Re: Proposal: An anti-phishing compromise
Recordon, David wrote: I think we all agree that talking about the method used is far more useful, though with this proposal we're really trying to balance it with simplicity in the authentication protocol itself. Maybe it is better to phrase the discussion around if the user provided a secret (password) to the OP or not to authenticate versus talking about phishing directly.?. If you were to define a URI for common authentication method values, could this value not be returned, simply, in the protocol? I'd hope that this parameter in the auth spec really helps bring the discussion around to the Assertion Quality Extension and that it be implemented to provide the more robust metadata such as what type of authentication was actually preformed. Agreed. If AQE is not mandated, however, I think that the original idea of allowing the OP to make a statement about the authentication method in the actual protocol is still a good one. If you start with a simple URI, returned directly in the protocol, can this not be linked to the equivalent statement in the AQE specification? - John --David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of john kemp Sent: Thursday, February 01, 2007 7:13 PM To: Granqvist, Hans Cc: OpenID specs list Subject: Re: Proposal: An anti-phishing compromise Granqvist, Hans wrote: Proposed Change === Add a single, required, boolean field to the authentication response that specifies whether or not the method the OP used to authenticate the user is phishable. The specification will have to provide guidelines on what properties an authentication mechanism needs to have in order to be non-phishable. The field is just meant to indicate that the authentication mechanism that was used is not a standard secret entered into a Web form. The receiver should decide what is 'non-phishable', not the sender, so it would be better if the OP just states what mechanism was used, perhaps. Agreed. A statement as to the phishability or not seems to be too vague to be useful. A simple statement of the authentication method used would seem to give the same effect, while providing potentially useful information (assuming the RP trusts statements from the OP at all.) Benefits ... Here's what I think: Since the association step is hardly ever used, it can much more easily be overloaded with extra information without breaking any implementation. If the OP would *require* an OpenID association step from the RP, then this step can include an exchange of meta-information between OP and RP. How does the association step between OP and RP change the relationship between the OP and the user (agent)? I thought the attack we were considering is when a rogue RP directs the user agent to a rogue OP, who then steals the user's credentials? How is that affected by the rogue RP and rogue OP associating? - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: Proposal: An anti-phishing compromise
Hi Josh, In addition to the protocol parameter that you have proposed, I'd hope that we can add something like what you wrote below as part of the security considerations section of the OpenID 2.0 Auth specification, as this text seems to capture quite succinctly the issues that RPs and OPs should be thinking about when attempting to deal with phishing: Josh Hoyt wrote: The ways that OpenID can potentially make phishing worse: * Redirects to your provider are a fundamental part of the flow of OpenID, so being redirected to a phishing site is easy to miss * Every relying party (necessarily) needs to know who the provider is in order to verify the authentication response. This means that the site knows what UI it needs to use to phish (and even worse, it can just proxy the user to the provider) I think these two issues cover what makes phishing potentially a greater threat when using OpenID. Although these problems are significant, if a user can authenticate to their OpenID provider through an non-phishable mechanism, OpenID can make the phishing problem *less* of a threat, because there are fewer places that will need to ask for credentials. Other relevant issues: * There is no universally deployed solution to the phishing problem * There is not even a universally *accepted* solution to the phishing problem * Any technology that prevents phishing will require user-agent support or else will fundamentally change the flow of OpenID (prevent relying-party-initiated sign-in) * OpenID is intended to be deployed without requiring specific technologies to be present in the user-agent It might also be helpful to add somewhere a specific definition of phishing, and the associated attack - that an OP can steal a user's credentials if they are passed to the OP. Mitigation can only really be performed by applying client-side changes that ensure that long-lived private information shared only between the OP and the user (such as a password) does not pass across the network. Regards, - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: Proposal: An anti-phishing compromise
Johnny Bufu wrote: On 2-Feb-07, at 7:05 AM, George Fletcher wrote: but I'm still not sure how this helps with the phishing problem. As you pointed out John, the issue is a rogue RP redirecting to a rogue OP. So the rogue OP just steals the credentials and returns whatever it wants. In this case, the rogue RP is not interested at in the the auth response from the rogue OP (or for that matter from the legitimate OP); just in stealing the user's credentials. The phishing field prevents the phisher to later use these credentials on a legitimate RP (which will be contacting the legitimate OP) to impersonate the user -- if the RP enforces phishable = no. I guess I'm confused by the above. If the OP has stolen the user's credentials, it can just say phishable = no and pass its assertion regarding those credentials to the RP. This is about a rogue OP, and the relationship between the OP and the user, not really about the relationship between the OP and RP (although if the RP knew whether or not it could trust the OP by some out-of-band means, it would simply not accept an assertion from the OP unless that trust was established). You might use a rogue RP to start the attack, but the attack is actually performed by the rogue OP, not the rogue RP. - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: Proposal: An anti-phishing compromise
Josh Hoyt wrote: On 2/2/07, john kemp [EMAIL PROTECTED] wrote: Don't get me wrong - I think it's a good idea for the OP to make a statement about the authentication method used (although I would prefer it to say something like authn_method=urn:openid:2.0:aqe:method:password, rather than phishable=yes). That points to AQE, as David mentioned already. A browser plug-in, like sxipper, that uses a username and (a generated, non-user-visible) password internally and will only submit it to the correct OP can't be phished. Is this a different kind of authentication than password? I don't think so. Is it phishable? I think that the OP can reasonably say that it is not. Therefore, I think that the authentication mechanism is (or at least can be) independent from whether the authentication channel is phishable. I will agree that the authentication channel might be separated from the authentication method, as you have described those concepts. I'm not sure if that's a meaningful distinction. For example - in Sxipper, does the password get moved across the network to the OP, or does Sxipper act as the OP (on the client side?) If the former, then I'd argue that Sxipper password auth is slightly less phishable, but not completely so. If the latter, then the trust is /really/ only between the RP and the user. - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: OpenID Auth 2.0 and user-agent neutrality (or, OpenID with REST/SOAP)
Dick Hardt wrote: On 16-Nov-06, at 11:41 PM, Matt Pelletier wrote: On Nov 17, 2006, at 1:24 AM, Dick Hardt wrote: Hi John So that a message can be more then 2K of data. Is it possible to update the language so 1) we don't deprecate HTTP redirects and 2) the form redirect method is described as the preferred/recommended approach, but is not required? You could even stress that HTTP redirects should only be used when the HTTP client's capabilities/limitations would adversely affect the application behavior (or user experience, whatever language is more appropriate for the spec) if form redirects were attempted. I agree with John on the basis that not all HTTP clients will have JS functionality, whether disabled or unavailable, and whether we can imagine what those clients would be or not (blackberry, mobile phone, iPod, Nike running shoe, car keys). The choice to deprecate HTTP redirects involves some assumptions that seem too broad: 1) Payloads will be 2K often enough that there is little value in supporting more than one way to redirect Supporting payloads larger then 2K is a requirement. I guess I don't understand what this 2K limit is (and this is not mentioned in the spec) - are you talking about limits on the URL size when doing an HTTP GET? If so, why not use POST instead? I foresee this to be fairly common in the future as digitally signed claims get moved around. 2) JS will be available to automate the user experience, or at least that automating the user experience isn't that important. JS is widely available now, and if not, the user needs to click a button, so things still work. In an AJAX world, this is pretty much a given now. As you note below, JS/AJAX-capable browsers may be present on high-end phones, but there aren't many of those in comparison to lower-end phones. 3) Deprecating functionality already built into the key spec (HTTP), that is already in use in OpenID 1.x, is preferred to supporting it, even though form redirects will involve more moving parts and specs (ECMA / JS) to maintain the same basic user experience. Moving to one way to do things instead of two is desirable. If POST turns out to be a real issue in the real world, then we can revisit. Libraries are supporting both. It also allows a good separation from URL parameters private to the RP and request parameters intended for the OP. What do you think? How would you suggest the servers determine wether to use GET or POST? Can't YADIS be used to obtain such metadata from the IdP? - John Dick, do you have a list of the browsers you tested against? We tested on a TREO and high end Nokia and Sony Ericson. The types of phones that people could do HTML browsing with. WAP is out of scope. ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: OpenID Auth 2.0 and user-agent neutrality (or, OpenID with REST/SOAP)
Dick Hardt wrote: Supporting payloads larger then 2K is a requirement. I guess I don't understand what this 2K limit is (and this is not mentioned in the spec) - are you talking about limits on the URL size when doing an HTTP GET? yes If so, why not use POST instead? Now I am really confused about what you are talking about I /think/ the limit you are talking about is that regarding the size of the URL. The reason you might approach or exceed that limit would be if you were sending an HTTP GET with parameters appended to the URL. The solution to that issue is to encode the data as an HTTP FORM POST, which, AFAIK has no such limit. As I understand it, that would be a separate issue than whether the protocol is transacted via HTTP 3XX redirects through the user-agent, vs. making the user-agent do the redirect manually. But as I mentioned, I have to admit I'm not totally sure what 2K limit you're actually talking about. How would you suggest the servers determine wether to use GET or POST? Can't YADIS be used to obtain such metadata from the IdP? That does not make any sense. You are stating that the user agent might not be able to support JS. Not supporting JS would mean the RP / OP would need to figure out what the user agent supported so that it could decide on using GET or POST. I'm expecting that the servers can use existing means (determining the browser type and version through the user-agent string) if they want to try that. But what I meant was that the servers (RP and IdP) have to agree on whether they will use one thing or the other, no? The RP could just try it out I guess, but it could also look in the IdP's YADIS doc. and find out whether the authentication service at the IdP supports GET, POST or both. - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: OpenID Auth 2.0 and user-agent neutrality (or, OpenID with REST/SOAP)
Hi, Sorry I'm just reading this, but I just wanted to put in a point very much in favour of NOT deprecating support for HTTP redirects in OpenID 2.0. I'll note that requiring the user to press a 'submit' button to push seems like a dodgy UI strategy. So then you require JavaScript to produce a reasonable user experience. Well, as a representative from the mobile community, I'll tell you that there are quite a few browsers out there (on deployed mobile phones) that still don't support JS in any useful way! So with OpenID 2.0, you may now be requiring many users to click a form submit. Regards, - John Johannes Ernst wrote: Well, as I've said before, I strongly believe that tying authentication to one particular HTTP verb is a bad idea, and unnecessary. I also believe that involving JavaScript in what is fundamentally an HTTP-level kind of protocol is a hack. It very likely is either unnecessary or points to a flaw in the conceptual model of the protocol, or both. The same may be true about having different protocols for thin-client and rich-client. Having said that, I am not making this point more strongly than I have because I don't think these issues are fatal and I don't want to raise more issues that delay OpenID 2.0 auth further. So I will log this as a bug against auth 2.0 as soon as it is published (and as soon as there is a place where to file bugs against the spec ;-)) but will bite my tongue in the meantime. On Nov 12, 2006, at 20:29, Dick Hardt wrote: On 12-Nov-06, at 8:19 PM, Adam Nelson wrote: Hi Dick: I think REST support is a really useful feature, and have described how that might happen in the past, but right now we are pretty focussed on getting browser based auth finalized, and I think the mechanisms for rich clients will be related, but slightly different. That all makes sense, thanks. Is that to say that rich client support isn't a goal of v2.0 of the spec, or just a goal subsequent to the conclusion of browser-based auth? Not a goal of OpenID Authentication 2.0 I think it would make sense to make it a separate document, and would value your involvement! -- Dick ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs Johannes Ernst NetMesh Inc. http://netmesh.info/jernst ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: Authentication Authority (was RE: IdP vs OP (WAS: RE: Editors Conference Call))
Hi Drummond, If what we're trying to express is merely that an OpenID can provide an authentication assertion, then I agree that authentication authority is quite appropriate. I would note that in SAML at least (as I understand it - correct me if I'm wrong Eve!), an authentication authority is not (in that role at least) being requested to actually authenticate the user (ie. to actually perform the authentication at that moment) - the request is only asking whether the authority can make an authentication assertion (ie. it's a query for authentication assertions, rather than an authentication request - which may have already been fulfilled). I don't know if that rather subtle difference is of any interest in OpenID? - John Drummond Reed wrote: Eve, Welcome, and thanks for delurking ;-) I'm fascinated by your suggestion that the SAML vocabulary includes the term authentication authority. I'd vote for the OpenID Authentication 2.0 specification (and the community at large) to adopt that term in a heartbeat because: a) I've many times thought that authentication authority was PRECISELY the role that the IdP/OP played in OpenID Authentication. b) I'm all for consistency with the SAML glossary because I know it was intended to be specification-neutral and I'm a big supporter of harmonizing vocabularies in a problem space (that's why we spent so long on the XRI glossary in the identifier problem space -- see appendix C of http://www.oasis-open.org/committees/download.php/15377). c) It allows us to step around all the semantic issues around whether an OpenID IdP is really providing an identity or not (and also whether OpenID is using classic identity federation or not.) =Drummond -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Eve L. Maler Sent: Tuesday, November 07, 2006 8:16 AM To: specs@openid.net Subject: Re: IdP vs OP (WAS: RE: Editors Conference Call) Delurking for the first time on this list: :-) Drummond and I are on the same page about many things, but John is right that SAML is agnostic as to the strength/significance of the service being provided and so the two cases are much more similar than different. On balance I prefer identity provider because it's intuitive in an English sense, it's used in several technology contexts (not just SAML and OpenID), and it avoids a terminological branding that would otherwise seem to suggest a conceptual divergence that doesn't -- to my mind -- exist. (By the way, there's another term SAML defines that seems to fit the bill of what Drummond is going for here: authentication authority. This is not quite synonymous with identity provider in SAML-land, but it's close -- much the way that relying party and service provider are often close to the same thing. I'm not seriously advocating using it -- just noting that the same software component in an actual deployment can be seen in various lights and have multiple names (roles!).) FWIW, Eve John Kemp wrote: Hi Drummond, Drummond Reed wrote: So why, indeed, is there so much interest in OpenID? I believe it's because of the trust model. To the best of my knowledge, it is radically different than the trust model assumed by the majority of use cases which led to SAML and the Liberty Alliance specs. As Eve Maler of Sun puts it, OpenID supports promiscuous federation -- RPs and OPs that don't know anything at all about each other. At http://www.openidp.org you'll find a promiscuous SAML IdP. While I agree with you that OpenID has been focused on this use-case, with an eye to the use-cases satisfied by SAML, I'd say that SAML has been developed with federated use-cases, but also with an eye to promiscuity. But to put it another way, the trust model used with SAML is out-of-scope for development of the SSO protocol itself. Just like it is for OpenID. And it doesn't stop there. OpenID also supports OPs that ***have zero control over the user's OpenID identifier***. The OP simply provides a service for authenticating that a user has control of the OpenID identifier about which the OP is being queried. And how does one authenticate that the user has control over an identifier? Is it not by having the OpenID IdP having some secret shared with the user - maybe a password, say? A SAML IdP also authenticates that an identifier (issued by the IdP in the SAML case) is bound to a particular user. This is a big deal. In fact, the closer you get to it, the bigger it is. As a result, even though an OP seems to fit the SAML definition of an IdP -- and many technical folks will be very comfortable treating the two as synonymous -- getting the semantics right to stress who really is in control of the identity ***right down to the identifier*** is very important. I don't think we need to worry about fitting the SAML glossary definition of an IdP, but rather
Re: IdP vs OP (WAS: RE: Editors Conference Call)
Hi Drummond, Drummond Reed wrote: So why, indeed, is there so much interest in OpenID? I believe it's because of the trust model. To the best of my knowledge, it is radically different than the trust model assumed by the majority of use cases which led to SAML and the Liberty Alliance specs. As Eve Maler of Sun puts it, OpenID supports promiscuous federation -- RPs and OPs that don't know anything at all about each other. At http://www.openidp.org you'll find a promiscuous SAML IdP. While I agree with you that OpenID has been focused on this use-case, with an eye to the use-cases satisfied by SAML, I'd say that SAML has been developed with federated use-cases, but also with an eye to promiscuity. But to put it another way, the trust model used with SAML is out-of-scope for development of the SSO protocol itself. Just like it is for OpenID. And it doesn't stop there. OpenID also supports OPs that ***have zero control over the user's OpenID identifier***. The OP simply provides a service for authenticating that a user has control of the OpenID identifier about which the OP is being queried. And how does one authenticate that the user has control over an identifier? Is it not by having the OpenID IdP having some secret shared with the user - maybe a password, say? A SAML IdP also authenticates that an identifier (issued by the IdP in the SAML case) is bound to a particular user. This is a big deal. In fact, the closer you get to it, the bigger it is. As a result, even though an OP seems to fit the SAML definition of an IdP -- and many technical folks will be very comfortable treating the two as synonymous -- getting the semantics right to stress who really is in control of the identity ***right down to the identifier*** is very important. I don't think we need to worry about fitting the SAML glossary definition of an IdP, but rather we should focus on making an OpenID glossary definition that makes sense for what OpenID is doing. Whatsmore, I don't think this should or will drive SAML and OpenID further apart. In factit could actually help pave the path to convergence: an OP can be defined as being a SAML IdP that provides identifier authentication services using the OpenID protocol, which may end out (3.0?) becoming a very specific set of SAML capabilities. As noted earlier, I think a SAML IdP also provides identifier authentication. I don't worry so much about convergence of these technologies (although that would be nice ;), but more about giving a converged message to users, developers, and purchasers of these technologies. Regards, - John =Drummond -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Recordon, David Sent: Monday, November 06, 2006 11:46 AM To: Dick Hardt; John Kemp; Patrick Harding Cc: specs@openid.net Subject: IdP vs OP (WAS: RE: Editors Conference Call) I see both sides of this discussion. I think John is correct that the role of an OP really is not that different than that of SAML's IdP. The difference comes down to the trust model. I certainly think reputation networks will exist which rate OPs, RPs, users, etc and will ultimately be needed for a technologies with promiscuous trust models to thrive in a large scale. I guess reading more of this is making me question if renaming IdP really is the best thing to do in OpenID. I think if anything we all, as a larger community, should be working to bring OpenID and SAML closer together versus driving them further apart. --David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Dick Hardt Sent: Wednesday, November 01, 2006 2:20 PM To: John Kemp Cc: specs@openid.net Subject: Re: Editors Conference Call On 1-Nov-06, at 12:28 PM, John Kemp wrote: OK. Just checking. So an IdP/OP can choose whether or not to trust a particular RP, based on some out-of-ban criteria. And an RP can choose whether or not to trust the assertions of a particular IdP/OP? OK good. Technically possible, yes for the RP to decide on an IdP/OP. Currently, there is no verified RP identity, so the IdP/OP cannot make that decision. I have not had a chance to wade into that discussion. I'd highly recommend it when you get the chance. in my queue :) I suspect the latter case will be unlikely, if OpenID is to be successful. And I do not. And that is the big driver why it should be OP instead of IdP. I think what you're trying to say is that OpenID won't depend on static trust relationships (like business contracts) between RPs and IdP/ OPs - is that right? In which case, sure, I get that. But I do think OpenID will depend on there emerging a way of some RP trusting (or not) some IdP (and vice-versa). Whitelists and blacklists seem like a scalable and dynamic way of doing that, and would seem to be a reasonable way of minimizing the presence of rogue IdPs. Don't take my
Re: IdP vs OP (WAS: RE: Editors Conference Call)
Dick Hardt wrote: On 7-Nov-06, at 7:59 AM, John Kemp wrote: I don't believe that trust is a differentiator between SAML specifications and OpenID Authentication specifications. It is AFAICT, in both cases, simply out of scope. I should have been more clear, IdP is a Federation term and implies trust between the IdP and the RP. That is the definition that many people have about an IdP Since trust is NOT required between an OP and an RP in OpenID, a different term helps clarify that important point I'll quit repeating myself after this go around, but: It [trust] is AFAICT, in both cases, simply out of scope. Cheers, - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: IdP vs OP (WAS: RE: Editors Conference Call)
Eve L. Maler wrote: On balance I prefer identity provider because it's intuitive in an English sense, it's used in several technology contexts (not just SAML and OpenID), and it avoids a terminological branding that would otherwise seem to suggest a conceptual divergence that doesn't -- to my mind -- exist. I agree. - John ___ specs mailing list specs@openid.net http://openid.net/mailman/listinfo/specs
Re: Editors Conference Call
Hi Dick, It would be nice to see a clear definition of an OP in order to determine the exact differences between such an entity and an IdP, but, in the absence of such, some questions: Dick Hardt wrote: Thanks David! ;-) Patrick, as you point out, Identity Provider is a well understood term in SAML and WS-*. Here is the definition from SAML 2.0 [1] Identity Provider: A kind of service provider that creates, maintains, and manages identity information for principals and provides principal authentication to other service providers within a federation, such as with web browser profiles. Per the definition, Identity Provider implies a federation or trust relationship between the IdP and RP. And I guess there is no similar concept in OpenID? Like perhaps an RP adds a particular (I hate to use the word) trusted IdP to a whitelist of IdPs from whom it accepts assertions? Or vice-versa? Admittedly, such federations might not be as linked to static business agreements perhaps as in a typical SAML deployment, but it seems they would be necessary unless you base trust on public key-based mechanisms, or decide that you'll accept assertions from no-password.com (to quote from a discussion on [EMAIL PROTECTED]) and anyone else. I suspect the latter case will be unlikely, if OpenID is to be successful. Additionally, IdPs often provide other assertions about the user. This is sometimes called attribute exchange. In OpenID, is it then not possible for an OP to exchange attributes related to a particular OpenID with an RP? There seems to be an attribute exchange specification (http://openid.net/specs/openid-attribute-exchange-1_0-02.html) where it says (for example, in section 2) Fetch retrieves attribute information from an Identity Provider, while store saves or updates attribute information on the Identity Provider.. In OpenID Authentication, there is no trust relationship requirement between the IdP and RP., and the only thing the IdP asserts is a binding between the user and an identifier (OpenID URL or i-name). And on what basis does the OP assert this binding to an RP? Doesn't the OP typically authenticate that binding, or does it simply take the users identifier on blind faith, and assert away? As people familiar with SAML / WS-* review the OpenID Authentication specification, there has been some confusion on exactly what the IdP does in OpenID. To make it clear that an IdP in OpenID is not the same as typical deployments in SAML, we decided to call it the OpenID Provider, which is more precise, and reduces ambiguity. I guess until we see this precise definition, we won't be able to see the precise differences. From what I can tell so far, it seems to me that the differences between an OP and an IdP are not significant. - John -- Dick [1] http://www.oasis-open.org/committees/download.php/11886/saml- glossary-2.0-os.pdf On 30-Oct-06, at 10:27 PM, Recordon, David wrote: I'll let Dick explain since it was his proposal and I didn't really care about if we changed the name or not. ;) --David From: Patrick Harding [mailto:[EMAIL PROTECTED] Sent: Monday, October 30, 2006 7:47 PM To: Recordon, David; specs@openid.net Subject: RE: Editors Conference Call Dave, Can you please clarify how an OpenID Provider is 'very' different from the role of Identity Provider as defined in SAML or WS-*. Thanks - Patrick Rename Identity Provider to OpenID Provider (IdP - OP) to add clarity to the term since IdP has a very different meaning in the SAML and WS-* worlds -Original Message- From: [EMAIL PROTECTED] on behalf of Recordon, David Sent: Mon 10/30/2006 7:51 PM To: specs@openid.net Subject: Editors Conference Call This morning Dick, Josh, and I got on Skype for 2.5 hours to try and hash through all the remaining proposals. Unfortunately Brad couldn't join us, though I did talk to him about some of this stuff as well beforehand. - Authentication Age will be developed as an extension due to questions around what is the best way for it to work, what features does it need, etc - The field setup_url will be removed from a checkid_immediate response, rather the RP should fallback to a checkid_setup request to complete the transaction. It has been found that in the, albeit few, implementations of checkid_immediate this is the behavior for the setup_url anyway. - Support bare requests by having the field openid.return_to as optional in checkid_* requests. There is a worry of user's not knowing when they'll be redirected back and when they won't, though that will only be worked out by allowing this functionality. - Clarify that the openid.realm parameter should be used to uniquely identifier relying parties - There are some places where it could be clear in step-by-step instructions of what an IdP needs to do in various parts of the protocol, like is done in section 12 for rp's. Sxip will
Re: Editors Conference Call
Dick Hardt wrote: It would be nice to see a clear definition of an OP in order to determine the exact differences between such an entity and an IdP, but, in the absence of such, some questions: Dick Hardt wrote: Thanks David! ;-) Patrick, as you point out, Identity Provider is a well understood term in SAML and WS-*. Here is the definition from SAML 2.0 [1] Identity Provider: A kind of service provider that creates, maintains, and manages identity information for principals and provides principal authentication to other service providers within a federation, such as with web browser profiles. Per the definition, Identity Provider implies a federation or trust relationship between the IdP and RP. And I guess there is no similar concept in OpenID? Like perhaps an RP adds a particular (I hate to use the word) trusted IdP to a whitelist of IdPs from whom it accepts assertions? Or vice-versa? Is it technically possible? OK. Just checking. So an IdP/OP can choose whether or not to trust a particular RP, based on some out-of-ban criteria. And an RP can choose whether or not to trust the assertions of a particular IdP/OP? OK good. Yes. Just like it is technically possible for SAML to be user-centric. :-) Admittedly, such federations might not be as linked to static business agreements perhaps as in a typical SAML deployment, but it seems they would be necessary unless you base trust on public key-based mechanisms, or decide that you'll accept assertions from no-password.com (to quote from a discussion on [EMAIL PROTECTED]) and anyone else. I have not had a chance to wade into that discussion. I'd highly recommend it when you get the chance. I suspect the latter case will be unlikely, if OpenID is to be successful. And I do not. And that is the big driver why it should be OP instead of IdP. I think what you're trying to say is that OpenID won't depend on static trust relationships (like business contracts) between RPs and IdP/OPs - is that right? In which case, sure, I get that. But I do think OpenID will depend on there emerging a way of some RP trusting (or not) some IdP (and vice-versa). Whitelists and blacklists seem like a scalable and dynamic way of doing that, and would seem to be a reasonable way of minimizing the presence of rogue IdPs. Don't take my word for it though - look at the discussion on [EMAIL PROTECTED] Additionally, IdPs often provide other assertions about the user. This is sometimes called attribute exchange. In OpenID, is it then not possible for an OP to exchange attributes related to a particular OpenID with an RP? There seems to be an attribute exchange specification (http://openid.net/specs/openid-attribute-exchange-1_0-02.html) where it says (for example, in section 2) Fetch retrieves attribute information from an Identity Provider, while store saves or updates attribute information on the Identity Provider.. When I talk about assertions, I mean third party claims, not self asserted data. The OP is not verifying the accuracy of any of the attributes in attribute exchange. A claim by my IdP/OP /might/ be a claim by a third-party, no? And if the IdP/OP makes such a claim on my behalf (and is not under my direct control), won't it at least want to verify that the subject of the claim is also the user whose identifier it asserted in OpenID Authentication? In OpenID Authentication, there is no trust relationship requirement between the IdP and RP., and the only thing the IdP asserts is a binding between the user and an identifier (OpenID URL or i-name). And on what basis does the OP assert this binding to an RP? Doesn't the OP typically authenticate that binding, or does it simply take the users identifier on blind faith, and assert away? The OP authenticates the user (how the OP authenticates the user is out of scope of the spec). OK - so the user probably maintains an account with the OP, very much like a user would with an IdP? Unless the user runs her own OP. As people familiar with SAML / WS-* review the OpenID Authentication specification, there has been some confusion on exactly what the IdP does in OpenID. To make it clear that an IdP in OpenID is not the same as typical deployments in SAML, we decided to call it the OpenID Provider, which is more precise, and reduces ambiguity. I guess until we see this precise definition, we won't be able to see the precise differences. How about: An OpenID Provider acts on behalf of the user in responding to OpenID Authentication requests from a Relying Party. What more do we need in the spec then that? What about: An OpenID Identity Provider acts on behalf of the user in responding to OpenID Authentication requests from a Relying Party.? I guess you might want to add something about such an entity typically authenticating the binding between an OpenID and some subject/principal/user. - John
