Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <1298408152.13953.35.camel@cthulhu.linuxasylum.net>
Date: Tue, 22 Feb 2011 21:55:52 +0100
From: Samuele Giovanni Tonon <samu@...uxasylum.net>
To: john-users <john-users@...ts.openwall.com>
Subject: rawsha1 e NSLDAPS jumbo 12 opencl patch

hello,
attached here a patch for jumbo12 to add rawsha1 and NSLDAPS formats
cracking with opencl.
With a salted hash opencl was able to go 2x time faster than original:

OpenCL dictionary attack with rules

#./john -rules -w=mydict ~/john/john-1.7.6/run/pippo 
OpenCL Platform: <<<ATI Stream>>> and device: <<<Juniper>>>
Loaded 10085 password hashes with 10085 different salts (Netscape LDAP
SSHA OPENCL [salted SHA-1])
guesses: 0  time: 0:00:09:57 1.34% (ETA: Wed Feb 23 09:25:30 2011)  c/s:
10350K  trying:  - unreluctantly00

normal dictionary attack with rules

#./john -rules --format=ssha -w=mydict ~/john/john-1.7.6/run/pippo 
Loaded 10085 password hashes with 10085 different salts (Netscape LDAP
SSHA [salted SHA-1])
guesses: 0  time: 0:00:09:58 0.29% (ETA: Fri Feb 25 06:32:58 2011)  c/s:
3943K  trying: gigantocyte

incremental mode

#./john -i:all ~/john/john-1.7.6/run/pippo 
OpenCL Platform: <<<ATI Stream>>> and device: <<<Juniper>>>
Loaded 10085 password hashes with 10085 different salts (Netscape LDAP
SSHA OPENCL [salted SHA-1])
guesses: 0  time: 0:00:09:59  c/s: 11462K  trying: 1952 - permath

#./john -i:all --format=ssha ~/john/john-1.7.6/run/pippo 
Loaded 10085 password hashes with 10085 different salts (Netscape LDAP
SSHA [salted SHA-1])
guesses: 0  time: 0:00:09:58  c/s: 4012K  trying: m1saji



so far i'd call this a good achievement because i'm pretty sure my code
can be optimized, if anyone is interested on taking a look to the code
he's more then welcomed.

now the bad part

problems: i'm still investigating the issue but single mode quits while
trying to eat all my RAM, lowering ssha_num_keys solves the issue but
the whole cracking process become damn slow; i suspect this is because
of the different approach in single cracking mode.
quick and dirt solution: use "non opencl" for single and revert to
opencl for dictionary+rules and incremental attack.

due to john behavior for salted password (get xx keys and 
for those xx keys try each salt one at time) i tried a different
approach by putting the password+salt cycle in to the kernel and
enqueing each time only the newer salt.
This, strangely instead lead to having really slow speed and i will
investigate a bit further: if i get rid of the 1024*1024*24 bytes per
crypt and instead i transfer only 8 bytes i should get rid of pci
bottle-ness.

I must say, so far it has been a lot of fun and new things to learn,
and now with salted passwords it has been rewarding.

to Dhiru Kholia: i've sent you a private email to integrate my code with
your but i didn't get any response, this patch doesn't have any of the
md* opencl code due to previous patch failing on jumbo 12 at NT_fmt.c
feel free to merge it with your changes . 

Samuele 


-- 
While various networks have become deeply rooted, and thoughts have been
sent out as light and electrons in a singular direction, this era has
yet to digitize/computerize to the degree necessary for individuals to
become a singular complex entity.
  KOUKAKU KIDOUTAI Stand Alone Complex


Download attachment "john-1.7.6-jumbo-12-opencl-11.diff.bz2" of type "application/x-bzip" (19589 bytes)

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.