Avoiding Network and Host Detection using Packet Bit-masking

George Stergiopoulos, Eirini Lygerou, Nikolaos Tsalis, Dimitris Tomaras, Dimitris Gritzalis


Current host and network intrusion detection and prevention systems mainly use deep packet inspection, signature analysis and behavior analytics on traffic and relevant software to detect and prevent malicious activity. Solutions are applied on both system and network level. We present an evasion attack to remotely control a shell and/or exfiltrate sensitive data that manages to avoid most popular host and network intrusion techniques. The idea is to use legitimate traffic and victim-generated packets that belong to different contexts and reuse it to communicate malicious content without tampering their payload or other information (except destination IP). We name the technique “bit-masking”. The attack seems able to exfiltrate any amount of data and execution time does not seem to affect detection rates. For proof, we develop the “Leaky-Faucet” software that allows us to (i) remotely control a reverse shell and (ii) transfer data unnoticed. The validation scope for the presented attack includes evading 5 popular NIDS, 8 of the most popular integrated end-point protection solutions and a Data Leakage Prevention system (DLP); both on the network and host session level. We present three different variations of the attack able to transfer (i) shell commands, (ii) large chunks of data, and (iii) malicious code to a remote command and control (CnC) center. During experiments, we also detected an NPcap library bug that allows resent packets to avoid logging from network analysis tools for Windows that use the Npcap library.


Paper Citation