mutex_lock(&id_priv->qp_mutex);
3859         if (!status && id_priv->id.qp) {
3860                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3861                                          be16_to_cpu(multicast->rec.mlid));
3862                 if (status)
3863                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3864                                              status);
3865         }
3866         mutex_unlock(&id_priv->qp_mutex);
3867
3868         memset(&event, 0, sizeof event);
3869         event.status = status;
3870         event.param.ud.private_data = mc->context;
3871         if (!status) {
3872                 struct rdma_dev_addr *dev_addr =
3873                         &id_priv->id.route.addr.dev_addr;
3874                 struct net_device *ndev =
3875                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3876                 enum ib_gid_type gid_type =
3877                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3878                         rdma_start_port(id_priv->cma_dev->device)];
3879
3880                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3881                 ib_init_ah_from_mcmember(id_priv->id.device,
3882                                          id_priv->id.port_num, &multicast->rec,
3883                                          ndev, gid_type,
3884                                          &event.param.ud.ah_attr);
3885                 event.param.ud.qp_num = 0xFFFFFF;
3886                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3887                 if (ndev)
3888                         dev_put(ndev);
3889         } else
3890                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3891
3892         ret = id_priv->id.event_handler(&id_priv->id, &event);
3893         if (ret) {
3894                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3895                 mutex_unlock(&id_priv->handler_mutex);
3896                 rdma_destroy_id(&id_priv->id);
3897                 return 0;
3898         }
3899
3900 out:
3901         mutex_unlock(&id_priv->handler_mutex);
3902         return 0;
3903 }
3904
3905 static void cma_set_mgid(struct rdma_id_private *id_priv,
3906                          struct sockaddr *addr, union ib_gid *mgid)
3907 {
3908         unsigned char mc_map[MAX_ADDR_LEN];
3909         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3910         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3911         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3912
3913         if (cma_any_addr(addr)) {
3914                 memset(mgid, 0, sizeof *mgid);
3915         } else if ((addr->sa_family == AF_INET6) &&
3916                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3917                                                                  0xFF10A01B)) {
3918                 /* IPv6 address is an SA assigned MGID. */
3919                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3920         } else if (addr->sa_family == AF_IB) {
3921                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3922         } else if ((addr->sa_family == AF_INET6)) {
3923                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3924                 if (id_priv->id.ps == RDMA_PS_UDP)
3925                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3926                 *mgid = *(union ib_gid *) (mc_map + 4);
3927         } else {
3928                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3929                 if (id_priv->id.ps == RDMA_PS_UDP)
3930                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3931                 *mgid = *(union ib_gid *) (mc_map + 4);
3932         }
3933 }
3934
3935 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3936                                  struct cma_multicast *mc)
3937 {
3938         struct ib_sa_mcmember_rec rec;
3939         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3940         ib_sa_comp_mask comp_mask;
3941         int ret;
3942
3943         ib_addr_get_mgid(dev_addr, &rec.mgid);
3944         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3945                                      &rec.mgid, &rec);
3946         if (ret)
3947                 return ret;
3948
3949         ret = cma_set_qkey(id_priv, 0);
3950         if (ret)
3951                 return ret;
3952
3953         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3954         rec.qkey = cpu_to_be32(id_priv->qkey);
3955         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3956         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3957         rec.join_state = mc->join_state;
3958
3959         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3960             (!ib_sa_sendonly_fullmem_support(&sa_client,
3961                                              id_priv->id.device,
3962                                              id_priv->id.port_num))) {
3963                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3964                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
3965                         id_priv->id.device->name, id_priv->id.port_num);
3966                 return -EOPNOTSUPP;
3967         }
3968
3969         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3970                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3971                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3972                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3973                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3974
3975         if (id_priv->id.ps == RDMA_PS_IPOIB)
3976                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3977                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3978                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3979                              IB_SA_MCMEMBER_REC_MTU |
3980                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3981
3982         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3983                                                 id_priv->id.port_num, &rec,
3984                                                 comp_mask, GFP_KERNEL,
3985                                                 cma_ib_mc_handler, mc);
3986         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3987 }
3988
3989 static void iboe_mcast_work_handler(struct work_struct *work)
3990 {
3991         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3992         struct cma_multicast *mc = mw->mc;
3993         struct ib_sa_multicast *m = mc->multicast.ib;
3994
3995         mc->multicast.ib->context = mc;
3996         cma_ib_mc_handler(0, m);
3997         kref_put(&mc->mcref, release_mc);
3998         kfree(mw);
3999 }
4000
4001 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4002 {
4003         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4004         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4005
4006         if (cma_any_addr(addr)) {
4007                 memset(mgid, 0, sizeof *mgid);
4008         } else if (addr->sa_family == AF_INET6) {
4009                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4010         } else {
4011                 mgid->raw[0] = 0xff;
4012                 mgid->raw[1] = 0x0e;
4013                 mgid->raw[2] = 0;
4014                 mgid->raw[3] = 0;
4015                 mgid->raw[4] = 0;
4016                 mgid->raw[5] = 0;
4017                 mgid->raw[6] = 0;
4018                 mgid->raw[7] = 0;
4019                 mgid->raw[8] = 0;
4020                 mgid->raw[9] = 0;
4021                 mgid->raw[10] = 0xff;
4022                 mgid->raw[11] = 0xff;
4023                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4024         }
4025 }
4026
4027 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4028                                    struct cma_multicast *mc)
4029 {
4030         struct iboe_mcast_work *work;
4031         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4032         int err = 0;
4033         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4034         struct net_device *ndev = NULL;
4035         enum ib_gid_type gid_type;
4036         bool send_only;
4037
4038         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4039
4040         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4041                 return -EINVAL;
4042
4043         work = kzalloc(sizeof *work, GFP_KERNEL);
4044         if (!work)
4045                 return -ENOMEM;
4046
4047         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4048         if (!mc->multicast.ib) {
4049                 err = -ENOMEM;
4050                 goto out1;
4051         }
4052
4053         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4054
4055         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4056         if (id_priv->id.ps == RDMA_PS_UDP)
4057                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4058
4059         if (dev_addr->bound_dev_if)
4060                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4061         if (!ndev) {
4062                 err = -ENODEV;
4063                 goto out2;
4064         }
4065         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4066         mc->multicast.ib->rec.hop_limit = 1;
4067         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4068
4069         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4070                    rdma_start_port(id_priv->cma_dev->device)];
4071         if (addr->sa_family == AF_INET) {
4072                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4073                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4074                         if (!send_only) {
4075                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4076                                                     true);
4077                                 if (!err)
4078                                         mc->igmp_joined = true;
4079                         }
4080                 }
4081         } else {
4082                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4083                         err = -ENOTSUPP;
4084         }
4085         dev_put(ndev);
4086         if (err || !mc->multicast.ib->rec.mtu) {
4087                 if (!err)
4088                         err = -EINVAL;
4089                 goto out2;
4090         }
4091         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4092                     &mc->multicast.ib->rec.port_gid);
4093         work->id = id_priv;
4094         work->mc = mc;
4095         INIT_WORK(&work->work, iboe_mcast_work_handler);
4096         kref_get(&mc->mcref);
4097         queue_work(cma_wq, &work->work);
4098
4099         return 0;
4100
4101 out2:
4102         kfree(mc->multicast.ib);
4103 out1:
4104         kfree(work);
4105         return err;
4106 }
4107
4108 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4109                         u8 join_state, void *context)
4110 {
4111         struct rdma_id_private *id_priv;
4112         struct cma_multicast *mc;
4113         int ret;
4114
4115         id_priv = container_of(id, struct rdma_id_private, id);
4116         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4117             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4118                 return -EINVAL;
4119
4120         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4121         if (!mc)
4122                 return -ENOMEM;
4123
4124         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4125         mc->context = context;
4126         mc->id_priv = id_priv;
4127         mc->igmp_joined = false;
4128         mc->join_state = join_state;
4129         spin_lock(&id_priv->lock);
4130         list_add(&mc->list, &id_priv->mc_list);
4131         spin_unlock(&id_priv->lock);
4132
4133         if (rdma_protocol_roce(id->device, id->port_num)) {
4134                 kref_init(&mc->mcref);
4135                 ret = cma_iboe_join_multicast(id_priv, mc);
4136         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4137                 ret = cma_join_ib_multicast(id_priv, mc);
4138         else
4139                 ret = -ENOSYS;
4140
4141         if (ret) {
4142                 spin_lock_irq(&id_priv->lock);
4143                 list_del(&mc->list);
4144                 spin_unlock_irq(&id_priv->lock);
4145                 kfree(mc);
4146         }
4147         return ret;
4148 }
4149 EXPORT_SYMBOL(rdma_join_multicast);
4150
4151 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4152 {
4153         struct rdma_id_private *id_priv;
4154         struct cma_multicast *mc;
4155
4156         id_priv = container_of(id, struct rdma_id_private, id);
4157         spin_lock_irq(&id_priv->lock);
4158         list_for_each_entry(mc, &id_priv->mc_list, list) {
4159                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4160                         list_del(&mc->list);
4161                         spin_unlock_irq(&id_priv->lock);
4162
4163                         if (id->qp)
4164                                 ib_detach_mcast(id->qp,
4165                                                 &mc->multicast.ib->rec.mgid,
4166                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4167
4168                         BUG_ON(id_priv->cma_dev->device != id->device);
4169
4170                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4171                                 ib_sa_free_multicast(mc->multicast.ib);
4172                                 kfree(mc);
4173                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4174                                 if (mc->igmp_joined) {
4175                                         struct rdma_dev_addr *dev_addr =
4176                                                 &id->route.addr.dev_addr;
4177                                         struct net_device *ndev = NULL;
4178
4179                                         if (dev_addr->bound_dev_if)
4180                                                 ndev = dev_get_by_index(&init_net,
4181                                                                         dev_addr->bound_dev_if);
4182                                         if (ndev) {
4183                                                 cma_igmp_send(ndev,
4184                                                               &mc->multicast.ib->rec.mgid,
4185                                                               false);
4186                                                 dev_put(ndev);
4187                                         }
4188                                         mc->igmp_joined = false;
4189                                 }
4190                                 kref_put(&mc->mcref, release_mc);
4191                         }
4192                         return;
4193                 }
4194         }
4195         spin_unlock_irq(&id_priv->lock);
4196 }
4197 EXPORT_SYMBOL(rdma_leave_multicast);
4198
4199 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4200 {
4201         struct rdma_dev_addr *dev_addr;
4202         struct cma_ndev_work *work;
4203
4204         dev_addr = &id_priv->id.route.addr.dev_addr;
4205
4206         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4207             (net_eq(dev_net(ndev), dev_addr->net)) &&
4208             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4209                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4210                         ndev->name, &id_priv->id);
4211                 work = kzalloc(sizeof *work, GFP_KERNEL);
4212                 if (!work)
4213                         return -ENOMEM;
4214
4215                 INIT_WORK(&work->work, cma_ndev_work_handler);
4216                 work->id = id_priv;
4217                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4218                 atomic_inc(&id_priv->refcount);
4219                 queue_work(cma_wq, &work->work);
4220         }
4221
4222         return 0;
4223 }
4224
4225 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4226                                void *ptr)
4227 {
4228         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4229         struct cma_device *cma_dev;
4230         struct rdma_id_private *id_priv;
4231         int ret = NOTIFY_DONE;
4232
4233         if (event != NETDEV_BONDING_FAILOVER)
4234                 return NOTIFY_DONE;
4235
4236         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4237                 return NOTIFY_DONE;
4238
4239         mutex_lock(&lock);
4240         list_for_each_entry(cma_dev, &dev_list, list)
4241                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4242                         ret = cma_netdev_change(ndev, id_priv);
4243                         if (ret)
4244                                 goto out;
4245                 }
4246
4247 out:
4248         mutex_unlock(&lock);
4249         return ret;
4250 }
4251
4252 static struct notifier_block cma_nb = {
4253         .notifier_call = cma_netdev_callback
4254 };
4255
4256 static void cma_add_one(struct ib_device *device)
4257 {
4258         struct cma_device *cma_dev;
4259         struct rdma_id_private *id_priv;
4260         unsigned int i;
4261         unsigned long supported_gids = 0;
4262
4263         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4264         if (!cma_dev)
4265                 return;
4266
4267         cma_dev->device = device;
4268         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4269                                             sizeof(*cma_dev->default_gid_type),
4270                                             GFP_KERNEL);
4271         if (!cma_dev->default_gid_type)
4272                 goto free_cma_dev;
4273
4274         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4275                                             sizeof(*cma_dev->default_roce_tos),
4276                                             GFP_KERNEL);
4277         if (!cma_dev->default_roce_tos)
4278                 goto free_gid_type;
4279
4280         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4281                 supported_gids = roce_gid_type_mask_support(device, i);
4282                 WARN_ON(!supported_gids);
4283                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4284                         find_first_bit(&supported_gids, BITS_PER_LONG);
4285                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4286         }
4287
4288         init_completion(&cma_dev->comp);
4289         atomic_set(&cma_dev->refcount, 1);
4290         INIT_LIST_HEAD(&cma_dev->id_list);
4291         ib_set_client_data(device, &cma_client, cma_dev);
4292
4293         mutex_lock(&lock);
4294         list_add_tail(&cma_dev->list, &dev_list);
4295         list_for_each_entry(id_priv, &listen_any_list, list)
4296                 cma_listen_on_dev(id_priv, cma_dev);
4297         mutex_unlock(&lock);
4298
4299         return;
4300
4301 free_gid_type:
4302         kfree(cma_dev->default_gid_type);
4303
4304 free_cma_dev:
4305         kfree(cma_dev);
4306
4307         return;
4308 }
4309
4310 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4311 {
4312         struct rdma_cm_event event;
4313         enum rdma_cm_state state;
4314         int ret = 0;
4315
4316         /* Record that we want to remove the device */
4317         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4318         if (state == RDMA_CM_DESTROYING)
4319                 return 0;
4320
4321         cma_cancel_operation(id_priv, state);
4322         mutex_lock(&id_priv->handler_mutex);
4323
4324         /* Check for destruction from another callback. */
4325         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4326                 goto out;
4327
4328         memset(&event, 0, sizeof event);
4329         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4330         ret = id_priv->id.event_handler(&id_priv->id, &event);
4331 out:
4332         mutex_unlock(&id_priv->handler_mutex);
4333         return ret;
4334 }
4335
4336 static void cma_process_remove(struct cma_device *cma_dev)
4337 {
4338         struct rdma_id_private *id_priv;
4339         int ret;
4340
4341         mutex_lock(&lock);
4342         while (!list_empty(&cma_dev->id_list)) {
4343                 id_priv = list_entry(cma_dev->id_list.next,
4344                                      struct rdma_id_private, list);
4345
4346                 list_del(&id_priv->listen_list);
4347                 list_del_init(&id_priv->list);
4348                 atomic_inc(&id_priv->refcount);
4349                 mutex_unlock(&lock);
4350
4351                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4352                 cma_deref_id(id_priv);
4353                 if (ret)
4354                         rdma_destroy_id(&id_priv->id);
4355
4356                 mutex_lock(&lock);
4357         }
4358         mutex_unlock(&lock);
4359
4360         cma_deref_dev(cma_dev);
4361         wait_for_completion(&cma_dev->comp);
4362 }
4363
4364 static void cma_remove_one(struct ib_device *device, void *client_data)
4365 {
4366         struct cma_device *cma_dev = client_data;
4367
4368         if (!cma_dev)
4369                 return;
4370
4371         mutex_lock(&lock);
4372         list_del(&cma_dev->list);
4373         mutex_unlock(&lock);
4374
4375         cma_process_remove(cma_dev);
4376         kfree(cma_dev->default_roce_tos);
4377         kfree(cma_dev->default_gid_type);
4378         kfree(cma_dev);
4379 }
4380
4381 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4382 {
4383         struct nlmsghdr *nlh;
4384         struct rdma_cm_id_stats *id_stats;
4385         struct rdma_id_private *id_priv;
4386         struct rdma_cm_id *id = NULL;
4387         struct cma_device *cma_dev;
4388         int i_dev = 0, i_id = 0;
4389
4390         /*
4391          * We export all of the IDs as a sequence of messages.  Each
4392          * ID gets its own netlink message.
4393          */
4394         mutex_lock(&lock);
4395
4396         list_for_each_entry(cma_dev, &dev_list, list) {
4397                 if (i_dev < cb->args[0]) {
4398                         i_dev++;
4399                         continue;
4400                 }
4401
4402                 i_id = 0;
4403                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4404                         if (i_id < cb->args[1]) {
4405                                 i_id++;
4406                                 continue;
4407                         }
4408
4409                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4410                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4411                                                 RDMA_NL_RDMA_CM_ID_STATS,
4412                                                 NLM_F_MULTI);
4413                         if (!id_stats)
4414                                 goto out;
4415
4416                         memset(id_stats, 0, sizeof *id_stats);
4417                         id = &id_priv->id;
4418                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4419                         id_stats->port_num = id->port_num;
4420                         id_stats->bound_dev_if =
4421                                 id->route.addr.dev_addr.bound_dev_if;
4422
4423                         if (ibnl_put_attr(skb, nlh,
4424                                           rdma_addr_size(cma_src_addr(id_priv)),
4425                                           cma_src_addr(id_priv),
4426                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4427                                 goto out;
4428                         if (ibnl_put_attr(skb, nlh,
4429                                           rdma_addr_size(cma_src_addr(id_priv)),
4430                                           cma_dst_addr(id_priv),
4431                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4432                                 goto out;
4433
4434                         id_stats->pid           = id_priv->owner;
4435                         id_stats->port_space    = id->ps;
4436                         id_stats->cm_state      = id_priv->state;
4437                         id_stats->qp_num        = id_priv->qp_num;
4438                         id_stats->qp_type       = id->qp_type;
4439
4440                         i_id++;
4441                 }
4442
4443                 cb->args[1] = 0;
4444                 i_dev++;
4445         }
4446
4447 out:
4448         mutex_unlock(&lock);
4449         cb->args[0] = i_dev;
4450         cb->args[1] = i_id;
4451
4452         return skb->len;
4453 }
4454
4455 static const struct ibnl_client_cbs cma_cb_table[] = {
4456         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4457                                        .module = THIS_MODULE },
4458 };
4459
4460 static int cma_init_net(struct net *net)
4461 {
4462         struct cma_pernet *pernet = cma_pernet(net);
4463
4464         idr_init(&pernet->tcp_ps);
4465         idr_init(&pernet->udp_ps);
4466         idr_init(&pernet->ipoib_ps);
4467         idr_init(&pernet->ib_ps);
4468
4469         return 0;
4470 }
4471
4472 static void cma_exit_net(struct net *net)
4473 {
4474         struct cma_pernet *pernet = cma_pernet(net);
4475
4476         idr_destroy(&pernet->tcp_ps);
4477         idr_destroy(&pernet->udp_ps);
4478         idr_destroy(&pernet->ipoib_ps);
4479         idr_destroy(&pernet->ib_ps);
4480 }
4481
4482 static struct pernet_operations cma_pernet_operations = {
4483         .init = cma_init_net,
4484         .exit = cma_exit_net,
4485         .id = &cma_pernet_id,
4486         .size = sizeof(struct cma_pernet),
4487 };
4488
4489 static int __init cma_init(void)
4490 {
4491         int ret;
4492
4493         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4494         if (!cma_wq)
4495                 return -ENOMEM;
4496
4497         ret = register_pernet_subsys(&cma_pernet_operations);
4498         if (ret)
4499                 goto err_wq;
4500
4501         ib_sa_register_client(&sa_client);
4502         rdma_addr_register_client(&addr_client);
4503         register_netdevice_notifier(&cma_nb);
4504
4505         ret = ib_register_client(&cma_client);
4506         if (ret)
4507                 goto err;
4508
4509         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4510                             cma_cb_table))
4511                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4512         cma_configfs_init();
4513
4514         return 0;
4515
4516 err:
4517         unregister_netdevice_notifier(&cma_nb);
4518         rdma_addr_unregister_client(&addr_client);
4519         ib_sa_unregister_client(&sa_client);
4520 err_wq:
4521         destroy_workqueue(cma_wq);
4522         return ret;
4523 }
4524
4525 static void __exit cma_cleanup(void)
4526 {
4527         cma_configfs_exit();
4528         ibnl_remove_client(RDMA_NL_RDMA_CM);
4529         ib_unregister_client(&cma_client);
4530         unregister_netdevice_notifier(&cma_nb);
4531         rdma_addr_unregister_client(&addr_client);
4532         ib_sa_unregister_client(&sa_client);
4533         unregister_pernet_subsys(&cma_pernet_operations);
4534         destroy_workqueue(cma_wq);
4535 }
4536
4537 module_init(cma_init);
4538 module_exit(cma_cleanup);