Call for testing: lvm2, dmsetup and friends

[Kees Cook]

On Sat, Feb 13, 2010 at 02:00:08PM +0100, Martin Pitt wrote:
> I wrote an udisks test suite the other day, which also covers LVM and
> LVM-on-RAID1 (not very deeply, though, but basic functionality). It

Ah, fantastic!  Can you get this into
lp:~ubuntu-bugcontrol/qa-regression-testing/master in scripts/ ?  That
would be a great addition.

> generally works well, but there's one smal regression (or at least
> behaviour change): RAID-1 mirror images (_mimage*, _mlog*) now get
> by-id/by-uuid symlinks, which creates some confusion/clutter in
> exposes them to userspace (see [1], and linked bug reports).

Ah, right, I didn't test LVM mirroring, just snapshots and MD mirroring.

> 56-lvm.rules covers this case:
> 
> ENV{DM_LV_NAME}=="?*_mimage_[0-9]*|pvmove?*|?*_vorigin", ENV{DM_UDEV_DISABLE_DISK_RULES_FLAG}="1"
> 
> and 60-persistent-storage-dm.rules also checks the
> DM_UDEV_DISABLE_DISK_RULES_FLAG property, but checks it too late, so
> that by-id/by-uuid symlinks are still created.
> 
> Do you know if that was intended? Since they are mostly useless in
> userspace, and are only an internal implementation details, I'd rather
> like to seem them hidden as much as possible.

Hm, I can't tell.  It creates the dm-name-* and dm-uuid-* symlinks, but
skips the symlinks that are based on the device content.  Regardless,
_mlog* isn't caught, I don't think?  I'm going to restore the prior
behavior and submit an update to Debian about it.

> This shouldn't block a lucid upload, though; if it's an intended
> change I can easily update the udisks test suite.

Okay, cool.  I'd like to hear back from at least one DM-RAID user, then
I'll upload it.

Thanks!

-Kees

-- 
Kees Cook
Ubuntu Security Team