#!@PYTHON@ # Copyright (C) 2010,2011 Internet Systems Consortium. # # Permission to use, copy, modify, and distribute this software for any # purpose with or without fee is hereby granted, provided that the above # copyright notice and this permission notice appear in all copies. # # THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SYSTEMS CONSORTIUM # DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL # INTERNET SYSTEMS CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT, # INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING # FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, # NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION # WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. """ This file implements the Boss of Bind (BoB, or bob) program. Its purpose is to start up the BIND 10 system, and then manage the processes, by starting and stopping processes, plus restarting processes that exit. To start the system, it first runs the c-channel program (msgq), then connects to that. It then runs the configuration manager, and reads its own configuration. Then it proceeds to starting other modules. The Python subprocess module is used for starting processes, but because this is not efficient for managing groups of processes, SIGCHLD signals are caught and processed using the signal module. Most of the logic is contained in the BoB class. However, since Python requires that signal processing happen in the main thread, we do signal handling outside of that class, in the code running for __main__. """ import sys; sys.path.append ('@@PYTHONPATH@@') import os # If B10_FROM_SOURCE is set in the environment, we use data files # from a directory relative to that, otherwise we use the ones # installed on the system if "B10_FROM_SOURCE" in os.environ: SPECFILE_LOCATION = os.environ["B10_FROM_SOURCE"] + "/src/bin/bind10/bob.spec" ADD_LIBEXEC_PATH = False else: PREFIX = "@prefix@" DATAROOTDIR = "@datarootdir@" SPECFILE_LOCATION = "@datadir@/@PACKAGE@/bob.spec".replace("${datarootdir}", DATAROOTDIR).replace("${prefix}", PREFIX) ADD_LIBEXEC_PATH = True import subprocess import signal import re import errno import time import select import random import socket from optparse import OptionParser, OptionValueError import io import pwd import posix import copy import isc.cc import isc.util.process import isc.net.parse import isc.log from isc.log_messages.bind10_messages import * import isc.bind10.component import isc.bind10.special_component isc.log.init("b10-boss") logger = isc.log.Logger("boss") # Pending system-wide debug level definitions, the ones we # use here are hardcoded for now DBG_PROCESS = logger.DBGLVL_TRACE_BASIC DBG_COMMANDS = logger.DBGLVL_TRACE_DETAIL # Assign this process some longer name isc.util.process.rename(sys.argv[0]) # This is the version that gets displayed to the user. # The VERSION string consists of the module name, the module version # number, and the overall BIND 10 version number (set in configure.ac). VERSION = "bind10 20110223 (BIND 10 @PACKAGE_VERSION@)" # This is for boot_time of Boss _BASETIME = time.gmtime() class RestartSchedule: """ Keeps state when restarting something (in this case, a process). When a process dies unexpectedly, we need to restart it. However, if it fails to restart for some reason, then we should not simply keep restarting it at high speed. A more sophisticated algorithm can be developed, but for now we choose a simple set of rules: * If a process was been running for >=10 seconds, we restart it right away. * If a process was running for <10 seconds, we wait until 10 seconds after it was started. To avoid programs getting into lockstep, we use a normal distribution to avoid being restarted at exactly 10 seconds.""" def __init__(self, restart_frequency=10.0): self.restart_frequency = restart_frequency self.run_start_time = None self.run_stop_time = None self.restart_time = None def set_run_start_time(self, when=None): if when is None: when = time.time() self.run_start_time = when sigma = self.restart_frequency * 0.05 self.restart_time = when + random.normalvariate(self.restart_frequency, sigma) def set_run_stop_time(self, when=None): """We don't actually do anything with stop time now, but it might be useful for future algorithms.""" if when is None: when = time.time() self.run_stop_time = when def get_restart_time(self, when=None): if when is None: when = time.time() return max(when, self.restart_time) class ProcessInfoError(Exception): pass class ProcessInfo: """Information about a process""" dev_null = open(os.devnull, "w") def __init__(self, name, args, env={}, dev_null_stdout=False, dev_null_stderr=False, uid=None, username=None): self.name = name self.args = args self.env = env self.dev_null_stdout = dev_null_stdout self.dev_null_stderr = dev_null_stderr self.restart_schedule = RestartSchedule() self.uid = uid self.username = username self.process = None self.pid = None def _preexec_work(self): """Function used before running a program that needs to run as a different user.""" # First, put us into a separate process group so we don't get # SIGINT signals on Ctrl-C (the boss will shut everthing down by # other means). os.setpgrp() # Second, set the user ID if one has been specified if self.uid is not None: try: posix.setuid(self.uid) except OSError as e: if e.errno == errno.EPERM: # if we failed to change user due to permission report that raise ProcessInfoError("Unable to change to user %s (uid %d)" % (self.username, self.uid)) else: # otherwise simply re-raise whatever error we found raise def _spawn(self): if self.dev_null_stdout: spawn_stdout = self.dev_null else: spawn_stdout = None if self.dev_null_stderr: spawn_stderr = self.dev_null else: spawn_stderr = None # Environment variables for the child process will be a copy of those # of the boss process with any additional specific variables given # on construction (self.env). spawn_env = copy.deepcopy(os.environ) spawn_env.update(self.env) if ADD_LIBEXEC_PATH: spawn_env['PATH'] = "@@LIBEXECDIR@@:" + spawn_env['PATH'] self.process = subprocess.Popen(self.args, stdin=subprocess.PIPE, stdout=spawn_stdout, stderr=spawn_stderr, close_fds=True, env=spawn_env, preexec_fn=self._preexec_work) self.pid = self.process.pid self.restart_schedule.set_run_start_time() # spawn() and respawn() are the same for now, but in the future they # may have different functionality def spawn(self): self._spawn() def respawn(self): self._spawn() class CChannelConnectError(Exception): pass class ProcessStartError(Exception): pass class BoB: """Boss of BIND class.""" def __init__(self, msgq_socket_file=None, data_path=None, config_filename=None, nocache=False, verbose=False, setuid=None, username=None, cmdctl_port=None, brittle=False, wait_time=10): """ Initialize the Boss of BIND. This is a singleton (only one can run). The msgq_socket_file specifies the UNIX domain socket file that the msgq process listens on. If verbose is True, then the boss reports what it is doing. Data path and config filename are passed through to config manager (if provided) and specify the config file to be used. The cmdctl_port is passed to cmdctl and specify on which port it should listen. brittle is a debug option that controls whether the Boss shuts down after any process dies. wait_time controls the amount of time (in seconds) that Boss waits for selected processes to initialize before continuing with the initialization. Currently this is only the configuration manager. """ self.cc_session = None self.ccs = None self.cfg_start_auth = True self.cfg_start_resolver = False self.cfg_start_dhcp6 = False self.cfg_start_dhcp4 = False self.started_auth_family = False self.started_resolver_family = False self.curproc = None # XXX: Not used now, waits for reintroduction of restarts. self.dead_processes = {} self.msgq_socket_file = msgq_socket_file self.nocache = nocache self.component_config = {} self.processes = {} self.runnable = False self.uid = setuid self.username = username self.verbose = verbose self.data_path = data_path self.config_filename = config_filename self.cmdctl_port = cmdctl_port self.brittle = brittle self.wait_time = wait_time self.sockcreator = None self._component_configurator = isc.bind10.component.Configurator(self, isc.bind10.special_component.get_specials()) # The priorities here make them start in the correct order. First # the socket creator (which would drop root privileges by then), # then message queue and after that the config manager (which uses # the config manager) self.__core_components = { 'sockcreator': { 'kind': 'core', 'special': 'sockcreator', 'priority': 200 }, 'msgq': { 'kind': 'core', 'special': 'msgq', 'priority': 199 }, 'cfgmgr': { 'kind': 'core', 'special': 'cfgmgr', 'priority': 198 } } self.__started = False self.exitcode = 0 # If -v was set, enable full debug logging. if self.verbose: logger.set_severity("DEBUG", 99) def __propagate_component_config(self, config): comps = dict(config) # Fill in the core components, so they stay alive for comp in self.__core_components: if comp in comps: raise Exception(comp + " is core component managed by " + "bind10 boss, do not set it") comps[comp] = self.__core_components[comp] # Update the configuration self._component_configurator.reconfigure(comps) def config_handler(self, new_config): # If this is initial update, don't do anything now, leave it to startup if not self.runnable: return # Now we declare few functions used only internally here. Besides the # benefit of not polluting the name space, they are closures, so we # don't need to pass some variables def start_stop(name, started, start, stop): if not'start_' + name in new_config: return if new_config['start_' + name]: if not started: if self.uid is not None: logger.info(BIND10_START_AS_NON_ROOT, name) start() else: stop() # These four functions are passed to start_stop (smells like functional # programming little bit) def resolver_on(): self.component_config['b10-resolver'] = { 'kind': 'needed', 'special': 'resolver' } self.__propagate_component_config(self.component_config) self.started_resolver_family = True def resolver_off(): if 'b10-resolver' in self.component_config: del self.component_config['b10-resolver'] self.__propagate_component_config(self.component_config) self.started_resolver_family = False def auth_on(): self.component_config['b10-auth'] = { 'kind': 'needed', 'special': 'auth' } self.component_config['b10-xfrout'] = { 'kind': 'dispensable', 'address': 'Xfrout' } self.component_config['b10-xfrin'] = { 'kind': 'dispensable', 'special': 'xfrin' } self.component_config['b10-zonemgr'] = { 'kind': 'dispensable', 'address': 'Zonemgr' } self.__propagate_component_config(self.component_config) self.started_auth_family = True def auth_off(): if 'b10-zonemgr' in self.component_config: del self.component_config['b10-zonemgr'] if 'b10-xfrin' in self.component_config: del self.component_config['b10-xfrin'] if 'b10-xfrout' in self.component_config: del self.component_config['b10-xfrout'] if 'b10-auth' in self.component_config: del self.component_config['b10-auth'] self.__propagate_component_config(self.component_config) self.started_auth_family = False # The real code of the config handler function follows here logger.debug(DBG_COMMANDS, BIND10_RECEIVED_NEW_CONFIGURATION, new_config) start_stop('resolver', self.started_resolver_family, resolver_on, resolver_off) start_stop('auth', self.started_auth_family, auth_on, auth_off) answer = isc.config.ccsession.create_answer(0) return answer def get_processes(self): pids = list(self.processes.keys()) pids.sort() process_list = [ ] for pid in pids: process_list.append([pid, self.processes[pid].name()]) return process_list def _get_stats_data(self): return { "owner": "Boss", "data": { 'boot_time': time.strftime('%Y-%m-%dT%H:%M:%SZ', _BASETIME) } } def command_handler(self, command, args): logger.debug(DBG_COMMANDS, BIND10_RECEIVED_COMMAND, command) answer = isc.config.ccsession.create_answer(1, "command not implemented") if type(command) != str: answer = isc.config.ccsession.create_answer(1, "bad command") else: if command == "shutdown": self.runnable = False answer = isc.config.ccsession.create_answer(0) elif command == "getstats": answer = isc.config.ccsession.create_answer(0, self._get_stats_data()) elif command == "sendstats": # send statistics data to the stats daemon immediately stats_data = self._get_stats_data() valid = self.ccs.get_module_spec().validate_statistics( True, stats_data["data"]) if valid: cmd = isc.config.ccsession.create_command('set', stats_data) seq = self.cc_session.group_sendmsg(cmd, 'Stats') # Consume the answer, in case it becomes a orphan message. try: self.cc_session.group_recvmsg(False, seq) except isc.cc.session.SessionTimeout: pass answer = isc.config.ccsession.create_answer(0) else: logger.fatal(BIND10_INVALID_STATISTICS_DATA); answer = isc.config.ccsession.create_answer( 1, "specified statistics data is invalid") elif command == "ping": answer = isc.config.ccsession.create_answer(0, "pong") elif command == "show_processes": answer = isc.config.ccsession. \ create_answer(0, self.get_processes()) else: answer = isc.config.ccsession.create_answer(1, "Unknown command") return answer def kill_started_processes(self): """ Called as part of the exception handling when a process fails to start, this runs through the list of started processes, killing each one. It then clears that list. """ logger.info(BIND10_KILLING_ALL_PROCESSES) self.stop_creator(True) for pid in self.processes: logger.info(BIND10_KILL_PROCESS, self.processes[pid].name()) self.processes[pid].kill(True) self.processes = {} def read_bind10_config(self): """ Reads the parameters associated with the BoB module itself. At present these are the components to start although arguably this information should be in the configuration for the appropriate module itself. (However, this would cause difficulty in the case of xfrin/xfrout and zone manager as we don't need to start those if we are not running the authoritative server.) """ logger.info(BIND10_READING_BOSS_CONFIGURATION) config_data = self.ccs.get_full_config() self.cfg_start_auth = config_data.get("start_auth") self.cfg_start_resolver = config_data.get("start_resolver") logger.info(BIND10_CONFIGURATION_START_AUTH, self.cfg_start_auth) logger.info(BIND10_CONFIGURATION_START_RESOLVER, self.cfg_start_resolver) def log_starting(self, process, port = None, address = None): """ A convenience function to output a "Starting xxx" message if the logging is set to DEBUG with debuglevel DBG_PROCESS or higher. Putting this into a separate method ensures that the output form is consistent across all processes. The process name (passed as the first argument) is put into self.curproc, and is used to indicate which process failed to start if there is an error (and is used in the "Started" message on success). The optional port and address information are appended to the message (if present). """ self.curproc = process if port is None and address is None: logger.info(BIND10_STARTING_PROCESS, self.curproc) elif address is None: logger.info(BIND10_STARTING_PROCESS_PORT, self.curproc, port) else: logger.info(BIND10_STARTING_PROCESS_PORT_ADDRESS, self.curproc, address, port) def log_started(self, pid = None): """ A convenience function to output a 'Started xxxx (PID yyyy)' message. As with starting_message(), this ensures a consistent format. """ if pid is None: logger.debug(DBG_PROCESS, BIND10_STARTED_PROCESS, self.curproc) else: logger.debug(DBG_PROCESS, BIND10_STARTED_PROCESS_PID, self.curproc, pid) def process_running(self, msg, who): """ Some processes return a message to the Boss after they have started to indicate that they are running. The form of the message is a dictionary with contents {"running:", ""}. This method checks the passed message and returns True if the "who" process is contained in the message (so is presumably running). It returns False for all other conditions and will log an error if appropriate. """ if msg is not None: try: if msg["running"] == who: return True else: logger.error(BIND10_STARTUP_UNEXPECTED_MESSAGE, msg) except: logger.error(BIND10_STARTUP_UNRECOGNISED_MESSAGE, msg) return False # The next few methods start the individual processes of BIND-10. They # are called via start_all_processes(). If any fail, an exception is # raised which is caught by the caller of start_all_processes(); this kills # processes started up to that point before terminating the program. def start_msgq(self): """ Start the message queue and connect to the command channel. """ self.log_starting("b10-msgq") msgq_proc = ProcessInfo("b10-msgq", ["b10-msgq"], self.c_channel_env, True, not self.verbose, uid=self.uid, username=self.username) msgq_proc.spawn() self.log_started(msgq_proc.pid) # Now connect to the c-channel cc_connect_start = time.time() while self.cc_session is None: # if we have been trying for "a while" give up if (time.time() - cc_connect_start) > 5: raise CChannelConnectError("Unable to connect to c-channel after 5 seconds") # try to connect, and if we can't wait a short while try: self.cc_session = isc.cc.Session(self.msgq_socket_file) except isc.cc.session.SessionError: time.sleep(0.1) # Subscribe to the message queue. The only messages we expect to receive # on this channel are once relating to process startup. self.cc_session.group_subscribe("Boss") return msgq_proc def start_cfgmgr(self): """ Starts the configuration manager process """ self.log_starting("b10-cfgmgr") args = ["b10-cfgmgr"] if self.data_path is not None: args.append("--data-path=" + self.data_path) if self.config_filename is not None: args.append("--config-filename=" + self.config_filename) bind_cfgd = ProcessInfo("b10-cfgmgr", args, self.c_channel_env, uid=self.uid, username=self.username) bind_cfgd.spawn() self.log_started(bind_cfgd.pid) # Wait for the configuration manager to start up as subsequent initialization # cannot proceed without it. The time to wait can be set on the command line. time_remaining = self.wait_time msg, env = self.cc_session.group_recvmsg() while time_remaining > 0 and not self.process_running(msg, "ConfigManager"): logger.debug(DBG_PROCESS, BIND10_WAIT_CFGMGR) time.sleep(1) time_remaining = time_remaining - 1 msg, env = self.cc_session.group_recvmsg() if not self.process_running(msg, "ConfigManager"): raise ProcessStartError("Configuration manager process has not started") return bind_cfgd def start_ccsession(self, c_channel_env): """ Start the CC Session The argument c_channel_env is unused but is supplied to keep the argument list the same for all start_xxx methods. With regards to logging, note that as the CC session is not a process, the log_starting/log_started methods are not used. """ logger.info(BIND10_STARTING_CC) self.ccs = isc.config.ModuleCCSession(SPECFILE_LOCATION, self.config_handler, self.command_handler, socket_file = self.msgq_socket_file) self.ccs.start() logger.debug(DBG_PROCESS, BIND10_STARTED_CC) # A couple of utility methods for starting processes... def start_process(self, name, args, c_channel_env, port=None, address=None): """ Given a set of command arguments, start the process and output appropriate log messages. If the start is successful, the process is added to the list of started processes. The port and address arguments are for log messages only. """ self.log_starting(name, port, address) newproc = ProcessInfo(name, args, c_channel_env) newproc.spawn() self.log_started(newproc.pid) return newproc def register_process(self, pid, component): """ Put another process into boss to watch over it. When the process dies, the component.failed() is called with the exit code. """ self.processes[pid] = component def start_simple(self, name): """ Most of the BIND-10 processes are started with the command: [-v] ... where -v is appended if verbose is enabled. This method generates the arguments from the name and starts the process. The port and address arguments are for log messages only. """ # Set up the command arguments. args = [name] if self.verbose: args += ['-v'] # ... and start the process return self.start_process(name, args, self.c_channel_env) # The next few methods start up the rest of the BIND-10 processes. # Although many of these methods are little more than a call to # start_simple, they are retained (a) for testing reasons and (b) as a place # where modifications can be made if the process start-up sequence changes # for a given process. def start_auth(self): """ Start the Authoritative server """ authargs = ['b10-auth'] if self.nocache: authargs += ['-n'] if self.uid: authargs += ['-u', str(self.uid)] if self.verbose: authargs += ['-v'] # ... and start return self.start_process("b10-auth", authargs, self.c_channel_env) def start_resolver(self): """ Start the Resolver. At present, all these arguments and switches are pure speculation. As with the auth daemon, they should be read from the configuration database. """ self.curproc = "b10-resolver" # XXX: this must be read from the configuration manager in the future resargs = ['b10-resolver'] if self.uid: resargs += ['-u', str(self.uid)] if self.verbose: resargs += ['-v'] # ... and start return self.start_process("b10-resolver", resargs, self.c_channel_env) def start_cmdctl(self): """ Starts the command control process """ args = ["b10-cmdctl"] if self.cmdctl_port is not None: args.append("--port=" + str(self.cmdctl_port)) return self.start_process("b10-cmdctl", args, self.c_channel_env, self.cmdctl_port) def start_xfrin(self): # XXX: a quick-hack workaround. xfrin will implicitly use dynamically # loadable data source modules, which will be installed in $(libdir). # On some OSes (including MacOS X and *BSDs) the main process (python) # cannot find the modules unless they are located in a common shared # object path or a path in the (DY)LD_LIBRARY_PATH. We should seek # a cleaner solution, but for a short term workaround we specify the # path here, unconditionally, and without even bothering which # environment variable should be used. # # We reuse the ADD_LIBEXEC_PATH variable to see whether we need to # do this, as the conditions that make this workaround needed are # the same as for the libexec path addition # TODO: Once #1292 is finished, remove this method and the special # component, use it as normal component. c_channel_env = dict(self.c_channel_env) if ADD_LIBEXEC_PATH: cur_path = os.getenv('DYLD_LIBRARY_PATH') cur_path = '' if cur_path is None else ':' + cur_path c_channel_env['DYLD_LIBRARY_PATH'] = "@@LIBDIR@@" + cur_path cur_path = os.getenv('LD_LIBRARY_PATH') cur_path = '' if cur_path is None else ':' + cur_path c_channel_env['LD_LIBRARY_PATH'] = "@@LIBDIR@@" + cur_path # Set up the command arguments. args = ['b10-xfrin'] if self.verbose: args += ['-v'] return self.start_process("b10-xfrin", args, c_channel_env) def start_all_processes(self): """ Starts up all the processes. Any exception generated during the starting of the processes is handled by the caller. """ # Start the real core (sockcreator, msgq, cfgmgr) self._component_configurator.startup(self.__core_components) # Connect to the msgq. This is not a process, so it's not handled # inside the configurator. c_channel_env = self.c_channel_env self.start_ccsession(c_channel_env) # Extract the parameters associated with Bob. This can only be # done after the CC Session is started. Note that the logging # configuration may override the "-v" switch set on the command line. self.read_bind10_config() # Continue starting the processes. The authoritative server (if # selected): component_config = {} if self.cfg_start_auth: component_config['b10-auth'] = { 'kind': 'needed', 'special': 'auth' } self.__propagate_component_config(component_config) # ... and resolver (if selected): if self.cfg_start_resolver: component_config['b10-resolver'] = { 'kind': 'needed', 'special': 'resolver' } self.started_resolver_family = True self.__propagate_component_config(component_config) # Everything after the main components can run as non-root. # TODO: this is only temporary - once the privileged socket creator is # fully working, nothing else will run as root. if self.uid is not None: posix.setuid(self.uid) # xfrin/xfrout and the zone manager are only meaningful if the # authoritative server has been started. if self.cfg_start_auth: component_config['b10-xfrout'] = { 'kind': 'dispensable', 'address': 'Xfrout' } component_config['b10-xfrin'] = { 'kind': 'dispensable', 'special': 'xfrin' } component_config['b10-zonemgr'] = { 'kind': 'dispensable', 'address': 'Zonemgr' } self.__propagate_component_config(component_config) self.started_auth_family = True # ... and finally start the remaining processes component_config['b10-stats'] = { 'kind': 'dispensable', 'address': 'Stats' } component_config['b10-stats-httpd'] = { 'kind': 'dispensable', 'address': 'StatsHttpd' } component_config['b10-cmdctl'] = { 'kind': 'needed', 'special': 'cmdctl' } if self.cfg_start_dhcp6: component_config['b10-dhcp6'] = { 'kind': 'dispensable', 'address': 'DHCP6' } self.__propagate_component_config(component_config) self.component_config = component_config def startup(self): """ Start the BoB instance. Returns None if successful, otherwise an string describing the problem. """ # Try to connect to the c-channel daemon, to see if it is already # running c_channel_env = {} if self.msgq_socket_file is not None: c_channel_env["BIND10_MSGQ_SOCKET_FILE"] = self.msgq_socket_file logger.debug(DBG_PROCESS, BIND10_CHECK_MSGQ_ALREADY_RUNNING) # try to connect, and if we can't wait a short while try: self.cc_session = isc.cc.Session(self.msgq_socket_file) logger.fatal(BIND10_MSGQ_ALREADY_RUNNING) return "b10-msgq already running, or socket file not cleaned , cannot start" except isc.cc.session.SessionError: # this is the case we want, where the msgq is not running pass # Start all processes. If any one fails to start, kill all started # processes and exit with an error indication. try: self.c_channel_env = c_channel_env self.start_all_processes() except Exception as e: self.kill_started_processes() return "Unable to start " + self.curproc + ": " + str(e) # Started successfully self.runnable = True self.__started = True return None def stop_process(self, process, recipient): """ Stop the given process, friendly-like. The process is the name it has (in logs, etc), the recipient is the address on msgq. """ logger.info(BIND10_STOP_PROCESS, process) self.cc_session.group_sendmsg({'command': ['shutdown']}, recipient, recipient) def component_shutdown(self, exitcode=0): """ Stop the Boss instance from a components' request. The exitcode indicates the desired exit code. If we did not start yet, it raises an exception, which is meant to propagate through the component and configurator to the startup routine and abort the startup imediatelly. If it is started up already, we just mark it so we terminate soon. It does set the exit code in both cases. """ self.exitcode = exitcode if not self.__started: raise Exception("Component failed during startup"); else: self.runnable = False # Series of stop_process wrappers def stop_resolver(self): self.stop_process('b10-resolver', 'Resolver') def stop_auth(self): self.stop_process('b10-auth', 'Auth') def stop_xfrout(self): self.stop_process('b10-xfrout', 'Xfrout') def stop_xfrin(self): self.stop_process('b10-xfrin', 'Xfrin') def stop_zonemgr(self): self.stop_process('b10-zonemgr', 'Zonemgr') def shutdown(self): """Stop the BoB instance.""" logger.info(BIND10_SHUTDOWN) # first try using the BIND 10 request to stop try: self._component_configurator.shutdown() except: pass # XXX: some delay probably useful... how much is uncertain # I have changed the delay from 0.5 to 1, but sometime it's # still not enough. time.sleep(1) self.reap_children() # next try sending a SIGTERM components_to_stop = list(self.processes.values()) for component in components_to_stop: logger.info(BIND10_SEND_SIGTERM, component.name(), component.pid()) try: component.kill() except OSError: # ignore these (usually ESRCH because the child # finally exited) pass # finally, send SIGKILL (unmaskable termination) until everybody dies while self.processes: # XXX: some delay probably useful... how much is uncertain time.sleep(0.1) self.reap_children() components_to_stop = list(self.processes.values()) for component in components_to_stop: logger.info(BIND10_SEND_SIGKILL, component.name(), component.pid()) try: component.kill(True) except OSError: # ignore these (usually ESRCH because the child # finally exited) pass logger.info(BIND10_SHUTDOWN_COMPLETE) def _get_process_exit_status(self): return os.waitpid(-1, os.WNOHANG) def reap_children(self): """Check to see if any of our child processes have exited, and note this for later handling. """ while True: try: (pid, exit_status) = self._get_process_exit_status() except OSError as o: if o.errno == errno.ECHILD: break # XXX: should be impossible to get any other error here raise if pid == 0: break if pid in self.processes: # One of the processes we know about. Get information on it. component = self.processes.pop(pid) if component.running() and self.runnable: # Tell it it failed. But only if it matters (we are # not shutting down and the component considers itself # to be running. component.failed(exit_status); else: logger.info(BIND10_UNKNOWN_CHILD_PROCESS_ENDED, pid) def restart_processes(self): """ Restart any dead processes: * Returns the time when the next process is ready to be restarted. * If the server is shutting down, returns 0. * If there are no processes, returns None. The values returned can be safely passed into select() as the timeout value. """ next_restart = None # if we're shutting down, then don't restart if not self.runnable: return 0 # otherwise look through each dead process and try to restart still_dead = {} now = time.time() for proc_info in self.dead_processes.values(): restart_time = proc_info.restart_schedule.get_restart_time(now) if restart_time > now: if (next_restart is None) or (next_restart > restart_time): next_restart = restart_time still_dead[proc_info.pid] = proc_info else: logger.info(BIND10_RESURRECTING_PROCESS, proc_info.name) try: proc_info.respawn() self.processes[proc_info.pid] = proc_info logger.info(BIND10_RESURRECTED_PROCESS, proc_info.name, proc_info.pid) except: still_dead[proc_info.pid] = proc_info # remember any processes that refuse to be resurrected self.dead_processes = still_dead # return the time when the next process is ready to be restarted return next_restart # global variables, needed for signal handlers options = None boss_of_bind = None def reaper(signal_number, stack_frame): """A child process has died (SIGCHLD received).""" # don't do anything... # the Python signal handler has been set up to write # down a pipe, waking up our select() bit pass def get_signame(signal_number): """Return the symbolic name for a signal.""" for sig in dir(signal): if sig.startswith("SIG") and sig[3].isalnum(): if getattr(signal, sig) == signal_number: return sig return "Unknown signal %d" % signal_number # XXX: perhaps register atexit() function and invoke that instead def fatal_signal(signal_number, stack_frame): """We need to exit (SIGINT or SIGTERM received).""" global options global boss_of_bind logger.info(BIND10_RECEIVED_SIGNAL, get_signame(signal_number)) signal.signal(signal.SIGCHLD, signal.SIG_DFL) boss_of_bind.runnable = False def process_rename(option, opt_str, value, parser): """Function that renames the process if it is requested by a option.""" isc.util.process.rename(value) def parse_args(args=sys.argv[1:], Parser=OptionParser): """ Function for parsing command line arguments. Returns the options object from OptionParser. """ parser = Parser(version=VERSION) parser.add_option("-m", "--msgq-socket-file", dest="msgq_socket_file", type="string", default=None, help="UNIX domain socket file the b10-msgq daemon will use") parser.add_option("-n", "--no-cache", action="store_true", dest="nocache", default=False, help="disable hot-spot cache in authoritative DNS server") parser.add_option("-u", "--user", dest="user", type="string", default=None, help="Change user after startup (must run as root)") parser.add_option("-v", "--verbose", dest="verbose", action="store_true", help="display more about what is going on") parser.add_option("--pretty-name", type="string", action="callback", callback=process_rename, help="Set the process name (displayed in ps, top, ...)") parser.add_option("-c", "--config-file", action="store", dest="config_file", default=None, help="Configuration database filename") parser.add_option("-p", "--data-path", dest="data_path", help="Directory to search for configuration files", default=None) parser.add_option("--cmdctl-port", dest="cmdctl_port", type="int", default=None, help="Port of command control") parser.add_option("--pid-file", dest="pid_file", type="string", default=None, help="file to dump the PID of the BIND 10 process") parser.add_option("--brittle", dest="brittle", action="store_true", help="debugging flag: exit if any component dies") parser.add_option("-w", "--wait", dest="wait_time", type="int", default=10, help="Time (in seconds) to wait for config manager to start up") (options, args) = parser.parse_args(args) if options.cmdctl_port is not None: try: isc.net.parse.port_parse(options.cmdctl_port) except ValueError as e: parser.error(e) if args: parser.print_help() sys.exit(1) return options def dump_pid(pid_file): """ Dump the PID of the current process to the specified file. If the given file is None this function does nothing. If the file already exists, the existing content will be removed. If a system error happens in creating or writing to the file, the corresponding exception will be propagated to the caller. """ if pid_file is None: return f = open(pid_file, "w") f.write('%d\n' % os.getpid()) f.close() def unlink_pid_file(pid_file): """ Remove the given file, which is basically expected to be the PID file created by dump_pid(). The specified may or may not exist; if it doesn't this function does nothing. Other system level errors in removing the file will be propagated as the corresponding exception. """ if pid_file is None: return try: os.unlink(pid_file) except OSError as error: if error.errno is not errno.ENOENT: raise def main(): global options global boss_of_bind # Enforce line buffering on stdout, even when not a TTY sys.stdout = io.TextIOWrapper(sys.stdout.detach(), line_buffering=True) options = parse_args() # Check user ID. setuid = None username = None if options.user: # Try getting information about the user, assuming UID passed. try: pw_ent = pwd.getpwuid(int(options.user)) setuid = pw_ent.pw_uid username = pw_ent.pw_name except ValueError: pass except KeyError: pass # Next try getting information about the user, assuming user name # passed. # If the information is both a valid user name and user number, we # prefer the name because we try it second. A minor point, hopefully. try: pw_ent = pwd.getpwnam(options.user) setuid = pw_ent.pw_uid username = pw_ent.pw_name except KeyError: pass if setuid is None: logger.fatal(BIND10_INVALID_USER, options.user) sys.exit(1) # Announce startup. logger.info(BIND10_STARTING, VERSION) # Create wakeup pipe for signal handlers wakeup_pipe = os.pipe() signal.set_wakeup_fd(wakeup_pipe[1]) # Set signal handlers for catching child termination, as well # as our own demise. signal.signal(signal.SIGCHLD, reaper) signal.siginterrupt(signal.SIGCHLD, False) signal.signal(signal.SIGINT, fatal_signal) signal.signal(signal.SIGTERM, fatal_signal) # Block SIGPIPE, as we don't want it to end this process signal.signal(signal.SIGPIPE, signal.SIG_IGN) # Go bob! boss_of_bind = BoB(options.msgq_socket_file, options.data_path, options.config_file, options.nocache, options.verbose, setuid, username, options.cmdctl_port, options.brittle, options.wait_time) startup_result = boss_of_bind.startup() if startup_result: logger.fatal(BIND10_STARTUP_ERROR, startup_result) sys.exit(1) logger.info(BIND10_STARTUP_COMPLETE) dump_pid(options.pid_file) # In our main loop, we check for dead processes or messages # on the c-channel. wakeup_fd = wakeup_pipe[0] ccs_fd = boss_of_bind.ccs.get_socket().fileno() while boss_of_bind.runnable: # clean up any processes that exited boss_of_bind.reap_children() # XXX: As we don't put anything into the processes to be restarted, # this is really a complicated NOP. But we will try to reintroduce # delayed restarts, so it stays here for now, until we find out if # it's useful. next_restart = boss_of_bind.restart_processes() if next_restart is None: wait_time = None else: wait_time = max(next_restart - time.time(), 0) # select() can raise EINTR when a signal arrives, # even if they are resumable, so we have to catch # the exception try: (rlist, wlist, xlist) = select.select([wakeup_fd, ccs_fd], [], [], wait_time) except select.error as err: if err.args[0] == errno.EINTR: (rlist, wlist, xlist) = ([], [], []) else: logger.fatal(BIND10_SELECT_ERROR, err) break for fd in rlist + xlist: if fd == ccs_fd: try: boss_of_bind.ccs.check_command() except isc.cc.session.ProtocolError: logger.fatal(BIND10_MSGQ_DISAPPEARED) self.runnable = False break elif fd == wakeup_fd: os.read(wakeup_fd, 32) # shutdown signal.signal(signal.SIGCHLD, signal.SIG_DFL) boss_of_bind.shutdown() unlink_pid_file(options.pid_file) sys.exit(0) if __name__ == "__main__": main()