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 bool 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 false;
1060         }
1061 }
1062
1063 static inline bool cma_loopback_addr(const struct sockaddr *addr)
1064 {
1065         switch (addr->sa_family) {
1066         case AF_INET:
1067                 return ipv4_is_loopback(
1068                         ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1069         case AF_INET6:
1070                 return ipv6_addr_loopback(
1071                         &((struct sockaddr_in6 *)addr)->sin6_addr);
1072         case AF_IB:
1073                 return ib_addr_loopback(
1074                         &((struct sockaddr_ib *)addr)->sib_addr);
1075         default:
1076                 return false;
1077         }
1078 }
1079
1080 static inline bool cma_any_addr(const struct sockaddr *addr)
1081 {
1082         return cma_zero_addr(addr) || cma_loopback_addr(addr);
1083 }
1084
1085 static int cma_addr_cmp(struct sockaddr *src, struct sockaddr *dst)
1086 {
1087         if (src->sa_family != dst->sa_family)
1088                 return -1;
1089
1090         switch (src->sa_family) {
1091         case AF_INET:
1092                 return ((struct sockaddr_in *) src)->sin_addr.s_addr !=
1093                        ((struct sockaddr_in *) dst)->sin_addr.s_addr;
1094         case AF_INET6:
1095                 return ipv6_addr_cmp(&((struct sockaddr_in6 *) src)->sin6_addr,
1096                                      &((struct sockaddr_in6 *) dst)->sin6_addr);
1097         default:
1098                 return ib_addr_cmp(&((struct sockaddr_ib *) src)->sib_addr,
1099                                    &((struct sockaddr_ib *) dst)->sib_addr);
1100         }
1101 }
1102
1103 static __be16 cma_port(const struct sockaddr *addr)
1104 {
1105         struct sockaddr_ib *sib;
1106
1107         switch (addr->sa_family) {
1108         case AF_INET:
1109                 return ((struct sockaddr_in *) addr)->sin_port;
1110         case AF_INET6:
1111                 return ((struct sockaddr_in6 *) addr)->sin6_port;
1112         case AF_IB:
1113                 sib = (struct sockaddr_ib *) addr;
1114                 return htons((u16) (be64_to_cpu(sib->sib_sid) &
1115                                     be64_to_cpu(sib->sib_sid_mask)));
1116         default:
1117                 return 0;
1118         }
1119 }
1120
1121 static inline int cma_any_port(const struct sockaddr *addr)
1122 {
1123         return !cma_port(addr);
1124 }
1125
1126 static void cma_save_ib_info(struct sockaddr *src_addr,
1127                              struct sockaddr *dst_addr,
1128                              const struct rdma_cm_id *listen_id,
1129                              const struct sa_path_rec *path)
1130 {
1131         struct sockaddr_ib *listen_ib, *ib;
1132
1133         listen_ib = (struct sockaddr_ib *) &listen_id->route.addr.src_addr;
1134         if (src_addr) {
1135                 ib = (struct sockaddr_ib *)src_addr;
1136                 ib->sib_family = AF_IB;
1137                 if (path) {
1138                         ib->sib_pkey = path->pkey;
1139                         ib->sib_flowinfo = path->flow_label;
1140                         memcpy(&ib->sib_addr, &path->sgid, 16);
1141                         ib->sib_sid = path->service_id;
1142                         ib->sib_scope_id = 0;
1143                 } else {
1144                         ib->sib_pkey = listen_ib->sib_pkey;
1145                         ib->sib_flowinfo = listen_ib->sib_flowinfo;
1146                         ib->sib_addr = listen_ib->sib_addr;
1147                         ib->sib_sid = listen_ib->sib_sid;
1148                         ib->sib_scope_id = listen_ib->sib_scope_id;
1149                 }
1150                 ib->sib_sid_mask = cpu_to_be64(0xffffffffffffffffULL);
1151         }
1152         if (dst_addr) {
1153                 ib = (struct sockaddr_ib *)dst_addr;
1154                 ib->sib_family = AF_IB;
1155                 if (path) {
1156                         ib->sib_pkey = path->pkey;
1157                         ib->sib_flowinfo = path->flow_label;
1158                         memcpy(&ib->sib_addr, &path->dgid, 16);
1159                 }
1160         }
1161 }
1162
1163 static void cma_save_ip4_info(struct sockaddr_in *src_addr,
1164                               struct sockaddr_in *dst_addr,
1165                               struct cma_hdr *hdr,
1166                               __be16 local_port)
1167 {
1168         if (src_addr) {
1169                 *src_addr = (struct sockaddr_in) {
1170                         .sin_family = AF_INET,
1171                         .sin_addr.s_addr = hdr->dst_addr.ip4.addr,
1172                         .sin_port = local_port,
1173                 };
1174         }
1175
1176         if (dst_addr) {
1177                 *dst_addr = (struct sockaddr_in) {
1178                         .sin_family = AF_INET,
1179                         .sin_addr.s_addr = hdr->src_addr.ip4.addr,
1180                         .sin_port = hdr->port,
1181                 };
1182         }
1183 }
1184
1185 static void cma_save_ip6_info(struct sockaddr_in6 *src_addr,
1186                               struct sockaddr_in6 *dst_addr,
1187                               struct cma_hdr *hdr,
1188                               __be16 local_port)
1189 {
1190         if (src_addr) {
1191                 *src_addr = (struct sockaddr_in6) {
1192                         .sin6_family = AF_INET6,
1193                         .sin6_addr = hdr->dst_addr.ip6,
1194                         .sin6_port = local_port,
1195                 };
1196         }
1197
1198         if (dst_addr) {
1199                 *dst_addr = (struct sockaddr_in6) {
1200                         .sin6_family = AF_INET6,
1201                         .sin6_addr = hdr->src_addr.ip6,
1202                         .sin6_port = hdr->port,
1203                 };
1204         }
1205 }
1206
1207 static u16 cma_port_from_service_id(__be64 service_id)
1208 {
1209         return (u16)be64_to_cpu(service_id);
1210 }
1211
1212 static int cma_save_ip_info(struct sockaddr *src_addr,
1213                             struct sockaddr *dst_addr,
1214                             const struct ib_cm_event *ib_event,
1215                             __be64 service_id)
1216 {
1217         struct cma_hdr *hdr;
1218         __be16 port;
1219
1220         hdr = ib_event->private_data;
1221         if (hdr->cma_version != CMA_VERSION)
1222                 return -EINVAL;
1223
1224         port = htons(cma_port_from_service_id(service_id));
1225
1226         switch (cma_get_ip_ver(hdr)) {
1227         case 4:
1228                 cma_save_ip4_info((struct sockaddr_in *)src_addr,
1229                                   (struct sockaddr_in *)dst_addr, hdr, port);
1230                 break;
1231         case 6:
1232                 cma_save_ip6_info((struct sockaddr_in6 *)src_addr,
1233                                   (struct sockaddr_in6 *)dst_addr, hdr, port);
1234                 break;
1235         default:
1236                 return -EAFNOSUPPORT;
1237         }
1238
1239         return 0;
1240 }
1241
1242 static int cma_save_net_info(struct sockaddr *src_addr,
1243                              struct sockaddr *dst_addr,
1244                              const struct rdma_cm_id *listen_id,
1245                              const struct ib_cm_event *ib_event,
1246                              sa_family_t sa_family, __be64 service_id)
1247 {
1248         if (sa_family == AF_IB) {
1249                 if (ib_event->event == IB_CM_REQ_RECEIVED)
1250                         cma_save_ib_info(src_addr, dst_addr, listen_id,
1251                                          ib_event->param.req_rcvd.primary_path);
1252                 else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1253                         cma_save_ib_info(src_addr, dst_addr, listen_id, NULL);
1254                 return 0;
1255         }
1256
1257         return cma_save_ip_info(src_addr, dst_addr, ib_event, service_id);
1258 }
1259
1260 static int cma_save_req_info(const struct ib_cm_event *ib_event,
1261                              struct cma_req_info *req)
1262 {
1263         const struct ib_cm_req_event_param *req_param =
1264                 &ib_event->param.req_rcvd;
1265         const struct ib_cm_sidr_req_event_param *sidr_param =
1266                 &ib_event->param.sidr_req_rcvd;
1267
1268         switch (ib_event->event) {
1269         case IB_CM_REQ_RECEIVED:
1270                 req->device     = req_param->listen_id->device;
1271                 req->port       = req_param->port;
1272                 memcpy(&req->local_gid, &req_param->primary_path->sgid,
1273                        sizeof(req->local_gid));
1274                 req->has_gid    = true;
1275                 req->service_id = req_param->primary_path->service_id;
1276                 req->pkey       = be16_to_cpu(req_param->primary_path->pkey);
1277                 if (req->pkey != req_param->bth_pkey)
1278                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and primary path P_Key (0x%x)\n"
1279                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1280                                             req_param->bth_pkey, req->pkey);
1281                 break;
1282         case IB_CM_SIDR_REQ_RECEIVED:
1283                 req->device     = sidr_param->listen_id->device;
1284                 req->port       = sidr_param->port;
1285                 req->has_gid    = false;
1286                 req->service_id = sidr_param->service_id;
1287                 req->pkey       = sidr_param->pkey;
1288                 if (req->pkey != sidr_param->bth_pkey)
1289                         pr_warn_ratelimited("RDMA CMA: got different BTH P_Key (0x%x) and SIDR request payload P_Key (0x%x)\n"
1290                                             "RDMA CMA: in the future this may cause the request to be dropped\n",
1291                                             sidr_param->bth_pkey, req->pkey);
1292                 break;
1293         default:
1294                 return -EINVAL;
1295         }
1296
1297         return 0;
1298 }
1299
1300 static bool validate_ipv4_net_dev(struct net_device *net_dev,
1301                                   const struct sockaddr_in *dst_addr,
1302                                   const struct sockaddr_in *src_addr)
1303 {
1304         __be32 daddr = dst_addr->sin_addr.s_addr,
1305                saddr = src_addr->sin_addr.s_addr;
1306         struct fib_result res;
1307         struct flowi4 fl4;
1308         int err;
1309         bool ret;
1310
1311         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1312             ipv4_is_lbcast(daddr) || ipv4_is_zeronet(saddr) ||
1313             ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr) ||
1314             ipv4_is_loopback(saddr))
1315                 return false;
1316
1317         memset(&fl4, 0, sizeof(fl4));
1318         fl4.flowi4_iif = net_dev->ifindex;
1319         fl4.daddr = daddr;
1320         fl4.saddr = saddr;
1321
1322         rcu_read_lock();
1323         err = fib_lookup(dev_net(net_dev), &fl4, &res, 0);
1324         ret = err == 0 && FIB_RES_DEV(res) == net_dev;
1325         rcu_read_unlock();
1326
1327         return ret;
1328 }
1329
1330 static bool validate_ipv6_net_dev(struct net_device *net_dev,
1331                                   const struct sockaddr_in6 *dst_addr,
1332                                   const struct sockaddr_in6 *src_addr)
1333 {
1334 #if IS_ENABLED(CONFIG_IPV6)
1335         const int strict = ipv6_addr_type(&dst_addr->sin6_addr) &
1336                            IPV6_ADDR_LINKLOCAL;
1337         struct rt6_info *rt = rt6_lookup(dev_net(net_dev), &dst_addr->sin6_addr,
1338                                          &src_addr->sin6_addr, net_dev->ifindex,
1339                                          NULL, strict);
1340         bool ret;
1341
1342         if (!rt)
1343                 return false;
1344
1345         ret = rt->rt6i_idev->dev == net_dev;
1346         ip6_rt_put(rt);
1347
1348         return ret;
1349 #else
1350         return false;
1351 #endif
1352 }
1353
1354 static bool validate_net_dev(struct net_device *net_dev,
1355                              const struct sockaddr *daddr,
1356                              const struct sockaddr *saddr)
1357 {
1358         const struct sockaddr_in *daddr4 = (const struct sockaddr_in *)daddr;
1359         const struct sockaddr_in *saddr4 = (const struct sockaddr_in *)saddr;
1360         const struct sockaddr_in6 *daddr6 = (const struct sockaddr_in6 *)daddr;
1361         const struct sockaddr_in6 *saddr6 = (const struct sockaddr_in6 *)saddr;
1362
1363         switch (daddr->sa_family) {
1364         case AF_INET:
1365                 return saddr->sa_family == AF_INET &&
1366                        validate_ipv4_net_dev(net_dev, daddr4, saddr4);
1367
1368         case AF_INET6:
1369                 return saddr->sa_family == AF_INET6 &&
1370                        validate_ipv6_net_dev(net_dev, daddr6, saddr6);
1371
1372         default:
1373                 return false;
1374         }
1375 }
1376
1377 static struct net_device *
1378 roce_get_net_dev_by_cm_event(const struct ib_cm_event *ib_event)
1379 {
1380         const struct ib_gid_attr *sgid_attr = NULL;
1381
1382         if (ib_event->event == IB_CM_REQ_RECEIVED)
1383                 sgid_attr = ib_event->param.req_rcvd.ppath_sgid_attr;
1384         else if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED)
1385                 sgid_attr = ib_event->param.sidr_req_rcvd.sgid_attr;
1386
1387         if (!sgid_attr)
1388                 return NULL;
1389         dev_hold(sgid_attr->ndev);
1390         return sgid_attr->ndev;
1391 }
1392
1393 static struct net_device *cma_get_net_dev(const struct ib_cm_event *ib_event,
1394                                           struct cma_req_info *req)
1395 {
1396         struct sockaddr *listen_addr =
1397                         (struct sockaddr *)&req->listen_addr_storage;
1398         struct sockaddr *src_addr = (struct sockaddr *)&req->src_addr_storage;
1399         struct net_device *net_dev;
1400         const union ib_gid *gid = req->has_gid ? &req->local_gid : NULL;
1401         int err;
1402
1403         err = cma_save_ip_info(listen_addr, src_addr, ib_event,
1404                                req->service_id);
1405         if (err)
1406                 return ERR_PTR(err);
1407
1408         if (rdma_protocol_roce(req->device, req->port))
1409                 net_dev = roce_get_net_dev_by_cm_event(ib_event);
1410         else
1411                 net_dev = ib_get_net_dev_by_params(req->device, req->port,
1412                                                    req->pkey,
1413                                                    gid, listen_addr);
1414         if (!net_dev)
1415                 return ERR_PTR(-ENODEV);
1416
1417         return net_dev;
1418 }
1419
1420 static enum rdma_ucm_port_space rdma_ps_from_service_id(__be64 service_id)
1421 {
1422         return (be64_to_cpu(service_id) >> 16) & 0xffff;
1423 }
1424
1425 static bool cma_match_private_data(struct rdma_id_private *id_priv,
1426                                    const struct cma_hdr *hdr)
1427 {
1428         struct sockaddr *addr = cma_src_addr(id_priv);
1429         __be32 ip4_addr;
1430         struct in6_addr ip6_addr;
1431
1432         if (cma_any_addr(addr) && !id_priv->afonly)
1433                 return true;
1434
1435         switch (addr->sa_family) {
1436         case AF_INET:
1437                 ip4_addr = ((struct sockaddr_in *)addr)->sin_addr.s_addr;
1438                 if (cma_get_ip_ver(hdr) != 4)
1439                         return false;
1440                 if (!cma_any_addr(addr) &&
1441                     hdr->dst_addr.ip4.addr != ip4_addr)
1442                         return false;
1443                 break;
1444         case AF_INET6:
1445                 ip6_addr = ((struct sockaddr_in6 *)addr)->sin6_addr;
1446                 if (cma_get_ip_ver(hdr) != 6)
1447                         return false;
1448                 if (!cma_any_addr(addr) &&
1449                     memcmp(&hdr->dst_addr.ip6, &ip6_addr, sizeof(ip6_addr)))
1450                         return false;
1451                 break;
1452         case AF_IB:
1453                 return true;
1454         default:
1455                 return false;
1456         }
1457
1458         return true;
1459 }
1460
1461 static bool cma_protocol_roce(const struct rdma_cm_id *id)
1462 {
1463         struct ib_device *device = id->device;
1464         const int port_num = id->port_num ?: rdma_start_port(device);
1465
1466         return rdma_protocol_roce(device, port_num);
1467 }
1468
1469 static bool cma_match_net_dev(const struct rdma_cm_id *id,
1470                               const struct net_device *net_dev,
1471                               u8 port_num)
1472 {
1473         const struct rdma_addr *addr = &id->route.addr;
1474
1475         if (!net_dev)
1476                 /* This request is an AF_IB request */
1477                 return (!id->port_num || id->port_num == port_num) &&
1478                        (addr->src_addr.ss_family == AF_IB);
1479
1480         /*
1481          * Net namespaces must match, and if the listner is listening
1482          * on a specific netdevice than netdevice must match as well.
1483          */
1484         if (net_eq(dev_net(net_dev), addr->dev_addr.net) &&
1485             (!!addr->dev_addr.bound_dev_if ==
1486              (addr->dev_addr.bound_dev_if == net_dev->ifindex)))
1487                 return true;
1488         else
1489                 return false;
1490 }
1491
1492 static struct rdma_id_private *cma_find_listener(
1493                 const struct rdma_bind_list *bind_list,
1494                 const struct ib_cm_id *cm_id,
1495                 const struct ib_cm_event *ib_event,
1496                 const struct cma_req_info *req,
1497                 const struct net_device *net_dev)
1498 {
1499         struct rdma_id_private *id_priv, *id_priv_dev;
1500
1501         if (!bind_list)
1502                 return ERR_PTR(-EINVAL);
1503
1504         hlist_for_each_entry(id_priv, &bind_list->owners, node) {
1505                 if (cma_match_private_data(id_priv, ib_event->private_data)) {
1506                         if (id_priv->id.device == cm_id->device &&
1507                             cma_match_net_dev(&id_priv->id, net_dev, req->port))
1508                                 return id_priv;
1509                         list_for_each_entry(id_priv_dev,
1510                                             &id_priv->listen_list,
1511                                             listen_list) {
1512                                 if (id_priv_dev->id.device == cm_id->device &&
1513                                     cma_match_net_dev(&id_priv_dev->id, net_dev, req->port))
1514                                         return id_priv_dev;
1515                         }
1516                 }
1517         }
1518
1519         return ERR_PTR(-EINVAL);
1520 }
1521
1522 static struct rdma_id_private *
1523 cma_id_from_event(struct ib_cm_id *cm_id,
1524                   const struct ib_cm_event *ib_event,
1525                   struct net_device **net_dev)
1526 {
1527         struct cma_req_info req;
1528         struct rdma_bind_list *bind_list;
1529         struct rdma_id_private *id_priv;
1530         int err;
1531
1532         err = cma_save_req_info(ib_event, &req);
1533         if (err)
1534                 return ERR_PTR(err);
1535
1536         *net_dev = cma_get_net_dev(ib_event, &req);
1537         if (IS_ERR(*net_dev)) {
1538                 if (PTR_ERR(*net_dev) == -EAFNOSUPPORT) {
1539                         /* Assuming the protocol is AF_IB */
1540                         *net_dev = NULL;
1541                 } else {
1542                         return ERR_CAST(*net_dev);
1543                 }
1544         }
1545
1546         /*
1547          * Net namespace might be getting deleted while route lookup,
1548          * cm_id lookup is in progress. Therefore, perform netdevice
1549          * validation, cm_id lookup under rcu lock.
1550          * RCU lock along with netdevice state check, synchronizes with
1551          * netdevice migrating to different net namespace and also avoids
1552          * case where net namespace doesn't get deleted while lookup is in
1553          * progress.
1554          * If the device state is not IFF_UP, its properties such as ifindex
1555          * and nd_net cannot be trusted to remain valid without rcu lock.
1556          * net/core/dev.c change_net_namespace() ensures to synchronize with
1557          * ongoing operations on net device after device is closed using
1558          * synchronize_net().
1559          */
1560         rcu_read_lock();
1561         if (*net_dev) {
1562                 /*
1563                  * If netdevice is down, it is likely that it is administratively
1564                  * down or it might be migrating to different namespace.
1565                  * In that case avoid further processing, as the net namespace
1566                  * or ifindex may change.
1567                  */
1568                 if (((*net_dev)->flags & IFF_UP) == 0) {
1569                         id_priv = ERR_PTR(-EHOSTUNREACH);
1570                         goto err;
1571                 }
1572
1573                 if (!validate_net_dev(*net_dev,
1574                                  (struct sockaddr *)&req.listen_addr_storage,
1575                                  (struct sockaddr *)&req.src_addr_storage)) {
1576                         id_priv = ERR_PTR(-EHOSTUNREACH);
1577                         goto err;
1578                 }
1579         }
1580
1581         bind_list = cma_ps_find(*net_dev ? dev_net(*net_dev) : &init_net,
1582                                 rdma_ps_from_service_id(req.service_id),
1583                                 cma_port_from_service_id(req.service_id));
1584         id_priv = cma_find_listener(bind_list, cm_id, ib_event, &req, *net_dev);
1585 err:
1586         rcu_read_unlock();
1587         if (IS_ERR(id_priv) && *net_dev) {
1588                 dev_put(*net_dev);
1589                 *net_dev = NULL;
1590         }
1591         return id_priv;
1592 }
1593
1594 static inline u8 cma_user_data_offset(struct rdma_id_private *id_priv)
1595 {
1596         return cma_family(id_priv) == AF_IB ? 0 : sizeof(struct cma_hdr);
1597 }
1598
1599 static void cma_cancel_route(struct rdma_id_private *id_priv)
1600 {
1601         if (rdma_cap_ib_sa(id_priv->id.device, id_priv->id.port_num)) {
1602                 if (id_priv->query)
1603                         ib_sa_cancel_query(id_priv->query_id, id_priv->query);
1604         }
1605 }
1606
1607 static void cma_cancel_listens(struct rdma_id_private *id_priv)
1608 {
1609         struct rdma_id_private *dev_id_priv;
1610
1611         /*
1612          * Remove from listen_any_list to prevent added devices from spawning
1613          * additional listen requests.
1614          */
1615         mutex_lock(&lock);
1616         list_del(&id_priv->list);
1617
1618         while (!list_empty(&id_priv->listen_list)) {
1619                 dev_id_priv = list_entry(id_priv->listen_list.next,
1620                                          struct rdma_id_private, listen_list);
1621                 /* sync with device removal to avoid duplicate destruction */
1622                 list_del_init(&dev_id_priv->list);
1623                 list_del(&dev_id_priv->listen_list);
1624                 mutex_unlock(&lock);
1625
1626                 rdma_destroy_id(&dev_id_priv->id);
1627                 mutex_lock(&lock);
1628         }
1629         mutex_unlock(&lock);
1630 }
1631
1632 static void cma_cancel_operation(struct rdma_id_private *id_priv,
1633                                  enum rdma_cm_state state)
1634 {
1635         switch (state) {
1636         case RDMA_CM_ADDR_QUERY:
1637                 rdma_addr_cancel(&id_priv->id.route.addr.dev_addr);
1638                 break;
1639         case RDMA_CM_ROUTE_QUERY:
1640                 cma_cancel_route(id_priv);
1641                 break;
1642         case RDMA_CM_LISTEN:
1643                 if (cma_any_addr(cma_src_addr(id_priv)) && !id_priv->cma_dev)
1644                         cma_cancel_listens(id_priv);
1645                 break;
1646         default:
1647                 break;
1648         }
1649 }
1650
1651 static void cma_release_port(struct rdma_id_private *id_priv)
1652 {
1653         struct rdma_bind_list *bind_list = id_priv->bind_list;
1654         struct net *net = id_priv->id.route.addr.dev_addr.net;
1655
1656         if (!bind_list)
1657                 return;
1658
1659         mutex_lock(&lock);
1660         hlist_del(&id_priv->node);
1661         if (hlist_empty(&bind_list->owners)) {
1662                 cma_ps_remove(net, bind_list->ps, bind_list->port);
1663                 kfree(bind_list);
1664         }
1665         mutex_unlock(&lock);
1666 }
1667
1668 static void cma_leave_roce_mc_group(struct rdma_id_private *id_priv,
1669                                     struct cma_multicast *mc)
1670 {
1671         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
1672         struct net_device *ndev = NULL;
1673
1674         if (dev_addr->bound_dev_if)
1675                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
1676         if (ndev) {
1677                 cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid, false);
1678                 dev_put(ndev);
1679         }
1680         kref_put(&mc->mcref, release_mc);
1681 }
1682
1683 static void cma_leave_mc_groups(struct rdma_id_private *id_priv)
1684 {
1685         struct cma_multicast *mc;
1686
1687         while (!list_empty(&id_priv->mc_list)) {
1688                 mc = container_of(id_priv->mc_list.next,
1689                                   struct cma_multicast, list);
1690                 list_del(&mc->list);
1691                 if (rdma_cap_ib_mcast(id_priv->cma_dev->device,
1692                                       id_priv->id.port_num)) {
1693                         ib_sa_free_multicast(mc->multicast.ib);
1694                         kfree(mc);
1695                 } else {
1696                         cma_leave_roce_mc_group(id_priv, mc);
1697                 }
1698         }
1699 }
1700
1701 void rdma_destroy_id(struct rdma_cm_id *id)
1702 {
1703         struct rdma_id_private *id_priv;
1704         enum rdma_cm_state state;
1705
1706         id_priv = container_of(id, struct rdma_id_private, id);
1707         state = cma_exch(id_priv, RDMA_CM_DESTROYING);
1708         cma_cancel_operation(id_priv, state);
1709
1710         /*
1711          * Wait for any active callback to finish.  New callbacks will find
1712          * the id_priv state set to destroying and abort.
1713          */
1714         mutex_lock(&id_priv->handler_mutex);
1715         mutex_unlock(&id_priv->handler_mutex);
1716
1717         if (id_priv->cma_dev) {
1718                 rdma_restrack_del(&id_priv->res);
1719                 if (rdma_cap_ib_cm(id_priv->id.device, 1)) {
1720                         if (id_priv->cm_id.ib)
1721                                 ib_destroy_cm_id(id_priv->cm_id.ib);
1722                 } else if (rdma_cap_iw_cm(id_priv->id.device, 1)) {
1723                         if (id_priv->cm_id.iw)
1724                                 iw_destroy_cm_id(id_priv->cm_id.iw);
1725                 }
1726                 cma_leave_mc_groups(id_priv);
1727                 cma_release_dev(id_priv);
1728         }
1729
1730         cma_release_port(id_priv);
1731         cma_deref_id(id_priv);
1732         wait_for_completion(&id_priv->comp);
1733
1734         if (id_priv->internal_id)
1735                 cma_deref_id(id_priv->id.context);
1736
1737         kfree(id_priv->id.route.path_rec);
1738
1739         if (id_priv->id.route.addr.dev_addr.sgid_attr)
1740                 rdma_put_gid_attr(id_priv->id.route.addr.dev_addr.sgid_attr);
1741
1742         put_net(id_priv->id.route.addr.dev_addr.net);
1743         kfree(id_priv);
1744 }
1745 EXPORT_SYMBOL(rdma_destroy_id);
1746
1747 static int cma_rep_recv(struct rdma_id_private *id_priv)
1748 {
1749         int ret;
1750
1751         ret = cma_modify_qp_rtr(id_priv, NULL);
1752         if (ret)
1753                 goto reject;
1754
1755         ret = cma_modify_qp_rts(id_priv, NULL);
1756         if (ret)
1757                 goto reject;
1758
1759         ret = ib_send_cm_rtu(id_priv->cm_id.ib, NULL, 0);
1760         if (ret)
1761                 goto reject;
1762
1763         return 0;
1764 reject:
1765         pr_debug_ratelimited("RDMA CM: CONNECT_ERROR: failed to handle reply. status %d\n", ret);
1766         cma_modify_qp_err(id_priv);
1767         ib_send_cm_rej(id_priv->cm_id.ib, IB_CM_REJ_CONSUMER_DEFINED,
1768                        NULL, 0, NULL, 0);
1769         return ret;
1770 }
1771
1772 static void cma_set_rep_event_data(struct rdma_cm_event *event,
1773                                    const struct ib_cm_rep_event_param *rep_data,
1774                                    void *private_data)
1775 {
1776         event->param.conn.private_data = private_data;
1777         event->param.conn.private_data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
1778         event->param.conn.responder_resources = rep_data->responder_resources;
1779         event->param.conn.initiator_depth = rep_data->initiator_depth;
1780         event->param.conn.flow_control = rep_data->flow_control;
1781         event->param.conn.rnr_retry_count = rep_data->rnr_retry_count;
1782         event->param.conn.srq = rep_data->srq;
1783         event->param.conn.qp_num = rep_data->remote_qpn;
1784 }
1785
1786 static int cma_ib_handler(struct ib_cm_id *cm_id,
1787                           const struct ib_cm_event *ib_event)
1788 {
1789         struct rdma_id_private *id_priv = cm_id->context;
1790         struct rdma_cm_event event;
1791         int ret = 0;
1792
1793         mutex_lock(&id_priv->handler_mutex);
1794         if ((ib_event->event != IB_CM_TIMEWAIT_EXIT &&
1795              id_priv->state != RDMA_CM_CONNECT) ||
1796             (ib_event->event == IB_CM_TIMEWAIT_EXIT &&
1797              id_priv->state != RDMA_CM_DISCONNECT))
1798                 goto out;
1799
1800         memset(&event, 0, sizeof event);
1801         switch (ib_event->event) {
1802         case IB_CM_REQ_ERROR:
1803         case IB_CM_REP_ERROR:
1804                 event.event = RDMA_CM_EVENT_UNREACHABLE;
1805                 event.status = -ETIMEDOUT;
1806                 break;
1807         case IB_CM_REP_RECEIVED:
1808                 if (cma_comp(id_priv, RDMA_CM_CONNECT) &&
1809                     (id_priv->id.qp_type != IB_QPT_UD))
1810                         ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
1811                 if (id_priv->id.qp) {
1812                         event.status = cma_rep_recv(id_priv);
1813                         event.event = event.status ? RDMA_CM_EVENT_CONNECT_ERROR :
1814                                                      RDMA_CM_EVENT_ESTABLISHED;
1815                 } else {
1816                         event.event = RDMA_CM_EVENT_CONNECT_RESPONSE;
1817                 }
1818                 cma_set_rep_event_data(&event, &ib_event->param.rep_rcvd,
1819                                        ib_event->private_data);
1820                 break;
1821         case IB_CM_RTU_RECEIVED:
1822         case IB_CM_USER_ESTABLISHED:
1823                 event.event = RDMA_CM_EVENT_ESTABLISHED;
1824                 break;
1825         case IB_CM_DREQ_ERROR:
1826                 event.status = -ETIMEDOUT; /* fall through */
1827         case IB_CM_DREQ_RECEIVED:
1828         case IB_CM_DREP_RECEIVED:
1829                 if (!cma_comp_exch(id_priv, RDMA_CM_CONNECT,
1830                                    RDMA_CM_DISCONNECT))
1831                         goto out;
1832                 event.event = RDMA_CM_EVENT_DISCONNECTED;
1833                 break;
1834         case IB_CM_TIMEWAIT_EXIT:
1835                 event.event = RDMA_CM_EVENT_TIMEWAIT_EXIT;
1836                 break;
1837         case IB_CM_MRA_RECEIVED:
1838                 /* ignore event */
1839                 goto out;
1840         case IB_CM_REJ_RECEIVED:
1841                 pr_debug_ratelimited("RDMA CM: REJECTED: %s\n", rdma_reject_msg(&id_priv->id,
1842                                                                                 ib_event->param.rej_rcvd.reason));
1843                 cma_modify_qp_err(id_priv);
1844                 event.status = ib_event->param.rej_rcvd.reason;
1845                 event.event = RDMA_CM_EVENT_REJECTED;
1846                 event.param.conn.private_data = ib_event->private_data;
1847                 event.param.conn.private_data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
1848                 break;
1849         default:
1850                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
1851                        ib_event->event);
1852                 goto out;
1853         }
1854
1855         ret = id_priv->id.event_handler(&id_priv->id, &event);
1856         if (ret) {
1857                 /* Destroy the CM ID by returning a non-zero value. */
1858                 id_priv->cm_id.ib = NULL;
1859                 cma_exch(id_priv, RDMA_CM_DESTROYING);
1860                 mutex_unlock(&id_priv->handler_mutex);
1861                 rdma_destroy_id(&id_priv->id);
1862                 return ret;
1863         }
1864 out:
1865         mutex_unlock(&id_priv->handler_mutex);
1866         return ret;
1867 }
1868
1869 static struct rdma_id_private *
1870 cma_new_conn_id(const struct rdma_cm_id *listen_id,
1871                 const struct ib_cm_event *ib_event,
1872                 struct net_device *net_dev)
1873 {
1874         struct rdma_id_private *listen_id_priv;
1875         struct rdma_id_private *id_priv;
1876         struct rdma_cm_id *id;
1877         struct rdma_route *rt;
1878         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1879         struct sa_path_rec *path = ib_event->param.req_rcvd.primary_path;
1880         const __be64 service_id =
1881                 ib_event->param.req_rcvd.primary_path->service_id;
1882         int ret;
1883
1884         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1885         id = __rdma_create_id(listen_id->route.addr.dev_addr.net,
1886                             listen_id->event_handler, listen_id->context,
1887                             listen_id->ps, ib_event->param.req_rcvd.qp_type,
1888                             listen_id_priv->res.kern_name);
1889         if (IS_ERR(id))
1890                 return NULL;
1891
1892         id_priv = container_of(id, struct rdma_id_private, id);
1893         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1894                               (struct sockaddr *)&id->route.addr.dst_addr,
1895                               listen_id, ib_event, ss_family, service_id))
1896                 goto err;
1897
1898         rt = &id->route;
1899         rt->num_paths = ib_event->param.req_rcvd.alternate_path ? 2 : 1;
1900         rt->path_rec = kmalloc_array(rt->num_paths, sizeof(*rt->path_rec),
1901                                      GFP_KERNEL);
1902         if (!rt->path_rec)
1903                 goto err;
1904
1905         rt->path_rec[0] = *path;
1906         if (rt->num_paths == 2)
1907                 rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path;
1908
1909         if (net_dev) {
1910                 rdma_copy_addr(&rt->addr.dev_addr, net_dev, NULL);
1911         } else {
1912                 if (!cma_protocol_roce(listen_id) &&
1913                     cma_any_addr(cma_src_addr(id_priv))) {
1914                         rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND;
1915                         rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid);
1916                         ib_addr_set_pkey(&rt->addr.dev_addr, be16_to_cpu(rt->path_rec[0].pkey));
1917                 } else if (!cma_any_addr(cma_src_addr(id_priv))) {
1918                         ret = cma_translate_addr(cma_src_addr(id_priv), &rt->addr.dev_addr);
1919                         if (ret)
1920                                 goto err;
1921                 }
1922         }
1923         rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid);
1924
1925         id_priv->state = RDMA_CM_CONNECT;
1926         return id_priv;
1927
1928 err:
1929         rdma_destroy_id(id);
1930         return NULL;
1931 }
1932
1933 static struct rdma_id_private *
1934 cma_new_udp_id(const struct rdma_cm_id *listen_id,
1935                const struct ib_cm_event *ib_event,
1936                struct net_device *net_dev)
1937 {
1938         const struct rdma_id_private *listen_id_priv;
1939         struct rdma_id_private *id_priv;
1940         struct rdma_cm_id *id;
1941         const sa_family_t ss_family = listen_id->route.addr.src_addr.ss_family;
1942         struct net *net = listen_id->route.addr.dev_addr.net;
1943         int ret;
1944
1945         listen_id_priv = container_of(listen_id, struct rdma_id_private, id);
1946         id = __rdma_create_id(net, listen_id->event_handler, listen_id->context,
1947                               listen_id->ps, IB_QPT_UD,
1948                               listen_id_priv->res.kern_name);
1949         if (IS_ERR(id))
1950                 return NULL;
1951
1952         id_priv = container_of(id, struct rdma_id_private, id);
1953         if (cma_save_net_info((struct sockaddr *)&id->route.addr.src_addr,
1954                               (struct sockaddr *)&id->route.addr.dst_addr,
1955                               listen_id, ib_event, ss_family,
1956                               ib_event->param.sidr_req_rcvd.service_id))
1957                 goto err;
1958
1959         if (net_dev) {
1960                 rdma_copy_addr(&id->route.addr.dev_addr, net_dev, NULL);
1961         } else {
1962                 if (!cma_any_addr(cma_src_addr(id_priv))) {
1963                         ret = cma_translate_addr(cma_src_addr(id_priv),
1964                                                  &id->route.addr.dev_addr);
1965                         if (ret)
1966                                 goto err;
1967                 }
1968         }
1969
1970         id_priv->state = RDMA_CM_CONNECT;
1971         return id_priv;
1972 err:
1973         rdma_destroy_id(id);
1974         return NULL;
1975 }
1976
1977 static void cma_set_req_event_data(struct rdma_cm_event *event,
1978                                    const struct ib_cm_req_event_param *req_data,
1979                                    void *private_data, int offset)
1980 {
1981         event->param.conn.private_data = private_data + offset;
1982         event->param.conn.private_data_len = IB_CM_REQ_PRIVATE_DATA_SIZE - offset;
1983         event->param.conn.responder_resources = req_data->responder_resources;
1984         event->param.conn.initiator_depth = req_data->initiator_depth;
1985         event->param.conn.flow_control = req_data->flow_control;
1986         event->param.conn.retry_count = req_data->retry_count;
1987         event->param.conn.rnr_retry_count = req_data->rnr_retry_count;
1988         event->param.conn.srq = req_data->srq;
1989         event->param.conn.qp_num = req_data->remote_qpn;
1990 }
1991
1992 static int cma_check_req_qp_type(const struct rdma_cm_id *id,
1993                                  const struct ib_cm_event *ib_event)
1994 {
1995         return (((ib_event->event == IB_CM_REQ_RECEIVED) &&
1996                  (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
1997                 ((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
1998                  (id->qp_type == IB_QPT_UD)) ||
1999                 (!id->qp_type));
2000 }
2001
2002 static int cma_req_handler(struct ib_cm_id *cm_id,
2003                            const struct ib_cm_event *ib_event)
2004 {
2005         struct rdma_id_private *listen_id, *conn_id = NULL;
2006         struct rdma_cm_event event;
2007         struct net_device *net_dev;
2008         u8 offset;
2009         int ret;
2010
2011         listen_id = cma_id_from_event(cm_id, ib_event, &net_dev);
2012         if (IS_ERR(listen_id))
2013                 return PTR_ERR(listen_id);
2014
2015         if (!cma_check_req_qp_type(&listen_id->id, ib_event)) {
2016                 ret = -EINVAL;
2017                 goto net_dev_put;
2018         }
2019
2020         mutex_lock(&listen_id->handler_mutex);
2021         if (listen_id->state != RDMA_CM_LISTEN) {
2022                 ret = -ECONNABORTED;
2023                 goto err1;
2024         }
2025
2026         memset(&event, 0, sizeof event);
2027         offset = cma_user_data_offset(listen_id);
2028         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2029         if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
2030                 conn_id = cma_new_udp_id(&listen_id->id, ib_event, net_dev);
2031                 event.param.ud.private_data = ib_event->private_data + offset;
2032                 event.param.ud.private_data_len =
2033                                 IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE - offset;
2034         } else {
2035                 conn_id = cma_new_conn_id(&listen_id->id, ib_event, net_dev);
2036                 cma_set_req_event_data(&event, &ib_event->param.req_rcvd,
2037                                        ib_event->private_data, offset);
2038         }
2039         if (!conn_id) {
2040                 ret = -ENOMEM;
2041                 goto err1;
2042         }
2043
2044         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2045         ret = cma_acquire_dev(conn_id, listen_id);
2046         if (ret)
2047                 goto err2;
2048
2049         conn_id->cm_id.ib = cm_id;
2050         cm_id->context = conn_id;
2051         cm_id->cm_handler = cma_ib_handler;
2052
2053         /*
2054          * Protect against the user destroying conn_id from another thread
2055          * until we're done accessing it.
2056          */
2057         atomic_inc(&conn_id->refcount);
2058         ret = conn_id->id.event_handler(&conn_id->id, &event);
2059         if (ret)
2060                 goto err3;
2061         /*
2062          * Acquire mutex to prevent user executing rdma_destroy_id()
2063          * while we're accessing the cm_id.
2064          */
2065         mutex_lock(&lock);
2066         if (cma_comp(conn_id, RDMA_CM_CONNECT) &&
2067             (conn_id->id.qp_type != IB_QPT_UD))
2068                 ib_send_cm_mra(cm_id, CMA_CM_MRA_SETTING, NULL, 0);
2069         mutex_unlock(&lock);
2070         mutex_unlock(&conn_id->handler_mutex);
2071         mutex_unlock(&listen_id->handler_mutex);
2072         cma_deref_id(conn_id);
2073         if (net_dev)
2074                 dev_put(net_dev);
2075         return 0;
2076
2077 err3:
2078         cma_deref_id(conn_id);
2079         /* Destroy the CM ID by returning a non-zero value. */
2080         conn_id->cm_id.ib = NULL;
2081 err2:
2082         cma_exch(conn_id, RDMA_CM_DESTROYING);
2083         mutex_unlock(&conn_id->handler_mutex);
2084 err1:
2085         mutex_unlock(&listen_id->handler_mutex);
2086         if (conn_id)
2087                 rdma_destroy_id(&conn_id->id);
2088
2089 net_dev_put:
2090         if (net_dev)
2091                 dev_put(net_dev);
2092
2093         return ret;
2094 }
2095
2096 __be64 rdma_get_service_id(struct rdma_cm_id *id, struct sockaddr *addr)
2097 {
2098         if (addr->sa_family == AF_IB)
2099                 return ((struct sockaddr_ib *) addr)->sib_sid;
2100
2101         return cpu_to_be64(((u64)id->ps << 16) + be16_to_cpu(cma_port(addr)));
2102 }
2103 EXPORT_SYMBOL(rdma_get_service_id);
2104
2105 void rdma_read_gids(struct rdma_cm_id *cm_id, union ib_gid *sgid,
2106                     union ib_gid *dgid)
2107 {
2108         struct rdma_addr *addr = &cm_id->route.addr;
2109
2110         if (!cm_id->device) {
2111                 if (sgid)
2112                         memset(sgid, 0, sizeof(*sgid));
2113                 if (dgid)
2114                         memset(dgid, 0, sizeof(*dgid));
2115                 return;
2116         }
2117
2118         if (rdma_protocol_roce(cm_id->device, cm_id->port_num)) {
2119                 if (sgid)
2120                         rdma_ip2gid((struct sockaddr *)&addr->src_addr, sgid);
2121                 if (dgid)
2122                         rdma_ip2gid((struct sockaddr *)&addr->dst_addr, dgid);
2123         } else {
2124                 if (sgid)
2125                         rdma_addr_get_sgid(&addr->dev_addr, sgid);
2126                 if (dgid)
2127                         rdma_addr_get_dgid(&addr->dev_addr, dgid);
2128         }
2129 }
2130 EXPORT_SYMBOL(rdma_read_gids);
2131
2132 static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
2133 {
2134         struct rdma_id_private *id_priv = iw_id->context;
2135         struct rdma_cm_event event;
2136         int ret = 0;
2137         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2138         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2139
2140         mutex_lock(&id_priv->handler_mutex);
2141         if (id_priv->state != RDMA_CM_CONNECT)
2142                 goto out;
2143
2144         memset(&event, 0, sizeof event);
2145         switch (iw_event->event) {
2146         case IW_CM_EVENT_CLOSE:
2147                 event.event = RDMA_CM_EVENT_DISCONNECTED;
2148                 break;
2149         case IW_CM_EVENT_CONNECT_REPLY:
2150                 memcpy(cma_src_addr(id_priv), laddr,
2151                        rdma_addr_size(laddr));
2152                 memcpy(cma_dst_addr(id_priv), raddr,
2153                        rdma_addr_size(raddr));
2154                 switch (iw_event->status) {
2155                 case 0:
2156                         event.event = RDMA_CM_EVENT_ESTABLISHED;
2157                         event.param.conn.initiator_depth = iw_event->ird;
2158                         event.param.conn.responder_resources = iw_event->ord;
2159                         break;
2160                 case -ECONNRESET:
2161                 case -ECONNREFUSED:
2162                         event.event = RDMA_CM_EVENT_REJECTED;
2163                         break;
2164                 case -ETIMEDOUT:
2165                         event.event = RDMA_CM_EVENT_UNREACHABLE;
2166                         break;
2167                 default:
2168                         event.event = RDMA_CM_EVENT_CONNECT_ERROR;
2169                         break;
2170                 }
2171                 break;
2172         case IW_CM_EVENT_ESTABLISHED:
2173                 event.event = RDMA_CM_EVENT_ESTABLISHED;
2174                 event.param.conn.initiator_depth = iw_event->ird;
2175                 event.param.conn.responder_resources = iw_event->ord;
2176                 break;
2177         default:
2178                 goto out;
2179         }
2180
2181         event.status = iw_event->status;
2182         event.param.conn.private_data = iw_event->private_data;
2183         event.param.conn.private_data_len = iw_event->private_data_len;
2184         ret = id_priv->id.event_handler(&id_priv->id, &event);
2185         if (ret) {
2186                 /* Destroy the CM ID by returning a non-zero value. */
2187                 id_priv->cm_id.iw = NULL;
2188                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2189                 mutex_unlock(&id_priv->handler_mutex);
2190                 rdma_destroy_id(&id_priv->id);
2191                 return ret;
2192         }
2193
2194 out:
2195         mutex_unlock(&id_priv->handler_mutex);
2196         return ret;
2197 }
2198
2199 static int iw_conn_req_handler(struct iw_cm_id *cm_id,
2200                                struct iw_cm_event *iw_event)
2201 {
2202         struct rdma_cm_id *new_cm_id;
2203         struct rdma_id_private *listen_id, *conn_id;
2204         struct rdma_cm_event event;
2205         int ret = -ECONNABORTED;
2206         struct sockaddr *laddr = (struct sockaddr *)&iw_event->local_addr;
2207         struct sockaddr *raddr = (struct sockaddr *)&iw_event->remote_addr;
2208
2209         listen_id = cm_id->context;
2210
2211         mutex_lock(&listen_id->handler_mutex);
2212         if (listen_id->state != RDMA_CM_LISTEN)
2213                 goto out;
2214
2215         /* Create a new RDMA id for the new IW CM ID */
2216         new_cm_id = __rdma_create_id(listen_id->id.route.addr.dev_addr.net,
2217                                      listen_id->id.event_handler,
2218                                      listen_id->id.context,
2219                                      RDMA_PS_TCP, IB_QPT_RC,
2220                                      listen_id->res.kern_name);
2221         if (IS_ERR(new_cm_id)) {
2222                 ret = -ENOMEM;
2223                 goto out;
2224         }
2225         conn_id = container_of(new_cm_id, struct rdma_id_private, id);
2226         mutex_lock_nested(&conn_id->handler_mutex, SINGLE_DEPTH_NESTING);
2227         conn_id->state = RDMA_CM_CONNECT;
2228
2229         ret = rdma_translate_ip(laddr, &conn_id->id.route.addr.dev_addr);
2230         if (ret) {
2231                 mutex_unlock(&conn_id->handler_mutex);
2232                 rdma_destroy_id(new_cm_id);
2233                 goto out;
2234         }
2235
2236         ret = cma_acquire_dev(conn_id, listen_id);
2237         if (ret) {
2238                 mutex_unlock(&conn_id->handler_mutex);
2239                 rdma_destroy_id(new_cm_id);
2240                 goto out;
2241         }
2242
2243         conn_id->cm_id.iw = cm_id;
2244         cm_id->context = conn_id;
2245         cm_id->cm_handler = cma_iw_handler;
2246
2247         memcpy(cma_src_addr(conn_id), laddr, rdma_addr_size(laddr));
2248         memcpy(cma_dst_addr(conn_id), raddr, rdma_addr_size(raddr));
2249
2250         memset(&event, 0, sizeof event);
2251         event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
2252         event.param.conn.private_data = iw_event->private_data;
2253         event.param.conn.private_data_len = iw_event->private_data_len;
2254         event.param.conn.initiator_depth = iw_event->ird;
2255         event.param.conn.responder_resources = iw_event->ord;
2256
2257         /*
2258          * Protect against the user destroying conn_id from another thread
2259          * until we're done accessing it.
2260          */
2261         atomic_inc(&conn_id->refcount);
2262         ret = conn_id->id.event_handler(&conn_id->id, &event);
2263         if (ret) {
2264                 /* User wants to destroy the CM ID */
2265                 conn_id->cm_id.iw = NULL;
2266                 cma_exch(conn_id, RDMA_CM_DESTROYING);
2267                 mutex_unlock(&conn_id->handler_mutex);
2268                 cma_deref_id(conn_id);
2269                 rdma_destroy_id(&conn_id->id);
2270                 goto out;
2271         }
2272
2273         mutex_unlock(&conn_id->handler_mutex);
2274         cma_deref_id(conn_id);
2275
2276 out:
2277         mutex_unlock(&listen_id->handler_mutex);
2278         return ret;
2279 }
2280
2281 static int cma_ib_listen(struct rdma_id_private *id_priv)
2282 {
2283         struct sockaddr *addr;
2284         struct ib_cm_id *id;
2285         __be64 svc_id;
2286
2287         addr = cma_src_addr(id_priv);
2288         svc_id = rdma_get_service_id(&id_priv->id, addr);
2289         id = ib_cm_insert_listen(id_priv->id.device, cma_req_handler, svc_id);
2290         if (IS_ERR(id))
2291                 return PTR_ERR(id);
2292         id_priv->cm_id.ib = id;
2293
2294         return 0;
2295 }
2296
2297 static int cma_iw_listen(struct rdma_id_private *id_priv, int backlog)
2298 {
2299         int ret;
2300         struct iw_cm_id *id;
2301
2302         id = iw_create_cm_id(id_priv->id.device,
2303                              iw_conn_req_handler,
2304                              id_priv);
2305         if (IS_ERR(id))
2306                 return PTR_ERR(id);
2307
2308         id->tos = id_priv->tos;
2309         id_priv->cm_id.iw = id;
2310
2311         memcpy(&id_priv->cm_id.iw->local_addr, cma_src_addr(id_priv),
2312                rdma_addr_size(cma_src_addr(id_priv)));
2313
2314         ret = iw_cm_listen(id_priv->cm_id.iw, backlog);
2315
2316         if (ret) {
2317                 iw_destroy_cm_id(id_priv->cm_id.iw);
2318                 id_priv->cm_id.iw = NULL;
2319         }
2320
2321         return ret;
2322 }
2323
2324 static int cma_listen_handler(struct rdma_cm_id *id,
2325                               struct rdma_cm_event *event)
2326 {
2327         struct rdma_id_private *id_priv = id->context;
2328
2329         id->context = id_priv->id.context;
2330         id->event_handler = id_priv->id.event_handler;
2331         return id_priv->id.event_handler(id, event);
2332 }
2333
2334 static void cma_listen_on_dev(struct rdma_id_private *id_priv,
2335                               struct cma_device *cma_dev)
2336 {
2337         struct rdma_id_private *dev_id_priv;
2338         struct rdma_cm_id *id;
2339         struct net *net = id_priv->id.route.addr.dev_addr.net;
2340         int ret;
2341
2342         if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
2343                 return;
2344
2345         id = __rdma_create_id(net, cma_listen_handler, id_priv, id_priv->id.ps,
2346                               id_priv->id.qp_type, id_priv->res.kern_name);
2347         if (IS_ERR(id))
2348                 return;
2349
2350         dev_id_priv = container_of(id, struct rdma_id_private, id);
2351
2352         dev_id_priv->state = RDMA_CM_ADDR_BOUND;
2353         memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
2354                rdma_addr_size(cma_src_addr(id_priv)));
2355
2356         _cma_attach_to_dev(dev_id_priv, cma_dev);
2357         list_add_tail(&dev_id_priv->listen_list, &id_priv->listen_list);
2358         atomic_inc(&id_priv->refcount);
2359         dev_id_priv->internal_id = 1;
2360         dev_id_priv->afonly = id_priv->afonly;
2361
2362         ret = rdma_listen(id, id_priv->backlog);
2363         if (ret)
2364                 pr_warn("RDMA CMA: cma_listen_on_dev, error %d, listening on device %s\n",
2365                         ret, cma_dev->device->name);
2366 }
2367
2368 static void cma_listen_on_all(struct rdma_id_private *id_priv)
2369 {
2370         struct cma_device *cma_dev;
2371
2372         mutex_lock(&lock);
2373         list_add_tail(&id_priv->list, &listen_any_list);
2374         list_for_each_entry(cma_dev, &dev_list, list)
2375                 cma_listen_on_dev(id_priv, cma_dev);
2376         mutex_unlock(&lock);
2377 }
2378
2379 void rdma_set_service_type(struct rdma_cm_id *id, int tos)
2380 {
2381         struct rdma_id_private *id_priv;
2382
2383         id_priv = container_of(id, struct rdma_id_private, id);
2384         id_priv->tos = (u8) tos;
2385         id_priv->tos_set = true;
2386 }
2387 EXPORT_SYMBOL(rdma_set_service_type);
2388
2389 static void cma_query_handler(int status, struct sa_path_rec *path_rec,
2390                               void *context)
2391 {
2392         struct cma_work *work = context;
2393         struct rdma_route *route;
2394
2395         route = &work->id->id.route;
2396
2397         if (!status) {
2398                 route->num_paths = 1;
2399                 *route->path_rec = *path_rec;
2400         } else {
2401                 work->old_state = RDMA_CM_ROUTE_QUERY;
2402                 work->new_state = RDMA_CM_ADDR_RESOLVED;
2403                 work->event.event = RDMA_CM_EVENT_ROUTE_ERROR;
2404                 work->event.status = status;
2405                 pr_debug_ratelimited("RDMA CM: ROUTE_ERROR: failed to query path. status %d\n",
2406                                      status);
2407         }
2408
2409         queue_work(cma_wq, &work->work);
2410 }
2411
2412 static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms,
2413                               struct cma_work *work)
2414 {
2415         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
2416         struct sa_path_rec path_rec;
2417         ib_sa_comp_mask comp_mask;
2418         struct sockaddr_in6 *sin6;
2419         struct sockaddr_ib *sib;
2420
2421         memset(&path_rec, 0, sizeof path_rec);
2422
2423         if (rdma_cap_opa_ah(id_priv->id.device, id_priv->id.port_num))
2424                 path_rec.rec_type = SA_PATH_REC_TYPE_OPA;
2425         else
2426                 path_rec.rec_type = SA_PATH_REC_TYPE_IB;
2427         rdma_addr_get_sgid(dev_addr, &path_rec.sgid);
2428         rdma_addr_get_dgid(dev_addr, &path_rec.dgid);
2429         path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
2430         path_rec.numb_path = 1;
2431         path_rec.reversible = 1;
2432         path_rec.service_id = rdma_get_service_id(&id_priv->id,
2433                                                   cma_dst_addr(id_priv));
2434
2435         comp_mask = IB_SA_PATH_REC_DGID | IB_SA_PATH_REC_SGID |
2436                     IB_SA_PATH_REC_PKEY | IB_SA_PATH_REC_NUMB_PATH |
2437                     IB_SA_PATH_REC_REVERSIBLE | IB_SA_PATH_REC_SERVICE_ID;
2438
2439         switch (cma_family(id_priv)) {
2440         case AF_INET:
2441                 path_rec.qos_class = cpu_to_be16((u16) id_priv->tos);
2442                 comp_mask |= IB_SA_PATH_REC_QOS_CLASS;
2443                 break;
2444         case AF_INET6:
2445                 sin6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
2446                 path_rec.traffic_class = (u8) (be32_to_cpu(sin6->sin6_flowinfo) >> 20);
2447                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2448                 break;
2449         case AF_IB:
2450                 sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
2451                 path_rec.traffic_class = (u8) (be32_to_cpu(sib->sib_flowinfo) >> 20);
2452                 comp_mask |= IB_SA_PATH_REC_TRAFFIC_CLASS;
2453                 break;
2454         }
2455
2456         id_priv->query_id = ib_sa_path_rec_get(&sa_client, id_priv->id.device,
2457                                                id_priv->id.port_num, &path_rec,
2458                                                comp_mask, timeout_ms,
2459                                                GFP_KERNEL, cma_query_handler,
2460                                                work, &id_priv->query);
2461
2462         return (id_priv->query_id < 0) ? id_priv->query_id : 0;
2463 }
2464
2465 static void cma_work_handler(struct work_struct *_work)
2466 {
2467         struct cma_work *work = container_of(_work, struct cma_work, work);
2468         struct rdma_id_private *id_priv = work->id;
2469         int destroy = 0;
2470
2471         mutex_lock(&id_priv->handler_mutex);
2472         if (!cma_comp_exch(id_priv, work->old_state, work->new_state))
2473                 goto out;
2474
2475         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2476                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2477                 destroy = 1;
2478         }
2479 out:
2480         mutex_unlock(&id_priv->handler_mutex);
2481         cma_deref_id(id_priv);
2482         if (destroy)
2483                 rdma_destroy_id(&id_priv->id);
2484         kfree(work);
2485 }
2486
2487 static void cma_ndev_work_handler(struct work_struct *_work)
2488 {
2489         struct cma_ndev_work *work = container_of(_work, struct cma_ndev_work, work);
2490         struct rdma_id_private *id_priv = work->id;
2491         int destroy = 0;
2492
2493         mutex_lock(&id_priv->handler_mutex);
2494         if (id_priv->state == RDMA_CM_DESTROYING ||
2495             id_priv->state == RDMA_CM_DEVICE_REMOVAL)
2496                 goto out;
2497
2498         if (id_priv->id.event_handler(&id_priv->id, &work->event)) {
2499                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2500                 destroy = 1;
2501         }
2502
2503 out:
2504         mutex_unlock(&id_priv->handler_mutex);
2505         cma_deref_id(id_priv);
2506         if (destroy)
2507                 rdma_destroy_id(&id_priv->id);
2508         kfree(work);
2509 }
2510
2511 static void cma_init_resolve_route_work(struct cma_work *work,
2512                                         struct rdma_id_private *id_priv)
2513 {
2514         work->id = id_priv;
2515         INIT_WORK(&work->work, cma_work_handler);
2516         work->old_state = RDMA_CM_ROUTE_QUERY;
2517         work->new_state = RDMA_CM_ROUTE_RESOLVED;
2518         work->event.event = RDMA_CM_EVENT_ROUTE_RESOLVED;
2519 }
2520
2521 static void cma_init_resolve_addr_work(struct cma_work *work,
2522                                        struct rdma_id_private *id_priv)
2523 {
2524         work->id = id_priv;
2525         INIT_WORK(&work->work, cma_work_handler);
2526         work->old_state = RDMA_CM_ADDR_QUERY;
2527         work->new_state = RDMA_CM_ADDR_RESOLVED;
2528         work->event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2529 }
2530
2531 static int cma_resolve_ib_route(struct rdma_id_private *id_priv, int timeout_ms)
2532 {
2533         struct rdma_route *route = &id_priv->id.route;
2534         struct cma_work *work;
2535         int ret;
2536
2537         work = kzalloc(sizeof *work, GFP_KERNEL);
2538         if (!work)
2539                 return -ENOMEM;
2540
2541         cma_init_resolve_route_work(work, id_priv);
2542
2543         route->path_rec = kmalloc(sizeof *route->path_rec, GFP_KERNEL);
2544         if (!route->path_rec) {
2545                 ret = -ENOMEM;
2546                 goto err1;
2547         }
2548
2549         ret = cma_query_ib_route(id_priv, timeout_ms, work);
2550         if (ret)
2551                 goto err2;
2552
2553         return 0;
2554 err2:
2555         kfree(route->path_rec);
2556         route->path_rec = NULL;
2557 err1:
2558         kfree(work);
2559         return ret;
2560 }
2561
2562 static enum ib_gid_type cma_route_gid_type(enum rdma_network_type network_type,
2563                                            unsigned long supported_gids,
2564                                            enum ib_gid_type default_gid)
2565 {
2566         if ((network_type == RDMA_NETWORK_IPV4 ||
2567              network_type == RDMA_NETWORK_IPV6) &&
2568             test_bit(IB_GID_TYPE_ROCE_UDP_ENCAP, &supported_gids))
2569                 return IB_GID_TYPE_ROCE_UDP_ENCAP;
2570
2571         return default_gid;
2572 }
2573
2574 /*
2575  * cma_iboe_set_path_rec_l2_fields() is helper function which sets
2576  * path record type based on GID type.
2577  * It also sets up other L2 fields which includes destination mac address
2578  * netdev ifindex, of the path record.
2579  * It returns the netdev of the bound interface for this path record entry.
2580  */
2581 static struct net_device *
2582 cma_iboe_set_path_rec_l2_fields(struct rdma_id_private *id_priv)
2583 {
2584         struct rdma_route *route = &id_priv->id.route;
2585         enum ib_gid_type gid_type = IB_GID_TYPE_ROCE;
2586         struct rdma_addr *addr = &route->addr;
2587         unsigned long supported_gids;
2588         struct net_device *ndev;
2589
2590         if (!addr->dev_addr.bound_dev_if)
2591                 return NULL;
2592
2593         ndev = dev_get_by_index(addr->dev_addr.net,
2594                                 addr->dev_addr.bound_dev_if);
2595         if (!ndev)
2596                 return NULL;
2597
2598         supported_gids = roce_gid_type_mask_support(id_priv->id.device,
2599                                                     id_priv->id.port_num);
2600         gid_type = cma_route_gid_type(addr->dev_addr.network,
2601                                       supported_gids,
2602                                       id_priv->gid_type);
2603         /* Use the hint from IP Stack to select GID Type */
2604         if (gid_type < ib_network_to_gid_type(addr->dev_addr.network))
2605                 gid_type = ib_network_to_gid_type(addr->dev_addr.network);
2606         route->path_rec->rec_type = sa_conv_gid_to_pathrec_type(gid_type);
2607
2608         route->path_rec->roce.route_resolved = true;
2609         sa_path_set_dmac(route->path_rec, addr->dev_addr.dst_dev_addr);
2610         return ndev;
2611 }
2612
2613 int rdma_set_ib_path(struct rdma_cm_id *id,
2614                      struct sa_path_rec *path_rec)
2615 {
2616         struct rdma_id_private *id_priv;
2617         struct net_device *ndev;
2618         int ret;
2619
2620         id_priv = container_of(id, struct rdma_id_private, id);
2621         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2622                            RDMA_CM_ROUTE_RESOLVED))
2623                 return -EINVAL;
2624
2625         id->route.path_rec = kmemdup(path_rec, sizeof(*path_rec),
2626                                      GFP_KERNEL);
2627         if (!id->route.path_rec) {
2628                 ret = -ENOMEM;
2629                 goto err;
2630         }
2631
2632         if (rdma_protocol_roce(id->device, id->port_num)) {
2633                 ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2634                 if (!ndev) {
2635                         ret = -ENODEV;
2636                         goto err_free;
2637                 }
2638                 dev_put(ndev);
2639         }
2640
2641         id->route.num_paths = 1;
2642         return 0;
2643
2644 err_free:
2645         kfree(id->route.path_rec);
2646         id->route.path_rec = NULL;
2647 err:
2648         cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_ADDR_RESOLVED);
2649         return ret;
2650 }
2651 EXPORT_SYMBOL(rdma_set_ib_path);
2652
2653 static int cma_resolve_iw_route(struct rdma_id_private *id_priv, int timeout_ms)
2654 {
2655         struct cma_work *work;
2656
2657         work = kzalloc(sizeof *work, GFP_KERNEL);
2658         if (!work)
2659                 return -ENOMEM;
2660
2661         cma_init_resolve_route_work(work, id_priv);
2662         queue_work(cma_wq, &work->work);
2663         return 0;
2664 }
2665
2666 static int iboe_tos_to_sl(struct net_device *ndev, int tos)
2667 {
2668         int prio;
2669         struct net_device *dev;
2670
2671         prio = rt_tos2priority(tos);
2672         dev = is_vlan_dev(ndev) ? vlan_dev_real_dev(ndev) : ndev;
2673         if (dev->num_tc)
2674                 return netdev_get_prio_tc_map(dev, prio);
2675
2676 #if IS_ENABLED(CONFIG_VLAN_8021Q)
2677         if (is_vlan_dev(ndev))
2678                 return (vlan_dev_get_egress_qos_mask(ndev, prio) &
2679                         VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
2680 #endif
2681         return 0;
2682 }
2683
2684 static int cma_resolve_iboe_route(struct rdma_id_private *id_priv)
2685 {
2686         struct rdma_route *route = &id_priv->id.route;
2687         struct rdma_addr *addr = &route->addr;
2688         struct cma_work *work;
2689         int ret;
2690         struct net_device *ndev;
2691
2692         u8 default_roce_tos = id_priv->cma_dev->default_roce_tos[id_priv->id.port_num -
2693                                         rdma_start_port(id_priv->cma_dev->device)];
2694         u8 tos = id_priv->tos_set ? id_priv->tos : default_roce_tos;
2695
2696
2697         work = kzalloc(sizeof *work, GFP_KERNEL);
2698         if (!work)
2699                 return -ENOMEM;
2700
2701         route->path_rec = kzalloc(sizeof *route->path_rec, GFP_KERNEL);
2702         if (!route->path_rec) {
2703                 ret = -ENOMEM;
2704                 goto err1;
2705         }
2706
2707         route->num_paths = 1;
2708
2709         ndev = cma_iboe_set_path_rec_l2_fields(id_priv);
2710         if (!ndev) {
2711                 ret = -ENODEV;
2712                 goto err2;
2713         }
2714
2715         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
2716                     &route->path_rec->sgid);
2717         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.dst_addr,
2718                     &route->path_rec->dgid);
2719
2720         if (((struct sockaddr *)&id_priv->id.route.addr.dst_addr)->sa_family != AF_IB)
2721                 /* TODO: get the hoplimit from the inet/inet6 device */
2722                 route->path_rec->hop_limit = addr->dev_addr.hoplimit;
2723         else
2724                 route->path_rec->hop_limit = 1;
2725         route->path_rec->reversible = 1;
2726         route->path_rec->pkey = cpu_to_be16(0xffff);
2727         route->path_rec->mtu_selector = IB_SA_EQ;
2728         route->path_rec->sl = iboe_tos_to_sl(ndev, tos);
2729         route->path_rec->traffic_class = tos;
2730         route->path_rec->mtu = iboe_get_mtu(ndev->mtu);
2731         route->path_rec->rate_selector = IB_SA_EQ;
2732         route->path_rec->rate = iboe_get_rate(ndev);
2733         dev_put(ndev);
2734         route->path_rec->packet_life_time_selector = IB_SA_EQ;
2735         route->path_rec->packet_life_time = CMA_IBOE_PACKET_LIFETIME;
2736         if (!route->path_rec->mtu) {
2737                 ret = -EINVAL;
2738                 goto err2;
2739         }
2740
2741         cma_init_resolve_route_work(work, id_priv);
2742         queue_work(cma_wq, &work->work);
2743
2744         return 0;
2745
2746 err2:
2747         kfree(route->path_rec);
2748         route->path_rec = NULL;
2749 err1:
2750         kfree(work);
2751         return ret;
2752 }
2753
2754 int rdma_resolve_route(struct rdma_cm_id *id, int timeout_ms)
2755 {
2756         struct rdma_id_private *id_priv;
2757         int ret;
2758
2759         id_priv = container_of(id, struct rdma_id_private, id);
2760         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED, RDMA_CM_ROUTE_QUERY))
2761                 return -EINVAL;
2762
2763         atomic_inc(&id_priv->refcount);
2764         if (rdma_cap_ib_sa(id->device, id->port_num))
2765                 ret = cma_resolve_ib_route(id_priv, timeout_ms);
2766         else if (rdma_protocol_roce(id->device, id->port_num))
2767                 ret = cma_resolve_iboe_route(id_priv);
2768         else if (rdma_protocol_iwarp(id->device, id->port_num))
2769                 ret = cma_resolve_iw_route(id_priv, timeout_ms);
2770         else
2771                 ret = -ENOSYS;
2772
2773         if (ret)
2774                 goto err;
2775
2776         return 0;
2777 err:
2778         cma_comp_exch(id_priv, RDMA_CM_ROUTE_QUERY, RDMA_CM_ADDR_RESOLVED);
2779         cma_deref_id(id_priv);
2780         return ret;
2781 }
2782 EXPORT_SYMBOL(rdma_resolve_route);
2783
2784 static void cma_set_loopback(struct sockaddr *addr)
2785 {
2786         switch (addr->sa_family) {
2787         case AF_INET:
2788                 ((struct sockaddr_in *) addr)->sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2789                 break;
2790         case AF_INET6:
2791                 ipv6_addr_set(&((struct sockaddr_in6 *) addr)->sin6_addr,
2792                               0, 0, 0, htonl(1));
2793                 break;
2794         default:
2795                 ib_addr_set(&((struct sockaddr_ib *) addr)->sib_addr,
2796                             0, 0, 0, htonl(1));
2797                 break;
2798         }
2799 }
2800
2801 static int cma_bind_loopback(struct rdma_id_private *id_priv)
2802 {
2803         struct cma_device *cma_dev, *cur_dev;
2804         union ib_gid gid;
2805         enum ib_port_state port_state;
2806         u16 pkey;
2807         int ret;
2808         u8 p;
2809
2810         cma_dev = NULL;
2811         mutex_lock(&lock);
2812         list_for_each_entry(cur_dev, &dev_list, list) {
2813                 if (cma_family(id_priv) == AF_IB &&
2814                     !rdma_cap_ib_cm(cur_dev->device, 1))
2815                         continue;
2816
2817                 if (!cma_dev)
2818                         cma_dev = cur_dev;
2819
2820                 for (p = 1; p <= cur_dev->device->phys_port_cnt; ++p) {
2821                         if (!ib_get_cached_port_state(cur_dev->device, p, &port_state) &&
2822                             port_state == IB_PORT_ACTIVE) {
2823                                 cma_dev = cur_dev;
2824                                 goto port_found;
2825                         }
2826                 }
2827         }
2828
2829         if (!cma_dev) {
2830                 ret = -ENODEV;
2831                 goto out;
2832         }
2833
2834         p = 1;
2835
2836 port_found:
2837         ret = rdma_query_gid(cma_dev->device, p, 0, &gid);
2838         if (ret)
2839                 goto out;
2840
2841         ret = ib_get_cached_pkey(cma_dev->device, p, 0, &pkey);
2842         if (ret)
2843                 goto out;
2844
2845         id_priv->id.route.addr.dev_addr.dev_type =
2846                 (rdma_protocol_ib(cma_dev->device, p)) ?
2847                 ARPHRD_INFINIBAND : ARPHRD_ETHER;
2848
2849         rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2850         ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey);
2851         id_priv->id.port_num = p;
2852         cma_attach_to_dev(id_priv, cma_dev);
2853         cma_set_loopback(cma_src_addr(id_priv));
2854 out:
2855         mutex_unlock(&lock);
2856         return ret;
2857 }
2858
2859 static void addr_handler(int status, struct sockaddr *src_addr,
2860                          struct rdma_dev_addr *dev_addr, void *context)
2861 {
2862         struct rdma_id_private *id_priv = context;
2863         struct rdma_cm_event event;
2864
2865         memset(&event, 0, sizeof event);
2866         mutex_lock(&id_priv->handler_mutex);
2867         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY,
2868                            RDMA_CM_ADDR_RESOLVED))
2869                 goto out;
2870
2871         memcpy(cma_src_addr(id_priv), src_addr, rdma_addr_size(src_addr));
2872         if (!status && !id_priv->cma_dev) {
2873                 status = cma_acquire_dev(id_priv, NULL);
2874                 if (status)
2875                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to acquire device. status %d\n",
2876                                              status);
2877         } else {
2878                 pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to resolve IP. status %d\n", status);
2879         }
2880
2881         if (status) {
2882                 if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_RESOLVED,
2883                                    RDMA_CM_ADDR_BOUND))
2884                         goto out;
2885                 event.event = RDMA_CM_EVENT_ADDR_ERROR;
2886                 event.status = status;
2887         } else
2888                 event.event = RDMA_CM_EVENT_ADDR_RESOLVED;
2889
2890         if (id_priv->id.event_handler(&id_priv->id, &event)) {
2891                 cma_exch(id_priv, RDMA_CM_DESTROYING);
2892                 mutex_unlock(&id_priv->handler_mutex);
2893                 cma_deref_id(id_priv);
2894                 rdma_destroy_id(&id_priv->id);
2895                 return;
2896         }
2897 out:
2898         mutex_unlock(&id_priv->handler_mutex);
2899         cma_deref_id(id_priv);
2900 }
2901
2902 static int cma_resolve_loopback(struct rdma_id_private *id_priv)
2903 {
2904         struct cma_work *work;
2905         union ib_gid gid;
2906         int ret;
2907
2908         work = kzalloc(sizeof *work, GFP_KERNEL);
2909         if (!work)
2910                 return -ENOMEM;
2911
2912         if (!id_priv->cma_dev) {
2913                 ret = cma_bind_loopback(id_priv);
2914                 if (ret)
2915                         goto err;
2916         }
2917
2918         rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
2919         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
2920
2921         cma_init_resolve_addr_work(work, id_priv);
2922         queue_work(cma_wq, &work->work);
2923         return 0;
2924 err:
2925         kfree(work);
2926         return ret;
2927 }
2928
2929 static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
2930 {
2931         struct cma_work *work;
2932         int ret;
2933
2934         work = kzalloc(sizeof *work, GFP_KERNEL);
2935         if (!work)
2936                 return -ENOMEM;
2937
2938         if (!id_priv->cma_dev) {
2939                 ret = cma_resolve_ib_dev(id_priv);
2940                 if (ret)
2941                         goto err;
2942         }
2943
2944         rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
2945                 &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
2946
2947         cma_init_resolve_addr_work(work, id_priv);
2948         queue_work(cma_wq, &work->work);
2949         return 0;
2950 err:
2951         kfree(work);
2952         return ret;
2953 }
2954
2955 static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2956                          const struct sockaddr *dst_addr)
2957 {
2958         if (!src_addr || !src_addr->sa_family) {
2959                 src_addr = (struct sockaddr *) &id->route.addr.src_addr;
2960                 src_addr->sa_family = dst_addr->sa_family;
2961                 if (IS_ENABLED(CONFIG_IPV6) &&
2962                     dst_addr->sa_family == AF_INET6) {
2963                         struct sockaddr_in6 *src_addr6 = (struct sockaddr_in6 *) src_addr;
2964                         struct sockaddr_in6 *dst_addr6 = (struct sockaddr_in6 *) dst_addr;
2965                         src_addr6->sin6_scope_id = dst_addr6->sin6_scope_id;
2966                         if (ipv6_addr_type(&dst_addr6->sin6_addr) & IPV6_ADDR_LINKLOCAL)
2967                                 id->route.addr.dev_addr.bound_dev_if = dst_addr6->sin6_scope_id;
2968                 } else if (dst_addr->sa_family == AF_IB) {
2969                         ((struct sockaddr_ib *) src_addr)->sib_pkey =
2970                                 ((struct sockaddr_ib *) dst_addr)->sib_pkey;
2971                 }
2972         }
2973         return rdma_bind_addr(id, src_addr);
2974 }
2975
2976 int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
2977                       const struct sockaddr *dst_addr, int timeout_ms)
2978 {
2979         struct rdma_id_private *id_priv;
2980         int ret;
2981
2982         id_priv = container_of(id, struct rdma_id_private, id);
2983         if (id_priv->state == RDMA_CM_IDLE) {
2984                 ret = cma_bind_addr(id, src_addr, dst_addr);
2985                 if (ret)
2986                         return ret;
2987         }
2988
2989         if (cma_family(id_priv) != dst_addr->sa_family)
2990                 return -EINVAL;
2991
2992         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_ADDR_QUERY))
2993                 return -EINVAL;
2994
2995         memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
2996         atomic_inc(&id_priv->refcount);
2997         if (cma_any_addr(dst_addr)) {
2998                 ret = cma_resolve_loopback(id_priv);
2999         } else {
3000                 if (dst_addr->sa_family == AF_IB) {
3001                         ret = cma_resolve_ib_addr(id_priv);
3002                 } else {
3003                         ret = rdma_resolve_ip(cma_src_addr(id_priv),
3004                                               dst_addr, &id->route.addr.dev_addr,
3005                                               timeout_ms, addr_handler, id_priv);
3006                 }
3007         }
3008         if (ret)
3009                 goto err;
3010
3011         return 0;
3012 err:
3013         cma_comp_exch(id_priv, RDMA_CM_ADDR_QUERY, RDMA_CM_ADDR_BOUND);
3014         cma_deref_id(id_priv);
3015         return ret;
3016 }
3017 EXPORT_SYMBOL(rdma_resolve_addr);
3018
3019 int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
3020 {
3021         struct rdma_id_private *id_priv;
3022         unsigned long flags;
3023         int ret;
3024
3025         id_priv = container_of(id, struct rdma_id_private, id);
3026         spin_lock_irqsave(&id_priv->lock, flags);
3027         if (reuse || id_priv->state == RDMA_CM_IDLE) {
3028                 id_priv->reuseaddr = reuse;
3029                 ret = 0;
3030         } else {
3031                 ret = -EINVAL;
3032         }
3033         spin_unlock_irqrestore(&id_priv->lock, flags);
3034         return ret;
3035 }
3036 EXPORT_SYMBOL(rdma_set_reuseaddr);
3037
3038 int rdma_set_afonly(struct rdma_cm_id *id, int afonly)
3039 {
3040         struct rdma_id_private *id_priv;
3041         unsigned long flags;
3042         int ret;
3043
3044         id_priv = container_of(id, struct rdma_id_private, id);
3045         spin_lock_irqsave(&id_priv->lock, flags);
3046         if (id_priv->state == RDMA_CM_IDLE || id_priv->state == RDMA_CM_ADDR_BOUND) {
3047                 id_priv->options |= (1 << CMA_OPTION_AFONLY);
3048                 id_priv->afonly = afonly;
3049                 ret = 0;
3050         } else {
3051                 ret = -EINVAL;
3052         }
3053         spin_unlock_irqrestore(&id_priv->lock, flags);
3054         return ret;
3055 }
3056 EXPORT_SYMBOL(rdma_set_afonly);
3057
3058 static void cma_bind_port(struct rdma_bind_list *bind_list,
3059                           struct rdma_id_private *id_priv)
3060 {
3061         struct sockaddr *addr;
3062         struct sockaddr_ib *sib;
3063         u64 sid, mask;
3064         __be16 port;
3065
3066         addr = cma_src_addr(id_priv);
3067         port = htons(bind_list->port);
3068
3069         switch (addr->sa_family) {
3070         case AF_INET:
3071                 ((struct sockaddr_in *) addr)->sin_port = port;
3072                 break;
3073         case AF_INET6:
3074                 ((struct sockaddr_in6 *) addr)->sin6_port = port;
3075                 break;
3076         case AF_IB:
3077                 sib = (struct sockaddr_ib *) addr;
3078                 sid = be64_to_cpu(sib->sib_sid);
3079                 mask = be64_to_cpu(sib->sib_sid_mask);
3080                 sib->sib_sid = cpu_to_be64((sid & mask) | (u64) ntohs(port));
3081                 sib->sib_sid_mask = cpu_to_be64(~0ULL);
3082                 break;
3083         }
3084         id_priv->bind_list = bind_list;
3085         hlist_add_head(&id_priv->node, &bind_list->owners);
3086 }
3087
3088 static int cma_alloc_port(enum rdma_ucm_port_space ps,
3089                           struct rdma_id_private *id_priv, unsigned short snum)
3090 {
3091         struct rdma_bind_list *bind_list;
3092         int ret;
3093
3094         bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL);
3095         if (!bind_list)
3096                 return -ENOMEM;
3097
3098         ret = cma_ps_alloc(id_priv->id.route.addr.dev_addr.net, ps, bind_list,
3099                            snum);
3100         if (ret < 0)
3101                 goto err;
3102
3103         bind_list->ps = ps;
3104         bind_list->port = (unsigned short)ret;
3105         cma_bind_port(bind_list, id_priv);
3106         return 0;
3107 err:
3108         kfree(bind_list);
3109         return ret == -ENOSPC ? -EADDRNOTAVAIL : ret;
3110 }
3111
3112 static int cma_port_is_unique(struct rdma_bind_list *bind_list,
3113                               struct rdma_id_private *id_priv)
3114 {
3115         struct rdma_id_private *cur_id;
3116         struct sockaddr  *daddr = cma_dst_addr(id_priv);
3117         struct sockaddr  *saddr = cma_src_addr(id_priv);
3118         __be16 dport = cma_port(daddr);
3119
3120         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3121                 struct sockaddr  *cur_daddr = cma_dst_addr(cur_id);
3122                 struct sockaddr  *cur_saddr = cma_src_addr(cur_id);
3123                 __be16 cur_dport = cma_port(cur_daddr);
3124
3125                 if (id_priv == cur_id)
3126                         continue;
3127
3128                 /* different dest port -> unique */
3129                 if (!cma_any_port(daddr) &&
3130                     !cma_any_port(cur_daddr) &&
3131                     (dport != cur_dport))
3132                         continue;
3133
3134                 /* different src address -> unique */
3135                 if (!cma_any_addr(saddr) &&
3136                     !cma_any_addr(cur_saddr) &&
3137                     cma_addr_cmp(saddr, cur_saddr))
3138                         continue;
3139
3140                 /* different dst address -> unique */
3141                 if (!cma_any_addr(daddr) &&
3142                     !cma_any_addr(cur_daddr) &&
3143                     cma_addr_cmp(daddr, cur_daddr))
3144                         continue;
3145
3146                 return -EADDRNOTAVAIL;
3147         }
3148         return 0;
3149 }
3150
3151 static int cma_alloc_any_port(enum rdma_ucm_port_space ps,
3152                               struct rdma_id_private *id_priv)
3153 {
3154         static unsigned int last_used_port;
3155         int low, high, remaining;
3156         unsigned int rover;
3157         struct net *net = id_priv->id.route.addr.dev_addr.net;
3158
3159         inet_get_local_port_range(net, &low, &high);
3160         remaining = (high - low) + 1;
3161         rover = prandom_u32() % remaining + low;
3162 retry:
3163         if (last_used_port != rover) {
3164                 struct rdma_bind_list *bind_list;
3165                 int ret;
3166
3167                 bind_list = cma_ps_find(net, ps, (unsigned short)rover);
3168
3169                 if (!bind_list) {
3170                         ret = cma_alloc_port(ps, id_priv, rover);
3171                 } else {
3172                         ret = cma_port_is_unique(bind_list, id_priv);
3173                         if (!ret)
3174                                 cma_bind_port(bind_list, id_priv);
3175                 }
3176                 /*
3177                  * Remember previously used port number in order to avoid
3178                  * re-using same port immediately after it is closed.
3179                  */
3180                 if (!ret)
3181                         last_used_port = rover;
3182                 if (ret != -EADDRNOTAVAIL)
3183                         return ret;
3184         }
3185         if (--remaining) {
3186                 rover++;
3187                 if ((rover < low) || (rover > high))
3188                         rover = low;
3189                 goto retry;
3190         }
3191         return -EADDRNOTAVAIL;
3192 }
3193
3194 /*
3195  * Check that the requested port is available.  This is called when trying to
3196  * bind to a specific port, or when trying to listen on a bound port.  In
3197  * the latter case, the provided id_priv may already be on the bind_list, but
3198  * we still need to check that it's okay to start listening.
3199  */
3200 static int cma_check_port(struct rdma_bind_list *bind_list,
3201                           struct rdma_id_private *id_priv, uint8_t reuseaddr)
3202 {
3203         struct rdma_id_private *cur_id;
3204         struct sockaddr *addr, *cur_addr;
3205
3206         addr = cma_src_addr(id_priv);
3207         hlist_for_each_entry(cur_id, &bind_list->owners, node) {
3208                 if (id_priv == cur_id)
3209                         continue;
3210
3211                 if ((cur_id->state != RDMA_CM_LISTEN) && reuseaddr &&
3212                     cur_id->reuseaddr)
3213                         continue;
3214
3215                 cur_addr = cma_src_addr(cur_id);
3216                 if (id_priv->afonly && cur_id->afonly &&
3217                     (addr->sa_family != cur_addr->sa_family))
3218                         continue;
3219
3220                 if (cma_any_addr(addr) || cma_any_addr(cur_addr))
3221                         return -EADDRNOTAVAIL;
3222
3223                 if (!cma_addr_cmp(addr, cur_addr))
3224                         return -EADDRINUSE;
3225         }
3226         return 0;
3227 }
3228
3229 static int cma_use_port(enum rdma_ucm_port_space ps,
3230                         struct rdma_id_private *id_priv)
3231 {
3232         struct rdma_bind_list *bind_list;
3233         unsigned short snum;
3234         int ret;
3235
3236         snum = ntohs(cma_port(cma_src_addr(id_priv)));
3237         if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
3238                 return -EACCES;
3239
3240         bind_list = cma_ps_find(id_priv->id.route.addr.dev_addr.net, ps, snum);
3241         if (!bind_list) {
3242                 ret = cma_alloc_port(ps, id_priv, snum);
3243         } else {
3244                 ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
3245                 if (!ret)
3246                         cma_bind_port(bind_list, id_priv);
3247         }
3248         return ret;
3249 }
3250
3251 static int cma_bind_listen(struct rdma_id_private *id_priv)
3252 {
3253         struct rdma_bind_list *bind_list = id_priv->bind_list;
3254         int ret = 0;
3255
3256         mutex_lock(&lock);
3257         if (bind_list->owners.first->next)
3258                 ret = cma_check_port(bind_list, id_priv, 0);
3259         mutex_unlock(&lock);
3260         return ret;
3261 }
3262
3263 static enum rdma_ucm_port_space
3264 cma_select_inet_ps(struct rdma_id_private *id_priv)
3265 {
3266         switch (id_priv->id.ps) {
3267         case RDMA_PS_TCP:
3268         case RDMA_PS_UDP:
3269         case RDMA_PS_IPOIB:
3270         case RDMA_PS_IB:
3271                 return id_priv->id.ps;
3272         default:
3273
3274                 return 0;
3275         }
3276 }
3277
3278 static enum rdma_ucm_port_space
3279 cma_select_ib_ps(struct rdma_id_private *id_priv)
3280 {
3281         enum rdma_ucm_port_space ps = 0;
3282         struct sockaddr_ib *sib;
3283         u64 sid_ps, mask, sid;
3284
3285         sib = (struct sockaddr_ib *) cma_src_addr(id_priv);
3286         mask = be64_to_cpu(sib->sib_sid_mask) & RDMA_IB_IP_PS_MASK;
3287         sid = be64_to_cpu(sib->sib_sid) & mask;
3288
3289         if ((id_priv->id.ps == RDMA_PS_IB) && (sid == (RDMA_IB_IP_PS_IB & mask))) {
3290                 sid_ps = RDMA_IB_IP_PS_IB;
3291                 ps = RDMA_PS_IB;
3292         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_TCP)) &&
3293                    (sid == (RDMA_IB_IP_PS_TCP & mask))) {
3294                 sid_ps = RDMA_IB_IP_PS_TCP;
3295                 ps = RDMA_PS_TCP;
3296         } else if (((id_priv->id.ps == RDMA_PS_IB) || (id_priv->id.ps == RDMA_PS_UDP)) &&
3297                    (sid == (RDMA_IB_IP_PS_UDP & mask))) {
3298                 sid_ps = RDMA_IB_IP_PS_UDP;
3299                 ps = RDMA_PS_UDP;
3300         }
3301
3302         if (ps) {
3303                 sib->sib_sid = cpu_to_be64(sid_ps | ntohs(cma_port((struct sockaddr *) sib)));
3304                 sib->sib_sid_mask = cpu_to_be64(RDMA_IB_IP_PS_MASK |
3305                                                 be64_to_cpu(sib->sib_sid_mask));
3306         }
3307         return ps;
3308 }
3309
3310 static int cma_get_port(struct rdma_id_private *id_priv)
3311 {
3312         enum rdma_ucm_port_space ps;
3313         int ret;
3314
3315         if (cma_family(id_priv) != AF_IB)
3316                 ps = cma_select_inet_ps(id_priv);
3317         else
3318                 ps = cma_select_ib_ps(id_priv);
3319         if (!ps)
3320                 return -EPROTONOSUPPORT;
3321
3322         mutex_lock(&lock);
3323         if (cma_any_port(cma_src_addr(id_priv)))
3324                 ret = cma_alloc_any_port(ps, id_priv);
3325         else
3326                 ret = cma_use_port(ps, id_priv);
3327         mutex_unlock(&lock);
3328
3329         return ret;
3330 }
3331
3332 static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
3333                                struct sockaddr *addr)
3334 {
3335 #if IS_ENABLED(CONFIG_IPV6)
3336         struct sockaddr_in6 *sin6;
3337
3338         if (addr->sa_family != AF_INET6)
3339                 return 0;
3340
3341         sin6 = (struct sockaddr_in6 *) addr;
3342
3343         if (!(ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL))
3344                 return 0;
3345
3346         if (!sin6->sin6_scope_id)
3347                         return -EINVAL;
3348
3349         dev_addr->bound_dev_if = sin6->sin6_scope_id;
3350 #endif
3351         return 0;
3352 }
3353
3354 int rdma_listen(struct rdma_cm_id *id, int backlog)
3355 {
3356         struct rdma_id_private *id_priv;
3357         int ret;
3358
3359         id_priv = container_of(id, struct rdma_id_private, id);
3360         if (id_priv->state == RDMA_CM_IDLE) {
3361                 id->route.addr.src_addr.ss_family = AF_INET;
3362                 ret = rdma_bind_addr(id, cma_src_addr(id_priv));
3363                 if (ret)
3364                         return ret;
3365         }
3366
3367         if (!cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_LISTEN))
3368                 return -EINVAL;
3369
3370         if (id_priv->reuseaddr) {
3371                 ret = cma_bind_listen(id_priv);
3372                 if (ret)
3373                         goto err;
3374         }
3375
3376         id_priv->backlog = backlog;
3377         if (id->device) {
3378                 if (rdma_cap_ib_cm(id->device, 1)) {
3379                         ret = cma_ib_listen(id_priv);
3380                         if (ret)
3381                                 goto err;
3382                 } else if (rdma_cap_iw_cm(id->device, 1)) {
3383                         ret = cma_iw_listen(id_priv, backlog);
3384                         if (ret)
3385                                 goto err;
3386                 } else {
3387                         ret = -ENOSYS;
3388                         goto err;
3389                 }
3390         } else
3391                 cma_listen_on_all(id_priv);
3392
3393         return 0;
3394 err:
3395         id_priv->backlog = 0;
3396         cma_comp_exch(id_priv, RDMA_CM_LISTEN, RDMA_CM_ADDR_BOUND);
3397         return ret;
3398 }
3399 EXPORT_SYMBOL(rdma_listen);
3400
3401 int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
3402 {
3403         struct rdma_id_private *id_priv;
3404         int ret;
3405         struct sockaddr  *daddr;
3406
3407         if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6 &&
3408             addr->sa_family != AF_IB)
3409                 return -EAFNOSUPPORT;
3410
3411         id_priv = container_of(id, struct rdma_id_private, id);
3412         if (!cma_comp_exch(id_priv, RDMA_CM_IDLE, RDMA_CM_ADDR_BOUND))
3413                 return -EINVAL;
3414
3415         ret = cma_check_linklocal(&id->route.addr.dev_addr, addr);
3416         if (ret)
3417                 goto err1;
3418
3419         memcpy(cma_src_addr(id_priv), addr, rdma_addr_size(addr));
3420         if (!cma_any_addr(addr)) {
3421                 ret = cma_translate_addr(addr, &id->route.addr.dev_addr);
3422                 if (ret)
3423                         goto err1;
3424
3425                 ret = cma_acquire_dev(id_priv, NULL);
3426                 if (ret)
3427                         goto err1;
3428         }
3429
3430         if (!(id_priv->options & (1 << CMA_OPTION_AFONLY))) {
3431                 if (addr->sa_family == AF_INET)
3432                         id_priv->afonly = 1;
3433 #if IS_ENABLED(CONFIG_IPV6)
3434                 else if (addr->sa_family == AF_INET6) {
3435                         struct net *net = id_priv->id.route.addr.dev_addr.net;
3436
3437                         id_priv->afonly = net->ipv6.sysctl.bindv6only;
3438                 }
3439 #endif
3440         }
3441         daddr = cma_dst_addr(id_priv);
3442         daddr->sa_family = addr->sa_family;
3443
3444         ret = cma_get_port(id_priv);
3445         if (ret)
3446                 goto err2;
3447
3448         return 0;
3449 err2:
3450         if (id_priv->cma_dev) {
3451                 rdma_restrack_del(&id_priv->res);
3452                 cma_release_dev(id_priv);
3453         }
3454 err1:
3455         cma_comp_exch(id_priv, RDMA_CM_ADDR_BOUND, RDMA_CM_IDLE);
3456         return ret;
3457 }
3458 EXPORT_SYMBOL(rdma_bind_addr);
3459
3460 static int cma_format_hdr(void *hdr, struct rdma_id_private *id_priv)
3461 {
3462         struct cma_hdr *cma_hdr;
3463
3464         cma_hdr = hdr;
3465         cma_hdr->cma_version = CMA_VERSION;
3466         if (cma_family(id_priv) == AF_INET) {
3467                 struct sockaddr_in *src4, *dst4;
3468
3469                 src4 = (struct sockaddr_in *) cma_src_addr(id_priv);
3470                 dst4 = (struct sockaddr_in *) cma_dst_addr(id_priv);
3471
3472                 cma_set_ip_ver(cma_hdr, 4);
3473                 cma_hdr->src_addr.ip4.addr = src4->sin_addr.s_addr;
3474                 cma_hdr->dst_addr.ip4.addr = dst4->sin_addr.s_addr;
3475                 cma_hdr->port = src4->sin_port;
3476         } else if (cma_family(id_priv) == AF_INET6) {
3477                 struct sockaddr_in6 *src6, *dst6;
3478
3479                 src6 = (struct sockaddr_in6 *) cma_src_addr(id_priv);
3480                 dst6 = (struct sockaddr_in6 *) cma_dst_addr(id_priv);
3481
3482                 cma_set_ip_ver(cma_hdr, 6);
3483                 cma_hdr->src_addr.ip6 = src6->sin6_addr;
3484                 cma_hdr->dst_addr.ip6 = dst6->sin6_addr;
3485                 cma_hdr->port = src6->sin6_port;
3486         }
3487         return 0;
3488 }
3489
3490 static int cma_sidr_rep_handler(struct ib_cm_id *cm_id,
3491                                 const struct ib_cm_event *ib_event)
3492 {
3493         struct rdma_id_private *id_priv = cm_id->context;
3494         struct rdma_cm_event event;
3495         const struct ib_cm_sidr_rep_event_param *rep =
3496                                 &ib_event->param.sidr_rep_rcvd;
3497         int ret = 0;
3498
3499         mutex_lock(&id_priv->handler_mutex);
3500         if (id_priv->state != RDMA_CM_CONNECT)
3501                 goto out;
3502
3503         memset(&event, 0, sizeof event);
3504         switch (ib_event->event) {
3505         case IB_CM_SIDR_REQ_ERROR:
3506                 event.event = RDMA_CM_EVENT_UNREACHABLE;
3507                 event.status = -ETIMEDOUT;
3508                 break;
3509         case IB_CM_SIDR_REP_RECEIVED:
3510                 event.param.ud.private_data = ib_event->private_data;
3511                 event.param.ud.private_data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
3512                 if (rep->status != IB_SIDR_SUCCESS) {
3513                         event.event = RDMA_CM_EVENT_UNREACHABLE;
3514                         event.status = ib_event->param.sidr_rep_rcvd.status;
3515                         pr_debug_ratelimited("RDMA CM: UNREACHABLE: bad SIDR reply. status %d\n",
3516                                              event.status);
3517                         break;
3518                 }
3519                 ret = cma_set_qkey(id_priv, rep->qkey);
3520                 if (ret) {
3521                         pr_debug_ratelimited("RDMA CM: ADDR_ERROR: failed to set qkey. status %d\n", ret);
3522                         event.event = RDMA_CM_EVENT_ADDR_ERROR;
3523                         event.status = ret;
3524                         break;
3525                 }
3526                 ib_init_ah_attr_from_path(id_priv->id.device,
3527                                           id_priv->id.port_num,
3528                                           id_priv->id.route.path_rec,
3529                                           &event.param.ud.ah_attr,
3530                                           rep->sgid_attr);
3531                 event.param.ud.qp_num = rep->qpn;
3532                 event.param.ud.qkey = rep->qkey;
3533                 event.event = RDMA_CM_EVENT_ESTABLISHED;
3534                 event.status = 0;
3535                 break;
3536         default:
3537                 pr_err("RDMA CMA: unexpected IB CM event: %d\n",
3538                        ib_event->event);
3539                 goto out;
3540         }
3541
3542         ret = id_priv->id.event_handler(&id_priv->id, &event);
3543
3544         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
3545         if (ret) {
3546                 /* Destroy the CM ID by returning a non-zero value. */
3547                 id_priv->cm_id.ib = NULL;
3548                 cma_exch(id_priv, RDMA_CM_DESTROYING);
3549                 mutex_unlock(&id_priv->handler_mutex);
3550                 rdma_destroy_id(&id_priv->id);
3551                 return ret;
3552         }
3553 out:
3554         mutex_unlock(&id_priv->handler_mutex);
3555         return ret;
3556 }
3557
3558 static int cma_resolve_ib_udp(struct rdma_id_private *id_priv,
3559                               struct rdma_conn_param *conn_param)
3560 {
3561         struct ib_cm_sidr_req_param req;
3562         struct ib_cm_id *id;
3563         void *private_data;
3564         u8 offset;
3565         int ret;
3566
3567         memset(&req, 0, sizeof req);
3568         offset = cma_user_data_offset(id_priv);
3569         req.private_data_len = offset + conn_param->private_data_len;
3570         if (req.private_data_len < conn_param->private_data_len)
3571                 return -EINVAL;
3572
3573         if (req.private_data_len) {
3574                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3575                 if (!private_data)
3576                         return -ENOMEM;
3577         } else {
3578                 private_data = NULL;
3579         }
3580
3581         if (conn_param->private_data && conn_param->private_data_len)
3582                 memcpy(private_data + offset, conn_param->private_data,
3583                        conn_param->private_data_len);
3584
3585         if (private_data) {
3586                 ret = cma_format_hdr(private_data, id_priv);
3587                 if (ret)
3588                         goto out;
3589                 req.private_data = private_data;
3590         }
3591
3592         id = ib_create_cm_id(id_priv->id.device, cma_sidr_rep_handler,
3593                              id_priv);
3594         if (IS_ERR(id)) {
3595                 ret = PTR_ERR(id);
3596                 goto out;
3597         }
3598         id_priv->cm_id.ib = id;
3599
3600         req.path = id_priv->id.route.path_rec;
3601         req.sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3602         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3603         req.timeout_ms = 1 << (CMA_CM_RESPONSE_TIMEOUT - 8);
3604         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3605
3606         ret = ib_send_cm_sidr_req(id_priv->cm_id.ib, &req);
3607         if (ret) {
3608                 ib_destroy_cm_id(id_priv->cm_id.ib);
3609                 id_priv->cm_id.ib = NULL;
3610         }
3611 out:
3612         kfree(private_data);
3613         return ret;
3614 }
3615
3616 static int cma_connect_ib(struct rdma_id_private *id_priv,
3617                           struct rdma_conn_param *conn_param)
3618 {
3619         struct ib_cm_req_param req;
3620         struct rdma_route *route;
3621         void *private_data;
3622         struct ib_cm_id *id;
3623         u8 offset;
3624         int ret;
3625
3626         memset(&req, 0, sizeof req);
3627         offset = cma_user_data_offset(id_priv);
3628         req.private_data_len = offset + conn_param->private_data_len;
3629         if (req.private_data_len < conn_param->private_data_len)
3630                 return -EINVAL;
3631
3632         if (req.private_data_len) {
3633                 private_data = kzalloc(req.private_data_len, GFP_ATOMIC);
3634                 if (!private_data)
3635                         return -ENOMEM;
3636         } else {
3637                 private_data = NULL;
3638         }
3639
3640         if (conn_param->private_data && conn_param->private_data_len)
3641                 memcpy(private_data + offset, conn_param->private_data,
3642                        conn_param->private_data_len);
3643
3644         id = ib_create_cm_id(id_priv->id.device, cma_ib_handler, id_priv);
3645         if (IS_ERR(id)) {
3646                 ret = PTR_ERR(id);
3647                 goto out;
3648         }
3649         id_priv->cm_id.ib = id;
3650
3651         route = &id_priv->id.route;
3652         if (private_data) {
3653                 ret = cma_format_hdr(private_data, id_priv);
3654                 if (ret)
3655                         goto out;
3656                 req.private_data = private_data;
3657         }
3658
3659         req.primary_path = &route->path_rec[0];
3660         if (route->num_paths == 2)
3661                 req.alternate_path = &route->path_rec[1];
3662
3663         req.ppath_sgid_attr = id_priv->id.route.addr.dev_addr.sgid_attr;
3664         /* Alternate path SGID attribute currently unsupported */
3665         req.service_id = rdma_get_service_id(&id_priv->id, cma_dst_addr(id_priv));
3666         req.qp_num = id_priv->qp_num;
3667         req.qp_type = id_priv->id.qp_type;
3668         req.starting_psn = id_priv->seq_num;
3669         req.responder_resources = conn_param->responder_resources;
3670         req.initiator_depth = conn_param->initiator_depth;
3671         req.flow_control = conn_param->flow_control;
3672         req.retry_count = min_t(u8, 7, conn_param->retry_count);
3673         req.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3674         req.remote_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3675         req.local_cm_response_timeout = CMA_CM_RESPONSE_TIMEOUT;
3676         req.max_cm_retries = CMA_MAX_CM_RETRIES;
3677         req.srq = id_priv->srq ? 1 : 0;
3678
3679         ret = ib_send_cm_req(id_priv->cm_id.ib, &req);
3680 out:
3681         if (ret && !IS_ERR(id)) {
3682                 ib_destroy_cm_id(id);
3683                 id_priv->cm_id.ib = NULL;
3684         }
3685
3686         kfree(private_data);
3687         return ret;
3688 }
3689
3690 static int cma_connect_iw(struct rdma_id_private *id_priv,
3691                           struct rdma_conn_param *conn_param)
3692 {
3693         struct iw_cm_id *cm_id;
3694         int ret;
3695         struct iw_cm_conn_param iw_param;
3696
3697         cm_id = iw_create_cm_id(id_priv->id.device, cma_iw_handler, id_priv);
3698         if (IS_ERR(cm_id))
3699                 return PTR_ERR(cm_id);
3700
3701         cm_id->tos = id_priv->tos;
3702         id_priv->cm_id.iw = cm_id;
3703
3704         memcpy(&cm_id->local_addr, cma_src_addr(id_priv),
3705                rdma_addr_size(cma_src_addr(id_priv)));
3706         memcpy(&cm_id->remote_addr, cma_dst_addr(id_priv),
3707                rdma_addr_size(cma_dst_addr(id_priv)));
3708
3709         ret = cma_modify_qp_rtr(id_priv, conn_param);
3710         if (ret)
3711                 goto out;
3712
3713         if (conn_param) {
3714                 iw_param.ord = conn_param->initiator_depth;
3715                 iw_param.ird = conn_param->responder_resources;
3716                 iw_param.private_data = conn_param->private_data;
3717                 iw_param.private_data_len = conn_param->private_data_len;
3718                 iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
3719         } else {
3720                 memset(&iw_param, 0, sizeof iw_param);
3721                 iw_param.qpn = id_priv->qp_num;
3722         }
3723         ret = iw_cm_connect(cm_id, &iw_param);
3724 out:
3725         if (ret) {
3726                 iw_destroy_cm_id(cm_id);
3727                 id_priv->cm_id.iw = NULL;
3728         }
3729         return ret;
3730 }
3731
3732 int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
3733 {
3734         struct rdma_id_private *id_priv;
3735         int ret;
3736
3737         id_priv = container_of(id, struct rdma_id_private, id);
3738         if (!cma_comp_exch(id_priv, RDMA_CM_ROUTE_RESOLVED, RDMA_CM_CONNECT))
3739                 return -EINVAL;
3740
3741         if (!id->qp) {
3742                 id_priv->qp_num = conn_param->qp_num;
3743                 id_priv->srq = conn_param->srq;
3744         }
3745
3746         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3747                 if (id->qp_type == IB_QPT_UD)
3748                         ret = cma_resolve_ib_udp(id_priv, conn_param);
3749                 else
3750                         ret = cma_connect_ib(id_priv, conn_param);
3751         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3752                 ret = cma_connect_iw(id_priv, conn_param);
3753         else
3754                 ret = -ENOSYS;
3755         if (ret)
3756                 goto err;
3757
3758         return 0;
3759 err:
3760         cma_comp_exch(id_priv, RDMA_CM_CONNECT, RDMA_CM_ROUTE_RESOLVED);
3761         return ret;
3762 }
3763 EXPORT_SYMBOL(rdma_connect);
3764
3765 static int cma_accept_ib(struct rdma_id_private *id_priv,
3766                          struct rdma_conn_param *conn_param)
3767 {
3768         struct ib_cm_rep_param rep;
3769         int ret;
3770
3771         ret = cma_modify_qp_rtr(id_priv, conn_param);
3772         if (ret)
3773                 goto out;
3774
3775         ret = cma_modify_qp_rts(id_priv, conn_param);
3776         if (ret)
3777                 goto out;
3778
3779         memset(&rep, 0, sizeof rep);
3780         rep.qp_num = id_priv->qp_num;
3781         rep.starting_psn = id_priv->seq_num;
3782         rep.private_data = conn_param->private_data;
3783         rep.private_data_len = conn_param->private_data_len;
3784         rep.responder_resources = conn_param->responder_resources;
3785         rep.initiator_depth = conn_param->initiator_depth;
3786         rep.failover_accepted = 0;
3787         rep.flow_control = conn_param->flow_control;
3788         rep.rnr_retry_count = min_t(u8, 7, conn_param->rnr_retry_count);
3789         rep.srq = id_priv->srq ? 1 : 0;
3790
3791         ret = ib_send_cm_rep(id_priv->cm_id.ib, &rep);
3792 out:
3793         return ret;
3794 }
3795
3796 static int cma_accept_iw(struct rdma_id_private *id_priv,
3797                   struct rdma_conn_param *conn_param)
3798 {
3799         struct iw_cm_conn_param iw_param;
3800         int ret;
3801
3802         if (!conn_param)
3803                 return -EINVAL;
3804
3805         ret = cma_modify_qp_rtr(id_priv, conn_param);
3806         if (ret)
3807                 return ret;
3808
3809         iw_param.ord = conn_param->initiator_depth;
3810         iw_param.ird = conn_param->responder_resources;
3811         iw_param.private_data = conn_param->private_data;
3812         iw_param.private_data_len = conn_param->private_data_len;
3813         if (id_priv->id.qp) {
3814                 iw_param.qpn = id_priv->qp_num;
3815         } else
3816                 iw_param.qpn = conn_param->qp_num;
3817
3818         return iw_cm_accept(id_priv->cm_id.iw, &iw_param);
3819 }
3820
3821 static int cma_send_sidr_rep(struct rdma_id_private *id_priv,
3822                              enum ib_cm_sidr_status status, u32 qkey,
3823                              const void *private_data, int private_data_len)
3824 {
3825         struct ib_cm_sidr_rep_param rep;
3826         int ret;
3827
3828         memset(&rep, 0, sizeof rep);
3829         rep.status = status;
3830         if (status == IB_SIDR_SUCCESS) {
3831                 ret = cma_set_qkey(id_priv, qkey);
3832                 if (ret)
3833                         return ret;
3834                 rep.qp_num = id_priv->qp_num;
3835                 rep.qkey = id_priv->qkey;
3836         }
3837         rep.private_data = private_data;
3838         rep.private_data_len = private_data_len;
3839
3840         return ib_send_cm_sidr_rep(id_priv->cm_id.ib, &rep);
3841 }
3842
3843 int __rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param,
3844                   const char *caller)
3845 {
3846         struct rdma_id_private *id_priv;
3847         int ret;
3848
3849         id_priv = container_of(id, struct rdma_id_private, id);
3850
3851         if (caller)
3852                 id_priv->res.kern_name = caller;
3853         else
3854                 rdma_restrack_set_task(&id_priv->res, current);
3855
3856         if (!cma_comp(id_priv, RDMA_CM_CONNECT))
3857                 return -EINVAL;
3858
3859         if (!id->qp && conn_param) {
3860                 id_priv->qp_num = conn_param->qp_num;
3861                 id_priv->srq = conn_param->srq;
3862         }
3863
3864         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3865                 if (id->qp_type == IB_QPT_UD) {
3866                         if (conn_param)
3867                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3868                                                         conn_param->qkey,
3869                                                         conn_param->private_data,
3870                                                         conn_param->private_data_len);
3871                         else
3872                                 ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
3873                                                         0, NULL, 0);
3874                 } else {
3875                         if (conn_param)
3876                                 ret = cma_accept_ib(id_priv, conn_param);
3877                         else
3878                                 ret = cma_rep_recv(id_priv);
3879                 }
3880         } else if (rdma_cap_iw_cm(id->device, id->port_num))
3881                 ret = cma_accept_iw(id_priv, conn_param);
3882         else
3883                 ret = -ENOSYS;
3884
3885         if (ret)
3886                 goto reject;
3887
3888         return 0;
3889 reject:
3890         cma_modify_qp_err(id_priv);
3891         rdma_reject(id, NULL, 0);
3892         return ret;
3893 }
3894 EXPORT_SYMBOL(__rdma_accept);
3895
3896 int rdma_notify(struct rdma_cm_id *id, enum ib_event_type event)
3897 {
3898         struct rdma_id_private *id_priv;
3899         int ret;
3900
3901         id_priv = container_of(id, struct rdma_id_private, id);
3902         if (!id_priv->cm_id.ib)
3903                 return -EINVAL;
3904
3905         switch (id->device->node_type) {
3906         case RDMA_NODE_IB_CA:
3907                 ret = ib_cm_notify(id_priv->cm_id.ib, event);
3908                 break;
3909         default:
3910                 ret = 0;
3911                 break;
3912         }
3913         return ret;
3914 }
3915 EXPORT_SYMBOL(rdma_notify);
3916
3917 int rdma_reject(struct rdma_cm_id *id, const void *private_data,
3918                 u8 private_data_len)
3919 {
3920         struct rdma_id_private *id_priv;
3921         int ret;
3922
3923         id_priv = container_of(id, struct rdma_id_private, id);
3924         if (!id_priv->cm_id.ib)
3925                 return -EINVAL;
3926
3927         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3928                 if (id->qp_type == IB_QPT_UD)
3929                         ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, 0,
3930                                                 private_data, private_data_len);
3931                 else
3932                         ret = ib_send_cm_rej(id_priv->cm_id.ib,
3933                                              IB_CM_REJ_CONSUMER_DEFINED, NULL,
3934                                              0, private_data, private_data_len);
3935         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3936                 ret = iw_cm_reject(id_priv->cm_id.iw,
3937                                    private_data, private_data_len);
3938         } else
3939                 ret = -ENOSYS;
3940
3941         return ret;
3942 }
3943 EXPORT_SYMBOL(rdma_reject);
3944
3945 int rdma_disconnect(struct rdma_cm_id *id)
3946 {
3947         struct rdma_id_private *id_priv;
3948         int ret;
3949
3950         id_priv = container_of(id, struct rdma_id_private, id);
3951         if (!id_priv->cm_id.ib)
3952                 return -EINVAL;
3953
3954         if (rdma_cap_ib_cm(id->device, id->port_num)) {
3955                 ret = cma_modify_qp_err(id_priv);
3956                 if (ret)
3957                         goto out;
3958                 /* Initiate or respond to a disconnect. */
3959                 if (ib_send_cm_dreq(id_priv->cm_id.ib, NULL, 0))
3960                         ib_send_cm_drep(id_priv->cm_id.ib, NULL, 0);
3961         } else if (rdma_cap_iw_cm(id->device, id->port_num)) {
3962                 ret = iw_cm_disconnect(id_priv->cm_id.iw, 0);
3963         } else
3964                 ret = -EINVAL;
3965
3966 out:
3967         return ret;
3968 }
3969 EXPORT_SYMBOL(rdma_disconnect);
3970
3971 static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast)
3972 {
3973         struct rdma_id_private *id_priv;
3974         struct cma_multicast *mc = multicast->context;
3975         struct rdma_cm_event event;
3976         int ret = 0;
3977
3978         id_priv = mc->id_priv;
3979         mutex_lock(&id_priv->handler_mutex);
3980         if (id_priv->state != RDMA_CM_ADDR_BOUND &&
3981             id_priv->state != RDMA_CM_ADDR_RESOLVED)
3982                 goto out;
3983
3984         if (!status)
3985                 status = cma_set_qkey(id_priv, be32_to_cpu(multicast->rec.qkey));
3986         else
3987                 pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to join multicast. status %d\n",
3988                                      status);
3989         mutex_lock(&id_priv->qp_mutex);
3990         if (!status && id_priv->id.qp) {
3991                 status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid,
3992                                          be16_to_cpu(multicast->rec.mlid));
3993                 if (status)
3994                         pr_debug_ratelimited("RDMA CM: MULTICAST_ERROR: failed to attach QP. status %d\n",
3995                                              status);
3996         }
3997         mutex_unlock(&id_priv->qp_mutex);
3998
3999         memset(&event, 0, sizeof event);
4000         event.status = status;
4001         event.param.ud.private_data = mc->context;
4002         if (!status) {
4003                 struct rdma_dev_addr *dev_addr =
4004                         &id_priv->id.route.addr.dev_addr;
4005                 struct net_device *ndev =
4006                         dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4007                 enum ib_gid_type gid_type =
4008                         id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4009                         rdma_start_port(id_priv->cma_dev->device)];
4010
4011                 event.event = RDMA_CM_EVENT_MULTICAST_JOIN;
4012                 ret = ib_init_ah_from_mcmember(id_priv->id.device,
4013                                                id_priv->id.port_num,
4014                                                &multicast->rec,
4015                                                ndev, gid_type,
4016                                                &event.param.ud.ah_attr);
4017                 if (ret)
4018                         event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4019
4020                 event.param.ud.qp_num = 0xFFFFFF;
4021                 event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey);
4022                 if (ndev)
4023                         dev_put(ndev);
4024         } else
4025                 event.event = RDMA_CM_EVENT_MULTICAST_ERROR;
4026
4027         ret = id_priv->id.event_handler(&id_priv->id, &event);
4028
4029         rdma_destroy_ah_attr(&event.param.ud.ah_attr);
4030         if (ret) {
4031                 cma_exch(id_priv, RDMA_CM_DESTROYING);
4032                 mutex_unlock(&id_priv->handler_mutex);
4033                 rdma_destroy_id(&id_priv->id);
4034                 return 0;
4035         }
4036
4037 out:
4038         mutex_unlock(&id_priv->handler_mutex);
4039         return 0;
4040 }
4041
4042 static void cma_set_mgid(struct rdma_id_private *id_priv,
4043                          struct sockaddr *addr, union ib_gid *mgid)
4044 {
4045         unsigned char mc_map[MAX_ADDR_LEN];
4046         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4047         struct sockaddr_in *sin = (struct sockaddr_in *) addr;
4048         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr;
4049
4050         if (cma_any_addr(addr)) {
4051                 memset(mgid, 0, sizeof *mgid);
4052         } else if ((addr->sa_family == AF_INET6) &&
4053                    ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) ==
4054                                                                  0xFF10A01B)) {
4055                 /* IPv6 address is an SA assigned MGID. */
4056                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4057         } else if (addr->sa_family == AF_IB) {
4058                 memcpy(mgid, &((struct sockaddr_ib *) addr)->sib_addr, sizeof *mgid);
4059         } else if (addr->sa_family == AF_INET6) {
4060                 ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map);
4061                 if (id_priv->id.ps == RDMA_PS_UDP)
4062                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4063                 *mgid = *(union ib_gid *) (mc_map + 4);
4064         } else {
4065                 ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map);
4066                 if (id_priv->id.ps == RDMA_PS_UDP)
4067                         mc_map[7] = 0x01;       /* Use RDMA CM signature */
4068                 *mgid = *(union ib_gid *) (mc_map + 4);
4069         }
4070 }
4071
4072 static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
4073                                  struct cma_multicast *mc)
4074 {
4075         struct ib_sa_mcmember_rec rec;
4076         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4077         ib_sa_comp_mask comp_mask;
4078         int ret;
4079
4080         ib_addr_get_mgid(dev_addr, &rec.mgid);
4081         ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num,
4082                                      &rec.mgid, &rec);
4083         if (ret)
4084                 return ret;
4085
4086         ret = cma_set_qkey(id_priv, 0);
4087         if (ret)
4088                 return ret;
4089
4090         cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid);
4091         rec.qkey = cpu_to_be32(id_priv->qkey);
4092         rdma_addr_get_sgid(dev_addr, &rec.port_gid);
4093         rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
4094         rec.join_state = mc->join_state;
4095
4096         if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
4097             (!ib_sa_sendonly_fullmem_support(&sa_client,
4098                                              id_priv->id.device,
4099                                              id_priv->id.port_num))) {
4100                 pr_warn("RDMA CM: %s port %u Unable to multicast join\n"
4101                         "RDMA CM: SM doesn't support Send Only Full Member option\n",
4102                         id_priv->id.device->name, id_priv->id.port_num);
4103                 return -EOPNOTSUPP;
4104         }
4105
4106         comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
4107                     IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
4108                     IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
4109                     IB_SA_MCMEMBER_REC_FLOW_LABEL |
4110                     IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
4111
4112         if (id_priv->id.ps == RDMA_PS_IPOIB)
4113                 comp_mask |= IB_SA_MCMEMBER_REC_RATE |
4114                              IB_SA_MCMEMBER_REC_RATE_SELECTOR |
4115                              IB_SA_MCMEMBER_REC_MTU_SELECTOR |
4116                              IB_SA_MCMEMBER_REC_MTU |
4117                              IB_SA_MCMEMBER_REC_HOP_LIMIT;
4118
4119         mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device,
4120                                                 id_priv->id.port_num, &rec,
4121                                                 comp_mask, GFP_KERNEL,
4122                                                 cma_ib_mc_handler, mc);
4123         return PTR_ERR_OR_ZERO(mc->multicast.ib);
4124 }
4125
4126 static void iboe_mcast_work_handler(struct work_struct *work)
4127 {
4128         struct iboe_mcast_work *mw = container_of(work, struct iboe_mcast_work, work);
4129         struct cma_multicast *mc = mw->mc;
4130         struct ib_sa_multicast *m = mc->multicast.ib;
4131
4132         mc->multicast.ib->context = mc;
4133         cma_ib_mc_handler(0, m);
4134         kref_put(&mc->mcref, release_mc);
4135         kfree(mw);
4136 }
4137
4138 static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
4139                               enum ib_gid_type gid_type)
4140 {
4141         struct sockaddr_in *sin = (struct sockaddr_in *)addr;
4142         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
4143
4144         if (cma_any_addr(addr)) {
4145                 memset(mgid, 0, sizeof *mgid);
4146         } else if (addr->sa_family == AF_INET6) {
4147                 memcpy(mgid, &sin6->sin6_addr, sizeof *mgid);
4148         } else {
4149                 mgid->raw[0] =
4150                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0xff;
4151                 mgid->raw[1] =
4152                         (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) ? 0 : 0x0e;
4153                 mgid->raw[2] = 0;
4154                 mgid->raw[3] = 0;
4155                 mgid->raw[4] = 0;
4156                 mgid->raw[5] = 0;
4157                 mgid->raw[6] = 0;
4158                 mgid->raw[7] = 0;
4159                 mgid->raw[8] = 0;
4160                 mgid->raw[9] = 0;
4161                 mgid->raw[10] = 0xff;
4162                 mgid->raw[11] = 0xff;
4163                 *(__be32 *)(&mgid->raw[12]) = sin->sin_addr.s_addr;
4164         }
4165 }
4166
4167 static int cma_iboe_join_multicast(struct rdma_id_private *id_priv,
4168                                    struct cma_multicast *mc)
4169 {
4170         struct iboe_mcast_work *work;
4171         struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr;
4172         int err = 0;
4173         struct sockaddr *addr = (struct sockaddr *)&mc->addr;
4174         struct net_device *ndev = NULL;
4175         enum ib_gid_type gid_type;
4176         bool send_only;
4177
4178         send_only = mc->join_state == BIT(SENDONLY_FULLMEMBER_JOIN);
4179
4180         if (cma_zero_addr((struct sockaddr *)&mc->addr))
4181                 return -EINVAL;
4182
4183         work = kzalloc(sizeof *work, GFP_KERNEL);
4184         if (!work)
4185                 return -ENOMEM;
4186
4187         mc->multicast.ib = kzalloc(sizeof(struct ib_sa_multicast), GFP_KERNEL);
4188         if (!mc->multicast.ib) {
4189                 err = -ENOMEM;
4190                 goto out1;
4191         }
4192
4193         gid_type = id_priv->cma_dev->default_gid_type[id_priv->id.port_num -
4194                    rdma_start_port(id_priv->cma_dev->device)];
4195         cma_iboe_set_mgid(addr, &mc->multicast.ib->rec.mgid, gid_type);
4196
4197         mc->multicast.ib->rec.pkey = cpu_to_be16(0xffff);
4198         if (id_priv->id.ps == RDMA_PS_UDP)
4199                 mc->multicast.ib->rec.qkey = cpu_to_be32(RDMA_UDP_QKEY);
4200
4201         if (dev_addr->bound_dev_if)
4202                 ndev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
4203         if (!ndev) {
4204                 err = -ENODEV;
4205                 goto out2;
4206         }
4207         mc->multicast.ib->rec.rate = iboe_get_rate(ndev);
4208         mc->multicast.ib->rec.hop_limit = 1;
4209         mc->multicast.ib->rec.mtu = iboe_get_mtu(ndev->mtu);
4210
4211         if (addr->sa_family == AF_INET) {
4212                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
4213                         mc->multicast.ib->rec.hop_limit = IPV6_DEFAULT_HOPLIMIT;
4214                         if (!send_only) {
4215                                 err = cma_igmp_send(ndev, &mc->multicast.ib->rec.mgid,
4216                                                     true);
4217                         }
4218                 }
4219         } else {
4220                 if (gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP)
4221                         err = -ENOTSUPP;
4222         }
4223         dev_put(ndev);
4224         if (err || !mc->multicast.ib->rec.mtu) {
4225                 if (!err)
4226                         err = -EINVAL;
4227                 goto out2;
4228         }
4229         rdma_ip2gid((struct sockaddr *)&id_priv->id.route.addr.src_addr,
4230                     &mc->multicast.ib->rec.port_gid);
4231         work->id = id_priv;
4232         work->mc = mc;
4233         INIT_WORK(&work->work, iboe_mcast_work_handler);
4234         kref_get(&mc->mcref);
4235         queue_work(cma_wq, &work->work);
4236
4237         return 0;
4238
4239 out2:
4240         kfree(mc->multicast.ib);
4241 out1:
4242         kfree(work);
4243         return err;
4244 }
4245
4246 int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr,
4247                         u8 join_state, void *context)
4248 {
4249         struct rdma_id_private *id_priv;
4250         struct cma_multicast *mc;
4251         int ret;
4252
4253         if (!id->device)
4254                 return -EINVAL;
4255
4256         id_priv = container_of(id, struct rdma_id_private, id);
4257         if (!cma_comp(id_priv, RDMA_CM_ADDR_BOUND) &&
4258             !cma_comp(id_priv, RDMA_CM_ADDR_RESOLVED))
4259                 return -EINVAL;
4260
4261         mc = kmalloc(sizeof *mc, GFP_KERNEL);
4262         if (!mc)
4263                 return -ENOMEM;
4264
4265         memcpy(&mc->addr, addr, rdma_addr_size(addr));
4266         mc->context = context;
4267         mc->id_priv = id_priv;
4268         mc->join_state = join_state;
4269
4270         if (rdma_protocol_roce(id->device, id->port_num)) {
4271                 kref_init(&mc->mcref);
4272                 ret = cma_iboe_join_multicast(id_priv, mc);
4273                 if (ret)
4274                         goto out_err;
4275         } else if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4276                 ret = cma_join_ib_multicast(id_priv, mc);
4277                 if (ret)
4278                         goto out_err;
4279         } else {
4280                 ret = -ENOSYS;
4281                 goto out_err;
4282         }
4283
4284         spin_lock(&id_priv->lock);
4285         list_add(&mc->list, &id_priv->mc_list);
4286         spin_unlock(&id_priv->lock);
4287
4288         return 0;
4289 out_err:
4290         kfree(mc);
4291         return ret;
4292 }
4293 EXPORT_SYMBOL(rdma_join_multicast);
4294
4295 void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr)
4296 {
4297         struct rdma_id_private *id_priv;
4298         struct cma_multicast *mc;
4299
4300         id_priv = container_of(id, struct rdma_id_private, id);
4301         spin_lock_irq(&id_priv->lock);
4302         list_for_each_entry(mc, &id_priv->mc_list, list) {
4303                 if (!memcmp(&mc->addr, addr, rdma_addr_size(addr))) {
4304                         list_del(&mc->list);
4305                         spin_unlock_irq(&id_priv->lock);
4306
4307                         if (id->qp)
4308                                 ib_detach_mcast(id->qp,
4309                                                 &mc->multicast.ib->rec.mgid,
4310                                                 be16_to_cpu(mc->multicast.ib->rec.mlid));
4311
4312                         BUG_ON(id_priv->cma_dev->device != id->device);
4313
4314                         if (rdma_cap_ib_mcast(id->device, id->port_num)) {
4315                                 ib_sa_free_multicast(mc->multicast.ib);
4316                                 kfree(mc);
4317                         } else if (rdma_protocol_roce(id->device, id->port_num)) {
4318                                 cma_leave_roce_mc_group(id_priv, mc);
4319                         }
4320                         return;
4321                 }
4322         }
4323         spin_unlock_irq(&id_priv->lock);
4324 }
4325 EXPORT_SYMBOL(rdma_leave_multicast);
4326
4327 static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id_priv)
4328 {
4329         struct rdma_dev_addr *dev_addr;
4330         struct cma_ndev_work *work;
4331
4332         dev_addr = &id_priv->id.route.addr.dev_addr;
4333
4334         if ((dev_addr->bound_dev_if == ndev->ifindex) &&
4335             (net_eq(dev_net(ndev), dev_addr->net)) &&
4336             memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) {
4337                 pr_info("RDMA CM addr change for ndev %s used by id %p\n",
4338                         ndev->name, &id_priv->id);
4339                 work = kzalloc(sizeof *work, GFP_KERNEL);
4340                 if (!work)
4341                         return -ENOMEM;
4342
4343                 INIT_WORK(&work->work, cma_ndev_work_handler);
4344                 work->id = id_priv;
4345                 work->event.event = RDMA_CM_EVENT_ADDR_CHANGE;
4346                 atomic_inc(&id_priv->refcount);
4347                 queue_work(cma_wq, &work->work);
4348         }
4349
4350         return 0;
4351 }
4352
4353 static int cma_netdev_callback(struct notifier_block *self, unsigned long event,
4354                                void *ptr)
4355 {
4356         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
4357         struct cma_device *cma_dev;
4358         struct rdma_id_private *id_priv;
4359         int ret = NOTIFY_DONE;
4360
4361         if (event != NETDEV_BONDING_FAILOVER)
4362                 return NOTIFY_DONE;
4363
4364         if (!netif_is_bond_master(ndev))
4365                 return NOTIFY_DONE;
4366
4367         mutex_lock(&lock);
4368         list_for_each_entry(cma_dev, &dev_list, list)
4369                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4370                         ret = cma_netdev_change(ndev, id_priv);
4371                         if (ret)
4372                                 goto out;
4373                 }
4374
4375 out:
4376         mutex_unlock(&lock);
4377         return ret;
4378 }
4379
4380 static struct notifier_block cma_nb = {
4381         .notifier_call = cma_netdev_callback
4382 };
4383
4384 static void cma_add_one(struct ib_device *device)
4385 {
4386         struct cma_device *cma_dev;
4387         struct rdma_id_private *id_priv;
4388         unsigned int i;
4389         unsigned long supported_gids = 0;
4390
4391         cma_dev = kmalloc(sizeof *cma_dev, GFP_KERNEL);
4392         if (!cma_dev)
4393                 return;
4394
4395         cma_dev->device = device;
4396         cma_dev->default_gid_type = kcalloc(device->phys_port_cnt,
4397                                             sizeof(*cma_dev->default_gid_type),
4398                                             GFP_KERNEL);
4399         if (!cma_dev->default_gid_type)
4400                 goto free_cma_dev;
4401
4402         cma_dev->default_roce_tos = kcalloc(device->phys_port_cnt,
4403                                             sizeof(*cma_dev->default_roce_tos),
4404                                             GFP_KERNEL);
4405         if (!cma_dev->default_roce_tos)
4406                 goto free_gid_type;
4407
4408         for (i = rdma_start_port(device); i <= rdma_end_port(device); i++) {
4409                 supported_gids = roce_gid_type_mask_support(device, i);
4410                 WARN_ON(!supported_gids);
4411                 if (supported_gids & (1 << CMA_PREFERRED_ROCE_GID_TYPE))
4412                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4413                                 CMA_PREFERRED_ROCE_GID_TYPE;
4414                 else
4415                         cma_dev->default_gid_type[i - rdma_start_port(device)] =
4416                                 find_first_bit(&supported_gids, BITS_PER_LONG);
4417                 cma_dev->default_roce_tos[i - rdma_start_port(device)] = 0;
4418         }
4419
4420         init_completion(&cma_dev->comp);
4421         atomic_set(&cma_dev->refcount, 1);
4422         INIT_LIST_HEAD(&cma_dev->id_list);
4423         ib_set_client_data(device, &cma_client, cma_dev);
4424
4425         mutex_lock(&lock);
4426         list_add_tail(&cma_dev->list, &dev_list);
4427         list_for_each_entry(id_priv, &listen_any_list, list)
4428                 cma_listen_on_dev(id_priv, cma_dev);
4429         mutex_unlock(&lock);
4430
4431         return;
4432
4433 free_gid_type:
4434         kfree(cma_dev->default_gid_type);
4435
4436 free_cma_dev:
4437         kfree(cma_dev);
4438
4439         return;
4440 }
4441
4442 static int cma_remove_id_dev(struct rdma_id_private *id_priv)
4443 {
4444         struct rdma_cm_event event;
4445         enum rdma_cm_state state;
4446         int ret = 0;
4447
4448         /* Record that we want to remove the device */
4449         state = cma_exch(id_priv, RDMA_CM_DEVICE_REMOVAL);
4450         if (state == RDMA_CM_DESTROYING)
4451                 return 0;
4452
4453         cma_cancel_operation(id_priv, state);
4454         mutex_lock(&id_priv->handler_mutex);
4455
4456         /* Check for destruction from another callback. */
4457         if (!cma_comp(id_priv, RDMA_CM_DEVICE_REMOVAL))
4458                 goto out;
4459
4460         memset(&event, 0, sizeof event);
4461         event.event = RDMA_CM_EVENT_DEVICE_REMOVAL;
4462         ret = id_priv->id.event_handler(&id_priv->id, &event);
4463 out:
4464         mutex_unlock(&id_priv->handler_mutex);
4465         return ret;
4466 }
4467
4468 static void cma_process_remove(struct cma_device *cma_dev)
4469 {
4470         struct rdma_id_private *id_priv;
4471         int ret;
4472
4473         mutex_lock(&lock);
4474         while (!list_empty(&cma_dev->id_list)) {
4475                 id_priv = list_entry(cma_dev->id_list.next,
4476                                      struct rdma_id_private, list);
4477
4478                 list_del(&id_priv->listen_list);
4479                 list_del_init(&id_priv->list);
4480                 atomic_inc(&id_priv->refcount);
4481                 mutex_unlock(&lock);
4482
4483                 ret = id_priv->internal_id ? 1 : cma_remove_id_dev(id_priv);
4484                 cma_deref_id(id_priv);
4485                 if (ret)
4486                         rdma_destroy_id(&id_priv->id);
4487
4488                 mutex_lock(&lock);
4489         }
4490         mutex_unlock(&lock);
4491
4492         cma_deref_dev(cma_dev);
4493         wait_for_completion(&cma_dev->comp);
4494 }
4495
4496 static void cma_remove_one(struct ib_device *device, void *client_data)
4497 {
4498         struct cma_device *cma_dev = client_data;
4499
4500         if (!cma_dev)
4501                 return;
4502
4503         mutex_lock(&lock);
4504         list_del(&cma_dev->list);
4505         mutex_unlock(&lock);
4506
4507         cma_process_remove(cma_dev);
4508         kfree(cma_dev->default_roce_tos);
4509         kfree(cma_dev->default_gid_type);
4510         kfree(cma_dev);
4511 }
4512
4513 static int cma_get_id_stats(struct sk_buff *skb, struct netlink_callback *cb)
4514 {
4515         struct nlmsghdr *nlh;
4516         struct rdma_cm_id_stats *id_stats;
4517         struct rdma_id_private *id_priv;
4518         struct rdma_cm_id *id = NULL;
4519         struct cma_device *cma_dev;
4520         int i_dev = 0, i_id = 0;
4521
4522         /*
4523          * We export all of the IDs as a sequence of messages.  Each
4524          * ID gets its own netlink message.
4525          */
4526         mutex_lock(&lock);
4527
4528         list_for_each_entry(cma_dev, &dev_list, list) {
4529                 if (i_dev < cb->args[0]) {
4530                         i_dev++;
4531                         continue;
4532                 }
4533
4534                 i_id = 0;
4535                 list_for_each_entry(id_priv, &cma_dev->id_list, list) {
4536                         if (i_id < cb->args[1]) {
4537                                 i_id++;
4538                                 continue;
4539                         }
4540
4541                         id_stats = ibnl_put_msg(skb, &nlh, cb->nlh->nlmsg_seq,
4542                                                 sizeof *id_stats, RDMA_NL_RDMA_CM,
4543                                                 RDMA_NL_RDMA_CM_ID_STATS,
4544                                                 NLM_F_MULTI);
4545                         if (!id_stats)
4546                                 goto out;
4547
4548                         memset(id_stats, 0, sizeof *id_stats);
4549                         id = &id_priv->id;
4550                         id_stats->node_type = id->route.addr.dev_addr.dev_type;
4551                         id_stats->port_num = id->port_num;
4552                         id_stats->bound_dev_if =
4553                                 id->route.addr.dev_addr.bound_dev_if;
4554
4555                         if (ibnl_put_attr(skb, nlh,
4556                                           rdma_addr_size(cma_src_addr(id_priv)),
4557                                           cma_src_addr(id_priv),
4558                                           RDMA_NL_RDMA_CM_ATTR_SRC_ADDR))
4559                                 goto out;
4560                         if (ibnl_put_attr(skb, nlh,
4561                                           rdma_addr_size(cma_dst_addr(id_priv)),
4562                                           cma_dst_addr(id_priv),
4563                                           RDMA_NL_RDMA_CM_ATTR_DST_ADDR))
4564                                 goto out;
4565
4566                         id_stats->pid   = task_pid_vnr(id_priv->res.task);
4567                         id_stats->port_space    = id->ps;
4568                         id_stats->cm_state      = id_priv->state;
4569                         id_stats->qp_num        = id_priv->qp_num;
4570                         id_stats->qp_type       = id->qp_type;
4571
4572                         i_id++;
4573                         nlmsg_end(skb, nlh);
4574                 }
4575
4576                 cb->args[1] = 0;
4577                 i_dev++;
4578         }
4579
4580 out:
4581         mutex_unlock(&lock);
4582         cb->args[0] = i_dev;
4583         cb->args[1] = i_id;
4584
4585         return skb->len;
4586 }
4587
4588 static const struct rdma_nl_cbs cma_cb_table[RDMA_NL_RDMA_CM_NUM_OPS] = {
4589         [RDMA_NL_RDMA_CM_ID_STATS] = { .dump = cma_get_id_stats},
4590 };
4591
4592 static int cma_init_net(struct net *net)
4593 {
4594         struct cma_pernet *pernet = cma_pernet(net);
4595
4596         idr_init(&pernet->tcp_ps);
4597         idr_init(&pernet->udp_ps);
4598         idr_init(&pernet->ipoib_ps);
4599         idr_init(&pernet->ib_ps);
4600
4601         return 0;
4602 }
4603
4604 static void cma_exit_net(struct net *net)
4605 {
4606         struct cma_pernet *pernet = cma_pernet(net);
4607
4608         idr_destroy(&pernet->tcp_ps);
4609         idr_destroy(&pernet->udp_ps);
4610         idr_destroy(&pernet->ipoib_ps);
4611         idr_destroy(&pernet->ib_ps);
4612 }
4613
4614 static struct pernet_operations cma_pernet_operations = {
4615         .init = cma_init_net,
4616         .exit = cma_exit_net,
4617         .id = &cma_pernet_id,
4618         .size = sizeof(struct cma_pernet),
4619 };
4620
4621 static int __init cma_init(void)
4622 {
4623         int ret;
4624
4625         cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM);
4626         if (!cma_wq)
4627                 return -ENOMEM;
4628
4629         ret = register_pernet_subsys(&cma_pernet_operations);
4630         if (ret)
4631                 goto err_wq;
4632
4633         ib_sa_register_client(&sa_client);
4634         register_netdevice_notifier(&cma_nb);
4635
4636         ret = ib_register_client(&cma_client);
4637         if (ret)
4638                 goto err;
4639
4640         rdma_nl_register(RDMA_NL_RDMA_CM, cma_cb_table);
4641         cma_configfs_init();
4642
4643         return 0;
4644
4645 err:
4646         unregister_netdevice_notifier(&cma_nb);
4647         ib_sa_unregister_client(&sa_client);
4648 err_wq:
4649         destroy_workqueue(cma_wq);
4650         return ret;
4651 }
4652
4653 static void __exit cma_cleanup(void)
4654 {
4655         cma_configfs_exit();
4656         rdma_nl_unregister(RDMA_NL_RDMA_CM);
4657         ib_unregister_client(&cma_client);
4658         unregister_netdevice_notifier(&cma_nb);
4659         ib_sa_unregister_client(&sa_client);
4660         unregister_pernet_subsys(&cma_pernet_operations);
4661         destroy_workqueue(cma_wq);
4662 }
4663
4664 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_RDMA_CM, 1);
4665
4666 module_init(cma_init);
4667 module_exit(cma_cleanup);