Hi John, I'd like to point out that the International Grid Trust Federation - http://igtf.net - was established specifically to provide a high-quality source of trusted CAs used in global research cyberinfrastructure and has been fulfilling this role for many years. These certificates, both for individuals and for hosts or services, provide the trust federation basis for large-scale computational grids and associated cloud projects, and might be a good source for your project. Generally obtaining grid certificates is available through national research and education initiatives or associated projects. OGF is on the verge of announcing free availability of these certificates to its members to make them available more broadly, and in addition will be supporting the availability of commercially trusted certificates (useful for those cases where you do want the certificate pre-loaded without going through installing an IGTF trust store to support your application) at discounted rates from the same supplier as the free grid ones. We do have a pretty well developed authorization community that can help with implementation issues like the ones that you mentioned. As for the widespread availability of certificates, this is generally seen as a good thing, depending on your needs. Obtaining a certificate from a commercial CA is easy, of course, and doesn't require the individual identity verification that is required by at least some of the IGTF profiles. For control-plane peering of service agents, I assume that simple possession of a valid certificate is not enough, and what you need is an authorization layer that says that a particular DN is allowed to carry out operations on the network function being controlled. We had planned to put together a workshop on identity management for software defined networking for SC'14, but I don't think we got that submitted in time. This sounds like a topic that would be good to discuss at an OGF meeting or other gathering of the NSI group. Alan On Jul 31, 2014, at 6:21 PM, John MacAuley <macauley@es.net> wrote:

Yes, I was indeed mixing authentication and basic authorization. I have solved the issue by adding certification DN authorization in Apache after the TLS session is established. It is just too bad TLS gets established in the first place with these wide ranging CAs. Seems a bit senseless in the grand scheme of security.

Java based implementations can override the default SSL Engine to give customized handling of the certificates, which solves my problem during the negotiation phase. Unfortunately, not everyone can do this.

"Self-signed certificates will not scale." - It really depends on the deployment requirements of the application. We are discussing control plane peering of service agents, of which an organization will typically have a handful to tens for the foreseeable future. I would not use self-signed for use cases where I am dealing with 100 - 1,000s of clients. In that case it definitely does not scale. However, having to provision 1,000s of access control lists to restrict access does not scale as well. If this was the case an entirely different solution would be required that does not depend on SSL/TLS for anything other than encryption.

As they say, six of one, half a dozen of the other.

Thank you for the feedback.

John

On 2014-07-31, at 10:58 AM, Mischa Salle <msalle@nikhef.nl> wrote:

...

Hi Alan,

I think he's mixing authentication and authorization. If I look at page 10, my reaction is, there is nothing wrong in trusting that client cert #2 is client cert #2, that's only the authentication part. That doesn't mean you also *allow* client cert #2. Same on page 11, trusting a certificate there should mean trusting that the identity is what it claims to be. Doesn't mean allowing it to enter. If this is indeed his whole use-case, I would say, go for (a subset of) public CAs and restrict access based on specific DNs. That gives you still all the revocation and renewal possibilities while at the same time allowing for restricted access. A private CA could work, but is a lot of work, and not trivial to keep safe... Self-signed certificates will not scale. They give no way of revocation, big problems with expiry etc. If they need more advanced authZ, such as based on certain roles, than indeed VOMS attribute certificates might be useful, although that would mean software adaptations.

Best wishes, Mischa Sallé

On Wed, Jul 30, 2014 at 03:34:08PM +0000, Sill, Alan wrote:

...

Dear folks in the OGF CAOPS, VOMS-PROC and NSI working groups.

I'd like to initiate some discussion among the participants in these working groups for the use case referred to in the talk at the link below.

Some review of the conditions for this use case would be helpful. Note this is also a use case that comes up in Internet-of-Things discussions, and has caused some discussion on the PKIX group list (though that group is now dormant of course) and other related lists lately.

To me this is a familiar situation with well-known parameters, but possibly some additional considerations, and might possibly lead to some useful communication among the members in these groups about solutions that could be applied using existing technologies that would avoid the possible downsides associated with the proposed use of self-signed certificates. (For example, extended attribute certificates as used in VOMS, though the same or perhaps through a different implementation, might be a good solution here; other solutions might be contemplated that would be more attractive than self-signed certificates for this situation.)

Your comments, discussion and input are recruited (by me -- I'm not speaking for the NIS-WG members per se!), and I hope that all parties will regard this as useful discussion for information exchange only.

Thanks, Alan

Begin forwarded message:

From: Guy Roberts <Guy.Roberts@dante.net<mailto:Guy.Roberts@dante.net>> Subject: RE: [Nsi-wg] Wednesday's NSI conf call Date: July 30, 2014 at 1:30:19 PM GMT+2 To: Alan Sill <kilohoku150@gmail.com<mailto:kilohoku150@gmail.com>>

Hi Alan,

Please find the slides on NSI security here:

https://redmine.ogf.org/dmsf/nsi-wg?folder_id=6592

The proposal is that NSAs will run their own private Certificate Authorities (self-signing) rather than using public Certificate Authorities. Participating NSAs will then exchange information about each other’s Certificates in an ad hoc way.

This solution does not scale well as private Certificates have to be manually shared, but it reduces the size of the certificate pool.

Guy

From: Alan Sill [mailto:kilohoku150@gmail.com] Sent: 30 July 2014 10:56 To: Guy Roberts Cc: Alan Sill Subject: Re: [Nsi-wg] Wednesday's NSI conf call

Guy,

On Jul 30, 2014, at 11:02 AM, Guy Roberts <Guy.Roberts@dante.net<mailto:Guy.Roberts@dante.net>> wrote:

- comments/feedback from last week’s presentation from John on ‘Secure Communications with Self Signed Certificates’

Are copies of these slides available? I would like to understand the context.

(In general, use of self-signed certificates is risky at best, so I would like to understand the use case here.)

Alan

...

_______________________________________________ voms-proc-wg mailing list voms-proc-wg@ogf.org https://www.ogf.org/mailman/listinfo/voms-proc-wg

-- Nikhef Room H155 Science Park 105 Tel. +31-20-592 5102 1098 XG Amsterdam Fax +31-20-592 5155 The Netherlands Email msalle@nikhef.nl __ .. ... _._. .... ._ ... ._ ._.. ._.. .._.. _______________________________________________ nsi-wg mailing list nsi-wg@ogf.org https://www.ogf.org/mailman/listinfo/nsi-wg

_______________________________________________ nsi-wg mailing list nsi-wg@ogf.org https://www.ogf.org/mailman/listinfo/nsi-wg