Input: matrix_keypad - use flush_delayed_work()
diff mbox series

Message ID 20190207224650.GA49861@dtor-ws
State Mainlined
Commit a342083abe576db43594a32d458a61fa81f7cb32
Headers show
Series
  • Input: matrix_keypad - use flush_delayed_work()
Related show

Commit Message

Dmitry Torokhov Feb. 7, 2019, 10:46 p.m. UTC
We should be using flush_delayed_work() instead of flush_work() in
matrix_keypad_stop() to ensure that we are not missing work that is
scheduled but not yet put in the workqueue (i.e. its delay timer has not
expired yet).

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
---
 drivers/input/keyboard/matrix_keypad.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Sven Van Asbroeck Feb. 10, 2019, 5:43 p.m. UTC | #1
Hi Dmitry,

On Thu, Feb 7, 2019 at 5:46 PM Dmitry Torokhov
<dmitry.torokhov@gmail.com> wrote:
>
> We should be using flush_delayed_work() instead of flush_work() in
> matrix_keypad_stop() to ensure that we are not missing work that is
> scheduled but not yet put in the workqueue (i.e. its delay timer has not
> expired yet).
>

Could the following scenario cause a use-after-free?
(I am adding comments on lines starting with -->)

a) user closes the device handle:

static void matrix_keypad_stop(struct input_dev *dev)
{
        struct matrix_keypad *keypad = input_get_drvdata(dev);

        spin_lock_irq(&keypad->lock);
        keypad->stopped = true;
        spin_unlock_irq(&keypad->lock);

        flush_work(&keypad->work.work);
-->
--> new interrupt comes in, and schedules new delayed keypad->work (1)
-->
        /*
         * matrix_keypad_scan() will leave IRQs enabled;
         * we should disable them now.
         */
        disable_row_irqs(keypad);
}

b) user removes the keypad, or unloads the module:

static int matrix_keypad_remove(struct platform_device *pdev)
{
        struct matrix_keypad *keypad = platform_get_drvdata(pdev);

        matrix_keypad_free_gpio(keypad);
        input_unregister_device(keypad->input_dev);
        kfree(keypad);

-->
--> delayed keypad->work scheduled at (1) above executes anywhere past here,
--> and causes a user-after-free.
-->

        return 0;
}
Dmitry Torokhov Feb. 11, 2019, 8:29 a.m. UTC | #2
Hi Sven,

On Sun, Feb 10, 2019 at 12:43:21PM -0500, Sven Van Asbroeck wrote:
> Hi Dmitry,
> 
> On Thu, Feb 7, 2019 at 5:46 PM Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> >
> > We should be using flush_delayed_work() instead of flush_work() in
> > matrix_keypad_stop() to ensure that we are not missing work that is
> > scheduled but not yet put in the workqueue (i.e. its delay timer has not
> > expired yet).
> >
> 
> Could the following scenario cause a use-after-free?
> (I am adding comments on lines starting with -->)
> 
> a) user closes the device handle:
> 
> static void matrix_keypad_stop(struct input_dev *dev)
> {
>         struct matrix_keypad *keypad = input_get_drvdata(dev);
> 
>         spin_lock_irq(&keypad->lock);
>         keypad->stopped = true;
>         spin_unlock_irq(&keypad->lock);
> 
>         flush_work(&keypad->work.work);
> -->
> --> new interrupt comes in, and schedules new delayed keypad->work (1)

It will not schedule new work because we check keypad->stopped flag
in ISR.

Thanks.
Sven Van Asbroeck Feb. 11, 2019, 12:42 p.m. UTC | #3
On Mon, Feb 11, 2019 at 3:29 AM Dmitry Torokhov
<dmitry.torokhov@gmail.com> wrote:
>
> It will not schedule new work because we check keypad->stopped flag
> in ISR.
>

Yes, you are correct. Sorry for the confusion.

Reviewed-by: Sven Van Asbroeck <TheSven73@gmail.com>

Patch
diff mbox series

diff --git a/drivers/input/keyboard/matrix_keypad.c b/drivers/input/keyboard/matrix_keypad.c
index 403452ef00e6..3d1cb7bf5e35 100644
--- a/drivers/input/keyboard/matrix_keypad.c
+++ b/drivers/input/keyboard/matrix_keypad.c
@@ -222,7 +222,7 @@  static void matrix_keypad_stop(struct input_dev *dev)
 	keypad->stopped = true;
 	spin_unlock_irq(&keypad->lock);
 
-	flush_work(&keypad->work.work);
+	flush_delayed_work(&keypad->work);
 	/*
 	 * matrix_keypad_scan() will leave IRQs enabled;
 	 * we should disable them now.