drivers/infiniband/core/multicast.c

   1 /*
   2  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include <linux/completion.h>
  34 #include <linux/dma-mapping.h>
  35 #include <linux/err.h>
  36 #include <linux/interrupt.h>
  37 #include <linux/pci.h>
  38 #include <linux/bitops.h>
  39 #include <linux/random.h>
  40
  41 #include <rdma/ib_cache.h>
  42 #include "sa.h"
  43
  44 static void mcast_add_one(struct ib_device *device);
  45 static void mcast_remove_one(struct ib_device *device);
  46
  47 static struct ib_client mcast_client = {
  48         .name   = "ib_multicast",
  49         .add    = mcast_add_one,
  50         .remove = mcast_remove_one
  51 };
  52
  53 static struct ib_sa_client      sa_client;
  54 static struct workqueue_struct  *mcast_wq;
  55 static union ib_gid mgid0;
  56
  57 struct mcast_device;
  58
  59 struct mcast_port {
  60         struct mcast_device     *dev;
  61         spinlock_t              lock;
  62         struct rb_root          table;
  63         atomic_t                refcount;
  64         struct completion       comp;
  65         u8                      port_num;
  66 };
  67
  68 struct mcast_device {
  69         struct ib_device        *device;
  70         struct ib_event_handler event_handler;
  71         int                     start_port;
  72         int                     end_port;
  73         struct mcast_port       port[0];
  74 };
  75
  76 enum mcast_state {
  77         MCAST_IDLE,
  78         MCAST_JOINING,
  79         MCAST_MEMBER,
  80         MCAST_BUSY,
  81         MCAST_ERROR
  82 };
  83
  84 struct mcast_member;
  85
  86 struct mcast_group {
  87         struct ib_sa_mcmember_rec rec;
  88         struct rb_node          node;
  89         struct mcast_port       *port;
  90         spinlock_t              lock;
  91         struct work_struct      work;
  92         struct list_head        pending_list;
  93         struct list_head        active_list;
  94         struct mcast_member     *last_join;
  95         int                     members[3];
  96         atomic_t                refcount;
  97         enum mcast_state        state;
  98         struct ib_sa_query      *query;
  99         int                     query_id;
 100 };
 101
 102 struct mcast_member {
 103         struct ib_sa_multicast  multicast;
 104         struct ib_sa_client     *client;
 105         struct mcast_group      *group;
 106         struct list_head        list;
 107         enum mcast_state        state;
 108         atomic_t                refcount;
 109         struct completion       comp;
 110 };
 111
 112 static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
 113                          void *context);
 114 static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
 115                           void *context);
 116
 117 static struct mcast_group *mcast_find(struct mcast_port *port,
 118                                       union ib_gid *mgid)
 119 {
 120         struct rb_node *node = port->table.rb_node;
 121         struct mcast_group *group;
 122         int ret;
 123
 124         while (node) {
 125                 group = rb_entry(node, struct mcast_group, node);
 126                 ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid);
 127                 if (!ret)
 128                         return group;
 129
 130                 if (ret < 0)
 131                         node = node->rb_left;
 132                 else
 133                         node = node->rb_right;
 134         }
 135         return NULL;
 136 }
 137
 138 static struct mcast_group *mcast_insert(struct mcast_port *port,
 139                                         struct mcast_group *group,
 140                                         int allow_duplicates)
 141 {
 142         struct rb_node **link = &port->table.rb_node;
 143         struct rb_node *parent = NULL;
 144         struct mcast_group *cur_group;
 145         int ret;
 146
 147         while (*link) {
 148                 parent = *link;
 149                 cur_group = rb_entry(parent, struct mcast_group, node);
 150
 151                 ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw,
 152                              sizeof group->rec.mgid);
 153                 if (ret < 0)
 154                         link = &(*link)->rb_left;
 155                 else if (ret > 0)
 156                         link = &(*link)->rb_right;
 157                 else if (allow_duplicates)
 158                         link = &(*link)->rb_left;
 159                 else
 160                         return cur_group;
 161         }
 162         rb_link_node(&group->node, parent, link);
 163         rb_insert_color(&group->node, &port->table);
 164         return NULL;
 165 }
 166
 167 static void deref_port(struct mcast_port *port)
 168 {
 169         if (atomic_dec_and_test(&port->refcount))
 170                 complete(&port->comp);
 171 }
 172
 173 static void release_group(struct mcast_group *group)
 174 {
 175         struct mcast_port *port = group->port;
 176         unsigned long flags;
 177
 178         spin_lock_irqsave(&port->lock, flags);
 179         if (atomic_dec_and_test(&group->refcount)) {
 180                 rb_erase(&group->node, &port->table);
 181                 spin_unlock_irqrestore(&port->lock, flags);
 182                 kfree(group);
 183                 deref_port(port);
 184         } else
 185                 spin_unlock_irqrestore(&port->lock, flags);
 186 }
 187
 188 static void deref_member(struct mcast_member *member)
 189 {
 190         if (atomic_dec_and_test(&member->refcount))
 191                 complete(&member->comp);
 192 }
 193
 194 static void queue_join(struct mcast_member *member)
 195 {
 196         struct mcast_group *group = member->group;
 197         unsigned long flags;
 198
 199         spin_lock_irqsave(&group->lock, flags);
 200         list_add(&member->list, &group->pending_list);
 201         if (group->state == MCAST_IDLE) {
 202                 group->state = MCAST_BUSY;
 203                 atomic_inc(&group->refcount);
 204                 queue_work(mcast_wq, &group->work);
 205         }
 206         spin_unlock_irqrestore(&group->lock, flags);
 207 }
 208
 209 /*
 210  * A multicast group has three types of members: full member, non member, and
 211  * send only member.  We need to keep track of the number of members of each
 212  * type based on their join state.  Adjust the number of members the belong to
 213  * the specified join states.
 214  */
 215 static void adjust_membership(struct mcast_group *group, u8 join_state, int inc)
 216 {
 217         int i;
 218
 219         for (i = 0; i < 3; i++, join_state >>= 1)
 220                 if (join_state & 0x1)
 221                         group->members[i] += inc;
 222 }
 223
 224 /*
 225  * If a multicast group has zero members left for a particular join state, but
 226  * the group is still a member with the SA, we need to leave that join state.
 227  * Determine which join states we still belong to, but that do not have any
 228  * active members.
 229  */
 230 static u8 get_leave_state(struct mcast_group *group)
 231 {
 232         u8 leave_state = 0;
 233         int i;
 234
 235         for (i = 0; i < 3; i++)
 236                 if (!group->members[i])
 237                         leave_state |= (0x1 << i);
 238
 239         return leave_state & group->rec.join_state;
 240 }
 241
 242 static int check_selector(ib_sa_comp_mask comp_mask,
 243                           ib_sa_comp_mask selector_mask,
 244                           ib_sa_comp_mask value_mask,
 245                           u8 selector, u8 src_value, u8 dst_value)
 246 {
 247         int err;
 248
 249         if (!(comp_mask & selector_mask) || !(comp_mask & value_mask))
 250                 return 0;
 251
 252         switch (selector) {
 253         case IB_SA_GT:
 254                 err = (src_value <= dst_value);
 255                 break;
 256         case IB_SA_LT:
 257                 err = (src_value >= dst_value);
 258                 break;
 259         case IB_SA_EQ:
 260                 err = (src_value != dst_value);
 261                 break;
 262         default:
 263                 err = 0;
 264                 break;
 265         }
 266
 267         return err;
 268 }
 269
 270 static int cmp_rec(struct ib_sa_mcmember_rec *src,
 271                    struct ib_sa_mcmember_rec *dst, ib_sa_comp_mask comp_mask)
 272 {
 273         /* MGID must already match */
 274
 275         if (comp_mask & IB_SA_MCMEMBER_REC_PORT_GID &&
 276             memcmp(&src->port_gid, &dst->port_gid, sizeof src->port_gid))
 277                 return -EINVAL;
 278         if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey)
 279                 return -EINVAL;
 280         if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid)
 281                 return -EINVAL;
 282         if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR,
 283                            IB_SA_MCMEMBER_REC_MTU, dst->mtu_selector,
 284                            src->mtu, dst->mtu))
 285                 return -EINVAL;
 286         if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS &&
 287             src->traffic_class != dst->traffic_class)
 288                 return -EINVAL;
 289         if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey)
 290                 return -EINVAL;
 291         if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR,
 292                            IB_SA_MCMEMBER_REC_RATE, dst->rate_selector,
 293                            src->rate, dst->rate))
 294                 return -EINVAL;
 295         if (check_selector(comp_mask,
 296                            IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR,
 297                            IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME,
 298                            dst->packet_life_time_selector,
 299                            src->packet_life_time, dst->packet_life_time))
 300                 return -EINVAL;
 301         if (comp_mask & IB_SA_MCMEMBER_REC_SL && src->sl != dst->sl)
 302                 return -EINVAL;
 303         if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL &&
 304             src->flow_label != dst->flow_label)
 305                 return -EINVAL;
 306         if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT &&
 307             src->hop_limit != dst->hop_limit)
 308                 return -EINVAL;
 309         if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE && src->scope != dst->scope)
 310                 return -EINVAL;
 311
 312         /* join_state checked separately, proxy_join ignored */
 313
 314         return 0;
 315 }
 316
 317 static int send_join(struct mcast_group *group, struct mcast_member *member)
 318 {
 319         struct mcast_port *port = group->port;
 320         int ret;
 321
 322         group->last_join = member;
 323         ret = ib_sa_mcmember_rec_query(&sa_client, port->dev->device,
 324                                        port->port_num, IB_MGMT_METHOD_SET,
 325                                        &member->multicast.rec,
 326                                        member->multicast.comp_mask,
 327                                        3000, GFP_KERNEL, join_handler, group,
 328                                        &group->query);
 329         if (ret >= 0) {
 330                 group->query_id = ret;
 331                 ret = 0;
 332         }
 333         return ret;
 334 }
 335
 336 static int send_leave(struct mcast_group *group, u8 leave_state)
 337 {
 338         struct mcast_port *port = group->port;
 339         struct ib_sa_mcmember_rec rec;
 340         int ret;
 341
 342         rec = group->rec;
 343         rec.join_state = leave_state;
 344
 345         ret = ib_sa_mcmember_rec_query(&sa_client, port->dev->device,
 346                                        port->port_num, IB_SA_METHOD_DELETE, &rec,
 347                                        IB_SA_MCMEMBER_REC_MGID     |
 348                                        IB_SA_MCMEMBER_REC_PORT_GID |
 349                                        IB_SA_MCMEMBER_REC_JOIN_STATE,
 350                                        3000, GFP_KERNEL, leave_handler,
 351                                        group, &group->query);
 352         if (ret >= 0) {
 353                 group->query_id = ret;
 354                 ret = 0;
 355         }
 356         return ret;
 357 }
 358
 359 static void join_group(struct mcast_group *group, struct mcast_member *member,
 360                        u8 join_state)
 361 {
 362         member->state = MCAST_MEMBER;
 363         adjust_membership(group, join_state, 1);
 364         group->rec.join_state |= join_state;
 365         member->multicast.rec = group->rec;
 366         member->multicast.rec.join_state = join_state;
 367         list_move(&member->list, &group->active_list);
 368 }
 369
 370 static int fail_join(struct mcast_group *group, struct mcast_member *member,
 371                      int status)
 372 {
 373         spin_lock_irq(&group->lock);
 374         list_del_init(&member->list);
 375         spin_unlock_irq(&group->lock);
 376         return member->multicast.callback(status, &member->multicast);
 377 }
 378
 379 static void process_group_error(struct mcast_group *group)
 380 {
 381         struct mcast_member *member;
 382         int ret;
 383
 384         spin_lock_irq(&group->lock);
 385         while (!list_empty(&group->active_list)) {
 386                 member = list_entry(group->active_list.next,
 387                                     struct mcast_member, list);
 388                 atomic_inc(&member->refcount);
 389                 list_del_init(&member->list);
 390                 adjust_membership(group, member->multicast.rec.join_state, -1);
 391                 member->state = MCAST_ERROR;
 392                 spin_unlock_irq(&group->lock);
 393
 394                 ret = member->multicast.callback(-ENETRESET,
 395                                                  &member->multicast);
 396                 deref_member(member);
 397                 if (ret)
 398                         ib_sa_free_multicast(&member->multicast);
 399                 spin_lock_irq(&group->lock);
 400         }
 401
 402         group->rec.join_state = 0;
 403         group->state = MCAST_BUSY;
 404         spin_unlock_irq(&group->lock);
 405 }
 406
 407 static void mcast_work_handler(struct work_struct *work)
 408 {
 409         struct mcast_group *group;
 410         struct mcast_member *member;
 411         struct ib_sa_multicast *multicast;
 412         int status, ret;
 413         u8 join_state;
 414
 415         group = container_of(work, typeof(*group), work);
 416 retest:
 417         spin_lock_irq(&group->lock);
 418         while (!list_empty(&group->pending_list) ||
 419                (group->state == MCAST_ERROR)) {
 420
 421                 if (group->state == MCAST_ERROR) {
 422                         spin_unlock_irq(&group->lock);
 423                         process_group_error(group);
 424                         goto retest;
 425                 }
 426
 427                 member = list_entry(group->pending_list.next,
 428                                     struct mcast_member, list);
 429                 multicast = &member->multicast;
 430                 join_state = multicast->rec.join_state;
 431                 atomic_inc(&member->refcount);
 432
 433                 if (join_state == (group->rec.join_state & join_state)) {
 434                         status = cmp_rec(&group->rec, &multicast->rec,
 435                                          multicast->comp_mask);
 436                         if (!status)
 437                                 join_group(group, member, join_state);
 438                         else
 439                                 list_del_init(&member->list);
 440                         spin_unlock_irq(&group->lock);
 441                         ret = multicast->callback(status, multicast);
 442                 } else {
 443                         spin_unlock_irq(&group->lock);
 444                         status = send_join(group, member);
 445                         if (!status) {
 446                                 deref_member(member);
 447                                 return;
 448                         }
 449                         ret = fail_join(group, member, status);
 450                 }
 451
 452                 deref_member(member);
 453                 if (ret)
 454                         ib_sa_free_multicast(&member->multicast);
 455                 spin_lock_irq(&group->lock);
 456         }
 457
 458         join_state = get_leave_state(group);
 459         if (join_state) {
 460                 group->rec.join_state &= ~join_state;
 461                 spin_unlock_irq(&group->lock);
 462                 if (send_leave(group, join_state))
 463                         goto retest;
 464         } else {
 465                 group->state = MCAST_IDLE;
 466                 spin_unlock_irq(&group->lock);
 467                 release_group(group);
 468         }
 469 }
 470
 471 /*
 472  * Fail a join request if it is still active - at the head of the pending queue.
 473  */
 474 static void process_join_error(struct mcast_group *group, int status)
 475 {
 476         struct mcast_member *member;
 477         int ret;
 478
 479         spin_lock_irq(&group->lock);
 480         member = list_entry(group->pending_list.next,
 481                             struct mcast_member, list);
 482         if (group->last_join == member) {
 483                 atomic_inc(&member->refcount);
 484                 list_del_init(&member->list);
 485                 spin_unlock_irq(&group->lock);
 486                 ret = member->multicast.callback(status, &member->multicast);
 487                 deref_member(member);
 488                 if (ret)
 489                         ib_sa_free_multicast(&member->multicast);
 490         } else
 491                 spin_unlock_irq(&group->lock);
 492 }
 493
 494 static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
 495                          void *context)
 496 {
 497         struct mcast_group *group = context;
 498
 499         if (status)
 500                 process_join_error(group, status);
 501         else {
 502                 spin_lock_irq(&group->port->lock);
 503                 group->rec = *rec;
 504                 if (!memcmp(&mgid0, &group->rec.mgid, sizeof mgid0)) {
 505                         rb_erase(&group->node, &group->port->table);
 506                         mcast_insert(group->port, group, 1);
 507                 }
 508                 spin_unlock_irq(&group->port->lock);
 509         }
 510         mcast_work_handler(&group->work);
 511 }
 512
 513 static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
 514                           void *context)
 515 {
 516         struct mcast_group *group = context;
 517
 518         mcast_work_handler(&group->work);
 519 }
 520
 521 static struct mcast_group *acquire_group(struct mcast_port *port,
 522                                          union ib_gid *mgid, gfp_t gfp_mask)
 523 {
 524         struct mcast_group *group, *cur_group;
 525         unsigned long flags;
 526         int is_mgid0;
 527
 528         is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
 529         if (!is_mgid0) {
 530                 spin_lock_irqsave(&port->lock, flags);
 531                 group = mcast_find(port, mgid);
 532                 if (group)
 533                         goto found;
 534                 spin_unlock_irqrestore(&port->lock, flags);
 535         }
 536
 537         group = kzalloc(sizeof *group, gfp_mask);
 538         if (!group)
 539                 return NULL;
 540
 541         group->port = port;
 542         group->rec.mgid = *mgid;
 543         INIT_LIST_HEAD(&group->pending_list);
 544         INIT_LIST_HEAD(&group->active_list);
 545         INIT_WORK(&group->work, mcast_work_handler);
 546         spin_lock_init(&group->lock);
 547
 548         spin_lock_irqsave(&port->lock, flags);
 549         cur_group = mcast_insert(port, group, is_mgid0);
 550         if (cur_group) {
 551                 kfree(group);
 552                 group = cur_group;
 553         } else
 554                 atomic_inc(&port->refcount);
 555 found:
 556         atomic_inc(&group->refcount);
 557         spin_unlock_irqrestore(&port->lock, flags);
 558         return group;
 559 }
 560
 561 /*
 562  * We serialize all join requests to a single group to make our lives much
 563  * easier.  Otherwise, two users could try to join the same group
 564  * simultaneously, with different configurations, one could leave while the
 565  * join is in progress, etc., which makes locking around error recovery
 566  * difficult.
 567  */
 568 struct ib_sa_multicast *
 569 ib_sa_join_multicast(struct ib_sa_client *client,
 570                      struct ib_device *device, u8 port_num,
 571                      struct ib_sa_mcmember_rec *rec,
 572                      ib_sa_comp_mask comp_mask, gfp_t gfp_mask,
 573                      int (*callback)(int status,
 574                                      struct ib_sa_multicast *multicast),
 575                      void *context)
 576 {
 577         struct mcast_device *dev;
 578         struct mcast_member *member;
 579         struct ib_sa_multicast *multicast;
 580         int ret;
 581
 582         dev = ib_get_client_data(device, &mcast_client);
 583         if (!dev)
 584                 return ERR_PTR(-ENODEV);
 585
 586         member = kmalloc(sizeof *member, gfp_mask);
 587         if (!member)
 588                 return ERR_PTR(-ENOMEM);
 589
 590         ib_sa_client_get(client);
 591         member->client = client;
 592         member->multicast.rec = *rec;
 593         member->multicast.comp_mask = comp_mask;
 594         member->multicast.callback = callback;
 595         member->multicast.context = context;
 596         init_completion(&member->comp);
 597         atomic_set(&member->refcount, 1);
 598         member->state = MCAST_JOINING;
 599
 600         member->group = acquire_group(&dev->port[port_num - dev->start_port],
 601                                       &rec->mgid, gfp_mask);
 602         if (!member->group) {
 603                 ret = -ENOMEM;
 604                 goto err;
 605         }
 606
 607         /*
 608          * The user will get the multicast structure in their callback.  They
 609          * could then free the multicast structure before we can return from
 610          * this routine.  So we save the pointer to return before queuing
 611          * any callback.
 612          */
 613         multicast = &member->multicast;
 614         queue_join(member);
 615         return multicast;
 616
 617 err:
 618         ib_sa_client_put(client);
 619         kfree(member);
 620         return ERR_PTR(ret);
 621 }
 622 EXPORT_SYMBOL(ib_sa_join_multicast);
 623
 624 void ib_sa_free_multicast(struct ib_sa_multicast *multicast)
 625 {
 626         struct mcast_member *member;
 627         struct mcast_group *group;
 628
 629         member = container_of(multicast, struct mcast_member, multicast);
 630         group = member->group;
 631
 632         spin_lock_irq(&group->lock);
 633         if (member->state == MCAST_MEMBER)
 634                 adjust_membership(group, multicast->rec.join_state, -1);
 635
 636         list_del_init(&member->list);
 637
 638         if (group->state == MCAST_IDLE) {
 639                 group->state = MCAST_BUSY;
 640                 spin_unlock_irq(&group->lock);
 641                 /* Continue to hold reference on group until callback */
 642                 queue_work(mcast_wq, &group->work);
 643         } else {
 644                 spin_unlock_irq(&group->lock);
 645                 release_group(group);
 646         }
 647
 648         deref_member(member);
 649         wait_for_completion(&member->comp);
 650         ib_sa_client_put(member->client);
 651         kfree(member);
 652 }
 653 EXPORT_SYMBOL(ib_sa_free_multicast);
 654
 655 int ib_sa_get_mcmember_rec(struct ib_device *device, u8 port_num,
 656                            union ib_gid *mgid, struct ib_sa_mcmember_rec *rec)
 657 {
 658         struct mcast_device *dev;
 659         struct mcast_port *port;
 660         struct mcast_group *group;
 661         unsigned long flags;
 662         int ret = 0;
 663
 664         dev = ib_get_client_data(device, &mcast_client);
 665         if (!dev)
 666                 return -ENODEV;
 667
 668         port = &dev->port[port_num - dev->start_port];
 669         spin_lock_irqsave(&port->lock, flags);
 670         group = mcast_find(port, mgid);
 671         if (group)
 672                 *rec = group->rec;
 673         else
 674                 ret = -EADDRNOTAVAIL;
 675         spin_unlock_irqrestore(&port->lock, flags);
 676
 677         return ret;
 678 }
 679 EXPORT_SYMBOL(ib_sa_get_mcmember_rec);
 680
 681 int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num,
 682                              struct ib_sa_mcmember_rec *rec,
 683                              struct ib_ah_attr *ah_attr)
 684 {
 685         int ret;
 686         u16 gid_index;
 687         u8 p;
 688
 689         ret = ib_find_cached_gid(device, &rec->port_gid, &p, &gid_index);
 690         if (ret)
 691                 return ret;
 692
 693         memset(ah_attr, 0, sizeof *ah_attr);
 694         ah_attr->dlid = be16_to_cpu(rec->mlid);
 695         ah_attr->sl = rec->sl;
 696         ah_attr->port_num = port_num;
 697         ah_attr->static_rate = rec->rate;
 698
 699         ah_attr->ah_flags = IB_AH_GRH;
 700         ah_attr->grh.dgid = rec->mgid;
 701
 702         ah_attr->grh.sgid_index = (u8) gid_index;
 703         ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label);
 704         ah_attr->grh.hop_limit = rec->hop_limit;
 705         ah_attr->grh.traffic_class = rec->traffic_class;
 706
 707         return 0;
 708 }
 709 EXPORT_SYMBOL(ib_init_ah_from_mcmember);
 710
 711 static void mcast_groups_lost(struct mcast_port *port)
 712 {
 713         struct mcast_group *group;
 714         struct rb_node *node;
 715         unsigned long flags;
 716
 717         spin_lock_irqsave(&port->lock, flags);
 718         for (node = rb_first(&port->table); node; node = rb_next(node)) {
 719                 group = rb_entry(node, struct mcast_group, node);
 720                 spin_lock(&group->lock);
 721                 if (group->state == MCAST_IDLE) {
 722                         atomic_inc(&group->refcount);
 723                         queue_work(mcast_wq, &group->work);
 724                 }
 725                 group->state = MCAST_ERROR;
 726                 spin_unlock(&group->lock);
 727         }
 728         spin_unlock_irqrestore(&port->lock, flags);
 729 }
 730
 731 static void mcast_event_handler(struct ib_event_handler *handler,
 732                                 struct ib_event *event)
 733 {
 734         struct mcast_device *dev;
 735
 736         dev = container_of(handler, struct mcast_device, event_handler);
 737
 738         switch (event->event) {
 739         case IB_EVENT_PORT_ERR:
 740         case IB_EVENT_LID_CHANGE:
 741         case IB_EVENT_SM_CHANGE:
 742         case IB_EVENT_CLIENT_REREGISTER:
 743                 mcast_groups_lost(&dev->port[event->element.port_num -
 744                                              dev->start_port]);
 745                 break;
 746         default:
 747                 break;
 748         }
 749 }
 750
 751 static void mcast_add_one(struct ib_device *device)
 752 {
 753         struct mcast_device *dev;
 754         struct mcast_port *port;
 755         int i;
 756
 757         if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
 758                 return;
 759
 760         dev = kmalloc(sizeof *dev + device->phys_port_cnt * sizeof *port,
 761                       GFP_KERNEL);
 762         if (!dev)
 763                 return;
 764
 765         if (device->node_type == RDMA_NODE_IB_SWITCH)
 766                 dev->start_port = dev->end_port = 0;
 767         else {
 768                 dev->start_port = 1;
 769                 dev->end_port = device->phys_port_cnt;
 770         }
 771
 772         for (i = 0; i <= dev->end_port - dev->start_port; i++) {
 773                 port = &dev->port[i];
 774                 port->dev = dev;
 775                 port->port_num = dev->start_port + i;
 776                 spin_lock_init(&port->lock);
 777                 port->table = RB_ROOT;
 778                 init_completion(&port->comp);
 779                 atomic_set(&port->refcount, 1);
 780         }
 781
 782         dev->device = device;
 783         ib_set_client_data(device, &mcast_client, dev);
 784
 785         INIT_IB_EVENT_HANDLER(&dev->event_handler, device, mcast_event_handler);
 786         ib_register_event_handler(&dev->event_handler);
 787 }
 788
 789 static void mcast_remove_one(struct ib_device *device)
 790 {
 791         struct mcast_device *dev;
 792         struct mcast_port *port;
 793         int i;
 794
 795         dev = ib_get_client_data(device, &mcast_client);
 796         if (!dev)
 797                 return;
 798
 799         ib_unregister_event_handler(&dev->event_handler);
 800         flush_workqueue(mcast_wq);
 801
 802         for (i = 0; i <= dev->end_port - dev->start_port; i++) {
 803                 port = &dev->port[i];
 804                 deref_port(port);
 805                 wait_for_completion(&port->comp);
 806         }
 807
 808         kfree(dev);
 809 }
 810
 811 int mcast_init(void)
 812 {
 813         int ret;
 814
 815         mcast_wq = create_singlethread_workqueue("ib_mcast");
 816         if (!mcast_wq)
 817                 return -ENOMEM;
 818
 819         ib_sa_register_client(&sa_client);
 820
 821         ret = ib_register_client(&mcast_client);
 822         if (ret)
 823                 goto err;
 824         return 0;
 825
 826 err:
 827         ib_sa_unregister_client(&sa_client);
 828         destroy_workqueue(mcast_wq);
 829         return ret;
 830 }
 831
 832 void mcast_cleanup(void)
 833 {
 834         ib_unregister_client(&mcast_client);
 835         ib_sa_unregister_client(&sa_client);
 836         destroy_workqueue(mcast_wq);
 837 }