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_DEBUG_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         static int warn_count;
 536
 537         if (!WIFEXITED(status)) {
 538                 snapshot_creation_status = SCS_READY;
 539                 compute_next_snapshot_time();
 540                 ret = -E_INVOLUNTARY_EXIT;
 541                 goto out;
 542         }
 543         es = WEXITSTATUS(status);
 544         if (es) {
 545                 if (!warn_count--) {
 546                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 547                                 conf.pre_create_hook_arg, es);
 548                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 549                         warn_count = 60; /* warn only once per hour */
 550                 }
 551                 snapshot_creation_status = SCS_READY;
 552                 compute_next_snapshot_time();
 553                 ret = 0;
 554                 goto out;
 555         }
 556         warn_count = 0;
 557         snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
 558         ret = 1;
 559 out:
 560         create_pid = 0;
 561         return ret;
 562 }
 563
 564 static int handle_sigchld(void)
 565 {
 566         pid_t pid;
 567         int status, ret = reap_child(&pid, &status);
 568
 569         if (ret <= 0)
 570                 return ret;
 571
 572         if (pid == create_pid) {
 573                 switch (snapshot_creation_status) {
 574                 case SCS_PRE_HOOK_RUNNING:
 575                         return handle_pre_create_hook_exit(status);
 576                 case SCS_RSYNC_RUNNING:
 577                         return handle_rsync_exit(status);
 578                 case SCS_POST_HOOK_RUNNING:
 579                         snapshot_creation_status = SCS_READY;
 580                         compute_next_snapshot_time();
 581                         return 1;
 582                 default:
 583                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 584                                 snapshot_creation_status);
 585                         return -E_BUG;
 586                 }
 587         }
 588         if (pid == rm_pid)
 589                 return handle_rm_exit(status);
 590         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 591         return -E_BUG;
 592 }
 593
 594 static int check_config(void)
 595 {
 596         if (conf.unit_interval_arg <= 0) {
 597                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 598                 return -E_INVALID_NUMBER;
 599         }
 600         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 601         if (conf.num_intervals_arg <= 0) {
 602                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 603                 return -E_INVALID_NUMBER;
 604         }
 605         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 606         return 1;
 607 }
 608
 609 static int parse_config_file(int override)
 610 {
 611         int ret;
 612         char *config_file;
 613         struct stat statbuf;
 614         char *old_logfile_arg = NULL;
 615         int old_daemon_given = 0;
 616
 617         if (conf.config_file_given)
 618                 config_file = dss_strdup(conf.config_file_arg);
 619         else {
 620                 char *home = get_homedir();
 621                 config_file = make_message("%s/.dssrc", home);
 622                 free(home);
 623         }
 624         if (override) { /* SIGHUP */
 625                 if (conf.logfile_given)
 626                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 627                 old_daemon_given = conf.daemon_given;
 628         }
 629
 630         ret = stat(config_file, &statbuf);
 631         if (ret && conf.config_file_given) {
 632                 ret = -ERRNO_TO_DSS_ERROR(errno);
 633                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 634                 goto out;
 635         }
 636         if (!ret) {
 637                 struct cmdline_parser_params params = {
 638                         .override = override,
 639                         .initialize = 0,
 640                         .check_required = 1,
 641                         .check_ambiguity = 0,
 642                         .print_errors = 1
 643                 };
 644                 if (override) { /* invalidate all rsync options */
 645                         int i;
 646
 647                         for (i = 0; i < conf.rsync_option_given; i++) {
 648                                 free(conf.rsync_option_arg[i]);
 649                                 conf.rsync_option_arg[i] = NULL;
 650                         }
 651                         conf.rsync_option_given = 0;
 652                 }
 653                 cmdline_parser_config_file(config_file, &conf, &params);
 654         }
 655         ret = check_config();
 656         if (ret < 0)
 657                 goto out;
 658         if (override) {
 659                 /* don't change daemon mode on SIGHUP */
 660                 conf.daemon_given = old_daemon_given;
 661                 close_log(logfile);
 662                 logfile = NULL;
 663                 if (conf.logfile_given)
 664                         free(old_logfile_arg);
 665                 else if (conf.daemon_given) { /* re-use old logfile */
 666                         conf.logfile_arg = old_logfile_arg;
 667                         conf.logfile_given = 1;
 668                 }
 669         }
 670         if (conf.logfile_given) {
 671                 logfile = open_log(conf.logfile_arg);
 672                 log_welcome(conf.loglevel_arg);
 673         }
 674         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 675 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 676 out:
 677         free(config_file);
 678         if (ret < 0)
 679                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 680         return ret;
 681 }
 682
 683 static int change_to_dest_dir(void)
 684 {
 685         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 686         return dss_chdir(conf.dest_dir_arg);
 687 }
 688
 689 static int handle_sighup(void)
 690 {
 691         int ret;
 692
 693         DSS_NOTICE_LOG("SIGHUP\n");
 694         ret = parse_config_file(1);
 695         if (ret < 0)
 696                 return ret;
 697         return change_to_dest_dir();
 698 }
 699
 700 static int handle_signal(void)
 701 {
 702         int sig, ret = next_signal();
 703
 704         if (ret <= 0)
 705                 goto out;
 706         sig = ret;
 707         switch (sig) {
 708         case SIGINT:
 709         case SIGTERM:
 710                 restart_create_process();
 711                 kill_process(create_pid);
 712                 kill_process(rm_pid);
 713                 ret = -E_SIGNAL;
 714                 break;
 715         case SIGHUP:
 716                 ret = handle_sighup();
 717                 break;
 718         case SIGCHLD:
 719                 ret = handle_sigchld();
 720                 break;
 721         }
 722 out:
 723         if (ret < 0)
 724                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 725         return ret;
 726 }
 727
 728 /*
 729  * We can not use rsync locally if the local user is different from the remote
 730  * user or if the src dir is not on the local host (or both).
 731  */
 732 static int use_rsync_locally(char *logname)
 733 {
 734         char *h = conf.remote_host_arg;
 735
 736         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 737                 return 0;
 738         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 739                 return 0;
 740         return 1;
 741 }
 742
 743 static void create_rsync_argv(char ***argv, int64_t *num)
 744 {
 745         char *logname, *newest;
 746         int i = 0, j;
 747         struct snapshot_list sl;
 748
 749         dss_get_snapshot_list(&sl);
 750         newest = name_of_newest_complete_snapshot(&sl);
 751         free_snapshot_list(&sl);
 752
 753         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 754         (*argv)[i++] = dss_strdup("rsync");
 755         (*argv)[i++] = dss_strdup("-aq");
 756         (*argv)[i++] = dss_strdup("--delete");
 757         for (j = 0; j < conf.rsync_option_given; j++)
 758                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 759         if (newest) {
 760                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 761                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 762                 free(newest);
 763         } else
 764                 DSS_INFO_LOG("no previous snapshot found\n");
 765         logname = dss_logname();
 766         if (use_rsync_locally(logname))
 767                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 768         else
 769                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 770                         conf.remote_user_arg : logname,
 771                         conf.remote_host_arg, conf.source_dir_arg);
 772         free(logname);
 773         *num = get_current_time();
 774         (*argv)[i++] = incomplete_name(*num);
 775         (*argv)[i++] = NULL;
 776         for (j = 0; j < i; j++)
 777                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 778 }
 779
 780 static void free_rsync_argv(char **argv)
 781 {
 782         int i;
 783
 784         if (!argv)
 785                 return;
 786         for (i = 0; argv[i]; i++)
 787                 free(argv[i]);
 788         free(argv);
 789 }
 790
 791 static int create_snapshot(char **argv)
 792 {
 793         int ret, fds[3] = {0, 0, 0};
 794         char *name;
 795
 796         name = incomplete_name(current_snapshot_creation_time);
 797         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 798         free(name);
 799         ret = dss_exec(&create_pid, argv[0], argv, fds);
 800         if (ret < 0)
 801                 return ret;
 802         snapshot_creation_status = SCS_RSYNC_RUNNING;
 803         return ret;
 804 }
 805
 806 static int select_loop(void)
 807 {
 808         int ret;
 809         /* check every 60 seconds for free disk space */
 810         struct timeval tv;
 811         char **rsync_argv = NULL;
 812
 813         for (;;) {
 814                 fd_set rfds;
 815                 int low_disk_space;
 816                 struct timeval now, *tvp;
 817
 818                 if (rm_pid)
 819                         tvp = NULL; /* sleep until rm process dies */
 820                 else { /* sleep one minute */
 821                         tv.tv_sec = 60;
 822                         tv.tv_usec = 0;
 823                         tvp = &tv;
 824                 }
 825                 FD_ZERO(&rfds);
 826                 FD_SET(signal_pipe, &rfds);
 827                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
 828                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
 829                 if (ret < 0)
 830                         goto out;
 831                 if (FD_ISSET(signal_pipe, &rfds)) {
 832                         ret = handle_signal();
 833                         if (ret < 0)
 834                                 goto out;
 835                 }
 836                 if (rm_pid)
 837                         continue;
 838                 ret = disk_space_low();
 839                 if (ret < 0)
 840                         goto out;
 841                 low_disk_space = ret;
 842                 ret = try_to_free_disk_space(low_disk_space);
 843                 if (ret < 0)
 844                         goto out;
 845                 if (rm_pid) {
 846                         stop_create_process();
 847                         continue;
 848                 }
 849                 restart_create_process();
 850                 gettimeofday(&now, NULL);
 851                 if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 852                         continue;
 853                 switch (snapshot_creation_status) {
 854                 case SCS_READY:
 855                         ret = pre_create_hook();
 856                         if (ret < 0)
 857                                 goto out;
 858                         continue;
 859                 case SCS_PRE_HOOK_RUNNING:
 860                         continue;
 861                 case SCS_RSYNC_NEEDS_RESTART:
 862                         ret = create_snapshot(rsync_argv);
 863                         if (ret < 0)
 864                                 goto out;
 865                         continue;
 866                 case SCS_PRE_HOOK_SUCCESS:
 867                         free_rsync_argv(rsync_argv);
 868                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
 869                         ret = create_snapshot(rsync_argv);
 870                         if (ret < 0)
 871                                 goto out;
 872                         continue;
 873                 case SCS_RSYNC_RUNNING:
 874                         continue;
 875                 case SCS_RSYNC_SUCCESS:
 876                         ret = post_create_hook();
 877                         if (ret < 0)
 878                                 goto out;
 879                         continue;
 880                 case SCS_POST_HOOK_RUNNING:
 881                         continue;
 882                 }
 883         }
 884 out:
 885         return ret;
 886 }
 887
 888 static void exit_hook(int exit_code)
 889 {
 890         int fds[3] = {0, 0, 0};
 891         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
 892         pid_t pid;
 893
 894         if (!conf.exit_hook_given)
 895                 return;
 896         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
 897         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
 898 }
 899
 900 static int com_run(void)
 901 {
 902         int ret;
 903
 904         if (conf.dry_run_given) {
 905                 DSS_ERROR_LOG("dry_run not supported by this command\n");
 906                 return -E_SYNTAX;
 907         }
 908         ret = install_sighandler(SIGHUP);
 909         if (ret < 0)
 910                 return ret;
 911         compute_next_snapshot_time();
 912         ret = select_loop();
 913         if (ret >= 0) /* impossible */
 914                 ret = -E_BUG;
 915         exit_hook(ret);
 916         return ret;
 917 }
 918
 919 static int com_prune(void)
 920 {
 921         int ret;
 922         struct snapshot_list sl;
 923         struct disk_space ds;
 924
 925         ret = get_disk_space(".", &ds);
 926         if (ret < 0)
 927                 return ret;
 928         log_disk_space(&ds);
 929         for (;;) {
 930                 dss_get_snapshot_list(&sl);
 931                 ret = remove_outdated_snapshot(&sl);
 932                 free_snapshot_list(&sl);
 933                 if (ret < 0)
 934                         return ret;
 935                 if (!ret)
 936                         break;
 937                 ret = wait_for_rm_process();
 938                 if (ret < 0)
 939                         goto out;
 940         }
 941         for (;;) {
 942                 dss_get_snapshot_list(&sl);
 943                 ret = remove_redundant_snapshot(&sl);
 944                 free_snapshot_list(&sl);
 945                 if (ret < 0)
 946                         return ret;
 947                 if (!ret)
 948                         break;
 949                 ret = wait_for_rm_process();
 950                 if (ret < 0)
 951                         goto out;
 952         }
 953         return 1;
 954 out:
 955         return ret;
 956 }
 957
 958 static int com_create(void)
 959 {
 960         int ret, status;
 961         char **rsync_argv;
 962
 963         if (conf.dry_run_given) {
 964                 int i;
 965                 char *msg = NULL;
 966                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
 967                 for (i = 0; rsync_argv[i]; i++) {
 968                         char *tmp = msg;
 969                         msg = make_message("%s%s%s", tmp? tmp : "",
 970                                 tmp? " " : "", rsync_argv[i]);
 971                         free(tmp);
 972                 }
 973                 free_rsync_argv(rsync_argv);
 974                 dss_msg("%s\n", msg);
 975                 free(msg);
 976                 return 1;
 977         }
 978         ret = pre_create_hook();
 979         if (ret < 0)
 980                 return ret;
 981         if (create_pid) {
 982                 ret = wait_for_process(create_pid, &status);
 983                 if (ret < 0)
 984                         return ret;
 985                 ret = handle_pre_create_hook_exit(status);
 986                 if (ret < 0)
 987                         return ret;
 988         }
 989         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
 990         ret = create_snapshot(rsync_argv);
 991         if (ret < 0)
 992                 goto out;
 993         ret = wait_for_process(create_pid, &status);
 994         if (ret < 0)
 995                 goto out;
 996         ret = handle_rsync_exit(status);
 997         if (ret < 0)
 998                 goto out;
 999         post_create_hook();
1000         if (create_pid)
1001                 ret = wait_for_process(create_pid, &status);
1002 out:
1003         free_rsync_argv(rsync_argv);
1004         return ret;
1005 }
1006
1007 static int com_ls(void)
1008 {
1009         int i;
1010         struct snapshot_list sl;
1011         struct snapshot *s;
1012
1013         dss_get_snapshot_list(&sl);
1014         FOR_EACH_SNAPSHOT(s, i, &sl) {
1015                 int64_t d = 0;
1016                 if (s->flags & SS_COMPLETE)
1017                         d = (s->completion_time - s->creation_time) / 60;
1018                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1019         };
1020         free_snapshot_list(&sl);
1021         return 1;
1022 }
1023
1024 static int setup_signal_handling(void)
1025 {
1026         int ret;
1027
1028         DSS_INFO_LOG("setting up signal handlers\n");
1029         signal_pipe = signal_init(); /* always successful */
1030         ret = install_sighandler(SIGINT);
1031         if (ret < 0)
1032                 return ret;
1033         ret = install_sighandler(SIGTERM);
1034         if (ret < 0)
1035                 return ret;
1036         return install_sighandler(SIGCHLD);
1037 }
1038
1039 /**
1040  * The main function of dss.
1041  *
1042  * \param argc Usual argument count.
1043  * \param argv Usual argument vector.
1044  */
1045 int main(int argc, char **argv)
1046 {
1047         int ret;
1048         struct cmdline_parser_params params = {
1049                 .override = 0,
1050                 .initialize = 1,
1051                 .check_required = 0,
1052                 .check_ambiguity = 0,
1053                 .print_errors = 1
1054         };
1055
1056         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1057         ret = parse_config_file(0);
1058         if (ret < 0)
1059                 goto out;
1060         if (conf.daemon_given)
1061                 daemon_init();
1062         ret = change_to_dest_dir();
1063         if (ret < 0)
1064                 goto out;
1065         ret = setup_signal_handling();
1066         if (ret < 0)
1067                 goto out;
1068         ret = call_command_handler();
1069 out:
1070         if (ret < 0)
1071                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1072         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1073 }