Re: Eric Rescorla's Discuss on draft-ietf-httpbis-expect-ct-07: (with DISCUSS and COMMENT)

[Eric Rescorla <[e-mail]>]

Following up, I don't see any response to these comments.

-Ekr


On Thu, Sep 13, 2018 at 7:22 AM Eric Rescorla <[e-mail]> wrote:

> I have one other non-blocking comment: Why is this document Experimental?
> People are already deploying CT without this. It seems like PS would make
> more sense or Informational.
>
> Alexey, I leave it to you.
> -Ekr
>
>
> On Wed, Sep 12, 2018 at 4:12 PM, Eric Rescorla <[e-mail]> wrote:
>
>> Eric Rescorla has entered the following ballot position for
>> draft-ietf-httpbis-expect-ct-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-httpbis-expect-ct/
>>
>>
>>
>> ----------------------------------------------------------------------
>> DISCUSS:
>> ----------------------------------------------------------------------
>>
>> Rich version of this review at:
>> https://mozphab-ietf.devsvcdev.mozaws.net/D4579
>>
>>
>> This generally seems like a sound mechanism, but I believe there are
>> some points here that are sufficiently unclear they might create
>> interop problems,s o I am balloting DISCUSS.
>>
>> Most importantly, this document just says you support CT, but that
>> creates a potential interop problem if say 6962-tris had a different
>> way of delivering CT information or a different syntax. I'm not saying
>> you need a version here, but you need to indicate that it's not
>> forward-looking.
>>
>> Also, see below.
>>
>> DETAIL
>> S 2.4.
>> >      beginning an HTTP conversation over the TLS channel.
>> >
>> >      If a connection to a Known Expect-CT Host violates the UA's CT
>> policy
>> >      (i.e., the connection is not CT-qualified), and if the Known
>> Expect-
>> >      CT Host's Expect-CT metadata indicates an "enforce" configuration,
>> >      the UA MUST treat the CT compliance failure as an error.
>>
>> Is this supposed to be a hard failure, as with HSTS. If not, how does
>> it interact with HSTS's hard failure reqs.
>>
>>
>> S 3.1.
>> >         (This may differ from the value of the
>> "served-certificate-chain"
>> >         key.)  The value is provided as an array of strings, which MUST
>> >         appear in the order matching the chain that the UA validated;
>> each
>> >         string in the array is the Privacy-Enhanced Mail (PEM)
>> >         representation of each X.509 certificate as described in
>> >         [RFC7468].
>>
>> What happens if you try to construct multiple paths?
>>
>>
>> S 3.1.
>> >            does not have or does not trust the public key of the log
>> from
>> >            which the SCT was issued), "valid" (indicating that the UA
>> >            successfully validated the SCT as described in Section 5.2 of
>> >            [RFC6962] or Section 8.2.3 of [I-D.ietf-trans-rfc6962-bis]),
>> or
>> >            "invalid" (indicating that the SCT validation failed because
>> >            of, e.g., a bad signature).
>>
>> Is "invalid" anything other than the specific cases listed above?
>>
>>
>> ----------------------------------------------------------------------
>> COMMENT:
>> ----------------------------------------------------------------------
>>
>>
>> >      allows web host operators to instruct user agents to expect valid
>> >      Signed Certificate Timestamps (SCTs) to be served on connections to
>> >      these hosts.  Expect-CT allows web host operators to discover
>> >      misconfigurations in their Certificate Transparency deployments and
>> >      ensure that misissued certificates accepted by UAs are discoverable
>> >      in Certificate Transparency logs.
>>
>> I don't believe that it does this. Consider a client which simply did
>> not support CT, then it would (a) accept a misissued certificate that
>> (b) was not discoverable
>>
>>
>> S 2.1.1.
>> >
>> >               Figure 2: Syntax of the report-uri directive value
>> >
>> >      "absolute-URI" is defined in Section 4.3 of [RFC3986].
>> >
>> >      Hosts may set "report-uri"s that use HTTP or HTTPS.  If the scheme
>> in
>>
>> Why are you allowing HTTP?
>>
>>
>> S 2.3.2.
>> >         the "enforce", "max-age", or "report-uri" header field value
>> >         directives convey information different from that already
>> >         maintained by the UA.  If the "max-age" directive has a value of
>> >         0, the UA MUST remove its cached Expect-CT information if the
>> host
>> >         was previously noted as a Known Expect-CT Host, and MUST NOT
>> note
>> >         this host as a Known Expect-CT Host if it is not already noted.
>>
>> As noted above, I think you need to clear the cache when you upgrade
>> to a potentially incompatible CT version, or otherwise reconfigure the
>> client.
>>
>>
>> S 2.3.2.1.
>> >         this host as a Known Expect-CT Host if it is not already noted.
>> >
>> >   2.3.2.1.  Noting Expect-CT
>> >
>> >      Upon receipt of the Expect-CT response header field over an error-
>> >      free TLS connection (including the validation adding in Section
>> 2.4),
>>
>> s/adding/added/?
>>
>>
>> S 2.3.2.1.
>> >      host's domain name and its associated Expect-CT directives in non-
>> >      volatile storage.
>> >
>> >      To note a host as a Known Expect-CT Host, the UA MUST set its
>> Expect-
>> >      CT metadata given in the most recently received valid Expect-CT
>> >      header field, as specified in Section 2.3.2.2.
>>
>> This seems ungrammatical. Set it where?
>>
>>
>> S 2.3.2.2.
>> >
>> >   2.3.2.2.  Storage Model
>> >
>> >      If the substring matching the host production from the Request-URI
>> >      (of the message to which the host responded) does not congruently
>> >      match an existing Known Expect-CT Host's domain name, per the
>>
>> I would say "exactly match" rather than "congruently match" unless
>> this ia term of art somewhere.
>>
>>
>> S 2.3.2.2.
>> >      understands them, the UA MAY note them as well.
>> >
>> >      UAs MAY set an upper limit on the value of max-age, so that UAs
>> that
>> >      have noted erroneous Expect-CT hosts (whether by accident or due to
>> >      attack) have some chance of recovering over time.  If the server
>> sets
>> >      a max-age greater than the UA's upper limit, the UA MAY behave as
>> if
>>
>> This MAY seems out of place, given that you already said MAY.
>>
>>
>> S 2.4.
>> >
>> >      When a UA connects to a Known Expect-CT Host using a TLS
>> connection,
>> >      if the TLS connection has no errors, then the UA will apply an
>> >      additional correctness check: compliance with a CT Policy.  A UA
>> >      should evaluate compliance with its CT Policy whenever connecting
>> to
>> >      a Known Expect-CT Host, as soon as possible.  However, the check
>> can
>>
>> What does "as soon as possible" mean?
>>
>>
>> S 2.4.
>> >      terminates the connection due to an Expect-CT failure, this could
>> >      cause the UA to skip subsequent correctness checks.  When the CT
>> >      compliance check is skipped or bypassed, Expect-CT reports
>> >      (Section 3) will not be sent.
>> >
>> >      When CT compliance is evaluted for a Known Expect-CT Host, the UA
>>
>> Nit: evaluated
>>
>>
>> S 2.4.1.
>> >      "report-uri" (Section 3).
>> >
>> >   2.4.1.  Skipping CT compliance checks
>> >
>> >      It is acceptable for a UA to skip CT compliance checks for some
>> hosts
>> >      according to local policy.  For example, a UA may disable CT
>>
>> Should this be MAY?
>>
>>
>> S 3.1.
>> >
>> >      o  "scts": the value represents the SCTs (if any) that the UA
>> >         received for the Expect-CT host and their validation statuses.
>> >         The value is provided as an array of JSON objects.  The SCTs may
>> >         appear in any order.  Each JSON object in the array has the
>> >         following keys:
>>
>> So these just apply to the EE cert? What about CT for the non-EE
>> certs?
>>
>>
>>
>>
>> S 3.1.
>> >            of, e.g., a bad signature).
>> >
>> >         *  The "source" key, with a string value that indicates from
>> where
>> >            the UA obtained the SCT, as defined in Section 3 of [RFC6962]
>> >            and Section 6 of [I-D.ietf-trans-rfc6962-bis].  The UA MUST
>> set
>> >            the value to one of "tls-extension", "ocsp", or "embedded".
>>
>> What do these mean? They seem obvious, but you don't say.
>>
>>
>>
>