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 = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2501         if (dev->num_tc)
2502                 return netdev_get_prio_tc_map(dev, prio);
2503
2504 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2505         if (is_vlan_dev(ndev))
2506                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2507                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2508 #endif
2509         return 0;
2510 }
2511
2512 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2513                                            unsigned long supported_gids,
2514                                            enum ib_gid_type default_gid)
2515 {
2516         if ((network_type == RDMA_NETWORK_IPV4 ||
2517              network_type == RDMA_NETWORK_IPV6) &&
2518             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2519                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2520
2521         return default_gid;
2522 }
2523
2524 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2525 {
2526         struct rdma_route *route = &id_priv->id.route;
2527         struct rdma_addr *addr = &route->addr;
2528         struct cma_work *work;
2529         int ret;
2530         struct net_device *ndev = NULL;
2531         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2532                                         rdma_start_port(id_priv->cma_dev->device)];
2533         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2534
2535
2536         work = kzalloc(sizeof *work, GFP_KERNEL);
2537         if (!work)
2538                 return -ENOMEM;
2539
2540         work->id = id_priv;
2541         INIT_WORK(&work->work, cma_work_handler);
2542
2543         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2544         if (!route->path_rec) {
2545                 ret = -ENOMEM;
2546                 goto err1;
2547         }
2548
2549         route->num_paths = 1;
2550
2551         if (addr->dev_addr.bound_dev_if) {
2552                 unsigned long supported_gids;
2553
2554                 ndev = dev_get_by_index(&init_net, addr->dev_addr.bound_dev_if);
2555                 if (!ndev) {
2556                         ret = -ENODEV;
2557                         goto err2;
2558                 }
2559
2560                 if (ndev->flags & IFF_LOOPBACK) {
2561                         dev_put(ndev);
2562                         if (!id_priv->id.device->get_netdev) {
2563                                 ret = -EOPNOTSUPP;
2564                                 goto err2;
2565                         }
2566
2567                         ndev = id_priv->id.device->get_netdev(id_priv->id.device,
2568                                                               id_priv->id.port_num);
2569                         if (!ndev) {
2570                                 ret = -ENODEV;
2571                                 goto err2;
2572                         }
2573                 }
2574
2575                 route->path_rec->net = &init_net;
2576                 route->path_rec->ifindex = ndev->ifindex;
2577                 supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2578                                                             id_priv->id.port_num);
2579                 route->path_rec->gid_type =
2580                         cma_route_gid_type(addr->dev_addr.network,
2581                                            supported_gids,
2582                                            id_priv->gid_type);
2583         }
2584         if (!ndev) {
2585                 ret = -ENODEV;
2586                 goto err2;
2587         }
2588
2589         memcpy(route->path_rec->dmac, addr->dev_addr.dst_dev_addr, ETH_ALEN);
2590
2591         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2592                     &route->path_rec->sgid);
2593         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2594                     &route->path_rec->dgid);
2595
2596         /* Use the hint from IP Stack to select GID Type */
2597         if (route->path_rec->gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2598                 route->path_rec->gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2599         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2600                 /* TODO: get the hoplimit from the inet/inet6 device */
2601                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2602         else
2603                 route->path_rec->hop_limit = 1;
2604         route->path_rec->reversible = 1;
2605         route->path_rec->pkey = cpu_to_be16(0xffff);
2606         route->path_rec->mtu_selector = IB_SA_EQ;
2607         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2608         route->path_rec->traffic_class = tos;
2609         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2610         route->path_rec->rate_selector = IB_SA_EQ;
2611         route->path_rec->rate = iboe_get_rate(ndev);
2612         dev_put(ndev);
2613         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2614         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2615         if (!route->path_rec->mtu) {
2616                 ret = -EINVAL;
2617                 goto err2;
2618         }
2619
2620         work->old_state = RDMA_CM_ROUTE_QUERY;
2621         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2622         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2623         work->event.status = 0;
2624
2625         queue_work(cma_wq, &work->work);
2626
2627         return 0;
2628
2629 err2:
2630         kfree(route->path_rec);
2631         route->path_rec = NULL;
2632 err1:
2633         kfree(work);
2634         return ret;
2635 }
2636
2637 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2638 {
2639         struct rdma_id_private *id_priv;
2640         int ret;
2641
2642         id_priv = container_of(id, struct rdma_id_private, id);
2643         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2644                 return -EINVAL;
2645
2646         atomic_inc(&id_priv->refcount);
2647         if (rdma_cap_ib_sa(id->device, id->port_num))
2648                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2649         else if (rdma_protocol_roce(id->device, id->port_num))
2650                 ret = cma_resolve_iboe_route(id_priv);
2651         else if (rdma_protocol_iwarp(id->device, id->port_num))
2652                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2653         else
2654                 ret = -ENOSYS;
2655
2656         if (ret)
2657                 goto err;
2658
2659         return 0;
2660 err:
2661         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2662         cma_deref_id(id_priv);
2663         return ret;
2664 }
2665 EXPORT_SYMBOL(rdma_resolve_route);
2666
2667 static void cma_set_loopback(struct sockaddr *addr)
2668 {
2669         switch (addr->sa_family) {
2670         case AF_INET:
2671                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2672                 break;
2673         case AF_INET6:
2674                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2675                               0, 0, 0, htonl(1));
2676                 break;
2677         default:
2678                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2679                             0, 0, 0, htonl(1));
2680                 break;
2681         }
2682 }
2683
2684 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2685 {
2686         struct cma_device *cma_dev, *cur_dev;
2687         union ib_gid gid;
2688         enum ib_port_state port_state;
2689         u16 pkey;
2690         int ret;
2691         u8 p;
2692
2693         cma_dev = NULL;
2694         mutex_lock(&lock);
2695         list_for_each_entry(cur_dev, &dev_list, list) {
2696                 if (cma_family(id_priv) == AF_IB &&
2697                     !rdma_cap_ib_cm(cur_dev->device, 1))
2698                         continue;
2699
2700                 if (!cma_dev)
2701                         cma_dev = cur_dev;
2702
2703                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2704                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2705                             port_state == IB_PORT_ACTIVE) {
2706                                 cma_dev = cur_dev;
2707                                 goto port_found;
2708                         }
2709                 }
2710         }
2711
2712         if (!cma_dev) {
2713                 ret = -ENODEV;
2714                 goto out;
2715         }
2716
2717         p = 1;
2718
2719 port_found:
2720         ret = ib_get_cached_gid(cma_dev->device, p, 0, &gid, NULL);
2721         if (ret)
2722                 goto out;
2723
2724         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2725         if (ret)
2726                 goto out;
2727
2728         id_priv->id.route.addr.dev_addr.dev_type =
2729                 (rdma_protocol_ib(cma_dev->device, p)) ?
2730                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2731
2732         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2733         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2734         id_priv->id.port_num = p;
2735         cma_attach_to_dev(id_priv, cma_dev);
2736         cma_set_loopback(cma_src_addr(id_priv));
2737 out:
2738         mutex_unlock(&lock);
2739         return ret;
2740 }
2741
2742 static void addr_handler(int status, struct sockaddr *src_addr,
2743                          struct rdma_dev_addr *dev_addr, void *context)
2744 {
2745         struct rdma_id_private *id_priv = context;
2746         struct rdma_cm_event event;
2747
2748         memset(&event, 0, sizeof event);
2749         mutex_lock(&id_priv->handler_mutex);
2750         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2751                            RDMA_CM_ADDR_RESOLVED))
2752                 goto out;
2753
2754         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2755         if (!status && !id_priv->cma_dev) {
2756                 status = cma_acquire_dev(id_priv, NULL);
2757                 if (status)
2758                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2759                                              status);
2760         } else {
2761                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2762         }
2763
2764         if (status) {
2765                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2766                                    RDMA_CM_ADDR_BOUND))
2767                         goto out;
2768                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2769                 event.status = status;
2770         } else
2771                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2772
2773         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2774                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2775                 mutex_unlock(&id_priv->handler_mutex);
2776                 cma_deref_id(id_priv);
2777                 rdma_destroy_id(&id_priv->id);
2778                 return;
2779         }
2780 out:
2781         mutex_unlock(&id_priv->handler_mutex);
2782         cma_deref_id(id_priv);
2783 }
2784
2785 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2786 {
2787         struct cma_work *work;
2788         union ib_gid gid;
2789         int ret;
2790
2791         work = kzalloc(sizeof *work, GFP_KERNEL);
2792         if (!work)
2793                 return -ENOMEM;
2794
2795         if (!id_priv->cma_dev) {
2796                 ret = cma_bind_loopback(id_priv);
2797                 if (ret)
2798                         goto err;
2799         }
2800
2801         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2802         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2803
2804         work->id = id_priv;
2805         INIT_WORK(&work->work, cma_work_handler);
2806         work->old_state = RDMA_CM_ADDR_QUERY;
2807         work->new_state = RDMA_CM_ADDR_RESOLVED;
2808         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2809         queue_work(cma_wq, &work->work);
2810         return 0;
2811 err:
2812         kfree(work);
2813         return ret;
2814 }
2815
2816 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2817 {
2818         struct cma_work *work;
2819         int ret;
2820
2821         work = kzalloc(sizeof *work, GFP_KERNEL);
2822         if (!work)
2823                 return -ENOMEM;
2824
2825         if (!id_priv->cma_dev) {
2826                 ret = cma_resolve_ib_dev(id_priv);
2827                 if (ret)
2828                         goto err;
2829         }
2830
2831         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2832                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2833
2834         work->id = id_priv;
2835         INIT_WORK(&work->work, cma_work_handler);
2836         work->old_state = RDMA_CM_ADDR_QUERY;
2837         work->new_state = RDMA_CM_ADDR_RESOLVED;
2838         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2839         queue_work(cma_wq, &work->work);
2840         return 0;
2841 err:
2842         kfree(work);
2843         return ret;
2844 }
2845
2846 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2847                          struct sockaddr *dst_addr)
2848 {
2849         if (!src_addr || !src_addr->sa_family) {
2850                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2851                 src_addr->sa_family = dst_addr->sa_family;
2852                 if (IS_ENABLED(CONFIG_IPV6) &&
2853                     dst_addr->sa_family == AF_INET6) {
2854                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2855                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2856                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2857                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2858                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2859                 } else if (dst_addr->sa_family == AF_IB) {
2860                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2861                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2862                 }
2863         }
2864         return rdma_bind_addr(id, src_addr);
2865 }
2866
2867 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2868                       struct sockaddr *dst_addr, int timeout_ms)
2869 {
2870         struct rdma_id_private *id_priv;
2871         int ret;
2872
2873         id_priv = container_of(id, struct rdma_id_private, id);
2874         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2875         if (id_priv->state == RDMA_CM_IDLE) {
2876                 ret = cma_bind_addr(id, src_addr, dst_addr);
2877                 if (ret) {
2878                         memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2879                         return ret;
2880                 }
2881         }
2882
2883         if (cma_family(id_priv) != dst_addr->sa_family) {
2884                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2885                 return -EINVAL;
2886         }
2887
2888         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY)) {
2889                 memset(cma_dst_addr(id_priv), 0, rdma_addr_size(dst_addr));
2890                 return -EINVAL;
2891         }
2892
2893         atomic_inc(&id_priv->refcount);
2894         if (cma_any_addr(dst_addr)) {
2895                 ret = cma_resolve_loopback(id_priv);
2896         } else {
2897                 if (dst_addr->sa_family == AF_IB) {
2898                         ret = cma_resolve_ib_addr(id_priv);
2899                 } else {
2900                         ret = rdma_resolve_ip(&addr_client, cma_src_addr(id_priv),
2901                                               dst_addr, &id->route.addr.dev_addr,
2902                                               timeout_ms, addr_handler, id_priv);
2903                 }
2904         }
2905         if (ret)
2906                 goto err;
2907
2908         return 0;
2909 err:
2910         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
2911         cma_deref_id(id_priv);
2912         return ret;
2913 }
2914 EXPORT_SYMBOL(rdma_resolve_addr);
2915
2916 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
2917 {
2918         struct rdma_id_private *id_priv;
2919         unsigned long flags;
2920         int ret;
2921
2922         id_priv = container_of(id, struct rdma_id_private, id);
2923         spin_lock_irqsave(&id_priv->lock, flags);
2924         if (reuse || id_priv->state == RDMA_CM_IDLE) {
2925                 id_priv->reuseaddr = reuse;
2926                 ret = 0;
2927         } else {
2928                 ret = -EINVAL;
2929         }
2930         spin_unlock_irqrestore(&id_priv->lock, flags);
2931         return ret;
2932 }
2933 EXPORT_SYMBOL(rdma_set_reuseaddr);
2934
2935 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
2936 {
2937         struct rdma_id_private *id_priv;
2938         unsigned long flags;
2939         int ret;
2940
2941         id_priv = container_of(id, struct rdma_id_private, id);
2942         spin_lock_irqsave(&id_priv->lock, flags);
2943         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
2944                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
2945                 id_priv->afonly = afonly;
2946                 ret = 0;
2947         } else {
2948                 ret = -EINVAL;
2949         }
2950         spin_unlock_irqrestore(&id_priv->lock, flags);
2951         return ret;
2952 }
2953 EXPORT_SYMBOL(rdma_set_afonly);
2954
2955 static void cma_bind_port(struct rdma_bind_list *bind_list,
2956                           struct rdma_id_private *id_priv)
2957 {
2958         struct sockaddr *addr;
2959         struct sockaddr_ib *sib;
2960         u64 sid, mask;
2961         __be16 port;
2962
2963         addr = cma_src_addr(id_priv);
2964         port = htons(bind_list->port);
2965
2966         switch (addr->sa_family) {
2967         case AF_INET:
2968                 ((struct sockaddr_in *) addr)->sin_port = port;
2969                 break;
2970         case AF_INET6:
2971                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
2972                 break;
2973         case AF_IB:
2974                 sib = (struct sockaddr_ib *) addr;
2975                 sid = be64_to_cpu(sib->sib_sid);
2976                 mask = be64_to_cpu(sib->sib_sid_mask);
2977                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
2978                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
2979                 break;
2980         }
2981         id_priv->bind_list = bind_list;
2982         hlist_add_head(&id_priv->node, &bind_list->owners);
2983 }
2984
2985 static int cma_alloc_port(enum rdma_port_space ps,
2986                           struct rdma_id_private *id_priv, unsigned short snum)
2987 {
2988         struct rdma_bind_list *bind_list;
2989         int ret;
2990
2991         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
2992         if (!bind_list)
2993                 return -ENOMEM;
2994
2995         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
2996                            snum);
2997         if (ret < 0)
2998                 goto err;
2999
3000         bind_list->ps = ps;
3001         bind_list->port = (unsigned short)ret;
3002         cma_bind_port(bind_list, id_priv);
3003         return 0;
3004 err:
3005         kfree(bind_list);
3006         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3007 }
3008
3009 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3010                               struct rdma_id_private *id_priv)
3011 {
3012         struct rdma_id_private *cur_id;
3013         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3014         struct sockaddr  *saddr = cma_src_addr(id_priv);
3015         __be16 dport = cma_port(daddr);
3016
3017         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3018                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3019                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3020                 __be16 cur_dport = cma_port(cur_daddr);
3021
3022                 if (id_priv == cur_id)
3023                         continue;
3024
3025                 /* different dest port -> unique */
3026                 if (!cma_any_port(cur_daddr) &&
3027                     (dport != cur_dport))
3028                         continue;
3029
3030                 /* different src address -> unique */
3031                 if (!cma_any_addr(saddr) &&
3032                     !cma_any_addr(cur_saddr) &&
3033                     cma_addr_cmp(saddr, cur_saddr))
3034                         continue;
3035
3036                 /* different dst address -> unique */
3037                 if (!cma_any_addr(cur_daddr) &&
3038                     cma_addr_cmp(daddr, cur_daddr))
3039                         continue;
3040
3041                 return -EADDRNOTAVAIL;
3042         }
3043         return 0;
3044 }
3045
3046 static int cma_alloc_any_port(enum rdma_port_space ps,
3047                               struct rdma_id_private *id_priv)
3048 {
3049         static unsigned int last_used_port;
3050         int low, high, remaining;
3051         unsigned int rover;
3052         struct net *net = id_priv->id.route.addr.dev_addr.net;
3053
3054         inet_get_local_port_range(net, &low, &high);
3055         remaining = (high - low) + 1;
3056         rover = prandom_u32() % remaining + low;
3057 retry:
3058         if (last_used_port != rover) {
3059                 struct rdma_bind_list *bind_list;
3060                 int ret;
3061
3062                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3063
3064                 if (!bind_list) {
3065                         ret = cma_alloc_port(ps, id_priv, rover);
3066                 } else {
3067                         ret = cma_port_is_unique(bind_list, id_priv);
3068                         if (!ret)
3069                                 cma_bind_port(bind_list, id_priv);
3070                 }
3071                 /*
3072                  * Remember previously used port number in order to avoid
3073                  * re-using same port immediately after it is closed.
3074                  */
3075                 if (!ret)
3076                         last_used_port = rover;
3077                 if (ret != -EADDRNOTAVAIL)
3078                         return ret;
3079         }
3080         if (--remaining) {
3081                 rover++;
3082                 if ((rover < low) || (rover > high))
3083                         rover = low;
3084                 goto retry;
3085         }
3086         return -EADDRNOTAVAIL;
3087 }
3088
3089 /*
3090  * Check that the requested port is available.  This is called when trying to
3091  * bind to a specific port, or when trying to listen on a bound port.  In
3092  * the latter case, the provided id_priv may already be on the bind_list, but
3093  * we still need to check that it's okay to start listening.
3094  */
3095 static int cma_check_port(struct rdma_bind_list *bind_list,
3096                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3097 {
3098         struct rdma_id_private *cur_id;
3099         struct sockaddr *addr, *cur_addr;
3100
3101         addr = cma_src_addr(id_priv);
3102         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3103                 if (id_priv == cur_id)
3104                         continue;
3105
3106                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3107                     cur_id->reuseaddr)
3108                         continue;
3109
3110                 cur_addr = cma_src_addr(cur_id);
3111                 if (id_priv->afonly && cur_id->afonly &&
3112                     (addr->sa_family != cur_addr->sa_family))
3113                         continue;
3114
3115                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3116                         return -EADDRNOTAVAIL;
3117
3118                 if (!cma_addr_cmp(addr, cur_addr))
3119                         return -EADDRINUSE;
3120         }
3121         return 0;
3122 }
3123
3124 static int cma_use_port(enum rdma_port_space ps,
3125                         struct rdma_id_private *id_priv)
3126 {
3127         struct rdma_bind_list *bind_list;
3128         unsigned short snum;
3129         int ret;
3130
3131         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3132         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3133                 return -EACCES;
3134
3135         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3136         if (!bind_list) {
3137                 ret = cma_alloc_port(ps, id_priv, snum);
3138         } else {
3139                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3140                 if (!ret)
3141                         cma_bind_port(bind_list, id_priv);
3142         }
3143         return ret;
3144 }
3145
3146 static int cma_bind_listen(struct rdma_id_private *id_priv)
3147 {
3148         struct rdma_bind_list *bind_list = id_priv->bind_list;
3149         int ret = 0;
3150
3151         mutex_lock(&lock);
3152         if (bind_list->owners.first->next)
3153                 ret = cma_check_port(bind_list, id_priv, 0);
3154         mutex_unlock(&lock);
3155         return ret;
3156 }
3157
3158 static enum rdma_port_space cma_select_inet_ps(
3159                 struct rdma_id_private *id_priv)
3160 {
3161         switch (id_priv->id.ps) {
3162         case RDMA_PS_TCP:
3163         case RDMA_PS_UDP:
3164         case RDMA_PS_IPOIB:
3165         case RDMA_PS_IB:
3166                 return id_priv->id.ps;
3167         default:
3168
3169                 return 0;
3170         }
3171 }
3172
3173 static enum rdma_port_space cma_select_ib_ps(struct rdma_id_private *id_priv)
3174 {
3175         enum rdma_port_space ps = 0;
3176         struct sockaddr_ib *sib;
3177         u64 sid_ps, mask, sid;
3178
3179         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3180         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3181         sid = be64_to_cpu(sib->sib_sid) & mask;
3182
3183         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3184                 sid_ps = RDMA_IB_IP_PS_IB;
3185                 ps = RDMA_PS_IB;
3186         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3187                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3188                 sid_ps = RDMA_IB_IP_PS_TCP;
3189                 ps = RDMA_PS_TCP;
3190         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3191                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3192                 sid_ps = RDMA_IB_IP_PS_UDP;
3193                 ps = RDMA_PS_UDP;
3194         }
3195
3196         if (ps) {
3197                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3198                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3199                                                 be64_to_cpu(sib->sib_sid_mask));
3200         }
3201         return ps;
3202 }
3203
3204 static int cma_get_port(struct rdma_id_private *id_priv)
3205 {
3206         enum rdma_port_space ps;
3207         int ret;
3208
3209         if (cma_family(id_priv) != AF_IB)
3210                 ps = cma_select_inet_ps(id_priv);
3211         else
3212                 ps = cma_select_ib_ps(id_priv);
3213         if (!ps)
3214                 return -EPROTONOSUPPORT;
3215
3216         mutex_lock(&lock);
3217         if (cma_any_port(cma_src_addr(id_priv)))
3218                 ret = cma_alloc_any_port(ps, id_priv);
3219         else
3220                 ret = cma_use_port(ps, id_priv);
3221         mutex_unlock(&lock);
3222
3223         return ret;
3224 }
3225
3226 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3227                                struct sockaddr *addr)
3228 {
3229 #if IS_ENABLED(CONFIG_IPV6)
3230         struct sockaddr_in6 *sin6;
3231
3232         if (addr->sa_family != AF_INET6)
3233                 return 0;
3234
3235         sin6 = (struct sockaddr_in6 *) addr;
3236
3237         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3238                 return 0;
3239
3240         if (!sin6->sin6_scope_id)
3241                         return -EINVAL;
3242
3243         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3244 #endif
3245         return 0;
3246 }
3247
3248 int rdma_listen(struct rdma_cm_id *id, int backlog)
3249 {
3250         struct rdma_id_private *id_priv;
3251         int ret;
3252
3253         id_priv = container_of(id, struct rdma_id_private, id);
3254         if (id_priv->state == RDMA_CM_IDLE) {
3255                 id->route.addr.src_addr.ss_family = AF_INET;
3256                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3257                 if (ret)
3258                         return ret;
3259         }
3260
3261         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3262                 return -EINVAL;
3263
3264         if (id_priv->reuseaddr) {
3265                 ret = cma_bind_listen(id_priv);
3266                 if (ret)
3267                         goto err;
3268         }
3269
3270         id_priv->backlog = backlog;
3271         if (id->device) {
3272                 if (rdma_cap_ib_cm(id->device, 1)) {
3273                         ret = cma_ib_listen(id_priv);
3274                         if (ret)
3275                                 goto err;
3276                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3277                         ret = cma_iw_listen(id_priv, backlog);
3278                         if (ret)
3279                                 goto err;
3280                 } else {
3281                         ret = -ENOSYS;
3282                         goto err;
3283                 }
3284         } else
3285                 cma_listen_on_all(id_priv);
3286
3287         return 0;
3288 err:
3289         id_priv->backlog = 0;
3290         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3291         return ret;
3292 }
3293 EXPORT_SYMBOL(rdma_listen);
3294
3295 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3296 {
3297         struct rdma_id_private *id_priv;
3298         int ret;
3299         struct sockaddr  *daddr;
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         daddr = cma_dst_addr(id_priv);
3340         daddr->sa_family = addr->sa_family;
3341
3342         return 0;
3343 err2:
3344         if (id_priv->cma_dev)
3345                 cma_release_dev(id_priv);
3346 err1:
3347         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3348         return ret;
3349 }
3350 EXPORT_SYMBOL(rdma_bind_addr);
3351
3352 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3353 {
3354         struct cma_hdr *cma_hdr;
3355
3356         cma_hdr = hdr;
3357         cma_hdr->cma_version = CMA_VERSION;
3358         if (cma_family(id_priv) == AF_INET) {
3359                 struct sockaddr_in *src4, *dst4;
3360
3361                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3362                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3363
3364                 cma_set_ip_ver(cma_hdr, 4);
3365                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3366                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3367                 cma_hdr->port = src4->sin_port;
3368         } else if (cma_family(id_priv) == AF_INET6) {
3369                 struct sockaddr_in6 *src6, *dst6;
3370
3371                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3372                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3373
3374                 cma_set_ip_ver(cma_hdr, 6);
3375                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3376                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3377                 cma_hdr->port = src6->sin6_port;
3378         }
3379         return 0;
3380 }
3381
3382 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3383                                 struct ib_cm_event *ib_event)
3384 {
3385         struct rdma_id_private *id_priv = cm_id->context;
3386         struct rdma_cm_event event;
3387         struct ib_cm_sidr_rep_event_param *rep = &ib_event->param.sidr_rep_rcvd;
3388         int ret = 0;
3389
3390         mutex_lock(&id_priv->handler_mutex);
3391         if (id_priv->state != RDMA_CM_CONNECT)
3392                 goto out;
3393
3394         memset(&event, 0, sizeof event);
3395         switch (ib_event->event) {
3396         case IB_CM_SIDR_REQ_ERROR:
3397                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3398                 event.status = -ETIMEDOUT;
3399                 break;
3400         case IB_CM_SIDR_REP_RECEIVED:
3401                 event.param.ud.private_data = ib_event->private_data;
3402                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3403                 if (rep->status != IB_SIDR_SUCCESS) {
3404                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3405                         event.status = ib_event->param.sidr_rep_rcvd.status;
3406                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3407                                              event.status);
3408                         break;
3409                 }
3410                 ret = cma_set_qkey(id_priv, rep->qkey);
3411                 if (ret) {
3412                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3413                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3414                         event.status = ret;
3415                         break;
3416                 }
3417                 ib_init_ah_from_path(id_priv->id.device, id_priv->id.port_num,
3418                                      id_priv->id.route.path_rec,
3419                                      &event.param.ud.ah_attr);
3420                 event.param.ud.qp_num = rep->qpn;
3421                 event.param.ud.qkey = rep->qkey;
3422                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3423                 event.status = 0;
3424                 break;
3425         default:
3426                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3427                        ib_event->event);
3428                 goto out;
3429         }
3430
3431         ret = id_priv->id.event_handler(&id_priv->id, &event);
3432         if (ret) {
3433                 /* Destroy the CM ID by returning a non-zero value. */
3434                 id_priv->cm_id.ib = NULL;
3435                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3436                 mutex_unlock(&id_priv->handler_mutex);
3437                 rdma_destroy_id(&id_priv->id);
3438                 return ret;
3439         }
3440 out:
3441         mutex_unlock(&id_priv->handler_mutex);
3442         return ret;
3443 }
3444
3445 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3446                               struct rdma_conn_param *conn_param)
3447 {
3448         struct ib_cm_sidr_req_param req;
3449         struct ib_cm_id *id;
3450         void *private_data;
3451         int offset, ret;
3452
3453         memset(&req, 0, sizeof req);
3454         offset = cma_user_data_offset(id_priv);
3455         req.private_data_len = offset + conn_param->private_data_len;
3456         if (req.private_data_len < conn_param->private_data_len)
3457                 return -EINVAL;
3458
3459         if (req.private_data_len) {
3460                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3461                 if (!private_data)
3462                         return -ENOMEM;
3463         } else {
3464                 private_data = NULL;
3465         }
3466
3467         if (conn_param->private_data && conn_param->private_data_len)
3468                 memcpy(private_data + offset, conn_param->private_data,
3469                        conn_param->private_data_len);
3470
3471         if (private_data) {
3472                 ret = cma_format_hdr(private_data, id_priv);
3473                 if (ret)
3474                         goto out;
3475                 req.private_data = private_data;
3476         }
3477
3478         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3479                              id_priv);
3480         if (IS_ERR(id)) {
3481                 ret = PTR_ERR(id);
3482                 goto out;
3483         }
3484         id_priv->cm_id.ib = id;
3485
3486         req.path = id_priv->id.route.path_rec;
3487         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3488         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3489         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3490
3491         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3492         if (ret) {
3493                 ib_destroy_cm_id(id_priv->cm_id.ib);
3494                 id_priv->cm_id.ib = NULL;
3495         }
3496 out:
3497         kfree(private_data);
3498         return ret;
3499 }
3500
3501 static int cma_connect_ib(struct rdma_id_private *id_priv,
3502                           struct rdma_conn_param *conn_param)
3503 {
3504         struct ib_cm_req_param req;
3505         struct rdma_route *route;
3506         void *private_data;
3507         struct ib_cm_id *id;
3508         int offset, ret;
3509
3510         memset(&req, 0, sizeof req);
3511         offset = cma_user_data_offset(id_priv);
3512         req.private_data_len = offset + conn_param->private_data_len;
3513         if (req.private_data_len < conn_param->private_data_len)
3514                 return -EINVAL;
3515
3516         if (req.private_data_len) {
3517                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3518                 if (!private_data)
3519                         return -ENOMEM;
3520         } else {
3521                 private_data = NULL;
3522         }
3523
3524         if (conn_param->private_data && conn_param->private_data_len)
3525                 memcpy(private_data + offset, conn_param->private_data,
3526                        conn_param->private_data_len);
3527
3528         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3529         if (IS_ERR(id)) {
3530                 ret = PTR_ERR(id);
3531                 goto out;
3532         }
3533         id_priv->cm_id.ib = id;
3534
3535         route = &id_priv->id.route;
3536         if (private_data) {
3537                 ret = cma_format_hdr(private_data, id_priv);
3538                 if (ret)
3539                         goto out;
3540                 req.private_data = private_data;
3541         }
3542
3543         req.primary_path = &route->path_rec[0];
3544         if (route->num_paths == 2)
3545                 req.alternate_path = &route->path_rec[1];
3546
3547         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3548         req.qp_num = id_priv->qp_num;
3549         req.qp_type = id_priv->id.qp_type;
3550         req.starting_psn = id_priv->seq_num;
3551         req.responder_resources = conn_param->responder_resources;
3552         req.initiator_depth = conn_param->initiator_depth;
3553         req.flow_control = conn_param->flow_control;
3554         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3555         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3556         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3557         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3558         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3559         req.srq = id_priv->srq ? 1 : 0;
3560
3561         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3562 out:
3563         if (ret && !IS_ERR(id)) {
3564                 ib_destroy_cm_id(id);
3565                 id_priv->cm_id.ib = NULL;
3566         }
3567
3568         kfree(private_data);
3569         return ret;
3570 }
3571
3572 static int cma_connect_iw(struct rdma_id_private *id_priv,
3573                           struct rdma_conn_param *conn_param)
3574 {
3575         struct iw_cm_id *cm_id;
3576         int ret;
3577         struct iw_cm_conn_param iw_param;
3578
3579         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3580         if (IS_ERR(cm_id))
3581                 return PTR_ERR(cm_id);
3582
3583         cm_id->tos = id_priv->tos;
3584         id_priv->cm_id.iw = cm_id;
3585
3586         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3587                rdma_addr_size(cma_src_addr(id_priv)));
3588         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3589                rdma_addr_size(cma_dst_addr(id_priv)));
3590
3591         ret = cma_modify_qp_rtr(id_priv, conn_param);
3592         if (ret)
3593                 goto out;
3594
3595         if (conn_param) {
3596                 iw_param.ord = conn_param->initiator_depth;
3597                 iw_param.ird = conn_param->responder_resources;
3598                 iw_param.private_data = conn_param->private_data;
3599                 iw_param.private_data_len = conn_param->private_data_len;
3600                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3601         } else {
3602                 memset(&iw_param, 0, sizeof iw_param);
3603                 iw_param.qpn = id_priv->qp_num;
3604         }
3605         ret = iw_cm_connect(cm_id, &iw_param);
3606 out:
3607         if (ret) {
3608                 iw_destroy_cm_id(cm_id);
3609                 id_priv->cm_id.iw = NULL;
3610         }
3611         return ret;
3612 }
3613
3614 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3615 {
3616         struct rdma_id_private *id_priv;
3617         int ret;
3618
3619         id_priv = container_of(id, struct rdma_id_private, id);
3620         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3621                 return -EINVAL;
3622
3623         if (!id->qp) {
3624                 id_priv->qp_num = conn_param->qp_num;
3625                 id_priv->srq = conn_param->srq;
3626         }
3627
3628         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3629                 if (id->qp_type == IB_QPT_UD)
3630                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3631                 else
3632                         ret = cma_connect_ib(id_priv, conn_param);
3633         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3634                 ret = cma_connect_iw(id_priv, conn_param);
3635         else
3636                 ret = -ENOSYS;
3637         if (ret)
3638                 goto err;
3639
3640         return 0;
3641 err:
3642         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3643         return ret;
3644 }
3645 EXPORT_SYMBOL(rdma_connect);
3646
3647 static int cma_accept_ib(struct rdma_id_private *id_priv,
3648                          struct rdma_conn_param *conn_param)
3649 {
3650         struct ib_cm_rep_param rep;
3651         int ret;
3652
3653         ret = cma_modify_qp_rtr(id_priv, conn_param);
3654         if (ret)
3655                 goto out;
3656
3657         ret = cma_modify_qp_rts(id_priv, conn_param);
3658         if (ret)
3659                 goto out;
3660
3661         memset(&rep, 0, sizeof rep);
3662         rep.qp_num = id_priv->qp_num;
3663         rep.starting_psn = id_priv->seq_num;
3664         rep.private_data = conn_param->private_data;
3665         rep.private_data_len = conn_param->private_data_len;
3666         rep.responder_resources = conn_param->responder_resources;
3667         rep.initiator_depth = conn_param->initiator_depth;
3668         rep.failover_accepted = 0;
3669         rep.flow_control = conn_param->flow_control;
3670         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3671         rep.srq = id_priv->srq ? 1 : 0;
3672
3673         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3674 out:
3675         return ret;
3676 }
3677
3678 static int cma_accept_iw(struct rdma_id_private *id_priv,
3679                   struct rdma_conn_param *conn_param)
3680 {
3681         struct iw_cm_conn_param iw_param;
3682         int ret;
3683
3684         if (!conn_param)
3685                 return -EINVAL;
3686
3687         ret = cma_modify_qp_rtr(id_priv, conn_param);
3688         if (ret)
3689                 return ret;
3690
3691         iw_param.ord = conn_param->initiator_depth;
3692         iw_param.ird = conn_param->responder_resources;
3693         iw_param.private_data = conn_param->private_data;
3694         iw_param.private_data_len = conn_param->private_data_len;
3695         if (id_priv->id.qp) {
3696                 iw_param.qpn = id_priv->qp_num;
3697         } else
3698                 iw_param.qpn = conn_param->qp_num;
3699
3700         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3701 }
3702
3703 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3704                              enum ib_cm_sidr_status status, u32 qkey,
3705                              const void *private_data, int private_data_len)
3706 {
3707         struct ib_cm_sidr_rep_param rep;
3708         int ret;
3709
3710         memset(&rep, 0, sizeof rep);
3711         rep.status = status;
3712         if (status == IB_SIDR_SUCCESS) {
3713                 ret = cma_set_qkey(id_priv, qkey);
3714                 if (ret)
3715                         return ret;
3716                 rep.qp_num = id_priv->qp_num;
3717                 rep.qkey = id_priv->qkey;
3718         }
3719         rep.private_data = private_data;
3720         rep.private_data_len = private_data_len;
3721
3722         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3723 }
3724
3725 int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3726 {
3727         struct rdma_id_private *id_priv;
3728         int ret;
3729
3730         id_priv = container_of(id, struct rdma_id_private, id);
3731
3732         id_priv->owner = task_pid_nr(current);
3733
3734         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3735                 return -EINVAL;
3736
3737         if (!id->qp && conn_param) {
3738                 id_priv->qp_num = conn_param->qp_num;
3739                 id_priv->srq = conn_param->srq;
3740         }
3741
3742         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3743                 if (id->qp_type == IB_QPT_UD) {
3744                         if (conn_param)
3745                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3746                                                         conn_param->qkey,
3747                                                         conn_param->private_data,
3748                                                         conn_param->private_data_len);
3749                         else
3750                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3751                                                         0, NULL, 0);
3752                 } else {
3753                         if (conn_param)
3754                                 ret = cma_accept_ib(id_priv, conn_param);
3755                         else
3756                                 ret = cma_rep_recv(id_priv);
3757                 }
3758         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3759                 ret = cma_accept_iw(id_priv, conn_param);
3760         else
3761                 ret = -ENOSYS;
3762
3763         if (ret)
3764                 goto reject;
3765
3766         return 0;
3767 reject:
3768         cma_modify_qp_err(id_priv);
3769         rdma_reject(id, NULL, 0);
3770         return ret;
3771 }
3772 EXPORT_SYMBOL(rdma_accept);
3773
3774 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3775 {
3776         struct rdma_id_private *id_priv;
3777         int ret;
3778
3779         id_priv = container_of(id, struct rdma_id_private, id);
3780         if (!id_priv->cm_id.ib)
3781                 return -EINVAL;
3782
3783         switch (id->device->node_type) {
3784         case RDMA_NODE_IB_CA:
3785                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3786                 break;
3787         default:
3788                 ret = 0;
3789                 break;
3790         }
3791         return ret;
3792 }
3793 EXPORT_SYMBOL(rdma_notify);
3794
3795 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3796                 u8 private_data_len)
3797 {
3798         struct rdma_id_private *id_priv;
3799         int ret;
3800
3801         id_priv = container_of(id, struct rdma_id_private, id);
3802         if (!id_priv->cm_id.ib)
3803                 return -EINVAL;
3804
3805         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3806                 if (id->qp_type == IB_QPT_UD)
3807                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3808                                                 private_data, private_data_len);
3809                 else
3810                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3811                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3812                                              0, private_data, private_data_len);
3813         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3814                 ret = iw_cm_reject(id_priv->cm_id.iw,
3815                                    private_data, private_data_len);
3816         } else
3817                 ret = -ENOSYS;
3818
3819         return ret;
3820 }
3821 EXPORT_SYMBOL(rdma_reject);
3822
3823 int rdma_disconnect(struct rdma_cm_id *id)
3824 {
3825         struct rdma_id_private *id_priv;
3826         int ret;
3827
3828         id_priv = container_of(id, struct rdma_id_private, id);
3829         if (!id_priv->cm_id.ib)
3830                 return -EINVAL;
3831
3832         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3833                 ret = cma_modify_qp_err(id_priv);
3834                 if (ret)
3835                         goto out;
3836                 /* Initiate or respond to a disconnect. */
3837                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3838                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3839         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3840                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3841         } else
3842                 ret = -EINVAL;
3843
3844 out:
3845         return ret;
3846 }
3847 EXPORT_SYMBOL(rdma_disconnect);
3848
3849 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3850 {
3851         struct rdma_id_private *id_priv;
3852         struct cma_multicast *mc = multicast->context;
3853         struct rdma_cm_event event;
3854         int ret = 0;
3855
3856         id_priv = mc->id_priv;
3857         mutex_lock(&id_priv->handler_mutex);
3858         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3859             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3860                 goto out;
3861
3862         if (!status)
3863                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3864         else
3865                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3866                                      status);
3867         mutex_lock(&id_priv->qp_mutex);
3868         if (!status && id_priv->id.qp) {
3869                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3870                                          be16_to_cpu(multicast->rec.mlid));
3871                 if (status)
3872                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3873                                              status);
3874         }
3875         mutex_unlock(&id_priv->qp_mutex);
3876
3877         memset(&event, 0, sizeof event);
3878         event.status = status;
3879         event.param.ud.private_data = mc->context;
3880         if (!status) {
3881                 struct rdma_dev_addr *dev_addr =
3882                         &id_priv->id.route.addr.dev_addr;
3883                 struct net_device *ndev =
3884                         dev_get_by_index(&init_net, dev_addr->bound_dev_if);
3885                 enum ib_gid_type gid_type =
3886                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3887                         rdma_start_port(id_priv->cma_dev->device)];
3888
3889                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
3890                 ib_init_ah_from_mcmember(id_priv->id.device,
3891                                          id_priv->id.port_num, &multicast->rec,
3892                                          ndev, gid_type,
3893                                          &event.param.ud.ah_attr);
3894                 event.param.ud.qp_num = 0xFFFFFF;
3895                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
3896                 if (ndev)
3897                         dev_put(ndev);
3898         } else
3899                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
3900
3901         ret = id_priv->id.event_handler(&id_priv->id, &event);
3902         if (ret) {
3903                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3904                 mutex_unlock(&id_priv->handler_mutex);
3905                 rdma_destroy_id(&id_priv->id);
3906                 return 0;
3907         }
3908
3909 out:
3910         mutex_unlock(&id_priv->handler_mutex);
3911         return 0;
3912 }
3913
3914 static void cma_set_mgid(struct rdma_id_private *id_priv,
3915                          struct sockaddr *addr, union ib_gid *mgid)
3916 {
3917         unsigned char mc_map[MAX_ADDR_LEN];
3918         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
3919         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
3920         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
3921
3922         if (cma_any_addr(addr)) {
3923                 memset(mgid, 0, sizeof *mgid);
3924         } else if ((addr->sa_family == AF_INET6) &&
3925                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
3926                                                                  0xFF10A01B)) {
3927                 /* IPv6 address is an SA assigned MGID. */
3928                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
3929         } else if (addr->sa_family == AF_IB) {
3930                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
3931         } else if ((addr->sa_family == AF_INET6)) {
3932                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
3933                 if (id_priv->id.ps == RDMA_PS_UDP)
3934                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3935                 *mgid = *(union ib_gid *) (mc_map + 4);
3936         } else {
3937                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
3938                 if (id_priv->id.ps == RDMA_PS_UDP)
3939                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
3940                 *mgid = *(union ib_gid *) (mc_map + 4);
3941         }
3942 }
3943
3944 static void cma_query_sa_classport_info_cb(int status,
3945                                            struct ib_class_port_info *rec,
3946                                            void *context)
3947 {
3948         struct class_port_info_context *cb_ctx = context;
3949
3950         WARN_ON(!context);
3951
3952         if (status || !rec) {
3953                 pr_debug("RDMA CM: %s port %u failed query ClassPortInfo status: %d\n",
3954                          cb_ctx->device->name, cb_ctx->port_num, status);
3955                 goto out;
3956         }
3957
3958         memcpy(cb_ctx->class_port_info, rec, sizeof(struct ib_class_port_info));
3959
3960 out:
3961         complete(&cb_ctx->done);
3962 }
3963
3964 static int cma_query_sa_classport_info(struct ib_device *device, u8 port_num,
3965                                        struct ib_class_port_info *class_port_info)
3966 {
3967         struct class_port_info_context *cb_ctx;
3968         int ret;
3969
3970         cb_ctx = kmalloc(sizeof(*cb_ctx), GFP_KERNEL);
3971         if (!cb_ctx)
3972                 return -ENOMEM;
3973
3974         cb_ctx->device = device;
3975         cb_ctx->class_port_info = class_port_info;
3976         cb_ctx->port_num = port_num;
3977         init_completion(&cb_ctx->done);
3978
3979         ret = ib_sa_classport_info_rec_query(&sa_client, device, port_num,
3980                                              CMA_QUERY_CLASSPORT_INFO_TIMEOUT,
3981                                              GFP_KERNEL, cma_query_sa_classport_info_cb,
3982                                              cb_ctx, &cb_ctx->sa_query);
3983         if (ret < 0) {
3984                 pr_err("RDMA CM: %s port %u failed to send ClassPortInfo query, ret: %d\n",
3985                        device->name, port_num, ret);
3986                 goto out;
3987         }
3988
3989         wait_for_completion(&cb_ctx->done);
3990
3991 out:
3992         kfree(cb_ctx);
3993         return ret;
3994 }
3995
3996 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
3997                                  struct cma_multicast *mc)
3998 {
3999         struct ib_sa_mcmember_rec rec;
4000         struct ib_class_port_info class_port_info;
4001         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4002         ib_sa_comp_mask comp_mask;
4003         int ret;
4004
4005         ib_addr_get_mgid(dev_addr, &rec.mgid);
4006         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4007                                      &rec.mgid, &rec);
4008         if (ret)
4009                 return ret;
4010
4011         ret = cma_set_qkey(id_priv, 0);
4012         if (ret)
4013                 return ret;
4014
4015         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4016         rec.qkey = cpu_to_be32(id_priv->qkey);
4017         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4018         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4019         rec.join_state = mc->join_state;
4020
4021         if (rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) {
4022                 ret = cma_query_sa_classport_info(id_priv->id.device,
4023                                                   id_priv->id.port_num,
4024                                                   &class_port_info);
4025
4026                 if (ret)
4027                         return ret;
4028
4029                 if (!(ib_get_cpi_capmask2(&class_port_info) &
4030                       IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT)) {
4031                         pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4032                                 "RDMA CM: SM doesn't support Send Only Full Member option\n",
4033                                 id_priv->id.device->name, id_priv->id.port_num);
4034                         return -EOPNOTSUPP;
4035                 }
4036         }
4037
4038         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4039                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4040                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4041                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4042                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4043
4044         if (id_priv->id.ps == RDMA_PS_IPOIB)
4045                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4046                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4047                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4048                              IB_SA_MCMEMBER_REC_MTU |
4049                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4050
4051         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4052                                                 id_priv->id.port_num, &rec,
4053                                                 comp_mask, GFP_KERNEL,
4054                                                 cma_ib_mc_handler, mc);
4055         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4056 }
4057
4058 static void iboe_mcast_work_handler(struct work_struct *work)
4059 {
4060         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4061         struct cma_multicast *mc = mw->mc;
4062         struct ib_sa_multicast *m = mc->multicast.ib;
4063
4064         mc->multicast.ib->context = mc;
4065         cma_ib_mc_handler(0, m);
4066         kref_put(&mc->mcref, release_mc);
4067         kfree(mw);
4068 }
4069
4070 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid)
4071 {
4072         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4073         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4074
4075         if (cma_any_addr(addr)) {
4076                 memset(mgid, 0, sizeof *mgid);
4077         } else if (addr->sa_family == AF_INET6) {
4078                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4079         } else {
4080                 mgid->raw[0] = 0xff;
4081                 mgid->raw[1] = 0x0e;
4082                 mgid->raw[2] = 0;
4083                 mgid->raw[3] = 0;
4084                 mgid->raw[4] = 0;
4085                 mgid->raw[5] = 0;
4086                 mgid->raw[6] = 0;
4087                 mgid->raw[7] = 0;
4088                 mgid->raw[8] = 0;
4089                 mgid->raw[9] = 0;
4090                 mgid->raw[10] = 0xff;
4091                 mgid->raw[11] = 0xff;
4092                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4093         }
4094 }
4095
4096 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4097                                    struct cma_multicast *mc)
4098 {
4099         struct iboe_mcast_work *work;
4100         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4101         int err = 0;
4102         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4103         struct net_device *ndev = NULL;
4104         enum ib_gid_type gid_type;
4105         bool send_only;
4106
4107         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4108
4109         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4110                 return -EINVAL;
4111
4112         work = kzalloc(sizeof *work, GFP_KERNEL);
4113         if (!work)
4114                 return -ENOMEM;
4115
4116         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4117         if (!mc->multicast.ib) {
4118                 err = -ENOMEM;
4119                 goto out1;
4120         }
4121
4122         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid);
4123
4124         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4125         if (id_priv->id.ps == RDMA_PS_UDP)
4126                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4127
4128         if (dev_addr->bound_dev_if)
4129                 ndev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
4130         if (!ndev) {
4131                 err = -ENODEV;
4132                 goto out2;
4133         }
4134         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4135         mc->multicast.ib->rec.hop_limit = 1;
4136         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4137
4138         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4139                    rdma_start_port(id_priv->cma_dev->device)];
4140         if (addr->sa_family == AF_INET) {
4141                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4142                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4143                         if (!send_only) {
4144                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4145                                                     true);
4146                                 if (!err)
4147                                         mc->igmp_joined = true;
4148                         }
4149                 }
4150         } else {
4151                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4152                         err = -ENOTSUPP;
4153         }
4154         dev_put(ndev);
4155         if (err || !mc->multicast.ib->rec.mtu) {
4156                 if (!err)
4157                         err = -EINVAL;
4158                 goto out2;
4159         }
4160         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4161                     &mc->multicast.ib->rec.port_gid);
4162         work->id = id_priv;
4163         work->mc = mc;
4164         INIT_WORK(&work->work, iboe_mcast_work_handler);
4165         kref_get(&mc->mcref);
4166         queue_work(cma_wq, &work->work);
4167
4168         return 0;
4169
4170 out2:
4171         kfree(mc->multicast.ib);
4172 out1:
4173         kfree(work);
4174         return err;
4175 }
4176
4177 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4178                         u8 join_state, void *context)
4179 {
4180         struct rdma_id_private *id_priv;
4181         struct cma_multicast *mc;
4182         int ret;
4183
4184         id_priv = container_of(id, struct rdma_id_private, id);
4185         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4186             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4187                 return -EINVAL;
4188
4189         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4190         if (!mc)
4191                 return -ENOMEM;
4192
4193         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4194         mc->context = context;
4195         mc->id_priv = id_priv;
4196         mc->igmp_joined = false;
4197         mc->join_state = join_state;
4198         spin_lock(&id_priv->lock);
4199         list_add(&mc->list, &id_priv->mc_list);
4200         spin_unlock(&id_priv->lock);
4201
4202         if (rdma_protocol_roce(id->device, id->port_num)) {
4203                 kref_init(&mc->mcref);
4204                 ret = cma_iboe_join_multicast(id_priv, mc);
4205         } else if (rdma_cap_ib_mcast(id->device, id->port_num))
4206                 ret = cma_join_ib_multicast(id_priv, mc);
4207         else
4208                 ret = -ENOSYS;
4209
4210         if (ret) {
4211                 spin_lock_irq(&id_priv->lock);
4212                 list_del(&mc->list);
4213                 spin_unlock_irq(&id_priv->lock);
4214                 kfree(mc);
4215         }
4216         return ret;
4217 }
4218 EXPORT_SYMBOL(rdma_join_multicast);
4219
4220 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4221 {
4222         struct rdma_id_private *id_priv;
4223         struct cma_multicast *mc;
4224
4225         id_priv = container_of(id, struct rdma_id_private, id);
4226         spin_lock_irq(&id_priv->lock);
4227         list_for_each_entry(mc, &id_priv->mc_list, list) {
4228                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4229                         list_del(&mc->list);
4230                         spin_unlock_irq(&id_priv->lock);
4231
4232                         if (id->qp)
4233                                 ib_detach_mcast(id->qp,
4234                                                 &mc->multicast.ib->rec.mgid,
4235                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4236
4237                         BUG_ON(id_priv->cma_dev->device != id->device);
4238
4239                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4240                                 ib_sa_free_multicast(mc->multicast.ib);
4241                                 kfree(mc);
4242                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4243                                 if (mc->igmp_joined) {
4244                                         struct rdma_dev_addr *dev_addr =
4245                                                 &id->route.addr.dev_addr;
4246                                         struct net_device *ndev = NULL;
4247
4248                                         if (dev_addr->bound_dev_if)
4249                                                 ndev = dev_get_by_index(&init_net,
4250                                                                         dev_addr->bound_dev_if);
4251                                         if (ndev) {
4252                                                 cma_igmp_send(ndev,
4253                                                               &mc->multicast.ib->rec.mgid,
4254                                                               false);
4255                                                 dev_put(ndev);
4256                                         }
4257                                         mc->igmp_joined = false;
4258                                 }
4259                                 kref_put(&mc->mcref, release_mc);
4260                         }
4261                         return;
4262                 }
4263         }
4264         spin_unlock_irq(&id_priv->lock);
4265 }
4266 EXPORT_SYMBOL(rdma_leave_multicast);
4267
4268 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4269 {
4270         struct rdma_dev_addr *dev_addr;
4271         struct cma_ndev_work *work;
4272
4273         dev_addr = &id_priv->id.route.addr.dev_addr;
4274
4275         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4276             (net_eq(dev_net(ndev), dev_addr->net)) &&
4277             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4278                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4279                         ndev->name, &id_priv->id);
4280                 work = kzalloc(sizeof *work, GFP_KERNEL);
4281                 if (!work)
4282                         return -ENOMEM;
4283
4284                 INIT_WORK(&work->work, cma_ndev_work_handler);
4285                 work->id = id_priv;
4286                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4287                 atomic_inc(&id_priv->refcount);
4288                 queue_work(cma_wq, &work->work);
4289         }
4290
4291         return 0;
4292 }
4293
4294 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4295                                void *ptr)
4296 {
4297         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4298         struct cma_device *cma_dev;
4299         struct rdma_id_private *id_priv;
4300         int ret = NOTIFY_DONE;
4301
4302         if (event != NETDEV_BONDING_FAILOVER)
4303                 return NOTIFY_DONE;
4304
4305         if (!(ndev->flags & IFF_MASTER) || !(ndev->priv_flags & IFF_BONDING))
4306                 return NOTIFY_DONE;
4307
4308         mutex_lock(&lock);
4309         list_for_each_entry(cma_dev, &dev_list, list)
4310                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4311                         ret = cma_netdev_change(ndev, id_priv);
4312                         if (ret)
4313                                 goto out;
4314                 }
4315
4316 out:
4317         mutex_unlock(&lock);
4318         return ret;
4319 }
4320
4321 static struct notifier_block cma_nb = {
4322         .notifier_call = cma_netdev_callback
4323 };
4324
4325 static void cma_add_one(struct ib_device *device)
4326 {
4327         struct cma_device *cma_dev;
4328         struct rdma_id_private *id_priv;
4329         unsigned int i;
4330         unsigned long supported_gids = 0;
4331
4332         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4333         if (!cma_dev)
4334                 return;
4335
4336         cma_dev->device = device;
4337         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4338                                             sizeof(*cma_dev->default_gid_type),
4339                                             GFP_KERNEL);
4340         if (!cma_dev->default_gid_type)
4341                 goto free_cma_dev;
4342
4343         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4344                                             sizeof(*cma_dev->default_roce_tos),
4345                                             GFP_KERNEL);
4346         if (!cma_dev->default_roce_tos)
4347                 goto free_gid_type;
4348
4349         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4350                 supported_gids = roce_gid_type_mask_support(device, i);
4351                 WARN_ON(!supported_gids);
4352                 cma_dev->default_gid_type[i - rdma_start_port(device)] =
4353                         find_first_bit(&supported_gids, BITS_PER_LONG);
4354                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4355         }
4356
4357         init_completion(&cma_dev->comp);
4358         atomic_set(&cma_dev->refcount, 1);
4359         INIT_LIST_HEAD(&cma_dev->id_list);
4360         ib_set_client_data(device, &cma_client, cma_dev);
4361
4362         mutex_lock(&lock);
4363         list_add_tail(&cma_dev->list, &dev_list);
4364         list_for_each_entry(id_priv, &listen_any_list, list)
4365                 cma_listen_on_dev(id_priv, cma_dev);
4366         mutex_unlock(&lock);
4367
4368         return;
4369
4370 free_gid_type:
4371         kfree(cma_dev->default_gid_type);
4372
4373 free_cma_dev:
4374         kfree(cma_dev);
4375
4376         return;
4377 }
4378
4379 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4380 {
4381         struct rdma_cm_event event;
4382         enum rdma_cm_state state;
4383         int ret = 0;
4384
4385         /* Record that we want to remove the device */
4386         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4387         if (state == RDMA_CM_DESTROYING)
4388                 return 0;
4389
4390         cma_cancel_operation(id_priv, state);
4391         mutex_lock(&id_priv->handler_mutex);
4392
4393         /* Check for destruction from another callback. */
4394         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4395                 goto out;
4396
4397         memset(&event, 0, sizeof event);
4398         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4399         ret = id_priv->id.event_handler(&id_priv->id, &event);
4400 out:
4401         mutex_unlock(&id_priv->handler_mutex);
4402         return ret;
4403 }
4404
4405 static void cma_process_remove(struct cma_device *cma_dev)
4406 {
4407         struct rdma_id_private *id_priv;
4408         int ret;
4409
4410         mutex_lock(&lock);
4411         while (!list_empty(&cma_dev->id_list)) {
4412                 id_priv = list_entry(cma_dev->id_list.next,
4413                                      struct rdma_id_private, list);
4414
4415                 list_del(&id_priv->listen_list);
4416                 list_del_init(&id_priv->list);
4417                 atomic_inc(&id_priv->refcount);
4418                 mutex_unlock(&lock);
4419
4420                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4421                 cma_deref_id(id_priv);
4422                 if (ret)
4423                         rdma_destroy_id(&id_priv->id);
4424
4425                 mutex_lock(&lock);
4426         }
4427         mutex_unlock(&lock);
4428
4429         cma_deref_dev(cma_dev);
4430         wait_for_completion(&cma_dev->comp);
4431 }
4432
4433 static void cma_remove_one(struct ib_device *device, void *client_data)
4434 {
4435         struct cma_device *cma_dev = client_data;
4436
4437         if (!cma_dev)
4438                 return;
4439
4440         mutex_lock(&lock);
4441         list_del(&cma_dev->list);
4442         mutex_unlock(&lock);
4443
4444         cma_process_remove(cma_dev);
4445         kfree(cma_dev->default_roce_tos);
4446         kfree(cma_dev->default_gid_type);
4447         kfree(cma_dev);
4448 }
4449
4450 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4451 {
4452         struct nlmsghdr *nlh;
4453         struct rdma_cm_id_stats *id_stats;
4454         struct rdma_id_private *id_priv;
4455         struct rdma_cm_id *id = NULL;
4456         struct cma_device *cma_dev;
4457         int i_dev = 0, i_id = 0;
4458
4459         /*
4460          * We export all of the IDs as a sequence of messages.  Each
4461          * ID gets its own netlink message.
4462          */
4463         mutex_lock(&lock);
4464
4465         list_for_each_entry(cma_dev, &dev_list, list) {
4466                 if (i_dev < cb->args[0]) {
4467                         i_dev++;
4468                         continue;
4469                 }
4470
4471                 i_id = 0;
4472                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4473                         if (i_id < cb->args[1]) {
4474                                 i_id++;
4475                                 continue;
4476                         }
4477
4478                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4479                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4480                                                 RDMA_NL_RDMA_CM_ID_STATS,
4481                                                 NLM_F_MULTI);
4482                         if (!id_stats)
4483                                 goto out;
4484
4485                         memset(id_stats, 0, sizeof *id_stats);
4486                         id = &id_priv->id;
4487                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4488                         id_stats->port_num = id->port_num;
4489                         id_stats->bound_dev_if =
4490                                 id->route.addr.dev_addr.bound_dev_if;
4491
4492                         if (ibnl_put_attr(skb, nlh,
4493                                           rdma_addr_size(cma_src_addr(id_priv)),
4494                                           cma_src_addr(id_priv),
4495                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4496                                 goto out;
4497                         if (ibnl_put_attr(skb, nlh,
4498                                           rdma_addr_size(cma_src_addr(id_priv)),
4499                                           cma_dst_addr(id_priv),
4500                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4501                                 goto out;
4502
4503                         id_stats->pid           = id_priv->owner;
4504                         id_stats->port_space    = id->ps;
4505                         id_stats->cm_state      = id_priv->state;
4506                         id_stats->qp_num        = id_priv->qp_num;
4507                         id_stats->qp_type       = id->qp_type;
4508
4509                         i_id++;
4510                 }
4511
4512                 cb->args[1] = 0;
4513                 i_dev++;
4514         }
4515
4516 out:
4517         mutex_unlock(&lock);
4518         cb->args[0] = i_dev;
4519         cb->args[1] = i_id;
4520
4521         return skb->len;
4522 }
4523
4524 static const struct ibnl_client_cbs cma_cb_table[] = {
4525         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats,
4526                                        .module = THIS_MODULE },
4527 };
4528
4529 static int cma_init_net(struct net *net)
4530 {
4531         struct cma_pernet *pernet = cma_pernet(net);
4532
4533         idr_init(&pernet->tcp_ps);
4534         idr_init(&pernet->udp_ps);
4535         idr_init(&pernet->ipoib_ps);
4536         idr_init(&pernet->ib_ps);
4537
4538         return 0;
4539 }
4540
4541 static void cma_exit_net(struct net *net)
4542 {
4543         struct cma_pernet *pernet = cma_pernet(net);
4544
4545         idr_destroy(&pernet->tcp_ps);
4546         idr_destroy(&pernet->udp_ps);
4547         idr_destroy(&pernet->ipoib_ps);
4548         idr_destroy(&pernet->ib_ps);
4549 }
4550
4551 static struct pernet_operations cma_pernet_operations = {
4552         .init = cma_init_net,
4553         .exit = cma_exit_net,
4554         .id = &cma_pernet_id,
4555         .size = sizeof(struct cma_pernet),
4556 };
4557
4558 static int __init cma_init(void)
4559 {
4560         int ret;
4561
4562         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4563         if (!cma_wq)
4564                 return -ENOMEM;
4565
4566         ret = register_pernet_subsys(&cma_pernet_operations);
4567         if (ret)
4568                 goto err_wq;
4569
4570         ib_sa_register_client(&sa_client);
4571         rdma_addr_register_client(&addr_client);
4572         register_netdevice_notifier(&cma_nb);
4573
4574         ret = ib_register_client(&cma_client);
4575         if (ret)
4576                 goto err;
4577
4578         if (ibnl_add_client(RDMA_NL_RDMA_CM, ARRAY_SIZE(cma_cb_table),
4579                             cma_cb_table))
4580                 pr_warn("RDMA CMA: failed to add netlink callback\n");
4581         cma_configfs_init();
4582
4583         return 0;
4584
4585 err:
4586         unregister_netdevice_notifier(&cma_nb);
4587         rdma_addr_unregister_client(&addr_client);
4588         ib_sa_unregister_client(&sa_client);
4589 err_wq:
4590         destroy_workqueue(cma_wq);
4591         return ret;
4592 }
4593
4594 static void __exit cma_cleanup(void)
4595 {
4596         cma_configfs_exit();
4597         ibnl_remove_client(RDMA_NL_RDMA_CM);
4598         ib_unregister_client(&cma_client);
4599         unregister_netdevice_notifier(&cma_nb);
4600         rdma_addr_unregister_client(&addr_client);
4601         ib_sa_unregister_client(&sa_client);
4602         unregister_pernet_subsys(&cma_pernet_operations);
4603         destroy_workqueue(cma_wq);
4604 }
4605
4606 module_init(cma_init);
4607 module_exit(cma_cleanup);