drivers/infiniband/core/cma.c

   1 /*
   2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
   3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
   4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
   5  * Copyright (c) 2005-2006 Intel Corporation.  All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/completion.h>
  37 #include <linux/in.h>
  38 #include <linux/in6.h>
  39 #include <linux/mutex.h>
  40 #include <linux/random.h>
  41 #include <linux/igmp.h>
  42 #include <linux/idr.h>
  43 #include <linux/inetdevice.h>
  44 #include <linux/slab.h>
  45 #include <linux/module.h>
  46 #include <net/route.h>
  47
  48 #include <net/net_namespace.h>
  49 #include <net/netns/generic.h>
  50 #include <net/tcp.h>
  51 #include <net/ipv6.h>
  52 #include <net/ip_fib.h>
  53 #include <net/ip6_route.h>
  54
  55 #include <rdma/rdma_cm.h>
  56 #include <rdma/rdma_cm_ib.h>
  57 #include <rdma/rdma_netlink.h>
  58 #include <rdma/ib.h>
  59 #include <rdma/ib_cache.h>
  60 #include <rdma/ib_cm.h>
  61 #include <rdma/ib_sa.h>
  62 #include <rdma/iw_cm.h>
  63
  64 #include "core_priv.h"
  65
  66 MODULE_AUTHOR("Sean Hefty");
  67 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  68 MODULE_LICENSE("Dual BSD/GPL");
  69
  70 #define CMA_CM_RESPONSE_TIMEOUT 20
  71 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  72 #define CMA_MAX_CM_RETRIES 15
  73 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  74 #define CMA_IBOE_PACKET_LIFETIME 18
  75
  76 static const char * const cma_events[] = {
  77         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  78         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  79         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  80         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  81         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  82         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  83         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  84         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  85         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  86         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  87         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  88         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  89         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  90         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  91         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  92         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  93 };
  94
  95 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  96 {
  97         size_t index = event;
  98
  99         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 100                         cma_events[index] : "unrecognized event";
 101 }
 102 EXPORT_SYMBOL(rdma_event_msg);
 103
 104 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 105                                                 int reason)
 106 {
 107         if (rdma_ib_or_roce(id->device, id->port_num))
 108                 return ibcm_reject_msg(reason);
 109
 110         if (rdma_protocol_iwarp(id->device, id->port_num))
 111                 return iwcm_reject_msg(reason);
 112
 113         WARN_ON_ONCE(1);
 114         return "unrecognized transport";
 115 }
 116 EXPORT_SYMBOL(rdma_reject_msg);
 117
 118 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 119 {
 120         if (rdma_ib_or_roce(id->device, id->port_num))
 121                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 122
 123         if (rdma_protocol_iwarp(id->device, id->port_num))
 124                 return reason == -ECONNREFUSED;
 125
 126         WARN_ON_ONCE(1);
 127         return false;
 128 }
 129 EXPORT_SYMBOL(rdma_is_consumer_reject);
 130
 131 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 132                                       struct rdma_cm_event *ev, u8 *data_len)
 133 {
 134         const void *p;
 135
 136         if (rdma_is_consumer_reject(id, ev->status)) {
 137                 *data_len = ev->param.conn.private_data_len;
 138                 p = ev->param.conn.private_data;
 139         } else {
 140                 *data_len = 0;
 141                 p = NULL;
 142         }
 143         return p;
 144 }
 145 EXPORT_SYMBOL(rdma_consumer_reject_data);
 146
 147 static void cma_add_one(struct ib_device *device);
 148 static void cma_remove_one(struct ib_device *device, void *client_data);
 149
 150 static struct ib_client cma_client = {
 151         .name   = "cma",
 152         .add    = cma_add_one,
 153         .remove = cma_remove_one
 154 };
 155
 156 static struct ib_sa_client sa_client;
 157 static struct rdma_addr_client addr_client;
 158 static LIST_HEAD(dev_list);
 159 static LIST_HEAD(listen_any_list);
 160 static DEFINE_MUTEX(lock);
 161 static struct workqueue_struct *cma_wq;
 162 static unsigned int cma_pernet_id;
 163
 164 struct cma_pernet {
 165         struct idr tcp_ps;
 166         struct idr udp_ps;
 167         struct idr ipoib_ps;
 168         struct idr ib_ps;
 169 };
 170
 171 static struct cma_pernet *cma_pernet(struct net *net)
 172 {
 173         return net_generic(net, cma_pernet_id);
 174 }
 175
 176 static struct idr *cma_pernet_idr(struct net *net, enum rdma_port_space ps)
 177 {
 178         struct cma_pernet *pernet = cma_pernet(net);
 179
 180         switch (ps) {
 181         case RDMA_PS_TCP:
 182                 return &pernet->tcp_ps;
 183         case RDMA_PS_UDP:
 184                 return &pernet->udp_ps;
 185         case RDMA_PS_IPOIB:
 186                 return &pernet->ipoib_ps;
 187         case RDMA_PS_IB:
 188                 return &pernet->ib_ps;
 189         default:
 190                 return NULL;
 191         }
 192 }
 193
 194 struct cma_device {
 195         struct list_head        list;
 196         struct ib_device        *device;
 197         struct completion       comp;
 198         atomic_t                refcount;
 199         struct list_head        id_list;
 200         enum ib_gid_type        *default_gid_type;
 201         u8                      *default_roce_tos;
 202 };
 203
 204 struct rdma_bind_list {
 205         enum rdma_port_space    ps;
 206         struct hlist_head       owners;
 207         unsigned short          port;
 208 };
 209
 210 struct class_port_info_context {
 211         struct ib_class_port_info       *class_port_info;
 212         struct ib_device                *device;
 213         struct completion               done;
 214         struct ib_sa_query              *sa_query;
 215         u8                              port_num;
 216 };
 217
 218 static int cma_ps_alloc(struct net *net, enum rdma_port_space ps,
 219                         struct rdma_bind_list *bind_list, int snum)
 220 {
 221         struct idr *idr = cma_pernet_idr(net, ps);
 222
 223         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 224 }
 225
 226 static struct rdma_bind_list *cma_ps_find(struct net *net,
 227                                           enum rdma_port_space ps, int snum)
 228 {
 229         struct idr *idr = cma_pernet_idr(net, ps);
 230
 231         return idr_find(idr, snum);
 232 }
 233
 234 static void cma_ps_remove(struct net *net, enum rdma_port_space ps, int snum)
 235 {
 236         struct idr *idr = cma_pernet_idr(net, ps);
 237
 238         idr_remove(idr, snum);
 239 }
 240
 241 enum {
 242         CMA_OPTION_AFONLY,
 243 };
 244
 245 void cma_ref_dev(struct cma_device *cma_dev)
 246 {
 247         atomic_inc(&cma_dev->refcount);
 248 }
 249
 250 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 251                                              void               *cookie)
 252 {
 253         struct cma_device *cma_dev;
 254         struct cma_device *found_cma_dev = NULL;
 255
 256         mutex_lock(&lock);
 257
 258         list_for_each_entry(cma_dev, &dev_list, list)
 259                 if (filter(cma_dev->device, cookie)) {
 260                         found_cma_dev = cma_dev;
 261                         break;
 262                 }
 263
 264         if (found_cma_dev)
 265                 cma_ref_dev(found_cma_dev);
 266         mutex_unlock(&lock);
 267         return found_cma_dev;
 268 }
 269
 270 int cma_get_default_gid_type(struct cma_device *cma_dev,
 271                              unsigned int port)
 272 {
 273         if (!rdma_is_port_valid(cma_dev->device, port))
 274                 return -EINVAL;
 275
 276         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 277 }
 278
 279 int cma_set_default_gid_type(struct cma_device *cma_dev,
 280                              unsigned int port,
 281                              enum ib_gid_type default_gid_type)
 282 {
 283         unsigned long supported_gids;
 284
 285         if (!rdma_is_port_valid(cma_dev->device, port))
 286                 return -EINVAL;
 287
 288         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 289
 290         if (!(supported_gids & 1 << default_gid_type))
 291                 return -EINVAL;
 292
 293         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 294                 default_gid_type;
 295
 296         return 0;
 297 }
 298
 299 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 300 {
 301         if (!rdma_is_port_valid(cma_dev->device, port))
 302                 return -EINVAL;
 303
 304         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 305 }
 306
 307 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 308                              u8 default_roce_tos)
 309 {
 310         if (!rdma_is_port_valid(cma_dev->device, port))
 311                 return -EINVAL;
 312
 313         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 314                  default_roce_tos;
 315
 316         return 0;
 317 }
 318 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 319 {
 320         return cma_dev->device;
 321 }
 322
 323 /*
 324  * Device removal can occur at anytime, so we need extra handling to
 325  * serialize notifying the user of device removal with other callbacks.
 326  * We do this by disabling removal notification while a callback is in process,
 327  * and reporting it after the callback completes.
 328  */
 329 struct rdma_id_private {
 330         struct rdma_cm_id       id;
 331
 332         struct rdma_bind_list   *bind_list;
 333         struct hlist_node       node;
 334         struct list_head        list; /* listen_any_list or cma_device.list */
 335         struct list_head        listen_list; /* per device listens */
 336         struct cma_device       *cma_dev;
 337         struct list_head        mc_list;
 338
 339         int                     internal_id;
 340         enum rdma_cm_state      state;
 341         spinlock_t              lock;
 342         struct mutex            qp_mutex;
 343
 344         struct completion       comp;
 345         atomic_t                refcount;
 346         struct mutex            handler_mutex;
 347
 348         int                     backlog;
 349         int                     timeout_ms;
 350         struct ib_sa_query      *query;
 351         int                     query_id;
 352         union {
 353                 struct ib_cm_id *ib;
 354                 struct iw_cm_id *iw;
 355         } cm_id;
 356
 357         u32                     seq_num;
 358         u32                     qkey;
 359         u32                     qp_num;
 360         pid_t                   owner;
 361         u32                     options;
 362         u8                      srq;
 363         u8                      tos;
 364         bool                    tos_set;
 365         u8                      reuseaddr;
 366         u8                      afonly;
 367         enum ib_gid_type        gid_type;
 368 };
 369
 370 struct cma_multicast {
 371         struct rdma_id_private *id_priv;
 372         union {
 373                 struct ib_sa_multicast *ib;
 374         } multicast;
 375         struct list_head        list;
 376         void                    *context;
 377         struct sockaddr_storage addr;
 378         struct kref             mcref;
 379         bool                    igmp_joined;
 380         u8                      join_state;
 381 };
 382
 383 struct cma_work {
 384         struct work_struct      work;
 385         struct rdma_id_private  *id;
 386         enum rdma_cm_state      old_state;
 387         enum rdma_cm_state      new_state;
 388         struct rdma_cm_event    event;
 389 };
 390
 391 struct cma_ndev_work {
 392         struct work_struct      work;
 393         struct rdma_id_private  *id;
 394         struct rdma_cm_event    event;
 395 };
 396
 397 struct iboe_mcast_work {
 398         struct work_struct       work;
 399         struct rdma_id_private  *id;
 400         struct cma_multicast    *mc;
 401 };
 402
 403 union cma_ip_addr {
 404         struct in6_addr ip6;
 405         struct {
 406                 __be32 pad[3];
 407                 __be32 addr;
 408         } ip4;
 409 };
 410
 411 struct cma_hdr {
 412         u8 cma_version;
 413         u8 ip_version;  /* IP version: 7:4 */
 414         __be16 port;
 415         union cma_ip_addr src_addr;
 416         union cma_ip_addr dst_addr;
 417 };
 418
 419 #define CMA_VERSION 0x00
 420
 421 struct cma_req_info {
 422         struct ib_device *device;
 423         int port;
 424         union ib_gid local_gid;
 425         __be64 service_id;
 426         u16 pkey;
 427         bool has_gid:1;
 428 };
 429
 430 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 431 {
 432         unsigned long flags;
 433         int ret;
 434
 435         spin_lock_irqsave(&id_priv->lock, flags);
 436         ret = (id_priv->state == comp);
 437         spin_unlock_irqrestore(&id_priv->lock, flags);
 438         return ret;
 439 }
 440
 441 static int cma_comp_exch(struct rdma_id_private *id_priv,
 442                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 443 {
 444         unsigned long flags;
 445         int ret;
 446
 447         spin_lock_irqsave(&id_priv->lock, flags);
 448         if ((ret = (id_priv->state == comp)))
 449                 id_priv->state = exch;
 450         spin_unlock_irqrestore(&id_priv->lock, flags);
 451         return ret;
 452 }
 453
 454 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 455                                    enum rdma_cm_state exch)
 456 {
 457         unsigned long flags;
 458         enum rdma_cm_state old;
 459
 460         spin_lock_irqsave(&id_priv->lock, flags);
 461         old = id_priv->state;
 462         id_priv->state = exch;
 463         spin_unlock_irqrestore(&id_priv->lock, flags);
 464         return old;
 465 }
 466
 467 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 468 {
 469         return hdr->ip_version >> 4;
 470 }
 471
 472 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 473 {
 474         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 475 }
 476
 477 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 478 {
 479         struct in_device *in_dev = NULL;
 480
 481         if (ndev) {
 482                 rtnl_lock();
 483                 in_dev = __in_dev_get_rtnl(ndev);
 484                 if (in_dev) {
 485                         if (join)
 486                                 ip_mc_inc_group(in_dev,
 487                                                 *(__be32 *)(mgid->raw + 12));
 488                         else
 489                                 ip_mc_dec_group(in_dev,
 490                                                 *(__be32 *)(mgid->raw + 12));
 491                 }
 492                 rtnl_unlock();
 493         }
 494         return (in_dev) ? 0 : -ENODEV;
 495 }
 496
 497 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 498                                struct cma_device *cma_dev)
 499 {
 500         cma_ref_dev(cma_dev);
 501         id_priv->cma_dev = cma_dev;
 502         id_priv->gid_type = 0;
 503         id_priv->id.device = cma_dev->device;
 504         id_priv->id.route.addr.dev_addr.transport =
 505                 rdma_node_get_transport(cma_dev->device->node_type);
 506         list_add_tail(&id_priv->list, &cma_dev->id_list);
 507 }
 508
 509 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 510                               struct cma_device *cma_dev)
 511 {
 512         _cma_attach_to_dev(id_priv, cma_dev);
 513         id_priv->gid_type =
 514                 cma_dev->default_gid_type[id_priv->id.port_num -
 515                                           rdma_start_port(cma_dev->device)];
 516 }
 517
 518 void cma_deref_dev(struct cma_device *cma_dev)
 519 {
 520         if (atomic_dec_and_test(&cma_dev->refcount))
 521                 complete(&cma_dev->comp);
 522 }
 523
 524 static inline void release_mc(struct kref *kref)
 525 {
 526         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 527
 528         kfree(mc->multicast.ib);
 529         kfree(mc);
 530 }
 531
 532 static void cma_release_dev(struct rdma_id_private *id_priv)
 533 {
 534         mutex_lock(&lock);
 535         list_del(&id_priv->list);
 536         cma_deref_dev(id_priv->cma_dev);
 537         id_priv->cma_dev = NULL;
 538         mutex_unlock(&lock);
 539 }
 540
 541 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 542 {
 543         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 544 }
 545
 546 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 547 {
 548         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 549 }
 550
 551 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 552 {
 553         return id_priv->id.route.addr.src_addr.ss_family;
 554 }
 555
 556 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 557 {
 558         struct ib_sa_mcmember_rec rec;
 559         int ret = 0;
 560
 561         if (id_priv->qkey) {
 562                 if (qkey && id_priv->qkey != qkey)
 563                         return -EINVAL;
 564                 return 0;
 565         }
 566
 567         if (qkey) {
 568                 id_priv->qkey = qkey;
 569                 return 0;
 570         }
 571
 572         switch (id_priv->id.ps) {
 573         case RDMA_PS_UDP:
 574         case RDMA_PS_IB:
 575                 id_priv->qkey = RDMA_UDP_QKEY;
 576                 break;
 577         case RDMA_PS_IPOIB:
 578                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 579                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 580                                              id_priv->id.port_num, &rec.mgid,
 581                                              &rec);
 582                 if (!ret)
 583                         id_priv->qkey = be32_to_cpu(rec.qkey);
 584                 break;
 585         default:
 586                 break;
 587         }
 588         return ret;
 589 }
 590
 591 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 592 {
 593         dev_addr->dev_type = ARPHRD_INFINIBAND;
 594         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 595         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 596 }
 597
 598 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 599 {
 600         int ret;
 601
 602         if (addr->sa_family != AF_IB) {
 603                 ret = rdma_translate_ip(addr, dev_addr, NULL);
 604         } else {
 605                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 606                 ret = 0;
 607         }
 608
 609         return ret;
 610 }
 611
 612 static inline int cma_validate_port(struct ib_device *device, u8 port,
 613                                     enum ib_gid_type gid_type,
 614                                       union ib_gid *gid, int dev_type,
 615                                       int bound_if_index)
 616 {
 617         int ret = -ENODEV;
 618         struct net_device *ndev = NULL;
 619
 620         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 621                 return ret;
 622
 623         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 624                 return ret;
 625
 626         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 627                 ndev = dev_get_by_index(&init_net, bound_if_index);
 628                 if (ndev && ndev->flags & IFF_LOOPBACK) {
 629                         pr_info("detected loopback device\n");
 630                         dev_put(ndev);
 631
 632                         if (!device->get_netdev)
 633                                 return -EOPNOTSUPP;
 634
 635                         ndev = device->get_netdev(device, port);
 636                         if (!ndev)
 637                                 return -ENODEV;
 638                 }
 639         } else {
 640                 gid_type = IB_GID_TYPE_IB;
 641         }
 642
 643         ret = ib_find_cached_gid_by_port(device, gid, gid_type, port,
 644                                          ndev, NULL);
 645
 646         if (ndev)
 647                 dev_put(ndev);
 648
 649         return ret;
 650 }
 651
 652 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 653                            struct rdma_id_private *listen_id_priv)
 654 {
 655         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 656         struct cma_device *cma_dev;
 657         union ib_gid gid, iboe_gid, *gidp;
 658         int ret = -ENODEV;
 659         u8 port;
 660
 661         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 662             id_priv->id.ps == RDMA_PS_IPOIB)
 663                 return -EINVAL;
 664
 665         mutex_lock(&lock);
 666         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 667                     &iboe_gid);
 668
 669         memcpy(&gid, dev_addr->src_dev_addr +
 670                rdma_addr_gid_offset(dev_addr), sizeof gid);
 671
 672         if (listen_id_priv) {
 673                 cma_dev = listen_id_priv->cma_dev;
 674                 port = listen_id_priv->id.port_num;
 675                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 676                        &iboe_gid : &gid;
 677
 678                 ret = cma_validate_port(cma_dev->device, port,
 679                                         rdma_protocol_ib(cma_dev->device, port) ?
 680                                         IB_GID_TYPE_IB :
 681                                         listen_id_priv->gid_type, gidp,
 682                                         dev_addr->dev_type,
 683                                         dev_addr->bound_dev_if);
 684                 if (!ret) {
 685                         id_priv->id.port_num = port;
 686                         goto out;
 687                 }
 688         }
 689
 690         list_for_each_entry(cma_dev, &dev_list, list) {
 691                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 692                         if (listen_id_priv &&
 693                             listen_id_priv->cma_dev == cma_dev &&
 694                             listen_id_priv->id.port_num == port)
 695                                 continue;
 696
 697                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 698                                &iboe_gid : &gid;
 699
 700                         ret = cma_validate_port(cma_dev->device, port,
 701                                                 rdma_protocol_ib(cma_dev->device, port) ?
 702                                                 IB_GID_TYPE_IB :
 703                                                 cma_dev->default_gid_type[port - 1],
 704                                                 gidp, dev_addr->dev_type,
 705                                                 dev_addr->bound_dev_if);
 706                         if (!ret) {
 707                                 id_priv->id.port_num = port;
 708                                 goto out;
 709                         }
 710                 }
 711         }
 712
 713 out:
 714         if (!ret)
 715                 cma_attach_to_dev(id_priv, cma_dev);
 716
 717         mutex_unlock(&lock);
 718         return ret;
 719 }
 720
 721 /*
 722  * Select the source IB device and address to reach the destination IB address.
 723  */
 724 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 725 {
 726         struct cma_device *cma_dev, *cur_dev;
 727         struct sockaddr_ib *addr;
 728         union ib_gid gid, sgid, *dgid;
 729         u16 pkey, index;
 730         u8 p;
 731         enum ib_port_state port_state;
 732         int i;
 733
 734         cma_dev = NULL;
 735         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 736         dgid = (union ib_gid *) &addr->sib_addr;
 737         pkey = ntohs(addr->sib_pkey);
 738
 739         list_for_each_entry(cur_dev, &dev_list, list) {
 740                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 741                         if (!rdma_cap_af_ib(cur_dev->device, p))
 742                                 continue;
 743
 744                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 745                                 continue;
 746
 747                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 748                                 continue;
 749                         for (i = 0; !ib_get_cached_gid(cur_dev->device, p, i,
 750                                                        &gid, NULL);
 751                              i++) {
 752                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 753                                         cma_dev = cur_dev;
 754                                         sgid = gid;
 755                                         id_priv->id.port_num = p;
 756                                         goto found;
 757                                 }
 758
 759                                 if (!cma_dev && (gid.global.subnet_prefix ==
 760                                     dgid->global.subnet_prefix) &&
 761                                     port_state == IB_PORT_ACTIVE) {
 762                                         cma_dev = cur_dev;
 763                                         sgid = gid;
 764                                         id_priv->id.port_num = p;
 765                                 }
 766                         }
 767                 }
 768         }
 769
 770         if (!cma_dev)
 771                 return -ENODEV;
 772
 773 found:
 774         cma_attach_to_dev(id_priv, cma_dev);
 775         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 776         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 777         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 778         return 0;
 779 }
 780
 781 static void cma_deref_id(struct rdma_id_private *id_priv)
 782 {
 783         if (atomic_dec_and_test(&id_priv->refcount))
 784                 complete(&id_priv->comp);
 785 }
 786
 787 struct rdma_cm_id *rdma_create_id(struct net *net,
 788                                   rdma_cm_event_handler event_handler,
 789                                   void *context, enum rdma_port_space ps,
 790                                   enum ib_qp_type qp_type)
 791 {
 792         struct rdma_id_private *id_priv;
 793
 794         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 795         if (!id_priv)
 796                 return ERR_PTR(-ENOMEM);
 797
 798         id_priv->owner = task_pid_nr(current);
 799         id_priv->state = RDMA_CM_IDLE;
 800         id_priv->id.context = context;
 801         id_priv->id.event_handler = event_handler;
 802         id_priv->id.ps = ps;
 803         id_priv->id.qp_type = qp_type;
 804         id_priv->tos_set = false;
 805         spin_lock_init(&id_priv->lock);
 806         mutex_init(&id_priv->qp_mutex);
 807         init_completion(&id_priv->comp);
 808         atomic_set(&id_priv->refcount, 1);
 809         mutex_init(&id_priv->handler_mutex);
 810         INIT_LIST_HEAD(&id_priv->listen_list);
 811         INIT_LIST_HEAD(&id_priv->mc_list);
 812         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 813         id_priv->id.route.addr.dev_addr.net = get_net(net);
 814
 815         return &id_priv->id;
 816 }
 817 EXPORT_SYMBOL(rdma_create_id);
 818
 819 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 820 {
 821         struct ib_qp_attr qp_attr;
 822         int qp_attr_mask, ret;
 823
 824         qp_attr.qp_state = IB_QPS_INIT;
 825         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 826         if (ret)
 827                 return ret;
 828
 829         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 830         if (ret)
 831                 return ret;
 832
 833         qp_attr.qp_state = IB_QPS_RTR;
 834         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 835         if (ret)
 836                 return ret;
 837
 838         qp_attr.qp_state = IB_QPS_RTS;
 839         qp_attr.sq_psn = 0;
 840         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 841
 842         return ret;
 843 }
 844
 845 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 846 {
 847         struct ib_qp_attr qp_attr;
 848         int qp_attr_mask, ret;
 849
 850         qp_attr.qp_state = IB_QPS_INIT;
 851         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 852         if (ret)
 853                 return ret;
 854
 855         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 856 }
 857
 858 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 859                    struct ib_qp_init_attr *qp_init_attr)
 860 {
 861         struct rdma_id_private *id_priv;
 862         struct ib_qp *qp;
 863         int ret;
 864
 865         id_priv = container_of(id, struct rdma_id_private, id);
 866         if (id->device != pd->device)
 867                 return -EINVAL;
 868
 869         qp_init_attr->port_num = id->port_num;
 870         qp = ib_create_qp(pd, qp_init_attr);
 871         if (IS_ERR(qp))
 872                 return PTR_ERR(qp);
 873
 874         if (id->qp_type == IB_QPT_UD)
 875                 ret = cma_init_ud_qp(id_priv, qp);
 876         else
 877                 ret = cma_init_conn_qp(id_priv, qp);
 878         if (ret)
 879                 goto err;
 880
 881         id->qp = qp;
 882         id_priv->qp_num = qp->qp_num;
 883         id_priv->srq = (qp->srq != NULL);
 884         return 0;
 885 err:
 886         ib_destroy_qp(qp);
 887         return ret;
 888 }
 889 EXPORT_SYMBOL(rdma_create_qp);
 890
 891 void rdma_destroy_qp(struct rdma_cm_id *id)
 892 {
 893         struct rdma_id_private *id_priv;
 894
 895         id_priv = container_of(id, struct rdma_id_private, id);
 896         mutex_lock(&id_priv->qp_mutex);
 897         ib_destroy_qp(id_priv->id.qp);
 898         id_priv->id.qp = NULL;
 899         mutex_unlock(&id_priv->qp_mutex);
 900 }
 901 EXPORT_SYMBOL(rdma_destroy_qp);
 902
 903 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 904                              struct rdma_conn_param *conn_param)
 905 {
 906         struct ib_qp_attr qp_attr;
 907         int qp_attr_mask, ret;
 908         union ib_gid sgid;
 909
 910         mutex_lock(&id_priv->qp_mutex);
 911         if (!id_priv->id.qp) {
 912                 ret = 0;
 913                 goto out;
 914         }
 915
 916         /* Need to update QP attributes from default values. */
 917         qp_attr.qp_state = IB_QPS_INIT;
 918         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 919         if (ret)
 920                 goto out;
 921
 922         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 923         if (ret)
 924                 goto out;
 925
 926         qp_attr.qp_state = IB_QPS_RTR;
 927         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 928         if (ret)
 929                 goto out;
 930
 931         ret = ib_query_gid(id_priv->id.device, id_priv->id.port_num,
 932                            rdma_ah_read_grh(&qp_attr.ah_attr)->sgid_index,
 933                            &sgid, NULL);
 934         if (ret)
 935                 goto out;
 936
 937         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 938
 939         if (conn_param)
 940                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 941         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 942 out:
 943         mutex_unlock(&id_priv->qp_mutex);
 944         return ret;
 945 }
 946
 947 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 948                              struct rdma_conn_param *conn_param)
 949 {
 950         struct ib_qp_attr qp_attr;
 951         int qp_attr_mask, ret;
 952
 953         mutex_lock(&id_priv->qp_mutex);
 954         if (!id_priv->id.qp) {
 955                 ret = 0;
 956                 goto out;
 957         }
 958
 959         qp_attr.qp_state = IB_QPS_RTS;
 960         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 961         if (ret)
 962                 goto out;
 963
 964         if (conn_param)
 965                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 966         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 967 out:
 968         mutex_unlock(&id_priv->qp_mutex);
 969         return ret;
 970 }
 971
 972 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 973 {
 974         struct ib_qp_attr qp_attr;
 975         int ret;
 976
 977         mutex_lock(&id_priv->qp_mutex);
 978         if (!id_priv->id.qp) {
 979                 ret = 0;
 980                 goto out;
 981         }
 982
 983         qp_attr.qp_state = IB_QPS_ERR;
 984         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 985 out:
 986         mutex_unlock(&id_priv->qp_mutex);
 987         return ret;
 988 }
 989
 990 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 991                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 992 {
 993         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 994         int ret;
 995         u16 pkey;
 996
 997         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 998                 pkey = 0xffff;
 999         else
1000                 pkey = ib_addr_get_pkey(dev_addr);
1001
1002         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
1003                                   pkey, &qp_attr->pkey_index);
1004         if (ret)
1005                 return ret;
1006
1007         qp_attr->port_num = id_priv->id.port_num;
1008         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1009
1010         if (id_priv->id.qp_type == IB_QPT_UD) {
1011                 ret = cma_set_qkey(id_priv, 0);
1012                 if (ret)
1013                         return ret;
1014
1015                 qp_attr->qkey = id_priv->qkey;
1016                 *qp_attr_mask |= IB_QP_QKEY;
1017         } else {
1018                 qp_attr->qp_access_flags = 0;
1019                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1020         }
1021         return 0;
1022 }
1023
1024 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1025                        int *qp_attr_mask)
1026 {
1027         struct rdma_id_private *id_priv;
1028         int ret = 0;
1029
1030         id_priv = container_of(id, struct rdma_id_private, id);
1031         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1032                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1033                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1034                 else
1035                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1036                                                  qp_attr_mask);
1037
1038                 if (qp_attr->qp_state == IB_QPS_RTR)
1039                         qp_attr->rq_psn = id_priv->seq_num;
1040         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1041                 if (!id_priv->cm_id.iw) {
1042                         qp_attr->qp_access_flags = 0;
1043                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1044                 } else
1045                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1046                                                  qp_attr_mask);
1047         } else
1048                 ret = -ENOSYS;
1049
1050         return ret;
1051 }
1052 EXPORT_SYMBOL(rdma_init_qp_attr);
1053
1054 static inline int cma_zero_addr(struct sockaddr *addr)
1055 {
1056         switch (addr->sa_family) {
1057         case AF_INET:
1058                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1059         case AF_INET6:
1060                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1061         case AF_IB:
1062                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1063         default:
1064                 return 0;
1065         }
1066 }
1067
1068 static inline int cma_loopback_addr(struct sockaddr *addr)
1069 {
1070         switch (addr->sa_family) {
1071         case AF_INET:
1072                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1073         case AF_INET6:
1074                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1075         case AF_IB:
1076                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1077         default:
1078                 return 0;
1079         }
1080 }
1081
1082 static inline int cma_any_addr(struct sockaddr *addr)
1083 {
1084         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1085 }
1086
1087 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1088 {
1089         if (src->sa_family != dst->sa_family)
1090                 return -1;
1091
1092         switch (src->sa_family) {
1093         case AF_INET:
1094                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1095                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1096         case AF_INET6:
1097                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1098                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1099         default:
1100                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1101                                    &((struct sockaddr_ib *) dst)->sib_addr);
1102         }
1103 }
1104
1105 static __be16 cma_port(struct sockaddr *addr)
1106 {
1107         struct sockaddr_ib *sib;
1108
1109         switch (addr->sa_family) {
1110         case AF_INET:
1111                 return ((struct sockaddr_in *) addr)->sin_port;
1112         case AF_INET6:
1113                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1114         case AF_IB:
1115                 sib = (struct sockaddr_ib *) addr;
1116                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1117                                     be64_to_cpu(sib->sib_sid_mask)));
1118         default:
1119                 return 0;
1120         }
1121 }
1122
1123 static inline int cma_any_port(struct sockaddr *addr)
1124 {
1125         return !cma_port(addr);
1126 }
1127
1128 static void cma_save_ib_info(struct sockaddr *src_addr,
1129                              struct sockaddr *dst_addr,
1130                              struct rdma_cm_id *listen_id,
1131                              struct sa_path_rec *path)
1132 {
1133         struct sockaddr_ib *listen_ib, *ib;
1134
1135         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1136         if (src_addr) {
1137                 ib = (struct sockaddr_ib *)src_addr;
1138                 ib->sib_family = AF_IB;
1139                 if (path) {
1140                         ib->sib_pkey = path->pkey;
1141                         ib->sib_flowinfo = path->flow_label;
1142                         memcpy(&ib->sib_addr, &path->sgid, 16);
1143                         ib->sib_sid = sa_path_get_service_id(path);
1144                         ib->sib_scope_id = 0;
1145                 } else {
1146                         ib->sib_pkey = listen_ib->sib_pkey;
1147                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1148                         ib->sib_addr = listen_ib->sib_addr;
1149                         ib->sib_sid = listen_ib->sib_sid;
1150                         ib->sib_scope_id = listen_ib->sib_scope_id;
1151                 }
1152                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1153         }
1154         if (dst_addr) {
1155                 ib = (struct sockaddr_ib *)dst_addr;
1156                 ib->sib_family = AF_IB;
1157                 if (path) {
1158                         ib->sib_pkey = path->pkey;
1159                         ib->sib_flowinfo = path->flow_label;
1160                         memcpy(&ib->sib_addr, &path->dgid, 16);
1161                 }
1162         }
1163 }
1164
1165 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1166                               struct sockaddr_in *dst_addr,
1167                               struct cma_hdr *hdr,
1168                               __be16 local_port)
1169 {
1170         if (src_addr) {
1171                 *src_addr = (struct sockaddr_in) {
1172                         .sin_family = AF_INET,
1173                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1174                         .sin_port = local_port,
1175                 };
1176         }
1177
1178         if (dst_addr) {
1179                 *dst_addr = (struct sockaddr_in) {
1180                         .sin_family = AF_INET,
1181                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1182                         .sin_port = hdr->port,
1183                 };
1184         }
1185 }
1186
1187 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1188                               struct sockaddr_in6 *dst_addr,
1189                               struct cma_hdr *hdr,
1190                               __be16 local_port)
1191 {
1192         if (src_addr) {
1193                 *src_addr = (struct sockaddr_in6) {
1194                         .sin6_family = AF_INET6,
1195                         .sin6_addr = hdr->dst_addr.ip6,
1196                         .sin6_port = local_port,
1197                 };
1198         }
1199
1200         if (dst_addr) {
1201                 *dst_addr = (struct sockaddr_in6) {
1202                         .sin6_family = AF_INET6,
1203                         .sin6_addr = hdr->src_addr.ip6,
1204                         .sin6_port = hdr->port,
1205                 };
1206         }
1207 }
1208
1209 static u16 cma_port_from_service_id(__be64 service_id)
1210 {
1211         return (u16)be64_to_cpu(service_id);
1212 }
1213
1214 static int cma_save_ip_info(struct sockaddr *src_addr,
1215                             struct sockaddr *dst_addr,
1216                             struct ib_cm_event *ib_event,
1217                             __be64 service_id)
1218 {
1219         struct cma_hdr *hdr;
1220         __be16 port;
1221
1222         hdr = ib_event->private_data;
1223         if (hdr->cma_version != CMA_VERSION)
1224                 return -EINVAL;
1225
1226         port = htons(cma_port_from_service_id(service_id));
1227
1228         switch (cma_get_ip_ver(hdr)) {
1229         case 4:
1230                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1231                                   (struct sockaddr_in *)dst_addr, hdr, port);
1232                 break;
1233         case 6:
1234                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1235                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1236                 break;
1237         default:
1238                 return -EAFNOSUPPORT;
1239         }
1240
1241         return 0;
1242 }
1243
1244 static int cma_save_net_info(struct sockaddr *src_addr,
1245                              struct sockaddr *dst_addr,
1246                              struct rdma_cm_id *listen_id,
1247                              struct ib_cm_event *ib_event,
1248                              sa_family_t sa_family, __be64 service_id)
1249 {
1250         if (sa_family == AF_IB) {
1251                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1252                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1253                                          ib_event->param.req_rcvd.primary_path);
1254                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1255                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1256                 return 0;
1257         }
1258
1259         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1260 }
1261
1262 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1263                              struct cma_req_info *req)
1264 {
1265         const struct ib_cm_req_event_param *req_param =
1266                 &ib_event->param.req_rcvd;
1267         const struct ib_cm_sidr_req_event_param *sidr_param =
1268                 &ib_event->param.sidr_req_rcvd;
1269
1270         switch (ib_event->event) {
1271         case IB_CM_REQ_RECEIVED:
1272                 req->device     = req_param->listen_id->device;
1273                 req->port       = req_param->port;
1274                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1275                        sizeof(req->local_gid));
1276                 req->has_gid    = true;
1277                 req->service_id =
1278                         sa_path_get_service_id(req_param->primary_path);
1279                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1280                 if (req->pkey != req_param->bth_pkey)
1281                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1282                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1283                                             req_param->bth_pkey, req->pkey);
1284                 break;
1285         case IB_CM_SIDR_REQ_RECEIVED:
1286                 req->device     = sidr_param->listen_id->device;
1287                 req->port       = sidr_param->port;
1288                 req->has_gid    = false;
1289                 req->service_id = sidr_param->service_id;
1290                 req->pkey       = sidr_param->pkey;
1291                 if (req->pkey != sidr_param->bth_pkey)
1292                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1293                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1294                                             sidr_param->bth_pkey, req->pkey);
1295                 break;
1296         default:
1297                 return -EINVAL;
1298         }
1299
1300         return 0;
1301 }
1302
1303 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1304                                   const struct sockaddr_in *dst_addr,
1305                                   const struct sockaddr_in *src_addr)
1306 {
1307         __be32 daddr = dst_addr->sin_addr.s_addr,
1308                saddr = src_addr->sin_addr.s_addr;
1309         struct fib_result res;
1310         struct flowi4 fl4;
1311         int err;
1312         bool ret;
1313
1314         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1315             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1316             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1317             ipv4_is_loopback(saddr))
1318                 return false;
1319
1320         memset(&fl4, 0, sizeof(fl4));
1321         fl4.flowi4_iif = net_dev->ifindex;
1322         fl4.daddr = daddr;
1323         fl4.saddr = saddr;
1324
1325         rcu_read_lock();
1326         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1327         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1328         rcu_read_unlock();
1329
1330         return ret;
1331 }
1332
1333 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1334                                   const struct sockaddr_in6 *dst_addr,
1335                                   const struct sockaddr_in6 *src_addr)
1336 {
1337 #if IS_ENABLED(CONFIG_IPV6)
1338         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1339                            IPV6_ADDR_LINKLOCAL;
1340         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1341                                          &src_addr->sin6_addr, net_dev->ifindex,
1342                                          strict);
1343         bool ret;
1344
1345         if (!rt)
1346                 return false;
1347
1348         ret = rt->rt6i_idev->dev == net_dev;
1349         ip6_rt_put(rt);
1350
1351         return ret;
1352 #else
1353         return false;
1354 #endif
1355 }
1356
1357 static bool validate_net_dev(struct net_device *net_dev,
1358                              const struct sockaddr *daddr,
1359                              const struct sockaddr *saddr)
1360 {
1361         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1362         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1363         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1364         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1365
1366         switch (daddr->sa_family) {
1367         case AF_INET:
1368                 return saddr->sa_family == AF_INET &&
1369                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1370
1371         case AF_INET6:
1372                 return saddr->sa_family == AF_INET6 &&
1373                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1374
1375         default:
1376                 return false;
1377         }
1378 }
1379
1380 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1381                                           const struct cma_req_info *req)
1382 {
1383         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1384         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1385                         *src_addr = (struct sockaddr *)&src_addr_storage;
1386         struct net_device *net_dev;
1387         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1388         int err;
1389
1390         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1391                                req->service_id);
1392         if (err)
1393                 return ERR_PTR(err);
1394
1395         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1396                                            gid, listen_addr);
1397         if (!net_dev)
1398                 return ERR_PTR(-ENODEV);
1399
1400         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1401                 dev_put(net_dev);
1402                 return ERR_PTR(-EHOSTUNREACH);
1403         }
1404
1405         return net_dev;
1406 }
1407
1408 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1409 {
1410         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1411 }
1412
1413 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1414                                    const struct cma_hdr *hdr)
1415 {
1416         struct sockaddr *addr = cma_src_addr(id_priv);
1417         __be32 ip4_addr;
1418         struct in6_addr ip6_addr;
1419
1420         if (cma_any_addr(addr) && !id_priv->afonly)
1421                 return true;
1422
1423         switch (addr->sa_family) {
1424         case AF_INET:
1425                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1426                 if (cma_get_ip_ver(hdr) != 4)
1427                         return false;
1428                 if (!cma_any_addr(addr) &&
1429                     hdr->dst_addr.ip4.addr != ip4_addr)
1430                         return false;
1431                 break;
1432         case AF_INET6:
1433                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1434                 if (cma_get_ip_ver(hdr) != 6)
1435                         return false;
1436                 if (!cma_any_addr(addr) &&
1437                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1438                         return false;
1439                 break;
1440         case AF_IB:
1441                 return true;
1442         default:
1443                 return false;
1444         }
1445
1446         return true;
1447 }
1448
1449 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1450 {
1451         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1452         enum rdma_transport_type transport =
1453                 rdma_node_get_transport(device->node_type);
1454
1455         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1456 }
1457
1458 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1459 {
1460         struct ib_device *device = id->device;
1461         const int port_num = id->port_num ?: rdma_start_port(device);
1462
1463         return cma_protocol_roce_dev_port(device, port_num);
1464 }
1465
1466 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1467                               const struct net_device *net_dev,
1468                               u8 port_num)
1469 {
1470         const struct rdma_addr *addr = &id->route.addr;
1471
1472         if (!net_dev)
1473                 /* This request is an AF_IB request or a RoCE request */
1474                 return (!id->port_num || id->port_num == port_num) &&
1475                        (addr->src_addr.ss_family == AF_IB ||
1476                         cma_protocol_roce_dev_port(id->device, port_num));
1477
1478         return !addr->dev_addr.bound_dev_if ||
1479                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1480                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1481 }
1482
1483 static struct rdma_id_private *cma_find_listener(
1484                 const struct rdma_bind_list *bind_list,
1485                 const struct ib_cm_id *cm_id,
1486                 const struct ib_cm_event *ib_event,
1487                 const struct cma_req_info *req,
1488                 const struct net_device *net_dev)
1489 {
1490         struct rdma_id_private *id_priv, *id_priv_dev;
1491
1492         if (!bind_list)
1493                 return ERR_PTR(-EINVAL);
1494
1495         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1496                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1497                         if (id_priv->id.device == cm_id->device &&
1498                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1499                                 return id_priv;
1500                         list_for_each_entry(id_priv_dev,
1501                                             &id_priv->listen_list,
1502                                             listen_list) {
1503                                 if (id_priv_dev->id.device == cm_id->device &&
1504                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1505                                         return id_priv_dev;
1506                         }
1507                 }
1508         }
1509
1510         return ERR_PTR(-EINVAL);
1511 }
1512
1513 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1514                                                  struct ib_cm_event *ib_event,
1515                                                  struct net_device **net_dev)
1516 {
1517         struct cma_req_info req;
1518         struct rdma_bind_list *bind_list;
1519         struct rdma_id_private *id_priv;
1520         int err;
1521
1522         err = cma_save_req_info(ib_event, &req);
1523         if (err)
1524                 return ERR_PTR(err);
1525
1526         *net_dev = cma_get_net_dev(ib_event, &req);
1527         if (IS_ERR(*net_dev)) {
1528                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1529                         /* Assuming the protocol is AF_IB */
1530                         *net_dev = NULL;
1531                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1532                         /* TODO find the net dev matching the request parameters
1533                          * through the RoCE GID table */
1534                         *net_dev = NULL;
1535                 } else {
1536                         return ERR_CAST(*net_dev);
1537                 }
1538         }
1539
1540         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1541                                 rdma_ps_from_service_id(req.service_id),
1542                                 cma_port_from_service_id(req.service_id));
1543         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1544         if (IS_ERR(id_priv) && *net_dev) {
1545                 dev_put(*net_dev);
1546                 *net_dev = NULL;
1547         }
1548
1549         return id_priv;
1550 }
1551
1552 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1553 {
1554         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1555 }
1556
1557 static void cma_cancel_route(struct rdma_id_private *id_priv)
1558 {
1559         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1560                 if (id_priv->query)
1561                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1562         }
1563 }
1564
1565 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1566 {
1567         struct rdma_id_private *dev_id_priv;
1568
1569         /*
1570          * Remove from listen_any_list to prevent added devices from spawning
1571          * additional listen requests.
1572          */
1573         mutex_lock(&lock);
1574         list_del(&id_priv->list);
1575
1576         while (!list_empty(&id_priv->listen_list)) {
1577                 dev_id_priv = list_entry(id_priv->listen_list.next,
1578                                          struct rdma_id_private, listen_list);
1579                 /* sync with device removal to avoid duplicate destruction */
1580                 list_del_init(&dev_id_priv->list);
1581                 list_del(&dev_id_priv->listen_list);
1582                 mutex_unlock(&lock);
1583
1584                 rdma_destroy_id(&dev_id_priv->id);
1585                 mutex_lock(&lock);
1586         }
1587         mutex_unlock(&lock);
1588 }
1589
1590 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1591                                  enum rdma_cm_state state)
1592 {
1593         switch (state) {
1594         case RDMA_CM_ADDR_QUERY:
1595                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1596                 break;
1597         case RDMA_CM_ROUTE_QUERY:
1598                 cma_cancel_route(id_priv);
1599                 break;
1600         case RDMA_CM_LISTEN:
1601                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1602                         cma_cancel_listens(id_priv);
1603                 break;
1604         default:
1605                 break;
1606         }
1607 }
1608
1609 static void cma_release_port(struct rdma_id_private *id_priv)
1610 {
1611         struct rdma_bind_list *bind_list = id_priv->bind_list;
1612         struct net *net = id_priv->id.route.addr.dev_addr.net;
1613
1614         if (!bind_list)
1615                 return;
1616
1617         mutex_lock(&lock);
1618         hlist_del(&id_priv->node);
1619         if (hlist_empty(&bind_list->owners)) {
1620                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1621                 kfree(bind_list);
1622         }
1623         mutex_unlock(&lock);
1624 }
1625
1626 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1627 {
1628         struct cma_multicast *mc;
1629
1630         while (!list_empty(&id_priv->mc_list)) {
1631                 mc = container_of(id_priv->mc_list.next,
1632                                   struct cma_multicast, list);
1633                 list_del(&mc->list);
1634                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1635                                       id_priv->id.port_num)) {
1636                         ib_sa_free_multicast(mc->multicast.ib);
1637                         kfree(mc);
1638                 } else {
1639                         if (mc->igmp_joined) {
1640                                 struct rdma_dev_addr *dev_addr =
1641                                         &id_priv->id.route.addr.dev_addr;
1642                                 struct net_device *ndev = NULL;
1643
1644                                 if (dev_addr->bound_dev_if)
1645                                         ndev = dev_get_by_index(&init_net,
1646                                                                 dev_addr->bound_dev_if);
1647                                 if (ndev) {
1648                                         cma_igmp_send(ndev,
1649                                                       &mc->multicast.ib->rec.mgid,
1650                                                       false);
1651                                         dev_put(ndev);
1652                                 }
1653                         }
1654                         kref_put(&mc->mcref, release_mc);
1655                 }
1656         }
1657 }
1658
1659 void rdma_destroy_id(struct rdma_cm_id *id)
1660 {
1661         struct rdma_id_private *id_priv;
1662         enum rdma_cm_state state;
1663
1664         id_priv = container_of(id, struct rdma_id_private, id);
1665         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1666         cma_cancel_operation(id_priv, state);
1667
1668         /*
1669          * Wait for any active callback to finish.  New callbacks will find
1670          * the id_priv state set to destroying and abort.
1671          */
1672         mutex_lock(&id_priv->handler_mutex);
1673         mutex_unlock(&id_priv->handler_mutex);
1674
1675         if (id_priv->cma_dev) {
1676                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1677                         if (id_priv->cm_id.ib)
1678                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1679                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1680                         if (id_priv->cm_id.iw)
1681                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1682                 }
1683                 cma_leave_mc_groups(id_priv);
1684                 cma_release_dev(id_priv);
1685         }
1686
1687         cma_release_port(id_priv);
1688         cma_deref_id(id_priv);
1689         wait_for_completion(&id_priv->comp);
1690
1691         if (id_priv->internal_id)
1692                 cma_deref_id(id_priv->id.context);
1693
1694         kfree(id_priv->id.route.path_rec);
1695         put_net(id_priv->id.route.addr.dev_addr.net);
1696         kfree(id_priv);
1697 }
1698 EXPORT_SYMBOL(rdma_destroy_id);
1699
1700 static int cma_rep_recv(struct rdma_id_private *id_priv)
1701 {
1702         int ret;
1703
1704         ret = cma_modify_qp_rtr(id_priv, NULL);
1705         if (ret)
1706                 goto reject;
1707
1708         ret = cma_modify_qp_rts(id_priv, NULL);
1709         if (ret)
1710                 goto reject;
1711
1712         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1713         if (ret)
1714                 goto reject;
1715
1716         return 0;
1717 reject:
1718         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1719         cma_modify_qp_err(id_priv);
1720         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1721                        NULL, 0, NULL, 0);
1722         return ret;
1723 }
1724
1725 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1726                                    struct ib_cm_rep_event_param *rep_data,
1727                                    void *private_data)
1728 {
1729         event->param.conn.private_data = private_data;
1730         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1731         event->param.conn.responder_resources = rep_data->responder_resources;
1732         event->param.conn.initiator_depth = rep_data->initiator_depth;
1733         event->param.conn.flow_control = rep_data->flow_control;
1734         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1735         event->param.conn.srq = rep_data->srq;
1736         event->param.conn.qp_num = rep_data->remote_qpn;
1737 }
1738
1739 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1740 {
1741         struct rdma_id_private *id_priv = cm_id->context;
1742         struct rdma_cm_event event;
1743         int ret = 0;
1744
1745         mutex_lock(&id_priv->handler_mutex);
1746         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1747              id_priv->state != RDMA_CM_CONNECT) ||
1748             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1749              id_priv->state != RDMA_CM_DISCONNECT))
1750                 goto out;
1751
1752         memset(&event, 0, sizeof event);
1753         switch (ib_event->event) {
1754         case IB_CM_REQ_ERROR:
1755         case IB_CM_REP_ERROR:
1756                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1757                 event.status = -ETIMEDOUT;
1758                 break;
1759         case IB_CM_REP_RECEIVED:
1760                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1761                     (id_priv->id.qp_type != IB_QPT_UD))
1762                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1763                 if (id_priv->id.qp) {
1764                         event.status = cma_rep_recv(id_priv);
1765                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1766                                                      RDMA_CM_EVENT_ESTABLISHED;
1767                 } else {
1768                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1769                 }
1770                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1771                                        ib_event->private_data);
1772                 break;
1773         case IB_CM_RTU_RECEIVED:
1774         case IB_CM_USER_ESTABLISHED:
1775                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1776                 break;
1777         case IB_CM_DREQ_ERROR:
1778                 event.status = -ETIMEDOUT; /* fall through */
1779         case IB_CM_DREQ_RECEIVED:
1780         case IB_CM_DREP_RECEIVED:
1781                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1782                                    RDMA_CM_DISCONNECT))
1783                         goto out;
1784                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1785                 break;
1786         case IB_CM_TIMEWAIT_EXIT:
1787                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1788                 break;
1789         case IB_CM_MRA_RECEIVED:
1790                 /* ignore event */
1791                 goto out;
1792         case IB_CM_REJ_RECEIVED:
1793                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1794                                                                                 ib_event->param.rej_rcvd.reason));
1795                 cma_modify_qp_err(id_priv);
1796                 event.status = ib_event->param.rej_rcvd.reason;
1797                 event.event = RDMA_CM_EVENT_REJECTED;
1798                 event.param.conn.private_data = ib_event->private_data;
1799                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1800                 break;
1801         default:
1802                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1803                        ib_event->event);
1804                 goto out;
1805         }
1806
1807         ret = id_priv->id.event_handler(&id_priv->id, &event);
1808         if (ret) {
1809                 /* Destroy the CM ID by returning a non-zero value. */
1810                 id_priv->cm_id.ib = NULL;
1811                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1812                 mutex_unlock(&id_priv->handler_mutex);
1813                 rdma_destroy_id(&id_priv->id);
1814                 return ret;
1815         }
1816 out:
1817         mutex_unlock(&id_priv->handler_mutex);
1818         return ret;
1819 }
1820
1821 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1822                                                struct ib_cm_event *ib_event,
1823                                                struct net_device *net_dev)
1824 {
1825         struct rdma_id_private *id_priv;
1826         struct rdma_cm_id *id;
1827         struct rdma_route *rt;
1828         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1829         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1830         const __be64 service_id = sa_path_get_service_id(path);
1831         int ret;
1832
1833         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1834                             listen_id->event_handler, listen_id->context,
1835                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1836         if (IS_ERR(id))
1837                 return NULL;
1838
1839         id_priv = container_of(id, struct rdma_id_private, id);
1840         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1841                               (struct sockaddr *)&id->route.addr.dst_addr,
1842                               listen_id, ib_event, ss_family, service_id))
1843                 goto err;
1844
1845         rt = &id->route;
1846         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1847         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1848                                GFP_KERNEL);
1849         if (!rt->path_rec)
1850                 goto err;
1851
1852         rt->path_rec[0] = *path;
1853         if (rt->num_paths == 2)
1854                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1855
1856         if (net_dev) {
1857                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1858                 if (ret)
1859                         goto err;
1860         } else {
1861                 if (!cma_protocol_roce(listen_id) &&
1862                     cma_any_addr(cma_src_addr(id_priv))) {
1863                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1864                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1865                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1866                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1867                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1868                         if (ret)
1869                                 goto err;
1870                 }
1871         }
1872         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1873
1874         id_priv->state = RDMA_CM_CONNECT;
1875         return id_priv;
1876
1877 err:
1878         rdma_destroy_id(id);
1879         return NULL;
1880 }
1881
1882 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1883                                               struct ib_cm_event *ib_event,
1884                                               struct net_device *net_dev)
1885 {
1886         struct rdma_id_private *id_priv;
1887         struct rdma_cm_id *id;
1888         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1889         struct net *net = listen_id->route.addr.dev_addr.net;
1890         int ret;
1891
1892         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1893                             listen_id->ps, IB_QPT_UD);
1894         if (IS_ERR(id))
1895                 return NULL;
1896
1897         id_priv = container_of(id, struct rdma_id_private, id);
1898         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1899                               (struct sockaddr *)&id->route.addr.dst_addr,
1900                               listen_id, ib_event, ss_family,
1901                               ib_event->param.sidr_req_rcvd.service_id))
1902                 goto err;
1903
1904         if (net_dev) {
1905                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1906                 if (ret)
1907                         goto err;
1908         } else {
1909                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1910                         ret = cma_translate_addr(cma_src_addr(id_priv),
1911                                                  &id->route.addr.dev_addr);
1912                         if (ret)
1913                                 goto err;
1914                 }
1915         }
1916
1917         id_priv->state = RDMA_CM_CONNECT;
1918         return id_priv;
1919 err:
1920         rdma_destroy_id(id);
1921         return NULL;
1922 }
1923
1924 static void cma_set_req_event_data(struct rdma_cm_event *event,
1925                                    struct ib_cm_req_event_param *req_data,
1926                                    void *private_data, int offset)
1927 {
1928         event->param.conn.private_data = private_data + offset;
1929         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1930         event->param.conn.responder_resources = req_data->responder_resources;
1931         event->param.conn.initiator_depth = req_data->initiator_depth;
1932         event->param.conn.flow_control = req_data->flow_control;
1933         event->param.conn.retry_count = req_data->retry_count;
1934         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1935         event->param.conn.srq = req_data->srq;
1936         event->param.conn.qp_num = req_data->remote_qpn;
1937 }
1938
1939 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1940 {
1941         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1942                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1943                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1944                  (id->qp_type == IB_QPT_UD)) ||
1945                 (!id->qp_type));
1946 }
1947
1948 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1949 {
1950         struct rdma_id_private *listen_id, *conn_id = NULL;
1951         struct rdma_cm_event event;
1952         struct net_device *net_dev;
1953         int offset, ret;
1954
1955         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1956         if (IS_ERR(listen_id))
1957                 return PTR_ERR(listen_id);
1958
1959         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1960                 ret = -EINVAL;
1961                 goto net_dev_put;
1962         }
1963
1964         mutex_lock(&listen_id->handler_mutex);
1965         if (listen_id->state != RDMA_CM_LISTEN) {
1966                 ret = -ECONNABORTED;
1967                 goto err1;
1968         }
1969
1970         memset(&event, 0, sizeof event);
1971         offset = cma_user_data_offset(listen_id);
1972         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1973         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1974                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1975                 event.param.ud.private_data = ib_event->private_data + offset;
1976                 event.param.ud.private_data_len =
1977                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1978         } else {
1979                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1980                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1981                                        ib_event->private_data, offset);
1982         }
1983         if (!conn_id) {
1984                 ret = -ENOMEM;
1985                 goto err1;
1986         }
1987
1988         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1989         ret = cma_acquire_dev(conn_id, listen_id);
1990         if (ret)
1991                 goto err2;
1992
1993         conn_id->cm_id.ib = cm_id;
1994         cm_id->context = conn_id;
1995         cm_id->cm_handler = cma_ib_handler;
1996
1997         /*
1998          * Protect against the user destroying conn_id from another thread
1999          * until we're done accessing it.
2000          */
2001         atomic_inc(&conn_id->refcount);
2002         ret = conn_id->id.event_handler(&conn_id->id, &event);
2003         if (ret)
2004                 goto err3;
2005         /*
2006          * Acquire mutex to prevent user executing rdma_destroy_id()
2007          * while we're accessing the cm_id.
2008          */
2009         mutex_lock(&lock);
2010         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2011             (conn_id->id.qp_type != IB_QPT_UD))
2012                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2013         mutex_unlock(&lock);
2014         mutex_unlock(&conn_id->handler_mutex);
2015         mutex_unlock(&listen_id->handler_mutex);
2016         cma_deref_id(conn_id);
2017         if (net_dev)
2018                 dev_put(net_dev);
2019         return 0;
2020
2021 err3:
2022         cma_deref_id(conn_id);
2023         /* Destroy the CM ID by returning a non-zero value. */
2024         conn_id->cm_id.ib = NULL;
2025 err2:
2026         cma_exch(conn_id, RDMA_CM_DESTROYING);
2027         mutex_unlock(&conn_id->handler_mutex);
2028 err1:
2029         mutex_unlock(&listen_id->handler_mutex);
2030         if (conn_id)
2031                 rdma_destroy_id(&conn_id->id);
2032
2033 net_dev_put:
2034         if (net_dev)
2035                 dev_put(net_dev);
2036
2037         return ret;
2038 }
2039
2040 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2041 {
2042         if (addr->sa_family == AF_IB)
2043                 return ((struct sockaddr_ib *) addr)->sib_sid;
2044
2045         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2046 }
2047 EXPORT_SYMBOL(rdma_get_service_id);
2048
2049 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2050 {
2051         struct rdma_id_private *id_priv = iw_id->context;
2052         struct rdma_cm_event event;
2053         int ret = 0;
2054         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2055         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2056
2057         mutex_lock(&id_priv->handler_mutex);
2058         if (id_priv->state != RDMA_CM_CONNECT)
2059                 goto out;
2060
2061         memset(&event, 0, sizeof event);
2062         switch (iw_event->event) {
2063         case IW_CM_EVENT_CLOSE:
2064                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2065                 break;
2066         case IW_CM_EVENT_CONNECT_REPLY:
2067                 memcpy(cma_src_addr(id_priv), laddr,
2068                        rdma_addr_size(laddr));
2069                 memcpy(cma_dst_addr(id_priv), raddr,
2070                        rdma_addr_size(raddr));
2071                 switch (iw_event->status) {
2072                 case 0:
2073                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2074                         event.param.conn.initiator_depth = iw_event->ird;
2075                         event.param.conn.responder_resources = iw_event->ord;
2076                         break;
2077                 case -ECONNRESET:
2078                 case -ECONNREFUSED:
2079                         event.event = RDMA_CM_EVENT_REJECTED;
2080                         break;
2081                 case -ETIMEDOUT:
2082                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2083                         break;
2084                 default:
2085                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2086                         break;
2087                 }
2088                 break;
2089         case IW_CM_EVENT_ESTABLISHED:
2090                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2091                 event.param.conn.initiator_depth = iw_event->ird;
2092                 event.param.conn.responder_resources = iw_event->ord;
2093                 break;
2094         default:
2095                 BUG_ON(1);
2096         }
2097
2098         event.status = iw_event->status;
2099         event.param.conn.private_data = iw_event->private_data;
2100         event.param.conn.private_data_len = iw_event->private_data_len;
2101         ret = id_priv->id.event_handler(&id_priv->id, &event);
2102         if (ret) {
2103                 /* Destroy the CM ID by returning a non-zero value. */
2104                 id_priv->cm_id.iw = NULL;
2105                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2106                 mutex_unlock(&id_priv->handler_mutex);
2107                 rdma_destroy_id(&id_priv->id);
2108                 return ret;
2109         }
2110
2111 out:
2112         mutex_unlock(&id_priv->handler_mutex);
2113         return ret;
2114 }
2115
2116 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2117                                struct iw_cm_event *iw_event)
2118 {
2119         struct rdma_cm_id *new_cm_id;
2120         struct rdma_id_private *listen_id, *conn_id;
2121         struct rdma_cm_event event;
2122         int ret = -ECONNABORTED;
2123         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2124         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2125
2126         listen_id = cm_id->context;
2127
2128         mutex_lock(&listen_id->handler_mutex);
2129         if (listen_id->state != RDMA_CM_LISTEN)
2130                 goto out;
2131
2132         /* Create a new RDMA id for the new IW CM ID */
2133         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2134                                    listen_id->id.event_handler,
2135                                    listen_id->id.context,
2136                                    RDMA_PS_TCP, IB_QPT_RC);
2137         if (IS_ERR(new_cm_id)) {
2138                 ret = -ENOMEM;
2139                 goto out;
2140         }
2141         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2142         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2143         conn_id->state = RDMA_CM_CONNECT;
2144
2145         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2146         if (ret) {
2147                 mutex_unlock(&conn_id->handler_mutex);
2148                 rdma_destroy_id(new_cm_id);
2149                 goto out;
2150         }
2151
2152         ret = cma_acquire_dev(conn_id, listen_id);
2153         if (ret) {
2154                 mutex_unlock(&conn_id->handler_mutex);
2155                 rdma_destroy_id(new_cm_id);
2156                 goto out;
2157         }
2158
2159         conn_id->cm_id.iw = cm_id;
2160         cm_id->context = conn_id;
2161         cm_id->cm_handler = cma_iw_handler;
2162
2163         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2164         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2165
2166         memset(&event, 0, sizeof event);
2167         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2168         event.param.conn.private_data = iw_event->private_data;
2169         event.param.conn.private_data_len = iw_event->private_data_len;
2170         event.param.conn.initiator_depth = iw_event->ird;
2171         event.param.conn.responder_resources = iw_event->ord;
2172
2173         /*
2174          * Protect against the user destroying conn_id from another thread
2175          * until we're done accessing it.
2176          */
2177         atomic_inc(&conn_id->refcount);
2178         ret = conn_id->id.event_handler(&conn_id->id, &event);
2179         if (ret) {
2180                 /* User wants to destroy the CM ID */
2181                 conn_id->cm_id.iw = NULL;
2182                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2183                 mutex_unlock(&conn_id->handler_mutex);
2184                 cma_deref_id(conn_id);
2185                 rdma_destroy_id(&conn_id->id);
2186                 goto out;
2187         }
2188
2189         mutex_unlock(&conn_id->handler_mutex);
2190         cma_deref_id(conn_id);
2191
2192 out:
2193         mutex_unlock(&listen_id->handler_mutex);
2194         return ret;
2195 }
2196
2197 static int cma_ib_listen(struct rdma_id_private *id_priv)
2198 {
2199         struct sockaddr *addr;
2200         struct ib_cm_id *id;
2201         __be64 svc_id;
2202
2203         addr = cma_src_addr(id_priv);
2204         svc_id = rdma_get_service_id(&id_priv->id, addr);
2205         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2206         if (IS_ERR(id))
2207                 return PTR_ERR(id);
2208         id_priv->cm_id.ib = id;
2209
2210         return 0;
2211 }
2212
2213 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2214 {
2215         int ret;
2216         struct iw_cm_id *id;
2217
2218         id = iw_create_cm_id(id_priv->id.device,
2219                              iw_conn_req_handler,
2220                              id_priv);
2221         if (IS_ERR(id))
2222                 return PTR_ERR(id);
2223
2224         id->tos = id_priv->tos;
2225         id_priv->cm_id.iw = id;
2226
2227         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2228                rdma_addr_size(cma_src_addr(id_priv)));
2229
2230         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2231
2232         if (ret) {
2233                 iw_destroy_cm_id(id_priv->cm_id.iw);
2234                 id_priv->cm_id.iw = NULL;
2235         }
2236
2237         return ret;
2238 }
2239
2240 static int cma_listen_handler(struct rdma_cm_id *id,
2241                               struct rdma_cm_event *event)
2242 {
2243         struct rdma_id_private *id_priv = id->context;
2244
2245         id->context = id_priv->id.context;
2246         id->event_handler = id_priv->id.event_handler;
2247         return id_priv->id.event_handler(id, event);
2248 }
2249
2250 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2251                               struct cma_device *cma_dev)
2252 {
2253         struct rdma_id_private *dev_id_priv;
2254         struct rdma_cm_id *id;
2255         struct net *net = id_priv->id.route.addr.dev_addr.net;
2256         int ret;
2257
2258         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2259                 return;
2260
2261         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2262                             id_priv->id.qp_type);
2263         if (IS_ERR(id))
2264                 return;
2265
2266         dev_id_priv = container_of(id, struct rdma_id_private, id);
2267
2268         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2269         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2270                rdma_addr_size(cma_src_addr(id_priv)));
2271
2272         _cma_attach_to_dev(dev_id_priv, cma_dev);
2273         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2274         atomic_inc(&id_priv->refcount);
2275         dev_id_priv->internal_id = 1;
2276         dev_id_priv->afonly = id_priv->afonly;
2277
2278         ret = rdma_listen(id, id_priv->backlog);
2279         if (ret)
2280                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2281                         ret, cma_dev->device->name);
2282 }
2283
2284 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2285 {
2286         struct cma_device *cma_dev;
2287
2288         mutex_lock(&lock);
2289         list_add_tail(&id_priv->list, &listen_any_list);
2290         list_for_each_entry(cma_dev, &dev_list, list)
2291                 cma_listen_on_dev(id_priv, cma_dev);
2292         mutex_unlock(&lock);
2293 }
2294
2295 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2296 {
2297         struct rdma_id_private *id_priv;
2298
2299         id_priv = container_of(id, struct rdma_id_private, id);
2300         id_priv->tos = (u8) tos;
2301         id_priv->tos_set = true;
2302 }
2303 EXPORT_SYMBOL(rdma_set_service_type);
2304
2305 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2306                               void *context)
2307 {
2308         struct cma_work *work = context;
2309         struct rdma_route *route;
2310
2311         route = &work->id->id.route;
2312
2313         if (!status) {
2314                 route->num_paths = 1;
2315                 *route->path_rec = *path_rec;
2316         } else {
2317                 work->old_state = RDMA_CM_ROUTE_QUERY;
2318                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2319                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2320                 work->event.status = status;
2321                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2322                                      status);
2323         }
2324
2325         queue_work(cma_wq, &work->work);
2326 }
2327
2328 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2329                               struct cma_work *work)
2330 {
2331         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2332         struct sa_path_rec path_rec;
2333         ib_sa_comp_mask comp_mask;
2334         struct sockaddr_in6 *sin6;
2335         struct sockaddr_ib *sib;
2336
2337         memset(&path_rec, 0, sizeof path_rec);
2338
2339         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2340                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2341         else
2342                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2343         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2344         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2345         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2346         path_rec.numb_path = 1;
2347         path_rec.reversible = 1;
2348         sa_path_set_service_id(&path_rec,
2349                                rdma_get_service_id(&id_priv->id,
2350                                                    cma_dst_addr(id_priv)));
2351
2352         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2353                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2354                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2355
2356         switch (cma_family(id_priv)) {
2357         case AF_INET:
2358                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2359                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2360                 break;
2361         case AF_INET6:
2362                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2363                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2364                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2365                 break;
2366         case AF_IB:
2367                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2368                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2369                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2370                 break;
2371         }
2372
2373         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2374                                                id_priv->id.port_num, &path_rec,
2375                                                comp_mask, timeout_ms,
2376                                                GFP_KERNEL, cma_query_handler,
2377                                                work, &id_priv->query);
2378
2379         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2380 }
2381
2382 static void cma_work_handler(struct work_struct *_work)
2383 {
2384         struct cma_work *work = container_of(_work, struct cma_work, work);
2385         struct rdma_id_private *id_priv = work->id;
2386         int destroy = 0;
2387
2388         mutex_lock(&id_priv->handler_mutex);
2389         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2390                 goto out;
2391
2392         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2393                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2394                 destroy = 1;
2395         }
2396 out:
2397         mutex_unlock(&id_priv->handler_mutex);
2398         cma_deref_id(id_priv);
2399         if (destroy)
2400                 rdma_destroy_id(&id_priv->id);
2401         kfree(work);
2402 }
2403
2404 static void cma_ndev_work_handler(struct work_struct *_work)
2405 {
2406         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2407         struct rdma_id_private *id_priv = work->id;
2408         int destroy = 0;
2409
2410         mutex_lock(&id_priv->handler_mutex);
2411         if (id_priv->state == RDMA_CM_DESTROYING ||
2412             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2413                 goto out;
2414
2415         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2416                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2417                 destroy = 1;
2418         }
2419
2420 out:
2421         mutex_unlock(&id_priv->handler_mutex);
2422         cma_deref_id(id_priv);
2423         if (destroy)
2424                 rdma_destroy_id(&id_priv->id);
2425         kfree(work);
2426 }
2427
2428 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2429 {
2430         struct rdma_route *route = &id_priv->id.route;
2431         struct cma_work *work;
2432         int ret;
2433
2434         work = kzalloc(sizeof *work, GFP_KERNEL);
2435         if (!work)
2436                 return -ENOMEM;
2437
2438         work->id = id_priv;
2439         INIT_WORK(&work->work, cma_work_handler);
2440         work->old_state = RDMA_CM_ROUTE_QUERY;
2441         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2442         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2443
2444         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2445         if (!route->path_rec) {
2446                 ret = -ENOMEM;
2447                 goto err1;
2448         }
2449
2450         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2451         if (ret)
2452                 goto err2;
2453
2454         return 0;
2455 err2:
2456         kfree(route->path_rec);
2457         route->path_rec = NULL;
2458 err1:
2459         kfree(work);
2460         return ret;
2461 }
2462
2463 int rdma_set_ib_paths(struct rdma_cm_id *id,
2464                       struct sa_path_rec *path_rec, int num_paths)
2465 {
2466         struct rdma_id_private *id_priv;
2467         int ret;
2468
2469         id_priv = container_of(id, struct rdma_id_private, id);
2470         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2471                            RDMA_CM_ROUTE_RESOLVED))
2472                 return -EINVAL;
2473
2474         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2475                                      GFP_KERNEL);
2476         if (!id->route.path_rec) {
2477                 ret = -ENOMEM;
2478                 goto err;
2479         }
2480
2481         id->route.num_paths = num_paths;
2482         return 0;
2483 err:
2484         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2485         return ret;
2486 }
2487 EXPORT_SYMBOL(rdma_set_ib_paths);
2488
2489 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2490 {
2491         struct cma_work *work;
2492
2493         work = kzalloc(sizeof *work, GFP_KERNEL);
2494         if (!work)
2495                 return -ENOMEM;
2496
2497         work->id = id_priv;
2498         INIT_WORK(&work->work, cma_work_handler);
2499         work->old_state = RDMA_CM_ROUTE_QUERY;
2500         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2501         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2502         queue_work(cma_wq, &work->work);
2503         return 0;
2504 }
2505
2506 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2507 {
2508         int prio;
2509         struct net_device *dev;
2510
2511         prio = rt_tos2priority(tos);
2512         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2513         if (dev->num_tc)
2514                 return netdev_get_prio_tc_map(dev, prio);
2515
2516 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2517         if (is_vlan_dev(ndev))
2518                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2519                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2520 #endif
2521         return 0;
2522 }
2523
2524 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2525                                            unsigned long supported_gids,
2526                                            enum ib_gid_type default_gid)
2527 {
2528         if ((network_type == RDMA_NETWORK_IPV4 ||
2529              network_type == RDMA_NETWORK_IPV6) &&
2530             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2531                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2532
2533         return default_gid;
2534 }
2535
2536 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2537 {
2538         struct rdma_route *route = &id_priv->id.route;
2539         struct rdma_addr *addr = &route->addr;
2540         struct cma_work *work;
2541         int ret;
2542         struct net_device *ndev = NULL;
2543         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
2544         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2545                                         rdma_start_port(id_priv->cma_dev->device)];
2546         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2547
2548
2549         work = kzalloc(sizeof *work, GFP_KERNEL);
2550         if (!work)
2551                 return -ENOMEM;
2552
2553         work->id = id_priv;
2554         INIT_WORK(&work->work, cma_work_handler);
2555
2556         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2557         if (!route->path_rec) {
2558                 ret = -ENOMEM;
2559                 goto err1;
2560         }
2561
2562         route->num_paths = 1;
2563
2564         if (addr->dev_addr.bound_dev_if) {
2565                 unsigned long supported_gids;
2566
2567                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2568                 if (!ndev) {
2569                         ret = -ENODEV;
2570                         goto err2;
2571                 }
2572
2573                 if (ndev->flags & IFF_LOOPBACK) {
2574                         dev_put(ndev);
2575                         if (!id_priv->id.device->get_netdev) {
2576                                 ret = -EOPNOTSUPP;
2577                                 goto err2;
2578                         }
2579
2580                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2581                                                               id_priv->id.port_num);
2582                         if (!ndev) {
2583                                 ret = -ENODEV;
2584                                 goto err2;
2585                         }
2586                 }
2587
2588                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2589                                                             id_priv->id.port_num);
2590                 gid_type = cma_route_gid_type(addr->dev_addr.network,
2591                                               supported_gids,
2592                                               id_priv->gid_type);
2593                 route->path_rec->rec_type =
2594                         sa_conv_gid_to_pathrec_type(gid_type);
2595                 sa_path_set_ndev(route->path_rec, &init_net);
2596                 sa_path_set_ifindex(route->path_rec, ndev->ifindex);
2597         }
2598         if (!ndev) {
2599                 ret = -ENODEV;
2600                 goto err2;
2601         }
2602
2603         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2604
2605         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2606                     &route->path_rec->sgid);
2607         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2608                     &route->path_rec->dgid);
2609
2610         /* Use the hint from IP Stack to select GID Type */
2611         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2612                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2613         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2614
2615         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2616                 /* TODO: get the hoplimit from the inet/inet6 device */
2617                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2618         else
2619                 route->path_rec->hop_limit = 1;
2620         route->path_rec->reversible = 1;
2621         route->path_rec->pkey = cpu_to_be16(0xffff);
2622         route->path_rec->mtu_selector = IB_SA_EQ;
2623         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2624         route->path_rec->traffic_class = tos;
2625         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2626         route->path_rec->rate_selector = IB_SA_EQ;
2627         route->path_rec->rate = iboe_get_rate(ndev);
2628         dev_put(ndev);
2629         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2630         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2631         if (!route->path_rec->mtu) {
2632                 ret = -EINVAL;
2633                 goto err2;
2634         }
2635
2636         work->old_state = RDMA_CM_ROUTE_QUERY;
2637         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2638         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2639         work->event.status = 0;
2640
2641         queue_work(cma_wq, &work->work);
2642
2643         return 0;
2644
2645 err2:
2646         kfree(route->path_rec);
2647         route->path_rec = NULL;
2648 err1:
2649         kfree(work);
2650         return ret;
2651 }
2652
2653 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2654 {
2655         struct rdma_id_private *id_priv;
2656         int ret;
2657
2658         id_priv = container_of(id, struct rdma_id_private, id);
2659         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2660                 return -EINVAL;
2661
2662         atomic_inc(&id_priv->refcount);
2663         if (rdma_cap_ib_sa(id->device, id->port_num))
2664                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2665         else if (rdma_protocol_roce(id->device, id->port_num))
2666                 ret = cma_resolve_iboe_route(id_priv);
2667         else if (rdma_protocol_iwarp(id->device, id->port_num))
2668                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2669         else
2670                 ret = -ENOSYS;
2671
2672         if (ret)
2673                 goto err;
2674
2675         return 0;
2676 err:
2677         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2678         cma_deref_id(id_priv);
2679         return ret;
2680 }
2681 EXPORT_SYMBOL(rdma_resolve_route);
2682
2683 static void cma_set_loopback(struct sockaddr *addr)
2684 {
2685         switch (addr->sa_family) {
2686         case AF_INET:
2687                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2688                 break;
2689         case AF_INET6:
2690                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2691                               0, 0, 0, htonl(1));
2692                 break;
2693         default:
2694                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2695                             0, 0, 0, htonl(1));
2696                 break;
2697         }
2698 }
2699
2700 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2701 {
2702         struct cma_device *cma_dev, *cur_dev;
2703         union ib_gid gid;
2704         enum ib_port_state port_state;
2705         u16 pkey;
2706         int ret;
2707         u8 p;
2708
2709         cma_dev = NULL;
2710         mutex_lock(&lock);
2711         list_for_each_entry(cur_dev, &dev_list, list) {
2712                 if (cma_family(id_priv) == AF_IB &&
2713                     !rdma_cap_ib_cm(cur_dev->device, 1))
2714                         continue;
2715
2716                 if (!cma_dev)
2717                         cma_dev = cur_dev;
2718
2719                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2720                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2721                             port_state == IB_PORT_ACTIVE) {
2722                                 cma_dev = cur_dev;
2723                                 goto port_found;
2724                         }
2725                 }
2726         }
2727
2728         if (!cma_dev) {
2729                 ret = -ENODEV;
2730                 goto out;
2731         }
2732
2733         p = 1;
2734
2735 port_found:
2736         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2737         if (ret)
2738                 goto out;
2739
2740         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2741         if (ret)
2742                 goto out;
2743
2744         id_priv->id.route.addr.dev_addr.dev_type =
2745                 (rdma_protocol_ib(cma_dev->device, p)) ?
2746                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2747
2748         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2749         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2750         id_priv->id.port_num = p;
2751         cma_attach_to_dev(id_priv, cma_dev);
2752         cma_set_loopback(cma_src_addr(id_priv));
2753 out:
2754         mutex_unlock(&lock);
2755         return ret;
2756 }
2757
2758 static void addr_handler(int status, struct sockaddr *src_addr,
2759                          struct rdma_dev_addr *dev_addr, void *context)
2760 {
2761         struct rdma_id_private *id_priv = context;
2762         struct rdma_cm_event event;
2763
2764         memset(&event, 0, sizeof event);
2765         mutex_lock(&id_priv->handler_mutex);
2766         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2767                            RDMA_CM_ADDR_RESOLVED))
2768                 goto out;
2769
2770         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2771         if (!status && !id_priv->cma_dev) {
2772                 status = cma_acquire_dev(id_priv, NULL);
2773                 if (status)
2774                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2775                                              status);
2776         } else {
2777                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2778         }
2779
2780         if (status) {
2781                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2782                                    RDMA_CM_ADDR_BOUND))
2783                         goto out;
2784                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2785                 event.status = status;
2786         } else
2787                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2788
2789         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2790                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2791                 mutex_unlock(&id_priv->handler_mutex);
2792                 cma_deref_id(id_priv);
2793                 rdma_destroy_id(&id_priv->id);
2794                 return;
2795         }
2796 out:
2797         mutex_unlock(&id_priv->handler_mutex);
2798         cma_deref_id(id_priv);
2799 }
2800
2801 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2802 {
2803         struct cma_work *work;
2804         union ib_gid gid;
2805         int ret;
2806
2807         work = kzalloc(sizeof *work, GFP_KERNEL);
2808         if (!work)
2809                 return -ENOMEM;
2810
2811         if (!id_priv->cma_dev) {
2812                 ret = cma_bind_loopback(id_priv);
2813                 if (ret)
2814                         goto err;
2815         }
2816
2817         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2818         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2819
2820         work->id = id_priv;
2821         INIT_WORK(&work->work, cma_work_handler);
2822         work->old_state = RDMA_CM_ADDR_QUERY;
2823         work->new_state = RDMA_CM_ADDR_RESOLVED;
2824         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2825         queue_work(cma_wq, &work->work);
2826         return 0;
2827 err:
2828         kfree(work);
2829         return ret;
2830 }
2831
2832 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2833 {
2834         struct cma_work *work;
2835         int ret;
2836
2837         work = kzalloc(sizeof *work, GFP_KERNEL);
2838         if (!work)
2839                 return -ENOMEM;
2840
2841         if (!id_priv->cma_dev) {
2842                 ret = cma_resolve_ib_dev(id_priv);
2843                 if (ret)
2844                         goto err;
2845         }
2846
2847         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2848                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2849
2850         work->id = id_priv;
2851         INIT_WORK(&work->work, cma_work_handler);
2852         work->old_state = RDMA_CM_ADDR_QUERY;
2853         work->new_state = RDMA_CM_ADDR_RESOLVED;
2854         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2855         queue_work(cma_wq, &work->work);
2856         return 0;
2857 err:
2858         kfree(work);
2859         return ret;
2860 }
2861
2862 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2863                          struct sockaddr *dst_addr)
2864 {
2865         if (!src_addr || !src_addr->sa_family) {
2866                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2867                 src_addr->sa_family = dst_addr->sa_family;
2868                 if (IS_ENABLED(CONFIG_IPV6) &&
2869                     dst_addr->sa_family == AF_INET6) {
2870                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2871                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2872                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2873                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2874                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2875                 } else if (dst_addr->sa_family == AF_IB) {
2876                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2877                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2878                 }
2879         }
2880         return rdma_bind_addr(id, src_addr);
2881 }
2882
2883 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2884                       struct sockaddr *dst_addr, int timeout_ms)
2885 {
2886         struct rdma_id_private *id_priv;
2887         int ret;
2888
2889         id_priv = container_of(id, struct rdma_id_private, id);
2890         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2891         if (id_priv->state == RDMA_CM_IDLE) {
2892                 ret = cma_bind_addr(id, src_addr, dst_addr);
2893                 if (ret) {
2894                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2895                         return ret;
2896                 }
2897         }
2898
2899         if (cma_family(id_priv) != dst_addr->sa_family) {
2900                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2901                 return -EINVAL;
2902         }
2903
2904         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2905                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2906                 return -EINVAL;
2907         }
2908
2909         atomic_inc(&id_priv->refcount);
2910         if (cma_any_addr(dst_addr)) {
2911                 ret = cma_resolve_loopback(id_priv);
2912         } else {
2913                 if (dst_addr->sa_family == AF_IB) {
2914                         ret = cma_resolve_ib_addr(id_priv);
2915                 } else {
2916                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2917                                               dst_addr, &id->route.addr.dev_addr,
2918                                               timeout_ms, addr_handler, id_priv);
2919                 }
2920         }
2921         if (ret)
2922                 goto err;
2923
2924         return 0;
2925 err:
2926         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2927         cma_deref_id(id_priv);
2928         return ret;
2929 }
2930 EXPORT_SYMBOL(rdma_resolve_addr);
2931
2932 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2933 {
2934         struct rdma_id_private *id_priv;
2935         unsigned long flags;
2936         int ret;
2937
2938         id_priv = container_of(id, struct rdma_id_private, id);
2939         spin_lock_irqsave(&id_priv->lock, flags);
2940         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2941                 id_priv->reuseaddr = reuse;
2942                 ret = 0;
2943         } else {
2944                 ret = -EINVAL;
2945         }
2946         spin_unlock_irqrestore(&id_priv->lock, flags);
2947         return ret;
2948 }
2949 EXPORT_SYMBOL(rdma_set_reuseaddr);
2950
2951 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2952 {
2953         struct rdma_id_private *id_priv;
2954         unsigned long flags;
2955         int ret;
2956
2957         id_priv = container_of(id, struct rdma_id_private, id);
2958         spin_lock_irqsave(&id_priv->lock, flags);
2959         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2960                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2961                 id_priv->afonly = afonly;
2962                 ret = 0;
2963         } else {
2964                 ret = -EINVAL;
2965         }
2966         spin_unlock_irqrestore(&id_priv->lock, flags);
2967         return ret;
2968 }
2969 EXPORT_SYMBOL(rdma_set_afonly);
2970
2971 static void cma_bind_port(struct rdma_bind_list *bind_list,
2972                           struct rdma_id_private *id_priv)
2973 {
2974         struct sockaddr *addr;
2975         struct sockaddr_ib *sib;
2976         u64 sid, mask;
2977         __be16 port;
2978
2979         addr = cma_src_addr(id_priv);
2980         port = htons(bind_list->port);
2981
2982         switch (addr->sa_family) {
2983         case AF_INET:
2984                 ((struct sockaddr_in *) addr)->sin_port = port;
2985                 break;
2986         case AF_INET6:
2987                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2988                 break;
2989         case AF_IB:
2990                 sib = (struct sockaddr_ib *) addr;
2991                 sid = be64_to_cpu(sib->sib_sid);
2992                 mask = be64_to_cpu(sib->sib_sid_mask);
2993                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2994                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2995                 break;
2996         }
2997         id_priv->bind_list = bind_list;
2998         hlist_add_head(&id_priv->node, &bind_list->owners);
2999 }
3000
3001 static int cma_alloc_port(enum rdma_port_space ps,
3002                           struct rdma_id_private *id_priv, unsigned short snum)
3003 {
3004         struct rdma_bind_list *bind_list;
3005         int ret;
3006
3007         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3008         if (!bind_list)
3009                 return -ENOMEM;
3010
3011         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3012                            snum);
3013         if (ret < 0)
3014                 goto err;
3015
3016         bind_list->ps = ps;
3017         bind_list->port = (unsigned short)ret;
3018         cma_bind_port(bind_list, id_priv);
3019         return 0;
3020 err:
3021         kfree(bind_list);
3022         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3023 }
3024
3025 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3026                               struct rdma_id_private *id_priv)
3027 {
3028         struct rdma_id_private *cur_id;
3029         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3030         struct sockaddr  *saddr = cma_src_addr(id_priv);
3031         __be16 dport = cma_port(daddr);
3032
3033         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3034                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3035                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3036                 __be16 cur_dport = cma_port(cur_daddr);
3037
3038                 if (id_priv == cur_id)
3039                         continue;
3040
3041                 /* different dest port -> unique */
3042                 if (!cma_any_port(cur_daddr) &&
3043                     (dport != cur_dport))
3044                         continue;
3045
3046                 /* different src address -> unique */
3047                 if (!cma_any_addr(saddr) &&
3048                     !cma_any_addr(cur_saddr) &&
3049                     cma_addr_cmp(saddr, cur_saddr))
3050                         continue;
3051
3052                 /* different dst address -> unique */
3053                 if (!cma_any_addr(cur_daddr) &&
3054                     cma_addr_cmp(daddr, cur_daddr))
3055                         continue;
3056
3057                 return -EADDRNOTAVAIL;
3058         }
3059         return 0;
3060 }
3061
3062 static int cma_alloc_any_port(enum rdma_port_space ps,
3063                               struct rdma_id_private *id_priv)
3064 {
3065         static unsigned int last_used_port;
3066         int low, high, remaining;
3067         unsigned int rover;
3068         struct net *net = id_priv->id.route.addr.dev_addr.net;
3069
3070         inet_get_local_port_range(net, &low, &high);
3071         remaining = (high - low) + 1;
3072         rover = prandom_u32() % remaining + low;
3073 retry:
3074         if (last_used_port != rover) {
3075                 struct rdma_bind_list *bind_list;
3076                 int ret;
3077
3078                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3079
3080                 if (!bind_list) {
3081                         ret = cma_alloc_port(ps, id_priv, rover);
3082                 } else {
3083                         ret = cma_port_is_unique(bind_list, id_priv);
3084                         if (!ret)
3085                                 cma_bind_port(bind_list, id_priv);
3086                 }
3087                 /*
3088                  * Remember previously used port number in order to avoid
3089                  * re-using same port immediately after it is closed.
3090                  */
3091                 if (!ret)
3092                         last_used_port = rover;
3093                 if (ret != -EADDRNOTAVAIL)
3094                         return ret;
3095         }
3096         if (--remaining) {
3097                 rover++;
3098                 if ((rover < low) || (rover > high))
3099                         rover = low;
3100                 goto retry;
3101         }
3102         return -EADDRNOTAVAIL;
3103 }
3104
3105 /*
3106  * Check that the requested port is available.  This is called when trying to
3107  * bind to a specific port, or when trying to listen on a bound port.  In
3108  * the latter case, the provided id_priv may already be on the bind_list, but
3109  * we still need to check that it's okay to start listening.
3110  */
3111 static int cma_check_port(struct rdma_bind_list *bind_list,
3112                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3113 {
3114         struct rdma_id_private *cur_id;
3115         struct sockaddr *addr, *cur_addr;
3116
3117         addr = cma_src_addr(id_priv);
3118         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3119                 if (id_priv == cur_id)
3120                         continue;
3121
3122                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3123                     cur_id->reuseaddr)
3124                         continue;
3125
3126                 cur_addr = cma_src_addr(cur_id);
3127                 if (id_priv->afonly && cur_id->afonly &&
3128                     (addr->sa_family != cur_addr->sa_family))
3129                         continue;
3130
3131                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3132                         return -EADDRNOTAVAIL;
3133
3134                 if (!cma_addr_cmp(addr, cur_addr))
3135                         return -EADDRINUSE;
3136         }
3137         return 0;
3138 }
3139
3140 static int cma_use_port(enum rdma_port_space ps,
3141                         struct rdma_id_private *id_priv)
3142 {
3143         struct rdma_bind_list *bind_list;
3144         unsigned short snum;
3145         int ret;
3146
3147         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3148         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3149                 return -EACCES;
3150
3151         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3152         if (!bind_list) {
3153                 ret = cma_alloc_port(ps, id_priv, snum);
3154         } else {
3155                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3156                 if (!ret)
3157                         cma_bind_port(bind_list, id_priv);
3158         }
3159         return ret;
3160 }
3161
3162 static int cma_bind_listen(struct rdma_id_private *id_priv)
3163 {
3164         struct rdma_bind_list *bind_list = id_priv->bind_list;
3165         int ret = 0;
3166
3167         mutex_lock(&lock);
3168         if (bind_list->owners.first->next)
3169                 ret = cma_check_port(bind_list, id_priv, 0);
3170         mutex_unlock(&lock);
3171         return ret;
3172 }
3173
3174 static enum rdma_port_space cma_select_inet_ps(
3175                 struct rdma_id_private *id_priv)
3176 {
3177         switch (id_priv->id.ps) {
3178         case RDMA_PS_TCP:
3179         case RDMA_PS_UDP:
3180         case RDMA_PS_IPOIB:
3181         case RDMA_PS_IB:
3182                 return id_priv->id.ps;
3183         default:
3184
3185                 return 0;
3186         }
3187 }
3188
3189 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3190 {
3191         enum rdma_port_space ps = 0;
3192         struct sockaddr_ib *sib;
3193         u64 sid_ps, mask, sid;
3194
3195         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3196         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3197         sid = be64_to_cpu(sib->sib_sid) & mask;
3198
3199         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3200                 sid_ps = RDMA_IB_IP_PS_IB;
3201                 ps = RDMA_PS_IB;
3202         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3203                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3204                 sid_ps = RDMA_IB_IP_PS_TCP;
3205                 ps = RDMA_PS_TCP;
3206         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3207                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3208                 sid_ps = RDMA_IB_IP_PS_UDP;
3209                 ps = RDMA_PS_UDP;
3210         }
3211
3212         if (ps) {
3213                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3214                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3215                                                 be64_to_cpu(sib->sib_sid_mask));
3216         }
3217         return ps;
3218 }
3219
3220 static int cma_get_port(struct rdma_id_private *id_priv)
3221 {
3222         enum rdma_port_space ps;
3223         int ret;
3224
3225         if (cma_family(id_priv) != AF_IB)
3226                 ps = cma_select_inet_ps(id_priv);
3227         else
3228                 ps = cma_select_ib_ps(id_priv);
3229         if (!ps)
3230                 return -EPROTONOSUPPORT;
3231
3232         mutex_lock(&lock);
3233         if (cma_any_port(cma_src_addr(id_priv)))
3234                 ret = cma_alloc_any_port(ps, id_priv);
3235         else
3236                 ret = cma_use_port(ps, id_priv);
3237         mutex_unlock(&lock);
3238
3239         return ret;
3240 }
3241
3242 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3243                                struct sockaddr *addr)
3244 {
3245 #if IS_ENABLED(CONFIG_IPV6)
3246         struct sockaddr_in6 *sin6;
3247
3248         if (addr->sa_family != AF_INET6)
3249                 return 0;
3250
3251         sin6 = (struct sockaddr_in6 *) addr;
3252
3253         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3254                 return 0;
3255
3256         if (!sin6->sin6_scope_id)
3257                         return -EINVAL;
3258
3259         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3260 #endif
3261         return 0;
3262 }
3263
3264 int rdma_listen(struct rdma_cm_id *id, int backlog)
3265 {
3266         struct rdma_id_private *id_priv;
3267         int ret;
3268
3269         id_priv = container_of(id, struct rdma_id_private, id);
3270         if (id_priv->state == RDMA_CM_IDLE) {
3271                 id->route.addr.src_addr.ss_family = AF_INET;
3272                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3273                 if (ret)
3274                         return ret;
3275         }
3276
3277         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3278                 return -EINVAL;
3279
3280         if (id_priv->reuseaddr) {
3281                 ret = cma_bind_listen(id_priv);
3282                 if (ret)
3283                         goto err;
3284         }
3285
3286         id_priv->backlog = backlog;
3287         if (id->device) {
3288                 if (rdma_cap_ib_cm(id->device, 1)) {
3289                         ret = cma_ib_listen(id_priv);
3290                         if (ret)
3291                                 goto err;
3292                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3293                         ret = cma_iw_listen(id_priv, backlog);
3294                         if (ret)
3295                                 goto err;
3296                 } else {
3297                         ret = -ENOSYS;
3298                         goto err;
3299                 }
3300         } else
3301                 cma_listen_on_all(id_priv);
3302
3303         return 0;
3304 err:
3305         id_priv->backlog = 0;
3306         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3307         return ret;
3308 }
3309 EXPORT_SYMBOL(rdma_listen);
3310
3311 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3312 {
3313         struct rdma_id_private *id_priv;
3314         int ret;
3315         struct sockaddr  *daddr;
3316
3317         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3318             addr->sa_family != AF_IB)
3319                 return -EAFNOSUPPORT;
3320
3321         id_priv = container_of(id, struct rdma_id_private, id);
3322         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3323                 return -EINVAL;
3324
3325         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3326         if (ret)
3327                 goto err1;
3328
3329         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3330         if (!cma_any_addr(addr)) {
3331                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3332                 if (ret)
3333                         goto err1;
3334
3335                 ret = cma_acquire_dev(id_priv, NULL);
3336                 if (ret)
3337                         goto err1;
3338         }
3339
3340         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3341                 if (addr->sa_family == AF_INET)
3342                         id_priv->afonly = 1;
3343 #if IS_ENABLED(CONFIG_IPV6)
3344                 else if (addr->sa_family == AF_INET6) {
3345                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3346
3347                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3348                 }
3349 #endif
3350         }
3351         ret = cma_get_port(id_priv);
3352         if (ret)
3353                 goto err2;
3354
3355         daddr = cma_dst_addr(id_priv);
3356         daddr->sa_family = addr->sa_family;
3357
3358         return 0;
3359 err2:
3360         if (id_priv->cma_dev)
3361                 cma_release_dev(id_priv);
3362 err1:
3363         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3364         return ret;
3365 }
3366 EXPORT_SYMBOL(rdma_bind_addr);
3367
3368 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3369 {
3370         struct cma_hdr *cma_hdr;
3371
3372         cma_hdr = hdr;
3373         cma_hdr->cma_version = CMA_VERSION;
3374         if (cma_family(id_priv) == AF_INET) {
3375                 struct sockaddr_in *src4, *dst4;
3376
3377                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3378                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3379
3380                 cma_set_ip_ver(cma_hdr, 4);
3381                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3382                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3383                 cma_hdr->port = src4->sin_port;
3384         } else if (cma_family(id_priv) == AF_INET6) {
3385                 struct sockaddr_in6 *src6, *dst6;
3386
3387                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3388                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3389
3390                 cma_set_ip_ver(cma_hdr, 6);
3391                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3392                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3393                 cma_hdr->port = src6->sin6_port;
3394         }
3395         return 0;
3396 }
3397
3398 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3399                                 struct ib_cm_event *ib_event)
3400 {
3401         struct rdma_id_private *id_priv = cm_id->context;
3402         struct rdma_cm_event event;
3403         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3404         int ret = 0;
3405
3406         mutex_lock(&id_priv->handler_mutex);
3407         if (id_priv->state != RDMA_CM_CONNECT)
3408                 goto out;
3409
3410         memset(&event, 0, sizeof event);
3411         switch (ib_event->event) {
3412         case IB_CM_SIDR_REQ_ERROR:
3413                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3414                 event.status = -ETIMEDOUT;
3415                 break;
3416         case IB_CM_SIDR_REP_RECEIVED:
3417                 event.param.ud.private_data = ib_event->private_data;
3418                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3419                 if (rep->status != IB_SIDR_SUCCESS) {
3420                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3421                         event.status = ib_event->param.sidr_rep_rcvd.status;
3422                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3423                                              event.status);
3424                         break;
3425                 }
3426                 ret = cma_set_qkey(id_priv, rep->qkey);
3427                 if (ret) {
3428                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3429                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3430                         event.status = ret;
3431                         break;
3432                 }
3433                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3434                                      id_priv->id.route.path_rec,
3435                                      &event.param.ud.ah_attr);
3436                 event.param.ud.qp_num = rep->qpn;
3437                 event.param.ud.qkey = rep->qkey;
3438                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3439                 event.status = 0;
3440                 break;
3441         default:
3442                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3443                        ib_event->event);
3444                 goto out;
3445         }
3446
3447         ret = id_priv->id.event_handler(&id_priv->id, &event);
3448         if (ret) {
3449                 /* Destroy the CM ID by returning a non-zero value. */
3450                 id_priv->cm_id.ib = NULL;
3451                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3452                 mutex_unlock(&id_priv->handler_mutex);
3453                 rdma_destroy_id(&id_priv->id);
3454                 return ret;
3455         }
3456 out:
3457         mutex_unlock(&id_priv->handler_mutex);
3458         return ret;
3459 }
3460
3461 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3462                               struct rdma_conn_param *conn_param)
3463 {
3464         struct ib_cm_sidr_req_param req;
3465         struct ib_cm_id *id;
3466         void *private_data;
3467         int offset, ret;
3468
3469         memset(&req, 0, sizeof req);
3470         offset = cma_user_data_offset(id_priv);
3471         req.private_data_len = offset + conn_param->private_data_len;
3472         if (req.private_data_len < conn_param->private_data_len)
3473                 return -EINVAL;
3474
3475         if (req.private_data_len) {
3476                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3477                 if (!private_data)
3478                         return -ENOMEM;
3479         } else {
3480                 private_data = NULL;
3481         }
3482
3483         if (conn_param->private_data && conn_param->private_data_len)
3484                 memcpy(private_data + offset, conn_param->private_data,
3485                        conn_param->private_data_len);
3486
3487         if (private_data) {
3488                 ret = cma_format_hdr(private_data, id_priv);
3489                 if (ret)
3490                         goto out;
3491                 req.private_data = private_data;
3492         }
3493
3494         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3495                              id_priv);
3496         if (IS_ERR(id)) {
3497                 ret = PTR_ERR(id);
3498                 goto out;
3499         }
3500         id_priv->cm_id.ib = id;
3501
3502         req.path = id_priv->id.route.path_rec;
3503         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3504         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3505         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3506
3507         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3508         if (ret) {
3509                 ib_destroy_cm_id(id_priv->cm_id.ib);
3510                 id_priv->cm_id.ib = NULL;
3511         }
3512 out:
3513         kfree(private_data);
3514         return ret;
3515 }
3516
3517 static int cma_connect_ib(struct rdma_id_private *id_priv,
3518                           struct rdma_conn_param *conn_param)
3519 {
3520         struct ib_cm_req_param req;
3521         struct rdma_route *route;
3522         void *private_data;
3523         struct ib_cm_id *id;
3524         int offset, ret;
3525
3526         memset(&req, 0, sizeof req);
3527         offset = cma_user_data_offset(id_priv);
3528         req.private_data_len = offset + conn_param->private_data_len;
3529         if (req.private_data_len < conn_param->private_data_len)
3530                 return -EINVAL;
3531
3532         if (req.private_data_len) {
3533                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3534                 if (!private_data)
3535                         return -ENOMEM;
3536         } else {
3537                 private_data = NULL;
3538         }
3539
3540         if (conn_param->private_data && conn_param->private_data_len)
3541                 memcpy(private_data + offset, conn_param->private_data,
3542                        conn_param->private_data_len);
3543
3544         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3545         if (IS_ERR(id)) {
3546                 ret = PTR_ERR(id);
3547                 goto out;
3548         }
3549         id_priv->cm_id.ib = id;
3550
3551         route = &id_priv->id.route;
3552         if (private_data) {
3553                 ret = cma_format_hdr(private_data, id_priv);
3554                 if (ret)
3555                         goto out;
3556                 req.private_data = private_data;
3557         }
3558
3559         req.primary_path = &route->path_rec[0];
3560         if (route->num_paths == 2)
3561                 req.alternate_path = &route->path_rec[1];
3562
3563         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3564         req.qp_num = id_priv->qp_num;
3565         req.qp_type = id_priv->id.qp_type;
3566         req.starting_psn = id_priv->seq_num;
3567         req.responder_resources = conn_param->responder_resources;
3568         req.initiator_depth = conn_param->initiator_depth;
3569         req.flow_control = conn_param->flow_control;
3570         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3571         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3572         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3573         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3574         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3575         req.srq = id_priv->srq ? 1 : 0;
3576
3577         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3578 out:
3579         if (ret && !IS_ERR(id)) {
3580                 ib_destroy_cm_id(id);
3581                 id_priv->cm_id.ib = NULL;
3582         }
3583
3584         kfree(private_data);
3585         return ret;
3586 }
3587
3588 static int cma_connect_iw(struct rdma_id_private *id_priv,
3589                           struct rdma_conn_param *conn_param)
3590 {
3591         struct iw_cm_id *cm_id;
3592         int ret;
3593         struct iw_cm_conn_param iw_param;
3594
3595         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3596         if (IS_ERR(cm_id))
3597                 return PTR_ERR(cm_id);
3598
3599         cm_id->tos = id_priv->tos;
3600         id_priv->cm_id.iw = cm_id;
3601
3602         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3603                rdma_addr_size(cma_src_addr(id_priv)));
3604         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3605                rdma_addr_size(cma_dst_addr(id_priv)));
3606
3607         ret = cma_modify_qp_rtr(id_priv, conn_param);
3608         if (ret)
3609                 goto out;
3610
3611         if (conn_param) {
3612                 iw_param.ord = conn_param->initiator_depth;
3613                 iw_param.ird = conn_param->responder_resources;
3614                 iw_param.private_data = conn_param->private_data;
3615                 iw_param.private_data_len = conn_param->private_data_len;
3616                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3617         } else {
3618                 memset(&iw_param, 0, sizeof iw_param);
3619                 iw_param.qpn = id_priv->qp_num;
3620         }
3621         ret = iw_cm_connect(cm_id, &iw_param);
3622 out:
3623         if (ret) {
3624                 iw_destroy_cm_id(cm_id);
3625                 id_priv->cm_id.iw = NULL;
3626         }
3627         return ret;
3628 }
3629
3630 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3631 {
3632         struct rdma_id_private *id_priv;
3633         int ret;
3634
3635         id_priv = container_of(id, struct rdma_id_private, id);
3636         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3637                 return -EINVAL;
3638
3639         if (!id->qp) {
3640                 id_priv->qp_num = conn_param->qp_num;
3641                 id_priv->srq = conn_param->srq;
3642         }
3643
3644         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3645                 if (id->qp_type == IB_QPT_UD)
3646                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3647                 else
3648                         ret = cma_connect_ib(id_priv, conn_param);
3649         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3650                 ret = cma_connect_iw(id_priv, conn_param);
3651         else
3652                 ret = -ENOSYS;
3653         if (ret)
3654                 goto err;
3655
3656         return 0;
3657 err:
3658         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3659         return ret;
3660 }
3661 EXPORT_SYMBOL(rdma_connect);
3662
3663 static int cma_accept_ib(struct rdma_id_private *id_priv,
3664                          struct rdma_conn_param *conn_param)
3665 {
3666         struct ib_cm_rep_param rep;
3667         int ret;
3668
3669         ret = cma_modify_qp_rtr(id_priv, conn_param);
3670         if (ret)
3671                 goto out;
3672
3673         ret = cma_modify_qp_rts(id_priv, conn_param);
3674         if (ret)
3675                 goto out;
3676
3677         memset(&rep, 0, sizeof rep);
3678         rep.qp_num = id_priv->qp_num;
3679         rep.starting_psn = id_priv->seq_num;
3680         rep.private_data = conn_param->private_data;
3681         rep.private_data_len = conn_param->private_data_len;
3682         rep.responder_resources = conn_param->responder_resources;
3683         rep.initiator_depth = conn_param->initiator_depth;
3684         rep.failover_accepted = 0;
3685         rep.flow_control = conn_param->flow_control;
3686         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3687         rep.srq = id_priv->srq ? 1 : 0;
3688
3689         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3690 out:
3691         return ret;
3692 }
3693
3694 static int cma_accept_iw(struct rdma_id_private *id_priv,
3695                   struct rdma_conn_param *conn_param)
3696 {
3697         struct iw_cm_conn_param iw_param;
3698         int ret;
3699
3700         if (!conn_param)
3701                 return -EINVAL;
3702
3703         ret = cma_modify_qp_rtr(id_priv, conn_param);
3704         if (ret)
3705                 return ret;
3706
3707         iw_param.ord = conn_param->initiator_depth;
3708         iw_param.ird = conn_param->responder_resources;
3709         iw_param.private_data = conn_param->private_data;
3710         iw_param.private_data_len = conn_param->private_data_len;
3711         if (id_priv->id.qp) {
3712                 iw_param.qpn = id_priv->qp_num;
3713         } else
3714                 iw_param.qpn = conn_param->qp_num;
3715
3716         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3717 }
3718
3719 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3720                              enum ib_cm_sidr_status status, u32 qkey,
3721                              const void *private_data, int private_data_len)
3722 {
3723         struct ib_cm_sidr_rep_param rep;
3724         int ret;
3725
3726         memset(&rep, 0, sizeof rep);
3727         rep.status = status;
3728         if (status == IB_SIDR_SUCCESS) {
3729                 ret = cma_set_qkey(id_priv, qkey);
3730                 if (ret)
3731                         return ret;
3732                 rep.qp_num = id_priv->qp_num;
3733                 rep.qkey = id_priv->qkey;
3734         }
3735         rep.private_data = private_data;
3736         rep.private_data_len = private_data_len;
3737
3738         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3739 }
3740
3741 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3742 {
3743         struct rdma_id_private *id_priv;
3744         int ret;
3745
3746         id_priv = container_of(id, struct rdma_id_private, id);
3747
3748         id_priv->owner = task_pid_nr(current);
3749
3750         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3751                 return -EINVAL;
3752
3753         if (!id->qp && conn_param) {
3754                 id_priv->qp_num = conn_param->qp_num;
3755                 id_priv->srq = conn_param->srq;
3756         }
3757
3758         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3759                 if (id->qp_type == IB_QPT_UD) {
3760                         if (conn_param)
3761                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3762                                                         conn_param->qkey,
3763                                                         conn_param->private_data,
3764                                                         conn_param->private_data_len);
3765                         else
3766                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3767                                                         0, NULL, 0);
3768                 } else {
3769                         if (conn_param)
3770                                 ret = cma_accept_ib(id_priv, conn_param);
3771                         else
3772                                 ret = cma_rep_recv(id_priv);
3773                 }
3774         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3775                 ret = cma_accept_iw(id_priv, conn_param);
3776         else
3777                 ret = -ENOSYS;
3778
3779         if (ret)
3780                 goto reject;
3781
3782         return 0;
3783 reject:
3784         cma_modify_qp_err(id_priv);
3785         rdma_reject(id, NULL, 0);
3786         return ret;
3787 }
3788 EXPORT_SYMBOL(rdma_accept);
3789
3790 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3791 {
3792         struct rdma_id_private *id_priv;
3793         int ret;
3794
3795         id_priv = container_of(id, struct rdma_id_private, id);
3796         if (!id_priv->cm_id.ib)
3797                 return -EINVAL;
3798
3799         switch (id->device->node_type) {
3800         case RDMA_NODE_IB_CA:
3801                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3802                 break;
3803         default:
3804                 ret = 0;
3805                 break;
3806         }
3807         return ret;
3808 }
3809 EXPORT_SYMBOL(rdma_notify);
3810
3811 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3812                 u8 private_data_len)
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                 if (id->qp_type == IB_QPT_UD)
3823                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3824                                                 private_data, private_data_len);
3825                 else
3826                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3827                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3828                                              0, private_data, private_data_len);
3829         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3830                 ret = iw_cm_reject(id_priv->cm_id.iw,
3831                                    private_data, private_data_len);
3832         } else
3833                 ret = -ENOSYS;
3834
3835         return ret;
3836 }
3837 EXPORT_SYMBOL(rdma_reject);
3838
3839 int rdma_disconnect(struct rdma_cm_id *id)
3840 {
3841         struct rdma_id_private *id_priv;
3842         int ret;
3843
3844         id_priv = container_of(id, struct rdma_id_private, id);
3845         if (!id_priv->cm_id.ib)
3846                 return -EINVAL;
3847
3848         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3849                 ret = cma_modify_qp_err(id_priv);
3850                 if (ret)
3851                         goto out;
3852                 /* Initiate or respond to a disconnect. */
3853                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3854                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3855         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3856                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3857         } else
3858                 ret = -EINVAL;
3859
3860 out:
3861         return ret;
3862 }
3863 EXPORT_SYMBOL(rdma_disconnect);
3864
3865 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3866 {
3867         struct rdma_id_private *id_priv;
3868         struct cma_multicast *mc = multicast->context;
3869         struct rdma_cm_event event;
3870         int ret = 0;
3871
3872         id_priv = mc->id_priv;
3873         mutex_lock(&id_priv->handler_mutex);
3874         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3875             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3876                 goto out;
3877
3878         if (!status)
3879                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3880         else
3881                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3882                                      status);
3883         mutex_lock(&id_priv->qp_mutex);
3884         if (!status && id_priv->id.qp) {
3885                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3886                                          be16_to_cpu(multicast->rec.mlid));
3887                 if (status)
3888                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3889                                              status);
3890         }
3891         mutex_unlock(&id_priv->qp_mutex);
3892
3893         memset(&event, 0, sizeof event);
3894         event.status = status;
3895         event.param.ud.private_data = mc->context;
3896         if (!status) {
3897                 struct rdma_dev_addr *dev_addr =
3898                         &id_priv->id.route.addr.dev_addr;
3899                 struct net_device *ndev =
3900                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3901                 enum ib_gid_type gid_type =
3902                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3903                         rdma_start_port(id_priv->cma_dev->device)];
3904
3905                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3906                 ib_init_ah_from_mcmember(id_priv->id.device,
3907                                          id_priv->id.port_num, &multicast->rec,
3908                                          ndev, gid_type,
3909                                          &event.param.ud.ah_attr);
3910                 event.param.ud.qp_num = 0xFFFFFF;
3911                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3912                 if (ndev)
3913                         dev_put(ndev);
3914         } else
3915                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3916
3917         ret = id_priv->id.event_handler(&id_priv->id, &event);
3918         if (ret) {
3919                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3920                 mutex_unlock(&id_priv->handler_mutex);
3921                 rdma_destroy_id(&id_priv->id);
3922                 return 0;
3923         }
3924
3925 out:
3926         mutex_unlock(&id_priv->handler_mutex);
3927         return 0;
3928 }
3929
3930 static void cma_set_mgid(struct rdma_id_private *id_priv,
3931                          struct sockaddr *addr, union ib_gid *mgid)
3932 {
3933         unsigned char mc_map[MAX_ADDR_LEN];
3934         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3935         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3936         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3937
3938         if (cma_any_addr(addr)) {
3939                 memset(mgid, 0, sizeof *mgid);
3940         } else if ((addr->sa_family == AF_INET6) &&
3941                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3942                                                                  0xFF10A01B)) {
3943                 /* IPv6 address is an SA assigned MGID. */
3944                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3945         } else if (addr->sa_family == AF_IB) {
3946                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3947         } else if ((addr->sa_family == AF_INET6)) {
3948                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3949                 if (id_priv->id.ps == RDMA_PS_UDP)
3950                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3951                 *mgid = *(union ib_gid *) (mc_map + 4);
3952         } else {
3953                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3954                 if (id_priv->id.ps == RDMA_PS_UDP)
3955                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3956                 *mgid = *(union ib_gid *) (mc_map + 4);
3957         }
3958 }
3959
3960 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3961                                  struct cma_multicast *mc)
3962 {
3963         struct ib_sa_mcmember_rec rec;
3964         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3965         ib_sa_comp_mask comp_mask;
3966         int ret;
3967
3968         ib_addr_get_mgid(dev_addr, &rec.mgid);
3969         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
3970                                      &rec.mgid, &rec);
3971         if (ret)
3972                 return ret;
3973
3974         ret = cma_set_qkey(id_priv, 0);
3975         if (ret)
3976                 return ret;
3977
3978         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
3979         rec.qkey = cpu_to_be32(id_priv->qkey);
3980         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
3981         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
3982         rec.join_state = mc->join_state;
3983
3984         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
3985             (!ib_sa_sendonly_fullmem_support(&sa_client,
3986                                              id_priv->id.device,
3987                                              id_priv->id.port_num))) {
3988                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
3989                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
3990                         id_priv->id.device->name, id_priv->id.port_num);
3991                 return -EOPNOTSUPP;
3992         }
3993
3994         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
3995                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
3996                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
3997                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
3998                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
3999
4000         if (id_priv->id.ps == RDMA_PS_IPOIB)
4001                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4002                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4003                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4004                              IB_SA_MCMEMBER_REC_MTU |
4005                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4006
4007         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4008                                                 id_priv->id.port_num, &rec,
4009                                                 comp_mask, GFP_KERNEL,
4010                                                 cma_ib_mc_handler, mc);
4011         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4012 }
4013
4014 static void iboe_mcast_work_handler(struct work_struct *work)
4015 {
4016         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4017         struct cma_multicast *mc = mw->mc;
4018         struct ib_sa_multicast *m = mc->multicast.ib;
4019
4020         mc->multicast.ib->context = mc;
4021         cma_ib_mc_handler(0, m);
4022         kref_put(&mc->mcref, release_mc);
4023         kfree(mw);
4024 }
4025
4026 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4027 {
4028         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4029         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4030
4031         if (cma_any_addr(addr)) {
4032                 memset(mgid, 0, sizeof *mgid);
4033         } else if (addr->sa_family == AF_INET6) {
4034                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4035         } else {
4036                 mgid->raw[0] = 0xff;
4037                 mgid->raw[1] = 0x0e;
4038                 mgid->raw[2] = 0;
4039                 mgid->raw[3] = 0;
4040                 mgid->raw[4] = 0;
4041                 mgid->raw[5] = 0;
4042                 mgid->raw[6] = 0;
4043                 mgid->raw[7] = 0;
4044                 mgid->raw[8] = 0;
4045                 mgid->raw[9] = 0;
4046                 mgid->raw[10] = 0xff;
4047                 mgid->raw[11] = 0xff;
4048                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4049         }
4050 }
4051
4052 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4053                                    struct cma_multicast *mc)
4054 {
4055         struct iboe_mcast_work *work;
4056         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4057         int err = 0;
4058         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4059         struct net_device *ndev = NULL;
4060         enum ib_gid_type gid_type;
4061         bool send_only;
4062
4063         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4064
4065         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4066                 return -EINVAL;
4067
4068         work = kzalloc(sizeof *work, GFP_KERNEL);
4069         if (!work)
4070                 return -ENOMEM;
4071
4072         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4073         if (!mc->multicast.ib) {
4074                 err = -ENOMEM;
4075                 goto out1;
4076         }
4077
4078         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4079
4080         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4081         if (id_priv->id.ps == RDMA_PS_UDP)
4082                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4083
4084         if (dev_addr->bound_dev_if)
4085                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4086         if (!ndev) {
4087                 err = -ENODEV;
4088                 goto out2;
4089         }
4090         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4091         mc->multicast.ib->rec.hop_limit = 1;
4092         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4093
4094         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4095                    rdma_start_port(id_priv->cma_dev->device)];
4096         if (addr->sa_family == AF_INET) {
4097                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4098                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4099                         if (!send_only) {
4100                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4101                                                     true);
4102                                 if (!err)
4103                                         mc->igmp_joined = true;
4104                         }
4105                 }
4106         } else {
4107                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4108                         err = -ENOTSUPP;
4109         }
4110         dev_put(ndev);
4111         if (err || !mc->multicast.ib->rec.mtu) {
4112                 if (!err)
4113                         err = -EINVAL;
4114                 goto out2;
4115         }
4116         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4117                     &mc->multicast.ib->rec.port_gid);
4118         work->id = id_priv;
4119         work->mc = mc;
4120         INIT_WORK(&work->work, iboe_mcast_work_handler);
4121         kref_get(&mc->mcref);
4122         queue_work(cma_wq, &work->work);
4123
4124         return 0;
4125
4126 out2:
4127         kfree(mc->multicast.ib);
4128 out1:
4129         kfree(work);
4130         return err;
4131 }
4132
4133 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4134                         u8 join_state, void *context)
4135 {
4136         struct rdma_id_private *id_priv;
4137         struct cma_multicast *mc;
4138         int ret;
4139
4140         id_priv = container_of(id, struct rdma_id_private, id);
4141         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4142             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4143                 return -EINVAL;
4144
4145         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4146         if (!mc)
4147                 return -ENOMEM;
4148
4149         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4150         mc->context = context;
4151         mc->id_priv = id_priv;
4152         mc->igmp_joined = false;
4153         mc->join_state = join_state;
4154         spin_lock(&id_priv->lock);
4155         list_add(&mc->list, &id_priv->mc_list);
4156         spin_unlock(&id_priv->lock);
4157
4158         if (rdma_protocol_roce(id->device, id->port_num)) {
4159                 kref_init(&mc->mcref);
4160                 ret = cma_iboe_join_multicast(id_priv, mc);
4161         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4162                 ret = cma_join_ib_multicast(id_priv, mc);
4163         else
4164                 ret = -ENOSYS;
4165
4166         if (ret) {
4167                 spin_lock_irq(&id_priv->lock);
4168                 list_del(&mc->list);
4169                 spin_unlock_irq(&id_priv->lock);
4170                 kfree(mc);
4171         }
4172         return ret;
4173 }
4174 EXPORT_SYMBOL(rdma_join_multicast);
4175
4176 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4177 {
4178         struct rdma_id_private *id_priv;
4179         struct cma_multicast *mc;
4180
4181         id_priv = container_of(id, struct rdma_id_private, id);
4182         spin_lock_irq(&id_priv->lock);
4183         list_for_each_entry(mc, &id_priv->mc_list, list) {
4184                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4185                         list_del(&mc->list);
4186                         spin_unlock_irq(&id_priv->lock);
4187
4188                         if (id->qp)
4189                                 ib_detach_mcast(id->qp,
4190                                                 &mc->multicast.ib->rec.mgid,
4191                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4192
4193                         BUG_ON(id_priv->cma_dev->device != id->device);
4194
4195                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4196                                 ib_sa_free_multicast(mc->multicast.ib);
4197                                 kfree(mc);
4198                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4199                                 if (mc->igmp_joined) {
4200                                         struct rdma_dev_addr *dev_addr =
4201                                                 &id->route.addr.dev_addr;
4202                                         struct net_device *ndev = NULL;
4203
4204                                         if (dev_addr->bound_dev_if)
4205                                                 ndev = dev_get_by_index(&init_net,
4206                                                                         dev_addr->bound_dev_if);
4207                                         if (ndev) {
4208                                                 cma_igmp_send(ndev,
4209                                                               &mc->multicast.ib->rec.mgid,
4210                                                               false);
4211                                                 dev_put(ndev);
4212                                         }
4213                                         mc->igmp_joined = false;
4214                                 }
4215                                 kref_put(&mc->mcref, release_mc);
4216                         }
4217                         return;
4218                 }
4219         }
4220         spin_unlock_irq(&id_priv->lock);
4221 }
4222 EXPORT_SYMBOL(rdma_leave_multicast);
4223
4224 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4225 {
4226         struct rdma_dev_addr *dev_addr;
4227         struct cma_ndev_work *work;
4228
4229         dev_addr = &id_priv->id.route.addr.dev_addr;
4230
4231         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4232             (net_eq(dev_net(ndev), dev_addr->net)) &&
4233             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4234                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4235                         ndev->name, &id_priv->id);
4236                 work = kzalloc(sizeof *work, GFP_KERNEL);
4237                 if (!work)
4238                         return -ENOMEM;
4239
4240                 INIT_WORK(&work->work, cma_ndev_work_handler);
4241                 work->id = id_priv;
4242                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4243                 atomic_inc(&id_priv->refcount);
4244                 queue_work(cma_wq, &work->work);
4245         }
4246
4247         return 0;
4248 }
4249
4250 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4251                                void *ptr)
4252 {
4253         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4254         struct cma_device *cma_dev;
4255         struct rdma_id_private *id_priv;
4256         int ret = NOTIFY_DONE;
4257
4258         if (event != NETDEV_BONDING_FAILOVER)
4259                 return NOTIFY_DONE;
4260
4261         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4262                 return NOTIFY_DONE;
4263
4264         mutex_lock(&lock);
4265         list_for_each_entry(cma_dev, &dev_list, list)
4266                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4267                         ret = cma_netdev_change(ndev, id_priv);
4268                         if (ret)
4269                                 goto out;
4270                 }
4271
4272 out:
4273         mutex_unlock(&lock);
4274         return ret;
4275 }
4276
4277 static struct notifier_block cma_nb = {
4278         .notifier_call = cma_netdev_callback
4279 };
4280
4281 static void cma_add_one(struct ib_device *device)
4282 {
4283         struct cma_device *cma_dev;
4284         struct rdma_id_private *id_priv;
4285         unsigned int i;
4286         unsigned long supported_gids = 0;
4287
4288         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4289         if (!cma_dev)
4290                 return;
4291
4292         cma_dev->device = device;
4293         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4294                                             sizeof(*cma_dev->default_gid_type),
4295                                             GFP_KERNEL);
4296         if (!cma_dev->default_gid_type)
4297                 goto free_cma_dev;
4298
4299         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4300                                             sizeof(*cma_dev->default_roce_tos),
4301                                             GFP_KERNEL);
4302         if (!cma_dev->default_roce_tos)
4303                 goto free_gid_type;
4304
4305         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4306                 supported_gids = roce_gid_type_mask_support(device, i);
4307                 WARN_ON(!supported_gids);
4308                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4309                         find_first_bit(&supported_gids, BITS_PER_LONG);
4310                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4311         }
4312
4313         init_completion(&cma_dev->comp);
4314         atomic_set(&cma_dev->refcount, 1);
4315         INIT_LIST_HEAD(&cma_dev->id_list);
4316         ib_set_client_data(device, &cma_client, cma_dev);
4317
4318         mutex_lock(&lock);
4319         list_add_tail(&cma_dev->list, &dev_list);
4320         list_for_each_entry(id_priv, &listen_any_list, list)
4321                 cma_listen_on_dev(id_priv, cma_dev);
4322         mutex_unlock(&lock);
4323
4324         return;
4325
4326 free_gid_type:
4327         kfree(cma_dev->default_gid_type);
4328
4329 free_cma_dev:
4330         kfree(cma_dev);
4331
4332         return;
4333 }
4334
4335 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4336 {
4337         struct rdma_cm_event event;
4338         enum rdma_cm_state state;
4339         int ret = 0;
4340
4341         /* Record that we want to remove the device */
4342         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4343         if (state == RDMA_CM_DESTROYING)
4344                 return 0;
4345
4346         cma_cancel_operation(id_priv, state);
4347         mutex_lock(&id_priv->handler_mutex);
4348
4349         /* Check for destruction from another callback. */
4350         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4351                 goto out;
4352
4353         memset(&event, 0, sizeof event);
4354         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4355         ret = id_priv->id.event_handler(&id_priv->id, &event);
4356 out:
4357         mutex_unlock(&id_priv->handler_mutex);
4358         return ret;
4359 }
4360
4361 static void cma_process_remove(struct cma_device *cma_dev)
4362 {
4363         struct rdma_id_private *id_priv;
4364         int ret;
4365
4366         mutex_lock(&lock);
4367         while (!list_empty(&cma_dev->id_list)) {
4368                 id_priv = list_entry(cma_dev->id_list.next,
4369                                      struct rdma_id_private, list);
4370
4371                 list_del(&id_priv->listen_list);
4372                 list_del_init(&id_priv->list);
4373                 atomic_inc(&id_priv->refcount);
4374                 mutex_unlock(&lock);
4375
4376                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4377                 cma_deref_id(id_priv);
4378                 if (ret)
4379                         rdma_destroy_id(&id_priv->id);
4380
4381                 mutex_lock(&lock);
4382         }
4383         mutex_unlock(&lock);
4384
4385         cma_deref_dev(cma_dev);
4386         wait_for_completion(&cma_dev->comp);
4387 }
4388
4389 static void cma_remove_one(struct ib_device *device, void *client_data)
4390 {
4391         struct cma_device *cma_dev = client_data;
4392
4393         if (!cma_dev)
4394                 return;
4395
4396         mutex_lock(&lock);
4397         list_del(&cma_dev->list);
4398         mutex_unlock(&lock);
4399
4400         cma_process_remove(cma_dev);
4401         kfree(cma_dev->default_roce_tos);
4402         kfree(cma_dev->default_gid_type);
4403         kfree(cma_dev);
4404 }
4405
4406 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4407 {
4408         struct nlmsghdr *nlh;
4409         struct rdma_cm_id_stats *id_stats;
4410         struct rdma_id_private *id_priv;
4411         struct rdma_cm_id *id = NULL;
4412         struct cma_device *cma_dev;
4413         int i_dev = 0, i_id = 0;
4414
4415         /*
4416          * We export all of the IDs as a sequence of messages.  Each
4417          * ID gets its own netlink message.
4418          */
4419         mutex_lock(&lock);
4420
4421         list_for_each_entry(cma_dev, &dev_list, list) {
4422                 if (i_dev < cb->args[0]) {
4423                         i_dev++;
4424                         continue;
4425                 }
4426
4427                 i_id = 0;
4428                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4429                         if (i_id < cb->args[1]) {
4430                                 i_id++;
4431                                 continue;
4432                         }
4433
4434                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4435                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4436                                                 RDMA_NL_RDMA_CM_ID_STATS,
4437                                                 NLM_F_MULTI);
4438                         if (!id_stats)
4439                                 goto out;
4440
4441                         memset(id_stats, 0, sizeof *id_stats);
4442                         id = &id_priv->id;
4443                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4444                         id_stats->port_num = id->port_num;
4445                         id_stats->bound_dev_if =
4446                                 id->route.addr.dev_addr.bound_dev_if;
4447
4448                         if (ibnl_put_attr(skb, nlh,
4449                                           rdma_addr_size(cma_src_addr(id_priv)),
4450                                           cma_src_addr(id_priv),
4451                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4452                                 goto out;
4453                         if (ibnl_put_attr(skb, nlh,
4454                                           rdma_addr_size(cma_src_addr(id_priv)),
4455                                           cma_dst_addr(id_priv),
4456                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4457                                 goto out;
4458
4459                         id_stats->pid           = id_priv->owner;
4460                         id_stats->port_space    = id->ps;
4461                         id_stats->cm_state      = id_priv->state;
4462                         id_stats->qp_num        = id_priv->qp_num;
4463                         id_stats->qp_type       = id->qp_type;
4464
4465                         i_id++;
4466                 }
4467
4468                 cb->args[1] = 0;
4469                 i_dev++;
4470         }
4471
4472 out:
4473         mutex_unlock(&lock);
4474         cb->args[0] = i_dev;
4475         cb->args[1] = i_id;
4476
4477         return skb->len;
4478 }
4479
4480 static const struct ibnl_client_cbs cma_cb_table[] = {
4481         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4482                                        .module = THIS_MODULE },
4483 };
4484
4485 static int cma_init_net(struct net *net)
4486 {
4487         struct cma_pernet *pernet = cma_pernet(net);
4488
4489         idr_init(&pernet->tcp_ps);
4490         idr_init(&pernet->udp_ps);
4491         idr_init(&pernet->ipoib_ps);
4492         idr_init(&pernet->ib_ps);
4493
4494         return 0;
4495 }
4496
4497 static void cma_exit_net(struct net *net)
4498 {
4499         struct cma_pernet *pernet = cma_pernet(net);
4500
4501         idr_destroy(&pernet->tcp_ps);
4502         idr_destroy(&pernet->udp_ps);
4503         idr_destroy(&pernet->ipoib_ps);
4504         idr_destroy(&pernet->ib_ps);
4505 }
4506
4507 static struct pernet_operations cma_pernet_operations = {
4508         .init = cma_init_net,
4509         .exit = cma_exit_net,
4510         .id = &cma_pernet_id,
4511         .size = sizeof(struct cma_pernet),
4512 };
4513
4514 static int __init cma_init(void)
4515 {
4516         int ret;
4517
4518         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4519         if (!cma_wq)
4520                 return -ENOMEM;
4521
4522         ret = register_pernet_subsys(&cma_pernet_operations);
4523         if (ret)
4524                 goto err_wq;
4525
4526         ib_sa_register_client(&sa_client);
4527         rdma_addr_register_client(&addr_client);
4528         register_netdevice_notifier(&cma_nb);
4529
4530         ret = ib_register_client(&cma_client);
4531         if (ret)
4532                 goto err;
4533
4534         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4535                             cma_cb_table))
4536                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4537         cma_configfs_init();
4538
4539         return 0;
4540
4541 err:
4542         unregister_netdevice_notifier(&cma_nb);
4543         rdma_addr_unregister_client(&addr_client);
4544         ib_sa_unregister_client(&sa_client);
4545 err_wq:
4546         destroy_workqueue(cma_wq);
4547         return ret;
4548 }
4549
4550 static void __exit cma_cleanup(void)
4551 {
4552         cma_configfs_exit();
4553         ibnl_remove_client(RDMA_NL_RDMA_CM);
4554         ib_unregister_client(&cma_client);
4555         unregister_netdevice_notifier(&cma_nb);
4556         rdma_addr_unregister_client(&addr_client);
4557         ib_sa_unregister_client(&sa_client);
4558         unregister_pernet_subsys(&cma_pernet_operations);
4559         destroy_workqueue(cma_wq);
4560 }
4561
4562 module_init(cma_init);
4563 module_exit(cma_cleanup);