class Reactor(object):
class Exit(object):
normal = 0
validate = 0
listening = 1
configuration = 1
privileges = 1
log = 1
pid = 1
socket = 1
io_error = 1
process = 1
select = 1
unknown = 1
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = concat_bytes_i(
character(int(c, 16))
for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a'])
def __init__(self, configurations):
self._ips = environment.settings().tcp.bind
self._port = environment.settings().tcp.port
self._stopping = environment.settings().tcp.once
self.exit_code = self.Exit.unknown
self.max_loop_time = environment.settings().reactor.speed
self._sleep_time = self.max_loop_time / 100
self._busyspin = {}
self._ratelimit = {}
self.early_drop = environment.settings().daemon.drop
self.processes = None
self.configuration = Configuration(configurations)
self.logger = Logger()
self.asynchronous = ASYNC()
self.signal = Signal()
self.daemon = Daemon(self)
self.listener = Listener(self)
self.api = API(self)
self._peers = {}
self._reload_processes = False
self._saved_pid = False
self._poller = select.poll()
def _termination(self, reason, exit_code):
self.exit_code = exit_code
self.signal.received = Signal.SHUTDOWN
self.logger.critical(reason, 'reactor')
def _prevent_spin(self):
second = int(time.time())
if not second in self._busyspin:
self._busyspin = {second: 0}
self._busyspin[second] += 1
if self._busyspin[second] > self.max_loop_time:
time.sleep(self._sleep_time)
return True
return False
def _rate_limited(self, peer, rate):
if rate <= 0:
return False
second = int(time.time())
ratelimit = self._ratelimit.get(peer, {})
if not second in ratelimit:
self._ratelimit[peer] = {second: rate - 1}
return False
if self._ratelimit[peer][second] > 0:
self._ratelimit[peer][second] -= 1
return False
return True
def _wait_for_io(self, sleeptime):
spin_prevention = False
try:
for fd, event in self._poller.poll(sleeptime):
if event & select.POLLIN or event & select.POLLPRI:
yield fd
continue
elif event & select.POLLHUP or event & select.POLLERR or event & select.POLLNVAL:
spin_prevention = True
continue
if spin_prevention:
self._prevent_spin()
except KeyboardInterrupt:
self._termination('^C received', self.Exit.normal)
return
except Exception:
self._prevent_spin()
return
# peer related functions
def active_peers(self):
peers = set()
for key, peer in self._peers.items():
if not peer.neighbor.passive or peer.proto:
peers.add(key)
return peers
def established_peers(self):
peers = set()
for key, peer in self._peers.items():
if peer.fsm == FSM.ESTABLISHED:
peers.add(key)
return peers
def peers(self):
return list(self._peers)
def handle_connection(self, peer_name, connection):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return
peer.handle_connection(connection)
def neighbor(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return
return peer.neighbor
def neighbor_name(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return ""
return peer.neighbor.name()
def neighbor_ip(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return ""
return str(peer.neighbor.peer_address)
def neighbor_cli_data(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return ""
return peer.cli_data()
def neighor_rib(self, peer_name, rib_name, advertised=False):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return []
families = None
if advertised:
families = peer.proto.negotiated.families if peer.proto else []
rib = peer.neighbor.rib.outgoing if rib_name == 'out' else peer.neighbor.rib.incoming
return list(rib.cached_changes(families))
def neighbor_rib_resend(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return
peer.neighbor.rib.outgoing.resend(None, peer.neighbor.route_refresh)
def neighbor_rib_out_withdraw(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return
peer.neighbor.rib.outgoing.withdraw(None, peer.neighbor.route_refresh)
def neighbor_rib_in_clear(self, peer_name):
peer = self._peers.get(peer_name, None)
if not peer:
self.logger.critical('could not find referenced peer', 'reactor')
return
peer.neighbor.rib.incoming.clear()
# ...
def _completed(self, peers):
for peer in peers:
if self._peers[peer].neighbor.rib.outgoing.pending():
return False
return True
def run(self, validate, root):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes()
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
for ip in self._ips:
if not self.listener.listen_on(ip, None, self._port, None, False,
None):
return self.Exit.listening
if not self.load():
return self.Exit.configuration
if validate: # only validate configuration
self.logger.warning('', 'configuration')
self.logger.warning('parsed Neighbors, un-templated',
'configuration')
self.logger.warning('------------------------------',
'configuration')
self.logger.warning('', 'configuration')
for key in self._peers:
self.logger.warning(str(self._peers[key].neighbor),
'configuration')
self.logger.warning('', 'configuration')
return self.Exit.validate
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
if not self.listener.listen_on(
neighbor.md5_ip, neighbor.peer_address,
neighbor.listen, neighbor.md5_password,
neighbor.md5_base64, neighbor.ttl_in):
return self.Exit.listening
if not self.early_drop:
self.processes.start(self.configuration.processes)
if not self.daemon.drop_privileges():
self.logger.critical(
'could not drop privileges to \'%s\' refusing to run as root' %
self.daemon.user, 'reactor')
self.logger.critical(
'set the environmemnt value exabgp.daemon.user to change the unprivileged user',
'reactor')
return self.Exit.privileges
if self.early_drop:
self.processes.start(self.configuration.processes)
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.critical('could not setup the logger, aborting',
'reactor')
return self.Exit.log
if not self.daemon.savepid():
return self.Exit.pid
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = False
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.debug(
'waiting for %d seconds before connecting' % sleeptime,
'reactor')
time.sleep(float(sleeptime))
workers = {}
peers = set()
api_fds = []
ms_sleep = int(self._sleep_time * 1000)
while True:
try:
if self.signal.received:
for key in self._peers:
if self._peers[key].neighbor.api['signal']:
self._peers[key].reactor.processes.signal(
self._peers[key].neighbor, self.signal.number)
signaled = self.signal.received
self.signal.rearm()
if signaled == Signal.SHUTDOWN:
self.exit_code = self.Exit.normal
self.shutdown()
break
if signaled == Signal.RESTART:
self.restart()
continue
if not reload_completed:
continue
if signaled == Signal.FULL_RELOAD:
self._reload_processes = True
if signaled in (Signal.RELOAD, Signal.FULL_RELOAD):
self.load()
self.processes.start(self.configuration.processes,
self._reload_processes)
self._reload_processes = False
continue
if self.listener.incoming():
# check all incoming connection
self.asynchronous.schedule(
str(uuid.uuid1()), 'checking for new connection(s)',
self.listener.new_connections())
peers = self.active_peers()
if self._completed(peers):
reload_completed = True
sleep = ms_sleep
# do not attempt to listen on closed sockets even if the peer is still here
for io in list(workers.keys()):
if io == -1:
self._poller.unregister(io)
del workers[io]
# give a turn to all the peers
for key in list(peers):
peer = self._peers[key]
# limit the number of message handling per second
if self._rate_limited(key, peer.neighbor.rate_limit):
peers.discard(key)
continue
# handle the peer
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
if key in self._peers:
del self._peers[key]
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
io = peer.socket()
if io != -1:
self._poller.register(
io,
select.POLLIN | select.POLLPRI | select.POLLHUP
| select.POLLNVAL | select.POLLERR)
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
elif action == ACTION.NOW:
sleep = 0
if not peers:
break
# read at least on message per process if there is some and parse it
for service, command in self.processes.received():
self.api.text(self, service, command)
sleep = 0
self.asynchronous.run()
if api_fds != self.processes.fds:
for fd in api_fds:
if fd == -1:
continue
if fd not in self.processes.fds:
self._poller.unregister(fd)
for fd in self.processes.fds:
if fd == -1:
continue
if fd not in api_fds:
self._poller.register(
fd,
select.POLLIN | select.POLLPRI | select.POLLHUP
| select.POLLNVAL | select.POLLERR)
api_fds = self.processes.fds
for io in self._wait_for_io(sleep):
if io not in api_fds:
peers.add(workers[io])
if self._stopping and not self._peers.keys():
self._termination('exiting on peer termination',
self.Exit.normal)
except KeyboardInterrupt:
self._termination('^C received', self.Exit.normal)
except SystemExit:
self._termination('exiting', self.Exit.normal)
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
self._termination('socket error received', self.Exit.socket)
except IOError:
self._termination(
'I/O Error received, most likely ^C during IO',
self.Exit.io_error)
except ProcessError:
self._termination(
'Problem when sending message(s) to helper program, stopping',
self.Exit.process)
except select.error:
self._termination('problem using select, stopping',
self.Exit.select)
return self.exit_code
def register_peer(self, name, peer):
self._peers[name] = peer
def teardown_peer(self, name, code):
self._peers[name].teardown(code)
def shutdown(self):
"""Terminate all the current BGP connections"""
self.logger.critical('performing shutdown', 'reactor')
if self.listener:
self.listener.stop()
self.listener = None
for key in self._peers.keys():
self._peers[key].shutdown()
self.asynchronous.clear()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load(self):
"""Reload the configuration and send to the peer the route which changed"""
self.logger.notice('performing reload of exabgp %s' % version,
'configuration')
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.error(
'not reloaded, no change found in the configuration',
'configuration')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.error(str(self.configuration.error), 'configuration')
return False
for key, peer in self._peers.items():
if key not in self.configuration.neighbors:
self.logger.debug('removing peer: %s' % peer.neighbor.name(),
'reactor')
peer.remove()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self._peers:
self.logger.debug('new peer: %s' % neighbor.name(), 'reactor')
peer = Peer(neighbor, self)
self._peers[key] = peer
# modified peer
elif self._peers[key].neighbor != neighbor:
self.logger.debug(
'peer definition change, establishing a new connection for %s'
% str(key), 'reactor')
self._peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.debug(
'peer definition identical, updating peer routes if required for %s'
% str(key), 'reactor')
self._peers[key].reconfigure(neighbor)
for ip in self._ips:
if ip.afi == neighbor.peer_address.afi:
self.listener.listen_on(ip, neighbor.peer_address,
self._port, neighbor.md5_password,
neighbor.md5_base64, None)
self.logger.notice('loaded new configuration successfully', 'reactor')
return True
def restart(self):
"""Kill the BGP session and restart it"""
self.logger.notice('performing restart of exabgp %s' % version,
'reactor')
# XXX: FIXME: Could return False, in case there is interference with old config...
reloaded = self.configuration.reload()
for key in self._peers.keys():
if key not in self.configuration.neighbors.keys():
peer = self._peers[key]
self.logger.debug('removing peer %s' % peer.neighbor.name(),
'reactor')
self._peers[key].remove()
else:
self._peers[key].reestablish()
self.processes.start(self.configuration.processes, True)
class Reactor (object):
class Exit (object):
normal = 0
validate = 0
listening = 1
configuration = 1
privileges = 1
log = 1
pid = 1
socket = 1
io_error = 1
process = 1
select = 1
unknown = 1
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = concat_bytes_i(character(int(c,16)) for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a'])
def __init__ (self, configurations):
self._ips = environment.settings().tcp.bind
self._port = environment.settings().tcp.port
self._stopping = environment.settings().tcp.once
self.exit_code = self.Exit.unknown
self.max_loop_time = environment.settings().reactor.speed
self._sleep_time = self.max_loop_time / 100
self._busyspin = {}
self.early_drop = environment.settings().daemon.drop
self.processes = None
self.configuration = Configuration(configurations)
self.logger = Logger()
self.asynchronous = ASYNC()
self.signal = Signal()
self.daemon = Daemon(self)
self.listener = Listener(self)
self.api = API(self)
self.peers = {}
self._reload_processes = False
self._saved_pid = False
def _termination (self,reason, exit_code):
self.exit_code = exit_code
self.signal.received = Signal.SHUTDOWN
self.logger.critical(reason,'reactor')
def _prevent_spin(self):
second = int(time.time())
if not second in self._busyspin:
self._busyspin = {second: 0}
self._busyspin[second] += 1
if self._busyspin[second] > self.max_loop_time:
time.sleep(self._sleep_time)
return True
return False
def _api_ready (self,sockets,sleeptime):
fds = self.processes.fds()
ios = fds + sockets
try:
read,_,_ = select.select(ios,[],[],sleeptime)
for fd in fds:
if fd in read:
read.remove(fd)
return read
except select.error as exc:
err_no,message = exc.args # pylint: disable=W0633
if err_no not in error.block:
raise exc
self._prevent_spin()
return []
except socket.error as exc:
# python 3 does not raise on closed FD, but python2 does
# we have lost a peer and it is causing the select
# to complain, the code will self-heal, ignore the issue
# (EBADF from python2 must be ignored if when checkign error.fatal)
# otherwise sending notification causes TCP to drop and cause
# this code to kill ExaBGP
self._prevent_spin()
return []
except ValueError as exc:
# The peer closing the TCP connection lead to a negative file descritor
self._prevent_spin()
return []
except KeyboardInterrupt:
self._termination('^C received',self.Exit.normal)
return []
def _active_peers (self):
peers = set()
for key,peer in self.peers.items():
if not peer.neighbor.passive or peer.proto:
peers.add(key)
return peers
def _completed (self,peers):
for peer in peers:
if self.peers[peer].neighbor.rib.outgoing.pending():
return False
return True
def run (self, validate, root):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes()
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
for ip in self._ips:
if not self.listener.listen_on(ip, None, self._port, None, False, None):
return self.Exit.listening
if not self.load():
return self.Exit.configuration
if validate: # only validate configuration
self.logger.warning('','configuration')
self.logger.warning('parsed Neighbors, un-templated','configuration')
self.logger.warning('------------------------------','configuration')
self.logger.warning('','configuration')
for key in self.peers:
self.logger.warning(str(self.peers[key].neighbor),'configuration')
self.logger.warning('','configuration')
return self.Exit.validate
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
if not self.listener.listen_on(neighbor.md5_ip, neighbor.peer_address, neighbor.listen, neighbor.md5_password, neighbor.md5_base64, neighbor.ttl_in):
return self.Exit.listening
if not self.early_drop:
self.processes.start(self.configuration.processes)
if not self.daemon.drop_privileges():
self.logger.critical('could not drop privileges to \'%s\' refusing to run as root' % self.daemon.user,'reactor')
self.logger.critical('set the environmemnt value exabgp.daemon.user to change the unprivileged user','reactor')
return self.Exit.privileges
if self.early_drop:
self.processes.start(self.configuration.processes)
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.critical('could not setup the logger, aborting','reactor')
return self.Exit.log
if not self.daemon.savepid():
return self.Exit.pid
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = False
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.debug('waiting for %d seconds before connecting' % sleeptime,'reactor')
time.sleep(float(sleeptime))
workers = {}
peers = set()
while True:
try:
if self.signal.received:
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self.peers[key].reactor.processes.signal(self.peers[key].neighbor,self.signal.number)
signaled = self.signal.received
self.signal.rearm()
if signaled == Signal.SHUTDOWN:
self.shutdown()
break
if signaled == Signal.RESTART:
self.restart()
continue
if not reload_completed:
continue
if signaled == Signal.FULL_RELOAD:
self._reload_processes = True
if signaled in (Signal.RELOAD, Signal.FULL_RELOAD):
self.load()
self.processes.start(self.configuration.processes,self._reload_processes)
self._reload_processes = False
continue
if self.listener.incoming():
# check all incoming connection
self.asynchronous.schedule(str(uuid.uuid1()),'checking for new connection(s)',self.listener.new_connections())
peers = self._active_peers()
if self._completed(peers):
reload_completed = True
sleep = self._sleep_time
# do not attempt to listen on closed sockets even if the peer is still here
for io in list(workers.keys()):
if io.fileno() == -1:
del workers[io]
# give a turn to all the peers
for key in list(peers):
peer = self.peers[key]
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
if key in self.peers:
del self.peers[key]
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
for io in peer.sockets():
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
elif action == ACTION.NOW:
sleep = 0
if not peers:
break
# read at least on message per process if there is some and parse it
for service,command in self.processes.received():
self.api.text(self,service,command)
sleep = 0
self.asynchronous.run()
for io in self._api_ready(list(workers),sleep):
peers.add(workers[io])
del workers[io]
if self._stopping and not self.peers.keys():
self._termination('exiting on peer termination',self.Exit.normal)
except KeyboardInterrupt:
self._termination('^C received',self.Exit.normal)
except SystemExit:
self._termination('exiting', self.Exit.normal)
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
self._termination('socket error received',self.Exit.socket)
except IOError:
self._termination('I/O Error received, most likely ^C during IO',self.Exit.io_error)
except ProcessError:
self._termination('Problem when sending message(s) to helper program, stopping',self.Exit.process)
except select.error:
self._termination('problem using select, stopping',self.Exit.select)
return self.exit_code
def shutdown (self):
"""Terminate all the current BGP connections"""
self.logger.critical('performing shutdown','reactor')
if self.listener:
self.listener.stop()
self.listener = None
for key in self.peers.keys():
self.peers[key].shutdown()
self.asynchronous.clear()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load (self):
"""Reload the configuration and send to the peer the route which changed"""
self.logger.notice('performing reload of exabgp %s' % version,'configuration')
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.error('problem with the configuration file, no change done','configuration')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.error(str(self.configuration.error),'configuration')
return False
for key, peer in self.peers.items():
if key not in self.configuration.neighbors:
self.logger.debug('removing peer: %s' % peer.neighbor.name(),'reactor')
peer.remove()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self.peers:
self.logger.debug('new peer: %s' % neighbor.name(),'reactor')
peer = Peer(neighbor,self)
self.peers[key] = peer
# modified peer
elif self.peers[key].neighbor != neighbor:
self.logger.debug('peer definition change, establishing a new connection for %s' % str(key),'reactor')
self.peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.debug('peer definition identical, updating peer routes if required for %s' % str(key),'reactor')
self.peers[key].reconfigure(neighbor)
for ip in self._ips:
if ip.afi == neighbor.peer_address.afi:
self.listener.listen_on(ip, neighbor.peer_address, self._port, neighbor.md5_password, neighbor.md5_base64, None)
self.logger.notice('loaded new configuration successfully','reactor')
return True
def restart (self):
"""Kill the BGP session and restart it"""
self.logger.notice('performing restart of exabgp %s' % version,'reactor')
self.configuration.reload()
for key in self.peers.keys():
if key not in self.configuration.neighbors.keys():
neighbor = self.configuration.neighbors[key]
self.logger.debug('removing Peer %s' % neighbor.name(),'reactor')
self.peers[key].remove()
else:
self.peers[key].reestablish()
self.processes.start(self.configuration.processes,True)
class Reactor(object):
class Exit(object):
normal = 0
validate = 0
listening = 1
configuration = 1
privileges = 1
log = 1
pid = 1
socket = 1
io_error = 1
process = 1
select = 1
unknown = 1
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = concat_bytes_i(
character(int(c, 16))
for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a'])
def __init__(self, configurations):
self._ips = environment.settings().tcp.bind
self._port = environment.settings().tcp.port
self._stopping = environment.settings().tcp.once
self.exit_code = self.Exit.unknown
self.max_loop_time = environment.settings().reactor.speed
self._sleep_time = self.max_loop_time / 100
self._busyspin = {}
self.early_drop = environment.settings().daemon.drop
self.processes = None
self.configuration = Configuration(configurations)
self.logger = Logger()
self.asynchronous = ASYNC()
self.signal = Signal()
self.daemon = Daemon(self)
self.listener = Listener(self)
self.api = API(self)
self.peers = {}
self._reload_processes = False
self._saved_pid = False
def _termination(self, reason, exit_code):
self.exit_code = exit_code
self.signal.received = Signal.SHUTDOWN
self.logger.critical(reason, 'reactor')
def _prevent_spin(self):
second = int(time.time())
if not second in self._busyspin:
self._busyspin = {second: 0}
self._busyspin[second] += 1
if self._busyspin[second] > self.max_loop_time:
time.sleep(self._sleep_time)
return True
return False
def _api_ready(self, sockets, sleeptime):
fds = self.processes.fds()
ios = fds + sockets
try:
read, _, _ = select.select(ios, [], [], sleeptime)
for fd in fds:
if fd in read:
read.remove(fd)
return read
except select.error as exc:
err_no, message = exc.args # pylint: disable=W0633
if err_no not in error.block:
raise exc
self._prevent_spin()
return []
except socket.error as exc:
# python 3 does not raise on closed FD, but python2 does
# we have lost a peer and it is causing the select
# to complain, the code will self-heal, ignore the issue
# (EBADF from python2 must be ignored if when checkign error.fatal)
# otherwise sending notification causes TCP to drop and cause
# this code to kill ExaBGP
self._prevent_spin()
return []
except ValueError as exc:
# The peer closing the TCP connection lead to a negative file descritor
self._prevent_spin()
return []
except KeyboardInterrupt:
self._termination('^C received', self.Exit.normal)
return []
def _active_peers(self):
peers = set()
for key, peer in self.peers.items():
if not peer.neighbor.passive or peer.proto:
peers.add(key)
return peers
def _completed(self, peers):
for peer in peers:
if self.peers[peer].neighbor.rib.outgoing.pending():
return False
return True
def run(self, validate, root):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes()
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
for ip in self._ips:
if not self.listener.listen_on(ip, None, self._port, None, False,
None):
return self.Exit.listening
if not self.load():
return self.Exit.configuration
if validate: # only validate configuration
self.logger.warning('', 'configuration')
self.logger.warning('parsed Neighbors, un-templated',
'configuration')
self.logger.warning('------------------------------',
'configuration')
self.logger.warning('', 'configuration')
for key in self.peers:
self.logger.warning(str(self.peers[key].neighbor),
'configuration')
self.logger.warning('', 'configuration')
return self.Exit.validate
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
if not self.listener.listen_on(
neighbor.md5_ip, neighbor.peer_address,
neighbor.listen, neighbor.md5_password,
neighbor.md5_base64, neighbor.ttl_in):
return self.Exit.listening
if not self.early_drop:
self.processes.start(self.configuration.processes)
if not self.daemon.drop_privileges():
self.logger.critical(
'could not drop privileges to \'%s\' refusing to run as root' %
self.daemon.user, 'reactor')
self.logger.critical(
'set the environmemnt value exabgp.daemon.user to change the unprivileged user',
'reactor')
return self.Exit.privileges
if self.early_drop:
self.processes.start(self.configuration.processes)
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.critical('could not setup the logger, aborting',
'reactor')
return self.Exit.log
if not self.daemon.savepid():
return self.Exit.pid
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = False
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.debug(
'waiting for %d seconds before connecting' % sleeptime,
'reactor')
time.sleep(float(sleeptime))
workers = {}
peers = set()
while True:
try:
if self.signal.received:
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self.peers[key].reactor.processes.signal(
self.peers[key].neighbor, self.signal.number)
signaled = self.signal.received
self.signal.rearm()
if signaled == Signal.SHUTDOWN:
self.shutdown()
break
if signaled == Signal.RESTART:
self.restart()
continue
if not reload_completed:
continue
if signaled == Signal.FULL_RELOAD:
self._reload_processes = True
if signaled in (Signal.RELOAD, Signal.FULL_RELOAD):
self.load()
self.processes.start(self.configuration.processes,
self._reload_processes)
self._reload_processes = False
continue
if self.listener.incoming():
# check all incoming connection
self.asynchronous.schedule(
str(uuid.uuid1()), 'checking for new connection(s)',
self.listener.new_connections())
peers = self._active_peers()
if self._completed(peers):
reload_completed = True
sleep = self._sleep_time
# do not attempt to listen on closed sockets even if the peer is still here
for io in list(workers.keys()):
if io.fileno() == -1:
del workers[io]
# give a turn to all the peers
for key in list(peers):
peer = self.peers[key]
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
if key in self.peers:
del self.peers[key]
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
for io in peer.sockets():
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
elif action == ACTION.NOW:
sleep = 0
if not peers:
break
# read at least on message per process if there is some and parse it
for service, command in self.processes.received():
self.api.text(self, service, command)
sleep = 0
self.asynchronous.run()
for io in self._api_ready(list(workers), sleep):
peers.add(workers[io])
del workers[io]
if self._stopping and not self.peers.keys():
self._termination('exiting on peer termination',
self.Exit.normal)
except KeyboardInterrupt:
self._termination('^C received', self.Exit.normal)
except SystemExit:
self._termination('exiting', self.Exit.normal)
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
self._termination('socket error received', self.Exit.socket)
except IOError:
self._termination(
'I/O Error received, most likely ^C during IO',
self.Exit.io_error)
except ProcessError:
self._termination(
'Problem when sending message(s) to helper program, stopping',
self.Exit.process)
except select.error:
self._termination('problem using select, stopping',
self.Exit.select)
return self.exit_code
def shutdown(self):
"""Terminate all the current BGP connections"""
self.logger.critical('performing shutdown', 'reactor')
if self.listener:
self.listener.stop()
self.listener = None
for key in self.peers.keys():
self.peers[key].shutdown()
self.asynchronous.clear()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load(self):
"""Reload the configuration and send to the peer the route which changed"""
self.logger.notice('performing reload of exabgp %s' % version,
'configuration')
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.error(
'problem with the configuration file, no change done',
'configuration')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.error(str(self.configuration.error), 'configuration')
return False
for key, peer in self.peers.items():
if key not in self.configuration.neighbors:
self.logger.debug('removing peer: %s' % peer.neighbor.name(),
'reactor')
peer.remove()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self.peers:
self.logger.debug('new peer: %s' % neighbor.name(), 'reactor')
peer = Peer(neighbor, self)
self.peers[key] = peer
# modified peer
elif self.peers[key].neighbor != neighbor:
self.logger.debug(
'peer definition change, establishing a new connection for %s'
% str(key), 'reactor')
self.peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.debug(
'peer definition identical, updating peer routes if required for %s'
% str(key), 'reactor')
self.peers[key].reconfigure(neighbor)
for ip in self._ips:
if ip.afi == neighbor.peer_address.afi:
self.listener.listen_on(ip, neighbor.peer_address,
self._port, neighbor.md5_password,
neighbor.md5_base64, None)
self.logger.notice('loaded new configuration successfully', 'reactor')
return True
def restart(self):
"""Kill the BGP session and restart it"""
self.logger.notice('performing restart of exabgp %s' % version,
'reactor')
self.configuration.reload()
for key in self.peers.keys():
if key not in self.configuration.neighbors.keys():
neighbor = self.configuration.neighbors[key]
self.logger.debug('removing Peer %s' % neighbor.name(),
'reactor')
self.peers[key].remove()
else:
self.peers[key].reestablish()
self.processes.start(self.configuration.processes, True)
class Reactor(object):
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = concat_bytes_i(
character(int(c, 16))
for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a'])
def __init__(self, configurations):
self.ip = environment.settings().tcp.bind
self.port = environment.settings().tcp.port
self.respawn = environment.settings().api.respawn
self.max_loop_time = environment.settings().reactor.speed
self.early_drop = environment.settings().daemon.drop
self.logger = Logger()
self.daemon = Daemon(self)
self.processes = None
self.listener = None
self.configuration = Configuration(configurations)
self.api = API(self)
self.peers = {}
self.route_update = False
self._stopping = environment.settings().tcp.once
self._shutdown = False
self._reload = False
self._reload_processes = False
self._restart = False
self._saved_pid = False
self._pending = deque()
self._running = None
signal.signal(signal.SIGTERM, self.sigterm)
signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGALRM, self.sigalrm)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
def sigterm(self, signum, frame):
self.logger.reactor('SIG TERM received - shutdown')
self._shutdown = True
def sighup(self, signum, frame):
self.logger.reactor('SIG HUP received - shutdown')
self._shutdown = True
def sigalrm(self, signum, frame):
self.logger.reactor('SIG ALRM received - restart')
self._restart = True
def sigusr1(self, signum, frame):
self.logger.reactor('SIG USR1 received - reload configuration')
self._reload = True
def sigusr2(self, signum, frame):
self.logger.reactor(
'SIG USR2 received - reload configuration and processes')
self._reload = True
self._reload_processes = True
def ready(self, sockets, ios, sleeptime=0):
# never sleep a negative number of second (if the rounding is negative somewhere)
# never sleep more than one second (should the clock time change during two time.time calls)
sleeptime = min(max(0.0, sleeptime), 1.0)
if not ios:
time.sleep(sleeptime)
return []
try:
read, _, _ = select.select(sockets + ios, [], [], sleeptime)
return read
except select.error as exc:
errno, message = exc.args # pylint: disable=W0633
if errno not in error.block:
raise exc
return []
except socket.error as exc:
if exc.errno in error.fatal:
raise exc
return []
def run(self):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes(self)
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
if not self.load():
return False
try:
self.listener = Listener()
if self.ip:
self.listener.listen(IP.create(self.ip), IP.create('0.0.0.0'),
self.port, None, False, None)
self.logger.reactor('Listening for BGP session(s) on %s:%d' %
(self.ip, self.port))
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
self.listener.listen(neighbor.md5_ip,
neighbor.peer_address,
neighbor.listen,
neighbor.md5_password,
neighbor.md5_base64, neighbor.ttl_in)
self.logger.reactor(
'Listening for BGP session(s) on %s:%d%s' %
(neighbor.md5_ip, neighbor.listen,
' with MD5' if neighbor.md5_password else ''))
except NetworkError as exc:
self.listener = None
if os.geteuid() != 0 and self.port <= 1024:
self.logger.reactor(
'Can not bind to %s:%d, you may need to run ExaBGP as root'
% (self.ip, self.port), 'critical')
else:
self.logger.reactor(
'Can not bind to %s:%d (%s)' %
(self.ip, self.port, str(exc)), 'critical')
self.logger.reactor(
'unset exabgp.tcp.bind if you do not want listen for incoming connections',
'critical')
self.logger.reactor(
'and check that no other daemon is already binding to port %d'
% self.port, 'critical')
sys.exit(1)
if not self.early_drop:
self.processes.start()
if not self.daemon.drop_privileges():
self.logger.reactor(
'Could not drop privileges to \'%s\' refusing to run as root' %
self.daemon.user, 'critical')
self.logger.reactor(
'Set the environmemnt value exabgp.daemon.user to change the unprivileged user',
'critical')
return
if self.early_drop:
self.processes.start()
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.reactor('Could not setup the logger, aborting',
'critical')
return
if not self.daemon.savepid():
return
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = True
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.reactor('waiting for %d seconds before connecting' %
sleeptime)
time.sleep(float(sleeptime))
workers = {}
peers = set()
scheduled = False
while True:
try:
finished = False
start = time.time()
end = start + self.max_loop_time
if self._shutdown:
self._shutdown = False
self.shutdown()
break
if self._reload and reload_completed:
self._reload = False
self.load()
self.processes.start(self._reload_processes)
self._reload_processes = False
elif self._restart:
self._restart = False
self.restart()
# We got some API routes to announce
if self.route_update:
self.route_update = False
self.route_send()
for peer in self.peers.keys():
peers.add(peer)
while start < time.time() < end and not finished:
if self.peers:
for key in list(peers):
peer = self.peers[key]
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
self.unschedule(key)
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
for io in peer.sockets():
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
if not peers:
reload_completed = True
if self.listener:
for connection in self.listener.connected():
# found
# * False, not peer found for this TCP connection
# * True, peer found
# * None, conflict found for this TCP connections
found = False
for key in self.peers:
peer = self.peers[key]
neighbor = peer.neighbor
# XXX: FIXME: Inet can only be compared to Inet
if connection.local == str(
neighbor.peer_address) and (
neighbor.auto_discovery
or connection.peer == str(
neighbor.local_address)):
if peer.incoming(connection):
found = True
break
found = None
break
if found:
self.logger.reactor(
'accepted connection from %s - %s' %
(connection.local, connection.peer))
elif found is False:
self.logger.reactor(
'no session configured for %s - %s' %
(connection.local, connection.peer))
connection.notification(
6, 3, 'no session configured for the peer')
connection.close()
elif found is None:
self.logger.reactor(
'connection refused (already connected to the peer) %s - %s'
% (connection.local, connection.peer))
connection.notification(
6, 5, 'could not accept the connection')
connection.close()
scheduled = self.schedule()
finished = not peers and not scheduled
# RFC state that we MUST not send more than one KEEPALIVE / sec
# And doing less could cause the session to drop
if finished:
for io in self.ready(list(peers), self.processes.fds(),
end - time.time()):
if io in workers:
peers.add(workers[io])
del workers[io]
if self._stopping and not self.peers.keys():
break
except KeyboardInterrupt:
while True:
try:
self._shutdown = True
self.logger.reactor('^C received')
break
except KeyboardInterrupt:
pass
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
try:
self._shutdown = True
self.logger.reactor('socket error received', 'warning')
break
except KeyboardInterrupt:
pass
except IOError:
while True:
try:
self._shutdown = True
self.logger.reactor(
'I/O Error received, most likely ^C during IO',
'warning')
break
except KeyboardInterrupt:
pass
except SystemExit:
try:
self._shutdown = True
self.logger.reactor('exiting')
break
except KeyboardInterrupt:
pass
except ProcessError:
try:
self._shutdown = True
self.logger.reactor(
'Problem when sending message(s) to helper program, stopping',
'error')
except KeyboardInterrupt:
pass
except select.error:
try:
self._shutdown = True
self.logger.reactor('problem using select, stopping',
'error')
except KeyboardInterrupt:
pass
# from exabgp.leak import objgraph
# print objgraph.show_most_common_types(limit=20)
# import random
# obj = objgraph.by_type('Route')[random.randint(0,2000)]
# objgraph.show_backrefs([obj], max_depth=10)
def shutdown(self):
"""terminate all the current BGP connections"""
self.logger.reactor('performing shutdown')
if self.listener:
self.listener.stop()
self.listener = None
for key in self.peers.keys():
self.peers[key].stop()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load(self):
"""reload the configuration and send to the peer the route which changed"""
self.logger.reactor('performing reload of exabgp %s' % version)
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.configuration(
'problem with the configuration file, no change done', 'error')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.configuration(str(self.configuration.error), 'error')
return False
for key, peer in self.peers.items():
if key not in self.configuration.neighbors:
self.logger.reactor('removing peer: %s' % peer.neighbor.name())
peer.stop()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self.peers:
self.logger.reactor('new peer: %s' % neighbor.name())
peer = Peer(neighbor, self)
self.peers[key] = peer
# modified peer
elif self.peers[key].neighbor != neighbor:
self.logger.reactor(
'peer definition change, establishing a new connection for %s'
% str(key))
self.peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.reactor(
'peer definition identical, updating peer routes if required for %s'
% str(key))
self.peers[key].reconfigure(neighbor)
self.logger.configuration('loaded new configuration successfully',
'info')
return True
def schedule(self):
try:
# read at least on message per process if there is some and parse it
for service, command in self.processes.received():
self.api.text(self, service, command)
# if we have nothing to do, return or save the work
if not self._running:
if not self._pending:
return False
self._running, name = self._pending.popleft()
self.logger.reactor('callback | installing %s' % name)
if self._running:
# run it
try:
self.logger.reactor('callback | running')
six.next(self._running) # run
# should raise StopIteration in most case
# and prevent us to have to run twice to run one command
six.next(self._running) # run
except StopIteration:
self._running = None
self.logger.reactor('callback | removing')
return True
except StopIteration:
pass
except KeyboardInterrupt:
self._shutdown = True
self.logger.reactor('^C received', 'error')
def route_send(self):
"""the process ran and we need to figure what routes to changes"""
self.logger.reactor('performing dynamic route update')
for key in self.configuration.neighbors.keys():
self.peers[key].send_new()
self.logger.reactor('updated peers dynamic routes successfully')
def restart(self):
"""kill the BGP session and restart it"""
self.logger.reactor('performing restart of exabgp %s' % version)
self.configuration.reload()
for key in self.peers.keys():
if key not in self.configuration.neighbors.keys():
neighbor = self.configuration.neighbors[key]
self.logger.reactor('removing Peer %s' % neighbor.name())
self.peers[key].stop()
else:
self.peers[key].reestablish()
self.processes.terminate()
self.processes.start()
def unschedule(self, peer):
if peer in self.peers:
del self.peers[peer]
def answer(self, service, string):
self.processes.write(service, string)
self.logger.reactor('responding to %s : %s' %
(service, string.replace('\n', '\\n')))
def api_shutdown(self):
self._shutdown = True
self._pending = deque()
self._running = None
def api_reload(self):
self._reload = True
self._pending = deque()
self._running = None
def api_restart(self):
self._restart = True
self._pending = deque()
self._running = None
@staticmethod
def match_neighbor(description, name):
for string in description:
if re.search(r'(^|[\s])%s($|[\s,])' % re.escape(string),
name) is None:
return False
return True
def match_neighbors(self, descriptions):
"""return the sublist of peers matching the description passed, or None if no description is given"""
if not descriptions:
return self.peers.keys()
returned = []
for key in self.peers:
for description in descriptions:
if Reactor.match_neighbor(description, key):
if key not in returned:
returned.append(key)
return returned
def nexthops(self, peers):
return dict(
(peer, self.peers[peer].neighbor.local_address) for peer in peers)
def plan(self, callback, name):
self._pending.append((callback, name))
class Reactor(object):
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = concat_bytes_i(
character(int(c, 16))
for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a'])
def __init__(self, configurations):
self.ips = environment.settings().tcp.bind
self.port = environment.settings().tcp.port
self.ack = environment.settings().api.ack
self.max_loop_time = environment.settings().reactor.speed
self.early_drop = environment.settings().daemon.drop
self.logger = Logger()
self.daemon = Daemon(self)
self.processes = None
self.listener = None
self.configuration = Configuration(configurations)
self.api = API(self)
self.peers = {}
self.route_update = False
self._stopping = environment.settings().tcp.once
self._shutdown = False
self._reload = False
self._reload_processes = False
self._restart = False
self._saved_pid = False
self._running = None
self._pending = deque()
self._async = deque()
self._signal = {}
signal.signal(signal.SIGTERM, self.sigterm)
signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGALRM, self.sigalrm)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
def _termination(self, reason):
while True:
try:
self._shutdown = True
self.logger.reactor(reason, 'warning')
break
except KeyboardInterrupt:
pass
def sigterm(self, signum, frame):
self.logger.reactor('SIG TERM received - shutdown')
self._shutdown = True
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self._signal[key] = signum
def sighup(self, signum, frame):
self.logger.reactor('SIG HUP received - shutdown')
self._shutdown = True
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self._signal[key] = signum
def sigalrm(self, signum, frame):
self.logger.reactor('SIG ALRM received - restart')
self._restart = True
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self._signal[key] = signum
def sigusr1(self, signum, frame):
self.logger.reactor('SIG USR1 received - reload configuration')
self._reload = True
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self._signal[key] = signum
def sigusr2(self, signum, frame):
self.logger.reactor(
'SIG USR2 received - reload configuration and processes')
self._reload = True
self._reload_processes = True
for key in self.peers:
if self.peers[key].neighbor.api['signal']:
self._signal[key] = signum
def _api_ready(self, sockets):
sleeptime = self.max_loop_time / 20
fds = self.processes.fds()
ios = fds + sockets
try:
read, _, _ = select.select(ios, [], [], sleeptime)
for fd in fds:
if fd in read:
read.remove(fd)
return read
except select.error as exc:
errno, message = exc.args # pylint: disable=W0633
if errno not in error.block:
raise exc
return []
except socket.error as exc:
if exc.errno in error.fatal:
raise exc
return []
except KeyboardInterrupt:
self._termination('^C received')
return []
def _setup_listener(self, local_addr, remote_addr, port, md5_password,
md5_base64, ttl_in):
try:
if not self.listener:
self.listener = Listener()
if not remote_addr:
remote_addr = IP.create(
'0.0.0.0') if local_addr.ipv4() else IP.create('::')
self.listener.listen(local_addr, remote_addr, port, md5_password,
md5_base64, ttl_in)
self.logger.reactor(
'Listening for BGP session(s) on %s:%d%s' %
(local_addr, port, ' with MD5' if md5_password else ''))
return True
except NetworkError as exc:
if os.geteuid() != 0 and port <= 1024:
self.logger.reactor(
'Can not bind to %s:%d, you may need to run ExaBGP as root'
% (local_addr, port), 'critical')
else:
self.logger.reactor(
'Can not bind to %s:%d (%s)' %
(local_addr, port, str(exc)), 'critical')
self.logger.reactor(
'unset exabgp.tcp.bind if you do not want listen for incoming connections',
'critical')
self.logger.reactor(
'and check that no other daemon is already binding to port %d'
% port, 'critical')
return False
def _handle_listener(self):
if not self.listener:
return
ranged_neighbor = []
for connection in self.listener.connected():
for key in self.peers:
peer = self.peers[key]
neighbor = peer.neighbor
connection_local = IP.create(connection.local).address()
neighbor_peer_start = neighbor.peer_address.address()
neighbor_peer_next = neighbor_peer_start + neighbor.range_size
if not neighbor_peer_start <= connection_local < neighbor_peer_next:
continue
connection_peer = IP.create(connection.peer).address()
neighbor_local = neighbor.local_address.address()
if connection_peer != neighbor_local:
if not neighbor.auto_discovery:
continue
# we found a range matching for this connection
# but the peer may already have connected, so
# we need to iterate all individual peers before
# handling "range" peers
if neighbor.range_size > 1:
ranged_neighbor.append(peer.neighbor)
continue
denied = peer.handle_connection(connection)
if denied:
self.logger.reactor(
'refused connection from %s due to the state machine' %
connection.name())
self._async.append(denied)
break
self.logger.reactor('accepted connection from %s' %
connection.name())
break
else:
# we did not break (and nothign was found/done or we have group match)
matched = len(ranged_neighbor)
if matched > 1:
self.logger.reactor(
'could not accept connection from %s (more than one neighbor match)'
% connection.name())
self._async.append(
connection.notification(
6, 5,
b'could not accept the connection (more than one neighbor match)'
))
return
if not matched:
self.logger.reactor('no session configured for %s' %
connection.name())
self._async.append(
connection.notification(
6, 3, b'no session configured for the peer'))
return
new_neighbor = copy.copy(ranged_neighbor[0])
new_neighbor.range_size = 1
new_neighbor.generated = True
new_neighbor.local_address = IP.create(connection.peer)
new_neighbor.peer_address = IP.create(connection.local)
new_peer = Peer(new_neighbor, self)
denied = new_peer.handle_connection(connection)
if denied:
self.logger.reactor(
'refused connection from %s due to the state machine' %
connection.name())
self._async.append(denied)
return
self.peers[new_neighbor.name()] = new_peer
return
def run(self, validate):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes(self)
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
for ip in self.ips:
if not self._setup_listener(ip, None, self.port, None, False,
None):
return False
if not self.load():
return False
if validate: # only validate configuration
self.logger.configuration('')
self.logger.configuration('Parsed Neighbors, un-templated')
self.logger.configuration('------------------------------')
self.logger.configuration('')
for key in self.peers:
self.logger.configuration(str(self.peers[key].neighbor))
self.logger.configuration('')
return True
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
if not self._setup_listener(
neighbor.md5_ip, neighbor.peer_address,
neighbor.listen, neighbor.md5_password,
neighbor.md5_base64, neighbor.ttl_in):
return False
if not self.early_drop:
self.processes.start()
if not self.daemon.drop_privileges():
self.logger.reactor(
'Could not drop privileges to \'%s\' refusing to run as root' %
self.daemon.user, 'critical')
self.logger.reactor(
'Set the environmemnt value exabgp.daemon.user to change the unprivileged user',
'critical')
return
if self.early_drop:
self.processes.start()
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.reactor('Could not setup the logger, aborting',
'critical')
return
if not self.daemon.savepid():
return
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = True
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.reactor('waiting for %d seconds before connecting' %
sleeptime)
time.sleep(float(sleeptime))
workers = {}
peers = set()
busy = False
while True:
try:
start = time.time()
end = start + self.max_loop_time
if self._shutdown:
self._shutdown = False
self.shutdown()
break
if self._reload and reload_completed:
self._reload = False
self.load()
self.processes.start(self._reload_processes)
self._reload_processes = False
elif self._restart:
self._restart = False
self.restart()
# We got some API routes to announce
if self.route_update:
self.route_update = False
self.route_send()
for key, peer in self.peers.items():
if not peer.neighbor.passive or peer.proto:
peers.add(key)
if key in self._signal:
self.peers[key].reactor.processes.signal(
self.peers[key].neighbor, self._signal[key])
self._signal = {}
# check all incoming connection
self._handle_listener()
# give a turn to all the peers
while start < time.time() < end:
for key in list(peers):
peer = self.peers[key]
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
self._unschedule(key)
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
for io in peer.sockets():
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
# handle API calls
busy = self._scheduled_api()
# handle new connections
busy |= self._scheduled_listener()
if not peers and not busy:
break
if not peers:
reload_completed = True
for io in self._api_ready(list(workers)):
peers.add(workers[io])
del workers[io]
if self._stopping and not self.peers.keys():
break
except KeyboardInterrupt:
self._termination('^C received')
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
self._termination('socket error received')
except IOError:
self._termination(
'I/O Error received, most likely ^C during IO')
except SystemExit:
self._termination('exiting')
except ProcessError:
self._termination(
'Problem when sending message(s) to helper program, stopping'
)
except select.error:
self._termination('problem using select, stopping')
def shutdown(self):
"""terminate all the current BGP connections"""
self.logger.reactor('performing shutdown')
if self.listener:
self.listener.stop()
self.listener = None
for key in self.peers.keys():
self.peers[key].stop()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load(self):
"""reload the configuration and send to the peer the route which changed"""
self.logger.reactor('performing reload of exabgp %s' % version)
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.configuration(
'problem with the configuration file, no change done', 'error')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.configuration(str(self.configuration.error), 'error')
return False
for key, peer in self.peers.items():
if key not in self.configuration.neighbors:
self.logger.reactor('removing peer: %s' % peer.neighbor.name())
peer.stop()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self.peers:
self.logger.reactor('new peer: %s' % neighbor.name())
peer = Peer(neighbor, self)
self.peers[key] = peer
# modified peer
elif self.peers[key].neighbor != neighbor:
self.logger.reactor(
'peer definition change, establishing a new connection for %s'
% str(key))
self.peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.reactor(
'peer definition identical, updating peer routes if required for %s'
% str(key))
self.peers[key].reconfigure(neighbor)
for ip in self.ips:
if ip.afi == neighbor.peer_address.afi:
self._setup_listener(ip, neighbor.peer_address, self.port,
neighbor.md5_password,
neighbor.md5_base64, None)
self.logger.configuration('loaded new configuration successfully',
'info')
return True
def _scheduled_listener(self, flipflop=[]):
try:
for generator in self._async:
try:
six.next(generator)
six.next(generator)
flipflop.append(generator)
except StopIteration:
pass
self._async, flipflop = flipflop, self._async
return len(self._async) > 0
except KeyboardInterrupt:
self._termination('^C received')
return False
def _scheduled_api(self):
try:
# read at least on message per process if there is some and parse it
for service, command in self.processes.received():
self.api.text(self, service, command)
# if we have nothing to do, return or save the work
if not self._running:
if not self._pending:
return False
self._running, name = self._pending.popleft()
self.logger.reactor('callback | installing %s' % name)
if self._running:
# run it
try:
self.logger.reactor('callback | running')
six.next(self._running) # run
# should raise StopIteration in most case
# and prevent us to have to run twice to run one command
six.next(self._running) # run
except StopIteration:
self._running = None
self.logger.reactor('callback | removing')
return True
return False
except KeyboardInterrupt:
self._termination('^C received')
return False
def route_send(self):
"""the process ran and we need to figure what routes to changes"""
self.logger.reactor('performing dynamic route update')
for key in self.configuration.neighbors.keys():
self.peers[key].send_new()
self.logger.reactor('updated peers dynamic routes successfully')
def restart(self):
"""kill the BGP session and restart it"""
self.logger.reactor('performing restart of exabgp %s' % version)
self.configuration.reload()
for key in self.peers.keys():
if key not in self.configuration.neighbors.keys():
neighbor = self.configuration.neighbors[key]
self.logger.reactor('removing Peer %s' % neighbor.name())
self.peers[key].stop()
else:
self.peers[key].reestablish()
self.processes.terminate()
self.processes.start()
def _unschedule(self, peer):
if peer in self.peers:
del self.peers[peer]
def answer(self, service, string):
if self.ack:
self.always_answer(service, string)
def always_answer(self, service, string):
self.processes.write(service, string)
self.logger.reactor('responding to %s : %s' %
(service, string.replace('\n', '\\n')))
def api_shutdown(self):
self._shutdown = True
self._pending = deque()
self._running = None
def api_reload(self):
self._reload = True
self._pending = deque()
self._running = None
def api_restart(self):
self._restart = True
self._pending = deque()
self._running = None
@staticmethod
def match_neighbor(description, name):
for string in description:
if re.search(r'(^|[\s])%s($|[\s,])' % re.escape(string),
name) is None:
return False
return True
def match_neighbors(self, descriptions):
"""return the sublist of peers matching the description passed, or None if no description is given"""
if not descriptions:
return self.peers.keys()
returned = []
for key in self.peers:
for description in descriptions:
if Reactor.match_neighbor(description, key):
if key not in returned:
returned.append(key)
return returned
def nexthops(self, peers):
return dict(
(peer, self.peers[peer].neighbor.local_address) for peer in peers)
def plan(self, callback, name):
self._pending.append((callback, name))
class Reactor (object):
# [hex(ord(c)) for c in os.popen('clear').read()]
clear = b''.join([chr_(int(c,16)) for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a']])
def __init__ (self, configurations):
self.ip = environment.settings().tcp.bind
self.port = environment.settings().tcp.port
self.respawn = environment.settings().api.respawn
self.max_loop_time = environment.settings().reactor.speed
self.early_drop = environment.settings().daemon.drop
self.logger = Logger()
self.daemon = Daemon(self)
self.processes = None
self.listener = None
self.configuration = Configuration(configurations)
self.api = API(self)
self.peers = {}
self.route_update = False
self._stopping = environment.settings().tcp.once
self._shutdown = False
self._reload = False
self._reload_processes = False
self._restart = False
self._saved_pid = False
self._pending = deque()
self._running = None
signal.signal(signal.SIGTERM, self.sigterm)
signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGALRM, self.sigalrm)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
def sigterm (self, signum, frame):
self.logger.reactor('SIG TERM received - shutdown')
self._shutdown = True
def sighup (self, signum, frame):
self.logger.reactor('SIG HUP received - shutdown')
self._shutdown = True
def sigalrm (self, signum, frame):
self.logger.reactor('SIG ALRM received - restart')
self._restart = True
def sigusr1 (self, signum, frame):
self.logger.reactor('SIG USR1 received - reload configuration')
self._reload = True
def sigusr2 (self, signum, frame):
self.logger.reactor('SIG USR2 received - reload configuration and processes')
self._reload = True
self._reload_processes = True
def ready (self, sockets, ios, sleeptime=0):
# never sleep a negative number of second (if the rounding is negative somewhere)
# never sleep more than one second (should the clock time change during two time.time calls)
sleeptime = min(max(0.0,sleeptime),1.0)
if not ios:
time.sleep(sleeptime)
return []
try:
read,_,_ = select.select(sockets+ios,[],[],sleeptime)
return read
except select.error as exc:
errno,message = exc.args # pylint: disable=W0633
if errno not in error.block:
raise exc
return []
except socket.error as exc:
if exc.errno in error.fatal:
raise exc
return []
def run (self):
self.daemon.daemonise()
# Make sure we create processes once we have closed file descriptor
# unfortunately, this must be done before reading the configuration file
# so we can not do it with dropped privileges
self.processes = Processes(self)
# we have to read the configuration possibly with root privileges
# as we need the MD5 information when we bind, and root is needed
# to bind to a port < 1024
# this is undesirable as :
# - handling user generated data as root should be avoided
# - we may not be able to reload the configuration once the privileges are dropped
# but I can not see any way to avoid it
if not self.load():
return False
try:
self.listener = Listener()
if self.ip:
self.listener.listen(IP.create(self.ip),IP.create('0.0.0.0'),self.port,None,None)
self.logger.reactor('Listening for BGP session(s) on %s:%d' % (self.ip,self.port))
for neighbor in self.configuration.neighbors.values():
if neighbor.listen:
self.listener.listen(neighbor.md5_ip,neighbor.peer_address,neighbor.listen,neighbor.md5_password,neighbor.ttl_in)
self.logger.reactor('Listening for BGP session(s) on %s:%d%s' % (neighbor.md5_ip,neighbor.listen,' with MD5' if neighbor.md5_password else ''))
except NetworkError as exc:
self.listener = None
if os.geteuid() != 0 and self.port <= 1024:
self.logger.reactor('Can not bind to %s:%d, you may need to run ExaBGP as root' % (self.ip,self.port),'critical')
else:
self.logger.reactor('Can not bind to %s:%d (%s)' % (self.ip,self.port,str(exc)),'critical')
self.logger.reactor('unset exabgp.tcp.bind if you do not want listen for incoming connections','critical')
self.logger.reactor('and check that no other daemon is already binding to port %d' % self.port,'critical')
sys.exit(1)
if not self.early_drop:
self.processes.start()
if not self.daemon.drop_privileges():
self.logger.reactor('Could not drop privileges to \'%s\' refusing to run as root' % self.daemon.user,'critical')
self.logger.reactor('Set the environmemnt value exabgp.daemon.user to change the unprivileged user','critical')
return
if self.early_drop:
self.processes.start()
# This is required to make sure we can write in the log location as we now have dropped root privileges
if not self.logger.restart():
self.logger.reactor('Could not setup the logger, aborting','critical')
return
if not self.daemon.savepid():
return
# did we complete the run of updates caused by the last SIGUSR1/SIGUSR2 ?
reload_completed = True
wait = environment.settings().tcp.delay
if wait:
sleeptime = (wait * 60) - int(time.time()) % (wait * 60)
self.logger.reactor('waiting for %d seconds before connecting' % sleeptime)
time.sleep(float(sleeptime))
workers = {}
peers = set()
scheduled = False
while True:
try:
finished = False
start = time.time()
end = start + self.max_loop_time
if self._shutdown:
self._shutdown = False
self.shutdown()
break
if self._reload and reload_completed:
self._reload = False
self.load()
self.processes.start(self._reload_processes)
self._reload_processes = False
elif self._restart:
self._restart = False
self.restart()
# We got some API routes to announce
if self.route_update:
self.route_update = False
self.route_send()
for peer in self.peers.keys():
peers.add(peer)
while start < time.time() < end and not finished:
if self.peers:
for key in list(peers):
peer = self.peers[key]
action = peer.run()
# .run() returns an ACTION enum:
# * immediate if it wants to be called again
# * later if it should be called again but has no work atm
# * close if it is finished and is closing down, or restarting
if action == ACTION.CLOSE:
self.unschedule(peer)
peers.discard(key)
# we are loosing this peer, not point to schedule more process work
elif action == ACTION.LATER:
for io in peer.sockets():
workers[io] = key
# no need to come back to it before a a full cycle
peers.discard(key)
if not peers:
reload_completed = True
if self.listener:
for connection in self.listener.connected():
# found
# * False, not peer found for this TCP connection
# * True, peer found
# * None, conflict found for this TCP connections
found = False
for key in self.peers:
peer = self.peers[key]
neighbor = peer.neighbor
# XXX: FIXME: Inet can only be compared to Inet
if connection.local == str(neighbor.peer_address) and connection.peer == str(neighbor.local_address):
if peer.incoming(connection):
found = True
break
found = None
break
if found:
self.logger.reactor('accepted connection from %s - %s' % (connection.local,connection.peer))
elif found is False:
self.logger.reactor('no session configured for %s - %s' % (connection.local,connection.peer))
connection.notification(6,3,'no session configured for the peer')
connection.close()
elif found is None:
self.logger.reactor('connection refused (already connected to the peer) %s - %s' % (connection.local,connection.peer))
connection.notification(6,5,'could not accept the connection')
connection.close()
scheduled = self.schedule()
finished = not peers and not scheduled
# RFC state that we MUST not send more than one KEEPALIVE / sec
# And doing less could cause the session to drop
if finished:
for io in self.ready(list(peers),self.processes.fds(),end-time.time()):
if io in workers:
peers.add(workers[io])
del workers[io]
if self._stopping and not self.peers.keys():
break
except KeyboardInterrupt:
while True:
try:
self._shutdown = True
self.logger.reactor('^C received')
break
except KeyboardInterrupt:
pass
# socket.error is a subclass of IOError (so catch it first)
except socket.error:
try:
self._shutdown = True
self.logger.reactor('socket error received','warning')
break
except KeyboardInterrupt:
pass
except IOError:
while True:
try:
self._shutdown = True
self.logger.reactor('I/O Error received, most likely ^C during IO','warning')
break
except KeyboardInterrupt:
pass
except SystemExit:
try:
self._shutdown = True
self.logger.reactor('exiting')
break
except KeyboardInterrupt:
pass
except ProcessError:
try:
self._shutdown = True
self.logger.reactor('Problem when sending message(s) to helper program, stopping','error')
except KeyboardInterrupt:
pass
except select.error:
try:
self._shutdown = True
self.logger.reactor('problem using select, stopping','error')
except KeyboardInterrupt:
pass
# from exabgp.leak import objgraph
# print objgraph.show_most_common_types(limit=20)
# import random
# obj = objgraph.by_type('Route')[random.randint(0,2000)]
# objgraph.show_backrefs([obj], max_depth=10)
def shutdown (self):
"""terminate all the current BGP connections"""
self.logger.reactor('performing shutdown')
if self.listener:
self.listener.stop()
self.listener = None
for key in self.peers.keys():
self.peers[key].stop()
self.processes.terminate()
self.daemon.removepid()
self._stopping = True
def load (self):
"""reload the configuration and send to the peer the route which changed"""
self.logger.reactor('performing reload of exabgp %s' % version)
reloaded = self.configuration.reload()
if not reloaded:
#
# Careful the string below is used but the QA code to check for sucess of failure
self.logger.configuration('problem with the configuration file, no change done','error')
# Careful the string above is used but the QA code to check for sucess of failure
#
self.logger.configuration(str(self.configuration.error),'error')
return False
for key, peer in self.peers.items():
if key not in self.configuration.neighbors:
self.logger.reactor('removing peer: %s' % peer.neighbor.name())
peer.stop()
for key, neighbor in self.configuration.neighbors.items():
# new peer
if key not in self.peers:
self.logger.reactor('new peer: %s' % neighbor.name())
peer = Peer(neighbor,self)
self.peers[key] = peer
# modified peer
elif self.peers[key].neighbor != neighbor:
self.logger.reactor('peer definition change, establishing a new connection for %s' % str(key))
self.peers[key].reestablish(neighbor)
# same peer but perhaps not the routes
else:
# finding what route changed and sending the delta is not obvious
self.logger.reactor('peer definition identical, updating peer routes if required for %s' % str(key))
self.peers[key].reconfigure(neighbor)
self.logger.configuration('loaded new configuration successfully','info')
return True
def schedule (self):
try:
# read at least on message per process if there is some and parse it
for service,command in self.processes.received():
self.api.text(self,service,command)
# if we have nothing to do, return or save the work
if not self._running:
if not self._pending:
return False
self._running,name = self._pending.popleft()
self.logger.reactor('callback | installing %s' % name)
if self._running:
# run it
try:
self.logger.reactor('callback | running')
six.next(self._running) # run
# should raise StopIteration in most case
# and prevent us to have to run twice to run one command
six.next(self._running) # run
except StopIteration:
self._running = None
self.logger.reactor('callback | removing')
return True
except StopIteration:
pass
except KeyboardInterrupt:
self._shutdown = True
self.logger.reactor('^C received','error')
def route_send (self):
"""the process ran and we need to figure what routes to changes"""
self.logger.reactor('performing dynamic route update')
for key in self.configuration.neighbors.keys():
self.peers[key].send_new()
self.logger.reactor('updated peers dynamic routes successfully')
def restart (self):
"""kill the BGP session and restart it"""
self.logger.reactor('performing restart of exabgp %s' % version)
self.configuration.reload()
for key in self.peers.keys():
if key not in self.configuration.neighbors.keys():
neighbor = self.configuration.neighbors[key]
self.logger.reactor('removing Peer %s' % neighbor.name())
self.peers[key].stop()
else:
self.peers[key].reestablish()
self.processes.terminate()
self.processes.start()
def unschedule (self, peer):
key = peer.neighbor.name()
if key in self.peers:
del self.peers[key]
def answer (self, service, string):
self.processes.write(service,string)
self.logger.reactor('responding to %s : %s' % (service,string.replace('\n','\\n')))
def api_shutdown (self):
self._shutdown = True
self._pending = deque()
self._running = None
def api_reload (self):
self._reload = True
self._pending = deque()
self._running = None
def api_restart (self):
self._restart = True
self._pending = deque()
self._running = None
@staticmethod
def match_neighbor (description, name):
for string in description:
if re.search(r'(^|[\s])%s($|[\s,])' % re.escape(string), name) is None:
return False
return True
def match_neighbors (self, descriptions):
"""return the sublist of peers matching the description passed, or None if no description is given"""
if not descriptions:
return self.peers.keys()
returned = []
for key in self.peers:
for description in descriptions:
if Reactor.match_neighbor(description,key):
if key not in returned:
returned.append(key)
return returned
def nexthops (self, peers):
return dict((peer,self.peers[peer].neighbor.local_address) for peer in peers)
def plan (self, callback,name):
self._pending.append((callback,name))