Message ID | 20210824023208.392670-1-david@fromorbit.com (mailing list archive) |
---|---|
State | Accepted |
Headers | show |
Series | xfs: fix I_DONTCACHE | expand |
On Tue, Aug 24, 2021 at 12:32:08PM +1000, Dave Chinner wrote: > From: Dave Chinner <dchinner@redhat.com> > > Yup, the VFS hoist broke it, and nobody noticed. Bulkstat workloads > make it clear that it doesn't work as it should. Is there an easy way to test the dontcache behavior so that we don't screw this up again? /me's brain is fried, will study this in more detail in the morning. --D > Fixes: dae2f8ed7992 ("fs: Lift XFS_IDONTCACHE to the VFS layer") > Signed-off-by: Dave Chinner <dchinner@redhat.com> > --- > fs/xfs/xfs_icache.c | 3 ++- > fs/xfs/xfs_iops.c | 2 +- > 2 files changed, 3 insertions(+), 2 deletions(-) > > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c > index a3fe4c5307d3..f2210d927481 100644 > --- a/fs/xfs/xfs_icache.c > +++ b/fs/xfs/xfs_icache.c > @@ -84,8 +84,9 @@ xfs_inode_alloc( > return NULL; > } > > - /* VFS doesn't initialise i_mode! */ > + /* VFS doesn't initialise i_mode or i_state! */ > VFS_I(ip)->i_mode = 0; > + VFS_I(ip)->i_state = 0; > > XFS_STATS_INC(mp, vn_active); > ASSERT(atomic_read(&ip->i_pincount) == 0); > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c > index 0ff0cca94092..a607d6aca5c4 100644 > --- a/fs/xfs/xfs_iops.c > +++ b/fs/xfs/xfs_iops.c > @@ -1344,7 +1344,7 @@ xfs_setup_inode( > gfp_t gfp_mask; > > inode->i_ino = ip->i_ino; > - inode->i_state = I_NEW; > + inode->i_state |= I_NEW; > > inode_sb_list_add(inode); > /* make the inode look hashed for the writeback code */ > -- > 2.31.1 >
On Mon, Aug 23, 2021 at 07:58:41PM -0700, Darrick J. Wong wrote: > On Tue, Aug 24, 2021 at 12:32:08PM +1000, Dave Chinner wrote: > > From: Dave Chinner <dchinner@redhat.com> > > > > Yup, the VFS hoist broke it, and nobody noticed. Bulkstat workloads > > make it clear that it doesn't work as it should. > > Is there an easy way to test the dontcache behavior so that we don't > screw this up again? > > /me's brain is fried, will study this in more detail in the morning. Perhaps. We can measure how many xfs inodes are cached via the filesystem stats e.g. $ pminfo -t xfs.vnodes.active xfs.vnodes.active [number of vnodes not on free lists] $ sudo grep xfs_inode /proc/slabinfo | awk '{ print $2 }' 243440 $ pminfo -f xfs.vnodes.active xfs.vnodes.active value 243166 $ And so we should be able to run a bulkstat from fstests on a filesystem with a known amount of files in it and measure the number of cached inodes before/after... I noticed this because I recently re-added the threaded per-ag bulkstat scan to my scalability workload (via the xfs_io bulkstat command) after I dropped it ages ago because per-ag threading of fstests::src/bulkstat.c was really messy. It appears nobody has been paying attention to bulkstat memory usage (and therefore I_DONTCACHE behaviour) for some time.... Cheers, Dave.
On Tue, Aug 24, 2021 at 12:32:08PM +1000, Dave Chinner wrote: > From: Dave Chinner <dchinner@redhat.com> > > Yup, the VFS hoist broke it, and nobody noticed. Bulkstat workloads > make it clear that it doesn't work as it should. > > Fixes: dae2f8ed7992 ("fs: Lift XFS_IDONTCACHE to the VFS layer") > Signed-off-by: Dave Chinner <dchinner@redhat.com> Assuming the RFC test that I just sent out looks reasonable to you, I think I have a sufficient understanding of what DONTCACHE is supposed to do to say: Reviewed-by: Darrick J. Wong <djwong@kernel.org> --D > --- > fs/xfs/xfs_icache.c | 3 ++- > fs/xfs/xfs_iops.c | 2 +- > 2 files changed, 3 insertions(+), 2 deletions(-) > > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c > index a3fe4c5307d3..f2210d927481 100644 > --- a/fs/xfs/xfs_icache.c > +++ b/fs/xfs/xfs_icache.c > @@ -84,8 +84,9 @@ xfs_inode_alloc( > return NULL; > } > > - /* VFS doesn't initialise i_mode! */ > + /* VFS doesn't initialise i_mode or i_state! */ > VFS_I(ip)->i_mode = 0; > + VFS_I(ip)->i_state = 0; > > XFS_STATS_INC(mp, vn_active); > ASSERT(atomic_read(&ip->i_pincount) == 0); > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c > index 0ff0cca94092..a607d6aca5c4 100644 > --- a/fs/xfs/xfs_iops.c > +++ b/fs/xfs/xfs_iops.c > @@ -1344,7 +1344,7 @@ xfs_setup_inode( > gfp_t gfp_mask; > > inode->i_ino = ip->i_ino; > - inode->i_state = I_NEW; > + inode->i_state |= I_NEW; > > inode_sb_list_add(inode); > /* make the inode look hashed for the writeback code */ > -- > 2.31.1 >
diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c index a3fe4c5307d3..f2210d927481 100644 --- a/fs/xfs/xfs_icache.c +++ b/fs/xfs/xfs_icache.c @@ -84,8 +84,9 @@ xfs_inode_alloc( return NULL; } - /* VFS doesn't initialise i_mode! */ + /* VFS doesn't initialise i_mode or i_state! */ VFS_I(ip)->i_mode = 0; + VFS_I(ip)->i_state = 0; XFS_STATS_INC(mp, vn_active); ASSERT(atomic_read(&ip->i_pincount) == 0); diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index 0ff0cca94092..a607d6aca5c4 100644 --- a/fs/xfs/xfs_iops.c +++ b/fs/xfs/xfs_iops.c @@ -1344,7 +1344,7 @@ xfs_setup_inode( gfp_t gfp_mask; inode->i_ino = ip->i_ino; - inode->i_state = I_NEW; + inode->i_state |= I_NEW; inode_sb_list_add(inode); /* make the inode look hashed for the writeback code */