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
  19 plc_lock = threading.Lock()
  20
  21 # Use our versions of telnetlib and pyssh
  22 sys.path.insert(0, os.path.dirname(sys.argv[0]))
  23 import telnetlib
  24 sys.path.insert(0, os.path.dirname(sys.argv[0]) + "/pyssh")
  25 import pyssh
  26
  27 # Timeouts in seconds
  28 TELNET_TIMEOUT = 30
  29
  30 # Event class ID from pcu events
  31 #NODE_POWER_CONTROL = 3
  32
  33 # Monitor user ID
  34 #MONITOR_USER_ID = 11142
  35
  36 import logging
  37 logger = logging.getLogger("monitor")
  38 verbose = 1
  39 #dryrun = 0;
  40
  41 class ExceptionNoTransport(Exception): pass
  42 class ExceptionNotFound(Exception): pass
  43 class ExceptionPassword(Exception): pass
  44 class ExceptionTimeout(Exception): pass
  45 class ExceptionPrompt(Exception): pass
  46 class ExceptionSequence(Exception): pass
  47 class ExceptionReset(Exception): pass
  48 class ExceptionPort(Exception): pass
  49 class ExceptionUsername(Exception): pass
  50
  51 def telnet_answer(telnet, expected, buffer):
  52         global verbose
  53
  54         output = telnet.read_until(expected, TELNET_TIMEOUT)
  55         #if verbose:
  56         #       logger.debug(output)
  57         if output.find(expected) == -1:
  58                 raise ExceptionNotFound, "'%s' not found" % expected
  59         else:
  60                 telnet.write(buffer + "\r\n")
  61
  62
  63 # PCU has model, host, preferred-port, user, passwd,
  64
  65 # This is an object derived directly form the PLCAPI DB fields
  66 class PCU(object):
  67         def __init__(self, plc_pcu_dict):
  68                 for field in ['username', 'password', 'site_id',
  69                                                 'hostname', 'ip',
  70                                                 'pcu_id', 'model',
  71                                                 'node_ids', 'ports', ]:
  72                         if field in plc_pcu_dict:
  73                                 self.__setattr__(field, plc_pcu_dict[field])
  74                         else:
  75                                 raise Exception("No such field %s in PCU object" % field)
  76
  77 # These are the convenience functions build around the PCU object.
  78 class PCUModel(PCU):
  79         def __init__(self, plc_pcu_dict):
  80                 PCU.__init__(self, plc_pcu_dict)
  81                 self.host = self.pcu_name()
  82
  83         def pcu_name(self):
  84                 if self.hostname is not None and self.hostname is not "":
  85                         return self.hostname
  86                 elif self.ip is not None and self.ip is not "":
  87                         return self.ip
  88                 else:
  89                         return None
  90
  91         def nodeidToPort(self, node_id):
  92                 if node_id in self.node_ids:
  93                         for i in range(0, len(self.node_ids)):
  94                                 if node_id == self.node_ids[i]:
  95                                         return self.ports[i]
  96
  97                 raise Exception("No such Node ID: %d" % node_id)
  98
  99 # This class captures the observed pcu records from FindBadPCUs.py
 100 class PCURecord:
 101         def __init__(self, pcu_record_dict):
 102                 for field in ['nodenames', 'portstatus',
 103                                                 'dnsmatch',
 104                                                 'complete_entry', ]:
 105                         if field in pcu_record_dict:
 106                                 if field == "reboot":
 107                                         self.__setattr__("reboot_str", pcu_record_dict[field])
 108                                 else:
 109                                         self.__setattr__(field, pcu_record_dict[field])
 110                         else:
 111                                 raise Exception("No such field %s in pcu record dict" % field)
 112
 113 class Transport:
 114         TELNET = 1
 115         SSH    = 2
 116         HTTP   = 3
 117
 118         TELNET_TIMEOUT = 60
 119
 120         def __init__(self, type, verbose):
 121                 self.type = type
 122                 self.verbose = verbose
 123                 self.transport = None
 124
 125 #       def __del__(self):
 126 #               if self.transport:
 127 #                       self.close()
 128
 129         def open(self, host, username=None, password=None, prompt="User Name"):
 130                 transport = None
 131
 132                 if self.type == self.TELNET:
 133                         transport = telnetlib.Telnet(host, timeout=self.TELNET_TIMEOUT)
 134                         transport.set_debuglevel(self.verbose)
 135                         if username is not None:
 136                                 self.transport = transport
 137                                 self.ifThenSend(prompt, username, ExceptionUsername)
 138
 139                 elif self.type == self.SSH:
 140                         if username is not None:
 141                                 transport = pyssh.Ssh(username, host)
 142                                 transport.set_debuglevel(self.verbose)
 143                                 transport.open()
 144                                 # TODO: have an ssh set_debuglevel() also...
 145                         else:
 146                                 raise Exception("Username cannot be None for ssh transport.")
 147                 elif self.type == self.HTTP:
 148                         self.url = "http://%s:%d/" % (host,80)
 149                         uri = "%s:%d" % (host,80)
 150
 151                         # create authinfo
 152                         authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 153                         authinfo.add_password (None, uri, username, password)
 154                         authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 155
 156                         transport = urllib2.build_opener(authhandler)
 157
 158                 else:
 159                         raise Exception("Unknown transport type: %s" % self.type)
 160
 161                 self.transport = transport
 162                 return True
 163
 164         def close(self):
 165                 if self.type == self.TELNET:
 166                         self.transport.close()
 167                 elif self.type == self.SSH:
 168                         self.transport.close()
 169                 elif self.type == self.HTTP:
 170                         pass
 171                 else:
 172                         raise Exception("Unknown transport type %s" % self.type)
 173                 self.transport = None
 174
 175         def sendHTTP(self, resource, data):
 176                 if self.verbose:
 177                         print "POSTing '%s' to %s" % (data,self.url + resource)
 178
 179                 try:
 180                         f = self.transport.open(self.url + resource ,data)
 181                         r = f.read()
 182                         if self.verbose:
 183                                 print r
 184
 185                 except urllib2.URLError,err:
 186                         logger.info('Could not open http connection', err)
 187                         return "http transport error"
 188
 189                 return 0
 190
 191         def sendPassword(self, password, prompt=None):
 192                 if self.type == self.TELNET:
 193                         if prompt == None:
 194                                 self.ifThenSend("Password", password, ExceptionPassword)
 195                         else:
 196                                 self.ifThenSend(prompt, password, ExceptionPassword)
 197                 elif self.type == self.SSH:
 198                         self.ifThenSend("password:", password, ExceptionPassword)
 199                 elif self.type == self.HTTP:
 200                         pass
 201                 else:
 202                         raise Exception("Unknown transport type: %s" % self.type)
 203
 204         def ifThenSend(self, expected, buffer, ErrorClass=ExceptionPrompt):
 205
 206                 if self.transport != None:
 207                         output = self.transport.read_until(expected, self.TELNET_TIMEOUT)
 208                         if output.find(expected) == -1:
 209                                 raise ErrorClass, "'%s' not found" % expected
 210                         else:
 211                                 self.transport.write(buffer + "\r\n")
 212                 else:
 213                         raise ExceptionNoTransport("transport object is type None")
 214
 215         def ifElse(self, expected, ErrorClass):
 216                 try:
 217                         self.transport.read_until(expected, self.TELNET_TIMEOUT)
 218                 except:
 219                         raise ErrorClass("Could not find '%s' within timeout" % expected)
 220
 221
 222 class PCUControl(Transport,PCUModel,PCURecord):
 223         def __init__(self, plc_pcu_record, verbose, supported_ports=[]):
 224                 PCUModel.__init__(self, plc_pcu_record)
 225                 PCURecord.__init__(self, plc_pcu_record)
 226                 if self.portstatus:
 227                         if '22' in supported_ports and self.portstatus['22'] == "open":
 228                                 type = Transport.SSH
 229                         elif '23' in supported_ports and self.portstatus['23'] == "open":
 230                                 type = Transport.TELNET
 231                         elif '80' in supported_ports and self.portstatus['80'] == "open":
 232                                 type = Transport.HTTP
 233                         elif '443' in supported_ports and self.portstatus['443'] == "open":
 234                                 type = Transport.HTTP
 235                         elif '5869' in supported_ports and self.portstatus['5869'] == "open":
 236                                 # For DRAC cards.  not sure how much it's used in the
 237                                 # protocol.. but racadm opens this port.
 238                                 type = Transport.HTTP
 239                         else:
 240                                 raise ExceptionPort("Unsupported Port: No transport from open ports")
 241                 Transport.__init__(self, type, verbose)
 242
 243         def run(self, node_port, dryrun):
 244                 """ This function is to be defined by the specific PCU instance.  """
 245                 pass
 246
 247         def reboot(self, node_port, dryrun):
 248                 try:
 249                         return self.run(node_port, dryrun)
 250                 except ExceptionNotFound, err:
 251                         return "error: " + str(err)
 252                 except ExceptionPassword, err:
 253                         return "password exception: " + str(err)
 254                 except ExceptionTimeout, err:
 255                         return "timeout exception: " + str(err)
 256                 except ExceptionUsername, err:
 257                         return "exception: no username prompt: " + str(err)
 258                 except ExceptionSequence, err:
 259                         return "sequence error: " + str(err)
 260                 except ExceptionPrompt, err:
 261                         return "prompt exception: " + str(err)
 262                 except ExceptionPort, err:
 263                         return "no ports exception: " + str(err)
 264                 except socket.error, err:
 265                         return "socket error: timeout: " + str(err)
 266                 except EOFError, err:
 267                         if self.verbose:
 268                                 logger.debug("reboot: EOF")
 269                                 logger.debug(err)
 270                         self.transport.close()
 271                         import traceback
 272                         traceback.print_exc()
 273                         return "EOF connection reset" + str(err)
 274                 #except Exception, err:
 275                 #       if self.verbose:
 276                 #               logger.debug("reboot: Exception")
 277                 #               logger.debug(err)
 278                 #       if self.transport:
 279                 #               self.transport.close()
 280                 #       import traceback
 281                 #       traceback.print_exc()
 282                 #       return  "generic exception; unknown problem."
 283
 284
 285 class IPAL(PCUControl):
 286         def run(self, node_port, dryrun):
 287                 self.open(self.host)
 288
 289                 # XXX Some iPals require you to hit Enter a few times first
 290                 self.ifThenSend("Password >", "\r\n\r\n", ExceptionNotFound)
 291
 292                 # Login
 293                 self.ifThenSend("Password >", self.password, ExceptionPassword)
 294                 self.transport.write("\r\n\r\n")
 295
 296                 if not dryrun: # P# - Pulse relay
 297                         self.ifThenSend("Enter >",
 298                                                         "P%d" % node_port,
 299                                                         ExceptionNotFound)
 300                 # Get the next prompt
 301                 self.ifElse("Enter >", ExceptionTimeout)
 302
 303                 self.close()
 304                 return 0
 305
 306 def ipal_reboot(ip, password, port, dryrun):
 307         global verbose
 308         global plc_lock
 309         telnet = None
 310
 311         try:
 312                 #plc_lock.acquire()
 313                 #print "lock acquired"
 314
 315                 #try:
 316                         #telnet = telnetlib.Telnet(ip) # , timeout=TELNET_TIMEOUT)
 317                 telnet = telnetlib.Telnet(ip, timeout=TELNET_TIMEOUT)
 318                 #except:
 319                 #       import traceback
 320                 #       traceback.print_exc()
 321
 322
 323                 telnet.set_debuglevel(verbose)
 324
 325                 # XXX Some iPals require you to hit Enter a few times first
 326                 telnet_answer(telnet, "Password >", "\r\n\r\n")
 327
 328                 # Login
 329                 telnet_answer(telnet, "Password >", password)
 330
 331                 # XXX Some iPals require you to hit Enter a few times first
 332                 telnet.write("\r\n\r\n")
 333
 334                 # P# - Pulse relay
 335                 if not dryrun:
 336                         telnet_answer(telnet, "Enter >", "P%d" % port)
 337
 338                 telnet.read_until("Enter >", TELNET_TIMEOUT)
 339
 340                 # Close
 341                 telnet.close()
 342
 343                 #print "lock released"
 344                 #plc_lock.release()
 345                 return 0
 346
 347         except EOFError, err:
 348                 if verbose:
 349                         logger.debug("ipal_reboot: EOF")
 350                         logger.debug(err)
 351                 telnet.close()
 352                 import traceback
 353                 traceback.print_exc()
 354                 #print "lock released"
 355                 #plc_lock.release()
 356                 return errno.ECONNRESET
 357         except socket.error, err:
 358                 logger.debug("ipal_reboot: Socket Error")
 359                 logger.debug(err)
 360                 import traceback
 361                 traceback.print_exc()
 362
 363                 return errno.ETIMEDOUT
 364
 365         except Exception, err:
 366                 if verbose:
 367                         logger.debug("ipal_reboot: Exception")
 368                         logger.debug(err)
 369                 if telnet:
 370                         telnet.close()
 371                 import traceback
 372                 traceback.print_exc()
 373                 #print "lock released"
 374                 #plc_lock.release()
 375                 return  "ipal error"
 376
 377 class APCEurope(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> ", str(node_port))
 385                 # 3- Immediate Reboot
 386                 self.ifThenSend("\r\n> ", "3")
 387
 388                 if not dryrun:
 389                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 390                                                         "YES\r\n",
 391                                                         ExceptionSequence)
 392                 else:
 393                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 394                                                         "", ExceptionSequence)
 395                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 396
 397                 self.close()
 398                 return 0
 399
 400 class APCFolsom(PCUControl):
 401         def run(self, node_port, dryrun):
 402                 self.open(self.host, self.username)
 403                 self.sendPassword(self.password)
 404
 405                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 406                 self.ifThenSend("\r\n> ", "2")
 407                 self.ifThenSend("\r\n> ", "1")
 408                 self.ifThenSend("\r\n> ", str(node_port))
 409                 self.ifThenSend("\r\n> ", "1")
 410
 411                 # 3- Immediate Reboot
 412                 self.ifThenSend("\r\n> ", "3")
 413
 414                 if not dryrun:
 415                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 416                                                         "YES\r\n",
 417                                                         ExceptionSequence)
 418                 else:
 419                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 420                                                         "", ExceptionSequence)
 421                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 422
 423                 self.close()
 424                 return 0
 425
 426 class APCMaster(PCUControl):
 427         def run(self, node_port, dryrun):
 428                 self.open(self.host, self.username)
 429                 self.sendPassword(self.password)
 430
 431                 # 1- Device Manager
 432                 self.ifThenSend("\r\n> ", "1", ExceptionPassword)
 433                 # 3- Outlet Control/Config
 434                 self.ifThenSend("\r\n> ", "3")
 435                 # n- Outlet n
 436                 self.ifThenSend("\r\n> ", str(node_port))
 437                 # 1- Control Outlet
 438                 self.ifThenSend("\r\n> ", "1")
 439                 # 3- Immediate Reboot
 440                 self.ifThenSend("\r\n> ", "3")
 441
 442                 if not dryrun:
 443                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 444                                                         "YES\r\n",
 445                                                         ExceptionSequence)
 446                 else:
 447                         self.ifThenSend("Enter 'YES' to continue or <ENTER> to cancel",
 448                                                         "", ExceptionSequence)
 449                 self.ifThenSend("Press <ENTER> to continue...", "", ExceptionSequence)
 450
 451                 self.close()
 452                 return 0
 453
 454 class APC(PCUControl):
 455         def __init__(self, plc_pcu_record, verbose):
 456                 PCUControl.__init__(self, plc_pcu_record, verbose)
 457
 458                 self.master = APCMaster(plc_pcu_record, verbose)
 459                 self.folsom = APCFolsom(plc_pcu_record, verbose)
 460                 self.europe = APCEurope(plc_pcu_record, verbose)
 461
 462         def run(self, node_port, dryrun):
 463                 try_again = True
 464                 sleep_time = 1
 465
 466                 for pcu in [self.master, self.europe, self.folsom]:
 467                         if try_again:
 468                                 try:
 469                                         print "-*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*_*"
 470                                         try_again = False
 471                                         print "sleeping 5"
 472                                         time.sleep(sleep_time)
 473                                         ret = pcu.reboot(node_port, dryrun)
 474                                 except ExceptionSequence, err:
 475                                         del pcu
 476                                         sleep_time = 130
 477                                         try_again = True
 478
 479                 if try_again:
 480                         return "Unknown reboot sequence for APC PCU"
 481                 else:
 482                         return ret
 483
 484 class DRACRacAdm(PCUControl):
 485         def run(self, node_port, dryrun):
 486
 487                 print "trying racadm_reboot..."
 488                 racadm_reboot(self.host, self.username, self.password, node_port, dryrun)
 489
 490                 return 0
 491
 492 class DRAC(PCUControl):
 493         def run(self, node_port, dryrun):
 494                 self.open(self.host, self.username)
 495                 self.sendPassword(self.password)
 496
 497                 print "logging in..."
 498                 self.transport.write("\r\n")
 499                 # Testing Reboot ?
 500                 if dryrun:
 501                         self.ifThenSend("[%s]#" % self.username, "getsysinfo")
 502                 else:
 503                         # Reset this machine
 504                         self.ifThenSend("[%s]#" % self.username, "serveraction powercycle")
 505
 506                 self.ifThenSend("[%s]#" % self.username, "exit")
 507
 508                 self.close()
 509                 return 0
 510
 511 class HPiLO(PCUControl):
 512         def run(self, node_port, dryrun):
 513                 self.open(self.host, self.username)
 514                 self.sendPassword(self.password)
 515
 516                 # </>hpiLO->
 517                 self.ifThenSend("</>hpiLO->", "cd system1")
 518
 519                 # Reboot Outlet  N        (Y/N)?
 520                 if dryrun:
 521                         self.ifThenSend("</system1>hpiLO->", "POWER")
 522                 else:
 523                         # Reset this machine
 524                         self.ifThenSend("</system1>hpiLO->", "reset")
 525
 526                 self.ifThenSend("</system1>hpiLO->", "exit")
 527
 528                 self.close()
 529                 return 0
 530
 531
 532 class HPiLOHttps(PCUControl):
 533         def run(self, node_port, dryrun):
 534
 535                 cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p %s" % (
 536                                         self.host, "iloxml/Get_Network.xml",
 537                                         self.username, self.password)
 538                 p_ilo  = Popen(cmd, stdout=PIPE, shell=True)
 539                 cmd2 = "/bin/grep 'MESSAGE' | /bin/grep -v 'No error'"
 540                 p_grep = Popen(cmd2, stdin=p_ilo.stdout, stdout=PIPE, stderr=PIPE, shell=True)
 541                 sout, serr = p_grep.communicate()
 542
 543                 p_ilo.wait()
 544                 p_grep.wait()
 545                 if sout.strip() != "":
 546                         print "sout: %s" % sout.strip()
 547                         return sout.strip()
 548
 549                 if not dryrun:
 550                         cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p %s" % (
 551                                         self.host, "iloxml/Reset_Server.xml",
 552                                         self.username, self.password)
 553                         print cmd
 554                         p_ilo = Popen(cmd, stdin=PIPE, stdout=PIPE, shell=True)
 555                         cmd2 = "/bin/grep 'MESSAGE' | /bin/grep -v 'No error'"
 556                         print cmd2
 557                         p_grep = Popen(cmd2, stdin=p_ilo.stdout, stdout=PIPE, stderr=PIPE, shell=True)
 558                         sout, serr = p_grep.communicate()
 559                         p_ilo.wait()
 560                         p_grep.wait()
 561
 562                         if sout.strip() != "":
 563                                 print "sout: %s" % sout.strip()
 564                                 return sout.strip()
 565
 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 BayTechCtrlC(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                 self.open(self.host, self.username)
 596                 self.sendPassword(self.password)
 597
 598                 #self.transport.write('\ 3')
 599                 self.transport.write("\r\n")
 600                 self.transport.write(pyssh.CTRL_C)
 601                 #self.transport.write(chr(3))
 602                 #self.transport.write(chr(24))
 603                 #self.transport.write(chr(26))
 604                 #self.transport.write('\18')
 605                 # Control Outlets  (5 ,1).........5
 606                 self.ifThenSend("Enter Request :", "5")
 607
 608                 # Reboot N
 609                 try:
 610                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port)
 611                 except ExceptionNotFound, msg:
 612                         # one machine is configured to ask for a username,
 613                         # even after login...
 614                         print "msg: %s" % msg
 615                         self.transport.write(self.username + "\r\n")
 616                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port)
 617
 618
 619                 # Reboot Outlet  N        (Y/N)?
 620                 if dryrun:
 621                         self.ifThenSend("(Y/N)?", "N")
 622                 else:
 623                         self.ifThenSend("(Y/N)?", "Y")
 624                 self.ifThenSend("DS-RPC>", "")
 625
 626                 self.close()
 627                 return 0
 628
 629 class BayTech(PCUControl):
 630         def run(self, node_port, dryrun):
 631                 self.open(self.host, self.username)
 632                 self.sendPassword(self.password)
 633
 634                 # Control Outlets  (5 ,1).........5
 635                 self.ifThenSend("Enter Request :", "5")
 636
 637                 # Reboot N
 638                 try:
 639                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port)
 640                 except ExceptionNotFound, msg:
 641                         # one machine is configured to ask for a username,
 642                         # even after login...
 643                         print "msg: %s" % msg
 644                         self.transport.write(self.username + "\r\n")
 645                         self.ifThenSend("DS-RPC>", "Reboot %d" % node_port)
 646
 647
 648                 # Reboot Outlet  N        (Y/N)?
 649                 if dryrun:
 650                         self.ifThenSend("(Y/N)?", "N")
 651                 else:
 652                         self.ifThenSend("(Y/N)?", "Y")
 653                 self.ifThenSend("DS-RPC>", "")
 654
 655                 self.close()
 656                 return 0
 657
 658 class ePowerSwitchGood(PCUControl):
 659         # NOTE:
 660         #               The old code used Python's HTTPPasswordMgrWithDefaultRealm()
 661         #               For some reason this both doesn't work and in some cases, actually
 662         #               hangs the PCU.  Definitely not what we want.
 663         #
 664         #               The code below is much simpler.  Just letting things fail first,
 665         #               and then, trying again with authentication string in the header.
 666         #
 667         def run(self, node_port, dryrun):
 668                 self.transport = None
 669                 self.url = "http://%s:%d/" % (self.host,80)
 670                 uri = "%s:%d" % (self.host,80)
 671
 672                 req = urllib2.Request(self.url)
 673                 try:
 674                         handle = urllib2.urlopen(req)
 675                 except IOError, e:
 676                         # NOTE: this is expected to fail initially
 677                         pass
 678                 else:
 679                         print self.url
 680                         print "-----------"
 681                         print handle.read()
 682                         print "-----------"
 683                         return "ERROR: not protected by HTTP authentication"
 684
 685                 if not hasattr(e, 'code') or e.code != 401:
 686                         return "ERROR: failed for: %s" % str(e)
 687
 688                 base64data = base64.encodestring("%s:%s" % (self.username, self.password))[:-1]
 689                 # NOTE: assuming basic realm authentication.
 690                 authheader = "Basic %s" % base64data
 691                 req.add_header("Authorization", authheader)
 692
 693                 try:
 694                         f = urllib2.urlopen(req)
 695                 except IOError, e:
 696                         # failing here means the User/passwd is wrong (hopefully)
 697                         raise ExceptionPassword("Incorrect username/password")
 698
 699                 # TODO: after verifying that the user/password is correct, we should
 700                 # actually reboot the given node.
 701
 702                 if not dryrun:
 703                         # add data to handler,
 704                         # fetch url one more time on cmd.html, econtrol.html or whatever.
 705                         pass
 706
 707                 if self.verbose: print f.read()
 708
 709                 self.close()
 710                 return 0
 711
 712
 713 class ePowerSwitchOld(PCUControl):
 714         def run(self, node_port, dryrun):
 715                 self.url = "http://%s:%d/" % (self.host,80)
 716                 uri = "%s:%d" % (self.host,80)
 717
 718                 # create authinfo
 719                 authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 720                 authinfo.add_password (None, uri, self.username, self.password)
 721                 authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 722
 723                 # NOTE: it doesn't seem to matter whether this authinfo is here or not.
 724                 transport = urllib2.build_opener(authinfo)
 725                 f = transport.open(self.url)
 726                 if self.verbose: print f.read()
 727
 728                 if not dryrun:
 729                         transport = urllib2.build_opener(authhandler)
 730                         f = transport.open(self.url + "cmd.html", "P%d=r" % node_port)
 731                         if self.verbose: print f.read()
 732
 733                 self.close()
 734                 return 0
 735
 736 class ePowerSwitch(PCUControl):
 737         def run(self, node_port, dryrun):
 738                 self.url = "http://%s:%d/" % (self.host,80)
 739                 uri = "%s:%d" % (self.host,80)
 740
 741                 # TODO: I'm still not sure what the deal is here.
 742                 #               two independent calls appear to need to be made before the
 743                 #               reboot will succeed.  It doesn't seem to be possible to do
 744                 #               this with a single call.  I have no idea why.
 745
 746                 # create authinfo
 747                 authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 748                 authinfo.add_password (None, uri, self.username, self.password)
 749                 authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 750
 751                 # NOTE: it doesn't seem to matter whether this authinfo is here or not.
 752                 transport = urllib2.build_opener()
 753                 f = transport.open(self.url + "elogin.html", "pwd=%s" % self.password)
 754                 if self.verbose: print f.read()
 755
 756                 if not dryrun:
 757                         transport = urllib2.build_opener(authhandler)
 758                         f = transport.open(self.url + "econtrol.html", "P%d=r" % node_port)
 759                         if self.verbose: print f.read()
 760
 761                 #       data= "P%d=r" % node_port
 762                 #self.open(self.host, self.username, self.password)
 763                 #self.sendHTTP("elogin.html", "pwd=%s" % self.password)
 764                 #self.sendHTTP("econtrol.html", data)
 765                 #self.sendHTTP("cmd.html", data)
 766
 767                 self.close()
 768                 return 0
 769
 770
 771 ### rebooting european BlackBox PSE boxes
 772 # Thierry Parmentelat - May 11 2005
 773 # tested on 4-ports models known as PSE505-FR
 774 # uses http to POST a data 'P<port>=r'
 775 # relies on basic authentication within http1.0
 776 # first curl-based script was
 777 # curl --http1.0 --basic --user <username>:<password> --data P<port>=r \
 778 #       http://<hostname>:<http_port>/cmd.html && echo OK
 779
 780 def bbpse_reboot (pcu_ip,username,password,port_in_pcu,http_port, dryrun):
 781
 782         global verbose
 783
 784         url = "http://%s:%d/cmd.html" % (pcu_ip,http_port)
 785         data= "P%d=r" % port_in_pcu
 786         if verbose:
 787                 logger.debug("POSTing '%s' on %s" % (data,url))
 788
 789         authinfo = urllib2.HTTPPasswordMgrWithDefaultRealm()
 790         uri = "%s:%d" % (pcu_ip,http_port)
 791         authinfo.add_password (None, uri, username, password)
 792         authhandler = urllib2.HTTPBasicAuthHandler( authinfo )
 793
 794         opener = urllib2.build_opener(authhandler)
 795         urllib2.install_opener(opener)
 796
 797         if (dryrun):
 798                 return 0
 799
 800         try:
 801                 f = urllib2.urlopen(url,data)
 802
 803                 r= f.read()
 804                 if verbose:
 805                         logger.debug(r)
 806                 return 0
 807
 808         except urllib2.URLError,err:
 809                 logger.info('Could not open http connection', err)
 810                 return "bbpse error"
 811
 812 ### rebooting x10toggle based systems addressed by port
 813 # Marc E. Fiuczynski - May 31 2005
 814 # tested on 4-ports models known as PSE505-FR
 815 # uses ssh and password to login to an account
 816 # that will cause the system to be powercycled.
 817
 818 def x10toggle_reboot(ip, username, password, port, dryrun):
 819         global verbose
 820
 821         ssh = None
 822         try:
 823                 ssh = pyssh.Ssh(username, ip)
 824                 ssh.open()
 825
 826                 # Login
 827                 telnet_answer(ssh, "password:", password)
 828
 829                 if not dryrun:
 830                         # Reboot
 831                         telnet_answer(ssh, "x10toggle>", "A%d" % port)
 832
 833                 # Close
 834                 output = ssh.close()
 835                 if verbose:
 836                         logger.debug(output)
 837                 return 0
 838
 839         except Exception, err:
 840                 if verbose:
 841                         logger.debug(err)
 842                 if ssh:
 843                         output = ssh.close()
 844                         if verbose:
 845                                 logger.debug(output)
 846                 return errno.ETIMEDOUT
 847
 848 ### rebooting Dell systems via RAC card
 849 # Marc E. Fiuczynski - June 01 2005
 850 # tested with David Lowenthal's itchy/scratchy nodes at UGA
 851 #
 852
 853 def runcmd(command, args, username, password, timeout = None):
 854
 855         result = [None]
 856         result_ready = threading.Condition()
 857
 858         def set_result(x):
 859
 860                 result_ready.acquire()
 861                 try:
 862                         result[0] = x
 863                 finally:
 864                         result_ready.notify()
 865                         result_ready.release()
 866
 867         def do_command(command, username, password):
 868
 869                 try:
 870                         # Popen4 is a popen-type class that combines stdout and stderr
 871                         p = popen2.Popen4(command)
 872
 873                         # read all output data
 874                         p.tochild.write("%s\n" % username)
 875                         p.tochild.write("%s\n" % password)
 876                         p.tochild.close()
 877                         data = p.fromchild.read()
 878
 879                         while True:
 880                                 # might get interrupted by a signal in poll() or waitpid()
 881                                 try:
 882                                         retval = p.wait()
 883                                         set_result((retval, data))
 884                                         break
 885                                 except OSError, ex:
 886                                         if ex.errno == errno.EINTR:
 887                                                 continue
 888                                         raise ex
 889                 except Exception, ex:
 890                         set_result(ex)
 891
 892         if args:
 893                 command = " ".join([command] + args)
 894
 895         worker = threading.Thread(target = do_command, args = (command, username, password, ))
 896         worker.setDaemon(True)
 897         result_ready.acquire()
 898         worker.start()
 899         result_ready.wait(timeout)
 900         try:
 901                 if result == [None]:
 902                         raise Exception, "command timed-out: '%s'" % command
 903         finally:
 904                 result_ready.release()
 905         result = result[0]
 906
 907         if isinstance(result, Exception):
 908                 raise result
 909         else:
 910                 (retval, data) = result
 911                 if os.WIFEXITED(retval) and os.WEXITSTATUS(retval) == 0:
 912                         return data
 913                 else:
 914                         out = "system command ('%s') " % command
 915                         if os.WIFEXITED(retval):
 916                                 out += "failed, rc = %d" % os.WEXITSTATUS(retval)
 917                         else:
 918                                 out += "killed by signal %d" % os.WTERMSIG(retval)
 919                         if data:
 920                                 out += "; output follows:\n" + data
 921                         raise Exception, out
 922
 923 def racadm_reboot(ip, username, password, port, dryrun):
 924         global verbose
 925
 926         try:
 927                 cmd = "/usr/sbin/racadm"
 928                 os.stat(cmd)
 929                 if not dryrun:
 930                         output = runcmd(cmd, ["-r %s -i serveraction powercycle" % ip],
 931                                 username, password)
 932                 else:
 933                         output = runcmd(cmd, ["-r %s -i getsysinfo" % ip],
 934                                 username, password)
 935
 936                 print "RUNCMD: %s" % output
 937                 if verbose:
 938                         logger.debug(output)
 939                 return 0
 940
 941         except Exception, err:
 942                 logger.debug("runcmd raised exception %s" % err)
 943                 if verbose:
 944                         logger.debug(err)
 945                 return -1
 946
 947 def pcu_name(pcu):
 948         if pcu['hostname'] is not None and pcu['hostname'] is not "":
 949                 return pcu['hostname']
 950         elif pcu['ip'] is not None and pcu['ip'] is not "":
 951                 return pcu['ip']
 952         else:
 953                 return None
 954
 955 def get_pcu_values(pcu_id):
 956         # TODO: obviously, this shouldn't be loaded each time...
 957         import soltesz
 958         fb =soltesz.dbLoad("findbadpcus")
 959
 960         try:
 961                 values = fb['nodes']["id_%s" % pcu_id]['values']
 962         except:
 963                 values = None
 964
 965         return values
 966
 967 def check_open_port(values, port_list):
 968         ret = False
 969
 970         if 'portstatus' in values:
 971                 for port in port_list:
 972                         if      port in values['portstatus'] and \
 973                                 values['portstatus'][port] == "open":
 974
 975                                 ret = True
 976
 977         return ret
 978
 979 def reboot(nodename):
 980         return reboot_policy(nodename, True, False)
 981
 982 def reboot_policy(nodename, continue_probe, dryrun):
 983         global verbose
 984
 985         pcu = plc.getpcu(nodename)
 986         if not pcu:
 987                 return False # "%s has no pcu" % nodename
 988
 989         values = get_pcu_values(pcu['pcu_id'])
 990         if values == None:
 991                 return False #"no info for pcu_id %s" % pcu['pcu_id']
 992
 993         # Try the PCU first
 994         logger.debug("Trying PCU %s %s" % (pcu['hostname'], pcu['model']))
 995
 996         ret = reboot_test(nodename, values, continue_probe, verbose, dryrun)
 997         print ret
 998
 999         if ret != 0:
1000                 return False
1001         else:
1002                 return True
1003
1004 def reboot_test(nodename, values, continue_probe, verbose, dryrun):
1005         rb_ret = ""
1006
1007         try:
1008                 # DataProbe iPal (many sites)
1009                 if  continue_probe and values['model'].find("Dataprobe IP-41x/IP-81x") >= 0:
1010                         ipal = IPAL(values, verbose, ['23'])
1011                         rb_ret = ipal.reboot(values[nodename], dryrun)
1012
1013                 # APC Masterswitch (Berkeley)
1014                 elif continue_probe and values['model'].find("APC AP79xx/Masterswitch") >= 0:
1015
1016                         # TODO: make a more robust version of APC
1017                         if values['pcu_id'] in [1163,1055,1111,1231,1113,1127,1128,1148]:
1018                                 apc = APCEurope(values, verbose, ['22', '23'])
1019                                 rb_ret = apc.reboot(values[nodename], dryrun)
1020
1021                         elif values['pcu_id'] in [1173,1221,1220,1225]:
1022                                 apc = APCFolsom(values, verbose, ['22', '23'])
1023                                 rb_ret = apc.reboot(values[nodename], dryrun)
1024
1025                         else:
1026                                 apc = APCMaster(values, verbose, ['22', '23'])
1027                                 rb_ret = apc.reboot(values[nodename], dryrun)
1028
1029                 # BayTech DS4-RPC
1030                 elif continue_probe and values['model'].find("Baytech DS4-RPC") >= 0:
1031                         if values['pcu_id'] in [1041,1209,1025,1052,1057]:
1032                                 # These  require a 'ctrl-c' to be sent...
1033                                 baytech = BayTechCtrlC(values, verbose, ['22', '23'])
1034                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1035
1036                         elif values['pcu_id'] in [1012]:
1037                                 # This pcu sometimes doesn't present the 'Username' prompt,
1038                                 # unless you immediately try again...
1039                                 try:
1040                                         baytech = BayTechGeorgeTown(values, verbose, ['22', '23'])
1041                                         rb_ret = baytech.reboot(values[nodename], dryrun)
1042                                 except:
1043                                         baytech = BayTechGeorgeTown(values, verbose, ['22', '23'])
1044                                         rb_ret = baytech.reboot(values[nodename], dryrun)
1045                         else:
1046                                 baytech = BayTech(values, verbose, ['22', '23'])
1047                                 rb_ret = baytech.reboot(values[nodename], dryrun)
1048
1049                 # iLO
1050                 elif continue_probe and values['model'].find("HP iLO") >= 0:
1051                         try:
1052                                 hpilo = HPiLO(values, verbose, ['22'])
1053                                 rb_ret = hpilo.reboot(0, dryrun)
1054                                 if rb_ret != 0:
1055                                         hpilo = HPiLOHttps(values, verbose, ['443'])
1056                                         rb_ret = hpilo.reboot(0, dryrun)
1057                         except:
1058                                 hpilo = HPiLOHttps(values, verbose, ['443'])
1059                                 rb_ret = hpilo.reboot(0, dryrun)
1060
1061                 # DRAC ssh
1062                 elif continue_probe and values['model'].find("Dell RAC") >= 0:
1063                         # TODO: I don't think DRACRacAdm will throw an exception for the
1064                         # default method to catch...
1065                         try:
1066                                 drac = DRACRacAdm(values, verbose, ['443', '5869'])
1067                                 rb_ret = drac.reboot(0, dryrun)
1068                         except:
1069                                 drac = DRAC(values, verbose, ['22'])
1070                                 rb_ret = drac.reboot(0, dryrun)
1071
1072                 # BlackBox PSExxx-xx (e.g. PSE505-FR)
1073                 elif continue_probe and \
1074                         (values['model'].find("BlackBox PS5xx") >= 0 or
1075                          values['model'].find("ePowerSwitch 1/4/8x") >=0 ):
1076
1077                         # TODO: allow a different port than http 80.
1078                         if values['pcu_id'] in [1089, 1071, 1046, 1035, 1118]:
1079                                 eps = ePowerSwitchGood(values, verbose, ['80'])
1080                         elif values['pcu_id'] in [1003]:
1081                                 eps = ePowerSwitch(values, verbose, ['80'])
1082                         else:
1083                                 eps = ePowerSwitchGood(values, verbose, ['80'])
1084
1085                         rb_ret = eps.reboot(values[nodename], dryrun)
1086
1087                 elif continue_probe:
1088                         rb_ret = "Unsupported_PCU"
1089
1090                 elif continue_probe == False:
1091                         if 'portstatus' in values:
1092                                 rb_ret = "NetDown"
1093                         else:
1094                                 rb_ret = "Not_Run"
1095                 else:
1096                         rb_ret = -1
1097
1098         except ExceptionPort, err:
1099                 rb_ret = str(err)
1100
1101         return rb_ret
1102         # ????
1103         #elif continue_probe and values['protocol'] == "racadm" and \
1104         #               values['model'] == "RAC":
1105         #       rb_ret = racadm_reboot(pcu_name(values),
1106         #                                                                 values['username'],
1107         #                                                                 values['password'],
1108         #                                                                 pcu[nodename],
1109         #                                                                 dryrun)
1110
1111 # Returns true if rebooted via PCU
1112 def reboot_old(nodename, dryrun):
1113         pcu = plc.getpcu(nodename)
1114         if not pcu:
1115                 plc.nodePOD(nodename)
1116                 return False
1117         # Try the PCU first
1118         logger.debug("Trying PCU %s %s" % (pcu['hostname'], pcu['model']))
1119
1120         # APC Masterswitch (Berkeley)
1121         if pcu['model'] == "APC Masterswitch":
1122                 err = apc_reboot(pcu['ip'], pcu['username'],pcu['password'],
1123                                 pcu[nodename], pcu['protocol'], dryrun)
1124
1125         # DataProbe iPal (many sites)
1126         elif pcu['protocol'] == "telnet" and pcu['model'].find("IP-4") >= 0:
1127                 err = ipal_reboot(pcu['ip'],pcu['password'], pcu[nodename], dryrun)
1128
1129         # BayTech DS4-RPC
1130         elif pcu['protocol'] == "ssh" and \
1131         (pcu['model'].find("Baytech") >= 0 or pcu['model'].find("DS4") >= 0):
1132                 err = baytech_reboot(pcu['ip'], pcu['username'],pcu['password'], pcu[nodename], dryrun)
1133
1134         # BlackBox PSExxx-xx (e.g. PSE505-FR)
1135         elif pcu['protocol'] == "http" and (pcu['model'] == "bbpse"):
1136                 err = bbpse_reboot(pcu['ip'], pcu['username'], pcu['password'], pcu[nodename],80, dryrun)
1137
1138         # x10toggle
1139         elif pcu['protocol'] == "ssh" and (pcu['model'] == "x10toggle"):
1140                 err = x10toggle_reboot(pcu['ip'], pcu['username'],pcu['password'], pcu[nodename], dryrun)
1141
1142         #
1143         elif pcu['protocol'] == "racadm" and (pcu['model'] == "RAC"):
1144                 err = racadm_reboot(pcu['ip'], pcu['username'],pcu['password'], pcu_[nodename], dryrun)
1145
1146         # Unknown or unsupported
1147         else:
1148                 err = errno.EPROTONOSUPPORT
1149                 return False
1150         return True
1151
1152
1153 def main():
1154         logger.setLevel(logging.DEBUG)
1155         ch = logging.StreamHandler()
1156         ch.setLevel(logging.DEBUG)
1157         formatter = logging.Formatter('LOGGER - %(message)s')
1158         ch.setFormatter(formatter)
1159         logger.addHandler(ch)
1160
1161         try:
1162                 reboot("planetlab2.cs.uchicago.edu")
1163                 reboot("alice.cs.princeton.edu")
1164         except Exception, err:
1165                 print err
1166
1167 if __name__ == '__main__':
1168         import plc
1169         logger = logging.getLogger("monitor")
1170         main()