Re: [jose] [apps-discuss] Appsdir review for draft-ietf-jose-json-web-algorithms-33

[Mike Jones <[e-mail]>]

Replies to your replies are inline below.  Thanks again for your thoughtful and detailed review.

				-- Mike

> -----Original Message-----
> From: Carsten Bormann [mailto:[e-mail]]
> Sent: Wednesday, October 15, 2014 2:55 AM
> To: Mike Jones
> Cc: [e-mail]; draft-ietf-jose-json-web-
> [e-mail]; Matt Miller; [e-mail]; [e-mail]
> Subject: Re: [apps-discuss] Appsdir review for draft-ietf-jose-json-web-
> algorithms-33
> 
> Hi Mike,
> 
> thanks for getting to this review - I don't envy you for this herculean job.
> A few more comments/clarifications inline below.
> 
> Grüße, Carsten
> 
> On 15 Oct 2014, at 10:51, Mike Jones <[e-mail]> wrote:
> 
> > Thanks for your review, Carsten.  I apologize for not responding to it until
> now.  It had gotten sorted into a folder and I wasn't aware of it until Kathleen
> brought it to my attention.  Replies are inline below...
> >
> >> -----Original Message-----
> >> From: apps-discuss [mailto:[e-mail]] On Behalf
> >> Of Carsten Bormann
> >> Sent: Thursday, October 02, 2014 12:22 AM
> >> To: [e-mail]; draft-ietf-jose-json-web-
> >> [e-mail]
> >> Cc: [e-mail]; [e-mail]
> >> Subject: [apps-discuss] Appsdir review for
> >> draft-ietf-jose-json-web-algorithms-
> >> 33
> >>
> >> I have been selected as the Applications Area Directorate reviewer
> >> for this draft (for background on appsdir, please see
> >> http://trac.tools.ietf.org/area/app/trac/wiki/ApplicationsAreaDirecto
> >> rate
> >> ).
> >>
> >> Please resolve these comments along with any other Last Call comments
> >> you may receive. Please wait for direction from your document
> >> shepherd or AD before posting a new version of the draft.
> >>
> >> Document:  draft-ietf-jose-json-web-algorithms-33
> >> Title: JSON Web Algorithms (JWA)
> >> Reviewer: Carsten Bormann
> >> Review Date: 2014-10-02
> >> IESG Telechat date: 2014-10-02
> >>
> >> Summary: This draft is ready for publication as a standards track
> >> RFC, with a few nits corrected.
> >>
> >> However, some additional editorial improvements might improve the
> >> security outcome when it is referenced by application developers.
> >>
> >> Major issues: None.
> >>
> >> Minor issues:
> >>
> >> 5.2:
> >> Add a reference that defines PKCS #7 padding.
> >
> > I'll plan on adding a reference to RFC 2315 for this.
> 
> That would work (section reference would be 10.3) Alternatively, you could
> reference RFC 5652 (section 6.3), which is an Internet Standard (STD70).
> 
> >> 5.2.2.2
> >> Does "the PKCS #7 padding is removed" entail checking all of its bytes?
> >
> > Should this be changed to "the PKCS #7 padding bytes are checked and then
> removed"?
> 
> I think it is good to alert the reader to this need.
> 
> >> 6.2.1
> >> Is the intention that the sentence containing "point compression is
> >> not supported" also applies to any future registered value of "crv"?
> >> A similar comment applies to other specifications in 6.2.1.x, e.g.,
> >> the reference to SEC1 representation to x and y.
> >
> > It would apply to any future curves registered that use the "x", "y" point
> representation.  Others could define new key types that use a different
> representation or we could refine the definition of "kty":"EC" to make the
> representation specific to the "crv" value.
> >
> > A discussion happened on the JOSE thread "JWK Elliptic Curve key
> representations and new curves".  The conclusion suggested by Richard Barnes
> and seconded by Stephen Farrell was to table defining any new key
> representations for new elliptic curves until the CRFG recommendations to TLS
> have been made.  That ended the discussion for now.
> 
> Right, I didn't want to anticipate that discussion; I would be happy if it didn't
> appear that the avenue for any evolution on point compression/compact points
> were closed.

I can add language saying that future curve registrations may use different parameters to represent the curve value.  That may get a little convoluted because currently it's the "kty" that determines which parameters are used.  But I agree with trying to accommodate new key representations for new curves.

> >> 6.2.1.1
> >> »Additional "crv" values MAY be
> >> used, provided they are understood by implementations using that
> >> Elliptic Curve key.« How are conflicts between such implementation
> >> defined values and future registered values handled?
> >
> > Conflicts are avoided using the IANA JSON Web Key Elliptic Curve registry
> defined in Section 7.6.
> 
> Oh, but that's not what the text says.  I read this as a general license to use
> random locally defined crv values.  Maybe the need for registration needs to be
> clarified.

I'll delete the sentence "Additional 'crv' values MAY be used, provided they are understood by implementations using that Elliptic Curve key."  The registry is already talked about in the previous sentences.

> >> 6.3.2:
> >> The MAY accept partially overrides the MUST include?
> >> Is the latter thus really a SHOULD?
> >
> > "d" is REQUIRED.  The others SHOULD be included, and if any are included, the
> others MUST be included.  Those are requirements on producers.
> >
> > Consumers may choose to accept keys that don't include all the CRT
> parameters.  Given that they can be computed from "n", "e", and "d" and
> sometimes they're not available, a previous reviewer had asked this language
> about consumers to be included.  It's a case of "Be conservative in what you
> send, be liberal in what you accept".
> 
> [I'm not inserting my usual rant here that this bon mot may apply to
> implementations, but never to specifications.]
> 
> > It's not clear to me that there's any value in relaxing the requirements on
> producers.
> 
> Well, if the producer MUST send "all of the others" if "any of the other"
> parameters are present, it is not clear to me what the point of "MAY choose
>    to accept an RSA private key that does not contain a complete set of
>    the private key parameters other than "d"" is other than enabling producers to
> send just that.
> (More importantly, it is not clear when a producer should be taking that license.)

I can delete the sentence " The consumer of a JWK MAY choose to accept an RSA private key that does not contain a complete set of the private key parameters..."

> >> 7.1:
> >> It is interesting that a mere registration (vetted only by a DE) can
> >> change the IETF consensus base specifications by making an algorithm
> "Required".
> >
> > It is expected that the designated experts will be appointed, in part, for their
> cryptographic expertise.
> 
> There has been more discussion about the requirements levels here that I don't
> want to repeat.
> My comment here was that a DE (even with the best crypto knowledge) is in a
> hard place updating the consensus about *desirability of implementation* -
> there are more aspects to this than the technical ones.

Thinking about this some more, I agree that "Required" is a special case.  I'm thinking that a reasonable safeguard for this case is to require that approval of both a designated expert and a sitting security area director be required to make an algorithm "Required" or to change it from "Required" to something else.  Kathleen and Carsten, would that work for you?

> >> 8.
> >> I am unable to find a "security considerations" section in NIST SP 800-38A.
> >> 800-38D at least has a "practical considerations" section, is that meant?
> >> (Etc., I haven't checked all the references.) In general, I believe a
> >> security considerations section is most useful where it provides more
> >> directed guidance instead of saying the equivalent of "here is a textbook".
> >
> > There are 14 subsections with directed guidance, plus more in the JWS, JWE,
> and JWK specifications.  If there is other directed guidance you'd like to see
> added, please supply proposed text.
> >
> > Having skimmed through it, I agree that 800-30A doesn't contain clear enough
> security guidance to merit inclusion in the list.  I'll remove it.  If there are other
> specifications that you'd like to see added or removed, please let me know.
> 
> This was more of a general comment that "Here is a reference to a textbook"
> type security considerations are less useful than specific, directed references.  I
> haven't checked all those referenced documents, so I'm not in a position to
> recommend changes here.  Maybe something for a checklist to apply to the next
> security spec.
> 
> >
> >> 8.7 is not clear: is it NOT RECOMMENDED to reuse an entire set of key
> >> material (including IV), or to reuse any part of it?
> >
> > I agree that this is ambiguous, as worded.  Its advice seems to be so broad as
> to be impractical, as never reusing a key means that a new key would have to be
> redistributed for each encryption, which then requires a key to use for the
> distribution...  This text came from http://tools.ietf.org/html/draft-miller-jose-
> jwe-protected-jwk-02#section-8.1, which was incorporated into the document
> after a working group decision to do so.  Matt Miller, you were the author of
> this text.  Could you clarify what the intent was?  Thanks.
> >
> >> Nits/editorial comments:
> >>
> >> 6.3.2.x:
> >> The constant repetition of »It is represented as the base64url
> >> encoding of the value's unsigned big endian representation as an octet
> sequence.
> >> The octet sequence MUST utilize the minimum number of octets to
> >> represent the value.« almost ensures that an implementer will stop
> >> reading the details (well, I did, and I did not write a program to
> >> verify the same phrase is used everywhere; if any parameter were
> >> using a different encoding, that sure would be missed).  Why not define
> another abstraction like base64url and use this?
> >
> > I did just verify that they are all consistent.  Yes, if you're reading the spec
> linearly I'm sure it feels repetitive, but if you're an implementer going straight to
> a definition to implement it, having a complete description all in one place is
> valuable.  I understand your request but it's not clear to me that this rises to the
> level that a separate definition that the developer then has to go find is
> warranted.
> 
> My point was corroborated by the tiny error Richard found.  Repeating the same
> information again and again drowns out the important information.  A single
> paragraph at the end of 6.3's intro of the form:
> 
>    Each of the positive integer parameters defined in this section
>    is represented as the base64url encoding
>    of the value's unsigned big endian representation as an octet
>    sequence.  The octet sequence MUST utilize the minimum number of
>    octets to represent the value.

OK - I'll see what I can do in this regard.  Editorially, it's a little bit complicated because there are different subsections for public and private key parameters.

> Note that the repetitive text already has a reference that needs to be looked up
> in that it uses the term "base64url encoding", which has a slightly refined
> meaning here from that in RFC 4648.  So an even better way to handle this
> would be to add a new term "posint encoding" or some such to section 1.1,
> turning e.g. the entirety of 6.3.2.2 into
> 
>            The "p" (first prime factor) member contains the first prime factor, in
> POSINT encoding.
> 
> 
> >
> >> 6.2.3.1: This is not a positive integer?  6.2.3.x mentions this otherwise.
> >
> > I may not understand this remark.  I'm guessing you're referring to 6.3.2.1 and
> the fact that the description includes the word "unsigned".  This is there to make
> it clear that no sign bit will be present in the representation - which is especially
> important since the high-order bit is always set for "n" and "d" for correctly
> generated keys.
> 
> (Sorry, this is indeed about 6.3.2.x, not 6.2.3.x.) That was not what I meant:  All
> 6.3.2.x have the form »The "p" (first prime factor) member contains the first
> prime factor, a positive integer.«
> 6.3.2.1 does not have the ", a positive integer", but I can't see a reason for that
> (d clearly cannot be zero, and it is still an unsigned value).
> This is likely to confuse an implementer that reads this text closely.
> (And I'm offering the fact that you are overlooking that inconsistency as another
> corroboration of my previous point.)

I can delete the unnecessary phrase "a positive integer", subject to how it does or doesn't fit into the edit above that you asked for.

> >> 7.1.1
> >> »Example description« is not a useful example for an "Algorithm Description".
> >> (Same comment for 7.x.1.)
> >
> > See the actual registrations in Section 7.1.2 for more useful examples.
> 
> Maybe just take out the not so useful examples in 7.x.1?

All other registration templates that I have seen include examples of all parameters.  This is only one of the parameters in the template, and all need to be retained.  Would you be happier with the e.g. text "Example algorithm"?

> >> 8.3:
> >> s/because it/because it is/
> >
> > Good catch
> >
> >> [sec1]
> >> (Given the date, this is probably referencing V2.0 of this spec.)
> >
> > Hmmm - the version I have cached locally came from
> http://www.secg.org/collateral/sec1_final.pdf and is dated September 20,
> 2000.  But the site appears to be down at present.  The version referenced by
> Wikipedia at http://www.secg.org/download/aid-385/sec1_final.pdf is missing
> too. http://secg.org/download/aid-780/sec1-v2.pdf is missing too.  Can you
> point me to a good link?
> 
> I just had a look and embarrassingly found that the copy I'm using comes from
> the personal space of a researcher:
> http://perso.univ-rennes1.fr/sylvain.duquesne/master/standards/sec1-v2.pdf
> Not very satisfying as a reference.
> 
> >
> > RFC 5915 includes this reference:
> >
> >   [SECG1]    Standards for Efficient Cryptography Group (SECG), "SEC 1:
> >              Elliptic Curve Cryptography", Version 2.0, May 2009.
> 
> That looks good: We don't need a URL, as long as we are certain which version
> we are referencing.
> 
> > Worst comes to worst, I'll either reference that or explicitly reference the 1.0
> version.  But if I'm going to reference the 2.0 version, I'd like to be able to read a
> copy so that I can verify that the way we're using it is consistent with the actual
> spec, including the section numbers.
> >
> >> [usascii]
> >> The reference to ANSI X3.4:1986 should probably be replaced by a
> >> reference to RFC 20.  There is little reason to reference a somewhat
> >> hard to obtain external document ($60!) when we have an RFC about the
> same subject.
> >
> > OK
> >
> >> (Tables in Appendix A need some formatting.)
> >
> > Agreed.  It's on my to-do list.  Jim Schaad recently told me to "Add a width
> attribute to the ttcol with either "digits" (for fixed width) or "digits%" (for
> percent width) to the column in question".  I plan to give it a try.
> >
> >> _______________________________________________
> >> apps-discuss mailing list
> >> [e-mail]
> >> https://www.ietf.org/mailman/listinfo/apps-discuss
> >
> > 				Thanks again,
> > 				-- Mike
> >
> >
> >