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 int64_t 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 int64_t compute_next_snapshot_time(void)
 155 {
 156         int64_t x = 0, now = get_current_time(), unit_interval
 157                 = 24 * 3600 * conf.unit_interval_arg, ret;
 158         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 159                 num_complete_snapshots = 0;
 160         int i;
 161         struct snapshot *s = NULL;
 162         struct snapshot_list sl;
 163
 164         dss_get_snapshot_list(&sl);
 165         FOR_EACH_SNAPSHOT(s, i, &sl) {
 166                 if (!(s->flags & SS_COMPLETE))
 167                         continue;
 168                 num_complete_snapshots++;
 169                 x += s->completion_time - s->creation_time;
 170         }
 171         assert(x >= 0);
 172
 173         ret = now;
 174         if (num_complete_snapshots == 0)
 175                 goto out;
 176         x /= num_complete_snapshots; /* avg time to create one snapshot */
 177         if (unit_interval < x * wanted) /* oops, no sleep at all */
 178                 goto out;
 179         ret = s->completion_time + unit_interval / wanted - x;
 180 out:
 181         free_snapshot_list(&sl);
 182         return ret;
 183 }
 184
 185 static inline void invalidate_next_snapshot_time(void)
 186 {
 187         next_snapshot_time = 0;
 188 }
 189
 190 static inline int next_snapshot_time_is_valid(void)
 191 {
 192         return next_snapshot_time != 0;
 193 }
 194
 195 static int next_snapshot_is_due(void)
 196 {
 197         int64_t now = get_current_time();
 198
 199         if (!next_snapshot_time_is_valid())
 200                 next_snapshot_time = compute_next_snapshot_time();
 201         if (next_snapshot_time <= now) {
 202                 DSS_DEBUG_LOG("next snapshot: now\n");
 203                 return 1;
 204         }
 205         DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
 206                 next_snapshot_time - now);
 207         return 0;
 208 }
 209
 210 static int pre_create_hook(void)
 211 {
 212         int ret, fds[3] = {0, 0, 0};
 213
 214         assert(snapshot_creation_status == HS_READY);
 215         /* make sure that the next snapshot time will be recomputed */
 216         invalidate_next_snapshot_time();
 217         if (!conf.pre_create_hook_given) {
 218                 snapshot_creation_status = HS_PRE_SUCCESS;
 219                 return 0;
 220         }
 221         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 222         ret = dss_exec_cmdline_pid(&create_pid,
 223                 conf.pre_create_hook_arg, fds);
 224         if (ret < 0)
 225                 return ret;
 226         snapshot_creation_status = HS_PRE_RUNNING;
 227         return ret;
 228 }
 229
 230 static int pre_remove_hook(struct snapshot *s, const char *why)
 231 {
 232         int ret, fds[3] = {0, 0, 0};
 233         char *cmd;
 234
 235         if (!s)
 236                 return 0;
 237         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 238         assert(snapshot_removal_status == HS_READY);
 239         assert(remove_pid == 0);
 240         assert(!snapshot_currently_being_removed);
 241
 242         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 243         *snapshot_currently_being_removed = *s;
 244         snapshot_currently_being_removed->name = dss_strdup(s->name);
 245
 246         if (!conf.pre_remove_hook_given) {
 247                 snapshot_removal_status = HS_PRE_SUCCESS;
 248                 return 0;
 249         }
 250         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 251                 conf.dest_dir_arg, s->name);
 252         DSS_DEBUG_LOG("executing %s\n", cmd);
 253         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 254         free(cmd);
 255         if (ret < 0)
 256                 return ret;
 257         snapshot_removal_status = HS_PRE_RUNNING;
 258         return ret;
 259 }
 260
 261 static int exec_rm(void)
 262 {
 263         struct snapshot *s = snapshot_currently_being_removed;
 264         int fds[3] = {0, 0, 0};
 265         char *new_name = being_deleted_name(s);
 266         char *argv[] = {"rm", "-rf", new_name, NULL};
 267         int ret;
 268
 269         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 270         assert(remove_pid == 0);
 271
 272         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 273         ret = dss_rename(s->name, new_name);
 274         if (ret < 0)
 275                 goto out;
 276         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 277         if (ret < 0)
 278                 goto out;
 279         snapshot_removal_status = HS_RUNNING;
 280 out:
 281         free(new_name);
 282         return ret;
 283 }
 284
 285 static int snapshot_is_being_created(struct snapshot *s)
 286 {
 287         return s->creation_time == current_snapshot_creation_time;
 288 }
 289
 290 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 291 {
 292         struct snapshot *s;
 293         int i;
 294
 295         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 296         FOR_EACH_SNAPSHOT(s, i, sl) {
 297                 if (snapshot_is_being_created(s))
 298                         continue;
 299                 /*
 300                  * We know that no rm is currently running, so if s is marked
 301                  * as being deleted, a previously started rm must have failed.
 302                  */
 303                 if (s->flags & SS_BEING_DELETED)
 304                         return s;
 305
 306                 if (s->flags & SS_COMPLETE) /* good snapshot */
 307                         continue;
 308                 /*
 309                  * This snapshot is incomplete and it is not the snapshot
 310                  * currently being created. However, we must not remove it if
 311                  * rsync is about to be restarted. As only the newest snapshot
 312                  * can be restarted, this snapshot is orphaned if it is not the
 313                  * newest snapshot or if we are not about to restart rsync.
 314                  */
 315                 if (get_newest_snapshot(sl) != s)
 316                         return s;
 317                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 318                         return s;
 319         }
 320         /* no orphaned snapshots */
 321         return NULL;
 322 }
 323
 324 static int is_reference_snapshot(struct snapshot *s)
 325 {
 326         if (!name_of_reference_snapshot)
 327                 return 0;
 328         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 329 }
 330
 331 /*
 332  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 333  */
 334 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 335 {
 336         int i, interval;
 337         struct snapshot *s;
 338         unsigned missing = 0;
 339
 340         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 341         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 342                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 343                 unsigned num = sl->interval_count[interval];
 344                 struct snapshot *victim = NULL, *prev = NULL;
 345                 int64_t score = LONG_MAX;
 346
 347                 if (keep >= num)
 348                         missing += keep - num;
 349 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 350 //                      interval, keep, num, missing);
 351                 if (keep + missing >= num)
 352                         continue;
 353                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 354                 FOR_EACH_SNAPSHOT(s, i, sl) {
 355                         int64_t this_score;
 356
 357                         if (snapshot_is_being_created(s))
 358                                 continue;
 359                         if (is_reference_snapshot(s))
 360                                 continue;
 361                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 362                         if (s->interval > interval) {
 363                                 prev = s;
 364                                 continue;
 365                         }
 366                         if (s->interval < interval)
 367                                 break;
 368                         if (!victim) {
 369                                 victim = s;
 370                                 prev = s;
 371                                 continue;
 372                         }
 373                         assert(prev);
 374                         /* check if s is a better victim */
 375                         this_score = s->creation_time - prev->creation_time;
 376                         assert(this_score >= 0);
 377                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 378                         if (this_score < score) {
 379                                 score = this_score;
 380                                 victim = s;
 381                         }
 382                         prev = s;
 383                 }
 384                 assert(victim);
 385                 return victim;
 386         }
 387         return NULL;
 388 }
 389
 390 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 391 {
 392         int i;
 393         struct snapshot *s;
 394
 395         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 396                 conf.num_intervals_arg);
 397         FOR_EACH_SNAPSHOT(s, i, sl) {
 398                 if (snapshot_is_being_created(s))
 399                         continue;
 400                 if (is_reference_snapshot(s))
 401                         continue;
 402                 if (s->interval < conf.num_intervals_arg)
 403                         continue;
 404                 return s;
 405         }
 406         return NULL;
 407 }
 408
 409 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 410 {
 411         int i;
 412         struct snapshot *s;
 413         FOR_EACH_SNAPSHOT(s, i, sl) {
 414                 if (snapshot_is_being_created(s))
 415                         continue;
 416                 if (is_reference_snapshot(s))
 417                         continue;
 418                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 419                 return s;
 420         }
 421         return NULL;
 422 }
 423
 424 static int rename_incomplete_snapshot(int64_t start)
 425 {
 426         char *old_name;
 427         int ret;
 428
 429         free(path_to_last_complete_snapshot);
 430         ret = complete_name(start, get_current_time(),
 431                 &path_to_last_complete_snapshot);
 432         if (ret < 0)
 433                 return ret;
 434         old_name = incomplete_name(start);
 435         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 436         if (ret >= 0)
 437                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 438                         path_to_last_complete_snapshot);
 439         free(old_name);
 440         return ret;
 441 }
 442
 443 static int try_to_free_disk_space(void)
 444 {
 445         int ret;
 446         struct snapshot_list sl;
 447         struct snapshot *victim;
 448         struct timeval now;
 449         const char *why;
 450         int low_disk_space;
 451
 452         ret = disk_space_low();
 453         if (ret < 0)
 454                 return ret;
 455         low_disk_space = ret;
 456         gettimeofday(&now, NULL);
 457         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 458                 return 0;
 459         if (!low_disk_space) {
 460                 if (conf.keep_redundant_given)
 461                         return 0;
 462                 if (snapshot_creation_status != HS_READY)
 463                         return 0;
 464                 if (next_snapshot_is_due())
 465                         return 0;
 466         }
 467         dss_get_snapshot_list(&sl);
 468         ret = 0;
 469         if (!low_disk_space && sl.num_snapshots <= 1)
 470                 goto out;
 471         why = "outdated";
 472         victim = find_outdated_snapshot(&sl);
 473         if (victim)
 474                 goto remove;
 475         why = "redundant";
 476         victim = find_redundant_snapshot(&sl);
 477         if (victim)
 478                 goto remove;
 479         /* try harder only if disk space is low */
 480         if (!low_disk_space)
 481                 goto out;
 482         why = "orphaned";
 483         victim = find_orphaned_snapshot(&sl);
 484         if (victim)
 485                 goto remove;
 486         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 487         victim = find_oldest_removable_snapshot(&sl);
 488         if (victim)
 489                 goto remove;
 490         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 491         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 492         goto out;
 493 remove:
 494         ret = pre_remove_hook(victim, why);
 495 out:
 496         free_snapshot_list(&sl);
 497         return ret;
 498 }
 499
 500 static int post_create_hook(void)
 501 {
 502         int ret, fds[3] = {0, 0, 0};
 503         char *cmd;
 504
 505         if (!conf.post_create_hook_given) {
 506                 create_pid = 0;
 507                 snapshot_creation_status = HS_READY;
 508                 return 0;
 509         }
 510         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 511                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 512         DSS_NOTICE_LOG("executing %s\n", cmd);
 513         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 514         free(cmd);
 515         if (ret < 0)
 516                 return ret;
 517         snapshot_creation_status = HS_POST_RUNNING;
 518         return ret;
 519 }
 520
 521 static int post_remove_hook(void)
 522 {
 523         int ret, fds[3] = {0, 0, 0};
 524         char *cmd;
 525         struct snapshot *s = snapshot_currently_being_removed;
 526
 527         assert(s);
 528
 529         if (!conf.post_remove_hook_given) {
 530                 snapshot_removal_status = HS_READY;
 531                 return 0;
 532         }
 533         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 534                 conf.dest_dir_arg, s->name);
 535         DSS_NOTICE_LOG("executing %s\n", cmd);
 536         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 537         free(cmd);
 538         if (ret < 0)
 539                 return ret;
 540         snapshot_removal_status = HS_POST_RUNNING;
 541         return ret;
 542 }
 543
 544 static void kill_process(pid_t pid)
 545 {
 546         if (!pid)
 547                 return;
 548         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 549         kill(pid, SIGTERM);
 550 }
 551
 552 static void stop_create_process(void)
 553 {
 554         if (!create_pid || create_process_stopped)
 555                 return;
 556         kill(SIGSTOP, create_pid);
 557         create_process_stopped = 1;
 558 }
 559
 560 static void restart_create_process(void)
 561 {
 562         if (!create_pid || !create_process_stopped)
 563                 return;
 564         kill (SIGCONT, create_pid);
 565         create_process_stopped = 0;
 566 }
 567
 568 /**
 569  * Print a log message about the exit status of a child.
 570  */
 571 static void log_termination_msg(pid_t pid, int status)
 572 {
 573         if (WIFEXITED(status))
 574                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 575                         WEXITSTATUS(status));
 576         else if (WIFSIGNALED(status))
 577                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 578                         WTERMSIG(status));
 579         else
 580                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 581 }
 582
 583 static int wait_for_process(pid_t pid, int *status)
 584 {
 585         int ret;
 586
 587         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 588         for (;;) {
 589                 fd_set rfds;
 590
 591                 FD_ZERO(&rfds);
 592                 FD_SET(signal_pipe, &rfds);
 593                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 594                 if (ret < 0)
 595                         break;
 596                 ret = next_signal();
 597                 if (!ret)
 598                         continue;
 599                 if (ret == SIGCHLD) {
 600                         ret = waitpid(pid, status, 0);
 601                         if (ret >= 0)
 602                                 break;
 603                         if (errno != EINTR) { /* error */
 604                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 605                                 break;
 606                         }
 607                 }
 608                 /* SIGINT or SIGTERM */
 609                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 610                 kill(pid, SIGTERM);
 611         }
 612         if (ret < 0)
 613                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 614         else
 615                 log_termination_msg(pid, *status);
 616         return ret;
 617 }
 618
 619 static void handle_pre_remove_exit(int status)
 620 {
 621         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 622                 snapshot_removal_status = HS_READY;
 623                 gettimeofday(&next_removal_check, NULL);
 624                 next_removal_check.tv_sec += 60;
 625                 return;
 626         }
 627         snapshot_removal_status = HS_PRE_SUCCESS;
 628 }
 629
 630 static int handle_rm_exit(int status)
 631 {
 632         if (!WIFEXITED(status)) {
 633                 snapshot_removal_status = HS_READY;
 634                 return -E_INVOLUNTARY_EXIT;
 635         }
 636         if (WEXITSTATUS(status)) {
 637                 snapshot_removal_status = HS_READY;
 638                 return -E_BAD_EXIT_CODE;
 639         }
 640         if (conf.post_remove_hook_given)
 641                 snapshot_removal_status = HS_SUCCESS;
 642         else
 643                 snapshot_removal_status = HS_READY;
 644         return 1;
 645 }
 646
 647 static void handle_post_remove_exit(void)
 648 {
 649         snapshot_removal_status = HS_READY;
 650 }
 651
 652 static int handle_remove_exit(int status)
 653 {
 654         int ret;
 655         struct snapshot *s = snapshot_currently_being_removed;
 656
 657         assert(s);
 658         switch (snapshot_removal_status) {
 659         case HS_PRE_RUNNING:
 660                 handle_pre_remove_exit(status);
 661                 ret = 1;
 662                 break;
 663         case HS_RUNNING:
 664                 ret = handle_rm_exit(status);
 665                 break;
 666         case HS_POST_RUNNING:
 667                 handle_post_remove_exit();
 668                 ret = 1;
 669                 break;
 670         default:
 671                 ret = -E_BUG;
 672         }
 673         if (snapshot_removal_status == HS_READY) {
 674                 free(s->name);
 675                 free(s);
 676                 snapshot_currently_being_removed = NULL;
 677         }
 678         remove_pid = 0;
 679         return ret;
 680 }
 681
 682 static int wait_for_remove_process(void)
 683 {
 684         int status, ret;
 685
 686         assert(remove_pid);
 687         assert(
 688                 snapshot_removal_status == HS_PRE_RUNNING ||
 689                 snapshot_removal_status == HS_RUNNING ||
 690                 snapshot_removal_status == HS_POST_RUNNING
 691         );
 692         ret = wait_for_process(remove_pid, &status);
 693         if (ret < 0)
 694                 return ret;
 695         return handle_remove_exit(status);
 696 }
 697
 698 static int handle_rsync_exit(int status)
 699 {
 700         int es, ret;
 701
 702         if (!WIFEXITED(status)) {
 703                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 704                 ret = -E_INVOLUNTARY_EXIT;
 705                 snapshot_creation_status = HS_READY;
 706                 goto out;
 707         }
 708         es = WEXITSTATUS(status);
 709         /*
 710          * Restart rsync on non-fatal errors:
 711          * 12: Error in rsync protocol data stream
 712          * 13: Errors with program diagnostics
 713          */
 714         if (es == 12 || es == 13) {
 715                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 716                         (int)create_pid, es);
 717                 snapshot_creation_status = HS_NEEDS_RESTART;
 718                 next_snapshot_time = get_current_time() + 60;
 719                 ret = 1;
 720                 goto out;
 721         }
 722         if (es != 0 && es != 23 && es != 24) {
 723                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 724                 ret = -E_BAD_EXIT_CODE;
 725                 snapshot_creation_status = HS_READY;
 726                 goto out;
 727         }
 728         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 729         if (ret < 0)
 730                 goto out;
 731         snapshot_creation_status = HS_SUCCESS;
 732         free(name_of_reference_snapshot);
 733         name_of_reference_snapshot = NULL;
 734 out:
 735         create_process_stopped = 0;
 736         return ret;
 737 }
 738
 739 static int handle_pre_create_hook_exit(int status)
 740 {
 741         int es, ret;
 742         static int warn_count;
 743
 744         if (!WIFEXITED(status)) {
 745                 snapshot_creation_status = HS_READY;
 746                 ret = -E_INVOLUNTARY_EXIT;
 747                 goto out;
 748         }
 749         es = WEXITSTATUS(status);
 750         if (es) {
 751                 if (!warn_count--) {
 752                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 753                                 conf.pre_create_hook_arg, es);
 754                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 755                         warn_count = 60; /* warn only once per hour */
 756                 }
 757                 next_snapshot_time = get_current_time() + 60;
 758                 snapshot_creation_status = HS_READY;
 759                 ret = 0;
 760                 goto out;
 761         }
 762         warn_count = 0;
 763         snapshot_creation_status = HS_PRE_SUCCESS;
 764         ret = 1;
 765 out:
 766         return ret;
 767 }
 768
 769 static int handle_sigchld(void)
 770 {
 771         pid_t pid;
 772         int status, ret = reap_child(&pid, &status);
 773
 774         if (ret <= 0)
 775                 return ret;
 776
 777         if (pid == create_pid) {
 778                 switch (snapshot_creation_status) {
 779                 case HS_PRE_RUNNING:
 780                         return handle_pre_create_hook_exit(status);
 781                 case HS_RUNNING:
 782                         return handle_rsync_exit(status);
 783                 case HS_POST_RUNNING:
 784                         snapshot_creation_status = HS_READY;
 785                         return 1;
 786                 default:
 787                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 788                                 snapshot_creation_status);
 789                         return -E_BUG;
 790                 }
 791                 create_pid = 0;
 792         }
 793         if (pid == remove_pid) {
 794                 ret = handle_remove_exit(status);
 795                 if (ret < 0)
 796                         return ret;
 797                 return ret;
 798         }
 799         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 800         return -E_BUG;
 801 }
 802
 803 static int check_config(void)
 804 {
 805         if (conf.unit_interval_arg <= 0) {
 806                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 807                 return -E_INVALID_NUMBER;
 808         }
 809         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 810         if (conf.num_intervals_arg <= 0) {
 811                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 812                 return -E_INVALID_NUMBER;
 813         }
 814         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 815         return 1;
 816 }
 817
 818 /*
 819  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 820  * file was read successfully.
 821  */
 822 static int parse_config_file(int override)
 823 {
 824         int ret, config_file_exists;
 825         char *config_file;
 826         struct stat statbuf;
 827         char *old_logfile_arg = NULL;
 828         int old_daemon_given = 0;
 829
 830         if (conf.config_file_given)
 831                 config_file = dss_strdup(conf.config_file_arg);
 832         else {
 833                 char *home = get_homedir();
 834                 config_file = make_message("%s/.dssrc", home);
 835                 free(home);
 836         }
 837         if (override) { /* SIGHUP */
 838                 if (conf.logfile_given)
 839                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 840                 old_daemon_given = conf.daemon_given;
 841         }
 842
 843         config_file_exists = !stat(config_file, &statbuf);
 844         if (!config_file_exists && conf.config_file_given) {
 845                 ret = -ERRNO_TO_DSS_ERROR(errno);
 846                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 847                 goto out;
 848         }
 849         if (config_file_exists) {
 850                 struct cmdline_parser_params params = {
 851                         .override = override,
 852                         .initialize = 0,
 853                         .check_required = 1,
 854                         .check_ambiguity = 0,
 855                         .print_errors = 1
 856                 };
 857                 if (override) { /* invalidate all rsync options */
 858                         int i;
 859
 860                         for (i = 0; i < conf.rsync_option_given; i++) {
 861                                 free(conf.rsync_option_arg[i]);
 862                                 conf.rsync_option_arg[i] = NULL;
 863                         }
 864                         conf.rsync_option_given = 0;
 865                 }
 866                 cmdline_parser_config_file(config_file, &conf, &params);
 867         }
 868         ret = check_config();
 869         if (ret < 0)
 870                 goto out;
 871         if (override) {
 872                 /* don't change daemon mode on SIGHUP */
 873                 conf.daemon_given = old_daemon_given;
 874                 close_log(logfile);
 875                 logfile = NULL;
 876                 if (conf.logfile_given)
 877                         free(old_logfile_arg);
 878                 else if (conf.daemon_given) { /* re-use old logfile */
 879                         conf.logfile_arg = old_logfile_arg;
 880                         conf.logfile_given = 1;
 881                 }
 882         }
 883         if (conf.logfile_given) {
 884                 logfile = open_log(conf.logfile_arg);
 885                 log_welcome(conf.loglevel_arg);
 886         }
 887         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 888         ret = config_file_exists;
 889 out:
 890         free(config_file);
 891         if (ret < 0)
 892                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 893         return ret;
 894 }
 895
 896 static int change_to_dest_dir(void)
 897 {
 898         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 899         return dss_chdir(conf.dest_dir_arg);
 900 }
 901
 902 static void dump_dss_config(const char *msg)
 903 {
 904         if (conf.loglevel_arg > INFO)
 905                 return;
 906         DSS_INFO_LOG("%s\n", msg);
 907         cmdline_parser_dump(logfile? logfile : stderr, &conf);
 908 }
 909
 910 static int handle_sighup(void)
 911 {
 912         int ret;
 913
 914         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
 915         dump_dss_config("current config");
 916         ret = parse_config_file(1);
 917         if (ret < 0)
 918                 return ret;
 919         dump_dss_config("new config");
 920         invalidate_next_snapshot_time();
 921         return change_to_dest_dir();
 922 }
 923
 924 static int handle_signal(void)
 925 {
 926         int sig, ret = next_signal();
 927
 928         if (ret <= 0)
 929                 goto out;
 930         sig = ret;
 931         switch (sig) {
 932         case SIGINT:
 933         case SIGTERM:
 934                 restart_create_process();
 935                 kill_process(create_pid);
 936                 kill_process(remove_pid);
 937                 ret = -E_SIGNAL;
 938                 break;
 939         case SIGHUP:
 940                 ret = handle_sighup();
 941                 break;
 942         case SIGCHLD:
 943                 ret = handle_sigchld();
 944                 break;
 945         }
 946 out:
 947         if (ret < 0)
 948                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 949         return ret;
 950 }
 951
 952 /*
 953  * We can not use rsync locally if the local user is different from the remote
 954  * user or if the src dir is not on the local host (or both).
 955  */
 956 static int use_rsync_locally(char *logname)
 957 {
 958         char *h = conf.remote_host_arg;
 959
 960         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 961                 return 0;
 962         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 963                 return 0;
 964         return 1;
 965 }
 966
 967 static int rename_resume_snap(int64_t creation_time)
 968 {
 969         struct snapshot_list sl = {.num_snapshots = 0};
 970         struct snapshot *s;
 971         char *new_name = incomplete_name(creation_time);
 972         int ret;
 973
 974         ret = 0;
 975         if (conf.no_resume_given)
 976                 goto out;
 977         dss_get_snapshot_list(&sl);
 978         s = get_newest_snapshot(&sl);
 979         if (!s)
 980                 goto out;
 981         if ((s->flags & SS_COMPLETE) != 0) /* complete */
 982                 goto out;
 983         DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
 984         ret = dss_rename(s->name, new_name);
 985 out:
 986         if (ret >= 0)
 987                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
 988         free(new_name);
 989         free_snapshot_list(&sl);
 990         return ret;
 991 }
 992
 993 static void create_rsync_argv(char ***argv, int64_t *num)
 994 {
 995         char *logname;
 996         int i = 0, j;
 997         struct snapshot_list sl;
 998
 999         dss_get_snapshot_list(&sl);
1000         assert(!name_of_reference_snapshot);
1001         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1002         free_snapshot_list(&sl);
1003
1004         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1005         (*argv)[i++] = dss_strdup("rsync");
1006         (*argv)[i++] = dss_strdup("-aq");
1007         (*argv)[i++] = dss_strdup("--delete");
1008         for (j = 0; j < conf.rsync_option_given; j++)
1009                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1010         if (name_of_reference_snapshot) {
1011                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1012                 (*argv)[i++] = make_message("--link-dest=../%s",
1013                         name_of_reference_snapshot);
1014         } else
1015                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1016         logname = dss_logname();
1017         if (use_rsync_locally(logname))
1018                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1019         else
1020                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1021                         conf.remote_user_arg : logname,
1022                         conf.remote_host_arg, conf.source_dir_arg);
1023         free(logname);
1024         *num = get_current_time();
1025         (*argv)[i++] = incomplete_name(*num);
1026         (*argv)[i++] = NULL;
1027         for (j = 0; j < i; j++)
1028                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1029 }
1030
1031 static void free_rsync_argv(char **argv)
1032 {
1033         int i;
1034
1035         if (!argv)
1036                 return;
1037         for (i = 0; argv[i]; i++)
1038                 free(argv[i]);
1039         free(argv);
1040 }
1041
1042 static int create_snapshot(char **argv)
1043 {
1044         int ret, fds[3] = {0, 0, 0};
1045
1046         ret = rename_resume_snap(current_snapshot_creation_time);
1047         if (ret < 0)
1048                 return ret;
1049         ret = dss_exec(&create_pid, argv[0], argv, fds);
1050         if (ret < 0)
1051                 return ret;
1052         snapshot_creation_status = HS_RUNNING;
1053         return ret;
1054 }
1055
1056 static int select_loop(void)
1057 {
1058         int ret;
1059         /* check every 60 seconds for free disk space */
1060         struct timeval tv;
1061         char **rsync_argv = NULL;
1062
1063         for (;;) {
1064                 fd_set rfds;
1065                 struct timeval *tvp;
1066
1067                 if (remove_pid)
1068                         tvp = NULL; /* sleep until rm hook/process dies */
1069                 else { /* sleep one minute */
1070                         tv.tv_sec = 60;
1071                         tv.tv_usec = 0;
1072                         tvp = &tv;
1073                 }
1074                 FD_ZERO(&rfds);
1075                 FD_SET(signal_pipe, &rfds);
1076                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1077                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1078                 if (ret < 0)
1079                         goto out;
1080                 if (FD_ISSET(signal_pipe, &rfds)) {
1081                         ret = handle_signal();
1082                         if (ret < 0)
1083                                 goto out;
1084                 }
1085                 if (remove_pid)
1086                         continue;
1087                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1088                         ret = exec_rm();
1089                         if (ret < 0)
1090                                 goto out;
1091                         continue;
1092                 }
1093                 if (snapshot_removal_status == HS_SUCCESS) {
1094                         ret = post_remove_hook();
1095                         if (ret < 0)
1096                                 goto out;
1097                         continue;
1098                 }
1099                 ret = try_to_free_disk_space();
1100                 if (ret < 0)
1101                         goto out;
1102                 if (snapshot_removal_status != HS_READY) {
1103                         stop_create_process();
1104                         continue;
1105                 }
1106                 restart_create_process();
1107                 switch (snapshot_creation_status) {
1108                 case HS_READY:
1109                         if (!next_snapshot_is_due())
1110                                 continue;
1111                         ret = pre_create_hook();
1112                         if (ret < 0)
1113                                 goto out;
1114                         continue;
1115                 case HS_PRE_RUNNING:
1116                 case HS_RUNNING:
1117                 case HS_POST_RUNNING:
1118                         continue;
1119                 case HS_PRE_SUCCESS:
1120                         if (!name_of_reference_snapshot) {
1121                                 free_rsync_argv(rsync_argv);
1122                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1123                         }
1124                         /* fall through */
1125                 case HS_NEEDS_RESTART:
1126                         if (!next_snapshot_is_due())
1127                                 continue;
1128                         ret = create_snapshot(rsync_argv);
1129                         if (ret < 0)
1130                                 goto out;
1131                         continue;
1132                 case HS_SUCCESS:
1133                         ret = post_create_hook();
1134                         if (ret < 0)
1135                                 goto out;
1136                         continue;
1137                 }
1138         }
1139 out:
1140         return ret;
1141 }
1142
1143 static void exit_hook(int exit_code)
1144 {
1145         int fds[3] = {0, 0, 0};
1146         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1147         pid_t pid;
1148
1149         if (!conf.exit_hook_given)
1150                 return;
1151         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1152         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1153 }
1154
1155 static int com_run(void)
1156 {
1157         int ret;
1158
1159         if (conf.dry_run_given) {
1160                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1161                 return -E_SYNTAX;
1162         }
1163         ret = install_sighandler(SIGHUP);
1164         if (ret < 0)
1165                 return ret;
1166         ret = select_loop();
1167         if (ret >= 0) /* impossible */
1168                 ret = -E_BUG;
1169         exit_hook(ret);
1170         return ret;
1171 }
1172
1173 static int com_prune(void)
1174 {
1175         int ret;
1176         struct snapshot_list sl;
1177         struct snapshot *victim;
1178         struct disk_space ds;
1179         const char *why;
1180
1181         ret = get_disk_space(".", &ds);
1182         if (ret < 0)
1183                 return ret;
1184         log_disk_space(&ds);
1185         dss_get_snapshot_list(&sl);
1186         why = "outdated";
1187         victim = find_outdated_snapshot(&sl);
1188         if (victim)
1189                 goto rm;
1190         why = "redundant";
1191         victim = find_redundant_snapshot(&sl);
1192         if (victim)
1193                 goto rm;
1194         ret = 0;
1195         goto out;
1196 rm:
1197         if (conf.dry_run_given) {
1198                 dss_msg("%s snapshot %s (interval = %i)\n",
1199                         why, victim->name, victim->interval);
1200                 ret = 0;
1201                 goto out;
1202         }
1203         ret = pre_remove_hook(victim, why);
1204         if (ret < 0)
1205                 goto out;
1206         if (snapshot_removal_status == HS_PRE_RUNNING) {
1207                 ret = wait_for_remove_process();
1208                 if (ret < 0)
1209                         goto out;
1210                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1211                         goto out;
1212         }
1213         ret = exec_rm();
1214         if (ret < 0)
1215                 goto out;
1216         ret = wait_for_remove_process();
1217         if (ret < 0)
1218                 goto out;
1219         if (snapshot_removal_status != HS_SUCCESS)
1220                 goto out;
1221         ret = post_remove_hook();
1222         if (ret < 0)
1223                 goto out;
1224         if (snapshot_removal_status != HS_POST_RUNNING)
1225                 goto out;
1226         ret = wait_for_remove_process();
1227         if (ret < 0)
1228                 goto out;
1229         ret = 1;
1230 out:
1231         free_snapshot_list(&sl);
1232         return ret;
1233 }
1234
1235 static int com_create(void)
1236 {
1237         int ret, status;
1238         char **rsync_argv;
1239
1240         if (conf.dry_run_given) {
1241                 int i;
1242                 char *msg = NULL;
1243                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1244                 for (i = 0; rsync_argv[i]; i++) {
1245                         char *tmp = msg;
1246                         msg = make_message("%s%s%s", tmp? tmp : "",
1247                                 tmp? " " : "", rsync_argv[i]);
1248                         free(tmp);
1249                 }
1250                 free_rsync_argv(rsync_argv);
1251                 dss_msg("%s\n", msg);
1252                 free(msg);
1253                 return 1;
1254         }
1255         ret = pre_create_hook();
1256         if (ret < 0)
1257                 return ret;
1258         if (create_pid) {
1259                 ret = wait_for_process(create_pid, &status);
1260                 if (ret < 0)
1261                         return ret;
1262                 ret = handle_pre_create_hook_exit(status);
1263                 if (ret <= 0) /* error, or pre-create failed */
1264                         return ret;
1265         }
1266         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1267         ret = create_snapshot(rsync_argv);
1268         if (ret < 0)
1269                 goto out;
1270         ret = wait_for_process(create_pid, &status);
1271         if (ret < 0)
1272                 goto out;
1273         ret = handle_rsync_exit(status);
1274         if (ret < 0)
1275                 goto out;
1276         post_create_hook();
1277         if (create_pid)
1278                 ret = wait_for_process(create_pid, &status);
1279 out:
1280         free_rsync_argv(rsync_argv);
1281         return ret;
1282 }
1283
1284 static int com_ls(void)
1285 {
1286         int i;
1287         struct snapshot_list sl;
1288         struct snapshot *s;
1289
1290         dss_get_snapshot_list(&sl);
1291         FOR_EACH_SNAPSHOT(s, i, &sl) {
1292                 int64_t d = 0;
1293                 if (s->flags & SS_COMPLETE)
1294                         d = (s->completion_time - s->creation_time) / 60;
1295                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1296         };
1297         free_snapshot_list(&sl);
1298         return 1;
1299 }
1300
1301 static int setup_signal_handling(void)
1302 {
1303         int ret;
1304
1305         DSS_INFO_LOG("setting up signal handlers\n");
1306         signal_pipe = signal_init(); /* always successful */
1307         ret = install_sighandler(SIGINT);
1308         if (ret < 0)
1309                 return ret;
1310         ret = install_sighandler(SIGTERM);
1311         if (ret < 0)
1312                 return ret;
1313         return install_sighandler(SIGCHLD);
1314 }
1315
1316 /**
1317  * The main function of dss.
1318  *
1319  * \param argc Usual argument count.
1320  * \param argv Usual argument vector.
1321  */
1322 int main(int argc, char **argv)
1323 {
1324         int ret;
1325         struct cmdline_parser_params params = {
1326                 .override = 0,
1327                 .initialize = 1,
1328                 .check_required = 0,
1329                 .check_ambiguity = 0,
1330                 .print_errors = 1
1331         };
1332
1333         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1334         ret = parse_config_file(0);
1335         if (ret < 0)
1336                 goto out;
1337         if (ret == 0) { /* no config file given */
1338                 /*
1339                  * Parse the command line options again, but this time check
1340                  * that all required options are given.
1341                  */
1342                 params = (struct cmdline_parser_params) {
1343                         .override = 1,
1344                         .initialize = 1,
1345                         .check_required = 1,
1346                         .check_ambiguity = 1,
1347                         .print_errors = 1
1348                 };
1349                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1350         }
1351         if (conf.daemon_given)
1352                 daemon_init();
1353         dump_dss_config("dss configuration");
1354         ret = change_to_dest_dir();
1355         if (ret < 0)
1356                 goto out;
1357         ret = setup_signal_handling();
1358         if (ret < 0)
1359                 goto out;
1360         ret = call_command_handler();
1361 out:
1362         if (ret < 0)
1363                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1364         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1365 }