zstd/tests/fuzz/fuzz.py

784 lines
24 KiB
Python
Executable File

#! /usr/bin/env python
# ################################################################
# Copyright (c) 2016-present, Facebook, Inc.
# All rights reserved.
#
# This source code is licensed under both the BSD-style license (found in the
# LICENSE file in the root directory of this source tree) and the GPLv2 (found
# in the COPYING file in the root directory of this source tree).
# ##########################################################################
import argparse
import contextlib
import os
import re
import shutil
import subprocess
import sys
import tempfile
def abs_join(a, *p):
return os.path.abspath(os.path.join(a, *p))
# Constants
FUZZ_DIR = os.path.abspath(os.path.dirname(__file__))
CORPORA_DIR = abs_join(FUZZ_DIR, 'corpora')
TARGETS = [
'simple_round_trip',
'stream_round_trip',
'block_round_trip',
'simple_decompress',
'stream_decompress',
'block_decompress',
]
ALL_TARGETS = TARGETS + ['all']
FUZZ_RNG_SEED_SIZE = 4
# Standard environment variables
CC = os.environ.get('CC', 'cc')
CXX = os.environ.get('CXX', 'c++')
CPPFLAGS = os.environ.get('CPPFLAGS', '')
CFLAGS = os.environ.get('CFLAGS', '-O3')
CXXFLAGS = os.environ.get('CXXFLAGS', CFLAGS)
LDFLAGS = os.environ.get('LDFLAGS', '')
MFLAGS = os.environ.get('MFLAGS', '-j')
# Fuzzing environment variables
LIB_FUZZING_ENGINE = os.environ.get('LIB_FUZZING_ENGINE', 'libregression.a')
AFL_FUZZ = os.environ.get('AFL_FUZZ', 'afl-fuzz')
DECODECORPUS = os.environ.get('DECODECORPUS',
abs_join(FUZZ_DIR, '..', 'decodecorpus'))
# Sanitizer environment variables
MSAN_EXTRA_CPPFLAGS = os.environ.get('MSAN_EXTRA_CPPFLAGS', '')
MSAN_EXTRA_CFLAGS = os.environ.get('MSAN_EXTRA_CFLAGS', '')
MSAN_EXTRA_CXXFLAGS = os.environ.get('MSAN_EXTRA_CXXFLAGS', '')
MSAN_EXTRA_LDFLAGS = os.environ.get('MSAN_EXTRA_LDFLAGS', '')
def create(r):
d = os.path.abspath(r)
if not os.path.isdir(d):
os.mkdir(d)
return d
def check(r):
d = os.path.abspath(r)
if not os.path.isdir(d):
return None
return d
@contextlib.contextmanager
def tmpdir():
dirpath = tempfile.mkdtemp()
try:
yield dirpath
finally:
shutil.rmtree(dirpath, ignore_errors=True)
def parse_targets(in_targets):
targets = set()
for target in in_targets:
if not target:
continue
if target == 'all':
targets = targets.union(TARGETS)
elif target in TARGETS:
targets.add(target)
else:
raise RuntimeError('{} is not a valid target'.format(target))
return list(targets)
def targets_parser(args, description):
parser = argparse.ArgumentParser(prog=args.pop(0), description=description)
parser.add_argument(
'TARGET',
nargs='*',
type=str,
help='Fuzz target(s) to build {{{}}}'.format(', '.join(ALL_TARGETS)))
args, extra = parser.parse_known_args(args)
args.extra = extra
args.TARGET = parse_targets(args.TARGET)
return args
def parse_env_flags(args, flags):
"""
Look for flags set by environment variables.
"""
san_flags = ','.join(re.findall('-fsanitize=((?:[a-z]+,?)+)', flags))
nosan_flags = ','.join(re.findall('-fno-sanitize=((?:[a-z]+,?)+)', flags))
def set_sanitizer(sanitizer, default, san, nosan):
if sanitizer in san and sanitizer in nosan:
raise RuntimeError('-fno-sanitize={s} and -fsanitize={s} passed'.
format(s=sanitizer))
if sanitizer in san:
return True
if sanitizer in nosan:
return False
return default
san = set(san_flags.split(','))
nosan = set(nosan_flags.split(','))
args.asan = set_sanitizer('address', args.asan, san, nosan)
args.msan = set_sanitizer('memory', args.msan, san, nosan)
args.ubsan = set_sanitizer('undefined', args.ubsan, san, nosan)
args.sanitize = args.asan or args.msan or args.ubsan
return args
def build_parser(args):
description = """
Cleans the repository and builds a fuzz target (or all).
Many flags default to environment variables (default says $X='y').
Options that aren't enabling features default to the correct values for
zstd.
Enable sanitizers with --enable-*san.
For regression testing just build.
For libFuzzer set LIB_FUZZING_ENGINE and pass --enable-coverage.
For AFL set CC and CXX to AFL's compilers and set
LIB_FUZZING_ENGINE='libregression.a'.
"""
parser = argparse.ArgumentParser(prog=args.pop(0), description=description)
parser.add_argument(
'--lib-fuzzing-engine',
dest='lib_fuzzing_engine',
type=str,
default=LIB_FUZZING_ENGINE,
help=('The fuzzing engine to use e.g. /path/to/libFuzzer.a '
"(default: $LIB_FUZZING_ENGINE='{})".format(LIB_FUZZING_ENGINE)))
parser.add_argument(
'--enable-coverage',
dest='coverage',
action='store_true',
help='Enable coverage instrumentation (-fsanitize-coverage)')
parser.add_argument(
'--enable-asan', dest='asan', action='store_true', help='Enable UBSAN')
parser.add_argument(
'--enable-ubsan',
dest='ubsan',
action='store_true',
help='Enable UBSAN')
parser.add_argument(
'--enable-ubsan-pointer-overflow',
dest='ubsan_pointer_overflow',
action='store_true',
help='Enable UBSAN pointer overflow check (known failure)')
parser.add_argument(
'--enable-msan', dest='msan', action='store_true', help='Enable MSAN')
parser.add_argument(
'--enable-msan-track-origins', dest='msan_track_origins',
action='store_true', help='Enable MSAN origin tracking')
parser.add_argument(
'--msan-extra-cppflags',
dest='msan_extra_cppflags',
type=str,
default=MSAN_EXTRA_CPPFLAGS,
help="Extra CPPFLAGS for MSAN (default: $MSAN_EXTRA_CPPFLAGS='{}')".
format(MSAN_EXTRA_CPPFLAGS))
parser.add_argument(
'--msan-extra-cflags',
dest='msan_extra_cflags',
type=str,
default=MSAN_EXTRA_CFLAGS,
help="Extra CFLAGS for MSAN (default: $MSAN_EXTRA_CFLAGS='{}')".format(
MSAN_EXTRA_CFLAGS))
parser.add_argument(
'--msan-extra-cxxflags',
dest='msan_extra_cxxflags',
type=str,
default=MSAN_EXTRA_CXXFLAGS,
help="Extra CXXFLAGS for MSAN (default: $MSAN_EXTRA_CXXFLAGS='{}')".
format(MSAN_EXTRA_CXXFLAGS))
parser.add_argument(
'--msan-extra-ldflags',
dest='msan_extra_ldflags',
type=str,
default=MSAN_EXTRA_LDFLAGS,
help="Extra LDFLAGS for MSAN (default: $MSAN_EXTRA_LDFLAGS='{}')".
format(MSAN_EXTRA_LDFLAGS))
parser.add_argument(
'--enable-sanitize-recover',
dest='sanitize_recover',
action='store_true',
help='Non-fatal sanitizer errors where possible')
parser.add_argument(
'--debug',
dest='debug',
type=int,
default=1,
help='Set ZSTD_DEBUG (default: 1)')
parser.add_argument(
'--force-memory-access',
dest='memory_access',
type=int,
default=0,
help='Set MEM_FORCE_MEMORY_ACCESS (default: 0)')
parser.add_argument(
'--fuzz-rng-seed-size',
dest='fuzz_rng_seed_size',
type=int,
default=4,
help='Set FUZZ_RNG_SEED_SIZE (default: 4)')
parser.add_argument(
'--disable-fuzzing-mode',
dest='fuzzing_mode',
action='store_false',
help='Do not define FUZZING_BUILD_MORE_UNSAFE_FOR_PRODUCTION')
parser.add_argument(
'--enable-stateful-fuzzing',
dest='stateful_fuzzing',
action='store_true',
help='Reuse contexts between runs (makes reproduction impossible)')
parser.add_argument(
'--cc',
dest='cc',
type=str,
default=CC,
help="CC (default: $CC='{}')".format(CC))
parser.add_argument(
'--cxx',
dest='cxx',
type=str,
default=CXX,
help="CXX (default: $CXX='{}')".format(CXX))
parser.add_argument(
'--cppflags',
dest='cppflags',
type=str,
default=CPPFLAGS,
help="CPPFLAGS (default: $CPPFLAGS='{}')".format(CPPFLAGS))
parser.add_argument(
'--cflags',
dest='cflags',
type=str,
default=CFLAGS,
help="CFLAGS (default: $CFLAGS='{}')".format(CFLAGS))
parser.add_argument(
'--cxxflags',
dest='cxxflags',
type=str,
default=CXXFLAGS,
help="CXXFLAGS (default: $CXXFLAGS='{}')".format(CXXFLAGS))
parser.add_argument(
'--ldflags',
dest='ldflags',
type=str,
default=LDFLAGS,
help="LDFLAGS (default: $LDFLAGS='{}')".format(LDFLAGS))
parser.add_argument(
'--mflags',
dest='mflags',
type=str,
default=MFLAGS,
help="Extra Make flags (default: $MFLAGS='{}')".format(MFLAGS))
parser.add_argument(
'TARGET',
nargs='*',
type=str,
help='Fuzz target(s) to build {{{}}}'.format(', '.join(ALL_TARGETS))
)
args = parser.parse_args(args)
args = parse_env_flags(args, ' '.join(
[args.cppflags, args.cflags, args.cxxflags, args.ldflags]))
# Check option sanitiy
if args.msan and (args.asan or args.ubsan):
raise RuntimeError('MSAN may not be used with any other sanitizers')
if args.msan_track_origins and not args.msan:
raise RuntimeError('--enable-msan-track-origins requires MSAN')
if args.ubsan_pointer_overflow and not args.ubsan:
raise RuntimeError('--enable-ubsan-pointer-overlow requires UBSAN')
if args.sanitize_recover and not args.sanitize:
raise RuntimeError('--enable-sanitize-recover but no sanitizers used')
return args
def build(args):
try:
args = build_parser(args)
except Exception as e:
print(e)
return 1
# The compilation flags we are setting
targets = args.TARGET
cc = args.cc
cxx = args.cxx
cppflags = [args.cppflags]
cflags = [args.cflags]
ldflags = [args.ldflags]
cxxflags = [args.cxxflags]
mflags = [args.mflags] if args.mflags else []
# Flags to be added to both cflags and cxxflags
common_flags = []
cppflags += [
'-DZSTD_DEBUG={}'.format(args.debug),
'-DMEM_FORCE_MEMORY_ACCESS={}'.format(args.memory_access),
'-DFUZZ_RNG_SEED_SIZE={}'.format(args.fuzz_rng_seed_size),
]
mflags += ['LIB_FUZZING_ENGINE={}'.format(args.lib_fuzzing_engine)]
# Set flags for options
if args.coverage:
common_flags += [
'-fsanitize-coverage=trace-pc-guard,indirect-calls,trace-cmp'
]
if args.sanitize_recover:
recover_flags = ['-fsanitize-recover=all']
else:
recover_flags = ['-fno-sanitize-recover=all']
if args.sanitize:
common_flags += recover_flags
if args.msan:
msan_flags = ['-fsanitize=memory']
if args.msan_track_origins:
msan_flags += ['-fsanitize-memory-track-origins']
common_flags += msan_flags
# Append extra MSAN flags (it might require special setup)
cppflags += [args.msan_extra_cppflags]
cflags += [args.msan_extra_cflags]
cxxflags += [args.msan_extra_cxxflags]
ldflags += [args.msan_extra_ldflags]
if args.asan:
common_flags += ['-fsanitize=address']
if args.ubsan:
ubsan_flags = ['-fsanitize=undefined']
if not args.ubsan_pointer_overflow:
ubsan_flags += ['-fno-sanitize=pointer-overflow']
common_flags += ubsan_flags
if args.stateful_fuzzing:
cppflags += ['-DSTATEFUL_FUZZING']
if args.fuzzing_mode:
cppflags += ['-DFUZZING_BUILD_MORE_UNSAFE_FOR_PRODUCTION']
if args.lib_fuzzing_engine == 'libregression.a':
targets = ['libregression.a'] + targets
# Append the common flags
cflags += common_flags
cxxflags += common_flags
# Prepare the flags for Make
cc_str = "CC={}".format(cc)
cxx_str = "CXX={}".format(cxx)
cppflags_str = "CPPFLAGS={}".format(' '.join(cppflags))
cflags_str = "CFLAGS={}".format(' '.join(cflags))
cxxflags_str = "CXXFLAGS={}".format(' '.join(cxxflags))
ldflags_str = "LDFLAGS={}".format(' '.join(ldflags))
# Print the flags
print('MFLAGS={}'.format(' '.join(mflags)))
print(cc_str)
print(cxx_str)
print(cppflags_str)
print(cflags_str)
print(cxxflags_str)
print(ldflags_str)
# Clean and build
clean_cmd = ['make', 'clean'] + mflags
print(' '.join(clean_cmd))
subprocess.check_call(clean_cmd)
build_cmd = [
'make',
cc_str,
cxx_str,
cppflags_str,
cflags_str,
cxxflags_str,
ldflags_str,
] + mflags + targets
print(' '.join(build_cmd))
subprocess.check_call(build_cmd)
return 0
def libfuzzer_parser(args):
description = """
Runs a libfuzzer binary.
Passes all extra arguments to libfuzzer.
The fuzzer should have been build with LIB_FUZZING_ENGINE pointing to
libFuzzer.a.
Generates output in the CORPORA directory, puts crashes in the ARTIFACT
directory, and takes extra input from the SEED directory.
To merge AFL's output pass the SEED as AFL's output directory and pass
'-merge=1'.
"""
parser = argparse.ArgumentParser(prog=args.pop(0), description=description)
parser.add_argument(
'--corpora',
type=str,
help='Override the default corpora dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET')))
parser.add_argument(
'--artifact',
type=str,
help='Override the default artifact dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET-crash')))
parser.add_argument(
'--seed',
type=str,
help='Override the default seed dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET-seed')))
parser.add_argument(
'TARGET',
type=str,
help='Fuzz target(s) to build {{{}}}'.format(', '.join(TARGETS)))
args, extra = parser.parse_known_args(args)
args.extra = extra
if args.TARGET and args.TARGET not in TARGETS:
raise RuntimeError('{} is not a valid target'.format(args.TARGET))
return args
def libfuzzer(target, corpora=None, artifact=None, seed=None, extra_args=None):
if corpora is None:
corpora = abs_join(CORPORA_DIR, target)
if artifact is None:
artifact = abs_join(CORPORA_DIR, '{}-crash'.format(target))
if seed is None:
seed = abs_join(CORPORA_DIR, '{}-seed'.format(target))
if extra_args is None:
extra_args = []
target = abs_join(FUZZ_DIR, target)
corpora = [create(corpora)]
artifact = create(artifact)
seed = check(seed)
corpora += [artifact]
if seed is not None:
corpora += [seed]
cmd = [target, '-artifact_prefix={}/'.format(artifact)]
cmd += corpora + extra_args
print(' '.join(cmd))
subprocess.check_call(cmd)
def libfuzzer_cmd(args):
try:
args = libfuzzer_parser(args)
except Exception as e:
print(e)
return 1
libfuzzer(args.TARGET, args.corpora, args.artifact, args.seed, args.extra)
return 0
def afl_parser(args):
description = """
Runs an afl-fuzz job.
Passes all extra arguments to afl-fuzz.
The fuzzer should have been built with CC/CXX set to the AFL compilers,
and with LIB_FUZZING_ENGINE='libregression.a'.
Takes input from CORPORA and writes output to OUTPUT.
Uses AFL_FUZZ as the binary (set from flag or environment variable).
"""
parser = argparse.ArgumentParser(prog=args.pop(0), description=description)
parser.add_argument(
'--corpora',
type=str,
help='Override the default corpora dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET')))
parser.add_argument(
'--output',
type=str,
help='Override the default AFL output dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET-afl')))
parser.add_argument(
'--afl-fuzz',
type=str,
default=AFL_FUZZ,
help='AFL_FUZZ (default: $AFL_FUZZ={})'.format(AFL_FUZZ))
parser.add_argument(
'TARGET',
type=str,
help='Fuzz target(s) to build {{{}}}'.format(', '.join(TARGETS)))
args, extra = parser.parse_known_args(args)
args.extra = extra
if args.TARGET and args.TARGET not in TARGETS:
raise RuntimeError('{} is not a valid target'.format(args.TARGET))
if not args.corpora:
args.corpora = abs_join(CORPORA_DIR, args.TARGET)
if not args.output:
args.output = abs_join(CORPORA_DIR, '{}-afl'.format(args.TARGET))
return args
def afl(args):
try:
args = afl_parser(args)
except Exception as e:
print(e)
return 1
target = abs_join(FUZZ_DIR, args.TARGET)
corpora = create(args.corpora)
output = create(args.output)
cmd = [args.afl_fuzz, '-i', corpora, '-o', output] + args.extra
cmd += [target, '@@']
print(' '.join(cmd))
subprocess.call(cmd)
return 0
def regression(args):
try:
description = """
Runs one or more regression tests.
The fuzzer should have been built with with
LIB_FUZZING_ENGINE='libregression.a'.
Takes input from CORPORA.
"""
args = targets_parser(args, description)
except Exception as e:
print(e)
return 1
for target in args.TARGET:
corpora = create(abs_join(CORPORA_DIR, target))
target = abs_join(FUZZ_DIR, target)
cmd = [target, corpora]
print(' '.join(cmd))
subprocess.check_call(cmd)
return 0
def gen_parser(args):
description = """
Generate a seed corpus appropiate for TARGET with data generated with
decodecorpus.
The fuzz inputs are prepended with a seed before the zstd data, so the
output of decodecorpus shouldn't be used directly.
Generates NUMBER samples prepended with FUZZ_RNG_SEED_SIZE random bytes and
puts the output in SEED.
DECODECORPUS is the decodecorpus binary, and must already be built.
"""
parser = argparse.ArgumentParser(prog=args.pop(0), description=description)
parser.add_argument(
'--number',
'-n',
type=int,
default=100,
help='Number of samples to generate')
parser.add_argument(
'--max-size-log',
type=int,
default=13,
help='Maximum sample size to generate')
parser.add_argument(
'--seed',
type=str,
help='Override the default seed dir (default: {})'.format(
abs_join(CORPORA_DIR, 'TARGET-seed')))
parser.add_argument(
'--decodecorpus',
type=str,
default=DECODECORPUS,
help="decodecorpus binary (default: $DECODECORPUS='{}')".format(
DECODECORPUS))
parser.add_argument(
'--fuzz-rng-seed-size',
type=int,
default=4,
help="FUZZ_RNG_SEED_SIZE used for generate the samples (must match)"
)
parser.add_argument(
'TARGET',
type=str,
help='Fuzz target(s) to build {{{}}}'.format(', '.join(TARGETS)))
args, extra = parser.parse_known_args(args)
args.extra = extra
if args.TARGET and args.TARGET not in TARGETS:
raise RuntimeError('{} is not a valid target'.format(args.TARGET))
if not args.seed:
args.seed = abs_join(CORPORA_DIR, '{}-seed'.format(args.TARGET))
if not os.path.isfile(args.decodecorpus):
raise RuntimeError("{} is not a file run 'make -C {} decodecorpus'".
format(args.decodecorpus, abs_join(FUZZ_DIR, '..')))
return args
def gen(args):
try:
args = gen_parser(args)
except Exception as e:
print(e)
return 1
seed = create(args.seed)
with tmpdir() as compressed:
with tmpdir() as decompressed:
cmd = [
args.decodecorpus,
'-n{}'.format(args.number),
'-p{}/'.format(compressed),
'-o{}'.format(decompressed),
]
if 'block_' in args.TARGET:
cmd += [
'--gen-blocks',
'--max-block-size-log={}'.format(args.max_size_log)
]
else:
cmd += ['--max-content-size-log={}'.format(args.max_size_log)]
print(' '.join(cmd))
subprocess.check_call(cmd)
if '_round_trip' in args.TARGET:
print('using decompressed data in {}'.format(decompressed))
samples = decompressed
elif '_decompress' in args.TARGET:
print('using compressed data in {}'.format(compressed))
samples = compressed
# Copy the samples over and prepend the RNG seeds
for name in os.listdir(samples):
samplename = abs_join(samples, name)
outname = abs_join(seed, name)
rng_seed = os.urandom(args.fuzz_rng_seed_size)
with open(samplename, 'rb') as sample:
with open(outname, 'wb') as out:
out.write(rng_seed)
CHUNK_SIZE = 131072
chunk = sample.read(CHUNK_SIZE)
while len(chunk) > 0:
out.write(chunk)
chunk = sample.read(CHUNK_SIZE)
return 0
def minimize(args):
try:
description = """
Runs a libfuzzer fuzzer with -merge=1 to build a minimal corpus in
TARGET_seed_corpus. All extra args are passed to libfuzzer.
"""
args = targets_parser(args, description)
except Exception as e:
print(e)
return 1
for target in args.TARGET:
# Merge the corpus + anything else into the seed_corpus
corpus = abs_join(CORPORA_DIR, target)
seed_corpus = abs_join(CORPORA_DIR, "{}_seed_corpus".format(target))
extra_args = [corpus, "-merge=1"] + args.extra
libfuzzer(target, corpora=seed_corpus, extra_args=extra_args)
seeds = set(os.listdir(seed_corpus))
# Copy all crashes directly into the seed_corpus if not already present
crashes = abs_join(CORPORA_DIR, '{}-crash'.format(target))
for crash in os.listdir(crashes):
if crash not in seeds:
shutil.copy(abs_join(crashes, crash), seed_corpus)
seeds.add(crash)
def zip_cmd(args):
try:
description = """
Zips up the seed corpus.
"""
args = targets_parser(args, description)
except Exception as e:
print(e)
return 1
for target in args.TARGET:
# Zip the seed_corpus
seed_corpus = abs_join(CORPORA_DIR, "{}_seed_corpus".format(target))
seeds = [abs_join(seed_corpus, f) for f in os.listdir(seed_corpus)]
zip_file = "{}.zip".format(seed_corpus)
cmd = ["zip", "-q", "-j", "-9", zip_file]
print(' '.join(cmd + [abs_join(seed_corpus, '*')]))
subprocess.check_call(cmd + seeds)
def short_help(args):
name = args[0]
print("Usage: {} [OPTIONS] COMMAND [ARGS]...\n".format(name))
def help(args):
short_help(args)
print("\tfuzzing helpers (select a command and pass -h for help)\n")
print("Options:")
print("\t-h, --help\tPrint this message")
print("")
print("Commands:")
print("\tbuild\t\tBuild a fuzzer")
print("\tlibfuzzer\tRun a libFuzzer fuzzer")
print("\tafl\t\tRun an AFL fuzzer")
print("\tregression\tRun a regression test")
print("\tgen\t\tGenerate a seed corpus for a fuzzer")
print("\tminimize\tMinimize the test corpora")
print("\tzip\t\tZip the minimized corpora up")
def main():
args = sys.argv
if len(args) < 2:
help(args)
return 1
if args[1] == '-h' or args[1] == '--help' or args[1] == '-H':
help(args)
return 1
command = args.pop(1)
args[0] = "{} {}".format(args[0], command)
if command == "build":
return build(args)
if command == "libfuzzer":
return libfuzzer_cmd(args)
if command == "regression":
return regression(args)
if command == "afl":
return afl(args)
if command == "gen":
return gen(args)
if command == "minimize":
return minimize(args)
if command == "zip":
return zip_cmd(args)
short_help(args)
print("Error: No such command {} (pass -h for help)".format(command))
return 1
if __name__ == "__main__":
sys.exit(main())