Refactor network monitor to wait for stop (#1992)

client/internal/networkmonitor/monitor.go

@@ -2,14 +2,20 @@ package networkmonitor
 
 import (
 	"context"
+	"errors"
+	"sync"
 )
 
-// NetworkWatcher watches for changes in network configuration.
-type NetworkWatcher struct {
+var ErrStopped = errors.New("monitor has been stopped")
+
+// NetworkMonitor watches for changes in network configuration.
+type NetworkMonitor struct {
 	cancel context.CancelFunc
+	wg     sync.WaitGroup
+	mu     sync.Mutex
 }
 
 // New creates a new network monitor.
-func New() *NetworkWatcher {
-	return &NetworkWatcher{}
+func New() *NetworkMonitor {
+	return &NetworkMonitor{}
 }

client/internal/networkmonitor/monitor_bsd.go

@@ -31,7 +31,7 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 	for {
 		select {
 		case <-ctx.Done():
-			return ctx.Err()
+			return ErrStopped
 		default:
 			buf := make([]byte, 2048)
 			n, err := unix.Read(fd, buf)
@@ -63,7 +63,7 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 				}
 
 				log.Infof("Network monitor: monitored interface (%s) is down.", ifinfo.Name)
-				callback()
+				go callback()
 
 			// handle route changes
 			case unix.RTM_ADD, syscall.RTM_DELETE:
@@ -84,11 +84,11 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 				switch msg.Type {
 				case unix.RTM_ADD:
 					log.Infof("Network monitor: default route changed: via %s, interface %s", route.Gw, intf)
-					callback()
+					go callback()
 				case unix.RTM_DELETE:
 					if intfv4 != nil && route.Gw.Compare(nexthopv4) == 0 || intfv6 != nil && route.Gw.Compare(nexthopv6) == 0 {
 						log.Infof("Network monitor: default route removed: via %s, interface %s", route.Gw, intf)
-						callback()
+						go callback()
 					}
 				}
 			}

client/internal/networkmonitor/monitor_generic.go

@@ -5,6 +5,7 @@ package networkmonitor
 import (
 	"context"
 	"errors"
+	"fmt"
 	"net"
 	"net/netip"
 	"runtime/debug"
@@ -15,20 +16,18 @@ import (
 	"github.com/netbirdio/netbird/client/internal/routemanager"
 )
 
-// Start begins watching for network changes and calls the callback function and stops when a change is detected.
-func (nw *NetworkWatcher) Start(ctx context.Context, callback func()) {
-	if nw.cancel != nil {
-		log.Warn("Network monitor: already running, stopping previous watcher")
-		nw.Stop()
-	}
-
+// Start begins monitoring network changes. When a change is detected, it calls the callback asynchronously and returns.
+func (nw *NetworkMonitor) Start(ctx context.Context, callback func()) (err error) {
 	if ctx.Err() != nil {
-		log.Info("Network monitor: not starting, context is already cancelled")
-		return
+		return ctx.Err()
 	}
 
+	nw.mu.Lock()
 	ctx, nw.cancel = context.WithCancel(ctx)
-	defer nw.Stop()
+	nw.mu.Unlock()
+
+	nw.wg.Add(1)
+	defer nw.wg.Done()
 
 	var nexthop4, nexthop6 netip.Addr
 	var intf4, intf6 *net.Interface
@@ -56,27 +55,30 @@ func (nw *NetworkWatcher) Start(ctx context.Context, callback func()) {
 	expBackOff := backoff.WithContext(backoff.NewExponentialBackOff(), ctx)
 
 	if err := backoff.Retry(operation, expBackOff); err != nil {
-		log.Errorf("Network monitor: failed to get default next hops: %v", err)
-		return
+		return fmt.Errorf("failed to get default next hops: %w", err)
 	}
 
 	// recover in case sys ops panic
 	defer func() {
 		if r := recover(); r != nil {
-			log.Errorf("Network monitor: panic occurred: %v, stack trace: %s", r, string(debug.Stack()))
+			err = fmt.Errorf("panic occurred: %v, stack trace: %s", r, string(debug.Stack()))
 		}
 	}()
 
-	if err := checkChange(ctx, nexthop4, intf4, nexthop6, intf6, callback); err != nil && !errors.Is(err, context.Canceled) {
-		log.Errorf("Network monitor: failed to start: %v", err)
+	if err := checkChange(ctx, nexthop4, intf4, nexthop6, intf6, callback); err != nil {
+		return fmt.Errorf("check change: %w", err)
 	}
+
+	return nil
 }
 
 // Stop stops the network monitor.
-func (nw *NetworkWatcher) Stop() {
+func (nw *NetworkMonitor) Stop() {
+	nw.mu.Lock()
+	defer nw.mu.Unlock()
+
 	if nw.cancel != nil {
 		nw.cancel()
-		nw.cancel = nil
-		log.Info("Network monitor: stopped")
+		nw.wg.Wait()
 	}
 }

client/internal/networkmonitor/monitor_linux.go

@@ -36,7 +36,7 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 	for {
 		select {
 		case <-ctx.Done():
-			return ctx.Err()
+			return ErrStopped
 
 		// handle interface state changes
 		case update := <-linkChan:
@@ -47,12 +47,12 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 			switch update.Header.Type {
 			case syscall.RTM_DELLINK:
 				log.Infof("Network monitor: monitored interface (%s) is gone", update.Link.Attrs().Name)
-				callback()
+				go callback()
 				return nil
 			case syscall.RTM_NEWLINK:
 				if (update.IfInfomsg.Flags&syscall.IFF_RUNNING) == 0 && update.Link.Attrs().OperState == netlink.OperDown {
 					log.Infof("Network monitor: monitored interface (%s) is down.", update.Link.Attrs().Name)
-					callback()
+					go callback()
 					return nil
 				}
 			}
@@ -67,12 +67,12 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 			// triggered on added/replaced routes
 			case syscall.RTM_NEWROUTE:
 				log.Infof("Network monitor: default route changed: via %s, interface %d", route.Gw, route.LinkIndex)
-				callback()
+				go callback()
 				return nil
 			case syscall.RTM_DELROUTE:
 				if intfv4 != nil && route.Gw.Equal(nexthopv4.AsSlice()) || intfv6 != nil && route.Gw.Equal(nexthop6.AsSlice()) {
 					log.Infof("Network monitor: default route removed: via %s, interface %d", route.Gw, route.LinkIndex)
-					callback()
+					go callback()
 					return nil
 				}
 			}

client/internal/networkmonitor/monitor_mobile.go

@@ -4,8 +4,9 @@ package networkmonitor
 
 import "context"
 
-func (nw *NetworkWatcher) Start(context.Context, func()) {
+func (nw *NetworkMonitor) Start(context.Context, func()) error {
+	return nil
 }
 
-func (nw *NetworkWatcher) Stop() {
+func (nw *NetworkMonitor) Stop() {
 }

client/internal/networkmonitor/monitor_windows.go

@@ -48,10 +48,10 @@ func checkChange(ctx context.Context, nexthopv4 netip.Addr, intfv4 *net.Interfac
 	for {
 		select {
 		case <-ctx.Done():
-			return ctx.Err()
+			return ErrStopped
 		case <-ticker.C:
 			if changed(nexthopv4, intfv4, neighborv4, nexthopv6, intfv6, neighborv6) {
-				callback()
+				go callback()
 				return nil
 			}
 		}