Protecting mobile agents from external replay attacks

Abstract

This paper presents a protocol for the protection of mobile agents against external replay attacks. This kind of attacks are performed by malicious platforms when dispatching an agent multiple times to a remote host, thus making it reexecute part of its itinerary. Current proposals aiming to address this problem are based on storing agent identifiers, or trip markers, inside agent platforms, so that future reexecutions can be detected and prevented. The problem of these solutions is that they do not allow the agent to perform legal migrations to the same platform several times. The aim of this paper is to address these issues by presenting a novel solution based on authorisation entities, which allow the agent to be reexecuted on the same platform a number of times determined at runtime. The proposed protocol is secure under the assumption that authorisation entities are trusted.

Introduction

Mobile agents can provide many benefits to the development of distributed applications, but their use also poses many security threats. Research on mobile agent technology has identified and solved a number of security-related issues, but there are still many remaining unsolved (Zachary, 2003).

Most of the research work undertaken on mobile agent security is concentrated on the problem of malicious platforms, as platforms have complete control over the agent execution, and can thus do almost anything with the agent code or data. Therefore, achieving a complete solution is considered impossible, although several problems can be mitigated. For example, even though platforms cannot be prevented from manipulating the results generated by the current execution of the agent, they can certainly be prevented from tampering with the results generated by the agent on other platforms.

Most proposed methods on mobile agent protection against malicious platforms try to provide a generic solution to cover as many security threats as possible, usually without putting the solution into practise in any real-life applications. On the other hand, most real-life agent-based applications do not take security into account. Subsequently, various proposals on mobile agent security have failed to address specific scenarios where their solutions may not be valid. In the context of this paper, the solutions presented in the literature for agent replay attacks have failed to address scenarios where the agent has to loop over a certain number of platforms an undetermined number of times (Tsipenyuk, 2004). Agent replay attacks can be classified in two different categories (Yee, 2003):

• These occur when the agent is forced to execute repeatedly on a single platform using different inputs, aiming to obtain different responses and draw conclusions about its behaviour.

• These are performed by malicious platforms by resending the agent to another platform, thus making the agent reexecute part of its itinerary.

Internal replay attacks are impossible to avoid because the platform has complete control over the execution, and can always reset the agent to its arrival state. On the contrary, external replay attacks can be avoided if platforms keep a record of previously executed agents.

The problem of current solutions (Yee, 2003, Westhoff et al., 1999, Wilhelm et al., 1998, Li et al., 2000, Suen, 2003, Mir and Borrell, 2003, Che et al., 2006, Cucurull et al., 2005) against external replay attacks is that they do not allow an agent to be executed n times on the same platform, especially if n is determined at runtime. However, the agent’s itinerary often contains roundtrips that require the same platform to be visited several times. Thus, current solutions force programmers to sacrifice some of the inherent flexibility in mobile agent itineraries.

In order to enhance security and flexibility, this paper presents a solution based on authorisation entities. The advantage is that these entities are entitled to generate new identifiers for the agent, thus allowing the repeatable migration of the agent to the same platform. Therefore, the proposed solution allows programmers to develop secure mobile agent applications and still maintaining the agents’ intrinsic flexibility.

The rest of the paper is structured as follows. Section 2 introduces the related work on preventing replay attacks. Section 3 presents the proposed protocol for replay attack prevention. Section 4 describes the implementation of the proof-of-concept of the proposed solution. Section 5 concludes the paper and points out future directions.