Merge branch 't/rename_system_files'
[dss.git] / dss.c
1 /*
2 * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6 #include <string.h>
7 #include <stdlib.h>
8 #include <stdarg.h>
9 #include <assert.h>
10 #include <errno.h>
11 #include <sys/types.h>
12 #include <signal.h>
13 #include <ctype.h>
14 #include <sys/stat.h>
15 #include <unistd.h>
16 #include <inttypes.h>
17 #include <sys/time.h>
18 #include <time.h>
19 #include <sys/wait.h>
20 #include <fnmatch.h>
21 #include <limits.h>
22
23
24 #include "gcc-compat.h"
25 #include "cmdline.h"
26 #include "log.h"
27 #include "str.h"
28 #include "err.h"
29 #include "file.h"
30 #include "exec.h"
31 #include "daemon.h"
32 #include "sig.h"
33 #include "df.h"
34 #include "tv.h"
35 #include "snap.h"
36 #include "ipc.h"
37
38 /** Command line and config file options. */
39 static struct gengetopt_args_info conf;
40 /** Non-NULL if we log to a file. */
41 static FILE *logfile;
42 /** The read end of the signal pipe */
43 static int signal_pipe;
44 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
45 static pid_t create_pid;
46 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
47 static int create_process_stopped;
48 /** Process id of current pre-remove/rm/post-remove process. */
49 static pid_t remove_pid;
50 /** When the next snapshot is due. */
51 static int64_t next_snapshot_time;
52 /** When to try to remove something. */
53 static struct timeval next_removal_check;
54 /** Creation time of the snapshot currently being created. */
55 static int64_t current_snapshot_creation_time;
56 /** The snapshot currently being removed. */
57 struct snapshot *snapshot_currently_being_removed;
58 /** Needed by the post-create hook. */
59 static char *path_to_last_complete_snapshot;
60 static char *name_of_reference_snapshot;
61 /** \sa \ref snap.h for details. */
62 enum hook_status snapshot_creation_status;
63 /** \sa \ref snap.h for details. */
64 enum hook_status snapshot_removal_status;
65
66
67 DEFINE_DSS_ERRLIST;
68 static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
69
70 /* may be called with ds == NULL. */
71 static int disk_space_low(struct disk_space *ds)
72 {
73 struct disk_space ds_struct;
74
75 if (!ds) {
76 int ret = get_disk_space(".", &ds_struct);
77 if (ret < 0)
78 return ret;
79 ds = &ds_struct;
80 }
81 if (conf.min_free_mb_arg)
82 if (ds->free_mb < conf.min_free_mb_arg)
83 return 1;
84 if (conf.min_free_percent_arg)
85 if (ds->percent_free < conf.min_free_percent_arg)
86 return 1;
87 if (conf.min_free_percent_inodes_arg)
88 if (ds->percent_free_inodes < conf.min_free_percent_inodes_arg)
89 return 1;
90 return 0;
91 }
92
93 static void dump_dss_config(const char *msg)
94 {
95 const char dash[] = "-----------------------------";
96 int ret;
97 FILE *log = logfile? logfile : stderr;
98 struct disk_space ds;
99 int64_t now = get_current_time();
100
101 if (conf.loglevel_arg > INFO)
102 return;
103
104 fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
105 fprintf(log, "\n*** disk space ***\n\n");
106 ret = get_disk_space(".", &ds);
107 if (ret >= 0) {
108 DSS_INFO_LOG(("disk space low: %s\n", disk_space_low(&ds)?
109 "yes" : "no"));
110 log_disk_space(&ds);
111 } else
112 DSS_ERROR_LOG(("can not get free disk space: %s\n",
113 dss_strerror(-ret)));
114
115 /* we continue on errors from get_disk_space */
116
117 fprintf(log, "\n*** command line and config file options ***\n\n");
118 cmdline_parser_dump(log, &conf);
119 fprintf(log, "\n*** internal state ***\n\n");
120 fprintf(log,
121 "pid: %d\n"
122 "logile: %s\n"
123 "snapshot_currently_being_removed: %s\n"
124 "path_to_last_complete_snapshot: %s\n"
125 "reference_snapshot: %s\n"
126 "snapshot_creation_status: %s\n"
127 "snapshot_removal_status: %s\n"
128 ,
129 (int) getpid(),
130 logfile? conf.logfile_arg : "stderr",
131 snapshot_currently_being_removed?
132 snapshot_currently_being_removed->name : "(none)",
133 path_to_last_complete_snapshot?
134 path_to_last_complete_snapshot : "(none)",
135 name_of_reference_snapshot?
136 name_of_reference_snapshot : "(none)",
137 hook_status_description[snapshot_creation_status],
138 hook_status_description[snapshot_removal_status]
139 );
140 if (create_pid != 0)
141 fprintf(log,
142 "create_pid: %" PRId32 "\n"
143 "create process is %sstopped\n"
144 ,
145 create_pid,
146 create_process_stopped? "" : "not "
147 );
148 if (remove_pid != 0)
149 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
150 if (next_snapshot_time != 0)
151 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
152 next_snapshot_time - now);
153 if (current_snapshot_creation_time != 0)
154 fprintf(log, "current_snapshot_creation_time: %"
155 PRId64 " (%" PRId64 " seconds ago)\n",
156 current_snapshot_creation_time,
157 now - current_snapshot_creation_time
158 );
159 if (next_removal_check.tv_sec != 0) {
160 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
161 (long long unsigned)next_removal_check.tv_sec,
162 now - (long long unsigned)next_removal_check.tv_sec
163 );
164
165 }
166 fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
167 }
168
169 /* a litte cpp magic helps to DRY */
170 #define COMMANDS \
171 COMMAND(ls) \
172 COMMAND(create) \
173 COMMAND(prune) \
174 COMMAND(run) \
175 COMMAND(kill) \
176 COMMAND(reload) \
177
178 #define COMMAND(x) static int com_ ##x(void);
179 COMMANDS
180 #undef COMMAND
181 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
182 static int call_command_handler(void)
183 {
184 COMMANDS
185 DSS_EMERG_LOG(("BUG: did not find command handler\n"));
186 return -E_BUG;
187 }
188 #undef COMMAND
189 #undef COMMANDS
190
191 static int loglevel = -1;
192 static const char *location_file = NULL;
193 static int location_line = -1;
194 static const char *location_func = NULL;
195
196 void dss_log_set_params(int ll, const char *file, int line, const char *func)
197 {
198 loglevel = ll;
199 location_file = file;
200 location_line = line;
201 location_func = func;
202 }
203
204 /**
205 * The log function of dss.
206 *
207 * \param ll Loglevel.
208 * \param fml Usual format string.
209 *
210 * All DSS_XXX_LOG() macros use this function.
211 */
212 __printf_1_2 void dss_log(const char* fmt,...)
213 {
214 va_list argp;
215 FILE *outfd;
216 struct tm *tm;
217 time_t t1;
218 char str[255] = "";
219
220 if (loglevel < conf.loglevel_arg)
221 return;
222 outfd = logfile? logfile : stderr;
223 time(&t1);
224 tm = localtime(&t1);
225 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
226 fprintf(outfd, "%s ", str);
227 if (conf.loglevel_arg <= INFO)
228 fprintf(outfd, "%i: ", loglevel);
229 #ifdef DSS_NO_FUNC_NAMES
230 fprintf(outfd, "%s:%d: ", location_file, location_line);
231 #else
232 fprintf(outfd, "%s: ", location_func);
233 #endif
234 va_start(argp, fmt);
235 vfprintf(outfd, fmt, argp);
236 va_end(argp);
237 }
238
239 /**
240 * Print a message either to stdout or to the log file.
241 */
242 static __printf_1_2 void dss_msg(const char* fmt,...)
243 {
244 FILE *outfd = conf.daemon_given? logfile : stdout;
245 va_list argp;
246 va_start(argp, fmt);
247 vfprintf(outfd, fmt, argp);
248 va_end(argp);
249 }
250
251 static char *get_config_file_name(void)
252 {
253 char *home, *config_file;
254
255 if (conf.config_file_given)
256 return dss_strdup(conf.config_file_arg);
257 home = get_homedir();
258 config_file = make_message("%s/.dssrc", home);
259 free(home);
260 return config_file;
261 }
262
263 static int send_signal(int sig)
264 {
265 pid_t pid;
266 char *config_file = get_config_file_name();
267 int ret = get_dss_pid(config_file, &pid);
268
269 free(config_file);
270 if (ret < 0)
271 return ret;
272 if (conf.dry_run_given) {
273 dss_msg("%d\n", (int)pid);
274 return 0;
275 }
276 ret = kill(pid, sig);
277 if (ret < 0)
278 return -ERRNO_TO_DSS_ERROR(errno);
279 return 1;
280 }
281
282 static int com_kill(void)
283 {
284 return send_signal(SIGTERM);
285 }
286
287 static int com_reload(void)
288 {
289 return send_signal(SIGHUP);
290 }
291
292 static void dss_get_snapshot_list(struct snapshot_list *sl)
293 {
294 get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
295 }
296
297 static int64_t compute_next_snapshot_time(void)
298 {
299 int64_t x = 0, now = get_current_time(), unit_interval
300 = 24 * 3600 * conf.unit_interval_arg, ret;
301 unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
302 num_complete_snapshots = 0;
303 int i;
304 struct snapshot *s = NULL;
305 struct snapshot_list sl;
306
307 dss_get_snapshot_list(&sl);
308 FOR_EACH_SNAPSHOT(s, i, &sl) {
309 if (!(s->flags & SS_COMPLETE))
310 continue;
311 num_complete_snapshots++;
312 x += s->completion_time - s->creation_time;
313 }
314 assert(x >= 0);
315
316 ret = now;
317 if (num_complete_snapshots == 0)
318 goto out;
319 x /= num_complete_snapshots; /* avg time to create one snapshot */
320 if (unit_interval < x * wanted) /* oops, no sleep at all */
321 goto out;
322 ret = s->completion_time + unit_interval / wanted - x;
323 out:
324 free_snapshot_list(&sl);
325 return ret;
326 }
327
328 static inline void invalidate_next_snapshot_time(void)
329 {
330 next_snapshot_time = 0;
331 }
332
333 static inline int next_snapshot_time_is_valid(void)
334 {
335 return next_snapshot_time != 0;
336 }
337
338 static int next_snapshot_is_due(void)
339 {
340 int64_t now = get_current_time();
341
342 if (!next_snapshot_time_is_valid())
343 next_snapshot_time = compute_next_snapshot_time();
344 if (next_snapshot_time <= now) {
345 DSS_DEBUG_LOG(("next snapshot: now\n"));
346 return 1;
347 }
348 DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
349 next_snapshot_time - now));
350 return 0;
351 }
352
353 static void pre_create_hook(void)
354 {
355 assert(snapshot_creation_status == HS_READY);
356 /* make sure that the next snapshot time will be recomputed */
357 invalidate_next_snapshot_time();
358 DSS_DEBUG_LOG(("executing %s\n", conf.pre_create_hook_arg));
359 dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
360 snapshot_creation_status = HS_PRE_RUNNING;
361 }
362
363 static void pre_remove_hook(struct snapshot *s, const char *why)
364 {
365 char *cmd;
366
367 if (!s)
368 return;
369 DSS_DEBUG_LOG(("%s snapshot %s\n", why, s->name));
370 assert(snapshot_removal_status == HS_READY);
371 assert(remove_pid == 0);
372 assert(!snapshot_currently_being_removed);
373
374 snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
375 *snapshot_currently_being_removed = *s;
376 snapshot_currently_being_removed->name = dss_strdup(s->name);
377
378 cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
379 conf.dest_dir_arg, s->name);
380 DSS_DEBUG_LOG(("executing %s\n", cmd));
381 dss_exec_cmdline_pid(&remove_pid, cmd);
382 free(cmd);
383 snapshot_removal_status = HS_PRE_RUNNING;
384 }
385
386 static int exec_rm(void)
387 {
388 struct snapshot *s = snapshot_currently_being_removed;
389 char *new_name = being_deleted_name(s);
390 char *argv[4];
391 int ret;
392
393 argv[0] = "rm";
394 argv[1] = "-rf";
395 argv[2] = new_name;
396 argv[3] = NULL;
397
398 assert(snapshot_removal_status == HS_PRE_SUCCESS);
399 assert(remove_pid == 0);
400
401 DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
402 ret = dss_rename(s->name, new_name);
403 if (ret < 0)
404 goto out;
405 dss_exec(&remove_pid, argv[0], argv);
406 snapshot_removal_status = HS_RUNNING;
407 out:
408 free(new_name);
409 return ret;
410 }
411
412 static int snapshot_is_being_created(struct snapshot *s)
413 {
414 return s->creation_time == current_snapshot_creation_time;
415 }
416
417 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
418 {
419 struct snapshot *s;
420 int i;
421
422 DSS_DEBUG_LOG(("looking for orphaned snapshots\n"));
423 FOR_EACH_SNAPSHOT(s, i, sl) {
424 if (snapshot_is_being_created(s))
425 continue;
426 /*
427 * We know that no rm is currently running, so if s is marked
428 * as being deleted, a previously started rm must have failed.
429 */
430 if (s->flags & SS_BEING_DELETED)
431 return s;
432
433 if (s->flags & SS_COMPLETE) /* good snapshot */
434 continue;
435 /*
436 * This snapshot is incomplete and it is not the snapshot
437 * currently being created. However, we must not remove it if
438 * rsync is about to be restarted. As only the newest snapshot
439 * can be restarted, this snapshot is orphaned if it is not the
440 * newest snapshot or if we are not about to restart rsync.
441 */
442 if (get_newest_snapshot(sl) != s)
443 return s;
444 if (snapshot_creation_status != HS_NEEDS_RESTART)
445 return s;
446 }
447 /* no orphaned snapshots */
448 return NULL;
449 }
450
451 static int is_reference_snapshot(struct snapshot *s)
452 {
453 if (!name_of_reference_snapshot)
454 return 0;
455 return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
456 }
457
458 /*
459 * return: 0: no redundant snapshots, 1: rm process started, negative: error
460 */
461 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
462 {
463 int i, interval;
464 struct snapshot *s;
465 unsigned missing = 0;
466
467 DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
468 for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
469 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
470 unsigned num = sl->interval_count[interval];
471 struct snapshot *victim = NULL, *prev = NULL;
472 int64_t score = LONG_MAX;
473
474 if (keep >= num)
475 missing += keep - num;
476 if (keep + missing >= num)
477 continue;
478 /* redundant snapshot in this interval, pick snapshot with lowest score */
479 FOR_EACH_SNAPSHOT(s, i, sl) {
480 int64_t this_score;
481
482 if (snapshot_is_being_created(s))
483 continue;
484 if (is_reference_snapshot(s))
485 continue;
486 if (s->interval > interval) {
487 prev = s;
488 continue;
489 }
490 if (s->interval < interval)
491 break;
492 if (!victim) {
493 victim = s;
494 prev = s;
495 continue;
496 }
497 assert(prev);
498 /* check if s is a better victim */
499 this_score = s->creation_time - prev->creation_time;
500 assert(this_score >= 0);
501 if (this_score < score) {
502 score = this_score;
503 victim = s;
504 }
505 prev = s;
506 }
507 assert(victim);
508 return victim;
509 }
510 return NULL;
511 }
512
513 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
514 {
515 int i;
516 struct snapshot *s;
517
518 DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
519 conf.num_intervals_arg));
520 FOR_EACH_SNAPSHOT(s, i, sl) {
521 if (snapshot_is_being_created(s))
522 continue;
523 if (is_reference_snapshot(s))
524 continue;
525 if (s->interval < conf.num_intervals_arg)
526 continue;
527 return s;
528 }
529 return NULL;
530 }
531
532 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
533 {
534 int i;
535 struct snapshot *s;
536 FOR_EACH_SNAPSHOT(s, i, sl) {
537 if (snapshot_is_being_created(s))
538 continue;
539 if (is_reference_snapshot(s))
540 continue;
541 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
542 return s;
543 }
544 return NULL;
545 }
546
547 static int rename_incomplete_snapshot(int64_t start)
548 {
549 char *old_name;
550 int ret;
551
552 free(path_to_last_complete_snapshot);
553 ret = complete_name(start, get_current_time(),
554 &path_to_last_complete_snapshot);
555 if (ret < 0)
556 return ret;
557 old_name = incomplete_name(start);
558 ret = dss_rename(old_name, path_to_last_complete_snapshot);
559 if (ret >= 0)
560 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
561 path_to_last_complete_snapshot));
562 free(old_name);
563 return ret;
564 }
565
566 static int try_to_free_disk_space(void)
567 {
568 int ret;
569 struct snapshot_list sl;
570 struct snapshot *victim;
571 struct timeval now;
572 const char *why;
573 int low_disk_space;
574
575 ret = disk_space_low(NULL);
576 if (ret < 0)
577 return ret;
578 low_disk_space = ret;
579 gettimeofday(&now, NULL);
580 if (tv_diff(&next_removal_check, &now, NULL) > 0)
581 return 0;
582 if (!low_disk_space) {
583 if (conf.keep_redundant_given)
584 return 0;
585 if (snapshot_creation_status != HS_READY)
586 return 0;
587 if (next_snapshot_is_due())
588 return 0;
589 }
590 dss_get_snapshot_list(&sl);
591 ret = 0;
592 if (!low_disk_space && sl.num_snapshots <= 1)
593 goto out;
594 why = "outdated";
595 victim = find_outdated_snapshot(&sl);
596 if (victim)
597 goto remove;
598 why = "redundant";
599 victim = find_redundant_snapshot(&sl);
600 if (victim)
601 goto remove;
602 /* try harder only if disk space is low */
603 if (!low_disk_space)
604 goto out;
605 why = "orphaned";
606 victim = find_orphaned_snapshot(&sl);
607 if (victim)
608 goto remove;
609 DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
610 victim = find_oldest_removable_snapshot(&sl);
611 if (victim)
612 goto remove;
613 DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
614 ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
615 goto out;
616 remove:
617 pre_remove_hook(victim, why);
618 out:
619 free_snapshot_list(&sl);
620 return ret;
621 }
622
623 static void post_create_hook(void)
624 {
625 char *cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
626 conf.dest_dir_arg, path_to_last_complete_snapshot);
627 DSS_NOTICE_LOG(("executing %s\n", cmd));
628 dss_exec_cmdline_pid(&create_pid, cmd);
629 free(cmd);
630 snapshot_creation_status = HS_POST_RUNNING;
631 }
632
633 static void post_remove_hook(void)
634 {
635 char *cmd;
636 struct snapshot *s = snapshot_currently_being_removed;
637
638 assert(s);
639
640 cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
641 conf.dest_dir_arg, s->name);
642 DSS_NOTICE_LOG(("executing %s\n", cmd));
643 dss_exec_cmdline_pid(&remove_pid, cmd);
644 free(cmd);
645 snapshot_removal_status = HS_POST_RUNNING;
646 }
647
648 static void dss_kill(pid_t pid, int sig, const char *msg)
649 {
650 const char *signame, *process_name;
651
652 if (pid == 0)
653 return;
654 switch (sig) {
655 case SIGTERM: signame = "TERM"; break;
656 case SIGSTOP: signame = "STOP"; break;
657 case SIGCONT: signame = "CONT"; break;
658 default: signame = "????";
659 }
660
661 if (pid == create_pid)
662 process_name = "create";
663 else if (pid == remove_pid)
664 process_name = "remove";
665 else process_name = "??????";
666
667 if (msg)
668 DSS_INFO_LOG(("%s\n", msg));
669 DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
670 sig, signame, (int)pid, process_name));
671 if (kill(pid, sig) >= 0)
672 return;
673 DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
674 sig, signame, (int)pid, process_name));
675 }
676
677 static void stop_create_process(void)
678 {
679 if (create_process_stopped)
680 return;
681 dss_kill(create_pid, SIGSTOP, "suspending create process");
682 create_process_stopped = 1;
683 }
684
685 static void restart_create_process(void)
686 {
687 if (!create_process_stopped)
688 return;
689 dss_kill(create_pid, SIGCONT, "resuming create process");
690 create_process_stopped = 0;
691 }
692
693 /**
694 * Print a log message about the exit status of a child.
695 */
696 static void log_termination_msg(pid_t pid, int status)
697 {
698 if (WIFEXITED(status))
699 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
700 WEXITSTATUS(status)));
701 else if (WIFSIGNALED(status))
702 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
703 WTERMSIG(status)));
704 else
705 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
706 }
707
708 static int wait_for_process(pid_t pid, int *status)
709 {
710 int ret;
711
712 DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
713 for (;;) {
714 fd_set rfds;
715
716 FD_ZERO(&rfds);
717 FD_SET(signal_pipe, &rfds);
718 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
719 if (ret < 0)
720 break;
721 ret = next_signal();
722 if (!ret)
723 continue;
724 if (ret == SIGCHLD) {
725 ret = waitpid(pid, status, 0);
726 if (ret >= 0)
727 break;
728 if (errno != EINTR) { /* error */
729 ret = -ERRNO_TO_DSS_ERROR(errno);
730 break;
731 }
732 }
733 /* SIGINT or SIGTERM */
734 dss_kill(pid, SIGTERM, "killing child process");
735 }
736 if (ret < 0)
737 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
738 else
739 log_termination_msg(pid, *status);
740 return ret;
741 }
742
743 static void handle_pre_remove_exit(int status)
744 {
745 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
746 snapshot_removal_status = HS_READY;
747 gettimeofday(&next_removal_check, NULL);
748 next_removal_check.tv_sec += 60;
749 return;
750 }
751 snapshot_removal_status = HS_PRE_SUCCESS;
752 }
753
754 static int handle_rm_exit(int status)
755 {
756 if (!WIFEXITED(status)) {
757 snapshot_removal_status = HS_READY;
758 return -E_INVOLUNTARY_EXIT;
759 }
760 if (WEXITSTATUS(status)) {
761 snapshot_removal_status = HS_READY;
762 return -E_BAD_EXIT_CODE;
763 }
764 snapshot_removal_status = HS_SUCCESS;
765 return 1;
766 }
767
768 static void handle_post_remove_exit(void)
769 {
770 snapshot_removal_status = HS_READY;
771 }
772
773 static int handle_remove_exit(int status)
774 {
775 int ret;
776 struct snapshot *s = snapshot_currently_being_removed;
777
778 assert(s);
779 switch (snapshot_removal_status) {
780 case HS_PRE_RUNNING:
781 handle_pre_remove_exit(status);
782 ret = 1;
783 break;
784 case HS_RUNNING:
785 ret = handle_rm_exit(status);
786 break;
787 case HS_POST_RUNNING:
788 handle_post_remove_exit();
789 ret = 1;
790 break;
791 default:
792 ret = -E_BUG;
793 }
794 if (snapshot_removal_status == HS_READY) {
795 free(s->name);
796 free(s);
797 snapshot_currently_being_removed = NULL;
798 }
799 remove_pid = 0;
800 return ret;
801 }
802
803 static int wait_for_remove_process(void)
804 {
805 int status, ret;
806
807 assert(remove_pid);
808 assert(
809 snapshot_removal_status == HS_PRE_RUNNING ||
810 snapshot_removal_status == HS_RUNNING ||
811 snapshot_removal_status == HS_POST_RUNNING
812 );
813 ret = wait_for_process(remove_pid, &status);
814 if (ret < 0)
815 return ret;
816 return handle_remove_exit(status);
817 }
818
819 static int handle_rsync_exit(int status)
820 {
821 int es, ret;
822
823 if (!WIFEXITED(status)) {
824 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
825 ret = -E_INVOLUNTARY_EXIT;
826 snapshot_creation_status = HS_READY;
827 goto out;
828 }
829 es = WEXITSTATUS(status);
830 /*
831 * Restart rsync on non-fatal errors:
832 * 12: Error in rsync protocol data stream
833 * 13: Errors with program diagnostics
834 */
835 if (es == 12 || es == 13) {
836 DSS_WARNING_LOG(("rsync process %d returned %d -- restarting\n",
837 (int)create_pid, es));
838 snapshot_creation_status = HS_NEEDS_RESTART;
839 next_snapshot_time = get_current_time() + 60;
840 ret = 1;
841 goto out;
842 }
843 if (es != 0 && es != 23 && es != 24) {
844 DSS_ERROR_LOG(("rsync process %d returned %d\n", (int)create_pid, es));
845 ret = -E_BAD_EXIT_CODE;
846 snapshot_creation_status = HS_READY;
847 goto out;
848 }
849 ret = rename_incomplete_snapshot(current_snapshot_creation_time);
850 if (ret < 0)
851 goto out;
852 snapshot_creation_status = HS_SUCCESS;
853 free(name_of_reference_snapshot);
854 name_of_reference_snapshot = NULL;
855 out:
856 create_process_stopped = 0;
857 return ret;
858 }
859
860 static int handle_pre_create_hook_exit(int status)
861 {
862 int es, ret;
863 static int warn_count;
864
865 if (!WIFEXITED(status)) {
866 snapshot_creation_status = HS_READY;
867 ret = -E_INVOLUNTARY_EXIT;
868 goto out;
869 }
870 es = WEXITSTATUS(status);
871 if (es) {
872 if (!warn_count--) {
873 DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
874 conf.pre_create_hook_arg, es));
875 DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
876 warn_count = 60; /* warn only once per hour */
877 }
878 next_snapshot_time = get_current_time() + 60;
879 snapshot_creation_status = HS_READY;
880 ret = 0;
881 goto out;
882 }
883 warn_count = 0;
884 snapshot_creation_status = HS_PRE_SUCCESS;
885 ret = 1;
886 out:
887 return ret;
888 }
889
890 static int handle_sigchld(void)
891 {
892 pid_t pid;
893 int status, ret = reap_child(&pid, &status);
894
895 if (ret <= 0)
896 return ret;
897
898 if (pid == create_pid) {
899 switch (snapshot_creation_status) {
900 case HS_PRE_RUNNING:
901 ret = handle_pre_create_hook_exit(status);
902 break;
903 case HS_RUNNING:
904 ret = handle_rsync_exit(status);
905 break;
906 case HS_POST_RUNNING:
907 snapshot_creation_status = HS_READY;
908 ret = 1;
909 break;
910 default:
911 DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
912 snapshot_creation_status));
913 return -E_BUG;
914 }
915 create_pid = 0;
916 return ret;
917 }
918 if (pid == remove_pid) {
919 ret = handle_remove_exit(status);
920 if (ret < 0)
921 return ret;
922 return ret;
923 }
924 DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
925 return -E_BUG;
926 }
927
928 static int check_config(void)
929 {
930 if (conf.unit_interval_arg <= 0) {
931 DSS_ERROR_LOG(("bad unit interval: %i\n", conf.unit_interval_arg));
932 return -E_INVALID_NUMBER;
933 }
934 DSS_DEBUG_LOG(("unit interval: %i day(s)\n", conf.unit_interval_arg));
935 if (conf.num_intervals_arg <= 0) {
936 DSS_ERROR_LOG(("bad number of intervals %i\n", conf.num_intervals_arg));
937 return -E_INVALID_NUMBER;
938 }
939 DSS_DEBUG_LOG(("number of intervals: %i\n", conf.num_intervals_arg));
940 return 1;
941 }
942
943 /*
944 * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
945 * file was read successfully.
946 */
947 static int parse_config_file(int override)
948 {
949 int ret, config_file_exists;
950 char *config_file = get_config_file_name();
951 struct stat statbuf;
952 char *old_logfile_arg = NULL;
953 int old_daemon_given = 0;
954
955 if (override) { /* SIGHUP */
956 if (conf.logfile_given)
957 old_logfile_arg = dss_strdup(conf.logfile_arg);
958 old_daemon_given = conf.daemon_given;
959 }
960
961 config_file_exists = !stat(config_file, &statbuf);
962 if (!config_file_exists && conf.config_file_given) {
963 ret = -ERRNO_TO_DSS_ERROR(errno);
964 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
965 goto out;
966 }
967 if (config_file_exists) {
968 struct cmdline_parser_params params;
969 params.override = override;
970 params.initialize = 0;
971 params.check_required = 1;
972 params.check_ambiguity = 0;
973 params.print_errors = 1;
974 if (override) { /* invalidate all rsync options */
975 int i;
976
977 for (i = 0; i < conf.rsync_option_given; i++) {
978 free(conf.rsync_option_arg[i]);
979 conf.rsync_option_arg[i] = NULL;
980 }
981 conf.rsync_option_given = 0;
982 }
983 cmdline_parser_config_file(config_file, &conf, &params);
984 }
985 ret = check_config();
986 if (ret < 0)
987 goto out;
988 if (override) {
989 /* don't change daemon mode on SIGHUP */
990 conf.daemon_given = old_daemon_given;
991 close_log(logfile);
992 logfile = NULL;
993 if (conf.logfile_given)
994 free(old_logfile_arg);
995 else if (conf.daemon_given) { /* re-use old logfile */
996 conf.logfile_arg = old_logfile_arg;
997 conf.logfile_given = 1;
998 }
999 }
1000 if (conf.logfile_given && conf.run_given && conf.daemon_given) {
1001 logfile = open_log(conf.logfile_arg);
1002 log_welcome(conf.loglevel_arg);
1003 }
1004 DSS_DEBUG_LOG(("loglevel: %d\n", conf.loglevel_arg));
1005 ret = config_file_exists;
1006 out:
1007 free(config_file);
1008 if (ret < 0)
1009 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1010 return ret;
1011 }
1012
1013 static int change_to_dest_dir(void)
1014 {
1015 DSS_INFO_LOG(("changing cwd to %s\n", conf.dest_dir_arg));
1016 return dss_chdir(conf.dest_dir_arg);
1017 }
1018
1019 static int handle_sighup(void)
1020 {
1021 int ret;
1022
1023 DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1024 dump_dss_config("old");
1025 ret = parse_config_file(1);
1026 if (ret < 0)
1027 return ret;
1028 dump_dss_config("reloaded");
1029 invalidate_next_snapshot_time();
1030 return change_to_dest_dir();
1031 }
1032
1033 static int handle_signal(void)
1034 {
1035 int sig, ret = next_signal();
1036
1037 if (ret <= 0)
1038 goto out;
1039 sig = ret;
1040 switch (sig) {
1041 case SIGINT:
1042 case SIGTERM:
1043 restart_create_process();
1044 dss_kill(create_pid, SIGTERM, NULL);
1045 dss_kill(remove_pid, SIGTERM, NULL);
1046 ret = -E_SIGNAL;
1047 break;
1048 case SIGHUP:
1049 ret = handle_sighup();
1050 break;
1051 case SIGCHLD:
1052 ret = handle_sigchld();
1053 break;
1054 }
1055 out:
1056 if (ret < 0)
1057 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1058 return ret;
1059 }
1060
1061 /*
1062 * We can not use rsync locally if the local user is different from the remote
1063 * user or if the src dir is not on the local host (or both).
1064 */
1065 static int use_rsync_locally(char *logname)
1066 {
1067 char *h = conf.remote_host_arg;
1068
1069 if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1070 return 0;
1071 if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1072 return 0;
1073 return 1;
1074 }
1075
1076 static int rename_resume_snap(int64_t creation_time)
1077 {
1078 struct snapshot_list sl;
1079 struct snapshot *s = NULL;
1080 char *new_name = incomplete_name(creation_time);
1081 int ret;
1082 const char *why;
1083
1084 sl.num_snapshots = 0;
1085
1086 ret = 0;
1087 if (conf.no_resume_given)
1088 goto out;
1089 dss_get_snapshot_list(&sl);
1090 /*
1091 * Snapshot recycling: We first look at the newest snapshot. If this
1092 * snapshot happens to be incomplete, the last rsync process was
1093 * aborted and we reuse this one. Otherwise we look at snapshots which
1094 * could be removed (outdated and redundant snapshots) as candidates
1095 * for recycling. If no outdated/redundant snapshot exists, we check if
1096 * there is an orphaned snapshot, which likely is useless anyway.
1097 *
1098 * Only if no existing snapshot is suitable for recycling, we bite the
1099 * bullet and create a new one.
1100 */
1101 s = get_newest_snapshot(&sl);
1102 if (!s) /* no snapshots at all */
1103 goto out;
1104 /* re-use last snapshot if it is incomplete */
1105 why = "aborted";
1106 if ((s->flags & SS_COMPLETE) == 0)
1107 goto out;
1108 why = "outdated";
1109 s = find_outdated_snapshot(&sl);
1110 if (s)
1111 goto out;
1112 why = "redundant";
1113 s = find_redundant_snapshot(&sl);
1114 if (s)
1115 goto out;
1116 why = "orphaned";
1117 s = find_orphaned_snapshot(&sl);
1118 out:
1119 if (s) {
1120 DSS_INFO_LOG(("reusing %s snapshot %s\n", why, s->name));
1121 ret = dss_rename(s->name, new_name);
1122 }
1123 if (ret >= 0)
1124 DSS_NOTICE_LOG(("creating new snapshot %s\n", new_name));
1125 free(new_name);
1126 free_snapshot_list(&sl);
1127 return ret;
1128 }
1129
1130 static void create_rsync_argv(char ***argv, int64_t *num)
1131 {
1132 char *logname;
1133 int i = 0, j;
1134 struct snapshot_list sl;
1135
1136 dss_get_snapshot_list(&sl);
1137 assert(!name_of_reference_snapshot);
1138 name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1139 free_snapshot_list(&sl);
1140
1141 *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1142 (*argv)[i++] = dss_strdup("rsync");
1143 (*argv)[i++] = dss_strdup("-aq");
1144 (*argv)[i++] = dss_strdup("--delete");
1145 for (j = 0; j < conf.rsync_option_given; j++)
1146 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1147 if (name_of_reference_snapshot) {
1148 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1149 (*argv)[i++] = make_message("--link-dest=../%s",
1150 name_of_reference_snapshot);
1151 } else
1152 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1153 logname = dss_logname();
1154 if (use_rsync_locally(logname))
1155 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1156 else
1157 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1158 conf.remote_user_arg : logname,
1159 conf.remote_host_arg, conf.source_dir_arg);
1160 free(logname);
1161 *num = get_current_time();
1162 (*argv)[i++] = incomplete_name(*num);
1163 (*argv)[i++] = NULL;
1164 for (j = 0; j < i; j++)
1165 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1166 }
1167
1168 static void free_rsync_argv(char **argv)
1169 {
1170 int i;
1171
1172 if (!argv)
1173 return;
1174 for (i = 0; argv[i]; i++)
1175 free(argv[i]);
1176 free(argv);
1177 }
1178
1179 static int create_snapshot(char **argv)
1180 {
1181 int ret;
1182
1183 ret = rename_resume_snap(current_snapshot_creation_time);
1184 if (ret < 0)
1185 return ret;
1186 dss_exec(&create_pid, argv[0], argv);
1187 snapshot_creation_status = HS_RUNNING;
1188 return ret;
1189 }
1190
1191 static int select_loop(void)
1192 {
1193 int ret;
1194 /* check every 60 seconds for free disk space */
1195 struct timeval tv;
1196 char **rsync_argv = NULL;
1197
1198 for (;;) {
1199 fd_set rfds;
1200 struct timeval *tvp;
1201
1202 if (remove_pid)
1203 tvp = NULL; /* sleep until rm hook/process dies */
1204 else { /* sleep one minute */
1205 tv.tv_sec = 60;
1206 tv.tv_usec = 0;
1207 tvp = &tv;
1208 }
1209 FD_ZERO(&rfds);
1210 FD_SET(signal_pipe, &rfds);
1211 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1212 if (ret < 0)
1213 goto out;
1214 if (FD_ISSET(signal_pipe, &rfds)) {
1215 ret = handle_signal();
1216 if (ret < 0)
1217 goto out;
1218 }
1219 if (remove_pid)
1220 continue;
1221 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1222 ret = exec_rm();
1223 if (ret < 0)
1224 goto out;
1225 continue;
1226 }
1227 if (snapshot_removal_status == HS_SUCCESS) {
1228 post_remove_hook();
1229 continue;
1230 }
1231 ret = try_to_free_disk_space();
1232 if (ret < 0)
1233 goto out;
1234 if (snapshot_removal_status != HS_READY) {
1235 stop_create_process();
1236 continue;
1237 }
1238 restart_create_process();
1239 switch (snapshot_creation_status) {
1240 case HS_READY:
1241 if (!next_snapshot_is_due())
1242 continue;
1243 pre_create_hook();
1244 continue;
1245 case HS_PRE_RUNNING:
1246 case HS_RUNNING:
1247 case HS_POST_RUNNING:
1248 continue;
1249 case HS_PRE_SUCCESS:
1250 if (!name_of_reference_snapshot) {
1251 free_rsync_argv(rsync_argv);
1252 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1253 }
1254 ret = create_snapshot(rsync_argv);
1255 if (ret < 0)
1256 goto out;
1257 continue;
1258 case HS_NEEDS_RESTART:
1259 if (!next_snapshot_is_due())
1260 continue;
1261 ret = create_snapshot(rsync_argv);
1262 if (ret < 0)
1263 goto out;
1264 continue;
1265 case HS_SUCCESS:
1266 post_create_hook();
1267 continue;
1268 }
1269 }
1270 out:
1271 return ret;
1272 }
1273
1274 static void exit_hook(int exit_code)
1275 {
1276 char *argv[3];
1277 pid_t pid;
1278
1279 argv[0] = conf.exit_hook_arg;
1280 argv[1] = dss_strerror(-exit_code);
1281 argv[2] = NULL;
1282
1283 DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
1284 dss_exec(&pid, conf.exit_hook_arg, argv);
1285 }
1286
1287 static void lock_dss_or_die(void)
1288 {
1289 char *config_file = get_config_file_name();
1290 int ret = lock_dss(config_file);
1291
1292 free(config_file);
1293 if (ret < 0) {
1294 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1295 exit(EXIT_FAILURE);
1296 }
1297 }
1298
1299 static int com_run(void)
1300 {
1301 int ret;
1302
1303 lock_dss_or_die();
1304 if (conf.dry_run_given) {
1305 DSS_ERROR_LOG(("dry_run not supported by this command\n"));
1306 return -E_SYNTAX;
1307 }
1308 ret = install_sighandler(SIGHUP);
1309 if (ret < 0)
1310 return ret;
1311 ret = select_loop();
1312 if (ret >= 0) /* impossible */
1313 ret = -E_BUG;
1314 exit_hook(ret);
1315 return ret;
1316 }
1317
1318 static int com_prune(void)
1319 {
1320 int ret;
1321 struct snapshot_list sl;
1322 struct snapshot *victim;
1323 struct disk_space ds;
1324 const char *why;
1325
1326 lock_dss_or_die();
1327 ret = get_disk_space(".", &ds);
1328 if (ret < 0)
1329 return ret;
1330 log_disk_space(&ds);
1331 dss_get_snapshot_list(&sl);
1332 why = "outdated";
1333 victim = find_outdated_snapshot(&sl);
1334 if (victim)
1335 goto rm;
1336 why = "redundant";
1337 victim = find_redundant_snapshot(&sl);
1338 if (victim)
1339 goto rm;
1340 ret = 0;
1341 goto out;
1342 rm:
1343 if (conf.dry_run_given) {
1344 dss_msg("%s snapshot %s (interval = %i)\n",
1345 why, victim->name, victim->interval);
1346 ret = 0;
1347 goto out;
1348 }
1349 pre_remove_hook(victim, why);
1350 if (snapshot_removal_status == HS_PRE_RUNNING) {
1351 ret = wait_for_remove_process();
1352 if (ret < 0)
1353 goto out;
1354 if (snapshot_removal_status != HS_PRE_SUCCESS)
1355 goto out;
1356 }
1357 ret = exec_rm();
1358 if (ret < 0)
1359 goto out;
1360 ret = wait_for_remove_process();
1361 if (ret < 0)
1362 goto out;
1363 if (snapshot_removal_status != HS_SUCCESS)
1364 goto out;
1365 post_remove_hook();
1366 if (snapshot_removal_status != HS_POST_RUNNING)
1367 goto out;
1368 ret = wait_for_remove_process();
1369 if (ret < 0)
1370 goto out;
1371 ret = 1;
1372 out:
1373 free_snapshot_list(&sl);
1374 return ret;
1375 }
1376
1377 static int com_create(void)
1378 {
1379 int ret, status;
1380 char **rsync_argv;
1381
1382 lock_dss_or_die();
1383 if (conf.dry_run_given) {
1384 int i;
1385 char *msg = NULL;
1386 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1387 for (i = 0; rsync_argv[i]; i++) {
1388 char *tmp = msg;
1389 msg = make_message("%s%s%s", tmp? tmp : "",
1390 tmp? " " : "", rsync_argv[i]);
1391 free(tmp);
1392 }
1393 free_rsync_argv(rsync_argv);
1394 dss_msg("%s\n", msg);
1395 free(msg);
1396 return 1;
1397 }
1398 pre_create_hook();
1399 if (create_pid) {
1400 ret = wait_for_process(create_pid, &status);
1401 if (ret < 0)
1402 return ret;
1403 ret = handle_pre_create_hook_exit(status);
1404 if (ret <= 0) /* error, or pre-create failed */
1405 return ret;
1406 }
1407 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1408 ret = create_snapshot(rsync_argv);
1409 if (ret < 0)
1410 goto out;
1411 ret = wait_for_process(create_pid, &status);
1412 if (ret < 0)
1413 goto out;
1414 ret = handle_rsync_exit(status);
1415 if (ret < 0)
1416 goto out;
1417 post_create_hook();
1418 if (create_pid)
1419 ret = wait_for_process(create_pid, &status);
1420 out:
1421 free_rsync_argv(rsync_argv);
1422 return ret;
1423 }
1424
1425 static int com_ls(void)
1426 {
1427 int i;
1428 struct snapshot_list sl;
1429 struct snapshot *s;
1430
1431 dss_get_snapshot_list(&sl);
1432 FOR_EACH_SNAPSHOT(s, i, &sl) {
1433 int64_t d = 0;
1434 if (s->flags & SS_COMPLETE)
1435 d = (s->completion_time - s->creation_time) / 60;
1436 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1437 };
1438 free_snapshot_list(&sl);
1439 return 1;
1440 }
1441
1442 static int setup_signal_handling(void)
1443 {
1444 int ret;
1445
1446 DSS_INFO_LOG(("setting up signal handlers\n"));
1447 signal_pipe = signal_init(); /* always successful */
1448 ret = install_sighandler(SIGINT);
1449 if (ret < 0)
1450 return ret;
1451 ret = install_sighandler(SIGTERM);
1452 if (ret < 0)
1453 return ret;
1454 return install_sighandler(SIGCHLD);
1455 }
1456
1457 /**
1458 * The main function of dss.
1459 *
1460 * \param argc Usual argument count.
1461 * \param argv Usual argument vector.
1462 */
1463 int main(int argc, char **argv)
1464 {
1465 int ret;
1466 struct cmdline_parser_params params;
1467
1468 params.override = 0;
1469 params.initialize = 1;
1470 params.check_required = 0;
1471 params.check_ambiguity = 0;
1472 params.print_errors = 1;
1473
1474 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1475 ret = parse_config_file(0);
1476 if (ret < 0)
1477 goto out;
1478 if (ret == 0) { /* no config file given */
1479 /*
1480 * Parse the command line options again, but this time check
1481 * that all required options are given.
1482 */
1483 struct cmdline_parser_params params;
1484 params.override = 1;
1485 params.initialize = 1;
1486 params.check_required = 1;
1487 params.check_ambiguity = 1;
1488 params.print_errors = 1;
1489 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1490 }
1491 if (conf.daemon_given)
1492 daemon_init();
1493 ret = change_to_dest_dir();
1494 if (ret < 0)
1495 goto out;
1496 dump_dss_config("startup");
1497 ret = setup_signal_handling();
1498 if (ret < 0)
1499 goto out;
1500 ret = call_command_handler();
1501 out:
1502 if (ret < 0)
1503 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1504 exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1505 }