Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 31 Mar 2018 21:55:38 +0200
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: OpenCL on Windows

Hello Tim,

I was hoping someone else would reply, but since no one did yet let me
reply with whatever little info I have on this:

On Thu, Mar 29, 2018 at 08:30:52PM +0200, timajus . wrote:
> this is tim from italy. I've cloned the github repository in my cygwin64,
> i've made symbolic links to the opencl libs and includes respectively to
> /usr/local/lib and /usr/local/include, then i've issued the following
> configure command

The way you've built JtR looks right to me.

> I've issued the following command (from the src directory):
> 
> $ ../run/john --wordlist=../run/wl.txt --format=descrypt-opencl
> ../run/hashes.txt --verbosity=5 --devices=0
> 
> and this is the result. (My device is a NVIDIA GeForce GTX 1050 Ti with
> CUDA)
[...]
> Note: building per-salt kernels. This takes e.g. 2 hours for 4096 salts.
> Salt compiled from Binary:258
> Salt compiled from Binary:517
> Alarm clock
> 
> $  echo $?
> 142

This looks like an incompatibility of our code with Cygwin, and in
particular I suspect the uses of the alarm timer in bt.c, which is
probably invoked by a call to create_perfect_hash_table() from
opencl_DES_bs_plug.c.  You may try to comment out or delete the
"alarm(3);" line in bt.c (currently line 396).  Does this help?

You may also try cracking hashes of another type, just to make sure the
problem is specific to formats that make use of bt.c.  For example,
download the "3107 of Unix crypt(3) hashes (4 flavors, plus bigcrypt for
length 9+) and corresponding plaintext passwords" file from here:

http://openwall.info/wiki/john/sample-hashes#Sample-password-hash-files

"gzip -d" it and use the "--format=md5crypt-opencl" option on it.

If/once you do get descrypt-opencl working, you'll also want to use
"mask mode" (the "--mask" option, possibly along with another cracking
mode, referring to that other mode's partial candidate passwords with
"?w" in the mask), or the performance would be poor.  That's because
descrypt with few different salts is fast enough that candidate
passwords should be generated at least partially on the GPU, and we're
only able to do that with mask.

I hope this helps, and I hope a jumbo developer will look into the issue
with alarm.

Alexander

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ