batchexp.in 33 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3
#
# EMULAB-COPYRIGHT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
4
# Copyright (c) 2000-2006 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 60
sub fatal($);

61
my $optlist = "iE:g:e:p:S:L:a:l:sfwqt:nzc:bx:y:m";	# 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
my $dpdb	   = 0; # A per-experiment DB on ops.
74 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 $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.
84 85 86 87 88

#
# Configure variables
#
my $TB       = "@prefix@";
89
my $PROJROOT = "/proj";
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;
Kevin Atkinson's avatar
Kevin Atkinson committed
103
use libtblog;
104
use libArchive;
105
use Experiment;
106
use Template;
107

108 109 110 111
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
112
my $errorstat=-1;
113 114
my $user_name;
my $user_email;
115
my $dbuid;
116
my $exptidx;
117
my $logname;
118

119 120 121 122 123 124 125
# 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;
126 127 128 129 130 131

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

132 133 134 135 136 137 138
#
# 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);

139 140
#
# Untaint the path
141
#
142 143
# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
144 145
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

146 147 148 149 150
my $eid;
my $pid;
my $gid;
my $description;
my $tempnsfile;
151 152 153
my $swappable    = 1;
my $noswap_reason;
my $idleswap     = 1;
154
my $idleswaptime = 60 * TBGetSiteVar("idle/threshold");
155
my $noidleswap_reason;
156
my $autoswap     = 0;
157
my $autoswaptime = 10 * 60;
158 159 160
my $idleignore   = 0;
my $priority     = TB_EXPTPRIORITY_LOW;
my $exptstate    = EXPTSTATE_NEW();
161
my $batchstate   = BATCHSTATE_UNLOCKED();
162 163
my $now          = localtime();
my $committed    = 0;
164 165
my $experiment;
my $copy_experiment;
166

167
#
168
# Verify user and get his DB uid.
169
#
170
if (! UNIX2DBUID($UID, \$dbuid)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
171
    tbdie("You do not exist in the Emulab Database!");
172 173
}

174 175 176 177
#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
178
    tbdie("Cannot determine your name and email address.");
179 180
}

181 182 183 184 185
#
# Before doing anything else, check for overquota ... lets not waste
# our time. Make sure user sees the error by exiting with 1.
#
if (system("$checkquota $dbuid") != 0) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
186 187
    tberror({cause => 'user'},
	    "You are over your disk quota on $CONTROL; please cleanup!");
188 189 190
    exit(1);
}

191
#
192
# Parse command arguments.
193
#
194
ParseArgs();
195
CheckCopyArgs();
196 197 198

#
# Sanity check them.
199
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
200 201 202
usage()
    if (!defined($pid) || !defined($eid));
usage()
203
    if (defined($tempnsfile) && defined($copyarg));
Leigh B. Stoller's avatar
Leigh B. Stoller committed
204

205 206 207
if (!defined($gid)) {
    $gid = $pid;
}
208
if (!defined($description)) {
209
    $description = "Created by $dbuid";
210
}
211
if (! $swappable && (!defined($noswap_reason) || $noswap_reason eq "")) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
212
    tbdie("Must provide a reason with -S option (not swappable reason)!");
213
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
214
if (! $idleswap && (!defined($noidleswap_reason) || $noidleswap_reason eq "")){
Kevin Atkinson's avatar
 
Kevin Atkinson committed
215
    tbdie("Must provide a reason with -L option (no idleswap reason)!");
216
}
217
if (!defined($tempnsfile) && !defined($copyarg) && !TBAdmin($dbuid)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
218
    tbdie("Only admins can create experiments with no NS file");
219
}
220 221
my $nsfile  = "$eid.ns";
my $repfile = "$eid.report";
222

223
# Defaults for the DB and for the email message. 
224
$noswap_reason = "None Given"
225
    if (!defined($noswap_reason));
226
$noidleswap_reason = "None Given"
227 228
    if (!defined($noidleswap_reason));

229
#
230
# Make sure UID is allowed to create experiments in this project.
231
#
232
if (! TBProjAccessCheck($dbuid, $pid, $gid, TB_PROJECT_CREATEEXPT)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
233
    die("You do not have permission to create experiments in $pid/$gid");
234 235
}

236 237 238
if (defined($copyarg)) {
    # This will be handled below.
    ;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
239 240 241
}
elsif (!defined($tempnsfile)) {
    # If no NS file, then override swap/idle stuff. 
242 243 244
    $swappable     = 0;
    $idleswap      = 0;
}
245
elsif (! -f $tempnsfile || ! -r $tempnsfile || -z $tempnsfile) {
246
    # Exit so that user sees the error, not us.
Kevin Atkinson's avatar
 
Kevin Atkinson committed
247
    tberror("$tempnsfile does not exist or is not a readable file!");
248 249
    exit(1);
}
250

251 252 253 254
#
# Batch jobs get a shorter idle time
#
my $swaptime = $idleswaptime;
255
if ($batchmode && TBSiteVarExists("idle/batch_threshold")) {
256 257 258 259 260 261
    my $batchidleswaptime = TBGetSiteVar("idle/batch_threshold");
    if ($swaptime > $batchidleswaptime) {
	$swaptime = $batchidleswaptime;
    }
}

262
#
263 264
# Grab me a secret key for accessing tar/rpm files via the web interface.
# Grab another secret key for the event system HMACs.
265
#
266 267
my $webkey   = TBGenSecretKey();
my $eventkey = TBGenSecretKey();
268

269 270 271 272 273 274 275 276 277
#
# In wait mode, block SIGINT until we spin off the background process.
#
if ($waitmode) {
    $SIG{QUIT} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

278
#
279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
# Create an arg array of parameters.
#
my %args = ();

$args{'gid'}	  	   = $gid;
$args{'expt_head_uid'}	   = $dbuid;
$args{'expt_swap_uid'}	   = $dbuid;
$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;
$args{'dpdb'}		   = $dpdb;
# 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))) {
    tbdie("Could not create a new experiment record!");
310 311
}

312 313 314 315 316 317
#
# 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
318 319 320 321 322
#
# Set error reporting info
# 
tblog_set_info($pid,$eid,$UID);

323
#
324
# Create a directory structure for the experiment.
325 326 327 328 329 330 331 332 333 334
# 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;
	}
	fatal("Failed to created experiment directory");
335
    }
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
}
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!");
351 352
}

353 354 355 356
#
# Grab the working directory path, and thats where we work.
# The user's experiment directory is off in /proj space.
#
357 358
my $workdir = $experiment->WorkDir();
my $userdir = $experiment->UserDir();
359 360 361 362

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

363 364 365 366
#
# 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
367 368 369 370 371 372 373
if ($template_mode && defined($archive_eid)) {
    fatal("Could not create experiment archive!")
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$pid, $archive_eid, undef)
	    < 0);
}
elsif ($copybranch) {
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393
    # Currently, support branching from existing experiment only.
    fatal("Could not create experiment archive!")
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$copypid,
						$copyeid, $copytag) < 0);
}
elsif (libArchive::TBCreateExperimentArchive($pid, $eid) < 0) {
    fatal("Could not create experiment archive!");
}

#
# 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();
}

394 395 396 397 398 399 400 401 402 403
#
# 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).
# 

404 405 406 407
#
# Dump the eventkey into a file in the experiment directory. 
#
if ($EVENTSYS) {
408
    open(KEY, ">" . $experiment->EventKeyPath()) or
409 410 411 412 413
	fatal("Could not create eventkey file: $!");
    print KEY $eventkey;
    close(KEY);
}

414
# And dump the web key too.
415
open(KEY, ">" . $experiment->WebKeyPath()) or
416 417 418 419
    fatal("Could not create webkey file: $!");
print KEY $webkey;
close(KEY);

420
#
421 422 423
# 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). 
424
#
425
if (!defined($tempnsfile)) {
426
    $experiment->Unlock(EXPTSTATE_NEW());
427 428
    exit(0);
}
429

430
#
431
# Now we can get the NS file!
432
#
433
if (system("/bin/cp", "$tempnsfile", "$nsfile")) {
434
    fatal("Could not copy $tempnsfile to $workdir/$nsfile");
435
}
436
chmod(0664, "$nsfile");
437

438
#
439 440
# Run parse in impotent mode on the NS file.  This has no effect but
# will display any errors.
441
#
442
if (system("$parser -n $zeeopt $pid $gid $eid $nsfile") != 0) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
443 444
    # Obey exit status protocol for web page.
    $errorstat = 1;
445
    fatal("NS Parse failed!");
446 447
}

448 449 450
#
# Gather statistics; start the clock ticking.
#
451
if ($frontend || $batchmode) {
452 453 454 455 456 457 458 459
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, 0,
		    TBDB_STATS_FLAGS_START);
}
else {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, 0,
		    TBDB_STATS_FLAGS_START);
}

460 461 462
goto skiplog
    if ($template_mode);

463 464 465 466 467 468
#
# 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.
#
469 470
if ($experiment->CreateLogFile("startexp", \$logname) != 0) {
    fatal("Could not create logfile!");
471
    
472 473 474 475
}
$experiment->SetLogFile($logname);
$experiment->OpenLogFile($logname);

476 477
if (my $childpid = TBBackGround($logname)) {
    #
478 479
    # Parent exits normally, unless in waitmode. We have to set
    # justexit to make sure the END block below does not run.
480
    #
481 482
    $justexit = 1;
    
483 484 485
    if (!$waitmode) {
	print("Experiment $pid/$eid is now configuring\n".
 	      "You will be notified via email when the experiment is ".
486 487
	      "ready to use\n")
	    if (! $quiet);
488 489
	exit(0);
    }
490
    print("Waiting for " . ($batchmode ? "batch " : "") . "experiment $eid ".
491 492 493 494 495 496 497
	  "to finish " . ($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");
    }
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514

    # 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);

515 516
    print("Done. Exited with status: $?\n")
	if (! $quiet);
Kevin Atkinson's avatar
 
Kevin Atkinson committed
517 518 519 520 521 522 523 524 525

    my $exit_code = $? >> 8;
    
    if ($exit_code != 0) {
	my $d = tblog_lookup_error();
	print tblog_format_error($d);
    }
    
    exit $exit_code;
526
}
527
TBdbfork();
528
skiplog:
529 530 531 532 533 534 535 536 537 538 539 540

# 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();
}

541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560
#
# 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';

561 562
#
# The guts of starting an experiment!
563
#
564 565
# A batch experiment is essentially preloaded (frontend mode) and then
# dropped into the batch queue, unless the user requested only preload.
566 567
#

568
#
569
# Run the various scripts. We want to propagate the error from tbprerun
570 571
# and tbrun back out, hence the bogus looking errorstat variable.
#
572
$experiment->SetState(EXPTSTATE_PRERUN) == 0
573
    or fatal("Failed to set experiment state to " . EXPTSTATE_PRERUN());
574 575

if ($experiment->PreRun($nsfile, ($zeeopt ? "-z" : ""))) {
576
    $errorstat = $? >> 8;
577
    fatal("tbprerun failed!");
578
}
579 580

$experiment->SetState(EXPTSTATE_SWAPPED) == 0
581
    or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
582 583 584 585

#
# If not in frontend mode (preload only) continue to swapping exp in.
# 
586
if (! ($frontend || $batchmode)) {
587
    $experiment->SetState(EXPTSTATE_ACTIVATING) == 0
588
	or fatal("Failed to set experiment state to ". EXPTSTATE_ACTIVATING());
589

590
    if ($experiment->Swap("in") != 0) {
591
	$errorstat = $? >> 8;
592
	fatal("tbswap in failed!");
593
    }
594 595
    
    $experiment->SetState(EXPTSTATE_ACTIVE) == 0
596
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
597 598 599 600 601

    #
    # Look for the unsual case of more than 2 nodes and no vlans. Send a
    # warning message.
    #
602 603 604 605
    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
606
    if (@localnodes && scalar(@localnodes) > 2) {
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
	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.
624
if ($experiment->Report($repfile, "-b") != 0) {
625
    fatal("tbreport failed!");
626 627
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
628 629 630 631 632
# Latest log is always called the same thing.
if (defined($logname)) {
    system("cp -fp $logname $workdir/" . EXPTLOGNAME());
}

633
#
634
# Save a copy of the files for testbed information gathering (long term).
635
#
636
$experiment->SaveLogFiles();
637 638 639 640 641

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files.
#
642
$experiment->CopyLogFiles();
643

644 645 646 647 648 649
#
# Copy out some files to the archive directory so that they get saved later.
#
my $archivedir = libArchive::TBUserFileArchiveDirectory($pid, $eid);

if (! -e "$archivedir/tbdata") {
650
    mkdir("$archivedir/tbdata", 0775)
651 652 653
	or fatal("Failed to mkdir $archivedir/tbdata");
}
if (! -e "$archivedir/nsdata") {
654
    mkdir("$archivedir/nsdata", 0775)
655 656 657
	or fatal("Failed to mkdir $archivedir/nsdata");
}

658
system("cp -fp $workdir/$nsfile $archivedir/nsdata/nsfile.ns") == 0
659 660
    or fatal("Failed to copy nsfile to $archivedir/tbdata");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
661
if (defined($logname)) {
662
    system("cp -fp $logname $archivedir/tbdata/log.batchexp") == 0
663
    or fatal("Failed to copy logfile to $archivedir/tbdata");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
664
}
665

666
system("cp -fp $workdir/$repfile $archivedir/tbdata/tbreport.batchexp") == 0
667 668 669 670 671
    or fatal("Failed to copy nsfile to $archivedir/tbdata");

# And tell the archive library about the above files.
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
672

673 674 675
#
# Do a SavePoint on the experiment files.
#
676
print "Doing a savepoint on the experiment archive ...\n";
677 678 679 680
if (libArchive::TBExperimentArchiveSavePoint($pid, $eid, "startexp") < 0) {
    fatal("Failed to do a savepoint on the experiment archive!");
}

681 682 683 684 685 686 687 688 689 690
#
# 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.
#
if ($copybranch) {
    print "Doing a commit on the experiment archive ...\n";
    libArchive::TBCommitExperimentArchive($pid, $eid, "branch_merge") == 0 or
	fatal("Failed to commit experiment archive!");
}

691 692 693 694 695 696 697 698 699 700
#
# Gather statistics.
#
if ($frontend || $batchmode) {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, 0);
}
else {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, 0);
}

701
#
702
# Set accounting stuff, but on success only, and *after* gathering swap stats!
703
#
704
$experiment->SetSwapInfo($dbuid);
705

706 707 708
#
# Close up the log file so the webpage stops.
#
709 710
if (!$template_mode) {
    print "Experiment $pid/$eid has been successfully created!\n";
711
    $experiment->CloseLogFile();
712
}
713 714 715 716

#
# Must unlock and drop batch experiments into the queue before exit.
#
717
if ($batchmode && !$frontend) {
718
    $experiment->Unlock(EXPTSTATE_QUEUED());
719 720
}
else {
721
    $experiment->Unlock();
722 723 724 725 726 727
}

#
# Clear the cancel flag now that the operation is complete. Must be
# done after we change the experiment state (above).
#
728
$experiment->SetCancelFlag(EXPTCANCEL_CLEAR());
729

730 731 732 733 734 735
#
# In template_mode we are done; the caller finishes up.
#
exit(0)
    if ($template_mode);

736 737 738
#
# Dump the report file and the log file to the user via email. 
#
739
my ($expt_created) = $experiment->created();
740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757
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";
758
}
759 760 761 762 763 764 765 766
$message .=
    "User:        $user_name\n" .
    "EID:         $eid\n" .
    "PID:         $pid\n" .
    "GID:         $gid\n" .
    "Description: $description\n" .
    "Swappable:   " . ($swappable ? "Yes\n" :
		                    "No  (Reason: $noswap_reason)\n") .
767
    "Idle-Swap:   " . ($idleswap  ? "Yes, at " . $idleswaptime/60.0 . " hours\n":
768
		                    "No  (Reason: $noidleswap_reason)\n") .
769
    "Auto-Swap:   " . ($autoswap  ? "Yes, at " . $autoswaptime/60.0 . " hours\n":
770 771 772 773 774 775 776 777 778 779 780 781 782 783
		                    "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: $TBLOGS",
784 785
	 ($repfile, $logname, $nsfile))
    if (!$zeemode);
786 787

# Done!
788
exit(0);
789

790 791 792
#
#
#
793
sub cleanup()
794
{
795
    #
796 797 798 799 800 801
    # 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 (!$committed) {
802 803 804
	# Completely remove all trace of the archive.
        libArchive::TBDestroyExperimentArchive($pid, $eid);

805
	#
806
	# Clear the experiment record and cleanup directories
807
	#
808 809
	$experiment->Delete(1)
	    if (defined($experiment));
810

811
	return;
812
    }
813

814 815 816 817 818 819 820 821 822
    #
    # Gather statistics.
    #
    if ($frontend) {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, $errorstat);
    }
    else {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, $errorstat);
    }
823
    $experiment->Refresh();
824

825 826 827
    #
    # Must clean up the experiment if it made it our of NEW state.
    #
828
    my $estate = $experiment->state();
829 830 831 832 833 834 835 836
    if ($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 && $signaled)) {
837 838
	    if ($experiment->Swap("out", "-force") != 0) {
		print "tbswap out -force failed!\n";
839
	    }
840
	    $experiment->SetState(EXPTSTATE_SWAPPED);
841
	}
842
	
843
	if ($experiment->End("-force") != 0) {
844
	    print "tbend failed!\n";
845 846
	}
    }
847
    $experiment->SetState(EXPTSTATE_TERMINATED);
848

849
    #
850 851 852 853
    # Okay, we *are* going to terminate the experiment.
    # 
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_TERMINATE, 0);

854 855
    # Clear the logfile so the webpage stops.
    $experiment->CloseLogFile();
856

Kevin Atkinson's avatar
 
Kevin Atkinson committed
857 858 859 860 861
    #
    # Figure out the error if possible
    #
    my $error_data = tblog_find_error();

862 863 864
    #
    # Send a message to the testbed list. 
    #
Kevin Atkinson's avatar
 
Kevin Atkinson committed
865 866 867 868 869 870
    tblog_email_error($error_data,
		      "$user_name <$user_email>",
		      "Config Failure", "$pid/$eid",
		      "$user_name <$user_email>",
		      "",
		      "Cc: $TBOPS",
Kevin Atkinson's avatar
 
Kevin Atkinson committed
871
		      "",
Kevin Atkinson's avatar
 
Kevin Atkinson committed
872
		      ($logname, "assign.log", "wanassign.log", $nsfile));
873

874
    #
875 876 877 878
    # Back up the work dir for post-mortem debugging. 
    #
    system("/bin/rm -rf  ${workdir}-failed");
    system("/bin/mv -f   $workdir ${workdir}-failed");
879

880 881
    #
    # Clear the record and cleanup.
882 883
    #
    $experiment->Delete();
884 885 886
}

#
887 888
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
889
#
890
sub ParseArgs()
891
{
892 893 894 895
    my %options = ();
    if (! getopts($optlist, \%options)) {
	usage();
    }
896

897
    if (@ARGV > 1) {
898 899
	usage();
    }
900 901
    if (@ARGV == 1) {
	$tempnsfile = $ARGV[0];
902

903
	# Note different taint check (allow /).
904
	if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
905 906 907
	    $tempnsfile = $1;
	}
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
908
	    tbdie("Bad data in nsfile: $tempnsfile");
909
	}
910

911 912 913 914 915 916 917 918 919 920
	#
	# Called from ops interactively. Make sure NS file in /proj or /users.
	#
	# Use realpath to resolve any symlinks.
	#
	my $translated = `realpath $tempnsfile`;
	if ($translated =~ /^([-\w\.\/]+)$/) {
	    $tempnsfile = $1;
	}
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
921
	    tbdie("Bad data returned by realpath: $translated");
922 923 924 925 926 927 928 929 930
	}

	#
	# The file must reside in /proj, /groups, or /users. Since this script
	# runs as the caller, regular file permission checks ensure its a file
	# the user is allowed to use. /tmp/$pid-$eid.nsfile.XXXXX also allowed
	# since this script is invoked directly from web interface.
	#
	if (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
931
	    ! ($tempnsfile =~ /^\/tmp\/\d+\.ns/) &&
932 933 934 935
	    ! ($tempnsfile =~ /^\/var\/tmp\/php\w+/) &&
	    ! ($tempnsfile =~ /^\/proj/) &&
	    ! ($tempnsfile =~ /^\/groups/) &&
	    ! ($tempnsfile =~ /^\/users/)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
936
	    tberror("$tempnsfile does not resolve to an allowed directory!");
937 938 939 940 941
	    # Note positive status; so error goes to user not tbops.
	    exit(1);
	}
    }
    
942
    if (defined($options{"i"})) {
943
	$batchmode = 0;
944
    }
945 946 947
    if (defined($options{"f"})) {
	$frontend = 1;
    }
948 949 950
    if (defined($options{"q"})) {
	$quiet = 1;
    }
951 952 953
    if (defined($options{"m"})) {
	$dpdb = 1;
    }
954 955 956 957
    if (defined($options{"z"})) {
	$zeemode = 1;
	$zeeopt  = "-p";
    }
958 959 960 961
    # This option should not be exported via the XMLRPC server. 
    if (defined($options{"n"})) {
	$idleignore = 1;
    }
962

963 964 965 966
    if (defined($options{"s"})) {
	$savestate = 1;
    }

Leigh B. Stoller's avatar
Leigh B. Stoller committed
967 968 969 970 971
    #
    # Clone an experiment, either an existing experiment or an old one
    # (using the archive). 
    #
    if (defined($options{"c"})) {
972
	$copyarg = $options{"c"};
Leigh B. Stoller's avatar
Leigh B. Stoller committed
973

974 975 976 977 978 979 980
	if (! (($copyarg =~ /^([-\w]+),([-\w]+)(?::[-\w]*)?$/) ||
	       ($copyarg =~ /^(\d+)(?::[-\w]*)?$/))) {
	    tbdie("Bad data in argument: $copyarg");
	}
	# This option only makes sense with -c option.
	if (defined($options{"b"})) {
	    $copybranch = 1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
981 982 983
	}
    }

984 985 986
    #
    # pid,eid,gid get passed along as shell commands args; must taint check.
    # 
987 988
    if (defined($options{"p"})) {
	$pid = $options{"p"};
989

990
	if ($pid =~ /^([-\w]+)$/) {
991 992
	    $pid = $1;
	}
993
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
994
	    tbdie("Bad data in argument: $pid.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
995
	}
996 997 998 999