Re: [PATCH] (0/4) Entropy accounting fixes

[Oliver Xymoron]

On Sat, Aug 17, 2002 at 11:51:52PM -0400, Robert Love wrote:
> On Sat, 2002-08-17 at 23:05, Linus Torvalds wrote:
> 
> > This is particularly true on things like embedded routers, where the 
> > machine usually doesn't actually _run_ much user-level software, but is 
> > just shuffling packets back and forth. Your logic seems to make it not add 
> > any entropy from those packets, which can be _deadly_ if then the router 
> > is also used for occasionally generating some random numbers for other 
> > things.
> 
> Agreed.  Further, embedded routers - since they are headless/diskless -
> have problems even with the _current_ /dev/random code.  They simply do
> not generate enough entropy to fulfill sshd requests [1].

This analysis actually stemmed from my work to port OpenSSH to a headless
(non-UNIX, non-POSIX, non-protected-memory, diskless) network appliance. SSH
only needs real entropy for the keys generated by ssh-keygen. It's
complete overkill for session keys.

And guess what? Stock Portable OpenSSH (v3.4p1) uses /dev/urandom
(configure.ac):

# Check for user-specified random device, otherwise check /dev/urandom
AC_ARG_WITH(random,
        [  --with-random=FILE      read entropy from FILE
	(default=/dev/urandom)],
 
> Saying "use /dev/urandom" in this case means we may as well not have a
> /dev/random.  There is a difference between incorrect accounting (which
> it seems you have identified) and just too strict gathering behavior.
> 
> 	Robert Love
> 
> [1] this is why I wrote my netdev-random patches.  some machines just
>     have to take the entropy from the network card... there is nothing
>     else.

This patch is perfectly compatible with your netdev-random patches, in
fact I encourage it's resubmission after this one gets
in. /dev/urandom users will get all the benefits of network sampling
without /dev/random suffering at all.

-- 
 "Love the dolphins," she advised him. "Write by W.A.S.T.E.."