Re: [PATCH v2 1/2] iomap: fix zero padding data issue in concurrent append writes

[Christoph Hellwig <hch@infradead.org>]

On Wed, Nov 13, 2024 at 11:13:49AM -0500, Brian Foster wrote:
> >  static bool
> >  iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next)
> >  {
> > +	size_t size = iomap_ioend_extent_size(ioend);
> > +
> 
> The function name is kind of misleading IMO because this may not
> necessarily reflect "extent size." Maybe something like
> _ioend_size_aligned() would be more accurate..?

Agreed.  What also would be useful is a comment describing the
function and why io_size is not aligned.

> 1. It kind of feels like a landmine in an area where block alignment is
> typically expected. I wonder if a rename to something like io_bytes
> would help at all with that.

Fine with me.

> Another randomish idea might be to define a flag like
> IOMAP_F_EOF_TRIMMED for ioends that are trimmed to EOF. Then perhaps we
> could make an explicit decision not to grow or merge such ioends, and
> let the associated code use io_size as is.

I don't think such a branch is any cheaper than the rounding..