dss.c

   1 /*
   2  * Copyright (C) 2008 Andre Noll <maan@systemlinux.org>
   3  *
   4  * Licensed under the GPL v2. For licencing details see COPYING.
   5  */
   6 #include <string.h>
   7 #include <stdlib.h>
   8 #include <stdarg.h>
   9 #include <assert.h>
  10 #include <errno.h>
  11 #include <sys/types.h>
  12 #include <signal.h>
  13 #include <ctype.h>
  14 #include <sys/stat.h>
  15 #include <unistd.h>
  16 #include <inttypes.h>
  17 #include <sys/time.h>
  18 #include <time.h>
  19 #include <sys/wait.h>
  20 #include <fnmatch.h>
  21 #include <limits.h>
  22
  23
  24 #include "gcc-compat.h"
  25 #include "cmdline.h"
  26 #include "log.h"
  27 #include "string.h"
  28 #include "error.h"
  29 #include "fd.h"
  30 #include "exec.h"
  31 #include "daemon.h"
  32 #include "signal.h"
  33 #include "df.h"
  34 #include "time.h"
  35 #include "snap.h"
  36
  37 /** Command line and config file options. */
  38 static struct gengetopt_args_info conf;
  39 /** Non-NULL if we log to a file. */
  40 static FILE *logfile;
  41 /** The read end of the signal pipe */
  42 static int signal_pipe;
  43 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
  44 static pid_t create_pid;
  45 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
  46 static int create_process_stopped;
  47 /** Process id of current pre-remove/rm/post-remove process. */
  48 static pid_t remove_pid;
  49 /** When the next snapshot is due. */
  50 static struct timeval next_snapshot_time;
  51 /** When to try to remove something. */
  52 static struct timeval next_removal_check;
  53 /** Creation time of the snapshot currently being created. */
  54 static int64_t current_snapshot_creation_time;
  55 /* The snapshot currently being removed. */
  56 struct snapshot *snapshot_currently_being_removed;
  57 /** Needed by the post-create hook. */
  58 static char *path_to_last_complete_snapshot;
  59 /** \sa \ref snap.h for details. */
  60 enum hook_status snapshot_creation_status;
  61 /** \sa \ref snap.h for details. */
  62 enum hook_status snapshot_removal_status;
  63
  64
  65 DEFINE_DSS_ERRLIST;
  66
  67
  68 /* a litte cpp magic helps to DRY */
  69 #define COMMANDS \
  70         COMMAND(ls) \
  71         COMMAND(create) \
  72         COMMAND(prune) \
  73         COMMAND(run)
  74 #define COMMAND(x) static int com_ ##x(void);
  75 COMMANDS
  76 #undef COMMAND
  77 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
  78 static int call_command_handler(void)
  79 {
  80         COMMANDS
  81         DSS_EMERG_LOG("BUG: did not find command handler\n");
  82         return -E_BUG;
  83 }
  84 #undef COMMAND
  85 #undef COMMANDS
  86
  87 /**
  88  * The log function of dss.
  89  *
  90  * \param ll Loglevel.
  91  * \param fml Usual format string.
  92  *
  93  * All DSS_XXX_LOG() macros use this function.
  94  */
  95 __printf_2_3 void dss_log(int ll, const char* fmt,...)
  96 {
  97         va_list argp;
  98         FILE *outfd;
  99         struct tm *tm;
 100         time_t t1;
 101         char str[255] = "";
 102
 103         if (ll < conf.loglevel_arg)
 104                 return;
 105         outfd = logfile? logfile : stderr;
 106         time(&t1);
 107         tm = localtime(&t1);
 108         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 109         fprintf(outfd, "%s ", str);
 110         if (conf.loglevel_arg <= INFO)
 111                 fprintf(outfd, "%i: ", ll);
 112         va_start(argp, fmt);
 113         vfprintf(outfd, fmt, argp);
 114         va_end(argp);
 115 }
 116
 117 /**
 118  * Print a message either to stdout or to the log file.
 119  */
 120 static __printf_1_2 void dss_msg(const char* fmt,...)
 121 {
 122         FILE *outfd = conf.daemon_given? logfile : stdout;
 123         va_list argp;
 124         va_start(argp, fmt);
 125         vfprintf(outfd, fmt, argp);
 126         va_end(argp);
 127 }
 128
 129 static int disk_space_low(void)
 130 {
 131         struct disk_space ds;
 132         int ret = get_disk_space(".", &ds);
 133
 134         if (ret < 0)
 135                 return ret;
 136         if (conf.min_free_mb_arg)
 137                 if (ds.free_mb < conf.min_free_mb_arg)
 138                         return 1;
 139         if (conf.min_free_percent_arg)
 140                 if (ds.percent_free < conf.min_free_percent_arg)
 141                         return 1;
 142         if (conf.min_free_percent_inodes_arg)
 143                 if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg)
 144                         return 1;
 145         return 0;
 146 }
 147
 148 static void dss_get_snapshot_list(struct snapshot_list *sl)
 149 {
 150         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 151 }
 152
 153 static int next_snapshot_is_due(void)
 154 {
 155         struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 156                 tmp, diff;
 157         int64_t x = 0;
 158         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 159                 num_complete_snapshots = 0;
 160         int i, ret;
 161         struct snapshot *s = NULL;
 162         struct snapshot_list sl;
 163
 164         gettimeofday(&now, NULL);
 165         if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 166                 return 0;
 167         assert(snapshot_creation_status == HS_READY);
 168         current_snapshot_creation_time = 0;
 169         dss_get_snapshot_list(&sl);
 170         FOR_EACH_SNAPSHOT(s, i, &sl) {
 171                 if (!(s->flags & SS_COMPLETE))
 172                         continue;
 173                 num_complete_snapshots++;
 174                 x += s->completion_time - s->creation_time;
 175         }
 176         assert(x >= 0);
 177         if (num_complete_snapshots)
 178                 x /= num_complete_snapshots; /* avg time to create one snapshot */
 179         x *= wanted; /* time to create all snapshots in interval 0 */
 180         tmp.tv_sec = x;
 181         tmp.tv_usec = 0;
 182         ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
 183         if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
 184                 ret = 1;
 185                 goto out;
 186         }
 187
 188         tv_divide(wanted, &diff, &tmp); /* sleep time betweeen two snapshots */
 189         diff.tv_sec = s->completion_time; /* completion time of the the latest snaphot */
 190         diff.tv_usec = 0;
 191         tv_add(&diff, &tmp, &next_snapshot_time);
 192         ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
 193 out:
 194         free_snapshot_list(&sl);
 195         if (ret == 0)
 196                 DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
 197                         tv2ms(&diff) / 1000);
 198         else
 199                 DSS_DEBUG_LOG("next snapshot: now\n");
 200         return ret;
 201 }
 202
 203 static int pre_create_hook(void)
 204 {
 205         int ret, fds[3] = {0, 0, 0};
 206
 207         assert(snapshot_creation_status == HS_READY);
 208         if (!conf.pre_create_hook_given) {
 209                 snapshot_creation_status = HS_PRE_SUCCESS;
 210                 return 0;
 211         }
 212         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 213         ret = dss_exec_cmdline_pid(&create_pid,
 214                 conf.pre_create_hook_arg, fds);
 215         if (ret < 0)
 216                 return ret;
 217         snapshot_creation_status = HS_PRE_RUNNING;
 218         return ret;
 219 }
 220
 221 static int pre_remove_hook(struct snapshot *s, char *why)
 222 {
 223         int ret, fds[3] = {0, 0, 0};
 224         char *cmd;
 225
 226         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 227         assert(snapshot_removal_status == HS_READY);
 228         assert(remove_pid == 0);
 229         assert(!snapshot_currently_being_removed);
 230
 231         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 232         *snapshot_currently_being_removed = *s;
 233         snapshot_currently_being_removed->name = dss_strdup(s->name);
 234
 235         if (!conf.pre_remove_hook_given) {
 236                 snapshot_removal_status = HS_PRE_SUCCESS;
 237                 return 0;
 238         }
 239         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 240                 conf.dest_dir_arg, s->name);
 241         DSS_DEBUG_LOG("executing %s\n", cmd);
 242         ret = dss_exec_cmdline_pid(&remove_pid,
 243                 conf.pre_remove_hook_arg, fds);
 244         free(cmd);
 245         if (ret < 0)
 246                 return ret;
 247         snapshot_removal_status = HS_PRE_RUNNING;
 248         return ret;
 249 }
 250
 251 static int exec_rm(void)
 252 {
 253         struct snapshot *s = snapshot_currently_being_removed;
 254         int fds[3] = {0, 0, 0};
 255         char *new_name = being_deleted_name(s);
 256         char *argv[] = {"rm", "-rf", new_name, NULL};
 257         int ret;
 258
 259         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 260         assert(remove_pid == 0);
 261
 262         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 263         ret = dss_rename(s->name, new_name);
 264         if (ret < 0)
 265                 goto out;
 266         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 267         if (ret < 0)
 268                 goto out;
 269         snapshot_removal_status = HS_RUNNING;
 270 out:
 271         free(new_name);
 272         return ret;
 273 }
 274
 275 static int snapshot_is_being_created(struct snapshot *s)
 276 {
 277         return s->creation_time == current_snapshot_creation_time;
 278 }
 279
 280 /*
 281  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 282  */
 283 static int remove_redundant_snapshot(struct snapshot_list *sl)
 284 {
 285         int ret, i, interval;
 286         struct snapshot *s;
 287         unsigned missing = 0;
 288
 289         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 290         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 291                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 292                 unsigned num = sl->interval_count[interval];
 293                 struct snapshot *victim = NULL, *prev = NULL;
 294                 int64_t score = LONG_MAX;
 295
 296                 if (keep >= num)
 297                         missing += keep - num;
 298 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 299 //                      interval, keep, num, missing);
 300                 if (keep + missing >= num)
 301                         continue;
 302                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 303                 FOR_EACH_SNAPSHOT(s, i, sl) {
 304                         int64_t this_score;
 305
 306                         if (snapshot_is_being_created(s))
 307                                 continue;
 308                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 309                         if (s->interval > interval) {
 310                                 prev = s;
 311                                 continue;
 312                         }
 313                         if (s->interval < interval)
 314                                 break;
 315                         if (!victim) {
 316                                 victim = s;
 317                                 prev = s;
 318                                 continue;
 319                         }
 320                         assert(prev);
 321                         /* check if s is a better victim */
 322                         this_score = s->creation_time - prev->creation_time;
 323                         assert(this_score >= 0);
 324                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 325                         if (this_score < score) {
 326                                 score = this_score;
 327                                 victim = s;
 328                         }
 329                         prev = s;
 330                 }
 331                 assert(victim);
 332                 if (conf.dry_run_given) {
 333                         dss_msg("%s would be removed (interval = %i)\n",
 334                                 victim->name, victim->interval);
 335                         continue;
 336                 }
 337                 ret = pre_remove_hook(victim, "redundant");
 338                 return ret < 0? ret : 1;
 339         }
 340         return 0;
 341 }
 342
 343 static int remove_outdated_snapshot(struct snapshot_list *sl)
 344 {
 345         int i, ret;
 346         struct snapshot *s;
 347
 348         DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n",
 349                 conf.num_intervals_arg);
 350         FOR_EACH_SNAPSHOT(s, i, sl) {
 351                 if (snapshot_is_being_created(s))
 352                         continue;
 353                 if (s->interval <= conf.num_intervals_arg)
 354                         continue;
 355                 if (conf.dry_run_given) {
 356                         dss_msg("%s would be removed (interval = %i)\n",
 357                                 s->name, s->interval);
 358                         continue;
 359                 }
 360                 ret = pre_remove_hook(s, "outdated");
 361                 if (ret < 0)
 362                         return ret;
 363                 return 1;
 364         }
 365         return 0;
 366 }
 367
 368 static int remove_oldest_snapshot(struct snapshot_list *sl)
 369 {
 370         struct snapshot *s = get_oldest_snapshot(sl);
 371
 372         if (!s) /* no snapshot found */
 373                 return 0;
 374         DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
 375         if (snapshot_is_being_created(s))
 376                 return 0;
 377         return pre_remove_hook(s, "oldest");
 378 }
 379
 380 static int rename_incomplete_snapshot(int64_t start)
 381 {
 382         char *old_name;
 383         int ret;
 384
 385         free(path_to_last_complete_snapshot);
 386         ret = complete_name(start, get_current_time(),
 387                 &path_to_last_complete_snapshot);
 388         if (ret < 0)
 389                 return ret;
 390         old_name = incomplete_name(start);
 391         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 392         if (ret >= 0)
 393                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 394                         path_to_last_complete_snapshot);
 395         free(old_name);
 396         return ret;
 397 }
 398
 399 static int try_to_free_disk_space(int low_disk_space)
 400 {
 401         int ret;
 402         struct snapshot_list sl;
 403         struct timeval now;
 404
 405         gettimeofday(&now, NULL);
 406         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 407                 return 0;
 408         if (!low_disk_space && conf.keep_redundant_given)
 409                 return 0;
 410         dss_get_snapshot_list(&sl);
 411         ret = remove_outdated_snapshot(&sl);
 412         if (ret) /* error, or we are removing something */
 413                 goto out;
 414         /* no outdated snapshot */
 415         ret = remove_redundant_snapshot(&sl);
 416         if (ret)
 417                 goto out;
 418         ret = 0;
 419         if (!low_disk_space)
 420                 goto out;
 421         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 422         ret = remove_oldest_snapshot(&sl);
 423         if (ret)
 424                 goto out;
 425         DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
 426         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 427 out:
 428         free_snapshot_list(&sl);
 429         return ret;
 430 }
 431
 432 static int post_create_hook(void)
 433 {
 434         int ret, fds[3] = {0, 0, 0};
 435         char *cmd;
 436
 437         if (!conf.post_create_hook_given) {
 438                 snapshot_creation_status = HS_READY;
 439                 return 0;
 440         }
 441         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 442                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 443         DSS_NOTICE_LOG("executing %s\n", cmd);
 444         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 445         free(cmd);
 446         if (ret < 0)
 447                 return ret;
 448         snapshot_creation_status = HS_POST_RUNNING;
 449         return ret;
 450 }
 451
 452 static int post_remove_hook(void)
 453 {
 454         int ret, fds[3] = {0, 0, 0};
 455         char *cmd;
 456         struct snapshot *s = snapshot_currently_being_removed;
 457
 458         assert(s);
 459
 460         if (!conf.post_remove_hook_given) {
 461                 snapshot_removal_status = HS_READY;
 462                 return 0;
 463         }
 464         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 465                 conf.dest_dir_arg, s->name);
 466         DSS_NOTICE_LOG("executing %s\n", cmd);
 467         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 468         free(cmd);
 469         if (ret < 0)
 470                 return ret;
 471         snapshot_removal_status = HS_POST_RUNNING;
 472         return ret;
 473 }
 474
 475 static void kill_process(pid_t pid)
 476 {
 477         if (!pid)
 478                 return;
 479         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 480         kill(pid, SIGTERM);
 481 }
 482
 483 static void stop_create_process(void)
 484 {
 485         if (!create_pid || create_process_stopped)
 486                 return;
 487         kill(SIGSTOP, create_pid);
 488         create_process_stopped = 1;
 489 }
 490
 491 static void restart_create_process(void)
 492 {
 493         if (!create_pid || !create_process_stopped)
 494                 return;
 495         kill (SIGCONT, create_pid);
 496         create_process_stopped = 0;
 497 }
 498
 499 /**
 500  * Print a log message about the exit status of a child.
 501  */
 502 static void log_termination_msg(pid_t pid, int status)
 503 {
 504         if (WIFEXITED(status))
 505                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 506                         WEXITSTATUS(status));
 507         else if (WIFSIGNALED(status))
 508                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 509                         WTERMSIG(status));
 510         else
 511                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 512 }
 513
 514 static int wait_for_process(pid_t pid, int *status)
 515 {
 516         int ret;
 517
 518         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 519         for (;;) {
 520                 fd_set rfds;
 521
 522                 FD_ZERO(&rfds);
 523                 FD_SET(signal_pipe, &rfds);
 524                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 525                 if (ret < 0)
 526                         break;
 527                 ret = next_signal();
 528                 if (!ret)
 529                         continue;
 530                 if (ret == SIGCHLD) {
 531                         ret = waitpid(pid, status, 0);
 532                         if (ret >= 0)
 533                                 break;
 534                         if (errno != EINTR) { /* error */
 535                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 536                                 break;
 537                         }
 538                 }
 539                 /* SIGINT or SIGTERM */
 540                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 541                 kill(pid, SIGTERM);
 542         }
 543         if (ret < 0)
 544                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 545         else
 546                 log_termination_msg(pid, *status);
 547         return ret;
 548 }
 549
 550 static void handle_pre_remove_exit(int status)
 551 {
 552         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 553                 snapshot_removal_status = HS_READY;
 554                 gettimeofday(&next_removal_check, NULL);
 555                 next_removal_check.tv_sec += 60;
 556                 return;
 557         }
 558         snapshot_removal_status = HS_PRE_SUCCESS;
 559 }
 560
 561 static int handle_rm_exit(int status)
 562 {
 563         if (!WIFEXITED(status)) {
 564                 snapshot_removal_status = HS_READY;
 565                 return -E_INVOLUNTARY_EXIT;
 566         }
 567         if (WEXITSTATUS(status)) {
 568                 snapshot_removal_status = HS_READY;
 569                 return -E_BAD_EXIT_CODE;
 570         }
 571         if (conf.post_remove_hook_given)
 572                 snapshot_removal_status = HS_SUCCESS;
 573         else
 574                 snapshot_removal_status = HS_READY;
 575         return 1;
 576 }
 577
 578 static void handle_post_remove_exit(void)
 579 {
 580         snapshot_removal_status = HS_READY;
 581 }
 582
 583 static int handle_remove_exit(int status)
 584 {
 585         int ret;
 586         struct snapshot *s = snapshot_currently_being_removed;
 587
 588         assert(s);
 589         switch (snapshot_removal_status) {
 590         case HS_PRE_RUNNING:
 591                 handle_pre_remove_exit(status);
 592                 ret = 1;
 593                 break;
 594         case HS_RUNNING:
 595                 ret = handle_rm_exit(status);
 596                 break;
 597         case HS_POST_RUNNING:
 598                 handle_post_remove_exit();
 599                 ret = 1;
 600                 break;
 601         default:
 602                 ret = -E_BUG;
 603         }
 604         if (snapshot_removal_status == HS_READY) {
 605                 free(s->name);
 606                 free(s);
 607                 snapshot_currently_being_removed = NULL;
 608         }
 609         remove_pid = 0;
 610         return ret;
 611 }
 612
 613 static int wait_for_remove_process(void)
 614 {
 615         int status, ret;
 616
 617         assert(remove_pid);
 618         assert(
 619                 snapshot_removal_status == HS_PRE_RUNNING ||
 620                 snapshot_removal_status == HS_RUNNING ||
 621                 snapshot_removal_status == HS_POST_RUNNING
 622         );
 623         ret = wait_for_process(remove_pid, &status);
 624         if (ret < 0)
 625                 return ret;
 626         return handle_remove_exit(status);
 627 }
 628
 629 static int handle_rsync_exit(int status)
 630 {
 631         int es, ret;
 632
 633         if (!WIFEXITED(status)) {
 634                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 635                 ret = -E_INVOLUNTARY_EXIT;
 636                 snapshot_creation_status = HS_READY;
 637                 goto out;
 638         }
 639         es = WEXITSTATUS(status);
 640         if (es == 13) { /* Errors with program diagnostics */
 641                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 642                         (int)create_pid, es);
 643                 snapshot_creation_status = HS_NEEDS_RESTART;
 644                 gettimeofday(&next_snapshot_time, NULL);
 645                 next_snapshot_time.tv_sec += 60;
 646                 ret = 1;
 647                 goto out;
 648         }
 649         if (es != 0 && es != 23 && es != 24) {
 650                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 651                 ret = -E_BAD_EXIT_CODE;
 652                 snapshot_creation_status = HS_READY;
 653                 goto out;
 654         }
 655         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 656         if (ret < 0)
 657                 goto out;
 658         snapshot_creation_status = HS_SUCCESS;
 659 out:
 660         create_pid = 0;
 661         create_process_stopped = 0;
 662         return ret;
 663 }
 664
 665 static int handle_pre_create_hook_exit(int status)
 666 {
 667         int es, ret;
 668         static int warn_count;
 669
 670         if (!WIFEXITED(status)) {
 671                 snapshot_creation_status = HS_READY;
 672                 ret = -E_INVOLUNTARY_EXIT;
 673                 goto out;
 674         }
 675         es = WEXITSTATUS(status);
 676         if (es) {
 677                 if (!warn_count--) {
 678                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 679                                 conf.pre_create_hook_arg, es);
 680                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 681                         warn_count = 60; /* warn only once per hour */
 682                 }
 683                 snapshot_creation_status = HS_READY;
 684                 ret = 0;
 685                 goto out;
 686         }
 687         warn_count = 0;
 688         snapshot_creation_status = HS_PRE_SUCCESS;
 689         ret = 1;
 690 out:
 691         create_pid = 0;
 692         return ret;
 693 }
 694
 695 static int handle_sigchld(void)
 696 {
 697         pid_t pid;
 698         int status, ret = reap_child(&pid, &status);
 699
 700         if (ret <= 0)
 701                 return ret;
 702
 703         if (pid == create_pid) {
 704                 switch (snapshot_creation_status) {
 705                 case HS_PRE_RUNNING:
 706                         return handle_pre_create_hook_exit(status);
 707                 case HS_RUNNING:
 708                         return handle_rsync_exit(status);
 709                 case HS_POST_RUNNING:
 710                         snapshot_creation_status = HS_READY;
 711                         return 1;
 712                 default:
 713                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 714                                 snapshot_creation_status);
 715                         return -E_BUG;
 716                 }
 717         }
 718         if (pid == remove_pid) {
 719                 ret = handle_remove_exit(status);
 720                 if (ret < 0)
 721                         return ret;
 722                 return ret;
 723         }
 724         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 725         return -E_BUG;
 726 }
 727
 728 static int check_config(void)
 729 {
 730         if (conf.unit_interval_arg <= 0) {
 731                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 732                 return -E_INVALID_NUMBER;
 733         }
 734         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 735         if (conf.num_intervals_arg <= 0) {
 736                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 737                 return -E_INVALID_NUMBER;
 738         }
 739         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 740         return 1;
 741 }
 742
 743 static int parse_config_file(int override)
 744 {
 745         int ret;
 746         char *config_file;
 747         struct stat statbuf;
 748         char *old_logfile_arg = NULL;
 749         int old_daemon_given = 0;
 750
 751         if (conf.config_file_given)
 752                 config_file = dss_strdup(conf.config_file_arg);
 753         else {
 754                 char *home = get_homedir();
 755                 config_file = make_message("%s/.dssrc", home);
 756                 free(home);
 757         }
 758         if (override) { /* SIGHUP */
 759                 if (conf.logfile_given)
 760                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 761                 old_daemon_given = conf.daemon_given;
 762         }
 763
 764         ret = stat(config_file, &statbuf);
 765         if (ret && conf.config_file_given) {
 766                 ret = -ERRNO_TO_DSS_ERROR(errno);
 767                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 768                 goto out;
 769         }
 770         if (!ret) {
 771                 struct cmdline_parser_params params = {
 772                         .override = override,
 773                         .initialize = 0,
 774                         .check_required = 1,
 775                         .check_ambiguity = 0,
 776                         .print_errors = 1
 777                 };
 778                 if (override) { /* invalidate all rsync options */
 779                         int i;
 780
 781                         for (i = 0; i < conf.rsync_option_given; i++) {
 782                                 free(conf.rsync_option_arg[i]);
 783                                 conf.rsync_option_arg[i] = NULL;
 784                         }
 785                         conf.rsync_option_given = 0;
 786                 }
 787                 cmdline_parser_config_file(config_file, &conf, &params);
 788         }
 789         ret = check_config();
 790         if (ret < 0)
 791                 goto out;
 792         if (override) {
 793                 /* don't change daemon mode on SIGHUP */
 794                 conf.daemon_given = old_daemon_given;
 795                 close_log(logfile);
 796                 logfile = NULL;
 797                 if (conf.logfile_given)
 798                         free(old_logfile_arg);
 799                 else if (conf.daemon_given) { /* re-use old logfile */
 800                         conf.logfile_arg = old_logfile_arg;
 801                         conf.logfile_given = 1;
 802                 }
 803         }
 804         if (conf.logfile_given) {
 805                 logfile = open_log(conf.logfile_arg);
 806                 log_welcome(conf.loglevel_arg);
 807         }
 808         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 809 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 810 out:
 811         free(config_file);
 812         if (ret < 0)
 813                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 814         return ret;
 815 }
 816
 817 static int change_to_dest_dir(void)
 818 {
 819         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 820         return dss_chdir(conf.dest_dir_arg);
 821 }
 822
 823 static int handle_sighup(void)
 824 {
 825         int ret;
 826
 827         DSS_NOTICE_LOG("SIGHUP\n");
 828         ret = parse_config_file(1);
 829         if (ret < 0)
 830                 return ret;
 831         return change_to_dest_dir();
 832 }
 833
 834 static int handle_signal(void)
 835 {
 836         int sig, ret = next_signal();
 837
 838         if (ret <= 0)
 839                 goto out;
 840         sig = ret;
 841         switch (sig) {
 842         case SIGINT:
 843         case SIGTERM:
 844                 restart_create_process();
 845                 kill_process(create_pid);
 846                 kill_process(remove_pid);
 847                 ret = -E_SIGNAL;
 848                 break;
 849         case SIGHUP:
 850                 ret = handle_sighup();
 851                 break;
 852         case SIGCHLD:
 853                 ret = handle_sigchld();
 854                 break;
 855         }
 856 out:
 857         if (ret < 0)
 858                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 859         return ret;
 860 }
 861
 862 /*
 863  * We can not use rsync locally if the local user is different from the remote
 864  * user or if the src dir is not on the local host (or both).
 865  */
 866 static int use_rsync_locally(char *logname)
 867 {
 868         char *h = conf.remote_host_arg;
 869
 870         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 871                 return 0;
 872         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 873                 return 0;
 874         return 1;
 875 }
 876
 877 static void create_rsync_argv(char ***argv, int64_t *num)
 878 {
 879         char *logname, *newest;
 880         int i = 0, j;
 881         struct snapshot_list sl;
 882
 883         dss_get_snapshot_list(&sl);
 884         newest = name_of_newest_complete_snapshot(&sl);
 885         free_snapshot_list(&sl);
 886
 887         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 888         (*argv)[i++] = dss_strdup("rsync");
 889         (*argv)[i++] = dss_strdup("-aq");
 890         (*argv)[i++] = dss_strdup("--delete");
 891         for (j = 0; j < conf.rsync_option_given; j++)
 892                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 893         if (newest) {
 894                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 895                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 896                 free(newest);
 897         } else
 898                 DSS_INFO_LOG("no previous snapshot found\n");
 899         logname = dss_logname();
 900         if (use_rsync_locally(logname))
 901                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 902         else
 903                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 904                         conf.remote_user_arg : logname,
 905                         conf.remote_host_arg, conf.source_dir_arg);
 906         free(logname);
 907         *num = get_current_time();
 908         (*argv)[i++] = incomplete_name(*num);
 909         (*argv)[i++] = NULL;
 910         for (j = 0; j < i; j++)
 911                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 912 }
 913
 914 static void free_rsync_argv(char **argv)
 915 {
 916         int i;
 917
 918         if (!argv)
 919                 return;
 920         for (i = 0; argv[i]; i++)
 921                 free(argv[i]);
 922         free(argv);
 923 }
 924
 925 static int create_snapshot(char **argv)
 926 {
 927         int ret, fds[3] = {0, 0, 0};
 928         char *name;
 929
 930         name = incomplete_name(current_snapshot_creation_time);
 931         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 932         free(name);
 933         ret = dss_exec(&create_pid, argv[0], argv, fds);
 934         if (ret < 0)
 935                 return ret;
 936         snapshot_creation_status = HS_RUNNING;
 937         return ret;
 938 }
 939
 940 static int select_loop(void)
 941 {
 942         int ret;
 943         /* check every 60 seconds for free disk space */
 944         struct timeval tv;
 945         char **rsync_argv = NULL;
 946
 947         for (;;) {
 948                 fd_set rfds;
 949                 int low_disk_space;
 950                 struct timeval *tvp;
 951
 952                 if (remove_pid)
 953                         tvp = NULL; /* sleep until rm hook/process dies */
 954                 else { /* sleep one minute */
 955                         tv.tv_sec = 60;
 956                         tv.tv_usec = 0;
 957                         tvp = &tv;
 958                 }
 959                 FD_ZERO(&rfds);
 960                 FD_SET(signal_pipe, &rfds);
 961                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
 962                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
 963                 if (ret < 0)
 964                         goto out;
 965                 if (FD_ISSET(signal_pipe, &rfds)) {
 966                         ret = handle_signal();
 967                         if (ret < 0)
 968                                 goto out;
 969                 }
 970                 if (remove_pid)
 971                         continue;
 972                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
 973                         ret = exec_rm();
 974                         if (ret < 0)
 975                                 goto out;
 976                         continue;
 977                 }
 978                 if (snapshot_removal_status == HS_SUCCESS) {
 979                         ret = post_remove_hook();
 980                         if (ret < 0)
 981                                 goto out;
 982                         continue;
 983                 }
 984                 ret = disk_space_low();
 985                 if (ret < 0)
 986                         goto out;
 987                 low_disk_space = ret;
 988                 ret = try_to_free_disk_space(low_disk_space);
 989                 if (ret < 0)
 990                         goto out;
 991                 if (snapshot_removal_status != HS_READY) {
 992                         stop_create_process();
 993                         continue;
 994                 }
 995                 restart_create_process();
 996                 switch (snapshot_creation_status) {
 997                 case HS_READY:
 998                         if (!next_snapshot_is_due())
 999                                 continue;
1000                         ret = pre_create_hook();
1001                         if (ret < 0)
1002                                 goto out;
1003                         continue;
1004                 case HS_PRE_RUNNING:
1005                 case HS_RUNNING:
1006                 case HS_POST_RUNNING:
1007                         continue;
1008                 case HS_PRE_SUCCESS:
1009                         free_rsync_argv(rsync_argv);
1010                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1011                         /* fall through */
1012                 case HS_NEEDS_RESTART:
1013                         ret = create_snapshot(rsync_argv);
1014                         if (ret < 0)
1015                                 goto out;
1016                         continue;
1017                 case HS_SUCCESS:
1018                         ret = post_create_hook();
1019                         if (ret < 0)
1020                                 goto out;
1021                         continue;
1022                 }
1023         }
1024 out:
1025         return ret;
1026 }
1027
1028 static void exit_hook(int exit_code)
1029 {
1030         int fds[3] = {0, 0, 0};
1031         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1032         pid_t pid;
1033
1034         if (!conf.exit_hook_given)
1035                 return;
1036         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1037         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1038 }
1039
1040 static int com_run(void)
1041 {
1042         int ret;
1043
1044         if (conf.dry_run_given) {
1045                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1046                 return -E_SYNTAX;
1047         }
1048         ret = install_sighandler(SIGHUP);
1049         if (ret < 0)
1050                 return ret;
1051         ret = select_loop();
1052         if (ret >= 0) /* impossible */
1053                 ret = -E_BUG;
1054         exit_hook(ret);
1055         return ret;
1056 }
1057
1058 static int com_prune(void)
1059 {
1060         int ret;
1061         struct snapshot_list sl;
1062         struct disk_space ds;
1063
1064         ret = get_disk_space(".", &ds);
1065         if (ret < 0)
1066                 return ret;
1067         log_disk_space(&ds);
1068         dss_get_snapshot_list(&sl);
1069         ret = remove_outdated_snapshot(&sl);
1070         if (ret < 0)
1071                 goto out;
1072         if (ret > 0)
1073                 goto rm;
1074         ret = remove_redundant_snapshot(&sl);
1075         if (ret < 0)
1076                 goto out;
1077         if (ret > 0)
1078                 goto rm;
1079         ret = 0;
1080         goto out;
1081 rm:
1082         if (snapshot_removal_status == HS_PRE_RUNNING) {
1083                 ret = wait_for_remove_process();
1084                 if (ret < 0)
1085                         goto out;
1086                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1087                         goto out;
1088         }
1089         ret = exec_rm();
1090         if (ret < 0)
1091                 goto out;
1092         ret = wait_for_remove_process();
1093         if (ret < 0)
1094                 goto out;
1095         if (snapshot_removal_status != HS_SUCCESS)
1096                 goto out;
1097         ret = post_remove_hook();
1098         if (ret < 0)
1099                 goto out;
1100         if (snapshot_removal_status != HS_POST_RUNNING)
1101                 goto out;
1102         ret = wait_for_remove_process();
1103         if (ret < 0)
1104                 goto out;
1105         ret = 1;
1106 out:
1107         free_snapshot_list(&sl);
1108         return ret;
1109 }
1110
1111 static int com_create(void)
1112 {
1113         int ret, status;
1114         char **rsync_argv;
1115
1116         if (conf.dry_run_given) {
1117                 int i;
1118                 char *msg = NULL;
1119                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1120                 for (i = 0; rsync_argv[i]; i++) {
1121                         char *tmp = msg;
1122                         msg = make_message("%s%s%s", tmp? tmp : "",
1123                                 tmp? " " : "", rsync_argv[i]);
1124                         free(tmp);
1125                 }
1126                 free_rsync_argv(rsync_argv);
1127                 dss_msg("%s\n", msg);
1128                 free(msg);
1129                 return 1;
1130         }
1131         ret = pre_create_hook();
1132         if (ret < 0)
1133                 return ret;
1134         if (create_pid) {
1135                 ret = wait_for_process(create_pid, &status);
1136                 if (ret < 0)
1137                         return ret;
1138                 ret = handle_pre_create_hook_exit(status);
1139                 if (ret <= 0) /* error, or pre-create failed */
1140                         return ret;
1141         }
1142         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1143         ret = create_snapshot(rsync_argv);
1144         if (ret < 0)
1145                 goto out;
1146         ret = wait_for_process(create_pid, &status);
1147         if (ret < 0)
1148                 goto out;
1149         ret = handle_rsync_exit(status);
1150         if (ret < 0)
1151                 goto out;
1152         post_create_hook();
1153         if (create_pid)
1154                 ret = wait_for_process(create_pid, &status);
1155 out:
1156         free_rsync_argv(rsync_argv);
1157         return ret;
1158 }
1159
1160 static int com_ls(void)
1161 {
1162         int i;
1163         struct snapshot_list sl;
1164         struct snapshot *s;
1165
1166         dss_get_snapshot_list(&sl);
1167         FOR_EACH_SNAPSHOT(s, i, &sl) {
1168                 int64_t d = 0;
1169                 if (s->flags & SS_COMPLETE)
1170                         d = (s->completion_time - s->creation_time) / 60;
1171                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1172         };
1173         free_snapshot_list(&sl);
1174         return 1;
1175 }
1176
1177 static int setup_signal_handling(void)
1178 {
1179         int ret;
1180
1181         DSS_INFO_LOG("setting up signal handlers\n");
1182         signal_pipe = signal_init(); /* always successful */
1183         ret = install_sighandler(SIGINT);
1184         if (ret < 0)
1185                 return ret;
1186         ret = install_sighandler(SIGTERM);
1187         if (ret < 0)
1188                 return ret;
1189         return install_sighandler(SIGCHLD);
1190 }
1191
1192 /**
1193  * The main function of dss.
1194  *
1195  * \param argc Usual argument count.
1196  * \param argv Usual argument vector.
1197  */
1198 int main(int argc, char **argv)
1199 {
1200         int ret;
1201         struct cmdline_parser_params params = {
1202                 .override = 0,
1203                 .initialize = 1,
1204                 .check_required = 0,
1205                 .check_ambiguity = 0,
1206                 .print_errors = 1
1207         };
1208
1209         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1210         ret = parse_config_file(0);
1211         if (ret < 0)
1212                 goto out;
1213         if (conf.daemon_given)
1214                 daemon_init();
1215         ret = change_to_dest_dir();
1216         if (ret < 0)
1217                 goto out;
1218         ret = setup_signal_handling();
1219         if (ret < 0)
1220                 goto out;
1221         ret = call_command_handler();
1222 out:
1223         if (ret < 0)
1224                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1225         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1226 }