Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 15 Sep 2012 06:36:58 +0400
From: Solar Designer <>
Cc:, Aggelos Kiayias <>,
	Vladimir Vorontsov <>,
	gifts <>
Subject: Re: Randomness Attacks Against PHP Applications

On Wed, Aug 22, 2012 at 02:31:07PM +0400, Solar Designer wrote:
> On Thu, Aug 09, 2012 at 11:19:14AM -0700, Yves-Alexis Perez wrote:
> > Paper authors tried to port this to PHP security team, but it seems the
> > answer was that it was an application problem.
> Here's a vulnerability in and attack on session IDs of PHP proper:
> This is not exactly the same topic (PHP apps vs. PHP itself), yet it's
> closely related and the timing of it was provoked by the same research.

FWIW, here's a PHP mt_rand() seed cracker that I wrote:

It finds possible seeds given the very first mt_rand() output after
being seeded with mt_srand().

Here's a sample run.  First, generate a sample "random" number (using
PHP 5.3.x in this case):

$ php5 -r 'mt_srand(1234567890); echo mt_rand(), "\n";'

Now build and run the cracker:

$ make
gcc -Wall -O2 -fomit-frame-pointer -fopenmp php_mt_seed.c -o php_mt_seed
$ time ./php_mt_seed 1328851649
Found 0, trying 654311424 - 671088639, speed 13631488 seeds per second
seed = 658126103
Found 1, trying 1224736768 - 1241513983, speed 13585543 seeds per second
seed = 1234567890
Found 2, trying 4278190080 - 4294967295, speed 13617003 seeds per second
Found 2

real    5m15.397s
user    41m58.185s
sys     0m0.044s

In 5 minutes of real time (on an FX-8120 CPU), it found the original
seed, another seed that also produces the same mt_rand() output, and it
searched the rest of the 32-bit seed space (not finding other matches).

Note that this is a lot slower than crackers for LCG PRNG seeds, which
were crackable in way under 1 second even in 1990s (IIRC, some IDS
products did that for potential Back Orifice backdoor traffic, to detect
it regardless of password used).  There's a 397 iterations loop per seed
tested here.  Of course, a rainbow table would be quick.

Here's an OpenCL implementation by Gifts:

According to Gifts, this one tests 190 million seeds per second on a GTX
560 Ti, for a total running time of 22 seconds.

Maybe these PoCs will help convince someone.


Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

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