From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755483Ab2AXQcc (ORCPT ); Tue, 24 Jan 2012 11:32:32 -0500 Received: from mx1.redhat.com ([209.132.183.28]:51761 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755403Ab2AXQca (ORCPT ); Tue, 24 Jan 2012 11:32:30 -0500 Date: Tue, 24 Jan 2012 11:32:24 -0500 From: Vivek Goyal To: Tejun Heo Cc: axboe@kernel.dk, ctalbott@google.com, rni@google.com, linux-kernel@vger.kernel.org, Kay Sievers Subject: Re: [PATCH 14/16] blkcg: don't allow or retain configuration of missing devices Message-ID: <20120124163224.GB17291@redhat.com> References: <1327360193-24679-1-git-send-email-tj@kernel.org> <1327360193-24679-15-git-send-email-tj@kernel.org> <20120124154224.GA17291@redhat.com> <20120124155309.GA12906@google.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20120124155309.GA12906@google.com> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Jan 24, 2012 at 07:53:09AM -0800, Tejun Heo wrote: > On Tue, Jan 24, 2012 at 10:42:24AM -0500, Vivek Goyal wrote: > > On Mon, Jan 23, 2012 at 03:09:51PM -0800, Tejun Heo wrote: > > > + disk = get_gendisk(dev, &part); > > > + > > > + if ((!disk || part) && temp) { > > > + ret = -ENODEV; > > > + goto out; > > > + } > > > + > > > + rcu_read_lock(); > > > + > > > + if (disk && !part) { > > > + spin_lock_irq(disk->queue->queue_lock); > > > + blkg = blkg_lookup_create(blkcg, disk->queue, plid); > > > + spin_unlock_irq(disk->queue->queue_lock); > > > > If queue is in bypass mode, is group creation and linking allowed when > > somebody is trying to set per device rules. (/me is thinking if there are > > any potential races between elevator switch and rule setting in cgroup). > > Yeap, there is. Nice catch. blkg_lookup_create() should be testing > blk_queue_bypass() instead of blk_queue_dead() and parse_and_set > should probably retry after a bit. Will update. So now group creation fails if queue is in bypass mode. So a per device rule setting will fail both for cfq and throttling logic while device is still around. So a user is supposed to retry? How would a user know that elevator switch is happening and he should retry and till when should he retry. Thanks Vivek