io_uring: io_wq_submit_work() should not touch req->rw

I've been chasing a weird and obscure crash that was userspace stack
corruption, and finally narrowed it down to a bit flip that made a
stack address invalid. io_wq_submit_work() unconditionally flips
the req->rw.ki_flags IOCB_NOWAIT bit, but since it's a generic work
handler, this isn't valid. Normal read/write operations own that
part of the request, on other types it could be something else.

Move the IOCB_NOWAIT clear to the read/write handlers where it belongs.

Signed-off-by: Jens Axboe <[email protected]>

Author: axboe

fs/io_uring.c

@@ -1817,6 +1817,10 @@ static int io_read(struct io_kiocb *req, struct io_kiocb **nxt,
 			return ret;
 	}
 
+	/* Ensure we clear previously set non-block flag */
+	if (!force_nonblock)
+		req->rw.ki_flags &= ~IOCB_NOWAIT;
+
 	file = req->file;
 	io_size = ret;
 	if (req->flags & REQ_F_LINK)
@@ -1906,6 +1910,10 @@ static int io_write(struct io_kiocb *req, struct io_kiocb **nxt,
 			return ret;
 	}
 
+	/* Ensure we clear previously set non-block flag */
+	if (!force_nonblock)
+		req->rw.ki_flags &= ~IOCB_NOWAIT;
+
 	file = kiocb->ki_filp;
 	io_size = ret;
 	if (req->flags & REQ_F_LINK)
@@ -3274,9 +3282,6 @@ static void io_wq_submit_work(struct io_wq_work **workptr)
 	struct io_kiocb *nxt = NULL;
 	int ret = 0;
 
-	/* Ensure we clear previously set non-block flag */
-	req->rw.ki_flags &= ~IOCB_NOWAIT;
-
 	if (work->flags & IO_WQ_WORK_CANCEL)
 		ret = -ECANCELED;