[systemd-devel] prevent unmount of single partitions on shutdown

[Zbigniew Jędrzejewski-Szmek]

On Wed, Aug 21, 2013 at 06:03:34PM +0800, Tom Gundersen wrote:
> On Wed, Aug 21, 2013 at 5:56 PM, Colin Guthrie <gmane at colin.guthr.ie> wrote:
> > 'Twas brillig, and Zbigniew Jędrzejewski-Szmek at 21/08/13 04:38 did
> > gyre and gimble:
> >> On Wed, Aug 21, 2013 at 05:21:59AM +0200, Stephan Raue wrote:
> >>> Hi,
> >>>
> >>> i try to port systemd on a own embedded OS which is stored in a
> >>> squashfs file. This file is on a fat partition (later mounted as
> >>> /flash) on the drive.
> >>>
> >>> In our own initramfs (which dont uses systemd) /flash will be
> >>> mounted and then the Squashfs file as /sysroot. later we do a
> >>> switch_root and start systemd. On shutdown systemd trys now (5
> >>> times) to unmount /flash and times out later with a error message.
> >>> This delays the shutdown/reboot much. Systemd also trys to cleanup
> >>> /dev/loop0. I need to prevent systemd to unmount /flash and clean
> >>> /dev/loop0 (which is the / mount from the squashfs file).
> >>>
> >>> can i actually prevent this in some way and if nout could i request
> >>> a feature to add a mount option which if avaible prevents systemd
> >>> from unmounting single partitions and cleanup /dev/loopX if its
> >>> still mounted as / ? I need this as mount option or a systemd unit
> >>> file but for fstab based systems it would be usefull as a fstab
> >>> option too (we dont use fstab)
> >> I don't think it's possible currently with fstab. But with
> >> a mount unit, I think
> >>   DefaultDependencies=no
> >>   RequiredBy=-.mount
> >> should work. Have you tried something like that?
> >
> > Hmm, I thought the umount logic was such that it just tries to unmount
> > everything from/proc/mounts rather than looking at units etc.
OTOH, if systemd manages to unmount something on top of which another 
filesystem is mounted, is a kernel bug, no?

> > I asked a similar question a while back and we figured at the time that
> > adding a mount option might be a solution as there is already a fstab
> > option to indicate a given filesystem should be mounted in the initrd,
> > and it would make sense to honor that during shutdown (by not trying to
> > umount it), but the problem was the umount loop doesn't read fstab or
> > any mount options to check...
> 
> I have also been thinking of this problem recently. I have some
> patches to not add a conflict with umount.target when a mount is
> marked with x-initrd.mount, but as you point out we need to do
> something to also cover the final umount loop. One option is to simply
> jump to the shutdown ramfs straight away if it exists and let that do
> all the final unmounting/killing, which should always work.