Re: [PATCH 4/4] xfs: use separate btree cursor slab for each btree type

From: "Darrick J. Wong" <djwong@kernel.org>
To: Dave Chinner <david@fromorbit.com>
Cc: chandan.babu@oracle.com, chandanrlinux@gmail.com,
	linux-xfs@vger.kernel.org
Subject: Re: [PATCH 4/4] xfs: use separate btree cursor slab for each btree type
Date: Mon, 27 Sep 2021 11:21:22 -0700	[thread overview]
Message-ID: <20210927182122.GT570615@magnolia> (raw)
In-Reply-To: <20210926004721.GD1756565@dread.disaster.area>

On Sun, Sep 26, 2021 at 10:47:21AM +1000, Dave Chinner wrote:
> On Thu, Sep 23, 2021 at 06:27:59PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@kernel.org>
> > 
> > Now that we have the infrastructure to track the max possible height of
> > each btree type, we can create a separate slab zone for cursors of each
> > type of btree.  For smaller indices like the free space btrees, this
> > means that we can pack more cursors into a slab page, improving slab
> > utilization.
> > 
> > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > ---
> >  fs/xfs/libxfs/xfs_btree.c |   12 ++++++------
> >  fs/xfs/libxfs/xfs_btree.h |    9 +--------
> >  fs/xfs/xfs_super.c        |   33 ++++++++++++++++++++++++---------
> >  3 files changed, 31 insertions(+), 23 deletions(-)
> > 
> > 
> > diff --git a/fs/xfs/libxfs/xfs_btree.c b/fs/xfs/libxfs/xfs_btree.c
> > index 120280c998f8..3131de9ae631 100644
> > --- a/fs/xfs/libxfs/xfs_btree.c
> > +++ b/fs/xfs/libxfs/xfs_btree.c
> > @@ -26,7 +26,6 @@
> >  /*
> >   * Cursor allocation zone.
> >   */
> > -kmem_zone_t	*xfs_btree_cur_zone;
> >  struct xfs_btree_cur_zone xfs_btree_cur_zones[XFS_BTNUM_MAX] = {
> >  	[XFS_BTNUM_BNO]		= { .name = "xfs_alloc_btree_cur" },
> >  	[XFS_BTNUM_INO]		= { .name = "xfs_ialloc_btree_cur" },
> > @@ -364,6 +363,7 @@ xfs_btree_del_cursor(
> >  	struct xfs_btree_cur	*cur,		/* btree cursor */
> >  	int			error)		/* del because of error */
> >  {
> > +	struct xfs_btree_cur_zone *bczone = &xfs_btree_cur_zones[cur->bc_btnum];
> >  	int			i;		/* btree level */
> >  
> >  	/*
> > @@ -386,10 +386,10 @@ xfs_btree_del_cursor(
> >  		kmem_free(cur->bc_ops);
> >  	if (!(cur->bc_flags & XFS_BTREE_LONG_PTRS) && cur->bc_ag.pag)
> >  		xfs_perag_put(cur->bc_ag.pag);
> > -	if (cur->bc_maxlevels > XFS_BTREE_CUR_ZONE_MAXLEVELS)
> > +	if (cur->bc_maxlevels > bczone->maxlevels)
> >  		kmem_free(cur);
> >  	else
> > -		kmem_cache_free(xfs_btree_cur_zone, cur);
> > +		kmem_cache_free(bczone->zone, cur);
> >  }
> >  
> >  /*
> > @@ -5021,12 +5021,12 @@ xfs_btree_alloc_cursor(
> >  {
> >  	struct xfs_btree_cur	*cur;
> >  	unsigned int		maxlevels = xfs_btree_maxlevels(mp, btnum);
> > +	struct xfs_btree_cur_zone *bczone = &xfs_btree_cur_zones[btnum];
> >  
> > -	if (maxlevels > XFS_BTREE_CUR_ZONE_MAXLEVELS)
> > +	if (maxlevels > bczone->maxlevels)
> >  		cur = kmem_zalloc(xfs_btree_cur_sizeof(maxlevels), KM_NOFS);
> >  	else
> > -		cur = kmem_cache_zalloc(xfs_btree_cur_zone,
> > -				GFP_NOFS | __GFP_NOFAIL);
> > +		cur = kmem_cache_zalloc(bczone->zone, GFP_NOFS | __GFP_NOFAIL);
> 
> When will maxlevels ever be greater than bczone->maxlevels? Isn't
> the bczone->maxlevels case always supposed to be the tallest
> possible height for that btree?

It should never happen, provided that the maxlevels computation and
verification are all correct.  I thought it was important to leave the
heap allocation in here as a fallback, since the consequence for getting
the size calculations wrong is corrupt kernel memory.

--D

> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com