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 between two snapshots */
 189         diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
 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, const char *why)
 222 {
 223         int ret, fds[3] = {0, 0, 0};
 224         char *cmd;
 225
 226         if (!s)
 227                 return 0;
 228         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 229         assert(snapshot_removal_status == HS_READY);
 230         assert(remove_pid == 0);
 231         assert(!snapshot_currently_being_removed);
 232
 233         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 234         *snapshot_currently_being_removed = *s;
 235         snapshot_currently_being_removed->name = dss_strdup(s->name);
 236
 237         if (!conf.pre_remove_hook_given) {
 238                 snapshot_removal_status = HS_PRE_SUCCESS;
 239                 return 0;
 240         }
 241         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 242                 conf.dest_dir_arg, s->name);
 243         DSS_DEBUG_LOG("executing %s\n", cmd);
 244         ret = dss_exec_cmdline_pid(&remove_pid,
 245                 conf.pre_remove_hook_arg, fds);
 246         free(cmd);
 247         if (ret < 0)
 248                 return ret;
 249         snapshot_removal_status = HS_PRE_RUNNING;
 250         return ret;
 251 }
 252
 253 static int exec_rm(void)
 254 {
 255         struct snapshot *s = snapshot_currently_being_removed;
 256         int fds[3] = {0, 0, 0};
 257         char *new_name = being_deleted_name(s);
 258         char *argv[] = {"rm", "-rf", new_name, NULL};
 259         int ret;
 260
 261         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 262         assert(remove_pid == 0);
 263
 264         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 265         ret = dss_rename(s->name, new_name);
 266         if (ret < 0)
 267                 goto out;
 268         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 269         if (ret < 0)
 270                 goto out;
 271         snapshot_removal_status = HS_RUNNING;
 272 out:
 273         free(new_name);
 274         return ret;
 275 }
 276
 277 static int snapshot_is_being_created(struct snapshot *s)
 278 {
 279         return s->creation_time == current_snapshot_creation_time;
 280 }
 281
 282 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 283 {
 284         struct snapshot *s;
 285         int i;
 286
 287         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 288         FOR_EACH_SNAPSHOT(s, i, sl) {
 289                 if (snapshot_is_being_created(s))
 290                         continue;
 291                 /*
 292                  * We know that no rm is currently running, so if s is marked
 293                  * as being deleted, a previously started rm must have failed.
 294                  */
 295                 if (s->flags & SS_BEING_DELETED)
 296                         return s;
 297
 298                 if (s->flags & SS_COMPLETE) /* good snapshot */
 299                         continue;
 300                 /*
 301                  * This snapshot is incomplete and it is not the snapshot
 302                  * currently being created. However, we must not remove it if
 303                  * rsync is about to be restarted. As only the newest snapshot
 304                  * can be restarted, this snapshot is orphaned if it is not the
 305                  * newest snapshot or if we are not about to restart rsync.
 306                  */
 307                 if (get_newest_snapshot(sl) != s)
 308                         return s;
 309                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 310                         return s;
 311         }
 312         /* no orphaned snapshots */
 313         return NULL;
 314 }
 315
 316 /*
 317  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 318  */
 319 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 320 {
 321         int i, interval;
 322         struct snapshot *s;
 323         unsigned missing = 0;
 324
 325         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 326         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 327                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 328                 unsigned num = sl->interval_count[interval];
 329                 struct snapshot *victim = NULL, *prev = NULL;
 330                 int64_t score = LONG_MAX;
 331
 332                 if (keep >= num)
 333                         missing += keep - num;
 334 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 335 //                      interval, keep, num, missing);
 336                 if (keep + missing >= num)
 337                         continue;
 338                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 339                 FOR_EACH_SNAPSHOT(s, i, sl) {
 340                         int64_t this_score;
 341
 342                         if (snapshot_is_being_created(s))
 343                                 continue;
 344                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 345                         if (s->interval > interval) {
 346                                 prev = s;
 347                                 continue;
 348                         }
 349                         if (s->interval < interval)
 350                                 break;
 351                         if (!victim) {
 352                                 victim = s;
 353                                 prev = s;
 354                                 continue;
 355                         }
 356                         assert(prev);
 357                         /* check if s is a better victim */
 358                         this_score = s->creation_time - prev->creation_time;
 359                         assert(this_score >= 0);
 360                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 361                         if (this_score < score) {
 362                                 score = this_score;
 363                                 victim = s;
 364                         }
 365                         prev = s;
 366                 }
 367                 assert(victim);
 368                 return victim;
 369         }
 370         return NULL;
 371 }
 372
 373 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 374 {
 375         int i;
 376         struct snapshot *s;
 377
 378         DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n",
 379                 conf.num_intervals_arg);
 380         FOR_EACH_SNAPSHOT(s, i, sl) {
 381                 if (snapshot_is_being_created(s))
 382                         continue;
 383                 if (s->interval <= conf.num_intervals_arg)
 384                         continue;
 385                 return s;
 386         }
 387         return NULL;
 388 }
 389
 390 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 391 {
 392         struct snapshot *s = get_oldest_snapshot(sl);
 393
 394         if (!s) /* no snapshot found */
 395                 return NULL;
 396         DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
 397         if (snapshot_is_being_created(s))
 398                 return NULL;
 399         return s;
 400 }
 401
 402 static int rename_incomplete_snapshot(int64_t start)
 403 {
 404         char *old_name;
 405         int ret;
 406
 407         free(path_to_last_complete_snapshot);
 408         ret = complete_name(start, get_current_time(),
 409                 &path_to_last_complete_snapshot);
 410         if (ret < 0)
 411                 return ret;
 412         old_name = incomplete_name(start);
 413         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 414         if (ret >= 0)
 415                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 416                         path_to_last_complete_snapshot);
 417         free(old_name);
 418         return ret;
 419 }
 420
 421 static int try_to_free_disk_space(int low_disk_space)
 422 {
 423         int ret;
 424         struct snapshot_list sl;
 425         struct snapshot *victim;
 426         struct timeval now;
 427         const char *why;
 428
 429         gettimeofday(&now, NULL);
 430         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 431                 return 0;
 432         if (!low_disk_space && conf.keep_redundant_given)
 433                 return 0;
 434         dss_get_snapshot_list(&sl);
 435         why = "outdated";
 436         victim = find_outdated_snapshot(&sl);
 437         if (victim)
 438                 goto remove;
 439         why = "redundant";
 440         victim = find_redundant_snapshot(&sl);
 441         if (victim)
 442                 goto remove;
 443         /* try harder only if disk space is low */
 444         ret = 0;
 445         if (!low_disk_space)
 446                 goto out;
 447         why = "orphaned";
 448         victim = find_orphaned_snapshot(&sl);
 449         if (victim)
 450                 goto remove;
 451         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 452         victim = find_oldest_removable_snapshot(&sl);
 453         if (victim)
 454                 goto remove;
 455         DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
 456         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 457         goto out;
 458 remove:
 459         ret = pre_remove_hook(victim, why);
 460 out:
 461         free_snapshot_list(&sl);
 462         return ret;
 463 }
 464
 465 static int post_create_hook(void)
 466 {
 467         int ret, fds[3] = {0, 0, 0};
 468         char *cmd;
 469
 470         if (!conf.post_create_hook_given) {
 471                 snapshot_creation_status = HS_READY;
 472                 return 0;
 473         }
 474         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 475                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 476         DSS_NOTICE_LOG("executing %s\n", cmd);
 477         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 478         free(cmd);
 479         if (ret < 0)
 480                 return ret;
 481         snapshot_creation_status = HS_POST_RUNNING;
 482         return ret;
 483 }
 484
 485 static int post_remove_hook(void)
 486 {
 487         int ret, fds[3] = {0, 0, 0};
 488         char *cmd;
 489         struct snapshot *s = snapshot_currently_being_removed;
 490
 491         assert(s);
 492
 493         if (!conf.post_remove_hook_given) {
 494                 snapshot_removal_status = HS_READY;
 495                 return 0;
 496         }
 497         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 498                 conf.dest_dir_arg, s->name);
 499         DSS_NOTICE_LOG("executing %s\n", cmd);
 500         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 501         free(cmd);
 502         if (ret < 0)
 503                 return ret;
 504         snapshot_removal_status = HS_POST_RUNNING;
 505         return ret;
 506 }
 507
 508 static void kill_process(pid_t pid)
 509 {
 510         if (!pid)
 511                 return;
 512         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 513         kill(pid, SIGTERM);
 514 }
 515
 516 static void stop_create_process(void)
 517 {
 518         if (!create_pid || create_process_stopped)
 519                 return;
 520         kill(SIGSTOP, create_pid);
 521         create_process_stopped = 1;
 522 }
 523
 524 static void restart_create_process(void)
 525 {
 526         if (!create_pid || !create_process_stopped)
 527                 return;
 528         kill (SIGCONT, create_pid);
 529         create_process_stopped = 0;
 530 }
 531
 532 /**
 533  * Print a log message about the exit status of a child.
 534  */
 535 static void log_termination_msg(pid_t pid, int status)
 536 {
 537         if (WIFEXITED(status))
 538                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 539                         WEXITSTATUS(status));
 540         else if (WIFSIGNALED(status))
 541                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 542                         WTERMSIG(status));
 543         else
 544                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 545 }
 546
 547 static int wait_for_process(pid_t pid, int *status)
 548 {
 549         int ret;
 550
 551         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 552         for (;;) {
 553                 fd_set rfds;
 554
 555                 FD_ZERO(&rfds);
 556                 FD_SET(signal_pipe, &rfds);
 557                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 558                 if (ret < 0)
 559                         break;
 560                 ret = next_signal();
 561                 if (!ret)
 562                         continue;
 563                 if (ret == SIGCHLD) {
 564                         ret = waitpid(pid, status, 0);
 565                         if (ret >= 0)
 566                                 break;
 567                         if (errno != EINTR) { /* error */
 568                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 569                                 break;
 570                         }
 571                 }
 572                 /* SIGINT or SIGTERM */
 573                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 574                 kill(pid, SIGTERM);
 575         }
 576         if (ret < 0)
 577                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 578         else
 579                 log_termination_msg(pid, *status);
 580         return ret;
 581 }
 582
 583 static void handle_pre_remove_exit(int status)
 584 {
 585         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 586                 snapshot_removal_status = HS_READY;
 587                 gettimeofday(&next_removal_check, NULL);
 588                 next_removal_check.tv_sec += 60;
 589                 return;
 590         }
 591         snapshot_removal_status = HS_PRE_SUCCESS;
 592 }
 593
 594 static int handle_rm_exit(int status)
 595 {
 596         if (!WIFEXITED(status)) {
 597                 snapshot_removal_status = HS_READY;
 598                 return -E_INVOLUNTARY_EXIT;
 599         }
 600         if (WEXITSTATUS(status)) {
 601                 snapshot_removal_status = HS_READY;
 602                 return -E_BAD_EXIT_CODE;
 603         }
 604         if (conf.post_remove_hook_given)
 605                 snapshot_removal_status = HS_SUCCESS;
 606         else
 607                 snapshot_removal_status = HS_READY;
 608         return 1;
 609 }
 610
 611 static void handle_post_remove_exit(void)
 612 {
 613         snapshot_removal_status = HS_READY;
 614 }
 615
 616 static int handle_remove_exit(int status)
 617 {
 618         int ret;
 619         struct snapshot *s = snapshot_currently_being_removed;
 620
 621         assert(s);
 622         switch (snapshot_removal_status) {
 623         case HS_PRE_RUNNING:
 624                 handle_pre_remove_exit(status);
 625                 ret = 1;
 626                 break;
 627         case HS_RUNNING:
 628                 ret = handle_rm_exit(status);
 629                 break;
 630         case HS_POST_RUNNING:
 631                 handle_post_remove_exit();
 632                 ret = 1;
 633                 break;
 634         default:
 635                 ret = -E_BUG;
 636         }
 637         if (snapshot_removal_status == HS_READY) {
 638                 free(s->name);
 639                 free(s);
 640                 snapshot_currently_being_removed = NULL;
 641         }
 642         remove_pid = 0;
 643         return ret;
 644 }
 645
 646 static int wait_for_remove_process(void)
 647 {
 648         int status, ret;
 649
 650         assert(remove_pid);
 651         assert(
 652                 snapshot_removal_status == HS_PRE_RUNNING ||
 653                 snapshot_removal_status == HS_RUNNING ||
 654                 snapshot_removal_status == HS_POST_RUNNING
 655         );
 656         ret = wait_for_process(remove_pid, &status);
 657         if (ret < 0)
 658                 return ret;
 659         return handle_remove_exit(status);
 660 }
 661
 662 static int handle_rsync_exit(int status)
 663 {
 664         int es, ret;
 665
 666         if (!WIFEXITED(status)) {
 667                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 668                 ret = -E_INVOLUNTARY_EXIT;
 669                 snapshot_creation_status = HS_READY;
 670                 goto out;
 671         }
 672         es = WEXITSTATUS(status);
 673         /*
 674          * Restart rsync on non-fatal errors:
 675          * 12: Error in rsync protocol data stream
 676          * 13: Errors with program diagnostics
 677          */
 678         if (es == 12 || es == 13) {
 679                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 680                         (int)create_pid, es);
 681                 snapshot_creation_status = HS_NEEDS_RESTART;
 682                 gettimeofday(&next_snapshot_time, NULL);
 683                 next_snapshot_time.tv_sec += 60;
 684                 ret = 1;
 685                 goto out;
 686         }
 687         if (es != 0 && es != 23 && es != 24) {
 688                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 689                 ret = -E_BAD_EXIT_CODE;
 690                 snapshot_creation_status = HS_READY;
 691                 goto out;
 692         }
 693         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 694         if (ret < 0)
 695                 goto out;
 696         snapshot_creation_status = HS_SUCCESS;
 697 out:
 698         create_pid = 0;
 699         create_process_stopped = 0;
 700         return ret;
 701 }
 702
 703 static int handle_pre_create_hook_exit(int status)
 704 {
 705         int es, ret;
 706         static int warn_count;
 707
 708         if (!WIFEXITED(status)) {
 709                 snapshot_creation_status = HS_READY;
 710                 ret = -E_INVOLUNTARY_EXIT;
 711                 goto out;
 712         }
 713         es = WEXITSTATUS(status);
 714         if (es) {
 715                 if (!warn_count--) {
 716                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 717                                 conf.pre_create_hook_arg, es);
 718                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 719                         warn_count = 60; /* warn only once per hour */
 720                 }
 721                 snapshot_creation_status = HS_READY;
 722                 ret = 0;
 723                 goto out;
 724         }
 725         warn_count = 0;
 726         snapshot_creation_status = HS_PRE_SUCCESS;
 727         ret = 1;
 728 out:
 729         create_pid = 0;
 730         return ret;
 731 }
 732
 733 static int handle_sigchld(void)
 734 {
 735         pid_t pid;
 736         int status, ret = reap_child(&pid, &status);
 737
 738         if (ret <= 0)
 739                 return ret;
 740
 741         if (pid == create_pid) {
 742                 switch (snapshot_creation_status) {
 743                 case HS_PRE_RUNNING:
 744                         return handle_pre_create_hook_exit(status);
 745                 case HS_RUNNING:
 746                         return handle_rsync_exit(status);
 747                 case HS_POST_RUNNING:
 748                         snapshot_creation_status = HS_READY;
 749                         return 1;
 750                 default:
 751                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 752                                 snapshot_creation_status);
 753                         return -E_BUG;
 754                 }
 755         }
 756         if (pid == remove_pid) {
 757                 ret = handle_remove_exit(status);
 758                 if (ret < 0)
 759                         return ret;
 760                 return ret;
 761         }
 762         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 763         return -E_BUG;
 764 }
 765
 766 static int check_config(void)
 767 {
 768         if (conf.unit_interval_arg <= 0) {
 769                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 770                 return -E_INVALID_NUMBER;
 771         }
 772         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 773         if (conf.num_intervals_arg <= 0) {
 774                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 775                 return -E_INVALID_NUMBER;
 776         }
 777         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 778         return 1;
 779 }
 780
 781 static int parse_config_file(int override)
 782 {
 783         int ret;
 784         char *config_file;
 785         struct stat statbuf;
 786         char *old_logfile_arg = NULL;
 787         int old_daemon_given = 0;
 788
 789         if (conf.config_file_given)
 790                 config_file = dss_strdup(conf.config_file_arg);
 791         else {
 792                 char *home = get_homedir();
 793                 config_file = make_message("%s/.dssrc", home);
 794                 free(home);
 795         }
 796         if (override) { /* SIGHUP */
 797                 if (conf.logfile_given)
 798                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 799                 old_daemon_given = conf.daemon_given;
 800         }
 801
 802         ret = stat(config_file, &statbuf);
 803         if (ret && conf.config_file_given) {
 804                 ret = -ERRNO_TO_DSS_ERROR(errno);
 805                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 806                 goto out;
 807         }
 808         if (!ret) {
 809                 struct cmdline_parser_params params = {
 810                         .override = override,
 811                         .initialize = 0,
 812                         .check_required = 1,
 813                         .check_ambiguity = 0,
 814                         .print_errors = 1
 815                 };
 816                 if (override) { /* invalidate all rsync options */
 817                         int i;
 818
 819                         for (i = 0; i < conf.rsync_option_given; i++) {
 820                                 free(conf.rsync_option_arg[i]);
 821                                 conf.rsync_option_arg[i] = NULL;
 822                         }
 823                         conf.rsync_option_given = 0;
 824                 }
 825                 cmdline_parser_config_file(config_file, &conf, &params);
 826         }
 827         ret = check_config();
 828         if (ret < 0)
 829                 goto out;
 830         if (override) {
 831                 /* don't change daemon mode on SIGHUP */
 832                 conf.daemon_given = old_daemon_given;
 833                 close_log(logfile);
 834                 logfile = NULL;
 835                 if (conf.logfile_given)
 836                         free(old_logfile_arg);
 837                 else if (conf.daemon_given) { /* re-use old logfile */
 838                         conf.logfile_arg = old_logfile_arg;
 839                         conf.logfile_given = 1;
 840                 }
 841         }
 842         if (conf.logfile_given) {
 843                 logfile = open_log(conf.logfile_arg);
 844                 log_welcome(conf.loglevel_arg);
 845         }
 846         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 847 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 848 out:
 849         free(config_file);
 850         if (ret < 0)
 851                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 852         return ret;
 853 }
 854
 855 static int change_to_dest_dir(void)
 856 {
 857         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 858         return dss_chdir(conf.dest_dir_arg);
 859 }
 860
 861 static int handle_sighup(void)
 862 {
 863         int ret;
 864
 865         DSS_NOTICE_LOG("SIGHUP\n");
 866         ret = parse_config_file(1);
 867         if (ret < 0)
 868                 return ret;
 869         return change_to_dest_dir();
 870 }
 871
 872 static int handle_signal(void)
 873 {
 874         int sig, ret = next_signal();
 875
 876         if (ret <= 0)
 877                 goto out;
 878         sig = ret;
 879         switch (sig) {
 880         case SIGINT:
 881         case SIGTERM:
 882                 restart_create_process();
 883                 kill_process(create_pid);
 884                 kill_process(remove_pid);
 885                 ret = -E_SIGNAL;
 886                 break;
 887         case SIGHUP:
 888                 ret = handle_sighup();
 889                 break;
 890         case SIGCHLD:
 891                 ret = handle_sigchld();
 892                 break;
 893         }
 894 out:
 895         if (ret < 0)
 896                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 897         return ret;
 898 }
 899
 900 /*
 901  * We can not use rsync locally if the local user is different from the remote
 902  * user or if the src dir is not on the local host (or both).
 903  */
 904 static int use_rsync_locally(char *logname)
 905 {
 906         char *h = conf.remote_host_arg;
 907
 908         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 909                 return 0;
 910         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 911                 return 0;
 912         return 1;
 913 }
 914
 915 static void create_rsync_argv(char ***argv, int64_t *num)
 916 {
 917         char *logname, *newest;
 918         int i = 0, j;
 919         struct snapshot_list sl;
 920
 921         dss_get_snapshot_list(&sl);
 922         newest = name_of_newest_complete_snapshot(&sl);
 923         free_snapshot_list(&sl);
 924
 925         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 926         (*argv)[i++] = dss_strdup("rsync");
 927         (*argv)[i++] = dss_strdup("-aq");
 928         (*argv)[i++] = dss_strdup("--delete");
 929         for (j = 0; j < conf.rsync_option_given; j++)
 930                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 931         if (newest) {
 932                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 933                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 934                 free(newest);
 935         } else
 936                 DSS_INFO_LOG("no previous snapshot found\n");
 937         logname = dss_logname();
 938         if (use_rsync_locally(logname))
 939                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 940         else
 941                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 942                         conf.remote_user_arg : logname,
 943                         conf.remote_host_arg, conf.source_dir_arg);
 944         free(logname);
 945         *num = get_current_time();
 946         (*argv)[i++] = incomplete_name(*num);
 947         (*argv)[i++] = NULL;
 948         for (j = 0; j < i; j++)
 949                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 950 }
 951
 952 static void free_rsync_argv(char **argv)
 953 {
 954         int i;
 955
 956         if (!argv)
 957                 return;
 958         for (i = 0; argv[i]; i++)
 959                 free(argv[i]);
 960         free(argv);
 961 }
 962
 963 static int create_snapshot(char **argv)
 964 {
 965         int ret, fds[3] = {0, 0, 0};
 966         char *name;
 967
 968         name = incomplete_name(current_snapshot_creation_time);
 969         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 970         free(name);
 971         ret = dss_exec(&create_pid, argv[0], argv, fds);
 972         if (ret < 0)
 973                 return ret;
 974         snapshot_creation_status = HS_RUNNING;
 975         return ret;
 976 }
 977
 978 static int select_loop(void)
 979 {
 980         int ret;
 981         /* check every 60 seconds for free disk space */
 982         struct timeval tv;
 983         char **rsync_argv = NULL;
 984
 985         for (;;) {
 986                 fd_set rfds;
 987                 int low_disk_space;
 988                 struct timeval *tvp;
 989
 990                 if (remove_pid)
 991                         tvp = NULL; /* sleep until rm hook/process dies */
 992                 else { /* sleep one minute */
 993                         tv.tv_sec = 60;
 994                         tv.tv_usec = 0;
 995                         tvp = &tv;
 996                 }
 997                 FD_ZERO(&rfds);
 998                 FD_SET(signal_pipe, &rfds);
 999                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1000                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1001                 if (ret < 0)
1002                         goto out;
1003                 if (FD_ISSET(signal_pipe, &rfds)) {
1004                         ret = handle_signal();
1005                         if (ret < 0)
1006                                 goto out;
1007                 }
1008                 if (remove_pid)
1009                         continue;
1010                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1011                         ret = exec_rm();
1012                         if (ret < 0)
1013                                 goto out;
1014                         continue;
1015                 }
1016                 if (snapshot_removal_status == HS_SUCCESS) {
1017                         ret = post_remove_hook();
1018                         if (ret < 0)
1019                                 goto out;
1020                         continue;
1021                 }
1022                 ret = disk_space_low();
1023                 if (ret < 0)
1024                         goto out;
1025                 low_disk_space = ret;
1026                 ret = try_to_free_disk_space(low_disk_space);
1027                 if (ret < 0)
1028                         goto out;
1029                 if (snapshot_removal_status != HS_READY) {
1030                         stop_create_process();
1031                         continue;
1032                 }
1033                 restart_create_process();
1034                 switch (snapshot_creation_status) {
1035                 case HS_READY:
1036                         if (!next_snapshot_is_due())
1037                                 continue;
1038                         ret = pre_create_hook();
1039                         if (ret < 0)
1040                                 goto out;
1041                         continue;
1042                 case HS_PRE_RUNNING:
1043                 case HS_RUNNING:
1044                 case HS_POST_RUNNING:
1045                         continue;
1046                 case HS_PRE_SUCCESS:
1047                         free_rsync_argv(rsync_argv);
1048                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1049                         /* fall through */
1050                 case HS_NEEDS_RESTART:
1051                         ret = create_snapshot(rsync_argv);
1052                         if (ret < 0)
1053                                 goto out;
1054                         continue;
1055                 case HS_SUCCESS:
1056                         ret = post_create_hook();
1057                         if (ret < 0)
1058                                 goto out;
1059                         continue;
1060                 }
1061         }
1062 out:
1063         return ret;
1064 }
1065
1066 static void exit_hook(int exit_code)
1067 {
1068         int fds[3] = {0, 0, 0};
1069         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1070         pid_t pid;
1071
1072         if (!conf.exit_hook_given)
1073                 return;
1074         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1075         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1076 }
1077
1078 static int com_run(void)
1079 {
1080         int ret;
1081
1082         if (conf.dry_run_given) {
1083                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1084                 return -E_SYNTAX;
1085         }
1086         ret = install_sighandler(SIGHUP);
1087         if (ret < 0)
1088                 return ret;
1089         ret = select_loop();
1090         if (ret >= 0) /* impossible */
1091                 ret = -E_BUG;
1092         exit_hook(ret);
1093         return ret;
1094 }
1095
1096 static int com_prune(void)
1097 {
1098         int ret;
1099         struct snapshot_list sl;
1100         struct snapshot *victim;
1101         struct disk_space ds;
1102         const char *why;
1103
1104         ret = get_disk_space(".", &ds);
1105         if (ret < 0)
1106                 return ret;
1107         log_disk_space(&ds);
1108         dss_get_snapshot_list(&sl);
1109         why = "outdated";
1110         victim = find_outdated_snapshot(&sl);
1111         if (victim)
1112                 goto rm;
1113         why = "redundant";
1114         victim = find_redundant_snapshot(&sl);
1115         if (victim)
1116                 goto rm;
1117         ret = 0;
1118         goto out;
1119 rm:
1120         if (conf.dry_run_given) {
1121                 dss_msg("%s snapshot %s (interval = %i)\n",
1122                         why, victim->name, victim->interval);
1123                 ret = 0;
1124                 goto out;
1125         }
1126         ret = pre_remove_hook(victim, why);
1127         if (ret < 0)
1128                 goto out;
1129         if (snapshot_removal_status == HS_PRE_RUNNING) {
1130                 ret = wait_for_remove_process();
1131                 if (ret < 0)
1132                         goto out;
1133                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1134                         goto out;
1135         }
1136         ret = exec_rm();
1137         if (ret < 0)
1138                 goto out;
1139         ret = wait_for_remove_process();
1140         if (ret < 0)
1141                 goto out;
1142         if (snapshot_removal_status != HS_SUCCESS)
1143                 goto out;
1144         ret = post_remove_hook();
1145         if (ret < 0)
1146                 goto out;
1147         if (snapshot_removal_status != HS_POST_RUNNING)
1148                 goto out;
1149         ret = wait_for_remove_process();
1150         if (ret < 0)
1151                 goto out;
1152         ret = 1;
1153 out:
1154         free_snapshot_list(&sl);
1155         return ret;
1156 }
1157
1158 static int com_create(void)
1159 {
1160         int ret, status;
1161         char **rsync_argv;
1162
1163         if (conf.dry_run_given) {
1164                 int i;
1165                 char *msg = NULL;
1166                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1167                 for (i = 0; rsync_argv[i]; i++) {
1168                         char *tmp = msg;
1169                         msg = make_message("%s%s%s", tmp? tmp : "",
1170                                 tmp? " " : "", rsync_argv[i]);
1171                         free(tmp);
1172                 }
1173                 free_rsync_argv(rsync_argv);
1174                 dss_msg("%s\n", msg);
1175                 free(msg);
1176                 return 1;
1177         }
1178         ret = pre_create_hook();
1179         if (ret < 0)
1180                 return ret;
1181         if (create_pid) {
1182                 ret = wait_for_process(create_pid, &status);
1183                 if (ret < 0)
1184                         return ret;
1185                 ret = handle_pre_create_hook_exit(status);
1186                 if (ret <= 0) /* error, or pre-create failed */
1187                         return ret;
1188         }
1189         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1190         ret = create_snapshot(rsync_argv);
1191         if (ret < 0)
1192                 goto out;
1193         ret = wait_for_process(create_pid, &status);
1194         if (ret < 0)
1195                 goto out;
1196         ret = handle_rsync_exit(status);
1197         if (ret < 0)
1198                 goto out;
1199         post_create_hook();
1200         if (create_pid)
1201                 ret = wait_for_process(create_pid, &status);
1202 out:
1203         free_rsync_argv(rsync_argv);
1204         return ret;
1205 }
1206
1207 static int com_ls(void)
1208 {
1209         int i;
1210         struct snapshot_list sl;
1211         struct snapshot *s;
1212
1213         dss_get_snapshot_list(&sl);
1214         FOR_EACH_SNAPSHOT(s, i, &sl) {
1215                 int64_t d = 0;
1216                 if (s->flags & SS_COMPLETE)
1217                         d = (s->completion_time - s->creation_time) / 60;
1218                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1219         };
1220         free_snapshot_list(&sl);
1221         return 1;
1222 }
1223
1224 static int setup_signal_handling(void)
1225 {
1226         int ret;
1227
1228         DSS_INFO_LOG("setting up signal handlers\n");
1229         signal_pipe = signal_init(); /* always successful */
1230         ret = install_sighandler(SIGINT);
1231         if (ret < 0)
1232                 return ret;
1233         ret = install_sighandler(SIGTERM);
1234         if (ret < 0)
1235                 return ret;
1236         return install_sighandler(SIGCHLD);
1237 }
1238
1239 /**
1240  * The main function of dss.
1241  *
1242  * \param argc Usual argument count.
1243  * \param argv Usual argument vector.
1244  */
1245 int main(int argc, char **argv)
1246 {
1247         int ret;
1248         struct cmdline_parser_params params = {
1249                 .override = 0,
1250                 .initialize = 1,
1251                 .check_required = 0,
1252                 .check_ambiguity = 0,
1253                 .print_errors = 1
1254         };
1255
1256         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1257         ret = parse_config_file(0);
1258         if (ret < 0)
1259                 goto out;
1260         if (conf.daemon_given)
1261                 daemon_init();
1262         ret = change_to_dest_dir();
1263         if (ret < 0)
1264                 goto out;
1265         ret = setup_signal_handling();
1266         if (ret < 0)
1267                 goto out;
1268         ret = call_command_handler();
1269 out:
1270         if (ret < 0)
1271                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1272         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1273 }