06d57af666349d0d5b09042fa28515e69c9ad85e
1 /**
2 * collectd - src/ceph.c
3 * Copyright (C) 2011 New Dream Network
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License as published by the
7 * Free Software Foundation; only version 2 of the License is applicable.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 *
18 * Authors:
19 * Colin McCabe <cmccabe@alumni.cmu.edu>
20 * Dennis Zou <yunzou@cisco.com>
21 * Dan Ryder <daryder@cisco.com>
22 **/
24 #define _BSD_SOURCE
26 #include "collectd.h"
27 #include "common.h"
28 #include "plugin.h"
30 #include <arpa/inet.h>
31 #include <errno.h>
32 #include <fcntl.h>
33 #include <yajl/yajl_parse.h>
34 #if HAVE_YAJL_YAJL_VERSION_H
35 #include <yajl/yajl_version.h>
36 #endif
38 #include <limits.h>
39 #include <poll.h>
40 #include <stdint.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <strings.h>
45 #include <sys/socket.h>
46 #include <sys/time.h>
47 #include <sys/types.h>
48 #include <sys/un.h>
49 #include <unistd.h>
50 #include <math.h>
51 #include <inttypes.h>
53 #define MAX_RRD_DS_NAME_LEN 20
55 #define RETRY_AVGCOUNT -1
57 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
58 # define HAVE_YAJL_V2 1
59 #endif
61 #define RETRY_ON_EINTR(ret, expr) \
62 while(1) { \
63 ret = expr; \
64 if(ret >= 0) \
65 break; \
66 ret = -errno; \
67 if(ret != -EINTR) \
68 break; \
69 }
71 /** Timeout interval in seconds */
72 #define CEPH_TIMEOUT_INTERVAL 1
74 /** Maximum path length for a UNIX domain socket on this system */
75 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
77 /** Yajl callback returns */
78 #define CEPH_CB_CONTINUE 1
79 #define CEPH_CB_ABORT 0
81 #if HAVE_YAJL_V2
82 typedef size_t yajl_len_t;
83 #else
84 typedef unsigned int yajl_len_t;
85 #endif
87 /******* ceph_daemon *******/
88 struct ceph_daemon
89 {
90 /** Version of the admin_socket interface */
91 uint32_t version;
92 /** daemon name **/
93 char name[DATA_MAX_NAME_LEN];
95 int dset_num;
97 /** Path to the socket that we use to talk to the ceph daemon */
98 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
100 /** The set of key/value pairs that this daemon reports
101 * dset.type The daemon name
102 * dset.ds_num Number of data sources (key/value pairs)
103 * dset.ds Dynamically allocated array of key/value pairs
104 */
105 /** Dynamically allocated array **/
106 struct data_set_s *dset;
107 int **pc_types;
108 };
110 /******* JSON parsing *******/
111 typedef int (*node_handler_t)(void *, const char*, const char*);
113 /** Track state and handler while parsing JSON */
114 struct yajl_struct
115 {
116 node_handler_t handler;
117 void * handler_arg;
118 struct {
119 char key[DATA_MAX_NAME_LEN];
120 int key_len;
121 } state[YAJL_MAX_DEPTH];
122 int depth;
123 };
124 typedef struct yajl_struct yajl_struct;
126 /**
127 * Keep track of last data for latency values so we can calculate rate
128 * since last poll.
129 */
130 struct last_data **last_poll_data = NULL;
131 int last_idx = 0;
133 enum perfcounter_type_d
134 {
135 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
136 };
138 /** Give user option to use default (long run = since daemon started) avg */
139 static int long_run_latency_avg = 0;
141 /**
142 * Give user option to use default type for special cases -
143 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
144 * type as a sum/count pair and will calculate it the same as a latency value.
145 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
146 * use the DERIVE type. Unless user specifies to use given type, convert this
147 * metric to use DERIVE.
148 */
149 static int convert_special_metrics = 1;
151 /** Array of daemons to monitor */
152 static struct ceph_daemon **g_daemons = NULL;
154 /** Number of elements in g_daemons */
155 static int g_num_daemons = 0;
157 struct values_holder
158 {
159 int values_len;
160 value_t *values;
161 };
163 /**
164 * A set of values_t data that we build up in memory while parsing the JSON.
165 */
166 struct values_tmp
167 {
168 struct ceph_daemon *d;
169 int holder_num;
170 struct values_holder vh[0];
171 uint64_t avgcount;
172 };
174 /**
175 * A set of count/sum pairs to keep track of latency types and get difference
176 * between this poll data and last poll data.
177 */
178 struct last_data
179 {
180 char dset_name[DATA_MAX_NAME_LEN];
181 char ds_name[MAX_RRD_DS_NAME_LEN];
182 double last_sum;
183 uint64_t last_count;
184 };
187 /******* network I/O *******/
188 enum cstate_t
189 {
190 CSTATE_UNCONNECTED = 0,
191 CSTATE_WRITE_REQUEST,
192 CSTATE_READ_VERSION,
193 CSTATE_READ_AMT,
194 CSTATE_READ_JSON,
195 };
197 enum request_type_t
198 {
199 ASOK_REQ_VERSION = 0,
200 ASOK_REQ_DATA = 1,
201 ASOK_REQ_SCHEMA = 2,
202 ASOK_REQ_NONE = 1000,
203 };
205 struct cconn
206 {
207 /** The Ceph daemon that we're talking to */
208 struct ceph_daemon *d;
210 /** Request type */
211 uint32_t request_type;
213 /** The connection state */
214 enum cstate_t state;
216 /** The socket we use to talk to this daemon */
217 int asok;
219 /** The amount of data remaining to read / write. */
220 uint32_t amt;
222 /** Length of the JSON to read */
223 uint32_t json_len;
225 /** Buffer containing JSON data */
226 unsigned char *json;
228 /** Keep data important to yajl processing */
229 struct yajl_struct yajl;
230 };
232 static int ceph_cb_null(void *ctx)
233 {
234 return CEPH_CB_CONTINUE;
235 }
237 static int ceph_cb_boolean(void *ctx, int bool_val)
238 {
239 return CEPH_CB_CONTINUE;
240 }
242 static int
243 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
244 {
245 yajl_struct *yajl = (yajl_struct*)ctx;
246 char buffer[number_len+1];
247 int i, latency_type = 0, result;
248 char key[128];
250 memcpy(buffer, number_val, number_len);
251 buffer[sizeof(buffer) - 1] = 0;
253 ssnprintf(key, yajl->state[0].key_len, "%s", yajl->state[0].key);
254 for(i = 1; i < yajl->depth; i++)
255 {
256 if((i == yajl->depth-1) && ((strcmp(yajl->state[i].key,"avgcount") == 0)
257 || (strcmp(yajl->state[i].key,"sum") == 0)))
258 {
259 if(convert_special_metrics)
260 {
261 /**
262 * Special case for filestore:JournalWrBytes. For some reason,
263 * Ceph schema encodes this as a count/sum pair while all
264 * other "Bytes" data (excluding used/capacity bytes for OSD
265 * space) uses a single "Derive" type. To spare further
266 * confusion, keep this KPI as the same type of other "Bytes".
267 * Instead of keeping an "average" or "rate", use the "sum" in
268 * the pair and assign that to the derive value.
269 */
270 if((strcmp(yajl->state[i-1].key, "journal_wr_bytes") == 0) &&
271 (strcmp(yajl->state[i-2].key,"filestore") == 0) &&
272 (strcmp(yajl->state[i].key,"avgcount") == 0))
273 {
274 DEBUG("Skipping avgcount for filestore.JournalWrBytes");
275 yajl->depth = (yajl->depth - 1);
276 return CEPH_CB_CONTINUE;
277 }
278 }
279 //probably a avgcount/sum pair. if not - we'll try full key later
280 latency_type = 1;
281 break;
282 }
283 strncat(key, ".", 1);
284 strncat(key, yajl->state[i].key, yajl->state[i].key_len+1);
285 }
287 result = yajl->handler(yajl->handler_arg, buffer, key);
289 if((result == RETRY_AVGCOUNT) && latency_type)
290 {
291 strncat(key, ".", 1);
292 strncat(key, yajl->state[yajl->depth-1].key,
293 yajl->state[yajl->depth-1].key_len+1);
294 result = yajl->handler(yajl->handler_arg, buffer, key);
295 }
297 if(result == -ENOMEM)
298 {
299 ERROR("ceph plugin: memory allocation failed");
300 return CEPH_CB_ABORT;
301 }
303 yajl->depth = (yajl->depth - 1);
304 return CEPH_CB_CONTINUE;
305 }
307 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
308 yajl_len_t string_len)
309 {
310 return CEPH_CB_CONTINUE;
311 }
313 static int ceph_cb_start_map(void *ctx)
314 {
315 return CEPH_CB_CONTINUE;
316 }
318 static int
319 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
320 {
321 yajl_struct *yajl = (yajl_struct*)ctx;
323 if((yajl->depth+1) >= YAJL_MAX_DEPTH)
324 {
325 ERROR("ceph plugin: depth exceeds max, aborting.");
326 return CEPH_CB_ABORT;
327 }
329 char buffer[string_len+1];
331 memcpy(buffer, key, string_len);
332 buffer[sizeof(buffer) - 1] = 0;
334 snprintf(yajl->state[yajl->depth].key, sizeof(buffer), "%s", buffer);
335 yajl->state[yajl->depth].key_len = sizeof(buffer);
336 yajl->depth = (yajl->depth + 1);
338 return CEPH_CB_CONTINUE;
339 }
341 static int ceph_cb_end_map(void *ctx)
342 {
343 yajl_struct *yajl = (yajl_struct*)ctx;
345 yajl->depth = (yajl->depth - 1);
346 return CEPH_CB_CONTINUE;
347 }
349 static int ceph_cb_start_array(void *ctx)
350 {
351 return CEPH_CB_CONTINUE;
352 }
354 static int ceph_cb_end_array(void *ctx)
355 {
356 return CEPH_CB_CONTINUE;
357 }
359 static yajl_callbacks callbacks = {
360 ceph_cb_null,
361 ceph_cb_boolean,
362 NULL,
363 NULL,
364 ceph_cb_number,
365 ceph_cb_string,
366 ceph_cb_start_map,
367 ceph_cb_map_key,
368 ceph_cb_end_map,
369 ceph_cb_start_array,
370 ceph_cb_end_array
371 };
373 static void ceph_daemon_print(const struct ceph_daemon *d)
374 {
375 DEBUG("name=%s, asok_path=%s", d->name, d->asok_path);
376 }
378 static void ceph_daemons_print(void)
379 {
380 int i;
381 for(i = 0; i < g_num_daemons; ++i)
382 {
383 ceph_daemon_print(g_daemons[i]);
384 }
385 }
387 static void ceph_daemon_free(struct ceph_daemon *d)
388 {
389 int i = 0;
390 for(; i < d->dset_num; i++)
391 {
392 plugin_unregister_data_set((d->dset + i)->type);
393 sfree(d->dset->ds);
394 sfree(d->pc_types[i]);
395 }
396 sfree(d->dset);
397 sfree(d->pc_types);
398 sfree(d);
399 }
401 static void compact_ds_name(char *source, char *dest)
402 {
403 int keys_num = 0, i;
404 char *save_ptr = NULL, *tmp_ptr = source;
405 char *keys[16];
406 char len_str[3];
407 char tmp[DATA_MAX_NAME_LEN];
408 int reserved = 0;
409 int offset = 0;
410 memset(tmp, 0, sizeof(tmp));
411 if(source == NULL || dest == NULL || source[0] == '\0' || dest[0] != '\0')
412 {
413 return;
414 }
415 size_t src_len = strlen(source);
416 snprintf(len_str, sizeof(len_str), "%zu", src_len);
417 unsigned char append_status = 0x0;
418 append_status |= (source[src_len - 1] == '-') ? 0x1 : 0x0;
419 append_status |= (source[src_len - 1] == '+') ? 0x2 : 0x0;
420 while ((keys[keys_num] = strtok_r(tmp_ptr, ":_-+", &save_ptr)) != NULL)
421 {
422 tmp_ptr = NULL;
423 /** capitalize 1st char **/
424 keys[keys_num][0] = toupper(keys[keys_num][0]);
425 keys_num++;
426 if(keys_num >= 16)
427 {
428 break;
429 }
430 }
431 /** concatenate each part of source string **/
432 for(i = 0; i < keys_num; i++)
433 {
434 strcat(tmp, keys[i]);
435 }
436 tmp[DATA_MAX_NAME_LEN - 1] = '\0';
437 /** to coordinate limitation of length of ds name from RRD
438 * we will truncate ds_name
439 * when the its length is more than
440 * MAX_RRD_DS_NAME_LEN
441 */
442 if(strlen(tmp) > MAX_RRD_DS_NAME_LEN - 1)
443 {
444 append_status |= 0x4;
445 /** we should reserve space for
446 * len_str
447 */
448 reserved += 2;
449 }
450 if(append_status & 0x1)
451 {
452 /** we should reserve space for
453 * "Minus"
454 */
455 reserved += 5;
456 }
457 if(append_status & 0x2)
458 {
459 /** we should reserve space for
460 * "Plus"
461 */
462 reserved += 4;
463 }
464 snprintf(dest, MAX_RRD_DS_NAME_LEN - reserved, "%s", tmp);
465 offset = strlen(dest);
466 switch (append_status)
467 {
468 case 0x1:
469 memcpy(dest + offset, "Minus", 5);
470 break;
471 case 0x2:
472 memcpy(dest + offset, "Plus", 5);
473 break;
474 case 0x4:
475 memcpy(dest + offset, len_str, 2);
476 break;
477 case 0x5:
478 memcpy(dest + offset, "Minus", 5);
479 memcpy(dest + offset + 5, len_str, 2);
480 break;
481 case 0x6:
482 memcpy(dest + offset, "Plus", 4);
483 memcpy(dest + offset + 4, len_str, 2);
484 break;
485 default:
486 break;
487 }
488 }
489 static int parse_keys(const char *key_str, char *dset_name, char *ds_name)
490 {
491 char *ptr, *rptr;
492 size_t dset_name_len = 0;
493 size_t ds_name_len = 0;
494 char tmp_ds_name[DATA_MAX_NAME_LEN];
495 memset(tmp_ds_name, 0, sizeof(tmp_ds_name));
496 if(dset_name == NULL || ds_name == NULL || key_str == NULL ||
497 key_str[0] == '\0' || dset_name[0] != '\0' || ds_name[0] != '\0')
498 {
499 return -1;
500 }
501 if((ptr = strchr(key_str, '.')) == NULL
502 || (rptr = strrchr(key_str, '.')) == NULL)
503 {
504 strncpy(dset_name, key_str, DATA_MAX_NAME_LEN - 1);
505 strncpy(tmp_ds_name, key_str, DATA_MAX_NAME_LEN - 1);
506 goto compact;
507 }
508 dset_name_len =
509 (ptr - key_str) > (DATA_MAX_NAME_LEN - 1) ?
510 (DATA_MAX_NAME_LEN - 1) : (ptr - key_str);
511 memcpy(dset_name, key_str, dset_name_len);
512 ds_name_len =
513 (rptr - ptr) > DATA_MAX_NAME_LEN ? DATA_MAX_NAME_LEN : (rptr - ptr);
514 if(ds_name_len == 0)
515 { /** only have two keys **/
516 if(!strncmp(rptr + 1, "type", 4))
517 {/** if last key is "type",ignore **/
518 strncpy(tmp_ds_name, dset_name, DATA_MAX_NAME_LEN - 1);
519 }
520 else
521 {/** if last key isn't "type", copy last key **/
522 strncpy(tmp_ds_name, rptr + 1, DATA_MAX_NAME_LEN - 1);
523 }
524 }
525 else if(!strncmp(rptr + 1, "type", 4))
526 {/** more than two keys **/
527 memcpy(tmp_ds_name, ptr + 1, ds_name_len - 1);
528 }
529 else
530 {/** copy whole keys **/
531 strncpy(tmp_ds_name, ptr + 1, DATA_MAX_NAME_LEN - 1);
532 }
533 compact: compact_ds_name(tmp_ds_name, ds_name);
534 return 0;
535 }
537 static int get_matching_dset(const struct ceph_daemon *d, const char *name)
538 {
539 int idx;
540 for(idx = 0; idx < d->dset_num; ++idx)
541 {
542 if(strcmp(d->dset[idx].type, name) == 0)
543 {
544 return idx;
545 }
546 }
547 return -1;
548 }
550 static int get_matching_value(const struct data_set_s *dset, const char *name,
551 int num_values)
552 {
553 int idx;
554 for(idx = 0; idx < num_values; ++idx)
555 {
556 if(strcmp(dset->ds[idx].name, name) == 0)
557 {
558 return idx;
559 }
560 }
561 return -1;
562 }
564 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
565 int pc_type)
566 {
567 struct data_source_s *ds;
568 struct data_set_s *dset;
569 struct data_set_s *dset_array;
570 int **pc_types_array = NULL;
571 int *pc_types;
572 int *pc_types_new;
573 int idx = 0;
574 if(strlen(name) + 1 > DATA_MAX_NAME_LEN)
575 {
576 return -ENAMETOOLONG;
577 }
578 char dset_name[DATA_MAX_NAME_LEN];
579 char ds_name[MAX_RRD_DS_NAME_LEN];
580 memset(dset_name, 0, sizeof(dset_name));
581 memset(ds_name, 0, sizeof(ds_name));
582 if(parse_keys(name, dset_name, ds_name))
583 {
584 return 1;
585 }
586 idx = get_matching_dset(d, dset_name);
587 if(idx == -1)
588 {/* need to add a dset **/
589 dset_array = realloc(d->dset,
590 sizeof(struct data_set_s) * (d->dset_num + 1));
591 if(!dset_array)
592 {
593 return -ENOMEM;
594 }
595 pc_types_array = realloc(d->pc_types,
596 sizeof(int *) * (d->dset_num + 1));
597 if(!pc_types_array)
598 {
599 return -ENOMEM;
600 }
601 dset = &dset_array[d->dset_num];
602 /** this step is very important, otherwise,
603 * realloc for dset->ds will tricky because of
604 * a random addr in dset->ds
605 */
606 memset(dset, 0, sizeof(struct data_set_s));
607 dset->ds_num = 0;
608 snprintf(dset->type, DATA_MAX_NAME_LEN, "%s", dset_name);
609 pc_types = pc_types_array[d->dset_num] = NULL;
610 d->dset = dset_array;
611 }
612 else
613 {
614 dset = &d->dset[idx];
615 pc_types = d->pc_types[idx];
616 }
617 struct data_source_s *ds_array = realloc(dset->ds,
618 sizeof(struct data_source_s) * (dset->ds_num + 1));
619 if(!ds_array)
620 {
621 return -ENOMEM;
622 }
623 pc_types_new = realloc(pc_types, sizeof(int) * (dset->ds_num + 1));
624 if(!pc_types_new)
625 {
626 return -ENOMEM;
627 }
628 dset->ds = ds_array;
630 if(convert_special_metrics)
631 {
632 /**
633 * Special case for filestore:JournalWrBytes. For some reason, Ceph
634 * schema encodes this as a count/sum pair while all other "Bytes" data
635 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
636 * type. To spare further confusion, keep this KPI as the same type of
637 * other "Bytes". Instead of keeping an "average" or "rate", use the
638 * "sum" in the pair and assign that to the derive value.
639 */
640 if((strcmp(dset_name,"filestore") == 0) &&
641 strcmp(ds_name, "JournalWrBytes") == 0)
642 {
643 pc_type = 10;
644 }
645 }
647 if(idx == -1)
648 {
649 pc_types_array[d->dset_num] = pc_types_new;
650 d->pc_types = pc_types_array;
651 d->pc_types[d->dset_num][dset->ds_num] = pc_type;
652 d->dset_num++;
653 }
654 else
655 {
656 d->pc_types[idx] = pc_types_new;
657 d->pc_types[idx][dset->ds_num] = pc_type;
658 }
659 ds = &ds_array[dset->ds_num++];
660 snprintf(ds->name, MAX_RRD_DS_NAME_LEN, "%s", ds_name);
661 ds->type = (pc_type & PERFCOUNTER_DERIVE) ? DS_TYPE_DERIVE : DS_TYPE_GAUGE;
663 /**
664 * Use min of 0 for DERIVE types so we don't get negative values on Ceph
665 * service restart
666 */
667 ds->min = (ds->type == DS_TYPE_DERIVE) ? 0 : NAN;
668 ds->max = NAN;
669 return 0;
670 }
672 /******* ceph_config *******/
673 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
674 {
675 const char *val;
676 if(item->values_num != 1)
677 {
678 return -ENOTSUP;
679 }
680 if(item->values[0].type != OCONFIG_TYPE_STRING)
681 {
682 return -ENOTSUP;
683 }
684 val = item->values[0].value.string;
685 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
686 {
687 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
688 item->key);
689 return -ENAMETOOLONG;
690 }
691 return 0;
692 }
694 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
695 {
696 if(item->values_num != 1)
697 {
698 return -ENOTSUP;
699 }
701 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
702 {
703 return -ENOTSUP;
704 }
706 *dest = (item->values[0].value.boolean) ? 1 : 0;
707 return 0;
708 }
710 static int cc_add_daemon_config(oconfig_item_t *ci)
711 {
712 int ret, i;
713 struct ceph_daemon *array, *nd, cd;
714 memset(&cd, 0, sizeof(struct ceph_daemon));
716 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
717 {
718 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
719 "argument.");
720 return (-1);
721 }
723 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
724 if(ret)
725 {
726 return ret;
727 }
729 for(i=0; i < ci->children_num; i++)
730 {
731 oconfig_item_t *child = ci->children + i;
733 if(strcasecmp("SocketPath", child->key) == 0)
734 {
735 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
736 if(ret)
737 {
738 return ret;
739 }
740 }
741 else
742 {
743 WARNING("ceph plugin: ignoring unknown option %s", child->key);
744 }
745 }
746 if(cd.name[0] == '\0')
747 {
748 ERROR("ceph plugin: you must configure a daemon name.\n");
749 return -EINVAL;
750 }
751 else if(cd.asok_path[0] == '\0')
752 {
753 ERROR("ceph plugin(name=%s): you must configure an administrative "
754 "socket path.\n", cd.name);
755 return -EINVAL;
756 }
757 else if(!((cd.asok_path[0] == '/') ||
758 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
759 {
760 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
761 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
762 return -EINVAL;
763 }
764 array = realloc(g_daemons,
765 sizeof(struct ceph_daemon *) * (g_num_daemons + 1));
766 if(array == NULL)
767 {
768 /* The positive return value here indicates that this is a
769 * runtime error, not a configuration error. */
770 return ENOMEM;
771 }
772 g_daemons = (struct ceph_daemon**) array;
773 nd = malloc(sizeof(struct ceph_daemon));
774 if(!nd)
775 {
776 return ENOMEM;
777 }
778 memcpy(nd, &cd, sizeof(struct ceph_daemon));
779 g_daemons[g_num_daemons++] = nd;
780 return 0;
781 }
783 static int ceph_config(oconfig_item_t *ci)
784 {
785 int ret, i;
787 for(i = 0; i < ci->children_num; ++i)
788 {
789 oconfig_item_t *child = ci->children + i;
790 if(strcasecmp("Daemon", child->key) == 0)
791 {
792 ret = cc_add_daemon_config(child);
793 if(ret)
794 {
795 return ret;
796 }
797 }
798 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
799 {
800 ret = cc_handle_bool(child, &long_run_latency_avg);
801 if(ret)
802 {
803 return ret;
804 }
805 }
806 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
807 {
808 ret = cc_handle_bool(child, &convert_special_metrics);
809 if(ret)
810 {
811 return ret;
812 }
813 }
814 else
815 {
816 WARNING("ceph plugin: ignoring unknown option %s", child->key);
817 }
818 }
819 return 0;
820 }
822 static int
823 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
824 {
825 yajl_status status = yajl_parse(hand, json, json_len);
826 unsigned char *msg;
828 switch(status)
829 {
830 case yajl_status_error:
831 msg = yajl_get_error(hand, /* verbose = */ 1,
832 /* jsonText = */ (unsigned char *) json,
833 (unsigned int) json_len);
834 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
835 yajl_free_error(hand, msg);
836 return 1;
837 case yajl_status_client_canceled:
838 return 1;
839 default:
840 return 0;
841 }
842 }
844 static int
845 node_handler_define_schema(void *arg, const char *val, const char *key)
846 {
847 struct ceph_daemon *d = (struct ceph_daemon *) arg;
848 int pc_type;
849 pc_type = atoi(val);
850 DEBUG("\nceph_daemon_add_ds_entry(d=%s,key=%s,pc_type=%04x)",
851 d->name, key, pc_type);
852 return ceph_daemon_add_ds_entry(d, key, pc_type);
853 }
855 static int add_last(const char *dset_n, const char *ds_n, double cur_sum,
856 uint64_t cur_count)
857 {
858 last_poll_data[last_idx] = malloc(1 * sizeof(struct last_data));
859 if(!last_poll_data[last_idx])
860 {
861 return -ENOMEM;
862 }
863 sstrncpy(last_poll_data[last_idx]->dset_name,dset_n,
864 sizeof(last_poll_data[last_idx]->dset_name));
865 sstrncpy(last_poll_data[last_idx]->ds_name,ds_n,
866 sizeof(last_poll_data[last_idx]->ds_name));
867 last_poll_data[last_idx]->last_sum = cur_sum;
868 last_poll_data[last_idx]->last_count = cur_count;
869 last_idx++;
870 return 0;
871 }
873 static int update_last(const char *dset_n, const char *ds_n, double cur_sum,
874 uint64_t cur_count)
875 {
876 int i;
877 for(i = 0; i < last_idx; i++)
878 {
879 if(strcmp(last_poll_data[i]->dset_name,dset_n) == 0 &&
880 (strcmp(last_poll_data[i]->ds_name,ds_n) == 0))
881 {
882 last_poll_data[i]->last_sum = cur_sum;
883 last_poll_data[i]->last_count = cur_count;
884 return 0;
885 }
886 }
888 if(!last_poll_data)
889 {
890 last_poll_data = malloc(1 * sizeof(struct last_data *));
891 if(!last_poll_data)
892 {
893 return -ENOMEM;
894 }
895 }
896 else
897 {
898 struct last_data **tmp_last = realloc(last_poll_data,
899 ((last_idx+1) * sizeof(struct last_data *)));
900 if(!tmp_last)
901 {
902 return -ENOMEM;
903 }
904 last_poll_data = tmp_last;
905 }
906 return add_last(dset_n,ds_n,cur_sum,cur_count);
907 }
909 static double get_last_avg(const char *dset_n, const char *ds_n,
910 double cur_sum, uint64_t cur_count)
911 {
912 int i;
913 double result = -1.1, sum_delt = 0.0;
914 uint64_t count_delt = 0;
915 for(i = 0; i < last_idx; i++)
916 {
917 if((strcmp(last_poll_data[i]->dset_name,dset_n) == 0) &&
918 (strcmp(last_poll_data[i]->ds_name,ds_n) == 0))
919 {
920 if(cur_count < last_poll_data[i]->last_count)
921 {
922 break;
923 }
924 sum_delt = (cur_sum - last_poll_data[i]->last_sum);
925 count_delt = (cur_count - last_poll_data[i]->last_count);
926 result = (sum_delt / count_delt);
927 break;
928 }
929 }
931 if(result == -1.1)
932 {
933 result = NAN;
934 }
935 if(update_last(dset_n,ds_n,cur_sum,cur_count) == -ENOMEM)
936 {
937 return -ENOMEM;
938 }
939 return result;
940 }
942 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
943 {
944 int dset_idx, ds_idx;
945 value_t *uv;
946 char dset_name[DATA_MAX_NAME_LEN];
947 char ds_name[MAX_RRD_DS_NAME_LEN];
948 struct values_tmp *vtmp = (struct values_tmp*) arg;
949 memset(dset_name, 0, sizeof(dset_name));
950 memset(ds_name, 0, sizeof(ds_name));
951 if(parse_keys(key, dset_name, ds_name))
952 {
953 DEBUG("enter node_handler_fetch_data");
954 return 1;
955 }
956 dset_idx = get_matching_dset(vtmp->d, dset_name);
957 if(dset_idx == -1)
958 {
959 return 1;
960 }
961 ds_idx = get_matching_value(&vtmp->d->dset[dset_idx], ds_name,
962 vtmp->d->dset[dset_idx].ds_num);
963 if(ds_idx == -1)
964 {
965 DEBUG("DSet:%s, DS:%s, DSet idx:%d, DS idx:%d",
966 dset_name,ds_name,dset_idx,ds_idx);
967 return RETRY_AVGCOUNT;
968 }
969 uv = &(vtmp->vh[dset_idx].values[ds_idx]);
971 if(vtmp->d->pc_types[dset_idx][ds_idx] & PERFCOUNTER_LATENCY)
972 {
973 if(vtmp->avgcount == -1)
974 {
975 sscanf(val, "%" PRIu64, &vtmp->avgcount);
976 }
977 else
978 {
979 double sum, result;
980 sscanf(val, "%lf", &sum);
981 DEBUG("avgcount:%ld",vtmp->avgcount);
982 DEBUG("sum:%lf",sum);
984 if(vtmp->avgcount == 0)
985 {
986 vtmp->avgcount = 1;
987 }
989 /** User wants latency values as long run avg */
990 if(long_run_latency_avg)
991 {
992 result = (sum / vtmp->avgcount);
993 DEBUG("uv->gauge = sumd / avgcounti = :%lf", result);
994 }
995 else
996 {
997 result = get_last_avg(dset_name, ds_name, sum, vtmp->avgcount);
998 if(result == -ENOMEM)
999 {
1000 return -ENOMEM;
1001 }
1002 DEBUG("uv->gauge = (sumd_now - sumd_last) / "
1003 "(avgcounti_now - avgcounti_last) = :%lf", result);
1004 }
1006 uv->gauge = result;
1007 vtmp->avgcount = -1;
1008 }
1009 }
1010 else if(vtmp->d->pc_types[dset_idx][ds_idx] & PERFCOUNTER_DERIVE)
1011 {
1012 uint64_t derive_val;
1013 sscanf(val, "%" PRIu64, &derive_val);
1014 uv->derive = derive_val;
1015 DEBUG("uv->derive %" PRIu64 "",(uint64_t)uv->derive);
1016 }
1017 else
1018 {
1019 double other_val;
1020 sscanf(val, "%lf", &other_val);
1021 uv->gauge = other_val;
1022 DEBUG("uv->gauge %lf",uv->gauge);
1023 }
1024 return 0;
1025 }
1027 static int cconn_connect(struct cconn *io)
1028 {
1029 struct sockaddr_un address;
1030 int flags, fd, err;
1031 if(io->state != CSTATE_UNCONNECTED)
1032 {
1033 ERROR("cconn_connect: io->state != CSTATE_UNCONNECTED");
1034 return -EDOM;
1035 }
1036 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1037 if(fd < 0)
1038 {
1039 int err = -errno;
1040 ERROR("cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) failed: "
1041 "error %d", err);
1042 return err;
1043 }
1044 memset(&address, 0, sizeof(struct sockaddr_un));
1045 address.sun_family = AF_UNIX;
1046 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1047 io->d->asok_path);
1048 RETRY_ON_EINTR(err,
1049 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1050 if(err < 0)
1051 {
1052 ERROR("cconn_connect: connect(%d) failed: error %d", fd, err);
1053 return err;
1054 }
1056 flags = fcntl(fd, F_GETFL, 0);
1057 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1058 {
1059 err = -errno;
1060 ERROR("cconn_connect: fcntl(%d, O_NONBLOCK) error %d", fd, err);
1061 return err;
1062 }
1063 io->asok = fd;
1064 io->state = CSTATE_WRITE_REQUEST;
1065 io->amt = 0;
1066 io->json_len = 0;
1067 io->json = NULL;
1068 return 0;
1069 }
1071 static void cconn_close(struct cconn *io)
1072 {
1073 io->state = CSTATE_UNCONNECTED;
1074 if(io->asok != -1)
1075 {
1076 int res;
1077 RETRY_ON_EINTR(res, close(io->asok));
1078 }
1079 io->asok = -1;
1080 io->amt = 0;
1081 io->json_len = 0;
1082 sfree(io->json);
1083 io->json = NULL;
1084 }
1086 /* Process incoming JSON counter data */
1087 static int
1088 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1089 {
1090 int i, ret = 0;
1091 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp)
1092 + (sizeof(struct values_holder)) * io->d->dset_num);
1093 if(!vtmp)
1094 {
1095 return -ENOMEM;
1096 }
1098 for(i = 0; i < io->d->dset_num; i++)
1099 {
1100 value_t *val = calloc(1, (sizeof(value_t) * io->d->dset[i].ds_num));
1101 vtmp->vh[i].values = val;
1102 vtmp->vh[i].values_len = io->d->dset[i].ds_num;
1103 }
1104 vtmp->d = io->d;
1105 vtmp->holder_num = io->d->dset_num;
1106 vtmp->avgcount = -1;
1107 yajl->handler_arg = vtmp;
1108 ret = traverse_json(io->json, io->json_len, hand);
1109 if(ret)
1110 {
1111 goto done;
1112 }
1113 for(i = 0; i < vtmp->holder_num; i++)
1114 {
1115 value_list_t vl = VALUE_LIST_INIT;
1116 sstrncpy(vl.host, hostname_g, sizeof(vl.host));
1117 sstrncpy(vl.plugin, "ceph", sizeof(vl.plugin));
1118 strncpy(vl.plugin_instance, io->d->name, sizeof(vl.plugin_instance));
1119 sstrncpy(vl.type, io->d->dset[i].type, sizeof(vl.type));
1120 vl.values = vtmp->vh[i].values;
1121 vl.values_len = io->d->dset[i].ds_num;
1122 DEBUG("cconn_process_data(io=%s): vl.values_len=%d, json=\"%s\"",
1123 io->d->name, vl.values_len, io->json);
1124 ret = plugin_dispatch_values(&vl);
1125 if(ret)
1126 {
1127 goto done;
1128 }
1129 }
1131 done: for(i = 0; i < vtmp->holder_num; i++)
1132 {
1133 sfree(vtmp->vh[i].values);
1134 }
1135 sfree(vtmp);
1136 return ret;
1137 }
1139 static int cconn_process_json(struct cconn *io)
1140 {
1141 if((io->request_type != ASOK_REQ_DATA) &&
1142 (io->request_type != ASOK_REQ_SCHEMA))
1143 {
1144 return -EDOM;
1145 }
1147 int result = 1;
1148 yajl_handle hand;
1149 yajl_status status;
1151 hand = yajl_alloc(&callbacks,
1152 #if HAVE_YAJL_V2
1153 /* alloc funcs = */ NULL,
1154 #else
1155 /* alloc funcs = */ NULL, NULL,
1156 #endif
1157 /* context = */ (void *)(&io->yajl));
1159 if(!hand)
1160 {
1161 ERROR ("ceph plugin: yajl_alloc failed.");
1162 return ENOMEM;
1163 }
1165 io->yajl.depth = 0;
1167 switch(io->request_type)
1168 {
1169 case ASOK_REQ_DATA:
1170 io->yajl.handler = node_handler_fetch_data;
1171 result = cconn_process_data(io, &io->yajl, hand);
1172 break;
1173 case ASOK_REQ_SCHEMA:
1174 io->yajl.handler = node_handler_define_schema;
1175 io->yajl.handler_arg = io->d;
1176 result = traverse_json(io->json, io->json_len, hand);
1177 break;
1178 }
1180 if(result)
1181 {
1182 goto done;
1183 }
1185 #if HAVE_YAJL_V2
1186 status = yajl_complete_parse(hand);
1187 #else
1188 status = yajl_parse_complete(hand);
1189 #endif
1191 if (status != yajl_status_ok)
1192 {
1193 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1194 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1195 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1196 (char *) errmsg);
1197 yajl_free_error (hand, errmsg);
1198 yajl_free (hand);
1199 return 1;
1200 }
1202 done:
1203 yajl_free (hand);
1204 return result;
1205 }
1207 static int cconn_validate_revents(struct cconn *io, int revents)
1208 {
1209 if(revents & POLLERR)
1210 {
1211 ERROR("cconn_validate_revents(name=%s): got POLLERR", io->d->name);
1212 return -EIO;
1213 }
1214 switch (io->state)
1215 {
1216 case CSTATE_WRITE_REQUEST:
1217 return (revents & POLLOUT) ? 0 : -EINVAL;
1218 case CSTATE_READ_VERSION:
1219 case CSTATE_READ_AMT:
1220 case CSTATE_READ_JSON:
1221 return (revents & POLLIN) ? 0 : -EINVAL;
1222 return (revents & POLLIN) ? 0 : -EINVAL;
1223 default:
1224 ERROR("cconn_validate_revents(name=%s) got to illegal state on "
1225 "line %d", io->d->name, __LINE__);
1226 return -EDOM;
1227 }
1228 }
1230 /** Handle a network event for a connection */
1231 static int cconn_handle_event(struct cconn *io)
1232 {
1233 int ret;
1234 switch (io->state)
1235 {
1236 case CSTATE_UNCONNECTED:
1237 ERROR("cconn_handle_event(name=%s) got to illegal state on line "
1238 "%d", io->d->name, __LINE__);
1240 return -EDOM;
1241 case CSTATE_WRITE_REQUEST:
1242 {
1243 char cmd[32];
1244 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1245 io->request_type, "\" }\n");
1246 size_t cmd_len = strlen(cmd);
1247 RETRY_ON_EINTR(ret,
1248 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1249 DEBUG("cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1250 io->d->name, io->state, io->amt, ret);
1251 if(ret < 0)
1252 {
1253 return ret;
1254 }
1255 io->amt += ret;
1256 if(io->amt >= cmd_len)
1257 {
1258 io->amt = 0;
1259 switch (io->request_type)
1260 {
1261 case ASOK_REQ_VERSION:
1262 io->state = CSTATE_READ_VERSION;
1263 break;
1264 default:
1265 io->state = CSTATE_READ_AMT;
1266 break;
1267 }
1268 }
1269 return 0;
1270 }
1271 case CSTATE_READ_VERSION:
1272 {
1273 RETRY_ON_EINTR(ret,
1274 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1275 sizeof(io->d->version) - io->amt));
1276 DEBUG("cconn_handle_event(name=%s,state=%d,ret=%d)",
1277 io->d->name, io->state, ret);
1278 if(ret < 0)
1279 {
1280 return ret;
1281 }
1282 io->amt += ret;
1283 if(io->amt >= sizeof(io->d->version))
1284 {
1285 io->d->version = ntohl(io->d->version);
1286 if(io->d->version != 1)
1287 {
1288 ERROR("cconn_handle_event(name=%s) not "
1289 "expecting version %d!", io->d->name, io->d->version);
1290 return -ENOTSUP;
1291 }DEBUG("cconn_handle_event(name=%s): identified as "
1292 "version %d", io->d->name, io->d->version);
1293 io->amt = 0;
1294 cconn_close(io);
1295 io->request_type = ASOK_REQ_SCHEMA;
1296 }
1297 return 0;
1298 }
1299 case CSTATE_READ_AMT:
1300 {
1301 RETRY_ON_EINTR(ret,
1302 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1303 sizeof(io->json_len) - io->amt));
1304 DEBUG("cconn_handle_event(name=%s,state=%d,ret=%d)",
1305 io->d->name, io->state, ret);
1306 if(ret < 0)
1307 {
1308 return ret;
1309 }
1310 io->amt += ret;
1311 if(io->amt >= sizeof(io->json_len))
1312 {
1313 io->json_len = ntohl(io->json_len);
1314 io->amt = 0;
1315 io->state = CSTATE_READ_JSON;
1316 io->json = calloc(1, io->json_len + 1);
1317 if(!io->json)
1318 {
1319 ERROR("ERR CALLOCING IO->JSON");
1320 return -ENOMEM;
1321 }
1322 }
1323 return 0;
1324 }
1325 case CSTATE_READ_JSON:
1326 {
1327 RETRY_ON_EINTR(ret,
1328 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1329 DEBUG("cconn_handle_event(name=%s,state=%d,ret=%d)",
1330 io->d->name, io->state, ret);
1331 if(ret < 0)
1332 {
1333 return ret;
1334 }
1335 io->amt += ret;
1336 if(io->amt >= io->json_len)
1337 {
1338 ret = cconn_process_json(io);
1339 if(ret)
1340 {
1341 return ret;
1342 }
1343 cconn_close(io);
1344 io->request_type = ASOK_REQ_NONE;
1345 }
1346 return 0;
1347 }
1348 default:
1349 ERROR("cconn_handle_event(name=%s) got to illegal state on "
1350 "line %d", io->d->name, __LINE__);
1351 return -EDOM;
1352 }
1353 }
1355 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1356 {
1357 int ret;
1358 if(io->request_type == ASOK_REQ_NONE)
1359 {
1360 /* The request has already been serviced. */
1361 return 0;
1362 }
1363 else if((io->request_type == ASOK_REQ_DATA) && (io->d->dset_num == 0))
1364 {
1365 /* If there are no counters to report on, don't bother
1366 * connecting */
1367 return 0;
1368 }
1370 switch (io->state)
1371 {
1372 case CSTATE_UNCONNECTED:
1373 ret = cconn_connect(io);
1374 if(ret > 0)
1375 {
1376 return -ret;
1377 }
1378 else if(ret < 0)
1379 {
1380 return ret;
1381 }
1382 fds->fd = io->asok;
1383 fds->events = POLLOUT;
1384 return 1;
1385 case CSTATE_WRITE_REQUEST:
1386 fds->fd = io->asok;
1387 fds->events = POLLOUT;
1388 return 1;
1389 case CSTATE_READ_VERSION:
1390 case CSTATE_READ_AMT:
1391 case CSTATE_READ_JSON:
1392 fds->fd = io->asok;
1393 fds->events = POLLIN;
1394 return 1;
1395 default:
1396 ERROR("cconn_prepare(name=%s) got to illegal state on line %d",
1397 io->d->name, __LINE__);
1398 return -EDOM;
1399 }
1400 }
1402 /** Returns the difference between two struct timevals in milliseconds.
1403 * On overflow, we return max/min int.
1404 */
1405 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1406 {
1407 int64_t ret;
1408 int sec_diff = t1->tv_sec - t2->tv_sec;
1409 int usec_diff = t1->tv_usec - t2->tv_usec;
1410 ret = usec_diff / 1000;
1411 ret += (sec_diff * 1000);
1412 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1413 }
1415 /** This handles the actual network I/O to talk to the Ceph daemons.
1416 */
1417 static int cconn_main_loop(uint32_t request_type)
1418 {
1419 int i, ret, some_unreachable = 0;
1420 struct timeval end_tv;
1421 struct cconn io_array[g_num_daemons];
1423 DEBUG("entering cconn_main_loop(request_type = %d)", request_type);
1425 /* create cconn array */
1426 memset(io_array, 0, sizeof(io_array));
1427 for(i = 0; i < g_num_daemons; ++i)
1428 {
1429 io_array[i].d = g_daemons[i];
1430 io_array[i].request_type = request_type;
1431 io_array[i].state = CSTATE_UNCONNECTED;
1432 }
1434 /** Calculate the time at which we should give up */
1435 gettimeofday(&end_tv, NULL);
1436 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1438 while (1)
1439 {
1440 int nfds, diff;
1441 struct timeval tv;
1442 struct cconn *polled_io_array[g_num_daemons];
1443 struct pollfd fds[g_num_daemons];
1444 memset(fds, 0, sizeof(fds));
1445 nfds = 0;
1446 for(i = 0; i < g_num_daemons; ++i)
1447 {
1448 struct cconn *io = io_array + i;
1449 ret = cconn_prepare(io, fds + nfds);
1450 if(ret < 0)
1451 {
1452 WARNING("ERROR: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1453 io->d->name, i, io->state, ret);
1454 cconn_close(io);
1455 io->request_type = ASOK_REQ_NONE;
1456 some_unreachable = 1;
1457 }
1458 else if(ret == 1)
1459 {
1460 DEBUG("did cconn_prepare(name=%s,i=%d,st=%d)",
1461 io->d->name, i, io->state);
1462 polled_io_array[nfds++] = io_array + i;
1463 }
1464 }
1465 if(nfds == 0)
1466 {
1467 /* finished */
1468 ret = 0;
1469 DEBUG("cconn_main_loop: no more cconn to manage.");
1470 goto done;
1471 }
1472 gettimeofday(&tv, NULL);
1473 diff = milli_diff(&end_tv, &tv);
1474 if(diff <= 0)
1475 {
1476 /* Timed out */
1477 ret = -ETIMEDOUT;
1478 WARNING("ERROR: cconn_main_loop: timed out.\n");
1479 goto done;
1480 }
1481 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1482 if(ret < 0)
1483 {
1484 ERROR("poll(2) error: %d", ret);
1485 goto done;
1486 }
1487 for(i = 0; i < nfds; ++i)
1488 {
1489 struct cconn *io = polled_io_array[i];
1490 int revents = fds[i].revents;
1491 if(revents == 0)
1492 {
1493 /* do nothing */
1494 }
1495 else if(cconn_validate_revents(io, revents))
1496 {
1497 WARNING("ERROR: cconn(name=%s,i=%d,st=%d): "
1498 "revents validation error: "
1499 "revents=0x%08x", io->d->name, i, io->state, revents);
1500 cconn_close(io);
1501 io->request_type = ASOK_REQ_NONE;
1502 some_unreachable = 1;
1503 }
1504 else
1505 {
1506 int ret = cconn_handle_event(io);
1507 if(ret)
1508 {
1509 WARNING("ERROR: cconn_handle_event(name=%s,"
1510 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1511 cconn_close(io);
1512 io->request_type = ASOK_REQ_NONE;
1513 some_unreachable = 1;
1514 }
1515 }
1516 }
1517 }
1518 done: for(i = 0; i < g_num_daemons; ++i)
1519 {
1520 cconn_close(io_array + i);
1521 }
1522 if(some_unreachable)
1523 {
1524 DEBUG("cconn_main_loop: some Ceph daemons were unreachable.");
1525 }
1526 else
1527 {
1528 DEBUG("cconn_main_loop: reached all Ceph daemons :)");
1529 }
1530 return ret;
1531 }
1533 static int ceph_read(void)
1534 {
1535 return cconn_main_loop(ASOK_REQ_DATA);
1536 }
1538 /******* lifecycle *******/
1539 static int ceph_init(void)
1540 {
1541 int i, ret, j;
1542 DEBUG("ceph_init");
1543 ceph_daemons_print();
1545 ret = cconn_main_loop(ASOK_REQ_VERSION);
1546 if(ret)
1547 {
1548 return ret;
1549 }
1550 for(i = 0; i < g_num_daemons; ++i)
1551 {
1552 struct ceph_daemon *d = g_daemons[i];
1553 for(j = 0; j < d->dset_num; j++)
1554 {
1555 ret = plugin_register_data_set(d->dset + j);
1556 if(ret)
1557 {
1558 ERROR("plugin_register_data_set(%s) failed!", d->name);
1559 }
1560 else
1561 {
1562 DEBUG("plugin_register_data_set(%s): "
1563 "(d->dset)[%d]->ds_num=%d",
1564 d->name, j, d->dset[j].ds_num);
1565 }
1566 }
1567 }
1568 return 0;
1569 }
1571 static int ceph_shutdown(void)
1572 {
1573 int i;
1574 for(i = 0; i < g_num_daemons; ++i)
1575 {
1576 ceph_daemon_free(g_daemons[i]);
1577 }
1578 sfree(g_daemons);
1579 g_daemons = NULL;
1580 g_num_daemons = 0;
1581 for(i = 0; i < last_idx; i++)
1582 {
1583 sfree(last_poll_data[i]);
1584 }
1585 sfree(last_poll_data);
1586 last_poll_data = NULL;
1587 last_idx = 0;
1588 DEBUG("finished ceph_shutdown");
1589 return 0;
1590 }
1592 void module_register(void)
1593 {
1594 plugin_register_complex_config("ceph", ceph_config);
1595 plugin_register_init("ceph", ceph_init);
1596 plugin_register_read("ceph", ceph_read);
1597 plugin_register_shutdown("ceph", ceph_shutdown);
1598 }