tbswap.in 36.1 KB
Newer Older
Chad Barb's avatar
 
Chad Barb committed
1 2 3 4
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2005 University of Utah and the Flux Group.
Chad Barb's avatar
 
Chad Barb committed
6 7 8 9
# All rights reserved.
#
use English;

Chad Barb's avatar
 
Chad Barb committed
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).
Chad Barb's avatar
 
Chad Barb committed
15

Chad Barb's avatar
 
Chad Barb committed
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)
Chad Barb's avatar
 
Chad Barb committed
24 25 26

sub usage()
{
27
    print STDERR "Usage: $0 -force { in | out | update [-reboot] [-eventsys_restart] } pid eid\n";
Chad Barb's avatar
 
Chad Barb committed
28 29 30 31 32 33 34 35 36
    exit(-1);
}

#
# Configure variables
#
my $TBROOT         = "@prefix@";
my $TESTMODE       = @TESTMODE@;
my $DISABLE_EVENTS = "@DISABLE_EVENT_SCHED@";
37
my $piper          = "$TBROOT/sbin/locpiper";
Chad Barb's avatar
 
Chad Barb committed
38 39 40 41 42 43 44 45 46 47 48

# 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;
49
use libadminctrl;
50
#require exitonwarn; # exitonwarn isn't really a module, so just require it
Chad Barb's avatar
 
Chad Barb committed
51 52 53 54

#
# Actual swap-in and swap-out functions, defined below.
#
Chad Barb's avatar
 
Chad Barb committed
55 56 57
sub doSwapout($);
sub doSwapin($);

58 59 60 61
#
# Firewall stuff
# XXX maybe should be elsewhere
#
62 63 64 65 66
sub FWSETUP()     { return 1; }
sub FWADDNODES()  { return 2; }
sub FWDELNODES()  { return 3; }
sub FWTEARDOWN()  { return 4; }
sub doFW($$$$);
67 68

# XXX fixme: should not be hardwired!
69
my $cnetstack = "-S Control";
70 71 72
my $cnetvlanname = "Control";


Chad Barb's avatar
 
Chad Barb committed
73 74 75 76
sub REAL()    { return 4; }
sub CLEANUP() { return 3; }
sub RETRY()   { return 2; }
sub UPDATE()  { return 1; }
Chad Barb's avatar
Chad Barb committed
77
sub UPDATE_RECOVER()  { return 0; }
Chad Barb's avatar
 
Chad Barb committed
78 79 80 81 82 83 84

#
# Turn off line buffering on output
#

$| = 1;

85 86
my $updateReboot   = 0;
my $updateReconfig = 1;
87 88
my $update_Eventsys_restart = 0;
my $elabinelab     = 0;
Chad Barb's avatar
 
Chad Barb committed
89 90
my $force  = 0;
my $errors = 0;
91
my $updatehosed = 0;
Chad Barb's avatar
 
Chad Barb committed
92
my $state;
93
my $canceled;
Chad Barb's avatar
 
Chad Barb committed
94 95
my $os_setup_pid;
my $cleanvlans;
96
my $nextState;
Chad Barb's avatar
 
Chad Barb committed
97 98

#
Chad Barb's avatar
 
Chad Barb committed
99
# First argument is either "in", "out", or "update";
Chad Barb's avatar
 
Chad Barb committed
100 101 102 103 104
# this value goes into $swapop.
#

my $swapop = shift;	

Chad Barb's avatar
 
Chad Barb committed
105 106 107 108
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
Chad Barb's avatar
 
Chad Barb committed
109 110 111 112 113 114 115 116 117 118 119
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
120 121
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
122
	$updateReconfig = 0;
123 124 125
    } elsif ($arg eq "-noreconfig") {
	$updateReboot   = 0;
	$updateReconfig = 0;
126 127
    } elsif ($arg eq "-eventsys_restart" && $swapop eq "update") {
	$update_Eventsys_restart = 1;
Chad Barb's avatar
 
Chad Barb committed
128 129 130 131 132 133 134 135 136
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152
#
# 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");
}

Chad Barb's avatar
 
Chad Barb committed
153 154 155 156 157 158 159 160 161 162 163 164 165
TBDebugTimeStampsOn();

#
# Print starting message.
#

print "Beginning swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
TBDebugTimeStamp("tbswap $swapop started");

#
# Get experiment state; verify that experiment exists.
#
if (! ($state = ExpState($pid, $eid))) {
166 167
    die("*** $0:\n".
	"    No such experiment $pid/$eid\n");
Chad Barb's avatar
 
Chad Barb committed
168
}
169 170 171 172 173 174
# Sanity check the current state. 
if (!$force) {
    if ($swapop eq "in") {
	die("*** $0:\n".
	    "    Experiment should be ACTIVATING. Currently $state.\n")
	    if ($state ne EXPTSTATE_ACTIVATING);
Chad Barb's avatar
 
Chad Barb committed
175
    }
176 177 178 179
    elsif ($swapop eq "out") {
	die("*** $0:\n".
	    "    Experiment should be SWAPPING. Currently $state.\n")
	    if ($state ne EXPTSTATE_SWAPPING);
Chad Barb's avatar
 
Chad Barb committed
180
    }
181 182 183 184
    elsif ($swapop eq "update") {
	die("*** $0:\n".
	    "    Experiment should be MODIFY_RESWAP. Currently $state.\n")
	    if ($state ne EXPTSTATE_MODIFY_RESWAP);
Chad Barb's avatar
 
Chad Barb committed
185 186
    }
}
187 188 189 190 191
# Get elabinelab status. See below.
if (! TBExptIsElabInElab($pid, $eid, \$elabinelab)) {
    die("*** $0:\n".
	"    Could not get elabinelab status for experiment $pid/$eid\n");
}
Chad Barb's avatar
 
Chad Barb committed
192

193 194 195 196 197
#
# See if the experiment is firewalled
#
my $firewalled = TBExptFirewall($pid, $eid);

Chad Barb's avatar
 
Chad Barb committed
198 199 200
#
# Do actual swapping
#
201 202 203 204 205
if ($swapop eq "out") {
    #
    # Swap out
    #
    $errors = doSwapout(REAL);
Chad Barb's avatar
 
Chad Barb committed
206
}
207
elsif ($swapop eq "update") {
208 209 210 211 212 213 214 215 216 217 218 219 220 221
    #
    # Before swapout, do cursory admission control to see if the
    # modified experiment will be swappable. assign_wrapper does a
    # more stringent check using assign.
    #
    print STDERR "Checking with Admission Control ...\n";
    if (! TBAdmissionControlCheck($UID, $pid, $eid, undef)) {
	print STDERR "Admission control failure!\n";
	print "Failingly finished swap-$swapop for $pid/$eid. " .
	    TBTimeStamp() . "\n";
	TBDebugTimeStamp("tbswap $swapop finished (failed)");
	exit(1);
    }
    
222 223 224 225 226 227 228
    #
    # Update.
    #
    # Phase One -- swap experiment partially out.
    #
    print STDERR "Backing up physical state...\n";
    TBExptBackupPhysicalState($pid,$eid);
Chad Barb's avatar
 
Chad Barb committed
229

230
    $errors = doSwapout(UPDATE);
Chad Barb's avatar
Chad Barb committed
231

232
    if ($errors) {
Chad Barb's avatar
 
Chad Barb committed
233
	#
234 235 236 237 238
	# Clean up the mess, leaving the experiment in the SWAPPED state,
	# 
	print STDERR "Cleaning up after errors.\n";
	doSwapout(CLEANUP);
	$updatehosed = 1;
Chad Barb's avatar
 
Chad Barb committed
239
    }
240
    else {
Chad Barb's avatar
 
Chad Barb committed
241
	#
242
	# Phase Two -- swap experiment back in.
Chad Barb's avatar
 
Chad Barb committed
243
	#
Chad Barb's avatar
Chad Barb committed
244 245
	$errors = doSwapin(UPDATE);

246
	if ($errors) {
Chad Barb's avatar
Chad Barb committed
247 248 249
	    #
	    # There were errors; see if we can recover.
	    #
250
	    my $CanRecover = 1;
Chad Barb's avatar
Chad Barb committed
251 252 253

	    if ($errors != 7) {
		print STDERR "Update failure occurred _after_ assign phase; ";
254
		$CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
255 256
	    }

257 258 259 260 261 262 263 264
	    if ($CanRecover) {
		print STDERR "Recovering virtual and physical state.\n";

		if (TBExptRemoveVirtualState($pid, $eid) ||
		    TBExptRestoreVirtualState($pid, $eid) ||
		    TBExptRestorePhysicalState($pid,$eid)) {
		    print STDERR "Could not restore backed-up state; ";
		    $CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
265
		}
266 267 268 269 270 271 272
		else {
		    print STDERR "Doing a recovery swap-in of old state.\n";

		    if (doSwapin(UPDATE_RECOVER)) {
			print STDERR "Could not swap in old physical state; ";
			$CanRecover = 0;
		    }
Chad Barb's avatar
Chad Barb committed
273 274
		}
	    }
275 276 277 278 279 280 281

	    #
	    # 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.
	    # 
	    if (! $CanRecover) {
282
		print STDERR "Recovery aborted! Swapping experiment out.\n";
Chad Barb's avatar
Chad Barb committed
283
		doSwapout(CLEANUP);
284 285 286 287
		$updatehosed = 1;
	    }
	    else {
		print STDERR "Update recovery successful.\n";
Chad Barb's avatar
Chad Barb committed
288
	    }
Chad Barb's avatar
 
Chad Barb committed
289
	}
Chad Barb's avatar
Chad Barb committed
290 291
    }
}
292 293 294 295 296
elsif ($swapop eq "in") {
    #
    # Swap in
    #
    my $retries = 2;
297 298 299 300 301 302 303 304 305 306 307 308 309

    #
    # Before real swapin, do cursory admission control. assign_wrapper does
    # a more stringent check using assign.
    #
    print STDERR "Checking with Admission Control ...\n";
    if (! TBAdmissionControlCheck($UID, $pid, $eid, undef)) {
	print STDERR "Admission control failure!\n";
	print "Failingly finished swap-$swapop for $pid/$eid. " .
	    TBTimeStamp() . "\n";
	TBDebugTimeStamp("tbswap $swapop finished (failed)");
	exit(1);
    }
310 311
    
    $errors = doSwapin(REAL);
Chad Barb's avatar
Chad Barb committed
312

313 314 315 316 317 318 319 320 321 322
    #
    # 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--;
Chad Barb's avatar
 
Chad Barb committed
323

324 325
	print STDERR "Cleaning up after errors; will try again.\n";
	doSwapout(RETRY);
Chad Barb's avatar
 
Chad Barb committed
326

327 328
	print STDERR "Trying again...\n";
	$errors = doSwapin(RETRY);
Chad Barb's avatar
 
Chad Barb committed
329
    }
330 331
    if ($errors || $canceled) {
	print STDERR "Cleaning up after " .
332
	    ($canceled ? "cancelation" : "errors") . ".\n";
333
	doSwapout(CLEANUP);
Chad Barb's avatar
 
Chad Barb committed
334 335 336 337 338 339 340
    }
}

#
# Write appropriate message and exit.
#
if ($errors) {
341
    print "Failingly finished swap-$swapop for $pid/$eid. ".TBTimeStamp()."\n";
Chad Barb's avatar
 
Chad Barb committed
342
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
Chad Barb's avatar
Chad Barb committed
343

344 345
    # Pass out magic value to indicate that update failed!
    exit(1 | ($updatehosed ? 0x40 : 0));
Chad Barb's avatar
 
Chad Barb committed
346
}
347
print "Successfully finished swap-$swapop for $pid/$eid. " .TBTimeStamp()."\n";
348 349
TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
exit(0);
Chad Barb's avatar
 
Chad Barb committed
350 351 352 353

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

##
Chad Barb's avatar
Chad Barb committed
354
#
Chad Barb's avatar
 
Chad Barb committed
355 356
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
357
#             If in REAL or CLEANUP,
Chad Barb's avatar
 
Chad Barb committed
358 359 360
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
361
#             If in RETRY or UDPATE,
Chad Barb's avatar
 
Chad Barb committed
362 363 364 365 366 367
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

Chad Barb's avatar
 
Chad Barb committed
368 369
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
Chad Barb's avatar
 
Chad Barb committed
370 371 372 373 374 375 376 377 378 379 380
    my $swapout_errors = 0;

    #
    # 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
381

Chad Barb's avatar
 
Chad Barb committed
382
    if (! $TESTMODE) { 
383 384 385
	if (! ($DISABLE_EVENTS || $elabinelab)) {
	    if ($type >= RETRY ||
		($update_Eventsys_restart && $type == UPDATE) ) {
Chad Barb's avatar
 
Chad Barb committed
386 387 388 389 390
		print "Stopping the event system\n";
		if (system("eventsys_control stop $pid $eid")) {
		    print STDERR "*** Failed to stop the event system.\n";
		    $swapout_errors = 1;
		}
391 392 393 394 395 396 397 398 399 400 401

		#
		# Stop the location piper.
		#
		if (-x $piper) {
		    print "Stopping the location piper\n";
		    if (system("$piper -k $pid $eid")) {
			print STDERR "*** Failed to stop location piper.\n";
			$swapout_errors = 1;
		    }
		}
Chad Barb's avatar
 
Chad Barb committed
402 403
	    }
	}
404 405 406 407 408 409
	
	#
	# 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).
	#
410
	if ($elabinelab && $type >= CLEANUP) {
411 412 413 414 415 416 417
	    print "Tearing down elabinelab. This could take a while.\n";
	    if (system("elabinelab -k $pid $eid")) {
		print STDERR "*** Failed to teardown elabinelab!\n";
		$swapout_errors = 1;
	    }
	}

Chad Barb's avatar
 
Chad Barb committed
418
	#
Chad Barb's avatar
Chad Barb committed
419
	# Clean up any VLANs in experiment.
Chad Barb's avatar
 
Chad Barb committed
420
	#
421 422 423 424 425 426 427 428 429 430 431 432 433
	# When modifying an elabinelab experiment, leave the vlans intact
	# so that the inner networks are not suddenly disconnected!
	#
	if (! ($elabinelab && $type == UPDATE)) {
	    TBDebugTimeStamp("snmpit started");
	    print STDERR "Removing VLANs.\n";
	    if (system("snmpit -r $pid $eid")) {
		print STDERR "*** Failed to reset VLANs\n";
		$swapout_errors = 1;
	    } else {
		$cleanvlans = 0;
	    }
	    TBDebugTimeStamp("snmpit finished");
Chad Barb's avatar
 
Chad Barb committed
434
	}
435
    }
Chad Barb's avatar
Chad Barb committed
436

437 438 439
    if ($type >= CLEANUP) {
	#
	# We're not attempting a retry;
Chad Barb's avatar
 
Chad Barb committed
440
	#
441
	# Stop all of the vnodes.
Chad Barb's avatar
 
Chad Barb committed
442
	#
443
	if (! $TESTMODE) { 	
Chad Barb's avatar
 
Chad Barb committed
444 445 446 447 448 449 450
	    print "Tearing down virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid")) {
		print STDERR "*** Failed to tear down vnodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup finished");
Chad Barb's avatar
 
Chad Barb committed
451 452
	}

453
	if ($firewalled) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
	    #
	    # If the panic button was pressed, put all nodes into admin
	    # mode and power them off. After the firewall is torn down,
	    # we can power them back up.
	    #
	    my @nodes     = ExpNodes($pid, $eid, 1);
	    my $ADMINOSID = TB_OSID_FREEBSD_MFS;
	    my $paniced;

	    TBExptGetPanicBit($pid, $eid, \$paniced);

	    if ($paniced) {
		print STDERR "Powering down nodes.\n";
		TBDebugTimeStamp("Powering down nodes");

		system("power off @nodes");
		if ($?) {
		    #
		    # If an error, cannot continue. Must leave firewall in
		    # place. 
		    #
		    print STDERR "*** Failed to power nodes off! Stopping.\n";
		    return 1;
		}

		#
		# XXX: node_admin should take a list of nodes, or pid,eid.
		#      Instead, call os_select directly for now. 
		#
		# Clear any one-shot boots and partition boots.
		#
		print STDERR "Changing OSID to admin MFS.\n";
		TBDebugTimeStamp("Changing OSID to admin MFS");

		# Order matters cause os_select sillyness.
		if (system("os_select -c -1 @nodes") ||
		    system("os_select -t $ADMINOSID @nodes") ||
		    system("os_select -c @nodes")) {

		    #
		    # If an error, cannot continue. Must leave firewall in
		    # place. 
		    #
		    print STDERR "*** Failed to reset OSIDs! Stopping.\n";
		    return 1;
		}

		#
		# Now we can power up the nodes again.
		#
		print STDERR "Powering up nodes.\n";
		TBDebugTimeStamp("Powering up nodes");

		system("power on @nodes");
		if ($?) {
		    #
		    # If an error, cannot continue. Must leave firewall in
		    # place. Eventually, we can continue past this, if we
		    # know what nodes failed. But for now we have to do the
		    # ISUP test to make sure nodes really got into the MFS.
		    #
		    print STDERR "*** Failed to power on nodes! Stopping.\n";
		    return 1;
		}
	    }
519 520 521 522 523

	    #
	    # Once all nodes are safely in the admin MFS, we can take
	    # down the firewall
	    #
524
	    doFW($pid, $eid, FWTEARDOWN, undef);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545

	    if ($paniced) {
		#
		# Now wait for ISUP. I do this here cause the reload daemon
		# will power cycle the nodes again if its reboot fails, and
		# that will happen if the nodes are not back into the MFS in
		# time. THe right thing to do is to store the power state in
		# the DB, and have the reload daemon turn the nodes back on.
		#
		print STDERR "Waiting for nodes to boot the MFS.\n";
		TBDebugTimeStamp("Waiting for nodes to boot the MFS");
		
		system("node_statewait -s " . TBDB_NODESTATE_ISUP . " @nodes");
		if ($?) {
		    #
		    # Okay to continue; reload daemon will probably send
		    # email later when the reload fails.
		    #
		    print STDERR "*** Some nodes failed to reboot. Continuing\n";
		}
	    }	    
546 547
	}

Chad Barb's avatar
 
Chad Barb committed
548 549
	#
	# remove all nodes from the experiment.
Chad Barb's avatar
 
Chad Barb committed
550
	# (nfree will send them to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
551 552 553 554 555 556 557 558
	#
	print STDERR "Freeing nodes.\n";
	TBDebugTimeStamp("nfree started");
	if (system("nfree $pid $eid")) {
	    print STDERR "*** Could not free nodes.\n";
	    $swapout_errors = 1;
	}
	TBDebugTimeStamp("nfree finished");
Chad Barb's avatar
 
Chad Barb committed
559 560 561 562 563

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
564
	DBQueryWarn("update experiments set swap_requests='',sim_reswap_count='0' ".
565
		    "where eid='$eid' and pid='$pid'");
Chad Barb's avatar
 
Chad Barb committed
566 567
    } else {
	#
Chad Barb's avatar
 
Chad Barb committed
568
	# $type == RETRY or $type == UPDATE.
Chad Barb's avatar
 
Chad Barb committed
569 570
	# 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
571
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
572
	#
573 574 575 576 577 578 579 580
	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 ".
Chad Barb's avatar
 
Chad Barb committed
581 582
			 "where rv.pid='$pid' and rv.eid='$eid'");

583
	while (my ($node,$allocstate,$isvirt) = $db_result->fetchrow_array) {
Chad Barb's avatar
 
Chad Barb committed
584
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
585 586 587 588 589 590
		if ($isvirt) {
		    push(@failedvnodes, $node);
		}
		else {
		    push(@failedpnodes, $node);
		}
Chad Barb's avatar
 
Chad Barb committed
591 592 593
	    }
	}

594 595 596 597 598 599 600 601 602 603 604 605
	#
	# 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")) {
		print STDERR "*** Failed to tear down vnodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup -k finished");
	}
Chad Barb's avatar
 
Chad Barb committed
606

607 608 609 610
	#
	# Release all failed nodes.
	# 
	if (@failedpnodes > 0 || @failedvnodes > 0) {
611 612
	    print STDERR "Freeing failed nodes.\n";
	    
Chad Barb's avatar
 
Chad Barb committed
613
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
 
Chad Barb committed
614 615 616 617 618
	    #
	    # 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.
	    #
619 620
	    if (system("nfree -x $pid $eid " .
		       join(" ", (@failedpnodes, @failedvnodes)))) {
Chad Barb's avatar
 
Chad Barb committed
621 622 623 624 625 626 627
		print STDERR "*** Could not free nodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("nfree finished");
	}
    }

628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651
    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) {
	    # Are there any dslice nodes left?
	    $db_result =
		DBQueryFatal("select n.node_id from nodes as n ".
			     "left join node_types as nt on n.type = nt.type ".
			     "left join reserved as r ".
			     " on r.node_id = n.node_id ".
			     "where r.pid='$pid' and r.eid='$eid' ".
			     " and nt.isplabdslice = 1");

	    if (!$db_result->numrows) {
		print "Destroying Planetlab slice.\n";
		TBDebugTimeStamp("plabslice destroy started");
652
		if (system("plabslice destroy $pid $eid")) {
653 654 655 656 657 658 659 660
		    print STDERR "*** Failed to destroy Plab dslice\n";
		    $swapout_errors = 1;
		}
		TBDebugTimeStamp("plabslice destroy finished");
	    }
	}
    }

Chad Barb's avatar
 
Chad Barb committed
661 662 663 664 665 666 667 668
    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
Chad Barb's avatar
 
Chad Barb committed
669
	# not a failed swapin(cleanup), update, or retry.
Chad Barb's avatar
 
Chad Barb committed
670
	#
Chad Barb's avatar
 
Chad Barb committed
671
	if ($type == REAL) {
Chad Barb's avatar
 
Chad Barb committed
672 673 674 675 676 677 678
	    print "Resetting mountpoints.\n";
	    TBDebugTimeStamp("exports started");
	    if (system("exports_setup")) {
		print STDERR "*** Failed to reset mountpoints.\n";
	    }
	    TBDebugTimeStamp("exports finished");
	}
Chad Barb's avatar
Chad Barb committed
679

Chad Barb's avatar
 
Chad Barb committed
680 681 682 683
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
Chad Barb's avatar
 
Chad Barb committed
684 685 686 687 688 689
	print "Resetting named maps.\n";
	TBDebugTimeStamp("named started");
	if (system("named_setup")) {
	    print "*** WARNING: Failed to reset named map.\n";
	}
	TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
690

Chad Barb's avatar
 
Chad Barb committed
691 692
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
693
	if (system("genelists -t")) {
Chad Barb's avatar
 
Chad Barb committed
694 695 696 697 698
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

Chad Barb's avatar
 
Chad Barb committed
699
    #
700 701 702
    # Wipe the DB clean except during UPDATE or RETRY. In those
    #    cases, assign_wrapper will reset the DB after reading
    #    the info.
Chad Barb's avatar
 
Chad Barb committed
703
    #
704 705 706 707
    if ( $type >= CLEANUP ) {
	print STDERR "Resetting DB.\n";
	TBExptRemovePhysicalState( $pid, $eid );
    }
Chad Barb's avatar
 
Chad Barb committed
708 709 710 711 712

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
713
#
Chad Barb's avatar
 
Chad Barb committed
714 715
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
716 717 718 719 720
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
Chad Barb's avatar
 
Chad Barb committed
721 722
##

Chad Barb's avatar
 
Chad Barb committed
723
sub doSwapin($) {
724 725 726
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0.
    # Just the physnodes ...
    my @deleted_pnodes = ();
Chad Barb's avatar
 
Chad Barb committed
727

Chad Barb's avatar
 
Chad Barb committed
728 729 730 731 732
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
733 734 735 736 737
    if ($type > UPDATE_RECOVER) {
	print "Mapping to physical reality ...\n";
	TBDebugTimeStamp("assign_wrapper started");

	#
738 739 740 741
	# 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
742 743
	#
	my $exitcode;
744 745 746
	my $wrapper = "assign_wrapper -u";
	$wrapper .= " -f"
	    if ($type == RETRY);
747
	
748
	if (system("$wrapper $pid $eid")) {
749 750 751 752
	    $exitcode = $? >> 8;

	    print STDERR "*** Failed ($exitcode) to map to reality.\n";

753 754
	    # Wrapper sets this bit when recovery is possible.
	    if ($exitcode & 64) {
755
		# We can recover. 
Chad Barb's avatar
Chad Barb committed
756
		return 7;
757 758
	    }
	    else {
759
		# No recovery, no retry.
Chad Barb's avatar
Chad Barb committed
760 761 762 763
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
Chad Barb's avatar
 
Chad Barb committed
764

Chad Barb's avatar
Chad Barb committed
765 766
	print "Mapped to physical reality!\n";
    }
Chad Barb's avatar
 
Chad Barb committed
767

768
    # Check cancel flag before continuing. No retry, 
769
    TBGetCancelFlag($pid, $eid, \$canceled);
770 771 772
    return 1
	if ($canceled);

773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
    #
    # 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.
    #
    if ($type == UPDATE) {
	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);
		}
	    }
808 809 810
	    # See below.
	    @deleted_pnodes = @physnodes;
	    
811 812 813
	    if (@virtnodes) {
		TBDebugTimeStamp("vnode_setup started");
		
814
		if (system("vnode_setup -k $pid $eid @virtnodes")) {
815 816 817 818 819 820 821 822 823 824 825
		    print "Failed to tear down unused virtnodes!\n";
		    return 1;
		}
		TBDebugTimeStamp("vnode_setup finished");
		
		if (system("nfree $pid $eid @virtnodes")) {
		    print "Failed to nfree unused virtnodes!\n";
		    return 1;
		}
	    }
	    if (@physnodes) {
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
		if ($elabinelab) {
		    print "Removing nodes from inner elab.\n";
		    if (system("elabinelab -r $pid $eid @physnodes")) {
			print STDERR "*** Failed to remove inner nodes!\n";
			return 1;
		    }
		}

		#
		# XXX Nodes should actually be cleaned up, but we are not
		# doing that in the swapout case either, unless the panic
		# button was pressed, and in that case swapmod is not
		# allowed anyway (just swapout).
		# 
		if ($firewalled && doFW($pid, $eid,
					FWDELNODES, \@deleted_pnodes)) {
		    return 1;
		}
		
845 846 847 848 849 850 851 852
		if (system("nfree $pid $eid @physnodes")) {
		    print "Failed to nfree unused physnodes!\n";
		    return 1;
		}
	    }
	}
    }

Chad Barb's avatar
 
Chad Barb committed
853 854 855 856 857 858
    # Exit here if we are testing.
    if ($TESTMODE) {
	print "Testing run - Stopping here.\n";
	return 0;
    }

859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874
    #
    # 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?
	#
	print STDERR "*** Failed to set up tarballs.\n";
	return 1;
    }
    TBDebugTimeStamp("tarfiles_setup finished");

875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902
    #
    # If there are any Plab dslice nodes in the experiment, create the
    # dslice now
    #
    if ($type > UPDATE_RECOVER) {
	# Are there any Plab nodes?
	$db_result =
	    DBQueryFatal("select n.node_id from nodes as n ".
			 "left join node_types as nt on n.type = nt.type ".
			 "left join reserved as r on r.node_id = n.node_id ".
			 "where r.pid='$pid' and r.eid='$eid' ".
			 " and nt.isplabdslice = 1");

	if ($db_result->numrows) {
	    # Does slice already exist?
	    $db_result =
		DBQueryFatal("select slicename from plab_slices ".
			     "where pid='$pid' and eid='$eid'");

	    if (! $db_result->numrows) {
		my @plabnodes = ();
		
		while (my ($node) = $db_result->fetchrow_array()) {
		    push(@plabnodes, $node);
		}
		
		print "Creating Planetlab slice.\n";
		TBDebugTimeStamp("plabslice create started");
903
		if (system("plabslice create $pid $eid")) {
904 905 906 907 908 909 910 911
		    print STDERR "*** Failed to create Plab dslice\n";
		    return 3;
		}
		TBDebugTimeStamp("plabslice alloc finished");
	    }
	}
    }

912
    # Check cancel flag before continuing. No retry, 
913
    TBGetCancelFlag($pid, $eid, \$canceled);
914 915 916
    return 1
	if ($canceled);

Chad Barb's avatar
 
Chad Barb committed
917 918 919 920 921 922 923 924 925 926 927 928 929
    #
    # 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")) {
	print STDERR "*** Failed to setup mountpoints.\n";
	return 1;
    }
    TBDebugTimeStamp("mountpoints finished");
Chad Barb's avatar
Chad Barb committed
930

Chad Barb's avatar
 
Chad Barb committed
931 932 933 934 935 936
    TBDebugTimeStamp("named started");
    print "Setting up named maps.\n";
    if (system("named_setup")) {
	print STDERR "*** WARNING: Failed to add node names to named map.\n";
	#
	# This is a non-fatal error.
Chad Barb's avatar
Chad Barb committed
937
	#
Chad Barb's avatar
 
Chad Barb committed
938 939
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
940

941
    # Check cancel flag before continuing. No retry, 
942
    TBGetCancelFlag($pid, $eid, \$canceled);
943 944
    return 1
	if ($canceled);
Chad Barb's avatar
Chad Barb committed
945

946 947 948 949
    #
    # Setup any control-net firewall.
    # This must be done before reloading and rebooting nodes.
    #
950 951
    if ($firewalled && ($type == REAL || $type == UPDATE) &&
	doFW($pid, $eid, (($type == UPDATE) ? FWADDNODES : FWSETUP), undef)) {
952 953 954
	return 1;
    }

Chad Barb's avatar
Chad Barb committed
955 956
    #
    # If user specified -reboot to update,
Chad Barb's avatar
 
Chad Barb committed
957
    # and we are successfully performing the update,
958
    # then mark all nodes in experiment so os_setup will reboot them.
Chad Barb's avatar
Chad Barb committed
959
    #
960 961 962
    if (($type == UPDATE) &&
	($updateReboot || $updateReconfig)) {
	print STDERR "Marking nodes for reboot/reconfig.\n";
Chad Barb's avatar
Chad Barb committed
963
	$db_result =
964 965 966
	    DBQueryFatal("select r.node_id,n.allocstate from reserved as r ".
			 "left join nodes as n on n.node_id=r.node_id ".
			 "where r.pid='$pid' and r.eid='$eid'");
Chad Barb's avatar
Chad Barb committed
967

968 969 970 971 972 973 974 975
	while (my ($node,$allocstate) = $db_result->fetchrow_array) {
	    #
	    # If the node is INIT_CLEAN, leave it alone. It will still get
	    # rebooted, but will not falsely be tagged as dirty. This is
	    # important for vnodes too, where INIT_CLEAN indicated the vnode
	    # does not even exist yet (plab nodes).
	    #
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
976 977 978 979
		TBSetNodeAllocState($node,
				    ($updateReboot ?
				     TBDB_ALLOCSTATE_RES_INIT_DIRTY() :
				     TBDB_ALLOCSTATE_RES_RECONFIG()));
980
	    }
Chad Barb's avatar
Chad Barb committed
981 982 983
	}
    }

Chad Barb's avatar
 
Chad Barb committed
984 985 986 987 988
    #
    # Since it'll take a while for the nodes to reboot, we'll start now, and
    # wait for the os_setup to finish, down below
    #
    print "Resetting OS and rebooting.\n";
989
    TBDebugTimeStamp("launching os_setup");
Chad Barb's avatar
 
Chad Barb committed
990 991 992 993 994 995
    if (!($os_setup_pid = fork())) { 
	exec("os_setup $pid $eid") or return 1;
    } elsif ($os_setup_pid == -1) {
	print STDERR "*** Fork failed.\n";
	return 1;
    }
Chad Barb's avatar
Chad Barb committed
996

Chad Barb's avatar
 
Chad Barb committed
997
    #
998 999 1000 1001 1002
    # XXX
    # Don't add any steps between here and the waitpid() call below
    # without verifying that 1) It's OK for nodes to come up before
    # the step has completed and 2) It's OK for the command to run in
    # parallel with os_setup (no DB dependencies, etc.)
Chad Barb's avatar
 
Chad Barb committed
1003 1004 1005 1006 1007 1008 1009 1010 1011
    #

    print "Setting up VLANs.\n";
    TBDebugTimeStamp("snmpit started");
    if (system("snmpit -t $pid $eid")) {
	print STDERR "*** Failed to set up VLANs.\n";
	return 1;
    }
    TBDebugTimeStamp("snmpit finished");
Chad Barb's avatar
Chad Barb committed
1012

Chad Barb's avatar
 
Chad Barb committed
1013 1014 1015 1016
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
1017

Chad Barb's avatar
 
Chad Barb committed
1018 1019
    print "Setting up email lists.\n";
    TBDebugTimeStamp("genelists started");
1020
    if (system("genelists -t")) {
Chad Barb's avatar
 
Chad Barb committed
1021 1022 1023 1024 1025 1026
	print STDERR "*** WARNING: Failed to update email lists.\n";
	#
	# This is a non-fatal error.
	# 
    }
    TBDebugTimeStamp("genelists finished");
Chad Barb's avatar
Chad Barb committed
1027

Chad Barb's avatar
 
Chad Barb committed
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
    #
    # Don't clear port counters on UPDATE.
    # (XXX should clear new nodes' port counters.)

    if ($type >= RETRY) {
	print "Clearing port counters.\n";
	TBDebugTimeStamp("portstats started");
	if (system("portstats -z -a -q $pid $eid")) {
	    print STDERR "*** WARNING: Failed to clear port counters.\n";
	    #
	    # This is a non-fatal error.
	    # 
	}
	TBDebugTimeStamp("portstats finished");
Chad Barb's avatar
 
Chad Barb committed
1042
    }
Chad Barb's avatar
Chad Barb committed
1043

Chad Barb's avatar
 
Chad Barb committed
1044 1045 1046 1047 1048 1049 1050 1051
    #
    # OK, let's see how that os_setup did
    #
    $kid = waitpid($os_setup_pid,0);
    if ($kid == $os_setup_pid) {
	undef $os_setup_pid; # Make sure doswapout() doesn't wait for it.
	if ($CHILD_ERROR) {
	    print STDERR "*** Failed to reset OS and reboot nodes.\n";
1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
	    #
	    # If there is a firewall involved, it could be that the
	    # firewall rules are preventing essential communication,
	    # so don't retry.
	    #
	    # XXX should only do this if the user has specified additional
	    # rules.  But right now, I may screw up too!
	    #
	    if ($firewalled) {
		print STDERR "Not retrying, ".
		    "firewall may be preventing setup.\n";
		return 1;
	    }

Chad Barb's avatar
 
Chad Barb committed
1066
	    #
Chad Barb's avatar
 
Chad Barb committed
1067 1068
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
Chad Barb's avatar
 
Chad Barb committed
1069 1070 1071
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
Chad Barb's avatar
 
Chad Barb committed
1072
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
1073
		return 3;
Chad Barb's avatar
 
Chad Barb committed
1074 1075
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
1076
		return 1;
Chad Barb's avatar
 
Chad Barb committed
1077
	    }
Chad Barb's avatar
 
Chad Barb committed
1078 1079 1080 1081 1082 1083 1084
	}
    } else {
	undef $os_setup_pid;
	print STDERR "*** Error waiting for os_setup to finish.\n";
	return 1;
    }
    TBDebugTimeStamp("os_setup finished");
Chad Barb's avatar
Chad Barb committed
1085

Chad Barb's avatar
 
Chad Barb committed
1086 1087 1088 1089
    #
    # Okay, start the event system now that we know all the nodes have
    # rebooted (os_setup is done). This only takes a moment (puts itself
    # in the background), so its not enough of a delay to worry about.
Chad Barb's avatar
 
Chad Barb committed
1090 1091
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
Chad Barb's avatar
 
Chad Barb committed
1092
    # 
1093
    if (! ($DISABLE_EVENTS || $elabinelab)) {
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
	#
	# For the robot testbed, start the location piper *before* the event
	# system.
	#
	if (-x $piper && ($type != UPDATE && $type != UPDATE_RECOVER)) {
	    print "Starting the location piper.\n";
	    if (system("$piper $pid $eid")) {
		print STDERR "*** Failed to start the location piper.\n";
		return 1;
	    }
	}
	
1106 1107
	if ( $update_Eventsys_restart || 
	    ($type != UPDATE && $type != UPDATE_RECOVER) ) {
Chad Barb's avatar
 
Chad Barb committed
1108 1109 1110 1111 1112 1113 1114
	    print "Starting the event system.\n";
	    TBDebugTimeStamp("eventsys_control started");
	    if (system("eventsys_control start $pid $eid")) {
		print STDERR "*** Failed to start the event system.\n";
		return 1;
	    }
	    TBDebugTimeStamp("eventsys_control finished");
Chad Barb's avatar
 
Chad Barb committed
1115 1116
	}
    }
Chad Barb's avatar
Chad Barb committed
1117

1118 1119 1120
    #
    # ElabinElab setup. This might not be the right place for this!
    #
1121 1122 1123
    if ($elabinelab && !$TESTMODE && ($type == REAL || $type == UPDATE)) {
	my $optarg = ($type == UPDATE ? "-u" : "");
	
1124 1125
	print "Setting up elabinelab. This could take a while!\n";
	TBDebugTimeStamp("elabinelab setup started");
1126
	if (system("elabinelab $optarg $pid $eid")) {
1127 1128 1129
	    print STDERR "*** Failed to setup elabinelab!\n";
	    return 1;
	}
1130
	TBDebugTimeStamp("ElabInElab setup finished");
1131 1132
    }

Chad Barb's avatar
 
Chad Barb committed
1133 1134
    return 0;
}
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146

#
# Setup and teardown experiment firewall.
#
# XXX note that right now, we just setup the switch infrastructure
# first, and then just let everything else go.  Firewalled nodes will
# not boot until the firewall is up since the VLAN is isolated til then.
# The firewall will boot ok since it still talks to the real control net.
#
# XXX for tearing down firewalls, we assume that nodes have been "cleansed"
# and it is safe to put ports back into the default control net VLAN.
#
1147 1148
sub doFW($$$$) {
    my ($pid, $eid, $action, $nodelist) = @_;
1149
    my ($fwnode, $fwvlanname, $fwvlan, $fwport, $fwvid);
1150
    my %nodenames;
1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163

    #
    # See if there is a firewall, fetching node/VLAN info if so.
    # If not, we are all done.
    #
    if (!TBExptFirewall($pid, $eid, \$fwnode, \$fwvid, \$fwvlan)) {
	return 0;
    }

    if ($action == FWSETUP) {
	$fwvid = TBGetUniqueIndex("cnet_vlanid");

	print "Setting up control net firewall.\n";
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174
    }
    else {
	if ($action == FWADDNODES) {
	    print "Adding nodes to control net firewall.\n";
	}
	elsif ($action == FWDELNODES) {
	    print "Removing nodes from control net firewall.\n";
	}
	else {
	    print "Tearing down control net firewall.\n";
	}
1175 1176 1177 1178 1179 1180 1181

	# Prior setup didn't succeed, nothing to do
	if (!defined($fwvid)) {
	    return 0;
	}
    }

1182 1183 1184 1185 1186 1187 1188
    # See below.
    if (defined($nodelist)) {
	foreach my $node (@$nodelist) {
	    $nodenames{$node} = $node;
	}
    }

1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
    # XXX vlanid in the DB is currently an int, we need a more unique name
    $fwvlanname = "fw$fwvid";

    #
    # Find all the experiment nodes and their control interface switch ports
    #
    # XXX this may be replaced by a call to SNMPIT that just specifies
    # the pid/eid.  In that case someone else will first have to poplulate
    # the vlans table with this same info. 
    #
    my $db_result =
1200 1201 1202 1203 1204 1205
	DBQueryWarn("SELECT r.node_id,w.card1 ".
		    "  FROM wires AS w, reserved AS r ".
		    "WHERE r.node_id=w.node_id1 AND r.pid='$pid' ".
		    "  AND r.eid='$eid' AND w.type='Control'");
    return 1
	if (!$db_result);
1206 1207 1208

    my $portlist = "";
    while (my ($node,$cif) = $db_result->fetchrow_array()) {
1209
	print "$node $cif\n";
1210 1211
	if ($node eq $fwnode) {
	    $fwport = "$node:$cif";
1212 1213 1214 1215 1216 1217 1218 1219
	}
	elsif (defined($nodelist)) {
 	    print "foo @$nodelist\n";
	    # Only nodes we are moving in/out of the experiment.
	    $portlist .= " $node:$cif"
		if (exists($nodenames{$node}));
	}
	else {
1220 1221 1222 1223
	    $portlist .= " $node:$cif";
	}
    }
    if (!defined($fwport)) {
1224
	print STDERR "*** Firewall node '$fwnode' not found in $pid/${eid}!\n";
1225 1226 1227
	return 0;
    }
    if ($portlist eq "") {
1228 1229 1230 1231 1232 1233
	#
	# We catch this up in swapexp; admin users can specify just a firewall,
	# but mere users must have at least one firewalled node. Just print
	# the warning though. 
	# 
	print STDERR "*** WARNING: No firewalled nodes in $pid/${eid}!\n";
1234 1235 1236 1237 1238 1239 1240
    }

    #
    # XXX hack commands til we nail down the API
    #
    my $fwsetupstr1 = "snmpit $cnetstack -m $fwvlanname $portlist";
    my $fwsetupstr2 = "snmpit $cnetstack -T $fwport $cnetvlanname $fwvlanname";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1241
    my $fwtakedownstr0 = "snmpit $cnetstack -e $fwport";
1242 1243
    my $fwtakedownstr1 = ($portlist eq "" ? "true" :
			  "snmpit $cnetstack -m $cnetvlanname $portlist");
1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
    my $fwtakedownstr2 = "snmpit $cnetstack -o $fwvlanname";
    my $fwtakedownstr3 = "snmpit $cnetstack -U $fwport";
    my $fwtakedownstr4 = "snmpit $cnetstack -m $cnetvlanname $fwport";

    if (0) {
    # XXX debug fer now
    TBDebugTimeStamp("    setup1: $fwsetupstr1");
    TBDebugTimeStamp("    setup2: $fwsetupstr2");
    TBDebugTimeStamp(" teardown1: $fwtakedownstr1");
    TBDebugTimeStamp(" teardown2: $fwtakedownstr2");
    TBDebugTimeStamp(" teardown3: $fwtakedownstr3");
    TBDebugTimeStamp(" teardown4: $fwtakedownstr4");
    if ($action == FWSETUP) {
	my $lan = 100 + $fwvid;
	$fwsetupstr1    = "/bin/echo VLAN \#$lan on cisco2";
	$fwsetupstr2    = "true";
    } else {
	$fwtakedownstr1 = "true";
	$fwtakedownstr2 = "true";
	$fwtakedownstr3 = "true";
	$fwtakedownstr4 = "true";
    }
    }

    if ($action == FWSETUP) {
	TBDebugTimeStamp("snmpit firewall setup: VLAN");
1270
	print "doFW: '$fwsetupstr1'\n";
1271 1272 1273 1274 1275 1276 1277 1278
	my $snmpit_out = `$fwsetupstr1`;
	if ($? != 0 || $snmpit_out !~ /VLAN #(\d+) on /) {
	    print STDERR "*** Failed to setup Firewall control net VLAN.\n";
	    return 1;
	}
	my $fwvlan = $1;

	TBDebugTimeStamp("snmpit firewall setup: trunk");
1279
	print "doFW: '$fwsetupstr2'\n";
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
	if (system($fwsetupstr2)) {
	    print STDERR "*** Failed to setup Firewall trunk on port $fwport.\n";
	    if (system($fwtakedownstr1)) {
		print STDERR
		    "*** Could not return $portlist to Control VLAN!\n";
	    }
	    if (system($fwtakedownstr2)) {
		print STDERR
		    "*** Could not destroy VLAN $fwvlanname ($fwvlan)!\n";
	    }
	    return 1;
	}
	TBDebugTimeStamp("snmpit firewall setup done");

	# Record VLAN info now that everything is done
	TBSetExptFirewallVlan($pid, $eid, $fwvid, $fwvlan);
1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
    }
    elsif ($action == FWADDNODES) {
	TBDebugTimeStamp("snmpit firewall port addition");
	print "doFW: '$fwsetupstr1'\n";
	my $snmpit_out = `$fwsetupstr1`;
	if ($?) {
	    print STDERR
	        "*** Failed to add nodes to Firewall control net VLAN.\n";
	    return 1;
	}
	TBDebugTimeStamp("snmpit firewall setup done");
    }
    elsif ($action == FWDELNODES) {
	TBDebugTimeStamp("snmpit firewall port deletion");
	print "doFW: '$fwtakedownstr1'\n";
	if (system($fwtakedownstr1)) {
	    print STDERR
		"*** Failed to remove nodes from Firewall control net VLAN.\n";
	    return 1;
	}
	TBDebugTimeStamp("snmpit firewall setup done");
    }
    else {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1319
	TBDebugTimeStamp("snmpit re-enable fw control port: $fwport");