Phantom Boundaries and Cross-Layer Illusions in 802.15.4 Digital Radio

Abstract

The classic design of protocol stacks, where each layer of the stack receives and unwraps the payload of the next layer, implies that each layer has a parser that accepts Protocol Data Units and extracts the intended Service Data Units from them. The PHY layer plays a special role, because it must create frames, i.e., original PDUs, from a stream of bits or symbols. An important property implicitly expected from these parsers is that SDUs are passed to the next layer only if the encapsulating PDUs from all previous layers were received exactly as transmitted by the sender and were syntactically correct. The Packet-in-packet attack (WOOT 2011) showed that this false assumption could be easily violated and exploited on IEEE 802.15.4 and similar PHY layers, however, it did not challenge the assumption that symbols and bytes recognized by the receiver were as transmitted by the sender. This work shows that even that assumption is wrong: in fact, a valid received frame may share no symbols with the sent one! This property is due to a particular choice of low-level chip encoding of 802.15.4, which enables the attacker to co-opt the receiver's error correction. This case study demonstrates that PHY layer logic is as susceptible to the input language manipulation attacks as other layers, or perhaps more so. Consequently, when designing protocol stacks, language-theoretic considerations must be taken into account from the very bottom of the PHY layer, no layer is too low to be considered "mere engineering".