Comments and potential additions to the Usage Guide

["Fitzgerald, Indra" <indra.fitzgerald@hp.com>]

All,

During last weeks call, I was asked to send out the comments I exchanged with Robert. Below is a list of outstanding concerns and comments. Please let us know if you have any comments, agree or disagree, and if you would like to see additional clarification and examples in the spec or Usage Guide.

Regards,
Indra

1.	p.14, 2.1.7.1: Clarify in the Usage Guide that Transparent Symmetric Key is essentially the same as Raw key type. Raw key type allows clients to avoid keeping the octet key in a structure.

2.	p.22,3.2, line 352: Name Type includes extension for vendor-specific implementation. Vendor-specific attributes or operations are avoided in the spec and Usage Guide. For example, the Usage Guide specifies that if clients require unique identifiers in a special form, out-of-bound registration/configuration can be used to communicate this to the server. We should use a similar approach for server-created Application Data (as part of the Application Specific Information attribute). Instead of relying on implied, side-effect functionality, the spec should either make server-generated Application Data vendor-specific or require clients to explicitly request the server to generate Application Data by setting an option specific for this purpose. As currently specified (in the proposal), the attribute is confusing, complex, and will lead to interoperability issues. 

3.	Secret Data is considered to be a Cryptographic Object. The problem is that this may not always be the case. The client could register a password without using it for crypto purposes (such as key derivation). If Secret Data is considered to be a Crypto Object, certain attributes must always have a value. For example, if the password is just an ordinary password string and not used for key derivation, clients are still expected to set the Crypto Usage Mask to 00000000000000 (the spec specifies that this attribute must always have a value for Cryptographic Objects). This can get confusing and it would make sense to clarify this in the Usage Guide.

4.	Provide additional guidance and clarification on how to specify Custom Attributes in a message. If required, an example could be provided in the Usage Guide.

The spec currently states the following:
"The tag type Custom Attribute cannot identify the particular attribute; hence, such an attribute can only appear in an Attribute Structure with its name as defined in Section 2.1.1."
	
We should clarify that Attribute Name (as defined in Section 2.1.1) should be x-<custom attribute> or y-<custom attribute> and not Custom Attribute. The table should be updated accordingly.

5.	p.45, 4.2, line 774: Cryptographic Length should be a required attribute for the Create operation. It is required by Register, Create Key Pair, and Derive Key.

6.	p.49, 4.3: Should we provide guidance on how to register a key pair? During Create Key Pair, the server is expected to set certain attributes. For example, the link attribute will be set and certain attributes, such as Cryptographic Parameters, must be set to the same value for both public and private key. All this occurs transparently and clients may not realize that the server is setting these attributes. It may be helpful to include an example of registering a key pair using the Register operation in the Usage Guide.

7.	p.49, 4.4, line 831: States that during a re-key, attributes are changed similarly to performing a Revoke with Revocation Reason of Superseded. According to p. 34, line 562, issuing a Revoke with Revocation Reason other than Compromised causes the state to transition from Active to Deactivated state. This will not work for most clients. Protect Stop Date usually occurs prior to reaching the Deactivation Date. Clients would usually perform a rekey, expect the new key to be used for protect purposes, and continue using the old key for processing purposes until it reaches the Deactivation Date.

8.	p.54, 4.6: Three concerns regarding Certify and Re-certify:
a.	The table needs to include the CA to be used to sign the Certificate Request. We are currently assuming that there exists only one CA on the server, unless the Certificate Issuer attribute is supposed to be used for this purpose. If this is the case, we need to add clarification. If this is not the case, this needs to be discussed. Should we allow the CA name to be specified or do we require one of the following: CA public key UID, CA private key UID, or certificate UID?

b.	Creating a Certificate Request requires the private key. We are assuming that the client stores the key pair outside of the server and has the capability to create a Certificate Request. Should the protocol cover the case where the key pair is stored by the server and the Certificate Request can be created by the server? This would not only be more convenient for clients, but would also benefit dumb clients who do not have the capability to create Certificate Requests.

c.	Line 918 states that if the information in the Certificate Request conflicts with the attributes specified in Template-Attribute, the information in the Certificate Request takes precedence. Should we provide additional guidance/clarification in the Usage Guide? This may affect Certificate Subject, Cryptographic Algorithm, and Crypto Length. I am not sure what other attributes can be explicitly specified under Attributes inside Certificate Request.