src/nepi/resources/linux/udptunnel.py

   1 #
   2 #    NEPI, a framework to manage network experiments
   3 #    Copyright (C) 2013 INRIA
   4 #
   5 #    This program is free software: you can redistribute it and/or modify
   6 #    it under the terms of the GNU General Public License as published by
   7 #    the Free Software Foundation, either version 3 of the License, or
   8 #    (at your option) any later version.
   9 #
  10 #    This program is distributed in the hope that it will be useful,
  11 #    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 #    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 #    GNU General Public License for more details.
  14 #
  15 #    You should have received a copy of the GNU General Public License
  16 #    along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17 #
  18 # Author: Alina Quereilhac <alina.quereilhac@inria.fr>
  19
  20 from nepi.execution.attribute import Attribute, Flags, Types
  21 from nepi.execution.resource import clsinit_copy, ResourceState, \
  22         reschedule_delay
  23 from nepi.resources.linux.application import LinuxApplication
  24 from nepi.util.sshfuncs import ProcStatus
  25 from nepi.util.timefuncs import tnow, tdiffsec
  26
  27 import os
  28 import socket
  29 import time
  30
  31 @clsinit_copy
  32 class UdpTunnel(LinuxApplication):
  33     _rtype = "UdpTunnel"
  34     _help = "Constructs a tunnel between two Linux endpoints using a UDP connection "
  35     _backend = "linux"
  36
  37
  38     @classmethod
  39     def _register_attributes(cls):
  40         cipher = Attribute("cipher",
  41                "Cipher to encript communication. "
  42                 "One of PLAIN, AES, Blowfish, DES, DES3. ",
  43                 default = None,
  44                 allowed = ["PLAIN", "AES", "Blowfish", "DES", "DES3"],
  45                 type = Types.Enumerate,
  46                 flags = Flags.ExecReadOnly)
  47
  48         cipher_key = Attribute("cipherKey",
  49                 "Specify a symmetric encryption key with which to protect "
  50                 "packets across the tunnel. python-crypto must be installed "
  51                 "on the system." ,
  52                 flags = Flags.ExecReadOnly)
  53
  54         txqueuelen = Attribute("txQueueLen",
  55                 "Specifies the interface's transmission queue length. "
  56                 "Defaults to 1000. ",
  57                 type = Types.Integer,
  58                 flags = Flags.ExecReadOnly)
  59
  60         bwlimit = Attribute("bwLimit",
  61                 "Specifies the interface's emulated bandwidth in bytes "
  62                 "per second.",
  63                 type = Types.Integer,
  64                 flags = Flags.ExecReadOnly)
  65
  66         cls._register_attribute(cipher)
  67         cls._register_attribute(cipher_key)
  68         cls._register_attribute(txqueuelen)
  69         cls._register_attribute(bwlimit)
  70
  71     def __init__(self, ec, guid):
  72         super(UdpTunnel, self).__init__(ec, guid)
  73         self._home = "udp-tunnel-%s" % self.guid
  74         self._pid1 = None
  75         self._ppid1 = None
  76         self._pid2 = None
  77         self._ppid2 = None
  78
  79     def log_message(self, msg):
  80         return " guid %d - tunnel %s - %s - %s " % (self.guid,
  81                 self.endpoint1.node.get("hostname"),
  82                 self.endpoint2.node.get("hostname"),
  83                 msg)
  84
  85     def get_endpoints(self):
  86         """ Returns the list of RM that are endpoints to the tunnel
  87         """
  88         connected = []
  89         for guid in self.connections:
  90             rm = self.ec.get_resource(guid)
  91             if hasattr(rm, "udp_connect_command"):
  92                 connected.append(rm)
  93         return connected
  94
  95     @property
  96     def endpoint1(self):
  97         endpoints = self.get_endpoints()
  98         if endpoints: return endpoints[0]
  99         return None
 100
 101     @property
 102     def endpoint2(self):
 103         endpoints = self.get_endpoints()
 104         if endpoints and len(endpoints) > 1: return endpoints[1]
 105         return None
 106
 107     def app_home(self, endpoint):
 108         return os.path.join(endpoint.node.exp_home, self._home)
 109
 110     def run_home(self, endpoint):
 111         return os.path.join(self.app_home(endpoint), self.ec.run_id)
 112
 113     def udp_connect(self, endpoint, remote_ip):
 114         # Get udp connect command
 115         local_port_file = os.path.join(self.run_home(endpoint),
 116                 "local_port")
 117         remote_port_file = os.path.join(self.run_home(endpoint),
 118                 "remote_port")
 119         ret_file = os.path.join(self.run_home(endpoint),
 120                 "ret_file")
 121         cipher = self.get("cipher")
 122         cipher_key = self.get("cipherKey")
 123         bwlimit = self.get("bwLimit")
 124         txqueuelen = self.get("txQueueLen")
 125         udp_connect_command = endpoint.udp_connect_command(
 126                 remote_ip, local_port_file, remote_port_file,
 127                 ret_file, cipher, cipher_key, bwlimit, txqueuelen)
 128
 129         # upload command to connect.sh script
 130         shfile = os.path.join(self.app_home(endpoint), "udp-connect.sh")
 131         endpoint.node.upload(udp_connect_command,
 132                 shfile,
 133                 text = True,
 134                 overwrite = False)
 135
 136         # invoke connect script
 137         cmd = "bash %s" % shfile
 138         (out, err), proc = endpoint.node.run(cmd, self.run_home(endpoint))
 139
 140         # check if execution errors occurred
 141         msg = " Failed to connect endpoints "
 142
 143         if proc.poll():
 144             self.fail()
 145             self.error(msg, out, err)
 146             raise RuntimeError, msg
 147
 148         # Wait for pid file to be generated
 149         pid, ppid = endpoint.node.wait_pid(self.run_home(endpoint))
 150
 151         # If the process is not running, check for error information
 152         # on the remote machine
 153         if not pid or not ppid:
 154             (out, err), proc = endpoint.node.check_errors(self.run_home(endpoint))
 155             # Out is what was written in the stderr file
 156             if err:
 157                 self.fail()
 158                 msg = " Failed to start command '%s' " % command
 159                 self.error(msg, out, err)
 160                 raise RuntimeError, msg
 161
 162         # wait until port is written to file
 163         port = self.wait_local_port(endpoint)
 164         return (port, pid, ppid)
 165
 166     def provision(self):
 167         # create run dir for tunnel on each node
 168         self.endpoint1.node.mkdir(self.run_home(self.endpoint1))
 169         self.endpoint2.node.mkdir(self.run_home(self.endpoint2))
 170
 171         # Invoke connect script in endpoint 1
 172         remote_ip1 = socket.gethostbyname(self.endpoint2.node.get("hostname"))
 173         (port1, self._pid1, self._ppid1) = self.udp_connect(self.endpoint1,
 174                 remote_ip1)
 175
 176         # Invoke connect script in endpoint 2
 177         remote_ip2 = socket.gethostbyname(self.endpoint1.node.get("hostname"))
 178         (port2, self._pid2, self._ppid2) = self.udp_connect(self.endpoint2,
 179                 remote_ip2)
 180
 181         # upload file with port 2 to endpoint 1
 182         self.upload_remote_port(self.endpoint1, port2)
 183
 184         # upload file with port 1 to endpoint 2
 185         self.upload_remote_port(self.endpoint2, port1)
 186
 187         # check if connection was successful on both sides
 188         self.wait_result(self.endpoint1)
 189         self.wait_result(self.endpoint2)
 190
 191         self.info("Provisioning finished")
 192
 193         self.set_provisioned()
 194
 195     def deploy(self):
 196         if (not self.endpoint1 or self.endpoint1.state < ResourceState.READY) or \
 197             (not self.endpoint2 or self.endpoint2.state < ResourceState.READY):
 198             self.ec.schedule(reschedule_delay, self.deploy)
 199         else:
 200             try:
 201                 self.discover()
 202                 self.provision()
 203             except:
 204                 self.fail()
 205                 raise
 206
 207             self.debug("----- READY ---- ")
 208             self.set_ready()
 209
 210     def start(self):
 211         if self.state == ResourceState.READY:
 212             command = self.get("command")
 213             self.info("Starting command '%s'" % command)
 214
 215             self.set_started()
 216         else:
 217             msg = " Failed to execute command '%s'" % command
 218             self.error(msg, out, err)
 219             self.fail()
 220             raise RuntimeError, msg
 221
 222     def stop(self):
 223         """ Stops application execution
 224         """
 225         if self.state == ResourceState.STARTED:
 226             self.info("Stopping tunnel")
 227
 228             # Only try to kill the process if the pid and ppid
 229             # were retrieved
 230             if self._pid1 and self._ppid1 and self._pid2 and self._ppid2:
 231                 (out1, err1), proc1 = self.endpoint1.node.kill(self._pid1,
 232                         self._ppid1, sudo = True)
 233                 (out2, err2), proc2 = self.endpoint2.node.kill(self._pid2,
 234                         self._ppid2, sudo = True)
 235
 236                 if err1 or err2 or proc1.poll() or proc2.poll():
 237                     # check if execution errors occurred
 238                     msg = " Failed to STOP tunnel"
 239                     self.error(msg, err1, err2)
 240                     self.fail()
 241
 242         if self.state == ResourceState.STARTED:
 243             self.set_stopped()
 244
 245     @property
 246     def state(self):
 247         """ Returns the state of the application
 248         """
 249         if self._state == ResourceState.STARTED:
 250             # In order to avoid overwhelming the remote host and
 251             # the local processor with too many ssh queries, the state is only
 252             # requested every 'state_check_delay' seconds.
 253             state_check_delay = 0.5
 254             if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
 255                 if self._pid1 and self._ppid1 and self._pid2 and self._ppid2:
 256                     # Make sure the process is still running in background
 257                     # No execution errors occurred. Make sure the background
 258                     # process with the recorded pid is still running.
 259                     status1 = self.endpoint1.node.status(self._pid1, self._ppid1)
 260                     status2 = self.endpoint2.node.status(self._pid2, self._ppid2)
 261
 262                     if status1 == ProcStatus.FINISHED and \
 263                             status2 == ProcStatus.FINISHED:
 264
 265                         # check if execution errors occurred
 266                         (out1, err1), proc1 = self.endpoint1.node.check_errors(
 267                                 self.run_home(self.endpoint1))
 268
 269                         (out2, err2), proc2 = self.endpoint2.node.check_errors(
 270                                 self.run_home(self.endpoint2))
 271
 272                         if err1 or err2:
 273                             msg = "Error occurred in tunnel"
 274                             self.error(msg, err1, err2)
 275                             self.fail()
 276                         else:
 277                             self.set_finished()
 278
 279                 self._last_state_check = tnow()
 280
 281         return self._state
 282
 283     def wait_local_port(self, endpoint):
 284         """ Waits until the local_port file for the endpoint is generated,
 285         and returns the port number
 286
 287         """
 288         return self.wait_file(endpoint, "local_port")
 289
 290     def wait_result(self, endpoint):
 291         """ Waits until the return code file for the endpoint is generated
 292
 293         """
 294         return self.wait_file(endpoint, "ret_file")
 295
 296     def wait_file(self, endpoint, filename):
 297         """ Waits until file on endpoint is generated """
 298         result = None
 299         delay = 1.0
 300
 301         for i in xrange(4):
 302             (out, err), proc = endpoint.node.check_output(
 303                     self.run_home(endpoint), filename)
 304
 305             if out:
 306                 result = out.strip()
 307                 break
 308             else:
 309                 time.sleep(delay)
 310                 delay = delay * 1.5
 311         else:
 312             msg = "Couldn't retrieve %s" % filename
 313             self.error(msg, out, err)
 314             self.fail()
 315             raise RuntimeError, msg
 316
 317         return result
 318
 319     def upload_remote_port(self, endpoint, port):
 320         # upload remote port number to file
 321         port = "%s\n" % port
 322         endpoint.node.upload(port,
 323                 os.path.join(self.run_home(endpoint), "remote_port"),
 324                 text = True,
 325                 overwrite = False)
 326
 327     def valid_connection(self, guid):
 328         # TODO: Validate!
 329         return True
 330