bf30f893fa36fb9749d00ab706311709a9d4a42f
[dss.git] / dss.c
1 /*
2 * Copyright (C) 2008-2010 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 "string.h"
28 #include "error.h"
29 #include "fd.h"
30 #include "exec.h"
31 #include "daemon.h"
32 #include "signal.h"
33 #include "df.h"
34 #include "time.h"
35 #include "snap.h"
36
37 /** Command line and config file options. */
38 static struct gengetopt_args_info conf;
39 /** Non-NULL if we log to a file. */
40 static FILE *logfile;
41 /** The read end of the signal pipe */
42 static int signal_pipe;
43 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
44 static pid_t create_pid;
45 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
46 static int create_process_stopped;
47 /** Process id of current pre-remove/rm/post-remove process. */
48 static pid_t remove_pid;
49 /** When the next snapshot is due. */
50 static int64_t next_snapshot_time;
51 /** When to try to remove something. */
52 static struct timeval next_removal_check;
53 /** Creation time of the snapshot currently being created. */
54 static int64_t current_snapshot_creation_time;
55 /** The snapshot currently being removed. */
56 struct snapshot *snapshot_currently_being_removed;
57 /** Needed by the post-create hook. */
58 static char *path_to_last_complete_snapshot;
59 static char *name_of_reference_snapshot;
60 /** \sa \ref snap.h for details. */
61 enum hook_status snapshot_creation_status;
62 /** \sa \ref snap.h for details. */
63 enum hook_status snapshot_removal_status;
64
65
66 DEFINE_DSS_ERRLIST;
67 static const char const *hook_status_description[] = {HOOK_STATUS_ARRAY};
68
69 /* may be called with ds == NULL. */
70 static int disk_space_low(struct disk_space *ds)
71 {
72 struct disk_space ds_struct;
73
74 if (!ds) {
75 int ret = get_disk_space(".", &ds_struct);
76 if (ret < 0)
77 return ret;
78 ds = &ds_struct;
79 }
80 if (conf.min_free_mb_arg)
81 if (ds->free_mb < conf.min_free_mb_arg)
82 return 1;
83 if (conf.min_free_percent_arg)
84 if (ds->percent_free < conf.min_free_percent_arg)
85 return 1;
86 if (conf.min_free_percent_inodes_arg)
87 if (ds->percent_free_inodes < conf.min_free_percent_inodes_arg)
88 return 1;
89 return 0;
90 }
91
92 static void dump_dss_config(const char *msg)
93 {
94 const char dash[] = "-----------------------------";
95 int ret;
96 FILE *log = logfile? logfile : stderr;
97 struct disk_space ds;
98 int64_t now = get_current_time();
99
100 if (conf.loglevel_arg > INFO)
101 return;
102
103 fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
104 fprintf(log, "\n*** disk space ***\n\n");
105 ret = get_disk_space(".", &ds);
106 if (ret >= 0) {
107 DSS_INFO_LOG("disk space low: %s\n", disk_space_low(&ds)?
108 "yes" : "no");
109 log_disk_space(&ds);
110 } else
111 DSS_ERROR_LOG("can not get free disk space: %s\n",
112 dss_strerror(-ret));
113
114 /* we continue on errors from get_disk_space */
115
116 fprintf(log, "\n*** command line and config file options ***\n\n");
117 cmdline_parser_dump(log, &conf);
118 fprintf(log, "\n*** internal state ***\n\n");
119 fprintf(log,
120 "pid: %d\n"
121 "logile: %s\n"
122 "snapshot_currently_being_removed: %s\n"
123 "path_to_last_complete_snapshot: %s\n"
124 "reference_snapshot: %s\n"
125 "snapshot_creation_status: %s\n"
126 "snapshot_removal_status: %s\n"
127 ,
128 (int) getpid(),
129 logfile? conf.logfile_arg : "stderr",
130 snapshot_currently_being_removed?
131 snapshot_currently_being_removed->name : "(none)",
132 path_to_last_complete_snapshot?
133 path_to_last_complete_snapshot : "(none)",
134 name_of_reference_snapshot?
135 name_of_reference_snapshot : "(none)",
136 hook_status_description[snapshot_creation_status],
137 hook_status_description[snapshot_removal_status]
138 );
139 if (create_pid != 0)
140 fprintf(log,
141 "create_pid: %" PRId32 "\n"
142 "create process is %sstopped\n"
143 ,
144 create_pid,
145 create_process_stopped? "" : "not "
146 );
147 if (remove_pid != 0)
148 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
149 if (next_snapshot_time != 0)
150 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
151 next_snapshot_time - now);
152 if (current_snapshot_creation_time != 0)
153 fprintf(log, "current_snapshot_creation_time: %"
154 PRId64 " (%" PRId64 " seconds ago)\n",
155 current_snapshot_creation_time,
156 now - current_snapshot_creation_time
157 );
158 if (next_removal_check.tv_sec != 0) {
159 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
160 (long long unsigned)next_removal_check.tv_sec,
161 now - (long long unsigned)next_removal_check.tv_sec
162 );
163
164 }
165 fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
166 }
167
168 /* a litte cpp magic helps to DRY */
169 #define COMMANDS \
170 COMMAND(ls) \
171 COMMAND(create) \
172 COMMAND(prune) \
173 COMMAND(run)
174 #define COMMAND(x) static int com_ ##x(void);
175 COMMANDS
176 #undef COMMAND
177 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
178 static int call_command_handler(void)
179 {
180 COMMANDS
181 DSS_EMERG_LOG("BUG: did not find command handler\n");
182 return -E_BUG;
183 }
184 #undef COMMAND
185 #undef COMMANDS
186
187 /**
188 * The log function of dss.
189 *
190 * \param ll Loglevel.
191 * \param fml Usual format string.
192 *
193 * All DSS_XXX_LOG() macros use this function.
194 */
195 __printf_2_3 void dss_log(int ll, const char* fmt,...)
196 {
197 va_list argp;
198 FILE *outfd;
199 struct tm *tm;
200 time_t t1;
201 char str[255] = "";
202
203 if (ll < conf.loglevel_arg)
204 return;
205 outfd = logfile? logfile : stderr;
206 time(&t1);
207 tm = localtime(&t1);
208 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
209 fprintf(outfd, "%s ", str);
210 if (conf.loglevel_arg <= INFO)
211 fprintf(outfd, "%i: ", ll);
212 va_start(argp, fmt);
213 vfprintf(outfd, fmt, argp);
214 va_end(argp);
215 }
216
217 /**
218 * Print a message either to stdout or to the log file.
219 */
220 static __printf_1_2 void dss_msg(const char* fmt,...)
221 {
222 FILE *outfd = conf.daemon_given? logfile : stdout;
223 va_list argp;
224 va_start(argp, fmt);
225 vfprintf(outfd, fmt, argp);
226 va_end(argp);
227 }
228
229 static char *get_config_file_name(void)
230 {
231 char *home, *config_file;
232
233 if (conf.config_file_given)
234 return dss_strdup(conf.config_file_arg);
235 home = get_homedir();
236 config_file = make_message("%s/.dssrc", home);
237 free(home);
238 return config_file;
239 }
240
241 static void dss_get_snapshot_list(struct snapshot_list *sl)
242 {
243 get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
244 }
245
246 static int64_t compute_next_snapshot_time(void)
247 {
248 int64_t x = 0, now = get_current_time(), unit_interval
249 = 24 * 3600 * conf.unit_interval_arg, ret;
250 unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
251 num_complete_snapshots = 0;
252 int i;
253 struct snapshot *s = NULL;
254 struct snapshot_list sl;
255
256 dss_get_snapshot_list(&sl);
257 FOR_EACH_SNAPSHOT(s, i, &sl) {
258 if (!(s->flags & SS_COMPLETE))
259 continue;
260 num_complete_snapshots++;
261 x += s->completion_time - s->creation_time;
262 }
263 assert(x >= 0);
264
265 ret = now;
266 if (num_complete_snapshots == 0)
267 goto out;
268 x /= num_complete_snapshots; /* avg time to create one snapshot */
269 if (unit_interval < x * wanted) /* oops, no sleep at all */
270 goto out;
271 ret = s->completion_time + unit_interval / wanted - x;
272 out:
273 free_snapshot_list(&sl);
274 return ret;
275 }
276
277 static inline void invalidate_next_snapshot_time(void)
278 {
279 next_snapshot_time = 0;
280 }
281
282 static inline int next_snapshot_time_is_valid(void)
283 {
284 return next_snapshot_time != 0;
285 }
286
287 static int next_snapshot_is_due(void)
288 {
289 int64_t now = get_current_time();
290
291 if (!next_snapshot_time_is_valid())
292 next_snapshot_time = compute_next_snapshot_time();
293 if (next_snapshot_time <= now) {
294 DSS_DEBUG_LOG("next snapshot: now\n");
295 return 1;
296 }
297 DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
298 next_snapshot_time - now);
299 return 0;
300 }
301
302 static int pre_create_hook(void)
303 {
304 int ret, fds[3] = {0, 0, 0};
305
306 assert(snapshot_creation_status == HS_READY);
307 /* make sure that the next snapshot time will be recomputed */
308 invalidate_next_snapshot_time();
309 DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
310 ret = dss_exec_cmdline_pid(&create_pid,
311 conf.pre_create_hook_arg, fds);
312 if (ret < 0)
313 return ret;
314 snapshot_creation_status = HS_PRE_RUNNING;
315 return ret;
316 }
317
318 static int pre_remove_hook(struct snapshot *s, const char *why)
319 {
320 int ret, fds[3] = {0, 0, 0};
321 char *cmd;
322
323 if (!s)
324 return 0;
325 DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
326 assert(snapshot_removal_status == HS_READY);
327 assert(remove_pid == 0);
328 assert(!snapshot_currently_being_removed);
329
330 snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
331 *snapshot_currently_being_removed = *s;
332 snapshot_currently_being_removed->name = dss_strdup(s->name);
333
334 cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
335 conf.dest_dir_arg, s->name);
336 DSS_DEBUG_LOG("executing %s\n", cmd);
337 ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
338 free(cmd);
339 if (ret < 0)
340 return ret;
341 snapshot_removal_status = HS_PRE_RUNNING;
342 return ret;
343 }
344
345 static int exec_rm(void)
346 {
347 struct snapshot *s = snapshot_currently_being_removed;
348 int fds[3] = {0, 0, 0};
349 char *new_name = being_deleted_name(s);
350 char *argv[] = {"rm", "-rf", new_name, NULL};
351 int ret;
352
353 assert(snapshot_removal_status == HS_PRE_SUCCESS);
354 assert(remove_pid == 0);
355
356 DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
357 ret = dss_rename(s->name, new_name);
358 if (ret < 0)
359 goto out;
360 ret = dss_exec(&remove_pid, argv[0], argv, fds);
361 if (ret < 0)
362 goto out;
363 snapshot_removal_status = HS_RUNNING;
364 out:
365 free(new_name);
366 return ret;
367 }
368
369 static int snapshot_is_being_created(struct snapshot *s)
370 {
371 return s->creation_time == current_snapshot_creation_time;
372 }
373
374 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
375 {
376 struct snapshot *s;
377 int i;
378
379 DSS_DEBUG_LOG("looking for orphaned snapshots\n");
380 FOR_EACH_SNAPSHOT(s, i, sl) {
381 if (snapshot_is_being_created(s))
382 continue;
383 /*
384 * We know that no rm is currently running, so if s is marked
385 * as being deleted, a previously started rm must have failed.
386 */
387 if (s->flags & SS_BEING_DELETED)
388 return s;
389
390 if (s->flags & SS_COMPLETE) /* good snapshot */
391 continue;
392 /*
393 * This snapshot is incomplete and it is not the snapshot
394 * currently being created. However, we must not remove it if
395 * rsync is about to be restarted. As only the newest snapshot
396 * can be restarted, this snapshot is orphaned if it is not the
397 * newest snapshot or if we are not about to restart rsync.
398 */
399 if (get_newest_snapshot(sl) != s)
400 return s;
401 if (snapshot_creation_status != HS_NEEDS_RESTART)
402 return s;
403 }
404 /* no orphaned snapshots */
405 return NULL;
406 }
407
408 static int is_reference_snapshot(struct snapshot *s)
409 {
410 if (!name_of_reference_snapshot)
411 return 0;
412 return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
413 }
414
415 /*
416 * return: 0: no redundant snapshots, 1: rm process started, negative: error
417 */
418 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
419 {
420 int i, interval;
421 struct snapshot *s;
422 unsigned missing = 0;
423
424 DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
425 for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
426 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
427 unsigned num = sl->interval_count[interval];
428 struct snapshot *victim = NULL, *prev = NULL;
429 int64_t score = LONG_MAX;
430
431 if (keep >= num)
432 missing += keep - num;
433 // DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
434 // interval, keep, num, missing);
435 if (keep + missing >= num)
436 continue;
437 /* redundant snapshot in this interval, pick snapshot with lowest score */
438 FOR_EACH_SNAPSHOT(s, i, sl) {
439 int64_t this_score;
440
441 if (snapshot_is_being_created(s))
442 continue;
443 if (is_reference_snapshot(s))
444 continue;
445 //DSS_DEBUG_LOG("checking %s\n", s->name);
446 if (s->interval > interval) {
447 prev = s;
448 continue;
449 }
450 if (s->interval < interval)
451 break;
452 if (!victim) {
453 victim = s;
454 prev = s;
455 continue;
456 }
457 assert(prev);
458 /* check if s is a better victim */
459 this_score = s->creation_time - prev->creation_time;
460 assert(this_score >= 0);
461 //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
462 if (this_score < score) {
463 score = this_score;
464 victim = s;
465 }
466 prev = s;
467 }
468 assert(victim);
469 return victim;
470 }
471 return NULL;
472 }
473
474 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
475 {
476 int i;
477 struct snapshot *s;
478
479 DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
480 conf.num_intervals_arg);
481 FOR_EACH_SNAPSHOT(s, i, sl) {
482 if (snapshot_is_being_created(s))
483 continue;
484 if (is_reference_snapshot(s))
485 continue;
486 if (s->interval < conf.num_intervals_arg)
487 continue;
488 return s;
489 }
490 return NULL;
491 }
492
493 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
494 {
495 int i;
496 struct snapshot *s;
497 FOR_EACH_SNAPSHOT(s, i, sl) {
498 if (snapshot_is_being_created(s))
499 continue;
500 if (is_reference_snapshot(s))
501 continue;
502 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
503 return s;
504 }
505 return NULL;
506 }
507
508 static int rename_incomplete_snapshot(int64_t start)
509 {
510 char *old_name;
511 int ret;
512
513 free(path_to_last_complete_snapshot);
514 ret = complete_name(start, get_current_time(),
515 &path_to_last_complete_snapshot);
516 if (ret < 0)
517 return ret;
518 old_name = incomplete_name(start);
519 ret = dss_rename(old_name, path_to_last_complete_snapshot);
520 if (ret >= 0)
521 DSS_NOTICE_LOG("%s -> %s\n", old_name,
522 path_to_last_complete_snapshot);
523 free(old_name);
524 return ret;
525 }
526
527 static int try_to_free_disk_space(void)
528 {
529 int ret;
530 struct snapshot_list sl;
531 struct snapshot *victim;
532 struct timeval now;
533 const char *why;
534 int low_disk_space;
535
536 ret = disk_space_low(NULL);
537 if (ret < 0)
538 return ret;
539 low_disk_space = ret;
540 gettimeofday(&now, NULL);
541 if (tv_diff(&next_removal_check, &now, NULL) > 0)
542 return 0;
543 if (!low_disk_space) {
544 if (conf.keep_redundant_given)
545 return 0;
546 if (snapshot_creation_status != HS_READY)
547 return 0;
548 if (next_snapshot_is_due())
549 return 0;
550 }
551 dss_get_snapshot_list(&sl);
552 ret = 0;
553 if (!low_disk_space && sl.num_snapshots <= 1)
554 goto out;
555 why = "outdated";
556 victim = find_outdated_snapshot(&sl);
557 if (victim)
558 goto remove;
559 why = "redundant";
560 victim = find_redundant_snapshot(&sl);
561 if (victim)
562 goto remove;
563 /* try harder only if disk space is low */
564 if (!low_disk_space)
565 goto out;
566 why = "orphaned";
567 victim = find_orphaned_snapshot(&sl);
568 if (victim)
569 goto remove;
570 DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
571 victim = find_oldest_removable_snapshot(&sl);
572 if (victim)
573 goto remove;
574 DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
575 ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
576 goto out;
577 remove:
578 ret = pre_remove_hook(victim, why);
579 out:
580 free_snapshot_list(&sl);
581 return ret;
582 }
583
584 static int post_create_hook(void)
585 {
586 int ret, fds[3] = {0, 0, 0};
587 char *cmd;
588
589 cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
590 conf.dest_dir_arg, path_to_last_complete_snapshot);
591 DSS_NOTICE_LOG("executing %s\n", cmd);
592 ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
593 free(cmd);
594 if (ret < 0)
595 return ret;
596 snapshot_creation_status = HS_POST_RUNNING;
597 return ret;
598 }
599
600 static int post_remove_hook(void)
601 {
602 int ret, fds[3] = {0, 0, 0};
603 char *cmd;
604 struct snapshot *s = snapshot_currently_being_removed;
605
606 assert(s);
607
608 cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
609 conf.dest_dir_arg, s->name);
610 DSS_NOTICE_LOG("executing %s\n", cmd);
611 ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
612 free(cmd);
613 if (ret < 0)
614 return ret;
615 snapshot_removal_status = HS_POST_RUNNING;
616 return ret;
617 }
618
619 static void dss_kill(pid_t pid, int sig, const char *msg)
620 {
621 const char *signame, *process_name;
622
623 if (pid == 0)
624 return;
625 switch (sig) {
626 case SIGTERM: signame = "TERM"; break;
627 case SIGSTOP: signame = "STOP"; break;
628 case SIGCONT: signame = "CONT"; break;
629 default: signame = "????";
630 }
631
632 if (pid == create_pid)
633 process_name = "create";
634 else if (pid == remove_pid)
635 process_name = "remove";
636 else process_name = "??????";
637
638 if (msg)
639 DSS_INFO_LOG("%s\n", msg);
640 DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n",
641 sig, signame, (int)pid, process_name);
642 if (kill(pid, sig) >= 0)
643 return;
644 DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
645 sig, signame, (int)pid, process_name);
646 }
647
648 static void stop_create_process(void)
649 {
650 if (create_process_stopped)
651 return;
652 dss_kill(create_pid, SIGSTOP, "suspending create process");
653 create_process_stopped = 1;
654 }
655
656 static void restart_create_process(void)
657 {
658 if (!create_process_stopped)
659 return;
660 dss_kill(create_pid, SIGCONT, "resuming create process");
661 create_process_stopped = 0;
662 }
663
664 /**
665 * Print a log message about the exit status of a child.
666 */
667 static void log_termination_msg(pid_t pid, int status)
668 {
669 if (WIFEXITED(status))
670 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
671 WEXITSTATUS(status));
672 else if (WIFSIGNALED(status))
673 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
674 WTERMSIG(status));
675 else
676 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
677 }
678
679 static int wait_for_process(pid_t pid, int *status)
680 {
681 int ret;
682
683 DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
684 for (;;) {
685 fd_set rfds;
686
687 FD_ZERO(&rfds);
688 FD_SET(signal_pipe, &rfds);
689 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
690 if (ret < 0)
691 break;
692 ret = next_signal();
693 if (!ret)
694 continue;
695 if (ret == SIGCHLD) {
696 ret = waitpid(pid, status, 0);
697 if (ret >= 0)
698 break;
699 if (errno != EINTR) { /* error */
700 ret = -ERRNO_TO_DSS_ERROR(errno);
701 break;
702 }
703 }
704 /* SIGINT or SIGTERM */
705 dss_kill(pid, SIGTERM, "killing child process");
706 }
707 if (ret < 0)
708 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
709 else
710 log_termination_msg(pid, *status);
711 return ret;
712 }
713
714 static void handle_pre_remove_exit(int status)
715 {
716 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
717 snapshot_removal_status = HS_READY;
718 gettimeofday(&next_removal_check, NULL);
719 next_removal_check.tv_sec += 60;
720 return;
721 }
722 snapshot_removal_status = HS_PRE_SUCCESS;
723 }
724
725 static int handle_rm_exit(int status)
726 {
727 if (!WIFEXITED(status)) {
728 snapshot_removal_status = HS_READY;
729 return -E_INVOLUNTARY_EXIT;
730 }
731 if (WEXITSTATUS(status)) {
732 snapshot_removal_status = HS_READY;
733 return -E_BAD_EXIT_CODE;
734 }
735 snapshot_removal_status = HS_SUCCESS;
736 return 1;
737 }
738
739 static void handle_post_remove_exit(void)
740 {
741 snapshot_removal_status = HS_READY;
742 }
743
744 static int handle_remove_exit(int status)
745 {
746 int ret;
747 struct snapshot *s = snapshot_currently_being_removed;
748
749 assert(s);
750 switch (snapshot_removal_status) {
751 case HS_PRE_RUNNING:
752 handle_pre_remove_exit(status);
753 ret = 1;
754 break;
755 case HS_RUNNING:
756 ret = handle_rm_exit(status);
757 break;
758 case HS_POST_RUNNING:
759 handle_post_remove_exit();
760 ret = 1;
761 break;
762 default:
763 ret = -E_BUG;
764 }
765 if (snapshot_removal_status == HS_READY) {
766 free(s->name);
767 free(s);
768 snapshot_currently_being_removed = NULL;
769 }
770 remove_pid = 0;
771 return ret;
772 }
773
774 static int wait_for_remove_process(void)
775 {
776 int status, ret;
777
778 assert(remove_pid);
779 assert(
780 snapshot_removal_status == HS_PRE_RUNNING ||
781 snapshot_removal_status == HS_RUNNING ||
782 snapshot_removal_status == HS_POST_RUNNING
783 );
784 ret = wait_for_process(remove_pid, &status);
785 if (ret < 0)
786 return ret;
787 return handle_remove_exit(status);
788 }
789
790 static int handle_rsync_exit(int status)
791 {
792 int es, ret;
793
794 if (!WIFEXITED(status)) {
795 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
796 ret = -E_INVOLUNTARY_EXIT;
797 snapshot_creation_status = HS_READY;
798 goto out;
799 }
800 es = WEXITSTATUS(status);
801 /*
802 * Restart rsync on non-fatal errors:
803 * 12: Error in rsync protocol data stream
804 * 13: Errors with program diagnostics
805 */
806 if (es == 12 || es == 13) {
807 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
808 (int)create_pid, es);
809 snapshot_creation_status = HS_NEEDS_RESTART;
810 next_snapshot_time = get_current_time() + 60;
811 ret = 1;
812 goto out;
813 }
814 if (es != 0 && es != 23 && es != 24) {
815 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
816 ret = -E_BAD_EXIT_CODE;
817 snapshot_creation_status = HS_READY;
818 goto out;
819 }
820 ret = rename_incomplete_snapshot(current_snapshot_creation_time);
821 if (ret < 0)
822 goto out;
823 snapshot_creation_status = HS_SUCCESS;
824 free(name_of_reference_snapshot);
825 name_of_reference_snapshot = NULL;
826 out:
827 create_process_stopped = 0;
828 return ret;
829 }
830
831 static int handle_pre_create_hook_exit(int status)
832 {
833 int es, ret;
834 static int warn_count;
835
836 if (!WIFEXITED(status)) {
837 snapshot_creation_status = HS_READY;
838 ret = -E_INVOLUNTARY_EXIT;
839 goto out;
840 }
841 es = WEXITSTATUS(status);
842 if (es) {
843 if (!warn_count--) {
844 DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
845 conf.pre_create_hook_arg, es);
846 DSS_NOTICE_LOG("deferring snapshot creation...\n");
847 warn_count = 60; /* warn only once per hour */
848 }
849 next_snapshot_time = get_current_time() + 60;
850 snapshot_creation_status = HS_READY;
851 ret = 0;
852 goto out;
853 }
854 warn_count = 0;
855 snapshot_creation_status = HS_PRE_SUCCESS;
856 ret = 1;
857 out:
858 return ret;
859 }
860
861 static int handle_sigchld(void)
862 {
863 pid_t pid;
864 int status, ret = reap_child(&pid, &status);
865
866 if (ret <= 0)
867 return ret;
868
869 if (pid == create_pid) {
870 switch (snapshot_creation_status) {
871 case HS_PRE_RUNNING:
872 ret = handle_pre_create_hook_exit(status);
873 break;
874 case HS_RUNNING:
875 ret = handle_rsync_exit(status);
876 break;
877 case HS_POST_RUNNING:
878 snapshot_creation_status = HS_READY;
879 ret = 1;
880 break;
881 default:
882 DSS_EMERG_LOG("BUG: create can't die in status %d\n",
883 snapshot_creation_status);
884 return -E_BUG;
885 }
886 create_pid = 0;
887 return ret;
888 }
889 if (pid == remove_pid) {
890 ret = handle_remove_exit(status);
891 if (ret < 0)
892 return ret;
893 return ret;
894 }
895 DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
896 return -E_BUG;
897 }
898
899 static int check_config(void)
900 {
901 if (conf.unit_interval_arg <= 0) {
902 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
903 return -E_INVALID_NUMBER;
904 }
905 DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
906 if (conf.num_intervals_arg <= 0) {
907 DSS_ERROR_LOG("bad number of intervals %i\n", conf.num_intervals_arg);
908 return -E_INVALID_NUMBER;
909 }
910 DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
911 return 1;
912 }
913
914 /*
915 * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
916 * file was read successfully.
917 */
918 static int parse_config_file(int override)
919 {
920 int ret, config_file_exists;
921 char *config_file = get_config_file_name();
922 struct stat statbuf;
923 char *old_logfile_arg = NULL;
924 int old_daemon_given = 0;
925
926 if (override) { /* SIGHUP */
927 if (conf.logfile_given)
928 old_logfile_arg = dss_strdup(conf.logfile_arg);
929 old_daemon_given = conf.daemon_given;
930 }
931
932 config_file_exists = !stat(config_file, &statbuf);
933 if (!config_file_exists && conf.config_file_given) {
934 ret = -ERRNO_TO_DSS_ERROR(errno);
935 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
936 goto out;
937 }
938 if (config_file_exists) {
939 struct cmdline_parser_params params = {
940 .override = override,
941 .initialize = 0,
942 .check_required = 1,
943 .check_ambiguity = 0,
944 .print_errors = 1
945 };
946 if (override) { /* invalidate all rsync options */
947 int i;
948
949 for (i = 0; i < conf.rsync_option_given; i++) {
950 free(conf.rsync_option_arg[i]);
951 conf.rsync_option_arg[i] = NULL;
952 }
953 conf.rsync_option_given = 0;
954 }
955 cmdline_parser_config_file(config_file, &conf, &params);
956 }
957 ret = check_config();
958 if (ret < 0)
959 goto out;
960 if (override) {
961 /* don't change daemon mode on SIGHUP */
962 conf.daemon_given = old_daemon_given;
963 close_log(logfile);
964 logfile = NULL;
965 if (conf.logfile_given)
966 free(old_logfile_arg);
967 else if (conf.daemon_given) { /* re-use old logfile */
968 conf.logfile_arg = old_logfile_arg;
969 conf.logfile_given = 1;
970 }
971 }
972 if (conf.logfile_given && conf.run_given && conf.daemon_given) {
973 logfile = open_log(conf.logfile_arg);
974 log_welcome(conf.loglevel_arg);
975 }
976 DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
977 ret = config_file_exists;
978 out:
979 free(config_file);
980 if (ret < 0)
981 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
982 return ret;
983 }
984
985 static int change_to_dest_dir(void)
986 {
987 DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
988 return dss_chdir(conf.dest_dir_arg);
989 }
990
991 static int handle_sighup(void)
992 {
993 int ret;
994
995 DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
996 dump_dss_config("old");
997 ret = parse_config_file(1);
998 if (ret < 0)
999 return ret;
1000 dump_dss_config("reloaded");
1001 invalidate_next_snapshot_time();
1002 return change_to_dest_dir();
1003 }
1004
1005 static int handle_signal(void)
1006 {
1007 int sig, ret = next_signal();
1008
1009 if (ret <= 0)
1010 goto out;
1011 sig = ret;
1012 switch (sig) {
1013 case SIGINT:
1014 case SIGTERM:
1015 restart_create_process();
1016 dss_kill(create_pid, SIGTERM, NULL);
1017 dss_kill(remove_pid, SIGTERM, NULL);
1018 ret = -E_SIGNAL;
1019 break;
1020 case SIGHUP:
1021 ret = handle_sighup();
1022 break;
1023 case SIGCHLD:
1024 ret = handle_sigchld();
1025 break;
1026 }
1027 out:
1028 if (ret < 0)
1029 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
1030 return ret;
1031 }
1032
1033 /*
1034 * We can not use rsync locally if the local user is different from the remote
1035 * user or if the src dir is not on the local host (or both).
1036 */
1037 static int use_rsync_locally(char *logname)
1038 {
1039 char *h = conf.remote_host_arg;
1040
1041 if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1042 return 0;
1043 if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1044 return 0;
1045 return 1;
1046 }
1047
1048 static int rename_resume_snap(int64_t creation_time)
1049 {
1050 struct snapshot_list sl = {.num_snapshots = 0};
1051 struct snapshot *s = NULL;
1052 char *new_name = incomplete_name(creation_time);
1053 int ret;
1054 const char *why;
1055
1056 ret = 0;
1057 if (conf.no_resume_given)
1058 goto out;
1059 dss_get_snapshot_list(&sl);
1060 /*
1061 * Snapshot recycling: We first look at the newest snapshot. If this
1062 * snapshot happens to be incomplete, the last rsync process was
1063 * aborted and we reuse this one. Otherwise we look at snapshots which
1064 * could be removed (outdated and redundant snapshots) as candidates
1065 * for recycling. If no outdated/redundant snapshot exists, we check if
1066 * there is an orphaned snapshot, which likely is useless anyway.
1067 *
1068 * Only if no existing snapshot is suitable for recycling, we bite the
1069 * bullet and create a new one.
1070 */
1071 s = get_newest_snapshot(&sl);
1072 if (!s) /* no snapshots at all */
1073 goto out;
1074 /* re-use last snapshot if it is incomplete */
1075 why = "aborted";
1076 if ((s->flags & SS_COMPLETE) == 0)
1077 goto out;
1078 why = "outdated";
1079 s = find_outdated_snapshot(&sl);
1080 if (s)
1081 goto out;
1082 why = "redundant";
1083 s = find_redundant_snapshot(&sl);
1084 if (s)
1085 goto out;
1086 why = "orphaned";
1087 s = find_orphaned_snapshot(&sl);
1088 out:
1089 if (s) {
1090 DSS_INFO_LOG("reusing %s snapshot %s\n", why, s->name);
1091 ret = dss_rename(s->name, new_name);
1092 }
1093 if (ret >= 0)
1094 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
1095 free(new_name);
1096 free_snapshot_list(&sl);
1097 return ret;
1098 }
1099
1100 static void create_rsync_argv(char ***argv, int64_t *num)
1101 {
1102 char *logname;
1103 int i = 0, j;
1104 struct snapshot_list sl;
1105
1106 dss_get_snapshot_list(&sl);
1107 assert(!name_of_reference_snapshot);
1108 name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1109 free_snapshot_list(&sl);
1110
1111 *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1112 (*argv)[i++] = dss_strdup("rsync");
1113 (*argv)[i++] = dss_strdup("-aq");
1114 (*argv)[i++] = dss_strdup("--delete");
1115 for (j = 0; j < conf.rsync_option_given; j++)
1116 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1117 if (name_of_reference_snapshot) {
1118 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1119 (*argv)[i++] = make_message("--link-dest=../%s",
1120 name_of_reference_snapshot);
1121 } else
1122 DSS_INFO_LOG("no suitable reference snapshot found\n");
1123 logname = dss_logname();
1124 if (use_rsync_locally(logname))
1125 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1126 else
1127 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1128 conf.remote_user_arg : logname,
1129 conf.remote_host_arg, conf.source_dir_arg);
1130 free(logname);
1131 *num = get_current_time();
1132 (*argv)[i++] = incomplete_name(*num);
1133 (*argv)[i++] = NULL;
1134 for (j = 0; j < i; j++)
1135 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1136 }
1137
1138 static void free_rsync_argv(char **argv)
1139 {
1140 int i;
1141
1142 if (!argv)
1143 return;
1144 for (i = 0; argv[i]; i++)
1145 free(argv[i]);
1146 free(argv);
1147 }
1148
1149 static int create_snapshot(char **argv)
1150 {
1151 int ret, fds[3] = {0, 0, 0};
1152
1153 ret = rename_resume_snap(current_snapshot_creation_time);
1154 if (ret < 0)
1155 return ret;
1156 ret = dss_exec(&create_pid, argv[0], argv, fds);
1157 if (ret < 0)
1158 return ret;
1159 snapshot_creation_status = HS_RUNNING;
1160 return ret;
1161 }
1162
1163 static int select_loop(void)
1164 {
1165 int ret;
1166 /* check every 60 seconds for free disk space */
1167 struct timeval tv;
1168 char **rsync_argv = NULL;
1169
1170 for (;;) {
1171 fd_set rfds;
1172 struct timeval *tvp;
1173
1174 if (remove_pid)
1175 tvp = NULL; /* sleep until rm hook/process dies */
1176 else { /* sleep one minute */
1177 tv.tv_sec = 60;
1178 tv.tv_usec = 0;
1179 tvp = &tv;
1180 }
1181 FD_ZERO(&rfds);
1182 FD_SET(signal_pipe, &rfds);
1183 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1184 if (ret < 0)
1185 goto out;
1186 if (FD_ISSET(signal_pipe, &rfds)) {
1187 ret = handle_signal();
1188 if (ret < 0)
1189 goto out;
1190 }
1191 if (remove_pid)
1192 continue;
1193 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1194 ret = exec_rm();
1195 if (ret < 0)
1196 goto out;
1197 continue;
1198 }
1199 if (snapshot_removal_status == HS_SUCCESS) {
1200 ret = post_remove_hook();
1201 if (ret < 0)
1202 goto out;
1203 continue;
1204 }
1205 ret = try_to_free_disk_space();
1206 if (ret < 0)
1207 goto out;
1208 if (snapshot_removal_status != HS_READY) {
1209 stop_create_process();
1210 continue;
1211 }
1212 restart_create_process();
1213 switch (snapshot_creation_status) {
1214 case HS_READY:
1215 if (!next_snapshot_is_due())
1216 continue;
1217 ret = pre_create_hook();
1218 if (ret < 0)
1219 goto out;
1220 continue;
1221 case HS_PRE_RUNNING:
1222 case HS_RUNNING:
1223 case HS_POST_RUNNING:
1224 continue;
1225 case HS_PRE_SUCCESS:
1226 if (!name_of_reference_snapshot) {
1227 free_rsync_argv(rsync_argv);
1228 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1229 }
1230 ret = create_snapshot(rsync_argv);
1231 if (ret < 0)
1232 goto out;
1233 continue;
1234 case HS_NEEDS_RESTART:
1235 if (!next_snapshot_is_due())
1236 continue;
1237 ret = create_snapshot(rsync_argv);
1238 if (ret < 0)
1239 goto out;
1240 continue;
1241 case HS_SUCCESS:
1242 ret = post_create_hook();
1243 if (ret < 0)
1244 goto out;
1245 continue;
1246 }
1247 }
1248 out:
1249 return ret;
1250 }
1251
1252 static void exit_hook(int exit_code)
1253 {
1254 int fds[3] = {0, 0, 0};
1255 char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1256 pid_t pid;
1257
1258 DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1259 dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1260 }
1261
1262 static int com_run(void)
1263 {
1264 int ret;
1265
1266 if (conf.dry_run_given) {
1267 DSS_ERROR_LOG("dry_run not supported by this command\n");
1268 return -E_SYNTAX;
1269 }
1270 ret = install_sighandler(SIGHUP);
1271 if (ret < 0)
1272 return ret;
1273 ret = select_loop();
1274 if (ret >= 0) /* impossible */
1275 ret = -E_BUG;
1276 exit_hook(ret);
1277 return ret;
1278 }
1279
1280 static int com_prune(void)
1281 {
1282 int ret;
1283 struct snapshot_list sl;
1284 struct snapshot *victim;
1285 struct disk_space ds;
1286 const char *why;
1287
1288 ret = get_disk_space(".", &ds);
1289 if (ret < 0)
1290 return ret;
1291 log_disk_space(&ds);
1292 dss_get_snapshot_list(&sl);
1293 why = "outdated";
1294 victim = find_outdated_snapshot(&sl);
1295 if (victim)
1296 goto rm;
1297 why = "redundant";
1298 victim = find_redundant_snapshot(&sl);
1299 if (victim)
1300 goto rm;
1301 ret = 0;
1302 goto out;
1303 rm:
1304 if (conf.dry_run_given) {
1305 dss_msg("%s snapshot %s (interval = %i)\n",
1306 why, victim->name, victim->interval);
1307 ret = 0;
1308 goto out;
1309 }
1310 ret = pre_remove_hook(victim, why);
1311 if (ret < 0)
1312 goto out;
1313 if (snapshot_removal_status == HS_PRE_RUNNING) {
1314 ret = wait_for_remove_process();
1315 if (ret < 0)
1316 goto out;
1317 if (snapshot_removal_status != HS_PRE_SUCCESS)
1318 goto out;
1319 }
1320 ret = exec_rm();
1321 if (ret < 0)
1322 goto out;
1323 ret = wait_for_remove_process();
1324 if (ret < 0)
1325 goto out;
1326 if (snapshot_removal_status != HS_SUCCESS)
1327 goto out;
1328 ret = post_remove_hook();
1329 if (ret < 0)
1330 goto out;
1331 if (snapshot_removal_status != HS_POST_RUNNING)
1332 goto out;
1333 ret = wait_for_remove_process();
1334 if (ret < 0)
1335 goto out;
1336 ret = 1;
1337 out:
1338 free_snapshot_list(&sl);
1339 return ret;
1340 }
1341
1342 static int com_create(void)
1343 {
1344 int ret, status;
1345 char **rsync_argv;
1346
1347 if (conf.dry_run_given) {
1348 int i;
1349 char *msg = NULL;
1350 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1351 for (i = 0; rsync_argv[i]; i++) {
1352 char *tmp = msg;
1353 msg = make_message("%s%s%s", tmp? tmp : "",
1354 tmp? " " : "", rsync_argv[i]);
1355 free(tmp);
1356 }
1357 free_rsync_argv(rsync_argv);
1358 dss_msg("%s\n", msg);
1359 free(msg);
1360 return 1;
1361 }
1362 ret = pre_create_hook();
1363 if (ret < 0)
1364 return ret;
1365 if (create_pid) {
1366 ret = wait_for_process(create_pid, &status);
1367 if (ret < 0)
1368 return ret;
1369 ret = handle_pre_create_hook_exit(status);
1370 if (ret <= 0) /* error, or pre-create failed */
1371 return ret;
1372 }
1373 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1374 ret = create_snapshot(rsync_argv);
1375 if (ret < 0)
1376 goto out;
1377 ret = wait_for_process(create_pid, &status);
1378 if (ret < 0)
1379 goto out;
1380 ret = handle_rsync_exit(status);
1381 if (ret < 0)
1382 goto out;
1383 post_create_hook();
1384 if (create_pid)
1385 ret = wait_for_process(create_pid, &status);
1386 out:
1387 free_rsync_argv(rsync_argv);
1388 return ret;
1389 }
1390
1391 static int com_ls(void)
1392 {
1393 int i;
1394 struct snapshot_list sl;
1395 struct snapshot *s;
1396
1397 dss_get_snapshot_list(&sl);
1398 FOR_EACH_SNAPSHOT(s, i, &sl) {
1399 int64_t d = 0;
1400 if (s->flags & SS_COMPLETE)
1401 d = (s->completion_time - s->creation_time) / 60;
1402 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1403 };
1404 free_snapshot_list(&sl);
1405 return 1;
1406 }
1407
1408 static int setup_signal_handling(void)
1409 {
1410 int ret;
1411
1412 DSS_INFO_LOG("setting up signal handlers\n");
1413 signal_pipe = signal_init(); /* always successful */
1414 ret = install_sighandler(SIGINT);
1415 if (ret < 0)
1416 return ret;
1417 ret = install_sighandler(SIGTERM);
1418 if (ret < 0)
1419 return ret;
1420 return install_sighandler(SIGCHLD);
1421 }
1422
1423 /**
1424 * The main function of dss.
1425 *
1426 * \param argc Usual argument count.
1427 * \param argv Usual argument vector.
1428 */
1429 int main(int argc, char **argv)
1430 {
1431 int ret;
1432 struct cmdline_parser_params params = {
1433 .override = 0,
1434 .initialize = 1,
1435 .check_required = 0,
1436 .check_ambiguity = 0,
1437 .print_errors = 1
1438 };
1439
1440 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1441 ret = parse_config_file(0);
1442 if (ret < 0)
1443 goto out;
1444 if (ret == 0) { /* no config file given */
1445 /*
1446 * Parse the command line options again, but this time check
1447 * that all required options are given.
1448 */
1449 params = (struct cmdline_parser_params) {
1450 .override = 1,
1451 .initialize = 1,
1452 .check_required = 1,
1453 .check_ambiguity = 1,
1454 .print_errors = 1
1455 };
1456 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1457 }
1458 if (conf.daemon_given)
1459 daemon_init();
1460 ret = change_to_dest_dir();
1461 if (ret < 0)
1462 goto out;
1463 dump_dss_config("startup");
1464 ret = setup_signal_handling();
1465 if (ret < 0)
1466 goto out;
1467 ret = call_command_handler();
1468 out:
1469 if (ret < 0)
1470 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1471 exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1472 }