[PATCH 14/21] Fix free/alloc reversal of btree block merge condition

[Valerie Aurora]

On Fri, Feb 14, 2025 at 9:20 PM Zach Brown <zab at zabbo.net> wrote:
>
> On Tue, Feb 11, 2025 at 10:19:11PM +0100, Valerie Aurora wrote:
> >       memmove_item_headers(tblk, bt, ind, 1);
> >       ngnfs_tblk_le16_add_cpu(tblk, &bt->tail_free, -bytes);
> > +     BUG_ON(le16_to_cpu(bt->tail_free) > NGNFS_BTREE_MAX_FREE);
> >       ngnfs_tblk_le16_add_cpu(tblk, &bt->total_free, -bytes);
> > +     BUG_ON(le16_to_cpu(bt->total_free) > NGNFS_BTREE_MAX_FREE);
>
> But this drove me a little bonkers.  Let's not sprinkle these amongst
> the block changes.
>
> If we're going to have these assertions, let's put them up at the
> opening of the function that'd be verifying the caller's promises.
> insert_item had the start of that.  delete_item didn't but could.
>
> I figured you'd want to tweak this instead of having my mangle the patch
> -- but I'm happy to throw it together if you'd rather not.  Whatever's
> easiest!

Makes sense, thanks for flagging it and explaining why! I'll include
the updated version in the next set, coming shortly.

Valerie