[Gnuk-users] FST-01 is going to be non-reproducible any more

[Bertrand Jacquin]

Hi,

I can't say I'm really a fan of Bluetooth for security concern, so if 
this feature can be disable, I will be OK with that. Will is it possible 
to disable the piezo ?

A few features I would like to see are:

- It would be great to make use of the internal storage chip as a read 
only storage and make available the public_key as a plain text file
- NFC on Yubikey is I think a great to have, but I made no research 
regarding the security concern and how far can you can data from
- Something more tiny and discrete

Cheers

On 05/01/2016 06:50, NIIBE Yutaka wrote:
> Hello,
> 
> While updating the Seeed wiki page:
> 
>     FST-01 Wiki:
>     http://www.seeedstudio.com/wiki/FST-01
> 
> I realized that one of the parts (specifically, U2, LDO V-Regulator,
> CAT6217-330TDGT3) is discontinued.
> 
> Still, I can see QTY=21,000 is avaiable at Rochester, though:
> 
>    https://www.rocelec.com/parts/results/all/?s=CAT6217-330TDGT3
> 
> ... and we have alternative parts, too.
> 
> 
> 			*	*	*
> 
> Well, 2016 would be the good timing to consider another hardware
> design.
> 
> In this holiday season, I was considering use of Bluetooth for a
> crypto token, but my conclusion was: it's not good idea.  I concluded
> that the use of Bluetooth should be limited to some HID feature, only.
> 
> If using Bluetooth, I think that the size and complexity of other
> parts (than public key crypto for OpenPGP) will be bigger because of
> the protocol stack of Bluetooth and its encrypted communication
> channel.  So, the major security risks will not be determined by Gnuk
> implementation itself.
> 
> I am considering to release Gnuk 1.2.0.  And to start development of
> new branch, removing RSA.  Perhaps, it will be only with Curve25519
> and Ed25519.
> 
> I think that Cortex-M0+ (with 1-cycle 32-bit x 32-bit -> 32-bit
> multiplier) running at 48MHz can be good token for OpenPGP (or SSH).
> 
> While Cortex-M3 does 32-bit x 32-bit -> 64-bit multiplication takes
> 3-5 cycles and it is not constant-time (it skips cycles for smaller
> values).  I don't think it is easy to mount this fact to build a real
> attack, but 1-cycle 32-bit x 32-bit -> 32-bit multiplier would sound
> good for constant-time lovers.
> 
> I'll port Chopstx to Cortex-M0+.  Then, I'll consider Gnuk25519.
> 
> User interface of a token is also important.  I'm considering
> something like HID device; a piezo buzzer for notification and a
> button for acknowledgment/confirmation.

-- 
Bertrand