Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sun, 25 Mar 2012 05:14:22 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: Self test woes

On Thu, Mar 22, 2012 at 09:30:19PM +0100, magnum wrote:
> I think we have to to adjust/add/do something with the self-tests for
> certain formats.

Yes, self-tests need to be improved in a variety of ways.

> My OpenCL RAR format runs for very long per crypt_all() call: One call
> can take half a minute or much more (though I'm hoping to do better).
> The upshot is that it may contain thousands of candidates, or hundreds
> of thousands, or more.
> 
> But the self-tests just tries one hash, crypts, tries another one, calls
> crypt, etc. which make the self-test take forever (since one hash takes
> just as long to do as half a million of them).

Yes.  We need to make the self-tests more similar to actual cracking.
On the other hand, testing more than one call to crypt_all() makes
sense.  What if we have a bug where only the first call works right?
This is not too unlikely (think uninitialized variables).  So two calls
to crypt_all() is the minimum, and these should be fast enough that the
wait is not annoying.

For testing of cmp_all(), we could want to reset the previously set keys
before the second call to crypt_all().

> I had to comment out the
> fmt_self_test(format) call from benchmark_format() (but left the
> benchmark in place) just to be able to work on my format.

Ouch.

> I do not have any particular proposition though. Any ideas? This format
> is extreme but I think the problem is noticable with most other GPU
> formats too. Just disabling self-test with a flag would be easy but
> that's not a good solution - we need the tests. Maybe we could have a
> flag that tells john to collate all tests into just one or two calls to
> crypt_all()?

Yes, and it doesn't have to be a flag - we can make it standard.

Alexander

Powered by blists - more mailing lists

Your e-mail address:

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