From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751840Ab1F3NKU (ORCPT ); Thu, 30 Jun 2011 09:10:20 -0400 Received: from adelie.canonical.com ([91.189.90.139]:50997 "EHLO adelie.canonical.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751211Ab1F3NKQ (ORCPT ); Thu, 30 Jun 2011 09:10:16 -0400 Date: Thu, 30 Jun 2011 14:10:12 +0100 From: Seth Forshee To: Christoph Hellwig Cc: Anton Salikhmetov , linux-kernel@vger.kernel.org Subject: Re: hfsplus mount regression in 2.6.38 Message-ID: <20110630131012.GA19309@thinkpad-t410> Mail-Followup-To: Christoph Hellwig , Anton Salikhmetov , linux-kernel@vger.kernel.org References: <20110525142521.GA32511@thinkpad-t410> <20110527092522.GA11600@infradead.org> <20110527132356.GA13908@thinkpad-t410> <20110527182406.GB13908@thinkpad-t410> <20110602215821.GA13099@thinkpad-t410> <20110630113520.GA6744@infradead.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20110630113520.GA6744@infradead.org> 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 Thu, Jun 30, 2011 at 07:35:20AM -0400, Christoph Hellwig wrote: > On Thu, Jun 02, 2011 at 04:58:21PM -0500, Seth Forshee wrote: > > I took a crack at converting the users of direct bio to use > > bdev_logical_block_size instead of HFSPLUS_SECTOR_SIZE. sb->s_blocksize > > doesn't turn out to work because it may change after reading the > > volume header. The patch is below; feedback is appreciated. > > > > So far I've only done light testing, and no testing with large-sector > > devices since I don't have any to test with. I'm still concerned about > > duplicating data also in the page cache with this approach. Any thoughts > > on whether or not this is something to be worried about? > > Did you manage to test it on a large sector device? I've gotten some test results back, but the results are mixed. I haven't had much time yet to try to find out what's going wrong, but I hope to do so soon. > I'm be rather surprised if we actually need the read modify write > cycles. I've not seen any filesystem that doesn't align it's metadata > to the sector size yet. I agree. What I was more concerned about was that if you had e.g. a 2 KB sector device, metadata could possibly end up in the filesystem cache along with some file data in an adjacent block, and that as a result you could overwrite that data structure with stale data.