swapexp.in 41.3 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3 4

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2006 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6 7 8
# All rights reserved.
#

9 10
use English;
use Getopt::Std;
11
use POSIX qw(isatty setsid);
12 13

#
Chad Barb's avatar
Chad Barb committed
14
# This gets invoked from the Web interface.
Chad Barb's avatar
 
Chad Barb committed
15
# Swap an experiment in, swap it out, restart or modify.
16
#
Chad Barb's avatar
Chad Barb committed
17

18 19
sub usage()
{
20
    print(STDERR
21
	  "Usage: swapexp [-q] [-b | -w] [-i | -a | -f] [-r] [-e]\n".
22 23 24 25
	  "               <-s in | out | restart | modify | pause>\n".
	  "               <pid> <eid> [<nsfile>]\n".
	  "switches and arguments:\n".
	  "-w       - wait for non-batchmode experiment swap/modify\n".
26
	  "-q       - be less chatty\n".
27 28 29 30 31 32
	  "-r       - reboot nodes when doing a modify experiment\n".
	  "-e       - restart event scheduler when doing a modify experiment\n".
	  "-s <op>  - Operation to perform; one of those listed above\n".
	  "<pid>    - The project the experiment belongs to\n".
	  "<eid>    - The experiment name (id)\n".
	  "<nsfile> - Optional NS file to parse for experiment modify\n");
33 34
    exit(-1);
}
35 36 37

sub fatal($);

38
my  $optlist = "biafres:wq";
39

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
#
# Exit codes are important; they tell the web page what has happened so
# it can say something useful to the user. Fatal errors are mostly done
# with die(), but expected errors use this routine. At some point we will
# use the DB to communicate the actual error.
#
# $status < 0 - Fatal error. Something went wrong we did not expect.
# $status = 0 - Termination is proceeding in the background. Notified later.
# $status > 0 - Expected error. User not allowed for some reason. 
# 
sub ExitWithStatus($$)
{
    my ($status, $message) = @_;
    
    if ($status < 0) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
55
	tbdie($message);
56 57
    }
    else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
58
	tbnotice($message);
59 60 61 62
    }
    exit($status);
}

63 64 65 66 67
#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
68
my $TBROBOCOPS = "@TBROBOCOPSEMAIL@";
69
my $TBLOGS = "@TBLOGSEMAIL@";
70
my $TBINFO = "$TB/expinfo";
71
my $TBDOCBASE = "@TBDOCBASE@";
72
my $TBBASE = "@TBBASE@";
73
my $CONTROL  = "@USERNODE@";
74 75 76 77 78 79 80

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
Kevin Atkinson's avatar
 
Kevin Atkinson committed
81
use libtblog;
82
use libArchive;
83

84 85 86 87 88 89 90 91 92
# Be careful not to exit on transient error; 0 means infinite retry.
$libdb::DBQUERY_MAXTRIES = 0;

# For the END block below.
my $cleaning = 0;
my $justexit = 1;
my $signaled = 0;

my $tbdir    = "$TB/bin";
93
my $tbdata   = "tbdata";
94
my $checkquota = "$TB/sbin/checkquota";
95
my $batch    = 0;
96
my $idleswap = 0;
97 98
my $autoswap = 0;
my $force    = 0;
Chad Barb's avatar
Chad Barb committed
99
my $reboot   = 0;
100
my $waitmode = 0;
101
my $quiet    = 0;
102
my $eventsys_restart   = 0;
103
my $errorstat= -1;
104 105
my $modifyHosed   = 0;
my $modifySwapped = 0;
106
my $robotexp = 0;
Chad Barb's avatar
 
Chad Barb committed
107

108 109 110 111 112
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
113
my @allnodes;
114
my @row;
115
my $action;
116
my $tag;
117
my $nextswapstate;
118
my $termswapstate;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
119
my $isadmin  = 0;
Chad Barb's avatar
 
Chad Barb committed
120

121 122 123
#
# Untaint the path
# 
124
$ENV{'PATH'} = "/bin:/usr/bin:$TB/libexec/vis";
125 126 127 128 129 130 131
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

132 133 134 135 136 137 138
#
# Set umask for start/swap. We want other members in the project to be
# able to swap/end experiments, so the log and intermediate files need
# to be 664 since some are opened for append.
#
umask(0002);

139 140 141 142 143 144 145 146
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
147 148 149
if (defined($options{"i"})) {
    $idleswap = 1;
}
150 151 152
if (defined($options{"w"})) {
    $waitmode = 1;
}
153 154 155 156 157 158
if (defined($options{"a"})) {
    $autoswap = 1;
}
if (defined($options{"f"})) {
    $force = 1;
}
159 160 161
if (defined($options{"b"})) {
    $batch = 1;
}
Chad Barb's avatar
 
Chad Barb committed
162 163 164
if (defined($options{"r"})) {
    $reboot = 1;
}
165 166 167
if (defined($options{"e"})) {
    $eventsys_restart = 1;
}
168 169 170
if (defined($options{"q"})) {
    $quiet = 1;
}
171 172 173
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
174 175 176
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
177
	$inout ne "pause"   &&
Chad Barb's avatar
 
Chad Barb committed
178
	$inout ne "modify") {
179 180 181 182 183 184 185
	usage();
    }
}
else {
    usage();
}

186 187 188 189 190
usage()
    if (($waitmode && $batch) ||
	($inout ne "modify" && @ARGV != 2) ||
	(($waitmode || $batch) && ($idleswap || $autoswap || $force)));

191 192 193 194 195
if ($eventsys_restart && $inout ne "modify") {
    print STDOUT "Usage: swapexp: -e (eventsys_restart) can be used ".
                 "only with -s modify\n";
    usage();
}
Chad Barb's avatar
 
Chad Barb committed
196 197 198
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

199 200 201
#
# Untaint the arguments.
#
202
if ($pid =~ /^([-\w\.]+)$/) {
203 204 205
    $pid = $1;
}
else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
206
    tbdie("Tainted argument $pid!");
207
}
208
if ($eid =~ /^([-\w\.]+)$/) {
209 210 211
    $eid = $1;
}
else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
212
    tbdie("Tainted argument $eid!");
213
}
214
my $repfile = "$eid.report";
215 216
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
217 218
my $tempnsfile;
my $modnsfile;
219
my $nsfile;
220

Leigh B. Stoller's avatar
Leigh B. Stoller committed
221
if ($inout eq "modify" && @ARGV > 2) {
222 223 224 225 226
    $tempnsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
227
    if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
228
	$tempnsfile = $1;
229 230
    }
    else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
231
	tbdie("Tainted nsfile name: $tempnsfile");
232 233 234 235 236 237 238 239 240
    }
    #
    # Called from ops interactively. Make sure NS file in /proj or /users.
    #
    # Use realpath to resolve any symlinks.
    #
    my $translated = `realpath $tempnsfile`;
    if ($translated =~ /^([-\w\.\/]+)$/) {
	$tempnsfile = $1;
241
    }
242
    else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
243
	tbdie("Tainted nsfile returned by realpath: $translated");
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258
    }

    #
    # The file must reside in /proj, /groups, or /users. Since this script
    # runs as the caller, regular file permission checks ensure its a file
    # the user is allowed to use. /tmp/$guid-$nsref.nsfile also allowed
    # since this script is invoked directly from web interface, which generates
    # a name that should not be guessable, so as long as it looks to be in
    # proper format, we accept it. 
    #
    if (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
	! ($tempnsfile =~ /^\/var\/tmp\/php\w+/) &&
	! ($tempnsfile =~ /^\/proj/) &&
	! ($tempnsfile =~ /^\/groups/) &&
	! ($tempnsfile =~ /^\/users/)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
259
	tbdie("$tempnsfile does not resolve to an appropriate directory!");
260 261 262
    }

    if (! -f $tempnsfile || -z $tempnsfile || ! -r $tempnsfile) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
263
	tbdie("$tempnsfile does not look like an NS file!");
264
    }
265 266
    $nsfile    = "$eid.ns";
    $modnsfile = "${eid}-modify.ns";
267
}
268 269 270 271 272

#
# Verify user and get his DB uid.
#
if (! UNIX2DBUID($UID, \$dbuid)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
273
    tbdie("You do not exist in the Emulab Database.");
274 275 276 277 278 279
}

#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
280
    tbdie("Cannot determine your name and email address.");
281
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
282
$isadmin = TBAdmin($UID);
283

Kevin Atkinson's avatar
 
Kevin Atkinson committed
284 285 286 287 288
#
# Set error reporting info
#
tblog_set_info($pid,$eid,$UID);

289
#
Chad Barb's avatar
 
Chad Barb committed
290
# Verify that this person can muck with the experiment.
291 292
# Note that any script down the line has to do an admin check also. 
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
293
if ($UID && !$isadmin &&
294
    !TBExptAccessCheck($dbuid, $pid, $eid, TB_EXPT_DESTROY)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
295
    tbdie("You do not have permission to swap or modify this experiment!");
296 297
}

298 299 300 301
# Must do this before lock tables!
# idleswap is in minutes, threshold is in hours
$idleswap_time = 60 * TBGetSiteVar("idle/threshold");

302 303 304 305 306 307 308 309 310
#
# In wait mode, block interrupt until we spin off the background process.
#
if ($waitmode) {
    $SIG{TERM} = 'IGNORE';
    $SIG{QUIT} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

311 312 313 314 315 316
#
# Check for overquota; we deal with it below, cause of the batch system.
#
my $overquota = system("$checkquota $dbuid");

if ($overquota) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
317
    tberror("You are over your disk quota on $CONTROL; please cleanup!");
318 319
}

320 321 322 323 324 325
#
# Temp fix; Disallow swapmod to firewalled experiments. This will come
# out later.
#
my $firewalled = TBExptFirewall($pid, $eid);

326 327 328 329 330 331 332 333 334 335
#
# see if we've got a robot exp (this isn't the only check; if this is a
# swapmod, we've got to check tbprerun as well...
#
$robotexp = 
    TBExptContainsNodeCT($pid,$eid,'robot') || 
    TBExptContainsNodeCT($pid,$eid,'mote') || 
    TBExptContainsNodeCT($pid,$eid,'motehost') || 
    TBExptContainsNodeCT($pid,$eid,'powermon');

336 337 338 339 340
#
# We have to protect against trying to end an experiment that is currently
# in the process of being terminated. We use a "wrapper" state (actually
# a timestamp so we can say when termination was requested) since
# terminating consists of a couple of different experiment states down inside
Chad Barb's avatar
Chad Barb committed
341
# the tb scripts.
342 343 344 345 346 347 348
#
DBQueryFatal("lock tables experiments write");

$query_result =
    DBQueryFatal("SELECT * FROM experiments WHERE eid='$eid' and pid='$pid'");

if (! $query_result->numrows) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
349
    tbdie("No such experiment $pid/$eid exists!");
350 351 352
}
my %hashrow = $query_result->fetchhash();
my $expt_head_login = $hashrow{'expt_head_uid'};
353
my $last_swap_uid   = $hashrow{'expt_swap_uid'};
354
my $estate          = $hashrow{'state'};
355
my $batchstate      = $hashrow{'batchstate'};
356
my $expt_path       = $hashrow{'path'};
357
my $expt_locked     = $hashrow{'expt_locked'};
358
my $isbatchexpt     = $hashrow{'batchmode'};
359
my $canceled        = $hashrow{'canceled'};
360 361 362 363 364 365 366 367 368 369
my $swappablebit= $hashrow{'swappable'};
my $idleswapbit = $hashrow{'idleswap'};
my $autoswapbit = $hashrow{'autoswap'};
my $swappablestr= ( $swappablebit ? "Yes" : "No" );
my $idleswapstr = ( $idleswapbit ? "Yes" : "No" );
my $autoswapstr = ( $autoswapbit ? "Yes" : "No" );
my $noswap      = $hashrow{'noswap_reason'};
my $noidleswap  = $hashrow{'noidleswap_reason'};
my $idleswaptime= $hashrow{'idleswap_timeout'} / 60.0;
my $autoswaptime= $hashrow{'autoswap_timeout'} / 60.0;
370
my $rendering   = $hashrow{'prerender_pid'};
371
my $elabinelab  = $hashrow{'elab_in_elab'};
372
my $lockdown    = $hashrow{'lockdown'};
373

374 375
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
376
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
377 378 379
    $idleswaptime = $idleswap_time / 60.0;
}

380 381
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
382
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
383 384
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
385

386
if (! chdir($workdir)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
387
    tbdie("Could not chdir to $workdir: $!");
388 389
}

390
#
391 392 393
# This script is called from the batch daemon.
# 
if ($batch) {
394
    #
395 396 397
    # Sanity Check. If called from the daemon, must already be locked,
    # must be a batch experiment, and must be in proper state for the
    # operation requested. 
398
    #
Kevin Atkinson's avatar
 
Kevin Atkinson committed
399
    tbdie("Experiment $pid/$eid is supposed to be a batch experiment!")
400 401
	if (!$isbatchexpt);
    
Kevin Atkinson's avatar
 
Kevin Atkinson committed
402
    tbdie("Batch experiment $pid/$eid should be locked!")
403 404
	if (!defined($expt_locked) ||
	    $batchstate ne BATCHSTATE_LOCKED());
Kevin Atkinson's avatar
 
Kevin Atkinson committed
405 406
    
    tbdie("Batch experiment $pid/$eid is locked down; cannot be swapped!")
407 408
	if ($lockdown);

409
    if ($inout eq "in") {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
410 411
	tbdie("Batch experiment $pid/$eid is not in the proper state!\n".
	      "Currently $estate, but should be QUEUED.")
412 413
	    if ($estate ne EXPTSTATE_QUEUED);
	
Kevin Atkinson's avatar
 
Kevin Atkinson committed
414
	tbdie("Batch experiment $pid/$eid has been canceled! Aborting.")
415
	    if ($canceled);
416 417

	# Do not allow it to swap in. What about swapout? 
Kevin Atkinson's avatar
 
Kevin Atkinson committed
418
	tbdie("Batch experiment cannot swap in when over quota! Aborting.")
419
	    if ($overquota);
420 421
    }
    elsif ($inout eq "out") {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
422 423
	tbdie("Batch experiment $pid/$eid is not in the proper state!\n".
	      "Currently $estate, but should be ACTIVE.")
424
	    if ($estate ne EXPTSTATE_ACTIVE);
425 426
    }
    else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
427
	tbdie("Improper request from batch daemon for $pid/$eid!\n");
428 429 430 431
    }
}
else {
    if ($isbatchexpt) {
432 433 434 435
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
436 437
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
438
	    if ($canceled);
439

440 441 442
	ExitWithStatus(1, "Batch experiment $pid/$eid is locked down!")
	    if ($lockdown);

443
	if ($inout eq "in") {
444
	    ExitWithStatus(1,
445 446 447
			   "Batch experiment $pid/$eid must be SWAPPED to\n".
			   "QUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_SWAPPED);
448 449 450 451 452 453

	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid cannot swap in when ".
			   "over quota!\n")
		if ($overquota);
	    
454
	    SetExpState($pid, $eid, EXPTSTATE_QUEUED);
455 456
	}
	elsif ($inout eq "out") {
457
	    ExitWithStatus(1,
458 459 460 461
			   "Batch experiment $pid/$eid must be ACTIVE or\n".
			   "ACTIVATING to swap out. Currently $estate.")
		if ($estate ne EXPTSTATE_ACTIVE &&
		    $estate ne EXPTSTATE_ACTIVATING);
462 463 464 465 466

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
467
	    TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
468 469
	}
	elsif ($inout eq "pause") {
470
	    ExitWithStatus(1,
471 472 473
			   "Batch experiment $pid/$eid must be QUEUED to\n".
			   "DEQUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_QUEUED);
474 475

	    #
476 477 478 479
	    # XXX. The batch daemon might already have the experiment, but
	    # not have shipped it off to startexp. Change the state
	    # anyway. The error will be noticed later when startexp dies,
	    # and the batch daemon gets the error back. This sucks.
480
	    #
481
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
482
	}
483
	elsif ($inout eq "modify") {
484
	    ExitWithStatus(1,
485 486 487 488
			   "Batch experiment $pid/$eid must be SWAPPED or\n".
			   "ACTIVE to modify. Currently $estate.")
		if (($estate ne EXPTSTATE_SWAPPED &&
		     $estate ne EXPTSTATE_ACTIVATING) ||
489
		    $batchstate ne BATCHSTATE_UNLOCKED());
490

491 492 493
	    ExitWithStatus(1,
			"Cannot modify an active firewalled experiment (yet).")
		if ($firewalled && $estate ne EXPTSTATE_SWAPPED && !$isadmin);
494

495 496 497
	    ExitWithStatus(1,
			"Cannot modify an active ElabInElab experiment (yet).")
		if ($elabinelab && $estate ne EXPTSTATE_SWAPPED && !$isadmin);
498 499 500 501 502 503

	    ExitWithStatus(1,
			   "Cannot modify batch experiment $pid/$eid when ".
			   "over quota!\n")
		if ($overquota);
	    
504
	    #
505
	    # Otherwise, proceed with the modify. The experiment will be
506 507
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
508 509 510 511
	    # the experiments table. The batch daemon will leave it alone
	    # until the modify is done. If the modify fails and cannot recover
	    # it is going to get swapped out; that is okay since the batch
	    # daemon does not keep state internally. 
512
	    #
513 514
	    goto doit;
	}
515
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
516
	    tbdie("Operation $inout not allowed on a batch experiment!");
517
	}
518 519
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
520
      doit:
521
    }
522 523 524 525 526 527 528 529 530 531
    else {
	#
	# If the cancel flag is set, then user must wait for that to
	# clear before we can do anything else.
	#
	ExitWithStatus(1,
		       "Experiment $pid/$eid has its cancel flag set!.\n".
		       "You must wait for that to clear before you can swap\n".
		       "or modify the experiment.\n")
	    if ($canceled);
532

533 534 535 536
 	ExitWithStatus(1,
		       "Experiment $pid/$eid is locked down; cannot swap!\n")
	    if ($lockdown);

537 538 539 540 541 542 543 544 545 546
	#
	# Check the state for the various operations.
	#
	if (!$force) {
	  SWITCH: for ($inout) {
	      /^in$/i && do {
		  if ($estate ne EXPTSTATE_SWAPPED()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped out!");
		  }
547 548 549 550 551
		  ExitWithStatus(1,
				 "Experiment $pid/$eid cannot swap in when ".
				 "over quota!\n")
		      if ($overquota);
		  
552 553 554 555
		  last SWITCH;
	      };
	      /^out$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
Leigh B. Stoller's avatar
Leigh B. Stoller committed
556
 		      $estate ne EXPTSTATE_PANICED() &&
557 558 559 560 561 562
		      $estate ne EXPTSTATE_ACTIVATING()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped in ".
				     "or activating!\n");
		  }
		  
Leigh B. Stoller's avatar
Leigh B. Stoller committed
563 564 565 566 567 568 569 570 571 572 573 574
 		  #
 		  # Must be an admin person to swap out an experiment that
 		  # has had its panic button pressed.
 		  #
 		  if ($estate eq EXPTSTATE_PANICED() && !$isadmin) {
 		      ExitWithStatus(1,
 				     "Experiment $pid/$eid had its panic ".
 				     "button pressed!\n".
 				     "Only a testbed administrator can swap ".
 				     "this experiment out.");
 		  }

575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603
		  if ($estate eq EXPTSTATE_ACTIVATING()) {
		      #
		      # All we can do is set the cancel flag and hope that
		      # it gets noticed. We do not wait. 
		      # 
		      TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
		      
		      ExitWithStatus(0,
				     "Experiment $pid/$eid swapin has been  ".
				     "marked for cancelation.\n".
				     "You will receive email when the original ".
				     "swap request has finished.");
		  }
		  last SWITCH;
	      };
	      /^restart$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped in!");
		  }
		  last SWITCH;
	      };
	      /^modify$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
		      $estate ne EXPTSTATE_SWAPPED()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid must be ACTIVE or\n".
				     "SWAPPED to modify!\n");
		  }
604
		  ExitWithStatus(1,
605 606 607 608
			"Cannot modify an active firewalled experiment (yet).")
		      if ($firewalled &&
			  $estate ne EXPTSTATE_SWAPPED && !$isadmin);

609
		  ExitWithStatus(1,
610 611 612
			"Cannot modify an active ElabInElab experiment (yet).")
		      if ($elabinelab &&
			  $estate ne EXPTSTATE_SWAPPED && !$isadmin);
613 614 615 616 617

		  ExitWithStatus(1,
				 "Experiment $pid/$eid cannot be modified ".
				 "when over quota!\n")
		      if ($overquota);
618
		  
619 620
		  last SWITCH;
	      };
Kevin Atkinson's avatar
 
Kevin Atkinson committed
621
	      tbdie("Missing state check for action: $action");
622
	  }
623 624
	}
    }
625 626
}

627 628 629 630 631 632 633
#
# Determine the temporary and next state for experiment. If the experiment
# is a batch experiment, then the next state is actually handled by the
# batch daemon, but we still have to deal with the temporary state. 
#
SWITCH: for ($inout) {
    /^in$/i && do {
634
	$nextswapstate = EXPTSTATE_ACTIVATING();
635 636 637
	last SWITCH;
    };
    /^out$/i && do {
638
	$nextswapstate = EXPTSTATE_SWAPPING();
639 640 641
	last SWITCH;
    };
    /^restart$/i && do {
642
	$nextswapstate = EXPTSTATE_RESTARTING();
643 644 645
	last SWITCH;
    };
    /^modify$/i && do {
646 647
	$nextswapstate = (($estate eq EXPTSTATE_SWAPPED()) ?
			  EXPTSTATE_MODIFY_PARSE() : EXPTSTATE_MODIFY_REPARSE());
648 649
	last SWITCH;
    };
Kevin Atkinson's avatar
 
Kevin Atkinson committed
650
    tbdie("Missing state check for action: $action");
651
}
652 653
 
# Update idleswap_timeout to whatever the current value is.
654
if ($inout ne "out") {
655 656 657
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
658

659 660 661 662 663 664
#
# On a failure, we go back to this swapstate. Might be modified below.
# 
$termswapstate = $estate;

# Lock the record, set the nextstate, and unlock the table.
665
TBLockExp($pid, $eid, $nextswapstate)
Kevin Atkinson's avatar
 
Kevin Atkinson committed
666
    or tbdie("Failed to set experiment state to $nextswapstate");
667 668 669 670 671 672 673

#
# At this point, we need to force a cleanup no matter how we exit.
# See the END block below.
#
$justexit = 0;

674 675 676 677
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
678 679 680 681
# inconsistent state. Be sure to call fatal() only since we are
# going into the background, and we have to send email since no
# one is going to see printed error messages (output goes into the
# log file, which will be sent along in the email). 
682 683
#

684 685
if ($inout eq "in") {
    $action = "swapped in";
686
    $tag    = "swapin";
687 688 689
}
if ($inout eq "out") {
    $action = "swapped out";
690
    $tag    = "swapout";
691 692 693 694
}
if ($inout eq "restart") {
    $action = "restarted";
}
Chad Barb's avatar
 
Chad Barb committed
695 696
if ($inout eq "modify") {
    $action = "modified";
697
    $tag    = "swapmod";
Chad Barb's avatar
 
Chad Barb committed
698
}
699

700 701 702 703 704 705 706 707 708
#
# Get email address of the experiment head, which may be different than
# the person who is actually terminating the experiment, since its polite
# to let the original creator know whats going on. 
#
my $expt_head_name;
my $expt_head_email;

if (! UserDBInfo($expt_head_login, \$expt_head_name, \$expt_head_email)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
709
    tbwarn("Could not determine name/email for $expt_head_login.");
710 711 712 713
    $expt_head_name  = "TBOPS";
    $expt_head_email = $TBOPS;
}

714 715 716
#
# Before going to background, we have to copy out the NS file!
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
717
if ($inout eq "modify" && defined($modnsfile)) {
718 719
    unlink($modnsfile);
    if (system("/bin/cp", "$tempnsfile", "$modnsfile")) {
720
	fatal("Could not copy $tempnsfile to $modnsfile");
721 722 723 724
    }
    chmod(0664, "$modnsfile");
}

725 726 727 728
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
729
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
730
    TBExptSetLogFile($pid, $eid, $logname);
731
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
732

733 734
    if (my $childpid = TBBackGround($logname)) {
	#
735 736
	# Parent exits normally, unless in waitmode. We have to set
	# justexit to make sure the END block below does not run.
737
	#
738 739
	$justexit = 1;

740
	if (!$waitmode) {
741 742 743
	    print("Experiment $pid/$eid is now being $action.\n".
		  "You will be notified via email when the this is done.\n")
		if (! $quiet);
744 745
	    exit(0);
	}
746 747 748 749 750 751 752 753
	print("Waiting for experiment $eid to finish its swap${action}\n")
	    if (! $quiet);
	    
	if (isatty(STDIN) && !$quiet) {
	    print("You may type ^C at anytime; you will be notified via email.".
		  "\n".
		  "You will not actually interrupt the experiment itself.\n");
	}
754 755 756 757 758 759 760 761 762 763 764 765
	
	# Give child a chance to run.
	select(undef, undef, undef, 0.25);
	
	#
	# Reset signal handlers. User can now kill this process, without
	# stopping the child.
	#
	$SIG{TERM} = 'DEFAULT';
	$SIG{INT}  = 'DEFAULT';
	$SIG{QUIT} = 'DEFAULT';

766
	#
767
	# Wait until child exits or until user gets bored and types ^C.
768
	#
769 770
	waitpid($childpid, 0);
	
771 772
	print("Done. Exited with status: $?\n")
	    if (! $quiet);
773
	exit($? >> 8);
774
    }
775
    TBdbfork();
776 777
}

778 779 780 781 782 783 784 785
#
# When in waitmode, must put ourselves in another process group so that
# an interrupt to the parent will not have any effect on the backend.
#
if ($waitmode) {
    POSIX::setsid();
}

786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
#
# We need to catch TERM cause sometimes shit happens and we have to kill
# an experiment swap that is hung or otherwise scrogged. Rather then 
# trying to kill off the children one by one, lets arrange to catch it
# here and send a killpg to the children. This is not to be done lightly,
# cause it can leave things worse then they were before!
#
sub handler ($) {
    my ($signame) = @_;
    
    $SIG{TERM} = 'IGNORE';
    my $pgrp = getpgrp(0);
    kill('TERM', -$pgrp);
    sleep(1);
    $signaled = 1;
    fatal("Caught SIG${signame}! Killing experiment setup ...");
}
$SIG{TERM} = \&handler;
$SIG{QUIT} = 'DEFAULT';

806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821
#
# Gather stats; start clock ticking
#
if ($inout eq "in") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0,
		    TBDB_STATS_FLAGS_START);
}
elsif ($inout eq "out") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
		    TBDB_STATS_FLAGS_START);
}
elsif ($inout eq "modify") {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0,
		    TBDB_STATS_FLAGS_START);
}

822 823 824
#
# Remove old report file since its contents are going to be invalid.
#
825
if ($inout ne "restart" && -e $repfile) {
826 827 828
    unlink("$repfile");
}

829 830 831 832
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
833
    my $optarg = (($force || $idleswap) ? "-force" : "");
834

835 836
    print STDOUT "Running 'tbswap out $optarg $pid $eid'\n";
    if (system("$tbdir/tbswap out $optarg $pid $eid") != 0) {
837
	$errorstat = $? >> 8;
838
	fatal("tbswap out failed!");
839
    }
840 841 842 843 844 845 846 847

    #
    # Add the files that have been detected by tracing to the archive.
    #
    if (libArchive::TBExperimentArchiveAddTracedFiles($pid, $eid) < 0) {
	fatal("Failed to add traced files to the experiment archive!");
    }

Leigh B. Stoller's avatar
Leigh B. Stoller committed
848 849 850 851 852 853 854
    #
    # Add the special per-experiment archive directory.
    #
    if (libArchive::TBExperimentArchiveAddUserFiles($pid, $eid) < 0) {
	fatal("Failed to add user specified files to the experiment archive!");
    }

855 856
    SetExpState($pid, $eid, EXPTSTATE_SWAPPED)
	or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
Leigh B. Stoller's avatar
Leigh B. Stoller committed
857
    TBExptClearPanicBit($pid, $eid);
858
}
859
elsif ($inout eq "in") {
860 861
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPIN, 0, TBDB_STATS_FLAGS_PRESWAPIN);
862 863 864 865

    # Set the swapper now so that nodes use the proper uid. If the swapin
    # fails, we leave the swapper as is, since its harmless and informative.
    TBExptSetSwapUID($pid, $eid, $dbuid);
866
    
867
    print STDOUT "Running 'tbswap in $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
868
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
869
	$errorstat = $? >> 8;
870
	fatal("tbswap in failed!");
871
    }
872 873
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE)
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
874

875
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
876
}
Chad Barb's avatar
 
Chad Barb committed
877
elsif ($inout eq "modify") {
878
    my $modifyError;
Chad Barb's avatar
Chad Barb committed
879

880 881 882
    #
    # Prepare the Archive for the swapmod, in case we have to "roll back".
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
883
    print "Doing a preswapmod on the experiment archive ...\n";
884 885 886 887
    if (libArchive::TBExperimentArchivePreSwapMod($pid, $eid) < 0) {
	fatal("Failed to do a preswapmod on the experiment archive!");
    }

888 889 890
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

891 892 893 894 895
    # Gather up some firewall state for later comparison.
    if (GatherFWinfo() < 0) {
	fatal("Could not gather firewall info; cannot safely continue!");
    }

Chad Barb's avatar
Chad Barb committed
896
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
897
    if (TBExptBackupVirtualState($pid, $eid)) {
898
	fatal("Could not backup experiment state; cannot safely continue!");
Chad Barb's avatar
Chad Barb committed
899 900 901
    }

    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
902
    # Rerun tbprerun if modifying, but only if new NS file provided.
903 904
    # Yep, we allow reswap without changing the NS file. For Shashi and SIM.
    # Note that tbprerun kills the renderer if its running.
Chad Barb's avatar
Chad Barb committed
905
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
906 907 908
    if (defined($modnsfile)) {
	print STDOUT "Running 'tbprerun $pid $eid $modnsfile'\n";
	if (system("$tbdir/tbprerun $pid $eid $modnsfile") != 0) {
909
	    print STDOUT "Modify Error: tbprerun failed.\n";
910
	  FWHOSED:
911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928
	    print STDOUT "Recovering experiment state...\n";

	    if (TBExptRemoveVirtualState($pid, $eid) ||
		TBExptRestoreVirtualState($pid, $eid)) {
		$modifyHosed = 1;
		fatal("Experiment state could not be restored!");
		# Never returns;
	    }
	    #
	    # If the renderer was running when we started the swapmod, then we
	    # want to restart it. If it was stopped, then the renderer info
	    # was captured with the rest of the virtual state (restored above).
	    #
	    system("prerender -t $pid $eid")
		if ($rendering);

	    fatal("Update aborted; old virtual state restored.");
	    # Never returns;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
929
	}
930 931 932 933 934 935 936 937 938
	#
	# Okay, whenever a new NS file is presented, we need to do some
	# checks on the firewall to make sure the user is not trying to
	# do something "unsafe". 
	#
	if (CheckFWinfo($estate) != 0) {
	    # All the stuff for recovering is right above, so go there. 
	    goto FWHOSED;
	}
Chad Barb's avatar
Chad Barb committed
939 940
    }

Chad Barb's avatar
 
Chad Barb committed
941
    #
942
    # Our next state depends on whether the experiment was active or swapped.
Chad Barb's avatar
 
Chad Barb committed
943
    #
944 945 946 947 948
    if ($estate eq EXPTSTATE_SWAPPED) {
	SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
    }
    else {
	SetExpState($pid, $eid, EXPTSTATE_MODIFY_RESWAP);
949

950 951 952 953 954
	# Set the swapper now so that nodes use the proper uid. If the
	# swapin fails, we need to reset the swapper back so that he
	# is charged appropriately.
	TBExptSetSwapUID($pid, $eid, $dbuid);

955 956 957 958 959 960 961 962 963 964 965 966
	my $optarg = "";
	#
	# For elabinelab experiments; ignore reboot/eventsys_restart,
	# and force noreconfig; none of it will work or make sense. 
	#
	if ($elabinelab) {
	    $optarg = "-noreconfig";
	}
	else {
	    $optarg  = ($reboot ? "-reboot" : "");
	    $optarg .= ($eventsys_restart ? " -eventsys_restart" : "");
	}
967

968 969 970
	print STDOUT "Running 'tbswap update $optarg $pid $eid'\n";
	if (system("$tbdir/tbswap update $optarg $pid $eid") == 0) {
	    #
971 972
	    # Success. Set the state back to active cause thats where it
	    # started.
973 974 975 976 977 978 979
	    # 
	    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
	    $estate = EXPTSTATE_ACTIVE;
	}
	else {
	    $modifyError = $errorstat = $? >> 8;
	    print STDOUT "Modify Error: tbswap update failed.\n";
Chad Barb's avatar
 
Chad Barb committed
980

981
	    #
982 983 984
	    # tbswap either restored the experiment to the way it was,
	    # or it swapped it out completely. In either case, it has
	    # also restored the virtual state. 
985
	    # 
986 987 988 989 990 991 992
	    # Icky. Magic return code that says tbswap swapped it out.
	    # We do not want tbswap to muck with states anymore, so
	    # need to know what it did. At some point we should clean
	    # up the exit reporting! Anyway, fatal() needs to know the
	    # the right state to go back to (no longer ACTIVE).
	    #
	    if ($errorstat & 0x40) {
993 994
		$estate = EXPTSTATE_SWAPPED;
		$termswapstate = EXPTSTATE_SWAPPED;
995
		$modifySwapped = 1;
996 997
                # Old accounting info.
		TBSetExpSwapTime($pid, $eid);
998 999 1000 1001
		$modifyError = "Update aborted; experiment swapped out.";
	    }
	    else {
		$modifyError = "Update aborted; old state restored.";
1002 1003 1004

		# Reset the swapper since the experiment is still running.
		TBExptSetSwapUID($pid, $eid, $last_swap_uid);
Chad Barb's avatar
 
Chad Barb committed
1005
	    }
1006
	}
Chad Barb's avatar
Chad Barb committed
1007 1008
    }

1009 1010 1011 1012 1013 1014 1015 1016
    #
    # We need to rerender only if the NS file was changed (ran tbprerun),
    # If the swapmod succeeded, then unconditionally run the renderer. If
    # swap failed, then need to run the renderer only if we stopped one in
    # progress.
    #
    if (defined($modnsfile)) {
	system("prerender -t $pid $eid")	
1017
	    if (!defined($modifyError) || $rendering);
Chad Barb's avatar
 
Chad Barb committed
1018
    }
1019