1. Secure Real Time Embedded Systems Sherif Khattab and Daniel Mossé University of Pittsburgh Computer Science Department

2. Embedded Systems Before: isolated, closed systems Later: connected thru dedicated phone lines Now, web connected; control can be done remotely Convenience costs LOTS of remote security issues  Safety and security are big issues, since these systems (now on the web) control actual industrial plants and other devices Attacker’s goal: compromise data and deadlines Defender’s goal: satisfy deadlines, despite overhead

3. POTS? Voice over IP? Assume VoIP is widespread (skype anyone?) Assume compromised nodes can attack POTS  Use VoIP to attack dialup control systems  Distributed Denial of Service: lots of VoIP clients compromised attack control system to a slow or fast death…

4. Denial of Service DoS attacks cause system overload, overloads cause timing failures (missed deadlines, control period) System needs to react when it cannot  Suggested approach: reserve security bandwidth? RTSs are a perfect candidate  Every new component creates a new vulnerability  Make detection a real-time task (temporally secure) characteristics? temporally vulnerable? Mitigating DoS attacks in RTSs or EmSys  Mixture of static and dynamic analysis?  Relation with imprecise, reward-based, version-based, elastic, …, computing? Power grids, sensor networks, industrial control systems…

5. DoS (dist system) More difficult problem:  Need to meet end to end deadlines  Ensure that all messages arrive safely  Network partitions are possible (common?)  Distributed and quick detection may be needed  Coordinated attacks are the norm  Each compromised node is undetected Cooperation among hosts, routers and other network entities is essential/crucial  Backward compatibility a must for early deployment

6. DoS (wireless system) Single attacker can influence many victims Physical proximity can also be compromised  Need more defenses.  Need localization services?

7. Requirements Need another property, namely security level  Do we need YARTM? (yet another RT task model?)  Include a measure of robustness and power/energy  Complete model includes attackers’ capabilities and constraints (battery, CPU, etc), attack model (correlated attacks, spoofing attacks, etc) However, security is on the eye of the system integrator  Need to provide tradeoffs  Specification is needed  Need to remember that data exists forever

8. Questions Define the difference between security and fault tolerance? Similar in RTSs? In EmSys? Find tradeoff of crypto/security deadline misses Need efficient intrusion detection mechanisms What is special (besides funding ) in secure embedded systems?? Similar, but for small devs  Cannot afford the power for public key crypto  Need adaptive security; does it compromise security?  Relatively light attacks may be crippling What detection mechanisms can we use that satisfy all restrictions of embedded systems?