Questions about signing key pair in IdP

[Tom Scavo]

On Tue, Jul 10, 2012 at 7:15 AM, Ina Müller
<ina.mueller at zdv.uni-tuebingen.de> wrote:
> Here some considerations coming in mind while reading about step 3 in
> https://wiki.shibboleth.net/confluence/display/SHIB2/IdPKeyRollover -

Here's another way to look at key rollover:

https://spaces.internet2.edu/x/vgEFAQ

> As I understand from the above article, the key pair for signing and the
> key pair for the SSL/TLS on port 8443 MUST always be the same and this
> single public key is published through <md:KeyDescriptor use="signing">
> in metadata, right?

Well, a signing key is also used for SSL/TLS on a back channel, that is true.

https://spaces.internet2.edu/x/_YuKAQ

The way you say it, however (using normative MUST), makes me feel
uncomfortable. I don't believe there is anything like that in any spec
that I've read.

Also, since the absence of a 'use' attribute implies both signing and
encryption, use="signing" is not strictly required (as you seem to be
saying above). So I guess what I'm saying is: I agree with your
sentiment but I'd probably say it differently (see above wiki page).

> If the metadata KeyDescriptor does not match the SSL/TLS credentials,
> the SP refuses communication to IdP on 8443 at all?

It had better if it cares at all about security.

> If the metadata KeyDescriptor does not match the signing credentials,
> the SP refuses to accept the content of the SAML answer?

Yes.

> And one more question about key pairs and TrustManagement:
> with <TrustEngine type="ExplicitKey"/> in the SP, there is none of the
> PKI mechanism like validation of RootCA/revocation lists/certificate CN
> or subjectAltName involved for both SSL/TLS on 8443 and SAML signing. So
> only the existence of <md:KeyDescriptor use="signing"> in metadata is
> sufficient for trust from SP side, right?

That last phrase is a little strong, but basically yes, that is true.

Tom