[Ecrit] Benjamin Kaduk's Discuss on draft-ietf-ecrit-data-only-ea-21: (with DISCUSS and COMMENT)

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

Benjamin Kaduk has entered the following ballot position for
draft-ietf-ecrit-data-only-ea-21: 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-ecrit-data-only-ea/



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

I support Roman's Discuss (and comment) and will expound on his first
point some more.

The use of URIs to refer to remote content, potentially hosted on a
third party, has numerous security considerations, some better known
than others.  We need to reference the security considerations of RFC
3986 to get some coverage of these, such as the "reliability and
consistency" point.

One of the more subtle security considerations of using a URI to a
remote resource as part of a request in cases such as this, is that the
binding between the identity of the entity referring to the remote URI
and the identity of the site providing the corresponding resource (i.e.,
the authority compoent of the URI) is not always clear.  Although RFC
7852 requires TLS mutual authentication and HTTPS transport for
information provided by reference, it makes no statement requiring the
TLS server certificate to correspond to the SIP initiator, or any other
indication of authorization that the indicated resource makes sense as
part of the indicated request.

I also have two other Discuss-level points:

We discuss the use of timestamps in protocol elements for detecting
replay; this requires that all participants have (secure and) accurate
time.  We need to document this dependency on accurate time, and
optionally point out that common internet time protocols such as NTP are
not particularly secure at present.

I'm also not sure where there's existing treatment of using SIP MESSAGEs
as a DoS attack vector; that seems particularly pronounced in the
emergency-services case since emergency messages may receive
preferential treatment from the network.


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

Introduction (and Abstract?)

   medical monitors.  These messages may be one-shot alerts to emergency
   authorities and do not require establishment of a session.  These
   type of interactions are called 'non-interactive emergency calls'.

Is this an existing/widespread definition or one new to this document?
It may be worth providing a reference or rewording to clarify that it is
a new definition.

Section 2

   CID is Content InDirection [RFC2392]

The string "InDirection" does not appear in RFC 2392.

Section 4.1

   A CAP message may be sent in the initial message of any SIP
   transaction.  However, this document only addresses sending a CAP
   message in a SIP MESSAGE transaction for a one-shot, non-interactive
   emergency call.  Behavior with other transactions is not defined.

I suggest rephrasing to "A CAP message is sent in the initial message of
a SIP transaction" to avoid potential concerns about inconsistency
between "may be sent" and "not defined".

   the CAP message.  Alternatively, the Call-Info header field may
   contain a Content Indirect url [RFC2392] and the CAP message included

nit: "Indirect" here vs. "InDirection" in Section 2

Section 4.2

      field.  If there is a need to copy the PIDF-LO structure
      referenced by 'geolocation' to <area>, implementers must be aware

When might this occur?  (It seems almost out of scope for this document
based on my limited understanding.)

      that <area> is limited to a circle or polygon, and conversion of
      other shapes will be required.  Points SHOULD be converted to a
      circle with a radius equal to the uncertainty of the point.  Arc-
      bands and ellipses SHOULD be converted to an equivalent polygon.
      3D locations SHOULD be converted to their equivalent 2D forms.

nit: what is an "equivalent" polygon to an ellipse or arc-band?  In a
mathematical sense it might mean "strict equivalence" or a large-n limit
of an n-gon, or perhaps there is a corresponding uncertainty on the
ellipse/arc-band that the polygon must stay within?  Rewording to
"corresponding" might be an easy fix.

Section 5.1

   The 425 response code is a rejection of the request due to its
   included alert content, indicating that it was malformed or not
   satisfactory for the recipient's purpose.

   A SIP intermediary can also reject an alert it receives from a User
   Agent (UA) when it detects that the provided alert is malformed.

(I assume there's an implicit "reject with this code" in here.)

   It is only appropriate to generate a 425 response when the responding
   entity has no other information in the request that is usable by the
   responder.

I'm not sure I'm parsing this sentence correctly.  My best guess is that
it's saying "if there's anything in the request that you can respond to,
don't use 425, even if there's some malformed CAP", but that doesn't
really seem to be what the actual words are saying.  [Section 5.2 seems
to support my guess.]

Section 5.2

      message-header   /= AlertMsg-Error
                              ; (message-header from 3261)

nit: mybe "RFC3261" to match the later comments?

   wrong with the alert in the request.  This error code has a

pedantic nit: the ABNF has 1-to-3-digit

   standardized -- meaning an operator can change the strings -- but
   MUST NOT change the meaning of the error code.  Similar to how RFC
   3261 specifies, there MUST NOT be more than one string per error
   code.

RFC 3261 is ... pretty long.  Maybe throw us a bone with a section
reference?  Also, I assume this "MUST NOT be more than one" is within a
single AlertMsg-Error construction, since a truly global interpretation
would mean that (since this document assigns strings) there cannot
actually be changes made to the strings by an operator, which we just
said is possible.

   This document defines an initial list of AlertMsg-Error values for
   any SIP response, including provisional responses (other than 100
   Trying) and the new 425 response.  There MUST be no more than one
   AlertMsg-Error code in a SIP response.  AlertMsg-Error values sent in

Is this redundant with the "MUST NOT" I was asking about above?

   Additionally, if an entity cannot or chooses not to process the alert
   message from a SIP request, a 500 (Server Internal Error) SHOULD be
   used with or without a configurable Retry-After header field.

I don't think I understand when it's recommended to use 500 vs 425 (and
vice versa).

Section 7

It's probably worth mentioning RFC 3428's note about:

%  The size of MESSAGE requests outside of a media session MUST NOT
%  exceed 1300 bytes, unless the UAC has positive knowledge that the
%  message will not traverse a congestion-unsafe link at any hop, or
%  that the message size is at least 200 bytes less than the lowest MTU
%  value found en route to the UAS

since that seems to still apply here.

Section 8

I'd consider using a more recent date than 2008-11-19 for the example
<sent> time.  Also, why is the timestamp in the additional information
2010-11-14, almost two years later?

                     <gml:pos>32.86726 -97.16054</gml:pos>

Is that ... someone's house?  Maybe some different coordinates would be
more appropriate.

Figure 3 has "Content-Disposition: by-reference;handling=optional" but
Figure 4 does not; is that significant?

Section 9

   In any case, for alerts initiated by sensors, the identity could play
   an important role in deciding whether to accept or ignore an incoming
   alert message.  With the scenario shown in Figure 1 it is very likely
   that only authenticated sensor input will be processed.  For this
   reason, it needs to be possible to refuse to accept alert messages
   from an unknown origin.  Two types of information elements can be
   used for this purpose:

   1.  SIP itself provides security mechanisms that allow the
       verification of the originator's identity, such as P-Asserted-
       Identity [RFC3325] or SIP Identity [RFC8224].  The latter
       provides a cryptographic assurance while the former relies on a
       chain of trust model.  These mechanisms can be reused.

   2.  CAP provides additional security mechanisms and the ability to
       carry further information about the sender's identity.
       Section 3.3.4.1 of [cap] specifies the signing algorithms of CAP
       documents.

I'd consider adding a note that "the specific policy and mechanisms used
in a given deployment are out of scope for this document", since the
actual requirements in practice are so broad that we cannot really make
detailed recommendations here.

   In addition to the desire to perform identity-based access control,
   the classic communication security threats need to be considered,
   including integrity protection to prevent forgery or replay of alert
   messages in transit.  To deal with replay of alerts, a CAP document
   contains the mandatory <identifier>, <sender>, <sent> elements and an
   optional <expire> element.  Together, these elements make the CAP
   document unique for a specific sender and provide time restrictions.
   An entity that has already received a CAP message within the
   indicated timeframe is able to detect a replayed message and, if the
   content of that message is unchanged, then no additional security

Is *able* to detect, sure.  But is it recommended to do so?  Required?

   vulnerability is created.  Additionally, it is RECOMMENDED to make
   use of SIP security mechanisms, such as the SIP Identity PASSporT
   [RFC8225], to tie the CAP message to the SIP message.  To provide

Not required?  Please describe the risks of not doing so (whether
directly or by reference).

   protection of the entire SIP message exchange between neighboring SIP
   entities, the usage of TLS is REQUIRED.

Just to check: this is a 100% always, for everyone, "REQUIRED", not just
in the subset of cases where protection of the entire exchange is
desired?

   Note that none of the security mechanism in this document protect
   against a compromised sensor sending crafted alerts.  Privacy
   provided for any emergency calls, including non-interactive messages,
   is subject to local regulations.

I suggest using "confidentiality protection" rather than "privacy"
(unless it's a term of art in this field), to avoid confusion with...
Privacy considerations for alerts (e.g., generated by sensors) that
should be discussed.  RFC 7852 has some extensive privacy considerations
that can be referenced to do the bulk of the work here.

I'd also consider reiterating the opportunistic nature of cryptographic
protection on emergency messages, as discussed in 6443 (since insecure
emergency messaging is preferred to no emergency messaging), but this is
not strictly speaking required.

Section 12.2

RFC 8225, as a RECOMMENDED feature, is more properly listed as a
normative reference, per
https://www.ietf.org/about/groups/iesg/statements/normative-informative-references/