check_ntp.pl 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405
  1. #!/usr/bin/perl -w
  2. # (c)1999 Ian Cass, Knowledge Matters Ltd.
  3. # Read the GNU copyright stuff for all the legalese
  4. #
  5. # Check NTP time servers plugin. This plugin requires the ntpdate utility to
  6. # be installed on the system, however since it's part of the ntp suite, you
  7. # should already have it installed.
  8. #
  9. # $Id$
  10. #
  11. # Nothing clever done in this program - its a very simple bare basics hack to
  12. # get the job done.
  13. #
  14. # Things to do...
  15. # check @words[9] for time differences greater than +/- x secs & return a
  16. # warning.
  17. #
  18. # (c) 1999 Mark Jewiss, Knowledge Matters Limited
  19. # 22-9-1999, 12:45
  20. #
  21. # Modified script to accept 2 parameters or set defaults.
  22. # Now issues warning or critical alert is time difference is greater than the
  23. # time passed.
  24. #
  25. # These changes have not been tested completely due to the unavailability of a
  26. # server with the incorrect time.
  27. #
  28. # (c) 1999 Bo Kersey, VirCIO - Managed Server Solutions <bo@vircio.com>
  29. # 22-10-99, 12:17
  30. #
  31. # Modified the script to give useage if no parameters are input.
  32. #
  33. # Modified the script to check for negative as well as positive
  34. # time differences.
  35. #
  36. # Modified the script to work with ntpdate 3-5.93e Wed Apr 14 20:23:03 EDT 1999
  37. #
  38. # Modified the script to work with ntpdate's that return adjust or offset...
  39. #
  40. #
  41. # Script modified 2000 June 01 by William Pietri <william@bianca.com>
  42. #
  43. # Modified script to handle weird cases:
  44. # o NTP server doesn't respond (e.g., has died)
  45. # o Server has correct time but isn't suitable synchronization
  46. # source. This happens while starting up and if contact
  47. # with master has been lost.
  48. #
  49. # Modifed to run under Embedded Perl (sghosh@users.sf.net)
  50. # - combined logic some blocks together..
  51. #
  52. # Added ntpdate check for stratum 16 desynch peer (James Fidell) Feb 03, 2003
  53. #
  54. # ntpdate - offset is in seconds
  55. # changed ntpdc to ntpq - jitter/dispersion is in milliseconds
  56. #
  57. # Patch for for regex for stratum1 refid.
  58. require 5.004;
  59. use POSIX;
  60. use strict;
  61. use Getopt::Long;
  62. use vars qw($opt_V $opt_h $opt_H $opt_t $opt_w $opt_c $opt_j $opt_k $verbose $PROGNAME $def_jitter);
  63. use lib utils.pm;
  64. use utils qw($TIMEOUT %ERRORS &print_revision &support);
  65. $PROGNAME="check_ntp";
  66. sub print_help ();
  67. sub print_usage ();
  68. $ENV{'PATH'}='';
  69. $ENV{'BASH_ENV'}='';
  70. $ENV{'ENV'}='';
  71. # defaults in sec
  72. my $DEFAULT_OFFSET_WARN = 60; # 1 minute
  73. my $DEFAULT_OFFSET_CRIT = 120; # 2 minutes
  74. # default in millisec
  75. my $DEFAULT_JITTER_WARN = 5000; # 5 sec
  76. my $DEFAULT_JITTER_CRIT = 10000; # 10 sec
  77. Getopt::Long::Configure('bundling');
  78. GetOptions
  79. ("V" => \$opt_V, "version" => \$opt_V,
  80. "h" => \$opt_h, "help" => \$opt_h,
  81. "v" => \$verbose, "verbose" => \$verbose,
  82. "w=f" => \$opt_w, "warning=f" => \$opt_w, # offset|adjust warning if above this number
  83. "c=f" => \$opt_c, "critical=f" => \$opt_c, # offset|adjust critical if above this number
  84. "j=s" => \$opt_j, "jwarn=i" => \$opt_j, # jitter warning if above this number
  85. "k=s" => \$opt_k, "jcrit=i" => \$opt_k, # jitter critical if above this number
  86. "t=s" => \$opt_t, "timeout=i" => \$opt_t,
  87. "H=s" => \$opt_H, "hostname=s" => \$opt_H);
  88. if ($opt_V) {
  89. print_revision($PROGNAME,'$Revision$ ');
  90. exit $ERRORS{'OK'};
  91. }
  92. if ($opt_h) {
  93. print_help();
  94. exit $ERRORS{'OK'};
  95. }
  96. # jitter test params specified
  97. if (defined $opt_j || defined $opt_k ) {
  98. $def_jitter = 1;
  99. }
  100. $opt_H = shift unless ($opt_H);
  101. my $host = $1 if ($opt_H && $opt_H =~ m/^([0-9]+\.[0-9]+\.[0-9]+\.[0-9]+|[a-zA-Z][-a-zA-Z0-9]+(\.[a-zA-Z][-a-zA-Z0-9]+)*)$/);
  102. unless ($host) {
  103. print "No target host specified\n";
  104. print_usage();
  105. exit $ERRORS{'UNKNOWN'};
  106. }
  107. my ($timeout, $owarn, $ocrit, $jwarn, $jcrit);
  108. $timeout = $TIMEOUT;
  109. ($opt_t) && ($opt_t =~ /^([0-9]+)$/) && ($timeout = $1);
  110. $owarn = $DEFAULT_OFFSET_WARN;
  111. ($opt_w) && ($opt_w =~ /^([0-9.]+)$/) && ($owarn = $1);
  112. $ocrit = $DEFAULT_OFFSET_CRIT;
  113. ($opt_c) && ($opt_c =~ /^([0-9.]+)$/) && ($ocrit = $1);
  114. $jwarn = $DEFAULT_JITTER_WARN;
  115. ($opt_j) && ($opt_j =~ /^([0-9]+)$/) && ($jwarn = $1);
  116. $jcrit = $DEFAULT_JITTER_CRIT;
  117. ($opt_k) && ($opt_k =~ /^([0-9]+)$/) && ($jcrit = $1);
  118. if ($ocrit < $owarn ) {
  119. print "Critical offset should be larger than warning offset\n";
  120. print_usage();
  121. exit $ERRORS{"UNKNOWN"};
  122. }
  123. if ($def_jitter) {
  124. if ($opt_k < $opt_j) {
  125. print "Critical jitter should be larger than warning jitter\n";
  126. print_usage();
  127. exit $ERRORS{'UNKNOWN'};
  128. }
  129. }
  130. my $stratum = -1;
  131. my $ignoreret = 0;
  132. my $answer = undef;
  133. my $offset = undef;
  134. my $jitter = undef;
  135. my $syspeer = undef;
  136. my $candidates = 0;
  137. my $msg; # first line of output to print if format is invalid
  138. my $state = $ERRORS{'UNKNOWN'};
  139. my $ntpdate_error = $ERRORS{'UNKNOWN'};
  140. my $jitter_error = $ERRORS{'UNKNOWN'};
  141. # some systems don't have a proper ntpq (migrated from ntpdc)
  142. my $have_ntpq = undef;
  143. if ($utils::PATH_TO_NTPQ && -x $utils::PATH_TO_NTPQ ) {
  144. $have_ntpq = 1;
  145. }else{
  146. $have_ntpq = 0;
  147. }
  148. # Just in case of problems, let's not hang Nagios
  149. $SIG{'ALRM'} = sub {
  150. print ("ERROR: No response from ntp server (alarm)\n");
  151. exit $ERRORS{"UNKNOWN"};
  152. };
  153. alarm($timeout);
  154. ###
  155. ###
  156. ### First, check ntpdate
  157. ###
  158. ###
  159. if (!open (NTPDATE, "$utils::PATH_TO_NTPDATE -q $host 2>&1 |")) {
  160. print "Could not open ntpdate\n";
  161. exit $ERRORS{"UNKNOWN"};
  162. }
  163. while (<NTPDATE>) {
  164. print if ($verbose);
  165. $msg = $_ unless ($msg);
  166. if (/stratum\s(\d+)/) {
  167. $stratum = $1;
  168. }
  169. if (/(offset|adjust)\s+([-.\d]+)/i) {
  170. $offset = $2;
  171. # An offset of 0.000000 with an error is probably bogus. Actually,
  172. # it's probably always bogus, but let's be paranoid here.
  173. if ($offset == 0) { undef $offset;}
  174. $ntpdate_error = defined ($offset) ? $ERRORS{"OK"} : $ERRORS{"CRITICAL"};
  175. print "ntperr = $ntpdate_error \n" if $verbose;
  176. }
  177. if (/no server suitable for synchronization found/) {
  178. if ($stratum == 16) {
  179. $ntpdate_error = $ERRORS{"WARNING"};
  180. $msg = "Desynchronized peer server found";
  181. $ignoreret=1;
  182. }
  183. else {
  184. $ntpdate_error = $ERRORS{"CRITICAL"};
  185. $msg = "No suitable peer server found - ";
  186. }
  187. }
  188. }
  189. close (NTPDATE);
  190. # declare an error if we also get a non-zero return code from ntpdate
  191. # unless already set to critical
  192. if ( $? && !$ignoreret ) {
  193. print "stderr = $? : $! \n" if $verbose;
  194. $ntpdate_error = $ntpdate_error == $ERRORS{"CRITICAL"} ? $ERRORS{"CRITICAL"} : $ERRORS{"UNKNOWN"} ;
  195. print "ntperr = $ntpdate_error : $!\n" if $verbose;
  196. }
  197. ###
  198. ###
  199. ### Then scan xntpq/ntpq if it exists
  200. ### and look in the 11th column for jitter
  201. ###
  202. # Field 1: Tally Code ( Space, 'x','.','-','+','#','*','o')
  203. # Only match for '*' which implies sys.peer
  204. # or 'o' which implies pps.peer
  205. # If both exist, the last one is picked.
  206. # Field 2: address of the remote peer
  207. # Field 3: Refid of the clock (0.0.0.0 if unknown, WWWV/PPS/GPS/ACTS/USNO/PCS/... if Stratum1)
  208. # Field 4: stratum (0-15)
  209. # Field 5: Type of the peer: local (l), unicast (u), multicast (m)
  210. # broadcast (b); not sure about multicast/broadcast
  211. # Field 6: last packet receive (in seconds)
  212. # Field 7: polling interval
  213. # Field 8: reachability resgister (octal)
  214. # Field 9: delay
  215. # Field 10: offset
  216. # Field 11: dispersion/jitter
  217. #
  218. if ($have_ntpq) {
  219. if ( open(NTPQ,"$utils::PATH_TO_NTPQ -np $host 2>&1 |") ) {
  220. while (<NTPQ>) {
  221. print $_ if ($verbose);
  222. if ( /timed out/ ){
  223. $have_ntpq = 0 ;
  224. last ;
  225. }
  226. # number of candidates on <host> for sys.peer
  227. if (/^(\*|\+|\#|o])/) {
  228. ++$candidates;
  229. print "Candiate count= $candidates\n" if ($verbose);
  230. }
  231. # match sys.peer or pps.peer
  232. if (/^(\*|o)([-0-9.\s]+)\s+([-0-9A-Za-z.]+)\s+([-0-9.]+)\s+([lumb]+)\s+([-0-9m.]+)\s+([-0-9.]+)\s+([-0-9.]+)\s+([-0-9.]+)\s+([-0-9.]+)\s+([-0-9.]+)/) {
  233. $syspeer = $2;
  234. $stratum = $4;
  235. $jitter = $11;
  236. print "match $_ \n" if $verbose;
  237. if ($jitter > $jcrit) {
  238. print "Jitter_crit = $11 :$jcrit\n" if ($verbose);
  239. $jitter_error = $ERRORS{'CRITICAL'};
  240. } elsif ($jitter > $jwarn ) {
  241. print "Jitter_warn = $11 :$jwarn \n" if ($verbose);
  242. $jitter_error = $ERRORS{'WARNING'};
  243. } else {
  244. $jitter_error = $ERRORS{'OK'};
  245. }
  246. }
  247. }
  248. close NTPQ;
  249. }
  250. }
  251. if ($ntpdate_error != $ERRORS{'OK'}) {
  252. $state = $ntpdate_error;
  253. if ($ntpdate_error == $ERRORS{'WARNING'} ) {
  254. $answer = $msg . "\n";
  255. }
  256. else {
  257. $answer = $msg . "Server for ntp probably down\n";
  258. }
  259. if (defined($offset) && abs($offset) > $ocrit) {
  260. $state = $ERRORS{'CRITICAL'};
  261. $answer = "Server Error and offset $offset sec > +/- $ocrit sec\n";
  262. } elsif (defined($offset) && abs($offset) > $owarn) {
  263. $answer = "Server error and offset $offset sec > +/- $owarn sec\n";
  264. } elsif (defined($jitter) && abs($jitter) > $jcrit) {
  265. $answer = "Server error and jitter $jitter msec > +/- $jcrit msec\n";
  266. } elsif (defined($jitter) && abs($jitter) > $jwarn) {
  267. $answer = "Server error and jitter $jitter msec > +/- $jwarn msec\n";
  268. }
  269. } elsif ($have_ntpq && $jitter_error != $ERRORS{'OK'}) {
  270. $state = $jitter_error;
  271. $answer = "Jitter $jitter too high\n";
  272. if (defined($offset) && abs($offset) > $ocrit) {
  273. $state = $ERRORS{'CRITICAL'};
  274. $answer = "Jitter error and offset $offset sec > +/- $ocrit sec\n";
  275. } elsif (defined($offset) && abs($offset) > $owarn) {
  276. $answer = "Jitter error and offset $offset sec > +/- $owarn sec\n";
  277. } elsif (defined($jitter) && abs($jitter) > $jcrit) {
  278. $answer = "Jitter error and jitter $jitter msec > +/- $jcrit msec\n";
  279. } elsif (defined($jitter) && abs($jitter) > $jwarn) {
  280. $answer = "Jitter error and jitter $jitter msec > +/- $jwarn msec\n";
  281. }
  282. } elsif( !$have_ntpq ) { # no errors from ntpdate and no ntpq or ntpq timed out
  283. if (abs($offset) > $ocrit) {
  284. $state = $ERRORS{'CRITICAL'};
  285. $answer = "Offset $offset msec > +/- $ocrit sec\n";
  286. } elsif (abs($offset) > $owarn) {
  287. $state = $ERRORS{'WARNING'};
  288. $answer = "Offset $offset msec > +/- $owarn sec\n";
  289. } elsif (( abs($offset) > $owarn) && $def_jitter ) {
  290. $state = $ERRORS{'WARNING'};
  291. $answer = "Offset $offset msec > +/- $owarn sec, ntpq timed out\n";
  292. } elsif ( $def_jitter ) {
  293. $state = $ERRORS{'WARNING'};
  294. $answer = "Offset $offset secs, ntpq timed out\n";
  295. } else{
  296. $state = $ERRORS{'OK'};
  297. $answer = "Offset $offset secs \n";
  298. }
  299. } else { # no errors from ntpdate or ntpq
  300. if (abs($offset) > $ocrit) {
  301. $state = $ERRORS{'CRITICAL'};
  302. $answer = "Offset $offset msec > +/- $ocrit sec, jitter $jitter msec\n";
  303. } elsif (abs($jitter) > $jcrit ) {
  304. $state = $ERRORS{'CRITICAL'};
  305. $answer = "Jitter $jitter msec> +/- $jcrit msec, offset $offset sec \n";
  306. } elsif (abs($offset) > $owarn) {
  307. $state = $ERRORS{'WARNING'};
  308. $answer = "Offset $offset msec > +/- $owarn sec, jitter $jitter msec\n";
  309. } elsif (abs($jitter) > $jwarn ) {
  310. $state = $ERRORS{'WARNING'};
  311. $answer = "Jitter $jitter msec> +/- $jwarn msec, offset $offset sec \n";
  312. } else {
  313. $state = $ERRORS{'OK'};
  314. $answer = "Offset $offset secs, jitter $jitter msec, peer is stratum $stratum\n";
  315. }
  316. }
  317. foreach my $key (keys %ERRORS) {
  318. if ($state==$ERRORS{$key}) {
  319. print ("NTP $key: $answer");
  320. last;
  321. }
  322. }
  323. exit $state;
  324. ####
  325. #### subs
  326. sub print_usage () {
  327. print "Usage: $PROGNAME -H <host> [-w <warn>] [-c <crit>] [-j <warn>] [-k <crit>] [-v verbose]\n";
  328. }
  329. sub print_help () {
  330. print_revision($PROGNAME,'$Revision$');
  331. print "Copyright (c) 2003 Bo Kersey/Karl DeBisschop\n";
  332. print "\n";
  333. print_usage();
  334. print "
  335. Checks the local timestamp offset versus <host> with ntpdate
  336. Checks the jitter/dispersion of clock signal between <host> and its sys.peer with ntpq\n
  337. -w ( --warning)
  338. Clock offset in seconds at which a warning message will be generated.\n Defaults to $DEFAULT_OFFSET_WARN.
  339. -c (--critical)
  340. Clock offset in seconds at which a critical message will be generated.\n Defaults to $DEFAULT_OFFSET_CRIT.
  341. -j (--jwarn)
  342. Clock jitter in milliseconds at which a warning message will be generated.\n Defaults to $DEFAULT_JITTER_WARN.
  343. -k (--jcrit)
  344. Clock jitter in milliseconds at which a warning message will be generated.\n Defaults to $DEFAULT_JITTER_CRIT.\n
  345. If jitter/dispersion is specified with -j or -k and ntpq times out, then a
  346. warning is returned.
  347. ";
  348. support();
  349. }