add guard time when scanning naked pods

Signed-off-by: Matthias Bertschy <matthias.bertschy@gmail.com>

config/config.go

@@ -102,6 +102,8 @@ type Config struct {
 	HTTPExporterConfig         *exporters.HTTPExporterConfig `mapstructure:"httpExporterConfig"`
 	ExcludeNamespaces          []string                      `mapstructure:"excludeNamespaces"`
 	IncludeNamespaces          []string                      `mapstructure:"includeNamespaces"`
+	// PodScanGuardTime is the time to wait before scanning a pod without a parent
+	PodScanGuardTime time.Duration `mapstructure:"podScanGuardTime"`
 }
 
 // IConfig is an interface for all config types used in the operator
@@ -121,6 +123,7 @@ type IConfig interface {
 	KubescapeURL() string
 	KubevulnURL() string
 	SkipNamespace(ns string) bool
+	GuardTime() time.Duration
 }
 
 // OperatorConfig implements IConfig
@@ -185,6 +188,7 @@ func (c *OperatorConfig) MatchingRulesFilename() string {
 func (c *OperatorConfig) GatewayWebsocketURL() string {
 	return c.clusterConfig.GatewayWebsocketURL
 }
+
 func (c *OperatorConfig) ConcurrencyWorkers() int {
 	return c.serviceConfig.ConcurrencyWorkers
 }
@@ -224,6 +228,10 @@ func (c *OperatorConfig) EventReceiverURL() string {
 	return c.eventReceiverRestURL
 }
 
+func (c *OperatorConfig) GuardTime() time.Duration {
+	return c.serviceConfig.PodScanGuardTime
+}
+
 func LoadConfig(path string) (Config, error) {
 	viper.AddConfigPath(path)
 	viper.SetConfigName("config")
@@ -236,6 +244,7 @@ func LoadConfig(path string) (Config, error) {
 	viper.SetDefault("triggerSecurityFramework", false)
 	viper.SetDefault("matchingRulesFilename", "/etc/config/matchingRules.json")
 	viper.SetDefault("eventDeduplicationInterval", 2*time.Minute)
+	viper.SetDefault("podScanGuardTime", 3*time.Hour)
 
 	viper.AutomaticEnv()
 

mainhandler/handlerequests.go

@@ -8,6 +8,8 @@ import (
 	"time"
 
 	"github.com/kubescape/backend/pkg/versioncheck"
+	"github.com/kubescape/k8s-interface/workloadinterface"
+	"k8s.io/apimachinery/pkg/runtime"
 
 	core1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -384,12 +386,24 @@ func (mainHandler *MainHandler) HandleImageScanningScopedRequest(ctx context.Con
 			pod.Kind = "Pod"
 
 			// get pod instanceIDs
-			instanceIDs, err := instanceidhandlerv1.GenerateInstanceIDFromPod(&pod)
+			unstructuredObj, err := runtime.DefaultUnstructuredConverter.ToUnstructured(&pod)
+			if err != nil {
+				logger.L().Ctx(ctx).Error("failed to convert pod to unstructured", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()), helpers.Error(err))
+				continue
+			}
+			wl := workloadinterface.NewWorkloadObj(unstructuredObj)
+			instanceIDs, err := instanceidhandlerv1.GenerateInstanceID(wl)
 			if err != nil {
 				logger.L().Ctx(ctx).Error("failed to generate instance ID for pod", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()), helpers.Error(err))
 				continue
 			}
 
+			// for naked pods, only handle if pod is older than guard time
+			if !k8sinterface.WorkloadHasParent(wl) && time.Now().Before(pod.CreationTimestamp.Add(mainHandler.config.GuardTime())) {
+				logger.L().Debug("naked pod younger than guard time detected, skipping scan", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()))
+				continue
+			}
+
 			for _, instanceID := range instanceIDs {
 				s, _ := instanceID.GetSlug(false)
 				if ok := slugs[s]; ok {

watcher/podwatcher.go

@@ -3,8 +3,10 @@ package watcher
 import (
 	"context"
 	"errors"
+	"time"
 
 	mapset "github.com/deckarep/golang-set/v2"
+	"github.com/kubescape/k8s-interface/workloadinterface"
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/client-go/tools/pager"
 
@@ -68,26 +70,44 @@ func (wh *WatchHandler) PodWatch(ctx context.Context, workerPool *ants.PoolWithF
 		if !ok {
 			continue
 		}
+		// handle pod events
 		switch event.Type {
 		case watch.Modified, watch.Added:
-			wh.handlePodWatcher(ctx, pod, workerPool)
-		case watch.Deleted:
+			unstructuredObj, err := runtime.DefaultUnstructuredConverter.ToUnstructured(pod)
+			if err != nil {
+				logger.L().Ctx(ctx).Error("failed to convert pod to unstructured", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()), helpers.Error(err))
+				continue
+			}
+			wl := workloadinterface.NewWorkloadObj(unstructuredObj)
+			if k8sinterface.WorkloadHasParent(wl) {
+				wh.handlePodWatcher(ctx, pod, wl, workerPool)
+			} else {
+				// for naked pods, only handle if pod still exists after guard time
+				logger.L().Debug("naked pod detected, delaying scan", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()))
+				time.AfterFunc(wh.cfg.GuardTime(), func() {
+					// use get to check if pod still exists, and refresh the pod object
+					pod, err := wh.k8sAPI.KubernetesClient.CoreV1().Pods(pod.Namespace).Get(context.Background(), pod.Name, v1.GetOptions{})
+					if err == nil {
+						wh.handlePodWatcher(ctx, pod, wl, workerPool)
+					}
+				})
+			}
+		default:
 			continue
 		}
-
 	}
 	return nil
 }
 
 // handlePodWatcher handles the pod watch events
-func (wh *WatchHandler) handlePodWatcher(ctx context.Context, pod *core1.Pod, workerPool *ants.PoolWithFunc) {
+func (wh *WatchHandler) handlePodWatcher(ctx context.Context, pod *core1.Pod, wl *workloadinterface.Workload, workerPool *ants.PoolWithFunc) {
 
 	// check if we need to add
 	pod.APIVersion = "v1"
 	pod.Kind = "Pod"
 
 	// get pod instanceIDs
-	instanceIDs, err := instanceidhandlerv1.GenerateInstanceIDFromPod(pod)
+	instanceIDs, err := instanceidhandlerv1.GenerateInstanceID(wl)
 	if err != nil {
 		logger.L().Ctx(ctx).Error("failed to generate instance ID for pod", helpers.String("pod", pod.GetName()), helpers.String("namespace", pod.GetNamespace()), helpers.Error(err))
 		return