From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx1.redhat.com ([209.132.183.28]:37318 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752707AbdJaKKA (ORCPT ); Tue, 31 Oct 2017 06:10:00 -0400 Date: Tue, 31 Oct 2017 18:09:58 +0800 From: Eryu Guan Subject: Re: [PATCH] xfs: flush the range before zero partial block range on truncate down Message-ID: <20171031100958.GL17339@eguan.usersys.redhat.com> References: <20171027125328.25001-1-eguan@redhat.com> <20171028060529.GA29505@infradead.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20171028060529.GA29505@infradead.org> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Christoph Hellwig Cc: linux-xfs@vger.kernel.org On Fri, Oct 27, 2017 at 11:05:29PM -0700, Christoph Hellwig wrote: > On Fri, Oct 27, 2017 at 08:53:28PM +0800, Eryu Guan wrote: > > But it's possible that a buffer write overwrites the unwritten > > extent, which won't be converted to a normal extent until I/O > > completion, and iomap_truncate_page() skips zeroing wrongly because > > of the not-converted unwritten extent. This would cause a subsequent > > mmap read sees non-zeros beyond EOF. > > I suspect the right fix is to look at the in-core state im the iomap > truncate helpers instead of doing a duplicate flush. I may (and very likely) miss something, but my understanding is that iomap_truncate_page() already looks at the in-core extent state provided by xfs_file_iomap_begin() xfs_setattr_size() iomap_truncate_page() iomap_zero_range() iomap_apply() xfs_file_iomap_begin() # finds extent according to the range iomap_zero_range_actor() # sees iomap->type == IOMAP_UNWRITTEN And this in-core extent state won't be converted to XFS_EXT_NORM until writeback & I/O completion, so I think a flush is required. Thanks, Eryu