Re: [gentoo-dev] 2004.1 will not include a secure portage.

[Peter Johanson]

On Tue, Mar 23, 2004 at 02:12:13PM +0100, Paul de Vrieze wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Tuesday 23 March 2004 13:21, Kurt Lieber wrote:
> > On Tue, Mar 23, 2004 at 03:11:47AM -0800 or thereabouts, Robin H.
> Johnson wrote:
> > > I wrote up a functional prototype patch Mon, 8 Dec 2003 and mailed
> > > it to gentoo-core when a discussion on the subject was in progress.
> > > This is the ONLY code I've seen produced by anybody on the subject
> > > of GPG signing to date.
> >
> > rac also had a proof-of-concept working.  However, as I understand it,
> > the issues preventing this from becoming a reality are not technical
> > in nature, but more process and policy oriented.  Quite frankly, I
> > think the only issue standing between us and getting this implemented
> > is having the Portage folks deem this as an important project and
> > prioritize it accordingly.
>
> Ok, the biggest problem is an idea for how the actual checking should
> function. (Signing is straightforward).
>
> I'll take a first shot at describing a key chain idea. Please shoot at it
> and try to find the holes. But remember it needs to stay workable. First
> the premises:
>
> - - All developers should be able to sign with their own key
> - - We can register the keys of the developers before allowing them
> - - We don't necessarilly (want to/can) sign the keys of all the developers
> - - The system is gentoo-specific so we can implement custom systems.
> - - A root key should be kept safe.
>
> What do I propose:
> - - one master key whose public key is put on every cd. It should be signed
>   by as many people as possible and also put on a ssl protected website
>   with a real certificate (payed for).
> - - a signing key. This key is signed by the master key and is rotated on a
>   regular basis (say once a month). The public key is put in the portage
>   tree.
> - - The signing key is used for signing a list of the fingerprints/public
>   keys of all developers with ebuild commit privileges.
> - - After an rsync all signatures for packages with changes are checked.
>   This will first check the signing key based on the known master key
>   (which should change as little as possible and be kept very secure).
>   It also verifies that the signing key is not too old. This is important
>   so that compromise of the signing key will be less of a security issue
>   even when it is not possible to get the revokation of the key to all
>   users.
>   Then it will verify that the list of developer keys is valid based on
>   the signing key.
>   Next it will check that the signature file for this package is correct
>   and validates that the key used for signing this signature is in the
>   list of acceptable keys.
>   If the package is found to be invalid we could start with removing the
>   package from the repository.
> - - As the checking is done post-rsync there is no problem with infrequent
>   rsync-ing.
>
> note:
> *) alternatively there could be a key between the signing key and the
>    master key, this root key could have a one month rotation. Then this
>    root key could be used to have a daily generated signing key.
>
>    The disadvantage is that the root key and signing key would be
>    passphraseless to allow for automatic signing of either the signing
>    key and the dev key list. (assuming that manually entering keyphrases
>    daily is too much)
>
>    *) It might be possible to keep the passphrase for the daily signing
>       key in memory (or even the whole secret part, which would mean
>       that changes in developer keys would only be visible with a one
>       day delay (max) or manual regeneration of a signing key)
>
> Paul

While this is being discussed, I think it would be a very good idea to
investigate the KeyMan project (http://keyman.aldigital.co.uk/) which
was designed by one of the core openSSL guys as a means for doing
distribution signing (it was originally written as a solution for the
apache folks). Ben Laurie (the openSSL guy) is willing (and excited) to
help implement using KeyMan for the signing. It already has a system of
"domains" and "subdomains" in which keys are valid, etc, etc. 

He's willing to port to python (it's in perl now) and would probably be
very proactive in this work. If people are interested, I can get him in
touch with the right people or vice versa.

-pete

>
> - --
> Paul de Vrieze
> Gentoo Developer
> Mail: pauldv@g.o
> Homepage: http://www.devrieze.net
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.2.4 (GNU/Linux)
>
> iD8DBQFAYDetbKx5DBjWFdsRArC0AJ989ls+w+RszK7FazB0I8hq9//swQCg2BUA
> sXkzItayvUSKux+/rfHFx1w=
> =N+pp
> -----END PGP SIGNATURE-----
>
> --
> gentoo-dev@g.o mailing list
>

-- 
Peter Johanson
<latexer@g.o>

Key ID = 0x6EFA3917
Key fingerprint = A90A 2518 57B1 9D20 9B71  A2FF 8649 439B 6EFA 3917