The draft currently has all three: standalone ChaCha, standalone Poly1305,
and AEAD.

Standalone Poly1305 has the issue that it requires a one-time key or a
nonce to generate it, but ESP only allows an IV for the cipher, not for the
MAC. The draft deals with it by using the replay counter as a nonce, but
then changes APIs. That is one reason why some people are opposed to
standalone Poly1305.

Yoav


On Mon, Mar 10, 2014 at 10:43 AM, Daniel Migault <[email protected]>wrote:

> Hi,
>
> My understanding is that Poly1305 and chacha20 are provided as
> "alternatives" to SHA* or AES*. Specific devices with AES
> accelerations may be willing, for performance optimization, to use
> Poly1305 instead of SHA with AES. For this reason it might be better
> to have:
>     - chacha20 as a stand-alone cipher
>     - Poly1305 as a stand-alone MAC
>
> On the other hand, providing the AEAD chacha20-poly1305 may be helpful
> for end users or admins. Especially if security consideration
> recommends AEAD. Would it bring too much complexity to also define
> AEAD chacha20-poly1305?
>
> BR
> Daniel
>
>
>
> On Mon, Mar 10, 2014 at 9:15 AM, Yoav Nir <[email protected]> wrote:
> >
> >
> >
> > On Mon, Mar 10, 2014 at 8:00 AM, Yaron Sheffer <[email protected]>
> > wrote:
> >>
> >> Hi Yoav,
> >>
> >> Can you explain why we need Poly1305 at all? We have SHA-2 and will
> >> probably adopt Keccak (SHA-3), so it's not like we don't have a backup.
> >
> >
> > Sure.  Poly1305 is fast.Faster than SHA-1, SHA-2, and Keccak. I haven't
> > compared it to GMAC on Intel, but that is fast only becuase it has
> special
> > Intel instructions like PCLMULQD. Both ChaCha and Poly1305 can be fast
> in a
> > plain C implementation, so they're fast on any platform.  Poly1305 needs
> > another algorithm to generate the per-message keys. That could be AES as
> in
> > DJB's original paper, or it can be ChaCha as in this draft (with or
> without
> > the AEAD).
> >
> >>
> >> Let me suggest that we adopt *only* ChaCha20, which can be combined with
> >> any integrity protection algorithm in the normal ESP way. Is there any
> extra
> >> value (maybe code sharing?) in predefining an AEAD?
> >
> >
> > The AEAD version is already in at least one crypto library (NSS as used
> in
> > Chrome) and there's a patch that AGL donated to OpenSSL (not in there
> yet).
> > So in addition to AEADs being fashionable, this combination makes sense
> for
> > performance, especially on non-Intel platforms.
> >
> > Yoav
> >
> >
> > _______________________________________________
> > IPsec mailing list
> > [email protected]
> > https://www.ietf.org/mailman/listinfo/ipsec
> >
>
>
>
> --
> Daniel Migault
> Orange Labs -- Security
> +33 6 70 72 69 58
>
_______________________________________________
IPsec mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/ipsec

Reply via email to