dss.c

   1 /*
   2  * Copyright (C) 2008-2009 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 static char *name_of_reference_snapshot;
  60 /** \sa \ref snap.h for details. */
  61 enum hook_status snapshot_creation_status;
  62 /** \sa \ref snap.h for details. */
  63 enum hook_status snapshot_removal_status;
  64
  65
  66 DEFINE_DSS_ERRLIST;
  67
  68
  69 /* a litte cpp magic helps to DRY */
  70 #define COMMANDS \
  71         COMMAND(ls) \
  72         COMMAND(create) \
  73         COMMAND(prune) \
  74         COMMAND(run)
  75 #define COMMAND(x) static int com_ ##x(void);
  76 COMMANDS
  77 #undef COMMAND
  78 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
  79 static int call_command_handler(void)
  80 {
  81         COMMANDS
  82         DSS_EMERG_LOG("BUG: did not find command handler\n");
  83         return -E_BUG;
  84 }
  85 #undef COMMAND
  86 #undef COMMANDS
  87
  88 /**
  89  * The log function of dss.
  90  *
  91  * \param ll Loglevel.
  92  * \param fml Usual format string.
  93  *
  94  * All DSS_XXX_LOG() macros use this function.
  95  */
  96 __printf_2_3 void dss_log(int ll, const char* fmt,...)
  97 {
  98         va_list argp;
  99         FILE *outfd;
 100         struct tm *tm;
 101         time_t t1;
 102         char str[255] = "";
 103
 104         if (ll < conf.loglevel_arg)
 105                 return;
 106         outfd = logfile? logfile : stderr;
 107         time(&t1);
 108         tm = localtime(&t1);
 109         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 110         fprintf(outfd, "%s ", str);
 111         if (conf.loglevel_arg <= INFO)
 112                 fprintf(outfd, "%i: ", ll);
 113         va_start(argp, fmt);
 114         vfprintf(outfd, fmt, argp);
 115         va_end(argp);
 116 }
 117
 118 /**
 119  * Print a message either to stdout or to the log file.
 120  */
 121 static __printf_1_2 void dss_msg(const char* fmt,...)
 122 {
 123         FILE *outfd = conf.daemon_given? logfile : stdout;
 124         va_list argp;
 125         va_start(argp, fmt);
 126         vfprintf(outfd, fmt, argp);
 127         va_end(argp);
 128 }
 129
 130 static int disk_space_low(void)
 131 {
 132         struct disk_space ds;
 133         int ret = get_disk_space(".", &ds);
 134
 135         if (ret < 0)
 136                 return ret;
 137         if (conf.min_free_mb_arg)
 138                 if (ds.free_mb < conf.min_free_mb_arg)
 139                         return 1;
 140         if (conf.min_free_percent_arg)
 141                 if (ds.percent_free < conf.min_free_percent_arg)
 142                         return 1;
 143         if (conf.min_free_percent_inodes_arg)
 144                 if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg)
 145                         return 1;
 146         return 0;
 147 }
 148
 149 static void dss_get_snapshot_list(struct snapshot_list *sl)
 150 {
 151         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 152 }
 153
 154 static int next_snapshot_is_due(void)
 155 {
 156         struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 157                 tmp, diff;
 158         int64_t x = 0;
 159         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 160                 num_complete_snapshots = 0;
 161         int i, ret;
 162         struct snapshot *s = NULL;
 163         struct snapshot_list sl;
 164
 165         gettimeofday(&now, NULL);
 166         if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 167                 return 0;
 168         assert(snapshot_creation_status == HS_READY);
 169         current_snapshot_creation_time = 0;
 170         dss_get_snapshot_list(&sl);
 171         FOR_EACH_SNAPSHOT(s, i, &sl) {
 172                 if (!(s->flags & SS_COMPLETE))
 173                         continue;
 174                 num_complete_snapshots++;
 175                 x += s->completion_time - s->creation_time;
 176         }
 177         assert(x >= 0);
 178         if (num_complete_snapshots)
 179                 x /= num_complete_snapshots; /* avg time to create one snapshot */
 180         x *= wanted; /* time to create all snapshots in interval 0 */
 181         tmp.tv_sec = x;
 182         tmp.tv_usec = 0;
 183         ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
 184         if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
 185                 ret = 1;
 186                 goto out;
 187         }
 188
 189         tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */
 190         diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
 191         diff.tv_usec = 0;
 192         tv_add(&diff, &tmp, &next_snapshot_time);
 193         ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
 194 out:
 195         free_snapshot_list(&sl);
 196         if (ret == 0)
 197                 DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
 198                         tv2ms(&diff) / 1000);
 199         else
 200                 DSS_DEBUG_LOG("next snapshot: now\n");
 201         return ret;
 202 }
 203
 204 static int pre_create_hook(void)
 205 {
 206         int ret, fds[3] = {0, 0, 0};
 207
 208         assert(snapshot_creation_status == HS_READY);
 209         if (!conf.pre_create_hook_given) {
 210                 snapshot_creation_status = HS_PRE_SUCCESS;
 211                 return 0;
 212         }
 213         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 214         ret = dss_exec_cmdline_pid(&create_pid,
 215                 conf.pre_create_hook_arg, fds);
 216         if (ret < 0)
 217                 return ret;
 218         snapshot_creation_status = HS_PRE_RUNNING;
 219         return ret;
 220 }
 221
 222 static int pre_remove_hook(struct snapshot *s, const char *why)
 223 {
 224         int ret, fds[3] = {0, 0, 0};
 225         char *cmd;
 226
 227         if (!s)
 228                 return 0;
 229         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 230         assert(snapshot_removal_status == HS_READY);
 231         assert(remove_pid == 0);
 232         assert(!snapshot_currently_being_removed);
 233
 234         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 235         *snapshot_currently_being_removed = *s;
 236         snapshot_currently_being_removed->name = dss_strdup(s->name);
 237
 238         if (!conf.pre_remove_hook_given) {
 239                 snapshot_removal_status = HS_PRE_SUCCESS;
 240                 return 0;
 241         }
 242         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 243                 conf.dest_dir_arg, s->name);
 244         DSS_DEBUG_LOG("executing %s\n", cmd);
 245         ret = dss_exec_cmdline_pid(&remove_pid,
 246                 conf.pre_remove_hook_arg, fds);
 247         free(cmd);
 248         if (ret < 0)
 249                 return ret;
 250         snapshot_removal_status = HS_PRE_RUNNING;
 251         return ret;
 252 }
 253
 254 static int exec_rm(void)
 255 {
 256         struct snapshot *s = snapshot_currently_being_removed;
 257         int fds[3] = {0, 0, 0};
 258         char *new_name = being_deleted_name(s);
 259         char *argv[] = {"rm", "-rf", new_name, NULL};
 260         int ret;
 261
 262         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 263         assert(remove_pid == 0);
 264
 265         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 266         ret = dss_rename(s->name, new_name);
 267         if (ret < 0)
 268                 goto out;
 269         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 270         if (ret < 0)
 271                 goto out;
 272         snapshot_removal_status = HS_RUNNING;
 273 out:
 274         free(new_name);
 275         return ret;
 276 }
 277
 278 static int snapshot_is_being_created(struct snapshot *s)
 279 {
 280         return s->creation_time == current_snapshot_creation_time;
 281 }
 282
 283 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 284 {
 285         struct snapshot *s;
 286         int i;
 287
 288         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 289         FOR_EACH_SNAPSHOT(s, i, sl) {
 290                 if (snapshot_is_being_created(s))
 291                         continue;
 292                 /*
 293                  * We know that no rm is currently running, so if s is marked
 294                  * as being deleted, a previously started rm must have failed.
 295                  */
 296                 if (s->flags & SS_BEING_DELETED)
 297                         return s;
 298
 299                 if (s->flags & SS_COMPLETE) /* good snapshot */
 300                         continue;
 301                 /*
 302                  * This snapshot is incomplete and it is not the snapshot
 303                  * currently being created. However, we must not remove it if
 304                  * rsync is about to be restarted. As only the newest snapshot
 305                  * can be restarted, this snapshot is orphaned if it is not the
 306                  * newest snapshot or if we are not about to restart rsync.
 307                  */
 308                 if (get_newest_snapshot(sl) != s)
 309                         return s;
 310                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 311                         return s;
 312         }
 313         /* no orphaned snapshots */
 314         return NULL;
 315 }
 316
 317 static int is_reference_snapshot(struct snapshot *s)
 318 {
 319         if (!name_of_reference_snapshot)
 320                 return 0;
 321         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 322 }
 323
 324 /*
 325  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 326  */
 327 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 328 {
 329         int i, interval;
 330         struct snapshot *s;
 331         unsigned missing = 0;
 332
 333         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 334         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 335                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 336                 unsigned num = sl->interval_count[interval];
 337                 struct snapshot *victim = NULL, *prev = NULL;
 338                 int64_t score = LONG_MAX;
 339
 340                 if (keep >= num)
 341                         missing += keep - num;
 342 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 343 //                      interval, keep, num, missing);
 344                 if (keep + missing >= num)
 345                         continue;
 346                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 347                 FOR_EACH_SNAPSHOT(s, i, sl) {
 348                         int64_t this_score;
 349
 350                         if (snapshot_is_being_created(s))
 351                                 continue;
 352                         if (is_reference_snapshot(s))
 353                                 continue;
 354                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 355                         if (s->interval > interval) {
 356                                 prev = s;
 357                                 continue;
 358                         }
 359                         if (s->interval < interval)
 360                                 break;
 361                         if (!victim) {
 362                                 victim = s;
 363                                 prev = s;
 364                                 continue;
 365                         }
 366                         assert(prev);
 367                         /* check if s is a better victim */
 368                         this_score = s->creation_time - prev->creation_time;
 369                         assert(this_score >= 0);
 370                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 371                         if (this_score < score) {
 372                                 score = this_score;
 373                                 victim = s;
 374                         }
 375                         prev = s;
 376                 }
 377                 assert(victim);
 378                 return victim;
 379         }
 380         return NULL;
 381 }
 382
 383 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 384 {
 385         int i;
 386         struct snapshot *s;
 387
 388         DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n",
 389                 conf.num_intervals_arg);
 390         FOR_EACH_SNAPSHOT(s, i, sl) {
 391                 if (snapshot_is_being_created(s))
 392                         continue;
 393                 if (is_reference_snapshot(s))
 394                         continue;
 395                 if (s->interval <= conf.num_intervals_arg)
 396                         continue;
 397                 return s;
 398         }
 399         return NULL;
 400 }
 401
 402 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 403 {
 404         int i;
 405         struct snapshot *s;
 406         FOR_EACH_SNAPSHOT(s, i, sl) {
 407                 if (snapshot_is_being_created(s))
 408                         continue;
 409                 if (is_reference_snapshot(s))
 410                         continue;
 411                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 412                 return s;
 413         }
 414         return NULL;
 415 }
 416
 417 static int rename_incomplete_snapshot(int64_t start)
 418 {
 419         char *old_name;
 420         int ret;
 421
 422         free(path_to_last_complete_snapshot);
 423         ret = complete_name(start, get_current_time(),
 424                 &path_to_last_complete_snapshot);
 425         if (ret < 0)
 426                 return ret;
 427         old_name = incomplete_name(start);
 428         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 429         if (ret >= 0)
 430                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 431                         path_to_last_complete_snapshot);
 432         free(old_name);
 433         return ret;
 434 }
 435
 436 static int try_to_free_disk_space(void)
 437 {
 438         int ret;
 439         struct snapshot_list sl;
 440         struct snapshot *victim;
 441         struct timeval now;
 442         const char *why;
 443         int low_disk_space;
 444
 445         ret = disk_space_low();
 446         if (ret < 0)
 447                 return ret;
 448         low_disk_space = ret;
 449         gettimeofday(&now, NULL);
 450         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 451                 return 0;
 452         if (!low_disk_space && conf.keep_redundant_given)
 453                 return 0;
 454         dss_get_snapshot_list(&sl);
 455         ret = 0;
 456         if (!low_disk_space && sl.num_snapshots <= 1)
 457                 goto out;
 458         why = "outdated";
 459         victim = find_outdated_snapshot(&sl);
 460         if (victim)
 461                 goto remove;
 462         why = "redundant";
 463         victim = find_redundant_snapshot(&sl);
 464         if (victim)
 465                 goto remove;
 466         /* try harder only if disk space is low */
 467         if (!low_disk_space)
 468                 goto out;
 469         why = "orphaned";
 470         victim = find_orphaned_snapshot(&sl);
 471         if (victim)
 472                 goto remove;
 473         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 474         victim = find_oldest_removable_snapshot(&sl);
 475         if (victim)
 476                 goto remove;
 477         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 478         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 479         goto out;
 480 remove:
 481         ret = pre_remove_hook(victim, why);
 482 out:
 483         free_snapshot_list(&sl);
 484         return ret;
 485 }
 486
 487 static int post_create_hook(void)
 488 {
 489         int ret, fds[3] = {0, 0, 0};
 490         char *cmd;
 491
 492         if (!conf.post_create_hook_given) {
 493                 snapshot_creation_status = HS_READY;
 494                 return 0;
 495         }
 496         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 497                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 498         DSS_NOTICE_LOG("executing %s\n", cmd);
 499         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 500         free(cmd);
 501         if (ret < 0)
 502                 return ret;
 503         snapshot_creation_status = HS_POST_RUNNING;
 504         return ret;
 505 }
 506
 507 static int post_remove_hook(void)
 508 {
 509         int ret, fds[3] = {0, 0, 0};
 510         char *cmd;
 511         struct snapshot *s = snapshot_currently_being_removed;
 512
 513         assert(s);
 514
 515         if (!conf.post_remove_hook_given) {
 516                 snapshot_removal_status = HS_READY;
 517                 return 0;
 518         }
 519         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 520                 conf.dest_dir_arg, s->name);
 521         DSS_NOTICE_LOG("executing %s\n", cmd);
 522         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 523         free(cmd);
 524         if (ret < 0)
 525                 return ret;
 526         snapshot_removal_status = HS_POST_RUNNING;
 527         return ret;
 528 }
 529
 530 static void kill_process(pid_t pid)
 531 {
 532         if (!pid)
 533                 return;
 534         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 535         kill(pid, SIGTERM);
 536 }
 537
 538 static void stop_create_process(void)
 539 {
 540         if (!create_pid || create_process_stopped)
 541                 return;
 542         kill(SIGSTOP, create_pid);
 543         create_process_stopped = 1;
 544 }
 545
 546 static void restart_create_process(void)
 547 {
 548         if (!create_pid || !create_process_stopped)
 549                 return;
 550         kill (SIGCONT, create_pid);
 551         create_process_stopped = 0;
 552 }
 553
 554 /**
 555  * Print a log message about the exit status of a child.
 556  */
 557 static void log_termination_msg(pid_t pid, int status)
 558 {
 559         if (WIFEXITED(status))
 560                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 561                         WEXITSTATUS(status));
 562         else if (WIFSIGNALED(status))
 563                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 564                         WTERMSIG(status));
 565         else
 566                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 567 }
 568
 569 static int wait_for_process(pid_t pid, int *status)
 570 {
 571         int ret;
 572
 573         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 574         for (;;) {
 575                 fd_set rfds;
 576
 577                 FD_ZERO(&rfds);
 578                 FD_SET(signal_pipe, &rfds);
 579                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 580                 if (ret < 0)
 581                         break;
 582                 ret = next_signal();
 583                 if (!ret)
 584                         continue;
 585                 if (ret == SIGCHLD) {
 586                         ret = waitpid(pid, status, 0);
 587                         if (ret >= 0)
 588                                 break;
 589                         if (errno != EINTR) { /* error */
 590                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 591                                 break;
 592                         }
 593                 }
 594                 /* SIGINT or SIGTERM */
 595                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 596                 kill(pid, SIGTERM);
 597         }
 598         if (ret < 0)
 599                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 600         else
 601                 log_termination_msg(pid, *status);
 602         return ret;
 603 }
 604
 605 static void handle_pre_remove_exit(int status)
 606 {
 607         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 608                 snapshot_removal_status = HS_READY;
 609                 gettimeofday(&next_removal_check, NULL);
 610                 next_removal_check.tv_sec += 60;
 611                 return;
 612         }
 613         snapshot_removal_status = HS_PRE_SUCCESS;
 614 }
 615
 616 static int handle_rm_exit(int status)
 617 {
 618         if (!WIFEXITED(status)) {
 619                 snapshot_removal_status = HS_READY;
 620                 return -E_INVOLUNTARY_EXIT;
 621         }
 622         if (WEXITSTATUS(status)) {
 623                 snapshot_removal_status = HS_READY;
 624                 return -E_BAD_EXIT_CODE;
 625         }
 626         if (conf.post_remove_hook_given)
 627                 snapshot_removal_status = HS_SUCCESS;
 628         else
 629                 snapshot_removal_status = HS_READY;
 630         return 1;
 631 }
 632
 633 static void handle_post_remove_exit(void)
 634 {
 635         snapshot_removal_status = HS_READY;
 636 }
 637
 638 static int handle_remove_exit(int status)
 639 {
 640         int ret;
 641         struct snapshot *s = snapshot_currently_being_removed;
 642
 643         assert(s);
 644         switch (snapshot_removal_status) {
 645         case HS_PRE_RUNNING:
 646                 handle_pre_remove_exit(status);
 647                 ret = 1;
 648                 break;
 649         case HS_RUNNING:
 650                 ret = handle_rm_exit(status);
 651                 break;
 652         case HS_POST_RUNNING:
 653                 handle_post_remove_exit();
 654                 ret = 1;
 655                 break;
 656         default:
 657                 ret = -E_BUG;
 658         }
 659         if (snapshot_removal_status == HS_READY) {
 660                 free(s->name);
 661                 free(s);
 662                 snapshot_currently_being_removed = NULL;
 663         }
 664         remove_pid = 0;
 665         return ret;
 666 }
 667
 668 static int wait_for_remove_process(void)
 669 {
 670         int status, ret;
 671
 672         assert(remove_pid);
 673         assert(
 674                 snapshot_removal_status == HS_PRE_RUNNING ||
 675                 snapshot_removal_status == HS_RUNNING ||
 676                 snapshot_removal_status == HS_POST_RUNNING
 677         );
 678         ret = wait_for_process(remove_pid, &status);
 679         if (ret < 0)
 680                 return ret;
 681         return handle_remove_exit(status);
 682 }
 683
 684 static int handle_rsync_exit(int status)
 685 {
 686         int es, ret;
 687
 688         if (!WIFEXITED(status)) {
 689                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 690                 ret = -E_INVOLUNTARY_EXIT;
 691                 snapshot_creation_status = HS_READY;
 692                 goto out;
 693         }
 694         es = WEXITSTATUS(status);
 695         /*
 696          * Restart rsync on non-fatal errors:
 697          * 12: Error in rsync protocol data stream
 698          * 13: Errors with program diagnostics
 699          */
 700         if (es == 12 || es == 13) {
 701                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 702                         (int)create_pid, es);
 703                 snapshot_creation_status = HS_NEEDS_RESTART;
 704                 gettimeofday(&next_snapshot_time, NULL);
 705                 next_snapshot_time.tv_sec += 60;
 706                 ret = 1;
 707                 goto out;
 708         }
 709         if (es != 0 && es != 23 && es != 24) {
 710                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 711                 ret = -E_BAD_EXIT_CODE;
 712                 snapshot_creation_status = HS_READY;
 713                 goto out;
 714         }
 715         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 716         if (ret < 0)
 717                 goto out;
 718         snapshot_creation_status = HS_SUCCESS;
 719         free(name_of_reference_snapshot);
 720         name_of_reference_snapshot = NULL;
 721 out:
 722         create_pid = 0;
 723         create_process_stopped = 0;
 724         return ret;
 725 }
 726
 727 static int handle_pre_create_hook_exit(int status)
 728 {
 729         int es, ret;
 730         static int warn_count;
 731
 732         if (!WIFEXITED(status)) {
 733                 snapshot_creation_status = HS_READY;
 734                 ret = -E_INVOLUNTARY_EXIT;
 735                 goto out;
 736         }
 737         es = WEXITSTATUS(status);
 738         if (es) {
 739                 if (!warn_count--) {
 740                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 741                                 conf.pre_create_hook_arg, es);
 742                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 743                         warn_count = 60; /* warn only once per hour */
 744                 }
 745                 gettimeofday(&next_snapshot_time, NULL);
 746                 next_snapshot_time.tv_sec += 60;
 747                 snapshot_creation_status = HS_READY;
 748                 ret = 0;
 749                 goto out;
 750         }
 751         warn_count = 0;
 752         snapshot_creation_status = HS_PRE_SUCCESS;
 753         ret = 1;
 754 out:
 755         create_pid = 0;
 756         return ret;
 757 }
 758
 759 static int handle_sigchld(void)
 760 {
 761         pid_t pid;
 762         int status, ret = reap_child(&pid, &status);
 763
 764         if (ret <= 0)
 765                 return ret;
 766
 767         if (pid == create_pid) {
 768                 switch (snapshot_creation_status) {
 769                 case HS_PRE_RUNNING:
 770                         return handle_pre_create_hook_exit(status);
 771                 case HS_RUNNING:
 772                         return handle_rsync_exit(status);
 773                 case HS_POST_RUNNING:
 774                         snapshot_creation_status = HS_READY;
 775                         return 1;
 776                 default:
 777                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 778                                 snapshot_creation_status);
 779                         return -E_BUG;
 780                 }
 781         }
 782         if (pid == remove_pid) {
 783                 ret = handle_remove_exit(status);
 784                 if (ret < 0)
 785                         return ret;
 786                 return ret;
 787         }
 788         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 789         return -E_BUG;
 790 }
 791
 792 static int check_config(void)
 793 {
 794         if (conf.unit_interval_arg <= 0) {
 795                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 796                 return -E_INVALID_NUMBER;
 797         }
 798         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 799         if (conf.num_intervals_arg <= 0) {
 800                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 801                 return -E_INVALID_NUMBER;
 802         }
 803         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 804         return 1;
 805 }
 806
 807 static int parse_config_file(int override)
 808 {
 809         int ret;
 810         char *config_file;
 811         struct stat statbuf;
 812         char *old_logfile_arg = NULL;
 813         int old_daemon_given = 0;
 814
 815         if (conf.config_file_given)
 816                 config_file = dss_strdup(conf.config_file_arg);
 817         else {
 818                 char *home = get_homedir();
 819                 config_file = make_message("%s/.dssrc", home);
 820                 free(home);
 821         }
 822         if (override) { /* SIGHUP */
 823                 if (conf.logfile_given)
 824                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 825                 old_daemon_given = conf.daemon_given;
 826         }
 827
 828         ret = stat(config_file, &statbuf);
 829         if (ret && conf.config_file_given) {
 830                 ret = -ERRNO_TO_DSS_ERROR(errno);
 831                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 832                 goto out;
 833         }
 834         if (!ret) {
 835                 struct cmdline_parser_params params = {
 836                         .override = override,
 837                         .initialize = 0,
 838                         .check_required = 1,
 839                         .check_ambiguity = 0,
 840                         .print_errors = 1
 841                 };
 842                 if (override) { /* invalidate all rsync options */
 843                         int i;
 844
 845                         for (i = 0; i < conf.rsync_option_given; i++) {
 846                                 free(conf.rsync_option_arg[i]);
 847                                 conf.rsync_option_arg[i] = NULL;
 848                         }
 849                         conf.rsync_option_given = 0;
 850                 }
 851                 cmdline_parser_config_file(config_file, &conf, &params);
 852         }
 853         ret = check_config();
 854         if (ret < 0)
 855                 goto out;
 856         if (override) {
 857                 /* don't change daemon mode on SIGHUP */
 858                 conf.daemon_given = old_daemon_given;
 859                 close_log(logfile);
 860                 logfile = NULL;
 861                 if (conf.logfile_given)
 862                         free(old_logfile_arg);
 863                 else if (conf.daemon_given) { /* re-use old logfile */
 864                         conf.logfile_arg = old_logfile_arg;
 865                         conf.logfile_given = 1;
 866                 }
 867         }
 868         if (conf.logfile_given) {
 869                 logfile = open_log(conf.logfile_arg);
 870                 log_welcome(conf.loglevel_arg);
 871         }
 872         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 873 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 874 out:
 875         free(config_file);
 876         if (ret < 0)
 877                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 878         return ret;
 879 }
 880
 881 static int change_to_dest_dir(void)
 882 {
 883         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 884         return dss_chdir(conf.dest_dir_arg);
 885 }
 886
 887 static int handle_sighup(void)
 888 {
 889         int ret;
 890
 891         DSS_NOTICE_LOG("SIGHUP\n");
 892         ret = parse_config_file(1);
 893         if (ret < 0)
 894                 return ret;
 895         return change_to_dest_dir();
 896 }
 897
 898 static int handle_signal(void)
 899 {
 900         int sig, ret = next_signal();
 901
 902         if (ret <= 0)
 903                 goto out;
 904         sig = ret;
 905         switch (sig) {
 906         case SIGINT:
 907         case SIGTERM:
 908                 restart_create_process();
 909                 kill_process(create_pid);
 910                 kill_process(remove_pid);
 911                 ret = -E_SIGNAL;
 912                 break;
 913         case SIGHUP:
 914                 ret = handle_sighup();
 915                 break;
 916         case SIGCHLD:
 917                 ret = handle_sigchld();
 918                 break;
 919         }
 920 out:
 921         if (ret < 0)
 922                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 923         return ret;
 924 }
 925
 926 /*
 927  * We can not use rsync locally if the local user is different from the remote
 928  * user or if the src dir is not on the local host (or both).
 929  */
 930 static int use_rsync_locally(char *logname)
 931 {
 932         char *h = conf.remote_host_arg;
 933
 934         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 935                 return 0;
 936         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 937                 return 0;
 938         return 1;
 939 }
 940
 941 static void create_rsync_argv(char ***argv, int64_t *num)
 942 {
 943         char *logname;
 944         int i = 0, j;
 945         struct snapshot_list sl;
 946
 947         dss_get_snapshot_list(&sl);
 948         assert(!name_of_reference_snapshot);
 949         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
 950         free_snapshot_list(&sl);
 951
 952         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 953         (*argv)[i++] = dss_strdup("rsync");
 954         (*argv)[i++] = dss_strdup("-aq");
 955         (*argv)[i++] = dss_strdup("--delete");
 956         for (j = 0; j < conf.rsync_option_given; j++)
 957                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 958         if (name_of_reference_snapshot) {
 959                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
 960                 (*argv)[i++] = make_message("--link-dest=../%s",
 961                         name_of_reference_snapshot);
 962         } else
 963                 DSS_INFO_LOG("no suitable reference snapshot found\n");
 964         logname = dss_logname();
 965         if (use_rsync_locally(logname))
 966                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 967         else
 968                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 969                         conf.remote_user_arg : logname,
 970                         conf.remote_host_arg, conf.source_dir_arg);
 971         free(logname);
 972         *num = get_current_time();
 973         (*argv)[i++] = incomplete_name(*num);
 974         (*argv)[i++] = NULL;
 975         for (j = 0; j < i; j++)
 976                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 977 }
 978
 979 static void free_rsync_argv(char **argv)
 980 {
 981         int i;
 982
 983         if (!argv)
 984                 return;
 985         for (i = 0; argv[i]; i++)
 986                 free(argv[i]);
 987         free(argv);
 988 }
 989
 990 static int create_snapshot(char **argv)
 991 {
 992         int ret, fds[3] = {0, 0, 0};
 993         char *name;
 994
 995         name = incomplete_name(current_snapshot_creation_time);
 996         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 997         free(name);
 998         ret = dss_exec(&create_pid, argv[0], argv, fds);
 999         if (ret < 0)
1000                 return ret;
1001         snapshot_creation_status = HS_RUNNING;
1002         return ret;
1003 }
1004
1005 static int select_loop(void)
1006 {
1007         int ret;
1008         /* check every 60 seconds for free disk space */
1009         struct timeval tv;
1010         char **rsync_argv = NULL;
1011
1012         for (;;) {
1013                 fd_set rfds;
1014                 struct timeval *tvp;
1015
1016                 if (remove_pid)
1017                         tvp = NULL; /* sleep until rm hook/process dies */
1018                 else { /* sleep one minute */
1019                         tv.tv_sec = 60;
1020                         tv.tv_usec = 0;
1021                         tvp = &tv;
1022                 }
1023                 FD_ZERO(&rfds);
1024                 FD_SET(signal_pipe, &rfds);
1025                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1026                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1027                 if (ret < 0)
1028                         goto out;
1029                 if (FD_ISSET(signal_pipe, &rfds)) {
1030                         ret = handle_signal();
1031                         if (ret < 0)
1032                                 goto out;
1033                 }
1034                 if (remove_pid)
1035                         continue;
1036                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1037                         ret = exec_rm();
1038                         if (ret < 0)
1039                                 goto out;
1040                         continue;
1041                 }
1042                 if (snapshot_removal_status == HS_SUCCESS) {
1043                         ret = post_remove_hook();
1044                         if (ret < 0)
1045                                 goto out;
1046                         continue;
1047                 }
1048                 ret = try_to_free_disk_space();
1049                 if (ret < 0)
1050                         goto out;
1051                 if (snapshot_removal_status != HS_READY) {
1052                         stop_create_process();
1053                         continue;
1054                 }
1055                 restart_create_process();
1056                 switch (snapshot_creation_status) {
1057                 case HS_READY:
1058                         if (!next_snapshot_is_due())
1059                                 continue;
1060                         ret = pre_create_hook();
1061                         if (ret < 0)
1062                                 goto out;
1063                         continue;
1064                 case HS_PRE_RUNNING:
1065                 case HS_RUNNING:
1066                 case HS_POST_RUNNING:
1067                         continue;
1068                 case HS_PRE_SUCCESS:
1069                         free_rsync_argv(rsync_argv);
1070                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1071                         /* fall through */
1072                 case HS_NEEDS_RESTART:
1073                         ret = create_snapshot(rsync_argv);
1074                         if (ret < 0)
1075                                 goto out;
1076                         continue;
1077                 case HS_SUCCESS:
1078                         ret = post_create_hook();
1079                         if (ret < 0)
1080                                 goto out;
1081                         continue;
1082                 }
1083         }
1084 out:
1085         return ret;
1086 }
1087
1088 static void exit_hook(int exit_code)
1089 {
1090         int fds[3] = {0, 0, 0};
1091         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1092         pid_t pid;
1093
1094         if (!conf.exit_hook_given)
1095                 return;
1096         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1097         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1098 }
1099
1100 static int com_run(void)
1101 {
1102         int ret;
1103
1104         if (conf.dry_run_given) {
1105                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1106                 return -E_SYNTAX;
1107         }
1108         ret = install_sighandler(SIGHUP);
1109         if (ret < 0)
1110                 return ret;
1111         ret = select_loop();
1112         if (ret >= 0) /* impossible */
1113                 ret = -E_BUG;
1114         exit_hook(ret);
1115         return ret;
1116 }
1117
1118 static int com_prune(void)
1119 {
1120         int ret;
1121         struct snapshot_list sl;
1122         struct snapshot *victim;
1123         struct disk_space ds;
1124         const char *why;
1125
1126         ret = get_disk_space(".", &ds);
1127         if (ret < 0)
1128                 return ret;
1129         log_disk_space(&ds);
1130         dss_get_snapshot_list(&sl);
1131         why = "outdated";
1132         victim = find_outdated_snapshot(&sl);
1133         if (victim)
1134                 goto rm;
1135         why = "redundant";
1136         victim = find_redundant_snapshot(&sl);
1137         if (victim)
1138                 goto rm;
1139         ret = 0;
1140         goto out;
1141 rm:
1142         if (conf.dry_run_given) {
1143                 dss_msg("%s snapshot %s (interval = %i)\n",
1144                         why, victim->name, victim->interval);
1145                 ret = 0;
1146                 goto out;
1147         }
1148         ret = pre_remove_hook(victim, why);
1149         if (ret < 0)
1150                 goto out;
1151         if (snapshot_removal_status == HS_PRE_RUNNING) {
1152                 ret = wait_for_remove_process();
1153                 if (ret < 0)
1154                         goto out;
1155                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1156                         goto out;
1157         }
1158         ret = exec_rm();
1159         if (ret < 0)
1160                 goto out;
1161         ret = wait_for_remove_process();
1162         if (ret < 0)
1163                 goto out;
1164         if (snapshot_removal_status != HS_SUCCESS)
1165                 goto out;
1166         ret = post_remove_hook();
1167         if (ret < 0)
1168                 goto out;
1169         if (snapshot_removal_status != HS_POST_RUNNING)
1170                 goto out;
1171         ret = wait_for_remove_process();
1172         if (ret < 0)
1173                 goto out;
1174         ret = 1;
1175 out:
1176         free_snapshot_list(&sl);
1177         return ret;
1178 }
1179
1180 static int com_create(void)
1181 {
1182         int ret, status;
1183         char **rsync_argv;
1184
1185         if (conf.dry_run_given) {
1186                 int i;
1187                 char *msg = NULL;
1188                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1189                 for (i = 0; rsync_argv[i]; i++) {
1190                         char *tmp = msg;
1191                         msg = make_message("%s%s%s", tmp? tmp : "",
1192                                 tmp? " " : "", rsync_argv[i]);
1193                         free(tmp);
1194                 }
1195                 free_rsync_argv(rsync_argv);
1196                 dss_msg("%s\n", msg);
1197                 free(msg);
1198                 return 1;
1199         }
1200         ret = pre_create_hook();
1201         if (ret < 0)
1202                 return ret;
1203         if (create_pid) {
1204                 ret = wait_for_process(create_pid, &status);
1205                 if (ret < 0)
1206                         return ret;
1207                 ret = handle_pre_create_hook_exit(status);
1208                 if (ret <= 0) /* error, or pre-create failed */
1209                         return ret;
1210         }
1211         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1212         ret = create_snapshot(rsync_argv);
1213         if (ret < 0)
1214                 goto out;
1215         ret = wait_for_process(create_pid, &status);
1216         if (ret < 0)
1217                 goto out;
1218         ret = handle_rsync_exit(status);
1219         if (ret < 0)
1220                 goto out;
1221         post_create_hook();
1222         if (create_pid)
1223                 ret = wait_for_process(create_pid, &status);
1224 out:
1225         free_rsync_argv(rsync_argv);
1226         return ret;
1227 }
1228
1229 static int com_ls(void)
1230 {
1231         int i;
1232         struct snapshot_list sl;
1233         struct snapshot *s;
1234
1235         dss_get_snapshot_list(&sl);
1236         FOR_EACH_SNAPSHOT(s, i, &sl) {
1237                 int64_t d = 0;
1238                 if (s->flags & SS_COMPLETE)
1239                         d = (s->completion_time - s->creation_time) / 60;
1240                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1241         };
1242         free_snapshot_list(&sl);
1243         return 1;
1244 }
1245
1246 static int setup_signal_handling(void)
1247 {
1248         int ret;
1249
1250         DSS_INFO_LOG("setting up signal handlers\n");
1251         signal_pipe = signal_init(); /* always successful */
1252         ret = install_sighandler(SIGINT);
1253         if (ret < 0)
1254                 return ret;
1255         ret = install_sighandler(SIGTERM);
1256         if (ret < 0)
1257                 return ret;
1258         return install_sighandler(SIGCHLD);
1259 }
1260
1261 /**
1262  * The main function of dss.
1263  *
1264  * \param argc Usual argument count.
1265  * \param argv Usual argument vector.
1266  */
1267 int main(int argc, char **argv)
1268 {
1269         int ret;
1270         struct cmdline_parser_params params = {
1271                 .override = 0,
1272                 .initialize = 1,
1273                 .check_required = 0,
1274                 .check_ambiguity = 0,
1275                 .print_errors = 1
1276         };
1277
1278         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1279         ret = parse_config_file(0);
1280         if (ret < 0)
1281                 goto out;
1282         if (conf.daemon_given)
1283                 daemon_init();
1284         ret = change_to_dest_dir();
1285         if (ret < 0)
1286                 goto out;
1287         ret = setup_signal_handling();
1288         if (ret < 0)
1289                 goto out;
1290         ret = call_command_handler();
1291 out:
1292         if (ret < 0)
1293                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1294         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1295 }