Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Design, Implementation and Evaluation of

a Novel Anti-Virus Parasitic Malware

Byungho Min Vijay Varadharajan


Advanced Cyber Security Research Centre
Macquarie University, Sydney, Australia
{byungho.min, vijay.varadharajan}@mq.edu.au

ABSTRACT to take appropriate defensive actions to commensurate with


In this paper, we propose an advanced malware, anti-virus their anticipated risk. Furthermore the malware themselves
parasitic malware (AV-Parmware). It attacks protected com- are becoming increasingly more sophisticated and evasive.
ponents of anti-virus software by their exploiting security In order to be able to protect against such sophisticated mal-
weaknesses, and compromises the target systems by being ware, we as security professionals need to better understand
a parasite on the anti-virus. We have investigated 18 anti- how new malware is being designed and embedded into var-
virus solutions from seven major anti-virus software ven- ious systems. Hence in this paper, our aim is first to look at
dors and have discovered that 12 products from four vendors some new techniques that can be used to develop advanced
(AVG, Avira, McAfee, and Symantec) have certain security malware which can evade existing anti-virus software, and
weaknesses that can be utilised in the proposed malware1 . then develop new defensive mechanisms to counteract them.
There are several advantages to being an anti-virus parasitic From the point of view of malware design, one of the ma-
malware, including longevity (anti-virus runs while its sys- jor hurdles is to perform its functions without being detected
tem is up), improved stealthy behaviour, highest privileges by anti-virus software. Most anti-virus solutions nowadays
and capability to bypass security measures such as Egress use both signatures and behavioural characteristics to de-
filtering. We have implemented our proposed parasitic mal- tect malware; they perform static and/or dynamic analy-
ware, and have shown that all these advantages are achieved sis on files and network traffic. Furthermore, efforts on de-
in practice. tecting obfuscated malware are continuously improving [33,
24, 36, 17, 18]. In addition, recent operating system de-
velopments have begun to incorporate security features like
Categories and Subject Descriptors Early Launch Anti-Malware [3] that activate the anti-virus
K.6.5 [Management of Computing and Information software as early as possible.
Systems]: Security and Protection—Invasive software In this paper, first we propose a new approach to load-
ing malicious components, and develop an advanced mal-
General Terms ware based using this approach, which we refer to as anti-
Security, Experimentation virus parasitic malware (AV-Parmware). It is a software
component-based parasitic malware that subverts and lives
Keywords on major anti-virus software by updating a protected com-
ponent of the target anti-virus with a malicious one. Our
Parasitic Malware, Anti-Virus, Software Components, Vul- scheme performs file update/replacement by exploiting var-
nerability ious security weaknesses that we have found in 12 current
anti-virus solutions from four major anti-virus vendors (AVG,
1. INTRODUCTION Avira, McAfee, and Symantec) with our assessment tool.
Recent malware trends [19, 20] indicate there is an in- We have conducted extensive experiments with the anti-
crease in the instances of malware on the web over the last virus parasitic malware and confirmed its properties.
few years. It is important for security professionals to be up The rest of this paper is structured as follows. We in-
to date on the current and future malware threats to be able troduce the concept of component-based parasitic malware,
1 and describe the advantages and obstacles to anti-virus par-
We are firm believers in responsible disclosure, and have re-
ported the issues disclosed in this paper to the corresponding asitic malware in Section 2. The design and operation of
vendors. AV-Parmware are explained in Section 3. Section 4 describes
the security weaknesses that we have found in several of the
major anti-virus software, which will be used in the devel-
Permission to make digital or hard copies of all or part of this work for opment of the proposed malware in the next section. Ex-
personal or classroom use is granted without fee provided that copies are perimental results are discussed in Section 5. Related work
not made or distributed for profit or commercial advantage and that copies and conclusions are given in Sections 6 and 7 respectively.
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee. 2. BACKGROUND
SAC’15 April 13-17, 2015, Salamanca, Spain. 2.1 Component-based Parasitic Malware
Copyright 2015 ACM 978-1-4503-3196-8/15/04...$15.00.
http://dx.doi.org/10.1145/2695664.2695683 ...$15.00. There have been many attacks in the real world that in-
2127
volve loading malicious components in the context of be- This implies that replacing one of anti-virus files hardly trig-
nign software. For instance, several of these attacks drop gers an alarm for such solutions. Sixth, in many cases, anti-
malicious components at a specific path, which gets loaded virus is the only security tool installed by the user (firewall
via DLL hijacking due to the unsafe path search of Win- is often built in with major operating systems). Therefore,
dows [16]. Also, many malware instances inject their code security of commodity PC systems heavily depends on the
into the target process using memory-related techniques such effectiveness of the anti-virus software installed on them.
as DLL injection thereby loading the malicious components. Besides these obvious advantages, anti-virus has other de-
Hence the anti-virus industry has been developing several sirable properties for parasitic malware such as minimised
techniques to detect such attacks [12, 13, 32, 9, 15]. user intervention and deep integration with the operating
In this paper, we propose a new approach to loading ma- system. These major merits have been achieved with the
licious software components, which we refer to as “software proposed AV-Parmware in Section 5.
component-based parasitic malware.” It is defined as ma-
licious code that updates/replaces existing components of 2.3 Self-protection of Anti-virus
installed software and becomes parasitic on it. Our method Though it is clear that anti-virus offers several advantages
is a file-based attack, and hence it is different from memory- as a host for a parasitic malware, components of anti-virus
based attacks. Furthermore, the proposed scheme is dis- are much harder to replace than those of other software.
tinguished from DLL hijacking even though both are file- This is because as one would anticipate anti-virus software
based attacks; our scheme is based on the fundamental rule, tends to have strong self-protection feature3 . Self-protection
followed in the software industry, that a software compo- module protects its processes, memory, files, registry keys
nent can be updated as long as the new version provides and values, and configurations from being killed, modified,
the full functionality of the current version, whereas DLL and/or deleted. For instance, non-anti-virus process cannot
hijacking uses the unsafe path search weakness. Moreover, terminate the processes of anti-virus or modify its memory
it is hard to determine whether an update is malicious or even if it has the highest privilege (SYSTEM). Furthermore,
not, because there are many possible update channels in- anti-virus related files and registry keys and values cannot
cluding centralised third-party software management tools, be modified or deleted by non-anti-virus processes.
application’s built-in update feature, and user manually re- Self-protection feature is usually implemented with one or
installing a newer version. Hence the fundamental objective more kernel-mode drivers and Windows services. For exam-
of the scheme is to drop a file and inject code into a target ple, a service and a couple of drivers are in charge of self-
process in such a way that it gets loaded by benign software. protection in Avira AntiVirus 2013 product line [22]. Real-
Time Protection service is crucial in Avira’s self-protection
2.2 Parasite on Anti-virus mechanism, because it provides real-time protection not only
There can be various software that can be the target of to the system (such as on-access detection of malware), but
component-based parasitic malware. However, anti-virus also to itself (self-protection such as prevention of unautho-
software is the best target for such malware, since there are rised alteration on Avira’s processes, files, and registry keys).
many advantages to anti-virus parasitic malware. First, an In particular, unloading the kernel-mode drivers and the fil-
anti-virus runs while the system (on which it is deployed) is ter driver is blocked by this service. Furthermore, users can-
running. Therefore, it is one of the best hosts for a parasitic not stop, pause, or restart the service, as the service ignores
malware. Second, it is stealthier, because it does not create such requests. In contrast, some anti-virus software such
any new process while it is running; the word parasitic comes as Sophos End Point Security does not protect its services,
from this characteristic. Also it does not need to install any thereby enabling users or malware to start/stop/restart its
malicious Windows service, driver, and/or rootkit for per- services. In this case, it is possible for a malware to stop
sistent compromise since it will be loaded by anti-virus on the service, perform any malicious activity, and then start
every boot. This implies that the system integrity monitor- the service. The kernel-mode drivers of Avira protect its
ing solutions that monitor new services, drivers and rootkits registry keys and files. In particular, a filter driver protects
do not detect AV-Parmware. In addition, in case of detection, Avira’s installation path so that any other process cannot
anti-virus software will be found to be as “malware” (and not add, delete, or modify any of Avira’s files. And the pro-
AV-Parmware). Third, it is able to acquire more capabilities cesses of Avira themselves are protected by the kernel-mode
as anti-virus related services, processes, and drivers usually driver in such a way that other processes cannot kill or termi-
have the highest privilege on a system; so parasitic malware nate them. Therefore, no one can modify any component of
achieves this without installing rootkit or other malicious Avira and other self-protected anti-virus without disabling
components. Hence, security features like User Access Con- the self-protection feature.
trol (UAC) becomes ineffective. Also, anti-virus has the This self-protection feature is a major obstacle to anti-
ability to protect itself (described in Section 2.3), so that virus parasitic malware, since the malware must update a
other processes cannot stop, kill or disable the anti-virus; protected component of its target anti-virus. Therefore, we
this means AV-Parmware receives this protection once it suc- have developed a tool to assess the self-protection of major
cessfully becomes parasitic on the anti-virus. Fourth, it can anti-virus solutions. This led to the discovery of several new
bypass any Egress packet filtering, because anti-virus solu- security weaknesses, which are described in Section 4.
tions require network communications for updates and sys-
tem protection, and are allowed by outgoing traffic rules. dates every four hours [1, 2].
3
Fifth, anti-virus files are frequently updated in order to be Some anti-virus’ self-protection feature prevents the system
up-to-date2 and these files are normally excluded from file from being restored using Windows System Restore. This
integrity monitoring by security solutions such as Tripwire. is why most anti-virus solutions provide a way to disable its
self-protection feature in its GUI. CLI-based methods are
2
For example, McAfee and Avast free products look for up- not provided as they can be abused by malware.
2128
ware’s installation, which is one of the core requirements of
a stealthy malware.

4. EVALUATION OF SELF-PROTECTION OF
MAJOR ANTI-VIRUS SOFTWARE
Methodology.
As mentioned earlier, AV-Parmware must be able to dis-
able self-protection feature of the target anti-virus. The self-
protection feature of all the anti-virus software that we have
investigated has been implemented in kernel-mode drivers
and/or Windows services. We have developed an assess-
ment tool that checks whether these self-protection modules
are properly protected or not. To sum up, this tool at-
tempts several techniques given below with different options
Figure 1: Overview of AV-Parmware’s design and con- to stop/pause/disable drivers and services, modify config-
trol flow urations, and manipulate files during an update; details of
each method are given in the following sections with the
corresponding vulnerable anti-virus software products:
3. DESIGN: STAGED MALWARE ARCHI-
1. Filter Manager: APIs and fltmc command (AVG)
TECTURE
Attackers have resorted to staged malware in order to over- 2. Manipulation on files during an update: I/O APIs
come the hurdles mentioned in Section 1. In particular, only (Avira and McAfee [2nd])
a dropper or a downloader is executed at the first intrusion; 3. Service Control Manager (SCM): APIs and sc com-
then the malware checks the target environment and mod- mand (McAfee [1st])
ifies its behaviour depending on the state of the environ-
ment (such as enabling a specific anti-virus bypassing tech- 4. Windows Management Instrumentation (WMI): code
nique or terminating itself to avoid detection). Most of the in VBS and C# and WMI console (Symantec Norton)
well-known modern malware including Zeus [13], Citadel [6], 5. net command (anti-virus w/o self-protection such as
Conficker [32] and Stuxnet [12] are staged ones. The staged Sophos)
approach at least gives two benefits to the attackers. First,
it allows them to minimise the risk of detection by security After running the tool, we analysed the results and trans-
tools. Second, it minimises the exposed portion and hides formed the newly found flaws into security vulnerabilities
the full functionality of the malware at the early stage of in- so that AV-Parmware disables these anti-virus solutions’ self-
trusion. Consequently, security professionals cannot reveal protection. Since the malware must update an anti-virus
the functionalities of a malware when they obtain only the related file in order to become a parasite, our research was
first stage such as downloader. focused on the evaluation of self-protection on file access.
AV-Parmware is a staged malware that is made up of two
components: a dropper and a payload DLL as shown in Evaluation results.
Figure 1. The dropper (light grey box in the figure) is the We have investigated self-protection feature of 18 anti-
first stage that is executed on the target system. In this virus solutions from seven major anti-virus vendors; our re-
paper, we assume that AV-Parmware has been introduced sults are shown in Table 1. Top seven worldwide market
to a victim machine by an initial attack such as client-side share vendors have been selected according to the OPSWAT
attack or spear phishing. This is valid because any malware anti-virus vendor market share report [27]. Their market
must be delivered and executed by an initial attack vector; share accounts for 68.3% of the entire anti-virus market.4
most related work assume the same situation [23, 26, 30, 25, As a result, 12 products from four vendors have been iden-
8, 7, 14, 29, 11, 38]. The role of the dropper is to disable tified to have security issues in their self-protection feature,
the self-protection of the target anti-virus and update a file which allows the malware to disable the feature and replace
of the anti-virus with its payload DLL embedded in itself. anti-virus-owned files. The security weaknesses found are
After a reboot or an update, the payload DLL is loaded by explained in the following sections. All the weaknesses ex-
a service of the target anti-virus, and AV-Parmware becomes cept for one of Avira’s (part 1 of Section 4.2) are publicly
parasitic. From this point on, the malware performs its ma- disclosed in this paper for the first time.
licious activities, such as Windows password hash dump and
data exfiltration, in the context of the anti-virus service pro- 4.1 AVG
cess. The overall flow of the execution of AV-Parmware is AVG’s self-protection feature is named as “AVG Self pro-
also illustrated in Figure 1; in this figure, the payload DLL, tection”; two filter drivers are in charge of protecting access
a target file to be replaced, and a target anti-virus service on AVG’s files. When a user or a process (even with the
are shaded in dark grey. The execution flow of AV-Parmware highest privilege) tries to delete or modify those files, then
is hard to detect, because all the operations performed by the two drivers block such I/O requests. Similarly, new files
its dropper – from configuring a service and/or a driver to cannot be added to the AVG folder.
triggering an update and to opening and copying files – are 4
We have excluded Microsoft, as it is not a security-specific
totally legitimate. It also ensures the stability of the mal- software vendor.
2129
Table 1: Evaluation results of anti-virus self-protection feature (SP: self-protection, IS: Internet Security)
Products (2013 versions) SP disarmable Security weakness
AVAST (3 products)
X -
(Free, Pro & IS)
AVG (3 products) O
Unloadable/detachable drivers
(AntiVirus Free, AntiVirus & IS) (Arbitrarily)
Avira (3 products) O
Service restart + restart coercion
(Free, Antivirus Premium & IS) (On every update)
ESET (1 product)
X -
(NOD32 Antivirus 6)
Kaspersky (2 products)
X -
(Anti-Virus & IS)
McAfee (3 products) O 1. Incomplete protection on SP service
(AntiVirus Plus, IS & Total Protection) (On next reboot) 2. Unprotected temporary service
Symantec Norton (3 products) O
Incomplete protection on SP driver
(AntiVirus, IS & 360) (On next reboot)

Weakness: Detachable drivers. This implies that the probability for a malware to be landed
AVG fails to protect the self-protection filter drivers, even on the system during a vulnerable time frame is 10.71%
though it must block any unauthorised attempt to unload (=3/28) at maximum. Therefore, it is impossible to assume
or detach such filter drivers. As a result, two filter drivers that a service restart will happen when the malware trig-
named avgtp and avgmfx86 can be detached by the user or gers an update. In conclusion, without a method to force
malware using Filter Manager Control (fltmc) command. a service restart, all the malware can do is to try to trigger
After the detachment, AV-Parmware replaces an AVG’s file an update, and hope that the Real-Time Protection service
with its payload DLL. Once the malicious DLL is loaded restarts.
by the AVG on the next reboot or at the restart of AVG, It is clear that it is necessary to restart the service, and
AV-Parmware becomes parasitic on AVG. This shows intro- this is where the second security weakness comes into play.
ducing new security modules (filter drivers) cannot provide We investigated the reason for service restart, and discov-
additional protection, if they are not protected by either ered that Real-Time Protection service is temporarily termi-
themselves (the cases of non-vulnerable products in Table 1) nated by the update process (update.exe) when the service
or other modules that protect themselves (the case of Avira). is locking one or more component files. We further anal-
ysed this issue and found that the update process terminates
4.2 Avira Real-Time Protection service even if some files related to the
Avira’s case is much more complicated than the AVG’s. service are opened by any random process; it does not check
As described in Section 2.3, Real-Time Protection service which process is opening these files, and falsely assumes that
prevents its filter driver from being unloaded or detached the files are being opened by the service. It is also worth not-
in all Avira’s anti-virus products. Therefore, AV-Parmware ing that any process can open and read Avira’s files, even
cannot arbitrarily unload the driver in one step. After in- though its installation path is protected by the file system
vestigation, we found two security weaknesses that may not filter driver, and hence files in it cannot be modified or re-
be severe threats in themselves, but become a critical one moved. AV-Parmware exploits this wrong assumption, and
when combined. The first weakness has a big impact (com- forces the service to restart. After triggering an update, it
promise of Avira and the operating system) with low prob- opens a few files that are opened by Real-Time Protection
ability, while the second one increases the probability, thus service. This makes Avira’s update process temporarily ter-
making the risk caused by the combined weaknesses to be minate the service. Then, the malware performs its chained
high and usable by AV-Parmware. exploit, and closes the files. If it does not close them, the
Weakness Pt. 1: Partial pause of self-protection (known). update process keeps stopping and starting the service. This
Byungho et al. have discovered that a core component behaviour can also be used for DoS on Avira, but it is almost
of Avira can be paused during an update, which leads to a meaningless for the same reason as the one given earlier.
total compromise of Avira; full details can be found in their Now AV-Parmware has the ability to compromise the vic-
work [22]. To summarise, Avira’s Real-Time Protection ser- tim system with much higher probability, because virus def-
vice is restarted on some updates. This restarting usually initions and detection engines are frequently updated as dis-
takes several seconds on a commodity personal computer; cussed in Section 2.2. As long as there is a file to be updated,
during this time period, the system and Avira become vul- the malware compromises its victim system by exploiting the
nerable, since the system virtually becomes identical to the two security weaknesses illustrated in this section.
one with no anti-virus. 4.3 McAfee
Weakness Pt. 2: Wrong assumption during pause. McAfee’s anti-virus products host a service named “McAfee
Even though the impact of the security weakness explained Access Protection” (McAPExe) to provide protection on McAfee-
in the previous section is big, its probability is low because owned files and services so they are not accidentally manip-
such a service restart doesn’t happen on every update. We ulated by the user or malware. This service also protects it-
observed Avira Free Antivirus 2013 for more than three self from being stopped or restarted. Therefore, the method
months, and found that the restart happens up to three used against AVG is ineffective against this service. Self-
times in a week (usually less than twice in a week). Because disarm coercion used against Avira is also not possible here
this product checks for an update every six hours (four times even though McAfee restarts Access Protection service on
a day), one week is divided into 28 (=4×7) time frames. some updates. This is because the frequency is too low and
2130
we could not find a way to force a service restart. However, ture for management data and operations on Windows sys-
we have discovered two methods that can be used to nullify tems, and is useful in enterprise applications and adminis-
McAfee’s self-protection feature. trative scripts [5]. We found that Symantec Norton products
fail to block service/driver configuration performed via this
1st Weakness: Incomplete protection on McAPExe.
interface. As a result, BHDrvx86 driver can be disabled by
McAfee products do not protect the configurations of their
AV-Parmware through WMI. Other steps are exactly same as
services. Even though the user (or malware) cannot stop
in the McAfee’s case; AV-Parmware queues a file move opera-
or restart Access Protection service, its configurations such
tion, and the payload DLL is dropped in C:\Program Files\
as “start mode” can be modified using SCM. Because the
Norton AntiVirus\Engine\20.4.0.40\ on the next reboot.
start mode is changed to “disabled”, the service will not start
The path is built at runtime since it includes a version num-
from next reboot. However, when AV-Parmware disables Ac-
ber (20.4.0.40).
cess Protection service, it cannot replace McAfee’s files since
the service is active at this moment. This hurdle can be 4.5 Other products
overcome using MoveFileEx() API with MOVEFILE_DELAY_
UNTIL_REBOOT flag. This flag makes the file move operation Even after running our assessment tool, and then manu-
happen on next reboot, and this operation succeeds since ally analysing them, we could not disable the self-protection
Access Protection service does not start in the next reboot. feature of the three vendors’ products (Table 1). Their
In the current implementation of AV-Parmware, it replaces a self-protection modules (“self-defense” in AVAST and ESET
file in C:\Program Files\Common Files\McAfee\Platform NOD32, “self-defence” in Kaspersky) blocked all the access
to load the payload DLL. As soon as it is loaded, the DLL requests to their components and configurations unless such
recovers start mode of Access Protection service thus mak- requests are originated from themselves; additionally we were
ing everything normal. not able to restart any of their services/drivers during an
This security weakness shows that not only it is important update. However, our assessment tool has found a few im-
to protect the running state of a service or a driver but also proper protection on services and temporary files from these
configurations need to be protected, so as not to be modified products, which we believe has the potential for developing
by malware or malicious users. the parasitic malware; we are in the process of analysing
these results.
2nd Weakness: Unprotected temporary service file.
We analysed McAfee products and found another weak- 5. EXPERIMENTAL RESULTS
ness that can be used by AV-Parmware. During some up-
dates, McAfee installs a temporary service to clean up the In this section, we present our experimental results that
leftovers of the update as well as itself (after the clean- show the advantages of our anti-virus parasitic malware.
up). Such a service creation happens quite frequently (about The victim environment is a virtual machine with fully patched
three times out of five updates according to our observa- Windows 7 Ultimate edition.
tions). This clean-up service is executed on the next reboot 5.1 Bypassing Egress Filtering
before any other McAfee service including Access Protection
This experiment shows that AV-Parmware bypasses Egress
service begins. It executes a file located in %WinDir%\Temp\
filtering rules, and connects back to the attacker. In or-
[dynamic_filename].exe. The problem is that this partic-
der to block malware communications, Egress filtering is
ular executable file is not protected, even though McAfee
required in PCI-DSS [28] and recommended in US-CERT
protects its other temporary files during the update. There-
guideline [34]. As a result, advanced modern malware in-
fore, AV-Parmware can use this weakness by replacing the
stances inject themselves into trusted processes so as to by-
original service file with the dropper. On the next reboot,
pass these security mitigations [12, 9, 15]. However, this is
the clean-up service starts and the dropper is executed in-
insufficient in some cases. For instance, Flame injects into
stead of the original executable file. The dropper replaces
Windows Explorer that does not make a network connection
a file of McAfee with its payload DLL, and AV-Parmware
in a usual situation. Stuxnet uses lsass.exe and winlogon.
becomes parasitic on McAfee when the payload is loaded
exe as its injection target, but they normally don’t con-
by the McAfee service. This method shows how incomplete
nect to the Internet, even though they can connect to LAN
protection on crucial temporary files leads to a compromise
servers and/or can be connected from LAN clients. In con-
of anti-virus and operating system.
trast, Avira’s update process and Web Protection service
4.4 Symantec Norton are required to connect to the Internet for their functionali-
ties like downloading new updates and protecting the system
In Symantec Norton software, a driver (BHDrvx86) pro-
from web threats.
tects its files. This driver is a part of “Norton Product
In the case of targeting Avira, AV-Parmware uses sqlite3.
Tamper Protection” feature that provides self-protection to
dll to compromise Real-Time Protection service (avguard.
Norton products. Unlike McAfee, Norton tries to block any
exe). The problem is that this service does not require
attempt to modify configurations of the driver. In partic-
network connections. However, sqlite3.dll is also loaded
ular, a user or malware cannot modify the start mode of
by Avira’s other processes such as Web Protection service
BHDrvx86 via SCM (using sc command or APIs) or net
(avwebgrd.exe). This service acts as a proxy server on the
command. In addition, the driver cannot be detached using
system, and connects to the Internet on behalf of applica-
any of the documented methods.
tions that make network connections to the Internet. As a
Weakness: Incomplete protection on BHDrvx86. result, Web Protection service must be allowed in outgoing
However, there is another technique that can be used traffic rules. In order to utilise this, the payload DLL of
to control configurations of drivers and services: Windows AV-Parmware behaves differently based on its caller process;
Management Instrumentation (WMI). It is the infrastruc- it checks the full path of its caller, and creates the mali-
2131
cious thread only if the caller is C:\Program Files\Avira\ 6. RELATED WORK
AntiVir Desktop\avwebgrd.exe. We verified that this pay-
load bypassed Egress filtering firewall rules forced by Co- Loading Malicious Components.
modo Firewall. Kwon and Su [16] have experimented with DLL hijacking
attack against major software suites such as Microsoft Of-
It is worth noting that a single file (ccSvcHst.exe) hosts fice and Google Desktop, and have suggested a method that
all the services in the case of Symantec Norton, thus making detects unsafe library use in software. Analyses on several
bypassing Egress filtering easier; any process of this file must malware instances have shown that malicious components
be allowed in outgoing traffic rules. are loaded via memory manipulation like DLL injection, and
anti-virus software vendors are trying to detect such injec-
5.2 Remaining Stealthy (even on detection) tion and recommending prevention methods such as remov-
There are two aspects to the stealth nature of the AV- ing debug privilege from non-administrative users. [12, 13,
Parmware. First, it does not create any process after the pay- 32, 9, 15].
load is loaded by anti-virus, and hence the user cannot see Attacking Anti-virus.
any suspicious process running on the system. Also, there Traditional implementation vulnerabilities such as buffer
is no additional file installed on the target system, as the overflow have been discovered from anti-virus and published [8,
malware updates anti-virus’ file with its payload. When it 7, 35, 37]. These attacks exploit such vulnerabilities, and
comes to network activity, the user and firewalls can hardly perform denial-of-service or arbitrary code execution. In
detect any anomaly as long as the Command and Control addition, there has been considerable research on anti-virus
server of the malware resides in an unsuspicious country - evasion techniques using code obfuscation, such as poly-
because some anti-virus services and processes (Web Pro- morphism, metamorphism and packing, as well as on de-
tection service in the case of Avira) have to connect to the tection schemes [33, 24, 36, 17, 18, 25]. File obfuscation
Internet. techniques for non-executable formats have also been pro-
The second aspect comes out in the worst case, which is posed. Porst suggested an evasion technique that generates
at the time of detection. When an anti-virus detects the script-embedded PDF files that are not recognised by anti-
malware, it reports itself as malicious because AV-Parmware virus software, but parsed by Adobe reader [29]. Jana and
is parasitic on it. This behaviour was confirmed using a Shmatikov [14] suggested two anti-virus evasion techniques
detectable payload. We implemented a payload DLL that that use various file formats. However, malware using any
loads another DLL located at c:\a.dll. This additional file obfuscation technique can bypass only the file detection,
DLL is a Metasploit’s Meterpreter shellcode that connects and thus its activities can be detected by anti-virus’ be-
back to the attacker. Upon the malware attack, malicious havioural engines.
sqlite3.dll was loaded by Avira’s Real-Time Protection
service, and a.dll was loaded by the payload DLL. Un- Attacking Software Update.
til this point, the obviously malicious file was not detected Bellissimo et al. [10] analysed update procedures of sev-
by Avira. This is an unusual behaviour because Avira’s eral software products, and showed some of them don’t use
Real-Time Protection service normally scans and blocks the secure network connections and/or binary verification. Sim-
access on a suspicious file when an access is requested. In ilar research was started by Infobyte security research in
other words, a.dll cannot be executed if non-Avira pro- 2008, and its product, evilgrade, is still actively updated
cesses try to load it. It turned out that Avira requires with vulnerable software that uses insecure network con-
SQLite to open its definition files prior to becoming able nections. However, these network-based man-in-the-middle
to perform on-access scanning, and a.dll is loaded before attacks require that the attacker is able to make traffic redi-
those definitions are loaded. As a result, we could control rection, which can be done in various ways such as DNS
the victim machine using a Meterpreter shell without any tampering, DNS cache poisoning, ARP spoofing, Wi-Fi ac-
problem, even though a.dll is totally detectable by Avira; cess point impersonation and DHCP hijacking. In contrast,
it proves AV-Parmware’s excellent ability to remain stealthy. our scheme is a local attack without such a requirement.
In order to make a.dll to be detected, we tried to ac-
cess a.dll using Windows Explorer. Finally, it was detected 7. CONCLUDING REMARKS
by Avira. However, even in this worst case, Avira detects its
In this paper, we have proposed a new advanced anti-
own processes (avscan.exe, avwebgrd.exe, and avguard.
virus parasitic malware AV-Parmware and have developed
exe) as Trojan horses. Neither sqlite3 nor a.dll is de-
practical implementations. AV-Parmware exploits the newly
tected. Therefore, the user can hardly find AV-Parmware’s
discovered security weaknesses of 12 anti-virus software so-
payload (sqlite3.dll); he even might ignore the detection
lutions, which enables attackers to subvert the anti-virus
after searching for this behaviour, because there have been
and the victim system. Our experiment results show that
false-positive cases that some anti-virus solutions detected
the merits of being a parasite on an anti-virus are signif-
and deleted their own components [4].
icant such as network communications being allowed in a
Another interesting point to note about this experiment is
highly secured environment, gaining more capabilities and
that a.dll successfully connected to the attacker and per-
becoming stealthier. We believe that this type of security
formed its malicious activities such as dumping Windows
weaknesses related to the self-protection feature might also
password hash and capturing screenshots, even after it was
be present in other anti-virus solutions, and this could form
detected. This means the attacker is able to update the
part of our future work.
malware to a new (undetectable) version, or clean-up the
We have developed an effective defence against AV-Parmware
malware from the victim system.
and implemented and evaluated the solution. The proposed
solution utilises digital signature and same origin policy [21],
2132
and is applicable to the current versions of the anti-virus [20] Microsoft. Security Intelligence Report. Technical report.
software without any modification5 . It is important to stress Tech. rep., 2013.
that it is essential that at least one anti-virus solution must [21] Min, B., and Varadharajan, V. Feature-Distributed
be present on every system, even if it is imperfect. Anti- Malware Attack: Risk and Defence. In 19th European
Symposium on Research in Computer Security (Wroclaw,
virus and other security tools raise the bar for attackers to Poland, 2014), Springer International Publishing,
breach the security of a system. Furthermore, there is con- pp. 457–474.
siderable research happening in the anti-malware research [22] Min, B., Varadharajan, V., Tupakula, U. K., and
community aimed at improving detection techniques, espe- Hitchens, M. Antivirus security: naked during updates.
cially behaviour-based ones [31]. Software: Practice and Experience 44, 10 (Oct. 2014),
1201–1222.
[23] Murad, K., Cheng, S.-M., Zikria, Y., and Ikram, N.
8. REFERENCES Evading Virus Detection Using Code Obfuscation. Future
Generation Information Technology (2010), 394–401.
[1] McAfee Communities. https://community.mcafee.com/. [24] Newsome, J., Karp, B., and Song, D. Polygraph:
[2] Avast Forum. http://forum.avast.com/index.php?topic= Automatically generating signatures for polymorphic
106560.msg848307#msg848307, 2012. worms. In IEEE Symposium on Security and Privacy
[3] Early Launch Anti-Malware. (S&P) 2005 (CA, USA, 2005), pp. 226–241.
http://msdn.microsoft.com/en- [25] Oberheide, J., Bailey, M., and Jahanian, F. PolyPack:
us/library/windows/hardware/br259096, 2012. an automated online packing service for optimal antivirus
[4] Sophos antimalware software detects itself as malware, evasion. In Proceedings of the 3rd USENIX Workshop on
deletes critical binaries. offensive technologies (Montreal, Canada, 2009).
http://thenextweb.com/insider/2012/09/20/sophos- [26] O’Kane, P., Sezer, S., and McLaughlin, K. Obfuscation:
antimalware-software-detects-malware-deletes- The Hidden Malware. Security & Privacy, IEEE 9, 5
critical-binaries/, 2012. (2011), 41–47.
[5] Windows Management Instrumentation. [27] OPSWAT. Market Share Report December 2012. Tech.
http://msdn.microsoft.com/en- rep., Dec. 2012.
us/library/windows/desktop/aa394582(v=vs.85).aspx, [28] PCI Security Standards Council. Payment Card
2013. Industry (PCI) Data Security Standard. Tech. rep., Oct.
[6] AhnLab Analysis Team. Malware Analysis: Citadel. 2010.
AhnLab ASEC (AhnLab Security Emergency response [29] Porst, S. How to really obfuscate your PDF malware. In
Center) (Dec. 2012). ReCon (Montreal, Canada, July 2010).
[7] Alvarez, S. Antivirus (In)Security. In CCC (Chaos [30] Rad, B. B., Masrom, M., and Ibrahim, S. Camouflage in
Communication Camp) (Finowfurt, Germany, 2007). Malware: from Encryption to Metamorphism. International
[8] Alvarez, S., and Zoller, T. The Death of AV Defense in Journal of Computer Science and Network Security 12, 8
Depth? - revisiting Anti-Virus Software. In CanSecWest (Aug. 2012), 74–83.
(Vancouver, B.C., Canada, 2008). [31] Sukwong, O., Kim, H., and Hoe, J. Commercial antivirus
[9] Anity Labs. Analysis Report on Flame Worm Samples. software effectiveness: an empirical study. Computer (Mar.
Tech. rep., July 2012. 2011), 63–70.
[10] Bellissimo, A., Burgess, J., and Fu, K. Secure software [32] Symantec Security Response. The Downadup Codex A
updates: disappointments and new challenges. In comprehensive guide to the threat’s mechanics. Tech. rep.,
Proceedings of USENIX Hot Topics in Security (HotSec) 2009.
(Vancouver, B.C., Canada, 2006). [33] Ször, P., and Ferrie, P. Hunting for metamorphic. In
[11] Bilge, L., and Dumitras, T. Before we knew it: an Virus Bulletin Conference (Prague, Czech, 2001).
empirical study of zero-day attacks in the real world. In [34] US-CERT. Malware Threats and Mitigation Strategies.
CCS ’12 (Raleigh, NC, USA, Oct. 2012). Tech. rep., May 2005.
[12] Falliere, N., Murchu, L. O., and Chien, E. [35] Wheeler, A., and Mehta, N. Owning Antivirus. In Black
W32.Stuxnet dossier. Tech. rep., 2011. Hat Europe Conference (Amsterdam, Netherlands, July
[13] IOActive. Reversal and Analysis of Zeus and SpyEye 2005).
Banking Trojans. Tech. rep., 2012. [36] Xu, J., Sung, A. H., Mukkamala, S., and Liu, Q.
[14] Jana, S., and Shmatikov, V. Abusing File Processing in Obfuscated Malicious Executable Scanner. Journal of
Malware Detectors for Fun and Profit. In IEEE Symposium Research and Practice in Information Technology 39, 3
on Security and Privacy (S&P) 2012 (San Francisco, CA, (2007), 181–198.
USA, 2012), pp. 80–94. [37] Xue, F. Attacking Antivirus. In Black Hat Europe
[15] Kaspersky Lab. Gauss: Abnormal Distribution. Tech. Conference (Amsterdam, Netherlands, 2008).
rep., Aug. 2012. [38] Zhou, Y., and Jiang, X. Dissecting android malware:
[16] Kwon, T., and Su, Z. Automatic detection of unsafe Characterization and evolution. In IEEE S&P (San
component loadings. In ISSTA ’10: Proceedings of the 19th Francisco, CA, USA, 2012).
international symposium on Software testing and analysis
(Trento, Italy, July 2010), pp. 107–118.
[17] Lee, J., Jeong, K., and Lee, H. Detecting metamorphic
malwares using code graphs. In SAC ’10: Proceedings of
the 2010 ACM Symposium on Applied Computing (Sierre,
Switzerland, Mar. 2010).
[18] Lin, D., and Stamp, M. Hunting for Undetectable
Metamorphic Viruses. Journal in Computer Virology 7, 3
(Dec. 2010), 201–214.
[19] McAfee. McAfee Labs Threats Report. Tech. rep., 2013.

5
Due to page restrictions, we are not able to describe the
solution in this paper.
2133

You might also like