From Exposed to Exploited: Drawing the Picture of Industrial Control
Systems Security Status in the Internet Age
Yixiong Wu
1
, Jianwei Zhuge
1,2
, Tingting Yin
1
, Tianyi Li
3
, Junmin Zhu
4
, Guannan Guo
5
, Yue Liu
6
and Jianju Hu
7
1
Institute of Network Science and Cyberspace, Tsinghua University, Beijing, China
2
Beijing National Research Center for Information Science and Technology, Beijing, China
3
Peking University, Beijing, China
4
Shanghai Jiao Tong University, Shanghai, China
5
School of Computer Science and Technology, University of Science and Technology of China, Hefei, China
6
Qi An Xin Technology Research Institute, Beijing, China
7
Siemens Ltd., China
Keywords:
Internet-facing ICS Devices, Passive Vulnerability Assessment, Device Search Engine.
Abstract:
The number of Internet-facing industrial control system(ICS) devices has risen rapidly due to remote control
demand. Going beyond benefits in maintenance, this also exposes the fragile ICS devices to cyber-attackers.
To characterize the security status of Internet-facing ICS devices, we analyze the exposed ICS devices and their
vulnerabilities. Considering the ethic, we design and implement ICScope, a passive vulnerability assessment
system based on device search engines. Firstly, ICScope extracts the ICS device information from the ban-
ners returned by multiple search engines. Then, ICScope filters out the possible ICS honeypots to guarantee
accuracy. Finally, ICScope associates ICS vulnerabilities with each ICS device. Over the past year, our mea-
surements cover more than 466,000 IPs. We first perform a comprehensive measurement of Internet-facing
ICS devices from Dec 2019 to Jan 2020. We find that there are about 49.58% of Internet-facing ICS devices
that can be identified are affected by one or more vulnerabilities. We also conduct three times experiments
from Jun 2020 to Dec 2020 to monitor the security status of Internet-facing ICS devices. We observe a slowly
decreasing trend in the number of vulnerable ICS devices during our experiment period.
1 INTRODUCTION
With the rapid growth and deep integration of In-
formation Technology and Operational Technology,
more and more industrial control system (ICS) de-
vices with computing and communication capabili-
ties are introduced to smart factories, buildings, and
cities. The proprietary protocols used in these devices
are originally designed to communicate between de-
vices in isolated operation LANs. As a result, secu-
rity protections for ICS communication are usually
missing. Besides, for economic considerations, ICS
devices are rarely updated or upgraded. Administra-
tors usually avoid upgrades to reduce the possibility
of system crashes. Thus, the life cycle of ICS vul-
nerabilities is much longer than other kinds of vulner-
abilities. However, more and more ICS devices are
exposed to the Internet nowadays. This is usually for
the convenience of data collection and remote mainte-
nance. However, it also makes the fragile ICS devices
exposed to the cyber-attacks. In 2017, Triton attacked
a petrochemical plant in the Middle East, resulting in
an emergency shutdown of a critical industrial con-
trol system (Di Pinto et al., 2018). In 2019, the Triton
was relaunched. Cyber-attacks like these towards ICS
have existed for a long time and have already caused
substantial damage.
To mitigate this type of threat, both the research
community and the industry take proactive actions.
ICS-CERT
1
provides annual assessment reports for
public ICS vulnerabilities. Kaspersky
2
further ana-
lyzes these reports and summarizes the vulnerability
characteristics. However, both of them did not an-
1
https://us-cert.cisa.gov/ics
2
Industrial Control Systems Vulnerabilities Statistics,
Reported by Kaspersky in 2019
Wu, Y., Zhuge, J., Yin, T., Li, T., Zhu, J., Guo, G., Liu, Y. and Hu, J.
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age.
DOI: 10.5220/0010327902370248
In Proceedings of the 7th International Conference on Information Systems Security and Privacy (ICISSP 2021), pages 237-248
ISBN: 978-989-758-491-6
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
237
alyze the distribution of vulnerabilities in the actual
network. Existing works also focus on the discov-
ery of online ICS devices. Both Mirian et al. (Mirian
et al., 2016) and Feng et al. (Feng et al., 2016) have
proposed their device discovery approaches to give an
Internet-Wide view of ICS devices. However, they
focus more on the number of exposed devices rather
than vulnerable devices. Further works behind on-
line ICS devices’ detection are to perform a vulnera-
bility assessment for the Internet-facing ICS devices.
Characterizing the vulnerable Internet-facing ICS de-
vices is highly desirable for the interests of securing
Internet-facing ICS devices.
To identify a device’s vulnerabilities, one instinc-
tive reaction is to perform a vulnerability assess-
ment using vulnerability scanners like Nessus (Ac-
tive Mode). However, such techniques cannot be ap-
plied to identify an ICS device’s vulnerability. Be-
cause sending many packets to ICS systems regards
as an intrusive behavior. Therefore, we prefer the pas-
sive vulnerability assessment (Passive Mode), which
is based on a search engine to identify a device’s vul-
nerabilities without any interaction. Most previous
works on passive vulnerability assessment focus on
common services, i.e., HTTP, IMAP. For example,
ShoVAT (Genge and En
˘
achescu, 2016) focuses on
several common services, and Scout (O’Hare et al.,
2019) only supports HTTP service. Because of the
inconsistent banner format between common services
and ICS devices, their methods are not considered
feasible. There are also existing works focus on ICS
services. Both Samtani et al. (Samtani et al., 2016)
and Leverett et al. (Leverett and Wightman, 2013)
identify the vulnerabilities for online ICS devices.
However, the scope of their works is limited to spe-
cial ICS services. Moreover, both of them scan target
ICS devices. In this paper, we aim to answer the fol-
lowing question:
Q: What is the security status of Internet-facing ICS
devices in the whole public IP address space?
To answer this question, we propose a passive vul-
nerability assessment system based on device search
engines. For ethical consideration, our system IC-
Scope uses passive mode to avoid intrusive behaviors.
We face three main technical challenges to build IC-
Scope. 1) The information returned by device search
engines is often incomplete. 2) There are many ICS
honeypots in the results of device search engines. 3)
It is not easy to accurately find the vulnerabilities that
affect the ICS devices.
To address these challenges, we build ICScope
with data enrichment module, honeypot detection
module, and vulnerability association module. For
the first challenge, the data enrichment module inte-
grates the device information extracted from multiple
search engines to enrich the incomplete device infor-
mation. It also uses the fingerprint database. For the
second challenge, the honeypot detection module uti-
lizes the non-interaction features to detect ICS honey-
pots, such as ISP, ICS honeypot fingerprinting. It also
detects ICS honeypots by comparing the results be-
tween multiple search engines. For the last challenge,
the vulnerability association module builds vulnera-
bility trees to identify each ICS device’s vulnerabili-
ties.
We have implemented ICScope and performed a
large-scale measurement of ICS devices’ security sta-
tus in the whole public address space. We collect
270,283 Internet-facing ICS devices on four device
search engines from Dec 2019 to Jan 2020. Among
these ICS devices, ICScope has detected 21,578 ICS
honeypots (about 7.98%). Excluding these ICS hon-
eypots, we observe that 106,382 ICS devices can ex-
tract complete ICS device entries. And ICScope re-
gards 52,739 of them (about 49.58%) as vulnerable
ICS devices. We further analyze the vulnerable ICS
devices and related vulnerabilities. We find that the
vulnerable ICS devices are distributed in the vast ma-
jority of countries. Among the vulnerabilities, most
of them are in high or critical severity level and can
be remotely exploitable with little or no limitation.
Over six months, our measurement shows a slowly
decreasing trend in the number of vulnerable ICS de-
vices. We also use ICScope to measure the impact of
a 0day vulnerability. The result shows that the vulner-
ability affects 3,999 ICS devices.
Contribution. Our contributions can be summarized
as follows:
1. We design and implement a passive vulnerability
assessment tool ICScope. ICScope can automat-
ically obtain ICS devices’ information from mul-
tiple device search engines and correlate the vul-
nerabilities of each device.
2. By using ICScope, we perform the large-scale
measurement on the security status of all Internet-
facing ICS devices in the whole public IP address
space and find that 49.58% of the Internet-facing
ICS devices with complete device entries are vul-
nerable. We also further characterizing the vul-
nerable ICS devices.
3. We proposed a non-interaction ICS honeypot de-
tection approach to detect the ICS honeypots
without intrusive behaviors. This method has de-
tected 21,578 ICS honeypots (about 7.98%) on
our datasets.
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
238
Our results reveal the severe and considerable se-
curity risks faced by ICS devices in the current cy-
berspace. At the end of the paper, we discuss the mit-
igation measures against this security threat.
2 MOTIVATION AND
CHALLENGES
2.1 Motivation
The risk of Internet-facing ICS devices is a persis-
tent problem. Most existing works focus on what
devices have been exposed to the Internet or the at-
tacker’s malicious behaviors (Fachkha et al., 2017).
Researchers perform vulnerability assessments for
Internet-facing ICS devices (Samtani et al., 2016;
Leverett and Wightman, 2013). However, these works
are usually limited in scale and scope. More impor-
tantly, their approaches cannot avoid intrusive behav-
iors.
In this paper, we seek to provide a more compre-
hensive and holistic view of Internet-facing ICS de-
vices’ security status via a large-scale empirical mea-
surement. We obtain more comprehensive Internet-
facing ICS devices by integrating multiple search en-
gines’ data and perform passive vulnerability assess-
ment on them. We further analyze the vulnerable ICS
devices and related vulnerabilities across multiple di-
mensions, e.g., geolocation distribution, vulnerability
ratio. We also conduct a non-interaction ICS honey-
pot detection to decrease false positives while previ-
ous vulnerability assessment works not.
2.2 Challenges
There are three main challenges in designing a passive
vulnerability assessment system.
Incomplete Banners. Not all banners returned by
device search engines contain complete device in-
formation. For example, the different banners of
the same ICS device are respectively returned by
Shodan and FOFA (see Figure 1a and Figure 1b).
Compared to Shodan, the FOFAs banner lacks the
Basic Firmware information used to extract version
information. Moreover, both of them do not contain
vendor and product information. Existing works that
reconstruct CPE names from banners are limited to
the lack of device information. Fortunately, we ob-
serve that device search engines’ capabilities are dif-
ferent, and part of device information can be identi-
fied through a particular field in ICS protocol. Based
on this observation, we try to extract complete device
Basic Hardware: 6ES7 313-6CE01-0AB0 v.0.1
Module: 6ES7 313-6CE01-0AB0 v.0.1
Basic Firmware: v.2.0.11
(a) Shodan banner
Vendor: SIEMENS
Product: SIMATIC S7-300 CPU 313C-2DP
Version: v2.0.11
(d) Device Information
Basic Hardware: 6ES7 313-6CE01-0AB0
Module: 6ES7 313-6CE01-0AB0
(b) FOFA banner
(c) SIEMENS
Online Shop
Figure 1: Device information for SIMATIC S7-300 with
banners returned by Shodan and FOFA.
information from banners returned by multiple device
search engines based on ICS protocol features. For
the ICS device in Figure 1, we extract its version from
the Shodan banner. Through the article number found
in the banner, we obtain its vendor and product from
the SIEMENS online shop.
Honeypot Detection. Industrial honeypots are wildly
used in the detection of industrial control security
threats (Serbanescu et al., 2015). The device search
engines also add honeypots to the query results to
monitor the abuse of them by attackers. Previous
works (Feng et al., 2016) that utilize network finger-
printing technology (Comer and Lin, 1994) are not
working here due to the lack of interaction. We lever-
age the non-interaction features to detect ICS hon-
eypots, such as the fingerprints of open-source ICS
honeypots and the device inconsistencies in different
search engines.
Associate Vulnerabilities. Previous works (Genge
and En
˘
achescu, 2016; O’Hare et al., 2019) utilize the
most similar CPE name reconstructed from banner to
find possible vulnerabilities. However, this approach
is limited to ICS. First, some ICS devices might have
more than one CPE name, while some have none. For
example, the device’s CPE name in Figure 1 can be
cpe:/h:siemens:simatic s7-300 cpu:2.0.11 or
cpe:/h:siemens:simatic s7-300 cpu 313:2.0.
11. Second, CPE data used to identify the impact
range of a vulnerability in the National Vulnerability
Database (NVD) is exaggerated or underesti-
mated (Dong et al., 2019). Rather than reconstructing
the CPE name, we construct vulnerability trees from
multiple public vulnerability libraries and use them
to associate vulnerabilities.
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age
239
Banner
Processing
Data
Enrichment
Data Acquisition Information Extraction Vulnerability Association
Banners
Shodan
Censys
FOFA
SiNan
ICS Device
Entries
Vulnerability
Tree
Associate
Vulnerabilities
Vulnerable
ICS Devices
Vulnerability
Databases
Honeypot
Detection
Figure 2: Architecture of the ICScope.
3 ICScope
In this work, we design and implement a search en-
gine based automatic passive vulnerability assess-
ment tool for ICS, ICScope, and present its design in
this section.
3.1 Overview
The overall architecture of ICScope is shaped in Fig-
ure 2, which consists of three modules. First, the data
acquisition module (§3.2) collects all Internet-facing
ICS device data via the Application Programming In-
terface (API) of search engines. Then, the informa-
tion extraction module (§3.3) extracts device infor-
mation, i.e., vendor names, product names, and ver-
sions, from the ICS device banners returned by search
engines. The information extraction module (§3.3.3)
also detects and excludes the ICS honeypots in the
results. Finally, the vulnerability association module
(§3.4) constructs vulnerability trees from the public
vulnerability libraries and utilizes it and the extracted
device information to associate possible vulnerabili-
ties.
3.2 Data Acquisition
The search engines provide APIs with search filters,
which are special keywords for special properties. For
example, protocol:s7 is used to find all Internet-
facing ICS devices in FOFA
3
, which use the S7 proto-
col. As reported by Guo et al., the data integration of
multiple device search engines can obtain more com-
prehensive Internet-facing ICS devices (Guo et al.,
2018). Table 1 shows the industrial control protocols
and device search engines supported by ICScope. IC-
Scope leverages the APIs and special search filters to
acquires all Internet-facing ICS devices from different
device search engines’ databases.
3
https://fofa.so/
Table 1: Protocols supported by ICScope. X: supported by
device search engine; 4: enriched by ICScope.
Protocol Name
Device Search Engines
Enrich
Shodan Censys FOFA SiNan
Modbus X X X X -
Siemens S7 X X X X 4
DNP3 X X X X -
Niagara Fox X X X X 4
BACnet X X X X 4
EtherNet/IP X - X X 4
GE-SRTP X - X X -
HART-IP X - X X -
PCWorx X - X X 4
MELSEC-Q X - X X -
OMRON FINS X - X X 4
Crimson v3 X - X X -
CoDeSys X - X X 4
IEC 60870-5-104 X - X X -
ProConOS X - X X -
VertX Edge - - X X 4
Moxa NPort - - X X -
Lantronix UDP - - X X -
Koyo DirectNet - - - X -
HollySys MACS - - - X -
Proficy-webspace - - - X -
Siemens License - - - X -
Total 15 5 18 22 8
3.3 Information Extraction
As shown in Figure 2, the information extraction
module consists of three submodules:
3.3.1 Banner Processing
This submodule takes the banners as input and out-
puts the device information. An ICS device entry
can be characterized by the tuple hvendor, product,
versioni. Except for the ICS device entry, device in-
formation also includes feature information for data
enrichment or honeypot detection, i.e., the article
number in the S7 protocol. However, the fields of de-
vice information are different in different ICS proto-
cols. We analyzed 22 ICS protocols and wrote parsers
for them separately.
3.3.2 Data Enrichment
The challenge here is that the banners returned by
search engines are incomplete. To address this, we
take the following three ways to enrich the incomplete
ICS device entry:
Data Enrichment based on Industrial Control Pro-
tocol. Some attributes of ICS device entry might show
as identifiers rather than the exact value in some ICS
protocols. For example, BACnet defines a numerical
”vendor identifier” to arbitrate between non-standard
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
240
messages of different manufacturers. Each vendor
identifier has been assigned to a vendor. For the pro-
prietary protocols, we can identify their vendors di-
rectly, e.g., SIEMENS for S7 protocols. Based on
these observations, we manually build several map-
pings between identifiers and exact values. ICScope
utilizes these mappings to supplement the missing at-
tributions.
Data Enrichment based on Fingerprint Database.
Some banner fields can be used to identify the de-
vice information, such as the product-ID system’s
field. An example is the article number 6ES7
313-6CE01-0AB0 in Figure 1, which is the SIEMENS
product-ID system’s identifier. We utilize these fields
to build the fingerprint database for ICS devices.
Data Enrichment based on Multiple Search En-
gines. Device search engines perform differently
on each ICS protocols. Excepting the number of
discovered devices, the banners can also be differ-
ent. As shown in Figure 1, Shodan can acquire
Basic Firmware, but FOFA not. The strategy is to
integrate the device information extracted from mul-
tiple search engines for complete ones. ICScope
supports four search engines: Shodan, Censys (Du-
rumeric et al., 2015), FOFA, and SiNan
4
.
3.3.3 Honeypot Detection
To achieving accuracy in discovering vulnerable
Internet-facing ICS devices, ICScope detects ICS
honeypots within the results in passive mode. IC-
Scope adopts the following approaches:
Multi-source Comparison. Although the search en-
gines’ scanning time is different, the industrial con-
trol system devices are usually neither upgraded nor
replaced for the duration. So the data information for
an ICS device should be the same at different search
engines. If the data information is inconsistent, we
regard this ICS device as an ICS honeypot. Notice
that if multiple ICS protocols are found for one ICS
device, we can not verify whether it is an ICS honey-
pot based on current features. We regard them as ICS
honeypots to decrease false positives. After inspect-
ing these IPs, we find they might be verified by the
non-ICS ports’ information, such as the HTTP service
ports. We leave it as our future work.
Fingerprinting Detection. We utilize fingerprint-
based methods to detect ICS honeypots. We extract
characteristics from the open-source industrial hon-
eypot as fingerprints, such as default configuration,
prompt information.We also find that some fields in
the ICS protocol are assigned to a unique device, i.e.,
4
An internal device search engine newly developed by
QiAnXin, china in 2018.
Vendor 1
Product 1 Product 2
Product n


<CVE, Version Range>
Figure 3: Data structure used to store vulnerabilities (Vul-
nerability Tree). Green color denotes the information stored
on the leaf.
the MAC in Moxa NPort protocol. ICScope leverages
such fingerprinting to detect ICS honeypots.
ISP-based Detection. ICS honeypots are often de-
ployed on cloud service. However, no cloud ser-
vice provider has provided online industrial control
device services. Therefore, we can reasonably infer
that the devices with an IP belonging to cloud service
providers are ICS honeypots.
3.4 Vulnerability Association
Vulnerability Tree. We adopt a multi-forked tree for-
est to store all vulnerability information, and each tree
represents all vulnerabilities that affected the same
vendor. As shown in Figure 3, each tree takes the
vendor as the root node. The intermediate nodes are
the vulnerable products that belong to the vendor, and
each leaf node represents a related vulnerability. Each
leaf node stores the CVE name and corresponding
version range of the vulnerability. Notice that the dif-
ferent intermediate nodes might denote the same de-
vice. For example, the ICS device in Figure 1 has two
different intermediate nodes (simatic s7-300 cpu
and simatic s7-300 cpu 313) on the vulnerability
tree. For more comprehensive vulnerabilities, we use
multiple vulnerability libraries (NVD, China National
Vulnerability Database of Information Security (CN-
NVD), and SecurityFocus) to construct the vulnera-
bility trees. We also use the data from Exploit-DB,
SecurityFocus, PacketStorm to determine whether
there is a public exploit for a vulnerability.
Associate Vulnerabilities. We utilize the information
extraction results and vulnerability tree to associate
ICS devices with vulnerabilities. We adopt a version-
based associating scheme. We assume that an ICS
device is affected by a vulnerability when its version
is in the vulnerability affected scope. The overall as-
sociation strategy is shaped in Algorithm 1. For each
ICS device, we first use the sim lookup() function
to find similar vendors and products in the vulnerabil-
ity tree. Through the vulnerability tree, we decrease
unnecessary comparisons. Second, we extract all vul-
nerabilities associated with similar vendors and prod-
ucts in the vulnerability tree. For each vulnerability,
we check whether the ICS device’s version is in the
vulnerability affected scope. If the version is affected,
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age
241
Algorithm 1: Associate ICS device with vulnerabilities.
Input: ICS device entry hvendor, product, versioni, Vul-
nerability Tree(V T )
Output: a set of possible CVE names (R
cve
)
R
cve
{}
vt vendors all root nodes of VT
for each vt vendor vt vendors do
if sim lookup(vendor, vt vend or) then
vt products all sub-nodes of vt vendor in V T
for each vt product vt products do
if sim lookup(product, vt product) then
vt vuls all leaves of vt product in V T
for each hcve, versionRangei vt vuls do
if match(version, versionRange) then
R
cve
R
cve
+ cve
end if
end for
end if
end for
end if
end for
we add the CVE name of the vulnerability into the set
of possible CVE names.
4 DEMYSTIFYING
INTERNET-FACING ICS
DEVICES SECURITY STATUS
In this section, we use ICScope to evaluate the se-
curity status of Internet-facing ICS devices in the
whole public IP address space. We first introduce the
datasets used by ICScope and then we analyze the se-
curity status of Internet-facing ICS devices. Finally,
we compare our work with the vulnerability detection
function provided by Shodan.
4.1 The Datasets of ICScope
The data collected by ICScope can be divided into two
parts: data of Internet-facing ICS devices from the de-
vice search engine and vulnerability information from
public vulnerability database. We retrieve Internet-
facing ICS devices from the four device search en-
gines between Dec 2019 and Jan 2020. We ob-
tain 76,489/219,238/78,363/85,531 device informa-
tion from Shodan, Censys, FOFA and SiNan, respec-
tively. The device number of FOFA is much larger
than the others because the time span of the FOFA
is one year, but the others not. The independent IPs
for the above results is 270,283. To monitor Internet-
facing ICS devices, we also conducted the experi-
ments per three months from Jun 2020 to Dec 2020.
We obtain 229,048/225,473/242,034 Internet-facing
ICS devices in Jun 2020, Sep 2020, and Dec 2020,
respectively. We crawl the published vulnerability in-
formation from NVD, CNNVD, Exploit-DB, Secu-
rityFocus and PacketStorm, and merged all the in-
formation into 286,269 distinct enriched vulnerability
records, of which 160,795 are from CVE and 129,653
are from NVD.
4.2 Discovery Approaches Validation
First, we evaluate the performance of our honeypot
detection module. Shodan provides an interface
5
to
query the probability that an IP is a honeypot. Among
the above ICScope datasets, Shodan only ensures
55 IPs as honeypots. We use these IPs as a cross-
validation dataset to validate the honeypot detection
module results. For these ICS honeypots, ICScope
can detect 52 of them (about 94.55%). We manu-
ally inspect the remaining 3 IPs. We notice that the
information extraction module returns none for two
of them. However, our approach is based on the de-
vice information returned by the information extrac-
tion module. The last one seems like a normal ICS
device. The reason might be the limitation of the fea-
tures in our approach.
Secondly, we evaluate the precision of our device
discovery approach. The ICS devices reside in the re-
mote Internet space. We cannot perform active-mode
scanning due to ethical considerations. We also try
to contact several vendors with their devices exposed,
but none of them accept our request to fill a survey
obtaining the ground truth of the exposed devices.
Therefore, we report the 319 vulnerable ICS devices
located in our own country to the relevant Computer
Emergency Response Team (CERT). Until the time of
paper submission, the CERT team has dealt with 59
of them and they confirm that ICScope identifies the
vendor, product and version information and the asso-
ciated vulnerabilities of all 59 devices correctly. The
results show that our device discovery approach has
high accuracy. And we plan to share the ICScope tool
and discovered vulnerable ICS devices to the inter-
ested CERT teams and help to deal with the exposed
and vulnerable ICS devices.
4.3 The Honeypots in Internet-facing
ICS Devices
Among the 270,283 Internet-facing ICS devices, IC-
Scope regards 21,578 of them (about 7.98%) as ICS
honeypots. Among these ICS honeypots, the multi-
source comparison-based approach detects 18,428 of
them (about 85.40%), while the ISP-based approach
5
https://honeyscore.shodan.io/
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
242
detects 3,653 (about 16.93%). Furthermore, the
fingerprinting-based approach detects 1,216 (about
5.64%). Notice that an ICS honeypot might occur
multiple times on the above results.
Vendor Product
Version
2.75% 6.60%
0.77%
2.40%
0.69% 12.86%
73.93%
Figure 4: Attribution-distribution of inconsistent ICS de-
vices. The three circles represent the percentage of devices
with vendor, product, and version inconsistency among all
inconsistent devices.
(a) Vendor recognition rate
at all ICS protocols
(b) Product recognition rate
at all ICS protocols
(c) Version recognition rate
at all ICS protocols
(d) Vendor recognition rate
at BACnet protocols
(e) Product recognition rate
at BACnet protocols
(f) Version recognition rate
at BACnet protocols
Figure 5: Recognition rate of different attributions.
4.4 The Advantage of Multiple Device
Search Engines
Inconsistencies in Device Search Engines. We
first analyze the inconsistencies between the different
search engines. Since the probability of upgrading or
replacing the device during the scanning cycle is ex-
tremely low, the differences indicate that it may be
a honeypot. We are mainly concerned about the in-
consistency of the vendor, product, and version. As
shown in Figure 4, most of the inconsistencies are
related to the version. Devices with version incon-
sistency account for 89.89% of the total inconsistent
devices. It is worth noting that search engines identi-
fied entirely different results on 2.4% of inconsistent
devices. We tag these inconsistent devices as honey-
pots.
Data Enrichment with Multiple Device Search En-
gines. As mentioned in Section 2.2, the results of
search engines, including incomplete data. This prob-
lem can be alleviated by data enrichment. Different
search engines have different ability to recognize dif-
ferent attributes of devices. We extract the attributes
that can be identified by different search engines and
combine them to get complete device information.
Figure 5 shows the different abilities of the four
search engines on recognizing different device at-
tributes. From Figure 5a, Figure 5b and Figure 5c,
we can know that the search engines, except for Cen-
sys, performs well in recognition of vendor and prod-
uct. But all the four search engines are weak in rec-
ognizing version, among which, Censys get the best
recognition rate of 56.6%. Taking the BACnet proto-
col as an example, we can obtain the version informa-
tion from Censys and the other information from the
other three search engines (See Figure 5d, Figure 5e,
Figure 5f).
4.5 The ICS Devices Affected by Public
Vulnerabilities
For the 22 industrial control protocols mentioned in
this paper, we can completely extract vendor, prod-
uct and version information in only 9 of them (See
Table 2). We obtain 215,649 ICS devices using one
of the 9 ICS protocols from device search engines.
After excluding 15,294 devices that may be indus-
trial controlled honeypots, 106,382 (about 53.10%)
devices can extract complete information. For those
devices with complete information, there are 52,739
(about 49.58%) devices affected by one or more vul-
nerabilities. Figure 6 shows the percentage of vul-
nerable ICS devices in different protocols. Next, we
focus on the analysis of the impact of public vulnera-
bilities on these 52,739 devices.
Protocol-distribution of Vulnerable ICS Devices.
Table 2 shows the impact of public vulnerabilities
in ICS devices. We find that about half of Internet-
facing ICS devices are still vulnerable, thus having
the risks of being compromised. More than half of
the ICS protocols have at least 40% vulnerable de-
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age
243
Figure 6: Vulnerable device percentage at protocol-level.
Table 2: Statistics result of vulnerable devices.
Protocol Name Vulnerable Devices
Vulnerable
Percentage
Modbus 3461 28600 12.10%
Siemens S7 4400 7663 57.42%
Niagara Fox 35669 37060 96.25%
BACnet 1918 14405 13.31%
EtherNet/IP 4006 9171 43.68%
PCWorx 1444 3610 40.00%
OMRON FINS 208 2032 10.24%
CoDeSys 1633 3526 46.31%
VertX Edge 0 315 0.00%
Total 52739 106382 49.58%
*Devices: refers to total number of ICS devices with complete vendor,
product and version information.
vices. In particular, the percentage of vulnerable ICS
devices using Niagara Fox protocol, the most seri-
ous protocol, is close to 100%. For Vertx Edge, we
only find one remote code execution vulnerability of
an HID VertX/Edge device in the public vulnerabil-
ity databases, more specific, from SecurityFocus and
PacketStorm. However, there is no detailed impact
scope of this vulnerability. Thus the assessment re-
sult for VertX Edge is 0.00%.
Geo-distribution of Vulnerable ICS Devices. We
adopt the Hilbert curve (Moon et al., 2001) to visu-
alize the vulnerable ICS device distribution in the In-
ternet space. As shown in Figure 7, the distribution
of vulnerable ICS devices is irregular. It is hard to
find any correlation between vulnerable ICS devices
and IP. Although we cannot find any correlation at IP-
layer, we can also investigate the correlation between
spatial location and vulnerable ICS device distribu-
tion. Figure 8 shows the geographical distribution of
Internet-facing ICS devices affected by public vulner-
abilities. These vulnerable ICS devices are distributed
in the vast majority of countries around the world,
among which the United States, Italy, Canada, France
and Spain are on the top. In particular, the United
States is the only country that has over 10,000 vulner-
Figure 7: Hilbert curve heatmap of vulnerable ICS devices
in the IPv4 addresses space.
Figure 8: Geo-distribution of vulnerable ICS devices.
able devices (21,156), which accounts for 40.11% of
all vulnerable Internet-facing ICS devices. We further
analyze the proportion of vulnerable ICS devices in
each country. Among the six continents, North Amer-
ica has the highest proportion of vulner- able ICS de-
vices, with an overall rate of 39.48%. This is mainly
because the proportion of vulnerable ICS devices in
the United States and Canada with more than 1000
ICS devices is as high as 40.72% and 35.48%, respec-
tively. The rest are Australia (20.99%), South Amer-
ica (20.23%), Europe (19.93%), Asia (11.66%) and
Africa (9.15%). As shown in Figure 9, more than 90%
of the Niagara Fox vulnerable devices can be found in
the top 10 countries, and there is a long-tail effect in
their vulnerable ICS devices distribution. There are
similar distribution characteristics in other vulnera-
ble ICS devices. At the city-level, Figure 10 shows
a long-tail effect in vulnerable ICS devices. The Ni-
agara Fox protocol has the most extensive coverage.
More than 43% of the Niagara Fox vulnerable devices
are located in the top 50 cities, 60% of the Niagara
Fox are in the top 134 cities. OMRON FINS has the
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
244
Figure 9: Location distribution of vulnerable ICS devices at
country-level.
Figure 10: Location distribution of vulnerable ICS devices
at city-level.
least coverage. The cities which have ICS devices us-
ing OMRON FINS protocol are only 34.
Vulnerable ICS Devices over Time. Among
the three experiments over six months, we find
46,489/38,466/37,027 vulnerable ICS devices, re-
spectively, and the details are shown in Figure 11.
There is a slowly decreasing trend in the number of
vulnerable ICS devices for most ICS protocols dur-
ing our measurement period. This indicates that peo-
ple had taken some measures to mitigate the risk
of Internet-facing ICS devices, but still not enough
counter-measures. However, the Siemens S7 pro-
tocol shows an anomalously increasing trend in the
number and percentage of vulnerable ICS devices.
This mainly due to the growing number of exposed
SIMATIC S7-1200 devices.
4.6 The Statistics of ICS Vulnerabilities
We then carry on the statistical analysis of the ICS
vulnerabilities mentioned above. We find that only
207 different vulnerabilities affect a total of 52,739
ICS devices. We classify these vulnerabilities accord-
ing to protocols and then further analyze the impact
scope, CVSSv3 score, and life cycle of vulnerabili-
ties under different protocols.
Figure 11: Vulnerable ICS devices over time. The solid
lines represent the number of vulnerable ICS devices. The
dash lines represent the percentage of vulnerable ICS de-
vices.
Top 5 Vulnerabilities in Different ICS Protocols. Ta-
ble 3 shows the top 5 ICS vulnerabilities in different
industrial control protocols. For Niagara Fox proto-
col, the three vulnerabilities we found affect about
96.25% of all Niagara Fox ICS devices with com-
plete information. There are similar characteristics in
PCWorx and CoDeSys protocols. For such vulnera-
bilities, ICS engineers should take measures to fix or
mitigate them as soon as possible.
Figure 12: Severity distribution of ICS vulnerabilities in
different protocols.
The Severity of ICS Vulnerabilities in Different ICS
Protocols. We use the Common Vulnerability Scor-
ing System(CVSS)(version 3) to assess the severity
of ICS vulnerabilities. CVSS version 3.0 was released
in June 2015, and vulnerabilities released before that
date have no CVSSv3 score. 46 out of 207 ICS vul-
nerabilities are released before June 2015, and the ICS
vulnerabilities of OMRON FINS and CoDeSys are all
among them. As shown in Figure 12, the ICS vulner-
abilities with high or critical severity in all ICS pro-
tocols are at least 60%. The severity of vulnerabili-
ties in the PCWorx protocol, the most severe proto-
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age
245
Table 3: Top 5 vulnerabilities in different ICS protocols.
Modbus Siemens S7 BACnet EtherNet/IP
CVE Numbers CVE Numbers CVE Numbers CVE Numbers
CVE-2015-7937 3095 CVE-2017-2680 4384 CVE-2019-18249 1770 CVE-2017-7898 3147
CVE-2015-6461 3077 CVE-2017-2681 4384 CVE-2016-4495 37 CVE-2017-7899 3147
CVE-2015-6462 3077 CVE-2017-12741 4337 CVE-2016-3155 36 CVE-2017-7901 3147
CVE-2018-7241 2462 CVE-2019-10936 4246 CVE-2018-7779 27 CVE-2017-7902 3147
CVE-2018-7242 2462 CVE-2018-13815 3021 CVE-2018-7795 25 CVE-2017-7903 3147
Total vul num: 22 Total vul num: 56 Total vul num: 93 Total vul num: 29
Niagara Fox PCWorx OMRON FINS CoDeSys
CVE Numbers CVE Numbers CVE Numbers CVE Numbers
CVE-2017-16744 35669 CVE-2019-9201 1444 CVE-2015-0987 208 CVE-2014-0760 1633
CVE-2017-16748 35669 CVE-2019-10953 605 CVE-2015-1015 195 CVE-2014-0769 1633
CVE-2018-18985 35669 - - - - - -
Total vul num: 3 Total vul num: 2 Total vul num: 2 Total vul num: 2
*Numbers: refers to the number of devices affected by the vulnerabilities.
Figure 13: Average score and outliers of ICS vulnerabilities
in different protocols.
Figure 14: Vulnerabilities Existing Time.
col, is high or critical. Figure 13 shows the CVSSv3
scores of vulnerabilities in different protocols. The
average scores revolve around 7.5, and the scores of
most vulnerabilities are spread across the 7.5-10 spec-
trum. Among the ICS vulnerabilities, 69.23% are re-
motely exploitable, 74.52% do not require any special
conditions, and 66.35% could be executed without au-
thorization.
The Timeline of ICS Vulnerabilities in Different ICS
Protocols. Figure 14 shows the disclosure time of
the vulnerabilities mentioned above. We can find
that most of the vulnerabilities are disclosed within 3
years, yet 19.80% have been disclosed for more than 5
years. Among these vulnerabilities, we find that there
are 40 vulnerabilities (about 19.32%) which exploits
are published on the Internet. This indicates that these
corresponding affected devices are in a highly danger-
ous state.
4.7 The ICS Devices Affected by 0day
Vulnerabilities
ICScope also supports the evaluation of the impact of
0day vulnerabilities. We use ICScope to evaluate the
impact of a 0day vulnerability we found in Schneider
Electric device. This vulnerability affects four dif-
ferent products, including Modicon M580 (version
3.10). We find that 3,999 Schneider Electric devices
exposed to the Internet are affected by this vulnerabil-
ity. We have reported this vulnerability to the vendor
and waiting for the vendor to fix it.
4.8 Comparison with Shodan
Vulnerability Detection
Shodan also tries to verify whether the target device
is affected by known vulnerabilities. We use this
function of Shodan to detect the vulnerability of the
ICS devices that the complete information can be ex-
tracted. For 106,382 different IP addresses, Shodan
finds only 3,472 (about 3.26%) associated with one
or more vulnerabilities. Moreover, the vulnerabili-
ties detected by Shodan are all Web-based vulnerabil-
ities instead of ICS vulnerabilities. Comparing with
Shodan, ICScope finds that 52,739 Internet-facing
ICS devices (about 49.58%) are affected by at least
one ICS vulnerabilities. As for ICS vulnerability de-
tection, ICScope performs better than Shodan.
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
246
5 RELATED WORK
Passive Vulnerability Assessment in Cyberspace.
Passive vulnerability assessment is an eco-friendly
and resource-conserving research method for large
scale security measurement towards Cyber-Space.
B Genge et al. are the first to propose the
Shodan-based passive vulnerability assessment tool,
ShoVAT (Genge and En
˘
achescu, 2016). The CPE
based vulnerability matching technique used by
ShoVAT is heavily dependent on the quality of its
CPE dictionary. Jamie O’Hare et al (O’Hare et al.,
2019) improved the previous work and proposed
Scout to perform a passive vulnerability assessment
of HTTP services on the Internet. Compared with
ShoVAT, Scout focuses on the accuracy of vulnerabil-
ity matching. By comparing with the scanning result
of OpenVAS, Scout’s results are more practical. How-
ever, neither of the works mentioned above is suit-
able for the industrial control scenario. Their match-
ing techniques depends on the quality of the banners
returned by search engines. However, it is common
that the banners are incomplete in ICS. We mitigate
this problem with our data enrichment module. More-
over, previous works on passive vulnerability assess-
ment rarely consider the honeypot problem. ICScope
try to address it to increase the accuracy of vulnera-
bility detection.
Online ICS Device Discovery. Previous works on the
discovery of online ICS devices are always a hot topic
on ICS security. Mirian et al. (Mirian et al., 2016) ex-
tend ZMap (Durumeric et al., 2013) to support four
ICS protocols and scan the public IPv4 address space
to discover online ICS devices. Differently, Feng et
al. (Feng et al., 2016) investigates 17 industrial con-
trol protocols and develop their ICS device discovery
system to conduct ICS device discovery. Although
both of them have tried to minimize the number of
probing packets, their systems still need to interact
with target devices. By using search engines, ICScope
can discover online ICS devices without any commu-
nication interactions. By using multiple search en-
gines, ICScope can better tolerate the limitation of
scanning nodes.
ICS Honeypot Detection. The honeypots in ICS se-
curity are often used to perform ICS threat analy-
sis (Mirian et al., 2016; Vasilomanolakis et al., 2016;
Serbanescu et al., 2015). There are few works con-
sidering ICS honeypot detection. Feng et al. (Feng
et al., 2016) propose a learning model to determine
the probability of an ICS honeypot and a heuristic al-
gorithm to verify it with the least number of packets.
However, both the learning model and the algorithm
are based on the probing packets. The ICS honeypot
detection approach used in ICScope is based on the
non-interaction features. So it has no probing packet
with ICS devices.
6 DISCUSSION
Ethical Considerations. From an ethical perspective,
we perform our measurements in a completely passive
mode to avoid any potential damage. The data used
for ICScope contains no sensitive data, and the data
provided by public search engines are publicly avail-
able. However, because the vulnerable ICS devices
are still unfixed, we cannot publish our datasets and
over 52K vulnerable ICS devices directly. Instead, we
report vulnerable ICS devices in our country to rele-
vant CERT, and we also try to contact several vendors.
Limitations. First, the above experiment results only
show a subset of the security status of Internet-facing
ICS devices. The actual situation should be more se-
rious. This is a limitation because it is difficult or
impossible to obtain all Internet-facing ICS devices
via search engines, and not all ICS devices can be ex-
tracted full information. Second, while we obtain vul-
nerable Internet-facing ICS devices, we still do not
confirm whether there is any defensive measure on
them. Because ICScope is based on version to iden-
tify vulnerabilities rather than proof of concept(PoC).
Third, the supporting ICS protocols of ICScope is
limited to the ability of search engines. ICScope sup-
ports one ICS protocols only if its banners contain key
fields. Fourth, a lack of raw probing packets makes
it harder to extract device information or detect ICS
honeypots because the banners generated by search
engines might lose some information.
Honeypot Detection. It is hard to detect ICS honey-
pots accurately in passive mode. For example, ICS
devices might share one IP by port forwarding. How-
ever, this is also a fingerprint to detect honeypots. We
cannot distinguish them without the help of probing
packets. So we regard all of them as ICS honeypots
to reduce false positives. Fortunately, we find that this
problem might be corrected by the data from non-ICS
port, and we leave it as our future work.
Mitigation. We need to take some defensive mea-
sures to mitigate the risk of Internet-facing ICS de-
vices. The intuitive measure is to keep the latest ver-
sion of each ICS device. However, this is difficult or
impossible in the actual ICS environment. Instead, we
can take the following defensive measures:
Add Firewall Policy. By adding firewall to restrict
source IP and destination port, we can prevent the at-
tack from accessing the vulnerable ICS devices.
Network Isolation. Isolate the ICS devices that need
From Exposed to Exploited: Drawing the Picture of Industrial Control Systems Security Status in the Internet Age
247
Internet connection from the ones that do not need
to prevent the attacker from accessing the vulnerable
devices.
7 CONCLUSION
As a crucial component of modern city infrastructure,
the industrial control system exists in every corner of
our life. However, the Internet-facing ICS devices
are under the risk of known vulnerabilities. In this
work, we develop ICScope to discover the vulnerable
Internet-facing ICS devices. Base on the results, we
perform a comprehensive analysis of the security sta-
tus for online ICS devices. We find that 49.58% of
Internet-facing ICS devices that we can extract com-
plete device information are affected by known vul-
nerabilities. The most serious ICS protocol is Niagara
Fox, which proportion of vulnerable devices is even
as high as 96.25%. We observe that most of the vul-
nerable devices are affected by the same vulnerabil-
ity in Niagara Fox, PCWorx, and CoDeSys protocols.
In all ICS protocols, at least 60% of the ICS vulner-
abilities are with high or critical level severity. We
also observe a slowly decreasing trend in the number
of vulnerable ICS devices during our six-month mea-
surement period. Moreover, our measurement results
only present the lower limit of the actual situation. In
response to these severe industrial control security is-
sues, we also discuss the mitigation measures, such as
add firewall policy.
ACKNOWLEDGEMENTS
We would like to thank all anonymous reviewers for
their valuable feedback that greatly helped us improve
this paper. Besides, we would like to thank Yuxiang
Lu, Zhenbang Ma, Yu Wang, for their helping in our
work.
REFERENCES
Comer, D. E. and Lin, J. C. (1994). Probing tcp implemen-
tations. In Usenix Summer, pages 245–255.
Di Pinto, A. A., Dragoni, Y., and Carcano, A. (2018). Tri-
ton: The first ics cyber attack on safety instrument sys-
tems. In Proc. Black Hat USA, pages 1–26.
Dong, Y., Guo, W., Chen, Y., Xing, X., Zhang, Y., and
Wang, G. (2019). Towards the detection of inconsis-
tencies in public security vulnerability reports. In 28th
{USENIX} Security Symposium ({USENIX} Security
19), pages 869–885.
Durumeric, Z., Adrian, D., Mirian, A., Bailey, M., and Hal-
derman, J. A. (2015). A search engine backed by
internet-wide scanning. In Proceedings of the 22nd
ACM SIGSAC Conference on Computer and Commu-
nications Security, pages 542–553. ACM.
Durumeric, Z., Wustrow, E., and Halderman, J. A. (2013).
Zmap: Fast internet-wide scanning and its security ap-
plications. In 22nd {USENIX} Security Symposium
({USENIX} Security 13), pages 605–620.
Fachkha, C., Bou-Harb, E., Keliris, A., Memon, N. D., and
Ahamad, M. (2017). Internet-scale probing of cps:
Inference, characterization and orchestration analysis.
In NDSS.
Feng, X., Li, Q., Wang, H., and Sun, L. (2016). Character-
izing industrial control system devices on the internet.
In 2016 IEEE 24th International Conference on Net-
work Protocols (ICNP), pages 1–10. IEEE.
Genge, B. and En
˘
achescu, C. (2016). Shovat: Shodan-
based vulnerability assessment tool for internet-facing
services. Security and communication networks,
9(15):2696–2714.
Guo, G., Zhuge, J., Yang, M., Zhou, G., and Wu, Y. (2018).
A survey of industrial control system devices on the
internet. In 2018 International Conference on Inter-
net of Things, Embedded Systems and Communica-
tions (IINTEC), pages 197–202. IEEE.
Leverett,
´
E. and Wightman, R. (2013). Vulnerability inheri-
tance programmable logic controllers. In Proceedings
of the Second International Symposium on Research
in Grey-Hat Hacking.
Mirian, A., Ma, Z., Adrian, D., Tischer, M., Chuenchujit,
T., Yardley, T., Berthier, R., Mason, J., Durumeric,
Z., Halderman, J. A., et al. (2016). An internet-wide
view of ics devices. In 2016 14th Annual Conference
on Privacy, Security and Trust (PST), pages 96–103.
IEEE.
Moon, B., Jagadish, H. V., Faloutsos, C., and Saltz, J. H.
(2001). Analysis of the clustering properties of the
hilbert space-filling curve. IEEE Transactions on
knowledge and data engineering, 13(1):124–141.
O’Hare, J., Macfarlane, R., and Lo, O. (2019). Identifying
vulnerabilities using internet-wide scanning data. In
2019 IEEE 12th International Conference on Global
Security, Safety and Sustainability (ICGS3), pages 1–
10. IEEE.
Samtani, S., Yu, S., Zhu, H., Patton, M., and Chen, H.
(2016). Identifying scada vulnerabilities using pas-
sive and active vulnerability assessment techniques.
In 2016 IEEE Conference on Intelligence and Secu-
rity Informatics (ISI), pages 25–30. IEEE.
Serbanescu, A. V., Obermeier, S., and Yu, D.-Y. (2015). Ics
threat analysis using a large-scale honeynet. In 3rd In-
ternational Symposium for ICS & SCADA Cyber Se-
curity Research 2015 (ICS-CSR 2015) 3, pages 20–30.
Vasilomanolakis, E., Srinivasa, S., Cordero, C. G., and
M
¨
uhlh
¨
auser, M. (2016). Multi-stage attack detec-
tion and signature generation with ics honeypots.
In NOMS 2016-2016 IEEE/IFIP Network Opera-
tions and Management Symposium, pages 1227–1232.
IEEE.
ICISSP 2021 - 7th International Conference on Information Systems Security and Privacy
248