dss.c

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