On Fri, Jul 29, 2016 at 1:47 PM, Robert Sparks
<rjsparks(_at_)nostrum(_dot_)com> wrote:
On 7/29/16 3:36 PM, Andy Bierman wrote:
Hi,
I will add this review to the list.
A new version in in progress.
Some comments inline
On Fri, Jul 29, 2016 at 1:11 PM, Robert Sparks
<rjsparks(_at_)nostrum(_dot_)com>
wrote:
I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please treat these comments just
like any other last call comments.
For more information, please see the FAQ at
<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.
Document: draft-ietf-netconf-restconf-15
Reviewer: Robert Sparks
Review Date: 28Jul2016
IETF LC End Date: 3Aug2016
IESG Telechat date: not yet scheduled
Summary:
Major issues:
* I am not finding any discussion in the Security Considerations or in
the text around what a server's options are if a client is asking it to
keep more state than it is willing or capable of holding. The possible
values of the "depth" query parameter (particularly "unbounded") points out
that a misconfigured or compromised client might start creating arbitrarily
deep trees. Should a server have the ability to say no?
I guess we need more text somewhere explaining the "depth" parameter is
a retrieval filter.
I got that. It's existence, however, caused me to think about the fact
that what is stored at the server can be arbitrarily deep. Clients using
POST can build trees that are arbitrarily deep, with bits at the node that
are arbitrarily large (subject to the constraints the YANG models put on
the node). There should be some discussion acknowledging that this can
happen, and discussion of what the server can do if some client starts
asking it to store more than it is willing to store.
Clients can build trees according to the YANG modules supported by the
server.
The YANG module has conformance requirements.
The protocols have 'resource-denied' errors.
Not sure what kind of warning they need.
An implementation may have no problem with 4 deep table entries,
but cannot store 100,000 flat table entries.
Andy
It is not used to create anything in the server.
The server does not maintain any state except during the processing of
the retrieval request
* The third paragraph of 3.7 paraphrases to "SHOULD NOT delete more than
one instance unless a proprietary query parameter says it's ok". This isn't
really helpful in a specification. Proprietary things are proprietary. The
SHOULD NOT already allows proprietary things to do something different
without trainwrecking the protocol. Please just delete the 2nd and 3rd
sentence from the paragraph.
OK
* Section 2.3 says "If X.509 certificate path validation fails and the
presented X.509 certificate does not match a locally configured certificate
fingerprint, the connection MUST be terminated as defined in [RFC5246]."
RFC5246 doesn't really talk about certificate validation, and it certainly
doesn't say "the connection MUST be terminated" when certificates fail to
validate. What are you trying to point to in RFC5246 here? Should you be
pointing somewhere else? (It's perfectly reasonable for the document to
reference RFC5246, and it does so elsewhere without problem).
Please suggest replacement text if we are citing the wrong RFC.
I will ask Kent to look into this issue
Minor issues:
* "A server MUST support XML or JSON encoding." is ambiguous. (2nd
paragraph of 5.2). Did you mean the server MUST support at least one of XML
or JSON but not necessarily both? I think you really intended that the
server support BOTH types of encoding.
No -- it will be clarified that the server must support at least 1 of the 2
* I _think_ I can infer that PUT can't be used with datastore resources.
Section 3.4 only speaks of POST and PATCH. Section 4.5 speaks about "target
data resource" and is silent about datastore resources. If I've understood
the intent, please be explicit about datastore resources in 4.5. If I've
misunderstood then more clarity is needed in both 3.4 and 4.5.
The next draft will be clarified to allow PUT on a datastore resource
Hrmm - that makes me less comfortable that you are actually aligned with
7231. It may just be that you need to be more precise with your
description, but per 7231, PUT never creates resources - it can create or
replace the state of a resource.
* In 3.5.3.1 you restrict identifiers with "MUST NOT start with 'xml' (or
any case variant of that string). Please call out why (or point to an
existing document that explains why).
OK
* The text in 5.3 about access control interacting with caching (added
based on my early review I think) doesn't mesh well with paragraph 3 of
section 5.5. There you tell the client to use Etag and Last-Modified, but
in 5.3 you say it won't work reliably when access permissions change. At
the very least 5.5 should point back into the paragraph in 5.3.
Nits/editorial comments:
* Introduction, 4th paragraph - please change "MAY provide" to
"provides". Section 3.6 explains the cases where there is choice in what to
provide.
* Section 2.3 paragraphs 1 and 2. There is edit-itis here left (I
suspect) from working in matching fingerprints. Consider combining and
simplifying these two paragraphs after improving the reference issue called
out above.
* Section 4 says "Access control mechanisms MUST be used to limit..."
This is not a good use of a 2119 MUST. I suggest replacing "MUST be" with
"are". The subsequent text already captures the actual normative
requirements on the server.
* Section 12 says "this protocol SHOULD be implemented carefully". That
is not a good use of a 2119 SHOULD. It is not a protocol requirement. I
suggest reformulating this into something like "There are many patterns of
attack that have been observed through operational practice with existing
management interfaces. It would be wise for implementers to research them,
and take them into account when implementing this protocol." It would be
far better to provide a pointer to where the implementer should start this
research.
* (micronit) Lots of examples are internally inconsistent wrt dates. For
instance, look at the 200 OK in section 3.3.3 - it says that back in 2012,
a server returned something talking about a library versioned in 2016.
Andy