On Mon, 2021-11-08 at 15:32 -0500, Stephen Frost wrote:
> Where does that leave us with what Jonathan is suggesting though? For
> my 2c, we shouldn't allow 'map=' to be used for scram or md5 because
> it'll just confuse users, until and unless we actually do the PGAUTHUSER
> thing and then we can allow 'map=' to check if that mapping is allowed
> and the actual SCRAM PW check is done against PGAUTHUSER and then the
> logged in user is the user as specified in the startup packet, assuming
> that mapping is allowed. For Jonathan's actual case though, we should
> add a 'certmap' option instead and have that be explicitly for the case
> where it's scram w/ clientcert=verify-full and then we check the mapping
> of the DN/CN to the startup-packet username. There's no reason this
> couldn't also work with a 'map' specified and PGAUTHUSER set and SCRAM
> used to verify against that at the same time.
Agreed.
> Perhaps not a surprise, but I continue to be against the idea of adding
> anything more to the insecure hack that is our LDAP auth method. We
> should be moving away from that, not adding to it.
Paraphrasing you from earlier, the "authenticate as one user and then
log in as another" use case is the one I'm trying to expand. LDAP is
just the auth method I happen to have present-day customer cases for.
> That this would also require a new connection option / envvar to tell us
> who the user wants to authenticate to LDAP as doesn't exactly make me
> any more thrilled with it.
Forgetting the LDAP part for a moment, do you have another suggestion
for how we can separate the role name from the user name? The
connection string seemed to be the most straightforward.
--Jacob