Hello, This patch series fixes a bug manifesting as occasional generic/418 failure when direct IO write extending a file races with buffered read. Patch 1 extends iomap DIO code to pass original IO size to ->end_dio handler so that ext4 can tell whether everything has succeeded and we don't need expensive DIO cleanup (possible truncate of leftover blocks beyond i_size). Patch 2 fixes the ext4 bug, patch 3 fixes unrelated problem I've found in ext4 DIO code. Honza
