RE: Signer certificate discovery for CRLs

["Stefan Santesson" <stefans@xxxxxxxxxxxxx>]

David,

Thanks for the clarifications, they make sense.
I did misread your pictures.

So can we conclude that for a re-keyed CA in accordance with RFC 3280, either the CRL issuer certificate itself, or the location of the CRL issuer certificate, will be clearly identified/available for the validating party in some cases, but not in others?

Can we also conclude that there is no real discovery solution for indirect CRLs?

Do you then agree then that it could be appropriate to allow AIA as a CRL extension to facilitate discovery of CRL signer certificate?

Stefan Santesson 
Microsoft Security Center of Excellence (SCOE) 
  
________________________________________
From: David A. Cooper [mailto:david.cooper@xxxxxxxx] 
Sent: den 12 oktober 2004 21:14
To: Stefan Santesson
Cc: pkix
Subject: Re: Signer certificate discovery for CRLs

Stefan,

I believe that you are misinterpreting the figures.  They really do represent three different cases, not three different certification paths that have been constructed through the same PKI architecture.

In figure 1, CA 1 has generated self-issued key rollover certificates.  The Root CA has issued a certificate to CA 1's new key, but not its old key (either the Root CA never issued a certificate to CA 1's old key or that certificate has expired).

In figure 2, CA 2 has also generated self-issued key rollover certificates.  The Root CA has issued a certificate to CA 2's old key, but not its new key.

In figure 3, when CA 3 performed key rollover, it requested a new CA certificate from the Root CA.  CA 3 did not generated self-issued key rollover certificates.

Of course, another realistic scenario would be one in which a CA generated self-issued key rollover certificates upon key rollover and then had the Root CA issue a CA certificate to its new key.  In this case, as you suggest, any of the certification paths from figures 1, 2, or 3 could be constructed.

As for the contents of the AIA extension, here is what I have specified in the "X.509 Certificate and CRL Extensions Profile for the Common Policy":

The authorityInfoAccess extension uses URIs for two purposes. When the id-ad-caIssuers access method is used, the access location specifies where certificates issued to the issuer of the certificate may be found. If LDAP is used, the URI must include the DN of the entry containing the relevant certificates and specify the directory attribute in which the certificates are located. If the directory in which the certificates are stored expects the "binary" option to be specified, then the attribute type must be followed by ";binary" in the URI. If HTTP is used, the URI must point to a file that has an extension of ".p7c" that contains a certs-only CMS message (see RFC 2633). The CMS message should include all certificates issued to the issuer of this certificate, but must at least contain all certificates issued to the issuer of this certificate in which the subject public key may be used to verify the signature on this certificate.
....
For a certificate issued by "Good CA", some examples of URIs that may appear as the access location in an authorityInfoAccess extension when the id-ad-caIssuers access method is used are: 
ldap://smime2.nist.gov/cn=Good%20CA,o=Test%20Certificates,c=US?cACertificate,crossCertificatePair
ldap://129.6.20.71/cn=Good%20CA,o=Test%20Certificates,c=US?cACertificate;binary,crossCertificatePair;binary
http://fictitious.nist.gov/fictitiousCertsOnlyCMSdirectory/certsIssuedToGoodCA.p7c
For both LDAP and HTTP, the URI provides the exact location where information is to be located, so there is no requirement for the relying party to try to figure out where information is located.

The HTTP URI points to a certs-only CMS message that includes all certificates issued to the CA identified in the issuer field of the certificate.

The LDAP URI points to the cACertificate and crossCertificatePair attributes of the directory entry of the CA identified in the issuer field of the certificate.  These two attributes together hold all of the certificates issued to the CA:  the cACertificate attribute holds the CA's self-issued certificates and the crossCertificatePair attribute holds the cross-certificates issued to the CA by other CAs.

Dave


Stefan Santesson wrote:

David,
 
Thanks for these good thoughts and very useful scenarios.
 
I have some comments and questions on this.
 
First of all we can conclude that in some scenarios (figure 1) where a self issued certificate is inserted into the path, you are likely to find the CRL issuer cert in the path. (given that the new CA have a common key and certificate for cert signing and CRL signing).
 
Figure 1, 2 and 3 describe the same case. It is just describing different path building strategies. An application that has access locally to all chaining options may however still choose path 2 for the certs and path 1 for the CRL independent of each other (which I think figure 3 tries to describe)
 
Another comment is the structure of AIA extensions. The use I'm familiar with doesn't use AIA to describe a directory entry where it is left to the validation application logic to be intelligent enough to find appropriate certificate data from the directory. The model I'm familiar with is when the AIA URL explicitly identifies the exact location of the appropriate CA certificate file, relieving the validation software from complex information queries.
If just location of explicit certificate files are identified through AIA, the presence of an AIA may not help finding the CRL signer cert if this is different from the CA certificate.
This is also the problem with Denis proposal.
 
I think we share the basic conclusion that the ability to locate the CRL signer certificate directly through the CRL could be very useful. At least in the case of indirect CRL but it could also be proven very useful in CA re-keying scenarios.
 
The easiest solution would probably be to allow AIA to be used in its current shape and structure as a CRL extension (MUST NOT be critical).
It would present a very clear and uncomplicated logic to certificate validating applications in many cases.
 
Stefan Santesson 
Microsoft Security Center of Excellence (SCOE) 
  
________________________________________
From: David A. Cooper [mailto:david.cooper@xxxxxxxx] 
Sent: den 7 oktober 2004 18:35
To: Stefan Santesson
Cc: pkix
Subject: Re: Signer certificate discovery for CRLs
 
Stefan,

I think what you are proposing is a good idea.  In most cases, path discovery algorithms assume that both the trust anchor (or trust anchors) and the end entity certificate are provided as input.  In this case, one may need to construct a certification path without a priori access to the end entity certificate (the one with the subject public key corresponding to the CRL signing key).  Including an AIA extension (or some other pointer) in the CRL would provide the relying party with a simple way to obtain the end entity certificate for the CRL signing key's certification path.
On the other hand, I believe that a relying party should be able to construct the certification path even without an AIA extension in the CRL, so long as it is not an indirect CRL.  Attached is a drawing of the three basic scenarios that I expect a relying party may encounter:

In each of these scenarios, the CA has performed key rollover and is only signing CRLs with its new key.  The diagrams would look similar, however, if the CA simply choose to use different keys to sign certificates and CRLs for some other reason.

If the PKI architecture resembled figure 1, then the certification path for the CRL signing key would just be a subset of the certification path for the EE certificate, so no addition path discovery would be needed.

If the PKI architecture resembled figure 1, then it would be necessary to obtain the new-signed-with-old self-issued certificate.  In building the certification path to the EE certificate, however, the relying party will obtain the certificates pointed to by the AIA extension in the EE certificate.  This AIA extension will point to a location containing all certificates issued to CA 2, which would include both the certificate issued by the Root to CA 2 and CA 2's self-issued certificate.  So, even though the self-issued certificate would not be part of the certification path to the EE certificate, it would be downloaded by the relying party during the construction of that certification path.  (Yes, there are circular dependency issues in figure 2, but that is another issue.)

A similar situation would happen if the PKI architecture resembled figure 3.  The AIA extension in the EE certificate would point to a location containing certificates issued to CA 3.  When the relying party downloaded these certificates, it would obtain both of the certificates issued by the Root to CA 3 and so again would have the certificate needed to validate the CRL signing key.

In the case of an indirect CRL, things may not work as well.  If indirect CRLs were used, and the PKI architecture resembled figures 2 or 3 (replacing "New Key" with a different CA), then the set of certificates pointed to by the AIA extension in the EE certificate would not include the certificate of the CRL signer.  One could find this certificate by building in the reverse direction and using the SIA extension, but that may not be the most convenient solution.

So, when indirect CRLs are being used, it seems that it would be very useful to have a pointer in the CRL to the CRL signing certificate.  When the CRL is not indirect, the need for this pointer does not seem to be as clear, but I can't see any harm in including it.

Dave

Stefan Santesson wrote:
All,
 
I'm interested in the opinion from members on this list about discovery
of CRL signer's certificate in non directory centric environments.
 
The problem is the following.
 
The relying party (RP) needs to check validity of a certificate and
finds a CDP extension with a URL to the CRL.
The RP retrieves this CRL which in this particular case is either signed
by another key of the CA (re-keyed CA) or another entity (indirect CRL).
 
In this case the relying party needs to obtain the certificate of the
CRL signer which may NOT be part of the original chain. In a directory
centric solution this is retrieved from the directory, but what if such
directory is not available or accessible.
 
The RP have thus no hint where to find the CRL issuers certificate
unless the RP already have possession of it by some other means.
 
Is seems that CRLs would need an AIA extension with the option to point
to the location of the signers certificate in the same manner as is
possible for certificates.
 
Maybe AIA should be defined as both cert and CRL extension and not only
certificate extension as today.
 
Thoughts and comments?
 
Stefan Santesson
Microsoft Security Center of Excellence (SCOE)