fs/ocfs2/dlm/dlmdomain.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmdomain.c
   5  *
   6  * defines domain join / leave apis
   7  *
   8  * Copyright (C) 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  *
  25  */
  26
  27 #include <linux/module.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/utsname.h>
  32 #include <linux/init.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/delay.h>
  35 #include <linux/err.h>
  36
  37 #include "cluster/heartbeat.h"
  38 #include "cluster/nodemanager.h"
  39 #include "cluster/tcp.h"
  40
  41 #include "dlmapi.h"
  42 #include "dlmcommon.h"
  43
  44 #include "dlmdebug.h"
  45 #include "dlmdomain.h"
  46
  47 #include "dlmver.h"
  48
  49 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
  50 #include "cluster/masklog.h"
  51
  52 /*
  53  *
  54  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
  55  *    dlm_domain_lock
  56  *    struct dlm_ctxt->spinlock
  57  *    struct dlm_lock_resource->spinlock
  58  *    struct dlm_ctxt->master_lock
  59  *    struct dlm_ctxt->ast_lock
  60  *    dlm_master_list_entry->spinlock
  61  *    dlm_lock->spinlock
  62  *
  63  */
  64
  65 spinlock_t dlm_domain_lock = SPIN_LOCK_UNLOCKED;
  66 LIST_HEAD(dlm_domains);
  67 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
  68
  69 #define DLM_DOMAIN_BACKOFF_MS 200
  70
  71 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data);
  72 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data);
  73 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data);
  74 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data);
  75
  76 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
  77
  78 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
  79 {
  80         hlist_del_init(&lockres->hash_node);
  81         dlm_lockres_put(lockres);
  82 }
  83
  84 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
  85                        struct dlm_lock_resource *res)
  86 {
  87         struct hlist_head *bucket;
  88         struct qstr *q;
  89
  90         assert_spin_locked(&dlm->spinlock);
  91
  92         q = &res->lockname;
  93         bucket = &(dlm->lockres_hash[q->hash % DLM_HASH_BUCKETS]);
  94
  95         /* get a reference for our hashtable */
  96         dlm_lockres_get(res);
  97
  98         hlist_add_head(&res->hash_node, bucket);
  99 }
 100
 101 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
 102                                                 const char *name,
 103                                                 unsigned int len,
 104                                                 unsigned int hash)
 105 {
 106         struct hlist_head *bucket;
 107         struct hlist_node *list;
 108
 109         mlog_entry("%.*s\n", len, name);
 110
 111         assert_spin_locked(&dlm->spinlock);
 112
 113         bucket = dlm->lockres_hash + full_name_hash(name, len) % DLM_HASH_BUCKETS;
 114         hlist_for_each(list, bucket) {
 115                 struct dlm_lock_resource *res = hlist_entry(list,
 116                         struct dlm_lock_resource, hash_node);
 117                 if (res->lockname.name[0] != name[0])
 118                         continue;
 119                 if (unlikely(res->lockname.len != len))
 120                         continue;
 121                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
 122                         continue;
 123                 dlm_lockres_get(res);
 124                 return res;
 125         }
 126         return NULL;
 127 }
 128
 129 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
 130                                     const char *name,
 131                                     unsigned int len)
 132 {
 133         struct dlm_lock_resource *res;
 134         unsigned int hash = dlm_lockid_hash(name, len);
 135
 136         spin_lock(&dlm->spinlock);
 137         res = __dlm_lookup_lockres(dlm, name, len, hash);
 138         spin_unlock(&dlm->spinlock);
 139         return res;
 140 }
 141
 142 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
 143 {
 144         struct dlm_ctxt *tmp = NULL;
 145         struct list_head *iter;
 146
 147         assert_spin_locked(&dlm_domain_lock);
 148
 149         /* tmp->name here is always NULL terminated,
 150          * but domain may not be! */
 151         list_for_each(iter, &dlm_domains) {
 152                 tmp = list_entry (iter, struct dlm_ctxt, list);
 153                 if (strlen(tmp->name) == len &&
 154                     memcmp(tmp->name, domain, len)==0)
 155                         break;
 156                 tmp = NULL;
 157         }
 158
 159         return tmp;
 160 }
 161
 162 /* For null terminated domain strings ONLY */
 163 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
 164 {
 165         assert_spin_locked(&dlm_domain_lock);
 166
 167         return __dlm_lookup_domain_full(domain, strlen(domain));
 168 }
 169
 170
 171 /* returns true on one of two conditions:
 172  * 1) the domain does not exist
 173  * 2) the domain exists and it's state is "joined" */
 174 static int dlm_wait_on_domain_helper(const char *domain)
 175 {
 176         int ret = 0;
 177         struct dlm_ctxt *tmp = NULL;
 178
 179         spin_lock(&dlm_domain_lock);
 180
 181         tmp = __dlm_lookup_domain(domain);
 182         if (!tmp)
 183                 ret = 1;
 184         else if (tmp->dlm_state == DLM_CTXT_JOINED)
 185                 ret = 1;
 186
 187         spin_unlock(&dlm_domain_lock);
 188         return ret;
 189 }
 190
 191 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
 192 {
 193         if (dlm->lockres_hash)
 194                 free_page((unsigned long) dlm->lockres_hash);
 195
 196         if (dlm->name)
 197                 kfree(dlm->name);
 198
 199         kfree(dlm);
 200 }
 201
 202 /* A little strange - this function will be called while holding
 203  * dlm_domain_lock and is expected to be holding it on the way out. We
 204  * will however drop and reacquire it multiple times */
 205 static void dlm_ctxt_release(struct kref *kref)
 206 {
 207         struct dlm_ctxt *dlm;
 208
 209         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
 210
 211         BUG_ON(dlm->num_joins);
 212         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
 213
 214         /* we may still be in the list if we hit an error during join. */
 215         list_del_init(&dlm->list);
 216
 217         spin_unlock(&dlm_domain_lock);
 218
 219         mlog(0, "freeing memory from domain %s\n", dlm->name);
 220
 221         wake_up(&dlm_domain_events);
 222
 223         dlm_free_ctxt_mem(dlm);
 224
 225         spin_lock(&dlm_domain_lock);
 226 }
 227
 228 void dlm_put(struct dlm_ctxt *dlm)
 229 {
 230         spin_lock(&dlm_domain_lock);
 231         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
 232         spin_unlock(&dlm_domain_lock);
 233 }
 234
 235 static void __dlm_get(struct dlm_ctxt *dlm)
 236 {
 237         kref_get(&dlm->dlm_refs);
 238 }
 239
 240 /* given a questionable reference to a dlm object, gets a reference if
 241  * it can find it in the list, otherwise returns NULL in which case
 242  * you shouldn't trust your pointer. */
 243 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
 244 {
 245         struct list_head *iter;
 246         struct dlm_ctxt *target = NULL;
 247
 248         spin_lock(&dlm_domain_lock);
 249
 250         list_for_each(iter, &dlm_domains) {
 251                 target = list_entry (iter, struct dlm_ctxt, list);
 252
 253                 if (target == dlm) {
 254                         __dlm_get(target);
 255                         break;
 256                 }
 257
 258                 target = NULL;
 259         }
 260
 261         spin_unlock(&dlm_domain_lock);
 262
 263         return target;
 264 }
 265
 266 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
 267 {
 268         int ret;
 269
 270         spin_lock(&dlm_domain_lock);
 271         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
 272                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
 273         spin_unlock(&dlm_domain_lock);
 274
 275         return ret;
 276 }
 277
 278 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
 279 {
 280         dlm_unregister_domain_handlers(dlm);
 281         dlm_complete_thread(dlm);
 282         dlm_complete_recovery_thread(dlm);
 283
 284         /* We've left the domain. Now we can take ourselves out of the
 285          * list and allow the kref stuff to help us free the
 286          * memory. */
 287         spin_lock(&dlm_domain_lock);
 288         list_del_init(&dlm->list);
 289         spin_unlock(&dlm_domain_lock);
 290
 291         /* Wake up anyone waiting for us to remove this domain */
 292         wake_up(&dlm_domain_events);
 293 }
 294
 295 static void dlm_migrate_all_locks(struct dlm_ctxt *dlm)
 296 {
 297         int i;
 298         struct dlm_lock_resource *res;
 299
 300         mlog(0, "Migrating locks from domain %s\n", dlm->name);
 301 restart:
 302         spin_lock(&dlm->spinlock);
 303         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
 304                 while (!hlist_empty(&dlm->lockres_hash[i])) {
 305                         res = hlist_entry(dlm->lockres_hash[i].first,
 306                                           struct dlm_lock_resource, hash_node);
 307                         /* need reference when manually grabbing lockres */
 308                         dlm_lockres_get(res);
 309                         /* this should unhash the lockres
 310                          * and exit with dlm->spinlock */
 311                         mlog(0, "purging res=%p\n", res);
 312                         if (dlm_lockres_is_dirty(dlm, res)) {
 313                                 /* HACK!  this should absolutely go.
 314                                  * need to figure out why some empty
 315                                  * lockreses are still marked dirty */
 316                                 mlog(ML_ERROR, "lockres %.*s dirty!\n",
 317                                      res->lockname.len, res->lockname.name);
 318
 319                                 spin_unlock(&dlm->spinlock);
 320                                 dlm_kick_thread(dlm, res);
 321                                 wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
 322                                 dlm_lockres_put(res);
 323                                 goto restart;
 324                         }
 325                         dlm_purge_lockres(dlm, res);
 326                         dlm_lockres_put(res);
 327                 }
 328         }
 329         spin_unlock(&dlm->spinlock);
 330
 331         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
 332 }
 333
 334 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
 335 {
 336         int ret;
 337
 338         spin_lock(&dlm->spinlock);
 339         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
 340         spin_unlock(&dlm->spinlock);
 341
 342         return ret;
 343 }
 344
 345 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
 346 {
 347         /* Yikes, a double spinlock! I need domain_lock for the dlm
 348          * state and the dlm spinlock for join state... Sorry! */
 349 again:
 350         spin_lock(&dlm_domain_lock);
 351         spin_lock(&dlm->spinlock);
 352
 353         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 354                 mlog(0, "Node %d is joining, we wait on it.\n",
 355                           dlm->joining_node);
 356                 spin_unlock(&dlm->spinlock);
 357                 spin_unlock(&dlm_domain_lock);
 358
 359                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
 360                 goto again;
 361         }
 362
 363         dlm->dlm_state = DLM_CTXT_LEAVING;
 364         spin_unlock(&dlm->spinlock);
 365         spin_unlock(&dlm_domain_lock);
 366 }
 367
 368 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
 369 {
 370         int node = -1;
 371
 372         assert_spin_locked(&dlm->spinlock);
 373
 374         mlog(ML_NOTICE, "Nodes in my domain (\"%s\"):\n", dlm->name);
 375
 376         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 377                                      node + 1)) < O2NM_MAX_NODES) {
 378                 mlog(ML_NOTICE, " node %d\n", node);
 379         }
 380 }
 381
 382 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data)
 383 {
 384         struct dlm_ctxt *dlm = data;
 385         unsigned int node;
 386         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
 387
 388         mlog_entry("%p %u %p", msg, len, data);
 389
 390         if (!dlm_grab(dlm))
 391                 return 0;
 392
 393         node = exit_msg->node_idx;
 394
 395         mlog(0, "Node %u leaves domain %s\n", node, dlm->name);
 396
 397         spin_lock(&dlm->spinlock);
 398         clear_bit(node, dlm->domain_map);
 399         __dlm_print_nodes(dlm);
 400
 401         /* notify anything attached to the heartbeat events */
 402         dlm_hb_event_notify_attached(dlm, node, 0);
 403
 404         spin_unlock(&dlm->spinlock);
 405
 406         dlm_put(dlm);
 407
 408         return 0;
 409 }
 410
 411 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
 412                                     unsigned int node)
 413 {
 414         int status;
 415         struct dlm_exit_domain leave_msg;
 416
 417         mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
 418                   node, dlm->name, dlm->node_num);
 419
 420         memset(&leave_msg, 0, sizeof(leave_msg));
 421         leave_msg.node_idx = dlm->node_num;
 422
 423         status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
 424                                     &leave_msg, sizeof(leave_msg), node,
 425                                     NULL);
 426
 427         mlog(0, "status return %d from o2net_send_message\n", status);
 428
 429         return status;
 430 }
 431
 432
 433 static void dlm_leave_domain(struct dlm_ctxt *dlm)
 434 {
 435         int node, clear_node, status;
 436
 437         /* At this point we've migrated away all our locks and won't
 438          * accept mastership of new ones. The dlm is responsible for
 439          * almost nothing now. We make sure not to confuse any joining
 440          * nodes and then commence shutdown procedure. */
 441
 442         spin_lock(&dlm->spinlock);
 443         /* Clear ourselves from the domain map */
 444         clear_bit(dlm->node_num, dlm->domain_map);
 445         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
 446                                      0)) < O2NM_MAX_NODES) {
 447                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
 448                  * -nodes cannot be added now as the
 449                  *   query_join_handlers knows to respond with OK_NO_MAP
 450                  * -we catch the right network errors if a node is
 451                  *   removed from the map while we're sending him the
 452                  *   exit message. */
 453                 spin_unlock(&dlm->spinlock);
 454
 455                 clear_node = 1;
 456
 457                 status = dlm_send_one_domain_exit(dlm, node);
 458                 if (status < 0 &&
 459                     status != -ENOPROTOOPT &&
 460                     status != -ENOTCONN) {
 461                         mlog(ML_NOTICE, "Error %d sending domain exit message "
 462                              "to node %d\n", status, node);
 463
 464                         /* Not sure what to do here but lets sleep for
 465                          * a bit in case this was a transient
 466                          * error... */
 467                         msleep(DLM_DOMAIN_BACKOFF_MS);
 468                         clear_node = 0;
 469                 }
 470
 471                 spin_lock(&dlm->spinlock);
 472                 /* If we're not clearing the node bit then we intend
 473                  * to loop back around to try again. */
 474                 if (clear_node)
 475                         clear_bit(node, dlm->domain_map);
 476         }
 477         spin_unlock(&dlm->spinlock);
 478 }
 479
 480 int dlm_joined(struct dlm_ctxt *dlm)
 481 {
 482         int ret = 0;
 483
 484         spin_lock(&dlm_domain_lock);
 485
 486         if (dlm->dlm_state == DLM_CTXT_JOINED)
 487                 ret = 1;
 488
 489         spin_unlock(&dlm_domain_lock);
 490
 491         return ret;
 492 }
 493
 494 int dlm_shutting_down(struct dlm_ctxt *dlm)
 495 {
 496         int ret = 0;
 497
 498         spin_lock(&dlm_domain_lock);
 499
 500         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
 501                 ret = 1;
 502
 503         spin_unlock(&dlm_domain_lock);
 504
 505         return ret;
 506 }
 507
 508 void dlm_unregister_domain(struct dlm_ctxt *dlm)
 509 {
 510         int leave = 0;
 511
 512         spin_lock(&dlm_domain_lock);
 513         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
 514         BUG_ON(!dlm->num_joins);
 515
 516         dlm->num_joins--;
 517         if (!dlm->num_joins) {
 518                 /* We mark it "in shutdown" now so new register
 519                  * requests wait until we've completely left the
 520                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
 521                  * want new domain joins to communicate with us at
 522                  * least until we've completed migration of our
 523                  * resources. */
 524                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
 525                 leave = 1;
 526         }
 527         spin_unlock(&dlm_domain_lock);
 528
 529         if (leave) {
 530                 mlog(0, "shutting down domain %s\n", dlm->name);
 531
 532                 /* We changed dlm state, notify the thread */
 533                 dlm_kick_thread(dlm, NULL);
 534
 535                 dlm_migrate_all_locks(dlm);
 536                 dlm_mark_domain_leaving(dlm);
 537                 dlm_leave_domain(dlm);
 538                 dlm_complete_dlm_shutdown(dlm);
 539         }
 540         dlm_put(dlm);
 541 }
 542 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
 543
 544 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data)
 545 {
 546         struct dlm_query_join_request *query;
 547         enum dlm_query_join_response response;
 548         struct dlm_ctxt *dlm = NULL;
 549
 550         query = (struct dlm_query_join_request *) msg->buf;
 551
 552         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
 553                   query->domain);
 554
 555         /*
 556          * If heartbeat doesn't consider the node live, tell it
 557          * to back off and try again.  This gives heartbeat a chance
 558          * to catch up.
 559          */
 560         if (!o2hb_check_node_heartbeating(query->node_idx)) {
 561                 mlog(0, "node %u is not in our live map yet\n",
 562                      query->node_idx);
 563
 564                 response = JOIN_DISALLOW;
 565                 goto respond;
 566         }
 567
 568         response = JOIN_OK_NO_MAP;
 569
 570         spin_lock(&dlm_domain_lock);
 571         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
 572         /* Once the dlm ctxt is marked as leaving then we don't want
 573          * to be put in someone's domain map.
 574          * Also, explicitly disallow joining at certain troublesome
 575          * times (ie. during recovery). */
 576         if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
 577                 int bit = query->node_idx;
 578                 spin_lock(&dlm->spinlock);
 579
 580                 if (dlm->dlm_state == DLM_CTXT_NEW &&
 581                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
 582                         /*If this is a brand new context and we
 583                          * haven't started our join process yet, then
 584                          * the other node won the race. */
 585                         response = JOIN_OK_NO_MAP;
 586                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
 587                         /* Disallow parallel joins. */
 588                         response = JOIN_DISALLOW;
 589                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
 590                         mlog(ML_NOTICE, "node %u trying to join, but recovery "
 591                              "is ongoing.\n", bit);
 592                         response = JOIN_DISALLOW;
 593                 } else if (test_bit(bit, dlm->recovery_map)) {
 594                         mlog(ML_NOTICE, "node %u trying to join, but it "
 595                              "still needs recovery.\n", bit);
 596                         response = JOIN_DISALLOW;
 597                 } else if (test_bit(bit, dlm->domain_map)) {
 598                         mlog(ML_NOTICE, "node %u trying to join, but it "
 599                              "is still in the domain! needs recovery?\n",
 600                              bit);
 601                         response = JOIN_DISALLOW;
 602                 } else {
 603                         /* Alright we're fully a part of this domain
 604                          * so we keep some state as to who's joining
 605                          * and indicate to him that needs to be fixed
 606                          * up. */
 607                         response = JOIN_OK;
 608                         __dlm_set_joining_node(dlm, query->node_idx);
 609                 }
 610
 611                 spin_unlock(&dlm->spinlock);
 612         }
 613         spin_unlock(&dlm_domain_lock);
 614
 615 respond:
 616         mlog(0, "We respond with %u\n", response);
 617
 618         return response;
 619 }
 620
 621 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data)
 622 {
 623         struct dlm_assert_joined *assert;
 624         struct dlm_ctxt *dlm = NULL;
 625
 626         assert = (struct dlm_assert_joined *) msg->buf;
 627
 628         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
 629                   assert->domain);
 630
 631         spin_lock(&dlm_domain_lock);
 632         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
 633         /* XXX should we consider no dlm ctxt an error? */
 634         if (dlm) {
 635                 spin_lock(&dlm->spinlock);
 636
 637                 /* Alright, this node has officially joined our
 638                  * domain. Set him in the map and clean up our
 639                  * leftover join state. */
 640                 BUG_ON(dlm->joining_node != assert->node_idx);
 641                 set_bit(assert->node_idx, dlm->domain_map);
 642                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 643
 644                 __dlm_print_nodes(dlm);
 645
 646                 /* notify anything attached to the heartbeat events */
 647                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
 648
 649                 spin_unlock(&dlm->spinlock);
 650         }
 651         spin_unlock(&dlm_domain_lock);
 652
 653         return 0;
 654 }
 655
 656 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data)
 657 {
 658         struct dlm_cancel_join *cancel;
 659         struct dlm_ctxt *dlm = NULL;
 660
 661         cancel = (struct dlm_cancel_join *) msg->buf;
 662
 663         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
 664                   cancel->domain);
 665
 666         spin_lock(&dlm_domain_lock);
 667         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
 668
 669         if (dlm) {
 670                 spin_lock(&dlm->spinlock);
 671
 672                 /* Yikes, this guy wants to cancel his join. No
 673                  * problem, we simply cleanup our join state. */
 674                 BUG_ON(dlm->joining_node != cancel->node_idx);
 675                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 676
 677                 spin_unlock(&dlm->spinlock);
 678         }
 679         spin_unlock(&dlm_domain_lock);
 680
 681         return 0;
 682 }
 683
 684 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
 685                                     unsigned int node)
 686 {
 687         int status;
 688         struct dlm_cancel_join cancel_msg;
 689
 690         memset(&cancel_msg, 0, sizeof(cancel_msg));
 691         cancel_msg.node_idx = dlm->node_num;
 692         cancel_msg.name_len = strlen(dlm->name);
 693         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
 694
 695         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
 696                                     &cancel_msg, sizeof(cancel_msg), node,
 697                                     NULL);
 698         if (status < 0) {
 699                 mlog_errno(status);
 700                 goto bail;
 701         }
 702
 703 bail:
 704         return status;
 705 }
 706
 707 /* map_size should be in bytes. */
 708 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
 709                                  unsigned long *node_map,
 710                                  unsigned int map_size)
 711 {
 712         int status, tmpstat;
 713         unsigned int node;
 714
 715         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
 716                          sizeof(unsigned long))) {
 717                 mlog(ML_ERROR,
 718                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
 719                      map_size, BITS_TO_LONGS(O2NM_MAX_NODES));
 720                 return -EINVAL;
 721         }
 722
 723         status = 0;
 724         node = -1;
 725         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
 726                                      node + 1)) < O2NM_MAX_NODES) {
 727                 if (node == dlm->node_num)
 728                         continue;
 729
 730                 tmpstat = dlm_send_one_join_cancel(dlm, node);
 731                 if (tmpstat) {
 732                         mlog(ML_ERROR, "Error return %d cancelling join on "
 733                              "node %d\n", tmpstat, node);
 734                         if (!status)
 735                                 status = tmpstat;
 736                 }
 737         }
 738
 739         if (status)
 740                 mlog_errno(status);
 741         return status;
 742 }
 743
 744 static int dlm_request_join(struct dlm_ctxt *dlm,
 745                             int node,
 746                             enum dlm_query_join_response *response)
 747 {
 748         int status, retval;
 749         struct dlm_query_join_request join_msg;
 750
 751         mlog(0, "querying node %d\n", node);
 752
 753         memset(&join_msg, 0, sizeof(join_msg));
 754         join_msg.node_idx = dlm->node_num;
 755         join_msg.name_len = strlen(dlm->name);
 756         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
 757
 758         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
 759                                     sizeof(join_msg), node, &retval);
 760         if (status < 0 && status != -ENOPROTOOPT) {
 761                 mlog_errno(status);
 762                 goto bail;
 763         }
 764
 765         /* -ENOPROTOOPT from the net code means the other side isn't
 766             listening for our message type -- that's fine, it means
 767             his dlm isn't up, so we can consider him a 'yes' but not
 768             joined into the domain.  */
 769         if (status == -ENOPROTOOPT) {
 770                 status = 0;
 771                 *response = JOIN_OK_NO_MAP;
 772         } else if (retval == JOIN_DISALLOW ||
 773                    retval == JOIN_OK ||
 774                    retval == JOIN_OK_NO_MAP) {
 775                 *response = retval;
 776         } else {
 777                 status = -EINVAL;
 778                 mlog(ML_ERROR, "invalid response %d from node %u\n", retval,
 779                      node);
 780         }
 781
 782         mlog(0, "status %d, node %d response is %d\n", status, node,
 783                   *response);
 784
 785 bail:
 786         return status;
 787 }
 788
 789 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
 790                                     unsigned int node)
 791 {
 792         int status;
 793         struct dlm_assert_joined assert_msg;
 794
 795         mlog(0, "Sending join assert to node %u\n", node);
 796
 797         memset(&assert_msg, 0, sizeof(assert_msg));
 798         assert_msg.node_idx = dlm->node_num;
 799         assert_msg.name_len = strlen(dlm->name);
 800         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
 801
 802         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
 803                                     &assert_msg, sizeof(assert_msg), node,
 804                                     NULL);
 805         if (status < 0)
 806                 mlog_errno(status);
 807
 808         return status;
 809 }
 810
 811 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
 812                                   unsigned long *node_map)
 813 {
 814         int status, node, live;
 815
 816         status = 0;
 817         node = -1;
 818         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
 819                                      node + 1)) < O2NM_MAX_NODES) {
 820                 if (node == dlm->node_num)
 821                         continue;
 822
 823                 do {
 824                         /* It is very important that this message be
 825                          * received so we spin until either the node
 826                          * has died or it gets the message. */
 827                         status = dlm_send_one_join_assert(dlm, node);
 828
 829                         spin_lock(&dlm->spinlock);
 830                         live = test_bit(node, dlm->live_nodes_map);
 831                         spin_unlock(&dlm->spinlock);
 832
 833                         if (status) {
 834                                 mlog(ML_ERROR, "Error return %d asserting "
 835                                      "join on node %d\n", status, node);
 836
 837                                 /* give us some time between errors... */
 838                                 if (live)
 839                                         msleep(DLM_DOMAIN_BACKOFF_MS);
 840                         }
 841                 } while (status && live);
 842         }
 843 }
 844
 845 struct domain_join_ctxt {
 846         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 847         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
 848 };
 849
 850 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
 851                                    struct domain_join_ctxt *ctxt,
 852                                    enum dlm_query_join_response response)
 853 {
 854         int ret;
 855
 856         if (response == JOIN_DISALLOW) {
 857                 mlog(0, "Latest response of disallow -- should restart\n");
 858                 return 1;
 859         }
 860
 861         spin_lock(&dlm->spinlock);
 862         /* For now, we restart the process if the node maps have
 863          * changed at all */
 864         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
 865                      sizeof(dlm->live_nodes_map));
 866         spin_unlock(&dlm->spinlock);
 867
 868         if (ret)
 869                 mlog(0, "Node maps changed -- should restart\n");
 870
 871         return ret;
 872 }
 873
 874 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
 875 {
 876         int status = 0, tmpstat, node;
 877         struct domain_join_ctxt *ctxt;
 878         enum dlm_query_join_response response;
 879
 880         mlog_entry("%p", dlm);
 881
 882         ctxt = kcalloc(1, sizeof(*ctxt), GFP_KERNEL);
 883         if (!ctxt) {
 884                 status = -ENOMEM;
 885                 mlog_errno(status);
 886                 goto bail;
 887         }
 888
 889         /* group sem locking should work for us here -- we're already
 890          * registered for heartbeat events so filling this should be
 891          * atomic wrt getting those handlers called. */
 892         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
 893
 894         spin_lock(&dlm->spinlock);
 895         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
 896
 897         __dlm_set_joining_node(dlm, dlm->node_num);
 898
 899         spin_unlock(&dlm->spinlock);
 900
 901         node = -1;
 902         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
 903                                      node + 1)) < O2NM_MAX_NODES) {
 904                 if (node == dlm->node_num)
 905                         continue;
 906
 907                 status = dlm_request_join(dlm, node, &response);
 908                 if (status < 0) {
 909                         mlog_errno(status);
 910                         goto bail;
 911                 }
 912
 913                 /* Ok, either we got a response or the node doesn't have a
 914                  * dlm up. */
 915                 if (response == JOIN_OK)
 916                         set_bit(node, ctxt->yes_resp_map);
 917
 918                 if (dlm_should_restart_join(dlm, ctxt, response)) {
 919                         status = -EAGAIN;
 920                         goto bail;
 921                 }
 922         }
 923
 924         mlog(0, "Yay, done querying nodes!\n");
 925
 926         /* Yay, everyone agree's we can join the domain. My domain is
 927          * comprised of all nodes who were put in the
 928          * yes_resp_map. Copy that into our domain map and send a join
 929          * assert message to clean up everyone elses state. */
 930         spin_lock(&dlm->spinlock);
 931         memcpy(dlm->domain_map, ctxt->yes_resp_map,
 932                sizeof(ctxt->yes_resp_map));
 933         set_bit(dlm->node_num, dlm->domain_map);
 934         spin_unlock(&dlm->spinlock);
 935
 936         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
 937
 938         /* Joined state *must* be set before the joining node
 939          * information, otherwise the query_join handler may read no
 940          * current joiner but a state of NEW and tell joining nodes
 941          * we're not in the domain. */
 942         spin_lock(&dlm_domain_lock);
 943         dlm->dlm_state = DLM_CTXT_JOINED;
 944         dlm->num_joins++;
 945         spin_unlock(&dlm_domain_lock);
 946
 947 bail:
 948         spin_lock(&dlm->spinlock);
 949         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
 950         if (!status)
 951                 __dlm_print_nodes(dlm);
 952         spin_unlock(&dlm->spinlock);
 953
 954         if (ctxt) {
 955                 /* Do we need to send a cancel message to any nodes? */
 956                 if (status < 0) {
 957                         tmpstat = dlm_send_join_cancels(dlm,
 958                                                         ctxt->yes_resp_map,
 959                                                         sizeof(ctxt->yes_resp_map));
 960                         if (tmpstat < 0)
 961                                 mlog_errno(tmpstat);
 962                 }
 963                 kfree(ctxt);
 964         }
 965
 966         mlog(0, "returning %d\n", status);
 967         return status;
 968 }
 969
 970 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
 971 {
 972         o2hb_unregister_callback(&dlm->dlm_hb_up);
 973         o2hb_unregister_callback(&dlm->dlm_hb_down);
 974         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
 975 }
 976
 977 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
 978 {
 979         int status;
 980
 981         mlog(0, "registering handlers.\n");
 982
 983         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
 984                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
 985         status = o2hb_register_callback(&dlm->dlm_hb_down);
 986         if (status)
 987                 goto bail;
 988
 989         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
 990                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
 991         status = o2hb_register_callback(&dlm->dlm_hb_up);
 992         if (status)
 993                 goto bail;
 994
 995         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
 996                                         sizeof(struct dlm_master_request),
 997                                         dlm_master_request_handler,
 998                                         dlm, &dlm->dlm_domain_handlers);
 999         if (status)
1000                 goto bail;
1001
1002         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1003                                         sizeof(struct dlm_assert_master),
1004                                         dlm_assert_master_handler,
1005                                         dlm, &dlm->dlm_domain_handlers);
1006         if (status)
1007                 goto bail;
1008
1009         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1010                                         sizeof(struct dlm_create_lock),
1011                                         dlm_create_lock_handler,
1012                                         dlm, &dlm->dlm_domain_handlers);
1013         if (status)
1014                 goto bail;
1015
1016         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1017                                         DLM_CONVERT_LOCK_MAX_LEN,
1018                                         dlm_convert_lock_handler,
1019                                         dlm, &dlm->dlm_domain_handlers);
1020         if (status)
1021                 goto bail;
1022
1023         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1024                                         DLM_UNLOCK_LOCK_MAX_LEN,
1025                                         dlm_unlock_lock_handler,
1026                                         dlm, &dlm->dlm_domain_handlers);
1027         if (status)
1028                 goto bail;
1029
1030         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1031                                         DLM_PROXY_AST_MAX_LEN,
1032                                         dlm_proxy_ast_handler,
1033                                         dlm, &dlm->dlm_domain_handlers);
1034         if (status)
1035                 goto bail;
1036
1037         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1038                                         sizeof(struct dlm_exit_domain),
1039                                         dlm_exit_domain_handler,
1040                                         dlm, &dlm->dlm_domain_handlers);
1041         if (status)
1042                 goto bail;
1043
1044         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1045                                         sizeof(struct dlm_migrate_request),
1046                                         dlm_migrate_request_handler,
1047                                         dlm, &dlm->dlm_domain_handlers);
1048         if (status)
1049                 goto bail;
1050
1051         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1052                                         DLM_MIG_LOCKRES_MAX_LEN,
1053                                         dlm_mig_lockres_handler,
1054                                         dlm, &dlm->dlm_domain_handlers);
1055         if (status)
1056                 goto bail;
1057
1058         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1059                                         sizeof(struct dlm_master_requery),
1060                                         dlm_master_requery_handler,
1061                                         dlm, &dlm->dlm_domain_handlers);
1062         if (status)
1063                 goto bail;
1064
1065         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1066                                         sizeof(struct dlm_lock_request),
1067                                         dlm_request_all_locks_handler,
1068                                         dlm, &dlm->dlm_domain_handlers);
1069         if (status)
1070                 goto bail;
1071
1072         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1073                                         sizeof(struct dlm_reco_data_done),
1074                                         dlm_reco_data_done_handler,
1075                                         dlm, &dlm->dlm_domain_handlers);
1076         if (status)
1077                 goto bail;
1078
1079         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1080                                         sizeof(struct dlm_begin_reco),
1081                                         dlm_begin_reco_handler,
1082                                         dlm, &dlm->dlm_domain_handlers);
1083         if (status)
1084                 goto bail;
1085
1086         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1087                                         sizeof(struct dlm_finalize_reco),
1088                                         dlm_finalize_reco_handler,
1089                                         dlm, &dlm->dlm_domain_handlers);
1090         if (status)
1091                 goto bail;
1092
1093 bail:
1094         if (status)
1095                 dlm_unregister_domain_handlers(dlm);
1096
1097         return status;
1098 }
1099
1100 static int dlm_join_domain(struct dlm_ctxt *dlm)
1101 {
1102         int status;
1103
1104         BUG_ON(!dlm);
1105
1106         mlog(0, "Join domain %s\n", dlm->name);
1107
1108         status = dlm_register_domain_handlers(dlm);
1109         if (status) {
1110                 mlog_errno(status);
1111                 goto bail;
1112         }
1113
1114         status = dlm_launch_thread(dlm);
1115         if (status < 0) {
1116                 mlog_errno(status);
1117                 goto bail;
1118         }
1119
1120         status = dlm_launch_recovery_thread(dlm);
1121         if (status < 0) {
1122                 mlog_errno(status);
1123                 goto bail;
1124         }
1125
1126         do {
1127                 unsigned int backoff;
1128                 status = dlm_try_to_join_domain(dlm);
1129
1130                 /* If we're racing another node to the join, then we
1131                  * need to back off temporarily and let them
1132                  * complete. */
1133                 if (status == -EAGAIN) {
1134                         if (signal_pending(current)) {
1135                                 status = -ERESTARTSYS;
1136                                 goto bail;
1137                         }
1138
1139                         /*
1140                          * <chip> After you!
1141                          * <dale> No, after you!
1142                          * <chip> I insist!
1143                          * <dale> But you first!
1144                          * ...
1145                          */
1146                         backoff = (unsigned int)(jiffies & 0x3);
1147                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1148                         mlog(0, "backoff %d\n", backoff);
1149                         msleep(backoff);
1150                 }
1151         } while (status == -EAGAIN);
1152
1153         if (status < 0) {
1154                 mlog_errno(status);
1155                 goto bail;
1156         }
1157
1158         status = 0;
1159 bail:
1160         wake_up(&dlm_domain_events);
1161
1162         if (status) {
1163                 dlm_unregister_domain_handlers(dlm);
1164                 dlm_complete_thread(dlm);
1165                 dlm_complete_recovery_thread(dlm);
1166         }
1167
1168         return status;
1169 }
1170
1171 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1172                                 u32 key)
1173 {
1174         int i;
1175         struct dlm_ctxt *dlm = NULL;
1176
1177         dlm = kcalloc(1, sizeof(*dlm), GFP_KERNEL);
1178         if (!dlm) {
1179                 mlog_errno(-ENOMEM);
1180                 goto leave;
1181         }
1182
1183         dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL);
1184         if (dlm->name == NULL) {
1185                 mlog_errno(-ENOMEM);
1186                 kfree(dlm);
1187                 dlm = NULL;
1188                 goto leave;
1189         }
1190
1191         dlm->lockres_hash = (struct hlist_head *) __get_free_page(GFP_KERNEL);
1192         if (!dlm->lockres_hash) {
1193                 mlog_errno(-ENOMEM);
1194                 kfree(dlm->name);
1195                 kfree(dlm);
1196                 dlm = NULL;
1197                 goto leave;
1198         }
1199
1200         for (i=0; i<DLM_HASH_BUCKETS; i++)
1201                 INIT_HLIST_HEAD(&dlm->lockres_hash[i]);
1202
1203         strcpy(dlm->name, domain);
1204         dlm->key = key;
1205         dlm->node_num = o2nm_this_node();
1206
1207         spin_lock_init(&dlm->spinlock);
1208         spin_lock_init(&dlm->master_lock);
1209         spin_lock_init(&dlm->ast_lock);
1210         INIT_LIST_HEAD(&dlm->list);
1211         INIT_LIST_HEAD(&dlm->dirty_list);
1212         INIT_LIST_HEAD(&dlm->reco.resources);
1213         INIT_LIST_HEAD(&dlm->reco.received);
1214         INIT_LIST_HEAD(&dlm->reco.node_data);
1215         INIT_LIST_HEAD(&dlm->purge_list);
1216         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1217         dlm->reco.state = 0;
1218
1219         INIT_LIST_HEAD(&dlm->pending_asts);
1220         INIT_LIST_HEAD(&dlm->pending_basts);
1221
1222         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1223                   dlm->recovery_map, &(dlm->recovery_map[0]));
1224
1225         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1226         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1227         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1228
1229         dlm->dlm_thread_task = NULL;
1230         dlm->dlm_reco_thread_task = NULL;
1231         init_waitqueue_head(&dlm->dlm_thread_wq);
1232         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1233         init_waitqueue_head(&dlm->reco.event);
1234         init_waitqueue_head(&dlm->ast_wq);
1235         init_waitqueue_head(&dlm->migration_wq);
1236         INIT_LIST_HEAD(&dlm->master_list);
1237         INIT_LIST_HEAD(&dlm->mle_hb_events);
1238
1239         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1240         init_waitqueue_head(&dlm->dlm_join_events);
1241
1242         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1243         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1244         atomic_set(&dlm->local_resources, 0);
1245         atomic_set(&dlm->remote_resources, 0);
1246         atomic_set(&dlm->unknown_resources, 0);
1247
1248         spin_lock_init(&dlm->work_lock);
1249         INIT_LIST_HEAD(&dlm->work_list);
1250         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work, dlm);
1251
1252         kref_init(&dlm->dlm_refs);
1253         dlm->dlm_state = DLM_CTXT_NEW;
1254
1255         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
1256
1257         mlog(0, "context init: refcount %u\n",
1258                   atomic_read(&dlm->dlm_refs.refcount));
1259
1260 leave:
1261         return dlm;
1262 }
1263
1264 /*
1265  * dlm_register_domain: one-time setup per "domain"
1266  */
1267 struct dlm_ctxt * dlm_register_domain(const char *domain,
1268                                u32 key)
1269 {
1270         int ret;
1271         struct dlm_ctxt *dlm = NULL;
1272         struct dlm_ctxt *new_ctxt = NULL;
1273
1274         if (strlen(domain) > O2NM_MAX_NAME_LEN) {
1275                 ret = -ENAMETOOLONG;
1276                 mlog(ML_ERROR, "domain name length too long\n");
1277                 goto leave;
1278         }
1279
1280         if (!o2hb_check_local_node_heartbeating()) {
1281                 mlog(ML_ERROR, "the local node has not been configured, or is "
1282                      "not heartbeating\n");
1283                 ret = -EPROTO;
1284                 goto leave;
1285         }
1286
1287         mlog(0, "register called for domain \"%s\"\n", domain);
1288
1289 retry:
1290         dlm = NULL;
1291         if (signal_pending(current)) {
1292                 ret = -ERESTARTSYS;
1293                 mlog_errno(ret);
1294                 goto leave;
1295         }
1296
1297         spin_lock(&dlm_domain_lock);
1298
1299         dlm = __dlm_lookup_domain(domain);
1300         if (dlm) {
1301                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
1302                         spin_unlock(&dlm_domain_lock);
1303
1304                         mlog(0, "This ctxt is not joined yet!\n");
1305                         wait_event_interruptible(dlm_domain_events,
1306                                                  dlm_wait_on_domain_helper(
1307                                                          domain));
1308                         goto retry;
1309                 }
1310
1311                 __dlm_get(dlm);
1312                 dlm->num_joins++;
1313
1314                 spin_unlock(&dlm_domain_lock);
1315
1316                 ret = 0;
1317                 goto leave;
1318         }
1319
1320         /* doesn't exist */
1321         if (!new_ctxt) {
1322                 spin_unlock(&dlm_domain_lock);
1323
1324                 new_ctxt = dlm_alloc_ctxt(domain, key);
1325                 if (new_ctxt)
1326                         goto retry;
1327
1328                 ret = -ENOMEM;
1329                 mlog_errno(ret);
1330                 goto leave;
1331         }
1332
1333         /* a little variable switch-a-roo here... */
1334         dlm = new_ctxt;
1335         new_ctxt = NULL;
1336
1337         /* add the new domain */
1338         list_add_tail(&dlm->list, &dlm_domains);
1339         spin_unlock(&dlm_domain_lock);
1340
1341         ret = dlm_join_domain(dlm);
1342         if (ret) {
1343                 mlog_errno(ret);
1344                 dlm_put(dlm);
1345                 goto leave;
1346         }
1347
1348         ret = 0;
1349 leave:
1350         if (new_ctxt)
1351                 dlm_free_ctxt_mem(new_ctxt);
1352
1353         if (ret < 0)
1354                 dlm = ERR_PTR(ret);
1355
1356         return dlm;
1357 }
1358 EXPORT_SYMBOL_GPL(dlm_register_domain);
1359
1360 static LIST_HEAD(dlm_join_handlers);
1361
1362 static void dlm_unregister_net_handlers(void)
1363 {
1364         o2net_unregister_handler_list(&dlm_join_handlers);
1365 }
1366
1367 static int dlm_register_net_handlers(void)
1368 {
1369         int status = 0;
1370
1371         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1372                                         sizeof(struct dlm_query_join_request),
1373                                         dlm_query_join_handler,
1374                                         NULL, &dlm_join_handlers);
1375         if (status)
1376                 goto bail;
1377
1378         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1379                                         sizeof(struct dlm_assert_joined),
1380                                         dlm_assert_joined_handler,
1381                                         NULL, &dlm_join_handlers);
1382         if (status)
1383                 goto bail;
1384
1385         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1386                                         sizeof(struct dlm_cancel_join),
1387                                         dlm_cancel_join_handler,
1388                                         NULL, &dlm_join_handlers);
1389
1390 bail:
1391         if (status < 0)
1392                 dlm_unregister_net_handlers();
1393
1394         return status;
1395 }
1396
1397 /* Domain eviction callback handling.
1398  *
1399  * The file system requires notification of node death *before* the
1400  * dlm completes it's recovery work, otherwise it may be able to
1401  * acquire locks on resources requiring recovery. Since the dlm can
1402  * evict a node from it's domain *before* heartbeat fires, a similar
1403  * mechanism is required. */
1404
1405 /* Eviction is not expected to happen often, so a per-domain lock is
1406  * not necessary. Eviction callbacks are allowed to sleep for short
1407  * periods of time. */
1408 static DECLARE_RWSEM(dlm_callback_sem);
1409
1410 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
1411                                         int node_num)
1412 {
1413         struct list_head *iter;
1414         struct dlm_eviction_cb *cb;
1415
1416         down_read(&dlm_callback_sem);
1417         list_for_each(iter, &dlm->dlm_eviction_callbacks) {
1418                 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
1419
1420                 cb->ec_func(node_num, cb->ec_data);
1421         }
1422         up_read(&dlm_callback_sem);
1423 }
1424
1425 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
1426                            dlm_eviction_func *f,
1427                            void *data)
1428 {
1429         INIT_LIST_HEAD(&cb->ec_item);
1430         cb->ec_func = f;
1431         cb->ec_data = data;
1432 }
1433 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
1434
1435 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
1436                               struct dlm_eviction_cb *cb)
1437 {
1438         down_write(&dlm_callback_sem);
1439         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
1440         up_write(&dlm_callback_sem);
1441 }
1442 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
1443
1444 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
1445 {
1446         down_write(&dlm_callback_sem);
1447         list_del_init(&cb->ec_item);
1448         up_write(&dlm_callback_sem);
1449 }
1450 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
1451
1452 static int __init dlm_init(void)
1453 {
1454         int status;
1455
1456         dlm_print_version();
1457
1458         status = dlm_init_mle_cache();
1459         if (status)
1460                 return -1;
1461
1462         status = dlm_register_net_handlers();
1463         if (status) {
1464                 dlm_destroy_mle_cache();
1465                 return -1;
1466         }
1467
1468         return 0;
1469 }
1470
1471 static void __exit dlm_exit (void)
1472 {
1473         dlm_unregister_net_handlers();
1474         dlm_destroy_mle_cache();
1475 }
1476
1477 MODULE_AUTHOR("Oracle");
1478 MODULE_LICENSE("GPL");
1479
1480 module_init(dlm_init);
1481 module_exit(dlm_exit);