drivers/infiniband/core/cma.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005-2006 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/completion.h>
  37 #include <linux/in.h>
  38 #include <linux/in6.h>
  39 #include <linux/mutex.h>
  40 #include <linux/random.h>
  41 #include <linux/igmp.h>
  42 #include <linux/idr.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/slab.h>
  45 #include <linux/module.h>
  46 #include <net/route.h>
  47
  48 #include <net/net_namespace.h>
  49 #include <net/netns/generic.h>
  50 #include <net/tcp.h>
  51 #include <net/ipv6.h>
  52 #include <net/ip_fib.h>
  53 #include <net/ip6_route.h>
  54
  55 #include <rdma/rdma_cm.h>
  56 #include <rdma/rdma_cm_ib.h>
  57 #include <rdma/rdma_netlink.h>
  58 #include <rdma/ib.h>
  59 #include <rdma/ib_cache.h>
  60 #include <rdma/ib_cm.h>
  61 #include <rdma/ib_sa.h>
  62 #include <rdma/iw_cm.h>
  63
  64 #include "core_priv.h"
  65
  66 MODULE_AUTHOR("Sean Hefty");
  67 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  68 MODULE_LICENSE("Dual BSD/GPL");
  69
  70 #define CMA_CM_RESPONSE_TIMEOUT 20
  71 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  72 #define CMA_MAX_CM_RETRIES 15
  73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  74 #define CMA_IBOE_PACKET_LIFETIME 18
  75
  76 static const char * const cma_events[] = {
  77         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  78         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  79         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  80         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  81         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  82         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  83         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  84         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  85         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  86         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  87         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  88         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  89         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  90         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  91         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  92         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  93 };
  94
  95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  96 {
  97         size_t index = event;
  98
  99         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 100                         cma_events[index] : "unrecognized event";
 101 }
 102 EXPORT_SYMBOL(rdma_event_msg);
 103
 104 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 105                                                 int reason)
 106 {
 107         if (rdma_ib_or_roce(id->device, id->port_num))
 108                 return ibcm_reject_msg(reason);
 109
 110         if (rdma_protocol_iwarp(id->device, id->port_num))
 111                 return iwcm_reject_msg(reason);
 112
 113         WARN_ON_ONCE(1);
 114         return "unrecognized transport";
 115 }
 116 EXPORT_SYMBOL(rdma_reject_msg);
 117
 118 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 119 {
 120         if (rdma_ib_or_roce(id->device, id->port_num))
 121                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 122
 123         if (rdma_protocol_iwarp(id->device, id->port_num))
 124                 return reason == -ECONNREFUSED;
 125
 126         WARN_ON_ONCE(1);
 127         return false;
 128 }
 129 EXPORT_SYMBOL(rdma_is_consumer_reject);
 130
 131 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 132                                       struct rdma_cm_event *ev, u8 *data_len)
 133 {
 134         const void *p;
 135
 136         if (rdma_is_consumer_reject(id, ev->status)) {
 137                 *data_len = ev->param.conn.private_data_len;
 138                 p = ev->param.conn.private_data;
 139         } else {
 140                 *data_len = 0;
 141                 p = NULL;
 142         }
 143         return p;
 144 }
 145 EXPORT_SYMBOL(rdma_consumer_reject_data);
 146
 147 static void cma_add_one(struct ib_device *device);
 148 static void cma_remove_one(struct ib_device *device, void *client_data);
 149
 150 static struct ib_client cma_client = {
 151         .name   = "cma",
 152         .add    = cma_add_one,
 153         .remove = cma_remove_one
 154 };
 155
 156 static struct ib_sa_client sa_client;
 157 static struct rdma_addr_client addr_client;
 158 static LIST_HEAD(dev_list);
 159 static LIST_HEAD(listen_any_list);
 160 static DEFINE_MUTEX(lock);
 161 static struct workqueue_struct *cma_wq;
 162 static unsigned int cma_pernet_id;
 163
 164 struct cma_pernet {
 165         struct idr tcp_ps;
 166         struct idr udp_ps;
 167         struct idr ipoib_ps;
 168         struct idr ib_ps;
 169 };
 170
 171 static struct cma_pernet *cma_pernet(struct net *net)
 172 {
 173         return net_generic(net, cma_pernet_id);
 174 }
 175
 176 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 177 {
 178         struct cma_pernet *pernet = cma_pernet(net);
 179
 180         switch (ps) {
 181         case RDMA_PS_TCP:
 182                 return &pernet->tcp_ps;
 183         case RDMA_PS_UDP:
 184                 return &pernet->udp_ps;
 185         case RDMA_PS_IPOIB:
 186                 return &pernet->ipoib_ps;
 187         case RDMA_PS_IB:
 188                 return &pernet->ib_ps;
 189         default:
 190                 return NULL;
 191         }
 192 }
 193
 194 struct cma_device {
 195         struct list_head        list;
 196         struct ib_device        *device;
 197         struct completion       comp;
 198         atomic_t                refcount;
 199         struct list_head        id_list;
 200         enum ib_gid_type        *default_gid_type;
 201         u8                      *default_roce_tos;
 202 };
 203
 204 struct rdma_bind_list {
 205         enum rdma_port_space    ps;
 206         struct hlist_head       owners;
 207         unsigned short          port;
 208 };
 209
 210 struct class_port_info_context {
 211         struct ib_class_port_info       *class_port_info;
 212         struct ib_device                *device;
 213         struct completion               done;
 214         struct ib_sa_query              *sa_query;
 215         u8                              port_num;
 216 };
 217
 218 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 219                         struct rdma_bind_list *bind_list, int snum)
 220 {
 221         struct idr *idr = cma_pernet_idr(net, ps);
 222
 223         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 224 }
 225
 226 static struct rdma_bind_list *cma_ps_find(struct net *net,
 227                                           enum rdma_port_space ps, int snum)
 228 {
 229         struct idr *idr = cma_pernet_idr(net, ps);
 230
 231         return idr_find(idr, snum);
 232 }
 233
 234 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 235 {
 236         struct idr *idr = cma_pernet_idr(net, ps);
 237
 238         idr_remove(idr, snum);
 239 }
 240
 241 enum {
 242         CMA_OPTION_AFONLY,
 243 };
 244
 245 void cma_ref_dev(struct cma_device *cma_dev)
 246 {
 247         atomic_inc(&cma_dev->refcount);
 248 }
 249
 250 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 251                                              void               *cookie)
 252 {
 253         struct cma_device *cma_dev;
 254         struct cma_device *found_cma_dev = NULL;
 255
 256         mutex_lock(&lock);
 257
 258         list_for_each_entry(cma_dev, &dev_list, list)
 259                 if (filter(cma_dev->device, cookie)) {
 260                         found_cma_dev = cma_dev;
 261                         break;
 262                 }
 263
 264         if (found_cma_dev)
 265                 cma_ref_dev(found_cma_dev);
 266         mutex_unlock(&lock);
 267         return found_cma_dev;
 268 }
 269
 270 int cma_get_default_gid_type(struct cma_device *cma_dev,
 271                              unsigned int port)
 272 {
 273         if (!rdma_is_port_valid(cma_dev->device, port))
 274                 return -EINVAL;
 275
 276         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 277 }
 278
 279 int cma_set_default_gid_type(struct cma_device *cma_dev,
 280                              unsigned int port,
 281                              enum ib_gid_type default_gid_type)
 282 {
 283         unsigned long supported_gids;
 284
 285         if (!rdma_is_port_valid(cma_dev->device, port))
 286                 return -EINVAL;
 287
 288         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 289
 290         if (!(supported_gids & 1 << default_gid_type))
 291                 return -EINVAL;
 292
 293         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 294                 default_gid_type;
 295
 296         return 0;
 297 }
 298
 299 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 300 {
 301         if (!rdma_is_port_valid(cma_dev->device, port))
 302                 return -EINVAL;
 303
 304         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 305 }
 306
 307 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 308                              u8 default_roce_tos)
 309 {
 310         if (!rdma_is_port_valid(cma_dev->device, port))
 311                 return -EINVAL;
 312
 313         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 314                  default_roce_tos;
 315
 316         return 0;
 317 }
 318 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 319 {
 320         return cma_dev->device;
 321 }
 322
 323 /*
 324  * Device removal can occur at anytime, so we need extra handling to
 325  * serialize notifying the user of device removal with other callbacks.
 326  * We do this by disabling removal notification while a callback is in process,
 327  * and reporting it after the callback completes.
 328  */
 329 struct rdma_id_private {
 330         struct rdma_cm_id       id;
 331
 332         struct rdma_bind_list   *bind_list;
 333         struct hlist_node       node;
 334         struct list_head        list; /* listen_any_list or cma_device.list */
 335         struct list_head        listen_list; /* per device listens */
 336         struct cma_device       *cma_dev;
 337         struct list_head        mc_list;
 338
 339         int                     internal_id;
 340         enum rdma_cm_state      state;
 341         spinlock_t              lock;
 342         struct mutex            qp_mutex;
 343
 344         struct completion       comp;
 345         atomic_t                refcount;
 346         struct mutex            handler_mutex;
 347
 348         int                     backlog;
 349         int                     timeout_ms;
 350         struct ib_sa_query      *query;
 351         int                     query_id;
 352         union {
 353                 struct ib_cm_id *ib;
 354                 struct iw_cm_id *iw;
 355         } cm_id;
 356
 357         u32                     seq_num;
 358         u32                     qkey;
 359         u32                     qp_num;
 360         pid_t                   owner;
 361         u32                     options;
 362         u8                      srq;
 363         u8                      tos;
 364         bool                    tos_set;
 365         u8                      reuseaddr;
 366         u8                      afonly;
 367         enum ib_gid_type        gid_type;
 368 };
 369
 370 struct cma_multicast {
 371         struct rdma_id_private *id_priv;
 372         union {
 373                 struct ib_sa_multicast *ib;
 374         } multicast;
 375         struct list_head        list;
 376         void                    *context;
 377         struct sockaddr_storage addr;
 378         struct kref             mcref;
 379         bool                    igmp_joined;
 380         u8                      join_state;
 381 };
 382
 383 struct cma_work {
 384         struct work_struct      work;
 385         struct rdma_id_private  *id;
 386         enum rdma_cm_state      old_state;
 387         enum rdma_cm_state      new_state;
 388         struct rdma_cm_event    event;
 389 };
 390
 391 struct cma_ndev_work {
 392         struct work_struct      work;
 393         struct rdma_id_private  *id;
 394         struct rdma_cm_event    event;
 395 };
 396
 397 struct iboe_mcast_work {
 398         struct work_struct       work;
 399         struct rdma_id_private  *id;
 400         struct cma_multicast    *mc;
 401 };
 402
 403 union cma_ip_addr {
 404         struct in6_addr ip6;
 405         struct {
 406                 __be32 pad[3];
 407                 __be32 addr;
 408         } ip4;
 409 };
 410
 411 struct cma_hdr {
 412         u8 cma_version;
 413         u8 ip_version;  /* IP version: 7:4 */
 414         __be16 port;
 415         union cma_ip_addr src_addr;
 416         union cma_ip_addr dst_addr;
 417 };
 418
 419 #define CMA_VERSION 0x00
 420
 421 struct cma_req_info {
 422         struct ib_device *device;
 423         int port;
 424         union ib_gid local_gid;
 425         __be64 service_id;
 426         u16 pkey;
 427         bool has_gid:1;
 428 };
 429
 430 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 431 {
 432         unsigned long flags;
 433         int ret;
 434
 435         spin_lock_irqsave(&id_priv->lock, flags);
 436         ret = (id_priv->state == comp);
 437         spin_unlock_irqrestore(&id_priv->lock, flags);
 438         return ret;
 439 }
 440
 441 static int cma_comp_exch(struct rdma_id_private *id_priv,
 442                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 443 {
 444         unsigned long flags;
 445         int ret;
 446
 447         spin_lock_irqsave(&id_priv->lock, flags);
 448         if ((ret = (id_priv->state == comp)))
 449                 id_priv->state = exch;
 450         spin_unlock_irqrestore(&id_priv->lock, flags);
 451         return ret;
 452 }
 453
 454 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 455                                    enum rdma_cm_state exch)
 456 {
 457         unsigned long flags;
 458         enum rdma_cm_state old;
 459
 460         spin_lock_irqsave(&id_priv->lock, flags);
 461         old = id_priv->state;
 462         id_priv->state = exch;
 463         spin_unlock_irqrestore(&id_priv->lock, flags);
 464         return old;
 465 }
 466
 467 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 468 {
 469         return hdr->ip_version >> 4;
 470 }
 471
 472 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 473 {
 474         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 475 }
 476
 477 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 478 {
 479         struct in_device *in_dev = NULL;
 480
 481         if (ndev) {
 482                 rtnl_lock();
 483                 in_dev = __in_dev_get_rtnl(ndev);
 484                 if (in_dev) {
 485                         if (join)
 486                                 ip_mc_inc_group(in_dev,
 487                                                 *(__be32 *)(mgid->raw + 12));
 488                         else
 489                                 ip_mc_dec_group(in_dev,
 490                                                 *(__be32 *)(mgid->raw + 12));
 491                 }
 492                 rtnl_unlock();
 493         }
 494         return (in_dev) ? 0 : -ENODEV;
 495 }
 496
 497 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 498                                struct cma_device *cma_dev)
 499 {
 500         cma_ref_dev(cma_dev);
 501         id_priv->cma_dev = cma_dev;
 502         id_priv->gid_type = 0;
 503         id_priv->id.device = cma_dev->device;
 504         id_priv->id.route.addr.dev_addr.transport =
 505                 rdma_node_get_transport(cma_dev->device->node_type);
 506         list_add_tail(&id_priv->list, &cma_dev->id_list);
 507 }
 508
 509 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 510                               struct cma_device *cma_dev)
 511 {
 512         _cma_attach_to_dev(id_priv, cma_dev);
 513         id_priv->gid_type =
 514                 cma_dev->default_gid_type[id_priv->id.port_num -
 515                                           rdma_start_port(cma_dev->device)];
 516 }
 517
 518 void cma_deref_dev(struct cma_device *cma_dev)
 519 {
 520         if (atomic_dec_and_test(&cma_dev->refcount))
 521                 complete(&cma_dev->comp);
 522 }
 523
 524 static inline void release_mc(struct kref *kref)
 525 {
 526         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 527
 528         kfree(mc->multicast.ib);
 529         kfree(mc);
 530 }
 531
 532 static void cma_release_dev(struct rdma_id_private *id_priv)
 533 {
 534         mutex_lock(&lock);
 535         list_del(&id_priv->list);
 536         cma_deref_dev(id_priv->cma_dev);
 537         id_priv->cma_dev = NULL;
 538         mutex_unlock(&lock);
 539 }
 540
 541 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 542 {
 543         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 544 }
 545
 546 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 547 {
 548         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 549 }
 550
 551 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 552 {
 553         return id_priv->id.route.addr.src_addr.ss_family;
 554 }
 555
 556 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 557 {
 558         struct ib_sa_mcmember_rec rec;
 559         int ret = 0;
 560
 561         if (id_priv->qkey) {
 562                 if (qkey && id_priv->qkey != qkey)
 563                         return -EINVAL;
 564                 return 0;
 565         }
 566
 567         if (qkey) {
 568                 id_priv->qkey = qkey;
 569                 return 0;
 570         }
 571
 572         switch (id_priv->id.ps) {
 573         case RDMA_PS_UDP:
 574         case RDMA_PS_IB:
 575                 id_priv->qkey = RDMA_UDP_QKEY;
 576                 break;
 577         case RDMA_PS_IPOIB:
 578                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 579                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 580                                              id_priv->id.port_num, &rec.mgid,
 581                                              &rec);
 582                 if (!ret)
 583                         id_priv->qkey = be32_to_cpu(rec.qkey);
 584                 break;
 585         default:
 586                 break;
 587         }
 588         return ret;
 589 }
 590
 591 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 592 {
 593         dev_addr->dev_type = ARPHRD_INFINIBAND;
 594         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 595         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 596 }
 597
 598 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 599 {
 600         int ret;
 601
 602         if (addr->sa_family != AF_IB) {
 603                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 604         } else {
 605                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 606                 ret = 0;
 607         }
 608
 609         return ret;
 610 }
 611
 612 static inline int cma_validate_port(struct ib_device *device, u8 port,
 613                                     enum ib_gid_type gid_type,
 614                                       union ib_gid *gid, int dev_type,
 615                                       int bound_if_index)
 616 {
 617         int ret = -ENODEV;
 618         struct net_device *ndev = NULL;
 619
 620         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 621                 return ret;
 622
 623         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 624                 return ret;
 625
 626         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port))
 627                 ndev = dev_get_by_index(&init_net, bound_if_index);
 628         else
 629                 gid_type = IB_GID_TYPE_IB;
 630
 631
 632         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 633                                          ndev, NULL);
 634
 635         if (ndev)
 636                 dev_put(ndev);
 637
 638         return ret;
 639 }
 640
 641 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 642                            struct rdma_id_private *listen_id_priv)
 643 {
 644         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 645         struct cma_device *cma_dev;
 646         union ib_gid gid, iboe_gid, *gidp;
 647         int ret = -ENODEV;
 648         u8 port;
 649
 650         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 651             id_priv->id.ps == RDMA_PS_IPOIB)
 652                 return -EINVAL;
 653
 654         mutex_lock(&lock);
 655         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 656                     &iboe_gid);
 657
 658         memcpy(&gid, dev_addr->src_dev_addr +
 659                rdma_addr_gid_offset(dev_addr), sizeof gid);
 660
 661         if (listen_id_priv) {
 662                 cma_dev = listen_id_priv->cma_dev;
 663                 port = listen_id_priv->id.port_num;
 664                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 665                        &iboe_gid : &gid;
 666
 667                 ret = cma_validate_port(cma_dev->device, port,
 668                                         rdma_protocol_ib(cma_dev->device, port) ?
 669                                         IB_GID_TYPE_IB :
 670                                         listen_id_priv->gid_type, gidp,
 671                                         dev_addr->dev_type,
 672                                         dev_addr->bound_dev_if);
 673                 if (!ret) {
 674                         id_priv->id.port_num = port;
 675                         goto out;
 676                 }
 677         }
 678
 679         list_for_each_entry(cma_dev, &dev_list, list) {
 680                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 681                         if (listen_id_priv &&
 682                             listen_id_priv->cma_dev == cma_dev &&
 683                             listen_id_priv->id.port_num == port)
 684                                 continue;
 685
 686                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 687                                &iboe_gid : &gid;
 688
 689                         ret = cma_validate_port(cma_dev->device, port,
 690                                                 rdma_protocol_ib(cma_dev->device, port) ?
 691                                                 IB_GID_TYPE_IB :
 692                                                 cma_dev->default_gid_type[port - 1],
 693                                                 gidp, dev_addr->dev_type,
 694                                                 dev_addr->bound_dev_if);
 695                         if (!ret) {
 696                                 id_priv->id.port_num = port;
 697                                 goto out;
 698                         }
 699                 }
 700         }
 701
 702 out:
 703         if (!ret)
 704                 cma_attach_to_dev(id_priv, cma_dev);
 705
 706         mutex_unlock(&lock);
 707         return ret;
 708 }
 709
 710 /*
 711  * Select the source IB device and address to reach the destination IB address.
 712  */
 713 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 714 {
 715         struct cma_device *cma_dev, *cur_dev;
 716         struct sockaddr_ib *addr;
 717         union ib_gid gid, sgid, *dgid;
 718         u16 pkey, index;
 719         u8 p;
 720         enum ib_port_state port_state;
 721         int i;
 722
 723         cma_dev = NULL;
 724         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 725         dgid = (union ib_gid *) &addr->sib_addr;
 726         pkey = ntohs(addr->sib_pkey);
 727
 728         list_for_each_entry(cur_dev, &dev_list, list) {
 729                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 730                         if (!rdma_cap_af_ib(cur_dev->device, p))
 731                                 continue;
 732
 733                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 734                                 continue;
 735
 736                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 737                                 continue;
 738                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 739                                                        &gid, NULL);
 740                              i++) {
 741                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 742                                         cma_dev = cur_dev;
 743                                         sgid = gid;
 744                                         id_priv->id.port_num = p;
 745                                         goto found;
 746                                 }
 747
 748                                 if (!cma_dev && (gid.global.subnet_prefix ==
 749                                     dgid->global.subnet_prefix) &&
 750                                     port_state == IB_PORT_ACTIVE) {
 751                                         cma_dev = cur_dev;
 752                                         sgid = gid;
 753                                         id_priv->id.port_num = p;
 754                                 }
 755                         }
 756                 }
 757         }
 758
 759         if (!cma_dev)
 760                 return -ENODEV;
 761
 762 found:
 763         cma_attach_to_dev(id_priv, cma_dev);
 764         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 765         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 766         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 767         return 0;
 768 }
 769
 770 static void cma_deref_id(struct rdma_id_private *id_priv)
 771 {
 772         if (atomic_dec_and_test(&id_priv->refcount))
 773                 complete(&id_priv->comp);
 774 }
 775
 776 struct rdma_cm_id *rdma_create_id(struct net *net,
 777                                   rdma_cm_event_handler event_handler,
 778                                   void *context, enum rdma_port_space ps,
 779                                   enum ib_qp_type qp_type)
 780 {
 781         struct rdma_id_private *id_priv;
 782
 783         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 784         if (!id_priv)
 785                 return ERR_PTR(-ENOMEM);
 786
 787         id_priv->owner = task_pid_nr(current);
 788         id_priv->state = RDMA_CM_IDLE;
 789         id_priv->id.context = context;
 790         id_priv->id.event_handler = event_handler;
 791         id_priv->id.ps = ps;
 792         id_priv->id.qp_type = qp_type;
 793         id_priv->tos_set = false;
 794         spin_lock_init(&id_priv->lock);
 795         mutex_init(&id_priv->qp_mutex);
 796         init_completion(&id_priv->comp);
 797         atomic_set(&id_priv->refcount, 1);
 798         mutex_init(&id_priv->handler_mutex);
 799         INIT_LIST_HEAD(&id_priv->listen_list);
 800         INIT_LIST_HEAD(&id_priv->mc_list);
 801         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 802         id_priv->id.route.addr.dev_addr.net = get_net(net);
 803
 804         return &id_priv->id;
 805 }
 806 EXPORT_SYMBOL(rdma_create_id);
 807
 808 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 809 {
 810         struct ib_qp_attr qp_attr;
 811         int qp_attr_mask, ret;
 812
 813         qp_attr.qp_state = IB_QPS_INIT;
 814         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 815         if (ret)
 816                 return ret;
 817
 818         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 819         if (ret)
 820                 return ret;
 821
 822         qp_attr.qp_state = IB_QPS_RTR;
 823         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 824         if (ret)
 825                 return ret;
 826
 827         qp_attr.qp_state = IB_QPS_RTS;
 828         qp_attr.sq_psn = 0;
 829         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 830
 831         return ret;
 832 }
 833
 834 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 835 {
 836         struct ib_qp_attr qp_attr;
 837         int qp_attr_mask, ret;
 838
 839         qp_attr.qp_state = IB_QPS_INIT;
 840         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 841         if (ret)
 842                 return ret;
 843
 844         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 845 }
 846
 847 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 848                    struct ib_qp_init_attr *qp_init_attr)
 849 {
 850         struct rdma_id_private *id_priv;
 851         struct ib_qp *qp;
 852         int ret;
 853
 854         id_priv = container_of(id, struct rdma_id_private, id);
 855         if (id->device != pd->device)
 856                 return -EINVAL;
 857
 858         qp_init_attr->port_num = id->port_num;
 859         qp = ib_create_qp(pd, qp_init_attr);
 860         if (IS_ERR(qp))
 861                 return PTR_ERR(qp);
 862
 863         if (id->qp_type == IB_QPT_UD)
 864                 ret = cma_init_ud_qp(id_priv, qp);
 865         else
 866                 ret = cma_init_conn_qp(id_priv, qp);
 867         if (ret)
 868                 goto err;
 869
 870         id->qp = qp;
 871         id_priv->qp_num = qp->qp_num;
 872         id_priv->srq = (qp->srq != NULL);
 873         return 0;
 874 err:
 875         ib_destroy_qp(qp);
 876         return ret;
 877 }
 878 EXPORT_SYMBOL(rdma_create_qp);
 879
 880 void rdma_destroy_qp(struct rdma_cm_id *id)
 881 {
 882         struct rdma_id_private *id_priv;
 883
 884         id_priv = container_of(id, struct rdma_id_private, id);
 885         mutex_lock(&id_priv->qp_mutex);
 886         ib_destroy_qp(id_priv->id.qp);
 887         id_priv->id.qp = NULL;
 888         mutex_unlock(&id_priv->qp_mutex);
 889 }
 890 EXPORT_SYMBOL(rdma_destroy_qp);
 891
 892 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 893                              struct rdma_conn_param *conn_param)
 894 {
 895         struct ib_qp_attr qp_attr;
 896         int qp_attr_mask, ret;
 897         union ib_gid sgid;
 898
 899         mutex_lock(&id_priv->qp_mutex);
 900         if (!id_priv->id.qp) {
 901                 ret = 0;
 902                 goto out;
 903         }
 904
 905         /* Need to update QP attributes from default values. */
 906         qp_attr.qp_state = IB_QPS_INIT;
 907         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 908         if (ret)
 909                 goto out;
 910
 911         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 912         if (ret)
 913                 goto out;
 914
 915         qp_attr.qp_state = IB_QPS_RTR;
 916         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 917         if (ret)
 918                 goto out;
 919
 920         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 921                            rdma_ah_read_grh(&qp_attr.ah_attr)->sgid_index,
 922                            &sgid, NULL);
 923         if (ret)
 924                 goto out;
 925
 926         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 927
 928         if (conn_param)
 929                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 930         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 931 out:
 932         mutex_unlock(&id_priv->qp_mutex);
 933         return ret;
 934 }
 935
 936 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 937                              struct rdma_conn_param *conn_param)
 938 {
 939         struct ib_qp_attr qp_attr;
 940         int qp_attr_mask, ret;
 941
 942         mutex_lock(&id_priv->qp_mutex);
 943         if (!id_priv->id.qp) {
 944                 ret = 0;
 945                 goto out;
 946         }
 947
 948         qp_attr.qp_state = IB_QPS_RTS;
 949         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 950         if (ret)
 951                 goto out;
 952
 953         if (conn_param)
 954                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 955         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 956 out:
 957         mutex_unlock(&id_priv->qp_mutex);
 958         return ret;
 959 }
 960
 961 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 962 {
 963         struct ib_qp_attr qp_attr;
 964         int ret;
 965
 966         mutex_lock(&id_priv->qp_mutex);
 967         if (!id_priv->id.qp) {
 968                 ret = 0;
 969                 goto out;
 970         }
 971
 972         qp_attr.qp_state = IB_QPS_ERR;
 973         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 974 out:
 975         mutex_unlock(&id_priv->qp_mutex);
 976         return ret;
 977 }
 978
 979 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 980                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 981 {
 982         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 983         int ret;
 984         u16 pkey;
 985
 986         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 987                 pkey = 0xffff;
 988         else
 989                 pkey = ib_addr_get_pkey(dev_addr);
 990
 991         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 992                                   pkey, &qp_attr->pkey_index);
 993         if (ret)
 994                 return ret;
 995
 996         qp_attr->port_num = id_priv->id.port_num;
 997         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 998
 999         if (id_priv->id.qp_type == IB_QPT_UD) {
1000                 ret = cma_set_qkey(id_priv, 0);
1001                 if (ret)
1002                         return ret;
1003
1004                 qp_attr->qkey = id_priv->qkey;
1005                 *qp_attr_mask |= IB_QP_QKEY;
1006         } else {
1007                 qp_attr->qp_access_flags = 0;
1008                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1009         }
1010         return 0;
1011 }
1012
1013 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1014                        int *qp_attr_mask)
1015 {
1016         struct rdma_id_private *id_priv;
1017         int ret = 0;
1018
1019         id_priv = container_of(id, struct rdma_id_private, id);
1020         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1021                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1022                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1023                 else
1024                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1025                                                  qp_attr_mask);
1026
1027                 if (qp_attr->qp_state == IB_QPS_RTR)
1028                         qp_attr->rq_psn = id_priv->seq_num;
1029         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1030                 if (!id_priv->cm_id.iw) {
1031                         qp_attr->qp_access_flags = 0;
1032                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1033                 } else
1034                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1035                                                  qp_attr_mask);
1036         } else
1037                 ret = -ENOSYS;
1038
1039         return ret;
1040 }
1041 EXPORT_SYMBOL(rdma_init_qp_attr);
1042
1043 static inline int cma_zero_addr(struct sockaddr *addr)
1044 {
1045         switch (addr->sa_family) {
1046         case AF_INET:
1047                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1048         case AF_INET6:
1049                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1050         case AF_IB:
1051                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1052         default:
1053                 return 0;
1054         }
1055 }
1056
1057 static inline int cma_loopback_addr(struct sockaddr *addr)
1058 {
1059         switch (addr->sa_family) {
1060         case AF_INET:
1061                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1062         case AF_INET6:
1063                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1064         case AF_IB:
1065                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1066         default:
1067                 return 0;
1068         }
1069 }
1070
1071 static inline int cma_any_addr(struct sockaddr *addr)
1072 {
1073         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1074 }
1075
1076 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1077 {
1078         if (src->sa_family != dst->sa_family)
1079                 return -1;
1080
1081         switch (src->sa_family) {
1082         case AF_INET:
1083                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1084                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1085         case AF_INET6:
1086                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1087                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1088         default:
1089                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1090                                    &((struct sockaddr_ib *) dst)->sib_addr);
1091         }
1092 }
1093
1094 static __be16 cma_port(struct sockaddr *addr)
1095 {
1096         struct sockaddr_ib *sib;
1097
1098         switch (addr->sa_family) {
1099         case AF_INET:
1100                 return ((struct sockaddr_in *) addr)->sin_port;
1101         case AF_INET6:
1102                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1103         case AF_IB:
1104                 sib = (struct sockaddr_ib *) addr;
1105                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1106                                     be64_to_cpu(sib->sib_sid_mask)));
1107         default:
1108                 return 0;
1109         }
1110 }
1111
1112 static inline int cma_any_port(struct sockaddr *addr)
1113 {
1114         return !cma_port(addr);
1115 }
1116
1117 static void cma_save_ib_info(struct sockaddr *src_addr,
1118                              struct sockaddr *dst_addr,
1119                              struct rdma_cm_id *listen_id,
1120                              struct sa_path_rec *path)
1121 {
1122         struct sockaddr_ib *listen_ib, *ib;
1123
1124         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1125         if (src_addr) {
1126                 ib = (struct sockaddr_ib *)src_addr;
1127                 ib->sib_family = AF_IB;
1128                 if (path) {
1129                         ib->sib_pkey = path->pkey;
1130                         ib->sib_flowinfo = path->flow_label;
1131                         memcpy(&ib->sib_addr, &path->sgid, 16);
1132                         ib->sib_sid = path->service_id;
1133                         ib->sib_scope_id = 0;
1134                 } else {
1135                         ib->sib_pkey = listen_ib->sib_pkey;
1136                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1137                         ib->sib_addr = listen_ib->sib_addr;
1138                         ib->sib_sid = listen_ib->sib_sid;
1139                         ib->sib_scope_id = listen_ib->sib_scope_id;
1140                 }
1141                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1142         }
1143         if (dst_addr) {
1144                 ib = (struct sockaddr_ib *)dst_addr;
1145                 ib->sib_family = AF_IB;
1146                 if (path) {
1147                         ib->sib_pkey = path->pkey;
1148                         ib->sib_flowinfo = path->flow_label;
1149                         memcpy(&ib->sib_addr, &path->dgid, 16);
1150                 }
1151         }
1152 }
1153
1154 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1155                               struct sockaddr_in *dst_addr,
1156                               struct cma_hdr *hdr,
1157                               __be16 local_port)
1158 {
1159         if (src_addr) {
1160                 *src_addr = (struct sockaddr_in) {
1161                         .sin_family = AF_INET,
1162                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1163                         .sin_port = local_port,
1164                 };
1165         }
1166
1167         if (dst_addr) {
1168                 *dst_addr = (struct sockaddr_in) {
1169                         .sin_family = AF_INET,
1170                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1171                         .sin_port = hdr->port,
1172                 };
1173         }
1174 }
1175
1176 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1177                               struct sockaddr_in6 *dst_addr,
1178                               struct cma_hdr *hdr,
1179                               __be16 local_port)
1180 {
1181         if (src_addr) {
1182                 *src_addr = (struct sockaddr_in6) {
1183                         .sin6_family = AF_INET6,
1184                         .sin6_addr = hdr->dst_addr.ip6,
1185                         .sin6_port = local_port,
1186                 };
1187         }
1188
1189         if (dst_addr) {
1190                 *dst_addr = (struct sockaddr_in6) {
1191                         .sin6_family = AF_INET6,
1192                         .sin6_addr = hdr->src_addr.ip6,
1193                         .sin6_port = hdr->port,
1194                 };
1195         }
1196 }
1197
1198 static u16 cma_port_from_service_id(__be64 service_id)
1199 {
1200         return (u16)be64_to_cpu(service_id);
1201 }
1202
1203 static int cma_save_ip_info(struct sockaddr *src_addr,
1204                             struct sockaddr *dst_addr,
1205                             struct ib_cm_event *ib_event,
1206                             __be64 service_id)
1207 {
1208         struct cma_hdr *hdr;
1209         __be16 port;
1210
1211         hdr = ib_event->private_data;
1212         if (hdr->cma_version != CMA_VERSION)
1213                 return -EINVAL;
1214
1215         port = htons(cma_port_from_service_id(service_id));
1216
1217         switch (cma_get_ip_ver(hdr)) {
1218         case 4:
1219                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1220                                   (struct sockaddr_in *)dst_addr, hdr, port);
1221                 break;
1222         case 6:
1223                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1224                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1225                 break;
1226         default:
1227                 return -EAFNOSUPPORT;
1228         }
1229
1230         return 0;
1231 }
1232
1233 static int cma_save_net_info(struct sockaddr *src_addr,
1234                              struct sockaddr *dst_addr,
1235                              struct rdma_cm_id *listen_id,
1236                              struct ib_cm_event *ib_event,
1237                              sa_family_t sa_family, __be64 service_id)
1238 {
1239         if (sa_family == AF_IB) {
1240                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1241                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1242                                          ib_event->param.req_rcvd.primary_path);
1243                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1244                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1245                 return 0;
1246         }
1247
1248         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1249 }
1250
1251 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1252                              struct cma_req_info *req)
1253 {
1254         const struct ib_cm_req_event_param *req_param =
1255                 &ib_event->param.req_rcvd;
1256         const struct ib_cm_sidr_req_event_param *sidr_param =
1257                 &ib_event->param.sidr_req_rcvd;
1258
1259         switch (ib_event->event) {
1260         case IB_CM_REQ_RECEIVED:
1261                 req->device     = req_param->listen_id->device;
1262                 req->port       = req_param->port;
1263                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1264                        sizeof(req->local_gid));
1265                 req->has_gid    = true;
1266                 req->service_id = req_param->primary_path->service_id;
1267                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1268                 if (req->pkey != req_param->bth_pkey)
1269                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1270                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1271                                             req_param->bth_pkey, req->pkey);
1272                 break;
1273         case IB_CM_SIDR_REQ_RECEIVED:
1274                 req->device     = sidr_param->listen_id->device;
1275                 req->port       = sidr_param->port;
1276                 req->has_gid    = false;
1277                 req->service_id = sidr_param->service_id;
1278                 req->pkey       = sidr_param->pkey;
1279                 if (req->pkey != sidr_param->bth_pkey)
1280                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1281                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1282                                             sidr_param->bth_pkey, req->pkey);
1283                 break;
1284         default:
1285                 return -EINVAL;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1292                                   const struct sockaddr_in *dst_addr,
1293                                   const struct sockaddr_in *src_addr)
1294 {
1295         __be32 daddr = dst_addr->sin_addr.s_addr,
1296                saddr = src_addr->sin_addr.s_addr;
1297         struct fib_result res;
1298         struct flowi4 fl4;
1299         int err;
1300         bool ret;
1301
1302         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1303             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1304             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1305             ipv4_is_loopback(saddr))
1306                 return false;
1307
1308         memset(&fl4, 0, sizeof(fl4));
1309         fl4.flowi4_iif = net_dev->ifindex;
1310         fl4.daddr = daddr;
1311         fl4.saddr = saddr;
1312
1313         rcu_read_lock();
1314         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1315         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1316         rcu_read_unlock();
1317
1318         return ret;
1319 }
1320
1321 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1322                                   const struct sockaddr_in6 *dst_addr,
1323                                   const struct sockaddr_in6 *src_addr)
1324 {
1325 #if IS_ENABLED(CONFIG_IPV6)
1326         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1327                            IPV6_ADDR_LINKLOCAL;
1328         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1329                                          &src_addr->sin6_addr, net_dev->ifindex,
1330                                          strict);
1331         bool ret;
1332
1333         if (!rt)
1334                 return false;
1335
1336         ret = rt->rt6i_idev->dev == net_dev;
1337         ip6_rt_put(rt);
1338
1339         return ret;
1340 #else
1341         return false;
1342 #endif
1343 }
1344
1345 static bool validate_net_dev(struct net_device *net_dev,
1346                              const struct sockaddr *daddr,
1347                              const struct sockaddr *saddr)
1348 {
1349         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1350         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1351         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1352         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1353
1354         switch (daddr->sa_family) {
1355         case AF_INET:
1356                 return saddr->sa_family == AF_INET &&
1357                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1358
1359         case AF_INET6:
1360                 return saddr->sa_family == AF_INET6 &&
1361                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1362
1363         default:
1364                 return false;
1365         }
1366 }
1367
1368 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1369                                           const struct cma_req_info *req)
1370 {
1371         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1372         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1373                         *src_addr = (struct sockaddr *)&src_addr_storage;
1374         struct net_device *net_dev;
1375         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1376         int err;
1377
1378         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1379                                req->service_id);
1380         if (err)
1381                 return ERR_PTR(err);
1382
1383         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1384                                            gid, listen_addr);
1385         if (!net_dev)
1386                 return ERR_PTR(-ENODEV);
1387
1388         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1389                 dev_put(net_dev);
1390                 return ERR_PTR(-EHOSTUNREACH);
1391         }
1392
1393         return net_dev;
1394 }
1395
1396 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1397 {
1398         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1399 }
1400
1401 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1402                                    const struct cma_hdr *hdr)
1403 {
1404         struct sockaddr *addr = cma_src_addr(id_priv);
1405         __be32 ip4_addr;
1406         struct in6_addr ip6_addr;
1407
1408         if (cma_any_addr(addr) && !id_priv->afonly)
1409                 return true;
1410
1411         switch (addr->sa_family) {
1412         case AF_INET:
1413                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1414                 if (cma_get_ip_ver(hdr) != 4)
1415                         return false;
1416                 if (!cma_any_addr(addr) &&
1417                     hdr->dst_addr.ip4.addr != ip4_addr)
1418                         return false;
1419                 break;
1420         case AF_INET6:
1421                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1422                 if (cma_get_ip_ver(hdr) != 6)
1423                         return false;
1424                 if (!cma_any_addr(addr) &&
1425                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1426                         return false;
1427                 break;
1428         case AF_IB:
1429                 return true;
1430         default:
1431                 return false;
1432         }
1433
1434         return true;
1435 }
1436
1437 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1438 {
1439         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1440         enum rdma_transport_type transport =
1441                 rdma_node_get_transport(device->node_type);
1442
1443         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1444 }
1445
1446 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1447 {
1448         struct ib_device *device = id->device;
1449         const int port_num = id->port_num ?: rdma_start_port(device);
1450
1451         return cma_protocol_roce_dev_port(device, port_num);
1452 }
1453
1454 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1455                               const struct net_device *net_dev,
1456                               u8 port_num)
1457 {
1458         const struct rdma_addr *addr = &id->route.addr;
1459
1460         if (!net_dev)
1461                 /* This request is an AF_IB request or a RoCE request */
1462                 return (!id->port_num || id->port_num == port_num) &&
1463                        (addr->src_addr.ss_family == AF_IB ||
1464                         cma_protocol_roce_dev_port(id->device, port_num));
1465
1466         return !addr->dev_addr.bound_dev_if ||
1467                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1468                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1469 }
1470
1471 static struct rdma_id_private *cma_find_listener(
1472                 const struct rdma_bind_list *bind_list,
1473                 const struct ib_cm_id *cm_id,
1474                 const struct ib_cm_event *ib_event,
1475                 const struct cma_req_info *req,
1476                 const struct net_device *net_dev)
1477 {
1478         struct rdma_id_private *id_priv, *id_priv_dev;
1479
1480         if (!bind_list)
1481                 return ERR_PTR(-EINVAL);
1482
1483         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1484                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1485                         if (id_priv->id.device == cm_id->device &&
1486                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1487                                 return id_priv;
1488                         list_for_each_entry(id_priv_dev,
1489                                             &id_priv->listen_list,
1490                                             listen_list) {
1491                                 if (id_priv_dev->id.device == cm_id->device &&
1492                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1493                                         return id_priv_dev;
1494                         }
1495                 }
1496         }
1497
1498         return ERR_PTR(-EINVAL);
1499 }
1500
1501 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1502                                                  struct ib_cm_event *ib_event,
1503                                                  struct net_device **net_dev)
1504 {
1505         struct cma_req_info req;
1506         struct rdma_bind_list *bind_list;
1507         struct rdma_id_private *id_priv;
1508         int err;
1509
1510         err = cma_save_req_info(ib_event, &req);
1511         if (err)
1512                 return ERR_PTR(err);
1513
1514         *net_dev = cma_get_net_dev(ib_event, &req);
1515         if (IS_ERR(*net_dev)) {
1516                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1517                         /* Assuming the protocol is AF_IB */
1518                         *net_dev = NULL;
1519                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1520                         /* TODO find the net dev matching the request parameters
1521                          * through the RoCE GID table */
1522                         *net_dev = NULL;
1523                 } else {
1524                         return ERR_CAST(*net_dev);
1525                 }
1526         }
1527
1528         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1529                                 rdma_ps_from_service_id(req.service_id),
1530                                 cma_port_from_service_id(req.service_id));
1531         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1532         if (IS_ERR(id_priv) && *net_dev) {
1533                 dev_put(*net_dev);
1534                 *net_dev = NULL;
1535         }
1536
1537         return id_priv;
1538 }
1539
1540 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1541 {
1542         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1543 }
1544
1545 static void cma_cancel_route(struct rdma_id_private *id_priv)
1546 {
1547         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1548                 if (id_priv->query)
1549                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1550         }
1551 }
1552
1553 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1554 {
1555         struct rdma_id_private *dev_id_priv;
1556
1557         /*
1558          * Remove from listen_any_list to prevent added devices from spawning
1559          * additional listen requests.
1560          */
1561         mutex_lock(&lock);
1562         list_del(&id_priv->list);
1563
1564         while (!list_empty(&id_priv->listen_list)) {
1565                 dev_id_priv = list_entry(id_priv->listen_list.next,
1566                                          struct rdma_id_private, listen_list);
1567                 /* sync with device removal to avoid duplicate destruction */
1568                 list_del_init(&dev_id_priv->list);
1569                 list_del(&dev_id_priv->listen_list);
1570                 mutex_unlock(&lock);
1571
1572                 rdma_destroy_id(&dev_id_priv->id);
1573                 mutex_lock(&lock);
1574         }
1575         mutex_unlock(&lock);
1576 }
1577
1578 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1579                                  enum rdma_cm_state state)
1580 {
1581         switch (state) {
1582         case RDMA_CM_ADDR_QUERY:
1583                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1584                 break;
1585         case RDMA_CM_ROUTE_QUERY:
1586                 cma_cancel_route(id_priv);
1587                 break;
1588         case RDMA_CM_LISTEN:
1589                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1590                         cma_cancel_listens(id_priv);
1591                 break;
1592         default:
1593                 break;
1594         }
1595 }
1596
1597 static void cma_release_port(struct rdma_id_private *id_priv)
1598 {
1599         struct rdma_bind_list *bind_list = id_priv->bind_list;
1600         struct net *net = id_priv->id.route.addr.dev_addr.net;
1601
1602         if (!bind_list)
1603                 return;
1604
1605         mutex_lock(&lock);
1606         hlist_del(&id_priv->node);
1607         if (hlist_empty(&bind_list->owners)) {
1608                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1609                 kfree(bind_list);
1610         }
1611         mutex_unlock(&lock);
1612 }
1613
1614 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1615 {
1616         struct cma_multicast *mc;
1617
1618         while (!list_empty(&id_priv->mc_list)) {
1619                 mc = container_of(id_priv->mc_list.next,
1620                                   struct cma_multicast, list);
1621                 list_del(&mc->list);
1622                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1623                                       id_priv->id.port_num)) {
1624                         ib_sa_free_multicast(mc->multicast.ib);
1625                         kfree(mc);
1626                 } else {
1627                         if (mc->igmp_joined) {
1628                                 struct rdma_dev_addr *dev_addr =
1629                                         &id_priv->id.route.addr.dev_addr;
1630                                 struct net_device *ndev = NULL;
1631
1632                                 if (dev_addr->bound_dev_if)
1633                                         ndev = dev_get_by_index(&init_net,
1634                                                                 dev_addr->bound_dev_if);
1635                                 if (ndev) {
1636                                         cma_igmp_send(ndev,
1637                                                       &mc->multicast.ib->rec.mgid,
1638                                                       false);
1639                                         dev_put(ndev);
1640                                 }
1641                         }
1642                         kref_put(&mc->mcref, release_mc);
1643                 }
1644         }
1645 }
1646
1647 void rdma_destroy_id(struct rdma_cm_id *id)
1648 {
1649         struct rdma_id_private *id_priv;
1650         enum rdma_cm_state state;
1651
1652         id_priv = container_of(id, struct rdma_id_private, id);
1653         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1654         cma_cancel_operation(id_priv, state);
1655
1656         /*
1657          * Wait for any active callback to finish.  New callbacks will find
1658          * the id_priv state set to destroying and abort.
1659          */
1660         mutex_lock(&id_priv->handler_mutex);
1661         mutex_unlock(&id_priv->handler_mutex);
1662
1663         if (id_priv->cma_dev) {
1664                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1665                         if (id_priv->cm_id.ib)
1666                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1667                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1668                         if (id_priv->cm_id.iw)
1669                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1670                 }
1671                 cma_leave_mc_groups(id_priv);
1672                 cma_release_dev(id_priv);
1673         }
1674
1675         cma_release_port(id_priv);
1676         cma_deref_id(id_priv);
1677         wait_for_completion(&id_priv->comp);
1678
1679         if (id_priv->internal_id)
1680                 cma_deref_id(id_priv->id.context);
1681
1682         kfree(id_priv->id.route.path_rec);
1683         put_net(id_priv->id.route.addr.dev_addr.net);
1684         kfree(id_priv);
1685 }
1686 EXPORT_SYMBOL(rdma_destroy_id);
1687
1688 static int cma_rep_recv(struct rdma_id_private *id_priv)
1689 {
1690         int ret;
1691
1692         ret = cma_modify_qp_rtr(id_priv, NULL);
1693         if (ret)
1694                 goto reject;
1695
1696         ret = cma_modify_qp_rts(id_priv, NULL);
1697         if (ret)
1698                 goto reject;
1699
1700         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1701         if (ret)
1702                 goto reject;
1703
1704         return 0;
1705 reject:
1706         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1707         cma_modify_qp_err(id_priv);
1708         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1709                        NULL, 0, NULL, 0);
1710         return ret;
1711 }
1712
1713 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1714                                    struct ib_cm_rep_event_param *rep_data,
1715                                    void *private_data)
1716 {
1717         event->param.conn.private_data = private_data;
1718         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1719         event->param.conn.responder_resources = rep_data->responder_resources;
1720         event->param.conn.initiator_depth = rep_data->initiator_depth;
1721         event->param.conn.flow_control = rep_data->flow_control;
1722         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1723         event->param.conn.srq = rep_data->srq;
1724         event->param.conn.qp_num = rep_data->remote_qpn;
1725 }
1726
1727 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1728 {
1729         struct rdma_id_private *id_priv = cm_id->context;
1730         struct rdma_cm_event event;
1731         int ret = 0;
1732
1733         mutex_lock(&id_priv->handler_mutex);
1734         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1735              id_priv->state != RDMA_CM_CONNECT) ||
1736             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1737              id_priv->state != RDMA_CM_DISCONNECT))
1738                 goto out;
1739
1740         memset(&event, 0, sizeof event);
1741         switch (ib_event->event) {
1742         case IB_CM_REQ_ERROR:
1743         case IB_CM_REP_ERROR:
1744                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1745                 event.status = -ETIMEDOUT;
1746                 break;
1747         case IB_CM_REP_RECEIVED:
1748                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1749                     (id_priv->id.qp_type != IB_QPT_UD))
1750                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1751                 if (id_priv->id.qp) {
1752                         event.status = cma_rep_recv(id_priv);
1753                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1754                                                      RDMA_CM_EVENT_ESTABLISHED;
1755                 } else {
1756                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1757                 }
1758                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1759                                        ib_event->private_data);
1760                 break;
1761         case IB_CM_RTU_RECEIVED:
1762         case IB_CM_USER_ESTABLISHED:
1763                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1764                 break;
1765         case IB_CM_DREQ_ERROR:
1766                 event.status = -ETIMEDOUT; /* fall through */
1767         case IB_CM_DREQ_RECEIVED:
1768         case IB_CM_DREP_RECEIVED:
1769                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1770                                    RDMA_CM_DISCONNECT))
1771                         goto out;
1772                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1773                 break;
1774         case IB_CM_TIMEWAIT_EXIT:
1775                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1776                 break;
1777         case IB_CM_MRA_RECEIVED:
1778                 /* ignore event */
1779                 goto out;
1780         case IB_CM_REJ_RECEIVED:
1781                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1782                                                                                 ib_event->param.rej_rcvd.reason));
1783                 cma_modify_qp_err(id_priv);
1784                 event.status = ib_event->param.rej_rcvd.reason;
1785                 event.event = RDMA_CM_EVENT_REJECTED;
1786                 event.param.conn.private_data = ib_event->private_data;
1787                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1788                 break;
1789         default:
1790                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1791                        ib_event->event);
1792                 goto out;
1793         }
1794
1795         ret = id_priv->id.event_handler(&id_priv->id, &event);
1796         if (ret) {
1797                 /* Destroy the CM ID by returning a non-zero value. */
1798                 id_priv->cm_id.ib = NULL;
1799                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1800                 mutex_unlock(&id_priv->handler_mutex);
1801                 rdma_destroy_id(&id_priv->id);
1802                 return ret;
1803         }
1804 out:
1805         mutex_unlock(&id_priv->handler_mutex);
1806         return ret;
1807 }
1808
1809 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1810                                                struct ib_cm_event *ib_event,
1811                                                struct net_device *net_dev)
1812 {
1813         struct rdma_id_private *id_priv;
1814         struct rdma_cm_id *id;
1815         struct rdma_route *rt;
1816         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1817         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1818         const __be64 service_id =
1819                 ib_event->param.req_rcvd.primary_path->service_id;
1820         int ret;
1821
1822         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1823                             listen_id->event_handler, listen_id->context,
1824                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1825         if (IS_ERR(id))
1826                 return NULL;
1827
1828         id_priv = container_of(id, struct rdma_id_private, id);
1829         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1830                               (struct sockaddr *)&id->route.addr.dst_addr,
1831                               listen_id, ib_event, ss_family, service_id))
1832                 goto err;
1833
1834         rt = &id->route;
1835         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1836         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1837                                GFP_KERNEL);
1838         if (!rt->path_rec)
1839                 goto err;
1840
1841         rt->path_rec[0] = *path;
1842         if (rt->num_paths == 2)
1843                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1844
1845         if (net_dev) {
1846                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1847                 if (ret)
1848                         goto err;
1849         } else {
1850                 if (!cma_protocol_roce(listen_id) &&
1851                     cma_any_addr(cma_src_addr(id_priv))) {
1852                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1853                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1854                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1855                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1856                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1857                         if (ret)
1858                                 goto err;
1859                 }
1860         }
1861         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1862
1863         id_priv->state = RDMA_CM_CONNECT;
1864         return id_priv;
1865
1866 err:
1867         rdma_destroy_id(id);
1868         return NULL;
1869 }
1870
1871 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1872                                               struct ib_cm_event *ib_event,
1873                                               struct net_device *net_dev)
1874 {
1875         struct rdma_id_private *id_priv;
1876         struct rdma_cm_id *id;
1877         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1878         struct net *net = listen_id->route.addr.dev_addr.net;
1879         int ret;
1880
1881         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1882                             listen_id->ps, IB_QPT_UD);
1883         if (IS_ERR(id))
1884                 return NULL;
1885
1886         id_priv = container_of(id, struct rdma_id_private, id);
1887         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1888                               (struct sockaddr *)&id->route.addr.dst_addr,
1889                               listen_id, ib_event, ss_family,
1890                               ib_event->param.sidr_req_rcvd.service_id))
1891                 goto err;
1892
1893         if (net_dev) {
1894                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1895                 if (ret)
1896                         goto err;
1897         } else {
1898                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1899                         ret = cma_translate_addr(cma_src_addr(id_priv),
1900                                                  &id->route.addr.dev_addr);
1901                         if (ret)
1902                                 goto err;
1903                 }
1904         }
1905
1906         id_priv->state = RDMA_CM_CONNECT;
1907         return id_priv;
1908 err:
1909         rdma_destroy_id(id);
1910         return NULL;
1911 }
1912
1913 static void cma_set_req_event_data(struct rdma_cm_event *event,
1914                                    struct ib_cm_req_event_param *req_data,
1915                                    void *private_data, int offset)
1916 {
1917         event->param.conn.private_data = private_data + offset;
1918         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1919         event->param.conn.responder_resources = req_data->responder_resources;
1920         event->param.conn.initiator_depth = req_data->initiator_depth;
1921         event->param.conn.flow_control = req_data->flow_control;
1922         event->param.conn.retry_count = req_data->retry_count;
1923         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1924         event->param.conn.srq = req_data->srq;
1925         event->param.conn.qp_num = req_data->remote_qpn;
1926 }
1927
1928 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1929 {
1930         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1931                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1932                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1933                  (id->qp_type == IB_QPT_UD)) ||
1934                 (!id->qp_type));
1935 }
1936
1937 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1938 {
1939         struct rdma_id_private *listen_id, *conn_id = NULL;
1940         struct rdma_cm_event event;
1941         struct net_device *net_dev;
1942         int offset, ret;
1943
1944         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1945         if (IS_ERR(listen_id))
1946                 return PTR_ERR(listen_id);
1947
1948         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1949                 ret = -EINVAL;
1950                 goto net_dev_put;
1951         }
1952
1953         mutex_lock(&listen_id->handler_mutex);
1954         if (listen_id->state != RDMA_CM_LISTEN) {
1955                 ret = -ECONNABORTED;
1956                 goto err1;
1957         }
1958
1959         memset(&event, 0, sizeof event);
1960         offset = cma_user_data_offset(listen_id);
1961         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1962         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1963                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1964                 event.param.ud.private_data = ib_event->private_data + offset;
1965                 event.param.ud.private_data_len =
1966                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1967         } else {
1968                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1969                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1970                                        ib_event->private_data, offset);
1971         }
1972         if (!conn_id) {
1973                 ret = -ENOMEM;
1974                 goto err1;
1975         }
1976
1977         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1978         ret = cma_acquire_dev(conn_id, listen_id);
1979         if (ret)
1980                 goto err2;
1981
1982         conn_id->cm_id.ib = cm_id;
1983         cm_id->context = conn_id;
1984         cm_id->cm_handler = cma_ib_handler;
1985
1986         /*
1987          * Protect against the user destroying conn_id from another thread
1988          * until we're done accessing it.
1989          */
1990         atomic_inc(&conn_id->refcount);
1991         ret = conn_id->id.event_handler(&conn_id->id, &event);
1992         if (ret)
1993                 goto err3;
1994         /*
1995          * Acquire mutex to prevent user executing rdma_destroy_id()
1996          * while we're accessing the cm_id.
1997          */
1998         mutex_lock(&lock);
1999         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2000             (conn_id->id.qp_type != IB_QPT_UD))
2001                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2002         mutex_unlock(&lock);
2003         mutex_unlock(&conn_id->handler_mutex);
2004         mutex_unlock(&listen_id->handler_mutex);
2005         cma_deref_id(conn_id);
2006         if (net_dev)
2007                 dev_put(net_dev);
2008         return 0;
2009
2010 err3:
2011         cma_deref_id(conn_id);
2012         /* Destroy the CM ID by returning a non-zero value. */
2013         conn_id->cm_id.ib = NULL;
2014 err2:
2015         cma_exch(conn_id, RDMA_CM_DESTROYING);
2016         mutex_unlock(&conn_id->handler_mutex);
2017 err1:
2018         mutex_unlock(&listen_id->handler_mutex);
2019         if (conn_id)
2020                 rdma_destroy_id(&conn_id->id);
2021
2022 net_dev_put:
2023         if (net_dev)
2024                 dev_put(net_dev);
2025
2026         return ret;
2027 }
2028
2029 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2030 {
2031         if (addr->sa_family == AF_IB)
2032                 return ((struct sockaddr_ib *) addr)->sib_sid;
2033
2034         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2035 }
2036 EXPORT_SYMBOL(rdma_get_service_id);
2037
2038 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2039 {
2040         struct rdma_id_private *id_priv = iw_id->context;
2041         struct rdma_cm_event event;
2042         int ret = 0;
2043         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2044         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2045
2046         mutex_lock(&id_priv->handler_mutex);
2047         if (id_priv->state != RDMA_CM_CONNECT)
2048                 goto out;
2049
2050         memset(&event, 0, sizeof event);
2051         switch (iw_event->event) {
2052         case IW_CM_EVENT_CLOSE:
2053                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2054                 break;
2055         case IW_CM_EVENT_CONNECT_REPLY:
2056                 memcpy(cma_src_addr(id_priv), laddr,
2057                        rdma_addr_size(laddr));
2058                 memcpy(cma_dst_addr(id_priv), raddr,
2059                        rdma_addr_size(raddr));
2060                 switch (iw_event->status) {
2061                 case 0:
2062                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2063                         event.param.conn.initiator_depth = iw_event->ird;
2064                         event.param.conn.responder_resources = iw_event->ord;
2065                         break;
2066                 case -ECONNRESET:
2067                 case -ECONNREFUSED:
2068                         event.event = RDMA_CM_EVENT_REJECTED;
2069                         break;
2070                 case -ETIMEDOUT:
2071                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2072                         break;
2073                 default:
2074                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2075                         break;
2076                 }
2077                 break;
2078         case IW_CM_EVENT_ESTABLISHED:
2079                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2080                 event.param.conn.initiator_depth = iw_event->ird;
2081                 event.param.conn.responder_resources = iw_event->ord;
2082                 break;
2083         default:
2084                 BUG_ON(1);
2085         }
2086
2087         event.status = iw_event->status;
2088         event.param.conn.private_data = iw_event->private_data;
2089         event.param.conn.private_data_len = iw_event->private_data_len;
2090         ret = id_priv->id.event_handler(&id_priv->id, &event);
2091         if (ret) {
2092                 /* Destroy the CM ID by returning a non-zero value. */
2093                 id_priv->cm_id.iw = NULL;
2094                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2095                 mutex_unlock(&id_priv->handler_mutex);
2096                 rdma_destroy_id(&id_priv->id);
2097                 return ret;
2098         }
2099
2100 out:
2101         mutex_unlock(&id_priv->handler_mutex);
2102         return ret;
2103 }
2104
2105 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2106                                struct iw_cm_event *iw_event)
2107 {
2108         struct rdma_cm_id *new_cm_id;
2109         struct rdma_id_private *listen_id, *conn_id;
2110         struct rdma_cm_event event;
2111         int ret = -ECONNABORTED;
2112         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2113         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2114
2115         listen_id = cm_id->context;
2116
2117         mutex_lock(&listen_id->handler_mutex);
2118         if (listen_id->state != RDMA_CM_LISTEN)
2119                 goto out;
2120
2121         /* Create a new RDMA id for the new IW CM ID */
2122         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2123                                    listen_id->id.event_handler,
2124                                    listen_id->id.context,
2125                                    RDMA_PS_TCP, IB_QPT_RC);
2126         if (IS_ERR(new_cm_id)) {
2127                 ret = -ENOMEM;
2128                 goto out;
2129         }
2130         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2131         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2132         conn_id->state = RDMA_CM_CONNECT;
2133
2134         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2135         if (ret) {
2136                 mutex_unlock(&conn_id->handler_mutex);
2137                 rdma_destroy_id(new_cm_id);
2138                 goto out;
2139         }
2140
2141         ret = cma_acquire_dev(conn_id, listen_id);
2142         if (ret) {
2143                 mutex_unlock(&conn_id->handler_mutex);
2144                 rdma_destroy_id(new_cm_id);
2145                 goto out;
2146         }
2147
2148         conn_id->cm_id.iw = cm_id;
2149         cm_id->context = conn_id;
2150         cm_id->cm_handler = cma_iw_handler;
2151
2152         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2153         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2154
2155         memset(&event, 0, sizeof event);
2156         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2157         event.param.conn.private_data = iw_event->private_data;
2158         event.param.conn.private_data_len = iw_event->private_data_len;
2159         event.param.conn.initiator_depth = iw_event->ird;
2160         event.param.conn.responder_resources = iw_event->ord;
2161
2162         /*
2163          * Protect against the user destroying conn_id from another thread
2164          * until we're done accessing it.
2165          */
2166         atomic_inc(&conn_id->refcount);
2167         ret = conn_id->id.event_handler(&conn_id->id, &event);
2168         if (ret) {
2169                 /* User wants to destroy the CM ID */
2170                 conn_id->cm_id.iw = NULL;
2171                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2172                 mutex_unlock(&conn_id->handler_mutex);
2173                 cma_deref_id(conn_id);
2174                 rdma_destroy_id(&conn_id->id);
2175                 goto out;
2176         }
2177
2178         mutex_unlock(&conn_id->handler_mutex);
2179         cma_deref_id(conn_id);
2180
2181 out:
2182         mutex_unlock(&listen_id->handler_mutex);
2183         return ret;
2184 }
2185
2186 static int cma_ib_listen(struct rdma_id_private *id_priv)
2187 {
2188         struct sockaddr *addr;
2189         struct ib_cm_id *id;
2190         __be64 svc_id;
2191
2192         addr = cma_src_addr(id_priv);
2193         svc_id = rdma_get_service_id(&id_priv->id, addr);
2194         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2195         if (IS_ERR(id))
2196                 return PTR_ERR(id);
2197         id_priv->cm_id.ib = id;
2198
2199         return 0;
2200 }
2201
2202 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2203 {
2204         int ret;
2205         struct iw_cm_id *id;
2206
2207         id = iw_create_cm_id(id_priv->id.device,
2208                              iw_conn_req_handler,
2209                              id_priv);
2210         if (IS_ERR(id))
2211                 return PTR_ERR(id);
2212
2213         id->tos = id_priv->tos;
2214         id_priv->cm_id.iw = id;
2215
2216         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2217                rdma_addr_size(cma_src_addr(id_priv)));
2218
2219         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2220
2221         if (ret) {
2222                 iw_destroy_cm_id(id_priv->cm_id.iw);
2223                 id_priv->cm_id.iw = NULL;
2224         }
2225
2226         return ret;
2227 }
2228
2229 static int cma_listen_handler(struct rdma_cm_id *id,
2230                               struct rdma_cm_event *event)
2231 {
2232         struct rdma_id_private *id_priv = id->context;
2233
2234         id->context = id_priv->id.context;
2235         id->event_handler = id_priv->id.event_handler;
2236         return id_priv->id.event_handler(id, event);
2237 }
2238
2239 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2240                               struct cma_device *cma_dev)
2241 {
2242         struct rdma_id_private *dev_id_priv;
2243         struct rdma_cm_id *id;
2244         struct net *net = id_priv->id.route.addr.dev_addr.net;
2245         int ret;
2246
2247         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2248                 return;
2249
2250         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2251                             id_priv->id.qp_type);
2252         if (IS_ERR(id))
2253                 return;
2254
2255         dev_id_priv = container_of(id, struct rdma_id_private, id);
2256
2257         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2258         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2259                rdma_addr_size(cma_src_addr(id_priv)));
2260
2261         _cma_attach_to_dev(dev_id_priv, cma_dev);
2262         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2263         atomic_inc(&id_priv->refcount);
2264         dev_id_priv->internal_id = 1;
2265         dev_id_priv->afonly = id_priv->afonly;
2266
2267         ret = rdma_listen(id, id_priv->backlog);
2268         if (ret)
2269                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2270                         ret, cma_dev->device->name);
2271 }
2272
2273 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2274 {
2275         struct cma_device *cma_dev;
2276
2277         mutex_lock(&lock);
2278         list_add_tail(&id_priv->list, &listen_any_list);
2279         list_for_each_entry(cma_dev, &dev_list, list)
2280                 cma_listen_on_dev(id_priv, cma_dev);
2281         mutex_unlock(&lock);
2282 }
2283
2284 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2285 {
2286         struct rdma_id_private *id_priv;
2287
2288         id_priv = container_of(id, struct rdma_id_private, id);
2289         id_priv->tos = (u8) tos;
2290         id_priv->tos_set = true;
2291 }
2292 EXPORT_SYMBOL(rdma_set_service_type);
2293
2294 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2295                               void *context)
2296 {
2297         struct cma_work *work = context;
2298         struct rdma_route *route;
2299
2300         route = &work->id->id.route;
2301
2302         if (!status) {
2303                 route->num_paths = 1;
2304                 *route->path_rec = *path_rec;
2305         } else {
2306                 work->old_state = RDMA_CM_ROUTE_QUERY;
2307                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2308                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2309                 work->event.status = status;
2310                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2311                                      status);
2312         }
2313
2314         queue_work(cma_wq, &work->work);
2315 }
2316
2317 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2318                               struct cma_work *work)
2319 {
2320         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2321         struct sa_path_rec path_rec;
2322         ib_sa_comp_mask comp_mask;
2323         struct sockaddr_in6 *sin6;
2324         struct sockaddr_ib *sib;
2325
2326         memset(&path_rec, 0, sizeof path_rec);
2327
2328         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2329                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2330         else
2331                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2332         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2333         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2334         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2335         path_rec.numb_path = 1;
2336         path_rec.reversible = 1;
2337         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2338                                                   cma_dst_addr(id_priv));
2339
2340         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2341                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2342                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2343
2344         switch (cma_family(id_priv)) {
2345         case AF_INET:
2346                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2347                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2348                 break;
2349         case AF_INET6:
2350                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2351                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2352                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2353                 break;
2354         case AF_IB:
2355                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2356                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2357                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2358                 break;
2359         }
2360
2361         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2362                                                id_priv->id.port_num, &path_rec,
2363                                                comp_mask, timeout_ms,
2364                                                GFP_KERNEL, cma_query_handler,
2365                                                work, &id_priv->query);
2366
2367         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2368 }
2369
2370 static void cma_work_handler(struct work_struct *_work)
2371 {
2372         struct cma_work *work = container_of(_work, struct cma_work, work);
2373         struct rdma_id_private *id_priv = work->id;
2374         int destroy = 0;
2375
2376         mutex_lock(&id_priv->handler_mutex);
2377         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2378                 goto out;
2379
2380         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2381                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2382                 destroy = 1;
2383         }
2384 out:
2385         mutex_unlock(&id_priv->handler_mutex);
2386         cma_deref_id(id_priv);
2387         if (destroy)
2388                 rdma_destroy_id(&id_priv->id);
2389         kfree(work);
2390 }
2391
2392 static void cma_ndev_work_handler(struct work_struct *_work)
2393 {
2394         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2395         struct rdma_id_private *id_priv = work->id;
2396         int destroy = 0;
2397
2398         mutex_lock(&id_priv->handler_mutex);
2399         if (id_priv->state == RDMA_CM_DESTROYING ||
2400             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2401                 goto out;
2402
2403         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2404                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2405                 destroy = 1;
2406         }
2407
2408 out:
2409         mutex_unlock(&id_priv->handler_mutex);
2410         cma_deref_id(id_priv);
2411         if (destroy)
2412                 rdma_destroy_id(&id_priv->id);
2413         kfree(work);
2414 }
2415
2416 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2417 {
2418         struct rdma_route *route = &id_priv->id.route;
2419         struct cma_work *work;
2420         int ret;
2421
2422         work = kzalloc(sizeof *work, GFP_KERNEL);
2423         if (!work)
2424                 return -ENOMEM;
2425
2426         work->id = id_priv;
2427         INIT_WORK(&work->work, cma_work_handler);
2428         work->old_state = RDMA_CM_ROUTE_QUERY;
2429         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2430         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2431
2432         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2433         if (!route->path_rec) {
2434                 ret = -ENOMEM;
2435                 goto err1;
2436         }
2437
2438         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2439         if (ret)
2440                 goto err2;
2441
2442         return 0;
2443 err2:
2444         kfree(route->path_rec);
2445         route->path_rec = NULL;
2446 err1:
2447         kfree(work);
2448         return ret;
2449 }
2450
2451 int rdma_set_ib_paths(struct rdma_cm_id *id,
2452                       struct sa_path_rec *path_rec, int num_paths)
2453 {
2454         struct rdma_id_private *id_priv;
2455         int ret;
2456
2457         id_priv = container_of(id, struct rdma_id_private, id);
2458         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2459                            RDMA_CM_ROUTE_RESOLVED))
2460                 return -EINVAL;
2461
2462         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2463                                      GFP_KERNEL);
2464         if (!id->route.path_rec) {
2465                 ret = -ENOMEM;
2466                 goto err;
2467         }
2468
2469         id->route.num_paths = num_paths;
2470         return 0;
2471 err:
2472         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2473         return ret;
2474 }
2475 EXPORT_SYMBOL(rdma_set_ib_paths);
2476
2477 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2478 {
2479         struct cma_work *work;
2480
2481         work = kzalloc(sizeof *work, GFP_KERNEL);
2482         if (!work)
2483                 return -ENOMEM;
2484
2485         work->id = id_priv;
2486         INIT_WORK(&work->work, cma_work_handler);
2487         work->old_state = RDMA_CM_ROUTE_QUERY;
2488         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2489         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2490         queue_work(cma_wq, &work->work);
2491         return 0;
2492 }
2493
2494 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2495 {
2496         int prio;
2497         struct net_device *dev;
2498
2499         prio = rt_tos2priority(tos);
2500         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2501         if (dev->num_tc)
2502                 return netdev_get_prio_tc_map(dev, prio);
2503
2504 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2505         if (is_vlan_dev(ndev))
2506                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2507                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2508 #endif
2509         return 0;
2510 }
2511
2512 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2513                                            unsigned long supported_gids,
2514                                            enum ib_gid_type default_gid)
2515 {
2516         if ((network_type == RDMA_NETWORK_IPV4 ||
2517              network_type == RDMA_NETWORK_IPV6) &&
2518             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2519                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2520
2521         return default_gid;
2522 }
2523
2524 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2525 {
2526         struct rdma_route *route = &id_priv->id.route;
2527         struct rdma_addr *addr = &route->addr;
2528         struct cma_work *work;
2529         int ret;
2530         struct net_device *ndev = NULL;
2531         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
2532         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2533                                         rdma_start_port(id_priv->cma_dev->device)];
2534         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2535
2536
2537         work = kzalloc(sizeof *work, GFP_KERNEL);
2538         if (!work)
2539                 return -ENOMEM;
2540
2541         work->id = id_priv;
2542         INIT_WORK(&work->work, cma_work_handler);
2543
2544         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2545         if (!route->path_rec) {
2546                 ret = -ENOMEM;
2547                 goto err1;
2548         }
2549
2550         route->num_paths = 1;
2551
2552         if (addr->dev_addr.bound_dev_if) {
2553                 unsigned long supported_gids;
2554
2555                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2556                 if (!ndev) {
2557                         ret = -ENODEV;
2558                         goto err2;
2559                 }
2560
2561                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2562                                                             id_priv->id.port_num);
2563                 gid_type = cma_route_gid_type(addr->dev_addr.network,
2564                                               supported_gids,
2565                                               id_priv->gid_type);
2566                 route->path_rec->rec_type =
2567                         sa_conv_gid_to_pathrec_type(gid_type);
2568                 sa_path_set_ndev(route->path_rec, &init_net);
2569                 sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2570         }
2571         if (!ndev) {
2572                 ret = -ENODEV;
2573                 goto err2;
2574         }
2575
2576         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2577
2578         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2579                     &route->path_rec->sgid);
2580         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2581                     &route->path_rec->dgid);
2582
2583         /* Use the hint from IP Stack to select GID Type */
2584         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2585                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2586         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2587
2588         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2589                 /* TODO: get the hoplimit from the inet/inet6 device */
2590                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2591         else
2592                 route->path_rec->hop_limit = 1;
2593         route->path_rec->reversible = 1;
2594         route->path_rec->pkey = cpu_to_be16(0xffff);
2595         route->path_rec->mtu_selector = IB_SA_EQ;
2596         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2597         route->path_rec->traffic_class = tos;
2598         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2599         route->path_rec->rate_selector = IB_SA_EQ;
2600         route->path_rec->rate = iboe_get_rate(ndev);
2601         dev_put(ndev);
2602         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2603         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2604         if (!route->path_rec->mtu) {
2605                 ret = -EINVAL;
2606                 goto err2;
2607         }
2608
2609         work->old_state = RDMA_CM_ROUTE_QUERY;
2610         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2611         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2612         work->event.status = 0;
2613
2614         queue_work(cma_wq, &work->work);
2615
2616         return 0;
2617
2618 err2:
2619         kfree(route->path_rec);
2620         route->path_rec = NULL;
2621 err1:
2622         kfree(work);
2623         return ret;
2624 }
2625
2626 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2627 {
2628         struct rdma_id_private *id_priv;
2629         int ret;
2630
2631         id_priv = container_of(id, struct rdma_id_private, id);
2632         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2633                 return -EINVAL;
2634
2635         atomic_inc(&id_priv->refcount);
2636         if (rdma_cap_ib_sa(id->device, id->port_num))
2637                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2638         else if (rdma_protocol_roce(id->device, id->port_num))
2639                 ret = cma_resolve_iboe_route(id_priv);
2640         else if (rdma_protocol_iwarp(id->device, id->port_num))
2641                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2642         else
2643                 ret = -ENOSYS;
2644
2645         if (ret)
2646                 goto err;
2647
2648         return 0;
2649 err:
2650         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2651         cma_deref_id(id_priv);
2652         return ret;
2653 }
2654 EXPORT_SYMBOL(rdma_resolve_route);
2655
2656 static void cma_set_loopback(struct sockaddr *addr)
2657 {
2658         switch (addr->sa_family) {
2659         case AF_INET:
2660                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2661                 break;
2662         case AF_INET6:
2663                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2664                               0, 0, 0, htonl(1));
2665                 break;
2666         default:
2667                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2668                             0, 0, 0, htonl(1));
2669                 break;
2670         }
2671 }
2672
2673 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2674 {
2675         struct cma_device *cma_dev, *cur_dev;
2676         union ib_gid gid;
2677         enum ib_port_state port_state;
2678         u16 pkey;
2679         int ret;
2680         u8 p;
2681
2682         cma_dev = NULL;
2683         mutex_lock(&lock);
2684         list_for_each_entry(cur_dev, &dev_list, list) {
2685                 if (cma_family(id_priv) == AF_IB &&
2686                     !rdma_cap_ib_cm(cur_dev->device, 1))
2687                         continue;
2688
2689                 if (!cma_dev)
2690                         cma_dev = cur_dev;
2691
2692                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2693                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2694                             port_state == IB_PORT_ACTIVE) {
2695                                 cma_dev = cur_dev;
2696                                 goto port_found;
2697                         }
2698                 }
2699         }
2700
2701         if (!cma_dev) {
2702                 ret = -ENODEV;
2703                 goto out;
2704         }
2705
2706         p = 1;
2707
2708 port_found:
2709         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2710         if (ret)
2711                 goto out;
2712
2713         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2714         if (ret)
2715                 goto out;
2716
2717         id_priv->id.route.addr.dev_addr.dev_type =
2718                 (rdma_protocol_ib(cma_dev->device, p)) ?
2719                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2720
2721         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2722         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2723         id_priv->id.port_num = p;
2724         cma_attach_to_dev(id_priv, cma_dev);
2725         cma_set_loopback(cma_src_addr(id_priv));
2726 out:
2727         mutex_unlock(&lock);
2728         return ret;
2729 }
2730
2731 static void addr_handler(int status, struct sockaddr *src_addr,
2732                          struct rdma_dev_addr *dev_addr, void *context)
2733 {
2734         struct rdma_id_private *id_priv = context;
2735         struct rdma_cm_event event;
2736
2737         memset(&event, 0, sizeof event);
2738         mutex_lock(&id_priv->handler_mutex);
2739         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2740                            RDMA_CM_ADDR_RESOLVED))
2741                 goto out;
2742
2743         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2744         if (!status && !id_priv->cma_dev) {
2745                 status = cma_acquire_dev(id_priv, NULL);
2746                 if (status)
2747                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2748                                              status);
2749         } else {
2750                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2751         }
2752
2753         if (status) {
2754                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2755                                    RDMA_CM_ADDR_BOUND))
2756                         goto out;
2757                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2758                 event.status = status;
2759         } else
2760                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2761
2762         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2763                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2764                 mutex_unlock(&id_priv->handler_mutex);
2765                 cma_deref_id(id_priv);
2766                 rdma_destroy_id(&id_priv->id);
2767                 return;
2768         }
2769 out:
2770         mutex_unlock(&id_priv->handler_mutex);
2771         cma_deref_id(id_priv);
2772 }
2773
2774 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2775 {
2776         struct cma_work *work;
2777         union ib_gid gid;
2778         int ret;
2779
2780         work = kzalloc(sizeof *work, GFP_KERNEL);
2781         if (!work)
2782                 return -ENOMEM;
2783
2784         if (!id_priv->cma_dev) {
2785                 ret = cma_bind_loopback(id_priv);
2786                 if (ret)
2787                         goto err;
2788         }
2789
2790         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2791         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2792
2793         work->id = id_priv;
2794         INIT_WORK(&work->work, cma_work_handler);
2795         work->old_state = RDMA_CM_ADDR_QUERY;
2796         work->new_state = RDMA_CM_ADDR_RESOLVED;
2797         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2798         queue_work(cma_wq, &work->work);
2799         return 0;
2800 err:
2801         kfree(work);
2802         return ret;
2803 }
2804
2805 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2806 {
2807         struct cma_work *work;
2808         int ret;
2809
2810         work = kzalloc(sizeof *work, GFP_KERNEL);
2811         if (!work)
2812                 return -ENOMEM;
2813
2814         if (!id_priv->cma_dev) {
2815                 ret = cma_resolve_ib_dev(id_priv);
2816                 if (ret)
2817                         goto err;
2818         }
2819
2820         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2821                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2822
2823         work->id = id_priv;
2824         INIT_WORK(&work->work, cma_work_handler);
2825         work->old_state = RDMA_CM_ADDR_QUERY;
2826         work->new_state = RDMA_CM_ADDR_RESOLVED;
2827         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2828         queue_work(cma_wq, &work->work);
2829         return 0;
2830 err:
2831         kfree(work);
2832         return ret;
2833 }
2834
2835 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2836                          struct sockaddr *dst_addr)
2837 {
2838         if (!src_addr || !src_addr->sa_family) {
2839                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2840                 src_addr->sa_family = dst_addr->sa_family;
2841                 if (IS_ENABLED(CONFIG_IPV6) &&
2842                     dst_addr->sa_family == AF_INET6) {
2843                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2844                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2845                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2846                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2847                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2848                 } else if (dst_addr->sa_family == AF_IB) {
2849                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2850                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2851                 }
2852         }
2853         return rdma_bind_addr(id, src_addr);
2854 }
2855
2856 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2857                       struct sockaddr *dst_addr, int timeout_ms)
2858 {
2859         struct rdma_id_private *id_priv;
2860         int ret;
2861
2862         id_priv = container_of(id, struct rdma_id_private, id);
2863         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2864         if (id_priv->state == RDMA_CM_IDLE) {
2865                 ret = cma_bind_addr(id, src_addr, dst_addr);
2866                 if (ret) {
2867                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2868                         return ret;
2869                 }
2870         }
2871
2872         if (cma_family(id_priv) != dst_addr->sa_family) {
2873                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2874                 return -EINVAL;
2875         }
2876
2877         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2878                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2879                 return -EINVAL;
2880         }
2881
2882         atomic_inc(&id_priv->refcount);
2883         if (cma_any_addr(dst_addr)) {
2884                 ret = cma_resolve_loopback(id_priv);
2885         } else {
2886                 if (dst_addr->sa_family == AF_IB) {
2887                         ret = cma_resolve_ib_addr(id_priv);
2888                 } else {
2889                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2890                                               dst_addr, &id->route.addr.dev_addr,
2891                                               timeout_ms, addr_handler, id_priv);
2892                 }
2893         }
2894         if (ret)
2895                 goto err;
2896
2897         return 0;
2898 err:
2899         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2900         cma_deref_id(id_priv);
2901         return ret;
2902 }
2903 EXPORT_SYMBOL(rdma_resolve_addr);
2904
2905 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2906 {
2907         struct rdma_id_private *id_priv;
2908         unsigned long flags;
2909         int ret;
2910
2911         id_priv = container_of(id, struct rdma_id_private, id);
2912         spin_lock_irqsave(&id_priv->lock, flags);
2913         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2914                 id_priv->reuseaddr = reuse;
2915                 ret = 0;
2916         } else {
2917                 ret = -EINVAL;
2918         }
2919         spin_unlock_irqrestore(&id_priv->lock, flags);
2920         return ret;
2921 }
2922 EXPORT_SYMBOL(rdma_set_reuseaddr);
2923
2924 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2925 {
2926         struct rdma_id_private *id_priv;
2927         unsigned long flags;
2928         int ret;
2929
2930         id_priv = container_of(id, struct rdma_id_private, id);
2931         spin_lock_irqsave(&id_priv->lock, flags);
2932         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2933                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2934                 id_priv->afonly = afonly;
2935                 ret = 0;
2936         } else {
2937                 ret = -EINVAL;
2938         }
2939         spin_unlock_irqrestore(&id_priv->lock, flags);
2940         return ret;
2941 }
2942 EXPORT_SYMBOL(rdma_set_afonly);
2943
2944 static void cma_bind_port(struct rdma_bind_list *bind_list,
2945                           struct rdma_id_private *id_priv)
2946 {
2947         struct sockaddr *addr;
2948         struct sockaddr_ib *sib;
2949         u64 sid, mask;
2950         __be16 port;
2951
2952         addr = cma_src_addr(id_priv);
2953         port = htons(bind_list->port);
2954
2955         switch (addr->sa_family) {
2956         case AF_INET:
2957                 ((struct sockaddr_in *) addr)->sin_port = port;
2958                 break;
2959         case AF_INET6:
2960                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2961                 break;
2962         case AF_IB:
2963                 sib = (struct sockaddr_ib *) addr;
2964                 sid = be64_to_cpu(sib->sib_sid);
2965                 mask = be64_to_cpu(sib->sib_sid_mask);
2966                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2967                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2968                 break;
2969         }
2970         id_priv->bind_list = bind_list;
2971         hlist_add_head(&id_priv->node, &bind_list->owners);
2972 }
2973
2974 static int cma_alloc_port(enum rdma_port_space ps,
2975                           struct rdma_id_private *id_priv, unsigned short snum)
2976 {
2977         struct rdma_bind_list *bind_list;
2978         int ret;
2979
2980         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2981         if (!bind_list)
2982                 return -ENOMEM;
2983
2984         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2985                            snum);
2986         if (ret < 0)
2987                 goto err;
2988
2989         bind_list->ps = ps;
2990         bind_list->port = (unsigned short)ret;
2991         cma_bind_port(bind_list, id_priv);
2992         return 0;
2993 err:
2994         kfree(bind_list);
2995         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
2996 }
2997
2998 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
2999                               struct rdma_id_private *id_priv)
3000 {
3001         struct rdma_id_private *cur_id;
3002         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3003         struct sockaddr  *saddr = cma_src_addr(id_priv);
3004         __be16 dport = cma_port(daddr);
3005
3006         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3007                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3008                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3009                 __be16 cur_dport = cma_port(cur_daddr);
3010
3011                 if (id_priv == cur_id)
3012                         continue;
3013
3014                 /* different dest port -> unique */
3015                 if (!cma_any_port(cur_daddr) &&
3016                     (dport != cur_dport))
3017                         continue;
3018
3019                 /* different src address -> unique */
3020                 if (!cma_any_addr(saddr) &&
3021                     !cma_any_addr(cur_saddr) &&
3022                     cma_addr_cmp(saddr, cur_saddr))
3023                         continue;
3024
3025                 /* different dst address -> unique */
3026                 if (!cma_any_addr(cur_daddr) &&
3027                     cma_addr_cmp(daddr, cur_daddr))
3028                         continue;
3029
3030                 return -EADDRNOTAVAIL;
3031         }
3032         return 0;
3033 }
3034
3035 static int cma_alloc_any_port(enum rdma_port_space ps,
3036                               struct rdma_id_private *id_priv)
3037 {
3038         static unsigned int last_used_port;
3039         int low, high, remaining;
3040         unsigned int rover;
3041         struct net *net = id_priv->id.route.addr.dev_addr.net;
3042
3043         inet_get_local_port_range(net, &low, &high);
3044         remaining = (high - low) + 1;
3045         rover = prandom_u32() % remaining + low;
3046 retry:
3047         if (last_used_port != rover) {
3048                 struct rdma_bind_list *bind_list;
3049                 int ret;
3050
3051                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3052
3053                 if (!bind_list) {
3054                         ret = cma_alloc_port(ps, id_priv, rover);
3055                 } else {
3056                         ret = cma_port_is_unique(bind_list, id_priv);
3057                         if (!ret)
3058                                 cma_bind_port(bind_list, id_priv);
3059                 }
3060                 /*
3061                  * Remember previously used port number in order to avoid
3062                  * re-using same port immediately after it is closed.
3063                  */
3064                 if (!ret)
3065                         last_used_port = rover;
3066                 if (ret != -EADDRNOTAVAIL)
3067                         return ret;
3068         }
3069         if (--remaining) {
3070                 rover++;
3071                 if ((rover < low) || (rover > high))
3072                         rover = low;
3073                 goto retry;
3074         }
3075         return -EADDRNOTAVAIL;
3076 }
3077
3078 /*
3079  * Check that the requested port is available.  This is called when trying to
3080  * bind to a specific port, or when trying to listen on a bound port.  In
3081  * the latter case, the provided id_priv may already be on the bind_list, but
3082  * we still need to check that it's okay to start listening.
3083  */
3084 static int cma_check_port(struct rdma_bind_list *bind_list,
3085                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3086 {
3087         struct rdma_id_private *cur_id;
3088         struct sockaddr *addr, *cur_addr;
3089
3090         addr = cma_src_addr(id_priv);
3091         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3092                 if (id_priv == cur_id)
3093                         continue;
3094
3095                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3096                     cur_id->reuseaddr)
3097                         continue;
3098
3099                 cur_addr = cma_src_addr(cur_id);
3100                 if (id_priv->afonly && cur_id->afonly &&
3101                     (addr->sa_family != cur_addr->sa_family))
3102                         continue;
3103
3104                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3105                         return -EADDRNOTAVAIL;
3106
3107                 if (!cma_addr_cmp(addr, cur_addr))
3108                         return -EADDRINUSE;
3109         }
3110         return 0;
3111 }
3112
3113 static int cma_use_port(enum rdma_port_space ps,
3114                         struct rdma_id_private *id_priv)
3115 {
3116         struct rdma_bind_list *bind_list;
3117         unsigned short snum;
3118         int ret;
3119
3120         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3121         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3122                 return -EACCES;
3123
3124         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3125         if (!bind_list) {
3126                 ret = cma_alloc_port(ps, id_priv, snum);
3127         } else {
3128                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3129                 if (!ret)
3130                         cma_bind_port(bind_list, id_priv);
3131         }
3132         return ret;
3133 }
3134
3135 static int cma_bind_listen(struct rdma_id_private *id_priv)
3136 {
3137         struct rdma_bind_list *bind_list = id_priv->bind_list;
3138         int ret = 0;
3139
3140         mutex_lock(&lock);
3141         if (bind_list->owners.first->next)
3142                 ret = cma_check_port(bind_list, id_priv, 0);
3143         mutex_unlock(&lock);
3144         return ret;
3145 }
3146
3147 static enum rdma_port_space cma_select_inet_ps(
3148                 struct rdma_id_private *id_priv)
3149 {
3150         switch (id_priv->id.ps) {
3151         case RDMA_PS_TCP:
3152         case RDMA_PS_UDP:
3153         case RDMA_PS_IPOIB:
3154         case RDMA_PS_IB:
3155                 return id_priv->id.ps;
3156         default:
3157
3158                 return 0;
3159         }
3160 }
3161
3162 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3163 {
3164         enum rdma_port_space ps = 0;
3165         struct sockaddr_ib *sib;
3166         u64 sid_ps, mask, sid;
3167
3168         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3169         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3170         sid = be64_to_cpu(sib->sib_sid) & mask;
3171
3172         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3173                 sid_ps = RDMA_IB_IP_PS_IB;
3174                 ps = RDMA_PS_IB;
3175         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3176                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3177                 sid_ps = RDMA_IB_IP_PS_TCP;
3178                 ps = RDMA_PS_TCP;
3179         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3180                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3181                 sid_ps = RDMA_IB_IP_PS_UDP;
3182                 ps = RDMA_PS_UDP;
3183         }
3184
3185         if (ps) {
3186                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3187                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3188                                                 be64_to_cpu(sib->sib_sid_mask));
3189         }
3190         return ps;
3191 }
3192
3193 static int cma_get_port(struct rdma_id_private *id_priv)
3194 {
3195         enum rdma_port_space ps;
3196         int ret;
3197
3198         if (cma_family(id_priv) != AF_IB)
3199                 ps = cma_select_inet_ps(id_priv);
3200         else
3201                 ps = cma_select_ib_ps(id_priv);
3202         if (!ps)
3203                 return -EPROTONOSUPPORT;
3204
3205         mutex_lock(&lock);
3206         if (cma_any_port(cma_src_addr(id_priv)))
3207                 ret = cma_alloc_any_port(ps, id_priv);
3208         else
3209                 ret = cma_use_port(ps, id_priv);
3210         mutex_unlock(&lock);
3211
3212         return ret;
3213 }
3214
3215 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3216                                struct sockaddr *addr)
3217 {
3218 #if IS_ENABLED(CONFIG_IPV6)
3219         struct sockaddr_in6 *sin6;
3220
3221         if (addr->sa_family != AF_INET6)
3222                 return 0;
3223
3224         sin6 = (struct sockaddr_in6 *) addr;
3225
3226         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3227                 return 0;
3228
3229         if (!sin6->sin6_scope_id)
3230                         return -EINVAL;
3231
3232         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3233 #endif
3234         return 0;
3235 }
3236
3237 int rdma_listen(struct rdma_cm_id *id, int backlog)
3238 {
3239         struct rdma_id_private *id_priv;
3240         int ret;
3241
3242         id_priv = container_of(id, struct rdma_id_private, id);
3243         if (id_priv->state == RDMA_CM_IDLE) {
3244                 id->route.addr.src_addr.ss_family = AF_INET;
3245                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3246                 if (ret)
3247                         return ret;
3248         }
3249
3250         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3251                 return -EINVAL;
3252
3253         if (id_priv->reuseaddr) {
3254                 ret = cma_bind_listen(id_priv);
3255                 if (ret)
3256                         goto err;
3257         }
3258
3259         id_priv->backlog = backlog;
3260         if (id->device) {
3261                 if (rdma_cap_ib_cm(id->device, 1)) {
3262                         ret = cma_ib_listen(id_priv);
3263                         if (ret)
3264                                 goto err;
3265                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3266                         ret = cma_iw_listen(id_priv, backlog);
3267                         if (ret)
3268                                 goto err;
3269                 } else {
3270                         ret = -ENOSYS;
3271                         goto err;
3272                 }
3273         } else
3274                 cma_listen_on_all(id_priv);
3275
3276         return 0;
3277 err:
3278         id_priv->backlog = 0;
3279         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3280         return ret;
3281 }
3282 EXPORT_SYMBOL(rdma_listen);
3283
3284 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3285 {
3286         struct rdma_id_private *id_priv;
3287         int ret;
3288         struct sockaddr  *daddr;
3289
3290         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3291             addr->sa_family != AF_IB)
3292                 return -EAFNOSUPPORT;
3293
3294         id_priv = container_of(id, struct rdma_id_private, id);
3295         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3296                 return -EINVAL;
3297
3298         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3299         if (ret)
3300                 goto err1;
3301
3302         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3303         if (!cma_any_addr(addr)) {
3304                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3305                 if (ret)
3306                         goto err1;
3307
3308                 ret = cma_acquire_dev(id_priv, NULL);
3309                 if (ret)
3310                         goto err1;
3311         }
3312
3313         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3314                 if (addr->sa_family == AF_INET)
3315                         id_priv->afonly = 1;
3316 #if IS_ENABLED(CONFIG_IPV6)
3317                 else if (addr->sa_family == AF_INET6) {
3318                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3319
3320                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3321                 }
3322 #endif
3323         }
3324         ret = cma_get_port(id_priv);
3325         if (ret)
3326                 goto err2;
3327
3328         daddr = cma_dst_addr(id_priv);
3329         daddr->sa_family = addr->sa_family;
3330
3331         return 0;
3332 err2:
3333         if (id_priv->cma_dev)
3334                 cma_release_dev(id_priv);
3335 err1:
3336         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3337         return ret;
3338 }
3339 EXPORT_SYMBOL(rdma_bind_addr);
3340
3341 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3342 {
3343         struct cma_hdr *cma_hdr;
3344
3345         cma_hdr = hdr;
3346         cma_hdr->cma_version = CMA_VERSION;
3347         if (cma_family(id_priv) == AF_INET) {
3348                 struct sockaddr_in *src4, *dst4;
3349
3350                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3351                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3352
3353                 cma_set_ip_ver(cma_hdr, 4);
3354                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3355                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3356                 cma_hdr->port = src4->sin_port;
3357         } else if (cma_family(id_priv) == AF_INET6) {
3358                 struct sockaddr_in6 *src6, *dst6;
3359
3360                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3361                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3362
3363                 cma_set_ip_ver(cma_hdr, 6);
3364                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3365                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3366                 cma_hdr->port = src6->sin6_port;
3367         }
3368         return 0;
3369 }
3370
3371 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3372                                 struct ib_cm_event *ib_event)
3373 {
3374         struct rdma_id_private *id_priv = cm_id->context;
3375         struct rdma_cm_event event;
3376         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3377         int ret = 0;
3378
3379         mutex_lock(&id_priv->handler_mutex);
3380         if (id_priv->state != RDMA_CM_CONNECT)
3381                 goto out;
3382
3383         memset(&event, 0, sizeof event);
3384         switch (ib_event->event) {
3385         case IB_CM_SIDR_REQ_ERROR:
3386                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3387                 event.status = -ETIMEDOUT;
3388                 break;
3389         case IB_CM_SIDR_REP_RECEIVED:
3390                 event.param.ud.private_data = ib_event->private_data;
3391                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3392                 if (rep->status != IB_SIDR_SUCCESS) {
3393                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3394                         event.status = ib_event->param.sidr_rep_rcvd.status;
3395                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3396                                              event.status);
3397                         break;
3398                 }
3399                 ret = cma_set_qkey(id_priv, rep->qkey);
3400                 if (ret) {
3401                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3402                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3403                         event.status = ret;
3404                         break;
3405                 }
3406                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3407                                      id_priv->id.route.path_rec,
3408                                      &event.param.ud.ah_attr);
3409                 event.param.ud.qp_num = rep->qpn;
3410                 event.param.ud.qkey = rep->qkey;
3411                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3412                 event.status = 0;
3413                 break;
3414         default:
3415                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3416                        ib_event->event);
3417                 goto out;
3418         }
3419
3420         ret = id_priv->id.event_handler(&id_priv->id, &event);
3421         if (ret) {
3422                 /* Destroy the CM ID by returning a non-zero value. */
3423                 id_priv->cm_id.ib = NULL;
3424                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3425                 mutex_unlock(&id_priv->handler_mutex);
3426                 rdma_destroy_id(&id_priv->id);
3427                 return ret;
3428         }
3429 out:
3430         mutex_unlock(&id_priv->handler_mutex);
3431         return ret;
3432 }
3433
3434 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3435                               struct rdma_conn_param *conn_param)
3436 {
3437         struct ib_cm_sidr_req_param req;
3438         struct ib_cm_id *id;
3439         void *private_data;
3440         int offset, ret;
3441
3442         memset(&req, 0, sizeof req);
3443         offset = cma_user_data_offset(id_priv);
3444         req.private_data_len = offset + conn_param->private_data_len;
3445         if (req.private_data_len < conn_param->private_data_len)
3446                 return -EINVAL;
3447
3448         if (req.private_data_len) {
3449                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3450                 if (!private_data)
3451                         return -ENOMEM;
3452         } else {
3453                 private_data = NULL;
3454         }
3455
3456         if (conn_param->private_data && conn_param->private_data_len)
3457                 memcpy(private_data + offset, conn_param->private_data,
3458                        conn_param->private_data_len);
3459
3460         if (private_data) {
3461                 ret = cma_format_hdr(private_data, id_priv);
3462                 if (ret)
3463                         goto out;
3464                 req.private_data = private_data;
3465         }
3466
3467         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3468                              id_priv);
3469         if (IS_ERR(id)) {
3470                 ret = PTR_ERR(id);
3471                 goto out;
3472         }
3473         id_priv->cm_id.ib = id;
3474
3475         req.path = id_priv->id.route.path_rec;
3476         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3477         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3478         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3479
3480         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3481         if (ret) {
3482                 ib_destroy_cm_id(id_priv->cm_id.ib);
3483                 id_priv->cm_id.ib = NULL;
3484         }
3485 out:
3486         kfree(private_data);
3487         return ret;
3488 }
3489
3490 static int cma_connect_ib(struct rdma_id_private *id_priv,
3491                           struct rdma_conn_param *conn_param)
3492 {
3493         struct ib_cm_req_param req;
3494         struct rdma_route *route;
3495         void *private_data;
3496         struct ib_cm_id *id;
3497         int offset, ret;
3498
3499         memset(&req, 0, sizeof req);
3500         offset = cma_user_data_offset(id_priv);
3501         req.private_data_len = offset + conn_param->private_data_len;
3502         if (req.private_data_len < conn_param->private_data_len)
3503                 return -EINVAL;
3504
3505         if (req.private_data_len) {
3506                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3507                 if (!private_data)
3508                         return -ENOMEM;
3509         } else {
3510                 private_data = NULL;
3511         }
3512
3513         if (conn_param->private_data && conn_param->private_data_len)
3514                 memcpy(private_data + offset, conn_param->private_data,
3515                        conn_param->private_data_len);
3516
3517         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3518         if (IS_ERR(id)) {
3519                 ret = PTR_ERR(id);
3520                 goto out;
3521         }
3522         id_priv->cm_id.ib = id;
3523
3524         route = &id_priv->id.route;
3525         if (private_data) {
3526                 ret = cma_format_hdr(private_data, id_priv);
3527                 if (ret)
3528                         goto out;
3529                 req.private_data = private_data;
3530         }
3531
3532         req.primary_path = &route->path_rec[0];
3533         if (route->num_paths == 2)
3534                 req.alternate_path = &route->path_rec[1];
3535
3536         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3537         req.qp_num = id_priv->qp_num;
3538         req.qp_type = id_priv->id.qp_type;
3539         req.starting_psn = id_priv->seq_num;
3540         req.responder_resources = conn_param->responder_resources;
3541         req.initiator_depth = conn_param->initiator_depth;
3542         req.flow_control = conn_param->flow_control;
3543         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3544         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3545         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3546         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3547         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3548         req.srq = id_priv->srq ? 1 : 0;
3549
3550         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3551 out:
3552         if (ret && !IS_ERR(id)) {
3553                 ib_destroy_cm_id(id);
3554                 id_priv->cm_id.ib = NULL;
3555         }
3556
3557         kfree(private_data);
3558         return ret;
3559 }
3560
3561 static int cma_connect_iw(struct rdma_id_private *id_priv,
3562                           struct rdma_conn_param *conn_param)
3563 {
3564         struct iw_cm_id *cm_id;
3565         int ret;
3566         struct iw_cm_conn_param iw_param;
3567
3568         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3569         if (IS_ERR(cm_id))
3570                 return PTR_ERR(cm_id);
3571
3572         cm_id->tos = id_priv->tos;
3573         id_priv->cm_id.iw = cm_id;
3574
3575         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3576                rdma_addr_size(cma_src_addr(id_priv)));
3577         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3578                rdma_addr_size(cma_dst_addr(id_priv)));
3579
3580         ret = cma_modify_qp_rtr(id_priv, conn_param);
3581         if (ret)
3582                 goto out;
3583
3584         if (conn_param) {
3585                 iw_param.ord = conn_param->initiator_depth;
3586                 iw_param.ird = conn_param->responder_resources;
3587                 iw_param.private_data = conn_param->private_data;
3588                 iw_param.private_data_len = conn_param->private_data_len;
3589                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3590         } else {
3591                 memset(&iw_param, 0, sizeof iw_param);
3592                 iw_param.qpn = id_priv->qp_num;
3593         }
3594         ret = iw_cm_connect(cm_id, &iw_param);
3595 out:
3596         if (ret) {
3597                 iw_destroy_cm_id(cm_id);
3598                 id_priv->cm_id.iw = NULL;
3599         }
3600         return ret;
3601 }
3602
3603 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3604 {
3605         struct rdma_id_private *id_priv;
3606         int ret;
3607
3608         id_priv = container_of(id, struct rdma_id_private, id);
3609         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3610                 return -EINVAL;
3611
3612         if (!id->qp) {
3613                 id_priv->qp_num = conn_param->qp_num;
3614                 id_priv->srq = conn_param->srq;
3615         }
3616
3617         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3618                 if (id->qp_type == IB_QPT_UD)
3619                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3620                 else
3621                         ret = cma_connect_ib(id_priv, conn_param);
3622         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3623                 ret = cma_connect_iw(id_priv, conn_param);
3624         else
3625                 ret = -ENOSYS;
3626         if (ret)
3627                 goto err;
3628
3629         return 0;
3630 err:
3631         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3632         return ret;
3633 }
3634 EXPORT_SYMBOL(rdma_connect);
3635
3636 static int cma_accept_ib(struct rdma_id_private *id_priv,
3637                          struct rdma_conn_param *conn_param)
3638 {
3639         struct ib_cm_rep_param rep;
3640         int ret;
3641
3642         ret = cma_modify_qp_rtr(id_priv, conn_param);
3643         if (ret)
3644                 goto out;
3645
3646         ret = cma_modify_qp_rts(id_priv, conn_param);
3647         if (ret)
3648                 goto out;
3649
3650         memset(&rep, 0, sizeof rep);
3651         rep.qp_num = id_priv->qp_num;
3652         rep.starting_psn = id_priv->seq_num;
3653         rep.private_data = conn_param->private_data;
3654         rep.private_data_len = conn_param->private_data_len;
3655         rep.responder_resources = conn_param->responder_resources;
3656         rep.initiator_depth = conn_param->initiator_depth;
3657         rep.failover_accepted = 0;
3658         rep.flow_control = conn_param->flow_control;
3659         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3660         rep.srq = id_priv->srq ? 1 : 0;
3661
3662         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3663 out:
3664         return ret;
3665 }
3666
3667 static int cma_accept_iw(struct rdma_id_private *id_priv,
3668                   struct rdma_conn_param *conn_param)
3669 {
3670         struct iw_cm_conn_param iw_param;
3671         int ret;
3672
3673         if (!conn_param)
3674                 return -EINVAL;
3675
3676         ret = cma_modify_qp_rtr(id_priv, conn_param);
3677         if (ret)
3678                 return ret;
3679
3680         iw_param.ord = conn_param->initiator_depth;
3681         iw_param.ird = conn_param->responder_resources;
3682         iw_param.private_data = conn_param->private_data;
3683         iw_param.private_data_len = conn_param->private_data_len;
3684         if (id_priv->id.qp) {
3685                 iw_param.qpn = id_priv->qp_num;
3686         } else
3687                 iw_param.qpn = conn_param->qp_num;
3688
3689         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3690 }
3691
3692 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3693                              enum ib_cm_sidr_status status, u32 qkey,
3694                              const void *private_data, int private_data_len)
3695 {
3696         struct ib_cm_sidr_rep_param rep;
3697         int ret;
3698
3699         memset(&rep, 0, sizeof rep);
3700         rep.status = status;
3701         if (status == IB_SIDR_SUCCESS) {
3702                 ret = cma_set_qkey(id_priv, qkey);
3703                 if (ret)
3704                         return ret;
3705                 rep.qp_num = id_priv->qp_num;
3706                 rep.qkey = id_priv->qkey;
3707         }
3708         rep.private_data = private_data;
3709         rep.private_data_len = private_data_len;
3710
3711         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3712 }
3713
3714 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3715 {
3716         struct rdma_id_private *id_priv;
3717         int ret;
3718
3719         id_priv = container_of(id, struct rdma_id_private, id);
3720
3721         id_priv->owner = task_pid_nr(current);
3722
3723         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3724                 return -EINVAL;
3725
3726         if (!id->qp && conn_param) {
3727                 id_priv->qp_num = conn_param->qp_num;
3728                 id_priv->srq = conn_param->srq;
3729         }
3730
3731         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3732                 if (id->qp_type == IB_QPT_UD) {
3733                         if (conn_param)
3734                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3735                                                         conn_param->qkey,
3736                                                         conn_param->private_data,
3737                                                         conn_param->private_data_len);
3738                         else
3739                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3740                                                         0, NULL, 0);
3741                 } else {
3742                         if (conn_param)
3743                                 ret = cma_accept_ib(id_priv, conn_param);
3744                         else
3745                                 ret = cma_rep_recv(id_priv);
3746                 }
3747         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3748                 ret = cma_accept_iw(id_priv, conn_param);
3749         else
3750                 ret = -ENOSYS;
3751
3752         if (ret)
3753                 goto reject;
3754
3755         return 0;
3756 reject:
3757         cma_modify_qp_err(id_priv);
3758         rdma_reject(id, NULL, 0);
3759         return ret;
3760 }
3761 EXPORT_SYMBOL(rdma_accept);
3762
3763 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3764 {
3765         struct rdma_id_private *id_priv;
3766         int ret;
3767
3768         id_priv = container_of(id, struct rdma_id_private, id);
3769         if (!id_priv->cm_id.ib)
3770                 return -EINVAL;
3771
3772         switch (id->device->node_type) {
3773         case RDMA_NODE_IB_CA:
3774                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3775                 break;
3776         default:
3777                 ret = 0;
3778                 break;
3779         }
3780         return ret;
3781 }
3782 EXPORT_SYMBOL(rdma_notify);
3783
3784 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3785                 u8 private_data_len)
3786 {
3787         struct rdma_id_private *id_priv;
3788         int ret;
3789
3790         id_priv = container_of(id, struct rdma_id_private, id);
3791         if (!id_priv->cm_id.ib)
3792                 return -EINVAL;
3793
3794         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3795                 if (id->qp_type == IB_QPT_UD)
3796                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3797                                                 private_data, private_data_len);
3798                 else
3799                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3800                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3801                                              0, private_data, private_data_len);
3802         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3803                 ret = iw_cm_reject(id_priv->cm_id.iw,
3804                                    private_data, private_data_len);
3805         } else
3806                 ret = -ENOSYS;
3807
3808         return ret;
3809 }
3810 EXPORT_SYMBOL(rdma_reject);
3811
3812 int rdma_disconnect(struct rdma_cm_id *id)
3813 {
3814         struct rdma_id_private *id_priv;
3815         int ret;
3816
3817         id_priv = container_of(id, struct rdma_id_private, id);
3818         if (!id_priv->cm_id.ib)
3819                 return -EINVAL;
3820
3821         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3822                 ret = cma_modify_qp_err(id_priv);
3823                 if (ret)
3824                         goto out;
3825                 /* Initiate or respond to a disconnect. */
3826                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3827                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3828         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3829                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3830         } else
3831                 ret = -EINVAL;
3832
3833 out:
3834         return ret;
3835 }
3836 EXPORT_SYMBOL(rdma_disconnect);
3837
3838 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3839 {
3840         struct rdma_id_private *id_priv;
3841         struct cma_multicast *mc = multicast->context;
3842         struct rdma_cm_event event;
3843         int ret = 0;
3844
3845         id_priv = mc->id_priv;
3846         mutex_lock(&id_priv->handler_mutex);
3847         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3848             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3849                 goto out;
3850
3851         if (!status)
3852                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3853         else
3854                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3855                                      status);
3856         mutex_lock(&id_priv->qp_mutex);
3857         if (!status && id_priv->id.qp) {
3858                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3859                                          be16_to_cpu(multicast->rec.mlid));
3860                 if (status)
3861                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3862                                              status);
3863         }
3864         mutex_unlock(&id_priv->qp_mutex);
3865
3866         memset(&event, 0, sizeof event);
3867         event.status = status;
3868         event.param.ud.private_data = mc->context;
3869         if (!status) {
3870                 struct rdma_dev_addr *dev_addr =
3871                         &id_priv->id.route.addr.dev_addr;
3872                 struct net_device *ndev =
3873                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3874                 enum ib_gid_type gid_type =
3875                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3876                         rdma_start_port(id_priv->cma_dev->device)];
3877
3878                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3879                 ib_init_ah_from_mcmember(id_priv->id.device,
3880                                          id_priv->id.port_num, &multicast->rec,
3881                                          ndev, gid_type,
3882                                          &event.param.ud.ah_attr);
3883                 event.param.ud.qp_num = 0xFFFFFF;
3884                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3885                 if (ndev)
3886                         dev_put(ndev);
3887         } else
3888                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3889
3890         ret = id_priv->id.event_handler(&id_priv->id, &event);
3891         if (ret) {
3892                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3893                 mutex_unlock(&id_priv->handler_mutex);
3894                 rdma_destroy_id(&id_priv->id);
3895                 return 0;
3896         }
3897
3898 out:
3899         mutex_unlock(&id_priv->handler_mutex);
3900         return 0;
3901 }
3902
3903 static void cma_set_mgid(struct rdma_id_private *id_priv,
3904                          struct sockaddr *addr, union ib_gid *mgid)
3905 {
3906         unsigned char mc_map[MAX_ADDR_LEN];
3907         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3908         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3909         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3910
3911         if (cma_any_addr(addr)) {
3912                 memset(mgid, 0, sizeof *mgid);
3913         } else if ((addr->sa_family == AF_INET6) &&
3914                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3915                                                                  0xFF10A01B)) {
3916                 /* IPv6 address is an SA assigned MGID. */
3917                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3918         } else if (addr->sa_family == AF_IB) {
3919                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3920         } else if ((addr->sa_family == AF_INET6)) {
3921                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3922                 if (id_priv->id.ps == RDMA_PS_UDP)
3923                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3924                 *mgid = *(union ib_gid *) (mc_map + 4);
3925         } else {
3926                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3927                 if (id_priv->id.ps == RDMA_PS_UDP)
3928                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3929                 *mgid = *(union ib_gid *) (mc_map + 4);
3930         }
3931 }
3932
3933 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3934                                  struct cma_multicast *mc)
3935 {
3936         struct ib_sa_mcmember_rec rec;
3937         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3938         ib_sa_comp_mask comp_mask;
3939         int ret;
3940
3941         ib_addr_get_mgid(dev_addr, &rec.mgid);
3942         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3943                                      &rec.mgid, &rec);
3944         if (ret)
3945                 return ret;
3946
3947         ret = cma_set_qkey(id_priv, 0);
3948         if (ret)
3949                 return ret;
3950
3951         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3952         rec.qkey = cpu_to_be32(id_priv->qkey);
3953         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3954         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3955         rec.join_state = mc->join_state;
3956
3957         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3958             (!ib_sa_sendonly_fullmem_support(&sa_client,
3959                                              id_priv->id.device,
3960                                              id_priv->id.port_num))) {
3961                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3962                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
3963                         id_priv->id.device->name, id_priv->id.port_num);
3964                 return -EOPNOTSUPP;
3965         }
3966
3967         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3968                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3969                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3970                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3971                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3972
3973         if (id_priv->id.ps == RDMA_PS_IPOIB)
3974                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
3975                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
3976                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
3977                              IB_SA_MCMEMBER_REC_MTU |
3978                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
3979
3980         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
3981                                                 id_priv->id.port_num, &rec,
3982                                                 comp_mask, GFP_KERNEL,
3983                                                 cma_ib_mc_handler, mc);
3984         return PTR_ERR_OR_ZERO(mc->multicast.ib);
3985 }
3986
3987 static void iboe_mcast_work_handler(struct work_struct *work)
3988 {
3989         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
3990         struct cma_multicast *mc = mw->mc;
3991         struct ib_sa_multicast *m = mc->multicast.ib;
3992
3993         mc->multicast.ib->context = mc;
3994         cma_ib_mc_handler(0, m);
3995         kref_put(&mc->mcref, release_mc);
3996         kfree(mw);
3997 }
3998
3999 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4000 {
4001         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4002         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4003
4004         if (cma_any_addr(addr)) {
4005                 memset(mgid, 0, sizeof *mgid);
4006         } else if (addr->sa_family == AF_INET6) {
4007                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4008         } else {
4009                 mgid->raw[0] = 0xff;
4010                 mgid->raw[1] = 0x0e;
4011                 mgid->raw[2] = 0;
4012                 mgid->raw[3] = 0;
4013                 mgid->raw[4] = 0;
4014                 mgid->raw[5] = 0;
4015                 mgid->raw[6] = 0;
4016                 mgid->raw[7] = 0;
4017                 mgid->raw[8] = 0;
4018                 mgid->raw[9] = 0;
4019                 mgid->raw[10] = 0xff;
4020                 mgid->raw[11] = 0xff;
4021                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4022         }
4023 }
4024
4025 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4026                                    struct cma_multicast *mc)
4027 {
4028         struct iboe_mcast_work *work;
4029         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4030         int err = 0;
4031         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4032         struct net_device *ndev = NULL;
4033         enum ib_gid_type gid_type;
4034         bool send_only;
4035
4036         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4037
4038         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4039                 return -EINVAL;
4040
4041         work = kzalloc(sizeof *work, GFP_KERNEL);
4042         if (!work)
4043                 return -ENOMEM;
4044
4045         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4046         if (!mc->multicast.ib) {
4047                 err = -ENOMEM;
4048                 goto out1;
4049         }
4050
4051         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4052
4053         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4054         if (id_priv->id.ps == RDMA_PS_UDP)
4055                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4056
4057         if (dev_addr->bound_dev_if)
4058                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4059         if (!ndev) {
4060                 err = -ENODEV;
4061                 goto out2;
4062         }
4063         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4064         mc->multicast.ib->rec.hop_limit = 1;
4065         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4066
4067         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4068                    rdma_start_port(id_priv->cma_dev->device)];
4069         if (addr->sa_family == AF_INET) {
4070                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4071                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4072                         if (!send_only) {
4073                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4074                                                     true);
4075                                 if (!err)
4076                                         mc->igmp_joined = true;
4077                         }
4078                 }
4079         } else {
4080                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4081                         err = -ENOTSUPP;
4082         }
4083         dev_put(ndev);
4084         if (err || !mc->multicast.ib->rec.mtu) {
4085                 if (!err)
4086                         err = -EINVAL;
4087                 goto out2;
4088         }
4089         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4090                     &mc->multicast.ib->rec.port_gid);
4091         work->id = id_priv;
4092         work->mc = mc;
4093         INIT_WORK(&work->work, iboe_mcast_work_handler);
4094         kref_get(&mc->mcref);
4095         queue_work(cma_wq, &work->work);
4096
4097         return 0;
4098
4099 out2:
4100         kfree(mc->multicast.ib);
4101 out1:
4102         kfree(work);
4103         return err;
4104 }
4105
4106 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4107                         u8 join_state, void *context)
4108 {
4109         struct rdma_id_private *id_priv;
4110         struct cma_multicast *mc;
4111         int ret;
4112
4113         id_priv = container_of(id, struct rdma_id_private, id);
4114         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4115             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4116                 return -EINVAL;
4117
4118         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4119         if (!mc)
4120                 return -ENOMEM;
4121
4122         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4123         mc->context = context;
4124         mc->id_priv = id_priv;
4125         mc->igmp_joined = false;
4126         mc->join_state = join_state;
4127         spin_lock(&id_priv->lock);
4128         list_add(&mc->list, &id_priv->mc_list);
4129         spin_unlock(&id_priv->lock);
4130
4131         if (rdma_protocol_roce(id->device, id->port_num)) {
4132                 kref_init(&mc->mcref);
4133                 ret = cma_iboe_join_multicast(id_priv, mc);
4134         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4135                 ret = cma_join_ib_multicast(id_priv, mc);
4136         else
4137                 ret = -ENOSYS;
4138
4139         if (ret) {
4140                 spin_lock_irq(&id_priv->lock);
4141                 list_del(&mc->list);
4142                 spin_unlock_irq(&id_priv->lock);
4143                 kfree(mc);
4144         }
4145         return ret;
4146 }
4147 EXPORT_SYMBOL(rdma_join_multicast);
4148
4149 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4150 {
4151         struct rdma_id_private *id_priv;
4152         struct cma_multicast *mc;
4153
4154         id_priv = container_of(id, struct rdma_id_private, id);
4155         spin_lock_irq(&id_priv->lock);
4156         list_for_each_entry(mc, &id_priv->mc_list, list) {
4157                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4158                         list_del(&mc->list);
4159                         spin_unlock_irq(&id_priv->lock);
4160
4161                         if (id->qp)
4162                                 ib_detach_mcast(id->qp,
4163                                                 &mc->multicast.ib->rec.mgid,
4164                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4165
4166                         BUG_ON(id_priv->cma_dev->device != id->device);
4167
4168                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4169                                 ib_sa_free_multicast(mc->multicast.ib);
4170                                 kfree(mc);
4171                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4172                                 if (mc->igmp_joined) {
4173                                         struct rdma_dev_addr *dev_addr =
4174                                                 &id->route.addr.dev_addr;
4175                                         struct net_device *ndev = NULL;
4176
4177                                         if (dev_addr->bound_dev_if)
4178                                                 ndev = dev_get_by_index(&init_net,
4179                                                                         dev_addr->bound_dev_if);
4180                                         if (ndev) {
4181                                                 cma_igmp_send(ndev,
4182                                                               &mc->multicast.ib->rec.mgid,
4183                                                               false);
4184                                                 dev_put(ndev);
4185                                         }
4186                                         mc->igmp_joined = false;
4187                                 }
4188                                 kref_put(&mc->mcref, release_mc);
4189                         }
4190                         return;
4191                 }
4192         }
4193         spin_unlock_irq(&id_priv->lock);
4194 }
4195 EXPORT_SYMBOL(rdma_leave_multicast);
4196
4197 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4198 {
4199         struct rdma_dev_addr *dev_addr;
4200         struct cma_ndev_work *work;
4201
4202         dev_addr = &id_priv->id.route.addr.dev_addr;
4203
4204         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4205             (net_eq(dev_net(ndev), dev_addr->net)) &&
4206             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4207                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4208                         ndev->name, &id_priv->id);
4209                 work = kzalloc(sizeof *work, GFP_KERNEL);
4210                 if (!work)
4211                         return -ENOMEM;
4212
4213                 INIT_WORK(&work->work, cma_ndev_work_handler);
4214                 work->id = id_priv;
4215                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4216                 atomic_inc(&id_priv->refcount);
4217                 queue_work(cma_wq, &work->work);
4218         }
4219
4220         return 0;
4221 }
4222
4223 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4224                                void *ptr)
4225 {
4226         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4227         struct cma_device *cma_dev;
4228         struct rdma_id_private *id_priv;
4229         int ret = NOTIFY_DONE;
4230
4231         if (event != NETDEV_BONDING_FAILOVER)
4232                 return NOTIFY_DONE;
4233
4234         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4235                 return NOTIFY_DONE;
4236
4237         mutex_lock(&lock);
4238         list_for_each_entry(cma_dev, &dev_list, list)
4239                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4240                         ret = cma_netdev_change(ndev, id_priv);
4241                         if (ret)
4242                                 goto out;
4243                 }
4244
4245 out:
4246         mutex_unlock(&lock);
4247         return ret;
4248 }
4249
4250 static struct notifier_block cma_nb = {
4251         .notifier_call = cma_netdev_callback
4252 };
4253
4254 static void cma_add_one(struct ib_device *device)
4255 {
4256         struct cma_device *cma_dev;
4257         struct rdma_id_private *id_priv;
4258         unsigned int i;
4259         unsigned long supported_gids = 0;
4260
4261         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4262         if (!cma_dev)
4263                 return;
4264
4265         cma_dev->device = device;
4266         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4267                                             sizeof(*cma_dev->default_gid_type),
4268                                             GFP_KERNEL);
4269         if (!cma_dev->default_gid_type)
4270                 goto free_cma_dev;
4271
4272         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4273                                             sizeof(*cma_dev->default_roce_tos),
4274                                             GFP_KERNEL);
4275         if (!cma_dev->default_roce_tos)
4276                 goto free_gid_type;
4277
4278         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4279                 supported_gids = roce_gid_type_mask_support(device, i);
4280                 WARN_ON(!supported_gids);
4281                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4282                         find_first_bit(&supported_gids, BITS_PER_LONG);
4283                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4284         }
4285
4286         init_completion(&cma_dev->comp);
4287         atomic_set(&cma_dev->refcount, 1);
4288         INIT_LIST_HEAD(&cma_dev->id_list);
4289         ib_set_client_data(device, &cma_client, cma_dev);
4290
4291         mutex_lock(&lock);
4292         list_add_tail(&cma_dev->list, &dev_list);
4293         list_for_each_entry(id_priv, &listen_any_list, list)
4294                 cma_listen_on_dev(id_priv, cma_dev);
4295         mutex_unlock(&lock);
4296
4297         return;
4298
4299 free_gid_type:
4300         kfree(cma_dev->default_gid_type);
4301
4302 free_cma_dev:
4303         kfree(cma_dev);
4304
4305         return;
4306 }
4307
4308 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4309 {
4310         struct rdma_cm_event event;
4311         enum rdma_cm_state state;
4312         int ret = 0;
4313
4314         /* Record that we want to remove the device */
4315         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4316         if (state == RDMA_CM_DESTROYING)
4317                 return 0;
4318
4319         cma_cancel_operation(id_priv, state);
4320         mutex_lock(&id_priv->handler_mutex);
4321
4322         /* Check for destruction from another callback. */
4323         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4324                 goto out;
4325
4326         memset(&event, 0, sizeof event);
4327         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4328         ret = id_priv->id.event_handler(&id_priv->id, &event);
4329 out:
4330         mutex_unlock(&id_priv->handler_mutex);
4331         return ret;
4332 }
4333
4334 static void cma_process_remove(struct cma_device *cma_dev)
4335 {
4336         struct rdma_id_private *id_priv;
4337         int ret;
4338
4339         mutex_lock(&lock);
4340         while (!list_empty(&cma_dev->id_list)) {
4341                 id_priv = list_entry(cma_dev->id_list.next,
4342                                      struct rdma_id_private, list);
4343
4344                 list_del(&id_priv->listen_list);
4345                 list_del_init(&id_priv->list);
4346                 atomic_inc(&id_priv->refcount);
4347                 mutex_unlock(&lock);
4348
4349                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4350                 cma_deref_id(id_priv);
4351                 if (ret)
4352                         rdma_destroy_id(&id_priv->id);
4353
4354                 mutex_lock(&lock);
4355         }
4356         mutex_unlock(&lock);
4357
4358         cma_deref_dev(cma_dev);
4359         wait_for_completion(&cma_dev->comp);
4360 }
4361
4362 static void cma_remove_one(struct ib_device *device, void *client_data)
4363 {
4364         struct cma_device *cma_dev = client_data;
4365
4366         if (!cma_dev)
4367                 return;
4368
4369         mutex_lock(&lock);
4370         list_del(&cma_dev->list);
4371         mutex_unlock(&lock);
4372
4373         cma_process_remove(cma_dev);
4374         kfree(cma_dev->default_roce_tos);
4375         kfree(cma_dev->default_gid_type);
4376         kfree(cma_dev);
4377 }
4378
4379 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4380 {
4381         struct nlmsghdr *nlh;
4382         struct rdma_cm_id_stats *id_stats;
4383         struct rdma_id_private *id_priv;
4384         struct rdma_cm_id *id = NULL;
4385         struct cma_device *cma_dev;
4386         int i_dev = 0, i_id = 0;
4387
4388         /*
4389          * We export all of the IDs as a sequence of messages.  Each
4390          * ID gets its own netlink message.
4391          */
4392         mutex_lock(&lock);
4393
4394         list_for_each_entry(cma_dev, &dev_list, list) {
4395                 if (i_dev < cb->args[0]) {
4396                         i_dev++;
4397                         continue;
4398                 }
4399
4400                 i_id = 0;
4401                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4402                         if (i_id < cb->args[1]) {
4403                                 i_id++;
4404                                 continue;
4405                         }
4406
4407                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4408                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4409                                                 RDMA_NL_RDMA_CM_ID_STATS,
4410                                                 NLM_F_MULTI);
4411                         if (!id_stats)
4412                                 goto out;
4413
4414                         memset(id_stats, 0, sizeof *id_stats);
4415                         id = &id_priv->id;
4416                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4417                         id_stats->port_num = id->port_num;
4418                         id_stats->bound_dev_if =
4419                                 id->route.addr.dev_addr.bound_dev_if;
4420
4421                         if (ibnl_put_attr(skb, nlh,
4422                                           rdma_addr_size(cma_src_addr(id_priv)),
4423                                           cma_src_addr(id_priv),
4424                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4425                                 goto out;
4426                         if (ibnl_put_attr(skb, nlh,
4427                                           rdma_addr_size(cma_src_addr(id_priv)),
4428                                           cma_dst_addr(id_priv),
4429                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4430                                 goto out;
4431
4432                         id_stats->pid           = id_priv->owner;
4433                         id_stats->port_space    = id->ps;
4434                         id_stats->cm_state      = id_priv->state;
4435                         id_stats->qp_num        = id_priv->qp_num;
4436                         id_stats->qp_type       = id->qp_type;
4437
4438                         i_id++;
4439                 }
4440
4441                 cb->args[1] = 0;
4442                 i_dev++;
4443         }
4444
4445 out:
4446         mutex_unlock(&lock);
4447         cb->args[0] = i_dev;
4448         cb->args[1] = i_id;
4449
4450         return skb->len;
4451 }
4452
4453 static const struct ibnl_client_cbs cma_cb_table[] = {
4454         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4455                                        .module = THIS_MODULE },
4456 };
4457
4458 static int cma_init_net(struct net *net)
4459 {
4460         struct cma_pernet *pernet = cma_pernet(net);
4461
4462         idr_init(&pernet->tcp_ps);
4463         idr_init(&pernet->udp_ps);
4464         idr_init(&pernet->ipoib_ps);
4465         idr_init(&pernet->ib_ps);
4466
4467         return 0;
4468 }
4469
4470 static void cma_exit_net(struct net *net)
4471 {
4472         struct cma_pernet *pernet = cma_pernet(net);
4473
4474         idr_destroy(&pernet->tcp_ps);
4475         idr_destroy(&pernet->udp_ps);
4476         idr_destroy(&pernet->ipoib_ps);
4477         idr_destroy(&pernet->ib_ps);
4478 }
4479
4480 static struct pernet_operations cma_pernet_operations = {
4481         .init = cma_init_net,
4482         .exit = cma_exit_net,
4483         .id = &cma_pernet_id,
4484         .size = sizeof(struct cma_pernet),
4485 };
4486
4487 static int __init cma_init(void)
4488 {
4489         int ret;
4490
4491         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4492         if (!cma_wq)
4493                 return -ENOMEM;
4494
4495         ret = register_pernet_subsys(&cma_pernet_operations);
4496         if (ret)
4497                 goto err_wq;
4498
4499         ib_sa_register_client(&sa_client);
4500         rdma_addr_register_client(&addr_client);
4501         register_netdevice_notifier(&cma_nb);
4502
4503         ret = ib_register_client(&cma_client);
4504         if (ret)
4505                 goto err;
4506
4507         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4508                             cma_cb_table))
4509                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4510         cma_configfs_init();
4511
4512         return 0;
4513
4514 err:
4515         unregister_netdevice_notifier(&cma_nb);
4516         rdma_addr_unregister_client(&addr_client);
4517         ib_sa_unregister_client(&sa_client);
4518 err_wq:
4519         destroy_workqueue(cma_wq);
4520         return ret;
4521 }
4522
4523 static void __exit cma_cleanup(void)
4524 {
4525         cma_configfs_exit();
4526         ibnl_remove_client(RDMA_NL_RDMA_CM);
4527         ib_unregister_client(&cma_client);
4528         unregister_netdevice_notifier(&cma_nb);
4529         rdma_addr_unregister_client(&addr_client);
4530         ib_sa_unregister_client(&sa_client);
4531         unregister_pernet_subsys(&cma_pernet_operations);
4532         destroy_workqueue(cma_wq);
4533 }
4534
4535 module_init(cma_init);
4536 module_exit(cma_cleanup);