Greybox Fuzzing

In the previous chapter, we have introduced mutation-based fuzzing, a technique that generates fuzz inputs by applying small mutations to given inputs. In this chapter, we show how to guide these mutations towards specific goals such as coverage. The algorithms in this chapter stem from the popular American Fuzzy Lop (AFL) fuzzer, in particular from its AFLFast and AFLGo flavors. We will explore the greybox fuzzing algorithm behind AFL and how we can exploit it to solve various problems for automated vulnerability detection.

from bookutils import YouTubeVideo



To use the code provided in this chapter, write

>>> from fuzzingbook.GreyboxFuzzer import <identifier>

and then make use of the following features.

This chapter introduces advanced methods for grey-box fuzzing inspired by the popular AFL fuzzer. The GreyboxFuzzer class has three arguments. First, a list of seed inputs:

>>> seed_input = ""
>>> seeds = [seed_input]

Second, a mutator that changes individual parts of the input.

>>> mutator = Mutator()

Third, a power schedule that assigns fuzzing effort across the population:

>>> schedule = PowerSchedule()

These three go into the GreyboxFuzzer constructor:

>>> greybox_fuzzer = GreyboxFuzzer(seeds=seeds, mutator=mutator, schedule=schedule)

The GreyboxFuzzer class is used in conjunction with a FunctionCoverageRunner:

>>> http_runner = FunctionCoverageRunner(http_program)
>>> outcomes = greybox_fuzzer.runs(http_runner, trials=10000)

After fuzzing, we can inspect the population:

>>> greybox_fuzzer.population[:20]
 http=/uww>gonlco\m'$sterc( fJ|kGziosg*,
 htp=uwD>god0co\estqrP(ct! Ckz+o{g,
 uww>g@onlo\m'$gs2;terc( fJ|ozixos*,
 uww>g@onlo\m'$g`s;t\erc( fJ\ozxks*,,,
 o#\om'Zswer< fJo{iH[/os*,
 http=uw>gnlNco\mz#$sterc( fJ|kG_ziosg,
 *tpwDgod0o\so5qrrDta(ct! ;Wz/ow,

Besides the simple PowerSchedule, we can have advanced power schedules.

  • AFLFastSchedule assigns high energy to "unusual" paths not taken very often.
  • AFLGoSchedule assigns high energy to paths close to uncovered program locations.

The AFLGoSchedule class constructor requires a distance metric from each node towards target locations, as determined via analysis of the program code. See the chapter for details.

CountingGreyboxFuzzer CountingGreyboxFuzzer run() reset() GreyboxFuzzer GreyboxFuzzer run() reset() CountingGreyboxFuzzer->GreyboxFuzzer AdvancedMutationFuzzer AdvancedMutationFuzzer __init__() fuzz() create_candidate() reset() GreyboxFuzzer->AdvancedMutationFuzzer Fuzzer Fuzzer __init__() fuzz() run() runs() AdvancedMutationFuzzer->Fuzzer AFLFastSchedule AFLFastSchedule __init__() assignEnergy() PowerSchedule PowerSchedule __init__() assignEnergy() choose() normalizedEnergy() AFLFastSchedule->PowerSchedule AFLGoSchedule AFLGoSchedule assignEnergy() DirectedSchedule DirectedSchedule __init__() __getFunctions__() assignEnergy() AFLGoSchedule->DirectedSchedule DirectedSchedule->PowerSchedule DictMutator DictMutator __init__() insert_from_dictionary() Mutator Mutator __init__() delete_random_character() flip_random_character() insert_random_character() mutate() DictMutator->Mutator Seed Seed __init__() __repr__() __str__() Legend Legend •  public_method() •  private_method() •  overloaded_method() Hover over names to see doc

AFL: An Effective Greybox Fuzzer

The algorithms in this chapter stem from the popular American Fuzzy Lop (AFL) fuzzer.

AFL is a mutation-based fuzzer. Meaning, AFL generates new inputs by slightly modifying a seed input (i.e., mutation), or by joining the first half of one input with the second half of another (i.e., splicing).

AFL is also a greybox fuzzer (not blackbox nor whitebox). Meaning, AFL leverages coverage-feedback to learn how to reach deeper into the program. It is not entirely blackbox because AFL leverages at least some program analysis. It is not entirely whitebox either because AFL does not build on heavyweight program analysis or constraint solving. Instead, AFL uses lightweight program instrumentation to glean some information about the (branch) coverage of a generated input. If a generated input increases coverage, it is added to the seed corpus for further fuzzing.

To instrument a program, AFL injects a piece of code right after every conditional jump instruction. When executed, this so-called trampoline assigns the exercised branch a unique identifier and increments a counter that is associated with this branch. For efficiency, only a coarse branch hit count is maintained. In other words, for each input the fuzzer knows which branches and roughly how often they are exercised. The instrumentation is usually done at compile-time, i.e., when the program source code is compiled to an executable binary. However, it is possible to run AFL on non-instrumented binaries using tools such as a virtual machine (e.g., QEMU) or a dynamic instrumentation tool (e.g., Intel PinTool). For Python programs, we can collect coverage information without any instrumentation (see chapter on collecting coverage).

Ingredients for Greybox Fuzzing

We start with discussing the most important parts we need for mutational testing and goal guidance.


We introduce specific classes for mutating a seed.

from typing import List, Set, Any, Tuple, Dict, Union
from import Sequence
import random
from Coverage import population_coverage

First, we'll introduce the Mutator class. Given a seed input inp, the mutator returns a slightly modified version of inp. In the chapter on greybox grammar fuzzing, we extend this class to consider the input grammar for smart greybox fuzzing.

class Mutator:
    """Mutate strings"""

    def __init__(self) -> None:
        self.mutators = [

For insertion, we add a random character in a random position.

class Mutator(Mutator):
    def insert_random_character(self, s: str) -> str:
        """Returns s with a random character inserted"""
        pos = random.randint(0, len(s))
        random_character = chr(random.randrange(32, 127))
        return s[:pos] + random_character + s[pos:]

For deletion, if the string is non-empty choose a random position and delete the character. Otherwise, use the insertion-operation.

class Mutator(Mutator):
    def delete_random_character(self, s: str) -> str:
        """Returns s with a random character deleted"""
        if s == "":
            return self.insert_random_character(s)

        pos = random.randint(0, len(s) - 1)
        return s[:pos] + s[pos + 1:]

For substitution, if the string is non-empty choose a random position and flip a random bit in the character. Otherwise, use the insertion-operation.

class Mutator(Mutator):
    def flip_random_character(self, s: str) -> str:
        """Returns s with a random bit flipped in a random position"""
        if s == "":
            return self.insert_random_character(s)

        pos = random.randint(0, len(s) - 1)
        c = s[pos]
        bit = 1 << random.randint(0, 6)
        new_c = chr(ord(c) ^ bit)
        return s[:pos] + new_c + s[pos + 1:]

The main method is mutate which chooses a random mutation operator from the list of operators.

class Mutator(Mutator):
    def mutate(self, inp: Any) -> Any:  # can be str or Seed (see below)
        """Return s with a random mutation applied. Can be overloaded in subclasses."""
        mutator = random.choice(self.mutators)
        return mutator(inp)

Let's try the mutator. You can actually interact with such a "cell" and try other inputs by loading this chapter as Jupyter notebook. After opening, run all cells in the notebook using "Kernel -> Restart & Run All".


Seeds and Power Schedules

Now we introduce a new concept; the power schedule. A power schedule distributes the precious fuzzing time among the seeds in the population. Our objective is to maximize the time spent fuzzing those (most progressive) seeds which lead to higher coverage increase in shorter time.

We call the likelihood with which a seed is chosen from the population as the seed's energy. Throughout a fuzzing campaign, we would like to prioritize seeds that are more promising. Simply said, we do not want to waste energy fuzzing non-progressive seeds. We call the procedure that decides a seed's energy as the fuzzer's power schedule. For instance, AFL's schedule assigns more energy to seeds that are shorter, that execute faster, and yield coverage increases more often.

First, there is some information that we need to attach to each seed in addition to the seed's data. Hence, we define the following Seed class.

from Coverage import Location
class Seed:
    """Represent an input with additional attributes"""

    def __init__(self, data: str) -> None:
        """Initialize from seed data""" = data

        # These will be needed for advanced power schedules
        self.coverage: Set[Location] = set()
        self.distance: Union[int, float] = -1 = 0.0

    def __str__(self) -> str:
        """Returns data as string representation of the seed"""

    __repr__ = __str__

The power schedule that is implemented below assigns each seed the same energy. Once a seed is in the population, it will be fuzzed as often as any other seed in the population.

In Python, we can squeeze long for-loops into much smaller statements.

  • lambda x: ... returns a function that takes x as input. Lambda allows for quick definitions unnamed functions.
  • map(f, l) returns a list where the function f is applied to each element in list l.
  • random.choices(l, weights)[0] returns element l[i] with probability in weights[i].
class PowerSchedule:
    """Define how fuzzing time should be distributed across the population."""

    def __init__(self) -> None:
        self.path_frequency: Dict = {}

    def assignEnergy(self, population: Sequence[Seed]) -> None:
        """Assigns each seed the same energy"""
        for seed in population:
   = 1

    def normalizedEnergy(self, population: Sequence[Seed]) -> List[float]:
        """Normalize energy"""
        energy = list(map(lambda seed:, population))
        sum_energy = sum(energy)  # Add up all values in energy
        assert sum_energy != 0
        norm_energy = list(map(lambda nrg: nrg / sum_energy, energy))
        return norm_energy

    def choose(self, population: Sequence[Seed]) -> Seed:
        """Choose weighted by normalized energy."""
        norm_energy = self.normalizedEnergy(population)
        seed: Seed = random.choices(population, weights=norm_energy)[0]
        return seed

Let's see whether this power schedule chooses seeds uniformly at random. We ask the schedule 10k times to choose a seed from the population of three seeds (A, B, C) and keep track of the number of times we have seen each seed. We should see each seed about 3.3k times.

population = [Seed("A"), Seed("B"), Seed("C")]
schedule = PowerSchedule()
hits = {
    "A": 0,
    "B": 0,
    "C": 0
for i in range(10000):
    seed = schedule.choose(population)
    hits[] += 1
{'A': 3387, 'B': 3255, 'C': 3358}

Looks good. Every seed has been chosen about a third of the time.

Runners and a Sample Program

We'll start with a small sample program of six lines. In order to collect coverage information during execution, we import the FunctionCoverageRunner class from the chapter on mutation-based fuzzing.

The FunctionCoverageRunner constructor takes a Python function to execute. The function run takes an input, passes it on to the Python function, and collects the coverage information for this execution. The function coverage() returns a list of tuples (function name, line number) for each statement that has been covered in the Python function.

from MutationFuzzer import FunctionCoverageRunner, http_program

The crashme() function raises an exception for the input "bad!". Let's see which statements are covered for the input "good".

def crashme(s: str) -> None:
    if len(s) > 0 and s[0] == 'b':
        if len(s) > 1 and s[1] == 'a':
            if len(s) > 2 and s[2] == 'd':
                if len(s) > 3 and s[3] == '!':
                    raise Exception()
crashme_runner = FunctionCoverageRunner(crashme)"good")
[('crashme', 2), ('run_function', 132)]

In crashme, the input "good" only covers the if-statement in line 2. The branch condition len(s) > 0 and s[0] == 'b' evaluates to False.

Advanced Blackbox Mutation-based Fuzzing

Let's integrate both the mutator and power schedule into a fuzzer. We'll start with a blackbox fuzzer -- which does not leverage any coverage information.

Our AdvancedMutationFuzzer class is an advanced and parameterized version of the MutationFuzzer class from the chapter on mutation-based fuzzing. It also inherits from the Fuzzer class. For now, we only need to know the functions fuzz() which returns a generated input and runs() which executes fuzz() a specified number of times. For our AdvancedMutationFuzzer class, we override the function fuzz().

from Fuzzer import Fuzzer

The AdvancedMutationFuzzer is constructed with a set of initial seeds, a mutator, and a power schedule. Throughout the fuzzing campaign, it maintains a seed corpus called population. The function fuzz returns either an unfuzzed seed from the initial seeds, or the result of fuzzing a seed in the population. The function create_candidate handles the latter. It randomly chooses an input from the population and applies a number of mutations.

class AdvancedMutationFuzzer(Fuzzer):
    """Base class for mutation-based fuzzing."""

    def __init__(self, seeds: List[str],
                 mutator: Mutator,
                 schedule: PowerSchedule) -> None:
        `seeds` - a list of (input) strings to mutate.
        `mutator` - the mutator to apply.
        `schedule` - the power schedule to apply.
        self.seeds = seeds
        self.mutator = mutator
        self.schedule = schedule
        self.inputs: List[str] = []

    def reset(self) -> None:
        """Reset the initial population and seed index"""
        self.population = list(map(lambda x: Seed(x), self.seeds))
        self.seed_index = 0

    def create_candidate(self) -> str:
        """Returns an input generated by fuzzing a seed in the population"""
        seed = self.schedule.choose(self.population)

        # Stacking: Apply multiple mutations to generate the candidate
        candidate =
        trials = min(len(candidate), 1 << random.randint(1, 5))
        for i in range(trials):
            candidate = self.mutator.mutate(candidate)
        return candidate

    def fuzz(self) -> str:
        """Returns first each seed once and then generates new inputs"""
        if self.seed_index < len(self.seeds):
            # Still seeding
            self.inp = self.seeds[self.seed_index]
            self.seed_index += 1
            # Mutating
            self.inp = self.create_candidate()

        return self.inp

Okay, let's take the mutation fuzzer for a spin. Given a single seed, we ask it to generate three inputs.

seed_input = "good"
mutation_fuzzer = AdvancedMutationFuzzer([seed_input], Mutator(), PowerSchedule())

Let's see how many statements the mutation-based blackbox fuzzer covers in a campaign with n=30k inputs.

The fuzzer function runs(crashme_runner, trials=n) generates n inputs and executes them on the crashme function via the crashme_runner. As stated earlier, the crashme_runner also collects coverage information.

import time
n = 30000
blackbox_fuzzer = AdvancedMutationFuzzer([seed_input], Mutator(), PowerSchedule())

start = time.time()
blackbox_fuzzer.runs(FunctionCoverageRunner(crashme), trials=n)
end = time.time()

"It took the blackbox mutation-based fuzzer %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the blackbox mutation-based fuzzer 0.34 seconds to generate and execute 30000 inputs.'

In order to measure coverage, we import the population_coverage function. It takes a set of inputs and a Python function, executes the inputs on that function and collects coverage information. Specifically, it returns a tuple (all_coverage, cumulative_coverage) where all_coverage is the set of statements covered by all inputs, and cumulative_coverage is the number of statements covered as the number of executed inputs increases. We are just interested in the latter to plot coverage over time.

We extract the generated inputs from the blackbox fuzzer and measure coverage as the number of inputs increases.

_, blackbox_coverage = population_coverage(blackbox_fuzzer.inputs, crashme)
bb_max_coverage = max(blackbox_coverage)

"The blackbox mutation-based fuzzer achieved a maximum coverage of %d statements." % bb_max_coverage
'The blackbox mutation-based fuzzer achieved a maximum coverage of 2 statements.'

The following generated inputs increased the coverage for our crashme example.

[seed_input] + \
        blackbox_fuzzer.inputs[idx] for idx in range(len(blackbox_coverage))
        if blackbox_coverage[idx] > blackbox_coverage[idx - 1]
['good', 'bo']

Summary. This is how a blackbox mutation-based fuzzer works. We have integrated the mutator to generate inputs by fuzzing a provided set of initial seeds and the power schedule to decide which seed to choose next.

Greybox Mutation-based Fuzzing

In contrast to a blackbox fuzzer, a greybox fuzzer like AFL does leverage coverage information. Specifically, a greybox fuzzer adds to the seed population generated inputs which increase code coverage.

The method run() is inherited from the Fuzzer class. It is called to generate and execute exactly one input. We override this function to add an input to the population that increases coverage. The greybox fuzzer attribute coverages_seen maintains the set of statements, that have previously been covered.

class GreyboxFuzzer(AdvancedMutationFuzzer):
    """Coverage-guided mutational fuzzing."""

    def reset(self):
        """Reset the initial population, seed index, coverage information"""
        self.coverages_seen = set()
        self.population = []  # population is filled during greybox fuzzing

    def run(self, runner: FunctionCoverageRunner) -> Tuple[Any, str]:
        """Run function(inp) while tracking coverage.
           If we reach new coverage,
           add inp to population and its coverage to population_coverage
        result, outcome = super().run(runner)
        new_coverage = frozenset(runner.coverage())
        if new_coverage not in self.coverages_seen:
            # We have new coverage
            seed = Seed(self.inp)
            seed.coverage = runner.coverage()

        return (result, outcome)

Let's take our greybox fuzzer for a spin.

seed_input = "good"
greybox_fuzzer = GreyboxFuzzer([seed_input], Mutator(), PowerSchedule())

start = time.time()
greybox_fuzzer.runs(FunctionCoverageRunner(crashme), trials=n)
end = time.time()

"It took the greybox mutation-based fuzzer %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the greybox mutation-based fuzzer 0.35 seconds to generate and execute 30000 inputs.'

Does the greybox fuzzer cover more statements after generating the same number of test inputs?

_, greybox_coverage = population_coverage(greybox_fuzzer.inputs, crashme)
gb_max_coverage = max(greybox_coverage)

"Our greybox mutation-based fuzzer covers %d more statements" % (gb_max_coverage - bb_max_coverage)
'Our greybox mutation-based fuzzer covers 2 more statements'

Our seed population for our example now contains the following seeds.

[good, bo, baof, bad4u]

Coverage-feedback is indeed helpful. The new seeds are like bread crumbs or milestones that guide the fuzzer to progress more quickly into deeper code regions. Following is a simple plot showing the coverage achieved over time for both fuzzers on our simple example.

%matplotlib inline
import matplotlib.pyplot as plt
line_bb, = plt.plot(blackbox_coverage, label="Blackbox")
line_gb, = plt.plot(greybox_coverage, label="Greybox")
plt.legend(handles=[line_bb, line_gb])
plt.title('Coverage over time')
plt.xlabel('# of inputs')
plt.ylabel('lines covered');

Summary. We have seen how a greybox fuzzer "discovers" interesting seeds that can lead to more progress. From the input good, our greybox fuzzer has slowly learned how to generate the input bad! which raises the exception. Now, how can we do that even faster?

Try it. How much coverage would be achieved over time using a blackbox generation-based fuzzer? Try plotting the coverage for all three fuzzers. You can define the blackbox generation-based fuzzer as follows.

from Fuzzer import RandomFuzzer
blackbox_gen_fuzzer = RandomFuzzer(min_length=4, max_length=4, char_start=32, char_range=96)

You can execute your own code by opening this chapter as Jupyter notebook.

Read. This is the high-level view how AFL works, one of the most successful vulnerability detection tools. If you are interested in the technical details, have a look at:

Boosted Greybox Fuzzing

Our boosted greybox fuzzer assigns more energy to seeds that promise to achieve more coverage. We change the power schedule such that seeds that exercise "unusual" paths have more energy. With unusual paths, we mean paths that are not exercised very often by generated inputs.

In order to identify which path is exercised by an input, we leverage the function getPathID from the section on trace coverage.

import pickle   # serializes an object by producing a byte array from all the information in the object
import hashlib  # produces a 128-bit hash value from a byte array

The function getPathID returns a unique hash for a coverage set.

def getPathID(coverage: Any) -> str:
    """Returns a unique hash for the covered statements"""
    pickled = pickle.dumps(sorted(coverage))
    return hashlib.md5(pickled).hexdigest()

There are several ways to assign energy based on how unusual the exercised path is. In this case, we implement an exponential power schedule which computes the energy $e(s)$ for a seed $s$ as follows $$e(s) = \frac{1}{f(p(s))^a}$$ where

  • $p(s)$ returns the ID of the path exercised by $s$,
  • $f(p)$ returns the number of times the path $p$ is exercised by generated inputs, and
  • $a$ is a given exponent.
class AFLFastSchedule(PowerSchedule):
    """Exponential power schedule as implemented in AFL"""

    def __init__(self, exponent: float) -> None:
        self.exponent = exponent

    def assignEnergy(self, population: Sequence[Seed]) -> None:
        """Assign exponential energy inversely proportional to path frequency"""
        for seed in population:
   = 1 / (self.path_frequency[getPathID(seed.coverage)] ** self.exponent)

In the greybox fuzzer, let's keep track of the number of times $f(p)$ each path $p$ is exercised, and update the power schedule.

class CountingGreyboxFuzzer(GreyboxFuzzer):
    """Count how often individual paths are exercised."""

    def reset(self):
        """Reset path frequency"""
        self.schedule.path_frequency = {}

    def run(self, runner: FunctionCoverageRunner) -> Tuple[Any, str]:
        """Inform scheduler about path frequency"""
        result, outcome = super().run(runner)

        path_id = getPathID(runner.coverage())
        if path_id not in self.schedule.path_frequency:
            self.schedule.path_frequency[path_id] = 1
            self.schedule.path_frequency[path_id] += 1

        return(result, outcome)

Okay, let's run our boosted greybox fuzzer $n=10k$ times on our simple example. We set the exponentent of our exponential power schedule to $a=5$.

n = 10000
seed_input = "good"
fast_schedule = AFLFastSchedule(5)
fast_fuzzer = CountingGreyboxFuzzer([seed_input], Mutator(), fast_schedule)
start = time.time()
fast_fuzzer.runs(FunctionCoverageRunner(crashme), trials=n)
end = time.time()

"It took the fuzzer w/ exponential schedule %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the fuzzer w/ exponential schedule 0.24 seconds to generate and execute 10000 inputs.'
import numpy as np
x_axis = np.arange(len(fast_schedule.path_frequency))
y_axis = list(fast_schedule.path_frequency.values()), y_axis)
plt.ylim(0, n)
# plt.yscale("log")
# plt.yticks([10,100,1000,10000])
print("             path id 'p'           : path frequency 'f(p)'")
             path id 'p'           : path frequency 'f(p)'
{'e014b68ad4f3bc2daf207e2498d14cbf': 5612,
 '0a1008773804033d8a4c0e3aba4b96a0': 2607,
 'eae4df5b039511eac56625f47c337d24': 1105,
 'b14f545c3b39716a455034d9a0c61b8c': 457,
 '11529f85aaa30be08110f3076748e420': 219}

How does it compare to our greybox fuzzer with the classical power schedule?

seed_input = "good"
orig_schedule = PowerSchedule()
orig_fuzzer = CountingGreyboxFuzzer([seed_input], Mutator(), orig_schedule)
start = time.time()
orig_fuzzer.runs(FunctionCoverageRunner(crashme), trials=n)
end = time.time()

"It took the fuzzer w/ original schedule %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the fuzzer w/ original schedule 0.16 seconds to generate and execute 10000 inputs.'
x_axis = np.arange(len(orig_schedule.path_frequency))
y_axis = list(orig_schedule.path_frequency.values()), y_axis)
plt.ylim(0, n)
# plt.yscale("log")
# plt.yticks([10,100,1000,10000])
print("             path id 'p'           : path frequency 'f(p)'")
             path id 'p'           : path frequency 'f(p)'
{'e014b68ad4f3bc2daf207e2498d14cbf': 6581,
 '0a1008773804033d8a4c0e3aba4b96a0': 2379,
 'eae4df5b039511eac56625f47c337d24': 737,
 'b14f545c3b39716a455034d9a0c61b8c': 241,
 '11529f85aaa30be08110f3076748e420': 62}

The exponential power schedule shaves some of the executions of the "high-frequency path" off and adds them to the lower-frequency paths. The path executed least often is either not at all exercised using the traditional power schedule or it is exercised much less often.

Let's have a look at the energy that is assigned to the discovered seeds.

orig_energy = orig_schedule.normalizedEnergy(orig_fuzzer.population)

for (seed, norm_energy) in zip(orig_fuzzer.population, orig_energy):
    print("'%s', %0.5f, %s" % (getPathID(seed.coverage),
                               norm_energy, repr(
'e014b68ad4f3bc2daf207e2498d14cbf', 0.20000, 'good'
'0a1008773804033d8a4c0e3aba4b96a0', 0.20000, 'bgI/d'
'eae4df5b039511eac56625f47c337d24', 0.20000, 'baI/dt'
'b14f545c3b39716a455034d9a0c61b8c', 0.20000, 'badtuS'
'11529f85aaa30be08110f3076748e420', 0.20000, 'bad!`tuS'
fast_energy = fast_schedule.normalizedEnergy(fast_fuzzer.population)

for (seed, norm_energy) in zip(fast_fuzzer.population, fast_energy):
    print("'%s', %0.5f, %s" % (getPathID(seed.coverage),
                               norm_energy, repr(
'e014b68ad4f3bc2daf207e2498d14cbf', 0.00000, 'good'
'0a1008773804033d8a4c0e3aba4b96a0', 0.00000, 'bnd'
'eae4df5b039511eac56625f47c337d24', 0.00030, 'ba.'
'b14f545c3b39716a455034d9a0c61b8c', 0.02464, 'bad.'
'11529f85aaa30be08110f3076748e420', 0.97506, 'bad!\\.'

Exactly. Our new exponential power schedule assigns most energy to the seed exercising the lowest-frequency path.

Let's compare them in terms of coverage achieved over time for our simple example.

_, orig_coverage = population_coverage(orig_fuzzer.inputs, crashme)
_, fast_coverage = population_coverage(fast_fuzzer.inputs, crashme)
line_orig, = plt.plot(orig_coverage, label="Original Greybox Fuzzer")
line_fast, = plt.plot(fast_coverage, label="Boosted Greybox Fuzzer")
plt.legend(handles=[line_orig, line_fast])
plt.title('Coverage over time')
plt.xlabel('# of inputs')
plt.ylabel('lines covered');

As expected, the boosted greybox fuzzer (with the exponential power schedule) achieves coverage much faster.

Summary. By fuzzing seeds more often that exercise low-frequency paths, we can explore program paths in a much more efficient manner.

Try it. You can try other exponents for the fast power schedule, or change the power schedule entirely. Note that a large exponent can lead to overflows and imprecisions in the floating point arithmetic producing unexpected results. You can execute your own code by opening this chapter as Jupyter notebook.

Read. You can find out more about fuzzer boosting in the paper "Coverage-based Greybox Fuzzing as Markov Chain" [Böhme et al, 2018] and check out the implementation into AFL at [].

A Complex Example: HTMLParser

Let's compare the three fuzzers on a more realistic example, the Python HTML parser. We run all three fuzzers $n=5k$ times on the HTMLParser, starting with the "empty" seed.

from html.parser import HTMLParser
# create wrapper function
def my_parser(inp: str) -> None:
    parser = HTMLParser()  # resets the HTMLParser object for every fuzz input
n = 5000
seed_input = " "  # empty seed
blackbox_fuzzer = AdvancedMutationFuzzer([seed_input], Mutator(), PowerSchedule())
greybox_fuzzer = GreyboxFuzzer([seed_input], Mutator(), PowerSchedule())
boosted_fuzzer = CountingGreyboxFuzzer([seed_input], Mutator(), AFLFastSchedule(5))
start = time.time()
blackbox_fuzzer.runs(FunctionCoverageRunner(my_parser), trials=n)
greybox_fuzzer.runs(FunctionCoverageRunner(my_parser), trials=n)
boosted_fuzzer.runs(FunctionCoverageRunner(my_parser), trials=n)
end = time.time()

"It took all three fuzzers %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took all three fuzzers 6.48 seconds to generate and execute 5000 inputs.'

How do the fuzzers compare in terms of coverage over time?

_, black_coverage = population_coverage(blackbox_fuzzer.inputs, my_parser)
_, grey_coverage = population_coverage(greybox_fuzzer.inputs, my_parser)
_, boost_coverage = population_coverage(boosted_fuzzer.inputs, my_parser)
line_black, = plt.plot(black_coverage, label="Blackbox Fuzzer")
line_grey, = plt.plot(grey_coverage, label="Greybox Fuzzer")
line_boost, = plt.plot(boost_coverage, label="Boosted Greybox Fuzzer")
plt.legend(handles=[line_boost, line_grey, line_black])
plt.title('Coverage over time')
plt.xlabel('# of inputs')
plt.ylabel('lines covered');

Both greybox fuzzers clearly outperform the blackbox fuzzer. The reason is that the greybox fuzzer "discovers" interesting inputs along the way. Let's have a look at the last 10 inputs generated by the greybox versus blackbox fuzzer.

[' H', '', '', '`', ' i', '', '(', 'j ', '', '0']
 'a]<;<PCb/Xo\n PHE>NM<9`',

The greybox fuzzer executes much more complicated inputs, many of which include special characters such as opening and closing brackets and chevrons (i.e., <, >, [, ]). Yet, many important keywords, such as <html> are still missing.

To inform the fuzzer about these important keywords, we will need grammars; in the section on smart greybox fuzzing, we combine them with the techniques above.

Try it. You can re-run these experiments to understand the variance of fuzzing experiments. Sometimes, the fuzzer that we claim to be superior does not seem to outperform the inferior fuzzer. In order to do this, you just need to open this chapter as Jupyter notebook.

Directed Greybox Fuzzing

Sometimes, you just want the fuzzer to reach some dangerous location in the source code. This could be a location where you expect a buffer overflow. Or you want to test a recent change in your code base. How do we direct the fuzzer towards these locations?

In this chapter, we introduce directed greybox fuzzing as an optimization problem.

Solving the Maze

To provide a meaningful example where you can easily change the code complexity and target location, we generate the maze source code from the maze provided as string. This example is loosely based on an old blog post on symbolic execution by Felipe Andres Manzano (Quick shout-out!).

You simply specify the maze as a string. Like so.

maze_string = """
|X|     |
| | --+ |
| |   | |
| +-- | |
|     |#|

The code is generated using the function generate_maze_code(). We'll hide the implementation and instead explain what it does. If you are interested in the coding, go here.

from ControlFlow import generate_maze_code
maze_code = generate_maze_code(maze_string)

The objective is to get the "X" to the "#" by providing inputs D for down, U for up, L for left, and R for right.

print(maze("DDDDRRRRUULLUURRRRDDDD"))  # Appending one more 'D', you have reached the target.

| |     |
| | --+ |
| |   | |
| +-- | |
|     |X|

Each character in maze_string represents a tile. For each tile, a tile-function is generated.

  • If the current tile is "benign" (), the tile-function corresponding to the next input character (D, U, L, R) is called. Unexpected input characters are ignored. If no more input characters are left, it returns "VALID" and the current maze state.
  • If the current tile is a "trap" (+,|,-), it returns "INVALID" and the current maze state.
  • If the current tile is the "target" (#), it returns "SOLVED" and the current maze state.

Try it. You can test other sequences of input characters, or even change the maze entirely. In order to execute your own code, you just need to open this chapter as Jupyter notebook.

To get an idea of the generated code, lets look at the static call graph. A call graph shows the order in which functions can be executed.

from ControlFlow import callgraph
G cluster_G cluster_callgraphX callgraph callgraphX callgraph callgraphX__maze maze ( callgraphX->callgraphX__maze callgraphX__print_maze print_maze ( callgraphX->callgraphX__print_maze callgraphX__target_tile target_tile ( callgraphX->callgraphX__target_tile callgraphX__tile_1_0 tile_1_0 ( callgraphX->callgraphX__tile_1_0 callgraphX__tile_1_1 tile_1_1 ( callgraphX->callgraphX__tile_1_1 callgraphX__tile_1_2 tile_1_2 ( callgraphX->callgraphX__tile_1_2 callgraphX__tile_1_3 tile_1_3 ( callgraphX->callgraphX__tile_1_3 callgraphX__tile_1_4 tile_1_4 ( callgraphX->callgraphX__tile_1_4 callgraphX__tile_1_5 tile_1_5 ( callgraphX->callgraphX__tile_1_5 callgraphX__tile_1_6 tile_1_6 ( callgraphX->callgraphX__tile_1_6 callgraphX__tile_1_7 tile_1_7 ( callgraphX->callgraphX__tile_1_7 callgraphX__tile_1_8 tile_1_8 ( callgraphX->callgraphX__tile_1_8 callgraphX__tile_2_0 tile_2_0 ( callgraphX->callgraphX__tile_2_0 callgraphX__tile_2_1 tile_2_1 ( callgraphX->callgraphX__tile_2_1 callgraphX__tile_2_2 tile_2_2 ( callgraphX->callgraphX__tile_2_2 callgraphX__tile_2_3 tile_2_3 ( callgraphX->callgraphX__tile_2_3 callgraphX__tile_2_4 tile_2_4 ( callgraphX->callgraphX__tile_2_4 callgraphX__tile_2_5 tile_2_5 ( callgraphX->callgraphX__tile_2_5 callgraphX__tile_2_6 tile_2_6 ( callgraphX->callgraphX__tile_2_6 callgraphX__tile_2_7 tile_2_7 ( callgraphX->callgraphX__tile_2_7 callgraphX__tile_2_8 tile_2_8 ( callgraphX->callgraphX__tile_2_8 callgraphX__tile_3_0 tile_3_0 ( callgraphX->callgraphX__tile_3_0 callgraphX__tile_3_1 tile_3_1 ( callgraphX->callgraphX__tile_3_1 callgraphX__tile_3_2 tile_3_2 ( callgraphX->callgraphX__tile_3_2 callgraphX__tile_3_3 tile_3_3 ( callgraphX->callgraphX__tile_3_3 callgraphX__tile_3_4 tile_3_4 ( callgraphX->callgraphX__tile_3_4 callgraphX__tile_3_5 tile_3_5 ( callgraphX->callgraphX__tile_3_5 callgraphX__tile_3_6 tile_3_6 ( callgraphX->callgraphX__tile_3_6 callgraphX__tile_3_7 tile_3_7 ( callgraphX->callgraphX__tile_3_7 callgraphX__tile_3_8 tile_3_8 ( callgraphX->callgraphX__tile_3_8 callgraphX__tile_4_0 tile_4_0 ( callgraphX->callgraphX__tile_4_0 callgraphX__tile_4_1 tile_4_1 ( callgraphX->callgraphX__tile_4_1 callgraphX__tile_4_2 tile_4_2 ( callgraphX->callgraphX__tile_4_2 callgraphX__tile_4_3 tile_4_3 ( callgraphX->callgraphX__tile_4_3 callgraphX__tile_4_4 tile_4_4 ( callgraphX->callgraphX__tile_4_4 callgraphX__tile_4_5 tile_4_5 ( callgraphX->callgraphX__tile_4_5 callgraphX__tile_4_6 tile_4_6 ( callgraphX->callgraphX__tile_4_6 callgraphX__tile_4_7 tile_4_7 ( callgraphX->callgraphX__tile_4_7 callgraphX__tile_4_8 tile_4_8 ( callgraphX->callgraphX__tile_4_8 callgraphX__tile_5_0 tile_5_0 ( callgraphX->callgraphX__tile_5_0 callgraphX__tile_5_1 tile_5_1 ( callgraphX->callgraphX__tile_5_1 callgraphX__tile_5_2 tile_5_2 ( callgraphX->callgraphX__tile_5_2 callgraphX__tile_5_3 tile_5_3 ( callgraphX->callgraphX__tile_5_3 callgraphX__tile_5_4 tile_5_4 ( callgraphX->callgraphX__tile_5_4 callgraphX__tile_5_5 tile_5_5 ( callgraphX->callgraphX__tile_5_5 callgraphX__tile_5_6 tile_5_6 ( callgraphX->callgraphX__tile_5_6 callgraphX__tile_5_7 tile_5_7 ( callgraphX->callgraphX__tile_5_7 callgraphX__tile_5_8 tile_5_8 ( callgraphX->callgraphX__tile_5_8 callgraphX__tile_6_0 tile_6_0 ( callgraphX->callgraphX__tile_6_0 callgraphX__tile_6_1 tile_6_1 ( callgraphX->callgraphX__tile_6_1 callgraphX__tile_6_2 tile_6_2 ( callgraphX->callgraphX__tile_6_2 callgraphX__tile_6_3 tile_6_3 ( callgraphX->callgraphX__tile_6_3 callgraphX__tile_6_4 tile_6_4 ( callgraphX->callgraphX__tile_6_4 callgraphX__tile_6_5 tile_6_5 ( callgraphX->callgraphX__tile_6_5 callgraphX__tile_6_6 tile_6_6 ( callgraphX->callgraphX__tile_6_6 callgraphX__tile_6_7 tile_6_7 ( callgraphX->callgraphX__tile_6_7 callgraphX__tile_6_8 tile_6_8 ( callgraphX->callgraphX__tile_6_8 callgraphX__tile_7_0 tile_7_0 ( callgraphX->callgraphX__tile_7_0 callgraphX__tile_7_1 tile_7_1 ( callgraphX->callgraphX__tile_7_1 callgraphX__tile_7_2 tile_7_2 ( callgraphX->callgraphX__tile_7_2 callgraphX__tile_7_3 tile_7_3 ( callgraphX->callgraphX__tile_7_3 callgraphX__tile_7_4 tile_7_4 ( callgraphX->callgraphX__tile_7_4 callgraphX__tile_7_5 tile_7_5 ( callgraphX->callgraphX__tile_7_5 callgraphX__tile_7_6 tile_7_6 ( callgraphX->callgraphX__tile_7_6 callgraphX__tile_7_7 tile_7_7 ( callgraphX->callgraphX__tile_7_7 callgraphX__tile_7_8 tile_7_8 ( callgraphX->callgraphX__tile_7_8 callgraphX__maze->callgraphX__tile_2_1 callgraphX__tile_1_0->callgraphX__print_maze callgraphX__tile_1_1->callgraphX__print_maze callgraphX__tile_1_2->callgraphX__print_maze callgraphX__tile_1_3->callgraphX__print_maze callgraphX__tile_1_4->callgraphX__print_maze callgraphX__tile_1_5->callgraphX__print_maze callgraphX__tile_1_6->callgraphX__print_maze callgraphX__tile_1_7->callgraphX__print_maze callgraphX__tile_1_8->callgraphX__print_maze callgraphX__tile_2_0->callgraphX__print_maze callgraphX__tile_2_1->callgraphX__print_maze callgraphX__tile_2_1->callgraphX__tile_1_1 callgraphX__tile_2_1->callgraphX__tile_2_0 callgraphX__tile_2_1->callgraphX__tile_2_1 callgraphX__tile_2_1->callgraphX__tile_2_2 callgraphX__tile_2_1->callgraphX__tile_3_1 callgraphX__tile_2_2->callgraphX__print_maze callgraphX__tile_2_3->callgraphX__print_maze callgraphX__tile_2_3->callgraphX__tile_1_3 callgraphX__tile_2_3->callgraphX__tile_2_2 callgraphX__tile_2_3->callgraphX__tile_2_3 callgraphX__tile_2_3->callgraphX__tile_2_4 callgraphX__tile_2_3->callgraphX__tile_3_3 callgraphX__tile_2_4->callgraphX__print_maze callgraphX__tile_2_4->callgraphX__tile_1_4 callgraphX__tile_2_4->callgraphX__tile_2_3 callgraphX__tile_2_4->callgraphX__tile_2_4 callgraphX__tile_2_4->callgraphX__tile_2_5 callgraphX__tile_2_4->callgraphX__tile_3_4 callgraphX__tile_2_5->callgraphX__print_maze callgraphX__tile_2_5->callgraphX__tile_1_5 callgraphX__tile_2_5->callgraphX__tile_2_4 callgraphX__tile_2_5->callgraphX__tile_2_5 callgraphX__tile_2_5->callgraphX__tile_2_6 callgraphX__tile_2_5->callgraphX__tile_3_5 callgraphX__tile_2_6->callgraphX__print_maze callgraphX__tile_2_6->callgraphX__tile_1_6 callgraphX__tile_2_6->callgraphX__tile_2_5 callgraphX__tile_2_6->callgraphX__tile_2_6 callgraphX__tile_2_6->callgraphX__tile_2_7 callgraphX__tile_2_6->callgraphX__tile_3_6 callgraphX__tile_2_7->callgraphX__print_maze callgraphX__tile_2_7->callgraphX__tile_1_7 callgraphX__tile_2_7->callgraphX__tile_2_6 callgraphX__tile_2_7->callgraphX__tile_2_7 callgraphX__tile_2_7->callgraphX__tile_2_8 callgraphX__tile_2_7->callgraphX__tile_3_7 callgraphX__tile_2_8->callgraphX__print_maze callgraphX__tile_3_0->callgraphX__print_maze callgraphX__tile_3_1->callgraphX__print_maze callgraphX__tile_3_1->callgraphX__tile_2_1 callgraphX__tile_3_1->callgraphX__tile_3_0 callgraphX__tile_3_1->callgraphX__tile_3_1 callgraphX__tile_3_1->callgraphX__tile_3_2 callgraphX__tile_3_1->callgraphX__tile_4_1 callgraphX__tile_3_2->callgraphX__print_maze callgraphX__tile_3_3->callgraphX__print_maze callgraphX__tile_3_3->callgraphX__tile_2_3 callgraphX__tile_3_3->callgraphX__tile_3_2 callgraphX__tile_3_3->callgraphX__tile_3_3 callgraphX__tile_3_3->callgraphX__tile_3_4 callgraphX__tile_3_3->callgraphX__tile_4_3 callgraphX__tile_3_4->callgraphX__print_maze callgraphX__tile_3_5->callgraphX__print_maze callgraphX__tile_3_6->callgraphX__print_maze callgraphX__tile_3_7->callgraphX__print_maze callgraphX__tile_3_7->callgraphX__tile_2_7 callgraphX__tile_3_7->callgraphX__tile_3_6 callgraphX__tile_3_7->callgraphX__tile_3_7 callgraphX__tile_3_7->callgraphX__tile_3_8 callgraphX__tile_3_7->callgraphX__tile_4_7 callgraphX__tile_3_8->callgraphX__print_maze callgraphX__tile_4_0->callgraphX__print_maze callgraphX__tile_4_1->callgraphX__print_maze callgraphX__tile_4_1->callgraphX__tile_3_1 callgraphX__tile_4_1->callgraphX__tile_4_0 callgraphX__tile_4_1->callgraphX__tile_4_1 callgraphX__tile_4_1->callgraphX__tile_4_2 callgraphX__tile_4_1->callgraphX__tile_5_1 callgraphX__tile_4_2->callgraphX__print_maze callgraphX__tile_4_3->callgraphX__print_maze callgraphX__tile_4_3->callgraphX__tile_3_3 callgraphX__tile_4_3->callgraphX__tile_4_2 callgraphX__tile_4_3->callgraphX__tile_4_3 callgraphX__tile_4_3->callgraphX__tile_4_4 callgraphX__tile_4_3->callgraphX__tile_5_3 callgraphX__tile_4_4->callgraphX__print_maze callgraphX__tile_4_4->callgraphX__tile_3_4 callgraphX__tile_4_4->callgraphX__tile_4_3 callgraphX__tile_4_4->callgraphX__tile_4_4 callgraphX__tile_4_4->callgraphX__tile_4_5 callgraphX__tile_4_4->callgraphX__tile_5_4 callgraphX__tile_4_5->callgraphX__print_maze callgraphX__tile_4_5->callgraphX__tile_3_5 callgraphX__tile_4_5->callgraphX__tile_4_4 callgraphX__tile_4_5->callgraphX__tile_4_5 callgraphX__tile_4_5->callgraphX__tile_4_6 callgraphX__tile_4_5->callgraphX__tile_5_5 callgraphX__tile_4_6->callgraphX__print_maze callgraphX__tile_4_7->callgraphX__print_maze callgraphX__tile_4_7->callgraphX__tile_3_7 callgraphX__tile_4_7->callgraphX__tile_4_6 callgraphX__tile_4_7->callgraphX__tile_4_7 callgraphX__tile_4_7->callgraphX__tile_4_8 callgraphX__tile_4_7->callgraphX__tile_5_7 callgraphX__tile_4_8->callgraphX__print_maze callgraphX__tile_5_0->callgraphX__print_maze callgraphX__tile_5_1->callgraphX__print_maze callgraphX__tile_5_1->callgraphX__tile_4_1 callgraphX__tile_5_1->callgraphX__tile_5_0 callgraphX__tile_5_1->callgraphX__tile_5_1 callgraphX__tile_5_1->callgraphX__tile_5_2 callgraphX__tile_5_1->callgraphX__tile_6_1 callgraphX__tile_5_2->callgraphX__print_maze callgraphX__tile_5_3->callgraphX__print_maze callgraphX__tile_5_4->callgraphX__print_maze callgraphX__tile_5_5->callgraphX__print_maze callgraphX__tile_5_5->callgraphX__tile_4_5 callgraphX__tile_5_5->callgraphX__tile_5_4 callgraphX__tile_5_5->callgraphX__tile_5_5 callgraphX__tile_5_5->callgraphX__tile_5_6 callgraphX__tile_5_5->callgraphX__tile_6_5 callgraphX__tile_5_6->callgraphX__print_maze callgraphX__tile_5_7->callgraphX__print_maze callgraphX__tile_5_7->callgraphX__tile_4_7 callgraphX__tile_5_7->callgraphX__tile_5_6 callgraphX__tile_5_7->callgraphX__tile_5_7 callgraphX__tile_5_7->callgraphX__tile_5_8 callgraphX__tile_5_7->callgraphX__tile_6_7 callgraphX__tile_5_8->callgraphX__print_maze callgraphX__tile_6_0->callgraphX__print_maze callgraphX__tile_6_1->callgraphX__print_maze callgraphX__tile_6_1->callgraphX__tile_5_1 callgraphX__tile_6_1->callgraphX__tile_6_0 callgraphX__tile_6_1->callgraphX__tile_6_1 callgraphX__tile_6_1->callgraphX__tile_6_2 callgraphX__tile_6_1->callgraphX__tile_7_1 callgraphX__tile_6_2->callgraphX__print_maze callgraphX__tile_6_2->callgraphX__tile_5_2 callgraphX__tile_6_2->callgraphX__tile_6_1 callgraphX__tile_6_2->callgraphX__tile_6_2 callgraphX__tile_6_2->callgraphX__tile_6_3 callgraphX__tile_6_2->callgraphX__tile_7_2 callgraphX__tile_6_3->callgraphX__print_maze callgraphX__tile_6_3->callgraphX__tile_5_3 callgraphX__tile_6_3->callgraphX__tile_6_2 callgraphX__tile_6_3->callgraphX__tile_6_3 callgraphX__tile_6_3->callgraphX__tile_6_4 callgraphX__tile_6_3->callgraphX__tile_7_3 callgraphX__tile_6_4->callgraphX__print_maze callgraphX__tile_6_4->callgraphX__tile_5_4 callgraphX__tile_6_4->callgraphX__tile_6_3 callgraphX__tile_6_4->callgraphX__tile_6_4 callgraphX__tile_6_4->callgraphX__tile_6_5 callgraphX__tile_6_4->callgraphX__tile_7_4 callgraphX__tile_6_5->callgraphX__print_maze callgraphX__tile_6_5->callgraphX__tile_5_5 callgraphX__tile_6_5->callgraphX__tile_6_4 callgraphX__tile_6_5->callgraphX__tile_6_5 callgraphX__tile_6_5->callgraphX__tile_6_6 callgraphX__tile_6_5->callgraphX__tile_7_5 callgraphX__tile_6_6->callgraphX__print_maze callgraphX__tile_6_7->callgraphX__print_maze callgraphX__tile_6_8->callgraphX__print_maze callgraphX__tile_7_0->callgraphX__print_maze callgraphX__tile_7_1->callgraphX__print_maze callgraphX__tile_7_2->callgraphX__print_maze callgraphX__tile_7_3->callgraphX__print_maze callgraphX__tile_7_4->callgraphX__print_maze callgraphX__tile_7_5->callgraphX__print_maze callgraphX__tile_7_6->callgraphX__print_maze callgraphX__tile_7_7->callgraphX__print_maze callgraphX__tile_7_8->callgraphX__print_maze

A First Attempt

We introduce a DictMutator class which mutates strings by inserting a keyword from a given dictionary:

class DictMutator(Mutator):
    """Variant of `Mutator` inserting keywords from a dictionary"""

    def __init__(self, dictionary: Sequence[str]) -> None:
        `dictionary` - a list of strings that can be used as keywords
        self.dictionary = dictionary

    def insert_from_dictionary(self, s: str) -> str:
        """Returns `s` with a keyword from the dictionary inserted"""
        pos = random.randint(0, len(s))
        random_keyword = random.choice(self.dictionary)
        return s[:pos] + random_keyword + s[pos:]

To fuzz the maze, we extend the DictMutator class to append dictionary keywords to the end of the seed and to remove a character from the end of the seed.

class MazeMutator(DictMutator):
    def __init__(self, dictionary: Sequence[str]) -> None:

    def append_from_dictionary(self, s: str) -> str:
        """Returns s with a keyword from the dictionary appended"""
        random_keyword = random.choice(self.dictionary)
        return s + random_keyword

    def delete_last_character(self, s: str) -> str:
        """Returns s without the last character"""
        if len(s) > 0:
            return s[:-1]
        return s

Let's try a standard greybox fuzzer with the classic power schedule and our extended maze mutator (n=20k).

n = 20000
seed_input = " "  # empty seed

maze_mutator = MazeMutator(["L", "R", "U", "D"])
maze_schedule = PowerSchedule()
maze_fuzzer = GreyboxFuzzer([seed_input], maze_mutator, maze_schedule)

start = time.time()
maze_fuzzer.runs(FunctionCoverageRunner(maze), trials=n)
end = time.time()

"It took the fuzzer %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the fuzzer 7.02 seconds to generate and execute 20000 inputs.'

We will need to print statistics for several fuzzers. Why don't we define a function for that?

def print_stats(fuzzer: GreyboxFuzzer) -> None:
    total = len(fuzzer.population)
    solved = 0
    invalid = 0
    valid = 0
    for seed in fuzzer.population:
        s = maze(str(
        if "INVALID" in s:
            invalid += 1
        elif "VALID" in s:
            valid += 1
        elif "SOLVED" in s:
            solved += 1
            if solved == 1:
                print("First solution: %s" % repr(seed))

    print("""Out of %d seeds,
* %4d solved the maze,
* %4d were valid but did not solve the maze, and
* %4d were invalid""" % (total, solved, valid, invalid))

How well does our good, old greybox fuzzer do?

Out of 1339 seeds,
*    0 solved the maze,
*  337 were valid but did not solve the maze, and
* 1002 were invalid

It probably didn't solve the maze a single time. How can we make the fuzzer aware how "far" a seed is from reaching the target? If we know that, we can just assign more energy to that seed.

Try it. Print the statistics for the boosted fuzzer using the AFLFastSchedule and the CountingGreyboxFuzzer. It will likely perform much better than the unboosted greybox fuzzer: The lowest-probablity path happens to be also the path which reaches the target. You can execute your own code by opening this chapter as Jupyter notebook.

Computing Function-Level Distance

Using the static call graph for the maze code and the target function, we can compute the distance of each function $f$ to the target $t$ as the length of the shortest path between $f$ and $t$.

Fortunately, the generated maze code includes a function called target_tile which returns the name of the target-function.

target = target_tile()

Now, we need to find the corresponding function in the call graph. The function get_callgraph returns the call graph for the maze code as networkx graph. Networkx provides some useful functions for graph analysis.

import networkx as nx
from ControlFlow import get_callgraph
cg = get_callgraph(maze_code)
for node in cg.nodes():
    if target in node:
        target_node = node

We can now generate the function-level distance. The dictionary distance contains for each function the distance to the target-function. If there is no path to the target, we assign a maximum distance (0xFFFF).

The function nx.shortest_path_length(CG, node, target_node) returns the length of the shortest path from function node to function target_node in the call graph CG.

distance = {}
for node in cg.nodes():
    if "__" in node:
        name = node.split("__")[-1]
        name = node
        distance[name] = nx.shortest_path_length(cg, node, target_node)
        distance[name] = 0xFFFF

These are the distance values for all tile-functions on the path to the target function.

{k: distance[k] for k in list(distance) if distance[k] < 0xFFFF}
{'callgraphX': 1,
 'maze': 23,
 'tile_2_1': 22,
 'tile_2_3': 8,
 'tile_2_4': 7,
 'tile_2_5': 6,
 'tile_2_6': 5,
 'tile_2_7': 4,
 'tile_3_1': 21,
 'tile_3_3': 9,
 'tile_3_7': 3,
 'tile_4_1': 20,
 'tile_4_3': 10,
 'tile_4_4': 11,
 'tile_4_5': 12,
 'tile_4_7': 2,
 'tile_5_1': 19,
 'tile_5_5': 13,
 'tile_5_7': 1,
 'tile_6_1': 18,
 'tile_6_2': 17,
 'tile_6_3': 16,
 'tile_6_4': 15,
 'tile_6_5': 14,
 'tile_6_7': 0}

Summary. Using the static call graph and the target function $t$, we have shown how to compute the function-level distance of each function $f$ to the target $t$.

Try it. You can try and execute your own code by opening this chapter as Jupyter notebook.

  • How do we compute distance if there are multiple targets? (Hint: Geometric Mean).
  • Given the call graph (CG) and the control-flow graph (CFG$_f$) for each function $f$, how do we compute basic-block (BB)-level distance? (Hint: In CFG$_f$, measure the BB-level distance to calls of functions on the path to the target function. Remember that BB-level distance in functions with higher function-level distance is higher, too.)

Read. If you are interested in other aspects of search, you can follow up by reading the chapter on Search-based Fuzzing. If you are interested, how to solve the problems above, you can have a look at our paper on "Directed Greybox Fuzzing".

Directed Power Schedule

Now that we know how to compute the function-level distance, let's try to implement a power schedule that assigns more energy to seeds with a lower average distance to the target function. Notice that the distance values are all pre-computed. These values are injected into the program binary, just like the coverage instrumentation. In practice, this makes the computation of the average distance extremely efficient.

If you really want to know. Given the function-level distance $d_f(s,t)$ of a function $s$ to a function $t$ in call graph $CG$, our directed power schedule computes the seed distance $d(i,t)$ for a seed $i$ to function $t$ as $d(i,t)=\dfrac{\sum_{s\in CG} d_f(s,t)}{|CG|}$ where $|CG|$ is the number of nodes in the call graph $CG$.

class DirectedSchedule(PowerSchedule):
    """Assign high energy to seeds close to some target"""

    def __init__(self, distance: Dict[str, int], exponent: float) -> None:
        self.distance = distance
        self.exponent = exponent

    def __getFunctions__(self, coverage: Set[Location]) -> Set[str]:
        functions = set()
        for f, _ in set(coverage):
        return functions

    def assignEnergy(self, population: Sequence[Seed]) -> None:
        """Assigns each seed energy inversely proportional
           to the average function-level distance to target."""
        for seed in population:
            if seed.distance < 0:
                num_dist = 0
                sum_dist = 0
                for f in self.__getFunctions__(seed.coverage):
                    if f in list(self.distance):
                        sum_dist += self.distance[f]
                        num_dist += 1
                seed.distance = sum_dist / num_dist
       = (1 / seed.distance) ** self.exponent

Let's see how the directed schedule performs against the good, old greybox fuzzer.

directed_schedule = DirectedSchedule(distance, 3)
directed_fuzzer = GreyboxFuzzer([seed_input], maze_mutator, directed_schedule)

start = time.time()
directed_fuzzer.runs(FunctionCoverageRunner(maze), trials=n)
end = time.time()

"It took the fuzzer %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the fuzzer 8.85 seconds to generate and execute 20000 inputs.'
Out of 2446 seeds,
*    0 solved the maze,
*  845 were valid but did not solve the maze, and
* 1601 were invalid

It probably didn't solve a single maze either, but we have more valid solutions. So, there is definitely progress.

Let's have a look at the distance values for each seed.

y = [seed.distance for seed in directed_fuzzer.population]
x = range(len(y))
plt.scatter(x, y)
plt.ylim(0, max(y))
plt.xlabel("Seed ID")

Let's normalize the y-axis and improve the importance of the small distance seeds.

Improved Directed Power Schedule

The improved directed schedule normalizes seed distance between the minimal and maximal distance. Again, if you really want to know. Given the seed distance $d(i,t)$ of a seed $i$ to a function $t$, our improved power schedule computes the new seed distance $d'(i,t)$ as $$ d'(i,t)=\begin{cases} 1 & \text{if } d(i,t) = \text{minD} = \text{maxD}\\ \text{maxD} - \text{minD} & \text{if } d(i,t) = \text{minD} \neq \text{maxD}\\ \frac{\text{maxD} - \text{minD}}{d(i,t)-\text{minD}} & \text{otherwise} \end{cases} $$ where $$\text{minD}=\min_{i\in T}[d(i,t)]$$ and $$\text{maxD}=\max_{i\in T}[d(i,t)]$$ where $T$ is the set of seeds (i.e., the population).

class AFLGoSchedule(DirectedSchedule):
    """Assign high energy to seeds close to the target"""

    def assignEnergy(self, population: Sequence[Seed]):
        """Assigns each seed energy inversely proportional
           to the average function-level distance to target."""
        min_dist: Union[int, float] = 0xFFFF
        max_dist: Union[int, float] = 0

        for seed in population:
            if seed.distance < 0:
                num_dist = 0
                sum_dist = 0
                for f in self.__getFunctions__(seed.coverage):
                    if f in list(self.distance):
                        sum_dist += self.distance[f]
                        num_dist += 1
                seed.distance = sum_dist / num_dist
            if seed.distance < min_dist:
                min_dist = seed.distance
            if seed.distance > max_dist:
                max_dist = seed.distance

        for seed in population:
            if seed.distance == min_dist:
                if min_dist == max_dist:
           = 1
           = max_dist - min_dist
       = (max_dist - min_dist) / (seed.distance - min_dist)

Let's see how the improved power schedule performs.

aflgo_schedule = AFLGoSchedule(distance, 3)
aflgo_fuzzer = GreyboxFuzzer([seed_input], maze_mutator, aflgo_schedule)

start = time.time()
aflgo_fuzzer.runs(FunctionCoverageRunner(maze), trials=n)
end = time.time()

"It took the fuzzer %0.2f seconds to generate and execute %d inputs." % (end - start, n)
'It took the fuzzer 15.21 seconds to generate and execute 20000 inputs.'
Out of 3038 seeds,
*  504 solved the maze,
*  399 were valid but did not solve the maze, and
* 2135 were invalid

In contrast to all previous power schedules, this one generates hundreds of solutions. It has generated many solutions.

Let's filter out all ignored input characters from the first solution. The function filter(f, returns a list of elements e in where the function f applied on e returns True.

for seed in aflgo_fuzzer.population:
    s = maze(str(
    if "SOLVED" in s:
        filtered = "".join(list(filter(lambda c: c in "UDLR",

This is definitely a solution for the maze specified at the beginning!

Summary. After pre-computing the function-level distance to the target, we can develop a power schedule that assigns more energy to a seed with a smaller average function-level distance to the target. By normalizing seed distance values between the minimum and maximum seed distance, we can further boost the directed power schedule.

Try it. Implement and evaluate a simpler directed power that uses the minimal (rather than average) function-level distance. What is the downside of using the minimal distance? In order to execute your code, you just need to open this chapter as Jupyter notebook.

Read. You can find out more about directed greybox fuzzing in the equally-named paper "Directed Greybox Fuzzing" [Böhme et al, 2017] and check out the implementation into AFL at

Lessons Learned

  • A greybox fuzzer generates thousands of inputs per second. Pre-processing and lightweight instrumentation
    • allows maintaining the efficiency during the fuzzing campaign, and
    • still provides enough information to control progress and slightly steer the fuzzer.
  • The power schedule allows steering and controlling the fuzzer. For instance,
    • Our boosted greybox fuzzer spends more energy on seeds that exercise "unlikely" paths. The hope is that the generated inputs exercise even more unlikely paths. This in turn increases the number of paths explored per unit time.
    • Our directed greybox fuzzer spends more energy on seeds that are "closer" to a target location. The hope is that the generated inputs get even closer to the target.
  • The mutator defines the fuzzer's search space. Customizing the mutator for the given program allows reducing the search space to only relevant inputs. In a couple of chapters, we'll learn about dictionary-based, and grammar-based mutators to increase the ratio of valid inputs generated.


Next Steps

Our aim is still to sufficiently cover functionality, such that we can trigger as many bugs as possible. To this end, we focus on two classes of techniques:

  1. Try to cover as much specified functionality as possible. Here, we would need a specification of the input format, distinguishing between individual input elements such as (in our case) numbers, operators, comments, and strings – and attempting to cover as many of these as possible. We will explore this as it comes to grammar-based testing, and especially in grammar-based mutations.

  2. Try to cover as much implemented functionality as possible. The concept of a "population" that is systematically "evolved" through "mutations" will be explored in depth when discussing search-based testing. Furthermore, symbolic testing introduces how to systematically reach program locations by solving the conditions that lie on their paths.

These two techniques make up the gist of the book; and, of course, they can also be combined with each other. As usual, we provide runnable code for all. Enjoy!

We're done, so we clean up:

import os
if os.path.exists(''):

if os.path.exists(''):


In previous version of the chapter, the AdvancedMutationFuzzer class was named simply MutationFuzzer, causing name confusion with the MutationFuzzer class in the introduction to mutation-based fuzzing. The following declaration introduces a backward-compatible alias.

class MutationFuzzer(AdvancedMutationFuzzer):


To be added. \todo{}

Creative Commons License The content of this project is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. The source code that is part of the content, as well as the source code used to format and display that content is licensed under the MIT License. Last change: 2023-11-11 18:18:06+01:00CiteImprint