io_uring: prevent sq_thread from spinning when it should stop
authorStefano Garzarella <sgarzare@redhat.com>
Fri, 21 Feb 2020 15:42:16 +0000 (16:42 +0100)
committerJens Axboe <axboe@kernel.dk>
Fri, 21 Feb 2020 16:16:10 +0000 (09:16 -0700)
This patch drops 'cur_mm' before calling cond_resched(), to prevent
the sq_thread from spinning even when the user process is finished.

Before this patch, if the user process ended without closing the
io_uring fd, the sq_thread continues to spin until the
'sq_thread_idle' timeout ends.

In the worst case where the 'sq_thread_idle' parameter is bigger than
INT_MAX, the sq_thread will spin forever.

Fixes: 6c271ce2f1d5 ("io_uring: add submission polling")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c

index 6e249aa..b43467b 100644 (file)
@@ -5142,6 +5142,18 @@ static int io_sq_thread(void *data)
                 * to enter the kernel to reap and flush events.
                 */
                if (!to_submit || ret == -EBUSY) {
+                       /*
+                        * Drop cur_mm before scheduling, we can't hold it for
+                        * long periods (or over schedule()). Do this before
+                        * adding ourselves to the waitqueue, as the unuse/drop
+                        * may sleep.
+                        */
+                       if (cur_mm) {
+                               unuse_mm(cur_mm);
+                               mmput(cur_mm);
+                               cur_mm = NULL;
+                       }
+
                        /*
                         * We're polling. If we're within the defined idle
                         * period, then let us spin without work before going
@@ -5156,18 +5168,6 @@ static int io_sq_thread(void *data)
                                continue;
                        }
 
-                       /*
-                        * Drop cur_mm before scheduling, we can't hold it for
-                        * long periods (or over schedule()). Do this before
-                        * adding ourselves to the waitqueue, as the unuse/drop
-                        * may sleep.
-                        */
-                       if (cur_mm) {
-                               unuse_mm(cur_mm);
-                               mmput(cur_mm);
-                               cur_mm = NULL;
-                       }
-
                        prepare_to_wait(&ctx->sqo_wait, &wait,
                                                TASK_INTERRUPTIBLE);