2 * RRDTool - src/rrd_daemon.c
3 * Copyright (C) 2008 Florian octo Forster
4 * Copyright (C) 2008 Kevin Brintnall
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.
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.
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
20 * Florian octo Forster <octo at verplant.org>
21 * kevin brintnall <kbrint@rufus.net>
26 * First tell the compiler to stick to the C99 and POSIX standards as close as
29 #ifndef __STRICT_ANSI__ /* {{{ */
30 # define __STRICT_ANSI__
33 #ifndef _ISOC99_SOURCE
34 # define _ISOC99_SOURCE
37 #ifdef _POSIX_C_SOURCE
38 # undef _POSIX_C_SOURCE
40 #define _POSIX_C_SOURCE 200112L
42 /* Single UNIX needed for strdup. */
46 #define _XOPEN_SOURCE 500
63 * Now for some includes..
66 #if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__) && !defined(HAVE_CONFIG_H)
67 #include "../win32/config.h"
70 #include "../rrd_config.h"
75 #include "rrd_client.h"
86 #include <sys/types.h>
90 #include <sys/socket.h>
101 #include <glib-2.0/glib.h>
104 #define RRDD_LOG(severity, ...) syslog ((severity), __VA_ARGS__)
107 # define __attribute__(x) /**/
119 typedef enum { RESP_ERR = -1, RESP_OK = 0 } response_code;
121 struct listen_socket_s
124 char addr[PATH_MAX + 1];
126 socket_privilege privilege;
128 /* state for BATCH processing */
140 typedef struct listen_socket_s listen_socket_t;
143 typedef struct cache_item_s cache_item_t;
149 time_t last_flush_time;
150 time_t last_update_stamp;
151 #define CI_FLAGS_IN_TREE (1<<0)
152 #define CI_FLAGS_IN_QUEUE (1<<1)
154 pthread_cond_t flushed;
159 struct callback_flush_data_s
166 typedef struct callback_flush_data_s callback_flush_data_t;
173 typedef enum queue_side_e queue_side_t;
175 /* max length of socket command or response */
177 #define RBUF_SIZE (CMD_MAX*2)
182 static int stay_foreground = 0;
183 static uid_t daemon_uid;
185 static listen_socket_t *listen_fds = NULL;
186 static size_t listen_fds_num = 0;
188 static int do_shutdown = 0;
190 static pthread_t queue_thread;
192 static pthread_t *connection_threads = NULL;
193 static pthread_mutex_t connection_threads_lock = PTHREAD_MUTEX_INITIALIZER;
194 static int connection_threads_num = 0;
197 static GTree *cache_tree = NULL;
198 static cache_item_t *cache_queue_head = NULL;
199 static cache_item_t *cache_queue_tail = NULL;
200 static pthread_mutex_t cache_lock = PTHREAD_MUTEX_INITIALIZER;
201 static pthread_cond_t cache_cond = PTHREAD_COND_INITIALIZER;
203 static int config_write_interval = 300;
204 static int config_write_jitter = 0;
205 static int config_flush_interval = 3600;
206 static int config_flush_at_shutdown = 0;
207 static char *config_pid_file = NULL;
208 static char *config_base_dir = NULL;
209 static size_t _config_base_dir_len = 0;
210 static int config_write_base_only = 0;
212 static listen_socket_t **config_listen_address_list = NULL;
213 static int config_listen_address_list_len = 0;
215 static uint64_t stats_queue_length = 0;
216 static uint64_t stats_updates_received = 0;
217 static uint64_t stats_flush_received = 0;
218 static uint64_t stats_updates_written = 0;
219 static uint64_t stats_data_sets_written = 0;
220 static uint64_t stats_journal_bytes = 0;
221 static uint64_t stats_journal_rotate = 0;
222 static pthread_mutex_t stats_lock = PTHREAD_MUTEX_INITIALIZER;
224 /* Journaled updates */
225 static char *journal_cur = NULL;
226 static char *journal_old = NULL;
227 static FILE *journal_fh = NULL;
228 static pthread_mutex_t journal_lock = PTHREAD_MUTEX_INITIALIZER;
229 static int journal_write(char *cmd, char *args);
230 static void journal_done(void);
231 static void journal_rotate(void);
236 static void sig_common (const char *sig) /* {{{ */
238 RRDD_LOG(LOG_NOTICE, "caught SIG%s", sig);
240 pthread_cond_broadcast(&cache_cond);
241 } /* }}} void sig_common */
243 static void sig_int_handler (int s __attribute__((unused))) /* {{{ */
246 } /* }}} void sig_int_handler */
248 static void sig_term_handler (int s __attribute__((unused))) /* {{{ */
251 } /* }}} void sig_term_handler */
253 static void sig_usr1_handler (int s __attribute__((unused))) /* {{{ */
255 config_flush_at_shutdown = 1;
257 } /* }}} void sig_usr1_handler */
259 static void sig_usr2_handler (int s __attribute__((unused))) /* {{{ */
261 config_flush_at_shutdown = 0;
263 } /* }}} void sig_usr2_handler */
265 static void install_signal_handlers(void) /* {{{ */
267 /* These structures are static, because `sigaction' behaves weird if the are
269 static struct sigaction sa_int;
270 static struct sigaction sa_term;
271 static struct sigaction sa_pipe;
272 static struct sigaction sa_usr1;
273 static struct sigaction sa_usr2;
275 /* Install signal handlers */
276 memset (&sa_int, 0, sizeof (sa_int));
277 sa_int.sa_handler = sig_int_handler;
278 sigaction (SIGINT, &sa_int, NULL);
280 memset (&sa_term, 0, sizeof (sa_term));
281 sa_term.sa_handler = sig_term_handler;
282 sigaction (SIGTERM, &sa_term, NULL);
284 memset (&sa_pipe, 0, sizeof (sa_pipe));
285 sa_pipe.sa_handler = SIG_IGN;
286 sigaction (SIGPIPE, &sa_pipe, NULL);
288 memset (&sa_pipe, 0, sizeof (sa_usr1));
289 sa_usr1.sa_handler = sig_usr1_handler;
290 sigaction (SIGUSR1, &sa_usr1, NULL);
292 memset (&sa_usr2, 0, sizeof (sa_usr2));
293 sa_usr2.sa_handler = sig_usr2_handler;
294 sigaction (SIGUSR2, &sa_usr2, NULL);
296 } /* }}} void install_signal_handlers */
298 static int open_pidfile(char *action, int oflag) /* {{{ */
303 file = (config_pid_file != NULL)
305 : LOCALSTATEDIR "/run/rrdcached.pid";
307 fd = open(file, oflag, S_IWUSR|S_IRUSR|S_IRGRP|S_IROTH);
309 fprintf(stderr, "rrdcached: can't %s pid file '%s' (%s)\n",
310 action, file, rrd_strerror(errno));
313 } /* }}} static int open_pidfile */
315 /* check existing pid file to see whether a daemon is running */
316 static int check_pidfile(void)
322 pid_fd = open_pidfile("open", O_RDWR);
326 if (read(pid_fd, pid_str, sizeof(pid_str)) <= 0)
333 /* another running process that we can signal COULD be
334 * a competing rrdcached */
335 if (pid != getpid() && kill(pid, 0) == 0)
338 "FATAL: Another rrdcached daemon is running?? (pid %d)\n", pid);
343 lseek(pid_fd, 0, SEEK_SET);
344 ftruncate(pid_fd, 0);
347 "rrdcached: removed stale PID file (no rrdcached on pid %d)\n"
348 "rrdcached: starting normally.\n", pid);
351 } /* }}} static int check_pidfile */
353 static int write_pidfile (int fd) /* {{{ */
360 fh = fdopen (fd, "w");
363 RRDD_LOG (LOG_ERR, "write_pidfile: fdopen() failed.");
368 fprintf (fh, "%i\n", (int) pid);
372 } /* }}} int write_pidfile */
374 static int remove_pidfile (void) /* {{{ */
379 file = (config_pid_file != NULL)
381 : LOCALSTATEDIR "/run/rrdcached.pid";
383 status = unlink (file);
387 } /* }}} int remove_pidfile */
389 static char *next_cmd (listen_socket_t *sock, ssize_t *len) /* {{{ */
393 eol = memchr(sock->rbuf + sock->next_cmd, '\n',
394 sock->next_read - sock->next_cmd);
398 /* no commands left, move remainder back to front of rbuf */
399 memmove(sock->rbuf, sock->rbuf + sock->next_cmd,
400 sock->next_read - sock->next_cmd);
401 sock->next_read -= sock->next_cmd;
408 char *cmd = sock->rbuf + sock->next_cmd;
411 sock->next_cmd = eol - sock->rbuf + 1;
413 if (eol > sock->rbuf && *(eol-1) == '\r')
414 *(--eol) = '\0'; /* handle "\r\n" EOL */
425 /* add the characters directly to the write buffer */
426 static int add_to_wbuf(listen_socket_t *sock, char *str, size_t len) /* {{{ */
430 assert(sock != NULL);
432 new_buf = realloc(sock->wbuf, sock->wbuf_len + len + 1);
435 RRDD_LOG(LOG_ERR, "add_to_wbuf: realloc failed");
439 strncpy(new_buf + sock->wbuf_len, str, len + 1);
441 sock->wbuf = new_buf;
442 sock->wbuf_len += len;
445 } /* }}} static int add_to_wbuf */
447 /* add the text to the "extra" info that's sent after the status line */
448 static int add_response_info(listen_socket_t *sock, char *fmt, ...) /* {{{ */
451 char buffer[CMD_MAX];
454 if (sock == NULL) return 0; /* journal replay mode */
455 if (sock->batch_start) return 0; /* no extra info returned when in BATCH */
458 #ifdef HAVE_VSNPRINTF
459 len = vsnprintf(buffer, sizeof(buffer)-1, fmt, argp);
461 len = vsprintf(buffer, fmt, argp);
466 RRDD_LOG(LOG_ERR, "add_response_info: vnsprintf failed");
470 return add_to_wbuf(sock, buffer, len);
471 } /* }}} static int add_response_info */
473 static int count_lines(char *str) /* {{{ */
479 while ((str = strchr(str, '\n')) != NULL)
487 } /* }}} static int count_lines */
489 /* send the response back to the user.
490 * returns 0 on success, -1 on error
491 * write buffer is always zeroed after this call */
492 static int send_response (listen_socket_t *sock, response_code rc,
493 char *fmt, ...) /* {{{ */
496 char buffer[CMD_MAX];
501 if (sock == NULL) return rc; /* journal replay mode */
503 if (sock->batch_start)
506 return rc; /* no response on success during BATCH */
507 lines = sock->batch_cmd;
509 else if (rc == RESP_OK)
510 lines = count_lines(sock->wbuf);
514 rclen = sprintf(buffer, "%d ", lines);
516 #ifdef HAVE_VSNPRINTF
517 len = vsnprintf(buffer+rclen, sizeof(buffer)-rclen-1, fmt, argp);
519 len = vsprintf(buffer+rclen, fmt, argp);
527 /* append the result to the wbuf, don't write to the user */
528 if (sock->batch_start)
529 return add_to_wbuf(sock, buffer, len);
531 /* first write must be complete */
532 if (len != write(sock->fd, buffer, len))
534 RRDD_LOG(LOG_INFO, "send_response: could not write status message");
538 if (sock->wbuf != NULL && rc == RESP_OK)
541 while (wrote < sock->wbuf_len)
543 ssize_t wb = write(sock->fd, sock->wbuf + wrote, sock->wbuf_len - wrote);
546 RRDD_LOG(LOG_INFO, "send_response: could not write results");
553 free(sock->wbuf); sock->wbuf = NULL;
559 static void wipe_ci_values(cache_item_t *ci, time_t when)
564 ci->last_flush_time = when;
565 if (config_write_jitter > 0)
566 ci->last_flush_time += (random() % config_write_jitter);
570 * remove a "cache_item_t" item from the queue.
571 * must hold 'cache_lock' when calling this
573 static void remove_from_queue(cache_item_t *ci) /* {{{ */
575 if (ci == NULL) return;
576 if ((ci->flags & CI_FLAGS_IN_QUEUE) == 0) return; /* not queued */
578 if (ci->prev == NULL)
579 cache_queue_head = ci->next; /* reset head */
581 ci->prev->next = ci->next;
583 if (ci->next == NULL)
584 cache_queue_tail = ci->prev; /* reset the tail */
586 ci->next->prev = ci->prev;
588 ci->next = ci->prev = NULL;
589 ci->flags &= ~CI_FLAGS_IN_QUEUE;
590 } /* }}} static void remove_from_queue */
592 /* free the resources associated with the cache_item_t
593 * must hold cache_lock when calling this function
595 static void *free_cache_item(cache_item_t *ci) /* {{{ */
597 if (ci == NULL) return NULL;
599 remove_from_queue(ci);
601 for (int i=0; i < ci->values_num; i++)
607 /* in case anyone is waiting */
608 pthread_cond_broadcast(&ci->flushed);
613 } /* }}} static void *free_cache_item */
616 * enqueue_cache_item:
617 * `cache_lock' must be acquired before calling this function!
619 static int enqueue_cache_item (cache_item_t *ci, /* {{{ */
625 if (ci->values_num == 0)
630 if (cache_queue_head == ci)
633 /* remove if further down in queue */
634 remove_from_queue(ci);
637 ci->next = cache_queue_head;
638 if (ci->next != NULL)
640 cache_queue_head = ci;
642 if (cache_queue_tail == NULL)
643 cache_queue_tail = cache_queue_head;
645 else /* (side == TAIL) */
647 /* We don't move values back in the list.. */
648 if (ci->flags & CI_FLAGS_IN_QUEUE)
651 assert (ci->next == NULL);
652 assert (ci->prev == NULL);
654 ci->prev = cache_queue_tail;
656 if (cache_queue_tail == NULL)
657 cache_queue_head = ci;
659 cache_queue_tail->next = ci;
661 cache_queue_tail = ci;
664 ci->flags |= CI_FLAGS_IN_QUEUE;
666 pthread_cond_broadcast(&cache_cond);
667 pthread_mutex_lock (&stats_lock);
668 stats_queue_length++;
669 pthread_mutex_unlock (&stats_lock);
672 } /* }}} int enqueue_cache_item */
675 * tree_callback_flush:
676 * Called via `g_tree_foreach' in `queue_thread_main'. `cache_lock' is held
677 * while this is in progress.
679 static gboolean tree_callback_flush (gpointer key, gpointer value, /* {{{ */
683 callback_flush_data_t *cfd;
685 ci = (cache_item_t *) value;
686 cfd = (callback_flush_data_t *) data;
688 if (ci->flags & CI_FLAGS_IN_QUEUE)
691 if ((ci->last_flush_time <= cfd->abs_timeout)
692 && (ci->values_num > 0))
694 enqueue_cache_item (ci, TAIL);
696 else if ((do_shutdown != 0)
697 && (ci->values_num > 0))
699 enqueue_cache_item (ci, TAIL);
701 else if (((cfd->now - ci->last_flush_time) >= config_flush_interval)
702 && (ci->values_num <= 0))
706 temp = (char **) realloc (cfd->keys,
707 sizeof (char *) * (cfd->keys_num + 1));
710 RRDD_LOG (LOG_ERR, "tree_callback_flush: realloc failed.");
714 /* Make really sure this points to the _same_ place */
715 assert ((char *) key == ci->file);
716 cfd->keys[cfd->keys_num] = (char *) key;
721 } /* }}} gboolean tree_callback_flush */
723 static int flush_old_values (int max_age)
725 callback_flush_data_t cfd;
728 memset (&cfd, 0, sizeof (cfd));
729 /* Pass the current time as user data so that we don't need to call
730 * `time' for each node. */
731 cfd.now = time (NULL);
736 cfd.abs_timeout = cfd.now - max_age;
738 cfd.abs_timeout = cfd.now + 2*config_write_jitter + 1;
740 /* `tree_callback_flush' will return the keys of all values that haven't
741 * been touched in the last `config_flush_interval' seconds in `cfd'.
742 * The char*'s in this array point to the same memory as ci->file, so we
743 * don't need to free them separately. */
744 g_tree_foreach (cache_tree, tree_callback_flush, (gpointer) &cfd);
746 for (k = 0; k < cfd.keys_num; k++)
748 /* should never fail, since we have held the cache_lock
750 assert( g_tree_remove(cache_tree, cfd.keys[k]) == TRUE );
753 if (cfd.keys != NULL)
760 } /* int flush_old_values */
762 static void *queue_thread_main (void *args __attribute__((unused))) /* {{{ */
765 struct timespec next_flush;
766 int final_flush = 0; /* make sure we only flush once on shutdown */
768 gettimeofday (&now, NULL);
769 next_flush.tv_sec = now.tv_sec + config_flush_interval;
770 next_flush.tv_nsec = 1000 * now.tv_usec;
772 pthread_mutex_lock (&cache_lock);
773 while ((do_shutdown == 0) || (cache_queue_head != NULL))
782 /* First, check if it's time to do the cache flush. */
783 gettimeofday (&now, NULL);
784 if ((now.tv_sec > next_flush.tv_sec)
785 || ((now.tv_sec == next_flush.tv_sec)
786 && ((1000 * now.tv_usec) > next_flush.tv_nsec)))
788 /* Flush all values that haven't been written in the last
789 * `config_write_interval' seconds. */
790 flush_old_values (config_write_interval);
792 /* Determine the time of the next cache flush. */
794 now.tv_sec + next_flush.tv_sec % config_flush_interval;
796 /* unlock the cache while we rotate so we don't block incoming
797 * updates if the fsync() blocks on disk I/O */
798 pthread_mutex_unlock(&cache_lock);
800 pthread_mutex_lock(&cache_lock);
803 /* Now, check if there's something to store away. If not, wait until
804 * something comes in or it's time to do the cache flush. if we are
805 * shutting down, do not wait around. */
806 if (cache_queue_head == NULL && !do_shutdown)
808 status = pthread_cond_timedwait (&cache_cond, &cache_lock, &next_flush);
809 if ((status != 0) && (status != ETIMEDOUT))
811 RRDD_LOG (LOG_ERR, "queue_thread_main: "
812 "pthread_cond_timedwait returned %i.", status);
816 /* We're about to shut down */
817 if (do_shutdown != 0 && !final_flush++)
819 if (config_flush_at_shutdown)
820 flush_old_values (-1); /* flush everything */
825 /* Check if a value has arrived. This may be NULL if we timed out or there
826 * was an interrupt such as a signal. */
827 if (cache_queue_head == NULL)
830 ci = cache_queue_head;
832 /* copy the relevant parts */
833 file = strdup (ci->file);
836 RRDD_LOG (LOG_ERR, "queue_thread_main: strdup failed.");
840 assert(ci->values != NULL);
841 assert(ci->values_num > 0);
844 values_num = ci->values_num;
846 wipe_ci_values(ci, time(NULL));
847 remove_from_queue(ci);
849 pthread_mutex_lock (&stats_lock);
850 assert (stats_queue_length > 0);
851 stats_queue_length--;
852 pthread_mutex_unlock (&stats_lock);
854 pthread_mutex_unlock (&cache_lock);
857 status = rrd_update_r (file, NULL, values_num, (void *) values);
860 RRDD_LOG (LOG_NOTICE, "queue_thread_main: "
861 "rrd_update_r (%s) failed with status %i. (%s)",
862 file, status, rrd_get_error());
865 journal_write("wrote", file);
866 pthread_cond_broadcast(&ci->flushed);
868 for (i = 0; i < values_num; i++)
876 pthread_mutex_lock (&stats_lock);
877 stats_updates_written++;
878 stats_data_sets_written += values_num;
879 pthread_mutex_unlock (&stats_lock);
882 pthread_mutex_lock (&cache_lock);
884 /* We're about to shut down */
885 if (do_shutdown != 0 && !final_flush++)
887 if (config_flush_at_shutdown)
888 flush_old_values (-1); /* flush everything */
892 } /* while ((do_shutdown == 0) || (cache_queue_head != NULL)) */
893 pthread_mutex_unlock (&cache_lock);
895 if (config_flush_at_shutdown)
897 assert(cache_queue_head == NULL);
898 RRDD_LOG(LOG_INFO, "clean shutdown; all RRDs flushed");
904 } /* }}} void *queue_thread_main */
906 static int buffer_get_field (char **buffer_ret, /* {{{ */
907 size_t *buffer_size_ret, char **field_ret)
916 buffer = *buffer_ret;
918 buffer_size = *buffer_size_ret;
922 if (buffer_size <= 0)
925 /* This is ensured by `handle_request'. */
926 assert (buffer[buffer_size - 1] == '\0');
929 while (buffer_pos < buffer_size)
931 /* Check for end-of-field or end-of-buffer */
932 if (buffer[buffer_pos] == ' ' || buffer[buffer_pos] == '\0')
934 field[field_size] = 0;
940 /* Handle escaped characters. */
941 else if (buffer[buffer_pos] == '\\')
943 if (buffer_pos >= (buffer_size - 1))
946 field[field_size] = buffer[buffer_pos];
950 /* Normal operation */
953 field[field_size] = buffer[buffer_pos];
957 } /* while (buffer_pos < buffer_size) */
962 *buffer_ret = buffer + buffer_pos;
963 *buffer_size_ret = buffer_size - buffer_pos;
967 } /* }}} int buffer_get_field */
969 /* if we're restricting writes to the base directory,
970 * check whether the file falls within the dir
971 * returns 1 if OK, otherwise 0
973 static int check_file_access (const char *file, listen_socket_t *sock) /* {{{ */
975 assert(file != NULL);
977 if (!config_write_base_only
978 || sock == NULL /* journal replay */
979 || config_base_dir == NULL)
982 if (strstr(file, "../") != NULL) goto err;
984 /* relative paths without "../" are ok */
985 if (*file != '/') return 1;
987 /* file must be of the format base + "/" + <1+ char filename> */
988 if (strlen(file) < _config_base_dir_len + 2) goto err;
989 if (strncmp(file, config_base_dir, _config_base_dir_len) != 0) goto err;
990 if (*(file + _config_base_dir_len) != '/') goto err;
995 if (sock != NULL && sock->fd >= 0)
996 send_response(sock, RESP_ERR, "%s\n", rrd_strerror(EACCES));
999 } /* }}} static int check_file_access */
1001 /* when using a base dir, convert relative paths to absolute paths.
1002 * if necessary, modifies the "filename" pointer to point
1003 * to the new path created in "tmp". "tmp" is provided
1004 * by the caller and sizeof(tmp) must be >= PATH_MAX.
1006 * this allows us to optimize for the expected case (absolute path)
1009 static void get_abs_path(char **filename, char *tmp)
1011 assert(tmp != NULL);
1012 assert(filename != NULL && *filename != NULL);
1014 if (config_base_dir == NULL || **filename == '/')
1017 snprintf(tmp, PATH_MAX, "%s/%s", config_base_dir, *filename);
1019 } /* }}} static int get_abs_path */
1021 /* returns 1 if we have the required privilege level,
1022 * otherwise issue an error to the user on sock */
1023 static int has_privilege (listen_socket_t *sock, /* {{{ */
1024 socket_privilege priv)
1026 if (sock == NULL) /* journal replay */
1029 if (sock->privilege >= priv)
1032 return send_response(sock, RESP_ERR, "%s\n", rrd_strerror(EACCES));
1033 } /* }}} static int has_privilege */
1035 static int flush_file (const char *filename) /* {{{ */
1039 pthread_mutex_lock (&cache_lock);
1041 ci = (cache_item_t *) g_tree_lookup (cache_tree, filename);
1044 pthread_mutex_unlock (&cache_lock);
1048 if (ci->values_num > 0)
1050 /* Enqueue at head */
1051 enqueue_cache_item (ci, HEAD);
1052 pthread_cond_wait(&ci->flushed, &cache_lock);
1055 /* DO NOT DO ANYTHING WITH ci HERE!! The entry
1056 * may have been purged during our cond_wait() */
1058 pthread_mutex_unlock(&cache_lock);
1061 } /* }}} int flush_file */
1063 static int handle_request_help (listen_socket_t *sock, /* {{{ */
1064 char *buffer, size_t buffer_size)
1070 char *help_help[2] =
1072 "Command overview\n"
1074 "HELP [<command>]\n"
1075 "FLUSH <filename>\n"
1077 "PENDING <filename>\n"
1078 "FORGET <filename>\n"
1079 "UPDATE <filename> <values> [<values> ...]\n"
1084 char *help_flush[2] =
1088 "Usage: FLUSH <filename>\n"
1090 "Adds the given filename to the head of the update queue and returns\n"
1091 "after is has been dequeued.\n"
1094 char *help_flushall[2] =
1096 "Help for FLUSHALL\n"
1100 "Triggers writing of all pending updates. Returns immediately.\n"
1103 char *help_pending[2] =
1105 "Help for PENDING\n"
1107 "Usage: PENDING <filename>\n"
1109 "Shows any 'pending' updates for a file, in order.\n"
1110 "The updates shown have not yet been written to the underlying RRD file.\n"
1113 char *help_forget[2] =
1117 "Usage: FORGET <filename>\n"
1119 "Removes the file completely from the cache.\n"
1120 "Any pending updates for the file will be lost.\n"
1123 char *help_update[2] =
1127 "Usage: UPDATE <filename> <values> [<values> ...]\n"
1129 "Adds the given file to the internal cache if it is not yet known and\n"
1130 "appends the given value(s) to the entry. See the rrdcached(1) manpage\n"
1133 "Each <values> has the following form:\n"
1134 " <values> = <time>:<value>[:<value>[...]]\n"
1135 "See the rrdupdate(1) manpage for details.\n"
1138 char *help_stats[2] =
1144 "Returns some performance counters, see the rrdcached(1) manpage for\n"
1145 "a description of the values.\n"
1148 char *help_batch[2] =
1152 "The 'BATCH' command permits the client to initiate a bulk load\n"
1153 " of commands to rrdcached.\n"
1158 " server: 0 Go ahead. End with dot '.' on its own line.\n"
1159 " client: command #1\n"
1160 " client: command #2\n"
1161 " client: ... and so on\n"
1163 " server: 2 errors\n"
1164 " server: 7 message for command #7\n"
1165 " server: 9 message for command #9\n"
1167 "For more information, consult the rrdcached(1) documentation.\n"
1170 status = buffer_get_field (&buffer, &buffer_size, &command);
1172 help_text = help_help;
1175 if (strcasecmp (command, "update") == 0)
1176 help_text = help_update;
1177 else if (strcasecmp (command, "flush") == 0)
1178 help_text = help_flush;
1179 else if (strcasecmp (command, "flushall") == 0)
1180 help_text = help_flushall;
1181 else if (strcasecmp (command, "pending") == 0)
1182 help_text = help_pending;
1183 else if (strcasecmp (command, "forget") == 0)
1184 help_text = help_forget;
1185 else if (strcasecmp (command, "stats") == 0)
1186 help_text = help_stats;
1187 else if (strcasecmp (command, "batch") == 0)
1188 help_text = help_batch;
1190 help_text = help_help;
1193 add_response_info(sock, help_text[1]);
1194 return send_response(sock, RESP_OK, help_text[0]);
1195 } /* }}} int handle_request_help */
1197 static int handle_request_stats (listen_socket_t *sock) /* {{{ */
1199 uint64_t copy_queue_length;
1200 uint64_t copy_updates_received;
1201 uint64_t copy_flush_received;
1202 uint64_t copy_updates_written;
1203 uint64_t copy_data_sets_written;
1204 uint64_t copy_journal_bytes;
1205 uint64_t copy_journal_rotate;
1207 uint64_t tree_nodes_number;
1208 uint64_t tree_depth;
1210 pthread_mutex_lock (&stats_lock);
1211 copy_queue_length = stats_queue_length;
1212 copy_updates_received = stats_updates_received;
1213 copy_flush_received = stats_flush_received;
1214 copy_updates_written = stats_updates_written;
1215 copy_data_sets_written = stats_data_sets_written;
1216 copy_journal_bytes = stats_journal_bytes;
1217 copy_journal_rotate = stats_journal_rotate;
1218 pthread_mutex_unlock (&stats_lock);
1220 pthread_mutex_lock (&cache_lock);
1221 tree_nodes_number = (uint64_t) g_tree_nnodes (cache_tree);
1222 tree_depth = (uint64_t) g_tree_height (cache_tree);
1223 pthread_mutex_unlock (&cache_lock);
1225 add_response_info(sock,
1226 "QueueLength: %"PRIu64"\n", copy_queue_length);
1227 add_response_info(sock,
1228 "UpdatesReceived: %"PRIu64"\n", copy_updates_received);
1229 add_response_info(sock,
1230 "FlushesReceived: %"PRIu64"\n", copy_flush_received);
1231 add_response_info(sock,
1232 "UpdatesWritten: %"PRIu64"\n", copy_updates_written);
1233 add_response_info(sock,
1234 "DataSetsWritten: %"PRIu64"\n", copy_data_sets_written);
1235 add_response_info(sock, "TreeNodesNumber: %"PRIu64"\n", tree_nodes_number);
1236 add_response_info(sock, "TreeDepth: %"PRIu64"\n", tree_depth);
1237 add_response_info(sock, "JournalBytes: %"PRIu64"\n", copy_journal_bytes);
1238 add_response_info(sock, "JournalRotate: %"PRIu64"\n", copy_journal_rotate);
1240 send_response(sock, RESP_OK, "Statistics follow\n");
1243 } /* }}} int handle_request_stats */
1245 static int handle_request_flush (listen_socket_t *sock, /* {{{ */
1246 char *buffer, size_t buffer_size)
1248 char *file, file_tmp[PATH_MAX];
1251 status = buffer_get_field (&buffer, &buffer_size, &file);
1254 return send_response(sock, RESP_ERR, "Usage: flush <filename>\n");
1258 pthread_mutex_lock(&stats_lock);
1259 stats_flush_received++;
1260 pthread_mutex_unlock(&stats_lock);
1262 get_abs_path(&file, file_tmp);
1263 if (!check_file_access(file, sock)) return 0;
1265 status = flush_file (file);
1267 return send_response(sock, RESP_OK, "Successfully flushed %s.\n", file);
1268 else if (status == ENOENT)
1270 /* no file in our tree; see whether it exists at all */
1271 struct stat statbuf;
1273 memset(&statbuf, 0, sizeof(statbuf));
1274 if (stat(file, &statbuf) == 0 && S_ISREG(statbuf.st_mode))
1275 return send_response(sock, RESP_OK, "Nothing to flush: %s.\n", file);
1277 return send_response(sock, RESP_ERR, "No such file: %s.\n", file);
1279 else if (status < 0)
1280 return send_response(sock, RESP_ERR, "Internal error.\n");
1282 return send_response(sock, RESP_ERR, "Failed with status %i.\n", status);
1287 } /* }}} int handle_request_flush */
1289 static int handle_request_flushall(listen_socket_t *sock) /* {{{ */
1293 status = has_privilege(sock, PRIV_HIGH);
1297 RRDD_LOG(LOG_DEBUG, "Received FLUSHALL");
1299 pthread_mutex_lock(&cache_lock);
1300 flush_old_values(-1);
1301 pthread_mutex_unlock(&cache_lock);
1303 return send_response(sock, RESP_OK, "Started flush.\n");
1304 } /* }}} static int handle_request_flushall */
1306 static int handle_request_pending(listen_socket_t *sock, /* {{{ */
1307 char *buffer, size_t buffer_size)
1310 char *file, file_tmp[PATH_MAX];
1313 status = buffer_get_field(&buffer, &buffer_size, &file);
1315 return send_response(sock, RESP_ERR,
1316 "Usage: PENDING <filename>\n");
1318 status = has_privilege(sock, PRIV_HIGH);
1322 get_abs_path(&file, file_tmp);
1324 pthread_mutex_lock(&cache_lock);
1325 ci = g_tree_lookup(cache_tree, file);
1328 pthread_mutex_unlock(&cache_lock);
1329 return send_response(sock, RESP_ERR, "%s\n", rrd_strerror(ENOENT));
1332 for (int i=0; i < ci->values_num; i++)
1333 add_response_info(sock, "%s\n", ci->values[i]);
1335 pthread_mutex_unlock(&cache_lock);
1336 return send_response(sock, RESP_OK, "updates pending\n");
1337 } /* }}} static int handle_request_pending */
1339 static int handle_request_forget(listen_socket_t *sock, /* {{{ */
1340 char *buffer, size_t buffer_size)
1344 char *file, file_tmp[PATH_MAX];
1346 status = buffer_get_field(&buffer, &buffer_size, &file);
1348 return send_response(sock, RESP_ERR,
1349 "Usage: FORGET <filename>\n");
1351 status = has_privilege(sock, PRIV_HIGH);
1355 get_abs_path(&file, file_tmp);
1356 if (!check_file_access(file, sock)) return 0;
1358 pthread_mutex_lock(&cache_lock);
1359 found = g_tree_remove(cache_tree, file);
1360 pthread_mutex_unlock(&cache_lock);
1365 journal_write("forget", file);
1367 return send_response(sock, RESP_OK, "Gone!\n");
1370 return send_response(sock, RESP_ERR, "%s\n", rrd_strerror(ENOENT));
1374 } /* }}} static int handle_request_forget */
1376 static int handle_request_update (listen_socket_t *sock, /* {{{ */
1378 char *buffer, size_t buffer_size)
1380 char *file, file_tmp[PATH_MAX];
1383 char orig_buf[CMD_MAX];
1387 status = has_privilege(sock, PRIV_HIGH);
1391 /* save it for the journal later */
1392 strncpy(orig_buf, buffer, sizeof(orig_buf)-1);
1394 status = buffer_get_field (&buffer, &buffer_size, &file);
1396 return send_response(sock, RESP_ERR,
1397 "Usage: UPDATE <filename> <values> [<values> ...]\n");
1399 pthread_mutex_lock(&stats_lock);
1400 stats_updates_received++;
1401 pthread_mutex_unlock(&stats_lock);
1403 get_abs_path(&file, file_tmp);
1404 if (!check_file_access(file, sock)) return 0;
1406 pthread_mutex_lock (&cache_lock);
1407 ci = g_tree_lookup (cache_tree, file);
1409 if (ci == NULL) /* {{{ */
1411 struct stat statbuf;
1413 /* don't hold the lock while we setup; stat(2) might block */
1414 pthread_mutex_unlock(&cache_lock);
1416 memset (&statbuf, 0, sizeof (statbuf));
1417 status = stat (file, &statbuf);
1420 RRDD_LOG (LOG_NOTICE, "handle_request_update: stat (%s) failed.", file);
1423 if (status == ENOENT)
1424 return send_response(sock, RESP_ERR, "No such file: %s\n", file);
1426 return send_response(sock, RESP_ERR,
1427 "stat failed with error %i.\n", status);
1429 if (!S_ISREG (statbuf.st_mode))
1430 return send_response(sock, RESP_ERR, "Not a regular file: %s\n", file);
1432 if (access(file, R_OK|W_OK) != 0)
1433 return send_response(sock, RESP_ERR, "Cannot read/write %s: %s\n",
1434 file, rrd_strerror(errno));
1436 ci = (cache_item_t *) malloc (sizeof (cache_item_t));
1439 RRDD_LOG (LOG_ERR, "handle_request_update: malloc failed.");
1441 return send_response(sock, RESP_ERR, "malloc failed.\n");
1443 memset (ci, 0, sizeof (cache_item_t));
1445 ci->file = strdup (file);
1446 if (ci->file == NULL)
1449 RRDD_LOG (LOG_ERR, "handle_request_update: strdup failed.");
1451 return send_response(sock, RESP_ERR, "strdup failed.\n");
1454 wipe_ci_values(ci, now);
1455 ci->flags = CI_FLAGS_IN_TREE;
1456 pthread_cond_init(&ci->flushed, NULL);
1458 pthread_mutex_lock(&cache_lock);
1459 g_tree_replace (cache_tree, (void *) ci->file, (void *) ci);
1461 assert (ci != NULL);
1463 /* don't re-write updates in replay mode */
1465 journal_write("update", orig_buf);
1467 while (buffer_size > 0)
1474 status = buffer_get_field (&buffer, &buffer_size, &value);
1477 RRDD_LOG (LOG_INFO, "handle_request_update: Error reading field.");
1481 /* make sure update time is always moving forward */
1482 stamp = strtol(value, &eostamp, 10);
1483 if (eostamp == value || eostamp == NULL || *eostamp != ':')
1485 pthread_mutex_unlock(&cache_lock);
1486 return send_response(sock, RESP_ERR,
1487 "Cannot find timestamp in '%s'!\n", value);
1489 else if (stamp <= ci->last_update_stamp)
1491 pthread_mutex_unlock(&cache_lock);
1492 return send_response(sock, RESP_ERR,
1493 "illegal attempt to update using time %ld when last"
1494 " update time is %ld (minimum one second step)\n",
1495 stamp, ci->last_update_stamp);
1498 ci->last_update_stamp = stamp;
1500 temp = (char **) realloc (ci->values,
1501 sizeof (char *) * (ci->values_num + 1));
1504 RRDD_LOG (LOG_ERR, "handle_request_update: realloc failed.");
1509 ci->values[ci->values_num] = strdup (value);
1510 if (ci->values[ci->values_num] == NULL)
1512 RRDD_LOG (LOG_ERR, "handle_request_update: strdup failed.");
1520 if (((now - ci->last_flush_time) >= config_write_interval)
1521 && ((ci->flags & CI_FLAGS_IN_QUEUE) == 0)
1522 && (ci->values_num > 0))
1524 enqueue_cache_item (ci, TAIL);
1527 pthread_mutex_unlock (&cache_lock);
1530 return send_response(sock, RESP_ERR, "No values updated.\n");
1532 return send_response(sock, RESP_OK,
1533 "errors, enqueued %i value(s).\n", values_num);
1538 } /* }}} int handle_request_update */
1540 /* we came across a "WROTE" entry during journal replay.
1541 * throw away any values that we have accumulated for this file
1543 static int handle_request_wrote (const char *buffer, time_t now) /* {{{ */
1547 const char *file = buffer;
1549 pthread_mutex_lock(&cache_lock);
1551 ci = g_tree_lookup(cache_tree, file);
1554 pthread_mutex_unlock(&cache_lock);
1560 for (i=0; i < ci->values_num; i++)
1561 free(ci->values[i]);
1566 wipe_ci_values(ci, now);
1567 remove_from_queue(ci);
1569 pthread_mutex_unlock(&cache_lock);
1571 } /* }}} int handle_request_wrote */
1573 /* start "BATCH" processing */
1574 static int batch_start (listen_socket_t *sock) /* {{{ */
1577 if (sock->batch_start)
1578 return send_response(sock, RESP_ERR, "Already in BATCH\n");
1580 status = send_response(sock, RESP_OK,
1581 "Go ahead. End with dot '.' on its own line.\n");
1582 sock->batch_start = time(NULL);
1583 sock->batch_cmd = 0;
1586 } /* }}} static int batch_start */
1588 /* finish "BATCH" processing and return results to the client */
1589 static int batch_done (listen_socket_t *sock) /* {{{ */
1591 assert(sock->batch_start);
1592 sock->batch_start = 0;
1593 sock->batch_cmd = 0;
1594 return send_response(sock, RESP_OK, "errors\n");
1595 } /* }}} static int batch_done */
1597 /* if sock==NULL, we are in journal replay mode */
1598 static int handle_request (listen_socket_t *sock, /* {{{ */
1600 char *buffer, size_t buffer_size)
1606 assert (buffer[buffer_size - 1] == '\0');
1608 buffer_ptr = buffer;
1610 status = buffer_get_field (&buffer_ptr, &buffer_size, &command);
1613 RRDD_LOG (LOG_INFO, "handle_request: Unable parse command.");
1617 if (sock != NULL && sock->batch_start)
1620 if (strcasecmp (command, "update") == 0)
1621 return (handle_request_update (sock, now, buffer_ptr, buffer_size));
1622 else if (strcasecmp (command, "wrote") == 0 && sock == NULL)
1624 /* this is only valid in replay mode */
1625 return (handle_request_wrote (buffer_ptr, now));
1627 else if (strcasecmp (command, "flush") == 0)
1628 return (handle_request_flush (sock, buffer_ptr, buffer_size));
1629 else if (strcasecmp (command, "flushall") == 0)
1630 return (handle_request_flushall(sock));
1631 else if (strcasecmp (command, "pending") == 0)
1632 return (handle_request_pending(sock, buffer_ptr, buffer_size));
1633 else if (strcasecmp (command, "forget") == 0)
1634 return (handle_request_forget(sock, buffer_ptr, buffer_size));
1635 else if (strcasecmp (command, "stats") == 0)
1636 return (handle_request_stats (sock));
1637 else if (strcasecmp (command, "help") == 0)
1638 return (handle_request_help (sock, buffer_ptr, buffer_size));
1639 else if (strcasecmp (command, "batch") == 0 && sock != NULL)
1640 return batch_start(sock);
1641 else if (strcasecmp (command, ".") == 0 && sock != NULL && sock->batch_start)
1642 return batch_done(sock);
1643 else if (strcasecmp (command, "quit") == 0)
1646 return send_response(sock, RESP_ERR, "Unknown command: %s\n", command);
1650 } /* }}} int handle_request */
1652 /* MUST NOT hold journal_lock before calling this */
1653 static void journal_rotate(void) /* {{{ */
1655 FILE *old_fh = NULL;
1658 if (journal_cur == NULL || journal_old == NULL)
1661 pthread_mutex_lock(&journal_lock);
1663 /* we rotate this way (rename before close) so that the we can release
1664 * the journal lock as fast as possible. Journal writes to the new
1665 * journal can proceed immediately after the new file is opened. The
1666 * fclose can then block without affecting new updates.
1668 if (journal_fh != NULL)
1670 old_fh = journal_fh;
1672 rename(journal_cur, journal_old);
1673 ++stats_journal_rotate;
1676 new_fd = open(journal_cur, O_WRONLY|O_CREAT|O_APPEND,
1677 S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH);
1680 journal_fh = fdopen(new_fd, "a");
1681 if (journal_fh == NULL)
1685 pthread_mutex_unlock(&journal_lock);
1690 if (journal_fh == NULL)
1693 "JOURNALING DISABLED: Cannot open journal file '%s' : (%s)",
1694 journal_cur, rrd_strerror(errno));
1697 "JOURNALING DISABLED: All values will be flushed at shutdown");
1698 config_flush_at_shutdown = 1;
1701 } /* }}} static void journal_rotate */
1703 static void journal_done(void) /* {{{ */
1705 if (journal_cur == NULL)
1708 pthread_mutex_lock(&journal_lock);
1709 if (journal_fh != NULL)
1715 if (config_flush_at_shutdown)
1717 RRDD_LOG(LOG_INFO, "removing journals");
1718 unlink(journal_old);
1719 unlink(journal_cur);
1723 RRDD_LOG(LOG_INFO, "expedited shutdown; "
1724 "journals will be used at next startup");
1727 pthread_mutex_unlock(&journal_lock);
1729 } /* }}} static void journal_done */
1731 static int journal_write(char *cmd, char *args) /* {{{ */
1735 if (journal_fh == NULL)
1738 pthread_mutex_lock(&journal_lock);
1739 chars = fprintf(journal_fh, "%s %s\n", cmd, args);
1740 pthread_mutex_unlock(&journal_lock);
1744 pthread_mutex_lock(&stats_lock);
1745 stats_journal_bytes += chars;
1746 pthread_mutex_unlock(&stats_lock);
1750 } /* }}} static int journal_write */
1752 static int journal_replay (const char *file) /* {{{ */
1758 char entry[CMD_MAX];
1761 if (file == NULL) return 0;
1766 struct stat statbuf;
1768 memset(&statbuf, 0, sizeof(statbuf));
1769 if (stat(file, &statbuf) != 0)
1771 if (errno == ENOENT)
1774 reason = "stat error";
1777 else if (!S_ISREG(statbuf.st_mode))
1779 reason = "not a regular file";
1782 if (statbuf.st_uid != daemon_uid)
1784 reason = "not owned by daemon user";
1787 if (statbuf.st_mode & (S_IWGRP|S_IWOTH))
1789 reason = "must not be user/group writable";
1795 RRDD_LOG(LOG_ERR, "journal_replay: %s : %s (%s)",
1796 file, rrd_strerror(status), reason);
1801 fh = fopen(file, "r");
1804 if (errno != ENOENT)
1805 RRDD_LOG(LOG_ERR, "journal_replay: cannot open journal file: '%s' (%s)",
1806 file, rrd_strerror(errno));
1810 RRDD_LOG(LOG_NOTICE, "replaying from journal: %s", file);
1819 if (fgets(entry, sizeof(entry), fh) == NULL)
1821 entry_len = strlen(entry);
1823 /* check \n termination in case journal writing crashed mid-line */
1826 else if (entry[entry_len - 1] != '\n')
1828 RRDD_LOG(LOG_NOTICE, "Malformed journal entry at line %"PRIu64, line);
1833 entry[entry_len - 1] = '\0';
1835 if (handle_request(NULL, now, entry, entry_len) == 0)
1843 RRDD_LOG(LOG_INFO, "Replayed %d entries (%d failures)",
1844 entry_cnt, fail_cnt);
1846 return entry_cnt > 0 ? 1 : 0;
1847 } /* }}} static int journal_replay */
1849 static void journal_init(void) /* {{{ */
1851 int had_journal = 0;
1853 if (journal_cur == NULL) return;
1855 pthread_mutex_lock(&journal_lock);
1857 RRDD_LOG(LOG_INFO, "checking for journal files");
1859 had_journal += journal_replay(journal_old);
1860 had_journal += journal_replay(journal_cur);
1862 /* it must have been a crash. start a flush */
1863 if (had_journal && config_flush_at_shutdown)
1864 flush_old_values(-1);
1866 pthread_mutex_unlock(&journal_lock);
1869 RRDD_LOG(LOG_INFO, "journal processing complete");
1871 } /* }}} static void journal_init */
1873 static void free_listen_socket(listen_socket_t *sock) /* {{{ */
1875 assert(sock != NULL);
1877 free(sock->rbuf); sock->rbuf = NULL;
1878 free(sock->wbuf); sock->wbuf = NULL;
1880 } /* }}} void free_listen_socket */
1882 static void close_connection(listen_socket_t *sock) /* {{{ */
1890 free_listen_socket(sock);
1892 } /* }}} void close_connection */
1894 static void *connection_thread_main (void *args) /* {{{ */
1896 listen_socket_t *sock;
1900 sock = (listen_socket_t *) args;
1903 /* init read buffers */
1904 sock->next_read = sock->next_cmd = 0;
1905 sock->rbuf = malloc(RBUF_SIZE);
1906 if (sock->rbuf == NULL)
1908 RRDD_LOG(LOG_ERR, "connection_thread_main: cannot malloc read buffer");
1909 close_connection(sock);
1913 pthread_mutex_lock (&connection_threads_lock);
1917 temp = (pthread_t *) realloc (connection_threads,
1918 sizeof (pthread_t) * (connection_threads_num + 1));
1921 RRDD_LOG (LOG_ERR, "connection_thread_main: realloc(++) failed.");
1925 connection_threads = temp;
1926 connection_threads[connection_threads_num] = pthread_self ();
1927 connection_threads_num++;
1930 pthread_mutex_unlock (&connection_threads_lock);
1932 while (do_shutdown == 0)
1939 struct pollfd pollfd;
1943 pollfd.events = POLLIN | POLLPRI;
1946 status = poll (&pollfd, 1, /* timeout = */ 500);
1949 else if (status == 0) /* timeout */
1951 else if (status < 0) /* error */
1954 if (status != EINTR)
1955 RRDD_LOG (LOG_ERR, "connection_thread_main: poll(2) failed.");
1959 if ((pollfd.revents & POLLHUP) != 0) /* normal shutdown */
1961 else if ((pollfd.revents & (POLLIN | POLLPRI)) == 0)
1963 RRDD_LOG (LOG_WARNING, "connection_thread_main: "
1964 "poll(2) returned something unexpected: %#04hx",
1969 rbytes = read(fd, sock->rbuf + sock->next_read,
1970 RBUF_SIZE - sock->next_read);
1973 RRDD_LOG(LOG_ERR, "connection_thread_main: read() failed.");
1976 else if (rbytes == 0)
1979 sock->next_read += rbytes;
1981 if (sock->batch_start)
1982 now = sock->batch_start;
1986 while ((cmd = next_cmd(sock, &cmd_len)) != NULL)
1988 status = handle_request (sock, now, cmd, cmd_len+1);
1995 close_connection(sock);
1997 /* Remove this thread from the connection threads list */
1998 pthread_mutex_lock (&connection_threads_lock);
2003 /* Find out own index in the array */
2004 self = pthread_self ();
2005 for (i = 0; i < connection_threads_num; i++)
2006 if (pthread_equal (connection_threads[i], self) != 0)
2008 assert (i < connection_threads_num);
2010 /* Move the trailing threads forward. */
2011 if (i < (connection_threads_num - 1))
2013 memmove (connection_threads + i,
2014 connection_threads + i + 1,
2015 sizeof (pthread_t) * (connection_threads_num - i - 1));
2018 connection_threads_num--;
2020 temp = realloc(connection_threads,
2021 sizeof(*connection_threads) * connection_threads_num);
2022 if (connection_threads_num > 0 && temp == NULL)
2023 RRDD_LOG(LOG_ERR, "connection_thread_main: realloc(--) failed.");
2025 connection_threads = temp;
2027 pthread_mutex_unlock (&connection_threads_lock);
2030 } /* }}} void *connection_thread_main */
2032 static int open_listen_socket_unix (const listen_socket_t *sock) /* {{{ */
2035 struct sockaddr_un sa;
2036 listen_socket_t *temp;
2041 if (strncmp(path, "unix:", strlen("unix:")) == 0)
2042 path += strlen("unix:");
2044 temp = (listen_socket_t *) realloc (listen_fds,
2045 sizeof (listen_fds[0]) * (listen_fds_num + 1));
2048 fprintf (stderr, "rrdcached: open_listen_socket_unix: realloc failed.\n");
2052 memcpy (listen_fds + listen_fds_num, sock, sizeof (listen_fds[0]));
2054 fd = socket (PF_UNIX, SOCK_STREAM, /* protocol = */ 0);
2057 fprintf (stderr, "rrdcached: unix socket(2) failed: %s\n",
2058 rrd_strerror(errno));
2062 memset (&sa, 0, sizeof (sa));
2063 sa.sun_family = AF_UNIX;
2064 strncpy (sa.sun_path, path, sizeof (sa.sun_path) - 1);
2066 /* if we've gotten this far, we own the pid file. any daemon started
2067 * with the same args must not be alive. therefore, ensure that we can
2068 * create the socket...
2072 status = bind (fd, (struct sockaddr *) &sa, sizeof (sa));
2075 fprintf (stderr, "rrdcached: bind(%s) failed: %s.\n",
2076 path, rrd_strerror(errno));
2081 status = listen (fd, /* backlog = */ 10);
2084 fprintf (stderr, "rrdcached: listen(%s) failed: %s.\n",
2085 path, rrd_strerror(errno));
2091 listen_fds[listen_fds_num].fd = fd;
2092 listen_fds[listen_fds_num].family = PF_UNIX;
2093 strncpy(listen_fds[listen_fds_num].addr, path,
2094 sizeof (listen_fds[listen_fds_num].addr) - 1);
2098 } /* }}} int open_listen_socket_unix */
2100 static int open_listen_socket_network(const listen_socket_t *sock) /* {{{ */
2102 struct addrinfo ai_hints;
2103 struct addrinfo *ai_res;
2104 struct addrinfo *ai_ptr;
2105 char addr_copy[NI_MAXHOST];
2110 strncpy (addr_copy, sock->addr, sizeof (addr_copy));
2111 addr_copy[sizeof (addr_copy) - 1] = 0;
2114 memset (&ai_hints, 0, sizeof (ai_hints));
2115 ai_hints.ai_flags = 0;
2116 #ifdef AI_ADDRCONFIG
2117 ai_hints.ai_flags |= AI_ADDRCONFIG;
2119 ai_hints.ai_family = AF_UNSPEC;
2120 ai_hints.ai_socktype = SOCK_STREAM;
2123 if (*addr == '[') /* IPv6+port format */
2125 /* `addr' is something like "[2001:780:104:2:211:24ff:feab:26f8]:12345" */
2128 port = strchr (addr, ']');
2131 fprintf (stderr, "rrdcached: Malformed address: %s\n", sock->addr);
2139 else if (*port == 0)
2143 fprintf (stderr, "rrdcached: Garbage after address: %s\n", port);
2146 } /* if (*addr = ']') */
2147 else if (strchr (addr, '.') != NULL) /* Hostname or IPv4 */
2149 port = rindex(addr, ':');
2157 status = getaddrinfo (addr,
2158 port == NULL ? RRDCACHED_DEFAULT_PORT : port,
2159 &ai_hints, &ai_res);
2162 fprintf (stderr, "rrdcached: getaddrinfo(%s) failed: %s\n",
2163 addr, gai_strerror (status));
2167 for (ai_ptr = ai_res; ai_ptr != NULL; ai_ptr = ai_ptr->ai_next)
2170 listen_socket_t *temp;
2173 temp = (listen_socket_t *) realloc (listen_fds,
2174 sizeof (listen_fds[0]) * (listen_fds_num + 1));
2178 "rrdcached: open_listen_socket_network: realloc failed.\n");
2182 memcpy (listen_fds + listen_fds_num, sock, sizeof (listen_fds[0]));
2184 fd = socket (ai_ptr->ai_family, ai_ptr->ai_socktype, ai_ptr->ai_protocol);
2187 fprintf (stderr, "rrdcached: network socket(2) failed: %s.\n",
2188 rrd_strerror(errno));
2192 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
2194 status = bind (fd, ai_ptr->ai_addr, ai_ptr->ai_addrlen);
2197 fprintf (stderr, "rrdcached: bind(%s) failed: %s.\n",
2198 sock->addr, rrd_strerror(errno));
2203 status = listen (fd, /* backlog = */ 10);
2206 fprintf (stderr, "rrdcached: listen(%s) failed: %s\n.",
2207 sock->addr, rrd_strerror(errno));
2209 freeaddrinfo(ai_res);
2213 listen_fds[listen_fds_num].fd = fd;
2214 listen_fds[listen_fds_num].family = ai_ptr->ai_family;
2216 } /* for (ai_ptr) */
2218 freeaddrinfo(ai_res);
2220 } /* }}} static int open_listen_socket_network */
2222 static int open_listen_socket (const listen_socket_t *sock) /* {{{ */
2224 assert(sock != NULL);
2225 assert(sock->addr != NULL);
2227 if (strncmp ("unix:", sock->addr, strlen ("unix:")) == 0
2228 || sock->addr[0] == '/')
2229 return (open_listen_socket_unix(sock));
2231 return (open_listen_socket_network(sock));
2232 } /* }}} int open_listen_socket */
2234 static int close_listen_sockets (void) /* {{{ */
2238 for (i = 0; i < listen_fds_num; i++)
2240 close (listen_fds[i].fd);
2242 if (listen_fds[i].family == PF_UNIX)
2243 unlink(listen_fds[i].addr);
2251 } /* }}} int close_listen_sockets */
2253 static void *listen_thread_main (void *args __attribute__((unused))) /* {{{ */
2255 struct pollfd *pollfds;
2260 if (listen_fds_num < 1)
2262 RRDD_LOG(LOG_ERR, "listen_thread_main: no listen_fds !");
2266 pollfds_num = listen_fds_num;
2267 pollfds = (struct pollfd *) malloc (sizeof (*pollfds) * pollfds_num);
2268 if (pollfds == NULL)
2270 RRDD_LOG (LOG_ERR, "listen_thread_main: malloc failed.");
2273 memset (pollfds, 0, sizeof (*pollfds) * pollfds_num);
2275 RRDD_LOG(LOG_INFO, "listening for connections");
2277 while (do_shutdown == 0)
2279 for (i = 0; i < pollfds_num; i++)
2281 pollfds[i].fd = listen_fds[i].fd;
2282 pollfds[i].events = POLLIN | POLLPRI;
2283 pollfds[i].revents = 0;
2286 status = poll (pollfds, pollfds_num, /* timeout = */ 1000);
2289 else if (status == 0) /* timeout */
2291 else if (status < 0) /* error */
2294 if (status != EINTR)
2296 RRDD_LOG (LOG_ERR, "listen_thread_main: poll(2) failed.");
2301 for (i = 0; i < pollfds_num; i++)
2303 listen_socket_t *client_sock;
2304 struct sockaddr_storage client_sa;
2305 socklen_t client_sa_size;
2307 pthread_attr_t attr;
2309 if (pollfds[i].revents == 0)
2312 if ((pollfds[i].revents & (POLLIN | POLLPRI)) == 0)
2314 RRDD_LOG (LOG_ERR, "listen_thread_main: "
2315 "poll(2) returned something unexpected for listen FD #%i.",
2320 client_sock = (listen_socket_t *) malloc (sizeof (listen_socket_t));
2321 if (client_sock == NULL)
2323 RRDD_LOG (LOG_ERR, "listen_thread_main: malloc failed.");
2326 memcpy(client_sock, &listen_fds[i], sizeof(listen_fds[0]));
2328 client_sa_size = sizeof (client_sa);
2329 client_sock->fd = accept (pollfds[i].fd,
2330 (struct sockaddr *) &client_sa, &client_sa_size);
2331 if (client_sock->fd < 0)
2333 RRDD_LOG (LOG_ERR, "listen_thread_main: accept(2) failed.");
2338 pthread_attr_init (&attr);
2339 pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);
2341 status = pthread_create (&tid, &attr, connection_thread_main,
2345 RRDD_LOG (LOG_ERR, "listen_thread_main: pthread_create failed.");
2346 close_connection(client_sock);
2349 } /* for (pollfds_num) */
2350 } /* while (do_shutdown == 0) */
2352 RRDD_LOG(LOG_INFO, "starting shutdown");
2354 close_listen_sockets ();
2356 pthread_mutex_lock (&connection_threads_lock);
2357 while (connection_threads_num > 0)
2361 wait_for = connection_threads[0];
2363 pthread_mutex_unlock (&connection_threads_lock);
2364 pthread_join (wait_for, /* retval = */ NULL);
2365 pthread_mutex_lock (&connection_threads_lock);
2367 pthread_mutex_unlock (&connection_threads_lock);
2372 } /* }}} void *listen_thread_main */
2374 static int daemonize (void) /* {{{ */
2379 daemon_uid = geteuid();
2381 pid_fd = open_pidfile("create", O_CREAT|O_EXCL|O_WRONLY);
2383 pid_fd = check_pidfile();
2387 /* open all the listen sockets */
2388 if (config_listen_address_list_len > 0)
2390 for (int i = 0; i < config_listen_address_list_len; i++)
2392 open_listen_socket (config_listen_address_list[i]);
2393 free_listen_socket (config_listen_address_list[i]);
2396 free(config_listen_address_list);
2400 listen_socket_t sock;
2401 memset(&sock, 0, sizeof(sock));
2402 strncpy(sock.addr, RRDCACHED_DEFAULT_ADDRESS, sizeof(sock.addr));
2403 open_listen_socket (&sock);
2406 if (listen_fds_num < 1)
2408 fprintf (stderr, "rrdcached: FATAL: cannot open any listen sockets\n");
2412 if (!stay_foreground)
2419 fprintf (stderr, "daemonize: fork(2) failed.\n");
2425 /* Become session leader */
2428 /* Open the first three file descriptors to /dev/null */
2433 open ("/dev/null", O_RDWR);
2436 } /* if (!stay_foreground) */
2438 /* Change into the /tmp directory. */
2439 base_dir = (config_base_dir != NULL)
2443 if (chdir (base_dir) != 0)
2445 fprintf (stderr, "daemonize: chdir (%s) failed.\n", base_dir);
2449 install_signal_handlers();
2451 openlog ("rrdcached", LOG_PID, LOG_DAEMON);
2452 RRDD_LOG(LOG_INFO, "starting up");
2454 cache_tree = g_tree_new_full ((GCompareDataFunc) strcmp, NULL, NULL,
2455 (GDestroyNotify) free_cache_item);
2456 if (cache_tree == NULL)
2458 RRDD_LOG (LOG_ERR, "daemonize: g_tree_new failed.");
2462 return write_pidfile (pid_fd);
2467 } /* }}} int daemonize */
2469 static int cleanup (void) /* {{{ */
2473 pthread_cond_signal (&cache_cond);
2474 pthread_join (queue_thread, /* return = */ NULL);
2478 free(config_base_dir);
2479 free(config_pid_file);
2483 pthread_mutex_lock(&cache_lock);
2484 g_tree_destroy(cache_tree);
2486 RRDD_LOG(LOG_INFO, "goodbye");
2490 } /* }}} int cleanup */
2492 static int read_options (int argc, char **argv) /* {{{ */
2497 while ((option = getopt(argc, argv, "gl:L:f:w:b:Bz:p:j:h?F")) != -1)
2508 listen_socket_t **temp;
2509 listen_socket_t *new;
2511 new = malloc(sizeof(listen_socket_t));
2514 fprintf(stderr, "read_options: malloc failed.\n");
2517 memset(new, 0, sizeof(listen_socket_t));
2519 temp = (listen_socket_t **) realloc (config_listen_address_list,
2520 sizeof (listen_socket_t *) * (config_listen_address_list_len + 1));
2523 fprintf (stderr, "read_options: realloc failed.\n");
2526 config_listen_address_list = temp;
2528 strncpy(new->addr, optarg, sizeof(new->addr)-1);
2529 new->privilege = (option == 'l') ? PRIV_HIGH : PRIV_LOW;
2531 temp[config_listen_address_list_len] = new;
2532 config_listen_address_list_len++;
2540 temp = atoi (optarg);
2542 config_flush_interval = temp;
2545 fprintf (stderr, "Invalid flush interval: %s\n", optarg);
2555 temp = atoi (optarg);
2557 config_write_interval = temp;
2560 fprintf (stderr, "Invalid write interval: %s\n", optarg);
2570 temp = atoi(optarg);
2572 config_write_jitter = temp;
2575 fprintf (stderr, "Invalid write jitter: -z %s\n", optarg);
2583 config_write_base_only = 1;
2589 char base_realpath[PATH_MAX];
2591 if (config_base_dir != NULL)
2592 free (config_base_dir);
2593 config_base_dir = strdup (optarg);
2594 if (config_base_dir == NULL)
2596 fprintf (stderr, "read_options: strdup failed.\n");
2600 /* make sure that the base directory is not resolved via
2601 * symbolic links. this makes some performance-enhancing
2602 * assumptions possible (we don't have to resolve paths
2603 * that start with a "/")
2605 if (realpath(config_base_dir, base_realpath) == NULL)
2607 fprintf (stderr, "Invalid base directory '%s'.\n", config_base_dir);
2610 else if (strncmp(config_base_dir,
2611 base_realpath, sizeof(base_realpath)) != 0)
2614 "Base directory (-b) resolved via file system links!\n"
2615 "Please consult rrdcached '-b' documentation!\n"
2616 "Consider specifying the real directory (%s)\n",
2621 len = strlen (config_base_dir);
2622 while ((len > 0) && (config_base_dir[len - 1] == '/'))
2624 config_base_dir[len - 1] = 0;
2630 fprintf (stderr, "Invalid base directory: %s\n", optarg);
2634 _config_base_dir_len = len;
2640 if (config_pid_file != NULL)
2641 free (config_pid_file);
2642 config_pid_file = strdup (optarg);
2643 if (config_pid_file == NULL)
2645 fprintf (stderr, "read_options: strdup failed.\n");
2652 config_flush_at_shutdown = 1;
2657 struct stat statbuf;
2658 const char *dir = optarg;
2660 status = stat(dir, &statbuf);
2663 fprintf(stderr, "Cannot stat '%s' : %s\n", dir, rrd_strerror(errno));
2667 if (!S_ISDIR(statbuf.st_mode)
2668 || access(dir, R_OK|W_OK|X_OK) != 0)
2670 fprintf(stderr, "Must specify a writable directory with -j! (%s)\n",
2671 errno ? rrd_strerror(errno) : "");
2675 journal_cur = malloc(PATH_MAX + 1);
2676 journal_old = malloc(PATH_MAX + 1);
2677 if (journal_cur == NULL || journal_old == NULL)
2679 fprintf(stderr, "malloc failure for journal files\n");
2684 snprintf(journal_cur, PATH_MAX, "%s/rrd.journal", dir);
2685 snprintf(journal_old, PATH_MAX, "%s/rrd.journal.old", dir);
2692 printf ("RRDCacheD %s Copyright (C) 2008 Florian octo Forster\n"
2694 "Usage: rrdcached [options]\n"
2696 "Valid options are:\n"
2697 " -l <address> Socket address to listen to.\n"
2698 " -L <address> Socket address to listen to ('FLUSH' only).\n"
2699 " -w <seconds> Interval in which to write data.\n"
2700 " -z <delay> Delay writes up to <delay> seconds to spread load\n"
2701 " -f <seconds> Interval in which to flush dead data.\n"
2702 " -p <file> Location of the PID-file.\n"
2703 " -b <dir> Base directory to change to.\n"
2704 " -B Restrict file access to paths within -b <dir>\n"
2705 " -g Do not fork and run in the foreground.\n"
2706 " -j <dir> Directory in which to create the journal files.\n"
2707 " -F Always flush all updates at shutdown\n"
2709 "For more information and a detailed description of all options "
2711 "to the rrdcached(1) manual page.\n",
2715 } /* switch (option) */
2716 } /* while (getopt) */
2718 /* advise the user when values are not sane */
2719 if (config_flush_interval < 2 * config_write_interval)
2720 fprintf(stderr, "WARNING: flush interval (-f) should be at least"
2721 " 2x write interval (-w) !\n");
2722 if (config_write_jitter > config_write_interval)
2723 fprintf(stderr, "WARNING: write delay (-z) should NOT be larger than"
2724 " write interval (-w) !\n");
2726 if (config_write_base_only && config_base_dir == NULL)
2727 fprintf(stderr, "WARNING: -B does not make sense without -b!\n"
2728 " Consult the rrdcached documentation\n");
2730 if (journal_cur == NULL)
2731 config_flush_at_shutdown = 1;
2734 } /* }}} int read_options */
2736 int main (int argc, char **argv)
2740 status = read_options (argc, argv);
2748 status = daemonize ();
2751 fprintf (stderr, "rrdcached: daemonize failed, exiting.\n");
2757 /* start the queue thread */
2758 memset (&queue_thread, 0, sizeof (queue_thread));
2759 status = pthread_create (&queue_thread,
2765 RRDD_LOG (LOG_ERR, "FATAL: cannot create queue thread");
2770 listen_thread_main (NULL);
2777 * vim: set sw=2 sts=2 ts=8 et fdm=marker :