Netlink timeouts

drivers/overlay/ov_network.go

@@ -696,6 +696,12 @@ func (n *network) initSandbox(restore bool) error {
 	var nlSock *nl.NetlinkSocket
 	sbox.InvokeFunc(func() {
 		nlSock, err = nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_NEIGH)
+		if err != nil {
+			return
+		}
+		// set the receive timeout to not remain stuck on the RecvFrom if the fd gets closed
+		tv := syscall.NsecToTimeval(soTimeout.Nanoseconds())
+		err = nlSock.SetReceiveTimeout(&tv)
 	})
 	n.setNetlinkSocket(nlSock)
 
@@ -721,6 +727,11 @@ func (n *network) watchMiss(nlSock *nl.NetlinkSocket) {
 				// The netlink socket got closed, simply exit to not leak this goroutine
 				return
 			}
+			// When the receive timeout expires the receive will return EAGAIN
+			if err == syscall.EAGAIN {
+				// we continue here to avoid spam for timeouts
+				continue
+			}
 			logrus.Errorf("Failed to receive from netlink: %v ", err)
 			continue
 		}

drivers/overlay/overlay_test.go

@@ -1,7 +1,9 @@
 package overlay
 
 import (
+	"context"
 	"net"
+	"syscall"
 	"testing"
 	"time"
 
@@ -12,6 +14,7 @@ import (
 	"github.com/docker/libnetwork/driverapi"
 	"github.com/docker/libnetwork/netlabel"
 	_ "github.com/docker/libnetwork/testutils"
+	"github.com/vishvananda/netlink/nl"
 )
 
 func init() {
@@ -135,3 +138,36 @@ func TestOverlayType(t *testing.T) {
 			dt.d.Type())
 	}
 }
+
+// Test that the netlink socket close unblock the watchMiss to avoid deadlock
+func TestNetlinkSocket(t *testing.T) {
+	// This is the same code used by the overlay driver to create the netlink interface
+	// for the watch miss
+	nlSock, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_NEIGH)
+	if err != nil {
+		t.Fatal()
+	}
+	// set the receive timeout to not remain stuck on the RecvFrom if the fd gets closed
+	tv := syscall.NsecToTimeval(soTimeout.Nanoseconds())
+	err = nlSock.SetReceiveTimeout(&tv)
+	if err != nil {
+		t.Fatal()
+	}
+	n := &network{id: "testnetid"}
+	ch := make(chan error)
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	go func() {
+		n.watchMiss(nlSock)
+		ch <- nil
+	}()
+	time.Sleep(5 * time.Second)
+	nlSock.Close()
+	select {
+	case <-ch:
+	case <-ctx.Done():
+		{
+			t.Fatalf("Timeout expired")
+		}
+	}
+}

ipvs/ipvs.go

@@ -5,12 +5,19 @@ package ipvs
 import (
 	"net"
 	"syscall"
+	"time"
 
 	"fmt"
+
 	"github.com/vishvananda/netlink/nl"
 	"github.com/vishvananda/netns"
 )
 
+const (
+	netlinkRecvSocketsTimeout = 3 * time.Second
+	netlinkSendSocketTimeout  = 30 * time.Second
+)
+
 // Service defines an IPVS service in its entirety.
 type Service struct {
 	// Virtual service address.
@@ -82,6 +89,15 @@ func New(path string) (*Handle, error) {
 	if err != nil {
 		return nil, err
 	}
+	// Add operation timeout to avoid deadlocks
+	tv := syscall.NsecToTimeval(netlinkSendSocketTimeout.Nanoseconds())
+	if err := sock.SetSendTimeout(&tv); err != nil {
+		return nil, err
+	}
+	tv = syscall.NsecToTimeval(netlinkRecvSocketsTimeout.Nanoseconds())
+	if err := sock.SetReceiveTimeout(&tv); err != nil {
+		return nil, err
+	}
 
 	return &Handle{sock: sock}, nil
 }

ipvs/netlink.go

@@ -203,10 +203,6 @@ func newGenlRequest(familyID int, cmd uint8) *nl.NetlinkRequest {
 }
 
 func execute(s *nl.NetlinkSocket, req *nl.NetlinkRequest, resType uint16) ([][]byte, error) {
-	var (
-		err error
-	)
-
 	if err := s.Send(req); err != nil {
 		return nil, err
 	}
@@ -222,6 +218,13 @@ done:
 	for {
 		msgs, err := s.Receive()
 		if err != nil {
+			if s.GetFd() == -1 {
+				return nil, fmt.Errorf("Socket got closed on receive")
+			}
+			if err == syscall.EAGAIN {
+				// timeout fired
+				continue
+			}
 			return nil, err
 		}
 		for _, m := range msgs {

vendor.conf

@@ -45,7 +45,7 @@ github.com/sirupsen/logrus v1.0.3
 github.com/stretchr/testify dab07ac62d4905d3e48d17dc549c684ac3b7c15a
 github.com/syndtr/gocapability db04d3cc01c8b54962a58ec7e491717d06cfcc16
 github.com/ugorji/go f1f1a805ed361a0e078bb537e4ea78cd37dcf065
-github.com/vishvananda/netlink bd6d5de5ccef2d66b0a26177928d0d8895d7f969
+github.com/vishvananda/netlink b2de5d10e38ecce8607e6b438b6d174f389a004e
 github.com/vishvananda/netns 604eaf189ee867d8c147fafc28def2394e878d25
 golang.org/x/crypto 558b6879de74bc843225cde5686419267ff707ca
 golang.org/x/net 7dcfb8076726a3fdd9353b6b8a1f1b6be6811bd6