This isn't a new problem. The underlying devices can change for any number of reasons:
- recabling or moving disks to different controllers. - attaching disks for cloned drivers (e.g. "sd* at scsibus?") that change the probe order (e.g. adding a SCSI disk with ID=1 when a disk with ID=2 is already present on the same bus). Historically solving this has been punted to the system administrator editing /etc/dumpdates when needed. I could be easily convinced that /etc/dumpdates should have the file system name as the first field, as that unambigously identifies what was dumped. What physical device it resides on really is immaterial. Maybe we should change the default. But that needs to be carefully tought out and managed. Meanwhile the -U flag of dump(8) might be helpful. --chris
