swapexp.in 22.7 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;
Chad Barb's avatar
 
Chad Barb committed
73

74 75 76 77 78
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
79
my @allnodes;
80
my @row;
81
my $action;
82 83 84
my $nextswapstate;
my $tempswapstate;

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.
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235
#
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'};
my $expt_path       = $hashrow{'path'};
my $isbatchexpt     = $hashrow{'batchmode'};
my $ebatchstate     = $hashrow{'batchstate'};
236
my $cancelflag  = $hashrow{'canceled'};
237 238 239 240 241 242 243 244 245 246
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;
247

248 249
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
250
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
251 252 253
    $idleswaptime = $idleswap_time / 60.0;
}

254 255
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
256
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
257 258
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
259

260
if (! chdir($workdir)) {
261
    die("*** $0:\n".
262
	"    Could not chdir to $workdir: $!\n");
263 264
}

265
#
266
# Batchmode.
267
#
268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294
if ($isbatchexpt) {
    #
    # When coming from the daemon, sanity check the batch state.
    #
    if ($batch) {
	if ($inout eq "in") {
	    die("*** $0:\n".
		"    Batch experiment $pid/$eid is not in the proper state!\n".
		"    Currently $ebatchstate, but should be ACTIVATING\n")
		if ($ebatchstate ne BATCHSTATE_ACTIVATING);
	}
	elsif ($inout eq "out") {
	    die("*** $0:\n".
		"    Batch experiment $pid/$eid is not in the proper state!\n".
		"    Currently $ebatchstate, but should be TERMINATING\n")
		if ($ebatchstate ne BATCHSTATE_TERMINATING);
	}
	else {
	    die("*** $0:\n".
		"    Improper request from batch daemon for $pid/$eid!\n");
	}
    }
    else {
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
295 296 297 298
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
	    if ($cancelflag);

299
	if ($inout eq "in") {
300 301 302
	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid must be PAUSED to\n".
			   "swap in. Currently $ebatchstate.")
303 304 305 306
		if ($ebatchstate ne BATCHSTATE_PAUSED);
	    TBSetBatchState($pid, $eid, BATCHSTATE_POSTED);
	}
	elsif ($inout eq "out") {
307 308 309 310 311 312 313 314 315 316
	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid must be RUNNING or\n".
			   "ACTIVATING to swap out. Currently $ebatchstate.")
		if ($ebatchstate ne BATCHSTATE_RUNNING &&
		    $ebatchstate ne BATCHSTATE_ACTIVATING);

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
317 318 319
	    TBSetBatchCancelFlag($pid, $eid, BATCHMODE_CANCELSWAP);
	}
	elsif ($inout eq "pause") {
320 321 322
	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid must be POSTED to\n".
			   "pause. Currently $ebatchstate.")
323
		if ($ebatchstate ne BATCHSTATE_POSTED);
324 325 326 327 328

	    #
	    # If the batchstate is POSTED, we can just set it to PAUSED
	    # since the batch_daemon is locked out from messing with it.
	    #
329
	    TBSetBatchState($pid, $eid, BATCHSTATE_PAUSED);
330
	}
331
	elsif ($inout eq "modify") {
332 333 334 335 336 337
	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid must be PAUSED or\n".
			   "RUNNING to modify. Currently $ebatchstate.")
		if ($ebatchstate ne BATCHSTATE_PAUSED &&
		    $ebatchstate ne BATCHSTATE_RUNNING);
	    #
338
	    # Otherwise, proceed with the modify. The experiment will be
339 340 341 342
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
	    # the experiments table.
	    #
343 344
	    goto doit;
	}
345 346
	else {
	    die("*** $0:\n",
347
		"    Operation $inout not allowed on a batch experiment!\n");
348
	}
349 350
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
351
    }
352
  doit:
353
}
354 355 356 357 358 359 360 361 362 363
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 or\n".
		   "or modify the experiment.\n")
	if ($cancelflag);
364

365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
    #
    # Check the state for the various operations.
    #
    if (!$force) {
        SWITCH: for ($inout) {
	    /^in$/i && do {
		if ($ebatchstate ne BATCHSTATE_PAUSED()) {
		    ExitWithStatus(1,
				   "Experiment $pid/$eid is not swapped out!");
		}
		last SWITCH;
	    };
	    /^out$/i && do {
		if ($ebatchstate eq BATCHSTATE_PAUSED()) {
		    ExitWithStatus(1,
				   "Experiment $pid/$eid is swapped out!");
		}
		elsif ($ebatchstate ne BATCHSTATE_RUNNING() &&
		       $ebatchstate ne BATCHSTATE_ACTIVATING()) {
		    ExitWithStatus(1,
				   "Experiment $pid/$eid is not swapped in!");
		}

		if ($ebatchstate eq BATCHSTATE_ACTIVATING()) {
		    #
		    # All we can do is set the cancel flag and hope that
		    # it gets noticed. We do not wait. 
		    # 
		    TBSetBatchCancelFlag($pid, $eid, BATCHMODE_CANCELSWAP);

		    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$/i && do {
		if ($ebatchstate ne BATCHSTATE_RUNNING()) {
		    ExitWithStatus(1,
				   "Experiment $pid/$eid is not swapped in!");
		}
		last SWITCH;
	    };
	    /^modify$/i && do {
		if ($ebatchstate ne BATCHSTATE_RUNNING() &&
		    $ebatchstate ne BATCHSTATE_PAUSED()) {
		    ExitWithStatus(1,
				   "Experiment $pid/$eid is in transition!");
		}
		last SWITCH;
	    };
	    die("*** $0:\n".
		"    Missing state check for action: $action\n");
	}
    }
422 423
}

424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
#
# 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 {
	$tempswapstate = BATCHSTATE_ACTIVATING();
	$nextswapstate = BATCHSTATE_RUNNING();
	last SWITCH;
    };
    /^out$/i && do {
	$tempswapstate = BATCHSTATE_TERMINATING();
	$nextswapstate = BATCHSTATE_PAUSED();
	last SWITCH;
    };
    /^restart$/i && do {
	$tempswapstate = BATCHSTATE_RUNNING_BUSY();
	$nextswapstate = BATCHSTATE_RUNNING();
	last SWITCH;
    };
    /^modify$/i && do {
	$tempswapstate = BATCHSTATE_MODIFYING();
	$nextswapstate = $ebatchstate;
	last SWITCH;
    };
450
    die("*** $0:\n".
451
	"    Missing state check for action: $action\n");
452
}
453 454
 
# Update idleswap_timeout to whatever the current value is.
455
if ($inout ne "out") {
456 457 458
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
459

460 461
# Lock the record, set the intermediate state, and unlock the table.
TBLockExp($pid, $eid, $tempswapstate);
462 463 464 465
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
466 467 468 469
# 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). 
470 471
#

472 473 474 475 476 477 478 479 480
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
481 482 483
if ($inout eq "modify") {
    $action = "modified";
}
484

485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
#
# 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;
}

500 501 502 503 504 505 506 507 508 509 510 511
#
# 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");
}

512 513 514 515
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
516
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
517
    TBExptSetLogFile($pid, $eid, $logname);
518
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
519

520 521 522 523
    if (TBBackGround($logname)) {
	#
	# Parent exits normally
	#
524 525
	print "Experiment $pid/$eid is now being $action.\n".
	    "You will be notified via email when the this is done.\n";
526 527 528 529
	exit(0);
    }
}

530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
#
# 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);
}

546 547 548
#
# Remove old report file since its contents are going to be invalid.
#
549
if ($inout ne "restart" && -e $repfile) {
550 551 552
    unlink("$repfile");
}

553 554 555 556
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
557
    if ($force || $idleswap) { $arg = "-force"; } else { $arg = ""; }
Chad Barb's avatar
 
Chad Barb committed
558
    print STDOUT "Running 'tbswap out' with arguments: $pid $eid\n";
559
    if (system("$tbdir/tbswap out $arg $pid $eid") != 0) {
560
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
561
	fatal("tbswap out failed!\n");
562
    }
Chad Barb's avatar
Chad Barb committed
563

564 565 566 567 568
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_SWAPPED) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
}
569
elsif ($inout eq "in") {
Chad Barb's avatar
 
Chad Barb committed
570 571
    print STDOUT "Running 'tbswap in' with arguments: $pid $eid\n";
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
572
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
573
	fatal("tbswap in failed!\n");
574
    }
Chad Barb's avatar
Chad Barb committed
575

576 577 578 579
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_ACTIVE) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
580

581
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
582
}
Chad Barb's avatar
 
Chad Barb committed
583
elsif ($inout eq "modify") {
Chad Barb's avatar
Chad Barb committed
584 585
    my $modifyError = "";

586 587 588
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

Chad Barb's avatar
Chad Barb committed
589
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
590
    if (TBExptBackupVirtualState($pid, $eid)) {
Chad Barb's avatar
Chad Barb committed
591 592 593
	fatal("*** $0:\n".
	      "    Could not backup experiment state; cannot safely continue!\n");
    }
594 595 596
    # Must deal with the prerender explicitly since it runs background.
    system("prerender -r $pid $eid");
    
Chad Barb's avatar
Chad Barb committed
597 598 599 600 601
    TBExptRemoveVirtualState($pid, $eid);

    #
    # Rerun tbprerun if modifying.
    #
602
    if (system("$tbdir/tbprerun -m $pid $eid $modnsfile") != 0) {
Chad Barb's avatar
Chad Barb committed
603 604 605
	$modifyError = "tbprerun failed!";
    }

Chad Barb's avatar
 
Chad Barb committed
606 607 608 609
    #
    # If experiment is currently swapped out, no need to do an update 
    # after modifying it.
    #
Chad Barb's avatar
Chad Barb committed
610
    if (! $modifyError && $estate eq EXPTSTATE_ACTIVE) {
Chad Barb's avatar
 
Chad Barb committed
611
	print STDOUT "Running 'tbswap update' with arguments: $pid $eid\n";
Chad Barb's avatar
Chad Barb committed
612 613 614 615

	my $rebootSwitch = "";
	if ($reboot) {
	    $rebootSwitch = "-reboot";
Chad Barb's avatar
 
Chad Barb committed
616 617
	}

Chad Barb's avatar
Chad Barb committed
618 619 620
	if (system("$tbdir/tbswap update $rebootSwitch $pid $eid") != 0) {
	    $errorstat = $? >> 8;
	    $modifyError = "tbswap update failed!";
Chad Barb's avatar
 
Chad Barb committed
621
	}
Chad Barb's avatar
 
Chad Barb committed
622

Chad Barb's avatar
Chad Barb committed
623 624 625 626
	if (! $modifyError) {
	    $estate = ExpState($pid, $eid);
	    if ($estate ne EXPTSTATE_ACTIVE) {
		fatal("Experiment is in the wrong state: $estate!");
Chad Barb's avatar
 
Chad Barb committed
627 628
	    }
	}
Chad Barb's avatar
Chad Barb committed
629 630 631 632 633 634 635 636 637 638

	if (! $modifyError) {
	    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
	}
    }

    if ($modifyError) {
	print STDERR "*** $0:\n".
	             "    $modifyError\n";
	print STDOUT "Recovering experiment state...\n";
639 640
	# Must deal with the prerender explicitly since it runs background.
	system("prerender -r $pid $eid");
Chad Barb's avatar
Chad Barb committed
641
	TBExptRemoveVirtualState($pid, $eid );
642
	
643 644
	if (TBExptRestoreVirtualState($pid, $eid) == 0) {
	    TBExptClearBackupState($pid, $eid);
645 646
	    # Must deal with the prerender explicitly since it runs background.
	    system("prerender -t $pid $eid");
Chad Barb's avatar
Chad Barb committed
647 648 649 650 651 652
	    fatal("*** Update aborted; old state restored.\n");
	} else {
	    # Set state to NEW so experiment will get wiped.
	    SetExpState($pid, $eid, EXPTSTATE_NEW);
	    fatal("*** Experiment state could not be restored!\n");
	}
Chad Barb's avatar
 
Chad Barb committed
653
    }
654
    TBExptClearBackupState($pid, $eid);
655
}
Chad Barb's avatar
 
Chad Barb committed
656
else { # $inout eq "restart" assumed.
657 658 659 660 661
    print STDOUT "Running tbrestart with arguments: $pid $eid\n";
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
	fatal("tbrestart failed!\n");
    }
}
662

663 664 665 666 667 668 669 670 671 672
#
# 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.
#
673
system("cp -Rfp $workdir/ $userdir/tbdata/");
674

675 676 677 678
#
# Gather stats. 
#
if ($inout eq "in") {
679
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
680 681
}
elsif ($inout eq "out") {
682
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
683
		    ($idleswap ? TBDB_STATS_FLAGS_IDLESWAP() : 0));
684 685
}
elsif ($inout eq "modify") {
686
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
687 688
}

689 690 691 692 693 694
#
# Set the swapper uid on success only, and *after* gathering swap stats!
#
TBExptSetSwapUID($pid, $eid, $dbuid);

#
695 696
# In batch mode, just exit without sending email or unlocking. The
# batch daemon will take care of that and setting the proper state. 
697 698 699 700 701
#
if ($batch) {
    exit(0);
}

Chad Barb's avatar
 
Chad Barb committed
702 703 704 705 706
#
# HACK! if successful, put new NS file in DB.
#

if ($inout eq "modify") {
707
    $nsdata_string = `cat $modnsfile`;
Chad Barb's avatar
 
Chad Barb committed
708 709 710 711 712 713
    if (defined($nsdata_string)) {
	$nsdata_string = DBQuoteSpecial($nsdata_string);

	DBQueryWarn("delete from nsfiles WHERE eid='$eid' and pid='$pid'");
	DBQueryWarn("insert into nsfiles (pid, eid, nsfile) ".
		    "VALUES('$pid', '$eid', $nsdata_string)");
714 715 716
    }
    else {
	print "Warning!! Could not read nsfile '$modnsfile'!\n";
Chad Barb's avatar
 
Chad Barb committed
717 718 719
    }
}

720 721 722 723 724 725 726
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

727 728 729 730 731 732 733 734 735 736 737 738 739
#
# Must unlock before exit.
#
TBUnLockExp($pid, $eid, $nextswapstate);

#
# Since the swap completed, clear the cancel flag. This must be done
# after we change the experiment state (above). 
#
TBSetBatchCancelFlag($pid, $eid, BATCHMODE_CANCELCLEAR);

print "Swap Success!\n";

740 741 742 743
#
# Send email notification to user.
#
my $message =
744 745
    "Experiment $eid in project $pid has been ";

746
if ($inout eq "out" && ($idleswap || $autoswap || $force) ) {
747
    $message .= "forcibly swapped out by\nEmulab";
748 749 750 751 752
    if ($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) {
753 754
	$message .= " because it exceeded its Maximum Duration.\n".
	  "(See also the Max. Duration info in \n".
755 756 757 758 759
	  "$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";
    }
760 761 762 763 764
}
else {
    $message .= "$action.\n";
}

765 766 767 768 769
if ($inout eq "in") {
    # Add the swap settings...
    $message .="\nCurrent swap settings:\n$swapsettings";
}

770 771
$message .=
    "\n".
772 773
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
774 775

SENDMAIL("$user_name <$user_email>",
776
	 "Experiment $pid/$eid \u$action",
777
	 $message,
778
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
779 780
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
781 782
	 (($inout eq "restart") ? ($logname) :
	  (($repfile, $logname), (defined($modnsfile) ? ($modnsfile) : ()))));
783 784 785 786 787 788

exit 0;

sub fatal($)
{
    my($mesg) = $_[0];
Chad Barb's avatar
Chad Barb committed
789

790 791
    print STDOUT $mesg;

792 793 794 795 796 797 798 799 800 801 802 803 804
    #
    # 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);
    }

Chad Barb's avatar
 
Chad Barb committed
805 806 807 808 809 810 811 812 813 814 815 816
    #
    # if $hosed == 1, we entirely terminate the experiment.
    #
    my $hosed = 0;

    # If we're doing a modify,
    # and tbprerun sent the experiment to "NEW",
    # we're hosed.
    if ($inout eq "modify" && ExpState($pid,$eid) eq EXPTSTATE_NEW) {
	$hosed = 1;
    }

Chad Barb's avatar
Chad Barb committed
817
    if ($hosed) {
Chad Barb's avatar
 
Chad Barb committed
818 819 820 821 822 823 824 825 826 827
	#
	# Note: $estate is still set to the state which the experiment was in
	# when we began.
	#
	if ($estate eq EXPTSTATE_ACTIVE) {
	    print "Running 'tbswap out' with arguments: $pid $eid\n";
	    if (system("$tbdir/tbswap out -force $pid $eid") != 0) {
		print "tbswap out failed!\n";
	    }
	}
Chad Barb's avatar
Chad Barb committed
828

Chad Barb's avatar
 
Chad Barb committed
829 830 831 832 833 834
	print "Running tbend with arguments: -force $pid $eid\n";
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
	}
    }

835 836 837
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

838
    #
839 840
    # In batch mode, exit without sending the email or unlocking. The
    # batch daemon will take care of that and setting the proper state. 
841 842
    #
    if ($batch) {
843
	exit($errorstat);
844 845
    }

846
    #
Chad Barb's avatar
Chad Barb committed
847
    # Clear the log file so the web page stops spewing.
848 849 850 851 852
    #
    if (defined($logname)) {
	TBExptCloseLogFile($pid, $eid);
    }

853 854 855 856 857 858 859 860 861
    # Unlock and reset state to its original value. 
    TBUnLockExp($pid, $eid, $ebatchstate);

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

862 863 864 865
    #
    # Send a message to the testbed list. Append the logfile.
    #
    SENDMAIL("$user_name <$user_email>",
866
	     "Swap ${inout} Failure: $pid/$eid",
867
	     $mesg,
868
	     ($idleswap ? $TBOPS : "$user_name <$user_email>"),
869
	     "Cc:  $expt_head_name <$expt_head_email>\n".
Leigh B. Stoller's avatar
Leigh B. Stoller committed
870
	     "Cc:  $TBOPS",
871
	     (($logname), (defined($modnsfile) ? ($modnsfile) : ())));
872

Chad Barb's avatar
 
Chad Barb committed
873 874 875 876 877 878 879 880 881 882
    if ($hosed) {
	#
	# 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 ${userdir}-failed");
Chad Barb's avatar
Chad Barb committed
883
	TBExptDestroy($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
884 885
    }

886
    exit($errorstat);
887
}