swapexp.in 26.2 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(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 STDOUT "Usage: swapexp [-b | -w] [-i | -a | -f] [-r] [-e] ".
21
	"<-s in | out | restart | modify | pause> <pid> <eid> [<nsfile>]\n";
22 23
    exit(-1);
}
24
my  $optlist = "biafres:w";
25

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
#
# 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);
}

50 51 52 53 54 55
#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
my $TBLOGS = "@TBLOGSEMAIL@";
56
my $TBINFO = "$TB/expinfo";
57
my $TBDOCBASE = "@TBDOCBASE@";
58 59 60 61 62 63 64 65 66

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

my $tbdir    = "$TB/bin/";
67
my $tbdata   = "tbdata";
68
my $batch    = 0;
69
my $idleswap = 0;
70 71
my $autoswap = 0;
my $force    = 0;
Chad Barb's avatar
Chad Barb committed
72
my $reboot   = 0;
73
my $waitmode = 0;
74
my $eventsys_restart   = 0;
75
my $errorstat= -1;
76
my $modifyHosed = 0;
Chad Barb's avatar
 
Chad Barb committed
77

78 79 80 81 82
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
83
my @allnodes;
84
my @row;
85
my $action;
86
my $nextswapstate;
87
my $termswapstate;
Chad Barb's avatar
 
Chad Barb committed
88

89 90 91
#
# Untaint the path
# 
92
$ENV{'PATH'} = "/bin:/usr/bin:$TB/libexec/vis";
93 94 95 96 97 98 99
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

100 101 102 103 104 105 106
#
# 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);

107 108 109 110 111 112 113 114
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
115 116 117
if (defined($options{"i"})) {
    $idleswap = 1;
}
118 119 120
if (defined($options{"w"})) {
    $waitmode = 1;
}
121 122 123 124 125 126
if (defined($options{"a"})) {
    $autoswap = 1;
}
if (defined($options{"f"})) {
    $force = 1;
}
127 128 129
if (defined($options{"b"})) {
    $batch = 1;
}
Chad Barb's avatar
 
Chad Barb committed
130 131 132
if (defined($options{"r"})) {
    $reboot = 1;
}
133 134 135
if (defined($options{"e"})) {
    $eventsys_restart = 1;
}
136 137 138
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
139 140 141
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
142
	$inout ne "pause"   &&
Chad Barb's avatar
 
Chad Barb committed
143
	$inout ne "modify") {
144 145 146 147 148 149 150
	usage();
    }
}
else {
    usage();
}

151 152 153 154 155
usage()
    if (($waitmode && $batch) ||
	($inout ne "modify" && @ARGV != 2) ||
	(($waitmode || $batch) && ($idleswap || $autoswap || $force)));

156 157 158 159 160
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
161 162 163
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

164 165 166
#
# Untaint the arguments.
#
167
if ($pid =~ /^([-\w\.]+)$/) {
168 169 170 171 172
    $pid = $1;
}
else {
    die("Tainted argument $pid!\n");
}
173
if ($eid =~ /^([-\w\.]+)$/) {
174 175 176 177 178
    $eid = $1;
}
else {
    die("Tainted argument $eid!\n");
}
179
my $repfile = "$eid.report";
180 181
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
182 183 184
my $tempnsfile;
my $modnsfile;

Leigh B. Stoller's avatar
Leigh B. Stoller committed
185
if ($inout eq "modify" && @ARGV > 2) {
186 187 188 189 190
    $tempnsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
191
    if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
192
	$tempnsfile = $1;
193 194
    }
    else {
195 196 197 198 199 200 201 202 203 204
	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;
205
    }
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
    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) {
227 228 229
	die("*** $0:\n".
	    "    $tempnsfile does not look like an NS file!\n");
    }
230 231
    $modnsfile = "$eid-modify.ns";
}
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249

#
# 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
250
# Verify that this person can muck with the experiment.
251 252 253 254 255
# 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
256
	"    You do not have permission to swap or modify this experiment!\n");
257 258
}

259 260 261 262
# Must do this before lock tables!
# idleswap is in minutes, threshold is in hours
$idleswap_time = 60 * TBGetSiteVar("idle/threshold");

263 264 265 266 267 268 269 270 271
#
# In wait mode, block interrupt until we spin off the background process.
#
if ($waitmode) {
    $SIG{TERM} = 'IGNORE';
    $SIG{QUIT} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

272 273 274 275 276
#
# 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
277
# the tb scripts.
278 279 280 281 282 283 284 285 286 287 288 289 290
#
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'};
291
my $batchstate      = $hashrow{'batchstate'};
292
my $expt_path       = $hashrow{'path'};
293
my $expt_locked     = $hashrow{'expt_locked'};
294
my $isbatchexpt     = $hashrow{'batchmode'};
295
my $canceled        = $hashrow{'canceled'};
296 297 298 299 300 301 302 303 304 305
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;
306

307 308
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
309
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
310 311 312
    $idleswaptime = $idleswap_time / 60.0;
}

313 314
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
315
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
316 317
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
318

319
if (! chdir($workdir)) {
320
    die("*** $0:\n".
321
	"    Could not chdir to $workdir: $!\n");
322 323
}

324
#
325 326 327
# This script is called from the batch daemon.
# 
if ($batch) {
328
    #
329 330 331
    # 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. 
332
    #
333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
    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);
357 358
    }
    else {
359 360 361 362 363 364
	die("*** $0:\n".
	    "    Improper request from batch daemon for $pid/$eid!\n");
    }
}
else {
    if ($isbatchexpt) {
365 366 367 368
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
369 370
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
371
	    if ($canceled);
372

373
	if ($inout eq "in") {
374
	    ExitWithStatus(1,
375 376 377 378
			   "Batch experiment $pid/$eid must be SWAPPED to\n".
			   "QUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_SWAPPED);
	    SetExpState($pid, $eid, EXPTSTATE_QUEUED);
379 380
	}
	elsif ($inout eq "out") {
381
	    ExitWithStatus(1,
382 383 384 385
			   "Batch experiment $pid/$eid must be ACTIVE or\n".
			   "ACTIVATING to swap out. Currently $estate.")
		if ($estate ne EXPTSTATE_ACTIVE &&
		    $estate ne EXPTSTATE_ACTIVATING);
386 387 388 389 390

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
391
	    TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
392 393
	}
	elsif ($inout eq "pause") {
394
	    ExitWithStatus(1,
395 396 397
			   "Batch experiment $pid/$eid must be QUEUED to\n".
			   "DEQUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_QUEUED);
398 399

	    #
400 401 402 403
	    # 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.
404
	    #
405
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
406
	}
407
	elsif ($inout eq "modify") {
408
	    ExitWithStatus(1,
409 410 411 412 413 414
			   "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());

415
	    #
416
	    # Otherwise, proceed with the modify. The experiment will be
417 418
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
419 420 421 422
	    # 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. 
423
	    #
424 425
	    goto doit;
	}
426 427
	else {
	    die("*** $0:\n",
428
		"    Operation $inout not allowed on a batch experiment!\n");
429
	}
430 431
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
432
      doit:
433
    }
434 435 436 437 438 439 440 441 442 443
    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);
444

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
	#
	# 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");
	  }
499 500
	}
    }
501 502
}

503 504 505 506 507 508 509
#
# 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 {
510
	$nextswapstate = EXPTSTATE_ACTIVATING();
511 512 513
	last SWITCH;
    };
    /^out$/i && do {
514
	$nextswapstate = EXPTSTATE_SWAPPING();
515 516 517
	last SWITCH;
    };
    /^restart$/i && do {
518
	$nextswapstate = EXPTSTATE_RESTARTING();
519 520 521
	last SWITCH;
    };
    /^modify$/i && do {
522 523
	$nextswapstate = (($estate eq EXPTSTATE_SWAPPED()) ?
			  EXPTSTATE_MODIFY_PARSE() : EXPTSTATE_MODIFY_REPARSE());
524 525
	last SWITCH;
    };
526
    die("*** $0:\n".
527
	"    Missing state check for action: $action\n");
528
}
529 530
 
# Update idleswap_timeout to whatever the current value is.
531
if ($inout ne "out") {
532 533 534
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
535

536 537 538 539 540 541 542
#
# 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.
TBLockExp($pid, $eid, $nextswapstate);
543 544 545 546
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
547 548 549 550
# 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). 
551 552
#

553 554 555 556 557 558 559 560 561
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
562 563 564
if ($inout eq "modify") {
    $action = "modified";
}
565

566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
#
# 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;
}

581 582 583
#
# Before going to background, we have to copy out the NS file!
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
584
if ($inout eq "modify" && defined($modnsfile)) {
585 586 587 588 589 590 591 592
    unlink($modnsfile);
    if (system("/bin/cp", "$tempnsfile", "$modnsfile")) {
	die("*** $0:\n".
	    "    Could not copy $tempnsfile to $modnsfile");
    }
    chmod(0664, "$modnsfile");
}

593 594 595 596
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
597
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
598
    TBExptSetLogFile($pid, $eid, $logname);
599
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
600

601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624
    if (my $childpid = TBBackGround($logname)) {
	#
	# Parent exits normally, except if in waitmode. 
	#
	if (!$waitmode) {
	    print "Experiment $pid/$eid is now being $action.\n".
		"You will be notified via email when the this is done.\n";
	    exit(0);
	}
	print("Waiting for experiment $eid to finish its swap${action}\n");
	print("You may type ^C at anytime; you will be notified via email;\n".
	      "later; you will not actually interrupt the experiment itself.\n");
	
	# 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';

625
	#
626
	# Wait until child exits or until user gets bored and types ^C.
627
	#
628 629 630 631
	waitpid($childpid, 0);
	
	print("Done. Exited with status: $?\n");
	exit($? >> 8);
632 633 634
    }
}

635 636 637 638 639 640 641 642
#
# 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();
}

643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
#
# 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);
}

659 660 661
#
# Remove old report file since its contents are going to be invalid.
#
662
if ($inout ne "restart" && -e $repfile) {
663 664 665
    unlink("$repfile");
}

666 667 668 669
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
670 671 672 673
    my $optarg = (($force || $idleswap) ? "-force" : "");
    
    print STDOUT "Running 'tbswap out $optarg $pid $eid'\n";
    if (system("$tbdir/tbswap out $optarg $pid $eid") != 0) {
674
	$errorstat = $? >> 8;
675
	fatal("tbswap out failed!");
676
    }
677
    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
678
}
679
elsif ($inout eq "in") {
680
    print STDOUT "Running 'tbswap in $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
681
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
682
	$errorstat = $? >> 8;
683
	fatal("tbswap in failed!");
684
    }
685
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
686

687
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
688
}
Chad Barb's avatar
 
Chad Barb committed
689
elsif ($inout eq "modify") {
Chad Barb's avatar
Chad Barb committed
690
    my $modifyError = "";
691
    my $oldstate    = $estate;
Chad Barb's avatar
Chad Barb committed
692

693 694 695
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

Chad Barb's avatar
Chad Barb committed
696
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
697
    if (TBExptBackupVirtualState($pid, $eid)) {
698
	fatal("Could not backup experiment state; cannot safely continue!");
Chad Barb's avatar
Chad Barb committed
699 700 701
    }

    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
702 703
    # Rerun tbprerun if modifying, but only if new NS file provided.
    # Yep, we allow reswap without changing the NS file. For Shashi and SIM. 
Chad Barb's avatar
Chad Barb committed
704
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
705 706 707 708 709
    if (defined($modnsfile)) {
	print STDOUT "Running 'tbprerun $pid $eid $modnsfile'\n";
	if (system("$tbdir/tbprerun $pid $eid $modnsfile") != 0) {
	    $modifyError = "tbprerun failed!";
	}
Chad Barb's avatar
Chad Barb committed
710 711
    }

Chad Barb's avatar
 
Chad Barb committed
712
    #
713
    # Our next state depends on whether the experiment was active or swapped.
Chad Barb's avatar
 
Chad Barb committed
714
    #
715 716 717
    if (! $modifyError) {
	if ($estate eq EXPTSTATE_SWAPPED) {
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
Chad Barb's avatar
 
Chad Barb committed
718
	}
719 720 721 722
	else {
	    SetExpState($pid, $eid, EXPTSTATE_MODIFY_RESWAP);
	    
	    my $optarg = ($reboot ? "-reboot" : "");
723
	    $optarg .= ($eventsys_restart ? " -eventsys_restart" : "");
724 725 726 727 728 729

	    print STDOUT "Running 'tbswap update $optarg $pid $eid'\n";
	    if (system("$tbdir/tbswap update $optarg $pid $eid") != 0) {
		$errorstat = $? >> 8;
		$modifyError = "tbswap update failed!";
	    }
Chad Barb's avatar
 
Chad Barb committed
730

731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750
	    #
	    # See what tbswap did. It might have swapped it out if there
	    # was an error. 
	    # 
	    if (! $modifyError) {
		SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
		$estate = EXPTSTATE_ACTIVE;
	    }
	    elsif ($errorstat & 0x40) {
		#
		# 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).
		#
		$estate = EXPTSTATE_SWAPPED;
		$termswapstate = EXPTSTATE_SWAPPED;
                # Old accounting info.
		TBSetExpSwapTime($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
751
	    }
752
	}
Chad Barb's avatar
Chad Barb committed
753 754 755
    }

    if ($modifyError) {
756
	print STDOUT "Modify Error: $modifyError\n";
Chad Barb's avatar
Chad Barb committed
757
	print STDOUT "Recovering experiment state...\n";
758
	
759 760
	# Must deal with the prerender explicitly since it runs background.
	system("prerender -r $pid $eid");
761
	TBExptRemoveVirtualState($pid, $eid);
762
	
763
	if (TBExptRestoreVirtualState($pid, $eid) == 0) {
764 765
	    # Must deal with the prerender explicitly since it runs background.
	    system("prerender -t $pid $eid");
766 767 768 769 770
	    fatal("Update aborted; old state restored.");
	}
	else {
	    $modifyHosed = 1;
	    fatal("Experiment state could not be restored!");
Chad Barb's avatar
Chad Barb committed
771
	}
Chad Barb's avatar
 
Chad Barb committed
772
    }
773
    
774
    TBExptClearBackupState($pid, $eid);
775
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
776
}
Chad Barb's avatar
 
Chad Barb committed
777
else { # $inout eq "restart" assumed.
778
    print STDOUT "Running 'tbrestart $pid $eid'\n";
779
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
780
	fatal("tbrestart failed!");
781
    }
782
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
783
}
784

785 786 787 788 789 790 791 792 793 794
#
# 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.
#
795
system("cp -Rfp $workdir/ $userdir/tbdata/");
796

797 798 799 800
#
# Gather stats. 
#
if ($inout eq "in") {
801
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
802 803
}
elsif ($inout eq "out") {
804
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
805
		    ($idleswap ? TBDB_STATS_FLAGS_IDLESWAP() : 0));
806 807
}
elsif ($inout eq "modify") {
808
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
809
}
810 811
# Old accounting info.
TBSetExpSwapTime($pid, $eid);
812

813 814 815 816 817 818
#
# Set the swapper uid on success only, and *after* gathering swap stats!
#
TBExptSetSwapUID($pid, $eid, $dbuid);

#
819
# In batch mode, just exit without sending email or unlocking. The
820
# batch daemon will take care of that.
821 822 823 824 825
#
if ($batch) {
    exit(0);
}

826 827 828 829 830 831 832
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

833 834 835
#
# Must unlock before exit.
#
836
TBUnLockExp($pid, $eid);
837 838 839 840 841

#
# Since the swap completed, clear the cancel flag. This must be done
# after we change the experiment state (above). 
#
842
TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
843 844 845

print "Swap Success!\n";

846 847 848 849
#
# Send email notification to user.
#
my $message =
850 851
    "Experiment $eid in project $pid has been ";

852
if ($inout eq "out" && ($idleswap || $autoswap || $force) ) {
853
    $message .= "forcibly swapped out by\nEmulab";
854 855 856 857 858
    if ($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) {
859 860
	$message .= " because it exceeded its Maximum Duration.\n".
	  "(See also the Max. Duration info in \n".
861 862 863 864 865
	  "$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";
    }
866 867 868 869 870
}
else {
    $message .= "$action.\n";
}

871 872 873 874 875
if ($inout eq "in") {
    # Add the swap settings...
    $message .="\nCurrent swap settings:\n$swapsettings";
}

876 877
$message .=
    "\n".
878 879
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
880 881

SENDMAIL("$user_name <$user_email>",
882
	 "Experiment $pid/$eid \u$action",
883
	 $message,
884
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
885 886
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
887 888
	 (($inout eq "restart") ? ($logname) :
	  (($repfile, $logname), (defined($modnsfile) ? ($modnsfile) : ()))));
889 890 891 892 893 894

exit 0;

sub fatal($)
{
    my($mesg) = $_[0];
Chad Barb's avatar
Chad Barb committed
895

896 897
    print STDOUT "*** $0:\n".
	         "    $mesg\n";
898

899 900 901 902 903 904 905 906 907 908 909 910 911
    #
    # 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") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, $errorstat);
    }

912 913 914 915 916 917 918
    #
    # Clear backup state since not needed anymore; experiment is toast. 
    # 
    if ($inout eq "modify") {
	TBExptClearBackupState($pid, $eid);
    }

Chad Barb's avatar
 
Chad Barb committed
919
    #
920
    # If hosed, we entirely terminate the experiment.
Chad Barb's avatar
 
Chad Barb committed
921
    #
922
    if ($modifyHosed) {
Chad Barb's avatar
 
Chad Barb committed
923
	#
924
	# Note: $estate is indeed still set appropriately!
Chad Barb's avatar
 
Chad Barb committed
925 926
	#
	if ($estate eq EXPTSTATE_ACTIVE) {
927
	    print "Running 'tbswap out -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
928 929 930 931
	    if (system("$tbdir/tbswap out -force $pid $eid") != 0) {
		print "tbswap out failed!\n";
	    }
	}
Chad Barb's avatar
Chad Barb committed
932

933
	print "Running 'tbend -force $pid $eid'\n";
Chad Barb's avatar
 
Chad Barb committed
934 935 936
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
	}
937
	# Must override since we are so badly hosed. 
938
	$termswapstate = EXPTSTATE_TERMINATED;
Chad Barb's avatar
 
Chad Barb committed
939 940
    }

941 942 943
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

944 945 946
    # Set proper state, which is typically the way we came in.
    SetExpState($pid, $eid, $termswapstate);

947
    #
948
    # In batch mode, exit without sending the email or unlocking. The
949
    # batch daemon will take care of that.
950 951
    #
    if ($batch) {
952
	exit($errorstat);
953 954
    }

955
    #
Chad Barb's avatar
Chad Barb committed
956
    # Clear the log file so the web page stops spewing.
957 958 959 960 961
    #
    if (defined($logname)) {
	TBExptCloseLogFile($pid, $eid);
    }

962 963
    # Unlock and reset state to its terminal value.
    TBUnLockExp($pid, $eid);
964 965 966 967 968

    #
    # Clear the cancel flag now that the operation is complete. Must be done
    # after we change the experiment state (above).
    #
969
    TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
970

971 972 973 974
    #
    # Send a message to the testbed list. Append the logfile.
    #
    SENDMAIL("$user_name <$user_email>",
975
	     "Swap ${inout} Failure: $pid/$eid",
976
	     $mesg,
977
	     ($idleswap ? $TBOPS : "$user_name <$user_email>"),
978
	     "Cc:  $expt_head_name <$expt_head_email>\n".
Leigh B. Stoller's avatar
Leigh B. Stoller committed
979
	     "Cc:  $TBOPS",
980
	     (($logname), (defined($modnsfile) ? ($modnsfile) : ())));
981

Leigh B. Stoller's avatar
Leigh B. Stoller committed
982
    if ($modifyHosed) {
Chad Barb's avatar
 
Chad Barb committed
983 984 985 986 987 988 989 990 991
	#
	# Copy off the workdir to the user directory, Then back up both of
	# them for post-mortem debugging.
	#
	system("/bin/cp -Rfp $workdir/ $userdir/tbdata");
	system("/bin/rm -rf  ${workdir}-failed");
	system("/bin/mv -f   $workdir ${workdir}-failed");
	system("/bin/rm -rf  ${userdir}-failed");
	system("/bin/mv -f   $userdir ${userdir}-failed");
Chad Barb's avatar
Chad Barb committed
992
	TBExptDestroy($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
993 994
    }

995
    exit($errorstat);
996
}