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->gid_type = 0;
 494         id_priv->id.device = cma_dev->device;
 495         id_priv->id.route.addr.dev_addr.transport =
 496                 rdma_node_get_transport(cma_dev->device->node_type);
 497         list_add_tail(&id_priv->list, &cma_dev->id_list);
 498         rdma_restrack_add(&id_priv->res);
 499 }
 500
 501 static void cma_attach_to_dev(struct rdma_id_private *id_priv,
 502                               struct cma_device *cma_dev)
 503 {
 504         _cma_attach_to_dev(id_priv, cma_dev);
 505         id_priv->gid_type =
 506                 cma_dev->default_gid_type[id_priv->id.port_num -
 507                                           rdma_start_port(cma_dev->device)];
 508 }
 509
 510 void cma_deref_dev(struct cma_device *cma_dev)
 511 {
 512         if (atomic_dec_and_test(&cma_dev->refcount))
 513                 complete(&cma_dev->comp);
 514 }
 515
 516 static inline void release_mc(struct kref *kref)
 517 {
 518         struct cma_multicast *mc = container_of(kref, struct cma_multicast, mcref);
 519
 520         kfree(mc->multicast.ib);
 521         kfree(mc);
 522 }
 523
 524 static void cma_release_dev(struct rdma_id_private *id_priv)
 525 {
 526         mutex_lock(&lock);
 527         list_del(&id_priv->list);
 528         cma_deref_dev(id_priv->cma_dev);
 529         id_priv->cma_dev = NULL;
 530         mutex_unlock(&lock);
 531 }
 532
 533 static inline struct sockaddr *cma_src_addr(struct rdma_id_private *id_priv)
 534 {
 535         return (struct sockaddr *) &id_priv->id.route.addr.src_addr;
 536 }
 537
 538 static inline struct sockaddr *cma_dst_addr(struct rdma_id_private *id_priv)
 539 {
 540         return (struct sockaddr *) &id_priv->id.route.addr.dst_addr;
 541 }
 542
 543 static inline unsigned short cma_family(struct rdma_id_private *id_priv)
 544 {
 545         return id_priv->id.route.addr.src_addr.ss_family;
 546 }
 547
 548 static int cma_set_qkey(struct rdma_id_private *id_priv, u32 qkey)
 549 {
 550         struct ib_sa_mcmember_rec rec;
 551         int ret = 0;
 552
 553         if (id_priv->qkey) {
 554                 if (qkey && id_priv->qkey != qkey)
 555                         return -EINVAL;
 556                 return 0;
 557         }
 558
 559         if (qkey) {
 560                 id_priv->qkey = qkey;
 561                 return 0;
 562         }
 563
 564         switch (id_priv->id.ps) {
 565         case RDMA_PS_UDP:
 566         case RDMA_PS_IB:
 567                 id_priv->qkey = RDMA_UDP_QKEY;
 568                 break;
 569         case RDMA_PS_IPOIB:
 570                 ib_addr_get_mgid(&id_priv->id.route.addr.dev_addr, &rec.mgid);
 571                 ret = ib_sa_get_mcmember_rec(id_priv->id.device,
 572                                              id_priv->id.port_num, &rec.mgid,
 573                                              &rec);
 574                 if (!ret)
 575                         id_priv->qkey = be32_to_cpu(rec.qkey);
 576                 break;
 577         default:
 578                 break;
 579         }
 580         return ret;
 581 }
 582
 583 static void cma_translate_ib(struct sockaddr_ib *sib, struct rdma_dev_addr *dev_addr)
 584 {
 585         dev_addr->dev_type = ARPHRD_INFINIBAND;
 586         rdma_addr_set_sgid(dev_addr, (union ib_gid *) &sib->sib_addr);
 587         ib_addr_set_pkey(dev_addr, ntohs(sib->sib_pkey));
 588 }
 589
 590 static int cma_translate_addr(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
 591 {
 592         int ret;
 593
 594         if (addr->sa_family != AF_IB) {
 595                 ret = rdma_translate_ip(addr, dev_addr);
 596         } else {
 597                 cma_translate_ib((struct sockaddr_ib *) addr, dev_addr);
 598                 ret = 0;
 599         }
 600
 601         return ret;
 602 }
 603
 604 static const struct ib_gid_attr *
 605 cma_validate_port(struct ib_device *device, u8 port,
 606                   enum ib_gid_type gid_type,
 607                   union ib_gid *gid,
 608                   struct rdma_id_private *id_priv)
 609 {
 610         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 611         int bound_if_index = dev_addr->bound_dev_if;
 612         const struct ib_gid_attr *sgid_attr;
 613         int dev_type = dev_addr->dev_type;
 614         struct net_device *ndev = NULL;
 615
 616         if ((dev_type == ARPHRD_INFINIBAND) && !rdma_protocol_ib(device, port))
 617                 return ERR_PTR(-ENODEV);
 618
 619         if ((dev_type != ARPHRD_INFINIBAND) && rdma_protocol_ib(device, port))
 620                 return ERR_PTR(-ENODEV);
 621
 622         if (dev_type == ARPHRD_ETHER && rdma_protocol_roce(device, port)) {
 623                 ndev = dev_get_by_index(dev_addr->net, bound_if_index);
 624                 if (!ndev)
 625                         return ERR_PTR(-ENODEV);
 626         } else {
 627                 gid_type = IB_GID_TYPE_IB;
 628         }
 629
 630         sgid_attr = rdma_find_gid_by_port(device, gid, gid_type, port, ndev);
 631         if (ndev)
 632                 dev_put(ndev);
 633         return sgid_attr;
 634 }
 635
 636 static void cma_bind_sgid_attr(struct rdma_id_private *id_priv,
 637                                const struct ib_gid_attr *sgid_attr)
 638 {
 639         WARN_ON(id_priv->id.route.addr.dev_addr.sgid_attr);
 640         id_priv->id.route.addr.dev_addr.sgid_attr = sgid_attr;
 641 }
 642
 643 static int cma_acquire_dev(struct rdma_id_private *id_priv,
 644                            const struct rdma_id_private *listen_id_priv)
 645 {
 646         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 647         const struct ib_gid_attr *sgid_attr;
 648         struct cma_device *cma_dev;
 649         union ib_gid gid, iboe_gid, *gidp;
 650         enum ib_gid_type gid_type;
 651         enum ib_gid_type default_type;
 652         int ret = -ENODEV;
 653         u8 port;
 654
 655         if (dev_addr->dev_type != ARPHRD_INFINIBAND &&
 656             id_priv->id.ps == RDMA_PS_IPOIB)
 657                 return -EINVAL;
 658
 659         mutex_lock(&lock);
 660         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
 661                     &iboe_gid);
 662
 663         memcpy(&gid, dev_addr->src_dev_addr +
 664                rdma_addr_gid_offset(dev_addr), sizeof gid);
 665
 666         if (listen_id_priv) {
 667                 cma_dev = listen_id_priv->cma_dev;
 668                 port = listen_id_priv->id.port_num;
 669                 gidp = rdma_protocol_roce(cma_dev->device, port) ?
 670                        &iboe_gid : &gid;
 671                 gid_type = rdma_protocol_ib(cma_dev->device, port) ?
 672                                             IB_GID_TYPE_IB :
 673                                             listen_id_priv->gid_type;
 674                 sgid_attr = cma_validate_port(cma_dev->device, port,
 675                                               gid_type, gidp, id_priv);
 676                 if (!IS_ERR(sgid_attr)) {
 677                         id_priv->id.port_num = port;
 678                         cma_bind_sgid_attr(id_priv, sgid_attr);
 679                         ret = 0;
 680                         goto out;
 681                 }
 682         }
 683
 684         list_for_each_entry(cma_dev, &dev_list, list) {
 685                 for (port = 1; port <= cma_dev->device->phys_port_cnt; ++port) {
 686                         if (listen_id_priv &&
 687                             listen_id_priv->cma_dev == cma_dev &&
 688                             listen_id_priv->id.port_num == port)
 689                                 continue;
 690
 691                         gidp = rdma_protocol_roce(cma_dev->device, port) ?
 692                                &iboe_gid : &gid;
 693                         default_type = cma_dev->default_gid_type[port - 1];
 694                         gid_type =
 695                                 rdma_protocol_ib(cma_dev->device, port) ?
 696                                                  IB_GID_TYPE_IB : default_type;
 697                         sgid_attr = cma_validate_port(cma_dev->device, port,
 698                                                       gid_type, gidp, id_priv);
 699                         if (!IS_ERR(sgid_attr)) {
 700                                 id_priv->id.port_num = port;
 701                                 cma_bind_sgid_attr(id_priv, sgid_attr);
 702                                 ret = 0;
 703                                 goto out;
 704                         }
 705                 }
 706         }
 707
 708 out:
 709         if (!ret)
 710                 cma_attach_to_dev(id_priv, cma_dev);
 711
 712         mutex_unlock(&lock);
 713         return ret;
 714 }
 715
 716 /*
 717  * Select the source IB device and address to reach the destination IB address.
 718  */
 719 static int cma_resolve_ib_dev(struct rdma_id_private *id_priv)
 720 {
 721         struct cma_device *cma_dev, *cur_dev;
 722         struct sockaddr_ib *addr;
 723         union ib_gid gid, sgid, *dgid;
 724         u16 pkey, index;
 725         u8 p;
 726         enum ib_port_state port_state;
 727         int i;
 728
 729         cma_dev = NULL;
 730         addr = (struct sockaddr_ib *) cma_dst_addr(id_priv);
 731         dgid = (union ib_gid *) &addr->sib_addr;
 732         pkey = ntohs(addr->sib_pkey);
 733
 734         list_for_each_entry(cur_dev, &dev_list, list) {
 735                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
 736                         if (!rdma_cap_af_ib(cur_dev->device, p))
 737                                 continue;
 738
 739                         if (ib_find_cached_pkey(cur_dev->device, p, pkey, &index))
 740                                 continue;
 741
 742                         if (ib_get_cached_port_state(cur_dev->device, p, &port_state))
 743                                 continue;
 744                         for (i = 0; !rdma_query_gid(cur_dev->device,
 745                                                     p, i, &gid);
 746                              i++) {
 747                                 if (!memcmp(&gid, dgid, sizeof(gid))) {
 748                                         cma_dev = cur_dev;
 749                                         sgid = gid;
 750                                         id_priv->id.port_num = p;
 751                                         goto found;
 752                                 }
 753
 754                                 if (!cma_dev && (gid.global.subnet_prefix ==
 755                                     dgid->global.subnet_prefix) &&
 756                                     port_state == IB_PORT_ACTIVE) {
 757                                         cma_dev = cur_dev;
 758                                         sgid = gid;
 759                                         id_priv->id.port_num = p;
 760                                 }
 761                         }
 762                 }
 763         }
 764
 765         if (!cma_dev)
 766                 return -ENODEV;
 767
 768 found:
 769         cma_attach_to_dev(id_priv, cma_dev);
 770         addr = (struct sockaddr_ib *) cma_src_addr(id_priv);
 771         memcpy(&addr->sib_addr, &sgid, sizeof sgid);
 772         cma_translate_ib(addr, &id_priv->id.route.addr.dev_addr);
 773         return 0;
 774 }
 775
 776 static void cma_deref_id(struct rdma_id_private *id_priv)
 777 {
 778         if (atomic_dec_and_test(&id_priv->refcount))
 779                 complete(&id_priv->comp);
 780 }
 781
 782 struct rdma_cm_id *__rdma_create_id(struct net *net,
 783                                     rdma_cm_event_handler event_handler,
 784                                     void *context, enum rdma_ucm_port_space ps,
 785                                     enum ib_qp_type qp_type, const char *caller)
 786 {
 787         struct rdma_id_private *id_priv;
 788
 789         id_priv = kzalloc(sizeof *id_priv, GFP_KERNEL);
 790         if (!id_priv)
 791                 return ERR_PTR(-ENOMEM);
 792
 793         if (caller)
 794                 id_priv->res.kern_name = caller;
 795         else
 796                 rdma_restrack_set_task(&id_priv->res, current);
 797         id_priv->res.type = RDMA_RESTRACK_CM_ID;
 798         id_priv->state = RDMA_CM_IDLE;
 799         id_priv->id.context = context;
 800         id_priv->id.event_handler = event_handler;
 801         id_priv->id.ps = ps;
 802         id_priv->id.qp_type = qp_type;
 803         id_priv->tos_set = false;
 804         spin_lock_init(&id_priv->lock);
 805         mutex_init(&id_priv->qp_mutex);
 806         init_completion(&id_priv->comp);
 807         atomic_set(&id_priv->refcount, 1);
 808         mutex_init(&id_priv->handler_mutex);
 809         INIT_LIST_HEAD(&id_priv->listen_list);
 810         INIT_LIST_HEAD(&id_priv->mc_list);
 811         get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num);
 812         id_priv->id.route.addr.dev_addr.net = get_net(net);
 813         id_priv->seq_num &= 0x00ffffff;
 814
 815         return &id_priv->id;
 816 }
 817 EXPORT_SYMBOL(__rdma_create_id);
 818
 819 static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 820 {
 821         struct ib_qp_attr qp_attr;
 822         int qp_attr_mask, ret;
 823
 824         qp_attr.qp_state = IB_QPS_INIT;
 825         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 826         if (ret)
 827                 return ret;
 828
 829         ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 830         if (ret)
 831                 return ret;
 832
 833         qp_attr.qp_state = IB_QPS_RTR;
 834         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
 835         if (ret)
 836                 return ret;
 837
 838         qp_attr.qp_state = IB_QPS_RTS;
 839         qp_attr.sq_psn = 0;
 840         ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
 841
 842         return ret;
 843 }
 844
 845 static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp)
 846 {
 847         struct ib_qp_attr qp_attr;
 848         int qp_attr_mask, ret;
 849
 850         qp_attr.qp_state = IB_QPS_INIT;
 851         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 852         if (ret)
 853                 return ret;
 854
 855         return ib_modify_qp(qp, &qp_attr, qp_attr_mask);
 856 }
 857
 858 int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd,
 859                    struct ib_qp_init_attr *qp_init_attr)
 860 {
 861         struct rdma_id_private *id_priv;
 862         struct ib_qp *qp;
 863         int ret;
 864
 865         id_priv = container_of(id, struct rdma_id_private, id);
 866         if (id->device != pd->device)
 867                 return -EINVAL;
 868
 869         qp_init_attr->port_num = id->port_num;
 870         qp = ib_create_qp(pd, qp_init_attr);
 871         if (IS_ERR(qp))
 872                 return PTR_ERR(qp);
 873
 874         if (id->qp_type == IB_QPT_UD)
 875                 ret = cma_init_ud_qp(id_priv, qp);
 876         else
 877                 ret = cma_init_conn_qp(id_priv, qp);
 878         if (ret)
 879                 goto err;
 880
 881         id->qp = qp;
 882         id_priv->qp_num = qp->qp_num;
 883         id_priv->srq = (qp->srq != NULL);
 884         return 0;
 885 err:
 886         ib_destroy_qp(qp);
 887         return ret;
 888 }
 889 EXPORT_SYMBOL(rdma_create_qp);
 890
 891 void rdma_destroy_qp(struct rdma_cm_id *id)
 892 {
 893         struct rdma_id_private *id_priv;
 894
 895         id_priv = container_of(id, struct rdma_id_private, id);
 896         mutex_lock(&id_priv->qp_mutex);
 897         ib_destroy_qp(id_priv->id.qp);
 898         id_priv->id.qp = NULL;
 899         mutex_unlock(&id_priv->qp_mutex);
 900 }
 901 EXPORT_SYMBOL(rdma_destroy_qp);
 902
 903 static int cma_modify_qp_rtr(struct rdma_id_private *id_priv,
 904                              struct rdma_conn_param *conn_param)
 905 {
 906         struct ib_qp_attr qp_attr;
 907         int qp_attr_mask, ret;
 908
 909         mutex_lock(&id_priv->qp_mutex);
 910         if (!id_priv->id.qp) {
 911                 ret = 0;
 912                 goto out;
 913         }
 914
 915         /* Need to update QP attributes from default values. */
 916         qp_attr.qp_state = IB_QPS_INIT;
 917         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 918         if (ret)
 919                 goto out;
 920
 921         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 922         if (ret)
 923                 goto out;
 924
 925         qp_attr.qp_state = IB_QPS_RTR;
 926         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 927         if (ret)
 928                 goto out;
 929
 930         BUG_ON(id_priv->cma_dev->device != id_priv->id.device);
 931
 932         if (conn_param)
 933                 qp_attr.max_dest_rd_atomic = conn_param->responder_resources;
 934         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 935 out:
 936         mutex_unlock(&id_priv->qp_mutex);
 937         return ret;
 938 }
 939
 940 static int cma_modify_qp_rts(struct rdma_id_private *id_priv,
 941                              struct rdma_conn_param *conn_param)
 942 {
 943         struct ib_qp_attr qp_attr;
 944         int qp_attr_mask, ret;
 945
 946         mutex_lock(&id_priv->qp_mutex);
 947         if (!id_priv->id.qp) {
 948                 ret = 0;
 949                 goto out;
 950         }
 951
 952         qp_attr.qp_state = IB_QPS_RTS;
 953         ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask);
 954         if (ret)
 955                 goto out;
 956
 957         if (conn_param)
 958                 qp_attr.max_rd_atomic = conn_param->initiator_depth;
 959         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, qp_attr_mask);
 960 out:
 961         mutex_unlock(&id_priv->qp_mutex);
 962         return ret;
 963 }
 964
 965 static int cma_modify_qp_err(struct rdma_id_private *id_priv)
 966 {
 967         struct ib_qp_attr qp_attr;
 968         int ret;
 969
 970         mutex_lock(&id_priv->qp_mutex);
 971         if (!id_priv->id.qp) {
 972                 ret = 0;
 973                 goto out;
 974         }
 975
 976         qp_attr.qp_state = IB_QPS_ERR;
 977         ret = ib_modify_qp(id_priv->id.qp, &qp_attr, IB_QP_STATE);
 978 out:
 979         mutex_unlock(&id_priv->qp_mutex);
 980         return ret;
 981 }
 982
 983 static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv,
 984                                struct ib_qp_attr *qp_attr, int *qp_attr_mask)
 985 {
 986         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
 987         int ret;
 988         u16 pkey;
 989
 990         if (rdma_cap_eth_ah(id_priv->id.device, id_priv->id.port_num))
 991                 pkey = 0xffff;
 992         else
 993                 pkey = ib_addr_get_pkey(dev_addr);
 994
 995         ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num,
 996                                   pkey, &qp_attr->pkey_index);
 997         if (ret)
 998                 return ret;
 999
1000         qp_attr->port_num = id_priv->id.port_num;
1001         *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT;
1002
1003         if (id_priv->id.qp_type == IB_QPT_UD) {
1004                 ret = cma_set_qkey(id_priv, 0);
1005                 if (ret)
1006                         return ret;
1007
1008                 qp_attr->qkey = id_priv->qkey;
1009                 *qp_attr_mask |= IB_QP_QKEY;
1010         } else {
1011                 qp_attr->qp_access_flags = 0;
1012                 *qp_attr_mask |= IB_QP_ACCESS_FLAGS;
1013         }
1014         return 0;
1015 }
1016
1017 int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr,
1018                        int *qp_attr_mask)
1019 {
1020         struct rdma_id_private *id_priv;
1021         int ret = 0;
1022
1023         id_priv = container_of(id, struct rdma_id_private, id);
1024         if (rdma_cap_ib_cm(id->device, id->port_num)) {
1025                 if (!id_priv->cm_id.ib || (id_priv->id.qp_type == IB_QPT_UD))
1026                         ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask);
1027                 else
1028                         ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr,
1029                                                  qp_attr_mask);
1030
1031                 if (qp_attr->qp_state == IB_QPS_RTR)
1032                         qp_attr->rq_psn = id_priv->seq_num;
1033         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
1034                 if (!id_priv->cm_id.iw) {
1035                         qp_attr->qp_access_flags = 0;
1036                         *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS;
1037                 } else
1038                         ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr,
1039                                                  qp_attr_mask);
1040                 qp_attr->port_num = id_priv->id.port_num;
1041                 *qp_attr_mask |= IB_QP_PORT;
1042         } else
1043                 ret = -ENOSYS;
1044
1045         return ret;
1046 }
1047 EXPORT_SYMBOL(rdma_init_qp_attr);
1048
1049 static inline int cma_zero_addr(const struct sockaddr *addr)
1050 {
1051         switch (addr->sa_family) {
1052         case AF_INET:
1053                 return ipv4_is_zeronet(((struct sockaddr_in *)addr)->sin_addr.s_addr);
1054         case AF_INET6:
1055                 return ipv6_addr_any(&((struct sockaddr_in6 *) addr)->sin6_addr);
1056         case AF_IB:
1057                 return ib_addr_any(&((struct sockaddr_ib *) addr)->sib_addr);
1058         default:
1059                 return 0;
1060         }
1061 }
1062
1063 static inline int cma_loopback_addr(const struct sockaddr *addr)
1064 {
1065         switch (addr->sa_family) {
1066         case AF_INET:
1067                 return ipv4_is_loopback(((struct sockaddr_in *) addr)->sin_addr.s_addr);
1068         case AF_INET6:
1069                 return ipv6_addr_loopback(&((struct sockaddr_in6 *) addr)->sin6_addr);
1070         case AF_IB:
1071                 return ib_addr_loopback(&((struct sockaddr_ib *) addr)->sib_addr);
1072         default:
1073                 return 0;
1074         }
1075 }
1076
1077 static inline int cma_any_addr(const struct sockaddr *addr)
1078 {
1079         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1080 }
1081
1082 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1083 {
1084         if (src->sa_family != dst->sa_family)
1085                 return -1;
1086
1087         switch (src->sa_family) {
1088         case AF_INET:
1089                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1090                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1091         case AF_INET6:
1092                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1093                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1094         default:
1095                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1096                                    &((struct sockaddr_ib *) dst)->sib_addr);
1097         }
1098 }
1099
1100 static __be16 cma_port(const struct sockaddr *addr)
1101 {
1102         struct sockaddr_ib *sib;
1103
1104         switch (addr->sa_family) {
1105         case AF_INET:
1106                 return ((struct sockaddr_in *) addr)->sin_port;
1107         case AF_INET6:
1108                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1109         case AF_IB:
1110                 sib = (struct sockaddr_ib *) addr;
1111                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1112                                     be64_to_cpu(sib->sib_sid_mask)));
1113         default:
1114                 return 0;
1115         }
1116 }
1117
1118 static inline int cma_any_port(const struct sockaddr *addr)
1119 {
1120         return !cma_port(addr);
1121 }
1122
1123 static void cma_save_ib_info(struct sockaddr *src_addr,
1124                              struct sockaddr *dst_addr,
1125                              const struct rdma_cm_id *listen_id,
1126                              const struct sa_path_rec *path)
1127 {
1128         struct sockaddr_ib *listen_ib, *ib;
1129
1130         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1131         if (src_addr) {
1132                 ib = (struct sockaddr_ib *)src_addr;
1133                 ib->sib_family = AF_IB;
1134                 if (path) {
1135                         ib->sib_pkey = path->pkey;
1136                         ib->sib_flowinfo = path->flow_label;
1137                         memcpy(&ib->sib_addr, &path->sgid, 16);
1138                         ib->sib_sid = path->service_id;
1139                         ib->sib_scope_id = 0;
1140                 } else {
1141                         ib->sib_pkey = listen_ib->sib_pkey;
1142                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1143                         ib->sib_addr = listen_ib->sib_addr;
1144                         ib->sib_sid = listen_ib->sib_sid;
1145                         ib->sib_scope_id = listen_ib->sib_scope_id;
1146                 }
1147                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1148         }
1149         if (dst_addr) {
1150                 ib = (struct sockaddr_ib *)dst_addr;
1151                 ib->sib_family = AF_IB;
1152                 if (path) {
1153                         ib->sib_pkey = path->pkey;
1154                         ib->sib_flowinfo = path->flow_label;
1155                         memcpy(&ib->sib_addr, &path->dgid, 16);
1156                 }
1157         }
1158 }
1159
1160 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1161                               struct sockaddr_in *dst_addr,
1162                               struct cma_hdr *hdr,
1163                               __be16 local_port)
1164 {
1165         if (src_addr) {
1166                 *src_addr = (struct sockaddr_in) {
1167                         .sin_family = AF_INET,
1168                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1169                         .sin_port = local_port,
1170                 };
1171         }
1172
1173         if (dst_addr) {
1174                 *dst_addr = (struct sockaddr_in) {
1175                         .sin_family = AF_INET,
1176                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1177                         .sin_port = hdr->port,
1178                 };
1179         }
1180 }
1181
1182 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1183                               struct sockaddr_in6 *dst_addr,
1184                               struct cma_hdr *hdr,
1185                               __be16 local_port)
1186 {
1187         if (src_addr) {
1188                 *src_addr = (struct sockaddr_in6) {
1189                         .sin6_family = AF_INET6,
1190                         .sin6_addr = hdr->dst_addr.ip6,
1191                         .sin6_port = local_port,
1192                 };
1193         }
1194
1195         if (dst_addr) {
1196                 *dst_addr = (struct sockaddr_in6) {
1197                         .sin6_family = AF_INET6,
1198                         .sin6_addr = hdr->src_addr.ip6,
1199                         .sin6_port = hdr->port,
1200                 };
1201         }
1202 }
1203
1204 static u16 cma_port_from_service_id(__be64 service_id)
1205 {
1206         return (u16)be64_to_cpu(service_id);
1207 }
1208
1209 static int cma_save_ip_info(struct sockaddr *src_addr,
1210                             struct sockaddr *dst_addr,
1211                             const struct ib_cm_event *ib_event,
1212                             __be64 service_id)
1213 {
1214         struct cma_hdr *hdr;
1215         __be16 port;
1216
1217         hdr = ib_event->private_data;
1218         if (hdr->cma_version != CMA_VERSION)
1219                 return -EINVAL;
1220
1221         port = htons(cma_port_from_service_id(service_id));
1222
1223         switch (cma_get_ip_ver(hdr)) {
1224         case 4:
1225                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1226                                   (struct sockaddr_in *)dst_addr, hdr, port);
1227                 break;
1228         case 6:
1229                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1230                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1231                 break;
1232         default:
1233                 return -EAFNOSUPPORT;
1234         }
1235
1236         return 0;
1237 }
1238
1239 static int cma_save_net_info(struct sockaddr *src_addr,
1240                              struct sockaddr *dst_addr,
1241                              const struct rdma_cm_id *listen_id,
1242                              const struct ib_cm_event *ib_event,
1243                              sa_family_t sa_family, __be64 service_id)
1244 {
1245         if (sa_family == AF_IB) {
1246                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1247                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1248                                          ib_event->param.req_rcvd.primary_path);
1249                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1250                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1251                 return 0;
1252         }
1253
1254         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1255 }
1256
1257 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1258                              struct cma_req_info *req)
1259 {
1260         const struct ib_cm_req_event_param *req_param =
1261                 &ib_event->param.req_rcvd;
1262         const struct ib_cm_sidr_req_event_param *sidr_param =
1263                 &ib_event->param.sidr_req_rcvd;
1264
1265         switch (ib_event->event) {
1266         case IB_CM_REQ_RECEIVED:
1267                 req->device     = req_param->listen_id->device;
1268                 req->port       = req_param->port;
1269                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1270                        sizeof(req->local_gid));
1271                 req->has_gid    = true;
1272                 req->service_id = req_param->primary_path->service_id;
1273                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1274                 if (req->pkey != req_param->bth_pkey)
1275                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1276                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1277                                             req_param->bth_pkey, req->pkey);
1278                 break;
1279         case IB_CM_SIDR_REQ_RECEIVED:
1280                 req->device     = sidr_param->listen_id->device;
1281                 req->port       = sidr_param->port;
1282                 req->has_gid    = false;
1283                 req->service_id = sidr_param->service_id;
1284                 req->pkey       = sidr_param->pkey;
1285                 if (req->pkey != sidr_param->bth_pkey)
1286                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1287                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1288                                             sidr_param->bth_pkey, req->pkey);
1289                 break;
1290         default:
1291                 return -EINVAL;
1292         }
1293
1294         return 0;
1295 }
1296
1297 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1298                                   const struct sockaddr_in *dst_addr,
1299                                   const struct sockaddr_in *src_addr)
1300 {
1301         __be32 daddr = dst_addr->sin_addr.s_addr,
1302                saddr = src_addr->sin_addr.s_addr;
1303         struct fib_result res;
1304         struct flowi4 fl4;
1305         int err;
1306         bool ret;
1307
1308         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1309             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1310             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1311             ipv4_is_loopback(saddr))
1312                 return false;
1313
1314         memset(&fl4, 0, sizeof(fl4));
1315         fl4.flowi4_iif = net_dev->ifindex;
1316         fl4.daddr = daddr;
1317         fl4.saddr = saddr;
1318
1319         rcu_read_lock();
1320         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1321         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1322         rcu_read_unlock();
1323
1324         return ret;
1325 }
1326
1327 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1328                                   const struct sockaddr_in6 *dst_addr,
1329                                   const struct sockaddr_in6 *src_addr)
1330 {
1331 #if IS_ENABLED(CONFIG_IPV6)
1332         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1333                            IPV6_ADDR_LINKLOCAL;
1334         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1335                                          &src_addr->sin6_addr, net_dev->ifindex,
1336                                          NULL, strict);
1337         bool ret;
1338
1339         if (!rt)
1340                 return false;
1341
1342         ret = rt->rt6i_idev->dev == net_dev;
1343         ip6_rt_put(rt);
1344
1345         return ret;
1346 #else
1347         return false;
1348 #endif
1349 }
1350
1351 static bool validate_net_dev(struct net_device *net_dev,
1352                              const struct sockaddr *daddr,
1353                              const struct sockaddr *saddr)
1354 {
1355         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1356         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1357         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1358         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1359
1360         switch (daddr->sa_family) {
1361         case AF_INET:
1362                 return saddr->sa_family == AF_INET &&
1363                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1364
1365         case AF_INET6:
1366                 return saddr->sa_family == AF_INET6 &&
1367                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1368
1369         default:
1370                 return false;
1371         }
1372 }
1373
1374 static struct net_device *
1375 roce_get_net_dev_by_cm_event(const struct ib_cm_event *ib_event)
1376 {
1377         const struct ib_gid_attr *sgid_attr = NULL;
1378
1379         if (ib_event->event == IB_CM_REQ_RECEIVED)
1380                 sgid_attr = ib_event->param.req_rcvd.ppath_sgid_attr;
1381         else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1382                 sgid_attr = ib_event->param.sidr_req_rcvd.sgid_attr;
1383
1384         if (!sgid_attr)
1385                 return NULL;
1386         dev_hold(sgid_attr->ndev);
1387         return sgid_attr->ndev;
1388 }
1389
1390 static struct net_device *cma_get_net_dev(const struct ib_cm_event *ib_event,
1391                                           struct cma_req_info *req)
1392 {
1393         struct sockaddr *listen_addr =
1394                         (struct sockaddr *)&req->listen_addr_storage;
1395         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1396         struct net_device *net_dev;
1397         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1398         int err;
1399
1400         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1401                                req->service_id);
1402         if (err)
1403                 return ERR_PTR(err);
1404
1405         if (rdma_protocol_roce(req->device, req->port))
1406                 net_dev = roce_get_net_dev_by_cm_event(ib_event);
1407         else
1408                 net_dev = ib_get_net_dev_by_params(req->device, req->port,
1409                                                    req->pkey,
1410                                                    gid, listen_addr);
1411         if (!net_dev)
1412                 return ERR_PTR(-ENODEV);
1413
1414         return net_dev;
1415 }
1416
1417 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1418 {
1419         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1420 }
1421
1422 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1423                                    const struct cma_hdr *hdr)
1424 {
1425         struct sockaddr *addr = cma_src_addr(id_priv);
1426         __be32 ip4_addr;
1427         struct in6_addr ip6_addr;
1428
1429         if (cma_any_addr(addr) && !id_priv->afonly)
1430                 return true;
1431
1432         switch (addr->sa_family) {
1433         case AF_INET:
1434                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1435                 if (cma_get_ip_ver(hdr) != 4)
1436                         return false;
1437                 if (!cma_any_addr(addr) &&
1438                     hdr->dst_addr.ip4.addr != ip4_addr)
1439                         return false;
1440                 break;
1441         case AF_INET6:
1442                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1443                 if (cma_get_ip_ver(hdr) != 6)
1444                         return false;
1445                 if (!cma_any_addr(addr) &&
1446                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1447                         return false;
1448                 break;
1449         case AF_IB:
1450                 return true;
1451         default:
1452                 return false;
1453         }
1454
1455         return true;
1456 }
1457
1458 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1459 {
1460         struct ib_device *device = id->device;
1461         const int port_num = id->port_num ?: rdma_start_port(device);
1462
1463         return rdma_protocol_roce(device, port_num);
1464 }
1465
1466 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1467                               const struct net_device *net_dev,
1468                               u8 port_num)
1469 {
1470         const struct rdma_addr *addr = &id->route.addr;
1471
1472         if (!net_dev)
1473                 /* This request is an AF_IB request */
1474                 return (!id->port_num || id->port_num == port_num) &&
1475                        (addr->src_addr.ss_family == AF_IB);
1476
1477         /*
1478          * Net namespaces must match, and if the listner is listening
1479          * on a specific netdevice than netdevice must match as well.
1480          */
1481         if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1482             (!!addr->dev_addr.bound_dev_if ==
1483              (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1484                 return true;
1485         else
1486                 return false;
1487 }
1488
1489 static struct rdma_id_private *cma_find_listener(
1490                 const struct rdma_bind_list *bind_list,
1491                 const struct ib_cm_id *cm_id,
1492                 const struct ib_cm_event *ib_event,
1493                 const struct cma_req_info *req,
1494                 const struct net_device *net_dev)
1495 {
1496         struct rdma_id_private *id_priv, *id_priv_dev;
1497
1498         if (!bind_list)
1499                 return ERR_PTR(-EINVAL);
1500
1501         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1502                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1503                         if (id_priv->id.device == cm_id->device &&
1504                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1505                                 return id_priv;
1506                         list_for_each_entry(id_priv_dev,
1507                                             &id_priv->listen_list,
1508                                             listen_list) {
1509                                 if (id_priv_dev->id.device == cm_id->device &&
1510                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1511                                         return id_priv_dev;
1512                         }
1513                 }
1514         }
1515
1516         return ERR_PTR(-EINVAL);
1517 }
1518
1519 static struct rdma_id_private *
1520 cma_id_from_event(struct ib_cm_id *cm_id,
1521                   const struct ib_cm_event *ib_event,
1522                   struct net_device **net_dev)
1523 {
1524         struct cma_req_info req;
1525         struct rdma_bind_list *bind_list;
1526         struct rdma_id_private *id_priv;
1527         int err;
1528
1529         err = cma_save_req_info(ib_event, &req);
1530         if (err)
1531                 return ERR_PTR(err);
1532
1533         *net_dev = cma_get_net_dev(ib_event, &req);
1534         if (IS_ERR(*net_dev)) {
1535                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1536                         /* Assuming the protocol is AF_IB */
1537                         *net_dev = NULL;
1538                 } else {
1539                         return ERR_CAST(*net_dev);
1540                 }
1541         }
1542
1543         /*
1544          * Net namespace might be getting deleted while route lookup,
1545          * cm_id lookup is in progress. Therefore, perform netdevice
1546          * validation, cm_id lookup under rcu lock.
1547          * RCU lock along with netdevice state check, synchronizes with
1548          * netdevice migrating to different net namespace and also avoids
1549          * case where net namespace doesn't get deleted while lookup is in
1550          * progress.
1551          * If the device state is not IFF_UP, its properties such as ifindex
1552          * and nd_net cannot be trusted to remain valid without rcu lock.
1553          * net/core/dev.c change_net_namespace() ensures to synchronize with
1554          * ongoing operations on net device after device is closed using
1555          * synchronize_net().
1556          */
1557         rcu_read_lock();
1558         if (*net_dev) {
1559                 /*
1560                  * If netdevice is down, it is likely that it is administratively
1561                  * down or it might be migrating to different namespace.
1562                  * In that case avoid further processing, as the net namespace
1563                  * or ifindex may change.
1564                  */
1565                 if (((*net_dev)->flags & IFF_UP) == 0) {
1566                         id_priv = ERR_PTR(-EHOSTUNREACH);
1567                         goto err;
1568                 }
1569
1570                 if (!validate_net_dev(*net_dev,
1571                                  (struct sockaddr *)&req.listen_addr_storage,
1572                                  (struct sockaddr *)&req.src_addr_storage)) {
1573                         id_priv = ERR_PTR(-EHOSTUNREACH);
1574                         goto err;
1575                 }
1576         }
1577
1578         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1579                                 rdma_ps_from_service_id(req.service_id),
1580                                 cma_port_from_service_id(req.service_id));
1581         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1582 err:
1583         rcu_read_unlock();
1584         if (IS_ERR(id_priv) && *net_dev) {
1585                 dev_put(*net_dev);
1586                 *net_dev = NULL;
1587         }
1588         return id_priv;
1589 }
1590
1591 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1592 {
1593         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1594 }
1595
1596 static void cma_cancel_route(struct rdma_id_private *id_priv)
1597 {
1598         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1599                 if (id_priv->query)
1600                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1601         }
1602 }
1603
1604 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1605 {
1606         struct rdma_id_private *dev_id_priv;
1607
1608         /*
1609          * Remove from listen_any_list to prevent added devices from spawning
1610          * additional listen requests.
1611          */
1612         mutex_lock(&lock);
1613         list_del(&id_priv->list);
1614
1615         while (!list_empty(&id_priv->listen_list)) {
1616                 dev_id_priv = list_entry(id_priv->listen_list.next,
1617                                          struct rdma_id_private, listen_list);
1618                 /* sync with device removal to avoid duplicate destruction */
1619                 list_del_init(&dev_id_priv->list);
1620                 list_del(&dev_id_priv->listen_list);
1621                 mutex_unlock(&lock);
1622
1623                 rdma_destroy_id(&dev_id_priv->id);
1624                 mutex_lock(&lock);
1625         }
1626         mutex_unlock(&lock);
1627 }
1628
1629 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1630                                  enum rdma_cm_state state)
1631 {
1632         switch (state) {
1633         case RDMA_CM_ADDR_QUERY:
1634                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1635                 break;
1636         case RDMA_CM_ROUTE_QUERY:
1637                 cma_cancel_route(id_priv);
1638                 break;
1639         case RDMA_CM_LISTEN:
1640                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1641                         cma_cancel_listens(id_priv);
1642                 break;
1643         default:
1644                 break;
1645         }
1646 }
1647
1648 static void cma_release_port(struct rdma_id_private *id_priv)
1649 {
1650         struct rdma_bind_list *bind_list = id_priv->bind_list;
1651         struct net *net = id_priv->id.route.addr.dev_addr.net;
1652
1653         if (!bind_list)
1654                 return;
1655
1656         mutex_lock(&lock);
1657         hlist_del(&id_priv->node);
1658         if (hlist_empty(&bind_list->owners)) {
1659                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1660                 kfree(bind_list);
1661         }
1662         mutex_unlock(&lock);
1663 }
1664
1665 static void cma_leave_roce_mc_group(struct rdma_id_private *id_priv,
1666                                     struct cma_multicast *mc)
1667 {
1668         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1669         struct net_device *ndev = NULL;
1670
1671         if (dev_addr->bound_dev_if)
1672                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
1673         if (ndev) {
1674                 cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid, false);
1675                 dev_put(ndev);
1676         }
1677         kref_put(&mc->mcref, release_mc);
1678 }
1679
1680 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1681 {
1682         struct cma_multicast *mc;
1683
1684         while (!list_empty(&id_priv->mc_list)) {
1685                 mc = container_of(id_priv->mc_list.next,
1686                                   struct cma_multicast, list);
1687                 list_del(&mc->list);
1688                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1689                                       id_priv->id.port_num)) {
1690                         ib_sa_free_multicast(mc->multicast.ib);
1691                         kfree(mc);
1692                 } else {
1693                         cma_leave_roce_mc_group(id_priv, mc);
1694                 }
1695         }
1696 }
1697
1698 void rdma_destroy_id(struct rdma_cm_id *id)
1699 {
1700         struct rdma_id_private *id_priv;
1701         enum rdma_cm_state state;
1702
1703         id_priv = container_of(id, struct rdma_id_private, id);
1704         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1705         cma_cancel_operation(id_priv, state);
1706
1707         /*
1708          * Wait for any active callback to finish.  New callbacks will find
1709          * the id_priv state set to destroying and abort.
1710          */
1711         mutex_lock(&id_priv->handler_mutex);
1712         mutex_unlock(&id_priv->handler_mutex);
1713
1714         if (id_priv->cma_dev) {
1715                 rdma_restrack_del(&id_priv->res);
1716                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1717                         if (id_priv->cm_id.ib)
1718                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1719                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1720                         if (id_priv->cm_id.iw)
1721                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1722                 }
1723                 cma_leave_mc_groups(id_priv);
1724                 cma_release_dev(id_priv);
1725         }
1726
1727         cma_release_port(id_priv);
1728         cma_deref_id(id_priv);
1729         wait_for_completion(&id_priv->comp);
1730
1731         if (id_priv->internal_id)
1732                 cma_deref_id(id_priv->id.context);
1733
1734         kfree(id_priv->id.route.path_rec);
1735
1736         if (id_priv->id.route.addr.dev_addr.sgid_attr)
1737                 rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
1738
1739         put_net(id_priv->id.route.addr.dev_addr.net);
1740         kfree(id_priv);
1741 }
1742 EXPORT_SYMBOL(rdma_destroy_id);
1743
1744 static int cma_rep_recv(struct rdma_id_private *id_priv)
1745 {
1746         int ret;
1747
1748         ret = cma_modify_qp_rtr(id_priv, NULL);
1749         if (ret)
1750                 goto reject;
1751
1752         ret = cma_modify_qp_rts(id_priv, NULL);
1753         if (ret)
1754                 goto reject;
1755
1756         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1757         if (ret)
1758                 goto reject;
1759
1760         return 0;
1761 reject:
1762         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1763         cma_modify_qp_err(id_priv);
1764         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1765                        NULL, 0, NULL, 0);
1766         return ret;
1767 }
1768
1769 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1770                                    const struct ib_cm_rep_event_param *rep_data,
1771                                    void *private_data)
1772 {
1773         event->param.conn.private_data = private_data;
1774         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1775         event->param.conn.responder_resources = rep_data->responder_resources;
1776         event->param.conn.initiator_depth = rep_data->initiator_depth;
1777         event->param.conn.flow_control = rep_data->flow_control;
1778         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1779         event->param.conn.srq = rep_data->srq;
1780         event->param.conn.qp_num = rep_data->remote_qpn;
1781 }
1782
1783 static int cma_ib_handler(struct ib_cm_id *cm_id,
1784                           const struct ib_cm_event *ib_event)
1785 {
1786         struct rdma_id_private *id_priv = cm_id->context;
1787         struct rdma_cm_event event;
1788         int ret = 0;
1789
1790         mutex_lock(&id_priv->handler_mutex);
1791         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1792              id_priv->state != RDMA_CM_CONNECT) ||
1793             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1794              id_priv->state != RDMA_CM_DISCONNECT))
1795                 goto out;
1796
1797         memset(&event, 0, sizeof event);
1798         switch (ib_event->event) {
1799         case IB_CM_REQ_ERROR:
1800         case IB_CM_REP_ERROR:
1801                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1802                 event.status = -ETIMEDOUT;
1803                 break;
1804         case IB_CM_REP_RECEIVED:
1805                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1806                     (id_priv->id.qp_type != IB_QPT_UD))
1807                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1808                 if (id_priv->id.qp) {
1809                         event.status = cma_rep_recv(id_priv);
1810                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1811                                                      RDMA_CM_EVENT_ESTABLISHED;
1812                 } else {
1813                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1814                 }
1815                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1816                                        ib_event->private_data);
1817                 break;
1818         case IB_CM_RTU_RECEIVED:
1819         case IB_CM_USER_ESTABLISHED:
1820                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1821                 break;
1822         case IB_CM_DREQ_ERROR:
1823                 event.status = -ETIMEDOUT; /* fall through */
1824         case IB_CM_DREQ_RECEIVED:
1825         case IB_CM_DREP_RECEIVED:
1826                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1827                                    RDMA_CM_DISCONNECT))
1828                         goto out;
1829                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1830                 break;
1831         case IB_CM_TIMEWAIT_EXIT:
1832                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1833                 break;
1834         case IB_CM_MRA_RECEIVED:
1835                 /* ignore event */
1836                 goto out;
1837         case IB_CM_REJ_RECEIVED:
1838                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1839                                                                                 ib_event->param.rej_rcvd.reason));
1840                 cma_modify_qp_err(id_priv);
1841                 event.status = ib_event->param.rej_rcvd.reason;
1842                 event.event = RDMA_CM_EVENT_REJECTED;
1843                 event.param.conn.private_data = ib_event->private_data;
1844                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1845                 break;
1846         default:
1847                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1848                        ib_event->event);
1849                 goto out;
1850         }
1851
1852         ret = id_priv->id.event_handler(&id_priv->id, &event);
1853         if (ret) {
1854                 /* Destroy the CM ID by returning a non-zero value. */
1855                 id_priv->cm_id.ib = NULL;
1856                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1857                 mutex_unlock(&id_priv->handler_mutex);
1858                 rdma_destroy_id(&id_priv->id);
1859                 return ret;
1860         }
1861 out:
1862         mutex_unlock(&id_priv->handler_mutex);
1863         return ret;
1864 }
1865
1866 static struct rdma_id_private *
1867 cma_new_conn_id(const struct rdma_cm_id *listen_id,
1868                 const struct ib_cm_event *ib_event,
1869                 struct net_device *net_dev)
1870 {
1871         struct rdma_id_private *listen_id_priv;
1872         struct rdma_id_private *id_priv;
1873         struct rdma_cm_id *id;
1874         struct rdma_route *rt;
1875         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1876         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1877         const __be64 service_id =
1878                 ib_event->param.req_rcvd.primary_path->service_id;
1879         int ret;
1880
1881         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1882         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1883                             listen_id->event_handler, listen_id->context,
1884                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1885                             listen_id_priv->res.kern_name);
1886         if (IS_ERR(id))
1887                 return NULL;
1888
1889         id_priv = container_of(id, struct rdma_id_private, id);
1890         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1891                               (struct sockaddr *)&id->route.addr.dst_addr,
1892                               listen_id, ib_event, ss_family, service_id))
1893                 goto err;
1894
1895         rt = &id->route;
1896         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1897         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
1898                                      GFP_KERNEL);
1899         if (!rt->path_rec)
1900                 goto err;
1901
1902         rt->path_rec[0] = *path;
1903         if (rt->num_paths == 2)
1904                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1905
1906         if (net_dev) {
1907                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1908         } else {
1909                 if (!cma_protocol_roce(listen_id) &&
1910                     cma_any_addr(cma_src_addr(id_priv))) {
1911                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1912                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1913                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1914                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1915                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1916                         if (ret)
1917                                 goto err;
1918                 }
1919         }
1920         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1921
1922         id_priv->state = RDMA_CM_CONNECT;
1923         return id_priv;
1924
1925 err:
1926         rdma_destroy_id(id);
1927         return NULL;
1928 }
1929
1930 static struct rdma_id_private *
1931 cma_new_udp_id(const struct rdma_cm_id *listen_id,
1932                const struct ib_cm_event *ib_event,
1933                struct net_device *net_dev)
1934 {
1935         const struct rdma_id_private *listen_id_priv;
1936         struct rdma_id_private *id_priv;
1937         struct rdma_cm_id *id;
1938         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1939         struct net *net = listen_id->route.addr.dev_addr.net;
1940         int ret;
1941
1942         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1943         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
1944                               listen_id->ps, IB_QPT_UD,
1945                               listen_id_priv->res.kern_name);
1946         if (IS_ERR(id))
1947                 return NULL;
1948
1949         id_priv = container_of(id, struct rdma_id_private, id);
1950         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1951                               (struct sockaddr *)&id->route.addr.dst_addr,
1952                               listen_id, ib_event, ss_family,
1953                               ib_event->param.sidr_req_rcvd.service_id))
1954                 goto err;
1955
1956         if (net_dev) {
1957                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1958         } else {
1959                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1960                         ret = cma_translate_addr(cma_src_addr(id_priv),
1961                                                  &id->route.addr.dev_addr);
1962                         if (ret)
1963                                 goto err;
1964                 }
1965         }
1966
1967         id_priv->state = RDMA_CM_CONNECT;
1968         return id_priv;
1969 err:
1970         rdma_destroy_id(id);
1971         return NULL;
1972 }
1973
1974 static void cma_set_req_event_data(struct rdma_cm_event *event,
1975                                    const struct ib_cm_req_event_param *req_data,
1976                                    void *private_data, int offset)
1977 {
1978         event->param.conn.private_data = private_data + offset;
1979         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1980         event->param.conn.responder_resources = req_data->responder_resources;
1981         event->param.conn.initiator_depth = req_data->initiator_depth;
1982         event->param.conn.flow_control = req_data->flow_control;
1983         event->param.conn.retry_count = req_data->retry_count;
1984         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1985         event->param.conn.srq = req_data->srq;
1986         event->param.conn.qp_num = req_data->remote_qpn;
1987 }
1988
1989 static int cma_check_req_qp_type(const struct rdma_cm_id *id,
1990                                  const struct ib_cm_event *ib_event)
1991 {
1992         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1993                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1994                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1995                  (id->qp_type == IB_QPT_UD)) ||
1996                 (!id->qp_type));
1997 }
1998
1999 static int cma_req_handler(struct ib_cm_id *cm_id,
2000                            const struct ib_cm_event *ib_event)
2001 {
2002         struct rdma_id_private *listen_id, *conn_id = NULL;
2003         struct rdma_cm_event event;
2004         struct net_device *net_dev;
2005         u8 offset;
2006         int ret;
2007
2008         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
2009         if (IS_ERR(listen_id))
2010                 return PTR_ERR(listen_id);
2011
2012         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
2013                 ret = -EINVAL;
2014                 goto net_dev_put;
2015         }
2016
2017         mutex_lock(&listen_id->handler_mutex);
2018         if (listen_id->state != RDMA_CM_LISTEN) {
2019                 ret = -ECONNABORTED;
2020                 goto err1;
2021         }
2022
2023         memset(&event, 0, sizeof event);
2024         offset = cma_user_data_offset(listen_id);
2025         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2026         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
2027                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
2028                 event.param.ud.private_data = ib_event->private_data + offset;
2029                 event.param.ud.private_data_len =
2030                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
2031         } else {
2032                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
2033                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
2034                                        ib_event->private_data, offset);
2035         }
2036         if (!conn_id) {
2037                 ret = -ENOMEM;
2038                 goto err1;
2039         }
2040
2041         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2042         ret = cma_acquire_dev(conn_id, listen_id);
2043         if (ret)
2044                 goto err2;
2045
2046         conn_id->cm_id.ib = cm_id;
2047         cm_id->context = conn_id;
2048         cm_id->cm_handler = cma_ib_handler;
2049
2050         /*
2051          * Protect against the user destroying conn_id from another thread
2052          * until we're done accessing it.
2053          */
2054         atomic_inc(&conn_id->refcount);
2055         ret = conn_id->id.event_handler(&conn_id->id, &event);
2056         if (ret)
2057                 goto err3;
2058         /*
2059          * Acquire mutex to prevent user executing rdma_destroy_id()
2060          * while we're accessing the cm_id.
2061          */
2062         mutex_lock(&lock);
2063         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2064             (conn_id->id.qp_type != IB_QPT_UD))
2065                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2066         mutex_unlock(&lock);
2067         mutex_unlock(&conn_id->handler_mutex);
2068         mutex_unlock(&listen_id->handler_mutex);
2069         cma_deref_id(conn_id);
2070         if (net_dev)
2071                 dev_put(net_dev);
2072         return 0;
2073
2074 err3:
2075         cma_deref_id(conn_id);
2076         /* Destroy the CM ID by returning a non-zero value. */
2077         conn_id->cm_id.ib = NULL;
2078 err2:
2079         cma_exch(conn_id, RDMA_CM_DESTROYING);
2080         mutex_unlock(&conn_id->handler_mutex);
2081 err1:
2082         mutex_unlock(&listen_id->handler_mutex);
2083         if (conn_id)
2084                 rdma_destroy_id(&conn_id->id);
2085
2086 net_dev_put:
2087         if (net_dev)
2088                 dev_put(net_dev);
2089
2090         return ret;
2091 }
2092
2093 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2094 {
2095         if (addr->sa_family == AF_IB)
2096                 return ((struct sockaddr_ib *) addr)->sib_sid;
2097
2098         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2099 }
2100 EXPORT_SYMBOL(rdma_get_service_id);
2101
2102 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2103                     union ib_gid *dgid)
2104 {
2105         struct rdma_addr *addr = &cm_id->route.addr;
2106
2107         if (!cm_id->device) {
2108                 if (sgid)
2109                         memset(sgid, 0, sizeof(*sgid));
2110                 if (dgid)
2111                         memset(dgid, 0, sizeof(*dgid));
2112                 return;
2113         }
2114
2115         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2116                 if (sgid)
2117                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2118                 if (dgid)
2119                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2120         } else {
2121                 if (sgid)
2122                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2123                 if (dgid)
2124                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2125         }
2126 }
2127 EXPORT_SYMBOL(rdma_read_gids);
2128
2129 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2130 {
2131         struct rdma_id_private *id_priv = iw_id->context;
2132         struct rdma_cm_event event;
2133         int ret = 0;
2134         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2135         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2136
2137         mutex_lock(&id_priv->handler_mutex);
2138         if (id_priv->state != RDMA_CM_CONNECT)
2139                 goto out;
2140
2141         memset(&event, 0, sizeof event);
2142         switch (iw_event->event) {
2143         case IW_CM_EVENT_CLOSE:
2144                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2145                 break;
2146         case IW_CM_EVENT_CONNECT_REPLY:
2147                 memcpy(cma_src_addr(id_priv), laddr,
2148                        rdma_addr_size(laddr));
2149                 memcpy(cma_dst_addr(id_priv), raddr,
2150                        rdma_addr_size(raddr));
2151                 switch (iw_event->status) {
2152                 case 0:
2153                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2154                         event.param.conn.initiator_depth = iw_event->ird;
2155                         event.param.conn.responder_resources = iw_event->ord;
2156                         break;
2157                 case -ECONNRESET:
2158                 case -ECONNREFUSED:
2159                         event.event = RDMA_CM_EVENT_REJECTED;
2160                         break;
2161                 case -ETIMEDOUT:
2162                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2163                         break;
2164                 default:
2165                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2166                         break;
2167                 }
2168                 break;
2169         case IW_CM_EVENT_ESTABLISHED:
2170                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2171                 event.param.conn.initiator_depth = iw_event->ird;
2172                 event.param.conn.responder_resources = iw_event->ord;
2173                 break;
2174         default:
2175                 goto out;
2176         }
2177
2178         event.status = iw_event->status;
2179         event.param.conn.private_data = iw_event->private_data;
2180         event.param.conn.private_data_len = iw_event->private_data_len;
2181         ret = id_priv->id.event_handler(&id_priv->id, &event);
2182         if (ret) {
2183                 /* Destroy the CM ID by returning a non-zero value. */
2184                 id_priv->cm_id.iw = NULL;
2185                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2186                 mutex_unlock(&id_priv->handler_mutex);
2187                 rdma_destroy_id(&id_priv->id);
2188                 return ret;
2189         }
2190
2191 out:
2192         mutex_unlock(&id_priv->handler_mutex);
2193         return ret;
2194 }
2195
2196 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2197                                struct iw_cm_event *iw_event)
2198 {
2199         struct rdma_cm_id *new_cm_id;
2200         struct rdma_id_private *listen_id, *conn_id;
2201         struct rdma_cm_event event;
2202         int ret = -ECONNABORTED;
2203         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2204         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2205
2206         listen_id = cm_id->context;
2207
2208         mutex_lock(&listen_id->handler_mutex);
2209         if (listen_id->state != RDMA_CM_LISTEN)
2210                 goto out;
2211
2212         /* Create a new RDMA id for the new IW CM ID */
2213         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2214                                      listen_id->id.event_handler,
2215                                      listen_id->id.context,
2216                                      RDMA_PS_TCP, IB_QPT_RC,
2217                                      listen_id->res.kern_name);
2218         if (IS_ERR(new_cm_id)) {
2219                 ret = -ENOMEM;
2220                 goto out;
2221         }
2222         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2223         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2224         conn_id->state = RDMA_CM_CONNECT;
2225
2226         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2227         if (ret) {
2228                 mutex_unlock(&conn_id->handler_mutex);
2229                 rdma_destroy_id(new_cm_id);
2230                 goto out;
2231         }
2232
2233         ret = cma_acquire_dev(conn_id, listen_id);
2234         if (ret) {
2235                 mutex_unlock(&conn_id->handler_mutex);
2236                 rdma_destroy_id(new_cm_id);
2237                 goto out;
2238         }
2239
2240         conn_id->cm_id.iw = cm_id;
2241         cm_id->context = conn_id;
2242         cm_id->cm_handler = cma_iw_handler;
2243
2244         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2245         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2246
2247         memset(&event, 0, sizeof event);
2248         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2249         event.param.conn.private_data = iw_event->private_data;
2250         event.param.conn.private_data_len = iw_event->private_data_len;
2251         event.param.conn.initiator_depth = iw_event->ird;
2252         event.param.conn.responder_resources = iw_event->ord;
2253
2254         /*
2255          * Protect against the user destroying conn_id from another thread
2256          * until we're done accessing it.
2257          */
2258         atomic_inc(&conn_id->refcount);
2259         ret = conn_id->id.event_handler(&conn_id->id, &event);
2260         if (ret) {
2261                 /* User wants to destroy the CM ID */
2262                 conn_id->cm_id.iw = NULL;
2263                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2264                 mutex_unlock(&conn_id->handler_mutex);
2265                 cma_deref_id(conn_id);
2266                 rdma_destroy_id(&conn_id->id);
2267                 goto out;
2268         }
2269
2270         mutex_unlock(&conn_id->handler_mutex);
2271         cma_deref_id(conn_id);
2272
2273 out:
2274         mutex_unlock(&listen_id->handler_mutex);
2275         return ret;
2276 }
2277
2278 static int cma_ib_listen(struct rdma_id_private *id_priv)
2279 {
2280         struct sockaddr *addr;
2281         struct ib_cm_id *id;
2282         __be64 svc_id;
2283
2284         addr = cma_src_addr(id_priv);
2285         svc_id = rdma_get_service_id(&id_priv->id, addr);
2286         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2287         if (IS_ERR(id))
2288                 return PTR_ERR(id);
2289         id_priv->cm_id.ib = id;
2290
2291         return 0;
2292 }
2293
2294 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2295 {
2296         int ret;
2297         struct iw_cm_id *id;
2298
2299         id = iw_create_cm_id(id_priv->id.device,
2300                              iw_conn_req_handler,
2301                              id_priv);
2302         if (IS_ERR(id))
2303                 return PTR_ERR(id);
2304
2305         id->tos = id_priv->tos;
2306         id_priv->cm_id.iw = id;
2307
2308         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2309                rdma_addr_size(cma_src_addr(id_priv)));
2310
2311         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2312
2313         if (ret) {
2314                 iw_destroy_cm_id(id_priv->cm_id.iw);
2315                 id_priv->cm_id.iw = NULL;
2316         }
2317
2318         return ret;
2319 }
2320
2321 static int cma_listen_handler(struct rdma_cm_id *id,
2322                               struct rdma_cm_event *event)
2323 {
2324         struct rdma_id_private *id_priv = id->context;
2325
2326         id->context = id_priv->id.context;
2327         id->event_handler = id_priv->id.event_handler;
2328         return id_priv->id.event_handler(id, event);
2329 }
2330
2331 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2332                               struct cma_device *cma_dev)
2333 {
2334         struct rdma_id_private *dev_id_priv;
2335         struct rdma_cm_id *id;
2336         struct net *net = id_priv->id.route.addr.dev_addr.net;
2337         int ret;
2338
2339         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2340                 return;
2341
2342         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2343                               id_priv->id.qp_type, id_priv->res.kern_name);
2344         if (IS_ERR(id))
2345                 return;
2346
2347         dev_id_priv = container_of(id, struct rdma_id_private, id);
2348
2349         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2350         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2351                rdma_addr_size(cma_src_addr(id_priv)));
2352
2353         _cma_attach_to_dev(dev_id_priv, cma_dev);
2354         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2355         atomic_inc(&id_priv->refcount);
2356         dev_id_priv->internal_id = 1;
2357         dev_id_priv->afonly = id_priv->afonly;
2358
2359         ret = rdma_listen(id, id_priv->backlog);
2360         if (ret)
2361                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2362                         ret, cma_dev->device->name);
2363 }
2364
2365 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2366 {
2367         struct cma_device *cma_dev;
2368
2369         mutex_lock(&lock);
2370         list_add_tail(&id_priv->list, &listen_any_list);
2371         list_for_each_entry(cma_dev, &dev_list, list)
2372                 cma_listen_on_dev(id_priv, cma_dev);
2373         mutex_unlock(&lock);
2374 }
2375
2376 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2377 {
2378         struct rdma_id_private *id_priv;
2379
2380         id_priv = container_of(id, struct rdma_id_private, id);
2381         id_priv->tos = (u8) tos;
2382         id_priv->tos_set = true;
2383 }
2384 EXPORT_SYMBOL(rdma_set_service_type);
2385
2386 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2387                               void *context)
2388 {
2389         struct cma_work *work = context;
2390         struct rdma_route *route;
2391
2392         route = &work->id->id.route;
2393
2394         if (!status) {
2395                 route->num_paths = 1;
2396                 *route->path_rec = *path_rec;
2397         } else {
2398                 work->old_state = RDMA_CM_ROUTE_QUERY;
2399                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2400                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2401                 work->event.status = status;
2402                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2403                                      status);
2404         }
2405
2406         queue_work(cma_wq, &work->work);
2407 }
2408
2409 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2410                               struct cma_work *work)
2411 {
2412         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2413         struct sa_path_rec path_rec;
2414         ib_sa_comp_mask comp_mask;
2415         struct sockaddr_in6 *sin6;
2416         struct sockaddr_ib *sib;
2417
2418         memset(&path_rec, 0, sizeof path_rec);
2419
2420         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2421                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2422         else
2423                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2424         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2425         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2426         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2427         path_rec.numb_path = 1;
2428         path_rec.reversible = 1;
2429         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2430                                                   cma_dst_addr(id_priv));
2431
2432         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2433                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2434                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2435
2436         switch (cma_family(id_priv)) {
2437         case AF_INET:
2438                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2439                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2440                 break;
2441         case AF_INET6:
2442                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2443                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2444                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2445                 break;
2446         case AF_IB:
2447                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2448                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2449                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2450                 break;
2451         }
2452
2453         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2454                                                id_priv->id.port_num, &path_rec,
2455                                                comp_mask, timeout_ms,
2456                                                GFP_KERNEL, cma_query_handler,
2457                                                work, &id_priv->query);
2458
2459         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2460 }
2461
2462 static void cma_work_handler(struct work_struct *_work)
2463 {
2464         struct cma_work *work = container_of(_work, struct cma_work, work);
2465         struct rdma_id_private *id_priv = work->id;
2466         int destroy = 0;
2467
2468         mutex_lock(&id_priv->handler_mutex);
2469         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2470                 goto out;
2471
2472         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2473                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2474                 destroy = 1;
2475         }
2476 out:
2477         mutex_unlock(&id_priv->handler_mutex);
2478         cma_deref_id(id_priv);
2479         if (destroy)
2480                 rdma_destroy_id(&id_priv->id);
2481         kfree(work);
2482 }
2483
2484 static void cma_ndev_work_handler(struct work_struct *_work)
2485 {
2486         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2487         struct rdma_id_private *id_priv = work->id;
2488         int destroy = 0;
2489
2490         mutex_lock(&id_priv->handler_mutex);
2491         if (id_priv->state == RDMA_CM_DESTROYING ||
2492             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2493                 goto out;
2494
2495         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2496                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2497                 destroy = 1;
2498         }
2499
2500 out:
2501         mutex_unlock(&id_priv->handler_mutex);
2502         cma_deref_id(id_priv);
2503         if (destroy)
2504                 rdma_destroy_id(&id_priv->id);
2505         kfree(work);
2506 }
2507
2508 static void cma_init_resolve_route_work(struct cma_work *work,
2509                                         struct rdma_id_private *id_priv)
2510 {
2511         work->id = id_priv;
2512         INIT_WORK(&work->work, cma_work_handler);
2513         work->old_state = RDMA_CM_ROUTE_QUERY;
2514         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2515         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2516 }
2517
2518 static void cma_init_resolve_addr_work(struct cma_work *work,
2519                                        struct rdma_id_private *id_priv)
2520 {
2521         work->id = id_priv;
2522         INIT_WORK(&work->work, cma_work_handler);
2523         work->old_state = RDMA_CM_ADDR_QUERY;
2524         work->new_state = RDMA_CM_ADDR_RESOLVED;
2525         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2526 }
2527
2528 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2529 {
2530         struct rdma_route *route = &id_priv->id.route;
2531         struct cma_work *work;
2532         int ret;
2533
2534         work = kzalloc(sizeof *work, GFP_KERNEL);
2535         if (!work)
2536                 return -ENOMEM;
2537
2538         cma_init_resolve_route_work(work, id_priv);
2539
2540         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2541         if (!route->path_rec) {
2542                 ret = -ENOMEM;
2543                 goto err1;
2544         }
2545
2546         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2547         if (ret)
2548                 goto err2;
2549
2550         return 0;
2551 err2:
2552         kfree(route->path_rec);
2553         route->path_rec = NULL;
2554 err1:
2555         kfree(work);
2556         return ret;
2557 }
2558
2559 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2560                                            unsigned long supported_gids,
2561                                            enum ib_gid_type default_gid)
2562 {
2563         if ((network_type == RDMA_NETWORK_IPV4 ||
2564              network_type == RDMA_NETWORK_IPV6) &&
2565             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2566                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2567
2568         return default_gid;
2569 }
2570
2571 /*
2572  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2573  * path record type based on GID type.
2574  * It also sets up other L2 fields which includes destination mac address
2575  * netdev ifindex, of the path record.
2576  * It returns the netdev of the bound interface for this path record entry.
2577  */
2578 static struct net_device *
2579 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2580 {
2581         struct rdma_route *route = &id_priv->id.route;
2582         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2583         struct rdma_addr *addr = &route->addr;
2584         unsigned long supported_gids;
2585         struct net_device *ndev;
2586
2587         if (!addr->dev_addr.bound_dev_if)
2588                 return NULL;
2589
2590         ndev = dev_get_by_index(addr->dev_addr.net,
2591                                 addr->dev_addr.bound_dev_if);
2592         if (!ndev)
2593                 return NULL;
2594
2595         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2596                                                     id_priv->id.port_num);
2597         gid_type = cma_route_gid_type(addr->dev_addr.network,
2598                                       supported_gids,
2599                                       id_priv->gid_type);
2600         /* Use the hint from IP Stack to select GID Type */
2601         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2602                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2603         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2604
2605         route->path_rec->roce.route_resolved = true;
2606         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2607         return ndev;
2608 }
2609
2610 int rdma_set_ib_path(struct rdma_cm_id *id,
2611                      struct sa_path_rec *path_rec)
2612 {
2613         struct rdma_id_private *id_priv;
2614         struct net_device *ndev;
2615         int ret;
2616
2617         id_priv = container_of(id, struct rdma_id_private, id);
2618         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2619                            RDMA_CM_ROUTE_RESOLVED))
2620                 return -EINVAL;
2621
2622         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2623                                      GFP_KERNEL);
2624         if (!id->route.path_rec) {
2625                 ret = -ENOMEM;
2626                 goto err;
2627         }
2628
2629         if (rdma_protocol_roce(id->device, id->port_num)) {
2630                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2631                 if (!ndev) {
2632                         ret = -ENODEV;
2633                         goto err_free;
2634                 }
2635                 dev_put(ndev);
2636         }
2637
2638         id->route.num_paths = 1;
2639         return 0;
2640
2641 err_free:
2642         kfree(id->route.path_rec);
2643         id->route.path_rec = NULL;
2644 err:
2645         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2646         return ret;
2647 }
2648 EXPORT_SYMBOL(rdma_set_ib_path);
2649
2650 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2651 {
2652         struct cma_work *work;
2653
2654         work = kzalloc(sizeof *work, GFP_KERNEL);
2655         if (!work)
2656                 return -ENOMEM;
2657
2658         cma_init_resolve_route_work(work, id_priv);
2659         queue_work(cma_wq, &work->work);
2660         return 0;
2661 }
2662
2663 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2664 {
2665         int prio;
2666         struct net_device *dev;
2667
2668         prio = rt_tos2priority(tos);
2669         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2670         if (dev->num_tc)
2671                 return netdev_get_prio_tc_map(dev, prio);
2672
2673 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2674         if (is_vlan_dev(ndev))
2675                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2676                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2677 #endif
2678         return 0;
2679 }
2680
2681 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2682 {
2683         struct rdma_route *route = &id_priv->id.route;
2684         struct rdma_addr *addr = &route->addr;
2685         struct cma_work *work;
2686         int ret;
2687         struct net_device *ndev;
2688
2689         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2690                                         rdma_start_port(id_priv->cma_dev->device)];
2691         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2692
2693
2694         work = kzalloc(sizeof *work, GFP_KERNEL);
2695         if (!work)
2696                 return -ENOMEM;
2697
2698         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2699         if (!route->path_rec) {
2700                 ret = -ENOMEM;
2701                 goto err1;
2702         }
2703
2704         route->num_paths = 1;
2705
2706         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2707         if (!ndev) {
2708                 ret = -ENODEV;
2709                 goto err2;
2710         }
2711
2712         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2713                     &route->path_rec->sgid);
2714         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2715                     &route->path_rec->dgid);
2716
2717         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2718                 /* TODO: get the hoplimit from the inet/inet6 device */
2719                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2720         else
2721                 route->path_rec->hop_limit = 1;
2722         route->path_rec->reversible = 1;
2723         route->path_rec->pkey = cpu_to_be16(0xffff);
2724         route->path_rec->mtu_selector = IB_SA_EQ;
2725         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2726         route->path_rec->traffic_class = tos;
2727         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2728         route->path_rec->rate_selector = IB_SA_EQ;
2729         route->path_rec->rate = iboe_get_rate(ndev);
2730         dev_put(ndev);
2731         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2732         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2733         if (!route->path_rec->mtu) {
2734                 ret = -EINVAL;
2735                 goto err2;
2736         }
2737
2738         cma_init_resolve_route_work(work, id_priv);
2739         queue_work(cma_wq, &work->work);
2740
2741         return 0;
2742
2743 err2:
2744         kfree(route->path_rec);
2745         route->path_rec = NULL;
2746 err1:
2747         kfree(work);
2748         return ret;
2749 }
2750
2751 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2752 {
2753         struct rdma_id_private *id_priv;
2754         int ret;
2755
2756         id_priv = container_of(id, struct rdma_id_private, id);
2757         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2758                 return -EINVAL;
2759
2760         atomic_inc(&id_priv->refcount);
2761         if (rdma_cap_ib_sa(id->device, id->port_num))
2762                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2763         else if (rdma_protocol_roce(id->device, id->port_num))
2764                 ret = cma_resolve_iboe_route(id_priv);
2765         else if (rdma_protocol_iwarp(id->device, id->port_num))
2766                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2767         else
2768                 ret = -ENOSYS;
2769
2770         if (ret)
2771                 goto err;
2772
2773         return 0;
2774 err:
2775         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2776         cma_deref_id(id_priv);
2777         return ret;
2778 }
2779 EXPORT_SYMBOL(rdma_resolve_route);
2780
2781 static void cma_set_loopback(struct sockaddr *addr)
2782 {
2783         switch (addr->sa_family) {
2784         case AF_INET:
2785                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2786                 break;
2787         case AF_INET6:
2788                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2789                               0, 0, 0, htonl(1));
2790                 break;
2791         default:
2792                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2793                             0, 0, 0, htonl(1));
2794                 break;
2795         }
2796 }
2797
2798 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2799 {
2800         struct cma_device *cma_dev, *cur_dev;
2801         union ib_gid gid;
2802         enum ib_port_state port_state;
2803         u16 pkey;
2804         int ret;
2805         u8 p;
2806
2807         cma_dev = NULL;
2808         mutex_lock(&lock);
2809         list_for_each_entry(cur_dev, &dev_list, list) {
2810                 if (cma_family(id_priv) == AF_IB &&
2811                     !rdma_cap_ib_cm(cur_dev->device, 1))
2812                         continue;
2813
2814                 if (!cma_dev)
2815                         cma_dev = cur_dev;
2816
2817                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2818                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2819                             port_state == IB_PORT_ACTIVE) {
2820                                 cma_dev = cur_dev;
2821                                 goto port_found;
2822                         }
2823                 }
2824         }
2825
2826         if (!cma_dev) {
2827                 ret = -ENODEV;
2828                 goto out;
2829         }
2830
2831         p = 1;
2832
2833 port_found:
2834         ret = rdma_query_gid(cma_dev->device, p, 0, &gid);
2835         if (ret)
2836                 goto out;
2837
2838         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2839         if (ret)
2840                 goto out;
2841
2842         id_priv->id.route.addr.dev_addr.dev_type =
2843                 (rdma_protocol_ib(cma_dev->device, p)) ?
2844                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2845
2846         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2847         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2848         id_priv->id.port_num = p;
2849         cma_attach_to_dev(id_priv, cma_dev);
2850         cma_set_loopback(cma_src_addr(id_priv));
2851 out:
2852         mutex_unlock(&lock);
2853         return ret;
2854 }
2855
2856 static void addr_handler(int status, struct sockaddr *src_addr,
2857                          struct rdma_dev_addr *dev_addr, void *context)
2858 {
2859         struct rdma_id_private *id_priv = context;
2860         struct rdma_cm_event event;
2861
2862         memset(&event, 0, sizeof event);
2863         mutex_lock(&id_priv->handler_mutex);
2864         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2865                            RDMA_CM_ADDR_RESOLVED))
2866                 goto out;
2867
2868         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2869         if (!status && !id_priv->cma_dev) {
2870                 status = cma_acquire_dev(id_priv, NULL);
2871                 if (status)
2872                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2873                                              status);
2874         } else {
2875                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2876         }
2877
2878         if (status) {
2879                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2880                                    RDMA_CM_ADDR_BOUND))
2881                         goto out;
2882                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2883                 event.status = status;
2884         } else
2885                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2886
2887         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2888                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2889                 mutex_unlock(&id_priv->handler_mutex);
2890                 cma_deref_id(id_priv);
2891                 rdma_destroy_id(&id_priv->id);
2892                 return;
2893         }
2894 out:
2895         mutex_unlock(&id_priv->handler_mutex);
2896         cma_deref_id(id_priv);
2897 }
2898
2899 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2900 {
2901         struct cma_work *work;
2902         union ib_gid gid;
2903         int ret;
2904
2905         work = kzalloc(sizeof *work, GFP_KERNEL);
2906         if (!work)
2907                 return -ENOMEM;
2908
2909         if (!id_priv->cma_dev) {
2910                 ret = cma_bind_loopback(id_priv);
2911                 if (ret)
2912                         goto err;
2913         }
2914
2915         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2916         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2917
2918         cma_init_resolve_addr_work(work, id_priv);
2919         queue_work(cma_wq, &work->work);
2920         return 0;
2921 err:
2922         kfree(work);
2923         return ret;
2924 }
2925
2926 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2927 {
2928         struct cma_work *work;
2929         int ret;
2930
2931         work = kzalloc(sizeof *work, GFP_KERNEL);
2932         if (!work)
2933                 return -ENOMEM;
2934
2935         if (!id_priv->cma_dev) {
2936                 ret = cma_resolve_ib_dev(id_priv);
2937                 if (ret)
2938                         goto err;
2939         }
2940
2941         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2942                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2943
2944         cma_init_resolve_addr_work(work, id_priv);
2945         queue_work(cma_wq, &work->work);
2946         return 0;
2947 err:
2948         kfree(work);
2949         return ret;
2950 }
2951
2952 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2953                          const struct sockaddr *dst_addr)
2954 {
2955         if (!src_addr || !src_addr->sa_family) {
2956                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2957                 src_addr->sa_family = dst_addr->sa_family;
2958                 if (IS_ENABLED(CONFIG_IPV6) &&
2959                     dst_addr->sa_family == AF_INET6) {
2960                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2961                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2962                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2963                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2964                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2965                 } else if (dst_addr->sa_family == AF_IB) {
2966                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2967                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2968                 }
2969         }
2970         return rdma_bind_addr(id, src_addr);
2971 }
2972
2973 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2974                       const struct sockaddr *dst_addr, int timeout_ms)
2975 {
2976         struct rdma_id_private *id_priv;
2977         int ret;
2978
2979         id_priv = container_of(id, struct rdma_id_private, id);
2980         if (id_priv->state == RDMA_CM_IDLE) {
2981                 ret = cma_bind_addr(id, src_addr, dst_addr);
2982                 if (ret)
2983                         return ret;
2984         }
2985
2986         if (cma_family(id_priv) != dst_addr->sa_family)
2987                 return -EINVAL;
2988
2989         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2990                 return -EINVAL;
2991
2992         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2993         atomic_inc(&id_priv->refcount);
2994         if (cma_any_addr(dst_addr)) {
2995                 ret = cma_resolve_loopback(id_priv);
2996         } else {
2997                 if (dst_addr->sa_family == AF_IB) {
2998                         ret = cma_resolve_ib_addr(id_priv);
2999                 } else {
3000                         ret = rdma_resolve_ip(cma_src_addr(id_priv),
3001                                               dst_addr, &id->route.addr.dev_addr,
3002                                               timeout_ms, addr_handler, id_priv);
3003                 }
3004         }
3005         if (ret)
3006                 goto err;
3007
3008         return 0;
3009 err:
3010         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
3011         cma_deref_id(id_priv);
3012         return ret;
3013 }
3014 EXPORT_SYMBOL(rdma_resolve_addr);
3015
3016 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
3017 {
3018         struct rdma_id_private *id_priv;
3019         unsigned long flags;
3020         int ret;
3021
3022         id_priv = container_of(id, struct rdma_id_private, id);
3023         spin_lock_irqsave(&id_priv->lock, flags);
3024         if (reuse || id_priv->state == RDMA_CM_IDLE) {
3025                 id_priv->reuseaddr = reuse;
3026                 ret = 0;
3027         } else {
3028                 ret = -EINVAL;
3029         }
3030         spin_unlock_irqrestore(&id_priv->lock, flags);
3031         return ret;
3032 }
3033 EXPORT_SYMBOL(rdma_set_reuseaddr);
3034
3035 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3036 {
3037         struct rdma_id_private *id_priv;
3038         unsigned long flags;
3039         int ret;
3040
3041         id_priv = container_of(id, struct rdma_id_private, id);
3042         spin_lock_irqsave(&id_priv->lock, flags);
3043         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3044                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3045                 id_priv->afonly = afonly;
3046                 ret = 0;
3047         } else {
3048                 ret = -EINVAL;
3049         }
3050         spin_unlock_irqrestore(&id_priv->lock, flags);
3051         return ret;
3052 }
3053 EXPORT_SYMBOL(rdma_set_afonly);
3054
3055 static void cma_bind_port(struct rdma_bind_list *bind_list,
3056                           struct rdma_id_private *id_priv)
3057 {
3058         struct sockaddr *addr;
3059         struct sockaddr_ib *sib;
3060         u64 sid, mask;
3061         __be16 port;
3062
3063         addr = cma_src_addr(id_priv);
3064         port = htons(bind_list->port);
3065
3066         switch (addr->sa_family) {
3067         case AF_INET:
3068                 ((struct sockaddr_in *) addr)->sin_port = port;
3069                 break;
3070         case AF_INET6:
3071                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3072                 break;
3073         case AF_IB:
3074                 sib = (struct sockaddr_ib *) addr;
3075                 sid = be64_to_cpu(sib->sib_sid);
3076                 mask = be64_to_cpu(sib->sib_sid_mask);
3077                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3078                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3079                 break;
3080         }
3081         id_priv->bind_list = bind_list;
3082         hlist_add_head(&id_priv->node, &bind_list->owners);
3083 }
3084
3085 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3086                           struct rdma_id_private *id_priv, unsigned short snum)
3087 {
3088         struct rdma_bind_list *bind_list;
3089         int ret;
3090
3091         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3092         if (!bind_list)
3093                 return -ENOMEM;
3094
3095         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3096                            snum);
3097         if (ret < 0)
3098                 goto err;
3099
3100         bind_list->ps = ps;
3101         bind_list->port = (unsigned short)ret;
3102         cma_bind_port(bind_list, id_priv);
3103         return 0;
3104 err:
3105         kfree(bind_list);
3106         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3107 }
3108
3109 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3110                               struct rdma_id_private *id_priv)
3111 {
3112         struct rdma_id_private *cur_id;
3113         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3114         struct sockaddr  *saddr = cma_src_addr(id_priv);
3115         __be16 dport = cma_port(daddr);
3116
3117         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3118                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3119                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3120                 __be16 cur_dport = cma_port(cur_daddr);
3121
3122                 if (id_priv == cur_id)
3123                         continue;
3124
3125                 /* different dest port -> unique */
3126                 if (!cma_any_port(daddr) &&
3127                     !cma_any_port(cur_daddr) &&
3128                     (dport != cur_dport))
3129                         continue;
3130
3131                 /* different src address -> unique */
3132                 if (!cma_any_addr(saddr) &&
3133                     !cma_any_addr(cur_saddr) &&
3134                     cma_addr_cmp(saddr, cur_saddr))
3135                         continue;
3136
3137                 /* different dst address -> unique */
3138                 if (!cma_any_addr(daddr) &&
3139                     !cma_any_addr(cur_daddr) &&
3140                     cma_addr_cmp(daddr, cur_daddr))
3141                         continue;
3142
3143                 return -EADDRNOTAVAIL;
3144         }
3145         return 0;
3146 }
3147
3148 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3149                               struct rdma_id_private *id_priv)
3150 {
3151         static unsigned int last_used_port;
3152         int low, high, remaining;
3153         unsigned int rover;
3154         struct net *net = id_priv->id.route.addr.dev_addr.net;
3155
3156         inet_get_local_port_range(net, &low, &high);
3157         remaining = (high - low) + 1;
3158         rover = prandom_u32() % remaining + low;
3159 retry:
3160         if (last_used_port != rover) {
3161                 struct rdma_bind_list *bind_list;
3162                 int ret;
3163
3164                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3165
3166                 if (!bind_list) {
3167                         ret = cma_alloc_port(ps, id_priv, rover);
3168                 } else {
3169                         ret = cma_port_is_unique(bind_list, id_priv);
3170                         if (!ret)
3171                                 cma_bind_port(bind_list, id_priv);
3172                 }
3173                 /*
3174                  * Remember previously used port number in order to avoid
3175                  * re-using same port immediately after it is closed.
3176                  */
3177                 if (!ret)
3178                         last_used_port = rover;
3179                 if (ret != -EADDRNOTAVAIL)
3180                         return ret;
3181         }
3182         if (--remaining) {
3183                 rover++;
3184                 if ((rover < low) || (rover > high))
3185                         rover = low;
3186                 goto retry;
3187         }
3188         return -EADDRNOTAVAIL;
3189 }
3190
3191 /*
3192  * Check that the requested port is available.  This is called when trying to
3193  * bind to a specific port, or when trying to listen on a bound port.  In
3194  * the latter case, the provided id_priv may already be on the bind_list, but
3195  * we still need to check that it's okay to start listening.
3196  */
3197 static int cma_check_port(struct rdma_bind_list *bind_list,
3198                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3199 {
3200         struct rdma_id_private *cur_id;
3201         struct sockaddr *addr, *cur_addr;
3202
3203         addr = cma_src_addr(id_priv);
3204         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3205                 if (id_priv == cur_id)
3206                         continue;
3207
3208                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3209                     cur_id->reuseaddr)
3210                         continue;
3211
3212                 cur_addr = cma_src_addr(cur_id);
3213                 if (id_priv->afonly && cur_id->afonly &&
3214                     (addr->sa_family != cur_addr->sa_family))
3215                         continue;
3216
3217                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3218                         return -EADDRNOTAVAIL;
3219
3220                 if (!cma_addr_cmp(addr, cur_addr))
3221                         return -EADDRINUSE;
3222         }
3223         return 0;
3224 }
3225
3226 static int cma_use_port(enum rdma_ucm_port_space ps,
3227                         struct rdma_id_private *id_priv)
3228 {
3229         struct rdma_bind_list *bind_list;
3230         unsigned short snum;
3231         int ret;
3232
3233         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3234         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3235                 return -EACCES;
3236
3237         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3238         if (!bind_list) {
3239                 ret = cma_alloc_port(ps, id_priv, snum);
3240         } else {
3241                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3242                 if (!ret)
3243                         cma_bind_port(bind_list, id_priv);
3244         }
3245         return ret;
3246 }
3247
3248 static int cma_bind_listen(struct rdma_id_private *id_priv)
3249 {
3250         struct rdma_bind_list *bind_list = id_priv->bind_list;
3251         int ret = 0;
3252
3253         mutex_lock(&lock);
3254         if (bind_list->owners.first->next)
3255                 ret = cma_check_port(bind_list, id_priv, 0);
3256         mutex_unlock(&lock);
3257         return ret;
3258 }
3259
3260 static enum rdma_ucm_port_space
3261 cma_select_inet_ps(struct rdma_id_private *id_priv)
3262 {
3263         switch (id_priv->id.ps) {
3264         case RDMA_PS_TCP:
3265         case RDMA_PS_UDP:
3266         case RDMA_PS_IPOIB:
3267         case RDMA_PS_IB:
3268                 return id_priv->id.ps;
3269         default:
3270
3271                 return 0;
3272         }
3273 }
3274
3275 static enum rdma_ucm_port_space
3276 cma_select_ib_ps(struct rdma_id_private *id_priv)
3277 {
3278         enum rdma_ucm_port_space ps = 0;
3279         struct sockaddr_ib *sib;
3280         u64 sid_ps, mask, sid;
3281
3282         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3283         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3284         sid = be64_to_cpu(sib->sib_sid) & mask;
3285
3286         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3287                 sid_ps = RDMA_IB_IP_PS_IB;
3288                 ps = RDMA_PS_IB;
3289         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3290                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3291                 sid_ps = RDMA_IB_IP_PS_TCP;
3292                 ps = RDMA_PS_TCP;
3293         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3294                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3295                 sid_ps = RDMA_IB_IP_PS_UDP;
3296                 ps = RDMA_PS_UDP;
3297         }
3298
3299         if (ps) {
3300                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3301                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3302                                                 be64_to_cpu(sib->sib_sid_mask));
3303         }
3304         return ps;
3305 }
3306
3307 static int cma_get_port(struct rdma_id_private *id_priv)
3308 {
3309         enum rdma_ucm_port_space ps;
3310         int ret;
3311
3312         if (cma_family(id_priv) != AF_IB)
3313                 ps = cma_select_inet_ps(id_priv);
3314         else
3315                 ps = cma_select_ib_ps(id_priv);
3316         if (!ps)
3317                 return -EPROTONOSUPPORT;
3318
3319         mutex_lock(&lock);
3320         if (cma_any_port(cma_src_addr(id_priv)))
3321                 ret = cma_alloc_any_port(ps, id_priv);
3322         else
3323                 ret = cma_use_port(ps, id_priv);
3324         mutex_unlock(&lock);
3325
3326         return ret;
3327 }
3328
3329 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3330                                struct sockaddr *addr)
3331 {
3332 #if IS_ENABLED(CONFIG_IPV6)
3333         struct sockaddr_in6 *sin6;
3334
3335         if (addr->sa_family != AF_INET6)
3336                 return 0;
3337
3338         sin6 = (struct sockaddr_in6 *) addr;
3339
3340         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3341                 return 0;
3342
3343         if (!sin6->sin6_scope_id)
3344                         return -EINVAL;
3345
3346         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3347 #endif
3348         return 0;
3349 }
3350
3351 int rdma_listen(struct rdma_cm_id *id, int backlog)
3352 {
3353         struct rdma_id_private *id_priv;
3354         int ret;
3355
3356         id_priv = container_of(id, struct rdma_id_private, id);
3357         if (id_priv->state == RDMA_CM_IDLE) {
3358                 id->route.addr.src_addr.ss_family = AF_INET;
3359                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3360                 if (ret)
3361                         return ret;
3362         }
3363
3364         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3365                 return -EINVAL;
3366
3367         if (id_priv->reuseaddr) {
3368                 ret = cma_bind_listen(id_priv);
3369                 if (ret)
3370                         goto err;
3371         }
3372
3373         id_priv->backlog = backlog;
3374         if (id->device) {
3375                 if (rdma_cap_ib_cm(id->device, 1)) {
3376                         ret = cma_ib_listen(id_priv);
3377                         if (ret)
3378                                 goto err;
3379                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3380                         ret = cma_iw_listen(id_priv, backlog);
3381                         if (ret)
3382                                 goto err;
3383                 } else {
3384                         ret = -ENOSYS;
3385                         goto err;
3386                 }
3387         } else
3388                 cma_listen_on_all(id_priv);
3389
3390         return 0;
3391 err:
3392         id_priv->backlog = 0;
3393         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3394         return ret;
3395 }
3396 EXPORT_SYMBOL(rdma_listen);
3397
3398 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3399 {
3400         struct rdma_id_private *id_priv;
3401         int ret;
3402         struct sockaddr  *daddr;
3403
3404         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3405             addr->sa_family != AF_IB)
3406                 return -EAFNOSUPPORT;
3407
3408         id_priv = container_of(id, struct rdma_id_private, id);
3409         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3410                 return -EINVAL;
3411
3412         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3413         if (ret)
3414                 goto err1;
3415
3416         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3417         if (!cma_any_addr(addr)) {
3418                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3419                 if (ret)
3420                         goto err1;
3421
3422                 ret = cma_acquire_dev(id_priv, NULL);
3423                 if (ret)
3424                         goto err1;
3425         }
3426
3427         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3428                 if (addr->sa_family == AF_INET)
3429                         id_priv->afonly = 1;
3430 #if IS_ENABLED(CONFIG_IPV6)
3431                 else if (addr->sa_family == AF_INET6) {
3432                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3433
3434                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3435                 }
3436 #endif
3437         }
3438         daddr = cma_dst_addr(id_priv);
3439         daddr->sa_family = addr->sa_family;
3440
3441         ret = cma_get_port(id_priv);
3442         if (ret)
3443                 goto err2;
3444
3445         return 0;
3446 err2:
3447         if (id_priv->cma_dev) {
3448                 rdma_restrack_del(&id_priv->res);
3449                 cma_release_dev(id_priv);
3450         }
3451 err1:
3452         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3453         return ret;
3454 }
3455 EXPORT_SYMBOL(rdma_bind_addr);
3456
3457 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3458 {
3459         struct cma_hdr *cma_hdr;
3460
3461         cma_hdr = hdr;
3462         cma_hdr->cma_version = CMA_VERSION;
3463         if (cma_family(id_priv) == AF_INET) {
3464                 struct sockaddr_in *src4, *dst4;
3465
3466                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3467                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3468
3469                 cma_set_ip_ver(cma_hdr, 4);
3470                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3471                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3472                 cma_hdr->port = src4->sin_port;
3473         } else if (cma_family(id_priv) == AF_INET6) {
3474                 struct sockaddr_in6 *src6, *dst6;
3475
3476                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3477                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3478
3479                 cma_set_ip_ver(cma_hdr, 6);
3480                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3481                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3482                 cma_hdr->port = src6->sin6_port;
3483         }
3484         return 0;
3485 }
3486
3487 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3488                                 const struct ib_cm_event *ib_event)
3489 {
3490         struct rdma_id_private *id_priv = cm_id->context;
3491         struct rdma_cm_event event;
3492         const struct ib_cm_sidr_rep_event_param *rep =
3493                                 &ib_event->param.sidr_rep_rcvd;
3494         int ret = 0;
3495
3496         mutex_lock(&id_priv->handler_mutex);
3497         if (id_priv->state != RDMA_CM_CONNECT)
3498                 goto out;
3499
3500         memset(&event, 0, sizeof event);
3501         switch (ib_event->event) {
3502         case IB_CM_SIDR_REQ_ERROR:
3503                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3504                 event.status = -ETIMEDOUT;
3505                 break;
3506         case IB_CM_SIDR_REP_RECEIVED:
3507                 event.param.ud.private_data = ib_event->private_data;
3508                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3509                 if (rep->status != IB_SIDR_SUCCESS) {
3510                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3511                         event.status = ib_event->param.sidr_rep_rcvd.status;
3512                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3513                                              event.status);
3514                         break;
3515                 }
3516                 ret = cma_set_qkey(id_priv, rep->qkey);
3517                 if (ret) {
3518                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3519                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3520                         event.status = ret;
3521                         break;
3522                 }
3523                 ib_init_ah_attr_from_path(id_priv->id.device,
3524                                           id_priv->id.port_num,
3525                                           id_priv->id.route.path_rec,
3526                                           &event.param.ud.ah_attr,
3527                                           rep->sgid_attr);
3528                 event.param.ud.qp_num = rep->qpn;
3529                 event.param.ud.qkey = rep->qkey;
3530                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3531                 event.status = 0;
3532                 break;
3533         default:
3534                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3535                        ib_event->event);
3536                 goto out;
3537         }
3538
3539         ret = id_priv->id.event_handler(&id_priv->id, &event);
3540
3541         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
3542         if (ret) {
3543                 /* Destroy the CM ID by returning a non-zero value. */
3544                 id_priv->cm_id.ib = NULL;
3545                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3546                 mutex_unlock(&id_priv->handler_mutex);
3547                 rdma_destroy_id(&id_priv->id);
3548                 return ret;
3549         }
3550 out:
3551         mutex_unlock(&id_priv->handler_mutex);
3552         return ret;
3553 }
3554
3555 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3556                               struct rdma_conn_param *conn_param)
3557 {
3558         struct ib_cm_sidr_req_param req;
3559         struct ib_cm_id *id;
3560         void *private_data;
3561         u8 offset;
3562         int ret;
3563
3564         memset(&req, 0, sizeof req);
3565         offset = cma_user_data_offset(id_priv);
3566         req.private_data_len = offset + conn_param->private_data_len;
3567         if (req.private_data_len < conn_param->private_data_len)
3568                 return -EINVAL;
3569
3570         if (req.private_data_len) {
3571                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3572                 if (!private_data)
3573                         return -ENOMEM;
3574         } else {
3575                 private_data = NULL;
3576         }
3577
3578         if (conn_param->private_data && conn_param->private_data_len)
3579                 memcpy(private_data + offset, conn_param->private_data,
3580                        conn_param->private_data_len);
3581
3582         if (private_data) {
3583                 ret = cma_format_hdr(private_data, id_priv);
3584                 if (ret)
3585                         goto out;
3586                 req.private_data = private_data;
3587         }
3588
3589         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3590                              id_priv);
3591         if (IS_ERR(id)) {
3592                 ret = PTR_ERR(id);
3593                 goto out;
3594         }
3595         id_priv->cm_id.ib = id;
3596
3597         req.path = id_priv->id.route.path_rec;
3598         req.sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3599         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3600         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3601         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3602
3603         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3604         if (ret) {
3605                 ib_destroy_cm_id(id_priv->cm_id.ib);
3606                 id_priv->cm_id.ib = NULL;
3607         }
3608 out:
3609         kfree(private_data);
3610         return ret;
3611 }
3612
3613 static int cma_connect_ib(struct rdma_id_private *id_priv,
3614                           struct rdma_conn_param *conn_param)
3615 {
3616         struct ib_cm_req_param req;
3617         struct rdma_route *route;
3618         void *private_data;
3619         struct ib_cm_id *id;
3620         u8 offset;
3621         int ret;
3622
3623         memset(&req, 0, sizeof req);
3624         offset = cma_user_data_offset(id_priv);
3625         req.private_data_len = offset + conn_param->private_data_len;
3626         if (req.private_data_len < conn_param->private_data_len)
3627                 return -EINVAL;
3628
3629         if (req.private_data_len) {
3630                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3631                 if (!private_data)
3632                         return -ENOMEM;
3633         } else {
3634                 private_data = NULL;
3635         }
3636
3637         if (conn_param->private_data && conn_param->private_data_len)
3638                 memcpy(private_data + offset, conn_param->private_data,
3639                        conn_param->private_data_len);
3640
3641         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3642         if (IS_ERR(id)) {
3643                 ret = PTR_ERR(id);
3644                 goto out;
3645         }
3646         id_priv->cm_id.ib = id;
3647
3648         route = &id_priv->id.route;
3649         if (private_data) {
3650                 ret = cma_format_hdr(private_data, id_priv);
3651                 if (ret)
3652                         goto out;
3653                 req.private_data = private_data;
3654         }
3655
3656         req.primary_path = &route->path_rec[0];
3657         if (route->num_paths == 2)
3658                 req.alternate_path = &route->path_rec[1];
3659
3660         req.ppath_sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3661         /* Alternate path SGID attribute currently unsupported */
3662         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3663         req.qp_num = id_priv->qp_num;
3664         req.qp_type = id_priv->id.qp_type;
3665         req.starting_psn = id_priv->seq_num;
3666         req.responder_resources = conn_param->responder_resources;
3667         req.initiator_depth = conn_param->initiator_depth;
3668         req.flow_control = conn_param->flow_control;
3669         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3670         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3671         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3672         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3673         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3674         req.srq = id_priv->srq ? 1 : 0;
3675
3676         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3677 out:
3678         if (ret && !IS_ERR(id)) {
3679                 ib_destroy_cm_id(id);
3680                 id_priv->cm_id.ib = NULL;
3681         }
3682
3683         kfree(private_data);
3684         return ret;
3685 }
3686
3687 static int cma_connect_iw(struct rdma_id_private *id_priv,
3688                           struct rdma_conn_param *conn_param)
3689 {
3690         struct iw_cm_id *cm_id;
3691         int ret;
3692         struct iw_cm_conn_param iw_param;
3693
3694         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3695         if (IS_ERR(cm_id))
3696                 return PTR_ERR(cm_id);
3697
3698         cm_id->tos = id_priv->tos;
3699         id_priv->cm_id.iw = cm_id;
3700
3701         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3702                rdma_addr_size(cma_src_addr(id_priv)));
3703         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3704                rdma_addr_size(cma_dst_addr(id_priv)));
3705
3706         ret = cma_modify_qp_rtr(id_priv, conn_param);
3707         if (ret)
3708                 goto out;
3709
3710         if (conn_param) {
3711                 iw_param.ord = conn_param->initiator_depth;
3712                 iw_param.ird = conn_param->responder_resources;
3713                 iw_param.private_data = conn_param->private_data;
3714                 iw_param.private_data_len = conn_param->private_data_len;
3715                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3716         } else {
3717                 memset(&iw_param, 0, sizeof iw_param);
3718                 iw_param.qpn = id_priv->qp_num;
3719         }
3720         ret = iw_cm_connect(cm_id, &iw_param);
3721 out:
3722         if (ret) {
3723                 iw_destroy_cm_id(cm_id);
3724                 id_priv->cm_id.iw = NULL;
3725         }
3726         return ret;
3727 }
3728
3729 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3730 {
3731         struct rdma_id_private *id_priv;
3732         int ret;
3733
3734         id_priv = container_of(id, struct rdma_id_private, id);
3735         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3736                 return -EINVAL;
3737
3738         if (!id->qp) {
3739                 id_priv->qp_num = conn_param->qp_num;
3740                 id_priv->srq = conn_param->srq;
3741         }
3742
3743         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3744                 if (id->qp_type == IB_QPT_UD)
3745                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3746                 else
3747                         ret = cma_connect_ib(id_priv, conn_param);
3748         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3749                 ret = cma_connect_iw(id_priv, conn_param);
3750         else
3751                 ret = -ENOSYS;
3752         if (ret)
3753                 goto err;
3754
3755         return 0;
3756 err:
3757         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3758         return ret;
3759 }
3760 EXPORT_SYMBOL(rdma_connect);
3761
3762 static int cma_accept_ib(struct rdma_id_private *id_priv,
3763                          struct rdma_conn_param *conn_param)
3764 {
3765         struct ib_cm_rep_param rep;
3766         int ret;
3767
3768         ret = cma_modify_qp_rtr(id_priv, conn_param);
3769         if (ret)
3770                 goto out;
3771
3772         ret = cma_modify_qp_rts(id_priv, conn_param);
3773         if (ret)
3774                 goto out;
3775
3776         memset(&rep, 0, sizeof rep);
3777         rep.qp_num = id_priv->qp_num;
3778         rep.starting_psn = id_priv->seq_num;
3779         rep.private_data = conn_param->private_data;
3780         rep.private_data_len = conn_param->private_data_len;
3781         rep.responder_resources = conn_param->responder_resources;
3782         rep.initiator_depth = conn_param->initiator_depth;
3783         rep.failover_accepted = 0;
3784         rep.flow_control = conn_param->flow_control;
3785         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3786         rep.srq = id_priv->srq ? 1 : 0;
3787
3788         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3789 out:
3790         return ret;
3791 }
3792
3793 static int cma_accept_iw(struct rdma_id_private *id_priv,
3794                   struct rdma_conn_param *conn_param)
3795 {
3796         struct iw_cm_conn_param iw_param;
3797         int ret;
3798
3799         if (!conn_param)
3800                 return -EINVAL;
3801
3802         ret = cma_modify_qp_rtr(id_priv, conn_param);
3803         if (ret)
3804                 return ret;
3805
3806         iw_param.ord = conn_param->initiator_depth;
3807         iw_param.ird = conn_param->responder_resources;
3808         iw_param.private_data = conn_param->private_data;
3809         iw_param.private_data_len = conn_param->private_data_len;
3810         if (id_priv->id.qp) {
3811                 iw_param.qpn = id_priv->qp_num;
3812         } else
3813                 iw_param.qpn = conn_param->qp_num;
3814
3815         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3816 }
3817
3818 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3819                              enum ib_cm_sidr_status status, u32 qkey,
3820                              const void *private_data, int private_data_len)
3821 {
3822         struct ib_cm_sidr_rep_param rep;
3823         int ret;
3824
3825         memset(&rep, 0, sizeof rep);
3826         rep.status = status;
3827         if (status == IB_SIDR_SUCCESS) {
3828                 ret = cma_set_qkey(id_priv, qkey);
3829                 if (ret)
3830                         return ret;
3831                 rep.qp_num = id_priv->qp_num;
3832                 rep.qkey = id_priv->qkey;
3833         }
3834         rep.private_data = private_data;
3835         rep.private_data_len = private_data_len;
3836
3837         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3838 }
3839
3840 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3841                   const char *caller)
3842 {
3843         struct rdma_id_private *id_priv;
3844         int ret;
3845
3846         id_priv = container_of(id, struct rdma_id_private, id);
3847
3848         if (caller)
3849                 id_priv->res.kern_name = caller;
3850         else
3851                 rdma_restrack_set_task(&id_priv->res, current);
3852
3853         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3854                 return -EINVAL;
3855
3856         if (!id->qp && conn_param) {
3857                 id_priv->qp_num = conn_param->qp_num;
3858                 id_priv->srq = conn_param->srq;
3859         }
3860
3861         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3862                 if (id->qp_type == IB_QPT_UD) {
3863                         if (conn_param)
3864                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3865                                                         conn_param->qkey,
3866                                                         conn_param->private_data,
3867                                                         conn_param->private_data_len);
3868                         else
3869                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3870                                                         0, NULL, 0);
3871                 } else {
3872                         if (conn_param)
3873                                 ret = cma_accept_ib(id_priv, conn_param);
3874                         else
3875                                 ret = cma_rep_recv(id_priv);
3876                 }
3877         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3878                 ret = cma_accept_iw(id_priv, conn_param);
3879         else
3880                 ret = -ENOSYS;
3881
3882         if (ret)
3883                 goto reject;
3884
3885         return 0;
3886 reject:
3887         cma_modify_qp_err(id_priv);
3888         rdma_reject(id, NULL, 0);
3889         return ret;
3890 }
3891 EXPORT_SYMBOL(__rdma_accept);
3892
3893 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3894 {
3895         struct rdma_id_private *id_priv;
3896         int ret;
3897
3898         id_priv = container_of(id, struct rdma_id_private, id);
3899         if (!id_priv->cm_id.ib)
3900                 return -EINVAL;
3901
3902         switch (id->device->node_type) {
3903         case RDMA_NODE_IB_CA:
3904                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3905                 break;
3906         default:
3907                 ret = 0;
3908                 break;
3909         }
3910         return ret;
3911 }
3912 EXPORT_SYMBOL(rdma_notify);
3913
3914 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3915                 u8 private_data_len)
3916 {
3917         struct rdma_id_private *id_priv;
3918         int ret;
3919
3920         id_priv = container_of(id, struct rdma_id_private, id);
3921         if (!id_priv->cm_id.ib)
3922                 return -EINVAL;
3923
3924         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3925                 if (id->qp_type == IB_QPT_UD)
3926                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3927                                                 private_data, private_data_len);
3928                 else
3929                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3930                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3931                                              0, private_data, private_data_len);
3932         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3933                 ret = iw_cm_reject(id_priv->cm_id.iw,
3934                                    private_data, private_data_len);
3935         } else
3936                 ret = -ENOSYS;
3937
3938         return ret;
3939 }
3940 EXPORT_SYMBOL(rdma_reject);
3941
3942 int rdma_disconnect(struct rdma_cm_id *id)
3943 {
3944         struct rdma_id_private *id_priv;
3945         int ret;
3946
3947         id_priv = container_of(id, struct rdma_id_private, id);
3948         if (!id_priv->cm_id.ib)
3949                 return -EINVAL;
3950
3951         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3952                 ret = cma_modify_qp_err(id_priv);
3953                 if (ret)
3954                         goto out;
3955                 /* Initiate or respond to a disconnect. */
3956                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3957                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3958         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3959                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3960         } else
3961                 ret = -EINVAL;
3962
3963 out:
3964         return ret;
3965 }
3966 EXPORT_SYMBOL(rdma_disconnect);
3967
3968 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3969 {
3970         struct rdma_id_private *id_priv;
3971         struct cma_multicast *mc = multicast->context;
3972         struct rdma_cm_event event;
3973         int ret = 0;
3974
3975         id_priv = mc->id_priv;
3976         mutex_lock(&id_priv->handler_mutex);
3977         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3978             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3979                 goto out;
3980
3981         if (!status)
3982                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3983         else
3984                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3985                                      status);
3986         mutex_lock(&id_priv->qp_mutex);
3987         if (!status && id_priv->id.qp) {
3988                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3989                                          be16_to_cpu(multicast->rec.mlid));
3990                 if (status)
3991                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3992                                              status);
3993         }
3994         mutex_unlock(&id_priv->qp_mutex);
3995
3996         memset(&event, 0, sizeof event);
3997         event.status = status;
3998         event.param.ud.private_data = mc->context;
3999         if (!status) {
4000                 struct rdma_dev_addr *dev_addr =
4001                         &id_priv->id.route.addr.dev_addr;
4002                 struct net_device *ndev =
4003                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4004                 enum ib_gid_type gid_type =
4005                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4006                         rdma_start_port(id_priv->cma_dev->device)];
4007
4008                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
4009                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
4010                                                id_priv->id.port_num,
4011                                                &multicast->rec,
4012                                                ndev, gid_type,
4013                                                &event.param.ud.ah_attr);
4014                 if (ret)
4015                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4016
4017                 event.param.ud.qp_num = 0xFFFFFF;
4018                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
4019                 if (ndev)
4020                         dev_put(ndev);
4021         } else
4022                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4023
4024         ret = id_priv->id.event_handler(&id_priv->id, &event);
4025
4026         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
4027         if (ret) {
4028                 cma_exch(id_priv, RDMA_CM_DESTROYING);
4029                 mutex_unlock(&id_priv->handler_mutex);
4030                 rdma_destroy_id(&id_priv->id);
4031                 return 0;
4032         }
4033
4034 out:
4035         mutex_unlock(&id_priv->handler_mutex);
4036         return 0;
4037 }
4038
4039 static void cma_set_mgid(struct rdma_id_private *id_priv,
4040                          struct sockaddr *addr, union ib_gid *mgid)
4041 {
4042         unsigned char mc_map[MAX_ADDR_LEN];
4043         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4044         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4045         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4046
4047         if (cma_any_addr(addr)) {
4048                 memset(mgid, 0, sizeof *mgid);
4049         } else if ((addr->sa_family == AF_INET6) &&
4050                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4051                                                                  0xFF10A01B)) {
4052                 /* IPv6 address is an SA assigned MGID. */
4053                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4054         } else if (addr->sa_family == AF_IB) {
4055                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4056         } else if (addr->sa_family == AF_INET6) {
4057                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4058                 if (id_priv->id.ps == RDMA_PS_UDP)
4059                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4060                 *mgid = *(union ib_gid *) (mc_map + 4);
4061         } else {
4062                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4063                 if (id_priv->id.ps == RDMA_PS_UDP)
4064                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4065                 *mgid = *(union ib_gid *) (mc_map + 4);
4066         }
4067 }
4068
4069 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4070                                  struct cma_multicast *mc)
4071 {
4072         struct ib_sa_mcmember_rec rec;
4073         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4074         ib_sa_comp_mask comp_mask;
4075         int ret;
4076
4077         ib_addr_get_mgid(dev_addr, &rec.mgid);
4078         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4079                                      &rec.mgid, &rec);
4080         if (ret)
4081                 return ret;
4082
4083         ret = cma_set_qkey(id_priv, 0);
4084         if (ret)
4085                 return ret;
4086
4087         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4088         rec.qkey = cpu_to_be32(id_priv->qkey);
4089         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4090         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4091         rec.join_state = mc->join_state;
4092
4093         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4094             (!ib_sa_sendonly_fullmem_support(&sa_client,
4095                                              id_priv->id.device,
4096                                              id_priv->id.port_num))) {
4097                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4098                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4099                         id_priv->id.device->name, id_priv->id.port_num);
4100                 return -EOPNOTSUPP;
4101         }
4102
4103         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4104                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4105                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4106                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4107                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4108
4109         if (id_priv->id.ps == RDMA_PS_IPOIB)
4110                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4111                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4112                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4113                              IB_SA_MCMEMBER_REC_MTU |
4114                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4115
4116         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4117                                                 id_priv->id.port_num, &rec,
4118                                                 comp_mask, GFP_KERNEL,
4119                                                 cma_ib_mc_handler, mc);
4120         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4121 }
4122
4123 static void iboe_mcast_work_handler(struct work_struct *work)
4124 {
4125         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4126         struct cma_multicast *mc = mw->mc;
4127         struct ib_sa_multicast *m = mc->multicast.ib;
4128
4129         mc->multicast.ib->context = mc;
4130         cma_ib_mc_handler(0, m);
4131         kref_put(&mc->mcref, release_mc);
4132         kfree(mw);
4133 }
4134
4135 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4136                               enum ib_gid_type gid_type)
4137 {
4138         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4139         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4140
4141         if (cma_any_addr(addr)) {
4142                 memset(mgid, 0, sizeof *mgid);
4143         } else if (addr->sa_family == AF_INET6) {
4144                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4145         } else {
4146                 mgid->raw[0] =
4147                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4148                 mgid->raw[1] =
4149                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4150                 mgid->raw[2] = 0;
4151                 mgid->raw[3] = 0;
4152                 mgid->raw[4] = 0;
4153                 mgid->raw[5] = 0;
4154                 mgid->raw[6] = 0;
4155                 mgid->raw[7] = 0;
4156                 mgid->raw[8] = 0;
4157                 mgid->raw[9] = 0;
4158                 mgid->raw[10] = 0xff;
4159                 mgid->raw[11] = 0xff;
4160                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4161         }
4162 }
4163
4164 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4165                                    struct cma_multicast *mc)
4166 {
4167         struct iboe_mcast_work *work;
4168         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4169         int err = 0;
4170         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4171         struct net_device *ndev = NULL;
4172         enum ib_gid_type gid_type;
4173         bool send_only;
4174
4175         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4176
4177         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4178                 return -EINVAL;
4179
4180         work = kzalloc(sizeof *work, GFP_KERNEL);
4181         if (!work)
4182                 return -ENOMEM;
4183
4184         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4185         if (!mc->multicast.ib) {
4186                 err = -ENOMEM;
4187                 goto out1;
4188         }
4189
4190         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4191                    rdma_start_port(id_priv->cma_dev->device)];
4192         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4193
4194         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4195         if (id_priv->id.ps == RDMA_PS_UDP)
4196                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4197
4198         if (dev_addr->bound_dev_if)
4199                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4200         if (!ndev) {
4201                 err = -ENODEV;
4202                 goto out2;
4203         }
4204         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4205         mc->multicast.ib->rec.hop_limit = 1;
4206         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4207
4208         if (addr->sa_family == AF_INET) {
4209                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4210                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4211                         if (!send_only) {
4212                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4213                                                     true);
4214                         }
4215                 }
4216         } else {
4217                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4218                         err = -ENOTSUPP;
4219         }
4220         dev_put(ndev);
4221         if (err || !mc->multicast.ib->rec.mtu) {
4222                 if (!err)
4223                         err = -EINVAL;
4224                 goto out2;
4225         }
4226         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4227                     &mc->multicast.ib->rec.port_gid);
4228         work->id = id_priv;
4229         work->mc = mc;
4230         INIT_WORK(&work->work, iboe_mcast_work_handler);
4231         kref_get(&mc->mcref);
4232         queue_work(cma_wq, &work->work);
4233
4234         return 0;
4235
4236 out2:
4237         kfree(mc->multicast.ib);
4238 out1:
4239         kfree(work);
4240         return err;
4241 }
4242
4243 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4244                         u8 join_state, void *context)
4245 {
4246         struct rdma_id_private *id_priv;
4247         struct cma_multicast *mc;
4248         int ret;
4249
4250         if (!id->device)
4251                 return -EINVAL;
4252
4253         id_priv = container_of(id, struct rdma_id_private, id);
4254         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4255             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4256                 return -EINVAL;
4257
4258         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4259         if (!mc)
4260                 return -ENOMEM;
4261
4262         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4263         mc->context = context;
4264         mc->id_priv = id_priv;
4265         mc->join_state = join_state;
4266
4267         if (rdma_protocol_roce(id->device, id->port_num)) {
4268                 kref_init(&mc->mcref);
4269                 ret = cma_iboe_join_multicast(id_priv, mc);
4270                 if (ret)
4271                         goto out_err;
4272         } else if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4273                 ret = cma_join_ib_multicast(id_priv, mc);
4274                 if (ret)
4275                         goto out_err;
4276         } else {
4277                 ret = -ENOSYS;
4278                 goto out_err;
4279         }
4280
4281         spin_lock(&id_priv->lock);
4282         list_add(&mc->list, &id_priv->mc_list);
4283         spin_unlock(&id_priv->lock);
4284
4285         return 0;
4286 out_err:
4287         kfree(mc);
4288         return ret;
4289 }
4290 EXPORT_SYMBOL(rdma_join_multicast);
4291
4292 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4293 {
4294         struct rdma_id_private *id_priv;
4295         struct cma_multicast *mc;
4296
4297         id_priv = container_of(id, struct rdma_id_private, id);
4298         spin_lock_irq(&id_priv->lock);
4299         list_for_each_entry(mc, &id_priv->mc_list, list) {
4300                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4301                         list_del(&mc->list);
4302                         spin_unlock_irq(&id_priv->lock);
4303
4304                         if (id->qp)
4305                                 ib_detach_mcast(id->qp,
4306                                                 &mc->multicast.ib->rec.mgid,
4307                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4308
4309                         BUG_ON(id_priv->cma_dev->device != id->device);
4310
4311                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4312                                 ib_sa_free_multicast(mc->multicast.ib);
4313                                 kfree(mc);
4314                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4315                                 cma_leave_roce_mc_group(id_priv, mc);
4316                         }
4317                         return;
4318                 }
4319         }
4320         spin_unlock_irq(&id_priv->lock);
4321 }
4322 EXPORT_SYMBOL(rdma_leave_multicast);
4323
4324 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4325 {
4326         struct rdma_dev_addr *dev_addr;
4327         struct cma_ndev_work *work;
4328
4329         dev_addr = &id_priv->id.route.addr.dev_addr;
4330
4331         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4332             (net_eq(dev_net(ndev), dev_addr->net)) &&
4333             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4334                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4335                         ndev->name, &id_priv->id);
4336                 work = kzalloc(sizeof *work, GFP_KERNEL);
4337                 if (!work)
4338                         return -ENOMEM;
4339
4340                 INIT_WORK(&work->work, cma_ndev_work_handler);
4341                 work->id = id_priv;
4342                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4343                 atomic_inc(&id_priv->refcount);
4344                 queue_work(cma_wq, &work->work);
4345         }
4346
4347         return 0;
4348 }
4349
4350 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4351                                void *ptr)
4352 {
4353         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4354         struct cma_device *cma_dev;
4355         struct rdma_id_private *id_priv;
4356         int ret = NOTIFY_DONE;
4357
4358         if (event != NETDEV_BONDING_FAILOVER)
4359                 return NOTIFY_DONE;
4360
4361         if (!netif_is_bond_master(ndev))
4362                 return NOTIFY_DONE;
4363
4364         mutex_lock(&lock);
4365         list_for_each_entry(cma_dev, &dev_list, list)
4366                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4367                         ret = cma_netdev_change(ndev, id_priv);
4368                         if (ret)
4369                                 goto out;
4370                 }
4371
4372 out:
4373         mutex_unlock(&lock);
4374         return ret;
4375 }
4376
4377 static struct notifier_block cma_nb = {
4378         .notifier_call = cma_netdev_callback
4379 };
4380
4381 static void cma_add_one(struct ib_device *device)
4382 {
4383         struct cma_device *cma_dev;
4384         struct rdma_id_private *id_priv;
4385         unsigned int i;
4386         unsigned long supported_gids = 0;
4387
4388         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4389         if (!cma_dev)
4390                 return;
4391
4392         cma_dev->device = device;
4393         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4394                                             sizeof(*cma_dev->default_gid_type),
4395                                             GFP_KERNEL);
4396         if (!cma_dev->default_gid_type)
4397                 goto free_cma_dev;
4398
4399         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4400                                             sizeof(*cma_dev->default_roce_tos),
4401                                             GFP_KERNEL);
4402         if (!cma_dev->default_roce_tos)
4403                 goto free_gid_type;
4404
4405         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4406                 supported_gids = roce_gid_type_mask_support(device, i);
4407                 WARN_ON(!supported_gids);
4408                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4409                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4410                                 CMA_PREFERRED_ROCE_GID_TYPE;
4411                 else
4412                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4413                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4414                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4415         }
4416
4417         init_completion(&cma_dev->comp);
4418         atomic_set(&cma_dev->refcount, 1);
4419         INIT_LIST_HEAD(&cma_dev->id_list);
4420         ib_set_client_data(device, &cma_client, cma_dev);
4421
4422         mutex_lock(&lock);
4423         list_add_tail(&cma_dev->list, &dev_list);
4424         list_for_each_entry(id_priv, &listen_any_list, list)
4425                 cma_listen_on_dev(id_priv, cma_dev);
4426         mutex_unlock(&lock);
4427
4428         return;
4429
4430 free_gid_type:
4431         kfree(cma_dev->default_gid_type);
4432
4433 free_cma_dev:
4434         kfree(cma_dev);
4435
4436         return;
4437 }
4438
4439 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4440 {
4441         struct rdma_cm_event event;
4442         enum rdma_cm_state state;
4443         int ret = 0;
4444
4445         /* Record that we want to remove the device */
4446         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4447         if (state == RDMA_CM_DESTROYING)
4448                 return 0;
4449
4450         cma_cancel_operation(id_priv, state);
4451         mutex_lock(&id_priv->handler_mutex);
4452
4453         /* Check for destruction from another callback. */
4454         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4455                 goto out;
4456
4457         memset(&event, 0, sizeof event);
4458         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4459         ret = id_priv->id.event_handler(&id_priv->id, &event);
4460 out:
4461         mutex_unlock(&id_priv->handler_mutex);
4462         return ret;
4463 }
4464
4465 static void cma_process_remove(struct cma_device *cma_dev)
4466 {
4467         struct rdma_id_private *id_priv;
4468         int ret;
4469
4470         mutex_lock(&lock);
4471         while (!list_empty(&cma_dev->id_list)) {
4472                 id_priv = list_entry(cma_dev->id_list.next,
4473                                      struct rdma_id_private, list);
4474
4475                 list_del(&id_priv->listen_list);
4476                 list_del_init(&id_priv->list);
4477                 atomic_inc(&id_priv->refcount);
4478                 mutex_unlock(&lock);
4479
4480                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4481                 cma_deref_id(id_priv);
4482                 if (ret)
4483                         rdma_destroy_id(&id_priv->id);
4484
4485                 mutex_lock(&lock);
4486         }
4487         mutex_unlock(&lock);
4488
4489         cma_deref_dev(cma_dev);
4490         wait_for_completion(&cma_dev->comp);
4491 }
4492
4493 static void cma_remove_one(struct ib_device *device, void *client_data)
4494 {
4495         struct cma_device *cma_dev = client_data;
4496
4497         if (!cma_dev)
4498                 return;
4499
4500         mutex_lock(&lock);
4501         list_del(&cma_dev->list);
4502         mutex_unlock(&lock);
4503
4504         cma_process_remove(cma_dev);
4505         kfree(cma_dev->default_roce_tos);
4506         kfree(cma_dev->default_gid_type);
4507         kfree(cma_dev);
4508 }
4509
4510 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4511 {
4512         struct nlmsghdr *nlh;
4513         struct rdma_cm_id_stats *id_stats;
4514         struct rdma_id_private *id_priv;
4515         struct rdma_cm_id *id = NULL;
4516         struct cma_device *cma_dev;
4517         int i_dev = 0, i_id = 0;
4518
4519         /*
4520          * We export all of the IDs as a sequence of messages.  Each
4521          * ID gets its own netlink message.
4522          */
4523         mutex_lock(&lock);
4524
4525         list_for_each_entry(cma_dev, &dev_list, list) {
4526                 if (i_dev < cb->args[0]) {
4527                         i_dev++;
4528                         continue;
4529                 }
4530
4531                 i_id = 0;
4532                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4533                         if (i_id < cb->args[1]) {
4534                                 i_id++;
4535                                 continue;
4536                         }
4537
4538                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4539                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4540                                                 RDMA_NL_RDMA_CM_ID_STATS,
4541                                                 NLM_F_MULTI);
4542                         if (!id_stats)
4543                                 goto out;
4544
4545                         memset(id_stats, 0, sizeof *id_stats);
4546                         id = &id_priv->id;
4547                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4548                         id_stats->port_num = id->port_num;
4549                         id_stats->bound_dev_if =
4550                                 id->route.addr.dev_addr.bound_dev_if;
4551
4552                         if (ibnl_put_attr(skb, nlh,
4553                                           rdma_addr_size(cma_src_addr(id_priv)),
4554                                           cma_src_addr(id_priv),
4555                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4556                                 goto out;
4557                         if (ibnl_put_attr(skb, nlh,
4558                                           rdma_addr_size(cma_dst_addr(id_priv)),
4559                                           cma_dst_addr(id_priv),
4560                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4561                                 goto out;
4562
4563                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4564                         id_stats->port_space    = id->ps;
4565                         id_stats->cm_state      = id_priv->state;
4566                         id_stats->qp_num        = id_priv->qp_num;
4567                         id_stats->qp_type       = id->qp_type;
4568
4569                         i_id++;
4570                         nlmsg_end(skb, nlh);
4571                 }
4572
4573                 cb->args[1] = 0;
4574                 i_dev++;
4575         }
4576
4577 out:
4578         mutex_unlock(&lock);
4579         cb->args[0] = i_dev;
4580         cb->args[1] = i_id;
4581
4582         return skb->len;
4583 }
4584
4585 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4586         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4587 };
4588
4589 static int cma_init_net(struct net *net)
4590 {
4591         struct cma_pernet *pernet = cma_pernet(net);
4592
4593         idr_init(&pernet->tcp_ps);
4594         idr_init(&pernet->udp_ps);
4595         idr_init(&pernet->ipoib_ps);
4596         idr_init(&pernet->ib_ps);
4597
4598         return 0;
4599 }
4600
4601 static void cma_exit_net(struct net *net)
4602 {
4603         struct cma_pernet *pernet = cma_pernet(net);
4604
4605         idr_destroy(&pernet->tcp_ps);
4606         idr_destroy(&pernet->udp_ps);
4607         idr_destroy(&pernet->ipoib_ps);
4608         idr_destroy(&pernet->ib_ps);
4609 }
4610
4611 static struct pernet_operations cma_pernet_operations = {
4612         .init = cma_init_net,
4613         .exit = cma_exit_net,
4614         .id = &cma_pernet_id,
4615         .size = sizeof(struct cma_pernet),
4616 };
4617
4618 static int __init cma_init(void)
4619 {
4620         int ret;
4621
4622         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4623         if (!cma_wq)
4624                 return -ENOMEM;
4625
4626         ret = register_pernet_subsys(&cma_pernet_operations);
4627         if (ret)
4628                 goto err_wq;
4629
4630         ib_sa_register_client(&sa_client);
4631         register_netdevice_notifier(&cma_nb);
4632
4633         ret = ib_register_client(&cma_client);
4634         if (ret)
4635                 goto err;
4636
4637         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4638         cma_configfs_init();
4639
4640         return 0;
4641
4642 err:
4643         unregister_netdevice_notifier(&cma_nb);
4644         ib_sa_unregister_client(&sa_client);
4645 err_wq:
4646         destroy_workqueue(cma_wq);
4647         return ret;
4648 }
4649
4650 static void __exit cma_cleanup(void)
4651 {
4652         cma_configfs_exit();
4653         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4654         ib_unregister_client(&cma_client);
4655         unregister_netdevice_notifier(&cma_nb);
4656         ib_sa_unregister_client(&sa_client);
4657         unregister_pernet_subsys(&cma_pernet_operations);
4658         destroy_workqueue(cma_wq);
4659 }
4660
4661 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4662
4663 module_init(cma_init);
4664 module_exit(cma_cleanup);