Return-Path: Received: from fieldses.org ([173.255.197.46]:41972 "EHLO fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932757AbdCUPuC (ORCPT ); Tue, 21 Mar 2017 11:50:02 -0400 Date: Tue, 21 Mar 2017 11:50:00 -0400 From: "J. Bruce Fields" To: Olga Kornievskaia Cc: Christoph Hellwig , linux-fsdevel@vger.kernel.org, Linux NFS Mailing List , ng-linux-team Subject: Re: [PATCH v1 2/3] VFS permit cross device vfs_copy_file_range Message-ID: <20170321155000.GB15402@fieldses.org> References: <20170302160211.30451-1-kolga@netapp.com> <20170302160211.30451-3-kolga@netapp.com> <20170302160714.GA4760@lst.de> <0ADCDD29-61BA-46CA-902C-06A6F7D06450@netapp.com> <20170315180913.GB18135@fieldses.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20170315180913.GB18135@fieldses.org> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Wed, Mar 15, 2017 at 02:09:13PM -0400, J. Bruce Fields wrote: > On Thu, Mar 02, 2017 at 11:38:03AM -0500, Olga Kornievskaia wrote: > > > On Mar 2, 2017, at 11:07 AM, Christoph Hellwig wrote: > > > NAK, we really should not do operations between different > > > superblocks. > > > > Can you provide some reasoning as to why? What would it break? The > > reasoning for including one is to allow for a file system to achieve > > better performance which seems like a feature that would be of great > > benefit. -- > > Yes, this has come up a few times. What's going on?: > > - There was an explanation, and I missed it. > - The explanation is complicated and Christoph hasn't had time > to write it up. > - Christoph has a strong suspicion there are issues without > being sure exactly where they are. > - Something else? As far as I can tell from talking at LSF: - file system operations crossing superblocks is unusual. No specific known issue. Also not sure I understand why splice isn't precedent. - implementation (with server acting as a client, long-running process, etc.) will be complicated and ugly. Sure, I guess we'll see how complicated and weigh that against any advantages. (Though I'm curious about case e.g. of btrfs clone--can't it easily clone across filesystem boundaries in some cases when they share storage?) Anyway. I'll read the patches. Probably not this week. --b.