On Thu, Dec 06, 2018 at 05:46:01AM -0600, Goldwyn Rodrigues wrote: > This is not worth with btrfs. With non-page aligned I/O on btrfs, we > need to copy the first/last page of the extents for CoW. We'll need to do the same for XFS with reflink support anyway. So I'd rather handle it in common code if we can. > So, we > would end up using the exported functions anyways. Believe me, I have > spent some time getting btrfs iomap compatible before giving up. The > problems are btrfs needs to carry a lot of information across > iomap_begin and iomap_end. While the added private variable helps in > this, it also needs hooks in bio_submit() functions for crc calculations > during direct writes. Although we'd need to find a smarter way to handle this vs adding more indirect calls everywhere.
