tbswap.in 45.2 KB
Newer Older
1 2 3 4
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2010 University of Utah and the Flux Group.
6 7 8 9
# All rights reserved.
#
use English;

10 11 12
# Returns 0 on success.
# Returns 1 on non-assign_wrapper failure.
# Returns (1 | assign_wrapper's errorcode) on assign_wrapper failure.
13 14
# Returns |0x40 if update caused a swapout. Icky.
# Returns -1 on uncontrolled error (die called).
15

16 17 18 19 20
# XXX: handle error cases for update? (backup the db?)
# XXX: Shouldn't do idempotent stuff twice for update.
# XXX: repush/calc routing for update??? (tbprerun)
# XXX: previz for update???              (tbprerun)
# XXX: make snmpit faster for update.
Chad Barb's avatar
Chad Barb committed
21 22 23
#
# XXX: for update, expt is swapped out on os_setup fail.
#      (we only recover if assign fails)
24 25 26

sub usage()
{
27 28 29
    print STDERR
	"Usage: $0 {in|out|modify [-reboot] [-eventsys_restart]} ".
	"[-noswapout] [-genimode] pid eid\n";
30 31 32 33 34 35 36
    exit(-1);
}

#
# Configure variables
#
my $TBROOT         = "@prefix@";
37
my $TBOPS          = "@TBOPSEMAIL@";
Timothy Stack's avatar
Timothy Stack committed
38
my $TBLOGS         = "@TBLOGSEMAIL@";
39
my $MAINSITE	   = @TBMAINSITE@;
40
my $ELABINELAB     = @ELABINELAB@;
41
my $THISHOMEBASE   = "@THISHOMEBASE@";
42 43
my $TESTMODE       = @TESTMODE@;
my $DISABLE_EVENTS = "@DISABLE_EVENT_SCHED@";
44
my $piper          = "$TBROOT/sbin/locpiper";
45 46
my $mapper         = "$TBROOT/bin/mapper";
my $wrapper        = "$TBROOT/libexec/assign_wrapper";
Timothy Stack's avatar
Timothy Stack committed
47
my $NFSTRACESUPPORT= @NFSTRACESUPPORT@;
48
my $PGENISUPPORT   = @PROTOGENI_SUPPORT@;
49 50 51 52 53 54 55 56 57 58 59

# Untaint the path
$ENV{'PATH'} = "/usr/bin:$TBROOT/libexec:$TBROOT/libexec/ns2ir" . 
    ":$TBROOT/sbin:$TBROOT/bin";

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
60
use libadminctrl;
61
use libadminmfs;
62
use libtblog;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
63
use libArchive;
64
use Experiment;
65
use User;
66
use Lan;
67
use Firewall;
68 69 70
if ($PGENISUPPORT) {
    require libGeni;
}
71 72 73 74

#
# Actual swap-in and swap-out functions, defined below.
#
75 76 77
sub doSwapout($);
sub doSwapin($);

78
# XXX fixme: should not be hardwired!
79
my $cnetstack = "-S Control";
80 81 82
my $cnetvlanname = "Control";


83 84 85 86
sub REAL()    { return 5; }
sub CLEANUP() { return 4; }
sub RETRY()   { return 3; }
sub MODIFY()  { return 2; }
87
sub UPDATE()  { return 1; }
88
sub MODIFY_RECOVER()  { return 0; }
89

90 91 92
#
# Grab global enable of linkdelays.
#
93
my $enablelinkdelays = TBGetSiteVar("general/linux_endnodeshaping");
94

95 96 97 98 99 100
#
# Turn off line buffering on output
#

$| = 1;

101 102
my $updateReboot   = 0;
my $updateReconfig = 1;
103 104
my $update_Eventsys_restart = 0;
my $elabinelab     = 0;
105
my $plabinelab     = 0;
106 107 108 109 110
my $force          = 0;
my $noswapout      = 0;
my $genimode       = 0;
my $errors         = 0;
my $updatehosed    = 0;
111
my $state;
112
my $canceled;
113
my $os_setup_pid;
114
my $nextState;
115 116

#
117
# First argument is either "in", "out", or "update";
118 119 120 121 122
# this value goes into $swapop.
#

my $swapop = shift;	

123 124 125
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
126
     ($swapop ne "modify") &&
127
     ($swapop ne "update"))) {
128 129 130 131 132 133 134 135 136 137 138
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
139 140
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
141
	$updateReconfig = 0;
142 143 144
    } elsif ($arg eq "-noreconfig") {
	$updateReboot   = 0;
	$updateReconfig = 0;
145
    } elsif ($arg eq "-eventsys_restart" && $swapop eq "modify") {
146
	$update_Eventsys_restart = 1;
147
    } elsif ($arg eq "-noswapout") {
148
	$noswapout = 1;
149 150 151 152 153 154 155 156 157
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173
#
# 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");
}

174 175 176 177 178 179
#
# Set Error reporting info
# 
tblog_set_info($pid,$eid,$UID);

#
180
# Turn on timestamps
181
#
182 183
TBDebugTimeStampsOn();

184 185 186 187 188 189
#
# Verify user and get his DB uid and other info for later.
#
my $this_user = User->ThisUser();
if (! defined($this_user)) {
    tbdie("You ($UID) do not exist!");
190
}
191 192 193
my $user_uid   = $this_user->uid();
my $user_name  = $this_user->name();
my $user_email = $this_user->email();
194

195 196 197 198 199
# Slowly convert to using Experiment module.
my $experiment = Experiment->Lookup($pid, $eid);
if (!defined($experiment)) {
    tbdie("Could not lookup experiment object!")
}
200
my $special = ($pid eq "testbed" || $pid eq "tbres" || $pid eq "geni" ||
201
	       $pid eq "emulab-ops" || $pid eq "utahstud" || $pid eq "ResiliNets");
202
my $newsetup = $special;
203

204 205 206
#
# Print starting message.
#
207 208
my $exptidx;
TBExptIDX($pid, $eid, \$exptidx);
209

210 211
print "Beginning swap-$swapop for $pid/$eid ($exptidx). " .
    TBTimeStampWithDate() . "\n";
212 213 214 215 216 217
TBDebugTimeStamp("tbswap $swapop started");

#
# Get experiment state; verify that experiment exists.
#
if (! ($state = ExpState($pid, $eid))) {
218
    tbdie "No such experiment $pid/$eid";
219
}
220 221 222
# Sanity check the current state. 
if (!$force) {
    if ($swapop eq "in") {
223
	tbdie("Experiment should be ACTIVATING. Currently $state.")
224
	    if ($state ne EXPTSTATE_ACTIVATING);
225
    }
226
    elsif ($swapop eq "out") {
227
	tbdie("Experiment should be SWAPPING. Currently $state.")
228
	    if ($state ne EXPTSTATE_SWAPPING);
229
    }
230
    elsif ($swapop eq "modify" || $swapop eq "update") {
231
	tbdie("Experiment should be MODIFY_RESWAP. Currently $state.")
232
	    if ($state ne EXPTSTATE_MODIFY_RESWAP);
233 234
    }
}
235 236
# Get elabinelab status. See below.
if (! TBExptIsElabInElab($pid, $eid, \$elabinelab)) {
237
    tbdie("Could not get elabinelab status for experiment $pid/$eid");
238
}
239 240 241 242 243 244 245 246 247 248 249
# and plabinelab status.
if (! TBExptIsPlabInElab($pid, $eid, \$plabinelab)) {
    tbdie("Could not get plabinelab status for experiment $pid/$eid");
}
if ($elabinelab && $plabinelab) {
    tberror "Cannot get my head around Plab in Elab in Elab!\n";
    print "Failingly finished swap-$swapop for $pid/$eid. " .
	TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
    exit(1);
}
250

251 252 253 254 255
#
# See if the experiment is firewalled
#
my $firewalled = TBExptFirewall($pid, $eid);

256 257 258
#
# Do actual swapping
#
259 260 261 262
if ($swapop eq "out") {
    #
    # Swap out
    #
263
    tblog_set_attempt(1);
264
    $errors = doSwapout(REAL);
265
}
266
elsif ($swapop eq "update" || $swapop eq "modify") {
267 268 269 270 271 272
    #
    # Before swapout, do cursory admission control to see if the
    # modified experiment will be swappable. assign_wrapper does a
    # more stringent check using assign.
    #
    print STDERR "Checking with Admission Control ...\n";
273
    if (! TBAdmissionControlCheck(undef, $experiment, undef)) {
274 275 276
	tberror({type => 'secondary', severity => SEV_SECONDARY,
		 error => ['admission_control_failure']},
		"Admission control failure!\n");
277 278 279 280 281 282
	print "Failingly finished swap-$swapop for $pid/$eid. " .
	    TBTimeStamp() . "\n";
	TBDebugTimeStamp("tbswap $swapop finished (failed)");
	exit(1);
    }
    
283 284
    #
    # Update.
285 286 287
    #
    tblog_set_attempt(1);

288 289 290 291
    #
    # Phase One -- swap experiment partially out.
    #
    print STDERR "Backing up physical state...\n";
292
    $experiment->BackupPhysicalState();
293

294 295 296 297 298
    #
    # Actually, in update mode this is not done cause we are anticipating
    # adding nodes only.
    #
    $errors = ($swapop eq "modify" ? doSwapout(MODIFY) : 0);
Chad Barb's avatar
Chad Barb committed
299

300 301 302 303 304 305
    if (0) {
	print STDERR "Doing a swapmodswapout on the experiment archive ...\n";
	if (libArchive::TBExperimentArchiveSwapModSwapOut($pid, $eid) < 0) {
	    tberror("Failed swapmodswapout on the experiment archive!");
	    $errors = 1;
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
306 307
    }

308
    if ($errors) {
309
	#
310
	# Clean up the mess, leaving the experiment in the SWAPPED state,
311
	#
312
	if ($noswapout || $elabinelab) {
313 314 315 316 317 318
	    print STDERR "Leaving experiment swapped in as directed.\n";
	}
	else {
	    print STDERR "Cleaning up after errors.\n";
	    doSwapout(CLEANUP);
	}
319
	$updatehosed = 1;
320
    }
321
    else {
322
	#
323
	# Phase Two -- swap experiment back in.
324
	#
325
	$errors = doSwapin(MODIFY);
Chad Barb's avatar
Chad Barb committed
326

327
	if ($errors) {
Chad Barb's avatar
Chad Barb committed
328 329 330
	    #
	    # There were errors; see if we can recover.
	    #
331
	    my $CanRecover = 1;
Chad Barb's avatar
Chad Barb committed
332 333 334

	    if ($errors != 7) {
		print STDERR "Update failure occurred _after_ assign phase; ";
335
		$CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
336 337
	    }

338 339 340
	    if ($CanRecover) {
		print STDERR "Recovering virtual and physical state.\n";

341 342 343
		# It is safe to remove the phystate since we know it was
		# backed up above, and cause we do not know if assign_wrapper
		# made it to that point before it failed. 
344 345 346 347 348
		if ($experiment->RemoveVirtualState() ||
		    $experiment->RestoreVirtualState() ||
		    $experiment->ReserveSharedBandwidth(1, 1) ||
		    $experiment->RemovePhysicalState() ||
		    $experiment->RestorePhysicalState()) {
349 350
		    print STDERR "Could not restore backed-up state; ";
		    $CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
351
		}
352 353 354
		else {
		    print STDERR "Doing a recovery swap-in of old state.\n";

355
		    if (doSwapin(MODIFY_RECOVER)) {
356 357 358
			print STDERR "Could not swap in old physical state; ";
			$CanRecover = 0;
		    }
Chad Barb's avatar
Chad Barb committed
359 360
		}
	    }
361 362 363 364 365 366

	    #
	    # Some part of the recovery failed; must swap it out. swapexp
	    # (caller) will then have to do more clean up, hence the special
	    # exit status indicated by $updatehosed.
	    # 
367
	    if (!$CanRecover) {
368
		if ($noswapout || $elabinelab) {
369 370 371 372 373 374 375 376 377
		    print STDERR
			"No Recovery, but leaving experiment swapped in.\n";
		}
		else {
		    print STDERR
			"Recovery aborted! Swapping experiment out.\n";
		    doSwapout(CLEANUP);
		    $updatehosed = 1;
		}
378 379 380
	    }
	    else {
		print STDERR "Update recovery successful.\n";
Chad Barb's avatar
Chad Barb committed
381
	    }
382
	}
Chad Barb's avatar
Chad Barb committed
383 384
    }
}
385 386 387 388 389
elsif ($swapop eq "in") {
    #
    # Swap in
    #
    my $retries = 2;
390 391 392 393 394 395

    #
    # Before real swapin, do cursory admission control. assign_wrapper does
    # a more stringent check using assign.
    #
    print STDERR "Checking with Admission Control ...\n";
396
    if (! TBAdmissionControlCheck(undef, $experiment, undef)) {
397 398 399
	tberror({type => 'secondary', severity => SEV_SECONDARY,
		 error => ['admission_control_failure']},
		"Admission control failure!\n");
400 401 402 403 404
	print "Failingly finished swap-$swapop for $pid/$eid. " .
	    TBTimeStamp() . "\n";
	TBDebugTimeStamp("tbswap $swapop finished (failed)");
	exit(1);
    }
405 406

    tblog_set_attempt(1);
407
    $errors = doSwapin(REAL);
Chad Barb's avatar
Chad Barb committed
408

409 410 411 412 413 414 415 416 417 418
    #
    # Attempt a retry if: 
    #   a) there were errors, 
    #   b) doswapin() indicated (via return code 3) a retry is appropriate,
    #   c) we haven't tried too many times already.
    #   d) The cancelflag has not been set.
    #   e) $TESTMODE == 0.
    #
    while ($errors == 3 && $retries && !$canceled && !$TESTMODE) {
	$retries--;
419
	tblog_inc_attempt();
420

421 422
	print STDERR "Cleaning up after errors; will try again.\n";
	doSwapout(RETRY);
423

424 425
	print STDERR "Trying again...\n";
	$errors = doSwapin(RETRY);
426
    }
427 428
    if ($errors || $canceled) {
	print STDERR "Cleaning up after " .
429
	    ($canceled ? "cancelation" : "errors") . ".\n";
430
	doSwapout(CLEANUP);
431 432 433
    }
}

434 435 436 437
tblog_set_attempt(0);

TBGetCancelFlag($pid, $eid, \$canceled);
if ($canceled) {
438 439
    tberror ({type=>'summary', cause=>'canceled', 
	      severity=>SEV_IMMEDIATE, error=>['cancel_flag']},
440 441 442
	     "Experiment swap-$swapop canceled by user.");
}

443 444 445 446
#
# Write appropriate message and exit.
#
if ($errors) {
447
    print "Failingly finished swap-$swapop for $pid/$eid. ".TBTimeStamp()."\n";
448
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
Chad Barb's avatar
Chad Barb committed
449

450 451
    # Pass out magic value to indicate that update failed!
    exit(1 | ($updatehosed ? 0x40 : 0));
452
}
453
print "Successfully finished swap-$swapop for $pid/$eid. " .TBTimeStamp()."\n";
454 455
TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
exit(0);
456 457 458 459

#################################

##
Chad Barb's avatar
Chad Barb committed
460
#
461 462
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
463
#             If in REAL or CLEANUP,
464 465 466
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
467
#             If in RETRY or UDPATE,
468 469 470 471 472 473
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

474
sub doSwapout($) {
475
    my $type = shift;
476 477
    my $swapout_errors = 0;

478 479
    tblog_set_cleanup(1) if $type == CLEANUP;

480 481 482 483 484 485 486 487 488
    #
    # wait for os_setup;
    # this only applies if called after a failed doswapin.
    #
    if ($os_setup_pid) {
	print "Waiting for os_setup to finish\n";
	waitpid($os_setup_pid, 0);
	undef $os_setup_pid;
    }
Chad Barb's avatar
Chad Barb committed
489

490
    if (0 && $NFSTRACESUPPORT && $type == REAL) {
Timothy Stack's avatar
Timothy Stack committed
491 492
	print "Getting files accessed via NFS.\n";
	TBDebugTimeStamp("nfstrace started");
493
	system("nfstrace transfer $pid $eid");
Timothy Stack's avatar
Timothy Stack committed
494
	TBDebugTimeStamp("nfstrace finished");
495
    }
496

497
    if (! $TESTMODE) { 
498 499
	if (! ($DISABLE_EVENTS || $elabinelab)) {
	    if ($type >= RETRY ||
500
		($update_Eventsys_restart && $type == MODIFY) ) {
501
		print "Stopping the event system\n";
502
		if (system("eventsys_control stop $pid,$eid")) {
503 504 505
		    tberror({type => 'secondary', severity => SEV_SECONDARY,
			     error => ['eventsys_stop_failed']},
			    "Failed to stop the event system.");
506 507
		    $swapout_errors = 1;
		}
508 509 510 511 512 513 514

		#
		# Stop the location piper.
		#
		if (-x $piper) {
		    print "Stopping the location piper\n";
		    if (system("$piper -k $pid $eid")) {
515 516 517
			tberror({type => 'secondary', severity => SEV_SECONDARY,
				 error => ['piper_stop_failed']},
				"Failed to stop location piper.");
518 519 520
			$swapout_errors = 1;
		    }
		}
521 522
	    }
	}
523 524 525 526 527 528
	
	#
	# Do teardown of inner elab. We must do this before we teardown the
	# vlans since the inner control network is a vlan, and we want that
	# active so inner boss can reboot the inner nodes (avoid power cycle).
	#
529
	if ($elabinelab && $type >= CLEANUP) {
530 531
	    print "Tearing down elabinelab. This could take a while.\n";
	    if (system("elabinelab -k $pid $eid")) {
532 533 534
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['elabinelab_tear_down_failed']},
			"Failed to teardown elabinelab!");
535 536 537 538
		$swapout_errors = 1;
	    }
	}

539
	#
Chad Barb's avatar
Chad Barb committed
540
	# Clean up any VLANs in experiment.
541
	#
542
	# When modifying an elabinelab experiment, leave the vlans intact
543
	# so that the inner networks are not suddenly disconnected.
544
	#
545 546
	if ($type != MODIFY ||
	    ($type == MODIFY && $ELABINELAB && Lan->GotTrunks($experiment))) {
547 548 549
	    TBDebugTimeStamp("snmpit started");
	    print STDERR "Removing VLANs.\n";
	    if (system("snmpit -r $pid $eid")) {
550 551 552
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['vlan_reset_failed']},
			"Failed to reset VLANs");
553 554 555
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("snmpit finished");
556
	}
557 558 559 560 561 562 563 564 565 566 567 568 569 570
	#
	# Must check for stale vlans that we kept around in the above clause
	# since they will not be in the lans table anymore.
	#
	if ($type == CLEANUP) {
	    my @stale;
	    if (VLan->StaleVlanList($experiment, \@stale) != 0) {
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['vlan_reset_failed']},
			"Failed to get stale VLANs");
		$swapout_errors = 1;
	    }
	    if (@stale) {
		print "Removing stale vlans @stale\n";
571
		system("snmpit -f ". join(" ", map("-o $_", @stale)));
572 573 574 575 576 577 578 579
		if ($?) {
		    tberror({type => 'summary', severity => SEV_SECONDARY,
			     error => ['vlan_reset_failed']},
			    "Failed to remove stale vlans");
		    $swapout_errors = 1;
		}
	    }
	}
580
    }
581
	
582
    if ($type >= MODIFY) {
583 584
	# XXX Do this before releasing the nodes. Needs more thought.
	$experiment->DeleteInternalProgramAgents();
585 586 587
    }

    if ($type >= CLEANUP) {
588 589 590 591 592 593 594 595
	#
	# Undo plab in elab specialness.
	# No need to worry about VLANs here, as all the special abilities
	# involve the control network.
	#
	if (! $TESTMODE && $plabinelab) {
	    print "Tearing down plabinelab.\n";
	    if (system("plabinelab -k $pid $eid")) {
596 597 598
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['plabinelab_tear_down_failed']},
			"Failed to teardown plabinelab!");
599 600 601 602
		$swapout_errors = 1;
	    }
	}

603 604
	#
	# We're not attempting a retry;
605
	#
606
	# Stop all of the vnodes.
607
	#
608
	if (! $TESTMODE) { 	
609 610 611
	    print "Tearing down virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid")) {
612 613 614
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['vnode_tear_down_failed']},
			"Failed to tear down vnodes.");
615 616 617
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup finished");
618 619
	}

620 621 622 623
	#
	# Nodes behind a firewall are treated special.
	# See undoFWNodes for details.
	#
624
	if ($firewalled && undoFWNodes($experiment)) {
625
	    tblog_set_cleanup(0);
626 627
	    return 1;
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
628

629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648
	#
	# Perform swapout time admin actions.  Right now there is at most
	# one of these.  It isn't really a general mechanism, just a hook
	# for state saving or data collection during swapout.
	# A couple of important "fer now" notes:
	#
	#	We don't do this for firewalled experiments.  We need a way
	#	to "tag" the saved disk state to ensure it doesn't get
	#	instantiated outside of a firewall.
	#
	#	We only do this on REAL swapouts, and not on CLEANUPs.
	#	There are some types of CLEANUPs where we may want to
	#	do this, in particular an invocation caused by a failed
	#	modify operation, where the admin action is to save the
	#	experiment state.  So we will need to revisit this.
	#
	my %soaction = ();
	if ($type == REAL && !$firewalled) {
	    TBExptGetSwapoutAction($pid, $eid, \%soaction);
	}
649
	if ($soaction{'command'} && doSwapoutAction($pid, $eid, %soaction)) {
650
	    tblog_set_cleanup(0);
651
	    return 1;
652 653
	}

654 655 656 657 658 659 660 661 662 663
	#
	# Clear away Geni slivers. Do not proceed if this fails.
	#
	if ($PGENISUPPORT) {
	    if (libGeni::DeleteAllSlivers($experiment, $this_user)) {
		tberror("Could not delete slivers\n");
		return 1;
	    }
	}

664 665
	#
	# remove all nodes from the experiment.
Chad Barb's avatar
Chad Barb committed
666
	# (nfree will send them to RES_FREE_DIRTY)
667 668 669
	#
	print STDERR "Freeing nodes.\n";
	TBDebugTimeStamp("nfree started");
670
	if (system("nfree -a $pid $eid")) {
671 672 673
	    tberror({type => 'secondary', severity => SEV_SECONDARY,
		     error => ['nfree_failed']},
		    "Could not free nodes.");
674 675 676
	    $swapout_errors = 1;
	}
	TBDebugTimeStamp("nfree finished");
677 678 679 680 681

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
682 683
	DBQueryWarn("update experiments set swap_requests='', ".
		    "   sim_reswap_count='0' ".
684
		    "where eid='$eid' and pid='$pid'");
685 686
    } else {
	#
687
	# $type == RETRY or $type == MODIFY.
688 689
	# Therefore, don't deallocate nodes which have been successfully
	# incorporated into the experiment (i.e., are RES_READY).
Chad Barb's avatar
Chad Barb committed
690
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
691
	#
692 693 694 695 696 697 698 699
	my @failedpnodes = ();
	my @failedvnodes = ();
	
	my $db_result =
	    DBQueryFatal("select rv.node_id,n.allocstate,nt.isvirtnode ".
                         "  from reserved as rv ".
			 "left join nodes as n on n.node_id = rv.node_id ".
			 "left join node_types as nt on nt.type=n.type ".
700 701
			 "where rv.pid='$pid' and rv.eid='$eid' and ".
			 "      rv.genisliver_idx is null");
702

703
	while (my ($node,$allocstate,$isvirt) = $db_result->fetchrow_array) {
704
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
705 706 707 708 709 710
		if ($isvirt) {
		    push(@failedvnodes, $node);
		}
		else {
		    push(@failedpnodes, $node);
		}
711 712 713
	    }
	}

714 715 716 717 718 719 720
	#
	# Tear down failed vnodes. Perhaps not needed?
	# 
	if (!$TESTMODE && @failedvnodes > 0) {
	    print "Tearing down failed virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid @failedvnodes")) {
721 722 723
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['vnode_tear_down_failed']},
			"Failed to tear down vnodes.");
724 725 726 727
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup -k finished");
	}
728

729 730 731 732 733 734 735 736 737 738 739
	#
	# Undo plabinelab setup
	#
	if (!$TESTMODE && $plabinelab && @failedpnodes > 0) {
	    print "Removing failed nodes from inner plab.\n";
	    if (system("plabinelab -r $pid $eid @failedpnodes")) {
		tberror "Failed to remove inner nodes!";
		$swapout_errors = 1;
	    }
	}

740 741 742 743
	#
	# Release all failed nodes.
	# 
	if (@failedpnodes > 0 || @failedvnodes > 0) {
744 745
	    print STDERR "Freeing failed nodes.\n";
	    
746
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
Chad Barb committed
747 748 749 750 751
	    #
	    # Specify -x switch so when a physical node gets freed,
	    # any virtual nodes (owned by this experiment)
	    # sitting on top of it are freed as well.
	    #
752 753
	    if (system("nfree -x $pid $eid " .
		       join(" ", (@failedpnodes, @failedvnodes)))) {
754 755 756
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['nfree_failed']},
			"Could not free nodes.");
757 758 759 760 761 762
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("nfree finished");
	}
    }

763 764 765 766 767 768 769 770 771 772 773 774
    if (! $TESTMODE) {
	#
	# If the experiment has no Plab dslice nodes left, but still has
	# a Plab slice, destroy the slice
	#

	# Does the slice exist?
	$db_result =
	    DBQueryFatal("select slicename from plab_slices ".
			 "where pid='$pid' and eid='$eid'");

	if ($db_result->numrows) {
775
	    # Are there any nodes left in the slice?
776
	    $db_result =
777 778 779 780 781 782 783 784 785
		DBQueryFatal("select r.node_id ".
			     " from plab_slices as ps".
			     " left join plab_slice_nodes as psn ".
			     "   on (ps.slicename=psn.slicename ".
			     "       and ps.plc_idx=psn.plc_idx) ".
			     " left join reserved as r ".
			     "   on psn.node_id=r.node_id ".
			     " where ps.pid='$pid' and ps.eid='$eid'".
			     "   and r.node_id is not NULL");
786 787

	    if (!$db_result->numrows) {
788
		print "Tearing down Slices.\n";
789
		TBDebugTimeStamp("plabslice destroy started");
790
		if (system("plabslice destroy $pid $eid")) {
791
		    tberror "Failed to tear down Slices.";
792 793 794 795 796 797 798
		    $swapout_errors = 1;
		}
		TBDebugTimeStamp("plabslice destroy finished");
	    }
	}
    }

799 800 801 802 803 804 805 806
    if (! $TESTMODE) {
	#
	# All of these errors are non-fatal on swapout. We find out about them
	# via email sent from the individual scripts.
	#

	#
	# Only reset mountpoints if this is an actual swapout, and
807
	# not a failed swapin(cleanup), update, or retry.
808
	#
809
	if ($type == REAL) {
810 811 812
	    print "Resetting mountpoints.\n";
	    TBDebugTimeStamp("exports started");
	    if (system("exports_setup")) {
813 814 815
		tberror({severity => SEV_WARNING,
			 error => ['mountpoint_reset_failed']},
			"Failed to reset mountpoints.");
816 817
	    }
	    TBDebugTimeStamp("exports finished");
Chad Barb's avatar
Chad Barb committed
818

819 820 821 822 823 824 825 826 827
	    #
	    # Ditto these two. 
	    #
	    print "Resetting named maps.\n";
	    TBDebugTimeStamp("named started");
	    if (system("named_setup")) {
		tbwarn "Failed to reset named map.";
	    }
	    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
828

829 830 831 832 833 834
	    print "Resetting email lists.\n";
	    TBDebugTimeStamp("genelists started");
	    if (system("genelists -t")) {
		tbwarn "Failed to reset email lists.";
	    }
	    TBDebugTimeStamp("genelists finished");
835 836 837
	}
    }

838
    #
839
    # Wipe the DB clean except during MODIFY or RETRY. In those
840 841
    #    cases, assign_wrapper will reset the DB after reading
    #    the info.
842
    #
843 844
    if ( $type >= CLEANUP ) {
	print STDERR "Resetting DB.\n";
845 846 847
	$experiment->RemovePhysicalState();
	# Special. 
	$experiment->ClearPortRegistration();
848
    }
849

850
    tblog_set_cleanup(0);
851 852 853 854
    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
855
#
856 857
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
858 859 860 861 862
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
863 864
##

865
sub doSwapin($) {
866
    my $type = shift; 
867 868
    # Just the physnodes ...
    my @deleted_pnodes = ();
869

870 871 872 873
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #
874
    
875
    if ($type > MODIFY_RECOVER) {
876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
        #
        # Hacky test to allow disabling of linkdelays if the node is going
        # to run Linux. See sitevar above.
        #
	if (! $enablelinkdelays) {
	    $db_result =
		DBQueryFatal("select distinct e.pid,e.eid,vl.vnode,vn.osname ".
			 "  from experiments as e ".
			 "left join virt_lans as vl on vl.pid=e.pid and ".
			 "     vl.eid=e.eid ".
			 "left join virt_nodes as vn on vn.pid=e.pid and ".
			 "     vn.eid=e.eid and vn.vname=vl.vnode ".
			 "left join os_info as o on o.osname=vn.osname and ".
			 "  (o.pid=vl.pid or o.pid='" . TBOPSPID() . "') ".
			 "where (vl.uselinkdelay!=0 or e.uselinkdelays!=0 or ".
			 "       e.forcelinkdelays!=0) and ".
892 893
			 "     (o.os is NULL or o.os='Linux' or ".
			 "      o.os='Fedora') and ".
894 895 896
			 "     e.pid='$pid' and e.eid='$eid'");

	    if ($db_result->numrows) {
897 898
		tberror "Endnodeshaping is disabled on Linux Images!";
		tberror "You must modify your experiment to swap it in.";
899 900 901 902
		return 1;
	    }
	}
	
Chad Barb's avatar
Chad Barb committed
903
	print "Mapping to physical reality ...\n";
904
	TBDebugTimeStamp("mapper wrapper started");
Chad Barb's avatar
Chad Barb committed
905 906

	#
907 908 909 910
	# Pass the -u (update) switch into assign_wrapper, which turns on
	# update mode. When doing a retry, must also fix the current nodes
	# to avoid stuff jumping around when simply trying to replace a node
	# that did not boot.
Chad Barb's avatar
Chad Barb committed
911 912
	#
	my $exitcode;
913
	my $cmd = "$wrapper";
914
	if ($special || $this_user->stud() || $experiment->virtnode_count()) {
915 916 917
	    $cmd = "$mapper";
	}
	$cmd .= " -f"
918
	    if ($type == RETRY);
919
	
920
	if (system("$cmd -u $pid $eid")) {
921 922
	    $exitcode = $? >> 8;

923
	    tberror "Failed ($exitcode) to map to reality.";
924

925 926
	    # Wrapper sets this bit when recovery is possible.
	    if ($exitcode & 64) {
927
		# We can recover. 
928
		tbreport(SEV_SECONDARY, 'assign_wrapper_failed', $exitcode);
Chad Barb's avatar
Chad Barb committed
929
		return 7;
930 931
	    }
	    else {
932
		# No recovery, no retry.
933
		tbreport(SEV_SECONDARY, 'assign_wrapper_failed', $exitcode);
Chad Barb's avatar
Chad Barb committed
934 935 936
		return 1;
	    }
	}
937
	TBDebugTimeStamp("mapper wrapper finished");
938

Chad Barb's avatar
Chad Barb committed
939 940
	print "Mapped to physical reality!\n";
    }
941

942
    # Check cancel flag before continuing. No retry, 
943
    TBGetCancelFlag($pid, $eid, \$canceled);
944 945 946 947
    if ($canceled) {
	tbreport(SEV_IMMEDIATE, 'cancel_flag');
	return 1
    }
948

949 950 951 952 953 954 955
    #
    # When doing a modify, we have to compare vlans to determine which
    # vlans actually changed and need to be deleted, before processing
    # the new vlans for the experiment. Note that vlans that already
    # exist on the switches will be left as is by snmpit.
    #
    # We must do this before the nfree of deleted nodes in the next section
956
    # because the new test in nfree that prevents nodes from accidentally
957 958 959 960 961
    # getting released when they are in a vlan.
    #
    if ($type == MODIFY) {
	my @diff = ();
	my @same = ();
962 963 964 965 966 967 968 969 970
	my $ret;

	if ($ELABINELAB || !$MAINSITE || !$special) {
	    $ret = Lan->CompareVlansWithSwitches($experiment, \@diff, \@same);
	}
	else {
	    $ret = Lan->CompareVlansWithSwitches2($experiment);
	}
	if ($ret) {
971 972 973 974 975
	    tberror({type => 'summary', severity => SEV_SECONDARY,
		     error => ['vlan_setup_failed']},
		    "Failed to compare old vlans");
	    return 1;
	}
976 977 978 979 980 981 982 983 984 985 986 987 988 989 990
	if ($ELABINELAB || !$MAINSITE || !$special) {
	    if (@diff) {
		print "Removing obsolete vlans @diff\n";
		system("snmpit -f ". join(" ", map("-o $_", @diff)));
		if ($?) {
		    tberror({type => 'summary', severity => SEV_SECONDARY,
			     error => ['vlan_setup_failed']},
			    "Failed to remove obsolete VLANs.");
		    return 1;
		}
	    }
	}
	else {
	    print "Synchronizing VLANs.\n";
	    system("snmpit -X $pid $eid");
991 992 993
	    if ($?) {
		tberror({type => 'summary', severity => SEV_SECONDARY,
			 error => ['vlan_setup_failed']},
994
			"Failed to synchronize VLANs.");
995 996 997 998 999
		return 1;
	    }
	}
    }

1000 1001 1002 1003 1004 1005 1006 1007 1008
    #
    # Look for any nodes in RES_TEARDOWN. These need to be released,
    # and if a virtnode, they need to be torn down. We cannot wait for
    # the virtnodes to go down with the physnode they are hosted on,
    # so teardown and release the virtnodes first, and then do the
    # physnodes.
    #
    # Errors are fatal; no recovery or retry.
    #
1009
    if ($type == MODIFY || $type == UPDATE) {
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
	my $allocstate = TBDB_ALLOCSTATE_RES_TEARDOWN();
	
	$db_result =
	    DBQueryFatal("select r.node_id,nt.isvirtnode,nt.isremotenode ".
			 "  from reserved as r ".
			 "left join nodes as n on n.node_id=r.node_id ".
			 "left join node_types as nt on nt.type=n.type ".
			 "where r.pid='$pid' and r.eid='$eid' and ".
			 "      n.allocstate='$allocstate'");

	if ($db_result->numrows) {
	    my @virtnodes = ();
	    my @physnodes = ();
	    
	    print "Tearing down and releasing unused nodes\n";

	    # First teardown/release virtnodes. 
	    while (my ($node,$isvirt,$isrem) = $db_result->fetchrow_array()) {
		if ($isvirt) {
		    push(@virtnodes, $node);
		}
		elsif (!$isrem) {
		    push(@physnodes, $node);
		}
	    }
1035 1036 1037
	    # See below.
	    @deleted_pnodes = @physnodes;
	    
1038 1039 1040
	    if (@virtnodes) {
		TBDebugTimeStamp("vnode_setup started");
		
1041
		if (system("vnode_setup -k $pid $eid @virtnodes")) {
1042
		    tberror "Failed to tear down unused virtnodes!\n";
1043 1044 1045 1046 1047
		    return 1;
		}
		TBDebugTimeStamp("vnode_setup finished");
		
		if (system("nfree $pid $eid @virtnodes")) {
1048
		    tberror "Failed to nfree unused virtnodes!\n";
1049 1050 1051 1052
		    return 1;
		}
	    }
	    if (@physnodes) {
1053 1054 1055
		if ($elabinelab) {
		    print "Removing nodes from inner elab.\n";
		    if (system("elabinelab -r $pid $eid @physnodes")) {
1056
			tberror "Failed to remove inner nodes!";
1057 1058
			return 1;
		    }
1059 1060 1061 1062 1063 1064
		} elsif ($plabinelab) {
		    print "Removing nodes from inner plab.\n";
		    if (system("plabinelab -r $pid $eid @physnodes")) {
			tberror "Failed to remove inner nodes!";
			return 1;
		    }
1065 1066 1067
		}

		#
1068 1069
		# If the experiment is firewalled, cleanup the nodes
		# we are releasing.
1070
		# 
1071 1072
		if ($firewalled && undoFWNodes($experiment, 1,
					       @deleted_pnodes)) {
1073 1074 1075
		    return 1;
		}
		
1076
		if (system("nfree $pid $eid @physnodes")) {
1077
		    tberror "Failed to nfree unused physnodes!\n";
1078 1079 1080 1081 1082 1083
		    return 1;
		}
	    }
	}
    }

1084 1085 1086 1087 1088 1089
    # Exit here if we are testing.
    if ($TESTMODE) {
	print "Testing run - Stopping here.\n";
	return 0;
    }

1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100
    #
    # Handle tarballs - we might need to fetch some from URLs if the user
    # asked for that.
    #
    print "Fetching tarballs and RPMs (if any) ...\n";
    TBDebugTimeStamp("tarfiles_setup started");

    if (system("tarfiles_setup $pid $eid")) {
	#
	# No recovery for now - what would we do?
	#
1101 1102 1103
	tberror({type => 'secondary', severity => SEV_SECONDARY,
		 error => ['tarfiles_setup_failed']},
		"Failed to set up tarballs.");
1104 1105 1106 1107
	return 1;
    }
    TBDebugTimeStamp("tarfiles_setup finished");

1108 1109 1110 1111
    #
    # If there are any Plab dslice nodes in the experiment, create the
    # dslice now
    #
1112
    if ($type > MODIFY_RECOVER) {
1113 1114 1115
	# Are there any Plab nodes?  First get a list of node types in the exp;
	# if any are types hosted by any of the PLCs we know about, create
	# all slices necessary for the experiment in a single plabslice call.
1116
	$db_result =
1117 1118 1119 1120 1121 1122 1123 1124
	    DBQueryFatal("select nt.type,ppi.plc_name from nodes as n ".
			 "  left join node_types as nt on n.type = nt.type ".
			 "  left join reserved as r on r.node_id = n.node_id ".
			 "  left join plab_plc_info as ppi ".
			 "    on nt.type = ppi.node_type".
			 " where r.pid='$pid' and r.eid='$eid'".
			 "    and ppi.node_type is not NULL".
			 " group by nt.type");
1125 1126

	if ($db_result->numrows) {
1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137
	    my $info_str = "";
	    while (my ($nt,$plc) = $db_result->fetchrow_array()) {
		$info_str .= " $plc($nt)";
	    }
	    print "Configuring Slices for $info_str.\n";
	    TBDebugTimeStamp("plabslice started");
	    if (system("plabslice create $pid $eid")) {
		tberror({type => 'secondary', severity => SEV_SECONDARY,
			 error => ['plabslice_create_failed']},
			"Failed to configure Slices.");
		return 3;
1138
	    }
1139
	    TBDebugTimeStamp("plabslice finished");
1140 1141 1142
	}
    }

1143
    # Check cancel flag before continuing. No retry, 
1144
    TBGetCancelFlag($pid, $eid, \$canceled);
1145 1146 1147 1148
    if ($canceled) {
	tbreport(SEV_IMMEDIATE, 'cancel_flag');
	return 1
    }
1149

1150 1151 1152 1153 1154 1155 1156 1157 1158
    #
    # These things need to get started before the nodes come up, so we'll
    # do them before the os_setup. Everything else can done in parallel with
    # os_setup. (Actually, these probably can too, since they should finish
    # long before the nodes reboot, but better safe than sorry)
    #
    print "Setting up mountpoints.\n";
    TBDebugTimeStamp("mountpoints started");
    if (system("exports_setup")) {
1159 1160 1161
	tberror({type => 'secondary', severity => SEV_SECONDARY,
		 error => ['mountpoints_setup_failed']},
		"Failed to setup mountpoints.");
1162 1163 1164
	return 1;
    }
    TBDebugTimeStamp("mountpoints finished");
Chad Barb's avatar
Chad Barb committed
1165

1166 1167 1168
    TBDebugTimeStamp("named started");
    print "Setting up named maps.\n";
    if (system("named_setup")) {
1169
	tbwarn "Failed to add node names to named map.";
1170 1171
	#
	# This is a non-fatal error.
Chad Barb's avatar
Chad Barb committed
1172
	#
1173 1174
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
1175

Timothy Stack's avatar
Timothy Stack committed
1176 1177 1178 1179
    if ($NFSTRACESUPPORT) {
	print "Cleaning NFS traces.\n";
	TBDebugTimeStamp("nfstrace gc started");
	if (system("nfstrace gc $pid $eid")) {
1180 1181 1182
	    tberror({type => 'secondary', severity => SEV_SECONDARY,
		     error => ['nfstrace_setup_failed']},
		    "Failed to setup nfstrace.");
Timothy Stack's avatar
Timothy Stack committed
1183 1184 1185
	    return 1;
	}
	TBDebugTimeStamp("nfstrace gc finished");