package netpol import ( "context" "net" "strings" "testing" "time" netv1 "k8s.io/api/networking/v1" "k8s.io/apimachinery/pkg/api/resource" "k8s.io/client-go/tools/cache" v1 "k8s.io/api/core/v1" metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" "k8s.io/apimachinery/pkg/labels" "k8s.io/client-go/informers" clientset "k8s.io/client-go/kubernetes" "k8s.io/client-go/kubernetes/fake" "github.com/cloudnativelabs/kube-router/pkg/options" ) // newFakeInformersFromClient creates the different informers used in the uneventful network policy controller func newFakeInformersFromClient(kubeClient clientset.Interface) (informers.SharedInformerFactory, cache.SharedIndexInformer, cache.SharedIndexInformer, cache.SharedIndexInformer) { informerFactory := informers.NewSharedInformerFactory(kubeClient, 0) podInformer := informerFactory.Core().V1().Pods().Informer() npInformer := informerFactory.Networking().V1().NetworkPolicies().Informer() nsInformer := informerFactory.Core().V1().Namespaces().Informer() return informerFactory, podInformer, nsInformer, npInformer } type tNamespaceMeta struct { name string labels labels.Set } // Add resources to Informer Store object to simulate updating the Informer func tAddToInformerStore(t *testing.T, informer cache.SharedIndexInformer, obj interface{}) { err := informer.GetStore().Add(obj) if err != nil { t.Fatalf("error injecting object to Informer Store: %v", err) } } type tNetpol struct { name string namespace string podSelector metav1.LabelSelector ingress []netv1.NetworkPolicyIngressRule egress []netv1.NetworkPolicyEgressRule } // createFakeNetpol is a helper to create the network policy from the tNetpol struct func (ns *tNetpol) createFakeNetpol(t *testing.T, informer cache.SharedIndexInformer) { polTypes := make([]netv1.PolicyType, 0) if len(ns.ingress) != 0 { polTypes = append(polTypes, netv1.PolicyTypeIngress) } if len(ns.egress) != 0 { polTypes = append(polTypes, netv1.PolicyTypeEgress) } tAddToInformerStore(t, informer, &netv1.NetworkPolicy{ObjectMeta: metav1.ObjectMeta{Name: ns.name, Namespace: ns.namespace}, Spec: netv1.NetworkPolicySpec{ PodSelector: ns.podSelector, PolicyTypes: polTypes, Ingress: ns.ingress, Egress: ns.egress, }}) } func (ns *tNetpol) findNetpolMatch(netpols *[]networkPolicyInfo) *networkPolicyInfo { for _, netpol := range *netpols { if netpol.namespace == ns.namespace && netpol.name == ns.name { return &netpol } } return nil } // tPodNamespaceMap is a helper to create sets of namespace,pod names type tPodNamespaceMap map[string]map[string]bool func (t tPodNamespaceMap) addPod(pod podInfo) { if _, ok := t[pod.namespace]; !ok { t[pod.namespace] = make(map[string]bool) } t[pod.namespace][pod.name] = true } func (t tPodNamespaceMap) delPod(pod podInfo) { delete(t[pod.namespace], pod.name) if len(t[pod.namespace]) == 0 { delete(t, pod.namespace) } } func (t tPodNamespaceMap) addNSPodInfo(ns, podname string) { if _, ok := t[ns]; !ok { t[ns] = make(map[string]bool) } t[ns][podname] = true } func (t tPodNamespaceMap) copy() tPodNamespaceMap { newMap := make(tPodNamespaceMap) for ns, pods := range t { for p := range pods { newMap.addNSPodInfo(ns, p) } } return newMap } func (t tPodNamespaceMap) toStrSlice() (r []string) { for ns, pods := range t { for pod := range pods { r = append(r, ns+":"+pod) } } return } // tNewPodNamespaceMapFromTC creates a new tPodNamespaceMap from the info of the testcase func tNewPodNamespaceMapFromTC(target map[string]string) tPodNamespaceMap { newMap := make(tPodNamespaceMap) for ns, pods := range target { for _, pod := range strings.Split(pods, ",") { newMap.addNSPodInfo(ns, pod) } } return newMap } // tCreateFakePods creates the Pods and Namespaces that will be affected by the network policies // returns a map like map[Namespace]map[PodName]bool func tCreateFakePods(t *testing.T, podInformer cache.SharedIndexInformer, nsInformer cache.SharedIndexInformer) tPodNamespaceMap { podNamespaceMap := make(tPodNamespaceMap) pods := []podInfo{ {name: "Aa", labels: labels.Set{"app": "a"}, namespace: "nsA", ip: "1.1"}, {name: "Aaa", labels: labels.Set{"app": "a", "component": "a"}, namespace: "nsA", ip: "1.2"}, {name: "Aab", labels: labels.Set{"app": "a", "component": "b"}, namespace: "nsA", ip: "1.3"}, {name: "Aac", labels: labels.Set{"app": "a", "component": "c"}, namespace: "nsA", ip: "1.4"}, {name: "Ba", labels: labels.Set{"app": "a"}, namespace: "nsB", ip: "2.1"}, {name: "Baa", labels: labels.Set{"app": "a", "component": "a"}, namespace: "nsB", ip: "2.2"}, {name: "Bab", labels: labels.Set{"app": "a", "component": "b"}, namespace: "nsB", ip: "2.3"}, {name: "Ca", labels: labels.Set{"app": "a"}, namespace: "nsC", ip: "3.1"}, } namespaces := []tNamespaceMeta{ {name: "nsA", labels: labels.Set{"name": "a", "team": "a"}}, {name: "nsB", labels: labels.Set{"name": "b", "team": "a"}}, {name: "nsC", labels: labels.Set{"name": "c"}}, {name: "nsD", labels: labels.Set{"name": "d"}}, } ipsUsed := make(map[string]bool) for _, pod := range pods { podNamespaceMap.addPod(pod) ipaddr := "1.1." + pod.ip if ipsUsed[ipaddr] { t.Fatalf("there is another pod with the same Ip address %s as this pod %s namespace %s", ipaddr, pod.name, pod.name) } ipsUsed[ipaddr] = true tAddToInformerStore(t, podInformer, &v1.Pod{ObjectMeta: metav1.ObjectMeta{Name: pod.name, Labels: pod.labels, Namespace: pod.namespace}, Status: v1.PodStatus{PodIP: ipaddr}}) } for _, ns := range namespaces { tAddToInformerStore(t, nsInformer, &v1.Namespace{ObjectMeta: metav1.ObjectMeta{Name: ns.name, Labels: ns.labels}}) } return podNamespaceMap } // newFakeNode is a helper function for creating Nodes for testing. func newFakeNode(name string, addr string) *v1.Node { return &v1.Node{ ObjectMeta: metav1.ObjectMeta{Name: name}, Status: v1.NodeStatus{ Capacity: v1.ResourceList{ v1.ResourceName(v1.ResourceCPU): resource.MustParse("1"), v1.ResourceName(v1.ResourceMemory): resource.MustParse("1G"), }, Addresses: []v1.NodeAddress{{Type: v1.NodeExternalIP, Address: addr}}, }, } } // newUneventfulNetworkPolicyController returns new NetworkPolicyController object without any event handler func newUneventfulNetworkPolicyController(podInformer cache.SharedIndexInformer, npInformer cache.SharedIndexInformer, nsInformer cache.SharedIndexInformer) (*NetworkPolicyController, error) { npc := NetworkPolicyController{} npc.syncPeriod = time.Hour npc.nodeHostName = "node" npc.nodeIP = net.IPv4(10, 10, 10, 10) npc.podLister = podInformer.GetIndexer() npc.nsLister = nsInformer.GetIndexer() npc.npLister = npInformer.GetIndexer() return &npc, nil } // tNetpolTestCase helper struct to define the inputs to the test case (netpols) and // the expected selected targets (targetPods, inSourcePods for ingress targets, and outDestPods // for egress targets) as maps with key being the namespace and a csv of pod names type tNetpolTestCase struct { name string netpol tNetpol targetPods tPodNamespaceMap inSourcePods tPodNamespaceMap outDestPods tPodNamespaceMap } // tGetNotTargetedPods finds set of pods that should not be targeted by netpol selectors func tGetNotTargetedPods(podsGot []podInfo, wanted tPodNamespaceMap) []string { unwanted := make(tPodNamespaceMap) for _, pod := range podsGot { if !wanted[pod.namespace][pod.name] { unwanted.addPod(pod) } } return unwanted.toStrSlice() } // tGetTargetPodsMissing returns the set of pods that should have been targeted but were missing by netpol selectors func tGetTargetPodsMissing(podsGot []podInfo, wanted tPodNamespaceMap) []string { missing := wanted.copy() for _, pod := range podsGot { if wanted[pod.namespace][pod.name] { missing.delPod(pod) } } return missing.toStrSlice() } func tListOfPodsFromTargets(target map[string]podInfo) (r []podInfo) { for _, pod := range target { r = append(r, pod) } return } func testForMissingOrUnwanted(t *testing.T, targetMsg string, got []podInfo, wanted tPodNamespaceMap) { if missing := tGetTargetPodsMissing(got, wanted); len(missing) != 0 { t.Errorf("Some Pods were not selected %s: %s", targetMsg, strings.Join(missing, ", ")) } if missing := tGetNotTargetedPods(got, wanted); len(missing) != 0 { t.Errorf("Some Pods NOT expected were selected on %s: %s", targetMsg, strings.Join(missing, ", ")) } } func newMinimalKubeRouterConfig(clusterIPCIDR string, nodePortRange string, hostNameOverride string, externalIPs []string) *options.KubeRouterConfig { kubeConfig := options.NewKubeRouterConfig() if clusterIPCIDR != "" { kubeConfig.ClusterIPCIDR = clusterIPCIDR } if nodePortRange != "" { kubeConfig.NodePortRange = nodePortRange } if hostNameOverride != "" { kubeConfig.HostnameOverride = hostNameOverride } if externalIPs != nil { kubeConfig.ExternalIPCIDRs = externalIPs } return kubeConfig } type tNetPolConfigTestCase struct { name string config *options.KubeRouterConfig expectError bool errorText string } func TestNewNetworkPolicySelectors(t *testing.T) { testCases := []tNetpolTestCase{ { name: "Non-Existent Namespace", netpol: tNetpol{name: "nsXX", podSelector: metav1.LabelSelector{}, namespace: "nsXX"}, targetPods: nil, }, { name: "Empty Namespace", netpol: tNetpol{name: "nsD", podSelector: metav1.LabelSelector{}, namespace: "nsD"}, targetPods: nil, }, { name: "All pods in nsA", netpol: tNetpol{name: "nsA", podSelector: metav1.LabelSelector{}, namespace: "nsA"}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aa,Aaa,Aab,Aac"}), }, { name: "All pods in nsB", netpol: tNetpol{name: "nsB", podSelector: metav1.LabelSelector{}, namespace: "nsB"}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsB": "Ba,Baa,Bab"}), }, { name: "All pods in nsC", netpol: tNetpol{name: "nsC", podSelector: metav1.LabelSelector{}, namespace: "nsC"}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsC": "Ca"}), }, { name: "All pods app=a in nsA using matchExpressions", netpol: tNetpol{ name: "nsA-app-a-matchExpression", namespace: "nsA", podSelector: metav1.LabelSelector{ MatchExpressions: []metav1.LabelSelectorRequirement{{ Key: "app", Operator: "In", Values: []string{"a"}, }}}}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aa,Aaa,Aab,Aac"}), }, { name: "All pods app=a in nsA using matchLabels", netpol: tNetpol{name: "nsA-app-a-matchLabels", namespace: "nsA", podSelector: metav1.LabelSelector{ MatchLabels: map[string]string{"app": "a"}}}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aa,Aaa,Aab,Aac"}), }, { name: "All pods app=a in nsA using matchLabels ingress allow from any pod in nsB", netpol: tNetpol{name: "nsA-app-a-matchLabels-2", namespace: "nsA", podSelector: metav1.LabelSelector{MatchLabels: map[string]string{"app": "a"}}, ingress: []netv1.NetworkPolicyIngressRule{{From: []netv1.NetworkPolicyPeer{{NamespaceSelector: &metav1.LabelSelector{MatchLabels: map[string]string{"name": "b"}}}}}}, }, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aa,Aaa,Aab,Aac"}), inSourcePods: tNewPodNamespaceMapFromTC(map[string]string{"nsB": "Ba,Baa,Bab"}), }, { name: "All pods app=a in nsA using matchLabels ingress allow from pod in nsB with component = b", netpol: tNetpol{name: "nsA-app-a-matchExpression-2", namespace: "nsA", podSelector: metav1.LabelSelector{MatchLabels: map[string]string{"app": "a"}}, ingress: []netv1.NetworkPolicyIngressRule{{From: []netv1.NetworkPolicyPeer{ { NamespaceSelector: &metav1.LabelSelector{MatchLabels: map[string]string{"name": "b"}}, PodSelector: &metav1.LabelSelector{ MatchExpressions: []metav1.LabelSelectorRequirement{{ Key: "component", Operator: "In", Values: []string{"b"}, }}}, }, }}}}, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aa,Aaa,Aab,Aac"}), inSourcePods: tNewPodNamespaceMapFromTC(map[string]string{"nsB": "Bab"}), }, { name: "All pods app=a,component=b or c in nsA", netpol: tNetpol{name: "nsA-app-a-matchExpression-3", namespace: "nsA", podSelector: metav1.LabelSelector{ MatchExpressions: []metav1.LabelSelectorRequirement{ { Key: "app", Operator: "In", Values: []string{"a"}, }, { Key: "component", Operator: "In", Values: []string{"b", "c"}, }}}, }, targetPods: tNewPodNamespaceMapFromTC(map[string]string{"nsA": "Aab,Aac"}), }, } client := fake.NewSimpleClientset(&v1.NodeList{Items: []v1.Node{*newFakeNode("node", "10.10.10.10")}}) informerFactory, podInformer, nsInformer, netpolInformer := newFakeInformersFromClient(client) ctx, cancel := context.WithCancel(context.Background()) defer cancel() informerFactory.Start(ctx.Done()) cache.WaitForCacheSync(ctx.Done(), podInformer.HasSynced) krNetPol, _ := newUneventfulNetworkPolicyController(podInformer, netpolInformer, nsInformer) tCreateFakePods(t, podInformer, nsInformer) for _, test := range testCases { test.netpol.createFakeNetpol(t, netpolInformer) } netpols, err := krNetPol.buildNetworkPoliciesInfo() if err != nil { t.Errorf("Problems building policies") } for _, test := range testCases { t.Run(test.name, func(t *testing.T) { np := test.netpol.findNetpolMatch(&netpols) testForMissingOrUnwanted(t, "targetPods", tListOfPodsFromTargets(np.targetPods), test.targetPods) for _, ingress := range np.ingressRules { testForMissingOrUnwanted(t, "ingress srcPods", ingress.srcPods, test.inSourcePods) } for _, egress := range np.egressRules { testForMissingOrUnwanted(t, "egress dstPods", egress.dstPods, test.outDestPods) } }) } } func TestNetworkPolicyController(t *testing.T) { testCases := []tNetPolConfigTestCase{ { "Default options are successful", newMinimalKubeRouterConfig("", "", "node", nil), false, "", }, { "Missing nodename fails appropriately", newMinimalKubeRouterConfig("", "", "", nil), true, "Failed to identify the node by NODE_NAME, hostname or --hostname-override", }, { "Test bad cluster CIDR (not properly formatting ip address)", newMinimalKubeRouterConfig("10.10.10", "", "node", nil), true, "failed to get parse --service-cluster-ip-range parameter: invalid CIDR address: 10.10.10", }, { "Test bad cluster CIDR (not using an ip address)", newMinimalKubeRouterConfig("foo", "", "node", nil), true, "failed to get parse --service-cluster-ip-range parameter: invalid CIDR address: foo", }, { "Test bad cluster CIDR (using an ip address that is not a CIDR)", newMinimalKubeRouterConfig("10.10.10.10", "", "node", nil), true, "failed to get parse --service-cluster-ip-range parameter: invalid CIDR address: 10.10.10.10", }, { "Test good cluster CIDR (using single IP with a /32)", newMinimalKubeRouterConfig("10.10.10.10/32", "", "node", nil), false, "", }, { "Test good cluster CIDR (using normal range with /24)", newMinimalKubeRouterConfig("10.10.10.0/24", "", "node", nil), false, "", }, { "Test bad node port specification (using commas)", newMinimalKubeRouterConfig("", "8080,8081", "node", nil), true, "failed to parse node port range given: '8080,8081' please see specification in help text", }, { "Test bad node port specification (not using numbers)", newMinimalKubeRouterConfig("", "foo:bar", "node", nil), true, "failed to parse node port range given: 'foo:bar' please see specification in help text", }, { "Test bad node port specification (using anything in addition to range)", newMinimalKubeRouterConfig("", "8080,8081-8090", "node", nil), true, "failed to parse node port range given: '8080,8081-8090' please see specification in help text", }, { "Test bad node port specification (using reversed range)", newMinimalKubeRouterConfig("", "8090-8080", "node", nil), true, "port 1 is greater than or equal to port 2 in range given: '8090-8080'", }, { "Test bad node port specification (port out of available range)", newMinimalKubeRouterConfig("", "132000-132001", "node", nil), true, "could not parse first port number from range given: '132000-132001'", }, { "Test good node port specification (using colon separator)", newMinimalKubeRouterConfig("", "8080:8090", "node", nil), false, "", }, { "Test good node port specification (using hyphen separator)", newMinimalKubeRouterConfig("", "8080-8090", "node", nil), false, "", }, { "Test bad external IP CIDR (not properly formatting ip address)", newMinimalKubeRouterConfig("", "", "node", []string{"199.10.10"}), true, "failed to get parse --service-external-ip-range parameter: '199.10.10'. Error: invalid CIDR address: 199.10.10", }, { "Test bad external IP CIDR (not using an ip address)", newMinimalKubeRouterConfig("", "", "node", []string{"foo"}), true, "failed to get parse --service-external-ip-range parameter: 'foo'. Error: invalid CIDR address: foo", }, { "Test bad external IP CIDR (using an ip address that is not a CIDR)", newMinimalKubeRouterConfig("", "", "node", []string{"199.10.10.10"}), true, "failed to get parse --service-external-ip-range parameter: '199.10.10.10'. Error: invalid CIDR address: 199.10.10.10", }, { "Test bad external IP CIDR (making sure that it processes all items in the list)", newMinimalKubeRouterConfig("", "", "node", []string{"199.10.10.10/32", "199.10.10.11"}), true, "failed to get parse --service-external-ip-range parameter: '199.10.10.11'. Error: invalid CIDR address: 199.10.10.11", }, { "Test good external IP CIDR (using single IP with a /32)", newMinimalKubeRouterConfig("", "", "node", []string{"199.10.10.10/32"}), false, "", }, { "Test good external IP CIDR (using normal range with /24)", newMinimalKubeRouterConfig("", "", "node", []string{"199.10.10.10/24"}), false, "", }, } client := fake.NewSimpleClientset(&v1.NodeList{Items: []v1.Node{*newFakeNode("node", "10.10.10.10")}}) _, podInformer, nsInformer, netpolInformer := newFakeInformersFromClient(client) for _, test := range testCases { t.Run(test.name, func(t *testing.T) { _, err := NewNetworkPolicyController(client, test.config, podInformer, netpolInformer, nsInformer) if err == nil && test.expectError { t.Error("This config should have failed, but it was successful instead") } else if err != nil { // Unfortunately without doing a ton of extra refactoring work, we can't remove this reference easily // from the controllers start up. Luckily it's one of the last items to be processed in the controller // so for now we'll consider that if we hit this error that we essentially didn't hit an error at all // TODO: refactor NPC to use an injectable interface for ipset operations if !test.expectError && err.Error() != "Ipset utility not found" { t.Errorf("This config should have been successful, but it failed instead. Error: %s", err) } else if test.expectError && err.Error() != test.errorText { t.Errorf("Expected error: '%s' but instead got: '%s'", test.errorText, err) } } }) } } // Ref: // https://github.com/kubernetes/kubernetes/blob/master/pkg/controller/podgc/gc_controller_test.go // https://github.com/kubernetes/kubernetes/blob/master/pkg/controller/testutil/test_utils.go