batchexp.in 32.5 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

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;
112 113
my $user_name;
my $user_email;
114
my $dbuid;
115
my $exptidx;
116
my $logname;
117

118 119 120 121 122 123 124
# 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;
125 126 127 128 129 130

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

131 132 133 134 135 136 137
#
# 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);

138 139
#
# Untaint the path
140
#
141 142
# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
143 144
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

145 146 147 148 149
my $eid;
my $pid;
my $gid;
my $description;
my $tempnsfile;
150 151 152
my $swappable    = 1;
my $noswap_reason;
my $idleswap     = 1;
153
my $idleswaptime = 60 * TBGetSiteVar("idle/threshold");
154
my $noidleswap_reason;
155
my $autoswap     = 0;
156
my $autoswaptime = 10 * 60;
157 158 159
my $idleignore   = 0;
my $priority     = TB_EXPTPRIORITY_LOW;
my $exptstate    = EXPTSTATE_NEW();
160
my $batchstate   = BATCHSTATE_UNLOCKED();
161 162
my $now          = localtime();
my $committed    = 0;
163 164
my $experiment;
my $copy_experiment;
165

166
#
167
# Verify user and get his DB uid.
168
#
169
if (! UNIX2DBUID($UID, \$dbuid)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
170
    tbdie("You do not exist in the Emulab Database!");
171 172
}

173 174 175 176
#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
177
    tbdie("Cannot determine your name and email address.");
178 179
}

180 181 182 183 184
#
# 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
185 186
    tberror({cause => 'user'},
	    "You are over your disk quota on $CONTROL; please cleanup!");
187 188 189
    exit(1);
}

190
#
191
# Parse command arguments.
192
#
193
ParseArgs();
194
CheckCopyArgs();
195 196 197

#
# Sanity check them.
198
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
199 200 201
usage()
    if (!defined($pid) || !defined($eid));
usage()
202
    if (defined($tempnsfile) && defined($copyarg));
Leigh B. Stoller's avatar
Leigh B. Stoller committed
203

204 205 206
if (!defined($gid)) {
    $gid = $pid;
}
207
if (!defined($description)) {
208
    $description = "Created by $dbuid";
209
}
210
if (! $swappable && (!defined($noswap_reason) || $noswap_reason eq "")) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
211
    tbdie("Must provide a reason with -S option (not swappable reason)!");
212
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
213
if (! $idleswap && (!defined($noidleswap_reason) || $noidleswap_reason eq "")){
Kevin Atkinson's avatar
 
Kevin Atkinson committed
214
    tbdie("Must provide a reason with -L option (no idleswap reason)!");
215
}
216
if (!defined($tempnsfile) && !defined($copyarg) && !TBAdmin($dbuid)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
217
    tbdie("Only admins can create experiments with no NS file");
218
}
219 220
my $nsfile  = "$eid.ns";
my $repfile = "$eid.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
# Make sure UID is allowed to create experiments in this project.
230
#
231
if (! TBProjAccessCheck($dbuid, $pid, $gid, TB_PROJECT_CREATEEXPT)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
232
    die("You do not have permission to create experiments in $pid/$gid");
233 234
}

235 236 237
if (defined($copyarg)) {
    # This will be handled below.
    ;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
238 239 240
}
elsif (!defined($tempnsfile)) {
    # If no NS file, then override swap/idle stuff. 
241 242 243
    $swappable     = 0;
    $idleswap      = 0;
}
244
elsif (! -f $tempnsfile || ! -r $tempnsfile || -z $tempnsfile) {
245
    # Exit so that user sees the error, not us.
Kevin Atkinson's avatar
 
Kevin Atkinson committed
246
    tberror("$tempnsfile does not exist or is not a readable file!");
247 248
    exit(1);
}
249

250 251 252 253
#
# Batch jobs get a shorter idle time
#
my $swaptime = $idleswaptime;
254
if ($batchmode && TBSiteVarExists("idle/batch_threshold")) {
255 256 257 258 259 260
    my $batchidleswaptime = TBGetSiteVar("idle/batch_threshold");
    if ($swaptime > $batchidleswaptime) {
	$swaptime = $batchidleswaptime;
    }
}

261
#
262 263
# Grab me a secret key for accessing tar/rpm files via the web interface.
# Grab another secret key for the event system HMACs.
264
#
265 266
my $webkey   = TBGenSecretKey();
my $eventkey = TBGenSecretKey();
267

268 269 270 271 272 273 274 275 276
#
# In wait mode, block SIGINT until we spin off the background process.
#
if ($waitmode) {
    $SIG{QUIT} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{INT}  = 'IGNORE';
}

277
#
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{'idx'}		   = $exptidx;
$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
if ($experiment->CreateDirectory() != 0) {
327 328 329 330
    if (($? >> 8) == EDQUOT()) {
	# Obey exit status protocol for web page; User should see this.
	$errorstat = 1;
    }
331
    fatal("Failed to created experiment directory");
332 333
}

334 335 336 337
#
# Grab the working directory path, and thats where we work.
# The user's experiment directory is off in /proj space.
#
338 339
my $workdir = $experiment->WorkDir();
my $userdir = $experiment->UserDir();
340 341 342 343

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

344 345 346 347
#
# 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
348 349 350 351 352 353 354
if ($template_mode && defined($archive_eid)) {
    fatal("Could not create experiment archive!")
	if (libArchive::TBForkExperimentArchive($pid, $eid,
						$pid, $archive_eid, undef)
	    < 0);
}
elsif ($copybranch) {
355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
    # 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();
}

375 376 377 378 379 380 381 382 383 384
#
# 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).
# 

385 386 387 388
#
# Dump the eventkey into a file in the experiment directory. 
#
if ($EVENTSYS) {
389
    open(KEY, ">" . $experiment->EventKeyPath()) or
390 391 392 393 394
	fatal("Could not create eventkey file: $!");
    print KEY $eventkey;
    close(KEY);
}

395
# And dump the web key too.
396
open(KEY, ">" . $experiment->WebKeyPath()) or
397 398 399 400
    fatal("Could not create webkey file: $!");
print KEY $webkey;
close(KEY);

401
#
402 403 404
# 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). 
405
#
406
if (!defined($tempnsfile)) {
407
    $experiment->Unlock(EXPTSTATE_NEW());
408 409
    exit(0);
}
410

411
#
412
# Now we can get the NS file!
413
#
414
if (system("/bin/cp", "$tempnsfile", "$nsfile")) {
415
    fatal("Could not copy $tempnsfile to $workdir/$nsfile");
416
}
417
chmod(0664, "$nsfile");
418

419
#
420 421
# Run parse in impotent mode on the NS file.  This has no effect but
# will display any errors.
422
#
423
if (system("$parser -n $zeeopt $pid $gid $eid $nsfile") != 0) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
424 425
    # Obey exit status protocol for web page.
    $errorstat = 1;
426
    fatal("NS Parse failed!");
427 428
}

429 430 431
#
# Gather statistics; start the clock ticking.
#
432
if ($frontend || $batchmode) {
433 434 435 436 437 438 439 440
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, 0,
		    TBDB_STATS_FLAGS_START);
}
else {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, 0,
		    TBDB_STATS_FLAGS_START);
}

441 442 443
goto skiplog
    if ($template_mode);

444 445 446 447 448 449
#
# 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.
#
450 451
if ($experiment->CreateLogFile("startexp", \$logname) != 0) {
    fatal("Could not create logfile!");
452
    
453 454 455 456
}
$experiment->SetLogFile($logname);
$experiment->OpenLogFile($logname);

457 458
if (my $childpid = TBBackGround($logname)) {
    #
459 460
    # Parent exits normally, unless in waitmode. We have to set
    # justexit to make sure the END block below does not run.
461
    #
462 463
    $justexit = 1;
    
464 465 466
    if (!$waitmode) {
	print("Experiment $pid/$eid is now configuring\n".
 	      "You will be notified via email when the experiment is ".
467 468
	      "ready to use\n")
	    if (! $quiet);
469 470
	exit(0);
    }
471
    print("Waiting for " . ($batchmode ? "batch " : "") . "experiment $eid ".
472 473 474 475 476 477 478
	  "to finish " . ($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");
    }
479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495

    # 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);

496 497
    print("Done. Exited with status: $?\n")
	if (! $quiet);
Kevin Atkinson's avatar
 
Kevin Atkinson committed
498 499 500 501 502 503 504 505 506

    my $exit_code = $? >> 8;
    
    if ($exit_code != 0) {
	my $d = tblog_lookup_error();
	print tblog_format_error($d);
    }
    
    exit $exit_code;
507
}
508
TBdbfork();
509
skiplog:
510 511 512 513 514 515 516 517 518 519 520 521

# 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();
}

522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
#
# 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';

542 543
#
# The guts of starting an experiment!
544
#
545 546
# A batch experiment is essentially preloaded (frontend mode) and then
# dropped into the batch queue, unless the user requested only preload.
547 548
#

549
#
550
# Run the various scripts. We want to propagate the error from tbprerun
551 552
# and tbrun back out, hence the bogus looking errorstat variable.
#
553
$experiment->SetState(EXPTSTATE_PRERUN) == 0
554
    or fatal("Failed to set experiment state to " . EXPTSTATE_PRERUN());
555 556

if ($experiment->PreRun($nsfile, ($zeeopt ? "-z" : ""))) {
557
    $errorstat = $? >> 8;
558
    fatal("tbprerun failed!");
559
}
560 561

$experiment->SetState(EXPTSTATE_SWAPPED) == 0
562
    or fatal("Failed to set experiment state to " . EXPTSTATE_SWAPPED());
563 564 565 566

#
# If not in frontend mode (preload only) continue to swapping exp in.
# 
567
if (! ($frontend || $batchmode)) {
568
    $experiment->SetState(EXPTSTATE_ACTIVATING) == 0
569
	or fatal("Failed to set experiment state to ". EXPTSTATE_ACTIVATING());
570

571
    if ($experiment->Swap("in") != 0) {
572
	$errorstat = $? >> 8;
573
	fatal("tbswap in failed!");
574
    }
575 576
    
    $experiment->SetState(EXPTSTATE_ACTIVE) == 0
577
	or fatal("Failed to set experiment state to " . EXPTSTATE_ACTIVE());
578 579 580 581 582

    #
    # Look for the unsual case of more than 2 nodes and no vlans. Send a
    # warning message.
    #
583 584 585 586
    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
587
    if (@localnodes && scalar(@localnodes) > 2) {
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
	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.
605
if ($experiment->Report($repfile, "-b") != 0) {
606
    fatal("tbreport failed!");
607 608
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
609 610 611 612 613
# Latest log is always called the same thing.
if (defined($logname)) {
    system("cp -fp $logname $workdir/" . EXPTLOGNAME());
}

614
#
615
# Save a copy of the files for testbed information gathering (long term).
616
#
617
$experiment->SaveLogFiles();
618 619 620 621 622

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files.
#
623
$experiment->CopyLogFiles();
624

625 626 627 628 629 630
#
# Copy out some files to the archive directory so that they get saved later.
#
my $archivedir = libArchive::TBUserFileArchiveDirectory($pid, $eid);

if (! -e "$archivedir/tbdata") {
631
    mkdir("$archivedir/tbdata", 0775)
632 633 634
	or fatal("Failed to mkdir $archivedir/tbdata");
}
if (! -e "$archivedir/nsdata") {
635
    mkdir("$archivedir/nsdata", 0775)
636 637 638
	or fatal("Failed to mkdir $archivedir/nsdata");
}

639
system("cp -fp $workdir/$nsfile $archivedir/nsdata/nsfile.ns") == 0
640 641
    or fatal("Failed to copy nsfile to $archivedir/tbdata");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
642
if (defined($logname)) {
643
    system("cp -fp $logname $archivedir/tbdata/log.batchexp") == 0
644
    or fatal("Failed to copy logfile to $archivedir/tbdata");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
645
}
646

647
system("cp -fp $workdir/$repfile $archivedir/tbdata/tbreport.batchexp") == 0
648 649 650 651 652
    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
653

654 655 656
#
# Do a SavePoint on the experiment files.
#
657
print "Doing a savepoint on the experiment archive ...\n";
658 659 660 661
if (libArchive::TBExperimentArchiveSavePoint($pid, $eid, "startexp") < 0) {
    fatal("Failed to do a savepoint on the experiment archive!");
}

662 663 664 665 666 667 668 669 670 671
#
# 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!");
}

672 673 674 675 676 677 678 679 680 681
#
# Gather statistics.
#
if ($frontend || $batchmode) {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, 0);
}
else {
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, 0);
}

682
#
683
# Set accounting stuff, but on success only, and *after* gathering swap stats!
684
#
685
$experiment->SetSwapInfo($dbuid);
686

687 688 689
#
# Close up the log file so the webpage stops.
#
690 691
if (!$template_mode) {
    print "Experiment $pid/$eid has been successfully created!\n";
692
    $experiment->CloseLogFile();
693
}
694 695 696 697

#
# Must unlock and drop batch experiments into the queue before exit.
#
698
if ($batchmode && !$frontend) {
699
    $experiment->Unlock(EXPTSTATE_QUEUED());
700 701
}
else {
702
    $experiment->Unlock();
703 704 705 706 707 708
}

#
# Clear the cancel flag now that the operation is complete. Must be
# done after we change the experiment state (above).
#
709
$experiment->SetCancelFlag(EXPTCANCEL_CLEAR());
710

711 712 713 714 715 716
#
# In template_mode we are done; the caller finishes up.
#
exit(0)
    if ($template_mode);

717 718 719
#
# Dump the report file and the log file to the user via email. 
#
720
my ($expt_created) = $experiment->created();
721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738
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";
739
}
740 741 742 743 744 745 746 747
$message .=
    "User:        $user_name\n" .
    "EID:         $eid\n" .
    "PID:         $pid\n" .
    "GID:         $gid\n" .
    "Description: $description\n" .
    "Swappable:   " . ($swappable ? "Yes\n" :
		                    "No  (Reason: $noswap_reason)\n") .
748
    "Idle-Swap:   " . ($idleswap  ? "Yes, at " . $idleswaptime/60.0 . " hours\n":
749
		                    "No  (Reason: $noidleswap_reason)\n") .
750
    "Auto-Swap:   " . ($autoswap  ? "Yes, at " . $autoswaptime/60.0 . " hours\n":
751 752 753 754 755 756 757 758 759 760 761 762 763 764
		                    "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: $TBLOGS",
765 766
	 ($repfile, $logname, $nsfile))
    if (!$zeemode);
767 768

# Done!
769
exit(0);
770

771 772 773
#
#
#
774
sub cleanup()
775
{
776
    #
777 778 779 780 781 782
    # 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 (!$committed) {
783 784 785
	# Completely remove all trace of the archive.
        libArchive::TBDestroyExperimentArchive($pid, $eid);

786
	#
787
	# Clear the experiment record and cleanup directories
788
	#
789 790
	$experiment->Delete(1)
	    if (defined($experiment));
791

792
	return;
793
    }
794

795 796 797 798 799 800 801 802 803
    #
    # Gather statistics.
    #
    if ($frontend) {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_PRELOAD, $errorstat);
    }
    else {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_START, $errorstat);
    }
804
    $experiment->Refresh();
805

806 807 808
    #
    # Must clean up the experiment if it made it our of NEW state.
    #
809
    my $estate = $experiment->state();
810 811 812 813 814 815 816 817
    if ($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 && $signaled)) {
818 819
	    if ($experiment->Swap("out", "-force") != 0) {
		print "tbswap out -force failed!\n";
820
	    }
821
	    $experiment->SetState(EXPTSTATE_SWAPPED);
822
	}
823
	
824
	if ($experiment->End("-force") != 0) {
825
	    print "tbend failed!\n";
826 827
	}
    }
828
    $experiment->SetState(EXPTSTATE_TERMINATED);
829

830
    #
831 832 833 834
    # Okay, we *are* going to terminate the experiment.
    # 
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_TERMINATE, 0);

835 836
    # Clear the logfile so the webpage stops.
    $experiment->CloseLogFile();
837

Kevin Atkinson's avatar
 
Kevin Atkinson committed
838 839 840 841 842
    #
    # Figure out the error if possible
    #
    my $error_data = tblog_find_error();

843 844 845
    #
    # Send a message to the testbed list. 
    #
Kevin Atkinson's avatar
 
Kevin Atkinson committed
846 847 848 849 850 851
    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
852
		      "",
Kevin Atkinson's avatar
 
Kevin Atkinson committed
853
		      ($logname, "assign.log", "wanassign.log", $nsfile));
854

855
    #
856 857 858 859
    # Back up the work dir for post-mortem debugging. 
    #
    system("/bin/rm -rf  ${workdir}-failed");
    system("/bin/mv -f   $workdir ${workdir}-failed");
860

861 862
    #
    # Clear the record and cleanup.
863 864
    #
    $experiment->Delete();
865 866 867
}

#
868 869
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
870
#
871
sub ParseArgs()
872
{
873 874 875 876
    my %options = ();
    if (! getopts($optlist, \%options)) {
	usage();
    }
877

878
    if (@ARGV > 1) {
879 880
	usage();
    }
881 882
    if (@ARGV == 1) {
	$tempnsfile = $ARGV[0];
883

884
	# Note different taint check (allow /).
885
	if ($tempnsfile =~ /^([-\w\.\/]+)$/) {
886 887 888
	    $tempnsfile = $1;
	}
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
889
	    tbdie("Bad data in nsfile: $tempnsfile");
890
	}
891

892 893 894 895 896 897 898 899 900 901
	#
	# 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
902
	    tbdie("Bad data returned by realpath: $translated");
903 904 905 906 907 908 909 910 911
	}

	#
	# 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 (! ($tempnsfile =~ /^\/tmp\/[-\w]+-\d+\.nsfile/) &&
912
	    ! ($tempnsfile =~ /^\/tmp\/\d+\.ns/) &&
913 914 915 916
	    ! ($tempnsfile =~ /^\/var\/tmp\/php\w+/) &&
	    ! ($tempnsfile =~ /^\/proj/) &&
	    ! ($tempnsfile =~ /^\/groups/) &&
	    ! ($tempnsfile =~ /^\/users/)) {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
917
	    tberror("$tempnsfile does not resolve to an allowed directory!");
918 919 920 921 922
	    # Note positive status; so error goes to user not tbops.
	    exit(1);
	}
    }
    
923
    if (defined($options{"i"})) {
924
	$batchmode = 0;
925
    }
926 927 928
    if (defined($options{"f"})) {
	$frontend = 1;
    }
929 930 931
    if (defined($options{"q"})) {
	$quiet = 1;
    }
932 933 934
    if (defined($options{"m"})) {
	$dpdb = 1;
    }
935 936 937 938
    if (defined($options{"z"})) {
	$zeemode = 1;
	$zeeopt  = "-p";
    }
939 940 941 942
    # This option should not be exported via the XMLRPC server. 
    if (defined($options{"n"})) {
	$idleignore = 1;
    }
943

944 945 946 947
    if (defined($options{"s"})) {
	$savestate = 1;
    }

Leigh B. Stoller's avatar
Leigh B. Stoller committed
948 949 950 951 952
    #
    # Clone an experiment, either an existing experiment or an old one
    # (using the archive). 
    #
    if (defined($options{"c"})) {
953
	$copyarg = $options{"c"};
Leigh B. Stoller's avatar
Leigh B. Stoller committed
954

955 956 957 958 959 960 961
	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
962 963 964
	}
    }

965 966 967
    #
    # pid,eid,gid get passed along as shell commands args; must taint check.
    # 
968 969
    if (defined($options{"p"})) {
	$pid = $options{"p"};
970

971
	if ($pid =~ /^([-\w]+)$/) {
972 973
	    $pid = $1;
	}
974
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
975
	    tbdie("Bad data in argument: $pid.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
976
	}
977 978 979 980
    }
    if (defined($options{"e"})) {
	$eid = $options{"e"};

981
	if ($eid =~ /^([-\w]+)$/) {
982 983 984
	    $eid = $1;
	}
	else {
Kevin Atkinson's avatar
 
Kevin Atkinson committed
985
	    tbdie("Bad data in argument: $eid.");