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

[Paul de Vrieze]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Wednesday 24 March 2004 15:24, Jesse Nelson wrote:
>
> signing @ a file level (even using signed sig files) stops a
> compromise of a rsync server from being an issue at all. imho a better
> solution. It does push the problem out to the devs keys or whatever
> you do for keys.. thats where the peer review and multiple sigs come
> into play.

I do not oppose normal key revokation to happen next to fast rotation 
based security. (So invalid dev keys can be revoked) We need however to 
have a way to distribute a list of those keys that can sign packages. 
The location of this list can allways be compromised. (One could do a 
DOS on the webnodes where it is stored). The only solution to this is to 
use a short lifetime for such a list.

It is not possible in all cases to actually revoke a key. What if a 
developer leaves gentoo. We can hardly require him to revoke his key if 
he uses it for other identities too. And even if we ask him to, there 
are no guarantees he will. Also using standard web of trust would mean 
that anyone whose key I have signed can sign a package. I don't want 
that as many of those people are not gentoo devs. The only thing my 
signature means is that I verified that they are themselves.

> 100% is totally impracticall but this does double the effort required
> to compromise integrity. thats not that bad of an optimization really.

It also more than quadruples the efforts needed for actually committing a 
change to an ebuild. As a start we can better have single signing that 
spending months on reorganization and devising a way to have practical 
multiple signing.

The problem is that signing can be done today. Multiple signing will need 
probably at least a half year.

> > Yes, it is. However with over 200 developers changing things
> > overnight is like trying to change the direction of a mammuth oil
> > tanker 90 degrees in 15 minutes. It is plain impossible.
>
> i totally agree, but im laying out my voice.. 1 sig and 1 signer is
> really quite easy to get around. 2 is alot harder.. and once you have
> the process in place for 2 sigs its is not difficult to extend it to
> 1+X sigs. where each itteratin of X increase trust in package quite
> alot.

The problem is that with each change to a package all previous signatures 
are invalid (they don't match the package anymore). This means that 
basically when one wants to change a package one must commit and sign it 
yourself. Then find someone who will check the packages again and have 
him sign it again.

We can probably devise a way in which we could allow multiple signatures 
if it were on a file instead of package base. However that would not 
mean that the latest greatest version would be guaranteed multiply 
signed.

Also single or multiple signing does not change the keyring part at all. 
The signing keys still would need to be allowed to sign, so there needs 
to be a list of all keys authorized to sign packages. There is no 
possibility that we can guarantee that we will be able to revoke such a 
list, so the list needs to have a short lifetime as an alternative 
security enhancement. The shorter the lifetime, the safer it is, but 
also more bothersome and resource intensive. 1 day seems to be a good 
compromise.

Paul

- -- 
Paul de Vrieze
Gentoo Developer
Mail: pauldv@g.o
Homepage: http://www.devrieze.net
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)

iD8DBQFAYaDJbKx5DBjWFdsRAsRkAKCF008sdGP9C9EjsY3MFyY6oz8BCACeIZgh
bzdrOmYBZ6VgrJ3/9nYwJ9M=
=/43w
-----END PGP SIGNATURE-----

--
gentoo-dev@g.o mailing list