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: Tue, 21 Jan 2014 11:36:35 -0500
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] Modified pseudo-random distribution in NoelKDF

On Mon, Jan 20, 2014 at 9:19 PM, Solar Designer <solar@...nwall.com> wrote:
> How do you implement thread-level parallelism, or is this scheme with
> the second loop at 1% only suitable for p=1?

With N threads, memory is broken into N equal parts, and each thread
gets it's own.  Threads are merged before the final loop, and data is
read based on the password across all memory.  I suppose I could have
N threads reading randomly from all memory at the end, and then
combine their hashes, however it becomes quickly bandwidth limited.

My "cheat killer" pass was a good idea when I thought "smoke" could be
used to hide the memory access pattern, but smoke didn't provide
enough protection from code that can see the individual cache lines I
read from.  Now I'm thinking that the final pass may need to be
independent of the password to avoid a timing attack.

>> The average recalculation for 10,000,000 nodes covered by evenly
>> spaced pebbles [...]
>
> What if they are not evenly spaced?  You're making the distribution of
> lookup indices highly non-uniform, so perhaps an attacker with limited
> memory can adjust the spacing accordingly and achieve a lower TMTO
> penalty?

I've tried many combinations, including random placement, weighted
random (more pebbles based on incoming edge density), weighted even
spacing, covering nodes pointed to by edges below a certain length,
and covering nodes with multiple edges pointing at them.  I've also
tried combinations of evenly spaced with these other schemes.  Nothing
seems to improve over evenly spaced so far.

This makes some sense.  The difference between 1 pebble every 4 and 1
every 8 is enormous.  If I use pebbles for anything other than
minimizing the average spacing between pebbles, it has to offer a
massive benefit, because the recalculation sub-DAG size seems to be
exponentially related to the spacing.  Even going from every 4 to
every 5 is a huge difference.  When an attacker has no knowledge of
what the next address will be, I seriously doubt it is possible for
him to achieve the same penalty as evenly spaced every 4 while only
using only 1/8th the number of pebbles as memory size.

Bill

Powered by blists - more mailing lists