From: Jan Kara Subject: Re: [PATCH 07/17] fs: Generic infrastructure for optional inode fields Date: Fri, 2 May 2014 21:44:23 +0200 Message-ID: <20140502194423.GD23636@quack.suse.cz> References: <1399036808-24521-1-git-send-email-jack@suse.cz> <1399036808-24521-8-git-send-email-jack@suse.cz> <20140502144409.GB5902@infradead.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , linux-fsdevel@vger.kernel.org, Al Viro To: Christoph Hellwig Return-path: Received: from cantor2.suse.de ([195.135.220.15]:60324 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752771AbaEBTo0 (ORCPT ); Fri, 2 May 2014 15:44:26 -0400 Content-Disposition: inline In-Reply-To: <20140502144409.GB5902@infradead.org> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Fri 02-05-14 07:44:09, Christoph Hellwig wrote: > On Fri, May 02, 2014 at 03:19:58PM +0200, Jan Kara wrote: > > There are parts of struct inode which are used only by a few filesystems > > (e.g. i_dquot pointers, i_mapping->private_list, ...). Thus all the > > Why don't you move the private_list into the fs inode like you did > in the previous few patches and remove the references from the generic > code? This would involve some heavy lifting in buffer.c, but that's > long overdue anyway. So I was looking into that option as well and even had some prelimiary patch. It is certainly doable but it seemed a bit ugly to pass around another pointer for the list of metadata buffers. It is especially ugly for stuff like remove_inode_buffers() or invalidate_inode_buffers(). There we'd have to provide new filesystem callbacks or something like that... Honza -- Jan Kara SUSE Labs, CR