On Wed, Aug 7, 2013 at 6:31 PM, Anand Avati <avati@redhat.com> wrote:
> Thanks for attaching the thread. Was very educative! I still do not quite
> understand - will umount() still work when inode_permission()[->getattr()]
> on the ancestors fail (with ESTALE etc.)?

Several cases:

1) mountpoint was removed
2) mountpoint's ancestor removed
3) mountpoint or ancestor moved

1) and 3) would only cause revalidate to fail, not inode operations
(permission or getattr) since the inodes of ancestors still exist.

2) could error out if the ->permission got to the server.  I'm not
sure it will and I don't have an NFS setup handy to test.

So it doesn't necessarily work but it can be made to work, and as
Linus pointed out, it might still be the least worst option.

BTW, attaching a little fuse filesystem that I've been testing this
with.  It's read-only, but the point here is that the tree is modified
*externally*, so that's not an issue.

Thanks,
Miklos