# Unix SMB/CIFS implementation. # Copyright (C) Jelmer Vernooij 2007-2010 # Copyright (C) Stefan Metzmacher 2014,2015 # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation; either version 3 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program. If not, see . # """Samba Python tests.""" import os import tempfile import traceback import collections import ldb import samba from samba import param from samba import credentials from samba.credentials import Credentials import subprocess import sys import unittest import re from enum import IntEnum, unique import samba.auth import samba.gensec import samba.dcerpc.base from random import randint from random import SystemRandom from contextlib import contextmanager import shutil import string try: from samba.samdb import SamDB except ImportError: # We are built without samdb support, # imitate it so that connect_samdb() can recover def SamDB(*args, **kwargs): return None import samba.ndr import samba.dcerpc.dcerpc import samba.dcerpc.epmapper from unittest import SkipTest BINDIR = os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../bin")) HEXDUMP_FILTER = bytearray([x if ((len(repr(chr(x))) == 3) and (x < 127)) else ord('.') for x in range(256)]) LDB_ERR_LUT = {v: k for k, v in vars(ldb).items() if k.startswith('ERR_')} RE_CAMELCASE = re.compile(r"([_\-])+") def ldb_err(v): if isinstance(v, ldb.LdbError): v = v.args[0] if v in LDB_ERR_LUT: return LDB_ERR_LUT[v] try: return f"[{', '.join(LDB_ERR_LUT.get(x, x) for x in v)}]" except TypeError as e: print(e) return v def DynamicTestCase(cls): cls.setUpDynamicTestCases() return cls class TestCase(unittest.TestCase): """A Samba test case.""" # Re-implement addClassCleanup to support Python versions older than 3.8. # Can be removed once these older Python versions are no longer needed. if sys.version_info.major == 3 and sys.version_info.minor < 8: _class_cleanups = [] @classmethod def addClassCleanup(cls, function, *args, **kwargs): cls._class_cleanups.append((function, args, kwargs)) @classmethod def tearDownClass(cls): teardown_exceptions = [] while cls._class_cleanups: function, args, kwargs = cls._class_cleanups.pop() try: function(*args, **kwargs) except Exception: teardown_exceptions.append(traceback.format_exc()) # ExceptionGroup would be better but requires Python 3.11 if teardown_exceptions: raise ValueError("tearDownClass failed:\n\n" + "\n".join(teardown_exceptions)) @classmethod def setUpClass(cls): """ Call setUpTestData, ensure tearDownClass is called on exceptions. This is only required on Python versions older than 3.8. """ try: cls.setUpTestData() except Exception: cls.tearDownClass() raise else: @classmethod def setUpClass(cls): """ setUpClass only needs to call setUpTestData. On Python 3.8 and above unittest will always call tearDownClass, even if an exception was raised in setUpClass. """ cls.setUpTestData() @classmethod def setUpTestData(cls): """Create class level test fixtures here.""" pass @classmethod def generate_dynamic_test(cls, fnname, suffix, *args, doc=None): """ fnname is something like "test_dynamic_sum" suffix is something like "1plus2" argstr could be (1, 2) This would generate a test case called "test_dynamic_sum_1plus2(self)" that calls self._test_dynamic_sum_with_args(1, 2) """ def fn(self): getattr(self, "_%s_with_args" % fnname)(*args) fn.__doc__ = doc attr = "%s_%s" % (fnname, suffix) if hasattr(cls, attr): raise RuntimeError(f"Dynamic test {attr} already exists!") setattr(cls, attr, fn) @classmethod def setUpDynamicTestCases(cls): """This can be implemented in order to call cls.generate_dynamic_test() In order to implement autogenerated testcase permutations. """ msg = "%s needs setUpDynamicTestCases() if @DynamicTestCase is used!" % (cls) raise NotImplementedError(msg) def unique_name(self): """Generate a unique name from within a test for creating objects. Used to ensure each test generates uniquely named objects that don't interfere with other tests. """ # name of calling function name = self.id().rsplit(".", 1)[1] # remove test_ prefix if name.startswith("test_"): name = name[5:] # finally, convert to camelcase name = RE_CAMELCASE.sub(" ", name).title().replace(" ", "") return "".join([name[0].lower(), name[1:]]) def setUp(self): super().setUp() test_debug_level = os.getenv("TEST_DEBUG_LEVEL") if test_debug_level is not None: test_debug_level = int(test_debug_level) self._old_debug_level = samba.get_debug_level() samba.set_debug_level(test_debug_level) self.addCleanup(samba.set_debug_level, test_debug_level) @classmethod def get_loadparm(cls, s3=False): return env_loadparm(s3=s3) def get_credentials(self): return cmdline_credentials @classmethod def get_env_credentials(cls, *, lp, env_username, env_password, env_realm=None, env_domain=None): creds = credentials.Credentials() # guess Credentials parameters here. Otherwise, workstation # and domain fields are NULL and gencache code segfaults creds.guess(lp) creds.set_username(env_get_var_value(env_username)) creds.set_password(env_get_var_value(env_password)) if env_realm is not None: creds.set_realm(env_get_var_value(env_realm)) if env_domain is not None: creds.set_domain(env_get_var_value(env_domain)) return creds def get_creds_ccache_name(self): creds = self.get_credentials() ccache = creds.get_named_ccache(self.get_loadparm()) ccache_name = ccache.get_name() return ccache_name def hexdump(self, src): N = 0 result = '' is_string = isinstance(src, str) while src: ll = src[:8] lr = src[8:16] src = src[16:] if is_string: hl = ' '.join(["%02X" % ord(x) for x in ll]) hr = ' '.join(["%02X" % ord(x) for x in lr]) ll = ll.translate(HEXDUMP_FILTER) lr = lr.translate(HEXDUMP_FILTER) else: hl = ' '.join(["%02X" % x for x in ll]) hr = ' '.join(["%02X" % x for x in lr]) ll = ll.translate(HEXDUMP_FILTER).decode('utf8') lr = lr.translate(HEXDUMP_FILTER).decode('utf8') result += "[%04X] %-*s %-*s %s %s\n" % (N, 8 * 3, hl, 8 * 3, hr, ll, lr) N += 16 return result def insta_creds(self, template=None, username=None, userpass=None, kerberos_state=None): if template is None: raise ValueError("you need to supply a Credentials template") if username is not None and userpass is None: raise ValueError( "you cannot set creds username without setting a password") if username is None: assert userpass is None username = template.get_username() userpass = template.get_password() simple_bind_dn = template.get_bind_dn() if kerberos_state is None: kerberos_state = template.get_kerberos_state() # get a copy of the global creds or the passed in creds c = Credentials() c.set_username(username) c.set_password(userpass) c.set_domain(template.get_domain()) c.set_realm(template.get_realm()) c.set_workstation(template.get_workstation()) c.set_gensec_features(c.get_gensec_features() | samba.gensec.FEATURE_SEAL) c.set_kerberos_state(kerberos_state) if simple_bind_dn: c.set_bind_dn(simple_bind_dn) return c def assertStringsEqual(self, a, b, msg=None, strip=False): """Assert equality between two strings and highlight any differences. If strip is true, leading and trailing whitespace is ignored.""" if strip: a = a.strip() b = b.strip() if a != b: sys.stderr.write("The strings differ %s(lengths %d vs %d); " "a diff follows\n" % ('when stripped ' if strip else '', len(a), len(b), )) from difflib import unified_diff diff = unified_diff(a.splitlines(True), b.splitlines(True), 'a', 'b') for line in diff: sys.stderr.write(line) self.fail(msg) def assertRaisesLdbError(self, errcode, message, f, *args, **kwargs): """Assert a function raises a particular LdbError.""" if message is None: message = f"{f.__name__}(*{args}, **{kwargs})" try: f(*args, **kwargs) except ldb.LdbError as e: (num, msg) = e.args if isinstance(errcode, collections.abc.Container): found = num in errcode else: found = num == errcode if not found: lut = {v: k for k, v in vars(ldb).items() if k.startswith('ERR_') and isinstance(v, int)} if isinstance(errcode, collections.abc.Container): errcode_name = ' '.join(lut.get(x) for x in errcode) else: errcode_name = lut.get(errcode) self.fail(f"{message}, expected " f"LdbError {errcode_name}, {errcode} " f"got {lut.get(num)} ({num}) " f"{msg}") else: lut = {v: k for k, v in vars(ldb).items() if k.startswith('ERR_') and isinstance(v, int)} if isinstance(errcode, collections.abc.Container): errcode_name = ' '.join(lut.get(x) for x in errcode) else: errcode_name = lut.get(errcode) self.fail("%s, expected " "LdbError %s, (%s) " "but we got success" % (message, errcode_name, errcode)) class LdbTestCase(TestCase): """Trivial test case for running tests against a LDB.""" def setUp(self): super().setUp() self.tempfile = tempfile.NamedTemporaryFile(delete=False) self.filename = self.tempfile.name self.ldb = samba.Ldb(self.filename) def set_modules(self, modules=None): """Change the modules for this Ldb.""" if modules is None: modules = [] m = ldb.Message() m.dn = ldb.Dn(self.ldb, "@MODULES") m["@LIST"] = ",".join(modules) self.ldb.add(m) self.ldb = samba.Ldb(self.filename) class TestCaseInTempDir(TestCase): def setUp(self): super().setUp() self.tempdir = tempfile.mkdtemp() self.addCleanup(self._remove_tempdir) def _remove_tempdir(self): # Note asserting here is treated as an error rather than a test failure self.assertEqual([], os.listdir(self.tempdir)) os.rmdir(self.tempdir) self.tempdir = None @contextmanager def mktemp(self): """Yield a temporary filename in the tempdir.""" try: fd, fn = tempfile.mkstemp(dir=self.tempdir) yield fn finally: try: os.close(fd) os.unlink(fn) except (OSError, IOError) as e: print("could not remove temporary file: %s" % e, file=sys.stderr) def rm_files(self, *files, allow_missing=False, _rm=os.remove): """Remove listed files from the temp directory. The files must be true files in the directory itself, not in sub-directories. By default a non-existent file will cause a test failure (or error if used outside a test in e.g. tearDown), but if allow_missing is true, the absence will be ignored. """ for f in files: path = os.path.join(self.tempdir, f) # os.path.join will happily step out of the tempdir, # so let's just check. if os.path.dirname(path) != self.tempdir: raise ValueError(f"{path} might be outside {self.tempdir}") try: _rm(path) except FileNotFoundError as e: if not allow_missing: raise AssertionError(f"{f} not in {self.tempdir}: {e}") print(f"{f} not in {self.tempdir}") def rm_dirs(self, *dirs, allow_missing=False): """Remove listed directories from temp directory. This works like rm_files, but only removes directories, including their contents. """ self.rm_files(*dirs, allow_missing=allow_missing, _rm=shutil.rmtree) def env_loadparm(s3=False): if s3: from samba.samba3 import param as s3param lp = s3param.get_context() else: lp = param.LoadParm() try: lp.load(os.environ["SMB_CONF_PATH"]) except KeyError: raise KeyError("SMB_CONF_PATH not set") return lp def env_get_var_value(var_name, allow_missing=False): """Returns value for variable in os.environ Function throws AssertionError if variable is undefined. Unit-test based python tests require certain input params to be set in environment, otherwise they can't be run """ if allow_missing: if var_name not in os.environ.keys(): return None assert var_name in os.environ.keys(), "Please supply %s in environment" % var_name return os.environ[var_name] cmdline_credentials = None class RpcInterfaceTestCase(TestCase): """DCE/RPC Test case.""" class BlackboxProcessError(Exception): """This is raised when check_output() process returns a non-zero exit status Exception instance should contain the exact exit code (S.returncode), command line (S.cmd), process output (S.stdout) and process error stream (S.stderr) """ def __init__(self, returncode, cmd, stdout, stderr, msg=None): self.returncode = returncode if isinstance(cmd, list): self.cmd = ' '.join(cmd) self.shell = False else: self.cmd = cmd self.shell = True self.stdout = stdout self.stderr = stderr self.msg = msg def __str__(self): s = ("Command '%s'; shell %s; exit status %d; " "stdout: '%s'; stderr: '%s'" % (self.cmd, self.shell, self.returncode, self.stdout, self.stderr)) if self.msg is not None: s = "%s; message: %s" % (s, self.msg) return s class BlackboxTestCase(TestCaseInTempDir): """Base test case for blackbox tests.""" @staticmethod def _make_cmdline(line): """Expand the called script into a fully resolved path in the bin directory.""" if isinstance(line, list): parts = line else: parts = line.split(" ", 1) cmd = parts[0] exe = os.path.join(BINDIR, cmd) python_cmds = ["samba-tool", "samba_dnsupdate", "samba_upgradedns", "script/traffic_replay", "script/traffic_learner"] if os.path.exists(exe): parts[0] = exe if cmd in python_cmds and os.getenv("PYTHON", False): parts.insert(0, os.environ["PYTHON"]) if not isinstance(line, list): line = " ".join(parts) return line @classmethod def check_run(cls, line, msg=None): cls.check_exit_code(line, 0, msg=msg) @classmethod def check_exit_code(cls, line, expected, msg=None): line = cls._make_cmdline(line) use_shell = not isinstance(line, list) p = subprocess.Popen(line, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=use_shell) stdoutdata, stderrdata = p.communicate() retcode = p.returncode if retcode != expected: if msg is None: msg = "expected return code %s; got %s" % (expected, retcode) raise BlackboxProcessError(retcode, line, stdoutdata, stderrdata, msg) return stdoutdata @classmethod def check_output(cls, line): use_shell = not isinstance(line, list) line = cls._make_cmdline(line) p = subprocess.Popen(line, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=use_shell, close_fds=True) stdoutdata, stderrdata = p.communicate() retcode = p.returncode if retcode: raise BlackboxProcessError(retcode, line, stdoutdata, stderrdata) return stdoutdata # # Run a command without checking the return code, returns the tuple # (ret, stdout, stderr) # where ret is the return code # stdout is a string containing the commands stdout # stderr is a string containing the commands stderr @classmethod def run_command(cls, line): line = cls._make_cmdline(line) use_shell = not isinstance(line, list) p = subprocess.Popen(line, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=use_shell) stdoutdata, stderrdata = p.communicate() retcode = p.returncode return (retcode, stdoutdata.decode('UTF8'), stderrdata.decode('UTF8')) # Generate a random password that can be safely passed on the command line # i.e. it does not contain any shell meta characters. def random_password(self, count=32): password = SystemRandom().choice(string.ascii_uppercase) password += SystemRandom().choice(string.digits) password += SystemRandom().choice(string.ascii_lowercase) password += ''.join(SystemRandom().choice(string.ascii_uppercase + string.ascii_lowercase + string.digits) for x in range(count - 3)) return password def connect_samdb(samdb_url, *, lp=None, session_info=None, credentials=None, flags=0, ldb_options=None, ldap_only=False, global_schema=True): """Create SamDB instance and connects to samdb_url database. :param samdb_url: Url for database to connect to. :param lp: Optional loadparm object :param session_info: Optional session information :param credentials: Optional credentials, defaults to anonymous. :param flags: Optional LDB flags :param ldap_only: If set, only remote LDAP connection will be created. :param global_schema: Whether to use global schema. Added value for tests is that we have a shorthand function to make proper URL for ldb.connect() while using default parameters for connection based on test environment """ if "://" not in samdb_url: if not ldap_only and os.path.isfile(samdb_url): samdb_url = "tdb://%s" % samdb_url else: samdb_url = "ldap://%s" % samdb_url # use 'paged_search' module when connecting remotely if samdb_url.startswith("ldap://"): ldb_options = ["modules:paged_searches"] elif ldap_only: raise AssertionError("Trying to connect to %s while remote " "connection is required" % samdb_url) # set defaults for test environment if lp is None: lp = env_loadparm() if session_info is None: session_info = samba.auth.system_session(lp) if credentials is None: credentials = cmdline_credentials return SamDB(url=samdb_url, lp=lp, session_info=session_info, credentials=credentials, flags=flags, options=ldb_options, global_schema=global_schema) def connect_samdb_ex(samdb_url, *, lp=None, session_info=None, credentials=None, flags=0, ldb_options=None, ldap_only=False): """Connects to samdb_url database :param samdb_url: Url for database to connect to. :param lp: Optional loadparm object :param session_info: Optional session information :param credentials: Optional credentials, defaults to anonymous. :param flags: Optional LDB flags :param ldap_only: If set, only remote LDAP connection will be created. :return: (sam_db_connection, rootDse_record) tuple """ sam_db = connect_samdb(samdb_url, lp=lp, session_info=session_info, credentials=credentials, flags=flags, ldb_options=ldb_options, ldap_only=ldap_only) # fetch RootDse res = sam_db.search(base="", expression="", scope=ldb.SCOPE_BASE, attrs=["*"]) return (sam_db, res[0]) def connect_samdb_env(env_url, env_username, env_password, lp=None): """Connect to SamDB by getting URL and Credentials from environment :param env_url: Environment variable name to get lsb url from :param env_username: Username environment variable :param env_password: Password environment variable :return: sam_db_connection """ samdb_url = env_get_var_value(env_url) creds = credentials.Credentials() if lp is None: # guess Credentials parameters here. Otherwise workstation # and domain fields are NULL and gencache code segfaults lp = param.LoadParm() creds.guess(lp) creds.set_username(env_get_var_value(env_username)) creds.set_password(env_get_var_value(env_password)) return connect_samdb(samdb_url, credentials=creds, lp=lp) def delete_force(samdb, dn, **kwargs): try: samdb.delete(dn, **kwargs) except ldb.LdbError as error: (num, errstr) = error.args assert num == ldb.ERR_NO_SUCH_OBJECT, "ldb.delete() failed: %s" % errstr def create_test_ou(samdb, name): """Creates a unique OU for the test""" # Add some randomness to the test OU. Replication between the testenvs is # constantly happening in the background. Deletion of the last test's # objects can be slow to replicate out. So the OU created by a previous # testenv may still exist at the point that tests start on another testenv. rand = randint(1, 10000000) dn = ldb.Dn(samdb, "OU=%s%d,%s" % (name, rand, samdb.get_default_basedn())) samdb.add({"dn": dn, "objectclass": "organizationalUnit"}) return dn @unique class OptState(IntEnum): NOOPT = 0 HYPHEN1 = 1 HYPHEN2 = 2 NAME = 3 def parse_help_consistency(out, options_start=None, options_end=None, optmap=None, max_leading_spaces=10): if options_start is None: opt_lines = [] else: opt_lines = None for raw_line in out.split('\n'): line = raw_line.lstrip() if line == '': continue if opt_lines is None: if line == options_start: opt_lines = [] else: continue if len(line) < len(raw_line) - max_leading_spaces: # for the case where we have: # # --foo frobnicate or barlify depending on # --bar option. # # where we want to ignore the --bar. continue if line[0] == '-': opt_lines.append(line) if line == options_end: break if opt_lines is None: # No --help options is not an error in *this* test. return is_longname_char = re.compile(r'^[\w-]$').match for line in opt_lines: state = OptState.NOOPT name = None prev = ' ' for c in line: if state == OptState.NOOPT: if c == '-' and prev.isspace(): state = OptState.HYPHEN1 prev = c continue if state == OptState.HYPHEN1: if c.isalnum(): name = '-' + c state = OptState.NAME elif c == '-': state = OptState.HYPHEN2 continue if state == OptState.HYPHEN2: if c.isalnum(): name = '--' + c state = OptState.NAME else: # WTF, perhaps '--' ending option list. state = OptState.NOOPT prev = c continue if state == OptState.NAME: if is_longname_char(c): name += c else: optmap.setdefault(name, []).append(line) state = OptState.NOOPT prev = c if state == OptState.NAME: optmap.setdefault(name, []).append(line) def check_help_consistency(out, options_start=None, options_end=None): """Ensure that options are not repeated and redefined in --help output. Returns None if everything is OK, otherwise a string indicating the problems. If options_start and/or options_end are provided, only the bit in the output between these two lines is considered. For example, with samba-tool, options_start='Options:', options_end='Available subcommands:' will prevent the test looking at the preamble which may contain examples using options. """ # Silly test, you might think, but this happens optmap = {} parse_help_consistency(out, options_start, options_end, optmap) errors = [] for k, values in sorted(optmap.items()): if len(values) > 1: for v in values: errors.append("%s: %s" % (k, v)) if errors: return "\n".join(errors) def get_env_dir(key): """A helper to pull a directory name from the environment, used in some tests that optionally write e.g. fuzz seeds into a directory named in an environment variable. """ dir = os.environ.get(key) if dir is None: return None if not os.path.isdir(dir): raise ValueError( f"{key} should name an existing directory (got '{dir}')") return dir