A fuzzing management tools collection
With this project, we aim to create a management toolchain for fuzzing. Unlike other toolchains and frameworks, we want to be modular in such a way that you can use those parts of FuzzManager that seem interesting to you without forcing a process upon you that does not fit your requirements.
CrashManager is the part of FuzzManager responsible for managing crash results submitted to the server. The main features are:
Store crash information gathered from various sources. See FTB.
Bucket crashes using flexible, human-readable signatures that can match a large number of symptoms of a crash, are proposed by the server but can be altered and tuned by the user. The server also includes semi-automatic optimization of signatures that helps you group duplicates into one bucket.
Report bugs directly to a bug tracker using the best testcase available. We support only Bugzilla as a bugtracker right now, but again the API is designed to be extendable.
FTB (Fuzzing Tool Box) is the underlying library that contains classes for parsing crash output from various tools (CrashInfo), bucketing crashes (CrashSignature), and parsing assertions (AssertionHelper). This can be used locally without having a running FuzzManager server instance to support crash logging and bucketing. FTB already supports a variety of tools like GDB, ASan and Minidumps but can be extended to support any form of crash information you would like.
Collector is a command-line utility or a Python class that can be used to communicate with a CrashManager server. Collector provides an easy client interface that allows your clients to submit crashes as well as download and match existing signatures to avoid reporting frequent issues repeatedly.
EC2SpotManager is another (optional) part of FuzzManager that allows you to manage large pools of spot instances in the Amazon Cloud. It supports hierachical configurations to avoid configuration duplication and ensures that your instances are always running in the desired quantity as well as in the cheapest zone.
Please send any questions regarding the project to choller-at-mozilla-dot-com.
The client portion of FuzzManager (FTB and Collector) can be installed with
pip install FuzzManager. This is all you need if you just need to talk to a FuzzManager server instance or use FTB locally.
The server part of FuzzManager is a Django application. Please note that it requires the full repository to be checked out, not just the server directory.
Server dependencies are listed in
server/requirements.txt. You can use pip to install these dependencies using
pip install -r server/requirements.txtand/or you can use your distribution's package management to fulfill them. A Redis server is also required for EC2SpotManager, and can be installed on a Debian-based Linux with:
sudo apt-get install redis-server.
You can set the server up just like any other Django project. The Django configuration file is found at
server/server/settings.py. The default will work, but for a production setup, you should at least review the database settings.
Afterwards, you should run the following commands
$ cd server $ python manage.py migrate
Create the fuzzmanager user.
$ python ./manage.py createsuperuser Username (leave blank to use 'user'): fuzzmanager Email address: [email protected] Password: Password (again): Superuser created successfully.Get fuzzmanager authorization token
$ python manage.py get_auth_token fuzzmanager 4a253efa90f514bd89ae9a86d1dc264aa3133945Since the fuzzmanager account is used as a service account, we need to set the http basic authentication password to the auth token.
htpasswd -cb .htpasswd fuzzmanager 4a253efa90f514bd89ae9a86d1dc264aa3133945`This .htpasswd file can be stored anywhere on your hard drive. Your Apache AuthUserFile line should be updated to reflect your path. See examples/apache2/default.vhost for an example
It is important that you edit FuzzManager/server/settings.py and adjust the following variables according to your needs.
ALLOWED_HOSTS = 
You may also want to increase the maximum size in bytes allowed in a request body. The default of 2.5MB may not be enough in some cases by adding the following variable.
For local testing, you can use the builtin debug webserver:
python manage.py runserver
For a production setup, see the next section about Apache+WSGI.
To properly run FuzzManager in a production setup, using Apache+WSGI is the recommended way.
examples/apache2/directory you'll find an example vhost file that shows you how to run FuzzManager in an Apache+WSGI setup. Of course you should adjust the configuration to use HTTPs if you don't plan to use any sort of TLS load balancer in front of it.
Use the following command to get an authentication token for a Django user:
python manage.py get_auth_token username
You can use the user that you created during
syncdbfor simple setups.
The following is an example crontab using
cronicto run several important FuzzManager jobs:
# Fetch the status of all bugs from our external bug tracker(s) */15 * * * * cd /path/to/FuzzManager/server && cronic python manage.py bug_update_status # Cleanup old crash entries and signatures according to configuration */30 * * * * cd /path/to/FuzzManager/server && cronic python manage.py cleanup_old_crashes # Attempt to fit recently added crash entries into existing buckets */5 * * * * cd /path/to/FuzzManager/server && cronic python manage.py triage_new_crashes # Export all signatures to a zip file for downloading by clients */30 * * * * cd /path/to/FuzzManager/server && cronic python manage.py export_signatures files/signatures.new.zip mv files/signatures.new.zip files/signatures.zip
A docker image is available by building the
You can easily run a local server (and Mysql database server) by using docker-composer:
On a first run, you must execute the database migrations:
docker-compose exec backend python manage.py migrate
And create a super user to be able to login on http://localhost:8000
docker-compose exec backend python manage.py createsuperuser
By default the docker image uses Django settings set in Python module
server.settings_docker, with the following settings: -
DEBUG = Falseto enable production mode -
ALLOWED_HOSTS = ["localhost", ]to allow development usage on
You can customize settings by mounting a file from your host into the container:
volumes: - "./settings_docker.py:/src/server/server/settings_docker.py:ro"
In order to talk to FuzzManager, your fuzzer should use the client interface provided, called the Collector. It can be used as a standalone command line tool or directly as a Python class in case your fuzzer is written in Python.
We'll first describe how to use the class interface directly from Python. If you want to use the command line interface instead, I still suggest that you read on because the command line interface is very similar to the class interface in terms of functionality and configuration.
For simple cases where you can just (re)run a command with a testcase that produces a crash, we also provide an easy report class that runs your command and figures out all the crash information on its own. You will find the description of this mode at the end of this section as it still requires configuration files to be setup properly, but tl;dr, it can be as easy as:
$ python Collector.py --autosubmit mybadprogram --someopt yourtest
And you're done submitting everything, crash information as well as program information.
The Collector constructor takes various arguments that are required for later operations. These arguments include a directory for signatures, server data such as hostname, port, etc. as well as authentication data and a client name. However, the preferred way to pass these options is not through the constructor, but through a configuration file. The constructor will try to read the configuration file located at ~/.fuzzmanagerconf and use any parameters from there if it hasn't been explicitly specified in the constructor call. This makes deployment very easy and saves time. An example configuration could look like this:
[Main] sigdir = /home/example/signatures serverhost = 127.0.0.1 serverport = 8000 serverproto = http serverauthtoken = 4a253efa90f514bd89ae9a86d1dc264aa3133945
With this file present and readable, instantiating the Collector doesn't require any further arguments.
Several methods of the collector work with the
CrashInfoclass. This class stores all the necessary data about a crash. In order to get a CrashInfo instance, you need:
The first three sets of data are typically already available in a fuzzer. Note that for GDB traces, the trace should contain first the stack trace, then a dump of all registers and then a dissassembly of the program counter (see also the FTB/Running/AutoRunner.py file which demonstrates how to output all information properly for FuzzManager).
The last thing required is the
ProgramConfiguration. This class is largely a container class storing various properties of the program, e.g. product name, the platform, version and runtime options. Instead of instantiating the class and providing all the data manually, it is again recommended to use the configuration file support. Assuming your binary is located at /home/example/foo then creating a configuration file at /home/example/foo.fuzzmanagerconf with the necessary data is recommended. Such a file could look like this:
[Main] platform = x86 product = mozilla-central product_version = 70de2960aa87 os = linux
[Metadata] pathPrefix = /srv/repos/mozilla-central/ buildFlags = --enable-optimize --enable-posix-nspr-emulation --enable-valgrind --enable-gczeal --target=i686-pc-linux-gnu --disable-tests --enable-debug
Once this file is present, you can call
ProgramConfiguration.fromBinarywith your binary path and the configuration will be created from the file. You can add program arguments and environment variables through the provided
addEnvironmentVariablesmethods afterwards. Finally, call
CrashInfo.fromRawCrashDatawith all of the described data. Here's a simple example:
# Note: This could fail and return None when the configuration is missing or throw if misconfigured configuration = ProgramConfiguration.fromBinary(opts.binary) configuration.addEnvironmentVariables(env) configuration.addProgramArguments(args) crashInfo = CrashInfo.fromRawCrashData(stdout, stderr, configuration, auxCrashData=crashdata)
refreshmethod of our Collector instance will download a zipfile from the server, containing the signatures and metadata exported by the server. Once the download is complete, the Collector will first delete all signatures including their metadata from the signature directory. Then the downloaded zipfile is extracted.
searchmethod is the first of a few methods requiring a
crashInfovariable. Create it as described above and the Collector will search inside the signature directory for any matching signatures. Upon match, it will return a tuple containing the filename of the signature matching as well as a metadata object corresponding to that signature.
submitmethod can be used to send a crash report to the FuzzManager server. Again the
crashInfoparameter works as described above. In addition, you can provide a file containing a test and an optional "quality" indicator of the test (best quality is 0). The use of this quality indicator largely depends on how your fuzzer/reducer works. The server will prefer better qualities when proposing test cases for filing bugs. Finally, the method accepts an additional metadata parameter which can contain arbitrary information that is stored with the crash on the server. Note that this metadata is combined with the metadata found in the
crashInfo. When using binary configuration files, this means that the metadata supplied in that configuration file is automatically submitted with the crash to the server.
Further methods of the Collector include
generatefor generating signatures locally and
downloadfor downloading testcases from the server. Both methods work as documented in the source code and are only useful in special cases depending on the application scenario.a
If your crashes can be reproduced on the command line by just running a command with your testcase, then you can use the automated submit method (
--autosubmitin the command line client) and just pass the failing command line to the client. The client will automatically run the target program, gather crash and program configuration and submit it to the server. Of course this mode requires that both the global configuration file as well as the binary configuration file are present.