fix(tray): stability, no-console, icon logic, Windows GUI build

2026-04-17 17:38:37 +07:00
parent a4d9f21553
commit a7703701b8
1 changed files with 154 additions and 37 deletions
@@ -44,13 +44,22 @@ type app struct {
 }

 func main() {
-	// Single instance mutex
+	// Single instance mutex. Lock the OS thread so that CreateMutex and
+	// the subsequent error check reference the same thread-local LastError.
+	// (Go may otherwise reschedule the goroutine onto a different OS thread
+	// between the two syscalls, yielding a meaningless LastError value.)
+	runtime.LockOSThread()
 	mutexName, _ := windows.UTF16PtrFromString("Global\\Tailscale-Custom-Tray-Mutex")
-	handle, err := windows.CreateMutex(nil, false, mutexName)
-	if err != nil {
+	handle, createErr := windows.CreateMutex(nil, false, mutexName)
+	lastErr := windows.GetLastError()
+	runtime.UnlockOSThread()
+	if handle == 0 {
+		// Real failure (access denied, invalid args, etc.)
+		log.Printf("CreateMutex failed: %v", createErr)
 		os.Exit(1)
 	}
-	if windows.GetLastError() == windows.ERROR_ALREADY_EXISTS {
+	if lastErr == windows.ERROR_ALREADY_EXISTS {
+		// Another instance is already running; exit silently with success.
 		windows.CloseHandle(handle)
 		os.Exit(0)
 	}
@@ -59,6 +68,14 @@ func main() {
 	setupLogging()
 	log.Println("tailscale-tray starting")

+	// Top-level panic recovery so a stray panic in any goroutine that
+	// eventually reaches main doesn't just vanish the tray icon silently.
+	defer func() {
+		if r := recover(); r != nil {
+			log.Printf("PANIC in main: %v", r)
+		}
+	}()
+
 	a := &app{
 		lc:        &local.Client{},
 		rebuildCh: make(chan struct{}, 1),
@@ -68,9 +85,44 @@ func main() {

 	a.updateState()
 	go a.watchIPNBus()
+	go a.periodicRefresh()
 	systray.Run(a.onReady, a.onExit)
 }

+// periodicRefresh acts as a safety-net against a silently-stuck IPN bus
+// watcher (e.g. broken pipe on tailscaled restart that does not surface as
+// an error on Next()). Every 15s it forces a fresh status snapshot and,
+// if the backend state actually changed, triggers a menu rebuild.
+func (a *app) periodicRefresh() {
+	defer func() {
+		if r := recover(); r != nil {
+			log.Printf("PANIC in periodicRefresh: %v", r)
+		}
+	}()
+	ticker := time.NewTicker(15 * time.Second)
+	defer ticker.Stop()
+	var lastState string
+	for {
+		select {
+		case <-a.bgCtx.Done():
+			return
+		case <-ticker.C:
+		}
+		a.updateState()
+		a.mu.Lock()
+		cur := ""
+		if a.status != nil {
+			cur = a.status.BackendState
+		}
+		a.mu.Unlock()
+		if cur != lastState {
+			log.Printf("periodicRefresh: state changed %q -> %q, rebuilding", lastState, cur)
+			lastState = cur
+			a.triggerRebuild()
+		}
+	}
+}
+
 func setupLogging() {
 	dirs := []string{
 		filepath.Join(os.Getenv("ProgramData"), "Tailscale-Custom", "Logs"),
@@ -163,9 +215,10 @@ func (a *app) rebuild() {

 	// Show IP on a separate line when connected
 	if isRunning && a.status != nil && a.status.Self != nil && len(a.status.Self.TailscaleIPs) > 0 {
-		ipLine := systray.AddMenuItem("  IP: "+a.status.Self.TailscaleIPs[0].String(), "Click to copy")
+		selfIP := a.status.Self.TailscaleIPs[0].String()
+		ipLine := systray.AddMenuItem("  IP: "+selfIP, "Click to copy")
 		onClick(ctx, ipLine, func() {
-			copyToClipboard(a.status.Self.TailscaleIPs[0].String())
+			copyToClipboard(selfIP)
 		})
 	}

@@ -413,7 +466,10 @@ func (a *app) doLogin() {
 	}
 	defer atomic.StoreInt32(&a.inAction, 0)

+	// Take a thread-safe snapshot of the current profile URL.
+	a.mu.Lock()
 	serverURL := a.curProfile.ControlURL
+	a.mu.Unlock()
 	if serverURL == "" {
 		serverURL = "https://vpn.softs.business"
 	}
@@ -497,14 +553,21 @@ func (a *app) addServer() {
 }

 // openAuthURL polls Status.AuthURL and opens the browser when available.
+// Uses bgCtx (not the caller's short operation context) because the auth URL
+// can take longer than 5s to materialise on a cold tailscaled.
 func (a *app) openAuthURL() {
-	for i := 0; i < 20; i++ { // poll up to ~5 seconds
+	deadline := time.Now().Add(30 * time.Second)
+	for time.Now().Before(deadline) {
+		if a.bgCtx.Err() != nil {
+			return
+		}
 		ctx, cancel := context.WithTimeout(a.bgCtx, 3*time.Second)
 		st, err := a.lc.Status(ctx)
 		cancel()
 		if err != nil {
-			log.Printf("openAuthURL: Status error: %v", err)
-			return
+			log.Printf("openAuthURL: Status error (will retry): %v", err)
+			time.Sleep(500 * time.Millisecond)
+			continue
 		}
 		if st.AuthURL != "" {
 			log.Printf("openAuthURL: opening %s", st.AuthURL)
@@ -515,7 +578,7 @@ func (a *app) openAuthURL() {
 			log.Println("openAuthURL: already running, no auth needed")
 			return
 		}
-		time.Sleep(250 * time.Millisecond)
+		time.Sleep(500 * time.Millisecond)
 	}
 	log.Println("openAuthURL: timed out waiting for AuthURL")
 }
@@ -523,11 +586,32 @@ func (a *app) openAuthURL() {
 // ── IPN Bus Watcher ─────────────────────────────────────

 func (a *app) watchIPNBus() {
-	for {
-		err := a.watchIPNBusInner()
-		if err != nil {
-			log.Printf("watchIPNBus error: %v", err)
+	defer func() {
+		if r := recover(); r != nil {
+			log.Printf("PANIC in watchIPNBus (outer): %v", r)
 		}
+	}()
+	for {
+		// Reset prevState before each connection attempt so stale state
+		// from a prior session does not trigger false transitions.
+		a.mu.Lock()
+		a.prevState = ""
+		a.mu.Unlock()
+
+		func() {
+			defer func() {
+				if r := recover(); r != nil {
+					log.Printf("PANIC in watchIPNBusInner: %v", r)
+				}
+			}()
+			if err := a.watchIPNBusInner(); err != nil {
+				log.Printf("watchIPNBus error: %v", err)
+			}
+		}()
+		// After the bus dies, force a status refresh + rebuild so the UI
+		// does not freeze on a stale snapshot while we wait to reconnect.
+		a.updateState()
+		a.triggerRebuild()
 		select {
 		case <-a.bgCtx.Done():
 			return
@@ -553,7 +637,7 @@ func (a *app) watchIPNBusInner() error {
 			return err
 		}

-		// Detect node removed: Running -> NeedsLogin transition
+		// Track state transitions for logging.
 		if n.State != nil {
 			newState := n.State.String()
 			a.mu.Lock()
@@ -561,12 +645,14 @@ func (a *app) watchIPNBusInner() error {
 			a.prevState = newState
 			a.mu.Unlock()

-			if prev == "Running" && newState == "NeedsLogin" {
-				log.Println("Detected state Running->NeedsLogin, performing auto-logout")
-				logoutCtx, cancel := context.WithTimeout(a.bgCtx, 5*time.Second)
-				_ = a.lc.Logout(logoutCtx)
-				cancel()
+			if prev != "" && prev != newState {
+				log.Printf("State transition: %s -> %s", prev, newState)
 			}
+			// NOTE: We intentionally do NOT auto-logout on Running->NeedsLogin.
+			// A NeedsLogin state can be transient (network hiccup, key rotation,
+			// server maintenance). Calling Logout() here would deregister the node
+			// entirely, forcing full re-authentication. Let the UI reflect the state
+			// and allow the user to decide.
 		}

 		if n.State != nil || n.Prefs != nil {
@@ -780,7 +866,10 @@ func buildInputDialogTemplate(title, prompt string) []byte {
 	d.w32(wsPopup | wsCaption | wsSysMenu | dsSetFont | dsModalFrm | ds3DLook)
 	d.w32(0)
 	d.w16(4)
-	d.ws16(0); d.ws16(0); d.ws16(280); d.ws16(90)
+	d.ws16(0)
+	d.ws16(0)
+	d.ws16(280)
+	d.ws16(90)
 	d.w16(0)
 	d.w16(0)
 	d.wstr(title)
@@ -789,31 +878,59 @@ func buildInputDialogTemplate(title, prompt string) []byte {

 	// Static label id=100
 	d.align(4)
-	d.w32(wsChild | wsVisible); d.w32(0)
-	d.ws16(12); d.ws16(12); d.ws16(256); d.ws16(14)
-	d.w16(100); d.w16(0xFFFF); d.w16(0x0082)
-	d.wstr(prompt); d.w16(0)
+	d.w32(wsChild | wsVisible)
+	d.w32(0)
+	d.ws16(12)
+	d.ws16(12)
+	d.ws16(256)
+	d.ws16(14)
+	d.w16(100)
+	d.w16(0xFFFF)
+	d.w16(0x0082)
+	d.wstr(prompt)
+	d.w16(0)

 	// Edit id=101
 	d.align(4)
-	d.w32(wsChild | wsVisible | wsTabStop | wsBorder | esAutoHS); d.w32(0)
-	d.ws16(12); d.ws16(32); d.ws16(256); d.ws16(14)
-	d.w16(101); d.w16(0xFFFF); d.w16(0x0081)
-	d.w16(0); d.w16(0)
+	d.w32(wsChild | wsVisible | wsTabStop | wsBorder | esAutoHS)
+	d.w32(0)
+	d.ws16(12)
+	d.ws16(32)
+	d.ws16(256)
+	d.ws16(14)
+	d.w16(101)
+	d.w16(0xFFFF)
+	d.w16(0x0081)
+	d.w16(0)
+	d.w16(0)

 	// OK id=1
 	d.align(4)
-	d.w32(wsChild | wsVisible | wsTabStop | bsPushBtn); d.w32(0)
-	d.ws16(150); d.ws16(62); d.ws16(55); d.ws16(16)
-	d.w16(1); d.w16(0xFFFF); d.w16(0x0080)
-	d.wstr("Connect"); d.w16(0)
+	d.w32(wsChild | wsVisible | wsTabStop | bsPushBtn)
+	d.w32(0)
+	d.ws16(150)
+	d.ws16(62)
+	d.ws16(55)
+	d.ws16(16)
+	d.w16(1)
+	d.w16(0xFFFF)
+	d.w16(0x0080)
+	d.wstr("Connect")
+	d.w16(0)

 	// Cancel id=2
 	d.align(4)
-	d.w32(wsChild | wsVisible | wsTabStop); d.w32(0)
-	d.ws16(213); d.ws16(62); d.ws16(55); d.ws16(16)
-	d.w16(2); d.w16(0xFFFF); d.w16(0x0080)
-	d.wstr("Cancel"); d.w16(0)
+	d.w32(wsChild | wsVisible | wsTabStop)
+	d.w32(0)
+	d.ws16(213)
+	d.ws16(62)
+	d.ws16(55)
+	d.ws16(16)
+	d.w16(2)
+	d.w16(0xFFFF)
+	d.w16(0x0080)
+	d.wstr("Cancel")
+	d.w16(0)

 	return d.buf
 }