Abstract:In order to solve the problem of blindness and poor efficiency of traditional fuzzing,a feedback fuzzing method was proposed.This method can improve the test effect by optimizing the process of traditional fuzzing.Firstly,original sample files were collected via a web crawler to ensure the diversity of them.Then,the best seed file was selected from a large number of sample files,to avoid redundant testing while ensuring adequate testing space.Finally,test cases were generated by mutating the seed file,and the mutation parameters was continuously adjusted according to the testing result based on improved roulette wheel selection strategy,in order to find as many software failures as possible.A prototype system named OSSRWSFuzzer was implemented,which had found 56 failures of WPS Office for Linux in experiments,and its test effect is obviously better than that of the existing fuzzers.