plugins/check_overcr.c

   1 /******************************************************************************
   2 *
   3 * CHECK_OVERCR.C
   4 *
   5 * Program: Over-CR collector plugin for Nagios
   6 * License: GPL
   7 * Copyright (c) 1999 Ethan Galstad (nagios@nagios.org)
   8 *
   9 * $Id$
  10 *
  11 * Description:
  12 *
  13 * Notes:
  14 * - This plugin requires that Eric Molitors' Over-CR collector daemon
  15 *        be running on any UNIX boxes you want to monitor.  Over-CR
  16 *        is available from * http://www.molitor.org/overcr/
  17 *
  18 * Modifications:
  19 *
  20 * 08-11-999 Jacob Lundqvist <jaclu@grm.se>
  21 * Load was presented as a one digit percentage - changed to two digit
  22 *       value * before load of 11.2 was presented as "1.2%" (not very
  23 *       high). Warning and Critical params were int's, not very good
  24 *       for load, changed to doubles, so we can trap loadlimits like
  25 *       1.5.  Also added more informative LOAD error messages.
  26 *
  27 * License Information:
  28 *
  29 * This program is free software; you can redistribute it and/or modify
  30 * it under the terms of the GNU General Public License as published by
  31 * the Free Software Foundation; either version 2 of the License, or
  32 * (at your option) any later version.
  33 *
  34 * This program is distributed in the hope that it will be useful,
  35 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  36 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  37 * GNU General Public License for more details.
  38 *
  39 * You should have received a copy of the GNU General Public License
  40 * along with this program; if not, write to the Free Software
  41 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  42 *
  43 *****************************************************************************/
  44
  45 #include "config.h"
  46 #include "common.h"
  47 #include "netutils.h"
  48 #include "utils.h"
  49
  50 #define CHECK_NONE      0
  51 #define CHECK_LOAD1     1
  52 #define CHECK_LOAD5     2
  53 #define CHECK_LOAD15    4
  54 #define CHECK_DPU       8
  55 #define CHECK_PROCS     16
  56 #define CHECK_NETSTAT   32
  57 #define CHECK_UPTIME    64
  58
  59 #define PORT    2000
  60
  61 #define PROGNAME "check_overcr"
  62
  63 char *server_address = NULL;
  64 int server_port = PORT;
  65 double warning_value = 0L;
  66 double critical_value = 0L;
  67 int check_warning_value = FALSE;
  68 int check_critical_value = FALSE;
  69 int vars_to_check = CHECK_NONE;
  70 int cmd_timeout = 1;
  71
  72 int netstat_port = 0;
  73 char *disk_name = NULL;
  74 char *process_name = NULL;
  75
  76 int process_arguments (int, char **);
  77 void print_usage (void);
  78 void print_help (void);
  79
  80 int
  81 main (int argc, char **argv)
  82 {
  83         int result;
  84         char send_buffer[MAX_INPUT_BUFFER];
  85         char recv_buffer[MAX_INPUT_BUFFER];
  86         char output_message[MAX_INPUT_BUFFER];
  87         char temp_buffer[MAX_INPUT_BUFFER];
  88         char *temp_ptr = NULL;
  89         int found_disk = FALSE;
  90         unsigned long percent_used_disk_space = 100;
  91         double load;
  92         double load_1min;
  93         double load_5min;
  94         double load_15min;
  95         int port_connections = 0;
  96         int processes = 0;
  97         double uptime_raw_hours;
  98         int uptime_raw_minutes = 0;
  99         int uptime_days = 0;
 100         int uptime_hours = 0;
 101         int uptime_minutes = 0;
 102
 103         if (process_arguments (argc, argv) == ERROR)
 104                 usage ("Could not parse arguments\n");
 105
 106         /* initialize alarm signal handling */
 107         signal (SIGALRM, socket_timeout_alarm_handler);
 108
 109         /* set socket timeout */
 110         alarm (socket_timeout);
 111
 112         result = STATE_OK;
 113
 114         if (vars_to_check == CHECK_LOAD1 || vars_to_check == CHECK_LOAD5
 115                         || vars_to_check == CHECK_LOAD15) {
 116
 117                 strcpy (send_buffer, "LOAD\r\nQUIT\r\n");
 118                 result =
 119                         process_tcp_request2 (server_address, server_port, send_buffer,
 120                                                                                                                 recv_buffer, sizeof (recv_buffer));
 121                 if (result != STATE_OK)
 122                         return result;
 123
 124                 temp_ptr = (char *) strtok (recv_buffer, "\r\n");
 125                 if (temp_ptr == NULL) {
 126                         printf ("Invalid response from server - no load information\n");
 127                         return STATE_CRITICAL;
 128                 }
 129                 load_1min = strtod (temp_ptr, NULL);
 130                 temp_ptr = (char *) strtok (NULL, "\r\n");
 131                 if (temp_ptr == NULL) {
 132                         printf ("Invalid response from server after load 1\n");
 133                         return STATE_CRITICAL;
 134                 }
 135                 load_5min = strtod (temp_ptr, NULL);
 136                 temp_ptr = (char *) strtok (NULL, "\r\n");
 137                 if (temp_ptr == NULL) {
 138                         printf ("Invalid response from server after load 5\n");
 139                         return STATE_CRITICAL;
 140                 }
 141                 load_15min = strtod (temp_ptr, NULL);
 142
 143
 144                 switch (vars_to_check) {
 145                 case CHECK_LOAD1:
 146                         strcpy (temp_buffer, "1");
 147                         load = load_1min;
 148                         break;
 149                 case CHECK_LOAD5:
 150                         strcpy (temp_buffer, "5");
 151                         load = load_5min;
 152                         break;
 153                 default:
 154                         strcpy (temp_buffer, "15");
 155                         load = load_15min;
 156                         break;
 157                 }
 158
 159                 if (check_critical_value == TRUE && (load >= critical_value))
 160                         result = STATE_CRITICAL;
 161                 else if (check_warning_value == TRUE && (load >= warning_value))
 162                         result = STATE_WARNING;
 163                 sprintf (output_message, "Load %s - %s-min load average = %0.2f",
 164                                                  (result == STATE_OK) ? "ok" : "problem", temp_buffer, load);
 165         }
 166
 167
 168         else if (vars_to_check == CHECK_DPU) {
 169
 170                 sprintf (send_buffer, "DISKSPACE\r\n");
 171                 result =
 172                         process_tcp_request2 (server_address, server_port, send_buffer,
 173                                                                                                                 recv_buffer, sizeof (recv_buffer));
 174                 if (result != STATE_OK)
 175                         return result;
 176
 177                 for (temp_ptr = (char *) strtok (recv_buffer, " "); temp_ptr != NULL;
 178                                  temp_ptr = (char *) strtok (NULL, " ")) {
 179
 180                         if (!strcmp (temp_ptr, disk_name)) {
 181                                 found_disk = TRUE;
 182                                 temp_ptr = (char *) strtok (NULL, "%");
 183                                 if (temp_ptr == NULL) {
 184                                         printf ("Invalid response from server\n");
 185                                         return STATE_CRITICAL;
 186                                 }
 187                                 percent_used_disk_space = strtoul (temp_ptr, NULL, 10);
 188                                 break;
 189                         }
 190
 191                         temp_ptr = (char *) strtok (NULL, "\r\n");
 192                 }
 193
 194                 /* error if we couldn't find the info for the disk */
 195                 if (found_disk == FALSE) {
 196                         sprintf (output_message, "Error: Disk '%s' non-existent or not mounted",
 197                                                          disk_name);
 198                         result = STATE_CRITICAL;
 199                 }
 200
 201                 /* else check the disk space used */
 202                 else {
 203
 204                         if (check_critical_value == TRUE
 205                                         && (percent_used_disk_space >= critical_value)) result =
 206                                         STATE_CRITICAL;
 207                         else if (check_warning_value == TRUE
 208                                                          && (percent_used_disk_space >= warning_value)) result =
 209                                         STATE_WARNING;
 210
 211                         sprintf (output_message, "Disk %s - %lu%% used on %s",
 212                                                          (result == STATE_OK) ? "ok" : "problem",
 213                                                          percent_used_disk_space, disk_name);
 214                 }
 215         }
 216
 217         else if (vars_to_check == CHECK_NETSTAT) {
 218
 219                 sprintf (send_buffer, "NETSTAT %d\r\n", netstat_port);
 220                 result =
 221                         process_tcp_request2 (server_address, server_port, send_buffer,
 222                                                                                                                 recv_buffer, sizeof (recv_buffer));
 223                 if (result != STATE_OK)
 224                         return result;
 225
 226                 port_connections = strtod (recv_buffer, NULL);
 227
 228                 if (check_critical_value == TRUE && (port_connections >= critical_value))
 229                         result = STATE_CRITICAL;
 230                 else if (check_warning_value == TRUE
 231                                                  && (port_connections >= warning_value)) result = STATE_WARNING;
 232
 233                 sprintf (output_message, "Net %s - %d connection%s on port %d",
 234                                                  (result == STATE_OK) ? "ok" : "problem", port_connections,
 235                                                  (port_connections == 1) ? "" : "s", netstat_port);
 236         }
 237
 238         else if (vars_to_check == CHECK_PROCS) {
 239
 240                 sprintf (send_buffer, "PROCESS %s\r\n", process_name);
 241                 result =
 242                         process_tcp_request2 (server_address, server_port, send_buffer,
 243                                                                                                                 recv_buffer, sizeof (recv_buffer));
 244                 if (result != STATE_OK)
 245                         return result;
 246
 247                 temp_ptr = (char *) strtok (recv_buffer, "(");
 248                 if (temp_ptr == NULL) {
 249                         printf ("Invalid response from server\n");
 250                         return STATE_CRITICAL;
 251                 }
 252                 temp_ptr = (char *) strtok (NULL, ")");
 253                 if (temp_ptr == NULL) {
 254                         printf ("Invalid response from server\n");
 255                         return STATE_CRITICAL;
 256                 }
 257                 processes = strtod (temp_ptr, NULL);
 258
 259                 if (check_critical_value == TRUE && (processes >= critical_value))
 260                         result = STATE_CRITICAL;
 261                 else if (check_warning_value == TRUE && (processes >= warning_value))
 262                         result = STATE_WARNING;
 263
 264                 sprintf (output_message, "Process %s - %d instance%s of %s running",
 265                                                  (result == STATE_OK) ? "ok" : "problem", processes,
 266                                                  (processes == 1) ? "" : "s", process_name);
 267         }
 268
 269         else if (vars_to_check == CHECK_UPTIME) {
 270
 271                 sprintf (send_buffer, "UPTIME\r\n");
 272                 result =
 273                         process_tcp_request2 (server_address, server_port, send_buffer,
 274                                                                                                                 recv_buffer, sizeof (recv_buffer));
 275                 if (result != STATE_OK)
 276                         return result;
 277
 278                 uptime_raw_hours = strtod (recv_buffer, NULL);
 279                 uptime_raw_minutes = (unsigned long) (uptime_raw_hours * 60.0);
 280
 281                 if (check_critical_value == TRUE
 282                                 && (uptime_raw_minutes <= critical_value)) result = STATE_CRITICAL;
 283                 else if (check_warning_value == TRUE
 284                                                  && (uptime_raw_minutes <= warning_value)) result = STATE_WARNING;
 285
 286                 uptime_days = uptime_raw_minutes / 1440;
 287                 uptime_raw_minutes %= 1440;
 288                 uptime_hours = uptime_raw_minutes / 60;
 289                 uptime_raw_minutes %= 60;
 290                 uptime_minutes = uptime_raw_minutes;
 291
 292                 sprintf (output_message, "Uptime %s - Up %d days %d hours %d minutes",
 293                                                  (result == STATE_OK) ? "ok" : "problem", uptime_days,
 294                                                  uptime_hours, uptime_minutes);
 295         }
 296
 297         else {
 298                 strcpy (output_message, "Nothing to check!\n");
 299                 result = STATE_UNKNOWN;
 300         }
 301
 302         /* reset timeout */
 303         alarm (0);
 304
 305         printf ("%s\n", output_message);
 306
 307         return result;
 308 }
 309
 310
 311
 312
 313
 314 /* process command-line arguments */
 315 int
 316 process_arguments (int argc, char **argv)
 317 {
 318         int c;
 319
 320 #ifdef HAVE_GETOPT_H
 321         int option_index = 0;
 322         static struct option long_options[] = {
 323                 {"port", required_argument, 0, 'p'},
 324                 {"timeout", required_argument, 0, 't'},
 325                 {"critical", required_argument, 0, 'c'},
 326                 {"warning", required_argument, 0, 'w'},
 327                 {"variable", required_argument, 0, 'v'},
 328                 {"hostname", required_argument, 0, 'H'},
 329                 {"version", no_argument, 0, 'V'},
 330                 {"help", no_argument, 0, 'h'},
 331                 {0, 0, 0, 0}
 332         };
 333 #endif
 334
 335         /* no options were supplied */
 336         if (argc < 2)
 337                 return ERROR;
 338
 339         /* backwards compatibility */
 340         if (!is_option (argv[1])) {
 341                 server_address = argv[1];
 342                 argv[1] = argv[0];
 343                 argv = &argv[1];
 344                 argc--;
 345         }
 346
 347         for (c = 1; c < argc; c++) {
 348                 if (strcmp ("-to", argv[c]) == 0)
 349                         strcpy (argv[c], "-t");
 350                 else if (strcmp ("-wv", argv[c]) == 0)
 351                         strcpy (argv[c], "-w");
 352                 else if (strcmp ("-cv", argv[c]) == 0)
 353                         strcpy (argv[c], "-c");
 354         }
 355
 356         while (1) {
 357 #ifdef HAVE_GETOPT_H
 358                 c =
 359                         getopt_long (argc, argv, "+hVH:t:c:w:p:v:", long_options,
 360                                                                          &option_index);
 361 #else
 362                 c = getopt (argc, argv, "+hVH:t:c:w:p:v:");
 363 #endif
 364
 365                 if (c == -1 || c == EOF || c == 1)
 366                         break;
 367
 368                 switch (c) {
 369                 case '?':                                                                       /* print short usage statement if args not parsable */
 370                         printf ("%s: Unknown argument: %s\n\n", my_basename (argv[0]), optarg);
 371                         print_usage ();
 372                         exit (STATE_UNKNOWN);
 373                 case 'h':                                                                       /* help */
 374                         print_help ();
 375                         exit (STATE_OK);
 376                 case 'V':                                                                       /* version */
 377                         print_revision (my_basename (argv[0]), "$Revision$");
 378                         exit (STATE_OK);
 379                 case 'H':                                                                       /* hostname */
 380                         server_address = optarg;
 381                         break;
 382                 case 'p':                                                                       /* port */
 383                         if (is_intnonneg (optarg))
 384                                 server_port = atoi (optarg);
 385                         else
 386                                 terminate (STATE_UNKNOWN,
 387                                                                          "Server port an integer (seconds)\nType '%s -h' for additional help\n",
 388                                                                          PROGNAME);
 389                         break;
 390                 case 'v':                                                                       /* variable */
 391                         if (strcmp (optarg, "LOAD1") == 0)
 392                                 vars_to_check = CHECK_LOAD1;
 393                         else if (strcmp (optarg, "LOAD5") == 0)
 394                                 vars_to_check = CHECK_LOAD5;
 395                         else if (strcmp (optarg, "LOAD15") == 0)
 396                                 vars_to_check = CHECK_LOAD15;
 397                         else if (strcmp (optarg, "UPTIME") == 0)
 398                                 vars_to_check = CHECK_UPTIME;
 399                         else if (strstr (optarg, "PROC") == optarg) {
 400                                 vars_to_check = CHECK_PROCS;
 401                                 process_name = strscpy (process_name, optarg + 4);
 402                         }
 403                         else if (strstr (optarg, "NET") == optarg) {
 404                                 vars_to_check = CHECK_NETSTAT;
 405                                 netstat_port = atoi (optarg + 3);
 406                         }
 407                         else if (strstr (optarg, "DPU") == optarg) {
 408                                 vars_to_check = CHECK_DPU;
 409                                 disk_name = strscpy (disk_name, optarg + 3);
 410                         }
 411                         else
 412                                 return ERROR;
 413                         break;
 414                 case 'w':                                                                       /* warning threshold */
 415                         warning_value = strtoul (optarg, NULL, 10);
 416                         check_warning_value = TRUE;
 417                         break;
 418                 case 'c':                                                                       /* critical threshold */
 419                         critical_value = strtoul (optarg, NULL, 10);
 420                         check_critical_value = TRUE;
 421                         break;
 422                 case 't':                                                                       /* timeout */
 423                         socket_timeout = atoi (optarg);
 424                         if (socket_timeout <= 0)
 425                                 return ERROR;
 426                 }
 427
 428         }
 429         return OK;
 430 }
 431
 432
 433
 434
 435
 436 void
 437 print_usage (void)
 438 {
 439         printf
 440                 ("Usage: %s -H host [-p port] [-v variable] [-w warning] [-c critical] [-t timeout]\n",
 441                  PROGNAME);
 442 }
 443
 444
 445
 446
 447
 448 void
 449 print_help (void)
 450 {
 451         print_revision (PROGNAME, "$Revision$");
 452         printf
 453                 ("Copyright (c) 2000 Ethan Galstad/Karl DeBisschop\n\n"
 454                  "This plugin attempts to contact the Over-CR collector daemon running on the\n"
 455                  "remote UNIX server in order to gather the requested system information. This\n"
 456                  "plugin requres that Eric Molitors' Over-CR collector daemon be running on the\n"
 457                  "remote server. Over-CR can be downloaded from http://www.molitor.org/overcr\n"
 458                  "(This plugin was tested with version 0.99.53 of the Over-CR collector)\n\n");
 459         print_usage ();
 460         printf
 461                 ("\nOptions:\n"
 462                  "-H, --hostname=HOST\n"
 463                  "   Name of the host to check\n"
 464                  "-p, --port=INTEGER\n"
 465                  "   Optional port number (default: %d)\n"
 466                  "-v, --variable=STRING\n"
 467                  "   Variable to check.  Valid variables include:\n"
 468                  "     LOAD1         = 1 minute average CPU load\n"
 469                  "     LOAD5         = 5 minute average CPU load\n"
 470                  "     LOAD15        = 15 minute average CPU load\n"
 471                  "     DPU<filesys>  = percent used disk space on filesystem <filesys>\n"
 472                  "     PROC<process> = number of running processes with name <process>\n"
 473                  "     NET<port>     = number of active connections on TCP port <port>\n"
 474                  "     UPTIME        = system uptime in seconds\n"
 475                  " -w, --warning=INTEGER\n"
 476                  "   Threshold which will result in a warning status\n"
 477                  " -c, --critical=INTEGER\n"
 478                  "   Threshold which will result in a critical status\n"
 479                  " -t, --timeout=INTEGER\n"
 480                  "   Seconds before connection attempt times out (default: %d)\n"
 481                  "-h, --help\n"
 482                  "   Print this help screen\n"
 483                  "-V, --version\n"
 484                  "   Print version information\n\n"
 485                  "Notes:\n"
 486                  " - For the available options, the critical threshold value should always be\n"
 487                  "   higher than the warning threshold value, EXCEPT with the uptime variable\n"
 488                  "   (i.e. lower uptimes are worse).\n", PORT, DEFAULT_SOCKET_TIMEOUT);
 489 }