phc-discussions - Re: [PHC] Overview of PHC Candidates and Garbage-Collector Attacks

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Thu, 30 Oct 2014 14:44:46 -0400
From: Bill Cox <waywardgeek@...il.com>
To: "discussions@...sword-hashing.net" <discussions@...sword-hashing.net>
Subject: Re: [PHC] Overview of PHC Candidates and Garbage-Collector Attacks

On Thu, Oct 30, 2014 at 1:03 PM, Jakob Wenzel <jakob.wenzel@...-weimar.de>
wrote:

> OK. I just had a close look to your code and now I am convinced. I
> just overlooked that PRK is overwritten for each increment of the
> garlic factor, which, when starting with MinGarlic = 0, is indeed at
> an early stage of the hash generation, rendering WGC attacks not
> applicable. Nevertheless, if I understood your code and explainations
> right, skinnycat does not support the parameters startMemCost and
> stopMemCost and thus, the proposed WGC attacks on the value PRK seems
> to work.
>

I agree.  SkinnyCat sacrificed many defenses (mostly suggested by
Alexander) to slim down.  It also lacks GPU defense through the small
unpredicable memory reads, and multiplication chain based compute time
hardening, parallel threads, adjustable data lanes (to match the SIMD
unit), and a lot of tunables to optimize for a particular platform.


> >
> >> By default, Catena runs with minGarlic == maxGarlic == 18.  In
> >> this mode, Catena does not begin to overwrite memory derived from
> >> the password until it has finished filling memory.  During this
> >> entire time, H(H(key material)) is present in memory.  If Catena
> >> tries to use too much memory, this memory might get swapped to
> >> disk.  If Catena is running continuously, as it might on an
> >> authentication server, with lambda == 3, there is about a 1 in 4
> >> chance that a cold-boot attack, DMA attack, or forced
> >> hibernation, will reveal H(H(key material)).
> >>
> >
> > For Catena-BRG, this is indeed an attack which succeeds with a
> > chance of $1/(lambda+1)$. Thanks for pointing this out! We will
> > recommend to use Catena-BRG with MinGarlic = 1 in the next version
> > of our submission paper to thwart this attack.
> >
> >
> > That would slow you down 2X, meaning you would lose another 2X in
> > memory*time defense against brute-force password guessing attacks.
> >  Instead, please consider doing what TwoCats does, and apply
> > Garlic starting at 0, but when you get close to the m_cost, just
> > skip ahead and to the last level of garlic.  This enables your
> > algorithm to trade-off defense against brute-force guessing attacks
> > vs memory-leak attacks.
> >
> > I certainly have copied many good ideas from Catena.  It would be
> > nice to contribute back something :-)
> >
>
> We will indeed consider this idea! Thanks!
>
> Best regards,
> Jakob
>

You're welcome.  Catena still rocks.  I learned a ton from Catena.

Bill

Content of type "text/html" skipped