Enabling zwp_idle_inhibit_manager_v1 causes Mir to crash with some applications

[OPNA2608]

(The latest releases of) Taisei & Teeworlds as well as some other closed-source applications that can spawn Wayland windows via SDL2 cause the following error in the logs when their window is closed.

[2023-02-13 15:05:53.822899] < -warning- > frontend:Wayland: Exception processing WlSurface::~WlSurface() request: /build/source/src/server/frontend_wayland/window_wl_surface_role.cpp(416): Throw in function surface_destroyed
Dynamic exception type: boost::wrapexcept<std::runtime_error>
std::exception::what: wl_surface@28 destroyed before associated role

error in client communication (pid 204373)

If

• there is no current X11/Wayland session (i.e. runlevel 3),
• a new Mir session is started (on a tty or via ssh) with the zwp_idle_inhibit_manager_v1 Wayland extension enabled, and
• within this session, one such application causes this error

then this instead crashes the entire Mir session. I've verified this behaviour on NixOS unstable (miral-app & Miriway) and Ubuntu 22.10 (ubuntu-frame snap & source-built miral-app), Mir 2.12.0 and current master (a35b397).

!!! Fatal signal received. Attempting cleanup, but deadlock may occur
Mir fatal error: Unsupported attempt to continue after a fatal signal: SIGSEGV
!!! Fatal signal received. Attempting cleanup, but deadlock may occur
Mir fatal error: Unsupported attempt to continue after a fatal signal: SIGABRT

Here is the gdb backtrace with debug information.

Thread 16 "Mir/Wayland" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fd37ee840 (LWP 435461)]
0x0000007ff6bdd5a0 in wl_closure_invoke (closure=closure@entry=0x7fc0052830, flags=flags@entry=2, target=<optimized out>, opcode=0, data=<optimized out>, data@entry=0x7fc0000fb0) at ../src/connection.c:1021
1021	../src/connection.c: No such file or directory.
(gdb) bt
#0  0x0000007ff6bdd5a0 in wl_closure_invoke (closure=closure@entry=0x7fc0052830, flags=flags@entry=2, target=<optimized out>, opcode=0, data=<optimized out>, data@entry=0x7fc0000fb0) at ../src/connection.c:1021
#1  0x0000007ff6bd97d0 in wl_client_connection_data (fd=<optimized out>, mask=<optimized out>, data=0x7fc0000fb0) at ../src/wayland-server.c:437
#2  0x0000007ff6bdb678 in wl_event_loop_dispatch (loop=0xe2e750, timeout=timeout@entry=-1) at ../src/event-loop.c:1027
#3  0x0000007ff6bd9a44 in wl_display_run (display=display@entry=0xe2e660) at ../src/wayland-server.c:1431
#4  0x0000007ff7736450 in <lambda(wl_display*)>::operator() (__closure=<optimized out>, d=0xe2e660) at /build/source/src/server/frontend_wayland/wayland_connector.cpp:394
#5  std::__invoke_impl<void, mir::frontend::WaylandConnector::start()::<lambda(wl_display*)>, wl_display*> (__f=...) at /nix/store/4mhhaggjp57h7sr8diiiskbb86nh4ppy-gcc-9.5.0/include/c++/9.5.0/bits/invoke.h:60
#6  std::__invoke<mir::frontend::WaylandConnector::start()::<lambda(wl_display*)>, wl_display*> (__fn=...) at /nix/store/4mhhaggjp57h7sr8diiiskbb86nh4ppy-gcc-9.5.0/include/c++/9.5.0/bits/invoke.h:95
#7  std::thread::_Invoker<std::tuple<mir::frontend::WaylandConnector::start()::<lambda(wl_display*)>, wl_display*> >::_M_invoke<0, 1> (this=<optimized out>) at /nix/store/4mhhaggjp57h7sr8diiiskbb86nh4ppy-gcc-9.5.0/include/c++/9.5.0/thread:244
#8  std::thread::_Invoker<std::tuple<mir::frontend::WaylandConnector::start()::<lambda(wl_display*)>, wl_display*> >::operator() (this=<optimized out>) at /nix/store/4mhhaggjp57h7sr8diiiskbb86nh4ppy-gcc-9.5.0/include/c++/9.5.0/thread:251
#9  std::thread::_State_impl<std::thread::_Invoker<std::tuple<mir::frontend::WaylandConnector::start()::<lambda(wl_display*)>, wl_display*> > >::_M_run(void) (this=<optimized out>) at /nix/store/4mhhaggjp57h7sr8diiiskbb86nh4ppy-gcc-9.5.0/include/c++/9.5.0/thread:195
#10 0x0000007ff7cb0984 in execute_native_thread_routine () from /nix/store/f01gwicpyspl4qf7h8cwvgzg5x4s2a0g-gcc-9.5.0-lib/lib/libstdc++.so.6
#11 0x0000007ff799bdb4 in start_thread () from /nix/store/yy19vx24hhn7snsv7pffnqh1mknfp1sq-glibc-2.35-224/lib/libc.so.6
#12 0x0000007ff7a0289c in thread_start () from /nix/store/yy19vx24hhn7snsv7pffnqh1mknfp1sq-glibc-2.35-224/lib/libc.so.6

---

https://wayland.app/protocols/idle-inhibit-unstable-v1 indicates that Sway also supports this extension, and these applications don't crash there.

Unsure if relevant to this, but the applications also print the following on startup.

[2023-02-12 23:02:27.012698] < -warning- > mirserver: WlRegion::subtract not implemented. ignoring.

[AlanGriffiths]

@OPNA2608, thanks for this. Clearly there are two issues:

1. SDL2 is not tearing down the window correctly; and,
2. this should not crash the Mir server.

To help us with reproducing this, could you confirm which version of SDL2 you are using?

[OPNA2608]

Nixpkgs unstable currently uses 2.24.2, Ubuntu 22.10 uses 2.24.0+dfsg-1.

[AlanGriffiths]

I'm currently testing on 23.04: that has SDL2 2.26.2.

I tried downloading the latest Taisei release from the website, but this doesn't bind to zwp_idle_inhibit_manager_v1 nor show the symptoms you describe. (It appears to statically link libSDL2, so that may well be an older version)

[AlanGriffiths]

Ah, but I've just reproduced (after several attempts) with Neverputt (from the Ubuntu archive)

[OPNA2608]

I've built Taisei from source against 22.10's system libraries, sorry for not stating so before. I also tried building it against a custom SDL2 2.26.2 on NixOS and it still errors there. The static build from their website does not exhibit these problems.

[AlanGriffiths]

OK, this is down to a flawed implementation in Mir of zwp_idle_inhibit_manager_v1.create_inhibitor.

With SDL2 clients this is called when the surface hasn't been mapped, so the Mir code (for reasons unknown) doesn't create an zwp_idle_inhibitor_v1.

When the client subsequently destroys the zwp_idle_inhibitor_v1 that it (quite reasonably) thinks was created Mir can't find the object it is trying to destroy.