def test_search_excludes(self):
"""Test SEARCH_EXCLUDES env var works."""
crash_stacktrace = ('aaa\nbbbbbbb\nccc\nddd\n\n')
self.assertFalse(crash_analyzer.ignore_stacktrace(crash_stacktrace))
os.environ['SEARCH_EXCLUDES'] = r'eeee'
self.assertFalse(crash_analyzer.ignore_stacktrace(crash_stacktrace))
os.environ['SEARCH_EXCLUDES'] = r'ccc'
self.assertTrue(crash_analyzer.ignore_stacktrace(crash_stacktrace))
def test_stack_blacklist_regexes(self):
"""Test stacktrace.stack_blacklist_regexes in project.yaml works."""
def _mock_config_get(_, param):
"""Handle test configuration options."""
if param == 'stacktrace.stack_blacklist_regexes':
return [r'.*[c]{3}']
return None
test_helpers.patch(self, ['config.local_config.ProjectConfig.get'])
self.mock.get.side_effect = _mock_config_get
crash_stacktrace = ('aaa\nbbbbbbb\nzzzccc\nddd\n\n')
self.assertTrue(crash_analyzer.ignore_stacktrace(crash_stacktrace))
crash_stacktrace = ('aaa\nbbbbbbb\nddd\n\n')
self.assertFalse(crash_analyzer.ignore_stacktrace(crash_stacktrace))
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.
unit_path = self._quarantine_unit(unit_path, quarantine_corpus_path)
num_bad_units += 1
continue
result = self._run_single_unit(unit_path)
if (not result.timed_out and
not crash_analyzer.is_memory_tool_crash(result.output)):
# Didn't crash or time out.
continue
if result.timed_out:
# Slow unit. Quarantine it.
unit_path = self._quarantine_unit(unit_path, quarantine_corpus_path)
num_bad_units += 1
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):
"""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)
# Setup testcase and get absolute testcase path.
file_list, _, testcase_file_path = setup.setup_testcase(testcase)
if not file_list:
return
# Set up a custom or regular build based on revision.
build_manager.setup_build(testcase.crash_revision)
# Check if we have an application path. If not, our build failed
# to setup correctly.
app_path = environment.get_value('APP_PATH')
if not 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:
close_invalid_testcase_and_update_status(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)
# 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
# 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)
close_invalid_testcase_and_update_status(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 http flag and re-run testcase to store dependencies (for bundled
# archives only).
testcase.http_flag = http_flag
if not store_testcase_dependencies_from_bundled_testcase_archive(
metadata, testcase, testcase_file_path):
return
# Update testcase crash parameters.
testcase.crash_type = state.crash_type
testcase.crash_address = state.crash_address
testcase.crash_state = state.crash_state
# 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):
close_invalid_testcase_and_update_status(testcase, metadata, 'Irrelavant')
return
# Test for reproducibility.
one_time_crasher_flag = not testcase_manager.test_for_reproducibility(
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.
project_name = data_handler.get_project_name(job_type)
existing_testcase = data_handler.find_testcase(
project_name, state.crash_type, state.crash_state, security_flag)
if existing_testcase:
# If the existing test case is unreproducible and we are, replace the
# existing test case with this one.
if existing_testcase.one_time_crasher_flag and not one_time_crasher_flag:
duplicate_testcase = existing_testcase
original_testcase = testcase
else:
duplicate_testcase = testcase
original_testcase = existing_testcase
metadata.status = 'Duplicate'
metadata.duplicate_of = existing_testcase.key.id()
duplicate_testcase.status = 'Duplicate'
duplicate_testcase.duplicate_of = original_testcase.key.id()
duplicate_testcase.put()
# Set testcase and metadata status if not set already.
if testcase.status != 'Duplicate':
testcase.status = 'Processed'
metadata.status = 'Confirmed'
# 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)
# Add application specific information in the trace.
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)
# Update the testcase values.
testcase.put()
# Update the upload metadata.
metadata.security_flag = security_flag
metadata.put()
# 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 crashers (stack).
task_creation.create_tasks(testcase)
def should_ignore(self):
"""Return True if this crash should be ignored."""
state = self.get_symbolized_data()
return crash_analyzer.ignore_stacktrace(state.crash_stacktrace)
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
at com.example.foo.CrashyClass.crash(CrashyClass.java:6)
at com.bugsnag.android.example.ExampleActivity.crashUnhandled(ExampleActivity.kt:55)
at com.bugsnag.android.example.ExampleActivity$onCreate$1.invoke(ExampleActivity.kt:33)
at com.bugsnag.android.example.ExampleActivity$onCreate$1.invoke(ExampleActivity.kt:14)
at com.bugsnag.android.example.ExampleActivity$sam$android_view_View_OnClickListener$0.onClick(ExampleActivity.kt)
at android.view.View.performClick(View.java:5637)
at android.view.View$PerformClick.run(View.java:22429)
at android.os.Handler.handleCallback(Handler.java:751)
at android.os.Handler.dispatchMessage(Handler.java:95)
at android.os.Looper.loop(Looper.java:154)
at android.app.ActivityThread.main(ActivityThread.java:6119)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:886)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:776)"""
application_command_line = ""
state = stack_analyzer.get_crash_data(unsymbolized_crash_stacktrace,
symbolize_flag=False,
already_symbolized=True)
security_flag = crash_analyzer.is_security_issue(unsymbolized_crash_stacktrace,
state.crash_type,
state.crash_address)
key = '%s,%s,%s' % (state.crash_type, state.crash_state, security_flag)
should_be_ignored = crash_analyzer.ignore_stacktrace(state.crash_stacktrace)
print("key:", key)
# sample output:
# key: Fatal Exception,com.example.foo.CrashyClass.sendMessage
# com.example.foo.CrashyClass.crash
# com.bugsnag.android.example.ExampleActivity.crashUnhandled
# ,False