Re: Make SSPI documentation clearer

Greetings,

* Tomas Pospisek (tpo2(at)sourcepole(dot)ch) wrote:
> what is the status of bringing the proposed SSPI documentation chages into
> the git repo?

> - is there anything to be improved or missing with the proposed changes?

... maybe, see below.

> - does the improvement need a hat tip of someone?

Needs a committer willing to include the change. I'm generally alright
with being that committer once we get to some finalized wording unless
there's comments from others.

> - anybody here that has the guts, time and willingness to take the
> supposed improvement and apply put it into the git repo?

This isn't helpful and frankly is detrimental to getting this change
included. We strive strongly to have a positive tone and focus on
technical excellence.

> Anything else needed here?

See below for comments on the change.

> If the change is in principle OK, then I can prepare a patch or such (I
> assume postgres doesn't take pull requests from https://github.com/postgres
> ?)?

If you'd like to work on a patch, that'd be great. The process would be
to make the change in your local git repo, create a patch with the change
(generally with git format-patch) and then post an email to the
pgsql-hackers mailing list with the proposed patch and finally register
it into the open commitfest (which is the app that we use for managing
the many, many requested changes to PG):

https://commitfest.postgresql.org

> On 13.03.23 10:00, Tomas Pospisek wrote:
> > On 13.03.23 01:36, Stephen Frost wrote:
> >
> > > * PG Doc comments form (noreply(at)postgresql(dot)org) wrote:
> > > > Page: https://www.postgresql.org/docs/15/sspi-auth.html
> > > > Description:
> > > >
> > > > The [current SSPI
> > > > documentation](https://www.postgresql.org/docs/current/sspi-auth.html)
> > > > reads:
> > > >
> > > > "SSPI authentication only works when both server and client are
> > > > running Windows, or, on non-Windows platforms, when GSSAPI is
> > > > available."
> > > >
> > > > I interpret that phrase like this:
> > > >
> > > > * there's a case where both server and client are running Windows
> > > > * there's a case where both are running non-Windows
> > >
> > > Yeah, that phrasing isn't great.
> > >
> > > > What about mixed cases? When the client is non-Windows, then can it
> > > > use SSPI? No, AFAIK not. So I'd suggest to make that phrase above
> > > > clearer and completely explicit:
> > >
> > > SSPI is Windows-specific, yeah.
> > >
> > > > "SSPI authentication works when both server and client are running
> > > > Windows.
> > > >
> > > > When the server is on a non-Windows platform then the server must
> > > > use GSSAPI if it wants to authenticate the client either via
> > > > Kerberos or via Active Directory. A client on a Windows platform
> > > > that connects to a non-Windows Postgresql server can either use SSPI
> > > > (strongly encouraged) or GSS (much more difficult to set up) if it
> > > > wants to authenticate via Kerberos or Active Directory. A client
> > > > from a non-Windows platform must use GSS if it wants to authenticate
> > > > via Kerberos or Active Directory."
> > >
> > > Rather than work in negative, I feel like it might make more sense to
> > > work in positives?  That is, perhaps this instead:
> > >
> > > On Windows platforms, SSPI is the default and most commonly used
> > > mechanism.  Note that an SSPI client can authenticate to a server
> > > which is using either SSPI or GSSAPI, and a GSSAPI client can
> > > authenticate to a server which is using either SSPI or GSSAPI.
> > > Generally speaking, clients and servers on Windows are recommended to
> > > use SSPI while clients and servers on Unix (non-Windows) platforms use
> > > GSSAPI.
> > >
> > > Stricltly speaking, this is all independent of if AD is being used as
> > > the KDC or not.
> >
> > I agree, that's a better formulation. I'd suggest to improve your
> > version in three ways:
> >
> > 1. replace "mechanism" with "authentication mechanism"

Sure.

> > 2. be explicit about Active Directory so there's no doubt wrt to setting
> >    up authentication

Not against the idea of mentioning AD...

> > 3. be explicit that GSSAPI should be used on non-Windows platform
> >    servers when one wants clients in an AD domain to seamlessly
> >    authenticate with the non-Windows server. I'd mention that because if
> >    the windows clients are *not* in an AD domain then they will *not* be
> >    able to authenticate to the non-Windows server with GSSAPI.

However, Windows systems not in an AD domain can certainly use GSSAPI to
authenticate with the appropriate libraries installed and Kerberos
configured on them.

> > So finally the whole start of the SSPI paragraph in the docu would look
> > like this:
> > ----------------------
> >
> > 21.7. SSPI Authentication
> >
> > On Windows platforms, SSPI is the default and most commonly used
> > authentication mechanism.  Note that an SSPI client can authenticate to
> > a server which is using either SSPI or GSSAPI, and a GSSAPI client can
> > authenticate to a server which is using either SSPI or GSSAPI.
> > Generally speaking, clients and servers on Windows are recommended to
> > use SSPI while clients and servers on Unix (non-Windows) platforms are
> > recommended to use GSSAPI if they want to interoperate seamlessly with
> > Active Directory or Kerberos authentication.
> >
> > When using Kerberos authentication, SSPI works the same way GSSAPI does;
> > see Section 21.6 for details.
> >
> > ----------------------

These changes generally seem alright to me and they don't seem to
include an invalid statement such as what you were proposing with #3
above.

> > If the docu is changed in this way, then the phrase "PostgreSQL will use
> > SSPI in negotiate mode" is dropped wrt to the previous documentation. I
> > have not been able to find out what "SSPI in negotion mode" is and
> > therefore if it's in any way relevant to mention that in the docs.

I'm not sure that it's necessary to delve into Negotiate mode either in
this documentation. For those curious, it's essentially another
wrapping of Kerberos, akin to SASL, to allow for other authentication
methods to be used instead of Kerberos (think NTLM). It's been a while
since I delved into it, but my last recollection was that using
Negotiate for any of the other valid options (which was mainly just
NTLM, again, as I recall) other than for Kerberos is basically
deprecated and strongly discouraged.

Thanks,

Stephen