On Mar 9, 2011, at 6:01 AM, Oscar Novo wrote:
Hi Ben,
More comments inline as [ON1]. :)
-----Original Message-----
From: Ben Campbell [mailto:ben(_at_)estacado(_dot_)net]
Sent: 8. maaliskuuta 2011 23:14
To: Oscar Novo
Cc:
draft-ietf-xcon-common-data-model(_dot_)all(_at_)tools(_dot_)ietf(_dot_)org;
General Area Review Team; The IETF
Subject: Re: Gen-ART LC Review of draft-ietf-xcon-common-data-model-23
Hi, thanks for the quick response. More comments inline. I've deleted any
sections on which I think we are in agreement.
On Mar 7, 2011, at 6:07 AM, Oscar Novo wrote:
Hello Ben,
Thanks for your comments. Answers to your comments inline.
Oscar
-----Original Message-----
From: Ben Campbell [mailto:ben(_at_)estacado(_dot_)net]
Sent: 5. maaliskuuta 2011 0:46
To:
draft-ietf-xcon-common-data-model(_dot_)all(_at_)tools(_dot_)ietf(_dot_)org
Cc: General Area Review Team; The IETF
Subject: Gen-ART LC Review of draft-ietf-xcon-common-data-model-23
I am the assigned Gen-ART reviewer for this draft. For background on
Gen-ART, please see the FAQ at
<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.
Please resolve these comments along with any other Last Call comments you
may receive.
Document: draft-ietf-xcon-common-data-model-23
Reviewer: Ben Campbell
Review Date: 2011-03-04
IETF LC End Date: 2011-03-04
IESG Telechat date: (if known)
Summary: This draft is almost ready for publication as a proposed standard.
I have a few minor comments that should be considered prior to publication.
[...]
Minor issues:
-- section 3.3.1: "XCON-URI can not be resolved to addresses and/or ports."
Then why does it include a port in the ABNF?
[ON] Note that URIs can not be resolved to addresses and then to ports.
That's why we state it very clear in the document that this isn't the case.
Besides that, an XCON-URI can be viewed as a key to access a specific
conference object. So, having a direct mapping to a URL can be useful some
times for some conferences.
I understand that it doesn't map to a port. But I don't understand why you
would include a "port" construction in a URI that can't map to a port.
Actually, to generalize, I don't understand why one would use "host" either,
since that construction is designed to carry addressing material, either in
the form of a DNS name or an IPv4 or v6 address.
What do you mean by "direct mapping to a URL?" Do you expect to contruct an
XCON URI from, say, a SIP URI?
[ON1] Ben, I think the XCON-URI identifier is unclear to you. I recommend you
to read section 3.3 of my document. A XCON-URI identifier is created by the
conference system and maintains a relationship between all the conference
object identifiers in the conference. Figure 2 of my document shows an
example and, in that case, the XCON-URI is a host identifier:
xcon:Ji092i(_at_)example(_dot_)com. A XCON-URI can be anything. However, I
can remove the port if it's unclear.
I think it would be best to remove "port" unless you have a concrete reason to
include it. I assume there must have been one at one time, but it's not
apparent to me from the text, or the discussion so far.
Also, can "host" be an IP address? If so, does that change the
comparison rules? (i.e. 192.168.0.1 vs 192.168.000.001, suppression of
zeros in an IPv6 address, etc?)
[ON]I'm not an URI expert. Ted Hardie was the person in charge to review and
verify the URIs of the document. For your question I would recommend you to
read RFC3986 section 6.1.
3986 says "string comparison, perhaps augmented by reference to additional
rules
provided by URI scheme definitions."
My point is that if you are really using the host and port constructions,
they need some of those additional rules. Certainly, 2 host:port
constructions that match character for character are equivilent--but there's
a number of ways host:port can be equivalent when it _doesn't_ match
character for character. And that's not even counting aliasing--I'm talking
about strict syntactic equivalence.
Wouldn't it better to just use some sort of string contruction (perhaps
"unreserved") that would allow you to put something in it that looks like
host:port, but without the meaning usually associated with those?
Was this concern discussed in Ted's review? If so, do you have a pointer to
it?
[ON1] From RFC3986 section 6.1: "comparison methods are designed to minimize
false negatives while strictly avoiding false positives."
3986 does not mandate that all URI schemes use a character by character
comparison. It lists that as a minimum choice. It also suggests that each
scheme will have it's own additional rules.
My issue with "host" is that there's a lot of implied baggage there. Host can
be an literal IP address (IPv6, IPv4, IPvFuture, a name, etc.) For some of
these, a character comparison may not yield the results that an implementor
would intuitively expect. "Host" can include internationalized domain
names--and that _really_ throws a kink into comparisons. Furthermore,
implementors are likely to already have libraries that they think "just do the
right thing" with the "host" syntax--but that right thing may be exactly the
wrong thing for XCON-URIs
I don't think that necessarily means you shouldn't use "host". But it would
probably be worth adding a non-normative note (perhaps as an indented
parenthetical note) pointing out that the character comparison rule may not
always match the the usual comparison expectations for certain kinds of
content, and giving an example or two. (For example, 192.168.0.1 vs
192.168.000.001).
OTOH, a simpler alternative (from the reader's perspective) would be to just
define some sort of string-token construction that uses the correct characters,
but doesn't that the "address semantics" baggage of "host".
Ted's agreed on the text mention in the document. He's an URI expert.
I certainly can't dispute Ted's expertise here, but did he offer a rationale
for his agreement?
-- 4.6.2, 1st paragraph:
Are two users with the same signaling protocol allowed to have different
authn mechanisms?
[ON] Answering that question is out of the scope of this document. The
Conference Control Protocol and/or RFC5239 (for instance section 11.1)
should answer this question.
My question is whether the data model allows it. Why wouldn't that be in
scope?
[ON1] From section 4.6.2 of the document: " Since a variety of signaling
protocols are possible, a variety of authentication mechanism - determined by
every individual conference servers - may need to be mapped from the
different protocols. The specific types of authentication mechanism are
beyond the scope of this document."
I'm concerned that <user-admission-policy> is a child of <users>, not <user>,
so it looks like all the <user> elements with the same <users> parent must
share the same admission policy. Am I misreading that? If not, how would you
render multiple policies in the same conference?
[ON1] You render multiple policies using the <allowed-users-list> and
<deny-users-list> elements. The <user-admission-policy> only lets the
organizer to choose the policy of the whole conference: closedAuthentication,
openAuthentication...
So you would never have "closedAuthenticated" for one set of users and
"openAutenticated" for another set for the same conference, right? If so, then
I think the text is okay here.
-- 4.6.3, 1st paragraph:
What if the user is using a protocol that doesn't use URIs?
[ON] This section talks about the SIP URI or the xcon-userid URI defined in
Section 4.6.5. The xcon-userid contains a unique conference user identifier
(XCON-USERID) within the scope of the conference. This URI will always exist
within the scope of the conference.
It lists SIP URIs and XCON-USERID URIs as examples. Are you limited to those
schemes? What if someone is using XMPP for their signaling protocol--are they
excluded since XMPP does not use URIs?
[ON1] A XCON-USERID is a unique conference user identifier assigned by the
conference system. It's explained in the section 4.6.5 of the document. If
there is an XMPP signaling protocol, such XMPP identifier will be mapped to
the XCON-USERID.
The first paragraph in 4.6.3 says "The <allowed-users-list> child element
contains a list of user URIs (e.g. SIP URI, xcon-userid...)
This led me to believe you meant to allow signaling protocol URIs, since SIP
was mentioned. Did you mean that to really be limited to XCON URIs? When I read
further down, I see the <user> child element seems to be limited to using
XCON-USERID, so this seems to be the case. If that is correct, then simply
removing the reference to SIP URIs in 1st paragraph of 4.6.3 would seem to fix
things.
-- 4.6.5.3: "The real information about the user is still stored in the data
model."
This could use some elaboration. Does this mean that clients subscribing to
the event package will get the real data, but be expected to conceal it from
the user? Or that the data is only stored internally by the focus and not
sent to subscribers?
[ON] The data model specifies a set of elements for different use cases in
every conference system. That doesn't imply all the elements has to be
defined in every conference, only those elements needed for every conference
system. RFC5239 section 11.2 explains a bit about the privacy concept in a
conference.
My question wasn't whether they are used for a particular conference. It's
more about how the data is protected when it is used by a conference. In
particular, the text says this affects the way the data is provided to other
participants. _How_ it effects it needs more detail.
I think that you mean that the focus should not provide the data to
participating clients. _Not_ that said clients can be expected to conceal it
from their users. This should be explicit either here or in the security
considerations.
If you think RFC 5239 explains this, then a reference to it would be in
order. I think it helps a bit, but I think you could use some stronger and
more explicit language here.
[ON1] What about modifying the text from:
"This element only affects the way the user information is provided to the
other participants. The real information about the user is still stored in
the data model."
To:
"This element only affects the way the user information is provided to the
other participants. The real user information is stored in the data model but
is not provide to the other participants of the conference."
My concern was what is meant by "is not provided" at a protocol level. But on
reflection, that probably doesn't belong here. OTOH, should that be a normative
"SHOULD not be provided"?
"'semi-private' value specifies that this user is anonymous to all users
with equal or lesser permissions (determined by local policy) in the
conference."
This also needs elaboration, even if the way permission systems work is out
of scope.
[ON] what information you think is missing in this sentence? Note that the
WG agree on not defining the local policy (roles semantic, permissions...)
in this document.
What do you mean, in abstract, by "equal or lesser" permissions? That implies
some assumptions about how permissions work, even though they are explicitly
out of scope. Equal or lesser to what? Why assume permissions are
hierarchical in the first place? Is it simply that the information should
only be provided to users who have permission to see it? (a much simpler
statement with fewer assumptions about how permissions work)
[ON1] So, what I meant with the "equal or lesser" is something like: "a
semi-private value implies a root role is anonymous to a user role". Since
the WG decided not to include the different roles and policies in this
document, we have to leave it as an abstract concept and every local policy
will define it as they think its appropriate.
Not all systems use "root" and "user" as the permissions model. So how about
removing the whole idea of "greater" and "lesser", and say something to the
effect of "... anonymous to all users who have not been granted permission to
see it"
[...]
-- 8, paragraph 6: "The confidentiality of the database SHOULD be
unauthorized users, given that the data model sensitive elements (e.g.,
passwords)."
Confidentiality of a database containing passwords only rates a SHOULD?
[ON] It might be the case in some particular scenarios where the
confidentiality can be ignored. For this reason, we decided not to impose
confidentiality (using SHOULD instead of MUST) in the document and to leave
that decision to the administrator of the conference.
Really? Can you give an example where it's safe to ignore confidentiality of
_passwords_?
[ON1] There are conference which don't define any password and any sensitive
information. Providing confidentiality at a conference may seem strange or
counterproductive sometimes, especially these days where event sessions are
routinely streamed and videoed for anyone who wants to watch.
Okay.
"Administrators of conferencing systems SHOULD also avoid disclosing
information to unauthorized parties when a conference is being cloned or
when a sidebar is being created."
This could use elaboration. Why is creating a sidebar more dangerous than
other operations not mentioned here?
[ON] Sidebars manipulation are described in RFC5239, section 9.4
If you think that answers the question, then a reference would be in order.
But on a quick scan, I don't see where that section explicitly discusses
security risks in creating sidebars.
[ON1] For instance, RFC5239, section 9.4.2 is talking about the external
sidebars. In those sidebars, users -which are external to the conference -
are *joining* the conference. Disclosing information to those unauthorized
users is essential. That's just an example.
The data model describes the data of the conference objects which are
described in RFC5239. So, the data model is *complementing* RFC5239. The aim
of the data model is not to repeat the information already written in RFC5239
rather than defining the missing information.
I'm not suggesting you recreate the definition of sidebars here. I'm suggesting
you explain the security consideration that you already mention. For example,
an additional sentence to the effect of "For example, an external sidebar as
defined in [RFC5239], section 9.4.2, may include participants who were not
authorized for the parent conference."
[...]
-- Section 1:
It might be nice to have a paragraph defining what you mean by "conference"
before jumping into conference attributes.
[ON] The definition of conference is already defined in RFC5239 section 4.
Sure, but it would be reader friendly to have a sentence describing it here.
I'm not talking about a detailed description.
[...]
[ON1] The terminology part of the document states:
"This document uses the terminology defined in the Centralized
Conferencing Framework [RFC5239], the SIPPING conferencing framework
[RFC4353] and the BFCP (Binary Floor Control Protocol) specification
[RFC4582]. Readers of this document should be familiar with the
terminology used in those documents."
While I expect a reader has to be familiar with 5239 to get much use of this
draft, it would still be more friendly to the reader if they can at least get a
sense of what this document is about prior to digging into the references. All
I am asking for is a sentence to the effect of the following:
"[RFC5239] defines the idea of a centralized conference as an association of
participants with a central focus. The state of a conference is represented by
a conference object. This document describes the data model to be used for
conference objects."
-- section 3.3, first paragraph after the diagram:
is XCON-URI intended to be an acronym? If so, you describe it as a "unique
conference object identifier", but never actually expand the acronym itself.
[ON] XCON-URI is not an acronym
Hmm. It doesn't stand for "Centralized Conferencing - Universal Resource
Identifier"?
[...]
[ON1] XCON-URI is a conference object Identifier.
That's the definition, not the expansion. But I think we've spent too much
effort on this one already. I'd leave it to the RFC editor at this point.
[...]
-- 4.6.2, last paragraph: "In all other cases, the appearance of an
<allowed-users-list> and <deny-users-list> MUST be ignored."
What other cases? I assume you don't mean to say that any extensions are
prohibited from using allowed-user-list or deny-user-list--but one can read
it to mean that.
[ON] That paragraph was an AD review proposal:
http://www.ietf.org/mail-archive/web/xcon/current/msg02495.html
With all due respect to Robert, I think the current text is unclear. I
interpret "All other cases" to mean "All cases except for those enumerated
above". That is, all cases other than the one it's talked about so far. The
text goes on to say some things about future extensions--which seems to
contradict the first sentence. I think you mean to say that, except in the
case where a future specification gives guidance otherwise, all other cases
MUST be ignored.
One simple fix would be to move the first sentence to the end of the
paragraph. Something to the effect of:
"Future specifications using the <allowed-users-list> and <deny-users-list>
lists must provide clear guidance... . In all other cases, the appearance of
an <allowed-users-list> and <deny-users-list> MUST be ignored."
Another approach would be to say "All other cases... MUST be ignored, except
as stated otherwise in a future specification describing..."
[...]
[ON1] I recommend you to talk with Robert about your proposed change. If he
accepts this change I'll modify my document. Otherwise, I will stick to the
AD proposal.
I talked to Robert offline. We came to the conclusion that my second example
above is the best choice. Here's a complete example:
"In all other cases, the appearance of an <allowed-users-list> and
<deny-users-list> MUST be ignored, except as otherwise described in a future
specification. Future specifications describing the use of these lists must
provide clear guidance on how to process the lists when they occur
concurrently, especially when both lists contain the same user."
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf