simseer and bugwise - web services for binary-level software similarity and defect detection

Click here to load reader

Post on 18-Nov-2014




1 download

Embed Size (px)




  • 1. Simseer and BugwiseWeb Services for Binary-level Software Similarity andDefect Detection SILVIO CESARE AND YANG XIANG DEAKIN UNIVERSITY

2. Introduction Defect detection Finds software bugs E.g., buffer overflows, divide-by-zeros, use-after-frees Malware variant detection Discover obfuscated, evolved, mutated copies of malware Software theft detection Discover illegitimate copies of software Plagiarism detection Discover unauthorized copying of software code. E.g., student assignments. 3. Motivation Defect detection External Auditing Verification of compilation and linkage Malware variant detection Increase predictive power of signatures Most new malware are variants of existing malware Software theft detection Protection of intellectual property Automated detection reduces costs of investigation Plagiarism detection Provide a deterrent through automated detection Manual approach not scalable 4. Innovation This research makes the following contributions: We propose an online web service, Bugwise, to performbinary-level defect detection. We propose an online web service, Simseer, to addressmalware variant detection, software theft detection andplagiarism detection. We use state-of-the-art algorithms in novel applications. We implement and make our services public 5. Related Work Defect detection Formal methods, program analysis, abstract interpretation,data flow analysis. Software similarity Features make a birthmark (fingerprint) Similarity function comparing birthmarks (euclidean distance,cosine similarity etc). Birthmarks Vectors, strings, sets, trees, graphs etc. Byte-level content, instructions, basic blocks, control flow, APIcalls etc. Our system uses control flow. 6. Our Approach Bugwise and Simseer use a unified backend from our previous work Malwise. We implement two web services using cloud-based virtual private servers. Simseer Uses control flow as a feature to generate a signature (birthmark). Bugwise Combines decompilation with traditional data flow analysis to detect several bug classes. 7. Web Services WorkflowWeb Frontend Scan Server Script SSH Tunnel Scheduler ScriptEvolutionarySSH Tunnel (Simseer)MalwiseTree Creation Store andDisplay SSH Tunnel (Bugwise)Results 8. The Web Frontend Accepts submission of archives and executables. Implemented with server side PHP programming language. PHP launches script to process submitted binary. Script performs validation. E.g., Filenames have no special characters. Launches C++ network client to submit binary to scan server. 9. The Web Frontend 10. The Scheduling Work Queue Listens to TCP port on scan server. Connects to web frontend via SSH tunnel. Accepts binaries from web frontend. Queues jobs so that only 1 is running at any time. Launches Simseer or Bugwise script to process binary. 11. Malwise Backend Malwise is a native C++ application of ~100,000 LOC. Plugin-based modular system. Simseer and Bugwise differ by their configuration and plugins. Configuation specified in XML. 12. The Simseer Backend Performs unpacking to remove malware obfsucation. Decompiles the control flow. 1st pass generates signatures. 2nd pass shows similarity between signatures. 13. The Bugwise Backend Performs decompilation of local variables. Performs compiler-style optimisations (dead code elimiation, copy propagation, constant folding etc). Performs data flow analysis (reaching defintions, upwards exposed uses etc). Detects double frees (deallocating the same memory twice) using the data flow analysis results. 14. Configuration - Simseer (l), Bugwise (r) ScanScanCode Optimsation 1Packer Detection Using EntropyLinux ArchUnpacker Using Application Level EmulationPre Decompiler Data Flow AnalysisStructuringX86 Decompiler Data Flow AnalysisNGram StructuringDecompiler Data Flow AnalysisCode Optimsation 2IRDataFlowAnalysisDouble Free Detection 15. Simseer Evolutionary Tree Visualization Phylogenetic tree e.g. tree of life. The closer nodes are in the tree, the more similar thosenodes are. Simseer backend generates distance/similarity matrix. PHYLIP software package takes matrix and generatestree. Tree is rendered to an image. 16. Program Realtionships Visualization 17. Results Processing Parse XML output from Malwise PHP parser Simseer Display evolutionary tree and similarity matrix Bugwise Display table showing address of double frees 18. Efficiency of Malwise as a Web Services Does a web service incur much overhead compared to command line usage? Test case is 9 samples submitted to Simseer. Python script sends samples and waits for results. We compare the times of command line versus the web service. Mean overhead is 0.64 seconds. 19. Processing timesSimseer Web Service (l), Malwise Command Line (r) 20. Availability http://www.FooCodeChu.Com Rate limiting of submissions. Limit of sample sizes and the number of samples in archives. We intend to relax these restrictions as we migrate to more scalable infrastructure. 21. Future Work Enterprise messaging to perform load balancing and queuing? More options to scans to exploit Malwise plugin system. Any-time clustering to cluster new samples incrementally in real-time? Bug detection could be developed as bug management system. 22. Conclusion We make available new services for bug detection andsoftware similarity. Our backend Malwise is versatile and allows plugins toimplement these services. Bugwise has found real bugs in Linux. The web service overhead is minimal. We believe web services in these applications will have futuregrowth.