io_uring: fix io_kiocb.flags modification race in IOPOLL mode

While testing io_uring in arm, we found sometimes io_sq_thread() keeps
polling io requests even though there are not inflight io requests in
block layer. After some investigations, found a possible race about
io_kiocb.flags, see below race codes:
  1) in the end of io_write() or io_read()
    req->flags &= ~REQ_F_NEED_CLEANUP;
    kfree(iovec);
    return ret;

  2) in io_complete_rw_iopoll()
    if (res != -EAGAIN)
        req->flags |= REQ_F_IOPOLL_COMPLETED;

In IOPOLL mode, io requests still maybe completed by interrupt, then
above codes are not safe, concurrent modifications to req->flags, which
is not protected by lock or is not atomic modifications. I also had
disassemble io_complete_rw_iopoll() in arm:
   req->flags |= REQ_F_IOPOLL_COMPLETED;
   0xffff000008387b18 <+76>:    ldr     w0, [x19,#104]
   0xffff000008387b1c <+80>:    orr     w0, w0, #0x1000
   0xffff000008387b20 <+84>:    str     w0, [x19,#104]

Seems that the "req->flags |= REQ_F_IOPOLL_COMPLETED;" is  load and
modification, two instructions, which obviously is not atomic.

To fix this issue, add a new iopoll_completed in io_kiocb to indicate
whether io request is completed.

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

fs/io_uring.c

@@ -529,7 +529,6 @@ enum {
 	REQ_F_INFLIGHT_BIT,
 	REQ_F_CUR_POS_BIT,
 	REQ_F_NOWAIT_BIT,
-	REQ_F_IOPOLL_COMPLETED_BIT,
 	REQ_F_LINK_TIMEOUT_BIT,
 	REQ_F_TIMEOUT_BIT,
 	REQ_F_ISREG_BIT,
@@ -574,8 +573,6 @@ enum {
 	REQ_F_CUR_POS		= BIT(REQ_F_CUR_POS_BIT),
 	/* must not punt to workers */
 	REQ_F_NOWAIT		= BIT(REQ_F_NOWAIT_BIT),
-	/* polled IO has completed */
-	REQ_F_IOPOLL_COMPLETED	= BIT(REQ_F_IOPOLL_COMPLETED_BIT),
 	/* has linked timeout */
 	REQ_F_LINK_TIMEOUT	= BIT(REQ_F_LINK_TIMEOUT_BIT),
 	/* timeout request */
@@ -640,6 +637,8 @@ struct io_kiocb {
 	struct io_async_ctx		*io;
 	int				cflags;
 	u8				opcode;
+	/* polled IO has completed */
+	u8				iopoll_completed;
 
 	u16				buf_index;
 
@@ -1798,7 +1797,7 @@ static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
 		 * If we find a request that requires polling, break out
 		 * and complete those lists first, if we have entries there.
 		 */
-		if (req->flags & REQ_F_IOPOLL_COMPLETED) {
+		if (READ_ONCE(req->iopoll_completed)) {
 			list_move_tail(&req->list, &done);
 			continue;
 		}
@@ -1979,7 +1978,7 @@ static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
 		req_set_fail_links(req);
 	req->result = res;
 	if (res != -EAGAIN)
-		req->flags |= REQ_F_IOPOLL_COMPLETED;
+		WRITE_ONCE(req->iopoll_completed, 1);
 }
 
 /*
@@ -2012,7 +2011,7 @@ static void io_iopoll_req_issued(struct io_kiocb *req)
 	 * For fast devices, IO may have already completed. If it has, add
 	 * it to the front so we find it first.
 	 */
-	if (req->flags & REQ_F_IOPOLL_COMPLETED)
+	if (READ_ONCE(req->iopoll_completed))
 		list_add(&req->list, &ctx->poll_list);
 	else
 		list_add_tail(&req->list, &ctx->poll_list);
@@ -2140,6 +2139,7 @@ static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 		kiocb->ki_flags |= IOCB_HIPRI;
 		kiocb->ki_complete = io_complete_rw_iopoll;
 		req->result = 0;
+		req->iopoll_completed = 0;
 	} else {
 		if (kiocb->ki_flags & IOCB_HIPRI)
 			return -EINVAL;