Artifact Binding Re: specs Digest, Vol 36, Issue 1

Nat Sakimura sakimura at gmail.com
Thu Aug 20 02:02:54 UTC 2009


John,

On Thu, Aug 20, 2009 at 10:11 AM, John Bradley <john.bradley at wingaa.com>wrote:

> Nat,
>
> You don't have the new Sec 9.4 hi-lighted like the other changes so I
> missed it.


Sorry, I fixed it.


>
>
> Lets see if I follow this now.
>
> The RP sends a direct request  openid.mode=art_res to the OP (9.4) with AX,
> PAPE parameters and the OP provides a (10.3) Artifact response.
>
> The RP then makes a indirect request with openid.artifact set to the value
> obtained from the artifact response.
>
> The OP then looks up the request based on the artifact.  (conflicting
> parameters are resolved in some way)


Direct request should take precedence.


>
>
> The User authenticates (must happen after the artifact is looked up or you
> need to include PAPE in the indirect request)


MUST happen after the artifact look up as the claimed id is in it.


>
> The user consents to releasing attributes.
>
> The OP returns a positive assertion with openid.artifact in the response
> (you need to include that in 10.1)


Sequence-wise, the OP does not return a positive assertion here but it
returns an Artifact Response (10.3).
Positive assertion is returned later.
I am not sure if openid.artifact is needed in the positive assertion since
it is returned as the response to the assertion request as openid.artifact
as one of the request parameter.



>
>
> The RP then makes another direct request openid.mode=assertion_req
> openid.artifact={artifact}
>
> The OP then returns the assertion with the extension payload.


Yes. It returns the auth and extension payload including claimed identifier
etc.

>
>
> Am I getting close to what you are thinking?


Quite.


>
>
> I don't think we should underestimate the synchronization issues that a OP
> will have across a cluster with this.


I agree. We need to be clever on how artifact should be hashed and stored.
e.g., store (artifact,payload) pair on distributed storage in such a way
that you can locate the location from artifact (e.g., if the Artifact is an
URL, it is trivial).

FYI, in CX, I have made the artifact created by the RP so that RP to OP
request is always with the Artifact, so that a load balancer can land the
request to the same server. Perhaps that's a better way but has a side
effect that the indirect request gets bigger.


>
>
> We would be bending the principals of REST with this.   I would like to get
> the opinion of some of the larger OPs on this.


I am not so sure about it (bending REST).

Think of it like this:

The authentication request over direct communication is stored on the server
side (the OP) as a resource and the OP returns a Resource URI, which is
called "Artifact".

The RP makes a GET request to a resource uri constructed from the OP End
Point URI and the Artifact.

Is this bending REST?


>
>
> The two requests coming from different IP will likely wind up on different
> servers.


Yes, but in principle, it should not be a problem.


>
>
> This on it's own makes data snooping worse not better.


I do not get it. Could you explain more?


>
>
> We need mutual TLS for the direct connection where the OP verifies the
> return_to URI against the cert of the incoming connection.
>
> Unless the claimed_id is only passed in the direct session it probably
> would not meet the no snooping requirement.  I need to consult on that.


The claimed_id is only passed in the direct session. In the Artifact
binding, it is only the Artifact and signature that are sent through
indirect communication.


>
>
> Without mutual TLS the artifact in the indirect response needs to be
> encrypted.


Does it have to be mutual TLS? Of course, it is better that way, but I do
not want to raise the bar too much.


>
> I am traditional and still prefer to encrypt the returned token with the
> cert you get from RP discovery to verify the return_to.


Are you talking about the Artifact or the Data?
For Data, yes, I do, too.
I was also going to encrypt the Artifact, but SAML was not doing that so I
left it unencrypted.


>
> Your proposal is simple in some ways but I don't know that it meets all of
> the potential use cases.


Without the list of use cases, it is difficult to estimate :-)
My primary use case is mobile. I think it satisfies it. For more security
oriented things, you can use something like CX ;-) Of course, that requires
more complexity, namely, the public key discovery, payload digital
signature, payload encryption, etc.


>
>
> If we are going to dig into this I don't know that doing it outside of WG
> IPR coverage.


You are right. Since CX covers artifact in its scope, it might be good
discussing it under CX WG IPR. In parallel, we can spin up Authn 2.1 WG and
move the discussion there.


>
>
> John B.
>
>
> On 19-Aug-09, at 8:10 PM, Nat Sakimura wrote:
>
>  Hi John,
>>
>> Ah! I think it is so called "Framing Problem".
>>
>> When I modified the 2.0 spec to create this 2.1 draft 0.001, I have
>> removed all the restrictions that authentication messages have to be
>> indirect. So, now, it can be direct as well. When using direct, to link it
>> with user action, the artifact is used.
>>
>> See inline:
>>
>> On Thu, Aug 20, 2009 at 1:52 AM, John Bradley <john.bradley at wingaa.com>
>> wrote:
>> Hi Nat,
>>
>> inline
>>
>>
>> On 19-Aug-09, at 12:32 PM, Nat Sakimura wrote:
>>
>> Hi John,
>>
>> Inline:
>>
>> On Wed, Aug 19, 2009 at 10:30 PM, John Bradley <john.bradley at wingaa.com>
>> wrote:
>> Nat,
>>
>> On a first read through.
>>
>> Your proposal only solves half the problem,  in that it only reduces the
>> size of the indirect response.   With extensions it is still possible to
>> likely that requests will go over 2K.
>>
>> Why is that so?
>> All the extensions can use this direct communication path.
>> What was sent over indirect communication is sent over direct
>> communication.
>>
>>
>> If the full request must be made indirectly that doesn't reduce the
>> request size.
>>
>> As stated above, the full request can be made directly. In that case, only
>> artifact and a few others moves indirectly. Thus, it will reduce the request
>> size drastically, putting upper bound for the indirect request size.
>>
>>
>>
>> Are you thinking that the authentication is done via a indirect request
>> but CX, AX etc all happen via direct communications?
>>
>> No. Main body of Authentication request, and thus extension requests, are
>> sent via direct request and only the artifact and a few others are sent via
>> indirect communications.
>>
>>
>>
>> Unless you send the attributes that are going to be requested in the
>> indirect request how would the user provide consent to release them to the
>> RP?
>>
>> The request is sent from the RP to the OP over the direct communication.
>> Then, the user is taken from the RP to the OP over the indirect
>> communication carrying the Artifact. The OP, upon receipt of the Artifact,
>> can reconstruct the main request from it. Then, the user consent etc.
>> happens as usual. Then, instead of the OP sending the positive assertion
>> back to the RP, it sends the Artifact and the user back to the RP over the
>> indirect communication signifying that it has completed the processing at
>> the OP. Using the Artifact, the RP fetches the (+ve or -ve) assertion from
>> the OP through the direct communication. Verification etc. goes on then as
>> usual there after.
>>
>>
>>
>>
>> Also openID relies on validating the users presence via a cookie.  That
>> would not be available to the OP in a direct session.
>>
>> Hopefully now you see why it works. It changes almost nothing. It just
>> pushes the main payload to the direct communication and that's it. Others do
>> not change.
>>
>>
>>
>>
>>
>> I would prefer not to have to revisit this again once the request size
>> becomes an issue.
>>
>> The OP needs to advertise that it supports the binding in it's XRD/S.
>>
>> In this draft, I made the support of direct communication mandatory and
>> the version of the OpenID Authn protocol was raised to 2.1. This is
>> advertising that it supports the binding in its XRD/S.
>>
>>
>> I don't know that making it mandatory is necessarily a good idea.  There
>> may be other things in 2.1 that may be useful aside from a artifact binding.
>>
>> I prefer the idea that a OP could optionally support the binding and it
>> would be discoverable.
>>
>> I don't feel super strong about it, but others may.
>>
>> Right. I just wanted to be a minimalist and also wanted the spec to be
>> fairly symmetric. If the artifact is to be an optional binding, then it
>> would have to define a new type URI. However, from the sake of being
>> symmetric, then, we should define type URI for the indirect binding as well
>> and list it on XRD/S.
>>
>>
>>
>>
>>
>> As you point out this doesn't do anything for security.    The artifact
>> will need to be encrypted or mutual TLS used for the direct connection.
>>
>> The encryption of the Artifact is an open question, as SAML Artifact
>> binding does not encrypt the Artifact either siting that in this limited
>> size that the encryption is unpractical.
>>
>> For the mutual authentication, I could incorporate relevant sections of CX
>> here as well. That will make the already thin CX spec even thinner.
>>
>>
>> You are going to make me read CX aren't you:)
>>
>> Yes. If you leave the contract schema alone, then it is extremely concise.
>>
>>
>>
>>
>> In testing something close to 1% of RP and OP have TLS implemented
>> correctly now.   Mutual TLS may impossible to implement in some
>> environments.
>>
>> It is easy to say just use TLS for that, and make it someone else's
>> problem.   Mutual TLS may be the best option but encrypting the fragment and
>> using normal TLS should also be considered.
>>
>> Having the OP POST to the RP directly should also be considered,  that
>> would work for LoA 2 but probably not LoA 3 without mutual TLS.
>>
>> That's unsolicited direct response, and it is not precluded in this draft.
>>
>>
>> No unsolicited assertions are still indirect.
>>
>> As I have explained above, in this draft, unsolicited assertions can be
>> direct. It may have no association with the user session as well. If it
>> does, then the user has to be at the OP to start with, and the User has to
>> be taken to the RP with the Artifact.
>>
>>
>> I was thinking of a flow where the OP makes two replies one indirect and
>> the other direct.
>>
>> That is exactly what is in my draft, though the indirect one is optional.
>>
>>
>>
>> The main reason not to do this is that it would not work with RP load
>> balancing.   Likely they go to different servers.
>>
>> We have that problem now with nonces.
>>
>> I think this is an implementation problem.
>>
>> The implementation should have some kind of shared storage behind the
>> server farm and store the direct response to with the Artifact as the key.
>> When the user landed on one of the server, the server can pull the data from
>> the shared storage, and that's it.
>>
>> Note that the unsolicited direct response without user being taken back to
>> the RP works. It works for AX update etc.
>>
>> Also, I have constructed the protocol to be "easier to implement" for RPs.
>> If they do not support unsolicited direct response, that is still OK.
>> It is only this feature that uses the OP to the RP communication unlike
>> SAML's case.
>> It is always the RP making request to the OP.
>>
>>
>>
>> Artifact binding is simple in principle but the devil is in the details.
>>
>> Is there anything else?
>> We can create an issue tracker and solve them one by one.
>> I actually do not foresee too many of them.
>> Also, we should not try to solve all the use cases.
>> Being able to satisfy 90% of them should be good enough.
>>
>>
>>
>> John B.
>>
>>
>>
>> There are a number of tradeoffs with different methods.
>>
>> A good attempt to show how this method would work.
>>
>> John B.
>>
>>
>> On 19-Aug-09, at 5:51 AM, Nat Sakimura wrote:
>>
>> Been sick and not following the various discussion around artifact since
>> last Saturday, so I might be out of sync but here is my shot for Artifact
>> Binding which I hoped to provide on Friday 14th.
>>
>> http://wiki.openid.net/OpenIDwithArtifactBinding
>>
>> It is about 40 lines of modification/addition. The portion that I
>> changed/added are in RED so it should be easy for you to find out.
>>
>> Its sequence is a bit different than SAML Artifact binding as I tried to
>> minimize the impact to the current deployments.
>>
>> It has done nothing about encryption. The direct communication should be
>> over the verified TLS channel. Security implication of the Artifact exposure
>> on the indirect communication should be further discussed, but my
>> preliminary assessment is that it should be ok.
>>
>> =nat
>>
>> On Wed, Aug 19, 2009 at 8:42 AM, Dick Hardt <Dick.Hardt at microsoft.com>
>> wrote:
>> my $0.02
>>
>> I expect the data moving between the RP and OP to become even larger over
>> time, therefore a standard, alternative mechanism for moving the data
>> directly between the RP and OP, particularly when bandwidth to the client is
>> constrained, seems desirable.
>>
>> I would generally prefer a proven, widely deployed encryption mechanism
>> such as TLS rather then adding functionality to OpenID
>>
>> -- Dick
>> ________________________________________
>> From: openid-specs-bounces at lists.openid.net [
>> openid-specs-bounces at lists.openid.net] on behalf of John Bradley [
>> john.bradley at wingaa.com]
>> Sent: Tuesday, August 18, 2009 3:27 PM
>> To: Allen Tom
>> Cc: openid-specs at lists.openid.net
>> Subject: Re: Artifact Binding Re: specs Digest, Vol 36, Issue 1
>>
>> One of the things you need for LoA 2 is to prevent eavesdropping.
>>
>> The choices are encrypt the response to the RP or use direct
>> communication with TLS (probably mutual) if the RP is going to make a
>> direct request to the OP.
>>
>> Using an artifact binding has advantages and disadvantages.   Using it
>> to get around the 2K URI limit in IE would put any RP not supporting
>> it at a disadvantage.
>>
>> It might be acceptable if the RP could indicate its support for
>> artifact binding in the request and allow the OP to use artifact
>> instead of post.
>>
>> With mobile devices becoming more common I can see people preferring
>> an artifact binding over the existing ones.
>>
>> It is a real change to the protocol and will add complexity supporting
>> another binding.
>>
>> One short term fix that Andrew Arnott implemented in DotNetOpenAuth is
>> a smart detection of OP's support for AX vs SREG and preferring SREG
>> if it is supported.   Most people are only using AX for the SREG
>> attributes anyway.
>>
>> I agree that the AX attribute URI need to get sorted out anyway.   We
>> could look at making them shorter when we mint new standard ones.
>>
>> John B.
>> On 18-Aug-09, at 6:02 PM, Allen Tom wrote:
>>
>> > Hi All,
>> >
>> > Sorry for the delayed response, I'm still catching up on mail after
>> > being on vacation last week.
>> >
>> > Breno - How would artifact binding help OpenID attain Loa2? I'm
>> > unclear as to how that would make a difference.
>> >
>> > The Yahoo OP was recently updated to return responses that are
>> > larger than 2KB using POST, and this has caused many users to see
>> > the ugly browser warning because most RPs don't support HTTPS.
>> > Displaying the ugly browser warning is really unacceptable, so we'll
>> > probably update the Yahoo OP to only use POST only for HTTPS
>> > return_to URLs.
>> >
>> > The excessively large responses are mostly due to AX being
>> > excessively verbose. It would be really nice if we could revise AX
>> > to be a lot more compact. Perhaps if we had a standardized AX
>> > schema, we'd be able to shorten the message size.
>> >
>> > Allen
>> >
>> >
>> >
>> > Breno de Medeiros wrote:
>> >>
>> >> Since Google was mentioned here as wanting artifact, let me make the
>> >> record clear to say that I spoke about artifact binding on my
>> >> personal
>> >> capacity.
>> >>
>> >> My very own personal view is that an artifact profile would be easy
>> >> to
>> >> spec out (the check_authentication or stateless mode is already the
>> >> artifact flow without the additional benefits of artifact) and would
>> >> make OpenID more robust. Currently long URLs require POST which only
>> >> gives you so much mileage. POST is ugly if the RP has a non-HTTPS
>> >> endpoint, with scary user confirmation dialogs.
>> >>
>> >> Also, I did not wish to express any personal opinion on whether
>> >> OpenID
>> >> should seek Loa2, just to note that artifact is the easiest route
>> >> there.
>> >>
>> >> On Thu, Aug 13, 2009 at 10:45 AM, Nat Sakimura<sakimura at gmail.com>
>> >> wrote:
>> >>
>> >>> John,
>> >>> You changed the topic of this thread.
>> >>> This thread was about artifact binding, not about Government LoA.
>> >>> That's another thread :-)
>> >>> Yes, Artifact helps LoA, but it is not only that.
>> >>> It helps the mobile space immensely.
>> >>> =nat
>> >>>
>> >>> On Fri, Aug 14, 2009 at 2:00 AM, John Bradley <jbradley at mac.com>
>> >>> wrote:
>> >>>
>> >>>> Chris
>> >>>> I think we are agreeing.  OpenID needs to play to it's strengths.
>> >>>> Chasing shiny things is tempting.
>> >>>> We need to carefully consider the impact of changes.
>> >>>> That is not to say that openID shouldn't evolve.
>> >>>> There are always tradeoffs.
>> >>>> Remember that a GSA LoA 2 or 3 profile is focused on the Gov
>> >>>> accepting the
>> >>>> assertions for specific uses.
>> >>>> Other people are free to make there own determinations for other
>> >>>> use
>> >>>> cases.
>> >>>> I am interested in finding out if IdP really want to be certified
>> >>>> at LoA 2
>> >>>> with all of the extra identity
>> >>>> proofing,  liability and other things that go with that.
>> >>>> A LoA 2 certification for a IdP involves a lot more than just
>> >>>> tweaking
>> >>>> some protocol peaces.
>> >>>> Are there OPs  that want that?
>> >>>> John B.
>> >>>> On 13-Aug-09, at 9:11 AM, Chris Messina wrote:
>> >>>>
>> >>>> On Thu, Aug 13, 2009 at 8:34 AM, John Bradley <jbradley at mac.com>
>> >>>> wrote:
>> >>>>
>> >>>>> Some may ask if we add artifact binding, signatures and
>> >>>>> encryption are we
>> >>>>> not reinventing SAML Web SSO, or something of equal complexity?
>> >>>>>
>> >>>> I would like to know more about this, but my instinct is always
>> >>>> to say
>> >>>> "NO" for as long as possible when any new feature will a) introduce
>> >>>> complexity and b) stifle or impair potential adoption.
>> >>>> That we've come as far as we have is a feat; maintaining that
>> >>>> momentum is
>> >>>> critical — and that means making good on the promise of what
>> >>>> OpenID offers
>> >>>> *today* — and only extending it with real world examples where
>> >>>> people are
>> >>>> implementing kludges (en masse) to serve a common need.
>> >>>>
>> >>>> Chris
>> >>>> --
>> >>>> Chris Messina
>> >>>> Open Web Advocate
>> >>>>
>> >>>> Personal: http://factoryjoe.com
>> >>>> Follow me on Twitter: http://twitter.com/chrismessina
>> >>>>
>> >>>> Citizen Agency: http://citizenagency.com
>> >>>> Diso Project: http://diso-project.org
>> >>>>
>> >>>> OpenID Foundation: http://openid.net
>> >>>>
>> >>>>
>> >>>> This email is:   [ ] bloggable    [X] ask first   [ ] private
>> >>>>
>> >>>>
>> >>>> _______________________________________________
>> >>>> specs mailing list
>> >>>> specs at lists.openid.net
>> >>>> http://lists.openid.net/mailman/listinfo/openid-specs
>> >>>>
>> >>>
>> >>> --
>> >>> Nat Sakimura (=nat)
>> >>> http://www.sakimura.org/en/
>> >>>
>> >>> _______________________________________________
>> >>> specs mailing list
>> >>> specs at lists.openid.net
>> >>> http://lists.openid.net/mailman/listinfo/openid-specs
>> >>>
>> >>>
>> >>>
>> >>
>> >>
>> >>
>> >>
>> >
>> > _______________________________________________
>> > specs mailing list
>> > specs at lists.openid.net
>> > http://lists.openid.net/mailman/listinfo/openid-specs
>>
>> _______________________________________________
>> specs mailing list
>> specs at lists.openid.net
>> http://lists.openid.net/mailman/listinfo/openid-specs
>>
>> _______________________________________________
>> specs mailing list
>> specs at lists.openid.net
>> http://lists.openid.net/mailman/listinfo/openid-specs
>>
>>
>>
>> --
>> Nat Sakimura (=nat)
>> http://www.sakimura.org/en/
>>
>>
>>
>>
>> --
>> Nat Sakimura (=nat)
>> http://www.sakimura.org/en/
>>
>>
>>
>>
>> --
>> Nat Sakimura (=nat)
>> http://www.sakimura.org/en/
>>
>
>


-- 
Nat Sakimura (=nat)
http://www.sakimura.org/en/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs/attachments/20090820/e62c847d/attachment-0001.htm>


More information about the specs mailing list