dmraid comments and a warning

Peter Jones pjones at redhat.com
Wed Feb 8 16:07:39 UTC 2006


On Tue, 2006-02-07 at 16:22 -0700, Lamont R. Peterson wrote:
> On Tuesday 07 February 2006 03:34pm, Peter Jones wrote:
> > On Tue, 2006-02-07 at 00:17 -0700, Dax Kelson wrote:
> [snip]
> > > And now it uses root=/dev/mapper/$DEV ?
> >
> > No, it still uses root=LABEL=/ (assuming no lvm), but the label
> > searching mechanism early in the boot process is now the same as that
> > used by mount, umount, swapon, etc.,  and it currently gives
> > device-mapper devices a higher "priority", which should guarantee that,
> > assuming it's possible to build the raid, all of those tools will use
> > the dm device instead of the normal disks.
> 
> Perhaps it would be better to have this defaulted on install to *not* use 
> LABEL= mounting syntax.

It'd be better to not have the physical disk partition devices even
exist when we're trying to bring up a system with a dmraid.  I've got a
patch to allow that, and I'm looking on getting it into the kernel (ours
and upstream's) now.

The patch adds a block device ioctl BLKRMPARTS, which simply turns all
of the subdevices on a block device off.  You can turn them back on with
BLKRRPART, which is exposed to sysadmins via "/sbin/blockdev
--rereadpt".  I'll look at adding --rmparts to that utility as well.

> In that case, the broken device would have prevented 
> mounting and, therefore, booting and the problem of dropping half the mirror 
> leading to this corruption would not happen without FRAID users knowing about 
> it.  Yet another case where mounting using LABEL= syntax causes more problems 
> than it solves.

It's not caused by mounting by label.  It's caused by having a device
that you can mount, but where mounting can *only* cause system
corruption.

(Don't get me wrong, I think mounting by label is a bit flawed, but
"just don't do it" isn't quite right either.)

-- 
  Peter




More information about the fedora-devel-list mailing list