pb-6681: Add support for PSA in kdmp job

- Added check and collected psa info from the namespace
- if psa enabled then extracted the uid and gid used by the POD.
  here the pod is choosen based on whichever PVC is used by that pod
- applied those uid and GID to all relevant job pod spec
- this is done only if the PSA mode enforced with "restricted" value
- for baseline and privilege mode no restriction on UID/GID and
  default setting of SElinux and secomp is adopted in the POD spec

Signed-off-by: Lalatendu Das <[email protected]>

pkg/controllers/dataexport/reconcile.go

@@ -1137,9 +1137,9 @@ func (c *Controller) stageLocalSnapshotRestore(ctx context.Context, dataExport *
 		}
 		// Already done with max retries, so moving to kdmp restore anyway
 		data := updateDataExportDetail{
-			stage:  kdmpapi.DataExportStageTransferScheduled,
-			status: kdmpapi.DataExportStatusInitial,
-			reason: "switching to restore from objectstore bucket as restoring from local snapshot did not happen",
+			stage:                     kdmpapi.DataExportStageTransferScheduled,
+			status:                    kdmpapi.DataExportStatusInitial,
+			reason:                    "switching to restore from objectstore bucket as restoring from local snapshot did not happen",
 			resetLocalSnapshotRestore: true,
 		}
 		logrus.Infof("%v: In stageLocalSnapshotRestore stage, local snapshot restore failed, trying KDMP restore.", dataExport.Name)
@@ -1872,13 +1872,21 @@ func startTransferJob(
 		nfsServerAddr  string
 		nfsExportPath  string
 		nfsMountOption string
+		psaIsEnabled   string
+		psaJobUid      string
+		psaJobGid      string
 	)
 
 	if backupLocation != nil {
 		nfsServerAddr = backupLocation.Location.NFSConfig.ServerAddr
 		nfsExportPath = backupLocation.Location.NFSConfig.SubPath
 		nfsMountOption = backupLocation.Location.NFSConfig.MountOptions
 	}
+	if dataExport != nil {
+		psaIsEnabled = getAnnotationValue(dataExport, utils.PsaEnabledKey)
+		psaJobUid = getAnnotationValue(dataExport, utils.PsaUIDKey)
+		psaJobGid = getAnnotationValue(dataExport, utils.PsaGIDKey)
+	}
 	switch drv.Name() {
 	case drivers.Rsync:
 		return drv.StartJob(
@@ -1927,6 +1935,9 @@ func startTransferJob(
 			drivers.WithNfsServer(nfsServerAddr),
 			drivers.WithNfsExportDir(nfsExportPath),
 			drivers.WithNfsMountOption(nfsMountOption),
+			drivers.WithPsaIsEnabled(psaIsEnabled),
+			drivers.WithPodUserId(psaJobUid),
+			drivers.WithPodGroupId(psaJobGid),
 		)
 	case drivers.KopiaRestore:
 		return drv.StartJob(
@@ -1945,6 +1956,9 @@ func startTransferJob(
 			drivers.WithJobConfigMapNs(jobConfigMapNs),
 			drivers.WithNfsServer(nfsServerAddr),
 			drivers.WithNfsExportDir(nfsExportPath),
+			drivers.WithPsaIsEnabled(psaIsEnabled),
+			drivers.WithPodUserId(psaJobUid),
+			drivers.WithPodGroupId(psaJobGid),
 		)
 	}
 
@@ -2353,6 +2367,17 @@ func startNfsCSIRestoreVolumeJob(
 	bl *storkapi.BackupLocation,
 ) (string, error) {
 
+	var (
+		psaIsEnabled string
+		psaJobUid    string
+		psaJobGid    string
+	)
+
+	if de != nil {
+		psaIsEnabled = getAnnotationValue(de, utils.PsaEnabledKey)
+		psaJobUid = getAnnotationValue(de, utils.PsaUIDKey)
+		psaJobGid = getAnnotationValue(de, utils.PsaGIDKey)
+	}
 	jobName := utils.GetCsiRestoreJobName(drivers.NFSCSIRestore, de.Name)
 	err := utils.CreateNfsSecret(utils.GetCredSecretName(jobName), bl, de.Namespace, nil)
 	if err != nil {
@@ -2371,6 +2396,9 @@ func startNfsCSIRestoreVolumeJob(
 			drivers.WithNfsExportDir(bl.Location.NFSConfig.SubPath),
 			drivers.WithNfsMountOption(bl.Location.NFSConfig.MountOptions),
 			drivers.WithNfsSubPath(bl.Location.Path),
+			drivers.WithPsaIsEnabled(psaIsEnabled),
+			drivers.WithPodUserId(psaJobUid),
+			drivers.WithPodGroupId(psaJobGid),
 		)
 	}
 	return "", fmt.Errorf("unknown driver for nfs csi volume restore: %s", drv.Name())

pkg/controllers/resourceexport/reconcile.go

@@ -38,6 +38,14 @@ type updateResourceExportFields struct {
 	LargeResourceEnabled  bool
 }
 
+func getAnnotationValue(re *kdmpapi.ResourceExport, key string) string {
+	var val string
+	if _, ok := re.Annotations[key]; ok {
+		val = re.Annotations[key]
+	}
+	return val
+}
+
 func (c *Controller) process(ctx context.Context, in *kdmpapi.ResourceExport) (bool, error) {
 	funct := "resourceExport.process"
 	if in == nil {
@@ -400,6 +408,8 @@ func startNfsResourceJob(
 	bl *storkapi.BackupLocation,
 ) (string, error) {
 
+	isPsaEnabled := getAnnotationValue(re, utils.PsaEnabledKey)
+
 	err := utils.CreateNfsSecret(utils.GetCredSecretName(re.Name), bl, re.Namespace, nil)
 	if err != nil {
 		logrus.Errorf("failed to create NFS cred secret: %v", err)
@@ -425,6 +435,7 @@ func startNfsResourceJob(
 			drivers.WithNfsExportDir(bl.Location.NFSConfig.SubPath),
 			drivers.WithJobConfigMap(jobConfigMap),
 			drivers.WithJobConfigMapNs(jobConfigMapNs),
+			drivers.WithPsaIsEnabled(isPsaEnabled),
 		)
 	case drivers.NFSRestore:
 		return drv.StartJob(
@@ -443,6 +454,7 @@ func startNfsResourceJob(
 			drivers.WithNfsExportDir(bl.Location.NFSConfig.SubPath),
 			drivers.WithJobConfigMap(jobConfigMap),
 			drivers.WithJobConfigMapNs(jobConfigMapNs),
+			drivers.WithPsaIsEnabled(isPsaEnabled),
 		)
 	}
 	return "", fmt.Errorf("unknown data transfer driver: %s", drv.Name())

pkg/drivers/kopiabackup/kopiabackup.go

@@ -318,6 +318,7 @@ func jobFor(
 		logrus.Errorf("failed to get the toleration details: %v", err)
 		return nil, fmt.Errorf("failed to get the toleration details for job [%s/%s]", jobOption.Namespace, jobName)
 	}
+
 	job := &batchv1.Job{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      jobName,
@@ -384,6 +385,13 @@ func jobFor(
 			},
 		},
 	}
+	// Add security Context only if the PSA is enabled.
+	if jobOption.PsaIsEnabled == "true" {
+		job, err = utils.AddSecurityContextToJob(job, jobOption.PodUserId, jobOption.PodGroupId)
+		if err != nil {
+			return nil, err
+		}
+	}
 
 	if len(nodeName) != 0 {
 		job.Spec.Template.Spec.NodeName = nodeName

pkg/drivers/kopiarestore/kopiarestore.go

@@ -219,6 +219,7 @@ func jobFor(
 		logrus.Errorf("failed to get the toleration details: %v", err)
 		return nil, fmt.Errorf("failed to get the toleration details for job [%s/%s]", jobOption.Namespace, jobName)
 	}
+
 	job := &batchv1.Job{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      jobName,
@@ -285,7 +286,13 @@ func jobFor(
 			},
 		},
 	}
-
+	// Add security Context only if the PSA is enabled.
+	if jobOption.PsaIsEnabled == "true" {
+		job, err = utils.AddSecurityContextToJob(job, jobOption.PodUserId, jobOption.PodGroupId)
+		if err != nil {
+			return nil, err
+		}
+	}
 	// Add the image secret in job spec only if it is present in the stork deployment.
 	if len(imageRegistrySecret) != 0 {
 		job.Spec.Template.Spec.ImagePullSecrets = utils.ToImagePullSecret(utils.GetImageSecretName(jobName))

pkg/drivers/nfsbackup/nfsbackup.go

@@ -199,7 +199,6 @@ func jobForBackupResource(
 	}, " ")
 
 	labels := addJobLabels(jobOption)
-
 	nfsExecutorImage, imageRegistrySecret, err := utils.GetExecutorImageAndSecret(drivers.NfsExecutorImage,
 		jobOption.NfsImageExecutorSource,
 		jobOption.NfsImageExecutorSourceNs,
@@ -216,6 +215,7 @@ func jobForBackupResource(
 		logrus.Errorf("failed to get the toleration details: %v", err)
 		return nil, fmt.Errorf("failed to get the toleration details for job [%s/%s]", jobOption.Namespace, jobOption.RestoreExportName)
 	}
+
 	job := &batchv1.Job{
 		ObjectMeta: metav1.ObjectMeta{
 			Name:      jobOption.RestoreExportName,
@@ -270,6 +270,18 @@ func jobForBackupResource(
 			},
 		},
 	}
+
+	// Add security Context only if the PSA is enabled.
+	if jobOption.PsaIsEnabled == "true" {
+		// The Job is intended to backup resources to  NFS backuplocation
+		// and it doesn't need a specific JOB uid/gid since it will be sqaushed at NFS server
+		// hence used a global hardcoded UID/GID.
+		job, err = utils.AddSecurityContextToJob(job, utils.KdmpJobUid, utils.KdmpJobGid)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	// Add the image secret in job spec only if it is present in the stork deployment.
 	if len(imageRegistrySecret) != 0 {
 		job.Spec.Template.Spec.ImagePullSecrets = utils.ToImagePullSecret(utils.GetImageSecretName(jobOption.RestoreExportName))

pkg/drivers/nfscsirestore/nfscsirestore.go

@@ -266,6 +266,13 @@ func jobForRestoreCSISnapshot(
 			},
 		},
 	}
+	// Add security Context only if the PSA is enabled.
+	if jobOption.PsaIsEnabled == "true" {
+		job, err = utils.AddSecurityContextToJob(job, jobOption.PodUserId, jobOption.PodGroupId)
+		if err != nil {
+			return nil, err
+		}
+	}
 	// Add the image secret in job spec only if it is present in the stork deployment.
 	if len(imageRegistrySecret) != 0 {
 		job.Spec.Template.Spec.ImagePullSecrets = utils.ToImagePullSecret(utils.GetImageSecretName(jobName))

pkg/drivers/nfsrestore/nfsrestore.go

@@ -313,6 +313,15 @@ func jobForRestoreResource(
 			},
 		},
 	}
+
+	// Add security Context only if the PSA is enabled.
+	if jobOption.PsaIsEnabled == "true" {
+		job, err = utils.AddSecurityContextToJob(job, utils.KdmpJobUid, utils.KdmpJobGid)
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	// Add the image secret in job spec only if it is present in the stork deployment.
 	if len(imageRegistrySecret) != 0 {
 		job.Spec.Template.Spec.ImagePullSecrets = utils.ToImagePullSecret(utils.GetImageSecretName(jobOption.RestoreExportName))

pkg/drivers/options.go

@@ -59,6 +59,10 @@ type JobOpts struct {
 	ResoureBackupName          string
 	ResoureBackupNamespace     string
 	S3DisableSSL               bool
+	// psa specifc option to be used by job
+	PsaIsEnabled string
+	PodUserId    string
+	PodGroupId   string
 }
 
 // WithS3DisableSSL is job parameter
@@ -517,3 +521,27 @@ func WithNodeAffinity(l map[string]string) JobOption {
 		return nil
 	}
 }
+
+// WithpsaIsEnabled is job parameter.
+func WithPsaIsEnabled(psaIsEnabled string) JobOption {
+	return func(opts *JobOpts) error {
+		opts.PsaIsEnabled = psaIsEnabled
+		return nil
+	}
+}
+
+// WithpsaIsEnabled is job parameter.
+func WithPodUserId(podUserId string) JobOption {
+	return func(opts *JobOpts) error {
+		opts.PodUserId = podUserId
+		return nil
+	}
+}
+
+// WithpsaIsEnabled is job parameter.
+func WithPodGroupId(PodGroupId string) JobOption {
+	return func(opts *JobOpts) error {
+		opts.PodGroupId = PodGroupId
+		return nil
+	}
+}

pkg/drivers/utils/utils.go

@@ -25,6 +25,7 @@ import (
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/apimachinery/pkg/util/validation"
 	"k8s.io/apimachinery/pkg/util/wait"
+	"k8s.io/utils/pointer"
 )
 
 const (
@@ -67,6 +68,11 @@ const (
 	IstioInjectLabel  = "sidecar.istio.io/inject"
 	// ProcessVMResourceSuccessMsg - vm resources processed successfully
 	ProcessVMResourceSuccessMsg = "vm resources processed successfully"
+	PsaEnabledKey               = "portworx.io/psa-enabled"
+	PsaUIDKey                   = "portworx.io/psa-uid"
+	PsaGIDKey                   = "portworx.io/psa-gid"
+	KdmpJobUid                  = "1013"
+	KdmpJobGid                  = "1013"
 )
 
 var (
@@ -966,3 +972,42 @@ func GetShortUID(uid string) string {
 	}
 	return uid[:8]
 }
+
+// Add container security Context to job pod if the PSA is enabled.
+// if static uids like kdmpJobUid or kdmpJobGid is used that means
+// these are dummy UIDs used for backing up resources to backuplocation
+// which doesn't need specific UID specific permission.
+func AddSecurityContextToJob(job *batchv1.Job, podUserId, podGroupId string) (*batchv1.Job, error) {
+	if job != nil {
+		if podUserId != "" {
+			uid, err := strconv.ParseInt(podUserId, 10, 64)
+			if err != nil {
+				logrus.Errorf("failed to convert the UID to int: %v", err)
+				return nil, fmt.Errorf("failed to convert the UID to int: %v", err)
+			}
+			job.Spec.Template.Spec.SecurityContext.RunAsUser = &uid
+		}
+		if podGroupId != "" {
+			gid, err := strconv.ParseInt(podGroupId, 10, 64)
+			if err != nil {
+				logrus.Errorf("failed to convert the GID to int: %v", err)
+				return nil, fmt.Errorf("failed to convert the GID to int: %v", err)
+			}
+			job.Spec.Template.Spec.SecurityContext.RunAsGroup = &gid
+		}
+		// Add RunAsNonRoot to true and drop all capabilities and seccomp profile and allowPrivilegeEscalation to false
+		job.Spec.Template.Spec.Containers[0].SecurityContext.RunAsNonRoot = pointer.Bool(true)
+		job.Spec.Template.Spec.Containers[0].SecurityContext.AllowPrivilegeEscalation = pointer.Bool(false)
+		job.Spec.Template.Spec.Containers[0].SecurityContext.SeccompProfile = &corev1.SeccompProfile{
+			Type: "RuntimeDefault",
+		}
+		job.Spec.Template.Spec.Containers[0].SecurityContext.Capabilities = &corev1.Capabilities{
+			Drop: []corev1.Capability{
+				"ALL",
+			},
+		}
+	} else {
+		return job, fmt.Errorf("recieved a nil job object to add security context")
+	}
+	return job, nil
+}