Security Property Modeling
Hiba Hnaini
1
, Luka Le Roux, Joel Champeau and Ciprian Teodorov
Lab STICC, SL Department, ENSTA Bretagne, Brest, France
Keywords:
Cyber-security, Modeling, Attacker, Methodology, Formal Methods, Model-checking, Property Specification,
Case Study.
Abstract:
With the increasing number of cyber-attacks on cyber-physical systems, many security precautions and so-
lutions have been suggested. However, most of these solutions aim to prevent the access of an adversary to
the system. Though, with the increasing number of elements used in a system, and thus vulnerabilities, it is
essential to study the risks introduced to the system to make the system itself efficient enough to react to the
attacks once an attacker has obtained access. Analyzing and discovering the risks is the first step to making
the system more resilient.
This paper proposes a methodology that combines the qualitative risk analysis with formal methods ( model
checking ) to identify the risks that were not recognized during testing or functional modeling phases.
To examine this methodology, a car reservation system is modeled with an attacker, and then its security prop-
erties are verified using UPPAAL model checking tool. As a result, some risks were identified and tested for
the possibility of them occurring and their effects on the system.
1 INTRODUCTION
The rapid advancement and progress of the Informa-
tion Technology sector over the past years has made
it possible to integrate technology in almost all do-
mains. Some of these domains require systems with
solid coordination between hardware and software or
physical and cyber elements, known as cyber-physical
systems. They are present in fields such as transporta-
tion, health care, and infrastructure. Due to their im-
pacts on the real world these systems form an appeal-
ing target to cyber attackers.
A cyber-physical system is composed of many el-
ements, each with its own weaknesses or vulnerabil-
ities, all of which are inherited by the system, thus
leading to a lot of opportunities for potential attacks.
To address this issue, several approaches have been
proposed (Wardell et al., 2016).
A multiple protective layer or “Defense in Depth”
strategy has been proposed to handle such threats
(Wardell et al., 2016). Regardless of the essential
role this strategy has in system protection, it only pro-
vides preventive measures that help control the ac-
cess to the system and its components. Previous work
around the use of formal methods in cyber-security
focuses on uncovering critical vulnerabilities in a sys-
tem (Mitchell and Bau, 2011; Yu et al., 2016; Wardell
et al., 2016; Kalaie et al., 2015). However, we be-
lieve it is important to use formal methods to unveil
the risks as well as the vulnerabilities. The qualitative
risk analysis approach may be more challenging, but
it helps render the system more resilient if an adver-
sary is to gain access to the system.
This paper proposes a methodology that uses for-
mal methods (model checking) and qualitative risk
analysis to identify the risks on the system after the
adversary has acquired access.
The proposed methodology is applied to a car
reservation system to discover the risks present in the
system. The case study presented in this article shows
the effectiveness of this methodology over other ones.
The paper is structured as follows. Section 2
overviews some related works. Section 3 introduces
a motivating example from the automotive domain.
Section 4 presents the proposed security modeling
methodology. Section 5 discusses the results obtained
on the car reservation system. Section 6 concludes
this study emphasizing open research questions.
2 RELATED WORK
This study is linked to work done on qualitative
risk analysis of cyber-physical systems using formal
methods based on model checking.
694
Hnaini, H., Roux, L., Champeau, J. and Teodorov, C.
Security Property Modeling.
DOI: 10.5220/0010388206940701
In Proceedings of the 7th International Conference on Information Systems Security and Privacy (ICISSP 2021), pages 694-701
ISBN: 978-989-758-491-6
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
A risk in cyber-security is known as a potential
loss or damage of an asset due to a threat exploiting a
vulnerability. Where an asset is a resource to be pro-
tected, a threat is anything that can exploit a weakness
and cause damage to an asset, and a vulnerability is a
weakness or gap in the security or protection of the
system(Kbar, 2008).
A system with no threat is not vulnerable. The
vulnerabilities of a system appear only in the exis-
tence of a threat. Also, a secured system or a sys-
tem that is not vulnerable to an activity representing
a threat is not threatened. Then, a threat is not con-
sidered as a genuine threat if the system is immune or
not vulnerable to its actions(Taveras, 2019).
Thus, a risk is the coexistence of a vulnerability
and a threat(Taveras, 2019). To reveal risks in a vul-
nerable system, a threat needs to be introduced.
Then, to test the possible impact of the risk, risk
assessment is used. There are two types of risk as-
sessment, qualitative and quantitative. In the quanti-
tative method, risk characterization produces numer-
ical estimates of the risk. Whereas in the qualita-
tive approach, the estimates are non-numerical. There
are two primary functions of qualitative risk assess-
ment: ”risk identification” and ”risk characterization
and analysis”. The first function, risk identification,
is to find, recognize, and describe the risks in natural
language or a narrative manner. The second function,
qualitative risk analysis, is the risk characterization
that generates a non-numerical evaluation of the risk
(US Army Corps of Engineers, 2018).
A way of evaluating risks is by using formal meth-
ods. Formal methods are built on consistent math-
ematical fundamentals. Formal verification is “the
act of proving or disproving the correctness of un-
derlying system algorithms concerning certain for-
mal specifications using formal methods of mathe-
matics.”.(Akhtar, 2014)
One type of formal methods is model-checking,
in which a system’s state space is extensively pro-
cessed. Model-checking tackles finite-state systems,
but in complex systems, state-explosion can cause a
limitation. However, errors that are undetected us-
ing testing and simulation are discovered by model
checking.(Akhtar, 2014)
As shown in figure 1 (Christel Baier, 2008),
model-checking necessities the existence of a sys-
tem model derived from the actual system and formal
properties deduced from the system requirements,
known as safety properties. Then, the properties are
evaluated on the system model to give two outcomes:
satisfied or violated.
A property is a statement that describes what the
system should or should not do. It translates a require-
Figure 1: Model-Checking Representation.
ment from natural or informal language into a formal
expression, whereas a model describes how the sys-
tem behaves.
A model-checker examines all pertinent states in
a system model to validate if they meet a property. A
counter-example of the property is given if it is not
satisfied, showing how the undesired state is reached
(As seen in figure 1)(Christel Baier, 2008).
Model-checking has many domains of applica-
tion, such as verifying military systems, aircraft
reservation systems, spacecraft, industrial control
systems, and storm surge barriers, as mentioned
in(Christel Baier, 2008).
There are different approaches to model a sys-
tem, and thus the meaning of the defined properties
changes accordingly. For example, a property in a
safety-critical system, such as an airplane, a car, or a
medical facility ensures the safety of the user where a
malfunction can cause a tragedy(Hsiung et al., 2007).
However, a property in network protocols security in-
tends to check if the protocol is secure against security
threats. (Basin et al., 2018)
There are also different purposes for using prop-
erties. The most common one is to verify that the
system satisfies all the requirements. Another is to
find the problems in a model of an algorithm, sys-
tem, protocol, etc..., and propose solutions that handle
these problems one by one(Lamport, 2000). However,
Security Property Modeling
695
these are not our objectives as we are defining proper-
ties to find security risks in the existing model, similar
to (Mitchell and Bau, 2011; Yu et al., 2016; Wardell
et al., 2016; Kalaie et al., 2015).
We can describe our work as close to (Mitchell
and Bau, 2011; Yu et al., 2016). However, our work
assumes that the adversary has all the knowledge to
access the system and studies the attack’s effects. Un-
like (Mitchell and Bau, 2011), that experiments if the
attacker can gain access based on protocol analysis.
(Yu et al., 2016) is very specific to the online shopping
business, whereas our methodology is more general.
There are many works around model checking and
cybersecurity analysis (Mitchell and Bau, 2011; Yu
et al., 2016; Wardell et al., 2016; Kalaie et al., 2015).
However, most of them focus on model-checking
techniques and tools, but not on properties modeling
related to the qualitative risk analysis method.
Through our research, we have found many pa-
pers related to qualitative risk analysis. We have also
found papers addressing formal methods. However,
to our knowledge, none of the papers found addressed
both at the same time. And thus, this paper presents
a methodology that targets using formal methods in
qualitative risk analysis, with emphasis on properties
modeling.
3 MOTIVATING EXAMPLE
This section introduces the case study used through-
out this paper to illustrate the proposed approach from
its concepts to expected outcomes. The Car Reserva-
tion Embedded System (CRES) assumes a provider
company (that owns the cars) and users (who are
equipped with a personal badge). Users can book a
car from the company through a website, then unlock,
drive, and return that car using their badge. To make
it possible, the cars are equipped with a system that
knows the bookings, controls the access to the car,
and can be interacted with using a badge.
From a cyber-security perspective, such a system
is a desirable target for malicious adversaries and ex-
poses a large attack surface. In this paper, we focus on
one car, its embedded system, and the environment.
In particular, we suppose that the online booking web-
site is secure.
This paper focuses on security analysis given a be-
havioral model and abstract specification of the ad-
versary goal and capabilities. Those are considered
entry-points to our contributions. Sub-section 3.1 pro-
vides a more detailed introduction to the system under
study. Sub-section 3.2 introduces its modeled behav-
iors.
3.1 System Presentation
To discover the functionalities and behaviors of the
system, we have studied a car reservation embed-
ded system. Its original specifications include desired
nominal scenarios and (C) source code.
Figure 2: System Architecture.
The entire system is composed of a system or
computer embedded in the car, a server or back of-
fice, and an ID badge for each user.
The reservation is first made and added to the
computer in the car. Then, the booking is started. The
user can now lock the car and resume the session later
or request to end the reservation if the trip has ended.
If the session is not ended, the user can unlock the car
again to continue the session.
The embedded system interacts with one compo-
nent at a time, either the server or the ID badge.
The driver uses the ID badge or user to lock and
unlock the vehicle, whereas the owning reservation
company uses the server to control the car.
These components (the embedded system in the
car, the server, and the badge) are connected through
many technologies and protocols. This signifies that
the system inherits all the vulnerabilities and weak-
nesses of these technologies and protocols. Adver-
saries can exploit them to target or access the system.
When applying the methodology discussed in sec-
tion 4, an interesting target in our system under study
(Car Reservations System) would be, for example, a
car lock. The feared events would be that the attacker
might want to keep the car locked or unlocked accord-
ing to the desired outcome.
3.2 System Model
As a first step, the system under study is modeled by
identifying the behaviors.
To obtain the formal model of the system, a func-
tional model is derived from the textual specification
of of the system, as explained in (Mitchell and Bau,
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
696
2011). A high-level of abstraction is considered since
the aim of the formal model is to check the behavioral
functions of the system.
Figure 3 illustrates the behavior of the software
deployed on the car in the Car Reservation System.
The figure shows the different states the system can
have, with the transitions from and to each state. The
states, represented by the circle, are the system’s con-
ditions after or before a transition. One state the sys-
tem can have is idle. It is the initial state of the sys-
tem that shows that there is no ongoing process or
reservation. Another state is ”Car Unlocked” which is
reached after receiving a Car Unlock or Start Reser-
vation message. This state shows that there is an on-
going reservation or that the car is currently unlocked
and that the engine can now be started. The other
states are Reserved, Car Locked, Ongoing Session,
Res End Requested, and Verify END Session.
The transitions are the messages exchanged be-
tween the different components of the system to com-
mand it to act in a certain way. The transition mes-
sages with a question mark (?) mean that the system
is waiting to receive this message from another com-
ponent (Back Office, Mobile App, Badge, etc.). The
other transition messages with an exclamation mark
(!) represent the ones sent by the system to another
component after receiving the expected message. The
other labels are guards to verify some system condi-
tions.
The first step is for the server to send an Add
Reservation request to the system embedded in the
car. Then, the reservation is initiated by the Start
Reservation request, where the car is unlocked, and a
driving session is started. The car at this stage can be
locked using a Lock Car request for an intermediate
stop. Then, the car is unlocked, using the Unlock Car
request, and the session resumes. The session can be
finished using the END RES Request. Then, the car
is locked using the Lock Car request, and it goes back
to idle after verifying the conditions of the car lock
and the network connection. Also, after each request,
a confirmation message is sent to the server.
Figure 4 shows an abstract model of the server
which, based on the requests previously define, com-
municates with the system embedded in the car.
We show another model in Figure 5. It represents
using an ID badge to control the car lock using the
Lock Car and Unlock Car requests. Then, the entire
model, with its different components, is verified to en-
sure the well functioning of the system.
Some of these functional requirements are:
- If the car is unlocked by the server or the ID, the
state in the server should also be car unlocked.
- If the car is locked by the server or the ID, the
state in the server should also be car locked.
- If the car is in the Ongoing Session state, the
state in the server should be car unlocked (In service
of the user).
- If the car is in the reservation end requested state,
the state in the server should be car unlocked.
- If the car is in the Verify end session state, the
state in the server should be car locked.
Using the methodology presented in Figure 1,
these functional requirements have been encoded in
queries and verified on this model. However, this
level of specification lacks the conditions necessary
for identifying and formalizing the security require-
ments. To achieve this, we identify the main method-
ological tasks needed to formalize the security re-
quirements, and we integrated them into the model-
checking methodology. The experimental section il-
lustrates these in the car-sharing system.
4 SECURITY MODELING
METHODOLOGY:
CONTRIBUTION
We adopt the Qualitative Risk Analysis approach in
our modeling methodology. The model comprises 3
principal components, the system, the attacker, and
the properties.
The aim of having a model with these three com-
ponents is to produce the co-existence of a vulnerable
system and a threat to reveal the risks.
The attacker model is considered being a threat.
It is used to uncover the risk in the system, which is
assumed to be vulnerable. Then, the properties which
represent the risks are checked on the entire model.
This serves to use model-checking in qualitative risk
analysis. This is represented in figure 6, where an at-
tacker model is added to figure 1. The figure shows
how the attacker is integrated into model checking.
The attacker model is composed of the system model
using an attack composition operator, and the attacker
interests are combined with the system requirements
to specify the properties. Then, the combined proper-
ties, requirements, and attacker interests are verified
on the composed model: system and attacker. This
results in identifying the genuineness of the risks and
if there is a possibility of them happening.
4.1 Attacker Model
After the system is modeled and verified, the vulner-
abilities, that form potential targets of the attacker in
Security Property Modeling
697
Figure 3: System Model (General Representation).
the system, are identified. By considering the adver-
sary’s perspective as in (Mitchell and Bau, 2011) and
(Yu et al., 2016), it is made possible to recognize some
behaviors that form appealing targets for the attacker.
The attacker is then modeled, representing the threat
to the system to exploit these vulnerabilities. This
step aims to reveal the risks on the system when an
attacker attacks.
An attacker is modeled assuming that they have
initial knowledge of the messages transmitted be-
tween the different components of the system. Thus,
the adversary is given the ability to manipulate system
states similarly to the legitimate system.
The attacker can be given all capabilities to manip-
ulate the system at any point in the execution. This
type of attack model can help discover all potential
attack scenarios (Figure 7). This method aims to as
many risks as possible in the system by introducing
an attacker that carries most of the threats that can
threaten the system. Also, this method is employed
because there is no ultimate way of modeling a single
attack scenario. And this is because there is no way
to be sure of the intentions of the attacker.
However, in larger systems, this can cause a state
explosion problem, so there are too many states and
scenarios. This makes large models impossible to ex-
plore exhaustively. A solution to this problem would
be to break down the attacks into several scenarios
where the attacker is not given full capabilities, but
only the ones enough to achieve a certain attack sce-
nario.
This can also be resolved using the Partially
Bounded Context-Aware Verification technique dis-
cussed in (Roux and Teodorov, 2019). This approach
aims to decompose the state-space and the verifica-
tion problem using environment-based guides that can
help select or remove attack scenarios.
Another approach is to use the scenarios to limit
the capabilities of the attacker and observe the effects
on the system. Or to use the scenarios to observe how
much damage an attacker with minimal capabilities
can cause. This approach helps identify the resistance
a system has to a different type of attacks represented
by different scenarios.
Another important aspect to be modeled in the
context of model checking and qualitative risk analy-
sis is the properties. They allow to verify the behavior
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
698
Figure 4: Server Model (General Representation).
Figure 5: ID badge Model (General Representation).
or functioning of the system in it is a normal case or
when it is under an attack. Section 4 discusses how
the properties are modeled and the link between them
and risk identification.
4.2 Security Properties
After determining the potential targets, the attacker’s
goals, or the feared events as defined in the EBIOS
methodology(Fortat et al., 2015), are translated into
properties. Each property represents one feared event.
Verifying the properties helps determine if the attack
issued on the system has been successful or not and
whether the system is resilient to this attack. The
properties check if a certain attack on the system
reached the feared event.
Thus, properties represent risks on the system. A
system would want to prevent the risk, but an adver-
sary would want the opposite, as in Table 1.
According to which perspective is considered,
there are two ways to write the security properties.
The first way is to take the system’s point of view
and create properties that ensure the well function-
ing of the system during or after an attack. After
the model is simulated and the property is checked,
if the property verification is successful, the system is
resilient to this attack and its risks. However, if the
property verification fails, the system is faulty, or the
attack issued was successful in changing the original
behavior of the system and thus reaching the risk.
Another way to write the properties is to consider
the attacker’s perspective. Here, properties ensure
that the goals of the attack or risks are reached. In
Security Property Modeling
699
Figure 6: Model-Checking with Attacker.
contrary to the previous case, where the system’s per-
spective was considered, the successful verification of
a property shows that the system is faulty, or that the
attack has successfully reached its goals of altering
the system’s behavior. This means that the attack has
introduced a risk to the system. However, the failed
verification of the property implies that the attack was
unsuccessful, and the system is resilient to the attack
the property describes. Thus, there is no risk caused
by this threat.
Table 1 summarizes the security property verifica-
tion from both perspectives (system and attacker).
From a system’s perspective, (in table 1) it is to be
made sure that the potential targets or assets are not
put at risk by the attacks or the threats.
However, from an adversary’s perspective, it is
the opposite. The properties, in this case, are written
to verify that the threats on the targeted assets were
successful in creating a risk by controlling, altering,
blocking... the target or the entire system.
To cover all the risks or properties, the methodol-
ogy can adopt both perspectives.
5 APPLICATION TO THE
MOTIVATING EXAMPLE
In this case-study, the attacker is given all possible ca-
pabilities to manipulate the system at any point in the
simulation. We have adopted this method because our
aim of modeling the attacker is not to abstract generic
behaviors based on classified attacks, but to reveal all
the risks on the system. This is represented in Figure
7 where the attacker can send all messages that can be
received by the car’s computer model (Figure 3).
Figure 7: Attacker Model (General Representation).
In our model, we are considering the system’s
point of view to construct the properties since we aim
to verify the well functioning and the resilience of the
system during or after an attack.
Considering the previously given example of a po-
tential target or targeted asset, the car lock, the follow-
ing property can be deduced:
- If the car is unlocked, it should be able to go back
to idle and be booked again.
This means that the attacker, representing the
threat, should not be able to keep the car unlocked
after the session has ended. Keeping a car unlocked
would cause the risk that the car becomes out of ser-
vice and cannot be booked again.
Then, the property is translated from natural lan-
guage into a query to be verified upon simulation of
the model.
6 CONCLUSION
The use of many components and protocols in one
system increases the number of vulnerabilities. This
is because the system inherits all of the vulnerabilities
of each element and protocol added. Even though re-
searchers have developed a multiple-protective-layer
strategy and other security protocols, these solutions
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
700
Table 1: Security Property Modeling.
* System Perspective (+) Attacker Perspective (-)
Property Successful (+) Threat causes no risk (+) Threat causes risk(-)
Property Failed (-) Threat causes risk(-) Threat causes no risk (+)
have focused on preventing the adversary from gain-
ing access to the system.
However, with the many vulnerabilities present in
a system, one must assume that the attacker is to gain
access at some point. And thus, in the presence of a
vulnerability and a threat, the attack will introduce a
risk on the system by changing its normal behavior.
This paper has introduced a methodology that em-
ploys formal methods and model checking to help
discover these risks through qualitative risk analysis,
where uncovering the risks is the first step to making
a system itself more resilient to attacks. This method-
ology was applied to a study case composed of a Car
Reservation System. As a result, we identified risks
in the system, such as the attacker could keep the car
locked or unlocked.
In the study case, the adversary is given full ca-
pabilities to attack the system. As for future work,
these capabilities are to be broken down into scenar-
ios. This approach helps determine the extent of dam-
age or risks an attacker with minimal or different ca-
pabilities can cause. Also, scenarios that attack all the
elements of the model are to be added.
ACKNOWLEDGMENTS
This project has been supported by the French Direc-
torate General of Armaments (DGA), the European
Regional Development Fund (ERDF) of the EU, the
Brittany Region, the Departmental Council of Fin-
ist
`
ere and Brest M
´
etropole as part of the Cyber-SSI
project within the framework of the Brittany 2015-
2020 State-Region Contract (CPER).
REFERENCES
Akhtar, N. (2014). Requirements, formal verification and
model transformations of an agent-based system: A
case study. Computer Engineering and Intelligent Sys-
tems - IISTE, 5:1–16.
Basin, D., Cremers, C., and Meadows, C. (2018).
Model Checking Security Protocols, pages 727–762.
Springer International Publishing, Cham.
Christel Baier, J.-P. K. (2008). Principles of Model Check-
ing. International series of monographs on physics.
The MIT Press.
Fortat, F., Laurent, M., and Simatic, M. (2015). Games
based on active nfc objects: Model and security re-
quirements. In 2015 International Workshop on Net-
work and Systems Support for Games (NetGames),
pages 1–3.
Hsiung, P.-A., Chen, Y.-R., and Lin, Y.-H. (2007). Model
checking safety-critical systems using safecharts.
Computers, IEEE Transactions on, 56:692–705.
Kalaie, A., Bhatia, S., Koutsoukos, X., Stouffer, K., Tang,
C., and Candell, R. (2015). Towards a systematic
threat modeling approach for cyber-physical systems.
volume 2015.
Kbar, G. (2008). Security risk analysis for asset in relation
to vulnerability, probability of threats and attacks. In
2008 International Conference on Innovations in In-
formation Technology, pages 668–672.
Lamport, L. (2000). The mutual exclusion problem part ii:
Statement and solutions. Journal of the ACM.
Mitchell, J. C. and Bau, J. (2011). Security modeling and
analysis. IEEE Security & Privacy, 9(03):18–25.
Roux, L. and Teodorov, C. (2019). Partially Bounded
Context-Aware Verification, pages 532–548.
Taveras, P. (2019). Cyber risk management, procedures and
considerations to address the threats of a cyber attack.
US Army Corps of Engineers (2018). Risk assessment -
quantitative methods. In Corps Risk Analysis Online
Training Modules, pages 1–49.
Wardell, D. C., Mills, R. F., Peterson, G. L., and Oxley,
M. E. (2016). A method for revealing and address-
ing security vulnerabilities in cyber-physical systems
by modeling malicious agent interactions with formal
verification. Procedia Computer Science, 95:24 – 31.
Complex Adaptive Systems Los Angeles, CA Novem-
ber 2-4, 2016.
Yu, W., Yan, C. G., Ding, Z., Jiang, C., and Zhou, M.
(2016). Modeling and verification of online shopping
business processes by considering malicious behavior
patterns. IEEE Transactions on Automation Science
and Engineering, 13(2):647–662.
Security Property Modeling
701