1 /**
2 * collectd - src/ceph.c
3 * Copyright (C) 2011 New Dream Network
4 * Copyright (C) 2015 Florian octo Forster
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; only version 2 of the License is applicable.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors:
20 * Colin McCabe <cmccabe at alumni.cmu.edu>
21 * Dennis Zou <yunzou at cisco.com>
22 * Dan Ryder <daryder at cisco.com>
23 * Florian octo Forster <octo at collectd.org>
24 **/
26 #define _DEFAULT_SOURCE
27 #define _BSD_SOURCE
29 #include "collectd.h"
31 #include "common.h"
32 #include "plugin.h"
34 #include <arpa/inet.h>
35 #include <errno.h>
36 #include <fcntl.h>
37 #include <yajl/yajl_parse.h>
38 #if HAVE_YAJL_YAJL_VERSION_H
39 #include <yajl/yajl_version.h>
40 #endif
41 #ifdef HAVE_SYS_CAPABILITY_H
42 # include <sys/capability.h>
43 #endif
45 #include <limits.h>
46 #include <poll.h>
47 #include <stdint.h>
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <strings.h>
52 #include <sys/time.h>
53 #include <sys/types.h>
54 #include <sys/un.h>
55 #include <unistd.h>
56 #include <math.h>
57 #include <inttypes.h>
59 #define RETRY_AVGCOUNT -1
61 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
62 # define HAVE_YAJL_V2 1
63 #endif
65 #define RETRY_ON_EINTR(ret, expr) \
66 while(1) { \
67 ret = expr; \
68 if(ret >= 0) \
69 break; \
70 ret = -errno; \
71 if(ret != -EINTR) \
72 break; \
73 }
75 /** Timeout interval in seconds */
76 #define CEPH_TIMEOUT_INTERVAL 1
78 /** Maximum path length for a UNIX domain socket on this system */
79 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
81 /** Yajl callback returns */
82 #define CEPH_CB_CONTINUE 1
83 #define CEPH_CB_ABORT 0
85 #if HAVE_YAJL_V2
86 typedef size_t yajl_len_t;
87 #else
88 typedef unsigned int yajl_len_t;
89 #endif
91 /** Number of types for ceph defined in types.db */
92 #define CEPH_DSET_TYPES_NUM 3
93 /** ceph types enum */
94 enum ceph_dset_type_d
95 {
96 DSET_LATENCY = 0,
97 DSET_BYTES = 1,
98 DSET_RATE = 2,
99 DSET_TYPE_UNFOUND = 1000
100 };
102 /** Valid types for ceph defined in types.db */
103 static const char * const ceph_dset_types [CEPH_DSET_TYPES_NUM] =
104 {"ceph_latency", "ceph_bytes", "ceph_rate"};
106 /******* ceph_daemon *******/
107 struct ceph_daemon
108 {
109 /** Version of the admin_socket interface */
110 uint32_t version;
111 /** daemon name **/
112 char name[DATA_MAX_NAME_LEN];
114 /** Path to the socket that we use to talk to the ceph daemon */
115 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
117 /** Number of counters */
118 int ds_num;
119 /** Track ds types */
120 uint32_t *ds_types;
121 /** Track ds names to match with types */
122 char **ds_names;
124 /**
125 * Keep track of last data for latency values so we can calculate rate
126 * since last poll.
127 */
128 struct last_data **last_poll_data;
129 /** index of last poll data */
130 int last_idx;
131 };
133 /******* JSON parsing *******/
134 typedef int (*node_handler_t)(void *, const char*, const char*);
136 /** Track state and handler while parsing JSON */
137 struct yajl_struct
138 {
139 node_handler_t handler;
140 void * handler_arg;
142 char *key;
143 char *stack[YAJL_MAX_DEPTH];
144 size_t depth;
145 };
146 typedef struct yajl_struct yajl_struct;
148 enum perfcounter_type_d
149 {
150 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
151 };
153 /** Give user option to use default (long run = since daemon started) avg */
154 static int long_run_latency_avg = 0;
156 /**
157 * Give user option to use default type for special cases -
158 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
159 * type as a sum/count pair and will calculate it the same as a latency value.
160 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
161 * use the DERIVE type. Unless user specifies to use given type, convert this
162 * metric to use DERIVE.
163 */
164 static int convert_special_metrics = 1;
166 /** Array of daemons to monitor */
167 static struct ceph_daemon **g_daemons = NULL;
169 /** Number of elements in g_daemons */
170 static size_t g_num_daemons = 0;
172 /**
173 * A set of data that we build up in memory while parsing the JSON.
174 */
175 struct values_tmp
176 {
177 /** ceph daemon we are processing data for*/
178 struct ceph_daemon *d;
179 /** track avgcount across counters for avgcount/sum latency pairs */
180 uint64_t avgcount;
181 /** current index of counters - used to get type of counter */
182 int index;
183 /** do we already have an avgcount for latency pair */
184 int avgcount_exists;
185 /**
186 * similar to index, but current index of latency type counters -
187 * used to get last poll data of counter
188 */
189 int latency_index;
190 /**
191 * values list - maintain across counters since
192 * host/plugin/plugin instance are always the same
193 */
194 value_list_t vlist;
195 };
197 /**
198 * A set of count/sum pairs to keep track of latency types and get difference
199 * between this poll data and last poll data.
200 */
201 struct last_data
202 {
203 char ds_name[DATA_MAX_NAME_LEN];
204 double last_sum;
205 uint64_t last_count;
206 };
208 /******* network I/O *******/
209 enum cstate_t
210 {
211 CSTATE_UNCONNECTED = 0,
212 CSTATE_WRITE_REQUEST,
213 CSTATE_READ_VERSION,
214 CSTATE_READ_AMT,
215 CSTATE_READ_JSON,
216 };
218 enum request_type_t
219 {
220 ASOK_REQ_VERSION = 0,
221 ASOK_REQ_DATA = 1,
222 ASOK_REQ_SCHEMA = 2,
223 ASOK_REQ_NONE = 1000,
224 };
226 struct cconn
227 {
228 /** The Ceph daemon that we're talking to */
229 struct ceph_daemon *d;
231 /** Request type */
232 uint32_t request_type;
234 /** The connection state */
235 enum cstate_t state;
237 /** The socket we use to talk to this daemon */
238 int asok;
240 /** The amount of data remaining to read / write. */
241 uint32_t amt;
243 /** Length of the JSON to read */
244 uint32_t json_len;
246 /** Buffer containing JSON data */
247 unsigned char *json;
249 /** Keep data important to yajl processing */
250 struct yajl_struct yajl;
251 };
253 static int ceph_cb_null(void *ctx)
254 {
255 return CEPH_CB_CONTINUE;
256 }
258 static int ceph_cb_boolean(void *ctx, int bool_val)
259 {
260 return CEPH_CB_CONTINUE;
261 }
263 #define BUFFER_ADD(dest, src) do { \
264 size_t dest_size = sizeof (dest); \
265 strncat ((dest), (src), dest_size - strlen (dest)); \
266 (dest)[dest_size - 1] = '\0'; \
267 } while (0)
269 static int
270 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
271 {
272 yajl_struct *state = (yajl_struct*) ctx;
273 char buffer[number_len+1];
274 char key[2 * DATA_MAX_NAME_LEN];
275 _Bool latency_type = 0;
276 int status;
278 key[0] = '\0';
279 memcpy(buffer, number_val, number_len);
280 buffer[sizeof(buffer) - 1] = '\0';
282 for (size_t i = 0; i < state->depth; i++)
283 {
284 if (state->stack[i] == NULL)
285 continue;
287 if (strlen (key) != 0)
288 BUFFER_ADD (key, ".");
289 BUFFER_ADD (key, state->stack[i]);
290 }
292 /* Special case for latency metrics. */
293 if ((strcmp ("avgcount", state->key) == 0)
294 || (strcmp ("sum", state->key) == 0))
295 {
296 latency_type = 1;
298 /* Super-special case for filestore.journal_wr_bytes.avgcount: For
299 * some reason, Ceph schema encodes this as a count/sum pair while all
300 * other "Bytes" data (excluding used/capacity bytes for OSD space) uses
301 * a single "Derive" type. To spare further confusion, keep this KPI as
302 * the same type of other "Bytes". Instead of keeping an "average" or
303 * "rate", use the "sum" in the pair and assign that to the derive
304 * value. */
305 if (convert_special_metrics && (state->depth >= 2)
306 && (strcmp("filestore", state->stack[state->depth - 2]) == 0)
307 && (strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0)
308 && (strcmp("avgcount", state->key) == 0))
309 {
310 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
311 return CEPH_CB_CONTINUE;
312 }
313 }
314 else /* not a latency type */
315 {
316 BUFFER_ADD (key, ".");
317 BUFFER_ADD (key, state->key);
318 }
320 status = state->handler(state->handler_arg, buffer, key);
321 if((status == RETRY_AVGCOUNT) && latency_type)
322 {
323 /* Add previously skipped part of the key, either "avgcount" or "sum",
324 * and try again. */
325 BUFFER_ADD (key, ".");
326 BUFFER_ADD (key, state->key);
328 status = state->handler(state->handler_arg, buffer, key);
329 }
331 if (status != 0)
332 {
333 ERROR("ceph plugin: JSON handler failed with status %d.", status);
334 return CEPH_CB_ABORT;
335 }
337 return CEPH_CB_CONTINUE;
338 }
340 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
341 yajl_len_t string_len)
342 {
343 return CEPH_CB_CONTINUE;
344 }
346 static int ceph_cb_start_map(void *ctx)
347 {
348 yajl_struct *state = (yajl_struct*) ctx;
350 /* Push key to the stack */
351 if (state->depth == YAJL_MAX_DEPTH)
352 return CEPH_CB_ABORT;
354 state->stack[state->depth] = state->key;
355 state->depth++;
356 state->key = NULL;
358 return CEPH_CB_CONTINUE;
359 }
361 static int ceph_cb_end_map(void *ctx)
362 {
363 yajl_struct *state = (yajl_struct*) ctx;
365 /* Pop key from the stack */
366 if (state->depth == 0)
367 return CEPH_CB_ABORT;
369 sfree (state->key);
370 state->depth--;
371 state->key = state->stack[state->depth];
372 state->stack[state->depth] = NULL;
374 return CEPH_CB_CONTINUE;
375 }
377 static int
378 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
379 {
380 yajl_struct *state = (yajl_struct*) ctx;
381 size_t sz = ((size_t) string_len) + 1;
383 sfree (state->key);
384 state->key = malloc (sz);
385 if (state->key == NULL)
386 {
387 ERROR ("ceph plugin: malloc failed.");
388 return CEPH_CB_ABORT;
389 }
391 memmove (state->key, key, sz - 1);
392 state->key[sz - 1] = 0;
394 return CEPH_CB_CONTINUE;
395 }
397 static int ceph_cb_start_array(void *ctx)
398 {
399 return CEPH_CB_CONTINUE;
400 }
402 static int ceph_cb_end_array(void *ctx)
403 {
404 return CEPH_CB_CONTINUE;
405 }
407 static yajl_callbacks callbacks = {
408 ceph_cb_null,
409 ceph_cb_boolean,
410 NULL,
411 NULL,
412 ceph_cb_number,
413 ceph_cb_string,
414 ceph_cb_start_map,
415 ceph_cb_map_key,
416 ceph_cb_end_map,
417 ceph_cb_start_array,
418 ceph_cb_end_array
419 };
421 static void ceph_daemon_print(const struct ceph_daemon *d)
422 {
423 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
424 }
426 static void ceph_daemons_print(void)
427 {
428 for(size_t i = 0; i < g_num_daemons; ++i)
429 {
430 ceph_daemon_print(g_daemons[i]);
431 }
432 }
434 static void ceph_daemon_free(struct ceph_daemon *d)
435 {
436 for(int i = 0; i < d->last_idx; i++)
437 {
438 sfree(d->last_poll_data[i]);
439 }
440 sfree(d->last_poll_data);
441 d->last_poll_data = NULL;
442 d->last_idx = 0;
444 for(int i = 0; i < d->ds_num; i++)
445 {
446 sfree(d->ds_names[i]);
447 }
448 sfree(d->ds_types);
449 sfree(d->ds_names);
450 sfree(d);
451 }
453 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
454 * intput string. Characters following these special characters are capitalized.
455 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
456 * "Minus". */
457 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
458 {
459 char *src_copy;
460 size_t src_len;
461 char *ptr = buffer;
462 size_t ptr_size = buffer_size;
463 _Bool append_plus = 0;
464 _Bool append_minus = 0;
466 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
467 return EINVAL;
469 src_copy = strdup (src);
470 src_len = strlen(src);
472 /* Remove trailing "+" and "-". */
473 if (src_copy[src_len - 1] == '+')
474 {
475 append_plus = 1;
476 src_len--;
477 src_copy[src_len] = 0;
478 }
479 else if (src_copy[src_len - 1] == '-')
480 {
481 append_minus = 1;
482 src_len--;
483 src_copy[src_len] = 0;
484 }
486 /* Split at special chars, capitalize first character, append to buffer. */
487 char *dummy = src_copy;
488 char *token;
489 char *save_ptr = NULL;
490 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
491 {
492 size_t len;
494 dummy = NULL;
496 token[0] = toupper ((int) token[0]);
498 assert (ptr_size > 1);
500 len = strlen (token);
501 if (len >= ptr_size)
502 len = ptr_size - 1;
504 assert (len > 0);
505 assert (len < ptr_size);
507 sstrncpy (ptr, token, len + 1);
508 ptr += len;
509 ptr_size -= len;
511 assert (*ptr == 0);
512 if (ptr_size <= 1)
513 break;
514 }
516 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
517 if (append_plus || append_minus)
518 {
519 char const *append = "Plus";
520 if (append_minus)
521 append = "Minus";
523 size_t offset = buffer_size - (strlen (append) + 1);
524 if (offset > strlen (buffer))
525 offset = strlen (buffer);
527 sstrncpy (buffer + offset, append, buffer_size - offset);
528 }
530 sfree (src_copy);
531 return 0;
532 }
534 static _Bool has_suffix (char const *str, char const *suffix)
535 {
536 size_t str_len = strlen (str);
537 size_t suffix_len = strlen (suffix);
538 size_t offset;
540 if (suffix_len > str_len)
541 return 0;
542 offset = str_len - suffix_len;
544 if (strcmp (str + offset, suffix) == 0)
545 return 1;
547 return 0;
548 }
550 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
551 static size_t count_parts (char const *key)
552 {
553 size_t parts_num = 0;
555 for (const char *ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
556 parts_num++;
558 return parts_num;
559 }
561 /**
562 * Parse key to remove "type" if this is for schema and initiate compaction
563 */
564 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
565 {
566 char tmp[2 * buffer_size];
568 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
569 return EINVAL;
571 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
572 {
573 /* strip ".type" suffix iff the key has more than two parts. */
574 size_t sz = strlen (key_str) - strlen (".type") + 1;
576 if (sz > sizeof (tmp))
577 sz = sizeof (tmp);
578 sstrncpy (tmp, key_str, sz);
579 }
580 else
581 {
582 sstrncpy (tmp, key_str, sizeof (tmp));
583 }
585 return compact_ds_name (buffer, buffer_size, tmp);
586 }
588 /**
589 * while parsing ceph admin socket schema, save counter name and type for later
590 * data processing
591 */
592 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
593 int pc_type)
594 {
595 uint32_t type;
596 char ds_name[DATA_MAX_NAME_LEN];
598 if(convert_special_metrics)
599 {
600 /**
601 * Special case for filestore:JournalWrBytes. For some reason, Ceph
602 * schema encodes this as a count/sum pair while all other "Bytes" data
603 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
604 * type. To spare further confusion, keep this KPI as the same type of
605 * other "Bytes". Instead of keeping an "average" or "rate", use the
606 * "sum" in the pair and assign that to the derive value.
607 */
608 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
609 {
610 pc_type = 10;
611 }
612 }
614 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
615 if(!d->ds_names)
616 {
617 return -ENOMEM;
618 }
620 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
621 if(!d->ds_types)
622 {
623 return -ENOMEM;
624 }
626 d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN);
627 if(!d->ds_names[d->ds_num])
628 {
629 return -ENOMEM;
630 }
632 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
633 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
634 d->ds_types[d->ds_num] = type;
636 if (parse_keys(ds_name, sizeof (ds_name), name))
637 {
638 return 1;
639 }
641 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
642 d->ds_num = (d->ds_num + 1);
644 return 0;
645 }
647 /******* ceph_config *******/
648 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
649 {
650 const char *val;
651 if(item->values_num != 1)
652 {
653 return -ENOTSUP;
654 }
655 if(item->values[0].type != OCONFIG_TYPE_STRING)
656 {
657 return -ENOTSUP;
658 }
659 val = item->values[0].value.string;
660 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
661 {
662 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
663 item->key);
664 return -ENAMETOOLONG;
665 }
666 return 0;
667 }
669 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
670 {
671 if(item->values_num != 1)
672 {
673 return -ENOTSUP;
674 }
676 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
677 {
678 return -ENOTSUP;
679 }
681 *dest = (item->values[0].value.boolean) ? 1 : 0;
682 return 0;
683 }
685 static int cc_add_daemon_config(oconfig_item_t *ci)
686 {
687 int ret;
688 struct ceph_daemon *nd, cd = { 0 };
689 struct ceph_daemon **tmp;
691 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
692 {
693 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
694 "argument.");
695 return (-1);
696 }
698 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
699 if(ret)
700 {
701 return ret;
702 }
704 for(int i=0; i < ci->children_num; i++)
705 {
706 oconfig_item_t *child = ci->children + i;
708 if(strcasecmp("SocketPath", child->key) == 0)
709 {
710 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
711 if(ret)
712 {
713 return ret;
714 }
715 }
716 else
717 {
718 WARNING("ceph plugin: ignoring unknown option %s", child->key);
719 }
720 }
721 if(cd.name[0] == '\0')
722 {
723 ERROR("ceph plugin: you must configure a daemon name.\n");
724 return -EINVAL;
725 }
726 else if(cd.asok_path[0] == '\0')
727 {
728 ERROR("ceph plugin(name=%s): you must configure an administrative "
729 "socket path.\n", cd.name);
730 return -EINVAL;
731 }
732 else if(!((cd.asok_path[0] == '/') ||
733 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
734 {
735 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
736 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
737 return -EINVAL;
738 }
740 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
741 if(tmp == NULL)
742 {
743 /* The positive return value here indicates that this is a
744 * runtime error, not a configuration error. */
745 return ENOMEM;
746 }
747 g_daemons = tmp;
749 nd = malloc(sizeof (*nd));
750 if(!nd)
751 {
752 return ENOMEM;
753 }
754 memcpy(nd, &cd, sizeof(*nd));
755 g_daemons[g_num_daemons] = nd;
756 g_num_daemons++;
757 return 0;
758 }
760 static int ceph_config(oconfig_item_t *ci)
761 {
762 int ret;
764 for(int i = 0; i < ci->children_num; ++i)
765 {
766 oconfig_item_t *child = ci->children + i;
767 if(strcasecmp("Daemon", child->key) == 0)
768 {
769 ret = cc_add_daemon_config(child);
770 if(ret == ENOMEM)
771 {
772 ERROR("ceph plugin: Couldn't allocate memory");
773 return ret;
774 }
775 else if(ret)
776 {
777 //process other daemons and ignore this one
778 continue;
779 }
780 }
781 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
782 {
783 ret = cc_handle_bool(child, &long_run_latency_avg);
784 if(ret)
785 {
786 return ret;
787 }
788 }
789 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
790 {
791 ret = cc_handle_bool(child, &convert_special_metrics);
792 if(ret)
793 {
794 return ret;
795 }
796 }
797 else
798 {
799 WARNING("ceph plugin: ignoring unknown option %s", child->key);
800 }
801 }
802 return 0;
803 }
805 /**
806 * Parse JSON and get error message if present
807 */
808 static int
809 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
810 {
811 yajl_status status = yajl_parse(hand, json, json_len);
812 unsigned char *msg;
814 switch(status)
815 {
816 case yajl_status_error:
817 msg = yajl_get_error(hand, /* verbose = */ 1,
818 /* jsonText = */ (unsigned char *) json,
819 (unsigned int) json_len);
820 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
821 yajl_free_error(hand, msg);
822 return 1;
823 case yajl_status_client_canceled:
824 return 1;
825 default:
826 return 0;
827 }
828 }
830 /**
831 * Add entry for each counter while parsing schema
832 */
833 static int
834 node_handler_define_schema(void *arg, const char *val, const char *key)
835 {
836 struct ceph_daemon *d = (struct ceph_daemon *) arg;
837 int pc_type;
838 pc_type = atoi(val);
839 return ceph_daemon_add_ds_entry(d, key, pc_type);
840 }
842 /**
843 * Latency counter does not yet have an entry in last poll data - add it.
844 */
845 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
846 uint64_t cur_count)
847 {
848 d->last_poll_data[d->last_idx] = malloc(sizeof (*d->last_poll_data[d->last_idx]));
849 if(!d->last_poll_data[d->last_idx])
850 {
851 return -ENOMEM;
852 }
853 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
854 sizeof(d->last_poll_data[d->last_idx]->ds_name));
855 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
856 d->last_poll_data[d->last_idx]->last_count = cur_count;
857 d->last_idx = (d->last_idx + 1);
858 return 0;
859 }
861 /**
862 * Update latency counter or add new entry if it doesn't exist
863 */
864 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
865 double cur_sum, uint64_t cur_count)
866 {
867 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
868 {
869 d->last_poll_data[index]->last_sum = cur_sum;
870 d->last_poll_data[index]->last_count = cur_count;
871 return 0;
872 }
874 if(!d->last_poll_data)
875 {
876 d->last_poll_data = malloc(sizeof (*d->last_poll_data));
877 if(!d->last_poll_data)
878 {
879 return -ENOMEM;
880 }
881 }
882 else
883 {
884 struct last_data **tmp_last = realloc(d->last_poll_data,
885 ((d->last_idx+1) * sizeof(struct last_data *)));
886 if(!tmp_last)
887 {
888 return -ENOMEM;
889 }
890 d->last_poll_data = tmp_last;
891 }
892 return add_last(d, ds_n, cur_sum, cur_count);
893 }
895 /**
896 * If using index guess failed (shouldn't happen, but possible if counters
897 * get rearranged), resort to searching for counter name
898 */
899 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
900 {
901 for(int i = 0; i < d->last_idx; i++)
902 {
903 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
904 {
905 return i;
906 }
907 }
908 return -1;
909 }
911 /**
912 * Calculate average b/t current data and last poll data
913 * if last poll data exists
914 */
915 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
916 double cur_sum, uint64_t cur_count)
917 {
918 double result = -1.1, sum_delt = 0.0;
919 uint64_t count_delt = 0;
920 int tmp_index = 0;
921 if(d->last_idx > index)
922 {
923 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
924 {
925 tmp_index = index;
926 }
927 //test previous index
928 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
929 {
930 tmp_index = (index - 1);
931 }
932 else
933 {
934 tmp_index = backup_search_for_last_avg(d, ds_n);
935 }
937 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
938 {
939 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
940 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
941 result = (sum_delt / count_delt);
942 }
943 }
945 if(result == -1.1)
946 {
947 result = NAN;
948 }
949 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
950 {
951 return -ENOMEM;
952 }
953 return result;
954 }
956 /**
957 * If using index guess failed, resort to searching for counter name
958 */
959 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
960 {
961 for(int i = 0; i < d->ds_num; i++)
962 {
963 if(strcmp(d->ds_names[i], ds_name) == 0)
964 {
965 return d->ds_types[i];
966 }
967 }
968 return DSET_TYPE_UNFOUND;
969 }
971 /**
972 * Process counter data and dispatch values
973 */
974 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
975 {
976 value_t uv;
977 double tmp_d;
978 uint64_t tmp_u;
979 struct values_tmp *vtmp = (struct values_tmp*) arg;
980 uint32_t type = DSET_TYPE_UNFOUND;
981 int index = vtmp->index;
983 char ds_name[DATA_MAX_NAME_LEN];
985 if (parse_keys (ds_name, sizeof (ds_name), key))
986 {
987 return 1;
988 }
990 if(index >= vtmp->d->ds_num)
991 {
992 //don't overflow bounds of array
993 index = (vtmp->d->ds_num - 1);
994 }
996 /**
997 * counters should remain in same order we parsed schema... we maintain the
998 * index variable to keep track of current point in list of counters. first
999 * use index to guess point in array for retrieving type. if that doesn't
1000 * work, use the old way to get the counter type
1001 */
1002 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
1003 {
1004 //found match
1005 type = vtmp->d->ds_types[index];
1006 }
1007 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
1008 {
1009 //try previous key
1010 type = vtmp->d->ds_types[index-1];
1011 }
1013 if(type == DSET_TYPE_UNFOUND)
1014 {
1015 //couldn't find right type by guessing, check the old way
1016 type = backup_search_for_type(vtmp->d, ds_name);
1017 }
1019 switch(type)
1020 {
1021 case DSET_LATENCY:
1022 if(vtmp->avgcount_exists == -1)
1023 {
1024 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1025 vtmp->avgcount_exists = 0;
1026 //return after saving avgcount - don't dispatch value
1027 //until latency calculation
1028 return 0;
1029 }
1030 else
1031 {
1032 double sum, result;
1033 sscanf(val, "%lf", &sum);
1035 if(vtmp->avgcount == 0)
1036 {
1037 vtmp->avgcount = 1;
1038 }
1040 /** User wants latency values as long run avg */
1041 if(long_run_latency_avg)
1042 {
1043 result = (sum / vtmp->avgcount);
1044 }
1045 else
1046 {
1047 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1048 if(result == -ENOMEM)
1049 {
1050 return -ENOMEM;
1051 }
1052 }
1054 uv.gauge = result;
1055 vtmp->avgcount_exists = -1;
1056 vtmp->latency_index = (vtmp->latency_index + 1);
1057 }
1058 break;
1059 case DSET_BYTES:
1060 sscanf(val, "%lf", &tmp_d);
1061 uv.gauge = tmp_d;
1062 break;
1063 case DSET_RATE:
1064 sscanf(val, "%" PRIu64, &tmp_u);
1065 uv.derive = tmp_u;
1066 break;
1067 case DSET_TYPE_UNFOUND:
1068 default:
1069 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1070 return -1;
1071 }
1073 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1074 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1075 vtmp->vlist.values = &uv;
1076 vtmp->vlist.values_len = 1;
1078 vtmp->index = (vtmp->index + 1);
1079 plugin_dispatch_values(&vtmp->vlist);
1081 return 0;
1082 }
1084 static int cconn_connect(struct cconn *io)
1085 {
1086 struct sockaddr_un address = { 0 };
1087 int flags, fd, err;
1088 if(io->state != CSTATE_UNCONNECTED)
1089 {
1090 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1091 return -EDOM;
1092 }
1093 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1094 if(fd < 0)
1095 {
1096 err = -errno;
1097 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1098 "failed: error %d", err);
1099 return err;
1100 }
1101 address.sun_family = AF_UNIX;
1102 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1103 io->d->asok_path);
1104 RETRY_ON_EINTR(err,
1105 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1106 if(err < 0)
1107 {
1108 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1109 fd, err);
1110 close(fd);
1111 return err;
1112 }
1114 flags = fcntl(fd, F_GETFL, 0);
1115 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1116 {
1117 err = -errno;
1118 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1119 fd, err);
1120 close(fd);
1121 return err;
1122 }
1123 io->asok = fd;
1124 io->state = CSTATE_WRITE_REQUEST;
1125 io->amt = 0;
1126 io->json_len = 0;
1127 io->json = NULL;
1128 return 0;
1129 }
1131 static void cconn_close(struct cconn *io)
1132 {
1133 io->state = CSTATE_UNCONNECTED;
1134 if(io->asok != -1)
1135 {
1136 int res;
1137 RETRY_ON_EINTR(res, close(io->asok));
1138 }
1139 io->asok = -1;
1140 io->amt = 0;
1141 io->json_len = 0;
1142 sfree(io->json);
1143 io->json = NULL;
1144 }
1146 /* Process incoming JSON counter data */
1147 static int
1148 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1149 {
1150 int ret;
1151 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1152 if(!vtmp)
1153 {
1154 return -ENOMEM;
1155 }
1157 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1158 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1159 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1160 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1162 vtmp->d = io->d;
1163 vtmp->avgcount_exists = -1;
1164 vtmp->latency_index = 0;
1165 vtmp->index = 0;
1166 yajl->handler_arg = vtmp;
1167 ret = traverse_json(io->json, io->json_len, hand);
1168 sfree(vtmp);
1169 return ret;
1170 }
1172 /**
1173 * Initiate JSON parsing and print error if one occurs
1174 */
1175 static int cconn_process_json(struct cconn *io)
1176 {
1177 if((io->request_type != ASOK_REQ_DATA) &&
1178 (io->request_type != ASOK_REQ_SCHEMA))
1179 {
1180 return -EDOM;
1181 }
1183 int result = 1;
1184 yajl_handle hand;
1185 yajl_status status;
1187 hand = yajl_alloc(&callbacks,
1188 #if HAVE_YAJL_V2
1189 /* alloc funcs = */ NULL,
1190 #else
1191 /* alloc funcs = */ NULL, NULL,
1192 #endif
1193 /* context = */ (void *)(&io->yajl));
1195 if(!hand)
1196 {
1197 ERROR ("ceph plugin: yajl_alloc failed.");
1198 return ENOMEM;
1199 }
1201 io->yajl.depth = 0;
1203 switch(io->request_type)
1204 {
1205 case ASOK_REQ_DATA:
1206 io->yajl.handler = node_handler_fetch_data;
1207 result = cconn_process_data(io, &io->yajl, hand);
1208 break;
1209 case ASOK_REQ_SCHEMA:
1210 //init daemon specific variables
1211 io->d->ds_num = 0;
1212 io->d->last_idx = 0;
1213 io->d->last_poll_data = NULL;
1214 io->yajl.handler = node_handler_define_schema;
1215 io->yajl.handler_arg = io->d;
1216 result = traverse_json(io->json, io->json_len, hand);
1217 break;
1218 }
1220 if(result)
1221 {
1222 goto done;
1223 }
1225 #if HAVE_YAJL_V2
1226 status = yajl_complete_parse(hand);
1227 #else
1228 status = yajl_parse_complete(hand);
1229 #endif
1231 if (status != yajl_status_ok)
1232 {
1233 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1234 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1235 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1236 (char *) errmsg);
1237 yajl_free_error (hand, errmsg);
1238 yajl_free (hand);
1239 return 1;
1240 }
1242 done:
1243 yajl_free (hand);
1244 return result;
1245 }
1247 static int cconn_validate_revents(struct cconn *io, int revents)
1248 {
1249 if(revents & POLLERR)
1250 {
1251 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1252 io->d->name);
1253 return -EIO;
1254 }
1255 switch (io->state)
1256 {
1257 case CSTATE_WRITE_REQUEST:
1258 return (revents & POLLOUT) ? 0 : -EINVAL;
1259 case CSTATE_READ_VERSION:
1260 case CSTATE_READ_AMT:
1261 case CSTATE_READ_JSON:
1262 return (revents & POLLIN) ? 0 : -EINVAL;
1263 default:
1264 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1265 "illegal state on line %d", io->d->name, __LINE__);
1266 return -EDOM;
1267 }
1268 }
1270 /** Handle a network event for a connection */
1271 static int cconn_handle_event(struct cconn *io)
1272 {
1273 int ret;
1274 switch (io->state)
1275 {
1276 case CSTATE_UNCONNECTED:
1277 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1278 "state on line %d", io->d->name, __LINE__);
1280 return -EDOM;
1281 case CSTATE_WRITE_REQUEST:
1282 {
1283 char cmd[32];
1284 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1285 io->request_type, "\" }\n");
1286 size_t cmd_len = strlen(cmd);
1287 RETRY_ON_EINTR(ret,
1288 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1289 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1290 io->d->name, io->state, io->amt, ret);
1291 if(ret < 0)
1292 {
1293 return ret;
1294 }
1295 io->amt += ret;
1296 if(io->amt >= cmd_len)
1297 {
1298 io->amt = 0;
1299 switch (io->request_type)
1300 {
1301 case ASOK_REQ_VERSION:
1302 io->state = CSTATE_READ_VERSION;
1303 break;
1304 default:
1305 io->state = CSTATE_READ_AMT;
1306 break;
1307 }
1308 }
1309 return 0;
1310 }
1311 case CSTATE_READ_VERSION:
1312 {
1313 RETRY_ON_EINTR(ret,
1314 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1315 sizeof(io->d->version) - io->amt));
1316 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1317 io->d->name, io->state, ret);
1318 if(ret < 0)
1319 {
1320 return ret;
1321 }
1322 io->amt += ret;
1323 if(io->amt >= sizeof(io->d->version))
1324 {
1325 io->d->version = ntohl(io->d->version);
1326 if(io->d->version != 1)
1327 {
1328 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1329 "expecting version %d!", io->d->name, io->d->version);
1330 return -ENOTSUP;
1331 }
1332 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1333 "version %d", io->d->name, io->d->version);
1334 io->amt = 0;
1335 cconn_close(io);
1336 io->request_type = ASOK_REQ_SCHEMA;
1337 }
1338 return 0;
1339 }
1340 case CSTATE_READ_AMT:
1341 {
1342 RETRY_ON_EINTR(ret,
1343 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1344 sizeof(io->json_len) - io->amt));
1345 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1346 io->d->name, io->state, ret);
1347 if(ret < 0)
1348 {
1349 return ret;
1350 }
1351 io->amt += ret;
1352 if(io->amt >= sizeof(io->json_len))
1353 {
1354 io->json_len = ntohl(io->json_len);
1355 io->amt = 0;
1356 io->state = CSTATE_READ_JSON;
1357 io->json = calloc(1, io->json_len + 1);
1358 if(!io->json)
1359 {
1360 ERROR("ceph plugin: error callocing io->json");
1361 return -ENOMEM;
1362 }
1363 }
1364 return 0;
1365 }
1366 case CSTATE_READ_JSON:
1367 {
1368 RETRY_ON_EINTR(ret,
1369 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1370 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1371 io->d->name, io->state, ret);
1372 if(ret < 0)
1373 {
1374 return ret;
1375 }
1376 io->amt += ret;
1377 if(io->amt >= io->json_len)
1378 {
1379 ret = cconn_process_json(io);
1380 if(ret)
1381 {
1382 return ret;
1383 }
1384 cconn_close(io);
1385 io->request_type = ASOK_REQ_NONE;
1386 }
1387 return 0;
1388 }
1389 default:
1390 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1391 "state on line %d", io->d->name, __LINE__);
1392 return -EDOM;
1393 }
1394 }
1396 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1397 {
1398 int ret;
1399 if(io->request_type == ASOK_REQ_NONE)
1400 {
1401 /* The request has already been serviced. */
1402 return 0;
1403 }
1404 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1405 {
1406 /* If there are no counters to report on, don't bother
1407 * connecting */
1408 return 0;
1409 }
1411 switch (io->state)
1412 {
1413 case CSTATE_UNCONNECTED:
1414 ret = cconn_connect(io);
1415 if(ret > 0)
1416 {
1417 return -ret;
1418 }
1419 else if(ret < 0)
1420 {
1421 return ret;
1422 }
1423 fds->fd = io->asok;
1424 fds->events = POLLOUT;
1425 return 1;
1426 case CSTATE_WRITE_REQUEST:
1427 fds->fd = io->asok;
1428 fds->events = POLLOUT;
1429 return 1;
1430 case CSTATE_READ_VERSION:
1431 case CSTATE_READ_AMT:
1432 case CSTATE_READ_JSON:
1433 fds->fd = io->asok;
1434 fds->events = POLLIN;
1435 return 1;
1436 default:
1437 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1438 "on line %d", io->d->name, __LINE__);
1439 return -EDOM;
1440 }
1441 }
1443 /** Returns the difference between two struct timevals in milliseconds.
1444 * On overflow, we return max/min int.
1445 */
1446 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1447 {
1448 int64_t ret;
1449 int sec_diff = t1->tv_sec - t2->tv_sec;
1450 int usec_diff = t1->tv_usec - t2->tv_usec;
1451 ret = usec_diff / 1000;
1452 ret += (sec_diff * 1000);
1453 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1454 }
1456 /** This handles the actual network I/O to talk to the Ceph daemons.
1457 */
1458 static int cconn_main_loop(uint32_t request_type)
1459 {
1460 int ret, some_unreachable = 0;
1461 struct timeval end_tv;
1462 struct cconn io_array[g_num_daemons];
1464 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1466 /* create cconn array */
1467 memset(io_array, 0, sizeof(io_array));
1468 for(size_t i = 0; i < g_num_daemons; ++i)
1469 {
1470 io_array[i].d = g_daemons[i];
1471 io_array[i].request_type = request_type;
1472 io_array[i].state = CSTATE_UNCONNECTED;
1473 }
1475 /** Calculate the time at which we should give up */
1476 gettimeofday(&end_tv, NULL);
1477 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1479 while (1)
1480 {
1481 int nfds, diff;
1482 struct timeval tv;
1483 struct cconn *polled_io_array[g_num_daemons];
1484 struct pollfd fds[g_num_daemons];
1485 memset(fds, 0, sizeof(fds));
1486 nfds = 0;
1487 for(size_t i = 0; i < g_num_daemons; ++i)
1488 {
1489 struct cconn *io = io_array + i;
1490 ret = cconn_prepare(io, fds + nfds);
1491 if(ret < 0)
1492 {
1493 WARNING("ceph plugin: cconn_prepare(name=%s,i=%zu,st=%d)=%d",
1494 io->d->name, i, io->state, ret);
1495 cconn_close(io);
1496 io->request_type = ASOK_REQ_NONE;
1497 some_unreachable = 1;
1498 }
1499 else if(ret == 1)
1500 {
1501 polled_io_array[nfds++] = io_array + i;
1502 }
1503 }
1504 if(nfds == 0)
1505 {
1506 /* finished */
1507 ret = 0;
1508 goto done;
1509 }
1510 gettimeofday(&tv, NULL);
1511 diff = milli_diff(&end_tv, &tv);
1512 if(diff <= 0)
1513 {
1514 /* Timed out */
1515 ret = -ETIMEDOUT;
1516 WARNING("ceph plugin: cconn_main_loop: timed out.");
1517 goto done;
1518 }
1519 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1520 if(ret < 0)
1521 {
1522 ERROR("ceph plugin: poll(2) error: %d", ret);
1523 goto done;
1524 }
1525 for(int i = 0; i < nfds; ++i)
1526 {
1527 struct cconn *io = polled_io_array[i];
1528 int revents = fds[i].revents;
1529 if(revents == 0)
1530 {
1531 /* do nothing */
1532 continue;
1533 }
1534 else if(cconn_validate_revents(io, revents))
1535 {
1536 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1537 "revents validation error: "
1538 "revents=0x%08x", io->d->name, i, io->state, revents);
1539 cconn_close(io);
1540 io->request_type = ASOK_REQ_NONE;
1541 some_unreachable = 1;
1542 }
1543 else
1544 {
1545 ret = cconn_handle_event(io);
1546 if(ret)
1547 {
1548 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1549 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1550 cconn_close(io);
1551 io->request_type = ASOK_REQ_NONE;
1552 some_unreachable = 1;
1553 }
1554 }
1555 }
1556 }
1557 done: for(size_t i = 0; i < g_num_daemons; ++i)
1558 {
1559 cconn_close(io_array + i);
1560 }
1561 if(some_unreachable)
1562 {
1563 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1564 }
1565 else
1566 {
1567 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1568 }
1569 return ret;
1570 }
1572 static int ceph_read(void)
1573 {
1574 return cconn_main_loop(ASOK_REQ_DATA);
1575 }
1577 /******* lifecycle *******/
1578 static int ceph_init(void)
1579 {
1580 int ret;
1582 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE)
1583 if (check_capability (CAP_DAC_OVERRIDE) != 0)
1584 {
1585 if (getuid () == 0)
1586 WARNING ("ceph plugin: Running collectd as root, but the "
1587 "CAP_DAC_OVERRIDE capability is missing. The plugin's read "
1588 "function will probably fail. Is your init system dropping "
1589 "capabilities?");
1590 else
1591 WARNING ("ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE "
1592 "capability. If you don't want to run collectd as root, try running "
1593 "\"setcap cap_dac_override=ep\" on the collectd binary.");
1594 }
1595 #endif
1597 ceph_daemons_print();
1599 ret = cconn_main_loop(ASOK_REQ_VERSION);
1601 return (ret) ? ret : 0;
1602 }
1604 static int ceph_shutdown(void)
1605 {
1606 for(size_t i = 0; i < g_num_daemons; ++i)
1607 {
1608 ceph_daemon_free(g_daemons[i]);
1609 }
1610 sfree(g_daemons);
1611 g_daemons = NULL;
1612 g_num_daemons = 0;
1613 DEBUG("ceph plugin: finished ceph_shutdown");
1614 return 0;
1615 }
1617 void module_register(void)
1618 {
1619 plugin_register_complex_config("ceph", ceph_config);
1620 plugin_register_init("ceph", ceph_init);
1621 plugin_register_read("ceph", ceph_read);
1622 plugin_register_shutdown("ceph", ceph_shutdown);
1623 }
1624 /* vim: set sw=4 sts=4 et : */