From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from ipmail06.adl2.internode.on.net ([150.101.137.129]:36565 "EHLO ipmail06.adl2.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753092AbeEVWF5 (ORCPT ); Tue, 22 May 2018 18:05:57 -0400 Date: Wed, 23 May 2018 08:05:52 +1000 From: Dave Chinner Subject: Re: [PATCH v3] fs: clear writeback errors in inode_init_always Message-ID: <20180522220552.GV10363@dastard> References: <20180518225037.GA26206@thunk.org> <630faadb74f608aa5a42649b81657e8b62d46bc3.camel@kernel.org> <20180519152700.GB4507@magnolia> <20180522164359.GB12940@magnolia> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180522164359.GB12940@magnolia> Sender: fstests-owner@vger.kernel.org To: "Darrick J. Wong" Cc: xfs , Matthew Wilcox , linux-fsdevel@vger.kernel.org, fstests@vger.kernel.org, Jeff Layton , Brian Foster List-ID: On Tue, May 22, 2018 at 09:43:59AM -0700, Darrick J. Wong wrote: > From: Darrick J. Wong > > In inode_init_always(), we clear the inode mapping flags, which clears > any retained error (AS_EIO, AS_ENOSPC) bits. Unfortunately, we do not > also clear wb_err, which means that old mapping errors can leak through > to new inodes. > > This is crucial for the XFS inode allocation path because we recycle old > in-core inodes and we do not want error state from an old file to leak > into the new file. This bug was discovered by running generic/036 and > generic/047 in a loop and noticing that the EIOs generated by the > collision of direct and buffered writes in generic/036 would survive the > remount between 036 and 047, and get reported to the fsyncs (on > different files!) in generic/047. > > Since we're changing the semantics of inode_init_always, we must also > change xfs_reinit_inode to retain the writeback error state when we go > to recover an inode that has been torn down in the vfs but not yet > disposed of by XFS. > > Signed-off-by: Darrick J. Wong > Reviewed-by: Jeff Layton > --- > v3: clear error state when allocating new inode > v2: retain AS_EIO/AS_ENOSPC across xfs inode reinit > --- > fs/inode.c | 1 + > fs/xfs/xfs_icache.c | 9 +++++++++ > fs/xfs/xfs_inode.c | 5 +++++ > 3 files changed, 15 insertions(+) > > diff --git a/fs/inode.c b/fs/inode.c > index 13ceb98c3bd3..3b55391072f3 100644 > --- a/fs/inode.c > +++ b/fs/inode.c > @@ -178,6 +178,7 @@ int inode_init_always(struct super_block *sb, struct inode *inode) > mapping->a_ops = &empty_aops; > mapping->host = inode; > mapping->flags = 0; > + mapping->wb_err = 0; > atomic_set(&mapping->i_mmap_writable, 0); > mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); > mapping->private_data = NULL; > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c > index 164350d91efc..d01f9544ff01 100644 > --- a/fs/xfs/xfs_icache.c > +++ b/fs/xfs/xfs_icache.c > @@ -298,6 +298,10 @@ xfs_reinit_inode( > uint64_t version = inode_peek_iversion(inode); > umode_t mode = inode->i_mode; > dev_t dev = inode->i_rdev; > + errseq_t old_err = inode->i_mapping->wb_err; > + bool as_eio = test_bit(AS_EIO, &inode->i_mapping->flags); > + bool as_enospc = test_bit(AS_ENOSPC, > + &inode->i_mapping->flags); > > error = inode_init_always(mp->m_super, inode); > > @@ -306,6 +310,11 @@ xfs_reinit_inode( > inode_set_iversion_queried(inode, version); > inode->i_mode = mode; > inode->i_rdev = dev; > + inode->i_mapping->wb_err = old_err; > + if (as_eio) > + set_bit(AS_EIO, &inode->i_mapping->flags); > + if (as_enospc) > + set_bit(AS_ENOSPC, &inode->i_mapping->flags); I just don't see this as valid. The inode has already been removed from the VFS cache when this happens (i.e. it's gone through iput_final()->evict->destroy_inode) and so any VFS-specific state is already invalid. We're retaining XFS-specific inode information across the recycling of the xfs_inode (i.e. stuff that doesn't change *on-disk* across recycling an inode, but is reset by inode_init_always()). This is a creating a new life cycle for the VFS inode (i.e. I_NEW is set), so *there is no VFS state retained* across this operation from it's previous life. Trying to do this is just going to lead us into nasty layering violations where the VFS state has to be hacked around just in case this recycling of the previous life cycle's state was the wrong thing to do. > diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c > index 02eae5059231..6c47ea3e577b 100644 > --- a/fs/xfs/xfs_inode.c > +++ b/fs/xfs/xfs_inode.c > @@ -835,6 +835,11 @@ xfs_ialloc( > inode->i_mode |= S_ISGID; > } > > + /* Reset all vfs error state. */ > + inode->i_mapping->wb_err = 0; > + clear_bit(AS_EIO, &inode->i_mapping->flags); > + clear_bit(AS_ENOSPC, &inode->i_mapping->flags); Just like this. This is a sure sign we're doing the wrong thing in xfs_reinit_inode(). VFS inode lifecycle state is trashed at ->destroy_inode. It does not persist into new instantiations of VFS inode state, regardless of that new inode lifecycle gets to the I_NEW state... Cheers, Dave. -- Dave Chinner david@fromorbit.com