When a mount entry in fstab uses the "nofail" option, systemd doesn't order the
corresponding mount unit with a "Before=local-fs.target" so if the mount unit
takes time to be activated, it won't delay the whole boot process.

However by doing this, we also remove ordering constraint for the shutdown
process. This can lead to the mount unit being stopped while some applications
are still using the corresponding mount point.

This problem was discussed in bug 1036962 but this bug was more about NFS or
remote fs. In this case the fix already exists since v218 and remote fs
automatically gain a Before=remote-fs.target after they're mounted.

A similar fix has been submitted upstream for local fs which can be found here:
https://github.com/systemd/systemd/pull/6102 and is waiting for approval.