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                            qp_attr.ah_attr.grh.sgid_index, &sgid, NULL);
 933         if (ret)
 934                 goto out;
 935
 936         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 937
 938         if (conn_param)
 939                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 940         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 941 out:
 942         mutex_unlock(&id_priv->qp_mutex);
 943         return ret;
 944 }
 945
 946 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 947                              struct rdma_conn_param *conn_param)
 948 {
 949         struct ib_qp_attr qp_attr;
 950         int qp_attr_mask, ret;
 951
 952         mutex_lock(&id_priv->qp_mutex);
 953         if (!id_priv->id.qp) {
 954                 ret = 0;
 955                 goto out;
 956         }
 957
 958         qp_attr.qp_state = IB_QPS_RTS;
 959         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 960         if (ret)
 961                 goto out;
 962
 963         if (conn_param)
 964                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 965         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 966 out:
 967         mutex_unlock(&id_priv->qp_mutex);
 968         return ret;
 969 }
 970
 971 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 972 {
 973         struct ib_qp_attr qp_attr;
 974         int ret;
 975
 976         mutex_lock(&id_priv->qp_mutex);
 977         if (!id_priv->id.qp) {
 978                 ret = 0;
 979                 goto out;
 980         }
 981
 982         qp_attr.qp_state = IB_QPS_ERR;
 983         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 984 out:
 985         mutex_unlock(&id_priv->qp_mutex);
 986         return ret;
 987 }
 988
 989 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 990                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 991 {
 992         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 993         int ret;
 994         u16 pkey;
 995
 996         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 997                 pkey = 0xffff;
 998         else
 999                 pkey = ib_addr_get_pkey(dev_addr);
1000
1001         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
1002                                   pkey, &qp_attr->pkey_index);
1003         if (ret)
1004                 return ret;
1005
1006         qp_attr->port_num = id_priv->id.port_num;
1007         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1008
1009         if (id_priv->id.qp_type == IB_QPT_UD) {
1010                 ret = cma_set_qkey(id_priv, 0);
1011                 if (ret)
1012                         return ret;
1013
1014                 qp_attr->qkey = id_priv->qkey;
1015                 *qp_attr_mask |= IB_QP_QKEY;
1016         } else {
1017                 qp_attr->qp_access_flags = 0;
1018                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1019         }
1020         return 0;
1021 }
1022
1023 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1024                        int *qp_attr_mask)
1025 {
1026         struct rdma_id_private *id_priv;
1027         int ret = 0;
1028
1029         id_priv = container_of(id, struct rdma_id_private, id);
1030         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1031                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1032                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1033                 else
1034                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1035                                                  qp_attr_mask);
1036
1037                 if (qp_attr->qp_state == IB_QPS_RTR)
1038                         qp_attr->rq_psn = id_priv->seq_num;
1039         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1040                 if (!id_priv->cm_id.iw) {
1041                         qp_attr->qp_access_flags = 0;
1042                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1043                 } else
1044                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1045                                                  qp_attr_mask);
1046         } else
1047                 ret = -ENOSYS;
1048
1049         return ret;
1050 }
1051 EXPORT_SYMBOL(rdma_init_qp_attr);
1052
1053 static inline int cma_zero_addr(struct sockaddr *addr)
1054 {
1055         switch (addr->sa_family) {
1056         case AF_INET:
1057                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1058         case AF_INET6:
1059                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1060         case AF_IB:
1061                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1062         default:
1063                 return 0;
1064         }
1065 }
1066
1067 static inline int cma_loopback_addr(struct sockaddr *addr)
1068 {
1069         switch (addr->sa_family) {
1070         case AF_INET:
1071                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1072         case AF_INET6:
1073                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1074         case AF_IB:
1075                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1076         default:
1077                 return 0;
1078         }
1079 }
1080
1081 static inline int cma_any_addr(struct sockaddr *addr)
1082 {
1083         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1084 }
1085
1086 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1087 {
1088         if (src->sa_family != dst->sa_family)
1089                 return -1;
1090
1091         switch (src->sa_family) {
1092         case AF_INET:
1093                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1094                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1095         case AF_INET6:
1096                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1097                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1098         default:
1099                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1100                                    &((struct sockaddr_ib *) dst)->sib_addr);
1101         }
1102 }
1103
1104 static __be16 cma_port(struct sockaddr *addr)
1105 {
1106         struct sockaddr_ib *sib;
1107
1108         switch (addr->sa_family) {
1109         case AF_INET:
1110                 return ((struct sockaddr_in *) addr)->sin_port;
1111         case AF_INET6:
1112                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1113         case AF_IB:
1114                 sib = (struct sockaddr_ib *) addr;
1115                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1116                                     be64_to_cpu(sib->sib_sid_mask)));
1117         default:
1118                 return 0;
1119         }
1120 }
1121
1122 static inline int cma_any_port(struct sockaddr *addr)
1123 {
1124         return !cma_port(addr);
1125 }
1126
1127 static void cma_save_ib_info(struct sockaddr *src_addr,
1128                              struct sockaddr *dst_addr,
1129                              struct rdma_cm_id *listen_id,
1130                              struct ib_sa_path_rec *path)
1131 {
1132         struct sockaddr_ib *listen_ib, *ib;
1133
1134         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1135         if (src_addr) {
1136                 ib = (struct sockaddr_ib *)src_addr;
1137                 ib->sib_family = AF_IB;
1138                 if (path) {
1139                         ib->sib_pkey = path->pkey;
1140                         ib->sib_flowinfo = path->flow_label;
1141                         memcpy(&ib->sib_addr, &path->sgid, 16);
1142                         ib->sib_sid = path->service_id;
1143                         ib->sib_scope_id = 0;
1144                 } else {
1145                         ib->sib_pkey = listen_ib->sib_pkey;
1146                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1147                         ib->sib_addr = listen_ib->sib_addr;
1148                         ib->sib_sid = listen_ib->sib_sid;
1149                         ib->sib_scope_id = listen_ib->sib_scope_id;
1150                 }
1151                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1152         }
1153         if (dst_addr) {
1154                 ib = (struct sockaddr_ib *)dst_addr;
1155                 ib->sib_family = AF_IB;
1156                 if (path) {
1157                         ib->sib_pkey = path->pkey;
1158                         ib->sib_flowinfo = path->flow_label;
1159                         memcpy(&ib->sib_addr, &path->dgid, 16);
1160                 }
1161         }
1162 }
1163
1164 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1165                               struct sockaddr_in *dst_addr,
1166                               struct cma_hdr *hdr,
1167                               __be16 local_port)
1168 {
1169         if (src_addr) {
1170                 *src_addr = (struct sockaddr_in) {
1171                         .sin_family = AF_INET,
1172                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1173                         .sin_port = local_port,
1174                 };
1175         }
1176
1177         if (dst_addr) {
1178                 *dst_addr = (struct sockaddr_in) {
1179                         .sin_family = AF_INET,
1180                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1181                         .sin_port = hdr->port,
1182                 };
1183         }
1184 }
1185
1186 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1187                               struct sockaddr_in6 *dst_addr,
1188                               struct cma_hdr *hdr,
1189                               __be16 local_port)
1190 {
1191         if (src_addr) {
1192                 *src_addr = (struct sockaddr_in6) {
1193                         .sin6_family = AF_INET6,
1194                         .sin6_addr = hdr->dst_addr.ip6,
1195                         .sin6_port = local_port,
1196                 };
1197         }
1198
1199         if (dst_addr) {
1200                 *dst_addr = (struct sockaddr_in6) {
1201                         .sin6_family = AF_INET6,
1202                         .sin6_addr = hdr->src_addr.ip6,
1203                         .sin6_port = hdr->port,
1204                 };
1205         }
1206 }
1207
1208 static u16 cma_port_from_service_id(__be64 service_id)
1209 {
1210         return (u16)be64_to_cpu(service_id);
1211 }
1212
1213 static int cma_save_ip_info(struct sockaddr *src_addr,
1214                             struct sockaddr *dst_addr,
1215                             struct ib_cm_event *ib_event,
1216                             __be64 service_id)
1217 {
1218         struct cma_hdr *hdr;
1219         __be16 port;
1220
1221         hdr = ib_event->private_data;
1222         if (hdr->cma_version != CMA_VERSION)
1223                 return -EINVAL;
1224
1225         port = htons(cma_port_from_service_id(service_id));
1226
1227         switch (cma_get_ip_ver(hdr)) {
1228         case 4:
1229                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1230                                   (struct sockaddr_in *)dst_addr, hdr, port);
1231                 break;
1232         case 6:
1233                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1234                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1235                 break;
1236         default:
1237                 return -EAFNOSUPPORT;
1238         }
1239
1240         return 0;
1241 }
1242
1243 static int cma_save_net_info(struct sockaddr *src_addr,
1244                              struct sockaddr *dst_addr,
1245                              struct rdma_cm_id *listen_id,
1246                              struct ib_cm_event *ib_event,
1247                              sa_family_t sa_family, __be64 service_id)
1248 {
1249         if (sa_family == AF_IB) {
1250                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1251                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1252                                          ib_event->param.req_rcvd.primary_path);
1253                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1254                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1255                 return 0;
1256         }
1257
1258         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1259 }
1260
1261 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1262                              struct cma_req_info *req)
1263 {
1264         const struct ib_cm_req_event_param *req_param =
1265                 &ib_event->param.req_rcvd;
1266         const struct ib_cm_sidr_req_event_param *sidr_param =
1267                 &ib_event->param.sidr_req_rcvd;
1268
1269         switch (ib_event->event) {
1270         case IB_CM_REQ_RECEIVED:
1271                 req->device     = req_param->listen_id->device;
1272                 req->port       = req_param->port;
1273                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1274                        sizeof(req->local_gid));
1275                 req->has_gid    = true;
1276                 req->service_id = req_param->primary_path->service_id;
1277                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1278                 if (req->pkey != req_param->bth_pkey)
1279                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1280                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1281                                             req_param->bth_pkey, req->pkey);
1282                 break;
1283         case IB_CM_SIDR_REQ_RECEIVED:
1284                 req->device     = sidr_param->listen_id->device;
1285                 req->port       = sidr_param->port;
1286                 req->has_gid    = false;
1287                 req->service_id = sidr_param->service_id;
1288                 req->pkey       = sidr_param->pkey;
1289                 if (req->pkey != sidr_param->bth_pkey)
1290                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1291                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1292                                             sidr_param->bth_pkey, req->pkey);
1293                 break;
1294         default:
1295                 return -EINVAL;
1296         }
1297
1298         return 0;
1299 }
1300
1301 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1302                                   const struct sockaddr_in *dst_addr,
1303                                   const struct sockaddr_in *src_addr)
1304 {
1305         __be32 daddr = dst_addr->sin_addr.s_addr,
1306                saddr = src_addr->sin_addr.s_addr;
1307         struct fib_result res;
1308         struct flowi4 fl4;
1309         int err;
1310         bool ret;
1311
1312         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1313             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1314             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1315             ipv4_is_loopback(saddr))
1316                 return false;
1317
1318         memset(&fl4, 0, sizeof(fl4));
1319         fl4.flowi4_iif = net_dev->ifindex;
1320         fl4.daddr = daddr;
1321         fl4.saddr = saddr;
1322
1323         rcu_read_lock();
1324         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1325         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1326         rcu_read_unlock();
1327
1328         return ret;
1329 }
1330
1331 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1332                                   const struct sockaddr_in6 *dst_addr,
1333                                   const struct sockaddr_in6 *src_addr)
1334 {
1335 #if IS_ENABLED(CONFIG_IPV6)
1336         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1337                            IPV6_ADDR_LINKLOCAL;
1338         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1339                                          &src_addr->sin6_addr, net_dev->ifindex,
1340                                          strict);
1341         bool ret;
1342
1343         if (!rt)
1344                 return false;
1345
1346         ret = rt->rt6i_idev->dev == net_dev;
1347         ip6_rt_put(rt);
1348
1349         return ret;
1350 #else
1351         return false;
1352 #endif
1353 }
1354
1355 static bool validate_net_dev(struct net_device *net_dev,
1356                              const struct sockaddr *daddr,
1357                              const struct sockaddr *saddr)
1358 {
1359         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1360         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1361         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1362         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1363
1364         switch (daddr->sa_family) {
1365         case AF_INET:
1366                 return saddr->sa_family == AF_INET &&
1367                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1368
1369         case AF_INET6:
1370                 return saddr->sa_family == AF_INET6 &&
1371                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1372
1373         default:
1374                 return false;
1375         }
1376 }
1377
1378 static struct net_device *cma_get_net_dev(struct ib_cm_event *ib_event,
1379                                           const struct cma_req_info *req)
1380 {
1381         struct sockaddr_storage listen_addr_storage, src_addr_storage;
1382         struct sockaddr *listen_addr = (struct sockaddr *)&listen_addr_storage,
1383                         *src_addr = (struct sockaddr *)&src_addr_storage;
1384         struct net_device *net_dev;
1385         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1386         int err;
1387
1388         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1389                                req->service_id);
1390         if (err)
1391                 return ERR_PTR(err);
1392
1393         net_dev = ib_get_net_dev_by_params(req->device, req->port, req->pkey,
1394                                            gid, listen_addr);
1395         if (!net_dev)
1396                 return ERR_PTR(-ENODEV);
1397
1398         if (!validate_net_dev(net_dev, listen_addr, src_addr)) {
1399                 dev_put(net_dev);
1400                 return ERR_PTR(-EHOSTUNREACH);
1401         }
1402
1403         return net_dev;
1404 }
1405
1406 static enum rdma_port_space rdma_ps_from_service_id(__be64 service_id)
1407 {
1408         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1409 }
1410
1411 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1412                                    const struct cma_hdr *hdr)
1413 {
1414         struct sockaddr *addr = cma_src_addr(id_priv);
1415         __be32 ip4_addr;
1416         struct in6_addr ip6_addr;
1417
1418         if (cma_any_addr(addr) && !id_priv->afonly)
1419                 return true;
1420
1421         switch (addr->sa_family) {
1422         case AF_INET:
1423                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1424                 if (cma_get_ip_ver(hdr) != 4)
1425                         return false;
1426                 if (!cma_any_addr(addr) &&
1427                     hdr->dst_addr.ip4.addr != ip4_addr)
1428                         return false;
1429                 break;
1430         case AF_INET6:
1431                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1432                 if (cma_get_ip_ver(hdr) != 6)
1433                         return false;
1434                 if (!cma_any_addr(addr) &&
1435                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1436                         return false;
1437                 break;
1438         case AF_IB:
1439                 return true;
1440         default:
1441                 return false;
1442         }
1443
1444         return true;
1445 }
1446
1447 static bool cma_protocol_roce_dev_port(struct ib_device *device, int port_num)
1448 {
1449         enum rdma_link_layer ll = rdma_port_get_link_layer(device, port_num);
1450         enum rdma_transport_type transport =
1451                 rdma_node_get_transport(device->node_type);
1452
1453         return ll == IB_LINK_LAYER_ETHERNET && transport == RDMA_TRANSPORT_IB;
1454 }
1455
1456 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1457 {
1458         struct ib_device *device = id->device;
1459         const int port_num = id->port_num ?: rdma_start_port(device);
1460
1461         return cma_protocol_roce_dev_port(device, port_num);
1462 }
1463
1464 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1465                               const struct net_device *net_dev,
1466                               u8 port_num)
1467 {
1468         const struct rdma_addr *addr = &id->route.addr;
1469
1470         if (!net_dev)
1471                 /* This request is an AF_IB request or a RoCE request */
1472                 return (!id->port_num || id->port_num == port_num) &&
1473                        (addr->src_addr.ss_family == AF_IB ||
1474                         cma_protocol_roce_dev_port(id->device, port_num));
1475
1476         return !addr->dev_addr.bound_dev_if ||
1477                (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1478                 addr->dev_addr.bound_dev_if == net_dev->ifindex);
1479 }
1480
1481 static struct rdma_id_private *cma_find_listener(
1482                 const struct rdma_bind_list *bind_list,
1483                 const struct ib_cm_id *cm_id,
1484                 const struct ib_cm_event *ib_event,
1485                 const struct cma_req_info *req,
1486                 const struct net_device *net_dev)
1487 {
1488         struct rdma_id_private *id_priv, *id_priv_dev;
1489
1490         if (!bind_list)
1491                 return ERR_PTR(-EINVAL);
1492
1493         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1494                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1495                         if (id_priv->id.device == cm_id->device &&
1496                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1497                                 return id_priv;
1498                         list_for_each_entry(id_priv_dev,
1499                                             &id_priv->listen_list,
1500                                             listen_list) {
1501                                 if (id_priv_dev->id.device == cm_id->device &&
1502                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1503                                         return id_priv_dev;
1504                         }
1505                 }
1506         }
1507
1508         return ERR_PTR(-EINVAL);
1509 }
1510
1511 static struct rdma_id_private *cma_id_from_event(struct ib_cm_id *cm_id,
1512                                                  struct ib_cm_event *ib_event,
1513                                                  struct net_device **net_dev)
1514 {
1515         struct cma_req_info req;
1516         struct rdma_bind_list *bind_list;
1517         struct rdma_id_private *id_priv;
1518         int err;
1519
1520         err = cma_save_req_info(ib_event, &req);
1521         if (err)
1522                 return ERR_PTR(err);
1523
1524         *net_dev = cma_get_net_dev(ib_event, &req);
1525         if (IS_ERR(*net_dev)) {
1526                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1527                         /* Assuming the protocol is AF_IB */
1528                         *net_dev = NULL;
1529                 } else if (cma_protocol_roce_dev_port(req.device, req.port)) {
1530                         /* TODO find the net dev matching the request parameters
1531                          * through the RoCE GID table */
1532                         *net_dev = NULL;
1533                 } else {
1534                         return ERR_CAST(*net_dev);
1535                 }
1536         }
1537
1538         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1539                                 rdma_ps_from_service_id(req.service_id),
1540                                 cma_port_from_service_id(req.service_id));
1541         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1542         if (IS_ERR(id_priv) && *net_dev) {
1543                 dev_put(*net_dev);
1544                 *net_dev = NULL;
1545         }
1546
1547         return id_priv;
1548 }
1549
1550 static inline int cma_user_data_offset(struct rdma_id_private *id_priv)
1551 {
1552         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1553 }
1554
1555 static void cma_cancel_route(struct rdma_id_private *id_priv)
1556 {
1557         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1558                 if (id_priv->query)
1559                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1560         }
1561 }
1562
1563 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1564 {
1565         struct rdma_id_private *dev_id_priv;
1566
1567         /*
1568          * Remove from listen_any_list to prevent added devices from spawning
1569          * additional listen requests.
1570          */
1571         mutex_lock(&lock);
1572         list_del(&id_priv->list);
1573
1574         while (!list_empty(&id_priv->listen_list)) {
1575                 dev_id_priv = list_entry(id_priv->listen_list.next,
1576                                          struct rdma_id_private, listen_list);
1577                 /* sync with device removal to avoid duplicate destruction */
1578                 list_del_init(&dev_id_priv->list);
1579                 list_del(&dev_id_priv->listen_list);
1580                 mutex_unlock(&lock);
1581
1582                 rdma_destroy_id(&dev_id_priv->id);
1583                 mutex_lock(&lock);
1584         }
1585         mutex_unlock(&lock);
1586 }
1587
1588 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1589                                  enum rdma_cm_state state)
1590 {
1591         switch (state) {
1592         case RDMA_CM_ADDR_QUERY:
1593                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1594                 break;
1595         case RDMA_CM_ROUTE_QUERY:
1596                 cma_cancel_route(id_priv);
1597                 break;
1598         case RDMA_CM_LISTEN:
1599                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1600                         cma_cancel_listens(id_priv);
1601                 break;
1602         default:
1603                 break;
1604         }
1605 }
1606
1607 static void cma_release_port(struct rdma_id_private *id_priv)
1608 {
1609         struct rdma_bind_list *bind_list = id_priv->bind_list;
1610         struct net *net = id_priv->id.route.addr.dev_addr.net;
1611
1612         if (!bind_list)
1613                 return;
1614
1615         mutex_lock(&lock);
1616         hlist_del(&id_priv->node);
1617         if (hlist_empty(&bind_list->owners)) {
1618                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1619                 kfree(bind_list);
1620         }
1621         mutex_unlock(&lock);
1622 }
1623
1624 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1625 {
1626         struct cma_multicast *mc;
1627
1628         while (!list_empty(&id_priv->mc_list)) {
1629                 mc = container_of(id_priv->mc_list.next,
1630                                   struct cma_multicast, list);
1631                 list_del(&mc->list);
1632                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1633                                       id_priv->id.port_num)) {
1634                         ib_sa_free_multicast(mc->multicast.ib);
1635                         kfree(mc);
1636                 } else {
1637                         if (mc->igmp_joined) {
1638                                 struct rdma_dev_addr *dev_addr =
1639                                         &id_priv->id.route.addr.dev_addr;
1640                                 struct net_device *ndev = NULL;
1641
1642                                 if (dev_addr->bound_dev_if)
1643                                         ndev = dev_get_by_index(&init_net,
1644                                                                 dev_addr->bound_dev_if);
1645                                 if (ndev) {
1646                                         cma_igmp_send(ndev,
1647                                                       &mc->multicast.ib->rec.mgid,
1648                                                       false);
1649                                         dev_put(ndev);
1650                                 }
1651                         }
1652                         kref_put(&mc->mcref, release_mc);
1653                 }
1654         }
1655 }
1656
1657 void rdma_destroy_id(struct rdma_cm_id *id)
1658 {
1659         struct rdma_id_private *id_priv;
1660         enum rdma_cm_state state;
1661
1662         id_priv = container_of(id, struct rdma_id_private, id);
1663         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1664         cma_cancel_operation(id_priv, state);
1665
1666         /*
1667          * Wait for any active callback to finish.  New callbacks will find
1668          * the id_priv state set to destroying and abort.
1669          */
1670         mutex_lock(&id_priv->handler_mutex);
1671         mutex_unlock(&id_priv->handler_mutex);
1672
1673         if (id_priv->cma_dev) {
1674                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1675                         if (id_priv->cm_id.ib)
1676                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1677                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1678                         if (id_priv->cm_id.iw)
1679                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1680                 }
1681                 cma_leave_mc_groups(id_priv);
1682                 cma_release_dev(id_priv);
1683         }
1684
1685         cma_release_port(id_priv);
1686         cma_deref_id(id_priv);
1687         wait_for_completion(&id_priv->comp);
1688
1689         if (id_priv->internal_id)
1690                 cma_deref_id(id_priv->id.context);
1691
1692         kfree(id_priv->id.route.path_rec);
1693         put_net(id_priv->id.route.addr.dev_addr.net);
1694         kfree(id_priv);
1695 }
1696 EXPORT_SYMBOL(rdma_destroy_id);
1697
1698 static int cma_rep_recv(struct rdma_id_private *id_priv)
1699 {
1700         int ret;
1701
1702         ret = cma_modify_qp_rtr(id_priv, NULL);
1703         if (ret)
1704                 goto reject;
1705
1706         ret = cma_modify_qp_rts(id_priv, NULL);
1707         if (ret)
1708                 goto reject;
1709
1710         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1711         if (ret)
1712                 goto reject;
1713
1714         return 0;
1715 reject:
1716         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1717         cma_modify_qp_err(id_priv);
1718         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1719                        NULL, 0, NULL, 0);
1720         return ret;
1721 }
1722
1723 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1724                                    struct ib_cm_rep_event_param *rep_data,
1725                                    void *private_data)
1726 {
1727         event->param.conn.private_data = private_data;
1728         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1729         event->param.conn.responder_resources = rep_data->responder_resources;
1730         event->param.conn.initiator_depth = rep_data->initiator_depth;
1731         event->param.conn.flow_control = rep_data->flow_control;
1732         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1733         event->param.conn.srq = rep_data->srq;
1734         event->param.conn.qp_num = rep_data->remote_qpn;
1735 }
1736
1737 static int cma_ib_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1738 {
1739         struct rdma_id_private *id_priv = cm_id->context;
1740         struct rdma_cm_event event;
1741         int ret = 0;
1742
1743         mutex_lock(&id_priv->handler_mutex);
1744         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1745              id_priv->state != RDMA_CM_CONNECT) ||
1746             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1747              id_priv->state != RDMA_CM_DISCONNECT))
1748                 goto out;
1749
1750         memset(&event, 0, sizeof event);
1751         switch (ib_event->event) {
1752         case IB_CM_REQ_ERROR:
1753         case IB_CM_REP_ERROR:
1754                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1755                 event.status = -ETIMEDOUT;
1756                 break;
1757         case IB_CM_REP_RECEIVED:
1758                 if (id_priv->id.qp) {
1759                         event.status = cma_rep_recv(id_priv);
1760                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1761                                                      RDMA_CM_EVENT_ESTABLISHED;
1762                 } else {
1763                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1764                 }
1765                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1766                                        ib_event->private_data);
1767                 break;
1768         case IB_CM_RTU_RECEIVED:
1769         case IB_CM_USER_ESTABLISHED:
1770                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1771                 break;
1772         case IB_CM_DREQ_ERROR:
1773                 event.status = -ETIMEDOUT; /* fall through */
1774         case IB_CM_DREQ_RECEIVED:
1775         case IB_CM_DREP_RECEIVED:
1776                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1777                                    RDMA_CM_DISCONNECT))
1778                         goto out;
1779                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1780                 break;
1781         case IB_CM_TIMEWAIT_EXIT:
1782                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1783                 break;
1784         case IB_CM_MRA_RECEIVED:
1785                 /* ignore event */
1786                 goto out;
1787         case IB_CM_REJ_RECEIVED:
1788                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1789                                                                                 ib_event->param.rej_rcvd.reason));
1790                 cma_modify_qp_err(id_priv);
1791                 event.status = ib_event->param.rej_rcvd.reason;
1792                 event.event = RDMA_CM_EVENT_REJECTED;
1793                 event.param.conn.private_data = ib_event->private_data;
1794                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1795                 break;
1796         default:
1797                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1798                        ib_event->event);
1799                 goto out;
1800         }
1801
1802         ret = id_priv->id.event_handler(&id_priv->id, &event);
1803         if (ret) {
1804                 /* Destroy the CM ID by returning a non-zero value. */
1805                 id_priv->cm_id.ib = NULL;
1806                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1807                 mutex_unlock(&id_priv->handler_mutex);
1808                 rdma_destroy_id(&id_priv->id);
1809                 return ret;
1810         }
1811 out:
1812         mutex_unlock(&id_priv->handler_mutex);
1813         return ret;
1814 }
1815
1816 static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id,
1817                                                struct ib_cm_event *ib_event,
1818                                                struct net_device *net_dev)
1819 {
1820         struct rdma_id_private *id_priv;
1821         struct rdma_cm_id *id;
1822         struct rdma_route *rt;
1823         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1824         const __be64 service_id =
1825                       ib_event->param.req_rcvd.primary_path->service_id;
1826         int ret;
1827
1828         id = rdma_create_id(listen_id->route.addr.dev_addr.net,
1829                             listen_id->event_handler, listen_id->context,
1830                             listen_id->ps, ib_event->param.req_rcvd.qp_type);
1831         if (IS_ERR(id))
1832                 return NULL;
1833
1834         id_priv = container_of(id, struct rdma_id_private, id);
1835         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1836                               (struct sockaddr *)&id->route.addr.dst_addr,
1837                               listen_id, ib_event, ss_family, service_id))
1838                 goto err;
1839
1840         rt = &id->route;
1841         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1842         rt->path_rec = kmalloc(sizeof *rt->path_rec * rt->num_paths,
1843                                GFP_KERNEL);
1844         if (!rt->path_rec)
1845                 goto err;
1846
1847         rt->path_rec[0] = *ib_event->param.req_rcvd.primary_path;
1848         if (rt->num_paths == 2)
1849                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1850
1851         if (net_dev) {
1852                 ret = rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1853                 if (ret)
1854                         goto err;
1855         } else {
1856                 if (!cma_protocol_roce(listen_id) &&
1857                     cma_any_addr(cma_src_addr(id_priv))) {
1858                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1859                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1860                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1861                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1862                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1863                         if (ret)
1864                                 goto err;
1865                 }
1866         }
1867         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1868
1869         id_priv->state = RDMA_CM_CONNECT;
1870         return id_priv;
1871
1872 err:
1873         rdma_destroy_id(id);
1874         return NULL;
1875 }
1876
1877 static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id,
1878                                               struct ib_cm_event *ib_event,
1879                                               struct net_device *net_dev)
1880 {
1881         struct rdma_id_private *id_priv;
1882         struct rdma_cm_id *id;
1883         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1884         struct net *net = listen_id->route.addr.dev_addr.net;
1885         int ret;
1886
1887         id = rdma_create_id(net, listen_id->event_handler, listen_id->context,
1888                             listen_id->ps, IB_QPT_UD);
1889         if (IS_ERR(id))
1890                 return NULL;
1891
1892         id_priv = container_of(id, struct rdma_id_private, id);
1893         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1894                               (struct sockaddr *)&id->route.addr.dst_addr,
1895                               listen_id, ib_event, ss_family,
1896                               ib_event->param.sidr_req_rcvd.service_id))
1897                 goto err;
1898
1899         if (net_dev) {
1900                 ret = rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1901                 if (ret)
1902                         goto err;
1903         } else {
1904                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1905                         ret = cma_translate_addr(cma_src_addr(id_priv),
1906                                                  &id->route.addr.dev_addr);
1907                         if (ret)
1908                                 goto err;
1909                 }
1910         }
1911
1912         id_priv->state = RDMA_CM_CONNECT;
1913         return id_priv;
1914 err:
1915         rdma_destroy_id(id);
1916         return NULL;
1917 }
1918
1919 static void cma_set_req_event_data(struct rdma_cm_event *event,
1920                                    struct ib_cm_req_event_param *req_data,
1921                                    void *private_data, int offset)
1922 {
1923         event->param.conn.private_data = private_data + offset;
1924         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1925         event->param.conn.responder_resources = req_data->responder_resources;
1926         event->param.conn.initiator_depth = req_data->initiator_depth;
1927         event->param.conn.flow_control = req_data->flow_control;
1928         event->param.conn.retry_count = req_data->retry_count;
1929         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1930         event->param.conn.srq = req_data->srq;
1931         event->param.conn.qp_num = req_data->remote_qpn;
1932 }
1933
1934 static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
1935 {
1936         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1937                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1938                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1939                  (id->qp_type == IB_QPT_UD)) ||
1940                 (!id->qp_type));
1941 }
1942
1943 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
1944 {
1945         struct rdma_id_private *listen_id, *conn_id = NULL;
1946         struct rdma_cm_event event;
1947         struct net_device *net_dev;
1948         int offset, ret;
1949
1950         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
1951         if (IS_ERR(listen_id))
1952                 return PTR_ERR(listen_id);
1953
1954         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
1955                 ret = -EINVAL;
1956                 goto net_dev_put;
1957         }
1958
1959         mutex_lock(&listen_id->handler_mutex);
1960         if (listen_id->state != RDMA_CM_LISTEN) {
1961                 ret = -ECONNABORTED;
1962                 goto err1;
1963         }
1964
1965         memset(&event, 0, sizeof event);
1966         offset = cma_user_data_offset(listen_id);
1967         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
1968         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
1969                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
1970                 event.param.ud.private_data = ib_event->private_data + offset;
1971                 event.param.ud.private_data_len =
1972                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
1973         } else {
1974                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
1975                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
1976                                        ib_event->private_data, offset);
1977         }
1978         if (!conn_id) {
1979                 ret = -ENOMEM;
1980                 goto err1;
1981         }
1982
1983         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
1984         ret = cma_acquire_dev(conn_id, listen_id);
1985         if (ret)
1986                 goto err2;
1987
1988         conn_id->cm_id.ib = cm_id;
1989         cm_id->context = conn_id;
1990         cm_id->cm_handler = cma_ib_handler;
1991
1992         /*
1993          * Protect against the user destroying conn_id from another thread
1994          * until we're done accessing it.
1995          */
1996         atomic_inc(&conn_id->refcount);
1997         ret = conn_id->id.event_handler(&conn_id->id, &event);
1998         if (ret)
1999                 goto err3;
2000         /*
2001          * Acquire mutex to prevent user executing rdma_destroy_id()
2002          * while we're accessing the cm_id.
2003          */
2004         mutex_lock(&lock);
2005         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2006             (conn_id->id.qp_type != IB_QPT_UD))
2007                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2008         mutex_unlock(&lock);
2009         mutex_unlock(&conn_id->handler_mutex);
2010         mutex_unlock(&listen_id->handler_mutex);
2011         cma_deref_id(conn_id);
2012         if (net_dev)
2013                 dev_put(net_dev);
2014         return 0;
2015
2016 err3:
2017         cma_deref_id(conn_id);
2018         /* Destroy the CM ID by returning a non-zero value. */
2019         conn_id->cm_id.ib = NULL;
2020 err2:
2021         cma_exch(conn_id, RDMA_CM_DESTROYING);
2022         mutex_unlock(&conn_id->handler_mutex);
2023 err1:
2024         mutex_unlock(&listen_id->handler_mutex);
2025         if (conn_id)
2026                 rdma_destroy_id(&conn_id->id);
2027
2028 net_dev_put:
2029         if (net_dev)
2030                 dev_put(net_dev);
2031
2032         return ret;
2033 }
2034
2035 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2036 {
2037         if (addr->sa_family == AF_IB)
2038                 return ((struct sockaddr_ib *) addr)->sib_sid;
2039
2040         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2041 }
2042 EXPORT_SYMBOL(rdma_get_service_id);
2043
2044 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2045 {
2046         struct rdma_id_private *id_priv = iw_id->context;
2047         struct rdma_cm_event event;
2048         int ret = 0;
2049         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2050         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2051
2052         mutex_lock(&id_priv->handler_mutex);
2053         if (id_priv->state != RDMA_CM_CONNECT)
2054                 goto out;
2055
2056         memset(&event, 0, sizeof event);
2057         switch (iw_event->event) {
2058         case IW_CM_EVENT_CLOSE:
2059                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2060                 break;
2061         case IW_CM_EVENT_CONNECT_REPLY:
2062                 memcpy(cma_src_addr(id_priv), laddr,
2063                        rdma_addr_size(laddr));
2064                 memcpy(cma_dst_addr(id_priv), raddr,
2065                        rdma_addr_size(raddr));
2066                 switch (iw_event->status) {
2067                 case 0:
2068                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2069                         event.param.conn.initiator_depth = iw_event->ird;
2070                         event.param.conn.responder_resources = iw_event->ord;
2071                         break;
2072                 case -ECONNRESET:
2073                 case -ECONNREFUSED:
2074                         event.event = RDMA_CM_EVENT_REJECTED;
2075                         break;
2076                 case -ETIMEDOUT:
2077                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2078                         break;
2079                 default:
2080                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2081                         break;
2082                 }
2083                 break;
2084         case IW_CM_EVENT_ESTABLISHED:
2085                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2086                 event.param.conn.initiator_depth = iw_event->ird;
2087                 event.param.conn.responder_resources = iw_event->ord;
2088                 break;
2089         default:
2090                 BUG_ON(1);
2091         }
2092
2093         event.status = iw_event->status;
2094         event.param.conn.private_data = iw_event->private_data;
2095         event.param.conn.private_data_len = iw_event->private_data_len;
2096         ret = id_priv->id.event_handler(&id_priv->id, &event);
2097         if (ret) {
2098                 /* Destroy the CM ID by returning a non-zero value. */
2099                 id_priv->cm_id.iw = NULL;
2100                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2101                 mutex_unlock(&id_priv->handler_mutex);
2102                 rdma_destroy_id(&id_priv->id);
2103                 return ret;
2104         }
2105
2106 out:
2107         mutex_unlock(&id_priv->handler_mutex);
2108         return ret;
2109 }
2110
2111 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2112                                struct iw_cm_event *iw_event)
2113 {
2114         struct rdma_cm_id *new_cm_id;
2115         struct rdma_id_private *listen_id, *conn_id;
2116         struct rdma_cm_event event;
2117         int ret = -ECONNABORTED;
2118         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2119         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2120
2121         listen_id = cm_id->context;
2122
2123         mutex_lock(&listen_id->handler_mutex);
2124         if (listen_id->state != RDMA_CM_LISTEN)
2125                 goto out;
2126
2127         /* Create a new RDMA id for the new IW CM ID */
2128         new_cm_id = rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2129                                    listen_id->id.event_handler,
2130                                    listen_id->id.context,
2131                                    RDMA_PS_TCP, IB_QPT_RC);
2132         if (IS_ERR(new_cm_id)) {
2133                 ret = -ENOMEM;
2134                 goto out;
2135         }
2136         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2137         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2138         conn_id->state = RDMA_CM_CONNECT;
2139
2140         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr, NULL);
2141         if (ret) {
2142                 mutex_unlock(&conn_id->handler_mutex);
2143                 rdma_destroy_id(new_cm_id);
2144                 goto out;
2145         }
2146
2147         ret = cma_acquire_dev(conn_id, listen_id);
2148         if (ret) {
2149                 mutex_unlock(&conn_id->handler_mutex);
2150                 rdma_destroy_id(new_cm_id);
2151                 goto out;
2152         }
2153
2154         conn_id->cm_id.iw = cm_id;
2155         cm_id->context = conn_id;
2156         cm_id->cm_handler = cma_iw_handler;
2157
2158         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2159         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2160
2161         memset(&event, 0, sizeof event);
2162         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2163         event.param.conn.private_data = iw_event->private_data;
2164         event.param.conn.private_data_len = iw_event->private_data_len;
2165         event.param.conn.initiator_depth = iw_event->ird;
2166         event.param.conn.responder_resources = iw_event->ord;
2167
2168         /*
2169          * Protect against the user destroying conn_id from another thread
2170          * until we're done accessing it.
2171          */
2172         atomic_inc(&conn_id->refcount);
2173         ret = conn_id->id.event_handler(&conn_id->id, &event);
2174         if (ret) {
2175                 /* User wants to destroy the CM ID */
2176                 conn_id->cm_id.iw = NULL;
2177                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2178                 mutex_unlock(&conn_id->handler_mutex);
2179                 cma_deref_id(conn_id);
2180                 rdma_destroy_id(&conn_id->id);
2181                 goto out;
2182         }
2183
2184         mutex_unlock(&conn_id->handler_mutex);
2185         cma_deref_id(conn_id);
2186
2187 out:
2188         mutex_unlock(&listen_id->handler_mutex);
2189         return ret;
2190 }
2191
2192 static int cma_ib_listen(struct rdma_id_private *id_priv)
2193 {
2194         struct sockaddr *addr;
2195         struct ib_cm_id *id;
2196         __be64 svc_id;
2197
2198         addr = cma_src_addr(id_priv);
2199         svc_id = rdma_get_service_id(&id_priv->id, addr);
2200         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2201         if (IS_ERR(id))
2202                 return PTR_ERR(id);
2203         id_priv->cm_id.ib = id;
2204
2205         return 0;
2206 }
2207
2208 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2209 {
2210         int ret;
2211         struct iw_cm_id *id;
2212
2213         id = iw_create_cm_id(id_priv->id.device,
2214                              iw_conn_req_handler,
2215                              id_priv);
2216         if (IS_ERR(id))
2217                 return PTR_ERR(id);
2218
2219         id->tos = id_priv->tos;
2220         id_priv->cm_id.iw = id;
2221
2222         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2223                rdma_addr_size(cma_src_addr(id_priv)));
2224
2225         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2226
2227         if (ret) {
2228                 iw_destroy_cm_id(id_priv->cm_id.iw);
2229                 id_priv->cm_id.iw = NULL;
2230         }
2231
2232         return ret;
2233 }
2234
2235 static int cma_listen_handler(struct rdma_cm_id *id,
2236                               struct rdma_cm_event *event)
2237 {
2238         struct rdma_id_private *id_priv = id->context;
2239
2240         id->context = id_priv->id.context;
2241         id->event_handler = id_priv->id.event_handler;
2242         return id_priv->id.event_handler(id, event);
2243 }
2244
2245 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2246                               struct cma_device *cma_dev)
2247 {
2248         struct rdma_id_private *dev_id_priv;
2249         struct rdma_cm_id *id;
2250         struct net *net = id_priv->id.route.addr.dev_addr.net;
2251         int ret;
2252
2253         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2254                 return;
2255
2256         id = rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2257                             id_priv->id.qp_type);
2258         if (IS_ERR(id))
2259                 return;
2260
2261         dev_id_priv = container_of(id, struct rdma_id_private, id);
2262
2263         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2264         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2265                rdma_addr_size(cma_src_addr(id_priv)));
2266
2267         _cma_attach_to_dev(dev_id_priv, cma_dev);
2268         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2269         atomic_inc(&id_priv->refcount);
2270         dev_id_priv->internal_id = 1;
2271         dev_id_priv->afonly = id_priv->afonly;
2272
2273         ret = rdma_listen(id, id_priv->backlog);
2274         if (ret)
2275                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2276                         ret, cma_dev->device->name);
2277 }
2278
2279 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2280 {
2281         struct cma_device *cma_dev;
2282
2283         mutex_lock(&lock);
2284         list_add_tail(&id_priv->list, &listen_any_list);
2285         list_for_each_entry(cma_dev, &dev_list, list)
2286                 cma_listen_on_dev(id_priv, cma_dev);
2287         mutex_unlock(&lock);
2288 }
2289
2290 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2291 {
2292         struct rdma_id_private *id_priv;
2293
2294         id_priv = container_of(id, struct rdma_id_private, id);
2295         id_priv->tos = (u8) tos;
2296         id_priv->tos_set = true;
2297 }
2298 EXPORT_SYMBOL(rdma_set_service_type);
2299
2300 static void cma_query_handler(int status, struct ib_sa_path_rec *path_rec,
2301                               void *context)
2302 {
2303         struct cma_work *work = context;
2304         struct rdma_route *route;
2305
2306         route = &work->id->id.route;
2307
2308         if (!status) {
2309                 route->num_paths = 1;
2310                 *route->path_rec = *path_rec;
2311         } else {
2312                 work->old_state = RDMA_CM_ROUTE_QUERY;
2313                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2314                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2315                 work->event.status = status;
2316                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2317                                      status);
2318         }
2319
2320         queue_work(cma_wq, &work->work);
2321 }
2322
2323 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2324                               struct cma_work *work)
2325 {
2326         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2327         struct ib_sa_path_rec path_rec;
2328         ib_sa_comp_mask comp_mask;
2329         struct sockaddr_in6 *sin6;
2330         struct sockaddr_ib *sib;
2331
2332         memset(&path_rec, 0, sizeof path_rec);
2333         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2334         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2335         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2336         path_rec.numb_path = 1;
2337         path_rec.reversible = 1;
2338         path_rec.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
2339
2340         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2341                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2342                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2343
2344         switch (cma_family(id_priv)) {
2345         case AF_INET:
2346                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2347                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2348                 break;
2349         case AF_INET6:
2350                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2351                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2352                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2353                 break;
2354         case AF_IB:
2355                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2356                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2357                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2358                 break;
2359         }
2360
2361         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2362                                                id_priv->id.port_num, &path_rec,
2363                                                comp_mask, timeout_ms,
2364                                                GFP_KERNEL, cma_query_handler,
2365                                                work, &id_priv->query);
2366
2367         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2368 }
2369
2370 static void cma_work_handler(struct work_struct *_work)
2371 {
2372         struct cma_work *work = container_of(_work, struct cma_work, work);
2373         struct rdma_id_private *id_priv = work->id;
2374         int destroy = 0;
2375
2376         mutex_lock(&id_priv->handler_mutex);
2377         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2378                 goto out;
2379
2380         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2381                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2382                 destroy = 1;
2383         }
2384 out:
2385         mutex_unlock(&id_priv->handler_mutex);
2386         cma_deref_id(id_priv);
2387         if (destroy)
2388                 rdma_destroy_id(&id_priv->id);
2389         kfree(work);
2390 }
2391
2392 static void cma_ndev_work_handler(struct work_struct *_work)
2393 {
2394         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2395         struct rdma_id_private *id_priv = work->id;
2396         int destroy = 0;
2397
2398         mutex_lock(&id_priv->handler_mutex);
2399         if (id_priv->state == RDMA_CM_DESTROYING ||
2400             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2401                 goto out;
2402
2403         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2404                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2405                 destroy = 1;
2406         }
2407
2408 out:
2409         mutex_unlock(&id_priv->handler_mutex);
2410         cma_deref_id(id_priv);
2411         if (destroy)
2412                 rdma_destroy_id(&id_priv->id);
2413         kfree(work);
2414 }
2415
2416 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2417 {
2418         struct rdma_route *route = &id_priv->id.route;
2419         struct cma_work *work;
2420         int ret;
2421
2422         work = kzalloc(sizeof *work, GFP_KERNEL);
2423         if (!work)
2424                 return -ENOMEM;
2425
2426         work->id = id_priv;
2427         INIT_WORK(&work->work, cma_work_handler);
2428         work->old_state = RDMA_CM_ROUTE_QUERY;
2429         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2430         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2431
2432         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2433         if (!route->path_rec) {
2434                 ret = -ENOMEM;
2435                 goto err1;
2436         }
2437
2438         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2439         if (ret)
2440                 goto err2;
2441
2442         return 0;
2443 err2:
2444         kfree(route->path_rec);
2445         route->path_rec = NULL;
2446 err1:
2447         kfree(work);
2448         return ret;
2449 }
2450
2451 int rdma_set_ib_paths(struct rdma_cm_id *id,
2452                       struct ib_sa_path_rec *path_rec, int num_paths)
2453 {
2454         struct rdma_id_private *id_priv;
2455         int ret;
2456
2457         id_priv = container_of(id, struct rdma_id_private, id);
2458         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2459                            RDMA_CM_ROUTE_RESOLVED))
2460                 return -EINVAL;
2461
2462         id->route.path_rec = kmemdup(path_rec, sizeof *path_rec * num_paths,
2463                                      GFP_KERNEL);
2464         if (!id->route.path_rec) {
2465                 ret = -ENOMEM;
2466                 goto err;
2467         }
2468
2469         id->route.num_paths = num_paths;
2470         return 0;
2471 err:
2472         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2473         return ret;
2474 }
2475 EXPORT_SYMBOL(rdma_set_ib_paths);
2476
2477 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2478 {
2479         struct cma_work *work;
2480
2481         work = kzalloc(sizeof *work, GFP_KERNEL);
2482         if (!work)
2483                 return -ENOMEM;
2484
2485         work->id = id_priv;
2486         INIT_WORK(&work->work, cma_work_handler);
2487         work->old_state = RDMA_CM_ROUTE_QUERY;
2488         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2489         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2490         queue_work(cma_wq, &work->work);
2491         return 0;
2492 }
2493
2494 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2495 {
2496         int prio;
2497         struct net_device *dev;
2498
2499         prio = rt_tos2priority(tos);
2500         dev = ndev->priv_flags & IFF_802_1Q_VLAN ?
2501                 vlan_dev_real_dev(ndev) : ndev;
2502
2503         if (dev->num_tc)
2504                 return netdev_get_prio_tc_map(dev, prio);
2505
2506 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2507         if (ndev->priv_flags & IFF_802_1Q_VLAN)
2508                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2509                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2510 #endif
2511         return 0;
2512 }
2513
2514 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2515                                            unsigned long supported_gids,
2516                                            enum ib_gid_type default_gid)
2517 {
2518         if ((network_type == RDMA_NETWORK_IPV4 ||
2519              network_type == RDMA_NETWORK_IPV6) &&
2520             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2521                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2522
2523         return default_gid;
2524 }
2525
2526 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2527 {
2528         struct rdma_route *route = &id_priv->id.route;
2529         struct rdma_addr *addr = &route->addr;
2530         struct cma_work *work;
2531         int ret;
2532         struct net_device *ndev = NULL;
2533         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2534                                         rdma_start_port(id_priv->cma_dev->device)];
2535         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2536
2537
2538         work = kzalloc(sizeof *work, GFP_KERNEL);
2539         if (!work)
2540                 return -ENOMEM;
2541
2542         work->id = id_priv;
2543         INIT_WORK(&work->work, cma_work_handler);
2544
2545         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2546         if (!route->path_rec) {
2547                 ret = -ENOMEM;
2548                 goto err1;
2549         }
2550
2551         route->num_paths = 1;
2552
2553         if (addr->dev_addr.bound_dev_if) {
2554                 unsigned long supported_gids;
2555
2556                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2557                 if (!ndev) {
2558                         ret = -ENODEV;
2559                         goto err2;
2560                 }
2561
2562                 if (ndev->flags & IFF_LOOPBACK) {
2563                         dev_put(ndev);
2564                         if (!id_priv->id.device->get_netdev) {
2565                                 ret = -EOPNOTSUPP;
2566                                 goto err2;
2567                         }
2568
2569                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2570                                                               id_priv->id.port_num);
2571                         if (!ndev) {
2572                                 ret = -ENODEV;
2573                                 goto err2;
2574                         }
2575                 }
2576
2577                 route->path_rec->net = &init_net;
2578                 route->path_rec->ifindex = ndev->ifindex;
2579                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2580                                                             id_priv->id.port_num);
2581                 route->path_rec->gid_type =
2582                         cma_route_gid_type(addr->dev_addr.network,
2583                                            supported_gids,
2584                                            id_priv->gid_type);
2585         }
2586         if (!ndev) {
2587                 ret = -ENODEV;
2588                 goto err2;
2589         }
2590
2591         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2592
2593         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2594                     &route->path_rec->sgid);
2595         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2596                     &route->path_rec->dgid);
2597
2598         /* Use the hint from IP Stack to select GID Type */
2599         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2600                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2601         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2602                 /* TODO: get the hoplimit from the inet/inet6 device */
2603                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2604         else
2605                 route->path_rec->hop_limit = 1;
2606         route->path_rec->reversible = 1;
2607         route->path_rec->pkey = cpu_to_be16(0xffff);
2608         route->path_rec->mtu_selector = IB_SA_EQ;
2609         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2610         route->path_rec->traffic_class = tos;
2611         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2612         route->path_rec->rate_selector = IB_SA_EQ;
2613         route->path_rec->rate = iboe_get_rate(ndev);
2614         dev_put(ndev);
2615         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2616         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2617         if (!route->path_rec->mtu) {
2618                 ret = -EINVAL;
2619                 goto err2;
2620         }
2621
2622         work->old_state = RDMA_CM_ROUTE_QUERY;
2623         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2624         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2625         work->event.status = 0;
2626
2627         queue_work(cma_wq, &work->work);
2628
2629         return 0;
2630
2631 err2:
2632         kfree(route->path_rec);
2633         route->path_rec = NULL;
2634 err1:
2635         kfree(work);
2636         return ret;
2637 }
2638
2639 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2640 {
2641         struct rdma_id_private *id_priv;
2642         int ret;
2643
2644         id_priv = container_of(id, struct rdma_id_private, id);
2645         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2646                 return -EINVAL;
2647
2648         atomic_inc(&id_priv->refcount);
2649         if (rdma_cap_ib_sa(id->device, id->port_num))
2650                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2651         else if (rdma_protocol_roce(id->device, id->port_num))
2652                 ret = cma_resolve_iboe_route(id_priv);
2653         else if (rdma_protocol_iwarp(id->device, id->port_num))
2654                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2655         else
2656                 ret = -ENOSYS;
2657
2658         if (ret)
2659                 goto err;
2660
2661         return 0;
2662 err:
2663         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2664         cma_deref_id(id_priv);
2665         return ret;
2666 }
2667 EXPORT_SYMBOL(rdma_resolve_route);
2668
2669 static void cma_set_loopback(struct sockaddr *addr)
2670 {
2671         switch (addr->sa_family) {
2672         case AF_INET:
2673                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2674                 break;
2675         case AF_INET6:
2676                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2677                               0, 0, 0, htonl(1));
2678                 break;
2679         default:
2680                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2681                             0, 0, 0, htonl(1));
2682                 break;
2683         }
2684 }
2685
2686 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2687 {
2688         struct cma_device *cma_dev, *cur_dev;
2689         union ib_gid gid;
2690         enum ib_port_state port_state;
2691         u16 pkey;
2692         int ret;
2693         u8 p;
2694
2695         cma_dev = NULL;
2696         mutex_lock(&lock);
2697         list_for_each_entry(cur_dev, &dev_list, list) {
2698                 if (cma_family(id_priv) == AF_IB &&
2699                     !rdma_cap_ib_cm(cur_dev->device, 1))
2700                         continue;
2701
2702                 if (!cma_dev)
2703                         cma_dev = cur_dev;
2704
2705                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2706                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2707                             port_state == IB_PORT_ACTIVE) {
2708                                 cma_dev = cur_dev;
2709                                 goto port_found;
2710                         }
2711                 }
2712         }
2713
2714         if (!cma_dev) {
2715                 ret = -ENODEV;
2716                 goto out;
2717         }
2718
2719         p = 1;
2720
2721 port_found:
2722         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2723         if (ret)
2724                 goto out;
2725
2726         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2727         if (ret)
2728                 goto out;
2729
2730         id_priv->id.route.addr.dev_addr.dev_type =
2731                 (rdma_protocol_ib(cma_dev->device, p)) ?
2732                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2733
2734         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2735         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2736         id_priv->id.port_num = p;
2737         cma_attach_to_dev(id_priv, cma_dev);
2738         cma_set_loopback(cma_src_addr(id_priv));
2739 out:
2740         mutex_unlock(&lock);
2741         return ret;
2742 }
2743
2744 static void addr_handler(int status, struct sockaddr *src_addr,
2745                          struct rdma_dev_addr *dev_addr, void *context)
2746 {
2747         struct rdma_id_private *id_priv = context;
2748         struct rdma_cm_event event;
2749
2750         memset(&event, 0, sizeof event);
2751         mutex_lock(&id_priv->handler_mutex);
2752         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2753                            RDMA_CM_ADDR_RESOLVED))
2754                 goto out;
2755
2756         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2757         if (!status && !id_priv->cma_dev) {
2758                 status = cma_acquire_dev(id_priv, NULL);
2759                 if (status)
2760                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2761                                              status);
2762         } else {
2763                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2764         }
2765
2766         if (status) {
2767                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2768                                    RDMA_CM_ADDR_BOUND))
2769                         goto out;
2770                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2771                 event.status = status;
2772         } else
2773                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2774
2775         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2776                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2777                 mutex_unlock(&id_priv->handler_mutex);
2778                 cma_deref_id(id_priv);
2779                 rdma_destroy_id(&id_priv->id);
2780                 return;
2781         }
2782 out:
2783         mutex_unlock(&id_priv->handler_mutex);
2784         cma_deref_id(id_priv);
2785 }
2786
2787 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2788 {
2789         struct cma_work *work;
2790         union ib_gid gid;
2791         int ret;
2792
2793         work = kzalloc(sizeof *work, GFP_KERNEL);
2794         if (!work)
2795                 return -ENOMEM;
2796
2797         if (!id_priv->cma_dev) {
2798                 ret = cma_bind_loopback(id_priv);
2799                 if (ret)
2800                         goto err;
2801         }
2802
2803         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2804         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2805
2806         work->id = id_priv;
2807         INIT_WORK(&work->work, cma_work_handler);
2808         work->old_state = RDMA_CM_ADDR_QUERY;
2809         work->new_state = RDMA_CM_ADDR_RESOLVED;
2810         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2811         queue_work(cma_wq, &work->work);
2812         return 0;
2813 err:
2814         kfree(work);
2815         return ret;
2816 }
2817
2818 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2819 {
2820         struct cma_work *work;
2821         int ret;
2822
2823         work = kzalloc(sizeof *work, GFP_KERNEL);
2824         if (!work)
2825                 return -ENOMEM;
2826
2827         if (!id_priv->cma_dev) {
2828                 ret = cma_resolve_ib_dev(id_priv);
2829                 if (ret)
2830                         goto err;
2831         }
2832
2833         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2834                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2835
2836         work->id = id_priv;
2837         INIT_WORK(&work->work, cma_work_handler);
2838         work->old_state = RDMA_CM_ADDR_QUERY;
2839         work->new_state = RDMA_CM_ADDR_RESOLVED;
2840         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2841         queue_work(cma_wq, &work->work);
2842         return 0;
2843 err:
2844         kfree(work);
2845         return ret;
2846 }
2847
2848 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2849                          struct sockaddr *dst_addr)
2850 {
2851         if (!src_addr || !src_addr->sa_family) {
2852                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2853                 src_addr->sa_family = dst_addr->sa_family;
2854                 if (dst_addr->sa_family == AF_INET6) {
2855                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2856                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2857                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2858                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2859                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2860                 } else if (dst_addr->sa_family == AF_IB) {
2861                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2862                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2863                 }
2864         }
2865         return rdma_bind_addr(id, src_addr);
2866 }
2867
2868 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2869                       struct sockaddr *dst_addr, int timeout_ms)
2870 {
2871         struct rdma_id_private *id_priv;
2872         int ret;
2873
2874         id_priv = container_of(id, struct rdma_id_private, id);
2875         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2876         if (id_priv->state == RDMA_CM_IDLE) {
2877                 ret = cma_bind_addr(id, src_addr, dst_addr);
2878                 if (ret) {
2879                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2880                         return ret;
2881                 }
2882         }
2883
2884         if (cma_family(id_priv) != dst_addr->sa_family) {
2885                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2886                 return -EINVAL;
2887         }
2888
2889         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2890                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2891                 return -EINVAL;
2892         }
2893
2894         atomic_inc(&id_priv->refcount);
2895         if (cma_any_addr(dst_addr)) {
2896                 ret = cma_resolve_loopback(id_priv);
2897         } else {
2898                 if (dst_addr->sa_family == AF_IB) {
2899                         ret = cma_resolve_ib_addr(id_priv);
2900                 } else {
2901                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2902                                               dst_addr, &id->route.addr.dev_addr,
2903                                               timeout_ms, addr_handler, id_priv);
2904                 }
2905         }
2906         if (ret)
2907                 goto err;
2908
2909         return 0;
2910 err:
2911         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2912         cma_deref_id(id_priv);
2913         return ret;
2914 }
2915 EXPORT_SYMBOL(rdma_resolve_addr);
2916
2917 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2918 {
2919         struct rdma_id_private *id_priv;
2920         unsigned long flags;
2921         int ret;
2922
2923         id_priv = container_of(id, struct rdma_id_private, id);
2924         spin_lock_irqsave(&id_priv->lock, flags);
2925         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2926                 id_priv->reuseaddr = reuse;
2927                 ret = 0;
2928         } else {
2929                 ret = -EINVAL;
2930         }
2931         spin_unlock_irqrestore(&id_priv->lock, flags);
2932         return ret;
2933 }
2934 EXPORT_SYMBOL(rdma_set_reuseaddr);
2935
2936 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2937 {
2938         struct rdma_id_private *id_priv;
2939         unsigned long flags;
2940         int ret;
2941
2942         id_priv = container_of(id, struct rdma_id_private, id);
2943         spin_lock_irqsave(&id_priv->lock, flags);
2944         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2945                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2946                 id_priv->afonly = afonly;
2947                 ret = 0;
2948         } else {
2949                 ret = -EINVAL;
2950         }
2951         spin_unlock_irqrestore(&id_priv->lock, flags);
2952         return ret;
2953 }
2954 EXPORT_SYMBOL(rdma_set_afonly);
2955
2956 static void cma_bind_port(struct rdma_bind_list *bind_list,
2957                           struct rdma_id_private *id_priv)
2958 {
2959         struct sockaddr *addr;
2960         struct sockaddr_ib *sib;
2961         u64 sid, mask;
2962         __be16 port;
2963
2964         addr = cma_src_addr(id_priv);
2965         port = htons(bind_list->port);
2966
2967         switch (addr->sa_family) {
2968         case AF_INET:
2969                 ((struct sockaddr_in *) addr)->sin_port = port;
2970                 break;
2971         case AF_INET6:
2972                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2973                 break;
2974         case AF_IB:
2975                 sib = (struct sockaddr_ib *) addr;
2976                 sid = be64_to_cpu(sib->sib_sid);
2977                 mask = be64_to_cpu(sib->sib_sid_mask);
2978                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2979                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2980                 break;
2981         }
2982         id_priv->bind_list = bind_list;
2983         hlist_add_head(&id_priv->node, &bind_list->owners);
2984 }
2985
2986 static int cma_alloc_port(enum rdma_port_space ps,
2987                           struct rdma_id_private *id_priv, unsigned short snum)
2988 {
2989         struct rdma_bind_list *bind_list;
2990         int ret;
2991
2992         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2993         if (!bind_list)
2994                 return -ENOMEM;
2995
2996         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2997                            snum);
2998         if (ret < 0)
2999                 goto err;
3000
3001         bind_list->ps = ps;
3002         bind_list->port = (unsigned short)ret;
3003         cma_bind_port(bind_list, id_priv);
3004         return 0;
3005 err:
3006         kfree(bind_list);
3007         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3008 }
3009
3010 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3011                               struct rdma_id_private *id_priv)
3012 {
3013         struct rdma_id_private *cur_id;
3014         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3015         struct sockaddr  *saddr = cma_src_addr(id_priv);
3016         __be16 dport = cma_port(daddr);
3017
3018         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3019                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3020                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3021                 __be16 cur_dport = cma_port(cur_daddr);
3022
3023                 if (id_priv == cur_id)
3024                         continue;
3025
3026                 /* different dest port -> unique */
3027                 if (!cma_any_port(cur_daddr) &&
3028                     (dport != cur_dport))
3029                         continue;
3030
3031                 /* different src address -> unique */
3032                 if (!cma_any_addr(saddr) &&
3033                     !cma_any_addr(cur_saddr) &&
3034                     cma_addr_cmp(saddr, cur_saddr))
3035                         continue;
3036
3037                 /* different dst address -> unique */
3038                 if (!cma_any_addr(cur_daddr) &&
3039                     cma_addr_cmp(daddr, cur_daddr))
3040                         continue;
3041
3042                 return -EADDRNOTAVAIL;
3043         }
3044         return 0;
3045 }
3046
3047 static int cma_alloc_any_port(enum rdma_port_space ps,
3048                               struct rdma_id_private *id_priv)
3049 {
3050         static unsigned int last_used_port;
3051         int low, high, remaining;
3052         unsigned int rover;
3053         struct net *net = id_priv->id.route.addr.dev_addr.net;
3054
3055         inet_get_local_port_range(net, &low, &high);
3056         remaining = (high - low) + 1;
3057         rover = prandom_u32() % remaining + low;
3058 retry:
3059         if (last_used_port != rover) {
3060                 struct rdma_bind_list *bind_list;
3061                 int ret;
3062
3063                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3064
3065                 if (!bind_list) {
3066                         ret = cma_alloc_port(ps, id_priv, rover);
3067                 } else {
3068                         ret = cma_port_is_unique(bind_list, id_priv);
3069                         if (!ret)
3070                                 cma_bind_port(bind_list, id_priv);
3071                 }
3072                 /*
3073                  * Remember previously used port number in order to avoid
3074                  * re-using same port immediately after it is closed.
3075                  */
3076                 if (!ret)
3077                         last_used_port = rover;
3078                 if (ret != -EADDRNOTAVAIL)
3079                         return ret;
3080         }
3081         if (--remaining) {
3082                 rover++;
3083                 if ((rover < low) || (rover > high))
3084                         rover = low;
3085                 goto retry;
3086         }
3087         return -EADDRNOTAVAIL;
3088 }
3089
3090 /*
3091  * Check that the requested port is available.  This is called when trying to
3092  * bind to a specific port, or when trying to listen on a bound port.  In
3093  * the latter case, the provided id_priv may already be on the bind_list, but
3094  * we still need to check that it's okay to start listening.
3095  */
3096 static int cma_check_port(struct rdma_bind_list *bind_list,
3097                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3098 {
3099         struct rdma_id_private *cur_id;
3100         struct sockaddr *addr, *cur_addr;
3101
3102         addr = cma_src_addr(id_priv);
3103         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3104                 if (id_priv == cur_id)
3105                         continue;
3106
3107                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3108                     cur_id->reuseaddr)
3109                         continue;
3110
3111                 cur_addr = cma_src_addr(cur_id);
3112                 if (id_priv->afonly && cur_id->afonly &&
3113                     (addr->sa_family != cur_addr->sa_family))
3114                         continue;
3115
3116                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3117                         return -EADDRNOTAVAIL;
3118
3119                 if (!cma_addr_cmp(addr, cur_addr))
3120                         return -EADDRINUSE;
3121         }
3122         return 0;
3123 }
3124
3125 static int cma_use_port(enum rdma_port_space ps,
3126                         struct rdma_id_private *id_priv)
3127 {
3128         struct rdma_bind_list *bind_list;
3129         unsigned short snum;
3130         int ret;
3131
3132         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3133         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3134                 return -EACCES;
3135
3136         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3137         if (!bind_list) {
3138                 ret = cma_alloc_port(ps, id_priv, snum);
3139         } else {
3140                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3141                 if (!ret)
3142                         cma_bind_port(bind_list, id_priv);
3143         }
3144         return ret;
3145 }
3146
3147 static int cma_bind_listen(struct rdma_id_private *id_priv)
3148 {
3149         struct rdma_bind_list *bind_list = id_priv->bind_list;
3150         int ret = 0;
3151
3152         mutex_lock(&lock);
3153         if (bind_list->owners.first->next)
3154                 ret = cma_check_port(bind_list, id_priv, 0);
3155         mutex_unlock(&lock);
3156         return ret;
3157 }
3158
3159 static enum rdma_port_space cma_select_inet_ps(
3160                 struct rdma_id_private *id_priv)
3161 {
3162         switch (id_priv->id.ps) {
3163         case RDMA_PS_TCP:
3164         case RDMA_PS_UDP:
3165         case RDMA_PS_IPOIB:
3166         case RDMA_PS_IB:
3167                 return id_priv->id.ps;
3168         default:
3169
3170                 return 0;
3171         }
3172 }
3173
3174 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3175 {
3176         enum rdma_port_space ps = 0;
3177         struct sockaddr_ib *sib;
3178         u64 sid_ps, mask, sid;
3179
3180         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3181         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3182         sid = be64_to_cpu(sib->sib_sid) & mask;
3183
3184         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3185                 sid_ps = RDMA_IB_IP_PS_IB;
3186                 ps = RDMA_PS_IB;
3187         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3188                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3189                 sid_ps = RDMA_IB_IP_PS_TCP;
3190                 ps = RDMA_PS_TCP;
3191         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3192                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3193                 sid_ps = RDMA_IB_IP_PS_UDP;
3194                 ps = RDMA_PS_UDP;
3195         }
3196
3197         if (ps) {
3198                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3199                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3200                                                 be64_to_cpu(sib->sib_sid_mask));
3201         }
3202         return ps;
3203 }
3204
3205 static int cma_get_port(struct rdma_id_private *id_priv)
3206 {
3207         enum rdma_port_space ps;
3208         int ret;
3209
3210         if (cma_family(id_priv) != AF_IB)
3211                 ps = cma_select_inet_ps(id_priv);
3212         else
3213                 ps = cma_select_ib_ps(id_priv);
3214         if (!ps)
3215                 return -EPROTONOSUPPORT;
3216
3217         mutex_lock(&lock);
3218         if (cma_any_port(cma_src_addr(id_priv)))
3219                 ret = cma_alloc_any_port(ps, id_priv);
3220         else
3221                 ret = cma_use_port(ps, id_priv);
3222         mutex_unlock(&lock);
3223
3224         return ret;
3225 }
3226
3227 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3228                                struct sockaddr *addr)
3229 {
3230 #if IS_ENABLED(CONFIG_IPV6)
3231         struct sockaddr_in6 *sin6;
3232
3233         if (addr->sa_family != AF_INET6)
3234                 return 0;
3235
3236         sin6 = (struct sockaddr_in6 *) addr;
3237
3238         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3239                 return 0;
3240
3241         if (!sin6->sin6_scope_id)
3242                         return -EINVAL;
3243
3244         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3245 #endif
3246         return 0;
3247 }
3248
3249 int rdma_listen(struct rdma_cm_id *id, int backlog)
3250 {
3251         struct rdma_id_private *id_priv;
3252         int ret;
3253
3254         id_priv = container_of(id, struct rdma_id_private, id);
3255         if (id_priv->state == RDMA_CM_IDLE) {
3256                 id->route.addr.src_addr.ss_family = AF_INET;
3257                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3258                 if (ret)
3259                         return ret;
3260         }
3261
3262         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3263                 return -EINVAL;
3264
3265         if (id_priv->reuseaddr) {
3266                 ret = cma_bind_listen(id_priv);
3267                 if (ret)
3268                         goto err;
3269         }
3270
3271         id_priv->backlog = backlog;
3272         if (id->device) {
3273                 if (rdma_cap_ib_cm(id->device, 1)) {
3274                         ret = cma_ib_listen(id_priv);
3275                         if (ret)
3276                                 goto err;
3277                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3278                         ret = cma_iw_listen(id_priv, backlog);
3279                         if (ret)
3280                                 goto err;
3281                 } else {
3282                         ret = -ENOSYS;
3283                         goto err;
3284                 }
3285         } else
3286                 cma_listen_on_all(id_priv);
3287
3288         return 0;
3289 err:
3290         id_priv->backlog = 0;
3291         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3292         return ret;
3293 }
3294 EXPORT_SYMBOL(rdma_listen);
3295
3296 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3297 {
3298         struct rdma_id_private *id_priv;
3299         int ret;
3300
3301         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3302             addr->sa_family != AF_IB)
3303                 return -EAFNOSUPPORT;
3304
3305         id_priv = container_of(id, struct rdma_id_private, id);
3306         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3307                 return -EINVAL;
3308
3309         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3310         if (ret)
3311                 goto err1;
3312
3313         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3314         if (!cma_any_addr(addr)) {
3315                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3316                 if (ret)
3317                         goto err1;
3318
3319                 ret = cma_acquire_dev(id_priv, NULL);
3320                 if (ret)
3321                         goto err1;
3322         }
3323
3324         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3325                 if (addr->sa_family == AF_INET)
3326                         id_priv->afonly = 1;
3327 #if IS_ENABLED(CONFIG_IPV6)
3328                 else if (addr->sa_family == AF_INET6) {
3329                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3330
3331                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3332                 }
3333 #endif
3334         }
3335         ret = cma_get_port(id_priv);
3336         if (ret)
3337                 goto err2;
3338
3339         return 0;
3340 err2:
3341         if (id_priv->cma_dev)
3342                 cma_release_dev(id_priv);
3343 err1:
3344         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3345         return ret;
3346 }
3347 EXPORT_SYMBOL(rdma_bind_addr);
3348
3349 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3350 {
3351         struct cma_hdr *cma_hdr;
3352
3353         cma_hdr = hdr;
3354         cma_hdr->cma_version = CMA_VERSION;
3355         if (cma_family(id_priv) == AF_INET) {
3356                 struct sockaddr_in *src4, *dst4;
3357
3358                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3359                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3360
3361                 cma_set_ip_ver(cma_hdr, 4);
3362                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3363                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3364                 cma_hdr->port = src4->sin_port;
3365         } else if (cma_family(id_priv) == AF_INET6) {
3366                 struct sockaddr_in6 *src6, *dst6;
3367
3368                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3369                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3370
3371                 cma_set_ip_ver(cma_hdr, 6);
3372                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3373                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3374                 cma_hdr->port = src6->sin6_port;
3375         }
3376         return 0;
3377 }
3378
3379 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3380                                 struct ib_cm_event *ib_event)
3381 {
3382         struct rdma_id_private *id_priv = cm_id->context;
3383         struct rdma_cm_event event;
3384         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3385         int ret = 0;
3386
3387         mutex_lock(&id_priv->handler_mutex);
3388         if (id_priv->state != RDMA_CM_CONNECT)
3389                 goto out;
3390
3391         memset(&event, 0, sizeof event);
3392         switch (ib_event->event) {
3393         case IB_CM_SIDR_REQ_ERROR:
3394                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3395                 event.status = -ETIMEDOUT;
3396                 break;
3397         case IB_CM_SIDR_REP_RECEIVED:
3398                 event.param.ud.private_data = ib_event->private_data;
3399                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3400                 if (rep->status != IB_SIDR_SUCCESS) {
3401                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3402                         event.status = ib_event->param.sidr_rep_rcvd.status;
3403                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3404                                              event.status);
3405                         break;
3406                 }
3407                 ret = cma_set_qkey(id_priv, rep->qkey);
3408                 if (ret) {
3409                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3410                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3411                         event.status = ret;
3412                         break;
3413                 }
3414                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3415                                      id_priv->id.route.path_rec,
3416                                      &event.param.ud.ah_attr);
3417                 event.param.ud.qp_num = rep->qpn;
3418                 event.param.ud.qkey = rep->qkey;
3419                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3420                 event.status = 0;
3421                 break;
3422         default:
3423                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3424                        ib_event->event);
3425                 goto out;
3426         }
3427
3428         ret = id_priv->id.event_handler(&id_priv->id, &event);
3429         if (ret) {
3430                 /* Destroy the CM ID by returning a non-zero value. */
3431                 id_priv->cm_id.ib = NULL;
3432                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3433                 mutex_unlock(&id_priv->handler_mutex);
3434                 rdma_destroy_id(&id_priv->id);
3435                 return ret;
3436         }
3437 out:
3438         mutex_unlock(&id_priv->handler_mutex);
3439         return ret;
3440 }
3441
3442 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3443                               struct rdma_conn_param *conn_param)
3444 {
3445         struct ib_cm_sidr_req_param req;
3446         struct ib_cm_id *id;
3447         void *private_data;
3448         int offset, ret;
3449
3450         memset(&req, 0, sizeof req);
3451         offset = cma_user_data_offset(id_priv);
3452         req.private_data_len = offset + conn_param->private_data_len;
3453         if (req.private_data_len < conn_param->private_data_len)
3454                 return -EINVAL;
3455
3456         if (req.private_data_len) {
3457                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3458                 if (!private_data)
3459                         return -ENOMEM;
3460         } else {
3461                 private_data = NULL;
3462         }
3463
3464         if (conn_param->private_data && conn_param->private_data_len)
3465                 memcpy(private_data + offset, conn_param->private_data,
3466                        conn_param->private_data_len);
3467
3468         if (private_data) {
3469                 ret = cma_format_hdr(private_data, id_priv);
3470                 if (ret)
3471                         goto out;
3472                 req.private_data = private_data;
3473         }
3474
3475         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3476                              id_priv);
3477         if (IS_ERR(id)) {
3478                 ret = PTR_ERR(id);
3479                 goto out;
3480         }
3481         id_priv->cm_id.ib = id;
3482
3483         req.path = id_priv->id.route.path_rec;
3484         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3485         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3486         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3487
3488         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3489         if (ret) {
3490                 ib_destroy_cm_id(id_priv->cm_id.ib);
3491                 id_priv->cm_id.ib = NULL;
3492         }
3493 out:
3494         kfree(private_data);
3495         return ret;
3496 }
3497
3498 static int cma_connect_ib(struct rdma_id_private *id_priv,
3499                           struct rdma_conn_param *conn_param)
3500 {
3501         struct ib_cm_req_param req;
3502         struct rdma_route *route;
3503         void *private_data;
3504         struct ib_cm_id *id;
3505         int offset, ret;
3506
3507         memset(&req, 0, sizeof req);
3508         offset = cma_user_data_offset(id_priv);
3509         req.private_data_len = offset + conn_param->private_data_len;
3510         if (req.private_data_len < conn_param->private_data_len)
3511                 return -EINVAL;
3512
3513         if (req.private_data_len) {
3514                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3515                 if (!private_data)
3516                         return -ENOMEM;
3517         } else {
3518                 private_data = NULL;
3519         }
3520
3521         if (conn_param->private_data && conn_param->private_data_len)
3522                 memcpy(private_data + offset, conn_param->private_data,
3523                        conn_param->private_data_len);
3524
3525         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3526         if (IS_ERR(id)) {
3527                 ret = PTR_ERR(id);
3528                 goto out;
3529         }
3530         id_priv->cm_id.ib = id;
3531
3532         route = &id_priv->id.route;
3533         if (private_data) {
3534                 ret = cma_format_hdr(private_data, id_priv);
3535                 if (ret)
3536                         goto out;
3537                 req.private_data = private_data;
3538         }
3539
3540         req.primary_path = &route->path_rec[0];
3541         if (route->num_paths == 2)
3542                 req.alternate_path = &route->path_rec[1];
3543
3544         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3545         req.qp_num = id_priv->qp_num;
3546         req.qp_type = id_priv->id.qp_type;
3547         req.starting_psn = id_priv->seq_num;
3548         req.responder_resources = conn_param->responder_resources;
3549         req.initiator_depth = conn_param->initiator_depth;
3550         req.flow_control = conn_param->flow_control;
3551         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3552         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3553         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3554         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3555         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3556         req.srq = id_priv->srq ? 1 : 0;
3557
3558         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3559 out:
3560         if (ret && !IS_ERR(id)) {
3561                 ib_destroy_cm_id(id);
3562                 id_priv->cm_id.ib = NULL;
3563         }
3564
3565         kfree(private_data);
3566         return ret;
3567 }
3568
3569 static int cma_connect_iw(struct rdma_id_private *id_priv,
3570                           struct rdma_conn_param *conn_param)
3571 {
3572         struct iw_cm_id *cm_id;
3573         int ret;
3574         struct iw_cm_conn_param iw_param;
3575
3576         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3577         if (IS_ERR(cm_id))
3578                 return PTR_ERR(cm_id);
3579
3580         cm_id->tos = id_priv->tos;
3581         id_priv->cm_id.iw = cm_id;
3582
3583         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3584                rdma_addr_size(cma_src_addr(id_priv)));
3585         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3586                rdma_addr_size(cma_dst_addr(id_priv)));
3587
3588         ret = cma_modify_qp_rtr(id_priv, conn_param);
3589         if (ret)
3590                 goto out;
3591
3592         if (conn_param) {
3593                 iw_param.ord = conn_param->initiator_depth;
3594                 iw_param.ird = conn_param->responder_resources;
3595                 iw_param.private_data = conn_param->private_data;
3596                 iw_param.private_data_len = conn_param->private_data_len;
3597                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3598         } else {
3599                 memset(&iw_param, 0, sizeof iw_param);
3600                 iw_param.qpn = id_priv->qp_num;
3601         }
3602         ret = iw_cm_connect(cm_id, &iw_param);
3603 out:
3604         if (ret) {
3605                 iw_destroy_cm_id(cm_id);
3606                 id_priv->cm_id.iw = NULL;
3607         }
3608         return ret;
3609 }
3610
3611 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3612 {
3613         struct rdma_id_private *id_priv;
3614         int ret;
3615
3616         id_priv = container_of(id, struct rdma_id_private, id);
3617         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3618                 return -EINVAL;
3619
3620         if (!id->qp) {
3621                 id_priv->qp_num = conn_param->qp_num;
3622                 id_priv->srq = conn_param->srq;
3623         }
3624
3625         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3626                 if (id->qp_type == IB_QPT_UD)
3627                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3628                 else
3629                         ret = cma_connect_ib(id_priv, conn_param);
3630         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3631                 ret = cma_connect_iw(id_priv, conn_param);
3632         else
3633                 ret = -ENOSYS;
3634         if (ret)
3635                 goto err;
3636
3637         return 0;
3638 err:
3639         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3640         return ret;
3641 }
3642 EXPORT_SYMBOL(rdma_connect);
3643
3644 static int cma_accept_ib(struct rdma_id_private *id_priv,
3645                          struct rdma_conn_param *conn_param)
3646 {
3647         struct ib_cm_rep_param rep;
3648         int ret;
3649
3650         ret = cma_modify_qp_rtr(id_priv, conn_param);
3651         if (ret)
3652                 goto out;
3653
3654         ret = cma_modify_qp_rts(id_priv, conn_param);
3655         if (ret)
3656                 goto out;
3657
3658         memset(&rep, 0, sizeof rep);
3659         rep.qp_num = id_priv->qp_num;
3660         rep.starting_psn = id_priv->seq_num;
3661         rep.private_data = conn_param->private_data;
3662         rep.private_data_len = conn_param->private_data_len;
3663         rep.responder_resources = conn_param->responder_resources;
3664         rep.initiator_depth = conn_param->initiator_depth;
3665         rep.failover_accepted = 0;
3666         rep.flow_control = conn_param->flow_control;
3667         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3668         rep.srq = id_priv->srq ? 1 : 0;
3669
3670         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3671 out:
3672         return ret;
3673 }
3674
3675 static int cma_accept_iw(struct rdma_id_private *id_priv,
3676                   struct rdma_conn_param *conn_param)
3677 {
3678         struct iw_cm_conn_param iw_param;
3679         int ret;
3680
3681         ret = cma_modify_qp_rtr(id_priv, conn_param);
3682         if (ret)
3683                 return ret;
3684
3685         iw_param.ord = conn_param->initiator_depth;
3686         iw_param.ird = conn_param->responder_resources;
3687         iw_param.private_data = conn_param->private_data;
3688         iw_param.private_data_len = conn_param->private_data_len;
3689         if (id_priv->id.qp) {
3690                 iw_param.qpn = id_priv->qp_num;
3691         } else
3692                 iw_param.qpn = conn_param->qp_num;
3693
3694         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3695 }
3696
3697 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3698                              enum ib_cm_sidr_status status, u32 qkey,
3699                              const void *private_data, int private_data_len)
3700 {
3701         struct ib_cm_sidr_rep_param rep;
3702         int ret;
3703
3704         memset(&rep, 0, sizeof rep);
3705         rep.status = status;
3706         if (status == IB_SIDR_SUCCESS) {
3707                 ret = cma_set_qkey(id_priv, qkey);
3708                 if (ret)
3709                         return ret;
3710                 rep.qp_num = id_priv->qp_num;
3711                 rep.qkey = id_priv->qkey;
3712         }
3713         rep.private_data = private_data;
3714         rep.private_data_len = private_data_len;
3715
3716         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3717 }
3718
3719 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3720 {
3721         struct rdma_id_private *id_priv;
3722         int ret;
3723
3724         id_priv = container_of(id, struct rdma_id_private, id);
3725
3726         id_priv->owner = task_pid_nr(current);
3727
3728         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3729                 return -EINVAL;
3730
3731         if (!id->qp && conn_param) {
3732                 id_priv->qp_num = conn_param->qp_num;
3733                 id_priv->srq = conn_param->srq;
3734         }
3735
3736         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3737                 if (id->qp_type == IB_QPT_UD) {
3738                         if (conn_param)
3739                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3740                                                         conn_param->qkey,
3741                                                         conn_param->private_data,
3742                                                         conn_param->private_data_len);
3743                         else
3744                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3745                                                         0, NULL, 0);
3746                 } else {
3747                         if (conn_param)
3748                                 ret = cma_accept_ib(id_priv, conn_param);
3749                         else
3750                                 ret = cma_rep_recv(id_priv);
3751                 }
3752         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3753                 ret = cma_accept_iw(id_priv, conn_param);
3754         else
3755                 ret = -ENOSYS;
3756
3757         if (ret)
3758                 goto reject;
3759
3760         return 0;
3761 reject:
3762         cma_modify_qp_err(id_priv);
3763         rdma_reject(id, NULL, 0);
3764         return ret;
3765 }
3766 EXPORT_SYMBOL(rdma_accept);
3767
3768 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3769 {
3770         struct rdma_id_private *id_priv;
3771         int ret;
3772
3773         id_priv = container_of(id, struct rdma_id_private, id);
3774         if (!id_priv->cm_id.ib)
3775                 return -EINVAL;
3776
3777         switch (id->device->node_type) {
3778         case RDMA_NODE_IB_CA:
3779                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3780                 break;
3781         default:
3782                 ret = 0;
3783                 break;
3784         }
3785         return ret;
3786 }
3787 EXPORT_SYMBOL(rdma_notify);
3788
3789 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3790                 u8 private_data_len)
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         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3800                 if (id->qp_type == IB_QPT_UD)
3801                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3802                                                 private_data, private_data_len);
3803                 else
3804                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3805                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3806                                              0, private_data, private_data_len);
3807         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3808                 ret = iw_cm_reject(id_priv->cm_id.iw,
3809                                    private_data, private_data_len);
3810         } else
3811                 ret = -ENOSYS;
3812
3813         return ret;
3814 }
3815 EXPORT_SYMBOL(rdma_reject);
3816
3817 int rdma_disconnect(struct rdma_cm_id *id)
3818 {
3819         struct rdma_id_private *id_priv;
3820         int ret;
3821
3822         id_priv = container_of(id, struct rdma_id_private, id);
3823         if (!id_priv->cm_id.ib)
3824                 return -EINVAL;
3825
3826         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3827                 ret = cma_modify_qp_err(id_priv);
3828                 if (ret)
3829                         goto out;
3830                 /* Initiate or respond to a disconnect. */
3831                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3832                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3833         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3834                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3835         } else
3836                 ret = -EINVAL;
3837
3838 out:
3839         return ret;
3840 }
3841 EXPORT_SYMBOL(rdma_disconnect);
3842
3843 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3844 {
3845         struct rdma_id_private *id_priv;
3846         struct cma_multicast *mc = multicast->context;
3847         struct rdma_cm_event event;
3848         int ret = 0;
3849
3850         id_priv = mc->id_priv;
3851         mutex_lock(&id_priv->handler_mutex);
3852         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3853             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3854                 goto out;
3855
3856         if (!status)
3857                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3858         else
3859                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3860                                      status);
3861         mutex_lock(&id_priv->qp_mutex);
3862         if (!status && id_priv->id.qp) {
3863                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3864                                          be16_to_cpu(multicast->rec.mlid));
3865                 if (status)
3866                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3867                                              status);
3868         }
3869         mutex_unlock(&id_priv->qp_mutex);
3870
3871         memset(&event, 0, sizeof event);
3872         event.status = status;
3873         event.param.ud.private_data = mc->context;
3874         if (!status) {
3875                 struct rdma_dev_addr *dev_addr =
3876                         &id_priv->id.route.addr.dev_addr;
3877                 struct net_device *ndev =
3878                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3879                 enum ib_gid_type gid_type =
3880                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3881                         rdma_start_port(id_priv->cma_dev->device)];
3882
3883                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3884                 ib_init_ah_from_mcmember(id_priv->id.device,
3885                                          id_priv->id.port_num, &multicast->rec,
3886                                          ndev, gid_type,
3887                                          &event.param.ud.ah_attr);
3888                 event.param.ud.qp_num = 0xFFFFFF;
3889                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3890                 if (ndev)
3891                         dev_put(ndev);
3892         } else
3893                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3894
3895         ret = id_priv->id.event_handler(&id_priv->id, &event);
3896         if (ret) {
3897                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3898                 mutex_unlock(&id_priv->handler_mutex);
3899                 rdma_destroy_id(&id_priv->id);
3900                 return 0;
3901         }
3902
3903 out:
3904         mutex_unlock(&id_priv->handler_mutex);
3905         return 0;
3906 }
3907
3908 static void cma_set_mgid(struct rdma_id_private *id_priv,
3909                          struct sockaddr *addr, union ib_gid *mgid)
3910 {
3911         unsigned char mc_map[MAX_ADDR_LEN];
3912         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3913         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3914         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3915
3916         if (cma_any_addr(addr)) {
3917                 memset(mgid, 0, sizeof *mgid);
3918         } else if ((addr->sa_family == AF_INET6) &&
3919                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3920                                                                  0xFF10A01B)) {
3921                 /* IPv6 address is an SA assigned MGID. */
3922                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3923         } else if (addr->sa_family == AF_IB) {
3924                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3925         } else if ((addr->sa_family == AF_INET6)) {
3926                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3927                 if (id_priv->id.ps == RDMA_PS_UDP)
3928                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3929                 *mgid = *(union ib_gid *) (mc_map + 4);
3930         } else {
3931                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3932                 if (id_priv->id.ps == RDMA_PS_UDP)
3933                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3934                 *mgid = *(union ib_gid *) (mc_map + 4);
3935         }
3936 }
3937
3938 static void cma_query_sa_classport_info_cb(int status,
3939                                            struct ib_class_port_info *rec,
3940                                            void *context)
3941 {
3942         struct class_port_info_context *cb_ctx = context;
3943
3944         WARN_ON(!context);
3945
3946         if (status || !rec) {
3947                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3948                          cb_ctx->device->name, cb_ctx->port_num, status);
3949                 goto out;
3950         }
3951
3952         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3953
3954 out:
3955         complete(&cb_ctx->done);
3956 }
3957
3958 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3959                                        struct ib_class_port_info *class_port_info)
3960 {
3961         struct class_port_info_context *cb_ctx;
3962         int ret;
3963
3964         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3965         if (!cb_ctx)
3966                 return -ENOMEM;
3967
3968         cb_ctx->device = device;
3969         cb_ctx->class_port_info = class_port_info;
3970         cb_ctx->port_num = port_num;
3971         init_completion(&cb_ctx->done);
3972
3973         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3974                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3975                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3976                                              cb_ctx, &cb_ctx->sa_query);
3977         if (ret < 0) {
3978                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3979                        device->name, port_num, ret);
3980                 goto out;
3981         }
3982
3983         wait_for_completion(&cb_ctx->done);
3984
3985 out:
3986         kfree(cb_ctx);
3987         return ret;
3988 }
3989
3990 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3991                                  struct cma_multicast *mc)
3992 {
3993         struct ib_sa_mcmember_rec rec;
3994         struct ib_class_port_info class_port_info;
3995         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3996         ib_sa_comp_mask comp_mask;
3997         int ret;
3998
3999         ib_addr_get_mgid(dev_addr, &rec.mgid);
4000         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4001                                      &rec.mgid, &rec);
4002         if (ret)
4003                 return ret;
4004
4005         ret = cma_set_qkey(id_priv, 0);
4006         if (ret)
4007                 return ret;
4008
4009         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4010         rec.qkey = cpu_to_be32(id_priv->qkey);
4011         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4012         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4013         rec.join_state = mc->join_state;
4014
4015         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
4016                 ret = cma_query_sa_classport_info(id_priv->id.device,
4017                                                   id_priv->id.port_num,
4018                                                   &class_port_info);
4019
4020                 if (ret)
4021                         return ret;
4022
4023                 if (!(ib_get_cpi_capmask2(&class_port_info) &
4024                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
4025                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4026                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
4027                                 id_priv->id.device->name, id_priv->id.port_num);
4028                         return -EOPNOTSUPP;
4029                 }
4030         }
4031
4032         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4033                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4034                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4035                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4036                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4037
4038         if (id_priv->id.ps == RDMA_PS_IPOIB)
4039                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4040                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4041                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4042                              IB_SA_MCMEMBER_REC_MTU |
4043                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4044
4045         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4046                                                 id_priv->id.port_num, &rec,
4047                                                 comp_mask, GFP_KERNEL,
4048                                                 cma_ib_mc_handler, mc);
4049         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4050 }
4051
4052 static void iboe_mcast_work_handler(struct work_struct *work)
4053 {
4054         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4055         struct cma_multicast *mc = mw->mc;
4056         struct ib_sa_multicast *m = mc->multicast.ib;
4057
4058         mc->multicast.ib->context = mc;
4059         cma_ib_mc_handler(0, m);
4060         kref_put(&mc->mcref, release_mc);
4061         kfree(mw);
4062 }
4063
4064 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4065 {
4066         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4067         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4068
4069         if (cma_any_addr(addr)) {
4070                 memset(mgid, 0, sizeof *mgid);
4071         } else if (addr->sa_family == AF_INET6) {
4072                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4073         } else {
4074                 mgid->raw[0] = 0xff;
4075                 mgid->raw[1] = 0x0e;
4076                 mgid->raw[2] = 0;
4077                 mgid->raw[3] = 0;
4078                 mgid->raw[4] = 0;
4079                 mgid->raw[5] = 0;
4080                 mgid->raw[6] = 0;
4081                 mgid->raw[7] = 0;
4082                 mgid->raw[8] = 0;
4083                 mgid->raw[9] = 0;
4084                 mgid->raw[10] = 0xff;
4085                 mgid->raw[11] = 0xff;
4086                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4087         }
4088 }
4089
4090 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4091                                    struct cma_multicast *mc)
4092 {
4093         struct iboe_mcast_work *work;
4094         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4095         int err = 0;
4096         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4097         struct net_device *ndev = NULL;
4098         enum ib_gid_type gid_type;
4099         bool send_only;
4100
4101         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4102
4103         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4104                 return -EINVAL;
4105
4106         work = kzalloc(sizeof *work, GFP_KERNEL);
4107         if (!work)
4108                 return -ENOMEM;
4109
4110         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4111         if (!mc->multicast.ib) {
4112                 err = -ENOMEM;
4113                 goto out1;
4114         }
4115
4116         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4117
4118         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4119         if (id_priv->id.ps == RDMA_PS_UDP)
4120                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4121
4122         if (dev_addr->bound_dev_if)
4123                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4124         if (!ndev) {
4125                 err = -ENODEV;
4126                 goto out2;
4127         }
4128         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4129         mc->multicast.ib->rec.hop_limit = 1;
4130         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4131
4132         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4133                    rdma_start_port(id_priv->cma_dev->device)];
4134         if (addr->sa_family == AF_INET) {
4135                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4136                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4137                         if (!send_only) {
4138                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4139                                                     true);
4140                                 if (!err)
4141                                         mc->igmp_joined = true;
4142                         }
4143                 }
4144         } else {
4145                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4146                         err = -ENOTSUPP;
4147         }
4148         dev_put(ndev);
4149         if (err || !mc->multicast.ib->rec.mtu) {
4150                 if (!err)
4151                         err = -EINVAL;
4152                 goto out2;
4153         }
4154         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4155                     &mc->multicast.ib->rec.port_gid);
4156         work->id = id_priv;
4157         work->mc = mc;
4158         INIT_WORK(&work->work, iboe_mcast_work_handler);
4159         kref_get(&mc->mcref);
4160         queue_work(cma_wq, &work->work);
4161
4162         return 0;
4163
4164 out2:
4165         kfree(mc->multicast.ib);
4166 out1:
4167         kfree(work);
4168         return err;
4169 }
4170
4171 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4172                         u8 join_state, void *context)
4173 {
4174         struct rdma_id_private *id_priv;
4175         struct cma_multicast *mc;
4176         int ret;
4177
4178         id_priv = container_of(id, struct rdma_id_private, id);
4179         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4180             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4181                 return -EINVAL;
4182
4183         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4184         if (!mc)
4185                 return -ENOMEM;
4186
4187         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4188         mc->context = context;
4189         mc->id_priv = id_priv;
4190         mc->igmp_joined = false;
4191         mc->join_state = join_state;
4192         spin_lock(&id_priv->lock);
4193         list_add(&mc->list, &id_priv->mc_list);
4194         spin_unlock(&id_priv->lock);
4195
4196         if (rdma_protocol_roce(id->device, id->port_num)) {
4197                 kref_init(&mc->mcref);
4198                 ret = cma_iboe_join_multicast(id_priv, mc);
4199         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4200                 ret = cma_join_ib_multicast(id_priv, mc);
4201         else
4202                 ret = -ENOSYS;
4203
4204         if (ret) {
4205                 spin_lock_irq(&id_priv->lock);
4206                 list_del(&mc->list);
4207                 spin_unlock_irq(&id_priv->lock);
4208                 kfree(mc);
4209         }
4210         return ret;
4211 }
4212 EXPORT_SYMBOL(rdma_join_multicast);
4213
4214 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4215 {
4216         struct rdma_id_private *id_priv;
4217         struct cma_multicast *mc;
4218
4219         id_priv = container_of(id, struct rdma_id_private, id);
4220         spin_lock_irq(&id_priv->lock);
4221         list_for_each_entry(mc, &id_priv->mc_list, list) {
4222                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4223                         list_del(&mc->list);
4224                         spin_unlock_irq(&id_priv->lock);
4225
4226                         if (id->qp)
4227                                 ib_detach_mcast(id->qp,
4228                                                 &mc->multicast.ib->rec.mgid,
4229                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4230
4231                         BUG_ON(id_priv->cma_dev->device != id->device);
4232
4233                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4234                                 ib_sa_free_multicast(mc->multicast.ib);
4235                                 kfree(mc);
4236                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4237                                 if (mc->igmp_joined) {
4238                                         struct rdma_dev_addr *dev_addr =
4239                                                 &id->route.addr.dev_addr;
4240                                         struct net_device *ndev = NULL;
4241
4242                                         if (dev_addr->bound_dev_if)
4243                                                 ndev = dev_get_by_index(&init_net,
4244                                                                         dev_addr->bound_dev_if);
4245                                         if (ndev) {
4246                                                 cma_igmp_send(ndev,
4247                                                               &mc->multicast.ib->rec.mgid,
4248                                                               false);
4249                                                 dev_put(ndev);
4250                                         }
4251                                         mc->igmp_joined = false;
4252                                 }
4253                                 kref_put(&mc->mcref, release_mc);
4254                         }
4255                         return;
4256                 }
4257         }
4258         spin_unlock_irq(&id_priv->lock);
4259 }
4260 EXPORT_SYMBOL(rdma_leave_multicast);
4261
4262 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4263 {
4264         struct rdma_dev_addr *dev_addr;
4265         struct cma_ndev_work *work;
4266
4267         dev_addr = &id_priv->id.route.addr.dev_addr;
4268
4269         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4270             (net_eq(dev_net(ndev), dev_addr->net)) &&
4271             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4272                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4273                         ndev->name, &id_priv->id);
4274                 work = kzalloc(sizeof *work, GFP_KERNEL);
4275                 if (!work)
4276                         return -ENOMEM;
4277
4278                 INIT_WORK(&work->work, cma_ndev_work_handler);
4279                 work->id = id_priv;
4280                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4281                 atomic_inc(&id_priv->refcount);
4282                 queue_work(cma_wq, &work->work);
4283         }
4284
4285         return 0;
4286 }
4287
4288 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4289                                void *ptr)
4290 {
4291         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4292         struct cma_device *cma_dev;
4293         struct rdma_id_private *id_priv;
4294         int ret = NOTIFY_DONE;
4295
4296         if (event != NETDEV_BONDING_FAILOVER)
4297                 return NOTIFY_DONE;
4298
4299         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4300                 return NOTIFY_DONE;
4301
4302         mutex_lock(&lock);
4303         list_for_each_entry(cma_dev, &dev_list, list)
4304                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4305                         ret = cma_netdev_change(ndev, id_priv);
4306                         if (ret)
4307                                 goto out;
4308                 }
4309
4310 out:
4311         mutex_unlock(&lock);
4312         return ret;
4313 }
4314
4315 static struct notifier_block cma_nb = {
4316         .notifier_call = cma_netdev_callback
4317 };
4318
4319 static void cma_add_one(struct ib_device *device)
4320 {
4321         struct cma_device *cma_dev;
4322         struct rdma_id_private *id_priv;
4323         unsigned int i;
4324         unsigned long supported_gids = 0;
4325
4326         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4327         if (!cma_dev)
4328                 return;
4329
4330         cma_dev->device = device;
4331         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4332                                             sizeof(*cma_dev->default_gid_type),
4333                                             GFP_KERNEL);
4334         if (!cma_dev->default_gid_type)
4335                 goto free_cma_dev;
4336
4337         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4338                                             sizeof(*cma_dev->default_roce_tos),
4339                                             GFP_KERNEL);
4340         if (!cma_dev->default_roce_tos)
4341                 goto free_gid_type;
4342
4343         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4344                 supported_gids = roce_gid_type_mask_support(device, i);
4345                 WARN_ON(!supported_gids);
4346                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4347                         find_first_bit(&supported_gids, BITS_PER_LONG);
4348                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4349         }
4350
4351         init_completion(&cma_dev->comp);
4352         atomic_set(&cma_dev->refcount, 1);
4353         INIT_LIST_HEAD(&cma_dev->id_list);
4354         ib_set_client_data(device, &cma_client, cma_dev);
4355
4356         mutex_lock(&lock);
4357         list_add_tail(&cma_dev->list, &dev_list);
4358         list_for_each_entry(id_priv, &listen_any_list, list)
4359                 cma_listen_on_dev(id_priv, cma_dev);
4360         mutex_unlock(&lock);
4361
4362         return;
4363
4364 free_gid_type:
4365         kfree(cma_dev->default_gid_type);
4366
4367 free_cma_dev:
4368         kfree(cma_dev);
4369
4370         return;
4371 }
4372
4373 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4374 {
4375         struct rdma_cm_event event;
4376         enum rdma_cm_state state;
4377         int ret = 0;
4378
4379         /* Record that we want to remove the device */
4380         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4381         if (state == RDMA_CM_DESTROYING)
4382                 return 0;
4383
4384         cma_cancel_operation(id_priv, state);
4385         mutex_lock(&id_priv->handler_mutex);
4386
4387         /* Check for destruction from another callback. */
4388         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4389                 goto out;
4390
4391         memset(&event, 0, sizeof event);
4392         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4393         ret = id_priv->id.event_handler(&id_priv->id, &event);
4394 out:
4395         mutex_unlock(&id_priv->handler_mutex);
4396         return ret;
4397 }
4398
4399 static void cma_process_remove(struct cma_device *cma_dev)
4400 {
4401         struct rdma_id_private *id_priv;
4402         int ret;
4403
4404         mutex_lock(&lock);
4405         while (!list_empty(&cma_dev->id_list)) {
4406                 id_priv = list_entry(cma_dev->id_list.next,
4407                                      struct rdma_id_private, list);
4408
4409                 list_del(&id_priv->listen_list);
4410                 list_del_init(&id_priv->list);
4411                 atomic_inc(&id_priv->refcount);
4412                 mutex_unlock(&lock);
4413
4414                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4415                 cma_deref_id(id_priv);
4416                 if (ret)
4417                         rdma_destroy_id(&id_priv->id);
4418
4419                 mutex_lock(&lock);
4420         }
4421         mutex_unlock(&lock);
4422
4423         cma_deref_dev(cma_dev);
4424         wait_for_completion(&cma_dev->comp);
4425 }
4426
4427 static void cma_remove_one(struct ib_device *device, void *client_data)
4428 {
4429         struct cma_device *cma_dev = client_data;
4430
4431         if (!cma_dev)
4432                 return;
4433
4434         mutex_lock(&lock);
4435         list_del(&cma_dev->list);
4436         mutex_unlock(&lock);
4437
4438         cma_process_remove(cma_dev);
4439         kfree(cma_dev->default_roce_tos);
4440         kfree(cma_dev->default_gid_type);
4441         kfree(cma_dev);
4442 }
4443
4444 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4445 {
4446         struct nlmsghdr *nlh;
4447         struct rdma_cm_id_stats *id_stats;
4448         struct rdma_id_private *id_priv;
4449         struct rdma_cm_id *id = NULL;
4450         struct cma_device *cma_dev;
4451         int i_dev = 0, i_id = 0;
4452
4453         /*
4454          * We export all of the IDs as a sequence of messages.  Each
4455          * ID gets its own netlink message.
4456          */
4457         mutex_lock(&lock);
4458
4459         list_for_each_entry(cma_dev, &dev_list, list) {
4460                 if (i_dev < cb->args[0]) {
4461                         i_dev++;
4462                         continue;
4463                 }
4464
4465                 i_id = 0;
4466                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4467                         if (i_id < cb->args[1]) {
4468                                 i_id++;
4469                                 continue;
4470                         }
4471
4472                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4473                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4474                                                 RDMA_NL_RDMA_CM_ID_STATS,
4475                                                 NLM_F_MULTI);
4476                         if (!id_stats)
4477                                 goto out;
4478
4479                         memset(id_stats, 0, sizeof *id_stats);
4480                         id = &id_priv->id;
4481                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4482                         id_stats->port_num = id->port_num;
4483                         id_stats->bound_dev_if =
4484                                 id->route.addr.dev_addr.bound_dev_if;
4485
4486                         if (ibnl_put_attr(skb, nlh,
4487                                           rdma_addr_size(cma_src_addr(id_priv)),
4488                                           cma_src_addr(id_priv),
4489                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4490                                 goto out;
4491                         if (ibnl_put_attr(skb, nlh,
4492                                           rdma_addr_size(cma_src_addr(id_priv)),
4493                                           cma_dst_addr(id_priv),
4494                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4495                                 goto out;
4496
4497                         id_stats->pid           = id_priv->owner;
4498                         id_stats->port_space    = id->ps;
4499                         id_stats->cm_state      = id_priv->state;
4500                         id_stats->qp_num        = id_priv->qp_num;
4501                         id_stats->qp_type       = id->qp_type;
4502
4503                         i_id++;
4504                 }
4505
4506                 cb->args[1] = 0;
4507                 i_dev++;
4508         }
4509
4510 out:
4511         mutex_unlock(&lock);
4512         cb->args[0] = i_dev;
4513         cb->args[1] = i_id;
4514
4515         return skb->len;
4516 }
4517
4518 static const struct ibnl_client_cbs cma_cb_table[] = {
4519         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4520                                        .module = THIS_MODULE },
4521 };
4522
4523 static int cma_init_net(struct net *net)
4524 {
4525         struct cma_pernet *pernet = cma_pernet(net);
4526
4527         idr_init(&pernet->tcp_ps);
4528         idr_init(&pernet->udp_ps);
4529         idr_init(&pernet->ipoib_ps);
4530         idr_init(&pernet->ib_ps);
4531
4532         return 0;
4533 }
4534
4535 static void cma_exit_net(struct net *net)
4536 {
4537         struct cma_pernet *pernet = cma_pernet(net);
4538
4539         idr_destroy(&pernet->tcp_ps);
4540         idr_destroy(&pernet->udp_ps);
4541         idr_destroy(&pernet->ipoib_ps);
4542         idr_destroy(&pernet->ib_ps);
4543 }
4544
4545 static struct pernet_operations cma_pernet_operations = {
4546         .init = cma_init_net,
4547         .exit = cma_exit_net,
4548         .id = &cma_pernet_id,
4549         .size = sizeof(struct cma_pernet),
4550 };
4551
4552 static int __init cma_init(void)
4553 {
4554         int ret;
4555
4556         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4557         if (!cma_wq)
4558                 return -ENOMEM;
4559
4560         ret = register_pernet_subsys(&cma_pernet_operations);
4561         if (ret)
4562                 goto err_wq;
4563
4564         ib_sa_register_client(&sa_client);
4565         rdma_addr_register_client(&addr_client);
4566         register_netdevice_notifier(&cma_nb);
4567
4568         ret = ib_register_client(&cma_client);
4569         if (ret)
4570                 goto err;
4571
4572         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4573                             cma_cb_table))
4574                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4575         cma_configfs_init();
4576
4577         return 0;
4578
4579 err:
4580         unregister_netdevice_notifier(&cma_nb);
4581         rdma_addr_unregister_client(&addr_client);
4582         ib_sa_unregister_client(&sa_client);
4583 err_wq:
4584         destroy_workqueue(cma_wq);
4585         return ret;
4586 }
4587
4588 static void __exit cma_cleanup(void)
4589 {
4590         cma_configfs_exit();
4591         ibnl_remove_client(RDMA_NL_RDMA_CM);
4592         ib_unregister_client(&cma_client);
4593         unregister_netdevice_notifier(&cma_nb);
4594         rdma_addr_unregister_client(&addr_client);
4595         ib_sa_unregister_client(&sa_client);
4596         unregister_pernet_subsys(&cma_pernet_operations);
4597         destroy_workqueue(cma_wq);
4598 }
4599
4600 module_init(cma_init);
4601 module_exit(cma_cleanup);