JDFSL V9N2 Accurate Modeling of the Siemens S7 SCADA ...
SCADA systems were originally designed for
serial communications, and were built on the
premise that all the operating entities would
be legitimate, properly installed, perform the
intended logic and follow the protocol. Thus,
many SCADA systems have almost no measures
for defending against deliberate attacks. Specif-
ically, SCADA network components do not ver-
ify the identity and permissions of other compo-
nents with which they interact (i.e., no authen-
tication and authorization mechanisms); they
do not verify message content and legitimacy
(i.e., no data integrity checks); and all the data
sent over the network is in plaintext (i.e., no en-
cryption to preserve confidentiality). Therefore,
deploying an Intrusion Detection Systems (IDS)
in a SCADA network is an important defensive
measure.
The Siemens S7 is one of the leading proto-
cols used in SCADA networks. Siemens S7 Pro-
grammable Logic Controllers (PLCs) (Siemens,
2014) are estimated to have over 30% of the
worldwide PLC market (Electrical Engineering
Blog, 2013). The platform is so popular that
other companies (e.g., (VIPA - A Yaskawa com-
pany, 2014)) offer compatible PLCs.
1.2 Related Work
Since the S7 protocol is proprietary, there is lit-
tle published information about attacks against
it. An exception is the work of (Beresford,
2011). The author showed that the standard
S7 protocol is not encrypted, or authenticated,
it is susceptible to spoofing, session hijacking,
Denial of Service (DoS) attacks, and other at-
tacks. Gaining access to the control network
gives the attacker full access to the PLCs and
allows attacks against the engineering worksta-
tion as well.
(Zhu, Joseph, & Sastry, 2011) evaluated sev-
eral SCADA-specific Network Intrusion Detec-
tion Systems (NIDSs), but they mentioned that,
to their best knowledge, none of the surveyed
systems has been tested on real operational
SCADA network. Due to the lack of access
to production ICS networks, many works deal
with the issue of building a SCADA testbed
that enables experimental capabilities of check-
ing vulnerabilities and validating security solu-
tions (Genge, Siaterlis, Nai Fovino, & Masera,
2012; Hahn et al., 2010; Mallouhi, Al-Nashif,
Cox, Chadaga, & Hariri, 2011). In contrast, one
of the important aspects of our work is that the
intrusion detection approach is evaluated using
real traffic from production SCADA networks.
(Yang, Usynin, & Hines, 2006) used an Auto
Associative Kernel Regression (AAKR) model
and applied it on a SCADA system looking for
matching patterns. The AAKR model used nu-
merous indicators, representing network traffic
and hardware-operating statistics to predict the
normal behavior. Hence, this model needs to
monitor different indicators for different intru-
sion methods, and must manage a large number
of potentially valuable variables.
Several recent studies (such as (Atassi, El-
hajj, Chehab, & Kayssi, 2014) & (Chen, Hsiao,
Yang, & Ou, 2013)) suggest anomaly-based de-
tection for SCADA systems that is based on
Markov chains. However, (Ye, Zhang, & Bor-
ror, 2004) showed that although the detection
accuracy of this technique is high, the number
of ‘false positive’ values is also high, as it is sen-
sitive to noise.
(Hadziosmanovic, Bolzoni, Hartel, & Etalle,
2011) used the logs generated by the control ap-
plication running on the HMI to detect anoma-
lous patterns of user actions on process control
application. The focus of this work was on the
threats that can be triggered by a single user
action. The authors acknowledged that “an at-
tacker could manipulate logs by sending false
data to the control application”. This model is
also susceptible to replay attacks.
(Barbosa, Sadre, & Pras, 2012) studied the
periodicity characteristics of SCADA traffic.
They measured dominant periods of between 1-
60 seconds in their datasets. They also observed
changes in the baseline patterns of the SCADA
traffic they collected, which they related to the
start (or end) of non periodic high throughput
flows. They speculated that these changes are
due to changes in the controlled environment,
such as water tanks becoming full and pipes be-
ing closed.
(Cheung et al., 2007) designed a model-based
Page 38
c
2014 ADFSL