openssh unix dev - Mar 2001 - arc4randomstir() in OpenSSH

I'm trying to understand the rational behind the arc4random() and
arc4random_stir() functions in the OpenSSH source tree.

On a system that has a good random number generator, say an in
kernel /dev/random what extra functionality is this stuff providing ?

Would it be acceptable to replace the calls to arc4random() with
reading from /dev/random and drop the arc4random_stir() all together ?

Note I'm not suggesting doing this to OpenSSH before anyone starts
a debate on the portability of the code and ties this into the
PRNGd/yada yada thread.  This is purely an exercise in understanding
why RC4 is used in getting random data.

Thanks.

--
Darren J Moffat

On Wed, 28 Mar 2001, Darren J Moffat wrote:
> I'm trying to understand the rational behind the arc4random() and
> arc4random_stir() functions in the OpenSSH source tree.
We are following the OpenBSD tree with as few changes as possible -
this is the approach used by them.
> On a system that has a good random number generator, say an in
> kernel /dev/random what extra functionality is this stuff providing ?
It minimises reads from /dev/random which usually contains a finite
amount of entropy. Most of the random numbers used in OpenSSH need
don't need the "information theoretic" strength of /dev/random.
> Would it be acceptable to replace the calls to arc4random() with
> reading from /dev/random and drop the arc4random_stir() all together ?
If you want to waste randomness, yes :)

Many of the calls to arc4random are fairly high volume users, e.g.
paddding which is done per-packet. These would deplete /dev/random
pretty quickly. If you used /dev/urandom (the 'endless' interface), then
you are back to doing something like arc4random only in the kernel.

-d

-- 
| Damien Miller <djm at mindrot.org> \ ``E-mail attachments are the poor
man's
| http://www.mindrot.org          /   distributed filesystem'' - Dan
Geer

In message <3AC24650.6BE258C at Eng.Sun.COM>, Darren J Moffat
writes:
>Would it be acceptable to replace the calls to arc4random() with
>reading from /dev/random and drop the arc4random_stir() all together ?
arc4random() does not block.  It is initalized with data from
/dev/arandom.  arc4randomstir() reintializes arc4random() with new
random data from the kernel.

If you replace all calls to arc4random() with a read from /dev/random,
you might cause OpenSSH to block for a long time.

Niels.