Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(253)

Unified Diff: build/android/pylib/linker/test_case.py

Issue 2101243005: Add a snapshot of flutter/engine/src/build to our sdk (Closed) Base URL: git@github.com:dart-lang/sdk.git@master
Patch Set: add README.dart Created 4 years, 6 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « build/android/pylib/linker/setup.py ('k') | build/android/pylib/linker/test_runner.py » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: build/android/pylib/linker/test_case.py
diff --git a/build/android/pylib/linker/test_case.py b/build/android/pylib/linker/test_case.py
new file mode 100644
index 0000000000000000000000000000000000000000..c7b0f50b59cc9f97440df027398cd31186008bd8
--- /dev/null
+++ b/build/android/pylib/linker/test_case.py
@@ -0,0 +1,496 @@
+# Copyright 2013 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""Base class for linker-specific test cases.
+
+ The custom dynamic linker can only be tested through a custom test case
+ for various technical reasons:
+
+ - It's an 'invisible feature', i.e. it doesn't expose a new API or
+ behaviour, all it does is save RAM when loading native libraries.
+
+ - Checking that it works correctly requires several things that do not
+ fit the existing GTest-based and instrumentation-based tests:
+
+ - Native test code needs to be run in both the browser and renderer
+ process at the same time just after loading native libraries, in
+ a completely asynchronous way.
+
+ - Each test case requires restarting a whole new application process
+ with a different command-line.
+
+ - Enabling test support in the Linker code requires building a special
+ APK with a flag to activate special test-only support code in the
+ Linker code itself.
+
+ Host-driven tests have also been tried, but since they're really
+ sub-classes of instrumentation tests, they didn't work well either.
+
+ To build and run the linker tests, do the following:
+
+ ninja -C out/Debug chromium_linker_test_apk
+ build/android/test_runner.py linker
+
+"""
+# pylint: disable=R0201
+
+import logging
+import os
+import re
+import time
+
+from pylib import constants
+from pylib.base import base_test_result
+from pylib.device import device_errors
+from pylib.device import intent
+
+
+ResultType = base_test_result.ResultType
+
+_PACKAGE_NAME = 'org.chromium.chromium_linker_test_apk'
+_ACTIVITY_NAME = '.ChromiumLinkerTestActivity'
+_COMMAND_LINE_FILE = '/data/local/tmp/chromium-linker-test-command-line'
+
+# Path to the Linker.java source file.
+_LINKER_JAVA_SOURCE_PATH = (
+ 'base/android/java/src/org/chromium/base/library_loader/Linker.java')
+
+# A regular expression used to extract the browser shared RELRO configuration
+# from the Java source file above.
+_RE_LINKER_BROWSER_CONFIG = re.compile(
+ r'.*BROWSER_SHARED_RELRO_CONFIG\s+=\s+' +
+ r'BROWSER_SHARED_RELRO_CONFIG_(\S+)\s*;.*',
+ re.MULTILINE | re.DOTALL)
+
+# Logcat filters used during each test. Only the 'chromium' one is really
+# needed, but the logs are added to the TestResult in case of error, and
+# it is handy to have the 'chromium_android_linker' ones as well when
+# troubleshooting.
+_LOGCAT_FILTERS = ['*:s', 'chromium:v', 'chromium_android_linker:v']
+#_LOGCAT_FILTERS = ['*:v'] ## DEBUG
+
+# Regular expression used to match status lines in logcat.
+_RE_BROWSER_STATUS_LINE = re.compile(r' BROWSER_LINKER_TEST: (FAIL|SUCCESS)$')
+_RE_RENDERER_STATUS_LINE = re.compile(r' RENDERER_LINKER_TEST: (FAIL|SUCCESS)$')
+
+# Regular expression used to mach library load addresses in logcat.
+_RE_LIBRARY_ADDRESS = re.compile(
+ r'(BROWSER|RENDERER)_LIBRARY_ADDRESS: (\S+) ([0-9A-Fa-f]+)')
+
+
+def _GetBrowserSharedRelroConfig():
+ """Returns a string corresponding to the Linker's configuration of shared
+ RELRO sections in the browser process. This parses the Java linker source
+ file to get the appropriate information.
+ Return:
+ None in case of error (e.g. could not locate the source file).
+ 'NEVER' if the browser process shall never use shared RELROs.
+ 'LOW_RAM_ONLY' if if uses it only on low-end devices.
+ 'ALWAYS' if it always uses a shared RELRO.
+ """
+ source_path = \
+ os.path.join(constants.DIR_SOURCE_ROOT, _LINKER_JAVA_SOURCE_PATH)
+ if not os.path.exists(source_path):
+ logging.error('Could not find linker source file: ' + source_path)
+ return None
+
+ with open(source_path) as f:
+ configs = _RE_LINKER_BROWSER_CONFIG.findall(f.read())
+ if not configs:
+ logging.error(
+ 'Can\'t find browser shared RELRO configuration value in ' + \
+ source_path)
+ return None
+
+ if configs[0] not in ['NEVER', 'LOW_RAM_ONLY', 'ALWAYS']:
+ logging.error('Unexpected browser config value: ' + configs[0])
+ return None
+
+ logging.info('Found linker browser shared RELRO config: ' + configs[0])
+ return configs[0]
+
+
+def _StartActivityAndWaitForLinkerTestStatus(device, timeout):
+ """Force-start an activity and wait up to |timeout| seconds until the full
+ linker test status lines appear in the logcat, recorded through |device|.
+ Args:
+ device: A DeviceUtils instance.
+ timeout: Timeout in seconds
+ Returns:
+ A (status, logs) tuple, where status is a ResultType constant, and logs
+ if the final logcat output as a string.
+ """
+
+ # 1. Start recording logcat with appropriate filters.
+ with device.GetLogcatMonitor(filter_specs=_LOGCAT_FILTERS) as logmon:
+
+ # 2. Force-start activity.
+ device.StartActivity(
+ intent.Intent(package=_PACKAGE_NAME, activity=_ACTIVITY_NAME),
+ force_stop=True)
+
+ # 3. Wait up to |timeout| seconds until the test status is in the logcat.
+ result = ResultType.PASS
+ try:
+ browser_match = logmon.WaitFor(_RE_BROWSER_STATUS_LINE, timeout=timeout)
+ logging.debug('Found browser match: %s', browser_match.group(0))
+ renderer_match = logmon.WaitFor(_RE_RENDERER_STATUS_LINE,
+ timeout=timeout)
+ logging.debug('Found renderer match: %s', renderer_match.group(0))
+ if (browser_match.group(1) != 'SUCCESS'
+ or renderer_match.group(1) != 'SUCCESS'):
+ result = ResultType.FAIL
+ except device_errors.CommandTimeoutError:
+ result = ResultType.TIMEOUT
+
+ return result, '\n'.join(device.adb.Logcat(dump=True))
+
+
+class LibraryLoadMap(dict):
+ """A helper class to pretty-print a map of library names to load addresses."""
+ def __str__(self):
+ items = ['\'%s\': 0x%x' % (name, address) for \
+ (name, address) in self.iteritems()]
+ return '{%s}' % (', '.join(items))
+
+ def __repr__(self):
+ return 'LibraryLoadMap(%s)' % self.__str__()
+
+
+class AddressList(list):
+ """A helper class to pretty-print a list of load addresses."""
+ def __str__(self):
+ items = ['0x%x' % address for address in self]
+ return '[%s]' % (', '.join(items))
+
+ def __repr__(self):
+ return 'AddressList(%s)' % self.__str__()
+
+
+def _ExtractLibraryLoadAddressesFromLogcat(logs):
+ """Extract the names and addresses of shared libraries loaded in the
+ browser and renderer processes.
+ Args:
+ logs: A string containing logcat output.
+ Returns:
+ A tuple (browser_libs, renderer_libs), where each item is a map of
+ library names (strings) to library load addresses (ints), for the
+ browser and renderer processes, respectively.
+ """
+ browser_libs = LibraryLoadMap()
+ renderer_libs = LibraryLoadMap()
+ for m in _RE_LIBRARY_ADDRESS.finditer(logs):
+ process_type, lib_name, lib_address = m.groups()
+ lib_address = int(lib_address, 16)
+ if process_type == 'BROWSER':
+ browser_libs[lib_name] = lib_address
+ elif process_type == 'RENDERER':
+ renderer_libs[lib_name] = lib_address
+ else:
+ assert False, 'Invalid process type'
+
+ return browser_libs, renderer_libs
+
+
+def _CheckLoadAddressRandomization(lib_map_list, process_type):
+ """Check that a map of library load addresses is random enough.
+ Args:
+ lib_map_list: a list of dictionaries that map library names (string)
+ to load addresses (int). Each item in the list corresponds to a
+ different run / process start.
+ process_type: a string describing the process type.
+ Returns:
+ (status, logs) tuple, where <status> is True iff the load addresses are
+ randomized, False otherwise, and <logs> is a string containing an error
+ message detailing the libraries that are not randomized properly.
+ """
+ # Collect, for each library, its list of load addresses.
+ lib_addr_map = {}
+ for lib_map in lib_map_list:
+ for lib_name, lib_address in lib_map.iteritems():
+ if lib_name not in lib_addr_map:
+ lib_addr_map[lib_name] = AddressList()
+ lib_addr_map[lib_name].append(lib_address)
+
+ logging.info('%s library load map: %s', process_type, lib_addr_map)
+
+ # For each library, check the randomness of its load addresses.
+ bad_libs = {}
+ for lib_name, lib_address_list in lib_addr_map.iteritems():
+ # If all addresses are different, skip to next item.
+ lib_address_set = set(lib_address_list)
+ # Consider that if there is more than one pair of identical addresses in
+ # the list, then randomization is broken.
+ if len(lib_address_set) < len(lib_address_list) - 1:
+ bad_libs[lib_name] = lib_address_list
+
+
+ if bad_libs:
+ return False, '%s libraries failed randomization: %s' % \
+ (process_type, bad_libs)
+
+ return True, '%s libraries properly randomized: %s' % \
+ (process_type, lib_addr_map)
+
+
+class LinkerTestCaseBase(object):
+ """Base class for linker test cases."""
+
+ def __init__(self, is_low_memory=False):
+ """Create a test case.
+ Args:
+ is_low_memory: True to simulate a low-memory device, False otherwise.
+ """
+ self.is_low_memory = is_low_memory
+ if is_low_memory:
+ test_suffix = 'ForLowMemoryDevice'
+ else:
+ test_suffix = 'ForRegularDevice'
+ class_name = self.__class__.__name__
+ self.qualified_name = '%s.%s' % (class_name, test_suffix)
+ self.tagged_name = self.qualified_name
+
+ def _RunTest(self, _device):
+ """Run the test, must be overriden.
+ Args:
+ _device: A DeviceUtils interface.
+ Returns:
+ A (status, log) tuple, where <status> is a ResultType constant, and <log>
+ is the logcat output captured during the test in case of error, or None
+ in case of success.
+ """
+ return ResultType.FAIL, 'Unimplemented _RunTest() method!'
+
+ def Run(self, device):
+ """Run the test on a given device.
+ Args:
+ device: Name of target device where to run the test.
+ Returns:
+ A base_test_result.TestRunResult() instance.
+ """
+ margin = 8
+ print '[ %-*s ] %s' % (margin, 'RUN', self.tagged_name)
+ logging.info('Running linker test: %s', self.tagged_name)
+
+ # Create command-line file on device.
+ command_line_flags = ''
+ if self.is_low_memory:
+ command_line_flags = '--low-memory-device'
+ device.WriteFile(_COMMAND_LINE_FILE, command_line_flags)
+
+ # Run the test.
+ status, logs = self._RunTest(device)
+
+ result_text = 'OK'
+ if status == ResultType.FAIL:
+ result_text = 'FAILED'
+ elif status == ResultType.TIMEOUT:
+ result_text = 'TIMEOUT'
+ print '[ %*s ] %s' % (margin, result_text, self.tagged_name)
+
+ results = base_test_result.TestRunResults()
+ results.AddResult(
+ base_test_result.BaseTestResult(
+ self.tagged_name,
+ status,
+ log=logs))
+
+ return results
+
+ def __str__(self):
+ return self.tagged_name
+
+ def __repr__(self):
+ return self.tagged_name
+
+
+class LinkerSharedRelroTest(LinkerTestCaseBase):
+ """A linker test case to check the status of shared RELRO sections.
+
+ The core of the checks performed here are pretty simple:
+
+ - Clear the logcat and start recording with an appropriate set of filters.
+ - Create the command-line appropriate for the test-case.
+ - Start the activity (always forcing a cold start).
+ - Every second, look at the current content of the filtered logcat lines
+ and look for instances of the following:
+
+ BROWSER_LINKER_TEST: <status>
+ RENDERER_LINKER_TEST: <status>
+
+ where <status> can be either FAIL or SUCCESS. These lines can appear
+ in any order in the logcat. Once both browser and renderer status are
+ found, stop the loop. Otherwise timeout after 30 seconds.
+
+ Note that there can be other lines beginning with BROWSER_LINKER_TEST:
+ and RENDERER_LINKER_TEST:, but are not followed by a <status> code.
+
+ - The test case passes if the <status> for both the browser and renderer
+ process are SUCCESS. Otherwise its a fail.
+ """
+ def _RunTest(self, device):
+ # Wait up to 30 seconds until the linker test status is in the logcat.
+ return _StartActivityAndWaitForLinkerTestStatus(device, timeout=30)
+
+
+class LinkerLibraryAddressTest(LinkerTestCaseBase):
+ """A test case that verifies library load addresses.
+
+ The point of this check is to ensure that the libraries are loaded
+ according to the following rules:
+
+ - For low-memory devices, they should always be loaded at the same address
+ in both browser and renderer processes, both below 0x4000_0000.
+
+ - For regular devices, the browser process should load libraries above
+ 0x4000_0000, and renderer ones below it.
+ """
+ def _RunTest(self, device):
+ result, logs = _StartActivityAndWaitForLinkerTestStatus(device, timeout=30)
+
+ # Return immediately in case of timeout.
+ if result == ResultType.TIMEOUT:
+ return result, logs
+
+ # Collect the library load addresses in the browser and renderer processes.
+ browser_libs, renderer_libs = _ExtractLibraryLoadAddressesFromLogcat(logs)
+
+ logging.info('Browser libraries: %s', browser_libs)
+ logging.info('Renderer libraries: %s', renderer_libs)
+
+ # Check that the same libraries are loaded into both processes:
+ browser_set = set(browser_libs.keys())
+ renderer_set = set(renderer_libs.keys())
+ if browser_set != renderer_set:
+ logging.error('Library set mistmach browser=%s renderer=%s',
+ browser_libs.keys(), renderer_libs.keys())
+ return ResultType.FAIL, logs
+
+ # And that there are not empty.
+ if not browser_set:
+ logging.error('No libraries loaded in any process!')
+ return ResultType.FAIL, logs
+
+ # Check that the renderer libraries are loaded at 'low-addresses'. i.e.
+ # below 0x4000_0000, for every kind of device.
+ memory_boundary = 0x40000000
+ bad_libs = []
+ for lib_name, lib_address in renderer_libs.iteritems():
+ if lib_address >= memory_boundary:
+ bad_libs.append((lib_name, lib_address))
+
+ if bad_libs:
+ logging.error('Renderer libraries loaded at high addresses: %s', bad_libs)
+ return ResultType.FAIL, logs
+
+ browser_config = _GetBrowserSharedRelroConfig()
+ if not browser_config:
+ return ResultType.FAIL, 'Bad linker source configuration'
+
+ if browser_config == 'ALWAYS' or \
+ (browser_config == 'LOW_RAM_ONLY' and self.is_low_memory):
+ # The libraries must all be loaded at the same addresses. This also
+ # implicitly checks that the browser libraries are at low addresses.
+ addr_mismatches = []
+ for lib_name, lib_address in browser_libs.iteritems():
+ lib_address2 = renderer_libs[lib_name]
+ if lib_address != lib_address2:
+ addr_mismatches.append((lib_name, lib_address, lib_address2))
+
+ if addr_mismatches:
+ logging.error('Library load address mismatches: %s',
+ addr_mismatches)
+ return ResultType.FAIL, logs
+
+ # Otherwise, check that libraries are loaded at 'high-addresses'.
+ # Note that for low-memory devices, the previous checks ensure that they
+ # were loaded at low-addresses.
+ else:
+ bad_libs = []
+ for lib_name, lib_address in browser_libs.iteritems():
+ if lib_address < memory_boundary:
+ bad_libs.append((lib_name, lib_address))
+
+ if bad_libs:
+ logging.error('Browser libraries loaded at low addresses: %s', bad_libs)
+ return ResultType.FAIL, logs
+
+ # Everything's ok.
+ return ResultType.PASS, logs
+
+
+class LinkerRandomizationTest(LinkerTestCaseBase):
+ """A linker test case to check that library load address randomization works
+ properly between successive starts of the test program/activity.
+
+ This starts the activity several time (each time forcing a new process
+ creation) and compares the load addresses of the libraries in them to
+ detect that they have changed.
+
+ In theory, two successive runs could (very rarely) use the same load
+ address, so loop 5 times and compare the values there. It is assumed
+ that if there are more than one pair of identical addresses, then the
+ load addresses are not random enough for this test.
+ """
+ def _RunTest(self, device):
+ max_loops = 5
+ browser_lib_map_list = []
+ renderer_lib_map_list = []
+ logs_list = []
+ for _ in range(max_loops):
+ # Start the activity.
+ result, logs = _StartActivityAndWaitForLinkerTestStatus(
+ device, timeout=30)
+ if result == ResultType.TIMEOUT:
+ # Something bad happened. Return immediately.
+ return result, logs
+
+ # Collect library addresses.
+ browser_libs, renderer_libs = _ExtractLibraryLoadAddressesFromLogcat(logs)
+ browser_lib_map_list.append(browser_libs)
+ renderer_lib_map_list.append(renderer_libs)
+ logs_list.append(logs)
+
+ # Check randomization in the browser libraries.
+ logs = '\n'.join(logs_list)
+
+ browser_status, browser_logs = _CheckLoadAddressRandomization(
+ browser_lib_map_list, 'Browser')
+
+ renderer_status, renderer_logs = _CheckLoadAddressRandomization(
+ renderer_lib_map_list, 'Renderer')
+
+ browser_config = _GetBrowserSharedRelroConfig()
+ if not browser_config:
+ return ResultType.FAIL, 'Bad linker source configuration'
+
+ if not browser_status:
+ if browser_config == 'ALWAYS' or \
+ (browser_config == 'LOW_RAM_ONLY' and self.is_low_memory):
+ return ResultType.FAIL, browser_logs
+
+ # IMPORTANT NOTE: The system's ASLR implementation seems to be very poor
+ # when starting an activity process in a loop with "adb shell am start".
+ #
+ # When simulating a regular device, loading libraries in the browser
+ # process uses a simple mmap(NULL, ...) to let the kernel device where to
+ # load the file (this is similar to what System.loadLibrary() does).
+ #
+ # Unfortunately, at least in the context of this test, doing so while
+ # restarting the activity with the activity manager very, very, often
+ # results in the system using the same load address for all 5 runs, or
+ # sometimes only 4 out of 5.
+ #
+ # This has been tested experimentally on both Android 4.1.2 and 4.3.
+ #
+ # Note that this behaviour doesn't seem to happen when starting an
+ # application 'normally', i.e. when using the application launcher to
+ # start the activity.
+ logging.info('Ignoring system\'s low randomization of browser libraries' +
+ ' for regular devices')
+
+ if not renderer_status:
+ return ResultType.FAIL, renderer_logs
+
+ return ResultType.PASS, logs
« no previous file with comments | « build/android/pylib/linker/setup.py ('k') | build/android/pylib/linker/test_runner.py » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698