Re: [secdir] secdir review of draft-ietf-storm-ipsec-ips-update-03

["Black, David" <[e-mail]>]

Tom,

I think we should agree to disagree on one point, and I'm prepared to make
text changes to address the others (in particular, you are clearly right
that the hardware implementation rationale for AES GCM instead of AES CTR
should be made clearer).  As previously noted, I plan to hold off on
submitting a revised draft until after the IESG telechat on August 29.

Again, thank you for the review and your comments.

-- Point of disagreement

I firmly believe that filing technical errata against RFC 4106 (AES GCM)
is the appropriate vehicle for the concerns about AES GCM nonce reuse,
including the comparison of security properties to AES CTR.  Such errata will
get the requisite level of review from security experts and cryptographers
who may not look at this draft, and will result in them being seen by
anyone who implements GCM, in contrast to this draft which may not even
be consulted by all IPsec implementers.

I propose leaving this to the Security ADs to determine at this juncture, as
I believe I understand your security concerns, (with which I do not disagree).

Rather, I am asserting that this draft is the wrong vehicle for them, because
(IMHO) they should be directly associated with the AES GCM RFC, and filing
technical errata against that RFC is the appropriate means to accomplish that.

-- Points of agreement

(1) Applicability of security considerations text in other RFCs

> > * OTOH, it would make sense to add a sentence to point out that all of the
> > referenced security RFCs and the security considerations in all of the
> > other referenced RFCs are applicable if that seems useful.
> 
> That would be useful, as would additional text calling out the
> security considerations sections for RFCs related to algorithms whose
> requirement levels this document changes, e.g., AES-CTR (RFC 3686),
> AES-GCM (RFC 4106), and AES-GMAC (RFC 4543).

Ok, I can do that.

(2) Rationale for replacement of AES CTR with AES GCM

> Am I correct in concluding from the above that there were no security
> issues leading to the changes of requirement levels for AES-CTR
> ("SHOULD" to "MAY"), and that you base the advice on hardware
> implementation considerations?  If so, this text in Section 2.2 seems
> potentially misleading:
> 
>    "This document changes both [3DES-CBC and AES-CTR] of these
>    requirements based on cryptographic developments since the
>    publication of RFC 3723."
> 
> unless "cryptographic developments" refers to both the block size
> vulnerability of 3DES and the standardization of AES-GCM.

That is the situation.

> In that
> case, I would suggest adding text to the "AES Counter mode" paragraph
> of Section 2.2 indicating that the new preference of AES-GCM above
> AES-CTR is because of hardware implementation considerations and not
> cryptographic weakness.

Good suggestion - I'll plan on doing that.

(3) Use of 3DES

> I agree that 3DES is no longer appropriate as the "MUST implement"
> confidentiality transform.  I think that because 3DES remains
> available as an optional algorithm, it could be valuable to provide
> advice about when its use might be an acceptable risk (low data rate
> sessions?).

Ok, I can write a sentence to the effect that when the data rate is low
enough that 3DES's rekeying frequency is not a concern, 3DES is a
reasonable algorithm to use.

(4) Key strength matching

As previously noted, I'll do the following:

> > * I'll add some advice about matching strengths of symmetric and asymmetric
> > keys with a pointer to NIST SP 800-57, as there are key length requirements
> > in this draft.

(5) ESP sequence numbers and rekeying

As previously noted, I'll do the following:

> > * ESP sequence number cycling is prohibited (MUST rekey before that happens)
> > by RFC 2404 and RFC 4303, so I don't see any point in discussing its
> > consequences.  OTOH, I think it would be good to call attention to this
> > requirement to rekey before the ESP sequence number rolls over and point out
> > that extended sequence numbers greatly reduce the frequency at which that
> > may be necessary (probably in Section 2 of this draft).

Thanks,
--David

> -----Original Message-----
> From: Tom Yu [mailto:[e-mail]]
> Sent: Thursday, August 22, 2013 7:45 PM
> To: Black, David
> Cc: [e-mail]; [e-mail]; draft-ietf-storm-ipsec-ips-
> [e-mail]
> Subject: Re: secdir review of draft-ietf-storm-ipsec-ips-update-03
> 
> "Black, David" <[e-mail]> writes:
> 
> > I would prefer not to repeat general advice applicable to all uses of
> > IPsec in this draft.  That applies to the exposure of AES GCM to nonce reuse
> > and means of preventing that reuse (e.g., use of IKE) which are covered in the
> > Security Considerations section of RFC 4106; if that advice is inadequate,
> > I'd suggest a revision of RFC 4106 rather than adding advice to this draft.
> 
> I would tend to agree about not repeating general advice applicable to
> all uses of IPsec.  This document does replace AES-CTR with AES-GCM,
> and the consequences of nonce reuse with these modes are qualitatively
> different and possibly bear mentioning.
> 
> I think the text in RFC 4106 about the risks of nonce reuse should be
> more specific about the consequences.  (RFC 4106 says "destroys the
> security guarantees of AES-GCM mode" instead of "reveals the
> authentication subkey".)  Compared to most other cipher modes, nonce
> reuse in GCM has surprisingly catastrophic consequences.  Perhaps RFC
> 4106 could use revision, but that would not benefit readers of this
> document until that revision was published.
> 
> > * OTOH, it would make sense to add a sentence to point out that all of the
> > referenced security RFCs and the security considerations in all of the
> > other referenced RFCs are applicable if that seems useful.
> 
> That would be useful, as would additional text calling out the
> security considerations sections for RFCs related to algorithms whose
> requirement levels this document changes, e.g., AES-CTR (RFC 3686),
> AES-GCM (RFC 4106), and AES-GMAC (RFC 4543).
> 
> Because this document replaces one "SHOULD" requirement level AES
> counter mode with another, I think it would also be useful to contrast
> the considerably different consequences of nonce reuse in AES-CTR
> (confidentiality loss limited to the blocks with colliding counter
> values) compared to AES-GCM (additionally reveals the authentication
> subkey).
> 
> The relative severity of these vulnerabilities might be debatable in
> the context of a block storage protocol.  If you believe the security
> impacts to block storage protocols caused by these nonce reuse issues
> are negligibly different, the document might not need to make that
> comparison.  (Although I would be interested in hearing the reasoning
> behind such a belief.)
> 
> > As explained in Section 2.2, the hardware implementation considerations
> > originally used to select AES-CTR as a "SHOULD implement" transform now
> > favor AES-GCM.  Based on those considerations, AES-GCM is a significantly
> > better choice due to its hardware-friendly hash, so retaining the "SHOULD"
> > for AES-CTR doesn't make sense.
> 
> Am I correct in concluding from the above that there were no security
> issues leading to the changes of requirement levels for AES-CTR
> ("SHOULD" to "MAY"), and that you base the advice on hardware
> implementation considerations?  If so, this text in Section 2.2 seems
> potentially misleading:
> 
>    "This document changes both [3DES-CBC and AES-CTR] of these
>    requirements based on cryptographic developments since the
>    publication of RFC 3723."
> 
> unless "cryptographic developments" refers to both the block size
> vulnerability of 3DES and the standardization of AES-GCM.  In that
> case, I would suggest adding text to the "AES Counter mode" paragraph
> of Section 2.2 indicating that the new preference of AES-GCM above
> AES-CTR is because of hardware implementation considerations and not
> cryptographic weakness.
> 
> > The comments on 3DES seems off the mark to me.  The IETF approach to
> > security, as I've understood it, is to require implementation of
> > sufficient security mechanisms so that they will be available if needed.
> > Since the protocols to which this draft's requirements apply will operate
> > at high data rates in some environments (e.g., see discussion in Section
> 2.2),
> > 3DES is no longer appropriate as the single "MUST implement" confidentiality
> > transform.  Further, the comparison to AES-GCM seems misplaced, as AES-CBC
> > is the new "MUST implement" transform, not AES-GCM.
> 
> I agree that 3DES is no longer appropriate as the "MUST implement"
> confidentiality transform.  I think that because 3DES remains
> available as an optional algorithm, it could be valuable to provide
> advice about when its use might be an acceptable risk (low data rate
> sessions?).
> 
> > * I'll add some advice about matching strengths of symmetric and asymmetric
> > keys with a pointer to NIST SP 800-57, as there are key length requirements
> > in this draft.
> >
> > * ESP sequence number cycling is prohibited (MUST rekey before that happens)
> > by RFC 2404 and RFC 4303, so I don't see any point in discussing its
> > consequences.  OTOH, I think it would be good to call attention to this
> > requirement to rekey before the ESP sequence number rolls over and point out
> > that extended sequence numbers greatly reduce the frequency at which that
> may
> > be necessary (probably in Section 2 of this draft).
> >
> > The three paragraphs noted by '*' above indicate the three changes that I'd
> > propose to make in response to this review - is that reasonable?
> >
> > With this draft in IESG Evaluation for next week's telechat (August 29),
> > I would plan to defer these changes until after the telechat.
> 
> Your proposed changes seem reasonable.