Re: [alto] Chair review of unified-props (Part 2 of 2)

["Randriamasy, Sabine (Nokia - FR/Paris-Saclay)" <[e-mail]>]

Hello Vijay,

Thank you for your review of Entity Property Maps. As promised, please find inline the answers from Jensen and myself on how your comments have been addressed, prefixed respectively by “[jensen]” and “(sabine)”.
Responses on sections 9.3 and 11, with “*****”, request further feedback from you.
Thanks,
Sabine


From: Vijay Gurbani <[e-mail]>
Sent: Wednesday, January 27, 2021 5:28 PM
To: [e-mail]
Cc: IETF ALTO <[e-mail]>
Subject: Chair review of unified-props (Part 2 of 2)

Chair review from Section 5 - end (inclusive).

Please go through the Major review points, they require some attention.

This concludes my review of this document.  Overall a well-written document
covering a range of important extensions to ALTO.

Major:

- S5.1.1: "The '.' separator MUST NOT be used unless specifically indicated in a
further extension document." ==> I don't understand this.  If the '.' must not
be used, then this should be an absolute condition (MUST NOT is the normative
strength in the sentence).  If this document allows the '.' to be used in a
further extension document, then this is a relative --- not absolute ---
condition, and thus a normative SHOULD NOT seem to be better.  However, as
currently written, this sentence seems rather wishy-washy.  Either say '.' is
out of bounds or say it is not, it seems weird to say that this document does
not use it, but others can.

One way to achieve this is to replace that sentence with:

  The '.' separator is not used by this document, however, future
  extensions may use it.  For the purpose of this document, the
  strings "ipv4", "ipv6", and "pid" are valid entity domain
  types, while "ipv4.anycast", and "pid.local" are invalid.

However, even then, I am not sure if the above is correct.  Later, in S5.1.2,
you go on to say that "Note that the '.' separator is not allowed in
EntityDomainType and hence there is no ambiguity on whether an entity domain
name refers to a resource-agnostic entity domain or a resource-specific entity
domain."  Thus it seems to me that future extensions could run into trouble if
they allow '.' in the EntityDomainType.

This needs to be resolved before publication.
[ [SR] ] [Jensen] Thanks for noticing this bug. The '.' separator is not allowed even in an extension document.
The paragraph has been revised to the following:

   An entity domain has a type, which is uniquely identified by a string
   that MUST be no more than 64 characters, and MUST NOT contain
   characters other than US-ASCII alphanumeric characters
   (U+0030-U+0039, U+0041-U+005A, and U+0061-U+007A), the hyphen ('-',
   U+002D), or the low line ('_', U+005F).


- S5.1.2, last paragraph: "Note that the resource ID format defined in Section
10.1 of [RFC7285]..." ==> Section 10.1 of RFC 7285 defines "PID Name", not
"Resource ID", which is defined in the next section.  Please clarify.
[ [SR] ] [Jensen] New paragraph rephrasing now:
   Note also that Section 10.1 of [RFC7285] specifies the format of the PID Name which is
   the format of the resource ID including the following specification: ...


- S9.1: "...it is RECOMMENDED that the EPS be deprecated in favour of ..." ==>
This is very important: If this draft is recommending a change in RFC 7285, then
the status of this draft must say "Updates: 7285" at the top of the draft (it
currently does not).  This will cause the RFC Editor to enter a "Updated by:
RFCXXXX" in the masthead of RFC7285.

Further, I presume that since this update is a recommendation, the processing
rules of property map and filtered property map are distinct enough that legacy
ALTO servers and clients will continue operations?  Please advise.
[ [SR] ] [Jensen]  We do not intend to update RFC7285 to deprecate EPS.
This recommendation is more like suggesting that the new ALTO servers that support UP
should also provide the EPS to be backwards compatible with legacy clients.
Since the keyword "RECOMMENDED" has a specific meaning, let's not use this keyword here.
The paragraph has been changed as follows:
NEW
Since the Property Map and the Filtered Property Map defined in this
   document provide a functionality that covers the Endpoint Property
   Service (EPS) defined in Section 11.4 of [RFC7285], ALTO servers may
   prefer to provide Property Map and Filtered Property Map in place of
   EPS.  However, for the legacy endpoint properties, it is recommended
   that ALTO servers also provide EPS so that legacy clients can still
   be supported.


- S9.3: What does "little or no impact on other previously defined properties"
mean?  Again, this appears wishy-washy for a standards document.  Either we know
that there is some impact, and we document what the impact is, or we state that
there is no impact.  But saying "little or no impact" is not helpful.  Please
state where there is impact and whether this impact will cause backwards
compatibility problems.

I note that S12.2.1 further expands on this "little or no impact".  Perhaps a
second order look at S9.3 is in order before publication to document the impacts
on previously defined properties.
[ [SR] ] (sabine)  Agree.
We have skipped this unclear first sentence and focused on the impact on the property semantics.
***** (Note: I did not find the text in 12.2.1 of version 15 that expands on this "little or no impact")
The paragraph was updates as follows:
NEW
In the present extension, properties can now be defined on sets of entity addresses, rather than just individual endpoint addresses as is is the case in RFC7285. This might change the semantics of a property. These sets can be for example hierachical IP address blocs. For instance, a property such as fictitious "geo-location", defined on a set of IP addresses would have a value corresponding to the barycenter of this set of addresses.


- S10: Please fill in all of the "###" for "Content-Length: ###" in all
examples.
[ [SR] ] [Jensen]  done


- S11: Did anyone in the WG do a threat analysis of the information exposed by
the property maps related to, say, domain type "ane"?  As the example in S10.9
demonstrates, there is a lot of information in the returned response that will
be of interest to malicious actors.

I note that there is a blanket paragraph in this section ("A particular
fundamental ... URI to provide the information.") that appears to address such a
scenario, but I don't see a well thought out threat-model that drives the
discussion in this section.  At the very least, the authors should spend some
time thinking about the information in the property maps and how it may be used
if it fell in the hands of a malicious actor.  If after reflection they come to
the conclusion that the second blanket paragraph outlines a mitigation strategy
that suffices, then that is fine.  Just wanted to make sure that the authors
have spent some time here.
[ [SR] ] (sabine) ***** agree
The text indeed proposes one option, that consists in managing access to information on a per property "P" value, matched against the IP address of the Client. Another, maybe more scalable option is to gather information that the provider considers sensitive in a delegate ALTO Server, whose access would be restricted.
If you think this makes sense, we will further check with the WG on such options, especially with operators and applications and hope to get back with new text that better reflects protection strategies, possibly see how access to sensitive information can be managed.

Minor:

- S5.1.3, last paragraph: I think you should append the following sentence to
the paragraph:

  Such equivalences should be established by the object represented
  by DomainTypeSpecificEntityID, for example, [RFC5952] establishes
  equivalence for IPv6 addresses, while [RFC4632] does so for IPv4
  addresses.
[ [SR] ] [Jensen] Thanks. I like this suggestion. Done.

Nits:

- Please run idnits; currently, it is reporting 3 errors having to do with using
obsoleted references (RFCs 5226 and 7159) and a downref (RFC 7921).
[ [SR] ] [Jensen] Done.

Thanks,

- vijay