Fix bug: networking between longhorn-csi-plugin and longhorn-manager …

…is broken after upgrading Longhorn to 1.7.0-rc3

1. Revert to set `HostPID: true` and mounting the host proc for CSI plugin pod
2. Revert to set `HostNetwork: false`
3. When RWX storage network is enabled, we mount NFS in the container namespace.
   When it is not we mount the NFS in host name space by nsenter the host namespace

longhorn-9223

Signed-off-by: Phan Le <phan.le@suse.com>

csi/deployment.go

@@ -307,8 +307,7 @@ func NewPluginDeployment(namespace, serviceAccount, nodeDriverRegistrarImage, li
 					Tolerations:        tolerations,
 					NodeSelector:       nodeSelector,
 					PriorityClassName:  priorityClass,
-					HostNetwork:        true,
-					DNSPolicy:          corev1.DNSClusterFirstWithHostNet,
+					HostPID:            true,
 					Containers: []corev1.Container{
 						{
 							Name:  "node-driver-registrar",
@@ -457,6 +456,11 @@ func NewPluginDeployment(namespace, serviceAccount, nodeDriverRegistrarImage, li
 									Name:      "host-sys",
 									MountPath: "/sys",
 								},
+								{
+									Name:             "host",
+									MountPath:        "/host",
+									MountPropagation: &MountPropagationBidirectional,
+								},
 								{
 									Name:      "lib-modules",
 									MountPath: "/lib/modules",
@@ -526,6 +530,14 @@ func NewPluginDeployment(namespace, serviceAccount, nodeDriverRegistrarImage, li
 								},
 							},
 						},
+						{
+							Name: "host",
+							VolumeSource: corev1.VolumeSource{
+								HostPath: &corev1.HostPathVolumeSource{
+									Path: "/",
+								},
+							},
+						},
 						{
 							Name: "lib-modules",
 							VolumeSource: corev1.VolumeSource{

csi/node_server.go

@@ -812,11 +812,51 @@ func getNodeServiceCapabilities(cs []csi.NodeServiceCapability_RPC_Type) []*csi.
 	return nscs
 }
 
+func (ns *NodeServer) requireHostNamespaceMounter(volume *longhornclient.Volume, volumeCapability *csi.VolumeCapability) (bool, error) {
+	if !requiresSharedAccess(volume, volumeCapability) {
+		return false, nil
+	}
+	if volume.Migratable {
+		return false, nil
+	}
+
+	storageNetworkSetting, err := ns.apiClient.Setting.ById(string(types.SettingNameStorageNetwork))
+	if err != nil {
+		return false, err
+	}
+	if storageNetworkSetting.Value == "" {
+		return false, nil
+	}
+
+	storageNetworkForRWXVolumeEnabledSetting, err := ns.apiClient.Setting.ById(string(types.SettingNameStorageNetworkForRWXVolumeEnabled))
+	if err != nil {
+		return false, err
+	}
+	storageNetworkForRWXVolumeEnabled, err := strconv.ParseBool(storageNetworkForRWXVolumeEnabledSetting.Value)
+	if err != nil {
+		return false, err
+	}
+	if !storageNetworkForRWXVolumeEnabled {
+		return false, nil
+	}
+
+	return true, nil
+}
+
 func (ns *NodeServer) getMounter(volume *longhornclient.Volume, volumeCapability *csi.VolumeCapability, volumeContext map[string]string) (mount.Interface, error) {
 	if volumeCapability.GetBlock() != nil {
 		return mount.New(""), nil
 	}
 
+	// HACK: to nsenter host namespaces for the nfs mounts to stay available after csi plugin dies
+	requireHostNamespaceMounter, err := ns.requireHostNamespaceMounter(volume, volumeCapability)
+	if err != nil {
+		return nil, err
+	}
+	if requireHostNamespaceMounter {
+		return mount.New("/usr/local/sbin/nsmounter"), nil
+	}
+
 	// mounter that can format and use hard coded filesystem params
 	if volumeCapability.GetMount() != nil {
 		fsType := volumeCapability.GetMount().GetFsType()

package/Dockerfile

@@ -12,7 +12,7 @@ ENV ARCH ${TARGETPLATFORM#linux/}
 
 COPY package/bin/longhorn-manager-${ARCH} /usr/local/sbin/longhorn-manager
 
-COPY package/launch-manager /usr/local/sbin/
+COPY package/launch-manager package/nsmounter /usr/local/sbin/
 
 RUN zypper -n ref && \
     zypper update -y

package/nsmounter

@@ -0,0 +1,46 @@
+#!/bin/bash
+
+PROC_DIR="/proc"
+
+os_distro_talos="talos"
+os_distro=""
+
+get_os_distro() {
+  local version_info=$(< $PROC_DIR/version)
+
+  [[ $version_info =~ $os_distro_talos ]] && os_distro=$os_distro_talos
+}
+
+target_pid=1
+
+get_pid() {
+  local process_name=$1
+  local pid
+  local status_file
+  local name
+
+  for dir in $PROC_DIR/*/; do
+    pid=$(basename "$dir")
+    status_file="$PROC_DIR/$pid/status"
+
+    if [ -f "$status_file" ]; then
+      while IFS= read -r line; do
+        if [[ $line == "Name:"* ]]; then
+          name="${line#*:}"
+          name="${name//[$'\t ']/}"  # Remove both spaces and tabs
+          break  # Exit the loop once the Name is found
+        fi
+      done < "$status_file"
+    fi
+
+    if [ "$name" = "$process_name" ]; then
+      target_pid=$pid
+    fi
+  done
+}
+
+get_os_distro
+
+[[ $os_distro = $os_distro_talos ]] && get_pid "kubelet"
+
+nsenter -t $target_pid -m -n -u -- "$@"

types/deploy.go

@@ -104,15 +104,6 @@ func UpdateDaemonSetTemplateBasedOnStorageNetwork(daemonSet *appsv1.DaemonSet, s
 
 	isContainerNetworkNamespace := IsStorageNetworkForRWXVolume(storageNetwork, isStorageNetworkForRWXVolumeEnabled)
 
-	updateHostNetwork := func() {
-		newHostNetwork := !isContainerNetworkNamespace
-		logger.WithFields(logrus.Fields{
-			"oldValue": daemonSet.Spec.Template.Spec.HostNetwork,
-			"newValue": newHostNetwork,
-		}).Debugf("Updating hostNetwork")
-		daemonSet.Spec.Template.Spec.HostNetwork = newHostNetwork
-	}
-
 	updateAnnotation := func() {
 		annotKey := string(CNIAnnotationNetworks)
 		annotValue := ""
@@ -137,6 +128,5 @@ func UpdateDaemonSetTemplateBasedOnStorageNetwork(daemonSet *appsv1.DaemonSet, s
 		}
 	}
 
-	updateHostNetwork()
 	updateAnnotation()
 }