plugins/check_overcr.c

   1 /******************************************************************************
   2
   3  This program is free software; you can redistribute it and/or modify
   4  it under the terms of the GNU General Public License as published by
   5  the Free Software Foundation; either version 2 of the License, or
   6  (at your option) any later version.
   7
   8  This program is distributed in the hope that it will be useful,
   9  but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  GNU General Public License for more details.
  12
  13  You should have received a copy of the GNU General Public License
  14  along with this program; if not, write to the Free Software
  15  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  16
  17 ******************************************************************************/
  18
  19 const char *progname = "check_overcr";
  20 const char *revision = "$Revision$";
  21 const char *copyright = "2000-2003";
  22 const char *email = "nagiosplug-devel@lists.sourceforge.net";
  23
  24 #include "common.h"
  25 #include "netutils.h"
  26 #include "utils.h"
  27
  28 enum checkvar {
  29         NONE,
  30         LOAD1,
  31         LOAD5,
  32         LOAD15,
  33         DPU,
  34         PROCS,
  35         NETSTAT,
  36         UPTIME
  37 };
  38
  39 enum {
  40         PORT = 2000
  41 };
  42
  43 char *server_address = NULL;
  44 int server_port = PORT;
  45 double warning_value = 0L;
  46 double critical_value = 0L;
  47 int check_warning_value = FALSE;
  48 int check_critical_value = FALSE;
  49 enum checkvar vars_to_check = NONE;
  50 int cmd_timeout = 1;
  51
  52 int netstat_port = 0;
  53 char *disk_name = NULL;
  54 char *process_name = NULL;
  55         char send_buffer[MAX_INPUT_BUFFER];
  56
  57 int process_arguments (int, char **);
  58 void print_usage (void);
  59 void print_help (void);
  60
  61 int
  62 main (int argc, char **argv)
  63 {
  64         int result;
  65         char recv_buffer[MAX_INPUT_BUFFER];
  66         char temp_buffer[MAX_INPUT_BUFFER];
  67         char *temp_ptr = NULL;
  68         int found_disk = FALSE;
  69         unsigned long percent_used_disk_space = 100;
  70         double load;
  71         double load_1min;
  72         double load_5min;
  73         double load_15min;
  74         int port_connections = 0;
  75         int processes = 0;
  76         double uptime_raw_hours;
  77         int uptime_raw_minutes = 0;
  78         int uptime_days = 0;
  79         int uptime_hours = 0;
  80         int uptime_minutes = 0;
  81
  82         if (process_arguments (argc, argv) == ERROR)
  83                 usage ("Could not parse arguments\n");
  84
  85         /* initialize alarm signal handling */
  86         signal (SIGALRM, socket_timeout_alarm_handler);
  87
  88         /* set socket timeout */
  89         alarm (socket_timeout);
  90
  91         result = process_tcp_request2 (server_address,
  92                                        server_port,
  93                                        send_buffer,
  94                                        recv_buffer,
  95                                        sizeof (recv_buffer));
  96
  97         switch (vars_to_check) {
  98
  99         case LOAD1:
 100         case LOAD5:
 101         case LOAD15:
 102
 103                 if (result != STATE_OK)
 104                         die (result, _("Unknown error fetching load data\n"));
 105
 106                 temp_ptr = (char *) strtok (recv_buffer, "\r\n");
 107                 if (temp_ptr == NULL)
 108                         die (STATE_CRITICAL, _("Invalid response from server - no load information\n"));
 109                 load_1min = strtod (temp_ptr, NULL);
 110
 111                 temp_ptr = (char *) strtok (NULL, "\r\n");
 112                 if (temp_ptr == NULL)
 113                         die (STATE_CRITICAL, _("Invalid response from server after load 1\n"));
 114                 load_5min = strtod (temp_ptr, NULL);
 115
 116                 temp_ptr = (char *) strtok (NULL, "\r\n");
 117                 if (temp_ptr == NULL)
 118                         die (STATE_CRITICAL, _("Invalid response from server after load 5\n"));
 119                 load_15min = strtod (temp_ptr, NULL);
 120
 121                 switch (vars_to_check) {
 122                 case LOAD1:
 123                         strcpy (temp_buffer, "1");
 124                         load = load_1min;
 125                         break;
 126                 case LOAD5:
 127                         strcpy (temp_buffer, "5");
 128                         load = load_5min;
 129                         break;
 130                 default:
 131                         strcpy (temp_buffer, "15");
 132                         load = load_15min;
 133                         break;
 134                 }
 135
 136                 if (check_critical_value == TRUE && (load >= critical_value))
 137                         result = STATE_CRITICAL;
 138                 else if (check_warning_value == TRUE && (load >= warning_value))
 139                         result = STATE_WARNING;
 140
 141                 die (result,
 142                           _("Load %s - %s-min load average = %0.2f"),
 143                                                          state_text(result),
 144                           temp_buffer,
 145                           load);
 146
 147                         break;
 148
 149         case DPU:
 150
 151                 if (result != STATE_OK)
 152                         die (result, _("Unknown error fetching disk data\n"));
 153
 154                 for (temp_ptr = (char *) strtok (recv_buffer, " ");
 155                      temp_ptr != NULL;
 156                      temp_ptr = (char *) strtok (NULL, " ")) {
 157
 158                         if (!strcmp (temp_ptr, disk_name)) {
 159                                 found_disk = TRUE;
 160                                 temp_ptr = (char *) strtok (NULL, "%");
 161                                 if (temp_ptr == NULL)
 162                                         die (STATE_CRITICAL, _("Invalid response from server\n"));
 163                                 percent_used_disk_space = strtoul (temp_ptr, NULL, 10);
 164                                 break;
 165                         }
 166
 167                         temp_ptr = (char *) strtok (NULL, "\r\n");
 168                 }
 169
 170                 /* error if we couldn't find the info for the disk */
 171                 if (found_disk == FALSE)
 172                         die (STATE_CRITICAL,
 173                                    "Error: Disk '%s' non-existent or not mounted",
 174                                    disk_name);
 175
 176                 if (check_critical_value == TRUE && (percent_used_disk_space >= critical_value))
 177                         result = STATE_CRITICAL;
 178                 else if (check_warning_value == TRUE && (percent_used_disk_space >= warning_value))
 179                         result = STATE_WARNING;
 180
 181                 die (result, "Disk %s - %lu%% used on %s", state_text(result), percent_used_disk_space, disk_name);
 182
 183                 break;
 184
 185         case NETSTAT:
 186
 187                 if (result != STATE_OK)
 188                         die (result, _("Unknown error fetching network status\n"));
 189
 190                 port_connections = strtod (recv_buffer, NULL);
 191
 192                 if (check_critical_value == TRUE && (port_connections >= critical_value))
 193                         result = STATE_CRITICAL;
 194                 else if (check_warning_value == TRUE && (port_connections >= warning_value))
 195                         result = STATE_WARNING;
 196
 197                 die (result,
 198                            _("Net %s - %d connection%s on port %d"),
 199                            state_text(result),
 200                            port_connections,
 201                            (port_connections == 1) ? "" : "s",
 202                            netstat_port);
 203
 204                 break;
 205
 206         case PROCS:
 207
 208                 if (result != STATE_OK)
 209                         die (result, _("Unknown error fetching process status\n"));
 210
 211                 temp_ptr = (char *) strtok (recv_buffer, "(");
 212                 if (temp_ptr == NULL)
 213                         die (STATE_CRITICAL, _("Invalid response from server\n"));
 214
 215                 temp_ptr = (char *) strtok (NULL, ")");
 216                 if (temp_ptr == NULL)
 217                         die (STATE_CRITICAL, _("Invalid response from server\n"));
 218
 219                 processes = strtod (temp_ptr, NULL);
 220
 221                 if (check_critical_value == TRUE && (processes >= critical_value))
 222                         result = STATE_CRITICAL;
 223                 else if (check_warning_value == TRUE && (processes >= warning_value))
 224                         result = STATE_WARNING;
 225
 226                 die (result,
 227                            _("Process %s - %d instance%s of %s running"),
 228                            state_text(result),
 229                            processes,
 230                            (processes == 1) ? "" : "s",
 231                            process_name);
 232                 break;
 233
 234         case UPTIME:
 235
 236                 if (result != STATE_OK)
 237                         return result;
 238
 239                 uptime_raw_hours = strtod (recv_buffer, NULL);
 240                 uptime_raw_minutes = (unsigned long) (uptime_raw_hours * 60.0);
 241
 242                 if (check_critical_value == TRUE && (uptime_raw_minutes <= critical_value))
 243                         result = STATE_CRITICAL;
 244                 else if (check_warning_value == TRUE && (uptime_raw_minutes <= warning_value))
 245                         result = STATE_WARNING;
 246
 247                 uptime_days = uptime_raw_minutes / 1440;
 248                 uptime_raw_minutes %= 1440;
 249                 uptime_hours = uptime_raw_minutes / 60;
 250                 uptime_raw_minutes %= 60;
 251                 uptime_minutes = uptime_raw_minutes;
 252
 253                 die (result,
 254                            _("Uptime %s - Up %d days %d hours %d minutes"),
 255                            state_text(result),
 256                            uptime_days,
 257                            uptime_hours,
 258                            uptime_minutes);
 259                 break;
 260
 261         default:
 262                 die (STATE_UNKNOWN, _("Nothing to check!\n"));
 263                 break;
 264         }
 265
 266         /* reset timeout */
 267         alarm (0);
 268
 269         printf (_("Reached end of program with no data returned\n"));
 270
 271         return result;
 272 }
 273
 274
 275
 276
 277
 278 /* process command-line arguments */
 279 int
 280 process_arguments (int argc, char **argv)
 281 {
 282         int c;
 283
 284         int option_index = 0;
 285         static struct option long_options[] = {
 286                 {"port", required_argument, 0, 'p'},
 287                 {"timeout", required_argument, 0, 't'},
 288                 {"critical", required_argument, 0, 'c'},
 289                 {"warning", required_argument, 0, 'w'},
 290                 {"variable", required_argument, 0, 'v'},
 291                 {"hostname", required_argument, 0, 'H'},
 292                 {"version", no_argument, 0, 'V'},
 293                 {"help", no_argument, 0, 'h'},
 294                 {0, 0, 0, 0}
 295         };
 296
 297         /* no options were supplied */
 298         if (argc < 2)
 299                 return ERROR;
 300
 301         /* backwards compatibility */
 302         if (!is_option (argv[1])) {
 303                 server_address = argv[1];
 304                 argv[1] = argv[0];
 305                 argv = &argv[1];
 306                 argc--;
 307         }
 308
 309         for (c = 1; c < argc; c++) {
 310                 if (strcmp ("-to", argv[c]) == 0)
 311                         strcpy (argv[c], "-t");
 312                 else if (strcmp ("-wv", argv[c]) == 0)
 313                         strcpy (argv[c], "-w");
 314                 else if (strcmp ("-cv", argv[c]) == 0)
 315                         strcpy (argv[c], "-c");
 316         }
 317
 318         while (1) {
 319                 c = getopt_long (argc, argv, "+hVH:t:c:w:p:v:", long_options,
 320                                                                          &option_index);
 321
 322                 if (c == -1 || c == EOF || c == 1)
 323                         break;
 324
 325                 switch (c) {
 326                 case '?':                                                                       /* print short usage statement if args not parsable */
 327                         printf (_("%s: Unknown argument: %s\n\n"), progname, optarg);
 328                         print_usage ();
 329                         exit (STATE_UNKNOWN);
 330                 case 'h':                                                                       /* help */
 331                         print_help ();
 332                         exit (STATE_OK);
 333                 case 'V':                                                                       /* version */
 334                         print_revision (progname, "$Revision$");
 335                         exit (STATE_OK);
 336                 case 'H':                                                                       /* hostname */
 337                         server_address = optarg;
 338                         break;
 339                 case 'p':                                                                       /* port */
 340                         if (is_intnonneg (optarg))
 341                                 server_port = atoi (optarg);
 342                         else
 343                                 die (STATE_UNKNOWN,
 344                                                                          _("Server port an integer (seconds)\nType '%s -h' for additional help\n"),
 345                                                                          progname);
 346                         break;
 347                 case 'v':                                                                       /* variable */
 348                         if (strcmp (optarg, "LOAD") == 0) {
 349                                 strcpy (send_buffer, "LOAD\r\nQUIT\r\n");
 350                                 if (strcmp (optarg, "LOAD1") == 0)
 351                                         vars_to_check = LOAD1;
 352                                 else if (strcmp (optarg, "LOAD5") == 0)
 353                                         vars_to_check = LOAD5;
 354                                 else if (strcmp (optarg, "LOAD15") == 0)
 355                                         vars_to_check = LOAD15;
 356                         }
 357                         else if (strcmp (optarg, "UPTIME") == 0) {
 358                                 vars_to_check = UPTIME;
 359                                 strcpy (send_buffer, "UPTIME\r\n");
 360                         }
 361                         else if (strstr (optarg, "PROC") == optarg) {
 362                                 vars_to_check = PROCS;
 363                                 process_name = strscpy (process_name, optarg + 4);
 364                                 sprintf (send_buffer, "PROCESS %s\r\n", process_name);
 365                         }
 366                         else if (strstr (optarg, "NET") == optarg) {
 367                                 vars_to_check = NETSTAT;
 368                                 netstat_port = atoi (optarg + 3);
 369                                 sprintf (send_buffer, "NETSTAT %d\r\n", netstat_port);
 370                         }
 371                         else if (strstr (optarg, "DPU") == optarg) {
 372                                 vars_to_check = DPU;
 373                                 strcpy (send_buffer, "DISKSPACE\r\n");
 374                                 disk_name = strscpy (disk_name, optarg + 3);
 375                         }
 376                         else
 377                                 return ERROR;
 378                         break;
 379                 case 'w':                                                                       /* warning threshold */
 380                         warning_value = strtoul (optarg, NULL, 10);
 381                         check_warning_value = TRUE;
 382                         break;
 383                 case 'c':                                                                       /* critical threshold */
 384                         critical_value = strtoul (optarg, NULL, 10);
 385                         check_critical_value = TRUE;
 386                         break;
 387                 case 't':                                                                       /* timeout */
 388                         socket_timeout = atoi (optarg);
 389                         if (socket_timeout <= 0)
 390                                 return ERROR;
 391                 }
 392
 393         }
 394         return OK;
 395 }
 396 \f
 397 void
 398 print_usage (void)
 399 {
 400         printf (_("\
 401 Usage: %s -H host [-p port] [-v variable] [-w warning] [-c critical]\n\
 402   [-t timeout]\n"),
 403                 progname);
 404         printf (_(UT_HLP_VRS), progname, progname);
 405 }
 406
 407 void
 408 print_help (void)
 409 {
 410         char *myport;
 411         asprintf (&myport, "%d", PORT);
 412
 413         print_revision (progname, revision);
 414
 415         printf (_("Copyright (c) 1999 Ethan Galstad <nagios@nagios.org>\n"));
 416         printf (_(COPYRIGHT), copyright, email);
 417
 418         printf (_("\
 419 This plugin attempts to contact the Over-CR collector daemon running on the\n\
 420 remote UNIX server in order to gather the requested system information.\n\n"));
 421
 422         print_usage ();
 423
 424         printf (_(UT_HELP_VRSN));
 425
 426         printf (_(UT_HOST_PORT), 'p', myport);
 427
 428         printf (_("\
 429 -v, --variable=STRING\n\
 430    Variable to check.  Valid variables include:\n\
 431      LOAD1         = 1 minute average CPU load\n\
 432      LOAD5         = 5 minute average CPU load\n\
 433      LOAD15        = 15 minute average CPU load\n\
 434      DPU<filesys>  = percent used disk space on filesystem <filesys>\n\
 435      PROC<process> = number of running processes with name <process>\n\
 436      NET<port>     = number of active connections on TCP port <port>\n\
 437      UPTIME        = system uptime in seconds\n"));
 438
 439         printf (_("\
 440  -w, --warning=INTEGER\n\
 441    Threshold which will result in a warning status\n\
 442  -c, --critical=INTEGER\n\
 443    Threshold which will result in a critical status\n"));
 444
 445         printf (_(UT_TIMEOUT), DEFAULT_SOCKET_TIMEOUT);
 446
 447         printf (_("\
 448 Notes:\n\
 449  - For the available options, the critical threshold value should always be\n\
 450    higher than the warning threshold value, EXCEPT with the uptime variable\n\n"));
 451
 452         printf (_("\
 453  - This plugin requres that Eric Molitors' Over-CR collector daemon be\n\
 454    running on the remote server. Over-CR can be downloaded from\n\
 455    http://www.molitor.org/overcr (This plugin was tested with version\n\
 456    0.99.53 of the Over-CR collector)\n\n"));
 457
 458         printf (_(UT_SUPPORT));
 459 }