The Connected Car Introduction The “Internet of Things” promises a future fuelled by the art of the possible. Connected Cars which can talk to each other and access the Internet will provide on-board features that we can’t imagine yet. Predictive systems which bypass traffic jams, reduce carbon emissions and improve safety could become the norm. BT Whitepaper: The Connected Car However, there is an alternate version of reality. A version of reality that threatens the privacy, security and safety of everyone who shares the road with this “Smart Car”. In this reality, remotely connected vehicles are compromised with malware, subject to hijacking, untrusted and unsafe. Therefore, it is vital that the automotive industry, the regulators and thought leaders not only consider the potential of the Connected Car, but also recognise the vulnerabilities and the potential for abuse. 2 Attack Surface and Threat Landscape While the attack vectors of the Connected Car can mirror those of other networked computer systems, there are some unique challenges. Here are a few of the key differences: Connected Cars Typical Networked Computer Computing resources are fixed for lifetime of vehicles – can’t rely on car to protect itself PCs and servers are replaced every 3 years or less, and can easily be upgraded Updating vehicles software, can’t rely on persistent connectivity, some files 1.3Gb can’t chop them up Virus guard updates run in background daily, with no disruption to end users Vehicle data could be compromised and provide untrustworthy information Computers are easier to physically secure Multiple CPUs all accessible via OBD2 port, CPUs can share same keys, some don’t even support PKI Single CPU, with no physical external access VPNs are not designed to protect millions of entities outside a firewall VPN protects small numbers of employees accessing systems outside normal firewall Sub-systems in cars need to be able to freely talk to each other, even for Infotainment Provide physical network separation High overheads to keep onboard signature database up to date, not all malware relevant Existing tools mature, and have little impact on computer processing In fact, the Connected Car threat landscape closely resembles the profile of critical infrastructure systems like SCADA (Supervisory Control and Data Acquisition) networks or those used in manufacturing, the smart grid and energy production & distribution. These all have the following in common: embedded systems from multiple vendors developed in isolation, are assumed to be deployed in “non-hostile” (secure) environments; can’t be patched. Whenever mechanical and digital systems are integrated or automated, new and novel threats must be considered, especially when human life is at stake. BT Whitepaper: The Connected Car Fact or Fantasy As defenders build their threat catalogues, they need to consider three things; the asset being protected, the known attack surface and adversary’s capabilities. Rarely can all of these be known to any degree of certainty at any given time. This forces defenders to be pragmatic in their operations but forward thinking in their architectural approach. The defender is well advised to consider theoretical vulnerabilities as part of its risk management practice, as a properly motivated and resourced adversary will develop a capability to exploit vulnerabilities. This allows for a more effective application of the predicative threat assessment methodology expressed in the diagram below. Defenders must temper the hype associated with theoretical attack scenarios with the more likely general system failure risks associated with infected software. How can defenders determine the right mix given the unique challenges facing the connected car without creating vehicles nobody wants to buy or trust? One approach would be to consider the most relevant theoretical vulnerabilities and current published research available to create a list of attack types. These can provided the necessary context for the various stakeholders to have an intelligent conversation about the specific threat agent. What follows is not an exhaustive list but can provide a starting point for such conversations. 3 TPMS Attacks: Navigation & Infotainment systems TPMS stands for Tyre Pressure Monitoring Sensor. They are plugs that are installed inside of tyres. They communicate on a simple wireless communication protocol that sends a unique ID along with the tyre pressure, rotation and temperature. It is possible to cause unwarranted errors and send signals about the tyre pressure being low (or not being low) regardless of the tyre's actual state. Not only can the unique ID be used to monitor the location of the vehicle it could be used to trigger targeted kinetic attacks such as improvised explosive devices that only detonate when the targeted vehicle is within the kill zone. The motivation for much of the early research into modifying navigation systems was to repurpose the navigation screen to play movies and video games. One proof of concept changed the splash screen, the DVD warning message and modified the binaries to allow for "backup" copies of the mapping software. Navigation systems are the same as any desktop system except they are often directly connected to the CAN bus network. This makes them especially susceptible to attacks. Security of such systems is rarely kept up to date. Car-Jacking It is possible to hijack vehicles over the CAN Bus system. The CAN (Controller Area Network) bus is a vehicle bus standard that uses a simple protocol allowing devices to communicate with each other without a host computer. CAN lines run throughout the entire vehicle and typically connect to around 150 sensors per vehicle. If an attacker can tap into any of these lines they can communicate directly with the car's other connected systems. In its crudest form, simply popping out a tail light to get to a CAN line, unlocking the door then running the signals to start a car, completely bypasses the immobiliser system. It’s not a stretch to believe that an adversary would desire the ability to launch this type of attack remotely. Many vendors provide such capabilities as premium services, which make their management systems a very attractive target. Why take over a single vehicle when you can commandeer all of them? ICSim The Instrument Cluster Simulator was written as a teaching mechanism to understand and reverse engineer CAN bus packets. On the CAN Bus every sensor can see every packet with no indication of the source. This tool teaches students how to RE CAN packets by giving them random Arbitration IDs and bytes for which the CAN bus system operates as well as a game controller interface to control the car. The student can then use the controller to operate the simulated vehicle while sniffing the CAN bus packets to practice finding and sniffing packets. https://github.com/zombieCraig/ICSim BT Whitepaper: The Connected Car 4 “ Related Research Reasonable Response While not exclusively the domain of the Connected Car, smart phone hacking has implications for the security and safety of smart cars. A compromised phone connecting via Bluetooth, USB, Wi-Fi or some other method can become the delivery vehicle for common or custom malware. This can provide the attacker remote access to sensitive systems without the risk of physically attacking the car. By using a compromised phone audio, video and location surveillance can also be affected. GPS spoofing is another area of research to consider. By sending specially crafted GPS signals an attacker could reroute drivers via the “trusted” navigation systems. Adversary’s Asymmetric Advantage Modern automobiles are pervasively computerized, and hence potentially vulnerable to attack…remote exploitation is feasible via a broad range of attack vectors (including mechanics tools, CD players, Bluetooth and cellular radio), and further, that wireless communications channels allow long distance vehicle control, location tracking, in-cabin audio exfiltration and theft....” http://www.autosec.org/pubs/cars-usenixsec2011.pdf BT Whitepaper: The Connected Car The asymmetric nature of the current attack & defend paradigm only adds to the defenders burden. They must be successful all of the time, while attackers only need to succeed once to gain a beachhead or pivot point. By monitoring the current state of public research one can track the shrinking gap between theoretical vulnerabilitties and a practical prototype. In many cases the hacking and open garage community is outpacing (arms race) the multi-billion dollar automotive industry. Since there is no such thing as 100% security, risk reward optimisation should be sought. This must considered as part of the total cost of production and is closely related to safety and quality control through the vehicle life-cycle. In parallel, a flexible capability to quickly detect [mean time to detection] and remediate [mean time to containment] from the inevitable exploitation of the Connected Car must be developed. Safety & Security by Design We know that bad things can and will happen. We need to ensure that we prepare for the most impactful “bad things” and continually improve our capabilities to properly respond to the evolving threat posed by motivated adversaries. Merging the discipline of car safety and cyber security is not a trivial exercise. Its complexity represents a significant threat in its own right. However, it is needed as they are no longer mutually exclusive domains! The engineering department needs to be directly connected to the security department. In many automotive firms engineering and electronics have their own people who look at "security" but the definitions tend to vary and lean more towards physical safety than cyber security. The internal IT department needs to ramp up its hardware capabilities and provide guidance to the engineers designing the components. 5 CAN Bus The CAN (Controller Area Network) bus protocol uses arbitration ID’s. An arbitration ID is a number that more or less represent a category. Each packet has up to 8 bytes. The goal of a CAN packet is to quickly send information to or from the appropriate sensor. The safety of CAN is focused on differential signalling, the idea of having a high and low signal to help eliminate noise on the line. When these protocols were developed the network was assumed to be closed and trusted, outside attacks were not considered a realistic threat to the protocol. Today there are many attack vectors: CAN bus, Navigations systems have wi-fi, XM, GPS, Bluetooth as well as other methods of wireless communications, all attached to the CAN bus. There are some fundamental architectural tenants and engineering practices we can adopt which can be very successful against the most significant class of attacks (control affinity). These simple approaches and practices can save lives in the connected car world and should be carefully considered: The Automobile “Blackbox” This is a read-only logging mechanism that is physically secure, resilient and tamper proof. Ideally the secure data would also be easily accessible by authorised parties, including the owner of the vehicle. This mechanism is great for forensics but vehicle owners need to know what is being recorded because of the potential for privacy abuse. The data collected would also be useful for the community. New tools and capabilities could be developed by the community further extending the value of the data beyond basic vehicle forensics. A standardised logging format should be developed (see standards). Using the OBD (On-board Diagnostics) is not recommended for Blackbox functionality, because of its exposure to abuse by various actors, including law enforcement. There is a high risk of having this system of recording inaccessible to the owners of the vehicle and only accessible to the car manufactures and/or law enforcement. This is not the ideal situation. It is BT Whitepaper: The Connected Car important to start treating vehicles the same way we treat other computing devices. The evidence needs to be recorded so that an attack can be verified however the owner of the vehicle needs a physical method to control how this information is used if we expect the public to trust these systems. If the owner wishes to share this information they can however in the end as it is *their* vehicle. If law enforcement requires access to the data they can go through the appropriate legal channels. This balances the needs of society and protects the privacy of the owner. Layer 1 We have to acknowledge that physical security (layer 1) is now and will always represent a primary attack vector which will be exploitable. This should not come as a surprise to the reader as car theft is an everyday occurrence. Therefore tactics and controls that have a high degree of affinity and effectiveness when executed properly should be considered in any Connected Car play book. From a cyber-standpoint, physically protecting the CAN lines from easy outside exposure should be a priority. We must make it difficult to simply pop out a tail light to splice into the CAN bus to achieve an easy hack. Keep CAN lines away from easy access from outside the vehicle. Full stop! Standards The industry should consider opening up vehicle specific protocol information for owners and researchers. This will crowd source findings and provide new solutions. Immobiliser systems need to be open and auditable. They should use the same standards developed for web based authentication mechanisms. Proprietary encryption did not work for network authentication and it will not work for automotive industry. Longer term, CAN needs to be updated to support a more standard secure protocol. This protocol should have sender info, support encryption and have separate lines to use for larger packets. Small quick packets will always be necessary but a new CAN bus protocol and/or additional CAN buses could provide physical and virtual segregation of data. 6 M2M Secure Machine to Machine (M2M) communications present some unique challenges. In most sensor networks, speed of communications and limited processing power require very efficient communications protocols. Security is often an afterthought. Machines can’t enter passwords but they can use signed certificates. The proper implementation of certificates requires Public Key Infrastructure (PKI) for the life-cycle management of certificates. On the web we have learned that using standards in encryption and authentication methods is ultimately more secure than the outdated notion of security through obscurity. Besides authentication, certificates can also be used to support encrypted communications between devices. Another technique known as “white listing”, can be used to limit which devices are authorised to communicate with each other. This technique doesn’t work well in Internet security applications but can be very effective in closed systems like the Connected Car. Embrace the Community Often new tools can be written by the community faster and with an “outside of the box” perspective. One only needs to look at the success and impact of the Open Source software movement to understand the potential. The simple act of maintaining a Responsible Disclosure web page for security researchers to submit findings creates value. It shows a forward thinking attitude to modern security research and provides a communication path to the community. Public researchers should be considered an asset to be cultivated and embraced. They are not the enemy. The enemy wouldn't tell you about your issues but instead would use them against you. Comprehensive Correlation Engine The Connected Car sensors provide limited security relevance by themselves, as CAN packets do not have state. However, when combined with other contextual data, correlation techniques and modelling predicative capabilities can be developed. Therefore, some form of automated threat modelling should be developed. By modelling complex interactions in ways we haven’t previously can identify conditions that would cause systematic failure affecting security and safety. BT Whitepaper: The Connected Car Open Garages Movement As the community of researchers, small independent garages and general public interest in Connected Cars grows so does the art of the possible. The following community contributions warrant further study to better understand their impact on safer connected cars; • SocketCAN - Linux based CAN drivers/sniffers • CAN in the Middle - HW CAN injection tool • CANiBUS - Multiuser web based CAN monitoring tools • rusEFI - Open source ECU system/replacement • CHT - CAN Hacking Tool • GoodThopter - CAN Bus Sniffing HW • Arduiono CAN Shields - CAN Bus sniffing HW • Kayak – CAN Bus diagnosis and monitoring Call to Action The automotive industry must understand the evolving threat landscape. Bringing new features, services and amenities to market also introduces new risks to security, privacy and safety. This is the nature of progress and intra-domain convergence. The automotive industry should embrace the community and even the odds of the adversary’s unfair advantage. They must understand that cyber security has a direct impact on vehicle safety. Software and embedded security systems need to be tested the same way we test other safety standards. On top of normal crash tests we need to see if a malicious adversary can also intentionally cause a remote crash. The automotive industry needs to share the details of the CAN packets as well as the wiring diagrams. Embracing the maker movement will not only increase sales but also increase safety. 7 About the Authors Craig Smith Craig Smith is the CEO of Theia Labs, core founder of Hive13 Hackerspace and founder of OpenGarages. He has specialised in security and reverse engineering for over 20 years. He has spent the last 4 years working with the automotive manufacturers and research companies on security vehicle infrastructure. He has taught car hacking classes at the US Cyber Camps and is the author of the 2014 Car Hacker's Manual. Bryan K. Fite A committed security practitioner and entrepreneur, Bryan is currently the Global Innovations Product Manager for Assure Intelligence at British Telecom (BT). Having spent over 25 years in mission-critical environments, Bryan is uniquely qualified to advise organisations on what works and what doesn’t. Bryan has worked with organisations in every major vertical throughout the world and has established himself as a trusted advisor. “The challenges facing organisations today require a business reasonable approach to managing risk, trust and limited resources while protecting what matters.” Offices worldwide The services described in this publication are subject to availability and may be modified from time to time. Services and equipment are provided subject to British Telecommunications plc’s respective standard conditions of contract. Nothing in this publication forms any part of any contract. British Telecommunications plc 2014. Registered office: 81 Newgate Street, London EC1A 7AJ Registered in England No: 1800000 27 June 2014