fix: validate snapshot data before migration restore

shim/restore.go

@@ -31,6 +31,7 @@ var (
 	ErrAlreadyRestored      = errors.New("container is already restored")
 	ErrRestoreRequestFailed = errors.New("restore request failed")
 	ErrRestoreDial          = errors.New("failed to connect to node socket")
+	ErrInvalidCheckpoint    = errors.New("checkpoint data invalid")
 )
 
 // Restore starts the container either from a checkpoint or from scratch
@@ -236,9 +237,11 @@ func MigrationRestore(ctx context.Context, r *task.CreateTaskRequest, cfg *Confi
 
 	log.G(ctx).Infof("restore response: %v", resp.MigrationInfo)
 
-	// TODO: validate that path is valid and contains image
+	if err := validateCheckpointData(nodev1.SnapshotPath(resp.MigrationInfo.ImageId)); err != nil {
+		return false, fmt.Errorf("%w: %w", ErrInvalidCheckpoint, err)
+	}
 	r.Checkpoint = nodev1.SnapshotPath(resp.MigrationInfo.ImageId)
-	log.G(ctx).Infof("setting checkpoint dir for restore: %s", nodev1.SnapshotPath(resp.MigrationInfo.ImageId))
+	log.G(ctx).Infof("setting checkpoint dir for restore: %s", r.Checkpoint)
 
 	// we set the criu work path for the live migration to work (the lazy pages
 	// socket needs to be there) and also so the restore stats are stored in the
@@ -323,3 +326,8 @@ func FinishRestore(ctx context.Context, id string, cfg *Config, startTime time.T
 
 	return nil
 }
+
+func validateCheckpointData(snapshotPath string) error {
+	_, err := os.Stat(filepath.Join(snapshotPath, "descriptors.json"))
+	return err
+}

shim/task/service_zeropod.go

@@ -160,13 +160,14 @@ func (w *wrapper) Create(ctx context.Context, r *taskAPI.CreateTaskRequest) (_ *
 		skipStart, err := zshim.MigrationRestore(ctx, r, cfg)
 		if err != nil {
 			if errors.Is(err, zshim.ErrRestoreRequestFailed) ||
-				errors.Is(err, zshim.ErrRestoreDial) {
+				errors.Is(err, zshim.ErrRestoreDial) ||
+				errors.Is(err, zshim.ErrInvalidCheckpoint) {
 				// if the restore fails with ErrRestoreRequestFailed it's very
 				// likely it simply did not find a matching migration. Equally,
 				// if the shim can't manage to dial the node service there's no
 				// chance it can be restored. We log it and create the container
 				// from scratch.
-				log.G(ctx).Errorf("restore request failed: %s", err)
+				log.G(ctx).WithError(err).Error("restore request failed")
 			} else {
 				return nil, err
 			}