On 12/30/20 2:34 PM, Pavel Begunkov wrote: > 1/4 and 2/4 are for hangs during files unregister. > > The other two fix not running task works during cancellation. > Instead of patching task_work it moves io_uring_files_cancel() > before PF_EXITING, should be less intrusive. Was there a > particular reasong for doing that in exit_files()? > > Pavel Begunkov (4): > io_uring: add a helper for setting a ref node > io_uring: fix io_sqe_files_unregister() hangs > kernel/io_uring: cancel io_uring before task works > io_uring: cancel requests enqueued as task_work's > > fs/file.c | 2 -- > fs/io_uring.c | 54 ++++++++++++++++++++++++++++++++++----------------- > kernel/exit.c | 2 ++ > 3 files changed, 38 insertions(+), 20 deletions(-) Applied 1-3, as they look good. Can you resend 4/4 with the return added? -- Jens Axboe
