batchexp.in 36.2 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2007 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:";	# 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 $savestate= 0;
70 71
my $template_mode  = 0;	# New stuff; experiment templates.
my $instance_idx   = 0;	# New stuff; experiment templates.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
72
my $archive_eid;	# New stuff; experiment templates.
73 74 75 76 77 78 79 80 81 82
# 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 $copybranch = 0;	# A branch instead of a duplicate
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.
83 84 85 86 87

#
# Configure variables
#
my $TB       = "@prefix@";
88
my $EVENTSYS = @EVENTSYS@;
89 90
my $TBOPS    = "@TBOPSEMAIL@";
my $TBLOGS   = "@TBLOGSEMAIL@";
91 92
my $TBDOCBASE = "@TBDOCBASE@";
my $TBBASE   = "@TBBASE@";
93
my $CONTROL  = "@USERNODE@";
94

95 96 97 98 99 100
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
Kevin Atkinson's avatar
Kevin Atkinson committed
101
use libtblog;
102
use libArchive;
103
use Experiment;
104
use Template;
105
use User;
106

107 108 109 110
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
111
my $errorstat=-1;
112
my $exptidx;
113
my $logname;
114

115 116 117 118 119 120 121
# 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;
122 123 124 125 126 127

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

142 143 144 145 146
my $eid;
my $pid;
my $gid;
my $description;
my $tempnsfile;
147 148 149
my $swappable    = 1;
my $noswap_reason;
my $idleswap     = 1;
150
my $idleswaptime = 60 * TBGetSiteVar("idle/threshold");
151
my $noidleswap_reason;
152
my $autoswap     = 0;
153
my $autoswaptime = 10 * 60;
154 155 156
my $idleignore   = 0;
my $priority     = TB_EXPTPRIORITY_LOW;
my $exptstate    = EXPTSTATE_NEW();
157
my $batchstate   = BATCHSTATE_UNLOCKED();
158 159
my $now          = localtime();
my $committed    = 0;
160 161
my $experiment;
my $copy_experiment;
162

163
#
164
# Verify user and get his DB uid and other info for later.
165
#
166 167 168
my $this_user = User->ThisUser();
if (! defined($this_user)) {
    tbdie("You ($UID) do not exist!");
169
}
170 171 172 173
my $user_dbid  = $this_user->dbid();
my $user_uid   = $this_user->uid();
my $user_name  = $this_user->name();
my $user_email = $this_user->email();
174

175 176 177 178
#
# Before doing anything else, check for overquota ... lets not waste
# our time. Make sure user sees the error by exiting with 1.
#
179
if (system("$checkquota $user_uid") != 0) {
180 181
    tberror({cause => 'user', type => 'primary', severity => SEV_ERROR,
	     error => ['over_disk_quota', $CONTROL]},
182 183
	    "You are over your disk quota on $CONTROL; ".
	    "please login there and cleanup!");
184 185 186
    exit(1);
}

187
#
188
# Parse command arguments.
189
#
190
ParseArgs();
191
CheckCopyArgs();
192 193 194

#
# Sanity check them.
195
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
196 197 198
usage()
    if (!defined($pid) || !defined($eid));
usage()
199
    if (defined($tempnsfile) && defined($copyarg));
Leigh B. Stoller's avatar
Leigh B. Stoller committed
200

201 202 203
if (!defined($gid)) {
    $gid = $pid;
}
204
if (!defined($description)) {
205
    $description = "Created by $user_uid";
206
}
207
if (! $swappable && (!defined($noswap_reason) || $noswap_reason eq "")) {
208
    tbdie("Must provide a reason with -S option (not swappable reason)!");
209
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
210
if (! $idleswap && (!defined($noidleswap_reason) || $noidleswap_reason eq "")){
211
    tbdie("Must provide a reason with -L option (no idleswap reason)!");
212
}
213
if (!defined($tempnsfile) && !defined($copyarg) && !TBAdmin()) {
214
    tbdie("Only admins can create experiments with no NS file");
215
}
216
my $nsfile  = "$eid.ns";
217
my $repfile = "report";
218

219
# Defaults for the DB and for the email message. 
220
$noswap_reason = "None Given"
221
    if (!defined($noswap_reason));
222
$noidleswap_reason = "None Given"
223 224
    if (!defined($noidleswap_reason));

225
#
226
# Make sure UID is allowed to create experiments in this project.
227
#
228
if (! TBProjAccessCheck($user_uid, $pid, $gid, TB_PROJECT_CREATEEXPT)) {
229
    die("You do not have permission to create experiments in $pid/$gid");
230 231
}

232 233 234
if (defined($copyarg)) {
    # This will be handled below.
    ;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
235 236 237
}
elsif (!defined($tempnsfile)) {
    # If no NS file, then override swap/idle stuff. 
238 239 240
    $swappable     = 0;
    $idleswap      = 0;
}
241
elsif (! -f $tempnsfile || ! -r $tempnsfile || -z $tempnsfile) {
242
    # Exit so that user sees the error, not us.
243 244 245
    tberror({type => 'primary', severity => SEV_ERROR,
	     error => ['bogus_ns_file', $tempnsfile]},
	    "$tempnsfile does not exist or is not a readable file!");
246 247
    exit(1);
}
248

249 250 251 252
#
# Batch jobs get a shorter idle time
#
my $swaptime = $idleswaptime;
253
if ($batchmode && TBSiteVarExists("idle/batch_threshold")) {
254 255 256 257 258 259
    my $batchidleswaptime = TBGetSiteVar("idle/batch_threshold");
    if ($swaptime > $batchidleswaptime) {
	$swaptime = $batchidleswaptime;
    }
}

260
#
261 262
# Grab me a secret key for accessing tar/rpm files via the web interface.
# Grab another secret key for the event system HMACs.
263
#
264 265
my $webkey   = TBGenSecretKey();
my $eventkey = TBGenSecretKey();
266

267 268 269 270 271 272 273 274 275
#
# In wait mode, block SIGINT until we spin off the background process.
#
if ($waitmode) {
    $SIG{QUIT} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

276
#
277 278 279 280 281
# Create an arg array of parameters.
#
my %args = ();

$args{'gid'}	  	   = $gid;
282 283 284 285
$args{'expt_head_uid'}	   = $user_uid;
$args{'expt_swap_uid'}	   = $user_uid;
$args{'creator_idx'}	   = $user_dbid;
$args{'swapper_idx'}	   = $user_dbid;
286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307
$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;
$args{'instance_idx'}	   = $instance_idx;
# These are special; the library will DBQuote them. 
$args{'expt_name'}	   = $description;
$args{'noswap_reason'}	   = $noswap_reason;
$args{'noidleswap_reason'} = $noidleswap_reason;

# Now create the experiment; we get back a perl class instance.
if (! ($experiment = Experiment->Create($pid, $eid, \%args))) {
308 309 310
    tbdie({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['create_experiment_record_failed']},
	  "Could not create a new experiment record!");
311 312
}

313 314 315 316 317 318
#
# 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
319 320 321 322 323
#
# Set error reporting info
# 
tblog_set_info($pid,$eid,$UID);

324
#
325
# Create a directory structure for the experiment.
326 327 328 329 330 331 332 333 334
# There is no need to do this for the template wrapper experiment; it was
# already done when the template was created.
#
if (!$template_mode || $instance_idx) {
    if ($experiment->CreateDirectory() != 0) {
	if (($? >> 8) == EDQUOT()) {
	    # Obey exit status protocol for web page; User should see this.
	    $errorstat = 1;
	}
335 336 337
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['create_experiment_directory_failed']},
	      "Failed to created experiment directory");
338
    }
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
}
else {
    #
    # But we do need to update the experiment record with the template path.
    #
    my $template = Template->LookupByPidEid($pid, $eid);

    fatal("Could not find template record for $pid/$eid")
	if (!defined($template));

    my %args = ();
    $args{'path'} = $template->path();
    
    $experiment->Update(\%args) == 0
	or fatal("Could not update experiment record!");
354 355
}

356 357 358 359
#
# Grab the working directory path, and thats where we work.
# The user's experiment directory is off in /proj space.
#
360 361
my $workdir = $experiment->WorkDir();
my $userdir = $experiment->UserDir();
362 363 364 365

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

366 367 368 369
#
# Create a new archive, which might actually be a branch of an existing one
# when doing an experiment branch (fork).
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
370
if ($template_mode && defined($archive_eid)) {
371 372 373
    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
374 375 376 377 378
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$pid, $archive_eid, undef)
	    < 0);
}
elsif ($copybranch) {
379
    # Currently, support branching from existing experiment only.
380 381 382
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['archive_op_failed', 'create', undef, undef]},
	  "Could not create experiment archive!")
383 384 385 386 387
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$copypid,
						$copyeid, $copytag) < 0);
}
elsif (libArchive::TBCreateExperimentArchive($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
    $experiment->Unlock(EXPTSTATE_NEW());
436 437
    exit(0);
}
438

439
#
440
# Now we can get the NS file!
441
#
442
if (system("/bin/cp", "$tempnsfile", "$nsfile")) {
443 444 445
    fatal({type => 'primary', severity => SEV_ERROR,
	   error => ['copy_ns_file_failed', $tempnsfile, $nsfile]},
	  "Could not copy $tempnsfile to $workdir/$nsfile");
446
}
447
chmod(0664, "$nsfile");
448

449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464
# 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");
}

# 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);
}

465
#
466 467
# Run parse in impotent mode on the NS file.  This has no effect but
# will display any errors.
468
#
469
if (system("$parser -n $zeeopt $pid $gid $eid $nsfile") != 0) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
470 471
    # Obey exit status protocol for web page.
    $errorstat = 1;
472 473 474
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['ns_parse_failed']},
	  "NS Parse failed!");
475 476
}

477 478 479
#
# Gather statistics; start the clock ticking.
#
480
if ($frontend || $batchmode) {
481
    GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_PRELOAD, 0,
482 483 484
		    TBDB_STATS_FLAGS_START);
}
else {
485
    GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_START, 0,
486 487 488
		    TBDB_STATS_FLAGS_START);
}

489 490 491
goto skiplog
    if ($template_mode);

492 493 494 495 496 497
#
# 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.
#
498 499
if ($experiment->CreateLogFile("startexp", \$logname) != 0) {
    fatal("Could not create logfile!");
500
    
501 502 503 504
}
$experiment->SetLogFile($logname);
$experiment->OpenLogFile($logname);

505 506
if (my $childpid = TBBackGround($logname)) {
    #
507 508
    # Parent exits normally, unless in waitmode. We have to set
    # justexit to make sure the END block below does not run.
509
    #
510 511
    $justexit = 1;
    
512 513 514
    if (!$waitmode) {
	print("Experiment $pid/$eid is now configuring\n".
 	      "You will be notified via email when the experiment is ".
515 516
	      "ready to use\n")
	    if (! $quiet);
517 518
	exit(0);
    }
519
    print("Waiting for " . ($batchmode ? "batch " : "") . "experiment $eid ".
520 521 522 523 524 525 526
	  "to finish " . ($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");
    }
527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543

    # 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);

544 545
    print("Done. Exited with status: $?\n")
	if (! $quiet);
546 547 548 549 550 551 552 553 554

    my $exit_code = $? >> 8;
    
    if ($exit_code != 0) {
	my $d = tblog_lookup_error();
	print tblog_format_error($d);
    }
    
    exit $exit_code;
555
}
556
TBdbfork();
557
skiplog:
558 559 560 561 562 563 564 565 566 567 568 569

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

#
# 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();
}

570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589
#
# 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';

590 591
#
# The guts of starting an experiment!
592
#
593 594
# A batch experiment is essentially preloaded (frontend mode) and then
# dropped into the batch queue, unless the user requested only preload.
595 596
#

597
#
598
# Run the various scripts. We want to propagate the error from tbprerun
599 600
# and tbrun back out, hence the bogus looking errorstat variable.
#
601
$experiment->SetState(EXPTSTATE_PRERUN) == 0
602
    or fatal("Failed to set experiment state to " . EXPTSTATE_PRERUN());
603 604

if ($experiment->PreRun($nsfile, ($zeeopt ? "-z" : ""))) {
605
    $errorstat = $? >> 8;
606 607 608
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbprerun_failed']},
	  "tbprerun failed!");
609
}
610 611

$experiment->SetState(EXPTSTATE_SWAPPED) == 0
612
    or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
613 614 615 616

#
# If not in frontend mode (preload only) continue to swapping exp in.
# 
617
if (! ($frontend || $batchmode)) {
618
    $experiment->SetState(EXPTSTATE_ACTIVATING) == 0
619
	or fatal("Failed to set experiment state to ". EXPTSTATE_ACTIVATING());
620

621
    if ($experiment->Swap("in") != 0) {
622
	$errorstat = $? >> 8;
623 624 625
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['tbswap_in_failed']},
	      "tbswap in failed!");
626
    }
627 628
    
    $experiment->SetState(EXPTSTATE_ACTIVE) == 0
629
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
630 631 632 633 634

    #
    # Look for the unsual case of more than 2 nodes and no vlans. Send a
    # warning message.
    #
635 636 637 638
    my @localnodes = ();
    fatal("Could not get local node list for $pid/$eid")
	if ($experiment->LocalNodeList(\@localnodes));

Leigh B. Stoller's avatar
Leigh B. Stoller committed
639
    if (@localnodes && scalar(@localnodes) > 2) {
640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656
	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.
657
if ($experiment->Report($repfile, "-b") != 0) {
658 659 660
    fatal({type => 'secondary', severity => SEV_SECONDARY,
	   error => ['tbreport_failed']},
	  "tbreport failed!");
661 662
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
663 664 665 666 667
# Latest log is always called the same thing.
if (defined($logname)) {
    system("cp -fp $logname $workdir/" . EXPTLOGNAME());
}

668
#
669
# Save a copy of the files for testbed information gathering (long term).
670
#
671
$experiment->SaveLogFiles();
672 673 674 675 676

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files.
#
677
$experiment->CopyLogFiles();
678

679
# Tell the archive library to add all files to the archive. 
680 681
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
682

683
#
684 685
# Do a SavePoint on the experiment files. In template mode, let the wrapper
# deal with this. Avoids duplication of work.
686
#
687 688 689 690 691 692 693
if (! $template_mode) {
    print "Doing a savepoint on the experiment archive ...\n";
    if (libArchive::TBExperimentArchiveSavePoint($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!");
    }
694 695
}

696 697 698 699
#
# If this is a branch, then do a commit. Otherwise, the archive looks
# wrong cause its a branch from the original and shows all those files.
#
700
if ($copybranch && !$template_mode) {
701 702
    print "Doing a commit on the experiment archive ...\n";
    libArchive::TBCommitExperimentArchive($pid, $eid, "branch_merge") == 0 or
703 704 705
	fatal({type => 'secondary', severity => SEV_SECONDARY,
	       error => ['archive_op_failed', 'commit', undef, undef]},
	      "Failed to commit experiment archive!");
706 707
}

708 709 710 711
#
# Gather statistics.
#
if ($frontend || $batchmode) {
712
    GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_PRELOAD, 0);
713 714
}
else {
715
    GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_START, 0);
716 717
}

718
#
719
# Set accounting stuff, but on success only, and *after* gathering swap stats!
720
#
721
$experiment->SetSwapInfo($this_user);
722

723 724 725
#
# Close up the log file so the webpage stops.
#
726 727
if (!$template_mode) {
    print "Experiment $pid/$eid has been successfully created!\n";
728
    $experiment->CloseLogFile();
729
}
730 731 732 733

#
# Must unlock and drop batch experiments into the queue before exit.
#
734
if ($batchmode && !$frontend) {
735
    $experiment->Unlock(EXPTSTATE_QUEUED());
736 737
}
else {
738
    $experiment->Unlock();
739 740 741 742 743 744
}

#
# Clear the cancel flag now that the operation is complete. Must be
# done after we change the experiment state (above).
#
745
$experiment->SetCancelFlag(EXPTCANCEL_CLEAR());
746

747 748 749 750 751 752
#
# In template_mode we are done; the caller finishes up.
#
exit(0)
    if ($template_mode);

753 754 755
#
# Dump the report file and the log file to the user via email. 
#
756
my ($expt_created) = $experiment->created();
757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774
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";
775
}
776 777 778 779 780 781 782 783
$message .=
    "User:        $user_name\n" .
    "EID:         $eid\n" .
    "PID:         $pid\n" .
    "GID:         $gid\n" .
    "Description: $description\n" .
    "Swappable:   " . ($swappable ? "Yes\n" :
		                    "No  (Reason: $noswap_reason)\n") .
784
    "Idle-Swap:   " . ($idleswap  ? "Yes, at " . $idleswaptime/60.0 . " hours\n":
785
		                    "No  (Reason: $noidleswap_reason)\n") .
786
    "Auto-Swap:   " . ($autoswap  ? "Yes, at " . $autoswaptime/60.0 . " hours\n":
787 788 789 790 791 792 793 794 795 796 797 798 799 800
		                    "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: $TBLOGS",
801 802
	 ($repfile, $logname, $nsfile))
    if (!$zeemode);
803 804

# Done!
805
exit(0);
806

807 808 809
#
#
#
810
sub cleanup()
811
{
812
    #
813 814 815 816 817 818
    # 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 (!$committed) {
819 820 821
	# Completely remove all trace of the archive.
        libArchive::TBDestroyExperimentArchive($pid, $eid);

822
	#
823
	# Clear the experiment record and cleanup directories
824
	#
825 826
	$experiment->Delete(1)
	    if (defined($experiment));
827

828
	return;
829
    }
830

831 832 833 834
    #
    # Gather statistics.
    #
    if ($frontend) {
835
	GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_PRELOAD, $errorstat);
836 837
    }
    else {
838
	GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_START, $errorstat);
839
    }
840
    $experiment->Refresh();
841

842 843 844
    #
    # Must clean up the experiment if it made it our of NEW state.
    #
845
    my $estate = $experiment->state();
846 847 848 849 850 851 852 853
    if ($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 && $signaled)) {
854 855
	    if ($experiment->Swap("out", "-force") != 0) {
		print "tbswap out -force failed!\n";
856
	    }
857
	    $experiment->SetState(EXPTSTATE_SWAPPED);
858
	}
859
	
860
	if ($experiment->End("-force") != 0) {
861
	    print "tbend failed!\n";
862 863
	}
    }
864
    $experiment->SetState(EXPTSTATE_TERMINATED);
865

866
    #
867 868
    # Okay, we *are* going to terminate the experiment.
    # 
869
    GatherSwapStats($pid, $eid, $user_uid, TBDB_STATS_TERMINATE, 0);
870

871 872
    # Clear the logfile so the webpage stops.
    $experiment->CloseLogFile();
873

Kevin Atkinson's avatar
Kevin Atkinson committed
874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892
    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));
    } 	
	
893
    #
894 895 896 897
    # Back up the work dir for post-mortem debugging. 
    #
    system("/bin/rm -rf  ${workdir}-failed");
    system("/bin/mv -f   $workdir ${workdir}-failed");
898

899 900
    #
    # Clear the record and cleanup.
901 902
    #
    $experiment->Delete();
903 904 905
}

#
906 907
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
908
#
909
sub ParseArgs()
910
{
911 912 913 914
    my %options = ();
    if (! getopts($optlist, \%options)) {
	usage();
    }
915

916
    if (@ARGV > 1) {
917 918
	usage();
    }
919 920
    if (@ARGV == 1) {
	$tempnsfile = $ARGV[0];
921

922
	# Note different taint check (allow /).
923
	if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
924 925 926
	    $tempnsfile = $1;
	}
	else {
927
	    tbdie("Bad data in nsfile: $tempnsfile");
928
	}
929

930
	#
931 932
	# Called from ops interactively. Make sure NS file resides in an
	# appropriate location.
933 934 935 936 937 938 939 940
	#
	# Use realpath to resolve any symlinks.
	#
	my $translated = `realpath $tempnsfile`;
	if ($translated =~ /^([-\w\.\/]+)$/) {
	    $tempnsfile = $1;
	}
	else {
941 942 943
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'realpath', $translated]},
		  "Bad data returned by realpath: $translated");
944 945 946
	}

	#
947 948 949 950 951 952 953
	# 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.
954 955
	#
	if (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
956
	    ! ($tempnsfile =~ /^\/tmp\/\d+\.ns/) &&
957
	    ! ($tempnsfile =~ /^\/var\/tmp\/php\w+/) &&
958
	    ! TBValidUserDir($tempnsfile, 0)) {
959 960 961
	    tberror({type => 'primary', severity => SEV_ERROR,
		     error => ['disallowed_directory', $tempnsfile]},
		    "$tempnsfile does not resolve to an allowed directory!");
962 963 964 965 966
	    # Note positive status; so error goes to user not tbops.
	    exit(1);
	}
    }
    
967
    if (defined($options{"i"})) {
968
	$batchmode = 0;
969
    }
970 971 972
    if (defined($options{"f"})) {
	$frontend = 1;
    }
973 974 975
    if (defined($options{"q"})) {
	$quiet = 1;
    }
976 977 978 979
    if (defined($options{"z"})) {
	$zeemode = 1;
	$zeeopt  = "-p";
    }
980 981 982 983
    # This option should not be exported via the XMLRPC server. 
    if (defined($options{"n"})) {
	$idleignore = 1;
    }
984

985 986 987 988
    if (defined($options{"s"})) {
	$savestate = 1;
    }

Leigh B. Stoller's avatar
Leigh B. Stoller committed
989 990 991 992 993
    #
    # Clone an experiment, either an existing experiment or an old one
    # (using the archive). 
    #
    if (defined($options{"c"})) {
994
	$copyarg = $options{"c"};
Leigh B. Stoller's avatar
Leigh B. Stoller committed
995

996 997
	if (! (($copyarg =~ /^([-\w]+),([-\w]+)(?::[-\w]*)?$/) ||
	       ($copyarg =~ /^(\d+)(?::[-\w]*)?$/))) {
998 999 1000
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $copyarg]},
		  "Bad data in argument: $copyarg");
1001 1002 1003 1004
	}
	# This option only makes sense with -c option.
	if (defined($options{"b"})) {
	    $copybranch = 1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1005 1006 1007
	}
    }

1008 1009 1010
    #
    # pid,eid,gid get passed along as shell commands args; must taint check.
    # 
1011 1012
    if (defined($options{"p"})) {
	$pid = $options{"p"};
1013

1014
	if ($pid =~ /^([-\w]+)$/) {
1015 1016
	    $pid = $1;
	}
1017
	else {
1018 1019 1020
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $pid]},
		  "Bad data in argument: $pid.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1021
	}
1022 1023 1024 1025
    }
    if (defined($options{"e"})) {
	$eid = $options{"e"};

1026
	if ($eid =~ /^([-\w]+)$/) {
1027 1028 1029
	    $eid = $1;
	}
	else {
1030 1031 1032
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $eid]},
		  "Bad data in argument: $eid.");
1033
	}
1034 1035
	if (! TBcheck_dbslot($eid, "experiments", "eid",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1036 1037 1038
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'eid', $eid]},
		  "Improper experiment name (id)!");
1039
	}
1040 1041 1042 1043
    }
    if (defined($options{"g"})) {
	$gid = $options{"g"};

1044
	if ($gid =~ /^([-\w]+)$/) {
1045
	    $gid = $1;
1046
	}
1047
	else {
1048 1049 1050
	    tbdie({type => 'primary', severity => SEV_ERROR,
		   error => ['bad_data', 'argument', $gid]},
		  "Bad data in argument: $gid.");
1051 1052
	}
    }
1053
    if (defined($options{"E"})) {
1054 1055
	if (! TBcheck_dbslot($options{"E"}, "experiments", "expt_name",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1056
	    tbdie("Improper experiment description!");
1057
	}
1058
	$description = $options{"E"};
1059
    }
1060 1061 1062
    if (defined($options{"S"})) {
	if (! TBcheck_dbslot($options{"S"}, "experiments", "noswap_reason",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1063
	    tbdie("Improper noswap reason!");
1064 1065
	}
	$swappable     = 0;
1066
	$noswap_reason = $options{"S"};
1067 1068 1069 1070
    }
    if (defined($options{"L"})) {
	if (! TBcheck_dbslot($options{"L"}, "experiments", "noidleswap_reason",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {
1071
	    tbdie("Improper noidleswap reason!");
1072 1073
	}
	$idleswap          = 0;
1074
	$noidleswap_reason = $options{"L"};
1075
    }
1076
    if (defined($options{"l"})) {
1077 1078
	if (! TBcheck_dbslot($options{"l"}, "experiments", "idleswap_timeout",
			   TBDB_CHECKDBSLOT_WARN|TBDB_CHECKDBSLOT_ERROR)) {