diff options
Diffstat (limited to 'test/cavp.py')
-rw-r--r-- | test/cavp.py | 70 |
1 files changed, 40 insertions, 30 deletions
diff --git a/test/cavp.py b/test/cavp.py index 059bcd2..2508bda 100644 --- a/test/cavp.py +++ b/test/cavp.py @@ -51,11 +51,11 @@ class _TestVectorsFile: self._archive = archive self._path = path self._fn = os.path.split(path)[1] - self._valid = False + self._recognized = False self._parse() - def valid(self): - return self._valid + def recognized(self): + return self._recognized def algorithm(self): return self._algo @@ -64,10 +64,11 @@ class _TestVectorsFile: return self._mode def parse(self): - self._parser = configparser.ConfigParser(dict_type=_MultiOrderedDict, - strict=False, - interpolation=None, - empty_lines_in_values=False) + self._parser = configparser.ConfigParser( + dict_type=_MultiOrderedDict, + strict=False, + interpolation=None, + empty_lines_in_values=False) self._parser.read_string(self._archive.read(self._path).decode('utf-8')) def _extract_test_data(self, section): @@ -80,16 +81,11 @@ class _TestVectorsFile: return keys, plaintexts, ciphertexts, init_vectors def _run_tests(self, tool, inputs, expected_output): - try: - for expected_output_chunk, input_chunk in _split_into_chunks(expected_output, list(inputs)): - actual_output = tool(self.algorithm(), self.mode(), input_chunk) - if not _assert_output(actual_output, expected_output_chunk): - return _TestExitCode.FAILURE - return _TestExitCode.SUCCESS - except toolkit.ToolkitError as e: - logging.error('Encountered an exception, skipping...') - logging.exception(e) - return _TestExitCode.ERROR + for expected_output_chunk, input_chunk in _split_into_chunks(expected_output, list(inputs)): + actual_output = tool(self.algorithm(), self.mode(), input_chunk) + if not _assert_output(actual_output, expected_output_chunk): + return _TestExitCode.FAILURE + return _TestExitCode.SUCCESS def run_encryption_tests(self, tools): logging.info('Running encryption tests...') @@ -113,7 +109,7 @@ class _TestVectorsFile: if not stub: return stub = self._strip_mode(stub) if not stub: return - self._valid = True + self._recognized = True def _strip_extension(self, stub): stub, ext = os.path.splitext(stub) @@ -125,12 +121,12 @@ class _TestVectorsFile: def _strip_algorithm(self, stub): algo_size = stub[-3:] maybe_algo = 'aes{0}'.format(algo_size) - self._algo = toolkit.to_supported_algorithm(maybe_algo) + self._algo = toolkit.is_algorithm_supported(maybe_algo) if self._algo: logging.info('\tAlgorithm: {0}'.format(self._algo)) return stub[0:-3] else: - logging.warn('Unknown or unsupported algorithm \'{0}\''.format(self._fn)) + logging.warn('Unknown or unsupported algorithm: ' + self._fn) return None def _strip_method(self, stub): @@ -138,26 +134,36 @@ class _TestVectorsFile: if stub.endswith(method): logging.info('\tMethod: {0}'.format(method)) return stub[0:len(stub) - len(method)] - logging.warn('Unknown or unsupported method \'{0}\''.format(self._fn)) + logging.warn('Unknown or unsupported method: ' + self._fn) def _strip_mode(self, stub): - self._mode = toolkit.to_supported_mode(stub) + self._mode = toolkit.is_mode_supported(stub) if self._mode: logging.info('\tMode: {0}'.format(self._mode)) return self._mode else: - logging.warn('Unknown or unsupported mode \'{0}\''.format(self._fn)) + logging.warn('Unknown or unsupported mode: ' + self._fn) return None -def _parse_test_vectors_archive(tools, archive_path='KAT_AES.zip'): +def _parse_archive_and_run_tests(tools, archive_path): archive = zipfile.ZipFile(archive_path) exit_codes = [] for fn in archive.namelist(): member = _TestVectorsFile(fn, archive) - if member.valid(): + if member.recognized(): member.parse() - exit_codes.append(member.run_encryption_tests(tools)) - exit_codes.append(member.run_decryption_tests(tools)) + try: + exit_codes.append(member.run_encryption_tests(tools)) + except Exception as e: + logging.error('Encountered an exception!') + logging.exception(e) + exit_codes.append(_TestExitCode.ERROR) + try: + exit_codes.append(member.run_decryption_tests(tools)) + except Exception as e: + logging.error('Encountered an exception!') + logging.exception(e) + exit_codes.append(_TestExitCode.ERROR) else: exit_codes.append(_TestExitCode.SKIPPED) logging.info('Test exit codes:') @@ -180,11 +186,15 @@ if __name__ == '__main__': help='use Intel SDE to run *.exe files') parser.add_argument('--box', '-b', action='store_true', help='use the "boxes" interface') + parser.add_argument('--archive', '-a', default='KAT_AES.zip', + help='set path of the archive with the test vectors') parser.add_argument('--log', '-l', help='set log file path') args = parser.parse_args() - logging_options = {'format': '%(asctime)s | %(module)s | %(levelname)s | %(message)s', - 'level': logging.DEBUG} + logging_options = { + 'format': '%(asctime)s | %(module)s | %(levelname)s | %(message)s', + 'level': logging.DEBUG } + if args.log is None: logging_options['filename'] = datetime.now().strftime('cavp_%Y-%m-%d_%H-%M-%S.log') else: @@ -192,4 +202,4 @@ if __name__ == '__main__': logging.basicConfig(**logging_options) tools = toolkit.Tools(args.path, use_sde=args.sde, use_boxes=args.box) - _parse_test_vectors_archive(tools) + _parse_archive_and_run_tests(tools, args.archive) |