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         /*
 610          * Idle and --keep_redundant not given, or low disk space. Look at
 611          * existing snapshots.
 612          */
 613         dss_get_snapshot_list(&sl);
 614         ret = 0;
 615         if (!low_disk_space && sl.num_snapshots <= 1)
 616                 goto out;
 617         why = "outdated";
 618         victim = find_outdated_snapshot(&sl);
 619         if (victim)
 620                 goto remove;
 621         why = "redundant";
 622         victim = find_redundant_snapshot(&sl);
 623         if (victim)
 624                 goto remove;
 625         /* try harder only if disk space is low */
 626         if (!low_disk_space)
 627                 goto out;
 628         why = "orphaned";
 629         victim = find_orphaned_snapshot(&sl);
 630         if (victim)
 631                 goto remove;
 632         DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 633         victim = find_oldest_removable_snapshot(&sl);
 634         if (victim)
 635                 goto remove;
 636         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 637         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 638         goto out;
 639 remove:
 640         pre_remove_hook(victim, why);
 641 out:
 642         free_snapshot_list(&sl);
 643         return ret;
 644 }
 645
 646 static void post_create_hook(void)
 647 {
 648         char *cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 649                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 650         DSS_NOTICE_LOG(("executing %s\n", cmd));
 651         dss_exec_cmdline_pid(&create_pid, cmd);
 652         free(cmd);
 653         snapshot_creation_status = HS_POST_RUNNING;
 654 }
 655
 656 static void post_remove_hook(void)
 657 {
 658         char *cmd;
 659         struct snapshot *s = snapshot_currently_being_removed;
 660
 661         assert(s);
 662
 663         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 664                 conf.dest_dir_arg, s->name);
 665         DSS_NOTICE_LOG(("executing %s\n", cmd));
 666         dss_exec_cmdline_pid(&remove_pid, cmd);
 667         free(cmd);
 668         snapshot_removal_status = HS_POST_RUNNING;
 669 }
 670
 671 static void dss_kill(pid_t pid, int sig, const char *msg)
 672 {
 673         const char *signame, *process_name;
 674
 675         if (pid == 0)
 676                 return;
 677         switch (sig) {
 678         case SIGTERM: signame = "TERM"; break;
 679         case SIGSTOP: signame = "STOP"; break;
 680         case SIGCONT: signame = "CONT"; break;
 681         default: signame = "????";
 682         }
 683
 684         if (pid == create_pid)
 685                 process_name = "create";
 686         else if (pid == remove_pid)
 687                 process_name = "remove";
 688         else process_name = "??????";
 689
 690         if (msg)
 691                 DSS_INFO_LOG(("%s\n", msg));
 692         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 693                 sig, signame, (int)pid, process_name));
 694         if (kill(pid, sig) >= 0)
 695                 return;
 696         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 697                 sig, signame, (int)pid, process_name));
 698 }
 699
 700 static void stop_create_process(void)
 701 {
 702         if (create_process_stopped)
 703                 return;
 704         dss_kill(create_pid, SIGSTOP, "suspending create process");
 705         create_process_stopped = 1;
 706 }
 707
 708 static void restart_create_process(void)
 709 {
 710         if (!create_process_stopped)
 711                 return;
 712         dss_kill(create_pid, SIGCONT, "resuming create process");
 713         create_process_stopped = 0;
 714 }
 715
 716 /**
 717  * Print a log message about the exit status of a child.
 718  */
 719 static void log_termination_msg(pid_t pid, int status)
 720 {
 721         if (WIFEXITED(status))
 722                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 723                         WEXITSTATUS(status)));
 724         else if (WIFSIGNALED(status))
 725                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 726                         WTERMSIG(status)));
 727         else
 728                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 729 }
 730
 731 static int wait_for_process(pid_t pid, int *status)
 732 {
 733         int ret;
 734
 735         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 736         for (;;) {
 737                 fd_set rfds;
 738
 739                 FD_ZERO(&rfds);
 740                 FD_SET(signal_pipe, &rfds);
 741                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 742                 if (ret < 0)
 743                         break;
 744                 ret = next_signal();
 745                 if (!ret)
 746                         continue;
 747                 if (ret == SIGCHLD) {
 748                         ret = waitpid(pid, status, 0);
 749                         if (ret >= 0)
 750                                 break;
 751                         if (errno != EINTR) { /* error */
 752                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 753                                 break;
 754                         }
 755                 }
 756                 /* SIGINT or SIGTERM */
 757                 dss_kill(pid, SIGTERM, "killing child process");
 758         }
 759         if (ret < 0)
 760                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 761         else
 762                 log_termination_msg(pid, *status);
 763         return ret;
 764 }
 765
 766 static void handle_pre_remove_exit(int status)
 767 {
 768         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 769                 snapshot_removal_status = HS_READY;
 770                 gettimeofday(&next_removal_check, NULL);
 771                 next_removal_check.tv_sec += 60;
 772                 return;
 773         }
 774         snapshot_removal_status = HS_PRE_SUCCESS;
 775 }
 776
 777 static int handle_rm_exit(int status)
 778 {
 779         if (!WIFEXITED(status)) {
 780                 snapshot_removal_status = HS_READY;
 781                 return -E_INVOLUNTARY_EXIT;
 782         }
 783         if (WEXITSTATUS(status)) {
 784                 snapshot_removal_status = HS_READY;
 785                 return -E_BAD_EXIT_CODE;
 786         }
 787         snapshot_removal_status = HS_SUCCESS;
 788         return 1;
 789 }
 790
 791 static void handle_post_remove_exit(void)
 792 {
 793         snapshot_removal_status = HS_READY;
 794 }
 795
 796 static int handle_remove_exit(int status)
 797 {
 798         int ret;
 799         struct snapshot *s = snapshot_currently_being_removed;
 800
 801         assert(s);
 802         switch (snapshot_removal_status) {
 803         case HS_PRE_RUNNING:
 804                 handle_pre_remove_exit(status);
 805                 ret = 1;
 806                 break;
 807         case HS_RUNNING:
 808                 ret = handle_rm_exit(status);
 809                 break;
 810         case HS_POST_RUNNING:
 811                 handle_post_remove_exit();
 812                 ret = 1;
 813                 break;
 814         default:
 815                 ret = -E_BUG;
 816         }
 817         if (snapshot_removal_status == HS_READY) {
 818                 free(s->name);
 819                 free(s);
 820                 snapshot_currently_being_removed = NULL;
 821         }
 822         remove_pid = 0;
 823         return ret;
 824 }
 825
 826 static int wait_for_remove_process(void)
 827 {
 828         int status, ret;
 829
 830         assert(remove_pid);
 831         assert(
 832                 snapshot_removal_status == HS_PRE_RUNNING ||
 833                 snapshot_removal_status == HS_RUNNING ||
 834                 snapshot_removal_status == HS_POST_RUNNING
 835         );
 836         ret = wait_for_process(remove_pid, &status);
 837         if (ret < 0)
 838                 return ret;
 839         return handle_remove_exit(status);
 840 }
 841
 842 static int handle_rsync_exit(int status)
 843 {
 844         int es, ret;
 845
 846         if (!WIFEXITED(status)) {
 847                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 848                 ret = -E_INVOLUNTARY_EXIT;
 849                 snapshot_creation_status = HS_READY;
 850                 goto out;
 851         }
 852         es = WEXITSTATUS(status);
 853         /*
 854          * Restart rsync on non-fatal errors:
 855          * 24: Partial transfer due to vanished source files
 856          */
 857         if (es != 0 && es != 24) {
 858                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 859                         es, ++num_consecutive_rsync_errors));
 860                 if (num_consecutive_rsync_errors > conf.max_rsync_errors_arg) {
 861                         ret = -E_TOO_MANY_RSYNC_ERRORS;
 862                         snapshot_creation_status = HS_READY;
 863                         goto out;
 864                 }
 865                 DSS_WARNING_LOG(("restarting rsync process\n"));
 866                 snapshot_creation_status = HS_NEEDS_RESTART;
 867                 next_snapshot_time = get_current_time() + 60;
 868                 ret = 1;
 869                 goto out;
 870         }
 871         num_consecutive_rsync_errors = 0;
 872         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 873         if (ret < 0)
 874                 goto out;
 875         snapshot_creation_status = HS_SUCCESS;
 876         free(name_of_reference_snapshot);
 877         name_of_reference_snapshot = NULL;
 878 out:
 879         create_process_stopped = 0;
 880         return ret;
 881 }
 882
 883 static int handle_pre_create_hook_exit(int status)
 884 {
 885         int es, ret;
 886         static int warn_count;
 887
 888         if (!WIFEXITED(status)) {
 889                 snapshot_creation_status = HS_READY;
 890                 ret = -E_INVOLUNTARY_EXIT;
 891                 goto out;
 892         }
 893         es = WEXITSTATUS(status);
 894         if (es) {
 895                 if (!warn_count--) {
 896                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
 897                                 conf.pre_create_hook_arg, es));
 898                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
 899                         warn_count = 60; /* warn only once per hour */
 900                 }
 901                 next_snapshot_time = get_current_time() + 60;
 902                 snapshot_creation_status = HS_READY;
 903                 ret = 0;
 904                 goto out;
 905         }
 906         warn_count = 0;
 907         snapshot_creation_status = HS_PRE_SUCCESS;
 908         ret = 1;
 909 out:
 910         return ret;
 911 }
 912
 913 static int handle_sigchld(void)
 914 {
 915         pid_t pid;
 916         int status, ret = reap_child(&pid, &status);
 917
 918         if (ret <= 0)
 919                 return ret;
 920
 921         if (pid == create_pid) {
 922                 switch (snapshot_creation_status) {
 923                 case HS_PRE_RUNNING:
 924                         ret = handle_pre_create_hook_exit(status);
 925                         break;
 926                 case HS_RUNNING:
 927                         ret = handle_rsync_exit(status);
 928                         break;
 929                 case HS_POST_RUNNING:
 930                         snapshot_creation_status = HS_READY;
 931                         ret = 1;
 932                         break;
 933                 default:
 934                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
 935                                 snapshot_creation_status));
 936                         return -E_BUG;
 937                 }
 938                 create_pid = 0;
 939                 return ret;
 940         }
 941         if (pid == remove_pid) {
 942                 ret = handle_remove_exit(status);
 943                 if (ret < 0)
 944                         return ret;
 945                 return ret;
 946         }
 947         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
 948         return -E_BUG;
 949 }
 950
 951 static int check_config(void)
 952 {
 953         if (conf.unit_interval_arg <= 0) {
 954                 DSS_ERROR_LOG(("bad unit interval: %i\n", conf.unit_interval_arg));
 955                 return -E_INVALID_NUMBER;
 956         }
 957         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", conf.unit_interval_arg));
 958         if (conf.num_intervals_arg <= 0 || conf.num_intervals_arg > 30) {
 959                 DSS_ERROR_LOG(("bad number of intervals: %i\n",
 960                         conf.num_intervals_arg));
 961                 return -E_INVALID_NUMBER;
 962         }
 963         DSS_DEBUG_LOG(("number of intervals: %i\n", conf.num_intervals_arg));
 964         return 1;
 965 }
 966
 967 /*
 968  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 969  * file was read successfully.
 970  */
 971 static int parse_config_file(int override)
 972 {
 973         int ret, config_file_exists;
 974         char *config_file = get_config_file_name();
 975         struct stat statbuf;
 976         char *old_logfile_arg = NULL;
 977         int old_daemon_given = 0;
 978
 979         if (override) { /* SIGHUP */
 980                 if (conf.logfile_given)
 981                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 982                 old_daemon_given = conf.daemon_given;
 983         }
 984
 985         config_file_exists = !stat(config_file, &statbuf);
 986         if (!config_file_exists && conf.config_file_given) {
 987                 ret = -ERRNO_TO_DSS_ERROR(errno);
 988                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
 989                 goto out;
 990         }
 991         if (config_file_exists) {
 992                 struct cmdline_parser_params params;
 993                 params.override = override;
 994                 params.initialize = 0;
 995                 params.check_required = 1;
 996                 params.check_ambiguity = 0;
 997                 params.print_errors = 1;
 998                 if (override) { /* invalidate all rsync options */
 999                         int i;
1000
1001                         for (i = 0; i < conf.rsync_option_given; i++) {
1002                                 free(conf.rsync_option_arg[i]);
1003                                 conf.rsync_option_arg[i] = NULL;
1004                         }
1005                         conf.rsync_option_given = 0;
1006                 }
1007                 cmdline_parser_config_file(config_file, &conf, &params);
1008         }
1009         ret = check_config();
1010         if (ret < 0)
1011                 goto out;
1012         if (override) {
1013                 /* don't change daemon mode on SIGHUP */
1014                 conf.daemon_given = old_daemon_given;
1015                 close_log(logfile);
1016                 logfile = NULL;
1017                 if (conf.logfile_given)
1018                         free(old_logfile_arg);
1019                 else if (conf.daemon_given) { /* re-use old logfile */
1020                         conf.logfile_arg = old_logfile_arg;
1021                         conf.logfile_given = 1;
1022                 }
1023         }
1024         if (conf.logfile_given && conf.run_given && conf.daemon_given) {
1025                 logfile = open_log(conf.logfile_arg);
1026                 log_welcome(conf.loglevel_arg);
1027         }
1028         DSS_DEBUG_LOG(("loglevel: %d\n", conf.loglevel_arg));
1029         ret = config_file_exists;
1030 out:
1031         free(config_file);
1032         if (ret < 0)
1033                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1034         return ret;
1035 }
1036
1037 static int change_to_dest_dir(void)
1038 {
1039         DSS_INFO_LOG(("changing cwd to %s\n", conf.dest_dir_arg));
1040         return dss_chdir(conf.dest_dir_arg);
1041 }
1042
1043 static int handle_sighup(void)
1044 {
1045         int ret;
1046
1047         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1048         dump_dss_config("old");
1049         ret = parse_config_file(1);
1050         if (ret < 0)
1051                 return ret;
1052         dump_dss_config("reloaded");
1053         invalidate_next_snapshot_time();
1054         return change_to_dest_dir();
1055 }
1056
1057 static void kill_children(void)
1058 {
1059         restart_create_process();
1060         dss_kill(create_pid, SIGTERM, NULL);
1061         dss_kill(remove_pid, SIGTERM, NULL);
1062 }
1063
1064 static int handle_signal(void)
1065 {
1066         int sig, ret = next_signal();
1067
1068         if (ret <= 0)
1069                 goto out;
1070         sig = ret;
1071         switch (sig) {
1072         case SIGINT:
1073         case SIGTERM:
1074                 kill_children();
1075                 ret = -E_SIGNAL;
1076                 break;
1077         case SIGHUP:
1078                 ret = handle_sighup();
1079                 break;
1080         case SIGCHLD:
1081                 ret = handle_sigchld();
1082                 break;
1083         }
1084 out:
1085         if (ret < 0)
1086                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1087         return ret;
1088 }
1089
1090 /*
1091  * We can not use rsync locally if the local user is different from the remote
1092  * user or if the src dir is not on the local host (or both).
1093  */
1094 static int use_rsync_locally(char *logname)
1095 {
1096         char *h = conf.remote_host_arg;
1097
1098         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1099                 return 0;
1100         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1101                 return 0;
1102         return 1;
1103 }
1104
1105 static int rename_resume_snap(int64_t creation_time)
1106 {
1107         struct snapshot_list sl;
1108         struct snapshot *s = NULL;
1109         char *new_name = incomplete_name(creation_time);
1110         int ret;
1111         const char *why;
1112
1113         sl.num_snapshots = 0;
1114
1115         ret = 0;
1116         if (conf.no_resume_given)
1117                 goto out;
1118         dss_get_snapshot_list(&sl);
1119         /*
1120          * Snapshot recycling: We first look at the newest snapshot. If this
1121          * snapshot happens to be incomplete, the last rsync process was
1122          * aborted and we reuse this one. Otherwise we look at snapshots which
1123          * could be removed (outdated and redundant snapshots) as candidates
1124          * for recycling. If no outdated/redundant snapshot exists, we check if
1125          * there is an orphaned snapshot, which likely is useless anyway.
1126          *
1127          * Only if no existing snapshot is suitable for recycling, we bite the
1128          * bullet and create a new one.
1129          */
1130         s = get_newest_snapshot(&sl);
1131         if (!s) /* no snapshots at all */
1132                 goto out;
1133         /* re-use last snapshot if it is incomplete */
1134         why = "aborted";
1135         if ((s->flags & SS_COMPLETE) == 0)
1136                 goto out;
1137         why = "outdated";
1138         s = find_outdated_snapshot(&sl);
1139         if (s)
1140                 goto out;
1141         why = "redundant";
1142         s = find_redundant_snapshot(&sl);
1143         if (s)
1144                 goto out;
1145         why = "orphaned";
1146         s = find_orphaned_snapshot(&sl);
1147 out:
1148         if (s) {
1149                 DSS_INFO_LOG(("reusing %s snapshot %s\n", why, s->name));
1150                 ret = dss_rename(s->name, new_name);
1151         }
1152         if (ret >= 0)
1153                 DSS_NOTICE_LOG(("creating new snapshot %s\n", new_name));
1154         free(new_name);
1155         free_snapshot_list(&sl);
1156         return ret;
1157 }
1158
1159 static void create_rsync_argv(char ***argv, int64_t *num)
1160 {
1161         char *logname;
1162         int i = 0, j;
1163         struct snapshot_list sl;
1164
1165         dss_get_snapshot_list(&sl);
1166         assert(!name_of_reference_snapshot);
1167         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1168         free_snapshot_list(&sl);
1169
1170         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1171         (*argv)[i++] = dss_strdup("rsync");
1172         (*argv)[i++] = dss_strdup("-a");
1173         (*argv)[i++] = dss_strdup("--delete");
1174         for (j = 0; j < conf.rsync_option_given; j++)
1175                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1176         if (name_of_reference_snapshot) {
1177                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1178                 (*argv)[i++] = make_message("--link-dest=../%s",
1179                         name_of_reference_snapshot);
1180         } else
1181                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1182         logname = dss_logname();
1183         if (use_rsync_locally(logname))
1184                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1185         else
1186                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1187                         conf.remote_user_arg : logname,
1188                         conf.remote_host_arg, conf.source_dir_arg);
1189         free(logname);
1190         *num = get_current_time();
1191         (*argv)[i++] = incomplete_name(*num);
1192         (*argv)[i++] = NULL;
1193         for (j = 0; j < i; j++)
1194                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1195 }
1196
1197 static void free_rsync_argv(char **argv)
1198 {
1199         int i;
1200
1201         if (!argv)
1202                 return;
1203         for (i = 0; argv[i]; i++)
1204                 free(argv[i]);
1205         free(argv);
1206 }
1207
1208 static int create_snapshot(char **argv)
1209 {
1210         int ret;
1211
1212         ret = rename_resume_snap(current_snapshot_creation_time);
1213         if (ret < 0)
1214                 return ret;
1215         dss_exec(&create_pid, argv[0], argv);
1216         snapshot_creation_status = HS_RUNNING;
1217         return ret;
1218 }
1219
1220 static int select_loop(void)
1221 {
1222         int ret;
1223         /* check every 60 seconds for free disk space */
1224         struct timeval tv;
1225         char **rsync_argv = NULL;
1226
1227         for (;;) {
1228                 fd_set rfds;
1229                 struct timeval *tvp;
1230
1231                 if (remove_pid)
1232                         tvp = NULL; /* sleep until rm hook/process dies */
1233                 else { /* sleep one minute */
1234                         tv.tv_sec = 60;
1235                         tv.tv_usec = 0;
1236                         tvp = &tv;
1237                 }
1238                 FD_ZERO(&rfds);
1239                 FD_SET(signal_pipe, &rfds);
1240                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1241                 if (ret < 0)
1242                         goto out;
1243                 if (FD_ISSET(signal_pipe, &rfds)) {
1244                         ret = handle_signal();
1245                         if (ret < 0)
1246                                 goto out;
1247                 }
1248                 if (remove_pid)
1249                         continue;
1250                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1251                         ret = exec_rm();
1252                         if (ret < 0)
1253                                 goto out;
1254                         continue;
1255                 }
1256                 if (snapshot_removal_status == HS_SUCCESS) {
1257                         post_remove_hook();
1258                         continue;
1259                 }
1260                 ret = try_to_free_disk_space();
1261                 if (ret < 0)
1262                         goto out;
1263                 if (snapshot_removal_status != HS_READY) {
1264                         stop_create_process();
1265                         continue;
1266                 }
1267                 restart_create_process();
1268                 switch (snapshot_creation_status) {
1269                 case HS_READY:
1270                         if (!next_snapshot_is_due())
1271                                 continue;
1272                         pre_create_hook();
1273                         continue;
1274                 case HS_PRE_RUNNING:
1275                 case HS_RUNNING:
1276                 case HS_POST_RUNNING:
1277                         continue;
1278                 case HS_PRE_SUCCESS:
1279                         if (!name_of_reference_snapshot) {
1280                                 free_rsync_argv(rsync_argv);
1281                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1282                         }
1283                         ret = create_snapshot(rsync_argv);
1284                         if (ret < 0)
1285                                 goto out;
1286                         continue;
1287                 case HS_NEEDS_RESTART:
1288                         if (!next_snapshot_is_due())
1289                                 continue;
1290                         ret = create_snapshot(rsync_argv);
1291                         if (ret < 0)
1292                                 goto out;
1293                         continue;
1294                 case HS_SUCCESS:
1295                         post_create_hook();
1296                         continue;
1297                 }
1298         }
1299 out:
1300         return ret;
1301 }
1302
1303 static void exit_hook(int exit_code)
1304 {
1305         char *argv[3];
1306         pid_t pid;
1307
1308         argv[0] = conf.exit_hook_arg;
1309         argv[1] = dss_strerror(-exit_code);
1310         argv[2] = NULL;
1311
1312         DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
1313         dss_exec(&pid, conf.exit_hook_arg, argv);
1314 }
1315
1316 static void lock_dss_or_die(void)
1317 {
1318         char *config_file = get_config_file_name();
1319         int ret = lock_dss(config_file);
1320
1321         free(config_file);
1322         if (ret < 0) {
1323                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1324                 exit(EXIT_FAILURE);
1325         }
1326 }
1327
1328 static int com_run(void)
1329 {
1330         int ret;
1331
1332         lock_dss_or_die();
1333         if (conf.dry_run_given) {
1334                 DSS_ERROR_LOG(("dry_run not supported by this command\n"));
1335                 return -E_SYNTAX;
1336         }
1337         ret = install_sighandler(SIGHUP);
1338         if (ret < 0)
1339                 return ret;
1340         ret = select_loop();
1341         if (ret >= 0) /* impossible */
1342                 ret = -E_BUG;
1343         kill_children();
1344         exit_hook(ret);
1345         return ret;
1346 }
1347
1348 static int com_prune(void)
1349 {
1350         int ret;
1351         struct snapshot_list sl;
1352         struct snapshot *victim;
1353         struct disk_space ds;
1354         const char *why;
1355
1356         lock_dss_or_die();
1357         ret = get_disk_space(".", &ds);
1358         if (ret < 0)
1359                 return ret;
1360         log_disk_space(&ds);
1361         dss_get_snapshot_list(&sl);
1362         why = "outdated";
1363         victim = find_outdated_snapshot(&sl);
1364         if (victim)
1365                 goto rm;
1366         why = "redundant";
1367         victim = find_redundant_snapshot(&sl);
1368         if (victim)
1369                 goto rm;
1370         ret = 0;
1371         goto out;
1372 rm:
1373         if (conf.dry_run_given) {
1374                 dss_msg("%s snapshot %s (interval = %i)\n",
1375                         why, victim->name, victim->interval);
1376                 ret = 0;
1377                 goto out;
1378         }
1379         pre_remove_hook(victim, why);
1380         if (snapshot_removal_status == HS_PRE_RUNNING) {
1381                 ret = wait_for_remove_process();
1382                 if (ret < 0)
1383                         goto out;
1384                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1385                         goto out;
1386         }
1387         ret = exec_rm();
1388         if (ret < 0)
1389                 goto out;
1390         ret = wait_for_remove_process();
1391         if (ret < 0)
1392                 goto out;
1393         if (snapshot_removal_status != HS_SUCCESS)
1394                 goto out;
1395         post_remove_hook();
1396         if (snapshot_removal_status != HS_POST_RUNNING)
1397                 goto out;
1398         ret = wait_for_remove_process();
1399         if (ret < 0)
1400                 goto out;
1401         ret = 1;
1402 out:
1403         free_snapshot_list(&sl);
1404         return ret;
1405 }
1406
1407 static int com_create(void)
1408 {
1409         int ret, status;
1410         char **rsync_argv;
1411
1412         lock_dss_or_die();
1413         if (conf.dry_run_given) {
1414                 int i;
1415                 char *msg = NULL;
1416                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1417                 for (i = 0; rsync_argv[i]; i++) {
1418                         char *tmp = msg;
1419                         msg = make_message("%s%s%s", tmp? tmp : "",
1420                                 tmp? " " : "", rsync_argv[i]);
1421                         free(tmp);
1422                 }
1423                 free_rsync_argv(rsync_argv);
1424                 dss_msg("%s\n", msg);
1425                 free(msg);
1426                 return 1;
1427         }
1428         pre_create_hook();
1429         if (create_pid) {
1430                 ret = wait_for_process(create_pid, &status);
1431                 if (ret < 0)
1432                         return ret;
1433                 ret = handle_pre_create_hook_exit(status);
1434                 if (ret <= 0) /* error, or pre-create failed */
1435                         return ret;
1436         }
1437         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1438         ret = create_snapshot(rsync_argv);
1439         if (ret < 0)
1440                 goto out;
1441         ret = wait_for_process(create_pid, &status);
1442         if (ret < 0)
1443                 goto out;
1444         ret = handle_rsync_exit(status);
1445         if (ret < 0)
1446                 goto out;
1447         post_create_hook();
1448         if (create_pid)
1449                 ret = wait_for_process(create_pid, &status);
1450 out:
1451         free_rsync_argv(rsync_argv);
1452         return ret;
1453 }
1454
1455 static int com_ls(void)
1456 {
1457         int i;
1458         struct snapshot_list sl;
1459         struct snapshot *s;
1460
1461         dss_get_snapshot_list(&sl);
1462         FOR_EACH_SNAPSHOT(s, i, &sl) {
1463                 int64_t d = 0;
1464                 if (s->flags & SS_COMPLETE)
1465                         d = (s->completion_time - s->creation_time) / 60;
1466                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1467         }
1468         free_snapshot_list(&sl);
1469         return 1;
1470 }
1471
1472 static int setup_signal_handling(void)
1473 {
1474         int ret;
1475
1476         DSS_INFO_LOG(("setting up signal handlers\n"));
1477         signal_pipe = signal_init(); /* always successful */
1478         ret = install_sighandler(SIGINT);
1479         if (ret < 0)
1480                 return ret;
1481         ret = install_sighandler(SIGTERM);
1482         if (ret < 0)
1483                 return ret;
1484         return install_sighandler(SIGCHLD);
1485 }
1486
1487 /**
1488  * The main function of dss.
1489  *
1490  * \param argc Usual argument count.
1491  * \param argv Usual argument vector.
1492  */
1493 int main(int argc, char **argv)
1494 {
1495         int ret;
1496         struct cmdline_parser_params params;
1497
1498         params.override = 0;
1499         params.initialize = 1;
1500         params.check_required = 0;
1501         params.check_ambiguity = 0;
1502         params.print_errors = 1;
1503
1504         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1505         ret = parse_config_file(0);
1506         if (ret < 0)
1507                 goto out;
1508         if (ret == 0) { /* no config file given */
1509                 /*
1510                  * Parse the command line options again, but this time check
1511                  * that all required options are given.
1512                  */
1513                 struct cmdline_parser_params params;
1514                 params.override = 1;
1515                 params.initialize = 1;
1516                 params.check_required = 1;
1517                 params.check_ambiguity = 1;
1518                 params.print_errors = 1;
1519                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1520         }
1521         if (conf.daemon_given)
1522                 daemon_init();
1523         ret = change_to_dest_dir();
1524         if (ret < 0)
1525                 goto out;
1526         dump_dss_config("startup");
1527         ret = setup_signal_handling();
1528         if (ret < 0)
1529                 goto out;
1530         ret = call_command_handler();
1531         signal_shutdown();
1532 out:
1533         if (ret < 0)
1534                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1535         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1536 }