==================
american fuzzy lop
==================

  Written and maintained by Michal Zalewski <lcamtuf@google.com>

  Copyright 2013, 2014, 2015 Google Inc. All rights reserved.
  Released under terms and conditions of Apache License, Version 2.0.

  For new versions and additional information, check out:
  http://lcamtuf.coredump.cx/afl/

  To compare notes with other users or get notified about major new features,
  send a mail to <afl-users+subscribe@googlegroups.com>.

1) Challenges of guided fuzzing
-------------------------------

Fuzzing is one of the most powerful and proven strategies for identifying
security issues in real-world software; it is responsible for the vast
majority of remote code execution and privilege escalation bugs found to date
in security-critical software.

Unfortunately, fuzzing also offers fairly shallow coverage, because many of the
mutations needed to reach new code paths are exceedingly unlikely to be hit 
purely by chance.

There have been numerous attempts to solve this problem by augmenting the
process with additional information about the behavior of the tested code,
ranging from simple corpus distillation, to flow analysis (aka "concolic"
execution), to pure symbolic execution, to static analysis.

The first method on that list has been demonstrated to work well, but depends
on the availability of a massive, high-quality corpus of valid input data. On
top of this, coverage measurements provide only a fairly simplistic view of
program state, making them less suited for guiding the fuzzing process later on.

The remaining techniques are extremely promising in experimental settings, but
frequently suffer from reliability problems or irreducible complexity. Most of
the high-value targets have enough internal states and possible execution paths
to make such tools fall apart and perform strictly worse than their traditional
counterparts, at least until fine-tuned with utmost care.

2) The afl-fuzz approach
------------------------

American Fuzzy Lop is a brute-force fuzzer coupled with an exceedingly simple
but rock-solid instrumentation-guided genetic algorithm. It uses an enhanced
form of edge coverage to easily detect subtle, local-scale changes to program
control flow, without being bogged down by complex comparisons between multiple
long-winded execution paths.

Simplifying a bit, the overall algorithm can be summed up as:

  1) Load user-supplied initial test cases into the queue,

  2) Take next input file from the queue,

  3) Attempt to trim the test case to the smallest size that doesn't alter
     the measured behavior of the program,

  4) Repeatedly mutate the file using a balanced and well-researched variety
     of traditional fuzzing strategies,

  5) If any of the generated mutations resulted in a new state transition
     recorded by the instrumentation, add mutated output as a new entry in the
     queue.

  6) Go to 2.

The discovered test cases are also periodically culled to eliminate ones that
have been obsoleted by newer, higher-coverage finds, and undergo several other
instrumentation-driven effort minimization steps.

The strategies mentioned in step 4 are fairly straightforward, but go well
beyond the functionality of tools such as zzuf and honggfuzz and lead to
additional finds; this is discussed in more detail in technical_notes.txt.

As a side result of the fuzzing process, the tool creates a small,
self-contained corpus of interesting test cases. These are extremely useful
for seeding other, labor- or resource-intensive testing regimes - for example,
for stress-testing browsers, office applications, graphics suites, or
closed-source tools.

The fuzzer is thoroughly tested to deliver coverage far superior to blind
fuzzing or coverage-only tools without the need to dial in any settings or
adjust any knobs.

3) Instrumenting programs for use with AFL
------------------------------------------

When source code is available, instrumentation can be injected by a companion
tool that works as a drop-in replacement for gcc or clang in any standard build
process for third-party code.

The instrumentation has a fairly modest performance impact; in conjunction with
other optimizations implemented by afl-fuzz, most programs can be fuzzed as fast
or even faster than possible with traditional tools.

The correct way to recompile the target program may vary depending on the
specifics of the build process, but a nearly-universal approach would be:

$ CC=/path/to/afl/afl-gcc ./configure
$ make clean all

For C++ programs, you will want:

$ CXX=/path/to/afl/afl-g++ ./configure

The clang wrappers (afl-clang and afl-clang++) are used in the same way; clang
users can also leverage a higher-performance instrumentation mode described in
llvm_mode/README.llvm.

When testing libraries, it is essential to either link the tested executable
against a static version of the instrumented library, or to set the right
LD_LIBRARY_PATH. Usually, the simplest option is just:

$ CC=/path/to/afl/afl-gcc ./configure --disable-shared

Setting AFL_HARDEN=1 when calling 'make' will cause the CC wrapper to
automatically enable code hardening options that make it easier to detect
simple memory bugs. The cost of this is a <5% performance drop.

Oh: when using ASAN, see the notes_for_asan.txt file for important caveats.

4) Instrumenting binary-only apps
---------------------------------

When fuzzing closed-source programs that can't be easily recompiled with
afl-gcc, the fuzzer offers experimental support for fast, on-the-fly
instrumentation of black-box binaries. This is accomplished with a
version of QEMU running in the lesser-known "user space emulation" mode.

QEMU is a project separate from AFL, but you can conveniently build the
feature by doing:

$ cd qemu_mode
$ ./build_qemu_support.sh

For additional instructions and caveats, see qemu_mode/README.qemu.

The mode isn't free; compared to compile-time instrumentation, the fuzzing
process will be approximately 2-5x slower; it is also less conductive to
parallelization on multiple cores.

5) Choosing initial test cases
------------------------------

To operate correctly, the fuzzer requires one or more starting file containing
the typical input normally expected by the targeted application. There are
two basic rules:

  - Keep the files small. Under 1 kB is ideal, although not strictly necessary.
    For a discussion of why size *really* matters, see perf_tips.txt.

  - Use multiple test cases only if they are fundamentally different from
    each other. There is no point in using fifty different vacation photos to
    fuzz an image library.

You can find quite a few good examples of starting files in the testcases/
subdirectory that comes with this tool.

If a large corpus of data is available for screening, you may want to use the
afl-cmin utility to reject redundant files - ideally, with an aggressive
timeout (-t); afl-showmap can be used to manually examine and compare execution
traces, too.

6) Fuzzing binaries
-------------------

The fuzzing process itself is carried out by the afl-fuzz utility. The program
requires a read-only directory with initial test cases, a separate place to
store its findings, plus a path to the binary to test.

For programs that accept input directly from stdin, the usual syntax may be:

$ ./afl-fuzz -i testcase_dir -o findings_dir /path/to/program [...params...]

For programs that take input from a file, use '@@' to mark the location where
the input file name should go. The fuzzer will substitute this for you:

$ ./afl-fuzz -i testcase_dir -o findings_dir /path/to/program -r @@

You can also use the -f option to have the mutated data written to a specific
file. This is useful if the program expects a particular file extension or so.

Non-instrumented binaries can be fuzzed in the QEMU mode by adding -Q in the
command line. It is also possible to use the -n flag to run afl-fuzz in plain
old non-guided mode. This gives you a fairly traditional fuzzer with a couple
of nice testing strategies.

You can use -t and -m to override the default timeout and memory limit for the
executed process; this is seldom necessary, perhaps except for video decoders
or compilers.

Tips for optimizing the performance of the process are discussed in
perf_tips.txt. Note that the fuzzer starts by meticulously performing an array
of deterministic fuzzing steps, which can take several days. If you want more
traditional behavior akin to zzuf or honggfuzz, use the -d option to get quick
but less systematic and less in-depth results right away.

7) Interpreting output
----------------------

The fuzzing process will continue until you press Ctrl-C. See the
status_screen.txt file for information on how to interpret the displayed stats
and monitor the health of the process. At the *very* minimum, you want to allow
the fuzzer to complete one queue cycle, which may take anywhere from a couple
of hours to a week or so.

There are three subdirectories created within the output directory and updated
in real time:

  - queue/   - test cases for every distinctive execution path, plus all the
               starting files given by the user. This is, in effect, the
               synthesized corpus mentioned in section 2.

               If desired, you can use afl-cmin to shrink the corpus to a much
               smaller size. This works by throwing away earlier inputs that
               used to trigger unique behaviors in the past, but have been made
               obsolete by better finds made by afl-fuzz later on.

  - hangs/   - unique test cases that cause the tested program to time out. Note
               that the default timeouts are fairly aggressive (set at 5x the
               average execution time) to keep things moving fast.

  - crashes/ - unique test cases that cause the tested program to receive a
               fatal signal (e.g., SIGSEGV, SIGILL, SIGABRT). The entries are 
               grouped by the received signal.

Crashes and hangs are considered "unique" if the associated execution paths
involve any state transitions not seen in previously-recorded faults. If a
single bug can be reached in multiple ways, there will be some count inflation
early in the process, but this should quickly taper off.

The file names for crashes and hangs should let you correlate them with the
parent, non-faulting queue entries. This should help with debugging.

When you can't reproduce a crash found by afl-fuzz, the most likely cause is
that you are not setting the same memory limit as used by the tool. Try:

$ LIMIT_MB=50
$ ( ulimit -Sv $[LIMIT_MB << 10]; /path/to/tested_binary ... )

Change LIMIT_MB to match the -m parameter passed to afl-fuzz. On OpenBSD,
also change -Sv to -Sd.

Any existing output directory can be also used to resume aborted jobs; try:

$ ./afl-fuzz -i- -o existing_output_dir [...etc...]

If you have gnuplot installed, you can also generate some pretty graphs for any
active fuzzing task using 'afl-plot'. For an example of how this looks like,
see http://lcamtuf.coredump.cx/afl/plot/.

8) Parallelized fuzzing
-----------------------

Every instance of afl-fuzz takes up roughly one core. This means that on
multi-core systems, parallelization is necessary to fully utilize the hardware.
For tips on how to fuzz a common target on multiple cores or multiple networked
machines, please refer to parallel_fuzzing.txt.

9) Fuzzer dictionaries
----------------------

By default, afl-fuzz mutation engine is optimized for compact data formats -
say, images, multimedia, compressed data, regular expression syntax, or shell
scripts. It is somewhat less suited for languages with particularly verbose and
redundant verbiage - notably including HTML, SQL, or JavaScript.

To avoid the hassle of building syntax-aware tools, afl-fuzz provides a way to
seed the fuzzing process with an optional dictionary of language keywords,
magic headers, or other special tokens associated with the targeted data type
- and use that to reconstruct the underlying grammar on the go:

  http://lcamtuf.blogspot.com/2015/01/afl-fuzz-making-up-grammar-with.html

To use this feature, you first need to create a dictionary in one of the two
formats discussed in testcases/README.testcases; and then point the fuzzer to
it via the -x option in the command line.

There is no way to provide more structured descriptions of the underlying
syntax, but the fuzzer will likely figure out some of this based on the
instrumentation feedback alone. This actually works in practice, say:

  http://lcamtuf.blogspot.com/2015/04/finding-bugs-in-sqlite-easy-way.html

PS. Even when no explicit dictionary is given, afl-fuzz will try to extract
existing syntax tokens in the input corpus by watching the instrumentation
very closely during deterministic byte flips. This works for some types of
parsers and grammars, but isn't nearly as good as the -x mode.

10) Crash triage
----------------

The coverage-based grouping of crashes usually produces a small data set that
can be quickly triaged manually or with a very simple GDB or Valgrind script.
Every crash is also traceable to its parent non-crashing test case in the
queue, making it easier to diagnose faults.

Having said that, it's important to acknowledge that some fuzzing crashes can be
difficult quickly evaluate for exploitability without a lot of debugging and
code analysis work. To assist with this task, afl-fuzz supports a very unique
"crash exploration" mode enabled with the -C flag.

In this mode, the fuzzer takes one or more crashing test cases as the input,
and uses its feedback-driven fuzzing strategies to very quickly enumerate all
code paths that can be reached in the program while keeping it in the
crashing state.

Mutations that do not result in a crash are rejected; so are any changes that
do not affect the execution path.

The output is a small corpus of files that can be very rapidly examined to see
what degree of control the attacker has over the faulting address, or whether
it is possible to get past an initial out-of-bounds read - and see what lies
beneath.

Oh, one more thing: for test case minimization, give afl-tmin a try. The tool
can be operated in a very simple way:

$ ./afl-tmin -i test_case -o minimized_result -- /path/to/program [...]

The tool works with crashing and non-crashing test cases alike. In the crash
mode, it will happily accept instrumented and non-instrumented binaries. In the
non-crashing mode, the minimizer relies on standard AFL instrumentation to make
the file simpler without altering the execution path.

The minimizer accepts the -m, -t, -f and @@ syntax in a manner compatible with
afl-fuzz.

11) Common-sense risks
----------------------

Please keep in mind that, similarly to many other computationally-intensive
tasks, fuzzing may put strain on your hardware and on the OS. In particular:

  - Your CPU will run hot and will need adequate cooling. In most cases, if
    cooling is insufficient or stops working properly, CPU speeds will be
    automatically throttled. That said, especially when fuzzing on less
    suitable hardware (laptops, smartphones, etc), it's not entirely impossible
    for something to blow up.

  - Targeted programs may end up erratically grabbing gigabytes of memory or
    filling up disk space with junk files. AFL tries to enforce basic memory
    limits, but can't prevent each and every possible mishap. The bottom line
    is that you shouldn't be fuzzing on systems where the prospect of data loss
    is not an acceptable risk.

  - Fuzzing involves billions of reads and writes to the filesystem. On modern
    systems, this will be usually heavily cached, resulting in fairly modest
    "physical" I/O - but there are many factors that may alter this equation.
    It is your responsibility to monitor for potential trouble; with very heavy
    I/O, the lifespan of many HDDs and SSDs may be reduced.

    A good way to monitor disk I/O on Linux is the 'iostat' command:

    $ iostat -d 3 -x -k [...optional disk ID...]

12) Known limitations & areas for improvement
---------------------------------------------

Here are some of the most important caveats for AFL:

  - AFL detects faults by checking for the first spawned process dying due to
    a signal (SIGSEGV, SIGABRT, etc). Programs that install custom handlers for
    these signals may need to have the relevant code commented out. In the same
    vein, faults in child processed spawned by the fuzzed target may evade
    detection unless you manually add some code to catch that.

  - As with any other brute-force tool, the fuzzer offers limited coverage if
    encryption, checksums, cryptographic signatures, or compression are used to
    wholly wrap the actual data format to be tested.

    To work around this, you can comment out the relevant checks (see
    experimental/libpng_no_checksum/ for inspiration); if this is not possible,
    you can also write a postprocessor, as explained in
    experimental/post_library/.

  - There are some unfortunate trade-offs with ASAN and 64-bit binaries. This
    isn't due to any specific fault of afl-fuzz; see notes_for_asan.txt for
    tips.

  - There is no direct support for fuzzing network services, background
    daemons, or interactive apps that require UI interaction to work. You may
    need to make simple code changes to make them behave in a more traditional
    way. Preeny may offer a relatively simple option, too - see:
    https://github.com/zardus/preeny

  - AFL doesn't output human-readable coverage data. If you want to monitor
    coverage, use afl-cov from Michael Rash: https://github.com/mrash/afl-cov

Beyond this, see INSTALL for platform-specific tips.

13) Special thanks
------------------

Many of the improvements to afl-fuzz wouldn't be possible without feedback,
bug reports, or patches from:

  Jann Horn                             Hanno Boeck
  Felix Groebert                        Jakub Wilk
  Richard W. M. Jones                   Alexander Cherepanov
  Tom Ritter                            Hovik Manucharyan
  Sebastian Roschke                     Eberhard Mattes
  Padraig Brady                         Ben Laurie
  @dronesec                             Luca Barbato
  Tobias Ospelt                         Thomas Jarosch
  Martin Carpenter                      Mudge Zatko
  Joe Zbiciak                           Ryan Govostes
  Michael Rash                          William Robinet
  Jonathan Gray                         Filipe Cabecinhas
  Nico Weber                            Jodie Cunningham
  Andrew Griffiths                      Parker Thompson
  Jonathan Neuschfer                    Tyler Nighswander
  Ben Nagy                              Samir Aguiar
  Aidan Thornton                        Aleksandar Nikolich
  Sam Hakim                             Laszlo Szekeres
  David A. Wheeler                      Turo Lamminen
  Andreas Stieger                       Richard Godbee
  Louis Dassy

Thank you!

14) Contact
-----------

Questions? Concerns? Bug reports? The author can be usually reached at
<lcamtuf@google.com>.

There is also a mailing list for the project; to join, send a mail to
<afl-users+subscribe@googlegroups.com>. Or, if you prefer to browse
archives first, try:

  https://groups.google.com/group/afl-users

PS. If you wish to submit raw code to be incorporated into the project, please
be aware that the copyright on most of AFL is claimed by Google. While you do
retain copyright on your contributions, they do ask people to agree to a simple
CLA first:

  https://cla.developers.google.com/clas

Sorry about the hassle. Of course, no CLA is required for feature requests or
bug reports.