]> git.uio.no Git - check_openmanage.git/blame - check_openmanage
ferdig med test
[check_openmanage.git] / check_openmanage
CommitLineData
669797e1 1#!/usr/bin/perl
2#
3# Nagios plugin
4#
5# Monitor Dell server hardware status using Dell OpenManage Server
6# Administrator, either locally via NRPE, or remotely via SNMP.
7#
8# $Id$
9#
f1728beb 10# Copyright (C) 2010 Trond H. Amundsen
669797e1 11#
12# This program is free software: you can redistribute it and/or modify
13# it under the terms of the GNU General Public License as published by
14# the Free Software Foundation, either version 3 of the License, or
15# (at your option) any later version.
16#
17# This program is distributed in the hope that it will be useful, but
18# WITHOUT ANY WARRANTY; without even the implied warranty of
19# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20# General Public License for more details.
21#
22# You should have received a copy of the GNU General Public License
23# along with this program. If not, see <http://www.gnu.org/licenses/>.
24#
25
26require 5.006; # Perl v5.6.0 or newer is required
27use strict;
28use warnings;
29use POSIX qw(isatty ceil);
c76b83db 30use Getopt::Long qw(:config no_ignore_case);
669797e1 31
32# Global (package) variables used throughout the code
33use vars qw( $NAME $VERSION $AUTHOR $CONTACT $E_OK $E_WARNING $E_CRITICAL
34 $E_UNKNOWN $FW_LOCK $USAGE $HELP $LICENSE
35 $snmp_session $snmp_error $omreport $globalstatus $global
36 $linebreak $omopt_chassis $omopt_system $blade
7c03958b 37 $exit_code $snmp
48aeec0b 38 %check %opt %reverse_exitcode %status2nagios
669797e1 39 %snmp_status %snmp_probestatus %probestatus2nagios %sysinfo
40 %blacklist %nagios_alert_count %count
48aeec0b 41 @perl_warnings @controllers @enclosures @perfdata
669797e1 42 @report_storage @report_chassis @report_other
43 );
44
45#---------------------------------------------------------------------
46# Initialization and global variables
47#---------------------------------------------------------------------
48
70ec369c 49# Collect perl warnings in an array
50$SIG{__WARN__} = sub { push @perl_warnings, [@_]; };
cbbc270f 51
669797e1 52# Version and similar info
53$NAME = 'check_openmanage';
c000d734 54$VERSION = '3.5.7-beta19';
669797e1 55$AUTHOR = 'Trond H. Amundsen';
56$CONTACT = 't.h.amundsen@usit.uio.no';
57
58# Exit codes
59$E_OK = 0;
60$E_WARNING = 1;
61$E_CRITICAL = 2;
62$E_UNKNOWN = 3;
63
64# Firmware update lock file [FIXME: location on Windows?]
65$FW_LOCK = '/var/lock/.spsetup'; # default on Linux
66
67# Usage text
68$USAGE = <<"END_USAGE";
69Usage: $NAME [OPTION]...
70END_USAGE
71
72# Help text
73$HELP = <<'END_HELP';
74
75GENERAL OPTIONS:
76
77 -p, --perfdata Output performance data
78 -t, --timeout Plugin timeout in seconds
79 -c, --critical Customise temperature critical limits
80 -w, --warning Customise temperature warning limits
81 -d, --debug Debug output, reports everything
82 -h, --help Display this help text
83 -V, --version Display version info
84
85SNMP OPTIONS:
86
a43b0da8 87 -H, --hostname Hostname or IP (required for SNMP)
669797e1 88 -C, --community SNMP community string
89 -P, --protocol SNMP protocol version
90 --port SNMP port number
91
92OUTPUT OPTIONS:
93
94 -i, --info Prefix any alerts with the service tag
95 -e, --extinfo Append system info to alerts
96 -s, --state Prefix alerts with alert state
057193f5 97 -S, --short-state Prefix alerts with alert state (abbreviated)
669797e1 98 -o, --okinfo Verbosity when check result is OK
bee55928 99 -I, --htmlinfo HTML output with clickable links
669797e1 100
101CHECK CONTROL AND BLACKLISTING:
102
103 -a, --all Check everything, even log content
104 -b, --blacklist Blacklist missing and/or failed components
105 --only Only check a certain component or alert type
106 --check Fine-tune which components are checked
107
108For more information and advanced options, see the manual page or URL:
109 http://folk.uio.no/trondham/software/check_openmanage.html
110END_HELP
111
112# Version and license text
113$LICENSE = <<"END_LICENSE";
114$NAME $VERSION
f1728beb 115Copyright (C) 2010 $AUTHOR
669797e1 116License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
117This is free software: you are free to change and redistribute it.
118There is NO WARRANTY, to the extent permitted by law.
119
120Written by $AUTHOR <$CONTACT>
121END_LICENSE
122
123# Options with default values
397acff6 124%opt = ( 'blacklist' => [], # blacklisting
125 'check' => [], # check control
126 'critical' => [], # temperature critical limits
127 'warning' => [], # temperature warning limits
128 'timeout' => 30, # default timeout is 30 seconds
129 'debug' => 0, # debugging / verbose output
130 'help' => 0, # display help output
131 'perfdata' => undef, # output performance data
132 'info' => 0, # display servicetag
133 'extinfo' => 0, # display extra info
134 'htmlinfo' => undef, # html tags in output
135 'postmsg' => undef, # post message
136 'state' => 0, # display alert type
137 'short-state' => 0, # display alert type (short)
138 'okinfo' => 0, # default "ok" output level
139 'linebreak' => undef, # specify linebreak
140 'version' => 0, # plugin version info
141 'all' => 0, # check everything
142 'only' => undef, # only one component
143 'omreport' => undef, # omreport path
144 'port' => 161, # default SNMP port
145 'hostname' => undef, # hostname or IP
146 'community' => 'public', # SMNP v1 or v2c
147 'protocol' => 2, # default SNMP protocol 2c
148 'username' => undef, # SMNP v3
149 'authpassword' => undef, # SMNP v3
150 'authkey' => undef, # SMNP v3
151 'authprotocol' => undef, # SMNP v3
152 'privpassword' => undef, # SMNP v3
153 'privkey' => undef, # SMNP v3
154 'privprotocol' => undef, # SMNP v3
155 'use_get_table' => 0, # hack for SNMPv3 on Windows with net-snmp
669797e1 156 );
157
158# Get options
159GetOptions('b|blacklist=s' => \@{ $opt{blacklist} },
160 'check=s' => \@{ $opt{check} },
161 'c|critical=s' => \@{ $opt{critical} },
162 'w|warning=s' => \@{ $opt{warning} },
163 't|timeout=i' => \$opt{timeout},
164 'd|debug' => \$opt{debug},
165 'h|help' => \$opt{help},
166 'V|version' => \$opt{version},
167 'p|perfdata:s' => \$opt{perfdata},
168 'i|info' => \$opt{info},
169 'e|extinfo' => \$opt{extinfo},
bee55928 170 'I|htmlinfo:s' => \$opt{htmlinfo},
669797e1 171 'postmsg=s' => \$opt{postmsg},
172 's|state' => \$opt{state},
057193f5 173 'S|short-state' => \$opt{shortstate},
669797e1 174 'o|ok-info=i' => \$opt{okinfo},
da64c4d4 175 'linebreak=s' => \$opt{linebreak},
669797e1 176 'a|all' => \$opt{all},
177 'only=s' => \$opt{only},
9ed0700c 178 'omreport=s' => \$opt{omreport},
669797e1 179 'port=i' => \$opt{port},
180 'H|hostname=s' => \$opt{hostname},
181 'C|community=s' => \$opt{community},
182 'P|protocol=i' => \$opt{protocol},
183 'U|username=s' => \$opt{username},
184 'authpassword=s' => \$opt{authpassword},
185 'authkey=s' => \$opt{authkey},
186 'authprotocol=s' => \$opt{authprotocol},
187 'privpassword=s' => \$opt{privpassword},
188 'privkey=s' => \$opt{privkey},
189 'privprotocol=s' => \$opt{privprotocol},
4cabd748 190 'use-get_table' => \$opt{use_get_table},
669797e1 191 ) or do { print $USAGE; exit $E_UNKNOWN };
192
193# If user requested help
194if ($opt{help}) {
195 print $USAGE, $HELP;
196 exit $E_OK;
197}
198
199# If user requested version info
200if ($opt{version}) {
201 print $LICENSE;
202 exit $E_OK;
203}
204
205# Setting timeout
206$SIG{ALRM} = sub {
0ae24325 207 print "PLUGIN TIMEOUT: $NAME timed out after $opt{timeout} seconds\n";
669797e1 208 exit $E_UNKNOWN;
209};
210alarm $opt{timeout};
211
212# If we're using SNMP
213$snmp = defined $opt{hostname} ? 1 : 0;
214
215# SNMP session variables
216$snmp_session = undef;
217$snmp_error = undef;
218
219# The omreport command
220$omreport = undef;
221
222# Check flags, override available with the --check option
223%check = ( 'storage' => 1, # check storage subsystem
224 'memory' => 1, # check memory (dimms)
225 'fans' => 1, # check fan status
226 'power' => 1, # check power supplies
227 'temp' => 1, # check temperature
228 'cpu' => 1, # check processors
229 'voltage' => 1, # check voltage
230 'batteries' => 1, # check battery probes
231 'amperage' => 1, # check power consumption
232 'intrusion' => 1, # check intrusion detection
233 'alertlog' => 0, # check the alert log
234 'esmlog' => 0, # check the ESM log (hardware log)
235 'esmhealth' => 1, # check the ESM log overall health
236 );
237
238# Default line break
51e99613 239$linebreak = isatty(*STDOUT) ? "\n" : '<br/>';
669797e1 240
241# Line break from option
242if (defined $opt{linebreak}) {
243 if ($opt{linebreak} eq 'REG') {
244 $linebreak = "\n";
245 }
246 elsif ($opt{linebreak} eq 'HTML') {
247 $linebreak = '<br/>';
248 }
249 else {
250 $linebreak = $opt{linebreak};
251 }
252}
253
254# Exit with status=UNKNOWN if there is firmware upgrade in progress
255if (!$snmp && -f $FW_LOCK) {
256 print "MONITORING DISABLED - Firmware update in progress ($FW_LOCK exists)\n";
257 exit $E_UNKNOWN;
258}
259
260# List of controllers and enclosures
261@controllers = (); # controllers
262@enclosures = (); # enclosures
263
264# Messages
265@report_storage = (); # messages with associated nagios level (storage)
266@report_chassis = (); # messages with associated nagios level (chassis)
267@report_other = (); # messages with associated nagios level (other)
268
269# Counters for everything
270%count
271 = (
272 'pdisk' => 0, # number of physical disks
273 'vdisk' => 0, # number of logical drives (virtual disks)
274 'temp' => 0, # number of temperature probes
275 'volt' => 0, # number of voltage probes
276 'amp' => 0, # number of amperage probes
277 'intr' => 0, # number of intrusion probes
278 'dimm' => 0, # number of memory modules
279 'fan' => 0, # number of fan probes
280 'cpu' => 0, # number of CPUs
281 'bat' => 0, # number of batteries
282 'power' => 0, # number of power supplies
283 'esm' => {
284 'Critical' => 0, # critical entries in ESM log
285 'Non-Critical' => 0, # warning entries in ESM log
286 'Ok' => 0, # ok entries in ESM log
287 },
288 'alert' => {
289 'Critical' => 0, # critical entries in alert log
290 'Non-Critical' => 0, # warning entries in alert log
291 'Ok' => 0, # ok entries in alert log
292 },
293 );
294
295# Performance data
48aeec0b 296@perfdata = ();
669797e1 297
298# Global health status
299$global = 1; # default is to check global status
300$globalstatus = $E_OK; # default global health status is "OK"
301
302# Nagios error levels reversed
303%reverse_exitcode
304 = (
305 $E_OK => 'OK',
306 $E_WARNING => 'WARNING',
307 $E_CRITICAL => 'CRITICAL',
308 $E_UNKNOWN => 'UNKNOWN',
309 );
310
311# OpenManage (omreport) and SNMP error levels
312%status2nagios
313 = (
314 'Unknown' => $E_CRITICAL,
315 'Critical' => $E_CRITICAL,
316 'Non-Critical' => $E_WARNING,
317 'Ok' => $E_OK,
318 'Non-Recoverable' => $E_CRITICAL,
319 'Other' => $E_CRITICAL,
320 );
321
322# Status via SNMP
323%snmp_status
324 = (
325 1 => 'Other',
326 2 => 'Unknown',
327 3 => 'Ok',
328 4 => 'Non-Critical',
329 5 => 'Critical',
330 6 => 'Non-Recoverable',
331 );
332
333# Probe Status via SNMP
334%snmp_probestatus
335 = (
336 1 => 'Other', # probe status is not one of the following:
337 2 => 'Unknown', # probe status is unknown (not known or monitored)
338 3 => 'Ok', # probe is reporting a value within the thresholds
339 4 => 'nonCriticalUpper', # probe has crossed upper noncritical threshold
340 5 => 'criticalUpper', # probe has crossed upper critical threshold
341 6 => 'nonRecoverableUpper', # probe has crossed upper non-recoverable threshold
342 7 => 'nonCriticalLower', # probe has crossed lower noncritical threshold
343 8 => 'criticalLower', # probe has crossed lower critical threshold
344 9 => 'nonRecoverableLower', # probe has crossed lower non-recoverable threshold
345 10 => 'failed', # probe is not functional
346 );
347
348# Probe status translated to Nagios alarm levels
349%probestatus2nagios
350 = (
351 'Other' => $E_CRITICAL,
352 'Unknown' => $E_CRITICAL,
353 'Ok' => $E_OK,
354 'nonCriticalUpper' => $E_WARNING,
355 'criticalUpper' => $E_CRITICAL,
356 'nonRecoverableUpper' => $E_CRITICAL,
357 'nonCriticalLower' => $E_WARNING,
358 'criticalLower' => $E_CRITICAL,
359 'nonRecoverableLower' => $E_CRITICAL,
360 'failed' => $E_CRITICAL,
361 );
362
363# System information gathered
364%sysinfo
365 = (
366 'bios' => 'N/A', # BIOS version
367 'biosdate' => 'N/A', # BIOS release date
368 'serial' => 'N/A', # serial number (service tag)
369 'model' => 'N/A', # system model
370 'osname' => 'N/A', # OS name
371 'osver' => 'N/A', # OS version
372 'om' => 'N/A', # OMSA version
373 'bmc' => 0, # HAS baseboard management controller (BMC)
374 'rac' => 0, # HAS remote access controller (RAC)
375 'rac_name' => 'N/A', # remote access controller (RAC)
376 'bmc_fw' => 'N/A', # BMC firmware
377 'rac_fw' => 'N/A', # RAC firmware
378 );
379
380# Adjust which checks to perform
381adjust_checks() if defined $opt{check};
382
383# Blacklisted components
384%blacklist = defined $opt{blacklist} ? %{ get_blacklist() } : ();
385
386# If blacklisting is in effect, don't check global health status
387if (scalar keys %blacklist > 0) {
388 $global = 0;
389}
390
391# Take into account new hardware and blades
392$omopt_chassis = 'chassis'; # default "chassis" option to omreport
393$omopt_system = 'system'; # default "system" option to omreport
394$blade = 0; # if this is a blade system
395
396# Some initializations and checking before we begin
397if ($snmp) {
398 snmp_initialize(); # initialize SNMP
399 snmp_check(); # check that SNMP works
400 snmp_detect_blade(); # detect blade via SNMP
401}
402else {
403 # Find the omreport binary
404 find_omreport();
405 # Check help output from omreport, see which options are available.
406 # Also detecting blade via omreport.
407 check_omreport_options();
408}
409
410
411#---------------------------------------------------------------------
412# Helper functions
413#---------------------------------------------------------------------
414
415#
416# Store a message in one of the message arrays
417#
418sub report {
419 my ($type, $msg, $exval, $id) = @_;
420 defined $id or $id = q{};
421
422 my %type2array
423 = (
424 'storage' => \@report_storage,
425 'chassis' => \@report_chassis,
426 'other' => \@report_other,
427 );
428
429 return push @{ $type2array{$type} }, [ $msg, $exval, $id ];
430}
431
432
433#
434# Run command, put resulting output lines in an array and return a
435# pointer to that array
436#
437sub run_command {
438 my $command = shift;
439
440 open my $CMD, '-|', $command
441 or do { report('other', "Couldn't run command '$command': $!", $E_UNKNOWN)
442 and return [] };
443 my @lines = <$CMD>;
444 close $CMD
445 or do { report('other', "Couldn't close filehandle for command '$command': $!", $E_UNKNOWN)
446 and return \@lines };
447 return \@lines;
448}
449
450#
451# Run command, put resulting output in a string variable and return it
452#
453sub slurp_command {
454 my $command = shift;
455
456 open my $CMD, '-|', $command
457 or do { report('other', "Couldn't run command '$command': $!", $E_UNKNOWN) and return };
458 my $rawtext = do { local $/ = undef; <$CMD> }; # slurping
459 close $CMD;
460
461 # NOTE: We don't check the return value of close() since omreport
462 # does something weird sometimes.
463
464 return $rawtext;
465}
466
467#
468# Initialize SNMP
469#
470sub snmp_initialize {
471 # Legal SNMP v3 protocols
472 my $snmp_v3_privprotocol = qr{\A des|aes|aes128|3des|3desde \z}xms;
473 my $snmp_v3_authprotocol = qr{\A md5|sha \z}xms;
474
475 # Parameters to Net::SNMP->session()
476 my %param
477 = (
478 '-port' => $opt{port},
479 '-hostname' => $opt{hostname},
480 '-version' => $opt{protocol},
481 );
482
483 # Parameters for SNMP v3
484 if ($opt{protocol} == 3) {
485
486 # Username is mandatory
487 if (defined $opt{username}) {
488 $param{'-username'} = $opt{username};
489 }
490 else {
491 print "SNMP ERROR: With SNMPv3 the username must be specified\n";
492 exit $E_UNKNOWN;
493 }
494
495 # Authpassword is optional
496 if (defined $opt{authpassword}) {
497 $param{'-authpassword'} = $opt{authpassword};
498 }
499
500 # Authkey is optional
501 if (defined $opt{authkey}) {
502 $param{'-authkey'} = $opt{authkey};
503 }
504
505 # Privpassword is optional
506 if (defined $opt{privpassword}) {
507 $param{'-privpassword'} = $opt{privpassword};
508 }
509
510 # Privkey is optional
511 if (defined $opt{privkey}) {
512 $param{'-privkey'} = $opt{privkey};
513 }
514
515 # Privprotocol is optional
516 if (defined $opt{privprotocol}) {
517 if ($opt{privprotocol} =~ m/$snmp_v3_privprotocol/xms) {
518 $param{'-privprotocol'} = $opt{privprotocol};
519 }
520 else {
521 print "SNMP ERROR: Unknown privprotocol '$opt{privprotocol}', "
522 . "must be one of [des|aes|aes128|3des|3desde]\n";
523 exit $E_UNKNOWN;
524 }
525 }
526
527 # Authprotocol is optional
528 if (defined $opt{authprotocol}) {
529 if ($opt{authprotocol} =~ m/$snmp_v3_authprotocol/xms) {
530 $param{'-authprotocol'} = $opt{authprotocol};
531 }
532 else {
533 print "SNMP ERROR: Unknown authprotocol '$opt{authprotocol}', "
534 . "must be one of [md5|sha]\n";
535 exit $E_UNKNOWN;
536 }
537 }
538 }
539 # Parameters for SNMP v2c or v1
540 elsif ($opt{protocol} == 2 or $opt{protocol} == 1) {
541 $param{'-community'} = $opt{community};
542 }
543 else {
544 print "SNMP ERROR: Unknown SNMP version '$opt{protocol}'\n";
545 exit $E_UNKNOWN;
546 }
547
548 # Try to initialize the SNMP session
549 if ( eval { require Net::SNMP; 1 } ) {
550 ($snmp_session, $snmp_error) = Net::SNMP->session( %param );
551 if (!defined $snmp_session) {
552 printf "SNMP: %s\n", $snmp_error;
553 exit $E_UNKNOWN;
554 }
555 }
556 else {
0ae24325 557 print "ERROR: You need perl module Net::SNMP to run $NAME in SNMP mode\n";
669797e1 558 exit $E_UNKNOWN;
559 }
560 return;
561}
562
563#
564# Checking if SNMP works by probing for "chassisModelName", which all
565# servers should have
566#
567sub snmp_check {
568 my $chassisModelName = '1.3.6.1.4.1.674.10892.1.300.10.1.9.1';
569 my $result = $snmp_session->get_request(-varbindlist => [$chassisModelName]);
570
571 # Typically if remote host isn't responding
572 if (!defined $result) {
0ae24325 573 printf "SNMP CRITICAL: %s\n", $snmp_session->error;
669797e1 574 exit $E_CRITICAL;
575 }
576
577 # If OpenManage isn't installed or is not working
578 if ($result->{$chassisModelName} =~ m{\A noSuch (Instance|Object) \z}xms) {
0ae24325 579 print "ERROR: (SNMP) OpenManage is not installed or is not working correctly\n";
669797e1 580 exit $E_UNKNOWN;
581 }
582 return;
583}
584
585#
586# Detecting blade via SNMP
587#
588sub snmp_detect_blade {
589 my $DellBaseBoardType = '1.3.6.1.4.1.674.10892.1.300.80.1.7.1.1';
590 my $result = $snmp_session->get_request(-varbindlist => [$DellBaseBoardType]);
591
592 # Identify blade. Older models (4th and 5th gen models) and/or old
593 # OMSA (4.x) don't have this OID. If we get "noSuchInstance" or
594 # similar, we assume that this isn't a blade
5c370da3 595 if (exists $result->{$DellBaseBoardType} && $result->{$DellBaseBoardType} eq '3') {
669797e1 596 $blade = 1;
597 }
598 return;
599}
600
601#
602# Locate the omreport binary
603#
604sub find_omreport {
ac760e0d 605 # If user has specified path to omreport
606 if (defined $opt{omreport} and -x $opt{omreport}) {
60994ca4 607 $omreport = qq{"$opt{omreport}"};
ac760e0d 608 return;
609 }
610
669797e1 611 # Possible full paths for omreport
612 my @omreport_paths
613 = (
614 '/usr/bin/omreport', # default on Linux
6a050646 615 '/opt/dell/srvadmin/bin/omreport', # default on Linux with OMSA 6.2.0
669797e1 616 '/opt/dell/srvadmin/oma/bin/omreport.sh', # alternate on Linux
617 '/opt/dell/srvadmin/oma/bin/omreport', # alternate on Linux
9025e83f 618 'C:\Program Files (x86)\Dell\SysMgt\oma\bin\omreport.exe', # default on Windows x64
619 'C:\Program Files\Dell\SysMgt\oma\bin\omreport.exe', # default on Windows x32
421b6c77 620 'c:\progra~1\dell\sysmgt\oma\bin\omreport.exe', # 8bit legacy default on Windows x32
621 'c:\progra~2\dell\sysmgt\oma\bin\omreport.exe', # 8bit legacy default on Windows x64
669797e1 622 );
623
624 # Find the one to use
625 OMREPORT_PATH:
626 foreach my $bin (@omreport_paths) {
627 if (-x $bin) {
60347693 628 $omreport = qq{"$bin"};
669797e1 629 last OMREPORT_PATH;
630 }
631 }
632
633 # Exit with status=UNKNOWN if OM is not installed, or we don't
634 # have permission to execute the binary
635 if (!defined $omreport) {
0ae24325 636 print "ERROR: Dell OpenManage Server Administrator (OMSA) is not installed\n";
669797e1 637 exit $E_UNKNOWN;
638 }
639 return;
640}
641
642#
643# Checks output from 'omreport -?' and searches for arguments to
644# omreport, to accommodate deprecated options "chassis" and "system"
645# (on newer hardware), as well as blade servers.
646#
647sub check_omreport_options {
648 foreach (@{ run_command("$omreport -? 2>&1") }) {
649 if (m/\A servermodule /xms) {
650 # If "servermodule" argument to omreport exists, use it
651 # instead of argument "system"
652 $omopt_system = 'servermodule';
653 }
654 elsif (m/\A mainsystem /xms) {
655 # If "mainsystem" argument to omreport exists, use it
656 # instead of argument "chassis"
657 $omopt_chassis = 'mainsystem';
658 }
659 elsif (m/\A modularenclosure /xms) {
660 # If "modularenclusure" argument to omreport exists, assume
661 # that this is a blade
662 $blade = 1;
663 }
664 }
665 return;
666}
667
668#
669# Read the blacklist option and return a hash containing the
670# blacklisted components
671#
672sub get_blacklist {
673 my @bl = ();
674 my %blacklist = ();
675
676 if (scalar @{ $opt{blacklist} } >= 0) {
677 foreach my $black (@{ $opt{blacklist} }) {
678 my $tmp = q{};
679 if (-f $black) {
680 open my $BL, '<', $black
681 or do { report('other', "Couldn't open blacklist file $black: $!", $E_UNKNOWN)
682 and return {} };
683 $tmp = <$BL>;
684 close $BL;
685 chomp $tmp;
686 }
687 else {
688 $tmp = $black;
689 }
690 push @bl, $tmp;
691 }
692 }
693
694 return {} if $#bl < 0;
695
696 # Parse blacklist string, put in hash
697 foreach my $black (@bl) {
698 my @comps = split m{/}xms, $black;
699 foreach my $c (@comps) {
700 next if $c !~ m/=/xms;
701 my ($key, $val) = split /=/xms, $c;
702 my @vals = split /,/xms, $val;
703 $blacklist{$key} = \@vals;
704 }
705 }
706
707 return \%blacklist;
708}
709
710#
711# Read the check option and adjust the hash %check, which is a rough
712# list of components to be checked
713#
714sub adjust_checks {
715 my @cl = ();
716
717 # Adjust checking based on the '--all' option
718 if ($opt{all}) {
719 # Check option usage
720 if (defined $opt{only} and $opt{only} !~ m{\A critical|warning \z}xms) {
721 print qq{ERROR: Wrong simultaneous usage of the "--all" and "--only" options\n};
722 exit $E_UNKNOWN;
723 }
724 if (scalar @{ $opt{check} } > 0) {
725 print qq{ERROR: Wrong simultaneous usage of the "--all" and "--check" options\n};
726 exit $E_UNKNOWN;
727 }
728
729 # set the check hash to check everything
730 map { $_ = 1 } values %check;
731
732 return;
733 }
734
735 # Adjust checking based on the '--only' option
736 if (defined $opt{only} and $opt{only} !~ m{\A critical|warning \z}xms) {
737 # Check option usage
738 if (scalar @{ $opt{check} } > 0) {
739 print qq{ERROR: Wrong simultaneous usage of the "--only" and "--check" options\n};
740 exit $E_UNKNOWN;
741 }
a2bbb2c1 742 if (! exists $check{$opt{only}} && $opt{only} ne 'chassis') {
669797e1 743 print qq{ERROR: "$opt{only}" is not a known keyword for the "--only" option\n};
744 exit $E_UNKNOWN;
745 }
746
747 # reset the check hash
748 map { $_ = 0 } values %check;
749
750 # adjust the check hash
751 if ($opt{only} eq 'chassis') {
752 map { $check{$_} = 1 } qw(memory fans power temp cpu voltage
753 batteries amperage intrusion esmhealth);
754 }
755 else {
756 $check{$opt{only}} = 1;
757 }
758
759 return;
760 }
761
762 # Adjust checking based on the '--check' option
763 if (scalar @{ $opt{check} } >= 0) {
764 foreach my $check (@{ $opt{check} }) {
765 my $tmp = q{};
766 if (-f $check) {
767 open my $CL, '<', $check
768 or do { report('other', "Couldn't open check file $check: $!", $E_UNKNOWN) and return };
769 $tmp = <$CL>;
770 close $CL;
771 }
772 else {
773 $tmp = $check;
774 }
775 push @cl, $tmp;
776 }
777 }
778
779 return if $#cl < 0;
780
781 # Parse checklist string, put in hash
782 foreach my $check (@cl) {
783 my @checks = split /,/xms, $check;
784 foreach my $c (@checks) {
785 next if $c !~ m/=/xms;
786 my ($key, $val) = split /=/xms, $c;
787 $check{$key} = $val;
788 }
789 }
790
791 # Check if we should check global health status
792 CHECK_KEY:
793 foreach (keys %check) {
794 next CHECK_KEY if $_ eq 'esmlog'; # not part of global status
795 next CHECK_KEY if $_ eq 'alertlog'; # not part of global status
796
797 if ($check{$_} == 0) { # found something with checking turned off
798 $global = 0;
799 last CHECK_KEY;
800 }
801 }
802
803 return;
804}
805
806#
807# Runs omreport and returns an array of anonymous hashes containing
808# the output.
809# Takes one argument: string containing parameters to omreport
810#
811sub run_omreport {
812 my $command = shift;
813 my @output = ();
814 my @keys = ();
815
816 # Errors that are OK. Some low-end poweredge (and blades) models
817 # don't have RAID controllers, intrusion detection sensor, or
818 # redundant/instrumented power supplies etc.
819 my $ok_errors
820 = qr{
821 Intrusion\sinformation\sis\snot\sfound\sfor\sthis\ssystem # No intrusion probe
822 | No\sinstrumented\spower\ssupplies\sfound\son\sthis\ssystem # No instrumented PS (blades/low-end)
823 | No\scontrollers\sfound # No RAID controller
824 | No\sbattery\sprobes\sfound\son\sthis\ssystem # No battery probes
825 | Invalid\scommand:\spwrmonitoring # Older OMSAs lack this command(?)
9df480be 826# | Current\sprobes\snot\sfound # OMSA + RHEL5.4 bug
669797e1 827 }xms;
828
829 # Errors that are OK on blade servers
830 my $ok_blade_errors
831 = qr{
832 No\sfan\sprobes\sfound\son\sthis\ssystem # No fan probes
833 }xms;
834
835 # Run omreport and fetch output
836 my $rawtext = slurp_command("$omreport $command -fmt ssv 2>&1");
837 return [] if !defined $rawtext;
838
839 # Workaround for Openmanage BUG introduced in OMSA 5.5.0
4a4baf82 840 $rawtext =~ s{\n;}{;}gxms if $command eq 'storage controller';
841
842 # Openmanage sometimes puts a linebreak between "Error" and the
843 # actual error text
49a51b07 844 $rawtext =~ s{^Error\s*\n}{Error: }xms;
669797e1 845
846 # Parse output, store in array
4a4baf82 847 for ((split m{\n}xms, $rawtext)) {
848 if (m{\AError}xms) {
669797e1 849 next if m{$ok_errors}xms;
850 next if ($blade and m{$ok_blade_errors}xms);
851 report('other', "Problem running 'omreport $command': $_", $E_UNKNOWN);
852 }
853
854 next if !m/(.*?;){2}/xms; # ignore lines with less than 3 fields
855 my @vals = split /;/xms;
8ce893fd 856 if ($vals[0] =~ m/\A (Index|ID|Severity|Processor|Current\sSpeed) \z/xms) {
669797e1 857 @keys = @vals;
858 }
859 else {
860 my $i = 0;
861 push @output, { map { $_ => $vals[$i++] } @keys };
862 }
863
864 }
865
866 # Finally, return the collected information
867 return \@output;
868}
869
870
871#
872# Checks if a component is blacklisted. Returns 1 if the component is
873# blacklisted, 0 otherwise. Takes two arguments:
874# arg1: component name
875# arg2: component id or index
876#
877sub blacklisted {
878 my $name = shift; # component name
879 my $id = shift; # component id
880 my $ret = 0; # return value
881
882 if (defined $blacklist{$name}) {
883 foreach my $comp (@{ $blacklist{$name} }) {
d4c27ad8 884 if (defined $id and ($comp eq $id or uc($comp) eq 'ALL')) {
669797e1 885 $ret = 1;
886 }
887 }
888 }
889
890 return $ret;
891}
892
893# Converts the NexusID from SNMP to our version
894sub convert_nexus {
895 my $nexus = shift;
896 $nexus =~ s{\A \\}{}xms;
897 $nexus =~ s{\\}{:}gxms;
898 return $nexus;
899}
900
901# Sets custom temperature thresholds based on user supplied options
902sub custom_temperature_thresholds {
903 my $type = shift; # type of threshold, either w (warning) or c (critical)
904 my %thres = (); # will contain the thresholds
905 my @limits = (); # holds the input
906
907 my @opt = $type eq 'w' ? @{ $opt{warning} } : @{ $opt{critical} };
908
909 if (scalar @opt >= 0) {
910 foreach my $t (@opt) {
911 my $tmp = q{};
912 if (-f $t) {
913 open my $F, '<', $t
914 or do { report('other', "Couldn't open temperature threshold file $t: $!",
915 $E_UNKNOWN) and return {} };
916 $tmp = <$F>;
917 close $F;
918 }
919 else {
920 $tmp = $t;
921 }
922 push @limits, $tmp;
923 }
924 }
925
926 # Parse checklist string, put in hash
927 foreach my $th (@limits) {
928 my @tmp = split m{,}xms, $th;
929 foreach my $t (@tmp) {
930 next if $t !~ m{=}xms;
931 my ($key, $val) = split m{=}xms, $t;
932 if ($val =~ m{/}xms) {
933 my ($max, $min) = split m{/}xms, $val;
934 $thres{$key}{max} = $max;
935 $thres{$key}{min} = $min;
936 }
937 else {
938 $thres{$key}{max} = $val;
939 }
940 }
941 }
942
943 return \%thres;
944}
945
946
947# Gets the output from SNMP result according to the OIDs checked
948sub get_snmp_output {
949 my ($result,$oidref) = @_;
b0e15fc9 950 my @temp = ();
669797e1 951 my @output = ();
952
953 foreach my $oid (keys %{ $result }) {
b0e15fc9 954 my $short = $oid;
f47687c4 955 $short =~ s{\s}{}gxms; # remove whitespace
956 $short =~ s{\A (.+) \. (\d+) \z}{$1}xms; # remove last number
b0e15fc9 957 my $id = $2;
958 if (exists $oidref->{$short}) {
959 $temp[$id]{$oidref->{$short}} = $result->{$oid};
669797e1 960 }
961 }
b0e15fc9 962
963 # Remove any empty indexes
964 foreach my $out (@temp) {
965 if (defined $out) {
966 push @output, $out;
967 }
968 }
969
669797e1 970 return \@output;
971}
972
973
974# Map the controller or other item in-place
975sub map_item {
976 my ($key, $val, $list) = @_;
977
978 foreach my $lst (@{ $list }) {
979 if (!exists $lst->{$key}) {
980 $lst->{$key} = $val;
981 }
982 }
983 return;
984}
985
986# Return the URL for official Dell documentation for a specific
987# PowerEdge server
988sub documentation_url {
989 my $model = shift;
990
991 # create model short form, e.g. "r710"
992 $model =~ s{\A PowerEdge \s (.+?) \z}{lc($1)}exms;
993
994 # special case for blades (e.g. M600, M710), they have common
995 # documentation
996 $model =~ s{\A m\d+ \z}{m}xms;
997
998 return 'http://support.dell.com/support/edocs/systems/pe' . $model . '/';
999}
1000
1001# Return the URL for warranty information for a server with a given
1002# serial number (servicetag)
1003sub warranty_url {
1004 my $tag = shift;
1005
1006 # Dell support sites for different parts of the world
1007 my %supportsite
1008 = (
1009 'emea' => 'http://support.euro.dell.com/support/topics/topic.aspx/emea/shared/support/my_systems_info/',
1010 'ap' => 'http://supportapj.dell.com/support/topics/topic.aspx/ap/shared/support/my_systems_info/en/details?',
1011 'glob' => 'http://support.dell.com/support/topics/global.aspx/support/my_systems_info/details?',
1012 );
1013
1014 # warranty URLs for different country codes
1015 my %url
1016 = (
1017 # EMEA
1018 'at' => $supportsite{emea} . 'de/details?c=at&l=de&ServiceTag=', # Austria
1019 'be' => $supportsite{emea} . 'nl/details?c=be&l=nl&ServiceTag=', # Belgium
1020 'cz' => $supportsite{emea} . 'cs/details?c=cz&l=cs&ServiceTag=', # Czech Republic
1021 'de' => $supportsite{emea} . 'de/details?c=de&l=de&ServiceTag=', # Germany
1022 'dk' => $supportsite{emea} . 'da/details?c=dk&l=da&ServiceTag=', # Denmark
1023 'es' => $supportsite{emea} . 'es/details?c=es&l=es&ServiceTag=', # Spain
1024 'fi' => $supportsite{emea} . 'fi/details?c=fi&l=fi&ServiceTag=', # Finland
1025 'fr' => $supportsite{emea} . 'fr/details?c=fr&l=fr&ServiceTag=', # France
1026 'gr' => $supportsite{emea} . 'en/details?c=gr&l=el&ServiceTag=', # Greece
1027 'it' => $supportsite{emea} . 'it/details?c=it&l=it&ServiceTag=', # Italy
1028 'il' => $supportsite{emea} . 'en/details?c=il&l=en&ServiceTag=', # Israel
1029 'me' => $supportsite{emea} . 'en/details?c=me&l=en&ServiceTag=', # Middle East
1030 'no' => $supportsite{emea} . 'no/details?c=no&l=no&ServiceTag=', # Norway
1031 'nl' => $supportsite{emea} . 'nl/details?c=nl&l=nl&ServiceTag=', # The Netherlands
1032 'pl' => $supportsite{emea} . 'pl/details?c=pl&l=pl&ServiceTag=', # Poland
1033 'pt' => $supportsite{emea} . 'en/details?c=pt&l=pt&ServiceTag=', # Portugal
1034 'ru' => $supportsite{emea} . 'ru/details?c=ru&l=ru&ServiceTag=', # Russia
1035 'se' => $supportsite{emea} . 'sv/details?c=se&l=sv&ServiceTag=', # Sweden
1036 'uk' => $supportsite{emea} . 'en/details?c=uk&l=en&ServiceTag=', # United Kingdom
1037 'za' => $supportsite{emea} . 'en/details?c=za&l=en&ServiceTag=', # South Africa
1038 # America
1039 'br' => $supportsite{glob} . 'c=br&l=pt&ServiceTag=', # Brazil
1040 'ca' => $supportsite{glob} . 'c=ca&l=en&ServiceTag=', # Canada
1041 'mx' => $supportsite{glob} . 'c=mx&l=es&ServiceTag=', # Mexico
1042 'us' => $supportsite{glob} . 'c=us&l=en&ServiceTag=', # USA
1043 # Asia/Pacific
1044 'au' => $supportsite{ap} . 'c=au&l=en&ServiceTag=', # Australia
1045 'cn' => $supportsite{ap} . 'c=cn&l=zh&ServiceTag=', # China
1046 'in' => $supportsite{ap} . 'c=in&l=en&ServiceTag=', # India
1047 # default fallback
1048 'XX' => $supportsite{glob} . 'ServiceTag=', # default
1049 );
1050
1051 if (exists $url{$opt{htmlinfo}}) {
1052 return $url{$opt{htmlinfo}} . $tag;
1053 }
1054 else {
1055 return $url{XX} . $tag;
1056 }
1057}
1058
1059
912d8679 1060# This helper function returns the corresponding value of a hash key,
1061# but takes into account that the key may not exist
1062sub get_hashval {
1063 my $key = shift || return undef;
1064 my $hash = shift;
1065 return exists $hash->{$key} ? $hash->{$key} : "Undefined value $key";
1066}
1067
1068
669797e1 1069
1070#---------------------------------------------------------------------
1071# Check functions
1072#---------------------------------------------------------------------
1073
1074#-----------------------------------------
1075# Check global health status
1076#-----------------------------------------
1077sub check_global {
1078 my $health = $E_OK;
1079
1080 if ($snmp) {
1081 #
1082 # Checks global status, i.e. both storage and chassis
1083 #
1084 my $systemStateGlobalSystemStatus = '1.3.6.1.4.1.674.10892.1.200.10.1.2.1';
1085 my $result = $snmp_session->get_request(-varbindlist => [$systemStateGlobalSystemStatus]);
1086 if (!defined $result) {
98b224a3 1087 printf "SNMP ERROR [global]: %s\n", $snmp_error;
669797e1 1088 exit $E_UNKNOWN;
1089 }
1090 $health = $status2nagios{$snmp_status{$result->{$systemStateGlobalSystemStatus}}};
1091 }
1092 else {
1093 #
1094 # NB! This does not check storage, only chassis...
1095 #
1096 foreach (@{ run_command("$omreport $omopt_system -fmt ssv") }) {
1097 next if !m/;/xms;
1098 next if m/\A SEVERITY;COMPONENT/xms;
1099 if (m/\A (.+?);Main\sSystem(\sChassis)? /xms) {
1100 $health = $status2nagios{$1};
1101 last;
1102 }
1103 }
1104 }
1105
1106 # Return the status
1107 return $health;
1108}
1109
1110
1111#-----------------------------------------
1112# STORAGE: Check controllers
1113#-----------------------------------------
1114sub check_controllers {
5ecf578c 1115 return if blacklisted('ctrl', 'all');
1116
669797e1 1117 my $id = undef;
1118 my $nexus = undef;
1119 my $name = undef;
1120 my $state = undef;
1121 my $status = undef;
1122 my $minfw = undef;
1123 my $mindr = undef;
1124 my $firmware = undef;
1125 my $driver = undef;
9df480be 1126 my $minstdr = undef; # Minimum required Storport driver version
1127 my $stdr = undef; # Storport driver version
669797e1 1128 my @output = ();
1129
1130 if ($snmp) {
1131 my %ctrl_oid
1132 = (
1133 '1.3.6.1.4.1.674.10893.1.20.130.1.1.1' => 'controllerNumber',
1134 '1.3.6.1.4.1.674.10893.1.20.130.1.1.2' => 'controllerName',
1135 '1.3.6.1.4.1.674.10893.1.20.130.1.1.5' => 'controllerState',
1136 '1.3.6.1.4.1.674.10893.1.20.130.1.1.8' => 'controllerFWVersion',
1137 '1.3.6.1.4.1.674.10893.1.20.130.1.1.38' => 'controllerComponentStatus',
1138 '1.3.6.1.4.1.674.10893.1.20.130.1.1.39' => 'controllerNexusID',
1139 '1.3.6.1.4.1.674.10893.1.20.130.1.1.41' => 'controllerDriverVersion',
1140 '1.3.6.1.4.1.674.10893.1.20.130.1.1.44' => 'controllerMinFWVersion',
1141 '1.3.6.1.4.1.674.10893.1.20.130.1.1.45' => 'controllerMinDriverVersion',
1b3f1f77 1142 '1.3.6.1.4.1.674.10893.1.20.130.1.1.55' => 'controllerStorportDriverVersion',
1143 '1.3.6.1.4.1.674.10893.1.20.130.1.1.56' => 'controllerMinRequiredStorportVer',
669797e1 1144 );
ba199ee0 1145
1146 # We use get_table() here for the odd case where a server has
1147 # two or more controllers, and where some OIDs are missing on
1148 # one of the controllers.
1149 my $controllerTable = '1.3.6.1.4.1.674.10893.1.20.130.1';
1150 my $result = $snmp_session->get_table(-baseoid => $controllerTable);
669797e1 1151
1152 # No controllers is OK
1153 return if !defined $result;
1154
1155 @output = @{ get_snmp_output($result, \%ctrl_oid) };
1156 }
1157 else {
1158 @output = @{ run_omreport('storage controller') };
1159 }
1160
1161 my %ctrl_state
1162 = (
1163 0 => 'Unknown',
1164 1 => 'Ready',
1165 2 => 'Failed',
1166 3 => 'Online',
1167 4 => 'Offline',
1168 6 => 'Degraded',
1169 );
1170
1171 CTRL:
1172 foreach my $out (@output) {
1173 if ($snmp) {
08c259f3 1174 $id = $out->{controllerNumber} - 1;
1175 $name = $out->{controllerName};
912d8679 1176 $state = get_hashval($out->{controllerState}, \%ctrl_state);
08c259f3 1177 $status = $snmp_status{$out->{controllerComponentStatus}};
1178 $minfw = exists $out->{controllerMinFWVersion}
1179 ? $out->{controllerMinFWVersion} : undef;
1180 $mindr = exists $out->{controllerMinDriverVersion}
1181 ? $out->{controllerMinDriverVersion} : undef;
669797e1 1182 $firmware = exists $out->{controllerFWVersion}
1183 ? $out->{controllerFWVersion} : 'N/A';
1184 $driver = exists $out->{controllerDriverVersion}
1185 ? $out->{controllerDriverVersion} : 'N/A';
1b3f1f77 1186 $minstdr = exists $out->{'controllerMinRequiredStorportVer'}
1187 ? $out->{controllerMinRequiredStorportVer} : undef;
1188 $stdr = exists $out->{controllerStorportDriverVersion}
1189 ? $out->{controllerStorportDriverVersion} : undef;
669797e1 1190 $nexus = convert_nexus($out->{controllerNexusID});
1191 }
1192 else {
1193 $id = $out->{ID};
1194 $name = $out->{Name};
1195 $state = $out->{State};
1196 $status = $out->{Status};
1197 $minfw = $out->{'Minimum Required Firmware Version'} ne 'Not Applicable'
1198 ? $out->{'Minimum Required Firmware Version'} : undef;
1199 $mindr = $out->{'Minimum Required Driver Version'} ne 'Not Applicable'
1200 ? $out->{'Minimum Required Driver Version'} : undef;
1201 $firmware = $out->{'Firmware Version'} ne 'Not Applicable'
1202 ? $out->{'Firmware Version'} : 'N/A';
1203 $driver = $out->{'Driver Version'} ne 'Not Applicable'
1204 ? $out->{'Driver Version'} : 'N/A';
f86e57b8 1205 $minstdr = (exists $out->{'Minimum Required Storport Driver Version'}
1206 and $out->{'Minimum Required Storport Driver Version'} ne 'Not Applicable')
08c259f3 1207 ? $out->{'Minimum Required Storport Driver Version'} : undef;
f86e57b8 1208 $stdr = (exists $out->{'Storport Driver Version'}
1209 and $out->{'Storport Driver Version'} ne 'Not Applicable')
956cf4d1 1210 ? $out->{'Storport Driver Version'} : undef;
669797e1 1211 $nexus = $id;
1212 }
1213
1214 $name =~ s{\s+\z}{}xms; # remove trailing whitespace
1215 push @controllers, $id;
1216
1217 # Collecting some storage info
1218 $sysinfo{'controller'}{$id}{'id'} = $nexus;
1219 $sysinfo{'controller'}{$id}{'name'} = $name;
1220 $sysinfo{'controller'}{$id}{'driver'} = $driver;
1221 $sysinfo{'controller'}{$id}{'firmware'} = $firmware;
956cf4d1 1222 $sysinfo{'controller'}{$id}{'storport'} = $stdr;
669797e1 1223
1224 next CTRL if blacklisted('ctrl', $nexus);
1225
1226 # Special case: old firmware
1227 if (!blacklisted('ctrl_fw', $id) && defined $minfw) {
1228 chomp $firmware;
98b224a3 1229 my $msg = sprintf q{Controller %d [%s]: Firmware '%s' is out of date},
669797e1 1230 $id, $name, $firmware;
1231 report('storage', $msg, $E_WARNING, $nexus);
1232 }
1233 # Special case: old driver
1234 if (!blacklisted('ctrl_driver', $id) && defined $mindr) {
1235 chomp $driver;
98b224a3 1236 my $msg = sprintf q{Controller %d [%s]: Driver '%s' is out of date},
669797e1 1237 $id, $name, $driver;
1238 report('storage', $msg, $E_WARNING, $nexus);
1239 }
08c259f3 1240 # Special case: old storport driver
1241 if (!blacklisted('ctrl_stdr', $id) && defined $minstdr) {
1242 chomp $stdr;
1243 my $msg = sprintf q{Controller %d [%s]: Storport driver '%s' is out of date},
1244 $id, $name, $stdr;
1245 report('storage', $msg, $E_WARNING, $nexus);
1246 }
669797e1 1247 # Ok
1248 if ($status eq 'Ok' or ($status eq 'Non-Critical'
babe647a 1249 and (defined $minfw or defined $mindr or defined $minstdr))) {
98b224a3 1250 my $msg = sprintf 'Controller %d [%s] is %s',
669797e1 1251 $id, $name, $state;
1252 report('storage', $msg, $E_OK, $nexus);
1253 }
1254 # Default
1255 else {
98b224a3 1256 my $msg = sprintf 'Controller %d [%s] needs attention: %s',
669797e1 1257 $id, $name, $state;
1258 report('storage', $msg, $status2nagios{$status}, $nexus);
1259 }
1260 }
1261 return;
1262}
1263
1264
1265#-----------------------------------------
1266# STORAGE: Check physical drives
1267#-----------------------------------------
1268sub check_physical_disks {
1269 return if $#controllers == -1;
c8eb5019 1270 return if blacklisted('pdisk', 'all');
669797e1 1271
1272 my $id = undef;
1273 my $nexus = undef;
1274 my $name = undef;
1275 my $state = undef;
1276 my $status = undef;
1277 my $fpred = undef;
1278 my $progr = undef;
1279 my $ctrl = undef;
1280 my $vendor = undef; # disk vendor
1281 my $product = undef; # product ID
1282 my $capacity = undef; # disk length (size) in bytes
1283 my @output = ();
1284
1285 if ($snmp) {
1286 my %pdisk_oid
1287 = (
1288 '1.3.6.1.4.1.674.10893.1.20.130.4.1.1' => 'arrayDiskNumber',
1289 '1.3.6.1.4.1.674.10893.1.20.130.4.1.2' => 'arrayDiskName',
1290 '1.3.6.1.4.1.674.10893.1.20.130.4.1.3' => 'arrayDiskVendor',
1291 '1.3.6.1.4.1.674.10893.1.20.130.4.1.4' => 'arrayDiskState',
1292 '1.3.6.1.4.1.674.10893.1.20.130.4.1.6' => 'arrayDiskProductID',
1293 '1.3.6.1.4.1.674.10893.1.20.130.4.1.9' => 'arrayDiskEnclosureID',
1294 '1.3.6.1.4.1.674.10893.1.20.130.4.1.10' => 'arrayDiskChannel',
1295 '1.3.6.1.4.1.674.10893.1.20.130.4.1.11' => 'arrayDiskLengthInMB',
1296 '1.3.6.1.4.1.674.10893.1.20.130.4.1.15' => 'arrayDiskTargetID',
1297 '1.3.6.1.4.1.674.10893.1.20.130.4.1.16' => 'arrayDiskLunID',
1298 '1.3.6.1.4.1.674.10893.1.20.130.4.1.24' => 'arrayDiskComponentStatus',
1299 '1.3.6.1.4.1.674.10893.1.20.130.4.1.26' => 'arrayDiskNexusID',
1300 '1.3.6.1.4.1.674.10893.1.20.130.4.1.31' => 'arrayDiskSmartAlertIndication',
669797e1 1301 '1.3.6.1.4.1.674.10893.1.20.130.5.1.7' => 'arrayDiskEnclosureConnectionControllerNumber',
c11849d6 1302 '1.3.6.1.4.1.674.10893.1.20.130.6.1.7' => 'arrayDiskChannelConnectionControllerNumber',
669797e1 1303 );
4cabd748 1304 my $result = undef;
1305 if ($opt{use_get_table}) {
1306 my $arrayDiskTable = '1.3.6.1.4.1.674.10893.1.20.130.4';
1307 my $arrayDiskEnclosureConnectionControllerNumber = '1.3.6.1.4.1.674.10893.1.20.130.5.1.7';
1308 my $arrayDiskChannelConnectionControllerNumber = '1.3.6.1.4.1.674.10893.1.20.130.6.1.7';
1309
1310 $result = $snmp_session->get_table(-baseoid => $arrayDiskTable);
1311 my $ext1 = $snmp_session->get_table(-baseoid => $arrayDiskEnclosureConnectionControllerNumber);
1312 my $ext2 = $snmp_session->get_table(-baseoid => $arrayDiskChannelConnectionControllerNumber);
1313
1314 if (defined $result) {
1315 defined $ext1 && map { $$result{$_} = $$ext1{$_} } keys %{ $ext1 };
1316 defined $ext2 && map { $$result{$_} = $$ext2{$_} } keys %{ $ext2 };
1317 }
1318 }
1319 else {
1320 $result = $snmp_session->get_entries(-columns => [keys %pdisk_oid]);
1321 }
669797e1 1322
1323 if (!defined $result) {
98b224a3 1324 printf "SNMP ERROR [storage / pdisk]: %s.\n", $snmp_session->error;
669797e1 1325 $snmp_session->close;
1326 exit $E_UNKNOWN;
1327 }
1328
1329 @output = @{ get_snmp_output($result, \%pdisk_oid) };
1330 }
1331 else {
1332 foreach my $c (@controllers) {
74177368 1333 # This blacklists disks with broken firmware, which includes
1334 # illegal XML characters that makes openmanage choke on itself
1335 next if blacklisted('ctrl_pdisk', $c);
1336
669797e1 1337 push @output, @{ run_omreport("storage pdisk controller=$c") };
1338 map_item('ctrl', $c, \@output);
1339 }
1340 }
1341
1342 my %pdisk_state
1343 = (
1344 0 => 'Unknown',
1345 1 => 'Ready',
1346 2 => 'Failed',
1347 3 => 'Online',
1348 4 => 'Offline',
1349 6 => 'Degraded',
1350 7 => 'Recovering',
1351 11 => 'Removed',
1352 15 => 'Resynching',
9113fb39 1353 22 => 'Replacing', # FIXME: this one is not defined in the OM 6.2.0.1 MIBs
669797e1 1354 24 => 'Rebuilding',
1355 25 => 'No Media',
1356 26 => 'Formatting',
1357 28 => 'Diagnostics',
1358 34 => 'Predictive failure',
1359 35 => 'Initializing',
1360 39 => 'Foreign',
1361 40 => 'Clear',
1362 41 => 'Unsupported',
1363 53 => 'Incompatible',
1364 );
1365
1366 # Check physical disks on each of the controllers
1367 PDISK:
1368 foreach my $out (@output) {
1369 if ($snmp) {
1370 $name = $out->{arrayDiskName};
07d224b2 1371 if (exists $out->{arrayDiskEnclosureID}) {
669797e1 1372 $id = join q{:}, ($out->{arrayDiskChannel}, $out->{arrayDiskEnclosureID},
07d224b2 1373 $out->{arrayDiskTargetID});
669797e1 1374 }
1375 else {
1376 $id = join q{:}, ($out->{arrayDiskChannel}, $out->{arrayDiskTargetID});
1377 }
912d8679 1378 $state = get_hashval($out->{arrayDiskState}, \%pdisk_state);
1379 $status = $snmp_status{$out->{arrayDiskComponentStatus}};
669797e1 1380 $fpred = $out->{arrayDiskSmartAlertIndication} == 2 ? 1 : 0;
1381 $progr = q{};
669797e1 1382 $nexus = convert_nexus($out->{arrayDiskNexusID});
1383 $vendor = $out->{arrayDiskVendor};
1384 $product = $out->{arrayDiskProductID};
1385 $capacity = $out->{arrayDiskLengthInMB} * 1024**2;
c11849d6 1386 if (exists $out->{arrayDiskEnclosureConnectionControllerNumber}) {
1387 $ctrl = $out->{arrayDiskEnclosureConnectionControllerNumber} - 1;
1388 }
1389 elsif (exists $out->{arrayDiskChannelConnectionControllerNumber}) {
1390 $ctrl = $out->{arrayDiskChannelConnectionControllerNumber} - 1;
1391 }
1392 else {
1393 $ctrl = -1;
1394 }
669797e1 1395 }
1396 else {
1397 $id = $out->{'ID'};
1398 $name = $out->{'Name'};
1399 $state = $out->{'State'};
1400 $status = $out->{'Status'};
1401 $fpred = lc($out->{'Failure Predicted'}) eq 'yes' ? 1 : 0;
1402 $progr = ' [' . $out->{'Progress'} . ']';
1403 $ctrl = $out->{'ctrl'};
1404 $nexus = join q{:}, $out->{ctrl}, $id;
1405 $vendor = $out->{'Vendor ID'};
1406 $product = $out->{'Product ID'};
1407 $capacity = $out->{'Capacity'};
1408 $capacity =~ s{\A .*? \((\d+) \s bytes\) \z}{$1}xms;
1409 }
1410
1411 next PDISK if blacklisted('pdisk', $nexus);
1412 $count{pdisk}++;
1413
1414 $vendor =~ s{\s+\z}{}xms; # remove trailing whitespace
1415 $product =~ s{\s+\z}{}xms; # remove trailing whitespace
1416
1417 # Calculate human readable capacity
1418 $capacity = ceil($capacity / 1000**3) >= 1000
49bf41a5 1419 ? sprintf '%.1fTB', ($capacity / 1000**4)
1420 : sprintf '%.0fGB', ($capacity / 1000**3);
1421 $capacity = '450GB' if $capacity eq '449GB'; # quick fix for 450GB disks
1422 $capacity = '300GB' if $capacity eq '299GB'; # quick fix for 300GB disks
1423 $capacity = '146GB' if $capacity eq '147GB'; # quick fix for 146GB disks
669797e1 1424
1425 # Capitalize only the first letter of the vendor name
1426 $vendor = (substr $vendor, 0, 1) . lc (substr $vendor, 1, length $vendor);
1427
1428 # Remove unnecessary trademark rubbish from vendor name
1429 $vendor =~ s{\(tm\)\z}{}xms;
1430
a8b24907 1431 # Special case: Failure predicted
1432 if ($fpred) {
ea0b94b8 1433 my $msg = sprintf '%s [%s %s, %s] on ctrl %d needs attention: Failure Predicted',
1434 $name, $vendor, $product, $capacity, $ctrl;
f2f69da2 1435 $msg .= " ($state)" if $state ne 'Predictive failure';
1436 report('storage', $msg,
1437 ($status2nagios{$status} == $E_CRITICAL ? $E_CRITICAL : $E_WARNING), $nexus);
ea0b94b8 1438 }
c5c69973 1439 # Special case: Rebuilding / Replacing
ea0b94b8 1440 elsif ($state =~ m{\A Rebuilding|Replacing \z}xms) {
ddeae63c 1441 my $msg = sprintf '%s [%s %s, %s] on ctrl %d is %s%s',
1442 $name, $vendor, $product, $capacity, $ctrl, $state, $progr;
669797e1 1443 report('storage', $msg, $E_WARNING, $nexus);
1444 }
1445 # Default
1446 elsif ($status ne 'Ok') {
c11849d6 1447 my $msg = sprintf '%s [%s %s, %s] on ctrl %d needs attention: %s',
1448 $name, $vendor, $product, $capacity, $ctrl, $state;
669797e1 1449 report('storage', $msg, $status2nagios{$status}, $nexus);
1450 }
1451 # Ok
1452 else {
c11849d6 1453 my $msg = sprintf '%s [%s] on ctrl %d is %s',
1454 $name, $capacity, $ctrl, $state;
669797e1 1455 report('storage', $msg, $E_OK, $nexus);
1456 }
1457 }
1458 return;
1459}
1460
1461
1462#-----------------------------------------
1463# STORAGE: Check logical drives
1464#-----------------------------------------
1465sub check_virtual_disks {
1466 return if $#controllers == -1;
5ecf578c 1467 return if blacklisted('vdisk', 'all');
669797e1 1468
1469 my $id = undef;
25d04c34 1470 my $name = undef;
669797e1 1471 my $nexus = undef;
1472 my $dev = undef;
1473 my $state = undef;
1474 my $status = undef;
1475 my $layout = undef;
1476 my $size = undef;
1477 my $progr = undef;
25d04c34 1478 my $ctrl = undef;
669797e1 1479 my @output = ();
1480
1481 if ($snmp) {
1482 my %vdisk_oid
1483 = (
669797e1 1484 '1.3.6.1.4.1.674.10893.1.20.140.1.1.3' => 'virtualDiskDeviceName',
1485 '1.3.6.1.4.1.674.10893.1.20.140.1.1.4' => 'virtualDiskState',
1486 '1.3.6.1.4.1.674.10893.1.20.140.1.1.6' => 'virtualDiskLengthInMB',
1487 '1.3.6.1.4.1.674.10893.1.20.140.1.1.13' => 'virtualDiskLayout',
25d04c34 1488 '1.3.6.1.4.1.674.10893.1.20.140.1.1.17' => 'virtualDiskTargetID',
669797e1 1489 '1.3.6.1.4.1.674.10893.1.20.140.1.1.20' => 'virtualDiskComponentStatus',
1490 '1.3.6.1.4.1.674.10893.1.20.140.1.1.21' => 'virtualDiskNexusID',
1491 );
4cabd748 1492 my $result = undef;
1493 if ($opt{use_get_table}) {
1494 my $virtualDiskTable = '1.3.6.1.4.1.674.10893.1.20.140.1';
1495 $result = $snmp_session->get_table(-baseoid => $virtualDiskTable);
1496 }
1497 else {
1498 $result = $snmp_session->get_entries(-columns => [keys %vdisk_oid]);
1499 }
669797e1 1500
1501 # No logical drives is OK
1502 return if !defined $result;
1503
1504 @output = @{ get_snmp_output($result, \%vdisk_oid) };
1505 }
1506 else {
1507 foreach my $c (@controllers) {
1508 push @output, @{ run_omreport("storage vdisk controller=$c") };
1509 map_item('ctrl', $c, \@output);
1510 }
1511 }
1512
1513 my %vdisk_state
1514 = (
1515 0 => 'Unknown',
1516 1 => 'Ready',
1517 2 => 'Failed',
1518 3 => 'Online',
1519 4 => 'Offline',
1520 6 => 'Degraded',
1521 15 => 'Resynching',
1522 16 => 'Regenerating',
1523 24 => 'Rebuilding',
1524 26 => 'Formatting',
1525 32 => 'Reconstructing',
1526 35 => 'Initializing',
1527 36 => 'Background Initialization',
1528 38 => 'Resynching Paused',
1529 52 => 'Permanently Degraded',
1530 54 => 'Degraded Redundancy',
1531 );
1532
1533 my %vdisk_layout
1534 = (
1535 1 => 'Concatenated',
1536 2 => 'RAID-0',
1537 3 => 'RAID-1',
1538 7 => 'RAID-5',
1539 8 => 'RAID-6',
1540 10 => 'RAID-10',
1541 12 => 'RAID-50',
9113fb39 1542 19 => 'Concatenated RAID-1',
669797e1 1543 24 => 'RAID-60',
1544 );
1545
1546 # Check virtual disks on each of the controllers
1547 VDISK:
1548 foreach my $out (@output) {
1549 if ($snmp) {
25d04c34 1550 $id = $out->{virtualDiskTargetID};
669797e1 1551 $dev = $out->{virtualDiskDeviceName};
912d8679 1552 $state = get_hashval($out->{virtualDiskState}, \%vdisk_state);
1553 $layout = get_hashval($out->{virtualDiskLayout}, \%vdisk_layout);
669797e1 1554 $status = $snmp_status{$out->{virtualDiskComponentStatus}};
669797e1 1555 $size = sprintf '%.2f GB', $out->{virtualDiskLengthInMB} / 1024;
1556 $progr = q{}; # can't get this from SNMP(?)
1557 $nexus = convert_nexus($out->{virtualDiskNexusID});
25d04c34 1558 $ctrl = $nexus; # We use the nexus id to get the controller id
1559 $ctrl =~ s{\A (\d+):\d+ \z}{$1}xms;
669797e1 1560 }
1561 else {
1562 $id = $out->{ID};
1563 $dev = $out->{'Device Name'};
1564 $state = $out->{State};
1565 $status = $out->{Status};
1566 $layout = $out->{Layout};
1567 $size = $out->{Size};
1568 $progr = ' [' . $out->{Progress} . ']';
1569 $size =~ s{\A (.*GB).* \z}{$1}xms;
1570 $nexus = join q{:}, $out->{ctrl}, $id;
25d04c34 1571 $ctrl = $out->{ctrl};
669797e1 1572 }
1573
1574 next VDISK if blacklisted('vdisk', $nexus);
1575 $count{vdisk}++;
1576
04b0f13b 1577 # The device name is undefined sometimes
1578 $dev = q{} if !defined $dev;
1579
669797e1 1580 # Special case: Regenerating
1581 if ($state eq 'Regenerating') {
98b224a3 1582 my $msg = sprintf q{Logical drive %d '%s' [%s, %s] on ctrl %d is %s%s},
1583 $id, $dev, $layout, $size, $ctrl, $state, $progr;
669797e1 1584 report('storage', $msg, $E_WARNING, $nexus);
1585 }
1586 # Default
1587 elsif ($status ne 'Ok') {
98b224a3 1588 my $msg = sprintf q{Logical drive %d '%s' [%s, %s] on ctrl %d needs attention: %s},
1589 $id, $dev, $layout, $size, $ctrl, $state;
669797e1 1590 report('storage', $msg, $status2nagios{$status}, $nexus);
1591 }
1592 # Ok
1593 else {
98b224a3 1594 my $msg = sprintf q{Logical drive %d '%s' [%s, %s] on ctrl %d is %s},
1595 $id, $dev, $layout, $size, $ctrl, $state;
669797e1 1596 report('storage', $msg, $E_OK, $nexus);
1597 }
1598 }
1599 return;
1600}
1601
1602
1603#-----------------------------------------
1604# STORAGE: Check cache batteries
1605#-----------------------------------------
1606sub check_cache_battery {
1607 return if $#controllers == -1;
5ecf578c 1608 return if blacklisted('bat', 'all');
669797e1 1609
1610 my $id = undef;
1611 my $nexus = undef;
1612 my $state = undef;
1613 my $status = undef;
1614 my $ctrl = undef;
1615 my $learn = undef; # learn state
1616 my $pred = undef; # battery's ability to be charged
1617 my @output = ();
1618
1619 if ($snmp) {
1620 my %bat_oid
1621 = (
669797e1 1622 '1.3.6.1.4.1.674.10893.1.20.130.15.1.4' => 'batteryState',
1623 '1.3.6.1.4.1.674.10893.1.20.130.15.1.6' => 'batteryComponentStatus',
1624 '1.3.6.1.4.1.674.10893.1.20.130.15.1.9' => 'batteryNexusID',
1625 '1.3.6.1.4.1.674.10893.1.20.130.15.1.10' => 'batteryPredictedCapacity',
1626 '1.3.6.1.4.1.674.10893.1.20.130.15.1.12' => 'batteryLearnState',
1627 '1.3.6.1.4.1.674.10893.1.20.130.16.1.5' => 'batteryConnectionControllerNumber',
1628 );
4cabd748 1629 my $result = undef;
1630 if ($opt{use_get_table}) {
1631 my $batteryTable = '1.3.6.1.4.1.674.10893.1.20.130.15';
c849fd4c 1632 my $batteryConnectionTable = '1.3.6.1.4.1.674.10893.1.20.130.16';
1633
4cabd748 1634 $result = $snmp_session->get_table(-baseoid => $batteryTable);
c849fd4c 1635 my $ext = $snmp_session->get_table(-baseoid => $batteryConnectionTable);
1636
1637 if (defined $result) {
1638 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
1639 }
4cabd748 1640 }
1641 else {
1642 $result = $snmp_session->get_entries(-columns => [keys %bat_oid]);
1643 }
669797e1 1644
1645 # No cache battery is OK
1646 return if !defined $result;
1647
1648 @output = @{ get_snmp_output($result, \%bat_oid) };
1649 }
1650 else {
1651 foreach my $c (@controllers) {
1652 push @output, @{ run_omreport("storage battery controller=$c") };
1653 map_item('ctrl', $c, \@output);
1654 }
1655 }
1656
1657 my %bat_state
1658 = (
1659 0 => 'Unknown',
1660 1 => 'Ready',
1661 2 => 'Failed',
1662 6 => 'Degraded',
1663 7 => 'Reconditioning',
1664 9 => 'High',
1665 10 => 'Power Low',
1666 12 => 'Charging',
1667 21 => 'Missing',
1668 36 => 'Learning',
1669 );
1670
a49bcfe8 1671 # Specifies the learn state activity of the battery
669797e1 1672 my %bat_learn_state
1673 = (
1674 1 => 'Failed',
1675 2 => 'Active',
1676 4 => 'Timed out',
1677 8 => 'Requested',
1678 16 => 'Idle',
1679 );
1680
a49bcfe8 1681 # This property displays the battery's ability to be charged
669797e1 1682 my %bat_pred_cap
1683 = (
1684 1 => 'Failed', # The battery cannot be charged and needs to be replaced
1685 2 => 'Ready', # The battery can be charged to full capacity
1686 4 => 'Unknown', # The battery is completing a Learn cycle. The charge capacity of the
1687 # battery cannot be determined until the Learn cycle is complete
1688 );
1689
1690 # Check battery on each of the controllers
1691 BATTERY:
1692 foreach my $out (@output) {
1693 if ($snmp) {
669797e1 1694 $status = $snmp_status{$out->{batteryComponentStatus}};
912d8679 1695 $state = get_hashval($out->{batteryState}, \%bat_state);
1696 $learn = get_hashval($out->{batteryLearnState}, \%bat_learn_state);
1697 $pred = get_hashval($out->{batteryPredictedCapacity}, \%bat_pred_cap);
669797e1 1698 $ctrl = $out->{batteryConnectionControllerNumber} - 1;
1699 $nexus = convert_nexus($out->{batteryNexusID});
25d04c34 1700 $id = $nexus;
1701 $id =~ s{\A \d+:(\d+) \z}{$1}xms;
669797e1 1702 }
1703 else {
1704 $id = $out->{'ID'};
1705 $state = $out->{'State'};
1706 $status = $out->{'Status'};
1707 $learn = $out->{'Learn State'};
1708 $pred = $out->{'Predicted Capacity Status'};
1709 $ctrl = $out->{'ctrl'};
1710 $nexus = join q{:}, $out->{ctrl}, $id;
1711 }
1712
1713 next BATTERY if blacklisted('bat', $nexus);
1714
1715 # Special case: Charging
1716 if ($state eq 'Charging') {
50d6bc4a 1717 if ($pred eq 'Failed') {
1718 my $msg = sprintf 'Cache battery %d in controller %d is %s (%s) [replace battery]',
1719 $id, $ctrl, $state, $pred;
1720 report('storage', $msg, $E_CRITICAL, $nexus);
1721 }
1722 else {
1723 next BATTERY if blacklisted('bat_charge', $nexus);
1724 my $msg = sprintf 'Cache battery %d in controller %d is %s (%s) [probably harmless]',
1725 $id, $ctrl, $state, $pred;
1726 report('storage', $msg, $E_WARNING, $nexus);
1727 }
669797e1 1728 }
1729 # Special case: Learning (battery learns its capacity)
1730 elsif ($state eq 'Learning') {
50d6bc4a 1731 if ($learn eq 'Failed') {
1732 my $msg = sprintf 'Cache battery %d in controller %d is %s (%s)',
1733 $id, $ctrl, $state, $learn;
1734 report('storage', $msg, $E_CRITICAL, $nexus);
1735 }
1736 else {
1737 next BATTERY if blacklisted('bat_charge', $nexus);
1738 my $msg = sprintf 'Cache battery %d in controller %d is %s (%s) [probably harmless]',
1739 $id, $ctrl, $state, $learn;
1740 report('storage', $msg, $E_WARNING, $nexus);
1741 }
669797e1 1742 }
1743 # Special case: Power Low (first part of recharge cycle)
1744 elsif ($state eq 'Power Low') {
5a28cf7f 1745 next BATTERY if blacklisted('bat_charge', $nexus);
669797e1 1746 my $msg = sprintf 'Cache battery %d in controller %d is %s [probably harmless]',
1747 $id, $ctrl, $state;
1748 report('storage', $msg, $E_WARNING, $nexus);
1749 }
5a28cf7f 1750 # Special case: Degraded and Non-Critical (usually part of recharge cycle)
1751 elsif ($state eq 'Degraded' && $status eq 'Non-Critical') {
1752 next BATTERY if blacklisted('bat_charge', $nexus);
1753 my $msg = sprintf 'Cache battery %d in controller %d is %s (%s) [probably harmless]',
1754 $id, $ctrl, $state, $status;
1755 report('storage', $msg, $E_WARNING, $nexus);
1756 }
669797e1 1757 # Default
1758 elsif ($status ne 'Ok') {
1759 my $msg = sprintf 'Cache battery %d in controller %d needs attention: %s (%s)',
1760 $id, $ctrl, $state, $status;
1761 report('storage', $msg, $status2nagios{$status}, $nexus);
1762 }
1763 # Ok
1764 else {
1765 my $msg = sprintf 'Cache battery %d in controller %d is %s',
1766 $id, $ctrl, $state;
1767 report('storage', $msg, $E_OK, $nexus);
1768 }
1769 }
1770 return;
1771}
1772
1773
1774#-----------------------------------------
1775# STORAGE: Check connectors (channels)
1776#-----------------------------------------
1777sub check_connectors {
1778 return if $#controllers == -1;
5ecf578c 1779 return if blacklisted('conn', 'all');
669797e1 1780
1781 my $id = undef;
1782 my $nexus = undef;
1783 my $name = undef;
1784 my $state = undef;
1785 my $status = undef;
1786 my $type = undef;
1787 my $ctrl = undef;
1788 my @output = ();
1789
1790 if ($snmp) {
1791 my %conn_oid
1792 = (
1793 '1.3.6.1.4.1.674.10893.1.20.130.2.1.1' => 'channelNumber',
1794 '1.3.6.1.4.1.674.10893.1.20.130.2.1.2' => 'channelName',
1795 '1.3.6.1.4.1.674.10893.1.20.130.2.1.3' => 'channelState',
1796 '1.3.6.1.4.1.674.10893.1.20.130.2.1.8' => 'channelComponentStatus',
1797 '1.3.6.1.4.1.674.10893.1.20.130.2.1.9' => 'channelNexusID',
1798 '1.3.6.1.4.1.674.10893.1.20.130.2.1.11' => 'channelBusType',
1799 );
4cabd748 1800 my $result = undef;
1801 if ($opt{use_get_table}) {
1802 my $channelTable = '1.3.6.1.4.1.674.10893.1.20.130.2';
1803 $result = $snmp_session->get_table(-baseoid => $channelTable);
1804 }
1805 else {
1806 $result = $snmp_session->get_entries(-columns => [keys %conn_oid]);
1807 }
669797e1 1808
1809 if (!defined $result) {
98b224a3 1810 printf "SNMP ERROR [storage / channel]: %s.\n", $snmp_session->error;
669797e1 1811 $snmp_session->close;
1812 exit $E_UNKNOWN;
1813 }
1814
1815 @output = @{ get_snmp_output($result, \%conn_oid) };
1816 }
1817 else {
1818 foreach my $c (@controllers) {
1819 push @output, @{ run_omreport("storage connector controller=$c") };
1820 map_item('ctrl', $c, \@output);
1821 }
1822 }
1823
1824 my %conn_state
1825 = (
1826 0 => 'Unknown',
1827 1 => 'Ready',
1828 2 => 'Failed',
1829 3 => 'Online',
1830 4 => 'Offline',
1831 6 => 'Degraded',
1832 );
1833
1834 my %conn_bustype
1835 = (
1836 1 => 'SCSI',
1837 2 => 'IDE',
1838 3 => 'Fibre Channel',
1839 4 => 'SSA',
1840 6 => 'USB',
1841 7 => 'SATA',
1842 8 => 'SAS',
1843 );
1844
1845 # Check connectors on each of the controllers
1846 CHANNEL:
1847 foreach my $out (@output) {
1848 if ($snmp) {
1849 $id = $out->{channelNumber} - 1;
1850 $name = $out->{channelName};
669797e1 1851 $status = $snmp_status{$out->{channelComponentStatus}};
912d8679 1852 $state = get_hashval($out->{channelState}, \%conn_state);
1853 $type = get_hashval($out->{channelBusType}, \%conn_bustype);
669797e1 1854 $nexus = convert_nexus($out->{channelNexusID});
1855 $ctrl = $nexus;
1856 $ctrl =~ s{(\d+):\d+}{$1}xms;
1857 }
1858 else {
1859 $id = $out->{'ID'};
1860 $name = $out->{'Name'};
1861 $state = $out->{'State'};
1862 $status = $out->{'Status'};
1863 $type = $out->{'Connector Type'};
1864 $ctrl = $out->{ctrl};
1865 $nexus = join q{:}, $out->{ctrl}, $id;
1866 }
1867
1868 next CHANNEL if blacklisted('conn', $nexus);
1869
98b224a3 1870 my $msg = sprintf '%s [%s] on controller %d is %s',
669797e1 1871 $name, $type, $ctrl, $state;
1872 report('storage', $msg, $status2nagios{$status}, $nexus);
1873 }
1874 return;
1875}
1876
1877
1878#-----------------------------------------
1879# STORAGE: Check enclosures
1880#-----------------------------------------
1881sub check_enclosures {
5ecf578c 1882 return if blacklisted('encl', 'all');
1883
669797e1 1884 my $id = undef;
1885 my $nexus = undef;
1886 my $name = undef;
1887 my $state = undef;
1888 my $status = undef;
1889 my $firmware = undef;
25d04c34 1890 my $ctrl = undef;
669797e1 1891 my @output = ();
1892
1893 if ($snmp) {
1894 my %encl_oid
1895 = (
1896 '1.3.6.1.4.1.674.10893.1.20.130.3.1.1' => 'enclosureNumber',
1897 '1.3.6.1.4.1.674.10893.1.20.130.3.1.2' => 'enclosureName',
1898 '1.3.6.1.4.1.674.10893.1.20.130.3.1.4' => 'enclosureState',
1899 '1.3.6.1.4.1.674.10893.1.20.130.3.1.19' => 'enclosureChannelNumber',
1900 '1.3.6.1.4.1.674.10893.1.20.130.3.1.24' => 'enclosureComponentStatus',
1901 '1.3.6.1.4.1.674.10893.1.20.130.3.1.25' => 'enclosureNexusID',
1902 '1.3.6.1.4.1.674.10893.1.20.130.3.1.26' => 'enclosureFirmwareVersion',
1903 );
4cabd748 1904 my $result = undef;
1905 if ($opt{use_get_table}) {
1906 my $enclosureTable = '1.3.6.1.4.1.674.10893.1.20.130.3';
1907 $result = $snmp_session->get_table(-baseoid => $enclosureTable);
1908 }
1909 else {
1910 $result = $snmp_session->get_entries(-columns => [keys %encl_oid]);
1911 }
669797e1 1912
1913 # No enclosures is OK
1914 return if !defined $result;
1915
1916 @output = @{ get_snmp_output($result, \%encl_oid) };
1917 }
1918 else {
1919 foreach my $c (@controllers) {
1920 push @output, @{ run_omreport("storage enclosure controller=$c") };
1921 map_item('ctrl', $c, \@output);
1922 }
1923 }
1924
1925 my %encl_state
1926 = (
1927 0 => 'Unknown',
1928 1 => 'Ready',
1929 2 => 'Failed',
1930 3 => 'Online',
1931 4 => 'Offline',
1932 6 => 'Degraded',
1933 );
1934
1935 ENCLOSURE:
1936 foreach my $out (@output) {
1937 if ($snmp) {
912d8679 1938 $id = $out->{enclosureNumber} - 1;
1939 $name = $out->{enclosureName};
1940 $state = get_hashval($out->{enclosureState}, \%encl_state);
1941 $status = $snmp_status{$out->{enclosureComponentStatus}};
669797e1 1942 $firmware = exists $out->{enclosureFirmwareVersion}
1943 ? $out->{enclosureFirmwareVersion} : 'N/A';
1944 $nexus = convert_nexus($out->{enclosureNexusID});
25d04c34 1945 $ctrl = $nexus;
1946 $ctrl =~ s{\A (\d+):.* \z}{$1}xms;
669797e1 1947 }
1948 else {
1949 $id = $out->{ID};
1950 $name = $out->{Name};
1951 $state = $out->{State};
1952 $status = $out->{Status};
1953 $firmware = $out->{'Firmware Version'} ne 'Not Applicable'
1954 ? $out->{'Firmware Version'} : 'N/A';
1955 $nexus = join q{:}, $out->{ctrl}, $id;
25d04c34 1956 $ctrl = $out->{ctrl};
669797e1 1957 }
1958
1959 $name =~ s{\s+\z}{}xms; # remove trailing whitespace
1960 $firmware =~ s{\s+\z}{}xms; # remove trailing whitespace
1961
1962 # store enclosure data for future use
1963 push @enclosures, { 'id' => $id,
1964 'ctrl' => $out->{ctrl},
1965 'name' => $name };
1966
1967 # Collecting some storage info
1968 $sysinfo{'enclosure'}{$nexus}{'id'} = $nexus;
1969 $sysinfo{'enclosure'}{$nexus}{'name'} = $name;
1970 $sysinfo{'enclosure'}{$nexus}{'firmware'} = $firmware;
1971
1972 next ENCLOSURE if blacklisted('encl', $nexus);
1973
98b224a3 1974 my $msg = sprintf 'Enclosure %s [%s] on controller %d is %s',
25d04c34 1975 $nexus, $name, $ctrl, $state;
669797e1 1976 report('storage', $msg, $status2nagios{$status}, $nexus);
1977 }
1978 return;
1979}
1980
1981
1982#-----------------------------------------
1983# STORAGE: Check enclosure fans
1984#-----------------------------------------
1985sub check_enclosure_fans {
1986 return if $#controllers == -1;
5ecf578c 1987 return if blacklisted('encl_fan', 'all');
669797e1 1988
1989 my $id = undef;
1990 my $nexus = undef;
1991 my $name = undef;
1992 my $state = undef;
1993 my $status = undef;
1994 my $speed = undef;
1995 my $encl_id = undef;
1996 my $encl_name = undef;
1997 my @output = ();
1998
1999 if ($snmp) {
2000 my %fan_oid
2001 = (
2002 '1.3.6.1.4.1.674.10893.1.20.130.7.1.1' => 'fanNumber',
2003 '1.3.6.1.4.1.674.10893.1.20.130.7.1.2' => 'fanName',
2004 '1.3.6.1.4.1.674.10893.1.20.130.7.1.4' => 'fanState',
2005 '1.3.6.1.4.1.674.10893.1.20.130.7.1.11' => 'fanProbeCurrValue',
2006 '1.3.6.1.4.1.674.10893.1.20.130.7.1.15' => 'fanComponentStatus',
2007 '1.3.6.1.4.1.674.10893.1.20.130.7.1.16' => 'fanNexusID',
2008 '1.3.6.1.4.1.674.10893.1.20.130.8.1.4' => 'fanConnectionEnclosureName',
2009 '1.3.6.1.4.1.674.10893.1.20.130.8.1.5' => 'fanConnectionEnclosureNumber',
2010 );
4cabd748 2011 my $result = undef;
2012 if ($opt{use_get_table}) {
2013 my $fanTable = '1.3.6.1.4.1.674.10893.1.20.130.7';
c849fd4c 2014 my $fanConnectionTable = '1.3.6.1.4.1.674.10893.1.20.130.8';
2015
4cabd748 2016 $result = $snmp_session->get_table(-baseoid => $fanTable);
c849fd4c 2017 my $ext = $snmp_session->get_table(-baseoid => $fanConnectionTable);
2018
2019 if (defined $result) {
2020 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
2021 }
4cabd748 2022 }
2023 else {
2024 $result = $snmp_session->get_entries(-columns => [keys %fan_oid]);
2025 }
669797e1 2026
2027 # No enclosure fans is OK
2028 return if !defined $result;
2029
2030 @output = @{ get_snmp_output($result, \%fan_oid) };
2031 }
2032 else {
2033 foreach my $enc (@enclosures) {
2034 push @output, @{ run_omreport("storage enclosure controller=$enc->{ctrl} enclosure=$enc->{id} info=fans") };
2035 map_item('ctrl', $enc->{ctrl}, \@output);
2036 map_item('encl_id', $enc->{id}, \@output);
2037 map_item('encl_name', $enc->{name}, \@output);
2038 }
2039 }
2040
2041 my %fan_state
2042 = (
2043 0 => 'Unknown',
2044 1 => 'Ready',
2045 2 => 'Failed',
2046 3 => 'Online',
2047 4 => 'Offline',
2048 6 => 'Degraded',
2049 21 => 'Missing',
2050 );
2051
2052 # Check fans on each of the enclosures
2053 FAN:
2054 foreach my $out (@output) {
2055 if ($snmp) {
2056 $id = $out->{fanNumber} - 1;
2057 $name = $out->{fanName};
912d8679 2058 $state = get_hashval($out->{fanState}, \%fan_state);
669797e1 2059 $status = $snmp_status{$out->{fanComponentStatus}};
2060 $speed = $out->{fanProbeCurrValue};
2061 $encl_id = $out->{fanConnectionEnclosureNumber} - 1;
2062 $encl_name = $out->{fanConnectionEnclosureName};
2063 $nexus = convert_nexus($out->{fanNexusID});
2064 }
2065 else {
2066 $id = $out->{'ID'};
2067 $name = $out->{'Name'};
2068 $state = $out->{'State'};
2069 $status = $out->{'Status'};
2070 $speed = $out->{'Speed'};
2071 $encl_id = join q{:}, $out->{ctrl}, $out->{'encl_id'};
2072 $encl_name = $out->{encl_name};
2073 $nexus = join q{:}, $out->{ctrl}, $out->{'encl_id'}, $id;
2074 }
2075
2076 next FAN if blacklisted('encl_fan', $nexus);
2077
2078 # Default
2079 if ($status ne 'Ok') {
98b224a3 2080 my $msg = sprintf '%s in enclosure %s [%s] needs attention: %s',
669797e1 2081 $name, $encl_id, $encl_name, $state;
2082 report('storage', $msg, $status2nagios{$status}, $nexus);
2083 }
2084 # Ok
2085 else {
98b224a3 2086 my $msg = sprintf '%s in enclosure %s [%s] is %s (speed=%s)',
669797e1 2087 $name, $encl_id, $encl_name, $state, $speed;
2088 report('storage', $msg, $E_OK, $nexus);
2089 }
2090 }
2091 return;
2092}
2093
2094
2095#-----------------------------------------
2096# STORAGE: Check enclosure power supplies
2097#-----------------------------------------
2098sub check_enclosure_pwr {
2099 return if $#controllers == -1;
5ecf578c 2100 return if blacklisted('encl_ps', 'all');
669797e1 2101
2102 my $id = undef;
2103 my $nexus = undef;
2104 my $name = undef;
2105 my $state = undef;
2106 my $status = undef;
2107 my $encl_id = undef;
2108 my $encl_name = undef;
2109 my @output = ();
2110
2111 if ($snmp) {
2112 my %ps_oid
2113 = (
2114 '1.3.6.1.4.1.674.10893.1.20.130.9.1.1' => 'powerSupplyNumber',
2115 '1.3.6.1.4.1.674.10893.1.20.130.9.1.2' => 'powerSupplyName',
2116 '1.3.6.1.4.1.674.10893.1.20.130.9.1.4' => 'powerSupplyState',
2117 '1.3.6.1.4.1.674.10893.1.20.130.9.1.9' => 'powerSupplyComponentStatus',
2118 '1.3.6.1.4.1.674.10893.1.20.130.9.1.10' => 'powerSupplyNexusID',
2119 '1.3.6.1.4.1.674.10893.1.20.130.10.1.4' => 'powerSupplyConnectionEnclosureName',
2120 '1.3.6.1.4.1.674.10893.1.20.130.10.1.5' => 'powerSupplyConnectionEnclosureNumber',
2121 );
4cabd748 2122 my $result = undef;
2123 if ($opt{use_get_table}) {
2124 my $powerSupplyTable = '1.3.6.1.4.1.674.10893.1.20.130.9';
c849fd4c 2125 my $powerSupplyConnectionTable = '1.3.6.1.4.1.674.10893.1.20.130.10';
2126
4cabd748 2127 $result = $snmp_session->get_table(-baseoid => $powerSupplyTable);
c849fd4c 2128 my $ext = $snmp_session->get_table(-baseoid => $powerSupplyConnectionTable);
2129
2130 if (defined $result) {
2131 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
2132 }
4cabd748 2133 }
2134 else {
2135 $result = $snmp_session->get_entries(-columns => [keys %ps_oid]);
2136 }
669797e1 2137
2138 # No enclosure power supplies is OK
2139 return if !defined $result;
2140
2141 @output = @{ get_snmp_output($result, \%ps_oid) };
2142 }
2143 else {
2144 foreach my $enc (@enclosures) {
2145 push @output, @{ run_omreport("storage enclosure controller=$enc->{ctrl} enclosure=$enc->{id} info=pwrsupplies") };
2146 map_item('ctrl', $enc->{ctrl}, \@output);
2147 map_item('encl_id', $enc->{id}, \@output);
2148 map_item('encl_name', $enc->{name}, \@output);
2149 }
2150 }
2151
2152 my %ps_state
2153 = (
2154 0 => 'Unknown',
2155 1 => 'Ready',
2156 2 => 'Failed',
2157 5 => 'Not Installed',
2158 6 => 'Degraded',
2159 11 => 'Removed',
2160 21 => 'Missing',
2161 );
2162
2163 # Check power supplies on each of the enclosures
2164 PS:
2165 foreach my $out (@output) {
2166 if ($snmp) {
2167 $id = $out->{powerSupplyNumber};
2168 $name = $out->{powerSupplyName};
912d8679 2169 $state = get_hashval($out->{powerSupplyState}, \%ps_state);
669797e1 2170 $status = $snmp_status{$out->{powerSupplyComponentStatus}};
2171 $encl_id = $out->{powerSupplyConnectionEnclosureNumber} - 1;
2172 $encl_name = $out->{powerSupplyConnectionEnclosureName};
2173 $nexus = convert_nexus($out->{powerSupplyNexusID});
2174 }
2175 else {
2176 $id = $out->{'ID'};
2177 $name = $out->{'Name'};
2178 $state = $out->{'State'};
2179 $status = $out->{'Status'};
2180 $encl_id = join q{:}, $out->{ctrl}, $out->{'encl_id'};
2181 $encl_name = $out->{encl_name};
2182 $nexus = join q{:}, $out->{ctrl}, $out->{'encl_id'}, $id;
2183 }
2184
2185 next PS if blacklisted('encl_ps', $nexus);
2186
2187 # Default
2188 if ($status ne 'Ok') {
98b224a3 2189 my $msg = sprintf '%s in enclosure %s [%s] needs attention: %s',
669797e1 2190 $name, $encl_id, $encl_name, $state;
2191 report('storage', $msg, $status2nagios{$status}, $nexus);
2192 }
2193 # Ok
2194 else {
98b224a3 2195 my $msg = sprintf '%s in enclosure %s [%s] is %s',
669797e1 2196 $name, $encl_id, $encl_name, $state;
2197 report('storage', $msg, $E_OK, $nexus);
2198 }
2199 }
2200 return;
2201}
2202
2203
2204#-----------------------------------------
2205# STORAGE: Check enclosure temperatures
2206#-----------------------------------------
2207sub check_enclosure_temp {
2208 return if $#controllers == -1;
5ecf578c 2209 return if blacklisted('encl_temp', 'all');
669797e1 2210
2211 my $id = undef;
2212 my $nexus = undef;
2213 my $name = undef;
2214 my $state = undef;
2215 my $status = undef;
2216 my $reading = undef;
2217 my $unit = undef;
2218 my $max_warn = undef;
2219 my $max_crit = undef;
2220 my $encl_id = undef;
2221 my $encl_name = undef;
2222 my @output = ();
2223
2224 if ($snmp) {
2225 my %temp_oid
2226 = (
2227 '1.3.6.1.4.1.674.10893.1.20.130.11.1.1' => 'temperatureProbeNumber',
2228 '1.3.6.1.4.1.674.10893.1.20.130.11.1.2' => 'temperatureProbeName',
2229 '1.3.6.1.4.1.674.10893.1.20.130.11.1.4' => 'temperatureProbeState',
2230 '1.3.6.1.4.1.674.10893.1.20.130.11.1.6' => 'temperatureProbeUnit',
2231 '1.3.6.1.4.1.674.10893.1.20.130.11.1.9' => 'temperatureProbeMaxWarning',
2232 '1.3.6.1.4.1.674.10893.1.20.130.11.1.10' => 'temperatureProbeMaxCritical',
2233 '1.3.6.1.4.1.674.10893.1.20.130.11.1.11' => 'temperatureProbeCurValue',
2234 '1.3.6.1.4.1.674.10893.1.20.130.11.1.13' => 'temperatureProbeComponentStatus',
2235 '1.3.6.1.4.1.674.10893.1.20.130.11.1.14' => 'temperatureProbeNexusID',
2236 '1.3.6.1.4.1.674.10893.1.20.130.12.1.4' => 'temperatureConnectionEnclosureName',
2237 '1.3.6.1.4.1.674.10893.1.20.130.12.1.5' => 'temperatureConnectionEnclosureNumber',
2238 );
4cabd748 2239 my $result = undef;
2240 if ($opt{use_get_table}) {
2241 my $temperatureProbeTable = '1.3.6.1.4.1.674.10893.1.20.130.11';
c849fd4c 2242 my $temperatureConnectionTable = '1.3.6.1.4.1.674.10893.1.20.130.12';
2243
4cabd748 2244 $result = $snmp_session->get_table(-baseoid => $temperatureProbeTable);
c849fd4c 2245 my $ext = $snmp_session->get_table(-baseoid => $temperatureConnectionTable);
2246
2247 if (defined $result) {
2248 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
2249 }
4cabd748 2250 }
2251 else {
2252 $result = $snmp_session->get_entries(-columns => [keys %temp_oid]);
2253 }
669797e1 2254
2255 # No enclosure temperature probes is OK
2256 return if !defined $result;
2257
2258 @output = @{ get_snmp_output($result, \%temp_oid) };
2259 }
2260 else {
2261 foreach my $enc (@enclosures) {
2262 push @output, @{ run_omreport("storage enclosure controller=$enc->{ctrl} enclosure=$enc->{id} info=temps") };
2263 map_item('ctrl', $enc->{ctrl}, \@output);
2264 map_item('encl_id', $enc->{id}, \@output);
2265 map_item('encl_name', $enc->{name}, \@output);
2266 }
2267 }
2268
2269 my %temp_state
2270 = (
2271 0 => 'Unknown',
2272 1 => 'Ready',
2273 2 => 'Failed',
2274 4 => 'Offline',
2275 6 => 'Degraded',
2276 9 => 'Inactive',
2277 21 => 'Missing',
2278 );
2279
2280 # Check temperature probes on each of the enclosures
2281 TEMP:
2282 foreach my $out (@output) {
2283 if ($snmp) {
2284 $id = $out->{temperatureProbeNumber} - 1;
2285 $name = $out->{temperatureProbeName};
912d8679 2286 $state = get_hashval($out->{temperatureProbeState}, \%temp_state);
669797e1 2287 $status = $snmp_status{$out->{temperatureProbeComponentStatus}};
2288 $unit = $out->{temperatureProbeUnit};
2289 $reading = $out->{temperatureProbeCurValue};
2290 $max_warn = $out->{temperatureProbeMaxWarning};
2291 $max_crit = $out->{temperatureProbeMaxCritical};
2292 $encl_id = $out->{temperatureConnectionEnclosureNumber} - 1;
2293 $encl_name = $out->{temperatureConnectionEnclosureName};
2294 $nexus = convert_nexus($out->{temperatureProbeNexusID});
2295 }
2296 else {
2297 $id = $out->{'ID'};
2298 $name = $out->{'Name'};
2299 $state = $out->{'State'};
2300 $status = $out->{'Status'};
2301 $unit = 'FIXME';
2302 $reading = $out->{'Reading'}; $reading =~ s{\s*C}{}xms;
2303 $max_warn = $out->{'Maximum Warning Threshold'}; $max_warn =~ s{\s*C}{}xms;
2304 $max_crit = $out->{'Maximum Failure Threshold'}; $max_crit =~ s{\s*C}{}xms;
2305 $encl_id = join q{:}, $out->{ctrl}, $out->{'encl_id'};
2306 $encl_name = $out->{encl_name};
2307 $nexus = join q{:}, $out->{ctrl}, $out->{'encl_id'}, $id;
2308 }
2309
2310 next TEMP if blacklisted('encl_temp', $nexus);
2311
2312 # Default
2313 if ($status ne 'Ok') {
8a565bfc 2314 my $msg = sprintf '%s in enclosure %s [%s] is %s C at %s (%s max)',
669797e1 2315 $name, $encl_id, $encl_name, $state, $reading, $max_crit;
2316 report('storage', $msg, $status2nagios{$status}, $nexus);
2317 }
2318 # Ok
2319 else {
8a565bfc 2320 my $msg = sprintf '%s in enclosure %s [%s]: %s C (%s max)',
669797e1 2321 $name, $encl_id, $encl_name, $reading, $max_crit;
2322 report('storage', $msg, $E_OK, $nexus);
2323 }
2324
2325 # Collect performance data
2326 if (defined $opt{perfdata}) {
2327 $name =~ s{\A Temperature\sProbe\s(\d+) \z}{temp_$1}gxms;
48aeec0b 2328 my $label = "enclosure_${encl_id}_${name}";
2329 my $mini = $label;
2330 $mini =~ s{enclosure_(.+?)_temp_(.+?)}{e$1t$2}xms;
2331 push @perfdata, {
2332 label => $label,
2333 mini => $mini,
2334 value => $reading,
c000d734 2335 unit => '',
48aeec0b 2336 warn => $max_warn,
2337 crit => $max_crit,
2338 };
669797e1 2339 }
2340 }
2341 return;
2342}
2343
2344
2345#-----------------------------------------
2346# STORAGE: Check enclosure management modules (EMM)
2347#-----------------------------------------
2348sub check_enclosure_emms {
2349 return if $#controllers == -1;
5ecf578c 2350 return if blacklisted('encl_emm', 'all');
669797e1 2351
2352 my $id = undef;
2353 my $nexus = undef;
2354 my $name = undef;
2355 my $state = undef;
2356 my $status = undef;
2357 my $encl_id = undef;
2358 my $encl_name = undef;
2359 my @output = ();
2360
2361 if ($snmp) {
2362 my %emms_oid
2363 = (
2364 '1.3.6.1.4.1.674.10893.1.20.130.13.1.1' => 'enclosureManagementModuleNumber',
2365 '1.3.6.1.4.1.674.10893.1.20.130.13.1.2' => 'enclosureManagementModuleName',
2366 '1.3.6.1.4.1.674.10893.1.20.130.13.1.4' => 'enclosureManagementModuleState',
2367 '1.3.6.1.4.1.674.10893.1.20.130.13.1.11' => 'enclosureManagementModuleComponentStatus',
2368 '1.3.6.1.4.1.674.10893.1.20.130.13.1.12' => 'enclosureManagementModuleNexusID',
2369 '1.3.6.1.4.1.674.10893.1.20.130.14.1.4' => 'enclosureManagementModuleConnectionEnclosureName',
2370 '1.3.6.1.4.1.674.10893.1.20.130.14.1.5' => 'enclosureManagementModuleConnectionEnclosureNumber',
2371 );
4cabd748 2372 my $result = undef;
2373 if ($opt{use_get_table}) {
2374 my $enclosureManagementModuleTable = '1.3.6.1.4.1.674.10893.1.20.130.13';
c849fd4c 2375 my $enclosureManagementModuleConnectionTable = '1.3.6.1.4.1.674.10893.1.20.130.14';
2376
4cabd748 2377 $result = $snmp_session->get_table(-baseoid => $enclosureManagementModuleTable);
c849fd4c 2378 my $ext = $snmp_session->get_table(-baseoid => $enclosureManagementModuleConnectionTable);
2379
2380 if (defined $result) {
2381 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
2382 }
4cabd748 2383 }
2384 else {
2385 $result = $snmp_session->get_entries(-columns => [keys %emms_oid]);
2386 }
669797e1 2387
2388 # No enclosure EMMs is OK
2389 return if !defined $result;
2390
2391 @output = @{ get_snmp_output($result, \%emms_oid) };
2392 }
2393 else {
2394 foreach my $enc (@enclosures) {
2395 push @output, @{ run_omreport("storage enclosure controller=$enc->{ctrl} enclosure=$enc->{id} info=emms") };
2396 map_item('ctrl', $enc->{ctrl}, \@output);
2397 map_item('encl_id', $enc->{id}, \@output);
2398 map_item('encl_name', $enc->{name}, \@output);
2399 }
2400 }
2401
2402 my %emms_state
2403 = (
2404 0 => 'Unknown',
2405 1 => 'Ready',
2406 2 => 'Failed',
2407 3 => 'Online',
2408 4 => 'Offline',
2409 5 => 'Not Installed',
2410 6 => 'Degraded',
2411 21 => 'Missing',
2412 );
2413
2414 # Check temperature probes on each of the enclosures
2415 EMM:
2416 foreach my $out (@output) {
2417 if ($snmp) {
2418 $id = $out->{enclosureManagementModuleNumber} - 1;
2419 $name = $out->{enclosureManagementModuleName};
912d8679 2420 $state = get_hashval($out->{enclosureManagementModuleState}, \%emms_state);
669797e1 2421 $status = $snmp_status{$out->{enclosureManagementModuleComponentStatus}};
2422 $encl_id = $out->{enclosureManagementModuleConnectionEnclosureNumber} - 1;
2423 $encl_name = $out->{enclosureManagementModuleConnectionEnclosureName};
2424 $nexus = convert_nexus($out->{enclosureManagementModuleNexusID});
2425 }
2426 else {
2427 $id = $out->{'ID'};
2428 $name = $out->{'Name'};
2429 $state = $out->{'State'};
2430 $status = $out->{'Status'};
2431 $encl_id = join q{:}, $out->{ctrl}, $out->{'encl_id'};
2432 $encl_name = $out->{encl_name};
2433 $nexus = join q{:}, $out->{ctrl}, $out->{'encl_id'}, $id;
2434 }
2435
2436 next EMM if blacklisted('encl_emm', $nexus);
2437
2438 # Default
2439 if ($status ne 'Ok') {
98b224a3 2440 my $msg = sprintf '%s in enclosure %s [%s] needs attention: %s',
669797e1 2441 $name, $encl_id, $encl_name, $state;
2442 report('storage', $msg, $status2nagios{$status}, $nexus);
2443 }
2444 # Ok
2445 else {
98b224a3 2446 my $msg = sprintf '%s in enclosure %s [%s] is %s',
669797e1 2447 $name, $encl_id, $encl_name, $state;
2448 report('storage', $msg, $E_OK, $nexus);
2449 }
2450 }
2451 return;
2452}
2453
2454
2455#-----------------------------------------
2456# CHASSIS: Check memory modules
2457#-----------------------------------------
2458sub check_memory {
5ecf578c 2459 return if blacklisted('dimm', 'all');
2460
669797e1 2461 my $index = undef;
2462 my $status = undef;
2463 my $location = undef;
2464 my $size = undef;
2465 my $modes = undef;
2466 my @failures = ();
2467 my @output = ();
2468
2469 if ($snmp) {
2470 my %dimm_oid
2471 = (
2472 '1.3.6.1.4.1.674.10892.1.1100.50.1.2.1' => 'memoryDeviceIndex',
2473 '1.3.6.1.4.1.674.10892.1.1100.50.1.5.1' => 'memoryDeviceStatus',
2474 '1.3.6.1.4.1.674.10892.1.1100.50.1.8.1' => 'memoryDeviceLocationName',
2475 '1.3.6.1.4.1.674.10892.1.1100.50.1.14.1' => 'memoryDeviceSize',
2476 '1.3.6.1.4.1.674.10892.1.1100.50.1.20.1' => 'memoryDeviceFailureModes',
2477 );
4cabd748 2478 my $result = undef;
2479 if ($opt{use_get_table}) {
2480 my $memoryDeviceTable = '1.3.6.1.4.1.674.10892.1.1100.50.1';
2481 $result = $snmp_session->get_table(-baseoid => $memoryDeviceTable);
2482 }
2483 else {
2484 $result = $snmp_session->get_entries(-columns => [keys %dimm_oid]);
2485 }
669797e1 2486
2487 if (!defined $result) {
98b224a3 2488 printf "SNMP ERROR [memory]: %s.\n", $snmp_session->error;
669797e1 2489 $snmp_session->close;
2490 exit $E_UNKNOWN;
2491 }
2492
2493 @output = @{ get_snmp_output($result, \%dimm_oid) };
2494 }
2495 else {
2496 @output = @{ run_omreport("$omopt_chassis memory") };
2497 }
2498
2499 # Note: These values are bit masks, so combination values are
2500 # possible. If value is 0 (zero), memory device has no faults.
2501 my %failure_mode
2502 = (
2503 1 => 'ECC single bit correction warning rate exceeded',
2504 2 => 'ECC single bit correction failure rate exceeded',
2505 4 => 'ECC multibit fault encountered',
2506 8 => 'ECC single bit correction logging disabled',
2507 16 => 'device disabled because of spare activation',
2508 );
2509
2510 DIMM:
2511 foreach my $out (@output) {
2512 @failures = (); # Initialize
2513 if ($snmp) {
2514 $index = $out->{memoryDeviceIndex};
2515 $status = $snmp_status{$out->{memoryDeviceStatus}};
2516 $location = $out->{memoryDeviceLocationName};
2517 $size = sprintf '%d MB', $out->{memoryDeviceSize}/1024;
2518 $modes = $out->{memoryDeviceFailureModes};
2519 if ($modes > 0) {
2520 foreach my $mask (sort keys %failure_mode) {
2521 if (($modes & $mask) != 0) { push @failures, $failure_mode{$mask}; }
2522 }
2523 }
2524 }
2525 else {
2526 $index = $out->{'Type'} eq '[Not Occupied]' ? undef : $out->{'Index'};
2527 $status = $out->{'Status'};
2528 $location = $out->{'Connector Name'};
2529 $size = $out->{'Size'};
2530 if (defined $size) {
2531 $size =~ s{\s\s}{ }gxms;
2532 }
2533 # Run 'omreport chassis memory index=X' to get the failures
2534 if ($status ne 'Ok' && defined $index) {
2535 foreach (@{ run_command("$omreport $omopt_chassis memory index=$index -fmt ssv") }) {
2536 if (m/\A Failures; (.+?) \z/xms) {
2537 chop(my $fail = $1);
2538 push @failures, split m{\.}xms, $fail;
2539 }
2540 }
2541 }
2542 }
2543 $location =~ s{\A \s*(.*?)\s* \z}{$1}xms;
2544
2545 next DIMM if blacklisted('dimm', $index);
2546
2547 # Ignore empty memory slots
2548 next DIMM if !defined $index;
2549 $count{dimm}++;
2550
2551 if ($status ne 'Ok') {
2552 my $msg = undef;
2553 if (scalar @failures == 0) {
98b224a3 2554 $msg = sprintf 'Memory module %d [%s, %s] needs attention (%s)',
669797e1 2555 $index, $location, $size, $status;
2556 }
2557 else {
98b224a3 2558 $msg = sprintf 'Memory module %d [%s, %s] needs attention: %s',
669797e1 2559 $index, $location, $size, (join q{, }, @failures);
2560 }
2561
2562 report('chassis', $msg, $status2nagios{$status}, $index);
2563 }
2564 # Ok
2565 else {
98b224a3 2566 my $msg = sprintf 'Memory module %d [%s, %s] is %s',
669797e1 2567 $index, $location, $size, $status;
2568 report('chassis', $msg, $E_OK, $index);
2569 }
2570 }
2571 return;
2572}
2573
2574
2575#-----------------------------------------
2576# CHASSIS: Check fans
2577#-----------------------------------------
2578sub check_fans {
5ecf578c 2579 return if blacklisted('fan', 'all');
2580
669797e1 2581 my $index = undef;
2582 my $status = undef;
2583 my $reading = undef;
2584 my $location = undef;
2585 my $max_crit = undef;
2586 my $max_warn = undef;
2587 my @output = ();
2588
2589 if ($snmp) {
2590 my %cool_oid
2591 = (
2592 '1.3.6.1.4.1.674.10892.1.700.12.1.2.1' => 'coolingDeviceIndex',
2593 '1.3.6.1.4.1.674.10892.1.700.12.1.5.1' => 'coolingDeviceStatus',
2594 '1.3.6.1.4.1.674.10892.1.700.12.1.6.1' => 'coolingDeviceReading',
2595 '1.3.6.1.4.1.674.10892.1.700.12.1.8.1' => 'coolingDeviceLocationName',
2596 '1.3.6.1.4.1.674.10892.1.700.12.1.10.1' => 'coolingDeviceUpperCriticalThreshold',
2597 '1.3.6.1.4.1.674.10892.1.700.12.1.11.1' => 'coolingDeviceUpperNonCriticalThreshold',
2598 );
4cabd748 2599 my $result = undef;
2600 if ($opt{use_get_table}) {
2601 my $coolingDeviceTable = '1.3.6.1.4.1.674.10892.1.700.12.1';
2602 $result = $snmp_session->get_table(-baseoid => $coolingDeviceTable);
2603 }
2604 else {
2605 $result = $snmp_session->get_entries(-columns => [keys %cool_oid]);
2606 }
669797e1 2607
2608 if ($blade && !defined $result) {
2609 return 0;
2610 }
2611 elsif (!$blade && !defined $result) {
98b224a3 2612 printf "SNMP ERROR [cooling]: %s.\n", $snmp_session->error;
669797e1 2613 $snmp_session->close;
2614 exit $E_UNKNOWN;
2615 }
2616
2617 @output = @{ get_snmp_output($result, \%cool_oid) };
2618 }
2619 else {
2620 @output = @{ run_omreport("$omopt_chassis fans") };
2621 }
2622
2623 FAN:
2624 foreach my $out (@output) {
2625 if ($snmp) {
2626 $index = $out->{coolingDeviceIndex};
2627 $status = $snmp_probestatus{$out->{coolingDeviceStatus}};
2628 $reading = $out->{coolingDeviceReading};
2629 $location = $out->{coolingDeviceLocationName};
2630 $max_crit = exists $out->{coolingDeviceUpperCriticalThreshold}
2631 ? $out->{coolingDeviceUpperCriticalThreshold} : 0;
2632 $max_warn = exists $out->{coolingDeviceUpperNonCriticalThreshold}
2633 ? $out->{coolingDeviceUpperNonCriticalThreshold} : 0;
2634 }
2635 else {
2636 $index = $out->{'Index'};
2637 $status = $out->{'Status'};
2638 $reading = $out->{'Reading'};
2639 $location = $out->{'Probe Name'};
2640 $max_crit = $out->{'Maximum Failure Threshold'} ne '[N/A]'
2641 ? $out->{'Maximum Failure Threshold'} : 0;
2642 $max_warn = $out->{'Maximum Warning Threshold'} ne '[N/A]'
2643 ? $out->{'Maximum Warning Threshold'} : 0;
2644 $reading =~ s{\A (\d+).* \z}{$1}xms;
2645 $max_warn =~ s{\A (\d+).* \z}{$1}xms;
2646 $max_crit =~ s{\A (\d+).* \z}{$1}xms;
2647 }
2648
2649 next FAN if blacklisted('fan', $index);
2650 $count{fan}++;
2651
2652 if ($status ne 'Ok') {
98b224a3 2653 my $msg = sprintf 'Chassis fan %d [%s] needs attention: %s',
669797e1 2654 $index, $location, $status;
2655 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2656 report('chassis', $msg, $err, $index);
2657 }
2658 else {
98b224a3 2659 my $msg = sprintf 'Chassis fan %d [%s]: %s',
669797e1 2660 $index, $location, $reading;
2661 report('chassis', $msg, $E_OK, $index);
2662 }
2663
2664 # Collect performance data
2665 if (defined $opt{perfdata}) {
2666 my $pname = lc $location;
2667 $pname =~ s{\s}{_}gxms;
2668 $pname =~ s{proc_}{cpu#}xms;
2669 my $pkey = join q{_}, 'fan', $index, $pname;
48aeec0b 2670 push @perfdata, {
2671 label => $pkey,
2672 mini => "f$index",
2673 value => $reading,
c000d734 2674 unit => '',
48aeec0b 2675 warn => $max_warn,
2676 crit => $max_crit,
2677 };
669797e1 2678 }
2679 }
2680 return;
2681}
2682
2683
2684#-----------------------------------------
2685# CHASSIS: Check power supplies
2686#-----------------------------------------
2687sub check_powersupplies {
5ecf578c 2688 return if blacklisted('ps', 'all');
2689
669797e1 2690 my $index = undef;
2691 my $status = undef;
2692 my $type = undef;
2693 my $err_type = undef;
2694 my $state = undef;
2695 my @states = ();
2696 my @output = ();
2697
2698 if ($snmp) {
2699 my %ps_oid
2700 = (
2701 '1.3.6.1.4.1.674.10892.1.600.12.1.2.1' => 'powerSupplyIndex',
2702 '1.3.6.1.4.1.674.10892.1.600.12.1.5.1' => 'powerSupplyStatus',
2703 '1.3.6.1.4.1.674.10892.1.600.12.1.7.1' => 'powerSupplyType',
2704 '1.3.6.1.4.1.674.10892.1.600.12.1.11.1' => 'powerSupplySensorState',
2705 '1.3.6.1.4.1.674.10892.1.600.12.1.12.1' => 'powerSupplyConfigurationErrorType',
2706 );
4cabd748 2707 my $result = undef;
2708 if ($opt{use_get_table}) {
2709 my $powerDeviceTable = '1.3.6.1.4.1.674.10892.1.600.12.1';
2710 $result = $snmp_session->get_table(-baseoid => $powerDeviceTable);
2711 }
2712 else {
2713 $result = $snmp_session->get_entries(-columns => [keys %ps_oid]);
2714 }
669797e1 2715
2716 # No instrumented PSU is OK (blades, low-end servers)
2717 return 0 if !defined $result;
2718
2719 @output = @{ get_snmp_output($result, \%ps_oid) };
2720 }
2721 else {
2722 @output = @{ run_omreport("$omopt_chassis pwrsupplies") };
2723 }
2724
2725 my %ps_type
2726 = (
2727 1 => 'Other',
2728 2 => 'Unknown',
2729 3 => 'Linear',
2730 4 => 'Switching',
2731 5 => 'Battery',
2732 6 => 'Uninterruptible Power Supply',
2733 7 => 'Converter',
2734 8 => 'Regulator',
2735 9 => 'AC',
2736 10 => 'DC',
2737 11 => 'VRM',
2738 );
2739
2740 my %ps_state
2741 = (
2742 1 => 'Presence detected',
2743 2 => 'Failure detected',
2744 4 => 'Predictive Failure',
2745 8 => 'AC lost',
2746 16 => 'AC lost or out-of-range',
2747 32 => 'AC out-of-range but present',
2748 64 => 'Configuration error',
2749 );
2750
2751 my %ps_config_error_type
2752 = (
2753 1 => 'Vendor mismatch',
2754 2 => 'Revision mismatch',
2755 3 => 'Processor missing',
2756 );
2757
2758 PS:
2759 foreach my $out (@output) {
2760 if ($snmp) {
2761 @states = (); # contains states for the PS
2762
2763 $index = $out->{powerSupplyIndex} - 1;
2764 $status = $snmp_status{$out->{powerSupplyStatus}};
912d8679 2765 $type = get_hashval($out->{powerSupplyType}, \%ps_type);
669797e1 2766 $err_type = defined $out->{powerSupplyConfigurationErrorType}
2767 ? $ps_config_error_type{$out->{powerSupplyConfigurationErrorType}} : undef;
2768
2769 # get the combined state from the StatusReading OID
2770 foreach my $mask (sort keys %ps_state) {
2771 if (($out->{powerSupplySensorState} & $mask) != 0) {
2772 push @states, $ps_state{$mask};
2773 }
2774 }
2775
2776 # If configuration error, also include the error type
2777 if (defined $err_type) {
2778 push @states, $err_type;
2779 }
2780
2781 # Finally, construct the state string
2782 $state = join q{, }, @states;
2783 }
2784 else {
2785 $index = $out->{'Index'};
2786 $status = $out->{'Status'};
2787 $type = $out->{'Type'};
2788 $state = $out->{'Online Status'};
2789 }
2790
2791 next PS if blacklisted('ps', $index);
2792 $count{power}++;
2793
2794 if ($status ne 'Ok') {
98b224a3 2795 my $msg = sprintf 'Power Supply %d [%s] needs attention: %s',
669797e1 2796 $index, $type, $state;
2797 report('chassis', $msg, $status2nagios{$status}, $index);
2798 }
2799 else {
98b224a3 2800 my $msg = sprintf 'Power Supply %d [%s]: %s',
669797e1 2801 $index, $type, $state;
2802 report('chassis', $msg, $E_OK, $index);
2803 }
2804 }
2805 return;
2806}
2807
2808
2809#-----------------------------------------
2810# CHASSIS: Check temperatures
2811#-----------------------------------------
2812sub check_temperatures {
5ecf578c 2813 return if blacklisted('temp', 'all');
2814
669797e1 2815 my $index = undef;
2816 my $status = undef;
2817 my $reading = undef;
2818 my $location = undef;
2819 my $max_crit = undef;
2820 my $max_warn = undef;
2821 my $min_warn = undef;
2822 my $min_crit = undef;
2823 my $type = undef;
2824 my $discrete = undef;
2825 my @output = ();
2826
2827 # Getting custom temperature thresholds (user option)
2828 my %warn_threshold = %{ custom_temperature_thresholds('w') };
2829 my %crit_threshold = %{ custom_temperature_thresholds('c') };
2830
2831 if ($snmp) {
2832 my %temp_oid
2833 = (
2834 '1.3.6.1.4.1.674.10892.1.700.20.1.2.1' => 'temperatureProbeIndex',
2835 '1.3.6.1.4.1.674.10892.1.700.20.1.5.1' => 'temperatureProbeStatus',
2836 '1.3.6.1.4.1.674.10892.1.700.20.1.6.1' => 'temperatureProbeReading',
2837 '1.3.6.1.4.1.674.10892.1.700.20.1.7.1' => 'temperatureProbeType',
2838 '1.3.6.1.4.1.674.10892.1.700.20.1.8.1' => 'temperatureProbeLocationName',
2839 '1.3.6.1.4.1.674.10892.1.700.20.1.10.1' => 'temperatureProbeUpperCriticalThreshold',
2840 '1.3.6.1.4.1.674.10892.1.700.20.1.11.1' => 'temperatureProbeUpperNonCriticalThreshold',
2841 '1.3.6.1.4.1.674.10892.1.700.20.1.12.1' => 'temperatureProbeLowerNonCriticalThreshold',
2842 '1.3.6.1.4.1.674.10892.1.700.20.1.13.1' => 'temperatureProbeLowerCriticalThreshold',
2843 '1.3.6.1.4.1.674.10892.1.700.20.1.16.1' => 'temperatureProbeDiscreteReading',
2844 );
ba199ee0 2845 # this didn't work well for some reason
2846 #my $result = $snmp_session->get_entries(-columns => [keys %temp_oid]);
2847
2848 # Getting values using the table
2849 my $temperatureProbeTable = '1.3.6.1.4.1.674.10892.1.700.20';
2850 my $result = $snmp_session->get_table(-baseoid => $temperatureProbeTable);
669797e1 2851
2852 if (!defined $result) {
98b224a3 2853 printf "SNMP ERROR [temperatures]: %s.\n", $snmp_session->error;
669797e1 2854 $snmp_session->close;
2855 exit $E_UNKNOWN;
2856 }
2857
2858 @output = @{ get_snmp_output($result, \%temp_oid) };
2859 }
2860 else {
2861 @output = @{ run_omreport("$omopt_chassis temps") };
2862 }
2863
2864 my %probe_type
2865 = (
2866 1 => 'Other', # type is other than following values
2867 2 => 'Unknown', # type is unknown
2868 3 => 'AmbientESM', # type is Ambient Embedded Systems Management temperature probe
2869 16 => 'Discrete', # type is temperature probe with discrete reading
2870 );
2871
2872 TEMP:
2873 foreach my $out (@output) {
2874 if ($snmp) {
2875 $index = $out->{temperatureProbeIndex} - 1;
2876 $status = $snmp_probestatus{$out->{temperatureProbeStatus}};
2877 $reading = $out->{temperatureProbeReading} / 10;
2878 $location = $out->{temperatureProbeLocationName};
2879 $max_crit = $out->{temperatureProbeUpperCriticalThreshold} / 10;
2880 $max_warn = $out->{temperatureProbeUpperNonCriticalThreshold} / 10;
2881 $min_crit = exists $out->{temperatureProbeLowerCriticalThreshold}
2882 ? $out->{temperatureProbeLowerCriticalThreshold} / 10 : '[N/A]';
2883 $min_warn = exists $out->{temperatureProbeLowerNonCriticalThreshold}
2884 ? $out->{temperatureProbeLowerNonCriticalThreshold} / 10 : '[N/A]';
912d8679 2885 $type = get_hashval($out->{temperatureProbeType}, \%probe_type);
669797e1 2886 $discrete = exists $out->{temperatureProbeDiscreteReading}
2887 ? $out->{temperatureProbeDiscreteReading} : undef;
2888 }
2889 else {
2890 $index = $out->{'Index'};
2891 $status = $out->{'Status'};
2892 $reading = $out->{'Reading'}; $reading =~ s{\.0\s+C}{}xms;
2893 $location = $out->{'Probe Name'};
2894 $max_crit = $out->{'Maximum Failure Threshold'}; $max_crit =~ s{\.0\s+C}{}xms;
2895 $max_warn = $out->{'Maximum Warning Threshold'}; $max_warn =~ s{\.0\s+C}{}xms;
2896 $min_crit = $out->{'Minimum Failure Threshold'}; $min_crit =~ s{\.0\s+C}{}xms;
2897 $min_warn = $out->{'Minimum Warning Threshold'}; $min_warn =~ s{\.0\s+C}{}xms;
2898 $type = $reading =~ m{\A\d+\z}xms ? 'AmbientESM' : 'Discrete';
2899 $discrete = $reading;
2900 }
2901
2902 next TEMP if blacklisted('temp', $index);
2903 $count{temp}++;
2904
2905 if ($type eq 'Discrete') {
2906 my $msg = sprintf 'Temperature probe %d (%s): is %s',
2907 $index, $location, $discrete;
2908 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2909 report('chassis', $msg, $err, $index);
2910 }
2911 else {
2912 # First check according to custom thresholds
2913 if (exists $crit_threshold{$index}{max} and $reading > $crit_threshold{$index}{max}) {
2914 # Custom critical MAX
98b224a3 2915 my $msg = sprintf 'Temperature Probe %d [%s] reads %d C (custom max=%d)',
669797e1 2916 $index, $location, $reading, $crit_threshold{$index}{max};
2917 report('chassis', $msg, $E_CRITICAL, $index);
2918 }
2919 elsif (exists $warn_threshold{$index}{max} and $reading > $warn_threshold{$index}{max}) {
2920 # Custom warning MAX
98b224a3 2921 my $msg = sprintf 'Temperature Probe %d [%s] reads %d C (custom max=%d)',
669797e1 2922 $index, $location, $reading, $warn_threshold{$index}{max};
2923 report('chassis', $msg, $E_WARNING, $index);
2924 }
2925 elsif (exists $crit_threshold{$index}{min} and $reading < $crit_threshold{$index}{min}) {
2926 # Custom critical MIN
98b224a3 2927 my $msg = sprintf 'Temperature Probe %d [%s] reads %d C (custom min=%d)',
669797e1 2928 $index, $location, $reading, $crit_threshold{$index}{min};
2929 report('chassis', $msg, $E_CRITICAL, $index);
2930 }
2931 elsif (exists $warn_threshold{$index}{min} and $reading < $warn_threshold{$index}{min}) {
2932 # Custom warning MIN
98b224a3 2933 my $msg = sprintf 'Temperature Probe %d [%s] reads %d C (custom min=%d)',
669797e1 2934 $index, $location, $reading, $warn_threshold{$index}{min};
2935 report('chassis', $msg, $E_WARNING, $index);
2936 }
2937 elsif ($status ne 'Ok' and $max_crit ne '[N/A]' and $reading > $max_crit) {
98b224a3 2938 my $msg = sprintf 'Temperature Probe %d [%s] is critically high at %d C',
669797e1 2939 $index, $location, $reading;
2940 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2941 report('chassis', $msg, $err, $index);
2942 }
2943 elsif ($status ne 'Ok' and $max_warn ne '[N/A]' and $reading > $max_warn) {
98b224a3 2944 my $msg = sprintf 'Temperature Probe %d [%s] is too high at %d C',
669797e1 2945 $index, $location, $reading;
2946 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2947 report('chassis', $msg, $err, $index);
2948 }
2949 elsif ($status ne 'Ok' and $min_crit ne '[N/A]' and $reading < $min_crit) {
98b224a3 2950 my $msg = sprintf 'Temperature Probe %d [%s] is critically low at %d C',
669797e1 2951 $index, $location, $reading;
2952 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2953 report('chassis', $msg, $err, $index);
2954 }
2955 elsif ($status ne 'Ok' and $min_warn ne '[N/A]' and $reading < $min_warn) {
98b224a3 2956 my $msg = sprintf 'Temperature Probe %d [%s] is too low at %d C',
669797e1 2957 $index, $location, $reading;
2958 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2959 report('chassis', $msg, $err, $index);
2960 }
2961 # Ok
2962 else {
304c4cba 2963 my $msg = sprintf 'Temperature Probe %d [%s] reads %d C',
2964 $index, $location, $reading;
2965 if ($min_warn eq '[N/A]' and $min_crit eq '[N/A]') {
2966 $msg .= sprintf ' (max=%s/%s)', $max_warn, $max_crit;
2967 }
2968 else {
2969 $msg .= sprintf ' (min=%s/%s, max=%s/%s)',
2970 $min_warn, $min_crit, $max_warn, $max_crit;
8ce893fd 2971 }
669797e1 2972 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
2973 report('chassis', $msg, $err, $index);
2974 }
2975
2976 # Collect performance data
2977 if (defined $opt{perfdata}) {
2978 my $pname = lc $location;
2979 $pname =~ s{\s}{_}gxms;
2980 $pname =~ s{_temp\z}{}xms;
2981 $pname =~ s{proc_}{cpu#}xms;
2982 my $pkey = join q{_}, 'temp', $index, $pname;
48aeec0b 2983 push @perfdata, {
2984 label => $pkey,
2985 mini => "t$index",
2986 value => $reading,
c000d734 2987 unit => '',
48aeec0b 2988 warn => $max_warn,
2989 crit => $max_crit,
2990 };
669797e1 2991 }
2992 }
2993 }
2994 return;
2995}
2996
2997
2998#-----------------------------------------
2999# CHASSIS: Check processors
3000#-----------------------------------------
3001sub check_processors {
5ecf578c 3002 return if blacklisted('cpu', 'all');
3003
669797e1 3004 my $index = undef;
3005 my $status = undef;
3006 my $state = undef;
8ce893fd 3007 my $brand = undef;
3008 my $family = undef;
3009 my $man = undef;
3010 my $speed = undef;
8ce893fd 3011 my @output = ();
669797e1 3012
3013 if ($snmp) {
3014
3015 # NOTE: For some reason, older models don't have the
8ce893fd 3016 # "Processor Device Status" OIDs. We check both the newer
3017 # (preferred) OIDs and the old ones.
669797e1 3018
8ce893fd 3019 my %cpu_oid
669797e1 3020 = (
8ce893fd 3021 '1.3.6.1.4.1.674.10892.1.1100.30.1.2.1' => 'processorDeviceIndex',
3022 '1.3.6.1.4.1.674.10892.1.1100.30.1.5.1' => 'processorDeviceStatus',
3023 '1.3.6.1.4.1.674.10892.1.1100.30.1.8.1' => 'processorDeviceManufacturerName',
3024 '1.3.6.1.4.1.674.10892.1.1100.30.1.9.1' => 'processorDeviceStatusState',
3025 '1.3.6.1.4.1.674.10892.1.1100.30.1.10.1' => 'processorDeviceFamily',
3026 '1.3.6.1.4.1.674.10892.1.1100.30.1.12.1' => 'processorDeviceCurrentSpeed',
3027 '1.3.6.1.4.1.674.10892.1.1100.30.1.23.1' => 'processorDeviceBrandName',
3028 '1.3.6.1.4.1.674.10892.1.1100.32.1.2.1' => 'processorDeviceStatusIndex',
3029 '1.3.6.1.4.1.674.10892.1.1100.32.1.5.1' => 'processorDeviceStatusStatus',
3030 '1.3.6.1.4.1.674.10892.1.1100.32.1.6.1' => 'processorDeviceStatusReading',
669797e1 3031 );
4cabd748 3032 my $result = undef;
3033 if ($opt{use_get_table}) {
3034 my $processorDeviceTable = '1.3.6.1.4.1.674.10892.1.1100.30.1';
3035 my $processorDeviceStatusTable = '1.3.6.1.4.1.674.10892.1.1100.32.1';
669797e1 3036
c849fd4c 3037 $result = $snmp_session->get_table(-baseoid => $processorDeviceTable);
3038 my $ext = $snmp_session->get_table(-baseoid => $processorDeviceStatusTable);
4cabd748 3039
c849fd4c 3040 defined $ext && map { $$result{$_} = $$ext{$_} } keys %{ $ext };
4cabd748 3041 }
3042 else {
3043 $result = $snmp_session->get_entries(-columns => [keys %cpu_oid]);
3044 }
669797e1 3045
3046 if (!defined $result) {
98b224a3 3047 printf "SNMP ERROR [processors]: %s.\n", $snmp_session->error;
669797e1 3048 $snmp_session->close;
3049 exit $E_UNKNOWN;
3050 }
3051
8ce893fd 3052 @output = @{ get_snmp_output($result, \%cpu_oid) };
669797e1 3053 }
3054 else {
3055 @output = @{ run_omreport("$omopt_chassis processors") };
3056 }
3057
3058 my %cpu_state
3059 = (
3060 1 => 'Other', # other than following values
3061 2 => 'Unknown', # unknown
3062 3 => 'Enabled', # enabled
3063 4 => 'User Disabled', # disabled by user via BIOS setup
3064 5 => 'BIOS Disabled', # disabled by BIOS (POST error)
3065 6 => 'Idle', # idle
3066 );
3067
3068 my %cpu_reading
3069 = (
3070 1 => 'Internal Error', # Internal Error
3071 2 => 'Thermal Trip', # Thermal Trip
3072 32 => 'Configuration Error', # Configuration Error
3073 128 => 'Present', # Processor Present
3074 256 => 'Disabled', # Processor Disabled
3075 512 => 'Terminator Present', # Terminator Present
3076 1024 => 'Throttled', # Processor Throttled
3077 );
3078
8ce893fd 3079 # Mapping between family numbers from SNMP and actual CPU family
3080 my %cpu_family
3081 = (
b221393e 3082 1 => 'Other', 2 => 'Unknown',
3083 3 => '8086', 4 => '80286',
3084 5 => '386', 6 => '486',
3085 7 => '8087', 8 => '80287',
3086 9 => '80387', 10 => '80487',
3087 11 => 'Pentium', 12 => 'Pentium Pro',
3088 13 => 'Pentium II', 14 => 'Pentium with MMX',
3089 15 => 'Celeron', 16 => 'Pentium II Xeon',
3090 17 => 'Pentium III', 18 => 'Pentium III Xeon',
3091 19 => 'Pentium III', 20 => 'Itanium',
3092 21 => 'Xeon', 22 => 'Pentium 4',
3093 23 => 'Xeon MP', 24 => 'Itanium 2',
3094 25 => 'K5', 26 => 'K6',
3095 27 => 'K6-2', 28 => 'K6-3',
3096 29 => 'Athlon', 30 => 'AMD2900',
3097 31 => 'K6-2+', 32 => 'Power PC',
3098 33 => 'Power PC 601', 34 => 'Power PC 603',
3099 35 => 'Power PC 603+', 36 => 'Power PC 604',
3100 37 => 'Power PC 620', 38 => 'Power PC x704',
3101 39 => 'Power PC 750', 40 => 'Core Duo',
3102 41 => 'Core Duo mobile', 42 => 'Core Solo mobile',
3103 43 => 'Intel Atom', 44 => undef,
3104 45 => undef, 46 => undef,
3105 47 => undef, 48 => 'Alpha',
3106 49 => 'Alpha 21064', 50 => 'Alpha 21066',
3107 51 => 'Alpha 21164', 52 => 'Alpha 21164PC',
3108 53 => 'Alpha 21164a', 54 => 'Alpha 21264',
3109 55 => 'Alpha 21364', 56 => 'Turion II Ultra Dual-Core Mobile M',
3110 57 => 'Turion II Dual-Core Mobile M', 58 => 'Athlon II Dual-Core Mobile M ',
3111 59 => 'Opteron 6100', 60 => undef,
3112 61 => undef, 62 => undef,
3113 63 => undef, 64 => 'MIPS',
3114 65 => 'MIPS R4000', 66 => 'MIPS R4200',
3115 67 => 'MIPS R4400', 68 => 'MIPS R4600',
3116 69 => 'MIPS R10000', 70 => undef,
3117 71 => undef, 72 => undef,
3118 73 => undef, 74 => undef,
3119 75 => undef, 76 => undef,
3120 77 => undef, 78 => undef,
3121 79 => undef, 80 => 'SPARC',
3122 81 => 'SuperSPARC', 82 => 'microSPARC II',
3123 83 => 'microSPARC IIep', 84 => 'UltraSPARC',
3124 85 => 'UltraSPARC II', 86 => 'UltraSPARC IIi',
3125 87 => 'UltraSPARC III', 88 => 'UltraSPARC IIIi',
3126 89 => undef, 90 => undef,
3127 91 => undef, 92 => undef,
3128 93 => undef, 94 => undef,
3129 95 => undef, 96 => '68040',
3130 97 => '68xxx', 98 => '68000',
3131 99 => '68010', 100 => '68020',
3132 101 => '68030', 102 => undef,
3133 103 => undef, 104 => undef,
3134 105 => undef, 106 => undef,
3135 107 => undef, 108 => undef,
3136 109 => undef, 110 => undef,
3137 111 => undef, 112 => 'Hobbit',
3138 113 => undef, 114 => undef,
3139 115 => undef, 116 => undef,
3140 117 => undef, 118 => undef,
3141 119 => undef, 120 => 'Crusoe TM5000',
3142 121 => 'Crusoe TM3000', 122 => 'Efficeon TM8000',
3143 123 => undef, 124 => undef,
3144 125 => undef, 126 => undef,
3145 127 => undef, 128 => 'Weitek',
3146 129 => undef, 130 => 'Celeron M',
3147 131 => 'Athlon 64', 132 => 'Opteron',
3148 133 => 'Sempron', 134 => 'Turion 64 Mobile',
3149 135 => 'Dual-Core Opteron', 136 => 'Athlon 64 X2 DC',
3150 137 => 'Turion 64 X2 M', 138 => 'Quad-Core Opteron',
3151 139 => '3rd gen Opteron', 140 => 'AMD Phenom FX Quad-Core',
3152 141 => 'AMD Phenom X4 Quad-Core', 142 => 'AMD Phenom X2 Dual-Core',
3153 143 => 'AMD Athlon X2 Dual-Core', 144 => 'PA-RISC',
3154 145 => 'PA-RISC 8500', 146 => 'PA-RISC 8000',
3155 147 => 'PA-RISC 7300LC', 148 => 'PA-RISC 7200',
3156 149 => 'PA-RISC 7100LC', 150 => 'PA-RISC 7100',
3157 151 => undef, 152 => undef,
3158 153 => undef, 154 => undef,
3159 155 => undef, 156 => undef,
3160 157 => undef, 158 => undef,
3161 159 => undef, 160 => 'V30',
3162 161 => 'Quad-Core Xeon 3200', 162 => 'Dual-Core Xeon 3000',
3163 163 => 'Quad-Core Xeon 5300', 164 => 'Dual-Core Xeon 5100',
3164 165 => 'Dual-Core Xeon 5000', 166 => 'Dual-Core Xeon LV',
3165 167 => 'Dual-Core Xeon ULV', 168 => 'Dual-Core Xeon 7100',
3166 169 => 'Quad-Core Xeon 5400', 170 => 'Quad-Core Xeon',
3167 171 => 'Dual-Core Xeon 5200', 172 => 'Dual-Core Xeon 7200',
3168 173 => 'Quad-Core Xeon 7300', 174 => 'Quad-Core Xeon 7400',
3169 175 => 'Multi-Core Xeon 7400', 176 => 'M1',
3170 177 => 'M2', 178 => undef,
3171 179 => 'Pentium 4 HT', 180 => 'AS400',
3172 181 => undef, 182 => 'Athlon XP',
3173 183 => 'Athlon MP', 184 => 'Duron',
3174 185 => 'Pentium M', 186 => 'Celeron D',
3175 187 => 'Pentium D', 188 => 'Pentium Extreme',
3176 189 => 'Core Solo', 190 => 'Core2',
3177 191 => 'Core2 Duo', 192 => 'Core2 Solo',
3178 193 => 'Core2 Extreme', 194 => 'Core2 Quad',
3179 195 => 'Core2 Extreme mobile', 196 => 'Core2 Duo mobile',
3180 197 => 'Core2 Solo mobile', 198 => 'Core i7',
3181 199 => 'Dual-Core Celeron', 200 => 'IBM390',
3182 201 => 'G4', 202 => 'G5',
3183 203 => 'ESA/390 G6', 204 => 'z/Architectur',
3184 205 => 'Core i5', 206 => undef,
3185 207 => undef, 208 => undef,
3186 209 => undef, 210 => 'C7-M',
3187 211 => 'C7-D', 212 => 'C7',
3188 213 => 'Eden', 214 => 'Multi-Core Xeon',
3189 215 => 'Dual-Core Xeon 3xxx', 216 => 'Quad-Core Xeon 3xxx',
3190 217 => undef, 218 => 'Dual-Core Xeon 5xxx',
3191 219 => 'Quad-Core Xeon 5xxx', 220 => undef,
3192 221 => 'Dual-Core Xeon 7xxx', 222 => 'Quad-Core Xeon 7xxx',
3193 223 => 'Multi-Core Xeon 7xxx', 224 => undef,
3194 225 => undef, 226 => undef,
3195 227 => undef, 228 => undef,
3196 229 => undef, 230 => 'Embedded AMD Opteron Quad-Core',
3197 231 => 'AMD Phenom Triple-Core', 232 => 'AMD Turion Ultra Dual-Core Mobile',
3198 233 => 'AMD Turion Dual-Core Mobile', 234 => 'AMD Athlon Dual-Core',
3199 235 => 'AMD Sempron SI', 236 => 'AMD Phenom II',
3200 237 => 'AMD Athlon II', 238 => 'Six-Core AMD Opteron',
3201 239 => 'AMD Sempron M', 240 => undef,
3202 241 => undef, 242 => undef,
3203 243 => undef, 244 => undef,
3204 245 => undef, 246 => undef,
3205 247 => undef, 248 => undef,
3206 249 => undef, 250 => 'i860',
3207 251 => 'i960',
8ce893fd 3208 );
669797e1 3209
3210 CPU:
3211 foreach my $out (@output) {
3212 if ($snmp) {
8ce893fd 3213 $index = exists $out->{processorDeviceStatusIndex}
3214 ? $out->{processorDeviceStatusIndex} - 1
3215 : $out->{processorDeviceIndex} - 1;
3216 $status = exists $out->{processorDeviceStatusStatus}
3217 ? $snmp_status{$out->{processorDeviceStatusStatus}}
3218 : $snmp_status{$out->{processorDeviceStatus}};
3219 if (exists $out->{processorDeviceStatusReading}) {
669797e1 3220 my @states = (); # contains states for the CPU
669797e1 3221
3222 # get the combined state from the StatusReading OID
3223 foreach my $mask (sort keys %cpu_reading) {
3224 if (($out->{processorDeviceStatusReading} & $mask) != 0) {
3225 push @states, $cpu_reading{$mask};
3226 }
3227 }
3228
3229 # Finally, create the state string
3230 $state = join q{, }, @states;
3231 }
3232 else {
912d8679 3233 $state = get_hashval($out->{processorDeviceStatusState}, \%cpu_state);
669797e1 3234 }
8ce893fd 3235 $man = $out->{processorDeviceManufacturerName};
87a0958c 3236 $family = (exists $out->{processorDeviceFamily}
04a878db 3237 and exists $cpu_family{$out->{processorDeviceFamily}})
3238 ? $cpu_family{$out->{processorDeviceFamily}} : undef;
8ce893fd 3239 $speed = $out->{processorDeviceCurrentSpeed};
3240 $brand = $out->{processorDeviceBrandName};
669797e1 3241 }
3242 else {
3243 $index = $out->{'Index'};
3244 $status = $out->{'Status'};
3245 $state = $out->{'State'};
8ce893fd 3246 $brand = exists $out->{'Processor Brand'} ? $out->{'Processor Brand'} : undef;
3247 $family = exists $out->{'Processor Family'} ? $out->{'Processor Family'} : undef;
3248 $man = exists $out->{'Processor Manufacturer'} ? $out->{'Processor Manufacturer'} : undef;
3249 $speed = exists $out->{'Current Speed'} ? $out->{'Current Speed'} : undef;
669797e1 3250 }
3251
3252 next CPU if blacklisted('cpu', $index);
3253
3254 # Ignore unoccupied CPU slots (omreport)
3255 next CPU if (defined $out->{'Processor Manufacturer'}
3256 and $out->{'Processor Manufacturer'} eq '[Not Occupied]')
3257 or (defined $out->{'Processor Brand'} and $out->{'Processor Brand'} eq '[Not Occupied]');
3258
3259 # Ignore unoccupied CPU slots (snmp)
3260 if ($snmp and exists $out->{processorDeviceStatusReading}
3261 and $out->{processorDeviceStatusReading} == 0) {
3262 next CPU;
3263 }
3264
3265 $count{cpu}++;
3266
8ce893fd 3267 if (defined $brand) {
3268 $brand =~ s{\s\s+}{ }gxms;
e7dc67d0 3269 $brand =~ s{\((R|tm)\)}{}gxms;
3270 $brand =~ s{\s(CPU|Processor)}{}xms;
8ce893fd 3271 $brand =~ s{\s\@}{}xms;
3272 }
3273 elsif (defined $family and defined $man and defined $speed) {
3274 $speed =~ s{\A (\d+) .*}{$1}xms;
49bf41a5 3275 $brand = sprintf '%s %s %.2fGHz', $man, $family, $speed / 1000;
8ce893fd 3276 }
3277 else {
3278 $brand = "unknown";
3279 }
3280
669797e1 3281 # Default
3282 if ($status ne 'Ok') {
0a0813de 3283 my $msg = sprintf 'Processor %d [%s] needs attention: %s',
8ce893fd 3284 $index, $brand, $state;
669797e1 3285 report('chassis', $msg, $status2nagios{$status}, $index);
3286 }
3287 # Ok
3288 else {
0a0813de 3289 my $msg = sprintf 'Processor %d [%s] is %s',
8ce893fd 3290 $index, $brand, $state;
669797e1 3291 report('chassis', $msg, $E_OK, $index);
3292 }
3293 }
3294 return;
3295}
3296
3297
3298#-----------------------------------------
3299# CHASSIS: Check voltage probes
3300#-----------------------------------------
3301sub check_volts {
5ecf578c 3302 return if blacklisted('volt', 'all');
3303
669797e1 3304 my $index = undef;
3305 my $status = undef;
3306 my $reading = undef;
3307 my $location = undef;
3308 my @output = ();
3309
3310 if ($snmp) {
3311 my %volt_oid
3312 = (
3313 '1.3.6.1.4.1.674.10892.1.600.20.1.2.1' => 'voltageProbeIndex',
3314 '1.3.6.1.4.1.674.10892.1.600.20.1.5.1' => 'voltageProbeStatus',
3315 '1.3.6.1.4.1.674.10892.1.600.20.1.6.1' => 'voltageProbeReading',
3316 '1.3.6.1.4.1.674.10892.1.600.20.1.8.1' => 'voltageProbeLocationName',
3317 '1.3.6.1.4.1.674.10892.1.600.20.1.16.1' => 'voltageProbeDiscreteReading',
3318 );
ba199ee0 3319
3320 my $voltageProbeTable = '1.3.6.1.4.1.674.10892.1.600.20.1';
3321 my $result = $snmp_session->get_table(-baseoid => $voltageProbeTable);
669797e1 3322
3323 if (!defined $result) {
98b224a3 3324 printf "SNMP ERROR [voltage]: %s.\n", $snmp_session->error;
669797e1 3325 $snmp_session->close;
3326 exit $E_UNKNOWN;
3327 }
3328
3329 @output = @{ get_snmp_output($result, \%volt_oid) };
3330 }
3331 else {
3332 @output = @{ run_omreport("$omopt_chassis volts") };
3333 }
3334
3335 my %volt_discrete_reading
3336 = (
3337 1 => 'Good',
3338 2 => 'Bad',
3339 );
3340
3341 VOLT:
3342 foreach my $out (@output) {
3343 if ($snmp) {
3344 $index = $out->{voltageProbeIndex} - 1;
ffa570fc 3345 $status = $snmp_probestatus{$out->{voltageProbeStatus}};
669797e1 3346 $reading = exists $out->{voltageProbeReading}
3347 ? sprintf('%.3f V', $out->{voltageProbeReading}/1000)
912d8679 3348 : get_hashval($out->{voltageProbeDiscreteReading}, \%volt_discrete_reading);
669797e1 3349 $location = $out->{voltageProbeLocationName};
3350 }
3351 else {
3352 $index = $out->{'Index'};
3353 $status = $out->{'Status'};
3354 $reading = $out->{'Reading'};
3355 $location = $out->{'Probe Name'};
3356 }
3357
3358 next VOLT if blacklisted('volt', $index);
3359 $count{volt}++;
3360
98b224a3 3361 my $msg = sprintf 'Voltage sensor %d [%s] is %s',
669797e1 3362 $index, $location, $reading;
3363 my $err = $snmp ? $probestatus2nagios{$status} : $status2nagios{$status};
3364 report('chassis', $msg, $err, $index);
3365 }
3366 return;
3367}
3368
3369
3370#-----------------------------------------
3371# CHASSIS: Check batteries
3372#-----------------------------------------
3373sub check_batteries {
5ecf578c 3374 return if blacklisted('bp', 'all');
3375
669797e1 3376 my $index = undef;
3377 my $status = undef;
3378 my $reading = undef;
3379 my $location = undef;
3380 my @output = ();
3381
3382 if ($snmp) {
3383 my %bat_oid
3384 = (
3385 '1.3.6.1.4.1.674.10892.1.600.50.1.2.1' => 'batteryIndex',
3386 '1.3.6.1.4.1.674.10892.1.600.50.1.5.1' => 'batteryStatus',
3387 '1.3.6.1.4.1.674.10892.1.600.50.1.6.1' => 'batteryReading',
3388 '1.3.6.1.4.1.674.10892.1.600.50.1.7.1' => 'batteryLocationName',
3389 );
4cabd748 3390 my $result = undef;
3391 if ($opt{use_get_table}) {
3392 my $batteryTable = '1.3.6.1.4.1.674.10892.1.600.50.1';
3393 $result = $snmp_session->get_table(-baseoid => $batteryTable);
3394 }
3395 else {
3396 $result = $snmp_session->get_entries(-columns => [keys %bat_oid]);
3397 }
669797e1 3398
3399 # No batteries is OK
3400 return 0 if !defined $result;
3401
3402 @output = @{ get_snmp_output($result, \%bat_oid) };
3403 }
3404 else {
3405 @output = @{ run_omreport("$omopt_chassis batteries") };
3406 }
3407
3408 my %bat_reading
3409 = (
3410 1 => 'Predictive Failure',
3411 2 => 'Failed',
3412 4 => 'Presence Detected',
3413 );
3414
3415 BATTERY:
3416 foreach my $out (@output) {
3417 if ($snmp) {
3418 $index = $out->{batteryIndex} - 1;
3419 $status = $snmp_status{$out->{batteryStatus}};
912d8679 3420 $reading = get_hashval($out->{batteryReading}, \%bat_reading);
669797e1 3421 $location = $out->{batteryLocationName};
3422 }
3423 else {
3424 $index = $out->{'Index'};
3425 $status = $out->{'Status'};
3426 $reading = $out->{'Reading'};
3427 $location = $out->{'Probe Name'};
3428 }
3429
3430 next BATTERY if blacklisted('bp', $index);
3431 $count{bat}++;
3432
98b224a3 3433 my $msg = sprintf 'Battery probe %d [%s] is %s',
669797e1 3434 $index, $location, $reading;
3435 report('chassis', $msg, $status2nagios{$status}, $index);
3436 }
3437 return;
3438}
3439
3440
3441#-----------------------------------------
3442# CHASSIS: Check amperage probes (power monitoring)
3443#-----------------------------------------
3444sub check_pwrmonitoring {
5ecf578c 3445 return if blacklisted('amp', 'all');
3446
669797e1 3447 my $index = undef;
3448 my $status = undef;
3449 my $reading = undef;
3450 my $location = undef;
3451 my $max_crit = undef;
3452 my $max_warn = undef;
3453 my $unit = undef;
3454 my @output = ();
3455
3456 if ($snmp) {
3457 my %amp_oid
3458 = (
3459 '1.3.6.1.4.1.674.10892.1.600.30.1.2.1' => 'amperageProbeIndex',
3460 '1.3.6.1.4.1.674.10892.1.600.30.1.5.1' => 'amperageProbeStatus',
3461 '1.3.6.1.4.1.674.10892.1.600.30.1.6.1' => 'amperageProbeReading',
3462 '1.3.6.1.4.1.674.10892.1.600.30.1.7.1' => 'amperageProbeType',
3463 '1.3.6.1.4.1.674.10892.1.600.30.1.8.1' => 'amperageProbeLocationName',
3464 '1.3.6.1.4.1.674.10892.1.600.30.1.10.1' => 'amperageProbeUpperCriticalThreshold',
3465 '1.3.6.1.4.1.674.10892.1.600.30.1.11.1' => 'amperageProbeUpperNonCriticalThreshold',
3466 '1.3.6.1.4.1.674.10892.1.600.30.1.16.1' => 'amperageProbeDiscreteReading',
3467 );
4cabd748 3468 my $result = undef;
3469 if ($opt{use_get_table}) {
3470 my $amperageProbeTable = '1.3.6.1.4.1.674.10892.1.600.30.1';
3471 $result = $snmp_session->get_table(-baseoid => $amperageProbeTable);
3472 }
3473 else {
3474 $result = $snmp_session->get_entries(-columns => [keys %amp_oid]);
3475 }
669797e1 3476
3477 # No pwrmonitoring is OK
3478 return 0 if !defined $result;
3479
3480 @output = @{ get_snmp_output($result, \%amp_oid) };
3481 }
3482 else {
3483 @output = @{ run_omreport("$omopt_chassis pwrmonitoring") };
3484 }
3485
3486 my %amp_type # Amperage probe types
3487 = (
3488 1 => 'amperageProbeTypeIsOther', # other than following values
3489 2 => 'amperageProbeTypeIsUnknown', # unknown
3490 3 => 'amperageProbeTypeIs1Point5Volt', # 1.5 amperage probe
3491 4 => 'amperageProbeTypeIs3Point3volt', # 3.3 amperage probe
3492 5 => 'amperageProbeTypeIs5Volt', # 5 amperage probe
3493 6 => 'amperageProbeTypeIsMinus5Volt', # -5 amperage probe
3494 7 => 'amperageProbeTypeIs12Volt', # 12 amperage probe
3495 8 => 'amperageProbeTypeIsMinus12Volt', # -12 amperage probe
3496 9 => 'amperageProbeTypeIsIO', # I/O probe
3497 10 => 'amperageProbeTypeIsCore', # Core probe
3498 11 => 'amperageProbeTypeIsFLEA', # FLEA (standby) probe
3499 12 => 'amperageProbeTypeIsBattery', # Battery probe
3500 13 => 'amperageProbeTypeIsTerminator', # SCSI Termination probe
3501 14 => 'amperageProbeTypeIs2Point5Volt', # 2.5 amperage probe
3502 15 => 'amperageProbeTypeIsGTL', # GTL (ground termination logic) probe
3503 16 => 'amperageProbeTypeIsDiscrete', # amperage probe with discrete reading
3504 23 => 'amperageProbeTypeIsPowerSupplyAmps', # Power Supply probe with reading in Amps
3505 24 => 'amperageProbeTypeIsPowerSupplyWatts', # Power Supply probe with reading in Watts
3506 25 => 'amperageProbeTypeIsSystemAmps', # System probe with reading in Amps
3507 26 => 'amperageProbeTypeIsSystemWatts', # System probe with reading in Watts
3508 );
3509
3510 my %amp_discrete
3511 = (
3512 1 => 'Good',
3513 2 => 'Bad',
3514 );
3515
3516 my %amp_unit
3517 = (
3518 'amperageProbeTypeIsPowerSupplyAmps' => 'hA', # tenths of Amps
3519 'amperageProbeTypeIsSystemAmps' => 'hA', # tenths of Amps
3520 'amperageProbeTypeIsPowerSupplyWatts' => 'W', # Watts
3521 'amperageProbeTypeIsSystemWatts' => 'W', # Watts
3522 'amperageProbeTypeIsDiscrete' => q{}, # discrete reading, no unit
3523 );
3524
3525 AMP:
3526 foreach my $out (@output) {
3527 if ($snmp) {
3528 $index = $out->{amperageProbeIndex} - 1;
3529 $status = $snmp_status{$out->{amperageProbeStatus}};
912d8679 3530 $reading = get_hashval($out->{amperageProbeType}, \%amp_type) eq 'amperageProbeTypeIsDiscrete'
3531 ? get_hashval($out->{amperageProbeDiscreteReading}, \%amp_discrete)
669797e1 3532 : $out->{amperageProbeReading};
3533 $location = $out->{amperageProbeLocationName};
3534 $max_crit = exists $out->{amperageProbeUpperCriticalThreshold}
3535 ? $out->{amperageProbeUpperCriticalThreshold} : 0;
3536 $max_warn = exists $out->{amperageProbeUpperNonCriticalThreshold}
3537 ? $out->{amperageProbeUpperNonCriticalThreshold} : 0;
3538 $unit = exists $amp_unit{$amp_type{$out->{amperageProbeType}}}
3539 ? $amp_unit{$amp_type{$out->{amperageProbeType}}} : 'mA';
3540 if ($unit eq 'hA') {
3541 $reading /= 10;
3542 $max_crit /= 10;
3543 $max_warn /= 10;
3544 $unit = 'A';
3545 }
3546 }
3547 else {
3548 $index = $out->{'Index'};
0be00f80 3549 next AMP if (!defined $index || $index !~ m/^\d+$/x);
669797e1 3550 $status = $out->{'Status'};
3551 $reading = $out->{'Reading'};
3552 $location = $out->{'Probe Name'};
3553 $max_crit = $out->{'Failure Threshold'} ne '[N/A]'
3554 ? $out->{'Failure Threshold'} : 0;
3555 $max_warn = $out->{'Warning Threshold'} ne '[N/A]'
3556 ? $out->{'Warning Threshold'} : 0;
3557 $reading =~ s{\A (\d+.*?)\s+([a-zA-Z]+) \s*\z}{$1}xms;
3558 $unit = $2;
3559 $max_warn =~ s{\A (\d+.*?)\s+[a-zA-Z]+ \s*\z}{$1}xms;
3560 $max_crit =~ s{\A (\d+.*?)\s+[a-zA-Z]+ \s*\z}{$1}xms;
3561 }
3562
78dbab97 3563 next AMP if blacklisted('amp', $index);
669797e1 3564 next AMP if $index !~ m{\A \d+ \z}xms;
3565 $count{amp}++;
3566
98b224a3 3567 my $msg = sprintf 'Amperage probe %d [%s] reads %s %s',
669797e1 3568 $index, $location, $reading, $unit, $status;
3569 report('chassis', $msg, $status2nagios{$status}, $index);
3570
3571 # Collect performance data
3572 if (defined $opt{perfdata}) {
3573 next AMP if $reading !~ m{\A \d+(\.\d+)? \z}xms; # discrete reading (not number)
48aeec0b 3574 my $label = join q{_}, 'pwr_mon', $index, lc $location;
3575 $label =~ s{\s}{_}gxms;
3576 push @perfdata, {
3577 label => $label,
3578 mini => "p$index",
3579 value => $reading,
c000d734 3580 unit => '',
48aeec0b 3581 warn => $max_warn,
3582 crit => $max_crit,
3583 };
669797e1 3584 }
3585 }
3586
3587 # Collect EXTRA performance data not found at first run. This is a
3588 # rather ugly hack
3589 if (defined $opt{perfdata} && !$snmp) {
3590 my $found = 0;
3591 my $index = 0;
3592 my %used = ();
3593
3594 # find used indexes
48aeec0b 3595 foreach (@perfdata) {
3596 if ($_->{label} =~ m/\A pwr_mon_(\d+)/xms) {
669797e1 3597 $used{$1} = 1;
3598 }
3599 }
3600
3601 AMP2:
3602 foreach my $line (@{ run_command("$omreport $omopt_chassis pwrmonitoring -fmt ssv") }) {
3603 chop $line;
3604 if ($line eq 'Location;Reading') {
3605 $found = 1;
3606 next AMP2;
3607 }
3608 if ($line eq q{}) {
3609 $found = 0;
3610 next AMP2;
3611 }
c000d734 3612 if ($found and $line =~ m/\A ([^;]+?) ; (\d*\.\d+) \s [AW] \z/xms) {
669797e1 3613 my $aname = lc $1;
3614 my $aval = $2;
669797e1 3615 $aname =~ s{\s}{_}gxms;
3616
3617 # don't use an existing index
3618 while (exists $used{$index}) { ++$index; }
3619
48aeec0b 3620 push @perfdata, {
3621 label => "pwr_mon_${index}_${aname}",
3622 mini => "p$index",
3623 value => $aval,
c000d734 3624 unit => '',
48aeec0b 3625 warn => 0,
3626 crit => 0,
3627 };
669797e1 3628 ++$index;
3629 }
3630 }
3631 }
3632
3633 return;
3634}
3635
3636
3637#-----------------------------------------
3638# CHASSIS: Check intrusion
3639#-----------------------------------------
3640sub check_intrusion {
5ecf578c 3641 return if blacklisted('intr', 'all');
3642
669797e1 3643 my $index = undef;
3644 my $status = undef;
3645 my $reading = undef;
3646 my @output = ();
3647
3648 if ($snmp) {
3649 my %int_oid
3650 = (
3651 '1.3.6.1.4.1.674.10892.1.300.70.1.2.1' => 'intrusionIndex',
3652 '1.3.6.1.4.1.674.10892.1.300.70.1.5.1' => 'intrusionStatus',
3653 '1.3.6.1.4.1.674.10892.1.300.70.1.6.1' => 'intrusionReading',
3654 );
4cabd748 3655 my $result = undef;
3656 if ($opt{use_get_table}) {
3657 my $intrusionTable = '1.3.6.1.4.1.674.10892.1.300.70.1';
3658 $result = $snmp_session->get_table(-baseoid => $intrusionTable);
3659 }
3660 else {
3661 $result = $snmp_session->get_entries(-columns => [keys %int_oid]);
3662 }
669797e1 3663
3664 # No intrusion is OK
3665 return 0 if !defined $result;
3666
3667 @output = @{ get_snmp_output($result, \%int_oid) };
3668 }
3669 else {
3670 @output = @{ run_omreport("$omopt_chassis intrusion") };
3671 }
3672
3673 my %int_reading
3674 = (
3675 1 => 'Not Breached', # chassis not breached and no uncleared breaches
3676 2 => 'Breached', # chassis currently breached
3677 3 => 'Breached Prior', # chassis breached prior to boot and has not been cleared
3678 4 => 'Breach Sensor Failure', # intrusion sensor has failed
3679 );
3680
3681 INTRUSION:
3682 foreach my $out (@output) {
3683 if ($snmp) {
3684 $index = $out->{intrusionIndex} - 1;
3685 $status = $snmp_status{$out->{intrusionStatus}};
912d8679 3686 $reading = get_hashval($out->{intrusionReading}, \%int_reading);
669797e1 3687 }
3688 else {
3689 $index = $out->{'Index'};
3690 $status = $out->{'Status'};
3691 $reading = $out->{'State'};
3692 }
3693
3694 next INTRUSION if blacklisted('intr', $index);
3695 $count{intr}++;
3696
3697 if ($status ne 'Ok') {
3698 my $msg = sprintf 'Chassis intrusion %d detected: %s',
3699 $index, $reading;
3700 report('chassis', $msg, $E_WARNING, $index);
3701 }
3702 # Ok
3703 else {
3704 my $msg = sprintf 'Chassis intrusion %d detection: %s (%s)',
3705 $index, $status, $reading;
3706 report('chassis', $msg, $E_OK, $index);
3707 }
3708 }
3709 return;
3710}
3711
3712
3713#-----------------------------------------
3714# CHASSIS: Check alert log
3715#-----------------------------------------
3716sub check_alertlog {
3717 return if $snmp; # Not supported with SNMP
3718
3719 my @output = @{ run_omreport("$omopt_system alertlog") };
3720 foreach my $out (@output) {
3721 ++$count{alert}{$out->{Severity}};
3722 }
3723
3724 # Create error messages and set exit value if appropriate
3725 my $err = 0;
3726 if ($count{alert}{'Critical'} > 0) { $err = $E_CRITICAL; }
3727 elsif ($count{alert}{'Non-Critical'} > 0) { $err = $E_WARNING; }
3728
3729 my $msg = sprintf 'Alert log content: %d critical, %d non-critical, %d ok',
3730 $count{alert}{'Critical'}, $count{alert}{'Non-Critical'}, $count{alert}{'Ok'};
3731 report('other', $msg, $err);
3732
3733 return;
3734}
3735
3736#-----------------------------------------
3737# CHASSIS: Check ESM log overall health
3738#-----------------------------------------
3739sub check_esmlog_health {
3740 my $health = 'Ok';
3741
3742 if ($snmp) {
3743 my $systemStateEventLogStatus = '1.3.6.1.4.1.674.10892.1.200.10.1.41.1';
3744 my $result = $snmp_session->get_request(-varbindlist => [$systemStateEventLogStatus]);
3745 if (!defined $result) {
98b224a3 3746 my $msg = sprintf 'SNMP ERROR [esmhealth]: %s',
669797e1 3747 $snmp_session->error;
3748 report('other', $msg, $E_UNKNOWN);
3749 }
3750 $health = $snmp_status{$result->{$systemStateEventLogStatus}};
3751 }
3752 else {
3753 foreach (@{ run_command("$omreport $omopt_system esmlog -fmt ssv") }) {
3754 if (m/\A Health;(.+) \z/xms) {
3755 $health = $1;
3756 chop $health;
3757 last;
3758 }
3759 }
3760 }
3761
3762 # If the overall health of the ESM log is other than "Ok", the
3763 # fill grade of the log is more than 80% and the log should be
3764 # cleared
3765 if ($health eq 'Ok') {
af7c7f76 3766 my $msg = sprintf 'ESM log health is Ok (less than 80%% full)';
669797e1 3767 report('other', $msg, $E_OK);
3768 }
3769 elsif ($health eq 'Critical') {
328d0a74 3770 my $msg = sprintf 'ESM log is 100%% full';
669797e1 3771 report('other', $msg, $status2nagios{$health});
3772 }
3773 else {
3774 my $msg = sprintf 'ESM log is more than 80%% full';
3775 report('other', $msg, $status2nagios{$health});
3776 }
3777
3778 return;
3779}
3780
3781#-----------------------------------------
3782# CHASSIS: Check ESM log
3783#-----------------------------------------
3784sub check_esmlog {
3785 my @output = ();
3786
3787 if ($snmp) {
3788 my %esm_oid
3789 = (
3790 '1.3.6.1.4.1.674.10892.1.300.40.1.7.1' => 'eventLogSeverityStatus',
3791 );
3792 my $result = $snmp_session->get_entries(-columns => [keys %esm_oid]);
3793
3794 # No entries is OK
3795 return if !defined $result;
3796
3797 @output = @{ get_snmp_output($result, \%esm_oid) };
3798 foreach my $out (@output) {
3799 ++$count{esm}{$snmp_status{$out->{eventLogSeverityStatus}}};
3800 }
3801 }
3802 else {
3803 @output = @{ run_omreport("$omopt_system esmlog") };
3804 foreach my $out (@output) {
3805 ++$count{esm}{$out->{Severity}};
3806 }
3807 }
3808
3809 # Create error messages and set exit value if appropriate
3810 my $err = 0;
3811 if ($count{esm}{'Critical'} > 0) { $err = $E_CRITICAL; }
3812 elsif ($count{esm}{'Non-Critical'} > 0) { $err = $E_WARNING; }
3813
3814 my $msg = sprintf 'ESM log content: %d critical, %d non-critical, %d ok',
3815 $count{esm}{'Critical'}, $count{esm}{'Non-Critical'}, $count{esm}{'Ok'};
3816 report('other', $msg, $err);
3817
3818 return;
3819}
3820
3821#
3822# Handy function for checking all storage components
3823#
3824sub check_storage {
3825 check_controllers();
3826 check_physical_disks();
3827 check_virtual_disks();
3828 check_cache_battery();
3829 check_connectors();
3830 check_enclosures();
3831 check_enclosure_fans();
3832 check_enclosure_pwr();
3833 check_enclosure_temp();
3834 check_enclosure_emms();
3835 return;
3836}
3837
3838
3839
3840#---------------------------------------------------------------------
3841# Info functions
3842#---------------------------------------------------------------------
3843
3844#
3845# Fetch output from 'omreport chassis info', put in sysinfo hash
3846#
3847sub get_omreport_chassis_info {
3848 if (open my $INFO, '-|', "$omreport $omopt_chassis info -fmt ssv") {
3849 my @lines = <$INFO>;
3850 close $INFO;
3851 foreach (@lines) {
3852 next if !m/\A (Chassis\sModel|Chassis\sService\sTag|Model|Service\sTag)/xms;
3853 my ($key, $val) = split /;/xms;
3854 $key =~ s{\s+\z}{}xms; # remove trailing whitespace
3855 $val =~ s{\s+\z}{}xms; # remove trailing whitespace
3856 if ($key eq 'Chassis Model' or $key eq 'Model') {
3857 $sysinfo{model} = $val;
3858 }
3859 if ($key eq 'Chassis Service Tag' or $key eq 'Service Tag') {
3860 $sysinfo{serial} = $val;
3861 }
3862 }
3863 }
3864 return;
3865}
3866
3867#
3868# Fetch output from 'omreport chassis bios', put in sysinfo hash
3869#
3870sub get_omreport_chassis_bios {
3871 if (open my $BIOS, '-|', "$omreport $omopt_chassis bios -fmt ssv") {
3872 my @lines = <$BIOS>;
3873 close $BIOS;
3874 foreach (@lines) {
3875 next if !m/;/xms;
3876 my ($key, $val) = split /;/xms;
3877 $key =~ s{\s+\z}{}xms; # remove trailing whitespace
3878 $val =~ s{\s+\z}{}xms; # remove trailing whitespace
3879 $sysinfo{bios} = $val if $key eq 'Version';
3880 $sysinfo{biosdate} = $val if $key eq 'Release Date';
3881 }
3882 }
3883 return;
3884}
3885
3886#
3887# Fetch output from 'omreport system operatingsystem', put in sysinfo hash
3888#
3889sub get_omreport_system_operatingsystem {
3890 if (open my $VER, '-|', "$omreport $omopt_system operatingsystem -fmt ssv") {
3891 my @lines = <$VER>;
3892 close $VER;
3893 foreach (@lines) {
3894 next if !m/;/xms;
3895 my ($key, $val) = split /;/xms;
3896 $key =~ s{\s+\z}{}xms; # remove trailing whitespace
3897 $val =~ s{\s+\z}{}xms; # remove trailing whitespace
3898 if ($key eq 'Operating System') {
3899 $sysinfo{osname} = $val;
3900 }
3901 elsif ($key eq 'Operating System Version') {
3902 $sysinfo{osver} = $val;
3903 }
3904 }
3905 }
3906 return;
3907}
3908
3909#
3910# Fetch output from 'omreport about', put in sysinfo hash
3911#
3912sub get_omreport_about {
3913 if (open my $OM, '-|', "$omreport about -fmt ssv") {
3914 my @lines = <$OM>;
3915 close $OM;
3916 foreach (@lines) {
3917 if (m/\A Version;(.+) \z/xms) {
3918 $sysinfo{om} = $1;
3919 chomp $sysinfo{om};
3920 }
3921 }
3922 }
3923 return;
3924}
3925
3926#
3927# Fetch chassis info via SNMP, put in sysinfo hash
3928#
3929sub get_snmp_chassis_info {
3930 my %chassis_oid
3931 = (
3932 '1.3.6.1.4.1.674.10892.1.300.10.1.9.1' => 'chassisModelName',
3933 '1.3.6.1.4.1.674.10892.1.300.10.1.11.1' => 'chassisServiceTagName',
3934 );
3935
3936 my $chassisInformationTable = '1.3.6.1.4.1.674.10892.1.300.10.1';
3937 my $result = $snmp_session->get_table(-baseoid => $chassisInformationTable);
3938
3939 if (defined $result) {
3940 foreach my $oid (keys %{ $result }) {
3941 if (exists $chassis_oid{$oid} and $chassis_oid{$oid} eq 'chassisModelName') {
3942 $sysinfo{model} = $result->{$oid};
3943 $sysinfo{model} =~ s{\s+\z}{}xms; # remove trailing whitespace
3944 }
3945 elsif (exists $chassis_oid{$oid} and $chassis_oid{$oid} eq 'chassisServiceTagName') {
3946 $sysinfo{serial} = $result->{$oid};
3947 }
3948 }
3949 }
3950 else {
3951 my $msg = sprintf 'SNMP ERROR getting chassis info: %s',
3952 $snmp_session->error;
3953 report('other', $msg, $E_UNKNOWN);
3954 }
3955 return;
3956}
3957
3958#
3959# Fetch BIOS info via SNMP, put in sysinfo hash
3960#
3961sub get_snmp_chassis_bios {
3962 my %bios_oid
3963 = (
3964 '1.3.6.1.4.1.674.10892.1.300.50.1.7.1.1' => 'systemBIOSReleaseDateName',
3965 '1.3.6.1.4.1.674.10892.1.300.50.1.8.1.1' => 'systemBIOSVersionName',
3966 );
3967
3968 my $systemBIOSTable = '1.3.6.1.4.1.674.10892.1.300.50.1';
3969 my $result = $snmp_session->get_table(-baseoid => $systemBIOSTable);
3970
3971 if (defined $result) {
3972 foreach my $oid (keys %{ $result }) {
3973 if (exists $bios_oid{$oid} and $bios_oid{$oid} eq 'systemBIOSReleaseDateName') {
3974 $sysinfo{biosdate} = $result->{$oid};
3975 $sysinfo{biosdate} =~ s{\A (\d{4})(\d{2})(\d{2}).*}{$2/$3/$1}xms;
3976 }
3977 elsif (exists $bios_oid{$oid} and $bios_oid{$oid} eq 'systemBIOSVersionName') {
3978 $sysinfo{bios} = $result->{$oid};
3979 }
3980 }
3981 }
3982 else {
3983 my $msg = sprintf 'SNMP ERROR getting BIOS info: %s',
3984 $snmp_session->error;
3985 report('other', $msg, $E_UNKNOWN);
3986 }
3987 return;
3988}
3989
3990#
3991# Fetch OS info via SNMP, put in sysinfo hash
3992#
3993sub get_snmp_system_operatingsystem {
3994 my %os_oid
3995 = (
3996 '1.3.6.1.4.1.674.10892.1.400.10.1.6.1' => 'operatingSystemOperatingSystemName',
3997 '1.3.6.1.4.1.674.10892.1.400.10.1.7.1' => 'operatingSystemOperatingSystemVersionName',
3998 );
3999
4000 my $operatingSystemTable = '1.3.6.1.4.1.674.10892.1.400.10.1';
4001 my $result = $snmp_session->get_table(-baseoid => $operatingSystemTable);
4002
4003 if (defined $result) {
4004 foreach my $oid (keys %{ $result }) {
4005 if (exists $os_oid{$oid} and $os_oid{$oid} eq 'operatingSystemOperatingSystemName') {
4006 $sysinfo{osname} = ($result->{$oid});
4007 }
4008 elsif (exists $os_oid{$oid} and $os_oid{$oid} eq 'operatingSystemOperatingSystemVersionName') {
4009 $sysinfo{osver} = $result->{$oid};
4010 }
4011 }
4012 }
4013 else {
4014 my $msg = sprintf 'SNMP ERROR getting OS info: %s',
4015 $snmp_session->error;
4016 report('other', $msg, $E_UNKNOWN);
4017 }
4018 return;
4019}
4020
4021#
4022# Fetch OMSA version via SNMP, put in sysinfo hash
4023#
4024sub get_snmp_about {
4025 my %omsa_oid
4026 = (
4027 '1.3.6.1.4.1.674.10892.1.100.10.0' => 'systemManagementSoftwareGlobalVersionName',
4028 );
4029 my $systemManagementSoftwareGroup = '1.3.6.1.4.1.674.10892.1.100';
4030 my $result = $snmp_session->get_table(-baseoid => $systemManagementSoftwareGroup);
4031 if (defined $result) {
4032 foreach my $oid (keys %{ $result }) {
4033 if (exists $omsa_oid{$oid} and $omsa_oid{$oid} eq 'systemManagementSoftwareGlobalVersionName') {
4034 $sysinfo{om} = ($result->{$oid});
4035 }
4036 }
4037 }
4038 else {
4039 my $msg = sprintf 'SNMP ERROR getting OMSA info: %s',
4040 $snmp_session->error;
4041 report('other', $msg, $E_UNKNOWN);
4042 }
4043 return;
4044}
4045
4046#
4047# Collects some information about the system
4048#
4049sub get_sysinfo
4050{
4051 # Get system model and serial number
4052 $snmp ? get_snmp_chassis_info() : get_omreport_chassis_info();
4053
4054 # Get BIOS information. Only if needed
4055 if ( $opt{okinfo} >= 1
4056 or $opt{debug}
4057 or (defined $opt{postmsg} and $opt{postmsg} =~ m/[%][bd]/xms) ) {
4058 $snmp ? get_snmp_chassis_bios() : get_omreport_chassis_bios();
4059 }
4060
f711f8c7 4061 # Get OMSA information. Only if needed
4062 if ($opt{okinfo} >= 3 or $opt{debug}) {
4063 $snmp ? get_snmp_about() : get_omreport_about();
4064 }
4065
669797e1 4066 # Return now if debug
4067 return if $opt{debug};
4068
4069 # Get OS information. Only if needed
4070 if (defined $opt{postmsg} and $opt{postmsg} =~ m/[%][or]/xms) {
4071 $snmp ? get_snmp_system_operatingsystem() : get_omreport_system_operatingsystem();
4072 }
4073
669797e1 4074 return;
4075}
4076
4077
4078# Helper function for running omreport when the results are strictly
4079# name=value pairs.
4080sub run_omreport_info {
4081 my $command = shift;
4082 my %output = ();
4083 my @keys = ();
4084
4085 # Run omreport and fetch output
4086 my $rawtext = slurp_command("$omreport $command -fmt ssv 2>&1");
4087
4088 # Parse output, store in array
4089 for ((split /\n/xms, $rawtext)) {
4090 if (m/\A Error/xms) {
4091 my $msg = "Problem running 'omreport $command': $_";
4092 report('other', $msg, $E_UNKNOWN);
4093 }
4094 next if !m/;/xms; # ignore lines with less than two fields
4095 my @vals = split m/;/xms;
4096 $output{$vals[0]} = $vals[1];
4097 }
4098
4099 # Finally, return the collected information
4100 return \%output;
4101}
4102
4103# Get various firmware information (BMC, RAC)
4104sub get_firmware_info {
4105 my @snmp_output = ();
4106 my %nrpe_output = ();
4107
4108 if ($snmp) {
4109 my %fw_oid
4110 = (
4111 '1.3.6.1.4.1.674.10892.1.300.60.1.7.1' => 'firmwareType',
4112 '1.3.6.1.4.1.674.10892.1.300.60.1.8.1' => 'firmwareTypeName',
4113 '1.3.6.1.4.1.674.10892.1.300.60.1.11.1' => 'firmwareVersionName',
4114 );
4115
4116 my $firmwareTable = '1.3.6.1.4.1.674.10892.1.300.60.1';
4117 my $result = $snmp_session->get_table(-baseoid => $firmwareTable);
4118
4119 # Some don't have this OID, this is ok
4120 if (!defined $result) {
4121 return;
4122 }
4123
4124 @snmp_output = @{ get_snmp_output($result, \%fw_oid) };
4125 }
4126 else {
4127 %nrpe_output = %{ run_omreport_info("$omopt_chassis info") };
4128 }
4129
4130 my %fw_type # Firmware types
4131 = (
4132 1 => 'other', # other than following values
4133 2 => 'unknown', # unknown
4134 3 => 'systemBIOS', # System BIOS
4135 4 => 'embeddedSystemManagementController', # Embedded System Management Controller
4136 5 => 'powerSupplyParallelingBoard', # Power Supply Paralleling Board
4137 6 => 'systemBackPlane', # System (Primary) Backplane
4138 7 => 'powerVault2XXSKernel', # PowerVault 2XXS Kernel
4139 8 => 'powerVault2XXSApplication', # PowerVault 2XXS Application
4140 9 => 'frontPanel', # Front Panel Controller
4141 10 => 'baseboardManagementController', # Baseboard Management Controller
4142 11 => 'hotPlugPCI', # Hot Plug PCI Controller
4143 12 => 'sensorData', # Sensor Data Records
4144 13 => 'peripheralBay', # Peripheral Bay Backplane
4145 14 => 'secondaryBackPlane', # Secondary Backplane for ESM 2 systems
4146 15 => 'secondaryBackPlaneESM3And4', # Secondary Backplane for ESM 3 and 4 systems
4147 16 => 'rac', # Remote Access Controller
4148 17 => 'imc' # Integrated Management Controller
4149 );
4150
4151
4152 if ($snmp) {
4153 foreach my $out (@snmp_output) {
4154 if ($fw_type{$out->{firmwareType}} eq 'baseboardManagementController') {
4155 $sysinfo{'bmc'} = 1;
4156 $sysinfo{'bmc_fw'} = $out->{firmwareVersionName};
4157 }
4158 elsif ($fw_type{$out->{firmwareType}} =~ m{\A rac|imc \z}xms) {
4159 my $name = $out->{firmwareTypeName}; $name =~ s/\s//gxms;
4160 $sysinfo{'rac'} = 1;
4161 $sysinfo{'rac_name'} = $name;
4162 $sysinfo{'rac_fw'} = $out->{firmwareVersionName};
4163 }
4164 }
4165 }
4166 else {
4167 foreach my $key (keys %nrpe_output) {
4168 next if !defined $nrpe_output{$key};
4169 if ($key eq 'BMC Version' or $key eq 'Baseboard Management Controller Version') {
4170 $sysinfo{'bmc'} = 1;
4171 $sysinfo{'bmc_fw'} = $nrpe_output{$key};
4172 }
4173 elsif ($key =~ m{\A (i?DRAC)\s*(\d?)\s+Version}xms) {
4174 my $name = "$1$2";
4175 $sysinfo{'rac'} = 1;
4176 $sysinfo{'rac_fw'} = $nrpe_output{$key};
4177 $sysinfo{'rac_name'} = $name;
4178 }
4179 }
4180 }
4181
4182 return;
4183}
4184
4185
4186
4187#=====================================================================
4188# Main program
4189#=====================================================================
4190
4191# Here we do the actual checking of components
4192# Check global status if applicable
4193if ($global) {
4194 $globalstatus = check_global();
4195}
4196
4197# Do multiple selected checks
4198if ($check{storage}) { check_storage(); }
4199if ($check{memory}) { check_memory(); }
4200if ($check{fans}) { check_fans(); }
4201if ($check{power}) { check_powersupplies(); }
4202if ($check{temp}) { check_temperatures(); }
4203if ($check{cpu}) { check_processors(); }
4204if ($check{voltage}) { check_volts(); }
4205if ($check{batteries}) { check_batteries(); }
4206if ($check{amperage}) { check_pwrmonitoring(); }
4207if ($check{intrusion}) { check_intrusion(); }
4208if ($check{alertlog}) { check_alertlog(); }
4209if ($check{esmlog}) { check_esmlog(); }
4210if ($check{esmhealth}) { check_esmlog_health(); }
4211
4212
4213#---------------------------------------------------------------------
4214# Finish up
4215#---------------------------------------------------------------------
4216
4217# Counter variable
4218%nagios_alert_count
4219 = (
4220 'OK' => 0,
4221 'WARNING' => 0,
4222 'CRITICAL' => 0,
4223 'UNKNOWN' => 0,
4224 );
4225
4226# Get system information
4227get_sysinfo();
4228
4229# Get firmware info if requested via option
4230if ($opt{okinfo} >= 1) {
4231 get_firmware_info();
4232}
4233
4234# Close SNMP session
4235if ($snmp) {
4236 $snmp_session->close;
4237}
4238
4239# Print messages
4240if ($opt{debug}) {
4241 print " System: $sysinfo{model}\n";
f711f8c7 4242 print " ServiceTag: $sysinfo{serial}";
4243 print q{ } x (25 - length $sysinfo{serial}), "OMSA version: $sysinfo{om}\n";
4244 print " BIOS/date: $sysinfo{bios} $sysinfo{biosdate}";
4245 print q{ } x (25 - length "$sysinfo{bios} $sysinfo{biosdate}"), "Plugin version: $VERSION\n";
669797e1 4246 if ($#report_storage >= 0) {
4247 print "-----------------------------------------------------------------------------\n";
4248 print " Storage Components \n";
4249 print "=============================================================================\n";
4250 print " STATE | ID | MESSAGE TEXT \n";
4251 print "---------+----------+--------------------------------------------------------\n";
4252 foreach (@report_storage) {
4253 my ($msg, $level, $nexus) = @{$_};
4254 print q{ } x (8 - length $reverse_exitcode{$level}) . "$reverse_exitcode{$level} | "
4255 . q{ } x (8 - length $nexus) . "$nexus | $msg\n";
4256 $nagios_alert_count{$reverse_exitcode{$level}}++;
4257 }
4258 }
4259 if ($#report_chassis >= 0) {
4260 print "-----------------------------------------------------------------------------\n";
4261 print " Chassis Components \n";
4262 print "=============================================================================\n";
1d003803 4263 print " STATE | ID | MESSAGE TEXT \n";
669797e1 4264 print "---------+------+------------------------------------------------------------\n";
4265 foreach (@report_chassis) {
4266 my ($msg, $level, $nexus) = @{$_};
4267 print q{ } x (8 - length $reverse_exitcode{$level}) . "$reverse_exitcode{$level} | "
4268 . q{ } x (4 - length $nexus) . "$nexus | $msg\n";
4269 $nagios_alert_count{$reverse_exitcode{$level}}++;
4270 }
4271 }
4272 if ($#report_other >= 0) {
4273 print "-----------------------------------------------------------------------------\n";
4274 print " Other messages \n";
4275 print "=============================================================================\n";
4276 print " STATE | MESSAGE TEXT \n";
4277 print "---------+-------------------------------------------------------------------\n";
4278 foreach (@report_other) {
4279 my ($msg, $level, $nexus) = @{$_};
4280 print q{ } x (8 - length $reverse_exitcode{$level}) . "$reverse_exitcode{$level} | $msg\n";
4281 $nagios_alert_count{$reverse_exitcode{$level}}++;
4282 }
4283 }
4284}
4285else {
4286 my $c = 0; # counter to determine linebreaks
4287
4288 # Run through each message, sorted by severity level
4289 ALERT:
4290 foreach (sort {$a->[1] < $b->[1]} (@report_storage, @report_chassis, @report_other)) {
4291 my ($msg, $level, $nexus) = @{ $_ };
4292 next ALERT if $level == $E_OK;
4293
4294 if (defined $opt{only}) {
4295 # If user wants only critical alerts
4296 next ALERT if ($opt{only} eq 'critical' and $level == $E_WARNING);
4297
4298 # If user wants only warning alerts
4299 next ALERT if ($opt{only} eq 'warning' and $level == $E_CRITICAL);
4300 }
4301
4302 # Prefix with service tag if specified with option '-i|--info'
4303 if ($opt{info}) {
4304 if (defined $opt{htmlinfo}) {
4305 $msg = '[<a href="' . warranty_url($sysinfo{serial})
4306 . "\">$sysinfo{serial}</a>] " . $msg;
4307 }
4308 else {
4309 $msg = "[$sysinfo{serial}] " . $msg;
4310 }
4311 }
4312
4313 # Prefix with nagios level if specified with option '--state'
4314 $msg = $reverse_exitcode{$level} . ": $msg" if $opt{state};
4315
4316 # Prefix with one-letter nagios level if specified with option '--short-state'
4317 $msg = (substr $reverse_exitcode{$level}, 0, 1) . ": $msg" if $opt{shortstate};
4318
4319 ($c++ == 0) ? print $msg : print $linebreak, $msg;
4320
4321 $nagios_alert_count{$reverse_exitcode{$level}}++;
4322 }
4323}
4324
4325# Determine our exit code
4326$exit_code = $E_OK;
4327$exit_code = $E_UNKNOWN if $nagios_alert_count{'UNKNOWN'} > 0;
4328$exit_code = $E_WARNING if $nagios_alert_count{'WARNING'} > 0;
4329$exit_code = $E_CRITICAL if $nagios_alert_count{'CRITICAL'} > 0;
4330
4331# Global status via SNMP.. extra safety check
4332if ($globalstatus != $E_OK && $exit_code == $E_OK && !defined $opt{only}) {
4333 print "OOPS! Something is wrong with this server, but I don't know what. ";
4334 print "The global system health status is $reverse_exitcode{$globalstatus}, ";
4335 print "but every component check is OK. This may be a bug in the Nagios plugin, ";
4336 print "please file a bug report.\n";
4337 exit $E_UNKNOWN;
4338}
4339
4340# Print OK message
4341if ($exit_code == $E_OK && defined $opt{only} && $opt{only} !~ m{\A critical|warning|chassis \z}xms && !$opt{debug}) {
4342 my %okmsg
4343 = ( 'storage' => "STORAGE OK - $count{pdisk} physical drives, $count{vdisk} logical drives",
4344 'fans' => $count{fan} == 0 && $blade ? 'OK - blade system with no fan probes' : "FANS OK - $count{fan} fan probes checked",
4345 'temp' => "TEMPERATURES OK - $count{temp} temperature probes checked",
4346 'memory' => "MEMORY OK - $count{dimm} memory modules checked",
4347 'power' => $count{power} == 0 ? 'OK - no instrumented power supplies found' : "POWER OK - $count{power} power supplies checked",
4348 'cpu' => "PROCESSORS OK - $count{cpu} processors checked",
4349 'voltage' => "VOLTAGE OK - $count{volt} voltage probes checked",
4350 'batteries' => $count{bat} == 0 ? 'OK - no batteries found' : "BATTERIES OK - $count{bat} batteries checked",
4351 'amperage' => $count{amp} == 0 ? 'OK - no power monitoring probes found' : "AMPERAGE OK - $count{amp} amperage (power monitoring) probes checked",
4352 'intrusion' => $count{intr} == 0 ? 'OK - no intrusion detection probes found' : "INTRUSION OK - $count{intr} intrusion detection probes checked",
4353 'alertlog' => $snmp ? 'OK - not supported via snmp' : "OK - Alert Log content: $count{alert}{Ok} ok, $count{alert}{'Non-Critical'} warning and $count{alert}{Critical} critical",
4354 'esmlog' => "OK - ESM Log content: $count{esm}{Ok} ok, $count{esm}{'Non-Critical'} warning and $count{esm}{Critical} critical",
4355 'esmhealth' => "ESM LOG OK - less than 80% used",
4356 );
4357
4358 print $okmsg{$opt{only}};
4359}
4360elsif ($exit_code == $E_OK && !$opt{debug}) {
4361 if (defined $opt{htmlinfo}) {
4362 printf q{OK - System: '<a href="%s">%s</a>', SN: '<a href="%s">%s</a>', hardware working fine},
4363 documentation_url($sysinfo{model}), $sysinfo{model},
4364 warranty_url($sysinfo{serial}), $sysinfo{serial};
4365 }
4366 else {
4367 printf q{OK - System: '%s', SN: '%s', hardware working fine},
4368 $sysinfo{model}, $sysinfo{serial};
4369 }
4370
4371 if ($check{storage}) {
4372 printf ', %d logical drives, %d physical drives',
4373 $count{vdisk}, $count{pdisk};
4374 }
4375 else {
4376 print ', not checking storage';
4377 }
4378
4379 if ($opt{okinfo} >= 1) {
4380 print $linebreak;
4381 printf q{----- BIOS='%s %s'}, $sysinfo{bios}, $sysinfo{biosdate};
4382
4383 if ($sysinfo{rac}) {
4384 printf q{, %s='%s'}, $sysinfo{rac_name}, $sysinfo{rac_fw};
4385 }
4386 if ($sysinfo{bmc}) {
4387 printf q{, BMC='%s'}, $sysinfo{bmc_fw};
4388 }
4389 }
4390
4391 if ($opt{okinfo} >= 2) {
4392 if ($check{storage}) {
4393 my @storageprint = ();
4394 foreach my $id (sort keys %{ $sysinfo{controller} }) {
4395 chomp $sysinfo{controller}{$id}{driver};
956cf4d1 4396 my $msg = sprintf q{----- Ctrl %s [%s]: Fw='%s', Dr='%s'},
669797e1 4397 $sysinfo{controller}{$id}{id}, $sysinfo{controller}{$id}{name},
4398 $sysinfo{controller}{$id}{firmware}, $sysinfo{controller}{$id}{driver};
956cf4d1 4399 if (defined $sysinfo{controller}{$id}{storport}) {
4400 $msg .= sprintf q{, Storport: '%s'}, $sysinfo{controller}{$id}{storport};
4401 }
4402 push @storageprint, $msg;
669797e1 4403 }
4404 foreach my $id (sort keys %{ $sysinfo{enclosure} }) {
956cf4d1 4405 push @storageprint, sprintf q{----- Encl %s [%s]: Fw='%s'},
669797e1 4406 $sysinfo{enclosure}{$id}->{id}, $sysinfo{enclosure}{$id}->{name},
4407 $sysinfo{enclosure}{$id}->{firmware};
4408 }
4409
4410 # print stuff
4411 foreach my $line (@storageprint) {
4412 print $linebreak, $line;
4413 }
4414 }
4415 }
4416
4417 if ($opt{okinfo} >= 3) {
4418 print "$linebreak----- OpenManage Server Administrator (OMSA) version: '$sysinfo{om}'";
4419 }
4420
4421}
4422else {
4423 if ($opt{extinfo}) {
4424 print $linebreak;
4425 if (defined $opt{htmlinfo}) {
4426 printf '------ SYSTEM: <a href="%s">%s</a>, SN: <a href="%s">%s</a>',
4427 documentation_url($sysinfo{model}), $sysinfo{model},
4428 warranty_url($sysinfo{serial}), $sysinfo{serial};
4429 }
4430 else {
4431 printf '------ SYSTEM: %s, SN: %s',
4432 $sysinfo{model}, $sysinfo{serial};
4433 }
4434 }
4435 if (defined $opt{postmsg}) {
4436 my $post = undef;
4437 if (-f $opt{postmsg}) {
4438 open my $POST, '<', $opt{postmsg}
4439 or ( print $linebreak
4440 and print "ERROR: Couldn't open post message file $opt{postmsg}: $!\n"
4441 and exit $E_UNKNOWN );
4442 $post = <$POST>;
4443 close $POST;
4444 chomp $post;
4445 }
4446 else {
4447 $post = $opt{postmsg};
4448 }
4449 if (defined $post) {
4450 print $linebreak;
4451 $post =~ s{[%]s}{$sysinfo{serial}}gxms;
4452 $post =~ s{[%]m}{$sysinfo{model}}gxms;
4453 $post =~ s{[%]b}{$sysinfo{bios}}gxms;
4454 $post =~ s{[%]d}{$sysinfo{biosdate}}gxms;
4455 $post =~ s{[%]o}{$sysinfo{osname}}gxms;
4456 $post =~ s{[%]r}{$sysinfo{osver}}gxms;
4457 $post =~ s{[%]p}{$count{pdisk}}gxms;
4458 $post =~ s{[%]l}{$count{vdisk}}gxms;
4459 $post =~ s{[%]n}{$linebreak}gxms;
4460 $post =~ s{[%]{2}}{%}gxms;
4461 print $post;
4462 }
4463 }
4464}
4465
7c03958b 4466# Reset the WARN signal
4467$SIG{__WARN__} = 'DEFAULT';
4468
cbbc270f 4469# Print any perl warnings that have occured
4470if (@perl_warnings) {
4471 foreach (@perl_warnings) {
4472 chop @$_;
4473 print "${linebreak}INTERNAL ERROR: @$_";
4474 }
4475 $exit_code = $E_UNKNOWN;
4476}
4477
669797e1 4478# Print performance data
48aeec0b 4479if (defined $opt{perfdata} && !$opt{debug} && @perfdata) {
669797e1 4480 my $lb = $opt{perfdata} eq 'multiline' ? "\n" : q{ }; # line break for perfdata
4481 print q{|};
4482
48aeec0b 4483 # Sort routine for performance data
4484 sub perfsort {
4485 my %order = ( fan => 0, pwr => 1, tem => 2, enc => 3, );
4486 return ($order{(substr $a->{label}, 0, 3)} cmp $order{(substr $b->{label}, 0, 3)}) ||
4487 $a->{label} cmp $b->{label};
669797e1 4488 }
4489
48aeec0b 4490 # Print performance data sorted
4491 my $type = $opt{perfdata} eq 'minimal' ? 'mini' : 'label';
e51ab9ec 4492 print join $lb, map { "$_->{$type}=$_->{value}$_->{unit};$_->{warn};$_->{crit}" } sort perfsort @perfdata;
669797e1 4493}
e133d101 4494
4495# Print a linebreak at the end
669797e1 4496print "\n" if !$opt{debug};
4497
4498# Exit with proper exit code
4499exit $exit_code;