libreboot.pm.in 27.2 KB
Newer Older
1 2 3
#!/usr/bin/perl -wT
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2008 University of Utah and the Flux Group.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
# All rights reserved.
#
# node reboot library. Basically the backend to the node_reboot script, but
# also used where we need finer control of rebooting nodes (and failure).
#
# XXX Only suitable for scripts that are already setuid and have not
# dropped privs by the time they call into nodereboot(). I think this
# makes the library somewhat useless.
#
package libreboot;

use strict;
use Exporter;
use vars qw(@ISA @EXPORT);

@ISA    = "Exporter";
@EXPORT = qw ( nodereboot nodereboot_wait );

# Must come after package declaration!
use lib '@prefix@/lib';
use libdb;
26
use libtblog;
27
use GeniEmulab;
28 29 30 31 32 33 34 35 36 37
use English;

#
# Configure variables
#
my $TB		= "@prefix@";
my $CLIENT_BIN  = "@CLIENT_BINDIR@";
my $BOSSNODE    = "@BOSSNODE@";

#
38 39
# The number of nodes we reboot at a time and the time to wait between batches. 
# The batch size determines the degree of parallelism.
40 41
#
my $BATCHCOUNT  = 12;
42 43 44 45 46 47 48 49 50
my $BATCHSLEEP	= 5;

#
# Default reboot waittime.
# If the user doesn't specify, we use this historic value.
#
# XXX should come from DB.
#
my $MAXWAITTIME	= (6 * 60);
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
use event;
use POSIX qw(strftime);
use IO::Handle;
use Fcntl;

# External Programs
my $ssh		= "$TB/bin/sshtb -n";
my $power	= "$TB/bin/power";
my $ipod	= "$TB/sbin/apod";
my $vnodesetup	= "$TB/sbin/vnode_setup";
my $bisend      = "$TB/sbin/bootinfosend";
my $logfile	= "$TB/log/power.log";
my $ping	= "/sbin/ping";
my $reboot      = "$TB/bin/node_reboot";

# Locals
my $debug       = 0;
75
my $silent      = 0;
76 77 78 79 80 81 82 83
my %children    = ();

#
# The actual library function. 
#
sub nodereboot($$)
{
    my ($args, $result) = @_;
84
    my @nodes       = @{ $args->{'nodelist'}};
85
    my %nodeobjects = ();	# Not in the mood to rewrite this function.
86 87 88 89 90 91 92 93 94

    #
    # Start a new logging sub-process
    #
    my $old_env = \%ENV;
    local %ENV;
    copy_hash %ENV, $old_env;
    tblog_sub_process("reboot", @nodes);

95 96
    $debug = $args->{'debug'}
        if (exists($args->{'debug'}));
97 98
    $silent = $args->{'silent'}
        if (exists($args->{'silent'}));
99 100 101 102 103 104 105 106

    if ($EUID != 0) {
	#
	# Use the exec version, but only if we do not already have
	# a pipeno in the environment, which would indicate a loop
	# caused by installing improperly.
	#
	if (exists($ENV{'REBOOTPIPENO'})) {
107
	    tberror "Must be root when using library!";
108 109 110 111 112 113 114 115
	    return -1;
	}
	print STDERR "reboot: no privs; invoking real nodereboot script!\n"
	    if ($debug);
	return nodereboot_exec($args, $result);
    }

    if (!defined($args->{'nodelist'})) {
116
	tberror "Must supply a node list!";
117 118 119
	return -1;
    }
    my $powercycle  = 0;
120
    my $rebootmode  = 0;
121
    my $waitmode    = 0;
122
    my $waittime    = $MAXWAITTIME;
123 124 125
    my $realmode    = 1;
    my $killmode    = 0;
    my $freemode    = 0;
126
    my $prepare     = 0;
127 128 129 130 131
    my $reconfig    = 0;
    my $asyncmode   = 0;
    my $pipemode    = 0;

    $powercycle  = $args->{'powercycle'} if (exists($args->{'powercycle'}));
132
    $rebootmode  = $args->{'rebootmode'} if (exists($args->{'rebootmode'}));
133
    $waitmode    = $args->{'waitmode'}   if (exists($args->{'waitmode'}));
134
    $waittime    = $args->{'waittime'}   if (exists($args->{'waittime'}));
135 136 137
    $realmode    = $args->{'realmode'}   if (exists($args->{'realmode'}));
    $killmode    = $args->{'killmode'}   if (exists($args->{'killmode'}));
    $freemode    = $args->{'freemode'}   if (exists($args->{'freemode'}));
138
    $prepare     = $args->{'prepare'}    if (exists($args->{'prepare'}));
139 140 141 142 143 144 145 146 147 148 149 150 151 152
    $reconfig    = $args->{'reconfig'}   if (exists($args->{'reconfig'}));
    $asyncmode   = $args->{'asyncmode'}  if (exists($args->{'asyncmode'}));

    #
    # If pipeno is specified in the environment, we write the results to
    # that pipe.
    # 
    if (exists($ENV{'REBOOTPIPENO'})) {
	$pipemode = $ENV{'REBOOTPIPENO'};

	if ($pipemode =~ /^(\d+)$/) {
	    $pipemode = $1;
	}
	else {
153
	    tberror "Bad pipeno in environment: $pipemode!";
154 155 156 157 158 159 160 161 162
	    return -1;
	}
    }

    #
    # Verify permission to reboot these nodes.
    #
    if ($UID && !TBAdmin($UID) &&
	! TBNodeAccessCheck($UID, TB_NODEACCESS_REBOOT, @nodes)) {
163
	tberror "You do not have permission to reboot some of the nodes!";
164 165 166 167 168 169 170
	return -1;
    }

    # Locals.
    my %pids	 = ();
    my $failed   = 0;

171 172 173 174
    # XXX - Wisconsin hack to avoid rebooting routers - this needs to be
    # replaced with the node_capabilities stuff or similar.
    my @temp = ();
    foreach my $node (@nodes) {
175 176
	my $nodeobject = Node->Lookup($node);
	my $rebootable = $nodeobject->rebootable();
177 178
	
	$nodeobjects{$node} = $nodeobject;
179 180 181 182 183
	push(@temp,$node) if $rebootable;
    }
    @nodes = @temp;
    # END XXX

184 185 186 187 188 189 190 191 192 193
    #
    # VIRTNODE HACK: Virtual nodes are special. We can reboot jailed
    # vnodes, but not old style (non-jail). Also, if we are going to
    # reboot the physical node that a vnode is on, do not bother with
    # rebooting the vnode since it will certainly get rebooted anyway!
    #
    my %realnodes = ();
    my %virtnodes = ();

    foreach my $node (@nodes) {
194 195 196 197 198 199 200 201 202 203
	my $nodeobj = $nodeobjects{$node};
	if (!defined($nodeobj)) {
	    tbdie("Could not map $node to its object");
	}
	my $jailed     = $nodeobj->jailflag();
	my $plab       = $nodeobj->isplabdslice();
	my $pnode      = $nodeobj->phys_nodeid();
	my $geninode   = $nodeobj->isfednode();
	my $rebootable = $nodeobj->rebootable();
	
204 205 206
	# All nodes start out as being successful; altered later as needed.
	$result->{$node} = 0;

207 208
	if ($nodeobj->isvirtnode()) {
	    if (!$jailed && !$plab && !$geninode && !$rebootable) {
209 210 211
		print "reboot ($node): Skipping old style virtual node\n";
		next;
	    }
212
	    if (!defined($pnode)) {
213
	        tberror "$node: No physical node!";
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
		return -1;
	    }
	    $virtnodes{$node} = $pnode;
	}
	else {
	    $realnodes{$node} = $node;
	}
    }
    for my $node ( keys(%virtnodes) ) {
	my $pnode = $virtnodes{$node};

	if (defined($realnodes{$pnode})) {
	    print "reboot ($node): Dropping since its host $pnode will reboot\n";
	    delete($virtnodes{$node});
	}
    }
    if (! keys(%realnodes) && ! keys(%virtnodes)) {
	print "reboot: No nodes to reboot.\n";
	return 0;
    }
    my @sortednodes = sort(keys(%realnodes));

    #
    # This stuff is incomplete and non-functional.
    # 
    if (!$realmode) {
	EventSendFatal(host      => $BOSSNODE ,
		       objtype   => TBDB_TBEVENT_COMMAND ,
		       eventtype => TBDB_COMMAND_REBOOT ,
		       objname   => join(",", @sortednodes));
	# In here we can do some output to tell the user what's going on.
	if ($waitmode) {
	    # Wait for [SHUTDOWN,ISUP]
	}
	else {
	    # Wait for [SHUTDOWN]
	}
	return 0;
    }

    #
    # This is somewhat hackish. To promote parallelism, we want to fork off
    # the reboot from the parent so it can do other things.  The problem is
    # how to return status via the results vector. Well, lets do it with
    # some simple IPC. Since the results vector is simply a hash of node
    # name to an integer value, its easy to pass that back.
    #
    # We return the pid to the caller, which it can wait on either directly
    # or by calling back into this library if it wants to actually get the
    # results from the child!
    #
    if ($asyncmode) {
	#
	# Create a pipe to read back results from the child we will create.
	#
	if (! pipe(PARENT_READER, CHILD_WRITER)) {
270
	    tberror "creating pipe: $!";
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296
	    return -1;
	}
	CHILD_WRITER->autoflush(1);

	if (my $childpid = fork()) {
	    close(CHILD_WRITER);
	    $children{$childpid} = [ *PARENT_READER, $result ];
	    return $childpid;
	}
	#
	# Child keeps going. 
	#
	close(PARENT_READER);
	TBdbfork();

	print STDERR "reboot: Running in asyncmode.\n"
	    if ($debug);
    }
    elsif ($pipemode) {
	#
	# We were invoked with an FD already opened to write the results to.
	# See nodereboot_exec() below. The operation from here is basically
	# the same as in asyncmode, but we do not need to create the pipe,
	# but rather just write the results to the pipe we got. 
	#
	if (! open(CHILD_WRITER, ">>&=${pipemode}")) {
297
	    tberror "reopening pipe: $!";
298 299 300 301 302 303 304 305 306
	    return -1;
	}
	CHILD_WRITER->autoflush(1);

	print STDERR "reboot: Running in pipemode ($pipemode).\n"
	    if ($debug);
    }

    #
307 308 309 310 311 312 313 314 315
    # We do not want lots of nodes all rebooting at the same time, it puts
    # a strain on UDP-based PXE/DHCP/TFTP protocols.  So we group them in
    # batches and wait a short time between batches.
    #
    # Currently batches are organized by increasing node order.  In the
    # future we may want to batch based on power controllers; e.g. either
    # reboot all nodes on a power controller at once or AVOID rebooting
    # all nodes on a power controller at once (if controller cannot handle
    # max surge of all attached nodes).
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
    #
    while (@sortednodes) {
	my @batch     = ();
	my $i         = 0;

	while ($i < $BATCHCOUNT && @sortednodes > 0) {
	    my $node = shift(@sortednodes);
	    push(@batch, $node);
	    $i++;
	}

	if ($powercycle) {
	    #
	    # In powercyle mode, call the power program for the whole
	    # batch, and continue on. We do not wait for them to go down or
	    # reboot.
	    #
	    info("*** reboot: Powercyle mode: power cycle ".join(" ",@batch));
	    if (PowerCycle(@batch)) {
335
		tberror "Powercyle failed for " . join(" ",@batch);
336 337 338 339 340 341 342 343 344 345 346 347
		foreach my $node (@batch) {
		    $result->{$node} = -1;
		    $failed++;
		}
	    }
	}
	else {
	    #
	    # Fire off a reboot process so that we can overlap them all.
	    # We need the pid so we can wait for them all before preceeding.
	    #
	    foreach my $node ( @batch ) {
348
		$pids{$node} = RebootNode($nodeobjects{$node}, $reconfig,
349
					  $killmode, $rebootmode, $prepare, $waittime);
350 351 352 353 354 355 356 357 358 359 360
	    }
	}

	#
	# If there are more nodes to go, then lets pause a bit so that we
	# do not get a flood of machines coming up all at the same exact
	# moment.
	#
	if (@sortednodes) {
	    print "reboot: Pausing to give some nodes time to reboot\n"
		if ($debug);
361
	    sleep($BATCHSLEEP);
362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
	}
    }

    #
    # Wait for all the reboot children to exit before continuing.
    #
    my @needPowercycle = ();
    if (scalar(keys(%pids))) {
	foreach my $node (sort(keys(%realnodes))) {
	    my $mypid     = $pids{$node};
	    my $status;

	    #
	    # Child may have avoided the fork, and returned status directly.
	    # Flip it and apply the same test.
	    #
	    if ($mypid <= 0) {
		$status = -$mypid;
	    }
	    else {
		waitpid($mypid, 0);
		$status = $? >> 8;
	    }

	    if ($status == 2) {
		# Child signaled to us that this node needs a power cycle
		push(@needPowercycle, $node);
	    }
390
	    elsif ($mypid != 0 && $?) {
391 392
		$failed++;
		$result->{$node} = -1;
393
		tberror "Failed ($?)!";
394 395 396 397 398 399 400 401 402 403 404 405
	    }
	    else {
		print STDOUT "reboot ($node): Successful!\n";
	    }
	}
    }

    #
    # Power cycle nodes that couldn't be brought down any other way
    #
    if (@needPowercycle) {
	if (PowerCycle(@needPowercycle)) {
406
	    tberror "Powercyle failed for " . join(" ",@needPowercycle);
407 408 409 410 411 412 413 414 415 416 417 418 419
	    foreach my $node (@needPowercycle) {
		$result->{$node} = -1;
		$failed++;
	    }
	}
    }

    #
    # Now do vnodes. Do these serially for now (simple).
    #
    for my $node ( sort(keys(%virtnodes)) ) {
	my $pnode = $virtnodes{$node};

420
	if (RebootVNode($nodeobjects{$node}, $pnode)) {
421 422
	    $failed++;
	    $result->{$node} = -1;
423
	    tberror "$node: Reboot failed (on $pnode)";
424 425
	}
	else {
Leigh Stoller's avatar
Leigh Stoller committed
426 427
	    print STDOUT "reboot ($node): rebooting (on $pnode).\n"
		if (!$silent);
428 429 430 431 432 433 434 435 436
	}
    }

    #
    # Wait for nodes to reboot. We wait only once, no reboots.
    #
    if ($waitmode) {
	my $waitstart = time;

437
	print STDOUT "reboot: Waiting ($waittime s) for nodes to come up.\n"
438
	    if (!$silent);
439 440 441 442 443 444

	# Wait for events to filter through stated! If we do not wait, then we
	# could see nodes still in ISUP.
	sleep(2);

	foreach my $node (sort(@nodes)) {
445 446
	    my $actual_state;
	    
447 448 449 450 451 452
	    #
	    # Skip if something failed earlier.
	    #
	    next
		if ($result->{node});
	    
453
	    if (!TBNodeStateWait($node, $waitstart, $waittime, \$actual_state,
454 455 456 457 458 459 460
				 (TBDB_NODESTATE_TBFAILED,
				  TBDB_NODESTATE_ISUP))) {
		if ($actual_state eq TBDB_NODESTATE_ISUP) {
		    print STDOUT "reboot ($node): alive and well.\n";
		    SetNodeBootStatus($node, NODEBOOTSTATUS_OKAY);
		    next;
		}
461
		tberror "$node reported a TBFAILED event.";
462 463 464 465 466 467 468
	    }
	    SetNodeBootStatus($node, NODEBOOTSTATUS_FAILED);
	    $result->{$node} = -1;
	    $failed++;
	}
    }
 done:    
469 470
    print "reboot: Done. There were $failed failures.\n"
	if (!$silent);
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492

    if ($asyncmode || $pipemode) {
	#
	# We are a child. Send back the results to the parent side
	# and *exit* with status instead of returning it.
	# 
	foreach my $node (keys(%{ $result })) {
	    my $status = $result->{$node};

	    print CHILD_WRITER "$node,$status\n";
	}
	close(CHILD_WRITER);
	exit($failed);
    }
    return $failed;
}

#
# Reboot a node in a child process. Return the pid to the parent so
# that it can wait on all the children later.
#
sub RebootNode {
493
    my ($nodeobject, $reconfig, $killmode, $rebootmode, $prepare) = @_;
494
    my ($status, $syspid, $mypid, $didipod, $nodestate);
495
    my $pc = $nodeobject->node_id();
496

497 498 499 500 501 502 503 504
    #
    # Start a new logging sub-process
    #
    my $old_env = \%ENV;
    local %ENV;
    copy_hash %ENV, $old_env;
    tblog_sub_process("reboot_node", $pc);

505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547
    if ($reconfig) {
	print "reboot ($pc): Attempting to reconfigure ...\n";
    }
    else {
	print "reboot ($pc): Attempting to reboot ...\n";
    }

    # Report some activity into last_ext_act
    TBActivityReport($pc);

    #
    # Is the node in PXEWAIT? If so we want to wake it up so that it can
    # query bootinfo and do what it is supposed to do, without a real reboot.
    # We send the initial wakeup from here, but let stated deal with it
    # failing (timeout) and resending it. That means we could be called
    # with the node in PXEWAKEUP, so send it another wakeup. The point is that
    # stated is controlling the timeouts. Eventually stated gives up and uses
    # the -k option to force a power cycle.
    #
    if (! TBGetNodeEventState($pc, \$nodestate)) {
	info("$pc has no event state: power cycle");
	print STDERR "*** reboot ($pc): no event state; will power cycle.\n"
	    if $debug;
	
	# Signal to the called that the node needs to be power cycled
	return -2;
    }
    if ($nodestate eq TBDB_NODESTATE_PXEWAIT() ||
	$nodestate eq TBDB_NODESTATE_PXEWAKEUP()) {
	#
	# In killmode, we do not want to bother with sending a wakeup event.
	# Just do the power cycle. This is used to unstick a machine that
	# is in waitmode, but not responding to the wakeups.
	#
	if ($killmode) {
	    info("$pc: in $nodestate: but power cycling in killmode");
	    print STDERR "reboot ($pc): in $nodestate, ".
		"but power cycling in killmode.\n" if $debug;
	    
	    # Signal to the caller that the node needs to be power cycled
	    return -2;
	}

548 549 550 551 552 553 554
	my $whol = 0;
	if ($nodeobject->NodeAttribute("wakeonlan_afterpower", \$whol) == 0) {
	    if ($whol) {
		$nodeobject->SimpleWOL();
	    }
	}
	
555 556 557
	#
	# The aux program sends the event to stated ...
	#
558
	my $reqarg = ($rebootmode ? "-r" : "-q");
559
	my $optarg = ($debug ? "-dd" : "");
560
	
561 562
	print STDERR "reboot ($pc): in $nodestate: sending wakeup command.\n"
	    if $debug;
563
	system("$bisend $optarg $reqarg $pc");
564 565
	if ($?) {
	    info("$pc: PXEWAKEUP failed ... power cycle");
566
	    tbnotice "$pc: PXEWAKEUP failed; will power cycle.\n";
567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
	    
	    # Signal to the caller that the node needs to be power cycled
	    return -2;
	}
	return 0;
    }

    #
    # Do the rest in a child process. After the fork, we do the ping test
    # before we reconnect to the DB. This cuts down on the flurry of connects
    # by a bunch of children (ping will take a moment to run).
    #
    $mypid = fork();
    if ($mypid) {
	return $mypid;
    }

    #
    # See if the machine is pingable. If its not pingable, then we just
    # power cycle the machine rather than wait for ssh to time out.
    #
588
    if (! DoesPing($pc, 0)) {
589
	info("$pc appears dead: power cycle");
590
	tbnotice "$pc appears dead; will power cycle.";
591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
	
	# Signal to the parent that the node needs to be power cycled
	exit(2);
    }
    TBdbfork();

    #
    # Machine is pingable at least. Try to reboot it gracefully,
    # or power cycle anyway if that does not work.
    #
    print STDERR "reboot ($pc): Trying ssh.\n" if $debug;

    #
    # Must change our real UID to root so that ssh will work. We save the old
    # UID so that we can restore it after we finish the ssh
    #
    my $oldUID = $UID;
    $UID = 0;

    #
    # If doing a reconfig, first try that in a child.
    #
    if ($reconfig) {
	TBSetNodeEventState($pc, "RECONFIG");
	$syspid = fork();

	if ($syspid) {
	    local $SIG{ALRM} = sub { kill("TERM", $syspid); };
619
	    alarm 30;
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643
	    waitpid($syspid, 0);
	    alarm 0;

	    #
	    # The ssh can return non-zero exit status, but still have worked.
	    # FreeBSD for example.
	    #
	    print "reboot ($pc): reconfig returned $?.\n" if $debug;

	    #
	    # Any failure, revert to plain reboot below.
	    #
	    if (! $?) {
		$UID = $oldUID;
		exit(0);
	    }
	}
	else {
	    exec("$ssh -host $pc ".
		 "/usr/local/etc/emulab/rc/rc.bootsetup -b reconfig");
	    exit(0);
	}
    }

644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674
    #
    # If any of the node's subnodes are being reloaded, wait for the operation
    # to finish before doing the reboot.
    #
    my @subnodes = TBNodeSubNodes($pc);
    foreach my $subnode (@subnodes) {
	my $opmode;
	
	if (TBGetNodeOpMode($subnode, \$opmode) &&
	    defined($opmode) &&
	    (($opmode eq TBDB_NODEOPMODE_RELOADING) ||
	     ($opmode eq TBDB_NODEOPMODE_RELOAD) ||
	     ($opmode eq TBDB_NODEOPMODE_RELOADMOTE))) {
	    my $startwait = time;
	    my $actual_state;

	    print "reboot ($pc): waiting for subnode '$subnode' to finish ".
		"reloading...\n";
	    sleep(5);
	    if (TBNodeStateWait($subnode,
				$startwait,
				(60*10),
				\$actual_state,
				(TBDB_NODESTATE_TBFAILED,
				 TBDB_NODESTATE_ISUP))) {
		print "reboot ($pc): subnode has not finished reloading, ".
		    "rebooting anyways...\n";
	    }
	}
    }

675 676 677 678 679 680 681 682
    #
    # Run an ssh command in a child process, protected by an alarm to
    # ensure that the ssh is not hung up forever if the machine is in
    # some funky state.
    #
    $syspid = fork();

    if ($syspid) {
683 684
	my $timedout = 0;
	local $SIG{ALRM} = sub { kill("TERM", $syspid); $timedout = 1; };
685
	alarm 20;;
686 687 688 689 690 691 692 693 694 695
	waitpid($syspid, 0);
	alarm 0;

	#
	# The ssh can return non-zero exit status, but still have worked.
	# FreeBSD for example.
	#
	print STDERR "reboot ($pc): reboot returned $?.\n" if $debug;

	#
696 697 698 699 700 701 702
	# We used to special case $?==256 here as meaning "ssh is not running"
	# but relying on any return code here is dubious.  Too much depends on
	# the timing of the reboot operation on the client.  So we just check
	# for a self-induced timeout here and immediately send a PoD in that
	# case.  Otherwise, we assume the reboot happened and we will catch
	# our error below if the node does not stop pinging within a couple
	# of seconds.
703
	#
704 705 706
	if ($timedout) {
	    print STDERR "*** reboot ($pc): wedged.\n" if $debug;
	    info("$pc: ssh reboot failed (hung) ... sending ipod");
707 708 709 710 711 712 713 714 715 716 717
	    print STDERR "*** reboot ($pc): Trying Ping-of-Death.\n" if $debug;

	    system("$ipod $pc");
	    $didipod = 1;
	}
	else {
	    info("$pc: ssh reboot ($?)");
	    $didipod = 0;
	}
    }
    else {
718 719 720 721 722
	my $cmd = "/sbin/reboot";
	$cmd = "'/usr/local/etc/emulab/reboot_prepare || $cmd'"
	    if ($prepare);
	
	exec("$ssh -host $pc $cmd");
723 724 725 726 727 728 729 730 731 732
	exit(0);
    }

    #
    # Restore the old UID so that scripts run from this point on get the
    # user's real UID
    #
    $UID = $oldUID;

    #
733 734 735
    # Okay, before we try IPoD or power cycle lets really make sure we need to.
    # We wait a while for the node to stop responding to pings, and if it never
    # goes silent, whack it with a bigger stick.
736
    #
737
    if (WaitTillDead($pc, ($prepare ? 200 : 30)) == 0) {
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
	my $state = TBDB_NODESTATE_SHUTDOWN;
	TBSetNodeEventState($pc,$state);
	exit(0);
    }

    #
    # Haven't yet tried an ipod, try that and wait again.
    # This further slows down reboot but is probably worth it
    # since this should be a rare case (reboot says it worked but
    # node doesn't reboot) and is vital if the nodes have no
    # power cycle capability to fall back on.
    #
    if (! $didipod) {
	info("$pc: reboot failed ... sending ipod");
	$UID = 0;
	system("$ipod $pc");
	$UID = $oldUID;
755
	if (WaitTillDead($pc, 20) == 0) {
756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771
	    my $state = TBDB_NODESTATE_SHUTDOWN;
	    TBSetNodeEventState($pc,$state);
	    exit(0);
	}
    }

    info("$pc: ipod failed ... power cycle");
    print STDERR "*** reboot ($pc): Still running; will power cycle.\n"
	if $debug;
    exit(2);
}

#
# Reboot a vnode in a child process, and wait for it.
#
sub RebootVNode($$) {
772
    my ($nodeobj, $pnode) = @_;
773 774
    my $syspid;

775 776 777 778
    my $vnode      = $nodeobj->node_id();
    my $jailed     = $nodeobj->jailflag();
    my $plab       = $nodeobj->isplabdslice();
    my $geninode   = $nodeobj->isfednode();
779

780 781 782 783 784
    print STDOUT "reboot ($vnode): Rebooting (on $pnode).\n"
	if (!$silent);

    #
    # If we're a plab node (must send the '-p' switch to vnodesetup
785
    # so that plab slivers reboot properly if vnodesetup has died.
786
    #
787

788 789 790 791 792 793 794 795 796
    #
    # Run an ssh command in a child process, protected by an alarm to
    # ensure that the ssh is not hung up forever if the machine is in
    # some funky state.
    #
    $syspid = fork();

    if ($syspid) {
	local $SIG{ALRM} = sub { kill("TERM", $syspid); };
797
	alarm 30;
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822
	waitpid($syspid, 0);
	alarm 0;
	my $exitstatus = $?;

	#
	# The ssh can return non-zero exit status, but still have worked.
	# FreeBSD for example.
	#
	print STDERR "*** reboot ($vnode): returned $exitstatus.\n" if $debug;

	#
	# Look for setup failure, reported back through ssh.
	#
	if ($exitstatus) {
	    if ($exitstatus == 256) {
		print STDERR "*** reboot ($vnode): $pnode is not running sshd.\n"
		    if $debug;
	    }
	    elsif ($exitstatus == 15) {
		print STDERR "*** reboot ($vnode): $pnode is wedged.\n"
		    if $debug;
	    }
	}
	return($exitstatus);
    }
823

824 825 826 827 828 829 830 831 832 833 834 835 836
    if ($geninode) {
	#
	# Do this via the Geni API and return that status to the caller.
	#
	my $experiment = $nodeobj->Reservation();
	if (!defined($experiment)) {
	    print STDERR "*** reboot ($vnode): geninode not allocated.\n";
	    exit(1);
	}
	$EUID = $UID;
	exit(GeniEmulab->StartSlivers($experiment, [ $nodeobj ]));
    }

837 838 839 840 841 842 843
    my $addargs = "";
    if ($plab) {
	$addargs .= "-p ";
    }
    elsif ($jailed) {
	$addargs .= "-j ";
    }
844 845 846
    else {
	$addargs .= "-i ";
    }	
847

848 849 850 851 852
    #
    # Must change our real UID to root so that ssh will work.
    #
    $UID = 0;

853
    exec("$ssh -host $vnode $CLIENT_BIN/vnodesetup -r $addargs $vnode");
854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872
    exit(0);
}

#
# Power cycle a PC using the testbed power program.
#
sub PowerCycle {
    my @pcs = @_;

    my $pcstring = join(" ",@pcs);

    system("$power cycle $pcstring");
    return $? >> 8;
}

#
# Wait until a machine stops returning ping packets.
#
sub WaitTillDead {
873
    my ($pc, $waittime) = @_;
874 875 876 877 878 879 880 881

    print STDERR "reboot ($pc): Waiting to die off.\n" if $debug;

    #
    # Sigh, a long ping results in the script waiting until all the
    # packets are sent from all the pings, before it will exit. So,
    # loop doing a bunch of shorter pings.
    #
882 883
    for (my $i = 0; $i < $waittime; $i++) {
	if (! DoesPing($pc, $i)) {
884 885 886 887 888 889 890 891 892 893 894 895 896
	    print STDERR "reboot ($pc): Died off.\n" if $debug;
	    return 0;
	}
    }
    print STDERR "reboot ($pc): still alive.\n" if $debug;
    return 1;
}

#
# Returns 1 if host is responding to pings, 0 otherwise
# This routine is NOT allowed to do any DB queries!
#
sub DoesPing {
897
    my ($pc, $index) = @_;
898 899 900 901 902 903 904 905 906 907 908 909 910 911
    my $status;
    my $saveuid;

    $saveuid = $UID;
    $UID = 0;
    system("$ping -q -i 0.25 -c 8 -t 2 $pc >/dev/null 2>&1");
    $UID = $saveuid;
    $status = $? >> 8;

    #
    # Ping returns 0 if any packets are returned. Returns 2 if pingable
    # but no packets are returned. Other non-zero error codes indicate
    # other problems.  Any non-zero return indicates "not pingable" to us.
    #
912
    print STDERR "reboot ($pc): $ping $index returned $status\n" if $debug;
913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
    if ($status) {
	return 0;
    }
    return 1;
}


sub info($) {
    my $message = shift;
    # Print out log entries like this:
    # Sep 20 09:36:00 $message
    open(LOG,">> $logfile");
    print LOG strftime("%b %e %H:%M:%S",localtime)." $message\n";
    close(LOG);
}

#
# This gets called in the parent, to wait for an async reboot that was
# launched earlier (asyncmode). The child will print the results back
# on the the pipe that was opened between the parent and child. They
# are stuffed into the original results array.
#
sub nodereboot_wait($)
{
    my ($childpid) = @_;

    if (!exists($children{$childpid})) {
940
	tberror "No such child pid $childpid!";
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
	return -1;
    }
    my ($PARENT_READER, $result) = @{ $children{$childpid}};

    #
    # Read back the results.
    # 
    while (<$PARENT_READER>) {
	chomp($_);

	if ($_ =~ /^([-\w]+),([-\d])+$/) {
	    $result->{$1} = $2;
	    print STDERR "reboot ($1): child returned $2 status.\n";
	}
	else {
956
	    tberror "Improper response from child: $_";
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986
	}
    }
    
    #
    # And get the actual exit status.
    # 
    waitpid($childpid, 0);
    return $? >> 8;
}

#
# Okay, this is horrible! We cannot include this library into non-setuid
# scripts cause a lot of this stuff needs to be run as root. So provide an
# interface that is equiv to the library interface above, but which execs
# an instance of node_reboot and waits for the results.
#
sub nodereboot_exec($$)
{
    my ($args, $result) = @_;
    my $asyncmode = 0;

    #
    # Create a suitable command line from arg hash.
    #
    my $cmdline = "";

    $cmdline .= "-d"
	if (exists($args->{'debug'}) && $args->{'debug'});
    $cmdline .= " -f"
	if (exists($args->{'powercycle'}) && $args->{'powercycle'});
987 988
    $cmdline .= " -b"
	if (exists($args->{'rebootmode'}) && $args->{'rebootmode'});
989 990 991 992 993 994 995 996 997 998
    $cmdline .= " -w"
	if (exists($args->{'waitmode'}) && $args->{'waitmode'});
    $cmdline .= " -r"
	if (exists($args->{'realmode'}) && $args->{'realmode'});
    $cmdline .= " -k"
	if (exists($args->{'killmode'}) && $args->{'killmode'});
    $cmdline .= " -a"
	if (exists($args->{'freemode'}) && $args->{'freemode'});
    $cmdline .= " -c"
	if (exists($args->{'reconfig'}) && $args->{'reconfig'});
999 1000
    $cmdline .= " -W$args->{waittime}"
	if (exists($args->{'waittime'}) && $args->{'waittime'});
1001 1002 1003 1004 1005 1006 1007 1008
    $cmdline .= " @{ $args->{'nodelist'}}";
    
    $asyncmode = $args->{'asyncmode'} if (exists($args->{'asyncmode'}));
	    
    #
    # Create a pipe to read back results from the child we will create.
    #
    if (! pipe(PARENT_READER, CHILD_WRITER)) {
1009
	tberror "creating pipe: $!";
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
	return -1;
    }
    CHILD_WRITER->autoflush(1);

    #
    # ACK! Perl defaults to close-on-exec? Why is that?
    # 
    fcntl(CHILD_WRITER, F_SETFD, 0);

    if (my $childpid = fork()) {
	close(CHILD_WRITER);

	#
	# Save the info we need for calling nodereboot_wait() above.
	# 
	$children{$childpid} = [ *PARENT_READER, $result ];

	#
	# If asyncmode we return right away. The caller will need to
	# use the nodereboot_wait() routine above.
	#
	if ($asyncmode) {
	    return $childpid;
	}

	#
	# Block waiting.
	#
	return nodereboot_wait($childpid);
    }
    close(PARENT_READER);

    #
    # Pass the fd number to the child.
    # 
    $ENV{'REBOOTPIPENO'} = fileno(CHILD_WRITER);
   
    #
    # Child execs the instance of node_reboot, with an extra open fd.
    #
    exec("$reboot $cmdline");
    die("Could not exec $reboot!");
    
}

# _Always_ make sure that this 1 is at the end of the file...
1;