This one fell through the cracks.  Simple fix, though.

http://mail-index.netbsd.org/netbsd-bugs/1997/12/01/0001.html

There's a very simple fix to this.  Every once in a while the disklabel
will get blown away for whatever reason; corruption, dd(8)'ing /dev/zero
into the first few sectors of the disk after using in in a CCD or
RAIDFrame set, etc., etc.

when you disklabel the disk, disklabel(8) will print out a message like
so:

	$ disklabel -e /dev/sd5
	[fullscreen $EDITOR mode works fine]
	disklabel: ioctl DIOCWDINFO: No disk label on disk;
	use "disklabel -I" to install initial label
	re-edit the label? [y]: n

However the cute recommendation to use -I is insuffcient

	$ disklabel -I -e /dev/sd5 

...will give you the same results, what you NEED is:

	$ disklabel -I -r -e /dev/sd5

Or as described in the ticket:

	zero# disklabel -i sd2
	partition> W
	Label disk [n]? y
	disklabel: ioctl DIOCWDINFO: No disk label on disk;
	use "disklabel -r" to install initial label
	Label not written
	partition> quit  4


The suggested fix:
	
	disklabel should write the in-kernel version (psuedo, whatever)
	disklabel to the drive even if it doesn't have a disklabel.  	This
	causes the end-user to have to exit out of this wonderful 	utility
	and defeats the -i usage.  "W" implies that it will write this
	label.. failing and telling the user to quite and run disklabel
	again is just.. lame, since it's displaying the disklabel it's
	supposed to write _anyway_!

This works, or a simple work-around would be to put a note in the
EXAMPLES section of the man page:

	Line 386:
	
	it appropriately (total size, partition offsets, types, etc.)
	.Dl disklabel -I -r -e /dev/sd0
	.Pp
	Recover and restore in-kernel disklabel after possibile 	corruption.
	.Sh DIAGNOSTICS