Re: [PATCH 2/3] xfs: speed up parent pointer operations when possible

[Christoph Hellwig <hch@infradead.org>]

On Tue, Jan 20, 2026 at 10:39:46PM -0800, Darrick J. Wong wrote:
> and fall back to the attr intent machinery if that doesn't work.  We can
> use the same helpers for regular xattrs and parent pointers.

Not just can, but do :)  This should really help with ACL inheritance
as well.

> +/* Can we bypass the attr intent mechanism for better performance? */
> +static inline bool
> +xfs_attr_can_shortcut(

This is really a could and not a can, it might still not be possible
and we bail out.  Maybe reflect that in at least the comment, if not
also the name?

> +	if (rmt_blks || !xfs_attr_can_shortcut(args->dp)) {
> +		xfs_attr_defer_add(args, XFS_ATTR_DEFER_REPLACE);
> +		return 0;
> +	}
> +
> +	args->op_flags |= XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE;
> +
> +	error = xfs_attr_sf_removename(args);
> +	if (error)
> +		return error;
> +	if (args->attr_filter & XFS_ATTR_PARENT) {
> +		/*
> +		 * Move the new name/value to the regular name/value slots and
> +		 * zero out the new name/value slots because we don't need to
> +		 * log them for a PPTR_SET operation.
> +		 */
> +		xfs_attr_update_pptr_replace_args(args);
> +		args->new_name = NULL;
> +		args->new_namelen = 0;
> +		args->new_value = NULL;
> +		args->new_valuelen = 0;
> +	}
> +	args->op_flags &= ~XFS_DA_OP_REPLACE;
> +
> +	error = xfs_attr_try_sf_addname(args);

This mirrors what the state machine would do.  Although I wonder if
we should simply try to do the replace in one go.  But I guess I can
look into that as an incremental optimization later.

Otherwise looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>