Resource Public Key Infrastructure (RPKI) Trust Anchor Locator
draft-ietf-sidrops-https-tal-08

Thanks for the easy to read diff with RFC7730.  A minor nit:

(1) Section 2.1, Typo.  s/implementors/implementers/

Thank you for this well written document. I have a few relatively minor comments (and questions) which I think you should address:

1.  Introduction

   This document obsoletes [RFC7730] by adding support for HTTPS URIs in
   a TAL.

If this document obsoletes RFC 7730, then I think you need to have "Changes since RFC 7730" section (Is this a BIS document?). If it only updates it, then the above (and the obsolete header at the top of the draft) is not correct.

2.2.  Trust Anchor Locator File Format

   In this document we define a Trust Anchor URI as a URI that can be
   used to retrieved a current Trust Anchor certificate.  This URI MUST
   be either an rsync URI [RFC5781], or an HTTPS URI [RFC7230].

I think the first mention of URI still needs a reference to RFC 3986.

   The TAL is an ordered sequence of:

   1.  an optional comment section consisting of one or more lines each
       starting with the '#' character, followed by human readable
       informational UTF-8 text, and ending with a line break,

Unless you think you want to use ASCII and Unicode Control characters in this field, I think you should recommend usage of RFC 5198 here.

2.3.  TAL and Trust Anchor Certificate Considerations

   The trust anchor MUST contain a stable key.  This key MUST NOT change

How does "MUST contain a stable key" differ from "key MUST NOT change"?

   when the certificate is reissued due to changes in the INR
   extension(s), when the certificate is renewed prior to expiration, or
   for any reason other than a key change.

This reads funny: “you must not change the key unless you decide to change the key”. Maybe talk about key compromise and key strength no longer being adequate instead?

4.  HTTPS Considerations

   o  This protocol does not require the use of SRV-IDs.

   o  This protocol does not require the use of URI-IDs.

I suspect this was copied from another RFC, but "does not require" is not right here, as it doesn't prevent it as an option. I think you should change "does not require the use" to "does not use"

I realize that this document inherits the text in Section 3 from RFC 6490, but can you tell me why there are SHOULDs and not MUSTs?  Why would one NOT do it the way Section 3 specifies?

Then I’ll ask the same question for the new https text in Section 4, especially about TLS certificate and host name validation.

Thank you for keeping the diff from RFC 7730 tidy!

Abstract

   their CA certificate.  In particular it allows TAs to change the set
   of Internet Number Resources included in the RFC3779 extension of
   their certificate.

Neither "Internet Number" nor "Number Resources" appears in RFC 3779 that I
can see.  (On a quick skim, I'm still not sure if we mean AS number or IP
address/prefix.)

Section 2.1

   the trust anchor per se.  In the RPKI, certificates contain
   extensions that represent Internet Number Resources (INRs) [RFC3779].

(As above, I don't see INRs mentioned in RFC 3779.)

Since comments are new in this rev of TAL, do we want to caution consumers
that implementations may not necessarily support comments yet?

Section 2.3

   The trust anchor MUST contain a stable key.  This key MUST NOT change
   when the certificate is reissued due to changes in the INR
   extension(s), when the certificate is renewed prior to expiration, or
   for any reason other than a key change.

(This seems a bit tautological...)

   If an entity wishes to withdraw a self-signed CA certificate as a
   putative trust anchor, for any reason, including key rollover, the
   entity MUST remove the object from the location referenced in the
   TAL.

Certain classes of attacker could continue to publish the last-known
certificate as a trust anchor and prevent this withdrawl from taking
effect; we should probably cover that in the security considerations.

Section 2.4

We say that it's RECOMMENDED to have different domains (so as to get
different IP addresses) but this example shows only a single domain.

Section 4

   Note that a Man in the Middle (MITM) cannot produce a CA certificate
   that would be considered valid according to the process described in
   Section 3.  [...]

I think the key part is that the attacker cannot produce a *new* CA
certificate that differs from a legitimate one, but they can MITM the HTTPS
connection and present a legitimate (but potentially stale) CA certificate.

   o  DNS names in Repository Server certificates SHOULD NOT contain the
      wildcard character "*".

Would a Relying Party ever reject the HTTPS connection (and thus, the
delivered TA) if a wildcard certificate is presented for the HTTPS
connection?

Section 5

   This TAL does not directly provide a list of resources covered by the
   referenced self-signed CA certificate.  Instead, the RP is referred
   to the trust anchor itself and the INR extension(s) within this
   certificate.  This provides necessary operational flexibility, but it
   also allows the certificate issuer to claim to be authoritative for
   any resource.  Relying parties should either have great confidence in
   the issuers of such certificates that they are configuring as trust
   anchors, or they should issue their own self-signed certificate as a
   trust anchor and, in doing so, impose constraints on the subordinate
   certificates.

Are there any external databases that a RP could consult to affect the
decision of whether to believe that a TA should actually be claiming the
indicated resource(s)?  (It would be a bit silly, given that this is the
RPKI already, but still...)

Usually we recommend to have a "Changes since RFC7730" section in bis documents... however, maybe the changes are small enough in this doc that that is not needed.