Introduction

UTopia is a tool for automatically generating fuzz drivers from unit tests.

UTopia will let developers perform fuzz testing without special knowledge about writing fuzzers. Even developers familiar with fuzzing can save significant time by generating fuzz drivers automatically.

UTopia supports C/C++ libraries which have unit tests with GoogleTest, Boost.Test or Tizen TCT.

Trophy

To see bugs found by UTopia, visit Trophy page. You can also see some UTopia-based fuzzers there.

Docker

For easy set up, we provide docker image for running UTopia. You can build UTopia and generate/run fuzzers inside docker container.

Build docker image with command below.

docker buildx build -f docker/Dockerfile -t utopia . #llvm-10
docker buildx build --build-arg LLVM_VERSION=12 -f docker/Dockerfile-12 -t utopia . #llvm-12

Build

UTopia depends on LLVM, Protobuf and GoogleTest. You can install dependencies manually, but we recommend to use given docker image.

To build UTopia, follow cmake process below.

cd $UTOPIA_HOME_DIR
cmake -B build -S .
cmake --build build -j$(nproc)

Run

For some selected projects, you can use helper script to run our tool without further effort. Please refer helper/README.md. For other projects, please check following manual.

target_analyzer

Target Analyzer analyzes target library code and generates a result as json file. Mandatory command line options are below.

target_analyzer --db ${builddb_path} --extern ${extern_path} --public ${api_json_path} --out ${output_path}

builddb_path

Build db is a json file that contains paths of AST and IR files of target library code. Its format looks like below.

{
  "bc": "/root/fuzz-test-generation/exp/sample/output/bc/libcommon.a.bc",
  "ast": [
    "/root/fuzz-test-generation/exp/sample/libcommon.a_ast/codec/common/src/ast1.o.ast",
    "/root/fuzz-test-generation/exp/sample/libcommon.a_ast/codec/common/src/ast2.o.ast"
  ],
  "project_dir": "/root/fuzz-test-generation/exp/sample"
}

LLVM bitcode file path of a specific library should be specified using "bc" keyword. We only accept one bitcode file so far, thus you may use llvm-link to link several bit code files to one bit code file.

AST file paths of a specific library should be specified using "ast" keyword. Note that a specified bitcode file and ast files are generated from same source codes for a specific library.

extern_path

Target Analyzer accepts target analyzer report of other libraries for an accurate result. This path should be a directory path where other reports stored, which means more than one library reports are allowed.

api_json_path

API function names to be analyzed. It should be json file formatted as below.

{
  "libcommon.a": [
    "API1",
    "API2",
    "API3"
  ]
}

You can get API list from a specific library using command below.

nm --no-demangle --defined-only -g ${librarypath} | awk '$2=="T" {k=""; for(i=3;i<=NF;i++) k=k $i""; print k}'

Report

Direction

The Direction property is an essential parameter that the target analyzer utilizes. It signifies whether a parameter is employed for reading (Dir_In), writing (Dir_Out), or both reading and writing (Dir_In | Dir_Out) within a function. The target analyzer delineates this property through an enumeration type comprising the following elements:

enum Dir {
  Dir_NoOp = 0x000,        // No operation
  Dir_In = 0x100,          // Input direction
  Dir_Out = 0x010,         // Output direction
  Dir_Unidentified = 0x001 // Unidentified direction
};

For instance, in the provided JSON file snippet:

{
  "Direction": {
    "BF_crypt(0)": 256,
    "BF_crypt(1)": 272,
    "BF_crypt(2)": 272,
    "BF_decode(1)": 272
  }
}

The entry "BF_crypt(0)": 256 indicates that the direction of the first parameter in the BF_crypt function is set to 256 (0x100), signifying an In direction, meaning it is used for input.
Similarly, "BF_crypt(1)": 272 reveals that the direction of the second parameter in the BF_crypt function is 272 (0x110), indicating it has both In and Out directions, meaning it is utilized for both input and output.

This notation helps in understanding how each parameter within a function is used, whether for input, output, or both, providing clear insights into the data flow and operations performed by the function.

ut_analyzer

UT Analyzer analyzes unit test code for a target library and generates a result as json file. Mandatory command line options are below.

ut_analyzer --entry ${entry_path} --extern ${extern_path} --ut ${ut_type} --name ${lib_name} --public ${api_json_path} --out ${output_path}

Most options are same as the target_analyzer. Note that, entry_path should specify AST/IR files for a unit test executable, not library.

ut_type

Framework used by the target project, could be tct, gtest, or boost.

fuzz_generator

fuzz_generator generates fuzz drivers using report files of target_anlayzer and ut_analyzer. Mandatory command line options are below.

fuzz_generator --src ${src_path} --target ${target_analyzer_report_path} --ut ${ut_analyzer_report_path} --public ${api_json_path} --out ${output_dir}

src_path

src_path is the directory path where unit test source code is stored. fuzz_generator copy this directory and generates fuzz driver by modifying these copied files.

Other options are same as the target_analyzer options.

How fuzz driver works

This section describes the functionality and implementation details of the generated fuzz driver, which is crucial for understanding how fuzz testing is integrated with the target source code.

fuzz_entry.cc (Auto-Generated File)

DEFINE_PROTO_FUZZER(const AutoFuzz::FuzzArgsProfile &autofuzz_mutation) {
  ... /* Values are assigned from autofuzz_mutation */
  enterAutofuzz();
}

This function serves as the entry point for the generated fuzz driver. It takes values generated by the fuzzer and assigns them to variables that are subsequently used to invoke library functions. Finally, the function calls enterAutofuzz(); to proceed with the fuzz testing process.

Source Code Defining the Target Test Case

#ifdef __cplusplus
extern "C" {
#endif
void enterAutofuzz() {
  class AutofuzzTest : public ::Parser_TestArray_Test {
  public:
    void runTest() {
      try {
        SetUpTestCase();
      } catch (std::exception &E) {}
      try {
        SetUp();
      } catch (std::exception &E) {}
      try {
        TestBody();
      } catch (std::exception &E) {}
      try {
        TearDown();
      } catch (std::exception &E) {}
      try {
        TearDownTestCase();
      } catch (std::exception &E) {}
    }
  };
  AutofuzzTest Fuzzer;
  Fuzzer.runTest();
}
#ifdef __cplusplus
}
#endif

In the final part of the source code that defines the target test case, UTopia injects the enterAutofuzz function. Within this function, the AutofuzzTest class is declared, inheriting from ::Parser_TestArray_Test. This parent class is defined by the GoogleTest framework and is specific to the test case being addressed, as shown below:

TEST(Parser, TestArray)
{
  ...
}

The runTest() method executes the test case independently of GoogleTest, invoking five functions that GoogleTest typically calls for each test case. This approach allows for direct test execution without relying on the GoogleTest framework.

Build generated fuzz drivers

Fuzz drivers are generated in ${output_dir} passed to fuzz_generator as command line option. You can build them using same compiler command for unit test executable. Note that, you should include fuzz_entry.cc, FuzzArgsProto.pb.cc files that are generated by fuzz_generator. You can find those files in ${output_dir}.

Reproduce Evaluation

Generate Fuzzer

python3 -m helper.make {library name}
python3 -m helper.build {library name}

You can find every outputs from the whole pipeline of 'UTopia' under following two directories:

exp/{library name}/output,
result/test/{library name}

Samsung / UTopia

readme