1. Software Tamper Resistance: Obstructing Static Analysis of Programs Chenxi Wang, Jonathan Hill, John Knight, Jack Davidson at university of Virginia This presentation consists of 4 parts:  Introduction  Techniques employed to hinder static analysis  Theoretical foundation and experimental results  Conclusion

2. Introduction Problem addressed: protecting trusted software against tampering on untrustworthy hosts. The solution: tamper resistant software. Why is access control infeasible? One aspect of software tamper resistance: prevention of static analysis of programs.

3. Static Analysis of Programs “Static analysis refers to techniques designed to extract [semantic] information from a static image of a computer program.” Entails two steps: control-flow analysis and data-flow analysis. Can be defeated by making the program control-flow data-dependent.

4. Control-flow Transformations Modify high-level control transfers to obstruct static detection of branch targets and call destinations. Two steps of transformation are illustrated in the following figures.

5. Dismantling of High-level Constructs int a,b; a=1; b=2; while(a 10) b--; a++; } use(b); a=1 b=2 b=a+b if(!(b>10)) goto L2 L2: a++ goto L1 L4: use (b) b-- L1: if(!(a<10)) goto L4

6. Transform to indirect control transfers L1: a=1; b=2; swVar=2; L3: b=b+a; if(!(b>10)) swVar=5; else swVar=4; L5: a++; swVar=2; goto switch switch(swVar) swVar=1 L2: if(!(a<10)) swVar=6; else swVar=3; L4: b--; swVar=5; L6: use(b);

7. Data-flow Transformations Dynamic computation of branch targets. The introduction of non-trivial aliases into the program.

8. dynamic computation of the switch variable L3: b=b+a; if(!(b>10)) swVar=5; else swVar=4; switch(swVar) goto switch; L3: b=b+a; if(!(b>10)) swVar=global_array[f1()]; else swVar=global_array[f2()]; switch(swVar) goto switch int global_array[];

9. Introducing aliases through pointers *p = a; a = a +b; *p = b b = 3; L2: *p = b; b = 3; swVar = f2(); L1: *p = a; a = a+b; swVar = f1(); A A

10. Complexity Evaluation Theorem 1: in the presence of general pointers, the problem of determining precise indirect branch target addresses is NP-hard. Proof: constructing a polynomial time reduction from 3SAT problem to it. Cannot be solved in polynomial time under the assumption that P!=NP.

11. Experimental Results Transformations considerably hindered the optimization that the compiler(gcc) is able to perform. Defeated PAF which is a static analysis tool from Rutgers university. Replacing 50% of the branches will result in an increase of a factor of 4 in the execution time and a factor of 2 in program size. Does the runtime cost justify the protection we gain?

12. Conclusion We have shown that static analysis can be defeated by making the control-flow analysis dependent on the data in the program. Future work includes establishing the practical lower bound on the time needed to analyze a transformed program. Thank you!