swapexp.in 38 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) 2000-2005 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
55
56
57
58
59
60
61
62
63
#
# 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);
}

64
65
66
67
68
69
#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
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
81

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

82
83
84
85
86
87
88
89
90
# 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";
91
my $tbdata   = "tbdata";
92
my $checkquota = "$TB/sbin/checkquota";
93
my $batch    = 0;
94
my $idleswap = 0;
95
96
my $autoswap = 0;
my $force    = 0;
Chad Barb's avatar
Chad Barb committed
97
my $reboot   = 0;
98
my $waitmode = 0;
99
my $quiet    = 0;
100
my $eventsys_restart   = 0;
101
my $errorstat= -1;
102
103
my $modifyHosed   = 0;
my $modifySwapped = 0;
Chad Barb's avatar
   
Chad Barb committed
104

105
106
107
108
109
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
110
my @allnodes;
111
my @row;
112
my $action;
113
my $nextswapstate;
114
my $termswapstate;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
115
my $isadmin  = 0;
Chad Barb's avatar
   
Chad Barb committed
116

117
118
119
#
# Untaint the path
# 
120
$ENV{'PATH'} = "/bin:/usr/bin:$TB/libexec/vis";
121
122
123
124
125
126
127
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

128
129
130
131
132
133
134
#
# 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);

135
136
137
138
139
140
141
142
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
143
144
145
if (defined($options{"i"})) {
    $idleswap = 1;
}
146
147
148
if (defined($options{"w"})) {
    $waitmode = 1;
}
149
150
151
152
153
154
if (defined($options{"a"})) {
    $autoswap = 1;
}
if (defined($options{"f"})) {
    $force = 1;
}
155
156
157
if (defined($options{"b"})) {
    $batch = 1;
}
Chad Barb's avatar
   
Chad Barb committed
158
159
160
if (defined($options{"r"})) {
    $reboot = 1;
}
161
162
163
if (defined($options{"e"})) {
    $eventsys_restart = 1;
}
164
165
166
if (defined($options{"q"})) {
    $quiet = 1;
}
167
168
169
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
170
171
172
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
173
	$inout ne "pause"   &&
Chad Barb's avatar
   
Chad Barb committed
174
	$inout ne "modify") {
175
176
177
178
179
180
181
	usage();
    }
}
else {
    usage();
}

182
183
184
185
186
usage()
    if (($waitmode && $batch) ||
	($inout ne "modify" && @ARGV != 2) ||
	(($waitmode || $batch) && ($idleswap || $autoswap || $force)));

187
188
189
190
191
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
192
193
194
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

195
196
197
#
# Untaint the arguments.
#
198
if ($pid =~ /^([-\w\.]+)$/) {
199
200
201
202
203
    $pid = $1;
}
else {
    die("Tainted argument $pid!\n");
}
204
if ($eid =~ /^([-\w\.]+)$/) {
205
206
207
208
209
    $eid = $1;
}
else {
    die("Tainted argument $eid!\n");
}
210
my $repfile = "$eid.report";
211
212
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
213
214
215
my $tempnsfile;
my $modnsfile;

Leigh B. Stoller's avatar
Leigh B. Stoller committed
216
if ($inout eq "modify" && @ARGV > 2) {
217
218
219
220
221
    $tempnsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
222
    if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
223
	$tempnsfile = $1;
224
225
    }
    else {
226
227
228
229
230
231
232
233
234
235
	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;
236
    }
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
    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) {
258
259
260
	die("*** $0:\n".
	    "    $tempnsfile does not look like an NS file!\n");
    }
261
262
    $modnsfile = "$eid-modify.ns";
}
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278

#
# Verify user and get his DB uid.
#
if (! UNIX2DBUID($UID, \$dbuid)) {
    die("*** $0:\n".
	"    You do not exist in the Emulab Database.\n");
}

#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
    die("*** $0:\n".
	"    Cannot determine your name and email address.\n");
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
279
$isadmin = TBAdmin($UID);
280
281

#
Chad Barb's avatar
   
Chad Barb committed
282
# Verify that this person can muck with the experiment.
283
284
# Note that any script down the line has to do an admin check also. 
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
285
if ($UID && !$isadmin &&
286
287
    !TBExptAccessCheck($dbuid, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
Chad Barb's avatar
   
Chad Barb committed
288
	"    You do not have permission to swap or modify this experiment!\n");
289
290
}

291
292
293
294
# Must do this before lock tables!
# idleswap is in minutes, threshold is in hours
$idleswap_time = 60 * TBGetSiteVar("idle/threshold");

295
296
297
298
299
300
301
302
303
#
# In wait mode, block interrupt until we spin off the background process.
#
if ($waitmode) {
    $SIG{TERM} = 'IGNORE';
    $SIG{QUIT} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

304
305
306
307
308
309
310
311
312
313
314
#
# Check for overquota; we deal with it below, cause of the batch system.
#
my $overquota = system("$checkquota $dbuid");

if ($overquota) {
    print STDERR
	"*** $0:\n".
	"    You are over your disk quota on $CONTROL; please cleanup!\n";
}

315
316
317
318
319
320
#
# Temp fix; Disallow swapmod to firewalled experiments. This will come
# out later.
#
my $firewalled = TBExptFirewall($pid, $eid);

321
322
323
324
325
#
# 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
326
# the tb scripts.
327
328
329
330
331
332
333
334
335
336
337
338
339
#
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'};
340
my $batchstate      = $hashrow{'batchstate'};
341
my $expt_path       = $hashrow{'path'};
342
my $expt_locked     = $hashrow{'expt_locked'};
343
my $isbatchexpt     = $hashrow{'batchmode'};
344
my $canceled        = $hashrow{'canceled'};
345
346
347
348
349
350
351
352
353
354
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;
355
my $rendering   = $hashrow{'prerender_pid'};
356
my $elabinelab  = $hashrow{'elab_in_elab'};
357
my $lockdown    = $hashrow{'lockdown'};
358

359
360
if ($inout ne "out") {
    # I'm going to update this below, so fix the value before I use it.
361
    $idleswap_time = min($idleswaptime * 60, $idleswap_time);
362
363
364
    $idleswaptime = $idleswap_time / 60.0;
}

365
366
my $swapsettings = 
  "Idle-Swap:   $idleswapstr".
367
  ($idleswapbit ? ", at $idleswaptime hours\n" : " (Reason: $noidleswap)\n").
368
369
  "Auto-Swap:   $autoswapstr".
  ($autoswapbit ? ", at $autoswaptime hours\n" : "\n");
370

371
if (! chdir($workdir)) {
372
    die("*** $0:\n".
373
	"    Could not chdir to $workdir: $!\n");
374
375
}

376
#
377
378
379
# This script is called from the batch daemon.
# 
if ($batch) {
380
    #
381
382
383
    # 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. 
384
    #
385
386
387
388
389
390
391
392
393
    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());

394
395
396
397
    die("*** $0:\n".
	"    Batch experiment $pid/$eid is locked down; cannot be swapped!\n")
	if ($lockdown);

398
399
400
401
402
403
404
405
406
    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);
407
408
409
410
411

	# Do not allow it to swap in. What about swapout? 
	die("*** $0:\n".
	    "    Batch experiment cannot swap in when over quota! Aborting.\n")
	    if ($overquota);
412
413
414
415
416
417
    }
    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);
418
419
    }
    else {
420
421
422
423
424
425
	die("*** $0:\n".
	    "    Improper request from batch daemon for $pid/$eid!\n");
    }
}
else {
    if ($isbatchexpt) {
426
427
428
429
	#
	# User is requesting that a batch either be injected or paused.
	# Sanity check the state, but otherwise let the batch daemon
	# handle it.
430
431
	#
	ExitWithStatus(1, "Batch experiment $pid/$eid is still canceling!")
432
	    if ($canceled);
433

434
435
436
	ExitWithStatus(1, "Batch experiment $pid/$eid is locked down!")
	    if ($lockdown);

437
	if ($inout eq "in") {
438
	    ExitWithStatus(1,
439
440
441
			   "Batch experiment $pid/$eid must be SWAPPED to\n".
			   "QUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_SWAPPED);
442
443
444
445
446
447

	    ExitWithStatus(1,
			   "Batch experiment $pid/$eid cannot swap in when ".
			   "over quota!\n")
		if ($overquota);
	    
448
	    SetExpState($pid, $eid, EXPTSTATE_QUEUED);
449
450
	}
	elsif ($inout eq "out") {
451
	    ExitWithStatus(1,
452
453
454
455
			   "Batch experiment $pid/$eid must be ACTIVE or\n".
			   "ACTIVATING to swap out. Currently $estate.")
		if ($estate ne EXPTSTATE_ACTIVE &&
		    $estate ne EXPTSTATE_ACTIVATING);
456
457
458
459
460

	    #
	    # Since the batch daemon has control, all we can do is set
	    # the cancel bit.
	    # 
461
	    TBSetCancelFlag($pid, $eid, EXPTCANCEL_SWAP);
462
463
	}
	elsif ($inout eq "pause") {
464
	    ExitWithStatus(1,
465
466
467
			   "Batch experiment $pid/$eid must be QUEUED to\n".
			   "DEQUEUE. Currently $estate.")
		if ($estate ne EXPTSTATE_QUEUED);
468
469

	    #
470
471
472
473
	    # 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.
474
	    #
475
	    SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
476
	}
477
	elsif ($inout eq "modify") {
478
	    ExitWithStatus(1,
479
480
481
482
			   "Batch experiment $pid/$eid must be SWAPPED or\n".
			   "ACTIVE to modify. Currently $estate.")
		if (($estate ne EXPTSTATE_SWAPPED &&
		     $estate ne EXPTSTATE_ACTIVATING) ||
483
		    $batchstate ne BATCHSTATE_UNLOCKED());
484

485
486
487
	    ExitWithStatus(1,
			"Cannot modify an active firewalled experiment (yet).")
		if ($firewalled && $estate ne EXPTSTATE_SWAPPED && !$isadmin);
488

489
490
491
	    ExitWithStatus(1,
			"Cannot modify an active ElabInElab experiment (yet).")
		if ($elabinelab && $estate ne EXPTSTATE_SWAPPED && !$isadmin);
492
493
494
495
496
497

	    ExitWithStatus(1,
			   "Cannot modify batch experiment $pid/$eid when ".
			   "over quota!\n")
		if ($overquota);
	    
498
	    #
499
	    # Otherwise, proceed with the modify. The experiment will be
500
501
	    # locked below, and so it cannot be injected or otherwise messed
	    # with since its state is going to be changed before we unlock
502
503
504
505
	    # 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. 
506
	    #
507
508
	    goto doit;
	}
509
510
	else {
	    die("*** $0:\n",
511
		"    Operation $inout not allowed on a batch experiment!\n");
512
	}
513
514
	ExitWithStatus(0, 
		       "Batch experiment $pid/$eid state has been changed.\n");
515
      doit:
516
    }
517
518
519
520
521
522
523
524
525
526
    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);
527

528
529
530
531
 	ExitWithStatus(1,
		       "Experiment $pid/$eid is locked down; cannot swap!\n")
	    if ($lockdown);

532
533
534
535
536
537
538
539
540
541
	#
	# 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!");
		  }
542
543
544
545
546
		  ExitWithStatus(1,
				 "Experiment $pid/$eid cannot swap in when ".
				 "over quota!\n")
		      if ($overquota);
		  
547
548
549
550
		  last SWITCH;
	      };
	      /^out$/i && do {
		  if ($estate ne EXPTSTATE_ACTIVE() &&
Leigh B. Stoller's avatar
Leigh B. Stoller committed
551
 		      $estate ne EXPTSTATE_PANICED() &&
552
553
554
555
556
557
		      $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
558
559
560
561
562
563
564
565
566
567
568
569
 		  #
 		  # 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.");
 		  }

570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
		  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");
		  }
599
		  ExitWithStatus(1,
600
601
602
603
			"Cannot modify an active firewalled experiment (yet).")
		      if ($firewalled &&
			  $estate ne EXPTSTATE_SWAPPED && !$isadmin);

604
		  ExitWithStatus(1,
605
606
607
			"Cannot modify an active ElabInElab experiment (yet).")
		      if ($elabinelab &&
			  $estate ne EXPTSTATE_SWAPPED && !$isadmin);
608
609
610
611
612

		  ExitWithStatus(1,
				 "Experiment $pid/$eid cannot be modified ".
				 "when over quota!\n")
		      if ($overquota);
613
		  
614
615
616
617
618
		  last SWITCH;
	      };
	      die("*** $0:\n".
		  "    Missing state check for action: $action\n");
	  }
619
620
	}
    }
621
622
}

623
624
625
626
627
628
629
#
# 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 {
630
	$nextswapstate = EXPTSTATE_ACTIVATING();
631
632
633
	last SWITCH;
    };
    /^out$/i && do {
634
	$nextswapstate = EXPTSTATE_SWAPPING();
635
636
637
	last SWITCH;
    };
    /^restart$/i && do {
638
	$nextswapstate = EXPTSTATE_RESTARTING();
639
640
641
	last SWITCH;
    };
    /^modify$/i && do {
642
643
	$nextswapstate = (($estate eq EXPTSTATE_SWAPPED()) ?
			  EXPTSTATE_MODIFY_PARSE() : EXPTSTATE_MODIFY_REPARSE());
644
645
	last SWITCH;
    };
646
    die("*** $0:\n".
647
	"    Missing state check for action: $action\n");
648
}
649
650
 
# Update idleswap_timeout to whatever the current value is.
651
if ($inout ne "out") {
652
653
654
    DBQueryFatal("update experiments set idleswap_timeout='$idleswap_time' ".
		 "where eid='$eid' and pid='$pid'");
}
655

656
657
658
659
660
661
#
# 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.
662
663
664
665
666
667
668
669
670
671
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;

672
673
674
675
DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
676
677
678
679
# 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). 
680
681
#

682
683
684
685
686
687
688
689
690
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
691
692
693
if ($inout eq "modify") {
    $action = "modified";
}
694

695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
#
# 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;
}

710
711
712
#
# Before going to background, we have to copy out the NS file!
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
713
if ($inout eq "modify" && defined($modnsfile)) {
714
715
    unlink($modnsfile);
    if (system("/bin/cp", "$tempnsfile", "$modnsfile")) {
716
	fatal("Could not copy $tempnsfile to $modnsfile");
717
718
719
720
    }
    chmod(0664, "$modnsfile");
}

721
722
723
724
#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
725
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
726
    TBExptSetLogFile($pid, $eid, $logname);
727
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
728

729
730
    if (my $childpid = TBBackGround($logname)) {
	#
731
732
	# Parent exits normally, unless in waitmode. We have to set
	# justexit to make sure the END block below does not run.
733
	#
734
735
	$justexit = 1;

736
	if (!$waitmode) {
737
738
739
	    print("Experiment $pid/$eid is now being $action.\n".
		  "You will be notified via email when the this is done.\n")
		if (! $quiet);
740
741
	    exit(0);
	}
742
743
744
745
746
747
748
749
	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");
	}
750
751
752
753
754
755
756
757
758
759
760
761
	
	# 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';

762
	#
763
	# Wait until child exits or until user gets bored and types ^C.
764
	#
765
766
	waitpid($childpid, 0);
	
767
768
	print("Done. Exited with status: $?\n")
	    if (! $quiet);
769
	exit($? >> 8);
770
    }
771
    TBdbfork();
772
773
}

774
775
776
777
778
779
780
781
#
# 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();
}

782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
#
# 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';

802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
#
# 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);
}

818
819
820
#
# Remove old report file since its contents are going to be invalid.
#
821
if ($inout ne "restart" && -e $repfile) {
822
823
824
    unlink("$repfile");
}

825
826
827
828
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
829
    my $optarg = (($force || $idleswap) ? "-force" : "");
830

831
832
    print STDOUT "Running 'tbswap out $optarg $pid $eid'\n";
    if (system("$tbdir/tbswap out $optarg $pid $eid") != 0) {
833
	$errorstat = $? >> 8;
834
	fatal("tbswap out failed!");
835
    }
836
837
    SetExpState($pid, $eid, EXPTSTATE_SWAPPED)
	or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
Leigh B. Stoller's avatar
Leigh B. Stoller committed
838
    TBExptClearPanicBit($pid, $eid);
839
}
840
elsif ($inout eq "in") {
841
842
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPIN, 0, TBDB_STATS_FLAGS_PRESWAPIN);
843
    
844
    print STDOUT "Running 'tbswap in $pid $eid'\n";
Chad Barb's avatar
   
Chad Barb committed
845
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
846
	$errorstat = $? >> 8;
847
	fatal("tbswap in failed!");
848
    }
849
850
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE)
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
851

852
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
853
}
Chad Barb's avatar
   
Chad Barb committed
854
elsif ($inout eq "modify") {
855
    my $modifyError;
Chad Barb's avatar
Chad Barb committed
856

857
858
859
    GatherSwapStats($pid, $eid, $dbuid,
		    TBDB_STATS_SWAPMODIFY, 0, TBDB_STATS_FLAGS_PREMODIFY);

860
861
862
863
864
    # 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
865
    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
866
    if (TBExptBackupVirtualState($pid, $eid)) {
867
	fatal("Could not backup experiment state; cannot safely continue!");
Chad Barb's avatar
Chad Barb committed
868
869
870
    }

    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
871
    # Rerun tbprerun if modifying, but only if new NS file provided.
872
873
    # 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
874
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
875
876
877
    if (defined($modnsfile)) {
	print STDOUT "Running 'tbprerun $pid $eid $modnsfile'\n";
	if (system("$tbdir/tbprerun $pid $eid $modnsfile") != 0) {
878
	    print STDOUT "Modify Error: tbprerun failed.\n";
879
	  FWHOSED:
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
	    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
898
	}
899
900
901
902
903
904
905
906
907
	#
	# 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
908
909
    }

Chad Barb's avatar
   
Chad Barb committed
910
    #
911
    # Our next state depends on whether the experiment was active or swapped.
Chad Barb's avatar
   
Chad Barb committed
912
    #
913
914
915
916
917
    if ($estate eq EXPTSTATE_SWAPPED) {
	SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
    }
    else {
	SetExpState($pid, $eid, EXPTSTATE_MODIFY_RESWAP);
918
919
920
921
922
923
924
925
926
927
928
929
930

	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" : "");
	}
931

932
933
934
935
936
937
938
939
940
941
942
	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
943

944
	    #
945
946
947
	    # 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. 
948
	    # 
949
950
951
952
953
954
955
	    # 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) {
956
957
		$estate = EXPTSTATE_SWAPPED;
		$termswapstate = EXPTSTATE_SWAPPED;
958
		$modifySwapped = 1;
959
960
                # Old accounting info.
		TBSetExpSwapTime($pid, $eid);
961
962
963
964
		$modifyError = "Update aborted; experiment swapped out.";
	    }
	    else {
		$modifyError = "Update aborted; old state restored.";
Chad Barb's avatar
   
Chad Barb committed
965
	    }
966
	}
Chad Barb's avatar
Chad Barb committed
967
968
    }

969
970
971
972
973
974
975
976
977
    #
    # 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
978
    }
979
980
981
982
983
984
985

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

986
    TBExptClearBackupState($pid, $eid);
987
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
988
}
Chad Barb's avatar
   
Chad Barb committed
989
else { # $inout eq "restart" assumed.
990
    print STDOUT "Running 'tbrestart $pid $eid'\n";
991
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
992
	fatal("tbrestart failed!");
993
    }
994
    SetExpState($pid, $eid, EXPTSTATE_ACTIVE);
995
}
996

997
998
999
1000
#
# Try to copy off the files for testbed information gathering.
#
TBSaveExpLogFiles($pid, $eid);