Batchprocessing to stdout…

Starting with rev50 it’s now possible to stream results directly to stdout while batchprocessing. For example this makes it possible to stream results directly into cowpatty:

./pyrit_cli.py -f – -e linksys batch | ~/cowpatty-4.3/cowpatty -d – -r ~/cowpatty-4.3/wpa2psk-linksys.dump -s linksys
cowpatty 4.3 – WPA-PSK dictionary attack. <jwright@hasborg.com>
Collected all necessary data to mount crack against WPA2/PSK passphrase.
Starting dictionary attack.  Please be patient.
Using STDIN for hashfile contents.

…

The PSK is “dictionary”.

3088258 passphrases tested in 356.50 seconds:  8662.76 passphrases/second

Here is another example, this time the box calculating the PMKs is a headless machine a few thousand miles away. It’s piping the GPGPU-calculated results through gzip and ssh back to the local cowpatty client. The lower speed is caused by low upload-bandwidth…

ssh ebfe@anotherhost.com “cd pyrit_svn; ./pyrit_cli.py -f – -e linksys batch | gzip” | gunzip | ./cowpatty -d – -r wpa2psk-linksys.dump -s linksys
…

The PSK is “dictionary”.

1831271 passphrases tested in 856.48 seconds:  2138.14 passphrases/second

New performance data

Some new performance data came in today. The GeForce GTX 280 takes the lead with more than 11.000 PMKs per second, about 100x times faster than what we’ve seen from tools about a year ago.

See the new blinkenlights on the googlecode.com page.

64-bit machines fixed…

The hanging/crashing in the Nvida-CUDA core should now be fixed, compilation on 64-bit machines was fixed. Performance benchmarks updated since the old values did not reflect the new code.

Padlock again

Updated code for VIA’s padlock engine. Turns out the words “must be” in the documentation have a flexible meaning. Performance increase is now about 5x compared to native x86 code on same cpu.

Support for VIA Padlock

Added support for VIA’s hardware crypto engine ‘padlock’. Compile with

setup.py install HAVE_PADLOCK’

to enable. Speedup is about 2.5x for me (which is slow – VIA’s xsha1 opcode implementation sucks).

Added some code to export resultsets to aircrack-ng’s airolib-database format:

./pyrit_cli.py -e some_essid -f hashdb.db export_hashdb

Your help required.

There are reports that the CUDA-core hangs on some configurations. I’d really like to hear from you if you experience that as I can’t reproduce this problem here.