fs/ocfs2/dlm/dlmthread.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dlmthread.c
   5  *
   6  * standalone DLM module
   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
  28 #include <linux/module.h>
  29 #include <linux/fs.h>
  30 #include <linux/types.h>
  31 #include <linux/highmem.h>
  32 #include <linux/init.h>
  33 #include <linux/sysctl.h>
  34 #include <linux/random.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/timer.h>
  39 #include <linux/kthread.h>
  40 #include <linux/delay.h>
  41
  42
  43 #include "cluster/heartbeat.h"
  44 #include "cluster/nodemanager.h"
  45 #include "cluster/tcp.h"
  46
  47 #include "dlmapi.h"
  48 #include "dlmcommon.h"
  49 #include "dlmdomain.h"
  50
  51 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
  52 #include "cluster/masklog.h"
  53
  54 static int dlm_thread(void *data);
  55 static void dlm_flush_asts(struct dlm_ctxt *dlm);
  56
  57 #define dlm_lock_is_remote(dlm, lock)     ((lock)->ml.node != (dlm)->node_num)
  58
  59 /* will exit holding res->spinlock, but may drop in function */
  60 /* waits until flags are cleared on res->state */
  61 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
  62 {
  63         DECLARE_WAITQUEUE(wait, current);
  64
  65         assert_spin_locked(&res->spinlock);
  66
  67         add_wait_queue(&res->wq, &wait);
  68 repeat:
  69         set_current_state(TASK_UNINTERRUPTIBLE);
  70         if (res->state & flags) {
  71                 spin_unlock(&res->spinlock);
  72                 schedule();
  73                 spin_lock(&res->spinlock);
  74                 goto repeat;
  75         }
  76         remove_wait_queue(&res->wq, &wait);
  77         __set_current_state(TASK_RUNNING);
  78 }
  79
  80 int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
  81 {
  82         if (list_empty(&res->granted) &&
  83             list_empty(&res->converting) &&
  84             list_empty(&res->blocked))
  85                 return 0;
  86         return 1;
  87 }
  88
  89 /* "unused": the lockres has no locks, is not on the dirty list,
  90  * has no inflight locks (in the gap between mastery and acquiring
  91  * the first lock), and has no bits in its refmap.
  92  * truly ready to be freed. */
  93 int __dlm_lockres_unused(struct dlm_lock_resource *res)
  94 {
  95         int bit;
  96
  97         assert_spin_locked(&res->spinlock);
  98
  99         if (__dlm_lockres_has_locks(res))
 100                 return 0;
 101
 102         /* Locks are in the process of being created */
 103         if (res->inflight_locks)
 104                 return 0;
 105
 106         if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
 107                 return 0;
 108
 109         if (res->state & (DLM_LOCK_RES_RECOVERING|
 110                         DLM_LOCK_RES_RECOVERY_WAITING))
 111                 return 0;
 112
 113         /* Another node has this resource with this node as the master */
 114         bit = find_next_bit(res->refmap, O2NM_MAX_NODES, 0);
 115         if (bit < O2NM_MAX_NODES)
 116                 return 0;
 117
 118         return 1;
 119 }
 120
 121
 122 /* Call whenever you may have added or deleted something from one of
 123  * the lockres queue's. This will figure out whether it belongs on the
 124  * unused list or not and does the appropriate thing. */
 125 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
 126                               struct dlm_lock_resource *res)
 127 {
 128         assert_spin_locked(&dlm->spinlock);
 129         assert_spin_locked(&res->spinlock);
 130
 131         if (__dlm_lockres_unused(res)){
 132                 if (list_empty(&res->purge)) {
 133                         mlog(0, "%s: Adding res %.*s to purge list\n",
 134                              dlm->name, res->lockname.len, res->lockname.name);
 135
 136                         res->last_used = jiffies;
 137                         dlm_lockres_get(res);
 138                         list_add_tail(&res->purge, &dlm->purge_list);
 139                         dlm->purge_count++;
 140                 }
 141         } else if (!list_empty(&res->purge)) {
 142                 mlog(0, "%s: Removing res %.*s from purge list\n",
 143                      dlm->name, res->lockname.len, res->lockname.name);
 144
 145                 list_del_init(&res->purge);
 146                 dlm_lockres_put(res);
 147                 dlm->purge_count--;
 148         }
 149 }
 150
 151 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
 152                             struct dlm_lock_resource *res)
 153 {
 154         spin_lock(&dlm->spinlock);
 155         spin_lock(&res->spinlock);
 156
 157         __dlm_lockres_calc_usage(dlm, res);
 158
 159         spin_unlock(&res->spinlock);
 160         spin_unlock(&dlm->spinlock);
 161 }
 162
 163 static void dlm_purge_lockres(struct dlm_ctxt *dlm,
 164                              struct dlm_lock_resource *res)
 165 {
 166         int master;
 167         int ret = 0;
 168
 169         assert_spin_locked(&dlm->spinlock);
 170         assert_spin_locked(&res->spinlock);
 171
 172         master = (res->owner == dlm->node_num);
 173
 174         mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
 175              res->lockname.len, res->lockname.name, master);
 176
 177         if (!master) {
 178                 res->state |= DLM_LOCK_RES_DROPPING_REF;
 179                 /* drop spinlock...  retake below */
 180                 spin_unlock(&res->spinlock);
 181                 spin_unlock(&dlm->spinlock);
 182
 183                 spin_lock(&res->spinlock);
 184                 /* This ensures that clear refmap is sent after the set */
 185                 __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
 186                 spin_unlock(&res->spinlock);
 187
 188                 /* clear our bit from the master's refmap, ignore errors */
 189                 ret = dlm_drop_lockres_ref(dlm, res);
 190                 if (ret < 0) {
 191                         if (!dlm_is_host_down(ret))
 192                                 BUG();
 193                 }
 194                 spin_lock(&dlm->spinlock);
 195                 spin_lock(&res->spinlock);
 196         }
 197
 198         if (!list_empty(&res->purge)) {
 199                 mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
 200                      dlm->name, res->lockname.len, res->lockname.name, master);
 201                 list_del_init(&res->purge);
 202                 dlm_lockres_put(res);
 203                 dlm->purge_count--;
 204         }
 205
 206         if (!master && ret != 0) {
 207                 mlog(0, "%s: deref %.*s in progress or master goes down\n",
 208                         dlm->name, res->lockname.len, res->lockname.name);
 209                 spin_unlock(&res->spinlock);
 210                 return;
 211         }
 212
 213         if (!__dlm_lockres_unused(res)) {
 214                 mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
 215                      dlm->name, res->lockname.len, res->lockname.name);
 216                 __dlm_print_one_lock_resource(res);
 217                 BUG();
 218         }
 219
 220         __dlm_unhash_lockres(dlm, res);
 221
 222         spin_lock(&dlm->track_lock);
 223         if (!list_empty(&res->tracking))
 224                 list_del_init(&res->tracking);
 225         else {
 226                 mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
 227                                 res->lockname.len, res->lockname.name);
 228                 __dlm_print_one_lock_resource(res);
 229         }
 230         spin_unlock(&dlm->track_lock);
 231
 232         /* lockres is not in the hash now.  drop the flag and wake up
 233          * any processes waiting in dlm_get_lock_resource. */
 234         if (!master) {
 235                 res->state &= ~DLM_LOCK_RES_DROPPING_REF;
 236                 spin_unlock(&res->spinlock);
 237                 wake_up(&res->wq);
 238         } else
 239                 spin_unlock(&res->spinlock);
 240 }
 241
 242 static void dlm_run_purge_list(struct dlm_ctxt *dlm,
 243                                int purge_now)
 244 {
 245         unsigned int run_max, unused;
 246         unsigned long purge_jiffies;
 247         struct dlm_lock_resource *lockres;
 248
 249         spin_lock(&dlm->spinlock);
 250         run_max = dlm->purge_count;
 251
 252         while(run_max && !list_empty(&dlm->purge_list)) {
 253                 run_max--;
 254
 255                 lockres = list_entry(dlm->purge_list.next,
 256                                      struct dlm_lock_resource, purge);
 257
 258                 spin_lock(&lockres->spinlock);
 259
 260                 purge_jiffies = lockres->last_used +
 261                         msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
 262
 263                 /* Make sure that we want to be processing this guy at
 264                  * this time. */
 265                 if (!purge_now && time_after(purge_jiffies, jiffies)) {
 266                         /* Since resources are added to the purge list
 267                          * in tail order, we can stop at the first
 268                          * unpurgable resource -- anyone added after
 269                          * him will have a greater last_used value */
 270                         spin_unlock(&lockres->spinlock);
 271                         break;
 272                 }
 273
 274                 /* Status of the lockres *might* change so double
 275                  * check. If the lockres is unused, holding the dlm
 276                  * spinlock will prevent people from getting and more
 277                  * refs on it. */
 278                 unused = __dlm_lockres_unused(lockres);
 279                 if (!unused ||
 280                     (lockres->state & DLM_LOCK_RES_MIGRATING) ||
 281                     (lockres->inflight_assert_workers != 0)) {
 282                         mlog(0, "%s: res %.*s is in use or being remastered, "
 283                              "used %d, state %d, assert master workers %u\n",
 284                              dlm->name, lockres->lockname.len,
 285                              lockres->lockname.name,
 286                              !unused, lockres->state,
 287                              lockres->inflight_assert_workers);
 288                         list_move_tail(&lockres->purge, &dlm->purge_list);
 289                         spin_unlock(&lockres->spinlock);
 290                         continue;
 291                 }
 292
 293                 dlm_lockres_get(lockres);
 294
 295                 dlm_purge_lockres(dlm, lockres);
 296
 297                 dlm_lockres_put(lockres);
 298
 299                 /* Avoid adding any scheduling latencies */
 300                 cond_resched_lock(&dlm->spinlock);
 301         }
 302
 303         spin_unlock(&dlm->spinlock);
 304 }
 305
 306 static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
 307                               struct dlm_lock_resource *res)
 308 {
 309         struct dlm_lock *lock, *target;
 310         int can_grant = 1;
 311
 312         /*
 313          * Because this function is called with the lockres
 314          * spinlock, and because we know that it is not migrating/
 315          * recovering/in-progress, it is fine to reserve asts and
 316          * basts right before queueing them all throughout
 317          */
 318         assert_spin_locked(&dlm->ast_lock);
 319         assert_spin_locked(&res->spinlock);
 320         BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
 321                               DLM_LOCK_RES_RECOVERING|
 322                               DLM_LOCK_RES_IN_PROGRESS)));
 323
 324 converting:
 325         if (list_empty(&res->converting))
 326                 goto blocked;
 327         mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
 328              res->lockname.len, res->lockname.name);
 329
 330         target = list_entry(res->converting.next, struct dlm_lock, list);
 331         if (target->ml.convert_type == LKM_IVMODE) {
 332                 mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
 333                      dlm->name, res->lockname.len, res->lockname.name);
 334                 BUG();
 335         }
 336         list_for_each_entry(lock, &res->granted, list) {
 337                 if (lock==target)
 338                         continue;
 339                 if (!dlm_lock_compatible(lock->ml.type,
 340                                          target->ml.convert_type)) {
 341                         can_grant = 0;
 342                         /* queue the BAST if not already */
 343                         if (lock->ml.highest_blocked == LKM_IVMODE) {
 344                                 __dlm_lockres_reserve_ast(res);
 345                                 __dlm_queue_bast(dlm, lock);
 346                         }
 347                         /* update the highest_blocked if needed */
 348                         if (lock->ml.highest_blocked < target->ml.convert_type)
 349                                 lock->ml.highest_blocked =
 350                                         target->ml.convert_type;
 351                 }
 352         }
 353
 354         list_for_each_entry(lock, &res->converting, list) {
 355                 if (lock==target)
 356                         continue;
 357                 if (!dlm_lock_compatible(lock->ml.type,
 358                                          target->ml.convert_type)) {
 359                         can_grant = 0;
 360                         if (lock->ml.highest_blocked == LKM_IVMODE) {
 361                                 __dlm_lockres_reserve_ast(res);
 362                                 __dlm_queue_bast(dlm, lock);
 363                         }
 364                         if (lock->ml.highest_blocked < target->ml.convert_type)
 365                                 lock->ml.highest_blocked =
 366                                         target->ml.convert_type;
 367                 }
 368         }
 369
 370         /* we can convert the lock */
 371         if (can_grant) {
 372                 spin_lock(&target->spinlock);
 373                 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
 374
 375                 mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
 376                      "%d => %d, node %u\n", dlm->name, res->lockname.len,
 377                      res->lockname.name,
 378                      dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
 379                      dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
 380                      target->ml.type,
 381                      target->ml.convert_type, target->ml.node);
 382
 383                 target->ml.type = target->ml.convert_type;
 384                 target->ml.convert_type = LKM_IVMODE;
 385                 list_move_tail(&target->list, &res->granted);
 386
 387                 BUG_ON(!target->lksb);
 388                 target->lksb->status = DLM_NORMAL;
 389
 390                 spin_unlock(&target->spinlock);
 391
 392                 __dlm_lockres_reserve_ast(res);
 393                 __dlm_queue_ast(dlm, target);
 394                 /* go back and check for more */
 395                 goto converting;
 396         }
 397
 398 blocked:
 399         if (list_empty(&res->blocked))
 400                 goto leave;
 401         target = list_entry(res->blocked.next, struct dlm_lock, list);
 402
 403         list_for_each_entry(lock, &res->granted, list) {
 404                 if (lock==target)
 405                         continue;
 406                 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
 407                         can_grant = 0;
 408                         if (lock->ml.highest_blocked == LKM_IVMODE) {
 409                                 __dlm_lockres_reserve_ast(res);
 410                                 __dlm_queue_bast(dlm, lock);
 411                         }
 412                         if (lock->ml.highest_blocked < target->ml.type)
 413                                 lock->ml.highest_blocked = target->ml.type;
 414                 }
 415         }
 416
 417         list_for_each_entry(lock, &res->converting, list) {
 418                 if (lock==target)
 419                         continue;
 420                 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
 421                         can_grant = 0;
 422                         if (lock->ml.highest_blocked == LKM_IVMODE) {
 423                                 __dlm_lockres_reserve_ast(res);
 424                                 __dlm_queue_bast(dlm, lock);
 425                         }
 426                         if (lock->ml.highest_blocked < target->ml.type)
 427                                 lock->ml.highest_blocked = target->ml.type;
 428                 }
 429         }
 430
 431         /* we can grant the blocked lock (only
 432          * possible if converting list empty) */
 433         if (can_grant) {
 434                 spin_lock(&target->spinlock);
 435                 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
 436
 437                 mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
 438                      "node %u\n", dlm->name, res->lockname.len,
 439                      res->lockname.name,
 440                      dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
 441                      dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
 442                      target->ml.type, target->ml.node);
 443
 444                 /* target->ml.type is already correct */
 445                 list_move_tail(&target->list, &res->granted);
 446
 447                 BUG_ON(!target->lksb);
 448                 target->lksb->status = DLM_NORMAL;
 449
 450                 spin_unlock(&target->spinlock);
 451
 452                 __dlm_lockres_reserve_ast(res);
 453                 __dlm_queue_ast(dlm, target);
 454                 /* go back and check for more */
 455                 goto converting;
 456         }
 457
 458 leave:
 459         return;
 460 }
 461
 462 /* must have NO locks when calling this with res !=NULL * */
 463 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
 464 {
 465         if (res) {
 466                 spin_lock(&dlm->spinlock);
 467                 spin_lock(&res->spinlock);
 468                 __dlm_dirty_lockres(dlm, res);
 469                 spin_unlock(&res->spinlock);
 470                 spin_unlock(&dlm->spinlock);
 471         }
 472         wake_up(&dlm->dlm_thread_wq);
 473 }
 474
 475 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
 476 {
 477         assert_spin_locked(&dlm->spinlock);
 478         assert_spin_locked(&res->spinlock);
 479
 480         /* don't shuffle secondary queues */
 481         if ((res->owner == dlm->node_num)) {
 482                 if (res->state & (DLM_LOCK_RES_MIGRATING |
 483                                   DLM_LOCK_RES_BLOCK_DIRTY))
 484                     return;
 485
 486                 if (list_empty(&res->dirty)) {
 487                         /* ref for dirty_list */
 488                         dlm_lockres_get(res);
 489                         list_add_tail(&res->dirty, &dlm->dirty_list);
 490                         res->state |= DLM_LOCK_RES_DIRTY;
 491                 }
 492         }
 493
 494         mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
 495              res->lockname.name);
 496 }
 497
 498
 499 /* Launch the NM thread for the mounted volume */
 500 int dlm_launch_thread(struct dlm_ctxt *dlm)
 501 {
 502         mlog(0, "Starting dlm_thread...\n");
 503
 504         dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm-%s",
 505                         dlm->name);
 506         if (IS_ERR(dlm->dlm_thread_task)) {
 507                 mlog_errno(PTR_ERR(dlm->dlm_thread_task));
 508                 dlm->dlm_thread_task = NULL;
 509                 return -EINVAL;
 510         }
 511
 512         return 0;
 513 }
 514
 515 void dlm_complete_thread(struct dlm_ctxt *dlm)
 516 {
 517         if (dlm->dlm_thread_task) {
 518                 mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
 519                 kthread_stop(dlm->dlm_thread_task);
 520                 dlm->dlm_thread_task = NULL;
 521         }
 522 }
 523
 524 static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
 525 {
 526         int empty;
 527
 528         spin_lock(&dlm->spinlock);
 529         empty = list_empty(&dlm->dirty_list);
 530         spin_unlock(&dlm->spinlock);
 531
 532         return empty;
 533 }
 534
 535 static void dlm_flush_asts(struct dlm_ctxt *dlm)
 536 {
 537         int ret;
 538         struct dlm_lock *lock;
 539         struct dlm_lock_resource *res;
 540         u8 hi;
 541
 542         spin_lock(&dlm->ast_lock);
 543         while (!list_empty(&dlm->pending_asts)) {
 544                 lock = list_entry(dlm->pending_asts.next,
 545                                   struct dlm_lock, ast_list);
 546                 /* get an extra ref on lock */
 547                 dlm_lock_get(lock);
 548                 res = lock->lockres;
 549                 mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
 550                      "node %u\n", dlm->name, res->lockname.len,
 551                      res->lockname.name,
 552                      dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
 553                      dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
 554                      lock->ml.type, lock->ml.node);
 555
 556                 BUG_ON(!lock->ast_pending);
 557
 558                 /* remove from list (including ref) */
 559                 list_del_init(&lock->ast_list);
 560                 dlm_lock_put(lock);
 561                 spin_unlock(&dlm->ast_lock);
 562
 563                 if (lock->ml.node != dlm->node_num) {
 564                         ret = dlm_do_remote_ast(dlm, res, lock);
 565                         if (ret < 0)
 566                                 mlog_errno(ret);
 567                 } else
 568                         dlm_do_local_ast(dlm, res, lock);
 569
 570                 spin_lock(&dlm->ast_lock);
 571
 572                 /* possible that another ast was queued while
 573                  * we were delivering the last one */
 574                 if (!list_empty(&lock->ast_list)) {
 575                         mlog(0, "%s: res %.*s, AST queued while flushing last "
 576                              "one\n", dlm->name, res->lockname.len,
 577                              res->lockname.name);
 578                 } else
 579                         lock->ast_pending = 0;
 580
 581                 /* drop the extra ref.
 582                  * this may drop it completely. */
 583                 dlm_lock_put(lock);
 584                 dlm_lockres_release_ast(dlm, res);
 585         }
 586
 587         while (!list_empty(&dlm->pending_basts)) {
 588                 lock = list_entry(dlm->pending_basts.next,
 589                                   struct dlm_lock, bast_list);
 590                 /* get an extra ref on lock */
 591                 dlm_lock_get(lock);
 592                 res = lock->lockres;
 593
 594                 BUG_ON(!lock->bast_pending);
 595
 596                 /* get the highest blocked lock, and reset */
 597                 spin_lock(&lock->spinlock);
 598                 BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
 599                 hi = lock->ml.highest_blocked;
 600                 lock->ml.highest_blocked = LKM_IVMODE;
 601                 spin_unlock(&lock->spinlock);
 602
 603                 /* remove from list (including ref) */
 604                 list_del_init(&lock->bast_list);
 605                 dlm_lock_put(lock);
 606                 spin_unlock(&dlm->ast_lock);
 607
 608                 mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
 609                      "blocked %d, node %u\n",
 610                      dlm->name, res->lockname.len, res->lockname.name,
 611                      dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
 612                      dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
 613                      hi, lock->ml.node);
 614
 615                 if (lock->ml.node != dlm->node_num) {
 616                         ret = dlm_send_proxy_bast(dlm, res, lock, hi);
 617                         if (ret < 0)
 618                                 mlog_errno(ret);
 619                 } else
 620                         dlm_do_local_bast(dlm, res, lock, hi);
 621
 622                 spin_lock(&dlm->ast_lock);
 623
 624                 /* possible that another bast was queued while
 625                  * we were delivering the last one */
 626                 if (!list_empty(&lock->bast_list)) {
 627                         mlog(0, "%s: res %.*s, BAST queued while flushing last "
 628                              "one\n", dlm->name, res->lockname.len,
 629                              res->lockname.name);
 630                 } else
 631                         lock->bast_pending = 0;
 632
 633                 /* drop the extra ref.
 634                  * this may drop it completely. */
 635                 dlm_lock_put(lock);
 636                 dlm_lockres_release_ast(dlm, res);
 637         }
 638         wake_up(&dlm->ast_wq);
 639         spin_unlock(&dlm->ast_lock);
 640 }
 641
 642
 643 #define DLM_THREAD_TIMEOUT_MS (4 * 1000)
 644 #define DLM_THREAD_MAX_DIRTY  100
 645 #define DLM_THREAD_MAX_ASTS   10
 646
 647 static int dlm_thread(void *data)
 648 {
 649         struct dlm_lock_resource *res;
 650         struct dlm_ctxt *dlm = data;
 651         unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
 652
 653         mlog(0, "dlm thread running for %s...\n", dlm->name);
 654
 655         while (!kthread_should_stop()) {
 656                 int n = DLM_THREAD_MAX_DIRTY;
 657
 658                 /* dlm_shutting_down is very point-in-time, but that
 659                  * doesn't matter as we'll just loop back around if we
 660                  * get false on the leading edge of a state
 661                  * transition. */
 662                 dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
 663
 664                 /* We really don't want to hold dlm->spinlock while
 665                  * calling dlm_shuffle_lists on each lockres that
 666                  * needs to have its queues adjusted and AST/BASTs
 667                  * run.  So let's pull each entry off the dirty_list
 668                  * and drop dlm->spinlock ASAP.  Once off the list,
 669                  * res->spinlock needs to be taken again to protect
 670                  * the queues while calling dlm_shuffle_lists.  */
 671                 spin_lock(&dlm->spinlock);
 672                 while (!list_empty(&dlm->dirty_list)) {
 673                         int delay = 0;
 674                         res = list_entry(dlm->dirty_list.next,
 675                                          struct dlm_lock_resource, dirty);
 676
 677                         /* peel a lockres off, remove it from the list,
 678                          * unset the dirty flag and drop the dlm lock */
 679                         BUG_ON(!res);
 680                         dlm_lockres_get(res);
 681
 682                         spin_lock(&res->spinlock);
 683                         /* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
 684                         list_del_init(&res->dirty);
 685                         spin_unlock(&res->spinlock);
 686                         spin_unlock(&dlm->spinlock);
 687                         /* Drop dirty_list ref */
 688                         dlm_lockres_put(res);
 689
 690                         /* lockres can be re-dirtied/re-added to the
 691                          * dirty_list in this gap, but that is ok */
 692
 693                         spin_lock(&dlm->ast_lock);
 694                         spin_lock(&res->spinlock);
 695                         if (res->owner != dlm->node_num) {
 696                                 __dlm_print_one_lock_resource(res);
 697                                 mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
 698                                      " dirty %d\n", dlm->name,
 699                                      !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
 700                                      !!(res->state & DLM_LOCK_RES_MIGRATING),
 701                                      !!(res->state & DLM_LOCK_RES_RECOVERING),
 702                                      !!(res->state & DLM_LOCK_RES_DIRTY));
 703                         }
 704                         BUG_ON(res->owner != dlm->node_num);
 705
 706                         /* it is now ok to move lockreses in these states
 707                          * to the dirty list, assuming that they will only be
 708                          * dirty for a short while. */
 709                         BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
 710                         if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
 711                                           DLM_LOCK_RES_RECOVERING |
 712                                           DLM_LOCK_RES_RECOVERY_WAITING)) {
 713                                 /* move it to the tail and keep going */
 714                                 res->state &= ~DLM_LOCK_RES_DIRTY;
 715                                 spin_unlock(&res->spinlock);
 716                                 spin_unlock(&dlm->ast_lock);
 717                                 mlog(0, "%s: res %.*s, inprogress, delay list "
 718                                      "shuffle, state %d\n", dlm->name,
 719                                      res->lockname.len, res->lockname.name,
 720                                      res->state);
 721                                 delay = 1;
 722                                 goto in_progress;
 723                         }
 724
 725                         /* at this point the lockres is not migrating/
 726                          * recovering/in-progress.  we have the lockres
 727                          * spinlock and do NOT have the dlm lock.
 728                          * safe to reserve/queue asts and run the lists. */
 729
 730                         /* called while holding lockres lock */
 731                         dlm_shuffle_lists(dlm, res);
 732                         res->state &= ~DLM_LOCK_RES_DIRTY;
 733                         spin_unlock(&res->spinlock);
 734                         spin_unlock(&dlm->ast_lock);
 735
 736                         dlm_lockres_calc_usage(dlm, res);
 737
 738 in_progress:
 739
 740                         spin_lock(&dlm->spinlock);
 741                         /* if the lock was in-progress, stick
 742                          * it on the back of the list */
 743                         if (delay) {
 744                                 spin_lock(&res->spinlock);
 745                                 __dlm_dirty_lockres(dlm, res);
 746                                 spin_unlock(&res->spinlock);
 747                         }
 748                         dlm_lockres_put(res);
 749
 750                         /* unlikely, but we may need to give time to
 751                          * other tasks */
 752                         if (!--n) {
 753                                 mlog(0, "%s: Throttling dlm thread\n",
 754                                      dlm->name);
 755                                 break;
 756                         }
 757                 }
 758
 759                 spin_unlock(&dlm->spinlock);
 760                 dlm_flush_asts(dlm);
 761
 762                 /* yield and continue right away if there is more work to do */
 763                 if (!n) {
 764                         cond_resched();
 765                         continue;
 766                 }
 767
 768                 wait_event_interruptible_timeout(dlm->dlm_thread_wq,
 769                                                  !dlm_dirty_list_empty(dlm) ||
 770                                                  kthread_should_stop(),
 771                                                  timeout);
 772         }
 773
 774         mlog(0, "quitting DLM thread\n");
 775         return 0;
 776 }