dss.c

   1 /*
   2  * Copyright (C) 2008-2009 Andre Noll <maan@systemlinux.org>
   3  *
   4  * Licensed under the GPL v2. For licencing details see COPYING.
   5  */
   6 #include <string.h>
   7 #include <stdlib.h>
   8 #include <stdarg.h>
   9 #include <assert.h>
  10 #include <errno.h>
  11 #include <sys/types.h>
  12 #include <signal.h>
  13 #include <ctype.h>
  14 #include <sys/stat.h>
  15 #include <unistd.h>
  16 #include <inttypes.h>
  17 #include <sys/time.h>
  18 #include <time.h>
  19 #include <sys/wait.h>
  20 #include <fnmatch.h>
  21 #include <limits.h>
  22
  23
  24 #include "gcc-compat.h"
  25 #include "cmdline.h"
  26 #include "log.h"
  27 #include "string.h"
  28 #include "error.h"
  29 #include "fd.h"
  30 #include "exec.h"
  31 #include "daemon.h"
  32 #include "signal.h"
  33 #include "df.h"
  34 #include "time.h"
  35 #include "snap.h"
  36
  37 /** Command line and config file options. */
  38 static struct gengetopt_args_info conf;
  39 /** Non-NULL if we log to a file. */
  40 static FILE *logfile;
  41 /** The read end of the signal pipe */
  42 static int signal_pipe;
  43 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
  44 static pid_t create_pid;
  45 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
  46 static int create_process_stopped;
  47 /** Process id of current pre-remove/rm/post-remove process. */
  48 static pid_t remove_pid;
  49 /** When the next snapshot is due. */
  50 static int64_t next_snapshot_time;
  51 /** When to try to remove something. */
  52 static struct timeval next_removal_check;
  53 /** Creation time of the snapshot currently being created. */
  54 static int64_t current_snapshot_creation_time;
  55 /** The snapshot currently being removed. */
  56 struct snapshot *snapshot_currently_being_removed;
  57 /** Needed by the post-create hook. */
  58 static char *path_to_last_complete_snapshot;
  59 static char *name_of_reference_snapshot;
  60 /** \sa \ref snap.h for details. */
  61 enum hook_status snapshot_creation_status;
  62 /** \sa \ref snap.h for details. */
  63 enum hook_status snapshot_removal_status;
  64
  65
  66 DEFINE_DSS_ERRLIST;
  67
  68
  69 /* a litte cpp magic helps to DRY */
  70 #define COMMANDS \
  71         COMMAND(ls) \
  72         COMMAND(create) \
  73         COMMAND(prune) \
  74         COMMAND(run)
  75 #define COMMAND(x) static int com_ ##x(void);
  76 COMMANDS
  77 #undef COMMAND
  78 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
  79 static int call_command_handler(void)
  80 {
  81         COMMANDS
  82         DSS_EMERG_LOG("BUG: did not find command handler\n");
  83         return -E_BUG;
  84 }
  85 #undef COMMAND
  86 #undef COMMANDS
  87
  88 /**
  89  * The log function of dss.
  90  *
  91  * \param ll Loglevel.
  92  * \param fml Usual format string.
  93  *
  94  * All DSS_XXX_LOG() macros use this function.
  95  */
  96 __printf_2_3 void dss_log(int ll, const char* fmt,...)
  97 {
  98         va_list argp;
  99         FILE *outfd;
 100         struct tm *tm;
 101         time_t t1;
 102         char str[255] = "";
 103
 104         if (ll < conf.loglevel_arg)
 105                 return;
 106         outfd = logfile? logfile : stderr;
 107         time(&t1);
 108         tm = localtime(&t1);
 109         strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 110         fprintf(outfd, "%s ", str);
 111         if (conf.loglevel_arg <= INFO)
 112                 fprintf(outfd, "%i: ", ll);
 113         va_start(argp, fmt);
 114         vfprintf(outfd, fmt, argp);
 115         va_end(argp);
 116 }
 117
 118 /**
 119  * Print a message either to stdout or to the log file.
 120  */
 121 static __printf_1_2 void dss_msg(const char* fmt,...)
 122 {
 123         FILE *outfd = conf.daemon_given? logfile : stdout;
 124         va_list argp;
 125         va_start(argp, fmt);
 126         vfprintf(outfd, fmt, argp);
 127         va_end(argp);
 128 }
 129
 130 static int disk_space_low(void)
 131 {
 132         struct disk_space ds;
 133         int ret = get_disk_space(".", &ds);
 134
 135         if (ret < 0)
 136                 return ret;
 137         if (conf.min_free_mb_arg)
 138                 if (ds.free_mb < conf.min_free_mb_arg)
 139                         return 1;
 140         if (conf.min_free_percent_arg)
 141                 if (ds.percent_free < conf.min_free_percent_arg)
 142                         return 1;
 143         if (conf.min_free_percent_inodes_arg)
 144                 if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg)
 145                         return 1;
 146         return 0;
 147 }
 148
 149 static void dss_get_snapshot_list(struct snapshot_list *sl)
 150 {
 151         get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 152 }
 153
 154 static int64_t compute_next_snapshot_time(void)
 155 {
 156         int64_t x = 0, now = get_current_time(), unit_interval
 157                 = 24 * 3600 * conf.unit_interval_arg, ret;
 158         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 159                 num_complete_snapshots = 0;
 160         int i;
 161         struct snapshot *s = NULL;
 162         struct snapshot_list sl;
 163
 164         current_snapshot_creation_time = 0;
 165         dss_get_snapshot_list(&sl);
 166         FOR_EACH_SNAPSHOT(s, i, &sl) {
 167                 if (!(s->flags & SS_COMPLETE))
 168                         continue;
 169                 num_complete_snapshots++;
 170                 x += s->completion_time - s->creation_time;
 171         }
 172         assert(x >= 0);
 173
 174         ret = now;
 175         if (num_complete_snapshots == 0)
 176                 goto out;
 177         x /= num_complete_snapshots; /* avg time to create one snapshot */
 178         if (unit_interval < x * wanted) /* oops, no sleep at all */
 179                 goto out;
 180         ret = s->completion_time + unit_interval / wanted - x;
 181 out:
 182         free_snapshot_list(&sl);
 183         return ret;
 184 }
 185
 186 static inline void invalidate_next_snapshot_time(void)
 187 {
 188         next_snapshot_time = 0;
 189 }
 190
 191 static inline int next_snapshot_time_is_valid(void)
 192 {
 193         return next_snapshot_time != 0;
 194 }
 195
 196 static int next_snapshot_is_due(void)
 197 {
 198         int64_t now = get_current_time();
 199
 200         if (!next_snapshot_time_is_valid())
 201                 next_snapshot_time = compute_next_snapshot_time();
 202         if (next_snapshot_time <= now) {
 203                 DSS_DEBUG_LOG("next snapshot: now\n");
 204                 return 1;
 205         }
 206         DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
 207                 next_snapshot_time - now);
 208         return 0;
 209 }
 210
 211 static int pre_create_hook(void)
 212 {
 213         int ret, fds[3] = {0, 0, 0};
 214
 215         assert(snapshot_creation_status == HS_READY);
 216         /* make sure that the next snapshot time will be recomputed */
 217         invalidate_next_snapshot_time();
 218         if (!conf.pre_create_hook_given) {
 219                 snapshot_creation_status = HS_PRE_SUCCESS;
 220                 return 0;
 221         }
 222         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 223         ret = dss_exec_cmdline_pid(&create_pid,
 224                 conf.pre_create_hook_arg, fds);
 225         if (ret < 0)
 226                 return ret;
 227         snapshot_creation_status = HS_PRE_RUNNING;
 228         return ret;
 229 }
 230
 231 static int pre_remove_hook(struct snapshot *s, const char *why)
 232 {
 233         int ret, fds[3] = {0, 0, 0};
 234         char *cmd;
 235
 236         if (!s)
 237                 return 0;
 238         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 239         assert(snapshot_removal_status == HS_READY);
 240         assert(remove_pid == 0);
 241         assert(!snapshot_currently_being_removed);
 242
 243         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 244         *snapshot_currently_being_removed = *s;
 245         snapshot_currently_being_removed->name = dss_strdup(s->name);
 246
 247         if (!conf.pre_remove_hook_given) {
 248                 snapshot_removal_status = HS_PRE_SUCCESS;
 249                 return 0;
 250         }
 251         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 252                 conf.dest_dir_arg, s->name);
 253         DSS_DEBUG_LOG("executing %s\n", cmd);
 254         ret = dss_exec_cmdline_pid(&remove_pid,
 255                 conf.pre_remove_hook_arg, fds);
 256         free(cmd);
 257         if (ret < 0)
 258                 return ret;
 259         snapshot_removal_status = HS_PRE_RUNNING;
 260         return ret;
 261 }
 262
 263 static int exec_rm(void)
 264 {
 265         struct snapshot *s = snapshot_currently_being_removed;
 266         int fds[3] = {0, 0, 0};
 267         char *new_name = being_deleted_name(s);
 268         char *argv[] = {"rm", "-rf", new_name, NULL};
 269         int ret;
 270
 271         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 272         assert(remove_pid == 0);
 273
 274         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 275         ret = dss_rename(s->name, new_name);
 276         if (ret < 0)
 277                 goto out;
 278         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 279         if (ret < 0)
 280                 goto out;
 281         snapshot_removal_status = HS_RUNNING;
 282 out:
 283         free(new_name);
 284         return ret;
 285 }
 286
 287 static int snapshot_is_being_created(struct snapshot *s)
 288 {
 289         return s->creation_time == current_snapshot_creation_time;
 290 }
 291
 292 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 293 {
 294         struct snapshot *s;
 295         int i;
 296
 297         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 298         FOR_EACH_SNAPSHOT(s, i, sl) {
 299                 if (snapshot_is_being_created(s))
 300                         continue;
 301                 /*
 302                  * We know that no rm is currently running, so if s is marked
 303                  * as being deleted, a previously started rm must have failed.
 304                  */
 305                 if (s->flags & SS_BEING_DELETED)
 306                         return s;
 307
 308                 if (s->flags & SS_COMPLETE) /* good snapshot */
 309                         continue;
 310                 /*
 311                  * This snapshot is incomplete and it is not the snapshot
 312                  * currently being created. However, we must not remove it if
 313                  * rsync is about to be restarted. As only the newest snapshot
 314                  * can be restarted, this snapshot is orphaned if it is not the
 315                  * newest snapshot or if we are not about to restart rsync.
 316                  */
 317                 if (get_newest_snapshot(sl) != s)
 318                         return s;
 319                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 320                         return s;
 321         }
 322         /* no orphaned snapshots */
 323         return NULL;
 324 }
 325
 326 static int is_reference_snapshot(struct snapshot *s)
 327 {
 328         if (!name_of_reference_snapshot)
 329                 return 0;
 330         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 331 }
 332
 333 /*
 334  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 335  */
 336 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 337 {
 338         int i, interval;
 339         struct snapshot *s;
 340         unsigned missing = 0;
 341
 342         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 343         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 344                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 345                 unsigned num = sl->interval_count[interval];
 346                 struct snapshot *victim = NULL, *prev = NULL;
 347                 int64_t score = LONG_MAX;
 348
 349                 if (keep >= num)
 350                         missing += keep - num;
 351 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 352 //                      interval, keep, num, missing);
 353                 if (keep + missing >= num)
 354                         continue;
 355                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 356                 FOR_EACH_SNAPSHOT(s, i, sl) {
 357                         int64_t this_score;
 358
 359                         if (snapshot_is_being_created(s))
 360                                 continue;
 361                         if (is_reference_snapshot(s))
 362                                 continue;
 363                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 364                         if (s->interval > interval) {
 365                                 prev = s;
 366                                 continue;
 367                         }
 368                         if (s->interval < interval)
 369                                 break;
 370                         if (!victim) {
 371                                 victim = s;
 372                                 prev = s;
 373                                 continue;
 374                         }
 375                         assert(prev);
 376                         /* check if s is a better victim */
 377                         this_score = s->creation_time - prev->creation_time;
 378                         assert(this_score >= 0);
 379                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 380                         if (this_score < score) {
 381                                 score = this_score;
 382                                 victim = s;
 383                         }
 384                         prev = s;
 385                 }
 386                 assert(victim);
 387                 return victim;
 388         }
 389         return NULL;
 390 }
 391
 392 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 393 {
 394         int i;
 395         struct snapshot *s;
 396
 397         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 398                 conf.num_intervals_arg);
 399         FOR_EACH_SNAPSHOT(s, i, sl) {
 400                 if (snapshot_is_being_created(s))
 401                         continue;
 402                 if (is_reference_snapshot(s))
 403                         continue;
 404                 if (s->interval < conf.num_intervals_arg)
 405                         continue;
 406                 return s;
 407         }
 408         return NULL;
 409 }
 410
 411 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 412 {
 413         int i;
 414         struct snapshot *s;
 415         FOR_EACH_SNAPSHOT(s, i, sl) {
 416                 if (snapshot_is_being_created(s))
 417                         continue;
 418                 if (is_reference_snapshot(s))
 419                         continue;
 420                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 421                 return s;
 422         }
 423         return NULL;
 424 }
 425
 426 static int rename_incomplete_snapshot(int64_t start)
 427 {
 428         char *old_name;
 429         int ret;
 430
 431         free(path_to_last_complete_snapshot);
 432         ret = complete_name(start, get_current_time(),
 433                 &path_to_last_complete_snapshot);
 434         if (ret < 0)
 435                 return ret;
 436         old_name = incomplete_name(start);
 437         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 438         if (ret >= 0)
 439                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 440                         path_to_last_complete_snapshot);
 441         free(old_name);
 442         return ret;
 443 }
 444
 445 static int try_to_free_disk_space(void)
 446 {
 447         int ret;
 448         struct snapshot_list sl;
 449         struct snapshot *victim;
 450         struct timeval now;
 451         const char *why;
 452         int low_disk_space;
 453
 454         ret = disk_space_low();
 455         if (ret < 0)
 456                 return ret;
 457         low_disk_space = ret;
 458         gettimeofday(&now, NULL);
 459         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 460                 return 0;
 461         if (!low_disk_space) {
 462                 if (conf.keep_redundant_given)
 463                         return 0;
 464                 if (snapshot_creation_status != HS_READY)
 465                         return 0;
 466                 if (next_snapshot_is_due())
 467                         return 0;
 468         }
 469         dss_get_snapshot_list(&sl);
 470         ret = 0;
 471         if (!low_disk_space && sl.num_snapshots <= 1)
 472                 goto out;
 473         why = "outdated";
 474         victim = find_outdated_snapshot(&sl);
 475         if (victim)
 476                 goto remove;
 477         why = "redundant";
 478         victim = find_redundant_snapshot(&sl);
 479         if (victim)
 480                 goto remove;
 481         /* try harder only if disk space is low */
 482         if (!low_disk_space)
 483                 goto out;
 484         why = "orphaned";
 485         victim = find_orphaned_snapshot(&sl);
 486         if (victim)
 487                 goto remove;
 488         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 489         victim = find_oldest_removable_snapshot(&sl);
 490         if (victim)
 491                 goto remove;
 492         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 493         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 494         goto out;
 495 remove:
 496         ret = pre_remove_hook(victim, why);
 497 out:
 498         free_snapshot_list(&sl);
 499         return ret;
 500 }
 501
 502 static int post_create_hook(void)
 503 {
 504         int ret, fds[3] = {0, 0, 0};
 505         char *cmd;
 506
 507         if (!conf.post_create_hook_given) {
 508                 create_pid = 0;
 509                 snapshot_creation_status = HS_READY;
 510                 return 0;
 511         }
 512         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 513                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 514         DSS_NOTICE_LOG("executing %s\n", cmd);
 515         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 516         free(cmd);
 517         if (ret < 0)
 518                 return ret;
 519         snapshot_creation_status = HS_POST_RUNNING;
 520         return ret;
 521 }
 522
 523 static int post_remove_hook(void)
 524 {
 525         int ret, fds[3] = {0, 0, 0};
 526         char *cmd;
 527         struct snapshot *s = snapshot_currently_being_removed;
 528
 529         assert(s);
 530
 531         if (!conf.post_remove_hook_given) {
 532                 snapshot_removal_status = HS_READY;
 533                 return 0;
 534         }
 535         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 536                 conf.dest_dir_arg, s->name);
 537         DSS_NOTICE_LOG("executing %s\n", cmd);
 538         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 539         free(cmd);
 540         if (ret < 0)
 541                 return ret;
 542         snapshot_removal_status = HS_POST_RUNNING;
 543         return ret;
 544 }
 545
 546 static void kill_process(pid_t pid)
 547 {
 548         if (!pid)
 549                 return;
 550         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 551         kill(pid, SIGTERM);
 552 }
 553
 554 static void stop_create_process(void)
 555 {
 556         if (!create_pid || create_process_stopped)
 557                 return;
 558         kill(SIGSTOP, create_pid);
 559         create_process_stopped = 1;
 560 }
 561
 562 static void restart_create_process(void)
 563 {
 564         if (!create_pid || !create_process_stopped)
 565                 return;
 566         kill (SIGCONT, create_pid);
 567         create_process_stopped = 0;
 568 }
 569
 570 /**
 571  * Print a log message about the exit status of a child.
 572  */
 573 static void log_termination_msg(pid_t pid, int status)
 574 {
 575         if (WIFEXITED(status))
 576                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 577                         WEXITSTATUS(status));
 578         else if (WIFSIGNALED(status))
 579                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 580                         WTERMSIG(status));
 581         else
 582                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 583 }
 584
 585 static int wait_for_process(pid_t pid, int *status)
 586 {
 587         int ret;
 588
 589         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 590         for (;;) {
 591                 fd_set rfds;
 592
 593                 FD_ZERO(&rfds);
 594                 FD_SET(signal_pipe, &rfds);
 595                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 596                 if (ret < 0)
 597                         break;
 598                 ret = next_signal();
 599                 if (!ret)
 600                         continue;
 601                 if (ret == SIGCHLD) {
 602                         ret = waitpid(pid, status, 0);
 603                         if (ret >= 0)
 604                                 break;
 605                         if (errno != EINTR) { /* error */
 606                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 607                                 break;
 608                         }
 609                 }
 610                 /* SIGINT or SIGTERM */
 611                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 612                 kill(pid, SIGTERM);
 613         }
 614         if (ret < 0)
 615                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 616         else
 617                 log_termination_msg(pid, *status);
 618         return ret;
 619 }
 620
 621 static void handle_pre_remove_exit(int status)
 622 {
 623         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 624                 snapshot_removal_status = HS_READY;
 625                 gettimeofday(&next_removal_check, NULL);
 626                 next_removal_check.tv_sec += 60;
 627                 return;
 628         }
 629         snapshot_removal_status = HS_PRE_SUCCESS;
 630 }
 631
 632 static int handle_rm_exit(int status)
 633 {
 634         if (!WIFEXITED(status)) {
 635                 snapshot_removal_status = HS_READY;
 636                 return -E_INVOLUNTARY_EXIT;
 637         }
 638         if (WEXITSTATUS(status)) {
 639                 snapshot_removal_status = HS_READY;
 640                 return -E_BAD_EXIT_CODE;
 641         }
 642         if (conf.post_remove_hook_given)
 643                 snapshot_removal_status = HS_SUCCESS;
 644         else
 645                 snapshot_removal_status = HS_READY;
 646         return 1;
 647 }
 648
 649 static void handle_post_remove_exit(void)
 650 {
 651         snapshot_removal_status = HS_READY;
 652 }
 653
 654 static int handle_remove_exit(int status)
 655 {
 656         int ret;
 657         struct snapshot *s = snapshot_currently_being_removed;
 658
 659         assert(s);
 660         switch (snapshot_removal_status) {
 661         case HS_PRE_RUNNING:
 662                 handle_pre_remove_exit(status);
 663                 ret = 1;
 664                 break;
 665         case HS_RUNNING:
 666                 ret = handle_rm_exit(status);
 667                 break;
 668         case HS_POST_RUNNING:
 669                 handle_post_remove_exit();
 670                 ret = 1;
 671                 break;
 672         default:
 673                 ret = -E_BUG;
 674         }
 675         if (snapshot_removal_status == HS_READY) {
 676                 free(s->name);
 677                 free(s);
 678                 snapshot_currently_being_removed = NULL;
 679         }
 680         remove_pid = 0;
 681         return ret;
 682 }
 683
 684 static int wait_for_remove_process(void)
 685 {
 686         int status, ret;
 687
 688         assert(remove_pid);
 689         assert(
 690                 snapshot_removal_status == HS_PRE_RUNNING ||
 691                 snapshot_removal_status == HS_RUNNING ||
 692                 snapshot_removal_status == HS_POST_RUNNING
 693         );
 694         ret = wait_for_process(remove_pid, &status);
 695         if (ret < 0)
 696                 return ret;
 697         return handle_remove_exit(status);
 698 }
 699
 700 static 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)create_pid);
 706                 ret = -E_INVOLUNTARY_EXIT;
 707                 snapshot_creation_status = HS_READY;
 708                 goto out;
 709         }
 710         es = WEXITSTATUS(status);
 711         /*
 712          * Restart rsync on non-fatal errors:
 713          * 12: Error in rsync protocol data stream
 714          * 13: Errors with program diagnostics
 715          */
 716         if (es == 12 || es == 13) {
 717                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 718                         (int)create_pid, es);
 719                 snapshot_creation_status = HS_NEEDS_RESTART;
 720                 next_snapshot_time = get_current_time() + 60;
 721                 ret = 1;
 722                 goto out;
 723         }
 724         if (es != 0 && es != 23 && es != 24) {
 725                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 726                 ret = -E_BAD_EXIT_CODE;
 727                 snapshot_creation_status = HS_READY;
 728                 goto out;
 729         }
 730         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 731         if (ret < 0)
 732                 goto out;
 733         snapshot_creation_status = HS_SUCCESS;
 734         free(name_of_reference_snapshot);
 735         name_of_reference_snapshot = NULL;
 736 out:
 737         create_process_stopped = 0;
 738         return ret;
 739 }
 740
 741 static int handle_pre_create_hook_exit(int status)
 742 {
 743         int es, ret;
 744         static int warn_count;
 745
 746         if (!WIFEXITED(status)) {
 747                 snapshot_creation_status = HS_READY;
 748                 ret = -E_INVOLUNTARY_EXIT;
 749                 goto out;
 750         }
 751         es = WEXITSTATUS(status);
 752         if (es) {
 753                 if (!warn_count--) {
 754                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 755                                 conf.pre_create_hook_arg, es);
 756                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 757                         warn_count = 60; /* warn only once per hour */
 758                 }
 759                 next_snapshot_time = get_current_time() + 60;
 760                 snapshot_creation_status = HS_READY;
 761                 ret = 0;
 762                 goto out;
 763         }
 764         warn_count = 0;
 765         snapshot_creation_status = HS_PRE_SUCCESS;
 766         ret = 1;
 767 out:
 768         return ret;
 769 }
 770
 771 static int handle_sigchld(void)
 772 {
 773         pid_t pid;
 774         int status, ret = reap_child(&pid, &status);
 775
 776         if (ret <= 0)
 777                 return ret;
 778
 779         if (pid == create_pid) {
 780                 switch (snapshot_creation_status) {
 781                 case HS_PRE_RUNNING:
 782                         return handle_pre_create_hook_exit(status);
 783                 case HS_RUNNING:
 784                         return handle_rsync_exit(status);
 785                 case HS_POST_RUNNING:
 786                         snapshot_creation_status = HS_READY;
 787                         return 1;
 788                 default:
 789                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 790                                 snapshot_creation_status);
 791                         return -E_BUG;
 792                 }
 793                 create_pid = 0;
 794         }
 795         if (pid == remove_pid) {
 796                 ret = handle_remove_exit(status);
 797                 if (ret < 0)
 798                         return ret;
 799                 return ret;
 800         }
 801         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 802         return -E_BUG;
 803 }
 804
 805 static int check_config(void)
 806 {
 807         if (conf.unit_interval_arg <= 0) {
 808                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 809                 return -E_INVALID_NUMBER;
 810         }
 811         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 812         if (conf.num_intervals_arg <= 0) {
 813                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 814                 return -E_INVALID_NUMBER;
 815         }
 816         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 817         return 1;
 818 }
 819
 820 /*
 821  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 822  * file was read successfully.
 823  */
 824 static int parse_config_file(int override)
 825 {
 826         int ret, config_file_exists;
 827         char *config_file;
 828         struct stat statbuf;
 829         char *old_logfile_arg = NULL;
 830         int old_daemon_given = 0;
 831
 832         if (conf.config_file_given)
 833                 config_file = dss_strdup(conf.config_file_arg);
 834         else {
 835                 char *home = get_homedir();
 836                 config_file = make_message("%s/.dssrc", home);
 837                 free(home);
 838         }
 839         if (override) { /* SIGHUP */
 840                 if (conf.logfile_given)
 841                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 842                 old_daemon_given = conf.daemon_given;
 843         }
 844
 845         config_file_exists = !stat(config_file, &statbuf);
 846         if (!config_file_exists && conf.config_file_given) {
 847                 ret = -ERRNO_TO_DSS_ERROR(errno);
 848                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 849                 goto out;
 850         }
 851         if (config_file_exists) {
 852                 struct cmdline_parser_params params = {
 853                         .override = override,
 854                         .initialize = 0,
 855                         .check_required = 1,
 856                         .check_ambiguity = 0,
 857                         .print_errors = 1
 858                 };
 859                 if (override) { /* invalidate all rsync options */
 860                         int i;
 861
 862                         for (i = 0; i < conf.rsync_option_given; i++) {
 863                                 free(conf.rsync_option_arg[i]);
 864                                 conf.rsync_option_arg[i] = NULL;
 865                         }
 866                         conf.rsync_option_given = 0;
 867                 }
 868                 cmdline_parser_config_file(config_file, &conf, &params);
 869         }
 870         ret = check_config();
 871         if (ret < 0)
 872                 goto out;
 873         if (override) {
 874                 /* don't change daemon mode on SIGHUP */
 875                 conf.daemon_given = old_daemon_given;
 876                 close_log(logfile);
 877                 logfile = NULL;
 878                 if (conf.logfile_given)
 879                         free(old_logfile_arg);
 880                 else if (conf.daemon_given) { /* re-use old logfile */
 881                         conf.logfile_arg = old_logfile_arg;
 882                         conf.logfile_given = 1;
 883                 }
 884         }
 885         if (conf.logfile_given) {
 886                 logfile = open_log(conf.logfile_arg);
 887                 log_welcome(conf.loglevel_arg);
 888         }
 889         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 890         ret = config_file_exists;
 891 out:
 892         free(config_file);
 893         if (ret < 0)
 894                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 895         return ret;
 896 }
 897
 898 static int change_to_dest_dir(void)
 899 {
 900         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 901         return dss_chdir(conf.dest_dir_arg);
 902 }
 903
 904 static void dump_dss_config(const char *msg)
 905 {
 906         if (conf.loglevel_arg > INFO)
 907                 return;
 908         DSS_INFO_LOG("%s\n", msg);
 909         cmdline_parser_dump(logfile? logfile : stderr, &conf);
 910 }
 911
 912 static int handle_sighup(void)
 913 {
 914         int ret;
 915
 916         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
 917         dump_dss_config("current config");
 918         ret = parse_config_file(1);
 919         if (ret < 0)
 920                 return ret;
 921         dump_dss_config("new config");
 922         invalidate_next_snapshot_time();
 923         return change_to_dest_dir();
 924 }
 925
 926 static int handle_signal(void)
 927 {
 928         int sig, ret = next_signal();
 929
 930         if (ret <= 0)
 931                 goto out;
 932         sig = ret;
 933         switch (sig) {
 934         case SIGINT:
 935         case SIGTERM:
 936                 restart_create_process();
 937                 kill_process(create_pid);
 938                 kill_process(remove_pid);
 939                 ret = -E_SIGNAL;
 940                 break;
 941         case SIGHUP:
 942                 ret = handle_sighup();
 943                 break;
 944         case SIGCHLD:
 945                 ret = handle_sigchld();
 946                 break;
 947         }
 948 out:
 949         if (ret < 0)
 950                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 951         return ret;
 952 }
 953
 954 /*
 955  * We can not use rsync locally if the local user is different from the remote
 956  * user or if the src dir is not on the local host (or both).
 957  */
 958 static int use_rsync_locally(char *logname)
 959 {
 960         char *h = conf.remote_host_arg;
 961
 962         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 963                 return 0;
 964         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 965                 return 0;
 966         return 1;
 967 }
 968
 969 static int rename_resume_snap(int64_t creation_time)
 970 {
 971         struct snapshot_list sl = {.num_snapshots = 0};
 972         struct snapshot *s;
 973         char *new_name = incomplete_name(creation_time);
 974         int ret;
 975
 976         ret = 0;
 977         if (conf.no_resume_given)
 978                 goto out;
 979         dss_get_snapshot_list(&sl);
 980         s = get_newest_snapshot(&sl);
 981         if (!s)
 982                 goto out;
 983         if ((s->flags & SS_COMPLETE) != 0) /* complete */
 984                 goto out;
 985         DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
 986         ret = dss_rename(s->name, new_name);
 987 out:
 988         if (ret >= 0)
 989                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
 990         free(new_name);
 991         free_snapshot_list(&sl);
 992         return ret;
 993 }
 994
 995 static void create_rsync_argv(char ***argv, int64_t *num)
 996 {
 997         char *logname;
 998         int i = 0, j;
 999         struct snapshot_list sl;
1000
1001         dss_get_snapshot_list(&sl);
1002         assert(!name_of_reference_snapshot);
1003         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1004         free_snapshot_list(&sl);
1005
1006         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1007         (*argv)[i++] = dss_strdup("rsync");
1008         (*argv)[i++] = dss_strdup("-aq");
1009         (*argv)[i++] = dss_strdup("--delete");
1010         for (j = 0; j < conf.rsync_option_given; j++)
1011                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1012         if (name_of_reference_snapshot) {
1013                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1014                 (*argv)[i++] = make_message("--link-dest=../%s",
1015                         name_of_reference_snapshot);
1016         } else
1017                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1018         logname = dss_logname();
1019         if (use_rsync_locally(logname))
1020                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1021         else
1022                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1023                         conf.remote_user_arg : logname,
1024                         conf.remote_host_arg, conf.source_dir_arg);
1025         free(logname);
1026         *num = get_current_time();
1027         (*argv)[i++] = incomplete_name(*num);
1028         (*argv)[i++] = NULL;
1029         for (j = 0; j < i; j++)
1030                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1031 }
1032
1033 static void free_rsync_argv(char **argv)
1034 {
1035         int i;
1036
1037         if (!argv)
1038                 return;
1039         for (i = 0; argv[i]; i++)
1040                 free(argv[i]);
1041         free(argv);
1042 }
1043
1044 static int create_snapshot(char **argv)
1045 {
1046         int ret, fds[3] = {0, 0, 0};
1047
1048         ret = rename_resume_snap(current_snapshot_creation_time);
1049         if (ret < 0)
1050                 return ret;
1051         ret = dss_exec(&create_pid, argv[0], argv, fds);
1052         if (ret < 0)
1053                 return ret;
1054         snapshot_creation_status = HS_RUNNING;
1055         return ret;
1056 }
1057
1058 static int select_loop(void)
1059 {
1060         int ret;
1061         /* check every 60 seconds for free disk space */
1062         struct timeval tv;
1063         char **rsync_argv = NULL;
1064
1065         for (;;) {
1066                 fd_set rfds;
1067                 struct timeval *tvp;
1068
1069                 if (remove_pid)
1070                         tvp = NULL; /* sleep until rm hook/process dies */
1071                 else { /* sleep one minute */
1072                         tv.tv_sec = 60;
1073                         tv.tv_usec = 0;
1074                         tvp = &tv;
1075                 }
1076                 FD_ZERO(&rfds);
1077                 FD_SET(signal_pipe, &rfds);
1078                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1079                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1080                 if (ret < 0)
1081                         goto out;
1082                 if (FD_ISSET(signal_pipe, &rfds)) {
1083                         ret = handle_signal();
1084                         if (ret < 0)
1085                                 goto out;
1086                 }
1087                 if (remove_pid)
1088                         continue;
1089                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1090                         ret = exec_rm();
1091                         if (ret < 0)
1092                                 goto out;
1093                         continue;
1094                 }
1095                 if (snapshot_removal_status == HS_SUCCESS) {
1096                         ret = post_remove_hook();
1097                         if (ret < 0)
1098                                 goto out;
1099                         continue;
1100                 }
1101                 ret = try_to_free_disk_space();
1102                 if (ret < 0)
1103                         goto out;
1104                 if (snapshot_removal_status != HS_READY) {
1105                         stop_create_process();
1106                         continue;
1107                 }
1108                 restart_create_process();
1109                 switch (snapshot_creation_status) {
1110                 case HS_READY:
1111                         if (!next_snapshot_is_due())
1112                                 continue;
1113                         ret = pre_create_hook();
1114                         if (ret < 0)
1115                                 goto out;
1116                         continue;
1117                 case HS_PRE_RUNNING:
1118                 case HS_RUNNING:
1119                 case HS_POST_RUNNING:
1120                         continue;
1121                 case HS_PRE_SUCCESS:
1122                         free_rsync_argv(rsync_argv);
1123                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1124                         /* fall through */
1125                 case HS_NEEDS_RESTART:
1126                         if (!next_snapshot_is_due())
1127                                 continue;
1128                         ret = create_snapshot(rsync_argv);
1129                         if (ret < 0)
1130                                 goto out;
1131                         continue;
1132                 case HS_SUCCESS:
1133                         ret = post_create_hook();
1134                         if (ret < 0)
1135                                 goto out;
1136                         continue;
1137                 }
1138         }
1139 out:
1140         return ret;
1141 }
1142
1143 static void exit_hook(int exit_code)
1144 {
1145         int fds[3] = {0, 0, 0};
1146         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1147         pid_t pid;
1148
1149         if (!conf.exit_hook_given)
1150                 return;
1151         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1152         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1153 }
1154
1155 static int com_run(void)
1156 {
1157         int ret;
1158
1159         if (conf.dry_run_given) {
1160                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1161                 return -E_SYNTAX;
1162         }
1163         ret = install_sighandler(SIGHUP);
1164         if (ret < 0)
1165                 return ret;
1166         ret = select_loop();
1167         if (ret >= 0) /* impossible */
1168                 ret = -E_BUG;
1169         exit_hook(ret);
1170         return ret;
1171 }
1172
1173 static int com_prune(void)
1174 {
1175         int ret;
1176         struct snapshot_list sl;
1177         struct snapshot *victim;
1178         struct disk_space ds;
1179         const char *why;
1180
1181         ret = get_disk_space(".", &ds);
1182         if (ret < 0)
1183                 return ret;
1184         log_disk_space(&ds);
1185         dss_get_snapshot_list(&sl);
1186         why = "outdated";
1187         victim = find_outdated_snapshot(&sl);
1188         if (victim)
1189                 goto rm;
1190         why = "redundant";
1191         victim = find_redundant_snapshot(&sl);
1192         if (victim)
1193                 goto rm;
1194         ret = 0;
1195         goto out;
1196 rm:
1197         if (conf.dry_run_given) {
1198                 dss_msg("%s snapshot %s (interval = %i)\n",
1199                         why, victim->name, victim->interval);
1200                 ret = 0;
1201                 goto out;
1202         }
1203         ret = pre_remove_hook(victim, why);
1204         if (ret < 0)
1205                 goto out;
1206         if (snapshot_removal_status == HS_PRE_RUNNING) {
1207                 ret = wait_for_remove_process();
1208                 if (ret < 0)
1209                         goto out;
1210                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1211                         goto out;
1212         }
1213         ret = exec_rm();
1214         if (ret < 0)
1215                 goto out;
1216         ret = wait_for_remove_process();
1217         if (ret < 0)
1218                 goto out;
1219         if (snapshot_removal_status != HS_SUCCESS)
1220                 goto out;
1221         ret = post_remove_hook();
1222         if (ret < 0)
1223                 goto out;
1224         if (snapshot_removal_status != HS_POST_RUNNING)
1225                 goto out;
1226         ret = wait_for_remove_process();
1227         if (ret < 0)
1228                 goto out;
1229         ret = 1;
1230 out:
1231         free_snapshot_list(&sl);
1232         return ret;
1233 }
1234
1235 static int com_create(void)
1236 {
1237         int ret, status;
1238         char **rsync_argv;
1239
1240         if (conf.dry_run_given) {
1241                 int i;
1242                 char *msg = NULL;
1243                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1244                 for (i = 0; rsync_argv[i]; i++) {
1245                         char *tmp = msg;
1246                         msg = make_message("%s%s%s", tmp? tmp : "",
1247                                 tmp? " " : "", rsync_argv[i]);
1248                         free(tmp);
1249                 }
1250                 free_rsync_argv(rsync_argv);
1251                 dss_msg("%s\n", msg);
1252                 free(msg);
1253                 return 1;
1254         }
1255         ret = pre_create_hook();
1256         if (ret < 0)
1257                 return ret;
1258         if (create_pid) {
1259                 ret = wait_for_process(create_pid, &status);
1260                 if (ret < 0)
1261                         return ret;
1262                 ret = handle_pre_create_hook_exit(status);
1263                 if (ret <= 0) /* error, or pre-create failed */
1264                         return ret;
1265         }
1266         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1267         ret = create_snapshot(rsync_argv);
1268         if (ret < 0)
1269                 goto out;
1270         ret = wait_for_process(create_pid, &status);
1271         if (ret < 0)
1272                 goto out;
1273         ret = handle_rsync_exit(status);
1274         if (ret < 0)
1275                 goto out;
1276         post_create_hook();
1277         if (create_pid)
1278                 ret = wait_for_process(create_pid, &status);
1279 out:
1280         free_rsync_argv(rsync_argv);
1281         return ret;
1282 }
1283
1284 static int com_ls(void)
1285 {
1286         int i;
1287         struct snapshot_list sl;
1288         struct snapshot *s;
1289
1290         dss_get_snapshot_list(&sl);
1291         FOR_EACH_SNAPSHOT(s, i, &sl) {
1292                 int64_t d = 0;
1293                 if (s->flags & SS_COMPLETE)
1294                         d = (s->completion_time - s->creation_time) / 60;
1295                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1296         };
1297         free_snapshot_list(&sl);
1298         return 1;
1299 }
1300
1301 static int setup_signal_handling(void)
1302 {
1303         int ret;
1304
1305         DSS_INFO_LOG("setting up signal handlers\n");
1306         signal_pipe = signal_init(); /* always successful */
1307         ret = install_sighandler(SIGINT);
1308         if (ret < 0)
1309                 return ret;
1310         ret = install_sighandler(SIGTERM);
1311         if (ret < 0)
1312                 return ret;
1313         return install_sighandler(SIGCHLD);
1314 }
1315
1316 /**
1317  * The main function of dss.
1318  *
1319  * \param argc Usual argument count.
1320  * \param argv Usual argument vector.
1321  */
1322 int main(int argc, char **argv)
1323 {
1324         int ret;
1325         struct cmdline_parser_params params = {
1326                 .override = 0,
1327                 .initialize = 1,
1328                 .check_required = 0,
1329                 .check_ambiguity = 0,
1330                 .print_errors = 1
1331         };
1332
1333         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1334         ret = parse_config_file(0);
1335         if (ret < 0)
1336                 goto out;
1337         if (ret == 0) { /* no config file given */
1338                 /*
1339                  * Parse the command line options again, but this time check
1340                  * that all required options are given.
1341                  */
1342                 params = (struct cmdline_parser_params) {
1343                         .override = 1,
1344                         .initialize = 1,
1345                         .check_required = 1,
1346                         .check_ambiguity = 1,
1347                         .print_errors = 1
1348                 };
1349                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1350         }
1351         if (conf.daemon_given)
1352                 daemon_init();
1353         dump_dss_config("dss configuration");
1354         ret = change_to_dest_dir();
1355         if (ret < 0)
1356                 goto out;
1357         ret = setup_signal_handling();
1358         if (ret < 0)
1359                 goto out;
1360         ret = call_command_handler();
1361 out:
1362         if (ret < 0)
1363                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1364         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1365 }