1675402 - [wayland] Crash on circular subsurface creation

Status: RESOLVED WORKSFORME

(Core :: Widget: Gtk, defect, P3)

Description

[Robert Mader [:rmader]]

Under very rare circumstances it appears that Firefox tries to make a surface a subsurface of itself - something like wl_subcompositor_get_subsurface(subcompositor, surface, surface). Right now this crashes GS, thus we are unlikely to get crash reports for it. I made a fix for Mutter to prevent that - once that lands I expect more crash reports here. Hopefully I'll be able to catch a backtrace of it soon now.

Observerd two times over the last week on Fedora 32, FF 82.

Mutter crash stack:

...
#391 0x00007f59ae6643d3 in meta_wayland_surface_notify_subsurface_state_changed (surface=0x398d6c0 [MetaWaylandSurface])
    at ../src/wayland/meta-wayland-surface.c:1852
#392 0x00007f59ae65f86b in meta_wayland_subsurface_notify_subsurface_state_changed (surface_role=0x3c7e900 [MetaWaylandSubsurface])
    at ../src/wayland/meta-wayland-subsurface.c:290
#393 0x00007f59ae6643a8 in meta_wayland_surface_role_notify_subsurface_state_changed (surface_role=0x3c7e900 [MetaWaylandSubsurface])
    at ../src/wayland/meta-wayland-surface.c:1845
#394 0x00007f59ae6643d3 in meta_wayland_surface_notify_subsurface_state_changed (surface=0x398d6c0 [MetaWaylandSurface])
    at ../src/wayland/meta-wayland-surface.c:1852
#395 0x00007f59ae65f86b in meta_wayland_subsurface_notify_subsurface_state_changed (surface_role=0x3c7e900 [MetaWaylandSubsurface])
    at ../src/wayland/meta-wayland-subsurface.c:290
#396 0x00007f59ae6643a8 in meta_wayland_surface_role_notify_subsurface_state_changed (surface_role=0x3c7e900 [MetaWaylandSubsurface])
...

and more until the stack is filled up.

See also https://gitlab.gnome.org/GNOME/mutter/-/issues/1521 and https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1550

Future error message for search engines:

wl_subcompositor@6: error 0: Circular relationship between wl_surface@12 and parent surface wl_surface@10

Comment 1

[Dzmitry Malyshau [:kvark]]

Martin, any thoughts on this?

Comment 2

[Martin Stránský [:stransky] (ni? me)]

I'll look at it when we get the firefox bt.

Comment 3

[Robert Mader [:rmader]]

This is super hard to track down. Just got a crash on Fedora 33 / FF 83, but all reports I could find were:

Nov 28 12:24:09 thinkpad-t460p gnome-shell[45748]: WL: unknown object (198), message get_subsurface(noo)
Nov 28 12:24:09 thinkpad-t460p gnome-shell[45748]: WL: error in client communication (pid 99620)
Nov 28 12:24:09 thinkpad-t460p firefox[99620]: Error 22 (Invalid argument) dispatching to Wayland display.

Comment 4

[Robert Mader [:rmader]]

Haven't seen this in a while (FF84), thus closing.