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 <stdbool.h>
  15 #include <sys/stat.h>
  16 #include <unistd.h>
  17 #include <inttypes.h>
  18 #include <sys/time.h>
  19 #include <time.h>
  20 #include <sys/wait.h>
  21 #include <fnmatch.h>
  22 #include <limits.h>
  23
  24
  25 #include "gcc-compat.h"
  26 #include "cmdline.h"
  27 #include "log.h"
  28 #include "str.h"
  29 #include "err.h"
  30 #include "file.h"
  31 #include "exec.h"
  32 #include "daemon.h"
  33 #include "sig.h"
  34 #include "df.h"
  35 #include "tv.h"
  36 #include "snap.h"
  37 #include "ipc.h"
  38
  39 /** Command line and config file options. */
  40 static struct gengetopt_args_info conf;
  41 /** Non-NULL if we log to a file. */
  42 static FILE *logfile;
  43 /** The read end of the signal pipe */
  44 static int signal_pipe;
  45 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
  46 static pid_t create_pid;
  47 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
  48 static int create_process_stopped;
  49 /** How many times in a row the rsync command failed. */
  50 static int num_consecutive_rsync_errors;
  51 /** Process id of current pre-remove/rm/post-remove process. */
  52 static pid_t remove_pid;
  53 /** When the next snapshot is due. */
  54 static int64_t next_snapshot_time;
  55 /** When to try to remove something. */
  56 static struct timeval next_removal_check;
  57 /** Creation time of the snapshot currently being created. */
  58 static int64_t current_snapshot_creation_time;
  59 /** The snapshot currently being removed. */
  60 struct snapshot *snapshot_currently_being_removed;
  61 /** Needed by the post-create hook. */
  62 static char *path_to_last_complete_snapshot;
  63 static char *name_of_reference_snapshot;
  64 /** \sa \ref snap.h for details. */
  65 enum hook_status snapshot_creation_status;
  66 /** \sa \ref snap.h for details. */
  67 enum hook_status snapshot_removal_status;
  68
  69
  70 DEFINE_DSS_ERRLIST;
  71 static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
  72
  73 /* may be called with ds == NULL. */
  74 static int disk_space_low(struct disk_space *ds)
  75 {
  76         struct disk_space ds_struct;
  77
  78         if (!ds) {
  79                 int ret = get_disk_space(".", &ds_struct);
  80                 if (ret < 0)
  81                         return ret;
  82                 ds = &ds_struct;
  83         }
  84         if (conf.min_free_mb_arg)
  85                 if (ds->free_mb < conf.min_free_mb_arg)
  86                         return 1;
  87         if (conf.min_free_percent_arg)
  88                 if (ds->percent_free < conf.min_free_percent_arg)
  89                         return 1;
  90         if (conf.min_free_percent_inodes_arg)
  91                 if (ds->percent_free_inodes < conf.min_free_percent_inodes_arg)
  92                         return 1;
  93         return 0;
  94 }
  95
  96 static void dump_dss_config(const char *msg)
  97 {
  98         const char dash[] = "-----------------------------";
  99         int ret;
 100         FILE *log = logfile? logfile : stderr;
 101         struct disk_space ds;
 102         int64_t now = get_current_time();
 103
 104         if (conf.loglevel_arg > INFO)
 105                 return;
 106
 107         fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
 108         fprintf(log, "\n*** disk space ***\n\n");
 109         ret = get_disk_space(".", &ds);
 110         if (ret >= 0) {
 111                 DSS_INFO_LOG(("disk space low: %s\n", disk_space_low(&ds)?
 112                         "yes" : "no"));
 113                 log_disk_space(&ds);
 114         } else
 115                 DSS_ERROR_LOG(("can not get free disk space: %s\n",
 116                         dss_strerror(-ret)));
 117
 118         /* we continue on errors from get_disk_space */
 119
 120         fprintf(log, "\n*** command line and config file options ***\n\n");
 121         cmdline_parser_dump(log, &conf);
 122         fprintf(log, "\n*** internal state ***\n\n");
 123         fprintf(log,
 124                 "pid: %d\n"
 125                 "logile: %s\n"
 126                 "snapshot_currently_being_removed: %s\n"
 127                 "path_to_last_complete_snapshot: %s\n"
 128                 "reference_snapshot: %s\n"
 129                 "snapshot_creation_status: %s\n"
 130                 "snapshot_removal_status: %s\n"
 131                 "num_consecutive_rsync_errors: %d\n"
 132                 ,
 133                 (int) getpid(),
 134                 logfile? conf.logfile_arg : "stderr",
 135                 snapshot_currently_being_removed?
 136                         snapshot_currently_being_removed->name : "(none)",
 137                 path_to_last_complete_snapshot?
 138                         path_to_last_complete_snapshot : "(none)",
 139                 name_of_reference_snapshot?
 140                         name_of_reference_snapshot : "(none)",
 141                 hook_status_description[snapshot_creation_status],
 142                 hook_status_description[snapshot_removal_status],
 143                 num_consecutive_rsync_errors
 144         );
 145         if (create_pid != 0)
 146                 fprintf(log,
 147                         "create_pid: %" PRId32 "\n"
 148                         "create process is %sstopped\n"
 149                         ,
 150                         create_pid,
 151                         create_process_stopped? "" : "not "
 152                 );
 153         if (remove_pid != 0)
 154                 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
 155         if (next_snapshot_time != 0)
 156                 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
 157                         next_snapshot_time - now);
 158         if (current_snapshot_creation_time != 0)
 159                 fprintf(log, "current_snapshot_creation_time: %"
 160                         PRId64 " (%" PRId64 " seconds ago)\n",
 161                         current_snapshot_creation_time,
 162                         now - current_snapshot_creation_time
 163                 );
 164         if (next_removal_check.tv_sec != 0) {
 165                 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
 166                         (long long unsigned)next_removal_check.tv_sec,
 167                         now - (long long unsigned)next_removal_check.tv_sec
 168                 );
 169
 170         }
 171         fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
 172 }
 173
 174 /* a litte cpp magic helps to DRY */
 175 #define COMMANDS \
 176         COMMAND(ls) \
 177         COMMAND(create) \
 178         COMMAND(prune) \
 179         COMMAND(run) \
 180         COMMAND(kill) \
 181         COMMAND(reload) \
 182
 183 #define COMMAND(x) static int com_ ##x(void);
 184 COMMANDS
 185 #undef COMMAND
 186 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
 187 static int call_command_handler(void)
 188 {
 189         COMMANDS
 190         DSS_EMERG_LOG(("BUG: did not find command handler\n"));
 191         return -E_BUG;
 192 }
 193 #undef COMMAND
 194 #undef COMMANDS
 195
 196 static int loglevel = -1;
 197 static const char *location_file = NULL;
 198 static int         location_line = -1;
 199 static const char *location_func = NULL;
 200
 201 void dss_log_set_params(int ll, const char *file, int line, const char *func)
 202 {
 203         loglevel = ll;
 204         location_file = file;
 205         location_line = line;
 206         location_func = func;
 207 }
 208
 209 /**
 210  * The log function of dss.
 211  *
 212  * \param ll Loglevel.
 213  * \param fml Usual format string.
 214  *
 215  * All DSS_XXX_LOG() macros use this function.
 216  */
 217 __printf_1_2 void dss_log(const char* fmt,...)
 218 {
 219         va_list argp;
 220         FILE *outfd;
 221         struct tm *tm;
 222         time_t t1;
 223         char str[255] = "";
 224
 225         if (loglevel < conf.loglevel_arg)
 226                 return;
 227         outfd = logfile? logfile : stderr;
 228         time(&t1);
 229         tm = localtime(&t1);
 230         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 231         fprintf(outfd, "%s ", str);
 232         if (conf.loglevel_arg <= INFO)
 233                 fprintf(outfd, "%i: ", loglevel);
 234 #ifdef DSS_NO_FUNC_NAMES
 235         fprintf(outfd, "%s:%d: ", location_file, location_line);
 236 #else
 237         fprintf(outfd, "%s: ", location_func);
 238 #endif
 239         va_start(argp, fmt);
 240         vfprintf(outfd, fmt, argp);
 241         va_end(argp);
 242 }
 243
 244 /**
 245  * Print a message either to stdout or to the log file.
 246  */
 247 static __printf_1_2 void dss_msg(const char* fmt,...)
 248 {
 249         FILE *outfd = conf.daemon_given? logfile : stdout;
 250         va_list argp;
 251         va_start(argp, fmt);
 252         vfprintf(outfd, fmt, argp);
 253         va_end(argp);
 254 }
 255
 256 static char *get_config_file_name(void)
 257 {
 258         char *home, *config_file;
 259
 260         if (conf.config_file_given)
 261                 return dss_strdup(conf.config_file_arg);
 262         home = get_homedir();
 263         config_file = make_message("%s/.dssrc", home);
 264         free(home);
 265         return config_file;
 266 }
 267
 268 static int send_signal(int sig)
 269 {
 270         pid_t pid;
 271         char *config_file = get_config_file_name();
 272         int ret = get_dss_pid(config_file, &pid);
 273
 274         free(config_file);
 275         if (ret < 0)
 276                 return ret;
 277         if (conf.dry_run_given) {
 278                 dss_msg("%d\n", (int)pid);
 279                 return 0;
 280         }
 281         ret = kill(pid, sig);
 282         if (ret < 0)
 283                 return -ERRNO_TO_DSS_ERROR(errno);
 284         return 1;
 285 }
 286
 287 static int com_kill(void)
 288 {
 289         return send_signal(SIGTERM);
 290 }
 291
 292 static int com_reload(void)
 293 {
 294         return send_signal(SIGHUP);
 295 }
 296
 297 static void dss_get_snapshot_list(struct snapshot_list *sl)
 298 {
 299         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 300 }
 301
 302 static int64_t compute_next_snapshot_time(void)
 303 {
 304         int64_t x = 0, now = get_current_time(), unit_interval
 305                 = 24 * 3600 * conf.unit_interval_arg, ret;
 306         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 307                 num_complete = 0;
 308         int i;
 309         struct snapshot *s = NULL;
 310         struct snapshot_list sl;
 311
 312         dss_get_snapshot_list(&sl);
 313         FOR_EACH_SNAPSHOT(s, i, &sl) {
 314                 if (!(s->flags & SS_COMPLETE))
 315                         continue;
 316                 num_complete++;
 317                 x += s->completion_time - s->creation_time;
 318         }
 319         assert(x >= 0);
 320
 321         ret = now;
 322         if (num_complete == 0)
 323                 goto out;
 324         x /= num_complete; /* avg time to create one snapshot */
 325         if (unit_interval < x * wanted) /* oops, no sleep at all */
 326                 goto out;
 327         ret = s->completion_time + unit_interval / wanted - x;
 328 out:
 329         free_snapshot_list(&sl);
 330         return ret;
 331 }
 332
 333 static inline void invalidate_next_snapshot_time(void)
 334 {
 335         next_snapshot_time = 0;
 336 }
 337
 338 static inline int next_snapshot_time_is_valid(void)
 339 {
 340         return next_snapshot_time != 0;
 341 }
 342
 343 static int next_snapshot_is_due(void)
 344 {
 345         int64_t now = get_current_time();
 346
 347         if (!next_snapshot_time_is_valid())
 348                 next_snapshot_time = compute_next_snapshot_time();
 349         if (next_snapshot_time <= now) {
 350                 DSS_DEBUG_LOG(("next snapshot: now\n"));
 351                 return 1;
 352         }
 353         DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
 354                 next_snapshot_time - now));
 355         return 0;
 356 }
 357
 358 static void pre_create_hook(void)
 359 {
 360         assert(snapshot_creation_status == HS_READY);
 361         /* make sure that the next snapshot time will be recomputed */
 362         invalidate_next_snapshot_time();
 363         DSS_DEBUG_LOG(("executing %s\n", conf.pre_create_hook_arg));
 364         dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
 365         snapshot_creation_status = HS_PRE_RUNNING;
 366 }
 367
 368 static void pre_remove_hook(struct snapshot *s, const char *why)
 369 {
 370         char *cmd;
 371
 372         if (!s)
 373                 return;
 374         DSS_DEBUG_LOG(("%s snapshot %s\n", why, s->name));
 375         assert(snapshot_removal_status == HS_READY);
 376         assert(remove_pid == 0);
 377         assert(!snapshot_currently_being_removed);
 378
 379         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 380         *snapshot_currently_being_removed = *s;
 381         snapshot_currently_being_removed->name = dss_strdup(s->name);
 382
 383         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 384                 conf.dest_dir_arg, s->name);
 385         DSS_DEBUG_LOG(("executing %s\n", cmd));
 386         dss_exec_cmdline_pid(&remove_pid, cmd);
 387         free(cmd);
 388         snapshot_removal_status = HS_PRE_RUNNING;
 389 }
 390
 391 static int exec_rm(void)
 392 {
 393         struct snapshot *s = snapshot_currently_being_removed;
 394         char *new_name = being_deleted_name(s);
 395         char *argv[4];
 396         int ret;
 397
 398         argv[0] = "rm";
 399         argv[1] = "-rf";
 400         argv[2] = new_name;
 401         argv[3] = NULL;
 402
 403         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 404         assert(remove_pid == 0);
 405
 406         DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
 407         ret = dss_rename(s->name, new_name);
 408         if (ret < 0)
 409                 goto out;
 410         dss_exec(&remove_pid, argv[0], argv);
 411         snapshot_removal_status = HS_RUNNING;
 412 out:
 413         free(new_name);
 414         return ret;
 415 }
 416
 417 static int snapshot_is_being_created(struct snapshot *s)
 418 {
 419         return s->creation_time == current_snapshot_creation_time;
 420 }
 421
 422 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 423 {
 424         struct snapshot *s;
 425         int i;
 426
 427         DSS_DEBUG_LOG(("looking for orphaned snapshots\n"));
 428         FOR_EACH_SNAPSHOT(s, i, sl) {
 429                 if (snapshot_is_being_created(s))
 430                         continue;
 431                 /*
 432                  * We know that no rm is currently running, so if s is marked
 433                  * as being deleted, a previously started rm must have failed.
 434                  */
 435                 if (s->flags & SS_BEING_DELETED)
 436                         return s;
 437
 438                 if (s->flags & SS_COMPLETE) /* good snapshot */
 439                         continue;
 440                 /*
 441                  * This snapshot is incomplete and it is not the snapshot
 442                  * currently being created. However, we must not remove it if
 443                  * rsync is about to be restarted. As only the newest snapshot
 444                  * can be restarted, this snapshot is orphaned if it is not the
 445                  * newest snapshot or if we are not about to restart rsync.
 446                  */
 447                 if (get_newest_snapshot(sl) != s)
 448                         return s;
 449                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 450                         return s;
 451         }
 452         /* no orphaned snapshots */
 453         return NULL;
 454 }
 455
 456 static int is_reference_snapshot(struct snapshot *s)
 457 {
 458         if (!name_of_reference_snapshot)
 459                 return 0;
 460         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 461 }
 462
 463 /*
 464  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 465  */
 466 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 467 {
 468         int i, interval;
 469         struct snapshot *s;
 470         unsigned missing = 0;
 471
 472         DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
 473         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 474                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 475                 unsigned num = sl->interval_count[interval];
 476                 struct snapshot *victim = NULL, *prev = NULL;
 477                 int64_t score = LONG_MAX;
 478
 479                 if (keep >= num)
 480                         missing += keep - num;
 481                 if (keep + missing >= num)
 482                         continue;
 483                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 484                 FOR_EACH_SNAPSHOT(s, i, sl) {
 485                         int64_t this_score;
 486
 487                         if (snapshot_is_being_created(s))
 488                                 continue;
 489                         if (is_reference_snapshot(s))
 490                                 continue;
 491                         if (s->interval > interval) {
 492                                 prev = s;
 493                                 continue;
 494                         }
 495                         if (s->interval < interval)
 496                                 break;
 497                         if (!victim) {
 498                                 victim = s;
 499                                 prev = s;
 500                                 continue;
 501                         }
 502                         assert(prev);
 503                         /* check if s is a better victim */
 504                         this_score = s->creation_time - prev->creation_time;
 505                         assert(this_score >= 0);
 506                         if (this_score < score) {
 507                                 score = this_score;
 508                                 victim = s;
 509                         }
 510                         prev = s;
 511                 }
 512                 assert(victim);
 513                 return victim;
 514         }
 515         return NULL;
 516 }
 517
 518 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 519 {
 520         int i;
 521         struct snapshot *s;
 522
 523         DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
 524                 conf.num_intervals_arg));
 525         FOR_EACH_SNAPSHOT(s, i, sl) {
 526                 if (snapshot_is_being_created(s))
 527                         continue;
 528                 if (is_reference_snapshot(s))
 529                         continue;
 530                 if (s->interval < conf.num_intervals_arg)
 531                         continue;
 532                 return s;
 533         }
 534         return NULL;
 535 }
 536
 537 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 538 {
 539         int i, num_complete;
 540         struct snapshot *s, *ref = NULL;
 541
 542         num_complete = num_complete_snapshots(sl);
 543         if (num_complete <= conf.min_complete_arg)
 544                 return NULL;
 545         FOR_EACH_SNAPSHOT(s, i, sl) {
 546                 if (snapshot_is_being_created(s))
 547                         continue;
 548                 if (is_reference_snapshot(s)) { /* avoid this one */
 549                         ref = s;
 550                         continue;
 551                 }
 552                 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 553                 return s;
 554         }
 555         assert(ref);
 556         DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
 557         return ref;
 558 }
 559
 560 static int rename_incomplete_snapshot(int64_t start)
 561 {
 562         char *old_name;
 563         int ret;
 564         int64_t now;
 565
 566         /*
 567          * We don't want the dss_rename() below to fail with EEXIST because the
 568          * last complete snapshot was created (and completed) in the same
 569          * second as this one.
 570          */
 571         while ((now = get_current_time()) == start)
 572                 sleep(1);
 573         free(path_to_last_complete_snapshot);
 574         ret = complete_name(start, now, &path_to_last_complete_snapshot);
 575         if (ret < 0)
 576                 return ret;
 577         old_name = incomplete_name(start);
 578         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 579         if (ret >= 0)
 580                 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
 581                         path_to_last_complete_snapshot));
 582         free(old_name);
 583         return ret;
 584 }
 585
 586 static int try_to_free_disk_space(void)
 587 {
 588         int ret;
 589         struct snapshot_list sl;
 590         struct snapshot *victim;
 591         struct timeval now;
 592         const char *why;
 593         int low_disk_space;
 594
 595         ret = disk_space_low(NULL);
 596         if (ret < 0)
 597                 return ret;
 598         low_disk_space = ret;
 599         gettimeofday(&now, NULL);
 600         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 601                 return 0;
 602         if (!low_disk_space) {
 603                 if (conf.keep_redundant_given)
 604                         return 0;
 605                 if (snapshot_creation_status != HS_READY)
 606                         return 0;
 607                 if (next_snapshot_is_due())
 608                         return 0;
 609         }
 610         /*
 611          * Idle and --keep_redundant not given, or low disk space. Look at
 612          * existing snapshots.
 613          */
 614         dss_get_snapshot_list(&sl);
 615         ret = 0;
 616         /*
 617          * Don't remove anything if there is free space and we have fewer
 618          * snapshots than configured, plus one. This way there is always one
 619          * snapshot that can be recycled.
 620          */
 621         if (!low_disk_space && sl.num_snapshots <= 1 << conf.num_intervals_arg)
 622                 goto out;
 623         why = "outdated";
 624         victim = find_outdated_snapshot(&sl);
 625         if (victim)
 626                 goto remove;
 627         why = "redundant";
 628         victim = find_redundant_snapshot(&sl);
 629         if (victim)
 630                 goto remove;
 631         why = "orphaned";
 632         victim = find_orphaned_snapshot(&sl);
 633         if (victim)
 634                 goto remove;
 635         /* try harder only if disk space is low */
 636         if (!low_disk_space)
 637                 goto out;
 638         DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 639         victim = find_oldest_removable_snapshot(&sl);
 640         if (victim)
 641                 goto remove;
 642         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 643         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 644         goto out;
 645 remove:
 646         pre_remove_hook(victim, why);
 647 out:
 648         free_snapshot_list(&sl);
 649         return ret;
 650 }
 651
 652 static void post_create_hook(void)
 653 {
 654         char *cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 655                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 656         DSS_NOTICE_LOG(("executing %s\n", cmd));
 657         dss_exec_cmdline_pid(&create_pid, cmd);
 658         free(cmd);
 659         snapshot_creation_status = HS_POST_RUNNING;
 660 }
 661
 662 static void post_remove_hook(void)
 663 {
 664         char *cmd;
 665         struct snapshot *s = snapshot_currently_being_removed;
 666
 667         assert(s);
 668
 669         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 670                 conf.dest_dir_arg, s->name);
 671         DSS_NOTICE_LOG(("executing %s\n", cmd));
 672         dss_exec_cmdline_pid(&remove_pid, cmd);
 673         free(cmd);
 674         snapshot_removal_status = HS_POST_RUNNING;
 675 }
 676
 677 static void dss_kill(pid_t pid, int sig, const char *msg)
 678 {
 679         const char *signame, *process_name;
 680
 681         if (pid == 0)
 682                 return;
 683         switch (sig) {
 684         case SIGTERM: signame = "TERM"; break;
 685         case SIGSTOP: signame = "STOP"; break;
 686         case SIGCONT: signame = "CONT"; break;
 687         default: signame = "????";
 688         }
 689
 690         if (pid == create_pid)
 691                 process_name = "create";
 692         else if (pid == remove_pid)
 693                 process_name = "remove";
 694         else process_name = "??????";
 695
 696         if (msg)
 697                 DSS_INFO_LOG(("%s\n", msg));
 698         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 699                 sig, signame, (int)pid, process_name));
 700         if (kill(pid, sig) >= 0)
 701                 return;
 702         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 703                 sig, signame, (int)pid, process_name));
 704 }
 705
 706 static void stop_create_process(void)
 707 {
 708         if (create_process_stopped)
 709                 return;
 710         dss_kill(create_pid, SIGSTOP, "suspending create process");
 711         create_process_stopped = 1;
 712 }
 713
 714 static void restart_create_process(void)
 715 {
 716         if (!create_process_stopped)
 717                 return;
 718         dss_kill(create_pid, SIGCONT, "resuming create process");
 719         create_process_stopped = 0;
 720 }
 721
 722 /**
 723  * Print a log message about the exit status of a child.
 724  */
 725 static void log_termination_msg(pid_t pid, int status)
 726 {
 727         if (WIFEXITED(status))
 728                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 729                         WEXITSTATUS(status)));
 730         else if (WIFSIGNALED(status))
 731                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 732                         WTERMSIG(status)));
 733         else
 734                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 735 }
 736
 737 static int wait_for_process(pid_t pid, int *status)
 738 {
 739         int ret;
 740
 741         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 742         for (;;) {
 743                 fd_set rfds;
 744
 745                 FD_ZERO(&rfds);
 746                 FD_SET(signal_pipe, &rfds);
 747                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 748                 if (ret < 0)
 749                         break;
 750                 ret = next_signal();
 751                 if (!ret)
 752                         continue;
 753                 if (ret == SIGCHLD) {
 754                         ret = waitpid(pid, status, 0);
 755                         if (ret >= 0)
 756                                 break;
 757                         if (errno != EINTR) { /* error */
 758                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 759                                 break;
 760                         }
 761                 }
 762                 /* SIGINT or SIGTERM */
 763                 dss_kill(pid, SIGTERM, "killing child process");
 764         }
 765         if (ret < 0)
 766                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 767         else
 768                 log_termination_msg(pid, *status);
 769         return ret;
 770 }
 771
 772 static void handle_pre_remove_exit(int status)
 773 {
 774         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 775                 snapshot_removal_status = HS_READY;
 776                 gettimeofday(&next_removal_check, NULL);
 777                 next_removal_check.tv_sec += 60;
 778                 return;
 779         }
 780         snapshot_removal_status = HS_PRE_SUCCESS;
 781 }
 782
 783 static int handle_rm_exit(int status)
 784 {
 785         if (!WIFEXITED(status)) {
 786                 snapshot_removal_status = HS_READY;
 787                 return -E_INVOLUNTARY_EXIT;
 788         }
 789         if (WEXITSTATUS(status)) {
 790                 snapshot_removal_status = HS_READY;
 791                 return -E_BAD_EXIT_CODE;
 792         }
 793         snapshot_removal_status = HS_SUCCESS;
 794         return 1;
 795 }
 796
 797 static void handle_post_remove_exit(void)
 798 {
 799         snapshot_removal_status = HS_READY;
 800 }
 801
 802 static int handle_remove_exit(int status)
 803 {
 804         int ret;
 805         struct snapshot *s = snapshot_currently_being_removed;
 806
 807         assert(s);
 808         switch (snapshot_removal_status) {
 809         case HS_PRE_RUNNING:
 810                 handle_pre_remove_exit(status);
 811                 ret = 1;
 812                 break;
 813         case HS_RUNNING:
 814                 ret = handle_rm_exit(status);
 815                 break;
 816         case HS_POST_RUNNING:
 817                 handle_post_remove_exit();
 818                 ret = 1;
 819                 break;
 820         default:
 821                 ret = -E_BUG;
 822         }
 823         if (snapshot_removal_status == HS_READY) {
 824                 free(s->name);
 825                 free(s);
 826                 snapshot_currently_being_removed = NULL;
 827         }
 828         remove_pid = 0;
 829         return ret;
 830 }
 831
 832 static int wait_for_remove_process(void)
 833 {
 834         int status, ret;
 835
 836         assert(remove_pid);
 837         assert(
 838                 snapshot_removal_status == HS_PRE_RUNNING ||
 839                 snapshot_removal_status == HS_RUNNING ||
 840                 snapshot_removal_status == HS_POST_RUNNING
 841         );
 842         ret = wait_for_process(remove_pid, &status);
 843         if (ret < 0)
 844                 return ret;
 845         return handle_remove_exit(status);
 846 }
 847
 848 static int handle_rsync_exit(int status)
 849 {
 850         int es, ret;
 851
 852         if (!WIFEXITED(status)) {
 853                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 854                 ret = -E_INVOLUNTARY_EXIT;
 855                 snapshot_creation_status = HS_READY;
 856                 goto out;
 857         }
 858         es = WEXITSTATUS(status);
 859         /*
 860          * Restart rsync on non-fatal errors:
 861          * 24: Partial transfer due to vanished source files
 862          */
 863         if (es != 0 && es != 24) {
 864                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 865                         es, ++num_consecutive_rsync_errors));
 866                 if (conf.create_given) {
 867                         ret = -E_BAD_EXIT_CODE;
 868                         goto out;
 869                 }
 870                 if (num_consecutive_rsync_errors > conf.max_rsync_errors_arg) {
 871                         ret = -E_TOO_MANY_RSYNC_ERRORS;
 872                         snapshot_creation_status = HS_READY;
 873                         goto out;
 874                 }
 875                 DSS_WARNING_LOG(("restarting rsync process\n"));
 876                 snapshot_creation_status = HS_NEEDS_RESTART;
 877                 next_snapshot_time = get_current_time() + 60;
 878                 ret = 1;
 879                 goto out;
 880         }
 881         num_consecutive_rsync_errors = 0;
 882         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 883         if (ret < 0)
 884                 goto out;
 885         snapshot_creation_status = HS_SUCCESS;
 886         free(name_of_reference_snapshot);
 887         name_of_reference_snapshot = NULL;
 888 out:
 889         create_process_stopped = 0;
 890         return ret;
 891 }
 892
 893 static int handle_pre_create_hook_exit(int status)
 894 {
 895         int es, ret;
 896         static int warn_count;
 897
 898         if (!WIFEXITED(status)) {
 899                 snapshot_creation_status = HS_READY;
 900                 ret = -E_INVOLUNTARY_EXIT;
 901                 goto out;
 902         }
 903         es = WEXITSTATUS(status);
 904         if (es) {
 905                 if (!warn_count--) {
 906                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
 907                                 conf.pre_create_hook_arg, es));
 908                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
 909                         warn_count = 60; /* warn only once per hour */
 910                 }
 911                 next_snapshot_time = get_current_time() + 60;
 912                 snapshot_creation_status = HS_READY;
 913                 ret = 0;
 914                 goto out;
 915         }
 916         warn_count = 0;
 917         snapshot_creation_status = HS_PRE_SUCCESS;
 918         ret = 1;
 919 out:
 920         return ret;
 921 }
 922
 923 static int handle_sigchld(void)
 924 {
 925         pid_t pid;
 926         int status, ret = reap_child(&pid, &status);
 927
 928         if (ret <= 0)
 929                 return ret;
 930
 931         if (pid == create_pid) {
 932                 switch (snapshot_creation_status) {
 933                 case HS_PRE_RUNNING:
 934                         ret = handle_pre_create_hook_exit(status);
 935                         break;
 936                 case HS_RUNNING:
 937                         ret = handle_rsync_exit(status);
 938                         break;
 939                 case HS_POST_RUNNING:
 940                         snapshot_creation_status = HS_READY;
 941                         ret = 1;
 942                         break;
 943                 default:
 944                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
 945                                 snapshot_creation_status));
 946                         return -E_BUG;
 947                 }
 948                 create_pid = 0;
 949                 return ret;
 950         }
 951         if (pid == remove_pid) {
 952                 ret = handle_remove_exit(status);
 953                 if (ret < 0)
 954                         return ret;
 955                 return ret;
 956         }
 957         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
 958         return -E_BUG;
 959 }
 960
 961 static int check_config(void)
 962 {
 963         if (conf.unit_interval_arg <= 0) {
 964                 DSS_ERROR_LOG(("bad unit interval: %i\n", conf.unit_interval_arg));
 965                 return -E_INVALID_NUMBER;
 966         }
 967         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", conf.unit_interval_arg));
 968         if (conf.num_intervals_arg <= 0 || conf.num_intervals_arg > 30) {
 969                 DSS_ERROR_LOG(("bad number of intervals: %i\n",
 970                         conf.num_intervals_arg));
 971                 return -E_INVALID_NUMBER;
 972         }
 973         DSS_DEBUG_LOG(("number of intervals: %i\n", conf.num_intervals_arg));
 974         return 1;
 975 }
 976
 977 /*
 978  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 979  * file was read successfully.
 980  */
 981 static int parse_config_file(bool sighup)
 982 {
 983         int ret, config_file_exists;
 984         char *config_file = get_config_file_name();
 985         struct stat statbuf;
 986         char *old_logfile_arg = NULL;
 987         int old_daemon_given = 0;
 988
 989         if (sighup) {
 990                 if (conf.logfile_given)
 991                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 992                 old_daemon_given = conf.daemon_given;
 993         }
 994
 995         config_file_exists = !stat(config_file, &statbuf);
 996         if (!config_file_exists && conf.config_file_given) {
 997                 ret = -ERRNO_TO_DSS_ERROR(errno);
 998                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
 999                 goto out;
1000         }
1001         if (config_file_exists) {
1002                 struct cmdline_parser_params params;
1003                 params.override = sighup;
1004                 params.initialize = 0;
1005                 params.check_required = 1;
1006                 params.check_ambiguity = 0;
1007                 params.print_errors = 1;
1008                 if (sighup) { /* invalidate all rsync options */
1009                         int i;
1010
1011                         for (i = 0; i < conf.rsync_option_given; i++) {
1012                                 free(conf.rsync_option_arg[i]);
1013                                 conf.rsync_option_arg[i] = NULL;
1014                         }
1015                         conf.rsync_option_given = 0;
1016                 }
1017                 cmdline_parser_config_file(config_file, &conf, &params);
1018         }
1019         ret = check_config();
1020         if (ret < 0)
1021                 goto out;
1022         if (sighup) {
1023                 /* don't change daemon mode on SIGHUP */
1024                 conf.daemon_given = old_daemon_given;
1025                 close_log(logfile);
1026                 logfile = NULL;
1027                 if (conf.logfile_given)
1028                         free(old_logfile_arg);
1029                 else if (conf.daemon_given) { /* re-use old logfile */
1030                         conf.logfile_arg = old_logfile_arg;
1031                         conf.logfile_given = 1;
1032                 }
1033         }
1034         if (conf.logfile_given && conf.run_given && conf.daemon_given) {
1035                 logfile = open_log(conf.logfile_arg);
1036                 log_welcome(conf.loglevel_arg);
1037         }
1038         DSS_DEBUG_LOG(("loglevel: %d\n", conf.loglevel_arg));
1039         ret = config_file_exists;
1040 out:
1041         free(config_file);
1042         if (ret < 0)
1043                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1044         return ret;
1045 }
1046
1047 static int change_to_dest_dir(void)
1048 {
1049         DSS_INFO_LOG(("changing cwd to %s\n", conf.dest_dir_arg));
1050         return dss_chdir(conf.dest_dir_arg);
1051 }
1052
1053 static int handle_sighup(void)
1054 {
1055         int ret;
1056
1057         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1058         dump_dss_config("old");
1059         ret = parse_config_file(true /* SIGHUP */);
1060         if (ret < 0)
1061                 return ret;
1062         dump_dss_config("reloaded");
1063         invalidate_next_snapshot_time();
1064         return change_to_dest_dir();
1065 }
1066
1067 static void kill_children(void)
1068 {
1069         restart_create_process();
1070         dss_kill(create_pid, SIGTERM, NULL);
1071         dss_kill(remove_pid, SIGTERM, NULL);
1072 }
1073
1074 static int handle_signal(void)
1075 {
1076         int sig, ret = next_signal();
1077
1078         if (ret <= 0)
1079                 goto out;
1080         sig = ret;
1081         switch (sig) {
1082         case SIGINT:
1083         case SIGTERM:
1084                 kill_children();
1085                 ret = -E_SIGNAL;
1086                 break;
1087         case SIGHUP:
1088                 ret = handle_sighup();
1089                 break;
1090         case SIGCHLD:
1091                 ret = handle_sigchld();
1092                 break;
1093         }
1094 out:
1095         if (ret < 0)
1096                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1097         return ret;
1098 }
1099
1100 /*
1101  * We can not use rsync locally if the local user is different from the remote
1102  * user or if the src dir is not on the local host (or both).
1103  */
1104 static int use_rsync_locally(char *logname)
1105 {
1106         char *h = conf.remote_host_arg;
1107
1108         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1109                 return 0;
1110         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
1111                 return 0;
1112         return 1;
1113 }
1114
1115 static int rename_resume_snap(int64_t creation_time)
1116 {
1117         struct snapshot_list sl;
1118         struct snapshot *s = NULL;
1119         char *new_name = incomplete_name(creation_time);
1120         int ret;
1121         const char *why;
1122
1123         sl.num_snapshots = 0;
1124
1125         ret = 0;
1126         if (conf.no_resume_given)
1127                 goto out;
1128         dss_get_snapshot_list(&sl);
1129         /*
1130          * Snapshot recycling: We first look at the newest snapshot. If this
1131          * snapshot happens to be incomplete, the last rsync process was
1132          * aborted and we reuse this one. Otherwise we look at snapshots which
1133          * could be removed (outdated and redundant snapshots) as candidates
1134          * for recycling. If no outdated/redundant snapshot exists, we check if
1135          * there is an orphaned snapshot, which likely is useless anyway.
1136          *
1137          * Only if no existing snapshot is suitable for recycling, we bite the
1138          * bullet and create a new one.
1139          */
1140         s = get_newest_snapshot(&sl);
1141         if (!s) /* no snapshots at all */
1142                 goto out;
1143         /* re-use last snapshot if it is incomplete */
1144         why = "aborted";
1145         if ((s->flags & SS_COMPLETE) == 0)
1146                 goto out;
1147         why = "outdated";
1148         s = find_outdated_snapshot(&sl);
1149         if (s)
1150                 goto out;
1151         why = "redundant";
1152         s = find_redundant_snapshot(&sl);
1153         if (s)
1154                 goto out;
1155         why = "orphaned";
1156         s = find_orphaned_snapshot(&sl);
1157 out:
1158         if (s) {
1159                 DSS_INFO_LOG(("reusing %s snapshot %s\n", why, s->name));
1160                 ret = dss_rename(s->name, new_name);
1161         }
1162         if (ret >= 0)
1163                 DSS_NOTICE_LOG(("creating new snapshot %s\n", new_name));
1164         free(new_name);
1165         free_snapshot_list(&sl);
1166         return ret;
1167 }
1168
1169 static void create_rsync_argv(char ***argv, int64_t *num)
1170 {
1171         char *logname;
1172         int i = 0, j;
1173         struct snapshot_list sl;
1174
1175         dss_get_snapshot_list(&sl);
1176         assert(!name_of_reference_snapshot);
1177         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1178         free_snapshot_list(&sl);
1179
1180         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1181         (*argv)[i++] = dss_strdup("rsync");
1182         (*argv)[i++] = dss_strdup("-a");
1183         (*argv)[i++] = dss_strdup("--delete");
1184         for (j = 0; j < conf.rsync_option_given; j++)
1185                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1186         if (name_of_reference_snapshot) {
1187                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1188                 (*argv)[i++] = make_message("--link-dest=../%s",
1189                         name_of_reference_snapshot);
1190         } else
1191                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1192         logname = dss_logname();
1193         if (use_rsync_locally(logname))
1194                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1195         else
1196                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1197                         conf.remote_user_arg : logname,
1198                         conf.remote_host_arg, conf.source_dir_arg);
1199         free(logname);
1200         *num = get_current_time();
1201         (*argv)[i++] = incomplete_name(*num);
1202         (*argv)[i++] = NULL;
1203         for (j = 0; j < i; j++)
1204                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1205 }
1206
1207 static void free_rsync_argv(char **argv)
1208 {
1209         int i;
1210
1211         if (!argv)
1212                 return;
1213         for (i = 0; argv[i]; i++)
1214                 free(argv[i]);
1215         free(argv);
1216 }
1217
1218 static int create_snapshot(char **argv)
1219 {
1220         int ret;
1221
1222         ret = rename_resume_snap(current_snapshot_creation_time);
1223         if (ret < 0)
1224                 return ret;
1225         dss_exec(&create_pid, argv[0], argv);
1226         snapshot_creation_status = HS_RUNNING;
1227         return ret;
1228 }
1229
1230 static int select_loop(void)
1231 {
1232         int ret;
1233         /* check every 60 seconds for free disk space */
1234         struct timeval tv;
1235         char **rsync_argv = NULL;
1236
1237         for (;;) {
1238                 fd_set rfds;
1239                 struct timeval *tvp;
1240
1241                 if (remove_pid)
1242                         tvp = NULL; /* sleep until rm hook/process dies */
1243                 else { /* sleep one minute */
1244                         tv.tv_sec = 60;
1245                         tv.tv_usec = 0;
1246                         tvp = &tv;
1247                 }
1248                 FD_ZERO(&rfds);
1249                 FD_SET(signal_pipe, &rfds);
1250                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1251                 if (ret < 0)
1252                         goto out;
1253                 if (FD_ISSET(signal_pipe, &rfds)) {
1254                         ret = handle_signal();
1255                         if (ret < 0)
1256                                 goto out;
1257                 }
1258                 if (remove_pid)
1259                         continue;
1260                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1261                         ret = exec_rm();
1262                         if (ret < 0)
1263                                 goto out;
1264                         continue;
1265                 }
1266                 if (snapshot_removal_status == HS_SUCCESS) {
1267                         post_remove_hook();
1268                         continue;
1269                 }
1270                 ret = try_to_free_disk_space();
1271                 if (ret < 0)
1272                         goto out;
1273                 if (snapshot_removal_status != HS_READY) {
1274                         stop_create_process();
1275                         continue;
1276                 }
1277                 restart_create_process();
1278                 switch (snapshot_creation_status) {
1279                 case HS_READY:
1280                         if (!next_snapshot_is_due())
1281                                 continue;
1282                         pre_create_hook();
1283                         continue;
1284                 case HS_PRE_RUNNING:
1285                 case HS_RUNNING:
1286                 case HS_POST_RUNNING:
1287                         continue;
1288                 case HS_PRE_SUCCESS:
1289                         if (!name_of_reference_snapshot) {
1290                                 free_rsync_argv(rsync_argv);
1291                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1292                         }
1293                         ret = create_snapshot(rsync_argv);
1294                         if (ret < 0)
1295                                 goto out;
1296                         continue;
1297                 case HS_NEEDS_RESTART:
1298                         if (!next_snapshot_is_due())
1299                                 continue;
1300                         ret = create_snapshot(rsync_argv);
1301                         if (ret < 0)
1302                                 goto out;
1303                         continue;
1304                 case HS_SUCCESS:
1305                         post_create_hook();
1306                         continue;
1307                 }
1308         }
1309 out:
1310         return ret;
1311 }
1312
1313 static void exit_hook(int exit_code)
1314 {
1315         char *argv[3];
1316         pid_t pid;
1317
1318         argv[0] = conf.exit_hook_arg;
1319         argv[1] = dss_strerror(-exit_code);
1320         argv[2] = NULL;
1321
1322         DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
1323         dss_exec(&pid, conf.exit_hook_arg, argv);
1324 }
1325
1326 static void lock_dss_or_die(void)
1327 {
1328         char *config_file = get_config_file_name();
1329         int ret = lock_dss(config_file);
1330
1331         free(config_file);
1332         if (ret < 0) {
1333                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1334                 exit(EXIT_FAILURE);
1335         }
1336 }
1337
1338 static int com_run(void)
1339 {
1340         int ret;
1341
1342         lock_dss_or_die();
1343         if (conf.dry_run_given) {
1344                 DSS_ERROR_LOG(("dry_run not supported by this command\n"));
1345                 return -E_SYNTAX;
1346         }
1347         ret = install_sighandler(SIGHUP);
1348         if (ret < 0)
1349                 return ret;
1350         ret = select_loop();
1351         if (ret >= 0) /* impossible */
1352                 ret = -E_BUG;
1353         kill_children();
1354         exit_hook(ret);
1355         return ret;
1356 }
1357
1358 static int com_prune(void)
1359 {
1360         int ret;
1361         struct snapshot_list sl;
1362         struct snapshot *victim;
1363         struct disk_space ds;
1364         const char *why;
1365
1366         lock_dss_or_die();
1367         ret = get_disk_space(".", &ds);
1368         if (ret < 0)
1369                 return ret;
1370         log_disk_space(&ds);
1371         dss_get_snapshot_list(&sl);
1372         why = "outdated";
1373         victim = find_outdated_snapshot(&sl);
1374         if (victim)
1375                 goto rm;
1376         why = "redundant";
1377         victim = find_redundant_snapshot(&sl);
1378         if (victim)
1379                 goto rm;
1380         ret = 0;
1381         goto out;
1382 rm:
1383         if (conf.dry_run_given) {
1384                 dss_msg("%s snapshot %s (interval = %i)\n",
1385                         why, victim->name, victim->interval);
1386                 ret = 0;
1387                 goto out;
1388         }
1389         pre_remove_hook(victim, why);
1390         if (snapshot_removal_status == HS_PRE_RUNNING) {
1391                 ret = wait_for_remove_process();
1392                 if (ret < 0)
1393                         goto out;
1394                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1395                         goto out;
1396         }
1397         ret = exec_rm();
1398         if (ret < 0)
1399                 goto out;
1400         ret = wait_for_remove_process();
1401         if (ret < 0)
1402                 goto out;
1403         if (snapshot_removal_status != HS_SUCCESS)
1404                 goto out;
1405         post_remove_hook();
1406         if (snapshot_removal_status != HS_POST_RUNNING)
1407                 goto out;
1408         ret = wait_for_remove_process();
1409         if (ret < 0)
1410                 goto out;
1411         ret = 1;
1412 out:
1413         free_snapshot_list(&sl);
1414         return ret;
1415 }
1416
1417 static int com_create(void)
1418 {
1419         int ret, status;
1420         char **rsync_argv;
1421
1422         lock_dss_or_die();
1423         if (conf.dry_run_given) {
1424                 int i;
1425                 char *msg = NULL;
1426                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1427                 for (i = 0; rsync_argv[i]; i++) {
1428                         char *tmp = msg;
1429                         msg = make_message("%s%s%s", tmp? tmp : "",
1430                                 tmp? " " : "", rsync_argv[i]);
1431                         free(tmp);
1432                 }
1433                 free_rsync_argv(rsync_argv);
1434                 dss_msg("%s\n", msg);
1435                 free(msg);
1436                 return 1;
1437         }
1438         pre_create_hook();
1439         if (create_pid) {
1440                 ret = wait_for_process(create_pid, &status);
1441                 if (ret < 0)
1442                         return ret;
1443                 ret = handle_pre_create_hook_exit(status);
1444                 if (ret <= 0) /* error, or pre-create failed */
1445                         return ret;
1446         }
1447         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1448         ret = create_snapshot(rsync_argv);
1449         if (ret < 0)
1450                 goto out;
1451         ret = wait_for_process(create_pid, &status);
1452         if (ret < 0)
1453                 goto out;
1454         ret = handle_rsync_exit(status);
1455         if (ret < 0)
1456                 goto out;
1457         post_create_hook();
1458         if (create_pid)
1459                 ret = wait_for_process(create_pid, &status);
1460 out:
1461         free_rsync_argv(rsync_argv);
1462         return ret;
1463 }
1464
1465 static int com_ls(void)
1466 {
1467         int i;
1468         struct snapshot_list sl;
1469         struct snapshot *s;
1470
1471         dss_get_snapshot_list(&sl);
1472         FOR_EACH_SNAPSHOT(s, i, &sl) {
1473                 int64_t d = 0;
1474                 if (s->flags & SS_COMPLETE)
1475                         d = (s->completion_time - s->creation_time) / 60;
1476                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1477         }
1478         free_snapshot_list(&sl);
1479         return 1;
1480 }
1481
1482 static int setup_signal_handling(void)
1483 {
1484         int ret;
1485
1486         DSS_INFO_LOG(("setting up signal handlers\n"));
1487         signal_pipe = signal_init(); /* always successful */
1488         ret = install_sighandler(SIGINT);
1489         if (ret < 0)
1490                 return ret;
1491         ret = install_sighandler(SIGTERM);
1492         if (ret < 0)
1493                 return ret;
1494         return install_sighandler(SIGCHLD);
1495 }
1496
1497 int main(int argc, char **argv)
1498 {
1499         int ret;
1500         struct cmdline_parser_params params;
1501
1502         params.override = 0;
1503         params.initialize = 1;
1504         params.check_required = 0;
1505         params.check_ambiguity = 0;
1506         params.print_errors = 1;
1507
1508         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1509         ret = parse_config_file(0);
1510         ret = parse_config_file(false /* no SIGHUP */);
1511         if (ret < 0)
1512                 goto out;
1513         if (ret == 0) { /* no config file given */
1514                 /*
1515                  * Parse the command line options again, but this time check
1516                  * that all required options are given.
1517                  */
1518                 struct cmdline_parser_params params;
1519                 params.override = 1;
1520                 params.initialize = 1;
1521                 params.check_required = 1;
1522                 params.check_ambiguity = 1;
1523                 params.print_errors = 1;
1524                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1525         }
1526         if (conf.daemon_given)
1527                 daemon_init();
1528         ret = change_to_dest_dir();
1529         if (ret < 0)
1530                 goto out;
1531         dump_dss_config("startup");
1532         ret = setup_signal_handling();
1533         if (ret < 0)
1534                 goto out;
1535         ret = call_command_handler();
1536         signal_shutdown();
1537 out:
1538         if (ret < 0)
1539                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1540         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1541 }