1. Convicting Exploitable Software Vulnerabilities: An Efficient Input Provenance Based Approach Zhiqiang Lin Xiangyu Zhang, Dongyan Xu Purdue University June 27 th, 2008 The 38th Annual IEEE/IFIP International Conference on Dependable Systems and Networks

2. Motivation Internet Worms (CodeRed, Slammer) Denial of Service (DoS) User DoS Viruses, Trojan Horses, Bots (Botnet) FC Vulnerability In Software Accidental Breaches in Security

3. Related Work  Dynamic analysis  Program shepherding (V. Kiriansky et al.) TaintCheck (J. Newsome et al.) Control Flow Integrity (M. Abadi et al.) Data Flow Integrity (M. Castro et al.)…  Run-time overhead, and waiting for attack  Static analysis  BOON (D. Wagner et al.), Splint (D. Larochelle et al.), Archer (Y. Xie et al.), RATS, Flawfinder  False positive  Recent automated multi-path exploration  DART (P. Godefroid et al.), Cute (K. Sen et al.), EXE (C. Cadar et al.), SAGE (P. Godefroid et al.)  Low Efficiency

4. Problem Statement and Our Technique  How to more efficiently discover/convict software vulnerability  An Efficient Input Provenance Based Approach  Conservative static analysis => Suspect  Dynamic analysis => Convicting the suspect and pruning false positives  Randomly mutation is avoided  No symbolic execution (can handle long execution)  Key idea  Data lineage tracing (Input Provenance)

5. Basic Idea fread(&imagehed,sizeof(imagehed),1,in);... width=(imagehed.wide_lo+256*imagehed.wide_hi) height=(imagehed.high_lo+256*imagehed.high_hi);... if((...(byte *)malloc(width*height))...) { fclose(in); return(_PICERR_NOMEM); }... 231 245 246 494 495 496 497 498 Input a.gif (256x128):xx...0x00 0x01 0x80 0x00... Input Data label (Offset): 6 789 An image viewer: Zgv-5.8/readgif.c Integer Overflow

6. Architecture Static- front End Input Lineage Tracer Input Mutator Run-time Detector Program/ binary Lineage Program Input Evidence Suspect New Input A piece of instruction which is exploitable to trigger the vulnerability

7. Component 1. Input Lineage Tracer  Label the input stream (using the offset)  Track their propagation mov 0xfffffffc(%ebp),%eax mov %eax, 0xfffffff8(%ebp) add %eax, %ecx mov %ecx, %edx

8. Component 1. Input Lineage Tracer  Key concept  Data Dependency (direct propagation)  Control dependency (indirect propagation) 1. b=a; 1. if (a==1) 2. b=1; 3. else 4. c=0; mov 0xfffffffc(%ebp),%eax mov %eax,0xfffffff8(%ebp) b=a cmpl $0x1,0xfffffffc(%ebp) jne 804832d movl $0x1,0xfffffff8(%ebp) movl $0x0,0xfffffff4(%ebp) jmp 8048334 a==1 b=1 c=0

9. Component 1. Data Lineage Tracer  DL(S i )=DL(def@s i )  DL(def@s i ) = get_new_id() if def is an input value U DL(use x @s i )otherwise Input data tracking (labeled with its offset in the input stream) DL Representation: reduced ordered Binary Decision Diagram (roBDD)

10. Component 1. Data Lineage Tracer  An Example fread(&imagehed,sizeof(imagehed),1,in);... width=(imagehed.wide_lo+256*imagehed.wide_hi) height=(imagehed.high_lo+256*imagehed.high_hi);... if((...(byte *)malloc(width*height))...) { fclose(in); return(_PICERR_NOMEM); }... 231 245 246 494 495 496 497 498 READ (buf,size,...), 0<= i < size, buf[i], DL(buf[i]@pc231) = get_new_id() DL(wide_lo@pc245)= DL(buf[6]@pc231) = {6}DL(wide_hi@pc245)=DL(buf[7]@pc231) = {7} DL(width@245) = DL(wide_hi@pc245) U DL(wide_lo@pc245) = {6; 7} DL(height@246) = DL(high_hi@pc246) U DL(high_lo@pc246) = {8; 9} DL((width*height)@494) = {6;7;8;9}

11. Component 2. Input Mutator Program Input Data Lineage Evidence Heuristics#1: Buffer overflow mutation (double buffer size …) Heuristics#2: Format string mutation (replace %s in format string argument) Heuristics#3: Integer overflow mutation (Boundary integer value: 0xffffffff,0,0x0fffffff) … Suspect

12. Implementation  Diablo:  Control flow graph  Statically generate Control dependency to facilitate Valgrind instrumentation  http://diablo.elis.ugent.be/  Valgrind:  Lineage tracing  http://valgrind.org/  RoBDD (Reduced ordered Binary Decision Diagram) to represent the data lineage.

13. Evaluation - Effectiveness  Static Detector  Known vulnerability  CVE-2001-1413 (ncompress 4.2.4, SO)  CVE-2001-1228 (gzip 1.2.4, SO)  CVE-2002-1496 (Nullhttpd 0.50, HO)  CVE-2002-1549 (lhttpd 0.1, SO)  CVE-2000-0573 (wu-ftpd-2.6.0, Format String)  CVE-2001-0609 (cfingerd-1.4.3, Format String)  CVE-2005-0226 (ngircd-0.8.2, Format String)  CVE-2004-0904 (xzgv-0.8, IO & HO)  CVE-2006-3082 (GnuPG 1.4.3, IO & HO)  RATS (Unknown)  Make extension to catch: buffer overflow, integer overflow (ipgrab-0.99, epstool-3.3, dcraw-7.94)

14. Evaluation - CVE-2006-3082 (GnuPG 1.4.3)  GnuPG Parse_User_ID Remote Buffer Overflow Vulnerability pktlen=in[2,3,4,5] =0x ff ff ff ff

15. Evaluation - CVE-2001-0609 (Cfingerd-1.4.3) syslog(LOG_NOTICE, "%s", (char *) syslog_str);

16. Evaluation - Ipgrab-0.99 (A New VUL)

17. Evaluation – Performance (Lineage Tracing) Platform: two 2.13 Ghz Pentium processors and 2G RAM running the Linux kernel 2.6.15

18. Evaluation - Performance

19. Evaluation - Space

20. Summary  An input lineage tracing and mutation system:  Capable of convicting known and unknown vulnerability.  Has reasonable overhead for the scenario of offline vulnerability conviction. Static-front End Data Lineage Tracer Input Mutator Run-time Detector Program/ binary Lineage New Input Program Input Evidence Suspect

21. Thank you For more information: {zlin, xyzhang, dxu}@cs.purdue.edu Q & A