Re: Direct SSL connection with ALPN and HBA rules

On 29/04/2024 21:43, Jacob Champion wrote:
> On Mon, Apr 29, 2024 at 1:38 AM Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:
>> If you only
>> have v17 servers in your environment, so you know all servers support
>> direct negotiation if they support SSL at all, but a mix of servers with
>> and without SSL, sslnegotiation=directonly would reduce roundtrips with
>> sslmode=prefer.
>
> But if you're in that situation, what does the use of directonly give
> you over `sslnegotiation=direct`? You already know that servers
> support direct, so there's no additional performance penalty from the
> less strict mode.

Well, by that argument we don't need requiredirect/directonly at all.
This goes back to whether it's a security feature or a performance feature.

There is a small benefit with sslmode=prefer if you connect to a server
that doesn't support SSL, though. With sslnegotiation=direct, if the
server rejects the direct SSL connection, the client will reconnect and
try SSL with SSLRequest. The server will respond with 'N', and the
client will proceed without encryption. sslnegotiation=directonly
removes that SSLRequest attempt, eliminating one roundtrip.

>> Making requiredirect to imply sslmode=require, or error out unless you
>> also set sslmode=require, feels like a cavalier way of forcing SSL. We
>> should have a serious discussion on making sslmode=require the default
>> instead. That would be a more direct way of nudging people to use SSL.
>> It would cause a lot of breakage, but it would also be a big improvement
>> to security.
>>
>> Consider how sslnegotiation=requiredirect/directonly would feel, if we
>> made sslmode=require the default. If you explicitly set "sslmode=prefer"
>> or "sslmode=disable", it would be annoying if you would also need to
>> remove "sslnegotiation=requiredirect" from your connection string.
>
> That's similar to how sslrootcert=system already works. To me, it
> feels great, because I don't have to worry about nonsensical
> combinations (with the exception of GSS, which we've touched on
> above). libpq complains loudly if I try to shoot myself in the foot,
> and if I'm using sslrootcert=system then it's a pretty clear signal
> that I care more about security than the temporary inconvenience of
> editing my connection string for one weird server that doesn't use SSL
> for some reason.

Oh I was not aware sslrootcert=system works like that. That's a bit
surprising, none of the other ssl-related settings imply or require that
SSL is actually used. Did we intend to set a precedence for new settings
with that?

(adding Daniel in case he has an opinion)

--
Heikki Linnakangas
Neon (https://neon.tech)