EVMS registers the base name for the top working object of a volume if the
object has its own device. However, EVMS engine fails to unregister such
names when discarding volumes. This prevents some operation on volumes.

The problem magnifies in cluster environments because rediscovery of
CSM containers would fail to add volumes (discarded moments ago) back to
the object tree, and the volumes would get deleted when committing changes.

Rediscovery can be triggered by an operation as simple as changing the
attributes of a CSM container.

I added code to unregister the base name, and tested on my cluster. 

"Changju Gao" <CGAO@novell.com>

--- evms-2.5.5/engine/internalAPI.c	2005-10-10 09:44:41.000000000 -0500
+++ evms-2.5.5-fix/engine/internalAPI.c	2006-11-01 17:11:42.113699000 -0600
@@ -2722,6 +2722,11 @@
 	if (rc == 0) {
 		engine_unregister_name(vol->name);
 
+		if( vol->object->feature_header != NULL )
+		{
+			engine_unregister_name(vol->name+EVMS_DEV_NODE_PATH_LEN);
+		}
+
 		/*
 		 * Clear out the volume pointer on this object and on all the child
 		 * objects in the tree.