On 8 Aug 2013 17:10, "Jan Engelhardt" <<a href="mailto:jengelh@inai.de">jengelh@inai.de</a>> wrote: > > > On Thursday 2013-08-08 16:26, Tom Gundersen wrote: > >On Thu, Aug 8, 2013 at 2:00 PM, Jan Engelhardt <<a href="mailto:jengelh@inai.de">jengelh@inai.de</a>> wrote: > >> > >> With systemd 195, an /etc/fstab line like > >> > >> /srv/www /home/www fuse.bindfs auto,group=foo,perms=g+rw 0 0 > >> > >> can fail to start if /dev/fuse does not exist yet. > > > >/dev/fuse should be created by udev (in 195, and by > >systemd-tmpflise-setup-dev in more recent versions), and this happens > >before the mounts from fstab are started. Can you verify that the > >ordering is correct for you with "systemctl show home-www.mount | grep > >After" (and so on until you hopefully find systemd-udevd.service)? > > home-www.mount: > After=local-fs-pre.target systemd-journald.socket -.mount > > local-fs-pre.target: > After=systemd-remount-fs.service md.service lvm.service dmraid.service > systemd-udevd.service > > > Even if systemd-udevd was listed in After=, there is not a guarantee > that udevd has already processed the event and managed to create > /dev/fuse, is there? /dev/fuse is a static node created unconditionally by udev as part of udevd startup (and not due to an uevent), so this should be guaranteed. Are you certain that a race with mknod is the problem? What output do you get? This was all recently refactored, so there is a small chance the problem will go away if you use a more recent version, but my guess is that something else is wrong... Cheers, Tom