Re: [PATCH 6/7] xfs: use iomap_valid method to detect stale cached iomaps

[Christoph Hellwig <hch@infradead.org>]

> +	*((int *)&iomap->private) = sequence;

> +static bool
> +xfs_buffered_write_iomap_valid(
> +	struct inode		*inode,
> +	const struct iomap	*iomap)
> +{
> +	int			seq = *((int *)&iomap->private);

I really hate this stuffing of the sequence into the private pointer.
The iomap structure isn't so size constrained that we have to do that,
so we can just add a sequence number field directly to it.  I don't
think that is a layering violation, as the concept of a sequence
numebr is pretty generic and we'll probably need it for all file systems
eventually.

> +
> +	if (seq != READ_ONCE(XFS_I(inode)->i_df.if_seq))
> +		return false;

Which makes me wonder if we could do away with the callback entirely
by adding an option sequence number pointer to the iomap_iter.  If set
the core code compares it against iomap->seq and we get rid of the
per-folio indirect call, and boilerplate code that would need to be
implemented in every file system.