Fix wait_pro_process().
[dss.git] / dss.c
1 #include <string.h>
2 #include <stdlib.h>
3 #include <stdarg.h>
4 #include <assert.h>
5 #include <errno.h>
6 #include <sys/types.h>
7 #include <signal.h>
8 #include <ctype.h>
9 #include <sys/stat.h>
10 #include <unistd.h>
11 #include <inttypes.h>
12 #include <sys/time.h>
13 #include <time.h>
14 #include <sys/wait.h>
15 #include <fnmatch.h>
16 #include <limits.h>
17
18
19 #include "gcc-compat.h"
20 #include "cmdline.h"
21 #include "log.h"
22 #include "string.h"
23 #include "error.h"
24 #include "fd.h"
25 #include "exec.h"
26 #include "daemon.h"
27 #include "signal.h"
28 #include "df.h"
29 #include "time.h"
30
31
32 struct gengetopt_args_info conf;
33 char *dss_error_txt = NULL;
34 static FILE *logfile;
35 static int signal_pipe;
36
37 /** Process id of current rsync process. */
38 static pid_t rsync_pid;
39 /** Whether the rsync process is currently stopped */
40 static int rsync_stopped;
41 /** Process id of current rm process. */
42 static pid_t rm_pid;
43 /** When the next snapshot is due. */
44 struct timeval next_snapshot_time;
45 /* Creation time of the snapshot currently being created. */
46 int64_t current_snapshot_creation_time;
47
48
49 DEFINE_DSS_ERRLIST;
50
51
52 /* a litte cpp magic helps to DRY */
53 #define COMMANDS \
54 COMMAND(ls) \
55 COMMAND(create) \
56 COMMAND(prune) \
57 COMMAND(run)
58 #define COMMAND(x) int com_ ##x(void);
59 COMMANDS
60 #undef COMMAND
61 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
62 int call_command_handler(void)
63 {
64 COMMANDS
65 DSS_EMERG_LOG("BUG: did not find command handler\n");
66 exit(EXIT_FAILURE);
67 }
68 #undef COMMAND
69 #undef COMMANDS
70
71 /*
72 * complete, not being deleted: 1204565370-1204565371.Sun_Mar_02_2008_14_33-Sun_Mar_02_2008_14_43
73 * complete, being deleted: 1204565370-1204565371.being_deleted
74 * incomplete, not being deleted: 1204565370-incomplete
75 * incomplete, being deleted: 1204565370-incomplete.being_deleted
76 */
77 enum snapshot_status_flags {
78 /** The rsync process terminated successfully. */
79 SS_COMPLETE = 1,
80 /** The rm process is running to remove this snapshot. */
81 SS_BEING_DELETED = 2,
82 };
83
84 struct snapshot {
85 char *name;
86 int64_t creation_time;
87 int64_t completion_time;
88 enum snapshot_status_flags flags;
89 unsigned interval;
90 };
91
92 /*
93 * An edge snapshot is either the oldest one or the newest one.
94 *
95 * We need to find either of them occasionally: The create code
96 * needs to know the newest snapshot because that is the one
97 * used as the link destination dir. The pruning code needs to
98 * find the oldest one in case disk space becomes low.
99 */
100 struct edge_snapshot_data {
101 int64_t now;
102 struct snapshot snap;
103 };
104
105 __printf_2_3 void dss_log(int ll, const char* fmt,...)
106 {
107 va_list argp;
108 FILE *outfd;
109 struct tm *tm;
110 time_t t1;
111 char str[255] = "";
112
113 if (ll < conf.loglevel_arg)
114 return;
115 outfd = logfile? logfile : stderr;
116 time(&t1);
117 tm = localtime(&t1);
118 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
119 fprintf(outfd, "%s ", str);
120 if (conf.loglevel_arg <= INFO)
121 fprintf(outfd, "%i: ", ll);
122 va_start(argp, fmt);
123 vfprintf(outfd, fmt, argp);
124 va_end(argp);
125 }
126
127 /**
128 * Print a message either to stdout or to the log file.
129 */
130 __printf_1_2 void dss_msg(const char* fmt,...)
131 {
132 FILE *outfd = conf.daemon_given? logfile : stdout;
133 va_list argp;
134 va_start(argp, fmt);
135 vfprintf(outfd, fmt, argp);
136 va_end(argp);
137 }
138
139 /**
140 * Return the desired number of snapshots of an interval.
141 */
142 unsigned num_snapshots(int interval)
143 {
144 unsigned n;
145
146 assert(interval >= 0);
147
148 if (interval >= conf.num_intervals_arg)
149 return 0;
150 n = conf.num_intervals_arg - interval - 1;
151 return 1 << n;
152 }
153
154 /* return: Whether dirname is a snapshot directory (0: no, 1: yes) */
155 int is_snapshot(const char *dirname, int64_t now, struct snapshot *s)
156 {
157 int i, ret;
158 char *dash, *dot, *tmp;
159 int64_t num;
160
161 assert(dirname);
162 dash = strchr(dirname, '-');
163 if (!dash || !dash[1] || dash == dirname)
164 return 0;
165 for (i = 0; dirname[i] != '-'; i++)
166 if (!isdigit(dirname[i]))
167 return 0;
168 tmp = dss_strdup(dirname);
169 tmp[i] = '\0';
170 ret = dss_atoi64(tmp, &num);
171 free(tmp);
172 if (ret < 0) {
173 free(dss_error_txt);
174 return 0;
175 }
176 assert(num >= 0);
177 if (num > now)
178 return 0;
179 s->creation_time = num;
180 //DSS_DEBUG_LOG("%s start time: %lli\n", dirname, (long long)s->creation_time);
181 s->interval = (long long) ((now - s->creation_time)
182 / conf.unit_interval_arg / 24 / 3600);
183 if (!strcmp(dash + 1, "incomplete")) {
184 s->completion_time = -1;
185 s->flags = 0; /* neither complete, nor being deleted */
186 goto success;
187 }
188 if (!strcmp(dash + 1, "incomplete.being_deleted")) {
189 s->completion_time = -1;
190 s->flags = SS_BEING_DELETED; /* mot cpmplete, being deleted */
191 goto success;
192 }
193 tmp = dash + 1;
194 dot = strchr(tmp, '.');
195 if (!dot || !dot[1] || dot == tmp)
196 return 0;
197 for (i = 0; tmp[i] != '.'; i++)
198 if (!isdigit(tmp[i]))
199 return 0;
200 tmp = dss_strdup(dash + 1);
201 tmp[i] = '\0';
202 ret = dss_atoi64(tmp, &num);
203 free(tmp);
204 if (ret < 0) {
205 free(dss_error_txt);
206 return 0;
207 }
208 if (num > now)
209 return 0;
210 s->completion_time = num;
211 s->flags = SS_COMPLETE;
212 if (!strcmp(dot + 1, "being_deleted"))
213 s->flags |= SS_BEING_DELETED;
214 success:
215 s->name = dss_strdup(dirname);
216 return 1;
217 }
218
219 int64_t get_current_time(void)
220 {
221 time_t now;
222 time(&now);
223 DSS_DEBUG_LOG("now: %lli\n", (long long) now);
224 return (int64_t)now;
225 }
226
227 char *incomplete_name(int64_t start)
228 {
229 return make_message("%lli-incomplete", (long long)start);
230 }
231
232 char *being_deleted_name(struct snapshot *s)
233 {
234 if (s->flags & SS_COMPLETE)
235 return make_message("%lli-%lli.being_deleted",
236 (long long)s->creation_time,
237 (long long)s->completion_time);
238 return make_message("%lli-incomplete.being_deleted",
239 (long long)s->creation_time);
240 }
241
242 int complete_name(int64_t start, int64_t end, char **result)
243 {
244 struct tm start_tm, end_tm;
245 time_t *start_seconds = (time_t *) (uint64_t *)&start; /* STFU, gcc */
246 time_t *end_seconds = (time_t *) (uint64_t *)&end; /* STFU, gcc */
247 char start_str[200], end_str[200];
248
249 if (!localtime_r(start_seconds, &start_tm)) {
250 make_err_msg("%lli", (long long)start);
251 return -E_LOCALTIME;
252 }
253 if (!localtime_r(end_seconds, &end_tm)) {
254 make_err_msg("%lli", (long long)end);
255 return -E_LOCALTIME;
256 }
257 if (!strftime(start_str, sizeof(start_str), "%a_%b_%d_%Y_%H_%M_%S", &start_tm)) {
258 make_err_msg("%lli", (long long)start);
259 return -E_STRFTIME;
260 }
261 if (!strftime(end_str, sizeof(end_str), "%a_%b_%d_%Y_%H_%M_%S", &end_tm)) {
262 make_err_msg("%lli", (long long)end);
263 return -E_STRFTIME;
264 }
265 *result = make_message("%lli-%lli.%s-%s", (long long) start, (long long) end,
266 start_str, end_str);
267 return 1;
268 }
269
270 struct snapshot_list {
271 int64_t now;
272 unsigned num_snapshots;
273 unsigned array_size;
274 struct snapshot **snapshots;
275 /**
276 * Array of size num_intervals + 1
277 *
278 * It contains the number of snapshots in each interval. interval_count[num_intervals]
279 * is the number of snapshots which belong to any interval greater than num_intervals.
280 */
281 unsigned *interval_count;
282 };
283
284 #define FOR_EACH_SNAPSHOT(s, i, sl) \
285 for ((i) = 0; (i) < (sl)->num_snapshots && ((s) = (sl)->snapshots[(i)]); (i)++)
286
287
288
289 #define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
290
291 static int compare_snapshots(const void *a, const void *b)
292 {
293 struct snapshot *s1 = *(struct snapshot **)a;
294 struct snapshot *s2 = *(struct snapshot **)b;
295 return NUM_COMPARE(s2->creation_time, s1->creation_time);
296 }
297
298 /** Compute the minimum of \a a and \a b. */
299 #define DSS_MIN(a,b) ((a) < (b) ? (a) : (b))
300
301 int add_snapshot(const char *dirname, void *private)
302 {
303 struct snapshot_list *sl = private;
304 struct snapshot s;
305 int ret = is_snapshot(dirname, sl->now, &s);
306
307 if (!ret)
308 return 1;
309 if (sl->num_snapshots >= sl->array_size) {
310 sl->array_size = 2 * sl->array_size + 1;
311 sl->snapshots = dss_realloc(sl->snapshots,
312 sl->array_size * sizeof(struct snapshot *));
313 }
314 sl->snapshots[sl->num_snapshots] = dss_malloc(sizeof(struct snapshot));
315 *(sl->snapshots[sl->num_snapshots]) = s;
316 sl->interval_count[DSS_MIN(s.interval, conf.num_intervals_arg)]++;
317 sl->num_snapshots++;
318 return 1;
319 }
320
321 void get_snapshot_list(struct snapshot_list *sl)
322 {
323 sl->now = get_current_time();
324 sl->num_snapshots = 0;
325 sl->array_size = 0;
326 sl->snapshots = NULL;
327 sl->interval_count = dss_calloc((conf.num_intervals_arg + 1) * sizeof(unsigned));
328 for_each_subdir(add_snapshot, sl);
329 qsort(sl->snapshots, sl->num_snapshots, sizeof(struct snapshot *),
330 compare_snapshots);
331 }
332
333 void free_snapshot_list(struct snapshot_list *sl)
334 {
335 int i;
336 struct snapshot *s;
337
338 FOR_EACH_SNAPSHOT(s, i, sl) {
339 free(s->name);
340 free(s);
341 }
342 free(sl->interval_count);
343 sl->interval_count = NULL;
344 free(sl->snapshots);
345 sl->snapshots = NULL;
346 sl->num_snapshots = 0;
347 }
348
349 void stop_rsync_process(void)
350 {
351 if (!rsync_pid || rsync_stopped)
352 return;
353 kill(SIGSTOP, rsync_pid);
354 rsync_stopped = 1;
355 }
356
357 void restart_rsync_process(void)
358 {
359 if (!rsync_pid || !rsync_stopped)
360 return;
361 kill (SIGCONT, rsync_pid);
362 rsync_stopped = 0;
363 }
364
365 /**
366 * Print a log message about the exit status of a child.
367 */
368 void log_termination_msg(pid_t pid, int status)
369 {
370 if (WIFEXITED(status))
371 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
372 WEXITSTATUS(status));
373 else if (WIFSIGNALED(status))
374 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
375 WTERMSIG(status));
376 else
377 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
378 }
379
380 int wait_for_process(pid_t pid, int *status)
381 {
382 int ret;
383
384 DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
385 for (;;) {
386 fd_set rfds;
387
388 FD_ZERO(&rfds);
389 FD_SET(signal_pipe, &rfds);
390 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
391 if (ret < 0)
392 break;
393 ret = next_signal();
394 if (!ret)
395 continue;
396 if (ret == SIGCHLD) {
397 ret = waitpid(pid, status, 0);
398 if (ret >= 0)
399 break;
400 if (errno != EINTR) { /* error */
401 ret = -ERRNO_TO_DSS_ERROR(errno);
402 break;
403 }
404 }
405 /* SIGINT or SIGTERM */
406 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
407 kill(pid, SIGTERM);
408 }
409 if (ret < 0)
410 make_err_msg("failed to wait for process %d", (int)pid);
411 else
412 log_termination_msg(pid, *status);
413 return ret;
414 }
415
416 int remove_snapshot(struct snapshot *s)
417 {
418 int fds[3] = {0, 0, 0};
419 assert(!rm_pid);
420 char *new_name = being_deleted_name(s);
421 int ret = dss_rename(s->name, new_name);
422 char *argv[] = {"rm", "-rf", new_name, NULL};
423
424 if (ret < 0)
425 goto out;
426 DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
427 stop_rsync_process();
428 ret = dss_exec(&rm_pid, argv[0], argv, fds);
429 out:
430 free(new_name);
431 return ret;
432 }
433
434 /*
435 * return: 0: no redundant snapshots, 1: rm process started, negative: error
436 */
437 int remove_redundant_snapshot(struct snapshot_list *sl)
438 {
439 int ret, i, interval;
440 struct snapshot *s;
441 unsigned missing = 0;
442
443 DSS_INFO_LOG("looking for intervals containing too many snapshots\n");
444 for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
445 unsigned keep = num_snapshots(interval);
446 unsigned num = sl->interval_count[interval];
447 struct snapshot *victim = NULL, *prev = NULL;
448 int64_t score = LONG_MAX;
449
450 if (keep >= num)
451 missing += keep - num;
452 DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
453 interval, keep, num, missing);
454 if (keep + missing >= num)
455 continue;
456 /* redundant snapshot in this interval, pick snapshot with lowest score */
457 FOR_EACH_SNAPSHOT(s, i, sl) {
458 int64_t this_score;
459
460 //DSS_DEBUG_LOG("checking %s\n", s->name);
461 if (s->interval > interval) {
462 prev = s;
463 continue;
464 }
465 if (s->interval < interval)
466 break;
467 if (!victim) {
468 victim = s;
469 prev = s;
470 continue;
471 }
472 assert(prev);
473 /* check if s is a better victim */
474 this_score = s->creation_time - prev->creation_time;
475 assert(this_score >= 0);
476 //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
477 if (this_score < score) {
478 score = this_score;
479 victim = s;
480 }
481 prev = s;
482 }
483 assert(victim);
484 if (conf.dry_run_given) {
485 dss_msg("%s would be removed (interval = %i)\n",
486 victim->name, victim->interval);
487 continue;
488 }
489 ret = remove_snapshot(victim);
490 return ret < 0? ret : 1;
491 }
492 return 0;
493 }
494
495 int remove_outdated_snapshot(struct snapshot_list *sl)
496 {
497 int i, ret;
498 struct snapshot *s;
499
500 DSS_INFO_LOG("looking for snapshots belonging to intervals greater than %d\n",
501 conf.num_intervals_arg);
502 FOR_EACH_SNAPSHOT(s, i, sl) {
503 if (s->interval <= conf.num_intervals_arg)
504 continue;
505 if (conf.dry_run_given) {
506 dss_msg("%s would be removed (interval = %i)\n",
507 s->name, s->interval);
508 continue;
509 }
510 ret = remove_snapshot(s);
511 if (ret < 0)
512 return ret;
513 return 1;
514 }
515 return 0;
516 }
517
518 int handle_rm_exit(int status)
519 {
520 int es, ret;
521
522 if (!WIFEXITED(status)) {
523 make_err_msg("rm process %d died involuntary", (int)rm_pid);
524 ret = -E_INVOLUNTARY_EXIT;
525 goto out;
526 }
527 es = WEXITSTATUS(status);
528 if (es) {
529 make_err_msg("rm process %d returned %d", (int)rm_pid, es);
530 ret = -E_BAD_EXIT_CODE;
531 goto out;
532 }
533 ret = 1;
534 rm_pid = 0;
535 out:
536 return ret;
537 }
538
539 int wait_for_rm_process(void)
540 {
541 int status, ret = wait_for_process(rm_pid, &status);
542
543 if (ret < 0)
544 return ret;
545 return handle_rm_exit(status);
546 }
547
548 void kill_process(pid_t pid)
549 {
550 if (!pid)
551 return;
552 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
553 kill(pid, SIGTERM);
554 }
555
556 int check_config(void)
557 {
558 if (conf.unit_interval_arg <= 0) {
559 make_err_msg("bad unit interval: %i", conf.unit_interval_arg);
560 return -E_INVALID_NUMBER;
561 }
562 DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
563 if (conf.num_intervals_arg <= 0) {
564 make_err_msg("bad number of intervals %i", conf.num_intervals_arg);
565 return -E_INVALID_NUMBER;
566 }
567 DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
568 return 1;
569 }
570
571 /* exits on errors */
572 void parse_config_file(int override)
573 {
574 int ret;
575 char *config_file;
576 struct stat statbuf;
577 char *old_logfile_arg = NULL;
578 int old_daemon_given = 0;
579
580 if (conf.config_file_given)
581 config_file = dss_strdup(conf.config_file_arg);
582 else {
583 char *home = get_homedir();
584 config_file = make_message("%s/.dssrc", home);
585 free(home);
586 }
587 if (override) { /* SIGHUP */
588 if (conf.logfile_given)
589 old_logfile_arg = dss_strdup(conf.logfile_arg);
590 old_daemon_given = conf.daemon_given;
591 }
592
593 ret = stat(config_file, &statbuf);
594 if (ret && conf.config_file_given) {
595 ret = -ERRNO_TO_DSS_ERROR(errno);
596 make_err_msg("failed to stat config file %s", config_file);
597 goto out;
598 }
599 if (!ret) {
600 struct cmdline_parser_params params = {
601 .override = override,
602 .initialize = 0,
603 .check_required = 1,
604 .check_ambiguity = 0
605 };
606 cmdline_parser_config_file(config_file, &conf, &params);
607 }
608 ret = check_config();
609 if (ret < 0)
610 goto out;
611 if (override) {
612 /* don't change daemon mode on SIGHUP */
613 conf.daemon_given = old_daemon_given;
614 close_log(logfile);
615 logfile = NULL;
616 if (conf.logfile_given)
617 free(old_logfile_arg);
618 else if (conf.daemon_given) { /* re-use old logfile */
619 conf.logfile_arg = old_logfile_arg;
620 conf.logfile_given = 1;
621 }
622 }
623 if (conf.logfile_given) {
624 logfile = open_log(conf.logfile_arg);
625 log_welcome(conf.loglevel_arg);
626 }
627 ret = dss_chdir(conf.dest_dir_arg);
628 out:
629 free(config_file);
630 if (ret >= 0)
631 return;
632 log_err_msg(EMERG, -ret);
633 exit(EXIT_FAILURE);
634 }
635
636 void handle_sighup(void)
637 {
638 DSS_NOTICE_LOG("SIGHUP\n");
639 parse_config_file(1);
640 }
641
642 int rename_incomplete_snapshot(int64_t start)
643 {
644 char *old_name, *new_name;
645 int ret;
646
647 ret = complete_name(start, get_current_time(), &new_name);
648 if (ret < 0)
649 return ret;
650 old_name = incomplete_name(start);
651 ret = dss_rename(old_name, new_name);
652 if (ret >= 0)
653 DSS_NOTICE_LOG("%s -> %s\n", old_name, new_name);
654 free(old_name);
655 free(new_name);
656 return ret;
657 }
658
659 int handle_rsync_exit(int status)
660 {
661 int es, ret;
662
663 if (!WIFEXITED(status)) {
664 make_err_msg("rsync process %d died involuntary", (int)rsync_pid);
665 ret = -E_INVOLUNTARY_EXIT;
666 goto out;
667 }
668 es = WEXITSTATUS(status);
669 if (es != 0 && es != 23 && es != 24) {
670 make_err_msg("rsync process %d returned %d", (int)rsync_pid, es);
671 ret = -E_BAD_EXIT_CODE;
672 goto out;
673 }
674 ret = rename_incomplete_snapshot(current_snapshot_creation_time);
675 out:
676 rsync_pid = 0;
677 current_snapshot_creation_time = 0;
678 rsync_stopped = 0;
679 return ret;
680 }
681
682 int get_newest_complete(const char *dirname, void *private)
683 {
684 struct edge_snapshot_data *esd = private;
685 struct snapshot s;
686 int ret = is_snapshot(dirname, esd->now, &s);
687
688 if (ret <= 0)
689 return 1;
690 if (s.flags != SS_COMPLETE) /* incomplete or being deleted */
691 return 1;
692 if (s.creation_time < esd->snap.creation_time)
693 return 1;
694 free(esd->snap.name);
695 esd->snap = s;
696 return 1;
697 }
698
699 __malloc char *name_of_newest_complete_snapshot(void)
700 {
701 struct edge_snapshot_data esd = {
702 .now = get_current_time(),
703 .snap = {.creation_time = -1}
704 };
705 for_each_subdir(get_newest_complete, &esd);
706 return esd.snap.name;
707 }
708
709 void create_rsync_argv(char ***argv, int64_t *num)
710 {
711 char *logname, *newest = name_of_newest_complete_snapshot();
712 int i = 0, j;
713
714 *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
715 (*argv)[i++] = dss_strdup("rsync");
716 (*argv)[i++] = dss_strdup("-aq");
717 (*argv)[i++] = dss_strdup("--delete");
718 for (j = 0; j < conf.rsync_option_given; j++)
719 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
720 if (newest) {
721 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
722 (*argv)[i++] = make_message("--link-dest=../%s", newest);
723 free(newest);
724 } else
725 DSS_INFO_LOG("no previous snapshot found\n");
726 if (conf.exclude_patterns_given) {
727 (*argv)[i++] = dss_strdup("--exclude-from");
728 (*argv)[i++] = dss_strdup(conf.exclude_patterns_arg);
729
730 }
731 logname = dss_logname();
732 if (conf.remote_user_given && !strcmp(conf.remote_user_arg, logname))
733 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
734 else
735 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
736 conf.remote_user_arg : logname,
737 conf.remote_host_arg, conf.source_dir_arg);
738 free(logname);
739 *num = get_current_time();
740 (*argv)[i++] = incomplete_name(*num);
741 (*argv)[i++] = NULL;
742 for (j = 0; j < i; j++)
743 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
744 }
745
746 void free_rsync_argv(char **argv)
747 {
748 int i;
749 for (i = 0; argv[i]; i++)
750 free(argv[i]);
751 free(argv);
752 }
753
754 int create_snapshot(char **argv)
755 {
756 int fds[3] = {0, 0, 0};
757 char *name = incomplete_name(current_snapshot_creation_time);
758
759 DSS_NOTICE_LOG("creating new snapshot %s\n", name);
760 free(name);
761 return dss_exec(&rsync_pid, argv[0], argv, fds);
762 }
763
764 void compute_next_snapshot_time(struct snapshot_list *sl)
765 {
766 struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
767 tmp, diff;
768 int64_t x = 0;
769 unsigned wanted = num_snapshots(0), num_complete_snapshots = 0;
770 int i, ret;
771 struct snapshot *s;
772
773 gettimeofday(&now, NULL);
774 FOR_EACH_SNAPSHOT(s, i, sl) {
775 if (!(s->flags & SS_COMPLETE))
776 continue;
777 num_complete_snapshots++;
778 x += s->completion_time - s->creation_time;
779 }
780 assert(x >= 0);
781 if (num_complete_snapshots)
782 x /= num_complete_snapshots; /* avg time to create one snapshot */
783 x *= wanted; /* time to create all snapshots in interval 0 */
784 tmp.tv_sec = x;
785 tmp.tv_usec = 0;
786 ret = tv_diff(&unit_interval, &tmp, &diff); /* time between creation */
787 if (ret < 0) {
788 next_snapshot_time = now;
789 return;
790 }
791 tv_divide(wanted, &diff, &tmp);
792 tv_add(&now, &tmp, &next_snapshot_time);
793 }
794
795 void handle_signal(void)
796 {
797 int sig, ret = next_signal();
798
799 if (ret <= 0)
800 goto out;
801 sig = ret;
802 switch (sig) {
803 int status;
804 pid_t pid;
805 case SIGINT:
806 case SIGTERM:
807 restart_rsync_process();
808 kill_process(rsync_pid);
809 kill_process(rm_pid);
810 exit(EXIT_FAILURE);
811 case SIGHUP:
812 handle_sighup();
813 ret = 1;
814 break;
815 case SIGCHLD:
816 ret = reap_child(&pid, &status);
817 if (ret <= 0)
818 break;
819 assert(pid == rsync_pid || pid == rm_pid);
820 if (pid == rsync_pid)
821 ret = handle_rsync_exit(status);
822 else
823 ret = handle_rm_exit(status);
824 }
825 out:
826 if (ret < 0)
827 log_err_msg(ERROR, -ret);
828 }
829
830 int get_oldest(const char *dirname, void *private)
831 {
832 struct edge_snapshot_data *esd = private;
833 struct snapshot s;
834 int ret = is_snapshot(dirname, esd->now, &s);
835
836 if (ret <= 0)
837 return 1;
838 if (s.creation_time > esd->snap.creation_time)
839 return 1;
840 free(esd->snap.name);
841 esd->snap = s;
842 return 1;
843 }
844
845 int remove_oldest_snapshot()
846 {
847 int ret;
848 struct edge_snapshot_data esd = {
849 .now = get_current_time(),
850 .snap = {.creation_time = LLONG_MAX}
851 };
852 for_each_subdir(get_oldest, &esd);
853 if (!esd.snap.name) /* no snapshot found */
854 return 0;
855 DSS_INFO_LOG("oldest snapshot: %s\n", esd.snap.name);
856 ret = 0;
857 if (esd.snap.creation_time == current_snapshot_creation_time)
858 goto out; /* do not remove the snapshot currently being created */
859 ret = remove_snapshot(&esd.snap);
860 out:
861 free(esd.snap.name);
862 return ret;
863 }
864
865 /* TODO: Also consider number of inodes. */
866 int disk_space_low(void)
867 {
868 struct disk_space ds;
869 int ret = get_disk_space(".", &ds);
870
871 if (ret < 0)
872 return ret;
873 if (conf.min_free_mb_arg)
874 if (ds.free_mb < conf.min_free_mb_arg)
875 return 1;
876 if (conf.min_free_percent_arg)
877 if (ds.percent_free < conf.min_free_percent_arg)
878 return 1;
879 return 0;
880 }
881
882 int try_to_free_disk_space(int low_disk_space, struct snapshot_list *sl)
883 {
884 int ret;
885
886 ret = remove_outdated_snapshot(sl);
887 if (ret) /* error, or we are removing something */
888 return ret;
889 /* no outdated snapshot */
890 ret = remove_redundant_snapshot(sl);
891 if (ret)
892 return ret;
893 if (!low_disk_space)
894 return 0;
895 DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
896 ret = remove_oldest_snapshot();
897 if (ret)
898 return ret;
899 make_err_msg("uhuhu: not enough disk space for a single snapshot");
900 return -ENOSPC;
901 }
902
903 int select_loop(void)
904 {
905 int ret;
906 struct timeval tv = {.tv_sec = 0, .tv_usec = 0};
907 struct snapshot_list sl = {.num_snapshots = 0};
908
909 for (;;) {
910 struct timeval now, *tvp = &tv;
911 fd_set rfds;
912 int low_disk_space;
913 char **rsync_argv;
914
915 free_snapshot_list(&sl);
916 get_snapshot_list(&sl);
917 compute_next_snapshot_time(&sl);
918 FD_ZERO(&rfds);
919 FD_SET(signal_pipe, &rfds);
920 if (rsync_pid)
921 tv.tv_sec = 60;
922 else if (rm_pid)
923 tvp = NULL;
924 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
925 if (ret < 0)
926 return ret;
927 if (FD_ISSET(signal_pipe, &rfds)) {
928 handle_signal();
929 continue;
930 }
931 if (rm_pid)
932 continue;
933 ret = disk_space_low();
934 if (ret < 0)
935 break;
936 low_disk_space = ret;
937 if (low_disk_space)
938 stop_rsync_process();
939 ret = try_to_free_disk_space(low_disk_space, &sl);
940 if (ret < 0)
941 break;
942 if (rm_pid)
943 continue;
944 if (rsync_pid) {
945 restart_rsync_process();
946 continue;
947 }
948 /* neither rsync nor rm are running. Start rsync? */
949 gettimeofday(&now, NULL);
950 if (tv_diff(&next_snapshot_time, &now, &tv) > 0)
951 continue;
952 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
953 ret = create_snapshot(rsync_argv);
954 free_rsync_argv(rsync_argv);
955 if (ret < 0)
956 break;
957 }
958 free_snapshot_list(&sl);
959 return ret;
960 }
961
962 int com_run(void)
963 {
964 int ret;
965
966 if (conf.dry_run_given) {
967 make_err_msg("dry_run not supported by this command");
968 return -E_SYNTAX;
969 }
970 ret = install_sighandler(SIGHUP);
971 if (ret < 0)
972 return ret;
973 return select_loop();
974 }
975
976 void log_disk_space(struct disk_space *ds)
977 {
978 DSS_INFO_LOG("free: %uM/%uM (%u%%), %u%% inodes unused\n",
979 ds->free_mb, ds->total_mb, ds->percent_free,
980 ds->percent_free_inodes);
981 }
982
983 int com_prune(void)
984 {
985 int ret;
986 struct snapshot_list sl;
987 struct disk_space ds;
988
989 ret = get_disk_space(".", &ds);
990 if (ret < 0)
991 return ret;
992 log_disk_space(&ds);
993 for (;;) {
994 get_snapshot_list(&sl);
995 ret = remove_outdated_snapshot(&sl);
996 free_snapshot_list(&sl);
997 if (ret < 0)
998 return ret;
999 if (!ret)
1000 break;
1001 ret = wait_for_rm_process();
1002 if (ret < 0)
1003 goto out;
1004 }
1005 for (;;) {
1006 get_snapshot_list(&sl);
1007 ret = remove_redundant_snapshot(&sl);
1008 free_snapshot_list(&sl);
1009 if (ret < 0)
1010 return ret;
1011 if (!ret)
1012 break;
1013 ret = wait_for_rm_process();
1014 if (ret < 0)
1015 goto out;
1016 }
1017 return 1;
1018 out:
1019 return ret;
1020 }
1021
1022 int com_create(void)
1023 {
1024 int ret, status;
1025 char **rsync_argv;
1026
1027 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1028 if (conf.dry_run_given) {
1029 int i;
1030 char *msg = NULL;
1031 for (i = 0; rsync_argv[i]; i++) {
1032 char *tmp = msg;
1033 msg = make_message("%s%s%s", tmp? tmp : "",
1034 tmp? " " : "", rsync_argv[i]);
1035 free(tmp);
1036 }
1037 dss_msg("%s\n", msg);
1038 free(msg);
1039 return 1;
1040 }
1041 ret = create_snapshot(rsync_argv);
1042 if (ret < 0)
1043 goto out;
1044 ret = wait_for_process(rsync_pid, &status);
1045 if (ret < 0)
1046 goto out;
1047 ret = handle_rsync_exit(status);
1048 out:
1049 free_rsync_argv(rsync_argv);
1050 return ret;
1051 }
1052
1053 int com_ls(void)
1054 {
1055 int i;
1056 struct snapshot_list sl;
1057 struct snapshot *s;
1058 get_snapshot_list(&sl);
1059 FOR_EACH_SNAPSHOT(s, i, &sl)
1060 dss_msg("%u\t%s\n", s->interval, s->name);
1061 free_snapshot_list(&sl);
1062 return 1;
1063 }
1064
1065 __noreturn void clean_exit(int status)
1066 {
1067 free(dss_error_txt);
1068 exit(status);
1069 }
1070 static void setup_signal_handling(void)
1071 {
1072 int ret;
1073
1074 DSS_INFO_LOG("setting up signal handlers\n");
1075 signal_pipe = signal_init(); /* always successful */
1076 ret = install_sighandler(SIGINT);
1077 if (ret < 0)
1078 goto err;
1079 ret = install_sighandler(SIGTERM);
1080 if (ret < 0)
1081 goto err;
1082 ret = install_sighandler(SIGCHLD);
1083 if (ret < 0)
1084 goto err;
1085 return;
1086 err:
1087 DSS_EMERG_LOG("could not install signal handlers\n");
1088 exit(EXIT_FAILURE);
1089 }
1090
1091 int main(int argc, char **argv)
1092 {
1093 int ret;
1094 struct cmdline_parser_params params = {
1095 .override = 0,
1096 .initialize = 1,
1097 .check_required = 0,
1098 .check_ambiguity = 0
1099 };
1100
1101 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1102 parse_config_file(0);
1103
1104 if (conf.daemon_given)
1105 daemon_init();
1106 setup_signal_handling();
1107 ret = call_command_handler();
1108 if (ret < 0)
1109 log_err_msg(EMERG, -ret);
1110 clean_exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1111 }