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