diff --git a/pkg/bypass4netns/bypass4netns.go b/pkg/bypass4netns/bypass4netns.go
index 44dea5a..9e42b40 100644
--- a/pkg/bypass4netns/bypass4netns.go
+++ b/pkg/bypass4netns/bypass4netns.go
@@ -417,7 +417,19 @@ func (h *notifHandler) registerSocket(pid int, sockfd int, syscallName string) (
 		sock.state = NotBypassable
 		logger.Debugf("failed to get socket args err=%q", err)
 	} else {
-		if sockDomain != syscall.AF_INET && sockDomain != syscall.AF_INET6 {
+		// compare process's netns is same or not with container init process's netns
+		isSameNetNS, err := util.SameNetNS(h.containerInitPid, pid)
+		if err != nil {
+			logger.Errorf("failed to check NetNS: err=%q", err)
+			sock.state = NotBypassable
+		}
+
+		// check the process is executed with container's netns.
+		// processes with nested netns are not handled by bypass4netns
+		if !isSameNetNS {
+			logger.Infof("process seems to be executed in other netns. socket is NotBypassable and ignored")
+			sock.state = NotBypassable
+		} else if sockDomain != syscall.AF_INET && sockDomain != syscall.AF_INET6 {
 			// non IP sockets are not handled.
 			sock.state = NotBypassable
 			logger.Debugf("socket domain=0x%x", sockDomain)
@@ -629,6 +641,8 @@ type Handler struct {
 
 	// key is child port
 	forwardingPorts map[int]ForwardPortMapping
+
+	ContainerInitPid int
 }
 
 // NewHandler creates new seccomp notif handler
@@ -640,6 +654,7 @@ func NewHandler(socketPath, comSocketPath, tracerAgentLogPath string) *Handler {
 		ignoredSubnets:     []net.IPNet{},
 		forwardingPorts:    map[int]ForwardPortMapping{},
 		readyFd:            -1,
+		ContainerInitPid:   -1,
 	}
 
 	return &handler
@@ -711,6 +726,10 @@ type notifHandler struct {
 
 	// cache pidfd to reduce latency. key is pid.
 	pidInfos map[int]pidInfo
+
+	// container init process's pid
+	// used to check whether netns is container or not.
+	containerInitPid int
 }
 
 type containerInterface struct {
@@ -732,14 +751,15 @@ type pidInfo struct {
 	tgid    int
 }
 
-func (h *Handler) newNotifHandler(fd uintptr, state *specs.ContainerProcessState) *notifHandler {
+func (h *Handler) newNotifHandler(fd uintptr, state *specs.ContainerProcessState, containerInitPid int) *notifHandler {
 	notifHandler := notifHandler{
-		fd:              libseccomp.ScmpFd(fd),
-		state:           state,
-		forwardingPorts: map[int]ForwardPortMapping{},
-		processes:       map[int]*processStatus{},
-		memfds:          map[int]int{},
-		pidInfos:        map[int]pidInfo{},
+		fd:               libseccomp.ScmpFd(fd),
+		state:            state,
+		forwardingPorts:  map[int]ForwardPortMapping{},
+		processes:        map[int]*processStatus{},
+		memfds:           map[int]int{},
+		pidInfos:         map[int]pidInfo{},
+		containerInitPid: containerInitPid,
 	}
 	notifHandler.nonBypassable = nonbypassable.New(h.ignoredSubnets)
 	notifHandler.nonBypassableAutoUpdate = h.ignoredSubnetsAutoUpdate
@@ -793,8 +813,18 @@ func (h *Handler) StartHandle(c2cConfig *C2CConnectionHandleConfig, multinodeCon
 			continue
 		}
 
+		// this may not work correctly when executed process with 'ip netns exec'.
+		// state.Pid can be the process in nested netns.
+		// so, we cannot distinguish container netns and nested netns with simply comparing state.Pid and hooked process's pid
+		// Instead of state.Pid, init process's pid should be used.
+		// bypass4netns recognizes the first process as a init process.
+		if h.ContainerInitPid < 0 {
+			h.ContainerInitPid = state.Pid
+			logrus.Infof("ContainerInitPid is %d", h.ContainerInitPid)
+		}
+
 		logrus.Infof("Received new seccomp fd: %v", newFd)
-		notifHandler := h.newNotifHandler(newFd, state)
+		notifHandler := h.newNotifHandler(newFd, state, h.ContainerInitPid)
 		notifHandler.c2cConnections = c2cConfig
 		notifHandler.multinode = multinodeConfig
 		if notifHandler.multinode.Enable {
diff --git a/pkg/util/util.go b/pkg/util/util.go
index 2aaf4af..9cc6888 100644
--- a/pkg/util/util.go
+++ b/pkg/util/util.go
@@ -20,8 +20,16 @@ func ShrinkID(id string) string {
 }
 
 func SameUserNS(pidX, pidY int) (bool, error) {
-	nsX := fmt.Sprintf("/proc/%d/ns/user", pidX)
-	nsY := fmt.Sprintf("/proc/%d/ns/user", pidY)
+	return sameNS(pidX, pidY, "user")
+}
+
+func SameNetNS(pidX, pidY int) (bool, error) {
+	return sameNS(pidX, pidY, "net")
+}
+
+func sameNS(pidX, pidY int, nsName string) (bool, error) {
+	nsX := fmt.Sprintf("/proc/%d/ns/%s", pidX, nsName)
+	nsY := fmt.Sprintf("/proc/%d/ns/%s", pidY, nsName)
 	nsXResolved, err := os.Readlink(nsX)
 	if err != nil {
 		return false, err
diff --git a/test/DockerfileNestedNetNS b/test/DockerfileNestedNetNS
new file mode 100644
index 0000000..8d08c6f
--- /dev/null
+++ b/test/DockerfileNestedNetNS
@@ -0,0 +1,4 @@
+FROM ubuntu:22.04
+
+RUN apt update && apt upgrade -y
+RUN apt install -y netcat iproute2 tcpdump iperf3
diff --git a/test/run_test.sh b/test/run_test.sh
index b0f0ba8..9eafc7b 100755
--- a/test/run_test.sh
+++ b/test/run_test.sh
@@ -19,10 +19,10 @@ if [ "$1" == "SYNC" ]; then
   exec $0 "FORK"
   exit 0
 fi
-cd ~/bypass4netns
-rm -f bypass4netns bypass4netnsd
-make
-sudo make install
+#cd ~/bypass4netns
+#rm -f bypass4netns bypass4netnsd
+#make
+#sudo make install
 set -u
 cd $SCRIPT_DIR
 
@@ -211,3 +211,47 @@ echo "===== multinode test (single node) ===="
   systemctl --user stop etcd.service
   systemctl --user reset-failed
 )
+
+echo "===== nested netns test ===="
+(
+  CONTAINER_NAME="test-nested"
+  set +e
+  nerdctl rm -f $CONTAINER_NAME
+  systemctl --user stop run-iperf3
+  systemctl --user stop run-bypass4netnsd
+  systemctl --user reset-failed
+  set -ex
+
+
+  IMAGE_NAME="b4ns:nested"
+  nerdctl build -f ./DockerfileNestedNetNS -t $IMAGE_NAME .
+
+  systemd-run --user --unit run-bypass4netnsd bypass4netnsd
+  sleep 1
+  nerdctl run --privileged --annotation nerdctl/bypass4netns=true -d -p 5202:5201 --name $CONTAINER_NAME $IMAGE_NAME sleep infinity
+
+  # with container's netns
+  systemd-run --user --unit run-iperf3 nerdctl exec $CONTAINER_NAME iperf3 -s
+  sleep 1
+  iperf3 -c localhost -t 1 -p 5202 --connect-timeout 1000 # it must success to connect.
+  systemctl --user stop run-iperf3
+  systemctl --user reset-failed
+
+  # with nested netns
+  nerdctl exec $CONTAINER_NAME mkdir /sys2
+  nerdctl exec $CONTAINER_NAME mount -t sysfs --make-private /sys2
+  nerdctl exec $CONTAINER_NAME ip netns add nested
+  systemd-run --user --unit run-iperf3 nerdctl exec $CONTAINER_NAME ip netns exec nested iperf3 -s
+  sleep 1
+  set +e
+  iperf3 -c localhost -t 1 -p 5202 --connect-timeout 1000 # it must fail
+  if [ $? -eq 0 ]; then
+    echo "iperf3 must not success to connect."
+    exit 1
+  fi
+  set -e
+  systemctl --user stop run-iperf3
+
+  nerdctl rm -f test-nested
+  systemctl --user reset-failed
+)