Power analysis may help secure devices on the Internet of Things

Side channel power, the new security front

elctrical plug shock

Side channels used to be avenues for cyber attacks. Today, one side channel has been elevated to a new front line for cyber defense, and it may go on to be a bulwark for the internet of things (IoT).

"The limited testing we have done has been very successful," says Joe Cordaro, engineer at the Savannah River National Laboratory in Aiken, S.C.

Cordaro is using a system that monitors a system's power side channel -- in other words, measures and analyzes the power consumed by any given device for the sake of ensuring security. His in-lab tests have looked for abnormalities in the functioning of programmable logic controllers.

Side channels are forms of information that a system generates as a side effect of its operations, such as radio emissions, sounds, variations in the time needed to perform certain operations and even error messages. In the past, this information was often used in attempts to break a system's security -- for instance, by helping to reconstruct encryption keys or passwords.

"Power attacks were originally developed for smartcards," explains Gary McGraw, CTO at security consultancy Cigital. "They had small processors and carried out limited computations. An attacker could tell what computations were being done from the power consumption, and determine what cryptographic keys the card was using.

"The notion of attack has been turned on its head and now you monitor power consumption to see if the system is behaving in an anomalous fashion, which is cool from a security perspective," says McGraw.

"The traditional method of cyber security is scanning and patching," agrees Cordaro. "But that patch was written because the vendor found some vulnerability, because of an attack that may have taken place months before the patch was available. In the meantime, scanning doesn't help you because you don't know what to look for. Power analysis is independent of that."

Two firms are currently known to be working on anti-malware systems based on power analysis: PFP Cybersecurity in Vienna, Va.; and Virta Labs in Ann Arbor, Mich.

Two commercial approaches

Both firms are targeting systems rather than smartcards, with Virta Labs specializing in medical devices and PFP Cybersecurity concentrating on programmable logic controllers (PLCs) and similar industrial devices -- similar to what Cordaro is using. (PFP, incidentally, stands for Power Fingerprinting.)

"We see ourselves as a software firm, as the analysis depends on our algorithms," says Thurston Brooks, vice president at PFP Cybersecurity. The firm uses off-the-shelf digitizers or oscilloscopes to gather the power consumption data, and probes can be connected to the monitored device in multiple ways. The firm hopes to eventually have inexpensive power monitors that can be installed at the chip level, Brooks adds.

Using PLCs, their system has proven able to detect the Stuxnet virus -- famous for compromising Iranian nuclear research -- within minutes, Brooks says.

"Attackers like to hide malware in places, like the root, where you can't see it," says Brooks. "But we see it when it executes. Normal activity is to do a task and then loop. Malware will execute snippets between tasks to see if it is time to load [the main body of the malware program], and we will see that."

Malware aside, Brooks says that the firm's test customers have been able to use power fingerprinting to confirm the configuration of products coming off the assembly line, or see if replacement parts that have been in storage for years have been altered -- presumably to cannibalize components.

Virta Labs, meanwhile, is working on an analyzer (still in test mode) that plugs in-line with the power cord of the monitored device.

"Our approach is to look at the power consumption signals from the plug to the device; we refer to our position as outside the box," says Ben Ransford, CTO at Virta Labs. "Different computing activities induce different patterns [of power consumption] and cause the components inside to do well-choreographed operations. We look for activity patterns that indicate what is going on. We can tell how hard a CPU is working, and that is often enough to figure out what programs are active. We are especially good at spotting periodic activity, which is good in the fight against malware since a lot of malware exhibits periodic behavior."

Overall, "Our goal is to offer visibility into devices that don't typically receive patches and don't have anti-virus software or other anti-malware measures," he says. The lack of security patches and anti-virus software is typical for medical devices, he adds, because it's a common policy of medical device vendors that once a device has been approved by the FDA it must remain unaltered. (The FDA itself actually allows patches and the addition of anti-malware for such devices, Ransford adds.)

"But unprotected devices can offer attackers a foothold on a network," he notes. "Most new devices are networked, and it's not uncommon to see medical devices running off-the-shelf operating systems. That introduces a potential vector for attack. Malware is rampant in the healthcare field, and visibility into the networks is a key problem," Ransford says.

IoT and desktops

IoT devices of the future will likely face situations similar to that of today's medical devices, with limited updates and little or no malware protection, making them prime candidates for protection through power analysis, Ransford notes.

"They will be building the devices on Linux, and if they go unpatched we will have a problem," he says. "Look at the wireless routers in your house" -- they probably run a version of Linux that is "many versions" behind the most recent one, "and there is no visibility into them, either."

Brooks agrees that IoT devices are prime use cases for power analysis because they typically incorporate small processors with limited functionality, with little ability to support anti-malware software. "We don't load the processor or steal resources, and a hacker cannot see us," he says of power analysis. "But we can see if something has been changed, or if the device is not acting right."

The target devices of the two existing firms are PLCs with limited functionality, or medical devices with a restricted software repertoire. Using power analysis with desktops is not promising, sources agree.

"In the old days we could only do [power analysis] with eight-bit processors. Now we can do it for more complicated devices that do one thing," says McGraw. "I am sure there are people researching power fingerprinting for multi-tasking machines, but we are not ready for that yet; I don't know that it will ever spread to the office," he says.

"For a web server that's just used to serve pages, it might work," adds McGraw. "But for a home computer that's used for everything from games to internet access to word processing, it would be hard [for the power monitoring system] to say what's normal activity. But that's the challenge facing any intrusion detection system based on anomaly recognition."

"We can detect things at a lower level, but at a higher level we are not able to recognize abnormal behavior," Brooks agrees. "We assume processes are legitimate, and look at the intervals between the processes. Then when you add a human operator there is an infinity of things they could do, so how do you know what is normal?"

"We haven't done any testing with Windows and, since it has randomness, using power fingerprinting would be very challenging," says Cordaro. "My vision is to see it permanently attached to multiple devices in a substation, so you can get broader picture of how the substation's devices are working. It adds another layer of defense, one that's independent from other layers of cyber-defense," he adds.

Wide-open side channels

"The lesson from power fingerprinting is that, when designing secure systems, you have to keep things like side channels in mind," says McGraw. "You can't just focus on silly bugs like SQL injection. You have to look at all angles of attack and defense."

His words take on more urgency when examining the many ways researchers have found to exploit side channel attacks. For instance, if you can reconstruct the user's keystrokes from the noise of the keys you could capture logins. Researchers have demonstrated ways to reconstruct the sounds in a room by pointing a high-speed video camera at a potted plant in that room and analyzing the sound-generated shivers going through its leaves. The trick also works with the surface of snack bags, filmed through a glass door.

Since an Ethernet cable has a solid connection with a unit, power consumption and radio emissions data from a laptop can be extracted by tapping the cable anywhere along its length. Radio emissions from a laptop otherwise have a range of only a few feet, but can be picked up within that range with simple equipment disguised as innocent items.

Display screens are especially noisy in the radio spectrum and their contents can be captured at a distance through electronic eavesdropping using relatively simple equipment.

"Consumer items are not TEMPEST-shielded," warns Ransford. (TEMPEST is the code name for the NATO anti-eavesdropping shielding standard.)

Chinese researchers have also noted that the sound a processor makes can be correlated to the functions it is performing.

Other issues worth considering:

  • The diffuse flickering reflections of a display off the walls of a room often contain enough information to reconstruct what's on the screen, making TEMPEST insufficient in some settings.
  • In some cases, analyzing the time it takes a server to perform encryption can be used to extract a key, partly or wholly, but only after several million iterations. Access to the memory cache, to see where in the process the CPU consults a look-up table, gives even better results.
  • Manually typed passwords, when eavesdropped on the way to a server, can also be at least partly reconstructed through timing analysis, since some keys are farther apart than others.
  • Fault analysis, which examines the operations of a cryptographic system during a transient hardware problem, has been used to break many cryptographic algorithms. But it requires circuit-level access to the hardware.
  • Analyzing error messages generated by fake cipher-text submitted to a secured system can sometimes expose the cipher being used, after perhaps a million iterations.

Consequently, "power fingerprinting is not going to win the war," says Zach Lanier, research scientist at software provider Optiv Security. "It just ups the ante and puts one more thing in the security practitioners' tool bag. Security is a never-ending cat-and-mouse game."

"It's just another tool in the war, it's not going to solve all security problems" agrees Jon Oberheide, CTO at software vendor Duo Security. "It's basing its information on the power signature, but more information would be available from software running on the device. But when you can't do that, it offers value and visibility you didn't have before."

This story, "Power analysis may help secure devices on the Internet of Things " was originally published by Computerworld.

Copyright © 2016 IDG Communications, Inc.

7 hot cybersecurity trends (and 2 going cold)