Effectiveness of Trust Reasoning for Attack Detection in
OLSR
Asmaa Adnane
1
, Christophe Bidan
1
and Rafael T. de Sousa Jr
2
1
SUPELEC, SSIR team (EA 4039), 5 Av de la Boulaie, 35510-Cesson-S´evign´e, France
2
University of Bras´ılia - SSIR team (EA 4039)
Av L3 Norte - FT - ENE, 70910-900 Bras´ılia, Brazil
Abstract. Previous works [2,3, 6] have proposed to check information consis-
tency and to detect misbehavior nodes for the OLSR protocol based on semantic
and trust properties. The basic idea is that each node uses only local observations
to detect attacks without having to collaborate with other nodes. The objective of
this paper is to prove the effectiveness of such approaches by presenting simula-
tion results.
1 Introduction
Several studies have been carried on to secure protocols for ad-hoc networks, where
nodes communicate directly with each other to relay messages without the support of a
central entity. We are interested by the Optimized Link State Routing protocol (OLSR)
[5]. This protocol presents an optimization of the classical link state algorithm adapted
to the requirements of a mobile wireless LAN. The concept used in the protocol is
the multipoint relays (MPRs). MPRs are nodes which broadcast messages during the
flooding process. This method substantially reduces the message overhead as compared
to a classical flooding mechanism, where every node retransmits each message when it
receives the first copy of the message.
In ad-hoc networks, the establishment of the routing table is associated to a pro-
cess of trust construction through cooperation among nodes for discovering neighbors,
selecting routers and announcing topology information. Thus, each node has to verify
the expected scheduling and content consistency of protocol messages, enabling the
mistrust of the other misbehavior nodes during this process [2,3].
In previous works [2,4] we have proposed for OLSR the integration of semantics
checking and trust reasonings into each node, so as to allow a self-organized control
to help nodes to detect attacks about modification of OLSR control messages. In ad-
dition, our proposal does not change the OLSR protocol and is compatible with the
bare OLSR. The objective of this paper is to prove the effectiveness of our approach to
the integration of trust reasoning in OLSR protocol to detect attacks by presenting the
simulation results.
The paper is organized as follows. Section 2 presents notations and the trust specifi-
cation language. Section 3 summarizes our previous works. Synthesis and simulation of
Adnane A., Bidan C. and T. de Sousa Jr R. (2008).
Effectiveness of Trust Reasoning for Attack Detection in OLSR.
In Proceedings of the 6th International Workshop on Security in Information Systems, pages 151-157
DOI: 10.5220/0001737501510157
Copyright
c
SciTePress
our approach are presented in section 4. Finally, we conclude in Section 5, and present
our future works.
2 Notations
In OLSR, each node maintains its local vision of the network. This vision consists in
the following sets:
MANET : the set of the whole MANET nodes,
NS
X
(Neighbor Set): the set of symmetric neighbors of the node X,
2HNS
X
(2-Hop Neighbor Set): the set of 2-hop neighbors of the node X,
MPRS
X
: the set of nodes selected as MPR by the node X (M P R
X
NS
X
),
that is the nodes that are in charge of routing and forwarding the packets sent by X.
MPRSS
X
(MPR Selection Set): the set of symmetric neighbors which have se-
lected the node X as MPR (MP RSS
X
N S
X
),
RT
X
(Routing Table): the routing table of the node X,
The node collects the information needed to maintain its local vision of the network by
exchanging HELLO and TC messages. For these messages, we note:
X
HELLO
Y , X
T C
Y
Y : respectively, the reception by node X of HELLO and
TC messages from node Y (HELLO = LS
Y
and T C
Y
= M P RSS
Y
),
X
T C
X
, X
DAT A
X
: The broadcast by X of a TC or respectively a data message
to be forwarded by its MPRs.
X
T C
Y
8 Y : absence of an awaited TC message from node Y ,
X
DAT A
X
8 Y : supposing that Y is MPR of X, this notation indicates the absence
of an awaited DATA message generated by X and forwarded by node Y .
For specifying the clauses concerning trust in the protocol, we use the language
proposed by [7] which allows to express trust by the fact that if an entity A trusts
an entity B in some respect, informally means that A believes that B will behave in
a certain way and will perform some action in certain specific circumstances. With
this language, the clauses relating to trust in routing operations are expressed with the
following notation:
the expression A trusts
fw
(Nodes) means that A trusts B (B Nodes) to for-
ward its messages. Otherwise, A not trusting relation is noted by ¬trusts,
3 Synthesis of Previous Works
In our previous works [2,4], we have specified the implicit trust in OLSR, i.e. the trust
relationships that should exist between the nodes according to the OLSR protocol. Then
we have focused on the detection of attacks on MPR selection, where the attacker abuses
the properties of the selection algorithm (HELLO message contents and scheduling) to
be selected as MPR. In this section we review the results of these works.
152
Thus, each node is able to verify the behavior of each neighbor. Such verification
can be improved by correlating the information provided by neighbors. First, a node
can check the consistency between the HELLO messages of its neighbors:
X
HELLO
Y, X
HELLO
Z, Z NS
Y
, Y / N S
Z
X¬trusts(Y, Z) (1)
Second, a node can check the consistency between the HELLO and TC messages of its
neighbors:
X
HELLO,T C
Y
Y, X
HELLO
Z, Z T C
Y
, Y / M P RS
Z
, X¬trusts(Y ) (2)
Third, a node can check the consistency of MPR selection and routing table of its neigh-
bors and verify that the resulting information they announce is coherent with the locally
observed information:
NS
A
N S
B
, Z T C
A
T C
B
X¬trusts(A, B, Z) (3)
It is worth to point out that the mistrust reasonings cannot every time allow the precise
identification of the misbehaving node, but allow the detection of inconsistent behavior
related to a group of nodes which includes the attacker.
4 Simulation Results
In previous works [2, 4], we have illustrated the effectiveness of trust reasoning for at-
tacks detection using small / trivial attack schemes, where the attacker’s position was
important to the success of the attack. However, the simulation with large scale ad-hoc
networks is necessary to prove the effectiveness and capability of the attack detection,
whatever the position of the attacker and the number of nodes in the network. In this
section, we discuss the simulation of OLSR with the integration of the previous formu-
lae. We evaluate the effectiveness of our approach under various attack scenarios, and
the capacity of mistrust based verifications to identify the attacker nodes.
4.1 Implementation
We have used the GlomoSim Simulator and the OLSR patch developed by the Niigata
University [1] to simulate the attacks and previous formulae. We have added to this
patch a module implementing mistrust rules, and several attack scenarios. In our sim-
ulations, ad-hoc networks are composed of different number of nodes (30, 50 and 100
nodes) which are placed randomly. Moreover, the attackers are selected randomly, and
each one selects randomly an attack scenario, as well as a set of targets according the
selected attack. However, since the ad-hoc networks have to be stabilized to allow the
attacker to perform the previous attacks, we have considered that nodes are not mobile.
In the following, we discuss only results with 100 nodes using the first attack scenario
which takes place according to the following steps:
1. The attacker A identifies target T , its neighbors and 2 hop neighbors.
153
2. The attacker A detects its common neighbors with the target T , and modifies its
HELLO messages to advertise their neighbors as symmetric neighbors as well as
an additional fictitious node X.
3. The attacker advertises as its MPR selectors the target’s 2 hop neighbors in its TC
messages: Y N S
A
NS
T
, Z N S
Y
, Z / N S
A
: T C
A
= T C
A
Z, X.
According to OLSR specification, the target has to select the attacker as MPR because
it provides reachability for the nodes Z and X, and so the attacker can control some
target’s flows.
4.2 Results and Discussion
Fig.1. Simulation results with 100 nodes.
The diagram (a) in (4.2) presents simulation results of the rate of nodes which are
able to detect the attack compared to the total number of nodes in the network. As we
can see, the percentage of detection never reaches 100%. However by analyzing the
results for each situation, we have deduced that the percentage compared to the total
number of nodes in the network was not significant, since the previous formulae do not
allow the detection by all the nodes of the network, but only by the concerned nodes
that are directly or indirectly impacted in the attack scenario (e.g., the target, the nodes
used by the attacker to perform the attack and the attacker neighbors).
According to this, in the second step of simulations, we have decided to study the
percentage of the concerned nodes that detect the attack. For that, we first have identi-
fied the concerned nodes for each attack scenario. Since the concerned nodes depend on
each attack scenario, we take the network presented in figure (2) as an example for the
first attack. In this scenario, the concerned nodes which have to detect the attack using
trust reasoning are :
1. The target node is the first and the most important concerned node, because it is the
target of the attack.
2. The target’s 2 hop neighbors have to detect the attack because they are advertised
as symmetric neighbors and MPR selectors by the attacker when they are not
3. The neighbors of the faulty neighbors advertised by the attacker detect the attack
by comparing local information with the attacker TC message.
154
Fig.2. Network example: A is the attacker, T is the Target.
4. The nodes that can compare the neighborhood of the attacker and the target neigh-
bors (which provide reachability to the 2 hop neighbors of the target advertised by
the attacker).
To illustrate the attack detection, we take the example presented in figure 2. The
attack takes place according to the following steps :
1. The attacker A identifies target T , its neighbors {N1, N2, N7, N 8, N20} and 2
hop neighbors {N 3, N7, N 8, N9, N12, N11, N21, N22}.
2. The attacker A detects its common neighbors with the target T : (N1, N 2, N20),
and modifies its HELLO messages to advertise their neighbors (N7, N8, N 9, N21,
N22) as its symmetric neighbors: NS
A
= {T, N1, N 2, N3, N5, N 6, N20, N7,
N8, N9, N21, N22, X} (X is the additional fictitious node).
3. The attacker advertises as its MPR selectors the target’s 2 hop neighbors in its TC
messages: N7, N 8, N9, N21, N22 T C
A
.
According to OLSR specification, the target T has to select the attacker A as MPR,
allowing the attacker to control some target’s flows. In this example, the concerned
nodes are the target T , the target’s 2 hop neighbors N7, N 8, N9, N21, N22 since they
are advertised as symmetric neighbors by the attacker, and the neighbors of these nodes
since they are indirectly impacted by the attack (they should be neighbors of the attacker
but they are not), and the nodes that are able to correlate the information advertised by
the attacker with other information. Using the previous formulae, all the concerned
nodes are able to detect the attack:
1. The target node T detect the attack using the following formulae:
Formula 1: the node detect inconsistencybetween HELLO messages of N 7, N8
and A, where N7, N8 N S
A
but A / N S
N7
, A / N S
N8
.
Formula 2: the node detect inconsistencybetween HELLO messages of N 7, N8
and TC message of A, where N 7, N8 T C
A
but A / NS
N7
and A / N S
N8
.
Formula 3: in this example, the node N9 will select N 2 as MPR and nodes
N21, N22 will select N20 as MPR. In the reception of the TC messages of N 2
and N20, the target will detect inconsistency (3), because the neighborhood of
N2 and N20 are included in the attacker neighborhood, and nodes N2 and
N20 should not be selected as MPR.
155
2. The target’s 2 hop neighbors advertised as symmetric neighbors and MPR selectors
by the attacker (N7, N8, N9, N21, N22 T C
A
) detect the attack. For instance
the node N7 detects the attack using the formulae 1 and 2: when it receives the TC
message of the attacker, it detects inconsistency because it has not selected attacker
as MPR and the attacker is not a symmetric neighbor A / MP RSS
N
7.
3. The neighbors of the faulty neighbors advertised by the attacker detect the at-
tack. For instance the node N2 detects inconsistency using the formula 2 in the
reception of the TC messages of the attacker because N7, N8, N9 T C
A
but
A / N S
N
7, A / NS
N
8, A / NS
N
9.
4. The nodes that can compare the neighborhood of the attacker and the target neigh-
bors (which provide reachability to the 2 hop neighbors of the target advertised by
the attacker). For instance when the node N1 receives TC messages of the attacker
and N2, it detects inconsistency using formula 3 because the neighborhood of N2
is included in the attacker neighborhood.
The diagrams (a) and (b) in the figure (4.2) allow to compare simulation results
regarding total nodes in the network , and concerned nodes by the attack. Simulation
results about concerned nodes detecting the attack show that using mistrust reasoning,
attacker is detected exactly or partially by all the concerned nodes. Partial detection
is the case where a node detects an inconsistency between several nodes, including
the attacker, but is unable to determine exactly who is attacking (for example formula
1). Certain nodes can detect the attacker partially and exactly using different formulae
(for example formulae 1 and 2). In this case, they can deduce exactly which node is
the attacker, and ignore the partial detection. However, partial attack detection provide
mistrust information, which can be considered by a node in future cooperation with
other nodes to take important decision (for example MPR selection) or correlated with
other partial detections in order to deduce exactly the misbehavior node.
The simulation allows us to prove that verifications are a matter of local node be-
havior producing a global effect on the ad hoc network. Indeed, each node can reason
locally on its direct observation to detect inconsistencies without the need for the opin-
ions of other nodes or to cooperate with them. These results reveal the effectiveness
of trust-based reasoning for detecting attacks and preventing from the problem of false
opinions that can occurs by sharing trust information.
5 Conclusions
Using simulation, we have demonstrated the effectiveness of the verification based on
mistrust reasoning in the attack detection. The results allow us to set up verifications
that each node can perform to assess the correct behavior of the other nodes and detect
attacks against OLSR. It is important to mention that the OLSR protocol (messages) is
unchanged, and so our approach still compatible with the bare OLSR.
These results motivate extending the approach for evaluating and distributing the
trust information in order to mitigate a partial detection and come to a total detection.
Indeed, cooperation between node by sharing trust information could be used to enforce
the detection of the misbehavior nodes based on distributed decision. However, it is
worth to point out that second-hand information can be subject to false accusations. To
156
mitigate this problem, we plan to set up a mechanism that allows each node to give a
proof of its mistrust opinion to participate in the propagation (distribution) of mistrust
towards the network. Such a mechanism could be used to enforce a reputation systems
by establishing (verifying) trust relationships before cooperating with the other nodes.
References
1. OLSR patch for GlomoSim. http://www.net.ie.niigata-u.ac.jp/mase/olsr/.
2. A. Adnane, R. T. D. Sousa, C. Bidan, and L. M´e. Analysis of the implicit trust within the
OLSR protocol. IFIPTM-2007, Joint iTrust and PST Conferences on Privacy, Trust Manage-
ment and Security, Moncton, New Brunswick, Canada.
3. A. Adnane, R. T. D. Sousa, C. Bidan, and L. M´e. Integrating trust reasonings into node
behavior in OLSR. the 3-rd ACM International Workshop on QoS and Security for Wireless
and Mobile Networks (Q2SWinet 2007), Chania, Crete Island, Greece., October 2007.
4. A. Adnane, R. T. D. Sousa, C. Bidan, and L. M´e. Autonomic trust reasoning enables mis-
behavior detection in olsr. SAC’08 : 23rd Annual ACM Symposium on Applied Computing,
Fortaleza, Cear´a, Brazil, March 2008.
5. T. Clausen and P. Jacquet. IETF RFC 3626: Optimized Link State Routing Protocol OLSR.
October 2003.
6. M. Wang, L. Lamont, P. Mason, and M. Gorlatova. An effective intrusion detection approach
for OLSR MANET protocol. First Workshop on Secure Network Protocols (NPSec). Boston,
Massachusetts, USA, July 2005.
7. R. Yahalom, B. Klein, and T. Beth. Trust relationships in secure systems A distributed
authentication perspective. In Proceedings of the IEEE Symposium on Security and Privacy,
1993.
157