A Method of Safety Analysis for Runtime Code Update

Abstract

In this paper, we present a novel method of safety analysis for runtime code update, i.e., updating a program at runtime without terminating its execution. Runtime code update is an emerging technique especially for increasing availability of the servers which should always be in service and free of any known bugs or security flaws. However, it may cause state inconsistency or unintended behaviors unless it is properly restricted. Although too much restriction enables us to easily ensure safety of updating code at runtime, it prevents us from coping with the realistic updates. To reveal appropriate restriction, we first construct a very precise model of safe runtime code update based on a framework of explicit data/control flow and dependency. Then, a class of analyses which statically estimates the set of safe update points is derived by approximating the model. We restrict only the timing: behaviorally safe runtime code update may occur only at safe update points. Moreover, we can relax the restriction by explicitly specifying non-critical points.

This research was supported by PRESTO research program of Japan Science and Technology Agency.