Ocfs2 devel - Jun 2020 - [Cluster-devel] [PATCH v11 16/25] fs: Convert mpage_readpages to mpage_readahead

Am Mi., 15. Apr. 2020 um 23:39 Uhr schrieb Matthew Wilcox <willy at
infradead.org>:
> From: "Matthew Wilcox (Oracle)" <willy at infradead.org>
>
> Implement the new readahead aop and convert all callers (block_dev,
> exfat, ext2, fat, gfs2, hpfs, isofs, jfs, nilfs2, ocfs2, omfs, qnx6,
> reiserfs & udf).  The callers are all trivial except for GFS2 &
OCFS2.
This patch leads to an ABBA deadlock in xfstest generic/095 on gfs2.

Our lock hierarchy is such that the inode cluster lock ("inode glock")
for an inode needs to be taken before any page locks in that inode's
address space. However, the readahead address space operation is
called with the pages already locked. When we try to grab the inode
glock inside gfs2_readahead, we'll deadlock with processes that are
holding that inode glock and trying to lock one of those same pages.

One possible solution is to use a trylock on the glock in
gfs2_readahead, and to give up the readahead in case of a locking
conflict. I have no idea how this is going to affect performance.

Any other ideas?

Thanks,
Andreas

On Wed, Jun 17, 2020 at 12:36:13AM +0200, Andreas Gruenbacher
wrote:
> Am Mi., 15. Apr. 2020 um 23:39 Uhr schrieb Matthew Wilcox <willy at
infradead.org>:
> > From: "Matthew Wilcox (Oracle)" <willy at
infradead.org>
> >
> > Implement the new readahead aop and convert all callers (block_dev,
> > exfat, ext2, fat, gfs2, hpfs, isofs, jfs, nilfs2, ocfs2, omfs, qnx6,
> > reiserfs & udf).  The callers are all trivial except for GFS2
& OCFS2.
> 
> This patch leads to an ABBA deadlock in xfstest generic/095 on gfs2.
> 
> Our lock hierarchy is such that the inode cluster lock ("inode
glock")
> for an inode needs to be taken before any page locks in that inode's
> address space.
How does that work for ...

writepage:              yes, unlocks (see below)
readpage:               yes, unlocks
invalidatepage:         yes
releasepage:            yes
freepage:               yes
isolate_page:           yes
migratepage:            yes (both)
putback_page:           yes
launder_page:           yes
is_partially_uptodate:  yes
error_remove_page:      yes

Is there a reason that you don't take the glock in the higher level
ops which are called before readhead gets called?  I'm looking at XFS,
and it takes the xfs_ilock SHARED in xfs_file_buffered_aio_read()
(called from xfs_file_read_iter).

Not that after -rc1 is a great time to be upending the locking model in
a filesystem ... but then, this has been baking in -mm for ten weeks and
the GFS2 mailing list has been on the cc for the patches for five months,
so I don't have a lot of sympathy for this.