An Internet Attribute Certificate Profile for Authorization
draft-ietf-pkix-3281update-05

[Ballot discuss]
If I understand well the current rules and the latest interpretation given by the Legal Counsel of the IETF, the code in Appendix B should include the (long version) copyright notice in "code" similarly to MIB modules and such.

[Ballot comment]
The diff between this and 3281 is pretty simple. The deletion of this sentence (near the end of page 32 of 3281 update) sticks out on casual inspection: "In this case, the digestedObjectType MUST be publicKey and the otherObjectTypeID field MUST NOT be present." I wanted to shine the light on it to make sure its deletion was intentional.

[Ballot discuss]
I don't think the following normative reference is stable enough:

[RFC3281-ERR] http://www.rfc-editor.org/errata_search.php?eid=302

1) The content behind that link can change significantly over time
2) The link, as expressed, puts a constraint (read burden for the maintainers) on how the errata_search tool works going forward.

Could the single correction in that errata be captured as a note in this document instead?

[Ballot comment]
In general I think this is a well written document and I was planning to vote "Yes" originally.
Here are some other comments that might be worth addressing:

1.2. AC Path Delegation

[...]

  Since the administration and processing associated with such AC
  chains is complex and the use of ACs in the Internet today is quite
  limited, this specification does NOT RECOMMEND the use of AC chains.

NOT RECOMMEND is not a RFC 2119 keyword ("NOT RECOMMENDED" is).


                        Figure 1: AC Exchanges

Having arrows with direction of flows would be helpful here




4.3.4. Authority Information Access

  The authorityInformationAccess extension, as defined in [PKIXPROF],
  MAY be used to assist the AC verifier in checking the revocation
  status of the AC.  Support for the id-ad-caIssuers accessMethod is
  NOT REQUIRED by this profile since AC chains are not expected.

NOT REQUIRED make it look as if this is an RFC 2119 contstruct, but it
isn't. I suggest rewording.


Also, a reference to section 4.2.2.1 of RFC 5280 is going to be useful here, as a search for "authorityInformationAccess" in RFC 5280 yields no result.


5. Attribute Certificate Validation

[...]

  Support for an extension in this context means:

  1. The AC verifier MUST be able to parse the extension value.

  2. Where the extension value SHOULD cause the AC to be rejected, the

Why use of SHOULD here? Suggest not using RFC 2119 in this case. I.e.
"Where the extension value causes the AC ...".

      AC verifier MUST reject the AC.


7.1. Attribute Encryption

[...]

  The procedure for an AC issuer when encrypting attributes is
  illustrated by the following (any other procedure that gives the same
  result MAY be used):

    1.    Identify the sets of attributes that are to be encrypted for
          each set of recipients.
    2.    For each attribute set which is to be encrypted:
        2.1. Create an EnvelopedData structure for the data for this
              set of recipients.
        2.2. Encode the ContentInfo containing the EnvelopedData as a
              value of the encAttrs attribute.
        2.3. Ensure the cleartext attributes are not present in the
              to-be-signed AC.
    3.  Add the encAttrs (with its multiple values) to the AC.

  Note that there may be more than one attribute of the same type (the
  same OBJECT IDENTIFIER) after decryption.  That is, an AC MAY contain
  the same attribute type both in clear and in encrypted form (and
  indeed several times if the same recipient is associated with more
  than one EnvelopedData).

I admit I might be confused here, but I have 2 questions:

1). Is this MAY requirement only stated in this section? I can't think of
a case when both an encrypted and an unencrypted attribute of the same type are present. Note that I can see a case when several encrypted attributes of the same type are present.

2). Shouldn't the text in () talk about *different* recipients?

  One approach implementers may choose, would
  be to merge attribute values following decryption in order to re-
  establish the "once only" constraint.


In section 8:

  There is often a requirement to map between the authentication
  supplied by a particular security protocol (e.g. TLS, S/MIME) and the
  AC holder's identity.  If the authentication uses PKCs, then this
  mapping is straightforward.  However, it is envisaged that ACs will
  also be used in environments where the holder may be authenticated
  using other means.  Implementers SHOULD be very careful in mapping
  the authenticated identity to the AC holder.

Can you elaborate on what problem the last sentence in Section 8
is trying to address? Is this referring to insecure mappings, or something else?

[Ballot discuss]
(Updated the DISCUSS portion)

4.2. Profile of Standard Fields

  GeneralName offers great flexibility.  To achieve interoperability,
  in spite of this flexibility, this profile imposes constraints on the
  use of GeneralName.

  Conforming implementations MUST be able to support the dNSName,
  directoryName, uniformResourceIdentifier, and iPAddress options. This
  is compatible with the GeneralName requirements in [PKIXPROF] (mainly
  in section 4.2.1.6).

What about SRVName define in:
  [X509-SRV]
              Santesson, S., "Internet X.509 Public Key Infrastructure
              Subject Alternative Name for Expression of Service Name",
              RFC 4985, August 2007.
?

Many Apps protocols are starting to recommend use of SRVName, so having a MUST/SHOULD here about them would improve consistency across protocols.

In Appendix A Object Identifiers:

  This (normative) appendix lists the new object identifiers which are
  defined in this specification.  Some of these are required only for
  support of optional features and are not required for conformance to
  this profile.  This specification mandates support for OIDs which
  have arc elements with values that are less than 2^32, (i.e. they
  MUST be between 0 and 4,294,967,295 inclusive) and SHOULD be less
  than 2^31 (i.e. less than or equal to 2,147,483,647).  This allows
  each arc element to be represented within a single 32 bit word.
  Implementations MUST also support OIDs where the length of the dotted
  decimal (see [LDAP], section 4.1.2) string representation can be up

RFC 4510 ([LDAP]) doesn't have section 4.1.2.
I think you want to reference Section 1.4 in RFC 4512.

In Appendix B ASN.1 Module:

  NOTE: The value for TBA will be included during AUTH48.

  //** RFC Editor: Remove this note prior to publication **//

  PKIXAttributeCertificate-2008 { iso(1) identified-organization(3)
    dod(6) internet(1) security(5) mechanisms(5) pkix(7) id-mod(0)
    id-mod-attribute-cert2(TBA) }

But the IANA section says that no IANA action is required.
Where is this TBA coming from?

[Ballot comment]
In general I think this is a well written document and I was planning to vote "Yes" originally.
Here are some other comments that might be worth addressing:

1.2. AC Path Delegation

[...]

  Since the administration and processing associated with such AC
  chains is complex and the use of ACs in the Internet today is quite
  limited, this specification does NOT RECOMMEND the use of AC chains.

NOT RECOMMEND is not a RFC 2119 keyword ("NOT RECOMMENDED" is).


                        Figure 1: AC Exchanges

Having arrows with direction of flows would be helpful here




4.3.4. Authority Information Access

  The authorityInformationAccess extension, as defined in [PKIXPROF],
  MAY be used to assist the AC verifier in checking the revocation
  status of the AC.  Support for the id-ad-caIssuers accessMethod is
  NOT REQUIRED by this profile since AC chains are not expected.

NOT REQUIRED make it look as if this is an RFC 2119 contstruct, but it
isn't. I suggest rewording.


Also, a reference to section 4.2.2.1 of RFC 5280 is going to be useful here, as a search for "authorityInformationAccess" in RFC 5280 yields no result.


5. Attribute Certificate Validation

[...]

  Support for an extension in this context means:

  1. The AC verifier MUST be able to parse the extension value.

  2. Where the extension value SHOULD cause the AC to be rejected, the

Why use of SHOULD here? Suggest not using RFC 2119 in this case. I.e.
"Where the extension value causes the AC ...".

      AC verifier MUST reject the AC.


7.1. Attribute Encryption

[...]

  The procedure for an AC issuer when encrypting attributes is
  illustrated by the following (any other procedure that gives the same
  result MAY be used):

    1.    Identify the sets of attributes that are to be encrypted for
          each set of recipients.
    2.    For each attribute set which is to be encrypted:
        2.1. Create an EnvelopedData structure for the data for this
              set of recipients.
        2.2. Encode the ContentInfo containing the EnvelopedData as a
              value of the encAttrs attribute.
        2.3. Ensure the cleartext attributes are not present in the
              to-be-signed AC.
    3.  Add the encAttrs (with its multiple values) to the AC.

  Note that there may be more than one attribute of the same type (the
  same OBJECT IDENTIFIER) after decryption.  That is, an AC MAY contain
  the same attribute type both in clear and in encrypted form (and
  indeed several times if the same recipient is associated with more
  than one EnvelopedData).

I admit I might be confused here, but I have 2 questions:

1). Is this MAY requirement only stated in this section? I can't think of
a case when both an encrypted and an unencrypted attribute of the same type are present. Note that I can see a case when several encrypted attributes of the same type are present.

2). Shouldn't the text in () talk about *different* recipients?

  One approach implementers may choose, would
  be to merge attribute values following decryption in order to re-
  establish the "once only" constraint.


In section 8:

  There is often a requirement to map between the authentication
  supplied by a particular security protocol (e.g. TLS, S/MIME) and the
  AC holder's identity.  If the authentication uses PKCs, then this
  mapping is straightforward.  However, it is envisaged that ACs will
  also be used in environments where the holder may be authenticated
  using other means.  Implementers SHOULD be very careful in mapping
  the authenticated identity to the AC holder.

Can you elaborate on what problem the last sentence in Section 8
is trying to address?

[Ballot discuss]
In Appendix A Object Identifiers:

  This (normative) appendix lists the new object identifiers which are
  defined in this specification.  Some of these are required only for
  support of optional features and are not required for conformance to
  this profile.  This specification mandates support for OIDs which
  have arc elements with values that are less than 2^32, (i.e. they
  MUST be between 0 and 4,294,967,295 inclusive) and SHOULD be less
  than 2^31 (i.e. less than or equal to 2,147,483,647).  This allows
  each arc element to be represented within a single 32 bit word.
  Implementations MUST also support OIDs where the length of the dotted
  decimal (see [LDAP], section 4.1.2) string representation can be up

RFC 4510 ([LDAP]) doesn't have section 4.1.2.


In Appendix B ASN.1 Module:

  NOTE: The value for TBA will be included during AUTH48.

  //** RFC Editor: Remove this note prior to publication **//

  PKIXAttributeCertificate-2008 { iso(1) identified-organization(3)
    dod(6) internet(1) security(5) mechanisms(5) pkix(7) id-mod(0)
    id-mod-attribute-cert2(TBA) }

But the IANA section says that no IANA action is required.
Where is this TBA coming from?

1.a - Steve Kent is the Shepherd. I have personally reviewed the
document and assert that it is ready for IESG publication.

1.b - The document has been reviewed by key WG members. There are no
concerns about depth or breadth of the reviews.

1.c - I see no need for wider review.

1.d - There are no specific concerns of which the AD and/or IESG should
be aware.

1.e - The WG consensus is solid.

1.f - There has been no threat of an appeal by an WG members.

1.g - I have personally verified that the document satisfies all ID
nits (although it does generate several spurious warnings).

1.h - The document splits it references into normative and informative
as required.

1.i - The document has an IANA consideration and it is consistent with
the main body (there are no IANA considerations).

1.j - Sean Truner assures me that the ASN.1 has been verified :-).

1.k - Write-up is as follows:

Technical Summary

This specification defines a profile for the use of X.509 Attribute
Certificates in Internet Protocols. Attribute certificates may be
used in a wide range of applications and environments covering a
broad spectrum of interoperability goals and a broader spectrum of
operational and assurance requirements. The goal of this document is
to establish a common baseline for generic applications requiring
broad interoperability as well as limited special purpose
requirements. The profile places emphasis on attribute certificate
support for Internet electronic mail, IPsec, and WWW security
applications. This document obsoletes RFC 3281.

Working Group Summary

This ID was discussed on the mailing list and at multiple meetings.
It is a simple update to address known errata posted to the RFC
editor and to address an issue identified as a result of developing
the draft-ietf-pkix-authorityclearanceconstraints ID. The clearance
attribute object identifier didn't match the attribute's X.509 object
identifier.

Document Quality

This document is a short update of an existing draft and is comparable
in quality to its predecessor. Note that most of the changes listed
in Appendix C of the document are a result of the long lag time
between updates (e.g., reference updates).

Personnel

Steve Kent is the document Shepherd. Tim Polk is the responsible
Security Area AD.