On Tue, Jan 07, 2025 at 07:46:37AM +0100, Christoph Hellwig wrote:
> On Mon, Jan 06, 2025 at 10:42:24PM -0800, Darrick J. Wong wrote:
> > So I guess b_io_remaining was the count of in-flight bios?
>
> Yes. Plus a bias of 1 for the submitting context so that the
> completion isn't run until all bios are submitted.
>
> > And making a
> > chain of bios basically just moves all that to the bio layer, so all
> > xfs_buf needs to know is whether or not a bio is in progress, right?
>
> Yes.
>
> > Eerrugh, I spent quite a while trying to wrap my head around the old
> > code when I was writing the in-memory xfs_buf support. This is much
> > less weird to look at...
> >
> > > + for (map = 0; map < bp->b_map_count - 1; map++) {
> >
> > ...but why isn't this "map < bp->b_map_count"?
>
> Because the final ("main") bio is submitted outside the loop as the
> loop body chains the ones before to it. I guess this should go into
> a comment to confuse the readers a little less.
Ah, right, because we start with one bio for N maps, which means we need
to split it N-1 times to end up with N (chained) bios, one for each map.
Yes, comment please. :)
--D
