dss.c

   1 /*
   2  * Copyright (C) 2008-2011 Andre Noll <maan@tuebingen.mpg.de>
   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 /** How many times in a row the rsync command failed. */
  49 static int num_consecutive_rsync_errors;
  50 /** Process id of current pre-remove/rm/post-remove process. */
  51 static pid_t remove_pid;
  52 /** When the next snapshot is due. */
  53 static int64_t next_snapshot_time;
  54 /** When to try to remove something. */
  55 static struct timeval next_removal_check;
  56 /** Creation time of the snapshot currently being created. */
  57 static int64_t current_snapshot_creation_time;
  58 /** The snapshot currently being removed. */
  59 struct snapshot *snapshot_currently_being_removed;
  60 /** Needed by the post-create hook. */
  61 static char *path_to_last_complete_snapshot;
  62 static char *name_of_reference_snapshot;
  63 /** \sa \ref snap.h for details. */
  64 enum hook_status snapshot_creation_status;
  65 /** \sa \ref snap.h for details. */
  66 enum hook_status snapshot_removal_status;
  67
  68
  69 DEFINE_DSS_ERRLIST;
  70 static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
  71
  72 /* may be called with ds == NULL. */
  73 static int disk_space_low(struct disk_space *ds)
  74 {
  75         struct disk_space ds_struct;
  76
  77         if (!ds) {
  78                 int ret = get_disk_space(".", &ds_struct);
  79                 if (ret < 0)
  80                         return ret;
  81                 ds = &ds_struct;
  82         }
  83         if (conf.min_free_mb_arg)
  84                 if (ds->free_mb < conf.min_free_mb_arg)
  85                         return 1;
  86         if (conf.min_free_percent_arg)
  87                 if (ds->percent_free < conf.min_free_percent_arg)
  88                         return 1;
  89         if (conf.min_free_percent_inodes_arg)
  90                 if (ds->percent_free_inodes < conf.min_free_percent_inodes_arg)
  91                         return 1;
  92         return 0;
  93 }
  94
  95 static void dump_dss_config(const char *msg)
  96 {
  97         const char dash[] = "-----------------------------";
  98         int ret;
  99         FILE *log = logfile? logfile : stderr;
 100         struct disk_space ds;
 101         int64_t now = get_current_time();
 102
 103         if (conf.loglevel_arg > INFO)
 104                 return;
 105
 106         fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
 107         fprintf(log, "\n*** disk space ***\n\n");
 108         ret = get_disk_space(".", &ds);
 109         if (ret >= 0) {
 110                 DSS_INFO_LOG(("disk space low: %s\n", disk_space_low(&ds)?
 111                         "yes" : "no"));
 112                 log_disk_space(&ds);
 113         } else
 114                 DSS_ERROR_LOG(("can not get free disk space: %s\n",
 115                         dss_strerror(-ret)));
 116
 117         /* we continue on errors from get_disk_space */
 118
 119         fprintf(log, "\n*** command line and config file options ***\n\n");
 120         cmdline_parser_dump(log, &conf);
 121         fprintf(log, "\n*** internal state ***\n\n");
 122         fprintf(log,
 123                 "pid: %d\n"
 124                 "logile: %s\n"
 125                 "snapshot_currently_being_removed: %s\n"
 126                 "path_to_last_complete_snapshot: %s\n"
 127                 "reference_snapshot: %s\n"
 128                 "snapshot_creation_status: %s\n"
 129                 "snapshot_removal_status: %s\n"
 130                 "num_consecutive_rsync_errors: %d\n"
 131                 ,
 132                 (int) getpid(),
 133                 logfile? conf.logfile_arg : "stderr",
 134                 snapshot_currently_being_removed?
 135                         snapshot_currently_being_removed->name : "(none)",
 136                 path_to_last_complete_snapshot?
 137                         path_to_last_complete_snapshot : "(none)",
 138                 name_of_reference_snapshot?
 139                         name_of_reference_snapshot : "(none)",
 140                 hook_status_description[snapshot_creation_status],
 141                 hook_status_description[snapshot_removal_status],
 142                 num_consecutive_rsync_errors
 143         );
 144         if (create_pid != 0)
 145                 fprintf(log,
 146                         "create_pid: %" PRId32 "\n"
 147                         "create process is %sstopped\n"
 148                         ,
 149                         create_pid,
 150                         create_process_stopped? "" : "not "
 151                 );
 152         if (remove_pid != 0)
 153                 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
 154         if (next_snapshot_time != 0)
 155                 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
 156                         next_snapshot_time - now);
 157         if (current_snapshot_creation_time != 0)
 158                 fprintf(log, "current_snapshot_creation_time: %"
 159                         PRId64 " (%" PRId64 " seconds ago)\n",
 160                         current_snapshot_creation_time,
 161                         now - current_snapshot_creation_time
 162                 );
 163         if (next_removal_check.tv_sec != 0) {
 164                 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
 165                         (long long unsigned)next_removal_check.tv_sec,
 166                         now - (long long unsigned)next_removal_check.tv_sec
 167                 );
 168
 169         }
 170         fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
 171 }
 172
 173 /* a litte cpp magic helps to DRY */
 174 #define COMMANDS \
 175         COMMAND(ls) \
 176         COMMAND(create) \
 177         COMMAND(prune) \
 178         COMMAND(run) \
 179         COMMAND(kill) \
 180         COMMAND(reload) \
 181
 182 #define COMMAND(x) static int com_ ##x(void);
 183 COMMANDS
 184 #undef COMMAND
 185 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
 186 static int call_command_handler(void)
 187 {
 188         COMMANDS
 189         DSS_EMERG_LOG(("BUG: did not find command handler\n"));
 190         return -E_BUG;
 191 }
 192 #undef COMMAND
 193 #undef COMMANDS
 194
 195 static int loglevel = -1;
 196 static const char *location_file = NULL;
 197 static int         location_line = -1;
 198 static const char *location_func = NULL;
 199
 200 void dss_log_set_params(int ll, const char *file, int line, const char *func)
 201 {
 202         loglevel = ll;
 203         location_file = file;
 204         location_line = line;
 205         location_func = func;
 206 }
 207
 208 /**
 209  * The log function of dss.
 210  *
 211  * \param ll Loglevel.
 212  * \param fml Usual format string.
 213  *
 214  * All DSS_XXX_LOG() macros use this function.
 215  */
 216 __printf_1_2 void dss_log(const char* fmt,...)
 217 {
 218         va_list argp;
 219         FILE *outfd;
 220         struct tm *tm;
 221         time_t t1;
 222         char str[255] = "";
 223
 224         if (loglevel < conf.loglevel_arg)
 225                 return;
 226         outfd = logfile? logfile : stderr;
 227         time(&t1);
 228         tm = localtime(&t1);
 229         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 230         fprintf(outfd, "%s ", str);
 231         if (conf.loglevel_arg <= INFO)
 232                 fprintf(outfd, "%i: ", loglevel);
 233 #ifdef DSS_NO_FUNC_NAMES
 234         fprintf(outfd, "%s:%d: ", location_file, location_line);
 235 #else
 236         fprintf(outfd, "%s: ", location_func);
 237 #endif
 238         va_start(argp, fmt);
 239         vfprintf(outfd, fmt, argp);
 240         va_end(argp);
 241 }
 242
 243 /**
 244  * Print a message either to stdout or to the log file.
 245  */
 246 static __printf_1_2 void dss_msg(const char* fmt,...)
 247 {
 248         FILE *outfd = conf.daemon_given? logfile : stdout;
 249         va_list argp;
 250         va_start(argp, fmt);
 251         vfprintf(outfd, fmt, argp);
 252         va_end(argp);
 253 }
 254
 255 static char *get_config_file_name(void)
 256 {
 257         char *home, *config_file;
 258
 259         if (conf.config_file_given)
 260                 return dss_strdup(conf.config_file_arg);
 261         home = get_homedir();
 262         config_file = make_message("%s/.dssrc", home);
 263         free(home);
 264         return config_file;
 265 }
 266
 267 static int send_signal(int sig)
 268 {
 269         pid_t pid;
 270         char *config_file = get_config_file_name();
 271         int ret = get_dss_pid(config_file, &pid);
 272
 273         free(config_file);
 274         if (ret < 0)
 275                 return ret;
 276         if (conf.dry_run_given) {
 277                 dss_msg("%d\n", (int)pid);
 278                 return 0;
 279         }
 280         ret = kill(pid, sig);
 281         if (ret < 0)
 282                 return -ERRNO_TO_DSS_ERROR(errno);
 283         return 1;
 284 }
 285
 286 static int com_kill(void)
 287 {
 288         return send_signal(SIGTERM);
 289 }
 290
 291 static int com_reload(void)
 292 {
 293         return send_signal(SIGHUP);
 294 }
 295
 296 static void dss_get_snapshot_list(struct snapshot_list *sl)
 297 {
 298         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 299 }
 300
 301 static int64_t compute_next_snapshot_time(void)
 302 {
 303         int64_t x = 0, now = get_current_time(), unit_interval
 304                 = 24 * 3600 * conf.unit_interval_arg, ret;
 305         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 306                 num_complete_snapshots = 0;
 307         int i;
 308         struct snapshot *s = NULL;
 309         struct snapshot_list sl;
 310
 311         dss_get_snapshot_list(&sl);
 312         FOR_EACH_SNAPSHOT(s, i, &sl) {
 313                 if (!(s->flags & SS_COMPLETE))
 314                         continue;
 315                 num_complete_snapshots++;
 316                 x += s->completion_time - s->creation_time;
 317         }
 318         assert(x >= 0);
 319
 320         ret = now;
 321         if (num_complete_snapshots == 0)
 322                 goto out;
 323         x /= num_complete_snapshots; /* avg time to create one snapshot */
 324         if (unit_interval < x * wanted) /* oops, no sleep at all */
 325                 goto out;
 326         ret = s->completion_time + unit_interval / wanted - x;
 327 out:
 328         free_snapshot_list(&sl);
 329         return ret;
 330 }
 331
 332 static inline void invalidate_next_snapshot_time(void)
 333 {
 334         next_snapshot_time = 0;
 335 }
 336
 337 static inline int next_snapshot_time_is_valid(void)
 338 {
 339         return next_snapshot_time != 0;
 340 }
 341
 342 static int next_snapshot_is_due(void)
 343 {
 344         int64_t now = get_current_time();
 345
 346         if (!next_snapshot_time_is_valid())
 347                 next_snapshot_time = compute_next_snapshot_time();
 348         if (next_snapshot_time <= now) {
 349                 DSS_DEBUG_LOG(("next snapshot: now\n"));
 350                 return 1;
 351         }
 352         DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
 353                 next_snapshot_time - now));
 354         return 0;
 355 }
 356
 357 static void pre_create_hook(void)
 358 {
 359         assert(snapshot_creation_status == HS_READY);
 360         /* make sure that the next snapshot time will be recomputed */
 361         invalidate_next_snapshot_time();
 362         DSS_DEBUG_LOG(("executing %s\n", conf.pre_create_hook_arg));
 363         dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
 364         snapshot_creation_status = HS_PRE_RUNNING;
 365 }
 366
 367 static void pre_remove_hook(struct snapshot *s, const char *why)
 368 {
 369         char *cmd;
 370
 371         if (!s)
 372                 return;
 373         DSS_DEBUG_LOG(("%s snapshot %s\n", why, s->name));
 374         assert(snapshot_removal_status == HS_READY);
 375         assert(remove_pid == 0);
 376         assert(!snapshot_currently_being_removed);
 377
 378         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 379         *snapshot_currently_being_removed = *s;
 380         snapshot_currently_being_removed->name = dss_strdup(s->name);
 381
 382         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 383                 conf.dest_dir_arg, s->name);
 384         DSS_DEBUG_LOG(("executing %s\n", cmd));
 385         dss_exec_cmdline_pid(&remove_pid, cmd);
 386         free(cmd);
 387         snapshot_removal_status = HS_PRE_RUNNING;
 388 }
 389
 390 static int exec_rm(void)
 391 {
 392         struct snapshot *s = snapshot_currently_being_removed;
 393         char *new_name = being_deleted_name(s);
 394         char *argv[4];
 395         int ret;
 396
 397         argv[0] = "rm";
 398         argv[1] = "-rf";
 399         argv[2] = new_name;
 400         argv[3] = NULL;
 401
 402         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 403         assert(remove_pid == 0);
 404
 405         DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
 406         ret = dss_rename(s->name, new_name);
 407         if (ret < 0)
 408                 goto out;
 409         dss_exec(&remove_pid, argv[0], argv);
 410         snapshot_removal_status = HS_RUNNING;
 411 out:
 412         free(new_name);
 413         return ret;
 414 }
 415
 416 static int snapshot_is_being_created(struct snapshot *s)
 417 {
 418         return s->creation_time == current_snapshot_creation_time;
 419 }
 420
 421 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 422 {
 423         struct snapshot *s;
 424         int i;
 425
 426         DSS_DEBUG_LOG(("looking for orphaned snapshots\n"));
 427         FOR_EACH_SNAPSHOT(s, i, sl) {
 428                 if (snapshot_is_being_created(s))
 429                         continue;
 430                 /*
 431                  * We know that no rm is currently running, so if s is marked
 432                  * as being deleted, a previously started rm must have failed.
 433                  */
 434                 if (s->flags & SS_BEING_DELETED)
 435                         return s;
 436
 437                 if (s->flags & SS_COMPLETE) /* good snapshot */
 438                         continue;
 439                 /*
 440                  * This snapshot is incomplete and it is not the snapshot
 441                  * currently being created. However, we must not remove it if
 442                  * rsync is about to be restarted. As only the newest snapshot
 443                  * can be restarted, this snapshot is orphaned if it is not the
 444                  * newest snapshot or if we are not about to restart rsync.
 445                  */
 446                 if (get_newest_snapshot(sl) != s)
 447                         return s;
 448                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 449                         return s;
 450         }
 451         /* no orphaned snapshots */
 452         return NULL;
 453 }
 454
 455 static int is_reference_snapshot(struct snapshot *s)
 456 {
 457         if (!name_of_reference_snapshot)
 458                 return 0;
 459         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 460 }
 461
 462 /*
 463  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 464  */
 465 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 466 {
 467         int i, interval;
 468         struct snapshot *s;
 469         unsigned missing = 0;
 470
 471         DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
 472         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 473                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 474                 unsigned num = sl->interval_count[interval];
 475                 struct snapshot *victim = NULL, *prev = NULL;
 476                 int64_t score = LONG_MAX;
 477
 478                 if (keep >= num)
 479                         missing += keep - num;
 480                 if (keep + missing >= num)
 481                         continue;
 482                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 483                 FOR_EACH_SNAPSHOT(s, i, sl) {
 484                         int64_t this_score;
 485
 486                         if (snapshot_is_being_created(s))
 487                                 continue;
 488                         if (is_reference_snapshot(s))
 489                                 continue;
 490                         if (s->interval > interval) {
 491                                 prev = s;
 492                                 continue;
 493                         }
 494                         if (s->interval < interval)
 495                                 break;
 496                         if (!victim) {
 497                                 victim = s;
 498                                 prev = s;
 499                                 continue;
 500                         }
 501                         assert(prev);
 502                         /* check if s is a better victim */
 503                         this_score = s->creation_time - prev->creation_time;
 504                         assert(this_score >= 0);
 505                         if (this_score < score) {
 506                                 score = this_score;
 507                                 victim = s;
 508                         }
 509                         prev = s;
 510                 }
 511                 assert(victim);
 512                 return victim;
 513         }
 514         return NULL;
 515 }
 516
 517 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 518 {
 519         int i;
 520         struct snapshot *s;
 521
 522         DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
 523                 conf.num_intervals_arg));
 524         FOR_EACH_SNAPSHOT(s, i, sl) {
 525                 if (snapshot_is_being_created(s))
 526                         continue;
 527                 if (is_reference_snapshot(s))
 528                         continue;
 529                 if (s->interval < conf.num_intervals_arg)
 530                         continue;
 531                 return s;
 532         }
 533         return NULL;
 534 }
 535
 536 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 537 {
 538         int i, num_complete;
 539         struct snapshot *s, *ref = NULL;
 540
 541         num_complete = num_complete_snapshots(sl);
 542         if (num_complete <= conf.min_complete_arg)
 543                 return NULL;
 544         FOR_EACH_SNAPSHOT(s, i, sl) {
 545                 if (snapshot_is_being_created(s))
 546                         continue;
 547                 if (is_reference_snapshot(s)) { /* avoid this one */
 548                         ref = s;
 549                         continue;
 550                 }
 551                 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 552                 return s;
 553         }
 554         assert(ref);
 555         DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
 556         return ref;
 557 }
 558
 559 static int rename_incomplete_snapshot(int64_t start)
 560 {
 561         char *old_name;
 562         int ret;
 563         int64_t now;
 564
 565         /*
 566          * We don't want the dss_rename() below to fail with EEXIST because the
 567          * last complete snapshot was created (and completed) in the same
 568          * second as this one.
 569          */
 570         while ((now = get_current_time()) == start)
 571                 sleep(1);
 572         free(path_to_last_complete_snapshot);
 573         ret = complete_name(start, now, &path_to_last_complete_snapshot);
 574         if (ret < 0)
 575                 return ret;
 576         old_name = incomplete_name(start);
 577         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 578         if (ret >= 0)
 579                 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
 580                         path_to_last_complete_snapshot));
 581         free(old_name);
 582         return ret;
 583 }
 584
 585 static int try_to_free_disk_space(void)
 586 {
 587         int ret;
 588         struct snapshot_list sl;
 589         struct snapshot *victim;
 590         struct timeval now;
 591         const char *why;
 592         int low_disk_space;
 593
 594         ret = disk_space_low(NULL);
 595         if (ret < 0)
 596                 return ret;
 597         low_disk_space = ret;
 598         gettimeofday(&now, NULL);
 599         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 600                 return 0;
 601         if (!low_disk_space) {
 602                 if (conf.keep_redundant_given)
 603                         return 0;
 604                 if (snapshot_creation_status != HS_READY)
 605                         return 0;
 606                 if (next_snapshot_is_due())
 607                         return 0;
 608         }
 609         dss_get_snapshot_list(&sl);
 610         ret = 0;
 611         if (!low_disk_space && sl.num_snapshots <= 1)
 612                 goto out;
 613         why = "outdated";
 614         victim = find_outdated_snapshot(&sl);
 615         if (victim)
 616                 goto remove;
 617         why = "redundant";
 618         victim = find_redundant_snapshot(&sl);
 619         if (victim)
 620                 goto remove;
 621         /* try harder only if disk space is low */
 622         if (!low_disk_space)
 623                 goto out;
 624         why = "orphaned";
 625         victim = find_orphaned_snapshot(&sl);
 626         if (victim)
 627                 goto remove;
 628         DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 629         victim = find_oldest_removable_snapshot(&sl);
 630         if (victim)
 631                 goto remove;
 632         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 633         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 634         goto out;
 635 remove:
 636         pre_remove_hook(victim, why);
 637 out:
 638         free_snapshot_list(&sl);
 639         return ret;
 640 }
 641
 642 static void post_create_hook(void)
 643 {
 644         char *cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 645                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 646         DSS_NOTICE_LOG(("executing %s\n", cmd));
 647         dss_exec_cmdline_pid(&create_pid, cmd);
 648         free(cmd);
 649         snapshot_creation_status = HS_POST_RUNNING;
 650 }
 651
 652 static void post_remove_hook(void)
 653 {
 654         char *cmd;
 655         struct snapshot *s = snapshot_currently_being_removed;
 656
 657         assert(s);
 658
 659         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 660                 conf.dest_dir_arg, s->name);
 661         DSS_NOTICE_LOG(("executing %s\n", cmd));
 662         dss_exec_cmdline_pid(&remove_pid, cmd);
 663         free(cmd);
 664         snapshot_removal_status = HS_POST_RUNNING;
 665 }
 666
 667 static void dss_kill(pid_t pid, int sig, const char *msg)
 668 {
 669         const char *signame, *process_name;
 670
 671         if (pid == 0)
 672                 return;
 673         switch (sig) {
 674         case SIGTERM: signame = "TERM"; break;
 675         case SIGSTOP: signame = "STOP"; break;
 676         case SIGCONT: signame = "CONT"; break;
 677         default: signame = "????";
 678         }
 679
 680         if (pid == create_pid)
 681                 process_name = "create";
 682         else if (pid == remove_pid)
 683                 process_name = "remove";
 684         else process_name = "??????";
 685
 686         if (msg)
 687                 DSS_INFO_LOG(("%s\n", msg));
 688         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 689                 sig, signame, (int)pid, process_name));
 690         if (kill(pid, sig) >= 0)
 691                 return;
 692         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 693                 sig, signame, (int)pid, process_name));
 694 }
 695
 696 static void stop_create_process(void)
 697 {
 698         if (create_process_stopped)
 699                 return;
 700         dss_kill(create_pid, SIGSTOP, "suspending create process");
 701         create_process_stopped = 1;
 702 }
 703
 704 static void restart_create_process(void)
 705 {
 706         if (!create_process_stopped)
 707                 return;
 708         dss_kill(create_pid, SIGCONT, "resuming create process");
 709         create_process_stopped = 0;
 710 }
 711
 712 /**
 713  * Print a log message about the exit status of a child.
 714  */
 715 static void log_termination_msg(pid_t pid, int status)
 716 {
 717         if (WIFEXITED(status))
 718                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 719                         WEXITSTATUS(status)));
 720         else if (WIFSIGNALED(status))
 721                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 722                         WTERMSIG(status)));
 723         else
 724                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 725 }
 726
 727 static int wait_for_process(pid_t pid, int *status)
 728 {
 729         int ret;
 730
 731         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 732         for (;;) {
 733                 fd_set rfds;
 734
 735                 FD_ZERO(&rfds);
 736                 FD_SET(signal_pipe, &rfds);
 737                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 738                 if (ret < 0)
 739                         break;
 740                 ret = next_signal();
 741                 if (!ret)
 742                         continue;
 743                 if (ret == SIGCHLD) {
 744                         ret = waitpid(pid, status, 0);
 745                         if (ret >= 0)
 746                                 break;
 747                         if (errno != EINTR) { /* error */
 748                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 749                                 break;
 750                         }
 751                 }
 752                 /* SIGINT or SIGTERM */
 753                 dss_kill(pid, SIGTERM, "killing child process");
 754         }
 755         if (ret < 0)
 756                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 757         else
 758                 log_termination_msg(pid, *status);
 759         return ret;
 760 }
 761
 762 static void handle_pre_remove_exit(int status)
 763 {
 764         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 765                 snapshot_removal_status = HS_READY;
 766                 gettimeofday(&next_removal_check, NULL);
 767                 next_removal_check.tv_sec += 60;
 768                 return;
 769         }
 770         snapshot_removal_status = HS_PRE_SUCCESS;
 771 }
 772
 773 static int handle_rm_exit(int status)
 774 {
 775         if (!WIFEXITED(status)) {
 776                 snapshot_removal_status = HS_READY;
 777                 return -E_INVOLUNTARY_EXIT;
 778         }
 779         if (WEXITSTATUS(status)) {
 780                 snapshot_removal_status = HS_READY;
 781                 return -E_BAD_EXIT_CODE;
 782         }
 783         snapshot_removal_status = HS_SUCCESS;
 784         return 1;
 785 }
 786
 787 static void handle_post_remove_exit(void)
 788 {
 789         snapshot_removal_status = HS_READY;
 790 }
 791
 792 static int handle_remove_exit(int status)
 793 {
 794         int ret;
 795         struct snapshot *s = snapshot_currently_being_removed;
 796
 797         assert(s);
 798         switch (snapshot_removal_status) {
 799         case HS_PRE_RUNNING:
 800                 handle_pre_remove_exit(status);
 801                 ret = 1;
 802                 break;
 803         case HS_RUNNING:
 804                 ret = handle_rm_exit(status);
 805                 break;
 806         case HS_POST_RUNNING:
 807                 handle_post_remove_exit();
 808                 ret = 1;
 809                 break;
 810         default:
 811                 ret = -E_BUG;
 812         }
 813         if (snapshot_removal_status == HS_READY) {
 814                 free(s->name);
 815                 free(s);
 816                 snapshot_currently_being_removed = NULL;
 817         }
 818         remove_pid = 0;
 819         return ret;
 820 }
 821
 822 static int wait_for_remove_process(void)
 823 {
 824         int status, ret;
 825
 826         assert(remove_pid);
 827         assert(
 828                 snapshot_removal_status == HS_PRE_RUNNING ||
 829                 snapshot_removal_status == HS_RUNNING ||
 830                 snapshot_removal_status == HS_POST_RUNNING
 831         );
 832         ret = wait_for_process(remove_pid, &status);
 833         if (ret < 0)
 834                 return ret;
 835         return handle_remove_exit(status);
 836 }
 837
 838 static int handle_rsync_exit(int status)
 839 {
 840         int es, ret;
 841
 842         if (!WIFEXITED(status)) {
 843                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 844                 ret = -E_INVOLUNTARY_EXIT;
 845                 snapshot_creation_status = HS_READY;
 846                 goto out;
 847         }
 848         es = WEXITSTATUS(status);
 849         /*
 850          * Restart rsync on non-fatal errors:
 851          * 24: Partial transfer due to vanished source files
 852          */
 853         if (es != 0 && es != 24) {
 854                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 855                         es, ++num_consecutive_rsync_errors));
 856                 if (num_consecutive_rsync_errors > conf.max_rsync_errors_arg) {
 857                         ret = -E_TOO_MANY_RSYNC_ERRORS;
 858                         snapshot_creation_status = HS_READY;
 859                         goto out;
 860                 }
 861                 DSS_WARNING_LOG(("restarting rsync process\n"));
 862                 snapshot_creation_status = HS_NEEDS_RESTART;
 863                 next_snapshot_time = get_current_time() + 60;
 864                 ret = 1;
 865                 goto out;
 866         }
 867         num_consecutive_rsync_errors = 0;
 868         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 869         if (ret < 0)
 870                 goto out;
 871         snapshot_creation_status = HS_SUCCESS;
 872         free(name_of_reference_snapshot);
 873         name_of_reference_snapshot = NULL;
 874 out:
 875         create_process_stopped = 0;
 876         return ret;
 877 }
 878
 879 static int handle_pre_create_hook_exit(int status)
 880 {
 881         int es, ret;
 882         static int warn_count;
 883
 884         if (!WIFEXITED(status)) {
 885                 snapshot_creation_status = HS_READY;
 886                 ret = -E_INVOLUNTARY_EXIT;
 887                 goto out;
 888         }
 889         es = WEXITSTATUS(status);
 890         if (es) {
 891                 if (!warn_count--) {
 892                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
 893                                 conf.pre_create_hook_arg, es));
 894                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
 895                         warn_count = 60; /* warn only once per hour */
 896                 }
 897                 next_snapshot_time = get_current_time() + 60;
 898                 snapshot_creation_status = HS_READY;
 899                 ret = 0;
 900                 goto out;
 901         }
 902         warn_count = 0;
 903         snapshot_creation_status = HS_PRE_SUCCESS;
 904         ret = 1;
 905 out:
 906         return ret;
 907 }
 908
 909 static int handle_sigchld(void)
 910 {
 911         pid_t pid;
 912         int status, ret = reap_child(&pid, &status);
 913
 914         if (ret <= 0)
 915                 return ret;
 916
 917         if (pid == create_pid) {
 918                 switch (snapshot_creation_status) {
 919                 case HS_PRE_RUNNING:
 920                         ret = handle_pre_create_hook_exit(status);
 921                         break;
 922                 case HS_RUNNING:
 923                         ret = handle_rsync_exit(status);
 924                         break;
 925                 case HS_POST_RUNNING:
 926                         snapshot_creation_status = HS_READY;
 927                         ret = 1;
 928                         break;
 929                 default:
 930                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
 931                                 snapshot_creation_status));
 932                         return -E_BUG;
 933                 }
 934                 create_pid = 0;
 935                 return ret;
 936         }
 937         if (pid == remove_pid) {
 938                 ret = handle_remove_exit(status);
 939                 if (ret < 0)
 940                         return ret;
 941                 return ret;
 942         }
 943         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
 944         return -E_BUG;
 945 }
 946
 947 static int check_config(void)
 948 {
 949         if (conf.unit_interval_arg <= 0) {
 950                 DSS_ERROR_LOG(("bad unit interval: %i\n", conf.unit_interval_arg));
 951                 return -E_INVALID_NUMBER;
 952         }
 953         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", conf.unit_interval_arg));
 954         if (conf.num_intervals_arg <= 0 || conf.num_intervals_arg > 30) {
 955                 DSS_ERROR_LOG(("bad number of intervals: %i\n",
 956                         conf.num_intervals_arg));
 957                 return -E_INVALID_NUMBER;
 958         }
 959         DSS_DEBUG_LOG(("number of intervals: %i\n", conf.num_intervals_arg));
 960         return 1;
 961 }
 962
 963 /*
 964  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 965  * file was read successfully.
 966  */
 967 static int parse_config_file(int override)
 968 {
 969         int ret, config_file_exists;
 970         char *config_file = get_config_file_name();
 971         struct stat statbuf;
 972         char *old_logfile_arg = NULL;
 973         int old_daemon_given = 0;
 974
 975         if (override) { /* SIGHUP */
 976                 if (conf.logfile_given)
 977                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 978                 old_daemon_given = conf.daemon_given;
 979         }
 980
 981         config_file_exists = !stat(config_file, &statbuf);
 982         if (!config_file_exists && conf.config_file_given) {
 983                 ret = -ERRNO_TO_DSS_ERROR(errno);
 984                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
 985                 goto out;
 986         }
 987         if (config_file_exists) {
 988                 struct cmdline_parser_params params;
 989                 params.override = override;
 990                 params.initialize = 0;
 991                 params.check_required = 1;
 992                 params.check_ambiguity = 0;
 993                 params.print_errors = 1;
 994                 if (override) { /* invalidate all rsync options */
 995                         int i;
 996
 997                         for (i = 0; i < conf.rsync_option_given; i++) {
 998                                 free(conf.rsync_option_arg[i]);
 999                                 conf.rsync_option_arg[i] = NULL;
1000                         }
1001                         conf.rsync_option_given = 0;
1002                 }
1003                 cmdline_parser_config_file(config_file, &conf, &params);
1004         }
1005         ret = check_config();
1006         if (ret < 0)
1007                 goto out;
1008         if (override) {
1009                 /* don't change daemon mode on SIGHUP */
1010                 conf.daemon_given = old_daemon_given;
1011                 close_log(logfile);
1012                 logfile = NULL;
1013                 if (conf.logfile_given)
1014                         free(old_logfile_arg);
1015                 else if (conf.daemon_given) { /* re-use old logfile */
1016                         conf.logfile_arg = old_logfile_arg;
1017                         conf.logfile_given = 1;
1018                 }
1019         }
1020         if (conf.logfile_given && conf.run_given && conf.daemon_given) {
1021                 logfile = open_log(conf.logfile_arg);
1022                 log_welcome(conf.loglevel_arg);
1023         }
1024         DSS_DEBUG_LOG(("loglevel: %d\n", conf.loglevel_arg));
1025         ret = config_file_exists;
1026 out:
1027         free(config_file);
1028         if (ret < 0)
1029                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1030         return ret;
1031 }
1032
1033 static int change_to_dest_dir(void)
1034 {
1035         DSS_INFO_LOG(("changing cwd to %s\n", conf.dest_dir_arg));
1036         return dss_chdir(conf.dest_dir_arg);
1037 }
1038
1039 static int handle_sighup(void)
1040 {
1041         int ret;
1042
1043         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1044         dump_dss_config("old");
1045         ret = parse_config_file(1);
1046         if (ret < 0)
1047                 return ret;
1048         dump_dss_config("reloaded");
1049         invalidate_next_snapshot_time();
1050         return change_to_dest_dir();
1051 }
1052
1053 static void kill_children(void)
1054 {
1055         restart_create_process();
1056         dss_kill(create_pid, SIGTERM, NULL);
1057         dss_kill(remove_pid, SIGTERM, NULL);
1058 }
1059
1060 static int handle_signal(void)
1061 {
1062         int sig, ret = next_signal();
1063
1064         if (ret <= 0)
1065                 goto out;
1066         sig = ret;
1067         switch (sig) {
1068         case SIGINT:
1069         case SIGTERM:
1070                 kill_children();
1071                 ret = -E_SIGNAL;
1072                 break;
1073         case SIGHUP:
1074                 ret = handle_sighup();
1075                 break;
1076         case SIGCHLD:
1077                 ret = handle_sigchld();
1078                 break;
1079         }
1080 out:
1081         if (ret < 0)
1082                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1083         return ret;
1084 }
1085
1086 /*
1087  * We can not use rsync locally if the local user is different from the remote
1088  * user or if the src dir is not on the local host (or both).
1089  */
1090 static int use_rsync_locally(char *logname)
1091 {
1092         char *h = conf.remote_host_arg;
1093
1094         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1095                 return 0;
1096         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1097                 return 0;
1098         return 1;
1099 }
1100
1101 static int rename_resume_snap(int64_t creation_time)
1102 {
1103         struct snapshot_list sl;
1104         struct snapshot *s = NULL;
1105         char *new_name = incomplete_name(creation_time);
1106         int ret;
1107         const char *why;
1108
1109         sl.num_snapshots = 0;
1110
1111         ret = 0;
1112         if (conf.no_resume_given)
1113                 goto out;
1114         dss_get_snapshot_list(&sl);
1115         /*
1116          * Snapshot recycling: We first look at the newest snapshot. If this
1117          * snapshot happens to be incomplete, the last rsync process was
1118          * aborted and we reuse this one. Otherwise we look at snapshots which
1119          * could be removed (outdated and redundant snapshots) as candidates
1120          * for recycling. If no outdated/redundant snapshot exists, we check if
1121          * there is an orphaned snapshot, which likely is useless anyway.
1122          *
1123          * Only if no existing snapshot is suitable for recycling, we bite the
1124          * bullet and create a new one.
1125          */
1126         s = get_newest_snapshot(&sl);
1127         if (!s) /* no snapshots at all */
1128                 goto out;
1129         /* re-use last snapshot if it is incomplete */
1130         why = "aborted";
1131         if ((s->flags & SS_COMPLETE) == 0)
1132                 goto out;
1133         why = "outdated";
1134         s = find_outdated_snapshot(&sl);
1135         if (s)
1136                 goto out;
1137         why = "redundant";
1138         s = find_redundant_snapshot(&sl);
1139         if (s)
1140                 goto out;
1141         why = "orphaned";
1142         s = find_orphaned_snapshot(&sl);
1143 out:
1144         if (s) {
1145                 DSS_INFO_LOG(("reusing %s snapshot %s\n", why, s->name));
1146                 ret = dss_rename(s->name, new_name);
1147         }
1148         if (ret >= 0)
1149                 DSS_NOTICE_LOG(("creating new snapshot %s\n", new_name));
1150         free(new_name);
1151         free_snapshot_list(&sl);
1152         return ret;
1153 }
1154
1155 static void create_rsync_argv(char ***argv, int64_t *num)
1156 {
1157         char *logname;
1158         int i = 0, j;
1159         struct snapshot_list sl;
1160
1161         dss_get_snapshot_list(&sl);
1162         assert(!name_of_reference_snapshot);
1163         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1164         free_snapshot_list(&sl);
1165
1166         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1167         (*argv)[i++] = dss_strdup("rsync");
1168         (*argv)[i++] = dss_strdup("-a");
1169         (*argv)[i++] = dss_strdup("--delete");
1170         for (j = 0; j < conf.rsync_option_given; j++)
1171                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1172         if (name_of_reference_snapshot) {
1173                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1174                 (*argv)[i++] = make_message("--link-dest=../%s",
1175                         name_of_reference_snapshot);
1176         } else
1177                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1178         logname = dss_logname();
1179         if (use_rsync_locally(logname))
1180                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1181         else
1182                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1183                         conf.remote_user_arg : logname,
1184                         conf.remote_host_arg, conf.source_dir_arg);
1185         free(logname);
1186         *num = get_current_time();
1187         (*argv)[i++] = incomplete_name(*num);
1188         (*argv)[i++] = NULL;
1189         for (j = 0; j < i; j++)
1190                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1191 }
1192
1193 static void free_rsync_argv(char **argv)
1194 {
1195         int i;
1196
1197         if (!argv)
1198                 return;
1199         for (i = 0; argv[i]; i++)
1200                 free(argv[i]);
1201         free(argv);
1202 }
1203
1204 static int create_snapshot(char **argv)
1205 {
1206         int ret;
1207
1208         ret = rename_resume_snap(current_snapshot_creation_time);
1209         if (ret < 0)
1210                 return ret;
1211         dss_exec(&create_pid, argv[0], argv);
1212         snapshot_creation_status = HS_RUNNING;
1213         return ret;
1214 }
1215
1216 static int select_loop(void)
1217 {
1218         int ret;
1219         /* check every 60 seconds for free disk space */
1220         struct timeval tv;
1221         char **rsync_argv = NULL;
1222
1223         for (;;) {
1224                 fd_set rfds;
1225                 struct timeval *tvp;
1226
1227                 if (remove_pid)
1228                         tvp = NULL; /* sleep until rm hook/process dies */
1229                 else { /* sleep one minute */
1230                         tv.tv_sec = 60;
1231                         tv.tv_usec = 0;
1232                         tvp = &tv;
1233                 }
1234                 FD_ZERO(&rfds);
1235                 FD_SET(signal_pipe, &rfds);
1236                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1237                 if (ret < 0)
1238                         goto out;
1239                 if (FD_ISSET(signal_pipe, &rfds)) {
1240                         ret = handle_signal();
1241                         if (ret < 0)
1242                                 goto out;
1243                 }
1244                 if (remove_pid)
1245                         continue;
1246                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1247                         ret = exec_rm();
1248                         if (ret < 0)
1249                                 goto out;
1250                         continue;
1251                 }
1252                 if (snapshot_removal_status == HS_SUCCESS) {
1253                         post_remove_hook();
1254                         continue;
1255                 }
1256                 ret = try_to_free_disk_space();
1257                 if (ret < 0)
1258                         goto out;
1259                 if (snapshot_removal_status != HS_READY) {
1260                         stop_create_process();
1261                         continue;
1262                 }
1263                 restart_create_process();
1264                 switch (snapshot_creation_status) {
1265                 case HS_READY:
1266                         if (!next_snapshot_is_due())
1267                                 continue;
1268                         pre_create_hook();
1269                         continue;
1270                 case HS_PRE_RUNNING:
1271                 case HS_RUNNING:
1272                 case HS_POST_RUNNING:
1273                         continue;
1274                 case HS_PRE_SUCCESS:
1275                         if (!name_of_reference_snapshot) {
1276                                 free_rsync_argv(rsync_argv);
1277                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1278                         }
1279                         ret = create_snapshot(rsync_argv);
1280                         if (ret < 0)
1281                                 goto out;
1282                         continue;
1283                 case HS_NEEDS_RESTART:
1284                         if (!next_snapshot_is_due())
1285                                 continue;
1286                         ret = create_snapshot(rsync_argv);
1287                         if (ret < 0)
1288                                 goto out;
1289                         continue;
1290                 case HS_SUCCESS:
1291                         post_create_hook();
1292                         continue;
1293                 }
1294         }
1295 out:
1296         return ret;
1297 }
1298
1299 static void exit_hook(int exit_code)
1300 {
1301         char *argv[3];
1302         pid_t pid;
1303
1304         argv[0] = conf.exit_hook_arg;
1305         argv[1] = dss_strerror(-exit_code);
1306         argv[2] = NULL;
1307
1308         DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
1309         dss_exec(&pid, conf.exit_hook_arg, argv);
1310 }
1311
1312 static void lock_dss_or_die(void)
1313 {
1314         char *config_file = get_config_file_name();
1315         int ret = lock_dss(config_file);
1316
1317         free(config_file);
1318         if (ret < 0) {
1319                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1320                 exit(EXIT_FAILURE);
1321         }
1322 }
1323
1324 static int com_run(void)
1325 {
1326         int ret;
1327
1328         lock_dss_or_die();
1329         if (conf.dry_run_given) {
1330                 DSS_ERROR_LOG(("dry_run not supported by this command\n"));
1331                 return -E_SYNTAX;
1332         }
1333         ret = install_sighandler(SIGHUP);
1334         if (ret < 0)
1335                 return ret;
1336         ret = select_loop();
1337         if (ret >= 0) /* impossible */
1338                 ret = -E_BUG;
1339         kill_children();
1340         exit_hook(ret);
1341         return ret;
1342 }
1343
1344 static int com_prune(void)
1345 {
1346         int ret;
1347         struct snapshot_list sl;
1348         struct snapshot *victim;
1349         struct disk_space ds;
1350         const char *why;
1351
1352         lock_dss_or_die();
1353         ret = get_disk_space(".", &ds);
1354         if (ret < 0)
1355                 return ret;
1356         log_disk_space(&ds);
1357         dss_get_snapshot_list(&sl);
1358         why = "outdated";
1359         victim = find_outdated_snapshot(&sl);
1360         if (victim)
1361                 goto rm;
1362         why = "redundant";
1363         victim = find_redundant_snapshot(&sl);
1364         if (victim)
1365                 goto rm;
1366         ret = 0;
1367         goto out;
1368 rm:
1369         if (conf.dry_run_given) {
1370                 dss_msg("%s snapshot %s (interval = %i)\n",
1371                         why, victim->name, victim->interval);
1372                 ret = 0;
1373                 goto out;
1374         }
1375         pre_remove_hook(victim, why);
1376         if (snapshot_removal_status == HS_PRE_RUNNING) {
1377                 ret = wait_for_remove_process();
1378                 if (ret < 0)
1379                         goto out;
1380                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1381                         goto out;
1382         }
1383         ret = exec_rm();
1384         if (ret < 0)
1385                 goto out;
1386         ret = wait_for_remove_process();
1387         if (ret < 0)
1388                 goto out;
1389         if (snapshot_removal_status != HS_SUCCESS)
1390                 goto out;
1391         post_remove_hook();
1392         if (snapshot_removal_status != HS_POST_RUNNING)
1393                 goto out;
1394         ret = wait_for_remove_process();
1395         if (ret < 0)
1396                 goto out;
1397         ret = 1;
1398 out:
1399         free_snapshot_list(&sl);
1400         return ret;
1401 }
1402
1403 static int com_create(void)
1404 {
1405         int ret, status;
1406         char **rsync_argv;
1407
1408         lock_dss_or_die();
1409         if (conf.dry_run_given) {
1410                 int i;
1411                 char *msg = NULL;
1412                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1413                 for (i = 0; rsync_argv[i]; i++) {
1414                         char *tmp = msg;
1415                         msg = make_message("%s%s%s", tmp? tmp : "",
1416                                 tmp? " " : "", rsync_argv[i]);
1417                         free(tmp);
1418                 }
1419                 free_rsync_argv(rsync_argv);
1420                 dss_msg("%s\n", msg);
1421                 free(msg);
1422                 return 1;
1423         }
1424         pre_create_hook();
1425         if (create_pid) {
1426                 ret = wait_for_process(create_pid, &status);
1427                 if (ret < 0)
1428                         return ret;
1429                 ret = handle_pre_create_hook_exit(status);
1430                 if (ret <= 0) /* error, or pre-create failed */
1431                         return ret;
1432         }
1433         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1434         ret = create_snapshot(rsync_argv);
1435         if (ret < 0)
1436                 goto out;
1437         ret = wait_for_process(create_pid, &status);
1438         if (ret < 0)
1439                 goto out;
1440         ret = handle_rsync_exit(status);
1441         if (ret < 0)
1442                 goto out;
1443         post_create_hook();
1444         if (create_pid)
1445                 ret = wait_for_process(create_pid, &status);
1446 out:
1447         free_rsync_argv(rsync_argv);
1448         return ret;
1449 }
1450
1451 static int com_ls(void)
1452 {
1453         int i;
1454         struct snapshot_list sl;
1455         struct snapshot *s;
1456
1457         dss_get_snapshot_list(&sl);
1458         FOR_EACH_SNAPSHOT(s, i, &sl) {
1459                 int64_t d = 0;
1460                 if (s->flags & SS_COMPLETE)
1461                         d = (s->completion_time - s->creation_time) / 60;
1462                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1463         }
1464         free_snapshot_list(&sl);
1465         return 1;
1466 }
1467
1468 static int setup_signal_handling(void)
1469 {
1470         int ret;
1471
1472         DSS_INFO_LOG(("setting up signal handlers\n"));
1473         signal_pipe = signal_init(); /* always successful */
1474         ret = install_sighandler(SIGINT);
1475         if (ret < 0)
1476                 return ret;
1477         ret = install_sighandler(SIGTERM);
1478         if (ret < 0)
1479                 return ret;
1480         return install_sighandler(SIGCHLD);
1481 }
1482
1483 /**
1484  * The main function of dss.
1485  *
1486  * \param argc Usual argument count.
1487  * \param argv Usual argument vector.
1488  */
1489 int main(int argc, char **argv)
1490 {
1491         int ret;
1492         struct cmdline_parser_params params;
1493
1494         params.override = 0;
1495         params.initialize = 1;
1496         params.check_required = 0;
1497         params.check_ambiguity = 0;
1498         params.print_errors = 1;
1499
1500         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1501         ret = parse_config_file(0);
1502         if (ret < 0)
1503                 goto out;
1504         if (ret == 0) { /* no config file given */
1505                 /*
1506                  * Parse the command line options again, but this time check
1507                  * that all required options are given.
1508                  */
1509                 struct cmdline_parser_params params;
1510                 params.override = 1;
1511                 params.initialize = 1;
1512                 params.check_required = 1;
1513                 params.check_ambiguity = 1;
1514                 params.print_errors = 1;
1515                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1516         }
1517         if (conf.daemon_given)
1518                 daemon_init();
1519         ret = change_to_dest_dir();
1520         if (ret < 0)
1521                 goto out;
1522         dump_dss_config("startup");
1523         ret = setup_signal_handling();
1524         if (ret < 0)
1525                 goto out;
1526         ret = call_command_handler();
1527         signal_shutdown();
1528 out:
1529         if (ret < 0)
1530                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1531         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1532 }