From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alessio Igor Bogani Subject: [PATCH 0/1] vfs: umount_begin BKL pushdown v2 Date: Fri, 24 Apr 2009 09:06:52 +0200 Message-ID: <1240556813-8739-1-git-send-email-abogani@texware.it> References: <20090423191934.GW1926@parisc-linux.org> Cc: Jonathan Corbet , =?utf-8?q?Fr=C3=A9d=C3=A9ric=20Weisbecker?= , Peter Zijlstra , LKML , LFSDEV , Ingo Molnar , Matthew Wilcox , Alessio Igor Bogani To: Alexander Viro Return-path: Received: from fg-out-1718.google.com ([72.14.220.157]:33010 "EHLO fg-out-1718.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750710AbZDXHHE (ORCPT ); Fri, 24 Apr 2009 03:07:04 -0400 In-Reply-To: <20090423191934.GW1926@parisc-linux.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: Push the BKL acquisition from vfs to every specific filesystems with hope that it can be eliminated in a second moment. Filesystems, which support umount_begin(), changed by this patch are 9p, nfs, cifs and fuse. Changes: Collapsed all patches into only one as requested by Al Viro. Moved CIFS_SB() down into BKL protected zone as requested by Matthew Wilcox. It is hard to say if tcon (into cifs's umount_begin() function) should be protected by BKL. Up to now umount_begin() is always called with BKL held so in uncertainty I maintain same logic. So I moved unlock_kernel() to at the bottom of that function as requested by Al Viro and Matthew Wilcox. Notes: About cifs's umount_begin() function I suspect that a deadlock or circular locking dependency can happens into kill-the-BKL tree if that patch is applied.