On Thu, Jul 20, 2023 at 01:41:11PM +0200, David Sterba wrote: > > Note that compress_file_range was even redirtying the locked_page twice > > given that extent_range_clear_dirty_for_io already redirties all pages > > in the range, which must include locked_page if there is one. > > This is probably the only scary patch in the series. I don't see > anything obviously wrong, the reditrying logic added due to the mmap > case is preserved. You'll be even more scared when I finally get to clearing the dirty patch and setting the writeback bit in the proper place so that the BTRFS_INODE_HAS_ASYNC_EXTENT can go away :) But that's still dozends of patches down the patch stack..