Why firewalls don’t cut it when protecting critical infrastructure

Andrew Ginter is vice president of industrial security at Waterfall Security Solutions, a Tel Aviv-based company that since about 2008 has been installing its SCADA (Supervisory Control and Data Acquisition) Monitoring Enablers and Unidirectional Gateways in critical infrastructure systems such as Canada's New Brunswick Power.

Ginter has guided development of commercial products for computer networking, industrial control systems, control system-to-enterprise middleware, and industrial cyber security. He represents the company to international standards bodies and speaks often on industrial control system cyber security.

Ginter recently talked with Defense Systems contributor David Walsh about the multiplying dangers to critical infrastructure. Given Stuxnet, Dragonfly and the cyberattack that took out power in Ukraine, he asserts, firewalls fall well short.

Defense Systems: So far references to “TPAs” have been few. How and why has it come to the fore? 

Andrew Ginter: Many still use the older term “advanced persistent threats (APT).” A number of organizations are trying to migrate to the “targeted persistent attack (TPA)” term because nobody believes they will ever be the target of an “advanced” attack. In fact, all of yesterday’s advanced attacks are now automated in widely available attack tools. APT/TPA is the modern attack methodology. Whenever you hear of a big new attack, odds are it is one of these.

DS: New variant attacks, then, are so worrisome they need a separate threat category?

Ginter: These attacks defeat “classic” [software-based] IT security mechanisms. They steal credentials. They use permissions rather than vulnerabilities. They are invisible to intrusion detection systems.

DS: Of the categories your “gateway” and other product lines are aimed at mitigating or defeating, which is likely the toughest to come to grips with and why?

Ginter: Security is a continuum, not a binary value. We are never “secure,” only “more secure” or “less secure.” Given time, money and talent, any “secure” site can be breached. This is why today’s targeted attacks are so difficult. [Previously], if we were better secured than our neighbors, the bad guys would give up and happily break into neighbors’ systems. Today, nemeses want us, target us, and are not distracted until they achieve their goal. 

DS: How does a combination toolset defeat, say, software bug exploits and code errors?

Ginter: Unidirectional security gateways are an alternative to the firewalls being deployed widely in critical infrastructure control system networks. Our gateways [combine] hardware and software. The hardware consists of two modules – the transmit module has a fiber-optic transmitter in it, but no receiver. This is not a standard fiber-optic chipset. The receive module has a fiber-optic receiver in it, but no transmitter.

The two modules are separated by about 18 inches of fiber-optic cable. Together, they enable us to send information out of a control system network, without permitting any information, any message, or any signal at all to reenter that network. No remote control attack messages, no nothing. We can monitor nuclear reactors from the Internet if we want, without introducing any risk of a message or attack leaking back into the critical network.

DS: What recent infrastructure cybersabotage or cyberattack events are most troubling?

Ginter: The power outages in the Ukraine. Attackers ... broke into two Ukrainian power distribution systems, turned a lot of substations off by remote control, and erased enough hard disks to make remote re-establishment of power impossible. A quarter million people were affected. The utility had to drive out to the dozens of affected substations, manually re-enabling power flows and disconnecting each substation from all networks to prevent a reoccurrence.

Forensic examination shows that this was a remote control attack – just like the attacks we see in the news stealing credit card numbers and such. Only this time, the consequences were physical: the lights went out in two different cities. 

DS: And how do gateways quash such episodes?

Ginter: Unidirectional security gateways improve security in a number of ways, but most importantly, it is precisely these modern, targeted attacks that are defeated by the gateways. If there is no way to send a command into a control system network, there is no way to acquire enough intelligence to sabotage the network, or to remotely misoperate the physical process.

DS: What are the advantages of “unidirectional” vs. “bi-directional” security protocols? Fewer vulnerability pathways?  And, the shortcomings of firewalls?

Ginter: All firewalls are bi-directional. Every path through a firewall that allows us to monitor an industrial network also lets [hackers] attack that network. Unidirectional security gateways are unidirectional at the hardware level. It is physically impossible to send any command back into a protected network through a unidirectional gateway.

When the only connection to a control system network is via unidirectional gateways, it takes a physical act at the site to compromise the industrial network: for example, carrying an infected laptop or USB key through security. Compromising those networks without either deliberate or accidental cooperation from someone at the site is impossible. Sabotaging the site while sipping coffee in a basement on the other side of the planet is physically impossible. No firewall can make that claim.

DS: What of separating network elements, so if constituent parts go down damage is dispersed, allowing other pieces to carry on or be patched?

Ginter: People are using unidirectional gateways to segment networks or a large power plant, for example. If the control system for every generating unit is truly independent, and unidirectionally protected, then even if one of the generating units is compromised by malware, however sophisticated, there is no way that compromise can propagate either automatically or by remote control into another generating unit. Damage and downtime is confined to one generating unit of typically 3-10 units at the plant.

DS: Are the gateways more geared to the military, federal agencies or private sector? Can the core product be tweaked per application?

Ginter: Unidirectional security gateways are deployed in industrial control system networks in a wide variety of industries. Waterfall provides software to move data out of a host of industrial control system vendors’ systems. Governments purchase unidirectional gateways, because many electric, water, natural gas and other utilities are government-owned. Military installations purchase them too, when they have industrial facilities of their own to operate.

Many military bases, for example, have both connections to their local power grid, as well as internal “micro-grids” complete with power generation and distribution infrastructures to fall back on, if the connection to the external grid fails.

DS: Have any systems been acquired/employed in the United States or deployed overseas?

Ginter: Currently, our biggest installed base is the North American electric sector, but these systems are deployed in every sector, worldwide.

DS: Can you offer a bit more on how gateways differ from firewalls. 

Ginter: Firewalls are routers – they forward messages. Some messages are attacks. Firewalls are routers with filters – the filters look at each message and try to figure out if the message is “good” or “bad.” Bad messages are dropped and the rest are allowed through. No filter is perfect. Inevitably, some residue of attack messages leaks through every firewall. For example, most of us have a firewall in our homes, yet phishing emails and other attacks ... show up on our computers.

Unidirectional security gateways allow us to monitor big, complex and dangerous industrial systems, without ever allowing any message back into the monitored networks.

DS: Can they mesh with existing software-based tools or must software-based cyber security solutions be dumped and replaced with gateways?

Ginter: Unidirectional security gateways cooperate with countless software security tools. The lesson here is not that software tools have no value; it is that relying on software exclusively to protect our most important control system networks is dangerous, because all software can be hacked. All software has bugs ... and can be hacked.

Unidirectional security gateways cannot be hacked by remote control (as happened in Ukraine). Hacking our gateway takes someone standing over it with a screwdriver in hand, or a hammer or a bucket of water. Software security has a place, but industrial security should begin with physical/hardware security.

DS: Cybertattacks and vulnerability probes emanate from a broadening field of actors: state, non-state, “lone wolves,” criminals, anarchists. Which actors are most concerning? 

Ginter: The most concerning are any entities carrying out targeted attacks. Hacktivist activity is generally targeted. State actors, or their contractors, similarly target infrastructure, and have more money to spend on sophisticated attack tools than hacktivists do. Organized crime is even getting into targeted attacks. Two hospitals were recently targeted with ransomware, and a ransom was extracted from the hospital as a whole to unlock the entire network, rather than from individual machines, as was the case in the past. As organized crime figures out how to reliably monetize cyber-sabotage attacks, we will see such attacks increasing in frequency dramatically.

DS: Have you identified any coalescing or alignment of these disparate attack elements?

Ginter: Nothing major. There were recent reports that Russian, and then Chinese, authorities arrested people involved in hacking overseas targets. This is curious, given that Russia and China are widely seen as state sponsors of cyber-attacks. These are anomalies, though – I don’t see them as a trend.