Research On Key Techniques Of Dynamic Program Analysis Framework And Framework Application

Computer software have made a great process,but the vulnerability goes incurable in the meantime.Software becomes powerful with different and complex functions,and so does the vulnerability.Like intrusion attack,denial of service attack and misuse of the resource,kinds of threatens are challenging our society.To defense them,researchers have proposed software analysis techniques from static methods to dynamic methods.We do benefit from these techniques,especially dynamic binary analysis since so many dynamic binary analysis(DBA)applications had been built.But a specific tool is somehow hard to expand or integrate with others unless we cost more,so we wish to build a dynamic binary analysis framework for various applications which can work efficiently and be built quickly,cheaply.In this paper,we propose a framework,named PinLLVM,which integrates kinds of key techniques for DBA framework with rich functions.PinLLVM now have applied into various DBA tools including Instruction Count,Memory Trace,Taint Analysis,Detection and Concolic Execution.In particular,we work much on concolic execution techniques which is developing but performed excellent in recent years.Concolic execution comes from the evolution of symbolic execution,and can explore the program execution path and potential defects through managing symbol information in execution and solving the path constraints with solvers.We build a concolic execution tool,named ConcTool,to support our related research and verify the framework as well.Concolic execution grows with the challenges,like path explosion.It becomes a challenge because of the large scale of real world softwares.To alleviate the problem,we propose the parallel concolic execution as a countermeasure,named PACCI(short for PArallelized Concolic execution based on Cloud Infrastructure).Experimental results show that PACCI is not only effective,but scalable(e.g.,gaining about 20 X speedup using 24 nodes).Furthermore,its efficiency declines slightly about 5% and 6.1% under resource fluctuations and node failures,respectively.