Characteristics of Buffer Overflow Attacks Tunneled in HTTP Traffic

Characteristics of Buffer Overflow Attacks Tunneled in HTTP Traffic

en

The purpose of this article is to describe characteristics of obfuscated network buffer overflow attacks in contrast with characteristics of directly simulated attacks. The obfuscation was performed by tunneling of malicious traffic in HTTP and HTTPS protocols. These protocols wrap a malicious communication between an attacker situated outside of an intranet and a callback located inside of an intranet. The detection analysis which we perform is based on features extraction from network packets dumps and it employs a behavioral and statistical analysis of communications' progress in time and packet index domain. There were performed experiments in four scenarios simulating traffic shaping, traffic policing and transmission on unreliable network channel to make properties of direct attacks and  obfuscated attacks as various as possible. Next part of this article is comparison of obfuscated and direct attacks classification by our previously designed ASNM network features with state-of-the-art features set of A. Moore, both representing statistical and behavioral based experimental academic kernels for NBA. Presented results show better classification accuracy of ASNM features in all kinds of experiments.

The purpose of this article is to describe characteristics of obfuscated network buffer overflow attacks in contrast with characteristics of directly simulated attacks. The obfuscation was performed by tunneling of malicious traffic in HTTP and HTTPS protocols. These protocols wrap a malicious communication between an attacker situated outside of an intranet and a callback located inside of an intranet. The detection analysis which we perform is based on features extraction from network packets dumps and it employs a behavioral and statistical analysis of communications' progress in time and packet index domain. There were performed experiments in four scenarios simulating traffic shaping, traffic policing and transmission on unreliable network channel to make properties of direct attacks and  obfuscated attacks as various as possible. Next part of this article is comparison of obfuscated and direct attacks classification by our previously designed ASNM network features with state-of-the-art features set of A. Moore, both representing statistical and behavioral based experimental academic kernels for NBA. Presented results show better classification accuracy of ASNM features in all kinds of experiments.