dvyukov/centipedeDistributed coverage-guided fuzzing engine compatible with libFuzzer targets; scales to thousands of concurrent jobs, uses sanitizers and corpus distillation to uncover bugs.
Why not? We are currently trying to fuzz some very large and very slow targets for which libFuzzer, AFL, and the like do not necessarily scale well. For one of our motivating examples see SiliFuzz. While working on Centipede we plan to experiment with new approaches to massive-scale differential fuzzing, that the existing fuzzing engines don't try to do.
Notable features:
Out-of-the-box support for libFuzzer-based fuzz targets. In order to use your
favourite LLVMFuzzerTestOneInput() you only need to build your target with
Centipede's compiler and linker options.
Work-in-progress. We test centipede within a small team on a couple of targets. Unless you are part of the Centipede project, or want to help us, you probably don't want to read further just yet.
Scale. The intent is to be able to run any number of jobs concurrently, with very little communication overhead. We currently test with 100 local jobs and with 10k jobs on a cluster.
Out of process. The target runs in a separate process. Any crashes in it will not affect the fuzzer. Centipede can be used in-process as well, but this mode is not the main goal. If your target is small and fast you probably still want libFuzzer.
Integration with the sanitizers is achieved via separate builds. If during fuzzing you want to find bugs with ASAN, MSAN, or TSAN, you will need to provide separate binaries for every sanitizer as well as one main binary for Centipede itself. The main binary should not use any of the sanitizers.
No part of the internal interface is stable. Anything may change at this stage.
A program that produces an infinite stream of inputs for a target and orchestrates the execution.
A binary, a library, an API, or rather anything that can consume bytes for input and produce some sort of coverage data as an output. A libFuzzer's target can be a Centipede's target. Read more here .
A sequence of bytes that can be fed to a target. The input can be an arbitrary bag of bytes, or some structured data, e.g. serialized proto.
A number that represents some unique behavior of the target. E.g. a feature 1234567 may represent the fact that a basic block number 987 in the target has been executed 7 times. When executing an input with the target, the fuzzer collects the features that were observed during execution.
A set of features associated with one specific input.
Some information about the behaviour of the target when it executes a given input. Coverage is usually represented as feature set that the input has triggered in the target.
A function that takes bytes as input and outputs a small random mutation of the input. See also: structure-aware fuzzing .
A function that knows how to feed an input into a target and get coverage in return (i.e. to execute).
A customizable fuzzing engine that allows the user to substitute the Mutator and the Executor.
A library that implements the executor interface expected by the Centipede fuzzer. The runner knows how to run a sancov-instrumented target, collect the resulting coverage, and pass it back to Centipede. Prospective Centipede fuzz targets can be linked with this library to make them runnable by Centipede.
A set of inputs.
A process of choosing a subset of a larger corpus, such that the subset has the same coverage features as the original corpus.
A file representing a subset of the corpus and another file representing feature sets for that same subset of the corpus.
To merge shard B into shard A means: for every input in shard B that has features missing in shard A, add that input to A.
A single fuzzer process. One job writes only to one shard, but may read multiple shards.
A local or remote directory that contains data produced or consumed by a fuzzer.
git clone https://github.com/google/centipede.git
cd centipede
CENTIPEDE_SRC=`pwd`
BIN_DIR=$CENTIPEDE_SRC/bazel-bin
bazel build -c opt :all
What you will need for the subsequent steps:
$BIN_DIR/centipede - the binary of the engine (the fuzzer).$BIN_DIR/libcentipede_runner.pic.a - the library you need to link with your fuzz target (the runner).$CENTIPEDE_SRC/clang-flags.txt - recommended clang compilation flags for the target.You can keep these files where they are or copy them somewhere.
We provide two examples of building the target: one tiny single-file target and libpng. Once you've built your target, proceed to the fuzz target running step.
This example uses one of the simple example fuzz targets, a.k.a. puzzles, included in the Centipede repo.
NOTE: The commands below use the flags from $CENTIPEDE_SRC/clang-flags.txt. You may choose to use some other set of instrumentation flags: clang-flags.txt only provides a simple default option.
FUZZ_TARGET=byte_cmp_4 # or any other source under $CENTIPEDE_SRC/puzzles
clang++ @$CENTIPEDE_SRC/clang-flags.txt -c $CENTIPEDE_SRC/puzzles/$FUZZ_TARGET.cc -o $BIN_DIR/$FUZZ_TARGET.o
This step links the just-built fuzz target with libcentipede_runner.pic.a and other required libraries.
clang++ $BIN_DIR/$FUZZ_TARGET.o $BIN_DIR/libcentipede_runner.pic.a \
-ldl -lrt -lpthread -o $BIN_DIR/$FUZZ_TARGET
Skip to the running step.
LIBPNG_BRANCH=v1.6.37 # You can experiment with other branches if you'd like
git clone --branch $LIBPNG_BRANCH --single-branch https://github.com/glennrp/libpng.git
cd libpng
CC=clang CFLAGS=@$CENTIPEDE_SRC/clang-flags.txt ./configure --disable-shared
make -j
FUZZ_TARGET=libpng_read_fuzzer
clang++ -include cstdlib \
./contrib/oss-fuzz/$FUZZ_TARGET.cc \
./.libs/libpng16.a \
$BIN_DIR/libcentipede_runner.pic.a \
-ldl -lrt -lpthread -lz \
-o $BIN_DIR/$FUZZ_TARGET
Running locally will not give the full scale, but it could be useful during the fuzzer development stage. We recommend that both the fuzzer and the target are copied to a local directory before running in order to avoid stressing a network file system.
WD=$HOME/centipede_run
mkdir -p $WD
NOTE: You may need to add
llvm-symbolizer
to your $PATH for some of the Centipede functionality to work. The
symbolizer can be installed as part of the LLVM
distribution:
sudo apt install llvm
which llvm-symbolizer # normally /usr/bin/llvm-symbolizer
rm -rf $WD/*
$BIN_DIR/centipede --binary=$BIN_DIR/$FUZZ_TARGET --workdir=$WD --num_runs=100
See what's in the working directory