This directory provides the prototype of the paper: "Titan: Efficient Multi-target Directed Greybox Fuzzing"[S&P 2024].

An easier way to run Titan on the fuzzing benchmark Magma is to move this repository into "magma/fuzzers" of magma repository and then follow the guidance to start fuzzing. For some specific modifications to ensure correct deployment, please refer to the build_targets repo.

For fuzzing other programs not included in Magma, you could refer to the following scripts.

• preinstall.sh: Support environment.
• instrument.sh: Generate binary for fuzzing.
• run.sh: Start fuzzing.

export TITAN=<path_of_TITAN_repository>

apt-get update --fix-missing && \
    apt-get install -y make build-essential git wget cmake gawk 

apt-get install -y libtinfo-dev 
apt-get install -y libcap-dev zlib1g-dev

# llvm-4.0
apt-get install -y libtinfo5
apt-get install -y xz-utils
wget -q https://releases.llvm.org/4.0.0/clang+llvm-4.0.0-x86_64-linux-gnu-ubuntu-16.10.tar.xz
tar -xf clang+llvm-4.0.0-x86_64-linux-gnu-ubuntu-16.10.tar.xz
rm clang+llvm-4.0.0-x86_64-linux-gnu-ubuntu-16.10.tar.xz

cp -r clang+llvm-4.0.0-x86_64-linux-gnu-ubuntu-16.10 /usr/llvm
cp -r /usr/llvm/bin/* /usr/bin 
cp -r /usr/llvm/lib/* /usr/lib
cp -r /usr/llvm/include/* /usr/include 
cp -r /usr/llvm/share/* /usr/share

apt-get install -y python3 python3-dev python3-pip
pip3 install --upgrade pip
pip3 install wllvm

It is recommended to run Titan under a new folder $TITAN/Outputs to make sure the output files are gathered in a common folder.

mkdir $TITAN/Outputs; cd $TITAN/Outputs

Generate the bitcode file for the target project by wllvm.

The static analysis engine used in Titan is similar to Beacon(S&P'22). You can have more details by accessing its repo.

$TITAN/prototype/precondInfer <target.bc> --target-file=<cstest.txt> --join-bound=1

Inputs:

• <target.bc> is the bitcode file for the target project.
• <cstest.txt> has multiple lines, which record the location of multiple targets. Each line is in the form of “fileName:lineNum” (e.g. parser.c:66 means that the target for directed fuzzing is at Line 66 of parser.c).

Outputs:

• range_res.txt: range analysis result.
• transed.bc: The slightly transformed bc for further processing.
• bug_conf_cluster: Cluster info for conflict relation.
• bug_over_cluster: Cluster info for over relation.

Notice that the independent information is included in the above two files.

$TITAN/prototype/Ins -output=$TITAN/Outputs/fuzz.bc -afl -res=$TITAN/Outputs -log=$TITAN/Outputs/log_Ins.txt -load=$TITAN/Outputs/range_res.txt $TITAN/Outputs/transed.bc

clang $TITAN/Outputs/fuzz.bc -o $TITAN/Outputs/fuzz_bin -lm -lz $TITAN/prototype/afl-llvm-rt.o

Finally, fuzz all the things!

$TITAN/prototype/afl-fuzz -i <seed_dir> -o $TITAN/Outputs/fuzz_out -s "$TITAN/Outputs/bug_conf_cluster" -k "$TITAN/Outputs/bug_over_cluster" -- $TITAN/Outputs/fuzz_bin <other_parameters> @@

Currently, Titan uses sequential static analysis for each target. Even though it is affordable as an offline one-time effort for the evaluation, it may become expensive in practice. One potential solution is to extend our static analysis as a multi-thread/process version, which can significantly reduce the analysis time. This orthogonal problem may also become a research question for efficient parallel static analysis in future work. For more details on implementation and potential discussion, please feel free to drop an email or open an issue in the issue track.