[openssl-users] WG: TLS Error in FreeRadius - eap_tls: ERROR: Failed in __FUNCTION__ (SSL_read): error:1417C086:SSL routines:tls_process_client_certificate:certificate verify failed

[Jeffrey Walton]

On Tue, Jan 23, 2018 at 12:43 PM, Viktor Dukhovni
<openssl-users at dukhovni.org> wrote:
>
>
>> On Jan 23, 2018, at 7:31 AM, Gladewitz, Robert via openssl-users <openssl-users at openssl.org> wrote:
>>
>> Despite being wrong it is also absolutely irrelevant, because FreeRADIUS
>> retrieves the OpenSSL rejection of the cacert.capf.pem before any end-entity
>> certifcate is ever seen.
>
> This is almost certainly not the case.  Why would FreeRADIUS be validating
> a stand-alone CA certificate that is not part of a chain from a leaf
> certificate to a trust anchor?  The chain is validated for a particular
> purpose, presumably "TLS client" (the error message in the subject line of
> this thread is about the processing of a client certificate).
>
>> We now have a very plain cacert.capf.pem. It only shows the following
>> attributes:
>> X509v3 Key Usage: critical
>>                Certificate Sign, CRL Sign
>> and things work.
>
> Great!  This confirms that the issue was the restricted EKU of the
> intermediate CA certificate.
>
>> So to sum up: there is a mistake, that we know of, but really it is not in
>> our hands to change it. And we do not need to change it, because it is of no
>> concern to the problem at hand.
>
> The previous CA certificate was explicitly not suitable for verifying TLS
> clients, based on the de facto interpretation of EKU in CA certificates in
> OpenSSL and various other TLS libraries.  So the issue was very much a
> "concern to the problem at hand".
>
>> Secondly the presented cacert.capf.pem is (by itself) a valid certificate.
>
> It is a valid intermediate CA certificate, that is (de facto) constrained
> to be used only for verifying for TLS server certificates, and NOT TLS
> client certificates.
>
>> It does not present a mistake. We should also not need to change it - but we
>> do. Why?
>
> You had to change it because you want to use this CA to issue TLS client
> certificates, and so its EKU needs to either be absent or to explicitly
> permit TLS client authentication.
>
>
>> Having read all the discussion i do not know why.
>
> Mostly because despite my best attempts to explain the above, perhaps
> a language barrier, and/or my inability to explain the issue sufficiently
> clearly, is preventing the reasons from being communicated effectively.
>
>> It is a CA certificate and Cisco somehow restricts this CA certificate to
>> a certain chosen purpose.
>
> Perhaps you mean that default software settings create such a certificate.
> If so, please raise this as a bug with the software vendor.  As you've
> already seen deploying a CA without an EKU works, so the previous EKU is
> not in fact a requirement.
>
>> OpenSSL should not interfere at this point.
>
> OpenSSL implements a widely practiced de facto CA certificate "purpose"
> policy based on the optional EKU extension in the CA certificate.  If
> you don't to restrict the purposes for which a CA can issue EE certificates
> (directly or indirectly), then DO NOT include an EKU in the CA certificate.
> If you do want to limit the CA to issue only certificates for particular
> purposes, then include all (and only) those purposes in the EKU.
>
> Good luck.  And thanks for reporting this, this discussion should
> help other users to quickly resolve similar issues in the future.

Your arguments are fallacious. How the browsers do things does not
constitute the "de facto" standard. Your just begging the claim.

The docs have _not_ changed: https://www.openssl.org/docs/standards.html.

Jeff