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