batchexp.in 36.6 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh Stoller's avatar
Leigh Stoller committed
2 3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2008 University of Utah and the Flux Group.
Leigh Stoller's avatar
Leigh 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 Stoller's avatar
Leigh 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:jk";	# Enough options?
62
my $batchmode= 1;
63 64
my $frontend = 0;
my $waitmode = 0;
65
my $quiet    = 0;
66
my $lockdown = 0;
67
my $linktest = 0;	# non-zero means level to run at.
68 69
my $zeemode  = 0;	# Hey, out of options.
my $zeeopt   = "";	# To pass along.
70
my $genimode;           # Allow creation with no NS file, uuid given.
71
my $savestate= 0;
72
my $nonsfile = 0;	# Admin only option to activate experiment anyway.
73 74 75
my $template;		# New stuff; experiment templates.
my $branch_template;    # New stuff; experiment templates.
my $instance;		# New stuff; experiment templates.
76 77 78 79 80 81 82 83 84
# 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.
85 86 87 88 89

#
# Configure variables
#
my $TB       = "@prefix@";
90
my $EVENTSYS = @EVENTSYS@;
91 92
my $TBOPS    = "@TBOPSEMAIL@";
my $TBLOGS   = "@TBLOGSEMAIL@";
93 94
my $TBDOCBASE = "@TBDOCBASE@";
my $TBBASE   = "@TBBASE@";
95
my $CONTROL  = "@USERNODE@";
96

97 98 99 100 101 102
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
103
use libtblog;
104
use libArchive;
105
use Experiment;
106
use Template;
107
use User;
108
use Project;
109
use Group;
110

111 112 113 114
my $parser      = "$TB/libexec/parse-ns";
my $checkquota  = "$TB/sbin/checkquota";
my $tbbindir    = "$TB/bin/";
my $RSYNC	= "/usr/local/bin/rsync";
Leigh Stoller's avatar
Leigh Stoller committed
115
my $errorstat=-1;
116
my $exptidx;
117
my $logfile;
118
my $logname;
119

120 121 122 123
# For the END block below.
my $cleaning = 0;
my $justexit = 1;
my $signaled = 0;
124 125 126 127 128 129

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

130 131 132 133 134 135 136
#
# 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);

137 138
#
# Untaint the path
139
#
140 141
# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
142 143
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

144 145 146 147 148
my $eid;
my $pid;
my $gid;
my $description;
my $tempnsfile;
149 150 151
my $swappable    = 1;
my $noswap_reason;
my $idleswap     = 1;
152
my $idleswaptime = 60 * TBGetSiteVar("idle/threshold");
153
my $noidleswap_reason;
154
my $autoswap     = 0;
155
my $autoswaptime = 10 * 60;
156 157 158
my $idleignore   = 0;
my $priority     = TB_EXPTPRIORITY_LOW;
my $exptstate    = EXPTSTATE_NEW();
159
my $batchstate   = BATCHSTATE_UNLOCKED();
160 161
my $now          = localtime();
my $committed    = 0;
162 163
my $experiment;
my $copy_experiment;
164

165
#
166
# Verify user and get his DB uid and other info for later.
167
#
168 169 170
my $this_user = User->ThisUser();
if (! defined($this_user)) {
    tbdie("You ($UID) do not exist!");
171
}
172 173 174 175
my $user_dbid  = $this_user->dbid();
my $user_uid   = $this_user->uid();
my $user_name  = $this_user->name();
my $user_email = $this_user->email();
176

177 178 179 180
#
# Before doing anything else, check for overquota ... lets not waste
# our time. Make sure user sees the error by exiting with 1.
#
181
if (system("$checkquota $user_uid") != 0) {
182 183
    tberror({cause => 'user', type => 'primary', severity => SEV_ERROR,
	     error => ['over_disk_quota', $CONTROL]},
184 185
	    "You are over your disk quota on $CONTROL; ".
	    "please login there and cleanup!");
186 187 188
    exit(1);
}

189
#
190
# Parse command arguments.
191
#
192
ParseArgs();
193
CheckCopyArgs();
194 195 196

#
# Sanity check them.
197
#
198 199 200
usage()
    if (!defined($pid) || !defined($eid));
usage()
201
    if (defined($tempnsfile) && defined($copyarg));
202

203 204 205
if (!defined($gid)) {
    $gid = $pid;
}
206
if (!defined($description)) {
207
    $description = "Created by $user_uid";
208
}
209
if (! $swappable && (!defined($noswap_reason) || $noswap_reason eq "")) {
210
    tbdie("Must provide a reason with -S option (not swappable reason)!");
211
}
212
if (! $idleswap && (!defined($noidleswap_reason) || $noidleswap_reason eq "")){
213
    tbdie("Must provide a reason with -L option (no idleswap reason)!");
214
}
215 216
if (!defined($tempnsfile) && !defined($copyarg)
    && !TBAdmin() && !defined($genimode)) {
217
    tbdie("Only admins can create experiments with no NS file");
218
}
219
my $nsfile  = "$eid.ns";
220
my $repfile = "report";
221

222
# Defaults for the DB and for the email message. 
223
$noswap_reason = "None Given"
224
    if (!defined($noswap_reason));
225
$noidleswap_reason = "None Given"
226 227
    if (!defined($noidleswap_reason));

228 229 230 231 232 233
# Need the group
my $group = Group->Lookup($pid, $gid);
if (!defined($group)) {
    die("No such group $pid/$gid!");
}

234
#
235
# Make sure UID is allowed to create experiments in this project.
236
#
237
if (! $group->AccessCheck($this_user, TB_PROJECT_CREATEEXPT)) {
238
    die("You do not have permission to create experiments in $pid/$gid");
239 240
}

241 242 243
if (defined($copyarg)) {
    # This will be handled below.
    ;
244 245 246
}
elsif (!defined($tempnsfile)) {
    # If no NS file, then override swap/idle stuff. 
247 248 249
    $swappable     = 0;
    $idleswap      = 0;
}
250
elsif (! -f $tempnsfile || ! -r $tempnsfile || -z $tempnsfile) {
251
    # Exit so that user sees the error, not us.
252 253 254
    tberror({type => 'primary', severity => SEV_ERROR,
	     error => ['bogus_ns_file', $tempnsfile]},
	    "$tempnsfile does not exist or is not a readable file!");
255 256
    exit(1);
}
257

258 259 260 261
#
# Batch jobs get a shorter idle time
#
my $swaptime = $idleswaptime;
262
if ($batchmode && TBSiteVarExists("idle/batch_threshold")) {
263 264 265 266 267 268
    my $batchidleswaptime = TBGetSiteVar("idle/batch_threshold");
    if ($swaptime > $batchidleswaptime) {
	$swaptime = $batchidleswaptime;
    }
}

269
#
270 271
# Grab me a secret key for accessing tar/rpm files via the web interface.
# Grab another secret key for the event system HMACs.
272
#
273 274
my $webkey   = TBGenSecretKey();
my $eventkey = TBGenSecretKey();
275

276 277 278 279 280 281 282 283 284
#
# In wait mode, block SIGINT until we spin off the background process.
#
if ($waitmode) {
    $SIG{QUIT} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

285
#
286 287 288 289
# Create an arg array of parameters.
#
my %args = ();

290 291 292 293
$args{'expt_head_uid'}	   = $user_uid;
$args{'expt_swap_uid'}	   = $user_uid;
$args{'creator_idx'}	   = $user_dbid;
$args{'swapper_idx'}	   = $user_dbid;
294 295 296 297 298 299 300 301 302
$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;
303
$args{'lockdown'}	   = $lockdown;
304 305 306 307 308
$args{'eventkey'}	   = $eventkey;
$args{'batchmode'}	   = $batchmode;
$args{'batchstate'}	   = $batchstate;
$args{'linktest_level'}    = $linktest;
$args{'savedisk'}	   = $savestate;
309
$args{'instance_idx'}	   = (defined($instance) ? $instance->idx() : 0);
310 311 312 313
# These are special; the library will DBQuote them. 
$args{'expt_name'}	   = $description;
$args{'noswap_reason'}	   = $noswap_reason;
$args{'noidleswap_reason'} = $noidleswap_reason;
314 315
$args{'eid_uuid'}          = $genimode
    if (defined($genimode));
316 317

# Now create the experiment; we get back a perl class instance.
318
if (! ($experiment = Experiment->Create($group, $eid, \%args))) {
319 320 321
    tbdie({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['create_experiment_record_failed']},
	  "Could not create a new experiment record!");
322 323
}

324 325 326 327 328 329
#
# At this point, we need to force a cleanup no matter how we exit.
# See the END block below.
#
$justexit = 0;

330 331 332 333 334
#
# Set error reporting info
# 
tblog_set_info($pid,$eid,$UID);

335
#
336
# Create a directory structure for the experiment.
337 338 339 340 341
#
if ($experiment->CreateDirectory() != 0) {
    if (($? >> 8) == EDQUOT()) {
	# Obey exit status protocol for web page; User should see this.
	$errorstat = 1;
342
    }
343 344 345
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['create_experiment_directory_failed']},
	  "Failed to created experiment directory");
346
}
347 348 349
if (defined($instance)) {
    # Need to cross-mark the instance right away so that it is flagged.
    # Would be better to do this with a plain flag.
350
    my %args = ();
351
    $args{'exptidx'} = $experiment->idx();
352
    
353 354 355 356 357
    $instance->Update(0, \%args) == 0
	or fatal("Could not update experiment instance record!");
}
elsif (defined($template)) {
    # Tell the template with the new experiment index.
358 359 360 361 362
    %args = ();
    $args{'exptidx'} = $experiment->idx();

    $template->Update(\%args) == 0
	or fatal("Could not update template record!");
363 364
}

365 366 367 368
#
# Grab the working directory path, and thats where we work.
# The user's experiment directory is off in /proj space.
#
369 370
my $workdir = $experiment->WorkDir();
my $userdir = $experiment->UserDir();
371 372 373 374

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

375 376 377 378
#
# Create a new archive, which might actually be a branch of an existing one
# when doing an experiment branch (fork).
#
379 380 381
if (defined($branch_template)) {
    my $archive_eid = $branch_template->eid();
    
382 383 384
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['archive_op_failed', 'create', undef, undef]},
	  "Could not create experiment archive!")
385 386 387 388
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$pid, $archive_eid, undef)
	    < 0);
}
389
elsif (libArchive::TBCreateExperimentArchive($pid, $eid) < 0) {
390 391 392
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['archive_op_failed', 'create', undef, undef]},
	  "Could not create experiment archive!");
393 394 395 396 397 398 399 400 401 402 403 404
}

#
# 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();
}

405 406 407 408 409 410 411 412 413 414
#
# 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).
# 

415 416 417 418
#
# Dump the eventkey into a file in the experiment directory. 
#
if ($EVENTSYS) {
419
    open(KEY, ">" . $experiment->EventKeyPath()) or
420 421 422 423 424
	fatal("Could not create eventkey file: $!");
    print KEY $eventkey;
    close(KEY);
}

425
# And dump the web key too.
426
open(KEY, ">" . $experiment->WebKeyPath()) or
427 428 429 430
    fatal("Could not create webkey file: $!");
print KEY $webkey;
close(KEY);

431
#
432 433 434
# 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). 
435
#
436
if (!defined($tempnsfile)) {
437 438 439 440 441 442 443 444 445 446
    #
    # In GeniMode, we skip past the nsfile stuff.
    #
    if (defined($genimode)) {
	$nonsfile = 1;
    }
    if (!$nonsfile) {
	$experiment->Unlock(EXPTSTATE_NEW());
	exit(0);
    }
447
}
448

449 450 451 452 453 454 455 456 457 458
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");
459

460 461 462 463 464 465
    # 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");
    }
466 467 468 469 470 471 472 473 474 475
}
# 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);
}

476
#
477 478
# Run parse in impotent mode on the NS file.  This has no effect but
# will display any errors.
479
#
480 481
if (!$nonsfile &&
    system("$parser -n $zeeopt $pid $gid $eid $nsfile") != 0) {
Leigh Stoller's avatar
Leigh Stoller committed
482 483
    # Obey exit status protocol for web page.
    $errorstat = 1;
484 485 486
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['ns_parse_failed']},
	  "NS Parse failed!");
487 488
}

489 490 491
#
# Gather statistics; start the clock ticking.
#
492
if ($frontend || $batchmode) {
Leigh Stoller's avatar
Leigh Stoller committed
493
    $experiment->PreSwap($this_user, TBDB_STATS_PRELOAD, $exptstate) == 0 or
494
	fatal("Preswap failed!");
495 496
}
else {
Leigh Stoller's avatar
Leigh Stoller committed
497
    $experiment->PreSwap($this_user, TBDB_STATS_START, $exptstate) == 0 or
498
	fatal("Preswap failed!");
499 500
}

501
goto skiplog
502
    if (defined($template));
503

504 505 506 507 508 509
#
# 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.
#
510 511
$logfile = $experiment->CreateLogFile("startexp");
if (!defined($logfile)) {
512 513
    fatal("Could not create logfile!");
}
514 515 516 517 518
$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();
519

520 521
if (my $childpid = TBBackGround($logname)) {
    #
522 523
    # Parent exits normally, unless in waitmode. We have to set
    # justexit to make sure the END block below does not run.
524
    #
525 526
    $justexit = 1;
    
527 528 529
    if (!$waitmode) {
	print("Experiment $pid/$eid is now configuring\n".
 	      "You will be notified via email when the experiment is ".
530 531
	      "ready to use\n")
	    if (! $quiet);
532 533
	exit(0);
    }
534
    print("Waiting for " . ($batchmode ? "batch " : "") . "experiment $eid ".
535 536 537 538 539 540 541
	  "to finish " . ($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");
    }
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558

    # 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);

559 560
    print("Done. Exited with status: $?\n")
	if (! $quiet);
561 562 563 564 565 566 567 568 569

    my $exit_code = $? >> 8;
    
    if ($exit_code != 0) {
	my $d = tblog_lookup_error();
	print tblog_format_error($d);
    }
    
    exit $exit_code;
570
}
571
TBdbfork();
572 573 574 575 576 577 578 579

#
# 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();
}
580 581 582 583
skiplog:

# We are committed now. Affects how fatal() operates.
$committed = 1;
584

585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
#
# 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';

605 606
#
# The guts of starting an experiment!
607
#
608 609
# A batch experiment is essentially preloaded (frontend mode) and then
# dropped into the batch queue, unless the user requested only preload.
610 611
#

612
#
613
# Run the various scripts. We want to propagate the error from tbprerun
614 615
# and tbrun back out, hence the bogus looking errorstat variable.
#
616
$experiment->SetState(EXPTSTATE_PRERUN) == 0
617
    or fatal("Failed to set experiment state to " . EXPTSTATE_PRERUN());
618

619 620
if (!$nonsfile &&
    $experiment->PreRun($nsfile, ($zeeopt ? "-z" : ""))) {
621
    $errorstat = $? >> 8;
622 623 624
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbprerun_failed']},
	  "tbprerun failed!");
625
}
626 627

$experiment->SetState(EXPTSTATE_SWAPPED) == 0
628
    or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
629 630 631 632

#
# If not in frontend mode (preload only) continue to swapping exp in.
# 
633
if (! ($frontend || $batchmode)) {
634
    $experiment->SetState(EXPTSTATE_ACTIVATING) == 0
635
	or fatal("Failed to set experiment state to ". EXPTSTATE_ACTIVATING());
636

637 638
    if (!$nonsfile &&
	$experiment->Swap($Experiment::EXPT_SWAPIN) != 0) {
639
	$errorstat = $? >> 8;
640 641 642
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['tbswap_in_failed']},
	      "tbswap in failed!");
643
    }
644 645
    
    $experiment->SetState(EXPTSTATE_ACTIVE) == 0
646
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
647 648 649 650 651

    #
    # Look for the unsual case of more than 2 nodes and no vlans. Send a
    # warning message.
    #
652 653
    my @localnodes = ();
    fatal("Could not get local node list for $pid/$eid")
654
	if ($experiment->LocalNodeListNames(\@localnodes));
655

Leigh Stoller's avatar
Leigh Stoller committed
656
    if (@localnodes && scalar(@localnodes) > 2) {
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673
	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.
674
if ($experiment->Report($repfile, "-b") != 0) {
675 676 677
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbreport_failed']},
	  "tbreport failed!");
678 679
}

680 681 682 683 684
# Latest log is always called the same thing.
if (defined($logname)) {
    system("cp -fp $logname $workdir/" . EXPTLOGNAME());
}

685
#
686
# Save a copy of the files for testbed information gathering (long term).
687
#
688

689
$experiment->SaveLogFiles();
690 691 692 693 694

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files.
#
695
$experiment->CopyLogFiles();
696

697
# Tell the archive library to add all files to the archive. 
698 699
libArchive::TBExperimentArchiveAddUserFiles($pid, $eid) == 0
    or fatal("Failed to add user archive files to the archive!");
700

701
#
702 703
# Do a SavePoint on the experiment files. In template mode, let the wrapper
# deal with this. Avoids duplication of work.
704
#
705
if (! defined($template)) {
706 707 708 709 710 711
    print "Doing a savepoint on the experiment archive ...\n";
    if (libArchive::TBExperimentArchiveSavePoint($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!");
    }
712 713
}

714
#
715 716
# 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!
717 718
#
if ($frontend || $batchmode) {
719 720
    $experiment->PostSwap($this_user, TBDB_STATS_PRELOAD) == 0 or
	fatal("Postswap failed!");
721 722
}
else {
723 724
    $experiment->PostSwap($this_user, TBDB_STATS_START) == 0 or
	fatal("Postswap failed!");
725 726
}

727
#
728
# Set accounting stuff, but on success only, and *after* gathering swap stats!
729
#
730
$experiment->SetSwapInfo($this_user);
731

732 733 734
#
# Close up the log file so the webpage stops.
#
735
if (!defined($template)) {
736
    print "Experiment $pid/$eid has been successfully created!\n";
737
    $experiment->CloseLogFile();
738
}
739 740 741 742

#
# Must unlock and drop batch experiments into the queue before exit.
#
743
if ($batchmode && !$frontend) {
744
    $experiment->Unlock(EXPTSTATE_QUEUED());
745 746
}
else {
747
    $experiment->Unlock();
748 749 750 751 752 753
}

#
# Clear the cancel flag now that the operation is complete. Must be
# done after we change the experiment state (above).
#
754
$experiment->SetCancelFlag(EXPTCANCEL_CLEAR());
755

756 757 758 759
#
# In template_mode we are done; the caller finishes up.
#
exit(0)
760
    if (defined($template));
761

762 763 764
#
# Dump the report file and the log file to the user via email. 
#
765
my ($expt_created) = $experiment->created();
766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
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";
784
}
785 786 787 788 789 790 791 792
$message .=
    "User:        $user_name\n" .
    "EID:         $eid\n" .
    "PID:         $pid\n" .
    "GID:         $gid\n" .
    "Description: $description\n" .
    "Swappable:   " . ($swappable ? "Yes\n" :
		                    "No  (Reason: $noswap_reason)\n") .
793
    "Idle-Swap:   " . ($idleswap  ? "Yes, at " . $idleswaptime/60.0 . " hours\n":
794
		                    "No  (Reason: $noidleswap_reason)\n") .
795
    "Auto-Swap:   " . ($autoswap  ? "Yes, at " . $autoswaptime/60.0 . " hours\n":
796 797 798 799 800 801 802 803 804 805 806 807 808 809
		                    "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: $TBLOGS",
810 811
	 ($repfile, $logname, $nsfile))
    if (!$zeemode);
812 813

# Done!
814
exit(0);
815

816 817 818
#
#
#
819
sub cleanup()
820
{
821
    #
822 823 824 825 826 827
    # 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 (!$committed) {
828 829 830
	# Completely remove all trace of the archive.
        libArchive::TBDestroyExperimentArchive($pid, $eid);

831
	#
832
	# Clear the experiment record and cleanup directories
833
	#
834 835
	$experiment->Delete(1)
	    if (defined($experiment));
836

837
	return;
838
    }
839

840 841 842 843
    #
    # Gather statistics.
    #
    if ($frontend) {
844
	$experiment->SwapFail($this_user, TBDB_STATS_PRELOAD, $errorstat);
845 846
    }
    else {
847
	$experiment->SwapFail($this_user, TBDB_STATS_START, $errorstat);
848
    }
849

850 851 852
    #
    # Must clean up the experiment if it made it our of NEW state.
    #
853
    my $estate = $experiment->state();
854 855 856 857 858 859 860 861
    if ($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 && $signaled)) {
862 863
	    if ($experiment->Swap("out", "-force") != 0) {
		print "tbswap out -force failed!\n";
864
	    }
865
	    $experiment->SetState(EXPTSTATE_SWAPPED);
866
	}
867
	
868
	if ($experiment->End("-f") != 0) {
869
	    print "tbend failed!\n";
870 871
	}
    }
872
    $experiment->SetState(EXPTSTATE_TERMINATED);
873

874 875
    # Old swap gathering stuff.
    $experiment->GatherSwapStats($this_user, TBDB_STATS_TERMINATE, 0);
876

877 878
    # Clear the logfile so the webpage stops.
    $experiment->CloseLogFile();
879

880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898
    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));
    } 	
	
899
    #
900 901 902 903
    # Back up the work dir for post-mortem debugging. 
    #
    system("/bin/rm -rf  ${workdir}-failed");
    system("/bin/mv -f   $workdir ${workdir}-failed");
904

905 906
    #
    # Clear the record and cleanup.
907 908
    #
    $experiment->Delete();
909 910 911
}

#
912 913
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
914
#
915
sub ParseArgs()
916
{
917 918 919 920
    my %options = ();
    if (! getopts($optlist, \%options)) {
	usage();
    }
921

922
    if (@ARGV > 1) {
923 924
	usage();
    }
925 926
    if (@ARGV == 1) {
	$tempnsfile = $ARGV[0];
927

928
	# Note different taint check (allow /).
929
	if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
930 931 932
	    $tempnsfile = $1;
	}
	else {
933
	    tbdie("Bad data in nsfile: $tempnsfile");
934
	}
935

936
	#
937 938
	# Called from ops interactively. Make sure NS file resides in an
	# appropriate location.
939 940 941 942 943 944 945 946
	#
	# Use realpath to resolve any symlinks.
	#
	my $translated = `realpath $tempnsfile`;
	if ($translated =~ /^([-\w\.\/]+)$/) {
	    $tempnsfile = $1;
	}
	else {
947 948 949
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'realpath', $translated]},
		  "Bad data returned by realpath: $translated");
950 951 952
	}

	#
953 954 955 956 957 958 959
	# 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.
960 961
	#
	if (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
962
	    ! ($tempnsfile =~ /^\/tmp\/\d+\.ns/) &&
963
	    ! ($tempnsfile =~ /^\/var\/tmp\/php[-\w]+/) &&
964
	    ! TBValidUserDir($tempnsfile, 0)) {
965 966 967
	    tberror({type => 'primary', severity => SEV_ERROR,
		     error => ['disallowed_directory', $tempnsfile]},
		    "$tempnsfile does not resolve to an allowed directory!");
968 969 970 971 972
	    # Note positive status; so error goes to user not tbops.
	    exit(1);
	}
    }
    
973
    if (defined($options{"i"})) {
974
	$batchmode = 0;
975
    }
976 977 978
    if (defined($options{"f"})) {
	$frontend = 1;
    }
979 980 981
    if (defined($options{"q"})) {
	$quiet = 1;
    }
982 983 984 985
    if (defined($options{"z"})) {
	$zeemode = 1;
	$zeeopt  = "-p";
    }
986 987 988 989
    # This option should not be exported via the XMLRPC server. 
    if (defined($options{"n"})) {
	$idleignore = 1;
    }
990

991 992 993
    if (defined($options{"s"})) {
	$savestate = 1;
    }
994 995 996 997 998 999 1000 1001 1002 1003 1004
    if (defined($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"})) {
	$nonsfile = 1;
    }
1005 1006 1007
    if (defined($options{"k"})) {
	$lockdown = 1;
    }
1008

1009 1010 1011 1012 1013
    #
    # Clone an experiment, either an existing experiment or an old one
    # (using the archive). 
    #
    if (defined($options{"c"})) {
1014
	$copyarg = $options{"c"};
1015

1016 1017
	if (! (($copyarg =~ /^([-\w]+),([-\w]+)(?::[-\w]*)?$/) ||
	       ($copyarg =~ /^(\d+)(?::[-\w]*)?$/))) {
1018 1019 1020
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $copyarg]},
		  "Bad data in argument: $copyarg");
1021
	}
1022 1023
    }

1024 1025 1026
    #
    # pid,eid,gid get passed along as shell commands args; must taint check.
    # 
1027 1028
    if (defined($options{"p"})) {
	$pid = $options{"p"};
1029

1030
	if ($pid =~ /^([-\w]+)$/) {
1031 1032
	    $pid = $1;
	}
1033
	else {
1034 1035 1036
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $pid]},
		  "Bad data in argument: $pid.");
Leigh Stoller's avatar
Leigh Stoller committed
1037
	}
1038 1039 1040 1041
    }
    if (defined($options{"e"})) {
	$eid = $options{"e"};

1042
	if ($eid =~ /^([-\w]+)$/) {
1043 1044 1045
	    $eid = $1;
	}
	else {
1046 1047 1048
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $eid]},
		  "Bad data in argument: $eid.");
1049
	}
1050 1051
	if (! TBcheck_dbslot($eid, "experiments", "eid",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1052 1053 1054
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'eid', $eid]},
		  "Improper experiment name (id)!");
1055
	}
1056 1057 1058 1059
    }
    if (defined($options{"g"})) {
	$gid = $options{"g"};

1060
	if ($gid =~ /^([-\w]+)$/) {
1061
	    $gid = $1;
1062
	}
1063
	else {
1064 1065 1066
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $gid]},
		  "Bad data in argument: $gid.");
1067 1068
	}
    }
1069
    if (defined($options{"E"})) {
1070 1071
	if (! TBcheck_dbslot($options{"E"}, "experiments", "expt_name",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1072
	    tbdie("Improper experiment description!");
1073
	}
1074
	$description = $options{"E"};
1075
    }
1076 1077 1078
    if (defined($options{"S"})) {
	if (! TBcheck_dbslot($options{"S"}, "experiments", "noswap_reason",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1079
	    tbdie("Improper noswap reason!");
1080 1081
	}
	$swappable     = 0;
1082
	$noswap_reason = $options{"S"};
1083 1084 1085 1086
    }
    if (defined($options{"L"})) {
	if (! TBcheck_dbslot($options{"L"}, "experiments", "noidleswap_reason",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1087
	    tbdie("Improper noidleswap reason!");
1088 1089
	}
	$idleswap          = 0;
1090
	$noidleswap_reason = $options{"L"};
1091
    }
1092
    if (defined($options{"l"})) {
1093 1094
	if (! TBcheck_dbslot($options{"l"}, "experiments", "idleswap_timeout",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1095
	    tbdie("Improper idleswap timeout!");
1096 1097
	}
        $idleswap     = 1;
1098 1099 1100
	$idleswaptime = $options{"l"};
    }
    if (defined($options{"a"})) {
1101 1102
	if (! TBcheck_dbslot($options{"a"}, "experiments", "autoswap_timeout",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1103
	    tbdie("Improper autoswap timeout!");
1104 1105
	}
        $autoswap     = 1;
1106 1107
	$autoswaptime = $options{"a"};
    }
1108 1109 1110
    if (defined($options{"t"})) {
	if (! TBcheck_dbslot($options{"t"}, "experiments", "linktest_level",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1111
	    tbdie("Improper linktest level!");
1112 1113 1114
	}
	$linktest = $options{"t"};
    }
1115 1116
    if (defined($options{"w"})) {
	$waitmode = 1;
1117
    }
1118
    if (defined($options{"x"})) {
1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136
	my $guid;
	my $vers;
	my $branch_guid;
	my $branch_vers;
	
	if ($options{"x"} =~ /^(\d*)\/(\d*)$/) {
	    $guid = $1;
	    $vers = $2;
	}
	elsif ($options{"x"} =~ /^(\d*)\/(\d*),(\d*)\/(\d*)$/) {
	    $guid = $1;
	    $vers = $2;
	    $branch_guid = $3;
	    $branch_vers = $4;
	}
	else {
	    tbdie("Bad arguments for -x option");
	}
1137

1138 1139 1140 1141 1142 1143 1144 1145 1146
	$template = Template->Lookup($guid, $vers);
	if (!defined($template)) {
	    tbdie("No such template $guid/$vers");
	}

	if (defined($branch_guid)) {
	    $branch_template = Template->Lookup($branch_guid, $branch_vers);
	    if (!defined($branch_template)) {
		tbdie("No such template $branch_guid/$branch_vers");
1147 1148
	    }
	}
1149 1150
	
	if (defined($options{"y"})) {
1151
	    my $instance_idx = $options{"y"};
1152 1153 1154 1155 1156
	    
	    if ($instance_idx =~ /^([\d]+)$/) {
		$instance_idx = $1;
	    }
	    else {
1157 1158 1159
		tbdie({type => 'primary', severity => SEV_ERROR,
		       error => ['bad_data', 'argument', $instance_idx]},
		      "Bad data in argument: $instance_idx.");
1160
	    }
1161
	    $instance = Template::Instance->LookupByID($instance_idx);
1162 1163 1164
	    if (!defined($instance)) {
		tbdie("No such template instance $instance_idx");
	    }
1165
	}
1166
    }
1167
}
1168

1169
#
1170
# Check the -c argument, and set some global variables at the same time.
1171
#
1172
sub CheckCopyArgs()
1173
{
1174 1175 1176 1177
    return 0
	if (!defined($copyarg));
    
    if ($copyarg =~ /^([-\w]+),([-\w]+)(?::([-\w]*))?$/) {
1178 1179 1180
	#
	# pid,eid of an existing experiment.
	#
1181 1182 1183 1184
	$copypid  = $1;
	$copyeid  = $2;
	$copytag  = (defined($3) ? (($3 eq "") ? undef : $3) : undef);
	$copyfrom = "exp";
1185

1186 1187 1188 1189
	$copy_experiment = Experiment->Lookup($copypid, $copyeid);

	tbdie("Could not get experiment index for $copypid/$copyeid")
	    if (!defined($copy_experiment));
1190 1191 1192 1193 1194 1195

	if (! $copy_experiment->AccessCheck($this_user, TB_EXPT_READINFO)) {
	    tberror("You do not have permission to copy $copy_experiment");
	    exit(1);
	}
	
1196 1197 1198 1199 1200
	#
	# If given a tag, must use the archive. To do that, we need the
	# experiment index of the experiment we are copying from.
	#
	if (defined($copytag)) {
1201
	    $copyidx  = $copy_experiment->idx();
1202
	    $copyfrom = "archive";
1203 1204
	}
    }
1205 1206 1207
    elsif ($copyarg =~ /^(\d+)(?::([-\w]*))?$/) {
	$copyidx  = $1;
	$copytag  = (defined($2) ? (($2 eq "") ? undef : $2) : undef);
1208

1209 1210 1211
	$copy_experiment = Experiment->LookupByIndex($copyidx);

	#
1212
	# Not a current experiment?
1213 1214 1215
	#
	if (!defined($copy_experiment)) {
	    my $query_result =
1216
		DBQueryFatal("select pid_idx from experiment_stats ".
1217
			     "where exptidx='$copyidx'");
1218

1219 1220 1221
	    if (! $query_result->numrows) {
		tbdie("No such experiment index: $copyidx");
	    }
1222 1223 1224 1225 1226 1227 1228 1229 1230
	    my ($copy_pid_idx) = $query_result->fetchrow_array();
	    my $copy_project = Project->Lookup($copy_pid_idx);
	    
	    if (! $copy_project->AccessCheck($this_user,
					     TB_PROJECT_READINFO)) {
		tberror("You do not have permission to copy $copy_experiment");
		exit(1);
	    }
	    $copyfrom = "archive";
1231 1232
	}
	else {
1233 1234 1235 1236 1237 1238 1239 1240 1241
	    $copypid  = $copy_experiment->pid();
	    $copyeid  = $copy_experiment->eid();
	    $copyfrom = "exp";

	    if (! $copy_experiment->AccessCheck($this_user,
						TB_EXPT_READINFO)) {
		tberror("You do not have permission to copy $copy_experiment");
		exit(1);
	    }
1242 1243 1244
	}
    }
    else {
1245
	tbdie("Bad data in -c option: $copyarg");
1246 1247 1248 1249
    }
}

#
1250
# Copy in what we need from another experiment (or archive).
1251
#
1252
sub CopyInArchive()
1253
{
1254 1255
    return 0
	if (!defined($copyarg));
1256

1257
    $tempnsfile = "/tmp/$$.ns";
1258

1259 1260
    if ($copyfrom eq "exp