From: Jan Kara Subject: Re: [PATCH 1/2 V2] Direct IO for holes and fallocate: add end_io callback Date: Tue, 25 Aug 2009 15:19:41 +0200 Message-ID: <20090825131941.GF405@duck.novell.com> References: <1250092470.18329.27.camel@mingming-laptop> <20090819141557.GA4705@atrey.karlin.mff.cuni.cz> <20090824191137.GA19624@mit.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , Mingming , linux-ext4@vger.kernel.org, Eric Sandeen To: Theodore Tso Return-path: Received: from cantor.suse.de ([195.135.220.2]:45265 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752338AbZHYNTn (ORCPT ); Tue, 25 Aug 2009 09:19:43 -0400 Content-Disposition: inline In-Reply-To: <20090824191137.GA19624@mit.edu> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon 24-08-09 15:11:38, Theodore Tso wrote: > On Wed, Aug 19, 2009 at 04:15:57PM +0200, Jan Kara wrote: > > > +static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, > > > + ssize_t size, void *private) > > > +{ > > > + ext4_io_end_t *io_end = iocb->private; > > > + struct workqueue_struct *wq; > > > + > > > + /* if not hole or unwritten extents, just simple return */ > > > + if (!io_end || !size) > > > + return; > > > + io_end->offset = offset; > > > + io_end->size = size; > > > + wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; > > > + > > > + /* We need to convert unwritten extents to written */ > > > + queue_work(wq, &io_end->work); > > > + > > > + if (is_sync_kiocb(iocb)) > > > + flush_workqueue(wq); > > I don't think you can flush_workqueue here. end_io is called from > > interrupt context and flush_workqueue blocks for a long time... > > The wait should be done in ext4_direct_IO IMHO... > > I don't see a problem here? This is a direct_io end_io callback, not > a bio callback; so it's only called from an interrupt context in the > async I/O case, and we only call flush_workqueue() when the kiocb is > synchronous. Ah, right, I didn't realize that DIO end_io callback gets called from an interrupt context only for async IO. That also explains why XFS is fine doing the same thing. Thanks for the explanation. Honza -- Jan Kara SUSE Labs, CR