Crash in wayfire-shell plugin when unplugging external monitor

[lcolitti]

Using wayfire from git at commit 27ddca3 . I opened my laptop and noticed wayfire had crashed. Looks like it happened when unplugging my external monitor. Simplified log below.

I notice that most of the methods in wfs_output check whether this->output is null. Should a null check be added to wfs_output::create_hotspot as well?

II 13-03-24 12:52:07.842 - [backend/drm/drm.c:1631] 'DP-3' disconnected
II 13-03-24 12:52:07.842 - [src/core/output-layout.cpp:1177] remove output: DP-3
II 13-03-24 12:52:07.842 - [src/core/output-layout.cpp:1090] new output: NOOP-1
II 13-03-24 12:52:07.842 - [src/core/output-layout.cpp:442] loaded mode auto
II 13-03-24 12:52:07.842 - [src/core/output-layout.cpp:615] Couldn't find matching mode 1280x720@0 for output NOOP-1. Trying to use custom mode(might not work)
...
EE 13-03-24 12:52:07.855 - [src/core/output-layout.cpp:514] disabling output: DP-3
II 13-03-24 12:52:07.859 - [src/core/output-layout.cpp:145] transfer views from DP-3 -> NOOP-1
II 13-03-24 12:52:07.971 - [backend/drm/drm.c:786] connector DP-3: Turning off
...
II 13-03-24 12:52:08.228 - [src/core/output-layout.cpp:1090] new output: NOOP-1
II 13-03-24 12:52:08.228 - [src/core/output-layout.cpp:442] loaded mode auto
II 13-03-24 12:52:08.228 - [src/core/output-layout.cpp:615] Couldn't find matching mode 1280x720@0 for output NOOP-1. Trying to use custom mode(might not work)
II 13-03-24 12:52:08.236 - [backend/drm/drm.c:782] connector eDP-1: Modesetting with 3840x2400 @ 60.000 Hz
...
II 13-03-24 12:52:09.550 - [src/core/output-layout.cpp:1130] remove output: NOOP-1
EE 13-03-24 12:52:09.550 - [src/core/output-layout.cpp:514] disabling output: NOOP-1
II 13-03-24 12:52:09.659 - [src/core/output-layout.cpp:145] transfer views from NOOP-1 -> eDP-1
II 13-03-24 12:52:10.068 - [backend/drm/drm.c:782] connector eDP-1: Modesetting with 3840x2400 @ 60.000 Hz
EE 13-03-24 12:52:10.235 - [src/main.cpp:131] Fatal error: Segmentation fault
EE 13-03-24 12:52:10.375 - #1  signal_handler(int) ../src/main.cpp:133
EE 13-03-24 12:52:11.196 - #2  __restore_rt libc_sigaction.c:?
EE 13-03-24 12:52:12.622 - #3  wf::output_t::get_layout_geometry() const ../src/output/output.cpp:182
EE 13-03-24 12:52:12.678 - #4  wfs_hotspot::calculate_hotspot_geometry(wf::output_t*, unsigned int, unsigned int) const ../plugins/protocols/wayfire-shell.cpp:108
EE 13-03-24 12:52:12.732 - #5  wfs_hotspot::wfs_hotspot(wf::output_t*, unsigned int, unsigned int, unsigned int, wl_client*, unsigned int) ../plugins/protocols/wayfire-shell.cpp:145
EE 13-03-24 12:52:12.785 - #6  wfs_output::create_hotspot(unsigned int, unsigned int, unsigned int, unsigned int) ../plugins/protocols/wayfire-shell.cpp:317 (discriminator 1)
EE 13-03-24 12:52:12.836 - #7  handle_zwf_output_create_hotspot(wl_client*, wl_resource*, unsigned int, unsigned int, unsigned int, unsigned int) ../plugins/protocols/wayfire-shell.cpp:339
EE 13-03-24 12:52:12.844 - #8  ffi_prep_go_closure ??:?

[ammen99]

We have recently had some fixes related to this, please test with latest master.

[ammen99]

Oh, I think I misread the stack trace. We had a similar problem in core, not in this plugin. I assume that yes, probably we ought to check for the output. Note however that we still have to create the hotspot object (otherwise we're violating the wayland protocol) but that object simply should be 'inert' - all requests to it should be ignored. I think it should be possible to adapt the wfs_hotspot object to support that.

[mark-herbert42]

Maybe the same issue is crashing wf-backround as it is also shell component...

[ammen99]

This is a race condition which I could reproduce with a specially crafted client, https://github.com/ammen99/wleird/blob/e0a81e4dd4c8df002016e60fffad03b908091ee5/wfshell-tester.c

I pushed a fix in #2211

[ammen99]

@mark-herbert42 That is a separate bug in the application (wf-background), please post your log there, not in the wayfire issue.