# Copyright (C) 2010 by the Massachusetts Institute of Technology.

# All rights reserved.

# Export of this software from the United States of America may

#   require a specific license from the United States Government.

#   It is the responsibility of any person or organization contemplating

#   export to obtain such a license before exporting.

#

# WITHIN THAT CONSTRAINT, permission to use, copy, modify, and

# distribute this software and its documentation for any purpose and

# without fee is hereby granted, provided that the above copyright

# notice appear in all copies and that both that copyright notice and

# this permission notice appear in supporting documentation, and that

# the name of M.I.T. not be used in advertising or publicity pertaining

# to distribution of the software without specific, written prior

# permission.  Furthermore if you modify this software you must label

# your software as modified software and not distribute it in such a

# fashion that it might be confused with the original M.I.T. software.

# M.I.T. makes no representations about the suitability of

# this software for any purpose.  It is provided "as is" without express

# or implied warranty.

"""A module for krb5 test scripts

To run test scripts during "make check" (if Python 2.5 or later is

available), add rules like the following to Makefile.in:

    check-pytests::

	$(RUNPYTEST) $(srcdir)/t_testname.py $(PYTESTFLAGS)

A sample test script:

    from k5test import *

    # Run a test program under a variety of configurations:

    for realm in multipass_realms():

        realm.run(['./testprog', 'arg'])

    # Run a test server and client under just the default configuration:

    realm = K5Realm()

    realm.start_server(['./serverprog'], 'starting...')

    realm.run(['./clientprog', realm.host_princ])

    # Inform framework that tests completed successfully.

    success('World peace and cure for cancer')

By default, the realm will have:

* The name KRBTEST.COM

* Listener ports starting at 61000

* krb5.conf and kdc.conf files

* A fresh DB2 KDB

* Running krb5kdc (but not kadmind)

* Principals named realm.user_princ and realm.admin_princ; call

  password('user') and password('admin') to get the password

* Credentials for realm.user_princ in realm.ccache

* Admin rights for realm.admin_princ in the kadmind acl file

* A host principal named realm.host_princ with a random key

* A keytab for the host principal in realm.keytab

The realm's behaviour can be modified with the following constructor

keyword arguments:

* realm='realmname': Override the realm name

* portbase=NNN: Override the listener port base; currently three ports are

  used

* testdir='dirname': Override the storage area for the realm's files

  (path may be specified relative to the current working dir)

* krb5_conf={ ... }: krb5.conf options, expressed as a nested

  dictionary, to be merged with the default krb5.conf settings.  A key

  may be mapped to None to delete a setting from the defaults.  A key

  may be mapped to a list in order to create multiple settings for the

  same variable name.  Keys and values undergo the following template

  substitutions:

    - $realm:    The realm name

    - $testdir:  The realm storage directory (absolute path)

    - $buildtop: The root of the build directory

    - $srctop:   The root of the source directory

    - $plugins:  The plugin directory in the build tree

    - $hostname: The FQDN of the host

    - $port0:    The first listener port (portbase)

    - ...

    - $port9:    The tenth listener port (portbase + 9)

  When choosing ports, note the following:

    - port0 is used in the default krb5.conf for the KDC

    - port1 is used in the default krb5.conf for kadmind

    - port2 is used in the default krb5.conf for kpasswd

    - port3 is used in the default krb5.conf for kpropd

    - port4 is used in the default krb5.conf for iprop (in kadmind)

    - port5 is the return value of realm.server_port()

* kdc_conf={...}: kdc.conf options, expressed as a nested dictionary,

  to be merged with the default kdc.conf settings.  The same

  conventions and substitutions for krb5_conf apply.

* create_kdb=False: Don't create a KDB.  Implicitly disables all of

  the other options since they all require a KDB.

* krbtgt_keysalt='enctype:salttype': After creating the KDB,

  regenerate the krbtgt key using the specified key/salt combination,

  using a kadmin.local cpw query.

* create_user=False: Don't create the user principal.  Implies

  get_creds=False.

* create_host=False: Don't create the host principal or the associated

  keytab.

* start_kdc=False: Don't start the KDC.  Implies get_creds=False.

* start_kadmind=True: Start kadmind.

* get_creds=False: Don't get user credentials.

* bdb_only=True: Use the DB2 KDB module even if K5TEST_LMDB is set in

  the environment.

Scripts may use the following functions and variables:

* fail(message): Display message (plus leading marker and trailing

  newline) and explanatory messages about debugging.

* success(message): Indicate that the test script has completed

  successfully.  Suppresses the display of explanatory debugging

  messages in the on-exit handler.  message should briefly summarize

  the operations tested; it will only be displayed (with leading

  marker and trailing newline) if the script is running verbosely.

* skipped(whatmsg, whymsg): Indicate that some tests were skipped.

  whatmsg should concisely say what was skipped (e.g. "LDAP KDB

  tests") and whymsg should give the reason (e.g. "because LDAP module

  not built").

* skip_rest(message): Indicate that some tests were skipped, then exit

  the current script.

* output(message, force_verbose=False): Place message (without any

  added newline) in testlog, and write it to stdout if running

  verbosely.

* mark(message): Place a divider message in the test output, to make

  it easier to determine what part of the test script a command

  invocation belongs to.  The last mark message will also be displayed

  if a command invocation fails.  Do not include a newline in message.

* which(progname): Return the location of progname in the executable

  path, or None if it is not found.

* password(name): Return a weakly random password based on name.  The

  password will be consistent across calls with the same name.

* stop_daemon(proc): Stop a daemon process started with

  realm.start_server() or realm.start_in_inetd().  Only necessary if

  the port needs to be reused; daemon processes will be stopped

  automatically when the script exits.

* multipass_realms(**keywords): This is an iterator function.  Yields

  a realm for each of the standard test passes, each of which alters

  the default configuration in some way to exercise different parts of

  the krb5 code base.  keywords may contain any K5Realm initializer

  keyword with the exception of krbtgt_keysalt, which will not be

  honored.  If keywords contains krb5_conf and/or kdc_conf fragments,

  they will be merged with the default and per-pass specifications.

* multidb_realms(**keywords): Yields a realm for multiple DB modules.

  Currently DB2 and LMDB are included.  Ideally LDAP would be

  included, but setting up a test LDAP server currently requires a

  one-second delay, so all LDAP tests are currently confined to

  t_kdb.py.  keywords may contain any K5Realm initializer.

* cross_realms(num, xtgts=None, args=None, **keywords): This function

  returns a list of num realms, where each realm's configuration knows

  how to contact all of the realms.  By default, each realm will

  contain cross TGTs in both directions for all other realms; this

  default may be overridden by specifying a collection of tuples in

  the xtgts parameter, where each tuple is a pair of zero-based realm

  indexes, indicating that the first realm can authenticate to the

  second (i.e. krbtgt/secondrealm@firstrealm exists in both realm's

  databases).  If args is given, it should be a list of keyword

  arguments specific to each realm; these will be merged with the

  global keyword arguments passed to cross_realms, with specific

  arguments taking priority.

* buildtop: The top of the build directory (absolute path).

* srctop: The top of the source directory (absolute path).

* plugins: The plugin directory in the build tree (absolute path).

* hostname: This machine's fully-qualified domain name.

* null_input: A file opened to read /dev/null.

* args: Positional arguments left over after flags are processed.

* runenv: The contents of $srctop/runenv.py, containing a dictionary

  'env' which specifies additional variables to be added to the realm

  environment, and a variable 'tls_impl', which indicates which TLS

  implementation (if any) is being used by libkrb5's support for

  contacting KDCs and kpasswd servers over HTTPS.

* verbose: Whether the script is running verbosely.

* testpass: The command-line test pass argument.  The script does not

  need to examine this argument in most cases; it will be honored in

  multipass_realms().

* Pathname variables for programs within the build directory:

  - krb5kdc

  - kadmind

  - kadmin

  - kadminl (kadmin.local)

  - kdb5_ldap_util

  - kdb5_util

  - ktutil

  - kinit

  - klist

  - kswitch

  - kvno

  - kdestroy

  - kpasswd

  - t_inetd

  - kproplog

  - kpropd

  - kprop

Scripts may use the following realm methods and attributes:

* realm.run(args, env=None, **keywords): Run a command in a specified

  environment (or the realm's environment by default), obeying the

  command-line debugging options.  Fail if the command does not return

  0.  Log the command output appropriately, and return it as a single

  multi-line string.  Keyword arguments can contain input='string' to

  send an input string to the command, expected_code=N to expect a

  return code other than 0, expected_msg=MSG to expect a substring in

  the command output, and expected_trace=('a', 'b', ...) to expect an

  ordered series of line substrings in the command's KRB5_TRACE

  output, or return_trace=True to return a tuple of the command output

  and the trace output.

* realm.kprop_port(): Returns a port number based on realm.portbase

  intended for use by kprop and kpropd.

* realm.server_port(): Returns a port number based on realm.portbase

  intended for use by server processes.

* realm.start_server(args, sentinel, env=None): Start a daemon

  process.  Wait until sentinel appears as a substring of a line in

  the server process's stdout or stderr (which are folded together).

  Returns a subprocess.Popen object which can be passed to

  stop_daemon() to stop the server, or used to read from the server's

  output.

* realm.start_in_inetd(args, port=None, env=None): Begin a t_inetd

  process which will spawn a server process after accepting a client

  connection.  If port is not specified, realm.server_port() will be

  used.  Returns a process object which can be passed to stop_daemon()

  to stop the server.

* realm.create_kdb(): Create a new KDB.

* realm.start_kdc(args=[], env=None): Start a krb5kdc process.  Errors

  if a KDC is already running.  If args is given, it contains a list

  of additional krb5kdc arguments.

* realm.stop_kdc(): Stop the krb5kdc process.  Errors if no KDC is

  running.

* realm.start_kadmind(env=None): Start a kadmind process.  Errors if a

  kadmind is already running.

* realm.stop_kadmind(): Stop the kadmind process.  Errors if no

  kadmind is running.

* realm.stop(): Stop any daemon processes running on behalf of the

  realm.

* realm.addprinc(princname, password=None): Using kadmin.local, create

  a principal in the KDB named princname, with either a random or

  specified key.

* realm.extract_keytab(princname, keytab): Using kadmin.local, create

  a keytab for princname in the filename keytab.  Uses the -norandkey

  option to avoid re-randomizing princname's key.

* realm.kinit(princname, password=None, flags=[]): Acquire credentials

  for princname using kinit, with additional flags [].  If password is

  specified, it will be used as input to the kinit process; otherwise

  flags must cause kinit not to need a password (e.g. by specifying a

  keytab).

* realm.klist(client_princ, service_princ=None, ccache=None): Using

  klist, list the credentials cache ccache (must be a filename;

  self.ccache if not specified) and verify that the output shows

  credentials for client_princ and service_princ (self.krbtgt_princ if

  not specified).

* realm.klist_keytab(princ, keytab=None): Using klist, list keytab

  (must be a filename; self.keytab if not specified) and verify that

  the output shows the keytab name and principal name.

* realm.prep_kadmin(princname=None, password=None, flags=[]): Populate

  realm.kadmin_ccache with a ticket which can be used to run kadmin.

  If princname is not specified, realm.admin_princ and its default

  password will be used.

* realm.run_kadmin(args, **keywords): Run the specified query in

  kadmin, using realm.kadmin_ccache to authenticate.  Accepts the same

  keyword arguments as run.

* realm.special_env(name, has_kdc_conf, krb5_conf=None,

  kdc_conf=None): Create an environment with a modified krb5.conf

  and/or kdc.conf.  The specified krb5_conf and kdc_conf fragments, if

  any, will be merged with the realm's existing configuration.  If

  has_kdc_conf is false, the new environment will have no kdc.conf.

  The environment returned by this method can be used with realm.run()

  or similar methods.

* realm.start_kpropd(env, args=[]): Start a kpropd process.  Pass an

  environment created with realm.special_env() for the replica.  If

  args is given, it contains a list of additional kpropd arguments.

  Returns a handle to the kpropd process.

* realm.run_kpropd_once(env, args=[]): Run kpropd once, using the -t

  flag.  Pass an environment created with realm.special_env() for the

  replica.  If args is given, it contains a list of additional kpropd

  arguments.  Returns the kpropd output.

* realm.realm: The realm's name.

* realm.testdir: The realm's storage directory (absolute path).

* realm.portbase: The realm's first listener port.

* realm.user_princ: The principal name user@<realmname>.

* realm.admin_princ: The principal name user/admin@<realmname>.

* realm.host_princ: The name of the host principal for this machine,

  with realm.

* realm.nfs_princ: The name of the nfs principal for this machine,

  with realm.

* realm.krbtgt_princ: The name of the krbtgt principal for the realm.

* realm.keytab: A keytab file in realm.testdir.  Initially contains a

  host keytab unless disabled by the realm construction options.

* realm.client_keytab: A keytab file in realm.testdir.  Initially

  nonexistent.

* realm.ccache: A ccache file in realm.testdir.  Initially contains

  credentials for user unless disabled by the realm construction

  options.

* realm.kadmin_ccache: The ccache file initialized by prep_kadmin and

  used by run_kadmin.

* env: The realm's environment, extended from os.environ to point at

  the realm's config files and the build tree's shared libraries.

When the test script is run, its behavior can be modified with

command-line flags.  These are documented in the --help output.

"""

import atexit

import fcntl

import optparse

import os

import shlex

import shutil

import signal

import socket

import string

import subprocess

import sys

import imp

# Used when most things go wrong (other than programming errors) so

# that the user sees an error message rather than a Python traceback,

# without help from the test script.  The on-exit handler will display

# additional explanatory text.

def fail(msg):

    """Print a message and exit with failure."""

    global _current_pass

    print("*** Failure:", msg)

    if _last_mark:

        print("*** Last mark: %s" % _last_mark)

    if _last_cmd:

        print("*** Last command (#%d): %s" % (_cmd_index - 1, _last_cmd))

    if _last_cmd_output:

        print("*** Output of last command:")

        sys.stdout.write(_last_cmd_output)

    if _current_pass:

        print("*** Failed in test pass:", _current_pass)

    if _current_db:

        print("*** Failed with db:", _current_db)

    sys.exit(1)

def success(msg):

    global _success

    _check_daemons()

    output('*** Success: %s\n' % msg)

    _success = True

def mark(msg):

    global _last_mark

    output('\n====== %s ======\n' % msg)

    _last_mark = msg

def skipped(whatmsg, whymsg):

    output('*** Skipping: %s: %s\n' % (whatmsg, whymsg), force_verbose=True)

    f = open(os.path.join(buildtop, 'skiptests'), 'a')

    f.write('Skipped %s: %s\n' % (whatmsg, whymsg))

    f.close()

def skip_rest(whatmsg, whymsg):

    global _success

    skipped(whatmsg, whymsg)

    _check_daemons()

    _success = True

    sys.exit(0)

def output(msg, force_verbose=False):

    """Output a message to testlog, and to stdout if running verbosely."""

    _outfile.write(msg)

    if verbose or force_verbose:

        sys.stdout.write(msg)

# Return the location of progname in the executable path, or None if

# it is not found.

def which(progname):

    for dir in os.environ["PATH"].split(os.pathsep):

        path = os.path.join(dir, progname)

        if os.access(path, os.X_OK):

            return path

    return None

def password(name):

    """Choose a weakly random password from name, consistent across calls."""

    return name + str(os.getpid())

# Exit handler which ensures processes are cleaned up and, on failure,

# prints messages to help developers debug the problem.

def _onexit():

    global _daemons, _success, srctop, verbose

    global _debug, _stop_before, _stop_after, _shell_before, _shell_after

    if _debug or _stop_before or _stop_after or _shell_before or _shell_after:

        # Wait before killing daemons in case one is being debugged.

        sys.stdout.write('*** Press return to kill daemons and exit script: ')

        sys.stdout.flush()

        sys.stdin.readline()

    for proc in _daemons:

        if _check_daemon(proc) is None:

            os.kill(proc.pid, signal.SIGTERM)

    if not _success:

        print

        if not verbose:

            testlogfile = os.path.join(os.getcwd(), 'testlog')

            utildir = os.path.join(srctop, 'util')

            print('For details, see: %s' % testlogfile)

            print('Or re-run this test script with the -v flag:')

            print('    cd %s' % os.getcwd())

            print('    PYTHONPATH=%s %s %s -v' %

                  (utildir, sys.executable, sys.argv[0]))

            print()

        print('Use --debug=NUM to run a command under a debugger.  Use')

        print('--stop-after=NUM to stop after a daemon is started in order to')

        print('attach to it with a debugger.  Use --help to see other')

        print('options.')

def _onsigint(signum, frame):

    # Exit without displaying a stack trace.  Suppress messages from _onexit.

    global _success

    _success = True

    sys.exit(1)

# Find the parent of dir which is at the root of a build or source directory.

def _find_root(dir):

    while True:

        if os.path.exists(os.path.join(dir, 'lib', 'krb5', 'krb')):

            break

        parent = os.path.dirname(dir)

        if (parent == dir):

            return None

        dir = parent

    return dir

def _find_buildtop():

    root = _find_root(os.getcwd())

    if root is None:

        fail('Cannot find root of krb5 build directory.')

    if not os.path.exists(os.path.join(root, 'config.status')):

        # Looks like an unbuilt source directory.

        fail('This script must be run inside a krb5 build directory.')

    return root

def _find_srctop():

    scriptdir = os.path.abspath(os.path.dirname(sys.argv[0]))

    if not scriptdir:

        scriptdir = os.getcwd()

    root = _find_root(scriptdir)

    if root is None:

        fail('Cannot find root of krb5 source directory.')

    return os.path.abspath(root)

# Return the local hostname as it will be canonicalized by

# krb5_sname_to_principal.  We can't simply use socket.getfqdn()

# because it explicitly prefers results containing periods and

# krb5_sname_to_principal doesn't care.

def _get_hostname():

    hostname = socket.gethostname()

    try:

        ai = socket.getaddrinfo(hostname, None, 0, 0, 0, socket.AI_CANONNAME)

    except socket.gaierror as e:

        fail('Local hostname "%s" does not resolve: %s.' % (hostname, e[1]))

    (family, socktype, proto, canonname, sockaddr) = ai[0]

    try:

        name = socket.getnameinfo(sockaddr, socket.NI_NAMEREQD)

    except socket.gaierror:

        return canonname.lower()

    return name[0].lower()

# Parse command line arguments, setting global option variables.  Also

# sets the global variable args to the positional arguments, which may

# be used by the test script.

def _parse_args():

    global args, verbose, testpass, _debug, _debugger_command

    global _stop_before, _stop_after, _shell_before, _shell_after

    parser = optparse.OptionParser()

    parser.add_option('-v', '--verbose', action='store_true', dest='verbose',

                      default=False, help='Display verbose output')

    parser.add_option('-p', '--pass', dest='testpass', metavar='PASS',

                      help='If a multi-pass test, run only PASS')

    parser.add_option('--debug', dest='debug', metavar='NUM',

                      help='Debug numbered command (or "all")')

    parser.add_option('--debugger', dest='debugger', metavar='COMMAND',

                      help='Debugger command (default is gdb --args)',

                      default='gdb --args')

    parser.add_option('--stop-before', dest='stopb', metavar='NUM',

                      help='Stop before numbered command (or "all")')

    parser.add_option('--stop-after', dest='stopa', metavar='NUM',

                      help='Stop after numbered command (or "all")')

    parser.add_option('--shell-before', dest='shellb', metavar='NUM',

                      help='Spawn shell before numbered command (or "all")')

    parser.add_option('--shell-after', dest='shella', metavar='NUM',

                      help='Spawn shell after numbered command (or "all")')

    (options, args) = parser.parse_args()

    verbose = options.verbose

    testpass = options.testpass

    _debug = _parse_cmdnum('--debug', options.debug)

    _debugger_command = shlex.split(options.debugger)

    _stop_before = _parse_cmdnum('--stop-before', options.stopb)

    _stop_after = _parse_cmdnum('--stop-after', options.stopa)

    _shell_before = _parse_cmdnum('--shell-before', options.shellb)

    _shell_after = _parse_cmdnum('--shell-after', options.shella)

# Translate a command number spec.  -1 means all, None means none.

def _parse_cmdnum(optname, str):

    if not str:

        return None

    if str == 'all':

        return -1

    try:

        return int(str)

    except ValueError:

        fail('%s value must be "all" or a number' % optname)

# Test if a command index matches a translated command number spec.

def _match_cmdnum(cmdnum, ind):

    if cmdnum is None:

        return False

    elif cmdnum == -1:

        return True

    else:

        return cmdnum == ind

# Return an environment suitable for running programs in the build

# tree.  It is safe to modify the result.

def _build_env():

    global buildtop, runenv

    env = os.environ.copy()

    for (k, v) in runenv.env.items():

        if v.find('./') == 0:

            env[k] = os.path.join(buildtop, v)

        else:

            env[k] = v

    # Make sure we don't get confused by translated messages

    # or localized times.

    env['LC_ALL'] = 'C'

    return env

def _import_runenv():

    global buildtop

    runenv_py = os.path.join(buildtop, 'runenv.py')

    if not os.path.exists(runenv_py):

        fail('You must run "make runenv.py" in %s first.' % buildtop)

    return imp.load_source('runenv', runenv_py)

# Merge the nested dictionaries cfg1 and cfg2 into a new dictionary.

# cfg1 or cfg2 may be None, in which case the other is returned.  If

# cfg2 contains keys mapped to None, the corresponding keys will be

# mapped to None in the result.  The result may contain references to

# parts of cfg1 or cfg2, so is not safe to modify.

def _cfg_merge(cfg1, cfg2):

    if not cfg2:

        return cfg1

    if not cfg1:

        return cfg2

    result = cfg1.copy()

    for key, value2 in cfg2.items():

        if value2 is None:

            result.pop(key, None)

        elif key not in result:

            result[key] = value2

        else:

            value1 = result[key]

            if isinstance(value1, dict):

                if not isinstance(value2, dict):

                    raise TypeError()

                result[key] = _cfg_merge(value1, value2)

            else:

                result[key] = value2

    return result

# Python gives us shlex.split() to turn a shell command into a list of

# arguments, but oddly enough, not the easier reverse operation.  For

# now, do a bad job of faking it.

def _shell_equiv(args):

    return " ".join(args)

# Add a valgrind prefix to the front of args if specified in the

# environment.  Under normal circumstances this just returns args.

def _valgrind(args):

    valgrind = os.getenv('VALGRIND')

    if valgrind:

        args = shlex.split(valgrind) + args

    return args

def _stop_or_shell(stop, shell, env, ind):

    if (_match_cmdnum(stop, ind)):

        sys.stdout.write('*** [%d] Waiting for return: ' % ind)

        sys.stdout.flush()

        sys.stdin.readline()

    if (_match_cmdnum(shell, ind)):

        output('*** [%d] Spawning shell\n' % ind, True)

        subprocess.call(os.getenv('SHELL'), env=env)

# Look for the expected strings in successive lines of trace.

def _check_trace(trace, expected):

    i = 0

    for line in trace.splitlines():

        if i < len(expected) and expected[i] in line:

            i += 1

    if i < len(expected):

        fail('Expected string not found in trace output: ' + expected[i])

def _run_cmd(args, env, input=None, expected_code=0, expected_msg=None,

             expected_trace=None, return_trace=False):

    global null_input, _cmd_index, _last_cmd, _last_cmd_output, _debug

    global _stop_before, _stop_after, _shell_before, _shell_after

    tracefile = None

    if expected_trace is not None or return_trace:

        tracefile = 'testtrace'

        if os.path.exists(tracefile):

            os.remove(tracefile)

        env = env.copy()

        env['KRB5_TRACE'] = tracefile

    if (_match_cmdnum(_debug, _cmd_index)):

        return _debug_cmd(args, env, input)

    args = _valgrind(args)

    _last_cmd = _shell_equiv(args)

    output('*** [%d] Executing: %s\n' % (_cmd_index, _last_cmd))

    _stop_or_shell(_stop_before, _shell_before, env, _cmd_index)

    if input:

        infile = subprocess.PIPE

    else:

        infile = null_input

    # Run the command and log the result, folding stderr into stdout.

    proc = subprocess.Popen(args, stdin=infile, stdout=subprocess.PIPE,

                            stderr=subprocess.STDOUT, env=env,

                            universal_newlines=True)

    (outdata, dummy_errdata) = proc.communicate(input)

    _last_cmd_output = outdata

    code = proc.returncode

    output(outdata)

    output('*** [%d] Completed with return code %d\n' % (_cmd_index, code))

    _stop_or_shell(_stop_after, _shell_after, env, _cmd_index)

    _cmd_index += 1

    # Check the return code and return the output.

    if code != expected_code:

        fail('%s failed with code %d.' % (args[0], code))

    if expected_msg is not None and expected_msg not in outdata:

        fail('Expected string not found in command output: ' + expected_msg)

    if tracefile is not None:

        with open(tracefile, 'r') as f:

            trace = f.read()

        output('*** Trace output for previous command:\n')

        output(trace)

        if expected_trace is not None:

            _check_trace(trace, expected_trace)

    return (outdata, trace) if return_trace else outdata

def _debug_cmd(args, env, input):

    global _cmd_index, _debugger_command

    args = _debugger_command + list(args)

    output('*** [%d] Executing in debugger: %s\n' %

           (_cmd_index, _shell_equiv(args)), True)

    if input:

        print

        print('*** Enter the following input when appropriate:')

        print()

        print(input)

        print()

    code = subprocess.call(args, env=env)

    output('*** [%d] Completed in debugger with return code %d\n' %

           (_cmd_index, code))

    _cmd_index += 1

# Start a daemon process with the specified args and env.  Wait until

# we see sentinel as a substring of a line on either stdout or stderr.

# Clean up the daemon process on exit.

def _start_daemon(args, env, sentinel):

    global null_input, _cmd_index, _last_cmd, _last_cmd_output, _debug

    global _stop_before, _stop_after, _shell_before, _shell_after

    if (_match_cmdnum(_debug, _cmd_index)):

        output('*** [%d] Warning: ' % _cmd_index, True)

        output( 'test script cannot proceed after debugging a daemon\n', True)

        _debug_cmd(args, env, None)

        output('*** Exiting after debugging daemon\n', True)

        sys.exit(1)

    args = _valgrind(args)

    _last_cmd = _shell_equiv(args)

    output('*** [%d] Starting: %s\n' % (_cmd_index, _last_cmd))

    _stop_or_shell(_stop_before, _shell_before, env, _cmd_index)

    # Start the daemon and look for the sentinel in stdout or stderr.

    proc = subprocess.Popen(args, stdin=null_input, stdout=subprocess.PIPE,

                            stderr=subprocess.STDOUT, env=env,

                            universal_newlines=True)

    _last_cmd_output = ''

    while True:

        line = proc.stdout.readline()

        _last_cmd_output += line

        if line == "":

            code = proc.wait()

            fail('%s failed to start with code %d.' % (args[0], code))

        output(line)

        if sentinel in line:

            break

    output('*** [%d] Started with pid %d\n' % (_cmd_index, proc.pid))

    _stop_or_shell(_stop_after, _shell_after, env, _cmd_index)

    _cmd_index += 1

    # Save the daemon in a list for cleanup.  Note that we won't read

    # any more of the daemon's output after the sentinel, which will

    # cause the daemon to block if it generates enough.  For now we

    # assume all daemon processes are quiet enough to avoid this

    # problem.  If it causes an issue, some alternatives are:

    #   - Output to a file and poll the file for the sentinel

    #     (undesirable because it slows down the test suite by the

    #     polling interval times the number of daemons started)

    #   - Create an intermediate subprocess which discards output

    #     after the sentinel.

    _daemons.append(proc)

    # Return the process; the caller can stop it with stop_daemon.

    return proc

# Check a daemon's status prior to terminating it.  Display its return

# code if it already exited, and display any output it has generated.

# Return the daemon's exit status or None if it is still running.

def _check_daemon(proc):

    exited = False

    code = proc.poll()

    if code is not None:

        output('*** Daemon pid %d exited with code %d\n' % (proc.pid, code))

    flags = fcntl.fcntl(proc.stdout, fcntl.F_GETFL)

    fcntl.fcntl(proc.stdout, fcntl.F_SETFL, flags | os.O_NONBLOCK)

    try:

        out = proc.stdout.read()

    except:

        return

    output('*** Daemon pid %d output:\n' % proc.pid)

    output(out)

    return code

# Check all tracked daemon processes.  If any daemons already exited,

# remove them from the list (so we don't try to terminate them again).

# If any daemons exited with an error, fail out.

def _check_daemons():

    exited = []

    daemon_error = False

    for proc in _daemons:

        code = _check_daemon(proc)

        if code is not None:

            exited.append(proc)

            if code != 0:

                daemon_error = True

    for proc in exited:

        _daemons.remove(proc)

    if daemon_error:

        fail('One or more daemon processes exited with an error')

def stop_daemon(proc):

    code = _check_daemon(proc)

    if code is not None:

        _daemons.remove(proc)

        if code != 0:

            fail('Daemon process %d exited early' % proc.pid)

    else:

        output('*** Terminating process %d\n' % proc.pid)

        os.kill(proc.pid, signal.SIGTERM)

        proc.wait()

        _daemons.remove(proc)

class K5Realm(object):

    """An object representing a functional krb5 test realm."""

    def __init__(self, realm='KRBTEST.COM', portbase=61000, testdir='testdir',

                 krb5_conf=None, kdc_conf=None, create_kdb=True,

                 krbtgt_keysalt=None, create_user=True, get_creds=True,

                 create_host=True, start_kdc=True, start_kadmind=False,

                 start_kpropd=False, bdb_only=False):

        global hostname, _default_krb5_conf, _default_kdc_conf

        global _lmdb_kdc_conf, _current_db