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