dss.c

   1 /*
   2  * Copyright (C) 2008 Andre Noll <maan@systemlinux.org>
   3  *
   4  * Licensed under the GPL v2. For licencing details see COPYING.
   5  */
   6 #include <string.h>
   7 #include <stdlib.h>
   8 #include <stdarg.h>
   9 #include <assert.h>
  10 #include <errno.h>
  11 #include <sys/types.h>
  12 #include <signal.h>
  13 #include <ctype.h>
  14 #include <sys/stat.h>
  15 #include <unistd.h>
  16 #include <inttypes.h>
  17 #include <sys/time.h>
  18 #include <time.h>
  19 #include <sys/wait.h>
  20 #include <fnmatch.h>
  21 #include <limits.h>
  22
  23
  24 #include "gcc-compat.h"
  25 #include "cmdline.h"
  26 #include "log.h"
  27 #include "string.h"
  28 #include "error.h"
  29 #include "fd.h"
  30 #include "exec.h"
  31 #include "daemon.h"
  32 #include "signal.h"
  33 #include "df.h"
  34 #include "time.h"
  35 #include "snap.h"
  36
  37 /** Command line and config file options. */
  38 static struct gengetopt_args_info conf;
  39 /** Non-NULL if we log to a file. */
  40 static FILE *logfile;
  41 /** The read end of the signal pipe */
  42 static int signal_pipe;
  43 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
  44 static pid_t create_pid;
  45 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
  46 static int create_process_stopped;
  47 /** Process id of current pre-remove/rm/post-remove process. */
  48 static pid_t remove_pid;
  49 /** When the next snapshot is due. */
  50 static struct timeval next_snapshot_time;
  51 /** When to try to remove something. */
  52 static struct timeval next_removal_check;
  53 /** Creation time of the snapshot currently being created. */
  54 static int64_t current_snapshot_creation_time;
  55 /* The snapshot currently being removed. */
  56 struct snapshot *snapshot_currently_being_removed;
  57 /** Needed by the post-create hook. */
  58 static char *path_to_last_complete_snapshot;
  59 /** \sa \ref snap.h for details. */
  60 enum hook_status snapshot_creation_status;
  61 /** \sa \ref snap.h for details. */
  62 enum hook_status snapshot_removal_status;
  63
  64
  65 DEFINE_DSS_ERRLIST;
  66
  67
  68 /* a litte cpp magic helps to DRY */
  69 #define COMMANDS \
  70         COMMAND(ls) \
  71         COMMAND(create) \
  72         COMMAND(prune) \
  73         COMMAND(run)
  74 #define COMMAND(x) static int com_ ##x(void);
  75 COMMANDS
  76 #undef COMMAND
  77 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
  78 static int call_command_handler(void)
  79 {
  80         COMMANDS
  81         DSS_EMERG_LOG("BUG: did not find command handler\n");
  82         return -E_BUG;
  83 }
  84 #undef COMMAND
  85 #undef COMMANDS
  86
  87 /**
  88  * The log function of dss.
  89  *
  90  * \param ll Loglevel.
  91  * \param fml Usual format string.
  92  *
  93  * All DSS_XXX_LOG() macros use this function.
  94  */
  95 __printf_2_3 void dss_log(int ll, const char* fmt,...)
  96 {
  97         va_list argp;
  98         FILE *outfd;
  99         struct tm *tm;
 100         time_t t1;
 101         char str[255] = "";
 102
 103         if (ll < conf.loglevel_arg)
 104                 return;
 105         outfd = logfile? logfile : stderr;
 106         time(&t1);
 107         tm = localtime(&t1);
 108         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 109         fprintf(outfd, "%s ", str);
 110         if (conf.loglevel_arg <= INFO)
 111                 fprintf(outfd, "%i: ", ll);
 112         va_start(argp, fmt);
 113         vfprintf(outfd, fmt, argp);
 114         va_end(argp);
 115 }
 116
 117 /**
 118  * Print a message either to stdout or to the log file.
 119  */
 120 static __printf_1_2 void dss_msg(const char* fmt,...)
 121 {
 122         FILE *outfd = conf.daemon_given? logfile : stdout;
 123         va_list argp;
 124         va_start(argp, fmt);
 125         vfprintf(outfd, fmt, argp);
 126         va_end(argp);
 127 }
 128
 129 static int disk_space_low(void)
 130 {
 131         struct disk_space ds;
 132         int ret = get_disk_space(".", &ds);
 133
 134         if (ret < 0)
 135                 return ret;
 136         if (conf.min_free_mb_arg)
 137                 if (ds.free_mb < conf.min_free_mb_arg)
 138                         return 1;
 139         if (conf.min_free_percent_arg)
 140                 if (ds.percent_free < conf.min_free_percent_arg)
 141                         return 1;
 142         if (conf.min_free_percent_inodes_arg)
 143                 if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg)
 144                         return 1;
 145         return 0;
 146 }
 147
 148 static void dss_get_snapshot_list(struct snapshot_list *sl)
 149 {
 150         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 151 }
 152
 153 static int next_snapshot_is_due(void)
 154 {
 155         struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 156                 tmp, diff;
 157         int64_t x = 0;
 158         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 159                 num_complete_snapshots = 0;
 160         int i, ret;
 161         struct snapshot *s = NULL;
 162         struct snapshot_list sl;
 163
 164         gettimeofday(&now, NULL);
 165         if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 166                 return 0;
 167         assert(snapshot_creation_status == HS_READY);
 168         current_snapshot_creation_time = 0;
 169         dss_get_snapshot_list(&sl);
 170         FOR_EACH_SNAPSHOT(s, i, &sl) {
 171                 if (!(s->flags & SS_COMPLETE))
 172                         continue;
 173                 num_complete_snapshots++;
 174                 x += s->completion_time - s->creation_time;
 175         }
 176         assert(x >= 0);
 177         if (num_complete_snapshots)
 178                 x /= num_complete_snapshots; /* avg time to create one snapshot */
 179         x *= wanted; /* time to create all snapshots in interval 0 */
 180         tmp.tv_sec = x;
 181         tmp.tv_usec = 0;
 182         ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
 183         if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
 184                 ret = 1;
 185                 goto out;
 186         }
 187
 188         tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */
 189         diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
 190         diff.tv_usec = 0;
 191         tv_add(&diff, &tmp, &next_snapshot_time);
 192         ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
 193 out:
 194         free_snapshot_list(&sl);
 195         if (ret == 0)
 196                 DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
 197                         tv2ms(&diff) / 1000);
 198         else
 199                 DSS_DEBUG_LOG("next snapshot: now\n");
 200         return ret;
 201 }
 202
 203 static int pre_create_hook(void)
 204 {
 205         int ret, fds[3] = {0, 0, 0};
 206
 207         assert(snapshot_creation_status == HS_READY);
 208         if (!conf.pre_create_hook_given) {
 209                 snapshot_creation_status = HS_PRE_SUCCESS;
 210                 return 0;
 211         }
 212         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 213         ret = dss_exec_cmdline_pid(&create_pid,
 214                 conf.pre_create_hook_arg, fds);
 215         if (ret < 0)
 216                 return ret;
 217         snapshot_creation_status = HS_PRE_RUNNING;
 218         return ret;
 219 }
 220
 221 static int pre_remove_hook(struct snapshot *s, char *why)
 222 {
 223         int ret, fds[3] = {0, 0, 0};
 224         char *cmd;
 225
 226         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 227         assert(snapshot_removal_status == HS_READY);
 228         assert(remove_pid == 0);
 229         assert(!snapshot_currently_being_removed);
 230
 231         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 232         *snapshot_currently_being_removed = *s;
 233         snapshot_currently_being_removed->name = dss_strdup(s->name);
 234
 235         if (!conf.pre_remove_hook_given) {
 236                 snapshot_removal_status = HS_PRE_SUCCESS;
 237                 return 0;
 238         }
 239         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 240                 conf.dest_dir_arg, s->name);
 241         DSS_DEBUG_LOG("executing %s\n", cmd);
 242         ret = dss_exec_cmdline_pid(&remove_pid,
 243                 conf.pre_remove_hook_arg, fds);
 244         free(cmd);
 245         if (ret < 0)
 246                 return ret;
 247         snapshot_removal_status = HS_PRE_RUNNING;
 248         return ret;
 249 }
 250
 251 static int exec_rm(void)
 252 {
 253         struct snapshot *s = snapshot_currently_being_removed;
 254         int fds[3] = {0, 0, 0};
 255         char *new_name = being_deleted_name(s);
 256         char *argv[] = {"rm", "-rf", new_name, NULL};
 257         int ret;
 258
 259         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 260         assert(remove_pid == 0);
 261
 262         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 263         ret = dss_rename(s->name, new_name);
 264         if (ret < 0)
 265                 goto out;
 266         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 267         if (ret < 0)
 268                 goto out;
 269         snapshot_removal_status = HS_RUNNING;
 270 out:
 271         free(new_name);
 272         return ret;
 273 }
 274
 275 static int snapshot_is_being_created(struct snapshot *s)
 276 {
 277         return s->creation_time == current_snapshot_creation_time;
 278 }
 279
 280 static int remove_orphaned_snapshot(struct snapshot_list *sl)
 281 {
 282         struct snapshot *s;
 283         int i, ret;
 284
 285         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 286         FOR_EACH_SNAPSHOT(s, i, sl) {
 287                 if (snapshot_is_being_created(s))
 288                         continue;
 289                 /*
 290                  * We know that no rm is currently running, so if s is marked
 291                  * as being deleted, a previously started rm must have failed.
 292                  */
 293                 if (s->flags & SS_BEING_DELETED)
 294                         goto remove;
 295
 296                 if (s->flags & SS_COMPLETE) /* good snapshot */
 297                         continue;
 298                 /*
 299                  * This snapshot is incomplete and it is not the snapshot
 300                  * currently being created. However, we must not remove it if
 301                  * rsync is about to be restarted. As only the newest snapshot
 302                  * can be restarted, this snapshot is orphaned if it is not the
 303                  * newest snapshot or if we are not about to restart rsync.
 304                  */
 305                 if (get_newest_snapshot(sl) != s)
 306                         goto remove;
 307                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 308                         goto remove;
 309         }
 310         return 0; /* no orphaned snapshots */
 311 remove:
 312         ret = pre_remove_hook(s, "orphaned");
 313         if (ret < 0)
 314                 return ret;
 315         return 1;
 316 }
 317
 318 /*
 319  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 320  */
 321 static int remove_redundant_snapshot(struct snapshot_list *sl)
 322 {
 323         int ret, i, interval;
 324         struct snapshot *s;
 325         unsigned missing = 0;
 326
 327         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 328         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 329                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 330                 unsigned num = sl->interval_count[interval];
 331                 struct snapshot *victim = NULL, *prev = NULL;
 332                 int64_t score = LONG_MAX;
 333
 334                 if (keep >= num)
 335                         missing += keep - num;
 336 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 337 //                      interval, keep, num, missing);
 338                 if (keep + missing >= num)
 339                         continue;
 340                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 341                 FOR_EACH_SNAPSHOT(s, i, sl) {
 342                         int64_t this_score;
 343
 344                         if (snapshot_is_being_created(s))
 345                                 continue;
 346                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 347                         if (s->interval > interval) {
 348                                 prev = s;
 349                                 continue;
 350                         }
 351                         if (s->interval < interval)
 352                                 break;
 353                         if (!victim) {
 354                                 victim = s;
 355                                 prev = s;
 356                                 continue;
 357                         }
 358                         assert(prev);
 359                         /* check if s is a better victim */
 360                         this_score = s->creation_time - prev->creation_time;
 361                         assert(this_score >= 0);
 362                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 363                         if (this_score < score) {
 364                                 score = this_score;
 365                                 victim = s;
 366                         }
 367                         prev = s;
 368                 }
 369                 assert(victim);
 370                 if (conf.dry_run_given) {
 371                         dss_msg("%s would be removed (interval = %i)\n",
 372                                 victim->name, victim->interval);
 373                         continue;
 374                 }
 375                 ret = pre_remove_hook(victim, "redundant");
 376                 return ret < 0? ret : 1;
 377         }
 378         return 0;
 379 }
 380
 381 static int remove_outdated_snapshot(struct snapshot_list *sl)
 382 {
 383         int i, ret;
 384         struct snapshot *s;
 385
 386         DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n",
 387                 conf.num_intervals_arg);
 388         FOR_EACH_SNAPSHOT(s, i, sl) {
 389                 if (snapshot_is_being_created(s))
 390                         continue;
 391                 if (s->interval <= conf.num_intervals_arg)
 392                         continue;
 393                 if (conf.dry_run_given) {
 394                         dss_msg("%s would be removed (interval = %i)\n",
 395                                 s->name, s->interval);
 396                         continue;
 397                 }
 398                 ret = pre_remove_hook(s, "outdated");
 399                 if (ret < 0)
 400                         return ret;
 401                 return 1;
 402         }
 403         return 0;
 404 }
 405
 406 static int remove_oldest_snapshot(struct snapshot_list *sl)
 407 {
 408         int ret;
 409         struct snapshot *s = get_oldest_snapshot(sl);
 410
 411         if (!s) /* no snapshot found */
 412                 return 0;
 413         DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
 414         if (snapshot_is_being_created(s))
 415                 return 0;
 416         ret = pre_remove_hook(s, "oldest");
 417         if (ret < 0)
 418                 return ret;
 419         return 1;
 420 }
 421
 422 static int rename_incomplete_snapshot(int64_t start)
 423 {
 424         char *old_name;
 425         int ret;
 426
 427         free(path_to_last_complete_snapshot);
 428         ret = complete_name(start, get_current_time(),
 429                 &path_to_last_complete_snapshot);
 430         if (ret < 0)
 431                 return ret;
 432         old_name = incomplete_name(start);
 433         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 434         if (ret >= 0)
 435                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 436                         path_to_last_complete_snapshot);
 437         free(old_name);
 438         return ret;
 439 }
 440
 441 static int try_to_free_disk_space(int low_disk_space)
 442 {
 443         int ret;
 444         struct snapshot_list sl;
 445         struct timeval now;
 446
 447         gettimeofday(&now, NULL);
 448         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 449                 return 0;
 450         if (!low_disk_space && conf.keep_redundant_given)
 451                 return 0;
 452         dss_get_snapshot_list(&sl);
 453         ret = remove_outdated_snapshot(&sl);
 454         if (ret) /* error, or we are removing something */
 455                 goto out;
 456         /* no outdated snapshot */
 457         ret = remove_redundant_snapshot(&sl);
 458         if (ret)
 459                 goto out;
 460         ret = 0;
 461         if (!low_disk_space)
 462                 goto out;
 463         ret = remove_orphaned_snapshot(&sl);
 464         if (ret)
 465                 goto out;
 466         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 467         ret = remove_oldest_snapshot(&sl);
 468         if (ret)
 469                 goto out;
 470         DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
 471         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 472 out:
 473         free_snapshot_list(&sl);
 474         return ret;
 475 }
 476
 477 static int post_create_hook(void)
 478 {
 479         int ret, fds[3] = {0, 0, 0};
 480         char *cmd;
 481
 482         if (!conf.post_create_hook_given) {
 483                 snapshot_creation_status = HS_READY;
 484                 return 0;
 485         }
 486         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 487                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 488         DSS_NOTICE_LOG("executing %s\n", cmd);
 489         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 490         free(cmd);
 491         if (ret < 0)
 492                 return ret;
 493         snapshot_creation_status = HS_POST_RUNNING;
 494         return ret;
 495 }
 496
 497 static int post_remove_hook(void)
 498 {
 499         int ret, fds[3] = {0, 0, 0};
 500         char *cmd;
 501         struct snapshot *s = snapshot_currently_being_removed;
 502
 503         assert(s);
 504
 505         if (!conf.post_remove_hook_given) {
 506                 snapshot_removal_status = HS_READY;
 507                 return 0;
 508         }
 509         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 510                 conf.dest_dir_arg, s->name);
 511         DSS_NOTICE_LOG("executing %s\n", cmd);
 512         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 513         free(cmd);
 514         if (ret < 0)
 515                 return ret;
 516         snapshot_removal_status = HS_POST_RUNNING;
 517         return ret;
 518 }
 519
 520 static void kill_process(pid_t pid)
 521 {
 522         if (!pid)
 523                 return;
 524         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 525         kill(pid, SIGTERM);
 526 }
 527
 528 static void stop_create_process(void)
 529 {
 530         if (!create_pid || create_process_stopped)
 531                 return;
 532         kill(SIGSTOP, create_pid);
 533         create_process_stopped = 1;
 534 }
 535
 536 static void restart_create_process(void)
 537 {
 538         if (!create_pid || !create_process_stopped)
 539                 return;
 540         kill (SIGCONT, create_pid);
 541         create_process_stopped = 0;
 542 }
 543
 544 /**
 545  * Print a log message about the exit status of a child.
 546  */
 547 static void log_termination_msg(pid_t pid, int status)
 548 {
 549         if (WIFEXITED(status))
 550                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 551                         WEXITSTATUS(status));
 552         else if (WIFSIGNALED(status))
 553                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 554                         WTERMSIG(status));
 555         else
 556                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 557 }
 558
 559 static int wait_for_process(pid_t pid, int *status)
 560 {
 561         int ret;
 562
 563         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 564         for (;;) {
 565                 fd_set rfds;
 566
 567                 FD_ZERO(&rfds);
 568                 FD_SET(signal_pipe, &rfds);
 569                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 570                 if (ret < 0)
 571                         break;
 572                 ret = next_signal();
 573                 if (!ret)
 574                         continue;
 575                 if (ret == SIGCHLD) {
 576                         ret = waitpid(pid, status, 0);
 577                         if (ret >= 0)
 578                                 break;
 579                         if (errno != EINTR) { /* error */
 580                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 581                                 break;
 582                         }
 583                 }
 584                 /* SIGINT or SIGTERM */
 585                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 586                 kill(pid, SIGTERM);
 587         }
 588         if (ret < 0)
 589                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 590         else
 591                 log_termination_msg(pid, *status);
 592         return ret;
 593 }
 594
 595 static void handle_pre_remove_exit(int status)
 596 {
 597         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 598                 snapshot_removal_status = HS_READY;
 599                 gettimeofday(&next_removal_check, NULL);
 600                 next_removal_check.tv_sec += 60;
 601                 return;
 602         }
 603         snapshot_removal_status = HS_PRE_SUCCESS;
 604 }
 605
 606 static int handle_rm_exit(int status)
 607 {
 608         if (!WIFEXITED(status)) {
 609                 snapshot_removal_status = HS_READY;
 610                 return -E_INVOLUNTARY_EXIT;
 611         }
 612         if (WEXITSTATUS(status)) {
 613                 snapshot_removal_status = HS_READY;
 614                 return -E_BAD_EXIT_CODE;
 615         }
 616         if (conf.post_remove_hook_given)
 617                 snapshot_removal_status = HS_SUCCESS;
 618         else
 619                 snapshot_removal_status = HS_READY;
 620         return 1;
 621 }
 622
 623 static void handle_post_remove_exit(void)
 624 {
 625         snapshot_removal_status = HS_READY;
 626 }
 627
 628 static int handle_remove_exit(int status)
 629 {
 630         int ret;
 631         struct snapshot *s = snapshot_currently_being_removed;
 632
 633         assert(s);
 634         switch (snapshot_removal_status) {
 635         case HS_PRE_RUNNING:
 636                 handle_pre_remove_exit(status);
 637                 ret = 1;
 638                 break;
 639         case HS_RUNNING:
 640                 ret = handle_rm_exit(status);
 641                 break;
 642         case HS_POST_RUNNING:
 643                 handle_post_remove_exit();
 644                 ret = 1;
 645                 break;
 646         default:
 647                 ret = -E_BUG;
 648         }
 649         if (snapshot_removal_status == HS_READY) {
 650                 free(s->name);
 651                 free(s);
 652                 snapshot_currently_being_removed = NULL;
 653         }
 654         remove_pid = 0;
 655         return ret;
 656 }
 657
 658 static int wait_for_remove_process(void)
 659 {
 660         int status, ret;
 661
 662         assert(remove_pid);
 663         assert(
 664                 snapshot_removal_status == HS_PRE_RUNNING ||
 665                 snapshot_removal_status == HS_RUNNING ||
 666                 snapshot_removal_status == HS_POST_RUNNING
 667         );
 668         ret = wait_for_process(remove_pid, &status);
 669         if (ret < 0)
 670                 return ret;
 671         return handle_remove_exit(status);
 672 }
 673
 674 static int handle_rsync_exit(int status)
 675 {
 676         int es, ret;
 677
 678         if (!WIFEXITED(status)) {
 679                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 680                 ret = -E_INVOLUNTARY_EXIT;
 681                 snapshot_creation_status = HS_READY;
 682                 goto out;
 683         }
 684         es = WEXITSTATUS(status);
 685         /*
 686          * Restart rsync on non-fatal errors:
 687          * 12: Error in rsync protocol data stream
 688          * 13: Errors with program diagnostics
 689          */
 690         if (es == 12 || es == 13) {
 691                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 692                         (int)create_pid, es);
 693                 snapshot_creation_status = HS_NEEDS_RESTART;
 694                 gettimeofday(&next_snapshot_time, NULL);
 695                 next_snapshot_time.tv_sec += 60;
 696                 ret = 1;
 697                 goto out;
 698         }
 699         if (es != 0 && es != 23 && es != 24) {
 700                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 701                 ret = -E_BAD_EXIT_CODE;
 702                 snapshot_creation_status = HS_READY;
 703                 goto out;
 704         }
 705         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 706         if (ret < 0)
 707                 goto out;
 708         snapshot_creation_status = HS_SUCCESS;
 709 out:
 710         create_pid = 0;
 711         create_process_stopped = 0;
 712         return ret;
 713 }
 714
 715 static int handle_pre_create_hook_exit(int status)
 716 {
 717         int es, ret;
 718         static int warn_count;
 719
 720         if (!WIFEXITED(status)) {
 721                 snapshot_creation_status = HS_READY;
 722                 ret = -E_INVOLUNTARY_EXIT;
 723                 goto out;
 724         }
 725         es = WEXITSTATUS(status);
 726         if (es) {
 727                 if (!warn_count--) {
 728                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 729                                 conf.pre_create_hook_arg, es);
 730                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 731                         warn_count = 60; /* warn only once per hour */
 732                 }
 733                 snapshot_creation_status = HS_READY;
 734                 ret = 0;
 735                 goto out;
 736         }
 737         warn_count = 0;
 738         snapshot_creation_status = HS_PRE_SUCCESS;
 739         ret = 1;
 740 out:
 741         create_pid = 0;
 742         return ret;
 743 }
 744
 745 static int handle_sigchld(void)
 746 {
 747         pid_t pid;
 748         int status, ret = reap_child(&pid, &status);
 749
 750         if (ret <= 0)
 751                 return ret;
 752
 753         if (pid == create_pid) {
 754                 switch (snapshot_creation_status) {
 755                 case HS_PRE_RUNNING:
 756                         return handle_pre_create_hook_exit(status);
 757                 case HS_RUNNING:
 758                         return handle_rsync_exit(status);
 759                 case HS_POST_RUNNING:
 760                         snapshot_creation_status = HS_READY;
 761                         return 1;
 762                 default:
 763                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 764                                 snapshot_creation_status);
 765                         return -E_BUG;
 766                 }
 767         }
 768         if (pid == remove_pid) {
 769                 ret = handle_remove_exit(status);
 770                 if (ret < 0)
 771                         return ret;
 772                 return ret;
 773         }
 774         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 775         return -E_BUG;
 776 }
 777
 778 static int check_config(void)
 779 {
 780         if (conf.unit_interval_arg <= 0) {
 781                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 782                 return -E_INVALID_NUMBER;
 783         }
 784         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 785         if (conf.num_intervals_arg <= 0) {
 786                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 787                 return -E_INVALID_NUMBER;
 788         }
 789         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 790         return 1;
 791 }
 792
 793 static int parse_config_file(int override)
 794 {
 795         int ret;
 796         char *config_file;
 797         struct stat statbuf;
 798         char *old_logfile_arg = NULL;
 799         int old_daemon_given = 0;
 800
 801         if (conf.config_file_given)
 802                 config_file = dss_strdup(conf.config_file_arg);
 803         else {
 804                 char *home = get_homedir();
 805                 config_file = make_message("%s/.dssrc", home);
 806                 free(home);
 807         }
 808         if (override) { /* SIGHUP */
 809                 if (conf.logfile_given)
 810                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 811                 old_daemon_given = conf.daemon_given;
 812         }
 813
 814         ret = stat(config_file, &statbuf);
 815         if (ret && conf.config_file_given) {
 816                 ret = -ERRNO_TO_DSS_ERROR(errno);
 817                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 818                 goto out;
 819         }
 820         if (!ret) {
 821                 struct cmdline_parser_params params = {
 822                         .override = override,
 823                         .initialize = 0,
 824                         .check_required = 1,
 825                         .check_ambiguity = 0,
 826                         .print_errors = 1
 827                 };
 828                 if (override) { /* invalidate all rsync options */
 829                         int i;
 830
 831                         for (i = 0; i < conf.rsync_option_given; i++) {
 832                                 free(conf.rsync_option_arg[i]);
 833                                 conf.rsync_option_arg[i] = NULL;
 834                         }
 835                         conf.rsync_option_given = 0;
 836                 }
 837                 cmdline_parser_config_file(config_file, &conf, &params);
 838         }
 839         ret = check_config();
 840         if (ret < 0)
 841                 goto out;
 842         if (override) {
 843                 /* don't change daemon mode on SIGHUP */
 844                 conf.daemon_given = old_daemon_given;
 845                 close_log(logfile);
 846                 logfile = NULL;
 847                 if (conf.logfile_given)
 848                         free(old_logfile_arg);
 849                 else if (conf.daemon_given) { /* re-use old logfile */
 850                         conf.logfile_arg = old_logfile_arg;
 851                         conf.logfile_given = 1;
 852                 }
 853         }
 854         if (conf.logfile_given) {
 855                 logfile = open_log(conf.logfile_arg);
 856                 log_welcome(conf.loglevel_arg);
 857         }
 858         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 859 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 860 out:
 861         free(config_file);
 862         if (ret < 0)
 863                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 864         return ret;
 865 }
 866
 867 static int change_to_dest_dir(void)
 868 {
 869         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 870         return dss_chdir(conf.dest_dir_arg);
 871 }
 872
 873 static int handle_sighup(void)
 874 {
 875         int ret;
 876
 877         DSS_NOTICE_LOG("SIGHUP\n");
 878         ret = parse_config_file(1);
 879         if (ret < 0)
 880                 return ret;
 881         return change_to_dest_dir();
 882 }
 883
 884 static int handle_signal(void)
 885 {
 886         int sig, ret = next_signal();
 887
 888         if (ret <= 0)
 889                 goto out;
 890         sig = ret;
 891         switch (sig) {
 892         case SIGINT:
 893         case SIGTERM:
 894                 restart_create_process();
 895                 kill_process(create_pid);
 896                 kill_process(remove_pid);
 897                 ret = -E_SIGNAL;
 898                 break;
 899         case SIGHUP:
 900                 ret = handle_sighup();
 901                 break;
 902         case SIGCHLD:
 903                 ret = handle_sigchld();
 904                 break;
 905         }
 906 out:
 907         if (ret < 0)
 908                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 909         return ret;
 910 }
 911
 912 /*
 913  * We can not use rsync locally if the local user is different from the remote
 914  * user or if the src dir is not on the local host (or both).
 915  */
 916 static int use_rsync_locally(char *logname)
 917 {
 918         char *h = conf.remote_host_arg;
 919
 920         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 921                 return 0;
 922         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 923                 return 0;
 924         return 1;
 925 }
 926
 927 static void create_rsync_argv(char ***argv, int64_t *num)
 928 {
 929         char *logname, *newest;
 930         int i = 0, j;
 931         struct snapshot_list sl;
 932
 933         dss_get_snapshot_list(&sl);
 934         newest = name_of_newest_complete_snapshot(&sl);
 935         free_snapshot_list(&sl);
 936
 937         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 938         (*argv)[i++] = dss_strdup("rsync");
 939         (*argv)[i++] = dss_strdup("-aq");
 940         (*argv)[i++] = dss_strdup("--delete");
 941         for (j = 0; j < conf.rsync_option_given; j++)
 942                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 943         if (newest) {
 944                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 945                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 946                 free(newest);
 947         } else
 948                 DSS_INFO_LOG("no previous snapshot found\n");
 949         logname = dss_logname();
 950         if (use_rsync_locally(logname))
 951                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 952         else
 953                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 954                         conf.remote_user_arg : logname,
 955                         conf.remote_host_arg, conf.source_dir_arg);
 956         free(logname);
 957         *num = get_current_time();
 958         (*argv)[i++] = incomplete_name(*num);
 959         (*argv)[i++] = NULL;
 960         for (j = 0; j < i; j++)
 961                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 962 }
 963
 964 static void free_rsync_argv(char **argv)
 965 {
 966         int i;
 967
 968         if (!argv)
 969                 return;
 970         for (i = 0; argv[i]; i++)
 971                 free(argv[i]);
 972         free(argv);
 973 }
 974
 975 static int create_snapshot(char **argv)
 976 {
 977         int ret, fds[3] = {0, 0, 0};
 978         char *name;
 979
 980         name = incomplete_name(current_snapshot_creation_time);
 981         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 982         free(name);
 983         ret = dss_exec(&create_pid, argv[0], argv, fds);
 984         if (ret < 0)
 985                 return ret;
 986         snapshot_creation_status = HS_RUNNING;
 987         return ret;
 988 }
 989
 990 static int select_loop(void)
 991 {
 992         int ret;
 993         /* check every 60 seconds for free disk space */
 994         struct timeval tv;
 995         char **rsync_argv = NULL;
 996
 997         for (;;) {
 998                 fd_set rfds;
 999                 int low_disk_space;
1000                 struct timeval *tvp;
1001
1002                 if (remove_pid)
1003                         tvp = NULL; /* sleep until rm hook/process dies */
1004                 else { /* sleep one minute */
1005                         tv.tv_sec = 60;
1006                         tv.tv_usec = 0;
1007                         tvp = &tv;
1008                 }
1009                 FD_ZERO(&rfds);
1010                 FD_SET(signal_pipe, &rfds);
1011                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1012                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1013                 if (ret < 0)
1014                         goto out;
1015                 if (FD_ISSET(signal_pipe, &rfds)) {
1016                         ret = handle_signal();
1017                         if (ret < 0)
1018                                 goto out;
1019                 }
1020                 if (remove_pid)
1021                         continue;
1022                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1023                         ret = exec_rm();
1024                         if (ret < 0)
1025                                 goto out;
1026                         continue;
1027                 }
1028                 if (snapshot_removal_status == HS_SUCCESS) {
1029                         ret = post_remove_hook();
1030                         if (ret < 0)
1031                                 goto out;
1032                         continue;
1033                 }
1034                 ret = disk_space_low();
1035                 if (ret < 0)
1036                         goto out;
1037                 low_disk_space = ret;
1038                 ret = try_to_free_disk_space(low_disk_space);
1039                 if (ret < 0)
1040                         goto out;
1041                 if (snapshot_removal_status != HS_READY) {
1042                         stop_create_process();
1043                         continue;
1044                 }
1045                 restart_create_process();
1046                 switch (snapshot_creation_status) {
1047                 case HS_READY:
1048                         if (!next_snapshot_is_due())
1049                                 continue;
1050                         ret = pre_create_hook();
1051                         if (ret < 0)
1052                                 goto out;
1053                         continue;
1054                 case HS_PRE_RUNNING:
1055                 case HS_RUNNING:
1056                 case HS_POST_RUNNING:
1057                         continue;
1058                 case HS_PRE_SUCCESS:
1059                         free_rsync_argv(rsync_argv);
1060                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1061                         /* fall through */
1062                 case HS_NEEDS_RESTART:
1063                         ret = create_snapshot(rsync_argv);
1064                         if (ret < 0)
1065                                 goto out;
1066                         continue;
1067                 case HS_SUCCESS:
1068                         ret = post_create_hook();
1069                         if (ret < 0)
1070                                 goto out;
1071                         continue;
1072                 }
1073         }
1074 out:
1075         return ret;
1076 }
1077
1078 static void exit_hook(int exit_code)
1079 {
1080         int fds[3] = {0, 0, 0};
1081         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1082         pid_t pid;
1083
1084         if (!conf.exit_hook_given)
1085                 return;
1086         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1087         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1088 }
1089
1090 static int com_run(void)
1091 {
1092         int ret;
1093
1094         if (conf.dry_run_given) {
1095                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1096                 return -E_SYNTAX;
1097         }
1098         ret = install_sighandler(SIGHUP);
1099         if (ret < 0)
1100                 return ret;
1101         ret = select_loop();
1102         if (ret >= 0) /* impossible */
1103                 ret = -E_BUG;
1104         exit_hook(ret);
1105         return ret;
1106 }
1107
1108 static int com_prune(void)
1109 {
1110         int ret;
1111         struct snapshot_list sl;
1112         struct disk_space ds;
1113
1114         ret = get_disk_space(".", &ds);
1115         if (ret < 0)
1116                 return ret;
1117         log_disk_space(&ds);
1118         dss_get_snapshot_list(&sl);
1119         ret = remove_outdated_snapshot(&sl);
1120         if (ret < 0)
1121                 goto out;
1122         if (ret > 0)
1123                 goto rm;
1124         ret = remove_redundant_snapshot(&sl);
1125         if (ret < 0)
1126                 goto out;
1127         if (ret > 0)
1128                 goto rm;
1129         ret = 0;
1130         goto out;
1131 rm:
1132         if (snapshot_removal_status == HS_PRE_RUNNING) {
1133                 ret = wait_for_remove_process();
1134                 if (ret < 0)
1135                         goto out;
1136                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1137                         goto out;
1138         }
1139         ret = exec_rm();
1140         if (ret < 0)
1141                 goto out;
1142         ret = wait_for_remove_process();
1143         if (ret < 0)
1144                 goto out;
1145         if (snapshot_removal_status != HS_SUCCESS)
1146                 goto out;
1147         ret = post_remove_hook();
1148         if (ret < 0)
1149                 goto out;
1150         if (snapshot_removal_status != HS_POST_RUNNING)
1151                 goto out;
1152         ret = wait_for_remove_process();
1153         if (ret < 0)
1154                 goto out;
1155         ret = 1;
1156 out:
1157         free_snapshot_list(&sl);
1158         return ret;
1159 }
1160
1161 static int com_create(void)
1162 {
1163         int ret, status;
1164         char **rsync_argv;
1165
1166         if (conf.dry_run_given) {
1167                 int i;
1168                 char *msg = NULL;
1169                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1170                 for (i = 0; rsync_argv[i]; i++) {
1171                         char *tmp = msg;
1172                         msg = make_message("%s%s%s", tmp? tmp : "",
1173                                 tmp? " " : "", rsync_argv[i]);
1174                         free(tmp);
1175                 }
1176                 free_rsync_argv(rsync_argv);
1177                 dss_msg("%s\n", msg);
1178                 free(msg);
1179                 return 1;
1180         }
1181         ret = pre_create_hook();
1182         if (ret < 0)
1183                 return ret;
1184         if (create_pid) {
1185                 ret = wait_for_process(create_pid, &status);
1186                 if (ret < 0)
1187                         return ret;
1188                 ret = handle_pre_create_hook_exit(status);
1189                 if (ret <= 0) /* error, or pre-create failed */
1190                         return ret;
1191         }
1192         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1193         ret = create_snapshot(rsync_argv);
1194         if (ret < 0)
1195                 goto out;
1196         ret = wait_for_process(create_pid, &status);
1197         if (ret < 0)
1198                 goto out;
1199         ret = handle_rsync_exit(status);
1200         if (ret < 0)
1201                 goto out;
1202         post_create_hook();
1203         if (create_pid)
1204                 ret = wait_for_process(create_pid, &status);
1205 out:
1206         free_rsync_argv(rsync_argv);
1207         return ret;
1208 }
1209
1210 static int com_ls(void)
1211 {
1212         int i;
1213         struct snapshot_list sl;
1214         struct snapshot *s;
1215
1216         dss_get_snapshot_list(&sl);
1217         FOR_EACH_SNAPSHOT(s, i, &sl) {
1218                 int64_t d = 0;
1219                 if (s->flags & SS_COMPLETE)
1220                         d = (s->completion_time - s->creation_time) / 60;
1221                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1222         };
1223         free_snapshot_list(&sl);
1224         return 1;
1225 }
1226
1227 static int setup_signal_handling(void)
1228 {
1229         int ret;
1230
1231         DSS_INFO_LOG("setting up signal handlers\n");
1232         signal_pipe = signal_init(); /* always successful */
1233         ret = install_sighandler(SIGINT);
1234         if (ret < 0)
1235                 return ret;
1236         ret = install_sighandler(SIGTERM);
1237         if (ret < 0)
1238                 return ret;
1239         return install_sighandler(SIGCHLD);
1240 }
1241
1242 /**
1243  * The main function of dss.
1244  *
1245  * \param argc Usual argument count.
1246  * \param argv Usual argument vector.
1247  */
1248 int main(int argc, char **argv)
1249 {
1250         int ret;
1251         struct cmdline_parser_params params = {
1252                 .override = 0,
1253                 .initialize = 1,
1254                 .check_required = 0,
1255                 .check_ambiguity = 0,
1256                 .print_errors = 1
1257         };
1258
1259         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1260         ret = parse_config_file(0);
1261         if (ret < 0)
1262                 goto out;
1263         if (conf.daemon_given)
1264                 daemon_init();
1265         ret = change_to_dest_dir();
1266         if (ret < 0)
1267                 goto out;
1268         ret = setup_signal_handling();
1269         if (ret < 0)
1270                 goto out;
1271         ret = call_command_handler();
1272 out:
1273         if (ret < 0)
1274                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1275         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1276 }