From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: xattr iops locking improvement Date: Mon, 7 Jul 2003 18:57:01 +0100 Sender: linux-fsdevel-owner@vger.kernel.org Message-ID: <20030707185701.A6224@infradead.org> References: <200307071948.04703.agruen@suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Marcelo Tossati , linux-fsdevel@vger.kernel.org, Andrew Morton , Andrea Arcangeli , Dave Kleikamp , Nathan Scott , Chris Mason , Tim Shimmin , Jeff Mahoney , Steve Best Return-path: Received: from carisma.slowglass.com ([195.224.96.167]:56845 "EHLO phoenix.infradead.org") by vger.kernel.org with ESMTP id S267147AbTGGRme (ORCPT ); Mon, 7 Jul 2003 13:42:34 -0400 To: Andreas Gruenbacher Content-Disposition: inline In-Reply-To: <200307071948.04703.agruen@suse.de>; from agruen@suse.de on Mon, Jul 07, 2003 at 07:48:04PM +0200 List-Id: linux-fsdevel.vger.kernel.org On Mon, Jul 07, 2003 at 07:48:04PM +0200, Andreas Gruenbacher wrote: > Hi Marcelo, > > Andrew Morton has found that taking i_sem for all xattr iops leads to lock > contention. Particularly annoying is that writers can block operations like > `ls -l' because both need i_sem. We have bene working on a fix for 2.4 and > 2.5. The solution is a series of patches that are in Andrew's current tree, > and have been sent to Linus. I will also shortly put them on > http://acl.bestbits.at/. ~o(/me remembers having argued for no VFS-level locking on them a while ago) Can we please get rid of the i_sem on the other two operations, and finally move all locking to the implementation?