[regext] Benjamin Kaduk's Discuss on draft-ietf-regext-data-escrow-07: (with DISCUSS and COMMENT)

[Benjamin Kaduk via Datatracker <[e-mail]>]

Benjamin Kaduk has entered the following ballot position for
draft-ietf-regext-data-escrow-07: Discuss

When responding, please keep the subject line intact and reply to all
email addresses included in the To and CC lines. (Feel free to cut this
introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-regext-data-escrow/



----------------------------------------------------------------------
DISCUSS:
----------------------------------------------------------------------

Let's have a discussion about the overall plans for providing guidance
on mechanisms for (e.g.) cryptographic confidentiality and integrity
protection of data both in transit and at-rest,
authentication/authorization requirements, etc..  In particular, what
it's appropriate and necessary to include in this document vs. other
documents, and how to provide some specific general baseline guidance
that can be used in the absence of conflictinc scenario-specific
deployment requirements.  Some further details in the COMMENT section,
but in general, it seems like there should be some "off-the-shelf"
mechanism that can be used without a need for every deployment to do a
custom thing, for cases where there are not custom requirements.  It may
not need to be part of this document, but we should have a plan for
where it will be.

Also, it's not clear to me whether we expect escrow of
credentials/verifiers used to authenticate/authorize fourth parties that
perform operations (e.g., registrations) at the registry (see comment on
Section 3), and that is pretty important for knowing what security
requirements to place on the escrow'd data.


----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

I agree with Roman that the link to the charter is tenuous; perhaps this
is a "data format for files exchanged between registration entities that
needs extraction from EPP or RDAP", but it's not a perfect fit.

Section 1

   The goal of data escrow is higher resiliency of registration
   services, for the benefit of Internet users.  The beneficiaries of a
   registry are not just those registering information there, but all
   relying parties that need to identify the owners of objects.

Only relying parties that need to identify *owners* specifically (as
opposed to consuming other registered data)?

   In the context of domain name registries, registration data escrow is
   a requirement for generic top-level domains and some country code
   top-level domain managers are also currently escrowing data.  There
   is also a similar requirement for ICANN-accredited domain registrars.

Are there easy references for these requirements being requirements?

Section 2

   Watermark.  If the Timeline Watermark of an Incremental Deposit were
   to cover the Timeline Watermark of another (Incremental or
   Differential) Deposit since the last Full Deposit, the more recent
   deposit MUST contain all the transactions of the earlier deposit.

Do we define what it means for one Timeline Watermark to "cover" another?

Section 3

   Specifications covering the objects used by registration
   organizations shall identify the format and contents of the deposits
   a registry has to make, such that a different registry would be able
   to rebuild the registration services of the former, without its help,
   in a timely manner, with minimum disruption to its users.

Rebuilding registration *services* as opposed to just *operation* sounds
like it will require also escrowing credentials/credential verifiers
used to authenticate to the registration organization in order to make
use of such services.  Such credentials would of course need very
careful protection both in transit and at rest, in order to avoid
compromising the integrity of the primary operations of the current
registration organization.

   Given the requirement for confidentiality and the importance of
   accuracy of the information that is handled in order to offer
   registration services, parties using this specification shall define
   confidentiality and integrity mechanisms for handling the
   registration data.

Do we have any recommendations/examples of such mechanisms in the
pipeline?

   Specifications covering the objects used by registration
   organizations shall not include in the specification transient
   objects that can be recreated by the new registry, particularly those
   of delicate confidentiality, e.g., DNSSEC KSK/ZSK private keys.

I'm not sure this is going to be terribly helpful guidance for
non-domain-name data.  I am acutely aware that cryptographic (symmetric
and/or private) keys are not the only type of data that this could apply
to, but it is the only type that I know of a concise/precise description
for.  E.g., "cryptographic key material that is vital to the integerity
and/or security of operation of the registry and the systems that rely
on it but that is not vital to the continuity of operations (e.g.,
DNSSEC KSK/ZSK private keys)".

Also, I would assume that the authenticatoin credentials I mention above
would not qualify for this requirement, since they are not things that
can be recreated by the new registry.

   Details that are a matter of policy should be identified as such for
   the benefit of the implementers.

(For human consumption or machine consumption?)

Section 4

Could we maybe use the section title "Conventions Used in This
Document"?  "General Conventions" could be interpreted as intended for a
broader scope, though based on (e.g.) RFC 8748 I don't think that's the
intent.

Section 5

   registry.  The deposits are represented in XML.  Only the format of
   the objects deposited is defined, nothing is prescribed about the
   method used to transfer such deposits between the registry and the
   escrow agent or vice versa.

nit: comma splice.

Also, we said earlier that "parties using this specification shall
define confidentiality and integrity mechanisms for handling the
registration data" without, AFAICT, conditions on when that should be
done.  So in some sense we do prescribe properties of the method used
for transfer, if I'm reading that correctly.

Section 5.1.1

   A REQUIRED <watermark> element contains the data-time corresponding
   to the Timeline Watermark of the deposit.

I will offer a third opinion, that this was meant to be dateTime.

Section 5.1.2

I'm confused about the relationship between the <objURI> elements in the
<rdeMenu> and the objects in the <deletes> and <contents> elements -- is
<rdeMenu> just a list of "here are all objects that this update
touchers", with the details in the other sections?

Section 5.1.4

   If an object is present in the <contents> section of several deposits
   (e.g.  Full and Differential) the registry data from the latest
   deposit (as defined by the Timeline Watermark) SHOULD be used when
   rebuilding the registry.

Why is this not a MUST?

Section 6.1

Why is there a 13-character maximum for the deposit ID?
But client IDs range from 3 to 16 characters (which seems kind of short
as a maximum, especially if a registrar wants to use a domain name as an
identifier)?

    <!-- A RDE version number is a dotted pair of decimal numbers -->
    <simpleType name="versionType">
      <restriction base="token">
        <pattern value="[1-9]+\.[0-9]+"/>
        <enumeration value="1.0"/>
      </restriction>
    </simpleType>

I assume the <pattern> is there to give the overall structure that
subsequently defined <enumeration>s will adhere to?

I see the rrType defined but not used anywhere; is it still needed in
this document?

Section 10

   Authentication of the parties passing data escrow deposit files is
   also of the utmost importance.  The escrow agent SHOULD properly
   authenticate the identity of the registry before accepting data
   escrow deposits.  In a similar manner, the registry SHOULD
   authenticate the identity of the escrow agent before submitting any
   data.

I am failing to come up with a scenario in which these SHOULDs would be
violated; should they be MUSTs instead?  (I'd like for the "encrypting
the data" in the previous paragraph to be a MUST as well, but that case
is less clear-cut.)

   Additionally, the registry and the escrow agent SHOULD use integrity
   checking mechanisms to ensure the data transmitted is what the source
   intended.  Validation of the contents by the escrow agent is

It seems like if there are no integrity checks then the escrow mechanism
is not really fit for purpose -- can't this be MUST?

Section 11

   This specification defines a format that may be used to escrow
   personal data.  The process of data escrow is governed by a legal
   document agreed by the parties, and such legal document must regulate
   the particularities regarding the protection of personal data.

"Regulate the particularities" is an interesting phrase, as it merely
specifies that some restrictions be made, but nothing about the nature
of them.  Shouldn't we be saying something like "ensure that
privacy-sensitive and/or personal data receives appropriate protection"?

Section 14, 15, 16

Could we not reuse the same deposit-id for different examples?
(Also, using 20191017001 with a watermark date of 2019-10-18
midnight-UTC is perhaps unnecessary cognitive dissonance.)

Is it appropriate to use the urn:ietf:params:xml:ns:rdeObj[N] namespaces
in the examples, vs. a dedicated "example" namespace?