From: Jan Kara Subject: Re: [RFC PATCH 1/2] bdi: Create a flag to indicate that a backing device needs stable page writes Date: Thu, 1 Nov 2012 23:42:16 +0100 Message-ID: <20121101224216.GA31937@quack.suse.cz> References: <20121030154844.1898f068@notabene.brown> <20121030201424.GD19559@blackbox.djwong.org> <20121031091441.5fc6b412@notabene.brown> <20121031085604.GC19591@blackbox.djwong.org> <20121031115614.GC18424@quack.suse.cz> <20121031193652.GF19591@blackbox.djwong.org> <20121101085942.GA6584@quack.suse.cz> <5092B060.9070604@panasas.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , "Darrick J. Wong" , NeilBrown , "Martin K. Petersen" , Theodore Ts'o , linux-ext4 , linux-fsdevel To: Boaz Harrosh Return-path: Received: from cantor2.suse.de ([195.135.220.15]:34428 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1762442Ab2KAWmS (ORCPT ); Thu, 1 Nov 2012 18:42:18 -0400 Content-Disposition: inline In-Reply-To: <5092B060.9070604@panasas.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Thu 01-11-12 10:24:48, Boaz Harrosh wrote: > On 11/01/2012 01:59 AM, Jan Kara wrote: > <> > > (all block device inodes share one superblock). > > > > Really? that is not so good is it, for other obvious reasons. > Why is it not one superblock per BDI? That would be more obvious > to me. > > > Thoughts? > > It's a really bad design. I think it is worth fixing. For the above > problem, as well as a much better fit with our current thread-per-bdi, > and the rest of the Kernel model. No? So the fact that there is one superblock of virtual filesystem containing all block device inodes is inconvenient at times (that's why we have to have inode_to_bdi() function in fs/fs-writeback.c) but OTOH you cannot really attach these virtual block device inodes (note that these are different from an inode for an object say /dev/sda in the filesystem) anywhere else and having one sb per block device would really be an overkill. Honza -- Jan Kara SUSE Labs, CR