Skip to content
Go Fuzzit Example
Shell Go
Branch: master
Clone or download
Latest commit b0b044d Aug 14, 2019

Build Status fuzzit

Continuous Fuzzing for Golang Example

This is an example of how to integrate your go-fuzz targets with the Fuzzit Continuous Fuzzing Platform (Go support is currently in Alpha).

This example will show the following steps:


  • Fuzzit will run the fuzz targets continuously on a daily basis with the latest release.
  • Fuzzit will run sanity tests on every pull-request with the generated corpus and crashes to catch bugs early on.

Fuzzing for go can help find both complex bugs, as well as correctness bugs. Go is a safe language so memory corruption bugs are very unlikely to happen, but some bugs can still have security implications.

This tutorial focuses less on how to build go-fuzz targets and more on how to integrate the targets with Fuzzit. A lot of great information is available at the go-fuzz repository.

Building go-fuzz Target

The targets that are currently supported on Fuzzit are targets that utilize the libFuzzer engine. This is why we will use the -libfuzzer flag of go-fuzz and compile it on a Linux machine (should also be supported on mac in the future)

Understanding the bug

The bug is located at parser_complex.go in the following code

package parser

func ParseComplex(data [] byte) bool {
	if len(data) == 5 {
		if data[0] == 'F' && data[1] == 'U' && data[2] == 'Z' && data[3] == 'Z' && data[4] == 'I' && data[5] == 'T' {
			return true
	return false

This is the simplest example to demonstrate a classic off-by-one/out-of-bounds error which causes the program to crash. Instead of len(data) == 5 the correct code will be len(data) == 6.

Understanding the fuzzer

the fuzzer is located at parse_complex_fuzz.go in the following code:

// +build gofuzz

package parser

func Fuzz(data []byte) int {
	return 0

Setting up the development environment

docker run -it golang:1.12.7-buster /bin/bash

# Download Clang-9
echo "deb llvm-toolchain-buster main" >> /etc/apt/sources.list
echo "deb-src llvm-toolchain-buster main" >> /etc/apt/sources.list
wget -O -| apt-key add -
apt update && apt install -y clang-9 lldb-9 lld-9

# Download go-fuzz and go-fuzz-build
go get -u

# Download this example
go get

Building the fuzzer

cd /go/src/
go-fuzz-build -libfuzzer -o fuzzer.a .
clang-9 -fsanitize=fuzzer fuzzer.a -o fuzzer

Running the fuzzer


Will print the following output and stacktrace:

INFO: Seed: 3709860458
INFO: 65536 Extra Counters
INFO: -max_len is not provided; libFuzzer will not generate inputs larger than 4096 bytes
INFO: A corpus is not provided, starting from an empty corpus
#2      INITED ft: 4 corp: 1/1b exec/s: 0 rss: 25Mb
#213    NEW    ft: 6 corp: 2/6b lim: 6 exec/s: 0 rss: 25Mb L: 5/5 MS: 1 CMP- DE: "\x01\x00\x00\x00"-
#13142  NEW    ft: 7 corp: 3/11b lim: 128 exec/s: 0 rss: 25Mb L: 5/5 MS: 4 EraseBytes-ChangeByte-ShuffleBytes-InsertByte-
#104833 NEW    ft: 8 corp: 4/16b lim: 1030 exec/s: 104833 rss: 25Mb L: 5/5 MS: 1 ChangeByte-
#262144 pulse  ft: 8 corp: 4/16b lim: 2589 exec/s: 87381 rss: 25Mb
#524288 pulse  ft: 8 corp: 4/16b lim: 4096 exec/s: 74898 rss: 25Mb
#1048576        pulse  ft: 8 corp: 4/16b lim: 4096 exec/s: 74898 rss: 25Mb
#1275694        NEW    ft: 9 corp: 5/21b lim: 4096 exec/s: 75040 rss: 25Mb L: 5/5 MS: 1 ChangeByte-
#1293550        NEW    ft: 10 corp: 6/26b lim: 4096 exec/s: 76091 rss: 25Mb L: 5/5 MS: 1 CopyPart-
panic: runtime error: index out of range

goroutine 17 [running, locked to thread]:
        /go/src/, 0x5, 0x5, 0xc00001e040)
        /go/src/ +0x1b2
main.LLVMFuzzerTestOneInput(0x2aabb20, 0x5, 0x545b78)
        /tmp/go-fuzz-build316206684/gopath/src/ +0x84
main._cgoexpwrap_90699947e885_LLVMFuzzerTestOneInput(0x2aabb20, 0x5, 0x2aaab10)
        _cgo_gotypes.go:64 +0x37
==4262== ERROR: libFuzzer: deadly signal
    #0 0x45c110 in __sanitizer_print_stack_trace (/go/src/
    #1 0x43b79b in fuzzer::PrintStackTrace() (/go/src/
    #2 0x422123 in fuzzer::Fuzzer::CrashCallback() (/go/src/
    #3 0x7f0ba60ff72f  (/lib/x86_64-linux-gnu/
    #4 0x4acc70 in runtime.raise /tmp/go-fuzz-build316206684/goroot/src/runtime/sys_linux_amd64.s:149

NOTE: libFuzzer has rudimentary signal handlers.
      Combine libFuzzer with AddressSanitizer or similar for better crash reports.
SUMMARY: libFuzzer: deadly signal
MS: 1 ChangeByte-; base unit: 89b92cdd9bcb9b861c47c0179eff7b3a9baafcde
artifact_prefix='./'; Test unit written to ./crash-df779ced6b712c5fca247e465de2de474d1d23b9
Base64: RlVaWkk=

Integrating with Fuzzit from CI

The best way to integrate with Fuzzit is by adding a two stages in your Continuous Build system (like Travis CI or Circle CI).

Fuzzing stage:

  • build a fuzzing target
  • download fuzzit cli
  • create a fuzzing job by uploading the fuzzing target

Regression stage

  • build a fuzzing target
  • download fuzzit cli
  • create a local regression fuzzing job - This will pull all the generated corpuses and run them through the fuzzing binary. If new bugs are introduced this will fail the CI and alert

Here is the relevant snippet from the which is being run by .travis.yml

if [ -z "${FUZZIT_API_KEY}" ]; then
    echo "Please set env variable FUZZIT_API_KEY to api key for your project"
    echo "Api key for your account:<ACCOUNT>/settings"
    exit 1

wget -q -O fuzzit
chmod a+x fuzzit

export TARGET=example-go
GIT_BRANCH=`git rev-parse --abbrev-ref HEAD`
GIT_COMMIT=`git rev-parse --short HEAD`
./fuzzit create job --type $1 --branch $GIT_BRANCH --revision $GIT_COMMIT $TARGET ./fuzzer

In production it is advised to download a pinned version of the CLI like in the example. In development you can use the latest version:${OS}_${ARCH}. Valid values for ${OS} are: Linux, Darwin, Windows. Valid values for ${ARCH} are: x86_64 and i386.

The steps are:

  • Authenticate with the API key (you should keep this secret) from the fuzzit settings dashboard.
  • Upload the fuzzer via create job command and create the fuzzing job. In This example we use two type of jobs:
    • A fuzzing job which is run on every push to master, that continues the previous job with the new release. This means the current corpus is kept and the fuzzer will try to find new paths in the newly added code.
    • In a Pull-Request the fuzzer will run a quick regression test running the fuzzer through all the generated corpuses and crashes to see if the Pull-Request doesnt introduce old or new crashes. This will be alred via the configured channel in the dashboard.
  • The name of the target is not a secret.

Each target has its own corpus and crashes.

You can’t perform that action at this time.