Merge branch 'master' of https://github.com/k8snetworkplumbingwg/srio…

…v-network-operator into sync/upstream-master-to-network-operator-99.9.9-beta.0
Mellanox · Jul 17, 2024 · 9f5347f · 9f5347f
2 parents 5d061d0 + 573367c
commit 9f5347f
Show file tree

Hide file tree

Showing 19 changed files with 162 additions and 10 deletions.
diff --git a/.github/workflows/pr-ci-triggers.yml b/.github/workflows/pr-ci-triggers.yml
@@ -1,5 +1,6 @@
 on:
   pull_request_target:
+    types: [opened, reopened]
 
 jobs:
   vendors-ci-triggers-list:
@@ -9,12 +10,12 @@ jobs:
         env:
           MESSAGE: >-
             Thanks for your PR,\n
-            To run vendors CIs use one of:\n
+            To run vendors CIs, Maintainers can use one of:\n
             * `/test-all`: To run all tests for all vendors.\n
             * `/test-e2e-all`: To run all E2E tests for all vendors.\n
             * `/test-e2e-nvidia-all`: To run all E2E tests for NVIDIA vendor.\n
             \n
-            To skip the vendors CIs use one of:\n
+            To skip the vendors CIs, Maintainers can use one of:\n
             * `/skip-all`: To skip all tests for all vendors.\n
             * `/skip-e2e-all`: To skip all E2E tests for all vendors.\n
             * `/skip-e2e-nvidia-all`: To skip all E2E tests for NVIDIA vendor.\n

diff --git a/bindata/manifests/daemon/daemonset.yaml b/bindata/manifests/daemon/daemonset.yaml
@@ -92,6 +92,23 @@ spec:
               mountPath: /host/etc/os-release
               readOnly: true
         {{- end }}
+        {{- if .RDMACNIImage }}
+        - name: rdma-cni
+          image: {{.RDMACNIImage}}
+          args: ["--no-sleep"]
+          securityContext:
+            privileged: true
+          resources:
+            requests:
+              cpu: 10m
+              memory: 10Mi
+          volumeMounts:
+            - name: cnibin
+              mountPath: /host/opt/cni/bin
+            - name: os-release
+              mountPath: /host/etc/os-release
+              readOnly: true
+        {{- end }}
       {{- if .UsedSystemdMode}}
         - name: sriov-service-copy
           image: {{.Image}}

diff --git a/controllers/sriovoperatorconfig_controller.go b/controllers/sriovoperatorconfig_controller.go
@@ -178,6 +178,7 @@ func (r *SriovOperatorConfigReconciler) syncConfigDaemonSet(ctx context.Context,
 	data.Data["SRIOVCNIImage"] = os.Getenv("SRIOV_CNI_IMAGE")
 	data.Data["SRIOVInfiniBandCNIImage"] = os.Getenv("SRIOV_INFINIBAND_CNI_IMAGE")
 	data.Data["OVSCNIImage"] = os.Getenv("OVS_CNI_IMAGE")
+	data.Data["RDMACNIImage"] = os.Getenv("RDMA_CNI_IMAGE")
 	data.Data["ReleaseVersion"] = os.Getenv("RELEASEVERSION")
 	data.Data["ClusterType"] = vars.ClusterType
 	data.Data["DevMode"] = os.Getenv("DEV_MODE")

diff --git a/controllers/suite_test.go b/controllers/suite_test.go
@@ -115,6 +115,8 @@ var _ = BeforeSuite(func() {
 	Expect(err).NotTo(HaveOccurred())
 	err = os.Setenv("OVS_CNI_IMAGE", "mock-image")
 	Expect(err).NotTo(HaveOccurred())
+	err = os.Setenv("RDMA_CNI_IMAGE", "mock-image")
+	Expect(err).NotTo(HaveOccurred())
 	err = os.Setenv("SRIOV_DEVICE_PLUGIN_IMAGE", "mock-image")
 	Expect(err).NotTo(HaveOccurred())
 	err = os.Setenv("NETWORK_RESOURCES_INJECTOR_IMAGE", "mock-image")

diff --git a/deploy/operator.yaml b/deploy/operator.yaml
@@ -58,6 +58,8 @@ spec:
               value: $SRIOV_INFINIBAND_CNI_IMAGE
             - name: OVS_CNI_IMAGE
               value: $OVS_CNI_IMAGE
+            - name: RDMA_CNI_IMAGE
+              value: $RDMA_CNI_IMAGE
             - name: SRIOV_DEVICE_PLUGIN_IMAGE
               value: $SRIOV_DEVICE_PLUGIN_IMAGE
             - name: NETWORK_RESOURCES_INJECTOR_IMAGE

diff --git a/deployment/sriov-network-operator-chart/README.md b/deployment/sriov-network-operator-chart/README.md
@@ -139,6 +139,7 @@ This section contains general parameters that apply to both the operator and dae
 | `images.sriovCni` | SR-IOV CNI image |
 | `images.ibSriovCni` | InfiniBand SR-IOV CNI image |
 | `images.ovsCni` | OVS CNI image |
+| `images.rdmaCni` | RDMA CNI image              |
 | `images.sriovDevicePlugin` | SR-IOV device plugin image |
 | `images.resourcesInjector` | Resources Injector image |
 | `images.webhook` | Operator Webhook image |

diff --git a/deployment/sriov-network-operator-chart/templates/operator.yaml b/deployment/sriov-network-operator-chart/templates/operator.yaml
@@ -58,6 +58,8 @@ spec:
               value: {{ .Values.images.ibSriovCni }}
             - name: OVS_CNI_IMAGE
               value: {{ .Values.images.ovsCni }}
+            - name: RDMA_CNI_IMAGE
+              value: {{ .Values.images.rdmaCni }}
             - name: SRIOV_DEVICE_PLUGIN_IMAGE
               value: {{ .Values.images.sriovDevicePlugin }}
             - name: NETWORK_RESOURCES_INJECTOR_IMAGE

diff --git a/deployment/sriov-network-operator-chart/values.yaml b/deployment/sriov-network-operator-chart/values.yaml
@@ -103,6 +103,7 @@ images:
   sriovCni: ghcr.io/k8snetworkplumbingwg/sriov-cni
   ibSriovCni: ghcr.io/k8snetworkplumbingwg/ib-sriov-cni
   ovsCni: ghcr.io/k8snetworkplumbingwg/ovs-cni-plugin
+  rdmaCni: ghcr.io/k8snetworkplumbingwg/rdma-cni
   sriovDevicePlugin: ghcr.io/k8snetworkplumbingwg/sriov-network-device-plugin
   resourcesInjector: ghcr.io/k8snetworkplumbingwg/network-resources-injector
   webhook: ghcr.io/k8snetworkplumbingwg/sriov-network-operator-webhook

diff --git a/hack/env.sh b/hack/env.sh
@@ -3,6 +3,8 @@ if [ -z $SKIP_VAR_SET ]; then
         export SRIOV_INFINIBAND_CNI_IMAGE=${SRIOV_INFINIBAND_CNI_IMAGE:-ghcr.io/k8snetworkplumbingwg/ib-sriov-cni}
         # OVS_CNI_IMAGE can be explicitly set to empty value, use default only if the var is not set
         export OVS_CNI_IMAGE=${OVS_CNI_IMAGE-ghcr.io/k8snetworkplumbingwg/ovs-cni-plugin}
+        # RDMA_CNI_IMAGE can be explicitly set to empty value, use default only if the var is not set
+        export RDMA_CNI_IMAGE=${RDMA_CNI_IMAGE-ghcr.io/k8snetworkplumbingwg/rdma-cni}
         export SRIOV_DEVICE_PLUGIN_IMAGE=${SRIOV_DEVICE_PLUGIN_IMAGE:-ghcr.io/k8snetworkplumbingwg/sriov-network-device-plugin}
         export NETWORK_RESOURCES_INJECTOR_IMAGE=${NETWORK_RESOURCES_INJECTOR_IMAGE:-ghcr.io/k8snetworkplumbingwg/network-resources-injector}
         export SRIOV_NETWORK_CONFIG_DAEMON_IMAGE=${SRIOV_NETWORK_CONFIG_DAEMON_IMAGE:-ghcr.io/k8snetworkplumbingwg/sriov-network-operator-config-daemon}
@@ -13,6 +15,8 @@ if [ -z $SKIP_VAR_SET ]; then
 else
         # ensure that OVS_CNI_IMAGE is set, empty string is a valid value
         OVS_CNI_IMAGE=${OVS_CNI_IMAGE:-}
+        # ensure that RDMA_CNI_IMAGE is set, empty string is a valid value
+        RDMA_CNI_IMAGE=${$RDMA_CNI_IMAGE:-}
         METRICS_EXPORTER_KUBE_RBAC_PROXY_IMAGE=${METRICS_EXPORTER_KUBE_RBAC_PROXY_IMAGE:-}
         [ -z $SRIOV_CNI_IMAGE ] && echo "SRIOV_CNI_IMAGE is empty but SKIP_VAR_SET is set" && exit 1
         [ -z $SRIOV_INFINIBAND_CNI_IMAGE ] && echo "SRIOV_INFINIBAND_CNI_IMAGE is empty but SKIP_VAR_SET is set" && exit 1

diff --git a/hack/release/chart-update.sh b/hack/release/chart-update.sh
@@ -45,6 +45,7 @@ OPERATOR_TAG=${GITHUB_TAG}
 IB_SRIOV_CNI_TAG=$(get_latest_github_tag k8snetworkplumbingwg ib-sriov-cni)
 SRIOV_CNI_TAG=$(get_latest_github_tag k8snetworkplumbingwg sriov-cni)
 OVS_CNI_TAG=$(get_latest_github_tag k8snetworkplumbingwg ovs-cni)
+RDMA_CNI_TAG=$(get_latest_github_tag k8snetworkplumbingwg rdma-cni)
 NETWORK_RESOURCE_INJECTOR_TAG=$(get_latest_github_tag k8snetworkplumbingwg network-resources-injector)
 SRIOV_DEVICE_PLUGIN_TAG=$(get_latest_github_tag k8snetworkplumbingwg sriov-network-device-plugin)
 METRICS_EXPORTER_TAG=$(get_latest_github_tag k8snetworkplumbingwg sriov-network-metrics-exporter)
@@ -61,6 +62,7 @@ $YQ_CMD -i ".images.webhook = \"ghcr.io/${OPERATOR_REPO}/sriov-network-operator-
 $YQ_CMD -i ".images.sriovCni = \"ghcr.io/k8snetworkplumbingwg/sriov-cni:${SRIOV_CNI_TAG}\"" ${HELM_VALUES}
 $YQ_CMD -i ".images.ibSriovCni = \"ghcr.io/k8snetworkplumbingwg/ib-sriov-cni:${IB_SRIOV_CNI_TAG}\"" ${HELM_VALUES}
 $YQ_CMD -i ".images.ovsCni = \"ghcr.io/k8snetworkplumbingwg/ovs-cni-plugin:${OVS_CNI_TAG}\"" ${HELM_VALUES}
+$YQ_CMD -i ".images.rdmaCni = \"ghcr.io/k8snetworkplumbingwg/rdma-cni:${RDMA_CNI_TAG}\"" ${HELM_VALUES}
 $YQ_CMD -i ".images.sriovDevicePlugin = \"ghcr.io/k8snetworkplumbingwg/sriov-network-device-plugin:${SRIOV_DEVICE_PLUGIN_TAG}\"" ${HELM_VALUES}
 $YQ_CMD -i ".images.resourcesInjector = \"ghcr.io/k8snetworkplumbingwg/network-resources-injector:${NETWORK_RESOURCE_INJECTOR_TAG}\"" ${HELM_VALUES}
 $YQ_CMD -i ".images.metricsExporter = \"ghcr.io/k8snetworkplumbingwg/sriov-network-metrics-exporter:${METRICS_EXPORTER_TAG}\"" ${HELM_VALUES}

diff --git a/hack/run-e2e-conformance-virtual-ocp.sh b/hack/run-e2e-conformance-virtual-ocp.sh
@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 set -xeo pipefail
 
-OCP_VERSION=${OCP_VERSION:-4.16.0-rc.2}
+OCP_VERSION=${OCP_VERSION:-4.16.0}
 cluster_name=${CLUSTER_NAME:-ocp-virt}
 domain_name=lab
 
@@ -237,7 +237,11 @@ echo ${auth} > registry-login.conf
 internal_registry="image-registry.openshift-image-registry.svc:5000"
 pass=$( jq .\"image-registry.openshift-image-registry.svc:5000\".auth registry-login.conf  )
 pass=`echo ${pass:1:-1} | base64 -d`
-podman login -u serviceaccount -p ${pass:15} $registry --tls-verify=false
+
+# dockercfg password is in the form `<token>:password`. We need to trim the `<token>:` prefix
+pass=${pass#"<token>:"}
+
+podman login -u serviceaccount -p ${pass} $registry --tls-verify=false
 
 MAX_RETRIES=20
 DELAY_SECONDS=10

diff --git a/hack/run-e2e-test.sh b/hack/run-e2e-test.sh
@@ -8,6 +8,7 @@ KUBECONFIG=${KUBECONFIG:-/root/env/ign/auth/kubeconfig}
 echo ${SRIOV_CNI_IMAGE}
 echo ${SRIOV_INFINIBAND_CNI_IMAGE}
 echo ${OVS_CNI_IMAGE}
+echo ${RDMA_CNI_IMAGE}
 echo ${SRIOV_DEVICE_PLUGIN_IMAGE}
 echo ${NETWORK_RESOURCES_INJECTOR_IMAGE}
 echo ${SRIOV_NETWORK_CONFIG_DAEMON_IMAGE}

diff --git a/pkg/helper/mock/mock_helper.go b/pkg/helper/mock/mock_helper.go
diff --git a/pkg/host/internal/sriov/sriov.go b/pkg/host/internal/sriov/sriov.go
@@ -817,6 +817,13 @@ func (s *sriov) checkForConfigAndReset(ifaceStatus sriovnetworkv1.InterfaceExt,
 		log.Log.V(2).Info("checkForConfigAndReset(): PF name with pci address was externally created skipping the device reset",
 			"pf-name", ifaceStatus.Name,
 			"address", ifaceStatus.PciAddress)
+
+		// remove pf status from host
+		err = storeManager.RemovePfAppliedStatus(ifaceStatus.PciAddress)
+		if err != nil {
+			return err
+		}
+
 		return nil
 	}
 	err = s.removeUdevRules(ifaceStatus.PciAddress)
@@ -828,6 +835,12 @@ func (s *sriov) checkForConfigAndReset(ifaceStatus sriovnetworkv1.InterfaceExt,
 		return err
 	}
 
+	// remove pf status from host
+	err = storeManager.RemovePfAppliedStatus(ifaceStatus.PciAddress)
+	if err != nil {
+		return err
+	}
+
 	return nil
 }
 

diff --git a/pkg/host/internal/sriov/sriov_test.go b/pkg/host/internal/sriov/sriov_test.go
@@ -452,6 +452,7 @@ var _ = Describe("SRIOV", func() {
 				PciAddress: "0000:d8:00.0",
 				NumVfs:     2,
 			}, true, nil)
+			storeManagerMode.EXPECT().RemovePfAppliedStatus("0000:d8:00.0").Return(nil)
 			netlinkLibMock.EXPECT().DevLinkGetDeviceByName("pci", "0000:d8:00.0").Return(
 				&netlink.DevlinkDevice{Attrs: netlink.DevlinkDevAttrs{Eswitch: netlink.DevlinkDevEswitchAttr{Mode: "legacy"}}},
 				nil)
@@ -479,6 +480,7 @@ var _ = Describe("SRIOV", func() {
 				NumVfs:            2,
 				ExternallyManaged: true,
 			}, true, nil)
+			storeManagerMode.EXPECT().RemovePfAppliedStatus("0000:d8:00.0").Return(nil)
 			Expect(s.ConfigSriovInterfaces(storeManagerMode,
 				[]sriovnetworkv1.Interface{},
 				[]sriovnetworkv1.InterfaceExt{

diff --git a/pkg/host/store/mock/mock_store.go b/pkg/host/store/mock/mock_store.go
diff --git a/pkg/host/store/store.go b/pkg/host/store/store.go
@@ -20,6 +20,7 @@ import (
 type ManagerInterface interface {
 	ClearPCIAddressFolder() error
 	SaveLastPfAppliedStatus(PfInfo *sriovnetworkv1.Interface) error
+	RemovePfAppliedStatus(pciAddress string) error
 	LoadPfsStatus(pciAddress string) (*sriovnetworkv1.Interface, bool, error)
 
 	GetCheckPointNodeState() (*sriovnetworkv1.SriovNetworkNodeState, error)
@@ -111,6 +112,17 @@ func (s *manager) SaveLastPfAppliedStatus(PfInfo *sriovnetworkv1.Interface) erro
 	return err
 }
 
+func (s *manager) RemovePfAppliedStatus(pciAddress string) error {
+	hostExtension := utils.GetHostExtension()
+	pathFile := filepath.Join(hostExtension, consts.PfAppliedConfig, pciAddress)
+	err := os.RemoveAll(pathFile)
+	if err != nil {
+		log.Log.Error(err, "failed to remove PF status", "pathFile", pathFile)
+		return err
+	}
+	return nil
+}
+
 // LoadPfsStatus convert the /etc/sriov-operator/pci/<pci-address> json to pfstatus
 // returns false if the file doesn't exist.
 func (s *manager) LoadPfsStatus(pciAddress string) (*sriovnetworkv1.Interface, bool, error) {

diff --git a/pkg/plugins/mellanox/mellanox_plugin.go b/pkg/plugins/mellanox/mellanox_plugin.go
@@ -143,6 +143,11 @@ func (p *MellanoxPlugin) OnNodeStateChange(new *sriovnetworkv1.SriovNetworkNodeS
 		processedNics[pciPrefix] = true
 		pciAddress := pciPrefix + "0"
 
+		// Skip devices not configured by the operator
+		if p.nicNotConfiguredByOperator(portsMap) {
+			continue
+		}
+
 		// Skip externally managed NICs
 		if p.nicHasExternallyManagedPFs(portsMap) {
 			continue
@@ -206,3 +211,19 @@ func (p *MellanoxPlugin) nicHasExternallyManagedPFs(nicPortsMap map[string]sriov
 	}
 	return false
 }
+
+// nicNotConfiguredByOperator returns true if one of the ports(interface) of the NIC is not configured by operator
+func (p *MellanoxPlugin) nicNotConfiguredByOperator(nicPortsMap map[string]sriovnetworkv1.InterfaceExt) bool {
+	for _, iface := range nicPortsMap {
+		_, exist, err := p.helpers.LoadPfsStatus(iface.PciAddress)
+		if err != nil {
+			log.Log.Error(err, "failed to load PF status from disk", "address", iface.PciAddress)
+			continue
+		}
+		if exist {
+			log.Log.V(2).Info("PF configured by the operator", "interface", iface)
+			return true
+		}
+	}
+	return false
+}