| 1 |
#!/usr/bin/perl |
| 2 |
# |
| 3 |
# Copyright (C) Nikhef 2011 |
| 4 |
# |
| 5 |
# Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 |
# you may not use this file except in compliance with the License. |
| 7 |
# You may obtain a copy of the License at |
| 8 |
# |
| 9 |
# http://www.apache.org/licenses/LICENSE-2.0 |
| 10 |
# |
| 11 |
# Unless required by applicable law or agreed to in writing, software |
| 12 |
# distributed under the License is distributed on an "AS IS" BASIS, |
| 13 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 |
# See the License for the specific language governing permissions and |
| 15 |
# limitations under the License. |
| 16 |
# |
| 17 |
# Author: |
| 18 |
# Mischa Sall\'e <msalle@nikhef.nl> |
| 19 |
# NIKHEF Amsterdam, the Netherlands |
| 20 |
# |
| 21 |
######################################################################## |
| 22 |
# |
| 23 |
# Nagios probe to test functioning of gLExec |
| 24 |
# |
| 25 |
# Nagios state can be one of the following: |
| 26 |
# - Missing glexec command: CRITICAL |
| 27 |
# - input proxies empty: UNKNOWN |
| 28 |
# - short timeout exceeded: WARNING |
| 29 |
# - timeout exceeded: CRITICAL |
| 30 |
# - gLExec exit codes: |
| 31 |
# 0 glexec succeeded: OK |
| 32 |
# 201 Client error: CRITICAL |
| 33 |
# 202 Internal error: CRITICAL |
| 34 |
# 203 Auth error: CRITICAL |
| 35 |
# 204 Overlap: CRITICAL |
| 36 |
# 126 execve failed: WARNING |
| 37 |
# 128+n signal: WARNING |
| 38 |
# !=0 rc of payload: WARNING |
| 39 |
# |
| 40 |
######################################################################## |
| 41 |
|
| 42 |
# DEFAULTS |
| 43 |
my $probeversion=0.2; |
| 44 |
|
| 45 |
# Note the following defaults can be overridden using cmdline options |
| 46 |
my $deftimeout=10; # Overall timeout for probe |
| 47 |
my $defcritical=8; # When to send SIGTERM |
| 48 |
my $defwarning=5; # When to warn about slow running |
| 49 |
my $defpayload="id -a"; # Which payload to run |
| 50 |
|
| 51 |
######################################################################## |
| 52 |
# Logging package |
| 53 |
# keeps internal log trace which can be dumped with dump_log |
| 54 |
######################################################################## |
| 55 |
package logger; |
| 56 |
use strict; |
| 57 |
use warnings; |
| 58 |
{ |
| 59 |
my $loglevel; |
| 60 |
my @logstring; |
| 61 |
|
| 62 |
# Constructor |
| 63 |
sub new { |
| 64 |
my $classname=shift; |
| 65 |
my $self={}; bless $self; |
| 66 |
my $level=shift; |
| 67 |
if (defined $level) { |
| 68 |
$self->set_loglevel($level); |
| 69 |
} else { |
| 70 |
$loglevel=0; |
| 71 |
} |
| 72 |
return $self; |
| 73 |
} |
| 74 |
|
| 75 |
# Sets loglevel |
| 76 |
sub set_loglevel($) { |
| 77 |
my $self=shift; |
| 78 |
my $level=shift; |
| 79 |
$loglevel=$level; |
| 80 |
} |
| 81 |
|
| 82 |
# Logging function: log_func(priority, "logstring\n"); |
| 83 |
sub log_func($@) { |
| 84 |
my $self=shift; |
| 85 |
my $prio=shift; |
| 86 |
return if ($prio > $loglevel); |
| 87 |
for my $line (@_) { |
| 88 |
push @logstring,$line; |
| 89 |
} |
| 90 |
} |
| 91 |
|
| 92 |
# Dumps log |
| 93 |
sub get_log(@) { |
| 94 |
my $self=shift; |
| 95 |
foreach my $myentry ( @logstring ) { |
| 96 |
print $myentry; |
| 97 |
} |
| 98 |
} |
| 99 |
} |
| 100 |
|
| 101 |
######################################################################## |
| 102 |
# Nagios status printing package |
| 103 |
# Can set and dump nagios status output |
| 104 |
######################################################################## |
| 105 |
package nagstat; |
| 106 |
{ |
| 107 |
my $code; |
| 108 |
my $summary; |
| 109 |
my $perfdata; |
| 110 |
my @stat; |
| 111 |
|
| 112 |
# Constructor |
| 113 |
sub new() { |
| 114 |
my $classname=shift; |
| 115 |
my $self={}; bless $self; |
| 116 |
$code=3; # Default status unknown |
| 117 |
$summary=undef; |
| 118 |
$perfdata=undef; |
| 119 |
@stat=("OK","WARNING","CRITICAL","UNKNOWN"); |
| 120 |
return $self; |
| 121 |
} |
| 122 |
|
| 123 |
# Set nagios code (0-3) plus summary |
| 124 |
sub set_status($$) { |
| 125 |
my $self=shift; |
| 126 |
if (!defined $summary) { |
| 127 |
$code=shift; |
| 128 |
$summary=shift; |
| 129 |
} |
| 130 |
} |
| 131 |
|
| 132 |
# Set internal performance data |
| 133 |
sub set_perfdata($) { |
| 134 |
my $self=shift; |
| 135 |
$perfdata=shift; |
| 136 |
} |
| 137 |
|
| 138 |
# Printout nagios status, summary and optionally performance data |
| 139 |
# return value is code (0-3) |
| 140 |
sub get_status { |
| 141 |
if (!defined $summary) { |
| 142 |
$summary="unknown status"; |
| 143 |
} |
| 144 |
if (defined $perfdata) { |
| 145 |
print $stat[$code].": ".$summary."|".$perfdata."\n"; |
| 146 |
} else { |
| 147 |
print $stat[$code].": ".$summary."\n"; |
| 148 |
} |
| 149 |
return $code; |
| 150 |
} |
| 151 |
} |
| 152 |
|
| 153 |
######################################################################## |
| 154 |
# Inter process communication package for nagios probes |
| 155 |
# Starts alarm handler when receiving alarm which checks status of |
| 156 |
# probe, and terminates or kills it. |
| 157 |
######################################################################## |
| 158 |
package probeipc; |
| 159 |
use POSIX ":sys_wait_h"; |
| 160 |
{ |
| 161 |
my $pid; |
| 162 |
my $wpid; |
| 163 |
my $status; |
| 164 |
my $numsent; |
| 165 |
my $killtime; |
| 166 |
my $termtime; |
| 167 |
my $exitfunc; |
| 168 |
|
| 169 |
# Constructor: new(exitfunc,[kill time], [term time]) |
| 170 |
sub new() { |
| 171 |
my $classname=shift; |
| 172 |
my $self={}; bless $self; |
| 173 |
my $exitfunc=shift or die ($classname."::new() needs exitfunc arg\n"); |
| 174 |
my $killtime=(shift or 10); # probe default timeout is 10 |
| 175 |
my $termtime=(shift or $killtime); |
| 176 |
$self->set_exitfunc($exitfunc); |
| 177 |
$self->set_killtime($killtime); |
| 178 |
$self->set_termtime($termtime); |
| 179 |
$pid=-1; |
| 180 |
$wpid=0; |
| 181 |
$status=0; |
| 182 |
$numsent=0; |
| 183 |
$SIG{'ALRM'} = \&alarm_handler; |
| 184 |
$SIG{'INT'} = \&int_handler; |
| 185 |
$SIG{'TERM'} = \&int_handler; |
| 186 |
return $self; |
| 187 |
} |
| 188 |
|
| 189 |
# Sets time after which to send SIGKILL |
| 190 |
sub set_killtime($) { |
| 191 |
my $self=shift; |
| 192 |
$killtime=shift; |
| 193 |
} |
| 194 |
|
| 195 |
# Sets time after which to send SIGTERM |
| 196 |
sub set_termtime($) { |
| 197 |
my $self=shift; |
| 198 |
$termtime=shift; |
| 199 |
} |
| 200 |
|
| 201 |
# Sets function to call when exiting after sending a SIGKILL |
| 202 |
sub set_exitfunc($) { |
| 203 |
my $self=shift; |
| 204 |
$exitfunc=shift; |
| 205 |
} |
| 206 |
|
| 207 |
# Signal handler for SIGALRM |
| 208 |
sub alarm_handler() { |
| 209 |
my ($sig) = @_; |
| 210 |
my $rc; |
| 211 |
if ($pid<0) { # No pid, nothing to do |
| 212 |
logger->log_func(2,"Payload hasn't started yet\n"); |
| 213 |
nagstat->set_status(2,"probe killtime exceeded"); |
| 214 |
&$exitfunc(); |
| 215 |
} |
| 216 |
# Either is or was a process: test status |
| 217 |
logger->log_func(2,"subprocess is/was running with pid ".$pid."\n"); |
| 218 |
if ($wpid!=0) { # process has finished (>0)or waitpid(-1) failed |
| 219 |
return; |
| 220 |
} |
| 221 |
# Get status |
| 222 |
$wpid=waitpid($pid,WNOHANG); |
| 223 |
$status=$?; |
| 224 |
if ($wpid==0) { # Still running |
| 225 |
if ($killtime<=$termtime || $numsent==1) { |
| 226 |
logger->log_func(2,"Sending SIGKILL to ".$pid."\n"); |
| 227 |
kill(9,$pid); |
| 228 |
nagstat->set_status(2,"probe timeout exceeded"); |
| 229 |
&$exitfunc(); |
| 230 |
} |
| 231 |
logger->log_func(2,"Sending SIGTERM to ".$pid."\n"); |
| 232 |
kill(15,$pid); |
| 233 |
$numsent=1; |
| 234 |
alarm($killtime-$termtime); |
| 235 |
nagstat->set_status(2,"probe critical time exceeded"); |
| 236 |
} |
| 237 |
return; |
| 238 |
} |
| 239 |
|
| 240 |
# Signal handler for SIGINT and SIGTERM |
| 241 |
sub int_handler() { |
| 242 |
my ($sig)=@_; |
| 243 |
|
| 244 |
logger->log_func(2,"Caught SIG$sig\n"); |
| 245 |
nagstat->set_status(2,"probe interrupted with SIG$sig"); |
| 246 |
if ($pid<0) { # No pid, nothing to do |
| 247 |
logger->log_func(2,"Payload hasn't started yet\n"); |
| 248 |
&$exitfunc(); |
| 249 |
} |
| 250 |
# Either is or was a process: test status |
| 251 |
if ($wpid!=0) { # process has finished (>0)or waitpid(-1) failed |
| 252 |
logger->log_func(2,"Subprocess with pid ".$pid." already finished\n"); |
| 253 |
&$exitfunc(); |
| 254 |
} |
| 255 |
# Get status |
| 256 |
$wpid=waitpid($pid,WNOHANG); |
| 257 |
$status=$?; |
| 258 |
if ($wpid==0) { # Still running: send SIGTERM |
| 259 |
logger->log_func(2,"Sending SIGTERM to ".$pid."\n"); |
| 260 |
kill(15,$pid); |
| 261 |
} |
| 262 |
&$exitfunc(); |
| 263 |
} |
| 264 |
|
| 265 |
# Wait for specified pid and return exitcode and signal number. |
| 266 |
sub wait_probe() { |
| 267 |
my $self=shift; |
| 268 |
my $rc; |
| 269 |
my $signo; |
| 270 |
|
| 271 |
$wpid=waitpid($pid,0) if ($wpid<=0); |
| 272 |
if ($wpid==$pid) { # probe exited here |
| 273 |
alarm(0); |
| 274 |
$rc=$? >> 8; |
| 275 |
$signo=$? & 127; |
| 276 |
} elsif ($wpid==-1) { # probe exited in sighandler |
| 277 |
$rc=$status >> 8; |
| 278 |
$signo=$status & 127; |
| 279 |
} |
| 280 |
return ($rc,$signo); |
| 281 |
} |
| 282 |
|
| 283 |
# Starts specified command: run_probe("command",\$rc,\$signo) and returns 0 |
| 284 |
# on normal exit, or 1 when command cannot be started. |
| 285 |
# Output of command is stored as log info. Nagios status is set when |
| 286 |
# applicable. |
| 287 |
sub run_probe($$$) { |
| 288 |
my $self=shift; |
| 289 |
my $command=shift; |
| 290 |
my $rc=shift; |
| 291 |
my $signo=shift; |
| 292 |
|
| 293 |
# Start command |
| 294 |
$pid = open(FOO, $command." 2>&1|"); |
| 295 |
if (!defined($pid)) { |
| 296 |
alarm(0); |
| 297 |
nagstat->set_status(2,"Failed to run $command"); |
| 298 |
return 1; |
| 299 |
} |
| 300 |
while (my $line=<FOO>) { |
| 301 |
logger->log_func(3,$line); |
| 302 |
} |
| 303 |
($$rc,$$signo)=$self->wait_probe(); |
| 304 |
return 0; |
| 305 |
} |
| 306 |
} |
| 307 |
|
| 308 |
######################################################################## |
| 309 |
# Running main probe package |
| 310 |
######################################################################## |
| 311 |
package main; |
| 312 |
use strict; |
| 313 |
use warnings; |
| 314 |
|
| 315 |
use Env qw(@PATH GLEXEC_LOCATION GLITE_LOCATION X509_USER_PROXY GLEXEC_CLIENT_CERT); |
| 316 |
use Getopt::Long qw(:config no_ignore_case bundling); |
| 317 |
|
| 318 |
my $timeout; # Total maximum runtime for probe |
| 319 |
my $critical; # Time after which to kill gLExec |
| 320 |
my $warning; # Time after which to warn about slow gLExec |
| 321 |
my $payload; # Payload plus arguments: relative uses $PATH to find |
| 322 |
my $verbose; # Verbosity level |
| 323 |
|
| 324 |
# Prints usage output |
| 325 |
sub usage() { |
| 326 |
(my $name = $0) =~ s/.*\///; |
| 327 |
print <<EOHELP; |
| 328 |
Usage: $name [options] |
| 329 |
|
| 330 |
Options: |
| 331 |
-t|--timeout <timeout> maximum runtime for probe, default: $deftimeout sec |
| 332 |
-w|--warning <timeout> runtime after which to warn, default: $defwarning sec |
| 333 |
-c|--critical <timeout> runtime after which to probe is to be killed, |
| 334 |
default: $defcritical sec |
| 335 |
-x|--x509-user-proxy <file> set X509_USER_PROXY to given file |
| 336 |
-g|--glexec-client-cert <file> set GLEXEC_CLIENT_CERT to given file |
| 337 |
default: value of variable X509_USER_PROXY |
| 338 |
-e|--execute <cmd> command to execute by gLExec |
| 339 |
default: \"$defpayload\" |
| 340 |
-v|--verbose be more verbose, more -v means more verbosity |
| 341 |
-V|--version print version |
| 342 |
--help show this helptext |
| 343 |
-h show short usage information |
| 344 |
EOHELP |
| 345 |
exit 0; |
| 346 |
} |
| 347 |
|
| 348 |
# Prints short usage output (oneline) |
| 349 |
sub shortusage() { |
| 350 |
(my $name = $0) =~ s/.*\///; |
| 351 |
print <<EOHELP; |
| 352 |
Usage: $name [options] |
| 353 |
EOHELP |
| 354 |
} |
| 355 |
|
| 356 |
# Prints probe version |
| 357 |
sub version() { |
| 358 |
(my $name = $0) =~ s/.*\///; |
| 359 |
print <<EOHELP; |
| 360 |
$name version: $probeversion |
| 361 |
EOHELP |
| 362 |
} |
| 363 |
|
| 364 |
# Parses command line options and sets global variables |
| 365 |
sub getopts() { |
| 366 |
my $x509proxy; |
| 367 |
my $clientcert; |
| 368 |
my $version; |
| 369 |
my $help; |
| 370 |
my $shorthelp; |
| 371 |
|
| 372 |
$timeout=$deftimeout; |
| 373 |
$critical=$defcritical; |
| 374 |
$warning=$defwarning; |
| 375 |
$payload=$defpayload; |
| 376 |
GetOptions( |
| 377 |
"t|timeout=i" => \$timeout, |
| 378 |
"c|critical=i" => \$critical, |
| 379 |
"w|warning=i" => \$warning, |
| 380 |
"x|x509-user-proxy=s" => \$X509_USER_PROXY, |
| 381 |
"g|glexec-client-cert=s" => \$GLEXEC_CLIENT_CERT, |
| 382 |
"e|execute=s" => \$payload, |
| 383 |
"v|verbose+" => \$verbose, |
| 384 |
"help+" => \$help, |
| 385 |
"h+" => \$shorthelp, |
| 386 |
"V|version+" => \$version, |
| 387 |
"H|host", |
| 388 |
"p|port" |
| 389 |
) or &usage and exit(1); |
| 390 |
|
| 391 |
$help and &usage and exit(0); |
| 392 |
$shorthelp and &shortusage and exit(0); |
| 393 |
$version and &version and exit(0); |
| 394 |
if (!defined $GLEXEC_CLIENT_CERT) { |
| 395 |
$GLEXEC_CLIENT_CERT=$X509_USER_PROXY; |
| 396 |
} |
| 397 |
} |
| 398 |
|
| 399 |
# Exit function: prints nagios status and dumps log |
| 400 |
sub nagios_exit() { |
| 401 |
my $rc=nagstat->get_status(); |
| 402 |
|
| 403 |
# Logging object |
| 404 |
logger->get_log(); |
| 405 |
|
| 406 |
exit $rc; |
| 407 |
} |
| 408 |
|
| 409 |
# Finds gLExec in path and pre-specified directories |
| 410 |
sub find_glexec { |
| 411 |
my $self=shift; |
| 412 |
my $glexloc; |
| 413 |
my $dir; |
| 414 |
my @DEFAULT_PATH=("/usr/local/sbin","/usr/sbin","/sbin", |
| 415 |
"/usr/local/bin","/usr/bin"); |
| 416 |
|
| 417 |
# Try GLEXEC_LOCATION |
| 418 |
if (defined $GLEXEC_LOCATION) { |
| 419 |
logger->log_func(3,"GLEXEC_LOCATION=".$GLEXEC_LOCATION."\n"); |
| 420 |
$glexloc=$GLEXEC_LOCATION."/sbin/glexec"; |
| 421 |
if (-x $glexloc) { |
| 422 |
logger->log_func(2,"gLExec found at ".$glexloc."\n"); |
| 423 |
return $glexloc; |
| 424 |
} |
| 425 |
logger->log_func(2,"gLExec NOT found at \$GLEXEC_LOCATION\n"); |
| 426 |
} |
| 427 |
|
| 428 |
# Try GLITE_LOCATION |
| 429 |
$GLITE_LOCATION="/opt/glite" if (!defined $GLITE_LOCATION); |
| 430 |
logger->log_func(3,"GLITE_LOCATION=".$GLITE_LOCATION."\n"); |
| 431 |
|
| 432 |
@PATH=(".") if (!$PATH[1]); |
| 433 |
|
| 434 |
for $dir (@PATH,$GLITE_LOCATION."/sbin",@DEFAULT_PATH) { |
| 435 |
logger->log_func(3,"Looking for glexec in ".$dir."\n"); |
| 436 |
$glexloc=$dir."/glexec"; |
| 437 |
if (-x $glexloc) { |
| 438 |
logger->log_func(2,"gLExec found at ".$glexloc."\n"); |
| 439 |
return $glexloc; |
| 440 |
} |
| 441 |
} |
| 442 |
return undef; |
| 443 |
} |
| 444 |
|
| 445 |
sub glexec_to_nagios($$$) { |
| 446 |
my $rc=shift; |
| 447 |
my $signo=shift; |
| 448 |
my $dt=shift; |
| 449 |
|
| 450 |
if ($rc==0) { |
| 451 |
nagstat->set_perfdata("${dt}s;$warning;$critical;0"); |
| 452 |
if ($dt>=$warning) { |
| 453 |
nagstat->set_status(1,"gLExec took long time to succeed"); |
| 454 |
return 0; |
| 455 |
} else { |
| 456 |
nagstat->set_status(0,"Success"); |
| 457 |
return 1; |
| 458 |
} |
| 459 |
} elsif ($rc==126) { |
| 460 |
nagstat->set_status(1,"executable $payload can't be executed ($rc)"); |
| 461 |
} elsif ($rc==201) { |
| 462 |
nagstat->set_status(2,"client error ($rc)"); |
| 463 |
} elsif ($rc==202) { |
| 464 |
nagstat->set_status(2,"system error ($rc)"); |
| 465 |
} elsif ($rc==203) { |
| 466 |
nagstat->set_status(2,"authorization error ($rc)"); |
| 467 |
} elsif ($rc==204) { |
| 468 |
nagstat->set_status(2,"exit code overlap error ($rc)"); |
| 469 |
} elsif ($signo!=0) { |
| 470 |
nagstat->set_status(2,"exit due to signal $signo ($rc)"); |
| 471 |
} else { |
| 472 |
nagstat->set_status(2, |
| 473 |
"executable $payload failed with non-zero exit code ($rc)"); |
| 474 |
} |
| 475 |
return 1; |
| 476 |
} |
| 477 |
|
| 478 |
# Find gLExec command, payload command (when relative), runs it and returns |
| 479 |
# status |
| 480 |
sub run_glexec() { |
| 481 |
my $glexec; |
| 482 |
my $exitcode; |
| 483 |
my $signo; |
| 484 |
my $t1; |
| 485 |
my $t2; |
| 486 |
|
| 487 |
# Make sure to have starttime |
| 488 |
$t1=time(); |
| 489 |
|
| 490 |
# Set alarm before looking for gLExec to prevent NFS timeouts |
| 491 |
alarm($critical); |
| 492 |
|
| 493 |
# Find glexec command |
| 494 |
if (!defined ($glexec=find_glexec)) { |
| 495 |
nagstat->set_status(2,"glexec command not found"); |
| 496 |
return 1; |
| 497 |
} |
| 498 |
|
| 499 |
# Check proxies |
| 500 |
if (!defined $X509_USER_PROXY) { |
| 501 |
nagstat->set_status(3,"\$X509_USER_PROXY is unset."); |
| 502 |
return 1; |
| 503 |
} |
| 504 |
if (! -e $X509_USER_PROXY || ! -s $X509_USER_PROXY) { |
| 505 |
nagstat->set_status(3, |
| 506 |
"\$X509_USER_PROXY does not point to a nonempty file."); |
| 507 |
return 1; |
| 508 |
} |
| 509 |
if (! -e $GLEXEC_CLIENT_CERT || ! -s $GLEXEC_CLIENT_CERT) { |
| 510 |
nagstat->set_status(3, |
| 511 |
"\$GLEXEC_CLIENT_CERT does not point to a nonempty file."); |
| 512 |
return 1; |
| 513 |
} |
| 514 |
|
| 515 |
# Find full path for payload if it's relative |
| 516 |
if ($payload !~ /^\/.*/) { |
| 517 |
(my $name=$payload) =~ s/ .*//; |
| 518 |
my $fullname; |
| 519 |
for my $dir (@PATH) { |
| 520 |
logger->log_func(3,"Looking for ".$name." in ".$dir."\n"); |
| 521 |
$fullname=$dir."/".$name; |
| 522 |
if (-x $fullname) { |
| 523 |
($payload=$payload) =~ s/^$name/$fullname/; |
| 524 |
logger->log_func(2,"Payload set to ".$payload."\n"); |
| 525 |
last; |
| 526 |
} |
| 527 |
} |
| 528 |
} |
| 529 |
|
| 530 |
# Run actual probe in child process |
| 531 |
if (probeipc->run_probe("$glexec $payload",\$exitcode,\$signo)!=0) { |
| 532 |
return 1; |
| 533 |
} |
| 534 |
|
| 535 |
# Probe exited: find exit status |
| 536 |
$t2=time(); |
| 537 |
return glexec_to_nagios($exitcode,$signo,$t2-$t1); |
| 538 |
} |
| 539 |
|
| 540 |
# Parse commandline options |
| 541 |
getopts(); |
| 542 |
|
| 543 |
# Initialize logger and set loglevel |
| 544 |
logger->new($verbose); |
| 545 |
|
| 546 |
# Initialize objects |
| 547 |
nagstat->new(); |
| 548 |
|
| 549 |
# Initialize signal handling |
| 550 |
probeipc->new(\&nagios_exit,$timeout,$critical); |
| 551 |
|
| 552 |
# run actual gLExec probe |
| 553 |
run_glexec(); |
| 554 |
|
| 555 |
# Dump nagios status, log and exit |
| 556 |
nagios_exit(); |
| 557 |
|
Parent Directory
| 
Revision Log