def sleep_until_next_sync(self):
"""Sleep until it is time to do the next sync."""
if self.last_sync_time is not None:
next_sync_time = (self.last_sync_time +
experiment_utils.get_snapshot_seconds())
sleep_time = next_sync_time - time.time()
if sleep_time < 0:
# Log error if a sync has taken longer than
# get_snapshot_seconds() and messed up our time
# synchronization.
logs.warning('Sleep time on cycle %d is %d', self.cycle,
sleep_time)
sleep_time = 0
else:
sleep_time = experiment_utils.get_snapshot_seconds()
logs.debug('Sleeping for %d seconds.', sleep_time)
time.sleep(sleep_time)
# last_sync_time is recorded before the sync so that each sync happens
# roughly get_snapshot_seconds() after each other.
self.last_sync_time = time.time()
def test_measure_snapshot_coverage( # pylint: disable=too-many-locals
self, mocked_is_cycle_unchanged, db, experiment, tmp_path):
"""Integration test for measure_snapshot_coverage."""
# WORK is set by experiment to a directory that only makes sense in a
# fakefs. A directory containing necessary llvm tools is also added to
# PATH.
llvm_tools_path = get_test_data_path('llvm_tools')
os.environ["PATH"] += os.pathsep + llvm_tools_path
os.environ['WORK'] = str(tmp_path)
mocked_is_cycle_unchanged.return_value = False
# Set up the coverage binary.
benchmark = 'freetype2-2017'
coverage_binary_src = get_test_data_path(
'test_measure_snapshot_coverage', benchmark + '-coverage')
benchmark_cov_binary_dir = os.path.join(
build_utils.get_coverage_binaries_dir(), benchmark)
os.makedirs(benchmark_cov_binary_dir)
coverage_binary_dst_dir = os.path.join(benchmark_cov_binary_dir,
'ftfuzzer')
shutil.copy(coverage_binary_src, coverage_binary_dst_dir)
# Set up entities in database so that the snapshot can be created.
experiment = models.Experiment(name=os.environ['EXPERIMENT'])
db_utils.add_all([experiment])
trial = models.Trial(fuzzer=FUZZER,
benchmark=benchmark,
experiment=os.environ['EXPERIMENT'])
db_utils.add_all([trial])
snapshot_measurer = measurer.SnapshotMeasurer(trial.fuzzer,
trial.benchmark,
trial.id,
SNAPSHOT_LOGGER)
# Set up the snapshot archive.
cycle = 1
archive = get_test_data_path('test_measure_snapshot_coverage',
'corpus-archive-%04d.tar.gz' % cycle)
corpus_dir = os.path.join(snapshot_measurer.trial_dir, 'corpus')
os.makedirs(corpus_dir)
shutil.copy(archive, corpus_dir)
with mock.patch('common.filestore_utils.cp') as mocked_cp:
mocked_cp.return_value = new_process.ProcessResult(0, '', False)
# TODO(metzman): Create a system for using actual buckets in
# integration tests.
snapshot = measurer.measure_snapshot_coverage(
snapshot_measurer.fuzzer, snapshot_measurer.benchmark,
snapshot_measurer.trial_num, cycle)
assert snapshot
assert snapshot.time == cycle * experiment_utils.get_snapshot_seconds()
assert snapshot.edges_covered == 13178
def test_measure_snapshot_coverage( # pylint: disable=too-many-locals
self, mocked_is_cycle_unchanged, create_measurer, db, experiment):
"""Integration test for measure_snapshot_coverage."""
mocked_is_cycle_unchanged.return_value = False
# Set up the coverage binary.
benchmark = 'freetype2-2017'
coverage_binary_src = get_test_data_path(
'test_measure_snapshot_coverage', benchmark + '-coverage')
benchmark_cov_binary_dir = os.path.join(
build_utils.get_coverage_binaries_dir(), benchmark)
os.makedirs(benchmark_cov_binary_dir)
coverage_binary_dst_dir = os.path.join(benchmark_cov_binary_dir,
'fuzz-target')
shutil.copy(coverage_binary_src, coverage_binary_dst_dir)
# Set up entities in database so that the snapshot can be created.
experiment = models.Experiment(name=os.environ['EXPERIMENT'])
db_utils.add_all([experiment])
trial = models.Trial(fuzzer=FUZZER,
benchmark=benchmark,
experiment=os.environ['EXPERIMENT'])
db_utils.add_all([trial])
snapshot_measurer = create_measurer(trial.fuzzer, trial.benchmark,
trial.id)
# Set up the snapshot archive.
cycle = 1
archive = get_test_data_path('test_measure_snapshot_coverage',
'corpus-archive-%04d.tar.gz' % cycle)
corpus_dir = os.path.join(snapshot_measurer.trial_dir, 'corpus')
os.makedirs(corpus_dir)
shutil.copy(archive, corpus_dir)
with mock.patch('common.gsutil.cp') as mocked_cp:
mocked_cp.return_value = new_process.ProcessResult(0, '', False)
# TODO(metzman): Create a system for using actual buckets in
# integration tests.
snapshot = measurer.measure_snapshot_coverage(
snapshot_measurer.fuzzer, snapshot_measurer.benchmark,
snapshot_measurer.trial_num, cycle)
assert snapshot
assert snapshot.time == cycle * experiment_utils.get_snapshot_seconds()
assert snapshot.edges_covered == 3798
logger = logs.Logger('run_coverage')
def find_crashing_units(artifacts_dir: str) -> List[str]:
"""Returns the crashing unit in coverage_binary_output."""
return [
# This assumes the artifacts are named {crash,oom,timeout,*}-$SHA1_HASH
# and that input units are also named with their hash.
filename.split('-')[1] for filename in os.listdir(artifacts_dir)
if os.path.isfile(os.path.join(artifacts_dir, filename))
]
RSS_LIMIT_MB = 2048
UNIT_TIMEOUT = 5
MAX_TOTAL_TIME = experiment_utils.get_snapshot_seconds()
def do_coverage_run( # pylint: disable=too-many-locals
coverage_binary: str, new_units_dir: List[str], sancov_dir: str,
crashes_dir: str) -> List[str]:
"""Does a coverage run of |coverage_binary| on |new_units_dir|. Writes
sancov files to |sancov_dir|. Returns a list of crashing units."""
with tempfile.TemporaryDirectory() as merge_dir:
command = [
coverage_binary, '-merge=1', '-dump_coverage=1',
'-artifact_prefix=%s/' % crashes_dir,
'-timeout=%d' % UNIT_TIMEOUT,
'-rss_limit_mb=%d' % RSS_LIMIT_MB,
'-max_total_time=%d' % MAX_TOTAL_TIME, merge_dir, new_units_dir
]
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 _time_to_cycle(time_in_seconds: float) -> int:
"""Converts |time_in_seconds| to the corresponding cycle and returns it."""
return time_in_seconds // experiment_utils.get_snapshot_seconds()
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
