My notes on making encrypted filesystems "Just Work(tm)"

David Zeuthen david at
Thu Dec 16 13:44:19 PST 2004

On Thu, 2004-12-16 at 22:15 +0100, Sjoerd Simons wrote:
> > If the kernel could just store the name we pass to dmsetup it would be
> > sufficient. E.g. we just have the invariant that the crypto device is to
> > be called sesame_crypto_<uuid> and the extremly ugly code can be made
> > nice again.
> Would be nice if it would also work with dm devices not created by hal :) Which
> isn't the case when you need the special name.

Well, at the end of the day this would only make sense if hal
understands the usage of devices that have a part in e.g. dm-crypt or
LVM2 schemees. I don't really see hald adding hal device objects for
random pieces of logical devices.

So, I thought about that for teaching hal about LVM2 volumes. It
shouldn't be too hard really; make volume_id understand the LVM2
metadata, populate the volume objects, have a callout check whether all
physical disks are there - when they are do the lvmsetup/dmsetup dance.
Hey presto, you have plug and play of mirrored/striped disks.

For bonus, put up a notification on the desktop saying "Dude, you need
to plug into the *other half* of your striped logical volume" if the
user is too slow :-)


hal mailing list
hal at

More information about the Hal mailing list