Bleh, my bad. Indeed it will be a problem. Thinking about it, I think it could be solvable for 4.3->4.4 - just honor the existing path. We get the source xml in migrationCreate so we can understand if the metadata’s payload already have a disk drive defined. 4.4->4.3 is more tricky. we can fix 4.3 code too..though i’d really like to keep compatiblity with 4.3 too. Then I *think* we can call the migrationCreate - it will create the payload in vmid-md5 location on destination. Then we can call the dumpXML (before starting the migration), get the actual destination path, communicate it down to a libvirt hook on source, and change the xml on the fly. We did that in the past…so may be doable. Ugly…