1 / 33

About Ari Takanen

Parallel Execution of Fuzzing Test Suites Study of maximum throughput, resource consumption and bottlenecks for fast-speed fuzzing. About Ari Takanen. The Past: Researcher and Lecturer 1998-2002 University of Oulu OUSPG/PROTOS research group Software Quality related lectures

jarvis
Download Presentation

About Ari Takanen

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Parallel Execution of Fuzzing Test Suites Study of maximum throughput, resource consumption and bottlenecks for fast-speed fuzzing

  2. About Ari Takanen • The Past: Researcher and Lecturer • 1998-2002 • University of Oulu • OUSPG/PROTOS research group • Software Quality related lectures • The Present: Entrepreneur and Evangelist • 2001-today • CTO of Codenomicon • Evangelist: 10 conference talks every year • Author of two books: • VoIP Security • Fuzzing

  3. Agenda • Intro: • Security Testing • Product Security • Fuzzing • Overview • Techniques • Test Coverage • Fuzzing Scalability and Performance • Performance case study • Multi-threaded execution • Test execution optimization • Command-line use • Results • Conclusions

  4. What is Testing - Three Flavors • Feature/Conformance tests validate positive requirements with use cases and test cases • Performance/Load testing repeats one of the feature tests to validate performance limits • Robustness/Fuzz testing uses “misuse cases” to verify software reliability with unexpected inputs

  5. Security Vulnerability = Just A Bug

  6. Codenomicon Labs Test Results Taken from: http://www.codenomicon.com/labs/results

  7. Fuzz Test Effectiveness against WiFi

  8. Some Helpful Definitions • Vulnerability – a weakness in software, a bug • Threat/Attack – exploit/worm/virus against a specific vulnerability • Protocol Modeling – Technique for explaining interface message sequences and message structures • Fuzzing – process and technique for security testing • Anomaly – abnormal or unexpected input • Failure – crash, busy-loop, memory corruption, or other indication of a bug in software

  9. Fuzzing In Short • Fuzzing means crash-testing • Also called: • Negative testing • Robustness testing • Grammar testing • Based on sending systematically broken (rarely random) inputs to a software, in order to crash it • We will ignore random mutator fuzzers for now • Two techniques of smart model-based fuzzers: • Template-based • Specification-based

  10. Model Based Fuzzing Techniques • Template Based Fuzzing (capture-replay, mutation) • Quality of tests is based on the used seed and modeling technique • Very quick to develop, but slow to run • Editing requires deep protocol know-how • Good for testing around known vulnerabilities • Specification Based Fuzzing (generational, model-based) • Full test coverage • Always repeatable • Short test cycle, more optimized tests • Easy to edit and add tests

  11. Coverage • Precision is about attack surface/protocol coverage • All interfaces/protocols tested? • All message sequences tested? • All message structures tested? • All data definitions tested? • All “tags” (values) tested? Accuracy is about anomaly coverage Anomaly categories? SQL? Buffer overflow? All values: 0..65k, a..z, 0x00..0x255 ? Combinations of anomalies?

  12. Anomaly Coverage Selection

  13. Traffic Capture Fuzzing

  14. Traffic Capture Fuzzing Results • Test against samba seems to find zero-day

  15. Key Questions in regards to Fuzzing • Test Coverage: • Q: Which interfaces and protocols are tested? • A: Understanding of threat models and attack surface are the most critical starting point to fuzzing • Q: How well something is tested? • A1: Specification coverage explains how deep knowledge the fuzzer has on a tested interface, and is typically represented by protocol models or protocol templates • A2: Anomaly coverage explains how well each protocol element is tested, and is the hardest metric to measure

  16. Fuzzing Scalability • Software-based Fuzzers scale for all testing needs • The throughput depends only on the available hardware • The entire Network Under Test (NUT) can be virtualized • Software fuzzers can attack every interface, and against all protocol layers • XML • HTTP • TLS • TCP/IP

  17. Why Fast Speed Fuzzing? • “One of the most important aspects of fuzzing is how fast you can execute test cases. The faster you can execute test cases, the more test cases you can run and the more vulnerabilities you will find.” - Dr. Charlie Miller from Independent Security Evaluators. " • Two means of speeding fuzz testing: • More processing power • More virtualization (e.g. in cloud)

  18. Generating Load with Defensics • Full model-based message sequences • Options for monitoring, instrumentation, fuzzing, … • Amount of available CPUs and Logging Level impact performance • Some speed records (test cases per second): • TLS: 2.400 tc/s • IPv6: 4.500 tc/s • HTTP: 16.000 tc/s • DNS: 41.000 tc/s (with capture replay)

  19. Motivation for our Performance Study • Test our test generation efficiency for different type of protocol fuzzers using high-end servers • Analyze benefits (test efficiency, ROI) of software-based fuzzers to appliances

  20. Raw Fuzzing Performance Metrics • Packets or bytes per second does not really work as each fuzz test case can consist of several packets back and forth • Test cases, or full protocol transactions, generated by each suite provides more insight • Performance is generated by having fuzzers running in parallel • These result in “test cases per second” or tc/s

  21. Parallel Fuzzing Benefits • No hardware constraints (fuzzing tools are just software) • No operating system constraints • Parallel processes and threads

  22. Test Setup • We used two different setups: • In the first setup, both test target and the test generator are in the same host. • In the second setup, test generator is in a separate machine from the SUT

  23. Performance Bottlenecks and Metrics • Network bandwidth: • Can be measured with number of full protocol transactions, packets, and data volume • CPU usage by different components of Defensics • Test generator(s) • Test driver(s) • Memory usage of Defensics components • Most interesting metric: • New type of bugs related to performance?

  24. Fuzz Test Can Generate 50x Load

  25. Generating Load with Dell R910

  26. Test Plans Using Command-Line • With some fuzzers, you can run the tools faster on command-line • With Defensics, the GUI is sometimes even faster • E.g. with a laptop with dual-core Intel T7500 processor (with default tool settings): • 10.000 x valid-case in loop • using GUI: 30 seconds • using command-line: 35 seconds

  27. Running Fuzzers in Parallel • Defensics GUI has options to run several suites in parallel • 2 x 5.000 x valid-case in loop • using GUI: 22 seconds (about 50-100% faster)

  28. Using Multi-threaded Mode • When CPU resources are available, multi-treaded mode is often best way to speed up testing • Uses less memory • Note that all parallel execution creates more false positives • Let’s look at that and other speed optimization hints now in live demo...

  29. Performance Value • Users can now… • Increase test coverage through parallel tests • Speed up testing • Find performance bottle-necks • Integrate Defensics also in other functional tests • Remember that Fuzzing ROI is measured in: • TCO vs. Value of Security Findings / Cost of Security Compromise • Reduced maintenance costs, failure reproduction costs • Better customer confidence, higher retention rate of customers • In performance: software-based tools can be significantly cheaper (and faster) than appliances

  30. Conclusions • Testautomationcanbereallyfast, and automated • Model-basedtesting is notnecessarilymanualwork • Scalability of testautomationdepends on the toolarchitecture, butalso on availableprocessingresources • Cloud-basedsolution vs. Hardware-basedsolution • Virtualization • Off-linegenerationversusonlinetestgeneration • Oh and finally: Fuzzingfindslots of bugs, the moreyoutest the morebugsyouwillfind

  31. More News from Codenomicon • Facebook: • Become fan of Codenomicon and Fuzzing • Twitter: • CodenomiconLTD • Codenomicon Website: • Newsletter every second month

  32. Our Book On Fuzzing! • http://www.fuzz-test.com/book/ • Takanen, DeMott and Miller: “Fuzzing for Software Security Testing and Quality Assurance” • Aimed at the general public, you do not need to be a security specialist to read this book • Purpose of the book is to teach next-gen testing approaches to: • Software practitioners • Security engineers • Academics

  33. PROACTIVE SECURITY AND ROBUSTNESS SOLUTIONS THANK YOU – QUESTIONS? “Thrill to the excitement of the chase! Stalk bugs with care, methodology, and reason. Build traps for them. ....Testers!Break that software (as you must) anddrive it to the ultimate- but don’t enjoy the programmer’s pain.” [from Boris Beizer]

More Related