> >> It looks to me, that the reason is part of autofs, the autofs
> >> filesystem, runs in the Linux kernel. Bad luck.
> >
> > Well, I'll correct myself: you can make it work if you're willing to
> > take a two order of magnitude slowdown.
>
> I disagree. I would implement the nfs mounts of autofs like autofs
> handles local links. That means, instead of mounting the remote
> directory directly onto /home it would mount it on
> /aufofs/host/user.... and then make a link from /home/user to
> /aufofs/host/user.
>
> That means, the access will be as fast as Linux can handle symbolic links. If
> that is slow, then autofs is slow. But how fast is NFS anyway.
But that doesn't solve your problem; you still can't umount the old
directory if it is busy. Besides, you get the path problems, which I
*definitely* don't want to re-introduce (in fact, people frequently
request lofs, which would completely eliminate them.)
-hpa
