fix: harden terminal WebSocket edge cases

- Wrap tryRefreshSession in try-finally to ensure authRecoveryInFlight is always reset
- Fix variable scope issue: use const tabState instead of reassigning d
- Add error handling for process.stdin.write() to catch BrokenPipeError/OSError
- Verify WebSocket instance before closing in ping timeout callback
- Properly await conn.wait_closed() to prevent SSH connection leaks
- Gracefully cancel and wait for read_task with 2s timeout

These fixes prevent race conditions, resource leaks, and improve error recovery.

dashboard/backend/routers/terminal.py

@@ -190,18 +190,31 @@ async def ws_endpoint(websocket: WebSocket, host: str = Query("nas")):
                         rows = struct.unpack("!H", data[3:5])[0]
                         process.channel.change_terminal_size(cols, rows)
                     else:
+                        try:
                             process.stdin.write(data)
+                        except (BrokenPipeError, OSError) as e:
+                            logger.warning("Failed to write to stdin for %s: %s", host, e)
+                            break
                 elif "text" in msg and msg["text"]:
                     if msg["text"] == "__ping__":
                         logger.debug("Received ping from %s, sending pong", host)
                         await websocket.send_text("__pong__")
                         continue
+                    try:
                         process.stdin.write(msg["text"].encode())
+                    except (BrokenPipeError, OSError) as e:
+                        logger.warning("Failed to write to stdin for %s: %s", host, e)
+                        break
         except Exception as e:
             logger.warning("WebSocket receive loop ended for %s: %s", host, e)
         finally:
             read_task.cancel()
+            try:
+                await asyncio.wait_for(read_task, timeout=2.0)
+            except (asyncio.TimeoutError, asyncio.CancelledError):
+                pass
             process.close()
     finally:
         conn.close()
+        await conn.wait_closed()
         await _release_session(session_id)

dashboard/frontend/src/routes/Terminal.svelte

@@ -300,7 +300,9 @@
               ws.send("__ping__");
               // Set timeout to detect missing pong
               pongTimeout = setTimeout(() => {
-                if (ws.readyState === 1) {
+                // Check if still the same WebSocket instance and still open
+                const currentTab = tabData.get(tabId);
+                if (currentTab?.ws === ws && ws.readyState === 1) {
                   ws.close(1000, "keepalive ping timeout");
                 }
               }, PONG_TIMEOUT);
@@ -343,8 +345,8 @@
         };
         ws.onclose = async (e) => {
           const reason = e.reason || `Connection closed (code ${e.code})`;
-          d = tabData.get(tabId);
+          const tabState = tabData.get(tabId);
-          if (d) d.ws = null;
+          if (tabState) tabState.ws = null;
 
           // Handle auth failures (code 1008)
           if (e.code === 1008) {
@@ -359,8 +361,8 @@
             // Try to refresh session
             if (!closedManually && !authRecoveryInFlight) {
               authRecoveryInFlight = true;
+              try {
                 const refreshed = await tryRefreshSession();
-              authRecoveryInFlight = false;
                 if (refreshed && !closedManually) {
                   reconnecting = true;
                   updateTab(tabId, {
@@ -372,6 +374,11 @@
                   void connect(false);
                   return;
                 }
+              } catch (err) {
+                console.error("Session refresh failed:", err);
+              } finally {
+                authRecoveryInFlight = false;
+              }
             }
             handleAuthExpired("Session expired — please log in again");
             return;
@@ -386,8 +393,8 @@
           }
         };
         ws.onerror = () => {
-          d = tabData.get(tabId);
+          const tabState = tabData.get(tabId);
-          if (d) d.ws = null;
+          if (tabState) tabState.ws = null;
           if (!closedManually) scheduleReconnect("Connection failed");
           else updateTab(tabId, { connected: false, error: "Connection failed" });
         };