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 int 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(int 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 return 0;
757 }
759 static int ceph_config(oconfig_item_t *ci)
760 {
761 int ret;
763 for(int i = 0; i < ci->children_num; ++i)
764 {
765 oconfig_item_t *child = ci->children + i;
766 if(strcasecmp("Daemon", child->key) == 0)
767 {
768 ret = cc_add_daemon_config(child);
769 if(ret == ENOMEM)
770 {
771 ERROR("ceph plugin: Couldn't allocate memory");
772 return ret;
773 }
774 else if(ret)
775 {
776 //process other daemons and ignore this one
777 continue;
778 }
779 }
780 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
781 {
782 ret = cc_handle_bool(child, &long_run_latency_avg);
783 if(ret)
784 {
785 return ret;
786 }
787 }
788 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
789 {
790 ret = cc_handle_bool(child, &convert_special_metrics);
791 if(ret)
792 {
793 return ret;
794 }
795 }
796 else
797 {
798 WARNING("ceph plugin: ignoring unknown option %s", child->key);
799 }
800 }
801 return 0;
802 }
804 /**
805 * Parse JSON and get error message if present
806 */
807 static int
808 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
809 {
810 yajl_status status = yajl_parse(hand, json, json_len);
811 unsigned char *msg;
813 switch(status)
814 {
815 case yajl_status_error:
816 msg = yajl_get_error(hand, /* verbose = */ 1,
817 /* jsonText = */ (unsigned char *) json,
818 (unsigned int) json_len);
819 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
820 yajl_free_error(hand, msg);
821 return 1;
822 case yajl_status_client_canceled:
823 return 1;
824 default:
825 return 0;
826 }
827 }
829 /**
830 * Add entry for each counter while parsing schema
831 */
832 static int
833 node_handler_define_schema(void *arg, const char *val, const char *key)
834 {
835 struct ceph_daemon *d = (struct ceph_daemon *) arg;
836 int pc_type;
837 pc_type = atoi(val);
838 return ceph_daemon_add_ds_entry(d, key, pc_type);
839 }
841 /**
842 * Latency counter does not yet have an entry in last poll data - add it.
843 */
844 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
845 uint64_t cur_count)
846 {
847 d->last_poll_data[d->last_idx] = malloc(sizeof (*d->last_poll_data[d->last_idx]));
848 if(!d->last_poll_data[d->last_idx])
849 {
850 return -ENOMEM;
851 }
852 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
853 sizeof(d->last_poll_data[d->last_idx]->ds_name));
854 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
855 d->last_poll_data[d->last_idx]->last_count = cur_count;
856 d->last_idx = (d->last_idx + 1);
857 return 0;
858 }
860 /**
861 * Update latency counter or add new entry if it doesn't exist
862 */
863 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
864 double cur_sum, uint64_t cur_count)
865 {
866 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
867 {
868 d->last_poll_data[index]->last_sum = cur_sum;
869 d->last_poll_data[index]->last_count = cur_count;
870 return 0;
871 }
873 if(!d->last_poll_data)
874 {
875 d->last_poll_data = malloc(sizeof (*d->last_poll_data));
876 if(!d->last_poll_data)
877 {
878 return -ENOMEM;
879 }
880 }
881 else
882 {
883 struct last_data **tmp_last = realloc(d->last_poll_data,
884 ((d->last_idx+1) * sizeof(struct last_data *)));
885 if(!tmp_last)
886 {
887 return -ENOMEM;
888 }
889 d->last_poll_data = tmp_last;
890 }
891 return add_last(d, ds_n, cur_sum, cur_count);
892 }
894 /**
895 * If using index guess failed (shouldn't happen, but possible if counters
896 * get rearranged), resort to searching for counter name
897 */
898 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
899 {
900 for(int i = 0; i < d->last_idx; i++)
901 {
902 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
903 {
904 return i;
905 }
906 }
907 return -1;
908 }
910 /**
911 * Calculate average b/t current data and last poll data
912 * if last poll data exists
913 */
914 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
915 double cur_sum, uint64_t cur_count)
916 {
917 double result = -1.1, sum_delt = 0.0;
918 uint64_t count_delt = 0;
919 int tmp_index = 0;
920 if(d->last_idx > index)
921 {
922 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
923 {
924 tmp_index = index;
925 }
926 //test previous index
927 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
928 {
929 tmp_index = (index - 1);
930 }
931 else
932 {
933 tmp_index = backup_search_for_last_avg(d, ds_n);
934 }
936 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
937 {
938 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
939 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
940 result = (sum_delt / count_delt);
941 }
942 }
944 if(result == -1.1)
945 {
946 result = NAN;
947 }
948 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
949 {
950 return -ENOMEM;
951 }
952 return result;
953 }
955 /**
956 * If using index guess failed, resort to searching for counter name
957 */
958 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
959 {
960 for(int i = 0; i < d->ds_num; i++)
961 {
962 if(strcmp(d->ds_names[i], ds_name) == 0)
963 {
964 return d->ds_types[i];
965 }
966 }
967 return DSET_TYPE_UNFOUND;
968 }
970 /**
971 * Process counter data and dispatch values
972 */
973 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
974 {
975 value_t uv;
976 double tmp_d;
977 uint64_t tmp_u;
978 struct values_tmp *vtmp = (struct values_tmp*) arg;
979 uint32_t type = DSET_TYPE_UNFOUND;
980 int index = vtmp->index;
982 char ds_name[DATA_MAX_NAME_LEN];
984 if (parse_keys (ds_name, sizeof (ds_name), key))
985 {
986 return 1;
987 }
989 if(index >= vtmp->d->ds_num)
990 {
991 //don't overflow bounds of array
992 index = (vtmp->d->ds_num - 1);
993 }
995 /**
996 * counters should remain in same order we parsed schema... we maintain the
997 * index variable to keep track of current point in list of counters. first
998 * use index to guess point in array for retrieving type. if that doesn't
999 * work, use the old way to get the counter type
1000 */
1001 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
1002 {
1003 //found match
1004 type = vtmp->d->ds_types[index];
1005 }
1006 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
1007 {
1008 //try previous key
1009 type = vtmp->d->ds_types[index-1];
1010 }
1012 if(type == DSET_TYPE_UNFOUND)
1013 {
1014 //couldn't find right type by guessing, check the old way
1015 type = backup_search_for_type(vtmp->d, ds_name);
1016 }
1018 switch(type)
1019 {
1020 case DSET_LATENCY:
1021 if(vtmp->avgcount_exists == -1)
1022 {
1023 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1024 vtmp->avgcount_exists = 0;
1025 //return after saving avgcount - don't dispatch value
1026 //until latency calculation
1027 return 0;
1028 }
1029 else
1030 {
1031 double sum, result;
1032 sscanf(val, "%lf", &sum);
1034 if(vtmp->avgcount == 0)
1035 {
1036 vtmp->avgcount = 1;
1037 }
1039 /** User wants latency values as long run avg */
1040 if(long_run_latency_avg)
1041 {
1042 result = (sum / vtmp->avgcount);
1043 }
1044 else
1045 {
1046 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1047 if(result == -ENOMEM)
1048 {
1049 return -ENOMEM;
1050 }
1051 }
1053 uv.gauge = result;
1054 vtmp->avgcount_exists = -1;
1055 vtmp->latency_index = (vtmp->latency_index + 1);
1056 }
1057 break;
1058 case DSET_BYTES:
1059 sscanf(val, "%lf", &tmp_d);
1060 uv.gauge = tmp_d;
1061 break;
1062 case DSET_RATE:
1063 sscanf(val, "%" PRIu64, &tmp_u);
1064 uv.derive = tmp_u;
1065 break;
1066 case DSET_TYPE_UNFOUND:
1067 default:
1068 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1069 return -1;
1070 }
1072 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1073 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1074 vtmp->vlist.values = &uv;
1075 vtmp->vlist.values_len = 1;
1077 vtmp->index = (vtmp->index + 1);
1078 plugin_dispatch_values(&vtmp->vlist);
1080 return 0;
1081 }
1083 static int cconn_connect(struct cconn *io)
1084 {
1085 struct sockaddr_un address = { 0 };
1086 int flags, fd, err;
1087 if(io->state != CSTATE_UNCONNECTED)
1088 {
1089 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1090 return -EDOM;
1091 }
1092 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1093 if(fd < 0)
1094 {
1095 err = -errno;
1096 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1097 "failed: error %d", err);
1098 return err;
1099 }
1100 address.sun_family = AF_UNIX;
1101 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1102 io->d->asok_path);
1103 RETRY_ON_EINTR(err,
1104 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1105 if(err < 0)
1106 {
1107 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1108 fd, err);
1109 close(fd);
1110 return err;
1111 }
1113 flags = fcntl(fd, F_GETFL, 0);
1114 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1115 {
1116 err = -errno;
1117 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1118 fd, err);
1119 close(fd);
1120 return err;
1121 }
1122 io->asok = fd;
1123 io->state = CSTATE_WRITE_REQUEST;
1124 io->amt = 0;
1125 io->json_len = 0;
1126 io->json = NULL;
1127 return 0;
1128 }
1130 static void cconn_close(struct cconn *io)
1131 {
1132 io->state = CSTATE_UNCONNECTED;
1133 if(io->asok != -1)
1134 {
1135 int res;
1136 RETRY_ON_EINTR(res, close(io->asok));
1137 }
1138 io->asok = -1;
1139 io->amt = 0;
1140 io->json_len = 0;
1141 sfree(io->json);
1142 io->json = NULL;
1143 }
1145 /* Process incoming JSON counter data */
1146 static int
1147 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1148 {
1149 int ret;
1150 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1151 if(!vtmp)
1152 {
1153 return -ENOMEM;
1154 }
1156 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1157 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1158 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1159 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1161 vtmp->d = io->d;
1162 vtmp->avgcount_exists = -1;
1163 vtmp->latency_index = 0;
1164 vtmp->index = 0;
1165 yajl->handler_arg = vtmp;
1166 ret = traverse_json(io->json, io->json_len, hand);
1167 sfree(vtmp);
1168 return ret;
1169 }
1171 /**
1172 * Initiate JSON parsing and print error if one occurs
1173 */
1174 static int cconn_process_json(struct cconn *io)
1175 {
1176 if((io->request_type != ASOK_REQ_DATA) &&
1177 (io->request_type != ASOK_REQ_SCHEMA))
1178 {
1179 return -EDOM;
1180 }
1182 int result = 1;
1183 yajl_handle hand;
1184 yajl_status status;
1186 hand = yajl_alloc(&callbacks,
1187 #if HAVE_YAJL_V2
1188 /* alloc funcs = */ NULL,
1189 #else
1190 /* alloc funcs = */ NULL, NULL,
1191 #endif
1192 /* context = */ (void *)(&io->yajl));
1194 if(!hand)
1195 {
1196 ERROR ("ceph plugin: yajl_alloc failed.");
1197 return ENOMEM;
1198 }
1200 io->yajl.depth = 0;
1202 switch(io->request_type)
1203 {
1204 case ASOK_REQ_DATA:
1205 io->yajl.handler = node_handler_fetch_data;
1206 result = cconn_process_data(io, &io->yajl, hand);
1207 break;
1208 case ASOK_REQ_SCHEMA:
1209 //init daemon specific variables
1210 io->d->ds_num = 0;
1211 io->d->last_idx = 0;
1212 io->d->last_poll_data = NULL;
1213 io->yajl.handler = node_handler_define_schema;
1214 io->yajl.handler_arg = io->d;
1215 result = traverse_json(io->json, io->json_len, hand);
1216 break;
1217 }
1219 if(result)
1220 {
1221 goto done;
1222 }
1224 #if HAVE_YAJL_V2
1225 status = yajl_complete_parse(hand);
1226 #else
1227 status = yajl_parse_complete(hand);
1228 #endif
1230 if (status != yajl_status_ok)
1231 {
1232 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1233 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1234 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1235 (char *) errmsg);
1236 yajl_free_error (hand, errmsg);
1237 yajl_free (hand);
1238 return 1;
1239 }
1241 done:
1242 yajl_free (hand);
1243 return result;
1244 }
1246 static int cconn_validate_revents(struct cconn *io, int revents)
1247 {
1248 if(revents & POLLERR)
1249 {
1250 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1251 io->d->name);
1252 return -EIO;
1253 }
1254 switch (io->state)
1255 {
1256 case CSTATE_WRITE_REQUEST:
1257 return (revents & POLLOUT) ? 0 : -EINVAL;
1258 case CSTATE_READ_VERSION:
1259 case CSTATE_READ_AMT:
1260 case CSTATE_READ_JSON:
1261 return (revents & POLLIN) ? 0 : -EINVAL;
1262 default:
1263 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1264 "illegal state on line %d", io->d->name, __LINE__);
1265 return -EDOM;
1266 }
1267 }
1269 /** Handle a network event for a connection */
1270 static int cconn_handle_event(struct cconn *io)
1271 {
1272 int ret;
1273 switch (io->state)
1274 {
1275 case CSTATE_UNCONNECTED:
1276 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1277 "state on line %d", io->d->name, __LINE__);
1279 return -EDOM;
1280 case CSTATE_WRITE_REQUEST:
1281 {
1282 char cmd[32];
1283 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1284 io->request_type, "\" }\n");
1285 size_t cmd_len = strlen(cmd);
1286 RETRY_ON_EINTR(ret,
1287 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1288 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1289 io->d->name, io->state, io->amt, ret);
1290 if(ret < 0)
1291 {
1292 return ret;
1293 }
1294 io->amt += ret;
1295 if(io->amt >= cmd_len)
1296 {
1297 io->amt = 0;
1298 switch (io->request_type)
1299 {
1300 case ASOK_REQ_VERSION:
1301 io->state = CSTATE_READ_VERSION;
1302 break;
1303 default:
1304 io->state = CSTATE_READ_AMT;
1305 break;
1306 }
1307 }
1308 return 0;
1309 }
1310 case CSTATE_READ_VERSION:
1311 {
1312 RETRY_ON_EINTR(ret,
1313 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1314 sizeof(io->d->version) - io->amt));
1315 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1316 io->d->name, io->state, ret);
1317 if(ret < 0)
1318 {
1319 return ret;
1320 }
1321 io->amt += ret;
1322 if(io->amt >= sizeof(io->d->version))
1323 {
1324 io->d->version = ntohl(io->d->version);
1325 if(io->d->version != 1)
1326 {
1327 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1328 "expecting version %d!", io->d->name, io->d->version);
1329 return -ENOTSUP;
1330 }
1331 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1332 "version %d", io->d->name, io->d->version);
1333 io->amt = 0;
1334 cconn_close(io);
1335 io->request_type = ASOK_REQ_SCHEMA;
1336 }
1337 return 0;
1338 }
1339 case CSTATE_READ_AMT:
1340 {
1341 RETRY_ON_EINTR(ret,
1342 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1343 sizeof(io->json_len) - io->amt));
1344 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1345 io->d->name, io->state, ret);
1346 if(ret < 0)
1347 {
1348 return ret;
1349 }
1350 io->amt += ret;
1351 if(io->amt >= sizeof(io->json_len))
1352 {
1353 io->json_len = ntohl(io->json_len);
1354 io->amt = 0;
1355 io->state = CSTATE_READ_JSON;
1356 io->json = calloc(1, io->json_len + 1);
1357 if(!io->json)
1358 {
1359 ERROR("ceph plugin: error callocing io->json");
1360 return -ENOMEM;
1361 }
1362 }
1363 return 0;
1364 }
1365 case CSTATE_READ_JSON:
1366 {
1367 RETRY_ON_EINTR(ret,
1368 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1369 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1370 io->d->name, io->state, ret);
1371 if(ret < 0)
1372 {
1373 return ret;
1374 }
1375 io->amt += ret;
1376 if(io->amt >= io->json_len)
1377 {
1378 ret = cconn_process_json(io);
1379 if(ret)
1380 {
1381 return ret;
1382 }
1383 cconn_close(io);
1384 io->request_type = ASOK_REQ_NONE;
1385 }
1386 return 0;
1387 }
1388 default:
1389 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1390 "state on line %d", io->d->name, __LINE__);
1391 return -EDOM;
1392 }
1393 }
1395 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1396 {
1397 int ret;
1398 if(io->request_type == ASOK_REQ_NONE)
1399 {
1400 /* The request has already been serviced. */
1401 return 0;
1402 }
1403 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1404 {
1405 /* If there are no counters to report on, don't bother
1406 * connecting */
1407 return 0;
1408 }
1410 switch (io->state)
1411 {
1412 case CSTATE_UNCONNECTED:
1413 ret = cconn_connect(io);
1414 if(ret > 0)
1415 {
1416 return -ret;
1417 }
1418 else if(ret < 0)
1419 {
1420 return ret;
1421 }
1422 fds->fd = io->asok;
1423 fds->events = POLLOUT;
1424 return 1;
1425 case CSTATE_WRITE_REQUEST:
1426 fds->fd = io->asok;
1427 fds->events = POLLOUT;
1428 return 1;
1429 case CSTATE_READ_VERSION:
1430 case CSTATE_READ_AMT:
1431 case CSTATE_READ_JSON:
1432 fds->fd = io->asok;
1433 fds->events = POLLIN;
1434 return 1;
1435 default:
1436 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1437 "on line %d", io->d->name, __LINE__);
1438 return -EDOM;
1439 }
1440 }
1442 /** Returns the difference between two struct timevals in milliseconds.
1443 * On overflow, we return max/min int.
1444 */
1445 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1446 {
1447 int64_t ret;
1448 int sec_diff = t1->tv_sec - t2->tv_sec;
1449 int usec_diff = t1->tv_usec - t2->tv_usec;
1450 ret = usec_diff / 1000;
1451 ret += (sec_diff * 1000);
1452 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1453 }
1455 /** This handles the actual network I/O to talk to the Ceph daemons.
1456 */
1457 static int cconn_main_loop(uint32_t request_type)
1458 {
1459 int ret, some_unreachable = 0;
1460 struct timeval end_tv;
1461 struct cconn io_array[g_num_daemons];
1463 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1465 /* create cconn array */
1466 memset(io_array, 0, sizeof(io_array));
1467 for(int i = 0; i < g_num_daemons; ++i)
1468 {
1469 io_array[i].d = g_daemons[i];
1470 io_array[i].request_type = request_type;
1471 io_array[i].state = CSTATE_UNCONNECTED;
1472 }
1474 /** Calculate the time at which we should give up */
1475 gettimeofday(&end_tv, NULL);
1476 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1478 while (1)
1479 {
1480 int nfds, diff;
1481 struct timeval tv;
1482 struct cconn *polled_io_array[g_num_daemons];
1483 struct pollfd fds[g_num_daemons];
1484 memset(fds, 0, sizeof(fds));
1485 nfds = 0;
1486 for(int i = 0; i < g_num_daemons; ++i)
1487 {
1488 struct cconn *io = io_array + i;
1489 ret = cconn_prepare(io, fds + nfds);
1490 if(ret < 0)
1491 {
1492 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1493 io->d->name, i, io->state, ret);
1494 cconn_close(io);
1495 io->request_type = ASOK_REQ_NONE;
1496 some_unreachable = 1;
1497 }
1498 else if(ret == 1)
1499 {
1500 polled_io_array[nfds++] = io_array + i;
1501 }
1502 }
1503 if(nfds == 0)
1504 {
1505 /* finished */
1506 ret = 0;
1507 goto done;
1508 }
1509 gettimeofday(&tv, NULL);
1510 diff = milli_diff(&end_tv, &tv);
1511 if(diff <= 0)
1512 {
1513 /* Timed out */
1514 ret = -ETIMEDOUT;
1515 WARNING("ceph plugin: cconn_main_loop: timed out.");
1516 goto done;
1517 }
1518 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1519 if(ret < 0)
1520 {
1521 ERROR("ceph plugin: poll(2) error: %d", ret);
1522 goto done;
1523 }
1524 for(int i = 0; i < nfds; ++i)
1525 {
1526 struct cconn *io = polled_io_array[i];
1527 int revents = fds[i].revents;
1528 if(revents == 0)
1529 {
1530 /* do nothing */
1531 }
1532 else if(cconn_validate_revents(io, revents))
1533 {
1534 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1535 "revents validation error: "
1536 "revents=0x%08x", io->d->name, i, io->state, revents);
1537 cconn_close(io);
1538 io->request_type = ASOK_REQ_NONE;
1539 some_unreachable = 1;
1540 }
1541 else
1542 {
1543 ret = cconn_handle_event(io);
1544 if(ret)
1545 {
1546 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1547 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1548 cconn_close(io);
1549 io->request_type = ASOK_REQ_NONE;
1550 some_unreachable = 1;
1551 }
1552 }
1553 }
1554 }
1555 done: for(int i = 0; i < g_num_daemons; ++i)
1556 {
1557 cconn_close(io_array + i);
1558 }
1559 if(some_unreachable)
1560 {
1561 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1562 }
1563 else
1564 {
1565 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1566 }
1567 return ret;
1568 }
1570 static int ceph_read(void)
1571 {
1572 return cconn_main_loop(ASOK_REQ_DATA);
1573 }
1575 /******* lifecycle *******/
1576 static int ceph_init(void)
1577 {
1578 int ret;
1580 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE)
1581 if (check_capability (CAP_DAC_OVERRIDE) != 0)
1582 {
1583 if (getuid () == 0)
1584 WARNING ("ceph plugin: Running collectd as root, but the "
1585 "CAP_DAC_OVERRIDE capability is missing. The plugin's read "
1586 "function will probably fail. Is your init system dropping "
1587 "capabilities?");
1588 else
1589 WARNING ("ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE "
1590 "capability. If you don't want to run collectd as root, try running "
1591 "\"setcap cap_dac_override=ep\" on the collectd binary.");
1592 }
1593 #endif
1595 ceph_daemons_print();
1597 ret = cconn_main_loop(ASOK_REQ_VERSION);
1599 return (ret) ? ret : 0;
1600 }
1602 static int ceph_shutdown(void)
1603 {
1604 for(int i = 0; i < g_num_daemons; ++i)
1605 {
1606 ceph_daemon_free(g_daemons[i]);
1607 }
1608 sfree(g_daemons);
1609 g_daemons = NULL;
1610 g_num_daemons = 0;
1611 DEBUG("ceph plugin: finished ceph_shutdown");
1612 return 0;
1613 }
1615 void module_register(void)
1616 {
1617 plugin_register_complex_config("ceph", ceph_config);
1618 plugin_register_init("ceph", ceph_init);
1619 plugin_register_read("ceph", ceph_read);
1620 plugin_register_shutdown("ceph", ceph_shutdown);
1621 }
1622 /* vim: set sw=4 sts=4 et : */