An Empirical Study of TCP Vulnerabilities in Critical Power System Devices

Abstract

Implementations of the TCP/IP protocol suite have been patched for decades to reduce the threat of TCP sequence number prediction attacks. TCP, in particular, has been adopted to many devices in the power grid as a transport layer for their applications since it provides reliability. Even though this threat has been well-known for almost three decades, this does not hold true in power grid networks; weak TCP sequence number generation can still be found in many devices used throughout the power grid. Although our analysis only covers one substation, we believe that this is without loss of generality given: 1) the pervasiveness of the flaws throughout the substation devices; and 2) the prominence of the vendors. In this paper, we show how much TCP initial sequence numbers (ISNs) are still predictable and how time is strongly correlated with TCP ISN generation. We collected power grid network traffic from a live substation for six months, and we measured TCP ISN differences and their time differences between TCP connection establishments. In the live substation, we found three unique vendors (135 devices, 68%) from a total of eight vendors (196 devices) running TCP that show strongly predictable patterns of TCP ISN generation.