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