swapexp.in 23.3 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3 4

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6 7 8
# All rights reserved.
#

9 10 11 12
use English;
use Getopt::Std;

#
Chad Barb's avatar
Chad Barb committed
13
# This gets invoked from the Web interface.
Chad Barb's avatar
 
Chad Barb committed
14
# Swap an experiment in, swap it out, restart or modify.
15
#
Chad Barb's avatar
Chad Barb committed
16

17 18
sub usage()
{
19
    print STDOUT "Usage: swapexp [-b] [-i | -a | -f] [-r] ".
20
	"<-s in | out | restart | modify | pause> <pid> <eid> [<nsfile>]\n";
21 22
    exit(-1);
}
23
my  $optlist = "biafrs:";
24

25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
#
# Exit codes are important; they tell the web page what has happened so
# it can say something useful to the user. Fatal errors are mostly done
# with die(), but expected errors use this routine. At some point we will
# use the DB to communicate the actual error.
#
# $status < 0 - Fatal error. Something went wrong we did not expect.
# $status = 0 - Termination is proceeding in the background. Notified later.
# $status > 0 - Expected error. User not allowed for some reason. 
# 
sub ExitWithStatus($$)
{
    my ($status, $message) = @_;
    
    if ($status < 0) {
	die("*** $0:\n".
	    "    $message\n");
    }
    else {
	print STDERR "$message\n";
    }
    exit($status);
}

49 50 51 52 53 54
#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
my $TBLOGS = "@TBLOGSEMAIL@";
55
my $TBINFO = "$TB/expinfo";
56
my $TBDOCBASE = "@TBDOCBASE@";
57 58 59 60 61 62 63 64 65

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

my $tbdir    = "$TB/bin/";
66
my $tbdata   = "tbdata";
67
my $batch    = 0;
68
my $idleswap = 0;
69 70
my $autoswap = 0;
my $force    = 0;
Chad Barb's avatar
Chad Barb committed
71
my $reboot   = 0;
72
my $errorstat= -1;
73
my $modifyHosed = 0;
Chad Barb's avatar
 
Chad Barb committed
74

75 76 77 78 79
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
80
my @allnodes;
81
my @row;
82
my $action;
83
my $nextswapstate;
84
my $termswapstate;
Chad Barb's avatar
 
Chad Barb committed
85

86 87 88
#
# Untaint the path
# 
89
$ENV{'PATH'} = "/bin:/usr/bin:$TB/libexec/vis";
90 91 92 93 94 95 96
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

#
# Turn off line buffering on output
#
$| = 1;

97 98 99 100 101 102 103
#
# Set umask for start/swap. We want other members in the project to be
# able to swap/end experiments, so the log and intermediate files need
# to be 664 since some are opened for append.
#
umask(0002);

104 105 106 107 108 109 110 111
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
112 113 114
if (defined($options{"i"})) {
    $idleswap = 1;
}
115 116 117 118 119 120
if (defined($options{"a"})) {
    $autoswap = 1;
}
if (defined($options{"f"})) {
    $force = 1;
}
121 122 123
if (defined($options{"b"})) {
    $batch = 1;
}
Chad Barb's avatar
 
Chad Barb committed
124 125 126
if (defined($options{"r"})) {
    $reboot = 1;
}
127 128 129
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
130 131 132
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
133
	$inout ne "pause"   &&
Chad Barb's avatar
 
Chad Barb committed
134
	$inout ne "modify") {
135 136 137 138 139 140 141
	usage();
    }
}
else {
    usage();
}

Chad Barb's avatar
 
Chad Barb committed
142 143 144 145 146 147
if (@ARGV != (($inout eq "modify") ? 3 : 2)) {
    usage();
}
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
#
# Untaint the arguments.
#
if ($pid =~ /^([-\@\w.]+)$/) {
    $pid = $1;
}
else {
    die("Tainted argument $pid!\n");
}
if ($eid =~ /^([-\@\w.]+)$/) {
    $eid = $1;
}
else {
    die("Tainted argument $eid!\n");
}
163
my $repfile = "$eid.report";
164 165
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
166 167 168 169 170 171 172 173 174 175 176
my $tempnsfile;
my $modnsfile;

if ($inout eq "modify") {
    $tempnsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
    if ($tempnsfile =~ /^([-\w.\/]+)$/) {
	$tempnsfile = $1;
177 178
    }
    else {
179 180 181 182
	die("Tainted nsfile name: $tempnsfile");
    }
    $modnsfile = "$eid-modify.ns";
}
183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200

#
# Verify user and get his DB uid.
#
if (! UNIX2DBUID($UID, \$dbuid)) {
    die("*** $0:\n".
	"    You do not exist in the Emulab Database.\n");
}

#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
    die("*** $0:\n".
	"    Cannot determine your name and email address.\n");
}

#
Chad Barb's avatar
 
Chad Barb committed
201
# Verify that this person can muck with the experiment.
202 203 204 205 206
# Note that any script down the line has to do an admin check also. 
#
if ($UID && !TBAdmin($UID) &&
    !TBExptAccessCheck($dbuid, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
Chad Barb's avatar
 
Chad Barb committed
207
	"    You do not have permission to swap or modify this experiment!\n");
208 209
}

210 211 212 213
# Must do this before lock tables!
# idleswap is in minutes, threshold is in hours
$idleswap_time = 60 * TBGetSiteVar("idle/threshold");

214 215 216 217 218
#
# We have to protect against trying to end an experiment that is currently
# in the process of being terminated. We use a "wrapper" state (actually
# a timestamp so we can say when termination was requested) since
# terminating consists of a couple of different experiment states down inside
Chad Barb's avatar
Chad Barb committed
219
# the tb scripts.
220 221 222 223 224 225 226 227 228 229 230 231 232
#
DBQueryFatal("lock tables experiments write");

$query_result =
    DBQueryFatal("SELECT * FROM experiments WHERE eid='$eid' and pid='$pid'");

if (! $query_result->numrows) {
    die("*** $0:\n".
	"    No such experiment $pid/$eid exists!\n");
}
my %hashrow = $query_result->fetchhash();
my $expt_head_login = $hashrow{'expt_head_uid'};
my $estate          = $hashrow{'state'};
233
my $batchstate      = $hashrow{'batchstate'};
234
my $expt_path       = $hashrow{'path'};
235
my $expt_locked     = $hashrow{'expt_locked'};
236
my $isbatchexpt     = $hashrow{'batchmode'};
237
my $canceled        = $hashrow{'canceled'};
238 239 240 241 242 243 244 245 246 247
my $swappablebit= $hashrow{'swappable'};
my $idleswapbit = $hashrow{'idleswap'};
my $autoswapbit = $hashrow{'autoswap'};
my $swappablestr= ( $swappablebit ? "Yes" : "No" );
my $idleswapstr = ( $idleswapbit ? "Yes" : "No" );
my $autoswapstr = ( $autoswapbit ? "Yes" : "No" );
my $noswap      = $hashrow{'noswap_reason'};
my $noidleswap  = $hashrow{'noidleswap_reason'};
my $idleswaptime= $hashrow{'idleswap_timeout'} / 60.0;
my $autoswaptime= $hashrow{'autoswap_timeout'} / 60.0;
248

249 250
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
251
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
252 253 254
    $idleswaptime = $idleswap_time / 60.0;
}

255 256
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
257
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
258 259
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
260

261
if (! chdir($workdir)) {
262
    die("*** $0:\n".
263
	"    Could not chdir to $workdir: $!\n");
264 265
}

266
#
267 268 269
# This script is called from the batch daemon.
# 
if ($batch) {
270
    #
271 272 273
    # Sanity Check. If called from the daemon, must already be locked,
    # must be a batch experiment, and must be in proper state for the
    # operation requested. 
274
    #
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
    die("*** $0:\n".
	"    Experiment $pid/$eid is supposed to be a batch experiment!\n")
	if (!$isbatchexpt);
    
    die("*** $0:\n".
	"    Batch experiment $pid/$eid should be locked!\n")
	if (!defined($expt_locked) ||
	    $batchstate ne BATCHSTATE_LOCKED());

    if ($inout eq "in") {
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid is not in the proper state!\n".
	    "    Currently $estate, but should be QUEUED.\n")
	    if ($estate ne EXPTSTATE_QUEUED);
	
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid has been canceled! Aborting.\n")
	    if ($canceled);
    }
    elsif ($inout eq "out") {
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid is not in the proper state!\n".
	    "    Currently $estate, but should be ACTIVE.\n")
	    if ($estate ne EXPTSTATE_ACTIVE);
299 300
    }
    else {
301 302 303 304 305 306
	die("*** $0:\n".
	    "    Improper request from batch daemon for $pid/$eid!\n");
    }
}
else {
    if ($isbatchexpt) {
307 308 309 310
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
311 312
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
313
	    if ($canceled);
314

315
	if ($inout eq "in") {
316
	    ExitWithStatus(1,
317 318 319 320
			   "Batch experiment $pid/$eid must be SWAPPED to\n".
			   "QUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_SWAPPED);
	    SetExpState($pid, $eid, EXPTSTATE_QUEUED);
321 322
	}
	elsif ($inout eq "out") {
323
	    ExitWithStatus(1,
324 325 326 327
			   "Batch experiment $pid/$eid must be ACTIVE or\n".
			   "ACTIVATING to swap out. Currently $estate.")
		if ($estate ne EXPTSTATE_ACTIVE &&
		    $estate ne EXPTSTATE_ACTIVATING);
328 329 330 331 332

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
333
	    TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
334 335
	}
	elsif ($inout eq "pause") {
336
	    ExitWithStatus(1,
337 338 339
			   "Batch experiment $pid/$eid must be QUEUED to\n".
			   "DEQUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_QUEUED);
340 341

	    #
342 343 344 345
	    # XXX. The batch daemon might already have the experiment, but
	    # not have shipped it off to startexp. Change the state
	    # anyway. The error will be noticed later when startexp dies,
	    # and the batch daemon gets the error back. This sucks.
346
	    #
347
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
348
	}
349
	elsif ($inout eq "modify") {
350
	    ExitWithStatus(1,
351 352 353 354 355 356
			   "Batch experiment $pid/$eid must be SWAPPED or\n".
			   "ACTIVE to modify. Currently $estate.")
		if (($estate ne EXPTSTATE_SWAPPED &&
		     $estate ne EXPTSTATE_ACTIVATING) ||
		    $batchstate != BATCHSTATE_UNLOCKED());

357
	    #
358
	    # Otherwise, proceed with the modify. The experiment will be
359 360
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
361 362 363 364
	    # the experiments table. The batch daemon will leave it alone
	    # until the modify is done. If the modify fails and cannot recover
	    # it is going to get swapped out; that is okay since the batch
	    # daemon does not keep state internally. 
365
	    #
366 367
	    goto doit;
	}
368 369
	else {
	    die("*** $0:\n",
370
		"    Operation $inout not allowed on a batch experiment!\n");
371
	}
372 373
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
374
      doit:
375
    }
376 377 378 379 380 381 382 383 384 385
    else {
	#
	# If the cancel flag is set, then user must wait for that to
	# clear before we can do anything else.
	#
	ExitWithStatus(1,
		       "Experiment $pid/$eid has its cancel flag set!.\n".
		       "You must wait for that to clear before you can swap\n".
		       "or modify the experiment.\n")
	    if ($canceled);
386

387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440
	#
	# Check the state for the various operations.
	#
	if (!$force) {
	  SWITCH: for ($inout) {
	      /^in$/i && do {
		  if ($estate ne EXPTSTATE_SWAPPED()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped out!");
		  }
		  last SWITCH;
	      };
	      /^out$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
		      $estate ne EXPTSTATE_ACTIVATING()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped in ".
				     "or activating!\n");
		  }
		  
		  if ($estate eq EXPTSTATE_ACTIVATING()) {
		      #
		      # All we can do is set the cancel flag and hope that
		      # it gets noticed. We do not wait. 
		      # 
		      TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
		      
		      ExitWithStatus(0,
				     "Experiment $pid/$eid swapin has been  ".
				     "marked for cancelation.\n".
				     "You will receive email when the original ".
				     "swap request has finished.");
		  }
		  last SWITCH;
	      };
	      /^restart$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped in!");
		  }
		  last SWITCH;
	      };
	      /^modify$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
		      $estate ne EXPTSTATE_SWAPPED()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid must be ACTIVE or\n".
				     "SWAPPED to modify!\n");
		  }
		  last SWITCH;
	      };
	      die("*** $0:\n".
		  "    Missing state check for action: $action\n");
	  }
441 442
	}
    }
443 444
}

445 446 447 448 449 450 451
#
# Determine the temporary and next state for experiment. If the experiment
# is a batch experiment, then the next state is actually handled by the
# batch daemon, but we still have to deal with the temporary state. 
#
SWITCH: for ($inout) {
    /^in$/i && do {
452
	$nextswapstate = EXPTSTATE_ACTIVATING();
453 454 455
	last SWITCH;
    };
    /^out$/i && do {
456
	$nextswapstate = EXPTSTATE_SWAPPING();
457 458 459
	last SWITCH;
    };
    /^restart$/i && do {
460
	$nextswapstate = EXPTSTATE_RESTARTING();
461 462 463
	last SWITCH;
    };
    /^modify$/i && do {
464 465
	$nextswapstate = (($estate eq EXPTSTATE_SWAPPED()) ?
			  EXPTSTATE_MODIFY_PARSE() : EXPTSTATE_MODIFY_REPARSE());
466 467
	last SWITCH;
    };
468
    die("*** $0:\n".
469
	"    Missing state check for action: $action\n");
470
}
471 472
 
# Update idleswap_timeout to whatever the current value is.
473
if ($inout ne "out") {
474 475 476
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
477

478 479 480 481 482 483 484
#
# On a failure, we go back to this swapstate. Might be modified below.
# 
$termswapstate = $estate;

# Lock the record, set the nextstate, and unlock the table.
TBLockExp($pid, $eid, $nextswapstate);
485 486 487 488
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
489 490 491 492
# inconsistent state. Be sure to call fatal() only since we are
# going into the background, and we have to send email since no
# one is going to see printed error messages (output goes into the
# log file, which will be sent along in the email). 
493 494
#

495 496 497 498 499 500 501 502 503
if ($inout eq "in") {
    $action = "swapped in";
}
if ($inout eq "out") {
    $action = "swapped out";
}
if ($inout eq "restart") {
    $action = "restarted";
}
Chad Barb's avatar
 
Chad Barb committed
504 505 506
if ($inout eq "modify") {
    $action = "modified";
}
507

508 509 510 511 512 513 514 515 516 517 518 519 520 521 522
#
# Get email address of the experiment head, which may be different than
# the person who is actually terminating the experiment, since its polite
# to let the original creator know whats going on. 
#
my $expt_head_name;
my $expt_head_email;

if (! UserDBInfo($expt_head_login, \$expt_head_name, \$expt_head_email)) {
    print STDERR "*** WARNING: ".
	         "Could not determine name/email for $expt_head_login.\n";
    $expt_head_name  = "TBOPS";
    $expt_head_email = $TBOPS;
}

523 524 525 526 527 528 529 530 531 532 533 534
#
# Before going to background, we have to copy out the NS file!
#
if ($inout eq "modify") {
    unlink($modnsfile);
    if (system("/bin/cp", "$tempnsfile", "$modnsfile")) {
	die("*** $0:\n".
	    "    Could not copy $tempnsfile to $modnsfile");
    }
    chmod(0664, "$modnsfile");
}

535 536 537 538
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
539
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
540
    TBExptSetLogFile($pid, $eid, $logname);
541
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
542

543 544 545 546
    if (TBBackGround($logname)) {
	#
	# Parent exits normally
	#
547 548
	print "Experiment $pid/$eid is now being $action.\n".
	    "You will be notified via email when the this is done.\n";
549 550 551 552
	exit(0);
    }
}

553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568
#
# Gather stats; start clock ticking
#
if ($inout eq "in") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0,
		    TBDB_STATS_FLAGS_START);
}
elsif ($inout eq "out") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
		    TBDB_STATS_FLAGS_START);
}
elsif ($inout eq "modify") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0,
		    TBDB_STATS_FLAGS_START);
}

569 570 571
#
# Remove old report file since its contents are going to be invalid.
#
572
if ($inout ne "restart" && -e $repfile) {
573 574 575
    unlink("$repfile");
}

576 577 578 579
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
580 581 582 583
    my $optarg = (($force || $idleswap) ? "-force" : "");
    
    print STDOUT "Running 'tbswap out $optarg $pid $eid'\n";
    if (system("$tbdir/tbswap out $optarg $pid $eid") != 0) {
584
	$errorstat = $? >> 8;
585
	fatal("tbswap out failed!");
586
    }
587
    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
588
}
589
elsif ($inout eq "in") {
590
    print STDOUT "Running 'tbswap in $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
591
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
592
	$errorstat = $? >> 8;
593
	fatal("tbswap in failed!");
594
    }
595
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
596

597
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
598
}
Chad Barb's avatar
 
Chad Barb committed
599
elsif ($inout eq "modify") {
Chad Barb's avatar
Chad Barb committed
600
    my $modifyError = "";
601
    my $oldstate    = $estate;
Chad Barb's avatar
Chad Barb committed
602

603 604 605
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

Chad Barb's avatar
Chad Barb committed
606
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
607
    if (TBExptBackupVirtualState($pid, $eid)) {
608
	fatal("Could not backup experiment state; cannot safely continue!");
Chad Barb's avatar
Chad Barb committed
609 610 611 612 613
    }

    #
    # Rerun tbprerun if modifying.
    #
614 615
    print STDOUT "Running 'tbprerun $pid $eid $modnsfile'\n";
    if (system("$tbdir/tbprerun $pid $eid $modnsfile") != 0) {
Chad Barb's avatar
Chad Barb committed
616 617 618
	$modifyError = "tbprerun failed!";
    }

Chad Barb's avatar
 
Chad Barb committed
619
    #
620
    # Our next state depends on whether the experiment was active or swapped.
Chad Barb's avatar
 
Chad Barb committed
621
    #
622 623 624
    if (! $modifyError) {
	if ($estate eq EXPTSTATE_SWAPPED) {
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
Chad Barb's avatar
 
Chad Barb committed
625
	}
626 627 628 629 630 631 632 633 634 635
	else {
	    SetExpState($pid, $eid, EXPTSTATE_MODIFY_RESWAP);
	    
	    my $optarg = ($reboot ? "-reboot" : "");

	    print STDOUT "Running 'tbswap update $optarg $pid $eid'\n";
	    if (system("$tbdir/tbswap update $optarg $pid $eid") != 0) {
		$errorstat = $? >> 8;
		$modifyError = "tbswap update failed!";
	    }
Chad Barb's avatar
 
Chad Barb committed
636

637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656
	    #
	    # See what tbswap did. It might have swapped it out if there
	    # was an error. 
	    # 
	    if (! $modifyError) {
		SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
		$estate = EXPTSTATE_ACTIVE;
	    }
	    elsif ($errorstat & 0x40) {
		#
		# Icky. Magic return code that says tbswap swapped it out.
		# We do not want tbswap to muck with states anymore, so
		# need to know what it did. At some point we should clean
		# up the exit reporting! Anyway, fatal() needs to know the
		# the right state to go back to (no longer ACTIVE).
		#
		$estate = EXPTSTATE_SWAPPED;
		$termswapstate = EXPTSTATE_SWAPPED;
                # Old accounting info.
		TBSetExpSwapTime($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
657
	    }
658
	}
Chad Barb's avatar
Chad Barb committed
659 660 661
    }

    if ($modifyError) {
662
	print STDOUT "Modify Error: $modifyError\n";
Chad Barb's avatar
Chad Barb committed
663
	print STDOUT "Recovering experiment state...\n";
664
	
665 666
	# Must deal with the prerender explicitly since it runs background.
	system("prerender -r $pid $eid");
667
	TBExptRemoveVirtualState($pid, $eid);
668
	
669
	if (TBExptRestoreVirtualState($pid, $eid) == 0) {
670 671
	    # Must deal with the prerender explicitly since it runs background.
	    system("prerender -t $pid $eid");
672 673 674 675 676
	    fatal("Update aborted; old state restored.");
	}
	else {
	    $modifyHosed = 1;
	    fatal("Experiment state could not be restored!");
Chad Barb's avatar
Chad Barb committed
677
	}
Chad Barb's avatar
 
Chad Barb committed
678
    }
679
    
680
    TBExptClearBackupState($pid, $eid);
681
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
682
}
Chad Barb's avatar
 
Chad Barb committed
683
else { # $inout eq "restart" assumed.
684
    print STDOUT "Running 'tbrestart $pid $eid'\n";
685
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
686
	fatal("tbrestart failed!");
687
    }
688
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
689
}
690

691 692 693 694 695 696 697 698 699 700
#
# Try to copy off the files for testbed information gathering.
#
TBSaveExpLogFiles($pid, $eid);

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files. This overwrites existing files of course,
# but thats okay.
#
701
system("cp -Rfp $workdir/ $userdir/tbdata/");
702

703 704 705 706
#
# Gather stats. 
#
if ($inout eq "in") {
707
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
708 709
}
elsif ($inout eq "out") {
710
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
711
		    ($idleswap ? TBDB_STATS_FLAGS_IDLESWAP() : 0));
712 713
}
elsif ($inout eq "modify") {
714
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
715
}
716 717
# Old accounting info.
TBSetExpSwapTime($pid, $eid);
718

719 720 721 722 723 724
#
# Set the swapper uid on success only, and *after* gathering swap stats!
#
TBExptSetSwapUID($pid, $eid, $dbuid);

#
725
# In batch mode, just exit without sending email or unlocking. The
726
# batch daemon will take care of that.
727 728 729 730 731
#
if ($batch) {
    exit(0);
}

732 733 734 735 736 737 738
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

739 740 741
#
# Must unlock before exit.
#
742
TBUnLockExp($pid, $eid);
743 744 745 746 747

#
# Since the swap completed, clear the cancel flag. This must be done
# after we change the experiment state (above). 
#
748
TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
749 750 751

print "Swap Success!\n";

752 753 754 755
#
# Send email notification to user.
#
my $message =
756 757
    "Experiment $eid in project $pid has been ";

758
if ($inout eq "out" && ($idleswap || $autoswap || $force) ) {
759
    $message .= "forcibly swapped out by\nEmulab";
760 761 762 763 764
    if ($idleswap) {
	$message .= " because it was idle for too long (Idle-Swap).\n".
	  "(See also the Idle-Swap info in \n".
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    } elsif ($autoswap) {
765 766
	$message .= " because it exceeded its Maximum Duration.\n".
	  "(See also the Max. Duration info in \n".
767 768 769 770 771
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    } elsif ($force) {
	$message .= ". (See also our Node Usage Policies in \n".
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    }
772 773 774 775 776
}
else {
    $message .= "$action.\n";
}

777 778 779 780 781
if ($inout eq "in") {
    # Add the swap settings...
    $message .="\nCurrent swap settings:\n$swapsettings";
}

782 783
$message .=
    "\n".
784 785
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
786 787

SENDMAIL("$user_name <$user_email>",
788
	 "Experiment $pid/$eid \u$action",
789
	 $message,
790
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
791 792
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
793 794
	 (($inout eq "restart") ? ($logname) :
	  (($repfile, $logname), (defined($modnsfile) ? ($modnsfile) : ()))));
795 796 797 798 799 800

exit 0;

sub fatal($)
{
    my($mesg) = $_[0];
Chad Barb's avatar
Chad Barb committed
801

802 803
    print STDOUT "*** $0:\n".
	         "    $mesg\n";
804

805 806 807 808 809 810 811 812 813 814 815 816 817
    #
    # Gather stats. 
    #
    if ($inout eq "in") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, $errorstat);
    }
    elsif ($inout eq "out") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, $errorstat);
    }
    elsif ($inout eq "modify") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, $errorstat);
    }

818 819 820 821 822 823 824
    #
    # Clear backup state since not needed anymore; experiment is toast. 
    # 
    if ($inout eq "modify") {
	TBExptClearBackupState($pid, $eid);
    }

Chad Barb's avatar
 
Chad Barb committed
825
    #
826
    # If hosed, we entirely terminate the experiment.
Chad Barb's avatar
 
Chad Barb committed
827
    #
828
    if ($modifyHosed) {
Chad Barb's avatar
 
Chad Barb committed
829
	#
830
	# Note: $estate is indeed still set appropriately!
Chad Barb's avatar
 
Chad Barb committed
831 832
	#
	if ($estate eq EXPTSTATE_ACTIVE) {
833
	    print "Running 'tbswap out -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
834 835 836 837
	    if (system("$tbdir/tbswap out -force $pid $eid") != 0) {
		print "tbswap out failed!\n";
	    }
	}
Chad Barb's avatar
Chad Barb committed
838

839
	print "Running 'tbend -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
840 841 842
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
	}
843
	# Must override since we are so badly hosed. 
844
	$termswapstate = EXPTSTATE_TERMINATED;
Chad Barb's avatar
 
Chad Barb committed
845 846
    }

847 848 849
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

850 851 852
    # Set proper state, which is typically the way we came in.
    SetExpState($pid, $eid, $termswapstate);

853
    #
854
    # In batch mode, exit without sending the email or unlocking. The
855
    # batch daemon will take care of that.
856 857
    #
    if ($batch) {
858
	exit($errorstat);
859 860
    }

861
    #
Chad Barb's avatar
Chad Barb committed
862
    # Clear the log file so the web page stops spewing.
863 864 865 866 867
    #
    if (defined($logname)) {
	TBExptCloseLogFile($pid, $eid);
    }

868 869
    # Unlock and reset state to its terminal value.
    TBUnLockExp($pid, $eid);
870 871 872 873 874

    #
    # Clear the cancel flag now that the operation is complete. Must be done
    # after we change the experiment state (above).
    #
875
    TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
876

877 878 879 880
    #
    # Send a message to the testbed list. Append the logfile.
    #
    SENDMAIL("$user_name <$user_email>",
881
	     "Swap ${inout} Failure: $pid/$eid",
882
	     $mesg,
883
	     ($idleswap ? $TBOPS : "$user_name <$user_email>"),
884
	     "Cc:  $expt_head_name <$expt_head_email>\n".
Leigh B. Stoller's avatar
Leigh B. Stoller committed
885
	     "Cc:  $TBOPS",
886
	     (($logname), (defined($modnsfile) ? ($modnsfile) : ())));
887

Leigh B. Stoller's avatar
Leigh B. Stoller committed
888
    if ($modifyHosed) {
Chad Barb's avatar
 
Chad Barb committed
889 890 891 892 893 894 895 896 897
	#
	# Copy off the workdir to the user directory, Then back up both of
	# them for post-mortem debugging.
	#
	system("/bin/cp -Rfp $workdir/ $userdir/tbdata");
	system("/bin/rm -rf  ${workdir}-failed");
	system("/bin/mv -f   $workdir ${workdir}-failed");
	system("/bin/rm -rf  ${userdir}-failed");
	system("/bin/mv -f   $userdir ${userdir}-failed");
Chad Barb's avatar
Chad Barb committed
898
	TBExptDestroy($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
899 900
    }

901
    exit($errorstat);
902
}