Folks,
Based on feedback from Randy, Geoff and Andrei, plus comments from Sean,
we plan to make the following changes to the CP:
- 1.2: we will delete the 3rd paragraph, but keep the first, as it
parallels the Arch doc (as Sean noted). Randy, can you elaborate on
the nature of current IANA allocations made to non-RIRs? Legacy
allocations are not addressed in this paragraph, just the current
allocation system.
- 1.4.1: we will remove the reference to PI space here (i.e., delete
the last sentence in the second paragraph).
- 1.6.3: we will delete this (equivalent to "not applicable")
section, since 3647 really focuses on the suitability of a CPS
relative to the policy expressed by a CP.
Andrei: I do not agree with your suggestion to remove the CPS MUSTs
in the CP. The intent is to help ensure that each CA (as amended
below, in 2.2) has a CPS, and that the CPS covers certain basic info.
If each parent registry were to act as an approver for a subordinate
CPS, this would not encourage RPKI-wide uniformity re CPS's.
- 1.7: we will remove the definition of PI space from this section.
- 2.2: it is the norm in a CP to require every CA covered by
the CP to publish a CPS. For orgs that avail themselves of managed CA
services, that burden is accepted by the supplier of that service.
However, we agree that a CA in the RPKI that does not issue
certificates to other subscribers really need not have to publish a
CPS (since it has no "subjects" outside of its own administrative
domain). So, how about the following change:
OLD : Each CA MUST publish its CPS.
NEW: Each CA that issues certificates to entities outside
of its administrative domain MUST create and publish a CPS that meets
the requirements set forth in this CP. Publication
means that the entities to which the CA issues certificates MUST
be able to acquire a copy of the CPS, and MUST be able to
ascertain when the CPS changes. (An organization that does not
allocate or assign INRs need not create nor publish a CPS.)
- 3.1.2: I disagree with Randy's recollection of the history of this
requirement. First it was no meaningful names for any subject other than IANA
and the RIRs. Then, at Randy's request (a reasonable one :-)) it was
changed to the current form, i.e., NO meaningful names for any
subject. It's now uniform and egalitarian. If, in Randy's question,
"uu" is uunet, then the answer is no, they ought not issue
certificates with customer subject names.
Geoff had a different concern here, i.e., that the text was
not clear. We propose the following change:
OLD: The Subject name in each certificate does not need to be meaningful.
The certificates issued under this PKI are used for authorization in
support of applications that make use of attestations of Internet
resource holdings, not for identification. The PKI binds a public
key to sets of INRs. The name of the holder of the INR is not
intended to be "meaningful".
NEW: The Subject name in each certificate SHOULD not be "meaningful", i.e.,
the name is NOT intended to convey the identity of the Subject
to relying parties. The rationale here is that certificates
issued under
this PKI are used for authorization in support of
applications that make
use of attestations of INRs. They are not used for to
identify Subjects.
- 3.1.5: we will add a reference to the cert profile here. That doc
includes an example of how to generate (meaningless) subject names in
a way that minimizes the likelihood of collisions. I understand that
this is what several instances of RIR software do anyway.
- 3.1.6: The new text re trademarks is typical for a CP, but it does
seem a bit harsh. The Subject name generation mechanisms described
in the
revised cert profile doc makes it unlikely that one would generate a
name that would conflict with a trademark, but nothing is impossible.
How about the following:
OLD: When generating subject names, CAs should use a mechanism that
is unlikely to result in Subject names that include trademarks,
service marks, etc. CAs MUST NOT knowingly issue a certificate with
a Subject name that a court of competent jurisdiction has determined
infringes on the trademark of another individual or organization.
NEW: When generating subject names, CAs SHOULD use a mechanism (such as
those described [cert-profile]) to minimize the likelihood that
the subject name is a trademark, service mark, etc. There is no
requirement that a CA perform a search to verify that a subject name
generated in this fashion does not infringe on a trademark, service
mark, etc.
- 4.10: The intent of this paragraph is to establish a minimum
interval for an RP to check the repository system (not a local
copy thereof), and to recommend a max rate of checking. The minimum
interval is tied to the timing established in Geoff's key rollover
document, the "staging period." The max is also a suggestion from
Geoff. I'm open to a better wording here, but, like, you, I don't
see a good fix.
- 3.2.3: The purpose of this section is to require a CA to have a
means of identifying and authenticating a person who acts as a
representative for the organization that is responsible for the INRs
for which the RPKI cert is issued. This is a typical requirement for
a CA, i.e., there needs to be someone at an organization who acts as
the liaison to the CA. Geoff is right that there is no explicit
mention of whether the liaison info is current. However, the final
sentence establishes the standard for satisfying this requirement,
i.e., whatever the organization operating the CA already does for INR
relationships.
- 4.6.1: Geoff, by definition, certificate renewal does not entail a
name change, so your first comment is not applicable. To clarify that
we want the new certificate to be issued at least 24 hours before
expiration of the old certificate, we propose the following change:
OLD: An overlap period of at least 24 hours is RECOMMENDED.
NEW: It is RECOMMENDED that the renewed certificate be issued and
published at least 24 hours prior to the expiration of the
certificate that it replaces.
- 4.8.1: Here is our proposed text modification:
OLD: Modification of a certificate occurs to implement changes to
selected attribute values in a certificate. In the context of this
PKI, the only changes that are accommodated by certificate
modification are additions to the INR holdings described by the RFC
3779 extension.
When previously distributed INRs are removed from a certificate,
then the old certificate MUST be revoked and a new certificate MUST
be issued.
When a certificate modification is approved, a new certificate is
issued. The new certificate MUST contain the same public key and the
same expiration date as the original certificate, but with the
incidental information corrected and/or the INRs expanded.
Revocation of the previous certificate is not required.
NEW: Modification of a certificate occurs to implement changes to
selected attribute values in a certificate. In the context of the
RPKI, the only changes that are accommodated by certificate
modification are changes to the INR holdings described by the RFC
3779 extension and changes to the SIA extension.
When a certificate modification is approved, a new certificate is
issued. If no INR holdings are removed from the certificate, the new
certificate MUST contain the same public key and the same expiration
date as the original certificate (but with the SIA extension and/or
and/or the INR set expanded). In this case, revocation of the previous
certificate is not required.
When previously distributed INRs are removed from a certificate,
then the old certificate MUST be revoked and a new certificate MUST
be issued, reflecting the changed INR holdings. (The SIA extension
also may be changed during this action, if requested.)
- 5.1: These are possible topics; their inclusion in a CPS is not
mandated. Also, , I reworded 2.2 to make it clear that a
non-registry/ISP need not publish a CPS, so the CPS mandates affect
fewer folks.
- 5.2: How about the following rewording:
OLD: Each CA MUST maintain procedural security controls that are
commensurate with those employed by the organization in the
management of INR distribution. The details for each CA MUST be
specified in its CPS. Topics covered SHOULD include:
NEW: Each CA MUST maintain procedural security controls that are
commensurate with those employed by the organization in the
management of INR distribution. The procedural security controls
employed for CA operation MUST be specified in its CPS. Possible
topics to be covered in the CPS include:
- 5.4.1: I think Sean addressed Randy's question here.
- 5.6: I'll reference the key rollover document and delete the second
sentence, I suggest the following text:
OLD: When a CA wishes to change keys, it MUST acquire a new certificate
containing its new public key, well in advance of the scheduled
change. This certificate SHOULD have a validity period that is at
least as long as that of any certificate being issued under that
certificate.
NEW: When a CA wishes to change keys, it MUST acquire a new certificate
containing its new public key. See [key-rollover] for a description
of how key changeover is effected on the RPKI.
- 6.3.2: Randy and Geoff do not agree on whether/how to change this.
I will delete the offending sentence, since it was not precise enough
to be normative anyway. Also, the next sentence is not consistent
with what we said in 4.8.1. I propose the following text for the 1st
paragraph:
OLD: The INRs held by a CA may periodically change because it receives
new distributions. Therefore it is appropriate for its certificate
to have a lifetime that matches the periodicity of these changes.
However, to minimize disruption, the key pairs SHOULD be maintained
across certificate changes (if no INRs are removed as a result of
the certificate change).
NEW: The INRs held by a CA may periodically change when it receives
new distributions. To minimize disruption, the CA key MUST NOT
change when INRs are added to it's certificate.
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr