Merge pull request openshift#1069 from SchSeba/merge-bot-master

Merge https://github.com/k8snetworkplumbingwg/sriov-network-operator:master into main

Author: openshift-merge-bot[bot]

bundle/manifests/sriov-network-operator.clusterserviceversion.yaml

@@ -100,7 +100,7 @@ metadata:
     categories: Networking
     certified: "false"
     containerImage: quay.io/openshift/origin-sriov-network-operator:4.19
-    createdAt: "2025-03-15T23:49:11Z"
+    createdAt: "2025-03-23T23:49:34Z"
     description: An operator for configuring SR-IOV components and initializing SRIOV
       network devices in Openshift cluster.
     features.operators.openshift.io/cnf: "false"

cmd/sriov-network-config-daemon/service.go

@@ -368,6 +368,7 @@ func (s *ServiceConfig) waitForDevicesInitialization() {
 	for time.Now().Before(deadline) {
 		for pciAddr, name := range devicesToWait {
 			if s.hostHelper.TryGetInterfaceName(pciAddr) == name {
+				s.log.Info("Device ready", "pci", pciAddr, "name", name)
 				delete(devicesToWait, pciAddr)
 			}
 		}

manifests/stable/sriov-network-operator.clusterserviceversion.yaml

@@ -100,7 +100,7 @@ metadata:
     categories: Networking
     certified: "false"
     containerImage: quay.io/openshift/origin-sriov-network-operator:4.19
-    createdAt: "2025-03-15T23:49:11Z"
+    createdAt: "2025-03-23T23:49:34Z"
     description: An operator for configuring SR-IOV components and initializing SRIOV
       network devices in Openshift cluster.
     features.operators.openshift.io/cnf: "false"

pkg/host/internal/network/network.go

@@ -75,7 +75,6 @@ func (n *network) TryToGetVirtualInterfaceName(pciAddr string) string {
 func (n *network) TryGetInterfaceName(pciAddr string) string {
 	names, err := n.dputilsLib.GetNetNames(pciAddr)
 	if err != nil || len(names) < 1 {
-		log.Log.V(2).Info("TryGetInterfaceName(): failed to get interface name", "err", err, "pciAddress", pciAddr)
 		return ""
 	}
 	netDevName := names[0]
@@ -96,7 +95,6 @@ func (n *network) TryGetInterfaceName(pciAddr string) string {
 		return name
 	}
 
-	log.Log.V(2).Info("TryGetInterfaceName()", "name", netDevName)
 	return netDevName
 }
 

pkg/host/internal/sriov/sriov.go

@@ -176,7 +176,7 @@ func (s *sriov) VFIsReady(pciAddr string) (netlink.Link, error) {
 	err = wait.PollImmediate(time.Second, 10*time.Second, func() (bool, error) {
 		vfIndex, err := s.networkHelper.GetInterfaceIndex(pciAddr)
 		if err != nil {
-			log.Log.Error(err, "VFIsReady(): invalid index number")
+			log.Log.Error(err, "VFIsReady(): invalid index number", "device", pciAddr)
 			return false, nil
 		}
 		vfLink, err = s.netlinkLib.LinkByIndex(vfIndex)
@@ -658,7 +658,7 @@ func (s *sriov) getConfigureAndReset(storeManager store.ManagerInterface, interf
 			}
 		}
 
-		if !configured && ifaceStatus.NumVfs > 0 {
+		if !configured {
 			toBeResetted = append(toBeResetted, ifaceStatus)
 		}
 	}
@@ -824,8 +824,10 @@ func (s *sriov) checkForConfigAndReset(ifaceStatus sriovnetworkv1.InterfaceExt,
 		return err
 	}
 
-	if err = s.ResetSriovDevice(ifaceStatus); err != nil {
-		return err
+	if ifaceStatus.NumVfs > 0 {
+		if err = s.ResetSriovDevice(ifaceStatus); err != nil {
+			return err
+		}
 	}
 
 	// remove pf status from host

pkg/host/internal/sriov/sriov_test.go

@@ -241,8 +241,13 @@ var _ = Describe("SRIOV", func() {
 			hostMock.EXPECT().HasDriver("0000:d8:00.3").Return(true, "vfio-pci").Times(2)
 			hostMock.EXPECT().UnbindDriverIfNeeded("0000:d8:00.3", false).Return(nil)
 			hostMock.EXPECT().BindDpdkDriver("0000:d8:00.3", "vfio-pci").Return(nil)
+			hostMock.EXPECT().RemoveDisableNMUdevRule("0000:d8:00.1").Return(nil)
+			hostMock.EXPECT().RemoveVfRepresentorUdevRule("0000:d8:00.1").Return(nil)
+			hostMock.EXPECT().RemovePersistPFNameUdevRule("0000:d8:00.1").Return(nil)
 
 			storeManagerMode.EXPECT().SaveLastPfAppliedStatus(gomock.Any()).Return(nil)
+			storeManagerMode.EXPECT().RemovePfAppliedStatus(gomock.Any()).Return(nil)
+			storeManagerMode.EXPECT().LoadPfsStatus("0000:d8:00.1").Return(&sriovnetworkv1.Interface{ExternallyManaged: false}, true, nil)
 
 			Expect(s.ConfigSriovInterfaces(storeManagerMode,
 				[]sriovnetworkv1.Interface{{

test/conformance/tests/test_sriov_operator.go

@@ -1711,6 +1711,54 @@ var _ = Describe("[sriov] operator", Ordered, func() {
 			})
 		})
 
+		Context("Daemon reset with shutdown", Ordered, func() {
+			var testNode string
+			var policy *sriovv1.SriovNetworkNodePolicy
+			resourceName := "resetresource"
+
+			BeforeAll(func() {
+				isSingleNode, err := cluster.IsSingleNode(clients)
+				Expect(err).ToNot(HaveOccurred())
+				if isSingleNode {
+					Skip("test not supported for single node")
+				}
+
+				sriovInfos, err := cluster.DiscoverSriov(clients, operatorNamespace)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(sriovInfos.Nodes)).ToNot(BeZero())
+				testNode = sriovInfos.Nodes[0]
+				iface, err := sriovInfos.FindOneSriovDevice(testNode)
+				Expect(err).ToNot(HaveOccurred())
+
+				policy, err = network.CreateSriovPolicy(clients, "test-policy-", operatorNamespace, iface.Name, testNode, 5, resourceName, "netdevice")
+				Expect(err).ToNot(HaveOccurred())
+				WaitForSRIOVStable()
+			})
+
+			It("should remove interface configuration from host even after reboot", func() {
+				By("force rebooting the node")
+				_, errOutput, err := runCommandOnConfigDaemon(testNode, "chroot", "/host", "reboot")
+				Expect(err).ToNot(HaveOccurred(), errOutput)
+				By("removing the policy")
+				err = clients.Delete(context.Background(), policy)
+				Expect(err).ToNot(HaveOccurred())
+
+				By("waiting for node to be not ready")
+				waitForNodeCondition(testNode, corev1.NodeReady, corev1.ConditionUnknown)
+
+				By("waiting for node to be ready")
+				waitForNodeCondition(testNode, corev1.NodeReady, corev1.ConditionTrue)
+
+				WaitForSRIOVStable()
+				By("Checking files on the host")
+				output, errOutput, err := runCommandOnConfigDaemon(testNode, "/bin/bash", "-c", "ls /host/etc/sriov-operator/pci/ | wc -l")
+				Expect(err).ToNot(HaveOccurred(), errOutput)
+				Expect(strings.HasPrefix(output, "0")).Should(BeTrue())
+				output, errOutput, err = runCommandOnConfigDaemon(testNode, "/bin/bash", "-c", "ls /host/etc/udev/rules.d/ | grep 10-nm-disable | wc -l")
+				Expect(err).ToNot(HaveOccurred(), errOutput)
+				Expect(strings.HasPrefix(output, "0")).Should(BeTrue())
+			})
+		})
 	})
 })
 
@@ -2403,3 +2451,16 @@ func getInterfaceFromNodeStateByPciAddress(node, pciAddress string) *sriovv1.Int
 	Expect(found).To(BeTrue())
 	return intf
 }
+
+func waitForNodeCondition(nodeName string, conditionType corev1.NodeConditionType, conditionStatus corev1.ConditionStatus) {
+	EventuallyWithOffset(1, func(g Gomega) bool {
+		node, err := clients.CoreV1Interface.Nodes().Get(context.Background(), nodeName, metav1.GetOptions{})
+		g.Expect(err).ToNot(HaveOccurred())
+		for _, con := range node.Status.Conditions {
+			if con.Type == conditionType && con.Status == conditionStatus {
+				return true
+			}
+		}
+		return false
+	}, waitingTime, 1*time.Second).Should(BeTrue())
+}