Libpq: Fail connection attempt if server didn't request a password

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Libpq: Fail connection attempt if server didn't request a password

Adrian Vogelsgesang

Dear Postgres community,

Today, I learned that in the Postgres protocol, the Postgres server first has to request a password from the client before libpq will send it.
In particular, this also means: if the server doesn’t request a password but simply accepts the connection, libpq will just establish the connection.


Is there some way to let libpq fail the connection attempt in such a case, i.e. if a password was provided but not requested from the server?
Can I, as the client, force the authentication used during creating a connection to be scram-sha-256?
Or can I at least disable MD5 and plain-text authentication on the client-side (which I both consider unsecure)?

You might ask: Why do I care to enforce the authentication mode on the client side? Two reasons:
1. To identify misconfigured server. In our use case, all servers should require a password. By failing hard and early, we can more easily find misconfigured servers
2. To establish some trust on the client-side that I am actually connected to the correct server. In scram-sha-256, the server also has to authenticate to the client. This would be a nice additional level of security. However, this is worthless if the client doesn’t enforce a scram-sha-256 authentication: A malicious server could simply accept the connection without asking for a password or use the MD5 method instead

Cheers,
Adrian

Reply | Threaded
Open this post in threaded view
|

Re: Libpq: Fail connection attempt if server didn't request a password

Tom Lane-2
Adrian Vogelsgesang <[hidden email]> writes:
> Is there some way to let libpq fail the connection attempt in such a case, i.e. if a password was provided but not requested from the server?

Why would that be a useful thing to do?

> 1. To identify misconfigured server. In our use case, all servers should require a password. By failing hard and early, we can more easily find misconfigured servers

I find that argument quite unconvincing.  You can easily look into
pg_hba.conf to see if any lines don't have an auth method you like.
Moreover, that approach scales to requirements that are more complicated
than this specific use-case, and you can be sure that *all* the entries
satisfy your business rule, not just the one you happened to match this
time.

> Can I, as the client, force the authentication used during creating a connection to be scram-sha-256?

There has been discussion of that, but no one has written a patch yet.

                        regards, tom lane