>>>kern/3021 describes poor handling of UNIT ATTENTION in the scsi
>>>driver.  The sense code handling recently got churned, but I don't
>>>think this has been changed.  At any rate, the bug doesn't seem to
>>>have been looked at.  (This one includes only skeletal code.)
>>
>>The real fix for this would be to update sd.c to actually parse errors
>>rather than let the (limited) general case suffice. I won't have time
>>to do this at present.
>
>I've thought about this somewhat, and I don't think that the low level disk
>driver can completely handle this situation.  For instance, if a drive
>loses power for some reason and comes back up, you may be able to verify
>that it is the exact same drive if it supports the serial number inquiry
>page or has a SCSI III WWID, but without a serial number or ID, what do you
>do? 

Inquiry/VPD info. Those devices that don't support this just lose (there
are very few of them now...)

>  Even if you have some identifier, it may not be sufficient for
>removable media devices unless the serial number is media specific.  There
>are still many common devices that don't have serial numbers (Jaz and Zip
>drives come to mind).

Well, now that I think about it, I agree with you on this,
for removable media.

>
>I think you have to have some type of generic media "verification" system
>in addition to properly handling the sense codes.  NT actually sticks some
>kind of identifier in an otherwise unused section of the partition table,
>but there may be other ways to handle this and have it work even if a disk
>doesn't contain a BSD label on it anywhere.
>

I rather think that in the U*ix domain that the filesystem should be
responsible for volume/partition validation.