On Tue, Aug 08, 2006 at 10:33:32AM -0700, Aaron J. Grier wrote:
> On Tue, Aug 08, 2006 at 09:29:29AM -0700, Bill Studenmund wrote:
> > On Mon, Aug 07, 2006 at 09:24:45PM -0700, Brian Buhrow wrote:
> > > 	Perhaps things have changed since last I tried, but as I remember,
> > > when I had a filesystem on /dev/raid0d of a raid5 set some years ago
> > > under NetBSD-1.5 and 1.6/i386, I couldn't run raidctl -P on that
> > > mounted filesystem because the system complained that the device was
> > > busy.  Also, if I started the raid check before the filesystem was
> > > mounted, mount would complain that the device was busy until after
> > > the parity calculation was completed.
> > 
> > Ahhh... That's a consequence of using the same device node for mounting 
> > and for RAID control. So that makes sense.
> 
> I think this behavior changed in netbsd-2; I recall not being able to
> even get "raidctl -s" from raw disk RAIDs.  2.0 fixed that.
> 
> (I've been using raidframe since 1.4.2 days...)

Yup, rev 1.35 of raidctl.c:
date: 2004/02/29 20:40:29;  author: oster;  state: Exp;  lines: +3 -3
As suggested by Paul Ripke, have raidctl use raw devices by default.

Looks like that made it into 2.0.

Cheers,
-- 
Paul Ripke