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         int ret;
 371         struct snapshot *s = get_oldest_snapshot(sl);
 372
 373         if (!s) /* no snapshot found */
 374                 return 0;
 375         DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
 376         if (snapshot_is_being_created(s))
 377                 return 0;
 378         ret = pre_remove_hook(s, "oldest");
 379         if (ret < 0)
 380                 return ret;
 381         return 1;
 382 }
 383
 384 static int rename_incomplete_snapshot(int64_t start)
 385 {
 386         char *old_name;
 387         int ret;
 388
 389         free(path_to_last_complete_snapshot);
 390         ret = complete_name(start, get_current_time(),
 391                 &path_to_last_complete_snapshot);
 392         if (ret < 0)
 393                 return ret;
 394         old_name = incomplete_name(start);
 395         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 396         if (ret >= 0)
 397                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 398                         path_to_last_complete_snapshot);
 399         free(old_name);
 400         return ret;
 401 }
 402
 403 static int try_to_free_disk_space(int low_disk_space)
 404 {
 405         int ret;
 406         struct snapshot_list sl;
 407         struct timeval now;
 408
 409         gettimeofday(&now, NULL);
 410         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 411                 return 0;
 412         if (!low_disk_space && conf.keep_redundant_given)
 413                 return 0;
 414         dss_get_snapshot_list(&sl);
 415         ret = remove_outdated_snapshot(&sl);
 416         if (ret) /* error, or we are removing something */
 417                 goto out;
 418         /* no outdated snapshot */
 419         ret = remove_redundant_snapshot(&sl);
 420         if (ret)
 421                 goto out;
 422         ret = 0;
 423         if (!low_disk_space)
 424                 goto out;
 425         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 426         ret = remove_oldest_snapshot(&sl);
 427         if (ret)
 428                 goto out;
 429         DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
 430         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 431 out:
 432         free_snapshot_list(&sl);
 433         return ret;
 434 }
 435
 436 static int post_create_hook(void)
 437 {
 438         int ret, fds[3] = {0, 0, 0};
 439         char *cmd;
 440
 441         if (!conf.post_create_hook_given) {
 442                 snapshot_creation_status = HS_READY;
 443                 return 0;
 444         }
 445         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 446                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 447         DSS_NOTICE_LOG("executing %s\n", cmd);
 448         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 449         free(cmd);
 450         if (ret < 0)
 451                 return ret;
 452         snapshot_creation_status = HS_POST_RUNNING;
 453         return ret;
 454 }
 455
 456 static int post_remove_hook(void)
 457 {
 458         int ret, fds[3] = {0, 0, 0};
 459         char *cmd;
 460         struct snapshot *s = snapshot_currently_being_removed;
 461
 462         assert(s);
 463
 464         if (!conf.post_remove_hook_given) {
 465                 snapshot_removal_status = HS_READY;
 466                 return 0;
 467         }
 468         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 469                 conf.dest_dir_arg, s->name);
 470         DSS_NOTICE_LOG("executing %s\n", cmd);
 471         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 472         free(cmd);
 473         if (ret < 0)
 474                 return ret;
 475         snapshot_removal_status = HS_POST_RUNNING;
 476         return ret;
 477 }
 478
 479 static void kill_process(pid_t pid)
 480 {
 481         if (!pid)
 482                 return;
 483         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 484         kill(pid, SIGTERM);
 485 }
 486
 487 static void stop_create_process(void)
 488 {
 489         if (!create_pid || create_process_stopped)
 490                 return;
 491         kill(SIGSTOP, create_pid);
 492         create_process_stopped = 1;
 493 }
 494
 495 static void restart_create_process(void)
 496 {
 497         if (!create_pid || !create_process_stopped)
 498                 return;
 499         kill (SIGCONT, create_pid);
 500         create_process_stopped = 0;
 501 }
 502
 503 /**
 504  * Print a log message about the exit status of a child.
 505  */
 506 static void log_termination_msg(pid_t pid, int status)
 507 {
 508         if (WIFEXITED(status))
 509                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 510                         WEXITSTATUS(status));
 511         else if (WIFSIGNALED(status))
 512                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 513                         WTERMSIG(status));
 514         else
 515                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 516 }
 517
 518 static int wait_for_process(pid_t pid, int *status)
 519 {
 520         int ret;
 521
 522         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 523         for (;;) {
 524                 fd_set rfds;
 525
 526                 FD_ZERO(&rfds);
 527                 FD_SET(signal_pipe, &rfds);
 528                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 529                 if (ret < 0)
 530                         break;
 531                 ret = next_signal();
 532                 if (!ret)
 533                         continue;
 534                 if (ret == SIGCHLD) {
 535                         ret = waitpid(pid, status, 0);
 536                         if (ret >= 0)
 537                                 break;
 538                         if (errno != EINTR) { /* error */
 539                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 540                                 break;
 541                         }
 542                 }
 543                 /* SIGINT or SIGTERM */
 544                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 545                 kill(pid, SIGTERM);
 546         }
 547         if (ret < 0)
 548                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 549         else
 550                 log_termination_msg(pid, *status);
 551         return ret;
 552 }
 553
 554 static void handle_pre_remove_exit(int status)
 555 {
 556         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 557                 snapshot_removal_status = HS_READY;
 558                 gettimeofday(&next_removal_check, NULL);
 559                 next_removal_check.tv_sec += 60;
 560                 return;
 561         }
 562         snapshot_removal_status = HS_PRE_SUCCESS;
 563 }
 564
 565 static int handle_rm_exit(int status)
 566 {
 567         if (!WIFEXITED(status)) {
 568                 snapshot_removal_status = HS_READY;
 569                 return -E_INVOLUNTARY_EXIT;
 570         }
 571         if (WEXITSTATUS(status)) {
 572                 snapshot_removal_status = HS_READY;
 573                 return -E_BAD_EXIT_CODE;
 574         }
 575         if (conf.post_remove_hook_given)
 576                 snapshot_removal_status = HS_SUCCESS;
 577         else
 578                 snapshot_removal_status = HS_READY;
 579         return 1;
 580 }
 581
 582 static void handle_post_remove_exit(void)
 583 {
 584         snapshot_removal_status = HS_READY;
 585 }
 586
 587 static int handle_remove_exit(int status)
 588 {
 589         int ret;
 590         struct snapshot *s = snapshot_currently_being_removed;
 591
 592         assert(s);
 593         switch (snapshot_removal_status) {
 594         case HS_PRE_RUNNING:
 595                 handle_pre_remove_exit(status);
 596                 ret = 1;
 597                 break;
 598         case HS_RUNNING:
 599                 ret = handle_rm_exit(status);
 600                 break;
 601         case HS_POST_RUNNING:
 602                 handle_post_remove_exit();
 603                 ret = 1;
 604                 break;
 605         default:
 606                 ret = -E_BUG;
 607         }
 608         if (snapshot_removal_status == HS_READY) {
 609                 free(s->name);
 610                 free(s);
 611                 snapshot_currently_being_removed = NULL;
 612         }
 613         remove_pid = 0;
 614         return ret;
 615 }
 616
 617 static int wait_for_remove_process(void)
 618 {
 619         int status, ret;
 620
 621         assert(remove_pid);
 622         assert(
 623                 snapshot_removal_status == HS_PRE_RUNNING ||
 624                 snapshot_removal_status == HS_RUNNING ||
 625                 snapshot_removal_status == HS_POST_RUNNING
 626         );
 627         ret = wait_for_process(remove_pid, &status);
 628         if (ret < 0)
 629                 return ret;
 630         return handle_remove_exit(status);
 631 }
 632
 633 static int handle_rsync_exit(int status)
 634 {
 635         int es, ret;
 636
 637         if (!WIFEXITED(status)) {
 638                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 639                 ret = -E_INVOLUNTARY_EXIT;
 640                 snapshot_creation_status = HS_READY;
 641                 goto out;
 642         }
 643         es = WEXITSTATUS(status);
 644         /*
 645          * Restart rsync on non-fatal errors:
 646          * 12: Error in rsync protocol data stream
 647          * 13: Errors with program diagnostics
 648          */
 649         if (es == 12 || es == 13) {
 650                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 651                         (int)create_pid, es);
 652                 snapshot_creation_status = HS_NEEDS_RESTART;
 653                 gettimeofday(&next_snapshot_time, NULL);
 654                 next_snapshot_time.tv_sec += 60;
 655                 ret = 1;
 656                 goto out;
 657         }
 658         if (es != 0 && es != 23 && es != 24) {
 659                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 660                 ret = -E_BAD_EXIT_CODE;
 661                 snapshot_creation_status = HS_READY;
 662                 goto out;
 663         }
 664         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 665         if (ret < 0)
 666                 goto out;
 667         snapshot_creation_status = HS_SUCCESS;
 668 out:
 669         create_pid = 0;
 670         create_process_stopped = 0;
 671         return ret;
 672 }
 673
 674 static int handle_pre_create_hook_exit(int status)
 675 {
 676         int es, ret;
 677         static int warn_count;
 678
 679         if (!WIFEXITED(status)) {
 680                 snapshot_creation_status = HS_READY;
 681                 ret = -E_INVOLUNTARY_EXIT;
 682                 goto out;
 683         }
 684         es = WEXITSTATUS(status);
 685         if (es) {
 686                 if (!warn_count--) {
 687                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 688                                 conf.pre_create_hook_arg, es);
 689                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 690                         warn_count = 60; /* warn only once per hour */
 691                 }
 692                 snapshot_creation_status = HS_READY;
 693                 ret = 0;
 694                 goto out;
 695         }
 696         warn_count = 0;
 697         snapshot_creation_status = HS_PRE_SUCCESS;
 698         ret = 1;
 699 out:
 700         create_pid = 0;
 701         return ret;
 702 }
 703
 704 static int handle_sigchld(void)
 705 {
 706         pid_t pid;
 707         int status, ret = reap_child(&pid, &status);
 708
 709         if (ret <= 0)
 710                 return ret;
 711
 712         if (pid == create_pid) {
 713                 switch (snapshot_creation_status) {
 714                 case HS_PRE_RUNNING:
 715                         return handle_pre_create_hook_exit(status);
 716                 case HS_RUNNING:
 717                         return handle_rsync_exit(status);
 718                 case HS_POST_RUNNING:
 719                         snapshot_creation_status = HS_READY;
 720                         return 1;
 721                 default:
 722                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 723                                 snapshot_creation_status);
 724                         return -E_BUG;
 725                 }
 726         }
 727         if (pid == remove_pid) {
 728                 ret = handle_remove_exit(status);
 729                 if (ret < 0)
 730                         return ret;
 731                 return ret;
 732         }
 733         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 734         return -E_BUG;
 735 }
 736
 737 static int check_config(void)
 738 {
 739         if (conf.unit_interval_arg <= 0) {
 740                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 741                 return -E_INVALID_NUMBER;
 742         }
 743         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 744         if (conf.num_intervals_arg <= 0) {
 745                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 746                 return -E_INVALID_NUMBER;
 747         }
 748         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 749         return 1;
 750 }
 751
 752 static int parse_config_file(int override)
 753 {
 754         int ret;
 755         char *config_file;
 756         struct stat statbuf;
 757         char *old_logfile_arg = NULL;
 758         int old_daemon_given = 0;
 759
 760         if (conf.config_file_given)
 761                 config_file = dss_strdup(conf.config_file_arg);
 762         else {
 763                 char *home = get_homedir();
 764                 config_file = make_message("%s/.dssrc", home);
 765                 free(home);
 766         }
 767         if (override) { /* SIGHUP */
 768                 if (conf.logfile_given)
 769                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 770                 old_daemon_given = conf.daemon_given;
 771         }
 772
 773         ret = stat(config_file, &statbuf);
 774         if (ret && conf.config_file_given) {
 775                 ret = -ERRNO_TO_DSS_ERROR(errno);
 776                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 777                 goto out;
 778         }
 779         if (!ret) {
 780                 struct cmdline_parser_params params = {
 781                         .override = override,
 782                         .initialize = 0,
 783                         .check_required = 1,
 784                         .check_ambiguity = 0,
 785                         .print_errors = 1
 786                 };
 787                 if (override) { /* invalidate all rsync options */
 788                         int i;
 789
 790                         for (i = 0; i < conf.rsync_option_given; i++) {
 791                                 free(conf.rsync_option_arg[i]);
 792                                 conf.rsync_option_arg[i] = NULL;
 793                         }
 794                         conf.rsync_option_given = 0;
 795                 }
 796                 cmdline_parser_config_file(config_file, &conf, &params);
 797         }
 798         ret = check_config();
 799         if (ret < 0)
 800                 goto out;
 801         if (override) {
 802                 /* don't change daemon mode on SIGHUP */
 803                 conf.daemon_given = old_daemon_given;
 804                 close_log(logfile);
 805                 logfile = NULL;
 806                 if (conf.logfile_given)
 807                         free(old_logfile_arg);
 808                 else if (conf.daemon_given) { /* re-use old logfile */
 809                         conf.logfile_arg = old_logfile_arg;
 810                         conf.logfile_given = 1;
 811                 }
 812         }
 813         if (conf.logfile_given) {
 814                 logfile = open_log(conf.logfile_arg);
 815                 log_welcome(conf.loglevel_arg);
 816         }
 817         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 818 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 819 out:
 820         free(config_file);
 821         if (ret < 0)
 822                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 823         return ret;
 824 }
 825
 826 static int change_to_dest_dir(void)
 827 {
 828         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 829         return dss_chdir(conf.dest_dir_arg);
 830 }
 831
 832 static int handle_sighup(void)
 833 {
 834         int ret;
 835
 836         DSS_NOTICE_LOG("SIGHUP\n");
 837         ret = parse_config_file(1);
 838         if (ret < 0)
 839                 return ret;
 840         return change_to_dest_dir();
 841 }
 842
 843 static int handle_signal(void)
 844 {
 845         int sig, ret = next_signal();
 846
 847         if (ret <= 0)
 848                 goto out;
 849         sig = ret;
 850         switch (sig) {
 851         case SIGINT:
 852         case SIGTERM:
 853                 restart_create_process();
 854                 kill_process(create_pid);
 855                 kill_process(remove_pid);
 856                 ret = -E_SIGNAL;
 857                 break;
 858         case SIGHUP:
 859                 ret = handle_sighup();
 860                 break;
 861         case SIGCHLD:
 862                 ret = handle_sigchld();
 863                 break;
 864         }
 865 out:
 866         if (ret < 0)
 867                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 868         return ret;
 869 }
 870
 871 /*
 872  * We can not use rsync locally if the local user is different from the remote
 873  * user or if the src dir is not on the local host (or both).
 874  */
 875 static int use_rsync_locally(char *logname)
 876 {
 877         char *h = conf.remote_host_arg;
 878
 879         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 880                 return 0;
 881         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 882                 return 0;
 883         return 1;
 884 }
 885
 886 static void create_rsync_argv(char ***argv, int64_t *num)
 887 {
 888         char *logname, *newest;
 889         int i = 0, j;
 890         struct snapshot_list sl;
 891
 892         dss_get_snapshot_list(&sl);
 893         newest = name_of_newest_complete_snapshot(&sl);
 894         free_snapshot_list(&sl);
 895
 896         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 897         (*argv)[i++] = dss_strdup("rsync");
 898         (*argv)[i++] = dss_strdup("-aq");
 899         (*argv)[i++] = dss_strdup("--delete");
 900         for (j = 0; j < conf.rsync_option_given; j++)
 901                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 902         if (newest) {
 903                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 904                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 905                 free(newest);
 906         } else
 907                 DSS_INFO_LOG("no previous snapshot found\n");
 908         logname = dss_logname();
 909         if (use_rsync_locally(logname))
 910                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 911         else
 912                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 913                         conf.remote_user_arg : logname,
 914                         conf.remote_host_arg, conf.source_dir_arg);
 915         free(logname);
 916         *num = get_current_time();
 917         (*argv)[i++] = incomplete_name(*num);
 918         (*argv)[i++] = NULL;
 919         for (j = 0; j < i; j++)
 920                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 921 }
 922
 923 static void free_rsync_argv(char **argv)
 924 {
 925         int i;
 926
 927         if (!argv)
 928                 return;
 929         for (i = 0; argv[i]; i++)
 930                 free(argv[i]);
 931         free(argv);
 932 }
 933
 934 static int create_snapshot(char **argv)
 935 {
 936         int ret, fds[3] = {0, 0, 0};
 937         char *name;
 938
 939         name = incomplete_name(current_snapshot_creation_time);
 940         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 941         free(name);
 942         ret = dss_exec(&create_pid, argv[0], argv, fds);
 943         if (ret < 0)
 944                 return ret;
 945         snapshot_creation_status = HS_RUNNING;
 946         return ret;
 947 }
 948
 949 static int select_loop(void)
 950 {
 951         int ret;
 952         /* check every 60 seconds for free disk space */
 953         struct timeval tv;
 954         char **rsync_argv = NULL;
 955
 956         for (;;) {
 957                 fd_set rfds;
 958                 int low_disk_space;
 959                 struct timeval *tvp;
 960
 961                 if (remove_pid)
 962                         tvp = NULL; /* sleep until rm hook/process dies */
 963                 else { /* sleep one minute */
 964                         tv.tv_sec = 60;
 965                         tv.tv_usec = 0;
 966                         tvp = &tv;
 967                 }
 968                 FD_ZERO(&rfds);
 969                 FD_SET(signal_pipe, &rfds);
 970                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
 971                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
 972                 if (ret < 0)
 973                         goto out;
 974                 if (FD_ISSET(signal_pipe, &rfds)) {
 975                         ret = handle_signal();
 976                         if (ret < 0)
 977                                 goto out;
 978                 }
 979                 if (remove_pid)
 980                         continue;
 981                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
 982                         ret = exec_rm();
 983                         if (ret < 0)
 984                                 goto out;
 985                         continue;
 986                 }
 987                 if (snapshot_removal_status == HS_SUCCESS) {
 988                         ret = post_remove_hook();
 989                         if (ret < 0)
 990                                 goto out;
 991                         continue;
 992                 }
 993                 ret = disk_space_low();
 994                 if (ret < 0)
 995                         goto out;
 996                 low_disk_space = ret;
 997                 ret = try_to_free_disk_space(low_disk_space);
 998                 if (ret < 0)
 999                         goto out;
1000                 if (snapshot_removal_status != HS_READY) {
1001                         stop_create_process();
1002                         continue;
1003                 }
1004                 restart_create_process();
1005                 switch (snapshot_creation_status) {
1006                 case HS_READY:
1007                         if (!next_snapshot_is_due())
1008                                 continue;
1009                         ret = pre_create_hook();
1010                         if (ret < 0)
1011                                 goto out;
1012                         continue;
1013                 case HS_PRE_RUNNING:
1014                 case HS_RUNNING:
1015                 case HS_POST_RUNNING:
1016                         continue;
1017                 case HS_PRE_SUCCESS:
1018                         free_rsync_argv(rsync_argv);
1019                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1020                         /* fall through */
1021                 case HS_NEEDS_RESTART:
1022                         ret = create_snapshot(rsync_argv);
1023                         if (ret < 0)
1024                                 goto out;
1025                         continue;
1026                 case HS_SUCCESS:
1027                         ret = post_create_hook();
1028                         if (ret < 0)
1029                                 goto out;
1030                         continue;
1031                 }
1032         }
1033 out:
1034         return ret;
1035 }
1036
1037 static void exit_hook(int exit_code)
1038 {
1039         int fds[3] = {0, 0, 0};
1040         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1041         pid_t pid;
1042
1043         if (!conf.exit_hook_given)
1044                 return;
1045         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1046         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1047 }
1048
1049 static int com_run(void)
1050 {
1051         int ret;
1052
1053         if (conf.dry_run_given) {
1054                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1055                 return -E_SYNTAX;
1056         }
1057         ret = install_sighandler(SIGHUP);
1058         if (ret < 0)
1059                 return ret;
1060         ret = select_loop();
1061         if (ret >= 0) /* impossible */
1062                 ret = -E_BUG;
1063         exit_hook(ret);
1064         return ret;
1065 }
1066
1067 static int com_prune(void)
1068 {
1069         int ret;
1070         struct snapshot_list sl;
1071         struct disk_space ds;
1072
1073         ret = get_disk_space(".", &ds);
1074         if (ret < 0)
1075                 return ret;
1076         log_disk_space(&ds);
1077         dss_get_snapshot_list(&sl);
1078         ret = remove_outdated_snapshot(&sl);
1079         if (ret < 0)
1080                 goto out;
1081         if (ret > 0)
1082                 goto rm;
1083         ret = remove_redundant_snapshot(&sl);
1084         if (ret < 0)
1085                 goto out;
1086         if (ret > 0)
1087                 goto rm;
1088         ret = 0;
1089         goto out;
1090 rm:
1091         if (snapshot_removal_status == HS_PRE_RUNNING) {
1092                 ret = wait_for_remove_process();
1093                 if (ret < 0)
1094                         goto out;
1095                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1096                         goto out;
1097         }
1098         ret = exec_rm();
1099         if (ret < 0)
1100                 goto out;
1101         ret = wait_for_remove_process();
1102         if (ret < 0)
1103                 goto out;
1104         if (snapshot_removal_status != HS_SUCCESS)
1105                 goto out;
1106         ret = post_remove_hook();
1107         if (ret < 0)
1108                 goto out;
1109         if (snapshot_removal_status != HS_POST_RUNNING)
1110                 goto out;
1111         ret = wait_for_remove_process();
1112         if (ret < 0)
1113                 goto out;
1114         ret = 1;
1115 out:
1116         free_snapshot_list(&sl);
1117         return ret;
1118 }
1119
1120 static int com_create(void)
1121 {
1122         int ret, status;
1123         char **rsync_argv;
1124
1125         if (conf.dry_run_given) {
1126                 int i;
1127                 char *msg = NULL;
1128                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1129                 for (i = 0; rsync_argv[i]; i++) {
1130                         char *tmp = msg;
1131                         msg = make_message("%s%s%s", tmp? tmp : "",
1132                                 tmp? " " : "", rsync_argv[i]);
1133                         free(tmp);
1134                 }
1135                 free_rsync_argv(rsync_argv);
1136                 dss_msg("%s\n", msg);
1137                 free(msg);
1138                 return 1;
1139         }
1140         ret = pre_create_hook();
1141         if (ret < 0)
1142                 return ret;
1143         if (create_pid) {
1144                 ret = wait_for_process(create_pid, &status);
1145                 if (ret < 0)
1146                         return ret;
1147                 ret = handle_pre_create_hook_exit(status);
1148                 if (ret <= 0) /* error, or pre-create failed */
1149                         return ret;
1150         }
1151         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1152         ret = create_snapshot(rsync_argv);
1153         if (ret < 0)
1154                 goto out;
1155         ret = wait_for_process(create_pid, &status);
1156         if (ret < 0)
1157                 goto out;
1158         ret = handle_rsync_exit(status);
1159         if (ret < 0)
1160                 goto out;
1161         post_create_hook();
1162         if (create_pid)
1163                 ret = wait_for_process(create_pid, &status);
1164 out:
1165         free_rsync_argv(rsync_argv);
1166         return ret;
1167 }
1168
1169 static int com_ls(void)
1170 {
1171         int i;
1172         struct snapshot_list sl;
1173         struct snapshot *s;
1174
1175         dss_get_snapshot_list(&sl);
1176         FOR_EACH_SNAPSHOT(s, i, &sl) {
1177                 int64_t d = 0;
1178                 if (s->flags & SS_COMPLETE)
1179                         d = (s->completion_time - s->creation_time) / 60;
1180                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1181         };
1182         free_snapshot_list(&sl);
1183         return 1;
1184 }
1185
1186 static int setup_signal_handling(void)
1187 {
1188         int ret;
1189
1190         DSS_INFO_LOG("setting up signal handlers\n");
1191         signal_pipe = signal_init(); /* always successful */
1192         ret = install_sighandler(SIGINT);
1193         if (ret < 0)
1194                 return ret;
1195         ret = install_sighandler(SIGTERM);
1196         if (ret < 0)
1197                 return ret;
1198         return install_sighandler(SIGCHLD);
1199 }
1200
1201 /**
1202  * The main function of dss.
1203  *
1204  * \param argc Usual argument count.
1205  * \param argv Usual argument vector.
1206  */
1207 int main(int argc, char **argv)
1208 {
1209         int ret;
1210         struct cmdline_parser_params params = {
1211                 .override = 0,
1212                 .initialize = 1,
1213                 .check_required = 0,
1214                 .check_ambiguity = 0,
1215                 .print_errors = 1
1216         };
1217
1218         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1219         ret = parse_config_file(0);
1220         if (ret < 0)
1221                 goto out;
1222         if (conf.daemon_given)
1223                 daemon_init();
1224         ret = change_to_dest_dir();
1225         if (ret < 0)
1226                 goto out;
1227         ret = setup_signal_handling();
1228         if (ret < 0)
1229                 goto out;
1230         ret = call_command_handler();
1231 out:
1232         if (ret < 0)
1233                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1234         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1235 }