Re: SCRAM with channel binding downgrade attack - Mailing list pgsql-hackers

On Wed, May 23, 2018 at 08:59:43AM +0200, Magnus Hagander wrote:
> On Wed, May 23, 2018 at 8:46 AM, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
>> "tls-unique" and "tls-server-end-point" are overly technical to users.
>> They don't care which one is used, there's no difference in security.
>> Furthermore, if we add another channel binding type in the future, perhaps
>> because someone finds a vulnerability in those types and a new one is added
>> to address it, users would surely like to be automatically switched over to
>> the new binding type. If they have "tls-unique" hardcoded in connection
>> strings, that won't happen.
>>
>> So I think the options should be "prefer", "disable", and "require".
>> That's also symmetrical with sslmode, which is nice.

OK, being able to introduce a new default if necessary is a good point.
Let's introduce a "require" mode then, which uses tls-unique
underground, while "tls-unique" and "tls-server-end-point" are
documented as developer-oriented. 

> As a general point, I still don't like being symmetrical with
> sslmode=prefer, because that's a silly setting for sslmode :) It basically
> says "I don't care about the security guarantees, I just want the overhead
> please". Now, granted, the over head for SCRAM channel-binding is certainly
> a lot less than it is for TLS. But if we just want to go with symmetry, it
> would perhaps make more sense to implement the "allow" mode which makes
> more sense on the TLS side as well.

Something that you may be forgetting here is that we still want to be
able to connect to a v10 backend with default options even with a
post-v11 libpq.  Or we will get complaints.

>> It would be nice to have a libpq setting like
>> "authenticate_server=require", which would mean "I want man-in-the-middle
>> protection".
>
> That seems like a bad name for such a thing. Shouldn't it be
> "authenticate_server=no_man_in_the_middle" (not those words but you get the
> idea). Specifically, protecting against man in the middle attack does not
> equal authenticating server -- you can still be connected to the wrong
> server just with no second attacker between you and the first
> attacker.

Still that's not something we want for v11, right?

>> With that, a connection would be allowed, if either the server's SSL
>> certificate is verified as with "sslmode=verify-full", *or* SCRAM
>> authentication with channel binding was used. Or perhaps cram it into
>> sslmode, "sslmode=verify-full-or-scram-channel-binding", just with a
>> nicer name. (We can do that after v11 though, I think.)
>
> sslmode=verify-full is very different from SCRAM with channel binding,
> isn't it? As in, SCRAM with channel binding at no point proves which server
> you're talking to -- only that you are talking to the SSL endpoint? It
> could be a rogue SSL endpoint unless you do certificate validation.

And the reverse is true as well, say the CA is compromised..