phc-discussions - Re: [PHC] Password hashing as a self-overwriting Turing machine: new version (July 3)

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]

Message-ID: <CAOLP8p5rvGoQoRpSbGn8c5sQ-vkwDMwcYq4FCH2vY0Tq_jhzCg@mail.gmail.com>
Date: Fri, 3 Jul 2015 05:55:16 -0700
From: Bill Cox <waywardgeek@...il.com>
To: "discussions@...sword-hashing.net" <discussions@...sword-hashing.net>
Subject: Re: [PHC] Password hashing as a self-overwriting Turing machine: new
 version (July 3)

On Fri, Jul 3, 2015 at 5:49 AM, Bill Cox <waywardgeek@...il.com> wrote:

> However, like Bcrypt, BusyBeaver in it's current form uses very little
> memory (it fits in L2 cache), and overwrites it many times.  This is good
> for GPU defense, and this makes a nice server-side algorithm.  However, for
> full-disk encryption or unwrapping user keys, I'd prefer to use _lots_ of
> memory - maybe a GiB.  In this case, the memory wont be overwritten many
> times like it is server-side.  Your algorithm would then become more
> sensitive to TMTO attacks, as it should.  There's no point lowering
> memory*time defense much due to paranoia about TMTO attacks.  Lyra2 suffers
> a bit in this regard, making it a bit slower than Yescrypt and Argon2d.  A
> reasonable defense is to introduce a cryptographic hash now and then into
> your computations, which force an attacker to move compute forward.  For
> example, you could compute hashState = SHA512(hashState || registerValue)
> every 10000 or so loop iterations, and do registerValue ^=
> lower64(hashState).
>
> However, this is really a minor concern, IMO, so feel free to leave it out
> ;-)
>
> Bill
>

If you do make the memory size a parameter, then you also need to address
cache-miss penalties.  The way to make an algorithm efficient at
large-memory hashing is to hash blocks of memory together, rather than
individual 64-bit values.  Of course, this adds additional complexity.
Maybe BusyBeaver should remain a simple low-memory server-side hash.

Bill

Content of type "text/html" skipped