Re: [PATCH RFC] NFSD: fix cannot umounting mount points under pseudo root

["J. Bruce Fields" <bfields@fieldses.org>]

On Wed, Apr 29, 2015 at 12:57:28PM +1000, NeilBrown wrote:
> In any case, holding a mutex while waiting for an upcall is probably a bad
> idea.  We should try to find a way to work around that...

Yeah, it sounds fragile even if we could fix this particular issue in
mountd.

> Any ideas Bruce ?-)

Looking at nfsd_buffered_readdir():

	/*
	 * Various filldir functions may end up calling back into
	 * lookup_one_len() and the file system's ->lookup() method.
	 * These expect i_mutex to be held, as it would within readdir.
	 */
        host_err = mutex_lock_killable(&dir_inode->i_mutex);

and as far as I can tell Kinglong's approach (adding an unlock and lock
around nfsd_cross_mnt() calls might actually be OK.

Though in general I expect

	lock()
	...code...
	unlock()

to mark a critical section and would be unpleasantly surprised to
discover that a function somewhere in the middle had a buried
unlock/lock pair.

Maybe drop the locking from nfsd_buffered_readdir and *just* take the
i_mutex around lookup_one_len(), if that's the only place we need it?

--b.