Comment # 2 on bug 1167342 from Neil Brown

I doubt that a kernel update would have fixed this.
I've seen this sort of thing before and I don't think we have a really
satisfactory solution.  One problem is that I don't remember exactly that the
problem is just now.
It relates to how systemd orders shutdown services, which must be the reverse
of startup.
But some NFS moutpoints start later (maybe after you log in), but also finish
late (after you have been logged out and are no longer using them).  That
doesn't follow the standard nested ordering.
THe tries to unmount things before killing uses, fails (because you are still
using the mountpoint), then logs you out, but doesn't try unmounting again.

Can you try to reproduce by shutting down while some process is running with
the mountpoint as its working directory?