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