Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 1 | #!/usr/bin/env python |
| 2 | # Copyright 2014 The Chromium Authors. All rights reserved. |
| 3 | # Use of this source code is governed by a BSD-style license that can be |
| 4 | # found in the LICENSE file. |
| 5 | |
| 6 | """A utility script for downloading versioned Syzygy binaries.""" |
| 7 | |
| 8 | import hashlib |
| 9 | import errno |
| 10 | import json |
| 11 | import logging |
| 12 | import optparse |
| 13 | import os |
| 14 | import re |
| 15 | import shutil |
| 16 | import stat |
| 17 | import sys |
| 18 | import subprocess |
| 19 | import tempfile |
| 20 | import time |
| 21 | import zipfile |
| 22 | |
| 23 | |
| 24 | _LOGGER = logging.getLogger(os.path.basename(__file__)) |
| 25 | |
| 26 | # The relative path where official builds are archived in their GS bucket. |
| 27 | _SYZYGY_ARCHIVE_PATH = ('/builds/official/%(revision)s') |
| 28 | |
| 29 | # A JSON file containing the state of the download directory. If this file and |
| 30 | # directory state do not agree, then the binaries will be downloaded and |
| 31 | # installed again. |
| 32 | _STATE = '.state' |
| 33 | |
| 34 | # This matches an integer (an SVN revision number) or a SHA1 value (a GIT hash). |
| 35 | # The archive exclusively uses lowercase GIT hashes. |
| 36 | _REVISION_RE = re.compile('^(?:\d+|[a-f0-9]{40})$') |
| 37 | |
| 38 | # This matches an MD5 hash. |
| 39 | _MD5_RE = re.compile('^[a-f0-9]{32}$') |
| 40 | |
| 41 | # List of reources to be downloaded and installed. These are tuples with the |
| 42 | # following format: |
| 43 | # (basename, logging name, relative installation path, extraction filter) |
| 44 | _RESOURCES = [ |
| 45 | ('benchmark.zip', 'benchmark', '', None), |
| 46 | ('binaries.zip', 'binaries', 'exe', None), |
| 47 | ('symbols.zip', 'symbols', 'exe', |
| 48 | lambda x: x.filename.endswith('.dll.pdb'))] |
| 49 | |
| 50 | |
| 51 | def _LoadState(output_dir): |
| 52 | """Loads the contents of the state file for a given |output_dir|, returning |
| 53 | None if it doesn't exist. |
| 54 | """ |
| 55 | path = os.path.join(output_dir, _STATE) |
| 56 | if not os.path.exists(path): |
| 57 | _LOGGER.debug('No state file found.') |
| 58 | return None |
| 59 | with open(path, 'rb') as f: |
| 60 | _LOGGER.debug('Reading state file: %s', path) |
| 61 | try: |
| 62 | return json.load(f) |
| 63 | except ValueError: |
| 64 | _LOGGER.debug('Invalid state file.') |
| 65 | return None |
| 66 | |
| 67 | |
| 68 | def _SaveState(output_dir, state, dry_run=False): |
| 69 | """Saves the |state| dictionary to the given |output_dir| as a JSON file.""" |
| 70 | path = os.path.join(output_dir, _STATE) |
| 71 | _LOGGER.debug('Writing state file: %s', path) |
| 72 | if dry_run: |
| 73 | return |
| 74 | with open(path, 'wb') as f: |
| 75 | f.write(json.dumps(state, sort_keys=True, indent=2)) |
| 76 | |
| 77 | |
| 78 | def _Md5(path): |
| 79 | """Returns the MD5 hash of the file at |path|, which must exist.""" |
| 80 | return hashlib.md5(open(path, 'rb').read()).hexdigest() |
| 81 | |
| 82 | |
| 83 | def _StateIsValid(state): |
| 84 | """Returns true if the given state structure is valid.""" |
| 85 | if not isinstance(state, dict): |
| 86 | _LOGGER.debug('State must be a dict.') |
| 87 | return False |
| 88 | r = state.get('revision', None) |
| 89 | if not isinstance(r, basestring) or not _REVISION_RE.match(r): |
| 90 | _LOGGER.debug('State contains an invalid revision.') |
| 91 | return False |
| 92 | c = state.get('contents', None) |
| 93 | if not isinstance(c, dict): |
| 94 | _LOGGER.debug('State must contain a contents dict.') |
| 95 | return False |
| 96 | for (relpath, md5) in c.iteritems(): |
| 97 | if not isinstance(relpath, basestring) or len(relpath) == 0: |
| 98 | _LOGGER.debug('State contents dict contains an invalid path.') |
| 99 | return False |
| 100 | if not isinstance(md5, basestring) or not _MD5_RE.match(md5): |
| 101 | _LOGGER.debug('State contents dict contains an invalid MD5 digest.') |
| 102 | return False |
| 103 | return True |
| 104 | |
| 105 | |
| 106 | def _BuildActualState(stored, revision, output_dir): |
| 107 | """Builds the actual state using the provided |stored| state as a template. |
| 108 | Only examines files listed in the stored state, causing the script to ignore |
| 109 | files that have been added to the directories locally. |stored| must be a |
| 110 | valid state dictionary. |
| 111 | """ |
| 112 | contents = {} |
| 113 | state = { 'revision': revision, 'contents': contents } |
| 114 | for relpath, md5 in stored['contents'].iteritems(): |
| 115 | abspath = os.path.abspath(os.path.join(output_dir, relpath)) |
| 116 | if os.path.isfile(abspath): |
| 117 | m = _Md5(abspath) |
| 118 | contents[relpath] = m |
| 119 | |
| 120 | return state |
| 121 | |
| 122 | |
| 123 | def _StatesAreConsistent(stored, actual): |
| 124 | """Validates whether two state dictionaries are consistent. Both must be valid |
| 125 | state dictionaries. Additional entries in |actual| are ignored. |
| 126 | """ |
| 127 | if stored['revision'] != actual['revision']: |
| 128 | _LOGGER.debug('Mismatched revision number.') |
| 129 | return False |
| 130 | cont_stored = stored['contents'] |
| 131 | cont_actual = actual['contents'] |
| 132 | for relpath, md5 in cont_stored.iteritems(): |
| 133 | if relpath not in cont_actual: |
| 134 | _LOGGER.debug('Missing content: %s', relpath) |
| 135 | return False |
| 136 | if md5 != cont_actual[relpath]: |
| 137 | _LOGGER.debug('Modified content: %s', relpath) |
| 138 | return False |
| 139 | return True |
| 140 | |
| 141 | |
| 142 | def _GetCurrentState(revision, output_dir): |
| 143 | """Loads the current state and checks to see if it is consistent. Returns |
| 144 | a tuple (state, bool). The returned state will always be valid, even if an |
| 145 | invalid state is present on disk. |
| 146 | """ |
| 147 | stored = _LoadState(output_dir) |
| 148 | if not _StateIsValid(stored): |
| 149 | _LOGGER.debug('State is invalid.') |
| 150 | # Return a valid but empty state. |
| 151 | return ({'revision': '0', 'contents': {}}, False) |
| 152 | actual = _BuildActualState(stored, revision, output_dir) |
| 153 | # If the script has been modified consider the state invalid. |
| 154 | path = os.path.join(output_dir, _STATE) |
| 155 | if os.path.getmtime(__file__) > os.path.getmtime(path): |
| 156 | return (stored, False) |
| 157 | # Otherwise, explicitly validate the state. |
| 158 | if not _StatesAreConsistent(stored, actual): |
| 159 | return (stored, False) |
| 160 | return (stored, True) |
| 161 | |
| 162 | |
| 163 | def _DirIsEmpty(path): |
| 164 | """Returns true if the given directory is empty, false otherwise.""" |
| 165 | for root, dirs, files in os.walk(path): |
| 166 | return not dirs and not files |
| 167 | |
| 168 | |
| 169 | def _RmTreeHandleReadOnly(func, path, exc): |
| 170 | """An error handling function for use with shutil.rmtree. This will |
| 171 | detect failures to remove read-only files, and will change their properties |
| 172 | prior to removing them. This is necessary on Windows as os.remove will return |
| 173 | an access error for read-only files, and git repos contain read-only |
| 174 | pack/index files. |
| 175 | """ |
| 176 | excvalue = exc[1] |
| 177 | if func in (os.rmdir, os.remove) and excvalue.errno == errno.EACCES: |
| 178 | _LOGGER.debug('Removing read-only path: %s', path) |
| 179 | os.chmod(path, stat.S_IRWXU | stat.S_IRWXG | stat.S_IRWXO) |
| 180 | func(path) |
| 181 | else: |
| 182 | raise |
| 183 | |
| 184 | |
| 185 | def _RmTree(path): |
| 186 | """A wrapper of shutil.rmtree that handles read-only files.""" |
| 187 | shutil.rmtree(path, ignore_errors=False, onerror=_RmTreeHandleReadOnly) |
| 188 | |
| 189 | |
| 190 | def _CleanState(output_dir, state, dry_run=False): |
| 191 | """Cleans up files/directories in |output_dir| that are referenced by |
| 192 | the given |state|. Raises an error if there are local changes. Returns a |
| 193 | dictionary of files that were deleted. |
| 194 | """ |
| 195 | _LOGGER.debug('Deleting files from previous installation.') |
| 196 | deleted = {} |
| 197 | |
| 198 | # Generate a list of files to delete, relative to |output_dir|. |
| 199 | contents = state['contents'] |
| 200 | files = sorted(contents.keys()) |
| 201 | |
| 202 | # Try to delete the files. Keep track of directories to delete as well. |
| 203 | dirs = {} |
| 204 | for relpath in files: |
| 205 | fullpath = os.path.join(output_dir, relpath) |
| 206 | fulldir = os.path.dirname(fullpath) |
| 207 | dirs[fulldir] = True |
| 208 | if os.path.exists(fullpath): |
| 209 | # If somehow the file has become a directory complain about it. |
| 210 | if os.path.isdir(fullpath): |
| 211 | raise Exception('Directory exists where file expected: %s' % fullpath) |
| 212 | |
| 213 | # Double check that the file doesn't have local changes. If it does |
| 214 | # then refuse to delete it. |
| 215 | if relpath in contents: |
| 216 | stored_md5 = contents[relpath] |
| 217 | actual_md5 = _Md5(fullpath) |
| 218 | if actual_md5 != stored_md5: |
| 219 | raise Exception('File has local changes: %s' % fullpath) |
| 220 | |
| 221 | # The file is unchanged so it can safely be deleted. |
| 222 | _LOGGER.debug('Deleting file "%s".', fullpath) |
| 223 | deleted[relpath] = True |
| 224 | if not dry_run: |
| 225 | os.unlink(fullpath) |
| 226 | |
| 227 | # Sort directories from longest name to shortest. This lets us remove empty |
| 228 | # directories from the most nested paths first. |
| 229 | dirs = sorted(dirs.keys(), key=lambda x: len(x), reverse=True) |
| 230 | for p in dirs: |
| 231 | if os.path.exists(p) and _DirIsEmpty(p): |
| 232 | _LOGGER.debug('Deleting empty directory "%s".', p) |
| 233 | if not dry_run: |
| 234 | _RmTree(p) |
| 235 | |
| 236 | return deleted |
| 237 | |
| 238 | |
| 239 | def _FindGsUtil(): |
| 240 | """Looks for depot_tools and returns the absolute path to gsutil.py.""" |
| 241 | for path in os.environ['PATH'].split(os.pathsep): |
| 242 | path = os.path.abspath(path) |
| 243 | git_cl = os.path.join(path, 'git_cl.py') |
| 244 | gs_util = os.path.join(path, 'gsutil.py') |
| 245 | if os.path.exists(git_cl) and os.path.exists(gs_util): |
| 246 | return gs_util |
| 247 | return None |
| 248 | |
| 249 | |
| 250 | def _GsUtil(*cmd): |
| 251 | """Runs the given command in gsutil with exponential backoff and retries.""" |
| 252 | gs_util = _FindGsUtil() |
| 253 | cmd = [sys.executable, gs_util] + list(cmd) |
| 254 | |
| 255 | retries = 3 |
| 256 | timeout = 4 # Seconds. |
| 257 | while True: |
| 258 | _LOGGER.debug('Running %s', cmd) |
| 259 | prog = subprocess.Popen(cmd, shell=False) |
| 260 | prog.communicate() |
| 261 | |
| 262 | # Stop retrying on success. |
| 263 | if prog.returncode == 0: |
| 264 | return |
| 265 | |
| 266 | # Raise a permanent failure if retries have been exhausted. |
| 267 | if retries == 0: |
| 268 | raise RuntimeError('Command "%s" returned %d.' % (cmd, prog.returncode)) |
| 269 | |
| 270 | _LOGGER.debug('Sleeping %d seconds and trying again.', timeout) |
| 271 | time.sleep(timeout) |
| 272 | retries -= 1 |
| 273 | timeout *= 2 |
| 274 | |
| 275 | |
| 276 | def _Download(resource): |
| 277 | """Downloads the given GS resource to a temporary file, returning its path.""" |
| 278 | tmp = tempfile.mkstemp(suffix='syzygy_archive') |
| 279 | os.close(tmp[0]) |
| 280 | url = 'gs://syzygy-archive' + resource |
| 281 | _GsUtil('cp', url, tmp[1]) |
| 282 | return tmp[1] |
| 283 | |
| 284 | |
| 285 | def _InstallBinaries(options, deleted={}): |
| 286 | """Installs Syzygy binaries. This assumes that the output directory has |
| 287 | already been cleaned, as it will refuse to overwrite existing files.""" |
| 288 | contents = {} |
| 289 | state = { 'revision': options.revision, 'contents': contents } |
| 290 | archive_path = _SYZYGY_ARCHIVE_PATH % { 'revision': options.revision } |
| 291 | if options.resources: |
| 292 | resources = [(resource, resource, '', None) |
| 293 | for resource in options.resources] |
| 294 | else: |
| 295 | resources = _RESOURCES |
| 296 | for (base, name, subdir, filt) in resources: |
| 297 | # Create the output directory if it doesn't exist. |
| 298 | fulldir = os.path.join(options.output_dir, subdir) |
| 299 | if os.path.isfile(fulldir): |
| 300 | raise Exception('File exists where a directory needs to be created: %s' % |
| 301 | fulldir) |
| 302 | if not os.path.exists(fulldir): |
| 303 | _LOGGER.debug('Creating directory: %s', fulldir) |
| 304 | if not options.dry_run: |
| 305 | os.makedirs(fulldir) |
| 306 | |
| 307 | # Download and read the archive. |
| 308 | resource = archive_path + '/' + base |
| 309 | _LOGGER.debug('Retrieving %s archive at "%s".', name, resource) |
| 310 | path = _Download(resource) |
| 311 | |
| 312 | _LOGGER.debug('Unzipping %s archive.', name) |
| 313 | with open(path, 'rb') as data: |
| 314 | archive = zipfile.ZipFile(data) |
| 315 | for entry in archive.infolist(): |
| 316 | if not filt or filt(entry): |
| 317 | fullpath = os.path.normpath(os.path.join(fulldir, entry.filename)) |
| 318 | relpath = os.path.relpath(fullpath, options.output_dir) |
| 319 | if os.path.exists(fullpath): |
| 320 | # If in a dry-run take into account the fact that the file *would* |
| 321 | # have been deleted. |
| 322 | if options.dry_run and relpath in deleted: |
| 323 | pass |
| 324 | else: |
| 325 | raise Exception('Path already exists: %s' % fullpath) |
| 326 | |
| 327 | # Extract the file and update the state dictionary. |
| 328 | _LOGGER.debug('Extracting "%s".', fullpath) |
| 329 | if not options.dry_run: |
| 330 | archive.extract(entry.filename, fulldir) |
| 331 | md5 = _Md5(fullpath) |
| 332 | contents[relpath] = md5 |
| 333 | if sys.platform == 'cygwin': |
| 334 | os.chmod(fullpath, os.stat(fullpath).st_mode | stat.S_IXUSR) |
| 335 | |
| 336 | _LOGGER.debug('Removing temporary file "%s".', path) |
| 337 | os.remove(path) |
| 338 | |
| 339 | return state |
| 340 | |
| 341 | |
| 342 | def _ParseCommandLine(): |
| 343 | """Parses the command-line and returns an options structure.""" |
| 344 | option_parser = optparse.OptionParser() |
| 345 | option_parser.add_option('--dry-run', action='store_true', default=False, |
| 346 | help='If true then will simply list actions that would be performed.') |
| 347 | option_parser.add_option('--force', action='store_true', default=False, |
| 348 | help='Force an installation even if the binaries are up to date.') |
| 349 | option_parser.add_option('--no-cleanup', action='store_true', default=False, |
| 350 | help='Allow installation on non-Windows platforms, and skip the forced ' |
| 351 | 'cleanup step.') |
| 352 | option_parser.add_option('--output-dir', type='string', |
| 353 | help='The path where the binaries will be replaced. Existing binaries ' |
| 354 | 'will only be overwritten if not up to date.') |
| 355 | option_parser.add_option('--overwrite', action='store_true', default=False, |
| 356 | help='If specified then the installation will happily delete and rewrite ' |
| 357 | 'the entire output directory, blasting any local changes.') |
| 358 | option_parser.add_option('--revision', type='string', |
| 359 | help='The SVN revision or GIT hash associated with the required version.') |
| 360 | option_parser.add_option('--revision-file', type='string', |
| 361 | help='A text file containing an SVN revision or GIT hash.') |
| 362 | option_parser.add_option('--resource', type='string', action='append', |
| 363 | dest='resources', help='A resource to be downloaded.') |
| 364 | option_parser.add_option('--verbose', dest='log_level', action='store_const', |
| 365 | default=logging.INFO, const=logging.DEBUG, |
| 366 | help='Enables verbose logging.') |
| 367 | option_parser.add_option('--quiet', dest='log_level', action='store_const', |
| 368 | default=logging.INFO, const=logging.ERROR, |
| 369 | help='Disables all output except for errors.') |
| 370 | options, args = option_parser.parse_args() |
| 371 | if args: |
| 372 | option_parser.error('Unexpected arguments: %s' % args) |
| 373 | if not options.output_dir: |
| 374 | option_parser.error('Must specify --output-dir.') |
| 375 | if not options.revision and not options.revision_file: |
| 376 | option_parser.error('Must specify one of --revision or --revision-file.') |
| 377 | if options.revision and options.revision_file: |
| 378 | option_parser.error('Must not specify both --revision and --revision-file.') |
| 379 | |
| 380 | # Configure logging. |
| 381 | logging.basicConfig(level=options.log_level) |
| 382 | |
| 383 | # If a revision file has been specified then read it. |
| 384 | if options.revision_file: |
| 385 | options.revision = open(options.revision_file, 'rb').read().strip() |
| 386 | _LOGGER.debug('Parsed revision "%s" from file "%s".', |
| 387 | options.revision, options.revision_file) |
| 388 | |
| 389 | # Ensure that the specified SVN revision or GIT hash is valid. |
| 390 | if not _REVISION_RE.match(options.revision): |
| 391 | option_parser.error('Must specify a valid SVN or GIT revision.') |
| 392 | |
| 393 | # This just makes output prettier to read. |
| 394 | options.output_dir = os.path.normpath(options.output_dir) |
| 395 | |
| 396 | return options |
| 397 | |
| 398 | |
| 399 | def _RemoveOrphanedFiles(options): |
| 400 | """This is run on non-Windows systems to remove orphaned files that may have |
| 401 | been downloaded by a previous version of this script. |
| 402 | """ |
| 403 | # Reconfigure logging to output info messages. This will allow inspection of |
| 404 | # cleanup status on non-Windows buildbots. |
| 405 | _LOGGER.setLevel(logging.INFO) |
| 406 | |
| 407 | output_dir = os.path.abspath(options.output_dir) |
| 408 | |
| 409 | # We only want to clean up the folder in 'src/third_party/syzygy', and we |
| 410 | # expect to be called with that as an output directory. This is an attempt to |
| 411 | # not start deleting random things if the script is run from an alternate |
| 412 | # location, or not called from the gclient hooks. |
| 413 | expected_syzygy_dir = os.path.abspath(os.path.join( |
| 414 | os.path.dirname(__file__), '..', 'third_party', 'syzygy')) |
| 415 | expected_output_dir = os.path.join(expected_syzygy_dir, 'binaries') |
| 416 | if expected_output_dir != output_dir: |
| 417 | _LOGGER.info('Unexpected output directory, skipping cleanup.') |
| 418 | return |
| 419 | |
| 420 | if not os.path.isdir(expected_syzygy_dir): |
| 421 | _LOGGER.info('Output directory does not exist, skipping cleanup.') |
| 422 | return |
| 423 | |
| 424 | def OnError(function, path, excinfo): |
| 425 | """Logs error encountered by shutil.rmtree.""" |
| 426 | _LOGGER.error('Error when running %s(%s)', function, path, exc_info=excinfo) |
| 427 | |
| 428 | _LOGGER.info('Removing orphaned files from %s', expected_syzygy_dir) |
| 429 | if not options.dry_run: |
| 430 | shutil.rmtree(expected_syzygy_dir, True, OnError) |
| 431 | |
| 432 | |
| 433 | def main(): |
| 434 | options = _ParseCommandLine() |
| 435 | |
| 436 | if options.dry_run: |
| 437 | _LOGGER.debug('Performing a dry-run.') |
| 438 | |
| 439 | # We only care about Windows platforms, as the Syzygy binaries aren't used |
| 440 | # elsewhere. However, there was a short period of time where this script |
| 441 | # wasn't gated on OS types, and those OSes downloaded and installed binaries. |
| 442 | # This will cleanup orphaned files on those operating systems. |
| 443 | if sys.platform not in ('win32', 'cygwin'): |
| 444 | if options.no_cleanup: |
| 445 | _LOGGER.debug('Skipping usual cleanup for non-Windows platforms.') |
| 446 | else: |
| 447 | return _RemoveOrphanedFiles(options) |
| 448 | |
| 449 | # Load the current installation state, and validate it against the |
| 450 | # requested installation. |
| 451 | state, is_consistent = _GetCurrentState(options.revision, options.output_dir) |
| 452 | |
| 453 | # Decide whether or not an install is necessary. |
| 454 | if options.force: |
| 455 | _LOGGER.debug('Forcing reinstall of binaries.') |
| 456 | elif is_consistent: |
| 457 | # Avoid doing any work if the contents of the directory are consistent. |
| 458 | _LOGGER.debug('State unchanged, no reinstall necessary.') |
| 459 | return |
| 460 | |
| 461 | # Under normal logging this is the only only message that will be reported. |
| 462 | _LOGGER.info('Installing revision %s Syzygy binaries.', |
| 463 | options.revision[0:12]) |
| 464 | |
| 465 | # Clean up the old state to begin with. |
| 466 | deleted = [] |
| 467 | if options.overwrite: |
| 468 | if os.path.exists(options.output_dir): |
| 469 | # If overwrite was specified then take a heavy-handed approach. |
| 470 | _LOGGER.debug('Deleting entire installation directory.') |
| 471 | if not options.dry_run: |
| 472 | _RmTree(options.output_dir) |
| 473 | else: |
| 474 | # Otherwise only delete things that the previous installation put in place, |
| 475 | # and take care to preserve any local changes. |
| 476 | deleted = _CleanState(options.output_dir, state, options.dry_run) |
| 477 | |
| 478 | # Install the new binaries. In a dry-run this will actually download the |
| 479 | # archives, but it won't write anything to disk. |
| 480 | state = _InstallBinaries(options, deleted) |
| 481 | |
| 482 | # Build and save the state for the directory. |
| 483 | _SaveState(options.output_dir, state, options.dry_run) |
| 484 | |
| 485 | |
| 486 | if __name__ == '__main__': |
| 487 | main() |