joshreyn / winafl Goto Github PK

======
WinAFL
======

   Original AFL code written by Michal Zalewski <[email protected]>

   Windows fork written and maintained by Ivan Fratric <[email protected]>

   Copyright 2016 Google Inc. All Rights Reserved.

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at

       http://www.apache.org/licenses/LICENSE-2.0

   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.


1) Background
-------------

AFL is a popular fuzzing tool for coverage-guided fuzzing. The tool combines
fast target execution with clever heuristics to find new execution paths in
the target binary. It has been successfully used to find a large number of
vulnerabilities in real products. For more info about the original project,
please refer to the original documentation at

http://lcamtuf.coredump.cx/afl/

Unfortunately, the original AFL does not work on Windows due to very
*nix-specific design (e.g. instrumentation, forkserver etc). This project is
a fork of AFL that uses different instrumentation approach which works on
Windows even for black box binary fuzzing.

2) The WinAFL approach
----------------------

Instead of instrumenting the code at compilation time, WinAFL relies on dynamic
instrumentation using DynamoRIO (http://dynamorio.org/) to measure and extract
target coverage. This approach has been found to introduce an overhead about 2x
compared to the native execution speed, which is comparable to the original AFL
in binary instrumentation mode.

To improve the process startup time, WinAFL relies heavily on persistant
fuzzing mode, that is, executing multiple input samples without restarting the
target process. This is accomplished by selecting a target function (that the
user wants to fuzz) and instrumenting it so that it runs in a loop.

3) Building WinAFL
------------------

1. Download and build DynamoRio sources or download DynamoRio Windows binary
package from https://github.com/DynamoRIO/dynamorio/wiki/Downloads 

2. Open Visual Studio Command Prompt (or Visual Studio x64 Win64 Command Prompt
if you want a 64-bit build). Note that you need a 64-bit winafl.dll build if
you are fuzzing 64-bit targets and vice versa.

3. Go to the directory containing the source

4. Type the following commands. Modify the -DDynamoRIO_DIR flag to point to the
location of your DynamoRIO cmake files (relative to the source directory).

For a 32-bit build:

mkdir build32
cd build32
cmake .. -DDynamoRIO_DIR=..\path\to\DynamoRIO\cmake
cmake --build . --config Release

For a 64-bit build:

mkdir build64
cd build64
cmake -G"Visual Studio 10 Win64" .. -DDynamoRIO_DIR=..\path\to\DynamoRIO\cmake
cmake --build . --config Release

4) Using WinAFL
---------------

The command line for afl-fuzz on Windows is different than on Linux. Instead of

%s [ afl options ] -- [instrumentation options] -- 

it now looks like this

afl-fuzz [afl options] -- [instrumentation options] -- target_cmd_line

The followin afl-fuzz options are supported:

  -i dir        - input directory with test cases
  -o dir        - output directory for fuzzer findings
  -D dir        - directory containing DynamoRIO binaries (drrun, drconfig)
  -t msec       - timeout for each run
  -f file       - location read by the fuzzed program
  -M \\ -S id   - distributed mode
  -x dir        - optional fuzzer dictionary

Please refer to the original AFL documentation for more info on these flags.

The following instrumentation options are used

  -covtype         - the type of coverage being recorded. Supported options are
                     bb (basic block) or edge (default).

  -coverage_module - module for which to record coverage. Multiple module flags
                     are supported.

  -target_module   - module which contains the target function to be fuzzed.
                     Either -target_method or -target_offset need to be
                     specified together with this option.

  -target_method   - name of the method to fuzz in persistent mode. A symbol
                     for the method needs to be exported for this to work.
                     Otherwise use -target_offset instead.

  -target_offset   - offset of the method to fuzz from the start of the module.

  -fuzz_iterations - Maximum nuber of iterations for the target function to run
                     before restarting the target process.

  -nargs           - Number of arguments the fuzzed method takes. This is used
                     to save/restore the arguments between runs.

  -debug           - Debug mode. Does not try to connect to the server. Outputs
                     a log file containing loaded modules, opened files and
                     coverage infrormation.

  -logdir          - specifies in which directory the log file will be written
                     (only to be used with -debug).

In general, you should perform the following steps when fuzzing a new target:

0. Make sure your target is running correctly without instrumentations.

1. Open the target binary in WinDbg and locate the function you want to fuzz.
Note the offset of the function from the start of the module. For example, if
you want to fuzz the main function and happen to have symbols around, you can
use the following windbg command:

x test!main

2. Make sure that the target is running correctly under DynamoRIO. For this
purpose you can use the standalone debug mode of WinAFL client which does not
require connecting to afl-fuzz. Make sure you use the drrun.exe and winafl.dll
version which corresponds to your target (32 vs. 64 bit).

Example command line

path\to\DynamoRIO\bin64\drrun.exe -c winafl.dll -debug
-target_module test_gdiplus.exe -target_offset 0x1270 -fuzz_iterations 10
-nargs 2 -- test_gdiplus.exe input.bmp

You should see the output corresponding to your target function being run 10
times after which the target executable will exit. A .log file should be
created in the current directory. The log file contains useful information
such as the files and modules loaded by the target as well as the dump of AFL
coverage map. You should see your input file being opened at least 10 times
(or more), as afl-fuzz requires that the file is opened for each fuzzing
iteration. Note the list of loaded modules for setting the -coverage_module
flag. Note that you must use the same values for module names as seen in the
log file (case sensitive).

3. Now you should be ready to fuzz the target. First, make sure that both
afl-fuzz.exe and winafl.dll are in the current directory. As stated earlier,
the command line for afl-fuzz on Windows is

afl-fuzz [afl options] -- [instrumentation options] -- target_cmd_line

Please refer above for the list of supported AFL and instrumentation options.

In AFL options, you must specify the DynamoRIO binaries directory via the new
-D option. You need to match the DynamoRIO and winafl.dll build (32 vs. 64 bit)
to the target binary. -t (timeout) option is mandatory for winafl as execution
time can vary significantly under instrumentation so it’s not a good idea to
rely on the auto-determined values.

You can use the same winafl options as in step 2 but remember to exclude the
-debug flag and you'll probably want to increase the iteration count.

As in afl-fuzz on Linux you can replace the input file param of the target
binary with @@

An example command line would look like

afl-fuzz.exe -i in -o out -D C:\work\winafl\DynamoRIO\bin64 -t 20000 --
-coverage_module gdiplus.dll -coverage_module WindowsCodecs.dll
-fuzz_iterations 5000 -target_module test_gdiplus.exe -target_offset 0x1270
-nargs 2 -- test_gdiplus.exe @@

That’s it. Happy fuzzing!