On Tue, Nov 24, 2015 at 02:24:21PM +0100, Jan Kara wrote:

Well, this problem seems to suggest that we have the freeze protection for
AIO writes wrong. We should call file_end_write() from aio_complete() and
not from aio_run_iocb()... I believe XFS and other filesystems may have
problems with this as well (CCed). Attached patch (so far only compile
tested since my test machine is pondering on something else) should fix
this.

Sounds like one way to do it, but we'd really want a vfs_* helper for
this so that it doesn't have to duplicated in other write_iter users
like the loop driver, which seems to be missing file
file_start_write/file_end_write entirely.