[OpenID] About Facebook, MySpace and OpenID

Sun Apr 5 14:59:04 UTC 2009

Out of interest, what happens the second time (on the same OP session)?

In the SAML websso protocol, as an RP I'm used to requesting an "ispassive-constrained" assertion (with fixed  attribute set) when folks on the RP site navigate to the internal home pages of certain site modules within the realm - where the authZ policy enforced by that particular module's class loader requires a recent act of user (re)authentication (using an RSA securid time-synced tokencode, typically) and reconfirmation that the member has an attribute indicating s/he is (still) in good standing with the IDP's membership policies. As our basic RP session is 60m or longer , the guards are necessary - to ensure for timeliness of user auth required by higher accountability and control requirements in those site areas.

Our OP/RP openid implementation happens to be a multi-stage gateway -  SAML and openid auth protocol engines operating in a pretty common co-resident fashion in a single process at a security domain boundary. If the SAML RP issues a authentication requests, the "fixed" attribute contract in the bilateral SAML metadata becomes a list of openid auth req "required" sreg/ax attributes. If the SAML RP makes a followup request 5m later, with ispassive=true (no UI allowed), the same set of attributes will be required. There is no context for the second protocol run, given the first - that enable it to ask for more or less attributes the second time.

Google OP seems to be making assumptions about how RPs are design and built - which is biasing how openid auth protocol operates in their case. Against conventional standards wisdom, intentionally or otherwise, their interpretation of what openid auth/sreg/ax conformance requires seems to be is forcing a particular security policy, persistence and session implementation architecture on RPs  - to know to use a particular code path when working with Google OP  concerning persistence of RP-session attributes.

Google is always a fascinating (and very POSTIIVE) case study for me. Only two years ago, the real estate bit of Google refused to contemplate ever being an IDP sharing such sensitive commercial consumer data as the Google account holders attributes with RPs like us  (they were obviously wrong, given the Google OP!), and we do already have the means for individual realtors acting in reverse handoff to assert their rights (as a member of the home listing service) to post a home listing to Google Base - for distribution of some of the listing attributes into that particular search engine -  in the hope it will generate a sales lead.

 I have hopes that I can put the two functions (consumers using openid websso to followup a lead listing in Google to get to us as RP, and realtors using our IDP/OP to posting off listing "lead" attributes to Google Base) together now, to create an web application which would migrate away from Google Base's proprietary auth mechanism to websso standards. What I cannot do, however, is be required to know of or use any particular implementation architecture, to accomplish that - as the same Rapattoni code base has to work according to the same technical standard with all the other (one day, similarly openid-powered) lead generation sites, too.

From: John Bradley [mailto:john.bradley at wingaa.com]
Sent: Saturday, April 04, 2009 9:12 PM
To: Peter Williams
Cc: Breno de Medeiros; Deron Meranda; general
Subject: Re: [OpenID] About Facebook, MySpace and OpenID

Peter,

If you ask google for a required AX attribute, email as an example and the user declines to send it to you Google sends back a negative assertion and you never find out the users claimed_id.

In my tests of Google a user has the option of continuing with the login and returning the required AX attributes or canceling and not logging in.

If the user cancels the RP would need to send a second checkid_setup without the required AX attribute request.  Log the user in and then perform some additional separate openID step to get the AX attributes.

For a RP to deal properly with a Google openID they should separate the Auth from the AX exchange otherwise if the the user declines sending the attributes to the RP they also decline sending the authentication.

Technically they are not violating the AX spec but to be useful the flow needs to be different from what RPs are doing with other OPs.

That is why I am hinting at the need for some higher-level definition/standard for the UX flows.

Regards
John Bradley

On 4-Apr-09, at 8:29 PM, Peter Williams wrote:

If I'm an RP taking authenticated comments from a google subscriber for the first time and the user denies the release of the optional email attribute, is there any way for the user to release it the next time s/he attempts to post a comment?

i.e. depending on the nature of the  comment or the thread, the user may wish  to release or not release the email identifier (to signal that followup email dialog is available, or to receive events that the comments received followup). Lets assume that the site processing the comments does not maintain an account for the commenting user.

From: general-bounces at openid.net<mailto:general-bounces at openid.net> [mailto:general-bounces at openid.net] On Behalf Of Breno de Medeiros
Sent: Saturday, April 04, 2009 6:38 PM
To: Deron Meranda
Cc: general; John Bradley
Subject: Re: [OpenID] About Facebook, MySpace and OpenID

It is not true that you need to request all your attributes for the Google OP or else you will be locked out. If you add a request for a new attribute we will prompt the user to authorize and then send it.  That also happens if the value of the attribute changed since the user approved (re-prompt and send).
On Apr 3, 2009 9:35 PM, "Deron Meranda" <deron.meranda at gmail.com<mailto:deron.meranda at gmail.com>> wrote:

On Fri, Apr 3, 2009 at 11:08 PM, Andrew Arnott <andrewarnott at gmail.com<mailto:andrewarnott at gmail.com>> wrote:
> ... All the RPs are upgrading their email requests to required

> so that they work with Google.  Apparently they really wanted email and they > were getting it unt...
Maybe its useful to think about attributes other than email to see
where "optional" perhaps makes more sense.

Consider a possible Time Zone attribute (yes SReg has this).  This is a case
where it is easier to see what an RP might mean by "optional" -- that if the
OP can provide it it wants it; but if not, the RP still wants the authentication
to proceed and not to fail, because it can deal easily enough with not
getting the time zone.

It could also be argued that the OP *should* ask the user which optional
attributes it wants to give.  You might not think time zone is important
enough to ask the user for permission; unless perhaps your time zone
happened to be America/Havana for example and you didn't want this
RP to know you're in Cuba.

I can see Google's perspective on this; and it is a combination of
trying to make a simple UI for non-technical users (which we all
agree is something OpenID desperately needs), as well as the
spec being somewhat silent on what its expectations are for the
OP and RP.

However, at this point, with Google's current interpretation, there is
effectively no utility in having optional attributes.  The distinction is
made meaningless.  And since we don't want RP's to special-case per
OP; then effectively, in practice today anyway, it is pointless for
OpenID AX to pretend there is an optional/required dimension.

Now one can argue that Google is right and the spec was trying to
make two categories when only one made sense; or that Google is
wrong and OPs should definitely treat optional attributes differently
than required ones.  I'm not sure which is correct; but I do think
that the spec should be made very clear one way or the other.
Because if the behavior is left up to the OP, as Google has done, then
the spec is made pointless in practice and its all just unnecessary
noise words.

I would say, another thing that Google does that may play into all
this is that they don't always send AX attributes back at all.  If the
RP and OP have communicated before concerning a certain
identity; then the RP may actually get no attributes whatsoever
on subsequent interactions (Google assumes that the RP will
remember these attributes the first time, which means that in
practice the RP will be forced to remember these attributes).

This also coerces the RP to try to request ALL the attributes it thinks
it might ever need the very first time it interacts with Google.  And since
it also using is directed identity (where the RP doesn't know the identity
before hand), this effectively means that the RP is going to have to
request all the possible AX attributes it might ever desire for any user,
effectively as a *requred* attribute, on every single request!  Because if
it guesses wrong and decides not to ask for a particular attribute even
once, it may then be locked out of ever getting that attribute in the future.

This makes it very hard to implement an RP.  Either it asks for too
many attributes, and the authentication fails because the user doesn''t
want to return ALL of them (or Google doesn't support a hypothetical
Time Zone AX attribute and treats optional as-if required so it fails).
--  Or on the other hand the RP doesn't ask for enough attributes; but
then has to live with never being able to ask for them in the future when
it later decides it does want to know them.

And the directed identity on top of this means the RP has to make this
difficult choice just once per OP, rather than a finer grained per identity.
I think this existing behavior, although defensible from Google's
perspective; pretty much renders a good portion of the AX spec
completely useless.

Either an RP is never going to use AX at all because it doesn't want
to risk causing an authentication to fail; or an RP is always going to
request as many attributes as possible all as required (even if it
doesn't need them right now), because it doesn't want to miss out on
it's one chance to get that information the first time.
--
Deron Meranda

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-general/attachments/20090405/693f9e20/attachment-0002.htm>