plugins/check_ide_smart.c

   1 /*
   2  *  check_ide-smart v.1 - hacked version of ide-smart for Nagios
   3  *  Copyright (C) 2000 Robert Dale <rdale@digital-mission.com>
   4  *
   5  *  Nagios - http://www.nagios.org
   6  *
   7  *  Notes:
   8  *         ide-smart has the same functionality as before. Some return
   9  *         values were changed, otherwise the --net-saint option was added.
  10  *
  11  *         Run with:  check_ide-smart --net-saint [-d] <DRIVE>
  12  *         Where DRIVE is an IDE drive, ie. /dev/hda, /dev/hdb, /dev/hdc
  13  *
  14  *           - Returns 0 on no errors
  15  *           - Returns 1 on advisories
  16  *           - Returns 2 on prefailure
  17  *           - Returns -1 not too often
  18  *
  19  *  ide-smart 1.3 - IDE S.M.A.R.T. checking tool
  20  *  Copyright (C) 1998-1999 Ragnar Hojland Espinosa <ragnar@lightside.dhis.org>
  21  *                1998      Gadi Oxman <gadio@netvision.net.il>
  22  *
  23  *  This program is free software; you can redistribute it and/or modify
  24  *  it under the terms of the GNU General Public License as published by
  25  *  the Free Software Foundation; either version 2 of the License, or
  26  *  (at your option) any later version.
  27  *
  28  *  This program is distributed in the hope that it will be useful,
  29  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  31  *  GNU General Public License for more details.
  32  *
  33  *  You should have received a copy of the GNU General Public License
  34  *  along with this program; if not, write to the Free Software
  35  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  36  *
  37  * $Id$
  38  */
  39
  40 const char *progname = "check_ide_smart";
  41 const char *revision = "$Revision$";
  42 const char *copyright = "2000-2004";
  43 const char *email = "nagiosplug-devel@lists.sourceforge.net";
  44
  45 #include "common.h"
  46 #include "utils.h"
  47
  48 #include <sys/stat.h>
  49 #include <sys/ioctl.h>
  50 #include <fcntl.h>
  51 #include <linux/hdreg.h>
  52 #include <linux/types.h>
  53 #include <errno.h>
  54
  55 #define NR_ATTRIBUTES   30
  56
  57 #ifndef TRUE
  58 #define TRUE 1
  59 #endif  /*  */
  60
  61 #define PREFAILURE 2
  62 #define ADVISORY 1
  63 #define OPERATIONAL 0
  64 #define UNKNOWN -1
  65
  66 typedef struct threshold_s
  67 {
  68         __u8 id;
  69         __u8 threshold;
  70         __u8 reserved[10];
  71 }
  72 __attribute__ ((packed)) threshold_t;
  73
  74 typedef struct thresholds_s
  75 {
  76         __u16 revision;
  77         threshold_t thresholds[NR_ATTRIBUTES];
  78         __u8 reserved[18];
  79         __u8 vendor[131];
  80         __u8 checksum;
  81 }
  82 __attribute__ ((packed)) thresholds_t;
  83
  84 typedef struct value_s
  85 {
  86         __u8 id;
  87         __u16 status;
  88         __u8 value;
  89         __u8 vendor[8];
  90 }
  91 __attribute__ ((packed)) value_t;
  92
  93 typedef struct values_s
  94 {
  95         __u16 revision;
  96         value_t values[NR_ATTRIBUTES];
  97         __u8 offline_status;
  98         __u8 vendor1;
  99         __u16 offline_timeout;
 100         __u8 vendor2;
 101         __u8 offline_capability;
 102         __u16 smart_capability;
 103         __u8 reserved[16];
 104         __u8 vendor[125];
 105         __u8 checksum;
 106 }
 107 __attribute__ ((packed)) values_t;
 108
 109 struct
 110 {
 111         __u8 value;
 112         char *text;
 113 }
 114
 115 offline_status_text[] =
 116         {
 117                 {0x00, "NeverStarted"},
 118                 {0x02, "Completed"},
 119                 {0x04, "Suspended"},
 120                 {0x05, "Aborted"},
 121                 {0x06, "Failed"},
 122                 {0, 0}
 123         };
 124
 125 struct
 126 {
 127         __u8 value;
 128         char *text;
 129 }
 130
 131 smart_command[] =
 132         {
 133                 {SMART_ENABLE, "SMART_ENABLE"},
 134                 {SMART_DISABLE, "SMART_DISABLE"},
 135                 {SMART_IMMEDIATE_OFFLINE, "SMART_IMMEDIATE_OFFLINE"},
 136                 {SMART_AUTO_OFFLINE, "SMART_AUTO_OFFLINE"}
 137         };
 138
 139
 140 /* Index to smart_command table, keep in order */
 141 enum SmartCommand
 142         { SMART_CMD_ENABLE,
 143                 SMART_CMD_DISABLE,
 144                 SMART_CMD_IMMEDIATE_OFFLINE,
 145                 SMART_CMD_AUTO_OFFLINE
 146         };
 147
 148
 149 int
 150 main (int argc, char *argv[])
 151 {
 152         char *device = NULL;
 153         int command = -1;
 154         int o, longindex;
 155         int retval = 0;
 156
 157         thresholds_t thresholds;
 158         values_t values;
 159         int fd;
 160
 161         static struct option longopts[] = {
 162                 {"device", required_argument, 0, 'd'},
 163                 {"immediate", no_argument, 0, 'i'},
 164                 {"quiet-check", no_argument, 0, 'q'},
 165                 {"auto-on", no_argument, 0, '1'},
 166                 {"auto-off", no_argument, 0, '0'},
 167                 {"net-saint", no_argument, 0, 'n'},
 168                 {"help", no_argument, 0, 'h'},
 169                 {"version", no_argument, 0, 'V'}, {0, 0, 0, 0}
 170         };
 171
 172         setlocale (LC_ALL, "");
 173         bindtextdomain (PACKAGE, LOCALEDIR);
 174         textdomain (PACKAGE);
 175
 176         while (1) {
 177
 178                 o = getopt_long (argc, argv, "+d:iq10nhV", longopts, &longindex);
 179
 180                 if (o == -1 || o == EOF)
 181                         break;
 182
 183                 switch (o) {
 184                 case 'd':
 185                         device = optarg;
 186                         break;
 187                 case 'q':
 188                         command = 3;
 189                         break;
 190                 case 'i':
 191                         command = 2;
 192                         break;
 193                 case '1':
 194                         command = 1;
 195                         break;
 196                 case '0':
 197                         command = 0;
 198                         break;
 199                 case 'n':
 200                         command = 4;
 201                         break;
 202                 case 'h':
 203                         print_help ();
 204                         return STATE_OK;
 205                 case 'V':
 206                         print_revision (progname, revision);
 207                         return STATE_OK;
 208                 default:
 209                         usage2 (_("Unknown argument"), optarg);
 210                 }
 211
 212                 if (optind < argc) {
 213                         device = argv[optind];
 214                 }
 215
 216                 if (!device) {
 217                         show_help ();
 218                         show_version ();
 219                         return -1;
 220                 }
 221
 222                 fd = open (device, O_RDONLY);
 223
 224                 if (fd < 0) {
 225                         printf (_("CRITICAL - Couldn't open device: %s\n"), strerror (errno));
 226                         return 2;
 227                 }
 228
 229                 if (smart_cmd_simple (fd, SMART_CMD_ENABLE, 0, TRUE)) {
 230                         printf (_("CRITICAL - SMART_CMD_ENABLE\n"));
 231                         return 2;
 232                 }
 233
 234                 switch (command) {
 235                 case 0:
 236                         retval = smart_cmd_simple (fd, SMART_CMD_AUTO_OFFLINE, 0, TRUE);
 237                         break;
 238                 case 1:
 239                         retval = smart_cmd_simple (fd, SMART_CMD_AUTO_OFFLINE, 0xF8, TRUE);
 240                         break;
 241                 case 2:
 242                         retval = smart_cmd_simple (fd, SMART_CMD_IMMEDIATE_OFFLINE, 0, TRUE);
 243                         break;
 244                 case 3:
 245                         smart_read_values (fd, &values);
 246                         smart_read_thresholds (fd, &thresholds);
 247                         retval = values_not_passed (&values, &thresholds);
 248                         break;
 249                 case 4:
 250                         smart_read_values (fd, &values);
 251                         smart_read_thresholds (fd, &thresholds);
 252                         retval = net_saint (&values, &thresholds);
 253                         break;
 254                 default:
 255                         smart_read_values (fd, &values);
 256                         smart_read_thresholds (fd, &thresholds);
 257                         print_values (&values, &thresholds);
 258                         break;
 259                 }
 260                 close (fd);
 261         }
 262         return retval;
 263 }
 264
 265
 266
 267 char *
 268 get_offline_text (int status)
 269 {
 270         int i;
 271         for (i = 0; offline_status_text[i].text; i++) {
 272                 if (offline_status_text[i].value == status) {
 273                         return offline_status_text[i].text;
 274                 }
 275         }
 276         return "UNKNOW";
 277 }
 278
 279
 280
 281 int
 282 smart_read_values (int fd, values_t * values)
 283 {
 284         int e;
 285         __u8 args[4 + 512];
 286         args[0] = WIN_SMART;
 287         args[1] = 0;
 288         args[2] = SMART_READ_VALUES;
 289         args[3] = 1;
 290         if (ioctl (fd, HDIO_DRIVE_CMD, &args)) {
 291                 e = errno;
 292                 printf (_("CRITICAL - SMART_READ_VALUES: %s\n"), strerror (errno));
 293                 return e;
 294         }
 295         memcpy (values, args + 4, 512);
 296         return 0;
 297 }
 298
 299
 300
 301 int
 302 values_not_passed (values_t * p, thresholds_t * t)
 303 {
 304         value_t * value = p->values;
 305         threshold_t * threshold = t->thresholds;
 306         int failed = 0;
 307         int passed = 0;
 308         int i;
 309         for (i = 0; i < NR_ATTRIBUTES; i++) {
 310                 if (value->id && threshold->id && value->id == threshold->id) {
 311                         if (value->value <= threshold->threshold) {
 312                                 ++failed;
 313                         }
 314                         else {
 315                                 ++passed;
 316                         }
 317                 }
 318                 ++value;
 319                 ++threshold;
 320         }
 321         return (passed ? -failed : 2);
 322 }
 323
 324
 325
 326 int
 327 net_saint (values_t * p, thresholds_t * t)
 328 {
 329         value_t * value = p->values;
 330         threshold_t * threshold = t->thresholds;
 331         int status = OPERATIONAL;
 332         int prefailure = 0;
 333         int advisory = 0;
 334         int failed = 0;
 335         int passed = 0;
 336         int total = 0;
 337         int i;
 338         for (i = 0; i < NR_ATTRIBUTES; i++) {
 339                 if (value->id && threshold->id && value->id == threshold->id) {
 340                         if (value->value <= threshold->threshold) {
 341                                 ++failed;
 342                                 if (value->status & 1) {
 343                                         status = PREFAILURE;
 344                                         ++prefailure;
 345                                 }
 346                                 else {
 347                                         status = ADVISORY;
 348                                         ++advisory;
 349                                 }
 350                         }
 351                         else {
 352                                 ++passed;
 353                         }
 354                         ++total;
 355                 }
 356                 ++value;
 357                 ++threshold;
 358         }
 359         switch (status) {
 360         case PREFAILURE:
 361                 printf (_("CRITICAL - %d Harddrive PreFailure%cDetected! %d/%d tests failed.\n"),
 362                         prefailure,
 363                         prefailure > 1 ? 's' : ' ',
 364                         failed,
 365                   total);
 366                 break;
 367         case ADVISORY:
 368                 printf (_("WARNING - %d Harddrive Advisor%s Detected. %d/%d tests failed.\n"),
 369                         advisory,
 370                         advisory > 1 ? "ies" : "y",
 371                         failed,
 372                         total);
 373                 break;
 374         case OPERATIONAL:
 375                 printf (_("OK - Operational (%d/%d tests passed)\n"), passed, total);
 376                 break;
 377         default:
 378                 printf (_("ERROR - Status '%d' uknown. %d/%d tests passed\n"), status,
 379                                                 passed, total);
 380                 status = -1;
 381                 break;
 382         }
 383         return status;
 384 }
 385
 386
 387
 388 void
 389 print_value (value_t * p, threshold_t * t)
 390 {
 391         printf ("Id=%3d, Status=%2d {%s , %s}, Value=%3d, Threshold=%3d, %s\n",
 392                                         p->id, p->status, p->status & 1 ? "PreFailure" : "Advisory   ",
 393                                         p->status & 2 ? "OnLine " : "OffLine", p->value, t->threshold,
 394                                         p->value > t->threshold ? "Passed" : "Failed");
 395 }
 396
 397
 398
 399 void
 400 print_values (values_t * p, thresholds_t * t)
 401 {
 402         value_t * value = p->values;
 403         threshold_t * threshold = t->thresholds;
 404         int i;
 405         for (i = 0; i < NR_ATTRIBUTES; i++) {
 406                 if (value->id && threshold->id && value->id == threshold->id) {
 407                         print_value (value++, threshold++);
 408                 }
 409         }
 410         printf
 411                 (_("OffLineStatus=%d {%s}, AutoOffLine=%s, OffLineTimeout=%d minutes\n"),
 412                  p->offline_status,
 413                  get_offline_text (p->offline_status & 0x7f),
 414                  (p->offline_status & 0x80 ? "Yes" : "No"),
 415                  p->offline_timeout / 60);
 416         printf
 417                 (_("OffLineCapability=%d {%s %s %s}\n"),
 418                  p->offline_capability,
 419                  p->offline_capability & 1 ? "Immediate" : "",
 420                  p->offline_capability & 2 ? "Auto" : "",
 421                  p->offline_capability & 4 ? "AbortOnCmd" : "SuspendOnCmd");
 422         printf
 423                 (_("SmartRevision=%d, CheckSum=%d, SmartCapability=%d {%s %s}\n"),
 424                  p->revision,
 425                  p->checksum,
 426                  p->smart_capability,
 427                  p->smart_capability & 1 ? "SaveOnStandBy" : "",
 428                  p->smart_capability & 2 ? "AutoSave" : "");
 429 }
 430
 431
 432
 433 void
 434 print_thresholds (thresholds_t * p)
 435 {
 436         threshold_t * threshold = p->thresholds;
 437         int i;
 438         printf ("\n");
 439         printf ("SmartRevision=%d\n", p->revision);
 440         for (i = 0; i < NR_ATTRIBUTES; i++) {
 441                 if (threshold->id) {
 442                         printf ("Id=%3d, Threshold=%3d\n", threshold->id,
 443                                                         threshold->threshold); }
 444                 ++threshold;
 445         }
 446         printf ("CheckSum=%d\n", p->checksum);
 447 }
 448
 449 int
 450 smart_cmd_simple (int fd, enum SmartCommand command, __u8 val0,
 451                                                                         char show_error)
 452 {
 453         int e = 0;
 454         __u8 args[4];
 455         args[0] = WIN_SMART;
 456         args[1] = val0;
 457         args[2] = smart_command[command].value;
 458         args[3] = 0;
 459         if (ioctl (fd, HDIO_DRIVE_CMD, &args)) {
 460                 e = errno;
 461                 if (show_error) {
 462                         printf (_("CRITICAL - %s: %s\n"), smart_command[command].text, strerror (errno));
 463                 }
 464         }
 465         return e;
 466 }
 467
 468
 469
 470 int
 471 smart_read_thresholds (int fd, thresholds_t * thresholds)
 472 {
 473         int e;
 474         __u8 args[4 + 512];
 475         args[0] = WIN_SMART;
 476   args[1] = 0;
 477   args[2] = SMART_READ_THRESHOLDS;
 478   args[3] = 1;
 479         if (ioctl (fd, HDIO_DRIVE_CMD, &args)) {
 480                 e = errno;
 481                 printf (_("CRITICAL - SMART_READ_THRESHOLDS: %s\n"), strerror (errno));
 482                 return e;
 483         }
 484         memcpy (thresholds, args + 4, 512);
 485         return 0;
 486 }
 487
 488
 489 void
 490 print_help ()
 491 {
 492         print_revision (progname, revision);
 493
 494         printf ("Nagios feature - 1999 Robert Dale <rdale@digital-mission.com>\n");
 495         printf ("(C) 1999 Ragnar Hojland Espinosa <ragnar@lightside.dhis.org>\n");
 496         printf (COPYRIGHT, copyright, email);
 497
 498         printf ("\
 499 Usage: %s [DEVICE] [OPTION]\n\
 500  -d, --device=DEVICE\n\
 501     Select device DEVICE\n\
 502  -i, --immediate\n\
 503     Perform immediately offline tests\n\
 504  -q, --quiet-check\n\
 505     Returns the number of failed tests\n\
 506  -1, --auto-on\n\
 507     Turn on automatic offline tests\n\
 508  -0, --auto-off\n\
 509     Turn off automatic offline tests\n\
 510  -n, --net-saint\n\
 511     Output suitable for Net Saint\n", progname);
 512 }
 513
 514
 515 void
 516 print_usage (void)
 517 {
 518         printf ("Usage: %s \n"), progname);
 519 }