Re: [sidr] Last Call: <draft-ietf-sidr-rpsl-sig-10.txt> (Securing RPSL Objects with RPKI Signatures) to Proposed Standard

[Brian Haberman <[e-mail]>]

Hi Tom,
     Thanks for the in-depth review and your efforts in creating another
implementation of this draft. Responses to your comments are below...

On 4/28/16 6:54 PM, Tom Harrison wrote:
> Section 5 requires that an EE certificate be used for the signing of
> the RPSL object.  An EE certificate must contain an SIA extension that
> points to an RPKI signed object (RFC 6487 [4.8.8.2]).  The draft does
> not define a profile for a new type of object, or specify an existing
> one that may be used instead.  There are a number of ways to deal with
> this: for example, by defining a new profile and changing the
> signature URL to suit, or by amending RFC 6487 such that object
> pointers in EE certificates are optional.

I would propose adding some text to this draft (probably as a
sub-section in section 2) that says that the SIA defined in RFC 6487 is
omitted when a certificate is used to sign RPSL objects. Given the
single-use nature of the key-pair (section 3.2, point #1), omitting the
SIA is straightforward.

> 
> Section 4 specifies sets of attributes that must be signed.  'org' is
> included as one of these attributes for the as-block, inet[6]num, and
> route[6] object types.  However, 'org' is not defined in either of the
> principal RPSL RFCs (2622 and 4012), and there are current
> implementations (e.g. APNIC's) that do not support it.  I think the
> references to 'org' should be omitted.

Agreed. I will remove 'org' from the listed objects.

> 
> Section 4 specifies 'signature' as an attribute that must be signed.
> 'signature' can appear multiple times in a single object, where e.g.
> two different resource holders sign a route[6] object.  Given that the
> text doesn't explicitly state that only the newest 'signature' must be
> signed, it would appear to require that any extant 'signature'
> attributes be signed as well.  That in turn would prevent previous
> signers from re-signing the object independently of the subsequent
> signers.  I think the text should be changed so that only the new
> signature attribute need be signed.

Agreed. I will update the text to explicitly refer to the signature
currently under construction.

> 
> Section 2.4 requires that "the Internet number resources present in
> [RFC3779] extensions of the certificate referred to in the "c" field
> of the signature must cover the resources in the primary key of the
> object".  This means that it's not possible to sign a route[6] object
> for a route where one resource holder has the ASN and another the
> prefix.  In revision 8 (and earlier), the possibility of there being
> multiple signatures, each with a certificate covering a subset of the
> primary key resources, was explicitly permitted.  I think that the
> previous text here should be restored.

I agree that the original text allowing multiple signatures supports the
case where the components of the primary key of the object (i.e.,
prefix+ASN) come from different resource holders. I will restore that text.

> 
> (The above points were the product of much discussion of this draft
> with Tim and Oleg from RIPE, not that I'm speaking for them.  We were
> able to write interoperable prototype signer/validator
> implementations, so the document is in pretty good shape on the
> whole.)

Thanks! Glad to hear this.

Regards,
Brian