Date: Tue, 14 Jul 2015 13:55:40 +0200
From: Jakob Wenzel <jakob.wenzel@...-weimar.de>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Overview of PHC Candidates and Garbage-Collector Attacks

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On 05.07.2015 21:57, Solar Designer wrote:
> Hi Jakob,
> 
> On Thu, Jul 02, 2015 at 05:05:50PM +0200, Jakob Wenzel wrote:
>> we have updated the classification document (including analysis 
>> regarding to (weak) garbage-collector attacks -- (W)GCA).
>> 
>> See: https://eprint.iacr.org/2014/881
>> 
>> Among other minor changes, the update includes: 1) Argon2d and
>> Argon2i (as two instantiations of the finalist Argon2) 2)
>> yescrypt now provides (W)GCA resistance under certain
>> requirements depending on the input parameter
> 
> It is unclear from your description whether you think yescrypt
> provides GC resistance at t > 0 and/or g > 0, and why.  Can you
> clarify?  You first list t = 0 among requirements for yescrypt's GC
> resistance, and then describe how things change at t > 0 and/or g >
> 0, but you seem to never clearly state whether it's GC attack
> resistant at those settings. (And there's a missing closing brace,
> but this doesn't affect meaning.)

Hi Alexander,

Thanks for the comment! My explaination was indeed not as clear as it
should have been. I updated the yescrypt part by considering the
following cases:

g - cost upgrade parameter (client-independent update increasing time)
V - array in RAM
N - defining the size of the array V (memory cost in RAM)
p - number of threads
r - memory per thread

1) no flags are set and g = 0
scrypt compatibility mode (when used without ROM) => same attacks as
for scrypt applicable.

2) no flags are set and g >= 0
yescrypt is vulnerable to WGC atacks since at least one full
invocation of the time- and memory-consuming core of yescrypt has to
be invoked before the password is overwritten.

3) YESCRYPT_RW is set and g = 0
The second loop of ROMix (Line 6-9) performs less than N writes to V
if t = 0 and if (t = 1 and N >= 8). GC attacks similar to that for
scrypt are applicable but with higher effort since V is at least
partially overwritten. For t > 1, it is most likely that the whole
state V is overwritten, making yescrypt resistant to GC attacks

4) g > 0
yescrypt provides resistance to GC attacks since V is overwritten at
least once (second invocation of first loop (Lines 2-5) of ROMix).

5) YESCRYPT_RW is set, p >= 1, N/p >= 256, N/p * r >= 2^17
Under these requirements, a 64-times smaller instance of yescrypt is
invoked before the full yescrypt. Then, yescrypt overwrites the
password significantly fast, hence, providing WGC resistance.


It would be really cool, if point 5) would also be mentioned in the
specification of yescrypt (and not only the reference code).


> On a related note, the tweaked yescrypt defines client-independent 
> updates, so you may check this in Table 2.

Done.


An updated version of the classification is available here:
http://eprint.iacr.org/2014/881



Best regards,
Jakob


- -- 
Jakob Wenzel
Research Assistant
Chair of Media Security (Prof. Lucks)
Bauhausstraße 11 (Room 217)
99423 Weimar
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2

iQEcBAEBCAAGBQJVpPi8AAoJEDFlRQsgEDnDR50H/RKvNdyqVx6pXlxClsOe/FXq
W6doRbJOEwg6QAlAN1vzx+40wChq1wbhyBPor8mgeEKy6ui4XQ72Ye4u/VAqST2G
EByRPueBTViUE5J0FyZtxAuiYhYpfHFRmQH8afN0vXJBgauxzG43z3Y2KI9BCeO2
qeUxfl+HzZnV3GHWtgdrAk4RCn0x2nfSrjpHtQi1PxwEgw+35JWfhYrQIOIM98jf
/ioWwRD/G+AFj/+oTpMhUJQM5RRWifvxa6aktWkVClH4uNf9o5ypqW9xhPiF6V02
wFCFR1rVwN1t80U53khu2OuaYI6Gd3sbrbsg7PMkJnk/TvGcY3zXxpSddremu8Y=
=+ZJC
-----END PGP SIGNATURE-----

Powered by blists - more mailing lists