EECS Presentation Web Tap: Intelligent Intrusion Detection Kevin Borders
EECS Presentation Overview Target Environment Threat Model Web Tap Design Results Future Work Conclusion Questions Demo
EECS Presentation Target Environment High-security corporate or government network –Very concerned about information leaks and intruders –Mail server and (optionally) proxy server on network perimeter –Strict firewall settings Only allow outgoing http traffic on port 80 from workstations Or use proxy server and block all traffic
EECS Presentation Threat Model A highly-skilled hacker compromises a vulnerable workstation – a link to a web page that exploits the browser – with a trojan in attachment –Hard to prevent due to multitude of browser vulnerabilities
EECS Presentation Threat Model (Part Two) Hacker needs to communicate with the compromised machine –Traditional Trojans do not work (Back Orifice, etc.) Incoming TCP requests blocked –Only two paths available: and Web (http) – is risky Logged Rapid two-way communication from remote shell can be easily detected –Web is a better way of communicating with machine Hard to detect Significantly more bandwidth is available (Without being detected)
EECS Presentation Threat Model (Part Three) Attacker places a custom Trojan Horse program on the machine –Trojan calls back to the hacker’s machine on port 80 (http) at predetermined times –Two-way communication follows in the form of web transactions –If proxy server is used, transactions must appear to be legitimate Later on: Demo of callback Trojan through a proxy
EECS Presentation Web Tap Design Web Tap is a Network-Based Anomaly Detection IDS Why Network-Based? –Host-Based intrusion detection systems are easily disabled Why Anomaly Detection? –Highly-skilled hackers use tools with unknown signatures
EECS Presentation Web Tap Design: Implementation Web Tap implemented as proxy server extension –Records web requests from all users –Extracts important statistics –Builds profile of each user –Raises an alert when it detects non-human web browsing behavior Note: Web Tap also detects spyware and adware in addition to Trojan Horse programs
EECS Presentation Web Tap Design: Statistics Web Tap calculates statistics to characterize human web browsing patterns –Delay between requests for the same site –Size of requests (mean, variance, maximum) –Bandwidth usage (upload) per site per five minutes and per day for each user –Total bandwidth usage (upload) per user per five minutes and per day
EECS Presentation Experimental Setup Statistics were collected from a proxy server with over 30 users (currently have 8 days of data available) –The population group consists of college students, faculty, friends and family members –Home computers with browser configured to use remote proxy server
EECS Presentation Results: Delay Times Aggregate delay times between accesses to a specific site by a specific user follow a distribution Jumps can be seen at certain times (30 seconds, 4 minutes, 5 minutes, etc.) –“Spyware” and other programs use proxy and call back regularly Trojans (and other programs) which call back regularly can be detected by examining distribution of delay times
EECS Presentation
Results: Request Size Outbound HTTP request size alone does not follow a predictable pattern like delay time –Whether a site is being accessed by a program or a person cannot be determined File uploads of over 3-4 KB can be detected –Only ten hosts with a request over 4 KB (four over 10 KB) Useful for detecting data leaks and enforcing “no upload” policy
EECS Presentation
Results: Bandwidth Usage Total upload bandwidth usage for single user shows activity time profile –Traffic during times when user is never active can raise an alarm –Will detect any callbacks that occur when user is usually away Bandwidth usage per site can show regular callbacks Daily upload bandwidth usage per site can detect site receiving a lot of data –An http callback Trojan will need a lot of information per day from the compromised machine
EECS Presentation
Future Work Develop an algorithm to detect entropy in strings –Greatly reduce the number of outbound bytes measured per request English words contain much less information than random bytes –Would help isolate intense, chaotic (encrypted or compressed) bandwidth usage associated with Trojans Apply concepts from Web Tap to other protocols –Thorough intrusion detection –Useful in more open networks
EECS Presentation Conclusion In a high security network, outbound http is the only good way to exfiltrate information Data exfiltration is done by a Trojan computer program using callbacks Web Tap is a Network-Based Anomaly Detection system –Human web browsing follows specific patterns which are hard to mimic –Web Tap takes advantage of patterns to hunt down Trojan and “ad/spyware” programs
EECS Presentation Questions?
EECS Presentation It’s Demo Time!