Re: [patch] random: make get_random_int() more random

From: Ingo Molnar <mingo@elte.hu>
To: Matt Mackall <mpm@selenic.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	Arjan van de Ven <arjan@infradead.org>, Jake Edge <jake@lwn.net>,
	security@kernel.org,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	James Morris <jmorris@namei.org>,
	linux-security-module@vger.kernel.org,
	Eric Paris <eparis@redhat.com>,
	Alan Cox <alan@lxorguk.ukuu.org.uk>,
	Roland McGrath <roland@redhat.com>,
	mingo@redhat.com, Andrew Morton <akpm@linux-foundation.org>,
	Greg KH <greg@kroah.com>, Dave Jones <davej@redhat.com>
Subject: Re: [patch] random: make get_random_int() more random
Date: Wed, 6 May 2009 22:51:45 +0200	[thread overview]
Message-ID: <20090506205145.GA1908@elte.hu> (raw)
In-Reply-To: <20090506204156.GB31071@waste.org>

* Matt Mackall <mpm@selenic.com> wrote:

> On Wed, May 06, 2009 at 10:09:54PM +0200, Ingo Molnar wrote:
> > I then ran the FIPS randomness test over the first 20,000 bits [2.5K 
> > data], which it passed:
> 
> That's proves nothing except that you have no idea what you're 
> talking about. People regularly break things that FIPS gives 
> flying colors. FIPS is nothing but a statistical sanity-check.

The current code didnt even pass the FIPS test, due to:

           <...>-1739  [000]   112.487579: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487583: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487584: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487585: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487586: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487588: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487589: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487590: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487592: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487593: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487594: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487595: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487597: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487598: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487599: sys_prctl: get_random_int(): d1f8a190
           <...>-1739  [000]   112.487601: sys_prctl: get_random_int(): d1f8a190

Linus's patch is a marked improvement, and it is really what we need 
here mostly.

We cannot afford true physical randomness (it's too expensive to get 
and not all hw has it), and even a 'good' PRNG is pretty expensive. 
Performance is the main reason why the networking stack has its own 
fast hack.

> Do you need this to be publicly broken again by someone who 
> actually knows something about cryptanalysis before you'll accept 
> that it's a bad idea? If so, then please move the code out of 
> random.c, so that I don't have to share in your embarassment a 
> second time.

I see your flame, but what's your technical point?

	Ingo