Bug#115767: I see this too on x86-xen running etch

[Justin Pryzby <justinpryzby@users.sourceforge.net>]

On Thu, Mar 09, 2006 at 05:13:34PM +0000, Andy Smith wrote:
> Hi,
> 
> I get this error message and sshd dying intermittently since I
> upgraded one of my sarge xen domains to etch.  It always happens in
> the middle of a prolonged dictionary attack on my sshd.  My other
> sarge domains on the same hardware get the dictionary attacks and
> weather them fine though.
> 
> I don't understand how it can be running out of random bytes when
> /dev/urandom is there and appears to be working.
> 
> Last time this happened I ran sshd from the console like so:
> 
> /usr/sbin/sshd -eD -o 'LogLevel VERBOSE'
This doesn't actually help much; the same error code was reported
before.  Would you consider trying to strace the processes?  This was
recommended for the other similar bug (assigned to "openssl"; there
are #115767, #155467).

Something like strace -f -o /var/log/ssh-strace/ssh-strace.log, where
you should be able to set the directory permissions to be sufficiently
tight.

This might also be a kernel bug, if read() returns short when it
shouldn't.  How reproducible is this for you?  What if you

  while :; do ssh otherhost true; done;

(with rsa or other noninteractive authentication mechanism enabled)

Justin