Detecting Targeted Attacks Using Shadow Honeypots K.G. Anagnostakis et al Presented by: Rui Peng
Outline Honeypots & anomaly detection systems Design of shadow honeypots Implementation of a shadow honeypot Performance evaluation Discussion and conclusion
Basic Concepts IPS: Intrusion Prevention Systems IDS: Intrusion Detection Systems Rule-based Limited for known attacks For previously unknown attacks Honeypots Anomaly detection systems (ADS)
A Simple Classification
What is a shadow honeypot? An instance of the protected application Shares all internal state with the normal instance Attacks will be detected Legitimate traffic misclassified as attacks will be validated
Key components Filtering: blocks known attacks Drops certain requests before processing ADS: labels traffic as malicious or benign Malicious traffic directed to shadow honeypot Benign traffic to normal application Shadow honeypot: detects attacks State changes by attacks discarded State changes by misclassified traffic preserved
Implementation Distributed Anomaly Detector Shadow honeypot Network Processor for load balancing An array of anomaly detector sensors Payload sifting and abstract payload execution Shadow honeypot Focuses on memory-violation attacks Code transformation tool takes original source code and generates shadow honeypot code
Creating a shadow honeypot Move all static memory buffers to the heap Dynamically allocate memory using pmalloc() Two additional write-protected pages to bracket the allocated buffer
Code transformation
Performance results Capable of processing all false-positives and detecting attacks. Instrumentation is expensive: 20% - 50% overhead. Still, overhead is within the processing budget.
Benefits Allow AD be tuned towards high sensitivity Less undetected attacks More false positives, but still ok because they will be processed as normal Self-train and fine-tune Attacks detected by shadow honeypot is used to train filtering component Benign traffic validated by shadow honeypot is used to train anomaly detectors
Limitations Creating a shadow honeypot requires source code transformation. Can only detect memory-violation attacks. Apache web server and Mozilla Firefox are the only tested applications. No mention of how filtering component and anomaly detectors can be trained.
Thank you! Questions?