Detecting malicious routers
The Internet is not a safe place. Unsecured hosts can expect to be compromised within minutes of connecting to the Internet and even well-protected hosts may be crippled with denial-of-service attacks. However, while such threats to host systems are widely understood, it is less well appreciated that the network infrastructure itself is subject to constant attack as well. Indeed, through combinations of social engineering and exploitation of weak passwords, attackers have seized control over of thousands of Internet routers. Once a router has been compromised in such a fashion, an attacker may interpose on the traffic stream and manipulate it maliciously to attack others—selectively dropping, modifying, or re-routing packets.
First, we specify this problem of detecting routers with incorrect packet forwarding behavior and we explore the design space of protocols that implement such a detector. We further present two concrete protocols that differ in accuracy, completeness, and overhead—one of which is likely inexpensive enough for practical implementation at scale. We present a prototype system that implements this approach on a PC router and describe our experiences with it. We believe our work is an important step in being able to tolerate attacks on key network infrastructure components.
Unfortunately, it is quite challenging to attribute a missing packet to a malicious action because normal network congestion can produce the same effect. Modern networks routinely drop packets when the load temporarily exceeds a router's buffering capacity. Previous detection protocols have tried to address this problem using a user-defined threshold. Recently, we have designed, developed and implemented a new compromised router detection protocol that dynamically infers, based on measured traffic rates and buffer sizes, the number of congestive packet losses that will occur. Once the ambiguity from congestion is removed, subsequent packet losses can be attributed to malicious actions. We have tested this protocol in Emulab and have studied its effectiveness in differentiating attacks from legitimate network behavior. We believe this protocol is the first to automatically predict congestion in a systematic manner and is necessary for making any such network fault detection practical.