vnet: add exponential backoff for failed reconnections (#5035)

pkg/plugin/visitor/plugin.go

@@ -23,11 +23,20 @@ import (
 	"github.com/fatedier/frp/pkg/vnet"
 )
 
+// PluginContext provides the necessary context and callbacks for visitor plugins.
 type PluginContext struct {
-	Name           string
-	Ctx            context.Context
+	// Name is the unique identifier for this visitor, used for logging and routing.
+	Name string
+
+	// Ctx manages the plugin's lifecycle and carries the logger for structured logging.
+	Ctx context.Context
+
+	// VnetController manages TUN device routing. May be nil if virtual networking is disabled.
 	VnetController *vnet.Controller
-	HandleConn     func(net.Conn)
+
+	// SendConnToVisitor sends a connection to the visitor's internal processing queue.
+	// Does not return error; failures are handled by closing the connection.
+	SendConnToVisitor func(net.Conn)
 }
 
 // Creators is used for create plugins to handle connections.

pkg/plugin/visitor/virtual_net.go

@@ -42,6 +42,8 @@ type VirtualNetPlugin struct {
 	controllerConn net.Conn
 	closeSignal    chan struct{}
 
+	consecutiveErrors int // Tracks consecutive connection errors for exponential backoff
+
 	ctx    context.Context
 	cancel context.CancelFunc
 }
@@ -98,7 +100,6 @@ func (p *VirtualNetPlugin) Start() {
 
 func (p *VirtualNetPlugin) run() {
 	xl := xlog.FromContextSafe(p.ctx)
-	reconnectDelay := 10 * time.Second
 
 	for {
 		currentCloseSignal := make(chan struct{})
@@ -121,7 +122,10 @@ func (p *VirtualNetPlugin) run() {
 		p.controllerConn = controllerConn
 		p.mu.Unlock()
 
-		pluginNotifyConn := netutil.WrapCloseNotifyConn(pluginConn, func() {
+		// Wrap with CloseNotifyConn which supports both close notification and error recording
+		var closeErr error
+		pluginNotifyConn := netutil.WrapCloseNotifyConn(pluginConn, func(err error) {
+			closeErr = err
 			close(currentCloseSignal) // Signal the run loop on close.
 		})
 
@@ -129,9 +133,9 @@ func (p *VirtualNetPlugin) run() {
 		p.pluginCtx.VnetController.RegisterClientRoute(p.ctx, p.pluginCtx.Name, p.routes, controllerConn)
 		xl.Infof("successfully registered client route for visitor [%s]. Starting connection handler with CloseNotifyConn.", p.pluginCtx.Name)
 
-		// Pass the CloseNotifyConn to HandleConn.
-		// HandleConn is responsible for calling Close() on pluginNotifyConn.
-		p.pluginCtx.HandleConn(pluginNotifyConn)
+		// Pass the CloseNotifyConn to the visitor for handling.
+		// The visitor can call CloseWithError to record the failure reason.
+		p.pluginCtx.SendConnToVisitor(pluginNotifyConn)
 
 		// Wait for context cancellation or connection close.
 		select {
@@ -140,8 +144,32 @@ func (p *VirtualNetPlugin) run() {
 			p.cleanupControllerConn(xl)
 			return
 		case <-currentCloseSignal:
-			xl.Infof("detected connection closed via CloseNotifyConn for visitor [%s].", p.pluginCtx.Name)
-			// HandleConn closed the plugin side. Close the controller side.
+			// Determine reconnect delay based on error with exponential backoff
+			var reconnectDelay time.Duration
+			if closeErr != nil {
+				p.consecutiveErrors++
+				xl.Warnf("connection closed with error for visitor [%s] (consecutive errors: %d): %v",
+					p.pluginCtx.Name, p.consecutiveErrors, closeErr)
+
+				// Exponential backoff: 60s, 120s, 240s, 300s (capped)
+				baseDelay := 60 * time.Second
+				reconnectDelay = baseDelay * time.Duration(1<<uint(p.consecutiveErrors-1))
+				if reconnectDelay > 300*time.Second {
+					reconnectDelay = 300 * time.Second
+				}
+			} else {
+				// Reset consecutive errors on successful connection
+				if p.consecutiveErrors > 0 {
+					xl.Infof("connection closed normally for visitor [%s], resetting error counter (was %d)",
+						p.pluginCtx.Name, p.consecutiveErrors)
+					p.consecutiveErrors = 0
+				} else {
+					xl.Infof("connection closed normally for visitor [%s]", p.pluginCtx.Name)
+				}
+				reconnectDelay = 10 * time.Second
+			}
+
+			// The visitor closed the plugin side. Close the controller side.
 			p.cleanupControllerConn(xl)
 
 			xl.Infof("waiting %v before attempting reconnection for visitor [%s]...", reconnectDelay, p.pluginCtx.Name)
@@ -184,7 +212,7 @@ func (p *VirtualNetPlugin) Close() error {
 	}
 
 	// Explicitly close the controller side of the pipe.
-	// This ensures the pipe is broken even if the run loop is stuck or HandleConn hasn't closed its end.
+	// This ensures the pipe is broken even if the run loop is stuck or the visitor hasn't closed its end.
 	p.cleanupControllerConn(xl)
 	xl.Infof("finished cleaning up connections during close for visitor [%s]", p.pluginCtx.Name)
 

pkg/util/net/conn.go

@@ -135,11 +135,11 @@ type CloseNotifyConn struct {
 	// 1 means closed
 	closeFlag int32
 
-	closeFn func()
+	closeFn func(error)
 }
 
-// closeFn will be only called once
-func WrapCloseNotifyConn(c net.Conn, closeFn func()) net.Conn {
+// closeFn will be only called once with the error (nil if Close() was called, non-nil if CloseWithError() was called)
+func WrapCloseNotifyConn(c net.Conn, closeFn func(error)) *CloseNotifyConn {
 	return &CloseNotifyConn{
 		Conn:    c,
 		closeFn: closeFn,
@@ -151,12 +151,25 @@ func (cc *CloseNotifyConn) Close() (err error) {
 	if pflag == 0 {
 		err = cc.Conn.Close()
 		if cc.closeFn != nil {
-			cc.closeFn()
+			cc.closeFn(nil)
 		}
 	}
 	return
 }
 
+// CloseWithError closes the connection and passes the error to the close callback.
+func (cc *CloseNotifyConn) CloseWithError(err error) error {
+	pflag := atomic.SwapInt32(&cc.closeFlag, 1)
+	if pflag == 0 {
+		closeErr := cc.Conn.Close()
+		if cc.closeFn != nil {
+			cc.closeFn(err)
+		}
+		return closeErr
+	}
+	return nil
+}
+
 type StatsConn struct {
 	net.Conn
 

pkg/util/net/websocket.go

@@ -32,7 +32,7 @@ func NewWebsocketListener(ln net.Listener) (wl *WebsocketListener) {
 	muxer := http.NewServeMux()
 	muxer.Handle(FrpWebsocketPath, websocket.Handler(func(c *websocket.Conn) {
 		notifyCh := make(chan struct{})
-		conn := WrapCloseNotifyConn(c, func() {
+		conn := WrapCloseNotifyConn(c, func(_ error) {
 			close(notifyCh)
 		})
 		wl.acceptCh <- conn