CAP 6135: Malware and Software Vulnerability Analysis Find Software Bugs Cliff Zou Spring 2012
Acknowledgement q This lecture is modified based on the lecture notes from: q Dr. Dawn Song: CS 161: computer security 2
Review q Memory-safety vulnerabilities q Buffer overflow Format string Integer overflow q Runtime bounds check q q q Runtime detection of overwrite q q q Purify, Jones & kelly (string bound check) Expensive Stackguard, etc. Practical, but only cover certain types of attacks Runtime mitigation to make attacks hard q q Randomization Practical, but not fool proof 3
This Class: Software Bug Finding q The i. Phone story q Blackbox bug finding q Whitebox bug finding 4
IPhone Security Flaw q q Jul 2007: “researchers at Independent Security Evaluators, said that they could take control of i. Phones through a Wi. Fi connection or by tricking users into going to a Web site that contains malicious code. The hack, the first reported, allowed them to tap the wealth of personal information the phones contain. ” Found by Charles Miller q q Dr. Charlie Miller presented the details of the exploit at Black. Hat in Las Vegas on August 2, 2007. The slides from this talk are also available. Details see: http: //securityevaluators. com/content/case-studies/iphone/ 5
i. Phone attack q i. Phone Safari downloads malicious web page q q q Arbitrary code is run with administrative privileges Can read SMS log, address book, call history, other data Can perform physical actions on the phone q q q system sound and vibrate the phone for a second could dial phone numbers, send text messages, or record audio (as a bugging device) Can transmit any collected data over network to attacker 6
0 days Are a Hacker Obsession q q An 0 day is a vulnerability that’s not publicly known Modern 0 days often combine multiple attack vectors & vulnerabilities into one exploit q q Many of these are used only once on high value targets 0 day statistics q Often open for months, sometimes years 7
How to Find a 0 day? q Step #1: obtain information Hardware, software information q Sometimes the hardest step q q Step #2: bug finding Manual audit q (semi)automated techniques/tools q q Fuzz testing (focus of this lecture) 8
The i. Phone Story q Step #1: Web. Kit opensource q q q “Web. Kit is an open source web browser engine. Web. Kit is also the name of the Mac OS X system framework version of the engine that's used by Safari, Dashboard, Mail, and many other OS X applications. ” http: //webkit. org/ Step #2: identify potential focus points From development site: The Java. Script. Core Tests “If you are making changes to Java. Script. Core, there is an additional test suite you must run before landing changes. This is the Mozilla Java. Script test suite. ” q q So we know what they use for code testing q q Use code coverage to see which portions of code is not well tested Tools gcov, icov, etc. , measure test coverage 9
Results q 59. 3% of 13622 lines in Java. Script. Core were covered q q q Next step: focus on PCRE q q 79. 3% of main engine covered 54. 7% of Perl Compatible Regular Expression (PCRE) covered Wrote a PCRE fuzzer (20 lines of perl) Ran it on standalone PCRE parser (pcredemo from PCRE library) Started getting errors: PCRE compilation failed at offset 6: internal error: code overflow Evil regular expressions crash mobile. Safari 10
The Art of Fuzzing q q q Automatically generate test cases Many slightly anomalous test cases are input into a target interface Application is monitored for errors Inputs are generally either file based (. pdf, . png, . wav, . mpg) Or network based… q http, SNMP, SOAP 11
Trivial Example q Standard HTTP GET request q q GET /index. html HTTP/1. 1 Anomalous requests AAAAAA. . . AAAA /index. html HTTP/1. 1 q GET ///////index. html HTTP/1. 1 q GET %n%n%n. html HTTP/1. 1 q GET /AAAAAAA. html HTTP/1. 1 q GET /index. html HTTTTTTTP/1. 1 q GET /index. html HTTP/1. 1. 1 q 12
Regression vs. Fuzzing q Regression: Run program on many normal inputs, look for badness. q q Goal: Prevent normal users from encountering errors Fuzzing: Run program on many abnormal inputs, look for badness. q Goal: Prevent attackers from encountering exploitable errors 13
Approach I: Black-box Fuzz Testing q q q Given a program, simply feed it random inputs, see whether it crashes Advantage: really easy Disadvantage: inefficient Input often requires structures, random inputs are likely to be malformed q Inputs that would trigger a crash is a very small fraction, probability of getting lucky may be very low q 14
Enhancement I: Mutation-Based Fuzzing q q Take a well-formed input, randomly perturb (flipping bit, etc. ) Little or no knowledge of the structure of the inputs is assumed Anomalies are added to existing valid inputs Anomalies may be completely random or follow some heuristics q q e. g. remove NUL, shift character forward Examples: q q ZZUF, very successful at finding bugs in many real-world programs, http: //sam. zoy. org/zzuf/ Taof, GPF, Proxy. Fuzz, Filep, etc. 15
Example: fuzzing a pdf viewer q q q Google for. pdf (about 1 billion results) Crawl pages to build a corpus Use fuzzing tool (or script to) 1. Grab a file q 2. Mutate that file q 3. Feed it to the program q 4. Record if it crashed (and input that crashed it) q 16
Mutation-based Fuzzing In Short q Strengths Super easy to setup and automate q Little to no protocol knowledge required q q Weaknesses Limited by initial corpus q May fail for protocols with checksums, those which depend on challenge response, etc. q 17
Enhancement II: Generation-Based Fuzzing q Test cases are generated from some description of the format: RFC, documentation, etc. q q Using specified protocols/file format info E. g. , SPIKE by Immunity http: //www. immunitysec. com/resourcesfreesoftw are. shtml Anomalies are added to each possible spot in the inputs Knowledge of protocol should give better results than random fuzzing 18
Generation-Based Fuzzing In Short q Strengths q q q completeness Can deal with complex dependencies e. g. checksums Weaknesses q Have to have spec of protocol q q q Often can find good tools for existing protocols e. g. http, SNMP Writing generator can be labor intensive for complex protocols The spec is not the code q Our goal is code testing, not spec testing 19
Fuzzing Tools q q q Hackers’ job made easy Input generation Input injection Bug detection Workflow automation 20
Input Generation q Existing generational fuzzers for common protocols (ftp, http, SNMP, etc. ) q q Fuzzing Frameworks: You provide a spec, they provide a fuzz set q q SPIKE, Peach, Sulley Dumb Fuzzing automated: you provide the files or packet traces, they provide the fuzz sets q q Mu-4000, Codenomicon, PROTOS, FTPFuzz Filep, Taof, GPF, Proxy. Fuzz, Peach. Shark Many special purpose fuzzers already exist as well q Active. X (Ax. Man), regular expressions, etc. 21
Input Injection q Simplest Run program on fuzzed file q Replay fuzzed packet trace q q Modify existing program/client q q Invoke fuzzer at appropriate point Use fuzzing framework e. g. Peach automates generating COM interface fuzzers q E. g. , Pai. Mei framework q q A reverse engineering framework 22
Bug Detection q See if program crashed q q Type of crash can tell a lot (SEGV vs. assert fail) Run program under dynamic memory error detector (valgrind/purify) q Catch more bugs, but more expensive per run. q See if program locks up Roll your own checker e. g. valgrind skins q Valgrind: http: //valgrind. org/ q q framework for building dynamic analysis tools. There are Valgrind tools that can automatically detect many memory management and threading bugs, and profile your programs in detail. 23
Workflow Automation q q Sulley, Peach, Mu-4000 all provide tools to aid setup, running, recording, etc. Virtual machines can help create reproducable workload 24
How Much Fuzz Is Enough? q q Mutation based fuzzers may generate an infinite number of test cases. . . When has the fuzzer run long enough? Generation based fuzzers may generate a finite number of test cases. What happens when they’re all run and no bugs are found? 25
Code Coverage q q q Some of the answers to these questions lie in code coverage Code coverage is a metric which can be used to determine how much code has been executed. Data can be obtained using a variety of profiling tools. e. g. gcov 26
Types of Code Coverage q Line/block coverage q q Branch coverage q q Measures how many lines of source code have been executed. Measures how many branches in code have been taken (conditional jmps) Path coverage q Measures how many paths have been taken 27
Example q Requires q q q 1 test case for line coverage 2 test cases for branch coverage 4 test cases for path coverage q i. e. (a, b) = {(0, 0), (3, 0), (0, 3), (3, 3)} 28
Code Coverage q Benefits: q q q How good is this initial file? Am I getting stuck somewhere? if(packet[0 x 10] < 7) { //hot path } else { //cold path } How good is fuzzer X vs. fuzzer Y Am I getting benefits from running a different fuzzer? Problems: q Code can be covered without revealing bugs q Maybe the bug logic is not revealed by any testing inputs so far. 29
Fuzzing Rules of Thumb q Protocol specific knowledge very helpful q q More fuzzers is better q q Each implementation will vary, different fuzzers find different bugs The longer you run, the more bugs you may find Best results come from guiding the process q q Generational tends to beat random, better spec’s make better fuzzers Notice where your getting stuck, use profiling! Code coverage can be very useful for guiding the process Can we do better? 30
Approach II: Constraint-based Automatic Test Case Generation q Look inside the box q q Use the code itself to guide the fuzzing Assert security/safety properties Explore different program execution paths to check for security properties Challenge: q q 1. For a given path, need to check whether an input can trigger the bug, i. e. , violate security property 2. Find inputs that will go down different program execution paths 31
Running Example f(unsigned int len){ unsigned int s; char *buf; if (len % 2==0) s = len; else s = len + 2; buf = malloc(s); read(fd, buf, len); … } q q Where’s the bug? What’s the security/safety property? q q s>=len = 232 - 1 What inputs will cause violation of the security property? How likely will random testing find the bug? 32
Symbolic Execution q q q Test input len=6 No assertion failure What about all inputs that takes the same path as len=6? 33
Symbolic Execution q q What about all inputs that takes the same path as len=6? Represent len as symbolic variable 34
Symbolic Execution q q Represent inputs as symbolic variables Perform each operation on symbolic variables symbolically q q q x = y + 5; Registers and memory values dependent on inputs become symoblic expressions Certain conditions for conditional jump become symbolic expressions as well 35
Symbolic Execution q q What about all inputs that takes the same path as len=6? Represent len as symbolic variable 36
Using a Solver q q q Is there a value for len s. t. len % 2 = 0 ^ s = len ^ s < len? Give the symbolic formula to a solver In this case, the solver returns No q q What does this mean? q q q The formula is not satisfiable For any len that follows the same path as len = 6, the execution will be safe Symbolic execution can check many inputs at the same time for the same path What to do next? q Try to explore different path 37
How to Explore Different Paths? q q Previous path constraint: len % 2 = 0 Flip the branch to go down a different path: q q len % 2 != 0 Using a solver for the formula q A satisfying assignment is a new input to go down the path 38
Checking Assertion in the Other Path q Is there a value for len s. t. q q q len % 2 != 0 ^ s = len+2 ^ s < len? Give the symbolic formula to a solver Solver returns satisfying assignment: len = 232 -1 q Found the bug! 39
Summary: Symbolic Execution for Bug Finding q Symbolically execute a path q q Create the formula representing: path constraint ^ assertion failure Give the solver the formula q q Reverse condition for a branch to go down a different path q q Give the solver the new path constraint If returns a satisfying assignment q q q If returns a satisfying assignment, a bug found The path is feasible Found a new input going down a different path Pioneer work q q q EXE, DART, CUTE “DART: directed automated random testing”, Conference on Programming Language Design and Implementation, 2005. “EXE: automatically generating inputs of death”, Conference on Computer and Communications Security, 2006. 40
Скачать презентацию CAP 6135 Malware and Software Vulnerability Analysis Find
e30519534a7d959e6e9cc42962d9fe37.ppt