Deliver input events only to window owner

[Peter Hutterer]

On Mon, Nov 23, 2015 at 04:20:05PM -0800, Keith Packard wrote:
> 
> X allows multiple clients to select for keyboard, motion and button
> release events (button press events generate implicit grabs, and are
> only selectable by one client). This means that a client receiving an
> input event has no way of knowing if some other client has received the
> same event, allowing snooping of the keyboard input stream by selecting
> for keyboard input on the right windows in the system.
> 
> Here's a patch which says that if the owner of the window receives a
> core input event, then don't go try and send it to other clients. Other
> clients can still ask for events, they will simply never receive
> them.
> 
> This has no effect on key grabbing; a grabbed key will still be
> delivered to the grabbing client, even if that is not the owner of the
> window and if the owner has selected for non-grabbed key events.
> 
> -keith
> 
> From 8c84031b5f33fa4574c5c9db6b1257ff0e34c8a9 Mon Sep 17 00:00:00 2001
> From: Keith Packard <keithp at keithp.com>
> Date: Mon, 23 Nov 2015 16:04:54 -0800
> Subject: [PATCH xserver] Deliver input events exclusively to window owner
> 
> When the window owner selects for core keyboard and mouse events,
> don't go looking for other clients to deliver those events to.  This
> means that if the window owner receives a key or mouse event, then it
> can know that no other client could have also received the same event.
> 
> This could be generalized to deliver input events to only one client,
> but which client would receive the event is less well defined as it
> would depend on the order that clients selected for input on
> particular windows.
> 
> Signed-off-by: Keith Packard <keithp at keithp.com>
> ---
>  dix/events.c | 11 ++++++++++-
>  1 file changed, 10 insertions(+), 1 deletion(-)
> 
> diff --git a/dix/events.c b/dix/events.c
> index 4114471..d223e97 100644
> --- a/dix/events.c
> +++ b/dix/events.c
> @@ -160,6 +160,12 @@ Equipment Corporation.
>  #define PropagateMask ( \
>  	KeyPressMask | KeyReleaseMask | ButtonPressMask | ButtonReleaseMask | \
>  	MotionMask )
> +
> +/* Deliver these events only to the owner if the owner selects them */
> +#define OwnerPriorityMask (\
> +        KeyPressMask | KeyReleaseMask | \
> +        ButtonPressMask | ButtonReleaseMask | PointerMotionMask)
> +
>  #define PointerGrabMask ( \
>  	ButtonPressMask | ButtonReleaseMask | \
>  	EnterWindowMask | LeaveWindowMask | \
> @@ -2253,6 +2259,7 @@ DeliverEventsToWindow(DeviceIntPtr pDev, WindowPtr pWin, xEvent
>      Mask deliveryMask = 0;      /* If a grab occurs due to a button press, then
>                                     this mask is the mask of the grab. */
>      int type = pEvents->u.u.type;
> +    Bool deliver_other_clients = TRUE;
>  
>      /* Deliver to window owner */
>      if ((filter == CantBeFiltered) || core_get_type(pEvents) != 0) {
> @@ -2269,6 +2276,8 @@ DeliverEventsToWindow(DeviceIntPtr pDev, WindowPtr pWin, xEvent
>          case EVENT_DELIVERED:
>              /* We delivered to the owner, with our event mask */
>              deliveries++;
> +            if (filter & OwnerPriorityMask)
> +                deliver_other_clients = FALSE;
>              client = wClient(pWin);
>              deliveryMask = pWin->eventMask;
>              break;
> @@ -2278,7 +2287,7 @@ DeliverEventsToWindow(DeviceIntPtr pDev, WindowPtr pWin, xEvent
>      }
>  
>      /* CantBeFiltered means only window owner gets the event */
> -    if (filter != CantBeFiltered) {
> +    if (filter != CantBeFiltered && deliver_other_clients) {
>          enum EventDeliveryState rc;
>  
>          rc = DeliverEventToWindowMask(pDev, pWin, pEvents, count, filter,
> -- 
> 2.6.1
 
looks correct, and it would also be fairly trivial to add test cases for
this to XIT. I didn't check the spec for the exact wording, but either way
we should add it.
Reviewed-by: Peter Hutterer <peter.hutterer at who-t.net>

note that this does not handle XI2 raw keyboard events, they will need to be
handled as well.

Cheers,
   Peter