From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ted Ts'o Subject: Re: [PATCH, RFC] ext4: fix race in xattr block allocation path Date: Sat, 29 Oct 2011 10:12:45 -0400 Message-ID: <20111029141245.GI19536@thunk.org> References: <4EA5E702.8030303@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: ext4 development To: Eric Sandeen Return-path: Received: from li9-11.members.linode.com ([67.18.176.11]:40636 "EHLO test.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933746Ab1J2SMj (ORCPT ); Sat, 29 Oct 2011 14:12:39 -0400 Content-Disposition: inline In-Reply-To: <4EA5E702.8030303@redhat.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon, Oct 24, 2011 at 05:30:26PM -0500, Eric Sandeen wrote: > Ceph users reported that when using Ceph on ext4, the filesystem > would often become corrupted, containing inodes with incorrect > i_blocks counters. > > ... > > In other words, the sync/flush thread pops in and sets > i_delalloc_reserved_flag on the inode, which makes the xattr thread > think that it's in a delalloc path in ext4_new_meta_blocks(), > and add the block for a second time, after already having added > it once in the !i_delalloc_reserved_flag case in ext4_mb_new_blocks > > I think this can be as simple as taking i_data_sem before we > go down the new metablock path, so that we won't be testing > i_delalloc_reserved_flag in a race with xt4_get_blocks setting it. > > Signed-off-by: Eric Sandeen Thanks, applied for now. > > This seems like slight abuse of i_data_sem though, since we are > not modifying i_data[] in the inode on the xattr path. But nothing > else protects readers of i_delalloc_reserved_flag in other > threads. Thoughts? Agreed, we really should be depending on the EXT4_GET_BLOCK_DELALLOC_RESERVE and EXT4_MB_DELALLOC_RESERVED flags. I can't remember what caused us to plumb in the i_state flag, but that's clearly a hack which is biting us hard. I'll take this patch for now since it fixes a real problem, but we should figure out a better way of fixing this, hopefully that results in the i_state flag going away. - Ted