ansible/bin/ansible-connection
Peter Sprygada 6fe9a5e40c functional updates to ansible-connection (#18574)
* sends the serialized play_context into an already established connection
* hooks the alarm_handler() method in the connection plugin if it exists
* added configuration options for connect interval and retries
* adds syslog logging to Server() instance

This update will send the updated play_context back into an already
established connection in case privilege escalation / descalation activities
need to be performed.  This change will also hook the alarm_handler() method
in the connection instance (if available) and call it in case of a
sigalarm raised.

This update adds two new configuration options

* PERSISTENT_CONNECT_INTERVAL - time to wait in between connection attempts
* PERSISTENT_CONNECT_RETRIES - max number of retries
2016-11-30 16:26:49 -05:00

341 lines
12 KiB
Python
Executable file

#!/usr/bin/env python
# (c) 2016, Ansible, Inc. <support@ansible.com>
#
# This file is part of Ansible
#
# Ansible is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# Ansible is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with Ansible. If not, see <http://www.gnu.org/licenses/>.
########################################################
from __future__ import (absolute_import, division, print_function)
__metaclass__ = type
__requires__ = ['ansible']
try:
import pkg_resources
except Exception:
pass
import fcntl
import hashlib
import os
import shlex
import signal
import socket
import struct
import sys
import time
import traceback
import syslog
import datetime
from io import BytesIO
from ansible import constants as C
from ansible.module_utils._text import to_bytes, to_native
from ansible.module_utils.six.moves import cPickle, StringIO
from ansible.playbook.play_context import PlayContext
from ansible.plugins import connection_loader
from ansible.utils.path import unfrackpath, makedirs_safe
def do_fork():
'''
Does the required double fork for a daemon process. Based on
http://code.activestate.com/recipes/66012-fork-a-daemon-process-on-unix/
'''
try:
pid = os.fork()
if pid > 0:
return pid
os.chdir("/")
os.setsid()
os.umask(0)
try:
pid = os.fork()
if pid > 0:
sys.exit(0)
os.close(sys.stdin.fileno())
os.close(sys.stdout.fileno())
os.close(sys.stderr.fileno())
return pid
except OSError as e:
sys.exit(1)
except OSError as e:
sys.exit(1)
def send_data(s, data):
packed_len = struct.pack('!Q',len(data))
return s.sendall(packed_len + data)
def recv_data(s):
header_len = 8 # size of a packed unsigned long long
data = b""
while len(data) < header_len:
d = s.recv(header_len - len(data))
if not d:
return None
data += d
data_len = struct.unpack('!Q',data[:header_len])[0]
data = data[header_len:]
while len(data) < data_len:
d = s.recv(data_len - len(data))
if not d:
return None
data += d
return data
class Server():
def __init__(self, path, play_context):
self.path = path
self.play_context = play_context
self._start_time = datetime.datetime.now()
try:
# FIXME: the connection loader here is created brand new,
# so it will not see any custom paths loaded (ie. via
# roles), so we will need to serialize the connection
# loader and send it over as we do the PlayContext
# in the main() method below.
self.log('loading connection plugin %s' % str(play_context.connection))
self.conn = connection_loader.get(play_context.connection, play_context, sys.stdin)
self.conn._connect()
self.socket = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
self.socket.bind(path)
self.socket.listen(1)
except Exception as exc:
self.log(exc)
return
signal.signal(signal.SIGALRM, self.alarm_handler)
def log(self, msg):
syslog_msg = '[%s] %s' % (self.play_context.remote_addr, msg)
facility = getattr(syslog, C.DEFAULT_SYSLOG_FACILITY, syslog.LOG_USER)
syslog.openlog('ansible-connection', 0, facility)
syslog.syslog(syslog.LOG_INFO, syslog_msg)
def dispatch(self, obj, name, *args, **kwargs):
meth = getattr(obj, name, None)
if meth:
return meth(*args, **kwargs)
def alarm_handler(self, signum, frame):
'''
Alarm handler
'''
# FIXME: this should also set internal flags for other
# areas of code to check, so they can terminate
# earlier than the socket going back to the accept
# call and failing there.
#
# hooks the connection plugin to handle any cleanup
self.dispatch(self.conn, 'alarm_handler', signum, frame)
self.socket.close()
def run(self):
try:
while True:
# set the alarm, if we don't get an accept before it
# goes off we exit (via an exception caused by the socket
# getting closed while waiting on accept())
# FIXME: is this the best way to exit? as noted above in the
# handler we should probably be setting a flag to check
# here and in other parts of the code
signal.alarm(C.PERSISTENT_CONNECT_TIMEOUT)
try:
(s, addr) = self.socket.accept()
# clear the alarm
# FIXME: potential race condition here between the accept and
# time to this call.
signal.alarm(0)
except:
break
while True:
data = recv_data(s)
if not data:
break
signal.alarm(C.DEFAULT_TIMEOUT)
rc = 255
try:
if data.startswith(b'EXEC: '):
cmd = data.split(b'EXEC: ')[1]
(rc, stdout, stderr) = self.conn.exec_command(cmd)
elif data.startswith(b'PUT: ') or data.startswith(b'FETCH: '):
(op, src, dst) = shlex.split(to_native(data))
stdout = stderr = ''
try:
if op == 'FETCH:':
self.conn.fetch_file(src, dst)
elif op == 'PUT:':
self.conn.put_file(src, dst)
rc = 0
except:
pass
elif data.startswith(b'CONTEXT: '):
pc_data = data.split(b'CONTEXT: ')[1]
src = StringIO(pc_data)
pc_data = cPickle.load(src)
src.close()
pc = PlayContext()
pc.deserialize(pc_data)
self.dispatch(self.conn, 'update_play_context', pc)
continue
else:
stdout = ''
stderr = 'Invalid action specified'
except:
stdout = ''
stderr = traceback.format_exc()
signal.alarm(0)
send_data(s, to_bytes(str(rc)))
send_data(s, to_bytes(stdout))
send_data(s, to_bytes(stderr))
s.close()
except Exception as e:
# FIXME: proper logging and error handling here
self.log('runtime exception: %s' % e)
print(traceback.format_exc())
finally:
# when done, close the connection properly and cleanup
# the socket file so it can be recreated
end_time = datetime.datetime.now()
delta = end_time - self._start_time
self.log('shutting down connection, connection was active for %s secs' % delta)
try:
self.conn.close()
self.socket.close()
except Exception as e:
pass
os.remove(self.path)
def main():
try:
# read the play context data via stdin, which means depickling it
# FIXME: as noted above, we will probably need to deserialize the
# connection loader here as well at some point, otherwise this
# won't find role- or playbook-based connection plugins
cur_line = sys.stdin.readline()
init_data = ''
while cur_line.strip() != '#END_INIT#':
if cur_line == '':
raise Exception("EOL found before init data was complete")
init_data += cur_line
cur_line = sys.stdin.readline()
src = BytesIO(to_bytes(init_data))
pc_data = cPickle.load(src)
#src.close()
pc = PlayContext()
pc.deserialize(pc_data)
except Exception as e:
# FIXME: better error message/handling/logging
print("FAIL: %s" % e)
print(traceback.format_exc())
sys.exit(1)
# here we create a hash to use later when creating the socket file,
# so we can hide the info about the target host/user/etc.
m = hashlib.sha256()
for attr in ('connection', 'remote_addr', 'port', 'remote_user'):
val = getattr(pc, attr, None)
if val:
m.update(to_bytes(val))
# create the persistent connection dir if need be and create the paths
# which we will be using later
tmp_path = unfrackpath("$HOME/.ansible/pc")
makedirs_safe(tmp_path)
lk_path = unfrackpath("%s/.ansible_pc_lock" % tmp_path)
sf_path = unfrackpath("%s/conn-%s" % (tmp_path, m.hexdigest()[0:12]))
# if the socket file doesn't exist, spin up the daemon process
lock_fd = os.open(lk_path, os.O_RDWR|os.O_CREAT, 0o600)
fcntl.lockf(lock_fd, fcntl.LOCK_EX)
if not os.path.exists(sf_path):
pid = do_fork()
if pid == 0:
server = Server(sf_path, pc)
fcntl.lockf(lock_fd, fcntl.LOCK_UN)
os.close(lock_fd)
server.run()
sys.exit(0)
fcntl.lockf(lock_fd, fcntl.LOCK_UN)
os.close(lock_fd)
# now connect to the daemon process
# FIXME: if the socket file existed but the daemonized process was killed,
# the connection will timeout here. Need to make this more resilient.
rc = 0
while rc == 0:
data = sys.stdin.readline()
if data == '':
break
if data.strip() == '':
continue
sf = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
attempts = 1
while True:
try:
sf.connect(sf_path)
break
except socket.error:
# FIXME: better error handling/logging/message here
time.sleep(C.PERSISTENT_CONNECT_INTERVAL)
attempts += 1
if attempts > C.PERSISTENT_CONNECT_RETRIES:
sys.stderr.write("failed to connect to the host, connection timeout")
sys.exit(255)
#
# send the play_context back into the connection so the connection
# can handle any privilege escalation or deescalation activities
#
pc_data = 'CONTEXT: %s' % src.getvalue()
send_data(sf, to_bytes(pc_data))
src.close()
send_data(sf, to_bytes(data.strip()))
rc = int(recv_data(sf), 10)
stdout = recv_data(sf)
stderr = recv_data(sf)
sys.stdout.write(to_native(stdout))
sys.stderr.write(to_native(stderr))
sf.close()
break
sys.exit(rc)
if __name__ == '__main__':
main()