So while this looks good to me,
> + for (i = 0; i < bp->b_folio_count; i++) {
> + if (bp->b_folios[i])
> + __folio_put(bp->b_folios[i]);
The __folio_put here really needs to be folio_put or page alloc
debugging gets very unhappy.
But even with that fixed on top of this patch the first mount just hangs
without a useful kernel backtrace in /proc/*/stack, although running
with the entire ѕeries applied it does pass the basic sanity checking
so far.
