From: Arnd Bergmann <arnd@arndb.de>
To: "J. Bruce Fields" <bfields@fieldses.org>
Subject: Re: nfsd changes for 2.6.37
Date: Tue, 26 Oct 2010 22:18:36 +0200
Cc: Linus Torvalds <torvalds@linux-foundation.org>, linux-nfs@vger.kernel.org,
        linux-kernel@vger.kernel.org, Bryan Schumaker <bjschuma@netapp.com>
References: <20101026164549.GD19445@fieldses.org>
In-Reply-To: <20101026164549.GD19445@fieldses.org>
Content-Type: Text/Plain;
  charset="iso-8859-1"
Message-Id: <201010262218.36940.arnd@arndb.de>
Sender: linux-nfs-owner@vger.kernel.org
MIME-Version: 1.0

On Tuesday 26 October 2010 18:45:50 J. Bruce Fields wrote:
> Bryan Schumaker (1):
>       lockd: Mostly remove BKL from the server

Could you explain the "mostly" part of this commit?

The commit message only says "This patch removes all
but one call to lock_kernel() from the server." This one
call is what keeps us from removing the BKL from fs/locks.c
because I can't tell if you still suspect that lockd
needs to lock against posix file locks or if there was
a different reason for leaving it in.

I can't think of anything else that this might be locking
against because everything that might interact with lockd
now does not use the BKL any more and lockd is
single-threaded by definition.

My guess is that the only thing that really needs to
lock_flocks() in lockd are the nlm_file_inuse and
nlm_traverse_locks functions because they traverse
the inode->i_flock list. All the exported functions
from fs/lock.c take care of locking in their own way
(possibly not lease_get_time, as I just discovered,
but that was never called under the BKL...).

	Arnd