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