[PATCH weston] xwm: Check whether the seat is NULL sometimes in weston_wm_handle_button

[Pekka Paalanen]

On Mon, 01 Sep 2014 20:07:46 +0800
Boyan Ding <stu_dby at 126.com> wrote:

> On Mon, 2014-09-01 at 12:14 +0300, Pekka Paalanen wrote:
> > On Fri, 29 Aug 2014 22:10:32 +0800
> > Boyan Ding <stu_dby at 126.com> wrote:
> > 
> > > Under some certain circumstances, pointer button may have been released
> > > when frame is still being resized/moved. When this happens, the picked
> > > seat is NULL and it will segfault when moving/resizing surfaces. Check
> > > whether the seat is NULL and ignore move/resize in that case.
> > > 
> > > Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=82827
> > > Signed-off-by: Boyan Ding <stu_dby at 126.com>
> > > ---
> > >  xwayland/window-manager.c | 6 ++++--
> > >  1 file changed, 4 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/xwayland/window-manager.c b/xwayland/window-manager.c
> > > index a216b76..f633324 100644
> > > --- a/xwayland/window-manager.c
> > > +++ b/xwayland/window-manager.c
> > > @@ -1648,12 +1648,14 @@ weston_wm_handle_button(struct weston_wm *wm, xcb_generic_event_t *event)
> > >  		weston_wm_window_schedule_repaint(window);
> > >  
> > >  	if (frame_status(window->frame) & FRAME_STATUS_MOVE) {
> > > -		shell_interface->move(window->shsurf, seat);
> > > +		if (seat != NULL)
> > > +			shell_interface->move(window->shsurf, seat);
> > >  		frame_status_clear(window->frame, FRAME_STATUS_MOVE);
> > >  	}
> > >  
> > >  	if (frame_status(window->frame) & FRAME_STATUS_RESIZE) {
> > > -		shell_interface->resize(window->shsurf, seat, location);
> > > +		if (seat != NULL)
> > > +			shell_interface->resize(window->shsurf, seat, location);
> > >  		frame_status_clear(window->frame, FRAME_STATUS_RESIZE);
> > >  	}
> > >  
> > 
> > Hi,
> > 
> > do you know if this condition is something that should be silently
> > ignored like in your patch, or should we at least print an error that
> > something unexpected is happening and being papered over?
> > 
> > Looking at how the seat is found:
> > 
> > static struct weston_seat *
> > weston_wm_pick_seat_for_window(struct weston_wm_window *window)
> > {
> > 	struct weston_wm *wm = window->wm;
> > 	struct weston_seat *seat, *s;
> > 
> > 	seat = NULL;
> > 	wl_list_for_each(s, &wm->server->compositor->seat_list, link) {
> > 		if (s->pointer != NULL &&
> > 		    s->pointer->focus == window->view &&
> > 		    s->pointer->button_count > 0 &&
> > 		    (seat == NULL ||
> > 		     s->pointer->grab_serial -
> > 		     seat->pointer->grab_serial < (1 << 30)))
> > 			seat = s;
> > 	}
> > 
> > 	return seat;
> > }
> > 
> > and that gets called as a response to an XCB input event via
> > weston_wm_handle_button...
> > 
> > The function will return NULL if there are no buttons pressed, even if
> > the pointer is focused on the window. Does that make sense in general?
> > Does it not cause every last-button-up event to hit seat==NULL?
> > So why don't we see this problem more often?
> > 
> > Could there be a problem in the shared frame code, maybe it makes
> > assumptions that won't work with X11?
> 
> Sorry I was in a rush when replying the last mail before leaving my
> table. Now I think it may be because the race in X and wayland.
> FRAME_STATUS_RESIZE (and so on) are set in frame.c, using wayland event
> handling, and are cleared in xwm, using XCB. The pointer may have been
> released between the two events. So the picked seat may be NULL while we
> have FRAME_STATUS_RESIZE set. I think we can safely ignore the event if
> pointer is released between the two.

Okay, so you believe the patch is really correct then?
Could you amend the commit message to include your understanding of
what is happening, rather than just "some certain circumstances"?
You don't have to be 100% confident that you are right.

That should make the patch perfect. :-)


Thanks,
pq