SPECIFICATION-BASED INTRUSION DETECTION SYSTEM FOR
CARRIER ETHERNET
Pan Jieke, Jo
˜
ao Redol
Siemens Networks S.A, Lisboa, Portugal
Miguel Correia
Faculty of Sciences, University of Lisboa, Lisboa, Portugal
Keywords:
Specification-based Intrusion Detection, Ethernet, Spanning Tree Protocol, Network Topology, Security.
Abstract:
Layered network architectures (OSI, TCP/IP) separate functionality in layers, allowing them to be designed
and implemented independently. However, from the security point of view, once a lower layer is compromised,
the reliability of the higher layers can be impaired. This paper is about the security of the Data Link Layer,
which can affect the reliability of higher layers, like TCP, HTTP and other World-Wide Web protocols. The
paper analyzes security-wise a layer 2 protocol the Spanning Tree Protocol (STP), part of the Ethernet suite
and presents a solution to detect attacks against this protocol using Specification-based Intrusion Detection.
1 INTRODUCTION
The Internet and the World-Wide Web play an in-
creasingly indispensable role in the modern society.
The dependence on networked computers is a phe-
nomenon that can be observed in several areas. This
trend could be observed during the last decade and
will certainly continue in the future. Nevertheless,
due to its popularity, the Internet has become the tar-
get of malicious hackers, which cause intrusions and
disseminate malware at a world-wide scale.
Layered network architectures, like the OSI and
TCP/IP models, separate functionality in layers,
where the lower layers provide services for the higher
layers. These flexible models provide a form of sep-
aration of concerns, which allows the layers to be
designed and implemented independently. However,
from the security point of view, once a lower layer
is compromised, the reliability of the higher layers
can also be affected. Most existing network secu-
rity paradigms and models are concerned with the
layers from 3 to 7 of the OSI model, i.e., from the
Network to the Application layers. In Internet terms,
this means security is mostly concerned with proto-
cols like IP, TCP, HTTP and SOAP, and issues like
user authentication, data integrity and confidentiality.
Less attention has been paid to the network infrastruc-
ture, i.e., to layer 1 and 2 protocols. However, attacks
against these two layers can “disable” the network,
causing unavailability of the higher layer protocols,
with a huge impact on a large-scale network, possibly
involving innumerable service providers and users.
The work presented in this paper has the purpose
of protecting the Carrier Ethernet network infrastruc-
ture using a distributed intrusion detection and re-
sponse system, since this kind of network technology
is being adopted around the world by Internet service
providers and other companies
1
. More specifically,
the work deals with the Network Elements NEs
Switches/Bridges and the Data Link Layer (layer
2) protocols. Currently, NEs are reasonably com-
plex hardware and software boxes, which therefore
can have vulnerabilities, like those often reported in
web servers, browsers, operating systems, etc. So, we
consider that a subset of the NEs of a network can be
compromised by hackers and launch attacks against
the network infrastructure, i.e., they can be malicious.
Previous protection schemes for STP did not consider
this possibility of NEs being malicious; they essen-
tially discard STP messages (BPDUs) (IEEE, 1998)
sent by non-NEs trying to emulate a NE (Marro,
2003; Cisco, 2005a; Cisco, 2005b).
To deal with this problem, we propose that NEs
1
The adoption of Carrier Ethernet is being pushed by
the Metro Ethernet Forum. The site of the forum is at:
http://www.metroethernetforum.org
426
Jieke P., Redol J. and Correia M. (2007).
SPECIFICATION-BASED INTRUSION DETECTION SYSTEM FOR CARRIER ETHERNET.
In Proceedings of the Third International Conference on Web Information Systems and Technologies - Internet Technology, pages 426-429
DOI: 10.5220/0001283204260429
Copyright
c
SciTePress
are equipped with a component which provides net-
work intrusion detection. Each of these detectors in-
spects the behavior of other NEs by inspecting the
messages received from them. Detection follows a
recent approach dubbed specification-based intrusion
detection, which relies on a specification of the proto-
col to detect deviations from it (Balepin et al., 2003;
Sekar et al., 2002; Uppuluri and Sekar, 2001).
The contribution of this paper is the design of the
first network intrusion detection scheme for carrier
Ethernets. It focuses on the original link management
protocol of the (switched) Ethernet, the Spanning
Tree Protocol (STP), as a case study (IEEE, 1998).
Detection is based on a specification-based intrusion
detection scheme, which we extend with annotations,
to detect attacks with certain time patterns.
2 THE SYSTEM
The paper proposes a distributed intrusion detection
system for Carrier Ethernet (Figure 1). Typically,
Carrier class NEs are used for providing connectiv-
ity from end users to ISPs or other service providers.
The proposal involves extending each NE with a net-
work intrusion detection component. This extension
can be done in two ways:
by inserting the component in the NE, in case its
internal resources (CPU, memory) allow it; or
by connecting the component to the NE using a
network port, and forwarding all BPDUs received
by the NE to the component.
Figure 1: System architecture.
We assume a subset of the NEs of a network
can be compromised by hackers and launch attacks
against the network infrastructure. We say those NEs
are malicious, while the rest are correct. To make
the assumption even stronger, if a NE is malicious, its
network intrusion detection component will be com-
promised too. Therefore, a subset of the NEs may
not follow the correct specification of STP, and this
incorrect behavior has to be detected by the correct
NEs/detectors. Previous protection schemes for STP
essentially discarded BPDUs sent by non-NEs trying
to emulate a NE (Marro, 2003; Cisco, 2005a; Cisco,
2005b).
2.1 STP Attacks
As one of the most commonly used Ethernet standard
configuration protocols, the Spanning Tree Protocol
(STP) is an obvious target for attacks, which can com-
promise the security of all the layers above layer 2 and
the network availability. STP is a low level network
link management protocol that provides path redun-
dancy while preventing undesirable loops in the net-
work. STP exchanges messages called Bridge Pro-
tocol Data Units (BPDUs), which contain the Bridge
ID (bridge’s unique identifier) of the sender. BPDUs
are used to learn about the existence of other bridges
and to obtain the information needed to calculate and
maintain the spanning tree. A few characteristics ren-
der the STP protocol vulnerable to several types of
attacks from hackers who have direct physical access
to the network equipment. These attacks use the le-
gitimate connection with network equipment to inject
attacks in the network. Recall that we say that a NE
is malicious if it is controlled by a hacker, thus can
execute arbitrary attacks against the STP protocol.
A careful analysis of the protocol and the literature
(Marro, 2003), lead us to conclude that the attacks
possible against STP are essentially the follow:
ID changing attacks: A NE modifies its bridge
identifier to cause (undesirable) topology changes
(since the root of the tree is the node with lowest
bridge ID).
Silent attacks: When the network topology is
stable every NE continues sending Configuration
BPDUs to others to indicate it is alive. A mali-
cious NE can omit sending these BPDUs to cause
an undesirable network reconfiguration. When
executed repetitively, this attack can impair the
availability of the network.
Faked failure attacks: A NE generates fake Topol-
ogy Change Notification BPDUs to cause network
reconfigurations.
BPDU flooding attacks: A flood of bogus BPDUs
from different addresses is injected in the net-
work, possibly causing undesirable network re-
configurations.
Invalid BPDU: Sending malformed BPDU, e.g.,
as an attempt to find a vulnerability in the target
NE.
SPECIFICATION-BASED INTRUSION DETECTION SYSTEM FOR CARRIER ETHERNET
427
2.2 STP Specification
This paper uses specification-based intrusion detec-
tion (Balepin et al., 2003; Sekar et al., 2002; Uppu-
luri and Sekar, 2001) which detect attacks as devia-
tions from a norm. By specifying correct behavior,
any other behaviors will be classified as anomalous.
The detector performs network intrusion detection
in run-time. The specification of the STP protocol is
modeled using a state machine. The states of this ma-
chine are the states of the protocol, and state transi-
tions are caused by the reception of BPDUs or expi-
ration of timeouts.
Instead of crafting the detailed description of the
protocol, an abstract specification of STP is used. De-
veloping a more precise specification requires more
effort and has been considered to have a negative im-
pact in specification-based intrusion detection (Sekar
et al., 2002). The reason is that real implementations
of a protocol tend to have differences between them-
selves and to deviate from the specification on the
standard or RFC. A more abstract specification tends
to omit the details that lead to these inconsistencies.
Figure 2: Spanning Tree Protocol state machine.
Figure 2 presents the state machine that specifies
STP. Each node in the diagram represents a state in
which a NE can be at a certain instant. Each arrow
represents a transition between two states. Arrows
have a label in the form of a fraction: the top part
is the event that causes the transition (message recep-
tion, timeout), while the bottom is the event generated
(message sent).
Annotations The state machine in Figure 2 is an in-
complete specification of the STP protocol since it al-
lows behavior that is not acceptable. Examples in-
clude a NE omitting sending Configuration BPDUs
repeatedly although it is alive, and sending many bo-
gus Topology Change BPDUs.
Distinguishing correct behavior from these devia-
tions requires extending the state machine with an-
notations. Each event timeout or send BPDU can
be annotated with the acceptable time pattern of that
event. For simplicity we consider that, given an
event e, this pattern is defined simply in terms of
the maximum number of repetitions of e per unit of
time, Rmax
e
, but more complex forms of specifi-
cation might be used. These values can be defined
for any event but only some of them are relevant
to detect the attacks presented in Section 2.1. Ex-
amples are Rmax
timeout
for the timeout in the
transition between the state Wait for CONF BPDU
and the state Wait TCNA BPDU and Rmax
tcn
for the send TCN BPDU events in state Wait for -
TCNA BPDU. These values have to be carefully de-
fined in order to avoid detecting false attacks (false
positives) and missing attacks (false negatives).
2.3 Intrusion Detection
Consider again the state machine in Figure 2. Each
NE stores a representation of the state machine of its
neighbors (i.e., of the NEs who are directly connected
to it) and has a variable that says which is the current
state of a neighbor. The state machine for every NE
starts at the Init state. The detection algorithm exe-
cuted by a pair NE/detector is the following:
1. When the NE receives a BPDU message: pass it
to the detector;
2. When the detector gets the BPDU: check if it is
an expected BPDU message in the current state
and if satisfies the pattern defined for it (i.e., its
Rmax
e
);
3. If the message is not expected, then suspect of the
NE that has sent it.
An attack is detected firstly by matching the re-
ceived message with expected messages in the cur-
rent state. For example, if a NE is in the Wait for -
CONF BPDU state, it can not do a transition to the
Init state directly (see Figure 2). In other words,
if a NE is waiting for a configuration BPDU in a
stable topology, it is incorrect to send a Topology
Change BPDU. Therefore, if the NE in the Wait for -
CONF BPDU sends a Topology Change BPDU to the
others, this will be detected to be an anomaly. The
same reasoning can be done to the other states and to
the patterns.
Let us now make an argument that the intrusion
detection scheme presented above detects the attacks
WEBIST 2007 - International Conference on Web Information Systems and Technologies
428
against STP presented in Section 2.1. All detection
actions are made by the detectors in the NEs, but we
simplify the presentation by forgetting this separation
between a NE and its detector.
ID changing attacks. Recall that each NE stores
a representation of the state machine of its neigh-
bors. This representation includes the Bridge ID
of the neighbors. If the NE receives two or more
BPDUs with different sender IDs from the same
port, it detects that its neighbor is malicious (if
the ID is modified by an administrator, other NEs
should be informed).
Silent attacks. For this type of attacks, when a ma-
licious NE stays silent on purpose, other NEs will
generate a timeout event, consider it as a legitim
NE failure and send Topology Change Notifica-
tion BPDUs (see Figure 2). So, the specification-
based intrusion detection scheme does not detect
it to be an attack. One solution would be the elim-
ination of the timeout event from the specifica-
tion, so a timeout would always be suspected of
being an attack, but it would generate suspicions
for every (non-malicious) failure. The solution is
not to detect if the NE simulates a failure but if it
does it too many times. This detection is done
using the pattern of the event timeout between
the states Wait for CONF BPDU and Wait for -
TCNA BPDU, i.e., if that timeout expires more
than Rmax
timeout
times in a unit of time then the
NE is malicious.
Faked failure attacks. This attack is done by send-
ing a Topology Change Notification BPDU re-
peatedly, so it is detected if that BPDU is not sent
in the state Wait for TCNA BPDU or if it is sent
more than Rmax
tcn
times in a unit of time.
BPDU flooding attacks. These attacks are de-
tected when events that send the BPDUs are done
more than the corresponding Rmax
e
times in a
unit of time.
Invalid BPDU. This attack is directly detected
by the specification-based intrusion detection
scheme, since an invalid BPDU is precisely one
that can not be sent in the current state of the NE.
When a NE is identified as being malicious, it will
be logically removed from the network, i.e., its neigh-
bor NEs will disconnect all ports connected to it. Nat-
urally this is only possible if the network has enough
redundancy to disconnect an NE and still keep the net-
work connected. Moreover, it requires that the NEs
correlate their detections (Kruegel et al., 2005), in
order to make agreement about which NE(s) is ma-
licious. This correlation and disconnection scheme is
still being investigated.
3 CONCLUSION
The Internet and the World-Wide Web have become
more and more important in modern society. How-
ever, they also became the target of a legion of ma-
licious hackers. Compromising a low layer, like the
Data Link Layer, can affect the reliability of higher
layers, like TCP/IP, HTTP, SOAP and other web pro-
tocols. This work gives a novel perspective about a se-
curity solution for low-level network protocols, based
on a specification-based intrusion detection system
enhanced with annotations. More precisely, we pre-
sented a solution for protecting Carrier Ethernet by
detecting attacks against the STP protocol. We argued
that protecting these protocols is crucial to ensure the
availability of the network so higher level protocols,
like HTTP and other Web-related protocols can be ex-
ecuted normally.
ACKNOWLEDGEMENTS
The work presented in the paper is funded by Siemens
Networks, S.A. Portugal.
REFERENCES
Balepin, I., Maltsev, S., Rowe, J., and Levitt, K. N. (2003).
Using specification-based intrusion detection for au-
tomated response. In Proceedings of the International
Symposium on Recent Advances in Intrusion Detec-
tion, pages 136–154.
Cisco (2005a). Spanning Tree PortFast BPDU Guard En-
hancement. Cisco Systems Inc. Document ID 10586.
Cisco (2005b). Spanning Tree Protocol Root Guard En-
hancement. Cisco Systems Inc. Document ID 10588.
IEEE (1998). ANSI/IEEE 802.1D-2004 standard - Part 3:
Media Access Control (MAC) Bridges.
Kruegel, C., Valeur, F., and Vigna, G. (2005). Intru-
sion Detection and Correlation: Challenges and Solu-
tions, volume 14 of Advances in Information Security.
Springer-Verlag.
Marro, G. M. (2003). Attacks at the data link layer. Master’s
thesis, University of California.
Sekar, R., Gupta, A., Frullo, J., Shanbhag, T., Tiwari, A.,
Yang, H., and Zhou, S. (2002). Specification-based
anomaly detection: a new approach for detecting net-
work intrusions. In Proceedings of the 9th ACM con-
ference on Computer and communications security,
pages 265–274.
Uppuluri, P. and Sekar, R. (2001). Experiences with
specification-based intrusion detection. Lecture Notes
in Computer Science, 2212:172–189.
SPECIFICATION-BASED INTRUSION DETECTION SYSTEM FOR CARRIER ETHERNET
429