ietf
[Top] [All Lists]

Re: [paws] Gen-art LC review: draft-ietf-paws-protocol-12

2014-07-15 16:43:25

On 7/8/14, 9:21 PM, Vincent Chen wrote:
Robert,

Thanks again for the review. I have place comments, questions, suggested text inline for each of your comments.
Please take a look.
I'll be in Toronto if you want to talk about any of these.

Some small comments inline:

-vince


On Thu, Jul 3, 2014 at 10:28 AM, Robert Sparks <rjsparks(_at_)nostrum(_dot_)com <mailto:rjsparks(_at_)nostrum(_dot_)com>> wrote:

    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-paws-protocol-12
    Reviewer: Robert Sparks
    Review Date: 3 July, 2014
    IETF LC End Date: 7 July, 2014
    IESG Telechat date: 10 July, 2014

    Summary: This document is not ready for publication as a Proposed
    Standard.

    I apologize in advance if I've missed where one of the questions
    below is already answered. There's a lot to take in here.

    Major Issues

    - The document says it "describes" the use of HTTP/TLS as the
    transport for the protocol. Was it the intent to allow others? If
    not, the language should be firmed up.


The protocol messages are meaningful independent of the transport, so we did not want to prohibit others. It that frowned on?

Well, your security considerations, at least, should talk about what you require of the transport if you don't use TLS. You're leaning heavily on TLS for things like server authentication for example.


    - The document still says "TBD Define message format" in the
    section on Listing Servers. I understand from reading the list
    that what the document is going to say about Listing Servers is
    going to change (to not include how you talk to them?). This
    change needs to be finished before the document can be reviewed
    for completeness.


Noted. Will ask again on the list.


    - It's not clear when a server should use the HTTP level
    redirection discussed in section 7 vs the databaseChange mechanism
    in the protocol's responses. There should be some discussion about
    what the Device should do when the databaseChange mechanism
    results in a redirect loop.


Thanks for pointing this out. I propose removing the HTTP-level 301 Moved Permanently mechanism and only have the databaseChange mechanism.
The group will have to chew on this I guess. You can't forbid HTTP-level 3xx responses, and you still want well defined application behavior when they're received.

Proposed modified text in Section 4.1:
   A Database MAY indicate that its URI will be changing by including
   the URI of one or more alternate databases (See DbUpdateSpec
   (Section 5.7)) in its responses to a Device.  Before a Database
   ceases operation, for example, it MUST include DbUpdateSpec in its
   responses to notify Devices.  A Device will update its preconfigured
   list of databases to replace (only) its entry for the responding
   Database with the URIs of the alternate databases; the list of
   alternate databases does not affect any other entries.  Note that the
   ordering of databases in the list does not imply any preference and
   does not need to remain the same for every request.  The Device
   SHOULD detect infinite redirection loops; if a suitable database
   cannot be contacted, the Device MUST treat this as equivalent to a
   response indicating no available spectrum.


    - The document needs to be clear where the primitive types (like
    string, float, and integer) in the UML-ish diagrams in section 4
    are defined. I'm guessing from context that you're assuming the
    definitions in RFC4627. If that's true, there are several places
    that you talk about string where your text should change. RFC4627
    says string is UNICODE, and may be encoded many ways (see section
    3 of that document). If your intent is to restrict all strings to
    UTF-8 encoding say that, and adjust the text you currently have
    that mentions UTF-8. (The various places where you say a string
    MAY contain UTF-8 do not make sense - if you're assuming the
    encoding is UTF-8 and trying to reinforce that there may be
    non-ASCII range UTF-8 here, say that explicitly). There are other
    related phrases that don't make sense such as where you say things
    like "The length of the string MUST NOT exceed 64 US-ASCII
    characters."


OK. Specified UTF-8 and removed meaningless statements.

Proposed added text in Section 4:

   The parameter tables in this section and Protocol Parameters
   (Section 5) are for reference and contain the name of each parameter,
   the data type of each parameter, and whether the existence of the
   parameter is required for the protocol transaction in question.  The
   diagrams are loosely based on UML, and the data types are defined
   either in Protocol Parameters (Section 5) or are one of the following
   primitive or structured types:

   string  A string, as defined by The JavaScript Object Notation (JSON)
      Data Interchange Format [RFC7159], restricted to the UTF-8
      encoding.

   int  A number, as defined by The JavaScript Object Notation (JSON)
      Data Interchange Format [RFC7159], without a fractional or
      exponent part.

   float  A number, as defined by The JavaScript Object Notation (JSON)
      Data Interchange Format [RFC7159].

   boolean  A boolean, as defined by The JavaScript Object Notation
      (JSON) Data Interchange Format [RFC7159].

   list  A structured type the represents a list of elements, as defined
      by The JavaScript Object Notation (JSON) Data Interchange Format
      [RFC7159] array type.  For each list parameter, its diagram and
      description include a reference to the data type its list
      elements.  The diagram notation and description may include
      additional constraints, such as minimum or maximum number of
      elements.

   NOTE: All parameter names are case sensitive.  Unless stated
   otherwise, all string values are case sensitive.


    - The descriptions of messages in section 4 all contain
    *other:any. None of those are reflected in the concrete schema of
    section 6. Should they be?


This was intended to be capture by the note in Section 6:

   NOTE: In general, all messages defined in this section are extensible
   by adding additional properties to support ruleset-specific and
   database-specific requirements.  In all cases, the Device or Database
   MUST ignore any parameter it does not understand.


    - The INIT_RESP description requires one or more RulesetInfo
    objects. What is a database supposed to do if it has no rulesets
    to return (possibly because it doesn't have anything overlapping
    with the list of ruleset IDs listed in the DeviceDescriptor in the
    INIT_REQ.). Should this have been 0..* (along with the
    corresponding change to the text), or is there an Error that the
    database should return when this happens. If the latter, it would
    be good to call it out in 4.2.2.


This was indicated in the prior section for INIT_REQ, but propose adding the following as first paragraph:

   The initialization response message communicates database parameters
   to the requesting device.  This response is returned only when there
   is at least one ruleset.  Otherwise, the Database returns an error
   response, as described in INIT_REQ (Section 4.2.1).


    - Section 4.4 talks about returning OUTSIDE_COVERAGE when the
    location specified in the request is outside the regulatory domain
    (could that be domains?) supported by the database. The sections
    on init and registration (4.2 and 4.3) don't have this discussion.
    Should they?


Yes. Added to those requests. Thanks.


    - A databaseChange (at least as described in 4.4) can provide one
    or more alternate database URIs, affecting the Device's
    configuration. When there's more than one, is there any preference
    to what order the Device should try to use them in? Is there any
    expectation that they will give the same answers to a given
    request? If not, do you want to say anything about discouraging
    devices from asking all the databases it knows about to find an
    answer it likes best?


In general, there should be no difference in answers, because each Database must also conform to the regulatory rules. There is no significance implied by the list ordering. On the other hand, I don't think we need language discouraging devices from asking multiple databases, since they should have an understanding of the rules as well.

See proposed text above that contains:

   Note that the
   ordering of databases in the list does not imply any preference and
   does not need to remain the same for every request.


    - If the requirements to act as if there is no available
    whitespace when you can't reach a Listing Server remain in the
    document, the security considerations should call out that any
    attack that would prevent reaching a Listing Server would result
    in all devices relying on that Listing Server ceasing their use of
    any whitespace.


Thanks. Added.


    - Please check the description of 'timeRange' in section 5.9
    (SpectrumSpec). I think you meant to say "in which there is _NO_
    available spectrum".


Thanks for the catch!


    - The definition of SpectrumProfile (section 5.12) needs
    clarification. Is this allowed?
      "profiles" : [
        {"hz": 5.18e8, "dbm": 30.0 },
        {"hz": 5.24e8, "dbm": 37.0 }
      ]
    If so, what does it mean? Do I do linear interpolation between
    points (33.5 dbm (~2.25 watts) at 521Mhz)?
    Similarly, does this specify a ramp up and then back down? (shaped
    like a ^)?
      [
        {"hz": 5.18e8, "dbm": 30.0 },
        {"hz": 5.21e8, "dbm": 33.5 },
        {"hz": 5.24e8, "dbm": 30.0 }
      ]
    If not, what text disallows it?


This is explicitly allowed. We changed the encoding to allow this. Propose modifying the first paragraph of 5.12 to the following:

   A spectrum profile is characterized by an ordered list of (frequency,
   power) points that represents the shape of maximum permissible power
   levels over a range of frequencies as a piecewise linear curve.

Also add after the list of constraints:

   NOTE: This encoding allows presentation of "ramps" where the slope of
   a line segment may be finite and non-zero.



    - You are using the schema language defined in
    draft-zyp-json-schema to define your message format. That makes it
    a normative reference. The draft is expired - are there plans to
    progress it?


Yikes. The intent is not to define a strict, formal schema, but just to have a way of defining the messages in a concise fashion. Do you have a recommendation of what to do here? Are the descriptions self-explanatory enough? or do I have to define
the "schema language".
Check with your AD on the best way to handle this.


    - Something needs to talk about case-sensitivity of the various
    protocol elements. JSON-RPC says that member names are case
    sensitive and is otherwise silent. The string "sensitive" doesn't
    appear in RFC4627. So, you have an example that says
    "authority":"us". Is that the same as "authority":"US", and where
    does the spec answer that question? My read of JSON-RPC says that
    "authority":"us" and "Authority":"us" are _not_ the same thing,
    and that the second would not be a recognized property of a
    RulesetInfo object.


You're right. See above proposed text defining the primitive types.


    Minor Issues

    - I'm not finding where you define the protocol version. I see
    "1.0" in the json examples in section 6. Where is it specified?
    Within a given method, the only extension point I find other than
    changing the protocol version is the *other concept in messages,
    which MUST be ignored when either side doesn't understand them. So
    there should be some discussion about what kind of change would
    require the protocol version number to change. Suppose you wanted
    to allow batching requests from several slaves into one request to
    the database (similar to AVAIL_SPECTRUM_BATCH_REQ but allowing a
    list of DeviceDescriptors perhaps). Does this require a new
    protocol version, or is it just a new request type in this
    version? If you think it's a new request-type, should there be a
    request and response type and/or method registry?  And yes, I see
    how you could do this with JSON-RPC batch, but if that's where
    you'd send this idea, why didn't you do AVAIL_SPECTRUM_BATCH_REQ
    that way? (Possibly related: I can't figure out what "The
    initialization message also represents extension points for
    database implementations or rulesets that require the explicit
    handshake." is trying to say. Can you rephrase that more simply?)


Good point. Propose adding a new section:

4.2.  PAWS Version

   PAWS version uses a "<major>.<minor>" numbering scheme to indicate
   versions of the protocol.  The protocol versioning policy is intended
   to allow the Device or Database to indicate the format of a message
   and its understanding of PAWS functionality defined by that version.
   No change is made to the version string for the addition of message
   components which only add to extensible field values.  The <minor>
   number is incremented when the changes made to the protocol add
   functionalities (methods), but do not change the existing
   functionalities.  The <major> number is incremented when incompatible
   changes are made to existing functionality.

   The current PAWS version is "1.0".


    - It's not clear what it means to "support" a ruleset. I infer
    that this means that the device has code that implements what's
    required by the name. Can you state that explicitly? Does a Master
    device have to have this code? Could it simply be a box that only
    serves to answer requests from Slave devices? If so, why does it
    care what the rulesets actually are. If a slave can ask and a
    database can answer, should a master just shovel the bits, or is
    there a requirement that the master device be configured to handle
    a ruleset before a slave can ask about it?


OK. Propose update Terminology section to distinguish between "ruleset" and "ruleset identifier":

   Ruleset:  A ruleset represents a set of rules that governs the
      operation of white space devices and Spectrum Databases.  A
      regulatory authority can define its own set of rules or adopt an
      existing ruleset.  When a Database or Device is said to "support a
      ruleset", it means that it contains out-of-band knowledge of the
      rules and that its hardware and software implementations conform
      to those rules.

   Ruleset Identifier:  A ruleset can be identified by an IANA-
      registered identifier (see PAWS Ruleset ID Registry
      (Section 9.1)).  When a Database or Device indicates it supports a
      ruleset identifier, it means that it conforms to the rules
      associated with that identifier.  A regulatory authority can
      define and register its own ruleset identifiers, or it can use a
      previously registered identifier if it adopts an existing ruleset.


    - In the last paragraph of section 4.1 (before 4.1.1 starts), "If
    the Device is already operating" assumes that the device could
    only be operating if it had previously contacted some database.
    The problem is that the device was able to reach a database at one
    point and now it can't reach any.  It would read more clearly if
    you said that explicitly.


Proposed change:

   If the Device had previously contacted a database to get available
   spectrum, but subsequently fails to contact a suitable database, the
   spectrum the Device is currently using can be used for as long as the
   spectrum data is valid; ...


    - Please point somewhere for a definition of the terms
    'uncertainty' and 'confidence' (I suggest
    draft-ietf-geopriv-uncertainty). The GEOPRIV working group has
    gone through many iterations of disagreement about what these
    terms mean and how they should be used. For a taste, skim some of
    
<https://mailarchive.ietf.org/arch/search/?email_list=geopriv&q=uncertainty>.
    If you don't point to a hard definition, your implementation
    community will have to go through the same arguments.


Done. Although regulatory authorities may prescribe their own definitions.


    - Should the security considerations section talk about the risk
    of collisions in serialNumber (since it is the only required
    element in DeviceDescriptor, and isn't a particularly secret
    thing)? Is there any harm at the database if two devices
    innocently end up sending the same serial number (without
    providing any of the optional information that would otherwise
    disambiguate the devices)? Can a device learn anything useful
    about another device by spoofing it? Is it possible that in some
    regulatory realm, some devices would get more access than others,
    encouraging devices to ask about what their competition gets to
    do? Can harm be done by a device sending SPECTRUM_USE_NOTIFY
    messages claiming to be some other serial number (and perhaps
    manufacturer) maliciously? I think this needs more discussion than
    what RFC6953 contains.


Re: Available spectrum.
The available spectrum depends on device type and location, not on serial number. Consequently, there is no advantage to spoofing, and there would be no additional information to be gained on spoofed devices.
So, you can't gain information about someone else by claiming to be their device type either?

If, in the future, there is some regulatory realm that would set up rules differently, then I think security considerations would be extended to handle those cases.

Re: SPECTRUM_USE_NOTIFY
I suppose this depends on the specific regulatory domains (or database implementations) that require it. Currently there is no harm, since notifications do not change the available spectrum answers that the Database returns to devices.

So should I add these statements to the Security Considerations section?
Consider calling out the possibility of SPECTRUM_USE_NOTIFY messages that are fraudulent (and perhaps how client authentication might protect against them)?


    - The use of the "id" parameter from JSON-RPC deserves more
    discussion. The JSON-RPC spec allows it to be string, numeric
    (without a fractional part), NULL or missing. You've chosen to
    require it (since you're not using json-rpc notifications), and
    not allowing numeric values (why?). Are you making any other
    assumptions about what it should contain? I think you're assuming
    a level of uniqueness that would let you use the Batch mechanism
    in section 6 of JSON-RPC (otherwise, the HTTP request/response
    context is enough to associate the request and response and the id
    might as well be constant).


Purely as a practical matter, handling a known type is easier than multi-type. It should improve interoperability. Otherwise, its value remains opaque to the Database.

What is your recommendation here?
Just adding a short description of the properties that you require of id.


    - Section 7 says a server can reject a GET with a 404 - wouldn't
    that have consequences for a later POST? Why wouldn't it use a 405?


Yes, it's supposed to be 405.


    - The draft calls for the creation of a special list for review
    requests for the IANA assignments. This may be ok (mailing lists
    are easy to set up), but is there not an existing list that would
    serve the purpose just as well?


I do not believe there is an existing list, unless it's standard practice to re-use this discussion group (paws(_at_)ietf(_dot_)org <mailto:paws(_at_)ietf(_dot_)org>).


    Nits

    - RFC 2616 has been obsoleted - the references should be updated.


Updating to RFC 7231


    - It would help to have an example of a ruleset in the definition
    in the Terminology section and perhaps for that definition to more
    strongly convey that it is a name in a namespace, and what that
    rule means is elsewhere.  (Right now the definition says that the
    ruleset is the actual set of rules, not a name, and it made
    reading the protocol overview much harder than it needed to be).
    At the very least, pointing to the examples in section 9 early
    would help.


See proposed text of Terminology above.


    - Is the document loosely borrowing UML, or are the diagrams used
    in section 4 of a format formally defined in some other RFC? A
    pointer to a definition of the format, or a brief description
    noting it's based on UML along with where the base types are
    defined would be useful.


See proposed text of primitive types above. Do I need a reference to UML?


    - "One approach to manage spectrum sharing" is awkward. Would "One
    approach to managing spectrum sharing" or "One approach to the
    management of spectrum sharing" work?


Thanks. Changing to "One approach to managing spectrum sharing"


    - There are several instances of "The Device needs to use the
    information to update its list". Consider clarifying 'needs to'.
    Should this have been MUST?


These were changed based on comments from our AD (Pete), since they are not a protocol requirement.


    - "The vertices MUST be defined in a counter-clockwise direction"
    assumes you are looking at them from above - please be explicit.


Thanks. Adding clarification.


    - Section 9.1.2's first paragraph should say "FCC and ETSI" the
    same way 9.2.2 does. You could generalize that to "any particular
    set of authorities".


Thanks for the suggestion. Changing both to "any particular set of authorities".


    _______________________________________________
    paws mailing list
    paws(_at_)ietf(_dot_)org <mailto:paws(_at_)ietf(_dot_)org>
    https://www.ietf.org/mailman/listinfo/paws




--
-vince