On Mon, Jan 25, 2016 at 04:59:59PM +0000, Jonathan Dowland wrote: > I *think* enabling lvmetad has indeed fixed the requirement to "vgchange -a y > <VG name>" upon unlocking the LUKS device. Great!
Since it is harmless to leave the vgchange in when lvmetad *is* available, I suppose I should leave it in. So I did: unfortunately this doesn't quite work right (see more below) > Less important for me but still to figure out, where to put the magic so > that 'systemctl stop /mountpoint' also disables the relevant VG and closes > the luks device. (I would like to do this with a removable drive in the > near future so it will be necessary then). OK; the necessary nonsense appears to be as follows. My specifics are * relevant mount point is /backup, defined as a systemd unit * I use 'backup' as the luks device name too * the vgname on top of my luks device is phobos_backup systemd-cryptsetup@backup.service: ... > [Unit] ... > BindsTo=backup.mount > Before=backup.mount ... > [Service] ... > # these two lines added by me, necessarily after the existing ExecStart and > before > # the existing ExecStop > ExecStart=/sbin/vgchange -a y phobos_backup > ExecStop=/sbin/vgchange -a n phobos_backup ... With the above: * "systemctl start /backup" works as it should * "systemctl stop /backup" de-activates the VG and LUKS device, thanks to the BindsTo * so does a plain "umount /backup"! Sadly * with use_lvmetad=1, I get a spurious failure message on the console on mount; however, nothing useful in the journal and both units are marked active/success * The BindsTo/Before has to be in the form "backup.mount" -- attempting to use "/backup" results in an "invalid argument" error on start-up. Systemd suggests that you should be able to use paths for mount units quite freely, sadly this doesn't work here. Still, with some warts, I've got what I need for now. -- Jonathan Dowland
