Shibboleth Developers

[Walter Hoehn <>]

On Jul 6, 2005, at 10:05 AM, Ian Young wrote:
> Scott Cantor wrote:
> > In practice, chains are dumb. The PKIX code doesn't care if the  
> > anchor is
> > self-signed. So you can trust the issuer directly, and I don't see  
> > why
> > anybody wouldn't be able to live with that. But fixing it doesn't  
> > really
> > seem possible in any other way.
>
> I've now tested that in our case, adding the intermediates in  
> question to the EntitiesDescriptor /Extensions /KeyAuthority list  
> allows the IdP to validate the "path".  I imagine, though, that it  
> is not validating the whole path to the self-signed root but only  
> to the immediate signer of the end certificate; do you agree?  Is  
> there a way to confirm this with the current codebase?

I've tried it, but it wouldn't hurt to have a jUnit test.

> I think you're right, that we should be able to live with this,  
> particularly if the alternative is holding our breath until someone  
> fixes AJP...  I'm assuming that adding intermediates to the  
> KeyAuthority list won't cause trouble for 1.2 entities; do you know  
> of any issues that might come up there?

The IdP doesn't do validation in 1.2.  It is handled completely by  
mod_ssl.  The 1.2 SP should work just fine with the intermediates in  
the metadata as long as the included chain goes all the way up to the  
root.

> Final question: I'm interested to hear that the code doesn't care  
> if the anchor is self-signed, as I thought that OpenSSL (as used by  
> the C++ SP?) had just such a limitation?  That implies to me that  
> we need to keep the whole chain of intermediates in the metadata  
> even if it shouldn't strictly be required (the chain I'm looking at  
> is strictly A->B->C->client with no branching, so A and B are  
> superfluous unless a self-signed root is required by someone).

The IdP trust module definitely allows anchoring with non-self-signed  
certificates.  This is correct behavior according to PKIX.  I believe  
that Scott worked around the OpenSSL limitations in 1.3 so that both  
code-based behave the same, but I'll let him confirm that.

-Walter