[BUG] remount ro on loopback mount leaves unmountable filesystem behind

[BUG] remount ro on loopback mount leaves unmountable filesystem behind

[David Chinner]

The following steps will leave an unmountable filesystem
behind. This test "works" using ext3 or XFS. XFS version is
as follows replacing steps with ext3 specific commands results
in the same unmountable filesystem problem:

$ mount -t xfs /dev/sdb6 /mnt/scratch
$ mkdir /mnt/scratch/mnt
$ touch /mnt/scratch/img
$ mkfs.xfs -f -d file,name=/mnt/scratch/img,size=1g
$ mount -t xfs -o loop /mnt/scratch/img /mnt/scratch/mnt

All normal up to here. I can unmount everything just fine.

$ umount /mnt/scratch/mnt
$ umount /mnt/scratch
$

Same thing, but remounting the loop filesystem readonly
before unmounting:

$ mount -t xfs /dev/sdb6 /mnt/scratch
$ mkdir /mnt/scratch/mnt
$ touch /mnt/scratch/img
$ mkfs.xfs -f -d file,name=/mnt/scratch/img,size=1g
$ mount -t xfs -o loop /mnt/scratch/img /mnt/scratch/mnt
$ mount -t xfs -o remount,ro /mnt/scratch/img /mnt/scratch/mnt
$ umount /mnt/scratch/mnt
$ umount /mnt/scratch
umount: /mnt/scratch: device is busy
umount: /mnt/scratch: device is busy
$

At this point /mnt/scratch is unmountable. lsof shows no open
references to the filesytem, so it would seem to be a reference
count leak somewhere....

Reproduced on 2.6.25-rc3 on ia64 and x86_64, and UML on
2.6.26-pre-rc1. Found when changing XFSQA test 073 to host
loopback images and mounts mounts on something other than /tmp.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: [BUG] remount ro on loopback mount leaves unmountable filesystem behind

[Christoph Hellwig]

On Tue, May 13, 2008 at 06:50:47PM +1000, David Chinner wrote:
> Same thing, but remounting the loop filesystem readonly
> before unmounting:
> 
> $ mount -t xfs /dev/sdb6 /mnt/scratch
> $ mkdir /mnt/scratch/mnt
> $ touch /mnt/scratch/img
> $ mkfs.xfs -f -d file,name=/mnt/scratch/img,size=1g
> $ mount -t xfs -o loop /mnt/scratch/img /mnt/scratch/mnt
> $ mount -t xfs -o remount,ro /mnt/scratch/img /mnt/scratch/mnt
> $ umount /mnt/scratch/mnt
> $ umount /mnt/scratch
> umount: /mnt/scratch: device is busy
> umount: /mnt/scratch: device is busy

This is a problem in mount, no the kernel.  Before the remount the
/etc mtab looks something like this:

/qemu/test.img /mnt xfs rw,loop=/dev/loop0 0 0

and after it looks something like this:

/qemu/test.img /mnt xfs ro 0 0

As a workaround do a losetup -d /dev/loop0 after unmounting the
filesystem.

--
To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [BUG] remount ro on loopback mount leaves unmountable filesystem behind

[David Chinner]

On Tue, May 13, 2008 at 06:23:40AM -0400, Christoph Hellwig wrote:
> On Tue, May 13, 2008 at 06:50:47PM +1000, David Chinner wrote:
> > Same thing, but remounting the loop filesystem readonly
> > before unmounting:
> > 
> > $ mount -t xfs /dev/sdb6 /mnt/scratch
> > $ mkdir /mnt/scratch/mnt
> > $ touch /mnt/scratch/img
> > $ mkfs.xfs -f -d file,name=/mnt/scratch/img,size=1g
> > $ mount -t xfs -o loop /mnt/scratch/img /mnt/scratch/mnt
> > $ mount -t xfs -o remount,ro /mnt/scratch/img /mnt/scratch/mnt
> > $ umount /mnt/scratch/mnt
> > $ umount /mnt/scratch
> > umount: /mnt/scratch: device is busy
> > umount: /mnt/scratch: device is busy
> 
> This is a problem in mount, no the kernel.  Before the remount the
> /etc mtab looks something like this:
> 
> /qemu/test.img /mnt xfs rw,loop=/dev/loop0 0 0
> 
> and after it looks something like this:
> 
> /qemu/test.img /mnt xfs ro 0 0
> 
> As a workaround do a losetup -d /dev/loop0 after unmounting the
> filesystem.

Hmmm - I even considered that and tried a '-o remount,ro,loop'
but that obviously doesn't work with dynamic loop device instantiation,
either.

/me didn't read all the way to the bottom of the extremely verbose
mount man page so didn't find the bit about losetup -d....

Anyway, I'll hard code loop device numbers into the script so
I can just forget about this problem. Thanks, Christoph.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: [BUG] remount ro on loopback mount leaves unmountable filesystem behind

[Christoph Hellwig]

On Tue, May 13, 2008 at 10:07:10PM +1000, David Chinner wrote:
> Anyway, I'll hard code loop device numbers into the script so
> I can just forget about this problem. Thanks, Christoph.

hardcoding them is a little dangerous because a system might aswell
have them in use already, e.g. for CD images.  Just use losetup -f to
find the first free one and keep it in a shell variable.
--
To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [BUG] remount ro on loopback mount leaves unmountable filesystem behind

[H. Peter Anvin]

Christoph Hellwig wrote:
> On Tue, May 13, 2008 at 10:07:10PM +1000, David Chinner wrote:
>> Anyway, I'll hard code loop device numbers into the script so
>> I can just forget about this problem. Thanks, Christoph.
> 
> hardcoding them is a little dangerous because a system might aswell
> have them in use already, e.g. for CD images.  Just use losetup -f to
> find the first free one and keep it in a shell variable.

losetup -f is racy, too.  losetup -sf is safe, however.

	-hpa
--
To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [BUG] remount ro on loopback mount leaves unmountable filesystem behind

[Karel Zak]

On Tue, May 13, 2008 at 06:23:40AM -0400, Christoph Hellwig wrote:
> On Tue, May 13, 2008 at 06:50:47PM +1000, David Chinner wrote:
> > Same thing, but remounting the loop filesystem readonly
> > before unmounting:
> > 
> > $ mount -t xfs /dev/sdb6 /mnt/scratch
> > $ mkdir /mnt/scratch/mnt
> > $ touch /mnt/scratch/img
> > $ mkfs.xfs -f -d file,name=/mnt/scratch/img,size=1g
> > $ mount -t xfs -o loop /mnt/scratch/img /mnt/scratch/mnt
> > $ mount -t xfs -o remount,ro /mnt/scratch/img /mnt/scratch/mnt
> > $ umount /mnt/scratch/mnt
> > $ umount /mnt/scratch
> > umount: /mnt/scratch: device is busy
> > umount: /mnt/scratch: device is busy
> 
> This is a problem in mount, no the kernel.  Before the remount the
> /etc mtab looks something like this:
> 
> /qemu/test.img /mnt xfs rw,loop=/dev/loop0 0 0
> 
> and after it looks something like this:
> 
> /qemu/test.img /mnt xfs ro 0 0

 Yes, that's mount(8) bug. It doesn't read old options from /etc/mtab
 when you remount by "mount -o remount <src> <dest>".

> As a workaround do a losetup -d /dev/loop0 after unmounting the
> filesystem.

 ... or use "mount -o remount <dest>", for example:

   mount -t xfs -o remount,ro /mnt/scratch/mnt


    Karel

-- 
 Karel Zak  <kzak@redhat.com>