reboot.py

   1 #!/usr/bin/python
   2 #
   3 # Reboot specified nodes
   4 #
   5
   6 import getpass, getopt
   7 import os, sys
   8 import xml, xmlrpclib
   9 import errno, time, traceback
  10 import urllib2
  11 import threading, popen2
  12 import array, struct
  13 #from socket import *
  14 import socket
  15 import plc
  16 import base64
  17 from subprocess import PIPE, Popen
  18 import ssh.pxssh as pxssh
  19 import ssh.pexpect as pexpect
  20 import socket
  21
  22 # Use our versions of telnetlib and pyssh
  23 sys.path.insert(0, os.path.dirname(sys.argv[0]))
  24 import telnetlib
  25 sys.path.insert(0, os.path.dirname(sys.argv[0]) + "/pyssh")
  26 import pyssh
  27
  28 # Timeouts in seconds
  29 TELNET_TIMEOUT = 45
  30
  31 # Event class ID from pcu events
  32 #NODE_POWER_CONTROL = 3
  33
  34 # Monitor user ID
  35 #MONITOR_USER_ID = 11142
  36
  37 import logging
  38 logger = logging.getLogger("monitor")
  39 verbose = 1
  40 #dryrun = 0;
  41
  42 class ExceptionNoTransport(Exception): pass
  43 class ExceptionNotFound(Exception): pass
  44 class ExceptionPassword(Exception): pass
  45 class ExceptionTimeout(Exception): pass
  46 class ExceptionPrompt(Exception): pass
  47 class ExceptionSequence(Exception): pass
  48 class ExceptionReset(Exception): pass
  49 class ExceptionPort(Exception): pass
  50 class ExceptionUsername(Exception): pass
  51
  52 def telnet_answer(telnet, expected, buffer):
  53         global verbose
  54
  55         output = telnet.read_until(expected, TELNET_TIMEOUT)
  56         #if verbose:
  57         #       logger.debug(output)
  58         if output.find(expected) == -1:
  59                 raise ExceptionNotFound, "'%s' not found" % expected
  60         else:
  61                 telnet.write(buffer + "\r\n")
  62
  63
  64 # PCU has model, host, preferred-port, user, passwd,
  65
  66 # This is an object derived directly form the PLCAPI DB fields
  67 class PCU(object):
  68         def __init__(self, plc_pcu_dict):
  69                 for field in ['username', 'password', 'site_id',
  70                                                 'hostname', 'ip',
  71                                                 'pcu_id', 'model',
  72                                                 'node_ids', 'ports', ]:
  73                         if field in plc_pcu_dict:
  74                                 self.__setattr__(field, plc_pcu_dict[field])
  75                         else:
  76                                 raise Exception("No such field %s in PCU object" % field)
  77
  78 # These are the convenience functions build around the PCU object.
  79 class PCUModel(PCU):
  80         def __init__(self, plc_pcu_dict):
  81                 PCU.__init__(self, plc_pcu_dict)
  82                 self.host = self.pcu_name()
  83
  84         def pcu_name(self):
  85                 if self.hostname is not None and self.hostname is not "":
  86                         return self.hostname
  87                 elif self.ip is not None and self.ip is not "":
  88                         return self.ip
  89                 else:
  90                         return None
  91
  92         def nodeidToPort(self, node_id):
  93                 if node_id in self.node_ids:
  94                         for i in range(0, len(self.node_ids)):
  95                                 if node_id == self.node_ids[i]:
  96                                         return self.ports[i]
  97
  98                 raise Exception("No such Node ID: %d" % node_id)
  99
 100 # This class captures the observed pcu records from FindBadPCUs.py
 101 class PCURecord:
 102         def __init__(self, pcu_record_dict):
 103                 for field in ['nodenames', 'portstatus',
 104                                                 'dnsmatch',
 105                                                 'complete_entry', ]:
 106                         if field in pcu_record_dict:
 107                                 if field == "reboot":
 108                                         self.__setattr__("reboot_str", pcu_record_dict[field])
 109                                 else:
 110                                         self.__setattr__(field, pcu_record_dict[field])
 111                         else:
 112                                 raise Exception("No such field %s in pcu record dict" % field)
 113
 114 class Transport:
 115         TELNET = 1
 116         SSH    = 2
 117         HTTP   = 3
 118
 119         TELNET_TIMEOUT = 60
 120
 121         def __init__(self, type, verbose):
 122                 self.type = type
 123                 self.verbose = verbose
 124                 self.transport = None
 125
 126 #       def __del__(self):
 127 #               if self.transport:
 128 #                       self.close()
 129
 130         def open(self, host, username=None, password=None, prompt="User Name"):
 131                 transport = None
 132
 133                 if self.type == self.TELNET:
 134                         transport = telnetlib.Telnet(host, timeout=self.TELNET_TIMEOUT)
 135                         transport.set_debuglevel(self.verbose)
 136                         if username is not None:
 137                                 self.transport = transport
 138                                 self.ifThenSend(prompt, username, ExceptionUsername)
 139
 140                 elif self.type == self.SSH:
 141                         if username is not None:
 142                                 transport = pyssh.Ssh(username, host)
 143                                 transport.set_debuglevel(self.verbose)
 144                                 transport.open()
 145                                 # TODO: have an ssh set_debuglevel() also...
 146                         else:
 147                                 raise Exception("Username cannot be None for ssh transport.")
 148                 elif self.type == self.HTTP:
 149                         self.url = "http://%s:%d/" % (host,80)
 150                         uri = "%s:%d" % (host,80)
 151
 152                         # create authinfo
 153                         authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 154                         authinfo.add_password (None, uri, username, password)
 155                         authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 156
 157                         transport = urllib2.build_opener(authhandler)
 158
 159                 else:
 160                         raise Exception("Unknown transport type: %s" % self.type)
 161
 162                 self.transport = transport
 163                 return True
 164
 165         def close(self):
 166                 if self.type == self.TELNET:
 167                         self.transport.close()
 168                 elif self.type == self.SSH:
 169                         self.transport.close()
 170                 elif self.type == self.HTTP:
 171                         pass
 172                 else:
 173                         raise Exception("Unknown transport type %s" % self.type)
 174                 self.transport = None
 175
 176         def sendHTTP(self, resource, data):
 177                 if self.verbose:
 178                         print "POSTing '%s' to %s" % (data,self.url + resource)
 179
 180                 try:
 181                         f = self.transport.open(self.url + resource ,data)
 182                         r = f.read()
 183                         if self.verbose:
 184                                 print r
 185
 186                 except urllib2.URLError,err:
 187                         logger.info('Could not open http connection', err)
 188                         return "http transport error"
 189
 190                 return 0
 191
 192         def sendPassword(self, password, prompt=None):
 193                 if self.type == self.TELNET:
 194                         if prompt == None:
 195                                 self.ifThenSend("Password", password, ExceptionPassword)
 196                         else:
 197                                 self.ifThenSend(prompt, password, ExceptionPassword)
 198                 elif self.type == self.SSH:
 199                         self.ifThenSend("password:", password, ExceptionPassword)
 200                 elif self.type == self.HTTP:
 201                         pass
 202                 else:
 203                         raise Exception("Unknown transport type: %s" % self.type)
 204
 205         def ifThenSend(self, expected, buffer, ErrorClass=ExceptionPrompt):
 206
 207                 if self.transport != None:
 208                         output = self.transport.read_until(expected, self.TELNET_TIMEOUT)
 209                         if output.find(expected) == -1:
 210                                 raise ErrorClass, "'%s' not found" % expected
 211                         else:
 212                                 self.transport.write(buffer + "\r\n")
 213                 else:
 214                         raise ExceptionNoTransport("transport object is type None")
 215
 216         def ifElse(self, expected, ErrorClass):
 217                 try:
 218                         self.transport.read_until(expected, self.TELNET_TIMEOUT)
 219                 except:
 220                         raise ErrorClass("Could not find '%s' within timeout" % expected)
 221
 222
 223 class PCUControl(Transport,PCUModel,PCURecord):
 224         def __init__(self, plc_pcu_record, verbose, supported_ports=[]):
 225                 PCUModel.__init__(self, plc_pcu_record)
 226                 PCURecord.__init__(self, plc_pcu_record)
 227                 type = None
 228                 if self.portstatus:
 229                         if '22' in supported_ports and self.portstatus['22'] == "open":
 230                                 type = Transport.SSH
 231                         elif '23' in supported_ports and self.portstatus['23'] == "open":
 232                                 type = Transport.TELNET
 233                         elif '80' in supported_ports and self.portstatus['80'] == "open":
 234                                 type = Transport.HTTP
 235                         elif '443' in supported_ports and self.portstatus['443'] == "open":
 236                                 type = Transport.HTTP
 237                         elif '5869' in supported_ports and self.portstatus['5869'] == "open":
 238                                 # For DRAC cards.  not sure how much it's used in the
 239                                 # protocol.. but racadm opens this port.
 240                                 type = Transport.HTTP
 241                         else:
 242                                 raise ExceptionPort("Unsupported Port: No transport from open ports")
 243                 Transport.__init__(self, type, verbose)
 244
 245         def run(self, node_port, dryrun):
 246                 """ This function is to be defined by the specific PCU instance.  """
 247                 pass
 248
 249         def reboot(self, node_port, dryrun):
 250                 try:
 251                         return self.run(node_port, dryrun)
 252                 except ExceptionNotFound, err:
 253                         return "error: " + str(err)
 254                 except ExceptionPassword, err:
 255                         return "password exception: " + str(err)
 256                 except ExceptionTimeout, err:
 257                         return "timeout exception: " + str(err)
 258                 except ExceptionUsername, err:
 259                         return "exception: no username prompt: " + str(err)
 260                 except ExceptionSequence, err:
 261                         return "sequence error: " + str(err)
 262                 except ExceptionPrompt, err:
 263                         return "prompt exception: " + str(err)
 264                 except ExceptionPort, err:
 265                         return "no ports exception: " + str(err)
 266                 except socket.error, err:
 267                         return "socket error: timeout: " + str(err)
 268                 except EOFError, err:
 269                         if self.verbose:
 270                                 logger.debug("reboot: EOF")
 271                                 logger.debug(err)
 272                         self.transport.close()
 273                         import traceback
 274                         traceback.print_exc()
 275                         return "EOF connection reset" + str(err)
 276                 #except Exception, err:
 277                 #       if self.verbose:
 278                 #               logger.debug("reboot: Exception")
 279                 #               logger.debug(err)
 280                 #       if self.transport:
 281                 #               self.transport.close()
 282                 #       import traceback
 283                 #       traceback.print_exc()
 284                 #       return  "generic exception; unknown problem."
 285
 286
 287 class IPAL(PCUControl):
 288         def run(self, node_port, dryrun):
 289                 self.open(self.host)
 290
 291                 # XXX Some iPals require you to hit Enter a few times first
 292                 self.ifThenSend("Password >", "\r\n\r\n", ExceptionNotFound)
 293
 294                 # Login
 295                 self.ifThenSend("Password >", self.password, ExceptionPassword)
 296                 self.transport.write("\r\n\r\n")
 297
 298                 if not dryrun: # P# - Pulse relay
 299                         self.ifThenSend("Enter >",
 300                                                         "P%d" % node_port,
 301                                                         ExceptionNotFound)
 302                 # Get the next prompt
 303                 self.ifElse("Enter >", ExceptionTimeout)
 304
 305                 self.close()
 306                 return 0
 307
 308 class APCEurope(PCUControl):
 309         def run(self, node_port, dryrun):
 310                 self.open(self.host, self.username)
 311                 self.sendPassword(self.password)
 312
 313                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 314                 self.ifThenSend("\r\n> ", "2")
 315                 self.ifThenSend("\r\n> ", str(node_port))
 316                 # 3- Immediate Reboot
 317                 self.ifThenSend("\r\n> ", "3")
 318
 319                 if not dryrun:
 320                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 321                                                         "YES\r\n",
 322                                                         ExceptionSequence)
 323                 else:
 324                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 325                                                         "", ExceptionSequence)
 326                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 327
 328                 self.close()
 329                 return 0
 330
 331 class APCBrazil(PCUControl):
 332         def run(self, node_port, dryrun):
 333                 self.open(self.host, self.username)
 334                 self.sendPassword(self.password)
 335
 336                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 337                 self.ifThenSend("\r\n> ", str(node_port))
 338                 # 4- Immediate Reboot
 339                 self.ifThenSend("\r\n> ", "4")
 340
 341                 if not dryrun:
 342                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 343                                                         "YES\r\n",
 344                                                         ExceptionSequence)
 345                 else:
 346                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 347                                                         "", ExceptionSequence)
 348                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 349
 350                 self.close()
 351                 return 0
 352
 353 class APCBerlin(PCUControl):
 354         def run(self, node_port, dryrun):
 355                 self.open(self.host, self.username)
 356                 self.sendPassword(self.password)
 357
 358                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 359                 self.ifThenSend("\r\n> ", "2")
 360                 self.ifThenSend("\r\n> ", "1")
 361                 self.ifThenSend("\r\n> ", str(node_port))
 362                 # 3- Immediate Reboot
 363                 self.ifThenSend("\r\n> ", "3")
 364
 365                 if not dryrun:
 366                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 367                                                         "YES\r\n",
 368                                                         ExceptionSequence)
 369                 else:
 370                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 371                                                         "", ExceptionSequence)
 372                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 373
 374                 self.close()
 375                 return 0
 376
 377 class APCFolsom(PCUControl):
 378         def run(self, node_port, dryrun):
 379                 self.open(self.host, self.username)
 380                 self.sendPassword(self.password)
 381
 382                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 383                 self.ifThenSend("\r\n> ", "2")
 384                 self.ifThenSend("\r\n> ", "1")
 385                 self.ifThenSend("\r\n> ", str(node_port))
 386                 self.ifThenSend("\r\n> ", "1")
 387
 388                 # 3- Immediate Reboot
 389                 self.ifThenSend("\r\n> ", "3")
 390
 391                 if not dryrun:
 392                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 393                                                         "YES\r\n",
 394                                                         ExceptionSequence)
 395                 else:
 396                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 397                                                         "", ExceptionSequence)
 398                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 399
 400                 self.close()
 401                 return 0
 402
 403 class APCMaster(PCUControl):
 404         def run(self, node_port, dryrun):
 405                 self.open(self.host, self.username)
 406                 self.sendPassword(self.password)
 407
 408                 # 1- Device Manager
 409                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 410                 # 3- Outlet Control/Config
 411                 self.ifThenSend("\r\n> ", "3")
 412                 # n- Outlet n
 413                 self.ifThenSend("\r\n> ", str(node_port))
 414                 # 1- Control Outlet
 415                 self.ifThenSend("\r\n> ", "1")
 416                 # 3- Immediate Reboot
 417                 self.ifThenSend("\r\n> ", "3")
 418
 419                 if not dryrun:
 420                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 421                                                         "YES\r\n",
 422                                                         ExceptionSequence)
 423                 else:
 424                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 425                                                         "", ExceptionSequence)
 426                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 427
 428                 self.close()
 429                 return 0
 430
 431 class APC(PCUControl):
 432         def __init__(self, plc_pcu_record, verbose):
 433                 PCUControl.__init__(self, plc_pcu_record, verbose)
 434
 435                 self.master = APCMaster(plc_pcu_record, verbose)
 436                 self.folsom = APCFolsom(plc_pcu_record, verbose)
 437                 self.europe = APCEurope(plc_pcu_record, verbose)
 438
 439         def run(self, node_port, dryrun):
 440                 try_again = True
 441                 sleep_time = 1
 442
 443                 for pcu in [self.master, self.europe, self.folsom]:
 444                         if try_again:
 445                                 try:
 446                                         print "-*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*"
 447                                         try_again = False
 448                                         print "sleeping 5"
 449                                         time.sleep(sleep_time)
 450                                         ret = pcu.reboot(node_port, dryrun)
 451                                 except ExceptionSequence, err:
 452                                         del pcu
 453                                         sleep_time = 130
 454                                         try_again = True
 455
 456                 if try_again:
 457                         return "Unknown reboot sequence for APC PCU"
 458                 else:
 459                         return ret
 460
 461 class DRACRacAdm(PCUControl):
 462         def run(self, node_port, dryrun):
 463
 464                 print "trying racadm_reboot..."
 465                 racadm_reboot(self.host, self.username, self.password, node_port, dryrun)
 466
 467                 return 0
 468
 469 class DRAC(PCUControl):
 470         def run(self, node_port, dryrun):
 471                 self.open(self.host, self.username)
 472                 self.sendPassword(self.password)
 473
 474                 print "logging in..."
 475                 self.transport.write("\r\n")
 476                 # Testing Reboot ?
 477                 if dryrun:
 478                         self.ifThenSend("[%s]#" % self.username, "getsysinfo")
 479                 else:
 480                         # Reset this machine
 481                         self.ifThenSend("[%s]#" % self.username, "serveraction powercycle")
 482
 483                 self.ifThenSend("[%s]#" % self.username, "exit")
 484
 485                 self.close()
 486                 return 0
 487
 488 class HPiLO(PCUControl):
 489         def run(self, node_port, dryrun):
 490                 self.open(self.host, self.username)
 491                 self.sendPassword(self.password)
 492
 493                 # </>hpiLO->
 494                 self.ifThenSend("</>hpiLO->", "cd system1")
 495
 496                 # Reboot Outlet  N        (Y/N)?
 497                 if dryrun:
 498                         self.ifThenSend("</system1>hpiLO->", "POWER")
 499                 else:
 500                         # Reset this machine
 501                         self.ifThenSend("</system1>hpiLO->", "reset")
 502
 503                 self.ifThenSend("</system1>hpiLO->", "exit")
 504
 505                 self.close()
 506                 return 0
 507
 508
 509 class HPiLOHttps(PCUControl):
 510         def run(self, node_port, dryrun):
 511
 512                 cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p %s" % (
 513                                         self.host, "iloxml/Get_Network.xml",
 514                                         self.username, self.password)
 515                 p_ilo  = Popen(cmd, stdout=PIPE, shell=True)
 516                 cmd2 = "grep 'MESSAGE' | grep -v 'No error'"
 517                 p_grep = Popen(cmd2, stdin=p_ilo.stdout, stdout=PIPE, stderr=PIPE, shell=True)
 518                 sout, serr = p_grep.communicate()
 519
 520                 p_ilo.wait()
 521                 p_grep.wait()
 522                 if sout.strip() != "":
 523                         print "sout: %s" % sout.strip()
 524                         return sout.strip()
 525
 526                 if not dryrun:
 527                         cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p %s" % (
 528                                         self.host, "iloxml/Reset_Server.xml",
 529                                         self.username, self.password)
 530                         print cmd
 531                         p_ilo = Popen(cmd, stdin=PIPE, stdout=PIPE, shell=True)
 532                         cmd2 = "grep 'MESSAGE' | grep -v 'No error'"
 533                         p_grep = Popen(cmd2, stdin=p_ilo.stdout, stdout=PIPE, stderr=PIPE)
 534                         sout, serr = p_grep.communicate()
 535                         try: p_ilo.wait()
 536                         except:
 537                                 print "p_ilo wait failed."
 538                                 pass
 539                         try: p_grep.wait()
 540                         except:
 541                                 print "p_grep wait failed."
 542                                 pass
 543
 544                         if sout.strip() != "":
 545                                 print "sout: %s" % sout.strip()
 546                                 return sout.strip()
 547
 548                 return 0
 549
 550 class BayTechAU(PCUControl):
 551         def run(self, node_port, dryrun):
 552                 self.open(self.host, self.username, None, "Enter user name:")
 553                 self.sendPassword(self.password, "Enter Password:")
 554
 555                 #self.ifThenSend("RPC-16>", "Status")
 556                 self.ifThenSend("RPC3-NC>", "Reboot %d" % node_port)
 557
 558                 # Reboot Outlet  N        (Y/N)?
 559                 if dryrun:
 560                         self.ifThenSend("(Y/N)?", "N")
 561                 else:
 562                         self.ifThenSend("(Y/N)?", "Y")
 563                 self.ifThenSend("RPC3-NC>", "")
 564
 565                 self.close()
 566                 return 0
 567
 568 class BayTechGeorgeTown(PCUControl):
 569         def run(self, node_port, dryrun):
 570                 self.open(self.host, self.username, None, "Enter user name:")
 571                 self.sendPassword(self.password, "Enter Password:")
 572
 573                 #self.ifThenSend("RPC-16>", "Status")
 574
 575                 self.ifThenSend("RPC-16>", "Reboot %d" % node_port)
 576
 577                 # Reboot Outlet  N        (Y/N)?
 578                 if dryrun:
 579                         self.ifThenSend("(Y/N)?", "N")
 580                 else:
 581                         self.ifThenSend("(Y/N)?", "Y")
 582                 self.ifThenSend("RPC-16>", "")
 583
 584                 self.close()
 585                 return 0
 586
 587 class BayTechCtrlCUnibe(PCUControl):
 588         """
 589                 For some reason, these units let you log in fine, but they hang
 590                 indefinitely, unless you send a Ctrl-C after the password.  No idea
 591                 why.
 592         """
 593         def run(self, node_port, dryrun):
 594                 print "BayTechCtrlC %s" % self.host
 595
 596                 ssh_options="-o StrictHostKeyChecking=no -o PasswordAuthentication=yes -o PubkeyAuthentication=no"
 597                 s = pxssh.pxssh()
 598                 if not s.login(self.host, self.username, self.password, ssh_options):
 599                         raise ExceptionPassword("Invalid Password")
 600                 # Otherwise, the login succeeded.
 601
 602                 # Send a ctrl-c to the remote process.
 603                 print "sending ctrl-c"
 604                 s.send(chr(3))
 605
 606                 # Control Outlets  (5 ,1).........5
 607                 try:
 608                         index = s.expect(["Enter Request :"])
 609
 610                         if index == 0:
 611                                 print "3"
 612                                 s.send("3\r\n")
 613                                 index = s.expect(["DS-RPC>", "Enter user name:"])
 614                                 if index == 1:
 615                                         s.send(self.username + "\r\n")
 616                                         index = s.expect(["DS-RPC>"])
 617
 618                                 if index == 0:
 619                                         print "Reboot %d" % node_port
 620                                         s.send("Reboot %d\r\n" % node_port)
 621
 622                                         index = s.expect(["(Y/N)?"])
 623                                         if index == 0:
 624                                                 if dryrun:
 625                                                         print "sending N"
 626                                                         s.send("N\r\n")
 627                                                 else:
 628                                                         print "sending Y"
 629                                                         s.send("Y\r\n")
 630
 631                                 #index = s.expect(["DS-RPC>"])
 632                                 #print "got prompt back"
 633
 634                         s.close()
 635
 636                 except pexpect.EOF:
 637                         raise ExceptionPrompt("EOF before 'Enter Request' Prompt")
 638                 except pexpect.TIMEOUT:
 639                         raise ExceptionPrompt("Timeout before 'Enter Request' Prompt")
 640
 641                 return 0
 642
 643 class BayTechCtrlC(PCUControl):
 644         """
 645                 For some reason, these units let you log in fine, but they hang
 646                 indefinitely, unless you send a Ctrl-C after the password.  No idea
 647                 why.
 648         """
 649         def run(self, node_port, dryrun):
 650                 print "BayTechCtrlC %s" % self.host
 651
 652                 ssh_options="-o StrictHostKeyChecking=no -o PasswordAuthentication=yes -o PubkeyAuthentication=no"
 653                 s = pxssh.pxssh()
 654                 if not s.login(self.host, self.username, self.password, ssh_options):
 655                         raise ExceptionPassword("Invalid Password")
 656                 # Otherwise, the login succeeded.
 657
 658                 # Send a ctrl-c to the remote process.
 659                 print "sending ctrl-c"
 660                 s.send(chr(3))
 661
 662                 # Control Outlets  (5 ,1).........5
 663                 try:
 664                         index = s.expect(["Enter Request :"])
 665
 666                         if index == 0:
 667                                 print "5"
 668                                 s.send("5\r\n")
 669                                 index = s.expect(["DS-RPC>", "Enter user name:"])
 670                                 if index == 1:
 671                                         print "sending username"
 672                                         s.send(self.username + "\r\n")
 673                                         index = s.expect(["DS-RPC>"])
 674
 675                                 if index == 0:
 676                                         print "Reboot %d" % node_port
 677                                         s.send("Reboot %d\r\n" % node_port)
 678
 679                                         index = s.expect(["(Y/N)?"])
 680                                         if index == 0:
 681                                                 if dryrun:
 682                                                         print "sending N"
 683                                                         s.send("N\r\n")
 684                                                 else:
 685                                                         print "sending Y"
 686                                                         s.send("Y\r\n")
 687
 688                                 #index = s.expect(["DS-RPC>"])
 689                                 #print "got prompt back"
 690
 691                         s.close()
 692
 693                 except pexpect.EOF:
 694                         raise ExceptionPrompt("EOF before 'Enter Request' Prompt")
 695                 except pexpect.TIMEOUT:
 696                         raise ExceptionPrompt("Timeout before 'Enter Request' Prompt")
 697
 698                 return 0
 699
 700 class BayTech(PCUControl):
 701         def run(self, node_port, dryrun):
 702                 self.open(self.host, self.username)
 703                 self.sendPassword(self.password)
 704
 705                 # Control Outlets  (5 ,1).........5
 706                 self.ifThenSend("Enter Request :", "5")
 707
 708                 # Reboot N
 709                 try:
 710                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port, ExceptionNotFound)
 711                 except ExceptionNotFound, msg:
 712                         # one machine is configured to ask for a username,
 713                         # even after login...
 714                         print "msg: %s" % msg
 715                         self.transport.write(self.username + "\r\n")
 716                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port)
 717
 718                 # Reboot Outlet  N        (Y/N)?
 719                 if dryrun:
 720                         self.ifThenSend("(Y/N)?", "N")
 721                 else:
 722                         self.ifThenSend("(Y/N)?", "Y")
 723                 self.ifThenSend("DS-RPC>", "")
 724
 725                 self.close()
 726                 return 0
 727
 728 class WTIIPS4(PCUControl):
 729         def run(self, node_port, dryrun):
 730                 self.open(self.host)
 731                 self.sendPassword(self.password, "Enter Password:")
 732
 733                 self.ifThenSend("IPS> ", "/Boot %s" % node_port)
 734                 if not dryrun:
 735                         self.ifThenSend("Sure? (Y/N): ", "N")
 736                 else:
 737                         self.ifThenSend("Sure? (Y/N): ", "Y")
 738
 739                 self.ifThenSend("IPS> ", "")
 740
 741                 self.close()
 742                 return 0
 743
 744 class ePowerSwitchGood(PCUControl):
 745         # NOTE:
 746         #               The old code used Python's HTTPPasswordMgrWithDefaultRealm()
 747         #               For some reason this both doesn't work and in some cases, actually
 748         #               hangs the PCU.  Definitely not what we want.
 749         #
 750         #               The code below is much simpler.  Just letting things fail first,
 751         #               and then, trying again with authentication string in the header.
 752         #
 753         def run(self, node_port, dryrun):
 754                 self.transport = None
 755                 self.url = "http://%s:%d/" % (self.host,80)
 756                 uri = "%s:%d" % (self.host,80)
 757
 758                 req = urllib2.Request(self.url)
 759                 try:
 760                         handle = urllib2.urlopen(req)
 761                 except IOError, e:
 762                         # NOTE: this is expected to fail initially
 763                         pass
 764                 else:
 765                         print self.url
 766                         print "-----------"
 767                         print handle.read()
 768                         print "-----------"
 769                         return "ERROR: not protected by HTTP authentication"
 770
 771                 if not hasattr(e, 'code') or e.code != 401:
 772                         return "ERROR: failed for: %s" % str(e)
 773
 774                 base64data = base64.encodestring("%s:%s" % (self.username, self.password))[:-1]
 775                 # NOTE: assuming basic realm authentication.
 776                 authheader = "Basic %s" % base64data
 777                 req.add_header("Authorization", authheader)
 778
 779                 try:
 780                         f = urllib2.urlopen(req)
 781                 except IOError, e:
 782                         # failing here means the User/passwd is wrong (hopefully)
 783                         raise ExceptionPassword("Incorrect username/password")
 784
 785                 # TODO: after verifying that the user/password is correct, we should
 786                 # actually reboot the given node.
 787
 788                 if not dryrun:
 789                         # add data to handler,
 790                         # fetch url one more time on cmd.html, econtrol.html or whatever.
 791                         pass
 792
 793                 if self.verbose: print f.read()
 794
 795                 self.close()
 796                 return 0
 797
 798
 799 class ePowerSwitchOld(PCUControl):
 800         def run(self, node_port, dryrun):
 801                 self.url = "http://%s:%d/" % (self.host,80)
 802                 uri = "%s:%d" % (self.host,80)
 803
 804                 # create authinfo
 805                 authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 806                 authinfo.add_password (None, uri, self.username, self.password)
 807                 authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 808
 809                 # NOTE: it doesn't seem to matter whether this authinfo is here or not.
 810                 transport = urllib2.build_opener(authinfo)
 811                 f = transport.open(self.url)
 812                 if self.verbose: print f.read()
 813
 814                 if not dryrun:
 815                         transport = urllib2.build_opener(authhandler)
 816                         f = transport.open(self.url + "cmd.html", "P%d=r" % node_port)
 817                         if self.verbose: print f.read()
 818
 819                 self.close()
 820                 return 0
 821
 822 class ePowerSwitch(PCUControl):
 823         def run(self, node_port, dryrun):
 824                 self.url = "http://%s:%d/" % (self.host,80)
 825                 uri = "%s:%d" % (self.host,80)
 826
 827                 # TODO: I'm still not sure what the deal is here.
 828                 #               two independent calls appear to need to be made before the
 829                 #               reboot will succeed.  It doesn't seem to be possible to do
 830                 #               this with a single call.  I have no idea why.
 831
 832                 # create authinfo
 833                 authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 834                 authinfo.add_password (None, uri, self.username, self.password)
 835                 authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 836
 837                 # NOTE: it doesn't seem to matter whether this authinfo is here or not.
 838                 transport = urllib2.build_opener()
 839                 f = transport.open(self.url + "elogin.html", "pwd=%s" % self.password)
 840                 if self.verbose: print f.read()
 841
 842                 if not dryrun:
 843                         transport = urllib2.build_opener(authhandler)
 844                         f = transport.open(self.url + "econtrol.html", "P%d=r" % node_port)
 845                         if self.verbose: print f.read()
 846
 847                 #       data= "P%d=r" % node_port
 848                 #self.open(self.host, self.username, self.password)
 849                 #self.sendHTTP("elogin.html", "pwd=%s" % self.password)
 850                 #self.sendHTTP("econtrol.html", data)
 851                 #self.sendHTTP("cmd.html", data)
 852
 853                 self.close()
 854                 return 0
 855
 856
 857 ### rebooting european BlackBox PSE boxes
 858 # Thierry Parmentelat - May 11 2005
 859 # tested on 4-ports models known as PSE505-FR
 860 # uses http to POST a data 'P<port>=r'
 861 # relies on basic authentication within http1.0
 862 # first curl-based script was
 863 # curl --http1.0 --basic --user <username>:<password> --data P<port>=r \
 864 #       http://<hostname>:<http_port>/cmd.html && echo OK
 865
 866 def bbpse_reboot (pcu_ip,username,password,port_in_pcu,http_port, dryrun):
 867
 868         global verbose
 869
 870         url = "http://%s:%d/cmd.html" % (pcu_ip,http_port)
 871         data= "P%d=r" % port_in_pcu
 872         if verbose:
 873                 logger.debug("POSTing '%s' on %s" % (data,url))
 874
 875         authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 876         uri = "%s:%d" % (pcu_ip,http_port)
 877         authinfo.add_password (None, uri, username, password)
 878         authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 879
 880         opener = urllib2.build_opener(authhandler)
 881         urllib2.install_opener(opener)
 882
 883         if (dryrun):
 884                 return 0
 885
 886         try:
 887                 f = urllib2.urlopen(url,data)
 888
 889                 r= f.read()
 890                 if verbose:
 891                         logger.debug(r)
 892                 return 0
 893
 894         except urllib2.URLError,err:
 895                 logger.info('Could not open http connection', err)
 896                 return "bbpse error"
 897
 898 ### rebooting x10toggle based systems addressed by port
 899 # Marc E. Fiuczynski - May 31 2005
 900 # tested on 4-ports models known as PSE505-FR
 901 # uses ssh and password to login to an account
 902 # that will cause the system to be powercycled.
 903
 904 def x10toggle_reboot(ip, username, password, port, dryrun):
 905         global verbose
 906
 907         ssh = None
 908         try:
 909                 ssh = pyssh.Ssh(username, ip)
 910                 ssh.open()
 911
 912                 # Login
 913                 telnet_answer(ssh, "password:", password)
 914
 915                 if not dryrun:
 916                         # Reboot
 917                         telnet_answer(ssh, "x10toggle>", "A%d" % port)
 918
 919                 # Close
 920                 output = ssh.close()
 921                 if verbose:
 922                         logger.debug(output)
 923                 return 0
 924
 925         except Exception, err:
 926                 if verbose:
 927                         logger.debug(err)
 928                 if ssh:
 929                         output = ssh.close()
 930                         if verbose:
 931                                 logger.debug(output)
 932                 return errno.ETIMEDOUT
 933
 934 ### rebooting Dell systems via RAC card
 935 # Marc E. Fiuczynski - June 01 2005
 936 # tested with David Lowenthal's itchy/scratchy nodes at UGA
 937 #
 938
 939 def runcmd(command, args, username, password, timeout = None):
 940
 941         result = [None]
 942         result_ready = threading.Condition()
 943
 944         def set_result(x):
 945
 946                 result_ready.acquire()
 947                 try:
 948                         result[0] = x
 949                 finally:
 950                         result_ready.notify()
 951                         result_ready.release()
 952
 953         def do_command(command, username, password):
 954
 955                 try:
 956                         # Popen4 is a popen-type class that combines stdout and stderr
 957                         p = popen2.Popen4(command)
 958
 959                         # read all output data
 960                         p.tochild.write("%s\n" % username)
 961                         p.tochild.write("%s\n" % password)
 962                         p.tochild.close()
 963                         data = p.fromchild.read()
 964
 965                         while True:
 966                                 # might get interrupted by a signal in poll() or waitpid()
 967                                 try:
 968                                         retval = p.wait()
 969                                         set_result((retval, data))
 970                                         break
 971                                 except OSError, ex:
 972                                         if ex.errno == errno.EINTR:
 973                                                 continue
 974                                         raise ex
 975                 except Exception, ex:
 976                         set_result(ex)
 977
 978         if args:
 979                 command = " ".join([command] + args)
 980
 981         worker = threading.Thread(target = do_command, args = (command, username, password, ))
 982         worker.setDaemon(True)
 983         result_ready.acquire()
 984         worker.start()
 985         result_ready.wait(timeout)
 986         try:
 987                 if result == [None]:
 988                         raise Exception, "command timed-out: '%s'" % command
 989         finally:
 990                 result_ready.release()
 991         result = result[0]
 992
 993         if isinstance(result, Exception):
 994                 raise result
 995         else:
 996                 (retval, data) = result
 997                 if os.WIFEXITED(retval) and os.WEXITSTATUS(retval) == 0:
 998                         return data
 999                 else:
1000                         out = "system command ('%s') " % command
1001                         if os.WIFEXITED(retval):
1002                                 out += "failed, rc = %d" % os.WEXITSTATUS(retval)
1003                         else:
1004                                 out += "killed by signal %d" % os.WTERMSIG(retval)
1005                         if data:
1006                                 out += "; output follows:\n" + data
1007                         raise Exception, out
1008
1009 def racadm_reboot(host, username, password, port, dryrun):
1010         global verbose
1011
1012         ip = socket.gethostbyname(host)
1013         try:
1014                 cmd = "/usr/sbin/racadm"
1015                 os.stat(cmd)
1016                 if not dryrun:
1017                         output = runcmd(cmd, ["-r %s -i serveraction powercycle" % ip],
1018                                 username, password)
1019                 else:
1020                         output = runcmd(cmd, ["-r %s -i getsysinfo" % ip],
1021                                 username, password)
1022
1023                 print "RUNCMD: %s" % output
1024                 if verbose:
1025                         logger.debug(output)
1026                 return 0
1027
1028         except Exception, err:
1029                 logger.debug("runcmd raised exception %s" % err)
1030                 if verbose:
1031                         logger.debug(err)
1032                 return -1
1033
1034 def pcu_name(pcu):
1035         if pcu['hostname'] is not None and pcu['hostname'] is not "":
1036                 return pcu['hostname']
1037         elif pcu['ip'] is not None and pcu['ip'] is not "":
1038                 return pcu['ip']
1039         else:
1040                 return None
1041
1042 def get_pcu_values(pcu_id):
1043         # TODO: obviously, this shouldn't be loaded each time...
1044         import soltesz
1045         fb =soltesz.dbLoad("findbadpcus")
1046
1047         try:
1048                 values = fb['nodes']["id_%s" % pcu_id]['values']
1049         except:
1050                 values = None
1051
1052         return values
1053
1054 def check_open_port(values, port_list):
1055         ret = False
1056
1057         if 'portstatus' in values:
1058                 for port in port_list:
1059                         if      port in values['portstatus'] and \
1060                                 values['portstatus'][port] == "open":
1061
1062                                 ret = True
1063
1064         return ret
1065
1066 def reboot_policy(nodename, continue_probe, dryrun):
1067         global verbose
1068
1069         pcu = plc.getpcu(nodename)
1070         if not pcu:
1071                 return False # "%s has no pcu" % nodename
1072
1073         values = get_pcu_values(pcu['pcu_id'])
1074         if values == None:
1075                 return False #"no info for pcu_id %s" % pcu['pcu_id']
1076
1077         # Try the PCU first
1078         logger.debug("Trying PCU %s %s" % (pcu['hostname'], pcu['model']))
1079
1080         ret = reboot_test(nodename, values, continue_probe, verbose, dryrun)
1081
1082         if ret != 0:
1083                 print ret
1084                 return False
1085         else:
1086                 return True
1087
1088 def reboot_test(nodename, values, continue_probe, verbose, dryrun):
1089         rb_ret = ""
1090
1091         try:
1092                 # DataProbe iPal (many sites)
1093                 if  continue_probe and values['model'].find("Dataprobe IP-41x/IP-81x") >= 0:
1094                         ipal = IPAL(values, verbose, ['23'])
1095                         rb_ret = ipal.reboot(values[nodename], dryrun)
1096
1097                 # APC Masterswitch (Berkeley)
1098                 elif continue_probe and values['model'].find("APC AP79xx/Masterswitch") >= 0:
1099
1100                         # TODO: make a more robust version of APC
1101                         if values['pcu_id'] in [1163,1055,1111,1231,1113,1127,1128,1148]:
1102                                 apc = APCEurope(values, verbose, ['22', '23'])
1103                                 rb_ret = apc.reboot(values[nodename], dryrun)
1104
1105                         elif values['pcu_id'] in [1110,86]:
1106                                 apc = APCBrazil(values, verbose, ['22', '23'])
1107                                 rb_ret = apc.reboot(values[nodename], dryrun)
1108
1109                         elif values['pcu_id'] in [1221,1225]:
1110                                 apc = APCBerlin(values, verbose, ['22', '23'])
1111                                 rb_ret = apc.reboot(values[nodename], dryrun)
1112
1113                         elif values['pcu_id'] in [1173,1221,1220]:
1114                                 apc = APCFolsom(values, verbose, ['22', '23'])
1115                                 rb_ret = apc.reboot(values[nodename], dryrun)
1116
1117                         else:
1118                                 apc = APCMaster(values, verbose, ['22', '23'])
1119                                 rb_ret = apc.reboot(values[nodename], dryrun)
1120
1121                 # BayTech DS4-RPC
1122                 elif continue_probe and values['model'].find("Baytech DS4-RPC") >= 0:
1123                         if values['pcu_id'] in [1237,1052,1209,1002,1008,1041,1013,1022]:
1124                                 # These  require a 'ctrl-c' to be sent...
1125                                 baytech = BayTechCtrlC(values, verbose, ['22', '23'])
1126                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1127
1128                         elif values['pcu_id'] in [93]:
1129                                 baytech = BayTechAU(values, verbose, ['22', '23'])
1130                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1131
1132                         elif values['pcu_id'] in [1057]:
1133                                 # These  require a 'ctrl-c' to be sent...
1134                                 baytech = BayTechCtrlCUnibe(values, verbose, ['22', '23'])
1135                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1136
1137                         elif values['pcu_id'] in [1012]:
1138                                 # This pcu sometimes doesn't present the 'Username' prompt,
1139                                 # unless you immediately try again...
1140                                 try:
1141                                         baytech = BayTechGeorgeTown(values, verbose, ['22', '23'])
1142                                         rb_ret = baytech.reboot(values[nodename], dryrun)
1143                                 except:
1144                                         baytech = BayTechGeorgeTown(values, verbose, ['22', '23'])
1145                                         rb_ret = baytech.reboot(values[nodename], dryrun)
1146                         else:
1147                                 baytech = BayTech(values, verbose, ['22', '23'])
1148                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1149
1150                 # iLO
1151                 elif continue_probe and values['model'].find("HP iLO") >= 0:
1152                         try:
1153                                 hpilo = HPiLO(values, verbose, ['22'])
1154                                 rb_ret = hpilo.reboot(0, dryrun)
1155                                 if rb_ret != 0:
1156                                         hpilo = HPiLOHttps(values, verbose, ['443'])
1157                                         rb_ret = hpilo.reboot(0, dryrun)
1158                         except:
1159                                 hpilo = HPiLOHttps(values, verbose, ['443'])
1160                                 rb_ret = hpilo.reboot(0, dryrun)
1161
1162                 # DRAC ssh
1163                 elif continue_probe and values['model'].find("Dell RAC") >= 0:
1164                         # TODO: I don't think DRACRacAdm will throw an exception for the
1165                         # default method to catch...
1166                         try:
1167                                 drac = DRACRacAdm(values, verbose, ['443', '5869'])
1168                                 rb_ret = drac.reboot(0, dryrun)
1169                         except:
1170                                 drac = DRAC(values, verbose, ['22'])
1171                                 rb_ret = drac.reboot(0, dryrun)
1172
1173                 elif continue_probe and values['model'].find("WTI IPS-4") >= 0:
1174                                 wti = WTIIPS4(values, verbose, ['23'])
1175                                 rb_ret = wti.reboot(values[nodename], dryrun)
1176
1177                 # BlackBox PSExxx-xx (e.g. PSE505-FR)
1178                 elif continue_probe and \
1179                         (values['model'].find("BlackBox PS5xx") >= 0 or
1180                          values['model'].find("ePowerSwitch 1/4/8x") >=0 ):
1181
1182                         # TODO: allow a different port than http 80.
1183                         if values['pcu_id'] in [1089, 1071, 1046, 1035, 1118]:
1184                                 eps = ePowerSwitchGood(values, verbose, ['80'])
1185                         elif values['pcu_id'] in [1003]:
1186                                 eps = ePowerSwitch(values, verbose, ['80'])
1187                         else:
1188                                 eps = ePowerSwitchGood(values, verbose, ['80'])
1189
1190                         rb_ret = eps.reboot(values[nodename], dryrun)
1191
1192                 elif continue_probe:
1193                         rb_ret = "Unsupported_PCU"
1194
1195                 elif continue_probe == False:
1196                         if 'portstatus' in values:
1197                                 rb_ret = "NetDown"
1198                         else:
1199                                 rb_ret = "Not_Run"
1200                 else:
1201                         rb_ret = -1
1202
1203         except ExceptionPort, err:
1204                 rb_ret = str(err)
1205
1206         return rb_ret
1207         # ????
1208         #elif continue_probe and values['protocol'] == "racadm" and \
1209         #               values['model'] == "RAC":
1210         #       rb_ret = racadm_reboot(pcu_name(values),
1211         #                                                                 values['username'],
1212         #                                                                 values['password'],
1213         #                                                                 pcu[nodename],
1214         #                                                                 dryrun)
1215
1216 def main():
1217         logger.setLevel(logging.DEBUG)
1218         ch = logging.StreamHandler()
1219         ch.setLevel(logging.DEBUG)
1220         formatter = logging.Formatter('LOGGER - %(message)s')
1221         ch.setFormatter(formatter)
1222         logger.addHandler(ch)
1223
1224         try:
1225                 for node in sys.argv[1:]:
1226                         print "Rebooting %s" % node
1227                         if reboot_policy(node, True, False):
1228                                 print "success"
1229                         else:
1230                                 print "failed"
1231         except Exception, err:
1232                 print err
1233
1234 if __name__ == '__main__':
1235         import plc
1236         logger = logging.getLogger("monitor")
1237         main()