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