[Ocfs2-commits] smushran commits r2888 - branches/ocfs2-1.2/fs/ocfs2/dlm
svn-commits@oss.oracle.com
svn-commits at oss.oracle.com
Wed Apr 19 21:20:37 CDT 2006
Author: smushran
Signed-off-by: mfasheh
Signed-off-by: khackel
Date: 2006-04-19 21:20:36 -0500 (Wed, 19 Apr 2006)
New Revision: 2888
Modified:
branches/ocfs2-1.2/fs/ocfs2/dlm/dlmmaster.c
Log:
Pending mastery asserts and migrations should block each other. Use the
existing structure for blocking migrations when ASTs are pending to achieve
the same result. If we can catch the assert before it goes on the wire,
just cancel it and let the migration continue.
Signed-off-by: mfasheh
Signed-off-by: khackel
Modified: branches/ocfs2-1.2/fs/ocfs2/dlm/dlmmaster.c
===================================================================
--- branches/ocfs2-1.2/fs/ocfs2/dlm/dlmmaster.c 2006-04-20 02:16:25 UTC (rev 2887)
+++ branches/ocfs2-1.2/fs/ocfs2/dlm/dlmmaster.c 2006-04-20 02:20:36 UTC (rev 2888)
@@ -1996,6 +1996,23 @@
}
}
+ /*
+ * If we're migrating this lock to someone else, we are no
+ * longer allowed to assert out own mastery. OTOH, we need to
+ * prevent migration from starting while we're still asserting
+ * our dominance. The reserved ast delays migration.
+ */
+ spin_lock(&res->spinlock);
+ if (res->state & DLM_LOCK_RES_MIGRATING) {
+ mlog(0, "Someone asked us to assert mastery, but we're "
+ "in the middle of migration. Skipping assert, "
+ "the new master will handle that.\n");
+ spin_unlock(&res->spinlock);
+ goto put;
+ } else
+ __dlm_lockres_reserve_ast(res);
+ spin_unlock(&res->spinlock);
+
/* this call now finishes out the nodemap
* even if one or more nodes die */
mlog(0, "worker about to master %.*s here, this=%u\n",
@@ -2008,6 +2025,10 @@
mlog_errno(ret);
}
+ /* Ok, we've asserted ourselves. Let's let migration start. */
+ dlm_lockres_release_ast(dlm, res);
+
+put:
dlm_lockres_put(res);
mlog(0, "finished with dlm_assert_master_worker\n");
More information about the Ocfs2-commits
mailing list