Merge pull request #21790 from vishh/seccomp

pkg/kubelet/dockertools/fake_docker_client.go

@@ -56,8 +56,12 @@ type FakeDockerClient struct {
 }
 
 func NewFakeDockerClient() *FakeDockerClient {
+	return NewFakeDockerClientWithVersion("1.8.1", "1.20")
+}
+
+func NewFakeDockerClientWithVersion(version, apiVersion string) *FakeDockerClient {
 	return &FakeDockerClient{
-		VersionInfo:   docker.Env{"Version=1.8.1", "ApiVersion=1.20"},
+		VersionInfo:   docker.Env{fmt.Sprintf("Version=%s", version), fmt.Sprintf("ApiVersion=%s", apiVersion)},
 		Errors:        make(map[string]error),
 		RemovedImages: sets.String{},
 		ContainerMap:  make(map[string]*docker.Container),

pkg/kubelet/dockertools/manager.go

@@ -79,13 +79,23 @@ const (
 	// networking). Must match the value returned by docker inspect -f
 	// '{{.HostConfig.NetworkMode}}'.
 	namespaceModeHost = "host"
+
+	// Remote API version for docker daemon version v1.10
+	// https://docs.docker.com/engine/reference/api/docker_remote_api/
+	dockerV110APIVersion = "1.22"
 )
 
-// DockerManager implements the Runtime interface.
-var _ kubecontainer.Runtime = &DockerManager{}
+var (
+	// DockerManager implements the Runtime interface.
+	_ kubecontainer.Runtime = &DockerManager{}
 
-// TODO: make this a TTL based pull (if image older than X policy, pull)
-var podInfraContainerImagePullPolicy = api.PullIfNotPresent
+	// TODO: make this a TTL based pull (if image older than X policy, pull)
+	podInfraContainerImagePullPolicy = api.PullIfNotPresent
+
+	// Default set of security options. Seccomp is disabled by default until
+	// github issue #20870 is resolved.
+	defaultSecurityOpt = []string{"seccomp:unconfined"}
+)
 
 type DockerManager struct {
 	client              DockerInterface
@@ -488,6 +498,11 @@ func (dm *DockerManager) runContainer(
 		ContainerName: container.Name,
 	}
 
+	securityOpts, err := dm.defaultSecurityOpt()
+	if err != nil {
+		return kubecontainer.ContainerID{}, err
+	}
+
 	// Pod information is recorded on the container as labels to preserve it in the event the pod is deleted
 	// while the Kubelet is down and there is no information available to recover the pod.
 	// TODO: keep these labels up to date if the pod changes
@@ -551,9 +566,10 @@ func (dm *DockerManager) runContainer(
 		PidMode:        pidMode,
 		ReadonlyRootfs: readOnlyRootFilesystem(container),
 		// Memory and CPU are set here for newer versions of Docker (1.6+).
-		Memory:     memoryLimit,
-		MemorySwap: -1,
-		CPUShares:  cpuShares,
+		Memory:      memoryLimit,
+		MemorySwap:  -1,
+		CPUShares:   cpuShares,
+		SecurityOpt: securityOpts,
 	}
 
 	if dm.cpuCFSQuota {
@@ -934,6 +950,22 @@ func (dm *DockerManager) nativeExecSupportExists() (bool, error) {
 	return false, err
 }
 
+func (dm *DockerManager) defaultSecurityOpt() ([]string, error) {
+	version, err := dm.APIVersion()
+	if err != nil {
+		return nil, err
+	}
+	// seccomp is to be disabled on docker versions >= v1.10
+	result, err := version.Compare(dockerV110APIVersion)
+	if err != nil {
+		return nil, err
+	}
+	if result >= 0 {
+		return defaultSecurityOpt, nil
+	}
+	return nil, nil
+}
+
 func (dm *DockerManager) getRunInContainerCommand(containerID kubecontainer.ContainerID, cmd []string) (*exec.Cmd, error) {
 	args := append([]string{"exec"}, cmd...)
 	command := exec.Command("/usr/sbin/nsinit", args...)

pkg/kubelet/dockertools/manager_test.go

@@ -82,8 +82,8 @@ func (f *fakeRuntimeHelper) GetClusterDNS(pod *api.Pod) ([]string, []string, err
 	return nil, nil, fmt.Errorf("not implemented")
 }
 
-func newTestDockerManagerWithHTTPClient(fakeHTTPClient *fakeHTTP) (*DockerManager, *FakeDockerClient) {
-	fakeDocker := NewFakeDockerClient()
+func newTestDockerManagerWithHTTPClientWithVersion(fakeHTTPClient *fakeHTTP, version, apiVersion string) (*DockerManager, *FakeDockerClient) {
+	fakeDocker := NewFakeDockerClientWithVersion(version, apiVersion)
 	fakeRecorder := &record.FakeRecorder{}
 	containerRefManager := kubecontainer.NewRefManager()
 	networkPlugin, _ := network.InitNetworkPlugin([]network.NetworkPlugin{}, "", network.NewFakeHost(nil))
@@ -104,6 +104,10 @@ func newTestDockerManagerWithHTTPClient(fakeHTTPClient *fakeHTTP) (*DockerManage
 	return dockerManager, fakeDocker
 }
 
+func newTestDockerManagerWithHTTPClient(fakeHTTPClient *fakeHTTP) (*DockerManager, *FakeDockerClient) {
+	return newTestDockerManagerWithHTTPClientWithVersion(fakeHTTPClient, "1.8.1", "1.20")
+}
+
 func newTestDockerManager() (*DockerManager, *FakeDockerClient) {
 	return newTestDockerManagerWithHTTPClient(&fakeHTTP{})
 }
@@ -634,7 +638,7 @@ func TestSyncPodCreateNetAndContainer(t *testing.T) {
 	if len(fakeDocker.Created) != 2 ||
 		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 }
@@ -670,13 +674,13 @@ func TestSyncPodCreatesNetAndContainerPullsImage(t *testing.T) {
 	fakeDocker.Lock()
 
 	if !reflect.DeepEqual(puller.ImagesPulled, []string{"pod_infra_image", "something"}) {
-		t.Errorf("Unexpected pulled containers: %v", puller.ImagesPulled)
+		t.Errorf("unexpected pulled containers: %v", puller.ImagesPulled)
 	}
 
 	if len(fakeDocker.Created) != 2 ||
 		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 }
@@ -713,7 +717,7 @@ func TestSyncPodWithPodInfraCreatesContainer(t *testing.T) {
 	fakeDocker.Lock()
 	if len(fakeDocker.Created) != 1 ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 }
@@ -1151,7 +1155,7 @@ func TestGetRestartCount(t *testing.T) {
 		runSyncPod(t, dm, fakeDocker, pod, nil, false)
 		status, err := dm.GetPodStatus(pod.UID, pod.Name, pod.Namespace)
 		if err != nil {
-			t.Fatalf("Unexpected error %v", err)
+			t.Fatalf("unexpected error %v", err)
 		}
 		cs := status.FindContainerStatusByName(containerName)
 		if cs == nil {
@@ -1166,7 +1170,7 @@ func TestGetRestartCount(t *testing.T) {
 	killOneContainer := func(pod *api.Pod) {
 		status, err := dm.GetPodStatus(pod.UID, pod.Name, pod.Namespace)
 		if err != nil {
-			t.Fatalf("Unexpected error %v", err)
+			t.Fatalf("unexpected error %v", err)
 		}
 		cs := status.FindContainerStatusByName(containerName)
 		if cs == nil {
@@ -1234,11 +1238,11 @@ func TestGetTerminationMessagePath(t *testing.T) {
 	containerList := fakeDocker.ContainerList
 	if len(containerList) != 2 {
 		// One for infra container, one for container "bar"
-		t.Fatalf("Unexpected container list length %d", len(containerList))
+		t.Fatalf("unexpected container list length %d", len(containerList))
 	}
 	inspectResult, err := dm.client.InspectContainer(containerList[0].ID)
 	if err != nil {
-		t.Fatalf("Unexpected inspect error: %v", err)
+		t.Fatalf("unexpected inspect error: %v", err)
 	}
 	containerInfo := getContainerInfoFromLabel(inspectResult.Config.Labels)
 	terminationMessagePath := containerInfo.TerminationMessagePath
@@ -1290,11 +1294,11 @@ func TestSyncPodWithPodInfraCreatesContainerCallsHandler(t *testing.T) {
 	fakeDocker.Lock()
 	if len(fakeDocker.Created) != 1 ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 	if fakeHTTPClient.url != "http://foo:8080/bar" {
-		t.Errorf("Unexpected handler: %q", fakeHTTPClient.url)
+		t.Errorf("unexpected handler: %q", fakeHTTPClient.url)
 	}
 }
 
@@ -1347,7 +1351,7 @@ func TestSyncPodEventHandlerFails(t *testing.T) {
 	}
 	dockerName, _, err := ParseDockerName(fakeDocker.Stopped[0])
 	if err != nil {
-		t.Errorf("Unexpected error: %v", err)
+		t.Errorf("unexpected error: %v", err)
 	}
 	if dockerName.ContainerName != "bar" {
 		t.Errorf("Wrong stopped container, expected: bar, get: %q", dockerName.ContainerName)
@@ -1417,7 +1421,7 @@ func TestSyncPodWithTerminationLog(t *testing.T) {
 	if len(fakeDocker.Created) != 2 ||
 		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 	newContainer, err := fakeDocker.InspectContainer(fakeDocker.Created[1])
@@ -1426,10 +1430,10 @@ func TestSyncPodWithTerminationLog(t *testing.T) {
 	}
 	parts := strings.Split(newContainer.HostConfig.Binds[0], ":")
 	if !matchString(t, testPodContainerDir+"/[a-f0-9]", parts[0]) {
-		t.Errorf("Unexpected host path: %s", parts[0])
+		t.Errorf("unexpected host path: %s", parts[0])
 	}
 	if parts[1] != "/dev/somepath" {
-		t.Errorf("Unexpected container path: %s", parts[1])
+		t.Errorf("unexpected container path: %s", parts[1])
 	}
 }
 
@@ -1464,7 +1468,7 @@ func TestSyncPodWithHostNetwork(t *testing.T) {
 	if len(fakeDocker.Created) != 2 ||
 		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
 		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 	fakeDocker.Unlock()
 
@@ -1687,7 +1691,7 @@ func TestSyncPodWithPullPolicy(t *testing.T) {
 	assert.Equal(t, []string{"pod_infra_image", "pull_always_image", "pull_if_not_present_image"}, pulledImageSorted)
 
 	if len(fakeDocker.Created) != 5 {
-		t.Errorf("Unexpected containers created %v", fakeDocker.Created)
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
 	}
 }
 
@@ -1781,3 +1785,81 @@ func verifySyncResults(t *testing.T, expectedResults []*kubecontainer.SyncResult
 		}
 	}
 }
+
+func TestSeccompIsDisabledWithDockerV110(t *testing.T) {
+	dm, fakeDocker := newTestDockerManagerWithHTTPClientWithVersion(&fakeHTTP{}, "1.10.1", "1.22")
+	pod := &api.Pod{
+		ObjectMeta: api.ObjectMeta{
+			UID:       "12345678",
+			Name:      "foo",
+			Namespace: "new",
+		},
+		Spec: api.PodSpec{
+			Containers: []api.Container{
+				{Name: "bar"},
+			},
+		},
+	}
+
+	runSyncPod(t, dm, fakeDocker, pod, nil, false)
+
+	verifyCalls(t, fakeDocker, []string{
+		// Create pod infra container.
+		"create", "start", "inspect_container", "inspect_container",
+		// Create container.
+		"create", "start", "inspect_container",
+	})
+
+	fakeDocker.Lock()
+	if len(fakeDocker.Created) != 2 ||
+		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
+		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
+	}
+	fakeDocker.Unlock()
+
+	newContainer, err := fakeDocker.InspectContainer(fakeDocker.Created[1])
+	if err != nil {
+		t.Fatalf("unexpected error %v", err)
+	}
+	assert.Contains(t, newContainer.HostConfig.SecurityOpt, "seccomp:unconfined", "Pods with Docker versions >= 1.10 must have seccomp disabled.")
+}
+
+func TestSecurityOptsAreNilWithDockerV19(t *testing.T) {
+	dm, fakeDocker := newTestDockerManagerWithHTTPClientWithVersion(&fakeHTTP{}, "1.9.1", "1.21")
+	pod := &api.Pod{
+		ObjectMeta: api.ObjectMeta{
+			UID:       "12345678",
+			Name:      "foo",
+			Namespace: "new",
+		},
+		Spec: api.PodSpec{
+			Containers: []api.Container{
+				{Name: "bar"},
+			},
+		},
+	}
+
+	runSyncPod(t, dm, fakeDocker, pod, nil, false)
+
+	verifyCalls(t, fakeDocker, []string{
+		// Create pod infra container.
+		"create", "start", "inspect_container", "inspect_container",
+		// Create container.
+		"create", "start", "inspect_container",
+	})
+
+	fakeDocker.Lock()
+	if len(fakeDocker.Created) != 2 ||
+		!matchString(t, "/k8s_POD\\.[a-f0-9]+_foo_new_", fakeDocker.Created[0]) ||
+		!matchString(t, "/k8s_bar\\.[a-f0-9]+_foo_new_", fakeDocker.Created[1]) {
+		t.Errorf("unexpected containers created %v", fakeDocker.Created)
+	}
+	fakeDocker.Unlock()
+
+	newContainer, err := fakeDocker.InspectContainer(fakeDocker.Created[1])
+	if err != nil {
+		t.Fatalf("unexpected error %v", err)
+	}
+	assert.NotContains(t, newContainer.HostConfig.SecurityOpt, "seccomp:unconfined", "Pods with Docker versions < 1.10 must not have seccomp disabled by default")
+}