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