RE: OCSP Algorithm Agility

[Stefan Santesson <[e-mail]>]

I'm generally supportive of the following statement.

> I'm generally in favor of simpler approaches, such as those suggested by
> Santosh, as they favor interoperability. Updating deployed applications to
> understand new elements of the protocol could take just as long to roll out
> as the new algorithm support that's purportedly lacking.

I've been trying to take a step back to think what I would need from a more practical viewpoint.


Security:
None of the presented threats is not a great concern for me. Downgrade attacks is only useful if we use algorithms that can be broken.
I would expect an organization to only allow algorithms that provide adequate security.
The stolen OCSP key is neither a problem for me from a protocol perspective. We have revocation.


Practical matters:
I can only find one practical matter that concerns me. I have an OCSP server and of some reason I want to gradually transition responses to be based on say ECC with SHA 256. Some clients can only verify RSA with SHA-1. The reason for doing this transition may be entirely policy oriented. Now I would like to know which clients that can actually verify my upgraded responses.

I see some value in the client knowing which algorithms the server can support, but not greatly. The client can mostly learn the capability of a server by observing its responses.


What can I do today:
The server can draw a number of conclusions today.

1) Looking at the hashAlgorithm of the CertID
2) Looking at the signatureAlgorithm used by the client to sign the response
3) Looking at the certificate actually being validated

None of these are 100% bulletproof to cover all corner cases. But are they good enough and is the remaining gap big enough to motivate a protocol update?


Possible updates:
To cover the corner cases I have seen the following options:
Client declaration - through a singleRequestExtension
Server declaration - through responseExtensions, or through an extension in the responder certificate.


Conclusion:
I'm not convinced I need any update of OCSP. If we do an update anyway, I'm more interested in the client declaration than the server declaration problem.



Stefan Santesson
Senior Program Manager
Windows Security, Standards


> -----Original Message-----
> From: [e-mail] [mailto:owner-ietf-
> [e-mail]] On Behalf Of Seth Hitchings
> Sent: den 10 oktober 2007 21:26
> To: Santosh Chokhani; Hallam-Baker, Phillip; [e-mail]
> Subject: RE: OCSP Algorithm Agility
>
> I'm generally in favor of simpler approaches, such as those suggested
> by
> Santosh, as they favor interoperability. Updating deployed applications
> to
> understand new elements of the protocol could take just as long to roll
> out
> as the new algorithm support that's purportedly lacking.
>
> I can see a situation in which the CA cannot assume that the entire
> user
> population understands a new algorithm such as SHA-256, and thus has to
> continue using SHA-1 with RSA when signing certificates. Individual
> clients
> that do understand SHA-256 may wish to receive OCSP responses signed
> using
> SHA-256 with RSA, and would need a way to signal their desires to the
> OCSP
> responder.
>
> The question is, then, is there any security benefit to having the
> signature
> algorithm on the OCSP response stronger than the signature algorithm on
> the
> certificate? If you can break the algorithm and cause me to accept a
> contrived OCSP response, could you not do the same thing with a
> certificate?
>
> Seth
>
> -----Original Message-----
> From: [e-mail] [mailto:owner-ietf-
> [e-mail]] On
> Behalf Of Santosh Chokhani
> Sent: Wednesday, October 10, 2007 2:28 PM
> To: Hallam-Baker, Phillip; [e-mail]
> Subject: RE: OCSP Algorithm Agility
>
>
> See responses in-line below.
>
> -----Original Message-----
> From: Hallam-Baker, Phillip [mailto:[e-mail]]
> Sent: Tuesday, October 09, 2007 2:07 PM
> To: Santosh Chokhani; [e-mail]
> Subject: RE: OCSP Algorithm Agility
>
> Not if they don't know that the client can understand them they can't.
> [Santosh Chokhani] The Responder has two ways to know what the client
> understands.  From certID or from CA.  If the client does not
> understand
> the algorithm used in certificate signing, revocation information is
> useless.
>
> And the issue is not just hashing algorithms, it is signature
> algorithms.
> Like your previous posts you continue to assume that cipher strength is
> a linear quantity with only one dimension and that any client which
> supports strength 2X must automatically support strength X.
> [Santosh Chokhani] I am making no such assumptions.  I am saying that
> if
> an algorithm is good enough for a certificate, it is good enough for
> revocation notification for that certificate.
>
> That is not the case if you consider the practicalities of deploying
> ECC
> in an RSA world.
>
> To give a practical example, VeriSign issues a cert for an ECC key
> signed with an ECC key. The email program on Alice's machine in the DHS
> supports ECC but not the DHS SCVP server or OCSP relay.
> [Santosh Chokhani] And what is the problem with that?  They have no
> choice but to use RSA.
>
> In the real world the OCSP service does not have a necessary connection
> to the CA. There are plenty of commercial OCSP offerings that report on
> certificates issued by other CAs.
> [Santosh Chokhani] And how does this relate to algorithm agility?  If
> the OCSP does not do ECDSA it won't.  If OCSP supports both RSA and
> ECDSA, it can take the cue from the CA CRL to determine what algorithm
> to use for signing OCSP responses.  If the OCSP does not consume CRL,
> the mechanism that tells it revocation information can also tell it the
> signature algorithm to use.
>
> > -----Original Message-----
> > From: [e-mail]
> > [mailto:[e-mail]] On Behalf Of Santosh Chokhani
> > Sent: Tuesday, October 09, 2007 11:16 AM
> > To: Hallam-Baker, Phillip; [e-mail]
> > Subject: RE: OCSP Algorithm Agility
> >
> >
> > The OCSP and SCVP can transition to stronger algorithms any
> > time they want.
> >
> > The OCSP can take their clue from the CA or the certID field
> > to decide the hashing algorithm.  When multiple certID is
> > present, OCSP can take low water mark of these.
> >
> > SCVP is no different than a client in terms of validating a
> > path.  In terms of signing a response, it can take the low
> > water mark of the hashing algorithms in the certificate chain.
> >
> > -----Original Message-----
> > From: [e-mail]
> > [mailto:[e-mail]]
> > On Behalf Of Hallam-Baker, Phillip
> > Sent: Monday, October 08, 2007 7:27 PM
> > To: Santosh Chokhani; Andrews, Rick; [e-mail]
> > Subject: RE: OCSP Algorithm Agility
> >
> >
> > The objective here is to overcome a real deployment obstacle
> > that creates a lockstep issue when use of stronger algorithms
> > is attempted.
> >
> > The CA cannot issue a certificate that employs a new
> > algorithm until it is known that all clients that might rely
> > on the certificate are capable of processing the new
> > certificate. Requiring lockstep updates to the OCSP and SCVP
> > infrastructure to be made at the same time renders transition
> > to stronger algorithms a much lengthier and much less
> > practical proposition.
> >
> >
> > > -----Original Message-----
> > > From: [e-mail]
> > > [mailto:[e-mail]] On Behalf Of Santosh Chokhani
> > > Sent: Saturday, October 06, 2007 4:12 PM
> > > To: Andrews, Rick; [e-mail]
> > > Subject: RE: OCSP Algorithm Agility
> > >
> > >
> > > Rick,
> > >
> > > SCVP or other intermediaries are red herring.  Whoever
> > processes the
> > > certificate processes the revocation information for that
> > certificate.
> > > It does not change the following:
> > >
> > > 1.  There is no security benefit in using a stronger algorithm for
> > > OCSP
> > > response than for the certificate in question.   Neither,
> > there is any
> > > benefit from interoperability viewpoint for these two to be
> > different.
> > >
> > > 2.  If the OCSP Responder does not get a CRL, it can use the same
> > > mechanism to get the hashing algorithm used as it uses to get the
> > > revocation information.
> > >
> > > -----Original Message-----
> > > From: Andrews, Rick [mailto:[e-mail]]
> > > Sent: Wednesday, October 03, 2007 3:29 PM
> > > To: Santosh Chokhani; [e-mail]
> > > Subject: RE: OCSP Algorithm Agility
> > >
> > > Santosh,
> > >
> > > Sorry for the long delay in responding - travel and vacation.
> > >
> > > Not all OCSP responders work from CRLs, so they won't take
> > their cue
> > > from the CRL. Nor should they take their cue from the
> > signature on the
> > > cert in question, I believe. Let me try to restate my argument in a
> > > different way.
> > >
> > > With SCVP delegated path validation, the client requesting a cert's
> > > status from an OCSP responder will be different from the
> > client at the
> > > other end of the SSL connection. Those two clients may have very
> > > different capabilities in terms of supported signature and hash
> > > algorithms. It's not realistic to expect that all SSL clients, all
> > > SCVP servers, and all CAs will be able to upgrade in
> > lockstep to new
> > > algorithms as they are developed. Allowing the OCSP client
> > and server
> > > to negotiate a mutually-acceptable set of algorithms is
> > essential to
> > > the deployment of newer, stronger algorithms.
> > >
> > > Likewise, companies that run large OCSP responders may wish to
> > > gradually move to ECC-based signatures for all their OCSP
> > responses,
> > > even those for certs with RSA or DSA keys, because ECC
> > signatures are
> > > cheaper to produce. If algorithm agility is added to OCSP, those
> > > companies can gradually achieve the move to ECC without
> > disrupting the
> > > installed base of OCSP clients that don't support ECC.
> > >
> > > -Rick Andrews
> > >
> > > > -----Original Message-----
> > > > From: [e-mail]
> > > > [mailto:[e-mail]] On Behalf Of
> > Santosh Chokhani
> > > > Sent: Friday, September 21, 2007 2:45 PM
> > > > To: Paul Hoffman; Stephen Kent; [e-mail]
> > > > Subject: RE: OCSP Algorithm Agility
> > > >
> > > >
> > > > Paul,
> > > >
> > > > Here are my views on this.
> > > >
> > > > The client should be first asking for the algorithm suite
> > > that signed
> > > > the certificate in question.  There is no need for the
> > > client to ask
> > > > for anything stronger.  The client can ask for stronger suites as
> > > > secondary, if client has them.
> > > >
> > > > In the scenario you cite, the Responder certificate will
> > > not include
> > > > RSA with SHA 1 any longer.  So, client will know that
> > > Responder only
> > > > supported his second choice and he should be ok with it.
> > > >
> > > > -----Original Message-----
> > > > From: [e-mail]
> > > > [mailto:[e-mail]]
> > > > On Behalf Of Paul Hoffman
> > > > Sent: Friday, September 21, 2007 4:39 PM
> > > > To: Stephen Kent; [e-mail]
> > > > Subject: RE: OCSP Algorithm Agility
> > > >
> > > >
> > > > At 2:07 PM -0400 9/21/07, Stephen Kent wrote:
> > > > >How about defining an extension to be included in the cert
> > > > issued to an
> > > > >OCSP responder by a CA.  The extension would have an
> > > ordered list of
> > > > >algorithms (hash and signature if we want to address more
> > > > than the hash
> > > > >agility issue) accepted by the OCSP responder.  An OCSP
> > > > client can use
> > > > >this info to determine what is the "best" algorithm (or alg
> > > > pair) that
> > > > >it and the responder share. The combination of this
> > > extension and an
> > > > >OCSP negotiation procedure will allow the client to detect MITM
> > > > >downgrade attacks. In fact, if the client acquires the
> > > > responder's cert
> > > > >prior to making a request, there would not even be a
> > need for real
> > > > >negotiation, since the client would know what alg to
> > request in a
> > > > >response.
> > > >
> > > > Imagine the list of algorithms is RSA-with-SHA1 first and
> > > > DSA-with-SHA1 second. How does your negotiation work? The
> > > client asks
> > > > for this message to be signed with RSA-with-SHA1.
> > > > But the server knows that RSA-with-SHA1 has been
> > > compromised since it
> > > > got that certificate from the CA. What does the server say to the
> > > > client to indicate that it only wants to sign with
> > > DSA-with-SHA1? What
> > > > prevents Mallory from saying the same thing to the client?
> > > >
> > > > --Paul Hoffman, Director
> > > > --VPN Consortium
> > > >
> > > >
> > > >
> > >
> > >
> >
> >
> >