[Openid-specs-ab] IdP initiated login

Thu Nov 3 16:48:18 UTC 2011

On Wed, Nov 2, 2011 at 21:39, Nat Sakimura <sakimura at gmail.com> wrote:

> 3. is what Justin is talking about and is the way to go, IMHO.

I agree. I think the IDP should simply redirect the user to a URL at the RP
with a clue about what IDP should be used. The RP will then initiate the
request. This is simpler to maintain (the IDP doesn't need to know what the
RP wants to ask in requests, which can evolve over time) and better for
security (the RP can implement other features such as XSRF protection to
prevent assertion leakage).

>
> =nat
>
>
> On Thu, Nov 3, 2011 at 6:44 AM, John Bradley <ve7jtb at ve7jtb.com> wrote:
>
>> No, it will require some sort of extension.
>>
>> A OAuth 2.0 response doesn't have enough information in it.
>> We are sending the id_token that dose have enough information in it.
>> However basic clients rely on the check_id endpoint.   They don't have
>> enough information to find it, without looking in the id_token.
>>
>> Possible solutions are:
>> 1. make every RP capable of directly inspecting the id_token,  all they
>> need to do is base64url decode the token body to find the issuer and lookup
>> the check_id endpoint if they can't process signatures.
>> 2. Add an additional response paramater of issuer to unsolicited
>> assertions.
>> 3. Add a special RP API to setup the login.
>>
>> 1&2 have the problem of working around the RP's XSRF protection, as well
>> as the IdP knowing what client_id and return URL to use,  they may change
>> over time.
>>
>> People weren't interested when I raised the issue the first time.   I
>> agree that we probably need to address this.
>>
>> There are a bunch of issues that need to be considered.
>>
>>
>> John B.
>>
>>
>> On 2011-11-02, at 6:14 PM, Allen Tom wrote:
>>
>> Hi John -
>>
>> The scenario that I'm talking about is where the user starts on IdP's
>> site in an authenticated state, and is presented with a list of links to
>> services hosted on other sites. The user clicks on one of the links and is
>> seamlessly logged into one of the services without having to
>> re-authenticate.
>>
>> So for example, imagine that the user is an employee and is authenticated
>> into her work intranet portal. The portal has a link to check the user's
>> mail, which is hosted on a different domain. The user should be able to
>> click on the [check mail] link and be able to use the mail application
>> without having to authenticate again.
>>
>> This is a pretty common use case - many companies outsource corporate
>> applications like mail, expense reports, calendaring, payroll, travel
>> bookings, etc  to 3rd parties.  The user logs into their employer's
>> corporate portal and should be able to click on the links to the different
>> apps, without having to re-authenticate.
>>
>> Does OpenID Connect have a solution for this use case?
>>
>> Allen
>>
>>
>>
>>
>>
>> On Tue, Nov 1, 2011 at 7:11 AM, John Bradley <ve7jtb at ve7jtb.com> wrote:
>>
>>> Yes but that, can't be done with the existing OAuth flow in connect.
>>> The OAuth presumption is that the user is starting at the RP and the RP
>>> stores state about what IdP the user is going to.
>>>
>>> There is not enough info in the authentication response alone for the RP
>>> to figure out where it came from.
>>>
>>> We need an extension to OAuth that would start the flow from the IdP and
>>> send the extra info,  as well as protecting against XSRF.
>>>
>>> John
>>> On 2011-11-01, at 2:21 AM, Harald Petersilka wrote:
>>>
>>> Hi John,****
>>> ** **
>>> I was more thinking of a kind of dashboard located at an IdP where I can
>>> collect my OpenID sites and just have to click them to be logged in.****
>>> There I’d send the user to the RP with the already known OpenID as
>>> GET/POST parameter to initiate the common authentication procedure just
>>> without forcing the user to enter his OpenID every time.****
>>> ** **
>>> BG, Harry****
>>> ** **
>>> ** **
>>>  *Von:* openid-specs-ab-bounces at lists.openid.net [mailto:
>>> openid-specs-ab-bounces at lists.openid.net] *Im Auftrag von *John Bradley
>>> *Gesendet:* Dienstag, 01. November 2011 01:32
>>> *An:* Allen Tom
>>> *Cc:* openid-specs-ab at lists.openid.net
>>> *Betreff:* Re: [Openid-specs-ab] IdP initiated login****
>>> ** **
>>> I agree that the user needs to go back to the IdP.  ****
>>> ** **
>>> I was saying the first step of getting the user to the RP can't be
>>> OAuth, because it doesn't have enough info.****
>>> ** **
>>> I was thinking that the IdP would need to have the user full frame and
>>> present some sort of dialog before sending the request to the RP.****
>>> Otherwise the attacker XSRFs the IdP to get to the RP.****
>>> ** **
>>> You seem to have another use case where the attacker is cooperating with
>>> a bad IdP to log the user into a RP as some 4th party that the attacker
>>> controls?****
>>> ** **
>>> John****
>>> On 2011-10-31, at 9:22 PM, Allen Tom wrote:****
>>>
>>>
>>> ****
>>> Hi John -****
>>> ** **
>>> I think that there's still an XRSF vulnerability even if the the IdP
>>> shows a dialog before generating the assertion.****
>>> ** **
>>> The attacker could login to the IdP with the attacker's own account,
>>> then go through the flow to see the dialog. The attacker could then click
>>> through the dialog and capture the assertion (possibly using a browser
>>> extension, or modified client) and then send the assertion to a victim
>>> (possibly via IM or email) to have the victim be signed into the RP using
>>> the attacker's account.****
>>> ** **
>>> This is why the RP needs to bounce the browser back to the OP to have
>>> the OP verify that the assertion was issued to the same browser before
>>> sending the browser back to the RP.****
>>> ** **
>>> I used to think that Login XRSF was not a big deal (who cares if an
>>> attacker can get someone else to login as the attacker?) - however there
>>> are a few interesting scenarios where a victim could have their privacy
>>> compromised.****
>>> ** **
>>> Allen****
>>> ** **
>>> ** **
>>> ** **
>>> On Mon, Oct 31, 2011 at 5:10 PM, John Bradley <ve7jtb at ve7jtb.com> wrote:
>>> ****
>>> If the IdP is not presenting a dialog then it leaves the user open to
>>> XSRF. ****
>>> ** **
>>> In a code flow you only get code and state from the IdP it is the RP's
>>> responsibility to figure out where it came from, so that wouldn't work.*
>>> ***
>>> If you used the Token flow the RP would need to look into the id_token
>>> to see where it came from.  ****
>>> That would be a problem for simple clients, because they need the
>>> introspection endpoint, and to get that they need to get the issuer from
>>> inside the token.****
>>> ** **
>>> I think we need a separate API that takes as its  parameters:****
>>>  issuer****
>>>  nonce****
>>>  hmac of issuer & nonce with client secret.****
>>> ** **
>>> That way the RP can tell that the IdP is sending the request and not a
>>> 3rd party.****
>>> ** **
>>> The RP would do a normal authentication with prompt=none to the issuer.*
>>> ***
>>> ** **
>>> There is a extra hop.****
>>> ** **
>>> We need to identify & authenticate the issuer/IdP somehow in the first
>>> step.****
>>> ** **
>>> John B.****
>>> ** **
>>> ** **
>>> ** **
>>> On 2011-10-31, at 5:42 PM, Allen Tom wrote:****
>>>
>>>
>>> ****
>>> Hi John -****
>>> ** **
>>> Is this the Unsolicited Assertion use case - where the user clicks on
>>> a sponsored link hosted on the IdP's site and gets authenticated on the
>>> RP's site?****
>>> ** **
>>> I think we had discussed this at IIW a couple years ago, and the general
>>> consensus was that upon receiving an unsolicited positive assertion, the RP
>>> would need to redirect the user's browser back to the OP to have the OP
>>> re-generate the assertion and resend it back to the RP. ****
>>> ** **
>>> The downside is that the UX would suffer due to the extra round trip.***
>>> *
>>> ** **
>>> Allen****
>>> ** **
>>> ** **
>>> ** **
>>> ** **
>>> On Mon, Oct 31, 2011 at 6:56 AM, John Bradley <ve7jtb at ve7jtb.com> wrote:
>>> ****
>>>
>>> Just a note on a possible idea.
>>>
>>> The when the RP registers a client ID it sets unsolicited_login_url:  to
>>> some return_url
>>>
>>> The IdP then sends the id_token with nonce set to  a time stamp +
>>> entropy , and a claim of idp_initiated: true .
>>>
>>> We probably need to restrict this to the code flow.
>>>
>>> RP could then check that the id_token was not generated by XSRF and set
>>> it's cookies.
>>>
>>> I don't see a general way that a unmodified RP is going to be able to
>>> safely.
>>>
>>> This should probably be an extension.
>>>
>>> John  B.
>>> _______________________________________________
>>> Openid-specs-ab mailing list
>>> Openid-specs-ab at lists.openid.net
>>> http://lists.openid.net/mailman/listinfo/openid-specs-ab****
>>> ** **
>>> ** **
>>> ** **
>>>
>>>
>>>
>>> _______________________________________________
>>> Openid-specs-ab mailing list
>>> Openid-specs-ab at lists.openid.net
>>> http://lists.openid.net/mailman/listinfo/openid-specs-ab
>>>
>>>
>>
>>
>> _______________________________________________
>> Openid-specs-ab mailing list
>> Openid-specs-ab at lists.openid.net
>> http://lists.openid.net/mailman/listinfo/openid-specs-ab
>>
>>
>
>
> --
> Nat Sakimura (=nat)
> Chairman, OpenID Foundation
> http://nat.sakimura.org/
> @_nat_en
>
>
> _______________________________________________
> Openid-specs-ab mailing list
> Openid-specs-ab at lists.openid.net
> http://lists.openid.net/mailman/listinfo/openid-specs-ab
>
>

-- 
--Breno
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-specs-ab/attachments/20111103/c5012390/attachment.html>