Re: [PATCH 2/3] xfs: drop IDONTCACHE on inodes when we mark them sick

[Dave Chinner <david@fromorbit.com>]

On Wed, Jun 02, 2021 at 08:12:46PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> When we decide to mark an inode sick, clear the DONTCACHE flag so that
> the incore inode will be kept around until memory pressure forces it out
> of memory.  This increases the chances that the sick status will be
> caught by someone compiling a health report later on.
> 
> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> ---
>  fs/xfs/xfs_health.c |    5 +++++
>  fs/xfs/xfs_icache.c |    3 ++-
>  2 files changed, 7 insertions(+), 1 deletion(-)
> 
> 
> diff --git a/fs/xfs/xfs_health.c b/fs/xfs/xfs_health.c
> index 8e0cb05a7142..824e0b781290 100644
> --- a/fs/xfs/xfs_health.c
> +++ b/fs/xfs/xfs_health.c
> @@ -231,6 +231,11 @@ xfs_inode_mark_sick(
>  	ip->i_sick |= mask;
>  	ip->i_checked |= mask;
>  	spin_unlock(&ip->i_flags_lock);
> +
> +	/* Keep this inode around so we don't lose the sickness report. */
> +	spin_lock(&VFS_I(ip)->i_lock);
> +	VFS_I(ip)->i_state &= ~I_DONTCACHE;
> +	spin_unlock(&VFS_I(ip)->i_lock);
>  }

Dentries will still be reclaimed, but the VFS will at least hold on
to the inode in this case.

>  /* Mark parts of an inode healed. */
> diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
> index c3f912a9231b..0e2b6c05e604 100644
> --- a/fs/xfs/xfs_icache.c
> +++ b/fs/xfs/xfs_icache.c
> @@ -23,6 +23,7 @@
>  #include "xfs_dquot.h"
>  #include "xfs_reflink.h"
>  #include "xfs_ialloc.h"
> +#include "xfs_health.h"
>  
>  #include <linux/iversion.h>
>  
> @@ -648,7 +649,7 @@ xfs_iget_cache_miss(
>  	 * time.
>  	 */
>  	iflags = XFS_INEW;
> -	if (flags & XFS_IGET_DONTCACHE)
> +	if ((flags & XFS_IGET_DONTCACHE) && xfs_inode_is_healthy(ip))

Hmmmm. xfs_inode_is_healthy() is kind of heavyweight for just
checking that ip->i_sick == 0. At this point, nobody else can be
accessing the inode, so we don't need masks nor a spinlock for
checking the sick field.

So why not:

	if ((flags & XFS_IGET_DONTCACHE) && !READ_ONCE(ip->i_sick))

Or maybe still use xfs_inode_is_healthy() but convert it to the
simpler, lockless sick check?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com