On Tue, Oct 08, 2024 at 08:36:02AM +1100, Dave Chinner wrote: > I think this map size limiting is completely unnecessary for > buffered writeback - buffered writes are throttled against writeback > by balance_dirty_pages(), not by extent allocation size. The size of > the delayed allocation or the overwrite map is largely irrelevant - > we're going to map the entire range during a write, do it just > doesn't matter what size the mapping is... Yes. This goes back all the way to your original iomap prototype, but even back then balance_dirty_pages should have done all the work.
