64d87855fbb0f820ea1a8e8f6d09eeb61894154c
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 size_t dest_len = strlen (dest); \
266 if (dest_size > dest_len) { \
267 sstrncpy ((dest) + dest_len, (src), dest_size - dest_len); \
268 } \
269 (dest)[dest_size - 1] = 0; \
270 } while (0)
272 static int
273 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
274 {
275 yajl_struct *state = (yajl_struct*) ctx;
276 char buffer[number_len+1];
277 char key[2 * DATA_MAX_NAME_LEN] = { 0 };
278 _Bool latency_type = 0;
279 int status;
281 memcpy(buffer, number_val, number_len);
282 buffer[sizeof(buffer) - 1] = '\0';
284 for (size_t i = 0; i < state->depth; i++)
285 {
286 if (state->stack[i] == NULL)
287 continue;
289 if (strlen (key) != 0)
290 BUFFER_ADD (key, ".");
291 BUFFER_ADD (key, state->stack[i]);
292 }
294 /* Special case for latency metrics. */
295 if ((strcmp ("avgcount", state->key) == 0)
296 || (strcmp ("sum", state->key) == 0))
297 {
298 latency_type = 1;
300 /* depth >= 2 => (stack[-1] != NULL && stack[-2] != NULL) */
301 assert ((state->depth < 2)
302 || ((state->stack[state->depth - 1] != NULL)
303 && (state->stack[state->depth - 2] != NULL)));
305 /* Super-special case for filestore.journal_wr_bytes.avgcount: For
306 * some reason, Ceph schema encodes this as a count/sum pair while all
307 * other "Bytes" data (excluding used/capacity bytes for OSD space) uses
308 * a single "Derive" type. To spare further confusion, keep this KPI as
309 * the same type of other "Bytes". Instead of keeping an "average" or
310 * "rate", use the "sum" in the pair and assign that to the derive
311 * value. */
312 if (convert_special_metrics && (state->depth >= 2)
313 && (strcmp("filestore", state->stack[state->depth - 2]) == 0)
314 && (strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0)
315 && (strcmp("avgcount", state->key) == 0))
316 {
317 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
318 return CEPH_CB_CONTINUE;
319 }
320 }
321 else /* not a latency type */
322 {
323 BUFFER_ADD (key, ".");
324 BUFFER_ADD (key, state->key);
325 }
327 status = state->handler(state->handler_arg, buffer, key);
328 if((status == RETRY_AVGCOUNT) && latency_type)
329 {
330 /* Add previously skipped part of the key, either "avgcount" or "sum",
331 * and try again. */
332 BUFFER_ADD (key, ".");
333 BUFFER_ADD (key, state->key);
335 status = state->handler(state->handler_arg, buffer, key);
336 }
338 if (status != 0)
339 {
340 ERROR("ceph plugin: JSON handler failed with status %d.", status);
341 return CEPH_CB_ABORT;
342 }
344 return CEPH_CB_CONTINUE;
345 }
347 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
348 yajl_len_t string_len)
349 {
350 return CEPH_CB_CONTINUE;
351 }
353 static int ceph_cb_start_map(void *ctx)
354 {
355 yajl_struct *state = (yajl_struct*) ctx;
357 /* Push key to the stack */
358 if (state->depth == YAJL_MAX_DEPTH)
359 return CEPH_CB_ABORT;
361 state->stack[state->depth] = state->key;
362 state->depth++;
363 state->key = NULL;
365 return CEPH_CB_CONTINUE;
366 }
368 static int ceph_cb_end_map(void *ctx)
369 {
370 yajl_struct *state = (yajl_struct*) ctx;
372 /* Pop key from the stack */
373 if (state->depth == 0)
374 return CEPH_CB_ABORT;
376 sfree (state->key);
377 state->depth--;
378 state->key = state->stack[state->depth];
379 state->stack[state->depth] = NULL;
381 return CEPH_CB_CONTINUE;
382 }
384 static int
385 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
386 {
387 yajl_struct *state = (yajl_struct*) ctx;
388 size_t sz = ((size_t) string_len) + 1;
390 sfree (state->key);
391 state->key = malloc (sz);
392 if (state->key == NULL)
393 {
394 ERROR ("ceph plugin: malloc failed.");
395 return CEPH_CB_ABORT;
396 }
398 memmove (state->key, key, sz - 1);
399 state->key[sz - 1] = 0;
401 return CEPH_CB_CONTINUE;
402 }
404 static int ceph_cb_start_array(void *ctx)
405 {
406 return CEPH_CB_CONTINUE;
407 }
409 static int ceph_cb_end_array(void *ctx)
410 {
411 return CEPH_CB_CONTINUE;
412 }
414 static yajl_callbacks callbacks = {
415 ceph_cb_null,
416 ceph_cb_boolean,
417 NULL,
418 NULL,
419 ceph_cb_number,
420 ceph_cb_string,
421 ceph_cb_start_map,
422 ceph_cb_map_key,
423 ceph_cb_end_map,
424 ceph_cb_start_array,
425 ceph_cb_end_array
426 };
428 static void ceph_daemon_print(const struct ceph_daemon *d)
429 {
430 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
431 }
433 static void ceph_daemons_print(void)
434 {
435 for(size_t i = 0; i < g_num_daemons; ++i)
436 {
437 ceph_daemon_print(g_daemons[i]);
438 }
439 }
441 static void ceph_daemon_free(struct ceph_daemon *d)
442 {
443 for(int i = 0; i < d->last_idx; i++)
444 {
445 sfree(d->last_poll_data[i]);
446 }
447 sfree(d->last_poll_data);
448 d->last_poll_data = NULL;
449 d->last_idx = 0;
451 for(int i = 0; i < d->ds_num; i++)
452 {
453 sfree(d->ds_names[i]);
454 }
455 sfree(d->ds_types);
456 sfree(d->ds_names);
457 sfree(d);
458 }
460 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
461 * intput string. Characters following these special characters are capitalized.
462 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
463 * "Minus". */
464 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
465 {
466 char *src_copy;
467 size_t src_len;
468 char *ptr = buffer;
469 size_t ptr_size = buffer_size;
470 _Bool append_plus = 0;
471 _Bool append_minus = 0;
473 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
474 return EINVAL;
476 src_copy = strdup (src);
477 src_len = strlen(src);
479 /* Remove trailing "+" and "-". */
480 if (src_copy[src_len - 1] == '+')
481 {
482 append_plus = 1;
483 src_len--;
484 src_copy[src_len] = 0;
485 }
486 else if (src_copy[src_len - 1] == '-')
487 {
488 append_minus = 1;
489 src_len--;
490 src_copy[src_len] = 0;
491 }
493 /* Split at special chars, capitalize first character, append to buffer. */
494 char *dummy = src_copy;
495 char *token;
496 char *save_ptr = NULL;
497 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
498 {
499 size_t len;
501 dummy = NULL;
503 token[0] = toupper ((int) token[0]);
505 assert (ptr_size > 1);
507 len = strlen (token);
508 if (len >= ptr_size)
509 len = ptr_size - 1;
511 assert (len > 0);
512 assert (len < ptr_size);
514 sstrncpy (ptr, token, len + 1);
515 ptr += len;
516 ptr_size -= len;
518 assert (*ptr == 0);
519 if (ptr_size <= 1)
520 break;
521 }
523 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
524 if (append_plus || append_minus)
525 {
526 char const *append = "Plus";
527 if (append_minus)
528 append = "Minus";
530 size_t offset = buffer_size - (strlen (append) + 1);
531 if (offset > strlen (buffer))
532 offset = strlen (buffer);
534 sstrncpy (buffer + offset, append, buffer_size - offset);
535 }
537 sfree (src_copy);
538 return 0;
539 }
541 static _Bool has_suffix (char const *str, char const *suffix)
542 {
543 size_t str_len = strlen (str);
544 size_t suffix_len = strlen (suffix);
545 size_t offset;
547 if (suffix_len > str_len)
548 return 0;
549 offset = str_len - suffix_len;
551 if (strcmp (str + offset, suffix) == 0)
552 return 1;
554 return 0;
555 }
557 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
558 static size_t count_parts (char const *key)
559 {
560 size_t parts_num = 0;
562 for (const char *ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
563 parts_num++;
565 return parts_num;
566 }
568 /**
569 * Parse key to remove "type" if this is for schema and initiate compaction
570 */
571 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
572 {
573 char tmp[2 * buffer_size];
575 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
576 return EINVAL;
578 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
579 {
580 /* strip ".type" suffix iff the key has more than two parts. */
581 size_t sz = strlen (key_str) - strlen (".type") + 1;
583 if (sz > sizeof (tmp))
584 sz = sizeof (tmp);
585 sstrncpy (tmp, key_str, sz);
586 }
587 else
588 {
589 sstrncpy (tmp, key_str, sizeof (tmp));
590 }
592 return compact_ds_name (buffer, buffer_size, tmp);
593 }
595 /**
596 * while parsing ceph admin socket schema, save counter name and type for later
597 * data processing
598 */
599 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
600 int pc_type)
601 {
602 uint32_t type;
603 char ds_name[DATA_MAX_NAME_LEN];
605 if(convert_special_metrics)
606 {
607 /**
608 * Special case for filestore:JournalWrBytes. For some reason, Ceph
609 * schema encodes this as a count/sum pair while all other "Bytes" data
610 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
611 * type. To spare further confusion, keep this KPI as the same type of
612 * other "Bytes". Instead of keeping an "average" or "rate", use the
613 * "sum" in the pair and assign that to the derive value.
614 */
615 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
616 {
617 pc_type = 10;
618 }
619 }
621 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
622 if(!d->ds_names)
623 {
624 return -ENOMEM;
625 }
627 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
628 if(!d->ds_types)
629 {
630 return -ENOMEM;
631 }
633 d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN);
634 if(!d->ds_names[d->ds_num])
635 {
636 return -ENOMEM;
637 }
639 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
640 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
641 d->ds_types[d->ds_num] = type;
643 if (parse_keys(ds_name, sizeof (ds_name), name))
644 {
645 return 1;
646 }
648 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
649 d->ds_num = (d->ds_num + 1);
651 return 0;
652 }
654 /******* ceph_config *******/
655 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
656 {
657 const char *val;
658 if(item->values_num != 1)
659 {
660 return -ENOTSUP;
661 }
662 if(item->values[0].type != OCONFIG_TYPE_STRING)
663 {
664 return -ENOTSUP;
665 }
666 val = item->values[0].value.string;
667 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
668 {
669 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
670 item->key);
671 return -ENAMETOOLONG;
672 }
673 return 0;
674 }
676 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
677 {
678 if(item->values_num != 1)
679 {
680 return -ENOTSUP;
681 }
683 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
684 {
685 return -ENOTSUP;
686 }
688 *dest = (item->values[0].value.boolean) ? 1 : 0;
689 return 0;
690 }
692 static int cc_add_daemon_config(oconfig_item_t *ci)
693 {
694 int ret;
695 struct ceph_daemon *nd, cd = { 0 };
696 struct ceph_daemon **tmp;
698 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
699 {
700 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
701 "argument.");
702 return (-1);
703 }
705 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
706 if(ret)
707 {
708 return ret;
709 }
711 for(int i=0; i < ci->children_num; i++)
712 {
713 oconfig_item_t *child = ci->children + i;
715 if(strcasecmp("SocketPath", child->key) == 0)
716 {
717 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
718 if(ret)
719 {
720 return ret;
721 }
722 }
723 else
724 {
725 WARNING("ceph plugin: ignoring unknown option %s", child->key);
726 }
727 }
728 if(cd.name[0] == '\0')
729 {
730 ERROR("ceph plugin: you must configure a daemon name.\n");
731 return -EINVAL;
732 }
733 else if(cd.asok_path[0] == '\0')
734 {
735 ERROR("ceph plugin(name=%s): you must configure an administrative "
736 "socket path.\n", cd.name);
737 return -EINVAL;
738 }
739 else if(!((cd.asok_path[0] == '/') ||
740 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
741 {
742 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
743 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
744 return -EINVAL;
745 }
747 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
748 if(tmp == NULL)
749 {
750 /* The positive return value here indicates that this is a
751 * runtime error, not a configuration error. */
752 return ENOMEM;
753 }
754 g_daemons = tmp;
756 nd = malloc(sizeof (*nd));
757 if(!nd)
758 {
759 return ENOMEM;
760 }
761 memcpy(nd, &cd, sizeof(*nd));
762 g_daemons[g_num_daemons] = nd;
763 g_num_daemons++;
764 return 0;
765 }
767 static int ceph_config(oconfig_item_t *ci)
768 {
769 int ret;
771 for(int i = 0; i < ci->children_num; ++i)
772 {
773 oconfig_item_t *child = ci->children + i;
774 if(strcasecmp("Daemon", child->key) == 0)
775 {
776 ret = cc_add_daemon_config(child);
777 if(ret == ENOMEM)
778 {
779 ERROR("ceph plugin: Couldn't allocate memory");
780 return ret;
781 }
782 else if(ret)
783 {
784 //process other daemons and ignore this one
785 continue;
786 }
787 }
788 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
789 {
790 ret = cc_handle_bool(child, &long_run_latency_avg);
791 if(ret)
792 {
793 return ret;
794 }
795 }
796 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
797 {
798 ret = cc_handle_bool(child, &convert_special_metrics);
799 if(ret)
800 {
801 return ret;
802 }
803 }
804 else
805 {
806 WARNING("ceph plugin: ignoring unknown option %s", child->key);
807 }
808 }
809 return 0;
810 }
812 /**
813 * Parse JSON and get error message if present
814 */
815 static int
816 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
817 {
818 yajl_status status = yajl_parse(hand, json, json_len);
819 unsigned char *msg;
821 switch(status)
822 {
823 case yajl_status_error:
824 msg = yajl_get_error(hand, /* verbose = */ 1,
825 /* jsonText = */ (unsigned char *) json,
826 (unsigned int) json_len);
827 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
828 yajl_free_error(hand, msg);
829 return 1;
830 case yajl_status_client_canceled:
831 return 1;
832 default:
833 return 0;
834 }
835 }
837 /**
838 * Add entry for each counter while parsing schema
839 */
840 static int
841 node_handler_define_schema(void *arg, const char *val, const char *key)
842 {
843 struct ceph_daemon *d = (struct ceph_daemon *) arg;
844 int pc_type;
845 pc_type = atoi(val);
846 return ceph_daemon_add_ds_entry(d, key, pc_type);
847 }
849 /**
850 * Latency counter does not yet have an entry in last poll data - add it.
851 */
852 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
853 uint64_t cur_count)
854 {
855 d->last_poll_data[d->last_idx] = malloc(sizeof (*d->last_poll_data[d->last_idx]));
856 if(!d->last_poll_data[d->last_idx])
857 {
858 return -ENOMEM;
859 }
860 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
861 sizeof(d->last_poll_data[d->last_idx]->ds_name));
862 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
863 d->last_poll_data[d->last_idx]->last_count = cur_count;
864 d->last_idx = (d->last_idx + 1);
865 return 0;
866 }
868 /**
869 * Update latency counter or add new entry if it doesn't exist
870 */
871 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
872 double cur_sum, uint64_t cur_count)
873 {
874 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
875 {
876 d->last_poll_data[index]->last_sum = cur_sum;
877 d->last_poll_data[index]->last_count = cur_count;
878 return 0;
879 }
881 if(!d->last_poll_data)
882 {
883 d->last_poll_data = malloc(sizeof (*d->last_poll_data));
884 if(!d->last_poll_data)
885 {
886 return -ENOMEM;
887 }
888 }
889 else
890 {
891 struct last_data **tmp_last = realloc(d->last_poll_data,
892 ((d->last_idx+1) * sizeof(struct last_data *)));
893 if(!tmp_last)
894 {
895 return -ENOMEM;
896 }
897 d->last_poll_data = tmp_last;
898 }
899 return add_last(d, ds_n, cur_sum, cur_count);
900 }
902 /**
903 * If using index guess failed (shouldn't happen, but possible if counters
904 * get rearranged), resort to searching for counter name
905 */
906 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
907 {
908 for(int i = 0; i < d->last_idx; i++)
909 {
910 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
911 {
912 return i;
913 }
914 }
915 return -1;
916 }
918 /**
919 * Calculate average b/t current data and last poll data
920 * if last poll data exists
921 */
922 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
923 double cur_sum, uint64_t cur_count)
924 {
925 double result = -1.1, sum_delt = 0.0;
926 uint64_t count_delt = 0;
927 int tmp_index = 0;
928 if(d->last_idx > index)
929 {
930 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
931 {
932 tmp_index = index;
933 }
934 //test previous index
935 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
936 {
937 tmp_index = (index - 1);
938 }
939 else
940 {
941 tmp_index = backup_search_for_last_avg(d, ds_n);
942 }
944 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
945 {
946 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
947 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
948 result = (sum_delt / count_delt);
949 }
950 }
952 if(result == -1.1)
953 {
954 result = NAN;
955 }
956 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
957 {
958 return -ENOMEM;
959 }
960 return result;
961 }
963 /**
964 * If using index guess failed, resort to searching for counter name
965 */
966 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
967 {
968 for(int i = 0; i < d->ds_num; i++)
969 {
970 if(strcmp(d->ds_names[i], ds_name) == 0)
971 {
972 return d->ds_types[i];
973 }
974 }
975 return DSET_TYPE_UNFOUND;
976 }
978 /**
979 * Process counter data and dispatch values
980 */
981 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
982 {
983 value_t uv;
984 double tmp_d;
985 uint64_t tmp_u;
986 struct values_tmp *vtmp = (struct values_tmp*) arg;
987 uint32_t type = DSET_TYPE_UNFOUND;
988 int index = vtmp->index;
990 char ds_name[DATA_MAX_NAME_LEN];
992 if (parse_keys (ds_name, sizeof (ds_name), key))
993 {
994 return 1;
995 }
997 if(index >= vtmp->d->ds_num)
998 {
999 //don't overflow bounds of array
1000 index = (vtmp->d->ds_num - 1);
1001 }
1003 /**
1004 * counters should remain in same order we parsed schema... we maintain the
1005 * index variable to keep track of current point in list of counters. first
1006 * use index to guess point in array for retrieving type. if that doesn't
1007 * work, use the old way to get the counter type
1008 */
1009 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
1010 {
1011 //found match
1012 type = vtmp->d->ds_types[index];
1013 }
1014 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
1015 {
1016 //try previous key
1017 type = vtmp->d->ds_types[index-1];
1018 }
1020 if(type == DSET_TYPE_UNFOUND)
1021 {
1022 //couldn't find right type by guessing, check the old way
1023 type = backup_search_for_type(vtmp->d, ds_name);
1024 }
1026 switch(type)
1027 {
1028 case DSET_LATENCY:
1029 if(vtmp->avgcount_exists == -1)
1030 {
1031 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1032 vtmp->avgcount_exists = 0;
1033 //return after saving avgcount - don't dispatch value
1034 //until latency calculation
1035 return 0;
1036 }
1037 else
1038 {
1039 double sum, result;
1040 sscanf(val, "%lf", &sum);
1042 if(vtmp->avgcount == 0)
1043 {
1044 vtmp->avgcount = 1;
1045 }
1047 /** User wants latency values as long run avg */
1048 if(long_run_latency_avg)
1049 {
1050 result = (sum / vtmp->avgcount);
1051 }
1052 else
1053 {
1054 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1055 if(result == -ENOMEM)
1056 {
1057 return -ENOMEM;
1058 }
1059 }
1061 uv.gauge = result;
1062 vtmp->avgcount_exists = -1;
1063 vtmp->latency_index = (vtmp->latency_index + 1);
1064 }
1065 break;
1066 case DSET_BYTES:
1067 sscanf(val, "%lf", &tmp_d);
1068 uv.gauge = tmp_d;
1069 break;
1070 case DSET_RATE:
1071 sscanf(val, "%" PRIu64, &tmp_u);
1072 uv.derive = tmp_u;
1073 break;
1074 case DSET_TYPE_UNFOUND:
1075 default:
1076 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1077 return -1;
1078 }
1080 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1081 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1082 vtmp->vlist.values = &uv;
1083 vtmp->vlist.values_len = 1;
1085 vtmp->index = (vtmp->index + 1);
1086 plugin_dispatch_values(&vtmp->vlist);
1088 return 0;
1089 }
1091 static int cconn_connect(struct cconn *io)
1092 {
1093 struct sockaddr_un address = { 0 };
1094 int flags, fd, err;
1095 if(io->state != CSTATE_UNCONNECTED)
1096 {
1097 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1098 return -EDOM;
1099 }
1100 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1101 if(fd < 0)
1102 {
1103 err = -errno;
1104 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1105 "failed: error %d", err);
1106 return err;
1107 }
1108 address.sun_family = AF_UNIX;
1109 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1110 io->d->asok_path);
1111 RETRY_ON_EINTR(err,
1112 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1113 if(err < 0)
1114 {
1115 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1116 fd, err);
1117 close(fd);
1118 return err;
1119 }
1121 flags = fcntl(fd, F_GETFL, 0);
1122 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1123 {
1124 err = -errno;
1125 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1126 fd, err);
1127 close(fd);
1128 return err;
1129 }
1130 io->asok = fd;
1131 io->state = CSTATE_WRITE_REQUEST;
1132 io->amt = 0;
1133 io->json_len = 0;
1134 io->json = NULL;
1135 return 0;
1136 }
1138 static void cconn_close(struct cconn *io)
1139 {
1140 io->state = CSTATE_UNCONNECTED;
1141 if(io->asok != -1)
1142 {
1143 int res;
1144 RETRY_ON_EINTR(res, close(io->asok));
1145 }
1146 io->asok = -1;
1147 io->amt = 0;
1148 io->json_len = 0;
1149 sfree(io->json);
1150 io->json = NULL;
1151 }
1153 /* Process incoming JSON counter data */
1154 static int
1155 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1156 {
1157 int ret;
1158 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1159 if(!vtmp)
1160 {
1161 return -ENOMEM;
1162 }
1164 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1165 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1166 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1167 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1169 vtmp->d = io->d;
1170 vtmp->avgcount_exists = -1;
1171 vtmp->latency_index = 0;
1172 vtmp->index = 0;
1173 yajl->handler_arg = vtmp;
1174 ret = traverse_json(io->json, io->json_len, hand);
1175 sfree(vtmp);
1176 return ret;
1177 }
1179 /**
1180 * Initiate JSON parsing and print error if one occurs
1181 */
1182 static int cconn_process_json(struct cconn *io)
1183 {
1184 if((io->request_type != ASOK_REQ_DATA) &&
1185 (io->request_type != ASOK_REQ_SCHEMA))
1186 {
1187 return -EDOM;
1188 }
1190 int result = 1;
1191 yajl_handle hand;
1192 yajl_status status;
1194 hand = yajl_alloc(&callbacks,
1195 #if HAVE_YAJL_V2
1196 /* alloc funcs = */ NULL,
1197 #else
1198 /* alloc funcs = */ NULL, NULL,
1199 #endif
1200 /* context = */ (void *)(&io->yajl));
1202 if(!hand)
1203 {
1204 ERROR ("ceph plugin: yajl_alloc failed.");
1205 return ENOMEM;
1206 }
1208 io->yajl.depth = 0;
1210 switch(io->request_type)
1211 {
1212 case ASOK_REQ_DATA:
1213 io->yajl.handler = node_handler_fetch_data;
1214 result = cconn_process_data(io, &io->yajl, hand);
1215 break;
1216 case ASOK_REQ_SCHEMA:
1217 //init daemon specific variables
1218 io->d->ds_num = 0;
1219 io->d->last_idx = 0;
1220 io->d->last_poll_data = NULL;
1221 io->yajl.handler = node_handler_define_schema;
1222 io->yajl.handler_arg = io->d;
1223 result = traverse_json(io->json, io->json_len, hand);
1224 break;
1225 }
1227 if(result)
1228 {
1229 goto done;
1230 }
1232 #if HAVE_YAJL_V2
1233 status = yajl_complete_parse(hand);
1234 #else
1235 status = yajl_parse_complete(hand);
1236 #endif
1238 if (status != yajl_status_ok)
1239 {
1240 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1241 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1242 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1243 (char *) errmsg);
1244 yajl_free_error (hand, errmsg);
1245 yajl_free (hand);
1246 return 1;
1247 }
1249 done:
1250 yajl_free (hand);
1251 return result;
1252 }
1254 static int cconn_validate_revents(struct cconn *io, int revents)
1255 {
1256 if(revents & POLLERR)
1257 {
1258 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1259 io->d->name);
1260 return -EIO;
1261 }
1262 switch (io->state)
1263 {
1264 case CSTATE_WRITE_REQUEST:
1265 return (revents & POLLOUT) ? 0 : -EINVAL;
1266 case CSTATE_READ_VERSION:
1267 case CSTATE_READ_AMT:
1268 case CSTATE_READ_JSON:
1269 return (revents & POLLIN) ? 0 : -EINVAL;
1270 default:
1271 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1272 "illegal state on line %d", io->d->name, __LINE__);
1273 return -EDOM;
1274 }
1275 }
1277 /** Handle a network event for a connection */
1278 static int cconn_handle_event(struct cconn *io)
1279 {
1280 int ret;
1281 switch (io->state)
1282 {
1283 case CSTATE_UNCONNECTED:
1284 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1285 "state on line %d", io->d->name, __LINE__);
1287 return -EDOM;
1288 case CSTATE_WRITE_REQUEST:
1289 {
1290 char cmd[32];
1291 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1292 io->request_type, "\" }\n");
1293 size_t cmd_len = strlen(cmd);
1294 RETRY_ON_EINTR(ret,
1295 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1296 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1297 io->d->name, io->state, io->amt, ret);
1298 if(ret < 0)
1299 {
1300 return ret;
1301 }
1302 io->amt += ret;
1303 if(io->amt >= cmd_len)
1304 {
1305 io->amt = 0;
1306 switch (io->request_type)
1307 {
1308 case ASOK_REQ_VERSION:
1309 io->state = CSTATE_READ_VERSION;
1310 break;
1311 default:
1312 io->state = CSTATE_READ_AMT;
1313 break;
1314 }
1315 }
1316 return 0;
1317 }
1318 case CSTATE_READ_VERSION:
1319 {
1320 RETRY_ON_EINTR(ret,
1321 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1322 sizeof(io->d->version) - io->amt));
1323 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1324 io->d->name, io->state, ret);
1325 if(ret < 0)
1326 {
1327 return ret;
1328 }
1329 io->amt += ret;
1330 if(io->amt >= sizeof(io->d->version))
1331 {
1332 io->d->version = ntohl(io->d->version);
1333 if(io->d->version != 1)
1334 {
1335 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1336 "expecting version %d!", io->d->name, io->d->version);
1337 return -ENOTSUP;
1338 }
1339 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1340 "version %d", io->d->name, io->d->version);
1341 io->amt = 0;
1342 cconn_close(io);
1343 io->request_type = ASOK_REQ_SCHEMA;
1344 }
1345 return 0;
1346 }
1347 case CSTATE_READ_AMT:
1348 {
1349 RETRY_ON_EINTR(ret,
1350 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1351 sizeof(io->json_len) - io->amt));
1352 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1353 io->d->name, io->state, ret);
1354 if(ret < 0)
1355 {
1356 return ret;
1357 }
1358 io->amt += ret;
1359 if(io->amt >= sizeof(io->json_len))
1360 {
1361 io->json_len = ntohl(io->json_len);
1362 io->amt = 0;
1363 io->state = CSTATE_READ_JSON;
1364 io->json = calloc(1, io->json_len + 1);
1365 if(!io->json)
1366 {
1367 ERROR("ceph plugin: error callocing io->json");
1368 return -ENOMEM;
1369 }
1370 }
1371 return 0;
1372 }
1373 case CSTATE_READ_JSON:
1374 {
1375 RETRY_ON_EINTR(ret,
1376 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1377 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1378 io->d->name, io->state, ret);
1379 if(ret < 0)
1380 {
1381 return ret;
1382 }
1383 io->amt += ret;
1384 if(io->amt >= io->json_len)
1385 {
1386 ret = cconn_process_json(io);
1387 if(ret)
1388 {
1389 return ret;
1390 }
1391 cconn_close(io);
1392 io->request_type = ASOK_REQ_NONE;
1393 }
1394 return 0;
1395 }
1396 default:
1397 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1398 "state on line %d", io->d->name, __LINE__);
1399 return -EDOM;
1400 }
1401 }
1403 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1404 {
1405 int ret;
1406 if(io->request_type == ASOK_REQ_NONE)
1407 {
1408 /* The request has already been serviced. */
1409 return 0;
1410 }
1411 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1412 {
1413 /* If there are no counters to report on, don't bother
1414 * connecting */
1415 return 0;
1416 }
1418 switch (io->state)
1419 {
1420 case CSTATE_UNCONNECTED:
1421 ret = cconn_connect(io);
1422 if(ret > 0)
1423 {
1424 return -ret;
1425 }
1426 else if(ret < 0)
1427 {
1428 return ret;
1429 }
1430 fds->fd = io->asok;
1431 fds->events = POLLOUT;
1432 return 1;
1433 case CSTATE_WRITE_REQUEST:
1434 fds->fd = io->asok;
1435 fds->events = POLLOUT;
1436 return 1;
1437 case CSTATE_READ_VERSION:
1438 case CSTATE_READ_AMT:
1439 case CSTATE_READ_JSON:
1440 fds->fd = io->asok;
1441 fds->events = POLLIN;
1442 return 1;
1443 default:
1444 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1445 "on line %d", io->d->name, __LINE__);
1446 return -EDOM;
1447 }
1448 }
1450 /** Returns the difference between two struct timevals in milliseconds.
1451 * On overflow, we return max/min int.
1452 */
1453 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1454 {
1455 int64_t ret;
1456 int sec_diff = t1->tv_sec - t2->tv_sec;
1457 int usec_diff = t1->tv_usec - t2->tv_usec;
1458 ret = usec_diff / 1000;
1459 ret += (sec_diff * 1000);
1460 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1461 }
1463 /** This handles the actual network I/O to talk to the Ceph daemons.
1464 */
1465 static int cconn_main_loop(uint32_t request_type)
1466 {
1467 int ret, some_unreachable = 0;
1468 struct timeval end_tv;
1469 struct cconn io_array[g_num_daemons];
1471 DEBUG ("ceph plugin: entering cconn_main_loop(request_type = %"PRIu32")", request_type);
1473 if (g_num_daemons < 1)
1474 {
1475 ERROR ("ceph plugin: No daemons configured. See the \"Daemon\" config option.");
1476 return ENOENT;
1477 }
1479 /* create cconn array */
1480 for (size_t i = 0; i < g_num_daemons; i++)
1481 {
1482 io_array[i] = (struct cconn) {
1483 .d = g_daemons[i],
1484 .request_type = request_type,
1485 .state = CSTATE_UNCONNECTED,
1486 };
1487 }
1489 /** Calculate the time at which we should give up */
1490 gettimeofday(&end_tv, NULL);
1491 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1493 while (1)
1494 {
1495 int nfds, diff;
1496 struct timeval tv;
1497 struct cconn *polled_io_array[g_num_daemons];
1498 struct pollfd fds[g_num_daemons];
1499 memset(fds, 0, sizeof(fds));
1500 nfds = 0;
1501 for(size_t i = 0; i < g_num_daemons; ++i)
1502 {
1503 struct cconn *io = io_array + i;
1504 ret = cconn_prepare(io, fds + nfds);
1505 if(ret < 0)
1506 {
1507 WARNING("ceph plugin: cconn_prepare(name=%s,i=%zu,st=%d)=%d",
1508 io->d->name, i, io->state, ret);
1509 cconn_close(io);
1510 io->request_type = ASOK_REQ_NONE;
1511 some_unreachable = 1;
1512 }
1513 else if(ret == 1)
1514 {
1515 polled_io_array[nfds++] = io_array + i;
1516 }
1517 }
1518 if(nfds == 0)
1519 {
1520 /* finished */
1521 ret = 0;
1522 goto done;
1523 }
1524 gettimeofday(&tv, NULL);
1525 diff = milli_diff(&end_tv, &tv);
1526 if(diff <= 0)
1527 {
1528 /* Timed out */
1529 ret = -ETIMEDOUT;
1530 WARNING("ceph plugin: cconn_main_loop: timed out.");
1531 goto done;
1532 }
1533 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1534 if(ret < 0)
1535 {
1536 ERROR("ceph plugin: poll(2) error: %d", ret);
1537 goto done;
1538 }
1539 for(int i = 0; i < nfds; ++i)
1540 {
1541 struct cconn *io = polled_io_array[i];
1542 int revents = fds[i].revents;
1543 if(revents == 0)
1544 {
1545 /* do nothing */
1546 continue;
1547 }
1548 else if(cconn_validate_revents(io, revents))
1549 {
1550 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1551 "revents validation error: "
1552 "revents=0x%08x", io->d->name, i, io->state, revents);
1553 cconn_close(io);
1554 io->request_type = ASOK_REQ_NONE;
1555 some_unreachable = 1;
1556 }
1557 else
1558 {
1559 ret = cconn_handle_event(io);
1560 if(ret)
1561 {
1562 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1563 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1564 cconn_close(io);
1565 io->request_type = ASOK_REQ_NONE;
1566 some_unreachable = 1;
1567 }
1568 }
1569 }
1570 }
1571 done: for(size_t i = 0; i < g_num_daemons; ++i)
1572 {
1573 cconn_close(io_array + i);
1574 }
1575 if(some_unreachable)
1576 {
1577 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1578 }
1579 else
1580 {
1581 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1582 }
1583 return ret;
1584 }
1586 static int ceph_read(void)
1587 {
1588 return cconn_main_loop(ASOK_REQ_DATA);
1589 }
1591 /******* lifecycle *******/
1592 static int ceph_init(void)
1593 {
1594 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE)
1595 if (check_capability (CAP_DAC_OVERRIDE) != 0)
1596 {
1597 if (getuid () == 0)
1598 WARNING ("ceph plugin: Running collectd as root, but the "
1599 "CAP_DAC_OVERRIDE capability is missing. The plugin's read "
1600 "function will probably fail. Is your init system dropping "
1601 "capabilities?");
1602 else
1603 WARNING ("ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE "
1604 "capability. If you don't want to run collectd as root, try running "
1605 "\"setcap cap_dac_override=ep\" on the collectd binary.");
1606 }
1607 #endif
1609 ceph_daemons_print();
1611 if (g_num_daemons < 1)
1612 {
1613 ERROR ("ceph plugin: No daemons configured. See the \"Daemon\" config option.");
1614 return ENOENT;
1615 }
1617 return cconn_main_loop(ASOK_REQ_VERSION);
1618 }
1620 static int ceph_shutdown(void)
1621 {
1622 for(size_t i = 0; i < g_num_daemons; ++i)
1623 {
1624 ceph_daemon_free(g_daemons[i]);
1625 }
1626 sfree(g_daemons);
1627 g_daemons = NULL;
1628 g_num_daemons = 0;
1629 DEBUG("ceph plugin: finished ceph_shutdown");
1630 return 0;
1631 }
1633 void module_register(void)
1634 {
1635 plugin_register_complex_config("ceph", ceph_config);
1636 plugin_register_init("ceph", ceph_init);
1637 plugin_register_read("ceph", ceph_read);
1638 plugin_register_shutdown("ceph", ceph_shutdown);
1639 }
1640 /* vim: set sw=4 sts=4 et : */