1 /**
2 * collectd-nagios - src/collectd-nagios.c
3 * Copyright (C) 2008-2010 Florian octo Forster
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Florian octo Forster <octo at collectd.org>
25 **/
27 #if HAVE_CONFIG_H
28 # include "config.h"
29 #endif
31 #if !defined(__GNUC__) || !__GNUC__
32 # define __attribute__(x) /**/
33 #endif
35 #include <stdlib.h>
36 #include <unistd.h>
37 #include <stdio.h>
38 #include <errno.h>
39 #include <string.h>
40 #include <strings.h>
41 #include <assert.h>
43 #if NAN_STATIC_DEFAULT
44 # include <math.h>
45 /* #endif NAN_STATIC_DEFAULT*/
46 #elif NAN_STATIC_ISOC
47 # ifndef __USE_ISOC99
48 # define DISABLE_ISOC99 1
49 # define __USE_ISOC99 1
50 # endif /* !defined(__USE_ISOC99) */
51 # include <math.h>
52 # if DISABLE_ISOC99
53 # undef DISABLE_ISOC99
54 # undef __USE_ISOC99
55 # endif /* DISABLE_ISOC99 */
56 /* #endif NAN_STATIC_ISOC */
57 #elif NAN_ZERO_ZERO
58 # include <math.h>
59 # ifdef NAN
60 # undef NAN
61 # endif
62 # define NAN (0.0 / 0.0)
63 # ifndef isnan
64 # define isnan(f) ((f) != (f))
65 # endif /* !defined(isnan) */
66 # ifndef isfinite
67 # define isfinite(f) (((f) - (f)) == 0.0)
68 # endif
69 # ifndef isinf
70 # define isinf(f) (!isfinite(f) && !isnan(f))
71 # endif
72 #endif /* NAN_ZERO_ZERO */
74 #include "libcollectdclient/collectd/client.h"
76 #define RET_OKAY 0
77 #define RET_WARNING 1
78 #define RET_CRITICAL 2
79 #define RET_UNKNOWN 3
81 #define CON_NONE 0
82 #define CON_AVERAGE 1
83 #define CON_SUM 2
84 #define CON_PERCENTAGE 3
86 struct range_s
87 {
88 double min;
89 double max;
90 int invert;
91 };
92 typedef struct range_s range_t;
94 extern char *optarg;
95 extern int optind, opterr, optopt;
97 static char *socket_file_g = NULL;
98 static char *value_string_g = NULL;
99 static char *hostname_g = NULL;
101 static range_t range_critical_g;
102 static range_t range_warning_g;
103 static int consolitation_g = CON_NONE;
104 static _Bool nan_is_error_g = 0;
106 static char **match_ds_g = NULL;
107 static size_t match_ds_num_g = 0;
109 /* `strdup' is an XSI extension. I don't want to pull in all of XSI just for
110 * that, so here's an own implementation.. It's easy enough. The GCC attributes
111 * are supposed to get good performance.. -octo */
112 __attribute__((malloc, nonnull (1)))
113 static char *cn_strdup (const char *str) /* {{{ */
114 {
115 size_t strsize;
116 char *ret;
118 strsize = strlen (str) + 1;
119 ret = (char *) malloc (strsize);
120 if (ret != NULL)
121 memcpy (ret, str, strsize);
122 return (ret);
123 } /* }}} char *cn_strdup */
125 static int filter_ds (size_t *values_num,
126 double **values, char ***values_names)
127 {
128 gauge_t *new_values;
129 char **new_names;
131 size_t i;
133 if (match_ds_g == NULL)
134 return (RET_OKAY);
136 new_values = (gauge_t *)calloc (match_ds_num_g, sizeof (*new_values));
137 if (new_values == NULL)
138 {
139 fprintf (stderr, "calloc failed: %s\n", strerror (errno));
140 return (RET_UNKNOWN);
141 }
143 new_names = (char **)calloc (match_ds_num_g, sizeof (*new_names));
144 if (new_names == NULL)
145 {
146 fprintf (stderr, "calloc failed: %s\n", strerror (errno));
147 free (new_values);
148 return (RET_UNKNOWN);
149 }
151 for (i = 0; i < match_ds_num_g; i++)
152 {
153 size_t j;
155 /* match_ds_g keeps pointers into argv but the names will be freed */
156 new_names[i] = cn_strdup (match_ds_g[i]);
157 if (new_names[i] == NULL)
158 {
159 fprintf (stderr, "cn_strdup failed: %s\n", strerror (errno));
160 free (new_values);
161 for (j = 0; j < i; j++)
162 free (new_names[j]);
163 free (new_names);
164 return (RET_UNKNOWN);
165 }
167 for (j = 0; j < *values_num; j++)
168 if (strcasecmp (new_names[i], (*values_names)[j]) == 0)
169 break;
171 if (j == *values_num)
172 {
173 printf ("ERROR: DS `%s' is not available.\n", new_names[i]);
174 free (new_values);
175 for (j = 0; j <= i; j++)
176 free (new_names[j]);
177 free (new_names);
178 return (RET_CRITICAL);
179 }
181 new_values[i] = (*values)[j];
182 }
184 free (*values);
185 for (i = 0; i < *values_num; i++)
186 free ((*values_names)[i]);
187 free (*values_names);
189 *values = new_values;
190 *values_names = new_names;
191 *values_num = match_ds_num_g;
192 return (RET_OKAY);
193 } /* int filter_ds */
195 static void parse_range (char *string, range_t *range)
196 {
197 char *min_ptr;
198 char *max_ptr;
200 if (*string == '@')
201 {
202 range->invert = 1;
203 string++;
204 }
206 max_ptr = strchr (string, ':');
207 if (max_ptr == NULL)
208 {
209 min_ptr = NULL;
210 max_ptr = string;
211 }
212 else
213 {
214 min_ptr = string;
215 *max_ptr = '\0';
216 max_ptr++;
217 }
219 assert (max_ptr != NULL);
221 /* `10' == `0:10' */
222 if (min_ptr == NULL)
223 range->min = 0.0;
224 /* :10 == ~:10 == -inf:10 */
225 else if ((*min_ptr == '\0') || (*min_ptr == '~'))
226 range->min = NAN;
227 else
228 range->min = atof (min_ptr);
230 if ((*max_ptr == '\0') || (*max_ptr == '~'))
231 range->max = NAN;
232 else
233 range->max = atof (max_ptr);
234 } /* void parse_range */
236 static int match_range (range_t *range, double value)
237 {
238 int ret = 0;
240 if (!isnan (range->min) && (range->min > value))
241 ret = 1;
242 if (!isnan (range->max) && (range->max < value))
243 ret = 1;
245 return (((ret - range->invert) == 0) ? 0 : 1);
246 } /* int match_range */
248 __attribute__((noreturn))
249 static void usage (const char *name)
250 {
251 fprintf (stderr, "Usage: %s <-s socket> <-n value_spec> <-H hostname> [options]\n"
252 "\n"
253 "Valid options are:\n"
254 " -s <socket> Path to collectd's UNIX-socket.\n"
255 " -n <v_spec> Value specification to get from collectd.\n"
256 " Format: `plugin-instance/type-instance'\n"
257 " -d <ds> Select the DS to examine. May be repeated to examine multiple\n"
258 " DSes. By default all DSes are used.\n"
259 " -g <consol> Method to use to consolidate several DSes.\n"
260 " See below for a list of valid arguments.\n"
261 " -H <host> Hostname to query the values for.\n"
262 " -c <range> Critical range\n"
263 " -w <range> Warning range\n"
264 " -m Treat \"Not a Number\" (NaN) as critical (default: warning)\n"
265 "\n"
266 "Consolidation functions:\n"
267 " none: Apply the warning- and critical-ranges to each data-source\n"
268 " individually.\n"
269 " average: Calculate the average of all matching DSes and apply the\n"
270 " warning- and critical-ranges to the calculated average.\n"
271 " sum: Apply the ranges to the sum of all DSes.\n"
272 " percentage: Apply the ranges to the ratio (in percent) of the first value\n"
273 " and the sum of all values."
274 "\n", name);
275 exit (1);
276 } /* void usage */
278 static int do_listval (lcc_connection_t *connection)
279 {
280 lcc_identifier_t *ret_ident = NULL;
281 size_t ret_ident_num = 0;
283 char *hostname = NULL;
285 int status;
286 size_t i;
288 status = lcc_listval (connection, &ret_ident, &ret_ident_num);
289 if (status != 0) {
290 printf ("UNKNOWN: %s\n", lcc_strerror (connection));
291 if (ret_ident != NULL)
292 free (ret_ident);
293 return (RET_UNKNOWN);
294 }
296 status = lcc_sort_identifiers (connection, ret_ident, ret_ident_num);
297 if (status != 0) {
298 printf ("UNKNOWN: %s\n", lcc_strerror (connection));
299 if (ret_ident != NULL)
300 free (ret_ident);
301 return (RET_UNKNOWN);
302 }
304 for (i = 0; i < ret_ident_num; ++i) {
305 char id[1024];
307 if ((hostname_g != NULL) && (strcasecmp (hostname_g, ret_ident[i].host)))
308 continue;
310 if ((hostname == NULL) || strcasecmp (hostname, ret_ident[i].host))
311 {
312 free (hostname);
313 hostname = strdup (ret_ident[i].host);
314 printf ("Host: %s\n", hostname);
315 }
317 /* empty hostname; not to be printed again */
318 ret_ident[i].host[0] = '\0';
320 status = lcc_identifier_to_string (connection,
321 id, sizeof (id), ret_ident + i);
322 if (status != 0) {
323 printf ("ERROR: listval: Failed to convert returned "
324 "identifier to a string: %s\n",
325 lcc_strerror (connection));
326 free (hostname);
327 hostname = NULL;
328 continue;
329 }
331 /* skip over the (empty) hostname and following '/' */
332 printf ("\t%s\n", id + 1);
333 }
335 free (ret_ident);
336 free (hostname);
337 return (RET_OKAY);
338 } /* int do_listval */
340 static int do_check_con_none (size_t values_num,
341 double *values, char **values_names)
342 {
343 int num_critical = 0;
344 int num_warning = 0;
345 int num_okay = 0;
346 const char *status_str = "UNKNOWN";
347 int status_code = RET_UNKNOWN;
348 size_t i;
350 for (i = 0; i < values_num; i++)
351 {
352 if (isnan (values[i]))
353 {
354 if (nan_is_error_g)
355 num_critical++;
356 else
357 num_warning++;
358 }
359 else if (match_range (&range_critical_g, values[i]) != 0)
360 num_critical++;
361 else if (match_range (&range_warning_g, values[i]) != 0)
362 num_warning++;
363 else
364 num_okay++;
365 }
367 if ((num_critical == 0) && (num_warning == 0) && (num_okay == 0))
368 {
369 printf ("WARNING: No defined values found\n");
370 return (RET_WARNING);
371 }
372 else if ((num_critical == 0) && (num_warning == 0))
373 {
374 status_str = "OKAY";
375 status_code = RET_OKAY;
376 }
377 else if (num_critical == 0)
378 {
379 status_str = "WARNING";
380 status_code = RET_WARNING;
381 }
382 else
383 {
384 status_str = "CRITICAL";
385 status_code = RET_CRITICAL;
386 }
388 printf ("%s: %i critical, %i warning, %i okay", status_str,
389 num_critical, num_warning, num_okay);
390 if (values_num > 0)
391 {
392 printf (" |");
393 for (i = 0; i < values_num; i++)
394 printf (" %s=%f;;;;", values_names[i], values[i]);
395 }
396 printf ("\n");
398 return (status_code);
399 } /* int do_check_con_none */
401 static int do_check_con_average (size_t values_num,
402 double *values, char **values_names)
403 {
404 size_t i;
405 double total;
406 int total_num;
407 double average;
408 const char *status_str = "UNKNOWN";
409 int status_code = RET_UNKNOWN;
411 total = 0.0;
412 total_num = 0;
413 for (i = 0; i < values_num; i++)
414 {
415 if (isnan (values[i]))
416 {
417 if (!nan_is_error_g)
418 continue;
420 printf ("CRITICAL: Data source \"%s\" is NaN\n",
421 values_names[i]);
422 return (RET_CRITICAL);
423 }
425 total += values[i];
426 total_num++;
427 }
429 if (total_num == 0)
430 {
431 printf ("WARNING: No defined values found\n");
432 return (RET_WARNING);
433 }
435 average = total / total_num;
437 if (match_range (&range_critical_g, average) != 0)
438 {
439 status_str = "CRITICAL";
440 status_code = RET_CRITICAL;
441 }
442 else if (match_range (&range_warning_g, average) != 0)
443 {
444 status_str = "WARNING";
445 status_code = RET_WARNING;
446 }
447 else
448 {
449 status_str = "OKAY";
450 status_code = RET_OKAY;
451 }
453 printf ("%s: %g average |", status_str, average);
454 for (i = 0; i < values_num; i++)
455 printf (" %s=%f;;;;", values_names[i], values[i]);
456 printf ("\n");
458 return (status_code);
459 } /* int do_check_con_average */
461 static int do_check_con_sum (size_t values_num,
462 double *values, char **values_names)
463 {
464 size_t i;
465 double total;
466 int total_num;
467 const char *status_str = "UNKNOWN";
468 int status_code = RET_UNKNOWN;
470 total = 0.0;
471 total_num = 0;
472 for (i = 0; i < values_num; i++)
473 {
474 if (isnan (values[i]))
475 {
476 if (!nan_is_error_g)
477 continue;
479 printf ("CRITICAL: Data source \"%s\" is NaN\n",
480 values_names[i]);
481 return (RET_CRITICAL);
482 }
484 total += values[i];
485 total_num++;
486 }
488 if (total_num == 0)
489 {
490 printf ("WARNING: No defined values found\n");
491 return (RET_WARNING);
492 }
494 if (match_range (&range_critical_g, total) != 0)
495 {
496 status_str = "CRITICAL";
497 status_code = RET_CRITICAL;
498 }
499 else if (match_range (&range_warning_g, total) != 0)
500 {
501 status_str = "WARNING";
502 status_code = RET_WARNING;
503 }
504 else
505 {
506 status_str = "OKAY";
507 status_code = RET_OKAY;
508 }
510 printf ("%s: %g sum |", status_str, total);
511 for (i = 0; i < values_num; i++)
512 printf (" %s=%f;;;;", values_names[i], values[i]);
513 printf ("\n");
515 return (status_code);
516 } /* int do_check_con_sum */
518 static int do_check_con_percentage (size_t values_num,
519 double *values, char **values_names)
520 {
521 size_t i;
522 double sum = 0.0;
523 double percentage;
525 const char *status_str = "UNKNOWN";
526 int status_code = RET_UNKNOWN;
528 if ((values_num < 1) || (isnan (values[0])))
529 {
530 printf ("WARNING: The first value is not defined\n");
531 return (RET_WARNING);
532 }
534 for (i = 0; i < values_num; i++)
535 {
536 if (isnan (values[i]))
537 {
538 if (!nan_is_error_g)
539 continue;
541 printf ("CRITICAL: Data source \"%s\" is NaN\n",
542 values_names[i]);
543 return (RET_CRITICAL);
544 }
546 sum += values[i];
547 }
549 if (sum == 0.0)
550 {
551 printf ("WARNING: Values sum up to zero\n");
552 return (RET_WARNING);
553 }
555 percentage = 100.0 * values[0] / sum;
557 if (match_range (&range_critical_g, percentage) != 0)
558 {
559 status_str = "CRITICAL";
560 status_code = RET_CRITICAL;
561 }
562 else if (match_range (&range_warning_g, percentage) != 0)
563 {
564 status_str = "WARNING";
565 status_code = RET_WARNING;
566 }
567 else
568 {
569 status_str = "OKAY";
570 status_code = RET_OKAY;
571 }
573 printf ("%s: %lf percent |", status_str, percentage);
574 for (i = 0; i < values_num; i++)
575 printf (" %s=%lf;;;;", values_names[i], values[i]);
576 return (status_code);
577 } /* int do_check_con_percentage */
579 static int do_check (lcc_connection_t *connection)
580 {
581 gauge_t *values;
582 char **values_names;
583 size_t values_num;
584 char ident_str[1024];
585 lcc_identifier_t ident;
586 size_t i;
587 int status;
589 snprintf (ident_str, sizeof (ident_str), "%s/%s",
590 hostname_g, value_string_g);
591 ident_str[sizeof (ident_str) - 1] = 0;
593 memset (&ident, 0, sizeof (ident));
594 status = lcc_string_to_identifier (connection, &ident, ident_str);
595 if (status != 0)
596 {
597 printf ("ERROR: Creating an identifier failed: %s.\n",
598 lcc_strerror (connection));
599 LCC_DESTROY (connection);
600 return (RET_CRITICAL);
601 }
603 status = lcc_getval (connection, &ident,
604 &values_num, &values, &values_names);
605 if (status != 0)
606 {
607 printf ("ERROR: Retrieving values from the daemon failed: %s.\n",
608 lcc_strerror (connection));
609 LCC_DESTROY (connection);
610 return (RET_CRITICAL);
611 }
613 LCC_DESTROY (connection);
615 status = filter_ds (&values_num, &values, &values_names);
616 if (status != RET_OKAY)
617 return (status);
619 status = RET_UNKNOWN;
620 if (consolitation_g == CON_NONE)
621 status = do_check_con_none (values_num, values, values_names);
622 else if (consolitation_g == CON_AVERAGE)
623 status = do_check_con_average (values_num, values, values_names);
624 else if (consolitation_g == CON_SUM)
625 status = do_check_con_sum (values_num, values, values_names);
626 else if (consolitation_g == CON_PERCENTAGE)
627 status = do_check_con_percentage (values_num, values, values_names);
629 free (values);
630 if (values_names != NULL)
631 for (i = 0; i < values_num; i++)
632 free (values_names[i]);
633 free (values_names);
635 return (status);
636 } /* int do_check */
638 int main (int argc, char **argv)
639 {
640 char address[1024];
641 lcc_connection_t *connection;
643 int status;
645 range_critical_g.min = NAN;
646 range_critical_g.max = NAN;
647 range_critical_g.invert = 0;
649 range_warning_g.min = NAN;
650 range_warning_g.max = NAN;
651 range_warning_g.invert = 0;
653 while (42)
654 {
655 int c;
657 c = getopt (argc, argv, "w:c:s:n:H:g:d:hm");
658 if (c < 0)
659 break;
661 switch (c)
662 {
663 case 'c':
664 parse_range (optarg, &range_critical_g);
665 break;
666 case 'w':
667 parse_range (optarg, &range_warning_g);
668 break;
669 case 's':
670 socket_file_g = optarg;
671 break;
672 case 'n':
673 value_string_g = optarg;
674 break;
675 case 'H':
676 hostname_g = optarg;
677 break;
678 case 'g':
679 if (strcasecmp (optarg, "none") == 0)
680 consolitation_g = CON_NONE;
681 else if (strcasecmp (optarg, "average") == 0)
682 consolitation_g = CON_AVERAGE;
683 else if (strcasecmp (optarg, "sum") == 0)
684 consolitation_g = CON_SUM;
685 else if (strcasecmp (optarg, "percentage") == 0)
686 consolitation_g = CON_PERCENTAGE;
687 else
688 {
689 fprintf (stderr, "Unknown consolidation function `%s'.\n",
690 optarg);
691 usage (argv[0]);
692 }
693 break;
694 case 'd':
695 {
696 char **tmp;
697 tmp = (char **) realloc (match_ds_g,
698 (match_ds_num_g + 1)
699 * sizeof (char *));
700 if (tmp == NULL)
701 {
702 fprintf (stderr, "realloc failed: %s\n",
703 strerror (errno));
704 return (RET_UNKNOWN);
705 }
706 match_ds_g = tmp;
707 match_ds_g[match_ds_num_g] = cn_strdup (optarg);
708 if (match_ds_g[match_ds_num_g] == NULL)
709 {
710 fprintf (stderr, "cn_strdup failed: %s\n",
711 strerror (errno));
712 return (RET_UNKNOWN);
713 }
714 match_ds_num_g++;
715 break;
716 }
717 case 'm':
718 nan_is_error_g = 1;
719 break;
720 default:
721 usage (argv[0]);
722 } /* switch (c) */
723 }
725 if ((socket_file_g == NULL) || (value_string_g == NULL)
726 || ((hostname_g == NULL) && (strcasecmp (value_string_g, "LIST"))))
727 {
728 fprintf (stderr, "Missing required arguments.\n");
729 usage (argv[0]);
730 }
732 snprintf (address, sizeof (address), "unix:%s", socket_file_g);
733 address[sizeof (address) - 1] = 0;
735 connection = NULL;
736 status = lcc_connect (address, &connection);
737 if (status != 0)
738 {
739 printf ("ERROR: Connecting to daemon at %s failed.\n",
740 socket_file_g);
741 return (RET_CRITICAL);
742 }
744 if (0 == strcasecmp (value_string_g, "LIST"))
745 return (do_listval (connection));
747 return (do_check (connection));
748 } /* int main */