From: Nick Piggin Subject: Re: [linux-cifs-client] Re: fsx-linux failing with latest cifs-2.6 git tree Date: Wed, 26 Nov 2008 14:09:43 +0100 Message-ID: <20081126130943.GF23649@wotan.suse.de> References: <20081121205151.1e3a09fe@tleilax.poochiereds.net> <524f69650811211802x74c0cadaided4c09f6c9e790e@mail.gmail.com> <1227329229.29807.9.camel@norville.austin.ibm.com> <20081122103916.6d9fc4e6@tleilax.poochiereds.net> <1227385664.10953.4.camel@norville.austin.ibm.com> <20081123065715.029b6ceb@tleilax.poochiereds.net> <524f69650811231832v12a03252ycbf53d9b06f178f8@mail.gmail.com> <20081124061918.7376c1ce@tleilax.poochiereds.net> <524f69650811252004nfc7936bga8aa0d34ef487cd@mail.gmail.com> <20081126071146.1bdcd630@tleilax.poochiereds.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Steve French , linux-fsdevel , pbadari@us.ibm.com, "linux-cifs-client@lists.samba.org" To: Jeff Layton Return-path: Received: from mx2.suse.de ([195.135.220.15]:33916 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752765AbYKZNJp (ORCPT ); Wed, 26 Nov 2008 08:09:45 -0500 Content-Disposition: inline In-Reply-To: <20081126071146.1bdcd630@tleilax.poochiereds.net> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Wed, Nov 26, 2008 at 07:11:46AM -0500, Jeff Layton wrote: > On Tue, 25 Nov 2008 22:04:04 -0600 > "Steve French" wrote: > > > Do you know why someone added the AOP_FLAG_UNITERRUPTIBLE check in > > cifs_write_begin instead of always marking a page up to date if we are > > writing the whole page? How often would that flag be set - I only see > > it in the path which calls generic_file_buffered_write > > > > /* If we are writing a full page it will be up to date, > > no need to read from the server */ > > if (len == PAGE_CACHE_SIZE && flags & AOP_FLAG_UNINTERRUPTIBLE) > > > > It seems restrictive > > Given that I now understand what AOP_FLAG_UNINTERRUPTIBLE is supposed > to do, this patch is probably what we need. Running tests on it now. That seems pretty reasonable, although keep in mind that AOP_FLAG_UNINTERRUPTIBLE is not going to be the common case (unless you're running loop or nfsd or something on the filesystem). It would be really nice to figure out a way to avoid the reads in the interruptible case as well.