Implement RDMA subsystem mode change

Now it's possible to configure RDMA subsystem mode using
SR-IOV Network Operator in systemd mode

We can't configure RDMA subsystem in a daemon mode because
it should be done on host before any network namespace is
created.

api/v1/sriovnetworknodestate_types.go

@@ -23,9 +23,14 @@ import (
 // EDIT THIS FILE!  THIS IS SCAFFOLDING FOR YOU TO OWN!
 // NOTE: json tags are required.  Any new fields you add must have json tags for the fields to be serialized.
 
+type System struct {
+	RdmaMode string `json:"rdmaMode,omitempty"`
+}
+
 // SriovNetworkNodeStateSpec defines the desired state of SriovNetworkNodeState
 type SriovNetworkNodeStateSpec struct {
 	Interfaces Interfaces `json:"interfaces,omitempty"`
+	System     System     `json:"system,omitempty"`
 }
 
 type Interfaces []Interface
@@ -87,6 +92,7 @@ type VirtualFunction struct {
 // SriovNetworkNodeStateStatus defines the observed state of SriovNetworkNodeState
 type SriovNetworkNodeStateStatus struct {
 	Interfaces    InterfaceExts `json:"interfaces,omitempty"`
+	System        System        `json:"system,omitempty"`
 	SyncStatus    string        `json:"syncStatus,omitempty"`
 	LastSyncError string        `json:"lastSyncError,omitempty"`
 }

api/v1/sriovnetworkpoolconfig_types.go

@@ -21,6 +21,10 @@ type SriovNetworkPoolConfigSpec struct {
 	// Drain will respect Pod Disruption Budgets (PDBs) such as etcd quorum guards,
 	// even if maxUnavailable is greater than one.
 	MaxUnavailable *intstr.IntOrString `json:"maxUnavailable,omitempty"`
+
+	// +kubebuilder:validation:Enum=shared;exclusive
+	// RDMA subsystem. Allowed value "shared", "exclusive".
+	RdmaMode string `json:"rdmaMode,omitempty"`
 }
 
 type OvsHardwareOffloadConfig struct {

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

@@ -22,6 +22,7 @@ import (
 
 	"github.com/go-logr/logr"
 	"github.com/spf13/cobra"
+	"github.com/vishvananda/netlink"
 	"sigs.k8s.io/controller-runtime/pkg/log"
 
 	sriovv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/api/v1"
@@ -152,6 +153,21 @@ func phasePre(setupLog logr.Logger, conf *systemd.SriovConfig, hostHelpers helpe
 	hostHelpers.TryEnableTun()
 	hostHelpers.TryEnableVhostNet()
 
+	if conf.Spec.System.RdmaMode != "" {
+		rdmaSubsystem, err := netlink.RdmaSystemGetNetnsMode()
+		if err != nil {
+			setupLog.Error(err, "failed to get RDMA subsystem mode")
+			return fmt.Errorf("failed to get RDMA subsystem mode: %v", err)
+		}
+		if rdmaSubsystem != conf.Spec.System.RdmaMode {
+			err = netlink.RdmaSystemSetNetnsMode(conf.Spec.System.RdmaMode)
+			if err != nil {
+				setupLog.Error(err, "failed to set RDMA subsystem mode")
+				return fmt.Errorf("failed to set RDMA subsystem mode: %v", err)
+			}
+		}
+	}
+
 	return callPlugin(setupLog, PhasePre, conf, hostHelpers)
 }
 

config/crd/bases/sriovnetwork.openshift.io_sriovnetworknodestates.yaml

@@ -89,6 +89,11 @@ spec:
                   - pciAddress
                   type: object
                 type: array
+              system:
+                properties:
+                  rdmaMode:
+                    type: string
+                type: object
             type: object
           status:
             description: SriovNetworkNodeStateStatus defines the observed state of
@@ -163,6 +168,11 @@ spec:
                 type: string
               syncStatus:
                 type: string
+              system:
+                properties:
+                  rdmaMode:
+                    type: string
+                type: object
             type: object
         type: object
     served: true

config/crd/bases/sriovnetwork.openshift.io_sriovnetworkpoolconfigs.yaml

@@ -103,6 +103,12 @@ spec:
                       offload'
                     type: string
                 type: object
+              rdmaMode:
+                description: RDMA subsystem. Allowed value "shared", "exclusive".
+                enum:
+                - shared
+                - exclusive
+                type: string
             type: object
           status:
             description: SriovNetworkPoolConfigStatus defines the observed state of

controllers/drain_controller.go

@@ -24,11 +24,8 @@ import (
 
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
-	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/apimachinery/pkg/types"
-	"k8s.io/apimachinery/pkg/util/intstr"
 	"k8s.io/client-go/tools/record"
 	"k8s.io/client-go/util/workqueue"
 	ctrl "sigs.k8s.io/controller-runtime"
@@ -48,13 +45,6 @@ import (
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/vars"
 )
 
-var (
-	oneNode     = intstr.FromInt32(1)
-	defaultNpcl = &sriovnetworkv1.SriovNetworkPoolConfig{Spec: sriovnetworkv1.SriovNetworkPoolConfigSpec{
-		MaxUnavailable: &oneNode,
-		NodeSelector:   &metav1.LabelSelector{}}}
-)
-
 type DrainReconcile struct {
 	client.Client
 	Scheme   *runtime.Scheme
@@ -346,94 +336,7 @@ func (dr *DrainReconcile) tryDrainNode(ctx context.Context, node *corev1.Node) (
 }
 
 func (dr *DrainReconcile) findNodePoolConfig(ctx context.Context, node *corev1.Node) (*sriovnetworkv1.SriovNetworkPoolConfig, []corev1.Node, error) {
-	logger := log.FromContext(ctx)
-	logger.Info("findNodePoolConfig():")
-	// get all the sriov network pool configs
-	npcl := &sriovnetworkv1.SriovNetworkPoolConfigList{}
-	err := dr.List(ctx, npcl)
-	if err != nil {
-		logger.Error(err, "failed to list sriovNetworkPoolConfig")
-		return nil, nil, err
-	}
-
-	selectedNpcl := []*sriovnetworkv1.SriovNetworkPoolConfig{}
-	nodesInPools := map[string]interface{}{}
-
-	for _, npc := range npcl.Items {
-		// we skip hw offload objects
-		if npc.Spec.OvsHardwareOffloadConfig.Name != "" {
-			continue
-		}
-
-		if npc.Spec.NodeSelector == nil {
-			npc.Spec.NodeSelector = &metav1.LabelSelector{}
-		}
-
-		selector, err := metav1.LabelSelectorAsSelector(npc.Spec.NodeSelector)
-		if err != nil {
-			logger.Error(err, "failed to create label selector from nodeSelector", "nodeSelector", npc.Spec.NodeSelector)
-			return nil, nil, err
-		}
-
-		if selector.Matches(labels.Set(node.Labels)) {
-			selectedNpcl = append(selectedNpcl, npc.DeepCopy())
-		}
-
-		nodeList := &corev1.NodeList{}
-		err = dr.List(ctx, nodeList, &client.ListOptions{LabelSelector: selector})
-		if err != nil {
-			logger.Error(err, "failed to list all the nodes matching the pool with label selector from nodeSelector",
-				"machineConfigPoolName", npc,
-				"nodeSelector", npc.Spec.NodeSelector)
-			return nil, nil, err
-		}
-
-		for _, nodeName := range nodeList.Items {
-			nodesInPools[nodeName.Name] = nil
-		}
-	}
-
-	if len(selectedNpcl) > 1 {
-		// don't allow the node to be part of multiple pools
-		err = fmt.Errorf("node is part of more then one pool")
-		logger.Error(err, "multiple pools founded for a specific node", "numberOfPools", len(selectedNpcl), "pools", selectedNpcl)
-		return nil, nil, err
-	} else if len(selectedNpcl) == 1 {
-		// found one pool for our node
-		logger.V(2).Info("found sriovNetworkPool", "pool", *selectedNpcl[0])
-		selector, err := metav1.LabelSelectorAsSelector(selectedNpcl[0].Spec.NodeSelector)
-		if err != nil {
-			logger.Error(err, "failed to create label selector from nodeSelector", "nodeSelector", selectedNpcl[0].Spec.NodeSelector)
-			return nil, nil, err
-		}
-
-		// list all the nodes that are also part of this pool and return them
-		nodeList := &corev1.NodeList{}
-		err = dr.List(ctx, nodeList, &client.ListOptions{LabelSelector: selector})
-		if err != nil {
-			logger.Error(err, "failed to list nodes using with label selector", "labelSelector", selector)
-			return nil, nil, err
-		}
-
-		return selectedNpcl[0], nodeList.Items, nil
-	} else {
-		// in this case we get all the nodes and remove the ones that already part of any pool
-		logger.V(1).Info("node doesn't belong to any pool, using default drain configuration with MaxUnavailable of one", "pool", *defaultNpcl)
-		nodeList := &corev1.NodeList{}
-		err = dr.List(ctx, nodeList)
-		if err != nil {
-			logger.Error(err, "failed to list all the nodes")
-			return nil, nil, err
-		}
-
-		defaultNodeLists := []corev1.Node{}
-		for _, nodeObj := range nodeList.Items {
-			if _, exist := nodesInPools[nodeObj.Name]; !exist {
-				defaultNodeLists = append(defaultNodeLists, nodeObj)
-			}
-		}
-		return defaultNpcl, defaultNodeLists, nil
-	}
+	return utils.FindNodePoolConfig(ctx, node, dr.Client)
 }
 
 // SetupWithManager sets up the controller with the Manager.

controllers/sriovnetworknodepolicy_controller.go

@@ -46,6 +46,7 @@ import (
 	sriovnetworkv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/api/v1"
 	constants "github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/consts"
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/featuregate"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/utils"
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/vars"
 )
 
@@ -269,6 +270,13 @@ func (r *SriovNetworkNodePolicyReconciler) syncAllSriovNetworkNodeStates(ctx con
 		ns.Name = node.Name
 		ns.Namespace = vars.Namespace
 		j, _ := json.Marshal(ns)
+		netPoolConfig, _, err := utils.FindNodePoolConfig(context.Background(), &node, r.Client)
+		if err != nil {
+			log.Log.Error(err, "nodeStateSyncHandler(): failed to get SriovNetworkPoolConfig for the current node")
+		}
+		if netPoolConfig != nil {
+			ns.Spec.System.RdmaMode = netPoolConfig.Spec.RdmaMode
+		}
 		logger.V(2).Info("SriovNetworkNodeState CR", "content", j)
 		if err := r.syncSriovNetworkNodeState(ctx, dc, npl, ns, &node); err != nil {
 			logger.Error(err, "Fail to sync", "SriovNetworkNodeState", ns.Name)

controllers/sriovnetworkpoolconfig_controller.go

@@ -73,6 +73,19 @@ func (r *SriovNetworkPoolConfigReconciler) Reconcile(ctx context.Context, req ct
 		return reconcile.Result{}, err
 	}
 
+	// RdmaMode could be set in systemd mode only
+	if instance.Spec.RdmaMode != "" {
+		operatorConfig := &sriovnetworkv1.SriovOperatorConfig{}
+		err := r.Get(ctx, types.NamespacedName{Namespace: vars.Namespace, Name: constants.DefaultConfigName}, operatorConfig)
+		if err != nil {
+			logger.Error(err, "failed to list SriovOperatorConfig")
+			return reconcile.Result{}, err
+		}
+		if operatorConfig.Spec.ConfigurationMode == sriovnetworkv1.DaemonConfigurationMode {
+			logger.Info("rdmaSpec is ignored in 'daemon' configuration mode")
+		}
+	}
+
 	// we don't need a finalizer for pools that doesn't use the ovs hardware offload feature
 	if instance.Spec.OvsHardwareOffloadConfig.Name == "" {
 		return ctrl.Result{}, nil

deployment/sriov-network-operator/crds/sriovnetwork.openshift.io_sriovnetworknodestates.yaml

@@ -89,6 +89,11 @@ spec:
                   - pciAddress
                   type: object
                 type: array
+              system:
+                properties:
+                  rdmaMode:
+                    type: string
+                type: object
             type: object
           status:
             description: SriovNetworkNodeStateStatus defines the observed state of
@@ -163,6 +168,11 @@ spec:
                 type: string
               syncStatus:
                 type: string
+              system:
+                properties:
+                  rdmaMode:
+                    type: string
+                type: object
             type: object
         type: object
     served: true

deployment/sriov-network-operator/crds/sriovnetwork.openshift.io_sriovnetworkpoolconfigs.yaml

@@ -103,6 +103,12 @@ spec:
                       offload'
                     type: string
                 type: object
+              rdmaMode:
+                description: RDMA subsystem. Allowed value "shared", "exclusive".
+                enum:
+                - shared
+                - exclusive
+                type: string
             type: object
           status:
             description: SriovNetworkPoolConfigStatus defines the observed state of

deployment/sriov-network-operator/templates/clusterrole.yaml

@@ -58,3 +58,6 @@ rules:
   - apiGroups: [ "config.openshift.io" ]
     resources: [ "infrastructures" ]
     verbs: [ "get", "list", "watch" ]
+  - apiGroups: [ "sriovnetwork.openshift.io" ]
+    resources: [ "sriovnetworkpoolconfigs" ]
+    verbs: [ "get", "list", "watch" ]

deployment/sriov-network-operator/templates/operator.yaml

@@ -59,7 +59,7 @@ spec:
             - name: OVS_CNI_IMAGE
               value: {{ .Values.images.ovsCni }}
             - name: RDMA_CNI_IMAGE
-              value: { { .Values.images.rdmaCni } }
+              value: {{ .Values.images.rdmaCni }}
             - name: SRIOV_DEVICE_PLUGIN_IMAGE
               value: {{ .Values.images.sriovDevicePlugin }}
             - name: NETWORK_RESOURCES_INJECTOR_IMAGE

pkg/consts/constants.go

@@ -48,6 +48,9 @@ const (
 	VdpaTypeVirtio      = "virtio"
 	VdpaTypeVhost       = "vhost"
 
+	RdmaSubsystemModeShared    = "shared"
+	RdmaSubsystemModeExclusive = "exclusive"
+
 	ClusterTypeOpenshift  = "openshift"
 	ClusterTypeKubernetes = "kubernetes"
 

pkg/daemon/daemon.go

@@ -9,6 +9,7 @@ import (
 	"time"
 
 	"golang.org/x/time/rate"
+	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
@@ -420,6 +421,13 @@ func (dn *Daemon) nodeStateSyncHandler() error {
 	// When using systemd configuration we write the file
 	if vars.UsingSystemdMode {
 		log.Log.V(0).Info("nodeStateSyncHandler(): writing systemd config file to host")
+		// get node object
+		node := &corev1.Node{}
+		err := dn.client.Get(context.TODO(), client.ObjectKey{Name: vars.NodeName}, node)
+		if err != nil {
+			log.Log.Error(err, "nodeStateSyncHandler(): failed to get node object")
+			return err
+		}
 		systemdConfModified, err := systemd.WriteConfFile(dn.desiredNodeState)
 		if err != nil {
 			log.Log.Error(err, "nodeStateSyncHandler(): failed to write configuration file for systemd mode")