elabinelab.in 39.3 KB
Newer Older
1 2 3
#!/usr/bin/perl -wT
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2004-2007 University of Utah and the Flux Group.
5 6 7 8 9 10 11 12
# All rights reserved.
#
# TODO: ntpinfo table.
#       Current source directory? From where?
#
use English;
use Getopt::Std;

13 14 15 16 17
# Load the Testbed support stuff.
use lib "@prefix@/lib";
use libdb;
use libtestbed;

18 19 20 21 22
#
# Do things necessary for setting up inner elab experiment. 
#
sub usage()
{
23
    print STDOUT "Usage: elabinelab [-d] [-g] [-u] pid eid\n";
24
    print STDOUT "       elabinelab [-d] [-k | -f] pid eid\n";
25
    print STDOUT "       elabinelab [-d] -r pid eid [node ...]\n";
26 27 28
 
    exit(-1);
}
29
my $optlist  = "dgkfur";
30
my $debug    = 1;
31
my $verbose  = 0;
32
my $killmode = 0;
33
my $fwboot   = 0;
34
my $dbgooonly= 0;
35 36
my $update   = 0;
my $remove   = 0;
37

38 39
sub DumpDBGoo();

40 41 42 43 44 45 46 47 48 49
#
# Configure variables
#
my $TB		= "@prefix@";
my $TBOPS       = "@TBOPSEMAIL@";
my $CONTROL	= "@USERNODE@";
my $TBOPSPID    = TBOPSPID();
my $SSH		= "$TB/bin/sshtb";
my $nodereboot  = "$TB/bin/node_reboot";
my $makeconf    = "$TB/sbin/dhcpd_makeconf";
50
my $nodewait    = "$TB/sbin/node_statewait";
51
my $snmpit      = "$TB/bin/snmpit";
52 53 54 55 56 57 58

# Locals
my $elabinelab;
my $elabinelab_eid;

# Protos
sub TearDownEmulab();
59 60
sub RemoveNodes();
sub UpdateEmulab();
61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78

# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin:/usr/site/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

#
# We don't want to run this script unless its the real version.
#
if ($EUID != 0) {
    die("*** $0:\n".
	"    Must be root! Maybe its a development version?\n");
}

79 80 81
# Be careful not to exit on transient error
$libdb::DBQUERY_MAXTRIES = 30;

82 83 84
# Locals
my $SAVEUID     = $UID;
my $workdir;
85
my $expdir;
86 87 88
my %noderoles	= ();
my $opsnode;
my $bossnode;
Mike Hibler's avatar
Mike Hibler committed
89 90
my $fsnode;
my $routernode;
91 92 93 94
my @expnodes    = ();
my $dbuid;
my $user_name;
my $user_email;
95 96
my $query_result;
my $exptinfo;
97 98 99 100 101 102 103 104 105 106 107 108 109 110 111

#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
if (defined($options{"g"})) {
    $dbgooonly = 1;
}
if (defined($options{"d"})) {
    $debug = 1;
}
112 113 114
if (defined($options{"k"})) {
    $killmode = 1;
}
115 116 117
if (defined($options{"f"})) {
    $fwboot = 1;
}
118 119 120 121 122 123
if (defined($options{"u"})) {
    $update = 1;
}
if (defined($options{"r"})) {
    $remove = 1;
}
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144
if (! @ARGV) {
    usage();
}
my ($pid,$eid) = @ARGV;

#
# 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");
}
$workdir = TBExptWorkDir($pid, $eid);
145
$expdir = PROJROOT() . "/$pid/exp/$eid";
146

147 148 149 150 151
# Build Logfile names
my $opslogfile  = "$workdir/opsnode.log";
my $fslogfile   = "$workdir/fsnode.log";
my $bosslogfile = "$workdir/bossnode.log";

152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
#
# 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");
}
Leigh Stoller's avatar
Leigh Stoller committed
167
TBDebugTimeStampsOn();
168

169 170 171 172 173 174 175 176 177 178 179
#
# Get elabinelab status to make sure, and to see if we need to fire off
# an experiment inside once its setup.
#
if (! TBExptIsElabInElab($pid, $eid, \$elabinelab, \$elabinelab_eid)) {
    die("*** $0:\n".
	"    Could not get elabinelab status for experiment $pid/$eid\n");
}
exit(0)
    if (!$elabinelab);

180 181 182 183 184 185
#
# See if the experiment is firewalled
#
my $firewall;
my $firewalled = TBExptFirewall($pid, $eid, \$firewall);

186 187 188 189
#
# XXX NOSETUP option needs more work ...
#
$query_result =
190 191
    DBQueryFatal("select elabinelab_nosetup,elabinelab_singlenet ".
		 "   from experiments ".
192
		 "where pid='$pid' and eid='$eid'");
193
my ($nosetup,$singlenet) = $query_result->fetchrow_array();
194

195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
#
# Presetup; turn off firewall.
#
if ($fwboot) {
    exit(0)
	if (!$firewalled);
    
    print "Turning off firewall rules on $firewall\n";
    $UID = 0;
    system("$SSH -host $firewall ipfw add 1 allow all from any to any");
    if ($?) {
	die("*** $0:\n".
	    "    Error turning off firewall rules ($firewall)!\n");
    }
    exit(0);
}

212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
#
# If we are going to start an inner experiment, grab the stuff we need
# from the DB and save it. 
#
if (defined($elabinelab_eid)) {
    $query_result =
	DBQueryFatal("select nsfile from nsfiles ".
		     "where pid='$pid' and eid='$elabinelab_eid'");

    die("*** $0:\n".
	"    No such experiment in DB for $pid/$elabinelab_eid\n")
	if (!$query_result->numrows);

    my ($nsfile) = $query_result->fetchrow_array();
    
    die("*** $0:\n".
	"    No nsfile in DB for $pid/$elabinelab_eid\n")
	if (!defined($nsfile) || $nsfile eq "");

    $query_result =
	DBQueryFatal("select * from experiments ".
		     "where pid='$pid' and eid='$elabinelab_eid'");

    die("*** $0:\n".
	"    No such experiment in DB for $pid/$elabinelab_eid\n")
	if (!$query_result->numrows);

    $exptinfo = $query_result->fetchrow_hashref();
    $exptinfo->{"nsfile"} = $nsfile;
}

#
# Get the role for each node.
#
$query_result =
    DBQueryFatal("select r.node_id,r.inner_elab_role from reserved as r ".
		 "where r.pid='$pid' and r.eid='$eid'");
while (my ($node_id,$role) = $query_result->fetchrow_array()) {
    # Like, the firewall node.
    next
	if (!defined($role));
	
    $noderoles{$node_id} = $role;
    $bossnode = $node_id
Mike Hibler's avatar
Mike Hibler committed
256 257 258
	if ($role eq 'boss' || $role eq 'boss+router');
    $routernode = $node_id
	if ($role eq 'router');
259
    $opsnode = $node_id
Mike Hibler's avatar
Mike Hibler committed
260 261 262
	if ($role eq 'ops' || $role eq 'ops+fs');
    $fsnode = $node_id
	if ($role eq 'fs');
263 264 265 266 267 268 269 270 271 272
    push(@expnodes, $node_id)
	if ($role eq 'node');
}

#
# Tear down an inner emulab.
# 
if ($killmode) {
    exit(TearDownEmulab());
}
273 274 275 276 277 278
elsif ($remove) {
    exit(RemoveNodes());
}
elsif ($update) {
    exit(UpdateEmulab());
}
279 280 281 282 283 284

#
# Get elabinelab info. If this is a container for an actual experiment,
# then need to fire off the experiment once the inner emulab is ready to
# go.
# 
Leigh Stoller's avatar
Leigh Stoller committed
285
TBDebugTimeStamp("Dumping DB state");
286 287 288 289 290
DumpDBGoo();
exit(0)
    if ($dbgooonly);

#
291
# For SSH and SCP below
292 293
#
$UID = 0;
Leigh Stoller's avatar
Leigh Stoller committed
294

295
#
296 297
# The firewall should be off at this point; called from os_setup with -f.
# 
298

Leigh Stoller's avatar
Leigh Stoller committed
299 300 301
#
# This is temporary. I think I will switch this over to grabbing the latest
# version from the web server.
302 303
#
my $mkelab = "$TB/etc/rc.mkelab";
304 305
if (-e "$expdir/rc.mkelab") {
    $mkelab = "$expdir/rc.mkelab";
306
}
307 308 309 310
print "Copying $mkelab to ${bossnode}/${opsnode}";
print "/${fsnode}"
    if (defined($fsnode));
print "\n";
311 312
system("scp $mkelab ${bossnode}:/tmp");
system("scp $mkelab ${opsnode}:/tmp");
313 314
system("scp $mkelab ${fsnode}:/tmp")
    if (defined($fsnode));
315

316 317 318
if (defined($fsnode)) {
    TBDebugTimeStamp("Setting up fsnode");
    print "Setting up fsnode on $fsnode\n";
319
    system("$SSH -host $fsnode /tmp/rc.mkelab -s -d > $fslogfile 2>&1");
320 321 322 323 324 325 326
    if ($?) {
	$UID = $SAVEUID;
	SENDMAIL("$user_name <$user_email>",
		 "ElabInElab Failure: $pid/$eid",
		 "Error building the fs node ($fsnode)",
		 $TBOPS,
		 "Cc: $TBOPS",
327
		 ($fslogfile));
328 329 330 331 332
	print STDERR "*** $0:\n".
	    "    Error building the fsnode ($fsnode)!\n";
	exit(($debug ? 0 : -1));
    }
}
Leigh Stoller's avatar
Leigh Stoller committed
333
TBDebugTimeStamp("Setting up opsnode");
334
print "Setting up opsnode on $opsnode\n";
335
system("$SSH -host $opsnode /tmp/rc.mkelab -s -d > $opslogfile 2>&1");
336 337 338 339
if ($?) {
    $UID = $SAVEUID;
    SENDMAIL("$user_name <$user_email>",
	     "ElabInElab Failure: $pid/$eid",
340
	     "Error building the ops node ($opsnode)",
341 342
	     $TBOPS,
	     "Cc: $TBOPS",
343
	     ($opslogfile));
344 345 346 347
    print STDERR "*** $0:\n".
	         "    Error building the opsnode ($opsnode)!\n";
    exit(($debug ? 0 : -1));
}
Leigh Stoller's avatar
Leigh Stoller committed
348
TBDebugTimeStamp("Setting up bossnode");
349
print "Setting up bossnode on $bossnode\n";
350
system("$SSH -host $bossnode /tmp/rc.mkelab -s -d > $bosslogfile 2>&1");
351 352 353 354
if ($?) {
    $UID = $SAVEUID;
    SENDMAIL("$user_name <$user_email>",
	     "ElabInElab Failure: $pid/$eid",
355
	     "Error building the boss node ($bossnode)",
356 357
	     $TBOPS,
	     "Cc: $TBOPS",
358
	     ($bosslogfile));
359 360 361 362 363
    print STDERR "*** $0:\n".
	         "    Error building the bossnode ($bossnode)!\n";
    exit(($debug ? 0 : -1));
}

364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
if ($verbose) {
    # Send these log files off now so that we can look at them.
    if (defined($fsnode)) {
	SENDMAIL("$user_name <$user_email>",
		 "ElabInElab Setup Log: $pid/$eid",
		 "Logs for building fs/ops/boss ($fsnode/$opsnode/$bossnode)",
		 $TBOPS,
		 "Cc: $TBOPS",
		 ($fslogfile, $opslogfile, $bosslogfile));
    }
    else {
	SENDMAIL("$user_name <$user_email>",
		 "ElabInElab Setup Log: $pid/$eid",
		 "Logs for building ops/boss ($opsnode/$bossnode)",
		 $TBOPS,
		 "Cc: $TBOPS",
		 ($opslogfile, $bosslogfile));
    }
382
}
383
$UID  = $SAVEUID;
384 385

# Run as real user for the next few scripts, which are setuid.
386
$EUID = $UID;
387

388 389 390
goto skipsetup
    if ($nosetup);

391
#
392 393 394 395 396
# Restart DHCPD, but first mark the nodes as being ready to boot inside
# the inner emulab, so that dhcpd_makeconf knows what nodes to change
# the entries for.
#
DBQueryFatal("update reserved set inner_elab_boot=1 ".
397
	     "where pid='$pid' and eid='$eid'");
398

399 400 401 402 403 404 405
print "Regenerating DHCPD config file and restarting daemon.\n";
system("$makeconf -i -r");
if ($?) {
    die("*** $0:\n".
	"    Failed to reconfig/restart DHCPD.\n");
}

406 407 408 409 410 411 412 413 414 415
if (defined($fsnode)) {
    # Reboot fs and wait for it to come back.
    print "Rebooting fsnode ($fsnode).\n";
    TBDebugTimeStamp("Rebooting fsnode");
    system("$nodereboot -w $fsnode");
    if ($?) {
	die("*** $0:\n".
	    "    Error rebooting the fsnode ($fsnode)!\n");
    }
}
416
# Reboot ops and wait for it to come back.
417
print "Rebooting opsnode ($opsnode).\n";
Leigh Stoller's avatar
Leigh Stoller committed
418
TBDebugTimeStamp("Rebooting opsnode");
419 420 421 422 423 424
system("$nodereboot -w $opsnode");
if ($?) {
    die("*** $0:\n".
	"    Error rebooting the opsnode ($opsnode)!\n");
}
# Reboot boss and wait for it to come back.
425
print "Rebooting bossnode ($bossnode).\n";
Leigh Stoller's avatar
Leigh Stoller committed
426
TBDebugTimeStamp("Rebooting bossnode");
427 428 429 430 431
system("$nodereboot -w $bossnode");
if ($?) {
    die("*** $0:\n".
	"    Error rebooting the bossnode ($bossnode)!\n");
}
432
$EUID = 0;
Leigh Stoller's avatar
Leigh Stoller committed
433

434 435 436
# Reboot the experimental nodes. They will come up inside the inner elab.
# DO NOT WAIT! They are not going to report ISUP from this point on. 
if (@expnodes) {
437
    print "Rebooting inner experimental nodes.\n";
Leigh Stoller's avatar
Leigh Stoller committed
438
    TBDebugTimeStamp("Rebooting experimental nodes");
439 440
    # Run as real user again.
    $EUID = $UID;
441 442 443 444 445
    system("$nodereboot @expnodes");
    if ($?) {
	die("*** $0:\n".
	    "    Error rebooting the expnodes (@expnodes)!\n");
    }
446
    $EUID = 0;
447 448 449 450 451

    #
    # Instead, we ssh into the node and use a utility script to determine
    # when the nodes have rebooted and are in PXEWAIT (part of the inner elab).
    #
452
    # Run as real root for ssh.
453 454 455 456 457 458 459 460 461 462 463
    $UID  = 0;

    print "Waiting for nodes to reboot and join the inner emulab.\n";
    TBDebugTimeStamp("Waiting for inner nodes to reboot");
    system("$SSH -host $bossnode /usr/testbed/sbin/node_statewait -t 180 -a");
    if ($?) {
	print STDERR "*** $0:\n".
	             "    Error waiting for inner nodes to join!\n";
	exit(($debug ? 0 : -1));
    }
    $UID  = $SAVEUID;
464 465 466 467 468 469 470 471 472

    #
    # To avoid confusion later (with swapmod, which wants them to be ISUP),
    # and so the web interface does not show the nodes as down, set the 
    # state to ISUP.
    #
    foreach my $node (@expnodes) {
	TBSetNodeEventState($node, TBDB_NODESTATE_ISUP());
    }
473
}
474 475

#
476 477 478 479 480
# Fire off inner elab experiment.
# 
if (defined($elabinelab_eid)) {
    # Formatted to make batchexp happy.
    my $nsfilename = "/tmp/$pid-$elabinelab_eid-$$.nsfile";
481
    
482 483 484 485 486 487 488 489 490 491 492 493 494 495
    #
    # Write NS file to temp file so we can send it over.
    #
    open(NS, "> /tmp/$$.ns")
	or die("*** $0:\n".
	       "    Could not write ns code to tmp file!\n");
    print NS $exptinfo->{"nsfile"};
    print NS "\n";
    close(NS);

    #
    # Copy the file over.
    #
    $UID = 0;
496
    print "Sending NS file to inner bossnode ($bossnode).\n";
497
    system("cat /tmp/$$.ns | $SSH -host $bossnode '(cat > $nsfilename)'");
498 499
    if ($?) {
	die("*** $0:\n".
500
	    "    Could not copy ns code to inner boss ($bossnode)!\n");
501
    }
502 503

    #
504 505
    # Now run batchexp on the node as the user. If firewalled, experiment
    # must start async (cause we have to turn the firewall back on). 
506
    #
507 508 509 510
    my $optarg = ($firewalled ? "" : "-w");
	
    print "Starting experiment $pid/$elabinelab_eid on inner emulab.\n";
    TBDebugTimeStamp("Starting inner experiment");
511
    system("$SSH -host $bossnode 'sudo -u $dbuid /usr/testbed/bin/batchexp ".
512
	   "  -q -i $optarg -S \"ElabInElab Experiment\" ".
513 514 515 516 517
	   "  -L \"ElabInElab ElabInElab\" -E \"ElabInElab Experiment\" ".
	   "  -p $pid -e $elabinelab_eid $nsfilename'");
    
    $UID = $SAVEUID;
    unlink("/tmp/$$.ns");
518
}
519
skipsetup:
520

521
#
522 523 524 525 526 527
# Turn the firewall back on.
#
# XXX If this fails, we have to do something much stronger! We do not want
# nodes coming up and starting something if the firewall is not active.
# Maybe hit the panic button from here (turning off the control network).
#
528 529 530 531 532 533
#
if ($firewalled) {
    print "Turning firewall back on\n";
    $UID = 0;
    system("$SSH -host $firewall ipfw delete 1");
    if ($?) {
534 535 536 537 538 539 540
	print STDERR "*** Error turning back on firewall rules ($firewall)!\n".
		     "    Will retry again.\n";
	system("$SSH -host $firewall ipfw delete 1");
	if ($?) {
	    die("*** $0:\n".
		"    Error turning back on firewall rules! Retry failed.\n");
	}
541 542 543 544
    }
    $UID = $SAVEUID;
}

545 546 547
TBDebugTimeStamp("ElabInElab setup done");
exit(0);

548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571
#
# Dump parts of the DB that are needed for inner elab to run. The idea
# is to create a set of files named by the table name. Note that mysqld
# cannot write to the project tree cause of directory permissions. Put the
# files into the workdir for now, and them copy them over. 
#
sub DumpDBGoo()
{
    my $statedir = "$workdir/elabinelab";

    if (-d $statedir) {
	system("rm -rf $statedir");
    }
    mkdir($statedir, 0777) or
	die("*** $0:\n".
	    "    Could not mkdir $statedir\n");
    
    chmod(0777, $statedir) or
	die("*** $0:\n".
	    "    Could not chmod $statedir\n");

    #
    # These tables are dumped completely.
    #
572
    my @FULLTABLES = ("node_types", "node_type_attributes", "interface_types",
573 574
		      "interface_capabilities",
		      "switch_paths", "switch_stack_types", "switch_stacks",
575 576
		      "node_type_features", "node_types_auxtypes", "osid_map",
		      "os_boot_cmd");
577 578 579

    #
    # These tables are dumped by role (node/ops). For each one dump the table
580 581
    # as is, unless its the fs or ops node. For those we want to change the
    # node_id to "fs" or "ops" and their type to ops.
582 583
    #
    my @NODETABLES = ("node_auxtypes", "node_status", "nodes",
Leigh Stoller's avatar
Leigh Stoller committed
584
		      "node_rusage", "node_hostkeys", "node_idlestats");
585 586 587 588 589 590 591 592 593 594 595 596 597

    #
    # These tables are dumped by project ID.
    #
    my @PROJTABLES = ("project_stats", "projects", "group_stats", "groups");

    #
    # These tables are dumped by user ID (for the project).
    #
    my @USERTABLES = ("users", "user_pubkeys", "user_stats");

    foreach my $table (@FULLTABLES) {
	unlink("$statedir/$table");
598 599 600 601 602
	DBQueryWarn("create temporary table temp_${table} ".
		    "select t.* from $table as t")
	    or die("*** $0:\n".
		   "    Could not dump table $table\n");

603 604
	if ($table eq "node_type_attributes") {
	    my $attributes_result =
605
		DBQueryFatal("select type,attrvalue from temp_${table} ".
606 607 608 609
			     "where attrkey='delay_capacity'");

	    while (my ($ntype,$value) = $attributes_result->fetchrow_array()) {
		my $newvalue = $value - 1;
610 611 612

		next
		    if ($newvalue < 0);
613 614 615
		
		DBQueryFatal("update temp_${table} set ".
			     "   attrvalue='$newvalue' ".
616 617
			     "where type='$ntype' and ".
			     "      attrkey='delay_capacity'");
618
	    }
619 620 621
	}

	DBQueryWarn("select * from temp_$table ".
622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
		    "into outfile '$statedir/$table'")
	    or die("*** $0:\n".
		   "    Could not dump table $table\n");
    }

    foreach my $table (@NODETABLES) {
	unlink("$statedir/$table");
	#
	# Create a temporary table.
	#
	DBQueryWarn("create temporary table temp_${table} ".
		    "select t.* from reserved as r ".
		    "left join $table as t on t.node_id=r.node_id ".
		    "left join virt_nodes as v on v.vname=r.vname and ".
		    "     v.pid=r.pid and v.eid=r.eid ".
		    "where r.pid='$pid' and r.eid='$eid' and ".
		    "      t.node_id is not null and ".
639
		    "      v.inner_elab_role in ('node','fs','ops','ops+fs')")
640 641 642
	    or die("*** $0:\n".
		   "    Could not create temporary table temp_$table\n");
	#
643 644
	# Rename the fs and ops node in each table. For the nodes table,
	# there is a bunch of other stuff to do.
645
	#
646 647 648
	DBQueryFatal("update temp_${table} set node_id='fs' ".
		     "where node_id='$fsnode'")
	    if (defined($fsnode));
649 650 651 652 653 654
	DBQueryFatal("update temp_${table} set node_id='ops' ".
		     "where node_id='$opsnode'");

	if ($table eq "nodes") {
	    DBQueryFatal("update temp_${table} set ".
			 " type='ops', ".
655
			 " phys_nodeid=node_id, ".
656 657
			 " role='ctrlnode', ".
			 " op_mode='OPSNODEBSD' ".
658
			 "where node_id in ('fs','ops')");
659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682

	    # Also add the nodes that correspond to the "trunk" wires.
	    DBQueryFatal("insert into temp_${table} ".
			 "select distinct n.* from wires as w ".
			 "left join nodes as n on w.node_id1=n.node_id or ".
			 "     w.node_id2=n.node_id ".
			 "where w.type='Trunk'");
	}
    
	DBQueryWarn("select * from temp_$table ".
		    "into outfile '$statedir/$table'")
	    or die("*** $0:\n".
		   "    Could not dump table $table\n");
    }

    foreach my $table (@PROJTABLES) {
	unlink("$statedir/$table");
	DBQueryWarn("select * from $table ".
		    "where pid='$pid' ".
		    "into outfile '$statedir/$table'")
	    or die("*** $0:\n".
		   "    Could not dump table $table\n");
    }

683 684 685 686 687 688 689 690 691 692 693 694
    #
    # Special case the group and user policy tables. Not sure what to
    # really do about this; should there be any restrictions inside the
    # inner elab?
    #
    unlink("$statedir/group_policies");
    DBQueryWarn("select * from group_policies ".
		"where pid='$pid' or pid='+' or pid='-' ".
		"into outfile '$statedir/group_policies'")
	or die("*** $0:\n".
	       "    Could not dump table group_policies\n");

695 696
    foreach my $table (@USERTABLES) {
	unlink("$statedir/$table");
697 698 699

	DBQueryWarn("create temporary table temp_$table ".
		    "select t.* from group_membership as gm ".
700 701
		    "left join users as u on u.uid_idx=gm.uid_idx ".
		    "left join $table as t on t.uid_idx=u.uid_idx ".
702
		    "where gm.pid='$pid' and gm.gid=gm.pid ".
703
		    " and t.uid_idx is not NULL and ".
704 705 706 707 708 709 710 711 712 713 714 715 716
		    " u.status='" . USERSTATUS_ACTIVE() . "'")
	    or die("*** $0:\n".
		   "    Could not create table temp_$table\n");

	if ($table eq "users") {
	    $creator = ExpLeader($pid, $eid);
	    
	    DBQueryFatal("update temp_${table} set ".
			 " admin=1 ".
			 "where uid='$creator'");
	}

	DBQueryWarn("select * from temp_$table ".
717 718
		    "into outfile '$statedir/$table'")
	    or die("*** $0:\n".
719
		   "    Could not dump table temp_$table\n");
720 721 722 723
    }

    # The group_membership is also special.
    DBQueryWarn("select gm.* from group_membership as gm ".
724
		"left join users as u on u.uid_idx=gm.uid_idx ".
725 726
		"where (gm.pid='$pid' or ".
		"       gm.pid='" . TBOPSPID() . "') and ".
727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750
		" u.status='" . USERSTATUS_ACTIVE() . "' ".
		"into outfile '$statedir/group_membership'")
	or die("*** $0:\n".
	       "    Could not dump table group_membership\n");

    #
    # Initial images; not that these images are not going to exist inside!
    # 
    DBQueryWarn("select * from images ".
		"where pid='$pid' or (pid='$TBOPSPID' and global=1) ".
		"into outfile '$statedir/images'")
	or die("*** $0:\n".
	       "    Could not dump table images\n");
	    
    DBQueryWarn("create temporary table temp_os_info ".
		"select * from os_info ".
		"where pid='$pid' or (pid='$TBOPSPID' and shared=1)")
	or die("*** $0:\n".
	       "    Could not create table temp_os_info\n");

    # Ack. The MFS paths have a hardcoded "boss" in them, but that is going
    # to resolve incorrectly to an inner control IP, which will not work
    # from the pxeboot kernel since it uses the outer control network.
    # Just remove the host spec; pxeboot will do the right thing.
751
    my $query_result =
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
	DBQueryFatal("select osid,path from temp_os_info ".
		     "where path like '%:%'");
    
    while (my ($osid,$hostpath) = $query_result->fetchrow_array()) {
	my ($host,$path) = $hostpath =~ /^(.*):(.*)$/;

	DBQueryFatal("update temp_os_info set path='$path' where osid='$osid'");
    }

    DBQueryWarn("select * from temp_os_info ".
		"into outfile '$statedir/os_info'")
	or die("*** $0:\n".
	       "    Could not dump table os_info\n");
	    
    DBQueryWarn("select o.* from osidtoimageid as o ".
		"left join images as i on i.imageid=o.imageid ".
		"where i.pid='$pid' or (i.pid='$TBOPSPID' and i.global=1) ".
		"into outfile '$statedir/osidtoimageid'")
	or die("*** $0:\n".
	       "    Could not dump table osidtoimageid\n");

    #
    # interfaces table. Need to tag the interfaces being used as the control
    # network, with the proper tag so they do not say they experimental
    # interfaces in the inner emulab. Use a temp table again.
    #
    DBQueryWarn("create temporary table temp_interfaces ".
		"select t.* from reserved as r ".
		"left join interfaces as t on t.node_id=r.node_id ".
		"left join virt_nodes as v on v.vname=r.vname and ".
		"     v.pid=r.pid and v.eid=r.eid ".
		"where r.pid='$pid' and r.eid='$eid' and ".
784
		"      v.inner_elab_role in ('node','ops','fs','ops+fs')")
785 786 787
	or die("*** $0:\n".
	       "    Could not create temporary table temp_interfaces\n");

788 789 790 791 792 793 794 795
    if (! $singlenet) {
	# First, mark the real control network as "other" to avoid it being
	# thought of as the control network!.
	DBQueryWarn("update temp_interfaces ".
		    "set role='" . TBDB_IFACEROLE_OUTER_CONTROL() . "' " .
		    "where role='" . TBDB_IFACEROLE_CONTROL() . "'")
	    or die("*** $0:\n".
		   "    Could not delete control ifaces from temp_interfaces\n");
796

797 798 799 800 801 802
	DBQueryWarn("update temp_interfaces set ".
		    " role='" . TBDB_IFACEROLE_CONTROL() . "' " .
		    "where IP!='' and role='" . TBDB_IFACEROLE_EXPERIMENT() . "'")
	    or die("*** $0:\n".
		   "    Could not update roles in temp_interfaces\n");
    }
803

804 805 806 807 808 809 810
    # And rename the fs/ops nodes as above.
    if (defined($fsnode)) {
	DBQueryWarn("update temp_interfaces set node_id='fs' ".
		    "where node_id='$fsnode'")
	    or die("*** $0:\n".
		   "    Could not fs node_id in temp_interfaces\n");
    }
811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
    DBQueryWarn("update temp_interfaces set node_id='ops' ".
		"where node_id='$opsnode'")
	or die("*** $0:\n".
	       "    Could not ops node_id in temp_interfaces\n");

    # Also add the interfaces that correspond to the "trunk" wires.
    DBQueryFatal("insert into temp_interfaces ".
		 "select distinct i.* from wires as w ".
		 "left join interfaces as i on w.node_id1=i.node_id or ".
		 "     w.node_id2=i.node_id ".
		 "where w.type='Trunk'");

    DBQueryWarn("select * from temp_interfaces ".
		"into outfile '$statedir/interfaces'")
	or die("*** $0:\n".
	       "    Could not dump table interfaces\n");

    # And the wires table. Strip out the control wires; not needed.
    DBQueryWarn("create temporary table temp_wires ".
		"select t.* from reserved as r ".
		"left join virt_nodes as v on v.vname=r.vname and ".
		"     v.pid=r.pid and v.eid=r.eid ".
833 834
		"left join wires as t on t.node_id1=r.node_id ".
		($singlenet == 0 ? " and t.type='Node' " : " ") .
835
		"where r.pid='$pid' and r.eid='$eid' and ".
836
		"      v.inner_elab_role in ('node','ops','fs','ops+fs') ")
837 838 839
	or die("*** $0:\n".
	       "    Could not create temporary table temp_wires\n");

840 841 842 843 844 845 846
    # And rename the fs/ops node as above.
    if (defined($fsnode)) {
	DBQueryWarn("update temp_wires set node_id1='fs' ".
		    "where node_id1='$fsnode'")
	    or die("*** $0:\n".
		   "    Could not fs node_id in temp_wires\n");
    }
847 848 849 850 851
    DBQueryWarn("update temp_wires set node_id1='ops' ".
		"where node_id1='$opsnode'")
	or die("*** $0:\n".
	       "    Could not ops node_id in temp_wires\n");

852 853 854 855 856 857 858 859 860 861 862 863
    if (! $singlenet) {
	# But we need to take out the wires that are being used as the
	# inner control network, or at least mark them as Control.
	$query_result =
	    DBQueryWarn("select node_id,card,port from temp_interfaces ".
			"where role='" . TBDB_IFACEROLE_CONTROL() . "' ");

	while (my ($node_id,$card,$port) = $query_result->fetchrow_array()) {
	    DBQueryWarn("update temp_wires set type='Control' ".
			"where node_id1='$node_id' and card1=$card and ".
			"      port1=$port");
	}
864 865 866 867 868 869 870 871 872 873 874 875 876
    }
    # Okay, now add the "trunk" wires in without any alteration.
    DBQueryWarn("insert into temp_wires ".
		"select * from wires where type='Trunk'") 
	or die("*** $0:\n".
	       "    Could not add trunk lines to temp_wires\n");

    DBQueryWarn("select * from temp_wires ".
		"into outfile '$statedir/wires'")
	or die("*** $0:\n".
	       "    Could not dump table wires\n");

    #
877 878
    # Ack, we need to create a reservation for the fs and ops nodes,
    # or else they will look free and it will not be able to check in.
879 880 881 882 883 884
    #
    DBQueryWarn("create temporary table temp_reserved ".
		"select r.* from reserved as r ".
		"left join virt_nodes as v on v.vname=r.vname and ".
		"     v.pid=r.pid and v.eid=r.eid ".
		"where r.pid='$pid' and r.eid='$eid' ".
885
		"      and v.inner_elab_role in ('fs','ops','ops+fs')")
886 887
	or die("*** $0:\n".
	       "    Could not create temporary table temp_reserved\n");
888 889 890 891
    if (defined($fsnode)) {
	DBQueryWarn("update temp_reserved set ".
		    "   node_id='fs', ".
		    "   pid='$TBOPSPID', ".
892 893
		    "   eid='opsnodes', ".
		    "   exptidx=1 ".
894 895 896 897
		    "where node_id='$fsnode'")
	    or die("*** $0:\n".
		   "    Could not update temporary table temp_reserved\n");
    }
898 899 900
    DBQueryWarn("update temp_reserved set ".
		"   node_id='ops', ".
		"   pid='$TBOPSPID', ".
901 902
		"   eid='opsnodes', ".
		"   exptidx=1 ".
903 904 905 906 907 908 909 910
		"where node_id='$opsnode'")
	or die("*** $0:\n".
	       "    Could not update temporary table temp_reserved\n");
    DBQueryWarn("select * from temp_reserved ".
		"into outfile '$statedir/reserved'")
	or die("*** $0:\n".
	       "    Could not dump table reserved\n");

911 912 913 914 915 916 917 918 919 920 921 922
    # Copy tiplines table for all nodes so web form gives us a console icon!
    DBQueryWarn("select t.tipname,t.node_id,'',0,0,NULL ".
		"from reserved as r ".
		"left join virt_nodes as v on v.vname=r.vname and ".
		"     v.pid=r.pid and v.eid=r.eid ".
		"left join tiplines as t on t.node_id=r.node_id ".
		"where r.pid='$pid' and r.eid='$eid' and ".
		"      v.inner_elab_role='node' ".
		"into outfile '$statedir/tiplines'")
	or die("*** $0:\n".
	       "    Could not dump table tiplines\n");

923
    #
924
    # Tar up the directory and send it over to (real) ops.
925 926 927
    #
    $UID = 0;
    system("tar cf - -C $statedir . | ".
928
	   "   gzip | $SSH -F /dev/null -host $CONTROL ".
929
	   "   '(cat > $expdir/dbstate.tar.gz)'");
930 931 932 933 934 935 936 937
    if ($?) {
	die("*** $0:\n".
	    "    Could not create dbstate.tar.gz\n");
    }
    $UID = $SAVEUID;
    return 0;
}

938
#
939 940
# Tear down an inner Emulab as cleanly as possible to avoid power cycling
# nodes.
941 942 943 944 945 946
# 
sub TearDownEmulab()
{
    my $tbdir      = "/usr/testbed";
    my $wap        = "$tbdir/sbin/withadminprivs";
    my $nodereboot = "$tbdir/bin/node_reboot";
947

948 949 950 951 952 953 954
    #
    # We want to rebuild the DHCPD file so that when we reboot the inner nodes
    # they come back to the outer emulab. We cannot just free the nodes, cause
    # then the reload daemon might beat us to it, and end up power cycling the
    # nodes, and that would be bad. So, munge the DB and clear the "role" slot
    # for inner nodes. 
    #
955
    DBQueryFatal("update reserved set inner_elab_role=NULL,inner_elab_boot=0 ".
956 957 958 959 960 961 962 963 964 965 966 967 968
		 "where pid='$pid' and eid='$eid'");

    #
    # XXX Failure at this point will leave things in an inconsistent state
    # cause we have just munged the reserved table. Since we were trying
    # to swap out the experiment, I think this will be okay. Wait and see.
    #
    return 0
	if (!defined($bossnode));

    #
    # Now regen the DHCPD file.
    #
969
    # Run as real user since script is setuid.
970 971 972 973 974 975 976 977 978 979
    $EUID = $UID;
    
    print "Regenerating DHCPD config file and restarting daemon.\n";
    system("$makeconf -i -r");
    if ($?) {
	die("*** $0:\n".
	    "    Failed to reconfig/restart DHCPD.\n");
    }
    $EUID = 0;

980 981 982 983 984 985 986 987 988
    #
    # Kill inner vlans table entries; this is the table that maps
    # inner to outer vlans. We do not care about that anymore since
    # all of the vlans are going to be torn down (using the outer
    # ids).
    #
    DBQueryFatal("delete from elabinelab_vlans ".
		 "where pid='$pid' and eid='$eid'");

989
    #
990 991
    # If firewalled, just return now since all nodes will be powered
    # off anyway.
992
    #
993 994
    if ($firewalled) {
	print "Skipping clean shutdown since experiment is firewalled.\n";
995 996 997
	return 0;
    }

998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
    #
    # When the nodes reboot, we want them to do something reasonable. We
    # have no idea what is loaded on the disk, so they should go into an
    # MFS and wait, but then a bunch of nodes will all try to load the big
    # MFS at once, and that could wreak havoc. So, clear the boot osids
    # so they go into PXEWAIT. I could use os_select, but clearing all the
    # OSIDs for a node is apparently a bad thing and generates warnings and
    # emails. Why is that? So just clear the DB state until I figure out
    # why that is.
    #
    DBQueryFatal("update nodes set ".
		 "  def_boot_osid='',next_boot_osid='',temp_boot_osid='' ".
Leigh Stoller's avatar
Leigh Stoller committed
1010 1011
		 "where " .
		    join(" or ", map("node_id='$_'",
1012 1013
				     ($bossnode, $opsnode,
				      defined($fsnode) ? $fsnode : (),
1014
				      @expnodes))));
Leigh Stoller's avatar
Leigh Stoller committed
1015
    
1016 1017 1018 1019 1020
    #
    # SSH in and kill the inner DHCPD daemon so that it does not reply
    # to rebooting nodes along the inner control network.
    #
    $UID = 0;
1021

1022 1023 1024
    print "Killing DHCPD on inner boss ($bossnode)\n";
    system("$SSH -host $bossnode /usr/local/etc/rc.d/2.dhcpd.sh stop");
    if ($?) {
1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
	#
	# This error is non-fatal. If DHCPD cannot be killed, then the inner
	# boss is scrogged or never set up properly. Just return and let
	# the nodes get power cycled (if need be). At some point we need a
	# state machine to control this setup stuff. 
	# 
	print STDERR "*** $0:\n".
	             "    Could not stop DHCPD on inner bossnode ($bossnode)!\n".
		     "    Continuing anyway; outer boss will use power cycle.\n";
	return 0;
1035
    }
1036

1037 1038 1039
    #
    # Now we ask inner boss to reboot all of the testnodes. Maybe need an
    # option to node_reboot, but for now just pass them on the command line.
1040 1041 1042 1043 1044 1045
    #
    if (! @expnodes) {
	$UID = $SAVEUID;
	return 0;
    }
    
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055
    print "Asking inner boss ($bossnode) to reboot inner nodes\n";
    system("$SSH -host $bossnode $wap $nodereboot -b @expnodes");
    if ($?) {
	#
	# This error is non-fatal; Outer boss will just resort to power cycle.
	#
	print STDERR "*** $0:\n".
	             "    Could not reboot some inner nodes!\n".
		     "    Continuing anyway; outer boss will use power cycle.\n";
    }
1056
    $UID = $SAVEUID;
1057

1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
    #
    # Now we wait for them to reach PXEWAIT. Again, use our utility script
    # instead of stated stuff.
    #
    $EUID = $UID;
    print "Waiting for inner nodes to reach PXEWAIT\n";
    system("$nodewait @expnodes");
    if ($?) {
	#
	# This error is non-fatal; Outer boss will just resort to power cycle.
	#
	print STDERR "*** $0:\n".
	             "    Some machines did not reboot properly!\n".
		     "    Continuing anyway; outer boss will use power cycle.\n";
    }
    return 0;
}
1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085

#
# Remove nodes from an inner Emulab.
# 
sub RemoveNodes()
{
    my $tbdir      = "/usr/testbed";
    my $wap        = "$tbdir/sbin/withadminprivs";
    my $nodereboot = "$tbdir/bin/node_reboot";
    my $deletenode = "$tbdir/sbin/deletenode";
    my @nodes	   = ();
1086
    my $paniced    = 0;
1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099

    #
    # If firewalled, check to see if paniced. Right now that means the nodes
    # are going to be powered off, so need to do the clean shutdown dance.
    # 
    if ($firewalled) {
	TBExptGetPanicBit($pid, $eid, \$paniced);
    }

    #
    # Actually, this should not even happen; a paniced experiment cannot be
    # modified at all.
    #
1100 1101
    if ($paniced) {
	print "A paniced experiment cannot be modified! What happened?\n";
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
	return -1;
    }

    #
    # Grab the list of nodes. We want to clear the reserved table bits so
    # that we can regen the DHCPD file. 
    #
    shift(@ARGV);	# pid
    shift(@ARGV);	# eid

    foreach my $node (@ARGV) {
	# Untaint the nodes.
	if ($node =~ /^([-\w]+)$/) {
	    $node = $1;
	}
	else {
	    die("*** Tainted node name: $node\n");
	}
	push(@nodes, $node);
    }
    return 0
	if (!@nodes);

    #
    # Grab the vlans table. We need to find any ports used by the nodes
    # getting deleted, and move them back to the default vlan. 
    #
    my %newvlans = ();
    my @todelete = ();
    
    my $query_result =
	DBQueryWarn("select v.*,e.inner_id from vlans as v ".
		    "left join elabinelab_vlans as e on ".
		    "   e.outer_id=v.id ".
		    "where v.pid='$pid' and v.eid='$eid'");
    return -1
	if (!$query_result);

    while (my (%row) = $query_result->fetchhash()) {
	my $members  = $row{"members"};
	my $id       = $row{"id"};
	my $inner_id = $row{"inner_id"};
	my @newports = ();
	my $changed  = 0;

	foreach my $port (split(/\s+/, $members)) {
	    my ($node,$eth) = split(":", $port);

	    # If this node is not in the list of nodes to be deleted,
	    # the node:port stays in the port list.
	    if (! grep {$_ eq $node} @nodes) {
		push(@newports, $port);
	    }
	    else {
		push(@todelete, $port);
		$changed = 1;
	    }
	}
	$newvlans{$id} = [ @newports ]
	    if ($changed);
    }

    # Remove ports from the vlans.
    if (@todelete) {
	print "Removing ports from deleted nodes: @todelete\n";
	system("$snmpit -m default @todelete");
	if ($?) {
	    return -1;
	}
    }

    # Only if the above succeeds, do we update the vlans table.
    foreach $id (keys(%newvlans)) {
	my $members = join(" ", @{ $newvlans{$id} });

	DBQueryWarn("update vlans set members='$members' ".
		    "where id=$id")
	    or return -1;
    }
    
    #
    # We want to rebuild the DHCPD file so that when we reboot the inner nodes
    # they come back to the outer emulab. We cannot just free the nodes, cause
    # then the reload daemon might beat us to it, and end up power cycling the
    # nodes, and that would be bad. So, munge the DB and clear the "role" and
    # boot slots for nodes about to be released (by tbswap).
    #
    DBQueryWarn("update reserved set inner_elab_role=NULL,inner_elab_boot=0 ".
		"where pid='$pid' and eid='$eid' and (".
		join(" or ", map("node_id='$_'", @nodes)) . ")")
	or return -1;

    #
    # Now regen the DHCPD file.
    #
    # Run as real user since script is setuid.
    $EUID = $UID;
    
    print "Regenerating DHCPD config file and restarting daemon.\n";
    system("$makeconf -i -r");
    if ($?) {
	die("*** $0:\n".
	    "    Failed to reconfig/restart DHCPD.\n");
    }
    $EUID = 0;

    #
    # When the nodes reboot, we want them to do something reasonable. We
    # have no idea what is loaded on the disk, so they should go into an
    # MFS and wait, but then a bunch of nodes will all try to load the big
    # MFS at once, and that could wreak havoc. So, clear the boot osids
    # so they go into PXEWAIT. I could use os_select, but clearing all the
    # OSIDs for a node is apparently a bad thing and generates warnings and
    # emails. Why is that? So just clear the DB state until I figure out
    # why that is.
    #
    DBQueryFatal("update nodes set ".
		 "  def_boot_osid='',next_boot_osid='',temp_boot_osid='' ".
		 "where " .
		   join(" or ", map("node_id='$_'", @nodes)));
    
    #
    # SSH in and kill the inner DHCPD daemon so that it does not reply
    # to rebooting nodes along the inner control network.
    #
    $UID = 0;

    #
    # We are going to do this in a loop, one node at a time. I do not like
    # doing it this way, but its the only reasonable thing to do until we
    # can reboot the inner nodes ourselves (via the outer control network).
    # The reason for doing it one node at a time, is that I cannot delete the
    # node from the inner testbed until its been rebooted. Note that the
    # delete node script regens the dhcpd.conf file, so no need to do that
    # explicitly.
    #
    foreach my $node (@nodes) {
	print "Asking inner boss ($bossnode) to reboot $node\n";
	system("$SSH -host $bossnode $wap $nodereboot -b $node");
	if ($?) {
	    #
	    # This error is non-fatal;
	    # Outer boss will just resort to power cycle.
	    #
	    print STDERR "*** $0:\n".
		         "    Could not reboot $node! Continuing anyway.\n".
			 "    Outer boss will use power cycle.\n";
	}
	print "Asking inner boss ($bossnode) to delete $node\n";
	system("$SSH -host $bossnode sudo -u elabman ".
1252
	       "     $wap $deletenode -b -q -f $node");
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
	if ($?) {
	    #
	    # This error is bad. 
	    #
	    print STDERR "*** $0:\n".
		         "    Could not delete $node! Modify will fail!\n";
	    return -1;
	}
    }
    $UID = $SAVEUID;

    #
    # Now we wait for them to reach PXEWAIT. Again, use our utility script
    # instead of stated stuff.
    #
    $EUID = $UID;
    print "Waiting for inner nodes to reach PXEWAIT\n";
    system("$nodewait @nodes");
    if ($?) {
	#
	# This error is non-fatal; Outer boss will just resort to power cycle.
	#
	print STDERR "*** $0:\n".
	             "    Some machines did not reboot properly!\n".
		     "    Continuing anyway; outer boss will use power cycle.\n";
    }
    return 0;
}

#
# Update an Emulab (add nodes).
# 
sub UpdateEmulab()
{
    my $tbdir      = "/usr/testbed";
    my $wap        = "$tbdir/sbin/withadminprivs";
    my $nodereboot = "$tbdir/bin/node_reboot";
    my $nodewait   = "$tbdir/sbin/node_statewait";
    my @nodes      = ();
1292
    my $paniced    = 0;
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305

    #
    # If firewalled, check to see if paniced. Right now that means the nodes
    # are going to be powered off, so need to do the clean shutdown dance.
    # 
    if ($firewalled) {
	TBExptGetPanicBit($pid, $eid, \$paniced);
    }

    #
    # Actually, this should not even happen; a paniced experiment cannot be
    # modified at all.
    #
1306 1307
    if ($paniced) {
	print "A paniced experiment cannot be modified! What happened?\n";
1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358
	return -1;
    }

    #
    # Grab the list of nodes that have been added to the inner elab.
    #
    my $query_result =
	DBQueryFatal("select node_id from reserved ".
		     "where pid='$pid' and eid='$eid' and ".
		     "      inner_elab_boot=0 and inner_elab_role='node'");
    
    while (my ($node) = $query_result->fetchrow_array()) {
	push(@nodes, $node);	
    }
    return 0
	if (!@nodes);

    # Run as real user for the next few scripts, which are setuid.
    $EUID = $UID;

    #
    # Restart DHCPD, but first mark the nodes as being ready to boot inside
    # the inner emulab, so that dhcpd_makeconf knows what nodes to change
    # the entries for.
    #
    DBQueryFatal("update reserved set inner_elab_boot=1 ".
		 "where pid='$pid' and eid='$eid' and ".
		 "      inner_elab_boot=0 and inner_elab_role='node'");

    print "Regenerating DHCPD config file and restarting daemon.\n";
    system("$makeconf -i -r");
    if ($?) {
	die("*** $0:\n".
	    "    Failed to reconfig/restart DHCPD.\n");
    }

    # Reboot the experimental nodes. They will come up inside the inner elab.
    # DO NOT WAIT! They are not going to report ISUP from this point on. 
    print "Rebooting inner new experimental nodes.\n";
    TBDebugTimeStamp("Rebooting experimental nodes");
    system("$nodereboot @nodes");
    if ($?) {
	die("*** $0:\n".
	    "    Error rebooting the nodes (@nodes)!\n");
    }
    $EUID = 0;

    #
    # At this point, not much I can think of do. The nodes will reboot and
    # enter the newnode MFS. I could add a script to wait for that in the
    # inner elab, but not going to bother yet. 
1359 1360 1361 1362 1363 1364 1365 1366
    #
    # To avoid confusion later (with swapmod, which wants them to be ISUP),
    # and so the web interface does not show the nodes as down, set the 
    # state to ISUP.
    #
    foreach my $node (@nodes) {
	TBSetNodeEventState($node, TBDB_NODESTATE_ISUP());
    }
1367 1368
    return 0;
}