atom feed17 messages in org.oasis-open.lists.security-servicesRE: [security-services] AssertionCons...
FromSent OnAttachments
Mishra, PrateekAug 25, 2004 9:44 am 
Conor P. CahillAug 25, 2004 10:06 am 
Scott CantorAug 25, 2004 10:16 am 
Scott CantorAug 25, 2004 10:19 am 
Conor P. CahillAug 25, 2004 10:31 am 
Scott CantorAug 25, 2004 11:02 am 
Conor P. CahillAug 25, 2004 11:20 am 
Scott CantorAug 25, 2004 11:28 am 
Conor P. CahillAug 25, 2004 11:42 am 
Scott CantorAug 25, 2004 11:51 am 
Conor P. CahillAug 25, 2004 9:13 pm 
Scott CantorAug 25, 2004 9:24 pm 
Conor P. CahillAug 25, 2004 9:26 pm 
Scott CantorAug 25, 2004 9:31 pm 
Scott CantorAug 25, 2004 9:39 pm 
Mishra, PrateekAug 30, 2004 1:24 pm 
Scott CantorAug 30, 2004 1:28 pm 
Subject:RE: [security-services] AssertionConsumerServiceIndex vs. AssertionConsumerURL
From:Scott Cantor (cant@osu.edu)
Date:Aug 25, 2004 11:28:48 am
List:org.oasis-open.lists.security-services

So here's how it's an issue:

I don't think so, but maybe I'm not clear on the attack yet...

So, the Principal somehow browses to BadProvider... BadProvider submits an AuthNRequest to IdP claiming he is SP and providing a consumerURL that points back to a BadProvider managed location. The IdP sends the response back to BadProvider at this location (and in this case we are doing a browser-post type operation, not artifact).

Right. And inside the signed assertion is:

<SubjectConfirmationData Recipient="URL submitted by bad provider">

In SAML 1.1, this was a Recipient in the Response, which was signed, but the basic approach is the same. Limit the location to which the token can be delivered.

In 2.0, I cast this as "limit the location to which the assertion can be presented in a profile and still satisfy the bearer confirmation".

ID-FF does not have this mechanism, and therefore this impersonation attack is dealt with in the manner you describe.

BadProvider can then act as a *browser* client of SP and submits the assertion as a response to the consumer URL of SP and now SP will let the BadProvider act as a bad guy on its site.

See above, this won't work.

So, the IdP shouldn't use a consumer URL unless there is some reason for it to trust it (either a signed request from a trusted party, or because of some trusted metadata or some other such equivalent).

I agree, but for other reasons, not to prevent impersonation via the SSO profile.

-- Scott