def beat(previous_state, log_filename):
"""Run a cycle of heartbeat checks to ensure bot is running."""
# Handle case when run_bot.py script is stuck. If yes, kill its process.
task_end_time = tasks.get_task_end_time()
if psutil and task_end_time and dates.time_has_expired(
task_end_time, seconds=tasks.TASK_COMPLETION_BUFFER):
# Get absolute path to |run_bot| script. We use this to identify unique
# instances of bot running on a particular host.
startup_scripts_directory = environment.get_startup_scripts_directory()
bot_file_path = os.path.join(startup_scripts_directory, 'run_bot')
for process in psutil.process_iter():
try:
command_line = ' '.join(process.cmdline())
except (psutil.AccessDenied, psutil.NoSuchProcess, OSError):
sys.exc_clear()
continue
# Find the process running the main bot script.
if bot_file_path not in command_line:
continue
process_id = process.pid
logs.log('Killing stale bot (pid %d) which seems to have stuck.' %
process_id)
try:
process_handler.terminate_root_and_child_processes(process_id)
except Exception:
logs.log_error('Failed to terminate stale bot processes.')
# Minor cleanup to avoid disk space issues on bot restart.
process_handler.terminate_stale_application_instances()
shell.clear_temp_directory()
shell.clear_testcase_directories()
# Concerned stale processes should be killed. Now, delete the stale task.
tasks.track_task_end()
# Figure out when the log file was last modified.
try:
current_state = str(os.path.getmtime(log_filename))
except Exception:
current_state = None
logs.log('Old state %s, current state %s.' %
(previous_state, current_state))
# Only update the heartbeat if the log file was modified.
if current_state and current_state != previous_state:
# Try updating the heartbeat. If an error occurs, just
# wait and return None.
if not data_handler.update_heartbeat():
return None
# Heartbeat is successfully updated.
return current_state
def test_clear_testcase_directories(self):
"""Test clearing test directories."""
fuzz_inputs = os.environ['FUZZ_INPUTS']
worker_fuzz_inputs = file_host.rebase_to_worker_root(fuzz_inputs)
fuzz_inputs_disk = os.environ['FUZZ_INPUTS_DISK']
worker_fuzz_inputs_disk = file_host.rebase_to_worker_root(fuzz_inputs_disk)
with open(os.path.join(worker_fuzz_inputs, 'file'), 'w') as f:
f.write('blah')
with open(os.path.join(worker_fuzz_inputs_disk, 'file'), 'w') as f:
f.write('blah2')
shell.clear_testcase_directories()
self.assertEqual(len(os.listdir(worker_fuzz_inputs)), 0)
self.assertEqual(len(os.listdir(worker_fuzz_inputs_disk)), 0)
def cleanup_task_state(): """Cleans state before and after a task is executed.""" # Cleanup stale processes. process_handler.cleanup_stale_processes() # Clear build urls, temp and testcase directories. shell.clear_build_urls_directory() shell.clear_crash_stacktraces_directory() shell.clear_testcase_directories() shell.clear_temp_directory() shell.clear_system_temp_directory() shell.clear_device_temp_directories() # Reset memory tool environment variables. environment.reset_current_memory_tool_options() # Call python's garbage collector. utils.python_gc()
def setup_testcase(testcase, fuzzer_override=None):
"""Sets up the testcase and needed dependencies like fuzzer,
data bundle, etc."""
fuzzer_name = fuzzer_override or testcase.fuzzer_name
job_type = testcase.job_type
task_name = environment.get_value('TASK_NAME')
testcase_fail_wait = environment.get_value('FAIL_WAIT')
testcase_id = testcase.key.id()
# Clear testcase directories.
shell.clear_testcase_directories()
# Adjust the test timeout value if this is coming from an user uploaded
# testcase.
_set_timeout_value_from_user_upload(testcase_id)
# Update the fuzzer if necessary in order to get the updated data bundle.
if fuzzer_name:
try:
update_successful = update_fuzzer_and_data_bundles(fuzzer_name)
except errors.InvalidFuzzerError:
# Close testcase and don't recreate tasks if this fuzzer is invalid.
testcase.open = False
testcase.fixed = 'NA'
testcase.set_metadata('fuzzer_was_deleted', True)
logs.log_error('Closed testcase %d with invalid fuzzer %s.' %
(testcase_id, fuzzer_name))
error_message = 'Fuzzer %s no longer exists' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
return None, None, None
if not update_successful:
error_message = 'Unable to setup fuzzer %s' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# Extract the testcase and any of its resources to the input directory.
file_list, input_directory, testcase_file_path = unpack_testcase(testcase)
if not file_list:
error_message = 'Unable to setup testcase %s' % testcase_file_path
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# For Android/Fuchsia, we need to sync our local testcases directory with the
# one on the device.
if environment.platform() == 'ANDROID':
_copy_testcase_to_device_and_setup_environment(testcase,
testcase_file_path)
# Push testcases to worker.
if environment.is_trusted_host():
from bot.untrusted_runner import file_host
file_host.push_testcases_to_worker()
# Copy global blacklist into local blacklist.
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Get local blacklist without this testcase's entry.
leak_blacklist.copy_global_to_local_blacklist(
excluded_testcase=testcase)
prepare_environment_for_testcase(testcase)
return file_list, input_directory, testcase_file_path
def execute_task(metadata_id, job_type):
"""Unpack a bundled testcase archive and create analyze jobs for each item."""
metadata = ndb.Key(data_types.BundledArchiveMetadata, int(metadata_id)).get()
if not metadata:
logs.log_error('Invalid bundle metadata id %s.' % metadata_id)
return
bot_name = environment.get_value('BOT_NAME')
upload_metadata = data_types.TestcaseUploadMetadata.query(
data_types.TestcaseUploadMetadata.blobstore_key ==
metadata.blobstore_key).get()
if not upload_metadata:
logs.log_error('Invalid upload metadata key %s.' % metadata.blobstore_key)
return
# Update the upload metadata with this bot name.
upload_metadata.bot_name = bot_name
upload_metadata.put()
# We can't use FUZZ_INPUTS directory since it is constrained
# by tmpfs limits.
testcases_directory = environment.get_value('FUZZ_INPUTS_DISK')
# Retrieve multi-testcase archive.
archive_path = os.path.join(testcases_directory, metadata.archive_filename)
if not blobs.read_blob_to_disk(metadata.blobstore_key, archive_path):
logs.log_error('Could not retrieve archive for bundle %d.' % metadata_id)
tasks.add_task('unpack', metadata_id, job_type)
return
try:
archive.unpack(archive_path, testcases_directory)
except:
logs.log_error('Could not unpack archive for bundle %d.' % metadata_id)
tasks.add_task('unpack', metadata_id, job_type)
return
archive_state = data_types.ArchiveStatus.NONE
bundled = True
file_list = archive.get_file_list(archive_path)
for file_path in file_list:
absolute_file_path = os.path.join(testcases_directory, file_path)
filename = os.path.basename(absolute_file_path)
# Only files are actual testcases. Skip directories.
if not os.path.isfile(absolute_file_path):
continue
try:
file_handle = open(absolute_file_path, 'rb')
blob_key = blobs.write_blob(file_handle)
file_handle.close()
except:
blob_key = None
if not blob_key:
logs.log_error(
'Could not write testcase %s to blobstore.' % absolute_file_path)
continue
data_handler.create_user_uploaded_testcase(
blob_key, metadata.blobstore_key, archive_state,
metadata.archive_filename, filename, metadata.timeout,
metadata.job_type, metadata.job_queue, metadata.http_flag,
metadata.gestures, metadata.additional_arguments,
metadata.bug_information, metadata.crash_revision,
metadata.uploader_email, metadata.platform_id,
metadata.app_launch_command, metadata.fuzzer_name,
metadata.overridden_fuzzer_name, metadata.fuzzer_binary_name, bundled,
upload_metadata.retries, upload_metadata.bug_summary_update_flag)
# The upload metadata for the archive is not needed anymore since we created
# one for each testcase.
upload_metadata.key.delete()
shell.clear_testcase_directories()
def setup_testcase(testcase):
"""Sets up the testcase and needed dependencies like fuzzer,
data bundle, etc."""
fuzzer_name = testcase.fuzzer_name
job_type = testcase.job_type
task_name = environment.get_value('TASK_NAME')
testcase_fail_wait = environment.get_value('FAIL_WAIT')
testcase_id = testcase.key.id()
# Clear testcase directories.
shell.clear_testcase_directories()
# Setup memory debugging tool environment.
environment.reset_current_memory_tool_options(
redzone_size=testcase.redzone)
# Adjust the test timeout value if this is coming from an user uploaded
# testcase.
_set_timeout_value_from_user_upload(testcase_id)
if task_name == 'minimize':
# Allow minimizing with a different fuzzer set up.
minimize_fuzzer_override = environment.get_value(
'MINIMIZE_FUZZER_OVERRIDE')
fuzzer_name = minimize_fuzzer_override or fuzzer_name
# Update the fuzzer if necessary in order to get the updated data bundle.
if fuzzer_name:
try:
update_successful = update_fuzzer_and_data_bundles(fuzzer_name)
except errors.InvalidFuzzerError:
# Close testcase and don't recreate tasks if this fuzzer is invalid.
testcase.open = False
testcase.fixed = 'NA'
testcase.set_metadata('fuzzer_was_deleted', True)
logs.log_error('Closed testcase %d with invalid fuzzer %s.' %
(testcase_id, fuzzer_name))
error_message = 'Fuzzer %s no longer exists.' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
return None, None, None
if not update_successful:
error_message = 'Unable to setup fuzzer %s.' % fuzzer_name
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# Extract the testcase and any of its resources to the input directory.
file_list, input_directory, testcase_file_path = unpack_testcase(testcase)
if not file_list:
error_message = 'Unable to setup testcase %s.' % testcase_file_path
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
error_message)
tasks.add_task(task_name,
testcase_id,
job_type,
wait_time=testcase_fail_wait)
return None, None, None
# For Android/Fuchsia, we need to sync our local testcases directory with the
# one on the device.
if environment.platform() == 'ANDROID':
_copy_testcase_to_device_and_setup_environment(testcase,
testcase_file_path)
if environment.platform() == 'FUCHSIA':
fuchsia.device.copy_testcase_to_device(testcase_file_path)
# Push testcases to worker.
if environment.is_trusted_host():
from bot.untrusted_runner import file_host
file_host.push_testcases_to_worker()
# Copy global blacklist into local blacklist.
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Get local blacklist without this testcase's entry.
leak_blacklist.copy_global_to_local_blacklist(
excluded_testcase=testcase)
# Setup environment variable for windows size and location properties.
# Explicitly use empty string to indicate use of default window properties.
if hasattr(testcase, 'window_argument'):
environment.set_value('WINDOW_ARG', testcase.window_argument)
# Adjust timeout based on the stored multiplier (if available).
if hasattr(testcase, 'timeout_multiplier') and testcase.timeout_multiplier:
test_timeout = environment.get_value('TEST_TIMEOUT')
environment.set_value('TEST_TIMEOUT',
int(test_timeout * testcase.timeout_multiplier))
# Override APP_ARGS with minimized arguments (if available).
if (hasattr(testcase, 'minimized_arguments')
and testcase.minimized_arguments):
environment.set_value('APP_ARGS', testcase.minimized_arguments)
# Add FUZZ_TARGET to environment if this is a fuzz target testcase.
fuzz_target = testcase.get_metadata('fuzzer_binary_name')
if fuzz_target:
environment.set_value('FUZZ_TARGET', fuzz_target)
return file_list, input_directory, testcase_file_path
def do_libfuzzer_minimization(testcase, testcase_file_path):
"""Use libFuzzer's built-in minimizer where appropriate."""
is_overriden_job = bool(environment.get_value('ORIGINAL_JOB_NAME'))
def handle_unreproducible():
# Be more lenient with marking testcases as unreproducible when this is a
# job override.
if is_overriden_job:
_skip_minimization(testcase, 'Unreproducible on overridden job.')
else:
task_creation.mark_unreproducible_if_flaky(testcase, True)
timeout = environment.get_value('LIBFUZZER_MINIMIZATION_TIMEOUT', 180)
rounds = environment.get_value('LIBFUZZER_MINIMIZATION_ROUNDS', 10)
current_testcase_path = testcase_file_path
last_crash_result = None
# Get initial crash state.
initial_crash_result = _run_libfuzzer_testcase(testcase, testcase_file_path)
if not initial_crash_result.is_crash():
logs.log_warn('Did not crash. Output:\n' +
initial_crash_result.get_stacktrace(symbolized=True))
handle_unreproducible()
return
if testcase.security_flag != initial_crash_result.is_security_issue():
logs.log_warn('Security flag does not match.')
handle_unreproducible()
return
task_creation.mark_unreproducible_if_flaky(testcase, False)
expected_state = initial_crash_result.get_symbolized_data()
logs.log('Initial crash state: %s\n' % expected_state.crash_state)
# We attempt minimization multiple times in case one round results in an
# incorrect state, or runs into another issue such as a slow unit.
for round_number in range(1, rounds + 1):
logs.log('Minimizing round %d.' % round_number)
output_file_path, crash_result = _run_libfuzzer_tool(
'minimize',
testcase,
current_testcase_path,
timeout,
expected_state.crash_state,
set_dedup_flags=True)
if output_file_path:
last_crash_result = crash_result
current_testcase_path = output_file_path
if not last_crash_result:
repro_command = tests.get_command_line_for_application(
file_to_run=testcase_file_path, needs_http=testcase.http_flag)
_skip_minimization(
testcase,
'LibFuzzer minimization failed.',
crash_result=initial_crash_result,
command=repro_command)
return
logs.log('LibFuzzer minimization succeeded.')
if utils.is_oss_fuzz():
# Scrub the testcase of non-essential data.
cleansed_testcase_path = do_libfuzzer_cleanse(
testcase, current_testcase_path, expected_state.crash_state)
if cleansed_testcase_path:
current_testcase_path = cleansed_testcase_path
# Finalize the test case if we were able to reproduce it.
repro_command = tests.get_command_line_for_application(
file_to_run=current_testcase_path, needs_http=testcase.http_flag)
finalize_testcase(testcase.key.id(), repro_command, last_crash_result)
# Clean up after we're done.
shell.clear_testcase_directories()
