itojun@iijlab.net wrote:

> 
> >>Is it possilbe that rsa support only works if "pseudo-device   rnd" is
> >>compiled in the kernel?
> >>
> >>At least I have it in my kernel on i386 and sparc where sshd works and don't
> >>have it in the kernel for amiga and hp300 (its not activated in GENERIC there
> >>either...). I will recompile a kernel with rnd and see if that fixes the
> >>sshd daemon, but its already late and it will take some time on m68k, so
> >>don't expect feedback that soon :)
> >	i'll try using a kernel without RND device and see if your issue
> >	repeats.  please hold.
> 
> 	hmm I see.
> 	- ssh/rsa.c:rsa_alive() calls RSA_generate_key()
> 	- RSA_generate_key() calls BN_generate_prime()
> 	- BN_generate_prime() calls BN_rand() via probable_prime()
> 	- BN_rand require us to have kernel random number device
> 	so openssh considers that we don't have RSA in the library.
> 
> 	similarly, DH_generate_key() does not work if there's no random number
> 	device.
> 
> 	I'm not sure if it is safe to allow userland-generated pseudo random
> 	number.  it looks intentional.  all I can do with confidence is to
> 	improve error message from ssh/sshd, to warn that you don't have random
> 	number device.

Don't we currently have rnd on by default in all kernels?  If so, then I
think that this is perhaps the best solution.

Simon.
--
Simon Burge                            <simonb@wasabisystems.com>
NetBSD Sales, Support and Service:  http://www.wasabisystems.com/