From: Chris Mason Subject: Re: [patch] fs: fix deadlocks in writeback_if_idle Date: Mon, 29 Nov 2010 19:50:46 -0500 Message-ID: <1291075500-sup-6320@think> References: <20101123100239.GA4232@amd> <1290515274-sup-3895@think> <20101124144740.bffb2716.akpm@linux-foundation.org> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cc: Nick Piggin , linux-fsdevel , Al Viro , linux-ext4 , linux-btrfs , Jan Kara , Eric Sandeen , "Theodore Ts'o" To: Andrew Morton Return-path: In-reply-to: <20101124144740.bffb2716.akpm@linux-foundation.org> Sender: linux-fsdevel-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org Excerpts from Andrew Morton's message of 2010-11-24 17:47:40 -0500: > On Tue, 23 Nov 2010 07:34:07 -0500 > Chris Mason wrote: > > > For btrfs there's only one bdi per SB, but for most everyone else a disk > > with a bunch of partitions is going to have multiple filesystems on the > > same bdi. > > um, please explain why that wasn't idiotic? The BDI is a > representation of a backing device and it's *supposed* to provide > visibility into what's happening against other partitions on the same > device. Creating a BDI per SB (it didn't even occur to me to think > that a filesystem was even able to do this) breaks that. > We don't really have visibility into the other partitions, we all just pretend they aren't there (this patch being the most recent example). Yes, it does help prevent N flushers seeking around on the drive but it does cause problems too. How is the btrfs one-bdi-per-super different from device mapper's one bdi per logical volume? We're all idiots together I suppose. As for having multiple bdis per FS, that was always a long term goal of mine when Jens was setting up the new flushers. I didn't want to confuse the initial implementation with it though. -chris