On 2006.02.18 00:38:02 +0100, Herbert Poetzl wrote:
> On Fri, Feb 17, 2006 at 01:50:41PM +0100, Oliver Welter wrote:
> > Hi Folks, Hi Bertl,
> >
> > the problem which seemed to be solved re-appeard today with recent
> > kernel patches (2.6.15-gentoo-r4-vs2.1.1-rc6)
> >
> > To reconstruct the problem
> > 1) add a drbd blockdevice to the guests fstab
> > 2) start and stop guest
> > 3) try to unmount the device
> > you get an error, telling that the drbd device is still mounted...
> >
> > So - any ideas ??
>
> okay, just as update, we tracked this down to
> the fact that even mainline 2.6.15/2.6.16-rc3
> does not release ext3 filesystems properly
> when the namespace is destroyed ... this can
> be easily verified with a mainline kernel and
> the following command sequence:
>
> vnamespace --new -- mount /dev/hda1 /mnt
>
> (assuming that /dev/hda1 contains an ext3 fs
> and /mnt exists, this will claim hda1 but not
> release it -- it works fine with ext2 though)
>
> now investigating ...
Bertl tracked the issue down to be caused by the kernel thread created
for the mount (kjournald), which is running in the new namespace. Once
all user processes in that namespace are gone, you loose access to it,
but the kjournald keeps it alive, thus the mount stays around which in
turn keeps the kjournald alive, oops :-)
A patch to fix this is in -mm now, the according threads on lkml are:
Message-Id: <20060218013547.GA32706@MAIL.13thfloor.at>
http://lkml.org/lkml/2006/2/17/323
Message-Id: <1140425218.2979.14.camel@laptopd505.fenrus.org>
http://lkml.org/lkml/2006/2/20/33
Björn
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Mon Feb 20 09:20:37 2006