BUG in tty_open when using containers and ptrace

Grzegorz Nosek root at localdomain.pl
Sun Jul 12 00:49:33 PDT 2009


On sob, lip 11, 2009 at 11:19:35 +0100, Alan Cox wrote:
> > Talking to myself (again), kmemleak doesn't report anything (apart from
> > a few unrelated early-init warnings) but running the startup/shutdown loop
> > on tty1 causes totally crazy behaviour on every VT I used. By "totally
> > crazy" I mean not responding properly to arrow keys, backspace etc., but
> 
> You are only seeing this with containers ?

Well, that's how I could reproduce it, didn't try any other way.
Multiple devpts instances don't matter here, but quite possibly
namespaces do.

If I let the loop run unattended, after it finishes the VTs are fine,
even the tty1 where the loop ran. However, if I try to use them with the
loop on another VT, they break and stay broken afterwards. Looks like
the container init (which has /dev/pts/0 mounted over /dev/console)
somehow interacts with the current VT. This may be a container isolation
problem rather than a TTY layer bug, though.

Didn't try (yet) running the loop while logged in via ssh or in a mc
subshell (so via a pty). I wonder if that changes anything.

BTW, kmemleak *did* report something after all but it was
networking-related (got the traces).

BTW2, I could not reproduce the fd count error while leaving the VM
unattended (and not letting its tty ops interleave with my tty ops),
so could you please have a look and say if the patch stands on its own?

Best regards,
 Grzegorz Nosek


More information about the Containers mailing list