Subject: Re: "LEXR" USB storage device
To: Alan Barrett <apb@cequrux.com>
From: Daniel Carosone <dan@geek.com.au>
List: current-users
Date: 05/24/2003 08:25:56
On Sat, May 17, 2003 at 10:24:31PM +0200, Alan Barrett wrote:
> > What is being used as the entropy source?  The data read or the 
> > rotational/seek delays?
> 
> Timing, and the block addresses used.  See the call to rnd_add_uint32 in
> sd.c.
> 
> The "rnd: ... attached ... (collecting)" message is printed only
> with options RND_VERBOSE, and indicates that it's both "collecting"
> and "estimating" entropy.  I'd prefer it to say "(collecting without
> estimation)", because I don't have enough trust in the randomness of
> timing variations in solid state storage devices.

I've run across this one too; I'm not sure how/whether sd.c should
decide at attach time what its parents are and what rnd flags to
set.

On an i386 machine with cycle counters for timing, there's going
to be some variation anyway - and the estimation is capped at "1
bit per sample".

If you want to turn off estimation because it bothers you, you
could do it manually with rndctl - but note that entropy estimation
in the rnd driver is basically bogus.

--
Dan.

PS: smb; your mailer seems to have reverted to thread-deprived mode again.