Re: SCRAM authentication, take three

On Fri, Apr 7, 2017 at 6:32 PM, Michael Paquier
<michael(dot)paquier(at)gmail(dot)com> wrote:
> On Sat, Apr 8, 2017 at 1:59 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>> On Fri, Apr 7, 2017 at 3:59 AM, Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:
>>> I think the "SCRAM" part is more important than "SHA-256", so -1 on that.
>>
>> I agree. The point here isn't that we're using a better hashing
>> method, even if a lot of people *think* that's the point. The point
>> is we're using a modern algorithm that has nice properties like "you
>> can't impersonate the client by steeling the verifier, or even by
>> snooping the exchange".
>>
>> But "sasl" might be even better.
>
> FWIW, my opinion has not changed much on the matter, I would still
> favor "sasl" as the keyword used in pg_hba.conf. What has changed in
> my mind though is that defining no mechanisms with an additional
> option mean that all possible choices are sent to the client. But if
> you define a list of mechanisms, then we'll just send back to the
> client the specified list as a possible choice of exchange mechanism:
> host all all blah.com sasl mechanism=scram-sha-256-plus
> Here for example the user would not be allowed to use SCRAM-SHA-256,
> just SCRAM with channel binding.
>
> Such an option makes sense once we add support for one more mechanism
> in SASL, like channel binding, but that's by far a generic approach
> that can serve us for years to come, and by admitting that nothing
> listed means all possible options we don't need any immediate action.

Yes, that all seems quite sensible. What exactly is the counter-argument?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company