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 #include "cma_priv.h"
  66
  67 MODULE_AUTHOR("Sean Hefty");
  68 MODULE_DESCRIPTION("Generic RDMA CM Agent");
  69 MODULE_LICENSE("Dual BSD/GPL");
  70
  71 #define CMA_CM_RESPONSE_TIMEOUT 20
  72 #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
  73 #define CMA_MAX_CM_RETRIES 15
  74 #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
  75 #define CMA_IBOE_PACKET_LIFETIME 18
  76 #define CMA_PREFERRED_ROCE_GID_TYPE IB_GID_TYPE_ROCE_UDP_ENCAP
  77
  78 static const char * const cma_events[] = {
  79         [RDMA_CM_EVENT_ADDR_RESOLVED]    = "address resolved",
  80         [RDMA_CM_EVENT_ADDR_ERROR]       = "address error",
  81         [RDMA_CM_EVENT_ROUTE_RESOLVED]   = "route resolved ",
  82         [RDMA_CM_EVENT_ROUTE_ERROR]      = "route error",
  83         [RDMA_CM_EVENT_CONNECT_REQUEST]  = "connect request",
  84         [RDMA_CM_EVENT_CONNECT_RESPONSE] = "connect response",
  85         [RDMA_CM_EVENT_CONNECT_ERROR]    = "connect error",
  86         [RDMA_CM_EVENT_UNREACHABLE]      = "unreachable",
  87         [RDMA_CM_EVENT_REJECTED]         = "rejected",
  88         [RDMA_CM_EVENT_ESTABLISHED]      = "established",
  89         [RDMA_CM_EVENT_DISCONNECTED]     = "disconnected",
  90         [RDMA_CM_EVENT_DEVICE_REMOVAL]   = "device removal",
  91         [RDMA_CM_EVENT_MULTICAST_JOIN]   = "multicast join",
  92         [RDMA_CM_EVENT_MULTICAST_ERROR]  = "multicast error",
  93         [RDMA_CM_EVENT_ADDR_CHANGE]      = "address change",
  94         [RDMA_CM_EVENT_TIMEWAIT_EXIT]    = "timewait exit",
  95 };
  96
  97 const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event)
  98 {
  99         size_t index = event;
 100
 101         return (index < ARRAY_SIZE(cma_events) && cma_events[index]) ?
 102                         cma_events[index] : "unrecognized event";
 103 }
 104 EXPORT_SYMBOL(rdma_event_msg);
 105
 106 const char *__attribute_const__ rdma_reject_msg(struct rdma_cm_id *id,
 107                                                 int reason)
 108 {
 109         if (rdma_ib_or_roce(id->device, id->port_num))
 110                 return ibcm_reject_msg(reason);
 111
 112         if (rdma_protocol_iwarp(id->device, id->port_num))
 113                 return iwcm_reject_msg(reason);
 114
 115         WARN_ON_ONCE(1);
 116         return "unrecognized transport";
 117 }
 118 EXPORT_SYMBOL(rdma_reject_msg);
 119
 120 bool rdma_is_consumer_reject(struct rdma_cm_id *id, int reason)
 121 {
 122         if (rdma_ib_or_roce(id->device, id->port_num))
 123                 return reason == IB_CM_REJ_CONSUMER_DEFINED;
 124
 125         if (rdma_protocol_iwarp(id->device, id->port_num))
 126                 return reason == -ECONNREFUSED;
 127
 128         WARN_ON_ONCE(1);
 129         return false;
 130 }
 131 EXPORT_SYMBOL(rdma_is_consumer_reject);
 132
 133 const void *rdma_consumer_reject_data(struct rdma_cm_id *id,
 134                                       struct rdma_cm_event *ev, u8 *data_len)
 135 {
 136         const void *p;
 137
 138         if (rdma_is_consumer_reject(id, ev->status)) {
 139                 *data_len = ev->param.conn.private_data_len;
 140                 p = ev->param.conn.private_data;
 141         } else {
 142                 *data_len = 0;
 143                 p = NULL;
 144         }
 145         return p;
 146 }
 147 EXPORT_SYMBOL(rdma_consumer_reject_data);
 148
 149 /**
 150  * rdma_iw_cm_id() - return the iw_cm_id pointer for this cm_id.
 151  * @id: Communication Identifier
 152  */
 153 struct iw_cm_id *rdma_iw_cm_id(struct rdma_cm_id *id)
 154 {
 155         struct rdma_id_private *id_priv;
 156
 157         id_priv = container_of(id, struct rdma_id_private, id);
 158         if (id->device->node_type == RDMA_NODE_RNIC)
 159                 return id_priv->cm_id.iw;
 160         return NULL;
 161 }
 162 EXPORT_SYMBOL(rdma_iw_cm_id);
 163
 164 /**
 165  * rdma_res_to_id() - return the rdma_cm_id pointer for this restrack.
 166  * @res: rdma resource tracking entry pointer
 167  */
 168 struct rdma_cm_id *rdma_res_to_id(struct rdma_restrack_entry *res)
 169 {
 170         struct rdma_id_private *id_priv =
 171                 container_of(res, struct rdma_id_private, res);
 172
 173         return &id_priv->id;
 174 }
 175 EXPORT_SYMBOL(rdma_res_to_id);
 176
 177 static void cma_add_one(struct ib_device *device);
 178 static void cma_remove_one(struct ib_device *device, void *client_data);
 179
 180 static struct ib_client cma_client = {
 181         .name   = "cma",
 182         .add    = cma_add_one,
 183         .remove = cma_remove_one
 184 };
 185
 186 static struct ib_sa_client sa_client;
 187 static LIST_HEAD(dev_list);
 188 static LIST_HEAD(listen_any_list);
 189 static DEFINE_MUTEX(lock);
 190 static struct workqueue_struct *cma_wq;
 191 static unsigned int cma_pernet_id;
 192
 193 struct cma_pernet {
 194         struct idr tcp_ps;
 195         struct idr udp_ps;
 196         struct idr ipoib_ps;
 197         struct idr ib_ps;
 198 };
 199
 200 static struct cma_pernet *cma_pernet(struct net *net)
 201 {
 202         return net_generic(net, cma_pernet_id);
 203 }
 204
 205 static struct idr *cma_pernet_idr(struct net *net, enum rdma_ucm_port_space ps)
 206 {
 207         struct cma_pernet *pernet = cma_pernet(net);
 208
 209         switch (ps) {
 210         case RDMA_PS_TCP:
 211                 return &pernet->tcp_ps;
 212         case RDMA_PS_UDP:
 213                 return &pernet->udp_ps;
 214         case RDMA_PS_IPOIB:
 215                 return &pernet->ipoib_ps;
 216         case RDMA_PS_IB:
 217                 return &pernet->ib_ps;
 218         default:
 219                 return NULL;
 220         }
 221 }
 222
 223 struct cma_device {
 224         struct list_head        list;
 225         struct ib_device        *device;
 226         struct completion       comp;
 227         atomic_t                refcount;
 228         struct list_head        id_list;
 229         enum ib_gid_type        *default_gid_type;
 230         u8                      *default_roce_tos;
 231 };
 232
 233 struct rdma_bind_list {
 234         enum rdma_ucm_port_space ps;
 235         struct hlist_head       owners;
 236         unsigned short          port;
 237 };
 238
 239 struct class_port_info_context {
 240         struct ib_class_port_info       *class_port_info;
 241         struct ib_device                *device;
 242         struct completion               done;
 243         struct ib_sa_query              *sa_query;
 244         u8                              port_num;
 245 };
 246
 247 static int cma_ps_alloc(struct net *net, enum rdma_ucm_port_space ps,
 248                         struct rdma_bind_list *bind_list, int snum)
 249 {
 250         struct idr *idr = cma_pernet_idr(net, ps);
 251
 252         return idr_alloc(idr, bind_list, snum, snum + 1, GFP_KERNEL);
 253 }
 254
 255 static struct rdma_bind_list *cma_ps_find(struct net *net,
 256                                           enum rdma_ucm_port_space ps, int snum)
 257 {
 258         struct idr *idr = cma_pernet_idr(net, ps);
 259
 260         return idr_find(idr, snum);
 261 }
 262
 263 static void cma_ps_remove(struct net *net, enum rdma_ucm_port_space ps,
 264                           int snum)
 265 {
 266         struct idr *idr = cma_pernet_idr(net, ps);
 267
 268         idr_remove(idr, snum);
 269 }
 270
 271 enum {
 272         CMA_OPTION_AFONLY,
 273 };
 274
 275 void cma_ref_dev(struct cma_device *cma_dev)
 276 {
 277         atomic_inc(&cma_dev->refcount);
 278 }
 279
 280 struct cma_device *cma_enum_devices_by_ibdev(cma_device_filter  filter,
 281                                              void               *cookie)
 282 {
 283         struct cma_device *cma_dev;
 284         struct cma_device *found_cma_dev = NULL;
 285
 286         mutex_lock(&lock);
 287
 288         list_for_each_entry(cma_dev, &dev_list, list)
 289                 if (filter(cma_dev->device, cookie)) {
 290                         found_cma_dev = cma_dev;
 291                         break;
 292                 }
 293
 294         if (found_cma_dev)
 295                 cma_ref_dev(found_cma_dev);
 296         mutex_unlock(&lock);
 297         return found_cma_dev;
 298 }
 299
 300 int cma_get_default_gid_type(struct cma_device *cma_dev,
 301                              unsigned int port)
 302 {
 303         if (!rdma_is_port_valid(cma_dev->device, port))
 304                 return -EINVAL;
 305
 306         return cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)];
 307 }
 308
 309 int cma_set_default_gid_type(struct cma_device *cma_dev,
 310                              unsigned int port,
 311                              enum ib_gid_type default_gid_type)
 312 {
 313         unsigned long supported_gids;
 314
 315         if (!rdma_is_port_valid(cma_dev->device, port))
 316                 return -EINVAL;
 317
 318         supported_gids = roce_gid_type_mask_support(cma_dev->device, port);
 319
 320         if (!(supported_gids & 1 << default_gid_type))
 321                 return -EINVAL;
 322
 323         cma_dev->default_gid_type[port - rdma_start_port(cma_dev->device)] =
 324                 default_gid_type;
 325
 326         return 0;
 327 }
 328
 329 int cma_get_default_roce_tos(struct cma_device *cma_dev, unsigned int port)
 330 {
 331         if (!rdma_is_port_valid(cma_dev->device, port))
 332                 return -EINVAL;
 333
 334         return cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)];
 335 }
 336
 337 int cma_set_default_roce_tos(struct cma_device *cma_dev, unsigned int port,
 338                              u8 default_roce_tos)
 339 {
 340         if (!rdma_is_port_valid(cma_dev->device, port))
 341                 return -EINVAL;
 342
 343         cma_dev->default_roce_tos[port - rdma_start_port(cma_dev->device)] =
 344                  default_roce_tos;
 345
 346         return 0;
 347 }
 348 struct ib_device *cma_get_ib_dev(struct cma_device *cma_dev)
 349 {
 350         return cma_dev->device;
 351 }
 352
 353 /*
 354  * Device removal can occur at anytime, so we need extra handling to
 355  * serialize notifying the user of device removal with other callbacks.
 356  * We do this by disabling removal notification while a callback is in process,
 357  * and reporting it after the callback completes.
 358  */
 359
 360 struct cma_multicast {
 361         struct rdma_id_private *id_priv;
 362         union {
 363                 struct ib_sa_multicast *ib;
 364         } multicast;
 365         struct list_head        list;
 366         void                    *context;
 367         struct sockaddr_storage addr;
 368         struct kref             mcref;
 369         u8                      join_state;
 370 };
 371
 372 struct cma_work {
 373         struct work_struct      work;
 374         struct rdma_id_private  *id;
 375         enum rdma_cm_state      old_state;
 376         enum rdma_cm_state      new_state;
 377         struct rdma_cm_event    event;
 378 };
 379
 380 struct cma_ndev_work {
 381         struct work_struct      work;
 382         struct rdma_id_private  *id;
 383         struct rdma_cm_event    event;
 384 };
 385
 386 struct iboe_mcast_work {
 387         struct work_struct       work;
 388         struct rdma_id_private  *id;
 389         struct cma_multicast    *mc;
 390 };
 391
 392 union cma_ip_addr {
 393         struct in6_addr ip6;
 394         struct {
 395                 __be32 pad[3];
 396                 __be32 addr;
 397         } ip4;
 398 };
 399
 400 struct cma_hdr {
 401         u8 cma_version;
 402         u8 ip_version;  /* IP version: 7:4 */
 403         __be16 port;
 404         union cma_ip_addr src_addr;
 405         union cma_ip_addr dst_addr;
 406 };
 407
 408 #define CMA_VERSION 0x00
 409
 410 struct cma_req_info {
 411         struct sockaddr_storage listen_addr_storage;
 412         struct sockaddr_storage src_addr_storage;
 413         struct ib_device *device;
 414         union ib_gid local_gid;
 415         __be64 service_id;
 416         int port;
 417         bool has_gid;
 418         u16 pkey;
 419 };
 420
 421 static int cma_comp(struct rdma_id_private *id_priv, enum rdma_cm_state comp)
 422 {
 423         unsigned long flags;
 424         int ret;
 425
 426         spin_lock_irqsave(&id_priv->lock, flags);
 427         ret = (id_priv->state == comp);
 428         spin_unlock_irqrestore(&id_priv->lock, flags);
 429         return ret;
 430 }
 431
 432 static int cma_comp_exch(struct rdma_id_private *id_priv,
 433                          enum rdma_cm_state comp, enum rdma_cm_state exch)
 434 {
 435         unsigned long flags;
 436         int ret;
 437
 438         spin_lock_irqsave(&id_priv->lock, flags);
 439         if ((ret = (id_priv->state == comp)))
 440                 id_priv->state = exch;
 441         spin_unlock_irqrestore(&id_priv->lock, flags);
 442         return ret;
 443 }
 444
 445 static enum rdma_cm_state cma_exch(struct rdma_id_private *id_priv,
 446                                    enum rdma_cm_state exch)
 447 {
 448         unsigned long flags;
 449         enum rdma_cm_state old;
 450
 451         spin_lock_irqsave(&id_priv->lock, flags);
 452         old = id_priv->state;
 453         id_priv->state = exch;
 454         spin_unlock_irqrestore(&id_priv->lock, flags);
 455         return old;
 456 }
 457
 458 static inline u8 cma_get_ip_ver(const struct cma_hdr *hdr)
 459 {
 460         return hdr->ip_version >> 4;
 461 }
 462
 463 static inline void cma_set_ip_ver(struct cma_hdr *hdr, u8 ip_ver)
 464 {
 465         hdr->ip_version = (ip_ver << 4) | (hdr->ip_version & 0xF);
 466 }
 467
 468 static int cma_igmp_send(struct net_device *ndev, union ib_gid *mgid, bool join)
 469 {
 470         struct in_device *in_dev = NULL;
 471
 472         if (ndev) {
 473                 rtnl_lock();
 474                 in_dev = __in_dev_get_rtnl(ndev);
 475                 if (in_dev) {
 476                         if (join)
 477                                 ip_mc_inc_group(in_dev,
 478                                                 *(__be32 *)(mgid->raw + 12));
 479                         else
 480                                 ip_mc_dec_group(in_dev,
 481                                                 *(__be32 *)(mgid->raw + 12));
 482                 }
 483                 rtnl_unlock();
 484         }
 485         return (in_dev) ? 0 : -ENODEV;
 486 }
 487
 488 static void _cma_attach_to_dev(struct rdma_id_private *id_priv,
 489                                struct cma_device *cma_dev)
 490 {
 491         cma_ref_dev(cma_dev);
 492         id_priv->cma_dev = cma_dev;
 493         id_priv->id.device = cma_dev->device;
 494         id_priv->id.route.addr.dev_addr.transport =
 495                 rdma_node_get_transport(cma_dev->device->node_type);
 496         list_add_tail(&id_priv->list, &cma_dev->id_list);
 497         rdma_restrack_add(&id_priv->res);
 498 }
 499
 500 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 501                               struct cma_device *cma_dev)
 502 {
 503         _cma_attach_to_dev(id_priv, cma_dev);
 504         id_priv->gid_type =
 505                 cma_dev->default_gid_type[id_priv->id.port_num -
 506                                           rdma_start_port(cma_dev->device)];
 507 }
 508
 509 void cma_deref_dev(struct cma_device *cma_dev)
 510 {
 511         if (atomic_dec_and_test(&cma_dev->refcount))
 512                 complete(&cma_dev->comp);
 513 }
 514
 515 static inline void release_mc(struct kref *kref)
 516 {
 517         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 518
 519         kfree(mc->multicast.ib);
 520         kfree(mc);
 521 }
 522
 523 static void cma_release_dev(struct rdma_id_private *id_priv)
 524 {
 525         mutex_lock(&lock);
 526         list_del(&id_priv->list);
 527         cma_deref_dev(id_priv->cma_dev);
 528         id_priv->cma_dev = NULL;
 529         mutex_unlock(&lock);
 530 }
 531
 532 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 533 {
 534         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 535 }
 536
 537 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 538 {
 539         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 540 }
 541
 542 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 543 {
 544         return id_priv->id.route.addr.src_addr.ss_family;
 545 }
 546
 547 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 548 {
 549         struct ib_sa_mcmember_rec rec;
 550         int ret = 0;
 551
 552         if (id_priv->qkey) {
 553                 if (qkey && id_priv->qkey != qkey)
 554                         return -EINVAL;
 555                 return 0;
 556         }
 557
 558         if (qkey) {
 559                 id_priv->qkey = qkey;
 560                 return 0;
 561         }
 562
 563         switch (id_priv->id.ps) {
 564         case RDMA_PS_UDP:
 565         case RDMA_PS_IB:
 566                 id_priv->qkey = RDMA_UDP_QKEY;
 567                 break;
 568         case RDMA_PS_IPOIB:
 569                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 570                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 571                                              id_priv->id.port_num, &rec.mgid,
 572                                              &rec);
 573                 if (!ret)
 574                         id_priv->qkey = be32_to_cpu(rec.qkey);
 575                 break;
 576         default:
 577                 break;
 578         }
 579         return ret;
 580 }
 581
 582 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 583 {
 584         dev_addr->dev_type = ARPHRD_INFINIBAND;
 585         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 586         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 587 }
 588
 589 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 590 {
 591         int ret;
 592
 593         if (addr->sa_family != AF_IB) {
 594                 ret = rdma_translate_ip(addr, dev_addr);
 595         } else {
 596                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 597                 ret = 0;
 598         }
 599
 600         return ret;
 601 }
 602
 603 static const struct ib_gid_attr *
 604 cma_validate_port(struct ib_device *device, u8 port,
 605                   enum ib_gid_type gid_type,
 606                   union ib_gid *gid,
 607                   struct rdma_id_private *id_priv)
 608 {
 609         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 610         int bound_if_index = dev_addr->bound_dev_if;
 611         const struct ib_gid_attr *sgid_attr;
 612         int dev_type = dev_addr->dev_type;
 613         struct net_device *ndev = NULL;
 614
 615         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 616                 return ERR_PTR(-ENODEV);
 617
 618         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 619                 return ERR_PTR(-ENODEV);
 620
 621         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 622                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 623                 if (!ndev)
 624                         return ERR_PTR(-ENODEV);
 625         } else {
 626                 gid_type = IB_GID_TYPE_IB;
 627         }
 628
 629         sgid_attr = rdma_find_gid_by_port(device, gid, gid_type, port, ndev);
 630         if (ndev)
 631                 dev_put(ndev);
 632         return sgid_attr;
 633 }
 634
 635 static void cma_bind_sgid_attr(struct rdma_id_private *id_priv,
 636                                const struct ib_gid_attr *sgid_attr)
 637 {
 638         WARN_ON(id_priv->id.route.addr.dev_addr.sgid_attr);
 639         id_priv->id.route.addr.dev_addr.sgid_attr = sgid_attr;
 640 }
 641
 642 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 643                            const struct rdma_id_private *listen_id_priv)
 644 {
 645         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 646         const struct ib_gid_attr *sgid_attr;
 647         struct cma_device *cma_dev;
 648         union ib_gid gid, iboe_gid, *gidp;
 649         enum ib_gid_type gid_type;
 650         int ret = -ENODEV;
 651         u8 port;
 652
 653         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 654             id_priv->id.ps == RDMA_PS_IPOIB)
 655                 return -EINVAL;
 656
 657         mutex_lock(&lock);
 658         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 659                     &iboe_gid);
 660
 661         memcpy(&gid, dev_addr->src_dev_addr +
 662                rdma_addr_gid_offset(dev_addr), sizeof gid);
 663
 664         if (listen_id_priv) {
 665                 cma_dev = listen_id_priv->cma_dev;
 666                 port = listen_id_priv->id.port_num;
 667                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 668                        &iboe_gid : &gid;
 669                 gid_type = listen_id_priv->gid_type;
 670                 sgid_attr = cma_validate_port(cma_dev->device, port,
 671                                               gid_type, gidp, id_priv);
 672                 if (!IS_ERR(sgid_attr)) {
 673                         id_priv->id.port_num = port;
 674                         cma_bind_sgid_attr(id_priv, sgid_attr);
 675                         ret = 0;
 676                         goto out;
 677                 }
 678         }
 679
 680         list_for_each_entry(cma_dev, &dev_list, list) {
 681                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 682                         if (listen_id_priv &&
 683                             listen_id_priv->cma_dev == cma_dev &&
 684                             listen_id_priv->id.port_num == port)
 685                                 continue;
 686
 687                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 688                                &iboe_gid : &gid;
 689                         gid_type = cma_dev->default_gid_type[port - 1];
 690                         sgid_attr = cma_validate_port(cma_dev->device, port,
 691                                                       gid_type, gidp, id_priv);
 692                         if (!IS_ERR(sgid_attr)) {
 693                                 id_priv->id.port_num = port;
 694                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 695                                 ret = 0;
 696                                 goto out;
 697                         }
 698                 }
 699         }
 700
 701 out:
 702         if (!ret)
 703                 cma_attach_to_dev(id_priv, cma_dev);
 704
 705         mutex_unlock(&lock);
 706         return ret;
 707 }
 708
 709 /*
 710  * Select the source IB device and address to reach the destination IB address.
 711  */
 712 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 713 {
 714         struct cma_device *cma_dev, *cur_dev;
 715         struct sockaddr_ib *addr;
 716         union ib_gid gid, sgid, *dgid;
 717         u16 pkey, index;
 718         u8 p;
 719         enum ib_port_state port_state;
 720         int i;
 721
 722         cma_dev = NULL;
 723         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 724         dgid = (union ib_gid *) &addr->sib_addr;
 725         pkey = ntohs(addr->sib_pkey);
 726
 727         mutex_lock(&lock);
 728         list_for_each_entry(cur_dev, &dev_list, list) {
 729                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 730                         if (!rdma_cap_af_ib(cur_dev->device, p))
 731                                 continue;
 732
 733                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 734                                 continue;
 735
 736                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 737                                 continue;
 738                         for (i = 0; !rdma_query_gid(cur_dev->device,
 739                                                     p, i, &gid);
 740                              i++) {
 741                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 742                                         cma_dev = cur_dev;
 743                                         sgid = gid;
 744                                         id_priv->id.port_num = p;
 745                                         goto found;
 746                                 }
 747
 748                                 if (!cma_dev && (gid.global.subnet_prefix ==
 749                                     dgid->global.subnet_prefix) &&
 750                                     port_state == IB_PORT_ACTIVE) {
 751                                         cma_dev = cur_dev;
 752                                         sgid = gid;
 753                                         id_priv->id.port_num = p;
 754                                         goto found;
 755                                 }
 756                         }
 757                 }
 758         }
 759         mutex_unlock(&lock);
 760         return -ENODEV;
 761
 762 found:
 763         cma_attach_to_dev(id_priv, cma_dev);
 764         mutex_unlock(&lock);
 765         addr = (struct sockaddr_ib *)cma_src_addr(id_priv);
 766         memcpy(&addr->sib_addr, &sgid, sizeof(sgid));
 767         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 768         return 0;
 769 }
 770
 771 static void cma_deref_id(struct rdma_id_private *id_priv)
 772 {
 773         if (atomic_dec_and_test(&id_priv->refcount))
 774                 complete(&id_priv->comp);
 775 }
 776
 777 struct rdma_cm_id *__rdma_create_id(struct net *net,
 778                                     rdma_cm_event_handler event_handler,
 779                                     void *context, enum rdma_ucm_port_space ps,
 780                                     enum ib_qp_type qp_type, const char *caller)
 781 {
 782         struct rdma_id_private *id_priv;
 783
 784         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 785         if (!id_priv)
 786                 return ERR_PTR(-ENOMEM);
 787
 788         if (caller)
 789                 id_priv->res.kern_name = caller;
 790         else
 791                 rdma_restrack_set_task(&id_priv->res, current);
 792         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 793         id_priv->state = RDMA_CM_IDLE;
 794         id_priv->id.context = context;
 795         id_priv->id.event_handler = event_handler;
 796         id_priv->id.ps = ps;
 797         id_priv->id.qp_type = qp_type;
 798         id_priv->tos_set = false;
 799         id_priv->gid_type = IB_GID_TYPE_IB;
 800         spin_lock_init(&id_priv->lock);
 801         mutex_init(&id_priv->qp_mutex);
 802         init_completion(&id_priv->comp);
 803         atomic_set(&id_priv->refcount, 1);
 804         mutex_init(&id_priv->handler_mutex);
 805         INIT_LIST_HEAD(&id_priv->listen_list);
 806         INIT_LIST_HEAD(&id_priv->mc_list);
 807         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 808         id_priv->id.route.addr.dev_addr.net = get_net(net);
 809         id_priv->seq_num &= 0x00ffffff;
 810
 811         return &id_priv->id;
 812 }
 813 EXPORT_SYMBOL(__rdma_create_id);
 814
 815 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 816 {
 817         struct ib_qp_attr qp_attr;
 818         int qp_attr_mask, ret;
 819
 820         qp_attr.qp_state = IB_QPS_INIT;
 821         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 822         if (ret)
 823                 return ret;
 824
 825         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 826         if (ret)
 827                 return ret;
 828
 829         qp_attr.qp_state = IB_QPS_RTR;
 830         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 831         if (ret)
 832                 return ret;
 833
 834         qp_attr.qp_state = IB_QPS_RTS;
 835         qp_attr.sq_psn = 0;
 836         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 837
 838         return ret;
 839 }
 840
 841 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 842 {
 843         struct ib_qp_attr qp_attr;
 844         int qp_attr_mask, ret;
 845
 846         qp_attr.qp_state = IB_QPS_INIT;
 847         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 848         if (ret)
 849                 return ret;
 850
 851         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 852 }
 853
 854 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 855                    struct ib_qp_init_attr *qp_init_attr)
 856 {
 857         struct rdma_id_private *id_priv;
 858         struct ib_qp *qp;
 859         int ret;
 860
 861         id_priv = container_of(id, struct rdma_id_private, id);
 862         if (id->device != pd->device)
 863                 return -EINVAL;
 864
 865         qp_init_attr->port_num = id->port_num;
 866         qp = ib_create_qp(pd, qp_init_attr);
 867         if (IS_ERR(qp))
 868                 return PTR_ERR(qp);
 869
 870         if (id->qp_type == IB_QPT_UD)
 871                 ret = cma_init_ud_qp(id_priv, qp);
 872         else
 873                 ret = cma_init_conn_qp(id_priv, qp);
 874         if (ret)
 875                 goto err;
 876
 877         id->qp = qp;
 878         id_priv->qp_num = qp->qp_num;
 879         id_priv->srq = (qp->srq != NULL);
 880         return 0;
 881 err:
 882         ib_destroy_qp(qp);
 883         return ret;
 884 }
 885 EXPORT_SYMBOL(rdma_create_qp);
 886
 887 void rdma_destroy_qp(struct rdma_cm_id *id)
 888 {
 889         struct rdma_id_private *id_priv;
 890
 891         id_priv = container_of(id, struct rdma_id_private, id);
 892         mutex_lock(&id_priv->qp_mutex);
 893         ib_destroy_qp(id_priv->id.qp);
 894         id_priv->id.qp = NULL;
 895         mutex_unlock(&id_priv->qp_mutex);
 896 }
 897 EXPORT_SYMBOL(rdma_destroy_qp);
 898
 899 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 900                              struct rdma_conn_param *conn_param)
 901 {
 902         struct ib_qp_attr qp_attr;
 903         int qp_attr_mask, ret;
 904
 905         mutex_lock(&id_priv->qp_mutex);
 906         if (!id_priv->id.qp) {
 907                 ret = 0;
 908                 goto out;
 909         }
 910
 911         /* Need to update QP attributes from default values. */
 912         qp_attr.qp_state = IB_QPS_INIT;
 913         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 914         if (ret)
 915                 goto out;
 916
 917         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 918         if (ret)
 919                 goto out;
 920
 921         qp_attr.qp_state = IB_QPS_RTR;
 922         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 923         if (ret)
 924                 goto out;
 925
 926         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 927
 928         if (conn_param)
 929                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 930         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 931 out:
 932         mutex_unlock(&id_priv->qp_mutex);
 933         return ret;
 934 }
 935
 936 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 937                              struct rdma_conn_param *conn_param)
 938 {
 939         struct ib_qp_attr qp_attr;
 940         int qp_attr_mask, ret;
 941
 942         mutex_lock(&id_priv->qp_mutex);
 943         if (!id_priv->id.qp) {
 944                 ret = 0;
 945                 goto out;
 946         }
 947
 948         qp_attr.qp_state = IB_QPS_RTS;
 949         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 950         if (ret)
 951                 goto out;
 952
 953         if (conn_param)
 954                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 955         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 956 out:
 957         mutex_unlock(&id_priv->qp_mutex);
 958         return ret;
 959 }
 960
 961 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 962 {
 963         struct ib_qp_attr qp_attr;
 964         int ret;
 965
 966         mutex_lock(&id_priv->qp_mutex);
 967         if (!id_priv->id.qp) {
 968                 ret = 0;
 969                 goto out;
 970         }
 971
 972         qp_attr.qp_state = IB_QPS_ERR;
 973         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 974 out:
 975         mutex_unlock(&id_priv->qp_mutex);
 976         return ret;
 977 }
 978
 979 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 980                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 981 {
 982         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 983         int ret;
 984         u16 pkey;
 985
 986         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 987                 pkey = 0xffff;
 988         else
 989                 pkey = ib_addr_get_pkey(dev_addr);
 990
 991         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 992                                   pkey, &qp_attr->pkey_index);
 993         if (ret)
 994                 return ret;
 995
 996         qp_attr->port_num = id_priv->id.port_num;
 997         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
 998
 999         if (id_priv->id.qp_type == IB_QPT_UD) {
1000                 ret = cma_set_qkey(id_priv, 0);
1001                 if (ret)
1002                         return ret;
1003
1004                 qp_attr->qkey = id_priv->qkey;
1005                 *qp_attr_mask |= IB_QP_QKEY;
1006         } else {
1007                 qp_attr->qp_access_flags = 0;
1008                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1009         }
1010         return 0;
1011 }
1012
1013 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1014                        int *qp_attr_mask)
1015 {
1016         struct rdma_id_private *id_priv;
1017         int ret = 0;
1018
1019         id_priv = container_of(id, struct rdma_id_private, id);
1020         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1021                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1022                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1023                 else
1024                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1025                                                  qp_attr_mask);
1026
1027                 if (qp_attr->qp_state == IB_QPS_RTR)
1028                         qp_attr->rq_psn = id_priv->seq_num;
1029         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1030                 if (!id_priv->cm_id.iw) {
1031                         qp_attr->qp_access_flags = 0;
1032                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1033                 } else
1034                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1035                                                  qp_attr_mask);
1036                 qp_attr->port_num = id_priv->id.port_num;
1037                 *qp_attr_mask |= IB_QP_PORT;
1038         } else
1039                 ret = -ENOSYS;
1040
1041         return ret;
1042 }
1043 EXPORT_SYMBOL(rdma_init_qp_attr);
1044
1045 static inline bool cma_zero_addr(const struct sockaddr *addr)
1046 {
1047         switch (addr->sa_family) {
1048         case AF_INET:
1049                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1050         case AF_INET6:
1051                 return ipv6_addr_any(&((struct sockaddr_in6 *)addr)->sin6_addr);
1052         case AF_IB:
1053                 return ib_addr_any(&((struct sockaddr_ib *)addr)->sib_addr);
1054         default:
1055                 return false;
1056         }
1057 }
1058
1059 static inline bool cma_loopback_addr(const struct sockaddr *addr)
1060 {
1061         switch (addr->sa_family) {
1062         case AF_INET:
1063                 return ipv4_is_loopback(
1064                         ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1065         case AF_INET6:
1066                 return ipv6_addr_loopback(
1067                         &((struct sockaddr_in6 *)addr)->sin6_addr);
1068         case AF_IB:
1069                 return ib_addr_loopback(
1070                         &((struct sockaddr_ib *)addr)->sib_addr);
1071         default:
1072                 return false;
1073         }
1074 }
1075
1076 static inline bool cma_any_addr(const struct sockaddr *addr)
1077 {
1078         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1079 }
1080
1081 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1082 {
1083         if (src->sa_family != dst->sa_family)
1084                 return -1;
1085
1086         switch (src->sa_family) {
1087         case AF_INET:
1088                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1089                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1090         case AF_INET6:
1091                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1092                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1093         default:
1094                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1095                                    &((struct sockaddr_ib *) dst)->sib_addr);
1096         }
1097 }
1098
1099 static __be16 cma_port(const struct sockaddr *addr)
1100 {
1101         struct sockaddr_ib *sib;
1102
1103         switch (addr->sa_family) {
1104         case AF_INET:
1105                 return ((struct sockaddr_in *) addr)->sin_port;
1106         case AF_INET6:
1107                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1108         case AF_IB:
1109                 sib = (struct sockaddr_ib *) addr;
1110                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1111                                     be64_to_cpu(sib->sib_sid_mask)));
1112         default:
1113                 return 0;
1114         }
1115 }
1116
1117 static inline int cma_any_port(const struct sockaddr *addr)
1118 {
1119         return !cma_port(addr);
1120 }
1121
1122 static void cma_save_ib_info(struct sockaddr *src_addr,
1123                              struct sockaddr *dst_addr,
1124                              const struct rdma_cm_id *listen_id,
1125                              const struct sa_path_rec *path)
1126 {
1127         struct sockaddr_ib *listen_ib, *ib;
1128
1129         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1130         if (src_addr) {
1131                 ib = (struct sockaddr_ib *)src_addr;
1132                 ib->sib_family = AF_IB;
1133                 if (path) {
1134                         ib->sib_pkey = path->pkey;
1135                         ib->sib_flowinfo = path->flow_label;
1136                         memcpy(&ib->sib_addr, &path->sgid, 16);
1137                         ib->sib_sid = path->service_id;
1138                         ib->sib_scope_id = 0;
1139                 } else {
1140                         ib->sib_pkey = listen_ib->sib_pkey;
1141                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1142                         ib->sib_addr = listen_ib->sib_addr;
1143                         ib->sib_sid = listen_ib->sib_sid;
1144                         ib->sib_scope_id = listen_ib->sib_scope_id;
1145                 }
1146                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1147         }
1148         if (dst_addr) {
1149                 ib = (struct sockaddr_ib *)dst_addr;
1150                 ib->sib_family = AF_IB;
1151                 if (path) {
1152                         ib->sib_pkey = path->pkey;
1153                         ib->sib_flowinfo = path->flow_label;
1154                         memcpy(&ib->sib_addr, &path->dgid, 16);
1155                 }
1156         }
1157 }
1158
1159 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1160                               struct sockaddr_in *dst_addr,
1161                               struct cma_hdr *hdr,
1162                               __be16 local_port)
1163 {
1164         if (src_addr) {
1165                 *src_addr = (struct sockaddr_in) {
1166                         .sin_family = AF_INET,
1167                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1168                         .sin_port = local_port,
1169                 };
1170         }
1171
1172         if (dst_addr) {
1173                 *dst_addr = (struct sockaddr_in) {
1174                         .sin_family = AF_INET,
1175                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1176                         .sin_port = hdr->port,
1177                 };
1178         }
1179 }
1180
1181 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1182                               struct sockaddr_in6 *dst_addr,
1183                               struct cma_hdr *hdr,
1184                               __be16 local_port)
1185 {
1186         if (src_addr) {
1187                 *src_addr = (struct sockaddr_in6) {
1188                         .sin6_family = AF_INET6,
1189                         .sin6_addr = hdr->dst_addr.ip6,
1190                         .sin6_port = local_port,
1191                 };
1192         }
1193
1194         if (dst_addr) {
1195                 *dst_addr = (struct sockaddr_in6) {
1196                         .sin6_family = AF_INET6,
1197                         .sin6_addr = hdr->src_addr.ip6,
1198                         .sin6_port = hdr->port,
1199                 };
1200         }
1201 }
1202
1203 static u16 cma_port_from_service_id(__be64 service_id)
1204 {
1205         return (u16)be64_to_cpu(service_id);
1206 }
1207
1208 static int cma_save_ip_info(struct sockaddr *src_addr,
1209                             struct sockaddr *dst_addr,
1210                             const struct ib_cm_event *ib_event,
1211                             __be64 service_id)
1212 {
1213         struct cma_hdr *hdr;
1214         __be16 port;
1215
1216         hdr = ib_event->private_data;
1217         if (hdr->cma_version != CMA_VERSION)
1218                 return -EINVAL;
1219
1220         port = htons(cma_port_from_service_id(service_id));
1221
1222         switch (cma_get_ip_ver(hdr)) {
1223         case 4:
1224                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1225                                   (struct sockaddr_in *)dst_addr, hdr, port);
1226                 break;
1227         case 6:
1228                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1229                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1230                 break;
1231         default:
1232                 return -EAFNOSUPPORT;
1233         }
1234
1235         return 0;
1236 }
1237
1238 static int cma_save_net_info(struct sockaddr *src_addr,
1239                              struct sockaddr *dst_addr,
1240                              const struct rdma_cm_id *listen_id,
1241                              const struct ib_cm_event *ib_event,
1242                              sa_family_t sa_family, __be64 service_id)
1243 {
1244         if (sa_family == AF_IB) {
1245                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1246                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1247                                          ib_event->param.req_rcvd.primary_path);
1248                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1249                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1250                 return 0;
1251         }
1252
1253         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1254 }
1255
1256 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1257                              struct cma_req_info *req)
1258 {
1259         const struct ib_cm_req_event_param *req_param =
1260                 &ib_event->param.req_rcvd;
1261         const struct ib_cm_sidr_req_event_param *sidr_param =
1262                 &ib_event->param.sidr_req_rcvd;
1263
1264         switch (ib_event->event) {
1265         case IB_CM_REQ_RECEIVED:
1266                 req->device     = req_param->listen_id->device;
1267                 req->port       = req_param->port;
1268                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1269                        sizeof(req->local_gid));
1270                 req->has_gid    = true;
1271                 req->service_id = req_param->primary_path->service_id;
1272                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1273                 if (req->pkey != req_param->bth_pkey)
1274                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1275                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1276                                             req_param->bth_pkey, req->pkey);
1277                 break;
1278         case IB_CM_SIDR_REQ_RECEIVED:
1279                 req->device     = sidr_param->listen_id->device;
1280                 req->port       = sidr_param->port;
1281                 req->has_gid    = false;
1282                 req->service_id = sidr_param->service_id;
1283                 req->pkey       = sidr_param->pkey;
1284                 if (req->pkey != sidr_param->bth_pkey)
1285                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1286                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1287                                             sidr_param->bth_pkey, req->pkey);
1288                 break;
1289         default:
1290                 return -EINVAL;
1291         }
1292
1293         return 0;
1294 }
1295
1296 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1297                                   const struct sockaddr_in *dst_addr,
1298                                   const struct sockaddr_in *src_addr)
1299 {
1300         __be32 daddr = dst_addr->sin_addr.s_addr,
1301                saddr = src_addr->sin_addr.s_addr;
1302         struct fib_result res;
1303         struct flowi4 fl4;
1304         int err;
1305         bool ret;
1306
1307         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1308             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1309             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1310             ipv4_is_loopback(saddr))
1311                 return false;
1312
1313         memset(&fl4, 0, sizeof(fl4));
1314         fl4.flowi4_iif = net_dev->ifindex;
1315         fl4.daddr = daddr;
1316         fl4.saddr = saddr;
1317
1318         rcu_read_lock();
1319         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1320         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1321         rcu_read_unlock();
1322
1323         return ret;
1324 }
1325
1326 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1327                                   const struct sockaddr_in6 *dst_addr,
1328                                   const struct sockaddr_in6 *src_addr)
1329 {
1330 #if IS_ENABLED(CONFIG_IPV6)
1331         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1332                            IPV6_ADDR_LINKLOCAL;
1333         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1334                                          &src_addr->sin6_addr, net_dev->ifindex,
1335                                          NULL, strict);
1336         bool ret;
1337
1338         if (!rt)
1339                 return false;
1340
1341         ret = rt->rt6i_idev->dev == net_dev;
1342         ip6_rt_put(rt);
1343
1344         return ret;
1345 #else
1346         return false;
1347 #endif
1348 }
1349
1350 static bool validate_net_dev(struct net_device *net_dev,
1351                              const struct sockaddr *daddr,
1352                              const struct sockaddr *saddr)
1353 {
1354         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1355         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1356         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1357         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1358
1359         switch (daddr->sa_family) {
1360         case AF_INET:
1361                 return saddr->sa_family == AF_INET &&
1362                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1363
1364         case AF_INET6:
1365                 return saddr->sa_family == AF_INET6 &&
1366                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1367
1368         default:
1369                 return false;
1370         }
1371 }
1372
1373 static struct net_device *
1374 roce_get_net_dev_by_cm_event(const struct ib_cm_event *ib_event)
1375 {
1376         const struct ib_gid_attr *sgid_attr = NULL;
1377
1378         if (ib_event->event == IB_CM_REQ_RECEIVED)
1379                 sgid_attr = ib_event->param.req_rcvd.ppath_sgid_attr;
1380         else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1381                 sgid_attr = ib_event->param.sidr_req_rcvd.sgid_attr;
1382
1383         if (!sgid_attr)
1384                 return NULL;
1385         dev_hold(sgid_attr->ndev);
1386         return sgid_attr->ndev;
1387 }
1388
1389 static struct net_device *cma_get_net_dev(const struct ib_cm_event *ib_event,
1390                                           struct cma_req_info *req)
1391 {
1392         struct sockaddr *listen_addr =
1393                         (struct sockaddr *)&req->listen_addr_storage;
1394         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1395         struct net_device *net_dev;
1396         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1397         int err;
1398
1399         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1400                                req->service_id);
1401         if (err)
1402                 return ERR_PTR(err);
1403
1404         if (rdma_protocol_roce(req->device, req->port))
1405                 net_dev = roce_get_net_dev_by_cm_event(ib_event);
1406         else
1407                 net_dev = ib_get_net_dev_by_params(req->device, req->port,
1408                                                    req->pkey,
1409                                                    gid, listen_addr);
1410         if (!net_dev)
1411                 return ERR_PTR(-ENODEV);
1412
1413         return net_dev;
1414 }
1415
1416 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1417 {
1418         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1419 }
1420
1421 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1422                                    const struct cma_hdr *hdr)
1423 {
1424         struct sockaddr *addr = cma_src_addr(id_priv);
1425         __be32 ip4_addr;
1426         struct in6_addr ip6_addr;
1427
1428         if (cma_any_addr(addr) && !id_priv->afonly)
1429                 return true;
1430
1431         switch (addr->sa_family) {
1432         case AF_INET:
1433                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1434                 if (cma_get_ip_ver(hdr) != 4)
1435                         return false;
1436                 if (!cma_any_addr(addr) &&
1437                     hdr->dst_addr.ip4.addr != ip4_addr)
1438                         return false;
1439                 break;
1440         case AF_INET6:
1441                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1442                 if (cma_get_ip_ver(hdr) != 6)
1443                         return false;
1444                 if (!cma_any_addr(addr) &&
1445                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1446                         return false;
1447                 break;
1448         case AF_IB:
1449                 return true;
1450         default:
1451                 return false;
1452         }
1453
1454         return true;
1455 }
1456
1457 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1458 {
1459         struct ib_device *device = id->device;
1460         const int port_num = id->port_num ?: rdma_start_port(device);
1461
1462         return rdma_protocol_roce(device, port_num);
1463 }
1464
1465 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1466                               const struct net_device *net_dev,
1467                               u8 port_num)
1468 {
1469         const struct rdma_addr *addr = &id->route.addr;
1470
1471         if (!net_dev)
1472                 /* This request is an AF_IB request */
1473                 return (!id->port_num || id->port_num == port_num) &&
1474                        (addr->src_addr.ss_family == AF_IB);
1475
1476         /*
1477          * Net namespaces must match, and if the listner is listening
1478          * on a specific netdevice than netdevice must match as well.
1479          */
1480         if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1481             (!!addr->dev_addr.bound_dev_if ==
1482              (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1483                 return true;
1484         else
1485                 return false;
1486 }
1487
1488 static struct rdma_id_private *cma_find_listener(
1489                 const struct rdma_bind_list *bind_list,
1490                 const struct ib_cm_id *cm_id,
1491                 const struct ib_cm_event *ib_event,
1492                 const struct cma_req_info *req,
1493                 const struct net_device *net_dev)
1494 {
1495         struct rdma_id_private *id_priv, *id_priv_dev;
1496
1497         if (!bind_list)
1498                 return ERR_PTR(-EINVAL);
1499
1500         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1501                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1502                         if (id_priv->id.device == cm_id->device &&
1503                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1504                                 return id_priv;
1505                         list_for_each_entry(id_priv_dev,
1506                                             &id_priv->listen_list,
1507                                             listen_list) {
1508                                 if (id_priv_dev->id.device == cm_id->device &&
1509                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1510                                         return id_priv_dev;
1511                         }
1512                 }
1513         }
1514
1515         return ERR_PTR(-EINVAL);
1516 }
1517
1518 static struct rdma_id_private *
1519 cma_ib_id_from_event(struct ib_cm_id *cm_id,
1520                      const struct ib_cm_event *ib_event,
1521                      struct net_device **net_dev)
1522 {
1523         struct cma_req_info req;
1524         struct rdma_bind_list *bind_list;
1525         struct rdma_id_private *id_priv;
1526         int err;
1527
1528         err = cma_save_req_info(ib_event, &req);
1529         if (err)
1530                 return ERR_PTR(err);
1531
1532         *net_dev = cma_get_net_dev(ib_event, &req);
1533         if (IS_ERR(*net_dev)) {
1534                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1535                         /* Assuming the protocol is AF_IB */
1536                         *net_dev = NULL;
1537                 } else {
1538                         return ERR_CAST(*net_dev);
1539                 }
1540         }
1541
1542         /*
1543          * Net namespace might be getting deleted while route lookup,
1544          * cm_id lookup is in progress. Therefore, perform netdevice
1545          * validation, cm_id lookup under rcu lock.
1546          * RCU lock along with netdevice state check, synchronizes with
1547          * netdevice migrating to different net namespace and also avoids
1548          * case where net namespace doesn't get deleted while lookup is in
1549          * progress.
1550          * If the device state is not IFF_UP, its properties such as ifindex
1551          * and nd_net cannot be trusted to remain valid without rcu lock.
1552          * net/core/dev.c change_net_namespace() ensures to synchronize with
1553          * ongoing operations on net device after device is closed using
1554          * synchronize_net().
1555          */
1556         rcu_read_lock();
1557         if (*net_dev) {
1558                 /*
1559                  * If netdevice is down, it is likely that it is administratively
1560                  * down or it might be migrating to different namespace.
1561                  * In that case avoid further processing, as the net namespace
1562                  * or ifindex may change.
1563                  */
1564                 if (((*net_dev)->flags & IFF_UP) == 0) {
1565                         id_priv = ERR_PTR(-EHOSTUNREACH);
1566                         goto err;
1567                 }
1568
1569                 if (!validate_net_dev(*net_dev,
1570                                  (struct sockaddr *)&req.listen_addr_storage,
1571                                  (struct sockaddr *)&req.src_addr_storage)) {
1572                         id_priv = ERR_PTR(-EHOSTUNREACH);
1573                         goto err;
1574                 }
1575         }
1576
1577         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1578                                 rdma_ps_from_service_id(req.service_id),
1579                                 cma_port_from_service_id(req.service_id));
1580         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1581 err:
1582         rcu_read_unlock();
1583         if (IS_ERR(id_priv) && *net_dev) {
1584                 dev_put(*net_dev);
1585                 *net_dev = NULL;
1586         }
1587         return id_priv;
1588 }
1589
1590 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1591 {
1592         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1593 }
1594
1595 static void cma_cancel_route(struct rdma_id_private *id_priv)
1596 {
1597         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1598                 if (id_priv->query)
1599                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1600         }
1601 }
1602
1603 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1604 {
1605         struct rdma_id_private *dev_id_priv;
1606
1607         /*
1608          * Remove from listen_any_list to prevent added devices from spawning
1609          * additional listen requests.
1610          */
1611         mutex_lock(&lock);
1612         list_del(&id_priv->list);
1613
1614         while (!list_empty(&id_priv->listen_list)) {
1615                 dev_id_priv = list_entry(id_priv->listen_list.next,
1616                                          struct rdma_id_private, listen_list);
1617                 /* sync with device removal to avoid duplicate destruction */
1618                 list_del_init(&dev_id_priv->list);
1619                 list_del(&dev_id_priv->listen_list);
1620                 mutex_unlock(&lock);
1621
1622                 rdma_destroy_id(&dev_id_priv->id);
1623                 mutex_lock(&lock);
1624         }
1625         mutex_unlock(&lock);
1626 }
1627
1628 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1629                                  enum rdma_cm_state state)
1630 {
1631         switch (state) {
1632         case RDMA_CM_ADDR_QUERY:
1633                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1634                 break;
1635         case RDMA_CM_ROUTE_QUERY:
1636                 cma_cancel_route(id_priv);
1637                 break;
1638         case RDMA_CM_LISTEN:
1639                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1640                         cma_cancel_listens(id_priv);
1641                 break;
1642         default:
1643                 break;
1644         }
1645 }
1646
1647 static void cma_release_port(struct rdma_id_private *id_priv)
1648 {
1649         struct rdma_bind_list *bind_list = id_priv->bind_list;
1650         struct net *net = id_priv->id.route.addr.dev_addr.net;
1651
1652         if (!bind_list)
1653                 return;
1654
1655         mutex_lock(&lock);
1656         hlist_del(&id_priv->node);
1657         if (hlist_empty(&bind_list->owners)) {
1658                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1659                 kfree(bind_list);
1660         }
1661         mutex_unlock(&lock);
1662 }
1663
1664 static void cma_leave_roce_mc_group(struct rdma_id_private *id_priv,
1665                                     struct cma_multicast *mc)
1666 {
1667         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1668         struct net_device *ndev = NULL;
1669
1670         if (dev_addr->bound_dev_if)
1671                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
1672         if (ndev) {
1673                 cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid, false);
1674                 dev_put(ndev);
1675         }
1676         kref_put(&mc->mcref, release_mc);
1677 }
1678
1679 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1680 {
1681         struct cma_multicast *mc;
1682
1683         while (!list_empty(&id_priv->mc_list)) {
1684                 mc = container_of(id_priv->mc_list.next,
1685                                   struct cma_multicast, list);
1686                 list_del(&mc->list);
1687                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1688                                       id_priv->id.port_num)) {
1689                         ib_sa_free_multicast(mc->multicast.ib);
1690                         kfree(mc);
1691                 } else {
1692                         cma_leave_roce_mc_group(id_priv, mc);
1693                 }
1694         }
1695 }
1696
1697 void rdma_destroy_id(struct rdma_cm_id *id)
1698 {
1699         struct rdma_id_private *id_priv;
1700         enum rdma_cm_state state;
1701
1702         id_priv = container_of(id, struct rdma_id_private, id);
1703         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1704         cma_cancel_operation(id_priv, state);
1705
1706         /*
1707          * Wait for any active callback to finish.  New callbacks will find
1708          * the id_priv state set to destroying and abort.
1709          */
1710         mutex_lock(&id_priv->handler_mutex);
1711         mutex_unlock(&id_priv->handler_mutex);
1712
1713         if (id_priv->cma_dev) {
1714                 rdma_restrack_del(&id_priv->res);
1715                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1716                         if (id_priv->cm_id.ib)
1717                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1718                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1719                         if (id_priv->cm_id.iw)
1720                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1721                 }
1722                 cma_leave_mc_groups(id_priv);
1723                 cma_release_dev(id_priv);
1724         }
1725
1726         cma_release_port(id_priv);
1727         cma_deref_id(id_priv);
1728         wait_for_completion(&id_priv->comp);
1729
1730         if (id_priv->internal_id)
1731                 cma_deref_id(id_priv->id.context);
1732
1733         kfree(id_priv->id.route.path_rec);
1734
1735         if (id_priv->id.route.addr.dev_addr.sgid_attr)
1736                 rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
1737
1738         put_net(id_priv->id.route.addr.dev_addr.net);
1739         kfree(id_priv);
1740 }
1741 EXPORT_SYMBOL(rdma_destroy_id);
1742
1743 static int cma_rep_recv(struct rdma_id_private *id_priv)
1744 {
1745         int ret;
1746
1747         ret = cma_modify_qp_rtr(id_priv, NULL);
1748         if (ret)
1749                 goto reject;
1750
1751         ret = cma_modify_qp_rts(id_priv, NULL);
1752         if (ret)
1753                 goto reject;
1754
1755         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1756         if (ret)
1757                 goto reject;
1758
1759         return 0;
1760 reject:
1761         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1762         cma_modify_qp_err(id_priv);
1763         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1764                        NULL, 0, NULL, 0);
1765         return ret;
1766 }
1767
1768 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1769                                    const struct ib_cm_rep_event_param *rep_data,
1770                                    void *private_data)
1771 {
1772         event->param.conn.private_data = private_data;
1773         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1774         event->param.conn.responder_resources = rep_data->responder_resources;
1775         event->param.conn.initiator_depth = rep_data->initiator_depth;
1776         event->param.conn.flow_control = rep_data->flow_control;
1777         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1778         event->param.conn.srq = rep_data->srq;
1779         event->param.conn.qp_num = rep_data->remote_qpn;
1780 }
1781
1782 static int cma_ib_handler(struct ib_cm_id *cm_id,
1783                           const struct ib_cm_event *ib_event)
1784 {
1785         struct rdma_id_private *id_priv = cm_id->context;
1786         struct rdma_cm_event event = {};
1787         int ret = 0;
1788
1789         mutex_lock(&id_priv->handler_mutex);
1790         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1791              id_priv->state != RDMA_CM_CONNECT) ||
1792             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1793              id_priv->state != RDMA_CM_DISCONNECT))
1794                 goto out;
1795
1796         switch (ib_event->event) {
1797         case IB_CM_REQ_ERROR:
1798         case IB_CM_REP_ERROR:
1799                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1800                 event.status = -ETIMEDOUT;
1801                 break;
1802         case IB_CM_REP_RECEIVED:
1803                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1804                     (id_priv->id.qp_type != IB_QPT_UD))
1805                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1806                 if (id_priv->id.qp) {
1807                         event.status = cma_rep_recv(id_priv);
1808                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1809                                                      RDMA_CM_EVENT_ESTABLISHED;
1810                 } else {
1811                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1812                 }
1813                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1814                                        ib_event->private_data);
1815                 break;
1816         case IB_CM_RTU_RECEIVED:
1817         case IB_CM_USER_ESTABLISHED:
1818                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1819                 break;
1820         case IB_CM_DREQ_ERROR:
1821                 event.status = -ETIMEDOUT; /* fall through */
1822         case IB_CM_DREQ_RECEIVED:
1823         case IB_CM_DREP_RECEIVED:
1824                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1825                                    RDMA_CM_DISCONNECT))
1826                         goto out;
1827                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1828                 break;
1829         case IB_CM_TIMEWAIT_EXIT:
1830                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1831                 break;
1832         case IB_CM_MRA_RECEIVED:
1833                 /* ignore event */
1834                 goto out;
1835         case IB_CM_REJ_RECEIVED:
1836                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1837                                                                                 ib_event->param.rej_rcvd.reason));
1838                 cma_modify_qp_err(id_priv);
1839                 event.status = ib_event->param.rej_rcvd.reason;
1840                 event.event = RDMA_CM_EVENT_REJECTED;
1841                 event.param.conn.private_data = ib_event->private_data;
1842                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1843                 break;
1844         default:
1845                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1846                        ib_event->event);
1847                 goto out;
1848         }
1849
1850         ret = id_priv->id.event_handler(&id_priv->id, &event);
1851         if (ret) {
1852                 /* Destroy the CM ID by returning a non-zero value. */
1853                 id_priv->cm_id.ib = NULL;
1854                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1855                 mutex_unlock(&id_priv->handler_mutex);
1856                 rdma_destroy_id(&id_priv->id);
1857                 return ret;
1858         }
1859 out:
1860         mutex_unlock(&id_priv->handler_mutex);
1861         return ret;
1862 }
1863
1864 static struct rdma_id_private *
1865 cma_ib_new_conn_id(const struct rdma_cm_id *listen_id,
1866                    const struct ib_cm_event *ib_event,
1867                    struct net_device *net_dev)
1868 {
1869         struct rdma_id_private *listen_id_priv;
1870         struct rdma_id_private *id_priv;
1871         struct rdma_cm_id *id;
1872         struct rdma_route *rt;
1873         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1874         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1875         const __be64 service_id =
1876                 ib_event->param.req_rcvd.primary_path->service_id;
1877         int ret;
1878
1879         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1880         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1881                             listen_id->event_handler, listen_id->context,
1882                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1883                             listen_id_priv->res.kern_name);
1884         if (IS_ERR(id))
1885                 return NULL;
1886
1887         id_priv = container_of(id, struct rdma_id_private, id);
1888         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1889                               (struct sockaddr *)&id->route.addr.dst_addr,
1890                               listen_id, ib_event, ss_family, service_id))
1891                 goto err;
1892
1893         rt = &id->route;
1894         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1895         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
1896                                      GFP_KERNEL);
1897         if (!rt->path_rec)
1898                 goto err;
1899
1900         rt->path_rec[0] = *path;
1901         if (rt->num_paths == 2)
1902                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1903
1904         if (net_dev) {
1905                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1906         } else {
1907                 if (!cma_protocol_roce(listen_id) &&
1908                     cma_any_addr(cma_src_addr(id_priv))) {
1909                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1910                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1911                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1912                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1913                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1914                         if (ret)
1915                                 goto err;
1916                 }
1917         }
1918         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1919
1920         id_priv->state = RDMA_CM_CONNECT;
1921         return id_priv;
1922
1923 err:
1924         rdma_destroy_id(id);
1925         return NULL;
1926 }
1927
1928 static struct rdma_id_private *
1929 cma_ib_new_udp_id(const struct rdma_cm_id *listen_id,
1930                   const struct ib_cm_event *ib_event,
1931                   struct net_device *net_dev)
1932 {
1933         const struct rdma_id_private *listen_id_priv;
1934         struct rdma_id_private *id_priv;
1935         struct rdma_cm_id *id;
1936         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1937         struct net *net = listen_id->route.addr.dev_addr.net;
1938         int ret;
1939
1940         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1941         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
1942                               listen_id->ps, IB_QPT_UD,
1943                               listen_id_priv->res.kern_name);
1944         if (IS_ERR(id))
1945                 return NULL;
1946
1947         id_priv = container_of(id, struct rdma_id_private, id);
1948         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1949                               (struct sockaddr *)&id->route.addr.dst_addr,
1950                               listen_id, ib_event, ss_family,
1951                               ib_event->param.sidr_req_rcvd.service_id))
1952                 goto err;
1953
1954         if (net_dev) {
1955                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1956         } else {
1957                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1958                         ret = cma_translate_addr(cma_src_addr(id_priv),
1959                                                  &id->route.addr.dev_addr);
1960                         if (ret)
1961                                 goto err;
1962                 }
1963         }
1964
1965         id_priv->state = RDMA_CM_CONNECT;
1966         return id_priv;
1967 err:
1968         rdma_destroy_id(id);
1969         return NULL;
1970 }
1971
1972 static void cma_set_req_event_data(struct rdma_cm_event *event,
1973                                    const struct ib_cm_req_event_param *req_data,
1974                                    void *private_data, int offset)
1975 {
1976         event->param.conn.private_data = private_data + offset;
1977         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1978         event->param.conn.responder_resources = req_data->responder_resources;
1979         event->param.conn.initiator_depth = req_data->initiator_depth;
1980         event->param.conn.flow_control = req_data->flow_control;
1981         event->param.conn.retry_count = req_data->retry_count;
1982         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1983         event->param.conn.srq = req_data->srq;
1984         event->param.conn.qp_num = req_data->remote_qpn;
1985 }
1986
1987 static int cma_ib_check_req_qp_type(const struct rdma_cm_id *id,
1988                                     const struct ib_cm_event *ib_event)
1989 {
1990         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1991                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1992                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1993                  (id->qp_type == IB_QPT_UD)) ||
1994                 (!id->qp_type));
1995 }
1996
1997 static int cma_ib_req_handler(struct ib_cm_id *cm_id,
1998                               const struct ib_cm_event *ib_event)
1999 {
2000         struct rdma_id_private *listen_id, *conn_id = NULL;
2001         struct rdma_cm_event event = {};
2002         struct net_device *net_dev;
2003         u8 offset;
2004         int ret;
2005
2006         listen_id = cma_ib_id_from_event(cm_id, ib_event, &net_dev);
2007         if (IS_ERR(listen_id))
2008                 return PTR_ERR(listen_id);
2009
2010         if (!cma_ib_check_req_qp_type(&listen_id->id, ib_event)) {
2011                 ret = -EINVAL;
2012                 goto net_dev_put;
2013         }
2014
2015         mutex_lock(&listen_id->handler_mutex);
2016         if (listen_id->state != RDMA_CM_LISTEN) {
2017                 ret = -ECONNABORTED;
2018                 goto err1;
2019         }
2020
2021         offset = cma_user_data_offset(listen_id);
2022         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2023         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
2024                 conn_id = cma_ib_new_udp_id(&listen_id->id, ib_event, net_dev);
2025                 event.param.ud.private_data = ib_event->private_data + offset;
2026                 event.param.ud.private_data_len =
2027                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
2028         } else {
2029                 conn_id = cma_ib_new_conn_id(&listen_id->id, ib_event, net_dev);
2030                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
2031                                        ib_event->private_data, offset);
2032         }
2033         if (!conn_id) {
2034                 ret = -ENOMEM;
2035                 goto err1;
2036         }
2037
2038         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2039         ret = cma_acquire_dev(conn_id, listen_id);
2040         if (ret)
2041                 goto err2;
2042
2043         conn_id->cm_id.ib = cm_id;
2044         cm_id->context = conn_id;
2045         cm_id->cm_handler = cma_ib_handler;
2046
2047         /*
2048          * Protect against the user destroying conn_id from another thread
2049          * until we're done accessing it.
2050          */
2051         atomic_inc(&conn_id->refcount);
2052         ret = conn_id->id.event_handler(&conn_id->id, &event);
2053         if (ret)
2054                 goto err3;
2055         /*
2056          * Acquire mutex to prevent user executing rdma_destroy_id()
2057          * while we're accessing the cm_id.
2058          */
2059         mutex_lock(&lock);
2060         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2061             (conn_id->id.qp_type != IB_QPT_UD))
2062                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2063         mutex_unlock(&lock);
2064         mutex_unlock(&conn_id->handler_mutex);
2065         mutex_unlock(&listen_id->handler_mutex);
2066         cma_deref_id(conn_id);
2067         if (net_dev)
2068                 dev_put(net_dev);
2069         return 0;
2070
2071 err3:
2072         cma_deref_id(conn_id);
2073         /* Destroy the CM ID by returning a non-zero value. */
2074         conn_id->cm_id.ib = NULL;
2075 err2:
2076         cma_exch(conn_id, RDMA_CM_DESTROYING);
2077         mutex_unlock(&conn_id->handler_mutex);
2078 err1:
2079         mutex_unlock(&listen_id->handler_mutex);
2080         if (conn_id)
2081                 rdma_destroy_id(&conn_id->id);
2082
2083 net_dev_put:
2084         if (net_dev)
2085                 dev_put(net_dev);
2086
2087         return ret;
2088 }
2089
2090 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2091 {
2092         if (addr->sa_family == AF_IB)
2093                 return ((struct sockaddr_ib *) addr)->sib_sid;
2094
2095         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2096 }
2097 EXPORT_SYMBOL(rdma_get_service_id);
2098
2099 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2100                     union ib_gid *dgid)
2101 {
2102         struct rdma_addr *addr = &cm_id->route.addr;
2103
2104         if (!cm_id->device) {
2105                 if (sgid)
2106                         memset(sgid, 0, sizeof(*sgid));
2107                 if (dgid)
2108                         memset(dgid, 0, sizeof(*dgid));
2109                 return;
2110         }
2111
2112         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2113                 if (sgid)
2114                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2115                 if (dgid)
2116                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2117         } else {
2118                 if (sgid)
2119                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2120                 if (dgid)
2121                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2122         }
2123 }
2124 EXPORT_SYMBOL(rdma_read_gids);
2125
2126 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2127 {
2128         struct rdma_id_private *id_priv = iw_id->context;
2129         struct rdma_cm_event event = {};
2130         int ret = 0;
2131         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2132         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2133
2134         mutex_lock(&id_priv->handler_mutex);
2135         if (id_priv->state != RDMA_CM_CONNECT)
2136                 goto out;
2137
2138         switch (iw_event->event) {
2139         case IW_CM_EVENT_CLOSE:
2140                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2141                 break;
2142         case IW_CM_EVENT_CONNECT_REPLY:
2143                 memcpy(cma_src_addr(id_priv), laddr,
2144                        rdma_addr_size(laddr));
2145                 memcpy(cma_dst_addr(id_priv), raddr,
2146                        rdma_addr_size(raddr));
2147                 switch (iw_event->status) {
2148                 case 0:
2149                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2150                         event.param.conn.initiator_depth = iw_event->ird;
2151                         event.param.conn.responder_resources = iw_event->ord;
2152                         break;
2153                 case -ECONNRESET:
2154                 case -ECONNREFUSED:
2155                         event.event = RDMA_CM_EVENT_REJECTED;
2156                         break;
2157                 case -ETIMEDOUT:
2158                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2159                         break;
2160                 default:
2161                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2162                         break;
2163                 }
2164                 break;
2165         case IW_CM_EVENT_ESTABLISHED:
2166                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2167                 event.param.conn.initiator_depth = iw_event->ird;
2168                 event.param.conn.responder_resources = iw_event->ord;
2169                 break;
2170         default:
2171                 goto out;
2172         }
2173
2174         event.status = iw_event->status;
2175         event.param.conn.private_data = iw_event->private_data;
2176         event.param.conn.private_data_len = iw_event->private_data_len;
2177         ret = id_priv->id.event_handler(&id_priv->id, &event);
2178         if (ret) {
2179                 /* Destroy the CM ID by returning a non-zero value. */
2180                 id_priv->cm_id.iw = NULL;
2181                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2182                 mutex_unlock(&id_priv->handler_mutex);
2183                 rdma_destroy_id(&id_priv->id);
2184                 return ret;
2185         }
2186
2187 out:
2188         mutex_unlock(&id_priv->handler_mutex);
2189         return ret;
2190 }
2191
2192 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2193                                struct iw_cm_event *iw_event)
2194 {
2195         struct rdma_cm_id *new_cm_id;
2196         struct rdma_id_private *listen_id, *conn_id;
2197         struct rdma_cm_event event = {};
2198         int ret = -ECONNABORTED;
2199         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2200         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2201
2202         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2203         event.param.conn.private_data = iw_event->private_data;
2204         event.param.conn.private_data_len = iw_event->private_data_len;
2205         event.param.conn.initiator_depth = iw_event->ird;
2206         event.param.conn.responder_resources = iw_event->ord;
2207
2208         listen_id = cm_id->context;
2209
2210         mutex_lock(&listen_id->handler_mutex);
2211         if (listen_id->state != RDMA_CM_LISTEN)
2212                 goto out;
2213
2214         /* Create a new RDMA id for the new IW CM ID */
2215         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2216                                      listen_id->id.event_handler,
2217                                      listen_id->id.context,
2218                                      RDMA_PS_TCP, IB_QPT_RC,
2219                                      listen_id->res.kern_name);
2220         if (IS_ERR(new_cm_id)) {
2221                 ret = -ENOMEM;
2222                 goto out;
2223         }
2224         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2225         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2226         conn_id->state = RDMA_CM_CONNECT;
2227
2228         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2229         if (ret) {
2230                 mutex_unlock(&conn_id->handler_mutex);
2231                 rdma_destroy_id(new_cm_id);
2232                 goto out;
2233         }
2234
2235         ret = cma_acquire_dev(conn_id, listen_id);
2236         if (ret) {
2237                 mutex_unlock(&conn_id->handler_mutex);
2238                 rdma_destroy_id(new_cm_id);
2239                 goto out;
2240         }
2241
2242         conn_id->cm_id.iw = cm_id;
2243         cm_id->context = conn_id;
2244         cm_id->cm_handler = cma_iw_handler;
2245
2246         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2247         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2248
2249         /*
2250          * Protect against the user destroying conn_id from another thread
2251          * until we're done accessing it.
2252          */
2253         atomic_inc(&conn_id->refcount);
2254         ret = conn_id->id.event_handler(&conn_id->id, &event);
2255         if (ret) {
2256                 /* User wants to destroy the CM ID */
2257                 conn_id->cm_id.iw = NULL;
2258                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2259                 mutex_unlock(&conn_id->handler_mutex);
2260                 cma_deref_id(conn_id);
2261                 rdma_destroy_id(&conn_id->id);
2262                 goto out;
2263         }
2264
2265         mutex_unlock(&conn_id->handler_mutex);
2266         cma_deref_id(conn_id);
2267
2268 out:
2269         mutex_unlock(&listen_id->handler_mutex);
2270         return ret;
2271 }
2272
2273 static int cma_ib_listen(struct rdma_id_private *id_priv)
2274 {
2275         struct sockaddr *addr;
2276         struct ib_cm_id *id;
2277         __be64 svc_id;
2278
2279         addr = cma_src_addr(id_priv);
2280         svc_id = rdma_get_service_id(&id_priv->id, addr);
2281         id = ib_cm_insert_listen(id_priv->id.device,
2282                                  cma_ib_req_handler, svc_id);
2283         if (IS_ERR(id))
2284                 return PTR_ERR(id);
2285         id_priv->cm_id.ib = id;
2286
2287         return 0;
2288 }
2289
2290 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2291 {
2292         int ret;
2293         struct iw_cm_id *id;
2294
2295         id = iw_create_cm_id(id_priv->id.device,
2296                              iw_conn_req_handler,
2297                              id_priv);
2298         if (IS_ERR(id))
2299                 return PTR_ERR(id);
2300
2301         id->tos = id_priv->tos;
2302         id_priv->cm_id.iw = id;
2303
2304         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2305                rdma_addr_size(cma_src_addr(id_priv)));
2306
2307         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2308
2309         if (ret) {
2310                 iw_destroy_cm_id(id_priv->cm_id.iw);
2311                 id_priv->cm_id.iw = NULL;
2312         }
2313
2314         return ret;
2315 }
2316
2317 static int cma_listen_handler(struct rdma_cm_id *id,
2318                               struct rdma_cm_event *event)
2319 {
2320         struct rdma_id_private *id_priv = id->context;
2321
2322         id->context = id_priv->id.context;
2323         id->event_handler = id_priv->id.event_handler;
2324         return id_priv->id.event_handler(id, event);
2325 }
2326
2327 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2328                               struct cma_device *cma_dev)
2329 {
2330         struct rdma_id_private *dev_id_priv;
2331         struct rdma_cm_id *id;
2332         struct net *net = id_priv->id.route.addr.dev_addr.net;
2333         int ret;
2334
2335         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2336                 return;
2337
2338         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2339                               id_priv->id.qp_type, id_priv->res.kern_name);
2340         if (IS_ERR(id))
2341                 return;
2342
2343         dev_id_priv = container_of(id, struct rdma_id_private, id);
2344
2345         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2346         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2347                rdma_addr_size(cma_src_addr(id_priv)));
2348
2349         _cma_attach_to_dev(dev_id_priv, cma_dev);
2350         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2351         atomic_inc(&id_priv->refcount);
2352         dev_id_priv->internal_id = 1;
2353         dev_id_priv->afonly = id_priv->afonly;
2354
2355         ret = rdma_listen(id, id_priv->backlog);
2356         if (ret)
2357                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2358                         ret, cma_dev->device->name);
2359 }
2360
2361 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2362 {
2363         struct cma_device *cma_dev;
2364
2365         mutex_lock(&lock);
2366         list_add_tail(&id_priv->list, &listen_any_list);
2367         list_for_each_entry(cma_dev, &dev_list, list)
2368                 cma_listen_on_dev(id_priv, cma_dev);
2369         mutex_unlock(&lock);
2370 }
2371
2372 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2373 {
2374         struct rdma_id_private *id_priv;
2375
2376         id_priv = container_of(id, struct rdma_id_private, id);
2377         id_priv->tos = (u8) tos;
2378         id_priv->tos_set = true;
2379 }
2380 EXPORT_SYMBOL(rdma_set_service_type);
2381
2382 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2383                               void *context)
2384 {
2385         struct cma_work *work = context;
2386         struct rdma_route *route;
2387
2388         route = &work->id->id.route;
2389
2390         if (!status) {
2391                 route->num_paths = 1;
2392                 *route->path_rec = *path_rec;
2393         } else {
2394                 work->old_state = RDMA_CM_ROUTE_QUERY;
2395                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2396                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2397                 work->event.status = status;
2398                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2399                                      status);
2400         }
2401
2402         queue_work(cma_wq, &work->work);
2403 }
2404
2405 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2406                               struct cma_work *work)
2407 {
2408         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2409         struct sa_path_rec path_rec;
2410         ib_sa_comp_mask comp_mask;
2411         struct sockaddr_in6 *sin6;
2412         struct sockaddr_ib *sib;
2413
2414         memset(&path_rec, 0, sizeof path_rec);
2415
2416         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2417                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2418         else
2419                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2420         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2421         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2422         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2423         path_rec.numb_path = 1;
2424         path_rec.reversible = 1;
2425         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2426                                                   cma_dst_addr(id_priv));
2427
2428         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2429                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2430                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2431
2432         switch (cma_family(id_priv)) {
2433         case AF_INET:
2434                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2435                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2436                 break;
2437         case AF_INET6:
2438                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2439                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2440                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2441                 break;
2442         case AF_IB:
2443                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2444                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2445                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2446                 break;
2447         }
2448
2449         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2450                                                id_priv->id.port_num, &path_rec,
2451                                                comp_mask, timeout_ms,
2452                                                GFP_KERNEL, cma_query_handler,
2453                                                work, &id_priv->query);
2454
2455         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2456 }
2457
2458 static void cma_work_handler(struct work_struct *_work)
2459 {
2460         struct cma_work *work = container_of(_work, struct cma_work, work);
2461         struct rdma_id_private *id_priv = work->id;
2462         int destroy = 0;
2463
2464         mutex_lock(&id_priv->handler_mutex);
2465         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2466                 goto out;
2467
2468         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2469                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2470                 destroy = 1;
2471         }
2472 out:
2473         mutex_unlock(&id_priv->handler_mutex);
2474         cma_deref_id(id_priv);
2475         if (destroy)
2476                 rdma_destroy_id(&id_priv->id);
2477         kfree(work);
2478 }
2479
2480 static void cma_ndev_work_handler(struct work_struct *_work)
2481 {
2482         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2483         struct rdma_id_private *id_priv = work->id;
2484         int destroy = 0;
2485
2486         mutex_lock(&id_priv->handler_mutex);
2487         if (id_priv->state == RDMA_CM_DESTROYING ||
2488             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2489                 goto out;
2490
2491         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2492                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2493                 destroy = 1;
2494         }
2495
2496 out:
2497         mutex_unlock(&id_priv->handler_mutex);
2498         cma_deref_id(id_priv);
2499         if (destroy)
2500                 rdma_destroy_id(&id_priv->id);
2501         kfree(work);
2502 }
2503
2504 static void cma_init_resolve_route_work(struct cma_work *work,
2505                                         struct rdma_id_private *id_priv)
2506 {
2507         work->id = id_priv;
2508         INIT_WORK(&work->work, cma_work_handler);
2509         work->old_state = RDMA_CM_ROUTE_QUERY;
2510         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2511         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2512 }
2513
2514 static void cma_init_resolve_addr_work(struct cma_work *work,
2515                                        struct rdma_id_private *id_priv)
2516 {
2517         work->id = id_priv;
2518         INIT_WORK(&work->work, cma_work_handler);
2519         work->old_state = RDMA_CM_ADDR_QUERY;
2520         work->new_state = RDMA_CM_ADDR_RESOLVED;
2521         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2522 }
2523
2524 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2525 {
2526         struct rdma_route *route = &id_priv->id.route;
2527         struct cma_work *work;
2528         int ret;
2529
2530         work = kzalloc(sizeof *work, GFP_KERNEL);
2531         if (!work)
2532                 return -ENOMEM;
2533
2534         cma_init_resolve_route_work(work, id_priv);
2535
2536         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2537         if (!route->path_rec) {
2538                 ret = -ENOMEM;
2539                 goto err1;
2540         }
2541
2542         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2543         if (ret)
2544                 goto err2;
2545
2546         return 0;
2547 err2:
2548         kfree(route->path_rec);
2549         route->path_rec = NULL;
2550 err1:
2551         kfree(work);
2552         return ret;
2553 }
2554
2555 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2556                                            unsigned long supported_gids,
2557                                            enum ib_gid_type default_gid)
2558 {
2559         if ((network_type == RDMA_NETWORK_IPV4 ||
2560              network_type == RDMA_NETWORK_IPV6) &&
2561             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2562                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2563
2564         return default_gid;
2565 }
2566
2567 /*
2568  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2569  * path record type based on GID type.
2570  * It also sets up other L2 fields which includes destination mac address
2571  * netdev ifindex, of the path record.
2572  * It returns the netdev of the bound interface for this path record entry.
2573  */
2574 static struct net_device *
2575 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2576 {
2577         struct rdma_route *route = &id_priv->id.route;
2578         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2579         struct rdma_addr *addr = &route->addr;
2580         unsigned long supported_gids;
2581         struct net_device *ndev;
2582
2583         if (!addr->dev_addr.bound_dev_if)
2584                 return NULL;
2585
2586         ndev = dev_get_by_index(addr->dev_addr.net,
2587                                 addr->dev_addr.bound_dev_if);
2588         if (!ndev)
2589                 return NULL;
2590
2591         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2592                                                     id_priv->id.port_num);
2593         gid_type = cma_route_gid_type(addr->dev_addr.network,
2594                                       supported_gids,
2595                                       id_priv->gid_type);
2596         /* Use the hint from IP Stack to select GID Type */
2597         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2598                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2599         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2600
2601         route->path_rec->roce.route_resolved = true;
2602         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2603         return ndev;
2604 }
2605
2606 int rdma_set_ib_path(struct rdma_cm_id *id,
2607                      struct sa_path_rec *path_rec)
2608 {
2609         struct rdma_id_private *id_priv;
2610         struct net_device *ndev;
2611         int ret;
2612
2613         id_priv = container_of(id, struct rdma_id_private, id);
2614         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2615                            RDMA_CM_ROUTE_RESOLVED))
2616                 return -EINVAL;
2617
2618         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2619                                      GFP_KERNEL);
2620         if (!id->route.path_rec) {
2621                 ret = -ENOMEM;
2622                 goto err;
2623         }
2624
2625         if (rdma_protocol_roce(id->device, id->port_num)) {
2626                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2627                 if (!ndev) {
2628                         ret = -ENODEV;
2629                         goto err_free;
2630                 }
2631                 dev_put(ndev);
2632         }
2633
2634         id->route.num_paths = 1;
2635         return 0;
2636
2637 err_free:
2638         kfree(id->route.path_rec);
2639         id->route.path_rec = NULL;
2640 err:
2641         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2642         return ret;
2643 }
2644 EXPORT_SYMBOL(rdma_set_ib_path);
2645
2646 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2647 {
2648         struct cma_work *work;
2649
2650         work = kzalloc(sizeof *work, GFP_KERNEL);
2651         if (!work)
2652                 return -ENOMEM;
2653
2654         cma_init_resolve_route_work(work, id_priv);
2655         queue_work(cma_wq, &work->work);
2656         return 0;
2657 }
2658
2659 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2660 {
2661         int prio;
2662         struct net_device *dev;
2663
2664         prio = rt_tos2priority(tos);
2665         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2666         if (dev->num_tc)
2667                 return netdev_get_prio_tc_map(dev, prio);
2668
2669 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2670         if (is_vlan_dev(ndev))
2671                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2672                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2673 #endif
2674         return 0;
2675 }
2676
2677 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2678 {
2679         struct rdma_route *route = &id_priv->id.route;
2680         struct rdma_addr *addr = &route->addr;
2681         struct cma_work *work;
2682         int ret;
2683         struct net_device *ndev;
2684
2685         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2686                                         rdma_start_port(id_priv->cma_dev->device)];
2687         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2688
2689
2690         work = kzalloc(sizeof *work, GFP_KERNEL);
2691         if (!work)
2692                 return -ENOMEM;
2693
2694         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2695         if (!route->path_rec) {
2696                 ret = -ENOMEM;
2697                 goto err1;
2698         }
2699
2700         route->num_paths = 1;
2701
2702         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2703         if (!ndev) {
2704                 ret = -ENODEV;
2705                 goto err2;
2706         }
2707
2708         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2709                     &route->path_rec->sgid);
2710         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2711                     &route->path_rec->dgid);
2712
2713         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2714                 /* TODO: get the hoplimit from the inet/inet6 device */
2715                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2716         else
2717                 route->path_rec->hop_limit = 1;
2718         route->path_rec->reversible = 1;
2719         route->path_rec->pkey = cpu_to_be16(0xffff);
2720         route->path_rec->mtu_selector = IB_SA_EQ;
2721         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2722         route->path_rec->traffic_class = tos;
2723         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2724         route->path_rec->rate_selector = IB_SA_EQ;
2725         route->path_rec->rate = iboe_get_rate(ndev);
2726         dev_put(ndev);
2727         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2728         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2729         if (!route->path_rec->mtu) {
2730                 ret = -EINVAL;
2731                 goto err2;
2732         }
2733
2734         cma_init_resolve_route_work(work, id_priv);
2735         queue_work(cma_wq, &work->work);
2736
2737         return 0;
2738
2739 err2:
2740         kfree(route->path_rec);
2741         route->path_rec = NULL;
2742 err1:
2743         kfree(work);
2744         return ret;
2745 }
2746
2747 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2748 {
2749         struct rdma_id_private *id_priv;
2750         int ret;
2751
2752         id_priv = container_of(id, struct rdma_id_private, id);
2753         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2754                 return -EINVAL;
2755
2756         atomic_inc(&id_priv->refcount);
2757         if (rdma_cap_ib_sa(id->device, id->port_num))
2758                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2759         else if (rdma_protocol_roce(id->device, id->port_num))
2760                 ret = cma_resolve_iboe_route(id_priv);
2761         else if (rdma_protocol_iwarp(id->device, id->port_num))
2762                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2763         else
2764                 ret = -ENOSYS;
2765
2766         if (ret)
2767                 goto err;
2768
2769         return 0;
2770 err:
2771         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2772         cma_deref_id(id_priv);
2773         return ret;
2774 }
2775 EXPORT_SYMBOL(rdma_resolve_route);
2776
2777 static void cma_set_loopback(struct sockaddr *addr)
2778 {
2779         switch (addr->sa_family) {
2780         case AF_INET:
2781                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2782                 break;
2783         case AF_INET6:
2784                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2785                               0, 0, 0, htonl(1));
2786                 break;
2787         default:
2788                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2789                             0, 0, 0, htonl(1));
2790                 break;
2791         }
2792 }
2793
2794 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2795 {
2796         struct cma_device *cma_dev, *cur_dev;
2797         union ib_gid gid;
2798         enum ib_port_state port_state;
2799         u16 pkey;
2800         int ret;
2801         u8 p;
2802
2803         cma_dev = NULL;
2804         mutex_lock(&lock);
2805         list_for_each_entry(cur_dev, &dev_list, list) {
2806                 if (cma_family(id_priv) == AF_IB &&
2807                     !rdma_cap_ib_cm(cur_dev->device, 1))
2808                         continue;
2809
2810                 if (!cma_dev)
2811                         cma_dev = cur_dev;
2812
2813                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2814                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2815                             port_state == IB_PORT_ACTIVE) {
2816                                 cma_dev = cur_dev;
2817                                 goto port_found;
2818                         }
2819                 }
2820         }
2821
2822         if (!cma_dev) {
2823                 ret = -ENODEV;
2824                 goto out;
2825         }
2826
2827         p = 1;
2828
2829 port_found:
2830         ret = rdma_query_gid(cma_dev->device, p, 0, &gid);
2831         if (ret)
2832                 goto out;
2833
2834         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2835         if (ret)
2836                 goto out;
2837
2838         id_priv->id.route.addr.dev_addr.dev_type =
2839                 (rdma_protocol_ib(cma_dev->device, p)) ?
2840                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2841
2842         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2843         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2844         id_priv->id.port_num = p;
2845         cma_attach_to_dev(id_priv, cma_dev);
2846         cma_set_loopback(cma_src_addr(id_priv));
2847 out:
2848         mutex_unlock(&lock);
2849         return ret;
2850 }
2851
2852 static void addr_handler(int status, struct sockaddr *src_addr,
2853                          struct rdma_dev_addr *dev_addr, void *context)
2854 {
2855         struct rdma_id_private *id_priv = context;
2856         struct rdma_cm_event event = {};
2857
2858         mutex_lock(&id_priv->handler_mutex);
2859         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2860                            RDMA_CM_ADDR_RESOLVED))
2861                 goto out;
2862
2863         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2864         if (!status && !id_priv->cma_dev) {
2865                 status = cma_acquire_dev(id_priv, NULL);
2866                 if (status)
2867                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2868                                              status);
2869         } else {
2870                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2871         }
2872
2873         if (status) {
2874                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2875                                    RDMA_CM_ADDR_BOUND))
2876                         goto out;
2877                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2878                 event.status = status;
2879         } else
2880                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2881
2882         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2883                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2884                 mutex_unlock(&id_priv->handler_mutex);
2885                 cma_deref_id(id_priv);
2886                 rdma_destroy_id(&id_priv->id);
2887                 return;
2888         }
2889 out:
2890         mutex_unlock(&id_priv->handler_mutex);
2891         cma_deref_id(id_priv);
2892 }
2893
2894 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2895 {
2896         struct cma_work *work;
2897         union ib_gid gid;
2898         int ret;
2899
2900         work = kzalloc(sizeof *work, GFP_KERNEL);
2901         if (!work)
2902                 return -ENOMEM;
2903
2904         if (!id_priv->cma_dev) {
2905                 ret = cma_bind_loopback(id_priv);
2906                 if (ret)
2907                         goto err;
2908         }
2909
2910         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2911         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2912
2913         cma_init_resolve_addr_work(work, id_priv);
2914         queue_work(cma_wq, &work->work);
2915         return 0;
2916 err:
2917         kfree(work);
2918         return ret;
2919 }
2920
2921 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2922 {
2923         struct cma_work *work;
2924         int ret;
2925
2926         work = kzalloc(sizeof *work, GFP_KERNEL);
2927         if (!work)
2928                 return -ENOMEM;
2929
2930         if (!id_priv->cma_dev) {
2931                 ret = cma_resolve_ib_dev(id_priv);
2932                 if (ret)
2933                         goto err;
2934         }
2935
2936         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2937                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2938
2939         cma_init_resolve_addr_work(work, id_priv);
2940         queue_work(cma_wq, &work->work);
2941         return 0;
2942 err:
2943         kfree(work);
2944         return ret;
2945 }
2946
2947 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2948                          const struct sockaddr *dst_addr)
2949 {
2950         if (!src_addr || !src_addr->sa_family) {
2951                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2952                 src_addr->sa_family = dst_addr->sa_family;
2953                 if (IS_ENABLED(CONFIG_IPV6) &&
2954                     dst_addr->sa_family == AF_INET6) {
2955                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2956                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2957                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2958                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2959                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2960                 } else if (dst_addr->sa_family == AF_IB) {
2961                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2962                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2963                 }
2964         }
2965         return rdma_bind_addr(id, src_addr);
2966 }
2967
2968 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2969                       const struct sockaddr *dst_addr, int timeout_ms)
2970 {
2971         struct rdma_id_private *id_priv;
2972         int ret;
2973
2974         id_priv = container_of(id, struct rdma_id_private, id);
2975         if (id_priv->state == RDMA_CM_IDLE) {
2976                 ret = cma_bind_addr(id, src_addr, dst_addr);
2977                 if (ret)
2978                         return ret;
2979         }
2980
2981         if (cma_family(id_priv) != dst_addr->sa_family)
2982                 return -EINVAL;
2983
2984         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2985                 return -EINVAL;
2986
2987         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2988         atomic_inc(&id_priv->refcount);
2989         if (cma_any_addr(dst_addr)) {
2990                 ret = cma_resolve_loopback(id_priv);
2991         } else {
2992                 if (dst_addr->sa_family == AF_IB) {
2993                         ret = cma_resolve_ib_addr(id_priv);
2994                 } else {
2995                         ret = rdma_resolve_ip(cma_src_addr(id_priv),
2996                                               dst_addr, &id->route.addr.dev_addr,
2997                                               timeout_ms, addr_handler, id_priv);
2998                 }
2999         }
3000         if (ret)
3001                 goto err;
3002
3003         return 0;
3004 err:
3005         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
3006         cma_deref_id(id_priv);
3007         return ret;
3008 }
3009 EXPORT_SYMBOL(rdma_resolve_addr);
3010
3011 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
3012 {
3013         struct rdma_id_private *id_priv;
3014         unsigned long flags;
3015         int ret;
3016
3017         id_priv = container_of(id, struct rdma_id_private, id);
3018         spin_lock_irqsave(&id_priv->lock, flags);
3019         if (reuse || id_priv->state == RDMA_CM_IDLE) {
3020                 id_priv->reuseaddr = reuse;
3021                 ret = 0;
3022         } else {
3023                 ret = -EINVAL;
3024         }
3025         spin_unlock_irqrestore(&id_priv->lock, flags);
3026         return ret;
3027 }
3028 EXPORT_SYMBOL(rdma_set_reuseaddr);
3029
3030 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3031 {
3032         struct rdma_id_private *id_priv;
3033         unsigned long flags;
3034         int ret;
3035
3036         id_priv = container_of(id, struct rdma_id_private, id);
3037         spin_lock_irqsave(&id_priv->lock, flags);
3038         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3039                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3040                 id_priv->afonly = afonly;
3041                 ret = 0;
3042         } else {
3043                 ret = -EINVAL;
3044         }
3045         spin_unlock_irqrestore(&id_priv->lock, flags);
3046         return ret;
3047 }
3048 EXPORT_SYMBOL(rdma_set_afonly);
3049
3050 static void cma_bind_port(struct rdma_bind_list *bind_list,
3051                           struct rdma_id_private *id_priv)
3052 {
3053         struct sockaddr *addr;
3054         struct sockaddr_ib *sib;
3055         u64 sid, mask;
3056         __be16 port;
3057
3058         addr = cma_src_addr(id_priv);
3059         port = htons(bind_list->port);
3060
3061         switch (addr->sa_family) {
3062         case AF_INET:
3063                 ((struct sockaddr_in *) addr)->sin_port = port;
3064                 break;
3065         case AF_INET6:
3066                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3067                 break;
3068         case AF_IB:
3069                 sib = (struct sockaddr_ib *) addr;
3070                 sid = be64_to_cpu(sib->sib_sid);
3071                 mask = be64_to_cpu(sib->sib_sid_mask);
3072                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3073                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3074                 break;
3075         }
3076         id_priv->bind_list = bind_list;
3077         hlist_add_head(&id_priv->node, &bind_list->owners);
3078 }
3079
3080 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3081                           struct rdma_id_private *id_priv, unsigned short snum)
3082 {
3083         struct rdma_bind_list *bind_list;
3084         int ret;
3085
3086         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3087         if (!bind_list)
3088                 return -ENOMEM;
3089
3090         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3091                            snum);
3092         if (ret < 0)
3093                 goto err;
3094
3095         bind_list->ps = ps;
3096         bind_list->port = (unsigned short)ret;
3097         cma_bind_port(bind_list, id_priv);
3098         return 0;
3099 err:
3100         kfree(bind_list);
3101         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3102 }
3103
3104 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3105                               struct rdma_id_private *id_priv)
3106 {
3107         struct rdma_id_private *cur_id;
3108         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3109         struct sockaddr  *saddr = cma_src_addr(id_priv);
3110         __be16 dport = cma_port(daddr);
3111
3112         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3113                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3114                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3115                 __be16 cur_dport = cma_port(cur_daddr);
3116
3117                 if (id_priv == cur_id)
3118                         continue;
3119
3120                 /* different dest port -> unique */
3121                 if (!cma_any_port(daddr) &&
3122                     !cma_any_port(cur_daddr) &&
3123                     (dport != cur_dport))
3124                         continue;
3125
3126                 /* different src address -> unique */
3127                 if (!cma_any_addr(saddr) &&
3128                     !cma_any_addr(cur_saddr) &&
3129                     cma_addr_cmp(saddr, cur_saddr))
3130                         continue;
3131
3132                 /* different dst address -> unique */
3133                 if (!cma_any_addr(daddr) &&
3134                     !cma_any_addr(cur_daddr) &&
3135                     cma_addr_cmp(daddr, cur_daddr))
3136                         continue;
3137
3138                 return -EADDRNOTAVAIL;
3139         }
3140         return 0;
3141 }
3142
3143 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3144                               struct rdma_id_private *id_priv)
3145 {
3146         static unsigned int last_used_port;
3147         int low, high, remaining;
3148         unsigned int rover;
3149         struct net *net = id_priv->id.route.addr.dev_addr.net;
3150
3151         inet_get_local_port_range(net, &low, &high);
3152         remaining = (high - low) + 1;
3153         rover = prandom_u32() % remaining + low;
3154 retry:
3155         if (last_used_port != rover) {
3156                 struct rdma_bind_list *bind_list;
3157                 int ret;
3158
3159                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3160
3161                 if (!bind_list) {
3162                         ret = cma_alloc_port(ps, id_priv, rover);
3163                 } else {
3164                         ret = cma_port_is_unique(bind_list, id_priv);
3165                         if (!ret)
3166                                 cma_bind_port(bind_list, id_priv);
3167                 }
3168                 /*
3169                  * Remember previously used port number in order to avoid
3170                  * re-using same port immediately after it is closed.
3171                  */
3172                 if (!ret)
3173                         last_used_port = rover;
3174                 if (ret != -EADDRNOTAVAIL)
3175                         return ret;
3176         }
3177         if (--remaining) {
3178                 rover++;
3179                 if ((rover < low) || (rover > high))
3180                         rover = low;
3181                 goto retry;
3182         }
3183         return -EADDRNOTAVAIL;
3184 }
3185
3186 /*
3187  * Check that the requested port is available.  This is called when trying to
3188  * bind to a specific port, or when trying to listen on a bound port.  In
3189  * the latter case, the provided id_priv may already be on the bind_list, but
3190  * we still need to check that it's okay to start listening.
3191  */
3192 static int cma_check_port(struct rdma_bind_list *bind_list,
3193                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3194 {
3195         struct rdma_id_private *cur_id;
3196         struct sockaddr *addr, *cur_addr;
3197
3198         addr = cma_src_addr(id_priv);
3199         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3200                 if (id_priv == cur_id)
3201                         continue;
3202
3203                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3204                     cur_id->reuseaddr)
3205                         continue;
3206
3207                 cur_addr = cma_src_addr(cur_id);
3208                 if (id_priv->afonly && cur_id->afonly &&
3209                     (addr->sa_family != cur_addr->sa_family))
3210                         continue;
3211
3212                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3213                         return -EADDRNOTAVAIL;
3214
3215                 if (!cma_addr_cmp(addr, cur_addr))
3216                         return -EADDRINUSE;
3217         }
3218         return 0;
3219 }
3220
3221 static int cma_use_port(enum rdma_ucm_port_space ps,
3222                         struct rdma_id_private *id_priv)
3223 {
3224         struct rdma_bind_list *bind_list;
3225         unsigned short snum;
3226         int ret;
3227
3228         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3229         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3230                 return -EACCES;
3231
3232         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3233         if (!bind_list) {
3234                 ret = cma_alloc_port(ps, id_priv, snum);
3235         } else {
3236                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3237                 if (!ret)
3238                         cma_bind_port(bind_list, id_priv);
3239         }
3240         return ret;
3241 }
3242
3243 static int cma_bind_listen(struct rdma_id_private *id_priv)
3244 {
3245         struct rdma_bind_list *bind_list = id_priv->bind_list;
3246         int ret = 0;
3247
3248         mutex_lock(&lock);
3249         if (bind_list->owners.first->next)
3250                 ret = cma_check_port(bind_list, id_priv, 0);
3251         mutex_unlock(&lock);
3252         return ret;
3253 }
3254
3255 static enum rdma_ucm_port_space
3256 cma_select_inet_ps(struct rdma_id_private *id_priv)
3257 {
3258         switch (id_priv->id.ps) {
3259         case RDMA_PS_TCP:
3260         case RDMA_PS_UDP:
3261         case RDMA_PS_IPOIB:
3262         case RDMA_PS_IB:
3263                 return id_priv->id.ps;
3264         default:
3265
3266                 return 0;
3267         }
3268 }
3269
3270 static enum rdma_ucm_port_space
3271 cma_select_ib_ps(struct rdma_id_private *id_priv)
3272 {
3273         enum rdma_ucm_port_space ps = 0;
3274         struct sockaddr_ib *sib;
3275         u64 sid_ps, mask, sid;
3276
3277         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3278         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3279         sid = be64_to_cpu(sib->sib_sid) & mask;
3280
3281         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3282                 sid_ps = RDMA_IB_IP_PS_IB;
3283                 ps = RDMA_PS_IB;
3284         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3285                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3286                 sid_ps = RDMA_IB_IP_PS_TCP;
3287                 ps = RDMA_PS_TCP;
3288         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3289                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3290                 sid_ps = RDMA_IB_IP_PS_UDP;
3291                 ps = RDMA_PS_UDP;
3292         }
3293
3294         if (ps) {
3295                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3296                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3297                                                 be64_to_cpu(sib->sib_sid_mask));
3298         }
3299         return ps;
3300 }
3301
3302 static int cma_get_port(struct rdma_id_private *id_priv)
3303 {
3304         enum rdma_ucm_port_space ps;
3305         int ret;
3306
3307         if (cma_family(id_priv) != AF_IB)
3308                 ps = cma_select_inet_ps(id_priv);
3309         else
3310                 ps = cma_select_ib_ps(id_priv);
3311         if (!ps)
3312                 return -EPROTONOSUPPORT;
3313
3314         mutex_lock(&lock);
3315         if (cma_any_port(cma_src_addr(id_priv)))
3316                 ret = cma_alloc_any_port(ps, id_priv);
3317         else
3318                 ret = cma_use_port(ps, id_priv);
3319         mutex_unlock(&lock);
3320
3321         return ret;
3322 }
3323
3324 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3325                                struct sockaddr *addr)
3326 {
3327 #if IS_ENABLED(CONFIG_IPV6)
3328         struct sockaddr_in6 *sin6;
3329
3330         if (addr->sa_family != AF_INET6)
3331                 return 0;
3332
3333         sin6 = (struct sockaddr_in6 *) addr;
3334
3335         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3336                 return 0;
3337
3338         if (!sin6->sin6_scope_id)
3339                         return -EINVAL;
3340
3341         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3342 #endif
3343         return 0;
3344 }
3345
3346 int rdma_listen(struct rdma_cm_id *id, int backlog)
3347 {
3348         struct rdma_id_private *id_priv;
3349         int ret;
3350
3351         id_priv = container_of(id, struct rdma_id_private, id);
3352         if (id_priv->state == RDMA_CM_IDLE) {
3353                 id->route.addr.src_addr.ss_family = AF_INET;
3354                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3355                 if (ret)
3356                         return ret;
3357         }
3358
3359         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3360                 return -EINVAL;
3361
3362         if (id_priv->reuseaddr) {
3363                 ret = cma_bind_listen(id_priv);
3364                 if (ret)
3365                         goto err;
3366         }
3367
3368         id_priv->backlog = backlog;
3369         if (id->device) {
3370                 if (rdma_cap_ib_cm(id->device, 1)) {
3371                         ret = cma_ib_listen(id_priv);
3372                         if (ret)
3373                                 goto err;
3374                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3375                         ret = cma_iw_listen(id_priv, backlog);
3376                         if (ret)
3377                                 goto err;
3378                 } else {
3379                         ret = -ENOSYS;
3380                         goto err;
3381                 }
3382         } else
3383                 cma_listen_on_all(id_priv);
3384
3385         return 0;
3386 err:
3387         id_priv->backlog = 0;
3388         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3389         return ret;
3390 }
3391 EXPORT_SYMBOL(rdma_listen);
3392
3393 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3394 {
3395         struct rdma_id_private *id_priv;
3396         int ret;
3397         struct sockaddr  *daddr;
3398
3399         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3400             addr->sa_family != AF_IB)
3401                 return -EAFNOSUPPORT;
3402
3403         id_priv = container_of(id, struct rdma_id_private, id);
3404         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3405                 return -EINVAL;
3406
3407         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3408         if (ret)
3409                 goto err1;
3410
3411         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3412         if (!cma_any_addr(addr)) {
3413                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3414                 if (ret)
3415                         goto err1;
3416
3417                 ret = cma_acquire_dev(id_priv, NULL);
3418                 if (ret)
3419                         goto err1;
3420         }
3421
3422         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3423                 if (addr->sa_family == AF_INET)
3424                         id_priv->afonly = 1;
3425 #if IS_ENABLED(CONFIG_IPV6)
3426                 else if (addr->sa_family == AF_INET6) {
3427                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3428
3429                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3430                 }
3431 #endif
3432         }
3433         daddr = cma_dst_addr(id_priv);
3434         daddr->sa_family = addr->sa_family;
3435
3436         ret = cma_get_port(id_priv);
3437         if (ret)
3438                 goto err2;
3439
3440         return 0;
3441 err2:
3442         if (id_priv->cma_dev) {
3443                 rdma_restrack_del(&id_priv->res);
3444                 cma_release_dev(id_priv);
3445         }
3446 err1:
3447         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3448         return ret;
3449 }
3450 EXPORT_SYMBOL(rdma_bind_addr);
3451
3452 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3453 {
3454         struct cma_hdr *cma_hdr;
3455
3456         cma_hdr = hdr;
3457         cma_hdr->cma_version = CMA_VERSION;
3458         if (cma_family(id_priv) == AF_INET) {
3459                 struct sockaddr_in *src4, *dst4;
3460
3461                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3462                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3463
3464                 cma_set_ip_ver(cma_hdr, 4);
3465                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3466                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3467                 cma_hdr->port = src4->sin_port;
3468         } else if (cma_family(id_priv) == AF_INET6) {
3469                 struct sockaddr_in6 *src6, *dst6;
3470
3471                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3472                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3473
3474                 cma_set_ip_ver(cma_hdr, 6);
3475                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3476                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3477                 cma_hdr->port = src6->sin6_port;
3478         }
3479         return 0;
3480 }
3481
3482 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3483                                 const struct ib_cm_event *ib_event)
3484 {
3485         struct rdma_id_private *id_priv = cm_id->context;
3486         struct rdma_cm_event event = {};
3487         const struct ib_cm_sidr_rep_event_param *rep =
3488                                 &ib_event->param.sidr_rep_rcvd;
3489         int ret = 0;
3490
3491         mutex_lock(&id_priv->handler_mutex);
3492         if (id_priv->state != RDMA_CM_CONNECT)
3493                 goto out;
3494
3495         switch (ib_event->event) {
3496         case IB_CM_SIDR_REQ_ERROR:
3497                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3498                 event.status = -ETIMEDOUT;
3499                 break;
3500         case IB_CM_SIDR_REP_RECEIVED:
3501                 event.param.ud.private_data = ib_event->private_data;
3502                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3503                 if (rep->status != IB_SIDR_SUCCESS) {
3504                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3505                         event.status = ib_event->param.sidr_rep_rcvd.status;
3506                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3507                                              event.status);
3508                         break;
3509                 }
3510                 ret = cma_set_qkey(id_priv, rep->qkey);
3511                 if (ret) {
3512                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3513                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3514                         event.status = ret;
3515                         break;
3516                 }
3517                 ib_init_ah_attr_from_path(id_priv->id.device,
3518                                           id_priv->id.port_num,
3519                                           id_priv->id.route.path_rec,
3520                                           &event.param.ud.ah_attr,
3521                                           rep->sgid_attr);
3522                 event.param.ud.qp_num = rep->qpn;
3523                 event.param.ud.qkey = rep->qkey;
3524                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3525                 event.status = 0;
3526                 break;
3527         default:
3528                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3529                        ib_event->event);
3530                 goto out;
3531         }
3532
3533         ret = id_priv->id.event_handler(&id_priv->id, &event);
3534
3535         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
3536         if (ret) {
3537                 /* Destroy the CM ID by returning a non-zero value. */
3538                 id_priv->cm_id.ib = NULL;
3539                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3540                 mutex_unlock(&id_priv->handler_mutex);
3541                 rdma_destroy_id(&id_priv->id);
3542                 return ret;
3543         }
3544 out:
3545         mutex_unlock(&id_priv->handler_mutex);
3546         return ret;
3547 }
3548
3549 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3550                               struct rdma_conn_param *conn_param)
3551 {
3552         struct ib_cm_sidr_req_param req;
3553         struct ib_cm_id *id;
3554         void *private_data;
3555         u8 offset;
3556         int ret;
3557
3558         memset(&req, 0, sizeof req);
3559         offset = cma_user_data_offset(id_priv);
3560         req.private_data_len = offset + conn_param->private_data_len;
3561         if (req.private_data_len < conn_param->private_data_len)
3562                 return -EINVAL;
3563
3564         if (req.private_data_len) {
3565                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3566                 if (!private_data)
3567                         return -ENOMEM;
3568         } else {
3569                 private_data = NULL;
3570         }
3571
3572         if (conn_param->private_data && conn_param->private_data_len)
3573                 memcpy(private_data + offset, conn_param->private_data,
3574                        conn_param->private_data_len);
3575
3576         if (private_data) {
3577                 ret = cma_format_hdr(private_data, id_priv);
3578                 if (ret)
3579                         goto out;
3580                 req.private_data = private_data;
3581         }
3582
3583         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3584                              id_priv);
3585         if (IS_ERR(id)) {
3586                 ret = PTR_ERR(id);
3587                 goto out;
3588         }
3589         id_priv->cm_id.ib = id;
3590
3591         req.path = id_priv->id.route.path_rec;
3592         req.sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3593         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3594         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3595         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3596
3597         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3598         if (ret) {
3599                 ib_destroy_cm_id(id_priv->cm_id.ib);
3600                 id_priv->cm_id.ib = NULL;
3601         }
3602 out:
3603         kfree(private_data);
3604         return ret;
3605 }
3606
3607 static int cma_connect_ib(struct rdma_id_private *id_priv,
3608                           struct rdma_conn_param *conn_param)
3609 {
3610         struct ib_cm_req_param req;
3611         struct rdma_route *route;
3612         void *private_data;
3613         struct ib_cm_id *id;
3614         u8 offset;
3615         int ret;
3616
3617         memset(&req, 0, sizeof req);
3618         offset = cma_user_data_offset(id_priv);
3619         req.private_data_len = offset + conn_param->private_data_len;
3620         if (req.private_data_len < conn_param->private_data_len)
3621                 return -EINVAL;
3622
3623         if (req.private_data_len) {
3624                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3625                 if (!private_data)
3626                         return -ENOMEM;
3627         } else {
3628                 private_data = NULL;
3629         }
3630
3631         if (conn_param->private_data && conn_param->private_data_len)
3632                 memcpy(private_data + offset, conn_param->private_data,
3633                        conn_param->private_data_len);
3634
3635         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3636         if (IS_ERR(id)) {
3637                 ret = PTR_ERR(id);
3638                 goto out;
3639         }
3640         id_priv->cm_id.ib = id;
3641
3642         route = &id_priv->id.route;
3643         if (private_data) {
3644                 ret = cma_format_hdr(private_data, id_priv);
3645                 if (ret)
3646                         goto out;
3647                 req.private_data = private_data;
3648         }
3649
3650         req.primary_path = &route->path_rec[0];
3651         if (route->num_paths == 2)
3652                 req.alternate_path = &route->path_rec[1];
3653
3654         req.ppath_sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3655         /* Alternate path SGID attribute currently unsupported */
3656         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3657         req.qp_num = id_priv->qp_num;
3658         req.qp_type = id_priv->id.qp_type;
3659         req.starting_psn = id_priv->seq_num;
3660         req.responder_resources = conn_param->responder_resources;
3661         req.initiator_depth = conn_param->initiator_depth;
3662         req.flow_control = conn_param->flow_control;
3663         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3664         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3665         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3666         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3667         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3668         req.srq = id_priv->srq ? 1 : 0;
3669
3670         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3671 out:
3672         if (ret && !IS_ERR(id)) {
3673                 ib_destroy_cm_id(id);
3674                 id_priv->cm_id.ib = NULL;
3675         }
3676
3677         kfree(private_data);
3678         return ret;
3679 }
3680
3681 static int cma_connect_iw(struct rdma_id_private *id_priv,
3682                           struct rdma_conn_param *conn_param)
3683 {
3684         struct iw_cm_id *cm_id;
3685         int ret;
3686         struct iw_cm_conn_param iw_param;
3687
3688         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3689         if (IS_ERR(cm_id))
3690                 return PTR_ERR(cm_id);
3691
3692         cm_id->tos = id_priv->tos;
3693         id_priv->cm_id.iw = cm_id;
3694
3695         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3696                rdma_addr_size(cma_src_addr(id_priv)));
3697         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3698                rdma_addr_size(cma_dst_addr(id_priv)));
3699
3700         ret = cma_modify_qp_rtr(id_priv, conn_param);
3701         if (ret)
3702                 goto out;
3703
3704         if (conn_param) {
3705                 iw_param.ord = conn_param->initiator_depth;
3706                 iw_param.ird = conn_param->responder_resources;
3707                 iw_param.private_data = conn_param->private_data;
3708                 iw_param.private_data_len = conn_param->private_data_len;
3709                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3710         } else {
3711                 memset(&iw_param, 0, sizeof iw_param);
3712                 iw_param.qpn = id_priv->qp_num;
3713         }
3714         ret = iw_cm_connect(cm_id, &iw_param);
3715 out:
3716         if (ret) {
3717                 iw_destroy_cm_id(cm_id);
3718                 id_priv->cm_id.iw = NULL;
3719         }
3720         return ret;
3721 }
3722
3723 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3724 {
3725         struct rdma_id_private *id_priv;
3726         int ret;
3727
3728         id_priv = container_of(id, struct rdma_id_private, id);
3729         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3730                 return -EINVAL;
3731
3732         if (!id->qp) {
3733                 id_priv->qp_num = conn_param->qp_num;
3734                 id_priv->srq = conn_param->srq;
3735         }
3736
3737         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3738                 if (id->qp_type == IB_QPT_UD)
3739                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3740                 else
3741                         ret = cma_connect_ib(id_priv, conn_param);
3742         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3743                 ret = cma_connect_iw(id_priv, conn_param);
3744         else
3745                 ret = -ENOSYS;
3746         if (ret)
3747                 goto err;
3748
3749         return 0;
3750 err:
3751         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3752         return ret;
3753 }
3754 EXPORT_SYMBOL(rdma_connect);
3755
3756 static int cma_accept_ib(struct rdma_id_private *id_priv,
3757                          struct rdma_conn_param *conn_param)
3758 {
3759         struct ib_cm_rep_param rep;
3760         int ret;
3761
3762         ret = cma_modify_qp_rtr(id_priv, conn_param);
3763         if (ret)
3764                 goto out;
3765
3766         ret = cma_modify_qp_rts(id_priv, conn_param);
3767         if (ret)
3768                 goto out;
3769
3770         memset(&rep, 0, sizeof rep);
3771         rep.qp_num = id_priv->qp_num;
3772         rep.starting_psn = id_priv->seq_num;
3773         rep.private_data = conn_param->private_data;
3774         rep.private_data_len = conn_param->private_data_len;
3775         rep.responder_resources = conn_param->responder_resources;
3776         rep.initiator_depth = conn_param->initiator_depth;
3777         rep.failover_accepted = 0;
3778         rep.flow_control = conn_param->flow_control;
3779         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3780         rep.srq = id_priv->srq ? 1 : 0;
3781
3782         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3783 out:
3784         return ret;
3785 }
3786
3787 static int cma_accept_iw(struct rdma_id_private *id_priv,
3788                   struct rdma_conn_param *conn_param)
3789 {
3790         struct iw_cm_conn_param iw_param;
3791         int ret;
3792
3793         if (!conn_param)
3794                 return -EINVAL;
3795
3796         ret = cma_modify_qp_rtr(id_priv, conn_param);
3797         if (ret)
3798                 return ret;
3799
3800         iw_param.ord = conn_param->initiator_depth;
3801         iw_param.ird = conn_param->responder_resources;
3802         iw_param.private_data = conn_param->private_data;
3803         iw_param.private_data_len = conn_param->private_data_len;
3804         if (id_priv->id.qp) {
3805                 iw_param.qpn = id_priv->qp_num;
3806         } else
3807                 iw_param.qpn = conn_param->qp_num;
3808
3809         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3810 }
3811
3812 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3813                              enum ib_cm_sidr_status status, u32 qkey,
3814                              const void *private_data, int private_data_len)
3815 {
3816         struct ib_cm_sidr_rep_param rep;
3817         int ret;
3818
3819         memset(&rep, 0, sizeof rep);
3820         rep.status = status;
3821         if (status == IB_SIDR_SUCCESS) {
3822                 ret = cma_set_qkey(id_priv, qkey);
3823                 if (ret)
3824                         return ret;
3825                 rep.qp_num = id_priv->qp_num;
3826                 rep.qkey = id_priv->qkey;
3827         }
3828         rep.private_data = private_data;
3829         rep.private_data_len = private_data_len;
3830
3831         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3832 }
3833
3834 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3835                   const char *caller)
3836 {
3837         struct rdma_id_private *id_priv;
3838         int ret;
3839
3840         id_priv = container_of(id, struct rdma_id_private, id);
3841
3842         if (caller)
3843                 id_priv->res.kern_name = caller;
3844         else
3845                 rdma_restrack_set_task(&id_priv->res, current);
3846
3847         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3848                 return -EINVAL;
3849
3850         if (!id->qp && conn_param) {
3851                 id_priv->qp_num = conn_param->qp_num;
3852                 id_priv->srq = conn_param->srq;
3853         }
3854
3855         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3856                 if (id->qp_type == IB_QPT_UD) {
3857                         if (conn_param)
3858                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3859                                                         conn_param->qkey,
3860                                                         conn_param->private_data,
3861                                                         conn_param->private_data_len);
3862                         else
3863                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3864                                                         0, NULL, 0);
3865                 } else {
3866                         if (conn_param)
3867                                 ret = cma_accept_ib(id_priv, conn_param);
3868                         else
3869                                 ret = cma_rep_recv(id_priv);
3870                 }
3871         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3872                 ret = cma_accept_iw(id_priv, conn_param);
3873         else
3874                 ret = -ENOSYS;
3875
3876         if (ret)
3877                 goto reject;
3878
3879         return 0;
3880 reject:
3881         cma_modify_qp_err(id_priv);
3882         rdma_reject(id, NULL, 0);
3883         return ret;
3884 }
3885 EXPORT_SYMBOL(__rdma_accept);
3886
3887 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3888 {
3889         struct rdma_id_private *id_priv;
3890         int ret;
3891
3892         id_priv = container_of(id, struct rdma_id_private, id);
3893         if (!id_priv->cm_id.ib)
3894                 return -EINVAL;
3895
3896         switch (id->device->node_type) {
3897         case RDMA_NODE_IB_CA:
3898                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3899                 break;
3900         default:
3901                 ret = 0;
3902                 break;
3903         }
3904         return ret;
3905 }
3906 EXPORT_SYMBOL(rdma_notify);
3907
3908 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3909                 u8 private_data_len)
3910 {
3911         struct rdma_id_private *id_priv;
3912         int ret;
3913
3914         id_priv = container_of(id, struct rdma_id_private, id);
3915         if (!id_priv->cm_id.ib)
3916                 return -EINVAL;
3917
3918         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3919                 if (id->qp_type == IB_QPT_UD)
3920                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3921                                                 private_data, private_data_len);
3922                 else
3923                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3924                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3925                                              0, private_data, private_data_len);
3926         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3927                 ret = iw_cm_reject(id_priv->cm_id.iw,
3928                                    private_data, private_data_len);
3929         } else
3930                 ret = -ENOSYS;
3931
3932         return ret;
3933 }
3934 EXPORT_SYMBOL(rdma_reject);
3935
3936 int rdma_disconnect(struct rdma_cm_id *id)
3937 {
3938         struct rdma_id_private *id_priv;
3939         int ret;
3940
3941         id_priv = container_of(id, struct rdma_id_private, id);
3942         if (!id_priv->cm_id.ib)
3943                 return -EINVAL;
3944
3945         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3946                 ret = cma_modify_qp_err(id_priv);
3947                 if (ret)
3948                         goto out;
3949                 /* Initiate or respond to a disconnect. */
3950                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3951                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3952         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3953                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3954         } else
3955                 ret = -EINVAL;
3956
3957 out:
3958         return ret;
3959 }
3960 EXPORT_SYMBOL(rdma_disconnect);
3961
3962 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3963 {
3964         struct rdma_id_private *id_priv;
3965         struct cma_multicast *mc = multicast->context;
3966         struct rdma_cm_event event = {};
3967         int ret = 0;
3968
3969         id_priv = mc->id_priv;
3970         mutex_lock(&id_priv->handler_mutex);
3971         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3972             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3973                 goto out;
3974
3975         if (!status)
3976                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3977         else
3978                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3979                                      status);
3980         mutex_lock(&id_priv->qp_mutex);
3981         if (!status && id_priv->id.qp) {
3982                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3983                                          be16_to_cpu(multicast->rec.mlid));
3984                 if (status)
3985                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3986                                              status);
3987         }
3988         mutex_unlock(&id_priv->qp_mutex);
3989
3990         event.status = status;
3991         event.param.ud.private_data = mc->context;
3992         if (!status) {
3993                 struct rdma_dev_addr *dev_addr =
3994                         &id_priv->id.route.addr.dev_addr;
3995                 struct net_device *ndev =
3996                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
3997                 enum ib_gid_type gid_type =
3998                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
3999                         rdma_start_port(id_priv->cma_dev->device)];
4000
4001                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
4002                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
4003                                                id_priv->id.port_num,
4004                                                &multicast->rec,
4005                                                ndev, gid_type,
4006                                                &event.param.ud.ah_attr);
4007                 if (ret)
4008                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4009
4010                 event.param.ud.qp_num = 0xFFFFFF;
4011                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
4012                 if (ndev)
4013                         dev_put(ndev);
4014         } else
4015                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4016
4017         ret = id_priv->id.event_handler(&id_priv->id, &event);
4018
4019         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
4020         if (ret) {
4021                 cma_exch(id_priv, RDMA_CM_DESTROYING);
4022                 mutex_unlock(&id_priv->handler_mutex);
4023                 rdma_destroy_id(&id_priv->id);
4024                 return 0;
4025         }
4026
4027 out:
4028         mutex_unlock(&id_priv->handler_mutex);
4029         return 0;
4030 }
4031
4032 static void cma_set_mgid(struct rdma_id_private *id_priv,
4033                          struct sockaddr *addr, union ib_gid *mgid)
4034 {
4035         unsigned char mc_map[MAX_ADDR_LEN];
4036         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4037         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4038         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4039
4040         if (cma_any_addr(addr)) {
4041                 memset(mgid, 0, sizeof *mgid);
4042         } else if ((addr->sa_family == AF_INET6) &&
4043                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4044                                                                  0xFF10A01B)) {
4045                 /* IPv6 address is an SA assigned MGID. */
4046                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4047         } else if (addr->sa_family == AF_IB) {
4048                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4049         } else if (addr->sa_family == AF_INET6) {
4050                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4051                 if (id_priv->id.ps == RDMA_PS_UDP)
4052                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4053                 *mgid = *(union ib_gid *) (mc_map + 4);
4054         } else {
4055                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4056                 if (id_priv->id.ps == RDMA_PS_UDP)
4057                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4058                 *mgid = *(union ib_gid *) (mc_map + 4);
4059         }
4060 }
4061
4062 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4063                                  struct cma_multicast *mc)
4064 {
4065         struct ib_sa_mcmember_rec rec;
4066         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4067         ib_sa_comp_mask comp_mask;
4068         int ret;
4069
4070         ib_addr_get_mgid(dev_addr, &rec.mgid);
4071         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4072                                      &rec.mgid, &rec);
4073         if (ret)
4074                 return ret;
4075
4076         ret = cma_set_qkey(id_priv, 0);
4077         if (ret)
4078                 return ret;
4079
4080         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4081         rec.qkey = cpu_to_be32(id_priv->qkey);
4082         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4083         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4084         rec.join_state = mc->join_state;
4085
4086         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4087             (!ib_sa_sendonly_fullmem_support(&sa_client,
4088                                              id_priv->id.device,
4089                                              id_priv->id.port_num))) {
4090                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4091                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4092                         id_priv->id.device->name, id_priv->id.port_num);
4093                 return -EOPNOTSUPP;
4094         }
4095
4096         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4097                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4098                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4099                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4100                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4101
4102         if (id_priv->id.ps == RDMA_PS_IPOIB)
4103                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4104                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4105                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4106                              IB_SA_MCMEMBER_REC_MTU |
4107                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4108
4109         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4110                                                 id_priv->id.port_num, &rec,
4111                                                 comp_mask, GFP_KERNEL,
4112                                                 cma_ib_mc_handler, mc);
4113         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4114 }
4115
4116 static void iboe_mcast_work_handler(struct work_struct *work)
4117 {
4118         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4119         struct cma_multicast *mc = mw->mc;
4120         struct ib_sa_multicast *m = mc->multicast.ib;
4121
4122         mc->multicast.ib->context = mc;
4123         cma_ib_mc_handler(0, m);
4124         kref_put(&mc->mcref, release_mc);
4125         kfree(mw);
4126 }
4127
4128 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4129                               enum ib_gid_type gid_type)
4130 {
4131         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4132         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4133
4134         if (cma_any_addr(addr)) {
4135                 memset(mgid, 0, sizeof *mgid);
4136         } else if (addr->sa_family == AF_INET6) {
4137                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4138         } else {
4139                 mgid->raw[0] =
4140                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4141                 mgid->raw[1] =
4142                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4143                 mgid->raw[2] = 0;
4144                 mgid->raw[3] = 0;
4145                 mgid->raw[4] = 0;
4146                 mgid->raw[5] = 0;
4147                 mgid->raw[6] = 0;
4148                 mgid->raw[7] = 0;
4149                 mgid->raw[8] = 0;
4150                 mgid->raw[9] = 0;
4151                 mgid->raw[10] = 0xff;
4152                 mgid->raw[11] = 0xff;
4153                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4154         }
4155 }
4156
4157 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4158                                    struct cma_multicast *mc)
4159 {
4160         struct iboe_mcast_work *work;
4161         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4162         int err = 0;
4163         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4164         struct net_device *ndev = NULL;
4165         enum ib_gid_type gid_type;
4166         bool send_only;
4167
4168         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4169
4170         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4171                 return -EINVAL;
4172
4173         work = kzalloc(sizeof *work, GFP_KERNEL);
4174         if (!work)
4175                 return -ENOMEM;
4176
4177         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4178         if (!mc->multicast.ib) {
4179                 err = -ENOMEM;
4180                 goto out1;
4181         }
4182
4183         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4184                    rdma_start_port(id_priv->cma_dev->device)];
4185         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4186
4187         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4188         if (id_priv->id.ps == RDMA_PS_UDP)
4189                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4190
4191         if (dev_addr->bound_dev_if)
4192                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4193         if (!ndev) {
4194                 err = -ENODEV;
4195                 goto out2;
4196         }
4197         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4198         mc->multicast.ib->rec.hop_limit = 1;
4199         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4200
4201         if (addr->sa_family == AF_INET) {
4202                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4203                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4204                         if (!send_only) {
4205                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4206                                                     true);
4207                         }
4208                 }
4209         } else {
4210                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4211                         err = -ENOTSUPP;
4212         }
4213         dev_put(ndev);
4214         if (err || !mc->multicast.ib->rec.mtu) {
4215                 if (!err)
4216                         err = -EINVAL;
4217                 goto out2;
4218         }
4219         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4220                     &mc->multicast.ib->rec.port_gid);
4221         work->id = id_priv;
4222         work->mc = mc;
4223         INIT_WORK(&work->work, iboe_mcast_work_handler);
4224         kref_get(&mc->mcref);
4225         queue_work(cma_wq, &work->work);
4226
4227         return 0;
4228
4229 out2:
4230         kfree(mc->multicast.ib);
4231 out1:
4232         kfree(work);
4233         return err;
4234 }
4235
4236 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4237                         u8 join_state, void *context)
4238 {
4239         struct rdma_id_private *id_priv;
4240         struct cma_multicast *mc;
4241         int ret;
4242
4243         if (!id->device)
4244                 return -EINVAL;
4245
4246         id_priv = container_of(id, struct rdma_id_private, id);
4247         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4248             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4249                 return -EINVAL;
4250
4251         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4252         if (!mc)
4253                 return -ENOMEM;
4254
4255         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4256         mc->context = context;
4257         mc->id_priv = id_priv;
4258         mc->join_state = join_state;
4259
4260         if (rdma_protocol_roce(id->device, id->port_num)) {
4261                 kref_init(&mc->mcref);
4262                 ret = cma_iboe_join_multicast(id_priv, mc);
4263                 if (ret)
4264                         goto out_err;
4265         } else if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4266                 ret = cma_join_ib_multicast(id_priv, mc);
4267                 if (ret)
4268                         goto out_err;
4269         } else {
4270                 ret = -ENOSYS;
4271                 goto out_err;
4272         }
4273
4274         spin_lock(&id_priv->lock);
4275         list_add(&mc->list, &id_priv->mc_list);
4276         spin_unlock(&id_priv->lock);
4277
4278         return 0;
4279 out_err:
4280         kfree(mc);
4281         return ret;
4282 }
4283 EXPORT_SYMBOL(rdma_join_multicast);
4284
4285 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4286 {
4287         struct rdma_id_private *id_priv;
4288         struct cma_multicast *mc;
4289
4290         id_priv = container_of(id, struct rdma_id_private, id);
4291         spin_lock_irq(&id_priv->lock);
4292         list_for_each_entry(mc, &id_priv->mc_list, list) {
4293                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4294                         list_del(&mc->list);
4295                         spin_unlock_irq(&id_priv->lock);
4296
4297                         if (id->qp)
4298                                 ib_detach_mcast(id->qp,
4299                                                 &mc->multicast.ib->rec.mgid,
4300                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4301
4302                         BUG_ON(id_priv->cma_dev->device != id->device);
4303
4304                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4305                                 ib_sa_free_multicast(mc->multicast.ib);
4306                                 kfree(mc);
4307                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4308                                 cma_leave_roce_mc_group(id_priv, mc);
4309                         }
4310                         return;
4311                 }
4312         }
4313         spin_unlock_irq(&id_priv->lock);
4314 }
4315 EXPORT_SYMBOL(rdma_leave_multicast);
4316
4317 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4318 {
4319         struct rdma_dev_addr *dev_addr;
4320         struct cma_ndev_work *work;
4321
4322         dev_addr = &id_priv->id.route.addr.dev_addr;
4323
4324         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4325             (net_eq(dev_net(ndev), dev_addr->net)) &&
4326             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4327                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4328                         ndev->name, &id_priv->id);
4329                 work = kzalloc(sizeof *work, GFP_KERNEL);
4330                 if (!work)
4331                         return -ENOMEM;
4332
4333                 INIT_WORK(&work->work, cma_ndev_work_handler);
4334                 work->id = id_priv;
4335                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4336                 atomic_inc(&id_priv->refcount);
4337                 queue_work(cma_wq, &work->work);
4338         }
4339
4340         return 0;
4341 }
4342
4343 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4344                                void *ptr)
4345 {
4346         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4347         struct cma_device *cma_dev;
4348         struct rdma_id_private *id_priv;
4349         int ret = NOTIFY_DONE;
4350
4351         if (event != NETDEV_BONDING_FAILOVER)
4352                 return NOTIFY_DONE;
4353
4354         if (!netif_is_bond_master(ndev))
4355                 return NOTIFY_DONE;
4356
4357         mutex_lock(&lock);
4358         list_for_each_entry(cma_dev, &dev_list, list)
4359                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4360                         ret = cma_netdev_change(ndev, id_priv);
4361                         if (ret)
4362                                 goto out;
4363                 }
4364
4365 out:
4366         mutex_unlock(&lock);
4367         return ret;
4368 }
4369
4370 static struct notifier_block cma_nb = {
4371         .notifier_call = cma_netdev_callback
4372 };
4373
4374 static void cma_add_one(struct ib_device *device)
4375 {
4376         struct cma_device *cma_dev;
4377         struct rdma_id_private *id_priv;
4378         unsigned int i;
4379         unsigned long supported_gids = 0;
4380
4381         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4382         if (!cma_dev)
4383                 return;
4384
4385         cma_dev->device = device;
4386         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4387                                             sizeof(*cma_dev->default_gid_type),
4388                                             GFP_KERNEL);
4389         if (!cma_dev->default_gid_type)
4390                 goto free_cma_dev;
4391
4392         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4393                                             sizeof(*cma_dev->default_roce_tos),
4394                                             GFP_KERNEL);
4395         if (!cma_dev->default_roce_tos)
4396                 goto free_gid_type;
4397
4398         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4399                 supported_gids = roce_gid_type_mask_support(device, i);
4400                 WARN_ON(!supported_gids);
4401                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4402                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4403                                 CMA_PREFERRED_ROCE_GID_TYPE;
4404                 else
4405                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4406                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4407                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4408         }
4409
4410         init_completion(&cma_dev->comp);
4411         atomic_set(&cma_dev->refcount, 1);
4412         INIT_LIST_HEAD(&cma_dev->id_list);
4413         ib_set_client_data(device, &cma_client, cma_dev);
4414
4415         mutex_lock(&lock);
4416         list_add_tail(&cma_dev->list, &dev_list);
4417         list_for_each_entry(id_priv, &listen_any_list, list)
4418                 cma_listen_on_dev(id_priv, cma_dev);
4419         mutex_unlock(&lock);
4420
4421         return;
4422
4423 free_gid_type:
4424         kfree(cma_dev->default_gid_type);
4425
4426 free_cma_dev:
4427         kfree(cma_dev);
4428
4429         return;
4430 }
4431
4432 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4433 {
4434         struct rdma_cm_event event = {};
4435         enum rdma_cm_state state;
4436         int ret = 0;
4437
4438         /* Record that we want to remove the device */
4439         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4440         if (state == RDMA_CM_DESTROYING)
4441                 return 0;
4442
4443         cma_cancel_operation(id_priv, state);
4444         mutex_lock(&id_priv->handler_mutex);
4445
4446         /* Check for destruction from another callback. */
4447         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4448                 goto out;
4449
4450         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4451         ret = id_priv->id.event_handler(&id_priv->id, &event);
4452 out:
4453         mutex_unlock(&id_priv->handler_mutex);
4454         return ret;
4455 }
4456
4457 static void cma_process_remove(struct cma_device *cma_dev)
4458 {
4459         struct rdma_id_private *id_priv;
4460         int ret;
4461
4462         mutex_lock(&lock);
4463         while (!list_empty(&cma_dev->id_list)) {
4464                 id_priv = list_entry(cma_dev->id_list.next,
4465                                      struct rdma_id_private, list);
4466
4467                 list_del(&id_priv->listen_list);
4468                 list_del_init(&id_priv->list);
4469                 atomic_inc(&id_priv->refcount);
4470                 mutex_unlock(&lock);
4471
4472                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4473                 cma_deref_id(id_priv);
4474                 if (ret)
4475                         rdma_destroy_id(&id_priv->id);
4476
4477                 mutex_lock(&lock);
4478         }
4479         mutex_unlock(&lock);
4480
4481         cma_deref_dev(cma_dev);
4482         wait_for_completion(&cma_dev->comp);
4483 }
4484
4485 static void cma_remove_one(struct ib_device *device, void *client_data)
4486 {
4487         struct cma_device *cma_dev = client_data;
4488
4489         if (!cma_dev)
4490                 return;
4491
4492         mutex_lock(&lock);
4493         list_del(&cma_dev->list);
4494         mutex_unlock(&lock);
4495
4496         cma_process_remove(cma_dev);
4497         kfree(cma_dev->default_roce_tos);
4498         kfree(cma_dev->default_gid_type);
4499         kfree(cma_dev);
4500 }
4501
4502 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4503 {
4504         struct nlmsghdr *nlh;
4505         struct rdma_cm_id_stats *id_stats;
4506         struct rdma_id_private *id_priv;
4507         struct rdma_cm_id *id = NULL;
4508         struct cma_device *cma_dev;
4509         int i_dev = 0, i_id = 0;
4510
4511         /*
4512          * We export all of the IDs as a sequence of messages.  Each
4513          * ID gets its own netlink message.
4514          */
4515         mutex_lock(&lock);
4516
4517         list_for_each_entry(cma_dev, &dev_list, list) {
4518                 if (i_dev < cb->args[0]) {
4519                         i_dev++;
4520                         continue;
4521                 }
4522
4523                 i_id = 0;
4524                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4525                         if (i_id < cb->args[1]) {
4526                                 i_id++;
4527                                 continue;
4528                         }
4529
4530                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4531                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4532                                                 RDMA_NL_RDMA_CM_ID_STATS,
4533                                                 NLM_F_MULTI);
4534                         if (!id_stats)
4535                                 goto out;
4536
4537                         memset(id_stats, 0, sizeof *id_stats);
4538                         id = &id_priv->id;
4539                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4540                         id_stats->port_num = id->port_num;
4541                         id_stats->bound_dev_if =
4542                                 id->route.addr.dev_addr.bound_dev_if;
4543
4544                         if (ibnl_put_attr(skb, nlh,
4545                                           rdma_addr_size(cma_src_addr(id_priv)),
4546                                           cma_src_addr(id_priv),
4547                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4548                                 goto out;
4549                         if (ibnl_put_attr(skb, nlh,
4550                                           rdma_addr_size(cma_dst_addr(id_priv)),
4551                                           cma_dst_addr(id_priv),
4552                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4553                                 goto out;
4554
4555                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4556                         id_stats->port_space    = id->ps;
4557                         id_stats->cm_state      = id_priv->state;
4558                         id_stats->qp_num        = id_priv->qp_num;
4559                         id_stats->qp_type       = id->qp_type;
4560
4561                         i_id++;
4562                         nlmsg_end(skb, nlh);
4563                 }
4564
4565                 cb->args[1] = 0;
4566                 i_dev++;
4567         }
4568
4569 out:
4570         mutex_unlock(&lock);
4571         cb->args[0] = i_dev;
4572         cb->args[1] = i_id;
4573
4574         return skb->len;
4575 }
4576
4577 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4578         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4579 };
4580
4581 static int cma_init_net(struct net *net)
4582 {
4583         struct cma_pernet *pernet = cma_pernet(net);
4584
4585         idr_init(&pernet->tcp_ps);
4586         idr_init(&pernet->udp_ps);
4587         idr_init(&pernet->ipoib_ps);
4588         idr_init(&pernet->ib_ps);
4589
4590         return 0;
4591 }
4592
4593 static void cma_exit_net(struct net *net)
4594 {
4595         struct cma_pernet *pernet = cma_pernet(net);
4596
4597         idr_destroy(&pernet->tcp_ps);
4598         idr_destroy(&pernet->udp_ps);
4599         idr_destroy(&pernet->ipoib_ps);
4600         idr_destroy(&pernet->ib_ps);
4601 }
4602
4603 static struct pernet_operations cma_pernet_operations = {
4604         .init = cma_init_net,
4605         .exit = cma_exit_net,
4606         .id = &cma_pernet_id,
4607         .size = sizeof(struct cma_pernet),
4608 };
4609
4610 static int __init cma_init(void)
4611 {
4612         int ret;
4613
4614         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4615         if (!cma_wq)
4616                 return -ENOMEM;
4617
4618         ret = register_pernet_subsys(&cma_pernet_operations);
4619         if (ret)
4620                 goto err_wq;
4621
4622         ib_sa_register_client(&sa_client);
4623         register_netdevice_notifier(&cma_nb);
4624
4625         ret = ib_register_client(&cma_client);
4626         if (ret)
4627                 goto err;
4628
4629         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4630         cma_configfs_init();
4631
4632         return 0;
4633
4634 err:
4635         unregister_netdevice_notifier(&cma_nb);
4636         ib_sa_unregister_client(&sa_client);
4637 err_wq:
4638         destroy_workqueue(cma_wq);
4639         return ret;
4640 }
4641
4642 static void __exit cma_cleanup(void)
4643 {
4644         cma_configfs_exit();
4645         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4646         ib_unregister_client(&cma_client);
4647         unregister_netdevice_notifier(&cma_nb);
4648         ib_sa_unregister_client(&sa_client);
4649         unregister_pernet_subsys(&cma_pernet_operations);
4650         destroy_workqueue(cma_wq);
4651 }
4652
4653 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4654
4655 module_init(cma_init);
4656 module_exit(cma_cleanup);