* Magnus Hagander (magnus@hagander.net) wrote:
> mark@mark.mielke.cc wrote:
> > On Sat, Dec 30, 2006 at 06:05:14PM +0100, Martijn van Oosterhout wrote:
> >> Except tht X.509 is already done (in a sense). The client can supply a
> >> certificate that the server can check, and vice-versa. You can't link
> >> this with the postgresql username yet, but I havn't seen any proposals
> >> about how to do that.
> >
> > I suggest associating the SHA-1 fingerprint with the ROLE. I would love
> > to have this.
>
> I would suggest a map based on the CN. Any org with a centralized PKI
> infrastructure is likely to assign certs with the userid or other unique
> identifier in the CN.
Right, this would be how I'd envision it as well. Basically provide a
CA which you trust and then a way to map from DN/CN to PostgreSQL users
(perhaps similar to the ident mapping in implementation?). I'd *also*
like to support full certificate matching (not just the fingerprint...)
but I think doing DN/CN would be a good, easier, first step.
Thanks,
Stephen