def run(self, round_number):
"""Run the testcase once."""
app_directory = environment.get_value('APP_DIR')
warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
run_timeout = warmup_timeout if round_number == 1 else self._test_timeout
if self._is_black_box:
return_code, crash_time, output = process_handler.run_process(
self._command,
timeout=run_timeout,
gestures=self._gestures,
current_working_directory=app_directory)
else:
result = engine_reproduce(self._engine_impl,
self._fuzz_target.binary,
self._testcase_path, self._arguments,
run_timeout)
return_code = result.return_code
crash_time = result.time_executed
log_header = engine_common.get_log_header(
result.command, environment.get_value('BOT_NAME'),
result.time_executed)
output = log_header + '\n' + result.output
process_handler.terminate_stale_application_instances()
crash_result = CrashResult(return_code, crash_time, output)
if not crash_result.is_crash():
logs.log('No crash occurred (round {round_number}).'.format(
round_number=round_number),
output=output)
return crash_result
def _pre_run_cleanup(self):
"""Common cleanup before running a testcase."""
# Cleanup any existing application instances and user profile directories.
# Cleaning up temp user profile directories. Should be done before calling
# |get_command_line_for_application| call since that creates dependencies in
# the profile folder.
process_handler.terminate_stale_application_instances()
shell.clear_temp_directory()
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 store_testcase_dependencies_from_bundled_testcase_archive(
metadata, testcase, testcase_file_path):
"""Store testcase dependencies from a bundled testcase archive
(with multiple testcases)."""
# Nothing to do if this is not a bundled testcase archive with
# multiple testcase.
if not metadata.bundled:
return 1
# Cleanup, before re-running the app to get resource list.
process_handler.terminate_stale_application_instances()
shell.clear_temp_directory()
# Increase the test timeout to warmup timeout. This is needed
# since we use warmup timeout in test_for_crash_with_retries.
warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
environment.set_value('TEST_TIMEOUT', warmup_timeout)
env_copy = environment.copy()
crash_queue = process_handler.get_queue()
thread_index = 0
thread = process_handler.get_process()(
target=testcase_manager.run_testcase_and_return_result_in_queue,
args=(crash_queue, thread_index, testcase_file_path, testcase.gestures,
env_copy))
thread.start()
thread.join(warmup_timeout)
if thread.is_alive():
try:
thread.terminate()
except:
logs.log_error('Process termination failed or was not needed.')
# We should be able to reproduce this reproducible crash. If not,
# something wrong happened and we would just try to redo task.
if not crash_queue.empty():
crash = crash_queue.get()
fuzzed_key, archived, absolute_path, archive_filename = (
setup.archive_testcase_and_dependencies_in_gcs(crash.resource_list,
testcase_file_path))
if archived:
testcase.archive_state = data_types.ArchiveStatus.FUZZED
else:
testcase.archive_state = 0
testcase.fuzzed_keys = fuzzed_key
metadata.blobstore_key = fuzzed_key
testcase.absolute_path = absolute_path
if archive_filename:
testcase.archive_filename = archive_filename
metadata.filename = os.path.basename(archive_filename)
else:
metadata.filename = os.path.basename(testcase_file_path)
metadata.bundled = False
metadata.put()
else:
logs.log_error('Could not get crash data from queue. Retrying task.')
tasks.add_task('analyze', testcase.key.id(), testcase.job_type)
return None
return 1
def test_terminate_stale_application_instances(self): """Test terminating stale application instances.""" # TODO(ochang): Improve this test once we use Docker. process_handler.terminate_stale_application_instances()
def TerminateStaleApplicationInstances(self, request, context): # pylint: disable=unused-argument
process_handler.terminate_stale_application_instances()
return untrusted_runner_pb2.TerminateStaleApplicationInstancesResponse(
)
def check_for_bad_build(job_type, crash_revision):
"""Return true if the build is bad, i.e. crashes on startup."""
# Check the bad build check flag to see if we want do this.
if not environment.get_value('BAD_BUILD_CHECK'):
return False
# Create a blank command line with no file to run and no http.
command = get_command_line_for_application(file_to_run='',
needs_http=False)
# When checking for bad builds, we use the default window size.
# We don't want to pick a custom size since it can potentially cause a
# startup crash and cause a build to be detected incorrectly as bad.
default_window_argument = environment.get_value('WINDOW_ARG', '')
if default_window_argument:
command = command.replace(' %s' % default_window_argument, '')
# TSAN is slow, and boots slow on first startup. Increase the warmup
# timeout for this case.
if environment.tool_matches('TSAN', job_type):
fast_warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
else:
fast_warmup_timeout = environment.get_value('FAST_WARMUP_TIMEOUT')
# Initialize helper variables.
is_bad_build = False
build_run_console_output = ''
app_directory = environment.get_value('APP_DIR')
# Exit all running instances.
process_handler.terminate_stale_application_instances()
# Check if the build is bad.
return_code, crash_time, output = process_handler.run_process(
command,
timeout=fast_warmup_timeout,
current_working_directory=app_directory)
crash_result = CrashResult(return_code, crash_time, output)
# 1. Need to account for startup crashes with no crash state. E.g. failed to
# load shared library. So, ignore state for comparison.
# 2. Ignore leaks as they don't block a build from reporting regular crashes
# and also don't impact regression range calculations.
if (crash_result.is_crash(ignore_state=True)
and not crash_result.should_ignore() and
not crash_result.get_type() in ['Direct-leak', 'Indirect-leak']):
is_bad_build = True
build_run_console_output = utils.get_crash_stacktrace_output(
command, crash_result.get_stacktrace(symbolized=True),
crash_result.get_stacktrace(symbolized=False))
logs.log('Bad build for %s detected at r%d.' %
(job_type, crash_revision),
output=build_run_console_output)
# Exit all running instances.
process_handler.terminate_stale_application_instances()
# Any of the conditions below indicate that bot is in a bad state and it is
# not caused by the build itself. In that case, just exit.
build_state = data_handler.get_build_state(job_type, crash_revision)
if is_bad_build and utils.sub_string_exists_in(BAD_STATE_HINTS, output):
logs.log_fatal_and_exit(
'Bad bot environment detected, exiting.',
output=build_run_console_output,
snapshot=process_handler.get_runtime_snapshot())
# If none of the other bots have added information about this build,
# then add it now.
if (build_state == data_types.BuildState.UNMARKED
and not crash_result.should_ignore()):
data_handler.add_build_metadata(job_type, crash_revision, is_bad_build,
build_run_console_output)
return is_bad_build
def execute_task(testcase_id, job_type):
"""Execute a symbolize command."""
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
# We should atleast have a symbolized debug or release build.
if not build_manager.has_symbolized_builds():
return
data_handler.update_testcase_comment(testcase, data_types.TaskState.STARTED)
# Setup testcase and its dependencies.
file_list, _, testcase_file_path = setup.setup_testcase(testcase, job_type)
if not file_list:
return
# Initialize variables.
build_fail_wait = environment.get_value("FAIL_WAIT")
old_crash_stacktrace = data_handler.get_stacktrace(testcase)
sym_crash_type = testcase.crash_type
sym_crash_address = testcase.crash_address
sym_crash_state = testcase.crash_state
sym_redzone = DEFAULT_REDZONE
warmup_timeout = environment.get_value("WARMUP_TIMEOUT")
# Decide which build revision to use.
if testcase.crash_stacktrace == "Pending":
# This usually happen when someone clicked the 'Update stacktrace from
# trunk' button on the testcase details page. In this case, we are forced
# to use trunk. No revision -> trunk build.
build_revision = None
else:
build_revision = testcase.crash_revision
# Set up a custom or regular build based on revision.
build_manager.setup_build(build_revision)
# Get crash revision used in setting up build.
crash_revision = environment.get_value("APP_REVISION")
if not build_manager.check_app_path():
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
"Build setup failed")
tasks.add_task(
"symbolize", testcase_id, job_type, wait_time=build_fail_wait)
return
# ASAN tool settings (if the tool is used).
# See if we can get better stacks with higher redzone sizes.
# A UAF might actually turn out to be OOB read/write with a bigger redzone.
if environment.tool_matches("ASAN", job_type) and testcase.security_flag:
redzone = MAX_REDZONE
while redzone >= MIN_REDZONE:
environment.reset_current_memory_tool_options(
redzone_size=testcase.redzone, disable_ubsan=testcase.disable_ubsan)
process_handler.terminate_stale_application_instances()
command = testcase_manager.get_command_line_for_application(
testcase_file_path, needs_http=testcase.http_flag)
return_code, crash_time, output = process_handler.run_process(
command, timeout=warmup_timeout, gestures=testcase.gestures)
crash_result = CrashResult(return_code, crash_time, output)
if crash_result.is_crash() and "AddressSanitizer" in output:
state = crash_result.get_symbolized_data()
security_flag = crash_result.is_security_issue()
if (not crash_analyzer.ignore_stacktrace(state.crash_stacktrace) and
security_flag == testcase.security_flag and
state.crash_type == testcase.crash_type and
(state.crash_type != sym_crash_type or
state.crash_state != sym_crash_state)):
logs.log("Changing crash parameters.\nOld : %s, %s, %s" %
(sym_crash_type, sym_crash_address, sym_crash_state))
sym_crash_type = state.crash_type
sym_crash_address = state.crash_address
sym_crash_state = state.crash_state
sym_redzone = redzone
old_crash_stacktrace = state.crash_stacktrace
logs.log("\nNew : %s, %s, %s" %
(sym_crash_type, sym_crash_address, sym_crash_state))
break
redzone /= 2
# We should have atleast a symbolized debug or a release build.
symbolized_builds = build_manager.setup_symbolized_builds(crash_revision)
if not symbolized_builds or (
not build_manager.check_app_path() and
not build_manager.check_app_path("APP_PATH_DEBUG")):
testcase = data_handler.get_testcase_by_id(testcase_id)
data_handler.update_testcase_comment(testcase, data_types.TaskState.ERROR,
"Build setup failed")
tasks.add_task(
"symbolize", testcase_id, job_type, wait_time=build_fail_wait)
return
# Increase malloc_context_size to get all stack frames. Default is 30.
environment.reset_current_memory_tool_options(
redzone_size=sym_redzone,
malloc_context_size=STACK_FRAME_COUNT,
symbolize_inline_frames=True,
disable_ubsan=testcase.disable_ubsan,
)
# TSAN tool settings (if the tool is used).
if environment.tool_matches("TSAN", job_type):
environment.set_tsan_max_history_size()
# Do the symbolization if supported by this application.
result, sym_crash_stacktrace = get_symbolized_stacktraces(
testcase_file_path, testcase, old_crash_stacktrace, sym_crash_state)
# Update crash parameters.
testcase = data_handler.get_testcase_by_id(testcase_id)
testcase.crash_type = sym_crash_type
testcase.crash_address = sym_crash_address
testcase.crash_state = sym_crash_state
testcase.crash_stacktrace = data_handler.filter_stacktrace(
sym_crash_stacktrace)
if not result:
data_handler.update_testcase_comment(
testcase,
data_types.TaskState.ERROR,
"Unable to reproduce crash, skipping "
"stacktrace update",
)
else:
# Switch build url to use the less-optimized symbolized build with better
# stacktrace.
build_url = environment.get_value("BUILD_URL")
if build_url:
testcase.set_metadata("build_url", build_url, update_testcase=False)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED)
testcase.symbolized = True
testcase.crash_revision = crash_revision
testcase.put()
# We might have updated the crash state. See if we need to marked as duplicate
# based on other testcases.
data_handler.handle_duplicate_entry(testcase)
task_creation.create_blame_task_if_needed(testcase)
# Switch current directory before builds cleanup.
root_directory = environment.get_value("ROOT_DIR")
os.chdir(root_directory)
# Cleanup symbolized builds which are space-heavy.
symbolized_builds.delete()
def get_symbolized_stacktraces(testcase_file_path, testcase,
old_crash_stacktrace, expected_state):
"""Use the symbolized builds to generate an updated stacktrace."""
# Initialize variables.
app_path = environment.get_value("APP_PATH")
app_path_debug = environment.get_value("APP_PATH_DEBUG")
long_test_timeout = environment.get_value("WARMUP_TIMEOUT")
retry_limit = environment.get_value("FAIL_RETRIES")
symbolized = False
debug_build_stacktrace = ""
release_build_stacktrace = old_crash_stacktrace
# Symbolize using the debug build first so that the debug build stacktrace
# comes after the more important release build stacktrace.
if app_path_debug:
for _ in range(retry_limit):
process_handler.terminate_stale_application_instances()
command = testcase_manager.get_command_line_for_application(
testcase_file_path,
app_path=app_path_debug,
needs_http=testcase.http_flag,
)
return_code, crash_time, output = process_handler.run_process(
command, timeout=long_test_timeout, gestures=testcase.gestures)
crash_result = CrashResult(return_code, crash_time, output)
if crash_result.is_crash():
state = crash_result.get_symbolized_data()
if crash_analyzer.ignore_stacktrace(state.crash_stacktrace):
continue
unsymbolized_crash_stacktrace = crash_result.get_stacktrace(
symbolized=False)
debug_build_stacktrace = utils.get_crash_stacktrace_output(
command,
state.crash_stacktrace,
unsymbolized_crash_stacktrace,
build_type="debug",
)
symbolized = True
break
# Symbolize using the release build.
if app_path:
for _ in range(retry_limit):
process_handler.terminate_stale_application_instances()
command = testcase_manager.get_command_line_for_application(
testcase_file_path, app_path=app_path, needs_http=testcase.http_flag)
return_code, crash_time, output = process_handler.run_process(
command, timeout=long_test_timeout, gestures=testcase.gestures)
crash_result = CrashResult(return_code, crash_time, output)
if crash_result.is_crash():
state = crash_result.get_symbolized_data()
if crash_analyzer.ignore_stacktrace(state.crash_stacktrace):
continue
if state.crash_state != expected_state:
continue
# Release stack's security flag has to match the symbolized release
# stack's security flag.
security_flag = crash_result.is_security_issue()
if security_flag != testcase.security_flag:
continue
unsymbolized_crash_stacktrace = crash_result.get_stacktrace(
symbolized=False)
release_build_stacktrace = utils.get_crash_stacktrace_output(
command,
state.crash_stacktrace,
unsymbolized_crash_stacktrace,
build_type="release",
)
symbolized = True
break
stacktrace = release_build_stacktrace
if debug_build_stacktrace:
stacktrace += "\n\n" + debug_build_stacktrace
return symbolized, stacktrace
def check_for_bad_build(job_type, crash_revision):
"""Return true if the build is bad, i.e. crashes on startup."""
# Check the bad build check flag to see if we want do this.
if not environment.get_value('BAD_BUILD_CHECK'):
return False
# Do not detect leaks while checking for bad builds.
environment.reset_current_memory_tool_options(leaks=False)
# Create a blank command line with no file to run and no http.
command = get_command_line_for_application(file_to_run='',
needs_http=False)
# When checking for bad builds, we use the default window size.
# We don't want to pick a custom size since it can potentially cause a
# startup crash and cause a build to be detected incorrectly as bad.
default_window_argument = environment.get_value('WINDOW_ARG', '')
if default_window_argument:
command = command.replace(' %s' % default_window_argument, '')
# Warmup timeout.
fast_warmup_timeout = environment.get_value('FAST_WARMUP_TIMEOUT')
# TSAN is slow, and boots slow on first startup. Increase the warmup
# timeout for this case.
if environment.tool_matches('TSAN', job_type):
fast_warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
# Initialize helper variables.
is_bad_build = False
build_run_console_output = ''
output = ''
app_directory = environment.get_value('APP_DIR')
# Check if the build is bad.
process_handler.terminate_stale_application_instances()
exit_code, _, output = process_handler.run_process(
command,
timeout=fast_warmup_timeout,
current_working_directory=app_directory)
output = utils.decode_to_unicode(output)
if crash_analyzer.is_crash(exit_code, output):
is_bad_build = True
build_run_console_output = (
'%s\n\n%s\n\n%s' %
(command, stack_symbolizer.symbolize_stacktrace(output), output))
logs.log('Bad build for %s detected at r%d.' %
(job_type, crash_revision),
output=build_run_console_output)
# Exit all running instances.
process_handler.terminate_stale_application_instances()
# Any of the conditions below indicate that bot is in a bad state and it is
# not caused by the build itself. In that case, just exit.
build_state = data_handler.get_build_state(job_type, crash_revision)
if (is_bad_build and ('cannot open display' in output
or 'logging service has stopped' in output
or 'Maximum number of clients reached' in output)):
logs.log_fatal_and_exit('Bad bot environment detected, exiting.',
output=build_run_console_output)
# If none of the other bots have added information about this build,
# then add it now.
if build_state == data_types.BuildState.UNMARKED:
data_handler.add_build_metadata(job_type, crash_revision, is_bad_build,
build_run_console_output)
# Reset memory tool options.
environment.reset_current_memory_tool_options()
return is_bad_build
def test_for_reproducibility(testcase_path, expected_state,
expected_security_flag, test_timeout, http_flag,
gestures):
"""Test to see if a crash is fully reproducible or is a one-time crasher."""
# Cleanup any existing application instances and user profile directories.
# Cleaning up temp clears user profile directories and should be done before
# calling |get_command_line_for_application| call since that creates
# dependencies in the profile folder.
process_handler.terminate_stale_application_instances()
shell.clear_temp_directory()
app_directory = environment.get_value('APP_DIR')
command = get_command_line_for_application(testcase_path,
needs_http=http_flag)
crash_count = 0
crash_retries = environment.get_value('CRASH_RETRIES')
reproducible_crash_target_count = crash_retries * REPRODUCIBILITY_FACTOR
warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
logs.log('Testing for crash (command="%s").' % command)
round_number = 0
for round_number in xrange(1, crash_retries + 1):
# Bail out early if there is no hope of finding a reproducible crash.
if (crash_retries - round_number + crash_count + 1 <
reproducible_crash_target_count):
break
run_timeout = warmup_timeout if round_number == 1 else test_timeout
return_code, crash_time, output = process_handler.run_process(
command,
timeout=run_timeout,
gestures=gestures,
current_working_directory=app_directory)
process_handler.terminate_stale_application_instances()
crash_result = CrashResult(return_code, crash_time, output)
if not crash_result.is_crash():
continue
state = crash_result.get_symbolized_data()
crash_state = state.crash_state
security_flag = crash_result.is_security_issue()
# If we don't have an expected crash state, set it to the one from initial
# crash.
if not expected_state:
expected_state = crash_state
if security_flag != expected_security_flag:
logs.log('Detected a crash without the correct security flag.')
continue
crash_comparer = CrashComparer(crash_state, expected_state)
if not crash_comparer.is_similar():
logs.log('Detected a crash with an unrelated state: '
'Expected(%s), Found(%s).' %
(expected_state, crash_state))
continue
crash_count += 1
if crash_count >= reproducible_crash_target_count:
logs.log('Crash is reproducible.')
return True
logs.log('Crash is not reproducible. Crash count: %d/%d.' %
(crash_count, round_number))
return False
def test_for_crash_with_retries(testcase,
testcase_path,
test_timeout,
http_flag=False,
compare_crash=True):
"""Test for a crash and return crash parameters like crash type, crash state,
crash stacktrace, etc."""
# Cleanup any existing application instances and user profile directories.
# Cleaning up temp clears user profile directories and should be done before
# calling |get_command_line_for_application| call since that creates
# dependencies in the profile folder.
process_handler.terminate_stale_application_instances()
shell.clear_temp_directory()
app_directory = environment.get_value('APP_DIR')
command = get_command_line_for_application(testcase_path,
needs_http=http_flag)
crash_retries = environment.get_value('CRASH_RETRIES')
flaky_stacktrace = testcase.flaky_stack
warmup_timeout = environment.get_value('WARMUP_TIMEOUT')
logs.log('Testing for crash (command="%s").' % command)
for round_number in xrange(1, crash_retries + 1):
run_timeout = warmup_timeout if round_number == 1 else test_timeout
return_code, crash_time, output = process_handler.run_process(
command,
timeout=run_timeout,
gestures=testcase.gestures,
current_working_directory=app_directory)
process_handler.terminate_stale_application_instances()
crash_result = CrashResult(return_code, crash_time, output)
if not crash_result.is_crash():
continue
state = crash_result.get_symbolized_data()
logs.log('Crash occurred in %d seconds (round %d). State:\n%s' %
(crash_time, round_number, state.crash_state))
if not compare_crash or not testcase.crash_state:
logs.log('Crash stacktrace comparison skipped.')
return crash_result
if flaky_stacktrace:
logs.log('Crash stacktrace is marked flaky, skipping comparison.')
return crash_result
if crash_result.should_ignore():
logs.log('Crash stacktrace matched ignore signatures, ignored.')
continue
if crash_result.is_security_issue() != testcase.security_flag:
logs.log('Crash security flag does not match, ignored.')
continue
crash_comparer = CrashComparer(state.crash_state, testcase.crash_state)
if crash_comparer.is_similar():
logs.log('Crash stacktrace is similar to original stacktrace.')
return crash_result
else:
logs.log('Crash stacktrace does not match original stacktrace.')
logs.log("Didn't crash at all.")
crash_result = CrashResult(return_code=0, crash_time=0, output=output)
return crash_result
