[PATCHSET] Cancel requests at ring close time

[Jens Axboe <axboe@kernel.dk>]

Hi,

As it stands, io_uring cancels requests in an async fashion when the
ring fd is closed. This can cause some application confusion, if the
application expects files to be fully closed by the time close(2) of the
ring fd returns. We've had reports before where someone doing IO to
/mnt/file, closes the ring and then attempts a umount, and gets -EBUSY
on the umount. Or more recently from the TigerBeetle folks, where their
shutdown is a backend kill, and that still ends up with file locks being
active until the background cancelations have been run.

This series attempts to make that more predictable, by canceling
everything sync, upfront, when the ring fd is closed.

Series can also be found here:

https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git/log/?h=io_uring-exit-cancel.4

 include/linux/io_uring_types.h |   2 +
 io_uring/cancel.c              |  89 +++++++++++++---
 io_uring/cancel.h              |  13 ++-
 io_uring/io_uring.c            | 180 ++++++++++++++++++++++++++-------
 io_uring/notif.c               |   2 +
 io_uring/timeout.c             |  18 ++++
 io_uring/timeout.h             |   2 +
 io_uring/uring_cmd.c           |   4 +-
 io_uring/uring_cmd.h           |   2 +-
 9 files changed, 260 insertions(+), 52 deletions(-)

-- 
Jens Axboe