ADVANCES IN NEWORK AND

DISTRIBUTED SYSTEMS SECURITY

IFIP - The International Federation for Information Processing

IFIP was founded in 1960 under the auspices of UNESCO, following the First World
Computer Congress held in Paris the previous year. An umbrella organization for societies
working in information processing, IFIP's aim is two-fold: to support information
processing within its member countries and to encourage technology transfer to developing
nations. As its mission statement clearly states,
IFIP's mission is to be the leading, truly international, apolitical organization which
encourages and assists in the development, exploitation and application of information
technology for the benefit of all people.
IFIP is a non-profitmaking organization, run almost solely by 2500 volunteers. It operates
through a number of technical committees, which organize events and publications. IFIP's
events range from an international congress to local seminars, but the most important are:
The IFIP World Computer Congress, held every second year;
open conferences;
working conferences.
The flagship event is the IFIP World Computer Congress, at which both invited and
contributed papers are presented. Contributed papers are rigorously refereed and the
rejection rate is high.
As with the Congress, participation in the open conferences is open to all and papers may
be invited or submitted. Again, submitted papers are stringently refereed.
The working conferences are structured differently. They are usually run by a working group
and attendance is small and by invitation only. Their purpose is to create an atmosphere
conducive to innovation and development. Refereeing is less rigorous and papers are
subjected to extensive group discussion.
Publications arising from IFIP events vary. The papers presented at the IFIP World
Computer Congress and at open conferences are published as conference proceedings, while
the results of the working conferences are often published as collections of sel ected and
edited papers.
Any national society whose primary activity is in information may apply to become a full
member of IFIP, although full membership is restricted to one society per country. Full
members are entitled to vote at the annual General Assembly, National societies preferring
a less committed involvement may apply for associate or corresponding membership.
Associate members enjoy the same benefits as full members, but without voting rights.
Corresponding members are not represented in IFIP bodies. Affiliated membership is open
to non-national societies, and individual and honorary membership schemes are also offered.

ADVANCES IN
NETWORK AND
DISTRIBUTED
SYSTEMS SECURITY
IFIP TC11 WG11.4
First Annual Working Conference on Network Security
November 26-27, 2001, Leuven, Belgium
Edited by

Bart De Decker
Katholieke Universiteit Leuven, DistriNet
Belgium

Frank Piessens
Katholieke Universiteit Leuven, DistriNet
Belgium

Jan Smits
Technische Universiteit Eindhoven
The Netherlands

Els Van Herreweghen

IBM Research Laboratory, Zurich
Switzerland

KLUWER ACADEMIC PUBLISHERS

NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW

eBook ISBN:
Print ISBN:

0-306-46958-8
0-792-37558-0

2002 Kluwer Academic Publishers

New York, Boston, Dordrecht, London, Moscow

All rights reserved

No part of this eBook may be reproduced or transmitted in any form or by any means, electronic,
mechanical, recording, or otherwise, without written consent from the Publisher

Created in the United States of America

Visit Kluwer Online at:

and Kluwer's eBookstore at:

http://www.kluweronline.com
http://www.ebooks.kluweronline.com

CONTENTS
Preface

vii

Acknowledgements

ix

Part One

- Reviewed Papers

1 A Role-Based Specification of the SET Payment

Transaction Protocol
Hideki Sakurada, Yasuyuki Tsukada

2 Information Security: Mutual Authentication in

E-Commerce
S.H. Von Solms, M. V. Kisimov

15

3 Software-Based Receipt-Freeness in On-Line Elections

Emmanouil Magkos, Vassilios Chrissikopoulos,
Nikos Alexandris

33

ID-Based Structured Multisignature Schemes

Chih- Y i n Lin, Tzong- Chen Wu, Jing- Jang Hwang

45

Probabilistic Relations for the Solitaire Keystream

Generator
Marina Pudovkina

61

6 Hazard Analysis for Security Protocol Requirements

Nathalie Foster, Jeremy Jacob

75

7 Securing RMI Communication

Vincent Naessens, Bart Vanhaute, Bart De Decker

93

Secure Java Development With UML

Jan Jrjens

9 Security Through Aspect-Oriented Programming

Bart De Win, Bart Vanhaute, Bart De Decker

107
125

10 Extending a Campus Network with Remote Bubbles

using IPsec
Aurlien Bonnet, Marc Lobelle

139

11 Combining World Wide Web and Wireless Security

Joris Claessens, Bart Preneel, Joos Vandewalle

153

vi

12 On Mobile Agent Based Transactions in Moderately

Hostile Environments
Niklas Borselius, Chris J. Mitchell, Aaron Wilson

173

13 SPARTA, A Mobile Agent Based Intrusion Detection

187

System
Christopher Krgel, Thomas Toth, Engin Kirda

Part Two - Invited Papers

1 Shells Trust Domain
Certification
Pieter van Dijken

Author Index

Infrastructure

Security

201

203

PREFACE
The first Annual Working Conference of WG11.4 of the International Federation for Information Processing (IFIP), focuses on
various state-of-the-art concepts in the field of Network and Distributed Systems Security.
Our society is rapidly evolving and irreversibly set on a course
governed by electronic interactions. We have seen the birth of email in the early seventies, and are now facing new challenging
applications such as e-commerce, e-government, . . . . The more our
society relies on electronic forms of communication, the more the
security of these communication networks is essential for its wellfunctioning. As a consequence, research on methods and techniques
to improve network security is of paramount importance.
This Working Conference brings together researchers and practioners of various disciplines, organisations and countries, to discuss
the latest developments in security protocols, secure software engineering, mobile agent security, e-commerce security and security for
distributed computing .
We are also pleased to have attracted two international speakers
to present two case studies, one dealing with Belgiums intention to
replace the identity card of its citizens by an electronic version, and
the other discussing the implications of the security certification in
a multinational corporation.
This Working Conference should also be considered as the kickoff activity of WG11.4, the aims of which can be summarized as
follows:
rn

to promote research on technical measures for securing computer networks, including both hardware- and software-based
techniques.
to promote dissemination of research results in the field of
network security in real-life networks in industry, academia
and administrative institutions.

viii

= to promote education in the application of security techniques,

and to promote general awareness about security problems in
the broad field of information technology.
Researchers and practioners who want to get involved in this
Working Group, are kindly requested to contact the chairman.
More information on the workings of WG11.4 is available from the
official IFIP-website: h t t p : //www . if i p . a t . org/.
Finally, we wish to express our gratitude to all those who have
contributed to this conference in one way or another. We are grateful to the international referee board who reviewed all the papers
and to the authors and invited speakers, whose contributions were
essential to the success of the conference. We would also like to
thank the participants whose presence and interest, together with
the changing imperatives of society, will prove a driving force for
future conferences to come.
P ROF . B. DE DECKER

ACKNOWLEDGEMENTS
Organised by:
K.U.Leuven, Dept. of Computer Science, DistriNet
IFIP/TC-11 Working Group 11.4 (Network Security)

Supported by:
Scientific Research Network on "Foundations of Software
Evolution", and as such, partially financed by the Fund for
Scientific Research - Flanders (Belgium)

Financially Supported by:

IBM Research
Telindus
Ubizen
Utimaco Safeware Belgium
Programme Committee:

Bart De Decker, (chair), K.U.Leuven, Belgium

Jan M. Smits, (co-chair), T.U.Eindhoven, The Netherlands
Els Van Herreweghen, (co-chair), IBM Research Lab, Zurich,
Switzerland
William J Caelli, Queensland Univ. of Technology, Australia
Herve Debar, France Telecom R&D, France
Serge Demeyer, Univ. of Antwerp, Belgium
Yves Deswarte, LAAS-CNRS, Toulouse, France
Jan Eloff, Rand Afrikaans Univ., South Africa
Dimitris Gritzalis, Athens Univ. of Economics & Business, Greece
Manfred Hauswirth, Technical Univ. of Vienna, Austria
Andrew Hutchison, MGX Consulting, South Africa
Guenter Karjoth, IBM Zurich Research Lab, Switzerland
Kwok-Yan Lam, PrivyLink International Limited, Hong Kong
Marc Lobelle, UCL, Belgium
Keith Martin, Royal Holloway, Univ. of London, United Kingdom
Refik Molva, Institut Eurcom, France
Frank Piessens, K.U.Leuven, Belgium

Hartmut Pohl, Univ. of Applied Sciences Bonn-Rhein-Sieg, Koln,

Germany
Reinhard Posch, Graz Univ. of Technology, Austria
Bart Preneel, K.U.Leuven, Belgium
Kai Rannenberg, Microsoft Research Cambridge, United Kingdom
Peter Ryan, Norwegian Computing Center, Oslo, Norway
Pierangela Samarati, Univ. of Milan, Italy
Einar Snekkenes, Norsk Regnesentral, Oslo, Norway
Henk van Tilborg, T.U.Eindhoven, The Netherlands
Vijay Varadharajan, Macquarie Univ., Australia
Basie Von Solms, Rand Afrikaans Univ., South Africa
Rossouw Von Solms, Port Elizabeth Technikon, South Africa
Jozef Vyskoc, VaF, Bratislava, Slovak Republic
Tatjana Welzer, Univ. of Maribor. Slovenia
Reviewers:

Bussard, Laurent, France

Debar, Herv, France
De Decker, Bart, Belgium
De Win, Bart, Belgium
Demeyer, Serge, Belgium
Deswarte, Yves, France
Druzovec, M arjan, Slovenia
Eloff, Jan, South Africa
Gritzalis, Dimitris, Greece
Hauswirth, Manfred, Austria
Hutchison, Andrew, South Africa
Karjoth, Guenter, Switzerland
Lam, Kwok-Yan, Hong Kong
Lobelle, Marc, Belgium
Martin, Keith, United Kingdom
Molva, Refik, France
Piessens, F'rank, Belgium
Pohl, Hartmut, Germany
Posch, Reinhard, Austria
Preneel, Bart, Belgium
Rannenberg, Kai, United Kingdom
Ryan, Peter, Norway
Samarati, Pierangela, Italy

xi

Schoenmakers, Berry, The Netherlands

Smits, Jan, The Netherlands
Snekkenes, Einar, Norway
Van Herreweghen, Els, Switzerland
Vanhaute, Bart, Belgium
van Tilborg, Henk, The Netherlands
Varadharajan, Vijay, Australia
Von Solms, Basie, South Africa
Von Solms, Rossouw, South Africa
Vyskoc, Josef, Slovak Republic
Welzer, Tatjana, Slovenia

This page intentionally left blank.

PART ONE
Reviewed Papers

This page intentionally left blank.

A ROLE-BASED SPECIFICATION OF
THE SET PAYMENT TRANSACTION
PROTOCOL
Hideki Sakurada
NTT Communication Science Laboratories,
NTT Corporation,
3-1 Morinosato- Wakamiya, Atsugi, Kanagawa, 243-0198 Japan
sakurada0theory.brl.ntt.co.jp

Yasuyuki Tsukada
NTT Communication Science Laboratories,
NTT Corporation,
3-1 Morinosato- Wakamiya, Atsugi, Kanagawa, 243-0198 Japan
[email protected]

Abstract

In this paper, we define a language for specifying security protocols

concisely and unambiguously. We use this language to formally specify
the protocol for payment transactions in Secure Electronic Transaction
(SET), which has been developed by Visa and MasterCard.
In our language, a protocol is specified as a collection of processes.
Each process expresses the role of a participant. In the role-based specification, the components that a participant sees in a message can be
stated explicitly. This is important in specifying protocols like that for
the SET payment transactions because in such protocols some message
components are encrypted and invisible to some participants.
We simplify the SET payment transaction protocol into the exchanges
of six messages. Because our future goal is to formally analyze the security properties that Meadows and Syverson discussed, we make the
simplified protocol contain the parameters used in their security properties. And we also refrain from excessive simplification. For example, we
use dual signature in the payment request message as it is specified in
the SET specification books, while most of the other works do not use
it. Our specification can serve as a starting point for a formal analysis
of the protocol.

Keywords:

Formal methods, security protocols, electronic commerce

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

1.

Introduction

Security protocols are used in distributed systems to protect the secrecy

of messages and to identify users. It is well known that designing them
is an error-prone task. The most significant issues concerning security
protocols are that (1) attacks on them may succeed even without breaking the cryptographic algorithms used and that (2) it may be difficult to
make sure of the correctness of a small protocol that involves exchanges
of only a few messages. Some examples of protocol failures are presented
in (Anderson and Needham, 1995; Clark and Jacob, 1997).
Formal methods can be used to analyze security protocols. With
the methods, protocols are specified and their security properties are
verified. Indeed, many formal methods have been developed (Meadows,
1996; Paulson, 1998; Denker et al., 2000) and succeeded in finding errors
in protocols or verifying their correctness (Burrows et al., 1990; Paulson,
1998). However, it is hard to apply these methods to large protocols.
This is because large protocols are complex and there are no appropriate tools for analyzing such complex protocols. With a tool designed for
small protocols, specifying complex protocols and their security properties is hard. Moreover, the obtained specifications tend to be lengthy and
unintuitive. To avoid these difficulties, protocols are usually simplified
and the simplified protocols are verified instead.
In this paper, we discuss the Secure Electronic Transaction (SET)
protocol (SET Secure Electronic Transaction LLC, 1997a; SET Secure
Electronic Transaction LLC, 1997b; SET Secure Electronic Transaction
LLC, 1997c). In particular, we formally specify the payment transaction
protocol that is a part of SET. This formal specification serves as a
starting point of a formal analysis of the protocol.
SET has been developed by Visa and MasterCard for secure electronic
commerce using payment cards. Over six hundred pages are needed to
explain and specify it. There are some works on the formal specification
and the analysis of the protocol (Lu and Smolka, 1999; Bolignano, 1997;
Kessler and Neumann, 1998). However, they simplified the protocol
excessively in order to reduce the complexity. For example, most of
these simplified protocols did not use dual signature, which is one of
the characteristics of SET. Since we aim at verifying security properties
that Meadows and Syverson discussed in (Meadows and Syverson, 1998),
we include in our simplified protocol the parameters that occur in the
properties. We also make the simplified protocol use dual signature. In
order to describe the specification concisely and unambiguously, we first
define a protocol specification language. In our language, a protocol
is specified as a collection of processes that express the roles of the

A role-based specification of the SET payment transaction protocol

Figure 1.

A typical message flow in the Needham-Schroeder shared-key protocol

participants in the protocol. This is useful for describing the specification

of the SET payment transaction protocol.
The rest of this paper is organized as follows. We first define a language for specifying large security protocols concisely and unambiguously (Section 2). We then use it to specify the SET payment transaction protocol (Section 3). We finally summarize our results and mention
some related works (Section 4).

2.

Protocol Specification Language

Before presenting our protocol specification language, we briefly explain our design policy for it.
Security protocols are often explained by showing a typical message
flow. For example, a typical message flow of the Needham-Schroeder
shared-key protocol (Needham and Schroeder, 1978) is shown in Figure
1. The first line means that a participant A sends a message composed of
her name, the name of the participant she wants to authenticate, and a
fresh nonce (random number) to the authentication server S . The second
K A B ,{ K A BA
, } K ~ ~ } K
line means that S replies with message {NA,B,
to A . This message is obtained by encrypting N A , B , a newly generated
key K A Bto be shared by A and B , and { K A B A
, } K with
~ ~the key K A S .
The { K A B , A } Kis~obtained
~
by encrypting K A B and A with the key
K B S . A, B , and N A on the second line refer to themselves on the first
line, respectively. Since A is assumed to know K A S and is not assumed
, } K ~ ~and
} can
K ~ ~
to know K B S ,she can decrypt { N A ,B , K A B ,{ K A B A
not decrypt {KAB,A}Kes.
Explanations by showing a typical message flow are concise and intuitive. However, they can not explicitly handle what each participant can
see in a message because each line expresses the sending and receiving
of a message at the same time. For example, on the second and the
third line in the previous example, A receives a messages that includes

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 2.

The initiator role in the Needham-Schroedcr shareti-key protocol

{ K A BA, } K and
~ ~sends it to B. Without assumptions on the knowledge
of A , it is not clear whether if she knows the content { K A H , A of
} the
message or not,. This ambiguity may cause human-errors in specifying
complex protocols that use cryptography frequently.
To avoid this problem, we specify a protocol as a collection of processes
that express the roles of the participants in the protocol. To illustrate
this, we show, in Figure 2, a process that are related to As role in the
previous example. Note that we use a variable X for the encrypted
component in the message from S to A. It is clear that A sends the
component X to B as it is.
Now we define our protocol specification language. Since we assume
the Dolev-Yao (Dolev and Yao, 1981) model, we define the set of messages as an algebra made from participants names, natural numbers
(including nonces), and keys with tupling and cryptographic operations.
The formal syntax of messages is as follows.

::= A

{ M1, . * * 7 M n )
{M)K
H(M)

I N

I
I

; participants name
; key
; natural number
; tuple
; encryption of message M using key K
; hash of message M

H is a collision-free one-way hash function. We write K- for the decryption key of a key K . For example, { A , NA}Kis a message obtained
by encrypting a tuple of A and N A with K , where A , N A ,and K are an
participants name, a nonce, and a key, respectively.

A role-based specification of the SET payment transaction protocol

Since our language has variables, we define the set of t e r m s by extending the previous syntax with variables.

.._ ...

; variable whose name is X

Because we usually use variables instead of concrete names, nonces, and

keys, we regard A, N A , K, etc. that occur in terms as variables unless
otherwise noted explicitly.
We finally define the set of processes with the following syntax. We
specify a protocol as the set of processes of its participants.

..I
I
I
I
I

End
Send T P
Recv T P
New X P
Let X = T P
Assert Q P

; silent process
; sending of message T
; receiving of message T
; generating of a fresh nonce X
; binding of T to the local variable X
; checking of proposition Q

We dont specify the receiver and the sender of a message in Send TP and
Recv TP, respectively because we assume that there exist intruders that
can capture any message on networks and can send any message they can
construct. We understand that a process of the form New X P binds free
occurrences of X in P. In other words, in a process New X P, the variables X that occur in P refer to the newly generated nonce X. We also
understand that a process of the form Recv T P does pattern-matching
and variable-binding. For example, a process Recv { N, H (N)} P accepts
(2001, H(2001)}, where variable N is bound to the number 2001. The
process however does not accept (2001, H(2002)} .
Assert Q P acts as P if proposition Q holds, otherwise it acts as End.
The set of propositions depends on the system used for analysis. Since
we use Isabelle (Paulson, 1994), a proof checker of higher-order logics,
we can use any proposition in Isabelle.
As an example, we specify the role of A, the initiator, in the NeedhamSchroeder shared-key protocol in Figure 3. The process is parametrized
by her name A, the responders name B , and the key K A S .

3.

A Specification of the SET Payment

Transaction Protocol

In this section, we give a formal specification of the SET payment

transaction protocol. Since our future goal is to verify security properties
that include those which Meadows and Syverson discussed in (Meadows
and Syverson, 1998), we simplify the protocol into the exchanges of six

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 3.
The process that specifies the initiator role of the Needham-Schroeder
shared-key protocol

messages that include the parameters used in their security properties.

Meadows and Syverson developed a method to describe security properties flexibly and discussed the security properties that the payment
transaction protocol is expected to satisfy. However, they did not specify the protocol formally. Our formal specification is needed in order
to verify the security properties. We also make the simplified protocol
use dual signature, which is one of the characteristics of the original
protocol.
We first overview the SET payment transaction protocol. Three
parties, a cardholder, a merchant, and a payment gateway, are involved
in a payment transaction in SET. This protocol is invoked after the
cardholder has completed browsing, selection, and ordering. One of the
purposes of the protocol is to securely send the payment information,
which includes the account number of the payment-card of the cardholder and the amount of money that he will pay for the order, to the
payment gateway.
A typical message flow of the protocol is shown in Figure 4. We show
only the six messages that our simplified protocol has. We also omit the
structures of the messages in the figure. The cardholder and the merchant first exchange the identifiers of the transaction in PInitReq and
PInitRes messages. The identifiers are referred to in subsequent messages. The cardholder then sends the purchase request message PReq
to the merchant. This message includes the amount of money that the
cardholder will pay and her payment-card number. She keeps the number secret from the merchant by encrypting a component that includes
it. The merchant sends the gateway AuthReq message that includes the
component. The gateway checks the validity of the payment-card number, processes the payment, and returns the result to the merchant in

A role-based specification of the SET payment transaction protocol

Cardholder

Merchant

Payment Gateway

PInitReq
[InitRes
PReg
AuthReg
AuthRes
t-PRes
c
--

Figure 4 .

A typical message flow in the SET payment transaction protocol

Figure 5. Operations on messages used in the SET payment transaction protocol

AuthRes message. The merchant receives it and sends the result to the
cardholder in PRes message.
Various cryptographic operations are used in SET. We define each of
the operations used in our protocol as a function on the set of messages
in our language. The definitions are essentially the same as what Bella et
al. did in their verification of the SET cardholder registration protocol
(Bella et al., 2000). We show the definitions in Figure 5 . The subscripts
r and s of names of participants indicate that the participants appear
as the receiver and the sender of a message, respectively. L(Ml,M2)
contains a linkage from message MI to message Mz.SO(A,, M ) is the
signature of a participant A, on message M . S ( A , , M ) is message M
with the signature of A s . Enc models a signed-then-encrypted message.
EncB models a signed-then-encrypted message with an external baggage.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Cardholder(C, M , P, O D , PurchAmt, P A N , PANSecret) =

New RRPIDl
New L I D c
New Challc
/ / PInitReq
Send (RRPID1, L I D c , Challc}
// PInitRes
Recv S ( M , { { L I D c , L I D M ,X I D } , RRPID1, Challc, ChallM})
Let TransID = { L I D c , L I D M ,X I D }
New ODSalt
New R R PID2
Let PANData = { P A N , PANSecret}
Let PIHead = { TransID, H ( O D ) ,PurchAmt}
Let OIData = {TransID, RRPID2, Challc, H(OD), ODSalt}
Let PIData = { PIHead, PA NData}
New K
// PReq
Send { { SO(C, {H(PIData),H(OIData)}),
EX(P, L(PIHead, OIData),PANData, K ) } ,
{ OIData, H(PIData)}}
// PRes
Recv ( S ( M ,{ DansID, RRPID2, Challc}))
Figure 6.

The cardholder process in the SET payment transaction protocol

EK and SK are the functions that relate each participant to his public
encryption key and his public signature key, respectively.
The processes of a cardholder, a merchant, and a payment gateway
are shown in Figures 6, 7 and 8, respectively.
Here, C, M , and P are the names of a cardholder, a merchant, and a
payment gateway, respectively. OD, P A N , PurchAmt and AuthReqAmt
are an order description, the account number of a payment-card, the
amount of money that a cardholder will pay, and the amount of money
that a merchant requires, respectively. PANSecret is used to prevent
guessing attacks on P A N . ValidPANSet is the set of valid PANS. It
does not appear in the SET specification books. We introduce it to
model the authentication of payment-cards. Dual signature is used in
the PReq message. The message is composed of the following three parts:
SO (C, {H(PIData),H ( OIData)}), EX(P, L(PIHead, OIData), PANData,
K ) and { OIData, H(PIData)}.

A role-based specification of the SET payment transaction protocol

Merchant(M,C, P, OD, ODSalt, AuthReqAmt) =

// PInitReq
Recv {RRPID 1 ,LIDc, ChaZZc}
New LIDM
New XID
New C h d M
Let TkansID = { L I D c , L I D M ,XID}
// PInitRes
Send S ( M , { FransID, RRPID1, Challc, ChallM})
Let OIData = { TkansID, RRPID2, Challc, H( OD), ODSalt}
// PReq
Recv { { S O ( C ,HPIData, H( OIData)),PIBody},
{ OIData, HPIData}}
New RRPID3
New K1
// AuthReq
Send EncB( M , P, { RRPID3, RansID, AuthReqAmt},
{SO(C, {HPIData, H( OlData)}),PIBody},Ki)
// AuthRes
Recv Enc(P, M , (RRPID3, lPransID,AuthAmt}, K2)
Assert AuthReqAmt = AuthAmt
// PRes
Send S ( M ,{ RansID, RRPID2, Challc})
Figure 7.

The merchant process in SET payment transactions

Gateway(P,C, M , ValidPANSet) =
// AuthReq
Recv EncB( M , P, (RRPID3, FransID, AuthReqAmt}
{ SO(C, { H({{ RansID, HOD, PurchAmt}, PANData}),
HOIData}),
EX(P, { { RansID, HOD, PurchAmt},
HOIData}, PANData, K l ) } )
Assert PANData E ValidPANSet
Assert PurchAmt = AuthReqAmt
Let AuthAmt = AuthReqAmt
New K2

//

AuthRes

Send Enc(P, M , (RRPID3, DansID, AuthAmt}, K2)

Figure 8.

The gateway process in SET payment transactions

10

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

The second part cannot be decrypted by a merchant and should be

passed to a payment gateway. The content of the third part should be
read by a merchant. The first part is the signature on {H (PIData),
H(OIData)}. A participant who receives either of the last two parts
can compute {H(PIData), H (OIData)} and can check the signature.

4.

Concluding Remarks

In this paper, we have defined a language for specifying security protocols and have used it to formally specify the SET payment transaction
protocol. In our language, a security protocol is specified as a collection
of processes. Each process defines the role of a participant. This is useful
in specifying complex protocols concisely and unambiguity.
We have simplified the SET payment transaction protocol and have
specified it formally. We aim at verifying various security properties of
the protocol including those that Meadows and Syversion discussed in
(Meadows and Syverson, 1998). Our specification can serve as a starting
point for a formal analysis that take into account dual signature of the
protocol.
We have already implemented our specification language on the Isabelle theorem prover (Paulson, 1994) and have written the specification
in it. We are also developing a protocol execution model and a language
to describe security properties concisely. In the execution model, a state
of a participant is modeled as a process in our language and an environment, a set of variable-value pairs. The environment corresponds to the
data that the participant uses. For example, in a key exchange protocol,
the environment of a participant may include the name of the agent that
a participant will talk with and the key she will exchange. The environments can also be used to describe security properties concisely. In the
previous example, the agreement between the participants about the key
can be expressed as coincidence between parts of the environments of
participants. We plan to describe security properties that the SET payment transaction protocol should satisfy in our language and to verify
them. We further have to make clear the correspondence between the
original payment transaction protocol used in actual e-commerce and
the simplified version we presented in this paper.
We finally mention some related works. There are a lot of works
applying formal methods to protocol analyses. We will mention a few
languages used to specify protocols in these works. CSP (Hoare, 1985) is
used to specify security protocols in many protocol verification systems
(Schneider, 1997; Roscoe, 1995). It seems that protocol specifications in

A role-based specification of the SET payment transaction protocol

11

our language can be easily translated into a collection of CSP processes

and that tools for CSP can be used to verify the security.
Cervesato (Cervesato, 2001a; Cervesato, 2001b) proposed a protocol
specification language, called Typed MSR. It is a kind of multiset rewriting system. His language also uses role-based descriptions. Protocol
specifications in our language are more concise than those in his language because, in his language, predicates that correspond to the state
of each participant must be explicitly written.
There are some works on security analyses of the SET protocol. Lu
and Smolka (Lu and Smolka, 1999) formally specified the protocol as
CSP processes and verified five correctness properties of the protocol
using the FDR (Formal Systems Ltd, 1998) model checker. They however did not analyze dual signature and did not assume the existence of
intruders in their analysis.
Meadows and Syverson (Meadows and Syverson, 1998) developed a
security specification language for their protocol analyzer (Meadows,
1996). They also discussed the security properties that the SET payment transaction protocol is expected to satisfy. However, they did not
give the specification of the protocol formally, and they left the actual
verification of the security for future work. As far as we know , no result
on the verification has been published yet. Our specification can serve
as a starting point of a formal verification of security properties they
discussed.
Bolignano (Bolignano, 1997) proposed a method to analyze security
protocols. He took a protocol that resembles SET as an example. He
has not completed the analysis of SET itself as far as we know.
Bella et al. (Bella et al., 2000) analyzed the cardholder registration
protocol in SET. The protocol is used to exchange certificates needed
in the payment transactions. They use the inductive method (Paulson,
1998) for their analysis.
Kessler and Neumann (Kessler and Neumann, 1998) defined a logic to
treat the accountability of participants in electronic commerce protocols.
They used their logic to analyze the accountability of a merchant in SET.
They took into account dual signature, although they treated only the
PReq message.

Acknowledgments
The authors thank Kazuo Ohta, Akira Takura, and Kiyoshi Shirayanagi for their helpful comments and encouragement.

12

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

References
Anderson, R. and Needham, R. (1995). Programming satan's computer. In Computer
Science Today: Recent Trends and Developments, volume 1000 of LNCS, pages
426-440. Springer-Verlag.
Bella, G., Massacci, F., Paulson, L. C., and 'Tkamontano, P. (2000). Formal verification of cardholder registration in SET. In 6th European Symposium on Research in Computer Security (ESORICS'00), volume 1895 of LNCS, pages 159-174.
Springer-Verlag.
Bolignano, D. (1997). Towards the formal verification of electronic commerce protocols. In 10th IEEE Computer Security Foundations Workshop, pages 133-146.
Burrows, M., Abadi, M., and Needham, R. (1990). A logic of authentication. ACM
Transactions on Computer Systems, 8(1):18-36.
Cervesato, I. (2001a). Typed MSR: Syntax and examples. In Information Assurance
in Computer Networks: Methods, Models, and Architectures for Network Security
(MMM-ACNS'01), volume 2052 of LNCS, pages 159-177. Springer-Verlag.
Cervesato, I. (2001b). Typed multiset rewriting specifications of security protocols.
In 1s Irish Conference on the Mathematical Foundations of Computer Science and
Information Technology (MFCSIT'00), ENTCS. Elsevier. To appear.
Clark, J. and Jacob, J. (1997). A survey of authentication protocol literature: Version
1 .0. Technical report, Department of Computer Science, University of York.
Denker, G., Millen, J., and Rue, H. (2000). The CAPSL integrated protocol environment. SRI Technical Report SRI-CSL-2000-02, SRI International.
Dolev, D. and Yao, A. C. (1981). On the security of public key protocols (extended
abstract). In 22nd Annual Symposium on Foundations of Computer Science, pages
350-357. IEEE.
Formal Systems Ltd (1998). FDR2 user manual.
Hoare, C. A. R. (1985). Communicating Sequential Processes. Prentice Hall.
Kessler, V. and Neumann, H. (1998). A sound logic for analysing electronic commerce protocols. In 5th European Symposium on Research i n Computer Security
(ESORICS'98), volume 1485 of LNCS, pages 345-360. Springer-Verlag.
Lu, S. and Smolka, S. (1999). Model checking SET Secure Electronic Transaction
Protocol. In 7th International Symposium on Modeling, Analysis and Simulation of
Computer and Telecommunication Systems (MASCOTS'99), pages 358-365. IEEE.
Meadows, C. (1996). The NRL protocol analyzer: an overview. Journal of Logic Programming, 26( 2):113-131.
Meadows, C. and Syverson, P. (1998). A formal specification of requirements for
payment transactions in the SET protocol. In Financial Cryptography '98, volume
1465 of LNCS, pages 122-140. Springer Verlag.
Needham, R. and Schroeder, M. (1978). Using encryption for authentication in large
networks of computers. Communications ofthe A C M , 2 1 ( 1 2 ) : 9 9 3 - 9 9 9 .
Paulson, L. C. (1994). Isabelle: A Generic Theorem Prover, volume 828 of LNCS.
Springer-Verlag.
Paulson, L. C. (1998). The inductive approach to verifying cryptographic protocols.
Journal of Computer Security, 6(1):85-128.
Roscoe, A. W. (1995). Modelling and verifying key-exchange protocols using CSP and
FDR. In 8th IEEE Computer Security Foundations Workshop, pages 98-107.

A role-based specification of the SET payment transaction protocol

13

Schneider, S. (1997). Verifying authentication protocols with CSP. In 10th IEEE

Computer Security Foundations Workshop, pages 3-17.
SET Secure Electronic Transaction LLC (1997a). SET secure electronic transaction
book 1: Business description.
SET Secure Electronic Transaction LLC (1997b). SET secure electronic transaction
book 2: Programmers guide.
SET Secure Electronic Transaction LLC (1997c). SET secure electronic transaction
book 3: Formal protocol definition.

This page intentionally left blank.

INFORMATION SECURITY: MUTUAL

AUTHENTICATION IN E-COMMERCE

S.H. Von Solms

Department of Computer Science
Rand Afrikaans University
PO Box 524, AUCKLAND PARK, 2006
South Africa
Tel: +27 11 489-2847 Fax: +27 11 489-2138
[email protected]

M.V.Kisimov
Department of Computer Science
Rand Afrikaans University
Johannesburg, South Africa
Tel + 27 11 673-0163 Fax + 27 1 I 673-0163
[email protected]

Abstract:

Information Security is ever increasingly becoming an important topic when it

comes to network communications. This greatly concerns areas of electronic
commerce, especially online shopping and money transfers. This paper outlines
a methodology for securing electronic communication between e-Merchants
and online shoppers. T h e methodology is based on a simple hierarchy of a
trusted third party and communicating hosts. The paper further explains how
the new methodology avoids e-commerce pitfalls of current technologies and
presents an approach for securing currently unsecured online shoppers, in the
process of making them capable of performing safe and secure network
transactions.

Keywords:

Certification Authority, Authentication, Guideline, Security, Digital

Certificates, Encryption, Asymmetric Cryptography, Digital Signature.

16

1.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

INTRODUCTION

In an ever-improving technological world, e-commerce is becoming an

increasingly popular a tool for communication, business and analysis.
Consequently the value of information being transmitted and its preservation
is of high importance to its owner. This paper presents a new methodology,
based on current technologies, which avoids security pitfalls to which current
e-commerce standards are prone. This methodology deals with outlining a
clear process for securing an average online shopper, with the necessary
attributes needed for performing a safe online transaction. Further it defines
an authentication process for verification of communicating parties
identities, using a trusted third party in the form of a Certification Authority
(CA). As a result this methodology provides a legal process for creating
nonrepudiation of performed transactions, which can be used in verifying the
origin and the occurrence of a transaction.

1.1

Outline

Section two of this document looks at background work done to improve

authentication between communicating parties as well as focusing on current
electronic commerce problems and security loopholes. Section three of the
document outlines the proposed methodology, which is the main focus of this
document. Finally section four serves as a logical end to the document
summarizing the important points made throughout.

2.

BACKGROUND AND SECURITY PROTOCOLS

This section presents certain pitfalls of current e-commerce strategies and

standards, in terms of security, customer satisfaction, authentication and
technological standards. It will further present certain security weaknesses of
the SSL protocol, which can be exploited by malicious parties. The points
discussed here, present an obstacle to companies and individuals in
establishing proper standards for electronic commerce and information
security.

Information Security: Mutual Authentication in E-Commerce

2.1

17

Customer Satisfaction

In a recently conducted study [PWC 98], statistics vital and worrying to

corporations conducting business over the Internet as well as to online
shoppers have emerged. The study showed that 60 percent of initiated online
transactions are abandoned due to lack of online support, necessary security
measures and lack of a standardised legal process for completing the online
transactions. The ratio of completed to initiated transactions should be very
discouraging to online merchants. Problems arising due to complex
techniques and unproven technologies, often lead potential customers
dropping transactions midway through and searching for different online
merchants. E-Merchants, who present customers with long and extended
processes for completing transactions, are usually the ones to suffer from lost
business [PWC 98]. Security is of high concern to as many as 58 percent of
online shoppers and only fewer than 10 percent of online shoppers are not
concerned with security while performing a sensitive Internet transaction.

2.2

Online Digital Certificate Verification

Research performed by the authors reveals that many commercial products

used for online transactions, which employs asymmetric cryptography and
Digital Certificates (DCs) as method of encryption and authentication, over
unprotected networks, do not provide methodology for online verification of
these DCs. The need for such verification is based on the fact that Digital
Certificates can be tampered with, corresponding private keys can be lost or
compromised. This can cause information secured with these keys, to be
compromised and to become volatile to malicious security attacks. Currently
existing Certification Authorities (CAs) and PKIs such as VeriSign and
Entrust [CTNS 00], [VS 01] implement special Certificate Revocation Lists
(CRLs) [BPKIC 01], which hold a list of certificates, which are registered or
issued by the CA or PKI. These lists represent DCs, which have been
compromised in any manner. A verification of the DCs in use between
communicating parties, in the issuing CA's CRL will confirm that in fact,
these certificates have not been reported to be compromised. This can serve
as a verification of the security of the data being transmitted. Such
verification is not a property of any of the commercial products, which
concern themselves with digital, network-based communication. Taking the
problem further, if a certain certificate has been compromised, but the
tampering has gone undetected to anyone, this certificate would not be
reported to the CA and consequently not listed in the CA's CRL. This would
leave any communication employing this DC compromised.

18

2.3

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Authentication of online customers

Credit card fraud is a common occurrence for e-Merchants [PWC 98].

Reasons for fraud vary from lost credit cards, falsely generated credit
information and duplicated or stolen credit cards being passed to the
Merchant. Currently true authentication of the online shoppers is not always
possible. Very few commercial or other products are in place, which deal
with authentication of communicating parties over an open network. The
latest version of the SSL protocol [SSL 96] provides for the possibility of
such authentication. This however is not a prerequisite for the functionality
of SSL. This leaves an opportunity for fraud on the side of a malicious online
shopper. The fact that the e-Merchant cannot certainly authenticate a client,
is enough for attempts at credit card fraud to be a persisting problem.
Resulting statistics [PWC 98] show that credit verification systems are not
advanced enough, resulting in false credit information being accepted as
genuine. This inexorably hurts financially any e-Merchant having accepted
fraudulent information as well as hurting unsuspecting people, whose credit
information is in the possession of a malicious party.

2.4

Security Protocol Characteristics and Exploits

Current e-commerce trends [PWC 98] for securing Internet transactions

reveal that the SSL protocol is seen and used by e-Merchants as the more
secure alternative in providing a secure channel for transmission of sensitive
information between online shoppers and electronic Merchants. The set of
procedures provided by SSL allow for different options for securing and
authenticating communicating parties [SSL 96]. There are three different
options, which the protocol supports for the purpose of authentication:
- Anonymous communication; no authentication of any of the
communicating parties.
- Server authentication; only the digital certificate of the server (eMerchant) is transmitted to the client for authentication.
- Complete authentication; there is a mutual exchange of certificates
between client and server.
The second and third option as listed above of the authentication process
provide for a relatively sound structure for verification of e-Merchant
(server) identification. The weakest option of the three listed is the
anonymous connection between communicating parties, where no certificates
are exchanged and thus no authentication is possible. This scenario is
vulnerable to man in the middle attacks [SSL 96]. This can present a great
cause for concern to any online shopper, as this weakness, if exploited

Information Security: Mutual Authentication in E-Commerce

19

properly can result in unsuspecting persons or entitys, credit information to

be transmitted to a malicious third party, pretending to be a genuine eMerchant [SON 97].

2.5

Secure Electronic Transaction (SET)

Based on the development of new technology such as smart cards, a new

security protocols SET has emerged, whose purpose is to provide
authentication and secure transactions between communicating parties [SET
97]. The main goal of SET is to allow specific cardholders and properly
equipped web merchants to perform business transactions over an open and
unprotected network. Such transactions, similar to many security payment
protocols, are based on use of a set of cryptographic techniques, for the
purpose of secure communication. The protocol further introduces a new
approach to digital signatures, although it does not introduce any new
algorithms or technologies. This approach sees the concept of dual
signatures. This is done with the purpose of encapsulating an eventual
payment to a merchant directly to the clients bank, as well with the purpose
of creating an offer for goods or services to the merchant. If this offer is
accepted, the merchant receives the full amount decided upon into his bank
account, without being aware of the customers credit particulars. In the same
breath, the bank is not aware of the types of goods or services being
purchased, or of their individual cost. This is all possible, with the existence
of specific client and merchant side certificates. These are issued by each
financial institution, which issues the credit smart card to clients and is in a
relationship with the specific web merchant. The client certificate is stored
on the clients smart card, but this certificate is optional and not compulsory.
This coupled with the fact that not too many individuals are in the possession
of a smart card reader, or in the case where they attempt to purchase goods or
services online from a different form their own computer, this protocol, will
not function properly in terms of authenticating the client as required by the
protocols functionality, thus presenting problems often encountered by web
merchants. Such problems deal with trust in the funds and validity of the
credit information provided, as well as the fact that the credit information
may be valid, but stolen from its original owner.

2.6

Summary

This section presented certain security weaknesses and methodologies, which

can be found in current e-commerce practices. The main concerns addressed
here represent a low level of customer satisfaction of e-traders, based on
poorly designed and implemented online trading practices, weak security

20

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

measures for transmission of sensitive information, as well as lack of

standardised practices for electronic transactions.

3.

PROPOSED MODEL

3.1

Introduction

The proposed model outlines solutions for the problems encountered and
described in the previous section as well as adding some extra features,
which improve the overall security of the model. The model presents a
methodology called Trusted Third Party (TTP), for securing a totally
unsecured client, willing to perform online purchases, the authentication of
communicating parties during this online transaction, as well as a secure
transmission of sensitive information between them in the process of
completing the online purchase.

Fig 1.

3.2

Overview

The figure above describes the functionality of the Methodology presented

here. An online transaction is usually initiated with an online shopper visiting

Information Security: Mutual Authentication in E-Commerce

21

the desired e-Merchant's web site (step 1). At this point the Merchant's
server initiates a SSL session as specified in [SSL 96], with the online
shopper. Once the initial SSL handshake procedure is initiated and the Server
DC is delivered to the Client, the Server requests similar DC from the Client
(step 2). If the Client is not in a possession of such certificate (step 3), the
SSL session with the Client is interrupted and the client is notified that he/she
needs to perform certain steps in order for the transaction to be secured. If he
does decide to take up these steps, the shopper is redirected to the trusted
CA's web site (step 4), while his session with the Merchant remains frozen.
At this point the root certificate of the trusted CA is delivered to the shopper
(step 5), followed by a small application, which is too installed at the client's
machine (step 6). Immediately after that a Java applet is delivered to the
client (step 7), which communicates with the installed application from step
6 and generates two pairs of asymmetric keys, followed by the generation of
corresponding DC. This completes the securing of the client and is followed
by resumption of the frozen Merchant session. This sees a different Java
applet delivered to the client (step 8) used for credit information gathering
and its encryption by the client residing application, as well as its
transmission to the Merchant (step 9). Steps 8 an 9 do not follow through
from entity to entity. This is done with the purpose of representing multiple
transmissions of data between clients and merchant, once a secure
communication between the two has been established and the appropriate
authentication has been performed on either side.
3.2.1

Trusted third party

Based on the principal of trust, the trusted third party does not participate in
any online transactions. Its sole purpose is to provide means of authentication
and encryption for other entities, in order for them to be able to perform
secure transactions over an unprotected network. Such attributes are provided
by Certification Authorities [BPKIC 01]. The trusted third party within this
methodology will be referred to as Master CA. The Master CA, consistent
with the requirements of a CA, has a root certificate. One difference, which
is vital to this section, is to mention that the root certificate of the Master CA
is not self-signed, which is generally the practice of most well known CAs, it
however is cross certified by a third party CA, which does not belong or is
connected in any way to the Master CA.

22

3.2.2

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

E-Merchant

An e-Merchant is an online trader, providing sale of products or services.

The Merchant requires payment in terms of a specific monetary currency, in
return for his product or services. In most cases this payment is in the form of
credit information, which is transmitted by the client to the Merchant. In
order for the e-Merchant to be authenticated, he will need to have two Digital
Certificates, holding the public keys for encryption and digital signature
respectively. Intuitively, the private keys for those two corresponding public
keys need to be in the possession of the Merchant and nobody else. The DCs
are registered with the Master CA, with an appropriate chain of trust, and
listed in this Master CAs Public Directory, if the Master CA itself has not
issued them. The DCs need to be listed in the Master CA directory if not
issued by the Mater CA, in order for the Master CA, serving as a point of
trust, to be able to verify, the identity of the Merchant. The chain of trust to
such a Digital Certificate needs to verifiable, in order for the Master CA to be
able to trust its origin.

3.2.3

Online Shoppers (Clients)

These are people or entities, which wish to perform online transactions, in

the form of purchases, from authentic e-Merchants. In order for an online
shopper to be able to provide his or her sensitive credit information to the eMerchant, he or she will require attributes similar to the Merchants. These
will be two pairs of public/private keys, for the purposes of digital signing of
data and encryption respectively. The public key of each respective pair will
need to be encapsulated in a Digital Certificate, which is either issued by the
Master CA and thus signed by it, or is issued by any other CA with a
verifiable chain of trust, and as with the e-Merchant scenario listed in the
Master CAs Public Certificate Directory.

3.3

Initial Steps

The previous subsection described the minimum attributes required by two

parties, in order for a secure communication to be established between them.
The described scenario involved the introduction of a trusted third party,
which does not take any part of any possible transactions involving an eMerchant and an online shopper. Even though the online shopper and the eMerchant are equipped with the necessary attributes to complete a secure
online transaction, the two parties dont have a methodology in place, which
will employ these attributes in a correct manner. Existing methodologies
such as SSL have certain pitfalls, such as no online verification of DCs in

Information Security: Mutual Authentication in E-Commerce

23

CRL, determining chain of trust of used certificates as well as guarantee of

an existing standard for processing online transactions. Having said all that,
most online shoppers are not equipped with any of the Listed minimum
attributes. Shoppers are purely restricted by the use of an Internet Browser
(IB) and their concern of security of the transaction.

3.4

Obtaining the Master CAs Root Digital Certificate.

Once the online shopper is redirected to the Master CAs web site, the CAs
Server detects his Internet Browsers make. That done, the shopper is further
redirected (the whole process is automated) to download the Master CAs
Root DC, which is cross certified by the maker of the shoppers Internet
Browser. This done, the shoppers Internet Browser verifies the digital
signature of the cross certifying third party CA (not the Master CA). This is
possible, because each Internet Browser comes with the root certificate of the
maker of the IB. This coupled with the fact that the root certificate of the
Master CA is cross certified by the private key of the maker of the online
shoppers IB makes this verification possible. From this point onwards the
following procedures become more automated.
3.4.1

The Master CAs root certificate

Following standard asymmetric cryptography techniques, in order for a

Digital Certificate to be generated there needs to be a public key of a
publid/private key pair encapsulated in it. The key pairs for the root
certificate of the Master CA are generated using the standard RSA algorithm
[PGP 95]. Use of other approved asymmetric algorithms can be equally as
effective. The key length is of 2048 bits size. The private key of this pair is
always kept with the Master CA. The public key is distributed to all known
Internet Browser manufacturers, who based on it generate a Digital
Certificate, which is signed with their own private key. Employing this
technique the online shopper can be asserted that the received Master CA
root certificate is indeed authentic and not fraudulent. Such approach can
prove to be expensive, but it server to right purpose of secure transmission
and identification of origin of transactions.

3.5

Background process

Once the Master CAs Root Certificate has been installed, any file or
application signed with the private key of the Master CA will be guaranteed
and be verifiable by the online shopper to be authentic and non malicious.
This is used for the base of downloading a small application, which is

24

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

installed and run on the clients machine. The application runs as a

background process to the IB and is active throughout the whole time the
clients machine is powered on. This application brings with itself the root
certificates of the major Certification Authorities from around the world.
These certificates are not hard wired into the application and are
exchangeable, once they expire or become compromised. The application has
networking capabilities, which will be discussed in the following sections.
As part of the installation process, the application gives security permission
to Java applets to interact with this background process. At no time however
will an applet be able to control the background process.

Application

<-

-1- - --

>

-----

MasterCA
-..
__---

Fig 3.
The applet will simply be able to pass information to the process in the form
of structured data.

3.6

Key pair generation

3.6.1

Review

The next step of the process sees the download of the application and its
installation followed by continuation of the connection with the Master CAs
server. After the installation procedure of the background process is
complete, the client is redirected by the CAs server to download a Java
applet.

3.6.2

Key generation and Digital Certificates

This applet is signed by the Master CAs private key. The purpose of this
applet is to generate two pairs of keys using the RSA algorithm, or a similar
asymmetric algorithm. These key pairs have the purpose of encryption and
digital signing respectively. Once the applet is downloaded, its digital
signature is verified by the IB. Following this, the two pairs of keys are
generated. The public keys are passed to the background process, which
signs them with the just generated signing private key, encrypts them with
the public key of the Master CA, obtained from its certificate and passes

Information Security: Mutual Authentication in E-Commerce

25

them back to the applet. The applet sends this encrypted information back to
the Master CA, which decodes this data and verifies the digital signature.
Digital Certificates are created, encapsulating these public keys. The
certificates are listed in the Master CAs Public Directory as well as these
DCs being sent back to the applet, which together with the corresponding
private keys are passed to the background process, for the purpose of storage
and further use.
3.6.3

Communication between applet and background process

The communication between the applet and the background process is

possible due to the fact that the applet has security permissions to
communicate with this process. Before any communication between
background process and applet is performed, the application verifies the
digital signature of the applet, for reconfirmation of its origin. The
communication between the background process and the applet is
emphasized in figure 4..

Shopper

Fig 4.

3.6.4

Summary

This is the last step for securing a client in preparation for secure
communication with a possible e-Merchant. This completes the process of
establishing the basis of a methodology for secure and correct authentication
of communicating parties, as well as for secure transmission of sensitive data
over an open network. It is important to note that the process of securing the
client, can be performed by anybody willing to adopt the methodology of
secure communication as offered by the Master CA. This process does not
have to be initiated by an e-Merchant who detects insufficient security on a
clients machine; any concerned online shopper can initiate it.

26

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

3.7

Communication between shopper and e-Merchant

3.7.1

Download of Merchant applet

Once the securing of the client is complete, there is no need for the abovedescribed procedures to be repeated ever again. The following step can be
part of a resumption of a frozen session between a previously unsecured
client with the e-Merchant, or as an initial step for submitting sensitive credit
information by the client to the Merchant in completing the online
transaction. This next step sees a Java applet downloaded from the eMerchants web site to the clients machine. The purpose of this applet is to
collect sensitive credit information from the client and return it to the
Merchant.

3.7.1.1
The Java applet
The Java applet is signed by the Master CAs private key. The applet takes as
an external component the Merchants Digital Certificate. The applet further
has security permissions to communicate with the clients background
process in the same manner described above as with the communication
between the applet used for key pair generation by the Master CA.
3.7.2

Merchant Authentication

Before any sensitive information is entered by the online shopper in the

downloaded applet, the IB first verifies the digital signature of the applet.
Following this, the applet passes the Merchants DC to the clients
background process. This triggers an authentication procedure by the clients
background process:
- Verification of the chain of trust of the Merchants digital certificate.
- Online check of the Merchants DCs ID in the issuer of this DCs
CRL.
- Final online procedure, involving download of the Merchants DC
from the issuing CAs Public Directory. Then at the clients machine,
a verification, of the main attributes of the DC,
DC, of the newly
downloaded certificate versus the one received from the applet is
performed.

Information Security: Mutual Authentication in E- Commerce

27

In the possibility that at any authentication step yields a negative result, this
would indicate an attempt for a security breach by a malicious party and the
transaction is discontinued.

3.7.3

Information encryption

Once the authentication process on the client side is complete the shopper is
prompted to enter credit information in the downloaded applet. This
information is then passed to the clients background process, which includes
the shoppers DC and encrypts the whole package with the Merchants public
key, signs it with his Digital Signature private key and passes it back to the
applet which in tum transmits it to the online Merchant.

3.7.4

Client Authentication

Once the encrypted data is received, alongside with the Clients Digital
Certificate, a chain of trust is established if possible, based on the existing
trusted root certificates on the Merchant side. Following this, the certificates
authenticity is checked in the issuing CAs CRL as well as this certificates
validity is checked, by downloading this certificate from the CAs Public
Directory and performing a comparison versus the certificate transmitted by
the Client. If this authentication process does not run into any problems then
credit card information, once decoded is verified using appropriate channels.
This completes the transaction and the online shopper is notified of the fact
that his/hers transaction has been performed or not.

3.7.5

Summary

The methodology outlined in this section (TTP), represents an effective

process for secure authentication of two parties over an open network. The
point of trust is a basic CA, which has established chain of trust. The
structure of the methodology is such that it does not allow for anonymous
communication between two parties, as complete mutual authentication is
required before a transaction can be performed. The drawback of the
proposed methodology is that it will affect the performance of any secure
transaction between hosts and it will require a permanent connection to Cas,
for the purpose of CRL verifications. The methodology avoids common
pitfalls displayed by implemented technologies now in practice and thus is
liable to raise consumer confidence in online trade.

28

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

4.

EXTENDING THE MODEL

4.1

Weaknesses of proposed model

The above described model serves the purpose of securing a client with the
required attributes for him or her to be able to perform a secure and
authenticated transaction with another party, equipped with similar attributes,
in a semi transparent manner. One of the attributes referred to is a pair of
Digital Certificates. These certificates are issued to every client who has
applied for them using the described above model or simply on his own
initiative applied for them. These two DCs have the purpose of creating a
digital identity for an applicant. This digital identity is created and based
upon information provided by each applicant. This varies from first name
and email address to surname and place of birth. Such information can be
easily falsified and thus the digital identity based on it becomes
untrustworthy. Examples of such digital identities, based on unverified
information are represented by most level one certificates issued by most
public CAs, to the general public.

4.2

Creating Trust

It becomes clear from the previous section that verification of user identity
becomes vital to the proposed models functionality. Such authentication of
user identity can only be performed by the trusted third party and that is the
Master CA. User authentication can be performed by a physical verification
of the user details, into public records or relative government departments.
This is assuming that the Master CA is based or has representation in each
country, in which it has clients or applicants. Even if this was the case,
physical verification of an applicants identity would take a reasonable
amount of time, far beyond what would be considered seamless and
transparent process for securing a client, as specified by the proposed model.
This would obviously not fit easily or at all in the described scenario of
previous sections and would seriously impede the theoretical and practical
flow of this methodology. Based on this, the need for an institution, which
can easily and quickly verify the identity of an applicant, is required.
Considering the fact that an applicant is at the point of purchasing goods or
services, before he or she is redirected to the TTP, it must be apparent that
this applicant is in the possession of some credit information such as a credit
card, which is issued by a reputable financial institution, which are banks in
most cases. Such institutions have performed a certain degree of

Information Security: Mutual Authentication in E-Commerce

29

identification of applicants, based on the fact that all applicants go through a

thorough identification process before being issued with appropriate
purchasing attributes, such as chequebooks or credit cards. This verification
process is already performed at the point of a client wanting to purchase
goods or services from and E-Merchant, as he or she is in the possession of at
least a credit card.
The placement of trust in an issued digital identity in the form of a certificate,
by the TTP, can be done by verifying that the person holding the creditpurchasing attribute is the same as the one to whom this credit attribute was
issued. Such verification would confirm the identity of the applicant for a DC
and thus place a great trust in the issued certificate. The trust in this
certificate will be great because the process for issuing credit purchasing
attributes such as credit cards, requires an extensive and thorough
identification of the applicant, his or hers financial status as well as previous
credit history. The above mentioned verification of credit attributes such as
credit card versus the identification details of the applicant for a DC
represent a simple mach of these two pieces of information in the financial
issuing authoritys database. This issuing authority could be represented by a
bank but does not necessarily have to be.

4.2.1

Verification details

The specific details required for an applicant to be issued with a high trust
certificate, deal with specific purchasing attributes e.g. credit card number,
coupled with the card owners name and a specific secret key. Such a key can
be the pin number for this card or some secret code known only by the
financial institution and the card owner. This would verify that the card is not
merely stolen but it is the possession of its rightful owner. Such supply of
information would be necessary for any other purchasing attribute apart from
a credit card.

4.2.2

Finalization of authorization

Once all the required information is supplied to the TTP this data is passed to
the relevant financial institution or issuing authority of purchasing attributes,
such as credit cards, in order for this data to be verified. Once this
information is verified, the financial institution can vouch for the identity of
the applicant. This will place a very high trust in the resulting Digital
Certificate(s). Following this approach, a financial institution such as a bank,

30

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

or going higher up the hierarchy a credit card issuing enterprise such as

Master Card or Visa, can serve as a authenticating parties in cooperation with
a TTP in the process of creation and implementation of the proposed model.

5.

CONCLUSION

This paper deals with presenting a new methodology for secure mutual
authentication between two communicating parties over an open network.
The process described here identifies and outlines clear steps for securing
network communications and the data being transmitted in them. It takes a
single point of trust in the name of a Certification Authority and with the
help of small network based applications and applets constructs authentic
way for identifying communicating hosts to one another via the use of
asymmetric cryptography and Digital Signatures. The end result represents
increased consumer confidence in a possible e-commerce environment,
avoidance of current security pitfalls and potential decrease in credit card
fraud.

6.

LIST OF SOURCES CONSULTED

[SSL 96]

[PGP 95]

[BPKIC 01]

The SSL Protocol Version 3.0, November 18 1996

http://home.netsacape.com/eng/ss13/draft302.txt
The official PGP Users Guide, P.R. Zimmermann, 1995, MIT
Press, USA
Basic Public-Key Infrastructure Characteristics, Marc
Branchaud
http://home.xcert.com/-marcnarc//PKYthesis/characteristics Jan
2001

[DS 97]
[SON 97]

Decrypted Secrets Methods and Maxims of Cryptology, F.L.

Bauer, Springer-Verlag Berlin Heidelberg, 1997
Security
on
the
Net,
Eddie
Rabinovitch,
http://www.cosmoc.org/ci/public/1997/mar/internet_column,html
1997

[CTNS 00]
[VS 01]

[SET 97]

The Concept of trust in Network Security, Entrust Technologies

White Paper, August 2000, http://www.entrust.com
Outsourced Authentication Administrators Guide, VeriSign,
January 2001. http://www.verisien.com
Secure Electronic Transaction version 1 .0, May 3 1, 1997, Master
Card & Visa, http://www.visa.com/nt/ecomm/set/setprot.html

Information Security: Mutual Authentication in E-Commerce

[PWC 98] Electronic Commerce/Internet Survey

http://www.pwcglobal.com/extweb/ncursvres.nsf

31

This page intentionally left blank.

SOFTWARE-BASED RECEIPT-FREENESS
IN ON-LINE ELECTIONS
Emmanouil Magkos*
Department of Informatics, University of Piraeus
80 Karaoli & Dimitriou, Piraeus, GREECE
[email protected]

Vassilios Chrissikopoulos
Department of Archiving and Libmry Studies, Ionian University
Corfu, 49100, GREECE
[email protected]

Nikos Alexandris
Department of Informatics, University of Piraeus
80 Karaoli & Dimitriou, Piraeus, GREECE
[email protected]

Abstract

Electronic elections could be a viable alternative for real-life elections in

a democratic society. In traditional elections, a voting booth does more
than allow voters to keep their vote secret. The voting booth actually
requires that voters vote secretly. If the privacy of the vote was allowed
but not required, then a voter could easily sell his vote to a vote-buyer,
or be coerced by a coercer. We present a receipt-free election scheme
without making any hardware or physical assumptions about the communication channel between the voter and the voting authorities. Our
solution is software-based i.e. voters are able to exercise their electoral
rights from their home by using a personal computer with Internet access. The only physical assumption we make use of is an untappable
channel between the two voting authorities that are employed in our
scheme. This scheme satisfies most requirements of a secure electronic
election. We make use of well-known cryptographic techniques such as
time-lock puzzles and blind signatures.

Keywords: Receipt-freeness, electronic voting, privacy.

*Research supported by the Secretariat for Research and Technology of Greece.

34

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

1.

Introduction

Due to the rapid growth of the Internet, electronic voting could be a viable alternative for governmental elections, especially in the case of geographically distributed voters with access to open computer networks. If
not carefully designed and implemented, e-voting systems can be easily
manipulated, thus corrupting election results or violating voters privacy.
In traditional elections, a voting booth does more than allow voters
to keep their vote secret. The voting booth actually requires that voters
vote secretly. If the privacy of the vote was allowed but not required,
then a voter could easily sell her vote to a vote-buyer, or be coerced by a
coercer. All receipt-free schemes met in the literature use hardware assumptions to achieve receipt-freeness. In [15] there are tamper-resistant
smartcards that keep some information secret from the voter. Most other
schemes 1, 2, 9, 10, 12, 13, 14, 19] make physical assumptions about the
communication channel between the voter and the election authorities.
More specifically, they assume the existence of :

Untappable channels from the voter to the authority [13, 14]

rn

Untappable channels from the authority to the voter [l, 9, 10, 19]
Physical Voting Booths [2, 12].

In [9], it is argued that one-way channels from the authorities to the

voters are the weakest physical assumption for which receipt-free voting
protocols are known to exist. We believe that these physical assumptions are unsatisfactory: If the underlying communication structure consists of untappable channels between the voting authority and secure
dedicated machines (where voters vote), then there is no point of quitting the traditional elections. Real life citizens in a democratic society,
who find it inconvenient to go to the polls (and so they finally abstain
from the elections) will find it equally inconvenient to cast their vote
from a physical voting booth in a dedicated computer network. Note
that untappable channels will also force the voter to use specified voting
locations.

Our Contribution.

We present a software-based receipt-free election scheme, which is secure against a coercer who has tapped all the
communication lines between a voter, say Victor, and the voting authorities. Victors vote is a computational time-lock puzzle [17], i.e.,
it requires a precise amount of time (real time, not CPU time) to be
solved, unless a trapdoor information is known in advance. In our election scheme, the trapdoor information is only known to a voting authority. A second authority exists to make sure that votes remain secret

Software-based Receipt-freeness in On-lane Elections

35

until the end of the voting period. A coercer, who wants to find out who
Victor voted for, has no other way than running a dedicated computer
continuously for a certain amount of time. Even if Victor has incentives to prove his vote to a vote-buyer, there are no means to prove it,
since he does not know the trapdoor information. We do not assume
any untappable channels between Victor and the voting authority, or
any hardware devices. The only physical assumption we make use of
is an untappable channel between the two voting authorities that are
employed in our system.
The cost paid for receipt-freeness is that the voter constructs her vote
inefficiently, by repeatedly squaring a given value, for a significantly large
amount of time. However, we believe that this is a minimal tradeoff for
a software-based receipt-free solution. To our knowledge, our scheme
is the only receipt-free scheme in the literature without the physical
assumption of an untappable channel between the voter and the voting
authority. Voters axe able to exercise their electoral rights from their
home by using a personal computer with Internet access. Furthermore,
our scheme satisfies most security requirements met in the literature.

2.

A Model for Software-based Receipt-Freeness

In our model we assume that a coercer may have tapped the communication channel between the voter and the voting authority. It is clear
enough that the vote should be encrypted, for vote secrecy. The trapdoor
information for the encrypted vote may consist of a secret decryption key
and/or the randomness used in a probabilistic encryption scheme. If this
trapdoor is in the possession of the voter (e.g. as in [3, 13, 14]) then it
could also serve as a receipt for the vote. Even if the voter lies about
the encrypted vote [l, 9, 10, 19], a coercer who taps the communication
channel will eventually find out the value of the vote by eavesdropping
on the confidential information exchanged between the voter and the
authority. Note that simple encryption does not serve our purposes: a
coercer will tap the encrypted message as it is being sent from the voter
to the authority (or vice versa), and then require the voter to reveal
the trapdoor information. Even worse, the coercer may demand that
the voter uses some specific randomness. To summarize: the simplest
bit of information that will make the voters life easier during the construction of the encrypted vote, may also make the coercers life easier.
Thus, software-based receipt-freeness in the presence of a coercer who
taps communication lines can only be achieved if the voter does not use
any secret information other than the vote itself.

36

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

We came up with a variation of the time-lock concept, as has been

described by Rivest, Shamir and Wagner [17]. The idea is based on
preprocessing: the voter selects a vote from a set of valid votes and constructs a time-lock puzzle of the vote by repetitively squaring a specific
value (which is not secret). The number of squarings is also a public
parameter. In [17] the user does this efficiently because she knows some
trapdoor information. In our model, the user does not know the trapdoor information. The trapdoor information is possessed by a voting
authority and it will be used at the end of the voting phase to reveal
the cleartext vote. Thus, the voter constructs his vote inefficiently by
executing an intrinsically sequential process.
A coercer, who taps the communication line between the voter and
the voting authority, will get the time-lock puzzle of the vote, as it is
being delivered to the authority. Even with the help of the voter, there
is no way to reverse the time-lock process: the voter does not know
the trapdoor information, so the coercer will have to run a dedicated
computer for a specific amount of time. This time can be determined
by an independent authority who sets the public parameters, e.g. the
number of squarings for each puzzle, so as to prevent massive coercion
in a large-scale election: assuming that each voter performs n squarings,
a coercer will have to perform nk squarings to coerce k voters. However
the voter too constructs his vote inefficiently, but we believe that this is a
minimal tradeoff for a software-based receipt-free solution that does not
employ untappable channels between the voter and the voting authority.
The only physical assumption we make use of is an untappable channel between the two voting authorities employed in our system. This is
acceptable, since our main goal was to abolish the necessity of a physically secure channel between the voter and the authority. We believe
that an untappable channel between two authorities that belong to a
distributed set of voting authorities, is a minimal physical assumption
for a receipt-free scheme. We could remove this physical assumption
by requiring that there is only one voting authority, but in that case,
and unless full trust was granted to this authority, fairness would have
been sacrificed: if the authority possesses the trapdoor information, then
votes may be revealed before the end of the voting period.

3.

Building Blocks

Our voting scheme makes use of blind signatures [4], which is a well
known technique, already implemented with the RSA algorithm [16, 20].
Blind signatures are the equivalent of signing carbon-paper envelopes:
a user seals a slip of a paper inside such an envelope, which is later

Software-based Receipt-freeness in On-line Elections

37

signed on the outside. When the envelope is opened, the slip bears the
carbon image of the signature. Furthermore, users in our scheme lock
their votes in a time-lock puzzle. The mechanism is a variation of a
well-known technique [17] and is presented below.

3.1.

Time-lock Puzzles

Suppose that Alice wants to encrypt a message M so that Bob can

decrypt it after a period of T seconds. T is a real (not CPU) time
period, given that Alice knows (or approximately assumes) in advance
the CPU power of Bob. In [17], Alice generates a composite modulus
n = pq as the product of two large primes p and q. Then, Alice computes
@(n) = ( p - l)(q - 1) and t = T S , where S is the number of squarings
modulo n per second that Bob can perform. Alice chooses a key K for
a symmetric cryptosystem and encrypts M with key K , thus getting
CM = Enc(K, M ) . In order to hide K , she picks a random a modulo n
and encrypts K as:
t
C K = K +a (mod n )
(1)
To do this efficiently, Alice uses the trapdoor information @(n)that
only she knows: She first computes e = 2t(mod a(.)) and then b =
ae(mod n). The public output of the puzzle is the set (n,a,t , C M ,C K ) .
Since Bob does not know the factors p and q, computing @ ( n from
)
n
is provably as hard as factoring n. Bob has no way of computing a2t,
other than starting with a and perform t sequential squarings, each time
squaring the previous result. The computational problem of performing
these squarings is not parallelizable: having two computers is not better
than having one computer.

Our variation.
In our model, Alice is the voter and Bob is the
coercer. Alice does not know the trapdoor information @ ( n )(if she
knew it she could hand it over to Bob, e.g. in a vote-selling scenario),
so she cannot construct the puzzle efficiently. In addition, there are two
voting authorities. The first authority selects n, p and q, and publishes
n and t , where t is the number of squarings that Alice has to perform.
Alice selects a as previously and computes a2t. Alice's vote v takes the
place of the key K in equation (l), thus yielding:

C"

=v

+ a2'(mod n )

(2)

The public information will now be the set ( n ,a ,t , Cv). When the time
comes, Alice uses a clear channel to submit the time-lock puzzle of her
vote to the second voting authority. The first voting authority, who
possesses the trapdoor information is(n),will later cooperate with the

38

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

second authority to decrypt the submitted votes. In Section 4, our voting

protocol is presented in detail.

An efficient solution with secure hardware.

Another solution would be each voter to be equipped with a tamper-resistant smartcard. During an off-line registration protocol, this smartcard would
be provided with the trapdoor information @ ( n ) .Later, during voting phase, the voter would provide the smartcard with her preferable
vote, and the smartcard would use the trapdoor information @(n)to
construct the time-lock puzzle in an efficient way. To reveal the vote,
the coercer would either have to tamper with the smartcard or solve the
time-lock puzzle.

4.

A Receipt-free E-voting Scheme

In our protocol there are N voters and two authorities, the Registrar
and the Voting Center. The Registrar acts as an intermediate between
the voter and the Voting Center, while the Voting Center is responsible
for tallying the votes. We assume that each authority is semi-trusted [7],
i.e., the authority may misbehave but will not conspire with another
party. We also make use of a bulletin board, which is publicly readable.
Only the Voting Center can write to it, and nobody can delete from it.
The Voting Center is committed to everything that is published on the
bulletin board.
There is a certificate infrastructure and all participants are legally
bound by their signatures. Voters and authorities possess a private/public key pair for signature and encryption as well as the corresponding
certificates, issued by a trusted Certification Authority. We also assume
that there is an untappable channel between the Registrar and the Voting Center. Communication between voters and authorities takes place
through an anonymous channel: voters can send/accept messages that
cannot be traced (e.g., by using traffic analysis). For example, e-mail
anonymity can be established with Mixmaster re-mailers [ 5 ] , and HTTP
anonymity can be established with services such as the Onion-Routing
system [8]. The election protocol is depicted on Figure 1. It is split into
four stages, the Authorizing stage (Steps 1-2), the Voting stage (Steps
3-5), the Claiming stage (Step 6) and the Tallying stage (Steps 7-8).

Authorizing Stage.
A voter, say Victor, wishes to get a certified
pseudonym that will identify him to the Voting Center. Victor creates
a private/public key pair (SKps, P K p s ) , blinds P K p s (the public tallying key) to create the blinding bl and then signs a message consisting
of bl and the unique election identification number Electid (Step 1).

Software- based Receipt-freeness in On-line Elections

39

No tat ion
RE = the Registrar
V = the voter's

true

identity

VC = the Voting Center

PS = the voter's pseudonym

blind(m) = blinding of a message m
PKx = the public signature/encryption key of X
[m] 'liX=public key encryption of m with PKx
SIGvr[TLP(vote)]

(4)

~~,

TLP(B) = time-lock puzzle of B

SKx = the secret signature/encryption key of X

SIGV[V.Elect-id]

SIGx[m] = signature on m with the key SKx

Elect, = the election's unique ID numbcr

shrdfedshuff l e d

.......... ............1

v v

=1..N

SIG",. SIGps,, TLP(vote,)

.............................................................

(7)

ii

.......... D

........Wn)
...... .b
I

CERT(PKx) = the public key Certificate of X

16)

@(n)

= trapdoor

__.*Normal

information for the puzzle

Channel

- - - - +Anonymous channel
.............
............. bUntappable Channel

shh f f 1 ed

Figure 1. A receipt-free election scheme (software-based)

Victor sends these to the Registrar and gets the blinding signed by the
Registrar (Step 2). Victor unblinds the Registrar's signature on bl and
is left with a certificate of the public tallying key, C E R T ( P K p s ) . This
certificate will be used later by the Voting Center to verify signatures
that are made with the secret tallying key, PKps The public key P K p s
will be Victor's official pseudonym.

Voting Stage. In Step 3, Victor, who has already constructed a

time-lock puzzle of his vote, TLP(vote),encrypts it with the public key
of the Voting Center, and signs the result using his secret tallying key,
thus producing SIGps[[TLP(vote)lPKVC].
He anonymously sends this
to the Voting Center, along with the certificate of his public tallying
key. The Voting Center verifies the signature, decrypts the message,
stores the time-lock puzzle in a local database and returns, in Step 4,
a signature on the puzzle, SIGvc[TLP(vote)].This can be seen as a
receipt that the Voting Center has accepted the time-lock puzzle of the
vote. At some time later, in Step 5 , Victor uses his authentic signature
key to sign a message consisting of his true identity V and the Electid
number. He then sends the signature to the Voting Center.

40

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Claiming Stage.
In step 6, the Voting Center publishes on the
board, in random order, the list of the authentic and pseudonymous
signatures S I G x and SIGps,, i = 1, ..., N. The Voting Center also
publishes all the time-lock puzzles TLP(votei) of the votes that have
been successfully submitted. In case Victors time-lock puzzle is not
published, he can protest by broadcasting SIGvc[TLP(vote)],with no
need to reveal in which way he actually voted. This is called an open
objection to the tally, introduced by Sako in [18].
Tallying Stage.

In Step 7, the Registrar sends the secret trapdoor

@ ( n )to the Voting Center, by using an untappable channel. No one,
except the Voting Center, can have access to @(n).The Voting Center
uses @(n)to solve the time-lock puzzles of the votes. In Step 8, the
Voting Center publishes in clear the results of the election, i.e. the list
of the votes vote*, i = 1,..., N . The Voting Center also publishes a list
with the corresponding time-lock puzzles of the votes, TLP(votei) =

[TLP(votel),..., TLP(vote, )].

5.

Security Analysis

We evaluate the security of our scheme by examining some basic requirements, which most researchers seem to agree upon [6, 20]:

Eligibility.

(Only authorized voters are able to vote). In Step 1,

Victor signs a message using his authentic signature key. The Registrar checks the eligibility of each user who submits a tallying key for
certification.

Unduplicability.

(No one is able to vote more than once). The Registrar will not issue more than one tallying keys for each voter. In Step
6, all the authentic signatures of the voters are published. Consequently,
it is not possible to exist more tallying keys than authentic public keys,
so the Registrar cannot misbehave without being caught.

Untraceability.

(All votes remain anonymous). When Victor submits a tallying key for certification, he signs a message and the Registrar
checks his identity. However, the tallying key is blindly signed by the
Registrar in Steps 1-2. Consequently, the Registrar cannot trace any
signature SIGps, published in Step 6 , back to Victors real identity.
Furthermore, Victor in Step 3 uses an anonymous channel to submit
his validated time-lock puzzle. The puzzle cannot be traced back to its
sender, since it is signed under a certified pseudonym (the tallying key).

Software- based Receipt-freeness an On-line Elections

41

The link between Victors pseudonym and his real identity cannot be
done by either authority.

Fairness.

(All ballots remain secret while voting is not completed).

The trapdoor information necessary to solve the puzzle, is in the possession of the Registrar. Victor encrypts the time-lock puzzle of his vote
with the public encryption key of the Voting Center, and sends it to
the Voting Center. The Voting Center will not publish the time-lock
puzzles until the end of the voting period. Fairness is achieved, as long
as the Registrar and the Voting Center do not combine their knowledge.
Neither the Registrar nor the Voting Center can break fairness by themselves. Since the Registrar and the Voting Center are assumed to be
semi-trusted, this requirement is satisfied.

Accuracy. (No one is able to alter/delete anyone elses vote). In Step

6, the Voting Center commits to the time-lock puzzles of all the votes
and cannot alter them, according to the properties of the bulletin board.
Every voter, whose time-lock puzzle has not been taken into account,
can make an open objection to the tally.

Atomic Verifiability. (Voters are able to verify that their vote has
been counted correctly). In Step 6, all the time-lock puzzles of the votes
are published by the Voting Center. Victor can check that his time-lock
puzzle has been published on the board. If not, Victor makes an open
objection: he anonymously broadcasts the receipt that was sent to him
in Step 4.
Receipt-Freeness. (No voter is able to prove the value of its vote).
The receipt freeness property is separately discussed in Section 2. It
must be noted that the scenario of a coercer who observes the voters at
the moment they vote, is not addressed at all. This attack cannot be
prevented by any e-voting scheme and is rather unrealistic in large-scale
elections.
Responsibility.

(Eligible voters who have not voted can be identified). This is an optional requirement, desirable in Australian elections [11]. All voters, who receive in Step 4, an acknowledgment of their
votes from the Voting Center, sign a message by using their authentic
signature keys and send this message to the Registrar, in Step 5 . The
Registrar has already received, in Step 1, the authentic signatures of all
eligible voters, so he is able to identify, by comparing the corresponding
lists, the eligible voters who have not voted.

42

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

6.

Discussion

We have presented a receipt-free election scheme, which satisfies most

requirements of a secure election. We do not assume any hardware
devices or untappable channels between the voter and the voting authorities. We make use of well-known cryptographic primitives that have
been implemented. Time-lock puzzles, while being very difficult in their
solution, are quite efficient in their construction. The problem with our
scheme is that we sacrifice efficiency in order to achieve software-based
receipt-freeness. While the computations during the election are done
quickly and in few steps, the computations made by the voter before the
election (the preprocessing for the time-lock puzzle) are not done in a
reasonable amount of time. This time is determined by an authority, and
has to be long enough to discourage massive coercion of voters. Yet, as
noted in Section 3.1, our scheme could be relaxed to become an efficient
scheme with smartcards. In such case, however, the scheme would be a
hardware-based solution.

References
[l] D. Alpert, D. Ellard, 0. Kavazovic, M. Scheff. Receipt-Free Secure Elections 6.857 Final Project, 6.857 Network and Computer Security, 1998,
http://www.eecs. harvard.edu/~ellard/6.857/final.ps.
[2] J. Benaloh, D. Tuinstra. Receipt-free secret-ballot elections, 26th Annual ACM
Symposium on the Theory of Computing, Proceedings, 1994, pp. 544-553.

[3] R. Canetti, C. Dwork, M. Naor, R. Ostrovsky. Deniable Encryption, Advances

in Cryptology - CRYPTO 97, Proceedings, Lecture Notes in Computer Science,
Vol. 1294 ,Springer-Verlag 1997, pp. 90-104.
[4] D. Chaum. Blind Signatures for Untraceable Payments, Advances in Cryptology
- CRYPTO 82, Proceedings, Plenum Press 1982, pp. 199-203.

[5] L. Cottrell. Mixmaster and Remailer Attacks, http://obscura.obscura.com/~loki/

remailer/remailer-essay. html.

[6] L. Cranor, R. Cytron. Sensus: A security-conscious electronic polling system for

the Internet, Hawaii International Conference on System Sciences, Proceedings,
1997, http://www.research.att.com/~lorrie/pubs/hicss/hicss.html.

[7] M.Franklin, M. Reiter. Fair exchange with a semi-trusted third party, 4th ACM
Conference on Computer and Communications Security, Proceedings, ACM 1997,
pp. 1-6.
[8] D. Goldschlag, M. Reed, P. Syverson. Onion Routing for Anonymous and Private
Communications, Communications of the ACM, Vol. 42(2), pp. ACM 1999, pp.
39-41.
[9] M. Hirt, K. Sako. Eficient Receipt-fie Voting Based on Homomorphic Encyption, Advances in Cryptology - EUROCRYPT 2000, Proceedings, Lecture Notes
in Computer Science, Vol. 1807, Springer-Verlag 2000, pp 539-556.

Software-based Receipt-freeness i n On-lane Elections

43

[10] B. Lee, K. Kim. Receipt-free Electronic Voting through Collaboration of Voter

and Honest Verifier, JWISC 2000, Proceedings, 2000, pp. 101-108.
[ll] Y. Mu, V. Varadharajan. Anonymous Secure e-voting over a network, 14th Annual Computer Security Application Conference, Proceedings, IEEE Computer
Society 1998, pp. 293-299.

[12] V. Niemi, A. Renvall. How to prevent Buying of Votes in Computer Elections,

Advances in Cryptology - ASIACRYPT 94, Proceedings, Lecture Notes in Computer Science, Vol. 917, Springer-Verlag 1994, pp. 141-148.
[13] T. Okamoto. An Electronic Voting Scheme, IFIP 96, Proceedings, Advanced IT
Tools, Chapman & Hall 1996, pp. 21-30.
[14] T. Okamoto. Receipt-Free Electronic Voting schemes for Large Scale Elections,
Workshop of Security Protocols 97, Proceedings, Lecture Notes in Computer
Science, Vol. 1163, Springer-Verlag 1996, pp. 125-132.
[15] A. Riera, J. Borrell, J. Rifa. An uncoercibleverifiable electronic voting protocol,
14th International Information Security Conference IFIP/SEC98, Proceedings,
1998, pp. 206-215.
[I6] R. Rivest, A. Shamir. A Method for Obtaining Digital Signatures and Public-Key
Cryptosystems, Communications of the ACM, Vol. 21, ACM 1978, pp. 120-126.
[17] R. Rivest, A. Shamir, D. Wagner. Time-Lock Puzzles and Timed-Released Crypto,
LCS Technical Memo MIT/LCS/TR-684, 1996, http://www.theory.Ics.rnit.edu/
~rivest /RivestShamirWagner-timelock.ps
[18] K. Sako. Electronic Voting Scheme Allowing Open Objection to the Tally, IEICE
Transactions on Fundamentals of Electronics, Communications and Computer
Sciences, Proceedings, Vol. E77-A( l ) , 1994, pp. 24-30.
[19] K. Sako, J. Killian. Receipt-Free Mix-TypeVoting Schemes - A practical solution
to the implementation of voting booth, Advances in Cryptology - EUROCRYPT
95, Lecture Notes in Computer Science, Vol. 921, Springer-Verlag 1995, pp. 393403.
[20] B. Schneier. Applied Cryptography - Protocols, Algorithms and Source Code in
C, 2nd Edition, 1996.

This page intentionally left blank.

ID-BASED STRUCTURED MULTISIGNATURE

SCHEMES

Chih-Yin Lin, Tzong-Chen Wu*and Jing-Jang Hwang

Institute of InformationManagement,National Chiao Tung University,Hsinchu, Taiwan, R.O.C.
*Departmentof Information Management,NationalTaiwan University of Science and
Technology, Taipei, Taiwan, R.O.C.

Abstract:

The signing structure of a multisignature scheme specifies the signing order for
all signers when signing messages, and any multisignature not obeying the
specified signing order will be verified as invalid. In accordance with the
different responsibilities of the participant signers, the signing structure of a
multisignature scheme could be further classified as the following three types:
serial, parallel and mixed, where the mixed structure is regarded as the mix of
the serial and the parallel. Based on the well-known ID-based public key
system, we will propose three ID-based structured multisignature schemes and
each scheme respectively realizes the serial, parallel and mixed signing
structures. In the proposed schemes, the length of a multisignature is fixed and
the verification of a multisignature is efficient, regardless of the number of
signers involved. Besides, any invalid partial multisignature can be effectively
identified during the generation of the multisignature.

Keywords:

Multisignature, structured multisignature, ID-based public key, signing

structure.

1.

INTRODUCTION

A multisignature scheme allows multiple signers to sign messages in

which all signers have to sign and individual signers identity can be identified from the multisignature [2-6, 8-10, 13-16, 19]. Furthermore, a structured
multisignature scheme [4, 6] is a multisignature scheme that additionally
requires all signers to obey a predefined signing structure when signing
messages, and any multisignature generated without obeying the specified
signing structure will be verified as invalid. The signing structure of a multi-

46

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

signature scheme indicates the signing order among all participant signers
when signing messages, As a consequence, the multisignature of a message
in a structured multisignature scheme is said to be valid when the following
conditions are satisfied: (i) All signers had signed the message; (ii) All
signers perform their signing operations in compliance with the specified
signing structure; (iii) The multisignature and all partial multisignatures
generated during the multisignature generation process have been successfully verified. Typical applications of the structured multisignature scheme
are multisignatures used in a corporate organization or hierarchical environment. For example, a legitimate working report should be signed accordingly
in the order of the operators, the section leader and the department manager.
Signing structures can be classified into three basic types: serial, parallel, and
mixed, where the mixed structure is the mix of the serial and the parallel. For
the serial structure, all signers sign messages in a predetermined sequence,
and hence the generated multisignatures are sensitive to the signing order. As
to the parallel structure, all signers sign messages in a parallel manner and
the generated multisignatures are independent of the signing order. In the
mixed structure, the signing structure is composed by substructures that
could be serial, parallel, or another mixed structure, and the generated
multisignatures are sensitive to the signing order specified in the
corresponding signing structure. Figure I depicts these three types of signing
structures.

( c ) Mired ( e x a m p l e )
__1c

: signing order

: signer

0: the group of signers

Figure 1 - Three types of the signing structure.

ID-based Structured Multisignature Schemes

47

Most of the previously proposed multisignature schemes are irrelevant to

signers signing order, while some others are order-sensitive. The schemes
presented in References 2, 3, 5, 8, 13, 14, and 19 are order-irrelevant, and the
schemes presented in References 4, 6, 9, 10, 15, and 16 are order-sensitive.
Among the order-sensitive schemes, the schemes proposed by Ham and
Kielser [9], Itakula and Nakamura [10], and Okamoto [16] are RSA-like
multisignature schemes in which the signers signing order has to be properly
arranged by different modules of their public keys; otherwise, messages to be
signed might be modularly truncated. Besides, the length of the
multisignature and the verification time required by these RSA-like schemes
varies proportionally with the amount of the signers participated. In 1998,
Doi et al. [6] firstly proposed a multisignature scheme considering the mixed
signing structure. They used structured group identity and proposed two
structured multisignature schemes for common modular RSA-type and
ElGamal-type signature schemes. However, the length of the multisignature
generated by their schemes varies with the number of the signers involved.
Later, Burmester et al. [4] proposed an EIGamal-type multisignature scheme
with a structured public key approach. In their scheme, the secret and the
public keys for each signer could be generated either by a trusted centre or by
cooperative signers using a distributed protocol. Moreover, Burmester et al.
assumed that there exists at least one honest signer for their scheme to be
secure. This assumption is somewhat less practical and incompatible
especially when applying to a delegation scheme, i.e. proxy signature [11-12],
in which the original signer has to consider the threat that all (proxy) signers
in the signing structure may commit frauds or collusions.
Based on the well-known ID-based public key systems [7, 18] , we
propose three structured multisignature schemes whose security is based on
the difficulty of solving discrete logarithm modulo a large composite
(DLMC) [ 1] and factorising a large composite (FAC) [I, 17]. Since ID-based
digital signature and multisignature schemes [7, 18-19] use the identity of the
signer as the public key, our scheme has the advantage that the signature
verification requires no extra interaction for public key verification. The
proposed schemes have the following merits:
( I ) The length of the multisignature is fixed to different messages.
(2) The length of the multisignature is fixed regardless of the number of
signers.
(3) The computation cost required for the multisignature is efficiently
fixed to the amount of signers participated.
(4) Any violation to the signing order will be detected and identified
immediately.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

48

This paper is sketched as follows. After the introduction, we specify the

notations, parameters and signing structures in section 2. In section 3, we will
propose the serial, the parallel and the mixed structure multisignature
schemes. Security analysis is given in section 4. We conclude the paper in
section 5 .

2.

NOTATIONS AND PARAMETERS

Let G=(u,,u2,

..., u,,)

be

a group consists of

signers and

ski = ( s k l i , s k 2 , ) be uis private key. The verification key for the partial
multisignatures generated by ui is named partial verification key and
denoted by P K , . The verification key for verifying the multisignatures
generated by G is named public verification key and denoted by V K , .
For each signer ui E G ,his identity IDi is a message digest of his/her
public identification information li using a one-way hash function, said F,
such that IDi = F ( l i ) . As defined in other ID-based crypto schemes [7, 18191, I i can be a combination of uis name, age, gender, telephone number or
home address, provided that this combination can uniquely identify ui . Note
that a system authority SA is assumed [7, 18-19] for setting up the ID-based
cryptosystem.

2.1

Signing Structure

Two types of notations are used for describing the signing structures.
SER[ ] denotes the serial structure; and PAR[ ] the parallel structure. For
GI =(u,,u2,u3], if the legal signing sequence is < u I , u 2 , u >
3 , then the

corresponding signing structure is SER[u, ,u,,u3} . Another example is, for

G2= { U ~ , U ~ , U ~ , with
U ~ } a mixed signing structure SER[u,, PAR[u2,
u3],u4] , there are exactly two legal signing sequences, which are
<uI,u2,u3,u4
> and <uI,u3,u2,u4
> . Furthermore, we can use a diagram to
represent the corresponding signing structure as in Figure 1 and 2. In the
diagram, each node indicates a signer and each arrow implies the signing
order for the two signers it connects. If an arrow points from ui to u j , it
means u, should sign after ui signs. In the above example group of signers
G , , it can be draw as in Figure 2(a). Notably, in order to facilitate the tasks
performed in the structured multisignature schemes described later, we add
two dummy nodes s and t to the diagram representation where s and t denote
the start node and terminate node, as shown in Figure 2(b). The general
diagrams for a group of serial signers, a group of parallel signers and an

ID-based Structured Multisignature Schemes

49

example diagram for a group of mixed-structure signers are shown (i.e.

SER[ PAR[ul,SER [ u 2 ,u 3 ,u4 11, PAR[u5, u6 1. u7 J ) respectively in Figure 1.

(a)

S E R [ U , . P A R [ U ~ . U ~ ] .( Ub ~) A]d d i n g n o d e s s a n d f r o (a)
0

: signer

0the group of signers

:

4
:
signing order

Figure 2 -The diagrams of an example signing structure.

2.2

System Parameters

SA initialises the ID-based public key system applicable for structured

multisignatures by first preparing the following parameters.
p, q : two large prime integers, where 2p +1 and 2q +1 are also primes.
N : the product of 2 p + 1 and 2q +I that N = ( 2 p + 1). (2 q +1).
w : the product of p and q that w = p.q .
a : a base of order w modulo N.
r : a random number, where r E Z i .
p : p=dmodN.
f,h : two hash functions, where f ( x ) c min( p , q ) and h ( x ) < min( p , q ) .
SA keeps p , q , w and r secret, while publishing N, a , p ,f and h. Note
that f is used to generate the public identities and verification keys and h is
used to produce the message digest of the message to be signed. Throughout
this paper, x-' denotes the inverse of x modulo w.

2.3

Public Verification Keys

SA generates VKG for G and P K , for each ui E G by the following rules.

For serial structure, signers' public identities, i.e. IDi 's, are concatenated,
and for parallel structure, signers' public identities are first sorted then

ADVANCES INNETWORK AND DISTR. SYSTEMS SECURITY

50

concatenated, to be the input of the function f. The output of f is the value of

the public verification key.
Notice that the reason why we sort and concatenate the identities of
signers in the parallel structure is to provide uniqueness of the verification
key. Consequently, the possibility of the existence of two identical
verification keys can be eliminated. To achieve this, we can use a function,
said [ ( ), that takes a variant numbers of values as input, sorts the input
values, and finally outputs the value of the concatenation of the sorted input
values. For example, the output of [(2983,9213,7615,1003,8714) will be
10032983761987149213. For G = ( U ~ , U , , U , , U ~ and
)
its signing structure
SER[u,,PAR[u,,u,],u,] , V K , = f ( I Q II <(iD,,ID,)
II ID,) , P K , =
f ( I D l ) , P K , = f(iDlII I D , ) , P K , = f ( I D l II ID,) , and P K , = V K , =
f(D,II {(iD2,1D3)
II I D , ) .

3.

THE PROPOSED MULTISIGNATURE SCHEMES

The multisignature schemes for serial, parallel and mixed signing

structures are presented respectively. Each proposed scheme consists of three
phases: key generation, multisignature generation and multisignature
verification. In key generation phase, the system authority generates the
private key for each signer. In the multisignature generation phase, each
signer follows the signing structure to sign messages after verifying the
partial multisignatures generated by the preceding signers. Finally in the
multisignature verification phase, the verifier verifies the validity of the
multisignature. Details are given in the followings.

3.1

For serial signing structure

Without loss of generality, assume the group of signers G= ( u ,,u2,...,un}

is associated with the signing structure SER[u,, u,, ..., u , ] . That is, all ui E G
have to sign messages by following the serial order u I , u 2...,
, u, for
generating a valid multisignature. The scheme is stated as follows.

Key generation phase:

SA prepares the partial verification keys and public verification key as
P K , = f(ID1 II ID, II ... II IDi) , for ui E G , i = 1,2 ,..., n , and VKG = PKn .
Then, he performs the following operations:
Step 1. Compute ki by the following equation, for i = l,2,..., n .

ID-based Structured Multisignature Schemes

ki = PK,-' . r mod w .
Step 2. Select a random number k0 ,where k, E Z: .
Step 3. Randomly select skl, E Z L , for i = 1,2,..., n .
Step 4. Calculate sk2, by the equation below, for i = 1,2,...,n .
sk2, =k,-k,-,.skl,modw
Step 5. Securely distribute ski = (skli,sk2,) to uiE G ,for i =1,2, ...,n
Step 6. Compute and deliver w = ako
mod N to uI .

Multisignature generation phase:

Suppose G = {ul ,up...., u, ) , with signing structure SER[u,,u, ,..., u,,] ,
want to generate a multisignature MS for a message m. Each ui , for
i = 1,2,..., n ,performs the signing operations as below.
(for i f 1 ) by
Step 1. Verify the partial multisignature S,-] signed by
testing if

(If the test fails, then the signing process is stopped and ui-] is
reported as a malicious signer.)
Step 2. Compute the partial multisignature Si by

where Si-, is generated by u,-] and So = wh("')mod N .

Step 3. Send Si to u i + ] for
, icn.
The partial multisignature S, generated by the last signer u, is treated as the
multisignature MS generated by G with SER[u,, u 2 ,..., u,] for message m.

Multisignature verification phase:

The multisignature MS of message m that signed by the signing group G
with signing structure SER[u,, u,,...,u,,] can be publicly verified by VK, as:

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

52

LEMMA 1. For any message m and its partial multisignature Si generated by

u, E G , S i = aki
h(m)(mod
N) in the serial approach.

Proof:
Multiplying h(m) to Equation 2 and raising both sides of it to exponents
with
base
a
,
it
yields
a
recursive
relation
ak, h ( m ) = ak,-t h ( m ) s k l , . a h ( m ) s k 2 , (modN) , where the k, E Z i is randomly
chosen. By the above fact and a k o h ( m ) = w ~ (=S,(modN)
~ )
, we can
Q.E.D.
conclude Si = akih'm)(mod N ) by mathematical inductions.

THEOREM 1 . If all ui E G honestly sign the message m by following

SER[u, , u 2 , . . . , u , ] , then the generated multisignature MS will be successfully
verified by Equation 5.
Proof:
Recall that MS = S, and V K G = P K n . By Equation I and Lemma 1 we
can obtain Equation 5

Q.E.D.
THEOREM 2. Any disorder signing operation regarding SER[u, ,u2 ,...,u n ]
will be identified with the probability of ( w - l ) / w.
Proof:
By following SER[uI,u2,.., u,], u i should sign the partial multisignature
Si-] generated by
for message rn after verifying S,-, 's validity. Assume
a disorder operation takes place before ui signs, whether by mistake or
intentionally, that u j , where i c j I n , signs S,-]instead of ui . Then, the
partial multisignature generated hereby will be

Si = S,-,

ski.
1

~
.ah ( 1 n ) . s k 2(mod
N) .

For S: to be successfully verified by Equation 5, it has to satisfy that

Sl = Si (mod N ) , which implies

By Lemma 1, the exponent part of Equation 6 indicates

k,-, . s k l j +sk2, = k , - , .ski, +sk2,(modw).

(7)

53

ID-based Structured Multisignature Schemes

In order for S,! to be valid, two distinct private keys (i.e. ski and s k i ) have
to satisfy Equation 7. Since the values of sk1,'s for all ui E G are randomly
selected and s k 2 , ' s are computed from Equation 2, it is to see that the
probability for ski and ski to satisfy Equation 7 is l / o . Therefore, the
probability for successfully identifying a disorder event is (W- 1)/ w . Q.E.D.

3.2

For parallel signing structure

Let G = ( u , ,u2,...,u,,} be a group of signers with signing structure

PAR[u,,u,, ..., u,] . The scheme is stated as follows.
Key Generation Phase:
SA prepares the partial verification keys and public verification key as
PK, = f(ID,),
for ui E G , i =1,2 ,.., n , and V K , = f(C(ZD,,lD,
...., ID,,)).
Then, he performs the following operations:
Step 1 . Compute k, by the following equation, for i = 1,2, . . . , n .

k, =PK,-'.rmodw.
Step 2. Select a random number ko, where k, E 2
., .
Step 3. Randomly select s k l , E 2: for i =1,2, ...,n .
Step 4. Calculate the value of sk2; as follows, for i =1,2,...,n .
sk2, = k , - k , .sklimodo.
Step 5. Securely distribute sk, = ( s k l j , s k 2 , )
Step 6 . Calculate the value of v by

to u, E

G , for i = 1,2, ...,n .

Step 7. Compute w = ah mod N and deliver w, v to all u , E G .

Multisignature generation phase:
Suppose the signing group G with signing structure PAR[u,, u2,...,u n ]
wants to generate the multisignature MS for message m. Each u, E G
performs the following tasks without concerning other signer's signing order.
Step 1.Compute the partial multisignature S, as

54

ADVANCES IN NETWORK ANDDISTR. SYSTEMS SECURITY

Step 2. Send Si to all u j E G , for j # i .

Step 3. Verify S j sent from u, ,for j f i ,by testing if
(Sj)PK=
1 ph'")(mod N ) .

(10)

(If the test fails, then the signing process is stopped and u j is
reported as a malicious signer.)
Step 4. Calculate the multisignature MS after receiving and verifying all
S j 's, for j # i , u j E G, by the following equation.
n
1=1

MS = vh("). (n . S i )mod N .

Multisignature verification phase:

The multisignature MS, generated by the signing group G with signing
structure PAR[u,,u2, ...,u,] , for message m can be verified by testing if

(MS)vKG= ph('")(modN ) .

(12)

THEOREM 3. If all 4 E G , for i = 1,2,..., n, honestly sign the message m by

following PAR[u,,u2,...,u , ] , then the multisignature MS generated by G will
be successfully verified by Equation 12.
Proof:
Based on the fact that all valid partial multisignatures can be successfully
verified by Equation 10, we can rewrite Equation 10 with Equation 8 as

Then, from Equation 9, 11 and 13, we can obtain that

ID-based Structured Multisignature Schemes

(n

MS = v ~ (. ~ )
1=1

55

Si)(modN)

This implies a valid MS will be successfully verified by Equation 12. Q.E.D.

3.3

For mixed signing structure

Assume G = ( u , ,u 2 ,...,u, } is a group consists of mixed-ordered signers.

In any real case, the partial verification keys PK, 's and the public
verification key VK G can be easily computed by following the rules
described in section 2.2. The diagram representation of the signing structure
is employed here to facilitate the key generation and multisignature
generation phases.
A new notation used here is prev(x), where x is a node in the signing
structure diagram and prev(x) indicates the set of nodes that directly
connect and point to node x in the diagram.
Key generation phase:
By observing the diagram of the signing structure of G, SA first prepares
the partial verification keys and public verification key, and then generates
the secret key s k ; for each u, E G as follows.
Step 1. Compute k, by ki = PK,-'. r mod w ,for i = 1.2,...,n .
Step 2 . Select a random number k, , where k, E Z,* .
Step 3. Randomly select s k l , , such that skl, E Z L , for i = 1,2,...,n .
Step 4. Calculate the value of sk2, for each u, E G as follows.
If prev(ui)= {s }, then sk2, = k, - k, .skl, mod w;
Otherwise,

~ k 2=, k; -(C

kj).skl,modw

u j E prev(u,)

Step 5. Distribute ski = (skli,sk2,) to each u, E G via a secure channel.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

56

Step 6. Calculate a value v as follows.

If I prev(t) I =1, then v = 1 ;
Otherwise,

mod N to all ui , for prev(ui)= (sJ

Step 7. Compute and deliver w = ako
Step 8. Send v to all u, E G .
Multisignature generation phase:
Supposing the signing group G = {ul,u2,...,u n } with a mixed signing
structure wants to generate a multisignature MS for a message m. Then, with
the help of the corresponding diagram, each ui E G performs the following
operations to compute and distribute the partial multisignature.
Step 1 . Compute the partial multisignature S, as:
For ui with p r e v ( e ) = Is),

for u, with prev(u,)f ( s ) ,

s; =<nu j E prev(ui)s ', ) r k l ,

ah(m).sk2,

mod N ,

where S j is the partial multisignature generated by u j E prev(u,) .

Step2. Distribute Si to all U ~ GE , for p r e v ( u j ) ' > { u i }, and to all for
{ u i , u k c prev(t) .
Afterwards, the multisignature MS for message m can be calculated by any
uk E prev(t) with the following equation:
MS = # * ) .

<nuk E prev(t)Sk)modN .

Note that before u , signs, he should have verified the validity of each
received partial multisignature S j for u j E prev(u,) by testing if

(Sj)pK,
= ph'"'(rnodN).

ID-based Structured Multisignature Schemes

57

If the test fails, the signing process is stopped and the corresponding u, is
reported as a malicious signer.

Multisignature verijication phase:

The multisignature M S , generated by the signing group G for the message
m can be publicly verified as below.

(MS)vKG= /?h(m)(mod
N).

4.

SECURITY ANALYSIS

Possible attacks to the proposed schemes include the attempts to disclose

the signer's private key and to forge a structured multisignature. Although
the proposed schemes solve three different signing structures, they adopted
the same techniques for key generation, multisignature generation and
multisignature verification. Hereby we will show that the proposed schemes
are secure against these attacks by focusing our discussion on the serial
approach. Note that the security of the proposed schemes relies on the
difficulty of solving discrete logarithm modulo a large composite (DLMC)
[ 1 ] and factorising a large composite (FAC) [1, 17].

ATTACK 1 . An attacker attempts to reveal the secret key ski = ( s k l ; , ~ k 2 ~ )

o f a signer ui E G from all available public information.
Analysis :
From Equation 1 and 2, it is to see that the secret key ski of ui would be
disclosed by the attacker only when he knows either the values of w , r, and
all PK,-' 's; or the values of w and all k , 's. However, given all public
information a , , N and all PKG,i's for uiE G , computing w from N is a
problem of FAC intractability and deducing r from p is a difficulty of
solving the problem of DLMC. In addition, the attacker may try to deduce
the value of k, from the result in Lemma 1 , i.e. S, =akt'h(m)(modN)
.
However, he will obviously face the problem of the DLMC intractability. 0
ATTACK 2. An attacker attempts to reveal the private key ski = (skl ,,sk 2, )
of a signer u iE G from the partial multisignatures S,'s (for all uiE G ) of

a message m.
Analysis:

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

58

Given h(m), a , p , N and all S, 's for u,E G , directly computing

( s k l , , s k 2 , )from Equation 4 in the serial ordered multisignature scheme is
an intractability of the DLMC problem. On the other hand, solving
(skl,, s k 2 , ) from Equation 2 is also infeasible since w and all k , for P, E G
n
are secret parameters and known only to SA.
ATTACK 3. An attacker attempts to directly forge a valid multisignature for
some message m f o r the signing group G = (u,,u2,..., u, ) .
Analysis
Since the private key of each u, E G is securely kept, an attacker cannot
create any partial multisignature or multisignature for some message m via
Equation 4. Moreover, we know that a forged multisignature has to satisfy
Equations 5 to be valid. However, with public information N, p , VK, and
h(m), it's obviously that the attacker will face the FAC problem to directly
n
solve MS from Equations 5.

5.

CONCLUSIONS

In this paper, we have addressed a new approach to multisignature

schemes that applicable for various signing structures based on ID-based
public keys. In addition to enforce the requirement that all signers in the
signing group have to follow the predefined signing structure when generating a multisignature, our scheme has the merits that both the length of
multisignature and the computation effort for multisignature verification are
fixed and independent to the amount of signers. Due to the intractability of
the DLMC problem and the FAC problem, the proposed scheme is secure
against the deduction of the signer's secret key and forgery to the multisignature.

ACKNOWLEDGEMENT
The authors would like to thank the anonymous referees for their useful
comments on improving our paper. This work was partially supported by the
Ministry of Education, Taiwan, Program of Excellence Research 90-E-FA041-1.

ID-based Structured Multisignature Schemes

59

REFERENCES
L.M. Adleman and K.S. McCurley, Open problems in number-theoretic complexity,
II, Proc. First Algorithmic Number Theory Symposium, Springer-Verlag, 1994,
pp.291-322.
C. Boyd, Digital Multisignatures. IMA Conference on Cryptography and Coding,
Oxford University Press, 1989, pp. 241 -246.
C. Boyd, Multisignatures based on Zero Knowledge Schemes, Electronics Letters. 27
(22), October 199 1 , pp. 2002-2004.
M. Burmester. Y. Desmedt, H. Doi, M. Mambo, E. Okamoto, M. Tada and Y. Yoshifuji,
A structured EIGamal-type multisignature scheme, Proc. Workshop on Practice and
Theory in Public Key Cryptosystems, LCNS 1751, Springer-Verlag, 2000. pp. 466-483.
Y.S. Chang, T.C. Wu and S.C. Huang, EIGamal-like digital signature and
multisignature schemes using self-certified public keys, The Journal of Systems and
Software, 50 (2). 2000, pp. 99-105.
H. Doi, E. Okamoto and M. Mambo, Multisignature schemes for various group
structures, The 36-th Annual Allerton Conference on Communication, Control, and
Computing, 1999, pp. 7 13-722.
A. Fiat and A. Shamir, How to prove yourself: practical solution to identification and
signature problems. Advances in Cryptology - CRYPTO86, Springer-Verlag, 1987, pp.
I 86- 194.
T. Hardjono and Y. Zheng, A practical digital multisignature scheme based on discrete
logarithms. Advance in Cryptology-A USCRYPT92, Springer-Verlag, 1993, pp. 122132.
L. Ham and T. Kielser, New scheme for digital multisignatures, Electronics Letters,
25 (15), 1989, pp. 1002-1003.
K. ltakura and K. Nakamura, A public-key cryptosystem suitable for digital
multisignature, NEC Research and Development, Vol. 71, October 1983, pp. 1-8.
M. Mambo, K. Usuda and E. Okamoto, Proxy signatures: Delegation of the power to
sign messages, IEICE Tran. Fundamentals, E97-A (9), 1996, pp. 1338-1353.
M. Mambo, K. Usuda and E. Okamoto, Proxy signatures for delegating signing
operation, Proc. Conf on Computer and Comm. Security, ACM press 1996, pp. 48-57.
K. Ohta, S. Micali and L. Reyzin, Accountable-subgroup Multisignatures,
Manuscript, Massachusetts Institute of Technology, Cambridge, MA, USA, Nov. 2000.
K. Ohta and T. Okamoto, A digital multisignature scheme based on the Fiat-Shamir
scheme, Advance in Cryptology-ASIACRYPT91, Springer-Verlag, 1993, pp. 139-148.
K. Ohta and T. Okamoto, Multisignature schemes secure against active insider
attacks, IEICE Transactions on Fundamentals, E82-A ( 1 ), 1999, pp. 21-31.
T. Okamoto, A digital multisignature scheme using bijective public-key
cryptosystems, ACM Tran. Computer Systems, 6 (8), 1988, pp. 432-441.
R.L. Rivest, A. Shamir and L. Adleman, A method for obtaining digital signatures and
public key cryptosystems, Comm. of the ACM, 21 (2), 1978, pp. 120-126.
A. Shamir, Identity-based cryptosystems and signature schemes, Advances in
Cryprology - CRYPTO84, Springer-Verlag, 1985, pp. 47-53.
T.C. Wu, S.L. Chou and T.S. Wu, "Two ID-based multisignature protocols for
sequential and broadcasting architectures, Computer Comm. 19 (9-10). 1996, pp. 851856.

This page intentionally left blank.

PROBABILISTIC RELATIONS FOR THE

SOLITAIRE KEYSTREAM GENERATOR

Marina Pudovkina
Moscow Engineering PhysicsInstitute
maripa @online.ru

Abstract:

Stream ciphers are often used in applications where high speed and low delay
are a requirement. The Solitaire keystream generator was developed by B.
Schneier as a paper-and-pencil cipher. Solitaire gets its security from the
inherent randomness in a shuffled deck of cards. In this paper we present
probabilistic relations for the Solitaire keystream generator and describe their
properties.

Keywords:

Solitaire. Probabilistic relations.

1.

INTRODUCTION

Many keystream generators proposed in the literature consist of a number

of possibly clocked linear feedback shift registers (LFSEs) that are
combined by a function with or without memory. LFSR-based generators are
often hardware oriented and for a variety of them it is known how to achieve
desired cryptographical properties [6]. For software implementation, a few
keystream generators have been designed which are not based on shift
registers. Such generators with mixing next-state functions are RC4[7], IA,
IBAA, ISAAC [8], SCOP [9].
The Solitaire keystream generator was developed by B. Schneier [ 1] as a
paper-and-pencil cipher. Solitaire gets its security from the inherent
randomness in a shuffled deck of cards. By manipulating this deck, a
communicant can create a string of "random" letters that he then combines
with his message. Solitaire can be simulated on a computer, but it is designed

62

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

to be implemented by hand. It was designed to be secure even against the

most well-funded military adversaries with the biggest computers and the
smartest cryptanalysts. It's not fast, though it can take an evening to encrypt
or decrypt a reasonably long message.
Solitaire is an output-feedback mode stream cipher. The next-state
function F is a composition of four transformations F = F4 F3 F2 FI which
permute of elements of a deck.
In [2] is considered cycle structure of Solitaire. It is proved that Solitaire
is not reversible and described all irreversible states. In [3] are analyzed
properties of the key scheduling algorithm which derives the initial state
from a variable size key, and described weaknesses of this process. One of
these weaknesses is the existence of large classes of equivalent keys.
In this paper we present probabilistic relations for the Solitaire keystream
generator and stress some their properties. These relations describe the jokers
location in a deck at any time t. We show that the number of elements
between the jokers at time t depends on t and the initial number of elements
between the A joker and the B joker.
The paper is organized in the following way. In section 2 we describe the
Solitaire cipher. In section 3 we consider probabilistic relations for the nextstate function and in section 4 we give them for the key scheduling
algorithm. We conclude in section 5 .

2.

DESCRIPTION OF THE SOLITAIRE CIPHER

Solitaire is in fact a family of algorithms indexed by parameter n, which

is a positive integer. Let m be a cardinality of an alphabet of a plaintext, then
n=2m+2. The internal state of Solitaire at time t consists of a table
S,=(s,[O],. ...,s,[n-1]} of n values. S is a permutation of integers between zero
and n-1.
B. Schneier takes n=54, m=26.
The next-state function F is a composition of four transformations F = F4
F3 F:! F,, which correspond to items 14 of the description given in [l]. The
transformations F4, F3, F2, F, permute elements of a table S=(s[O],. ..,s[n-l]).
Let one joker A=n-2 and the other B=n-1.

The next-state function F

1. The transformation F1: Si + X =(x[O], ...,x[n-11). Let sU]=n-l(A). I f j z
n1 then move the A joker one element down: x[j]=s[j+l], xfi+l]=n-I,
and x[k]=s[k], k=O ...n-1, k#j, j + l . If j=n-I move it just below s[0]:
x[0]=s[0], x[ l]=A, x[2]=s[ 1 ] , . . ., x[k]=s[k-I],.. .,x[n-l]=s[n-2].

Probabilistic Relations for the Solitaire Keystream Generator

63

2. The transformation FZ:X+Y=(y[O],. ..,y[n-1]). Let x[j]=n-2(B). If j# n1 and n-2 then move the B joker two elements down: y[j]=x[j+1],
y[j+l]=x[j+2], y[j+2]=n-2 (B). If j=n-1, move the B joker just below
x[ 1]. If j=n-2, move it just below x[0].
3. The transformation F3: Y 4 Z=(z[0], ...,z[n-1]). Perform a triple cut.
That is, swap the elements above the first joker with the elements below
the second joker. "First" and "second" jokers refer to whatever joker is
nearest to, and furthest from, the top of the deck. Ignore the "A" and "B"
designations for this step. The jokers and the elements between them
don't move; the other elements move around them.
4. The transformation F4: Z 3 S,+,. Perform a count cut. Let z[n-1]=k.
Swap the elements z[0],. ..,z[k] with the elements z[k+l],.. ., z[n-2]. The
element z[n-1] does not swap. A deck with a joker as z[n-l] will remain
unchanged by this step.
The output function f
Let si+,[O]=q.
If S~+~[O]=A
or S~+~[O]=B
then we have not an output element.
B then the output element ki=si+1[q] (mod m).
If S~+~[O]+A,

Let M=m 1 m 2 ...m L be a plaintext and C=cIc2...cLbe a ciphertext.

Encryption:
c,=(mi+ki) (mod m).
Decryption:
ki=(ci-mi) (mod m).
Key Scheduling Algorithm
Key is an initial deck ordering. A passphrase is used to order the deck.
This method uses the Solitaire algorithm to create an initial deck ordering.
Both the sender and receiver share a passphrase. (For example, "SECRET
KEY.") Start with the deck in a fixed order; (0, 1, 2, . . ., n-3, A, B). Perform
the Solitaire operation, but instead of Step 4, do another count cut based on
the first character of the passphrase. In other words, do step 4 a second time,
using the character of the passphrase as the cut number instead of the last
card.
Repeat the four steps of the Solitaire algorithm once for each character of
the passphrase. That is, the second time through the Solitaire steps use the
second character of the passphrase, the third time through use the third
character, etc. Use the final two characters to set the positions of the jokers.

64

3.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

PROBABILISTIC RELATIONS FOR THE NEXTSTATE FUNCTION

In this section we describe probabilistic relations and their properties for

the next-state function. Let dAbe the number of elements before the A joker
and dBbe the number of elements before the B joker. We shall say that dA is
called the A joker distance and dHis called the B joker distance.

THEOREM 1
Let dA(j)be the A joker distance at time j and dBQ) be the B joker distance at
time j . Let Sj=(F)( So) and kj =s,[n-l]. If for any t<j: dH[t]#n-i, dB[t]#n-2,
d~[t]
dB[t] #dA[t]+l and dA[t] #dB[t]+l, then the A joker distance and
the B joker distance satisfy the following relations.
1. If j=2i then
i-1

dA(2i)=[dA(0)+z (k2j+,- k2j+2)- i] (mod n-1)

j=O

1-1

d&i)=[d~(O)+

(k*j+l- k*j+2)+i](mod n-1)

j=0

2. If j=2i+1 then
i-l

d ~ ( 2 i + l ) =[-d~(O
(k2,+2
)+
-c
k*j+,)-(1+k2i+1)(mod n)-i-2] (mod n-1)
j=0

i- l

d ~ ( 2 i + l ) =[ - d ~ ( O ) + x (k2,+2 - k2j+l)+i-l-( l+k 2i+l )(mod n)] (mod n-1).

j =O

(2)
Proof.
We conduct the proof by induction. Let dlA be the A joker distance in a
permutation Y and dlHbe the B joker distance in a permutation Y Let dAbe
the A joker distance in a permutation Z and dlB be the B joker distance in a
permutation Z. Consider j=2k+l.
Let us remark that

Probabilistic Relations for the Solitaire Keystream Generator

65

d~(2k+l)=d~(2k)+l,

Recall that F2 FI(S2k)=Y and F3(Y) =Z. Note that the distances in
permutations SZk, Y, and Z satisfy the following relations.
d ~ ( 2 k +1)=n-l-d~(2k+ 1 )=n-3-dB(2k) ,

de( 2k+ 1)=n- 1 -d~( 2k+ 1)=n-2-d~(2k).

The distances in permutations S2k,Y, Z and
relations.

&+I

satisfy the following

d&k+ 1)=[d~(2k+I)-( k2k+l+1) (mod n)] (mod n- I )=[n- 1-dB(2k+ 1 )(k2k+\+1) (mod n)] (mod n-l)= [-d~(2k+1)-( k2k+,+l)] (mod n-1)=
[-de(2k)-2-( k2k+i+l)l(mod n-1)d&k+ 1)=[dB(2k+ I)-( k2k+l
+ 1) (mod n)] (mod n- 1)= [n-1 -d,42k+ 1)(kZk+]+l)(mod n)](mod n-1)=[+fA(2k+1)-( k2k+l+l)(modn)] (mod n-1)
= [-d~(2k)-1-( k2k+l+1) (mod n)] (mod n-1).

It follows that
dA(2k+I)= -[dB(2k)+( I+k2k+l)(modn)+2] (mod n-I),
d ~ ( 2 k+I)= -[d~(2k)+(1+k2k+l)(modn)+l] (mod n-I).
Therefore,

de(2k +2)=-[d~(2k+l)+( l+k2k+2)(mOd n)+l] (mod n-l)= -[-d~(2k)(I+k2k+i)(mOd n)-2+( I +k2k+z)(mOd n)+l] (mod n-l)=-[-d&k)
+(k2k+2kzk+i)-l] (mod n-l)= [dB(2k) + (k2k+l-kzk+2)+1] (mod n-1).
We apply an induction over k and obtain.

66

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

dA(2k+2)=[ d ~ ( 2 k +
) (k2k+l-k2k+2)- I 1 (mod n- 1)= [dA(2k-2)+(k2k-l+k2k+lk

k2k-k2k+2)-2](mod n-I)=. ...=[dA(0)+x (k2j+I- k2j+2)-k+l](mod n-1).

j =O

d ~ ( 2 k+2)= [d~(2k)+ (k2k+l-k2k+2)+I ](mod n-l)=[de(2k-2)+(k2k-l+kzk+1-

x
k

k2k-k2k+2)+2](mod n-1)=....=[dB(0)+

(k2j+l- k2j+2)+k+l](modn-1).

j=O

d~(zk+l)=-[d,(2k)+( 1+k2k+l)(mod n)+2]

c
c

(mod

n-l)=

-[dB(O)+

k-1

(k2j+l-k2j+z)+k+2+(1+ k2k+l) (mod n)] (mod n-l)= -[dB(O)+

j=O

k-1

(k2j+2-k2j+I)-k-2-( l+k2k+l)(mOdn)] (mod n-1).

j=0

d~(zk+l)=-[d~(2k)+(
l+k2k+l)(mod

n)+l]

(mod

n-l)=

-[dB(0)+

k-1

(k2j+l-k2j++k+I+( l+k2k+l)

(mod

n)]

(mod

n-l)=[-dB(0)+

j =O

(k2j+2-k2j+l)+k-l-(l+k 2k +l)(mod n)] (mod n-1).

j=0

This completes the proof.

Proof.
Really, P( d~=n-l,d~=n-2,d~=n-I,d~=
dA +l,dA=dB +1}5341-l)!/n! +2*
(n-2)!/n!=3/n+2/(n- 1) n W n .
Let Prob(j) be a probability that the probabilistic relations at time j are
true.

Probabilistic Relations for the Solitaire Keystream Generator

67

REMARK 2

Prob(j)L (1-4/n)'
Proof.
Note that the probabilistic relations at time j are true if for any t<j:
d~[f]#n-l, d~[t]#n-2, d ~ [ t #n-l,
]
d ~ [ t ]#dA[t]+l and d ~ [ f ]#dB[t]+l- Using
remark 1 , we have P 2 ( 1 4 n y .
Let us consider some properties that obtained from the presented
probabilistic relations. By distAB(t)denote the number of elements between
the A joker and the B joker at time t. Proposition 1 and proposition 2 show
that diStAB(t) depends on t and dA(0)- dB(0).
PROPOSITION 1
Let x=(dA(i)- dB(i))(modn-1) then
diStAB(i)E { X-1, n-2-x}
Proof.
Let US remark that diStAB(i)=l dA(i)- dB(i)l-l.
Consider two possible cases.
a ) I f dA(i)>dB(i) then diStAB(i)= ( dA(i)- dB(i))(mOd n-l)-l=x-l
b) If dA(i)<dB(i) then X = ( dA(i)- dB(i))(mod n-1)= n-1+ dA(i)- dB(i).
Therefore, distAB(i)=dB(i)- dA(i)-l= n-2-X
The proposition is proved.
PROPOSITION 2
Let y=( dA(0)- dB(0)-k) (mod n-1) then
diStAB(k)E { y-1, n-2-y}

Proof.
Let x=( dA(i)- dB(i))(modn-1). By proposition I and (1), (2) we obtain.
a) If k=2i+1 then

b) If k=2i then

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

68

i-l

x=( dA(k)- dB(k))(mod n-1)=(

dA(O)+

(kZj+l- k2j+2)-i+b(O)+
j=O

(k2j+l - kZj+2)+i)(mod n-1)= ( dA(0)- dB(0)- 2i)(mod n-1)= ( dA(0)-

j=O

dB(0)-k) (mod n-1).

Therefore, x=( dA(k)-dB(k))(modn-1)=( dA(0)+iB(O)-k)(mod n-1) = y.
We have diStAg(k)E { y-1, n-2-y}
The proposition is proved.

REMARK 3
If we take x=(dB(k)+iA(k))(mod n-1) and y=(dB(O)dA(O)+k) (mod n-1)
then proposition 1 and proposition 2 remain true.

PROPOSITION 3
If d~(o)-d~(O)=2
then
diStAB(k)E {k +1, n-k-4}.
This proposition can be proved by direct calculations.
In propositions 4-7 we describe some properties which allow finding the
jokers location or elements of the permutation with high probabilities. Let
So, S, be two initial states. By dA, dB denote the A joker distance and the
B joker distance for S and by dA,dB denote distances for S".

P ROPOSITION 4
1 . If dA(O)= dA(O), dB(O)= dg(0) and dA(k)= dA(k) then dB(k)= dB(k).
2. If dA(O)= dA(O), dB(O)= dg(0) and dB(k)= dB(k) then dA(k)=dA(k).
Proof.
Let us prove item 1.
a) Let k=2i.
B y dA(k)= dA(k) and ( 1 ) we get
.i-l .

1-1 .
.

(kzj+l- k2j+2)-i]=[dA(0)+

[dlA(0)+

(k2j+l- kzj+2)-i] (mod n-1).

j =O

j=O

Therefore,

2
j=O

i-1

(k2j+l- k2j+2)=
j=O

(k2j+I-kz,+2)(mod n-1).

Probabilistic Relations for the Solitaire Keystream Generator

69

Note that
i-1

(k2j+l- k12j+2)+i](mod n-1) and d&i)= [dB(O)+

d&i)= [dB(O)+

j=O

(k2,+] - k2,+2)+i] (mod n-1).

j=O

Therefore, dB(k)= dB(k).

b) Let k=2i+1.

From (2) and dA(k)=dA(k) it follows that

i-1

[-dtB(0)+

(k2,+2- kzj+])-( I+k2i+l)(modn)-i-2] =

j=O

i-1

(k2,+2- k2j+l)+i-l-( 1 +k2i+l)(mod n)] (mod n-1).

[-dA(O)+
j=O

1-1

dB(k)= [-d~(o)+

(k2j+2-

k~j+~)+i-l-(

1+kt2i+1)(mod n)](mod n-1),

j=O
i-1

(k2j+2-k2j+l)+i-l-( l+k2i+l)(mod n)] (mod n-1)

dB(k)= [-d~(0)+
j=O

we have

( l+k2i+l)(mod n). Therefore, dA(k)=dA(k).

Item 2 is proved similarly.

The proposition is proved.

PROPOSITION 5
If we know either (dA(k), dB(k+l)) or (dB(k), dA(k+l)) then we can
determine the value of kk+l.

70

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

This proposition can be proved by direct calculations.

PROPOSITION 6
1. If we know either (dA(0), dA(2i) } or [de(0), d&i)
determine

} then we can

2. If we know either (dB(O), dA(2i+l) }or { dA(0), d ~ ( 2 i + l )} then we can

determine

j=O

Proof.
Let us prove item 1.
Note that (1) we can rewrite as
i-1

dA(2i)=

(kzj+l-k2j+2)+dA(0)-i(mod n-1)=I+dA(0)-i (mod n-1),

j =O

i- l

d ~ ( 2 i ) ' C (k2,+,- k2,+z)+dB(0)+i (mod n-l)=I+dB(0)+i (mod n-1).

j=O

This yields that

i-l

I=

(kzj+l-k2,+2)= d ~ ( 2 i ) A ~ ( O ) (mod
+ i n-1),
j=O

Item 2 is proved similarly.

The proposition is proved.

Probabilistic Relationsfor the Solitaire Keystream Generator

71

PROPOSITION 7
1 . If dA(O)=dA(k) then dB(k)=( dB(O)+k) (mod n-1).
2. If dB(O)=dB(k) then dA(k)=(dA(0)- k) (mod n-1).
This proposition can be proved by direct calculations.
Let c be an element of the permutation S and CE (A, B ) and by dCQ)
denote the number of elements before c at time j. In proposition 8 we find
dC(1).

PROPOSITION 8
1 . I f either dA(O)< dc(0)< dB(0) or dB(O)< dc(0)< dA(0) then dc( 1)= [-d~(o)dB(O)-kl+ dc(0)-4] (mod n-1).
2. If either dc(0)c dA(0)<dB(O)or dB(0)< dA(0)< dc(0) then dc(l)= [dc(0)dA(0)-kl-2] (mod n-I).
3. If either d,(0)<dB(O)<dA(O) or dA(0)cdB(O)<dc(0) then dc(1)= [dc(0)dB(0)-kl-2] (mod n-1).
The proof is straightforward.

4.

PROBABILISTIC RELATIONS FOR THE KEY

SCHEDULING ALGORITHM

In this section we present probabilistic relations for the key scheduling

algorithm.

THEOREM 2
Let K=kl,..., kL be a passphrase, where L is its length. Let So=(O, 1, 2,
. . .,n-3, A, B) and Sj=(Fj( So). Let dAQ)be the A joker distance at time j and
dB(i) be the B joker distance at time j . If for any t<j: ds[t]#n-l, dB[t]+n-2,
dA[t] #n-1, d ~ [ t +
] d ~ [ t ] + land dA[t]#dB[t]+], then the A joker distance and
the B joker distance satisfy the following relations.
1 . If j=2i then

2. If j=2i+1 then

72

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

1-1

dA(2i+l)=[ado)+

(k2,+2 -

k2j+l)-kzi+i-i-3] (mod n),

j=O

i-1

dB(2i+l)= [-dA(0)+

(k2,+2- k2j+l)+i-2-k2,+1] (mod n).

j=O

This theorem is proved as theorem 1 .

We stress that the propositions which are proved in the previous section
remain true for the key scheduling algorithm but the operation mod (n-1) is
changed by (mod n).

5.

CONCLUSION

In this paper we presented probabilistic relations for the jokers in Solitaire

key stream generator and described some their properties. It analyzes the
probability distribution of distance between the two jokers in a deck at
different time periods. We found that the number of elements between jokers
at time t depends on t and the initial number of elements between the A joker
and the B joker. Presented results with bit changes are applied to the keyscheduling algorithm of Solitaire.
We hope that results described in [2], [3] and this paper allow mounting
an attack on this cipher which is more effective than exhaustive search, and
this will be the object of another paper.

REFERENCES
Schneier B., "The Solitaire Encryption Algorithm, http://www.counterpane.com/
solitaire. html.
Pudovkina M, Varfolomeev A.A. A Cycle Structure of the Solitaire Keystream
Generator. 3nd International Workshop on Computer Science and Information
Technologies CSIT'2001, YFA, 2001.
Pudovkina M., Weakness in the Key Scheduling Algorithm of the Solitaire Keystream
Generator, SIBCOM-2001, TOMSK. (the paper is being published)
Varfolomeev A.A., Zhukov A.E., Pudovkina M., Analysis of Stream Ciphers ,
Moscow, MEPhI, 2000.
Crowley
P.
Problems
with
Bruce
Schneier
Solitaire.,
http://www.hedonism.demon.co.uk/ paul/solitaire/
Rueppel R.A. Analysis and Design of Stream Ciphers, Springer-Verlag,
Communications and Control Engineering Series, 1986.

Probabilistic Relations for the Solitaire Keystream Generator

73

[7]
[8]

Rivest R.L.,"TheRC4 encryption algorithm, RSA Data Security, Inc., Mar. 1992
R.J. Jenkins, ISAAC, Fast Software Encryption -Cambridge 1996, vol. 1039, D.
Gollmann ed.. Springer-Verlag.

[9]

Simeon V. Maltchev and Peter T. Antonov. "The SCOP Stream Cipher,

ftp://ftp.funet.fi/pub/crypt/cryptography/symmetric/scop/scop..tar.gz, Dec. 1997.

This page intentionally left blank.

HAZARD ANALYSIS FOR SECURITY

PROTOCOL REQUIREMENTS
Nathalie Foster*
Jeremy Jacob
Department of Computer Science
University of York
UK
{nathalie.foster, jeremy.jacob}@cs.york.ac.uk

Abstract

This paper describes a process for the generation and analysis of security
protocol requirements. It addresses some of the problems resulting from
the inadequacies of present development methods. It is based on a
hazard analysis technique which has been developed for safety critical
systems engineering. This provides a structured m e t h o d of analysis of
the requirements whilst avoiding the problems of being too restrictive.

Keywords: security protocols, software engineering, requirements gathering and

analysis, hazard analysis.

1.
1.1.

Introduction
Security Protocol Development

In comparison to the process of general software development, security

protocol development is relatively unstructured. It is therefore hardly
surprising that protocols are still being published that are later found to
be vulnerable to attacks. Sound engineering practices need to be applied
to protocol development and we need to consider the whole development
lifecycle for protocols.
General models of software development have been proposed, the most
famous of these are the Waterfall model by Royce[12] and the Spiral
model by Boehm[5]. Many variations on these models have been suggested but common to all the models are a number of distinct activities:
of these requirements gathering and analysis are the first [9, 13]. Al*Supported by an Engineering and Physical Sciences Research Council (EPSRC) studentship

76

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

though not perfect, these models go someway towards the development

of higher quality software.
Research in protocol development has focused on the use of formal
methods and logics in the design and verification of protocols at a late
stage in the development. Little work has been carried out into the relevance to protocol development of other activities found in the software
engineering process, such as requirements engineering.
We believe that many attacks and flaws could be guarded against
through the proper gathering and analysis of the protocol requirements
before the actual design. This would reduce the incidence of security
failures due to the opportunistic exploitation of elementary design flaws
or implementation and management design errors [ 1, 2].
In this paper, we address the issue of requirements gathering and
analysis in protocol development. Our approach is based on a hazard
analysis technique and is described in Section 1.2 and 2. It is used to analyse the goals of the protocol with reference to the security requirements
of the protocol, in order to generate the low level protocol requirements.
Through the use of this technique, we are able to begin designing
a protocol with a more thorough understanding about what it should
do. We also have a higher level of confidence about the security of
the protocol designed based on the requirements, before we carry out
any verification of the protocol. This is necessary for the verification of
the protocols: we have requirements which we can verify the protocol
against, rather than having to guess what the requiremerits are before
we can start verification.

1.2.
HAZOP.

Hazard Analysis

We propose the use of a hazard analysis technique which

has its foundations in a method called Hazard and Operability Study
(HAZOP) [6, 7]. HAZOP was developed by Imperial Chemical Industries
(ICI) for the identification of hazards in process plant designs within the
chemical industries, where the analysis is carried out on the pipework
and instrumentation design of the plant. It has since been applied in the
food-processing, pharmaceutical, nuclear, oil and gas industries and has
also been adapted for use in the development of safety critical systems
[11].
In a HAZOP study, a team identifies the entities and attributes of a
design. A standard list of guide words is used to suggest deviations to
these attributes. The deviations are analysed to determine their possible
causes and effects and to consider what actions need to be taken to avoid
or minimise the effects of the deviations. The results of the analysis are

Hazard Analysis for Security Protocol Requirements

GUIDE
WORD
Omission

77

GENERIC MEANING

The service is never delivered, i.e. there is no communication. These are classified as either total or partial.
Commission A service is delivered when not required, i.e. there is
an unexpected communication. These are classified as
either spurious or repetition.
Early
The service (communication) occurs earlier than intended. This may be absolute (i.e. early compared to
a real-time deadline) or relative (early with respect to
other events or communications in the system).
Late
The service (communication) occurs later than intended. As with early, this may be absolute or relative.
Value
The information (data) delivered has the wrong value.
Table 1.

The SHARD guide words [10]

recorded in a table detailing the deviations, causes, effects, detection and

protection, and the justification and recommendations. The analysis
documentation is used to improve the safety of the system under study
and may also be used in further investigation of the safety of the system.

SHARD.

Software Hazard Analysis and Resolution in Design

(SHARD) [10],is a projective computer system safety analysis technique based on HAZOP. It is used to analyse designs and to obtain
system safety related requirements for the detailed development of those
designs. The guide words in SHARD are based on the communication
of pieces of information, with specific values, at particular points in time
(Table 1).
The analysis process in SHARD is even more structured than in
HAZOP, with extra steps to be carried out in the analysis. The SHARD
process is shown in the flow diagram in Figure 1.
The analysis is recorded in a table with at least the following column
headings: Guide word; Deviation; Possible Causes; Effects; Detection
and Protect ion; Just ificat i o n / Design Recommend at ions.
The structured SHARD process and the more appropriate guide words
for a system involving information flows lends itself to the analysis of
security protocol requirements with some modifications. The application
of SHARD to protocol requirements gathering and analysis is described

78

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 1.

SHARD analysis [10]

in Section 2. An example use of the analysis of the requirements for an

electronic commerce protocol is given in Section 3

2.
2.1.

Requirement Analysis for Security Protocol

Introduction

The aim of a requirements analysis process for security protocols is to

analyse the high level requirements of the protocol to obtain the low level
functional requirements of the protocol. These low level requirements
can then be used in the design phase of the protocol development. This

Hazard Analysis for Security Protocol Requirements

79

ensures that the security requirements are carefully considered at the

early stages of development and the features which are built into the
protocol are justified.
The analysis is split into two levels, based on the distinction between
high and low level protocol requirements:

High Level Requirements These state what must be achieved

by the end of the protocol run. The requirements are in terms
of the differences in the knowledge state of the principals between
the start and end of the protocol session, they indicate what the
principals should and should not know. They are equivalent to the
pre- and post-conditions of the protocol.
The high level requirements can be subdivided into functional and
non-functional requirements. The functional requirements indicate the functionality of the system under development; for example, By the end of the protocol, Principal A should have received an order from Principal B " . At this high level we are not
interested in how this is achieved, nor what the order looks like.
Non-functional requirements are more difficult to analyse, these
requirements include safety, security and reliability requirements.
In protocol requirements analysis, we are concerned with ensuring
that protocols maintain a number of security properties, which are
determined by the purpose of the protocol. We have designed this
analysis method with reference to the following security properties:
confidentiality; authenticity; integrity; non-repudiation; availability; timeliness; non-replicability. An example of a non-functional
requirement is The order must be kept confidential between principals A and B . ".

Low Level Requirements These are the low level functional

requirements of the protocol and are derived from the high level
functional and non-functional protocol requirements. They state
details such as what each protocol message will contain, how it
will be constructed, any interactions between messages, such as if
a particular message component is dependent on another message,
and what checks will need to be carried out on the messages. An
example of a low level requirement is: The message should contain
a component (such as a timestamp) to avoid replay attacks and to
ensure timeliness of messages.
The analysis explores how an implementation may fail to meet its
requirements, including how the external environment can affect the
protocol. This may prompt further requirements of the protocol to de-

80

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 2.

Hazard Analysis for Security Protocols Requirements

tect when such situations arise and also to protect or mitigate against
the violations of the requirements.
Our method takes into account the differing views of the stakeholders
in the protocol through the use of a team in the analysis. The team
should consist of representatives of each of the stakeholders in the protocol and, ideally, someone who is familiar with attacks and flaws which
are common in protocols, as well as the different verification techniques
which can be used on the protocols.

2.2.

Hazard Analysis for Security Protocols

The analysis process is based on the SHARD process using guide

words to prompt deviations to the requirements and identifying the
causes, effects, detection and mitigation mechanisms associated with
these deviations.
Our analysis is carried out at the both the high and low levels of
functional requirements. The analysis of the requirements will prompt
further high level and low level functional requirements which will be

subject to further analysis. Thus the analysis is an iterative process. It

is outlined in Figure 2.
The guide words in the analysis process have been adapted to relate
to message transfers, contents and checks on messages, to prompt deviations which make the requirements vulnerable to attacks which violate
the required security properties. Once these vulnerabilities have been
identified, measures can be taken, through the introduction of further

Hazard Analysis for Security Protocol Requirements

81

requirements to avoid the incorporation of the vulnerabilities into the

design and implementation.
Table 2 contains the guide words and interpretations used to generate
the deviations to the protocol requirements, these were influenced by the
SHARD guide words in Table 1. In this table we also identify the security
property violations which could result from the deviations suggested by
the guide words. The guide words we have selected may not be the only
guide words which could be used for security protocol analysis. Further
guide words can be added to the process to reflect different properties
required of different types of protocol.
Some of the steps in the analysis process of Figure 2 are described in
more detail below:

Identification of high level functional requirements. The high

level functional requirements are elicited from the informal scenario description which details the situation for which we wish to design a protocol. By identifying the principals, their actions and the objects on
which they act, we can extract more structured requirements which describe the scenario which contain the following: initiating principal, responding principal, action, object.
Identification of Causes. This is based on the primary - secondary
- command rule for identifying causes in SHARD. We interpret this as:
Primary (P) causes are due to the failure of the principal who
carries out the service. For instance, the principal may not have
sent out the message, or may have sent out an incorrect message.
Secondary (S) causes are due to the failure of the medium over
which communication is made or a n action or event is carried out,
such as the network. Cases where an intermediary party, such as
an intruder causes a deviation are also classed as secondary causes.
Command (C) causes are due to the failure of the command
which prompts the action to be carried out. Earlier messages in a
protocol session act as a command, or prompt, to the principal to
send out the next message. Therefore, if an incorrect message is
received then a response dependent on that message may also be
incorrect.
Identification of Effects. The immediate effects of the deviations
are noted. Any possible actions (A) which can be carried out by the
principals as a result of the deviation are identified and the consequences
(C) of these actions are identified. The actions and their consequences

82

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

SECURITY
PROPERTIES
VIOLATED
Availability
The event does not take place.
Omission
Commissior The event which takes place is not Spurious:
as expected. The different types of Authenticity,
Non-repudiat ion
commission:
Repetition:
spurious: a one off event.
Authenticity
repetition: a repeated event.
Non-repudiation
Non-redication.
The data obtained in the event has Integrity,
Value
the wrong value arid this can be de- Authenticity,
Non-repudiation
tected. This could be:
total: the data dclivered in the
event is totally corrupted.
extra: an event occurs as expected but with some unexpectedextra data/behaviour.
partial: parts of the expected
event are omitted.
The data in this event has b e e ndi- Confidentiality.
Disclosure
vulged to an unauthorised party.
The event occurs earlier than inten- Authenticity
Early
Timeliness.
ded. Early can be interpreted as:
absolute: early cornpared to a
real-time deadline.
relative: early with respect to
other events or communicationsin
the svstem.
Late
The event occurs later than inten- Authenticity ,
Timeliness,
ded. Late can be interpreted as:
absolute: late compared to a real- Availability.
time deadline.
relative: late with respect to
other events or communications in
the svstem.
GUIDE
WORD

GENERIC MEANING

Table 2. The protocol analysis guide words

Hazard Analysis for Security Protocol Requirements

83

must be considered because it is often a chain of events following from

a deviation which results in an insecurity in the protocol. As noted in
the SHARD analysis, effects may contribute to or be the causes of other
deviations, therefore any dependencies should be recorded.

Recommendations. In order to protect the security of the protocol,

recommendations to address the deviations are made. The recommendations are of three types:

Prevention (P) Measures to prevent a potential violation of a

security property, are incorporated into the design.
Detection (D) Mechanisms to detect when, how and who violated
the security of the protocol.

rn

Reaction (R) If we can detect when a violation has taken place

then we can recover from the security violation by correction or
mitigation mechanisms.

The recommendations depend upon the security properties which have

been breached. In some circumstances we can react to the security
violation and carry out an action to return the protocol to a secure
state. However, in some cases it is impossible to recover from a security
violation, such as when a confidentiality breach occurs. In such
we must attempt t o find protection mechanisms t o prevent such security
breaches. Similarly there are also situations in which detection of a
security violation is very difficult. The choice of recommendations must
be carefully considered to deal with such cases.
The recommendations can be implemented using a variety of methods.
Software or hardware controls could be used to ensure that the protocol
security is maintained; for example, encryption can be used to maintain
integrity and confidentiality. Policies and physical controls can be used
to govern the application of the protocols and information in a wider
context .
Further high and low level requirements are elicited from the recommendations. These are then added to the list of requirements and are,
in turn, analysed. If there are multiple recommendations to address the
same problem, then design decisions about which recommendations to
use will need to be taken and these should be documented.

Analysis Documentation.
The analysis is documented in a table
such as that in SHARD and HAZOP. The documentation table may also
contain a column for recording comments arising in the course of the
discussion, this is useful for recording other issues and cross references

84

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

relating to other parts of the analysis or design phase. Example headers

in the documentation table are: Guide word; Deviation; Causes; Effects;
Recommendations; Comments.

3.

Example Application

In this section, we provide a partial example of the use of the requirements analysis process for security protocols.

Scenario

3.1.

A vendor wishes to sell goods to its customers over the internet using
an electronic commerce protocol. It is envisaged that the customer will
send to the vendor an order for the goods and also payment details. The
vendor will then be able to obtain payment through the customers credit
card company. In return the customer will obtain the goods ordered.

Example Analysis

3.2.

Identification of High Level Functional Requirements.

We
extracted the following high level requirements from the scenario above
by identifying the principals, actions and objects and their interactions:
1 Customer sends an order for goods to the vendor.

2 Customer sends payment details to the vendor.

3 Vendor submits the payment details to the payment authority.

4 Vendor obtains payment for the goods from the payment authority.
5 Vendor distributes the goods to customer.

High Level Analysis. Table 3 contains an analysis of the high level

requirements "Customer sends and order for goods to the vendor" using
the hazard analysis process. In a full analysis, a similar table is produced
for each of the requirements.
Omission:
Cause:

Effect:
~~

No order is made.
(P) Customer doesnt send an order.
(S) Order lost by network/intruder actions.
(A) Customer waits indefinitely for response
from vendor.
( C ) Vendor looses an order if not detected.
~~

~~

Table 3. Analysis of Customer sends order for goods to vendor

Hazard Analysis for Security Protocol Requirements

Recommendations:

Comments:

Commission
(Spurious):
Cause:

85

(D) Timeout on waiting for response to order so

customer does not wait indefinitely.
(R) Recovery session to resend order.
(D) A pre-protocol exchange enables vendor to
detect if an order is missing.
(P) Use a reliable network.
Network reliability is out of the scope of this protocol since we have no control over the reliability
of the internet.
Prevention of intruder attacks is impossible, protection should make attacks infeasible.
An order takes place unexpectedly.

(P) Customer accidentally sends the order.

(S) An intruder fakes an order.
(S) Network fault results in spurious order.
(A) Vendor treats order as valid and waits indefEffect:
initely for a payment message which will not take
place (if payment is before delivery).
Recommendations: (D) Tinieouts on waiting for payment message so
vendor doesnt wait indefinitely.
(D) Order authentication.
(D) Customer feedback to check ordeir is correct.
(P) A pre-protocol exchangeso valid orders received by vendor are riot unexpected.
As for Omission
C o m me n t s
An order is repeated.
Commission
(Repetition):
(P) Customer repeats an order intentionally.
Cause:
(P) Customer accidentally sends a repeat order.
(S) An intruder replays the order maliciously.
(S) Network fault causes message to be resent.
(A) Vendor treats order as valid and customer
Effect:
receives unwanted goods.
(A) Vendor rejects order and customer waits indefinitely for response from vendor.
Recommendations: (D) Use of a fresh element (nonce or timestamp)
to detect replay of an order. This allows valid
repeat orders to take place.
(D) Customer feedback to check order is correct.
Table 3. Analysis of Customer sends order for goods to vendor"

86

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Value (Total):
Cause:
Effect:

Recommendations:

Comment:
Value (Extra):
Cause:

Effect:

Recommendations:

Comment:
Value (Partial):
Cause:
Effect:

Recommendations:

Order is totally corrupted

(S) Corrupted on the network or by intruder.
(A,C) Vendor rejects message as it is not identifiable as an order and customer waits indefinitely
for response from vendor.
(A) Message interpreted as an order, but not that
intended by the customer.
(D) Check integrity of order.
(D) Customer feedback to check order is correct.
(P) Avoid indefinite waiting by timing out waiting for a response to order.
(R) Recovery session to resend the order
(P) Use a reliable network.
As for Omission.
Order is valid but there is some extra information
with it.
(P) Extra information added by customer.
(S) Result of corruption on the network/by an
intruder.
(A) Order interpreted by vendor as an order with
unwanted extra items included.
(A) Order rejected by vendor and customer waits
indefinitely for response from vendor.
(D) Check integrity of order.
(D) Customer feedback to check order is correct.
(P) Avoid indefinite waiting by timing out waiting for a response to order.
(R) Recovery session to resend the order.
(P) Use a reliable network.
As for Omission.
Only part of order message is received.
(P) Customer missed off parts of order message.
(S) Components of order message are lost on network/by an intruder.
(A) Order accepted but parts of customers order
are missing.
(A) Order rejected and customer waits indefinitely for response from vendor.
(D) Check integrity of order.
(D) Customer feedback to check order is correct.

Table 3. Analysis of Customer sends order for goods to vendor

Hazard Analysis for Security Protocol Requirements

87

(R) Recovery session to resend the order.

(P) Use a reliable network.
Comment:
As for Omission.
Disclosure:
Order is disclosed.
Cause:
(C) Order is not protected and can be read by
eavesdropper on network.
(C) Customers privacy is violated as order is
Effect:
public knowledge.
(C) Vendors order details are available to everyone, including their competitors.
Recommendations: (P) Confidentiality protection of the order.
Early:
Order is received early.
As for Commission (spurious).
Order is received late.
Late:
(S) Delay on network or by an intruder.
Cause:
(A) Customer waits indefinitely for vendors reEffect:
sponse to order.
Recommendations: (D) Inclusion of a fresh component to enable the
vendor to determine if a message is late.
(D) Customer times-out waiting for messages for
vendors response to avoid indefinite waiting.
(R) Recovery session to resend order message.
(P) Use a reliable network.
As for Omission
Comment
Table 3. Analysis of "Customer sends order for goods to vendor"

Extraction of Further Requirements from High Level Analysis.

The following requirements were extracted from the analysis of the
requirement Customer sends an order for goods to vendor. In a full
analysis, these are analysed in later iterations of the Hazard Analysis
process.

High level requirements:

1 A recovery session should be available in case that order needs
to be resent, if it is detected that order is incorrect or has not
been received by the vendor.

2 Pre-protocol exchange to ensure that vendor is alive and accepting orders and also so that vendor is able to anticipate
receipt of orders.

88

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

3 Provide feedback (a confirmation of order) so customer can

check that order is correct.

Low level requirements:

1 Time-outs on waiting for orders and responses to orders to
avoid principals waiting indefinitely.

2 Authentication of order messages.

3 A fresh element in order provides uniqueness of order, giving assurance that it has been created recently and allowing
orders to be repeated.
4 Integrity checking and correction of order.
5 Confidentiality protection of order to protect customers pri-

vacy.

6 Incorporation of a time component to detect if order is late.

Low Level Requirements.
Table 4 contains an example of the
low level analysis stage in the Hazard Analysis for Security Protocols
process. This table shows the analysis of the low level requirement "A
fresh element in order provides uniqueness of the order message and
allows orders to be repeated. From this analysis we obtain further
requirements for the protocol.
No fresh element is included in the order message.
(P) Not included by customer.
(S) Unavailability of fresh element generator.
Effect:
(C) Vendor cannot check if order was created recently.
(A) Intruder is able to replay order message.
Recommendations: (D) Vendor checks for fresh element in order and
reject order if it contains no fresh element.
(P) Use of reliable fresh element generator.
Commission
Fresh element is unexpectedly in order message.
(spurious):
Comment
Not applicable since message is expected to contain a fresh element.
Commission
A fresh element is reused in order message.
R e p e t i tion) :
Cause:
(P) Reused by principal.
(S) Element replayed by intruder/network.

Omission:
Cause:

Table 4. Analysis of "A fresh element is included in order message"

Hazard Analysis for Security Protocol Requirements

Effect:

89

(A) Rejection of message by vendor.

(C) Customer does not receive goods.

Recommendations: (D) Check fresh element and reject if repeated.

(R) Recovery session to deal with invalid fresh
elements.
A fresh element of unexpected format is in order
Value (Total):
message.
Expected fresh element plus extra information is
Value (Extra):
in order message.
Value (Partial): Partial fresh element is in order message.
Cause:
(P) Included by customer.
(S) Element in format provided by generator.
(A) Order message is rejected by vendor.
Effect:
(C) Customer does not receive goods.
Recommendations: (D) Check fresh element and reject if invalid.
(R) Recovery session for cases where fresh element is invalid.
Fresh element is disclosed.
Disclosure
Cause:
(P) Not protected by principal.
(S) Disclosed on network/by intruder.
None. Public knowledge should reveal nothing.
Effect:
Fresh element in order message is early.
Early
Cause:
(S) Generator dispenses fresh items too early.
(S) Other messages have not yet been received.
Effect:
(C) It is known that order message has been created recently and so is valid.
Recommendations: (P) The fresh element generator for customer and
vendor should be periodically synchronised.
Fresh element in order message is late.
Late
(P) Principal sends order message late.
Cause:
(S) Fresh element generator generates late.
(S) Message delayed by intruder/network.
(A) Order is rejected because it is too late.
Effect:
Recommendations: (D) Check that messages are timely/fresh and
reject if late.
(P) Periodic synchronisation of customer and
vendor fresh element generators.
(R) Recovery session in case of late messages.
Table 4 . Analysis of A fresh element is included in order message"

90

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

E x t r a c t i o n of Further R e q u i r e m e n t s f r o m Low Level Analysis.

The analysis of the low level requirement A fresh element is included in
the order message identified the following requirements of the protocol:
H i g h level r e q u i r e m e n t s

1 Recovery session to deal with messages with invalid or late

fresh elements.

Low level r e q u i r e m e n t s
1 Periodic synchronisation of fresh element generator.

2 Use of reliable fresh element generator.

3 Checks to ensure fresh elements are of valid format/timely
and reject if not.

4 Checks for the fresh element in order message and rejection

if no fresh element.

3.3.

What has been gained from this analysis?

From this fragment of an example of a Hazard Analysis for Security

Protocol requirements, we can gain insight into the intuitive steps taken
by the designer. We can identify items which need to be kept confidential, checked for authenticity, integrity and freshness, recovery sessions
and feedback to the principals which is required. Using this analysis process, we can trace the generation of requirements and justify the features
which are built into the protocol.
In a full analysis, each of the recommendations would be justified
in more detail and labelled to make it easier to trace and refer to the
protocol requirements during the later development phases.

4.

Conclusions

In this paper, we have described a process for the gathering and analysis of the requirements of security protocols before the actual design
of the protocol. This is the traditional starting point in the software
engineering life cycle. It is preferable to spend time in the early stages
of the protocol development than to risk a compromise of security, when
the protocol is put into use. Our approach differs from previous research
into the requirements of protocols which focused on the use of requirements in the verification of protocols [14]; for example, Syverson and
Meadows [15] formalised the requirements of authentication protocols
and used them to verify and find attacks on the Neuman-Stubblebine
protocol.

Hazard Analysis for Security Protocol Requirements

91

The hazard analysis approach described in this paper provides a simpler, more structured and systematic approach to deviation identification
than the heuristic methods in the literature. Work on inquiry-based requirements analysis [8] relies on the use of what-if? questions to prompt
deviations. In goal-based requirements analysis [3, 4], trivial obstacles
are assigned to each goal to investigate the possible ways in which a goal
may fail to complete. These obstacles are identified through the use of an
extensive set of heuristics. The obstacle analysis is further elaborated
through scenario analysis which examines the concrete circumstances
under which goals may fail. Lamsweerde and leitier [16] present formal
and heuristic methods for obstacle identification and resolution based
on temporal logic.
An advantage of the hazard analysis approach for protocol requirements over the temporal logic approach is the focus of the analysis on
security features of the protocol. The temporal logic approach is very
formal, requiring the gathering of the preconditions for the negation of
the goal expressed in logic, these preconditions are obstacles to the goal.
Some formal techniques have missed attacks due to their over abstraction of the protocols, since security attacks may be the result of the
exploitation of properties which are not easily expressible in logic.
Our approach to the analysis of the requirements does not, of course,
guarantee that all the attacks are avoided and secure protocols will be
designed. The requirements analysis process is useful for highlighting
weaknesses and flaws which have previously occurred in protocols.
Attack and threat avoidance techniques prompted by the guidelines
may not be appropriate, for instance, if the recommendations would be
too costly or time consuming. Consideration of the recommendations
should be carefully evaluated with respect to the requirements of the
protocol stakeholders. However, just being aware of potential problems
which may be caused by a particular requirement is an important benefit
of using the method. In such situations, if it is considered appropriate,
higher level requirements may be weakened in the light of the analysis.
Our method is suitable for identifying and investigating common
threats and attacks on protocols and prompting protection mechanisms
against them. This method is a step forward in providing a more structured approach to the development of secure protocols and we believe
that this approach to requirements analysis can be applied more widely
in the field of computer security.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Acknowledgments
We would like to thank John Clark, Jonathan Moffett, colleagues in
the High Integrity Systems Engineering research group and the anonymous reviewers for their helpful comments.

References
[1] Ross Anderson. How to Cheat at the Lottery (or, Massively Parallel Requirements Engineering). Invited Talk a t the 15th Annual Computer Security Applications Conference, Phoenix, Arizona, December 1999.
[2] Ross J. Anderson. Why Cryptosystems Fail. Communications of the ACM,
37(11):32-40, 1994.
[3] Annie I. Antn. Goal-Based Requirements Analysis. In 2nd IEEE International
Conference on Requirements Engineering, pages 136-144, April 1996.
[4] Annie I. Antn. Goal Identification and Refinement in the Specification of
Software-Based Information Systems. PhD thesis, Georgia Institute of Technology, Atlanta, June 1997.
[5] Barry W. Boehm. A Spiral Model of Software Development and Enhancement.
IEEE Computer, pages 61-72, May 1988.
[6] CISHEC. A Guide to Hazard and Operability Studies. The Chemical Industry
Safety and Health Council of the Chemical Industries Association Ltd, 1977.
[7] T. Kletz. HAZOP and HAZAN: Identifying and Assessing Process Industry
Hazards. Institution of Chemical Engineers, third edition, 1992.
[8] Colin Potts, Kenji Takahashi, and Annie I. Antn. Inquiry-Based Requirements
Analysis. IEEE Software , 11(2):21-32, March 1994.
[9] Roger S. Pressman. Software Engineering: A Practitioners Approach. McGraw
Hill, 5th edition, 2000. (European adaption by Daryl Ince).
[10] D.J. Pumfrey. The Principled Design of Computer System Safety Analyses.
Dphil Thesis, University of York, 2000.
[11] Felix Redmill, Morris Chudleigh, and James Catmur. System Safety: HAZOP
and Software HAZOP. Wiley, 1999.
[12] W. W. Royce. Managing the Development of Large Software Systems. In Proceedings of IEEE WESCON, pages 1-9, 1970. Reprinted in Thayer R.H.(ed.)
(1988) IEEE Tutorial on Software Engineering Project Management.
[13] Ian Sommerville. Software Engineering. Addison Wesley, 6th edition, 2000.
[14] Paul Syverson and Catherine Meadows. A Logical Language for Specifying
Cryptographic Protocol Requirements. In IEEE Symposium on Research into
Security and Privacy, pages 165-177. IEEE Computer Society Press, 1993.
[15] Paul Syverson and Catherine Meadows. Formal Requirements for Key Distribution Protocols. In Alfredo De Santis, editor, Advances in Cryptology - EUROCRYPT 94, volume 950 of Lecture Notes in Computer Science, pages 320-331.
Springer, May 1994.

[16] Axel van Lamsweerde and Emmanuel Letier. Handling Obstacles in GoalOriented Requirements Engineering. IEEE Transactions on Software Engineering, 26(10):978-1005, October 2000.

SECURING RMI COMMUNICATION

Vincent Naessens
K. U.Leuven, Campus Kortrijk (KULAK)
[email protected]. be

Bart Vanhaute
K.U.Leuven, Dept. of Computer Science, DistriNet
bart. [email protected]

Bart De Decker
K. U.Leuven, Dept. of Computer Science, DistriNet
[email protected]

Abstract:

Application programmers often have to protect their applications themselves in

order to achieve secure applications. Therefore, they have to possess a lot of
knowledge about security related issues. The solution to this problem is to
separate the security-related modules as much as possible from the real
application and transparently invoke these security modules. By doing this, the
application programmer can build his distributed application without
considering the security requirements.
The case study presents how to achieve transparent security in the RMI
(remote method invocation) system, an API provided by Java to implement
applications in a distributed environment. The presented framework is also
flexible enough to support different levels of security.

Keywords:

open distributed system, security framework

94

1.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

INTRODUCTION

Enterprises are increasingly dependent on their information systems to

support their business activities. Compromise of these systems either-in
terms of loss or inaccuracy of information or competitors gaining access to
it- can be extremely costly to the enterprise. Security breaches, are
becoming more frequent and varied. These may often be due to accidental
misuse of the system, such as users accidentally gaining unauthorized access
to information. Commercial as well as government systems may also be
subject to malicious attacks (for example, to gain access to sensitive
information). Distributed systems are more vulnerable to security breaches
than the more traditional systems, as there are more places where the system
can be attacked. Therefore, security is needed in distributed systems. This
case study presents how to achieve transparent security in the RMI system.
Security protects an information system from unauthorized attempts to
access information or interfere with its operation. The key security features
we are concerned with are:
- identification and authentication to verify parties who they claim to be.
- authorization and access control to decide whether some party can
execute some action.
- protection of communication between parties. This requires trust to be
established between the client and the server, which involves
authentication of clients to servers and authentication of servers to clients.
It also requires integrity and confidentiality protection of messages in
transit,
- audit trail of actions.
Apart from these security requirements, administration of security
information is also needed.

In client/server applications, objects located at one host are

communicating with objects running on other hosts. The key security
features can be provided at two levels: at the location1 level and at the object
level. Security features provided at the location level secure communication
between two hosts. This kind of security is independent of the objects
communicating between these hosts. Each object can also be individually
protected if security is provided at the object level. It is clear that security
provided at the object level is more fine-grained than security provided at the

1
Locations will mostly correspond with hosts; more precisely, they correspond to Java Virtual
Machine instantiations.

Securing RMI Communication

95

location level. We will discuss how each of these security features can be
built into the system.
The main goal of this case study is to provide a flexible and transparent
security framework for the RMI system. Flexibility means that it must be
possible to incorporate different mechanisms and services, according to the
degree of security that is required. Transparency mains that applications are
not aware of the security aspects built into the system. Hence, each of the
security features should be implemented into the RMI system itself. That
way, application programmers do not have to recompile their applications to
work with the secured framework.
A first section briefly describes the architecture of the RMI system. The
second section introduces the security components and discusses where these
services should be added in the RMI system. By including these components
in the RMI system itself, they are transparent with respect to the application.
The third section presents a security framework for RMI that is flexible
enough to support different levels of security. The next two sections discuss
the transparency and the flexibility of the framework. Next, we refer to some
related work in this area. The paper ends with a general conclusion.

2.

THE RMI SYSTEM

application

Stub/ skeleton layer

Remote Refbrence Layer

Transpht
Layer
I
I

Figure 1. the RMI system

The RMI system [1] consists of three layers: the stub/skeleton layer, the
remote reference layer and the transport layer. The application itself runs on
top of this RMI system. When a client invokes an operation on a server
object, a stub object passes the method to the reference layer that initiates the

96

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

call. The remote references are mapped to locations. A specific reference

semantics is executed at that moment depending on the implementation of
the reference layer. For instance, this layer can support point-to-point calls,
calls to replicated objects, etc. The remote reference layer also sets up a
connection to the server side by creating a new connection or reusing an
existing connection. Depending on the implementation of the transport layer,
TCP [2], UDP [3] or other types of connections are supported. When a
server receives information on an incoming connection, the information will
be forwarded to the reference layer that executes code according to a specific
semantics. Finally, the remote object executes the method and sends the
result back to the client side in the same way.

3.

SECURITY COMPONENTS

To achieve a secure execution environment, some security components

must be added into the distributed system. The security components
discussed in this paper are the association component, the authentication
component, the access control component, and the audit trail component.
This section shows where these four security services are added into the RMI
system. By including these services in the RMI system, they are transparent
to the application.
Services can be added at two levels: the location level and the object
level. Services provided at the location level are executed between hosts.
Information provided at that level are the IP addresses of the communicating
hosts, the principals executing at each of the two hosts, etc. Services
provided at the object level are executed between objects. More information
is available at that level. The method name and parameters of the remote
invocation are known. Moreover, an object can be running on behalf of a
certain principal. An access controller at the object level can make use of
this information.

5?-Cli fnt.

Figure 2. Security services in the RMI system.

Securing RMI Communication

3.1

97

Secure association service

Before messages are sent over the wire, a secure association must be
established between two hosts: the client and the server. This service is
provided at the location level. As a result of this phase, both parties possess
a key that will be used to exchange further messages. Thus, setting up a
secure association guarantees the confidentiality of the data that is exchanged
between both parties. The secure RMI system performs this task after the
connection is established and before the actual method invocation from the
stub object to the server object takes place. This task can be fully executed at
the transport layer, making use of the connection. The resulting key is also
kept at the transport layer. As this service does not require any information
about the objects, the same secure association can and will be reused over
multiple calls between the two hosts.

3.2

Authentication service

Once a secure association is set up, an authentication service can be

executed. Often, both parties will want to know the correct identity of the
party they are dealing with, for instance as basis for authorization decisions.
Alternatively, they may want to act anonymously. Authentication can be
performed in a kind of handshake phase where trust is gained in the other
partys identity and where security attributes are exchanged. This service
can be fully performed at the transport layer, immediately after a secure
association is set up. The resulting security attributes are also stored at the
transport layer. Depending on the implementation, authentication is executed
at the location level and/or at the object level. The presented framework only
presents authentication at the location level.. This corresponds to the idea that
users are typically controlling locations, and they are the principals we want
to authenticate.

3.3

Access control service

The access control service (or authorisation service) gives a party the
possibility to allow/disallow an action of the other party involved in the
communication. In an object oriented environment, access decisions can be
based on the method and the parameters that are sent to the server. This
service is performed at the object level. Thus, access control must be
performed at the reference layer, after the necessary information is
unmarshalled and before the method will be invoked. This service can also
make use of the security information that is stored at the transport layer.

98

3.4

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Audit trail service

The audit trail service is responsible for logging information. Two types
of logging are introduced. In the transport layer (i.e. at location level),
information about the authentication procedure is logged. At the reference
level (i.e. at object level), information about the authorisation and the method
invocation is logged.

4.

THE SECURITY FRAMEWORK

We developped a security framework for RMI that is flexible enough to

support different security levels and mechanisms. By consulting a property
file, the security components are loaded into the RMI system at runtime. By
changing the values of this property file, other components are loaded into
the system. On the one hand, objects are loaded that are responsible for
holding security information. They are called security context objects (or
security contexts). On the other hand, objects are loaded that are responsible
for executing a specific security service. They are called security service
objects (or security services). Security services can modify the information
stored in the security contexts and query them to make decisions.

Figure 3 . The security framework

Securing RMI Communication

4.1

99

Security context objects

To obtain a secure execution environment, two types of security contexts

are introduced in the RMI system: a connection security context and an
invocation security context. They are responsible for storing security-related
data.
A connection security context contains security information specific for
a particular connection. This context contains information exchanged during
the secure association phase and the authentication phase at location level.
More specifically, a connection security context can hold a session key, the
time when the connection is created, the user or client that makes use of the
connection, etc. Thus, every time a new connection is created, a
corresponding new connection security context is initiated at the same level
in the RMI system i.e. at the transport level. A connection security context
disappears when the corresponding connection is closed.
An invocation security context holds information that is specific for a
particular invocation such as the time the invocation is executed, the
operation that must be executed and the parameters that belong to the
operation. Thus, a new invocation security context is created each time a
new call is initiated and is removed when the method call is finished. This is
analogous to the first type of security context. When authentication is
executed at object level, additional information is added into this context.
Remark that a connection security context can be considered a part of an
invocation security context. Every invocation security context holds a
pointer to a connection security context. However, the lifetime of a
connection security context can be longer than the lifetime of an invocation
security context. This is because the same connection can be reused during
subsequent method calls.

4.2

Security service objects

Security service objects are responsible for executing some kind of

security service. When a client invokes a method on a server object, a secure
association is established and a particular authentication protocol is
performed between the client and the server. To achieve these two tasks, a
vault object [4] is introduced at the transport level. A vault object can
perform these two tasks itself or delegate the work to an association object
and an authentication object.

100

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 4 . TheVault Object

When a call is initiated at the client side and a new unsafe connection is
created, the association object can decide to exchange a session key with the
association object on the server side. Several encryption libraries provide
implementations of key agreement algorithms [ 5 ] . The resulting session key
is stored at both sides in the connection security context. As a result of this
step, further information can be sent in encrypted form to the connection
object. In other words, encryption is done on top of a connection and
therefore, it does not affect the implementation of a connection type.
Moreover, if the association object sees that the connection itself is
implemented to support secure communication (for instance by using SSL
secure sockets), it can decide not to execute this first step. When a
connection already exists, the association object can decide to update the
connection security context if necessary. For instance when the time a
particular key is valid, is exceeded, the vault object can ask for a new key
agreement session to take place.
After this, the vault object calls an authentication object if
authentication is not already done. Depending on its implementation, the
authentication object explicitly asks the user for authentication information
or makes use of credentials that are created when the user logs in on the
system. These credentials are generated automaticalIy when the user logs in
on the system. It can happen that authentication is performed in several
successive steps. For instance, the server side can ask for additional
credentials or can conclude that the authentication data are not valid any
more. In these two cases, the authentication continues. Authentication

Securing RMI

Communication

101

information can be sent along a secure data stream making use of the session
key obtained in the previous step. The authentication information is stored in
the connection security context and can later be used to make access
decisions.
Access control in an object-oriented environment mostly depends on the
method that must be executed and the parameters of the method call. At that
point in the execution, the information must be in an unmarshalled form.
Marshalling and unmarshalling happens in the remote reference layer. This
information is passed to the invocation security context object. After this
information is set, an access control object can make a decision using the
information kept by the security context. At the client side, access control
can be checked just before marshalling information; at the server side access
control executed after unmarshalling the operation and parameters and just
before the information is dispatched to the application level.

posh

poririrn

posirirn

negative veto

ncgrtivcvcto

Figure 5. Access controller

To provide a flexible access control mechanism, the access control object

can be implemented using the composite design pattern [6]. A tree of access
controllers makes an access decision. At the leaf level, the access controllers
give a negative veto or advice, or a positive veto or advice to the intermediate
nodes in the tree. This information is propagated to the top level of the tree
that makes a final decision. Each access controller makes a particular
decision. For instance, there can be user access controllers, rights access
controllers, ... These access controllers can be implemented totally
independent of the actual application. To give the application the possibility
to attach his own access controller to the tree, it can give a series of access
controllers to the constructor of an application object. The constructor then
appends the controllers to the tree in a predefined way.

102

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Two types of log objects are introduced in the system: association log
objects and operation log objects. Association loggers are introduced at the
transport level and log information concerning the association. For instance,
an association logger can save which client is trying to make a connection, if
the authentication is successful, etc ... Operation loggers are introduced at
the reference layer and log information about the operations that have to be
executed or that have already been executed. For instance, an operation
logger can store the method a client tries to execute, the return value of the
access control decision object, the result of the method call, etc. In contrast
to vault objects and access control objects, we want to provide the possibility
to pick up several log objects at each level.

5.

TRANSPARENCY

Because the presented security features are all built into the RMI system,
it can be reused for every application. Access control and operation logging
happens at the reference layer; setting up a secure association and logging
associations happens at the transport layer. This also implies that stub
objects remain the same. Therefore, the rmic compiler that generates stubs,
does not have to be changed. This implies full transparency from the point of
view of the application programmer.
Providing full transparency to the end user of the application is difficult
to achieve. A secure distributed system wants the user to be authenticated at
some point in the execution. Depending on the implementation of the
authentication object, the user has to do it explicitly during the application
runtime or the authentication object can make use of the credentials created
when the user logs on the system.
From the point of view of the administrator of the system, one can say
that he has to make a decision about which security components have to be
loaded into the RMI system. He has to make a property file. The RMI
system consults this property file at runtime in order to know which instances
of the security components to create.
The presented framework can also be considered to be relatively
transparent to the RMI implementation because security components are
added to the system by loading security related objects and not by adapting
the implementation of existing objects in the system. For instance, a typical
connection implementation (UDP or TCP) does not have to be adapted
because encryption is provided on top of it.

Securing RMI Communication

6.

103

FLEXIBILITY

Four types of objects are introduced in the framework: security context

objects, access control objects, log objects and vault objects. In turn, a vault
object can call an association object and an authentication object. An
appropriate interface for each of these object types is available so that the
RMI system can invoke a method of an object via this interface. The
property file indicates which objects to load at runtime in the system.
Separating the security components from the RMI system this way provides
us a flexible way of working. Although a secure RMI package can provide
us with implementations of each of these objects, new implementations can
be introduced as long as they implement methods of the interface in an
appropriate way.
Flexibility is also needed within the proposed security components. For
instance, by implementing an access decision object as a tree of access
controllers, new access controllers can be added dynamically. Vault objects
present a similar degree of flexibility in that way they can decide to contact
an association object and an authentication object, contact one of those two
types of objects or contact no other object at all according to the level of
security that is preferred in the system.

7.

RELATED WORK

The Java Secure Socket Extension (JSSE) [7] is a Java optional package
that provides Secure Socket Layer (SSL) and Transport Layer Security (TLS)
support for the Java 2 Platform. Using JSSE, developers can provide for the
secure passage of data between a client and a server. Secure sockets can be
added into the RMI system at transport level to set up a secure association.
This way, they are transparent in front of application programmers. In the
presented framework, the Vault object is responsible for setting up a secure
association between two hosts. An implementation of that Vault object can
use JSSE.
The Java Authentication and Authorization Service (JAAS) [7] is a
framework that supplements the Java 2 platform with principal-based
authentication and access control capabilities. It includes a Java
implementation of the standard Pluggable Authentication Module (PAM)
architecture, and provides support for user-based, group-based, or role-based
access controls. These modules can also be added transparently into the
presented framework.
The Java Authentication Service provides

104

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

authentication at object level. The framework we presented provides

authentication at location level. However, we can extend the RMI security
framework with authentication at object level as suggested in paragraph 3.
The authorisation modules of JAAS can also be inserted into the framework
in the Access Decision Object. But the security framework is flexible
enough to support other types of access control. For instance, access control
can also be based on the parameters and the operation that is invoked.
Because Java has not specified standards for other types of authorisation, we
have to make an own implementation of each of these services if that is
required.
The Common Object Services specification (CORBASec) [4] describes
security related tasks and requirements needed for CORBA. The
specification is quite long and attempts to address an extremely wide range
of security issues. The topic of distributed objects is complicated enough
when considered on its own and it certainly does not get any simpler with the
addition of security. Due to this, there are many issues that are
underspecified and open to interpretation at this time, which gives scope for
R&D in this area. To further extend the RMI security architecture with more
advanced security services like delegation, a lot of inspiration can be found
in this specification. Depending of the implementation of an ORB, different
services are provided. This is similar with the flexibility of the presented
RMI security framework.
The Java Community [ 8 ] is working on the definition of a high-level

API for network security in JavaTM 2 Standard Edition RMI, covering basic
security mechanisms: authentication (including delegation), confidentiality,
and integrity. The main problem is that the proposals are not transparent
enough towards applications. Our framework tries to achieve more
transparency towards application programmers because all of the security
features are built into the RMI system itself. However, the framework also
enables application programmers to load their own security modules into the
RMI system.

8.

CONCLUSION

The presented framework gives the possibility to add different security

services to the RMI system: setting up a secure association, authentication,
authorisation and logging. These services are added to the RMI system in a
transparent and flexible way. The implementation of the suggested objects in
the framework depends on the level of security and the degree of complexity

Securing RMI Communication

105

that is needed. A simple implementation can already provide a good level of

security. For a more advanced implementation of each of these objects, a lot
of principles suggested by security specifications of other distributed systems
such as CORBA [4], can be used.

REFERENCES
1.
2.
3.
4.
5.

6.
7.
8.

Java Remote Method Invocation Specification - JDK1.2 Beta 1, October 1997

RMI implementation provided by SUN using TCP connections.
http://iava.sun.com/products/jdk/rmi/
RMI implementation provided by Ninja using UDP connections.
http://ninja.cs.berkelev.edu/ninja/
Corba Security Service Specification - november 1996.
IAIK: http://www.iaik.tu-graz.ac.at/
ABA: http://aba.net.au/
Erich Gamma, Richard Helm, Ralph Johnson and John Vlissides, 'Design Patterns,
elements of reusable objectoriented software', Addison-Wesley 1995
http://www.java.sun.com/security/
http://java.sun.com/aboutJava/communityprocess/jsr/jsr 076 rmisecuritv.html

This page intentionally left blank.

SECURE JAVA DEVELOPMENT

WITH UML
Jan Jrjens*
Computing Laboratory, University of Oxford, GB
http://www.jurjens.de/jan - [email protected]

Abstract

Developing secure software systems is difficult and error-prone. Numerous implementations have been found vulnerable in the past; a recent
example is the unauthorised access to millions of online account details
at an American bank.
We aim to address this general problem in the context of development
in Java. While the JDK 1.2 security architecture offers features (such
as guarded objects) that provide a high degree of flexibility and the
possibility to perform fine-grained access control, these features are not
so easy to use correctly.
We show how to use a formal core of the Unified Modeling Language
(UML), the de-facto industry-standard in object-oriented modelling, to
correctly employ Java security concepts as such as signing, sealing, and
guarding objects. We prove results for verification of specifications wrt.
security requirements. We illustrate our approach with a (simplified)
account of the development of a web-based financial application from
formal specifications.

Keywords: Distributed systems security, access control, mobile code, Java security,
secure software engineering, Unified Modeling Language.

1.

Introduction

The need to consider security aspects in the development of many

systems today is not always met by adequate knowledge on the side of
the developer. This is problematic since in practice, security is compromised most often not by breaking the dedicated mechanisms (such
as encryption or access control), but by exploiting weaknesses in the
way they are being used [And01]. Thus security mechanisms cannot be

*Supported by the Studienstiftung des deutschen Volkes and the Computing Laboratory.

108

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

blindly inserted into a security-critical system, but the overall system

development must take security aspects into account.
Especially dynamic access control mechanisms such as provided by
Java since the JDK 1.2 security architecture [Gon99; Kar00b] in the form
of GuardedObjectscan be difficult to administer since it is easy to forget
an access check [Gon98; BV99]. If the appropriate access controls are
not performed, the security of the entire system may be compromised.
Additionally, access control may be granted indirectly and unintentionally by granting access to an object containing the signature key that
enables access to another object. In this work, we aim to address these
problems by providing means of reasoning about the correct deployment
of security mechanisms such as signed, sealed and guarded objects using
a formal core of the widely used object-oriented design language Unified
Modeling Language (UML), extending previous work [Jr01f; Jr01a].
The more general aim of this work is to use UML to encapsulate
knowledge on prudent security engineering and thereby make it available
to developers not specialised in security [Jr01b]. Thus the approach
to use UML for security covers not just access control, but also other
security functions and requirements.

Overview.
After presenting some background on access control in
Java in the following section, we summarise our use of UML in section 3.
In Section 4 we outline the part of a design process relevant to enforcing
access control in Java and give some results on verifying access control
requirements. In Section 5 we illustrate our approach with the example
of the development of a web-based financial application from formal
specifications. We end with an account of related work, a conclusion
and indication of future work. Proofs have to be omitted due to space
reasons and will appear in an extended version.

2.

Access control in Java

Authorisation or access control [SS94] is one of the corner-stones of

computer security. The objective is to determine whether the source of
a request is authorzsed to be granted the request. Distributed systems
offer additional challenges: The trusted computing bases (TCBs) may be
in various locations and under different controls. Communication is in
presence of possible adversaries. Mobile code is employed that is possibly
malicious. Further complications arise from the need for delegation (i. e.
entities acting on behalf of other entities) and the fact that many security
requirements are location-dependent (e.g., a user may have more rights
at the office terminal than when logging in from home).

Secure Java Development with UML

109

Object-oriented systems offer a very suitable framework for considering security due to their encapsulation and modularisation principles
[FDR94; Var95; ND97; Gol99; Sam00].
In the JDK 1.0 security architecture, the challenges posed by mobile code were addressed by letting code from remote locations execute
within a sandbox offering strong limitations on its execution. However,
this model turned out to be too simplistic and restrictive. From JDK
1.2, a more fine-grained security architecture is employed which offers a
user-definable access control, and the sophisticated concepts of signing,
sealing, and guarding objects [Gon99; Kar00b].
A protection domain [SS75] is a set of entities accessible by a principal.
In the JDK 1.2, permissions are granted to protection domains (which
consist of classes and objects). Each object or class belongs to exactly
one domain.
The system security policy set by the user (or a system administrator) is represented by a policy object instantiated from the class
java.security.Policy. The security policy maps sets of running code (protection domains) to sets of access permissions given to the code. It is
specified depending on the origin of the code (as given by a URL) and
on the set of public keys corresponding to the private keys with which
the code is signed.
There is a hierarchy of typed and parameterised access permissions, of
which the root class is java.security.Permission and other permissions are
subclassed either from the root class or one of its subclasses. Permissions
consist of a target and an action. For file access permissions in the class
FilePermission, the targets can be directories or files, and the actions
include read, write, execute, and delete.
An access permission is granted if all callers in the current thread
history belong to domains that have been granted the said permission.
The history of a thread includes all classes on the current stack and
also transitively inherits all classes in its parent thread when the current
thread is created. This mechanism can be temporarily overridden using
the static method doPrivileged() .
Also, access modifiers protect sensitive fields of the JVM: For example,
system classes cannot be replaced by subtyping since they are declared
with access modifier final.
The sophisticated JDK 1.2 access control mechanisms are not so easy
to use. The granting of permissions depends on the execution context
(which however is overridden by doPrivileged(), which creates other subtleties). Sometimes, access control decisions rely on multiple threads. A
thread may involve several protection domains. Thus it is not always
easy to see if a given class will be granted a certain permission.

110

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

This complexity is increased by the new and rather powerful concepts

of signed, sealed and guarded objects [Gon99]. A SignedObject contains
the (to-be-)signed object and its signature.' It can be used internally as
an authorisation token or to sign and serialise data or objects for storage
outside the Java runtime. Nested SignedObjects can be used to construct
sequences of signatures (similar to certificate chains).
Similarly, a SealedObject is an encrypted object ensuring confidentiality.
If the supplier of a resource is not in the same thread as the consumer, and the consumer thread cannot provide the access control context information, one can use a GuardedObject to protect access to the
resource. The supplier of the resource creates an object representing
the resource and a Guardedobject containing the resource object, and
then hands the GuardedObject to the consumer. A specified Guard object incorporates checks that need to be met so that the resource object
can be obtained. For this, the Guard interface contains the method
checkGuard, taking an Object argument and performing the checks. To
grant access the Guard objects simply returns, to deny access is throws
a SecurityException. GuardedObjects are a quite powerful access control
mechanism. However, their use can be difficult to administer [Gon98].
For example, access to an object may be granted indirectly (and possibly unintentionally) by giving access to another object containing the
signature key for which the corresponding signature provides access to
the first object.

3.

Developing Secure Systems with UML

To address these issues, we extend previous work [Jr01f; Jr01a] to

employ a formal core of the Unified Modeling Language (UML) [UML01],
the de-facto industry standard in object-oriented modelling (an excellent
introduction is given in [SP00]). We would like to ensure that the protection mechanisms that are in place do offer the required level of security.
Specifically, we check the specified dynamic behaviour against expressed
security policies. We do this on the level of specification (rather than
the implementation level) because design mistakes can so be corrected
as early as possible, and because formal reasoning is more feasible at a
more abstract level.
UML consists of several kinds of diagrams describing the different
views on a system. We use only a simplified fragment of UML (together with a formal semantics) to enable formal reasoning and keep
1Note that signing object is different from the signing of JAR files.

Secure Java Development with UML

111

Op 1(arg 1:ATy 1 ):RTy1

Figure

1. Class diagram

the presentation concise. We use its standard extension mechanisms to

express security aspects. As a formal semantics for UML is subject of
ongoing research, we use a (simplified) semantics tailored to our needs
for the time being, just to illustrate our ideas. Note, however, that our
approach does not rely on use of a formal semantics; in fact we aim for
a tool to automatically check the considered security notions, and then
these may also be explained informally (which is more accessible, but
may be more prone to misunderstanding).
We use the following kinds of diagrams: class diagrams, statechart
diagrams, and deployment diagrams.
We define the diagrams using their abstract syntax for conciseness
and to enable formal reasoning. We also give the concrete syntax (in a
way that the translation between the two should be apparent).

3.1.

Class Diagrams

Using class diagrams we can model which objects are signed or sealed
with which keys, and which are guarded by which Guard objects.
An attribute specification A = (att_name, att_type, init_value) is given
by a name att-name, a type att-type and an initial value init-value.
An operation specification 0 = (op_name, Arguments, op_type) is given
by a name op_name, a set of Arguments and the type op-type of the return
value. The set of arguments may be empty and the return type may be
the empty type denoting absence of a return value. An argument
A = (arg_name,arg_type) is given by its name arg_name and its type
arg_type.

A class model C = (class-name, (tag, value), AttSpecs, OpSpecs, State)

(tag, value)
value) pair (written
is given by a name class-name, an optional (tag,
in curly brackets), a set of attribute specifications AttSpecs, a set of
operation specifications OpSpecs and a statechart diagram State giving
the object behaviour. The tag may be either of signed, sealed or guarded
(indicating a signed, sealed or guarded object), and the value is either
the public key corresponding to the private key with which the object
was signed or sealed, or it is the name of the corresponding Guard object.

112

ADVANCES IN N E T W O R K AND DISTR. SYSTEMS SECURITY

message(x)[condition]

Transition
Figure 2.

Initialstate

State

Start marker

Statechart Diagrams

A class diagram D = (Cls, Dependencies) is given by a set Cls of class

models and a set of Dependencies. A dependency is a tuple (client, supplier,
stereotype) consisting of class names client and supplier and a label (called
stereotype) indicating the kind of dependency (e.g. call).

3.2.

Statechart diagrams

We use statechart diagrams to specify the behaviour of objects, in

particular of the Guards.
We fix a set Var of (typed) variables x, z , y, . . .. We define the notion of a statechart diagram for a given class model C: A statechart
diagram S = (States,init_state, Transitions) is given by a set of States
(that includes the initial state init_state) and a set of Transitions. (In the
concrete syntax, the initial state is signified with a start marker.)
A statechart transition t = (source, event, condition, Actions, target) has
a source state, an event, a condition, a list of Actions and a target state.
An event is the name of an operation with a list of distinct variables as
arguments (e.g. op(x, y, z ) ) . Let the set Assignments consist of all partial
functions that assign to each variable and each attribute of the class C
a value of its type. A condition 2 is a function g : Assigments + Bool
evaluating each assignment to a boolean value. We write it as a sequence
of Boolean propositions with variables and attribute names that is interpreted as their conjunction; conditions are written in square brackets.
An action can be either to assign a value v to an attribute a (written a := v), to call an operation op with values v1, ... ,vn (written
op(v1,. . . ,vn), to return values v1,. . . , vn as a response to an earlier call
of the operation op (written returnop(vl,. . . ,w,)), or to throw an exception. In each case, the values can be constants, variables or attributes.
In the concrete syntax, actions are preceded by a backslash.

3.3.

Deployment diagrams

Deployment diagrams describing the physical layer of a system are

security-relevant in so far as they give the locations of the different components of the system (used in the access permissions) and they give
2
2We
We do not use the UML term guard here to avoid confusion with guard objects.

113

Secure Java Development with UML

Dependency

Figure 3.

Deployment diagram

information on kinds of the communication links between different components, inducing threat scenarios wrt. the physical security.
A system node N = (location, Components) is given by its location
(e.g. a URL or local system) and a set of contained Components.
A deployment diagram D = (Nodes, Links, Dependencies) is given by a
set of Nodes, a set of communication Links between nodes and a set of
logical Dependencies between components. A link l = (nds, stereo) consists of a two-element set nds of nodes being linked and a label (called
stereotype) indicating the kind of the link (e.g. Internet). Here a
dependency is a tuple (client, supplier, interface, tag) consisting of components client and supplier and a label (called tag) indicating the kind
of dependency (e.g.
ee { rmi}).

4.

Design process

We sketch the part of a design process for secure systems using UML
that is concerned with access control enforcement using guarded objects.

(1) Formulate the permission sets for access control for sensitive objects.
(2) Use statecharts to specify Guard objects that enforce appropriate
access control checks.

(3) Verify that the Guard objects protect the sensitive objects sufficiently by showing that they only grant access implied by the
security requirements.

(4) Ensure that the access control mechanisms are consistent with the
functionality required by the system by showing that the other
objects may perform their intended behaviour.
(5) Verify that mobile objects are sufficiently protected by considering
the threat scenario arising from the physical layer given in the
deployment diagram.

114

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Here the access control requirements in step (1) can be of the following
form:
origin of requesting object (based on URL)
signatures of requesting object
external variables (such as time of day etc. ).
In Section 5 we sketch a formal verification of a specification following
these steps. They enforce the following two requirements.

Security requirement: Check that the access control requirements

are strong enough to prevent unauthorised influence, given the
threat scenario arising from the physical layer.

Functionality requirement: Check that the access control requirements formulated are not overly restrictive, denying legitimate access from other components of the specification.
The functionality requirement is important since it is not always easy
to see if stated security requirements are at all implementable. If their
inconsistency is only noticed during implementation then, firstly, resources are wasted since work has to be redone. Secondly, most likely
security will be degraded in order to reduce this extra work.

4.1.

Verification

In this subsection, we sketch results to be applied in the above approach. The idea is to verify security properties by linking the different
views on a system given by the various kinds of diagrams. We convey
our ideas using a simplified semantics for UML statechart diagrams.
Any statechart diagram S defines a function US] from sequences of
input events to sets of sequences of output actions, each possibly with
arguments, often involving use of cryptographic operations (as detailed
in [Jr01f]). We say that S may eventually output a value v if there exists
a sequence e' of input events and a sequence a' E [Sg(Z)of corresponding
output actions such that v is output by one of the actions in a' (in
cleartex t ) [Jr01e].
The following definition uses the notion of an adversary from [Jr01e],
which is a function from sequences of output actions of the statechart
S to sequences of input events of S that captures the capabilities of an
3In future work we intend to formalise these requirements using an abstract security policy
specification language, enabling automatic generation of the corresponding guard object specifications.

Secure Java Development with UML

115

adversary intercepting the Internet communication links between S

and the other objects (the exact definition of adversary , without prior
knowledge and of the composition 3of the statechart interpretation [Sj
with the adversary A can be found in [Jr01e]).

Definition 1 A statechart diagram S preserves the secrecy of a value

K if there is no adversary A (eavesdropping on the Internet links)
without prior knowledge of K such that IS] 63 A may eventually output
K.
This definition is extended to system components by composing the functions arising from the statechart diagrams specifying the objects of a
given component.
Intuitively, then, a system component C preserves the secrecy of K if
no adversary can find out K in interaction with the system modeled by
C, following the approach of Dolev and Yao (1983), cf. [Aba00; Jr01e].
The following result is applied within the approach of subsection 4 to
the UML specification of a security-critical system (for a proof of this as
well as the following results cf. [Jr01d]).

Theorem 1 Suppose that the access to a certain resource is according

to the Guard object specifications granted only to objects signed with a
key K. Suppose all components preserve the security of K. Then only
objects signed with K according to the specification will be granted access
to the resource.
Before coming to the main example in the next section, we give a
short example to point out that the kind of weaknesses in Java security
access control can be quite subtle (rather than just mistakingly sending
out secret keys or forgetting to set access rules):

Example.
The statechart in Figure 4 describes the behaviour of a
guard object grd enforcing a slightly more complicated access control
policy. The idea is that an entity named req may establish a shared key
K M in order to submit keys KSprotected by KM such that objects signed
with KS should be granted access to the guarded object. Here we assume
that the keys KS may be updated frequently, so that it is more efficient
to use the symmetric key KM to protect KS (rather than the public
key associated with grd). The identity of req is taken as given and is
bound to a public key in the certificate cert signed with the key Kc of
a certification authority (assuming RSA-type encryption and signing).
On request cert(), the guard object sends out a self-signed certificate
certifying its public key K. The object req sends back the symmetric key

116

ADVANCES I N N E T W O R K AND DISTR. SYSTEMS SECURITY

recMaster

Init
\return(k,Dec

(grd,K))

mst(key,cert)
[fst( (cert) )-req]

[otherwise] \throw new SecurityException()

Figure 4

Statechart Example

K M signed with its private key (corresponding to the public key in cert)
and encrypted under K, together with the certificate cert (the functions
fst resp. snd applied to a pair returns its first resp. second component).
The guard object can receive the signature key KS encrypted under K M
and will then grant access to those objects signed by Ks.
Thus a typical message exchange to establish Ks may look like the
following:

return( K,DecK-

1 (grd,K))

grd

req

Unfortunately, this access control mechanism contains a flaw: An

adversary A intercepting the communication between req and grd (and
modifying the exchanged values) can find out K M and thus make grd
accept a key KS chosen by A . The critical part of the message exchange
corresponding to this attack is as follows:
grd

return(K,DecK-l (grd,K))

return(KA,Dec

*A

l(grd,KA))
Kii

* req

117

Secure Java Development with UML

<<Internet>>

Server

Figure 5.

Deployment diagram

Here the theorem above does not apply since the security of the signing
key Ks is violated (in a subtle way). With our approach one can exhibit
subtle flaws like this (in this case, one would notice the flaw e.g. when
trying to show formally that the assumptions of the above theorem are
fulfilled). - This example is quite realistic; in fact it is derived from a
published protocol which was found to be flawed in [Jr01e] (cf. there
for details).

5.

Example Financial Application

We illustrate our approach with the example of a web-based financial application. The example was chosen to be tractable enough given
the space restrictions but still realistic in that it points out some typical
issues when considering access control for web-based e-commerce applications (namely to have several entities - service-providers and customers
- interacting with each other while granting the other parties a limited
amount of trust and by enforcing this using credentials).
We first describe the physical layer of the application in a UML diagram and state its security requirements. We show in UML diagrams
how to employ GuardedObjectsto enforce these security requirements.
We prove that the specification given by the UML diagrams is secure
by showing that it does not grant any access not implied by the security requirements. We end the section by giving supplementary results
regarding consistency of the security requirements.
Two (fictional) institutions offer services over the Internet to local
users: an Internet bank, Bankeasy, and a financial advisor, Finance.
The physical layer is thus given in Figure 5.
To make use of these services, a local client needs to grant the applets
from the respective sites certain privileges.

118

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

StoredFinan
BankData: FDal
Read():FDat

-1 SignedReq: Obj

Figure 6.

Class diagram

(1) Applets that originate at and are signed by the bank can read
and write the financial data stored in the local database, but only
between 1 pm and 2 pm (when the user usually manages her bank
account).

(2) Applets from (and signed by) the financial advisor may read an
excerpt of the local financial data created for this purpose. Since
this information should only be used locally, they additionally have
to be signed4 by a certification company, CertiFlow, certifying that
they do not leak out information via covert channels.
(3) Applets originating at and signed by the financial advisor may use
the micropayment signature key of the local user (to purchase stock
rate information on behalf of the user), but this access should only
be granted five times a week.

Financial data sent over the Internet is signed and sealed to ensure
integrity and confidentiality. Access to the local financial data is realised
using GuardedObjects. Thus the relevant part of the class diagram is
given in Figure 6.
[orig in=signed=bankeasy,timeslot]ketum

[otherwise] \throw new SecurityExceptionO

Figure 7.

Statechart FinGd

Here we assume that SignedObject is subclassed to allow multiple signatures on the same
object [Gon99].
[Gon99].

Secure Java Development w i t h U M E

119

[origin=finance,signed=( finance.certiflow )]]\return

[otherwise]\throw new SecurityException()

Figure 8.

Statechart ExcGd

As specified in the class diagram, the access controls are realised by

the Guard objects FinGd, ExpGd and MicGd, whose behaviour is specified
in Figures 7, 8 and 9 (we assume that the condition timeslot is fulfilled
if and only if the time is between lpm and 2pm, that the condition
weeklimit is fulfilled if and only if the access to the micropayment key
has been granted less than five times in the current calendar week, and
that the method incThisWeek increments the relevant counter).
[origin=signed=finance,weeklimit]]\incThisWeek \return

[otherwise] \throw new SecurityException()

Figure 9.

Statechart MicGd

Now according to step (3) in Section 4, we prove that the specification

given by UML diagrams is secure in the following sense.

Theorem 2 The specification given b y UML diagrams for the guard objects does not grant any permissions not implied by the access permission
requirements given in (1)-(3).
Regarding step (4) in Section 4, we exemplarily prove that InfoAp
In
can
purchase the article on behalf of the user, as intended.

Theorem 3 Suppose all applets in the current execution context originate from and are signed by Finance, and that use of the micropayment
key is requested, which has happened less than five times before in the
current week. Then the current applet is permitted to purchase articles
on behalf of the user.
Finally, following (5) in Section 4, the mobile objects are sufficiently
protected since all objects sent over the Internet were required to be
signed and sealed (a more detailed discussion has to be omitted).

120

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

6.

Related Work

In [Jr01f; Jr01a] we considered how to model various aspects of

general systems security (including multi-level security, secure information flow and security protocols) with UML. [Jr01c] applies UML to
reason about audit-security in a smart-card based payment scheme and
[Jr01b] shows how to use UML to enforce general principles of secure
systems design from [SS75]. There seems to be little other systematic
work yet in applying UML to security.
Java 2 security and in particular the advanced topics of signed, sealed
and guarded objects is explained in [Gon99]. There has also been some
work giving formal reference models for Java 2 access control mechanisms, thus clarifying possible ambiguities in the informal accounts
and enabling proof of compiler conformance to the specification [KG98;
WF98; Kar00b] (but without considering signed, sealed or guarded objects). To our knowledge, the use of signed, sealed or guarded objects in
JDK 1.2 has not previously been considered in a formal model.
[HKK00] introduces higher-level abstractions for Java security policy
rules, simplifies security management and gives additional functionality.
General Java security is considered e.g. in [GAS99].
There has been extensive work regarding formal models for security,
mostly about security protocols (for an overview cf. [GSG99; RSG+01]).
A logic for access control was introduced in [ABLP93].

Conclusion and Future Work

7.

To summarise, we used a core of UML, the industry standard in

object-oriented modelling, to specify and reason about access control
in distributed Java-based systems. We have concentrated on advanced
JDK 1.2 access control mechanisms such as signing, sealing and guarding objects. We show how to specify security requirements and to prove
that modelled access control mechanisms such as guarded objects meet
their goals and that these mechanisms are consistent with the overall
functionality required from the system.
In conclusion, it seems that our approach is both worthwhile and
feasible:

Using the JDK 1.2 access control mechanisms can be rather complicated in practice (especially when indirect access permissions
using authorisation tokens are employed), thus providing support
for correct specification of the relevant mechanisms in the context
of a widely used specification as UML seems quite useful.

Secure Java Development with UML

m

121

In this paper, we could only illustrate our approach using a rather

simple example. However, UML allows a high degree of abstraction
in modelling systems. So we expect the approach to scale up rather
well. This is currently validated in practice in a Masters thesis
developing an Internet-based auction system [Mea01].

A further benefit is that by using a widely accepted notation, our approach to secure Java development can be integrated with other work on
secure systems using UML (e.g. on electronic purse systems [Jr01c]).
As to the limitations of this first step in this direction of research, our
account remains relatively abstract for space restrictions and conciseness
of presentation. As a next step, one should consider more details of
Java security, such as the use of access modifiers (private, final,...), the
doPrivileged() method and the implies() method. Also, an extension to
JAAS [LGK+99; Kar00b] is planned.
Work in progress aims to provide tool support to validate UML specifications of access control guards against security requirements, building on work in [CCR01].
Regarding future work, it would be very useful to have a way to generate the correct behaviour specification of guard objects in statechart diagrams automatically from the (formalised) security requirements. Also,
it would be interesting to try to extend our approach to the extension
of the Java security architecture proposed in [HKK00]. We intend to
address CORBA security (cf. e.g. [VH96; Kar00a]) in a similar way.

Acknowledgments
The idea for the line of work using UML for security arose when doing
security consulting for a project during a research visit with M. Abadi
at Bell Labs (Lucent Tech.), Palo Alto, whose hospitality is gratefully
acknowledged. This work benefitted from discussions at the summer
school Foundations of Security Analysis and Design 2000 (in particular, Li Gong suggested to apply the UML-based approach to security to
guarded objects) and the Dagstuhl seminar "Security through Analysis
and Verification (in particular with D. Gollmann and B. Pfitzmann).
The work was presented in two talks at the Computing Laboratory at
the University of Oxford. Comments from S. Abramsky, C. Crichton
and G. Lowe are gratefully acknowledged. Finally, comments by the
anonymous referees have been very helpful.

122

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

References
[Aba00]

M. Abadi. Security protocols and their properties. In F. Bauer and

R. Steinbrueggen, editors, Foundations of Secure Computation. IOS Press,
2000.

[ABLP93] M. Abadi, Michael Burrows, Butler Lampson, and Gordon Plotkin. A

calculus for access control in distributed systems. ACM Transactions on
Programming Languages and Systems, 15(4):706-734, 1993.
[And01]

R. Anderson. Security Engineering: A Guide to Building Dependable Distributed Systems . Wiley, 2001.

B. Bokowski and J. Vitek. Confined types. In 14th Annual ACM SIGPLAN Conference on Object Oriented Programming Systems, Languages,
and Applications (OOPSLA '99), 1999.
[CCR01] R. Campo, A. Cavarra, and E. Riccobene. Simulating UML state machines.
In E. Borger and U. Glsser, editors, ASM'2001 , LNCS. Springer-Verlag,
2001. To be published.
[FDR94] J. C. Fabre, Y. Deswarte, and B. Randell. Designing secure and reliable applications using fragmentation-redundancy-scattering: an object-oriented
approach. In PDCS 2: Open Conference, pages 343-362, Newcastle-uponTyne, 1994. Dept of Computing Science, University of Newcastle, NE1
7RU, UK.
[GAS99] Stefanos Gritzalis, George Aggelis, and Diomidis Spinellis. Architectures
for secure portable executable content. Internet Research, 9( 1):16-24,
1999.
D. Gollmann. Computer Security. J. Wiley, 1999.
[Go199]
Gong.
Security
Architecture
(JDK1.2).
[Gon98] Li
http://java.sun.com/products/jdk/1.2/docs/guide/security/spec/securityspec.doc.html, October 2 1998.
[BV99]

[Gon99]

[GSG99]

Li Gong. Inside Java 2 Platform Security - Architecture, API Design, and

Implementation. Addison-Wesley, 1999.
Stefanos Gritzalis, Diomidis Spinellis, and Panagiotis Georgiadis. Security
protocols over open networks and distributed systems: Formal methods
for their analysis, design, and verification. Computer Communications
Journal, 22(8):695-707, 1999.

[HKK00]

Manfred Hauswirth, Clemens Kerer, and Roman Kurmanowytsch. A secure execution framework for Java. In ACM conference on Computer and
communications security, 2000.

(HuB01]

H. HuBmann, editor. Fundamental Approaches to Software Engineering (FASE/ETA PS, International Conference) , volume 2029 of LNCS.
Springer-Verlag, 2001.

[Jr01a]

Jan Jrjens. Developing secure systems with UMLsec - from business

processes to implementation. In VIS 2001. Vieweg-Verlag, 2001. To appear.

[Jr 01b]

Jan Jiirjens. Encapsulating rules of prudent security engineering. In International Workshop on Security Protocols, LNCS. Springer-Verlag, 2001.
To be published.

Secure Java Development with UML

123

[Jr01c] Jan Jurjens. Modelling audit security for smart-card payment schemes
with UMLsec. In M. Dupuy and P. Paradinas, editors, Trusted Information: The New Decade Challenge, pages 93-108. International Federation for Information Processing (IFIP), Kluwer Academic Publishers, 2001.
Proceedings of SEC 2001 - 16th International Conference on Information
Security.
[Jr01d] Jan Jrjens. Principles for Secure Systems Design . PhD thesis, Oxford
University Computing Laboratory, 2001. In preparation.
[Jr01e] Jan Jrjens. Secrecy-preserving refinement. In Formal Methods
Europe (International Symposium), volume 2021 of LNCS, pages 135-152.
Springer-Verlag, 2001.
[Jr01f] Jan Jurjens. Towards development of secure systems using UMLsec. In
HuDmann [HuDOl], pages 187-200. Also OUCL TR-9-00 (Nov. 2000),
http://web.cornla b.ox.ac.u k/oucl/publications/tr/tr-9-00.html.

[Kar00a] G. Karjoth. Authorization in CORBA security. Journal of Computer

Security, 8(2,3):89-108, 2000.
[Kar00b] G. Karjoth. Java and mobile code security - an operational semantics of
Java 2 access control. In IEEE Computer Security Foundations Workshop,
2000.
[KG98) L. Kassab and S. Greenwald. Towards formalizing the Java Security Architecture in JDK 1.2. In European Symposium on Research in Computer
Security (ESORICS), LNCS. Springer-Verlag, 1998.
[LGK+99] C. Lai, L. Gong, L. Koved, A. Nadalin, and R. Schemers. User authentication and authorization in the Java platform. In IEEE Annual Computer
Security Applications Conference, 1999.
[Mea01] W. Measor. Secure byzantine agreement - design, implementation and
verification. Masters thesis, Oxford University Computing Laboratory,
2001.
[ND97] V. Nicomette and Y. Deswarte. An Authorization scheme for distributed
object systems. In I EEE Symposium on Security and Privacy, 1997.
[RSG+Ol] P. Ryan, S. Schneider, M. Goldsmith, G. Lowe, and B. Roscoe. The Modelling and Analysis of Security Protocols: the CSP Approach. AddisonWesley, 2001.
[Sam00] P. Samarati. Access control: Policies, models, architectures, and mechanisms. Lecture Notes, 2000.
P. Stevens and R. Pooley. Using VML. Addison-Wesley, 2000.
[SP00]
J. Saltzer and M. Schroeder. The protection of information in computer
[SS75]
systems. Proceedings of the IEEE, 63(9):1278-1308, September 1975.
R. Sandhu and P. Samarati. Access control: Principles and practice. IEEE
[SS94]
Communications, 32(9), 1994.
[UML01] UML Revision Task Force. OMG UML Specification v. 1.4 (draft). OMG
Document ad/01-02-14. Available at http : //www.omg.org/uml, February
2001.
[Var95] V. Varadharajan. Distributed object system security. In H.P. Eloff and
S.H. von Solms, editors, Information Security - the next Decade, pages
305-321. Chapman & Hall, 1995.

124

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

[VH96]

V. Varadharajan and T. Hardjono. Security model for distributed object

framework and its applicability to CORBA. In 12th International Information Security Conference IFIP SEC96, 1996.
D. Wallach and E. Felten. Understanding Java Stack Inspection. In IEEE
Security and Privacy, 1998.

[WF98]

SECURITY THROUGH ASPECTORIENTED PROGRAMMING

Bart De Win, Bart Vanhaute and Bart De Decker
K.U.Leuven, Departement of Computer Science
Celestijnenlaan

200A,

B-3001

Heverlee,

Belgium

[email protected]

Abstract

Since many applications are too complex to be solved ad hoc, mechanisms are being developed to deal with different concerns separately.
An interesting case of this separation is security. The implementation
of security mechanisms often interacts or even interferes with the core
functionality of the application. This results in tangled, unmanageable
code with a higher risk of security bugs.
Aspect-oriented programming promises to tackle this problem by offering several abstractions that help to reason about and specify the
concerns one a t a time. In this paper we make use of this approach
to introduce security into an application. By means of the example of
access control, we investigate how well the state of the art in aspectoriented programming can deal with the separation of security from an
application. We also discuss the benefits and drawbacks of this approach, and how it relates t o similar techniques.

Keywords:

aspect-oriented programming, security, separation of concerns

1.

Introduction

In the open world of the Internet it is very important to use secure applications, servers and operating systems in order to avoid losing valuable
assets. According to different sources (e.g. CERT [cer, 2001]) updating
and patching these systems to fix security holes is necessary frequently.
The fact is that writing a secure application in an open, distributed environment is a far from straightforward task. There are several reasons
why this is so hard to achieve.
First, securing an application is a very complex matter and requires a
thorough understanding of what can go wrong and might be exploited.
An average application programmer has not enough expertise in this
area to know the exact requirements for his specific case. Moreover,

126

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

compared to the core functionality of an application, security is often

considered as less important and as such it is only added afterwards.
This results in overlooking some details which might introduce security
holes. Furthermore, in the supposition that the security requirements
are clearly understood and known in advance, it is necessary to implement them in the correct way. Several examples exist (e.g. the Netscape
random generator bug [Demailly, 1996]) where a small weakness or bug
in security related code could bring down the security of the whole application.
A second important reason why applications are difficult to secure is
the structural difference between an application and the required security solution. Confidentiality for instance requires sealing and unsealing
of sensitive information. Although they are logically joint and in fact
very similar, they are typically separated in the implementation. Contrary to the application logic, security deals with principals that use
particular services of the application and by doing so exchange sensitive information. Clearly, this information is scattered throughout the
functionality of the application. This structural mismatch often leads
to duplication of security code over different locations. Management of
software is complicated considerably in this way, which unfortunately
often introduces security problems.
To solve these problems, this paper uses aspect-oriented programming
to implement security. The separation of concerns offered by this technique allows a programmer to only focus on the core functionality of his
application. At the same time, a security engineer can analyze the security requirements and add these to the application without difficulty.
As an extra advantage, the security requirement implementations can be
reused for other applications when properly designed. Moreover, since
the security related aspects are separated from the actual application,
there is no reason to fear losing the overall security picture and as such
forgetting important issues.
The structure of this paper is as follows. We will first give a short
introduction to AspectJ, an aspect-oriented programming language for
Java, and explain how this can be used to secure an application by
means of a concrete example. This mechanism will be generalized in
order to construct a framework of security aspects, after which the advantages/disadvantages of the approach will be discussed. We end this
paper with a section on related work where we compare the aspectoriented approach with other existing techniques.

Security through Aspect-Oriented Programming

2.

127

Security as an aspect

For many problems, the modularity as offered by Object-Oriented

programming is not powerful enough to clearly capture some important
design decisions, which results in code that is scattered throughout the
program. As a short introduction to aspect-oriented programming, we
will briefly discuss one particular system, AspectJ [asp, 2001]. This is
a Java language extension to support the separate definition of crosscutting concerns. In AspectJ, pointcuts define specific points in the
dynamic execution of a java program. Pointcut definitions are specified
using primitive pointcuts such as the reception of a method call, the
creation of a specific type of object, etc. Primitive pointcuts can be
combined using logical operators.
On pointcuts, advice can then be defined in order to execute certain
code. AspectJ supports before and after advice, depending on the time
the code is executed. In addition, around advice enables the combination
of the former and the latter into one advice. The use of the pointcut
and advice constructs will become clearer when we discuss a concrete
example.
The definition of pointcuts and the specification of advice on these
pointcuts together form an aspect definition. Besides these special constructs1, an aspect is similar to a class and can as such contain data
members, methods, etc. To deploy the aspects in a concrete application, AspectJ provides a special compiler that parses all application and
aspect code and transforms them into normal Java code, which is then
compiled using a standard Java compiler.
The technique of aspect-oriented programming helps us considerably
with the problems described in the introduction. On the one hand, it
provides a mechanism to combine separate pieces of code easily, which
encourages the separate implementation of non-functional issues like security. Using this divide and conquer strategy, the overall complexity of
the problem is reduced considerably. Moreover, it allows different specialists (e.g. an application engineer, a security engineer, . . . ) to work
simultaneously and to concentrate on their field only.
On the other hand, security concerns are often interwoven throughout the application. Aspect-oriented programming is particularly aimed
at these crosscutting concerns. It enables interweaving these concerns
into the application based on particular rules and automates as such a
difficult task that is normally performed by the programmer manually.
Hence, it eases the reflection about logically joint, but physically distinct
parts.

128

2.1.

ADVANCES IN N E T W O R K AND DISTR. SYSTEMS SECURITY

An example: Access Control

The example presented in this section discusses how to perform access control in an application. We have chosen this problem because it
very clearly shows that security related code can be separated from the
functionality of the application in an elegant way.
Basically, access control can be described as follows: at a certain
point, the application asks the user to authenticate himself, after which
it can allow/deny access based on his identity. However, this abstract
view hides several difficult details. The key to turn the above description
into an aspect-oriented application is the identification of the important
domain concepts and their mutual dependencies.
First, what is the exact entity that has to be authenticated? From a
user-oriented view the user of the global application might be a reasonable decision here. In this case the user has to login once, after which this
identity is used during the rest of the application. However, the granularity of this approach will clearly not suffice for some applications, e.g.
a multi-user or a multi-agent system. A second approach consists of
linking the identity to a certain object in the application. Here, login information will be reused as long as the actions are initiated by the same
object. However, the identity of the user might change over time. It is
then necessary to associate the identity with the initiator of a certain
action. In this case, an authentication procedure is required every time
the specific action is initiated.
Next, for what resources do we want to enforce access control? Again,
one can think of different scenarios. An identity might require access
to one resource instance (e.g. a printer). When more instances are
available, one could have access to the whole group or to only a particular
subgroup. In case of different resource types the identity could require
access to a specific combination of these resources. In general, this will
often correspond to a combination of (some parts of) application objects.
A last but not less important consideration deals with specifying
where the two previous concepts meet each other. This path from the authenticated entity to the resources is necessary to pass login information
to the access control mechanism. In a distributed system for instance,
authentication and access control might be performed on different hosts.
In that case, authentication information must evidently be passed to the
access control mechanism in order to ensure correct execution. One obvious example of such access path is the invocation of a specific service
of a resource.
Each of the above concepts (identity, resource and access path)is actually a crosscutting entity to the application and maps closely to an

Security through Aspect-Oriented Programming

129

aspect. In fact, the three concepts capture the conceptual model of

access control and they can as such be used for every access control
problem. Note however that we did not discuss any issues concerning
concrete mechanisms for authentication and access control. Although
certainly relevant, it is important to realize that these are implementation decisions and they will as such depend on the underlying security
architecture.
Figure 1 shows the details of one particular case of access control,
where each object is authenticated once and where access is checked
for each invocation of a particular service. The implementation of the
other discussed access control mechanisms would be fairly similar. The
observant reader will notice that the aspect code is written for a minimal
application that consists of a Server implementing a Serverlnterface
with a method service and a Client invoking this service.
The Identification aspect is used to tag the entities that must be authenticated. In this case, every object of the class Client is considered as
a possible candidate. Furthermore, the aspect implementation contains
a field Subject that is used to store the identity information. As such,
this information will be available as if it were glued to the particular
Client object.
The authenticationcall pointcut of the Authentication aspect specifies all places where the service method of the ServerInterface is invoked. Through the use of cflowroot the Authentication aspect travels
along with the invocation. Before the method is actually invoked, the
identity information from the Identification aspect is copied to a local
field of this aspect. As such, it is able to pass the authentication information to the access control mechanism. If the Client was not yet
authenticated, this is the right place to do this.
Finally, the Authorization aspect checks access based on the identity information received through the Authentication aspect. This check
is performed for every execution of the service method (checkedMethods
pointcut). In this example, as you can see, the login and access control
phase are written in pseudo code. The actual code will depend on the
underlying security architecture as discussed before. In our implementation, we have used the Java Authentication and Authorization Service
[Lai et al., 1999] for this purpose.
Weaving the above aspects into the application will result in a new,
more secure version of the application. In the latter, the access controlling code defined in the Authorization aspect will be executed before
every invocation of service(). At this point, the application will continue
its normal execution if access is granted, however an exception will be

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

130

aspect Identification of eachobject(instanceof(C1ient))

public Subject subject null ;

}
aspect Authentication of
private Subject subject;

eachcflowroot(authenticationCall())

pointcut

serviceRequest()

pointcut

authenticationCall()
:
hasaspect (Identification)

calls(ServerInterface,

&&

service(..))

serviceRequest()

before (Object caller) : instanceof (caller) && authenticationCall

final Identification id =Identification.aspectOf(caller);
if(id.subject == null) {
<login> ;
subject = id.subject ;

public Subject getSubject

return subject ;

}
}
aspect Authorization {
pointcut checkedMethods () : executions(* service(..))

before0 returns Object : checkedMethods() {

Authentication au = Authentication.aspectOf() ;
au.getSubject() ;
Subject subject
boolean allowed = <check access control> ;
if (allowed){
return proceed();

}
else{
throw new Exception("Access denied") ;

}
}
}
Figure 1.

Aspect code for object-based access control

thrown if the (un)authenticated entity is not allowed to do so. As such,

conventional4 use of the method service() will be restricted to certain
users depending on the security policy, just as would have been the case
by coding the access control mechanism directly into the applition code.

Security through A spect- Oriented Programming

2.2.

131

Generalization of the example

The deployment of each of the crosscutting entities described in the

previous section depends heavily on the actual type and implementation
of the particular application. For example, an email client will work
on behalf of one user, while a multi-user agenda system will want to
distinguish his users. Also, objects representing a user will clearly differ
in structure and behaviour between separate applications. In general, it
is impossible to define one set of aspects that will be applicable to all
possible applications. Therefore, a more generic mechanism is desirable
that separates the implementation of security mechanisms from these
choices.
Given the previous example you might notice that the deployment
decisions are actually contained in the pointcut definitions, which define
where and when an advice or an aspect has to be applied. For this
purpose, AspectJ provides us with the possibility to declare pointcuts
abstract and afterwards extend them to define the actual join points.
Using this mechanism, it is possible to build a general authorization aspect and redefine the included abstract pointcuts depending on a specific
application. To illustrate this technique, we have applied it to the experiment of the previous section. The result is shown in figure 2. In order
to use these generic aspects in a concrete situation, one has to extend
the abstract aspects and fill in the necessary pointcuts depending on the
specific security requirements of the application.
A major advantage of this generalization phase is the ability to reuse
the core structure of the security requirement. Since this will be similar
for every situation, it is not necessary to reinvent the wheel for every
case. It should be properly designed by a qualified person only once,
after which aspect inheritance enables easy reuse.

2.3.

Towards a framework of security aspects

For a secure distributed application, other security requirements besides authentication and authorization must be considered, such as confidentiality, non-repudiation etc. We will now briefly describe how they
could be implemented using aspects.
Encryption of objects is required for confidentiality and integrity. This
is a quite straightforward task using the Java JCA/JCE [Gong, 19981.
Two issues have to be considered. First, one has to decide where and how
to insert this into the application. One possibility is to encrypt objects
while they are written to a specific stream. For this case, the stream
can be wrapped by a specific encryption stream. Another possibility is
to encrypt objects whenever they are serialized. Therefore, the readOb-

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

132

abstract
aspect
Identification
of
eachobject(entities())
abstract
pointcut
entities()
;

public Subject subject null ;

}
abstract
aspect
Authentication
private Subject subject;
abstract

pointcut

eachcflowroot(authenticationCall())

of

serviceRequest()

...
}
abstract aspect Authorization {
abstract pointcut checkedMethods()

Figure 2.

Generalized aspect code for access control

ject() and writeObject() methods of the object should be overridden to

include encryption here. Second, there is the issue of how to get or store
the cryptographic keys. Similar to the identity in the previous section,
one has to find some entity in the application with which the keys will
be associated. The implementation will vary according to how the keys
are to be acquired.
Non-repudiation requires the generation of proof for certain events in
the system, e.g. the invocation of a specific method. This is quite similar
to the problem of access control described above. One crosscutting entity
defines the identity that wants to generate the proof. Another entity
stores and manages these proofs. And finally, a third entity defines
where and how proofs should be generated and passed along.
In the end, a combination of all the security aspects could form the
basis of an aspect framework for application security. This framework
will consist of generalized aspects for each of the security requirements.
Note that several aspect implementations, depending on different underlying security mechanisms, may be included for the same security
requirement. The deployment of the framework for a concrete application will then come down to choosing the necessary aspects and defining
concrete pointcut designators for them. This technique actually suggests three distinct tasks to develop a secure application : build the
application, develop a generic security aspect architecture and specify

Security through Aspect- Oriented Programming

133

the aspect deployment pointcuts. A more elaborate discussion on this

security framework can be found in [Vanhaute et al., 2001].

3.

Discussion

Being able to specify security concerns in a separate way, and still

having them applied throughout the whole application is a noble goal.
Aspect oriented programming techniques hold a promise of achieving
this goal.
Although the technology has not yet fully matured, the current possibilities of AspectJ already give us a number of interesting advantages:
Security should be applied at all times, if it is to be applied correctly. By looking at the definition of the pointcuts in the aspect
that implements that particular security concern, a security engineer immediately knows all the places where this concern will be
used, given that the AspectJ compiler does its job correctly.
The implementation of the security mechanisms does not have
to be copied several times. All the implementation code can be
gathered within a small number of advices, perhaps all within one
source file. As a result, when changes have to be made, the programmer can focus on that one part.
Another result of the separate specification of the aspect code is
that management of the different packages of an application is
easier. There can be separate packages for pure application functionality, one for pure security code and a package that defines the
points where the security code is to be applied.
Although it could be argued that simple text substitution tools
would also be able to insert code in a generic way into a program,
the aspect oriented approach has much less chance of introducing
bugs. The constructs aspect oriented transformers work on, are
not mere text elements, but language constructs. These map more
naturally onto the entities a security policy would speak about.
By making the aspects more generic with respect to an application,
we obtain a good combination of both application independent implementations of security concerns and the use of these implementations within the context of a specific application. The support for
abstract pointcuts within AspectJ makes it possible to specify the
two in separate files.

For our work, the use of the current version of AspectJ (0.7) has also
some drawbacks (see further). On the one hand there are some technical

134

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

issues regarding the current implementation of AspectJ. It is expected

that these will be solved in later releases of the tool. On the other hand
there are problems that are more fundamental in nature. For this, at
least a redesign of the aspect-oriented tool is needed.
rn

For each method call that has some security concern, the AspectJ
compiler will insert one or more extra calls. Therefore, the generated code is less efficient, and introduces more overhead than a
direct implementation would have. Unfortunately, this is the price
to pay for the genericity of our approach. However, it is certainly
riot, worse than some other systems discussed in the next section.
Building a more complex, but less general aspect combination tool
could solve this.
If not all code in the application can be trusted, one has to be very
certain the generated code does not add any security holes. For
instance in the case of authorization: it should not be possible for a
client to call the end-functionality of a server through some other,
by the aspect tool generated method, in order to circumvent the
authorization checks. This means the security implemeriter has to
have a very clear idea of what and how exactly the aspect tool
produces. At this moment, the output of the AspectJ compiler
cannot be trusted yet, because the original functionality, without
the new aspect code, is only moved into a new method with a
special name. However, this is only a problem if not all source code
is under the control of the AspectJ compiler. The fact that AspectJ
is not a formal proven language only increases this problem. This
might become one of our topics for future work.

rn

Another issue related to the generated code comes up when debugging. As the runtime code does not directly correspond to the code
the programmer wrote, it can be harder to figure out what is going
on. The aspect-oriented research community is a t the moment trying to build better support tools that would help the programmer
in relating runtime events with the source code it stems from.

rn

The implementer of the security code still has to have very detailed
knowledge of security mechanisms, their strong and weak points,
how to implement them. As AspectJ is a generic tool, it does not
help the programmer here, apart from providing a better modularization of the problem. However, this is not a particular problem
of AspectJ, but rather of our approach to the problem.

Security through Aspect- Oriented Programming

4.

135

Related work
There are already a large number of security architectures proposed

or implemented in Java, e.g. [De Win et al., 2000]. Sun for one has for

instance released JAAS [Lai et al., 1999] for authentication and authorization, SSE for secured network communication, and there are proposals
for a secure RMI implementation. These will often already realize the
intended result, and can therefore be used in the implementation of the
security aspects. The combination of existing technologies with aspectoriented programming is not expected t o pose severe problems. The
added value of aspects in this case is the possibility to have a much
more flexible security policy, and this at a granularity that corresponds
better with the application, i.e. at the level of method calls and objects.
Some of the proposed architectures also have a fine granularity, but the
configuration and mapping onto what happens inside an application can
be fairly difficult.
By using a number of object-oriented design patterns [Gamma et al.,
1994], the existing security architectures also try to be independent of an
application structure, and they all succeed in this to some degree. The
drawback of this design is that the structure of the solution becomes
more complex and harder to understand. With an aspect-oriented approach these implementations can be designed in a more natural way.
Transformations in AspectJ happen on the level of source code. Other
tools are available that work on the level of byte code [Cohen et al.,
1998, Keller and Holzle, 1998]. This has the advantage that you can
add your own aspects even when no source code is available for the
application. The disadvantage is that on the level of byte code, a lot of
the application logic is already lost. Reconstructing this is often hard,
and giving correct descriptions of how a series of byte codes has to be
changed to for instance implement authentication will be even harder.
Checking and debugging the result will also be difficult.
There is also research into a more declarative description of security
properties for an application [Evans and Twyman, 1999, Hagimont and
Ismail, 1997]. This corresponds to an aspect-oriented language of the
first type of section 2. The real challenge here is to think of the right
abstractions the description will consist of. This is not at all an evident
matter, certainly if a goal is to be generic. We think it is better to first
experiment with a generic aspect-oriented language as described in this
paper. From these experiments, we would hope to distill the important
abstractions.
Meta level architectures [Chiba, 1995, Robben et al., 1999, Stroud and
Wue, 1996] also make it possible to separate application from security

136

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

implementation [Ancona et al., 1999, Welch and Stroud, 2000]. They

offer a complete reification of what is going on in the application: the
events of sending a message, starting the execution, creating an object
all get reified into a first class object. Because the meta-program has
control over these reified entities, it can intervene in the execution of the
base application. In comparison to aspect-oriented programming this
mechanism is much more powerful, but it is also heavier. Moreover, the
development of meta-programs for security is more complex, because the
programmer is forced to think in terms of meta-elements, which is only
indirectly related to the application.
Other approaches [Fraser et al., 1999] also use the basic idea of introducing an interceptor between clients and services, for instance to do
extra access control checks. They are similar to meta-level architectures
in that they also intervene in the communication between client and service, but the intervention is less generic (and heavy): the interceptors
are mere decorators around the services. In simple situations, they can
be specified fairly easy, perhaps through some declarative description.
However, when more and more application state need to be taken into
account, writing decorators becomes very hard, or even impossible due
to the bounded possibilities of the declarative language.

5.

Summary

This paper presented the use of aspect-oriented programming to add

security to an application. By means of the example of access control, we first demonstrated the feasibility of this approach. In order
to construct a more generic solution, we suggested to abstract relevant
pointcuts out of the aspect implementation. This enabled us to separate the security mechanisms from the actual policy, which promotes the
reuse of the mechanism implementations. After briefly discussing some
other security requirements, we touched upon the feasibility to build a
security aspect framework. Finally, we discussed the advantages and
disadvantages of our approach.
The most important advantage of this approach is the separation of
the application and the security related code. This considerably simplifies the job of the application programmer. Moreover, the security
policies are gathered in one place, which makes it easier to check whether
all the requirements are met. Still, we think that the deployment of these
generalized aspects remains quite difficult. We would like to focus our
research in the future on this issue, for example by automating the generation of concrete pointcuts based on a simplified high level description.

137

Security through Aspect-Oriented Programming

1. AspectJ also supports other constructs like Introduction. Since they are not used
directly in this paper, we will not discuss them here.
2. From another point of view, application code might be the subject of authentication.
While the mechanism to establish the correct identity of the code originator might be different,
the overall authorization mechanism described in this paper will still be applicable.
3. cflowroot is a predefined keyword in AspectJ t h a t denotes every control flow leading
to that particular pointcut. Using this keyword, it is possible to 'follow' the invocation stack
and pass as such information from the caller to the callee.
4. By predicting the output of the aspect weaver, one might be able to circumvent this
access control mechanism under certain circumstances. We discuss this problem in detail in
section 3.
5. A similar separation of these tasks h a s been described in [Robben et al., 1999].

References
[asp, 2001] (2001). AspectJ Website. http://www.aspectj.org/.
[cer, 2001] (2001). CERT Website. http://www.cert.org/.
[Ancona et al., 1999] Ancona, M., Cazzola, W., and Fernandez, E. (1999). Reflective
Authorization Systems: Possibilities, Benifits anti Drawbacks. In Secure Internet
Programming: Security Issues for Mobile and Distributed Objects.
[Chiba, 1995] Chiba, S. (1995). A MetaObject Protocol for C++. In Proceedings of
the 1995 Conference on Object-Oriented Programming.
[Cohen et al., 1998] Cohen, S., Chase, J., and Kaminsky, D. (1998). Automatic Program Transformation with JOIE. In Proceedings of the 1998 USENZX Annual
Technical Symposium.
[De Win et al., 2000] De Win, B., Van den Bergh, J., Matthijs, F., De Decker, B., and
Joosen, W. (2000). A Security Architecture for Electronic Commerce Applications.
In Information Security for Global Information Infrastructures, pages 491-500.
IFIP TC11, Kluwer Academic Publishers.
[Demailly, 1996] Demailly, L. (1996).
http://www.demailly.com/ dl/netscapesec/ .

Netscape

Security

(problems).

[Evans and Twyman, 19991 Evans, D. and Twyman, A. (1999). Flexible PolicyDirected Code Safety. In Proceedings of the 1999 IEEE Symposium on Security
and Privacy.

[Fraser et al., 1999] Fraser, T., Badger, L., and Feldman, M. (1999). Hardening
COTS Software with Generic Software Wrappers. In Proceedings of the 1999 IEEE
Symposium on Security and Privacy.
[Gamma et al., 1994] Gamma, E., Helm, R., Johnson, R., and Vlissides, J. (1994).
Design Patterns: Elements of Reusable Object-Oriented Software. Addison Wesley
Longman.

L.
[Gong,
1998]
Gong,
http://java.sun.com/security.

(1998).

Java

Security

Architecture.

[Hagimont and Ismail, 1997] Hagimont, D. and Ismail, L. (1997). A Protection

Scheme for Mobile Agents on Java. In Proceedings of the International Conference on Mobile Computing and Networking.

138

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

[Keller and Holzle, 1998] Keller, R. and Holzle, U. (1998). Binary Code Adaptation.
In Proceeding of the 1998 European Conference on Object-Oriented Programming.
[Lai et al., 1999] Lai, C., Gong, L., Nadalin, A., and Schemers, R. (1999). User Authentication and Authorization in the Java Platform. In Proceedings of the 15th
Annual Computer Security Applications Conference.
[Robben et al., 19991 Robben, B., Vanhaute, B., Joosen, W., and Verbaeten, P.
(1999). Non-functional Policies. In Cointe, P., editor, Meta-Level Architectures
and Reflection, volume 1616 of Lecture Notes in Computer Science, pages 74-92.
Springer-Verlag.
[Stroud and Wue, 19961 Stroud, R. and Wue, Z. (1996). Using Metaobject Protocols
to Satifsy Non-functional Requirements. In Advances in Object-Oriented Metalevel
Architectures and Reflection.
[Vanhaute et al., 2001] Vanhaute, B., De Win, B., and De Decker, B. (2001). Building Frameworks in AspectJ. ECOOP2001 Workshop on Advanced Separation of
Concerns.
[Welch and Stroud, 2000] Welch, I. and Stroud, R. (2000). Using Reflection as a
Mechanism for Enforcing Security Policies in Mobile Code. In Proceedings of the
Sizth European Symposium on Research in Computer Security.

EXTENDING A CAMPUS NETWORK

WITH REMOTE BUBBLES USING IPSEC
Aurlien Bonnet
Universit catholique de Louvain, Dept INGI
Place Ste-Barbe, 2, 1348 Louvain-la-Neuve, Belgium
[email protected]

Marc Lobelle
Universit catholique de Louvain, Dept INGI
Place Ste-Barbe, 2, 1348 Louvain-la-Neuve, Belgium
[email protected]

Abstract

There is an increasing demand for high speed remote connections

(e.g. ADSL, cable, etc.) to university networks both from students
and staff members. In many instances, the need is to connect not a
single computer but a remote subnetwork and to be able to have not
only clients but also servers on the remote subnetwork (for instance to
access files on computers on the remote network from workstations in
the university). We present here a solution to meet these needs through
a simple dial-up-like connection to which the access provider allocates
only a single temporary I P address t o a single remote machine. The
solution allows remote networks to be started anytime like little bubbles
and be integrated dynamically into the big bubble that is the university
network. It is based on the use of IPSec, DHCP and address translation.
Beside providing confidentiality, this allows allocating dynamically I P
addresses in the range of the university to computers in the remote
bubble, and binding them to permanent DNS names in the domain
of the university. After configuration, the computers in the remote
subnetwork appear as if they were located inside the university. The
solution can, of course, be applied to any organization.

Keywords: Remote networks, Virtual private networks, IPSec, DHCP, NAT, RSIP.

140

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Introduction: Remote access to university networks

New learning technologies intensively use computers and computer
networks. Universities have to provide an increasing number of workstations and personal computers to their students. All these computers are
connected to the university network, and through this, to the Internet.
On the other hand, an increasing number of students own or have
access to a computer at home or in their residence near the university.
In many instances, more than one computer is available: when a new
computer is bought, the old one is rarely discarded, but it is rather restricted to the less demanding applications (E-mail, web access, word
processing), so that several members of a household can be simultaneously working on computers. In multi-room flats used as student accomodation, several computers are also often available.
However, many learning tasks cannot be performed at home because
they require permanent or quasi permanent access to servers of the university, or simply to the Internet. Other activities are only possible
with computers that are part of the university network, because of, for
instance, software licensing restrictions.
The availability of high speed remote connections (ADSL, Cable)
makes it possible to have at home the same kind of access speed as
in ones office or in a room of workstations. However, all the other restrictions of dial-up connections remain since remote users get only one
dynamically allocated I P address in the range of the provider.
I n this paper, we shall discuss how to connect a remote subnetwork
when the provider allocates only one dynamic IP address to it. The
purpose is to make the remote subnetwork, called remote bubble, really
look like part of the university network, with IP addresses belonging to
this network. These addresses are centrally and dynamically allocated
(central and dynamic allocation avoids wasting IP addresses). A s f a r as
we know, this problem has never been addressed before.

1.

Connection of a subnetwork through a single

IP address: state of the art

There are different solutions to connect subnetworks through a single

IP address. The first is to use NAT/PAT (Network Address Translation
and Port Address Translation); software as well or hardware products
(soho routers) implementing this solution are available. They are often based on Linux I P masquerading. Other solutions use tunneling
protocols to build a Virtual Private Network.

Extending a Campus Network with Remote Bubbles using IPSec

Figure 1.

1.1.

141

A VPN across the Internet

IP Masquerading

I P Masquerade is a Linux networking function similar to one-to-many

NAT (Network Address Translation) found in many firewalls and network routers. For example, if a Linux host, called the gateway, is connected to the Internet, the I P Masquerade feature allows client applications on other computers connected to the internal interface of this
gateway and using non-routable I P addresses to reach the Internet.
This system allows a set of machines to invisibly access the Internet
through the gateway. To other machines on the Internet, all this traffic
will appear to be from or to the gateway (for more information see
[Ran00]). Not even all client applications work with this scheme (e.g.
FTP). It is therefore often complemented with specified application level
gateway programs.

1.2.

Building Virtual Private Networks

The problem that VPNs are trying to solve is that of letting two
networks communicate securely when the only connection between them
is over a third network which they dont trust. VPNs use a gateway
between each of the communicating networks and the untrusted network.
Most of the current VPN packages use tunneling to create a private
network. The principle of tunneling is to encapsulate a packet within a
packet.
The gateway machines can encrypt packets entering the untrusted net
and decrypt packets leaving it, creating a secure tunnel through it (see
figure 1).

142

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

1.2.1
Simple tunneling protocol.
Most of the current operating system can enable simple tunnels between two gateways (without
any authentication or encryption: the tunnel is thus not secure). This
system is very simple. Gateways on each network encapsulate packets
destined to the distant network in a packet destined to the remote gateway. But this solution is limited to static IP addresses on gateways. In
our context, IP addresses of gateways will be dynamically allocated by
the provider. So we need to authenticate these gateways dynamically
and by other mean than its IP address.
1.2.2
The IPSec Protocol and its use in Virtual Private
Networks
IPSec is a mechanism for adding security to IP and up-

perlayer protocols. It can protect traffic between hosts, between network

security gateways (routers, firewalls,..) and between hosts and security
gateways. IPSec hosts and gateways are authenticated by cryptographic
technics independantly of their IP addresses, which can be allocated
dynamically.
IPSec defines two different protocols: The Encapsulating Security
Payload (ESP), and the Authentication Header (AH). AH provides proofof-data-origin, data integrity and antireplay protection on received packets. ESP provides, in addition, data confidentiality and limited traffic
flow confidentiality.
More information on the IPSec protocol, can be found in [DH99],
[RFC2401], which defines the base architecture upon which all implementation are built, [RFC2402] which explain AH functionalities, and
[RFC2406] which talks about ESP.
The VPN can be built by deploying IPSec gateways. The protected
network to which access is controlled is on one side of the gateway; the
unsafe and unsecured network (usually Internet) is on the other. IPSec
must be used in tunnel mode between the gateways, because the VPN
is protecting the traffic between two different networks.

Using IPSec VPNs to connect remote

bubbles to a university network

2.

We want to integrate computers at home (or in student residences)

transparently in the university network. The homes or the residences
are connected to a provider and obtain from it a single temporary IP
address: this I P address may be different each time they connect. Several
uses of this single temporary I P address are possible:
m

One single computer is connected to the modem (ASDL/Cable),

and uses the temporary I P address. This solution is the easiest,

Extending a Campus Network with Remote Bubbles using IPSec

143

DSL Modem

a.b.c.d Provider address

Figure 2.

Single connection with a DSL Modem

Real network prt with

provider IP address

university 1P address

Figure 3.
computer

Using IPSec to add an address of the university network to a remote

DSL Modern

Figure 4.

Architecture of a NAT network

144

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

VPN Terminatiir

I
,

.-.i

DSL Modem

130 :04 x y

Figure 5.

Remote bubble with IPSec

but the computer appears with a provider address, not an address

of the university network, as requested (see figure 2). To overcome
this problem, an internal address of the university network could
be allocated to a virtual interface of the computer and routed
through an IPSec VPN across the provider network to a gateway
in the university (See figure 3). This is actually a particular case
of the third situation.
rn

One computer is a gateway, uses I P masquerading, and connects a

subnetwork to the Internet (a dedicated NAT box is equivalent to
this. See figure 4). This solution allows to connect more than one
computer through a connection with a single routable I P address.
It is integrated in Linux, free, and easy to build (several companies market black-boxes with this kind of functionality), but the
machines in the subnetwork are not accessible from the Internet
(they can only be clients, not servers), and some protocols are not
compatible with it (H323, IPSec,...).

rn

One virtual private network is set up between a gateway to the

remote subnetwork and a gateway to the university. This solution
doesnt impose any constraint on the address allocation policy in
the remote subnetwork: the computers of the remote subnetwork

Extending a Campus Network with Remote Bubbles using IPSec

145

can have routable IP addresses. This solution is the only one

allowing remote bubbles to merge transparently in the university
network (see figure 5). Thats why we will develop this solution in
this section.

2.1.

Requirements of the V P N

The VPN used to connect the remote subnetwork to the university

network must meet the following requirements:
We want to be able to decide from within the university what
addresses will be allocated to the computers on the subnetwork.
These addresses may be non routable or routable (for instance
taken from the university range).
We want also to be able to allocate these addresses dynamically
but to be bound to permanent domain names. We will not discuss
this last aspect in this paper.
We want the remote network to merge automatically in the university network when the gateway is started. The computers on
the remote network must then be undiscernible by third parties
from computers located inside the university.
We want the same security beween computers in the remote network and other computers in the same department of the university
as between computers in this department.

2.2.

SubNetwork with IPSec Gateways

Using an IPSec tunnel between the gateway to the remote subnetwork

(lets call it Hawser) and a gateway in a department of the university
(lets call it Bollard) already allows to meet all but the second of the
above requirements (see figure 6).
The computers of the subnetwork are logically neighbours of the
other end of the tunnel.
rn

IPSec is a standard protocol available for any decent platform.

rn

IPSec authentication of the gateway is not based on its I P address

that can be dynamically allocated by the provider.

rn

IPSec provides the required security. If confidentiality is needed,

ESP can be used, otherwise AH is sufficient. If security is definitely not a problem (for instance if the ADSL links are connected

146

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure 6.

Architecture of the solution

directly by the provider to the university through a VPN instead

through the public Internet), then AH with null encoding can be
used. The choice is a matter of balance between security and performance. Benchmarks will tell us if providing less than maximum
security makes any difference from a performance point of view.
Meeting all these requirements is not straightforward. Let us look
more closely at what should happen when Hawser boots:
8

Hawser boots.

rn

Hawser logs into the provider network and gets a dynamic provider
address from, say, a radius server.

rn

Hawser calls Bollard and sets up an IPSec tunnel.

All the packets destined to the subnetwork will be routed through

Bollard and Hawser, and all external packets from the subnetwork will
be routed through Hawser and Bollard.

3.

Dynamic allocation and address translation

The Dynamic Host Configuration Protocol (CHCP) automates the

process of configuring devices on I P networks. DHCP performs many of

Extending a Campus Network with Remote Bubbles using IPSec

147

the functions a network administrator could carry out manually when

connecting a new computer to a network (see [DL99]). With DHCP relay
agents, remote machines can also be configured. We decided to use the
DHCP protocol with relay agents to configure the differents subnetworks
for differents reasons:
addresses can be leased temporarily when needed which simplifies
network administration of nomadic computers (laptops),
rn

subnetworks can be created without any administrative overhead

for address allocation,

the network configuration of the computer is easier (most of the

parameters are transmitted by the protocol),

rn

the DHCP protocol is available on many operating systems.

3.1.

The relay agent

A relay agent is designed to forward DHCP messages between clients

and server when the server and the client are not in the same network.
When the relay agent receives a client message, it forwards it to the
server, and when the server answers to a request, it also uses the relay
agent to contact the client. In our solution, the relay agent runs on
Hawser, and the DHCP server can run on any machine of the same
network as Bollard, but it is easier to put it directly on Bollard.

3.2.

DHCP configuration

As specified in [RFC2131], the technique is to assign statically a range

of addresses (subnetwork class) to each network behind a relay agent.
This solution would be the easiest to deploy but it wastes a lot of IP
addresses.
With a modified relay agent, another solution would be to assign
addresses to the devices on the different subnetworks without regard for
their localization. This solution is more economical in IP addresses, but
it is more difficult to deploy, since routes must be explicitly configured
for each individual device.
When a device asks for a new DHCP configuration, a dedicated IPSec
tunnel must be opened between Bollard and Hawser for this new IP
address. This will be done by the relay agent, as it has all the required
informations.

148

3.3.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Address translation

The solution proposed in the previous paragraph has a problem. All

the packets sent by a device on a subnetwork will be routed through
Bollard and Hawser, even the packets destined to same remote bubble,
as the differents devices are not considered to be on the same virtual
subnetwork. The solution we use to avoid this problem is the address
translation (More information available on [Dut0l]).
Instead of sending address like a.b.c.d, the DHCP server will send
address like 10.b.c.d with a class A subnetwork mask. Thus all devices
in the same remote bubble can see each other. Hawser will have to
NAT (translate address) l0.b.c.d and a.b.c.d for incoming and outgoing messages. Devices in other bubbles are known with their addresses
allocated and not 10.b.c.d.

3.4.

Summary

When a new computer is started on a remote bubble:

rn

the computer boots.

rn

it broadcasts a DHCP request.

rn

Hawser forwards the request to Bollard (assuming it is the DHCP

server).

rn

Bollard replies to the request with the address l0.b.c.d

rn

Hawser forwards the reply to the computer, starts address translation between 10.b.c.d and a.b.c.d, and enables the tunnel for the
address a.b.c.d

rn

the computer is connected to the network.

When a computer uses the network:

rn

if the address is local (10.x.x.x) it connects directly the specified

computer, else it sends the packet to Hawser, which it considers
as its default router.

rn

Hawser exchanges the source address (10.b.c.d to a.b.c.d)

Hawser sends the new packet to Bollard through the tunnel.

Bollard routes the packet it receives like a normal packet.

When a computer wants to connect or reply to a computer on a remote
bubble:

Extending a Campus Network with Remote Bubbles using IPSec

149

the packet is sent to Bollard.

Bollard sends it to the Hawser concerned (several remote bubbles

can be attached to the same Bollard via different Hawser) through
a tunnel.
Hawser translates the destination address (a.b.c.d to l0.b.c.d).
Hawser forwards the packet to the computer concerned in the remote bubble.

4.

Related Works

We did not find any other published paper on the problem of setting
up a remote subnetwork connected with a single temporary address allocated by a provider and using addresses belonging to a main network
and dynamically allocated by a server in this network. However, results
has been published about closely related problems.
Realm Specific IP (RSIP) is a new protocol, designed as an alternative
to NAT that preserves end to end packet integrity. RSIP allocates on
demand in a network A, addresses belonging to another network B. The
gateway between networks A and B has a pool of IP addresses of B and
will allocate them to hosts in A (called RSIP hosts) when requested.
Routes or tunnels must be provided in A for these RSIP hosts. They
can be considered as if they were really on the network B.
The advantage of this solution over NAT is that nothing in the packet
is modified. The drawback is that a driver is needed in each RSIP host
of the subnetwork.
RSIP was designed to solve a different problem than the one addressed
in this paper. However, it provides a similar service with different mean.
The main diference of the service provided by RSIP and the service
provided in our solution is that, since the RSIP server is on the gateway,
the range of addresses allocated to each remote subnetwork must be
allocated statically. Since our solution uses a central server, addresses
can be distributed on demand to several remote networks, which is a
more efficient allocation technique.
On the other hand, RSIP preserves the end to end integrity of packets
while our solution uses NAT coupled with DHCP which means that
packets are modified in the path. The price RSIP has to pay for this
advantage is to add some software to each host in the subnetwork.
More Information on RSIP can be found in [RSIP-PROTO].

150

ADVANCES IN N E T W O R K A N D DISTR. SYSTEMS SECURITY

5.

Conclusion

The paper identifies a new problem, which is to make a remote subnetwork, called remote bubble, really look like merged in another network,
with I P addresses belonging to this network centrally and dynamically
allocated.
It proposes a solution based on existing protocols (IPSec, DHCP,
NAT). Coupled DHCP relay and NAT protocol engines provide the hosts
of the remote bubble with addresses from the main network. IPSec allows the gateway itself to have a dynamic I P address and hide the contents of packets on the way between the remote bubble and the main
network. The situation of the hosts in the remote bubble is exactly the
same (except for communications delays) as if they were located inside
the main network.
The solution we propose presents many advantages. In addition to
secure communication, with the creation of an IPSec tunnel between
the remote bubble and the university network, its dynamic configuration
allows to manage easily many bubbles of any size, without wasting I P
addresses.
The administration of a bubble is quite simple: all the computers
plugged behind the gateway (Hawser) will be automatically on the university network.

Acknowledgements
This work has been performed in the framework of the SIN project
(Secure I P Networking) supported by grant FIRST 981/3749 by MRWDGTRE (Travaux raliss avec laide du Ministre de la Rgion Wallonne, en application du programme de formation et dimpulsion la
recherche scientifique et technologique ).
The authors express their thanks to the anonymous reviewers for their
useful comments.

References
[Ran00] Ranch, D. (2000). Linux IP Masquerading H O W T O . Technical report.

http://www.linuxdoc.org/HOWTO/IP-Masquerade-HOWTO.html.
[DH99] Doraswamy, N.; Harkins, D. (1999). IPSec :The New Security Standard for
the Internet, Intranets and Virtual Private Network. Prentice Hall PTR.
[RFC2401] Kent, S.; Atkinson, R. (1998). Security Architecture for the Internet Protocol. Request for Comments 2401. Network Working Group.
[RFC2402] Kent, S.; Atkinson, R. (1998). IP Authentication Header (AH). Request
for Comments 2402. Network Working Group.
[RFC2406] Kent, S.; Atkinson, R. (1998). IP Encapsulating Security Payload (ESP).
Request for Comments 2406. Network Working Group.

Extending a Campus Network with Remote Bubbles using IPSec

151

[RFC2131] Droms, R. (1997). Dynamic Host Configuration Protocol (DHCP). Request

for Comments 2131. Network Working Group.

[DL99] Droms, R.; Lemon, T. (1999). The DHCP Handbook. Macmillan Technical
Publishing.
[Dut01] Dutcher, B. (2001). The N A T Handbook . Wiley Computer Publishing.
[RSIP-PROTO] Borella M.; Grabelsky D.; Taniguchi K. (2000). Realm Specific IP:
Protocol Specification. Draft. Network Working Group.

This page intentionally left blank.

COMBINING WORLD WIDE WEB AND

WIRELESS SECURITY
Joris Claessens, Bart Preneel, Joos Vandewalle
C O m p u t e r Security and Industrial Cryptography (COSIC)
Dept. of Electrical Engineering - ESAT
Katholieke Uniuersiteit Leuven
http://www.esat.kuleuven.ac.be/cosic/

joris.claessens@esat. kuleuven.ac. be

Abstract

In current electronic commerce systems, customers have an on-line interaction with merchants via a browser on their personal computer. Also
payment is done electronically via the Internet, mostly with a credit
card. In parallel to this, e-services via wireless-only systems are emerging. This paper identifies security and functionality weaknesses in both
of these current approaches. The paper discusses why and how generalpurpose mobile devices could be used as an extension to PC based systems, to provide more security and functionality. General-purpose mobile devices are shown to be an alternative to costly special-purpose
hardware. This combined approach has in many cases more interesting properties than when using mobile devices only. As an example
of the combined approach, a GSM based electronic payment system is
proposed and investigated. The system enables users to order goods
through the World Wide Web and pay by using their mobile phone.

Keywords: WWW security, wireless security, m-commerce

1.

Introduction

In current electronic commerce systems, customers have an on-line

interaction with merchants via a browser on their personal computer.
Also payment is done electronically via the Internet, mostly by sending
a credit card number to the merchant. This basic system is in widespread
use today, and most people are familiar with buying books and music,
booking flights, ordering PCs, etc. There are however some important
security problems. For example, credit card numbers are often stolen by
hackers from merchants computers, orders and confirmations are usually
not digitally signed and can be repudiated afterwards. In parallel to

154

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

the fixed PC based systems, e-services are also emerging in the wireless
world. Current mobile devices have however rather limited functionality,
and in many applications, they are not suited to be used on their own.
This paper suggests a combined approach in which mobile devices are
used as an extension to the World Wide Web environment. The paper
starts with a description of the security properties of the World Wide
Web in Sect. 2, and the security features in some wireless systems, i.e.,
GSM and WAP, in Sect. 3. Section 4 discusses security and functionality weaknesses in both worlds, and suggests a combined approach. An
example of this approach is given in Sect. 5: a GSM based electronic
payment system for the WWW is proposed and investigated. Further
analysis of this system is presented in Sect. 6.

2.

World Wide Web security

There are many security issues related to the WWW. Within the
scope of this paper, we will only discuss the communications security
aspect, both at the network and the application level, and the payment
security aspect.

2.1.

Communications security

The communication between a web browser and a web server is secured by the SSL/TLS protocol. Historically, Secure Sockets Layer
(SSL) was an initiative of Netscape Communications. SSL 2.0 contains
a number of security flaws which are solved in SSL 3.0. SSL 3.0 was
adopted by the IETF Transport Layer Security (TLS) working group,
which made some small improvements and published the TLS 1.0 [8]
standard. SSL/TLS is used in this paper, as SSL is an acronym
everyone is quite familiar with; however, the use of TLS in applications
is certainly preferred to the use of the SSL protocols.
Within the protocol stack, SSL/TLS is situated underneath the application layer. It can in principle be used to secure the communication of any application, and not only between a web browser and
server. SSL/TLS provides entity authentication, data authentication,
and data confidentiality. In short, SSL/TLS works as follows: publickey cryptography is used to authenticate the participating entities, and
to establish cryptographic keys; symmetric key cryptography is used
for encrypting the communication and adding Message Authentication
Codes (MACs), to provide data confidentiality and data authentication
respectively. Thus, SSL/TLS depends on a Public Key Infrastructure.
Participating entities should have a public/private key pair and a certificate. Root certificates (the certification authorities certificates that

Combining World Wide Web and wireless security

155

are needed to verify the entities certificates) should be securely distributed in advance (e.g., they are shipped with the browsers). Private keys
should be properly protected. Note that these two elements, i.e., distribution of root certificates in browsers and the protection of private keys,
is actually one of the weak and exploited points with respect to WWW
security (see further).
More detailed information on SSL/TLS, the security flaws in SSL 2.0,
and the differences between SSL 3.0 and TLS 1.0, can be found in [27].

2.2.

Application security

SSL/TLS only protects data while it is in transit. Moreover, exchanged messages are not digitally signed. Therefore it does not provide
non-repudiation. Both customers and merchants can always deny later
on having sent or received requests or confirmations from each other.
In addition to SSL/TLS, critical messages should thus be digitally
signed before they are sent through the secure channel. The concept
of digitally signing messages is not really integrated yet in todays web
browsers. Netscape though allows the content of forms to be digitally
signed using the Javascript signText() function. XML will be more
and more used on the WWW to represent content instead of the basic
HTML. In the future, browsers are therefore expected to implement
Signed XML [10], which specifies how XML documents should be digitally signed.
Note that an alternative protocol to secure the communication on the
WWW has been proposed in the past: S-HTTP [26]. This protocol is
situated at the application layer, and is specifically intended for HTTP.
It secures HTTP messages in a very similar way to the protocols for
secure email, and provides non-repudiation. SSL/TLS has however become the de-facto standard on the web, and S-HTTP was not a success.

2.3.

Payment security

Although numerous different electronic payment systems have been

proposed that can be or are used on the WWW, including micro-payment
systems and cash-like systems, most transactions on the web are paid
using credit cards. Mostly, customers just have to send their credit card
number to the merchants web server. This is normally done securely
over SSL/TLS, but some serious problems can still be identified. Users
have to disclose their credit card number to each merchant. This is
quite contradictory to the fact that the credit card number is actually
the secret on which the whole payment system is based (note that there
is no electronic equivalent of the additional security mechanisms present

156

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

in real world credit card transactions, such as face-to-face interaction,

physical cards and handwritten signatures). Even if the merchant is
trusted and honest this is risky, as one can obtain huge lists of credit
card numbers by hacking into (trustworthy, but less protected) merchants web servers. Moreover, it is possible to generate fake but valid
credit card numbers, which is of great concern for the on-line merchants.
Thus, merchants bear risk in card-not-present transactions.
Secure Electronic Transaction, SET [29], is a more advanced standard
for credit card based payments. One of its core features is that merchants
only see encrypted credit card numbers, which can only be decrypted by
the issuers. This system is conceptually much better, but until now it
has not become popular due to its complexity.
American Express offers a `one-time credit card solution [1] with
which customers can protect their privacy, but which also solves some
of the above mentioned problems. Alternatively, several similar systems
exist (e.g., InternetCash [16]) in which customers can obtain some prepaid value identified and protected with a number and PIN, and use it
on-line in cooperation with a central server. Finally, real-life electronic
payment means (e.g., Proton [25] and debit cards) are also starting to
be deployed on the WWW (e.g., [2]).

3.

Wireless security

GSM and WAP are currently probably the two most popular and
widely used wireless technologies. They are briefly presented in the
following paragraphs. Thereafter, some other systems and initiatives in
the wireless world are discussed.

3.1.

GSM

GSM, Global System for Mobile communications, is the currently

very popular digital cellular telecommunications system specified by
the European Telecommunications Standards Institute (ETSI). In short,
GSM intends to provide three security services [32]: temporary identities, for the confidentiality of the user identity; entity authentication, that
is, to verify the identity of the user; and encryption, for the confidentiality of user-related data (note that data can be contained in a traffic
channel, e.g., voice, or signaling channel, e.g., SMS messages).
The Subscriber Identity Module (SIM) is a security device, a smart
card which contains all the necessary information and algorithms to authenticate the subscriber to the network. It is a removable module and
may be used in any mobile equipment [32]. Note that the encryption
algorithms are integrated into the mobile equipment as dedicated hard-

Combining World Wide Web and wireless security

157

ware. GSM does not use public-key cryptography. Symmetric keys are
derived from user related data using an algorithm under the control of
a master key.
The electronic payment system described in the example later in this
paper, requires the SIM to contain a small payment application, based on
the SIM Application Toolkit. The SIM Application Toolkit [13] provides
mechanisms which allow applications, existing in the SIM, to interact
and operate with any compliant mobile equipment. These mechanisms
include displaying text from the SIM to the mobile phone, sending and
receiving SMS messages, and initiating a dialogue with the user. In addition to the GSM security mechanisms, special SIM Application Toolkit
security features have been defined [ll, 12]. The security requirements
that have been considered are: (entity) authentication, message integrity, replay detection and sequence integrity, proof of receipt and proof of
execution, message confidentiality, and indication of the security mechanisms used. According to the standard, digital signatures can be used
to implement some of these requirements.
Note that the same distinction between communications security and
application security as made in the WWW security context, can be made
here: standard GSM security at the communications level, and SIM
Application Toolkit security at the application level.

3.2.

WAP

The Wireless Application Protocol (WAP) is a protocol stack for wireless communication networks. WAP is bearer independent; the most
common bearer is currently GSM.
Similar to SSL/TLS for the Internet, WTLS [39] is WAPs communications security solution. It also relies on a Public Key Infrastructure
[35, 34]. The main differences are that WTLS supports by default algorithms based on elliptic-curve cryptography, is adapted for datagram
communication (instead of connection), and supports its own certificate
format, besides X.509v3, optimized for size. TLS was as such modified
to make it more suitable in an environment where there are bandwidth,
memory, and processing limitations.
At the application layer, WAP provides digital signature functionality through the WMLScript Crypto Library [40], which is similar to
Netscapes Javascript signing. Comparable to the GSMs SIM, WAP
devices will use a Wireless Identity Module (WIM) [38] which can contain the necessary private and public keys to perform digital signatures
and certificate verification respectively.

158

3.3.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Other systems and init it iatives

GSM is a second-generation system (2G) . UMTS, Universal Mobile

Telecommunications System [31], is part of a global family of thirdgeneration (3G) mobile communications systems. These systems provide
high-capacity and more secure [33] communication. A competitor of
WAP is NTT DoCoMos i-mode [23]. Bluetooth [5] is a wireless protocol for communication between devices that are in close proximity.
The Internet itself is also expanding to the wireless world. The IETF
is currently defining standards for Mobile I P [15], and is working on
extensions (including wireless) for TLS [4].
The Mobile Electronic Signature Consortium has defined mSign [21],
which should provide a standardized interface between Primary Service
Providers (e.g., merchants) and Mobile Operators. It allows Primary
Service Providers to request signatures from end-users through the Mobile Operators. The Mobile electronic Transactions initiative - MeT
[22] - intends to establish a consistent and coherent framework for secure mobile transactions, based on existing standards and specifications;
where needed, new functionality will be submitted to relevant standardization and specification organizations. There are numerous other fora
concerned with mobile secure payments, see [7] for a description and
comparison of these.

4.

Combining WWW and wireless

Both the World Wide Web and the wireless world on their own have security and/or functionality problems. These shortcomings are explained
in the following paragraphs. An approach in which the two worlds and
their advantages are combined, is then motivated.

4.1.

WWW: problems

It is very common that only web servers have certificates with which
they are authenticated. In case user authentication is needed, it is almost
never done via SSL/TLS client authentication. Users are often authenticated via their I P address, which is vulnerable to I P spoofing [3], which
certainly does not provide mobility, and which is just not usable in an
open system. Fixed passwords are frequently used, which provide mobility, but which are vulnerable to guessing, dictionary attacks and social
engineering. Passwords that are only used once are not frequently used.
They would be more secure, but certainly less convenient.
Root certificates are needed when verifying a web server certificate.
It is very important that a user has an authentic copy of these certific-

Combining World Wide Web and wireless security

159

ates. This is more or less ensured by shipping them together with the
browsers. It is however easy to add more or even replace root certificates. Moreover, the browser trust model causes a server certificate to be
trusted if it is successfully verified by any of the root certificates (since
there is usually no central policy management, this might easily include
an attackers root certificate). Finally, browsers generally also do not
yet check if a certificate has been revoked.
If the user has a public/private key pair - for SSL/TLS client authentication, for SET, or for digitally signing documents - the private key
will mostly reside on the hard disk of the machine. Even if it is protected
by a pass phrase, it is still very vulnerable, for example due to Trojan
horses. Users with such a software token are also hardly mobile. Smart
cards are a solution, but for particular applications, they might be inconvenient. Moreover, smart card readers are currently not installed on
each machine. Other special-purpose hardware, such as a Digipass [9], as
sometimes used in e-banking, might be too costly for small applications,
i.e., the investment for the customers and/or merchants would just be
too high compared to the expected benefits.
Current end-user computing systems tend to offer more functionality at the cost of security. This is actually the reason why for example
root certificates and private keys are so vulnerable on current end-user
machines. Specifically, there is currently a lack of secure operating systems [19] and trusted components [30]. Todays PC and browser offer
advanced functionality, but are therefore an insecure environment.

4.2.

Wireless: problems

While the security problems on the WWW are currently more related
to the secure management of the end-points, the security problems in
some wireless systems are still with the protocols and algorithms themselves. For example, algorithms used by many GSM providers have been
broken and over-the-air cloning and real-time eavesdropping have been
shown (at least in theory) to be feasible [28]. Security problems have
been discovered in other mobile systems too [6, 17]. Most of these problems are due to non-public design of the algorithms and protocols, leakage and/or publication of the details to the general public afterwards,
and discovery of flaws by the cryptographic community.
More conceptually, both GSM and WAP do not offer end-to-end security. GSM security only applies on the wireless link, i.e., from mobile
phone to base station, but not from mobile phone to mobile phone. The
fixed network is considered to be secure (more precisely, GSM intends
to offer the same security level as the fixed network). In the WAP ar-

160

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

chitecture, WAP devices communicate with web servers through a WAP

gateway. WTLS is only used between the device and the gateway, while
SSL/TLS can be used between the gateway and the server. F'rom a security point of view, this means that the gateway should be considered as
a person-in-the-middle. Note that WAP is now evolving into end-to-end
security [37, 36].
Security seems to evolve in the good direction though. From a usability point of view on the other hand, mobile devices have still a rather
limited functionality. They are not performant, and have often a quite
poor human-device interface. Although mobile devices are getting more
advanced, they will always be outsmarted by desktop PCs. Note that
the complexity of the PC (e.g., multi-user operating system, data with
executable content, ...) is the main reason why securing the end-points
of the communication is such a difficult task, and remains an important
problem on the WWW. As long as mobile devices stay quite simple and
do not provide too much functionality, their security as an end-point will
be more easy to cope with.

4.3.

Motivation for a combined approach

By combining the World Wide Web with a wireless system, we want to

come to practical and low-cost electronic commerce applications, which
can fully exploit the broad functionality of the WWW. Two goals should
hereby be achieved at the same time: security and mobility.
The WWW on its own does not seem to be sufficient for these applications. It surely provides broad functionality. When for example only
fixed passwords are used, the WWW also offers mobility, i.e., a user
can initiate transactions from any computer (e.g., a public terminal).
Strong security is in that case however not achieved. Stronger security
can be achieved by using for example cryptographic keys stored on the
computer's hard disk. However, this does not allow for practical mobility. Special-purpose hardware tokens would increase the security of
the application and provide mobility again. However, in an electronic
commerce environment, consumers do not likely want to pay for a token
that can only be used in the context of that application.
Wireless systems on their own are not suitable either. By definition,
they offer mobility. Although there are some weaknesses in current systems, security in wireless systems tends to improve substantially. It is
however clear that the GSM system is a rather limited environment.
WAP offers a more general and WWW-like functionality, but in practice today's devices and networks do not satisfy the needs of merchants

Combining World Wade Web and wireless security

161

and customers. Mobile devices are generally expected to stay inferior to

desktop computers.
This brings us to the motivation for a combined approach. Mobile
devices are general-purpose devices which can be used as an extension
to the WWW - instead of special-purpose devices - to offer more security
and mobility without any extra cost. These mobile devices can be personalized and can store secret information such as cryptographic keys.
They can be used in combination with any computer, i.e., the personal
computer at the user's home, but also a public terminal, hereby providing mobility. Moreover, the computer terminal must not necessarily be
completely trusted, as (part of) the security will rely on trusted and/or
secret information that is securely stored in the device (and never leaves
it, in case of secrecy).
In the remainder of this paper, this combined approach will be illustrated with an electronic payment system for the WWW that makes
use of a mobile phone. This GSM based system is an alternative to
the widely spread credit card based solution, offering more security and
equivalent mobility and complexity (assuming that a mobile phone is
standard equipment of many users). In addition, it might be suited for
lower-price transactions.

5.

GSM based payment for the WWW

The main goal of the remaining part of the paper is to present a

system in which the WWW and GSM environment are combined to
improve overall security, mobility, and functionality. In particular, an
architecture and protocol are developed in which: (1) a customer can
initiate and complete an electronic payment over the GSM network where
the network operator is an active participant; (2) the pre-payment related
interaction is done via the WWW; (3) the customer receives a receipt
with which he/she can pick up the goods (post-payment).

5.1.

Involved entities

The following entities play an active role in this e-commerce system:

Customer.
The Customer wants to buy something via the WWW.
Payment will be done via his/her GSM. The Customer will receive a
receipt, with which he/she can pick up the goods (the system must
work with both physically deliverable goods and electronically available
goods). Obviously, the Customer should have a PC with Internet connection. This can also be a public terminal. He/she needs a mobile phone
with SIM Application Toolkit functionality. The SIM card should be

162

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

issued by a Network Operator that is running this electronic payment

service. Optionally, there should be a connection between the mobile
phone and the PC, and accordingly some extra software on the PC.

Merchant.
The Merchant wants to sell something via the WWW.
He/she should have a web server, and an access point to the mobile
network. Examples are an on-line bookstore, a pizza delivery chain, an
electronic parts shop, etc.
Deliverer. The Deliverer is the local (with respect to the Customer)
representative of the Merchant. It will deliver the goods after having
verified the receipt the Customer has obtained from the Merchant. The
Deliverer should have some equipment to verify this receipt. An example
is the pizza delivery boy/girl, etc. The Deliverer can also be another
company that made an agreement with the Merchant. For example, the
Merchant can send the goods to a gas station near the Customer; in this
case, the gas station is the Deliverer where the Customer can pick up
the goods.
Network Operator. The N.O. plays the role of the bank. It will deduct the necessary amount of money from the Customers balance (can
be credit or pre-payment based), and add this amount to the Merchants
balance. A commission on this amount will be taken, or a periodical fee
will be requested from the Customer and/or Merchant. In practice there
will be multiple N.O.s: N.O.(C), N.O.(M) and N.O.(D), for the Customer, the Merchant and the Deliverer respectively (as shown in Fig. 1).
Note that in reality, and from a non-technical point of view, it might
not be easy for any Network Operator to deploy an electronic payment
service (e.g., banking license). Alternatively, the Network Operator
could in this system be replaced by a real financial institution, which
makes an agreement with one or more operators.

5.2.

Architecture and protocol

From a high-level point of view, the different entities perform the

following interactions (see Fig. 1): after browsing and negotiating, the
Customer requests a purchase; via an SMS message, the Merchant asks
the Customer to pay the purchase; the Customer pays by sending an
SMS message to the Network Operator; the Network Operator informs
the Merchant about the successful payment; the Merchant sends a receipt to the Customer (also an SMS message); the Customer can use
this receipt to pick up the goods at the Deliverer.

163

Combining World Wade Web and wireless security

=I&,

of money

CLIENT ME

www

CUSTOMER

12

DELIVERER

MERCHANT

CLIENT PC

1.
2.
3.
4.

Purchase Request
Purchase Confirm
Verification by Customer
Debit Account

5. Inter-N.O.
6. Delivery OK
Figure 1.

7. Verification by Merchant
8. Receipt
9. Presentation of Receipt
10. Verification by Deliverer

11. Delivery of Goods

12. Confirmation of Reception

GSM based payment for the WWW: architecture and protocol

The protocol contains the following steps (see Fig. 1):

1. Purchase Request.
After browsing and negotiating (0), the
Customer makes a Purchase Request via the WWW (1). The Merchant
can choose the format and encoding of the message. It should at least
contain a description of the goods, the amount of money to be paid,
and the Customers GSM number (in order to be able to send an SMS
message to the Customer). The message will normally be sent through
submission of an HTML form. The level of protection can be chosen by
the Merchant, but it will normally be protected in transit by SSL/TLS.
The form could also be digitally signed by the Customer (e.g., Netscapes
Javascript signing capability, or Signed XML; note that a mobile device
might in fact not provide any added valu e in this case).
2. Purchase Confirm. The Merchant sends a PurchaseConfirm via
S M S (2) to the Customers mobile phone. This message should be in
a standard format, and is optionally digitally signed by the Merchant.
The message contains: (optionally) a description of the goods (either a

164

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

hashed form of the description, or an abbreviated yet unique description

of the goods, e.g., as in supermarket receipts), a Transaction ID (TID), a
unique Merchant ID, the ID of N.O.(M), and the amount of money to be
paid. The Merchant also sends a Purchase Confirm via the WWW (2).
Note that this could already be included in the reply to the submission
of the Purchase Request form.
3. Verification by the Customer. The Customer verifies whether
all the ordered goods are listed, and whether the amount of money requested equals the amount agreed on. The information in the SMS message should be the same as the information displayed in the browser.
Authentication of the Merchant thus relies on both GSM (we assume
that the Customer knows the number of the Merchant) and SSL/TLS,
so the Customers trust in the correct execution of the transaction increases. If the reply in the browser and/or the SMS message are digitally
signed, the signatures are verified. Note that in current GSM phones
such a signature must possibly be verified using additional software on
the computer. This requires a connection between the mobile phone
and the PC which can for example be provided by Bluetooth. An automatic verification and comparison of the reply in the browser and the
SMS message can then also be made. The interface to the Customer
is provided by the SIM Application Toolkit. A payment application is
installed on the SIM card, which is invoked on receipt of a Purchase
Confirm message.

4. Debit Account.
The SIM Application Toolkit application asks
the Customer a confirmation for sending a Debit Account message (4) to
the N.O.(C). This message includes the amount of money to be paid, the
TID, the Merchants ID and N.O.(M)s ID. The authentication of the
Customer relies on GSM entity authentication (the Customers mobile
phone number should be in the Merchants database). The TID will
allow verification by the Merchant afterwards.
5. Inter-N.O. The N.O.(C) deducts the proper amount of money
from the Customers balance, and forwards the Debit Account message
to N.O.(M). The N.O.(M) adds the amount to the Merchants account.
6. Delivery OK. The N.O.(M) sends a Delivery OK (6) to the Merchant. This message contains the amount of money and the TID, and
can be digitally signed by the N.O.(M).

Combining World Wide Web and wireless security

165

7. Verification by the Merchant.

The Merchant verifies if the
Delivery OK message originates from the N.O.(M) (relying on GSM entity authentication). If added, the digital signature of the N.O.(M) is
verified. The Merchant looks up the TID in his transaction database,
and checks if the amount of money is the same as included in the corresponding Purchase Confirm messages.
8. Receipt. The Merchant sends a Receipt (8) to the Customer
via SMS. It contains: a (hashed) description of the goods, the TID, a
timestamp (in order for the Deliverer to verify the freshness of the receipt), information on the Deliverer (optionally depending on the Customers cell location, and including the Deliverers GSM number), and
information on the Customer (optionally including its GSM number, to
allow verification of ownership of the receipt). The receipt is digitally
signed by the Merchant. The receipt can only be used for the intended Deliverer as indicated. The TID and timestamp ensure that the
receipt cannot be replayed by the Customer (i.e., the Deliverer should
keep a list of previously received TIDs and should not accept receipts
that are too old). GSM authentication is relied upon for authenticating
the Customer.
9. Presentation of the receipt.
If goods are electronic and delivered via the WWW, a receipt is not needed. Goods are then downloaded using the TID. The Merchant keeps a list of which TIDs correspond to transactions for which a payment has been received. Physical
goods should be retrieved at the Deliverer. The receipt is forwarded to
the Deliverer (9), manually or through the SIM Application Toolkit, or
the Customer just presents the receipt to the Deliverer on the screen of
his/her own GSM.
10. Verification by the Deliverer.
The Deliverer just reads the
receipt from the screen of the Customers or his/her own GSM, or he/she
verifies the receipt more properly by checking if the signature of the
Merchant is valid. The Deliverer needs some infrastructure with GSM
access point for this (e.g., a GSM connected to a laptop).

11. Delivery of goods. If the receipt is valid, the Deliverer can be

sure that the Customer is the one that has made (and paid) the purchase.
The goods can thus be delivered (11). In case of electronic goods which
are delivered directly by the Merchants web site (not necessarily though,
as the Deliverer might have its own web site), the Customer should be
granted access based on the TID: after a Delivery OK message has been

166

ADVANCES IN N E T W O R K A N D DISTR. SYSTEMS SECURITY

received, the Merchant enables the access to the information; the TID
should not be known to other entities (however, note that the N.O.
should be trusted not to misuse its knowledge of the TID).
12. Confirmation of reception. After the Customer has obtained
the goods, it can optionally be required that he/she confirms the reception of the goods (12), e.g., by digitally signing a specific message. This
will prevent Customers from denying later on having received the goods.

6.

Analysis and remarks

The proposed GSM based electronic payment system for the WWW is
analyzed further in this section. Some GSM specific comments are given,
the security and privacy of the system is evaluated, and a comparison
with a number of similar systems is made. Note that this section only
intends to discuss this particular example, and not the general combined
approach.

6.1.

GSM functionality

The protocol relies on SMS messages. These can only contain 160
characters, which should be taken into account when defining the exact
content of the protocol messages. Note that GSM provides a mechanism to send long messages as a concatenation of multiple SMS messages. Since the protocol involves on-line bi-directional communication
between the entities, there should be not much latency between sending
and receiving SMS messages. This might be a problem in the case of
international roaming.

6.2.

Security

The security features of SSL/TLS and GSM form together a basis for
the security of the proposed electronic payment system. By having a
close link between the two, the security is even improved.
The Customer can securely request a purchase via SSL/TLS. The
Customer will receive a confirmation via this same secure channel, and
also on its mobile phone. Therefore, the Customer can double-check
the Merchants identity, and the contents of the purchase, including the
amount of money to be paid.
The Merchant can rely on the GSM network to be sure to receive
an authenticated payment from the Customer via the Network Operator later on. Moreover, the Customer cannot cheat by requesting its
Network Operator to deduct a smaller amount of money than origin-

Combining World Wide Web and wireless security

167

ally requested by the Merchant. The Merchant would notice the smaller
amount of money and not send a receipt.
The Deliverer can validate a receipt by verifying the digital signature
of the Merchant, and by checking if the receipt is fresh. Thus, receipts
cannot be forged, and cannot be replayed. Moreover, if the Customers
mobile phone number is included in the receipt, the Deliverer could rely
on GSM authentication and check if the receipt is actually presented
by the original initiator of the transaction (provided that the Customer
allows its own number to be sent to the receiving end; note also that
for some applications, Customers might desire to be able to forward the
receipt to another party that in its turn can pick up the goods).
As on top of SSL/TLS and GSM, some crucial messages are digitally
signed; this decreases the need for Customers and Merchants to trust
each other (i.e., they only need to trust they use the right public key,
which should be ensured by the certificates that are issued by mutually
trusted CAs). For example, since the receipt is digitally signed, it cannot
only be verified by the Deliverer, but also by a Judge, in case of a dispute.
Note that the latter also requires that the receipt includes a unique and
indisputable description of the goods that should be delivered.
The Network Operator is trusted to transfer the proper amount of
money from the Customers to the Merchants balance. It is expected to
do so, as its business would otherwise quickly collapse due to negative
publicity.
In some sense, the Customers mobile phone can be considered as a
secure and personal device (and care should therefore be taken that it is
not easily stolen or lost). The strength of the electronic payment system
proposed in this example relies particularly on the security of such a
device, which is combined with the advanced yet insecure environment
provided by the P C and the browser.

6.3.

Privacy

The presented electronic payment system seems to offer more security

than todays widely used mechanisms; however, it does not really offer
more privacy. Merchants know at least the mobile phone number of their
Customers. This number does not necessarily reveal a Customers real
identity (as opposed to an ordinary credit card payment). There already
exist phone books with GSM numbers though. One would for example
certainly not be happy when this number would be used for advertisement purposes. In fact, for this reason, some people will be reluctant to
release their phone number, while they freely disclose their credit card
number to merchants. The ability of hiding numbers or anonymizing

168

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

customers in another way, would thus be an improvement of the system.

Just as with credit card payments, the Network Operator knows exactly
which Customers are buying goods from which Merchants and for what
amount of money. The Network Operator will not necessarily know the
actual nature of the goods though.

6.4.

Other approaches

Other GSM based payment systems exist. GiSMo [14] is a system

intended for the Internet in which customers receive a random code
through SMS via a central server. This random code is then entered
via the computer in order to pay. Mint [20] is a system in which each
terminal/shop has a unique phone number which the customer should
just call at the time of payment. Similar alternatives are Jalda [18] and
Paybox [24].
In the system presented in this paper, more payment related information is exchanged via GSM, which results in a closer link between the
WWW and the GSM interaction. Conceptually, it is also more general
and independent of the wireless system. With more advanced mobile
devices and networks, such as UMTS, more secure schemes would be
possible, following the same architecture and protocol, but with different content of (and another exchange mechanism of) the messages.
For example, instead of an account based protocol, electronic cash like
schemes could be used. Mobile devices with built-in smart card readers
would be very useful for integrating smart card based payment means
as used in the physical world.

7.

Conclusion

Electronic commerce is already a normal part of peoples ordinary

life. Mobile devices, and certainly mobile phones, are currently widely
spread. This paper gave a brief overview of the security properties of the
World Wide Web and some existing mobile systems. The main purpose
of this paper was to suggest to use a wireless system as an extension to
the WWW, to provide more security and functionality. To demonstrate
this combined approach, a GSM based electronic payment for the WWW
was presented.
Unlike most mobile phones, some mobile devices are powerful and advanced enough to allow more or less convenient browsing and shopping.
Future mobile systems will also be more secure and will offer more functionality than the GSM system or than WAP. Yet, the concept of using
an out-of-band channel for electronic payment, and the combined use
of a mobile device together with a normal PC, will remain very useful.

Combining World Wide Web and wireless security

169

For the PC and its big screen will always be far more advanced than the
mobile device, but will never be mobile.

Acknowledgments
Joris Claessens is funded by a research grant of the Flemish Institute
for the Promotion of Industrial Scientific and Technological Research
(IWT). This work was also partially supported by the Concerted Research Action (GOA) Mefisto-666 of the Flemish Government.
This work was mainly initiated during a research visit at the Information Security Research Centre (ISRC) of Queensland University
of Technology (QUT), Brisbane, Australia. The first author wants to
thank Dr. Mark Looi of the ISRC for the interesting discussions on this
topic. The authors also want to thank Dr. Silke Holtmanns of Ericsson
for the various pointers to existing mobile payment systems, and the
anonymous reviewers for their constructive remarks and suggestions.

References
[1] American Express. Private Payments. ht tp:// www. americanexpress.com
/privatepayments/.
[2] Banxafe. htt p://www.banxafe.com/.
[3] S. M. Bellovin. Security Problems in the TCP/IP Protocol Suite. Computer
Communication Review, 19(2):32-48, April 1989.
[4] Simon Blake-Wilson, Magnus Nystrom, David Hopwood, Jan Mikkelsen, and
Tim Wright. TLS Extensions. IETF Internet Draft, June 2001.
[5] Bluetooth SIG. http://www.bluetooth.com/.
[6] Nikita Borisov, Ian Goldberg, and David Wagner.
Intercepting
Mobile
Communications:
The
Insecurity
of
802.11.
http://www.isaac.cs.berkeley.edu/isaac/wep-draft.pdf.
[7] Clara Centeno. Mobile Payment Industry Fora - Consolidation of Initiatives
Expected. Electronic Payment Systems Observatory - Newsletter, ePSO-N,
(8):8-12, July 2001. Available at http://epso.jrc.es/.
[8] T. Dierks and C. Allen. The TLS Protocol Version 1.0. IETF Request for
Comments, RFC 2246, January 1999.
[9] Digipass. http://www.vasco.com/.
[10] D. Eastlake, J. Reagle, and D. Solo. XML-Signature Syntax and Processing.
IETF Request for Comments, RFC 3075, March 2001.
[ll] ETSI. Digital cellular telecommunications system (Phase 2+); Security mechanisms for the SIM Application Toolkit; Stage 1. ETSI TS 1 01 180 (GSM 02.48).
[12] ETSI. Digital cellular telecommunications system (Phase 2+); Security mechanisms for the SIM Application Toolkit; Stage 2. ETSI TS 101 181 (GSM 03.48).
[13] ETSI. Digital cellular telecommunications system (Phase 2+); Specification
of the SIM Application Toolkit for the Subscriber Identity Module - Mobile
Equipment (SIM - ME) interface. ETSI TS 101 267 (GSM 11.14).

170

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

[14]
[15]
[16]
[17]

GiSMo. http://www.gismo.net/.
IETF Working Group. IP Routing for Wireless/Mobile Hosts (mobileip).
InternetCash. http://www.internetcash.com/.
Markus Jakobsson and Susanne Wetzel. Security Weaknesses in Bluetooth. In
D. Naccache, editor, Progress an Cryptology - Proceedings of the Cryptographers'
Drack at RSA 2001, Lecture Notes in Computer Science, LNCS 2020, pages 176191. Springer-Verlag, 2001.
[18] Jalda. http://www.jalda.com/.
[19] Peter A. Loscocco, Stephen D. Smalley, Patrick A. Muckelbauer, and Ruth C.
Taylor. The Inevitability of Failure: The Flawed Assumption of Security in Modern Computing Environments. In Proceedings of the 21 st National Information
Systems Security Conference, pages 303-314, October 1998.
[20]
[21]
[22]
[23]
[24]
[25]
[26]

Mint. http://www.mint.nu/.
Mobile Electronic Signature Consortium. http://www.msign.org/.
Mobile electronic Transactions. http://www.mobiletransaction.org/.
NTT DoCoMo. i-mode. http://www.nttdocomo.co.jp/i/.
Paybox. http://www.paybox.de/.
Proton. http://www.protonworld.com/.
E. Rescorla and A. Schiffman. The Secure HyperText Transfer Protocol. IETF
Request for Comments, RFC 2660, August 1999.
[27] Eric Rescorla. SSL and TLS: Designing and Building Secure Systems. AddisonWesley, 2000.
[28] Bruce Schneier. European Cellular Encryption Algorithms. Crypto-Gram,
December 1999.
[29] SET Secure Electronic Transaction LLC. SET Secure Electronic Transaction
Specification. http://www.setco.org/.
[30] Trusted Computing Platform Alliance. http://www.trustedpc.org/.
[31] The UMTS Forum. http://www.umts-forum.org/.
[32] Klaus Vedder. GSM: Security, Services and the SIM. In B. Preneel and V. Rijmen, editors, State of the Art in Applied Cryptography, Lecture Notes in Computer Science, LNCS 1528, pages 227-243. Springer-Verlag, 1998.
[33] M. Walker. On the security of 3GPP networks. Invited talk at Eurocrypt 2000.
[34] Wireless Application Protocol Forum. WAP Certificate and CRL Profiles. Approved 2 2-May-200 1.
[35] Wireless Application Protocol Forum. WAP Public Key Infrastructure. Version
24-Apr-2001.
[36] Wireless Application Protocol Forum. WAP TLS Profile and Tunneling. Version
11-April-2001.
[37] Wireless Application Protocol Forum. WAP Transport Layer End-to-end Security. Approved Version 28-June-2001.
[38] Wireless Application Protocol Forum. WAP Wireless Identity Module, Part:
Security. Version 12-July-2001.
[39] Wireless Application Protocol Forum. WAP Wireless Transport Layer Security.
Version 06-Apr-2001.

Combining World Wide Web and wireless security

171

[40] Wireless Application Protocol Forum. WAP WMLScript Crypto Library. Ver-

sion 20-Jun-2001.

This page intentionally left blank.

ON MOBILE AGENT BASED

TRANSACTIONS IN MODERATELY
HOSTILE ENVIRONMENTS *
Niklas Borselius, Chris J. Mitchell, Aaron Wilson
Mobile VCE Research Group, Information Security Group,
Royal Holloway , University of London
Egham, Surrey

TWZO OEX, UK

[email protected], C.MitchellQrhul.ac.uk, [email protected]

Abstract

When using mobile agents, numerous security issues must be considered.

In this note we propose two methods t o improve t h e security and reliability of mobile agent based transactions in a n environment which may
contain some malicious hosts.

Keywords: mobile agent, digital signature, transaction, security

1.

Introduction

In this paper we consider strategies for the deployment of mobile

trading agents to reduce certain security threats to their operation. In a
future world of co-operating mobile and fixed devices, the mobile agent
computing model is expected to become an increasingly important one.
In the domain of e-commerce/m-commerce transactions, mobile trading
agents could play a very useful role. Users could launch such agents to
make transactions on their behalf, and the agents would look for the
best buy by visiting multiple merchant sites without any direct user
intervention. Indeed such activity could take place while the user has
no current network connectivity.
The mobile agent computing model gives rise to a range of security
threats. These threats can be divided into two main classes:
*The work reported in this paper has formed part of the Software Based Systems work area
of the Core 2 Research Programme of the Virtual Centre of Excellence in Mobile & Personal
Communications, Mobile VCE, www.mobilevce.co.uk, whose funding support, including that
of EPSRC, is gratefully acknowledged. More detailed technical reports on this research are
available to Industrial Members of Mobile VCE.

174

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

threats to the platform from malicious and/or unauthorised agents,
including threats to the integrity of the platform and other agents,
threats to the confidentiality of stored data, and denial of service
threats, and
threats to the agent from malicious platforms, including threats to
the confidentiality of agent stored data, and threats to the integrity
of the agent and its computations.

In this paper we are concerned with the second class of threats, and
in particular with threats to agents deployed for trading applications.
Specifically, users will need to give trading agents certain authority to
authorise transactions, whilst at the same time users will wish to protect
themselves against malicious merchants forcing an agent to make a nonoptimal purchase.
We consider simple ways in which deployment of multiple agents can
reduce the threat to trading agents from platforms outside of their direct
control. We consider two general approaches. In the first approach
multiple agents are equipped with shares of the means to commit to
a transaction. In the second approach a single trusted host provides
a location for multiple agents to report back information enabling a
purchasing decision to be made.
The paper has the following structure. The next section explores
threats to trading agents in more detail. This is followed in Sections 3
and 4 by a discussion of the models used here for agent platforms and
for trading agents. Sections 5 and 6 then explore the two approaches to
enhancing trading agent security.

Agent Security Issues

2.

The use of mobile agents raises a number of security concerns. Agents

need protection from other agents and from the hosts on which they
execute. Similarly, hosts need to be protected from agents and from
any party which can communicate with the platform. The problems
associated with the protection of hosts from malicious code are quite
well understood.
The problem of malicious hosts seems the hardest to solve. In fact
some people hold the opinion that it is insoluble. The particular attacks
that a malicious host can make have been described in [Hoh98a] and
[Has00], and can be summarised as follows.
m

Observation of code, data and flow control,

Manipulation of code, data and flow control
lating the route of an agent,

including manipu-

On mobile agent based transactions in moderately hostile environments

rn

Incorrect execution of code - including re-execution,

rn

Denial of Execution - either in part or whole,

rn

Masquerading as a different host,

rn

Eavesdropping of agent communication,

rn

Manipulation of agent communication,

175

False system call return values.

There have been many attempts to address these threats either completely or in part. Most of these attempts fall into one of the following
broad categories.
The first category comprises approaches that do not allow an agent
to leave a trusted environment. Solutions to this include using a
host infrastructure that is operated by a single party, allowing
agents to migrate only to trusted hosts [FGS96], or possibly hosts
with a good reputation [RJ96].
The second category is pragmatic; it consists of solutions to a single
part of the malicious host problem. These consist of agents detecting when they have been modified [Vig97], and proof verification
techniques [Yee97].
The third class consists of assuming that there is special, tamperproof hardware available, see for example [Yee97] or [WSB98].
The final category uses software methods to obscure the code from
the host. Approaches include obfuscation [Hoh98b] [Ng00], mobile
cryptography [ST98, ST97] and using environmental conditions to
hide parts of the code [RS98].
The approaches described in this paper, based on replicating agents, do
not fit into any of the above four classes. There appears to be relatively
little literature devoted to this approach to dealing with the threats to
agent security.
We now consider the threats to a trading agent in more detail.

2.1.

Threats to trading agents

We now turn to look at the particular threats to an agent which wishes

to purchase an item (or a service) from a merchant. These all fall into
the categories above. We concentrate on the threats to an agent involved
in a trade, rather than more general threats.

176

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

1 A malicious host lies about offer.

Here a host lies about the offer it makes to an agent, in order to
get the trade. The host would then charge a higher price at a later
date. One way around this is to force the host to sign its bid,
thereby committing to it.

2 A malicious host learns other offers and undercuts them.

If a host knows that all offers but its own have been collected and
finds out the best standing offer, it can undercut the best standing
offer slightly (in fact the host need not know all other offers, it
could just undercut the current offers). (In some circumstances
letting hosts undercut each other might be considered a desirable
feature.)

3 A malicious host learns the price a user is prepared to pay and

bids just under this.
In a similar fashion the host may charge more than its normal
price, if it knows the maximum price the user is prepared to pay.
Thus a host must be kept from learning the maximum price a user
is prepared to pay, either by encrypting this information or by not
sending this information with the agent.
4 A malicious host manipulates the requirements.

This is when the host changes the requirements to favour its bid.
For example, it could add a requirement to buy from a certain
host, or remove constraints from the agent.

5 A malicious host alters the agents route.

Here, the host keeps the agent away from its competitors, and
thus secures the agents trade. One way to prevent this is to use
more than one agent (possibly an agent per host), send each agent
on a different route and combine the offers on the agents return.
Another way is to use one agent with a star like route - it returns
home after visiting each host before being sent out to a different
host .
6 A malicious host commits to purchases that the user does not wish
to make.
This happens when a host can abuse the committal function that
an agent has. A method to discourage this is to force the host to
sign a transaction, as well as the user (thus providing traceability).
7 A malicious host denies the agent a service.

On mobile agent based transactions in moderately hostile environments

Figure 1.

177

A model for agent platforms

Here a host would stop an agent from moving further on its route.
This of course could be traced if an agent reports when it arrives
on a host.

8 A malicious host captures electronic money.

Here a host would remove the electronic money that an agent may
have to purchase an item and either steal the money outright, or
use it for a different purchase.
We do not consider the payment process here, as we are concerned
only with the part of a transaction involved in selecting a merchant and
committing to the transaction.

3.

Models of Agent Platforms

Mobile agents roam between platforms. However, they can also communicate with each other, and with other hosts. This leads to the question as to the best "platform" model to use for trading (or indeed any
other) agents. There are clearly two basic approaches which we now
describe.
The first approach (see Figure 1) is to have a designated platform
(or a collection of such platforms) to which we can send an agent to
execute. This agent then communicates with merchant servers to seek
information and commit to purchases.
The second model (see Figure 2) is to have an agent roam to each
merchant server in turn and collect the information it requires. After
collecting all the information the agent can then either return to the
user to make the purchase, return to the chosen merchant to make the
purchase or make the purchase from the final host.
In a mobile telecommunications environment it may also be beneficial
to have a third model. This is where the requirements for a purchase are
communicated to a 'home platform' (the user's home PC or a network

178 ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

operator controlled device) which then forms the agent and conforms to
one of the above models.
In the above, any of the platforms may be malicious, with the possible
exception of the home platform. The solutions proposed below can be
made to fit into any of the above situations, although they both fit better
into the first model.
The security risks associated with the above two models clearly differ.
In the first case, the designated platform might be trusted to keep
secret certain agent information. An example of where this might be
useful is when the agent contains details of the user expected price (or
maximum price), which it would be helpful not to reveal to the merchant.
Of course, the threat then arises that one of the designated platforms
will collude with one or more of the merchants. In the second case, it
is clearly impossible to try and keep any information in the agent secret
from the merchants. In both cases, however, as we will show in the
remainder of this paper, there are potential benefits to be gained from
the use of multiple agents, albeit not from the confidentiality perspective.

4.

Model for a trading agent

We consider the information that an agent wishing to trade must

know. Firstly, when initiating a purchase, a user will have a set of
requirements (for instance the item to be purchased, the maximum price
for that item, a time limit within which the purchase to be made). We
will assume that a user encodes these requirements into a string R which
is understood by all parties. When a server quotes for a given purchase,
it will also produce a similar string with its offer.
The agent, if it is to perform the purchase on behalf of the user, must
also carry a function which will commit to the trade. This could be
performed by, for instance, signing the details of the trade. One scheme
to allow an agent to perform a signature operation on behalf of a user
without revealing the users private key to a host is proposed in [KBCOO].

Merchant

Merchant

Figure 2.

HMerchant

Merchant

A second model for agent platforms

On mobile agent based transactions in moderately hostile environments

179

In this scheme, using RSA, an agent carries both the hash value, h , of
the requirements and the signed hash value, hd mod n, where ( d , n ) is
the users private RSA key. To commit to a transaction for the user the
agent calculates
(hd)= hxd = (h)dmod n
effectively signing h where 2 is the servers offer. An alternative to this
where the agent carries its own private key which the user certifies is
given in [BMW01a].
Thus we assume that a trading agent will carry the following information:
User Identifier

Requirements for purchase

m

A committal function - C. The committal function is used by the

agent to commit to a transaction on the users behalf. C could be
a signature function using a special private key provided to the
agent by the user. Alternatively, C could be a function of the type
described above, derived from the users own private signature
key. In any event, we assume that the function is designed so that
only transactions within constraints defined by the user can be
author ised.

Note that, if a single trading agent is deployed there are a number

of problems which might arise. Firstly, although the committal function
will typically be limited to transactions conforming to user-defined parameters, there is still the possibility that the agent platform will force the
agent to commit to a transaction which is less than optimal. It may also
commit to more than one transaction, even if the user only intended to
make at most one purchase.
One way to reduce this threat is to deploy multiple agents, a subset
of which must agree to the transaction before it can be authorised. Such
an approach is the focus of the remainder of this paper.

5.

Threshold Scheme

We attempt to solve the malicious host problem by using multiple

agents each of which has a vote. If one of the possible transactions
receives enough votes, then a transaction will be authorised with the
relevant merchant. We begin by outlining the scheme, and then consider
the details of what a secure vote can consist. We assume use of a (k, n )
scheme - i . e . a server will need k votes out of a possible n to win.

180

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

The Scheme
Let T = {TI,
T2,.
. . ,Tn}be a set of agent platforms. The user then
sets up a (k,n ) voting scheme with shares s1,s2,. . . ,sn. Clearly k should
5.1.

exceed the number of suspected malicious hosts. Given no information

about the system a sensible value would probably be n/2 + 1. The value
of k reflects the level of trust in the system.
The user then forms n agents Ai (1 5 i 5 n ) containing the following
information
User Identifier

Requirements for purchase

A vote

si

Each agent is then dispatched to its agent platform. At the platform

there are two modes of execution:
1 The agent contacts each merchant itself, and gathers bids that
meet the requirements.

2 The agent contacts a subset of the merchants and communicates

the best bid to its peers.
We note that case (2), unless only contacting a single host, is a situation
that must be carefully thought out. This is because if there is no overlap
in the merchants, collusion may mean that attacking less than k servers
is necessary.
When each agent has received all the information about each bid, the
agent sends its vote to the merchant with the best offer. On receipt of
the correct number of votes, the merchant or a nominated third party
can construct (and verify) from the votes the authorisation for the bid.
The merchant or nominated third party can then use this as evidence
that the user has committed to transaction.
We now consider the security of the above scheme. The major advantage of the scheme is the need to corrupt either n - k + 1 agents
to prevent the transaction or k hosts to divert or alter the transaction.
Thus the choice of k is crucial.
This also means that a denial of service attack is harder as a server
or set of colluding servers will need to terminate (or prevent from communicating their vote) ( n - k ) + 1 agents. Again to force a purchase a
host or hosts must force k agents to offer their vote.
If an agent visits a subset of the servers involved, the information
could then be used to help identify any malicious hosts.

On mobile agent based transactions in moderately hostile environments

5.2.

181

The Votes

As mentioned above, the votes can be assembled by either the selected merchant or a nominated third party. Note that there are clear
risks associated with giving votes to the merchant, since the merchant
could now possibly commit the user to a transaction of the merchants
choice (within any constraints imposed by the string R). That is, the
merchant is not forced to commit to the transaction as offered to the
agents. Hence the use of a nominated third party to reconstruct the
votes is the preferred approach. The possibility that this may not be
feasible in practise leads to an alternative approach.
One approach is thresholdcryptography.Threshold cryptography was
first proposed by Desmedt [Des88]. A typical example of a threshold
cryptosystem is one that would allow a set of t parties to sign any document such that any coalition of less than t parties cannot sign any other
document. Schemes tend to rely on a combiner which does not necessarily need to be trusted. Schemes based on both RSA and El Gamal
have been proposed.
Recently Shoup [Sho00] proposed an RSA scheme which is as efficient
as possible; the scheme uses only one level of secret sharing, each server
sends a single part signature to a combiner and must do work that is
equivalent, up to a constant factor, to computing a single RSA signature.
Although not perfect as a threshold signature scheme (as it relies on a
trusted party to form the shares) this scheme is ideal in our setting.
(Note that an alternative scheme without a trusted dealer is given in
[DK01]. This scheme also improves on Shoups scheme by not relying on
an RSA modulus made up of safe primes). An example of an El Gamal
scheme is given in [Lan95]. We note that a (n, n) threshold signature
scheme is just a multisignature; such schemes have been studied for many
years - see, for example, page 488 of [MvOV96].
We note, however, that such a threshold signature scheme does not
provide a means for the shares to incorporate an encoding of the string
R . Thus, if there were k colluding hosts they could sign (and reconstruct
a signature) for any document. One solution to this problem is for the
user to generate a special signature key pair for the particular purchase
(i.e. for this particular set of agents), and then to generate a certificate
for the public key incorporating a copy of R. When the signature is
reconstructed from the signature shares, it can be verified using this
certificate. However, it is possible to merge the undetachable signature
scheme given in [KBC00] with the threshold signature scheme of Shoup
[Sho00] and details of this are given in [BMW01b].

182

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

6.

Using one trusted host

We consider a second solution to the problem, which employs a single

trusted host. We note that the solution described below involves a user
sending out agent(s) to individual merchant servers, whereas it could
just communicate with them to ask for their bids. However, in a wireless
communications setting where communication is expensive, slow, and/or
unreliable, it is believed to be beneficial to be able to dispatch an agent
into the fixed network. When the agent has finished its task it contacts
the user or waits for the user to collect the result.
Let S = {Sl, S2,. . . ,Sn} be a collection of servers offering a service
that a user wishes to purchase. Let T be a host that the user trusts to
act honestly in this transaction. (Note that we do not need to trust this
host fully - it just needs to be neutral in this transaction). Before the
transaction commences we assume that each server Si and T securely establishes a shared secret key Ki. Optionally, a key for message integrity
checks could also be established.
The user despatches an agent A to the trusted host containing the
information outlined in $4. We note that the committal function C may
be of any form with which the user is prepared to trust the host T .
However, to reduce the trust requirements we envisage that this will be
the scheme outlined in either [BMW01a] or [KBC00].
There are now several approaches for T . The first is to form a single
subagent containing the following information
rn

Agent identifier - I

rn

Requirements for purchase - R

rn

Host identifier

which would then visit each of the servers in S in turn. We note that
the requirements sent out do not need to include pricing information
(that is the maximum price the user is prepared to pay) or any other
information that the user wishes to be used to help make the decision,
but does not wish to communicate to the server. Another approach is
to form a single agent for each server. A third approach has the above
agent visiting a subset S' c S of the above servers. Whichever strategy
is employed, at each host the agent performs the following actions:
1 Find out the server's bid Bi for the item specified in the requirements R.

2 Encrypts the concatenation of Bi, R, Si and I using either Ki.

At this point the server could also, optionally, attach a symmetric

On mobile agent based transactions in moderately hostile environments

183

MAC (Message Authentication Code) to the bid to protect the

integrity of the servers bid. Label the encrypted string E,.
3 The agent then stores the pair (Si,Ei).

The agent returns to T when it has finished visiting all of its servers.
The agent on T then decides the best offer and commits to it using the
committal function.
We note some of the features of the above scheme.
rn

Using an agent per server really alleviates the need to encrypt

anything, assuming that agents are always transferred between
hosts in encrypted form.

rn

Using a single agent leaves yourself open to some attacks.

rn

Using more than one agent that does not visit all the hosts could
then be used to (help) identify a malicious host.

If we use a single agent and it visits all the hosts, or we have an agent
that visits more than one host, the agent is subject to the following
at tacks:
An approach to enable a malicious host to underbid its competitors, is as follows. The host forms a new agent containing the
users requirements, a fictitious user identifier, and its own host
identifier. This agent would then traverse the route of the users
agent, and discover the bids offered for that set of requirements.
The host could then under bid its competitors, but the users agent
would have had to have been kept on the malicious host in the interim period. Thus monitoring the progress of an agent could help
determine if such an attack was being used.

A simple denial of service attack: stop the agent in its tracks. If

there is no progress monitoring (e.g. agent at host Si) then this
attack is hard to defeat.
A malicious host could alter the pair (Si,Ei) to read (Sj,junk)
(where junk is a random string of the correct length) to stop the
decryption of a bid. However as the host cannot read the bid, for
this to be successful (i.e. to delete those bids more attractive than
those of the malicious host) the host would have to have knowledge
of all the bids - which it would have to gather itself (possibly by
cloning the agent).
If we use an agent that visits a subset of the hosts, and assume that
the malicious host already knows the best offer at any given point, it

184

ADVANCES IN N E T W O R K A N D DISTR. SYSTEMS SECURITY

will then try to undercut this (this undercut is a lie). If we then require,
we can apply rules to the results of the other agents, and attempt to
identify the malicious host. This also requires careful choice of agent
destinations and routing.
Note that to force T to purchase from a malicious host, the host has
to lie and then be unscrupulous, or just lie and possibly not profit as
much as it would expect. That is if the malicious host M wants to force
a user to trade with it, then it must have the best price. So it must either
charge more than its advertised price (possibly breaking the committal
function) or make less profit than it expects (because the price advertised
is less than the host should sell for).
We now consider the extent to which the user must trust the host T.
The user must trust that T does not favour a particular server for this
transaction. However, with a sufficiently good committal function then
this is the only trust requirement. For example using the Kotzanikolaou
et al. undetachable signature scheme [KBC00], as a committal function,
T can be given the means to commit to the transaction without being
trusted with a copy of the users private signature key. This may be a
situation where using an undetachable signature scheme has advantages
over the creation of a separate signature key for each agent.

7.

Conclusions

We have considered two different ways in which the deployment of

multiple agents can reduce the threat to trading agents from potentially
malicious agent platforms. In the first approach multiple agents are
equipped with shares of the means to commit to a transaction. A
method implementing this idea using a threshold signature scheme, e.g.
the recently proposed scheme of Shoup, [Sho00], was outlined. In the
second approach a single trusted host is employed to collect information
from multiple agents on possible transactions. This host then chooses
the optimal transaction and commits to it.
The two approaches each have their own advantages. The first approach avoids the need for a single trusted host. However, implementing
the first approach requires use of some potentially complex cryptographic
signature functions. The second approach is potentially less complex
from a cryptographic perspective, but does require a host which, if not
completely trusted, is at least required to act neutrally with respect to
the set of merchants. Both approaches are of potential practical importance in future mobile computing environments.

On mobile agent based transactions an moderatelyhostile environments

185

References
[BMW01a]
[BMW01b]

[Des88]

[DK01]

[FGS96]

[Has00]
[Hoh98a]

[Hoh98b]

[KBC00]

[Lan95]

[MvOV96]

[Ng00]
[RJ96]

Niklas Borselius, Chris J. Mitchell, and Aaron Wilson. A pragmatic

alternative to undetachable signatures. Preprint, 2001.
Niklas Borselius, Chris J. Mitchell, and Aaron Wilson. Undetachable
threshold signatures. To be presented at the IMA Conference on Cryptography and Coding, December 2001 (proceedings to be published in
the Springer-Verlag LNCS series), 2001.
Y. Desmedt. Society and group oriented cryptography. In C. Pomerance, editor, Advances in Cryptology - Crypto 87 proceedings, number
293 in LNCS, pages 120-127. Springer-Verlag, Berlin, 1988.
Ivan Damkd and Maciej Koprowski. Practical threshold RSA signatures without a trusted dealer. In Birgit Pfitzmann, editor, Advances in
Cryptology - EUROCRYPT 2001, number 2045 in LNCS, pages 152165. Springer-Verlag, Berlin, 2001.
William Farmer, Joshua Guttmann, and Vipin Swarup. Security for
mobile agents: Authentication and state appraisal. In Proceedings of the
European Symposium on Research in Computer Security (ESORICS),
number 1146 in LNCS, pages 118-130. Springer-Verlag, Berlin, 1996.
Vesna Hassler. Security Fundamentals for E-commerce. Artech House,
2000.
Fritz Hohl. A model of attacks of malicious hosts against mobile agents.
In Proceedings of the ECOOP Workshop on Distributed Object Security
and 4th Workshop on Mobile Object Systems: Secure Internet Mobile
Computations , pages 105-120, 1998.
Fritz Hohl. Time limited blackbox security: Protecting mobile agents
from malicious hosts. In Giovanni Vigna, editor, Mobile Agents and
Security, number 1419 in LNCS, pages 92-113. Springer-Verlag, Berlin,
1998.
Panayiotis Kotzanikolaou, Mike Burmester, and Vassilios Chrissikopoulos. Secure transactions with mobile agents in hostile environments.
In E. Dawson, A. Clark, and C. Boyd, editors, Information Security and
Privacy, Proceedings of the 5th Australasian Conference A CISP 2000,
number 1841 in LNCS, pages 289-297. Springer-Verlag, Berlin, 2000.
Susan K. Langford. Threshold DSS signatures without a trusted party.
In D. Coppersmith, editor, Advances in Cryptology - Cypto 95 proceedings, number 963 in LNCS, pages 397-409. Springer-Verlag, Berlin,
1995.
A. Menezes, P. van Oorschot, and S. Vanstone. Handbook
of Applied Cryptography . Discrete Mathematics and Its Applications. CRC Press, October 1996. Available on-line at
http://www.cacr.math.uwaterloo.ca/hac.
Sau-Koon Ng. Protecting mobile agents against malicious hosts. Masters thesis, The Chinese University of Hong Kong, June 2000.
Lars Rasmusson and Sverker Jansson. Simulated social control for secure internet commerce. In New Security Paradigms 96, pages 18-26.
ACM Press, 1996.

186
[RS98]

[Sho00]

[ST971

[ST98]

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

James Riordan and Bruce Schneier. Environmental key generation towards clueless agents. In G. Vigna, editor, Mobile Agents and Security,
volume 1419 of LNCS , pages 15-24. Springer-Verlag, Berlin, 1998.
Victor Shoup. Practical threshold signatures. In Bart Preneel, editor,
Proceedings of EuroCrypt 2000, number 1807 in LNCS, pages 207 -220.
Springer-Verlag, Berlin, 2000.
Tomas Sander and Christian Tschudin.
Towards mobile cryptography.
Technical Report 97-049, International
Computer Science Institute, Berkeley, 1997.
Available at
http://ww.icsi.berkeley.edu/ sander/publications/tr-97-049.ps.
Tomas Sander and Christian Tschudin.
Protecting mobile agents against malicious hosts.
In Giovanni Vigna, editor, Mobile Agents and Security, number 1419 in LNCS,
Available from
pages 44-60. Springer-Verlag, Berlin, 1998.

http://www.icsi.berkley.edu/sander/publications/MA-protect.ps.
Giovanni Vigna. Protecting mobile agents through tracing. In Proceed[Vig97]
ings ofthe Third E C O O P Workshop on Operating System support for
Mobile Object Systems, 1997.
[WSB98]
U. G. Wilhelm, S. Staamann, and L. Buttyin.
On the
problem of trust in mobile agent systems.
Available from
http://www.isoc.org/isoc/conferences/ndss/98/ndss98.htm,
1998.
Network and Distributed System Security (NDSS98) Symposium.
[Yee97] Bennet Yee. A sanctuary for mobile agents. In DARPA Workshop o n Foundations for Secure Mobile code, 1997. Available from
http://www.cs.nps.navy.mil/research/languages/statemensts/bsy.ps.

SPARTA
A Mobile Agent based Intrusion Detection System
Christopher Krgel*
Distributed Systems Group, Technical University Vienna
[email protected]

Thomas Toth
Distributed Systems Group, Technical University Vienna
[email protected]

Engin Kirda
Distributed Systems Group, Technical University Vienna
[email protected]

Abstract

The large number of machines with different operating systems and applications
in an enterprise network makes it very difficult for the system administrator to
close all security holes and install the latest OS and software patches. When
the network is connected to the Internet and services are remotely available they
become a potential target for hackers. As the number of security related incidents
is constantly increasing at an alarming rate the need for automated tools to detect
intrusions becomes evident. Such tools are called intrusion detection systems.
We present Sparta, a system that allows to detect security policy violations
and network intrusions in a heterogeneous, networked environment. We have
designed a pattern language in order to express intrusions (i.e. offending event
patterns) in a declarative manner. This allows to specify what to detect instead of
how to detect. A fully distributed approach to find the given patterns is presented
as well. We use mobile agents to correlate event data instead of moving the
whole information to a central location. This increases the fault tolerance and
scalability of our system.

Keywords:

Intrusion Detection, Mobile Agents, Pattern Specification Language, Event Correlation, Network Security

*Contact Author

188

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Introduction
Virtually every organization depends on sensitive data which has to be protected against unauthorized access. Such data is often stored on machines
which are remotely available over a network. The growth of the Internet has
caused an increase of the size of individual networks as well as an increase of
transported traffic. This makes it extremely difficult to manually manage and
protect valuable assets. Combined with an alarming rise of attacks and hacking attempts, organizations need tools like intrusion detection systems (IDS) to
enforce security and detect hacking attempts.
Sparta(an acronym for Security Policy Adaptation Reinforced Through
Agents) is the name of a system architecture which is capable of monitoring a
network to detect network intrusions and security policy violations. The system
monitors local events at hosts which are connected by a network, relates them
and provides an interface where the user can query the gathered information.
This makes it possible to apply our design to a broad range of applications and
use it for a number of network related tasks, ranging from network management
to intrusion detection.
The contribution of this paper is the description of an architecture to collect
and relate distributed data in an efficient way by using mobile agents and its
application to network intrusion detection. In contrast to traditional designs
where data is gathered and analyzed at a central location, the application of mobile agents allows distributed analysis. This approach improves the scalability
and increases the fault tolerance in our opinion.

1.

Functional Description

Sparta is an architectural framework which allows to identify and relate interesting events that may occur at different hosts on a network. A single event
is described by specifying appropriate values for its attributes. A number of
events can be connected by defining temporal or spatial relationships between
them or imposing certain constraints on their attributes thereby creating a pattern. In order to deal with complex patterns and systems, it is not sufficient
to select events based on content alone. It is necessary to consider multiple
events at the same time and deduce knowledge that is beyond the scope of an
individual event. The process of detecting a set of events with given properties
is called correlation.
This general correlation capability allows the Sparta architecture to be used
for different distributed applications, ranging from network security to network
management implementations. We currently build a security policy and ID
application based on our design.
The basic functionality can be described as follows. Interesting events are
locally collected and stored. The collection of all local information can be

189

Sparta

considered as a distributed database with horizontal fragmentation. For each

relation (i.e. event type), the tuples (i.e. actual events) are stored at different
locations. A user may issue queries in our Event Query Language (EQL) to
search for a set of events that fulfill his desired constraints. In addition to this
the system can also be used to gather statistical information. It is possible to
find the number of pattern instances at each host and to calculate the maximum
or minimum for event attribute values as well as their sums over a set of hosts.
The query is carried out by mobile agents which return their results to the user.
For our intrusion detection system, a failed authentication attempt or the start
of a root shell might be examples of interesting events. Sparta allows to count
the number of failed telnet logins for a certain user throughout the network (to
detect distributed door knob rattling attempts) or to find tree-like connection
patterns between hosts (to identify a spreading worm). It is important to notice
that event correlation might yield information that is impossible to gain by just
looking at a single node. Consider an intruder who tries to cover his tracks by
performing several consecutive telnet logins (i.e. producing a telnet chain).
This is an often observable behavior that exploits the fact that different machines
are administered by different people and dont have synchronized local clocks.
Tracing an attacker by having to look at all these logfiles is rather difficult.
On each local machine only a simple incoming and outgoing connection is
noticed but when looking at the entire network the offending pattern becomes
evident. GrIDS (Staniford-Chen et al., 1996) is a well known ID system which
bases its detection solely on looking for connection patterns but uses a different
mechanism to collect and relate data.

2.

System Architecture

The system consists of a set of hosts connected by a network where each

node has the following components installed (see Figure 1 ) .
Local event generator (sensor)
rn Event storage component
m Mobile agent platform
rn Agent launch and query unit (optional)
rn

The local event generation is done by sensors which monitor interesting

occurrences on the network (network based) or at the host itself (host based
detection). The exact types of events and their attributes as well as the implementation of the sensors are mainly determined by the applications needs.
The type of an event is represented by the type of the class in the implementation (i.e. Java class), with the events attributes being stored by the members
of the corresponding class. It is possible to extend an event by subclassing
from an existing one and add the desired additional information. This allows to

190

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Figure I.

Sparta Architecture

write patterns which relate high level events and have the system automatically
consider all actual instances (i.e. subclasses) of such generic events.
Sensors store their generated data in a local data storage component, prefer-

ably a database. The data storage component must be able to support the
inheritance relationship of events. When queries specify parent class events,
derived events have to be returned as well.
The mobile agent subsystem is responsible for providing a communication
system to move the state and the code of agents between different hosts and for
providing an execution environment for them. Additionally, the system has to
provide protection against security risks involved when utilizing mobile code
(see Section 5 for more details). An important task of the agent subsystem is the
provision of a directory service. When agents have to look for event patterns,
they need to access a list of all hosts with an installed agent platform. The
agent platform also provides clock synchronization with a maximum guaranteed
deviation. This is needed to be able to temporally relate events at different nodes.
The user interface allows users to specify queries and claim the results.
The agent launch and query unit initiates the launch of appropriate agents and
provides a way for them to communicate back their results. Queries are written
in a language called Event Query Language (EQL), which we have developed to
conveniently specify patterns that reflect a security violation. This is described
in more detail below in Section 3.1. The user interface itself is realized as a
web interface using HTML and JavaScript on the client side and Servlets on the
server side. The communication between the client and the server is secured
by using SSL connections. This setup allows a user to access the system via a
standard browser from any computer that needs no Sparta components installed.

3.

Pattern Specification

The design of our pattern specification language is guided by two conflicting

goals. The first goal states that the language should be as expressive as possible.

191

Sparta

It would be desirable to allow the description of complex relationships between

events on different hosts using regular or tree grammars. Unfortunately, the
evaluation of complex patterns makes it necessary for each local host to send a
huge amount of data to a central site. This conflicts with the second goal, which
demands that the amount of data that has to be transferred between hosts should
be as small as possible. When a system uses mobile code (i.e. mobile agents),
it should aim at performing flexible computation remotely at the location where
the interesting data is stored instead of abusing agents as simple data containers.
When the interesting patterns do not change frequently, it would be desirable
to wire them directly into local components at each host. For our application,
users intend to specify many different patterns and perform a lot of ad-hoc
queries. Therefore, the application of mobile code is reasonable.
The basic building block of a pattern is a set of local events. One can specify
a list of events on a local host by enumerating them and imposing certain
constraints on their attributes. A constraint can have two different formats.
One format allows to relate an event attribute with a constant value, using one
of the standard logical operators or one of our self-defined ones (in, range).
The other format allows to relate an attribute of one event with another attribute
of the same or a different event, again using the full range of operators. This
allows to select a number of events with a common context. A connection
between events on different hosts is established by connection events.
Definition:
A pattern P, relating events that occur at n distinct hosts, consists of n sets of
events, one for each node. A set of events SA at host A is linked to a set of events

SB at host B, iff SA contains a send event and SB contains the corresponding

receive event. When node A opens a channel to B for data transmission (e.g.
open a TCP connection, send a UDP packet, send an Ethernet frame), a pair of
corresponding events (send at A, receive at BJ is created.

Definition:
Pattern P is valid, iff the following properties hold.

1 Each set of events is at least linked to one other set.

2 Every set except one (called the root set) contains exactly one send event.
The root set contains no send event.
3 The connection graph contains no cycles. The connection graph is built
by considering each event set as a vertex and each link between two sets
as an edge between the corresponding vertices.

These definitions actually only allow tree-like pattern structures (i.e. the
connection graph is a tree), where the node with the root set is the root of the
tree. Although this restriction seems limiting at a first glance, most desirable
situationscan still be described. Usually, activity at a target host only depends on
events that have occurred earlier at several other hosts. This situation can easily
be described by our tree patterns where connection links from those several hosts

192

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

end at the root node. The opposite case, where events on two different nodes
both depend on the occurrence of a single event at a third node is more difficult.
In this case, the connection links do not end at the root node, but have their origin
there. Such a situation cannot be directly expressed in our pattern language (as
the root node set would contain two send events). Nevertheless, an application
might split the original, illegal pattern into subpatterns (each representing a
legal tree like structure) and relate the results itself. This allows to define
arbitrary complex patterns at the expense of performance and network traffic.
The major advantage of the proposed limitation is the possible implementation
of an efficient search algorithm (for details, see Section 4) which transfers as
little data as possible over the network.
Our query language allows to combine pattern specifications with the possibility to extract statistical data. A pattern instance is a set of events that satisfy
the constraints of a valid pattern. Obviously, it might be possible that a single
pattern is satisfied by more than one event set. Two event sets are said to be
distinct, if they contain at least one distinct event element. An event element
can be uniquely identified by its timestamp and the host, where it occurred.
Statistical data can be computed for the set of all distinct instances of a given
pattern. One can obtain the number of elements in that set (i.e. valid instances)
or the maximum or minimum values for the number of instances at each host.
Additionally, one can query attribute values of a certain single event of the
pattern. The sum, maximum or minimum for an attribute may be calculated.

3.1.

Event Query Language

This section describes the syntax and semantics of our Event Query Language
(EQL) in more detail. We omit the complete language grammar, instead we
gradually introduce the language by giving explanations on several examples.
A query is written as follows (similar to SQL).
SELECT results FROM nodes WHERE conditions

The results section is used to define the type of information the user is interested in. The operator COUNT can be used for patterns and returns a list of
all nodes with the number (i.e. count) of found pattern instances at each one.
The operators SUM, MAX and MIN may be used for complete patterns or for an
attribute of a single event. When used for patterns, these operators return the
sum, the maximum or the minimum number of detected pattern instances per
node, respectively. When used for an event attribute, the sum or the extreme
value (maximum/minimum) for a certain attribute value over all instances is
returned.
The nodes section is used to assign an identifier to each node that is later
used in the pattern definition. Additionally, one can impose restrictions on each
node to have the agents only consider a limited set of actual hosts.

193

Sparta
SELECT COUNT FROM host_1 range (10.1.17.0, 10.1.17.255)
#
#

return the number of pattern instances for

10.1.17.* subnet (i.e. has an IP between

each host which is on the

10.1.17.0 and 10.1.17.255

The conditions section specifies the pattern. It consists of a list of event sets,
one for each node that appears in the node section. The event set is a list of
identifiers, each describing an event. In order to be able to specify statistics
operations on event attributes, one can assign identifiers (i.e. a label) to each
of them. Two predefined labels called send and r e c e i v e are used to identify
the send and receive events, respectively, for linking event sets (see Section 3).
Each event can optionally be defined more precisely by constraints on the
events attribute values. These attribute values can be related to constant values
or to variables by standard operators (=, !=, c , > , >= and <= with their
usual semantic meaning) or by a range or an i n operator as defined below.
x range

H zo

(~0~~1)

x in (zo,z~,

...,z,,)

C)

5 z 5 21
3i (0 5 i 5 n) and x =xi

A variable is defined the first time it is used. One must assign a value (bind
an attribute value) to each defined variable exactly once while it may be used
arbitrarily often as a right value in constraint definitions. The scope of variables
is global and its type is inherited from the defining attribute.
With these explanations, we may introduce the syntax (in BNF) of the conditions section (all identifiers represent strings).
conditions
event set
event
constraint
assignment
relation
value
operator
connection
label

: {event set}+
{I
{event)+
)
[connection] event-identif
[ {constraint ; )* 1
assignment I [label] relation
$variable-identifier
I=
( attribute-id I constant )
attribute-id operator
[(] {value , 1* value [)I
constant | $variable-identifier
< I > I I <= I I > = I a I ! = I I in I range
send(target-id):
I receive(source-id):
label-identifier:

: node-identifier
:
:

:
:
:
:
:
:

The following example shows a classical telnet chain pattern that describes a
connection from Node I to port 23 at Node2 and from there to port 23 at Node 3.
Node3 describes the root node set (i.e. has no outgoing send event).
Telnet Chain:
Note1 { send(node2): tcp-connect
[]
Node2 { receive(node1): tcp-accept
[ port = 23; 3
send(node3) : tcp-connect []
1
Node3 { receive(node2):
tcp_accept [ port = 23 ; 1 1

194

4.

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

Pattern Detection

Usually, patterns are discovered by gathering event data from distributed

nodes at a central host where pattern matching algorithms distill the desired
information. Our approach differs from the usual setup by detecting patterns in
a fully distributed way using mobile agents. Mobile agents roam the network
to search for suspicious events to start a more detailed investigation. When an
agent spots the mark of a possible intrusion, it decides which data to carry with
it on its next hop and which place to visit next.
The advantages of such a pattern detection scheme base on the fact, that no
central entity is needed for data correlation. This increases the fault tolerance
and robustness of the system, which is especially important for security relevant
systems. When the host where a centralized IDS performs its correlation is
taken out of action (e.g. by a DoS attack) the detection mechanism is actually
blinded. When an attack renders some hosts in the network unavailable, agents
can still search the remaining ones for signs of intrusions. Even when an attacker
takes over a few hosts and manages to modify the agent platform in a way that
it delivers wrong data (simply bringing it down is suspicious by itself), only
intrusions where parts of the pattern occur at the compromised hosts are not
detectable any more. The remaining system can still detect security violations.
Our approach also improves the scalability of the system because new hosts
on the network wont automatically cause additional traffic to a single existing
server machine. While traditional approaches like hierarchical installations
and redundant servers allow to process more traffic than a single machine, a
distributed approach is still desirable. We think that we can exploit the locality of
network accesses. Most connections in large companies are between machines
of the same department (like references to internal web servers or file shares)
while connections between departments are rare. This allows agents to look for
patterns in small areas and then move on. In a system with a central root node,
all traffic would need to be forwarded (even with prefiltering and reduction over
several hierarchies) to it.
The detection is done in the following way. An agent is started by the user
interface with a given pattern (representing a security violation) that it has to
look for. It starts its task by contacting the directory service to obtain a list of
all hosts with an installed agent platform that match the constraints given in the
patterns FROM clause. These nodes are then visited in arbitrary order.
When an agent arrives at a host, it looks for events that fulfill the constraints
given for the root node of the pattern it is currently investigating. In the case
of the telnet chain introduced above, the agent would have to look for accepted
TCP connections at port 23 (see Step 1 of Figure 2). The result of this process
are a number of events (representing different instances of the pattern) which
satisfy the root node constraint.

Sparta

195

Ncde

Step 2

Step 1

Step 4

Step 3

Figure 2 .

Pattern Detection

When no events are found, the agent immediately continues its journey. Otherwise, all receive events are identified. In the telnet chain case only receive
events exist but more complex patterns are possible. For the distributed detection algorithm, a receive event is important because it establishes a relationship
between nodes which is used to select a promising next place where an agent
should look further. For each receive event, a helper agent is spawned which
follows the link to the host with the corresponding send event (see Step 2 of
Figure 2). By using the send event (determined by the receive event of the
node the agent was coming from) it looks for events which fulfill the current
part of the pattern. In the case of our telnet chain example, the agent already
knows that the send event to Node 3 exists (as it is coming from there) and now
searches for another accepted TCP connection at port 23 (from Node 1).
When the local event set contains receive events itself, the process recursively
repeats by having the agent spawning helper agents and waiting for their return
(see Step 3 of Figure 2). When the helper agents return, they report their
findings (i.e. pattern instances) back to the agent waiting at the originating
node (helper agents only move over a single hop). When pattern instances are
returned, the waiting agent processes them (e.g. match variables or perform
statistical evaluation) and eventually continues. When all helper agents have
returned a pattern might be detected by the agent waiting at the root node as all
information is available (see Step 4 of Figure 2).
Usually, only a small amount of data has to be transferred as it is not necessary
to transport all pattern instances themselves but merely time stamps or single
attribute values which have been assigned to variables.
Variables are treated in the following way. When a variable has already
been bound to a value, it is straightforward to use this value directly for the

196

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

attributes constraint. This is the case when a value is assigned to a variable

at a node which is closer to the root of the connection graph than the node,
where it is used. On the other hand, when a constraint depends on a variable
which has not been resolved yet, one has to temporary ignore that constraint.
First, all actual bindings of the variable have to be determined and are then
matched against the instances where the variable has not been assigned yet.
This allows to filter out pattern instances which do not satisfy the previously
ignored constraint. Notice that this may cause agents to transfer unnecessary
pattern instances but the runtime complexity is still linear with regard to the
number of send/receive events.
Instead of having a single agent visit all nodes sequentially, the task could
be parallelized easily by partitioning the set of interesting hosts. Each partition
is visited by a dedicated agent which all have to agree on a destination node,
where they meet and merge their results.
We have installed a first prototype version at our departments network as
universities are traditionally favorite targets of hackers. We are currently able to
detect about a hundred local events (by looking for well known attack signatures
and network connections) and a dozen distributed patterns. The results are
promising as a couple of incidents have already been detected. The network
overhead of the traveling agents is negligible and the processing overhead at
each node is reasonably low.

5.

Security

Mobile code introduces a number of security issues that our design has to
deal with. Especially when building systems for security sensitive applications (like our intrusion detection system), it is important not to introduce new
vulnerabilities by the security monitoring tool itself. The security threats to
mobile agents are classified by four categories, namely agent-to-agent, agentto-platform, platform-to-agent and other-to-agent.

Agent-to-agent threats describe the set of attacks, where one agent exploits the vulnerabilities of another agent. In Sparta, agents only locally
communicate with helper agents they have previously spawned. As arbitrary interagent communication is prohibited, possible vulnerabilities
cannot be exploited and agent-to-agent attacks can be prevented.
Agent-to-platform threats describe attacks, where an agent performs malicious activities against its environment (i.e. platform). To prevent these
kind of attacks, the runtime permissions of agents are rigorously restricted. They are not allowed to access resources directly. Instead, agents
gain information by querying the data storage component.
Platform-to-agent threats describe situations, where the platform compromises the agents security. This sort of threats is extremely difficult to

197

Sparta

defend against when agents need unrestricted movement around the network (Jansen and Karygiannis, 1999) and Sparta has no special mechanisms to defend against such attacks. In contrast to a central server system,
we still have the advantage that even if a single node is compromised, all
patterns which do not touch this host are still detectable.

Other-to-agent attacks involve threats against agents performed by external entities while they are in transit over the network (e.g. eavesdropping or tampering). Sparta uses an asymmetric (public/private key
pair) cryptosystem to secure agents when they are transferred over the
network. The agent code is signed and can be authenticated before it is
executed. In order to manage the asymmetric cryptosystem, a Public Key
Infrastructure (PKI) is provided.

6.

Related Work

The idea of correlating events which occur at different places in a network

and to formalize patterns to describe such correlations is not new. The Complex
Event Processor which is developed at Stanford University is capable of correlating causally and temporally related events. It bases on the theory of partial
ordered multisets (Pratt, 1986) and is used for intrusion detection (Perrochon
et al., 2000) and network management (Perrochon et al., 1999). Patterns are
described using the Rapide Pattern Language (RAPIDE, 1997). The difference
between our approach and their work is the fact, that we use mobile agents to
perform the pattern detection in a distributed fashion without any central server.
In contrast to that, they collect data from different client sites and process it at
a server. Rapide has clearly influenced our work as their system as well as ours
try to correlate generic events and target a broad spectrum of applications.
State-of-the-art ID systems like EMERALD (Porras and Neumann, 1997),
NStat (Kemmerer, 1997)or AAFID (Balasubramaniyan et al., 1998) can gather
and relate data from different sources. In contrast to our distributed design, they
have a hierarchical architecture where sensors located at different hosts collect
data and send it to a central entity where events are related. Unlike Rapide
and our design, they completely focus on intrusion detection events and are
not applicable to different domains. The same is true for network management
software (e.g. HP OpenView (Sheers, 1996)).
Commercial intrusion detection systems IikeNetwork Flight Recorder (NFR,
2001) or Real Secure (RealSecure, 2001) perform their analysis on packet level
by monitoring network traffic. This allows only simple correlation, but their
output can be used as our basic events.
IDA (Asaka et al., 1999) uses mobile agents to trace a possible attacker back
to its origin, while Micael (de Queiroz et al., 1999) pursues a more ambitious
aim where each system component is realized as a mobile agent. Unfortunately,

198

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

only a high-level system design has been presented. The possible advantages of
mobile agents in intrusion detection systems are summarized in (Jansen et al.,
1999) and (Kriigel and Toth, 2001).

7.

Conclusion

Relating distributed events and deducing knowledge from different hosts is

especially important in the field of network management and intrusion detection.
We present a solution, where mobile agents perform the task of correlating
data in a fully decentralized manner. In order to prevent a tremendous increase
in network traffic, the expressiveness of our pattern description language had
to be slightly restricted. This allows an efficient detection algorithm and a fault
tolerant and scalable system design.

References
Asaka, M., Taguchi, A., and Goto, S. (1999). The implementation of ida: An intrusion detection
agent system. In Proceedings of the 11th FIRST Conference.
Balasubramaniyan, J. S., Garcia-Fernandez, J. O., Isacoff, D., Spafford, E., and Zamboni, D.
(1998). An architecture for intrusion detection using autonomous agents. In 14th IEEE Computer Security Applications Conference.
de Queiroz, J. D., da Costa Carmo, L. F. R., and Pirmez, L. (1999). Micael: Anautonomousmobile
agent system to protect new generation networked applications. In 2nd Annual Workshop on
Recent Advances in Intrusion Detection.
Jansen, W. and Karygiannis, T. (1999). Mobile agents and security. Special Pub. 800-19, NIST.
Jansen. W., Mell, P., Karygiannis. and Marks, D. (1999). Applying mobile agents to intrusion
detection and response. Interim Report (IR) 6416. NIST.
Kemmerer, R. A. (1997). A model-based real-time network intrusion detection system. Technical
report, Computer Science Dep., University of California Santa Barbara, November.
Krgel, C. and Toth, T. (2001). Applying mobile agent technology to intrusion detection. In
ICSE Workshop on Software Engineering and Mobility.
NFR (2001).Network Flight Recorder. http://www.nfr.net/.
Perrochon. L., Jang, E., and Luckham, D. C. (2000). Enlisting event patterns for cyber battlefield
awareness. In ARPA Information Survivability Conference and Exposition (DISCEX'00).
Perrochon, L., Kasriel, S., and Luckham, D. C. (1999). Managing event processing networks.
Technical Report CSL-TR-99-877, Stanford Computer Systems Laboratory.
Porras, P. A. and Neumann. P. G. (1997). Emerald: Event monitoring enabling responses to
anomalous live disturbances. In Proceedings of the 20th NISSecurity Conference.
Pratt, V. (1 986). Modelling concurrency with partial orders. Int. Journal of Parallel Programming,
15(1):33-7 1.
RAPIDE (1997). Rapide 1.0 Pattern Language Reference. Stanford University.
_
RealSecure (2001). Realsecure. http://www.iss.net/customer_care/resource_center/product_lit/.
Sheers, K. R. (1996). HP OpenView Event correlation. Hewlett-Packard Journal.
Staniford-Chen, S., Cheung, S.. Crawford, R., Dilger, M., Frank, J., Hoagland, J., Levitt, K., Wee,
C., Yip, R., and Zerkle, D. (1996). Grids - a graph based intrusion detection system for large
networks. In Proceedings of the 20th National Information Systems Security Conference,
volume 1, pages 361-370.

PART TWO
Invited Papers

This page intentionally left blank.

SHELLS TRUST DOMAIN

INFRASTRUCTURE SECURITY
CERTIFICATION

Linking security management to business objectives

Pieter van Dijken
Shell Services International
[email protected]

Abstract

Shell companies worldwide completed in 2000 a security programme,

covering certification of their IT infrastructure against a subset of ISO
17799 and related ISO standards on certification and audit. Objective
was to provide the Shell Group with a secure environment to do (e)
business in, i.e. sharing of knowledge, enabling support for global applications and supporting virtual teamworking. The scheme is now u p
and running in more than 160 countries and 240 Shell companies. In this
presentation I will describe background considerations on the Scheme as
an example of business linked information security management. I will
go as well into practical issues regarding roll-out and implementation of
a global scheme like this. I will conclude with outlook for the Scheme,
planned activities and issues.

Pieter van Dijken (51, Dutch) manages the global information security consultancy team in Shell Services International. This team
operates from locations around the world in support of the information security requirements of the Royal Dutch/Shell Group
of companies. His team facilitated very recently a strategic, world
wide security certification programme, called Trust Domain. Objective of this programme is to establish a common set of information security standards and controls for IT infrastructure throughout the Shell Group, based on a subset of the BS 7799 standard.
Benefits of having such a common set are numerous, e.g. facilitate information sharing across the Shell Group without making
Shell companies vulnerable to unauthorised access; having lower
cost and more capabilities through the use of standard security

202

ADVANCES IN NETWORK AND DISTR. SYSTEMS SECURITY

protocols and tools and finally by avoiding unnecessary security
controls between Shell companies.
Pieter joined Shell in 1988 and has been involved in numerous international policy and standardisation efforts with regard to trust
and confidence in IT since. He was directly responsible for the
translation of BS 7799 in Dutch, took part or chaired a host of
related initiatives (e.g. the Dutch BS 7799 certification Scheme
and many others). Pieter has degrees in business law and police
administration and published on criminal justice implications of
IT. He lives in the Netherlands with his wive and three children.

AUTHOR INDEX
Alexandris, Nikos
Bonnet, Aurlien
Borselius, Niklas
Chrissikopoulos, Vassilios
Claessens, Joris
De Decker, Bart
De Win, Bart
Foster, Nathalie
Hwang, Jing-Jang
Jacob, Jeremy
Jrjes, Jan
Kirda, Engin
Kisimov, M.V.
Krgel, Christopher
Lin, Chih-Yin
Lobelle, Marc
Magkos, Emmanouil
Mitchell, Chris J.
Naessens, Vincent
Preneel, Bart
Pudovkina, Marina
Sakurada, Hideki
Toth, Thomas
Tsukada, Yasuyuki
Vandewalle, Joos
van Dijken, Pieter
Vanhaute, Bart
Von Solms, S.H.
Wilson, Aaron
Wu, Tzong-Chen

33
139
173
33
153
93, 125
125
75
45
75
107
187
15
187
45
139
33
173
93
153
61
1
187
1

153
20 1
93, 125
15
173
45