class ChildFactory:
def preExec(self):
os.setpgrp()
def __init__(self, configuration, name):
self.log = Logger('worker ' + str(name), configuration.log.worker)
def createProcess(self, program, universal=False):
try:
process = subprocess.Popen(
program.split(' '),
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=universal,
preexec_fn=self.preExec,
)
self.log.debug('spawn process %s' % program)
except KeyboardInterrupt:
process = None
except (subprocess.CalledProcessError, OSError, ValueError):
self.log.error('could not spawn process %s' % program)
process = None
if process:
try:
fcntl.fcntl(process.stderr, fcntl.F_SETFL, os.O_NONBLOCK)
except IOError:
self.destroyProcess(process)
process = None
return process
def destroyProcess(self, process):
try:
process.terminate()
process.wait()
self.log.info('terminated process PID %s' % process.pid)
except OSError, e:
# No such processs
if e[0] != errno.ESRCH:
self.log.error('PID %s died' % process.pid)
class ChildFactory:
def preExec (self):
os.setpgrp()
def __init__ (self, configuration, name):
self.log = Logger('worker ' + str(name), configuration.log.worker)
def createProcess (self, program, universal=False):
try:
process = subprocess.Popen([program],
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=universal,
preexec_fn=self.preExec,
)
self.log.debug('spawn process %s' % program)
except KeyboardInterrupt:
process = None
except (subprocess.CalledProcessError,OSError,ValueError):
self.log.error('could not spawn process %s' % program)
process = None
if process:
try:
fcntl.fcntl(process.stderr, fcntl.F_SETFL, os.O_NONBLOCK)
except IOError:
self.destroyProcess(process)
process = None
return process
def destroyProcess (self, process):
try:
process.terminate()
process.wait()
self.log.info('terminated process PID %s' % process.pid)
except OSError, e:
# No such processs
if e[0] != errno.ESRCH:
self.log.error('PID %s died' % process.pid)
class Supervisor (object):
alarm_time = 0.1 # regular backend work
second_frequency = int(1/alarm_time) # when we record history
minute_frequency = int(60/alarm_time) # when we want to average history
increase_frequency = int(5/alarm_time) # when we add workers
decrease_frequency = int(60/alarm_time) # when we remove workers
saturation_frequency = int(20/alarm_time) # when we report connection saturation
interface_frequency = int(300/alarm_time) # when we check for new interfaces
# import os
# clear = [hex(ord(c)) for c in os.popen('clear').read()]
# clear = ''.join([chr(int(c,16)) for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a']])
def __init__ (self,configuration):
configuration = load()
self.configuration = configuration
# Only here so the introspection code can find them
self.log = Logger('supervisor', configuration.log.supervisor)
self.log.error('Starting exaproxy version %s' % configuration.proxy.version)
self.signal_log = Logger('signal', configuration.log.signal)
self.log_writer = SysLogWriter('log', configuration.log.destination, configuration.log.enable, level=configuration.log.level)
self.usage_writer = UsageWriter('usage', configuration.usage.destination, configuration.usage.enable)
sys.exitfunc = self.log_writer.writeMessages
self.log_writer.setIdentifier(configuration.daemon.identifier)
#self.usage_writer.setIdentifier(configuration.daemon.identifier)
if configuration.debug.log:
self.log_writer.toggleDebug()
self.usage_writer.toggleDebug()
self.log.error('python version %s' % sys.version.replace(os.linesep,' '))
self.log.debug('starting %s' % sys.argv[0])
self.pid = PID(self.configuration)
self.daemon = Daemon(self.configuration)
self.poller = Poller(self.configuration.daemon)
self.poller.setupRead('read_proxy') # Listening proxy sockets
self.poller.setupRead('read_web') # Listening webserver sockets
self.poller.setupRead('read_icap') # Listening icap sockets
self.poller.setupRead('read_redirector') # Pipes carrying responses from the redirector process
self.poller.setupRead('read_resolver') # Sockets currently listening for DNS responses
self.poller.setupRead('read_client') # Active clients
self.poller.setupRead('opening_client') # Clients we have not yet read a request from
self.poller.setupWrite('write_client') # Active clients with buffered data to send
self.poller.setupWrite('write_resolver') # Active DNS requests with buffered data to send
self.poller.setupRead('read_download') # Established connections
self.poller.setupWrite('write_download') # Established connections we have buffered data to send to
self.poller.setupWrite('opening_download') # Opening connections
self.monitor = Monitor(self)
self.page = Page(self)
self.content = ContentManager(self,configuration)
self.client = ClientManager(self.poller, configuration)
self.resolver = ResolverManager(self.poller, self.configuration, configuration.dns.retries*10)
self.proxy = Server('http proxy',self.poller,'read_proxy', configuration.http.connections)
self.web = Server('web server',self.poller,'read_web', configuration.web.connections)
self.icap = Server('icap server',self.poller,'read_icap', configuration.icap.connections)
self._shutdown = True if self.daemon.filemax == 0 else False # stop the program
self._softstop = False # stop once all current connection have been dealt with
self._reload = False # unimplemented
self._toggle_debug = False # start logging a lot
self._decrease_spawn_limit = 0
self._increase_spawn_limit = 0
self._refork = False # unimplemented
self._pdb = False # turn on pdb debugging
self._listen = None # listening change ? None: no, True: listen, False: stop listeing
self.wait_time = 5.0 # how long do we wait at maximum once we have been soft-killed
self.local = set() # what addresses are on our local interfaces
if not self.initialise():
self._shutdown = True
elif self.daemon.drop_privileges():
self.log.critical('Could not drop privileges to \'%s\'. Refusing to run as root' % self.daemon.user)
self.log.critical('Set the environment value USER to change the unprivileged user')
self._shutdown = True
# fork the redirector process before performing any further setup
redirector = fork_redirector(self.poller, self.configuration)
# create threads _after_ all forking is done
self.redirector = redirector_message_thread(redirector)
self.reactor = Reactor(self.configuration, self.web, self.proxy, self.icap, self.redirector, self.content, self.client, self.resolver, self.log_writer, self.usage_writer, self.poller)
self.interfaces()
signal.signal(signal.SIGQUIT, self.sigquit)
signal.signal(signal.SIGINT, self.sigterm)
signal.signal(signal.SIGTERM, self.sigterm)
# signal.signal(signal.SIGABRT, self.sigabrt)
# signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGTRAP, self.sigtrap)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
signal.signal(signal.SIGTTOU, self.sigttou)
signal.signal(signal.SIGTTIN, self.sigttin)
signal.signal(signal.SIGALRM, self.sigalrm)
# make sure we always have data in history
# (done in zero for dependencies reasons)
self.monitor.zero()
def exit (self):
sys.exit()
def sigquit (self,signum, frame):
if self._softstop:
self.signal_log.critical('multiple SIG INT received, shutdown')
self._shutdown = True
else:
self.signal_log.critical('SIG INT received, soft-stop')
self._softstop = True
self._listen = False
def sigterm (self,signum, frame):
self.signal_log.critical('SIG TERM received, shutdown request')
if os.environ.get('PDB',False):
self._pdb = True
else:
self._shutdown = True
# def sigabrt (self,signum, frame):
# self.signal_log.info('SIG INFO received, refork request')
# self._refork = True
# def sighup (self,signum, frame):
# self.signal_log.info('SIG HUP received, reload request')
# self._reload = True
def sigtrap (self,signum, frame):
self.signal_log.critical('SIG TRAP received, toggle debug')
self._toggle_debug = True
def sigusr1 (self,signum, frame):
self.signal_log.critical('SIG USR1 received, decrease worker number')
self._decrease_spawn_limit += 1
def sigusr2 (self,signum, frame):
self.signal_log.critical('SIG USR2 received, increase worker number')
self._increase_spawn_limit += 1
def sigttou (self,signum, frame):
self.signal_log.critical('SIG TTOU received, stop listening')
self._listen = False
def sigttin (self,signum, frame):
self.signal_log.critical('SIG IN received, star listening')
self._listen = True
def sigalrm (self,signum, frame):
self.reactor.running = False
signal.setitimer(signal.ITIMER_REAL,self.alarm_time,self.alarm_time)
def interfaces (self):
local = set(['127.0.0.1','::1'])
for interface in getifaddrs():
if interface.family not in (AF_INET,AF_INET6):
continue
if interface.address not in self.local:
self.log.info('found new local ip %s (%s)' % (interface.address,interface.name))
local.add(interface.address)
for ip in self.local:
if ip not in local:
self.log.info('removed local ip %s' % ip)
if local == self.local:
self.log.info('no ip change')
else:
self.local = local
def run (self):
signal.setitimer(signal.ITIMER_REAL,self.alarm_time,self.alarm_time)
count_second = 0
count_minute = 0
count_saturation = 0
count_interface = 0
while True:
count_second = (count_second + 1) % self.second_frequency
count_minute = (count_minute + 1) % self.minute_frequency
count_saturation = (count_saturation + 1) % self.saturation_frequency
count_interface = (count_interface + 1) % self.interface_frequency
try:
if self._pdb:
self._pdb = False
import pdb
pdb.set_trace()
# check for IO change with select
status = self.reactor.run()
if status is False:
self._shutdown = True
# must follow the reactor so we are sure to go through the reactor at least once
# and flush any logs
if self._shutdown:
self._shutdown = False
self.shutdown()
break
elif self._reload:
self._reload = False
self.reload()
elif self._refork:
self._refork = False
self.signal_log.warning('refork not implemented')
# stop listening to new connections
# refork the program (as we have been updated)
# just handle current open connection
if self._softstop:
if self._listen == False:
self.proxy.rejecting()
self._listen = None
if self.client.softstop():
self._shutdown = True
# only change listening if we are not shutting down
elif self._listen is not None:
if self._listen:
self._shutdown = not self.proxy.accepting()
self._listen = None
else:
self.proxy.rejecting()
self._listen = None
if self._toggle_debug:
self._toggle_debug = False
self.log_writer.toggleDebug()
if self._decrease_spawn_limit:
count = self._decrease_spawn_limit
self.redirector.decreaseSpawnLimit(count)
self._decrease_spawn_limit = 0
if self._increase_spawn_limit:
count = self._increase_spawn_limit
self.redirector.increaseSpawnLimit(count)
self._increase_spawn_limit = 0
# save our monitoring stats
if count_second == 0:
self.monitor.second()
expired = self.reactor.client.expire()
else:
expired = 0
if expired:
self.proxy.notifyClose(None, count=expired)
if count_minute == 0:
self.monitor.minute()
# report if we saw too many connections
if count_saturation == 0:
self.proxy.saturation()
self.web.saturation()
if self.configuration.daemon.poll_interfaces and count_interface == 0:
self.interfaces()
except KeyboardInterrupt:
self.log.critical('^C received')
self._shutdown = True
except OSError,e:
# This shoould never happen as we are limiting how many connections we accept
if e.errno == 24: # Too many open files
self.log.critical('Too many opened files, shutting down')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
else:
self.log.critical('unrecoverable io error')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
finally:
try:
poll([f], [], [f], 0.1)
except socket.error:
print "CANNOT POLL (read): %s" % str(f)
log.error('can not poll (read) : %s' % str(f))
for f in write:
try:
poll([], [f], [f], 0.1)
except socket.error:
print "CANNOT POLL (write): %s" % str(f)
log.error('can not poll (write) : %s' % str(f))
raise e
except (ValueError, AttributeError, TypeError), e:
log.error("fatal error encountered during select - %s %s" % (type(e),str(e)))
raise e
except select.error, e:
if e.args[0] in errno_block:
return [], [], []
log.error("fatal error encountered during select - %s %s" % (type(e),str(e)))
raise e
except KeyboardInterrupt,e:
raise e
except Exception, e:
log.critical("fatal error encountered during select - %s %s" % (type(e),str(e)))
raise e
return r, w, x
class ResolverManager(object):
resolverFactory = DNSResolver
def __init__(self, poller, configuration, max_workers):
self.poller = poller
self.configuration = configuration
self.resolver_factory = self.resolverFactory(configuration)
# The actual work is done in the worker
self.worker = self.resolver_factory.createUDPClient()
# All currently active clients (one UDP and many TCP)
self.workers = {}
self.workers[self.worker.socket] = self.worker
self.poller.addReadSocket('read_resolver', self.worker.socket)
# Track the clients currently expecting results
self.clients = {} # client_id : identifier
# Key should be the hostname rather than the request ID?
self.resolving = {} # identifier, worker_id :
# TCP workers that have not yet sent a complete request
self.sending = {} # sock :
# Maximum number of entry we will cache (1024 DNS lookup per second !)
# assuming 1k per entry, which is a lot, it mean 20Mb of memory
# which at the default of 900 seconds of cache is 22 new host per seonds
self.max_entries = 1024 * 20
# track the current queries and when they were started
self.active = []
self.cache = {}
self.cached = deque()
self.max_workers = max_workers
self.worker_count = len(self.workers) # the UDP client
self.waiting = []
self.log = Logger('resolver', configuration.log.resolver)
self.chained = {}
def cacheDestination(self, hostname, ip):
if hostname not in self.cache:
expire_time = time.time() + self.configuration.dns.ttl
expire_time = expire_time - expire_time % 5 # group the DNS record per buckets 5 seconds
latest_time, latest_hosts = self.cached[-1] if self.cached else (
-1, None)
if expire_time > latest_time:
hosts = []
self.cached.append((expire_time, hosts))
else:
hosts = latest_hosts
self.cache[hostname] = ip
hosts.append(hostname)
def expireCache(self):
# expire only one set of cache entries at a time
if self.cached:
current_time = time.time()
expire_time, hosts = self.cached[0]
if current_time >= expire_time or len(
self.cache) > self.max_entries:
expire_time, hosts = self.cached.popleft()
for hostname in hosts:
self.cache.pop(hostname, None)
def cleanup(self):
now = time.time()
cutoff = now - self.configuration.dns.timeout
count = 0
for timestamp, client_id, sock in self.active:
if timestamp > cutoff:
break
count += 1
cli_data = self.clients.pop(client_id, None)
worker = self.workers.get(sock)
tcpudp = 'udp' if worker is self.worker else 'tcp'
if cli_data is not None:
w_id, identifier, active_time, resolve_count = cli_data
data = self.resolving.pop((w_id, identifier), None)
if not data:
data = self.sending.pop(sock, None)
if data:
client_id, original, hostname, command, decision = data
self.log.error(
'timeout when requesting address for %s using the %s client - attempt %s'
% (hostname, tcpudp, resolve_count))
if resolve_count < self.configuration.dns.retries and worker is self.worker:
self.log.info(
'going to retransmit request for %s - attempt %s of %s'
% (hostname, resolve_count + 1,
self.configuration.dns.retries))
self.startResolving(client_id,
command,
decision,
resolve_count + 1,
identifier=identifier)
continue
self.log.error(
'given up trying to resolve %s after %s attempts' %
(hostname, self.configuration.dns.retries))
yield client_id, 'rewrite', '\0'.join(
('503', 'dns.html', '', '', '', hostname, 'peer'))
if worker is not None:
if worker is not self.worker:
worker.close()
self.workers.pop(sock)
if count:
self.active = self.active[count:]
def resolves(self, command, decision):
if command in ('download', 'connect'):
hostname = decision.split('\0')[0]
if isip(hostname):
res = False
else:
res = True
else:
res = False
return res
def extractHostname(self, command, decision):
data = decision.split('\0')
if command == 'download':
hostname = data[0]
elif command == 'connect':
hostname = decision.split('\0')[0]
else:
hostname = None
return hostname
def resolveDecision(self, command, decision, ip):
if command in ('download', 'connect'):
hostname, args = decision.split('\0', 1)
newdecision = '\0'.join((ip, args))
else:
newdecision = None
return newdecision
def startResolving(self,
client_id,
command,
decision,
resolve_count=1,
identifier=None):
hostname = self.extractHostname(command, decision)
if hostname:
# Resolution is already in our cache
if hostname in self.cache:
identifier = None
ip = self.cache[hostname]
if ip is not None:
resolved = self.resolveDecision(command, decision, ip)
response = client_id, command, resolved
else:
newdecision = '\0'.join(
('503', 'dns.html', 'http', '', '', hostname, 'peer'))
response = client_id, 'rewrite', newdecision
# do not try to resolve domains which are not FQDN
elif self.configuration.dns.fqdn and '.' not in hostname:
identifier = None
newdecision = '\0'.join(
('200', 'dns.html', 'http', '', '', hostname, 'peer'))
response = client_id, 'rewrite', newdecision
# each DNS part (between the dots) must be under 256 chars
elif max(len(p) for p in hostname.split('.')) > 255:
identifier = None
self.log.info('jumbo hostname: %s' % hostname)
newdecision = '\0'.join(
('503', 'dns.html', 'http', '', '', hostname, 'peer'))
response = client_id, 'rewrite', newdecision
# Lookup that DNS name
else:
identifier, _ = self.worker.resolveHost(hostname,
identifier=identifier)
response = None
active_time = time.time()
self.resolving[(
self.worker.w_id, identifier
)] = client_id, hostname, hostname, command, decision
self.clients[client_id] = (self.worker.w_id, identifier,
active_time, resolve_count)
self.active.append(
(active_time, client_id, self.worker.socket))
else:
identifier = None
response = None
return identifier, response
def beginResolvingTCP(self, client_id, command, decision, resolve_count):
if self.worker_count < self.max_workers:
identifier = self.newTCPResolver(client_id, command, decision,
resolve_count)
self.worker_count += 1
else:
self.waiting.append((client_id, command, decision, resolve_count))
identifier = None
return identifier
def notifyClose(self):
paused = self.worker_count >= self.max_workers
self.worker_count -= 1
if paused and self.worker_count < self.max_workers:
for _ in range(self.worker_count, self.max_workers):
if self.waiting:
data, self.waiting = self.waiting[0], self.waiting[1:]
client_id, command, decision, resolve_count = data
identifier = self.newTCPResolver(client_id, command,
decision, resolve_count)
self.worker_count += 1
def newTCPResolver(self, client_id, command, decision, resolve_count):
hostname = self.extractHostname(command, decision)
if hostname:
worker = self.resolver_factory.createTCPClient()
self.workers[worker.socket] = worker
identifier, all_sent = worker.resolveHost(hostname)
active_time = time.time()
self.resolving[(
worker.w_id,
identifier)] = client_id, hostname, hostname, command, decision
self.clients[client_id] = (worker.w_id, identifier, active_time,
resolve_count)
self.active.append((active_time, client_id, self.worker.socket))
if all_sent:
self.poller.addReadSocket('read_resolver', worker.socket)
self.resolving[(
worker.w_id, identifier
)] = client_id, hostname, hostname, command, decision
else:
self.poller.addWriteSocket('write_resolver', worker.socket)
self.sending[
worker.
socket] = client_id, hostname, hostname, command, decision
else:
identifier = None
return identifier
def getResponse(self, sock):
worker = self.workers.get(sock)
if worker:
result = worker.getResponse(self.chained)
if result:
identifier, forhost, ip, completed, newidentifier, newhost, newcomplete = result
data = self.resolving.pop((worker.w_id, identifier), None)
chain_count = self.chained.pop(identifier, 0)
if newidentifier:
self.chained[newidentifier] = chain_count + 1
if not data:
self.log.info(
'ignoring response for %s (%s) with identifier %s' %
(forhost, ip, identifier))
else:
# unable to parse response
self.log.error('unable to parse response')
data = None
if data:
client_id, original, hostname, command, decision = data
clidata = self.clients.pop(client_id, None)
if completed:
if clidata is not None:
key = clidata[2], client_id, worker.socket
if key in self.active:
self.active.remove(key)
# check to see if we received an incomplete response
if not completed:
newidentifier = self.beginResolvingTCP(
client_id, command, decision, 1)
newhost = hostname
response = None
# check to see if the worker started a new request
if newidentifier:
if completed:
active_time = time.time()
self.resolving[(
worker.w_id, newidentifier
)] = client_id, original, newhost, command, decision
self.clients[client_id] = (worker.w_id, newidentifier,
active_time, 1)
self.active.append(
(active_time, client_id, worker.socket))
response = None
if completed and newcomplete:
self.poller.addReadSocket('read_resolver',
worker.socket)
elif completed and not newcomplete:
self.poller.addWriteSocket('write_resolver',
worker.socket)
self.sending[
worker.
socket] = client_id, original, hostname, command, decision
# we just started a new (TCP) request and have not yet completely sent it
# make sure we still know who the request is for
elif not completed:
response = None
# maybe we read the wrong response?
elif forhost != hostname:
_, _, _, resolve_count = clidata
active_time = time.time()
self.resolving[(
worker.w_id, identifier
)] = client_id, original, hostname, command, decision
self.clients[client_id] = (worker.w_id, identifier,
active_time, resolve_count)
self.active.append((active_time, client_id, worker.socket))
response = None
# success
elif ip is not None:
resolved = self.resolveDecision(command, decision, ip)
response = client_id, command, resolved
self.cacheDestination(original, ip)
# not found
else:
newdecision = '\0'.join(
('503', 'dns.html', 'http', '', '', hostname, 'peer'))
response = client_id, 'rewrite', newdecision
#self.cacheDestination(original, ip)
else:
response = None
if response or result is None:
if worker is not self.worker:
self.poller.removeReadSocket('read_resolver', sock)
self.poller.removeWriteSocket('write_resolver', sock)
worker.close()
self.workers.pop(sock)
self.notifyClose()
else:
response = None
return response
def continueSending(self, sock):
"""Continue sending data over the connected TCP socket"""
data = self.sending.get(sock)
if data:
client_id, original, hostname, command, decision = data
else:
client_id, original, hostname, command, decision = None, None, None, None, None, None
worker = self.workers[sock]
res = worker.continueSending()
if res is False: # we've sent all we need to send
self.poller.removeWriteSocket('write_resolver', sock)
if client_id in self.clients:
w_id, identifier, active_time, resolve_count = self.clients[
client_id]
tmp = self.sending.pop(sock)
self.resolving[(w_id, identifier)] = tmp
self.poller.addReadSocket('read_resolver', sock)
else:
self.log.error(
'could not find client for dns request for %s. request is being left to timeout.'
% str(hostname))
class ResolverManager (object):
resolverFactory = DNSResolver
def __init__ (self, poller, configuration, max_workers):
self.poller = poller
self.configuration = configuration
self.resolver_factory = self.resolverFactory(configuration)
# The actual work is done in the worker
self.worker = self.resolver_factory.createUDPClient()
# All currently active clients (one UDP and many TCP)
self.workers = {}
self.workers[self.worker.socket] = self.worker
self.poller.addReadSocket('read_resolver', self.worker.socket)
# Track the clients currently expecting results
self.clients = {} # client_id : identifier
# Key should be the hostname rather than the request ID?
self.resolving = {} # identifier, worker_id :
# TCP workers that have not yet sent a complete request
self.sending = {} # sock :
# Maximum number of entry we will cache (1024 DNS lookup per second !)
# assuming 1k per entry, which is a lot, it mean 20Mb of memory
# which at the default of 900 seconds of cache is 22 new host per seonds
self.max_entries = 1024*20
# track the current queries and when they were started
self.active = []
self.cache = {}
self.cached = deque()
self.max_workers = max_workers
self.worker_count = len(self.workers) # the UDP client
self.waiting = []
self.log = Logger('resolver', configuration.log.resolver)
self.chained = {}
def cacheDestination (self, hostname, ip):
if hostname not in self.cache:
expire_time = time.time() + self.configuration.dns.ttl
expire_time = expire_time - expire_time % 5 # group the DNS record per buckets 5 seconds
latest_time, latest_hosts = self.cached[-1] if self.cached else (-1, None)
if expire_time > latest_time:
hosts = []
self.cached.append((expire_time, hosts))
else:
hosts = latest_hosts
self.cache[hostname] = ip
hosts.append(hostname)
def expireCache (self):
# expire only one set of cache entries at a time
if self.cached:
current_time = time.time()
expire_time, hosts = self.cached[0]
if current_time >= expire_time or len(self.cache) > self.max_entries:
expire_time, hosts = self.cached.popleft()
for hostname in hosts:
self.cache.pop(hostname, None)
def cleanup(self):
now = time.time()
cutoff = now - self.configuration.dns.timeout
count = 0
for timestamp, client_id, sock in self.active:
if timestamp > cutoff:
break
count += 1
cli_data = self.clients.pop(client_id, None)
worker = self.workers.get(sock)
tcpudp = 'udp' if worker is self.worker else 'tcp'
if cli_data is not None:
w_id, identifier, active_time, resolve_count = cli_data
data = self.resolving.pop((w_id, identifier), None)
if not data:
data = self.sending.pop(sock, None)
if data:
client_id, original, hostname, command, decision = data
self.log.error('timeout when requesting address for %s using the %s client - attempt %s' % (hostname, tcpudp, resolve_count))
if resolve_count < self.configuration.dns.retries and worker is self.worker:
self.log.info('going to retransmit request for %s - attempt %s of %s' % (hostname, resolve_count+1, self.configuration.dns.retries))
self.startResolving(client_id, command, decision, resolve_count+1, identifier=identifier)
continue
self.log.error('given up trying to resolve %s after %s attempts' % (hostname, self.configuration.dns.retries))
yield client_id, 'rewrite', ('503', 'dns.html', '', '', '', hostname, 'peer')
if worker is not None:
if worker is not self.worker:
worker.close()
self.workers.pop(sock)
if count:
self.active = self.active[count:]
def resolves(self, command, decision):
if command in ('download', 'connect'):
hostname = decision[0]
if isip(hostname):
res = False
else:
res = True
else:
res = False
return res
def extractHostname(self, command, decision):
if command in ('download', 'connect'):
hostname = decision[0]
else:
hostname = None
return hostname
def resolveDecision(self, command, decision, ip):
if command in ('download', 'connect'):
hostname, args = decision[0], decision[1:]
newdecision = (ip,) + args
else:
newdecision = None
return newdecision
def startResolving(self, client_id, command, decision, resolve_count=1, identifier=None):
hostname = self.extractHostname(command, decision)
if hostname:
# Resolution is already in our cache
if hostname in self.cache and identifier is None:
ip = self.cache[hostname]
if ip is not None:
resolved = self.resolveDecision(command, decision, ip)
response = (client_id, command) + resolved
else:
response = client_id, 'rewrite', '503', 'dns.html', 'http', '', '', hostname, 'peer'
# do not try to resolve domains which are not FQDN
elif self.configuration.dns.fqdn and '.' not in hostname:
identifier = None
response = client_id, 'rewrite', '200', 'dns.html', 'http', '', '', hostname, 'peer'
# each DNS part (between the dots) must be under 256 chars
elif max(len(p) for p in hostname.split('.')) > 255:
identifier = None
self.log.info('jumbo hostname: %s' % hostname)
response = client_id, 'rewrite', '503', 'dns.html', 'http', '', '', hostname, 'peer'
# Lookup that DNS name
else:
identifier, _ = self.worker.resolveHost(hostname, identifier=identifier)
response = None
active_time = time.time()
self.resolving[(self.worker.w_id, identifier)] = client_id, hostname, hostname, command, decision
self.clients[client_id] = (self.worker.w_id, identifier, active_time, resolve_count)
self.active.append((active_time, client_id, self.worker.socket))
else:
identifier = None
response = None
return identifier, response
def beginResolvingTCP (self, client_id, command, decision, resolve_count):
if self.worker_count < self.max_workers:
identifier = self.newTCPResolver(client_id, command, decision, resolve_count)
self.worker_count += 1
else:
self.waiting.append((client_id, command, decision, resolve_count))
identifier = None
return identifier
def notifyClose (self):
paused = self.worker_count >= self.max_workers
self.worker_count -= 1
if paused and self.worker_count < self.max_workers:
for _ in range(self.worker_count, self.max_workers):
if self.waiting:
data, self.waiting = self.waiting[0], self.waiting[1:]
client_id, command, decision, resolve_count = data
identifier = self.newTCPResolver(client_id, command, decision, resolve_count)
self.worker_count += 1
def newTCPResolver (self, client_id, command, decision, resolve_count):
hostname = self.extractHostname(command, decision)
if hostname:
worker = self.resolver_factory.createTCPClient()
self.workers[worker.socket] = worker
identifier, all_sent = worker.resolveHost(hostname)
active_time = time.time()
self.resolving[(worker.w_id, identifier)] = client_id, hostname, hostname, command, decision
self.clients[client_id] = (worker.w_id, identifier, active_time, resolve_count)
self.active.append((active_time, client_id, self.worker.socket))
if all_sent:
self.poller.addReadSocket('read_resolver', worker.socket)
self.resolving[(worker.w_id, identifier)] = client_id, hostname, hostname, command, decision
else:
self.poller.addWriteSocket('write_resolver', worker.socket)
self.sending[worker.socket] = client_id, hostname, hostname, command, decision
else:
identifier = None
return identifier
def getResponse(self, sock):
worker = self.workers.get(sock)
if worker:
result = worker.getResponse(self.chained)
if result:
identifier, forhost, ip, completed, newidentifier, newhost, newcomplete = result
data = self.resolving.pop((worker.w_id, identifier), None)
chain_count = self.chained.pop(identifier, 0)
if newidentifier:
self.chained[newidentifier] = chain_count + 1
if not data:
self.log.info('ignoring response for %s (%s) with identifier %s' % (forhost, ip, identifier))
else:
# unable to parse response
self.log.error('unable to parse response')
data = None
if data:
client_id, original, hostname, command, decision = data
clidata = self.clients.pop(client_id, None)
if completed:
if clidata is not None:
key = clidata[2], client_id, worker.socket
if key in self.active:
self.active.remove(key)
# check to see if we received an incomplete response
if not completed:
newidentifier = self.beginResolvingTCP(client_id, command, decision, 1)
newhost = hostname
response = None
# check to see if the worker started a new request
if newidentifier:
if completed:
active_time = time.time()
self.resolving[(worker.w_id, newidentifier)] = client_id, original, newhost, command, decision
self.clients[client_id] = (worker.w_id, newidentifier, active_time, 1)
self.active.append((active_time, client_id, worker.socket))
response = None
if completed and newcomplete:
self.poller.addReadSocket('read_resolver', worker.socket)
elif completed and not newcomplete:
self.poller.addWriteSocket('write_resolver', worker.socket)
self.sending[worker.socket] = client_id, original, hostname, command, decision
# we just started a new (TCP) request and have not yet completely sent it
# make sure we still know who the request is for
elif not completed:
response = None
# maybe we read the wrong response?
elif forhost != hostname:
_, _, _, resolve_count = clidata
active_time = time.time()
self.resolving[(worker.w_id, identifier)] = client_id, original, hostname, command, decision
self.clients[client_id] = (worker.w_id, identifier, active_time, resolve_count)
self.active.append((active_time, client_id, worker.socket))
response = None
# success
elif ip is not None:
resolved = self.resolveDecision(command, decision, ip)
response = (client_id, command) + resolved
self.cacheDestination(original, ip)
# not found
else:
response = client_id, 'rewrite', '503', 'dns.html', 'http', '', '', hostname, 'peer'
#self.cacheDestination(original, ip)
else:
response = None
if response or result is None:
if worker is not self.worker:
self.poller.removeReadSocket('read_resolver', sock)
self.poller.removeWriteSocket('write_resolver', sock)
worker.close()
self.workers.pop(sock)
self.notifyClose()
else:
response = None
return response
def continueSending(self, sock):
"""Continue sending data over the connected TCP socket"""
data = self.sending.get(sock)
if data:
client_id, original, hostname, command, decision = data
else:
client_id, original, hostname, command, decision = None, None, None, None, None, None
worker = self.workers[sock]
res = worker.continueSending()
if res is False: # we've sent all we need to send
self.poller.removeWriteSocket('write_resolver', sock)
if client_id in self.clients:
w_id, identifier, active_time, resolve_count = self.clients[client_id]
tmp = self.sending.pop(sock)
self.resolving[(w_id, identifier)] = tmp
self.poller.addReadSocket('read_resolver', sock)
else:
self.log.error('could not find client for dns request for %s. request is being left to timeout.' % str(hostname))
try:
poll([f], [], [f], 0.1)
except socket.error:
print "CANNOT POLL (read): %s" % str(f)
log.error('can not poll (read) : %s' % str(f))
for f in write:
try:
poll([], [f], [f], 0.1)
except socket.error:
print "CANNOT POLL (write): %s" % str(f)
log.error('can not poll (write) : %s' % str(f))
raise e
except (ValueError, AttributeError, TypeError), e:
log.error("fatal error encountered during select - %s %s" %
(type(e), str(e)))
raise e
except select.error, e:
if e.args[0] in errno_block:
return [], [], []
log.error("fatal error encountered during select - %s %s" %
(type(e), str(e)))
raise e
except KeyboardInterrupt, e:
raise e
except Exception, e:
log.critical("fatal error encountered during select - %s %s" %
(type(e), str(e)))
raise e
return r, w, x
class Supervisor(object):
alarm_time = 0.1 # regular backend work
second_frequency = int(1 / alarm_time) # when we record history
minute_frequency = int(60 / alarm_time) # when we want to average history
increase_frequency = int(5 / alarm_time) # when we add workers
decrease_frequency = int(60 / alarm_time) # when we remove workers
saturation_frequency = int(
20 / alarm_time) # when we report connection saturation
interface_frequency = int(300 /
alarm_time) # when we check for new interfaces
# import os
# clear = [hex(ord(c)) for c in os.popen('clear').read()]
# clear = ''.join([chr(int(c,16)) for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a']])
def __init__(self, configuration):
self.configuration = configuration
# Only here so the introspection code can find them
self.log = Logger('supervisor', configuration.log.supervisor)
self.log.error('Starting exaproxy version %s' %
configuration.proxy.version)
self.signal_log = Logger('signal', configuration.log.signal)
self.log_writer = SysLogWriter('log',
configuration.log.destination,
configuration.log.enable,
level=configuration.log.level)
self.usage_writer = UsageWriter('usage',
configuration.usage.destination,
configuration.usage.enable)
sys.exitfunc = self.log_writer.writeMessages
self.log_writer.setIdentifier(configuration.daemon.identifier)
#self.usage_writer.setIdentifier(configuration.daemon.identifier)
if configuration.debug.log:
self.log_writer.toggleDebug()
self.usage_writer.toggleDebug()
self.log.error('python version %s' %
sys.version.replace(os.linesep, ' '))
self.log.debug('starting %s' % sys.argv[0])
self.pid = PID(self.configuration)
self.daemon = Daemon(self.configuration)
self.poller = Poller(self.configuration.daemon)
self.poller.setupRead('read_proxy') # Listening proxy sockets
self.poller.setupRead('read_web') # Listening webserver sockets
self.poller.setupRead('read_icap') # Listening icap sockets
self.poller.setupRead('read_tls') # Listening tls sockets
self.poller.setupRead('read_passthrough') # Listening raw data sockets
self.poller.setupRead(
'read_redirector'
) # Pipes carrying responses from the redirector process
self.poller.setupRead(
'read_resolver') # Sockets currently listening for DNS responses
self.poller.setupRead('read_client') # Active clients
self.poller.setupRead(
'opening_client') # Clients we have not yet read a request from
self.poller.setupWrite(
'write_client') # Active clients with buffered data to send
self.poller.setupWrite(
'write_resolver') # Active DNS requests with buffered data to send
self.poller.setupRead('read_download') # Established connections
self.poller.setupWrite(
'write_download'
) # Established connections we have buffered data to send to
self.poller.setupWrite('opening_download') # Opening connections
self.poller.setupRead('read_interrupt') # Scheduled events
self.poller.setupRead(
'read_control'
) # Responses from commands sent to the redirector process
self.monitor = Monitor(self)
self.page = Page(self)
self.content = ContentManager(self, configuration)
self.client = ClientManager(self.poller, configuration)
self.resolver = ResolverManager(self.poller, self.configuration,
configuration.dns.retries * 10)
self.proxy = Server('http proxy', self.poller, 'read_proxy',
configuration.http)
self.web = Server('web server', self.poller, 'read_web',
configuration.web)
self.icap = Server('icap server', self.poller, 'read_icap',
configuration.icap)
self.tls = Server('tls server', self.poller, 'read_tls',
configuration.tls)
self.passthrough = InterceptServer('passthrough server', self.poller,
'read_passthrough',
configuration.passthrough)
self._shutdown = True if self.daemon.filemax == 0 else False # stop the program
self._softstop = False # stop once all current connection have been dealt with
self._reload = False # unimplemented
self._toggle_debug = False # start logging a lot
self._decrease_spawn_limit = 0
self._increase_spawn_limit = 0
self._refork = False # unimplemented
self._pdb = False # turn on pdb debugging
self._listen = None # listening change ? None: no, True: listen, False: stop listeing
self.wait_time = 5.0 # how long do we wait at maximum once we have been soft-killed
self.local = set() # what addresses are on our local interfaces
if not self.initialise():
self._shutdown = True
elif self.daemon.drop_privileges():
self.log.critical(
'Could not drop privileges to \'%s\'. Refusing to run as root'
% self.daemon.user)
self.log.critical(
'Set the environment value USER to change the unprivileged user'
)
self._shutdown = True
# fork the redirector process before performing any further setup
redirector = fork_redirector(self.poller, self.configuration)
# use simple blocking IO for communication with the redirector process
self.redirector = redirector_message_thread(redirector)
# NOTE: create threads _after_ all forking is done
# regularly interrupt the reactor for maintenance
self.interrupt_scheduler = alarm_thread(self.poller, self.alarm_time)
self.reactor = Reactor(self.configuration, self.web, self.proxy,
self.passthrough, self.icap, self.tls,
self.redirector, self.content, self.client,
self.resolver, self.log_writer,
self.usage_writer, self.poller)
self.interfaces()
signal.signal(signal.SIGQUIT, self.sigquit)
signal.signal(signal.SIGINT, self.sigterm)
signal.signal(signal.SIGTERM, self.sigterm)
# signal.signal(signal.SIGABRT, self.sigabrt)
# signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGTRAP, self.sigtrap)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
signal.signal(signal.SIGTTOU, self.sigttou)
signal.signal(signal.SIGTTIN, self.sigttin)
# make sure we always have data in history
# (done in zero for dependencies reasons)
if self._shutdown is False:
self.redirector.requestStats()
command, control_data = self.redirector.readResponse()
stats_data = control_data if command == 'STATS' else None
stats = self.monitor.statistics(stats_data)
ok = self.monitor.zero(stats)
if ok:
self.redirector.requestStats()
else:
self._shutdown = True
def exit(self):
sys.exit()
def sigquit(self, signum, frame):
if self._softstop:
self.signal_log.critical('multiple SIG INT received, shutdown')
self._shutdown = True
else:
self.signal_log.critical('SIG INT received, soft-stop')
self._softstop = True
self._listen = False
def sigterm(self, signum, frame):
self.signal_log.critical('SIG TERM received, shutdown request')
if os.environ.get('PDB', False):
self._pdb = True
else:
self._shutdown = True
# def sigabrt (self,signum, frame):
# self.signal_log.info('SIG INFO received, refork request')
# self._refork = True
# def sighup (self,signum, frame):
# self.signal_log.info('SIG HUP received, reload request')
# self._reload = True
def sigtrap(self, signum, frame):
self.signal_log.critical('SIG TRAP received, toggle debug')
self._toggle_debug = True
def sigusr1(self, signum, frame):
self.signal_log.critical('SIG USR1 received, decrease worker number')
self._decrease_spawn_limit += 1
def sigusr2(self, signum, frame):
self.signal_log.critical('SIG USR2 received, increase worker number')
self._increase_spawn_limit += 1
def sigttou(self, signum, frame):
self.signal_log.critical('SIG TTOU received, stop listening')
self._listen = False
def sigttin(self, signum, frame):
self.signal_log.critical('SIG IN received, star listening')
self._listen = True
def interfaces(self):
local = {'127.0.0.1', '::1'}
for interface in getifaddrs():
if interface.family not in (AF_INET, AF_INET6):
continue
if interface.address not in self.local:
self.log.info('found new local ip %s (%s)' %
(interface.address, interface.name))
local.add(interface.address)
for ip in self.local:
if ip not in local:
self.log.info('removed local ip %s' % ip)
if local == self.local:
self.log.info('no ip change')
else:
self.local = local
def run(self):
count_second = 0
count_minute = 0
count_saturation = 0
count_interface = 0
events = {'read_interrupt'}
while True:
count_second = (count_second + 1) % self.second_frequency
count_minute = (count_minute + 1) % self.minute_frequency
count_saturation = (count_saturation +
1) % self.saturation_frequency
count_interface = (count_interface + 1) % self.interface_frequency
try:
if self._pdb:
self._pdb = False
import pdb
pdb.set_trace()
# prime the alarm
if 'read_interrupt' in events:
self.interrupt_scheduler.setAlarm()
# check for IO change with select
status, events = self.reactor.run()
# shut down the server if a child process disappears
if status is False:
self._shutdown = True
# respond to control responses immediately
if 'read_control' in events:
command, control_data = self.redirector.readResponse()
if command == 'STATS':
ok = self.doStats(count_second, count_minute,
control_data)
if ok is False:
self._shutdown = True
# jump straight back into the reactor if we haven't yet received an
# interrupt event
if 'read_interrupt' not in events:
continue
# clear the alarm condition
self.interrupt_scheduler.acknowledgeAlarm()
# must follow the reactor so we are sure to go through the reactor at least once
# and flush any logs
if self._shutdown:
self._shutdown = False
self.shutdown()
break
elif self._reload:
self._reload = False
self.reload()
elif self._refork:
self._refork = False
self.signal_log.warning('refork not implemented')
# stop listening to new connections
# refork the program (as we have been updated)
# just handle current open connection
# ask the redirector process for stats
self.redirector.requestStats()
if self._softstop:
if self._listen == False:
self.proxy.rejecting()
self._listen = None
if self.client.softstop():
self._shutdown = True
# only change listening if we are not shutting down
elif self._listen is not None:
if self._listen:
self._shutdown = not self.proxy.accepting()
self._listen = None
else:
self.proxy.rejecting()
self._listen = None
if self._toggle_debug:
self._toggle_debug = False
self.log_writer.toggleDebug()
if self._decrease_spawn_limit:
count = self._decrease_spawn_limit
self.redirector.decreaseSpawnLimit(count)
self._decrease_spawn_limit = 0
if self._increase_spawn_limit:
count = self._increase_spawn_limit
self.redirector.increaseSpawnLimit(count)
self._increase_spawn_limit = 0
# cleanup idle connections
# TODO: track all idle connections, not just the ones that have never sent data
expired = self.reactor.client.expire()
for expire_source, expire_count in expired.items():
if expire_source == 'proxy':
self.proxy.notifyClose(None, count=expire_count)
elif expire_source == 'icap':
self.icap.notifyClose(None, count=expire_count)
elif expire_source == 'passthrough':
self.passthrough.notifyClose(None, count=expire_count)
elif expire_source == 'tls':
self.tls.notifyClose(None, count=expire_count)
elif expire_source == 'web':
self.web.notifyClose(None, count=expire_count)
# report if we saw too many connections
if count_saturation == 0:
self.proxy.saturation()
self.web.saturation()
if self.configuration.daemon.poll_interfaces and count_interface == 0:
self.interfaces()
except KeyboardInterrupt:
self.log.critical('^C received')
self._shutdown = True
except OSError, e:
# This shoould never happen as we are limiting how many connections we accept
if e.errno == 24: # Too many open files
self.log.critical('Too many opened files, shutting down')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
else:
self.log.critical('unrecoverable io error')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
finally:
class ClientManager(object):
unproxy = ProxyProtocol().parseRequest
def __init__(self, poller, configuration):
self.total_sent4 = 0L
self.total_sent6 = 0L
self.total_requested = 0L
self.norequest = TimeCache(configuration.http.idle_connect)
self.bysock = {}
self.byname = {}
self.buffered = []
self._nextid = 0
self.poller = poller
self.log = Logger('client', configuration.log.client)
self.proxied = configuration.http.proxied
self.max_buffer = configuration.http.header_size
def __contains__(self, item):
return item in self.byname
def getnextid(self):
self._nextid += 1
return str(self._nextid)
def expire(self, number=100):
count = 0
for sock in self.norequest.expired(number):
client = self.norequest.get(sock, [
None,
])[0]
if client:
self.cleanup(sock, client.name)
count += 1
return count
def newConnection(self, sock, peer, source):
name = self.getnextid()
client = Client(name, sock, peer, self.log, self.max_buffer)
self.norequest[sock] = client, source
self.byname[name] = client, source
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def readRequest(self, sock):
"""Read only the initial HTTP headers sent by the client"""
client, source = self.norequest.get(sock, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# headers can be read only once
self.norequest.pop(sock, (None, None))
# we have now read the client's opening request
self.poller.removeReadSocket('opening_client', client.sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error(
'trying to read headers from a client that does not exist %s' %
sock)
name, peer, request, content, source = None, None, None, None, None
if request and self.proxied is True and source == 'proxy':
client_ip, client_request = self.unproxy(request)
if client_ip and client_request:
peer = client_ip
request = client_request
client.setPeer(client_ip)
return name, peer, request, content, source
def readDataBySocket(self, sock):
client, source = self.bysock.get(sock, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error(
'trying to read from a client that does not exist %s' % sock)
name, peer, request, content = None, None, None, None
return name, peer, request, content, source
def readDataByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', client.sock)
elif request is None:
self.cleanup(client.sock, name)
else:
self.log.error(
'trying to read from a client that does not exist %s' % name)
name, peer, request, content = None, None, None, None
return name, peer, request, content
def sendDataBySocket(self, sock, data):
client, source = self.bysock.get(sock, (None, None))
if client:
name = client.name
res = client.writeData(data)
if res is None:
# close the client connection
self.cleanup(sock, client.name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if sock not in self.buffered:
self.buffered.append(sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and sock in self.buffered:
self.buffered.remove(sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
name = None
return result, buffer_change, name, source
def sendDataByName(self, name, data):
client, source = self.byname.get(name, (None, None))
if client:
res = client.writeData(data)
if res is None:
# we cannot write to the client so clean it up
self.cleanup(client.sock, name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if client.sock not in self.buffered:
self.buffered.append(client.sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and client.sock in self.buffered:
self.buffered.remove(client.sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
return result, buffer_change, client
def startData(self, name, data, remaining):
# NOTE: soo ugly but fast to code
nb_to_read = 0
if type(remaining) == type(''):
if 'chunked' in remaining:
mode = 'chunked'
else:
mode = 'passthrough'
elif remaining > 0:
mode = 'transfer'
nb_to_read = remaining
elif remaining == 0:
mode = 'request'
else:
mode = 'passthrough'
client, source = self.byname.get(name, (None, None))
if client:
try:
command, d = data
except (ValueError, TypeError):
self.log.error('invalid command sent to client %s' % name)
self.cleanup(client.sock, name)
res = None
else:
if client.sock not in self.bysock:
# Start checking for content sent by the client
self.bysock[client.sock] = client, source
# watch for the client sending new data
self.poller.addReadSocket('read_client', client.sock)
# make sure we don't somehow end up with this still here
self.norequest.pop(client.sock, (None, None))
# NOTE: always done already in readRequest
self.poller.removeReadSocket('opening_client', client.sock)
res = client.startData(command, d)
else:
res = client.restartData(command, d)
# If we are here then we must have prohibited reading from the client
# and it must otherwise have been in a readable state
self.poller.uncorkReadSocket('read_client', client.sock)
if res is not None:
buffered, had_buffer, sent4, sent6 = res
# buffered data we read with the HTTP headers
name, peer, request, content = client.readRelated(
mode, nb_to_read)
if request:
self.total_requested += 1
self.log.info('reading multiple requests')
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
elif request is None:
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
else:
# we cannot write to the client so clean it up
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
if buffered:
if client.sock not in self.buffered:
self.buffered.append(client.sock)
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
elif had_buffer and client.sock in self.buffered:
self.buffered.remove(client.sock)
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
content = None
return client, content, source
def corkUploadByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
self.poller.corkReadSocket('read_client', client.sock)
def uncorkUploadByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
if client.sock in self.bysock:
self.poller.uncorkReadSocket('read_client', client.sock)
def cleanup(self, sock, name):
self.log.debug('cleanup for socket %s' % sock)
client, source = self.bysock.get(sock, (None, None))
client, source = (client, None) if client else self.norequest.get(
sock, (None, None))
client, source = (client, None) or self.byname.get(name, (None, None))
self.bysock.pop(sock, None)
self.norequest.pop(sock, (None, None))
self.byname.pop(name, None)
if client:
self.poller.removeWriteSocket('write_client', client.sock)
self.poller.removeReadSocket('read_client', client.sock)
self.poller.removeReadSocket('opening_client', client.sock)
client.shutdown()
else:
self.log.error('COULD NOT CLEAN UP SOCKET %s' % sock)
if sock in self.buffered:
self.buffered.remove(sock)
def softstop(self):
if len(self.byname) > 0 or len(self.norequest) > 0:
return False
self.log.critical('no more client connection, exiting.')
return True
def stop(self):
for client, source in self.bysock.itervalues():
client.shutdown()
for client, source in self.norequest.itervalues():
client.shutdown()
self.poller.clearRead('read_client')
self.poller.clearRead('opening_client')
self.poller.clearWrite('write_client')
self.bysock = {}
self.norequest = {}
self.byname = {}
self.buffered = []
class Redirector(Thread):
# TODO : if the program is a function, fork and run :)
ICAPParser = ICAPParser
def __init__(self, configuration, name, request_box, program):
self.configuration = configuration
self.icap_parser = self.ICAPParser(configuration)
self.enabled = configuration.redirector.enable
self.protocol = configuration.redirector.protocol
self._transparent = configuration.http.transparent
self.log = Logger('worker ' + str(name), configuration.log.worker)
self.usage = UsageLogger('usage', configuration.log.worker)
self.universal = True if self.protocol == 'url' else False
self.icap = self.protocol[len('icap://'):].split(
'/')[0] if self.protocol.startswith('icap://') else ''
r, w = os.pipe() # pipe for communication with the main thread
self.response_box_write = os.fdopen(w, 'w',
0) # results are written here
self.response_box_read = os.fdopen(r, 'r',
0) # read from the main thread
self.wid = name # a unique name
self.creation = time.time() # when the thread was created
# self.last_worked = self.creation # when the thread last picked a task
self.request_box = request_box # queue with HTTP headers to process
self.program = program # the squid redirector program to fork
self.running = True # the thread is active
self.stats_timestamp = None # time of the most recent outstanding request to generate stats
self._proxy = 'ExaProxy-%s-id-%d' % (configuration.proxy.version,
os.getpid())
if self.protocol == 'url':
self.classify = self._classify_url
if self.protocol.startswith('icap://'):
self.classify = self._classify_icap
# Do not move, we need the forking AFTER the setup
self.process = self._createProcess(
) # the forked program to handle classification
Thread.__init__(self)
def _createProcess(self):
if not self.enabled:
return
def preexec(): # Don't forward signals.
os.setpgrp()
try:
process = subprocess.Popen(
[
self.program,
],
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=self.universal,
preexec_fn=preexec,
)
self.log.debug('spawn process %s' % self.program)
except KeyboardInterrupt:
process = None
except (subprocess.CalledProcessError, OSError, ValueError):
self.log.error('could not spawn process %s' % self.program)
process = None
if process:
try:
fcntl.fcntl(process.stderr, fcntl.F_SETFL, os.O_NONBLOCK)
except IOError:
self.destroyProcess()
process = None
return process
def destroyProcess(self):
if not self.enabled:
return
self.log.debug('destroying process %s' % self.program)
if not self.process:
return
try:
if self.process:
self.process.terminate()
self.process.wait()
self.log.info('terminated process PID %s' % self.process.pid)
except OSError, e:
# No such processs
if e[0] != errno.ESRCH:
self.log.error('PID %s died' % self.process.pid)
class ClientManager (object):
def __init__(self, poller, configuration):
self.total_sent4 = 0L
self.total_sent6 = 0L
self.total_requested = 0L
self.norequest = TimeCache(configuration.http.idle_connect)
self.bysock = {}
self.byname = {}
self.buffered = []
self._nextid = 0
self.poller = poller
self.log = Logger('client', configuration.log.client)
self.http_max_buffer = configuration.http.header_size
self.icap_max_buffer = configuration.icap.header_size
self.proxied = {
'proxy' : configuration.http.proxied,
'icap' : configuration.icap.proxied,
}
def __contains__(self, item):
return item in self.bysock
def lookupSocket (self, item):
return self.byname.get(item, None)
def getnextid(self):
self._nextid += 1
return str(self._nextid)
def expire (self,number=100):
count = 0
for sock in self.norequest.expired(number):
client = self.norequest.get(sock,[None,])[0]
if client:
self.cleanup(sock,client.name)
count += 1
return count
def httpConnection (self, sock, peer, source):
name = self.getnextid()
client = HTTPClient(name, sock, peer, self.log, self.http_max_buffer, self.proxied.get(source))
self.norequest[sock] = client, source
self.byname[name] = sock
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def icapConnection (self, sock, peer, source):
name = self.getnextid()
client = ICAPClient(name, sock, peer, self.log, self.icap_max_buffer, self.proxied.get(source))
self.norequest[sock] = client, source
self.byname[name] = sock
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def readRequest (self, sock):
"""Read only the initial HTTP headers sent by the client"""
client, source = self.norequest.get(sock, (None, None))
if client:
name, peer, request, subrequest, content = client.readData()
if request:
self.total_requested += 1
# headers can be read only once
self.norequest.pop(sock, (None, None))
self.bysock[sock] = client, source
# watch for the client sending new data
self.poller.addReadSocket('read_client', client.sock)
# we have now read the client's opening request
self.poller.removeReadSocket('opening_client', client.sock)
# do not read more data until we have properly handled the request
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read headers from a client that does not exist %s' % sock)
name, peer, request, subrequest, content, source = None, None, None, None, None, None
return name, peer, request, subrequest, content, source
def readData (self, sock):
client, source = self.bysock.get(sock, (None, None))
if client:
name, peer, request, subrequest, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read from a client that does not exist %s' % sock)
name, peer, request, subrequest, content = None, None, None, None, None
return name, peer, request, subrequest, content, source
def sendData (self, sock, data):
client, source = self.bysock.get(sock, (None, None))
if client:
name = client.name
res = client.writeData(data)
if res is None:
# close the client connection
self.cleanup(sock, client.name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if sock not in self.buffered:
self.buffered.append(sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and sock in self.buffered:
self.buffered.remove(sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
name = None
return result, buffer_change, name, source
def parseRemaining (self, remaining):
nb_to_read = 0
if isinstance(remaining, basestring):
mode = 'chunked' if remaining == 'chunked' else 'passthrough'
elif remaining > 0:
mode = 'transfer'
nb_to_read = remaining
elif remaining == 0:
mode = ''
else:
mode = 'passthrough'
return mode, nb_to_read
def startData(self, sock, data, remaining):
client, source = self.bysock.get(sock, (None, None))
try:
mode, nb_to_read = self.parseRemaining(remaining)
command, d = data if client is not None else (None, None)
except (ValueError, TypeError), e:
self.log.error('invalid command sent to client %s' % client.name)
command, d = None, None
if not client or command is None:
return None, source
name, peer, res = client.startData(command, d)
if res is not None:
name, peer, request, subrequest, content = client.readRelated(mode, nb_to_read)
buffered, had_buffer, sent4, sent6 = res
self.poller.uncorkReadSocket('read_client', client.sock)
self.total_sent4 += sent4
self.total_sent6 += sent6
else:
self.cleanup(client.sock, name)
return None, source
if request:
self.total_requested += 1
self.log.info('reading multiple requests')
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
elif request is None:
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
if buffered is True and had_buffer is False:
self.buffered.append(client.sock)
self.poller.addWriteSocket('write_client', client.sock)
elif buffered is False and had_buffer is True:
self.buffered.remove(client.sock)
self.poller.removeWriteSocket('write_client', client.sock)
return content, source
for events in res:
fd = events.ident
name, poller, sockets, fdtosock = self.pollers[fd]
events = poller.control(None, self.max_events, 0)
if (len(events) == self.max_events):
log.warning("polled max_events from queue %s" % (name))
for sock_events in events:
sock_fd = sock_events.ident
try:
response[name].append(fdtosock[sock_fd])
except KeyError, e:
log.error("KQueue register called before fdtosock registered! Skipping event")
continue
if sock_events.flags & select.KQ_EV_ERROR:
log.warning("%s KQ_EV_ERROR: fd=%d filter=%d fflags=%d flags=%d data=%d udata=%d" % (
str(datetime.datetime.now()),
sock_events.ident, sock_events.filter, sock_events.flags, sock_events.fflags,
sock_events.data, sock_events.udata))
sock = fdtosock.pop(sock_fd, None)
poller.control([kevent(sock, sock_events.filter, KQ_EV_DELETE)], 0)
sockets.pop(sock)
if sock not in self.errors:
self.errors[sock] = name
for events in res:
fd = events.ident
name, poller, sockets, fdtosock = self.pollers[fd]
events = poller.control(None, self.max_events, 0)
if len(events) == self.max_events:
log.warning("polled max_events from queue %s" % name)
for sock_events in events:
sock_fd = sock_events.ident
try:
response[name].append(fdtosock[sock_fd])
except KeyError:
log.error("KQueue register called before fdtosock registered! Skipping event")
continue
if sock_events.flags & select.KQ_EV_ERROR:
log.warning("%s KQ_EV_ERROR: fd=%d filter=%d fflags=%d flags=%d data=%d udata=%d" % (
str(datetime.datetime.now()),
sock_events.ident, sock_events.filter, sock_events.flags, sock_events.fflags,
sock_events.data, sock_events.udata))
sock = fdtosock.pop(sock_fd, None)
poller.control([kevent(sock, sock_events.filter, KQ_EV_DELETE)], 0)
sockets.pop(sock)
if sock not in self.errors:
self.errors[sock] = name
class Redirector (Thread):
# TODO : if the program is a function, fork and run :)
def __init__ (self, configuration, name, request_box, program):
self.configuration = configuration
self.enabled = configuration.redirector.enable
self.protocol = configuration.redirector.protocol
self._transparent = configuration.http.transparent
self.log = Logger('worker ' + str(name), configuration.log.worker)
self.usage = UsageLogger('usage', configuration.log.worker)
self.universal = True if self.protocol == 'url' else False
self.icap = self.protocol[len('icap://'):].split('/')[0] if self.protocol.startswith('icap://') else ''
r, w = os.pipe() # pipe for communication with the main thread
self.response_box_write = os.fdopen(w,'w',0) # results are written here
self.response_box_read = os.fdopen(r,'r',0) # read from the main thread
self.wid = name # a unique name
self.creation = time.time() # when the thread was created
# self.last_worked = self.creation # when the thread last picked a task
self.request_box = request_box # queue with HTTP headers to process
self.program = program # the squid redirector program to fork
self.running = True # the thread is active
self.stats_timestamp = None # time of the most recent outstanding request to generate stats
self._proxy = 'ExaProxy-%s-id-%d' % (configuration.proxy.version,os.getpid())
if self.protocol == 'url':
self.classify = self._classify_url
if self.protocol.startswith('icap://'):
self.classify = self._classify_icap
# Do not move, we need the forking AFTER the setup
self.process = self._createProcess() # the forked program to handle classification
Thread.__init__(self)
def _createProcess (self):
if not self.enabled:
return
def preexec(): # Don't forward signals.
os.setpgrp()
try:
process = subprocess.Popen([self.program,],
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
universal_newlines=self.universal,
preexec_fn=preexec,
)
self.log.debug('spawn process %s' % self.program)
except KeyboardInterrupt:
process = None
except (subprocess.CalledProcessError,OSError,ValueError):
self.log.error('could not spawn process %s' % self.program)
process = None
if process:
try:
fcntl.fcntl(process.stderr, fcntl.F_SETFL, os.O_NONBLOCK)
except IOError:
self.destroyProcess()
process = None
return process
def destroyProcess (self):
if not self.enabled:
return
self.log.debug('destroying process %s' % self.program)
if not self.process:
return
try:
if self.process:
self.process.terminate()
self.process.wait()
self.log.info('terminated process PID %s' % self.process.pid)
except OSError, e:
# No such processs
if e[0] != errno.ESRCH:
self.log.error('PID %s died' % self.process.pid)
class ClientManager (object):
unproxy = ProxyProtocol().parseRequest
def __init__(self, poller, configuration):
self.total_sent4 = 0L
self.total_sent6 = 0L
self.total_requested = 0L
self.norequest = TimeCache(configuration.http.idle_connect)
self.bysock = {}
self.byname = {}
self.buffered = []
self._nextid = 0
self.poller = poller
self.log = Logger('client', configuration.log.client)
self.proxied = configuration.http.proxied
self.max_buffer = configuration.http.header_size
def __contains__(self, item):
return item in self.byname
def getnextid(self):
self._nextid += 1
return str(self._nextid)
def expire (self,number=100):
count = 0
for sock in self.norequest.expired(number):
client = self.norequest.get(sock,[None,])[0]
if client:
self.cleanup(sock,client.name)
count += 1
return count
def newConnection(self, sock, peer, source):
name = self.getnextid()
client = Client(name, sock, peer, self.log, self.max_buffer)
self.norequest[sock] = client, source
self.byname[name] = client, source
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def readRequest(self, sock):
"""Read only the initial HTTP headers sent by the client"""
client, source = self.norequest.get(sock, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# headers can be read only once
self.norequest.pop(sock, (None, None))
# we have now read the client's opening request
self.poller.removeReadSocket('opening_client', client.sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read headers from a client that does not exist %s' % sock)
name, peer, request, content, source = None, None, None, None, None
if request and self.proxied is True and source == 'proxy':
client_ip, client_request = self.unproxy(request)
if client_ip and client_request:
peer = client_ip
request = client_request
client.setPeer(client_ip)
return name, peer, request, content, source
def readDataBySocket(self, sock):
client, source = self.bysock.get(sock, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read from a client that does not exist %s' % sock)
name, peer, request, content = None, None, None, None
return name, peer, request, content, source
def readDataByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
name, peer, request, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', client.sock)
elif request is None:
self.cleanup(client.sock, name)
else:
self.log.error('trying to read from a client that does not exist %s' % name)
name, peer, request, content = None, None, None, None
return name, peer, request, content
def sendDataBySocket(self, sock, data):
client, source = self.bysock.get(sock, (None, None))
if client:
name = client.name
res = client.writeData(data)
if res is None:
# close the client connection
self.cleanup(sock, client.name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if sock not in self.buffered:
self.buffered.append(sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and sock in self.buffered:
self.buffered.remove(sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
name = None
return result, buffer_change, name, source
def sendDataByName(self, name, data):
client, source = self.byname.get(name, (None, None))
if client:
res = client.writeData(data)
if res is None:
# we cannot write to the client so clean it up
self.cleanup(client.sock, name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if client.sock not in self.buffered:
self.buffered.append(client.sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and client.sock in self.buffered:
self.buffered.remove(client.sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
return result, buffer_change, client
def startData(self, name, data, remaining):
# NOTE: soo ugly but fast to code
nb_to_read = 0
if type(remaining) == type(''):
if 'chunked' in remaining:
mode = 'chunked'
else:
mode = 'passthrough'
elif remaining > 0:
mode = 'transfer'
nb_to_read = remaining
elif remaining == 0:
mode = 'request'
else:
mode = 'passthrough'
client, source = self.byname.get(name, (None, None))
if client:
try:
command, d = data
except (ValueError, TypeError):
self.log.error('invalid command sent to client %s' % name)
self.cleanup(client.sock, name)
res = None
else:
if client.sock not in self.bysock:
# Start checking for content sent by the client
self.bysock[client.sock] = client, source
# watch for the client sending new data
self.poller.addReadSocket('read_client', client.sock)
# make sure we don't somehow end up with this still here
self.norequest.pop(client.sock, (None,None))
# NOTE: always done already in readRequest
self.poller.removeReadSocket('opening_client', client.sock)
res = client.startData(command, d)
else:
res = client.restartData(command, d)
# If we are here then we must have prohibited reading from the client
# and it must otherwise have been in a readable state
self.poller.uncorkReadSocket('read_client', client.sock)
if res is not None:
buffered, had_buffer, sent4, sent6 = res
# buffered data we read with the HTTP headers
name, peer, request, content = client.readRelated(mode,nb_to_read)
if request:
self.total_requested += 1
self.log.info('reading multiple requests')
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
elif request is None:
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
else:
# we cannot write to the client so clean it up
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
if buffered:
if client.sock not in self.buffered:
self.buffered.append(client.sock)
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
elif had_buffer and client.sock in self.buffered:
self.buffered.remove(client.sock)
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
content = None
return client, content, source
def corkUploadByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
self.poller.corkReadSocket('read_client', client.sock)
def uncorkUploadByName(self, name):
client, source = self.byname.get(name, (None, None))
if client:
if client.sock in self.bysock:
self.poller.uncorkReadSocket('read_client', client.sock)
def cleanup(self, sock, name):
self.log.debug('cleanup for socket %s' % sock)
client, source = self.bysock.get(sock, (None,None))
client, source = (client,None) if client else self.norequest.get(sock, (None,None))
client, source = (client,None) or self.byname.get(name, (None,None))
self.bysock.pop(sock, None)
self.norequest.pop(sock, (None,None))
self.byname.pop(name, None)
if client:
self.poller.removeWriteSocket('write_client', client.sock)
self.poller.removeReadSocket('read_client', client.sock)
self.poller.removeReadSocket('opening_client', client.sock)
client.shutdown()
else:
self.log.error('COULD NOT CLEAN UP SOCKET %s' % sock)
if sock in self.buffered:
self.buffered.remove(sock)
def softstop (self):
if len(self.byname) > 0 or len(self.norequest) > 0:
return False
self.log.critical('no more client connection, exiting.')
return True
def stop(self):
for client, source in self.bysock.itervalues():
client.shutdown()
for client, source in self.norequest.itervalues():
client.shutdown()
self.poller.clearRead('read_client')
self.poller.clearRead('opening_client')
self.poller.clearWrite('write_client')
self.bysock = {}
self.norequest = {}
self.byname = {}
self.buffered = []
class RedirectorManager (object):
def __init__ (self, configuration, poller):
self.low = configuration.redirector.minimum # minimum concurrent redirector workers
self.high = configuration.redirector.maximum # maximum concurrent redirector workers
self.poller = poller
self.configuration = configuration
self.queue = Queue() # store requests we do not immediately have the resources to process
self.nextid = 1 # unique id to give to the next spawned worker
self.worker = {} # worker tasks for each spawned child
self.processes = {} # worker tasks indexed by file descriptors we can poll
self.available = set() # workers that are currently available to handle new requests
self.active = {} # workers that are currently busy waiting for a response from the spawned process
self.stopping = set() # workers we want to stop as soon as they stop being active
program = configuration.redirector.program
protocol = configuration.redirector.protocol
self.redirector_factory = RedirectorFactory(configuration, program, protocol)
self.log = Logger('manager', configuration.log.manager)
def _getid(self):
wid = str(self.nextid)
self.nextid += 1
return wid
def _spawn (self):
"""add one worker to the pool"""
wid = self._getid()
worker = self.redirector_factory.create(wid)
self.worker[wid] = worker
self.available.add(wid)
if worker.process is not None:
identifier = worker.process.stdout
self.processes[identifier] = worker
self.poller.addReadSocket('read_workers', identifier)
self.log.info("added a worker")
self.log.info("we have %d workers. defined range is ( %d / %d )" % (len(self.worker), self.low, self.high))
def spawn (self, number=1):
"""create the request number of worker processes"""
self.log.info("spawning %d more workers" % number)
for _ in range(number):
self._spawn()
def respawn (self):
"""make sure we reach the minimum number of workers"""
number = max(min(len(self.worker), self.high), self.low)
for wid in set(self.worker):
self.stopWorker(wid)
self.spawn(number)
def stopWorker (self, wid):
self.log.info('want worker %s to go away' % wid)
if wid not in self.active:
self.reap(wid)
else:
self.stopping.add(wid)
def reap (self, wid):
self.log.info('we are killing worker %s' % wid)
worker = self.worker[wid]
if wid in self.active:
self.log.error('reaping worker %s even though it is still active' % wid)
self.active.pop(wid)
if wid in self.stopping:
self.stopping.remove(wid)
if wid in self.available:
self.available.remove(wid)
if worker.process is not None:
self.poller.removeReadSocket('read_workers', worker.process.stdout)
self.processes.pop(worker.process.stdout)
worker.shutdown()
self.worker.pop(wid)
def _decrease (self):
if self.low < len(self.worker):
wid = self._oldest()
if wid:
self.stopWorker(wid)
def _increase (self):
if len(self.worker) < self.high:
self.spawn()
def decrease (self, count=1):
for _ in xrange(count):
self._decrease()
def increase (self, count=1):
for _ in xrange(count):
self._increase()
def start (self):
"""spawn our minimum number of workers"""
self.log.info("starting workers.")
self.spawn(max(0,self.low-len(self.worker)))
def stop (self):
"""tell all our worker to stop reading the queue and stop"""
for wid in self.worker:
self.reap(wid)
self.worker = {}
def _oldest (self):
"""find the oldest worker"""
oldest = None
past = time.time()
for wid in set(self.worker):
creation = self.worker[wid].creation
if creation < past and wid not in self.stopping:
past = creation
oldest = wid
return oldest
def provision (self):
"""manage our workers to make sure we have enough to consume the queue"""
size = self.queue.qsize()
num_workers = len(self.worker)
# bad we are bleeding workers !
if num_workers < self.low:
self.log.info("we lost some workers, respawing %d new workers" % (self.low - num_workers))
self.spawn(self.low - num_workers)
# we need more workers
if size >= num_workers:
# nothing we can do we have reach our limit
if num_workers >= self.high:
self.log.warning("help ! we need more workers but we reached our ceiling ! %d request are queued for %d processes" % (size,num_workers))
return
# try to figure a good number to add ..
# no less than one, no more than to reach self.high, lower between self.low and a quarter of the allowed growth
nb_to_add = int(min(max(1,min(self.low,(self.high-self.low)/4)),self.high-num_workers))
self.log.warning("we are low on workers adding a few (%d), the queue has %d unhandled url" % (nb_to_add,size))
self.spawn(nb_to_add)
def deprovision (self):
"""manage our workers to make sure we have enough to consume the queue"""
size = self.queue.qsize()
num_workers = len(self.worker)
# we are now overprovisioned
if size < 2 and num_workers > self.low:
self.log.info("we have too many workers (%d), stopping the oldest" % num_workers)
# if we have to kill one, at least stop the one who had the most chance to memory leak :)
wid = self._oldest()
if wid:
self.stopWorker(wid)
def acquire (self):
if self.available:
identifier = self.available.pop()
worker = self.worker[identifier]
else:
worker = None
return worker
def release (self, wid):
if wid not in self.stopping:
self.available.add(wid)
else:
self.reap(wid)
def persist (self, wid, client_id, peer, data, header, subheader, source, tainted):
self.active[wid] = client_id, peer, data, header, subheader, source, tainted
def progress (self, wid):
return self.active.pop(wid)
def doqueue (self):
if self.available and not self.queue.isempty():
client_id, peer, header, subheader, source, tainted = self.queue.get()
_, command, decision = self.request(client_id, peer, header, subheader, source, tainted=tainted)
else:
client_id, command, decision = None, None, None
return client_id, command, decision
def request (self, client_id, peer, header, subheader, source, tainted=False):
worker = self.acquire()
if worker is not None:
try:
_, command, decision = worker.decide(client_id, peer, header, subheader, source)
except:
command, decision = None, None
if command is None:
self.reap(worker.wid)
if tainted is False:
_, command, decision = self.request(client_id, peer, header, subheader, source, tainted=True)
else:
_, command, decision = Respond.close(client_id)
else:
command, decision = None, None
self.queue.put((client_id, peer, header, subheader, source, tainted))
if command == 'defer':
self.persist(worker.wid, client_id, peer, decision, header, subheader, source, tainted)
command, decision = None, None
elif worker is not None:
self.release(worker.wid)
return client_id, command, decision
def getDecision (self, pipe_in):
worker = self.processes.get(pipe_in, None)
if worker is not None and worker.wid in self.active:
client_id, peer, request, header, subheader, source, tainted = self.progress(worker.wid)
try:
_, command, decision = worker.progress(client_id, peer, request, header, subheader, source)
except Exception, e:
command, decision = None, None
self.release(worker.wid)
if command is None:
self.reap(worker.wid)
if tainted is False:
_, command, decision = self.request(client_id, peer, header, subheader, source, tainted=True)
else:
_, command, decision = Respond.close(client_id)
else:
class Supervisor (object):
alarm_time = 0.1 # regular backend work
second_frequency = int(1/alarm_time) # when we record history
minute_frequency = int(60/alarm_time) # when we want to average history
increase_frequency = int(5/alarm_time) # when we add workers
decrease_frequency = int(60/alarm_time) # when we remove workers
saturation_frequency = int(20/alarm_time) # when we report connection saturation
interface_frequency = int(300/alarm_time) # when we check for new interfaces
# import os
# clear = [hex(ord(c)) for c in os.popen('clear').read()]
# clear = ''.join([chr(int(c,16)) for c in ['0x1b', '0x5b', '0x48', '0x1b', '0x5b', '0x32', '0x4a']])
def __init__ (self,configuration):
configuration = load()
self.configuration = configuration
# Only here so the introspection code can find them
self.log = Logger('supervisor', configuration.log.supervisor)
self.log.error('Starting exaproxy version %s' % configuration.proxy.version)
self.signal_log = Logger('signal', configuration.log.signal)
self.log_writer = SysLogWriter('log', configuration.log.destination, configuration.log.enable, level=configuration.log.level)
self.usage_writer = UsageWriter('usage', configuration.usage.destination, configuration.usage.enable)
self.log_writer.setIdentifier(configuration.daemon.identifier)
#self.usage_writer.setIdentifier(configuration.daemon.identifier)
if configuration.debug.log:
self.log_writer.toggleDebug()
self.usage_writer.toggleDebug()
self.log.error('python version %s' % sys.version.replace(os.linesep,' '))
self.log.debug('starting %s' % sys.argv[0])
self.pid = PID(self.configuration)
self.daemon = Daemon(self.configuration)
self.poller = Poller(self.configuration.daemon)
self.poller.setupRead('read_proxy') # Listening proxy sockets
self.poller.setupRead('read_web') # Listening webserver sockets
self.poller.setupRead('read_icap') # Listening icap sockets
self.poller.setupRead('read_workers') # Pipes carrying responses from the child processes
self.poller.setupRead('read_resolver') # Sockets currently listening for DNS responses
self.poller.setupRead('read_client') # Active clients
self.poller.setupRead('opening_client') # Clients we have not yet read a request from
self.poller.setupWrite('write_client') # Active clients with buffered data to send
self.poller.setupWrite('write_resolver') # Active DNS requests with buffered data to send
self.poller.setupRead('read_download') # Established connections
self.poller.setupWrite('write_download') # Established connections we have buffered data to send to
self.poller.setupWrite('opening_download') # Opening connections
self.monitor = Monitor(self)
self.page = Page(self)
self.manager = RedirectorManager(
self.configuration,
self.poller,
)
self.content = ContentManager(self,configuration)
self.client = ClientManager(self.poller, configuration)
self.resolver = ResolverManager(self.poller, self.configuration, configuration.dns.retries*10)
self.proxy = Server('http proxy',self.poller,'read_proxy', configuration.http.connections)
self.web = Server('web server',self.poller,'read_web', configuration.web.connections)
self.icap = Server('icap server',self.poller,'read_icap', configuration.icap.connections)
self.reactor = Reactor(self.configuration, self.web, self.proxy, self.icap, self.manager, self.content, self.client, self.resolver, self.log_writer, self.usage_writer, self.poller)
self._shutdown = True if self.daemon.filemax == 0 else False # stop the program
self._softstop = False # stop once all current connection have been dealt with
self._reload = False # unimplemented
self._toggle_debug = False # start logging a lot
self._decrease_spawn_limit = 0
self._increase_spawn_limit = 0
self._refork = False # unimplemented
self._pdb = False # turn on pdb debugging
self._listen = None # listening change ? None: no, True: listen, False: stop listeing
self.wait_time = 5.0 # how long do we wait at maximum once we have been soft-killed
self.local = set() # what addresses are on our local interfaces
self.interfaces()
signal.signal(signal.SIGQUIT, self.sigquit)
signal.signal(signal.SIGINT, self.sigterm)
signal.signal(signal.SIGTERM, self.sigterm)
# signal.signal(signal.SIGABRT, self.sigabrt)
# signal.signal(signal.SIGHUP, self.sighup)
signal.signal(signal.SIGTRAP, self.sigtrap)
signal.signal(signal.SIGUSR1, self.sigusr1)
signal.signal(signal.SIGUSR2, self.sigusr2)
signal.signal(signal.SIGTTOU, self.sigttou)
signal.signal(signal.SIGTTIN, self.sigttin)
signal.signal(signal.SIGALRM, self.sigalrm)
# make sure we always have data in history
# (done in zero for dependencies reasons)
self.monitor.zero()
def sigquit (self,signum, frame):
if self._softstop:
self.signal_log.critical('multiple SIG INT received, shutdown')
self._shutdown = True
else:
self.signal_log.critical('SIG INT received, soft-stop')
self._softstop = True
self._listen = False
def sigterm (self,signum, frame):
self.signal_log.critical('SIG TERM received, shutdown request')
if os.environ.get('PDB',False):
self._pdb = True
else:
self._shutdown = True
# def sigabrt (self,signum, frame):
# self.signal_log.info('SIG INFO received, refork request')
# self._refork = True
# def sighup (self,signum, frame):
# self.signal_log.info('SIG HUP received, reload request')
# self._reload = True
def sigtrap (self,signum, frame):
self.signal_log.critical('SIG TRAP received, toggle debug')
self._toggle_debug = True
def sigusr1 (self,signum, frame):
self.signal_log.critical('SIG USR1 received, decrease worker number')
self._decrease_spawn_limit += 1
def sigusr2 (self,signum, frame):
self.signal_log.critical('SIG USR2 received, increase worker number')
self._increase_spawn_limit += 1
def sigttou (self,signum, frame):
self.signal_log.critical('SIG TTOU received, stop listening')
self._listen = False
def sigttin (self,signum, frame):
self.signal_log.critical('SIG IN received, star listening')
self._listen = True
def sigalrm (self,signum, frame):
self.signal_log.debug('SIG ALRM received, timed actions')
self.reactor.running = False
signal.setitimer(signal.ITIMER_REAL,self.alarm_time,self.alarm_time)
def interfaces (self):
local = set(['127.0.0.1','::1'])
for interface in getifaddrs():
if interface.family not in (AF_INET,AF_INET6):
continue
if interface.address not in self.local:
self.log.info('found new local ip %s (%s)' % (interface.address,interface.name))
local.add(interface.address)
for ip in self.local:
if ip not in local:
self.log.info('removed local ip %s' % ip)
if local == self.local:
self.log.info('no ip change')
else:
self.local = local
def run (self):
if self.daemon.drop_privileges():
self.log.critical('Could not drop privileges to \'%s\'. Refusing to run as root' % self.daemon.user)
self.log.critical('Set the environment value USER to change the unprivileged user')
self._shutdown = True
elif not self.initialise():
self._shutdown = True
signal.setitimer(signal.ITIMER_REAL,self.alarm_time,self.alarm_time)
count_second = 0
count_minute = 0
count_increase = 0
count_decrease = 0
count_saturation = 0
count_interface = 0
while True:
count_second = (count_second + 1) % self.second_frequency
count_minute = (count_minute + 1) % self.minute_frequency
count_increase = (count_increase + 1) % self.increase_frequency
count_decrease = (count_decrease + 1) % self.decrease_frequency
count_saturation = (count_saturation + 1) % self.saturation_frequency
count_interface = (count_interface + 1) % self.interface_frequency
try:
if self._pdb:
self._pdb = False
import pdb
pdb.set_trace()
# check for IO change with select
self.reactor.run()
# must follow the reactor so we are sure to go through the reactor at least once
# and flush any logs
if self._shutdown:
self._shutdown = False
self.shutdown()
break
elif self._reload:
self._reload = False
self.reload()
elif self._refork:
self._refork = False
self.signal_log.warning('refork not implemented')
# stop listening to new connections
# refork the program (as we have been updated)
# just handle current open connection
if self._softstop:
if self._listen == False:
self.proxy.rejecting()
self._listen = None
if self.client.softstop():
self._shutdown = True
# only change listening if we are not shutting down
elif self._listen is not None:
if self._listen:
self._shutdown = not self.proxy.accepting()
self._listen = None
else:
self.proxy.rejecting()
self._listen = None
if self._toggle_debug:
self._toggle_debug = False
self.log_writer.toggleDebug()
if self._increase_spawn_limit:
number = self._increase_spawn_limit
self._increase_spawn_limit = 0
self.manager.low += number
self.manager.high = max(self.manager.low,self.manager.high)
for _ in range(number):
self.manager.increase()
if self._decrease_spawn_limit:
number = self._decrease_spawn_limit
self._decrease_spawn_limit = 0
self.manager.high = max(1,self.manager.high-number)
self.manager.low = min(self.manager.high,self.manager.low)
for _ in range(number):
self.manager.decrease()
# save our monitoring stats
if count_second == 0:
self.monitor.second()
expired = self.reactor.client.expire()
self.reactor.log.debug('events : ' + ', '.join('%s:%d' % (k,len(v)) for (k,v) in self.reactor.events.items()))
else:
expired = 0
if expired:
self.proxy.notifyClose(None, count=expired)
if count_minute == 0:
self.monitor.minute()
# make sure we have enough workers
if count_increase == 0:
self.manager.provision()
# and every so often remove useless workers
if count_decrease == 0:
self.manager.deprovision()
# report if we saw too many connections
if count_saturation == 0:
self.proxy.saturation()
self.web.saturation()
if self.configuration.daemon.poll_interfaces and count_interface == 0:
self.interfaces()
except KeyboardInterrupt:
self.log.critical('^C received')
self._shutdown = True
except OSError,e:
# This shoould never happen as we are limiting how many connections we accept
if e.errno == 24: # Too many open files
self.log.critical('Too many opened files, shutting down')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
else:
self.log.critical('unrecoverable io error')
for line in traceback.format_exc().split('\n'):
self.log.critical(line)
self._shutdown = True
finally:
class ClientManager (object):
def __init__(self, poller, configuration):
self.total_sent4 = 0L
self.total_sent6 = 0L
self.total_requested = 0L
self.norequest = TimeCache(configuration.http.idle_connect)
self.bysock = {}
self.byname = {}
self.buffered = []
self._nextid = 0
self.poller = poller
self.log = Logger('client', configuration.log.client)
self.http_max_buffer = configuration.http.header_size
self.icap_max_buffer = configuration.icap.header_size
self.tls_max_buffer = configuration.tls.header_size
self.passthrough_max_buffer = 0
self.proxied = {
'proxy' : configuration.http.proxied,
'icap' : configuration.icap.proxied,
'tls' : configuration.tls.proxied,
}
def __contains__(self, item):
return item in self.bysock
def lookupSocket (self, item):
return self.byname.get(item, None)
def getnextid(self):
self._nextid += 1
return str(self._nextid)
def expire (self,number=100):
counts = {}
for sock in self.norequest.expired(number):
client, source = self.norequest.get(sock,(None,None))
if client:
self.cleanup(sock,client.name)
counts[source] = counts.get(source, 0) + 1
return counts
def httpConnection (self, sock, peer, source):
name = self.getnextid()
client = HTTPClient(name, sock, peer, self.log, self.http_max_buffer, self.proxied.get(source))
self.norequest[sock] = client, source
self.byname[name] = sock
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def icapConnection (self, sock, peer, source):
name = self.getnextid()
client = ICAPClient(name, sock, peer, self.log, self.icap_max_buffer, self.proxied.get(source))
self.norequest[sock] = client, source
self.byname[name] = sock
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def tlsConnection (self, sock, peer, source):
name = self.getnextid()
client = TLSClient(name, sock, peer, self.log, self.tls_max_buffer, self.proxied.get(source))
self.norequest[sock] = client, source
self.byname[name] = sock
# watch for the opening request
self.poller.addReadSocket('opening_client', client.sock)
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return peer
def passthroughConnection (self, sock, peer, source):
name = self.getnextid()
client = PassthroughClient(name, sock, peer, self.log, self.passthrough_max_buffer, self.proxied.get(source))
self.bysock[sock] = client, source
self.byname[name] = sock
# watch for the opening data
self.poller.addReadSocket('read_client', client.sock)
accept_addr, accept_port = client.getAcceptAddress()
#self.log.info('new id %s (socket %s) in clients : %s' % (name, sock, sock in self.bysock))
return name, accept_addr, accept_port
def readRequest (self, sock):
"""Read only the initial HTTP headers sent by the client"""
client, source = self.norequest.get(sock, (None, None))
if client:
name, accept_addr, accept_port, peer, request, subrequest, content = client.readData()
if request:
self.total_requested += 1
# headers can be read only once
self.norequest.pop(sock, (None, None))
self.bysock[sock] = client, source
# watch for the client sending new data
self.poller.addReadSocket('read_client', client.sock)
# we have now read the client's opening request
self.poller.removeReadSocket('opening_client', client.sock)
# do not read more data until we have properly handled the request
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read headers from a client that does not exist %s' % sock)
name, accept_addr, accept_port, peer, request, subrequest, content, source = None, None, None, None, None, None, None, None
return name, accept_addr, accept_port, peer, request, subrequest, content, source
def readData (self, sock):
client, source = self.bysock.get(sock, (None, None))
if client:
name, accept_addr, accept_port, peer, request, subrequest, content = client.readData()
if request:
self.total_requested += 1
# Parsing of the new request will be handled asynchronously. Ensure that
# we do not read anything from the client until a request has been sent
# to the remote webserver.
# Since we just read a request, we know that the cork is not currently
# set and so there's no risk of it being erroneously removed.
self.poller.corkReadSocket('read_client', sock)
elif request is None:
self.cleanup(sock, client.name)
else:
self.log.error('trying to read from a client that does not exist %s' % sock)
name, accept_addr, accept_port, peer, request, subrequest, content = None, None, None, None, None, None, None
return name, accept_addr, accept_port, peer, request, subrequest, content, source
def sendData (self, sock, data):
client, source = self.bysock.get(sock, (None, None))
if client:
name = client.name
res = client.writeData(data)
if res is None:
# close the client connection
self.cleanup(sock, client.name)
buffered, had_buffer, sent4, sent6 = None, None, 0, 0
result = None
buffer_change = None
else:
buffered, had_buffer, sent4, sent6 = res
self.total_sent4 += sent4
self.total_sent6 += sent6
result = buffered
if buffered:
if sock not in self.buffered:
self.buffered.append(sock)
buffer_change = True
# watch for the socket's send buffer becoming less than full
self.poller.addWriteSocket('write_client', client.sock)
else:
buffer_change = False
elif had_buffer and sock in self.buffered:
self.buffered.remove(sock)
buffer_change = True
# we no longer care about writing to the client
self.poller.removeWriteSocket('write_client', client.sock)
else:
buffer_change = False
else:
result = None
buffer_change = None
name = None
return result, buffer_change, name, source
def parseRemaining (self, remaining):
nb_to_read = 0
if isinstance(remaining, basestring):
mode = 'chunked' if remaining == 'chunked' else 'passthrough'
elif remaining > 0:
mode = 'transfer'
nb_to_read = remaining
elif remaining == 0:
mode = ''
else:
mode = 'passthrough'
return mode, nb_to_read
def startData(self, sock, data, remaining):
client, source = self.bysock.get(sock, (None, None))
try:
mode, nb_to_read = self.parseRemaining(remaining)
command, d = data if client is not None else (None, None)
except (ValueError, TypeError), e:
self.log.error('invalid command sent to client %s' % client.name)
command, d = None, None
if not client or command is None:
return None, source
name, peer, res = client.startData(command, d)
if res is not None:
name, accept_addr, peer, request, subrequest, content = client.readRelated(mode, nb_to_read)
buffered, had_buffer, sent4, sent6 = res
self.poller.uncorkReadSocket('read_client', client.sock)
self.total_sent4 += sent4
self.total_sent6 += sent6
else:
self.cleanup(client.sock, name)
return None, source
if request:
self.total_requested += 1
self.log.info('reading multiple requests')
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
elif request is None:
self.cleanup(client.sock, name)
buffered, had_buffer = None, None
content = None
if buffered is True and had_buffer is False:
self.buffered.append(client.sock)
self.poller.addWriteSocket('write_client', client.sock)
elif buffered is False and had_buffer is True:
self.buffered.remove(client.sock)
self.poller.removeWriteSocket('write_client', client.sock)
return content, source
