swapexp.in 31.8 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) 2000-2004 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
my  $optlist = "biafres:wq";
36

37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
#
# 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) {
	die("*** $0:\n".
	    "    $message\n");
    }
    else {
	print STDERR "$message\n";
    }
    exit($status);
}

61 62 63 64 65 66
#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
my $TBLOGS = "@TBLOGSEMAIL@";
67
my $TBINFO = "$TB/expinfo";
68
my $TBDOCBASE = "@TBDOCBASE@";
69
my $TBBASE = "@TBBASE@";
70 71 72 73 74 75 76 77

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

78 79 80 81 82 83 84 85 86
# 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";
87
my $tbdata   = "tbdata";
88
my $batch    = 0;
89
my $idleswap = 0;
90 91
my $autoswap = 0;
my $force    = 0;
Chad Barb's avatar
Chad Barb committed
92
my $reboot   = 0;
93
my $waitmode = 0;
94
my $quiet    = 0;
95
my $eventsys_restart   = 0;
96
my $errorstat= -1;
97 98
my $modifyHosed   = 0;
my $modifySwapped = 0;
Chad Barb's avatar
 
Chad Barb committed
99

100 101 102 103 104
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
105
my @allnodes;
106
my @row;
107
my $action;
108
my $nextswapstate;
109
my $termswapstate;
Chad Barb's avatar
 
Chad Barb committed
110

111 112 113
#
# Untaint the path
# 
114
$ENV{'PATH'} = "/bin:/usr/bin:$TB/libexec/vis";
115 116 117 118 119 120 121
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

122 123 124 125 126 127 128
#
# 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);

129 130 131 132 133 134 135 136
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
137 138 139
if (defined($options{"i"})) {
    $idleswap = 1;
}
140 141 142
if (defined($options{"w"})) {
    $waitmode = 1;
}
143 144 145 146 147 148
if (defined($options{"a"})) {
    $autoswap = 1;
}
if (defined($options{"f"})) {
    $force = 1;
}
149 150 151
if (defined($options{"b"})) {
    $batch = 1;
}
Chad Barb's avatar
 
Chad Barb committed
152 153 154
if (defined($options{"r"})) {
    $reboot = 1;
}
155 156 157
if (defined($options{"e"})) {
    $eventsys_restart = 1;
}
158 159 160
if (defined($options{"q"})) {
    $quiet = 1;
}
161 162 163
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
164 165 166
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
167
	$inout ne "pause"   &&
Chad Barb's avatar
 
Chad Barb committed
168
	$inout ne "modify") {
169 170 171 172 173 174 175
	usage();
    }
}
else {
    usage();
}

176 177 178 179 180
usage()
    if (($waitmode && $batch) ||
	($inout ne "modify" && @ARGV != 2) ||
	(($waitmode || $batch) && ($idleswap || $autoswap || $force)));

181 182 183 184 185
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
186 187 188
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

189 190 191
#
# Untaint the arguments.
#
192
if ($pid =~ /^([-\w\.]+)$/) {
193 194 195 196 197
    $pid = $1;
}
else {
    die("Tainted argument $pid!\n");
}
198
if ($eid =~ /^([-\w\.]+)$/) {
199 200 201 202 203
    $eid = $1;
}
else {
    die("Tainted argument $eid!\n");
}
204
my $repfile = "$eid.report";
205 206
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
207 208 209
my $tempnsfile;
my $modnsfile;

Leigh B. Stoller's avatar
Leigh B. Stoller committed
210
if ($inout eq "modify" && @ARGV > 2) {
211 212 213 214 215
    $tempnsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
216
    if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
217
	$tempnsfile = $1;
218 219
    }
    else {
220 221 222 223 224 225 226 227 228 229
	die("Tainted nsfile name: $tempnsfile\n");
    }
    #
    # 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;
230
    }
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
    else {
	die("Tainted nsfile returned by realpath: $translated\n");
    }

    #
    # 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/) &&
	! ($tempnsfile =~ /^\/users/)) {
	die("$tempnsfile does not resolve to an appropriate directory!\n");
    }

    if (! -f $tempnsfile || -z $tempnsfile || ! -r $tempnsfile) {
252 253 254
	die("*** $0:\n".
	    "    $tempnsfile does not look like an NS file!\n");
    }
255 256
    $modnsfile = "$eid-modify.ns";
}
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274

#
# 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");
}

#
Chad Barb's avatar
 
Chad Barb committed
275
# Verify that this person can muck with the experiment.
276 277 278 279 280
# Note that any script down the line has to do an admin check also. 
#
if ($UID && !TBAdmin($UID) &&
    !TBExptAccessCheck($dbuid, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
Chad Barb's avatar
 
Chad Barb committed
281
	"    You do not have permission to swap or modify this experiment!\n");
282 283
}

284 285 286 287
# Must do this before lock tables!
# idleswap is in minutes, threshold is in hours
$idleswap_time = 60 * TBGetSiteVar("idle/threshold");

288 289 290 291 292 293 294 295 296
#
# In wait mode, block interrupt until we spin off the background process.
#
if ($waitmode) {
    $SIG{TERM} = 'IGNORE';
    $SIG{QUIT} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

297 298 299 300 301
#
# 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
302
# the tb scripts.
303 304 305 306 307 308 309 310 311 312 313 314 315
#
DBQueryFatal("lock tables experiments write");

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

if (! $query_result->numrows) {
    die("*** $0:\n".
	"    No such experiment $pid/$eid exists!\n");
}
my %hashrow = $query_result->fetchhash();
my $expt_head_login = $hashrow{'expt_head_uid'};
my $estate          = $hashrow{'state'};
316
my $batchstate      = $hashrow{'batchstate'};
317
my $expt_path       = $hashrow{'path'};
318
my $expt_locked     = $hashrow{'expt_locked'};
319
my $isbatchexpt     = $hashrow{'batchmode'};
320
my $canceled        = $hashrow{'canceled'};
321
my $linktest_level  = $hashrow{'linktest_level'};
322 323 324 325 326 327 328 329 330 331
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;
332
my $rendering   = $hashrow{'prerender_pid'};
333

334 335
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
336
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
337 338 339
    $idleswaptime = $idleswap_time / 60.0;
}

340 341
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
342
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
343 344
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
345

346
if (! chdir($workdir)) {
347
    die("*** $0:\n".
348
	"    Could not chdir to $workdir: $!\n");
349 350
}

351
#
352 353 354
# This script is called from the batch daemon.
# 
if ($batch) {
355
    #
356 357 358
    # 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. 
359
    #
360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383
    die("*** $0:\n".
	"    Experiment $pid/$eid is supposed to be a batch experiment!\n")
	if (!$isbatchexpt);
    
    die("*** $0:\n".
	"    Batch experiment $pid/$eid should be locked!\n")
	if (!defined($expt_locked) ||
	    $batchstate ne BATCHSTATE_LOCKED());

    if ($inout eq "in") {
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid is not in the proper state!\n".
	    "    Currently $estate, but should be QUEUED.\n")
	    if ($estate ne EXPTSTATE_QUEUED);
	
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid has been canceled! Aborting.\n")
	    if ($canceled);
    }
    elsif ($inout eq "out") {
	die("*** $0:\n".
	    "    Batch experiment $pid/$eid is not in the proper state!\n".
	    "    Currently $estate, but should be ACTIVE.\n")
	    if ($estate ne EXPTSTATE_ACTIVE);
384 385
    }
    else {
386 387 388 389 390 391
	die("*** $0:\n".
	    "    Improper request from batch daemon for $pid/$eid!\n");
    }
}
else {
    if ($isbatchexpt) {
392 393 394 395
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
396 397
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
398
	    if ($canceled);
399

400
	if ($inout eq "in") {
401
	    ExitWithStatus(1,
402 403 404 405
			   "Batch experiment $pid/$eid must be SWAPPED to\n".
			   "QUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_SWAPPED);
	    SetExpState($pid, $eid, EXPTSTATE_QUEUED);
406 407
	}
	elsif ($inout eq "out") {
408
	    ExitWithStatus(1,
409 410 411 412
			   "Batch experiment $pid/$eid must be ACTIVE or\n".
			   "ACTIVATING to swap out. Currently $estate.")
		if ($estate ne EXPTSTATE_ACTIVE &&
		    $estate ne EXPTSTATE_ACTIVATING);
413 414 415 416 417

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
418
	    TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
419 420
	}
	elsif ($inout eq "pause") {
421
	    ExitWithStatus(1,
422 423 424
			   "Batch experiment $pid/$eid must be QUEUED to\n".
			   "DEQUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_QUEUED);
425 426

	    #
427 428 429 430
	    # 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.
431
	    #
432
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
433
	}
434
	elsif ($inout eq "modify") {
435
	    ExitWithStatus(1,
436 437 438 439 440 441
			   "Batch experiment $pid/$eid must be SWAPPED or\n".
			   "ACTIVE to modify. Currently $estate.")
		if (($estate ne EXPTSTATE_SWAPPED &&
		     $estate ne EXPTSTATE_ACTIVATING) ||
		    $batchstate != BATCHSTATE_UNLOCKED());

442
	    #
443
	    # Otherwise, proceed with the modify. The experiment will be
444 445
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
446 447 448 449
	    # 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. 
450
	    #
451 452
	    goto doit;
	}
453 454
	else {
	    die("*** $0:\n",
455
		"    Operation $inout not allowed on a batch experiment!\n");
456
	}
457 458
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
459
      doit:
460
    }
461 462 463 464 465 466 467 468 469 470
    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);
471

472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525
	#
	# 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!");
		  }
		  last SWITCH;
	      };
	      /^out$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
		      $estate ne EXPTSTATE_ACTIVATING()) {
		      ExitWithStatus(1,
				     "Experiment $pid/$eid is not swapped in ".
				     "or activating!\n");
		  }
		  
		  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");
		  }
		  last SWITCH;
	      };
	      die("*** $0:\n".
		  "    Missing state check for action: $action\n");
	  }
526 527
	}
    }
528 529
}

530 531 532 533 534 535 536
#
# 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 {
537
	$nextswapstate = EXPTSTATE_ACTIVATING();
538 539 540
	last SWITCH;
    };
    /^out$/i && do {
541
	$nextswapstate = EXPTSTATE_SWAPPING();
542 543 544
	last SWITCH;
    };
    /^restart$/i && do {
545
	$nextswapstate = EXPTSTATE_RESTARTING();
546 547 548
	last SWITCH;
    };
    /^modify$/i && do {
549 550
	$nextswapstate = (($estate eq EXPTSTATE_SWAPPED()) ?
			  EXPTSTATE_MODIFY_PARSE() : EXPTSTATE_MODIFY_REPARSE());
551 552
	last SWITCH;
    };
553
    die("*** $0:\n".
554
	"    Missing state check for action: $action\n");
555
}
556 557
 
# Update idleswap_timeout to whatever the current value is.
558
if ($inout ne "out") {
559 560 561
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
562

563 564 565 566 567 568
#
# 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.
569 570 571 572 573 574 575 576 577 578
TBLockExp($pid, $eid, $nextswapstate)
    or die("*** $0:\n".
	   "Failed to set experiment state to $nextswapstate\n");

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

579 580 581 582
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
583 584 585 586
# 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). 
587 588
#

589 590 591 592 593 594 595 596 597
if ($inout eq "in") {
    $action = "swapped in";
}
if ($inout eq "out") {
    $action = "swapped out";
}
if ($inout eq "restart") {
    $action = "restarted";
}
Chad Barb's avatar
 
Chad Barb committed
598 599 600
if ($inout eq "modify") {
    $action = "modified";
}
601

602 603 604 605 606 607 608 609 610 611 612 613 614 615 616
#
# 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)) {
    print STDERR "*** WARNING: ".
	         "Could not determine name/email for $expt_head_login.\n";
    $expt_head_name  = "TBOPS";
    $expt_head_email = $TBOPS;
}

617 618 619
#
# Before going to background, we have to copy out the NS file!
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
620
if ($inout eq "modify" && defined($modnsfile)) {
621 622
    unlink($modnsfile);
    if (system("/bin/cp", "$tempnsfile", "$modnsfile")) {
623
	fatal("Could not copy $tempnsfile to $modnsfile");
624 625 626 627
    }
    chmod(0664, "$modnsfile");
}

628 629 630 631
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
632
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
633
    TBExptSetLogFile($pid, $eid, $logname);
634
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
635

636 637
    if (my $childpid = TBBackGround($logname)) {
	#
638 639
	# Parent exits normally, unless in waitmode. We have to set
	# justexit to make sure the END block below does not run.
640
	#
641 642
	$justexit = 1;

643
	if (!$waitmode) {
644 645 646
	    print("Experiment $pid/$eid is now being $action.\n".
		  "You will be notified via email when the this is done.\n")
		if (! $quiet);
647 648
	    exit(0);
	}
649 650 651 652 653 654 655 656
	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");
	}
657 658 659 660 661 662 663 664 665 666 667 668
	
	# 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';

669
	#
670
	# Wait until child exits or until user gets bored and types ^C.
671
	#
672 673
	waitpid($childpid, 0);
	
674 675
	print("Done. Exited with status: $?\n")
	    if (! $quiet);
676
	exit($? >> 8);
677
    }
678
    TBdbfork();
679 680
}

681 682 683 684 685 686 687 688
#
# 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();
}

689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708
#
# 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';

709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724
#
# 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);
}

725 726 727
#
# Remove old report file since its contents are going to be invalid.
#
728
if ($inout ne "restart" && -e $repfile) {
729 730 731
    unlink("$repfile");
}

732 733 734 735
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
736 737 738 739
    my $optarg = (($force || $idleswap) ? "-force" : "");
    
    print STDOUT "Running 'tbswap out $optarg $pid $eid'\n";
    if (system("$tbdir/tbswap out $optarg $pid $eid") != 0) {
740
	$errorstat = $? >> 8;
741
	fatal("tbswap out failed!");
742
    }
743 744
    SetExpState($pid, $eid, EXPTSTATE_SWAPPED)
	or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
745
}
746
elsif ($inout eq "in") {
747
    print STDOUT "Running 'tbswap in $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
748
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
749
	$errorstat = $? >> 8;
750
	fatal("tbswap in failed!");
751
    }
752 753
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE)
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
754

755
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
756
}
Chad Barb's avatar
 
Chad Barb committed
757
elsif ($inout eq "modify") {
758
    my $modifyError;
Chad Barb's avatar
Chad Barb committed
759

760 761 762
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

Chad Barb's avatar
Chad Barb committed
763
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
764
    if (TBExptBackupVirtualState($pid, $eid)) {
765
	fatal("Could not backup experiment state; cannot safely continue!");
Chad Barb's avatar
Chad Barb committed
766 767 768
    }

    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
769
    # Rerun tbprerun if modifying, but only if new NS file provided.
770 771
    # 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
772
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
773 774 775
    if (defined($modnsfile)) {
	print STDOUT "Running 'tbprerun $pid $eid $modnsfile'\n";
	if (system("$tbdir/tbprerun $pid $eid $modnsfile") != 0) {
776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794
	    print STDOUT "Modify Error: tbprerun failed.\n";
	    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
795
	}
Chad Barb's avatar
Chad Barb committed
796 797
    }

Chad Barb's avatar
 
Chad Barb committed
798
    #
799
    # Our next state depends on whether the experiment was active or swapped.
Chad Barb's avatar
 
Chad Barb committed
800
    #
801 802 803 804 805
    if ($estate eq EXPTSTATE_SWAPPED) {
	SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
    }
    else {
	SetExpState($pid, $eid, EXPTSTATE_MODIFY_RESWAP);
806
	    
807 808
	my $optarg = ($reboot ? "-reboot" : "");
	$optarg .= ($eventsys_restart ? " -eventsys_restart" : "");
809

810 811 812 813 814 815 816 817 818 819 820
	print STDOUT "Running 'tbswap update $optarg $pid $eid'\n";
	if (system("$tbdir/tbswap update $optarg $pid $eid") == 0) {
	    #
	    # Success. Set the state back to active cause thats where it started.
	    # 
	    SetExpState($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
821

822
	    #
823 824 825
	    # 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. 
826
	    # 
827 828 829 830 831 832 833
	    # 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) {
834 835
		$estate = EXPTSTATE_SWAPPED;
		$termswapstate = EXPTSTATE_SWAPPED;
836
		$modifySwapped = 1;
837 838
                # Old accounting info.
		TBSetExpSwapTime($pid, $eid);
839 840 841 842
		$modifyError = "Update aborted; experiment swapped out.";
	    }
	    else {
		$modifyError = "Update aborted; old state restored.";
Chad Barb's avatar
 
Chad Barb committed
843
	    }
844
	}
Chad Barb's avatar
Chad Barb committed
845 846
    }

847 848 849 850 851 852 853 854 855
    #
    # 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 $eid")	
	    if (!defined($modifyError) || $rendering)
Chad Barb's avatar
 
Chad Barb committed
856
    }
857 858 859 860 861 862 863

    #
    # Swapmod failed ...
    #
    fatal($modifyError)
	if (defined($modifyError));

864
    TBExptClearBackupState($pid, $eid);
865
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
866
}
Chad Barb's avatar
 
Chad Barb committed
867
else { # $inout eq "restart" assumed.
868
    print STDOUT "Running 'tbrestart $pid $eid'\n";
869
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
870
	fatal("tbrestart failed!");
871
    }
872
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
873
}
874

875 876 877 878 879 880 881 882 883 884
#
# Try to copy off the files for testbed information gathering.
#
TBSaveExpLogFiles($pid, $eid);

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files. This overwrites existing files of course,
# but thats okay.
#
885
system("cp -Rfp $workdir/ $userdir/tbdata/");
886

887 888 889 890 891 892 893 894 895 896 897
#
# Deal with linktest. If requested, swapping in or modifying, and experiment
# is indeed active, then run it!
#
# XXX - linktest uses files from $userdir/tbdata/, so the above cp must
#       happen first!
#
if ($linktest_level && ExpState($pid, $eid) eq EXPTSTATE_ACTIVE) {
    #
    # Run it. No worries about failures.
    #
898
    my $output = "$userdir/logs/linktest.log";
899 900
    my $optarg = "-l $linktest_level -o $output";
    
901 902
    print STDOUT "Running 'linktest_control $optarg $pid $eid'\n";
    if (system("$TB/sbin/linktest_control $optarg $pid $eid") != 0) {
903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
	system("cat $output")
	    if (-r $output);
	
	print STDERR "*** WARNING: ".
	             "Linktest run returned non-zero status!\n";
	
	SENDMAIL("$user_name <$user_email>",
		 "Linktest Failure: $pid/$eid",
		 "Failure in linktest (level $linktest_level); ".
		 "returned non-zero status",
		 "$user_name <$user_email>",
		 "Cc: $expt_head_name <$expt_head_email>\n".
		 "Cc: $TBOPS",
		 ($output));
    }
}

920 921 922 923
#
# Gather stats. 
#
if ($inout eq "in") {
924
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
925 926
}
elsif ($inout eq "out") {
927
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
928
		    ($idleswap ? TBDB_STATS_FLAGS_IDLESWAP() : 0));
929 930
}
elsif ($inout eq "modify") {
931
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
932
}
933 934
# Old accounting info.
TBSetExpSwapTime($pid, $eid);
935

936 937 938 939 940 941
#
# Set the swapper uid on success only, and *after* gathering swap stats!
#
TBExptSetSwapUID($pid, $eid, $dbuid);

#
942
# In batch mode, just exit without sending email or unlocking. The
943
# batch daemon will take care of that.
944 945 946 947 948
#
if ($batch) {
    exit(0);
}

949 950 951 952 953 954 955
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

956 957 958
#
# Must unlock before exit.
#
959
TBUnLockExp($pid, $eid);
960 961 962 963 964

#
# Since the swap completed, clear the cancel flag. This must be done
# after we change the experiment state (above). 
#
965
TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
966 967 968

print "Swap Success!\n";

969 970 971 972
#
# Send email notification to user.
#
my $message =
973 974
    "Experiment $eid in project $pid has been ";

975
if ($inout eq "out" && ($idleswap || $autoswap || $force) ) {
976
    $message .= "forcibly swapped out by\nEmulab";
977 978 979 980 981
    if ($idleswap) {
	$message .= " because it was idle for too long (Idle-Swap).\n".
	  "(See also the Idle-Swap info in \n".
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    } elsif ($autoswap) {
982 983
	$message .= " because it exceeded its Maximum Duration.\n".
	  "(See also the Max. Duration info in \n".
984 985 986 987 988
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    } elsif ($force) {
	$message .= ". (See also our Node Usage Policies in \n".
	  "$TBDOCBASE/docwrapper.php3?docname=swapping.html )\n";
    }
989 990 991 992 993
}
else {
    $message .= "$action.\n";
}

994 995 996 997 998
if ($inout eq "in") {
    # Add the swap settings...
    $message .="\nCurrent swap settings:\n$swapsettings";
}

999 1000
$message .=
    "\n".
1001 1002
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
1003 1004

SENDMAIL("$user_name <$user_email>",
1005
	 "Experiment $pid/$eid \u$action",
1006
	 $message,
1007
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
1008 1009
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
1010 1011
	 (($inout eq "restart") ? ($logname) :
	  (($repfile, $logname), (defined($modnsfile) ? ($modnsfile) : ()))));
1012

1013
exit(0);
1014

1015
sub cleanup()
1016
{
1017 1018 1019 1020 1021 1022 1023 1024 1025 1026
    #
    # Gather stats. 
    #
    if ($inout eq "in") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, $errorstat);
    }
    elsif ($inout eq "out") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, $errorstat);
    }
    elsif ($inout eq "modify") {
1027 1028 1029 1030 1031 1032
	#
	# If the modify fails, and the experiment is swapped out, then
	# insert a record for that since that is different then a modify
	# that fails, but results in the experiment being restored the
	# way it was. 
	#
1033
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, $errorstat);
1034 1035 1036 1037
	
	if ($modifySwapped) {
	    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0);
	}
1038 1039
    }

1040 1041 1042 1043 1044 1045 1046
    #
    # Clear backup state since not needed anymore; experiment is toast. 
    # 
    if ($inout eq "modify") {
	TBExptClearBackupState($pid, $eid);
    }

Chad Barb's avatar
 
Chad Barb committed
1047
    #
1048
    # If hosed, we entirely terminate the experiment.
Chad Barb's avatar
 
Chad Barb committed
1049
    #
1050
    if ($modifyHosed) {
1051 1052
	my $stat = 0;
	    
Chad Barb's avatar
 
Chad Barb committed
1053
	#
1054
	# Note: $estate is indeed still set appropriately!
Chad Barb's avatar
 
Chad Barb committed
1055 1056
	#
	if ($estate eq EXPTSTATE_ACTIVE) {
1057
	    print "Running 'tbswap out -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
1058 1059
	    if (system("$tbdir/tbswap out -force $pid $eid") != 0) {
		print "tbswap out failed!\n";
1060
		$stat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
1061
	    }
1062
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
1063
	    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, $stat);
Chad Barb's avatar
 
Chad Barb committed
1064
	}
Chad Barb's avatar
Chad Barb committed
1065

1066
	$stat = 0;
1067
	print "Running 'tbend -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
1068 1069
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
1070
	    $stat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
1071
	}
1072 1073 1074 1075 1076 1077
	
	#
	# Okay, we are going to destroy the experiment below.
	# 
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_TERMINATE, 0);
	
1078
	# Must override since we are so badly hosed. 
1079
	$termswapstate = EXPTSTATE_TERMINATED;
Chad Barb's avatar
 
Chad Barb committed
1080
    }
1081 1082
    # Set proper state, which is typically the way we came in.
    SetExpState($pid, $eid, $termswapstate);
Chad Barb's avatar
 
Chad Barb committed
1083

1084 1085 1086
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

1087
    #
Leigh B. Stoller's avatar