def build_measurer(benchmark: str) -> bool:
"""Do a coverage build for a benchmark."""
try:
logger.info('Building measurer for benchmark: %s.', benchmark)
buildlib.build_coverage(benchmark)
docker_name = benchmark_utils.get_docker_name(benchmark)
archive_name = 'coverage-build-%s.tar.gz' % docker_name
coverage_binaries_dir = build_utils.get_coverage_binaries_dir()
benchmark_coverage_binary_dir = coverage_binaries_dir / benchmark
os.mkdir(benchmark_coverage_binary_dir)
cloud_bucket_archive_path = exp_path.gcs(coverage_binaries_dir /
archive_name)
gsutil.cp(cloud_bucket_archive_path,
str(benchmark_coverage_binary_dir),
parallel=False,
write_to_stdout=False)
archive_path = benchmark_coverage_binary_dir / archive_name
tar = tarfile.open(archive_path, 'r:gz')
tar.extractall(benchmark_coverage_binary_dir)
os.remove(archive_path)
logs.info('Done building measurer for benchmark: %s.', benchmark)
return True
except Exception: # pylint: disable=broad-except
logger.error('Failed to build measurer for %s.', benchmark)
return False
def store_build_logs(build_config, build_result):
"""Save build results in the build logs bucket."""
build_output = ('Command returned {retcode}.\nOutput: {output}'.format(
retcode=build_result.retcode, output=build_result.output))
with tempfile.NamedTemporaryFile(mode='w') as tmp:
tmp.write(build_output)
tmp.flush()
build_log_filename = build_config + '.txt'
gsutil.cp(tmp.name,
exp_path.gcs(get_build_logs_dir() / build_log_filename),
parallel=False)
def save_corpus_archive(self, archive):
"""Save corpus |archive| to GCS and delete when done."""
if not self.gcs_sync_dir:
return
basename = os.path.basename(archive)
gcs_path = posixpath.join(self.gcs_sync_dir, self.corpus_dir, basename)
# Don't use parallel to avoid stability issues.
gsutil.cp(archive, gcs_path, parallel=False)
# Delete corpus archive so disk doesn't fill up.
os.remove(archive)
def add_oss_fuzz_corpus(benchmark, oss_fuzz_corpora_dir):
"""Add latest public corpus from OSS-Fuzz as the seed corpus for various
fuzz targets."""
project = benchmark_utils.get_project(benchmark)
fuzz_target = benchmark_utils.get_fuzz_target(benchmark)
if not fuzz_target.startswith(project):
full_fuzz_target = '%s_%s' % (project, fuzz_target)
else:
full_fuzz_target = fuzz_target
src_corpus_url = _OSS_FUZZ_CORPUS_BACKUP_URL_FORMAT.format(
project=project, fuzz_target=full_fuzz_target)
dest_corpus_url = os.path.join(oss_fuzz_corpora_dir, f'{benchmark}.zip')
gsutil.cp(src_corpus_url, dest_corpus_url, parallel=True, expect_zero=False)
def archive_crashes(self, cycle):
"""Archive this cycle's crashes into cloud bucket."""
if not os.listdir(self.crashes_dir):
logs.info('No crashes found for cycle %d.', cycle)
return
logs.info('Archiving crashes for cycle %d.', cycle)
crashes_archive_name = experiment_utils.get_crashes_archive_name(cycle)
archive = os.path.join(os.path.dirname(self.crashes_dir),
crashes_archive_name)
with tarfile.open(archive, 'w:gz') as tar:
tar.add(self.crashes_dir,
arcname=os.path.basename(self.crashes_dir))
gcs_path = exp_path.gcs(
posixpath.join(self.trial_dir, 'crashes', crashes_archive_name))
gsutil.cp(archive, gcs_path)
os.remove(archive)
def set_up_coverage_binary(benchmark):
"""Set up coverage binaries for |benchmark|."""
initialize_logs()
coverage_binaries_dir = build_utils.get_coverage_binaries_dir()
benchmark_coverage_binary_dir = coverage_binaries_dir / benchmark
if not os.path.exists(benchmark_coverage_binary_dir):
os.mkdir(benchmark_coverage_binary_dir)
archive_name = 'coverage-build-%s.tar.gz' % benchmark
cloud_bucket_archive_path = exp_path.gcs(coverage_binaries_dir /
archive_name)
gsutil.cp(cloud_bucket_archive_path,
str(benchmark_coverage_binary_dir),
write_to_stdout=False)
archive_path = benchmark_coverage_binary_dir / archive_name
tar = tarfile.open(archive_path, 'r:gz')
tar.extractall(benchmark_coverage_binary_dir)
os.remove(archive_path)
def copy_resources_to_bucket(config_dir: str, config: Dict):
"""Copy resources the dispatcher will need for the experiment to the
cloud_experiment_bucket."""
def filter_file(tar_info):
"""Filter out unnecessary directories."""
if FILTER_SOURCE_REGEX.match(tar_info.name):
return None
return tar_info
cloud_experiment_path = os.path.join(config['cloud_experiment_bucket'],
config['experiment'])
base_destination = os.path.join(cloud_experiment_path, 'input')
# Send the local source repository to the cloud for use by dispatcher.
# Local changes to any file will propagate.
source_archive = 'src.tar.gz'
with tarfile.open(source_archive, 'w:gz') as tar:
tar.add(utils.ROOT_DIR, arcname='', recursive=True, filter=filter_file)
gsutil.cp(source_archive, base_destination + '/', parallel=True)
os.remove(source_archive)
# Send config files.
destination = os.path.join(base_destination, 'config')
gsutil.rsync(config_dir, destination, parallel=True)
def copy_coverage_binaries(benchmark):
"""Copy coverage binaries in a local experiment."""
shared_coverage_binaries_dir = get_shared_coverage_binaries_dir()
mount_arg = '{0}:{0}'.format(shared_coverage_binaries_dir)
builder_image_url = benchmark_utils.get_builder_image_url(
benchmark, 'coverage', environment.get('CLOUD_PROJECT'))
coverage_build_archive = 'coverage-build-{}.tar.gz'.format(benchmark)
coverage_build_archive_shared_dir_path = os.path.join(
shared_coverage_binaries_dir, coverage_build_archive)
command = 'cd /out; tar -czvf {} *'.format(
coverage_build_archive_shared_dir_path)
new_process.execute([
'docker', 'run', '-v', mount_arg, builder_image_url, '/bin/bash', '-c',
command
])
coverage_binaries_dir = build_utils.get_coverage_binaries_dir()
coverage_build_archive_gcs_path = posixpath.join(
exp_path.gcs(coverage_binaries_dir), coverage_build_archive)
return gsutil.cp(coverage_build_archive_shared_dir_path,
coverage_build_archive_gcs_path)
def measure_snapshot_coverage(fuzzer: str, benchmark: str, trial_num: int,
cycle: int) -> models.Snapshot:
"""Measure coverage of the snapshot for |cycle| for |trial_num| of |fuzzer|
and |benchmark|."""
snapshot_logger = logs.Logger('measurer',
default_extras={
'fuzzer': fuzzer,
'benchmark': benchmark,
'trial_id': str(trial_num),
'cycle': str(cycle),
})
snapshot_measurer = SnapshotMeasurer(fuzzer, benchmark, trial_num,
snapshot_logger)
measuring_start_time = time.time()
snapshot_logger.info('Measuring cycle: %d.', cycle)
this_time = cycle * experiment_utils.get_snapshot_seconds()
if snapshot_measurer.is_cycle_unchanged(cycle):
snapshot_logger.info('Cycle: %d is unchanged.', cycle)
current_pcs = snapshot_measurer.get_current_pcs()
return models.Snapshot(time=this_time,
trial_id=trial_num,
edges_covered=len(current_pcs))
corpus_archive_dst = os.path.join(
snapshot_measurer.trial_dir, 'corpus',
experiment_utils.get_corpus_archive_name(cycle))
corpus_archive_src = exp_path.gcs(corpus_archive_dst)
corpus_archive_dir = os.path.dirname(corpus_archive_dst)
if not os.path.exists(corpus_archive_dir):
os.makedirs(corpus_archive_dir)
if gsutil.cp(corpus_archive_src,
corpus_archive_dst,
expect_zero=False,
parallel=False,
write_to_stdout=False)[0] != 0:
snapshot_logger.warning('Corpus not found for cycle: %d.', cycle)
return None
snapshot_measurer.initialize_measurement_dirs()
snapshot_measurer.extract_corpus(corpus_archive_dst)
# Don't keep corpus archives around longer than they need to be.
os.remove(corpus_archive_dst)
# Get the coverage of the new corpus units.
snapshot_measurer.run_cov_new_units()
all_pcs = snapshot_measurer.merge_new_pcs()
snapshot = models.Snapshot(time=this_time,
trial_id=trial_num,
edges_covered=len(all_pcs))
# Record the new corpus files.
snapshot_measurer.update_measured_files()
# Archive crashes directory.
snapshot_measurer.archive_crashes(cycle)
measuring_time = round(time.time() - measuring_start_time, 2)
snapshot_logger.info('Measured cycle: %d in %d seconds.', cycle,
measuring_time)
return snapshot
def copy_unchanged_cycles_file():
result = gsutil.cp(exp_path.gcs(self.unchanged_cycles_path),
self.unchanged_cycles_path)
return result.retcode == 0
def test_integration_runner(self, mocked_error, tmp_path, environ):
"""Test that runner can run libFuzzer and saves snapshots to GCS."""
# Switch cwd so that fuzzers don't create tons of files in the repo.
os.chdir(tmp_path)
# Set env variables that would be set by the Dockerfile.
file_directory = pathlib.Path(__file__).parent
root_dir = file_directory.parent
os.environ['ROOT_DIR'] = str(root_dir)
seed_corpus_dir = tmp_path / 'seeds'
os.mkdir(seed_corpus_dir)
os.environ['SEED_CORPUS_DIR'] = str(seed_corpus_dir)
output_corpus_dir = tmp_path / 'corpus'
os.mkdir(output_corpus_dir)
os.environ['OUTPUT_CORPUS_DIR'] = str(output_corpus_dir)
fuzzer = 'libfuzzer'
fuzzer_variant = fuzzer + '_variant'
fuzzer_parent_path = root_dir / 'fuzzers' / fuzzer
benchmark = 'MultipleConstraintsOnSmallInputTest'
test_experiment_bucket = os.environ['TEST_CLOUD_EXPERIMENT_BUCKET']
experiment = 'integration-test-experiment'
gcs_directory = posixpath.join(test_experiment_bucket, experiment,
'experiment-folders',
'%s-%s' % (benchmark, fuzzer_variant),
'trial-1')
gsutil.rm(gcs_directory, force=True)
# Add fuzzer directory to make it easy to run fuzzer.py in local
# configuration.
os.environ['PYTHONPATH'] = ':'.join(
[str(root_dir), str(fuzzer_parent_path)])
# Set env variables that would set by the scheduler.
os.environ['FUZZER'] = fuzzer
os.environ['FUZZER_VARIANT_NAME'] = fuzzer_variant
os.environ['BENCHMARK'] = benchmark
os.environ['CLOUD_EXPERIMENT_BUCKET'] = test_experiment_bucket
os.environ['EXPERIMENT'] = experiment
os.environ['TRIAL_ID'] = str(TRIAL_NUM)
max_total_time = 10
os.environ['MAX_TOTAL_TIME'] = str(max_total_time)
target_binary_path = (file_directory / 'test_data' / 'test_runner' /
benchmark)
with mock.patch('common.fuzzer_utils.get_fuzz_target_binary',
return_value=str(target_binary_path)):
with mock.patch('common.experiment_utils.SNAPSHOT_PERIOD',
max_total_time / 10):
runner.main()
gcs_corpus_directory = posixpath.join(gcs_directory, 'corpus')
returncode, snapshots = gsutil.ls(gcs_corpus_directory)
# Ensure that test works.
assert returncode == 0, 'gsutil ls %s failed.' % gcs_corpus_directory
assert len(snapshots) >= 2
# Check that the archives are deleted after being copied to GCS.
assert not os.path.exists(
tmp_path / 'corpus-archives' / 'corpus-archive-0001.tar.gz')
local_gcs_corpus_dir_copy = tmp_path / 'gcs_corpus_dir'
os.mkdir(local_gcs_corpus_dir_copy)
gsutil.cp('-r', posixpath.join(gcs_corpus_directory, '*'),
str(local_gcs_corpus_dir_copy))
archive_size = os.path.getsize(local_gcs_corpus_dir_copy /
'corpus-archive-0001.tar.gz')
assert archive_size > 500
assert len(os.listdir(output_corpus_dir)) > 5
mocked_error.assert_not_called()
