[PATCH] OCFS2: The Second Oracle Cluster Filesystem

A distributed lock manager built with the cluster file system use case
in mind. The OCFS2 dlm exposes a VMS style API, though things have
been simplified internally. The only lock levels implemented currently
are NLMODE, PRMODE and EXMODE.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>

1 files changed, 695 insertions, 0 deletions

diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
new file mode 100644
index 000000000000..92cd5cd66db8
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -0,0 +1,695 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmthread.c
+ *
+ * standalone DLM module
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/utsname.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/timer.h>
+#include <linux/kthread.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+
+#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
+#include "cluster/masklog.h"
+
+extern spinlock_t dlm_domain_lock;
+extern struct list_head dlm_domains;
+
+static int dlm_thread(void *data);
+
+static void dlm_flush_asts(struct dlm_ctxt *dlm);
+
+#define dlm_lock_is_remote(dlm, lock)     ((lock)->ml.node != (dlm)->node_num)
+
+/* will exit holding res->spinlock, but may drop in function */
+/* waits until flags are cleared on res->state */
+void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
+{
+        DECLARE_WAITQUEUE(wait, current);
+
+        assert_spin_locked(&res->spinlock);
+
+        add_wait_queue(&res->wq, &wait);
+repeat:
+        set_current_state(TASK_UNINTERRUPTIBLE);
+        if (res->state & flags) {
+                spin_unlock(&res->spinlock);
+                schedule();
+                spin_lock(&res->spinlock);
+                goto repeat;
+        }
+        remove_wait_queue(&res->wq, &wait);
+        current->state = TASK_RUNNING;
+}
+
+
+static int __dlm_lockres_unused(struct dlm_lock_resource *res)
+{
+        if (list_empty(&res->granted) &&
+            list_empty(&res->converting) &&
+            list_empty(&res->blocked) &&
+            list_empty(&res->dirty))
+                return 1;
+        return 0;
+}
+
+
+/* Call whenever you may have added or deleted something from one of
+ * the lockres queue's. This will figure out whether it belongs on the
+ * unused list or not and does the appropriate thing. */
+void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+                              struct dlm_lock_resource *res)
+{
+        mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
+
+        assert_spin_locked(&dlm->spinlock);
+        assert_spin_locked(&res->spinlock);
+
+        if (__dlm_lockres_unused(res)){
+                if (list_empty(&res->purge)) {
+                        mlog(0, "putting lockres %.*s from purge list\n",
+                             res->lockname.len, res->lockname.name);
+
+                        res->last_used = jiffies;
+                        list_add_tail(&res->purge, &dlm->purge_list);
+                        dlm->purge_count++;
+                }
+        } else if (!list_empty(&res->purge)) {
+                mlog(0, "removing lockres %.*s from purge list\n",
+                     res->lockname.len, res->lockname.name);
+
+                list_del_init(&res->purge);
+                dlm->purge_count--;
+        }
+}
+
+void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+                            struct dlm_lock_resource *res)
+{
+        mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
+        spin_lock(&dlm->spinlock);
+        spin_lock(&res->spinlock);
+
+        __dlm_lockres_calc_usage(dlm, res);
+
+        spin_unlock(&res->spinlock);
+        spin_unlock(&dlm->spinlock);
+}
+
+/* TODO: Eventual API: Called with the dlm spinlock held, may drop it
+ * to do migration, but will re-acquire before exit. */
+void dlm_purge_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *lockres)
+{
+        int master;
+        int ret;
+
+        spin_lock(&lockres->spinlock);
+        master = lockres->owner == dlm->node_num;
+        spin_unlock(&lockres->spinlock);
+
+        mlog(0, "purging lockres %.*s, master = %d\n", lockres->lockname.len,
+             lockres->lockname.name, master);
+
+        /* Non master is the easy case -- no migration required, just
+         * quit. */
+        if (!master)
+                goto finish;
+
+        /* Wheee! Migrate lockres here! */
+        spin_unlock(&dlm->spinlock);
+again:
+
+        ret = dlm_migrate_lockres(dlm, lockres, O2NM_MAX_NODES);
+        if (ret == -ENOTEMPTY) {
+                mlog(ML_ERROR, "lockres %.*s still has local locks!\n",
+                     lockres->lockname.len, lockres->lockname.name);
+
+                BUG();
+        } else if (ret < 0) {
+                mlog(ML_NOTICE, "lockres %.*s: migrate failed, retrying\n",
+                     lockres->lockname.len, lockres->lockname.name);
+                goto again;
+        }
+
+        spin_lock(&dlm->spinlock);
+
+finish:
+        if (!list_empty(&lockres->purge)) {
+                list_del_init(&lockres->purge);
+                dlm->purge_count--;
+        }
+        __dlm_unhash_lockres(lockres);
+}
+
+static void dlm_run_purge_list(struct dlm_ctxt *dlm,
+                               int purge_now)
+{
+        unsigned int run_max, unused;
+        unsigned long purge_jiffies;
+        struct dlm_lock_resource *lockres;
+
+        spin_lock(&dlm->spinlock);
+        run_max = dlm->purge_count;
+
+        while(run_max && !list_empty(&dlm->purge_list)) {
+                run_max--;
+
+                lockres = list_entry(dlm->purge_list.next,
+                                     struct dlm_lock_resource, purge);
+
+                /* Status of the lockres *might* change so double
+                 * check. If the lockres is unused, holding the dlm
+                 * spinlock will prevent people from getting and more
+                 * refs on it -- there's no need to keep the lockres
+                 * spinlock. */
+                spin_lock(&lockres->spinlock);
+                unused = __dlm_lockres_unused(lockres);
+                spin_unlock(&lockres->spinlock);
+
+                if (!unused)
+                        continue;
+
+                purge_jiffies = lockres->last_used +
+                        msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
+
+                /* Make sure that we want to be processing this guy at
+                 * this time. */
+                if (!purge_now && time_after(purge_jiffies, jiffies)) {
+                        /* Since resources are added to the purge list
+                         * in tail order, we can stop at the first
+                         * unpurgable resource -- anyone added after
+                         * him will have a greater last_used value */
+                        break;
+                }
+
+                list_del_init(&lockres->purge);
+                dlm->purge_count--;
+
+                /* This may drop and reacquire the dlm spinlock if it
+                 * has to do migration. */
+                mlog(0, "calling dlm_purge_lockres!\n");
+                dlm_purge_lockres(dlm, lockres);
+                mlog(0, "DONE calling dlm_purge_lockres!\n");
+
+                /* Avoid adding any scheduling latencies */
+                cond_resched_lock(&dlm->spinlock);
+        }
+
+        spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
+                              struct dlm_lock_resource *res)
+{
+        struct dlm_lock *lock, *target;
+        struct list_head *iter;
+        struct list_head *head;
+        int can_grant = 1;
+
+        //mlog(0, "res->lockname.len=%d\n", res->lockname.len);
+        //mlog(0, "res->lockname.name=%p\n", res->lockname.name);
+        //mlog(0, "shuffle res %.*s\n", res->lockname.len,
+        //        res->lockname.name);
+
+        /* because this function is called with the lockres
+         * spinlock, and because we know that it is not migrating/
+         * recovering/in-progress, it is fine to reserve asts and
+         * basts right before queueing them all throughout */
+        assert_spin_locked(&res->spinlock);
+        BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
+                              DLM_LOCK_RES_RECOVERING|
+                              DLM_LOCK_RES_IN_PROGRESS)));
+
+converting:
+        if (list_empty(&res->converting))
+                goto blocked;
+        mlog(0, "res %.*s has locks on a convert queue\n", res->lockname.len,
+             res->lockname.name);
+
+        target = list_entry(res->converting.next, struct dlm_lock, list);
+        if (target->ml.convert_type == LKM_IVMODE) {
+                mlog(ML_ERROR, "%.*s: converting a lock with no "
+                     "convert_type!\n", res->lockname.len, res->lockname.name);
+                BUG();
+        }
+        head = &res->granted;
+        list_for_each(iter, head) {
+                lock = list_entry(iter, struct dlm_lock, list);
+                if (lock==target)
+                        continue;
+                if (!dlm_lock_compatible(lock->ml.type,
+                                         target->ml.convert_type)) {
+                        can_grant = 0;
+                        /* queue the BAST if not already */
+                        if (lock->ml.highest_blocked == LKM_IVMODE) {
+                                __dlm_lockres_reserve_ast(res);
+                                dlm_queue_bast(dlm, lock);
+                        }
+                        /* update the highest_blocked if needed */
+                        if (lock->ml.highest_blocked < target->ml.convert_type)
+                                lock->ml.highest_blocked =
+                                        target->ml.convert_type;
+                }
+        }
+        head = &res->converting;
+        list_for_each(iter, head) {
+                lock = list_entry(iter, struct dlm_lock, list);
+                if (lock==target)
+                        continue;
+                if (!dlm_lock_compatible(lock->ml.type,
+                                         target->ml.convert_type)) {
+                        can_grant = 0;
+                        if (lock->ml.highest_blocked == LKM_IVMODE) {
+                                __dlm_lockres_reserve_ast(res);
+                                dlm_queue_bast(dlm, lock);
+                        }
+                        if (lock->ml.highest_blocked < target->ml.convert_type)
+                                lock->ml.highest_blocked =
+                                        target->ml.convert_type;
+                }
+        }
+
+        /* we can convert the lock */
+        if (can_grant) {
+                spin_lock(&target->spinlock);
+                BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
+
+                mlog(0, "calling ast for converting lock: %.*s, have: %d, "
+                     "granting: %d, node: %u\n", res->lockname.len,
+                     res->lockname.name, target->ml.type,
+                     target->ml.convert_type, target->ml.node);
+
+                target->ml.type = target->ml.convert_type;
+                target->ml.convert_type = LKM_IVMODE;
+                list_del_init(&target->list);
+                list_add_tail(&target->list, &res->granted);
+
+                BUG_ON(!target->lksb);
+                target->lksb->status = DLM_NORMAL;
+
+                spin_unlock(&target->spinlock);
+
+                __dlm_lockres_reserve_ast(res);
+                dlm_queue_ast(dlm, target);
+                /* go back and check for more */
+                goto converting;
+        }
+
+blocked:
+        if (list_empty(&res->blocked))
+                goto leave;
+        target = list_entry(res->blocked.next, struct dlm_lock, list);
+
+        head = &res->granted;
+        list_for_each(iter, head) {
+                lock = list_entry(iter, struct dlm_lock, list);
+                if (lock==target)
+                        continue;
+                if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
+                        can_grant = 0;
+                        if (lock->ml.highest_blocked == LKM_IVMODE) {
+                                __dlm_lockres_reserve_ast(res);
+                                dlm_queue_bast(dlm, lock);
+                        }
+                        if (lock->ml.highest_blocked < target->ml.type)
+                                lock->ml.highest_blocked = target->ml.type;
+                }
+        }
+
+        head = &res->converting;
+        list_for_each(iter, head) {
+                lock = list_entry(iter, struct dlm_lock, list);
+                if (lock==target)
+                        continue;
+                if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
+                        can_grant = 0;
+                        if (lock->ml.highest_blocked == LKM_IVMODE) {
+                                __dlm_lockres_reserve_ast(res);
+                                dlm_queue_bast(dlm, lock);
+                        }
+                        if (lock->ml.highest_blocked < target->ml.type)
+                                lock->ml.highest_blocked = target->ml.type;
+                }
+        }
+
+        /* we can grant the blocked lock (only
+         * possible if converting list empty) */
+        if (can_grant) {
+                spin_lock(&target->spinlock);
+                BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
+
+                mlog(0, "calling ast for blocked lock: %.*s, granting: %d, "
+                     "node: %u\n", res->lockname.len, res->lockname.name,
+                     target->ml.type, target->ml.node);
+
+                // target->ml.type is already correct
+                list_del_init(&target->list);
+                list_add_tail(&target->list, &res->granted);
+
+                BUG_ON(!target->lksb);
+                target->lksb->status = DLM_NORMAL;
+
+                spin_unlock(&target->spinlock);
+
+                __dlm_lockres_reserve_ast(res);
+                dlm_queue_ast(dlm, target);
+                /* go back and check for more */
+                goto converting;
+        }
+
+leave:
+        return;
+}
+
+/* must have NO locks when calling this with res !=NULL * */
+void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+        mlog_entry("dlm=%p, res=%p\n", dlm, res);
+        if (res) {
+                spin_lock(&dlm->spinlock);
+                spin_lock(&res->spinlock);
+                __dlm_dirty_lockres(dlm, res);
+                spin_unlock(&res->spinlock);
+                spin_unlock(&dlm->spinlock);
+        }
+        wake_up(&dlm->dlm_thread_wq);
+}
+
+void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+        mlog_entry("dlm=%p, res=%p\n", dlm, res);
+
+        assert_spin_locked(&dlm->spinlock);
+        assert_spin_locked(&res->spinlock);
+
+        /* don't shuffle secondary queues */
+        if ((res->owner == dlm->node_num) &&
+            !(res->state & DLM_LOCK_RES_DIRTY)) {
+                list_add_tail(&res->dirty, &dlm->dirty_list);
+                res->state |= DLM_LOCK_RES_DIRTY;
+        }
+}
+
+
+/* Launch the NM thread for the mounted volume */
+int dlm_launch_thread(struct dlm_ctxt *dlm)
+{
+        mlog(0, "starting dlm thread...\n");
+
+        dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm_thread");
+        if (IS_ERR(dlm->dlm_thread_task)) {
+                mlog_errno(PTR_ERR(dlm->dlm_thread_task));
+                dlm->dlm_thread_task = NULL;
+                return -EINVAL;
+        }
+
+        return 0;
+}
+
+void dlm_complete_thread(struct dlm_ctxt *dlm)
+{
+        if (dlm->dlm_thread_task) {
+                mlog(ML_KTHREAD, "waiting for dlm thread to exit\n");
+                kthread_stop(dlm->dlm_thread_task);
+                dlm->dlm_thread_task = NULL;
+        }
+}
+
+static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
+{
+        int empty;
+
+        spin_lock(&dlm->spinlock);
+        empty = list_empty(&dlm->dirty_list);
+        spin_unlock(&dlm->spinlock);
+
+        return empty;
+}
+
+static void dlm_flush_asts(struct dlm_ctxt *dlm)
+{
+        int ret;
+        struct dlm_lock *lock;
+        struct dlm_lock_resource *res;
+        u8 hi;
+
+        spin_lock(&dlm->ast_lock);
+        while (!list_empty(&dlm->pending_asts)) {
+                lock = list_entry(dlm->pending_asts.next,
+                                  struct dlm_lock, ast_list);
+                /* get an extra ref on lock */
+                dlm_lock_get(lock);
+                res = lock->lockres;
+                mlog(0, "delivering an ast for this lockres\n");
+
+                BUG_ON(!lock->ast_pending);
+
+                /* remove from list (including ref) */
+                list_del_init(&lock->ast_list);
+                dlm_lock_put(lock);
+                spin_unlock(&dlm->ast_lock);
+
+                if (lock->ml.node != dlm->node_num) {
+                        ret = dlm_do_remote_ast(dlm, res, lock);
+                        if (ret < 0)
+                                mlog_errno(ret);
+                } else
+                        dlm_do_local_ast(dlm, res, lock);
+
+                spin_lock(&dlm->ast_lock);
+
+                /* possible that another ast was queued while
+                 * we were delivering the last one */
+                if (!list_empty(&lock->ast_list)) {
+                        mlog(0, "aha another ast got queued while "
+                             "we were finishing the last one.  will "
+                             "keep the ast_pending flag set.\n");
+                } else
+                        lock->ast_pending = 0;
+
+                /* drop the extra ref.
+                 * this may drop it completely. */
+                dlm_lock_put(lock);
+                dlm_lockres_release_ast(dlm, res);
+        }
+
+        while (!list_empty(&dlm->pending_basts)) {
+                lock = list_entry(dlm->pending_basts.next,
+                                  struct dlm_lock, bast_list);
+                /* get an extra ref on lock */
+                dlm_lock_get(lock);
+                res = lock->lockres;
+
+                BUG_ON(!lock->bast_pending);
+
+                /* get the highest blocked lock, and reset */
+                spin_lock(&lock->spinlock);
+                BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
+                hi = lock->ml.highest_blocked;
+                lock->ml.highest_blocked = LKM_IVMODE;
+                spin_unlock(&lock->spinlock);
+
+                /* remove from list (including ref) */
+                list_del_init(&lock->bast_list);
+                dlm_lock_put(lock);
+                spin_unlock(&dlm->ast_lock);
+
+                mlog(0, "delivering a bast for this lockres "
+                     "(blocked = %d\n", hi);
+
+                if (lock->ml.node != dlm->node_num) {
+                        ret = dlm_send_proxy_bast(dlm, res, lock, hi);
+                        if (ret < 0)
+                                mlog_errno(ret);
+                } else
+                        dlm_do_local_bast(dlm, res, lock, hi);
+
+                spin_lock(&dlm->ast_lock);
+
+                /* possible that another bast was queued while
+                 * we were delivering the last one */
+                if (!list_empty(&lock->bast_list)) {
+                        mlog(0, "aha another bast got queued while "
+                             "we were finishing the last one.  will "
+                             "keep the bast_pending flag set.\n");
+                } else
+                        lock->bast_pending = 0;
+
+                /* drop the extra ref.
+                 * this may drop it completely. */
+                dlm_lock_put(lock);
+                dlm_lockres_release_ast(dlm, res);
+        }
+        wake_up(&dlm->ast_wq);
+        spin_unlock(&dlm->ast_lock);
+}
+
+
+#define DLM_THREAD_TIMEOUT_MS (4 * 1000)
+#define DLM_THREAD_MAX_DIRTY  100
+#define DLM_THREAD_MAX_ASTS   10
+
+static int dlm_thread(void *data)
+{
+        struct dlm_lock_resource *res;
+        struct dlm_ctxt *dlm = data;
+        unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
+
+        mlog(0, "dlm thread running for %s...\n", dlm->name);
+
+        while (!kthread_should_stop()) {
+                int n = DLM_THREAD_MAX_DIRTY;
+
+                /* dlm_shutting_down is very point-in-time, but that
+                 * doesn't matter as we'll just loop back around if we
+                 * get false on the leading edge of a state
+                 * transition. */
+                dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
+
+                /* We really don't want to hold dlm->spinlock while
+                 * calling dlm_shuffle_lists on each lockres that
+                 * needs to have its queues adjusted and AST/BASTs
+                 * run.  So let's pull each entry off the dirty_list
+                 * and drop dlm->spinlock ASAP.  Once off the list,
+                 * res->spinlock needs to be taken again to protect
+                 * the queues while calling dlm_shuffle_lists.  */
+                spin_lock(&dlm->spinlock);
+                while (!list_empty(&dlm->dirty_list)) {
+                        int delay = 0;
+                        res = list_entry(dlm->dirty_list.next,
+                                         struct dlm_lock_resource, dirty);
+
+                        /* peel a lockres off, remove it from the list,
+                         * unset the dirty flag and drop the dlm lock */
+                        BUG_ON(!res);
+                        dlm_lockres_get(res);
+
+                        spin_lock(&res->spinlock);
+                        res->state &= ~DLM_LOCK_RES_DIRTY;
+                        list_del_init(&res->dirty);
+                        spin_unlock(&res->spinlock);
+                        spin_unlock(&dlm->spinlock);
+
+                        /* lockres can be re-dirtied/re-added to the
+                         * dirty_list in this gap, but that is ok */
+
+                        spin_lock(&res->spinlock);
+                        if (res->owner != dlm->node_num) {
+                                __dlm_print_one_lock_resource(res);
+                                mlog(ML_ERROR, "inprog:%s, mig:%s, reco:%s, dirty:%s\n",
+                                     res->state & DLM_LOCK_RES_IN_PROGRESS ? "yes" : "no",
+                                     res->state & DLM_LOCK_RES_MIGRATING ? "yes" : "no",
+                                     res->state & DLM_LOCK_RES_RECOVERING ? "yes" : "no",
+                                     res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
+                        }
+                        BUG_ON(res->owner != dlm->node_num);
+
+                        /* it is now ok to move lockreses in these states
+                         * to the dirty list, assuming that they will only be
+                         * dirty for a short while. */
+                        if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
+                                          DLM_LOCK_RES_MIGRATING |
+                                          DLM_LOCK_RES_RECOVERING)) {
+                                /* move it to the tail and keep going */
+                                spin_unlock(&res->spinlock);
+                                mlog(0, "delaying list shuffling for in-"
+                                     "progress lockres %.*s, state=%d\n",
+                                     res->lockname.len, res->lockname.name,
+                                     res->state);
+                                delay = 1;
+                                goto in_progress;
+                        }
+
+                        /* at this point the lockres is not migrating/
+                         * recovering/in-progress.  we have the lockres
+                         * spinlock and do NOT have the dlm lock.
+                         * safe to reserve/queue asts and run the lists. */
+
+                        mlog(0, "calling dlm_shuffle_lists with dlm=%p, "
+                             "res=%p\n", dlm, res);
+
+                        /* called while holding lockres lock */
+                        dlm_shuffle_lists(dlm, res);
+                        spin_unlock(&res->spinlock);
+
+                        dlm_lockres_calc_usage(dlm, res);
+
+in_progress:
+
+                        spin_lock(&dlm->spinlock);
+                        /* if the lock was in-progress, stick
+                         * it on the back of the list */
+                        if (delay) {
+                                spin_lock(&res->spinlock);
+                                list_add_tail(&res->dirty, &dlm->dirty_list);
+                                res->state |= DLM_LOCK_RES_DIRTY;
+                                spin_unlock(&res->spinlock);
+                        }
+                        dlm_lockres_put(res);
+
+                        /* unlikely, but we may need to give time to
+                         * other tasks */
+                        if (!--n) {
+                                mlog(0, "throttling dlm_thread\n");
+                                break;
+                        }
+                }
+
+                spin_unlock(&dlm->spinlock);
+                dlm_flush_asts(dlm);
+
+                /* yield and continue right away if there is more work to do */
+                if (!n) {
+                        yield();
+                        continue;
+                }
+
+                wait_event_interruptible_timeout(dlm->dlm_thread_wq,
+                                                 !dlm_dirty_list_empty(dlm) ||
+                                                 kthread_should_stop(),
+                                                 timeout);
+        }
+
+        mlog(0, "quitting DLM thread\n");
+        return 0;
+}