From: Jens Axboe <axboe@xxxxxxxxx> No upstream commit exists for this patch. If we drop the lock right after adding it to the timeout list, then someone attempting to kill timeouts will find it in an indeterminate state. That means that cancelation could attempt to cancel and remove a timeout, and then io_timeout() proceeds to init and add the timer afterwards. Ensure the timeout request is fully setup before we drop the completion lock, which guards cancelation as well. Reported-and-tested-by: Lee Jones <lee@xxxxxxxxxx> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- fs/io_uring.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) --- a/fs/io_uring.c +++ b/fs/io_uring.c @@ -2079,12 +2079,12 @@ static int io_timeout(struct io_kiocb *r req->sequence -= span; add: list_add(&req->list, entry); - spin_unlock_irq(&ctx->completion_lock); hrtimer_init(&req->timeout.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); req->timeout.timer.function = io_timeout_fn; hrtimer_start(&req->timeout.timer, timespec64_to_ktime(ts), HRTIMER_MODE_REL); + spin_unlock_irq(&ctx->completion_lock); return 0; }
