Re: ssh, /dev/urandom

[tb@becket.net (Thomas Bushnell, BSG)]

"Alfred M. Szmidt" <ams@kemisten.nu> writes:

>    >    Telnet has worse security than even a buggy miserably fake ssh.
>    > 
>    > Telnet has _no_ security.  It doesn't have fake security, which you
>    > get by using crappy random bits and Open SSH.  That is a huge
>    > difference.  Open SSH was designed for security, telnet was _not_.
> 
>    What?  So you are saying that telnet is better than a fake ssh?  
> 
> Yes, in the sense that it does _NOT_ give the user a sense of fake
> security.

This is an excellent reason to document what we do carefully and
completely.  

> The kind of security that I do _not_ stand up for is the kind that
> gives the user a fake feeling.  Which is what you want todo with
> adding weirdo hacks.  The best suggestion has been to compile Open SSH
> with its own flags for gathering random bits on systems that do not
> support /dev/random or /dev/urandom.

No, not at all.  I don't want to give the user a fake feeling.  I want
the user to be able to make a judgement "in this case, the security is
not important, but telnet is a major hassle, so I choose the fake
ssh". 

> Are you even following this discussion?  I have not said a single word
> of the exlusion of ssh, not even muttered it, or implied it.  I am
> against including a unsecure random translator!!!

Geez, there are enough proposals on the table already.  urandom isn't
guaranteed anything anyway, really, but I agree that we should do the
best we can, which might mean something nicely pseudo-random based on
something like the clock or the process table.  I'm not in favor of
just linking it to bash.

But I do *not* agree that linking it to bash is bad on the grounds
that we should never ever do such a thing, but only because we can do
better with little extra work.