def csv_filestore_helper(file_name, df):
"""Helper method for storing csv files in filestore."""
src = os.path.join(coverage_utils.get_coverage_info_dir(), 'data',
file_name)
dst = exp_path.filestore(src)
df.to_csv(src, index=False, compression='infer')
filestore_utils.cp(src, dst)
def coverage_steps(benchmark):
"""Returns GCB run steps for coverage builds."""
coverage_binaries_dir = exp_path.filestore(
build_utils.get_coverage_binaries_dir())
steps = [{
'name':
DOCKER_IMAGE,
'args': [
'run',
'-v',
'/workspace/out:/host-out',
# TODO(metzman): Get rid of this and use one source of truth
# for tags.
posixpath.join(get_docker_registry(), 'builders', 'coverage',
benchmark) + ':' +
experiment_utils.get_experiment_name(),
'/bin/bash',
'-c',
'cd /out; tar -czvf /host-out/coverage-build-' + benchmark +
'.tar.gz * /src /work'
]
}]
step = {'name': 'gcr.io/cloud-builders/gsutil'}
step['args'] = [
'-m', 'cp', '/workspace/out/coverage-build-' + benchmark + '.tar.gz',
coverage_binaries_dir + '/'
]
steps.append(step)
return steps
def copy_unchanged_cycles_file():
unchanged_cycles_filestore_path = exp_path.filestore(
self.unchanged_cycles_path)
result = filestore_utils.cp(unchanged_cycles_filestore_path,
self.unchanged_cycles_path,
expect_zero=False)
return result.retcode == 0
def copy_unchanged_cycles_file():
unchanged_cycles_filestore_path = exp_path.filestore(
self.unchanged_cycles_path)
try:
filestore_utils.cp(unchanged_cycles_filestore_path,
self.unchanged_cycles_path)
return True
except subprocess.CalledProcessError:
return False
def get_fuzzer_stats(self, cycle):
"""Get the fuzzer stats for |cycle|."""
stats_filename = experiment_utils.get_stats_filename(cycle)
stats_filestore_path = exp_path.filestore(
os.path.join(self.trial_dir, stats_filename))
try:
return get_fuzzer_stats(stats_filestore_path)
except (ValueError, json.decoder.JSONDecodeError):
logger.error('Stats are invalid.')
return None
def _build_benchmark_coverage(benchmark: str) -> Tuple[int, str]:
"""Build a coverage build of |benchmark| on GCB."""
coverage_binaries_dir = exp_path.filestore(
build_utils.get_coverage_binaries_dir())
substitutions = {
'_GCS_COVERAGE_BINARIES_DIR': coverage_binaries_dir,
'_BENCHMARK': benchmark,
}
config_file = get_build_config_file('coverage.yaml')
config_name = 'benchmark-{benchmark}-coverage'.format(benchmark=benchmark)
return _build(config_file, config_name, substitutions)
def build_coverage(benchmark):
"""Build coverage image for benchmark on GCB."""
coverage_binaries_dir = exp_path.filestore(
build_utils.get_coverage_binaries_dir())
substitutions = {
'_GCS_COVERAGE_BINARIES_DIR': coverage_binaries_dir,
'_BENCHMARK': benchmark,
}
config_file = get_build_config_file('coverage.yaml')
config_name = 'benchmark-{benchmark}-coverage'.format(benchmark=benchmark)
_build(config_file, config_name, substitutions)
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'
filestore_utils.cp(
tmp.name,
exp_path.filestore(get_build_logs_dir() / build_log_filename))
def generate_coverage_regions_json(self):
"""Stores the coverage data in a json file."""
covered_regions = extract_covered_regions_from_summary_json(
self.merged_summary_json_file)
coverage_json_src = os.path.join(self.data_dir, 'covered_regions.json')
coverage_json_dst = exp_path.filestore(coverage_json_src)
filesystem.create_directory(self.data_dir)
with open(coverage_json_src, 'w') as file_handle:
json.dump(covered_regions, file_handle)
filestore_utils.cp(coverage_json_src,
coverage_json_dst,
expect_zero=False)
def store_coverage_data(experiment_config: dict):
"""Generate the specific coverage data and store in cloud bucket."""
logger.info('Start storing coverage data')
with multiprocessing.Pool() as pool, multiprocessing.Manager() as manager:
q = manager.Queue() # pytype: disable=attribute-error
covered_regions = get_all_covered_regions(experiment_config, pool, q)
json_src_dir = get_experiment_folders_dir()
json_src = os.path.join(json_src_dir, 'covered_regions.json')
with open(json_src, 'w') as src_file:
json.dump(covered_regions, src_file)
json_dst = exp_path.filestore(json_src)
filestore_utils.cp(json_src, json_dst)
logger.info('Finished storing coverage data')
def save_crash_files(self, cycle):
"""Save crashes in per-cycle crash archive."""
crashes_archive_name = experiment_utils.get_crashes_archive_name(cycle)
archive_path = os.path.join(os.path.dirname(self.crashes_dir),
crashes_archive_name)
with tarfile.open(archive_path, 'w:gz') as tar:
tar.add(self.crashes_dir,
arcname=os.path.basename(self.crashes_dir))
trial_crashes_dir = posixpath.join(self.trial_dir, 'crashes')
archive_filestore_path = exp_path.filestore(
posixpath.join(trial_crashes_dir, crashes_archive_name))
filestore_utils.cp(archive_path, archive_filestore_path)
os.remove(archive_path)
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
filesystem.create_directory(benchmark_coverage_binary_dir)
archive_name = 'coverage-build-%s.tar.gz' % benchmark
archive_filestore_path = exp_path.filestore(coverage_binaries_dir /
archive_name)
filestore_utils.cp(archive_filestore_path,
str(benchmark_coverage_binary_dir))
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 archive_crashes(self, cycle):
"""Archive this cycle's crashes into filestore."""
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_path = os.path.join(os.path.dirname(self.crashes_dir),
crashes_archive_name)
with tarfile.open(archive_path, 'w:gz') as tar:
tar.add(self.crashes_dir,
arcname=os.path.basename(self.crashes_dir))
archive_filestore_path = exp_path.filestore(
posixpath.join(self.trial_dir, 'crashes', crashes_archive_name))
filestore_utils.cp(archive_path, archive_filestore_path)
os.remove(archive_path)
def _build_oss_fuzz_project_coverage(benchmark: str) -> Tuple[int, str]:
"""Build a coverage build of OSS-Fuzz-based benchmark |benchmark| on GCB."""
project = benchmark_utils.get_project(benchmark)
oss_fuzz_builder_hash = benchmark_utils.get_oss_fuzz_builder_hash(
benchmark)
coverage_binaries_dir = exp_path.filestore(
build_utils.get_coverage_binaries_dir())
substitutions = {
'_GCS_COVERAGE_BINARIES_DIR': coverage_binaries_dir,
'_BENCHMARK': benchmark,
'_OSS_FUZZ_PROJECT': project,
'_OSS_FUZZ_BUILDER_HASH': oss_fuzz_builder_hash,
}
config_file = get_build_config_file('oss-fuzz-coverage.yaml')
config_name = 'oss-fuzz-{project}-coverage-hash-{hash}'.format(
project=project, hash=oss_fuzz_builder_hash)
return _build(config_file, config_name, substitutions)
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.filestore(coverage_binaries_dir), coverage_build_archive)
return filestore_utils.cp(coverage_build_archive_shared_dir_path,
coverage_build_archive_gcs_path)
def generate_coverage_report(self):
"""Generates the coverage report and stores in bucket."""
command = [
'llvm-cov', 'show', '-format=html',
'-path-equivalence=/,{prefix}'.format(
prefix=self.source_files_dir),
'-output-dir={dst_dir}'.format(dst_dir=self.report_dir),
'-Xdemangler', 'c++filt', '-Xdemangler', '-n', self.binary_file,
'-instr-profile={profdata}'.format(
profdata=self.merged_profdata_file)
]
result = new_process.execute(command, expect_zero=False)
if result.retcode != 0:
logger.error('Coverage report generation failed for '
'fuzzer: {fuzzer},benchmark: {benchmark}.'.format(
fuzzer=self.fuzzer, benchmark=self.benchmark))
return
src_dir = self.report_dir
dst_dir = exp_path.filestore(self.report_dir)
filestore_utils.cp(src_dir, dst_dir, recursive=True, parallel=True)
def exists_in_experiment_filestore(path: pathlib.Path) -> bool:
"""Returns True if |path| exists in the experiment_filestore."""
return filestore_utils.ls(exp_path.filestore(path),
must_exist=False).retcode == 0
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)
regions_covered = snapshot_measurer.get_current_coverage()
fuzzer_stats_data = snapshot_measurer.get_fuzzer_stats(cycle)
return models.Snapshot(time=this_time,
trial_id=trial_num,
edges_covered=regions_covered,
fuzzer_stats=fuzzer_stats_data)
corpus_archive_dst = os.path.join(
snapshot_measurer.trial_dir, 'corpus',
experiment_utils.get_corpus_archive_name(cycle))
corpus_archive_src = exp_path.filestore(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 filestore_utils.cp(corpus_archive_src,
corpus_archive_dst,
expect_zero=False).retcode:
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)
# Run coverage on the new corpus units.
snapshot_measurer.run_cov_new_units()
# Generate profdata and transform it into json form.
snapshot_measurer.generate_coverage_information(cycle)
# Get the coverage of the new corpus units.
regions_covered = snapshot_measurer.get_current_coverage()
fuzzer_stats_data = snapshot_measurer.get_fuzzer_stats(cycle)
snapshot = models.Snapshot(time=this_time,
trial_id=trial_num,
edges_covered=regions_covered,
fuzzer_stats=fuzzer_stats_data)
# 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 %f seconds.', cycle,
measuring_time)
return snapshot
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.filestore(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)
try:
filestore_utils.cp(corpus_archive_src, corpus_archive_dst)
except subprocess.CalledProcessError:
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
