phc-discussions - A possible bandwidth limitation attack

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>] [day] [month] [year] [list]

Date: Sat, 1 Mar 2014 11:00:25 -0500
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: A possible bandwidth limitation attack

If I were a government trying to crack Script or one of it's ancestors
that has a fast and efficient hardware implementation for the hash
function (such as pretty much all the ARX and SHA2 families), I'd
force a large DRAM maker to integrate the hashing core directly on the
chip for me, dramatically reducing the I/O bandwidth bottleneck.

DRAM chips support 16 gigabits per die, if I'm not mistaken.  That's
2GB per chip.  Any memory-hard KDF hashing less than 2GB is
susceptible to this bandwidth limitation attack.  We can expect this
to grow in the future, like by around 4X to 8GB per chip.  Since most
users will likely use smaller amounts of memory, I'd just replicate
the hashing core so that I'd have high bandwidth down as small as say
64MiB (or whatever the basic repeated DRAM block size is on the die),
which would allow parallel attacks on the die against these smaller
hashes.

I think this could only be done by large governments.  No one else
could force Samsung or Micron to let them muck with their cutting edge
products.  Doing a similar attack with lower integration chips is
possible through exotic 3D interconnect, but I'm not sure it would be
cost effective.

I think this is a another argument for compute-time hardening.  Feel
free to rely on multiplication chains in your hash function the way I
do.  I do one multiply followed by an XOR in a long chain to force an
attacker to take his time computing the hash, tying up the memory for
that long.  My code is at:

https://github.com/waywardgeek/tigerkdf

Bill