diff --git a/pkg/server/sandbox_portforward.go b/pkg/server/sandbox_portforward.go
index 7106cb673..e544dadb3 100644
--- a/pkg/server/sandbox_portforward.go
+++ b/pkg/server/sandbox_portforward.go
@@ -23,7 +23,6 @@ import (
 	"os/exec"
 	"strings"
 
-	"github.com/containernetworking/plugins/pkg/ns"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 	"golang.org/x/net/context"
@@ -45,31 +44,28 @@ func (c *criService) PortForward(ctx context.Context, r *runtime.PortForwardRequ
 	return c.streamServer.GetPortForward(r)
 }
 
-// portForward requires `socat` on the node. It uses netns to enter the sandbox namespace,
-// and run `socat` insidethe namespace to forward stream for a specific port. The `socat`
-// command keeps running until it exits or client disconnect.
+// portForward requires `socat` on the node. It runs `socat` to forward stream for a
+// specific port. The `socat` command keeps running until it exits or client disconnect.
 func (c *criService) portForward(id string, port int32, stream io.ReadWriteCloser) error {
 	s, err := c.sandboxStore.Get(id)
 	if err != nil {
 		return errors.Wrapf(err, "failed to find sandbox %q in store", id)
 	}
-	var netNSDo func(func(ns.NetNS) error) error
-	// netNSPath is the network namespace path for logging.
-	var netNSPath string
+	if s.Status.Get().State != sandboxstore.StateReady {
+		return errors.Wrap(err, "sandbox container is not running")
+	}
+	// We use pod ip instead of netns.Do + localhost. One reason is that
+	// for sandbox container e.g. gvisor and kata, the network stack is
+	// inside sandbox, netns.Do + localhost will not work. However, accessing
+	// pod ip from host should always work, because it is an assumption of
+	// Kubernetes network model. (Related issue containerd/cri-containerd#524)
+	var addr string
 	securityContext := s.Config.GetLinux().GetSecurityContext()
 	hostNet := securityContext.GetNamespaceOptions().GetNetwork() == runtime.NamespaceMode_NODE
 	if !hostNet {
-		if s.NetNS == nil || s.NetNS.Closed() {
-			return errors.Errorf("network namespace for sandbox %q is closed", id)
-		}
-		netNSDo = s.NetNS.GetNs().Do
-		netNSPath = s.NetNS.GetPath()
+		addr = s.IP
 	} else {
-		// Run the function directly for host network.
-		netNSDo = func(do func(_ ns.NetNS) error) error {
-			return do(nil)
-		}
-		netNSPath = "host"
+		addr = "localhost"
 	}
 
 	socat, err := exec.LookPath("socat")
@@ -78,46 +74,39 @@ func (c *criService) portForward(id string, port int32, stream io.ReadWriteClose
 	}
 
 	// Check https://linux.die.net/man/1/socat for meaning of the options.
-	args := []string{socat, "-", fmt.Sprintf("TCP4:localhost:%d", port)}
+	args := []string{socat, "-", fmt.Sprintf("TCP4:%s:%d", addr, port)}
 
-	logrus.Infof("Executing port forwarding command %q in network namespace %q", strings.Join(args, " "), netNSPath)
-	err = netNSDo(func(_ ns.NetNS) error {
-		cmd := exec.Command(args[0], args[1:]...)
-		cmd.Stdout = stream
+	logrus.Infof("Executing port forwarding command %q", strings.Join(args, " "))
+	cmd := exec.Command(args[0], args[1:]...)
+	cmd.Stdout = stream
 
-		stderr := new(bytes.Buffer)
-		cmd.Stderr = stderr
+	stderr := new(bytes.Buffer)
+	cmd.Stderr = stderr
 
-		// If we use Stdin, command.Run() won't return until the goroutine that's copying
-		// from stream finishes. Unfortunately, if you have a client like telnet connected
-		// via port forwarding, as long as the user's telnet client is connected to the user's
-		// local listener that port forwarding sets up, the telnet session never exits. This
-		// means that even if socat has finished running, command.Run() won't ever return
-		// (because the client still has the connection and stream open).
-		//
-		// The work around is to use StdinPipe(), as Wait() (called by Run()) closes the pipe
-		// when the command (socat) exits.
-		in, err := cmd.StdinPipe()
-		if err != nil {
-			return errors.Wrap(err, "failed to create stdin pipe")
-		}
-		go func() {
-			if _, err := io.Copy(in, stream); err != nil {
-				logrus.WithError(err).Errorf("Failed to copy port forward input for %q port %d", id, port)
-			}
-			in.Close()
-			logrus.Debugf("Finish copying port forward input for %q port %d", id, port)
-		}()
-
-		if err := cmd.Run(); err != nil {
-			return errors.Errorf("socat command returns error: %v, stderr: %q", err, stderr.String())
-		}
-		return nil
-	})
+	// If we use Stdin, command.Run() won't return until the goroutine that's copying
+	// from stream finishes. Unfortunately, if you have a client like telnet connected
+	// via port forwarding, as long as the user's telnet client is connected to the user's
+	// local listener that port forwarding sets up, the telnet session never exits. This
+	// means that even if socat has finished running, command.Run() won't ever return
+	// (because the client still has the connection and stream open).
+	//
+	// The work around is to use StdinPipe(), as Wait() (called by Run()) closes the pipe
+	// when the command (socat) exits.
+	in, err := cmd.StdinPipe()
 	if err != nil {
-		return errors.Wrapf(err, "failed to execute portforward in network namespace %q", netNSPath)
+		return errors.Wrap(err, "failed to create stdin pipe")
+	}
+	go func() {
+		if _, err := io.Copy(in, stream); err != nil {
+			logrus.WithError(err).Errorf("Failed to copy port forward input for %q port %d", id, port)
+		}
+		in.Close()
+		logrus.Debugf("Finish copying port forward input for %q port %d", id, port)
+	}()
+
+	if err := cmd.Run(); err != nil {
+		return errors.Errorf("socat command returns error: %v, stderr: %q", err, stderr.String())
 	}
 	logrus.Infof("Finish port forwarding for %q port %d", id, port)
-
 	return nil
 }