cri-o · openshift-merge-robot · Mar 29, 2021 · Mar 29, 2021
diff --git a/cmd/crio/main.go b/cmd/crio/main.go
@@ -249,6 +249,16 @@ func main() {
 				logrus.Error(err)
 			}
 
+			// Write "$CleanShutdownFile".supported to show crio-wipe that
+			// we should be wiping if the CleanShutdownFile wasn't found.
+			// This protects us from wiping after an upgrade from a version that don't support
+			// CleanShutdownFile.
+			f, err := os.Create(config.CleanShutdownSupportedFileName())
+			if err != nil {
+				logrus.Errorf("Writing clean shutdown supported file: %v", err)
+			}
+			f.Close()
+
 			// and sync the changes to disk
 			if err := utils.SyncParent(config.CleanShutdownFile); err != nil {
 				logrus.Errorf("failed to sync parent directory of clean shutdown file: %v", err)

diff --git a/cmd/crio/wipe.go b/cmd/crio/wipe.go
@@ -7,6 +7,7 @@ import (
 	"github.com/cri-o/cri-o/internal/criocli"
 	"github.com/cri-o/cri-o/internal/storage"
 	"github.com/cri-o/cri-o/internal/version"
+	crioconf "github.com/cri-o/cri-o/pkg/config"
 	json "github.com/json-iterator/go"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
@@ -36,41 +37,18 @@ func crioWipe(c *cli.Context) error {
 	if err != nil {
 		return err
 	}
+	shouldWipeImages := true
+	shouldWipeContainers := true
 
-	// First, check if the node was rebooted.
-	// We know this happened because the VersionFile (which lives in a tmpfs)
-	// will not be there.
-	shouldWipeContainers, err := version.ShouldCrioWipe(config.VersionFile)
-	if err != nil {
-		logrus.Infof("checking whether cri-o should wipe containers: %v", err)
-	}
-
-	// Then, check whether crio has shutdown with time to sync.
-	// Note: this is only needed if the node rebooted.
-	// If there wasn't time to sync, we should clear the storage directory
-	if shouldWipeContainers && config.CleanShutdownFile != "" {
-		if _, err := os.Stat(config.CleanShutdownFile); err != nil {
-			logrus.Infof("file %s not found. Wiping storage directory %s because of suspected dirty shutdown", config.CleanShutdownFile, store.GraphRoot())
-			// If we do not do this, we may leak other resources that are not directly in the graphroot.
-			// Erroring here should not be fatal though, it's a best effort cleanup
-			if err := store.Wipe(); err != nil {
-				logrus.Infof("failed to wipe storage cleanly: %v", err)
-			}
-			// unmount storage or else we will fail with EBUSY
-			if _, err := store.Shutdown(false); err != nil {
-				return errors.Errorf("failed to shutdown storage before wiping: %v", err)
-			}
-			// totally remove storage, whatever is left (possibly orphaned layers)
-			if err := os.RemoveAll(store.GraphRoot()); err != nil {
-				return errors.Errorf("failed to remove storage directory: %v", err)
-			}
-			return nil
+	if !c.IsSet("force") {
+		// First, check if the node was rebooted.
+		// We know this happened because the VersionFile (which lives in a tmpfs)
+		// will not be there.
+		shouldWipeContainers, err = version.ShouldCrioWipe(config.VersionFile)
+		if err != nil {
+			logrus.Infof("checking whether cri-o should wipe containers: %v", err)
 		}
-	}
 
-	shouldWipeImages := true
-	// First, check if we need to upgrade at all
-	if !c.IsSet("force") {
 		// there are two locations we check before wiping:
 		// one in a temporary directory. This is to check whether the node has rebooted.
 		// if so, we should remove containers
@@ -82,6 +60,13 @@ func crioWipe(c *cli.Context) error {
 		}
 	}
 
+	// Then, check whether crio has shutdown with time to sync.
+	// Note: this is only needed if the node rebooted.
+	// If there wasn't time to sync, we should clear the storage directory
+	if shouldWipeContainers && shutdownWasUnclean(config) {
+		return handleCleanShutdown(config, store)
+	}
+
 	// if we should not wipe, exit with no error
 	if !shouldWipeContainers {
 		// we should not wipe images without wiping containers
@@ -103,6 +88,40 @@ func crioWipe(c *cli.Context) error {
 	return nil
 }
 
+func shutdownWasUnclean(config *crioconf.Config) bool {
+	// CleanShutdownFile not configured, skip
+	if config.CleanShutdownFile == "" {
+		return false
+	}
+	// CleanShutdownFile isn't supported, skip
+	if _, err := os.Stat(config.CleanShutdownSupportedFileName()); err != nil {
+		return false
+	}
+	// CleanShutdownFile is present, indicating clean shutdown
+	if _, err := os.Stat(config.CleanShutdownFile); err == nil {
+		return false
+	}
+	return true
+}
+
+func handleCleanShutdown(config *crioconf.Config, store cstorage.Store) error {
+	logrus.Infof("file %s not found. Wiping storage directory %s because of suspected dirty shutdown", config.CleanShutdownFile, store.GraphRoot())
+	// If we do not do this, we may leak other resources that are not directly in the graphroot.
+	// Erroring here should not be fatal though, it's a best effort cleanup
+	if err := store.Wipe(); err != nil {
+		logrus.Infof("failed to wipe storage cleanly: %v", err)
+	}
+	// unmount storage or else we will fail with EBUSY
+	if _, err := store.Shutdown(false); err != nil {
+		return errors.Errorf("failed to shutdown storage before wiping: %v", err)
+	}
+	// totally remove storage, whatever is left (possibly orphaned layers)
+	if err := os.RemoveAll(store.GraphRoot()); err != nil {
+		return errors.Errorf("failed to remove storage directory: %v", err)
+	}
+	return nil
+}
+
 type ContainerStore struct {
 	store cstorage.Store
 }

diff --git a/pkg/config/config.go b/pkg/config/config.go
@@ -762,6 +762,10 @@ func (c *RootConfig) Validate(onExecution bool) error {
 	return nil
 }
 
+func (c *RootConfig) CleanShutdownSupportedFileName() string {
+	return c.CleanShutdownFile + ".supported"
+}
+
 // Validate is the main entry point for runtime configuration validation
 // The parameter `onExecution` specifies if the validation should include
 // execution checks. It returns an `error` on validation failure, otherwise

diff --git a/test/crio-wipe.bats b/test/crio-wipe.bats
@@ -184,3 +184,17 @@ function start_crio_with_stopped_pod() {
 	test_crio_did_not_wipe_containers
 	test_crio_did_not_wipe_images
 }
+
+@test "don't clear containers if clean shutdown supported file not present" {
+	start_crio_with_stopped_pod
+	stop_crio_no_clean
+
+	rm "$CONTAINER_CLEAN_SHUTDOWN_FILE.supported"
+
+	run_crio_wipe
+
+	start_crio_no_setup
+
+	test_crio_did_not_wipe_containers
+	test_crio_did_not_wipe_images
+}