]> git.tuebingen.mpg.de Git - dss.git/blob - dss.c
Remove some debug messages.
[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 "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 *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 void dss_get_snapshot_list(struct snapshot_list *sl)
230 {
231         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
232 }
233
234 static int64_t compute_next_snapshot_time(void)
235 {
236         int64_t x = 0, now = get_current_time(), unit_interval
237                 = 24 * 3600 * conf.unit_interval_arg, ret;
238         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
239                 num_complete_snapshots = 0;
240         int i;
241         struct snapshot *s = NULL;
242         struct snapshot_list sl;
243
244         dss_get_snapshot_list(&sl);
245         FOR_EACH_SNAPSHOT(s, i, &sl) {
246                 if (!(s->flags & SS_COMPLETE))
247                         continue;
248                 num_complete_snapshots++;
249                 x += s->completion_time - s->creation_time;
250         }
251         assert(x >= 0);
252
253         ret = now;
254         if (num_complete_snapshots == 0)
255                 goto out;
256         x /= num_complete_snapshots; /* avg time to create one snapshot */
257         if (unit_interval < x * wanted) /* oops, no sleep at all */
258                 goto out;
259         ret = s->completion_time + unit_interval / wanted - x;
260 out:
261         free_snapshot_list(&sl);
262         return ret;
263 }
264
265 static inline void invalidate_next_snapshot_time(void)
266 {
267         next_snapshot_time = 0;
268 }
269
270 static inline int next_snapshot_time_is_valid(void)
271 {
272         return next_snapshot_time != 0;
273 }
274
275 static int next_snapshot_is_due(void)
276 {
277         int64_t now = get_current_time();
278
279         if (!next_snapshot_time_is_valid())
280                 next_snapshot_time = compute_next_snapshot_time();
281         if (next_snapshot_time <= now) {
282                 DSS_DEBUG_LOG("next snapshot: now\n");
283                 return 1;
284         }
285         DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
286                 next_snapshot_time - now);
287         return 0;
288 }
289
290 static void pre_create_hook(void)
291 {
292         assert(snapshot_creation_status == HS_READY);
293         /* make sure that the next snapshot time will be recomputed */
294         invalidate_next_snapshot_time();
295         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
296         dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
297         snapshot_creation_status = HS_PRE_RUNNING;
298 }
299
300 static void pre_remove_hook(struct snapshot *s, const char *why)
301 {
302         char *cmd;
303
304         if (!s)
305                 return;
306         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
307         assert(snapshot_removal_status == HS_READY);
308         assert(remove_pid == 0);
309         assert(!snapshot_currently_being_removed);
310
311         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
312         *snapshot_currently_being_removed = *s;
313         snapshot_currently_being_removed->name = dss_strdup(s->name);
314
315         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
316                 conf.dest_dir_arg, s->name);
317         DSS_DEBUG_LOG("executing %s\n", cmd);
318         dss_exec_cmdline_pid(&remove_pid, cmd);
319         free(cmd);
320         snapshot_removal_status = HS_PRE_RUNNING;
321 }
322
323 static int exec_rm(void)
324 {
325         struct snapshot *s = snapshot_currently_being_removed;
326         char *new_name = being_deleted_name(s);
327         char *argv[4];
328         int ret;
329
330         argv[0] = "rm";
331         argv[1] = "-rf";
332         argv[2] = new_name;
333         argv[3] = NULL;
334
335         assert(snapshot_removal_status == HS_PRE_SUCCESS);
336         assert(remove_pid == 0);
337
338         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
339         ret = dss_rename(s->name, new_name);
340         if (ret < 0)
341                 goto out;
342         dss_exec(&remove_pid, argv[0], argv);
343         snapshot_removal_status = HS_RUNNING;
344 out:
345         free(new_name);
346         return ret;
347 }
348
349 static int snapshot_is_being_created(struct snapshot *s)
350 {
351         return s->creation_time == current_snapshot_creation_time;
352 }
353
354 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
355 {
356         struct snapshot *s;
357         int i;
358
359         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
360         FOR_EACH_SNAPSHOT(s, i, sl) {
361                 if (snapshot_is_being_created(s))
362                         continue;
363                 /*
364                  * We know that no rm is currently running, so if s is marked
365                  * as being deleted, a previously started rm must have failed.
366                  */
367                 if (s->flags & SS_BEING_DELETED)
368                         return s;
369
370                 if (s->flags & SS_COMPLETE) /* good snapshot */
371                         continue;
372                 /*
373                  * This snapshot is incomplete and it is not the snapshot
374                  * currently being created. However, we must not remove it if
375                  * rsync is about to be restarted. As only the newest snapshot
376                  * can be restarted, this snapshot is orphaned if it is not the
377                  * newest snapshot or if we are not about to restart rsync.
378                  */
379                 if (get_newest_snapshot(sl) != s)
380                         return s;
381                 if (snapshot_creation_status != HS_NEEDS_RESTART)
382                         return s;
383         }
384         /* no orphaned snapshots */
385         return NULL;
386 }
387
388 static int is_reference_snapshot(struct snapshot *s)
389 {
390         if (!name_of_reference_snapshot)
391                 return 0;
392         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
393 }
394
395 /*
396  * return: 0: no redundant snapshots, 1: rm process started, negative: error
397  */
398 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
399 {
400         int i, interval;
401         struct snapshot *s;
402         unsigned missing = 0;
403
404         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
405         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
406                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
407                 unsigned num = sl->interval_count[interval];
408                 struct snapshot *victim = NULL, *prev = NULL;
409                 int64_t score = LONG_MAX;
410
411                 if (keep >= num)
412                         missing += keep - num;
413                 if (keep + missing >= num)
414                         continue;
415                 /* redundant snapshot in this interval, pick snapshot with lowest score */
416                 FOR_EACH_SNAPSHOT(s, i, sl) {
417                         int64_t this_score;
418
419                         if (snapshot_is_being_created(s))
420                                 continue;
421                         if (is_reference_snapshot(s))
422                                 continue;
423                         if (s->interval > interval) {
424                                 prev = s;
425                                 continue;
426                         }
427                         if (s->interval < interval)
428                                 break;
429                         if (!victim) {
430                                 victim = s;
431                                 prev = s;
432                                 continue;
433                         }
434                         assert(prev);
435                         /* check if s is a better victim */
436                         this_score = s->creation_time - prev->creation_time;
437                         assert(this_score >= 0);
438                         if (this_score < score) {
439                                 score = this_score;
440                                 victim = s;
441                         }
442                         prev = s;
443                 }
444                 assert(victim);
445                 return victim;
446         }
447         return NULL;
448 }
449
450 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
451 {
452         int i;
453         struct snapshot *s;
454
455         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
456                 conf.num_intervals_arg);
457         FOR_EACH_SNAPSHOT(s, i, sl) {
458                 if (snapshot_is_being_created(s))
459                         continue;
460                 if (is_reference_snapshot(s))
461                         continue;
462                 if (s->interval < conf.num_intervals_arg)
463                         continue;
464                 return s;
465         }
466         return NULL;
467 }
468
469 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
470 {
471         int i;
472         struct snapshot *s;
473         FOR_EACH_SNAPSHOT(s, i, sl) {
474                 if (snapshot_is_being_created(s))
475                         continue;
476                 if (is_reference_snapshot(s))
477                         continue;
478                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
479                 return s;
480         }
481         return NULL;
482 }
483
484 static int rename_incomplete_snapshot(int64_t start)
485 {
486         char *old_name;
487         int ret;
488
489         free(path_to_last_complete_snapshot);
490         ret = complete_name(start, get_current_time(),
491                 &path_to_last_complete_snapshot);
492         if (ret < 0)
493                 return ret;
494         old_name = incomplete_name(start);
495         ret = dss_rename(old_name, path_to_last_complete_snapshot);
496         if (ret >= 0)
497                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
498                         path_to_last_complete_snapshot);
499         free(old_name);
500         return ret;
501 }
502
503 static int try_to_free_disk_space(void)
504 {
505         int ret;
506         struct snapshot_list sl;
507         struct snapshot *victim;
508         struct timeval now;
509         const char *why;
510         int low_disk_space;
511
512         ret = disk_space_low(NULL);
513         if (ret < 0)
514                 return ret;
515         low_disk_space = ret;
516         gettimeofday(&now, NULL);
517         if (tv_diff(&next_removal_check, &now, NULL) > 0)
518                 return 0;
519         if (!low_disk_space) {
520                 if (conf.keep_redundant_given)
521                         return 0;
522                 if (snapshot_creation_status != HS_READY)
523                         return 0;
524                 if (next_snapshot_is_due())
525                         return 0;
526         }
527         dss_get_snapshot_list(&sl);
528         ret = 0;
529         if (!low_disk_space && sl.num_snapshots <= 1)
530                 goto out;
531         why = "outdated";
532         victim = find_outdated_snapshot(&sl);
533         if (victim)
534                 goto remove;
535         why = "redundant";
536         victim = find_redundant_snapshot(&sl);
537         if (victim)
538                 goto remove;
539         /* try harder only if disk space is low */
540         if (!low_disk_space)
541                 goto out;
542         why = "orphaned";
543         victim = find_orphaned_snapshot(&sl);
544         if (victim)
545                 goto remove;
546         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
547         victim = find_oldest_removable_snapshot(&sl);
548         if (victim)
549                 goto remove;
550         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
551         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
552         goto out;
553 remove:
554         pre_remove_hook(victim, why);
555 out:
556         free_snapshot_list(&sl);
557         return ret;
558 }
559
560 static void post_create_hook(void)
561 {
562         char *cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
563                 conf.dest_dir_arg, path_to_last_complete_snapshot);
564         DSS_NOTICE_LOG("executing %s\n", cmd);
565         dss_exec_cmdline_pid(&create_pid, cmd);
566         free(cmd);
567         snapshot_creation_status = HS_POST_RUNNING;
568 }
569
570 static void post_remove_hook(void)
571 {
572         char *cmd;
573         struct snapshot *s = snapshot_currently_being_removed;
574
575         assert(s);
576
577         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
578                 conf.dest_dir_arg, s->name);
579         DSS_NOTICE_LOG("executing %s\n", cmd);
580         dss_exec_cmdline_pid(&remove_pid, cmd);
581         free(cmd);
582         snapshot_removal_status = HS_POST_RUNNING;
583 }
584
585 static void dss_kill(pid_t pid, int sig, const char *msg)
586 {
587         const char *signame, *process_name;
588
589         if (pid == 0)
590                 return;
591         switch (sig) {
592         case SIGTERM: signame = "TERM"; break;
593         case SIGSTOP: signame = "STOP"; break;
594         case SIGCONT: signame = "CONT"; break;
595         default: signame = "????";
596         }
597
598         if (pid == create_pid)
599                 process_name = "create";
600         else if (pid == remove_pid)
601                 process_name = "remove";
602         else process_name = "??????";
603
604         if (msg)
605                 DSS_INFO_LOG("%s\n", msg);
606         DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n",
607                 sig, signame, (int)pid, process_name);
608         if (kill(pid, sig) >= 0)
609                 return;
610         DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
611                 sig, signame, (int)pid, process_name);
612 }
613
614 static void stop_create_process(void)
615 {
616         if (create_process_stopped)
617                 return;
618         dss_kill(create_pid, SIGSTOP, "suspending create process");
619         create_process_stopped = 1;
620 }
621
622 static void restart_create_process(void)
623 {
624         if (!create_process_stopped)
625                 return;
626         dss_kill(create_pid, SIGCONT, "resuming create process");
627         create_process_stopped = 0;
628 }
629
630 /**
631  * Print a log message about the exit status of a child.
632  */
633 static void log_termination_msg(pid_t pid, int status)
634 {
635         if (WIFEXITED(status))
636                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
637                         WEXITSTATUS(status));
638         else if (WIFSIGNALED(status))
639                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
640                         WTERMSIG(status));
641         else
642                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
643 }
644
645 static int wait_for_process(pid_t pid, int *status)
646 {
647         int ret;
648
649         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
650         for (;;) {
651                 fd_set rfds;
652
653                 FD_ZERO(&rfds);
654                 FD_SET(signal_pipe, &rfds);
655                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
656                 if (ret < 0)
657                         break;
658                 ret = next_signal();
659                 if (!ret)
660                         continue;
661                 if (ret == SIGCHLD) {
662                         ret = waitpid(pid, status, 0);
663                         if (ret >= 0)
664                                 break;
665                         if (errno != EINTR) { /* error */
666                                 ret = -ERRNO_TO_DSS_ERROR(errno);
667                                 break;
668                         }
669                 }
670                 /* SIGINT or SIGTERM */
671                 dss_kill(pid, SIGTERM, "killing child process");
672         }
673         if (ret < 0)
674                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
675         else
676                 log_termination_msg(pid, *status);
677         return ret;
678 }
679
680 static void handle_pre_remove_exit(int status)
681 {
682         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
683                 snapshot_removal_status = HS_READY;
684                 gettimeofday(&next_removal_check, NULL);
685                 next_removal_check.tv_sec += 60;
686                 return;
687         }
688         snapshot_removal_status = HS_PRE_SUCCESS;
689 }
690
691 static int handle_rm_exit(int status)
692 {
693         if (!WIFEXITED(status)) {
694                 snapshot_removal_status = HS_READY;
695                 return -E_INVOLUNTARY_EXIT;
696         }
697         if (WEXITSTATUS(status)) {
698                 snapshot_removal_status = HS_READY;
699                 return -E_BAD_EXIT_CODE;
700         }
701         snapshot_removal_status = HS_SUCCESS;
702         return 1;
703 }
704
705 static void handle_post_remove_exit(void)
706 {
707         snapshot_removal_status = HS_READY;
708 }
709
710 static int handle_remove_exit(int status)
711 {
712         int ret;
713         struct snapshot *s = snapshot_currently_being_removed;
714
715         assert(s);
716         switch (snapshot_removal_status) {
717         case HS_PRE_RUNNING:
718                 handle_pre_remove_exit(status);
719                 ret = 1;
720                 break;
721         case HS_RUNNING:
722                 ret = handle_rm_exit(status);
723                 break;
724         case HS_POST_RUNNING:
725                 handle_post_remove_exit();
726                 ret = 1;
727                 break;
728         default:
729                 ret = -E_BUG;
730         }
731         if (snapshot_removal_status == HS_READY) {
732                 free(s->name);
733                 free(s);
734                 snapshot_currently_being_removed = NULL;
735         }
736         remove_pid = 0;
737         return ret;
738 }
739
740 static int wait_for_remove_process(void)
741 {
742         int status, ret;
743
744         assert(remove_pid);
745         assert(
746                 snapshot_removal_status == HS_PRE_RUNNING ||
747                 snapshot_removal_status == HS_RUNNING ||
748                 snapshot_removal_status == HS_POST_RUNNING
749         );
750         ret = wait_for_process(remove_pid, &status);
751         if (ret < 0)
752                 return ret;
753         return handle_remove_exit(status);
754 }
755
756 static int handle_rsync_exit(int status)
757 {
758         int es, ret;
759
760         if (!WIFEXITED(status)) {
761                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
762                 ret = -E_INVOLUNTARY_EXIT;
763                 snapshot_creation_status = HS_READY;
764                 goto out;
765         }
766         es = WEXITSTATUS(status);
767         /*
768          * Restart rsync on non-fatal errors:
769          * 12: Error in rsync protocol data stream
770          * 13: Errors with program diagnostics
771          */
772         if (es == 12 || es == 13) {
773                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
774                         (int)create_pid, es);
775                 snapshot_creation_status = HS_NEEDS_RESTART;
776                 next_snapshot_time = get_current_time() + 60;
777                 ret = 1;
778                 goto out;
779         }
780         if (es != 0 && es != 23 && es != 24) {
781                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
782                 ret = -E_BAD_EXIT_CODE;
783                 snapshot_creation_status = HS_READY;
784                 goto out;
785         }
786         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
787         if (ret < 0)
788                 goto out;
789         snapshot_creation_status = HS_SUCCESS;
790         free(name_of_reference_snapshot);
791         name_of_reference_snapshot = NULL;
792 out:
793         create_process_stopped = 0;
794         return ret;
795 }
796
797 static int handle_pre_create_hook_exit(int status)
798 {
799         int es, ret;
800         static int warn_count;
801
802         if (!WIFEXITED(status)) {
803                 snapshot_creation_status = HS_READY;
804                 ret = -E_INVOLUNTARY_EXIT;
805                 goto out;
806         }
807         es = WEXITSTATUS(status);
808         if (es) {
809                 if (!warn_count--) {
810                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
811                                 conf.pre_create_hook_arg, es);
812                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
813                         warn_count = 60; /* warn only once per hour */
814                 }
815                 next_snapshot_time = get_current_time() + 60;
816                 snapshot_creation_status = HS_READY;
817                 ret = 0;
818                 goto out;
819         }
820         warn_count = 0;
821         snapshot_creation_status = HS_PRE_SUCCESS;
822         ret = 1;
823 out:
824         return ret;
825 }
826
827 static int handle_sigchld(void)
828 {
829         pid_t pid;
830         int status, ret = reap_child(&pid, &status);
831
832         if (ret <= 0)
833                 return ret;
834
835         if (pid == create_pid) {
836                 switch (snapshot_creation_status) {
837                 case HS_PRE_RUNNING:
838                         ret = handle_pre_create_hook_exit(status);
839                         break;
840                 case HS_RUNNING:
841                         ret = handle_rsync_exit(status);
842                         break;
843                 case HS_POST_RUNNING:
844                         snapshot_creation_status = HS_READY;
845                         ret = 1;
846                         break;
847                 default:
848                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
849                                 snapshot_creation_status);
850                         return -E_BUG;
851                 }
852                 create_pid = 0;
853                 return ret;
854         }
855         if (pid == remove_pid) {
856                 ret = handle_remove_exit(status);
857                 if (ret < 0)
858                         return ret;
859                 return ret;
860         }
861         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
862         return -E_BUG;
863 }
864
865 static int check_config(void)
866 {
867         if (conf.unit_interval_arg <= 0) {
868                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
869                 return -E_INVALID_NUMBER;
870         }
871         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
872         if (conf.num_intervals_arg <= 0) {
873                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
874                 return -E_INVALID_NUMBER;
875         }
876         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
877         return 1;
878 }
879
880 /*
881  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
882  * file was read successfully.
883  */
884 static int parse_config_file(int override)
885 {
886         int ret, config_file_exists;
887         char *config_file;
888         struct stat statbuf;
889         char *old_logfile_arg = NULL;
890         int old_daemon_given = 0;
891
892         if (conf.config_file_given)
893                 config_file = dss_strdup(conf.config_file_arg);
894         else {
895                 char *home = get_homedir();
896                 config_file = make_message("%s/.dssrc", home);
897                 free(home);
898         }
899         if (override) { /* SIGHUP */
900                 if (conf.logfile_given)
901                         old_logfile_arg = dss_strdup(conf.logfile_arg);
902                 old_daemon_given = conf.daemon_given;
903         }
904
905         config_file_exists = !stat(config_file, &statbuf);
906         if (!config_file_exists && conf.config_file_given) {
907                 ret = -ERRNO_TO_DSS_ERROR(errno);
908                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
909                 goto out;
910         }
911         if (config_file_exists) {
912                 struct cmdline_parser_params params;
913                 params.override = override;
914                 params.initialize = 0;
915                 params.check_required = 1;
916                 params.check_ambiguity = 0;
917                 params.print_errors = 1;
918                 if (override) { /* invalidate all rsync options */
919                         int i;
920
921                         for (i = 0; i < conf.rsync_option_given; i++) {
922                                 free(conf.rsync_option_arg[i]);
923                                 conf.rsync_option_arg[i] = NULL;
924                         }
925                         conf.rsync_option_given = 0;
926                 }
927                 cmdline_parser_config_file(config_file, &conf, &params);
928         }
929         ret = check_config();
930         if (ret < 0)
931                 goto out;
932         if (override) {
933                 /* don't change daemon mode on SIGHUP */
934                 conf.daemon_given = old_daemon_given;
935                 close_log(logfile);
936                 logfile = NULL;
937                 if (conf.logfile_given)
938                         free(old_logfile_arg);
939                 else if (conf.daemon_given) { /* re-use old logfile */
940                         conf.logfile_arg = old_logfile_arg;
941                         conf.logfile_given = 1;
942                 }
943         }
944         if (conf.logfile_given && conf.run_given && conf.daemon_given) {
945                 logfile = open_log(conf.logfile_arg);
946                 log_welcome(conf.loglevel_arg);
947         }
948         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
949         ret = config_file_exists;
950 out:
951         free(config_file);
952         if (ret < 0)
953                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
954         return ret;
955 }
956
957 static int change_to_dest_dir(void)
958 {
959         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
960         return dss_chdir(conf.dest_dir_arg);
961 }
962
963 static int handle_sighup(void)
964 {
965         int ret;
966
967         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
968         dump_dss_config("old");
969         ret = parse_config_file(1);
970         if (ret < 0)
971                 return ret;
972         dump_dss_config("reloaded");
973         invalidate_next_snapshot_time();
974         return change_to_dest_dir();
975 }
976
977 static int handle_signal(void)
978 {
979         int sig, ret = next_signal();
980
981         if (ret <= 0)
982                 goto out;
983         sig = ret;
984         switch (sig) {
985         case SIGINT:
986         case SIGTERM:
987                 restart_create_process();
988                 dss_kill(create_pid, SIGTERM, NULL);
989                 dss_kill(remove_pid, SIGTERM, NULL);
990                 ret = -E_SIGNAL;
991                 break;
992         case SIGHUP:
993                 ret = handle_sighup();
994                 break;
995         case SIGCHLD:
996                 ret = handle_sigchld();
997                 break;
998         }
999 out:
1000         if (ret < 0)
1001                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
1002         return ret;
1003 }
1004
1005 /*
1006  * We can not use rsync locally if the local user is different from the remote
1007  * user or if the src dir is not on the local host (or both).
1008  */
1009 static int use_rsync_locally(char *logname)
1010 {
1011         char *h = conf.remote_host_arg;
1012
1013         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1014                 return 0;
1015         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1016                 return 0;
1017         return 1;
1018 }
1019
1020 static int rename_resume_snap(int64_t creation_time)
1021 {
1022         struct snapshot_list sl;
1023         struct snapshot *s = NULL;
1024         char *new_name = incomplete_name(creation_time);
1025         int ret;
1026         const char *why;
1027
1028         sl.num_snapshots = 0;
1029
1030         ret = 0;
1031         if (conf.no_resume_given)
1032                 goto out;
1033         dss_get_snapshot_list(&sl);
1034         /*
1035          * Snapshot recycling: We first look at the newest snapshot. If this
1036          * snapshot happens to be incomplete, the last rsync process was
1037          * aborted and we reuse this one. Otherwise we look at snapshots which
1038          * could be removed (outdated and redundant snapshots) as candidates
1039          * for recycling. If no outdated/redundant snapshot exists, we check if
1040          * there is an orphaned snapshot, which likely is useless anyway.
1041          *
1042          * Only if no existing snapshot is suitable for recycling, we bite the
1043          * bullet and create a new one.
1044          */
1045         s = get_newest_snapshot(&sl);
1046         if (!s) /* no snapshots at all */
1047                 goto out;
1048         /* re-use last snapshot if it is incomplete */
1049         why = "aborted";
1050         if ((s->flags & SS_COMPLETE) == 0)
1051                 goto out;
1052         why = "outdated";
1053         s = find_outdated_snapshot(&sl);
1054         if (s)
1055                 goto out;
1056         why = "redundant";
1057         s = find_redundant_snapshot(&sl);
1058         if (s)
1059                 goto out;
1060         why = "orphaned";
1061         s = find_orphaned_snapshot(&sl);
1062 out:
1063         if (s) {
1064                 DSS_INFO_LOG("reusing %s snapshot %s\n", why, s->name);
1065                 ret = dss_rename(s->name, new_name);
1066         }
1067         if (ret >= 0)
1068                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
1069         free(new_name);
1070         free_snapshot_list(&sl);
1071         return ret;
1072 }
1073
1074 static void create_rsync_argv(char ***argv, int64_t *num)
1075 {
1076         char *logname;
1077         int i = 0, j;
1078         struct snapshot_list sl;
1079
1080         dss_get_snapshot_list(&sl);
1081         assert(!name_of_reference_snapshot);
1082         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1083         free_snapshot_list(&sl);
1084
1085         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1086         (*argv)[i++] = dss_strdup("rsync");
1087         (*argv)[i++] = dss_strdup("-aq");
1088         (*argv)[i++] = dss_strdup("--delete");
1089         for (j = 0; j < conf.rsync_option_given; j++)
1090                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1091         if (name_of_reference_snapshot) {
1092                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1093                 (*argv)[i++] = make_message("--link-dest=../%s",
1094                         name_of_reference_snapshot);
1095         } else
1096                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1097         logname = dss_logname();
1098         if (use_rsync_locally(logname))
1099                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1100         else
1101                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1102                         conf.remote_user_arg : logname,
1103                         conf.remote_host_arg, conf.source_dir_arg);
1104         free(logname);
1105         *num = get_current_time();
1106         (*argv)[i++] = incomplete_name(*num);
1107         (*argv)[i++] = NULL;
1108         for (j = 0; j < i; j++)
1109                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1110 }
1111
1112 static void free_rsync_argv(char **argv)
1113 {
1114         int i;
1115
1116         if (!argv)
1117                 return;
1118         for (i = 0; argv[i]; i++)
1119                 free(argv[i]);
1120         free(argv);
1121 }
1122
1123 static int create_snapshot(char **argv)
1124 {
1125         int ret;
1126
1127         ret = rename_resume_snap(current_snapshot_creation_time);
1128         if (ret < 0)
1129                 return ret;
1130         dss_exec(&create_pid, argv[0], argv);
1131         snapshot_creation_status = HS_RUNNING;
1132         return ret;
1133 }
1134
1135 static int select_loop(void)
1136 {
1137         int ret;
1138         /* check every 60 seconds for free disk space */
1139         struct timeval tv;
1140         char **rsync_argv = NULL;
1141
1142         for (;;) {
1143                 fd_set rfds;
1144                 struct timeval *tvp;
1145
1146                 if (remove_pid)
1147                         tvp = NULL; /* sleep until rm hook/process dies */
1148                 else { /* sleep one minute */
1149                         tv.tv_sec = 60;
1150                         tv.tv_usec = 0;
1151                         tvp = &tv;
1152                 }
1153                 FD_ZERO(&rfds);
1154                 FD_SET(signal_pipe, &rfds);
1155                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1156                 if (ret < 0)
1157                         goto out;
1158                 if (FD_ISSET(signal_pipe, &rfds)) {
1159                         ret = handle_signal();
1160                         if (ret < 0)
1161                                 goto out;
1162                 }
1163                 if (remove_pid)
1164                         continue;
1165                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1166                         ret = exec_rm();
1167                         if (ret < 0)
1168                                 goto out;
1169                         continue;
1170                 }
1171                 if (snapshot_removal_status == HS_SUCCESS) {
1172                         post_remove_hook();
1173                         continue;
1174                 }
1175                 ret = try_to_free_disk_space();
1176                 if (ret < 0)
1177                         goto out;
1178                 if (snapshot_removal_status != HS_READY) {
1179                         stop_create_process();
1180                         continue;
1181                 }
1182                 restart_create_process();
1183                 switch (snapshot_creation_status) {
1184                 case HS_READY:
1185                         if (!next_snapshot_is_due())
1186                                 continue;
1187                         pre_create_hook();
1188                         continue;
1189                 case HS_PRE_RUNNING:
1190                 case HS_RUNNING:
1191                 case HS_POST_RUNNING:
1192                         continue;
1193                 case HS_PRE_SUCCESS:
1194                         if (!name_of_reference_snapshot) {
1195                                 free_rsync_argv(rsync_argv);
1196                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1197                         }
1198                         ret = create_snapshot(rsync_argv);
1199                         if (ret < 0)
1200                                 goto out;
1201                         continue;
1202                 case HS_NEEDS_RESTART:
1203                         if (!next_snapshot_is_due())
1204                                 continue;
1205                         ret = create_snapshot(rsync_argv);
1206                         if (ret < 0)
1207                                 goto out;
1208                         continue;
1209                 case HS_SUCCESS:
1210                         post_create_hook();
1211                         continue;
1212                 }
1213         }
1214 out:
1215         return ret;
1216 }
1217
1218 static void exit_hook(int exit_code)
1219 {
1220         char *argv[3];
1221         pid_t pid;
1222
1223         argv[0] = conf.exit_hook_arg;
1224         argv[1] = dss_strerror(-exit_code);
1225         argv[2] = NULL;
1226
1227         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1228         dss_exec(&pid, conf.exit_hook_arg, argv);
1229 }
1230
1231 static int com_run(void)
1232 {
1233         int ret;
1234
1235         if (conf.dry_run_given) {
1236                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1237                 return -E_SYNTAX;
1238         }
1239         ret = install_sighandler(SIGHUP);
1240         if (ret < 0)
1241                 return ret;
1242         ret = select_loop();
1243         if (ret >= 0) /* impossible */
1244                 ret = -E_BUG;
1245         exit_hook(ret);
1246         return ret;
1247 }
1248
1249 static int com_prune(void)
1250 {
1251         int ret;
1252         struct snapshot_list sl;
1253         struct snapshot *victim;
1254         struct disk_space ds;
1255         const char *why;
1256
1257         ret = get_disk_space(".", &ds);
1258         if (ret < 0)
1259                 return ret;
1260         log_disk_space(&ds);
1261         dss_get_snapshot_list(&sl);
1262         why = "outdated";
1263         victim = find_outdated_snapshot(&sl);
1264         if (victim)
1265                 goto rm;
1266         why = "redundant";
1267         victim = find_redundant_snapshot(&sl);
1268         if (victim)
1269                 goto rm;
1270         ret = 0;
1271         goto out;
1272 rm:
1273         if (conf.dry_run_given) {
1274                 dss_msg("%s snapshot %s (interval = %i)\n",
1275                         why, victim->name, victim->interval);
1276                 ret = 0;
1277                 goto out;
1278         }
1279         pre_remove_hook(victim, why);
1280         if (snapshot_removal_status == HS_PRE_RUNNING) {
1281                 ret = wait_for_remove_process();
1282                 if (ret < 0)
1283                         goto out;
1284                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1285                         goto out;
1286         }
1287         ret = exec_rm();
1288         if (ret < 0)
1289                 goto out;
1290         ret = wait_for_remove_process();
1291         if (ret < 0)
1292                 goto out;
1293         if (snapshot_removal_status != HS_SUCCESS)
1294                 goto out;
1295         post_remove_hook();
1296         if (snapshot_removal_status != HS_POST_RUNNING)
1297                 goto out;
1298         ret = wait_for_remove_process();
1299         if (ret < 0)
1300                 goto out;
1301         ret = 1;
1302 out:
1303         free_snapshot_list(&sl);
1304         return ret;
1305 }
1306
1307 static int com_create(void)
1308 {
1309         int ret, status;
1310         char **rsync_argv;
1311
1312         if (conf.dry_run_given) {
1313                 int i;
1314                 char *msg = NULL;
1315                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1316                 for (i = 0; rsync_argv[i]; i++) {
1317                         char *tmp = msg;
1318                         msg = make_message("%s%s%s", tmp? tmp : "",
1319                                 tmp? " " : "", rsync_argv[i]);
1320                         free(tmp);
1321                 }
1322                 free_rsync_argv(rsync_argv);
1323                 dss_msg("%s\n", msg);
1324                 free(msg);
1325                 return 1;
1326         }
1327         pre_create_hook();
1328         if (create_pid) {
1329                 ret = wait_for_process(create_pid, &status);
1330                 if (ret < 0)
1331                         return ret;
1332                 ret = handle_pre_create_hook_exit(status);
1333                 if (ret <= 0) /* error, or pre-create failed */
1334                         return ret;
1335         }
1336         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1337         ret = create_snapshot(rsync_argv);
1338         if (ret < 0)
1339                 goto out;
1340         ret = wait_for_process(create_pid, &status);
1341         if (ret < 0)
1342                 goto out;
1343         ret = handle_rsync_exit(status);
1344         if (ret < 0)
1345                 goto out;
1346         post_create_hook();
1347         if (create_pid)
1348                 ret = wait_for_process(create_pid, &status);
1349 out:
1350         free_rsync_argv(rsync_argv);
1351         return ret;
1352 }
1353
1354 static int com_ls(void)
1355 {
1356         int i;
1357         struct snapshot_list sl;
1358         struct snapshot *s;
1359
1360         dss_get_snapshot_list(&sl);
1361         FOR_EACH_SNAPSHOT(s, i, &sl) {
1362                 int64_t d = 0;
1363                 if (s->flags & SS_COMPLETE)
1364                         d = (s->completion_time - s->creation_time) / 60;
1365                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1366         };
1367         free_snapshot_list(&sl);
1368         return 1;
1369 }
1370
1371 static int setup_signal_handling(void)
1372 {
1373         int ret;
1374
1375         DSS_INFO_LOG("setting up signal handlers\n");
1376         signal_pipe = signal_init(); /* always successful */
1377         ret = install_sighandler(SIGINT);
1378         if (ret < 0)
1379                 return ret;
1380         ret = install_sighandler(SIGTERM);
1381         if (ret < 0)
1382                 return ret;
1383         return install_sighandler(SIGCHLD);
1384 }
1385
1386 /**
1387  * The main function of dss.
1388  *
1389  * \param argc Usual argument count.
1390  * \param argv Usual argument vector.
1391  */
1392 int main(int argc, char **argv)
1393 {
1394         int ret;
1395         struct cmdline_parser_params params;
1396
1397         params.override = 0;
1398         params.initialize = 1;
1399         params.check_required = 0;
1400         params.check_ambiguity = 0;
1401         params.print_errors = 1;
1402
1403         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1404         ret = parse_config_file(0);
1405         if (ret < 0)
1406                 goto out;
1407         if (ret == 0) { /* no config file given */
1408                 /*
1409                  * Parse the command line options again, but this time check
1410                  * that all required options are given.
1411                  */
1412                 struct cmdline_parser_params params;
1413                 params.override = 1;
1414                 params.initialize = 1;
1415                 params.check_required = 1;
1416                 params.check_ambiguity = 1;
1417                 params.print_errors = 1;
1418                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1419         }
1420         if (conf.daemon_given)
1421                 daemon_init();
1422         ret = change_to_dest_dir();
1423         if (ret < 0)
1424                 goto out;
1425         dump_dss_config("startup");
1426         ret = setup_signal_handling();
1427         if (ret < 0)
1428                 goto out;
1429         ret = call_command_handler();
1430 out:
1431         if (ret < 0)
1432                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1433         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1434 }