dss.c

   1 #include <string.h>
   2 #include <stdlib.h>
   3 #include <stdarg.h>
   4 #include <assert.h>
   5 #include <errno.h>
   6 #include <sys/types.h>
   7 #include <signal.h>
   8 #include <ctype.h>
   9 #include <sys/stat.h>
  10 #include <unistd.h>
  11 #include <inttypes.h>
  12 #include <sys/time.h>
  13 #include <time.h>
  14 #include <sys/wait.h>
  15 #include <fnmatch.h>
  16 #include <limits.h>
  17
  18
  19 #include "gcc-compat.h"
  20 #include "cmdline.h"
  21 #include "log.h"
  22 #include "string.h"
  23 #include "error.h"
  24 #include "fd.h"
  25 #include "exec.h"
  26 #include "daemon.h"
  27 #include "signal.h"
  28 #include "df.h"
  29 #include "time.h"
  30
  31
  32 struct gengetopt_args_info conf;
  33 static FILE *logfile;
  34 static int signal_pipe;
  35
  36 /** Process id of current rsync process. */
  37 static pid_t rsync_pid;
  38 /** Whether the rsync process is currently stopped */
  39 static int rsync_stopped;
  40 /** Process id of current rm process. */
  41 static pid_t rm_pid;
  42 /** When the next snapshot is due. */
  43 struct timeval next_snapshot_time;
  44 /** The pid of the pre-create hook. */
  45 pid_t pre_create_hook_pid;
  46 /** The pid of the post-create hook. */
  47 pid_t post_create_hook_pid;
  48
  49 /* Creation time of the snapshot currently being created. */
  50 int64_t current_snapshot_creation_time;
  51
  52 static char *path_to_last_complete_snapshot;
  53
  54 enum {
  55         /** We are ready to take the next snapshot. */
  56         SCS_READY,
  57         /** The pre-creation hook has been started. */
  58         SCS_PRE_HOOK_RUNNING,
  59         /** The pre-creation hook exited successfully. */
  60         SCS_PRE_HOOK_SUCCESS,
  61         /** The rsync process is running. */
  62         SCS_RSYNC_RUNNING,
  63         /** The rsync process exited successfully. */
  64         SCS_RSYNC_SUCCESS,
  65         /** The post-create hook has been started- */
  66         SCS_POST_HOOK_RUNNING,
  67 };
  68
  69 static unsigned snapshot_creation_status;
  70
  71
  72 DEFINE_DSS_ERRLIST;
  73
  74
  75 /* a litte cpp magic helps to DRY */
  76 #define COMMANDS \
  77         COMMAND(ls) \
  78         COMMAND(create) \
  79         COMMAND(prune) \
  80         COMMAND(run)
  81 #define COMMAND(x) int com_ ##x(void);
  82 COMMANDS
  83 #undef COMMAND
  84 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
  85 int call_command_handler(void)
  86 {
  87         COMMANDS
  88         DSS_EMERG_LOG("BUG: did not find command handler\n");
  89         exit(EXIT_FAILURE);
  90 }
  91 #undef COMMAND
  92 #undef COMMANDS
  93
  94 /*
  95  * complete, not being deleted: 1204565370-1204565371.Sun_Mar_02_2008_14_33-Sun_Mar_02_2008_14_43
  96  * complete, being deleted: 1204565370-1204565371.being_deleted
  97  * incomplete, not being deleted: 1204565370-incomplete
  98  * incomplete, being deleted: 1204565370-incomplete.being_deleted
  99  */
 100 enum snapshot_status_flags {
 101         /** The rsync process terminated successfully. */
 102         SS_COMPLETE = 1,
 103         /** The rm process is running to remove this snapshot. */
 104         SS_BEING_DELETED = 2,
 105 };
 106
 107 struct snapshot {
 108         char *name;
 109         int64_t creation_time;
 110         int64_t completion_time;
 111         enum snapshot_status_flags flags;
 112         unsigned interval;
 113 };
 114
 115 __printf_2_3 void dss_log(int ll, const char* fmt,...)
 116 {
 117         va_list argp;
 118         FILE *outfd;
 119         struct tm *tm;
 120         time_t t1;
 121         char str[255] = "";
 122
 123         if (ll < conf.loglevel_arg)
 124                 return;
 125         outfd = logfile? logfile : stderr;
 126         time(&t1);
 127         tm = localtime(&t1);
 128         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 129         fprintf(outfd, "%s ", str);
 130         if (conf.loglevel_arg <= INFO)
 131                 fprintf(outfd, "%i: ", ll);
 132         va_start(argp, fmt);
 133         vfprintf(outfd, fmt, argp);
 134         va_end(argp);
 135 }
 136
 137 /**
 138  * Print a message either to stdout or to the log file.
 139  */
 140 __printf_1_2 void dss_msg(const char* fmt,...)
 141 {
 142         FILE *outfd = conf.daemon_given? logfile : stdout;
 143         va_list argp;
 144         va_start(argp, fmt);
 145         vfprintf(outfd, fmt, argp);
 146         va_end(argp);
 147 }
 148
 149 /**
 150  * Return the desired number of snapshots of an interval.
 151  */
 152 unsigned num_snapshots(int interval)
 153 {
 154         unsigned n;
 155
 156         assert(interval >= 0);
 157
 158         if (interval >= conf.num_intervals_arg)
 159                 return 0;
 160         n = conf.num_intervals_arg - interval - 1;
 161         return 1 << n;
 162 }
 163
 164 /* return: Whether dirname is a snapshot directory (0: no, 1: yes) */
 165 int is_snapshot(const char *dirname, int64_t now, struct snapshot *s)
 166 {
 167         int i, ret;
 168         char *dash, *dot, *tmp;
 169         int64_t num;
 170
 171         assert(dirname);
 172         dash = strchr(dirname, '-');
 173         if (!dash || !dash[1] || dash == dirname)
 174                 return 0;
 175         for (i = 0; dirname[i] != '-'; i++)
 176                 if (!isdigit(dirname[i]))
 177                         return 0;
 178         tmp = dss_strdup(dirname);
 179         tmp[i] = '\0';
 180         ret = dss_atoi64(tmp, &num);
 181         free(tmp);
 182         if (ret < 0)
 183                 return 0;
 184         assert(num >= 0);
 185         if (num > now)
 186                 return 0;
 187         s->creation_time = num;
 188         //DSS_DEBUG_LOG("%s start time: %lli\n", dirname, (long long)s->creation_time);
 189         s->interval = (long long) ((now - s->creation_time)
 190                 / conf.unit_interval_arg / 24 / 3600);
 191         if (!strcmp(dash + 1, "incomplete")) {
 192                 s->completion_time = -1;
 193                 s->flags = 0; /* neither complete, nor being deleted */
 194                 goto success;
 195         }
 196         if (!strcmp(dash + 1, "incomplete.being_deleted")) {
 197                 s->completion_time = -1;
 198                 s->flags = SS_BEING_DELETED; /* mot cpmplete, being deleted */
 199                 goto success;
 200         }
 201         tmp = dash + 1;
 202         dot = strchr(tmp, '.');
 203         if (!dot || !dot[1] || dot == tmp)
 204                 return 0;
 205         for (i = 0; tmp[i] != '.'; i++)
 206                 if (!isdigit(tmp[i]))
 207                         return 0;
 208         tmp = dss_strdup(dash + 1);
 209         tmp[i] = '\0';
 210         ret = dss_atoi64(tmp, &num);
 211         free(tmp);
 212         if (ret < 0)
 213                 return 0;
 214         if (num > now)
 215                 return 0;
 216         s->completion_time = num;
 217         s->flags = SS_COMPLETE;
 218         if (!strcmp(dot + 1, "being_deleted"))
 219                 s->flags |= SS_BEING_DELETED;
 220 success:
 221         s->name = dss_strdup(dirname);
 222         return 1;
 223 }
 224
 225 int64_t get_current_time(void)
 226 {
 227         time_t now;
 228         time(&now);
 229         DSS_DEBUG_LOG("now: %lli\n", (long long) now);
 230         return (int64_t)now;
 231 }
 232
 233 char *incomplete_name(int64_t start)
 234 {
 235         return make_message("%lli-incomplete", (long long)start);
 236 }
 237
 238 char *being_deleted_name(struct snapshot *s)
 239 {
 240         if (s->flags & SS_COMPLETE)
 241                 return make_message("%lli-%lli.being_deleted",
 242                         (long long)s->creation_time,
 243                         (long long)s->completion_time);
 244         return make_message("%lli-incomplete.being_deleted",
 245                 (long long)s->creation_time);
 246 }
 247
 248 int complete_name(int64_t start, int64_t end, char **result)
 249 {
 250         struct tm start_tm, end_tm;
 251         time_t *start_seconds = (time_t *) (uint64_t *)&start; /* STFU, gcc */
 252         time_t *end_seconds = (time_t *) (uint64_t *)&end; /* STFU, gcc */
 253         char start_str[200], end_str[200];
 254
 255         if (!localtime_r(start_seconds, &start_tm))
 256                 return -E_LOCALTIME;
 257         if (!localtime_r(end_seconds, &end_tm))
 258                 return -E_LOCALTIME;
 259         if (!strftime(start_str, sizeof(start_str), "%a_%b_%d_%Y_%H_%M_%S", &start_tm))
 260                 return -E_STRFTIME;
 261         if (!strftime(end_str, sizeof(end_str), "%a_%b_%d_%Y_%H_%M_%S", &end_tm))
 262                 return -E_STRFTIME;
 263         *result = make_message("%lli-%lli.%s-%s", (long long) start, (long long) end,
 264                 start_str, end_str);
 265         return 1;
 266 }
 267
 268 struct snapshot_list {
 269         int64_t now;
 270         unsigned num_snapshots;
 271         unsigned array_size;
 272         struct snapshot **snapshots;
 273         /**
 274          * Array of size num_intervals + 1
 275          *
 276          * It contains the number of snapshots in each interval. interval_count[num_intervals]
 277          * is the number of snapshots which belong to any interval greater than num_intervals.
 278          */
 279         unsigned *interval_count;
 280 };
 281
 282 #define FOR_EACH_SNAPSHOT(s, i, sl) \
 283         for ((i) = 0; (i) < (sl)->num_snapshots && ((s) = (sl)->snapshots[(i)]); (i)++)
 284
 285 #define FOR_EACH_SNAPSHOT_REVERSE(s, i, sl) \
 286         for ((i) = (sl)->num_snapshots; (i) > 0 && ((s) = (sl)->snapshots[(i - 1)]); (i)--)
 287
 288 static inline struct snapshot *oldest_snapshot(struct snapshot_list *sl)
 289 {
 290         if (!sl->num_snapshots)
 291                 return NULL;
 292         return sl->snapshots[0];
 293 }
 294
 295 #define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
 296
 297 static int compare_snapshots(const void *a, const void *b)
 298 {
 299         struct snapshot *s1 = *(struct snapshot **)a;
 300         struct snapshot *s2 = *(struct snapshot **)b;
 301         return NUM_COMPARE(s2->creation_time, s1->creation_time);
 302 }
 303
 304 /** Compute the minimum of \a a and \a b. */
 305 #define DSS_MIN(a,b) ((a) < (b) ? (a) : (b))
 306
 307 int add_snapshot(const char *dirname, void *private)
 308 {
 309         struct snapshot_list *sl = private;
 310         struct snapshot s;
 311         int ret = is_snapshot(dirname, sl->now, &s);
 312
 313         if (!ret)
 314                 return 1;
 315         if (sl->num_snapshots >= sl->array_size) {
 316                 sl->array_size = 2 * sl->array_size + 1;
 317                 sl->snapshots = dss_realloc(sl->snapshots,
 318                         sl->array_size * sizeof(struct snapshot *));
 319         }
 320         sl->snapshots[sl->num_snapshots] = dss_malloc(sizeof(struct snapshot));
 321         *(sl->snapshots[sl->num_snapshots]) = s;
 322         sl->interval_count[DSS_MIN(s.interval, conf.num_intervals_arg)]++;
 323         sl->num_snapshots++;
 324         return 1;
 325 }
 326
 327 void get_snapshot_list(struct snapshot_list *sl)
 328 {
 329         sl->now = get_current_time();
 330         sl->num_snapshots = 0;
 331         sl->array_size = 0;
 332         sl->snapshots = NULL;
 333         sl->interval_count = dss_calloc((conf.num_intervals_arg + 1) * sizeof(unsigned));
 334         for_each_subdir(add_snapshot, sl);
 335         qsort(sl->snapshots, sl->num_snapshots, sizeof(struct snapshot *),
 336                 compare_snapshots);
 337 }
 338
 339 void free_snapshot_list(struct snapshot_list *sl)
 340 {
 341         int i;
 342         struct snapshot *s;
 343
 344         FOR_EACH_SNAPSHOT(s, i, sl) {
 345                 free(s->name);
 346                 free(s);
 347         }
 348         free(sl->interval_count);
 349         sl->interval_count = NULL;
 350         free(sl->snapshots);
 351         sl->snapshots = NULL;
 352         sl->num_snapshots = 0;
 353 }
 354
 355 void stop_rsync_process(void)
 356 {
 357         if (!rsync_pid || rsync_stopped)
 358                 return;
 359         kill(SIGSTOP, rsync_pid);
 360         rsync_stopped = 1;
 361 }
 362
 363 void restart_rsync_process(void)
 364 {
 365         if (!rsync_pid || !rsync_stopped)
 366                 return;
 367         kill (SIGCONT, rsync_pid);
 368         rsync_stopped = 0;
 369 }
 370
 371 /**
 372  * Print a log message about the exit status of a child.
 373  */
 374 void log_termination_msg(pid_t pid, int status)
 375 {
 376         if (WIFEXITED(status))
 377                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 378                         WEXITSTATUS(status));
 379         else if (WIFSIGNALED(status))
 380                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 381                         WTERMSIG(status));
 382         else
 383                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 384 }
 385
 386 int wait_for_process(pid_t pid, int *status)
 387 {
 388         int ret;
 389
 390         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 391         for (;;) {
 392                 fd_set rfds;
 393
 394                 FD_ZERO(&rfds);
 395                 FD_SET(signal_pipe, &rfds);
 396                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 397                 if (ret < 0)
 398                         break;
 399                 ret = next_signal();
 400                 if (!ret)
 401                         continue;
 402                 if (ret == SIGCHLD) {
 403                         ret = waitpid(pid, status, 0);
 404                         if (ret >= 0)
 405                                 break;
 406                         if (errno != EINTR) { /* error */
 407                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 408                                 break;
 409                         }
 410                 }
 411                 /* SIGINT or SIGTERM */
 412                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 413                 kill(pid, SIGTERM);
 414         }
 415         if (ret < 0)
 416                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 417         else
 418                 log_termination_msg(pid, *status);
 419         return ret;
 420 }
 421
 422 int remove_snapshot(struct snapshot *s)
 423 {
 424         int fds[3] = {0, 0, 0};
 425         assert(!rm_pid);
 426         char *new_name = being_deleted_name(s);
 427         int ret = dss_rename(s->name, new_name);
 428         char *argv[] = {"rm", "-rf", new_name, NULL};
 429
 430         if (ret < 0)
 431                 goto out;
 432         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 433         stop_rsync_process();
 434         ret = dss_exec(&rm_pid, argv[0], argv, fds);
 435 out:
 436         free(new_name);
 437         return ret;
 438 }
 439
 440 /*
 441  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 442  */
 443 int remove_redundant_snapshot(struct snapshot_list *sl)
 444 {
 445         int ret, i, interval;
 446         struct snapshot *s;
 447         unsigned missing = 0;
 448
 449         DSS_INFO_LOG("looking for intervals containing too many snapshots\n");
 450         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 451                 unsigned keep = num_snapshots(interval);
 452                 unsigned num = sl->interval_count[interval];
 453                 struct snapshot *victim = NULL, *prev = NULL;
 454                 int64_t score = LONG_MAX;
 455
 456                 if (keep >= num)
 457                         missing += keep - num;
 458 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 459 //                      interval, keep, num, missing);
 460                 if (keep + missing >= num)
 461                         continue;
 462                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 463                 FOR_EACH_SNAPSHOT(s, i, sl) {
 464                         int64_t this_score;
 465
 466                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 467                         if (s->interval > interval) {
 468                                 prev = s;
 469                                 continue;
 470                         }
 471                         if (s->interval < interval)
 472                                 break;
 473                         if (!victim) {
 474                                 victim = s;
 475                                 prev = s;
 476                                 continue;
 477                         }
 478                         assert(prev);
 479                         /* check if s is a better victim */
 480                         this_score = s->creation_time - prev->creation_time;
 481                         assert(this_score >= 0);
 482                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 483                         if (this_score < score) {
 484                                 score = this_score;
 485                                 victim = s;
 486                         }
 487                         prev = s;
 488                 }
 489                 assert(victim);
 490                 if (conf.dry_run_given) {
 491                         dss_msg("%s would be removed (interval = %i)\n",
 492                                 victim->name, victim->interval);
 493                         continue;
 494                 }
 495                 ret = remove_snapshot(victim);
 496                 return ret < 0? ret : 1;
 497         }
 498         return 0;
 499 }
 500
 501 int remove_outdated_snapshot(struct snapshot_list *sl)
 502 {
 503         int i, ret;
 504         struct snapshot *s;
 505
 506         DSS_INFO_LOG("looking for snapshots belonging to intervals greater than %d\n",
 507                 conf.num_intervals_arg);
 508         FOR_EACH_SNAPSHOT(s, i, sl) {
 509                 if (s->interval <= conf.num_intervals_arg)
 510                         continue;
 511                 if (conf.dry_run_given) {
 512                         dss_msg("%s would be removed (interval = %i)\n",
 513                                 s->name, s->interval);
 514                         continue;
 515                 }
 516                 ret = remove_snapshot(s);
 517                 if (ret < 0)
 518                         return ret;
 519                 return 1;
 520         }
 521         return 0;
 522 }
 523
 524 int handle_rm_exit(int status)
 525 {
 526         rm_pid = 0;
 527         if (!WIFEXITED(status))
 528                 return -E_INVOLUNTARY_EXIT;
 529         if (WEXITSTATUS(status))
 530                 return -E_BAD_EXIT_CODE;
 531         return 1;
 532 }
 533
 534 int wait_for_rm_process(void)
 535 {
 536         int status, ret = wait_for_process(rm_pid, &status);
 537
 538         if (ret < 0)
 539                 return ret;
 540         return handle_rm_exit(status);
 541 }
 542
 543 void kill_process(pid_t pid)
 544 {
 545         if (!pid)
 546                 return;
 547         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 548         kill(pid, SIGTERM);
 549 }
 550
 551 int check_config(void)
 552 {
 553         if (conf.unit_interval_arg <= 0) {
 554                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 555                 return -E_INVALID_NUMBER;
 556         }
 557         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 558         if (conf.num_intervals_arg <= 0) {
 559                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 560                 return -E_INVALID_NUMBER;
 561         }
 562         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 563         return 1;
 564 }
 565
 566 /* exits on errors */
 567 void parse_config_file(int override)
 568 {
 569         int ret;
 570         char *config_file;
 571         struct stat statbuf;
 572         char *old_logfile_arg = NULL;
 573         int old_daemon_given = 0;
 574
 575         if (conf.config_file_given)
 576                 config_file = dss_strdup(conf.config_file_arg);
 577         else {
 578                 char *home = get_homedir();
 579                 config_file = make_message("%s/.dssrc", home);
 580                 free(home);
 581         }
 582         if (override) { /* SIGHUP */
 583                 if (conf.logfile_given)
 584                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 585                 old_daemon_given = conf.daemon_given;
 586         }
 587
 588         ret = stat(config_file, &statbuf);
 589         if (ret && conf.config_file_given) {
 590                 ret = -ERRNO_TO_DSS_ERROR(errno);
 591                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 592                 goto out;
 593         }
 594         if (!ret) {
 595                 struct cmdline_parser_params params = {
 596                         .override = override,
 597                         .initialize = 0,
 598                         .check_required = 1,
 599                         .check_ambiguity = 0
 600                 };
 601                 cmdline_parser_config_file(config_file, &conf, &params);
 602         }
 603         ret = check_config();
 604         if (ret < 0)
 605                 goto out;
 606         if (override) {
 607                 /* don't change daemon mode on SIGHUP */
 608                 conf.daemon_given = old_daemon_given;
 609                 close_log(logfile);
 610                 logfile = NULL;
 611                 if (conf.logfile_given)
 612                         free(old_logfile_arg);
 613                 else if (conf.daemon_given) { /* re-use old logfile */
 614                         conf.logfile_arg = old_logfile_arg;
 615                         conf.logfile_given = 1;
 616                 }
 617         }
 618         if (conf.logfile_given) {
 619                 logfile = open_log(conf.logfile_arg);
 620                 log_welcome(conf.loglevel_arg);
 621         }
 622         DSS_EMERG_LOG("loglevel: %d\n", conf.loglevel_arg);
 623 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 624         ret = dss_chdir(conf.dest_dir_arg);
 625 out:
 626         free(config_file);
 627         if (ret >= 0)
 628                 return;
 629         DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 630         exit(EXIT_FAILURE);
 631 }
 632
 633 void handle_sighup(void)
 634 {
 635         DSS_NOTICE_LOG("SIGHUP\n");
 636         parse_config_file(1);
 637 }
 638
 639 int rename_incomplete_snapshot(int64_t start)
 640 {
 641         char *old_name;
 642         int ret;
 643
 644         free(path_to_last_complete_snapshot);
 645         ret = complete_name(start, get_current_time(),
 646                 &path_to_last_complete_snapshot);
 647         if (ret < 0)
 648                 return ret;
 649         old_name = incomplete_name(start);
 650         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 651         if (ret >= 0)
 652                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 653                         path_to_last_complete_snapshot);
 654         free(old_name);
 655         return ret;
 656 }
 657
 658 void compute_next_snapshot_time(void)
 659 {
 660         struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 661                 tmp, diff;
 662         int64_t x = 0;
 663         unsigned wanted = num_snapshots(0), num_complete_snapshots = 0;
 664         int i, ret;
 665         struct snapshot *s = NULL;
 666         struct snapshot_list sl;
 667
 668         assert(snapshot_creation_status == SCS_READY);
 669         current_snapshot_creation_time = 0;
 670         get_snapshot_list(&sl);
 671         FOR_EACH_SNAPSHOT(s, i, &sl) {
 672                 if (!(s->flags & SS_COMPLETE))
 673                         continue;
 674                 num_complete_snapshots++;
 675                 x += s->completion_time - s->creation_time;
 676         }
 677         assert(x >= 0);
 678         if (num_complete_snapshots)
 679                 x /= num_complete_snapshots; /* avg time to create one snapshot */
 680         x *= wanted; /* time to create all snapshots in interval 0 */
 681         tmp.tv_sec = x;
 682         tmp.tv_usec = 0;
 683         ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
 684         gettimeofday(&now, NULL);
 685         if (ret < 0 || !s)
 686                 goto min_sleep;
 687         tv_divide(wanted, &diff, &tmp); /* sleep time betweeen two snapshots */
 688         diff.tv_sec = s->completion_time;
 689         diff.tv_usec = 0;
 690         tv_add(&diff, &tmp, &next_snapshot_time);
 691         if (tv_diff(&now, &next_snapshot_time, NULL) < 0)
 692                 goto out;
 693 min_sleep:
 694         next_snapshot_time = now;
 695         next_snapshot_time.tv_sec += 60;
 696 out:
 697         free_snapshot_list(&sl);
 698 }
 699
 700 int handle_rsync_exit(int status)
 701 {
 702         int es, ret;
 703
 704         if (!WIFEXITED(status)) {
 705                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)rsync_pid);
 706                 ret = -E_INVOLUNTARY_EXIT;
 707                 snapshot_creation_status = SCS_READY;
 708                 compute_next_snapshot_time();
 709                 goto out;
 710         }
 711         es = WEXITSTATUS(status);
 712         if (es != 0 && es != 23 && es != 24) {
 713                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)rsync_pid, es);
 714                 ret = -E_BAD_EXIT_CODE;
 715                 snapshot_creation_status = SCS_READY;
 716                 compute_next_snapshot_time();
 717                 goto out;
 718         }
 719         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 720         if (ret < 0)
 721                 goto out;
 722         snapshot_creation_status = SCS_RSYNC_SUCCESS;
 723 out:
 724         rsync_pid = 0;
 725         rsync_stopped = 0;
 726         return ret;
 727 }
 728
 729 __malloc char *name_of_newest_complete_snapshot(void)
 730 {
 731         struct snapshot_list sl;
 732         struct snapshot *s;
 733         int i;
 734         char *name = NULL;
 735
 736         get_snapshot_list(&sl);
 737
 738         FOR_EACH_SNAPSHOT_REVERSE(s, i, &sl) {
 739                 if (s->flags != SS_COMPLETE) /* incomplete or being deleted */
 740                         continue;
 741                 name = dss_strdup(s->name);
 742                 break;
 743         }
 744         free_snapshot_list(&sl);
 745         return name;
 746 }
 747
 748 void create_rsync_argv(char ***argv, int64_t *num)
 749 {
 750         char *logname, *newest = name_of_newest_complete_snapshot();
 751         int i = 0, j;
 752
 753         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 754         (*argv)[i++] = dss_strdup("rsync");
 755         (*argv)[i++] = dss_strdup("-aq");
 756         (*argv)[i++] = dss_strdup("--delete");
 757         for (j = 0; j < conf.rsync_option_given; j++)
 758                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 759         if (newest) {
 760                 DSS_INFO_LOG("using %s as reference snapshot\n", newest);
 761                 (*argv)[i++] = make_message("--link-dest=../%s", newest);
 762                 free(newest);
 763         } else
 764                 DSS_INFO_LOG("no previous snapshot found\n");
 765         if (conf.exclude_patterns_given) {
 766                 (*argv)[i++] = dss_strdup("--exclude-from");
 767                 (*argv)[i++] = dss_strdup(conf.exclude_patterns_arg);
 768
 769         }
 770         logname = dss_logname();
 771         if (conf.remote_user_given && !strcmp(conf.remote_user_arg, logname))
 772                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 773         else
 774                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 775                         conf.remote_user_arg : logname,
 776                         conf.remote_host_arg, conf.source_dir_arg);
 777         free(logname);
 778         *num = get_current_time();
 779         (*argv)[i++] = incomplete_name(*num);
 780         (*argv)[i++] = NULL;
 781         for (j = 0; j < i; j++)
 782                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 783 }
 784
 785 void free_rsync_argv(char **argv)
 786 {
 787         int i;
 788         for (i = 0; argv[i]; i++)
 789                 free(argv[i]);
 790         free(argv);
 791 }
 792
 793 int pre_create_hook(void)
 794 {
 795         int ret, fds[3] = {0, 0, 0};
 796
 797         if (!conf.pre_create_hook_given) {
 798                 snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
 799                 return 0;
 800         }
 801         DSS_NOTICE_LOG("executing %s\n", conf.pre_create_hook_arg);
 802         ret = dss_exec_cmdline_pid(&pre_create_hook_pid,
 803                 conf.pre_create_hook_arg, fds);
 804         if (ret < 0)
 805                 return ret;
 806         snapshot_creation_status = SCS_PRE_HOOK_RUNNING;
 807         return ret;
 808 }
 809
 810 int post_create_hook(void)
 811 {
 812         int ret, fds[3] = {0, 0, 0};
 813         char *cmd;
 814
 815         if (!conf.post_create_hook_given) {
 816                 snapshot_creation_status = SCS_READY;
 817                 compute_next_snapshot_time();
 818                 return 0;
 819         }
 820         cmd = make_message("%s %s", conf.post_create_hook_arg,
 821                 path_to_last_complete_snapshot);
 822         DSS_NOTICE_LOG("executing %s\n", cmd);
 823         ret = dss_exec_cmdline_pid(&post_create_hook_pid, cmd, fds);
 824         free(cmd);
 825         if (ret < 0)
 826                 return ret;
 827         snapshot_creation_status = SCS_POST_HOOK_RUNNING;
 828         return ret;
 829 }
 830
 831 int create_snapshot(char **argv)
 832 {
 833         int ret, fds[3] = {0, 0, 0};
 834         char *name;
 835
 836         name = incomplete_name(current_snapshot_creation_time);
 837         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 838         free(name);
 839         ret = dss_exec(&rsync_pid, argv[0], argv, fds);
 840         if (ret < 0)
 841                 return ret;
 842         snapshot_creation_status = SCS_RSYNC_RUNNING;
 843         return ret;
 844 }
 845
 846 int handle_pre_create_hook_exit(int status)
 847 {
 848         int es, ret;
 849
 850         if (!WIFEXITED(status)) {
 851                 snapshot_creation_status = SCS_READY;
 852                 compute_next_snapshot_time();
 853                 ret = -E_INVOLUNTARY_EXIT;
 854                 goto out;
 855         }
 856         es = WEXITSTATUS(status);
 857         if (es) {
 858                 snapshot_creation_status = SCS_READY;
 859                 compute_next_snapshot_time();
 860                 ret = -E_BAD_EXIT_CODE;
 861                 goto out;
 862         }
 863         snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
 864         ret = 1;
 865 out:
 866         pre_create_hook_pid = 0;
 867         return ret;
 868 }
 869
 870 int handle_sigchld()
 871 {
 872         pid_t pid;
 873         int status, ret = reap_child(&pid, &status);
 874
 875         if (ret <= 0)
 876                 return ret;
 877         if (pid == rsync_pid)
 878                 return handle_rsync_exit(status);
 879         if (pid == rm_pid)
 880                 return handle_rm_exit(status);
 881         if (pid == pre_create_hook_pid)
 882                 return handle_pre_create_hook_exit(status);
 883         if (pid == post_create_hook_pid) {
 884                 snapshot_creation_status = SCS_READY;
 885                 compute_next_snapshot_time();
 886                 return 1;
 887         }
 888         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 889         exit(EXIT_FAILURE);
 890 }
 891
 892 void handle_signal(void)
 893 {
 894         int sig, ret = next_signal();
 895
 896         if (ret <= 0)
 897                 goto out;
 898         sig = ret;
 899         switch (sig) {
 900         case SIGINT:
 901         case SIGTERM:
 902                 restart_rsync_process();
 903                 kill_process(rsync_pid);
 904                 kill_process(rm_pid);
 905                 exit(EXIT_FAILURE);
 906         case SIGHUP:
 907                 handle_sighup();
 908                 ret = 1;
 909                 break;
 910         case SIGCHLD:
 911                 ret = handle_sigchld();
 912                 break;
 913         }
 914 out:
 915         if (ret < 0)
 916                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 917 }
 918
 919 int remove_oldest_snapshot(struct snapshot_list *sl)
 920 {
 921         struct snapshot *s = oldest_snapshot(sl);
 922
 923         if (!s) /* no snapshot found */
 924                 return 0;
 925         DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
 926         if (s->creation_time == current_snapshot_creation_time)
 927                 return 0; /* do not remove the snapshot currently being created */
 928         return remove_snapshot(s);
 929 }
 930
 931 /* TODO: Also consider number of inodes. */
 932 int disk_space_low(void)
 933 {
 934         struct disk_space ds;
 935         int ret = get_disk_space(".", &ds);
 936
 937         if (ret < 0)
 938                 return ret;
 939         if (conf.min_free_mb_arg)
 940                 if (ds.free_mb < conf.min_free_mb_arg)
 941                         return 1;
 942         if (conf.min_free_percent_arg)
 943                 if (ds.percent_free < conf.min_free_percent_arg)
 944                         return 1;
 945         return 0;
 946 }
 947
 948 int try_to_free_disk_space(int low_disk_space)
 949 {
 950         int ret;
 951         struct snapshot_list sl;
 952
 953         get_snapshot_list(&sl);
 954         ret = remove_outdated_snapshot(&sl);
 955         if (ret) /* error, or we are removing something */
 956                 goto out;
 957         /* no outdated snapshot */
 958         ret = remove_redundant_snapshot(&sl);
 959         if (ret)
 960                 goto out;
 961         ret = 0;
 962         if (!low_disk_space)
 963                 goto out;
 964         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 965         ret = remove_oldest_snapshot(&sl);
 966         if (ret)
 967                 goto out;
 968         DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
 969         ret= -ENOSPC;
 970 out:
 971         free_snapshot_list(&sl);
 972         return ret;
 973 }
 974
 975 int select_loop(void)
 976 {
 977         int ret;
 978         struct timeval tv = {.tv_sec = 0, .tv_usec = 0};
 979
 980         for (;;) {
 981                 fd_set rfds;
 982                 int low_disk_space;
 983                 char **rsync_argv;
 984                 struct timeval now, *tvp = &tv;
 985
 986                 if (rsync_pid)
 987                         tv.tv_sec = 60; /* check every 60 seconds for free disk space */
 988                 else if (rm_pid)
 989                         tvp = NULL; /* sleep until rm process dies */
 990                 FD_ZERO(&rfds);
 991                 FD_SET(signal_pipe, &rfds);
 992                 DSS_DEBUG_LOG("tvp: %p, tv_sec: %lu\n", tvp, (long unsigned) tv.tv_sec);
 993                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
 994                 if (ret < 0)
 995                         return ret;
 996                 if (FD_ISSET(signal_pipe, &rfds))
 997                         handle_signal();
 998                 if (rm_pid)
 999                         continue;
1000                 ret = disk_space_low();
1001                 if (ret < 0)
1002                         break;
1003                 low_disk_space = ret;
1004                 if (low_disk_space)
1005                         stop_rsync_process();
1006                 ret = try_to_free_disk_space(low_disk_space);
1007                 if (ret < 0)
1008                         break;
1009                 if (rm_pid)
1010                         continue;
1011                 restart_rsync_process();
1012                 gettimeofday(&now, NULL);
1013                 if (tv_diff(&next_snapshot_time, &now, &tv) > 0)
1014                         continue;
1015                 switch (snapshot_creation_status) {
1016                 case SCS_READY:
1017                         ret = pre_create_hook();
1018                         if (ret < 0)
1019                                 goto out;
1020                         continue;
1021                 case SCS_PRE_HOOK_RUNNING:
1022                         continue;
1023                 case SCS_PRE_HOOK_SUCCESS:
1024                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1025                         ret = create_snapshot(rsync_argv);
1026                         free_rsync_argv(rsync_argv);
1027                         if (ret < 0)
1028                                 goto out;
1029                         continue;
1030                 case SCS_RSYNC_RUNNING:
1031                         continue;
1032                 case SCS_RSYNC_SUCCESS:
1033                         ret = post_create_hook();
1034                         if (ret < 0)
1035                                 goto out;
1036                         continue;
1037                 case SCS_POST_HOOK_RUNNING:
1038                         continue;
1039                 }
1040         }
1041 out:
1042         return ret;
1043 }
1044
1045 int com_run(void)
1046 {
1047         int ret;
1048
1049         if (conf.dry_run_given) {
1050                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1051                 return -E_SYNTAX;
1052         }
1053         ret = install_sighandler(SIGHUP);
1054         if (ret < 0)
1055                 return ret;
1056         compute_next_snapshot_time();
1057         return select_loop();
1058 }
1059
1060 int com_prune(void)
1061 {
1062         int ret;
1063         struct snapshot_list sl;
1064         struct disk_space ds;
1065
1066         ret = get_disk_space(".", &ds);
1067         if (ret < 0)
1068                 return ret;
1069         log_disk_space(&ds);
1070         for (;;) {
1071                 get_snapshot_list(&sl);
1072                 ret = remove_outdated_snapshot(&sl);
1073                 free_snapshot_list(&sl);
1074                 if (ret < 0)
1075                         return ret;
1076                 if (!ret)
1077                         break;
1078                 ret = wait_for_rm_process();
1079                 if (ret < 0)
1080                         goto out;
1081         }
1082         for (;;) {
1083                 get_snapshot_list(&sl);
1084                 ret = remove_redundant_snapshot(&sl);
1085                 free_snapshot_list(&sl);
1086                 if (ret < 0)
1087                         return ret;
1088                 if (!ret)
1089                         break;
1090                 ret = wait_for_rm_process();
1091                 if (ret < 0)
1092                         goto out;
1093         }
1094         return 1;
1095 out:
1096         return ret;
1097 }
1098
1099 int com_create(void)
1100 {
1101         int ret, status;
1102         char **rsync_argv;
1103
1104         if (conf.dry_run_given) {
1105                 int i;
1106                 char *msg = NULL;
1107                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1108                 for (i = 0; rsync_argv[i]; i++) {
1109                         char *tmp = msg;
1110                         msg = make_message("%s%s%s", tmp? tmp : "",
1111                                 tmp? " " : "", rsync_argv[i]);
1112                         free(tmp);
1113                 }
1114                 free_rsync_argv(rsync_argv);
1115                 dss_msg("%s\n", msg);
1116                 free(msg);
1117                 return 1;
1118         }
1119         ret = pre_create_hook();
1120         if (ret < 0)
1121                 return ret;
1122         if (pre_create_hook_pid) {
1123                 ret = wait_for_process(pre_create_hook_pid, &status);
1124                 if (ret < 0)
1125                         return ret;
1126                 ret = handle_pre_create_hook_exit(status);
1127                 if (ret < 0)
1128                         return ret;
1129         }
1130         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1131         ret = create_snapshot(rsync_argv);
1132         if (ret < 0)
1133                 goto out;
1134         ret = wait_for_process(rsync_pid, &status);
1135         if (ret < 0)
1136                 goto out;
1137         ret = handle_rsync_exit(status);
1138         if (ret < 0)
1139                 goto out;
1140         post_create_hook();
1141         if (post_create_hook_pid)
1142                 ret = wait_for_process(post_create_hook_pid, &status);
1143 out:
1144         free_rsync_argv(rsync_argv);
1145         return ret;
1146 }
1147
1148 int com_ls(void)
1149 {
1150         int i;
1151         struct snapshot_list sl;
1152         struct snapshot *s;
1153         get_snapshot_list(&sl);
1154         FOR_EACH_SNAPSHOT(s, i, &sl)
1155                 dss_msg("%u\t%s\n", s->interval, s->name);
1156         free_snapshot_list(&sl);
1157         return 1;
1158 }
1159
1160 static void setup_signal_handling(void)
1161 {
1162         int ret;
1163
1164         DSS_INFO_LOG("setting up signal handlers\n");
1165         signal_pipe = signal_init(); /* always successful */
1166         ret = install_sighandler(SIGINT);
1167         if (ret < 0)
1168                 goto err;
1169         ret = install_sighandler(SIGTERM);
1170         if (ret < 0)
1171                 goto err;
1172         ret = install_sighandler(SIGCHLD);
1173         if (ret < 0)
1174                 goto err;
1175         return;
1176 err:
1177         DSS_EMERG_LOG("could not install signal handlers\n");
1178         exit(EXIT_FAILURE);
1179 }
1180
1181 int main(int argc, char **argv)
1182 {
1183         int ret;
1184         struct cmdline_parser_params params = {
1185                 .override = 0,
1186                 .initialize = 1,
1187                 .check_required = 0,
1188                 .check_ambiguity = 0
1189         };
1190
1191         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1192         parse_config_file(0);
1193
1194         if (conf.daemon_given)
1195                 daemon_init();
1196         setup_signal_handling();
1197         ret = call_command_handler();
1198         if (ret < 0)
1199                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1200         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1201 }