def test(self):
"""Test that ASan instrumentation can be successfully set up on device."""
adb.revert_asan_device_setup_if_needed()
environment.reset_current_memory_tool_options()
sanitizer.setup_asan_if_needed()
self.assertEqual(0, self.mock.log_error.call_count)
self.assertTrue(adb.file_exists('/system/bin/asanwrapper'))
def test_doesnt_mutate_options(self):
"""Test that calling get_memory_tool_options followed by
set_memory_tool_options does not mutate sanitizer options unless we
do so explicitly."""
# Make environment module use the Windows symbolizer, since its path is
# hard to get right.
test_helpers.patch_environ(self)
os.environ['JOB_NAME'] = 'windows_libfuzzer_chrome_asan'
test_helpers.patch(self, [
'clusterfuzz._internal.system.environment.platform',
'clusterfuzz._internal.system.environment.get_llvm_symbolizer_path'
])
self.mock.platform.return_value = 'WINDOWS'
windows_symbolizer_path = (
r'c:\clusterfuzz\resources\platform\windows\llvm-symbolizer.exe')
self.mock.get_llvm_symbolizer_path.return_value = windows_symbolizer_path
environment.reset_current_memory_tool_options()
memory_tool_var = 'ASAN_OPTIONS'
first_asan_options_dict = environment.get_memory_tool_options(
memory_tool_var)
environment.set_memory_tool_options(memory_tool_var,
first_asan_options_dict)
second_asan_options_dict = environment.get_memory_tool_options(
memory_tool_var)
self.assertDictEqual(first_asan_options_dict, second_asan_options_dict)
def test_ubsan_disabled(self):
"""Test reset_current_memory_tool_options when ubsan is disabled."""
os.environ['JOB_NAME'] = 'libfuzzer_chrome_asan'
os.environ['UBSAN'] = 'True'
environment.reset_current_memory_tool_options(disable_ubsan=True)
self.assertDictEqual(
{
'halt_on_error': 0,
'print_stacktrace': 0,
'print_suppressions': 0
}, environment.get_memory_tool_options('UBSAN_OPTIONS'))
def _setup_memory_tools_environment(testcase):
"""Set up environment for various memory tools used."""
env = testcase.get_metadata('env')
if not env:
environment.reset_current_memory_tool_options(
redzone_size=testcase.redzone,
disable_ubsan=testcase.disable_ubsan)
return
for options_name, options_value in six.iteritems(env):
if not options_value:
environment.remove_key(options_name)
continue
environment.set_memory_tool_options(options_name, options_value)
def test_windows_symbolizer(self):
"""Test that the reset_current_memory_tool_options returns the expected path
to the llvm symbolizer on Windows."""
os.environ['JOB_NAME'] = 'windows_libfuzzer_chrome_asan'
test_helpers.patch(self, [
'clusterfuzz._internal.system.environment.platform',
'clusterfuzz._internal.system.environment.get_llvm_symbolizer_path'
])
self.mock.platform.return_value = 'WINDOWS'
windows_symbolizer_path = (
r'c:\clusterfuzz\resources\platform\windows\llvm-symbolizer.exe')
self.mock.get_llvm_symbolizer_path.return_value = windows_symbolizer_path
environment.reset_current_memory_tool_options()
self.assertIn(
'external_symbolizer_path="%s"' % windows_symbolizer_path,
os.environ['ASAN_OPTIONS'])
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 run(self, initial_corpus_path, minimized_corpus_path, bad_units_path):
"""Run corpus pruning. Output result to directory."""
if not shell.get_directory_file_count(initial_corpus_path):
# Empty corpus, nothing to do.
return None
# Set memory tool options and fuzzer arguments.
engine_common.unpack_seed_corpus_if_needed(self.runner.target_path,
initial_corpus_path,
force_unpack=True)
environment.reset_current_memory_tool_options(redzone_size=MIN_REDZONE,
leaks=True)
self.runner.process_sanitizer_options()
additional_args = self.runner.get_libfuzzer_flags()
# Execute fuzzer with arguments for corpus pruning.
logs.log('Running merge...')
try:
result = self.runner.minimize_corpus(additional_args,
[initial_corpus_path],
minimized_corpus_path,
bad_units_path,
CORPUS_PRUNING_TIMEOUT)
except TimeoutError as e:
raise CorpusPruningException(
'Corpus pruning timed out while minimizing corpus\n' + repr(e))
except engine.Error as e:
raise CorpusPruningException(
'Corpus pruning failed to minimize corpus\n' + repr(e))
symbolized_output = stack_symbolizer.symbolize_stacktrace(result.logs)
# Sanity check that there are files in minimized corpus after merging.
if not shell.get_directory_file_count(minimized_corpus_path):
raise CorpusPruningException(
'Corpus pruning failed to minimize corpus\n' +
symbolized_output)
logs.log('Corpus merge finished successfully.',
output=symbolized_output)
return result.stats
def test_ubsan_enabled(self):
"""Test reset_current_memory_tool_options when ubsan is enabled."""
os.environ['JOB_NAME'] = 'libfuzzer_chrome_asan'
os.environ['UBSAN'] = 'True'
environment.reset_current_memory_tool_options(disable_ubsan=False)
self.assertDictEqual(
{
'halt_on_error': 1,
'handle_abort': 1,
'handle_segv': 1,
'handle_sigbus': 1,
'handle_sigfpe': 1,
'handle_sigill': 1,
'print_stacktrace': 1,
'print_summary': 1,
'print_suppressions': 0,
'silence_unsigned_overflow': 1,
'use_sigaltstack': 1
}, environment.get_memory_tool_options('UBSAN_OPTIONS'))
def run(self, timeout):
"""Merge testcases from corpus from other fuzz targets."""
if not shell.get_directory_file_count(self.context.shared_corpus_path):
logs.log('No files found in shared corpus, skip merge.')
return None
# Run pruning on the shared corpus and log the result in case of error.
logs.log('Merging shared corpus...')
environment.reset_current_memory_tool_options(
redzone_size=DEFAULT_REDZONE)
self.runner.process_sanitizer_options()
additional_args = self.runner.get_libfuzzer_flags()
try:
result = self.runner.minimize_corpus(
additional_args, [self.context.shared_corpus_path],
self.context.minimized_corpus_path,
self.context.bad_units_path, timeout)
symbolized_output = stack_symbolizer.symbolize_stacktrace(
result.logs)
logs.log('Shared corpus merge finished successfully.',
output=symbolized_output)
except TimeoutError as e:
# Other cross pollinated fuzzer corpuses can have unexpected test cases
# that time us out. This is expected, so bail out.
logs.log_warn(
'Corpus pruning timed out while merging shared corpus\n' +
repr(e))
return None
except engine.Error as e:
# Other cross pollinated fuzzer corpuses can be large, so we can run out
# of disk space and exception out. This is expected, so bail out.
logs.log_warn('Corpus pruning failed to merge shared corpus\n' +
repr(e))
return None
return result.stats
def execute_task(testcase_id, job_type):
"""Run analyze task."""
# Reset redzones.
environment.reset_current_memory_tool_options(redzone_size=128)
# Unset window location size and position properties so as to use default.
environment.set_value('WINDOW_ARG', '')
# Locate the testcase associated with the id.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
data_handler.update_testcase_comment(testcase,
data_types.TaskState.STARTED)
metadata = data_types.TestcaseUploadMetadata.query(
data_types.TestcaseUploadMetadata.testcase_id == int(
testcase_id)).get()
if not metadata:
logs.log_error('Testcase %s has no associated upload metadata.' %
testcase_id)
testcase.key.delete()
return
is_lsan_enabled = environment.get_value('LSAN')
if is_lsan_enabled:
# Creates empty local blacklist so all leaks will be visible to uploader.
leak_blacklist.create_empty_local_blacklist()
# Store the bot name and timestamp in upload metadata.
bot_name = environment.get_value('BOT_NAME')
metadata.bot_name = bot_name
metadata.timestamp = datetime.datetime.utcnow()
metadata.put()
# Adjust the test timeout, if user has provided one.
if metadata.timeout:
environment.set_value('TEST_TIMEOUT', metadata.timeout)
# Adjust the number of retries, if user has provided one.
if metadata.retries is not None:
environment.set_value('CRASH_RETRIES', metadata.retries)
# Set up testcase and get absolute testcase path.
file_list, _, testcase_file_path = setup.setup_testcase(testcase, job_type)
if not file_list:
return
# Set up build.
setup_build(testcase)
# Check if we have an application path. If not, our build failed
# to setup correctly.
if not build_manager.check_app_path():
data_handler.update_testcase_comment(testcase,
data_types.TaskState.ERROR,
'Build setup failed')
if data_handler.is_first_retry_for_task(testcase):
build_fail_wait = environment.get_value('FAIL_WAIT')
tasks.add_task('analyze',
testcase_id,
job_type,
wait_time=build_fail_wait)
else:
data_handler.close_invalid_uploaded_testcase(
testcase, metadata, 'Build setup failed')
return
# Update initial testcase information.
testcase.absolute_path = testcase_file_path
testcase.job_type = job_type
testcase.binary_flag = utils.is_binary_file(testcase_file_path)
testcase.queue = tasks.default_queue()
testcase.crash_state = ''
# Set initial testcase metadata fields (e.g. build url, etc).
data_handler.set_initial_testcase_metadata(testcase)
# Update minimized arguments and use ones provided during user upload.
if not testcase.minimized_arguments:
minimized_arguments = environment.get_value('APP_ARGS') or ''
additional_command_line_flags = testcase.get_metadata(
'uploaded_additional_args')
if additional_command_line_flags:
minimized_arguments += ' %s' % additional_command_line_flags
environment.set_value('APP_ARGS', minimized_arguments)
testcase.minimized_arguments = minimized_arguments
# Update other fields not set at upload time.
testcase.crash_revision = environment.get_value('APP_REVISION')
data_handler.set_initial_testcase_metadata(testcase)
testcase.put()
# Initialize some variables.
gestures = testcase.gestures
http_flag = testcase.http_flag
test_timeout = environment.get_value('TEST_TIMEOUT')
# Get the crash output.
result = testcase_manager.test_for_crash_with_retries(testcase,
testcase_file_path,
test_timeout,
http_flag=http_flag,
compare_crash=False)
# If we don't get a crash, try enabling http to see if we can get a crash.
# Skip engine fuzzer jobs (e.g. libFuzzer, AFL) for which http testcase paths
# are not applicable.
if (not result.is_crash() and not http_flag
and not environment.is_engine_fuzzer_job()):
result_with_http = testcase_manager.test_for_crash_with_retries(
testcase,
testcase_file_path,
test_timeout,
http_flag=True,
compare_crash=False)
if result_with_http.is_crash():
logs.log('Testcase needs http flag for crash.')
http_flag = True
result = result_with_http
# Refresh our object.
testcase = data_handler.get_testcase_by_id(testcase_id)
if not testcase:
return
# Set application command line with the correct http flag.
application_command_line = (
testcase_manager.get_command_line_for_application(
testcase_file_path, needs_http=http_flag))
# Get the crash data.
crashed = result.is_crash()
crash_time = result.get_crash_time()
state = result.get_symbolized_data()
unsymbolized_crash_stacktrace = result.get_stacktrace(symbolized=False)
# Get crash info object with minidump info. Also, re-generate unsymbolized
# stacktrace if needed.
crash_info, _ = (crash_uploader.get_crash_info_and_stacktrace(
application_command_line, state.crash_stacktrace, gestures))
if crash_info:
testcase.minidump_keys = crash_info.store_minidump()
if not crashed:
# Could not reproduce the crash.
log_message = ('Testcase didn\'t crash in %d seconds (with retries)' %
test_timeout)
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED,
log_message)
# In the general case, we will not attempt to symbolize if we do not detect
# a crash. For user uploads, we should symbolize anyway to provide more
# information about what might be happening.
crash_stacktrace_output = utils.get_crash_stacktrace_output(
application_command_line, state.crash_stacktrace,
unsymbolized_crash_stacktrace)
testcase.crash_stacktrace = data_handler.filter_stacktrace(
crash_stacktrace_output)
# For an unreproducible testcase, retry once on another bot to confirm
# our results and in case this bot is in a bad state which we didn't catch
# through our usual means.
if data_handler.is_first_retry_for_task(testcase):
testcase.status = 'Unreproducible, retrying'
testcase.put()
tasks.add_task('analyze', testcase_id, job_type)
return
data_handler.close_invalid_uploaded_testcase(testcase, metadata,
'Unreproducible')
# A non-reproducing testcase might still impact production branches.
# Add the impact task to get that information.
task_creation.create_impact_task_if_needed(testcase)
return
# Update testcase crash parameters.
testcase.http_flag = http_flag
testcase.crash_type = state.crash_type
testcase.crash_address = state.crash_address
testcase.crash_state = state.crash_state
crash_stacktrace_output = utils.get_crash_stacktrace_output(
application_command_line, state.crash_stacktrace,
unsymbolized_crash_stacktrace)
testcase.crash_stacktrace = data_handler.filter_stacktrace(
crash_stacktrace_output)
# Try to guess if the bug is security or not.
security_flag = crash_analyzer.is_security_issue(state.crash_stacktrace,
state.crash_type,
state.crash_address)
testcase.security_flag = security_flag
# If it is, guess the severity.
if security_flag:
testcase.security_severity = severity_analyzer.get_security_severity(
state.crash_type, state.crash_stacktrace, job_type, bool(gestures))
log_message = ('Testcase crashed in %d seconds (r%d)' %
(crash_time, testcase.crash_revision))
data_handler.update_testcase_comment(testcase,
data_types.TaskState.FINISHED,
log_message)
# See if we have to ignore this crash.
if crash_analyzer.ignore_stacktrace(state.crash_stacktrace):
data_handler.close_invalid_uploaded_testcase(testcase, metadata,
'Irrelavant')
return
# Test for reproducibility.
one_time_crasher_flag = not testcase_manager.test_for_reproducibility(
testcase.fuzzer_name, testcase.actual_fuzzer_name(),
testcase_file_path, state.crash_state, security_flag, test_timeout,
http_flag, gestures)
testcase.one_time_crasher_flag = one_time_crasher_flag
# Check to see if this is a duplicate.
data_handler.check_uploaded_testcase_duplicate(testcase, metadata)
# Set testcase and metadata status if not set already.
if testcase.status == 'Duplicate':
# For testcase uploaded by bots (with quiet flag), don't create additional
# tasks.
if metadata.quiet_flag:
data_handler.close_invalid_uploaded_testcase(
testcase, metadata, 'Duplicate')
return
else:
# New testcase.
testcase.status = 'Processed'
metadata.status = 'Confirmed'
# Reset the timestamp as well, to respect
# data_types.MIN_ELAPSED_TIME_SINCE_REPORT. Otherwise it may get filed by
# triage task prematurely without the grouper having a chance to run on this
# testcase.
testcase.timestamp = utils.utcnow()
# Add new leaks to global blacklist to avoid detecting duplicates.
# Only add if testcase has a direct leak crash and if it's reproducible.
if is_lsan_enabled:
leak_blacklist.add_crash_to_global_blacklist_if_needed(testcase)
# Update the testcase values.
testcase.put()
# Update the upload metadata.
metadata.security_flag = security_flag
metadata.put()
_add_default_issue_metadata(testcase)
# Create tasks to
# 1. Minimize testcase (minimize).
# 2. Find regression range (regression).
# 3. Find testcase impact on production branches (impact).
# 4. Check whether testcase is fixed (progression).
# 5. Get second stacktrace from another job in case of
# one-time crashes (stack).
task_creation.create_tasks(testcase)
def process_bad_units(self, bad_units_path, quarantine_corpus_path,
crashes):
"""Process bad units found during merge."""
# TODO(ochang): A lot of this function is similar to parts of fuzz_task.
# Ideally fuzz_task can be refactored in a way that lets us share the common
# code.
environment.reset_current_memory_tool_options(
redzone_size=DEFAULT_REDZONE)
self.runner.process_sanitizer_options()
logs.log('Processing bad units.')
corpus_file_paths = _get_corpus_file_paths(bad_units_path)
num_bad_units = 0
# Run each corpus item individually.
for i, unit_path in enumerate(corpus_file_paths, 1):
if i % 100 == 0:
logs.log('Up to %d' % i)
unit_name = os.path.basename(unit_path)
if unit_name.startswith('timeout-') or unit_name.startswith(
'oom-'):
# Don't waste time re-running timeout or oom testcases.
self._quarantine_unit(unit_path, quarantine_corpus_path)
num_bad_units += 1
continue
try:
result = self._run_single_unit(unit_path)
except TimeoutError:
# Slow unit. Quarantine it.
self._quarantine_unit(unit_path, quarantine_corpus_path)
num_bad_units += 1
continue
if not crash_analyzer.is_memory_tool_crash(result.output):
# Didn't crash.
continue
# Get memory tool crash information.
state = stack_analyzer.get_crash_data(result.output,
symbolize_flag=True)
# Crashed or caused a leak. Quarantine it.
unit_path = self._quarantine_unit(unit_path,
quarantine_corpus_path)
num_bad_units += 1
if crash_analyzer.ignore_stacktrace(state.crash_stacktrace):
continue
# Local de-duplication.
if state.crash_state not in crashes:
security_flag = crash_analyzer.is_security_issue(
state.crash_stacktrace, state.crash_type,
state.crash_address)
crashes[state.crash_state] = CorpusCrash(
state.crash_state, state.crash_type, state.crash_address,
state.crash_stacktrace, unit_path, security_flag)
logs.log('Found %d bad units, %d unique crashes.' %
(num_bad_units, len(crashes)))
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()