The Optimal Research And Implementation Of Fuzz Strategy Based On Peach

Fuzz testing is a software testing technique whose core is the generation of fuzzed data.Fuzz strategy is a kind of strategy which controls how to generate fuzzed data.The existing Sequential strategy and RandomDeterministic strategy are single field test,and Random strategy is more fields test.However,Random strategy generates infinite mutations and runs forever.What's more,even single field test generates large number of mutations in complex scenarios which takes too much time to test.In view of the above problem,the main work of this thesis are as follows:1.Propose Combinatorial strategy which supports both single field test and more fields test.Combinatorial strategy can replace Sequential strategy and RandomDeterministic strategy for single field test.For more fields test,Combinatorial strategy sequentially traverses all the combinations of fields and sequentially traverses all the combinations of mutations for each combination of fields.Therefore,Combinatorial strategy generates finite mutations.Experiment shows that Combinatorial strategy generates relatively few mutations and the test can be finished in limited time when the number of combined fields is two.The number of generated mutations and flaws by Combinatorial strategy is far greater than Sequential strategy and RandomDeterministic strategy.Therefore,Combinatorial strategy is feasible and effective.2.For combinatorial explosion problem,this thesis presents three solutions as follows:only combine those related fields;reduce the number of mutators;limit the maximum number of mutations for each combination of fields.To some extent,the solutions reduce the number of mutations and solve the combinatorial explosion problem.3.This thesis proposes using sample in Fuzz testing to reduce the test time in complex scenarios and comes up with benefit sample and model sample.Benefit sample selects those mutators and mutations which found more flaws.Experiment shows that the test time of benefit sample is much less than normal test and the number of flaws found by benefit sample in the same time is far greater than normal test.Therefore,using benefit sample in Fuzz testing is feasible and effective.Based on the fields of data model,model sample chooses those mutations which led to more flaws for each field.Model sample requires researchers to have professional in-depth knowledge,so it generates targeted mutations.