2 * Copyright (C) 2008-2011 Andre Noll <maan@tuebingen.mpg.de>
4 * Licensed under the GPL v2. For licencing details see COPYING.
12 #include <sys/types.h>
28 #include "gcc-compat.h"
42 #define CMD_PTR(_cname) lls_cmd(LSG_DSS_CMD_ ## _cname, dss_suite)
43 #define OPT_RESULT(_cname, _oname) (lls_opt_result(\
44 LSG_DSS_ ## _cname ## _OPT_ ## _oname, (CMD_PTR(_cname) == CMD_PTR(DSS))? lpr : sublpr))
45 #define OPT_GIVEN(_cname, _oname) (lls_opt_given(OPT_RESULT(_cname, _oname)))
46 #define OPT_STRING_VAL(_cname, _oname) (lls_string_val(0, \
47 OPT_RESULT(_cname, _oname)))
48 #define OPT_UINT32_VAL(_cname, _oname) (lls_uint32_val(0, \
49 OPT_RESULT(_cname, _oname)))
51 struct dss_user_data {int (*handler)(void);};
52 #define EXPORT_CMD_HANDLER(_cmd) const struct dss_user_data \
53 lsg_dss_com_ ## _cmd ## _user_data = { \
54 .handler = com_ ## _cmd \
58 * Command line and active options. We need to keep a copy of the parsed
59 * command line options for the SIGHUP case where we merge the command line
60 * options and the new config file options.
62 static struct lls_parse_result *cmdline_lpr, *lpr;
64 /** Parsed subcommand options. */
65 static struct lls_parse_result *cmdline_sublpr, *sublpr;
66 /** Wether daemon_init() was called. */
67 static bool daemonized;
68 /** Non-NULL if we log to a file. */
70 /** The read end of the signal pipe */
71 static int signal_pipe;
72 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
73 static pid_t create_pid;
74 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
75 static int create_process_stopped;
76 /** How many times in a row the rsync command failed. */
77 static int num_consecutive_rsync_errors;
78 /** Process id of current pre-remove/rm/post-remove process. */
79 static pid_t remove_pid;
80 /** When the next snapshot is due. */
81 static int64_t next_snapshot_time;
82 /** When to try to remove something. */
83 static struct timeval next_removal_check;
84 /** Creation time of the snapshot currently being created. */
85 static int64_t current_snapshot_creation_time;
86 /** The snapshot currently being removed. */
87 struct snapshot *snapshot_currently_being_removed;
88 /** Needed by the post-create hook. */
89 static char *path_to_last_complete_snapshot;
90 static char *name_of_reference_snapshot;
91 /** \sa \ref snap.h for details. */
92 enum hook_status snapshot_creation_status;
93 /** \sa \ref snap.h for details. */
94 enum hook_status snapshot_removal_status;
98 static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
100 /* may be called with ds == NULL. */
101 static int disk_space_low(struct disk_space *ds)
103 struct disk_space ds_struct;
107 int ret = get_disk_space(".", &ds_struct);
112 val = OPT_UINT32_VAL(DSS, MIN_FREE_MB);
114 if (ds->free_mb < val)
116 val = OPT_UINT32_VAL(DSS, MIN_FREE_PERCENT);
118 if (ds->percent_free < val)
120 val = OPT_UINT32_VAL(DSS, MIN_FREE_PERCENT_INODES);
122 if (ds->percent_free_inodes < val)
127 static void dump_dss_config(const char *msg)
129 const char dash[] = "-----------------------------";
132 FILE *log = logfile? logfile : stderr;
133 struct disk_space ds;
134 int64_t now = get_current_time();
136 if (OPT_UINT32_VAL(DSS, LOGLEVEL) > INFO)
139 fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
140 fprintf(log, "\n*** disk space ***\n\n");
141 ret = get_disk_space(".", &ds);
143 DSS_INFO_LOG(("disk space low: %s\n", disk_space_low(&ds)?
147 DSS_ERROR_LOG(("can not get free disk space: %s\n",
148 dss_strerror(-ret)));
150 /* we continue on errors from get_disk_space */
152 fprintf(log, "\n*** non-default options ***\n\n");
153 lopsub_dump = lls_dump_parse_result(lpr, CMD_PTR(DSS), true);
154 fprintf(log, "%s", lopsub_dump);
156 fprintf(log, "\n*** non-default options for \"run\" ***\n\n");
157 lopsub_dump = lls_dump_parse_result(lpr, CMD_PTR(RUN), true);
158 fprintf(log, "%s", lopsub_dump);
160 fprintf(log, "\n*** internal state ***\n\n");
164 "snapshot_currently_being_removed: %s\n"
165 "path_to_last_complete_snapshot: %s\n"
166 "reference_snapshot: %s\n"
167 "snapshot_creation_status: %s\n"
168 "snapshot_removal_status: %s\n"
169 "num_consecutive_rsync_errors: %d\n"
172 logfile? OPT_STRING_VAL(RUN, LOGFILE) : "stderr",
173 snapshot_currently_being_removed?
174 snapshot_currently_being_removed->name : "(none)",
175 path_to_last_complete_snapshot?
176 path_to_last_complete_snapshot : "(none)",
177 name_of_reference_snapshot?
178 name_of_reference_snapshot : "(none)",
179 hook_status_description[snapshot_creation_status],
180 hook_status_description[snapshot_removal_status],
181 num_consecutive_rsync_errors
185 "create_pid: %" PRId32 "\n"
186 "create process is %sstopped\n"
189 create_process_stopped? "" : "not "
192 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
193 if (next_snapshot_time != 0)
194 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
195 next_snapshot_time - now);
196 if (current_snapshot_creation_time != 0)
197 fprintf(log, "current_snapshot_creation_time: %"
198 PRId64 " (%" PRId64 " seconds ago)\n",
199 current_snapshot_creation_time,
200 now - current_snapshot_creation_time
202 if (next_removal_check.tv_sec != 0) {
203 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
204 (long long unsigned)next_removal_check.tv_sec,
205 now - (long long unsigned)next_removal_check.tv_sec
209 fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
212 static int loglevel = -1;
213 static const char *location_file = NULL;
214 static int location_line = -1;
215 static const char *location_func = NULL;
217 void dss_log_set_params(int ll, const char *file, int line, const char *func)
220 location_file = file;
221 location_line = line;
222 location_func = func;
226 * The log function of dss.
228 * \param ll Loglevel.
229 * \param fml Usual format string.
231 * All DSS_XXX_LOG() macros use this function.
233 __printf_1_2 void dss_log(const char* fmt,...)
240 int lpr_ll = lpr? OPT_UINT32_VAL(DSS, LOGLEVEL) : WARNING;
242 if (loglevel < lpr_ll)
244 outfd = logfile? logfile : stderr;
247 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
248 fprintf(outfd, "%s ", str);
250 fprintf(outfd, "%i: ", loglevel);
251 #ifdef DSS_NO_FUNC_NAMES
252 fprintf(outfd, "%s:%d: ", location_file, location_line);
254 fprintf(outfd, "%s: ", location_func);
257 vfprintf(outfd, fmt, argp);
262 * Print a message either to stdout or to the log file.
264 static __printf_1_2 void dss_msg(const char* fmt,...)
266 FILE *outfd = logfile? logfile : stdout;
269 vfprintf(outfd, fmt, argp);
273 static char *get_config_file_name(void)
275 char *home, *config_file;
277 if (OPT_GIVEN(DSS, CONFIG_FILE))
278 return dss_strdup(OPT_STRING_VAL(DSS, CONFIG_FILE));
279 home = get_homedir();
280 config_file = make_message("%s/.dssrc", home);
285 static int send_signal(int sig, bool wait)
288 char *config_file = get_config_file_name();
289 int ret = get_dss_pid(config_file, &pid);
296 if (OPT_GIVEN(DSS, DRY_RUN)) {
297 dss_msg("%d\n", (int)pid);
300 ret = kill(pid, sig);
302 return -ERRNO_TO_DSS_ERROR(errno);
306 ts.tv_sec = ms / 1000;
307 ts.tv_nsec = (ms % 1000) * 1000 * 1000;
308 ret = nanosleep(&ts, NULL);
310 return -ERRNO_TO_DSS_ERROR(errno);
314 return -ERRNO_TO_DSS_ERROR(errno);
319 return -E_KILL_TIMEOUT;
323 const char * const name;
328 * The table below was taken 2016 from proc/sig.c of procps-3.2.8. Copyright
329 * 1998-2003 by Albert Cahalan, GPLv2.
331 static const struct signal_info signal_table[] = {
332 {"ABRT", SIGABRT}, /* IOT */
335 {"CHLD", SIGCHLD}, /* CLD */
344 {"POLL", SIGPOLL}, /* IO */
353 {"STKFLT", SIGSTKFLT},
356 {"SYS", SIGSYS}, /* UNUSED */
365 {"VTALRM", SIGVTALRM},
371 #define SIGNAL_TABLE_SIZE (sizeof(signal_table) / sizeof(signal_table[0]))
376 static int com_kill(void)
378 bool w_given = OPT_GIVEN(KILL, WAIT);
379 const char *arg = OPT_STRING_VAL(KILL, SIGNAL);
382 if (*arg >= '0' && *arg <= '9') {
384 ret = dss_atoi64(arg, &val);
387 if (val < 0 || val > SIGRTMAX)
388 return -ERRNO_TO_DSS_ERROR(EINVAL);
389 return send_signal(val, w_given);
391 if (strncasecmp(arg, "sig", 3) == 0)
393 if (strcasecmp(arg, "CLD") == 0)
394 return send_signal(SIGCHLD, w_given);
395 if (strcasecmp(arg, "IOT") == 0)
396 return send_signal(SIGABRT, w_given);
397 for (i = 0; i < SIGNAL_TABLE_SIZE; i++)
398 if (strcasecmp(arg, signal_table[i].name) == 0)
399 return send_signal(signal_table[i].num, w_given);
400 DSS_ERROR_LOG(("invalid sigspec: %s\n", arg));
401 return -ERRNO_TO_DSS_ERROR(EINVAL);
403 EXPORT_CMD_HANDLER(kill);
405 static void dss_get_snapshot_list(struct snapshot_list *sl)
407 get_snapshot_list(sl, OPT_UINT32_VAL(DSS, UNIT_INTERVAL),
408 OPT_UINT32_VAL(DSS, NUM_INTERVALS));
411 static int64_t compute_next_snapshot_time(void)
413 int64_t x = 0, now = get_current_time(), unit_interval
414 = 24 * 3600 * OPT_UINT32_VAL(DSS, UNIT_INTERVAL), ret;
415 unsigned wanted = desired_number_of_snapshots(0,
416 OPT_UINT32_VAL(DSS, NUM_INTERVALS)),
419 struct snapshot *s = NULL;
420 struct snapshot_list sl;
422 dss_get_snapshot_list(&sl);
423 FOR_EACH_SNAPSHOT(s, i, &sl) {
424 if (!(s->flags & SS_COMPLETE))
427 x += s->completion_time - s->creation_time;
432 if (num_complete == 0)
434 x /= num_complete; /* avg time to create one snapshot */
435 if (unit_interval < x * wanted) /* oops, no sleep at all */
437 ret = s->completion_time + unit_interval / wanted - x;
439 free_snapshot_list(&sl);
443 static inline void invalidate_next_snapshot_time(void)
445 next_snapshot_time = 0;
448 static inline int next_snapshot_time_is_valid(void)
450 return next_snapshot_time != 0;
453 static int next_snapshot_is_due(void)
455 int64_t now = get_current_time();
457 if (!next_snapshot_time_is_valid())
458 next_snapshot_time = compute_next_snapshot_time();
459 if (next_snapshot_time <= now) {
460 DSS_DEBUG_LOG(("next snapshot: now\n"));
463 DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
464 next_snapshot_time - now));
468 static void pre_create_hook(void)
470 assert(snapshot_creation_status == HS_READY);
471 /* make sure that the next snapshot time will be recomputed */
472 invalidate_next_snapshot_time();
473 DSS_DEBUG_LOG(("executing %s\n", OPT_STRING_VAL(DSS, PRE_CREATE_HOOK)));
474 dss_exec_cmdline_pid(&create_pid, OPT_STRING_VAL(DSS, PRE_CREATE_HOOK));
475 snapshot_creation_status = HS_PRE_RUNNING;
478 static void pre_remove_hook(struct snapshot *s, const char *why)
484 DSS_DEBUG_LOG(("%s snapshot %s\n", why, s->name));
485 assert(snapshot_removal_status == HS_READY);
486 assert(remove_pid == 0);
487 assert(!snapshot_currently_being_removed);
489 snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
490 *snapshot_currently_being_removed = *s;
491 snapshot_currently_being_removed->name = dss_strdup(s->name);
493 cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, PRE_REMOVE_HOOK),
494 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
495 DSS_DEBUG_LOG(("executing %s\n", cmd));
496 dss_exec_cmdline_pid(&remove_pid, cmd);
498 snapshot_removal_status = HS_PRE_RUNNING;
501 static int exec_rm(void)
503 struct snapshot *s = snapshot_currently_being_removed;
504 char *new_name = being_deleted_name(s);
513 assert(snapshot_removal_status == HS_PRE_SUCCESS);
514 assert(remove_pid == 0);
516 DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
517 ret = dss_rename(s->name, new_name);
520 dss_exec(&remove_pid, argv[0], argv);
521 snapshot_removal_status = HS_RUNNING;
527 static int snapshot_is_being_created(struct snapshot *s)
529 return s->creation_time == current_snapshot_creation_time;
532 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
537 DSS_DEBUG_LOG(("looking for orphaned snapshots\n"));
538 FOR_EACH_SNAPSHOT(s, i, sl) {
539 if (snapshot_is_being_created(s))
542 * We know that no rm is currently running, so if s is marked
543 * as being deleted, a previously started rm must have failed.
545 if (s->flags & SS_BEING_DELETED)
548 if (s->flags & SS_COMPLETE) /* good snapshot */
551 * This snapshot is incomplete and it is not the snapshot
552 * currently being created. However, we must not remove it if
553 * rsync is about to be restarted. As only the newest snapshot
554 * can be restarted, this snapshot is orphaned if it is not the
555 * newest snapshot or if we are not about to restart rsync.
557 if (get_newest_snapshot(sl) != s)
559 if (snapshot_creation_status != HS_NEEDS_RESTART)
562 /* no orphaned snapshots */
566 static int is_reference_snapshot(struct snapshot *s)
568 if (!name_of_reference_snapshot)
570 return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
574 * return: 0: no redundant snapshots, 1: rm process started, negative: error
576 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
580 unsigned missing = 0;
581 uint32_t N = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
583 DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
584 for (interval = N - 1; interval >= 0; interval--) {
585 unsigned keep = desired_number_of_snapshots(interval, N);
586 unsigned num = sl->interval_count[interval];
587 struct snapshot *victim = NULL, *prev = NULL;
588 int64_t score = LONG_MAX;
591 missing += keep - num;
592 if (keep + missing >= num)
594 /* redundant snapshot in this interval, pick snapshot with lowest score */
595 FOR_EACH_SNAPSHOT(s, i, sl) {
598 if (snapshot_is_being_created(s))
600 if (is_reference_snapshot(s))
602 if (s->interval > interval) {
606 if (s->interval < interval)
614 /* check if s is a better victim */
615 this_score = s->creation_time - prev->creation_time;
616 assert(this_score >= 0);
617 if (this_score < score) {
629 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
634 DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
635 OPT_UINT32_VAL(DSS, NUM_INTERVALS)));
636 FOR_EACH_SNAPSHOT(s, i, sl) {
637 if (snapshot_is_being_created(s))
639 if (is_reference_snapshot(s))
641 if (s->interval < OPT_UINT32_VAL(DSS, NUM_INTERVALS))
648 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
651 struct snapshot *s, *ref = NULL;
653 num_complete = num_complete_snapshots(sl);
654 if (num_complete <= OPT_UINT32_VAL(DSS, MIN_COMPLETE))
656 FOR_EACH_SNAPSHOT(s, i, sl) {
657 if (snapshot_is_being_created(s))
659 if (is_reference_snapshot(s)) { /* avoid this one */
663 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
667 DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
671 static int rename_incomplete_snapshot(int64_t start)
678 * We don't want the dss_rename() below to fail with EEXIST because the
679 * last complete snapshot was created (and completed) in the same
680 * second as this one.
682 while ((now = get_current_time()) == start)
684 free(path_to_last_complete_snapshot);
685 ret = complete_name(start, now, &path_to_last_complete_snapshot);
688 old_name = incomplete_name(start);
689 ret = dss_rename(old_name, path_to_last_complete_snapshot);
691 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
692 path_to_last_complete_snapshot));
697 static int try_to_free_disk_space(void)
700 struct snapshot_list sl;
701 struct snapshot *victim;
706 ret = disk_space_low(NULL);
709 low_disk_space = ret;
710 gettimeofday(&now, NULL);
711 if (tv_diff(&next_removal_check, &now, NULL) > 0)
713 if (!low_disk_space) {
714 if (OPT_GIVEN(DSS, KEEP_REDUNDANT))
716 if (snapshot_creation_status != HS_READY)
718 if (next_snapshot_is_due())
722 * Idle and --keep_redundant not given, or low disk space. Look at
723 * existing snapshots.
725 dss_get_snapshot_list(&sl);
728 * Don't remove anything if there is free space and we have fewer
729 * snapshots than configured, plus one. This way there is always one
730 * snapshot that can be recycled.
732 if (!low_disk_space && sl.num_snapshots <=
733 1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
736 victim = find_outdated_snapshot(&sl);
740 victim = find_redundant_snapshot(&sl);
744 victim = find_orphaned_snapshot(&sl);
747 /* try harder only if disk space is low */
750 DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
751 victim = find_oldest_removable_snapshot(&sl);
754 DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
755 ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
758 pre_remove_hook(victim, why);
760 free_snapshot_list(&sl);
764 static void post_create_hook(void)
766 char *cmd = make_message("%s %s/%s",
767 OPT_STRING_VAL(DSS, POST_CREATE_HOOK),
768 OPT_STRING_VAL(DSS, DEST_DIR), path_to_last_complete_snapshot);
769 DSS_NOTICE_LOG(("executing %s\n", cmd));
770 dss_exec_cmdline_pid(&create_pid, cmd);
772 snapshot_creation_status = HS_POST_RUNNING;
775 static void post_remove_hook(void)
778 struct snapshot *s = snapshot_currently_being_removed;
782 cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, POST_REMOVE_HOOK),
783 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
784 DSS_NOTICE_LOG(("executing %s\n", cmd));
785 dss_exec_cmdline_pid(&remove_pid, cmd);
787 snapshot_removal_status = HS_POST_RUNNING;
790 static void dss_kill(pid_t pid, int sig, const char *msg)
792 const char *signame, *process_name;
797 case SIGTERM: signame = "TERM"; break;
798 case SIGSTOP: signame = "STOP"; break;
799 case SIGCONT: signame = "CONT"; break;
800 default: signame = "????";
803 if (pid == create_pid)
804 process_name = "create";
805 else if (pid == remove_pid)
806 process_name = "remove";
807 else process_name = "??????";
810 DSS_INFO_LOG(("%s\n", msg));
811 DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
812 sig, signame, (int)pid, process_name));
813 if (kill(pid, sig) >= 0)
815 DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
816 sig, signame, (int)pid, process_name));
819 static void stop_create_process(void)
821 if (create_process_stopped)
823 dss_kill(create_pid, SIGSTOP, "suspending create process");
824 create_process_stopped = 1;
827 static void restart_create_process(void)
829 if (!create_process_stopped)
831 dss_kill(create_pid, SIGCONT, "resuming create process");
832 create_process_stopped = 0;
836 * Print a log message about the exit status of a child.
838 static void log_termination_msg(pid_t pid, int status)
840 if (WIFEXITED(status))
841 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
842 WEXITSTATUS(status)));
843 else if (WIFSIGNALED(status))
844 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
847 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
850 static int wait_for_process(pid_t pid, int *status)
854 DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
859 FD_SET(signal_pipe, &rfds);
860 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
866 if (ret == SIGCHLD) {
867 ret = waitpid(pid, status, 0);
870 if (errno != EINTR) { /* error */
871 ret = -ERRNO_TO_DSS_ERROR(errno);
875 /* SIGINT or SIGTERM */
876 dss_kill(pid, SIGTERM, "killing child process");
879 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
881 log_termination_msg(pid, *status);
885 static void handle_pre_remove_exit(int status)
887 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
888 snapshot_removal_status = HS_READY;
889 gettimeofday(&next_removal_check, NULL);
890 next_removal_check.tv_sec += 60;
893 snapshot_removal_status = HS_PRE_SUCCESS;
896 static int handle_rm_exit(int status)
898 if (!WIFEXITED(status)) {
899 snapshot_removal_status = HS_READY;
900 return -E_INVOLUNTARY_EXIT;
902 if (WEXITSTATUS(status)) {
903 snapshot_removal_status = HS_READY;
904 return -E_BAD_EXIT_CODE;
906 snapshot_removal_status = HS_SUCCESS;
910 static void handle_post_remove_exit(void)
912 snapshot_removal_status = HS_READY;
915 static int handle_remove_exit(int status)
918 struct snapshot *s = snapshot_currently_being_removed;
921 switch (snapshot_removal_status) {
923 handle_pre_remove_exit(status);
927 ret = handle_rm_exit(status);
929 case HS_POST_RUNNING:
930 handle_post_remove_exit();
936 if (snapshot_removal_status == HS_READY) {
939 snapshot_currently_being_removed = NULL;
945 static int wait_for_remove_process(void)
951 snapshot_removal_status == HS_PRE_RUNNING ||
952 snapshot_removal_status == HS_RUNNING ||
953 snapshot_removal_status == HS_POST_RUNNING
955 ret = wait_for_process(remove_pid, &status);
958 return handle_remove_exit(status);
961 static int handle_rsync_exit(int status)
965 if (!WIFEXITED(status)) {
966 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
967 ret = -E_INVOLUNTARY_EXIT;
968 snapshot_creation_status = HS_READY;
971 es = WEXITSTATUS(status);
973 * Restart rsync on non-fatal errors:
974 * 24: Partial transfer due to vanished source files
976 if (es != 0 && es != 24) {
977 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
978 es, ++num_consecutive_rsync_errors));
979 if (!logfile) { /* called by com_run() */
980 ret = -E_BAD_EXIT_CODE;
983 if (num_consecutive_rsync_errors >
984 OPT_UINT32_VAL(RUN, MAX_RSYNC_ERRORS)) {
985 ret = -E_TOO_MANY_RSYNC_ERRORS;
986 snapshot_creation_status = HS_READY;
989 DSS_WARNING_LOG(("restarting rsync process\n"));
990 snapshot_creation_status = HS_NEEDS_RESTART;
991 next_snapshot_time = get_current_time() + 60;
995 num_consecutive_rsync_errors = 0;
996 ret = rename_incomplete_snapshot(current_snapshot_creation_time);
999 snapshot_creation_status = HS_SUCCESS;
1000 free(name_of_reference_snapshot);
1001 name_of_reference_snapshot = NULL;
1003 create_process_stopped = 0;
1007 static int handle_pre_create_hook_exit(int status)
1010 static int warn_count;
1012 if (!WIFEXITED(status)) {
1013 snapshot_creation_status = HS_READY;
1014 ret = -E_INVOLUNTARY_EXIT;
1017 es = WEXITSTATUS(status);
1019 if (!warn_count--) {
1020 DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
1021 OPT_STRING_VAL(DSS, PRE_CREATE_HOOK), es));
1022 DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
1023 warn_count = 60; /* warn only once per hour */
1025 next_snapshot_time = get_current_time() + 60;
1026 snapshot_creation_status = HS_READY;
1031 snapshot_creation_status = HS_PRE_SUCCESS;
1037 static int handle_sigchld(void)
1040 int status, ret = reap_child(&pid, &status);
1045 if (pid == create_pid) {
1046 switch (snapshot_creation_status) {
1047 case HS_PRE_RUNNING:
1048 ret = handle_pre_create_hook_exit(status);
1051 ret = handle_rsync_exit(status);
1053 case HS_POST_RUNNING:
1054 snapshot_creation_status = HS_READY;
1058 DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
1059 snapshot_creation_status));
1065 if (pid == remove_pid) {
1066 ret = handle_remove_exit(status);
1071 DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
1075 static int change_to_dest_dir(void)
1078 const char *dd = OPT_STRING_VAL(DSS, DEST_DIR);
1080 DSS_INFO_LOG(("changing cwd to %s\n", dd));
1083 ret = -ERRNO_TO_DSS_ERROR(errno);
1084 DSS_ERROR_LOG(("could not change cwd to %s\n", dd));
1088 static int check_config(const struct lls_command *cmd)
1091 uint32_t unit_interval = OPT_UINT32_VAL(DSS, UNIT_INTERVAL);
1092 uint32_t num_intervals = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
1094 if (unit_interval == 0) {
1095 DSS_ERROR_LOG(("bad unit interval: %i\n", unit_interval));
1096 return -E_INVALID_NUMBER;
1098 DSS_DEBUG_LOG(("unit interval: %i day(s)\n", unit_interval));
1100 if (num_intervals == 0 || num_intervals > 30) {
1101 DSS_ERROR_LOG(("bad number of intervals: %i\n", num_intervals));
1102 return -E_INVALID_NUMBER;
1104 if (cmd == CMD_PTR(RUN) || cmd == CMD_PTR(CREATE))
1105 if (!OPT_GIVEN(DSS, SOURCE_DIR)) {
1106 DSS_ERROR_LOG(("--source-dir required\n"));
1109 if (cmd == CMD_PTR(RUN) || cmd == CMD_PTR(CREATE)
1110 || cmd == CMD_PTR(LS) || cmd == CMD_PTR(PRUNE)) {
1111 if (!OPT_GIVEN(DSS, DEST_DIR)) {
1112 DSS_ERROR_LOG(("--dest-dir required\n"));
1115 ret = change_to_dest_dir();
1119 DSS_DEBUG_LOG(("number of intervals: %i\n", num_intervals));
1123 static int lopsub_error(int lopsub_ret, char **errctx)
1125 const char *msg = lls_strerror(-lopsub_ret);
1127 DSS_ERROR_LOG(("%s: %s\n", *errctx, msg));
1129 DSS_ERROR_LOG(("%s\n", msg));
1135 static int parse_config_file(bool sighup, const struct lls_command *cmd)
1138 char *config_file = get_config_file_name();
1139 struct stat statbuf;
1143 char **cf_argv, *errctx = NULL;
1144 struct lls_parse_result *cf_lpr, *merged_lpr, *clpr;
1145 const char *subcmd_name;
1147 ret = open(config_file, O_RDONLY);
1149 if (errno != ENOENT || OPT_GIVEN(DSS, CONFIG_FILE)) {
1150 ret = -ERRNO_TO_DSS_ERROR(errno);
1151 DSS_ERROR_LOG(("config file %s can not be opened\n",
1155 /* no config file -- nothing to do */
1160 ret = fstat(fd, &statbuf);
1162 ret = -ERRNO_TO_DSS_ERROR(errno);
1163 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
1166 sz = statbuf.st_size;
1167 if (sz == 0) { /* config file is empty -- nothing to do */
1171 map = mmap(NULL, sz, PROT_READ, MAP_PRIVATE, fd, 0);
1172 if (map == MAP_FAILED) {
1173 ret = -ERRNO_TO_DSS_ERROR(errno);
1174 DSS_ERROR_LOG(("failed to mmap config file %s\n",
1178 if (cmd == CMD_PTR(DSS))
1181 subcmd_name = lls_command_name(cmd);
1182 ret = lls_convert_config(map, sz, subcmd_name, &cf_argv, &errctx);
1185 DSS_ERROR_LOG(("failed to convert config file %s\n",
1187 ret = lopsub_error(ret, &errctx);
1191 ret = lls_parse(cf_argc, cf_argv, cmd, &cf_lpr, &errctx);
1192 lls_free_argv(cf_argv);
1194 ret = lopsub_error(ret, &errctx);
1197 clpr = cmd == CMD_PTR(DSS)? cmdline_lpr : cmdline_sublpr;
1198 if (sighup) /* config file overrides command line */
1199 ret = lls_merge(cf_lpr, clpr, cmd, &merged_lpr, &errctx);
1200 else /* command line options overrride config file options */
1201 ret = lls_merge(clpr, cf_lpr, cmd, &merged_lpr, &errctx);
1202 lls_free_parse_result(cf_lpr, cmd);
1204 ret = lopsub_error(ret, &errctx);
1210 DSS_DEBUG_LOG(("loglevel: %d\n", OPT_UINT32_VAL(DSS, LOGLEVEL)));
1211 if (cmd != CMD_PTR(DSS)) {
1213 if (sublpr != cmdline_sublpr)
1214 lls_free_parse_result(sublpr, cmd);
1215 sublpr = merged_lpr;
1217 sublpr = cmdline_sublpr;
1220 if (lpr != cmdline_lpr)
1221 lls_free_parse_result(lpr, cmd);
1232 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1236 static int handle_sighup(void)
1240 DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1241 dump_dss_config("old");
1242 ret = parse_config_file(true /* SIGHUP */, CMD_PTR(DSS));
1245 ret = parse_config_file(true /* SIGHUP */, CMD_PTR(RUN));
1248 ret = check_config(CMD_PTR(RUN));
1253 if (OPT_GIVEN(RUN, DAEMON) || daemonized) {
1254 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1255 log_welcome(OPT_UINT32_VAL(DSS, LOGLEVEL));
1258 dump_dss_config("reloaded");
1259 invalidate_next_snapshot_time();
1263 static void kill_children(void)
1265 restart_create_process();
1266 dss_kill(create_pid, SIGTERM, NULL);
1267 dss_kill(remove_pid, SIGTERM, NULL);
1270 static int handle_signal(void)
1272 int sig, ret = next_signal();
1284 ret = handle_sighup();
1287 ret = handle_sigchld();
1292 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1297 * We can not use rsync locally if the local user is different from the remote
1298 * user or if the src dir is not on the local host (or both).
1300 static int use_rsync_locally(char *logname)
1302 const char *h = OPT_STRING_VAL(DSS, REMOTE_HOST);
1304 if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1306 if (OPT_GIVEN(DSS, REMOTE_USER) &&
1307 strcmp(OPT_STRING_VAL(DSS, REMOTE_USER), logname))
1312 static int rename_resume_snap(int64_t creation_time)
1314 struct snapshot_list sl;
1315 struct snapshot *s = NULL;
1316 char *new_name = incomplete_name(creation_time);
1320 sl.num_snapshots = 0;
1323 dss_get_snapshot_list(&sl);
1325 * Snapshot recycling: We first look at the newest snapshot. If this
1326 * snapshot happens to be incomplete, the last rsync process was
1327 * aborted and we reuse this one. Otherwise we look at snapshots which
1328 * could be removed (outdated and redundant snapshots) as candidates
1329 * for recycling. If no outdated/redundant snapshot exists, we check if
1330 * there is an orphaned snapshot, which likely is useless anyway.
1332 * Only if no existing snapshot is suitable for recycling, we bite the
1333 * bullet and create a new one.
1335 s = get_newest_snapshot(&sl);
1336 if (!s) /* no snapshots at all */
1338 /* re-use last snapshot if it is incomplete */
1340 if ((s->flags & SS_COMPLETE) == 0)
1343 s = find_outdated_snapshot(&sl);
1347 s = find_redundant_snapshot(&sl);
1351 s = find_orphaned_snapshot(&sl);
1354 DSS_NOTICE_LOG(("recycling %s snapshot %s\n", why, s->name));
1355 ret = dss_rename(s->name, new_name);
1358 DSS_NOTICE_LOG(("creating %s\n", new_name));
1360 free_snapshot_list(&sl);
1364 static void create_rsync_argv(char ***argv, int64_t *num)
1367 int i = 0, j, N = OPT_GIVEN(DSS, RSYNC_OPTION);
1368 struct snapshot_list sl;
1371 dss_get_snapshot_list(&sl);
1372 assert(!name_of_reference_snapshot);
1373 name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1374 free_snapshot_list(&sl);
1376 *argv = dss_malloc((15 + N) * sizeof(char *));
1377 (*argv)[i++] = dss_strdup("rsync");
1378 (*argv)[i++] = dss_strdup("-a");
1379 (*argv)[i++] = dss_strdup("--delete");
1381 srandom((unsigned)time(NULL)); /* no need to be fancy here */
1384 if (1000 * (random() / (RAND_MAX + 1.0)) < OPT_UINT32_VAL(DSS, CHECKSUM)) {
1385 DSS_NOTICE_LOG(("adding --checksum to rsync options\n"));
1386 (*argv)[i++] = dss_strdup("--checksum");
1388 for (j = 0; j < N; j++)
1389 (*argv)[i++] = dss_strdup(lls_string_val(j,
1390 OPT_RESULT(DSS, RSYNC_OPTION)));
1391 if (name_of_reference_snapshot) {
1392 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1393 (*argv)[i++] = make_message("--link-dest=../%s",
1394 name_of_reference_snapshot);
1396 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1397 logname = dss_logname();
1398 if (use_rsync_locally(logname))
1399 (*argv)[i++] = dss_strdup(OPT_STRING_VAL(DSS, SOURCE_DIR));
1401 (*argv)[i++] = make_message("%s@%s:%s/",
1402 OPT_GIVEN(DSS, REMOTE_USER)?
1403 OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
1404 OPT_STRING_VAL(DSS, REMOTE_HOST),
1405 OPT_STRING_VAL(DSS, SOURCE_DIR));
1407 *num = get_current_time();
1408 (*argv)[i++] = incomplete_name(*num);
1409 (*argv)[i++] = NULL;
1410 for (j = 0; j < i; j++)
1411 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1414 static void free_rsync_argv(char **argv)
1420 for (i = 0; argv[i]; i++)
1425 static int create_snapshot(char **argv)
1429 ret = rename_resume_snap(current_snapshot_creation_time);
1432 dss_exec(&create_pid, argv[0], argv);
1433 snapshot_creation_status = HS_RUNNING;
1437 static int select_loop(void)
1440 /* check every 60 seconds for free disk space */
1442 char **rsync_argv = NULL;
1446 struct timeval *tvp;
1449 tvp = NULL; /* sleep until rm hook/process dies */
1450 else { /* sleep one minute */
1456 FD_SET(signal_pipe, &rfds);
1457 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1460 if (FD_ISSET(signal_pipe, &rfds)) {
1461 ret = handle_signal();
1467 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1473 if (snapshot_removal_status == HS_SUCCESS) {
1477 ret = try_to_free_disk_space();
1480 if (snapshot_removal_status != HS_READY) {
1481 stop_create_process();
1484 restart_create_process();
1485 switch (snapshot_creation_status) {
1487 if (!next_snapshot_is_due())
1491 case HS_PRE_RUNNING:
1493 case HS_POST_RUNNING:
1495 case HS_PRE_SUCCESS:
1496 if (!name_of_reference_snapshot) {
1497 free_rsync_argv(rsync_argv);
1498 create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
1500 ret = create_snapshot(rsync_argv);
1504 case HS_NEEDS_RESTART:
1505 if (!next_snapshot_is_due())
1507 ret = create_snapshot(rsync_argv);
1520 static void exit_hook(int exit_code)
1522 const char *argv[3];
1525 argv[0] = OPT_STRING_VAL(DSS, EXIT_HOOK);
1526 argv[1] = dss_strerror(-exit_code);
1529 DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
1530 dss_exec(&pid, argv[0], (char **)argv);
1533 static void lock_dss_or_die(void)
1535 char *config_file = get_config_file_name();
1536 int ret = lock_dss(config_file);
1540 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1545 static int com_run(void)
1551 if (OPT_GIVEN(DSS, DRY_RUN)) {
1552 DSS_ERROR_LOG(("dry run not supported by this command\n"));
1555 config_file = get_config_file_name();
1556 ret = get_dss_pid(config_file, &pid);
1559 DSS_ERROR_LOG(("pid %d\n", (int)pid));
1560 return -E_ALREADY_RUNNING;
1562 if (OPT_GIVEN(RUN, DAEMON)) {
1565 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1568 dump_dss_config("startup");
1569 ret = install_sighandler(SIGHUP);
1573 ret = write(fd, "\0", 1);
1575 DSS_ERROR_LOG(("write to daemon pipe returned %d\n",
1578 return -ERRNO_TO_DSS_ERROR(errno);
1582 ret = select_loop();
1583 if (ret >= 0) /* impossible */
1587 while (wait(NULL) >= 0 || errno != ECHILD)
1588 ; /* still have children to wait for */
1591 EXPORT_CMD_HANDLER(run);
1593 static int com_prune(void)
1596 struct snapshot_list sl;
1597 struct snapshot *victim;
1598 struct disk_space ds;
1602 ret = get_disk_space(".", &ds);
1605 log_disk_space(&ds);
1606 dss_get_snapshot_list(&sl);
1608 victim = find_outdated_snapshot(&sl);
1612 victim = find_redundant_snapshot(&sl);
1618 if (OPT_GIVEN(DSS, DRY_RUN)) {
1619 dss_msg("%s snapshot %s (interval = %i)\n",
1620 why, victim->name, victim->interval);
1624 pre_remove_hook(victim, why);
1625 if (snapshot_removal_status == HS_PRE_RUNNING) {
1626 ret = wait_for_remove_process();
1629 if (snapshot_removal_status != HS_PRE_SUCCESS)
1635 ret = wait_for_remove_process();
1638 if (snapshot_removal_status != HS_SUCCESS)
1641 if (snapshot_removal_status != HS_POST_RUNNING)
1643 ret = wait_for_remove_process();
1648 free_snapshot_list(&sl);
1651 EXPORT_CMD_HANDLER(prune);
1653 static int com_create(void)
1659 if (OPT_GIVEN(DSS, DRY_RUN)) {
1662 create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
1663 for (i = 0; rsync_argv[i]; i++) {
1665 msg = make_message("%s%s%s", tmp? tmp : "",
1666 tmp? " " : "", rsync_argv[i]);
1669 free_rsync_argv(rsync_argv);
1670 dss_msg("%s\n", msg);
1676 ret = wait_for_process(create_pid, &status);
1679 ret = handle_pre_create_hook_exit(status);
1680 if (ret <= 0) /* error, or pre-create failed */
1683 create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
1684 ret = create_snapshot(rsync_argv);
1687 ret = wait_for_process(create_pid, &status);
1690 ret = handle_rsync_exit(status);
1695 ret = wait_for_process(create_pid, &status);
1697 free_rsync_argv(rsync_argv);
1700 EXPORT_CMD_HANDLER(create);
1702 static int com_ls(void)
1705 struct snapshot_list sl;
1708 dss_get_snapshot_list(&sl);
1709 FOR_EACH_SNAPSHOT(s, i, &sl) {
1711 if (s->flags & SS_COMPLETE)
1712 d = (s->completion_time - s->creation_time) / 60;
1713 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1715 free_snapshot_list(&sl);
1718 EXPORT_CMD_HANDLER(ls);
1720 static int setup_signal_handling(void)
1724 DSS_INFO_LOG(("setting up signal handlers\n"));
1725 signal_pipe = signal_init(); /* always successful */
1726 ret = install_sighandler(SIGINT);
1729 ret = install_sighandler(SIGTERM);
1732 return install_sighandler(SIGCHLD);
1735 static void handle_version_and_help(void)
1739 if (OPT_GIVEN(DSS, DETAILED_HELP))
1740 txt = lls_long_help(CMD_PTR(DSS));
1741 else if (OPT_GIVEN(DSS, HELP))
1742 txt = lls_short_help(CMD_PTR(DSS));
1743 else if (OPT_GIVEN(DSS, VERSION))
1744 txt = dss_strdup(VERSION_STRING);
1752 static void show_subcommand_summary(void)
1754 const struct lls_command *cmd;
1757 printf("Available subcommands:\n");
1758 for (i = 1; (cmd = lls_cmd(i, dss_suite)); i++) {
1759 const char *name = lls_command_name(cmd);
1760 const char *purpose = lls_purpose(cmd);
1761 printf("%-10s%s\n", name, purpose);
1766 int main(int argc, char **argv)
1769 const struct lls_command *cmd = CMD_PTR(DSS);
1770 char *errctx = NULL;
1771 unsigned num_inputs;
1772 const struct dss_user_data *ud = NULL;
1774 ret = lls_parse(argc, argv, cmd, &cmdline_lpr, &errctx);
1776 ret = lopsub_error(ret, &errctx);
1780 ret = parse_config_file(false /* no SIGHUP */, cmd);
1783 handle_version_and_help();
1784 num_inputs = lls_num_inputs(lpr);
1785 if (num_inputs == 0)
1786 show_subcommand_summary();
1787 ret = lls_lookup_subcmd(argv[argc - num_inputs], dss_suite, &errctx);
1789 ret = lopsub_error(ret, &errctx);
1792 cmd = lls_cmd(ret, dss_suite);
1793 ret = lls_parse(num_inputs, argv + argc - num_inputs, cmd,
1794 &cmdline_sublpr, &errctx);
1796 ret = lopsub_error(ret, &errctx);
1799 sublpr = cmdline_sublpr;
1800 ret = parse_config_file(false /* no SIGHUP */, cmd);
1803 ret = check_config(cmd);
1806 ret = setup_signal_handling();
1809 ud = lls_user_data(cmd);
1810 ret = ud->handler();
1815 DSS_ERROR_LOG(("%s\n", errctx));
1816 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1819 lls_free_parse_result(lpr, CMD_PTR(DSS));
1820 if (lpr != cmdline_lpr)
1821 lls_free_parse_result(cmdline_lpr, CMD_PTR(DSS));
1822 lls_free_parse_result(sublpr, cmd);
1823 if (sublpr != cmdline_sublpr)
1824 lls_free_parse_result(cmdline_sublpr, cmd);
1825 exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);