fix: linux shutdown handling by YUCLing · Pull Request #55666 · electron/electron

YUCLing · 2026-11-08T06:28:54Z

Description of Change

Fixes #55560
Fixes partially #54547 (we already tries to quit normally when logout, but cannot guarantee due to race)
Related #55551

This PR moves D-Bus shutdown listener out of the PowerObserverLinux, into a new SessionEndListenerLinux class, which gets created at startup by ElectronBrowserMainParts.

After the change, Electron will always hold a shutdown inhibitor of the logind and use Chromium's dbus_xdg::InhibitSessionMonitor to listen for logouts when available.

We hold a shutdown inhibitor to ensure the clean app quit when shutting down. Previously, Electron only held this inhibitor when the app registered powerMonitor's shutdown event, and does not quit if app doesn't quit in the handler, same if the app doesn't even register the event. Now we hold it since startup, and when we receive the shutdown from logind, we emit the shutdown event and delay the shutdown if a handler calls preventDefault(), otherwise, we simply quit normally, instead of waiting for SIGTERM to kill us.

For logouts, we listen for Ending state using dbus_xdg::InhibitSessionMonitor. At this state, logout is not cancellable, and is not delayable. We still call shutdown event handlers and still wait for app code if preventDefault is called, otherwise we quit normally. There's a state QueryEnd, where we can delay the logouts, but the logouts are also cancellable. I chose the Ending because of that.

Checklist

I have built and tested this change
I have filled out the PR description
I have reviewed and verified the changes
npm test passes
tests are changed or added
relevant API documentation, tutorials, and examples are updated and follow the documentation style guide
PR release notes describe the change in a way relevant to app developers, and are capitalized, punctuated, past tense, and no longer than 80 characters.

Release Notes

Notes: Fixed an issue where Electron did not respond to shutdown on Linux.

ckerr

Thanks for writing this. This is the right idea for the fix & moving this out of PowerObserverLinux makes sense to me.

There are two blockers, but both have easy fixes:

the macOS and Windows builds break
a shutdown listener added while the main script loads is never registered. The delay lock staying held after a cancelled quit also needs a decision.

I have inline suggestions for both of these + a few other nits.

ckerr · 2026-11-09T18:48:06Z

 #if BUILDFLAG(IS_LINUX)
-  void SetListeningForShutdown(bool);
+  // Asked when the session starts ending. False means the app is handling it.
+  bool ShouldQuitForSessionEnd();
 #endif


This breaks the macOS and Windows builds. ShouldQuitForSessionEnd() is only declared on Linux but is defined for all platforms in the .cc, and electron_api_power_monitor_mac.mm:112 still binds the removed ShouldShutdown(). Keeping the original name and declaring it on all platforms fixes both:

Suggested change

#if BUILDFLAG(IS_LINUX)

void SetListeningForShutdown(bool);

// Asked when the session starts ending. False means the app is handling it.

bool ShouldQuitForSessionEnd();

#endif

// Called by native callers when the session starts ending. False means the

// app is handling it.

bool ShouldShutdown();

ckerr · 2026-11-09T18:48:07Z

 }

-bool PowerMonitor::ShouldShutdown() {
+bool PowerMonitor::ShouldQuitForSessionEnd() {


Matches the header suggestion.

Suggested change

bool PowerMonitor::ShouldQuitForSessionEnd() {

bool PowerMonitor::ShouldShutdown() {

ckerr · 2026-11-09T18:48:07Z

  auto builder =
      gin_helper::EventEmitterMixin<PowerMonitor>::GetObjectTemplateBuilder(
          isolate);
-#if BUILDFLAG(IS_LINUX)
-  builder.SetMethod("setListeningForShutdown",
-                    &PowerMonitor::SetListeningForShutdown);
-#endif
  return builder;


This local was only needed for the Linux SetMethod() call that was removed.

Suggested change

auto builder =

gin_helper::EventEmitterMixin<PowerMonitor>::GetObjectTemplateBuilder(

isolate);

#if BUILDFLAG(IS_LINUX)

builder.SetMethod("setListeningForShutdown",

&PowerMonitor::SetListeningForShutdown);

#endif

return builder;

return gin_helper::EventEmitterMixin<PowerMonitor>::GetObjectTemplateBuilder(

isolate);

ckerr · 2026-11-09T18:48:07Z

+
+namespace {
+
+const char kLogindServiceName[] = "org.freedesktop.login1";


nit: these constants and GetExecutableBaseName() duplicate the ones in power_observer_linux.cc. Could they be shared?

ckerr · 2026-11-09T18:48:07Z

+    bool success) {
+  DCHECK_CALLED_ON_VALID_SEQUENCE(sequence_checker_);
+  if (!success) {
+    VLOG(1) << "Failed to connect to " << interface_name << "." << signal_name;


OnLogindAvailable() still takes the delay lock after this failure, so logind would wait for a process that can't receive PrepareForShutdown. Clearing the proxy makes RequestInhibitor() return early:

Suggested change

VLOG(1) << "Failed to connect to " << interface_name << "." << signal_name;

VLOG(1) << "Failed to connect to " << interface_name << "." << signal_name;

// Don't make logind wait for a process that can't hear the shutdown.

logind_proxy_ = nullptr;

ckerr · 2026-11-09T19:22:35Z

  BrowserProcessImpl* browser_process() { return fake_browser_process_.get(); }

+#if BUILDFLAG(IS_LINUX)
+  // Null when the D-Bus buses are unavailable.


The listener is created unconditionally, so this is only null before it exists.

Suggested change

// Null when the D-Bus buses are unavailable.

// Null until PostEarlyInitialization() creates the listener.

ckerr · 2026-11-09T19:22:36Z

  dbus_thread_linux::Initialize(base::BindRepeating(&ExitOnSessionLoss));

+  session_end_listener_ = std::make_unique<SessionEndListenerLinux>(
+      dbus_thread_linux::GetSharedSystemBus(),
+      dbus_thread_linux::GetSharedSessionBus(),
+      base::RepeatingCallback<bool()>(), base::BindRepeating([] {
+        if (auto* browser = Browser::Get())
+          browser->Quit();
+      }));


A shutdown listener added while the main script is loading never gets registered. LoadEnvironment() runs the main script in PostEarlyInitialization(), before this listener is created, so for a top-level powerMonitor.on('shutdown', ...) the lookup in the PowerMonitor constructor returns null and nothing retries it. shutdown then never fires and the app always quits, even if its handler would have called preventDefault().

Creating the listener before the main script runs fixes this. PostEarlyInitialization() is outside this diff, so I can't use a suggestion. These statements can move to just before LoadEnvironment():

#if BUILDFLAG(IS_LINUX) JoinSystemFontConfigInit(); + dbus_thread_linux::Initialize(base::BindRepeating(&ExitOnSessionLoss)); + + // Must exist before the main script runs, since it may use powerMonitor. + session_end_listener_ = std::make_unique<SessionEndListenerLinux>( + dbus_thread_linux::GetSharedSystemBus(), + dbus_thread_linux::GetSharedSessionBus(), + base::RepeatingCallback<bool()>(), base::BindRepeating([] { + if (auto* browser = Browser::Get()) + browser->Quit(); + })); #endif // Load everything. node_bindings_->LoadEnvironment(node_env_.get());

ckerr · 2026-11-09T19:22:36Z

+
+  SessionEndListenerLinux(scoped_refptr<dbus::Bus> system_bus,
+                          scoped_refptr<dbus::Bus> session_bus,
+                          base::RepeatingCallback<bool()> should_quit,


nit: this is always empty, because PowerMonitor sets the handler through SetShouldQuitHandler(). If you drop it, also remove the base::RepeatingCallback<bool()>() argument where the listener is created.

Suggested change

base::RepeatingCallback<bool()> should_quit,

ckerr · 2026-11-09T19:22:37Z

+SessionEndListenerLinux::SessionEndListenerLinux(
+    scoped_refptr<dbus::Bus> system_bus,
+    scoped_refptr<dbus::Bus> session_bus,
+    base::RepeatingCallback<bool()> should_quit,


Goes with the should_quit suggestion in the header.

Suggested change

base::RepeatingCallback<bool()> should_quit,

ckerr · 2026-11-09T19:22:37Z

+    base::RepeatingClosure on_quit)
+    : system_bus_(std::move(system_bus)),
+      session_bus_(std::move(session_bus)),
+      should_quit_(std::move(should_quit)),


Goes with the should_quit suggestion in the header.

Suggested change

should_quit_(std::move(should_quit)),

ShouldQuitForSessionEnd() was declared for Linux only but defined unconditionally, and electron_api_power_monitor_mac.mm still binds ShouldShutdown(). Restore the original name and declare it on all platforms. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

LoadEnvironment() runs the app's main script in PostEarlyInitialization(), so a top-level powerMonitor.on('shutdown') was registered before the listener existed and never reached it. Move the D-Bus setup and the listener to PostEarlyInitialization(), and drop the listener constructor argument that was always empty. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

If PrepareForShutdown could not be subscribed, taking the delay lock would only make logind wait for a process that cannot react to it. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

A "before-quit", "will-quit" or window close handler can cancel the quit started for a session end, leaving logind waiting for a process that is not exiting. Notify the listener from Browser and drop the delay lock. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

Note in the breaking changes that shutdown is also emitted for a portal logout and that Electron quits unless a handler prevents it. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

Share the logind service, object path and interface constants and the executable name helper between the power observer and the session end listener. Include <string> where it is used, remove the unused testing accessors, and drop the object template builder local. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

The fake session bus portal only implemented FileChooser, so the InhibitSessionMonitor path never ran. Add org.freedesktop.portal.Inhibit with CreateMonitor, a session state signal and session objects, and test that a portal logout quits the app, and that a shutdown handler that prevents it keeps the app alive. Rename the template since it now hosts more than FileChooser. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

fix: linux shutdown handling

2de6726

ckerr self-requested a review November 9, 2026 18:29

ckerr requested changes Nov 9, 2026

View reviewed changes

YUCLing and others added 8 commits November 10, 2026 21:53

fix: skip the shutdown inhibitor when the signal connection failed

39692d6

If PrepareForShutdown could not be subscribed, taking the delay lock would only make logind wait for a process that cannot react to it. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

docs: add the Linux shutdown behavior change

361d9db

Note in the breaking changes that shutdown is also emitted for a portal logout and that Electron quits unless a handler prevents it. Co-authored-by: Charles Kerr <70381+ckerr@users.noreply.github.com>

chore: code format

7a1f01b

YUCLing requested a review from a team as a code owner November 10, 2026 14:11

Navigation Menu

Repository navigation

fix: linux shutdown handling - #55666

fix: linux shutdown handling#55666
YUCLing wants to merge 9 commits into
electron:mainfrom
YUCLing:fix/linux-shutdown-handling

YUCLing commented Nov 8, 2026

Uh oh!

ckerr left a comment

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

ckerr Nov 9, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	bool PowerMonitor::ShouldQuitForSessionEnd() {
	bool PowerMonitor::ShouldShutdown() {


		namespace {

		const char kLogindServiceName[] = "org.freedesktop.login1";

	// Null when the D-Bus buses are unavailable.
	// Null until PostEarlyInitialization() creates the listener.

Conversation

YUCLing commented Nov 8, 2026

Description of Change

Checklist

Release Notes

Uh oh!

ckerr left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants