batchexp.in 36.5 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2008 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
5 6
# All rights reserved.
#
7 8
use English;
use Getopt::Std;
9
use POSIX qw(isatty setsid);
10
use POSIX qw(strftime);
11
use Errno qw(EDQUOT);
12 13

#
14 15 16 17
# Create an experiment. The experiment is either run immediately, or
# placed into the batch system to be run later. If no NS file is
# supplied, an experiment shell is created (this is currently an admin
# only option).
18
#
19
# TODO: Remove expt_expires and priority.
20
#       Add calls to check_slot() to verify inputs.
21
# 
Leigh B. Stoller's avatar
Leigh B. Stoller committed
22 23 24 25 26 27 28 29 30
# 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 - Everything okay.
# $status > 0 - Expected error. User not allowed for some reason. 
# 
31 32
sub usage()
{
33
    print(STDERR
34
	  "Usage: batchexp [-q] [-i [-w]] [-n] [-f] [-E description] [-g gid]\n".
35
	  "                [-S reason] [-L reason] [-a <time>] [-l <time>]\n".
36
	  "                -p <pid> -e <eid> <nsfile>\n".
37
	  "switches and arguments:\n".
38 39 40
	  "-i       - swapin immediately; by default experiment is batched\n".
	  "-w       - wait for non-batchmode experiment to preload or swapin\n".
	  "-f       - preload experiment (do not swapin or queue yet)\n".
41
	  "-q       - be less chatty\n".
42 43
	  "-S <str> - Experiment cannot be swapped; must provide reason\n".
	  "-L <str> - Experiment cannot be IDLE swapped; must provide reason\n".
44
	  "-n       - Do not send idle email (internal option only)\n".
45 46
	  "-a <nnn> - Auto swapout nnn minutes after experiment is swapped in\n".
	  "-l <nnn> - Auto swapout nnn minutes after experiment goes idle\n".
47
	  "-s       - Save disk state on swapout\n".
48 49 50 51 52 53 54 55
	  "-E <str> - A pithy sentence describing your experiment\n".
	  "-p <pid> - The project in which to create the experiment\n".
	  "-g <gid> - The group in which to create the experiment\n".
	  "-e <eid> - The experiment name (unique, alphanumeric, no blanks)\n".
	  "<nsfile> - NS file to parse for experiment.\n");
    exit(-1);
}

56
sub ParseArgs();
57 58
sub CheckCopyArgs();
sub CopyInArchive();
59
sub fatal($;$);
60

61
my $optlist = "iE:g:e:p:S:L:a:l:sfwqt:nzc:bx:y:h:j";	# Enough options?
62
my $batchmode= 1;
63 64
my $frontend = 0;
my $waitmode = 0;
65
my $quiet    = 0;
66
my $linktest = 0;	# non-zero means level to run at.
67 68
my $zeemode  = 0;	# Hey, out of options.
my $zeeopt   = "";	# To pass along.
69
my $genimode;           # Allow creation with no NS file, uuid given.
70
my $savestate= 0;
71
my $nonsfile = 0;	# Admin only option to activate experiment anyway.
72 73 74
my $template;		# New stuff; experiment templates.
my $branch_template;    # New stuff; experiment templates.
my $instance;		# New stuff; experiment templates.
75 76 77 78 79 80 81 82 83
# All of these are for experiment dup and branch. Really mucks things up.
# These globals are set when we parse the -c argument, but used later
my $copyfrom;		# Copy from where, archive or current experiment.
my $copyarg;            # The -c argument.
my $copyidx;            # The index of the experiment copied.
my $copypid;		# The pid of the experiment copied.
my $copyeid;		# The eid of the experiment copied.
my $copytag;		# The archive tag to us.
my $copydir;		# Directory extracted from archive, to delete.
84 85 86 87 88

#
# Configure variables
#
my $TB       = "@prefix@";
89
my $EVENTSYS = @EVENTSYS@;
90 91
my $TBOPS    = "@TBOPSEMAIL@";
my $TBLOGS   = "@TBLOGSEMAIL@";
92 93
my $TBDOCBASE = "@TBDOCBASE@";
my $TBBASE   = "@TBBASE@";
94
my $CONTROL  = "@USERNODE@";
95

96 97 98 99 100 101
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
Kevin Atkinson's avatar
Kevin Atkinson committed
102
use libtblog;
103
use libArchive;
104
use Experiment;
105
use Template;
106
use User;
107
use Project;
108
use Group;
109

110 111 112 113
my $parser      = "$TB/libexec/parse-ns";
my $checkquota  = "$TB/sbin/checkquota";
my $tbbindir    = "$TB/bin/";
my $RSYNC	= "/usr/local/bin/rsync";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
114
my $errorstat=-1;
115
my $exptidx;
116
my $logfile;
117
my $logname;
118

119 120 121 122
# For the END block below.
my $cleaning = 0;
my $justexit = 1;
my $signaled = 0;
123 124 125 126 127 128

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

129 130 131 132 133 134 135
#
# 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);

136 137
#
# Untaint the path
138
#
139 140
# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
141 142
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

143 144 145 146 147
my $eid;
my $pid;
my $gid;
my $description;
my $tempnsfile;
148 149 150
my $swappable    = 1;
my $noswap_reason;
my $idleswap     = 1;
151
my $idleswaptime = 60 * TBGetSiteVar("idle/threshold");
152
my $noidleswap_reason;
153
my $autoswap     = 0;
154
my $autoswaptime = 10 * 60;
155 156 157
my $idleignore   = 0;
my $priority     = TB_EXPTPRIORITY_LOW;
my $exptstate    = EXPTSTATE_NEW();
158
my $batchstate   = BATCHSTATE_UNLOCKED();
159 160
my $now          = localtime();
my $committed    = 0;
161 162
my $experiment;
my $copy_experiment;
163

164
#
165
# Verify user and get his DB uid and other info for later.
166
#
167 168 169
my $this_user = User->ThisUser();
if (! defined($this_user)) {
    tbdie("You ($UID) do not exist!");
170
}
171 172 173 174
my $user_dbid  = $this_user->dbid();
my $user_uid   = $this_user->uid();
my $user_name  = $this_user->name();
my $user_email = $this_user->email();
175

176 177 178 179
#
# Before doing anything else, check for overquota ... lets not waste
# our time. Make sure user sees the error by exiting with 1.
#
180
if (system("$checkquota $user_uid") != 0) {
181 182
    tberror({cause => 'user', type => 'primary', severity => SEV_ERROR,
	     error => ['over_disk_quota', $CONTROL]},
183 184
	    "You are over your disk quota on $CONTROL; ".
	    "please login there and cleanup!");
185 186 187
    exit(1);
}

188
#
189
# Parse command arguments.
190
#
191
ParseArgs();
192
CheckCopyArgs();
193 194 195

#
# Sanity check them.
196
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
197 198 199
usage()
    if (!defined($pid) || !defined($eid));
usage()
200
    if (defined($tempnsfile) && defined($copyarg));
Leigh B. Stoller's avatar
Leigh B. Stoller committed
201

202 203 204
if (!defined($gid)) {
    $gid = $pid;
}
205
if (!defined($description)) {
206
    $description = "Created by $user_uid";
207
}
208
if (! $swappable && (!defined($noswap_reason) || $noswap_reason eq "")) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
209
    tbdie("Must provide a reason with -S option (not swappable reason)!");
210
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
211
if (! $idleswap && (!defined($noidleswap_reason) || $noidleswap_reason eq "")){
Kevin Atkinson's avatar
 
Kevin Atkinson committed
212
    tbdie("Must provide a reason with -L option (no idleswap reason)!");
213
}
214 215
if (!defined($tempnsfile) && !defined($copyarg)
    && !TBAdmin() && !defined($genimode)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
216
    tbdie("Only admins can create experiments with no NS file");
217
}
218
my $nsfile  = "$eid.ns";
219
my $repfile = "report";
220

221
# Defaults for the DB and for the email message. 
222
$noswap_reason = "None Given"
223
    if (!defined($noswap_reason));
224
$noidleswap_reason = "None Given"
225 226
    if (!defined($noidleswap_reason));

227 228 229 230 231 232
# Need the group
my $group = Group->Lookup($pid, $gid);
if (!defined($group)) {
    die("No such group $pid/$gid!");
}

233
#
234
# Make sure UID is allowed to create experiments in this project.
235
#
236
if (! $group->AccessCheck($this_user, TB_PROJECT_CREATEEXPT)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
237
    die("You do not have permission to create experiments in $pid/$gid");
238 239
}

240 241 242
if (defined($copyarg)) {
    # This will be handled below.
    ;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
243 244 245
}
elsif (!defined($tempnsfile)) {
    # If no NS file, then override swap/idle stuff. 
246 247 248
    $swappable     = 0;
    $idleswap      = 0;
}
249
elsif (! -f $tempnsfile || ! -r $tempnsfile || -z $tempnsfile) {
250
    # Exit so that user sees the error, not us.
251 252 253
    tberror({type => 'primary', severity => SEV_ERROR,
	     error => ['bogus_ns_file', $tempnsfile]},
	    "$tempnsfile does not exist or is not a readable file!");
254 255
    exit(1);
}
256

257 258 259 260
#
# Batch jobs get a shorter idle time
#
my $swaptime = $idleswaptime;
261
if ($batchmode && TBSiteVarExists("idle/batch_threshold")) {
262 263 264 265 266 267
    my $batchidleswaptime = TBGetSiteVar("idle/batch_threshold");
    if ($swaptime > $batchidleswaptime) {
	$swaptime = $batchidleswaptime;
    }
}

268
#
269 270
# Grab me a secret key for accessing tar/rpm files via the web interface.
# Grab another secret key for the event system HMACs.
271
#
272 273
my $webkey   = TBGenSecretKey();
my $eventkey = TBGenSecretKey();
274

275 276 277 278 279 280 281 282 283
#
# In wait mode, block SIGINT until we spin off the background process.
#
if ($waitmode) {
    $SIG{QUIT} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

284
#
285 286 287 288
# Create an arg array of parameters.
#
my %args = ();

289 290 291 292
$args{'expt_head_uid'}	   = $user_uid;
$args{'expt_swap_uid'}	   = $user_uid;
$args{'creator_idx'}	   = $user_dbid;
$args{'swapper_idx'}	   = $user_dbid;
293 294 295 296 297 298 299 300 301 302 303 304 305 306
$args{'state'}		   = $exptstate;
$args{'priority'}	   = $priority;
$args{'swappable'}	   = $swappable;
$args{'idleswap'}	   = $idleswap;
$args{'idleswap_timeout'}  = $swaptime;
$args{'autoswap'}	   = $autoswap;
$args{'autoswap_timeout'}  = $autoswaptime;
$args{'idle_ignore'}	   = $idleignore;
$args{'keyhash'}	   = $webkey;
$args{'eventkey'}	   = $eventkey;
$args{'batchmode'}	   = $batchmode;
$args{'batchstate'}	   = $batchstate;
$args{'linktest_level'}    = $linktest;
$args{'savedisk'}	   = $savestate;
307
$args{'instance_idx'}	   = (defined($instance) ? $instance->idx() : 0);
308 309 310 311
# These are special; the library will DBQuote them. 
$args{'expt_name'}	   = $description;
$args{'noswap_reason'}	   = $noswap_reason;
$args{'noidleswap_reason'} = $noidleswap_reason;
312 313
$args{'eid_uuid'}          = $genimode
    if (defined($genimode));
314 315

# Now create the experiment; we get back a perl class instance.
316
if (! ($experiment = Experiment->Create($group, $eid, \%args))) {
317 318 319
    tbdie({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['create_experiment_record_failed']},
	  "Could not create a new experiment record!");
320 321
}

322 323 324 325 326 327
#
# At this point, we need to force a cleanup no matter how we exit.
# See the END block below.
#
$justexit = 0;

Kevin Atkinson's avatar
Kevin Atkinson committed
328 329 330 331 332
#
# Set error reporting info
# 
tblog_set_info($pid,$eid,$UID);

333
#
334
# Create a directory structure for the experiment.
335 336 337 338 339
#
if ($experiment->CreateDirectory() != 0) {
    if (($? >> 8) == EDQUOT()) {
	# Obey exit status protocol for web page; User should see this.
	$errorstat = 1;
340
    }
341 342 343
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['create_experiment_directory_failed']},
	  "Failed to created experiment directory");
344
}
345 346 347
if (defined($instance)) {
    # Need to cross-mark the instance right away so that it is flagged.
    # Would be better to do this with a plain flag.
348
    my %args = ();
349
    $args{'exptidx'} = $experiment->idx();
350
    
351 352 353 354 355
    $instance->Update(0, \%args) == 0
	or fatal("Could not update experiment instance record!");
}
elsif (defined($template)) {
    # Tell the template with the new experiment index.
356 357 358 359 360
    %args = ();
    $args{'exptidx'} = $experiment->idx();

    $template->Update(\%args) == 0
	or fatal("Could not update template record!");
361 362
}

363 364 365 366
#
# Grab the working directory path, and thats where we work.
# The user's experiment directory is off in /proj space.
#
367 368
my $workdir = $experiment->WorkDir();
my $userdir = $experiment->UserDir();
369 370 371 372

chdir("$workdir") or
    fatal("Could not chdir to $workdir: $!");

373 374 375 376
#
# Create a new archive, which might actually be a branch of an existing one
# when doing an experiment branch (fork).
#
377 378 379
if (defined($branch_template)) {
    my $archive_eid = $branch_template->eid();
    
380 381 382
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['archive_op_failed', 'create', undef, undef]},
	  "Could not create experiment archive!")
Leigh B. Stoller's avatar
Leigh B. Stoller committed
383 384 385 386
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$pid, $archive_eid, undef)
	    < 0);
}
387
elsif (libArchive::TBCreateExperimentArchive($pid, $eid) < 0) {
388 389 390
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['archive_op_failed', 'create', undef, undef]},
	  "Could not create experiment archive!");
391 392 393 394 395 396 397 398 399 400 401 402
}

#
# Okay, if copying/branching an experiment, we have to go find the
# NS file, extracting the special (currently by convention) archive
# directory into the new experiment. This will set the tempnsfile
# variable needed below.
#
if ($copyarg) {
    CopyInArchive();
}

403 404 405 406 407 408 409 410 411 412
#
# It would be nice to check for overquota during the rest of this
# setup, but thats going to be a pain. Besides, its very unlikely
# that someone goes overquota at this point. Typically, the person
# is already overquota at this point. If this turns out to be wrong,
# it will be easy to add a little test to make sure there is at least
# a reasonable amount of room to proceed (create temp file and stick
# some data in it).
# 

413 414 415 416
#
# Dump the eventkey into a file in the experiment directory. 
#
if ($EVENTSYS) {
417
    open(KEY, ">" . $experiment->EventKeyPath()) or
418 419 420 421 422
	fatal("Could not create eventkey file: $!");
    print KEY $eventkey;
    close(KEY);
}

423
# And dump the web key too.
424
open(KEY, ">" . $experiment->WebKeyPath()) or
425 426 427 428
    fatal("Could not create webkey file: $!");
print KEY $webkey;
close(KEY);

429
#
430 431 432
# If no NS file, we are done. We must unlock it and reset its state
# appropriately. We leave the experiment in the "new" state so that
# the user is forced to do a modify first (to give it a topology). 
433
#
434
if (!defined($tempnsfile)) {
435 436 437 438 439 440 441 442 443 444
    #
    # In GeniMode, we skip past the nsfile stuff.
    #
    if (defined($genimode)) {
	$nonsfile = 1;
    }
    if (!$nonsfile) {
	$experiment->Unlock(EXPTSTATE_NEW());
	exit(0);
    }
445
}
446

447 448 449 450 451 452 453 454 455 456
if (!$nonsfile) {
    #
    # Now we can get the NS file!
    #
    if (system("/bin/cp", "$tempnsfile", "$nsfile")) {
	fatal({type => 'primary', severity => SEV_ERROR,
	       error => ['copy_ns_file_failed', $tempnsfile, $nsfile]},
	      "Could not copy $tempnsfile to $workdir/$nsfile");
    }
    chmod(0664, "$nsfile");
457

458 459 460 461 462 463
    # Future; do not name the ns as above, but use generic name.
    if (system("/bin/cp", "$tempnsfile", "nsfile.ns")) {
	fatal({type => 'primary', severity => SEV_ERROR,
	       error => ['copy_ns_file_failed', $tempnsfile, "nsfile.ns"]},
	      "Could not copy $tempnsfile to $workdir/$nsfile");
    }
464 465 466 467 468 469 470 471 472 473
}
# We created this file below so kill it.
if ($copyarg) {
    # This is tainted for reasons I do not understand.
    if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
	$tempnsfile = $1;
    }
    unlink($tempnsfile);
}

474
#
475 476
# Run parse in impotent mode on the NS file.  This has no effect but
# will display any errors.
477
#
478 479
if (!$nonsfile &&
    system("$parser -n $zeeopt $pid $gid $eid $nsfile") != 0) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
480 481
    # Obey exit status protocol for web page.
    $errorstat = 1;
482 483 484
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['ns_parse_failed']},
	  "NS Parse failed!");
485 486
}

487 488 489
#
# Gather statistics; start the clock ticking.
#
490
if ($frontend || $batchmode) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
491
    $experiment->PreSwap($this_user, TBDB_STATS_PRELOAD, $exptstate) == 0 or
492
	fatal("Preswap failed!");
493 494
}
else {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
495
    $experiment->PreSwap($this_user, TBDB_STATS_START, $exptstate) == 0 or
496
	fatal("Preswap failed!");
497 498
}

499
goto skiplog
500
    if (defined($template));
501

502 503 504 505 506 507
#
# The rest of this goes into the background so that the user sees
# immediate response. We will send email later when the experiment
# is ready. In waitmode, we hold the parent waiting so that the user
# can script it. Must protect against async (^C) termination though.
#
508 509
$logfile = $experiment->CreateLogFile("startexp");
if (!defined($logfile)) {
510 511
    fatal("Could not create logfile!");
}
512 513 514 515 516
$logname = $logfile->filename();
# We want it to spew to the web.
$experiment->SetLogFile($logfile);
# Mark it open since we are going to start using it right away.
$logfile->Open();
517

518 519
if (my $childpid = TBBackGround($logname)) {
    #
520 521
    # Parent exits normally, unless in waitmode. We have to set
    # justexit to make sure the END block below does not run.
522
    #
523 524
    $justexit = 1;
    
525 526 527
    if (!$waitmode) {
	print("Experiment $pid/$eid is now configuring\n".
 	      "You will be notified via email when the experiment is ".
528 529
	      "ready to use\n")
	    if (! $quiet);
530 531
	exit(0);
    }
532
    print("Waiting for " . ($batchmode ? "batch " : "") . "experiment $eid ".
533 534 535 536 537 538 539
	  "to finish " . ($frontend ? "preloading." : "swapping in.") . "\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");
    }
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556

    # 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{QUIT} = 'DEFAULT';
    $SIG{INT}  = 'DEFAULT';

    #
    # Wait until child exits or until user gets bored and types ^C.
    #
    waitpid($childpid, 0);

557 558
    print("Done. Exited with status: $?\n")
	if (! $quiet);
Kevin Atkinson's avatar
 
Kevin Atkinson committed
559 560 561 562 563 564 565 566 567

    my $exit_code = $? >> 8;
    
    if ($exit_code != 0) {
	my $d = tblog_lookup_error();
	print tblog_format_error($d);
    }
    
    exit $exit_code;
568
}
569
TBdbfork();
570 571 572 573 574 575 576 577

#
# 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();
}
578 579 580 581
skiplog:

# We are committed now. Affects how fatal() operates.
$committed = 1;
582

583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
#
# We need to catch TERM cause sometimes shit happens and we have to kill
# an experiment setup 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';

603 604
#
# The guts of starting an experiment!
605
#
606 607
# A batch experiment is essentially preloaded (frontend mode) and then
# dropped into the batch queue, unless the user requested only preload.
608 609
#

610
#
611
# Run the various scripts. We want to propagate the error from tbprerun
612 613
# and tbrun back out, hence the bogus looking errorstat variable.
#
614
$experiment->SetState(EXPTSTATE_PRERUN) == 0
615
    or fatal("Failed to set experiment state to " . EXPTSTATE_PRERUN());
616

617 618
if (!$nonsfile &&
    $experiment->PreRun($nsfile, ($zeeopt ? "-z" : ""))) {
619
    $errorstat = $? >> 8;
620 621 622
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbprerun_failed']},
	  "tbprerun failed!");
623
}
624 625

$experiment->SetState(EXPTSTATE_SWAPPED) == 0
626
    or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
627 628 629 630

#
# If not in frontend mode (preload only) continue to swapping exp in.
# 
631
if (! ($frontend || $batchmode)) {
632
    $experiment->SetState(EXPTSTATE_ACTIVATING) == 0
633
	or fatal("Failed to set experiment state to ". EXPTSTATE_ACTIVATING());
634

635 636
    if (!$nonsfile &&
	$experiment->Swap($Experiment::EXPT_SWAPIN) != 0) {
637
	$errorstat = $? >> 8;
638 639 640
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['tbswap_in_failed']},
	      "tbswap in failed!");
641
    }
642 643
    
    $experiment->SetState(EXPTSTATE_ACTIVE) == 0
644
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
645 646 647 648 649

    #
    # Look for the unsual case of more than 2 nodes and no vlans. Send a
    # warning message.
    #
650 651
    my @localnodes = ();
    fatal("Could not get local node list for $pid/$eid")
652
	if ($experiment->LocalNodeListNames(\@localnodes));
653

Leigh B. Stoller's avatar
Leigh B. Stoller committed
654
    if (@localnodes && scalar(@localnodes) > 2) {
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671
	my $vlans_result =
	    DBQueryFatal("select pid from virt_lans ".
			 "where pid='$pid' and eid='$eid'");
    
	if (!$vlans_result->numrows) {
	    SENDMAIL("$user_name <$user_email>",
		     "WARNING: Experiment Configuration: $pid/$eid",
		     "This experiment has zero network links defined.\n".
		     "Please check your NS file to verify this is what you ".
		     "want!\n",
		     "$user_name <$user_email>",
		     "Cc: $TBOPS", ($nsfile));
	}
    }
}

# We append this report in the email message below.
672
if ($experiment->Report($repfile, "-b") != 0) {
673 674 675
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbreport_failed']},
	  "tbreport failed!");
676 677
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
678 679 680 681 682
# Latest log is always called the same thing.
if (defined($logname)) {
    system("cp -fp $logname $workdir/" . EXPTLOGNAME());
}

683
#
684
# Save a copy of the files for testbed information gathering (long term).
685
#
686

687
$experiment->SaveLogFiles();
688 689 690 691 692

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files.
#
693
$experiment->CopyLogFiles();
694

695
# Tell the archive library to add all files to the archive. 
696 697
libArchive::TBExperimentArchiveAddUserFiles($pid, $eid) == 0
    or fatal("Failed to add user archive files to the archive!");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
698

699
#
700 701
# Do a SavePoint on the experiment files. In template mode, let the wrapper
# deal with this. Avoids duplication of work.
702
#
703
if (! defined($template)) {
704 705 706 707 708 709
    print "Doing a savepoint on the experiment archive ...\n";
    if (libArchive::TBExperimentArchiveSavePoint($pid, $eid, "startexp") < 0) {
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['archive_op_failed', 'savepoint', undef, undef]},
	      "Failed to do a savepoint on the experiment archive!");
    }
710 711
}

712
#
713 714
# Gather statistics. This is not likely to fail, but if it does I want to
# bail cause the inconsistent records are a pain in the ass to deal with!
715 716
#
if ($frontend || $batchmode) {
717 718
    $experiment->PostSwap($this_user, TBDB_STATS_PRELOAD) == 0 or
	fatal("Postswap failed!");
719 720
}
else {
721 722
    $experiment->PostSwap($this_user, TBDB_STATS_START) == 0 or
	fatal("Postswap failed!");
723 724
}

725
#
726
# Set accounting stuff, but on success only, and *after* gathering swap stats!
727
#
728
$experiment->SetSwapInfo($this_user);
729

730 731 732
#
# Close up the log file so the webpage stops.
#
733
if (!defined($template)) {
734
    print "Experiment $pid/$eid has been successfully created!\n";
735
    $experiment->CloseLogFile();
736
}
737 738 739 740

#
# Must unlock and drop batch experiments into the queue before exit.
#
741
if ($batchmode && !$frontend) {
742
    $experiment->Unlock(EXPTSTATE_QUEUED());
743 744
}
else {
745
    $experiment->Unlock();
746 747 748 749 750 751
}

#
# Clear the cancel flag now that the operation is complete. Must be
# done after we change the experiment state (above).
#
752
$experiment->SetCancelFlag(EXPTCANCEL_CLEAR());
753

754 755 756 757
#
# In template_mode we are done; the caller finishes up.
#
exit(0)
758
    if (defined($template));
759

760 761 762
#
# Dump the report file and the log file to the user via email. 
#
763
my ($expt_created) = $experiment->created();
764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781
my $message;

if ($frontend) {
    $message =
	"Your experiment `$eid' in project `$pid' has been created.\n" .
	"You can check the web interface to see if it looks the way\n" .
	"you expected it to. If so, you may swap the experiment in,\n" .
	"or terminate it, at any time.\n" .
        "\n";
}
else {
    $message =
	"Your experiment `$eid' in project `$pid' has been started.\n" .
	"Here is the experiment summary detailing the nodes that were\n" .
	"allocated to you. You may use the `Qualified Name' to log on\n" .
	"to your nodes. See /etc/hosts on your nodes (when running\n" .
	"FreeBSD, Linux, or NetBSD) for the IP mapping on each node.\n" .
        "\n";
782
}
783 784 785 786 787 788 789 790
$message .=
    "User:        $user_name\n" .
    "EID:         $eid\n" .
    "PID:         $pid\n" .
    "GID:         $gid\n" .
    "Description: $description\n" .
    "Swappable:   " . ($swappable ? "Yes\n" :
		                    "No  (Reason: $noswap_reason)\n") .
791
    "Idle-Swap:   " . ($idleswap  ? "Yes, at " . $idleswaptime/60.0 . " hours\n":
792
		                    "No  (Reason: $noidleswap_reason)\n") .
793
    "Auto-Swap:   " . ($autoswap  ? "Yes, at " . $autoswaptime/60.0 . " hours\n":
794 795 796 797 798 799 800 801 802 803 804 805 806 807
		                    "No\n") .
    "Created:     $expt_created\n".
    "Directory:   $userdir\n".
    "\n".
    "Appended at the end is the output of the experiment setup. If you\n" .
    "have any questions or comments, please include the output below\n" .
    "in your message to $TBOPS";

SENDMAIL("$user_name <$user_email>",
	 "New Experiment " . (($frontend == 0) ? "Started" : "Created") .
	 ": $pid/$eid",
	 $message,
	 "$user_name <$user_email>",
	 "Bcc: $TBLOGS",
808 809
	 ($repfile, $logname, $nsfile))
    if (!$zeemode);
810 811

# Done!
812
exit(0);
813

814 815 816
#
#
#
817
sub cleanup()
818
{
819
    #
820 821 822 823 824 825
    # Failed early (say, in parsing). No point in keeping any of the
    # stats or resource records. Just a waste of space since the
    # testbed_stats log indicates there was a failure and why (sorta,
    # via the exit code).
    # 
    if (!$committed) {
826 827 828
	# Completely remove all trace of the archive.
        libArchive::TBDestroyExperimentArchive($pid, $eid);

829
	#
830
	# Clear the experiment record and cleanup directories
831
	#
832 833
	$experiment->Delete(1)
	    if (defined($experiment));
834

835
	return;
836
    }
837

838 839 840 841
    #
    # Gather statistics.
    #
    if ($frontend) {
842
	$experiment->SwapFail($this_user, TBDB_STATS_PRELOAD, $errorstat);
843 844
    }
    else {
845
	$experiment->SwapFail($this_user, TBDB_STATS_START, $errorstat);
846
    }
847

848 849 850
    #
    # Must clean up the experiment if it made it our of NEW state.
    #
851
    my $estate = $experiment->state();
852 853 854 855 856 857 858 859
    if ($estate ne EXPTSTATE_NEW) {
	#
	# We do not know exactly where things stopped, so if the
	# experiment was activating when the signal was delivered,
	# run tbswap on it. 
	# 
	if ($estate eq EXPTSTATE_ACTIVE ||
	    ($estate eq EXPTSTATE_ACTIVATING && $signaled)) {
860 861
	    if ($experiment->Swap("out", "-force") != 0) {
		print "tbswap out -force failed!\n";
862
	    }
863
	    $experiment->SetState(EXPTSTATE_SWAPPED);
864
	}
865
	
866
	if ($experiment->End("-f") != 0) {
867
	    print "tbend failed!\n";
868 869
	}
    }
870
    $experiment->SetState(EXPTSTATE_TERMINATED);
871

872 873
    # Old swap gathering stuff.
    $experiment->GatherSwapStats($this_user, TBDB_STATS_TERMINATE, 0);
874

875 876
    # Clear the logfile so the webpage stops.
    $experiment->CloseLogFile();
877

Kevin Atkinson's avatar
Kevin Atkinson committed
878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896
    if (!$ENV{'TBAUDITON'}) {
	#
	# Figure out the error if possible
	#
	my $error_data = tblog_find_error();
	
	#
	# Send a message to the testbed list. 
	#
	tblog_email_error($error_data,
			  "$user_name <$user_email>",
			  "Config Failure", "$pid/$eid",
			  "$user_name <$user_email>",
			  "",
			  "Cc: $TBOPS",
			  "",
			  ($logname, "assign.log", "wanassign.log", $nsfile));
    } 	
	
897
    #
898 899 900 901
    # Back up the work dir for post-mortem debugging. 
    #
    system("/bin/rm -rf  ${workdir}-failed");
    system("/bin/mv -f   $workdir ${workdir}-failed");
902

903 904
    #
    # Clear the record and cleanup.
905 906
    #
    $experiment->Delete();
907 908 909
}

#
910 911
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
912
#
913
sub ParseArgs()
914
{
915 916 917 918
    my %options = ();
    if (! getopts($optlist, \%options)) {
	usage();
    }
919

920
    if (@ARGV > 1) {
921 922
	usage();
    }
923 924
    if (@ARGV == 1) {
	$tempnsfile = $ARGV[0];
925

926
	# Note different taint check (allow /).
927
	if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
928 929 930
	    $tempnsfile = $1;
	}
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
931
	    tbdie("Bad data in nsfile: $tempnsfile");
932
	}
933

934
	#
935 936
	# Called from ops interactively. Make sure NS file resides in an
	# appropriate location.
937 938 939 940 941 942 943 944
	#
	# Use realpath to resolve any symlinks.
	#
	my $translated = `realpath $tempnsfile`;
	if ($translated =~ /^([-\w\.\/]+)$/) {
	    $tempnsfile = $1;
	}
	else {
945 946 947
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'realpath', $translated]},
		  "Bad data returned by realpath: $translated");
948 949 950
	}

	#
951 952 953 954 955 956 957
	# The file must reside in an acceptible location. Since this script
	# runs as the caller, regular file permission checks ensure it is a
	# file the user is allowed to use.  So we don't have to be too tight
	# with the RE matching /tmp and /var/tmp files.  Note that
	# /tmp/$guid-$nsref.nsfile is also allowed since this script is
	# invoked directly from web interface which generates a name that
	# should not be guessable.
958 959
	#
	if (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
960
	    ! ($tempnsfile =~ /^\/tmp\/\d+\.ns/) &&
961
	    ! ($tempnsfile =~ /^\/var\/tmp\/php[-\w]+/) &&
962
	    ! TBValidUserDir($tempnsfile, 0)) {
963 964 965
	    tberror({type => 'primary', severity => SEV_ERROR,
		     error => ['disallowed_directory', $tempnsfile]},
		    "$tempnsfile does not resolve to an allowed directory!");
966 967 968 969 970
	    # Note positive status; so error goes to user not tbops.
	    exit(1);
	}
    }
    
971
    if (defined($options{"i"})) {
972
	$batchmode = 0;
973
    }
974 975 976
    if (defined($options{"f"})) {
	$frontend = 1;
    }
977 978 979
    if (defined($options{"q"})) {
	$quiet = 1;
    }
980 981 982 983
    if (defined($options{"z"})) {
	$zeemode = 1;
	$zeeopt  = "-p";
    }
984 985 986 987
    # This option should not be exported via the XMLRPC server. 
    if (defined($options{"n"})) {
	$idleignore = 1;
    }
988

989 990 991
    if (defined($options{"s"})) {
	$savestate = 1;
    }
992 993 994 995 996 997 998 999 1000 1001 1002
    if (defined($options{"h"})) {
	$genimode = $options{"h"};
	if (! ($genimode =~ /^[-\w]+$/)) {
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $genimode]},
		  "Bad data in argument: $genimode.");
	}
    }
    if (defined($options{"j"})) {
	$nonsfile = 1;
    }
1003

Leigh B. Stoller's avatar
Leigh B. Stoller committed
1004 1005 1006 1007 1008
    #
    # Clone an experiment, either an existing experiment or an old one
    # (using the archive). 
    #
    if (defined($options{"c"})) {
1009
	$copyarg = $options{"c"};
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1010

1011 1012
	if (! (($copyarg =~ /^([-\w]+),([-\w]+)(?::[-\w]*)?$/) ||
	       ($copyarg =~ /^(\d+)(?::[-\w]*)?$/))) {
1013 1014 1015
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $copyarg]},
		  "Bad data in argument: $copyarg");
1016
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1017 1018
    }

1019 1020 1021
    #
    # pid,eid,gid get passed along as shell commands args; must taint check.
    # 
1022 1023
    if (defined($options{"p"})) {
	$pid = $options{"p"};
1024

1025
	if ($pid =~ /^([-\w]+)$/) {
1026 1027
	    $pid = $1;
	}
1028
	else {
1029 1030 1031
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $pid]},
		  "Bad data in argument: $pid.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1032
	}
1033 1034 1035 1036
    }
    if (defined($options{"e"})) {
	$eid = $options{"e"};

1037
	if ($eid =~ /^([-\w]+)$/) {
1038 1039 1040
	    $eid = $1;
	}
	else {
1041 1042 1043
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $eid]},
		  "Bad data in argument: $eid.");
1044
	}
1045 1046
	if (! TBcheck_dbslot($eid, "experiments", "eid",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1047 1048 1049
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'eid', $eid]},
		  "Improper experiment name (id)!");
1050
	}
1051 1052 1053 1054
    }
    if (defined($options{"g"})) {
	$gid = $options{"g"};

1055
	if ($gid =~ /^([-\w]+)$/) {
1056
	    $gid = $1;
1057
	}
1058
	else {
1059 1060 1061
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $gid]},
		  "Bad data in argument: $gid.");
1062 1063
	}
    }
1064
    if (defined($options{"E"})) {