Bug#478598: partman-crypto: problems with using random keys
On Saturday 03 May 2008, Max Vozeler wrote:
> When you select "Random key" for loop-AES, the actual keys
> are generated from /dev/urandom by mount or swapon. We don't
> use cdebconf-entropy for such setups.
Does that mean that I should not have been shown *either* of the two dialogs
(passphrase and random typing) with the "incorrect" method? Should
cdebconf-entropy be used only with dm-crypt?
If that is the case then the current logic is _really_ broken...
It would be nice to have this fixed before Beta2, but not at the risk of
breaking other things.
> > "Incorrect" method:
[...]
> > After this I am first asked to enter an encryption passphrase, even
> > though there is no partition that uses one. This is a bug.
>
> Indeed, this is arguably non-intuitive.
>
> Your earlier choice of random keytype was reset to the default
> for loop-AES, gnupg keyfile, when you changed the encryption
> method.
>
> FWIW, the partman dialog should reflect the reset keytype after
> switching the encryption type.
IIRC it did not. My test should be trivial to reproduce though.
> > After that I *am* asked to enter random characters, with the progress
> > bar at only 2%. Getting sufficient entropy litterally takes ages:
> > getting from 5 to 10% takes 20 seconds. I don't remember it taking that
> > long with previous tests I've done.
>
> Were the earlier tests done in the same environment?
Yes. Exactly the same (I'd done a snapshot in VirtualBox just before
partitioning and based both cases on that).
> Lots of factors contribute to how well (or how badly) the entropy
> pool is being fed by device drivers. IIRC some disk drivers do,
> some don't, some network drivers do, others don't etc.
>
> Apart from that I don't recall any changes that should have made
> key generation more painful than it already was. :-/
Well, IIUC I should not have seen the dialog at all, so it's somewhat
academic. I'll test again with dm-crypt and random keys some time.
> > The interface does allow
> > it, but I seem to remember that supporting random keys was the reason
> > why we still needed support for loop-aes.
>
> No. loop-AES is not a "legacy" for lack of features in dm-crypt.
It might be a good idea to document those in the README file in
partman-crypto (which needs updating anyway as the comments at the top of
that file are completely outdated).
Reply to: