[Lxc-users] (no subject)

[Daniel Lezcano]

On 09/14/2010 09:51 PM, lxc at jelmail.com wrote:

[ cut ]

> #
> # Terminals
> lxc.tty =1                           #                allow one tty
>    

Add this line in the configuration file:

lxc.pts = 1   # make a new pts instance
> lxc.cgroup.devices.allow = c 4:0 rwm # /dev/tty0      current virtual
> terminal
>    

[ cut ]

> # cat /etc/lxc/mycontainer.fstab
>    
And remove this one.
> none /srv/lxc/mycontainer/dev/pts devpts defaults 0 0
>    

Make sure you don't have, in your container /lib/init/fstab, a devpts 
mount point specified.

That should fix your problem.

A bit of explanation follows ;)

The devpts fs is a bit weird because you have to specify the 
"new_instance" option to create a ... new instance. If not specified, 
you will get the host devpts mounted in the container, it will be 
shared. Of course, most of the distros are not aware of this option, 
hence by default the unmodified containers distro will mount the host's 
devpts :(

The lxc.pts = 1 does this "new instance" mount + a few things to make 
available a fresh /dev/pts for the container.

The last trace you gave ...

":: Unmounting Filesystems
[BUSY]
umount: devpts busy - remounted read-only"


... make me thing it's the host's devpts which is mounted. So at the 
container shutdown time, devpts is busy because your host is using it 
(and probably some other containers are using it too).

However, I am curious to understand why a remount as read-only is 
propagated in all the system as we are running in our own mount 
namespace. I will ask to the kernel mailing list ...

Let us know if the problem is solved.

Thanks for the traces.
   -- Daniel