Wait on daemon down (#2279)

client/cmd/down.go

@@ -26,7 +26,7 @@ var downCmd = &cobra.Command{
 			return err
 		}
 
-		ctx, cancel := context.WithTimeout(context.Background(), time.Second*3)
+		ctx, cancel := context.WithTimeout(context.Background(), time.Second*7)
 		defer cancel()
 
 		conn, err := DialClientGRPCServer(ctx, daemonAddr)

client/internal/engine.go

@@ -266,8 +266,23 @@ func (e *Engine) Stop() error {
 
 	e.close()
 	e.wgConnWorker.Wait()
-	log.Infof("stopped Netbird Engine")
-	return nil
+
+	maxWaitTime := 5 * time.Second
+	timeout := time.After(maxWaitTime)
+
+	for {
+		if !e.IsWGIfaceUp() {
+			log.Infof("stopped Netbird Engine")
+			return nil
+		}
+
+		select {
+		case <-timeout:
+			return fmt.Errorf("timeout when waiting for interface shutdown")
+		default:
+			time.Sleep(100 * time.Millisecond)
+		}
+	}
 }
 
 // Start creates a new WireGuard tunnel interface and listens to events from Signal and Management services
@@ -1533,3 +1548,20 @@ func isChecksEqual(checks []*mgmProto.Checks, oChecks []*mgmProto.Checks) bool {
 		return slices.Equal(checks.Files, oChecks.Files)
 	})
 }
+
+func (e *Engine) IsWGIfaceUp() bool {
+	if e == nil || e.wgInterface == nil {
+		return false
+	}
+	iface, err := net.InterfaceByName(e.wgInterface.Name())
+	if err != nil {
+		log.Debugf("failed to get interface by name %s: %v", e.wgInterface.Name(), err)
+		return false
+	}
+
+	if iface.Flags&net.FlagUp != 0 {
+		return true
+	}
+
+	return false
+}

client/internal/networkmonitor/monitor_bsd.go

@@ -4,6 +4,7 @@ package networkmonitor
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"syscall"
 	"unsafe"
@@ -21,11 +22,20 @@ func checkChange(ctx context.Context, nexthopv4, nexthopv6 systemops.Nexthop, ca
 		return fmt.Errorf("failed to open routing socket: %v", err)
 	}
 	defer func() {
-		if err := unix.Close(fd); err != nil {
+		err := unix.Close(fd)
+		if err != nil && !errors.Is(err, unix.EBADF) {
 			log.Errorf("Network monitor: failed to close routing socket: %v", err)
 		}
 	}()
 
+	go func() {
+		<-ctx.Done()
+		err := unix.Close(fd)
+		if err != nil && !errors.Is(err, unix.EBADF) {
+			log.Debugf("Network monitor: closed routing socket")
+		}
+	}()
+
 	for {
 		select {
 		case <-ctx.Done():
@@ -34,7 +44,9 @@ func checkChange(ctx context.Context, nexthopv4, nexthopv6 systemops.Nexthop, ca
 			buf := make([]byte, 2048)
 			n, err := unix.Read(fd, buf)
 			if err != nil {
-				log.Errorf("Network monitor: failed to read from routing socket: %v", err)
+				if !errors.Is(err, unix.EBADF) && !errors.Is(err, unix.EINVAL) {
+					log.Errorf("Network monitor: failed to read from routing socket: %v", err)
+				}
 				continue
 			}
 			if n < unix.SizeofRtMsghdr {

client/server/server.go

@@ -582,7 +582,7 @@ func (s *Server) Up(callerCtx context.Context, _ *proto.UpRequest) (*proto.UpRes
 }
 
 // Down engine work in the daemon.
-func (s *Server) Down(_ context.Context, _ *proto.DownRequest) (*proto.DownResponse, error) {
+func (s *Server) Down(ctx context.Context, _ *proto.DownRequest) (*proto.DownResponse, error) {
 	s.mutex.Lock()
 	defer s.mutex.Unlock()
 
@@ -593,7 +593,25 @@ func (s *Server) Down(_ context.Context, _ *proto.DownRequest) (*proto.DownRespo
 	state := internal.CtxGetState(s.rootCtx)
 	state.Set(internal.StatusIdle)
 
-	return &proto.DownResponse{}, nil
+	maxWaitTime := 5 * time.Second
+	timeout := time.After(maxWaitTime)
+
+	engine := s.connectClient.Engine()
+
+	for {
+		if !engine.IsWGIfaceUp() {
+			return &proto.DownResponse{}, nil
+		}
+
+		select {
+		case <-ctx.Done():
+			return &proto.DownResponse{}, nil
+		case <-timeout:
+			return nil, fmt.Errorf("failed to shut down properly")
+		default:
+			time.Sleep(100 * time.Millisecond)
+		}
+	}
 }
 
 // Status returns the daemon status