Fuzzing with Code Fragments Christian Holler Kim Herzig Andreas Zeller Mozilla Corporation∗ Saarland University Saarland University
[email protected] [email protected] [email protected] Abstract JavaScript interpreter must follow the syntactic rules of JavaScript. Otherwise, the JavaScript interpreter will re- Fuzz testing is an automated technique providing random ject the input as invalid, and effectively restrict the test- data as input to a software system in the hope to expose ing to its lexical and syntactic analysis, never reaching a vulnerability. In order to be effective, the fuzzed input areas like code transformation, in-time compilation, or must be common enough to pass elementary consistency actual execution. To address this issue, fuzzing frame- checks; a JavaScript interpreter, for instance, would only works include strategies to model the structure of the de- accept a semantically valid program. On the other hand, sired input data; for fuzz testing a JavaScript interpreter, the fuzzed input must be uncommon enough to trigger this would require a built-in JavaScript grammar. exceptional behavior, such as a crash of the interpreter. Surprisingly, the number of fuzzing frameworks that The LangFuzz approach resolves this conflict by using generate test inputs on grammar basis is very limited [7, a grammar to randomly generate valid programs; the 17, 22]. For JavaScript, jsfunfuzz [17] is amongst the code fragments, however, partially stem from programs most popular fuzzing tools, having discovered more that known to have caused invalid behavior before. LangFuzz 1,000 defects in the Mozilla JavaScript engine. jsfunfuzz is an effective tool for security testing: Applied on the is effective because it is hardcoded to target a specific Mozilla JavaScript interpreter, it discovered a total of interpreter making use of specific knowledge about past 105 new severe vulnerabilities within three months of and common vulnerabilities.