Detecting and dealing with Advanced Persistent Threats to embedded systems
4 mins read
The vision of 15billion devices being connected to the Internet of Things by 2015 is getting closer to reality. As it does, major concerns are emerging about how secure the infrastructure is that makes it all happen.
High profile targeted attacks through connected devices have highlighted the fact that, whilst industry is broadly aware of the need to protect against common malware attacks, there has been insufficient awareness of the need to deal with direct and targeted attacks on specific pieces of infrastructure – be it terminals or servers – and the challenges involved.
When connected embedded devices use the same operating systems as IT endpoints (ATMs and POS terminals, for example), then hackers can use well tried techniques to attack the embedded infrastructure. This is particularly alarming when you consider that connected embedded devices are being used to control strategic infrastructure – the national grid for example. Welcome to the world of Advanced Persistent Threats (APTs).
The APT detection gap
The corporate IT world has been looking at the issue of APTs for some time, but with little success. While new solutions to tackle APTs are being introduced continuously, the detection gap remains alarmingly long. The main reason is that common security solutions fail to detect the actual APT infection. Instead, they focus on failed prevention attempts (using conventional anti malware technologies) and on monitoring targets that are already infected.
So the question remains: how can such attacks be detected effectively and averted in the embedded world, where timely detection is paramount? Using a new and unconventional method of detection – namely a secure embedded hypervisor – can resolve that problem.
While security vendors are responding more quickly to new methods of infection and evasion, it still takes months to detect APTs – the industry's accepted average is from 6 to 9 months.
The main reason for the APT detection gap is the sophistication of infection techniques used by the attackers. Most infections occur beneath the infected operating system and, as such, cannot be seen in real time by common detection technologies like anti-malware applications and sandboxes.
The principal stages of APT attack
Before anything happens, the attackers will use reconnaissance techniques to identify the target, obtain user contacts and prepare the ground for the attack. However, the APT attack begins only when it reaches the intended target – usually an endpoint.
An APT attack consists of three phases:
* Penetration. Exploiting vulnerabilities in an operating system and/or an application in order to allow the APT to be installed on the endpoint. This is a glaring weakness in the embedded world as embedded terminals typically run an outdated operating system (say Windows XP), which is updated infrequently with security patches, or not at all.
* Infection. Installation of the APT, commonly referred to as 'dropping', with the APT component (mostly with a rootkit module) is known as 'the payload'. This is the critical stage at which the target is compromised and the APT gains enough control over the target machine to carry out its malicious tasks.
* APT activity. Malicious activity on the infected and compromised machine, including communication with the C&C server, gathering personal information, deleting data and so on.
A gaping hole
APTs are currently detected using:
* Common anti-malware products (client applications, gateways, sandboxes and cloud services), which try to detect and prevent penetration.
* Existing anti-APT solutions, which focus on the APT's activity in the infected machine by discovering and monitoring the APT's network activity (mostly outbound traffic). These solutions do not prevent infection, nor are they capable of detecting the infection prior to its network activity.
There is no solution capable of detecting the actual APT infection – the most critical and vital stage of APT attacks – and issue an alert when it happens. Hence the APT detection gap.
Detection challenges
Most APTs use low level and sub operating system rootkits, which are designed to be undetectable by the operating system or any security application installed in or on the OS (and thus perform the 'P' – persistence – element of APT).
In order to be undetectable, yet gain enough control over vital OS functions, a rootkit typically needs two things:
* To install itself in parts of the hard disk hidden from view (the unpartitioned sectors between the disk partitions and the last disk sector) and to access the OS.
* To obtain superior security privilege over the infected OS (subvert the boot sequence of the OS, launch itself before the OS by altering OS original boot sectors – master boot record (MBR), volume boot record (VBR) and the Unified Extensible Firmware Interface (UEFI).
These two rootkit traits are deadly; while the infectors used to penetrate the OS are polymorphing frequently, rootkits change more slowly. Because they are stealthy and undetectable, new versions of these rootkits may only appear once every 12 to 18 months because there is little need to change.
Hypervisor honeypot
A new approach is needed to perform two critical tasks: detecting an APT infection in real time; and providing threat response personnel with live forensics data in order to cut their analysis and response time significantly.
* Detection: Given the evasive nature of APTs, detection must be carried out at a level lower than their level of infection and activity. That level can only be a bare metal hypervisor (such as LynxSecure), separating the hardware from the software, presenting only bare virtual hardware to the installed OS. Effectively a 'virtual motherboard', such a hypervisor will be invisible to APTs and undetectable by them.
The hypervisor must be designed specifically to serve as the means of detection – a honeypot – and hardened rigorously so that it will not become a target for those attacks. This approach also removes any OS dependency (a deficiency of some existing solutions), meaning detection becomes OS agnostic. As the most privileged monitor in the platform, it will be able to detect any changes to the monitored hardware.
A properly designed embedded hypervisor, with a very small inherently secure code base, can be installed on typical PC based embedded systems, as these tend to be very humble in terms of compute power and memory size. The hypervisor's small size will further strengthen the security of the entire system and reduce the attack surface significantly.
In this way, the stealthiest rootkits – such as MBR wipers (Dark Seoul), MBR infectors (TLD4), VBR infectors (XPAJ) and malware using hidden file systems (ZeroAccess) – will be intercepted immediately.
* Live forensics. Currently, finding the exact details of such infections requires arduous and lengthy forensic analysis; the forensics data of the uninfected hardware is not available and the entire hardware needs to be inspected to find the infection. It is like looking for a needle in a haystack.
However, a secure embedded hypervisor allows the generation of immediate fine tuned forensics reports, containing only the infected sections, with an automated analysis of the clean and infected states – a hypervisor always maintains a clean uninfected image.
To contain APT attacks successfully, an out of the box approach is required. Using an embedded secure hypervisor as a proactive detection layer provides that facility.
Avishai Ziv is vice president of cyber security solutions for Lynx Software Technologies (formerly known as LynuxWorks).