Research And Application On Fine-grained Memory Access Tracing Model

Software application has been penetrated in all areas of politics, economy, military, culture and everyday life. Improving the ability of analyzing and understanding software may not only help developers maintain and transplant software and reduce the cost of software development, but also help software analyzers test software system, judge whether there are potential threats and exploits hidden inside and finally carry out protection methods to help improve its reliability and stability.Since the state of the art may not meet the demand mentioned above very well, one technology named memory access analysis was proposed to help solve this problem based on the memory access information of executables. Firstly, a fine-grained memory access tracing(short for FMAT) model was established to track and record memory access information, then three algorithms were proposed based on this model:a new memory access tracing algorithm named sokoban was proposed based on deep first traversing, this algorithm was designed for FMAT model and was used to avoid re-recording information of instruction execution of program; an improved variable recognition algorithm named sokocovery was proposed for fine grained intra variable recognition based on semi-naive algorithm of IDA Pro; an intra variable-coupling analysis algorithm named sokocoupling was proposed based on randomization factor and inter-variable coupling, this model was used for quantified description of coupling relationship between variables.The collected memory access information was arranged for application retarget on data type recovery and program protection. The improved data type recovery system named ITIE based on sokocovery algorithm could improve the result accuracy of data type recovery by10%; the program protection method named SLR(stack layout randomization) based on sokocoupling algorithm could reduce the success rate of attacks on programs from different compilers to3%and reduce the success rate of attacks on different types of variables to8%.