1 / 26

An Attack Simulator for Systematically Testing Program-Based Security Mechanisms

An Attack Simulator for Systematically Testing Program-Based Security Mechanisms. Ben Breech Lori Pollock. Mike Tegtmeyer. University of Delaware. Army Research Lab. Insecure Programming Practices. Lead to malicious and costly exploits Zotob worm, 2005 ($10,000 - $100,000 / company)

EllenMixel
Télécharger la présentation

An Attack Simulator for Systematically Testing Program-Based Security Mechanisms

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. An Attack Simulator for Systematically Testing Program-Based Security Mechanisms Ben Breech Lori Pollock Mike Tegtmeyer University of Delaware Army Research Lab

  2. Insecure Programming Practices • Lead to malicious and costly exploits • Zotob worm, 2005 ($10,000 - $100,000 / company) • Sasser worm, 2004 (kept ~40 Delta planes on ground, shutdown part of UK Coastguard) • Examples: • gets() • strcpy() • strncpy() (with wrong n)

  3. Example Attack: Stack Smashing Runtime Stack Caller int f (parameters) { local_vars char buf2 [5]; gets (buf2) // vulnerability …. } 0x1200 return address f local vars buf2

  4. Possible Solutions • Fix vulnerability • Can be expensive to locate • Rewrite entire code with more secure language/libraries • Really expensive • Use security mechanism to prevent exploit

  5. Security Mechanisms • Any procedure in program’s environment to prevent successful exploitation of a vulnerability • Program based security mechanisms compiled into program • Problem: Testing of program based security mechanisms tends to be poor

  6. Example Security Mechanism: ProPolice Runtime Stack Caller int f (parameters) { local_vars char buf2 [5]; gets (buf2) // vulnerability …. } 0x1200 return address f canary canary local vars buf2 (Etoh and Yoda, 2000)

  7. Current testing of security mechanisms Implement Mechanism Exploit no Attack Prevented? Execute Vulnerable Prog w/ Sec. Mechanism Compile yes Vuln. Prog.c

  8. What’s Wrong with Security Mechanism Testing? • Limited amount of “exploitable” programs available • Not a lot of testing performed • Some mistakes may go unnoticed • movb instead of movl Our research: create attack simulator to test program based security mechanisms

  9. Goals of Security Mechanism Testing • Systematic -- insert attacks at any appropriate point • Reliable -- accurately report effectiveness of security mechanism tested • Low Overhead -- should be efficient and have reasonable overhead

  10. Testing Framework Test Case Existing Input Program w/ Sec. Mechanism Attack Spec Attack Simulator Coverage Report Monitor Oracle Test Report Framework (Breech and Pollock, SESS 2005)

  11. Attack Simulator Approach • Utilize a dynamic compiler to analyze code during execution • Test security mechanisms by dynamically adding code to simulate attack • No modification of program code • Allows testing of mechanisms without vulnerable programs

  12. Architecture of Attack Simulator Dynamic Compiler Program w/ Sec. Mech. Create Basic Block Attack Simulator Should Attack? Program Input yes no Insert Attack Execute Block on CPU

  13. Recovering During Testing • Most attacks terminate test run • Only do one attack / test run • Can “recover” from some simulated attacks by undoing damage done • Enables more efficient testing of security mechanisms by avoiding startup costs • Requires some extra code to be compiled into program

  14. Testing Stack Smashing Mechanisms • Goal: modify return address • each function call is possible attack point • 3 modes of testing • Direct -- only modify return address • Tailored -- modify enough to trigger security mechanism • Random -- overflow a buffer by writing random values into it

  15. Attack Mode: Direct Runtime Stack Modify return address directly Caller 0x1200 0x5555 return address • Final outcome identical to stack smashing • Can recover • Does not exactly simulate stack smashing attack Current 0xff12 canary local vars

  16. Attack Mode: Tailored Runtime Stack Modify both return address and canary (tailor attack to sec. mechanism) Caller 0x1200 0x5555 return address • Mechanism should identify all such attacks • Can recover • Does not exactly simulate stack smashing attack Current 0xff12 0xf0ba canary local vars

  17. Attack Mode: Random Runtime Stack Write random values into (and beyond) local buffer Caller yz 0x1200 uvwx return address • Exactly simulates buffer overflow • Can’t recover qrst Current canary 0xff12 mnop abcd efgh ijkl local vars

  18. Evaluation Questions • How well does the attack simulator identify attack points and provide systematic testing? • How well do some known security mechanisms identify attacks? • How much overhead is incurred?

  19. Evaluation Methodology • Implemented attack modes as DynamoRIO1 modules • Security Mechanisms • ProPolice2 from GCC 4.1 • RAD3 (testing details in paper) • Executed 4 C programs • None had known vulnerabilities • ~1k - 10k LOC 1 (Bruening et al. CGO 2003) 2(Etoh and Yoda, 2000) 3(Chiueh and Hsu, ICDCS 2001)

  20. Results: Identifying Attack Points Found almost all attack points in direct and tailored modes • Missed main

  21. Results: ProPolice Detection of Simulated Attacks ProPolice detected all attacks it should have • 4 nondetected strcpy wrote to global variables (can’t be used for stack smashing attack)

  22. Results: Tailored Mode Overhead (seconds) Reasonable overhead

  23. Results: Summary • Attack Simulator identified almost all possible attack points (missed main) • ProPolice detected all attacks designed to detect • Reasonable overhead for attack simulator

  24. Conclusions and Future Work • Presented attack simulator to provide more systematic testing of security mechanisms • Demonstrated on known security mechs. • Does not require vulnerable programs • Future work: • Simulate other attacks (function pointers, etc) • Seed faults in security mechanisms to further evaluate attack simulator

  25. Example Vulnerability Caller AR 42 int blah (int a, int b, int c) { char buf1 [3]; int z; char buf2 [5]; gets (buf2) …. } c -> 6 b -> a -> 7 return address -> 0x1200 QRST saved fp -> 0x150 MNOP blah AR JKL buf1 [] -> z -> FGHI Input: ABCDEFGHIJKLMNOPQRSTUVWX ABCDE buf2 [] ->

  26. Example Mechanism: ProPolice Caller AR 42 c -> int blah (int a, int b, int c) { char buf1 [3]; int z; char buf2 [5]; gets (buf2) …. } 6 b -> a -> 7 return address -> UVWX 0x1200 saved fp -> 0x150 QRST MNOP canary blah AR JKL buf1 [] -> z -> FGHI Input: ABCDEFGHIJKLMNOPQRSTUVWX ABCDE buf2 [] ->

More Related