ipn/ipnserver: on Windows in unattended mode, wait for Engine forever

Updates #1187

ipn/ipnserver/server.go

@@ -503,12 +503,80 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 	}()
 	logf("Listening on %v", listen.Addr())
 
+	var store ipn.StateStore
+	if opts.StatePath != "" {
+		store, err = ipn.NewFileStore(opts.StatePath)
+		if err != nil {
+			return fmt.Errorf("ipn.NewFileStore(%q): %v", opts.StatePath, err)
+		}
+		if opts.AutostartStateKey == "" {
+			autoStartKey, err := store.ReadState(ipn.ServerModeStartKey)
+			if err != nil && err != ipn.ErrStateNotExist {
+				return fmt.Errorf("calling ReadState on %s: %w", opts.StatePath, err)
+			}
+			key := string(autoStartKey)
+			if strings.HasPrefix(key, "user-") {
+				uid := strings.TrimPrefix(key, "user-")
+				u, err := server.lookupUserFromID(uid)
+				if err != nil {
+					logf("ipnserver: found server mode auto-start key %q; failed to load user: %v", key, err)
+				} else {
+					logf("ipnserver: found server mode auto-start key %q (user %s)", key, u.Username)
+					server.serverModeUser = u
+				}
+				opts.AutostartStateKey = ipn.StateKey(key)
+			}
+		}
+	} else {
+		store = &ipn.MemoryStore{}
+	}
+
 	bo := backoff.NewBackoff("ipnserver", logf, 30*time.Second)
 	var unservedConn net.Conn // if non-nil, accepted, but hasn't served yet
 
 	eng, err := getEngine()
 	if err != nil {
 		logf("ipnserver: initial getEngine call: %v", err)
+
+		// Issue 1187: on Windows, in unattended mode,
+		// sometimes we try 5 times and fail to create the
+		// engine before the system's ready. Hack until the
+		// bug if fixed properly: if we're running in
+		// unattended mode on Windows, keep trying forever,
+		// waiting for the machine to be ready (networking to
+		// come up?) and then dial our own safesocket TCP
+		// listener to wake up the usual mechanism that lets
+		// us surface getEngine errors to UI clients. (We
+		// don't want to just call getEngine in a loop without
+		// the listener.Accept, as we do want to handle client
+		// connections so we can tell them about errors)
+
+		bootRaceWaitForEngine, bootRaceWaitForEngineCancel := context.WithTimeout(context.Background(), time.Minute)
+		if runtime.GOOS == "windows" && opts.AutostartStateKey != "" {
+			logf("ipnserver: in unattended mode, waiting for engine availability")
+			getEngine = getEngineUntilItWorksWrapper(getEngine)
+			// Wait for it to be ready.
+			go func() {
+				defer bootRaceWaitForEngineCancel()
+				t0 := time.Now()
+				for {
+					time.Sleep(10 * time.Second)
+					if _, err := getEngine(); err != nil {
+						logf("ipnserver: unattended mode engine load: %v", err)
+						continue
+					}
+					c, err := net.Dial("tcp", listen.Addr().String())
+					logf("ipnserver: engine created after %v; waking up Accept: Dial error: %v", time.Since(t0).Round(time.Second), err)
+					if err == nil {
+						c.Close()
+					}
+					break
+				}
+			}()
+		} else {
+			bootRaceWaitForEngineCancel()
+		}
+
 		for i := 1; ctx.Err() == nil; i++ {
 			c, err := listen.Accept()
 			if err != nil {
@@ -516,6 +584,7 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 				bo.BackOff(ctx, err)
 				continue
 			}
+			<-bootRaceWaitForEngine.Done()
 			logf("ipnserver: try%d: trying getEngine again...", i)
 			eng, err = getEngine()
 			if err == nil {
@@ -538,34 +607,6 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 		}
 	}
 
-	var store ipn.StateStore
-	if opts.StatePath != "" {
-		store, err = ipn.NewFileStore(opts.StatePath)
-		if err != nil {
-			return fmt.Errorf("ipn.NewFileStore(%q): %v", opts.StatePath, err)
-		}
-		if opts.AutostartStateKey == "" {
-			autoStartKey, err := store.ReadState(ipn.ServerModeStartKey)
-			if err != nil && err != ipn.ErrStateNotExist {
-				return fmt.Errorf("calling ReadState on %s: %w", opts.StatePath, err)
-			}
-			key := string(autoStartKey)
-			if strings.HasPrefix(key, "user-") {
-				uid := strings.TrimPrefix(key, "user-")
-				u, err := server.lookupUserFromID(uid)
-				if err != nil {
-					logf("ipnserver: found server mode auto-start key %q; failed to load user: %v", key, err)
-				} else {
-					logf("ipnserver: found server mode auto-start key %q (user %s)", key, u.Username)
-					server.serverModeUser = u
-				}
-				opts.AutostartStateKey = ipn.StateKey(key)
-			}
-		}
-	} else {
-		store = &ipn.MemoryStore{}
-	}
-
 	b, err := ipn.NewLocalBackend(logf, logid, store, eng)
 	if err != nil {
 		return fmt.Errorf("NewLocalBackend: %v", err)
@@ -756,6 +797,27 @@ func FixedEngine(eng wgengine.Engine) func() (wgengine.Engine, error) {
 	return func() (wgengine.Engine, error) { return eng, nil }
 }
 
+// getEngineUntilItWorksWrapper returns a getEngine wrapper that does
+// not call getEngine concurrently and stops calling getEngine once
+// it's returned a working engine.
+func getEngineUntilItWorksWrapper(getEngine func() (wgengine.Engine, error)) func() (wgengine.Engine, error) {
+	var mu sync.Mutex
+	var engGood wgengine.Engine
+	return func() (wgengine.Engine, error) {
+		mu.Lock()
+		defer mu.Unlock()
+		if engGood != nil {
+			return engGood, nil
+		}
+		e, err := getEngine()
+		if err != nil {
+			return nil, err
+		}
+		engGood = e
+		return e, nil
+	}
+}
+
 type dummyAddr string
 type oneConnListener struct {
 	conn net.Conn