def main():
arg_parser = get_arg_parser_with_db('CTF Gameserver Controller')
arg_parser.add_argument(
'--nonstop',
action='store_true',
help='Use current time as start time and '
'ignore CTF end time from the database. Useful for testing checkers.')
arg_parser.add_argument(
'--metrics-listen',
help='Expose Prometheus metrics via HTTP ("<host>:<port>")')
args = arg_parser.parse_args()
logging.basicConfig(format='[%(levelname)s] %(message)s')
numeric_loglevel = getattr(logging, args.loglevel.upper())
logging.getLogger().setLevel(numeric_loglevel)
try:
db_conn = psycopg2.connect(host=args.dbhost,
database=args.dbname,
user=args.dbuser,
password=args.dbpassword)
except psycopg2.OperationalError as e:
logging.error('Could not establish database connection: %s', e)
return os.EX_UNAVAILABLE
logging.info('Established database connection')
# Keep our mental model easy by always using (timezone-aware) UTC for dates and times
with transaction_cursor(db_conn) as cursor:
cursor.execute('SET TIME ZONE "UTC"')
# Check database grants
try:
try:
database.get_control_info(db_conn, prohibit_changes=True)
except DBDataError as e:
logging.warning('Invalid database state: %s', e)
database.increase_tick(db_conn, prohibit_changes=True)
except psycopg2.ProgrammingError as e:
if e.pgcode == postgres_errors.INSUFFICIENT_PRIVILEGE:
# Log full exception because only the backtrace will tell which kind of permission is missing
logging.exception('Missing database permissions:')
return os.EX_NOPERM
else:
raise
if args.metrics_listen is not None:
try:
metrics_host, metrics_port, metrics_family = parse_host_port(
args.metrics_listen)
except ValueError:
logging.error(
'Metrics listen address needs to be specified as "<host>:<port>"'
)
return os.EX_USAGE
start_metrics_server(metrics_host, metrics_port, metrics_family)
metrics = make_metrics(db_conn)
metrics['start_timestamp'].set(time.time())
daemon.notify('READY=1')
while True:
main_loop_step(db_conn, metrics, args.nonstop)
def test_ipv4(self):
host, port, family = args.parse_host_port('127.0.0.1:22')
self.assertEqual(host, '127.0.0.1')
self.assertEqual(port, 22)
self.assertEqual(family, socket.AF_INET)
def test_invalid(self):
with self.assertRaises(ValueError):
args.parse_host_port('::1')
def test_hostname(self):
parsed = args.parse_host_port('localhost:1337')
self.assertEqual(parsed[1], 1337)
def test_ipv6(self):
host, port, family = args.parse_host_port('[::1]:8000')
self.assertEqual(host, '::1')
self.assertEqual(port, 8000)
self.assertEqual(family, socket.AF_INET6)
def main():
arg_parser = get_arg_parser_with_db('CTF Gameserver Checker Master')
arg_parser.add_argument('--ippattern', type=str, required=True,
help='(Old-style) Python formatstring for building the IP to connect to')
arg_parser.add_argument('--flagsecret', type=str, required=True,
help='Base64 string used as secret in flag generation')
arg_parser.add_argument('--metrics-listen', type=str, help='Expose Prometheus metrics via HTTP '
'("<host>:<port>")')
group = arg_parser.add_argument_group('check', 'Check parameters')
group.add_argument('--service', type=str, required=True,
help='Slug of the service')
group.add_argument('--checkerscript', type=str, required=True,
help='Path of the Checker Script')
group.add_argument('--sudouser', type=str, help=' User to excute the Checker Scripts as, will be passed '
'to `sudo -u`')
group.add_argument('--stddeviations', type=float, default=2.0,
help='Consider past runtimes within this number of standard deviations when '
'estimating Checker Script runtime (default: 2)')
group.add_argument('--checkercount', type=int, required=True,
help='Number of Checker Masters running for this service')
group.add_argument('--interval', type=float, required=True,
help='Time between launching batches of Checker Scripts in seconds')
group = arg_parser.add_argument_group('logging', 'Checker Script logging')
group.add_argument('--journald', action='store_true', help='Log Checker Script messages to journald')
group.add_argument('--gelf-server', help='Log Checker Script messages to the specified GELF (Graylog) '
'server ("<host>:<port>")')
args = arg_parser.parse_args()
logging.basicConfig(format='[%(levelname)s] %(message)s [%(name)s]')
numeric_loglevel = getattr(logging, args.loglevel.upper())
logging.getLogger().setLevel(numeric_loglevel)
if args.interval < 3:
logging.error('`--interval` must be at least 3 seconds')
return os.EX_USAGE
logging_params = {}
# Configure logging
if args.journald:
try:
# pylint: disable=import-outside-toplevel,unused-import,import-error
from systemd.journal import JournalHandler
except ImportError:
logging.error('systemd module is required for journald logging')
return os.EX_USAGE
logging_params['journald'] = True
if args.gelf_server is not None:
try:
# pylint: disable=import-outside-toplevel,unused-import,import-error
import graypy
except ImportError:
logging.error('graypy module is required for GELF logging')
return os.EX_USAGE
try:
gelf_host, gelf_port, gelf_family = parse_host_port(args.gelf_server)
except ValueError:
logging.error('GELF server needs to be specified as "<host>:<port>"')
return os.EX_USAGE
logging_params['gelf'] = {'host': gelf_host, 'port': gelf_port, 'family': gelf_family}
# Configure metrics
if args.metrics_listen is not None:
try:
metrics_host, metrics_port, metrics_family = parse_host_port(args.metrics_listen)
except ValueError:
logging.error('Metrics listen address needs to be specified as "<host>:<port>"')
return os.EX_USAGE
metrics_queue = multiprocessing.Queue()
metrics_recv, metrics_send = multiprocessing.Pipe()
metrics_collector_process = multiprocessing.Process(
target=metrics.run_collector,
args=(args.service, metrics.checker_metrics_factory, metrics_queue, metrics_send)
)
# Terminate the process when the parent process exits
metrics_collector_process.daemon = True
metrics_collector_process.start()
logging.info('Started metrics collector process')
metrics_server_process = multiprocessing.Process(
target=metrics.run_http_server,
args=(metrics_host, metrics_port, metrics_family, metrics_queue, metrics_recv)
)
metrics_server_process.daemon = True
metrics_server_process.start()
logging.info('Started metrics HTTP server process')
metrics.set(metrics_queue, 'interval_length_seconds', args.interval)
metrics.set(metrics_queue, 'start_timestamp', time.time())
else:
metrics_queue = metrics.DummyQueue()
flag_secret = base64.b64decode(args.flagsecret)
# Connect to databases
try:
db_conn = psycopg2.connect(host=args.dbhost, database=args.dbname, user=args.dbuser,
password=args.dbpassword)
except psycopg2.OperationalError as e:
logging.error('Could not establish connection to database: %s', e)
return os.EX_UNAVAILABLE
logging.info('Established connection to database')
# Keep our mental model easy by always using (timezone-aware) UTC for dates and times
with transaction_cursor(db_conn) as cursor:
cursor.execute('SET TIME ZONE "UTC"')
# Check database grants
try:
try:
database.get_control_info(db_conn, prohibit_changes=True)
except DBDataError as e:
logging.warning('Invalid database state: %s', e)
try:
service_id = database.get_service_attributes(db_conn, args.service,
prohibit_changes=True)['id']
except DBDataError as e:
logging.warning('Invalid database state: %s', e)
service_id = 1337 # Use dummy value for subsequent grant checks
try:
database.get_current_tick(db_conn, prohibit_changes=True)
except DBDataError as e:
logging.warning('Invalid database state: %s', e)
database.get_task_count(db_conn, service_id, prohibit_changes=True)
database.get_new_tasks(db_conn, service_id, 1, prohibit_changes=True)
database.commit_result(db_conn, service_id, 1, 2147483647, 0, prohibit_changes=True, fake_team_id=1)
database.set_flagid(db_conn, service_id, 1, 0, 'id', prohibit_changes=True, fake_team_id=1)
database.load_state(db_conn, service_id, 1, 'key', prohibit_changes=True)
database.store_state(db_conn, service_id, 1, 'key', 'data', prohibit_changes=True)
except psycopg2.ProgrammingError as e:
if e.pgcode == postgres_errors.INSUFFICIENT_PRIVILEGE:
# Log full exception because only the backtrace will tell which kind of permission is missing
logging.exception('Missing database permissions:')
return os.EX_NOPERM
else:
raise
daemon.notify('READY=1')
while True:
try:
master_loop = MasterLoop(db_conn, args.service, args.checkerscript, args.sudouser,
args.stddeviations, args.checkercount, args.interval, args.ippattern,
flag_secret, logging_params, metrics_queue)
break
except DBDataError as e:
logging.warning('Waiting for valid database state: %s', e)
time.sleep(60)
# Graceful shutdown to prevent loss of check results
def sigterm_handler(_, __):
logging.info('Shutting down, waiting for %d Checker Scripts to finish',
master_loop.get_running_script_count())
master_loop.shutting_down = True
signal.signal(signal.SIGTERM, sigterm_handler)
try:
while True:
master_loop.step()
if master_loop.shutting_down and master_loop.get_running_script_count() == 0:
break
except: # noqa, pylint: disable=bare-except
logging.exception('Aborting due to unexpected error:')
master_loop.supervisor.terminate_runners()
return os.EX_SOFTWARE
return os.EX_OK
