Re: [Cfrg] RGLC on draft-irtf-cfrg-kangarootwelve-01

[Stephen Farrell <[e-mail]>]

Hiya,

On 20/02/2020 11:37, Benoît Viguier wrote:
> Hi Stephen,
> 
> To answer your question:
> 
> Yep, I had a quick look and didn't find what I thought
> I was remembering;-) It may be that what I was thinking
> about was HKDF (RFC 5869), so let me ask it that way:
> is there a clear benefit in using k12 instead of using
> HKDF to get the output length wanted? If there is, and
> it's possible to state that in a short paragraph, that
> might be a useful addition. (While HKDF doesn't have
> arbitrary output length, it can produce long enough
> outputs for many uses I guess.)
> 
> -------------------------------------------------------------------------
> 
> K12 has the merit of being a simpler construction and can be used in
> place of HKDF.
> 
> In the scope of IoT or embedded devices, another advantage of K12 when
> its input contains a key is that it is easier to protect against DPA
> than SHA-1 or SHA-2.

Fair enough. Doesn't seem like a compelling addition to
the document in the end, but happy to leave it to you
and others to figure if adding that is useful or not.

Cheers,
S.

PS: Just to be clear, the discussion has moved me from
an "unsure" state to where I think this is ok to publish
in an RFC.


> 
> Kind Regards
> 
> On 2/18/20 4:18 PM, Stephen Farrell wrote:
>> Hiya,
>>
>> Coupla follow ups below...
>>
>> On 18/02/2020 14:45, Benoît Viguier wrote:
>>> Dear Stephen,
>>>
>>> Thank you for your comments on our draft!
>>>
>>> Please find below some answers and comments.
>>>
>>> On 2/16/20 3:53 PM, Stephen Farrell wrote:
>>>> Hiya,
>>>>
>>>> On 16/02/2020 11:16, Alexey Melnikov wrote:
>>>>> Dear CFRG participants,
>>>>>
>>>>> This message is starting 2 weeks RGLC on
>>>>> draft-irtf-cfrg-kangarootwelve-01 ("KangarooTwelve"), that will end
>>>>> on March 1st 2020. If you've read the document and think that it is
>>>>> ready (or not ready) for publication as an RFC, please send a message
>>>>> in reply to this email or directly to CFRG chairs
>>>>> ([e-mail]). If you have detailed comments, these would
>>>>> also be very helpful at this point.
>>>> I had my 1st read of this and think it needs a bit of
>>>> work, but am overall unclear if it ought be published
>>>> now. (I also looked back at the list archive messages
>>>> referring to k12.)
>>>>
>>>> I don't think CFRG ought be publishing very novel
>>>> algorithm RFCs and am unclear how much study k12 has
>>>> gotten outside the author team. The main reference
>>>> given [1] has pointers to lots of work with titles that
>>>> mention reduced-round keccak but it's unclear (to me,
>>>> not having read 'em;-) how  relevant those are to k12.
>>> Actually, all cryptanalysis of Keccak/SHA-3 is relevant to K12.
>>> Cryptanalysis resources are scarce, we chose as an explicit design goal
>>> to make K12 rely on cryptanalysis of Keccak/SHA-3.
>>>
>>> To be more precise, K12 is made of two layers:
>>>
>>> 1) The inner function F. This layer relies on cryptanalysis. K12's F
>>> function is exactly Keccak[r=1344, c=256] (as in SHAKE128) reduced to 12
>>> rounds (no tweaks!). Hence, any reduced-round cryptanalysis on Keccak is
>>> also a reduced round cryptanalysis of K12's F (provided the number of
>>> rounds attacked is not higher than 12 of course).
>>>
>>> 2) The tree hashing over F. This layer is a mode on top of F that does
>>> not introduce any vulnerability thanks to the use of Sakura coding
>>> proven secure in the paper [Bertoni et al., ACNS 2014].
>>>
>>> This reasoning is detailed and formalized in the paper [Bertoni et al.,
>>> ACNS 2018], which is peer-reviewed.
>> Thanks. It sounds like referring to that section
>> of that paper would help so. (Your explanation there
>> helped me fwiw.)
>>
>>>> [1] is also used as [KECCAK_CRYPTANALYSIS] in the draft
>>>> and is where the authors are pointing us to find
>>>> security analysis of k12. I don't think an author-
>>>> maintained web page like that is really good enough
>>>> as the key reference for an RFC like this.
>>> Good point indeed!
>>>
>>> One option would be that we put all the references in the RFC. We are
>>> not sure this is the right place for it, as we see an RFC more as
>>> implementation guide rather than an analysis paper. For example:
>>> https://tools.ietf.org/html/rfc7748 does not mention anything with
>>> respect to the security of X25519 which is detailed in section 3 of
>>> https://www.iacr.org/cryptodb/archive/2006/PKC/3351/3351.pdf.
>>> Nevertheless, we are open to this option.
>>>
>>> Another option would be to refer to [K12, Section 5]. In that section
>>> attacks with the highest number of rounds are discussed and
>>> corresponding paper are referred to. 
>> Yep, that sounds good to me.
>>
>>> However with the possible
>>> appearance of new cryptanalysis results, this section would become outdated.
>> Just to be clear - including the URL as an additional
>> reference as well seems like a good idea. It's only
>> depending on that as the main source that seems a bit
>> iffy.
>>
>>>> I totally get why the authors would find that
>>>> easier/better, but that won't be true for a reader in
>>>> 20 years time, so better to put in the precise
>>>> references now. That should also clarify the extent to
>>>> which those are about k12 (as defined here) and not
>>>> about something else.
>>> We agree with this, however Keccak cryptanalysis directly applies to K12.
>>>> I think the references really need fixing before this
>>>> ought be published. I'm not sure where the right line
>>>> ought be drawn in terms of maturity of an algorithm
>>>> before CFRG blesses it with an RFC. <4 years does seem
>>>> short, even if this is strongly based on keccak. (Hence
>>>> me being unsure overall.)
>>> For us the maturity amounts to more than 4 years. Being SHA-3, Keccak is
>>> a high-profile cryptanalysis target. As we did not tweak the round
>>> function, K12 relies on sustained cryptanalysis since 2008.
>> Ack. If nobody's gotten near breaking 12 rounds and
>> since you didn't tweak it, that does sound convincing
>> to me. Thanks.
>>
>>>> Separately, the draft could benefit from some guidance
>>>> as to when this is thought to be useful - presumably
>>>> that's when one wants a hash output that's >512 bits.
>>>> IIRC, there are other RFCs (forget numbers, sorry),
>>>> that describe how to get such outputs using standard
>>>> hash functions. If there are such RFCs, it'd be good
>>>> to reference those. Either way saying why and when this
>>>> is preferable to use of standard hash functions would
>>>> be good.
>>> The standard hash functions (SHA-1 and SHA-2) are subject to length
>>> extension attacks.
>> Yep, I had a quick look and didn't find what I thought
>> I was remembering;-) It may be that what I was thinking
>> about was HKDF (RFC 5869), so let me ask it that way:
>> is there a clear benefit in using k12 instead of using
>> HKDF to get the output length wanted? If there is, and
>> it's possible to state that in a short paragraph, that
>> might be a useful addition. (While HKDF doesn't have
>> arbitrary output length, it can produce long enough
>> outputs for many uses I guess.)
>>
>> Cheers,
>> S.
>>
>>
>>> And in all cases, K12 will be faster than SHA-3. If
>>> that makes it clearer, we could add a short guidance as to when the user
>>> can benefit the most from K12. We could simply reuse the arguments from
>>> these slides: https://benoit.viguier.nl/files/K12atMontreal.pdf
>>>> Other than the above, the draft seems clear (though
>>>> I didn't try implement) and many thanks for not
>>>> defining the usual pile of pointless variants/options:-)
>>> Thanks, yes, that was another design goal for K12.
>>>