Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 5 Jun 2015 15:39:20 +0300
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: poor man's fuzzer

On Fri, Jun 05, 2015 at 08:20:25PM +0800, Kai Zhao wrote:
> Do you get this at the begin ?
> 
> $ ./fuzz_solar.pl
> Argument "PRIO_PROCESS" isn't numeric in setpriority at ./fuzz_solar.pl
> line 15.

No.  It just works for me, and the priority is in fact changed.  BTW, I
also set "Idle = N" in john.conf, and let john inherit the nice +19 from
fuzz.pl.  I think it's better that way, when all of these processes are
at equal low priority and do not explicitly yield CPU to one another.

And I am currently running this on a --disable-openmp --disable-opencl
--disable-cuda build of john.  Fuzzing of OpenCL and CUDA formats should
also be done, separately.  I don't have a goal to fully do this myself -
rather my goal is to show to you that there's more work on this, and
that there's still plenty of low-hanging fruit (that you somehow missed?
or/and that got added after you were done with fuzzing?)

A shortcoming of fuzz.pl is that some of the child processes terminate
much sooner than others.  This appears to be mostly because the list of
test vectors is grouped by format, and some formats are slow to
self-test (oh, BTW I forgot --skip-self-tests) and slow to test even a
couple of candidate passwords.  Need to add --skip-self-tests to the
main invocations of john, and pass the output of "--list=format-tests
--format=cpu" through "| shuf |" to spread the slow test vectors across
child processes more evenly.

Alexander

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.