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 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 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 int next_snapshot_is_due(void)
 155 {
 156         struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 157                 tmp, diff;
 158         int64_t x = 0;
 159         unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
 160                 num_complete_snapshots = 0;
 161         int i, ret;
 162         struct snapshot *s = NULL;
 163         struct snapshot_list sl;
 164
 165         gettimeofday(&now, NULL);
 166         if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 167                 return 0;
 168         assert(snapshot_creation_status == HS_READY);
 169         current_snapshot_creation_time = 0;
 170         dss_get_snapshot_list(&sl);
 171         FOR_EACH_SNAPSHOT(s, i, &sl) {
 172                 if (!(s->flags & SS_COMPLETE))
 173                         continue;
 174                 num_complete_snapshots++;
 175                 x += s->completion_time - s->creation_time;
 176         }
 177         assert(x >= 0);
 178         if (num_complete_snapshots)
 179                 x /= num_complete_snapshots; /* avg time to create one snapshot */
 180         x *= wanted; /* time to create all snapshots in interval 0 */
 181         tmp.tv_sec = x;
 182         tmp.tv_usec = 0;
 183         ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
 184         if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
 185                 ret = 1;
 186                 goto out;
 187         }
 188
 189         tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */
 190         diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
 191         diff.tv_usec = 0;
 192         tv_add(&diff, &tmp, &next_snapshot_time);
 193         ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
 194 out:
 195         free_snapshot_list(&sl);
 196         if (ret == 0)
 197                 DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
 198                         tv2ms(&diff) / 1000);
 199         else
 200                 DSS_DEBUG_LOG("next snapshot: now\n");
 201         return ret;
 202 }
 203
 204 static int pre_create_hook(void)
 205 {
 206         int ret, fds[3] = {0, 0, 0};
 207
 208         assert(snapshot_creation_status == HS_READY);
 209         if (!conf.pre_create_hook_given) {
 210                 snapshot_creation_status = HS_PRE_SUCCESS;
 211                 return 0;
 212         }
 213         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 214         ret = dss_exec_cmdline_pid(&create_pid,
 215                 conf.pre_create_hook_arg, fds);
 216         if (ret < 0)
 217                 return ret;
 218         snapshot_creation_status = HS_PRE_RUNNING;
 219         return ret;
 220 }
 221
 222 static int pre_remove_hook(struct snapshot *s, const char *why)
 223 {
 224         int ret, fds[3] = {0, 0, 0};
 225         char *cmd;
 226
 227         if (!s)
 228                 return 0;
 229         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 230         assert(snapshot_removal_status == HS_READY);
 231         assert(remove_pid == 0);
 232         assert(!snapshot_currently_being_removed);
 233
 234         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 235         *snapshot_currently_being_removed = *s;
 236         snapshot_currently_being_removed->name = dss_strdup(s->name);
 237
 238         if (!conf.pre_remove_hook_given) {
 239                 snapshot_removal_status = HS_PRE_SUCCESS;
 240                 return 0;
 241         }
 242         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 243                 conf.dest_dir_arg, s->name);
 244         DSS_DEBUG_LOG("executing %s\n", cmd);
 245         ret = dss_exec_cmdline_pid(&remove_pid,
 246                 conf.pre_remove_hook_arg, fds);
 247         free(cmd);
 248         if (ret < 0)
 249                 return ret;
 250         snapshot_removal_status = HS_PRE_RUNNING;
 251         return ret;
 252 }
 253
 254 static int exec_rm(void)
 255 {
 256         struct snapshot *s = snapshot_currently_being_removed;
 257         int fds[3] = {0, 0, 0};
 258         char *new_name = being_deleted_name(s);
 259         char *argv[] = {"rm", "-rf", new_name, NULL};
 260         int ret;
 261
 262         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 263         assert(remove_pid == 0);
 264
 265         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 266         ret = dss_rename(s->name, new_name);
 267         if (ret < 0)
 268                 goto out;
 269         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 270         if (ret < 0)
 271                 goto out;
 272         snapshot_removal_status = HS_RUNNING;
 273 out:
 274         free(new_name);
 275         return ret;
 276 }
 277
 278 static int snapshot_is_being_created(struct snapshot *s)
 279 {
 280         return s->creation_time == current_snapshot_creation_time;
 281 }
 282
 283 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 284 {
 285         struct snapshot *s;
 286         int i;
 287
 288         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 289         FOR_EACH_SNAPSHOT(s, i, sl) {
 290                 if (snapshot_is_being_created(s))
 291                         continue;
 292                 /*
 293                  * We know that no rm is currently running, so if s is marked
 294                  * as being deleted, a previously started rm must have failed.
 295                  */
 296                 if (s->flags & SS_BEING_DELETED)
 297                         return s;
 298
 299                 if (s->flags & SS_COMPLETE) /* good snapshot */
 300                         continue;
 301                 /*
 302                  * This snapshot is incomplete and it is not the snapshot
 303                  * currently being created. However, we must not remove it if
 304                  * rsync is about to be restarted. As only the newest snapshot
 305                  * can be restarted, this snapshot is orphaned if it is not the
 306                  * newest snapshot or if we are not about to restart rsync.
 307                  */
 308                 if (get_newest_snapshot(sl) != s)
 309                         return s;
 310                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 311                         return s;
 312         }
 313         /* no orphaned snapshots */
 314         return NULL;
 315 }
 316
 317 static int is_reference_snapshot(struct snapshot *s)
 318 {
 319         if (!name_of_reference_snapshot)
 320                 return 0;
 321         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 322 }
 323
 324 /*
 325  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 326  */
 327 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 328 {
 329         int i, interval;
 330         struct snapshot *s;
 331         unsigned missing = 0;
 332
 333         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 334         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 335                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 336                 unsigned num = sl->interval_count[interval];
 337                 struct snapshot *victim = NULL, *prev = NULL;
 338                 int64_t score = LONG_MAX;
 339
 340                 if (keep >= num)
 341                         missing += keep - num;
 342 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 343 //                      interval, keep, num, missing);
 344                 if (keep + missing >= num)
 345                         continue;
 346                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 347                 FOR_EACH_SNAPSHOT(s, i, sl) {
 348                         int64_t this_score;
 349
 350                         if (snapshot_is_being_created(s))
 351                                 continue;
 352                         if (is_reference_snapshot(s))
 353                                 continue;
 354                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 355                         if (s->interval > interval) {
 356                                 prev = s;
 357                                 continue;
 358                         }
 359                         if (s->interval < interval)
 360                                 break;
 361                         if (!victim) {
 362                                 victim = s;
 363                                 prev = s;
 364                                 continue;
 365                         }
 366                         assert(prev);
 367                         /* check if s is a better victim */
 368                         this_score = s->creation_time - prev->creation_time;
 369                         assert(this_score >= 0);
 370                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 371                         if (this_score < score) {
 372                                 score = this_score;
 373                                 victim = s;
 374                         }
 375                         prev = s;
 376                 }
 377                 assert(victim);
 378                 return victim;
 379         }
 380         return NULL;
 381 }
 382
 383 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 384 {
 385         int i;
 386         struct snapshot *s;
 387
 388         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 389                 conf.num_intervals_arg);
 390         FOR_EACH_SNAPSHOT(s, i, sl) {
 391                 if (snapshot_is_being_created(s))
 392                         continue;
 393                 if (is_reference_snapshot(s))
 394                         continue;
 395                 if (s->interval < conf.num_intervals_arg)
 396                         continue;
 397                 return s;
 398         }
 399         return NULL;
 400 }
 401
 402 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 403 {
 404         int i;
 405         struct snapshot *s;
 406         FOR_EACH_SNAPSHOT(s, i, sl) {
 407                 if (snapshot_is_being_created(s))
 408                         continue;
 409                 if (is_reference_snapshot(s))
 410                         continue;
 411                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 412                 return s;
 413         }
 414         return NULL;
 415 }
 416
 417 static int rename_incomplete_snapshot(int64_t start)
 418 {
 419         char *old_name;
 420         int ret;
 421
 422         free(path_to_last_complete_snapshot);
 423         ret = complete_name(start, get_current_time(),
 424                 &path_to_last_complete_snapshot);
 425         if (ret < 0)
 426                 return ret;
 427         old_name = incomplete_name(start);
 428         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 429         if (ret >= 0)
 430                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 431                         path_to_last_complete_snapshot);
 432         free(old_name);
 433         return ret;
 434 }
 435
 436 static int try_to_free_disk_space(void)
 437 {
 438         int ret;
 439         struct snapshot_list sl;
 440         struct snapshot *victim;
 441         struct timeval now;
 442         const char *why;
 443         int low_disk_space;
 444
 445         ret = disk_space_low();
 446         if (ret < 0)
 447                 return ret;
 448         low_disk_space = ret;
 449         gettimeofday(&now, NULL);
 450         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 451                 return 0;
 452         if (!low_disk_space) {
 453                 if (conf.keep_redundant_given)
 454                         return 0;
 455                 if (snapshot_creation_status != HS_READY)
 456                         return 0;
 457                 if (next_snapshot_is_due())
 458                         return 0;
 459         }
 460         dss_get_snapshot_list(&sl);
 461         ret = 0;
 462         if (!low_disk_space && sl.num_snapshots <= 1)
 463                 goto out;
 464         why = "outdated";
 465         victim = find_outdated_snapshot(&sl);
 466         if (victim)
 467                 goto remove;
 468         why = "redundant";
 469         victim = find_redundant_snapshot(&sl);
 470         if (victim)
 471                 goto remove;
 472         /* try harder only if disk space is low */
 473         if (!low_disk_space)
 474                 goto out;
 475         why = "orphaned";
 476         victim = find_orphaned_snapshot(&sl);
 477         if (victim)
 478                 goto remove;
 479         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 480         victim = find_oldest_removable_snapshot(&sl);
 481         if (victim)
 482                 goto remove;
 483         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 484         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 485         goto out;
 486 remove:
 487         ret = pre_remove_hook(victim, why);
 488 out:
 489         free_snapshot_list(&sl);
 490         return ret;
 491 }
 492
 493 static int post_create_hook(void)
 494 {
 495         int ret, fds[3] = {0, 0, 0};
 496         char *cmd;
 497
 498         if (!conf.post_create_hook_given) {
 499                 snapshot_creation_status = HS_READY;
 500                 return 0;
 501         }
 502         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 503                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 504         DSS_NOTICE_LOG("executing %s\n", cmd);
 505         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 506         free(cmd);
 507         if (ret < 0)
 508                 return ret;
 509         snapshot_creation_status = HS_POST_RUNNING;
 510         return ret;
 511 }
 512
 513 static int post_remove_hook(void)
 514 {
 515         int ret, fds[3] = {0, 0, 0};
 516         char *cmd;
 517         struct snapshot *s = snapshot_currently_being_removed;
 518
 519         assert(s);
 520
 521         if (!conf.post_remove_hook_given) {
 522                 snapshot_removal_status = HS_READY;
 523                 return 0;
 524         }
 525         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 526                 conf.dest_dir_arg, s->name);
 527         DSS_NOTICE_LOG("executing %s\n", cmd);
 528         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 529         free(cmd);
 530         if (ret < 0)
 531                 return ret;
 532         snapshot_removal_status = HS_POST_RUNNING;
 533         return ret;
 534 }
 535
 536 static void kill_process(pid_t pid)
 537 {
 538         if (!pid)
 539                 return;
 540         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 541         kill(pid, SIGTERM);
 542 }
 543
 544 static void stop_create_process(void)
 545 {
 546         if (!create_pid || create_process_stopped)
 547                 return;
 548         kill(SIGSTOP, create_pid);
 549         create_process_stopped = 1;
 550 }
 551
 552 static void restart_create_process(void)
 553 {
 554         if (!create_pid || !create_process_stopped)
 555                 return;
 556         kill (SIGCONT, create_pid);
 557         create_process_stopped = 0;
 558 }
 559
 560 /**
 561  * Print a log message about the exit status of a child.
 562  */
 563 static void log_termination_msg(pid_t pid, int status)
 564 {
 565         if (WIFEXITED(status))
 566                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 567                         WEXITSTATUS(status));
 568         else if (WIFSIGNALED(status))
 569                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 570                         WTERMSIG(status));
 571         else
 572                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 573 }
 574
 575 static int wait_for_process(pid_t pid, int *status)
 576 {
 577         int ret;
 578
 579         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 580         for (;;) {
 581                 fd_set rfds;
 582
 583                 FD_ZERO(&rfds);
 584                 FD_SET(signal_pipe, &rfds);
 585                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 586                 if (ret < 0)
 587                         break;
 588                 ret = next_signal();
 589                 if (!ret)
 590                         continue;
 591                 if (ret == SIGCHLD) {
 592                         ret = waitpid(pid, status, 0);
 593                         if (ret >= 0)
 594                                 break;
 595                         if (errno != EINTR) { /* error */
 596                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 597                                 break;
 598                         }
 599                 }
 600                 /* SIGINT or SIGTERM */
 601                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 602                 kill(pid, SIGTERM);
 603         }
 604         if (ret < 0)
 605                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 606         else
 607                 log_termination_msg(pid, *status);
 608         return ret;
 609 }
 610
 611 static void handle_pre_remove_exit(int status)
 612 {
 613         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 614                 snapshot_removal_status = HS_READY;
 615                 gettimeofday(&next_removal_check, NULL);
 616                 next_removal_check.tv_sec += 60;
 617                 return;
 618         }
 619         snapshot_removal_status = HS_PRE_SUCCESS;
 620 }
 621
 622 static int handle_rm_exit(int status)
 623 {
 624         if (!WIFEXITED(status)) {
 625                 snapshot_removal_status = HS_READY;
 626                 return -E_INVOLUNTARY_EXIT;
 627         }
 628         if (WEXITSTATUS(status)) {
 629                 snapshot_removal_status = HS_READY;
 630                 return -E_BAD_EXIT_CODE;
 631         }
 632         if (conf.post_remove_hook_given)
 633                 snapshot_removal_status = HS_SUCCESS;
 634         else
 635                 snapshot_removal_status = HS_READY;
 636         return 1;
 637 }
 638
 639 static void handle_post_remove_exit(void)
 640 {
 641         snapshot_removal_status = HS_READY;
 642 }
 643
 644 static int handle_remove_exit(int status)
 645 {
 646         int ret;
 647         struct snapshot *s = snapshot_currently_being_removed;
 648
 649         assert(s);
 650         switch (snapshot_removal_status) {
 651         case HS_PRE_RUNNING:
 652                 handle_pre_remove_exit(status);
 653                 ret = 1;
 654                 break;
 655         case HS_RUNNING:
 656                 ret = handle_rm_exit(status);
 657                 break;
 658         case HS_POST_RUNNING:
 659                 handle_post_remove_exit();
 660                 ret = 1;
 661                 break;
 662         default:
 663                 ret = -E_BUG;
 664         }
 665         if (snapshot_removal_status == HS_READY) {
 666                 free(s->name);
 667                 free(s);
 668                 snapshot_currently_being_removed = NULL;
 669         }
 670         remove_pid = 0;
 671         return ret;
 672 }
 673
 674 static int wait_for_remove_process(void)
 675 {
 676         int status, ret;
 677
 678         assert(remove_pid);
 679         assert(
 680                 snapshot_removal_status == HS_PRE_RUNNING ||
 681                 snapshot_removal_status == HS_RUNNING ||
 682                 snapshot_removal_status == HS_POST_RUNNING
 683         );
 684         ret = wait_for_process(remove_pid, &status);
 685         if (ret < 0)
 686                 return ret;
 687         return handle_remove_exit(status);
 688 }
 689
 690 static int handle_rsync_exit(int status)
 691 {
 692         int es, ret;
 693
 694         if (!WIFEXITED(status)) {
 695                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 696                 ret = -E_INVOLUNTARY_EXIT;
 697                 snapshot_creation_status = HS_READY;
 698                 goto out;
 699         }
 700         es = WEXITSTATUS(status);
 701         /*
 702          * Restart rsync on non-fatal errors:
 703          * 12: Error in rsync protocol data stream
 704          * 13: Errors with program diagnostics
 705          */
 706         if (es == 12 || es == 13) {
 707                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 708                         (int)create_pid, es);
 709                 snapshot_creation_status = HS_NEEDS_RESTART;
 710                 gettimeofday(&next_snapshot_time, NULL);
 711                 next_snapshot_time.tv_sec += 60;
 712                 ret = 1;
 713                 goto out;
 714         }
 715         if (es != 0 && es != 23 && es != 24) {
 716                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 717                 ret = -E_BAD_EXIT_CODE;
 718                 snapshot_creation_status = HS_READY;
 719                 goto out;
 720         }
 721         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 722         if (ret < 0)
 723                 goto out;
 724         snapshot_creation_status = HS_SUCCESS;
 725         free(name_of_reference_snapshot);
 726         name_of_reference_snapshot = NULL;
 727 out:
 728         create_pid = 0;
 729         create_process_stopped = 0;
 730         return ret;
 731 }
 732
 733 static int handle_pre_create_hook_exit(int status)
 734 {
 735         int es, ret;
 736         static int warn_count;
 737
 738         if (!WIFEXITED(status)) {
 739                 snapshot_creation_status = HS_READY;
 740                 ret = -E_INVOLUNTARY_EXIT;
 741                 goto out;
 742         }
 743         es = WEXITSTATUS(status);
 744         if (es) {
 745                 if (!warn_count--) {
 746                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 747                                 conf.pre_create_hook_arg, es);
 748                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 749                         warn_count = 60; /* warn only once per hour */
 750                 }
 751                 gettimeofday(&next_snapshot_time, NULL);
 752                 next_snapshot_time.tv_sec += 60;
 753                 snapshot_creation_status = HS_READY;
 754                 ret = 0;
 755                 goto out;
 756         }
 757         warn_count = 0;
 758         snapshot_creation_status = HS_PRE_SUCCESS;
 759         ret = 1;
 760 out:
 761         create_pid = 0;
 762         return ret;
 763 }
 764
 765 static int handle_sigchld(void)
 766 {
 767         pid_t pid;
 768         int status, ret = reap_child(&pid, &status);
 769
 770         if (ret <= 0)
 771                 return ret;
 772
 773         if (pid == create_pid) {
 774                 switch (snapshot_creation_status) {
 775                 case HS_PRE_RUNNING:
 776                         return handle_pre_create_hook_exit(status);
 777                 case HS_RUNNING:
 778                         return handle_rsync_exit(status);
 779                 case HS_POST_RUNNING:
 780                         snapshot_creation_status = HS_READY;
 781                         return 1;
 782                 default:
 783                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 784                                 snapshot_creation_status);
 785                         return -E_BUG;
 786                 }
 787         }
 788         if (pid == remove_pid) {
 789                 ret = handle_remove_exit(status);
 790                 if (ret < 0)
 791                         return ret;
 792                 return ret;
 793         }
 794         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 795         return -E_BUG;
 796 }
 797
 798 static int check_config(void)
 799 {
 800         if (conf.unit_interval_arg <= 0) {
 801                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 802                 return -E_INVALID_NUMBER;
 803         }
 804         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 805         if (conf.num_intervals_arg <= 0) {
 806                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 807                 return -E_INVALID_NUMBER;
 808         }
 809         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 810         return 1;
 811 }
 812
 813 static int parse_config_file(int override)
 814 {
 815         int ret;
 816         char *config_file;
 817         struct stat statbuf;
 818         char *old_logfile_arg = NULL;
 819         int old_daemon_given = 0;
 820
 821         if (conf.config_file_given)
 822                 config_file = dss_strdup(conf.config_file_arg);
 823         else {
 824                 char *home = get_homedir();
 825                 config_file = make_message("%s/.dssrc", home);
 826                 free(home);
 827         }
 828         if (override) { /* SIGHUP */
 829                 if (conf.logfile_given)
 830                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 831                 old_daemon_given = conf.daemon_given;
 832         }
 833
 834         ret = stat(config_file, &statbuf);
 835         if (ret && conf.config_file_given) {
 836                 ret = -ERRNO_TO_DSS_ERROR(errno);
 837                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 838                 goto out;
 839         }
 840         if (!ret) {
 841                 struct cmdline_parser_params params = {
 842                         .override = override,
 843                         .initialize = 0,
 844                         .check_required = 1,
 845                         .check_ambiguity = 0,
 846                         .print_errors = 1
 847                 };
 848                 if (override) { /* invalidate all rsync options */
 849                         int i;
 850
 851                         for (i = 0; i < conf.rsync_option_given; i++) {
 852                                 free(conf.rsync_option_arg[i]);
 853                                 conf.rsync_option_arg[i] = NULL;
 854                         }
 855                         conf.rsync_option_given = 0;
 856                 }
 857                 cmdline_parser_config_file(config_file, &conf, &params);
 858         }
 859         ret = check_config();
 860         if (ret < 0)
 861                 goto out;
 862         if (override) {
 863                 /* don't change daemon mode on SIGHUP */
 864                 conf.daemon_given = old_daemon_given;
 865                 close_log(logfile);
 866                 logfile = NULL;
 867                 if (conf.logfile_given)
 868                         free(old_logfile_arg);
 869                 else if (conf.daemon_given) { /* re-use old logfile */
 870                         conf.logfile_arg = old_logfile_arg;
 871                         conf.logfile_given = 1;
 872                 }
 873         }
 874         if (conf.logfile_given) {
 875                 logfile = open_log(conf.logfile_arg);
 876                 log_welcome(conf.loglevel_arg);
 877         }
 878         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 879 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 880 out:
 881         free(config_file);
 882         if (ret < 0)
 883                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 884         return ret;
 885 }
 886
 887 static int change_to_dest_dir(void)
 888 {
 889         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 890         return dss_chdir(conf.dest_dir_arg);
 891 }
 892
 893 static int handle_sighup(void)
 894 {
 895         int ret;
 896
 897         DSS_NOTICE_LOG("SIGHUP\n");
 898         ret = parse_config_file(1);
 899         if (ret < 0)
 900                 return ret;
 901         return change_to_dest_dir();
 902 }
 903
 904 static int handle_signal(void)
 905 {
 906         int sig, ret = next_signal();
 907
 908         if (ret <= 0)
 909                 goto out;
 910         sig = ret;
 911         switch (sig) {
 912         case SIGINT:
 913         case SIGTERM:
 914                 restart_create_process();
 915                 kill_process(create_pid);
 916                 kill_process(remove_pid);
 917                 ret = -E_SIGNAL;
 918                 break;
 919         case SIGHUP:
 920                 ret = handle_sighup();
 921                 break;
 922         case SIGCHLD:
 923                 ret = handle_sigchld();
 924                 break;
 925         }
 926 out:
 927         if (ret < 0)
 928                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 929         return ret;
 930 }
 931
 932 /*
 933  * We can not use rsync locally if the local user is different from the remote
 934  * user or if the src dir is not on the local host (or both).
 935  */
 936 static int use_rsync_locally(char *logname)
 937 {
 938         char *h = conf.remote_host_arg;
 939
 940         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 941                 return 0;
 942         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 943                 return 0;
 944         return 1;
 945 }
 946
 947 static void create_rsync_argv(char ***argv, int64_t *num)
 948 {
 949         char *logname;
 950         int i = 0, j;
 951         struct snapshot_list sl;
 952
 953         dss_get_snapshot_list(&sl);
 954         assert(!name_of_reference_snapshot);
 955         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
 956         free_snapshot_list(&sl);
 957
 958         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 959         (*argv)[i++] = dss_strdup("rsync");
 960         (*argv)[i++] = dss_strdup("-aq");
 961         (*argv)[i++] = dss_strdup("--delete");
 962         for (j = 0; j < conf.rsync_option_given; j++)
 963                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 964         if (name_of_reference_snapshot) {
 965                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
 966                 (*argv)[i++] = make_message("--link-dest=../%s",
 967                         name_of_reference_snapshot);
 968         } else
 969                 DSS_INFO_LOG("no suitable reference snapshot found\n");
 970         logname = dss_logname();
 971         if (use_rsync_locally(logname))
 972                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 973         else
 974                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 975                         conf.remote_user_arg : logname,
 976                         conf.remote_host_arg, conf.source_dir_arg);
 977         free(logname);
 978         *num = get_current_time();
 979         (*argv)[i++] = incomplete_name(*num);
 980         (*argv)[i++] = NULL;
 981         for (j = 0; j < i; j++)
 982                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 983 }
 984
 985 static void free_rsync_argv(char **argv)
 986 {
 987         int i;
 988
 989         if (!argv)
 990                 return;
 991         for (i = 0; argv[i]; i++)
 992                 free(argv[i]);
 993         free(argv);
 994 }
 995
 996 static int create_snapshot(char **argv)
 997 {
 998         int ret, fds[3] = {0, 0, 0};
 999         char *name;
1000
1001         name = incomplete_name(current_snapshot_creation_time);
1002         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
1003         free(name);
1004         ret = dss_exec(&create_pid, argv[0], argv, fds);
1005         if (ret < 0)
1006                 return ret;
1007         snapshot_creation_status = HS_RUNNING;
1008         return ret;
1009 }
1010
1011 static int select_loop(void)
1012 {
1013         int ret;
1014         /* check every 60 seconds for free disk space */
1015         struct timeval tv;
1016         char **rsync_argv = NULL;
1017
1018         for (;;) {
1019                 fd_set rfds;
1020                 struct timeval *tvp;
1021
1022                 if (remove_pid)
1023                         tvp = NULL; /* sleep until rm hook/process dies */
1024                 else { /* sleep one minute */
1025                         tv.tv_sec = 60;
1026                         tv.tv_usec = 0;
1027                         tvp = &tv;
1028                 }
1029                 FD_ZERO(&rfds);
1030                 FD_SET(signal_pipe, &rfds);
1031                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1032                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1033                 if (ret < 0)
1034                         goto out;
1035                 if (FD_ISSET(signal_pipe, &rfds)) {
1036                         ret = handle_signal();
1037                         if (ret < 0)
1038                                 goto out;
1039                 }
1040                 if (remove_pid)
1041                         continue;
1042                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1043                         ret = exec_rm();
1044                         if (ret < 0)
1045                                 goto out;
1046                         continue;
1047                 }
1048                 if (snapshot_removal_status == HS_SUCCESS) {
1049                         ret = post_remove_hook();
1050                         if (ret < 0)
1051                                 goto out;
1052                         continue;
1053                 }
1054                 ret = try_to_free_disk_space();
1055                 if (ret < 0)
1056                         goto out;
1057                 if (snapshot_removal_status != HS_READY) {
1058                         stop_create_process();
1059                         continue;
1060                 }
1061                 restart_create_process();
1062                 switch (snapshot_creation_status) {
1063                 case HS_READY:
1064                         if (!next_snapshot_is_due())
1065                                 continue;
1066                         ret = pre_create_hook();
1067                         if (ret < 0)
1068                                 goto out;
1069                         continue;
1070                 case HS_PRE_RUNNING:
1071                 case HS_RUNNING:
1072                 case HS_POST_RUNNING:
1073                         continue;
1074                 case HS_PRE_SUCCESS:
1075                         free_rsync_argv(rsync_argv);
1076                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1077                         /* fall through */
1078                 case HS_NEEDS_RESTART:
1079                         ret = create_snapshot(rsync_argv);
1080                         if (ret < 0)
1081                                 goto out;
1082                         continue;
1083                 case HS_SUCCESS:
1084                         ret = post_create_hook();
1085                         if (ret < 0)
1086                                 goto out;
1087                         continue;
1088                 }
1089         }
1090 out:
1091         return ret;
1092 }
1093
1094 static void exit_hook(int exit_code)
1095 {
1096         int fds[3] = {0, 0, 0};
1097         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1098         pid_t pid;
1099
1100         if (!conf.exit_hook_given)
1101                 return;
1102         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1103         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1104 }
1105
1106 static int com_run(void)
1107 {
1108         int ret;
1109
1110         if (conf.dry_run_given) {
1111                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1112                 return -E_SYNTAX;
1113         }
1114         ret = install_sighandler(SIGHUP);
1115         if (ret < 0)
1116                 return ret;
1117         ret = select_loop();
1118         if (ret >= 0) /* impossible */
1119                 ret = -E_BUG;
1120         exit_hook(ret);
1121         return ret;
1122 }
1123
1124 static int com_prune(void)
1125 {
1126         int ret;
1127         struct snapshot_list sl;
1128         struct snapshot *victim;
1129         struct disk_space ds;
1130         const char *why;
1131
1132         ret = get_disk_space(".", &ds);
1133         if (ret < 0)
1134                 return ret;
1135         log_disk_space(&ds);
1136         dss_get_snapshot_list(&sl);
1137         why = "outdated";
1138         victim = find_outdated_snapshot(&sl);
1139         if (victim)
1140                 goto rm;
1141         why = "redundant";
1142         victim = find_redundant_snapshot(&sl);
1143         if (victim)
1144                 goto rm;
1145         ret = 0;
1146         goto out;
1147 rm:
1148         if (conf.dry_run_given) {
1149                 dss_msg("%s snapshot %s (interval = %i)\n",
1150                         why, victim->name, victim->interval);
1151                 ret = 0;
1152                 goto out;
1153         }
1154         ret = pre_remove_hook(victim, why);
1155         if (ret < 0)
1156                 goto out;
1157         if (snapshot_removal_status == HS_PRE_RUNNING) {
1158                 ret = wait_for_remove_process();
1159                 if (ret < 0)
1160                         goto out;
1161                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1162                         goto out;
1163         }
1164         ret = exec_rm();
1165         if (ret < 0)
1166                 goto out;
1167         ret = wait_for_remove_process();
1168         if (ret < 0)
1169                 goto out;
1170         if (snapshot_removal_status != HS_SUCCESS)
1171                 goto out;
1172         ret = post_remove_hook();
1173         if (ret < 0)
1174                 goto out;
1175         if (snapshot_removal_status != HS_POST_RUNNING)
1176                 goto out;
1177         ret = wait_for_remove_process();
1178         if (ret < 0)
1179                 goto out;
1180         ret = 1;
1181 out:
1182         free_snapshot_list(&sl);
1183         return ret;
1184 }
1185
1186 static int com_create(void)
1187 {
1188         int ret, status;
1189         char **rsync_argv;
1190
1191         if (conf.dry_run_given) {
1192                 int i;
1193                 char *msg = NULL;
1194                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1195                 for (i = 0; rsync_argv[i]; i++) {
1196                         char *tmp = msg;
1197                         msg = make_message("%s%s%s", tmp? tmp : "",
1198                                 tmp? " " : "", rsync_argv[i]);
1199                         free(tmp);
1200                 }
1201                 free_rsync_argv(rsync_argv);
1202                 dss_msg("%s\n", msg);
1203                 free(msg);
1204                 return 1;
1205         }
1206         ret = pre_create_hook();
1207         if (ret < 0)
1208                 return ret;
1209         if (create_pid) {
1210                 ret = wait_for_process(create_pid, &status);
1211                 if (ret < 0)
1212                         return ret;
1213                 ret = handle_pre_create_hook_exit(status);
1214                 if (ret <= 0) /* error, or pre-create failed */
1215                         return ret;
1216         }
1217         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1218         ret = create_snapshot(rsync_argv);
1219         if (ret < 0)
1220                 goto out;
1221         ret = wait_for_process(create_pid, &status);
1222         if (ret < 0)
1223                 goto out;
1224         ret = handle_rsync_exit(status);
1225         if (ret < 0)
1226                 goto out;
1227         post_create_hook();
1228         if (create_pid)
1229                 ret = wait_for_process(create_pid, &status);
1230 out:
1231         free_rsync_argv(rsync_argv);
1232         return ret;
1233 }
1234
1235 static int com_ls(void)
1236 {
1237         int i;
1238         struct snapshot_list sl;
1239         struct snapshot *s;
1240
1241         dss_get_snapshot_list(&sl);
1242         FOR_EACH_SNAPSHOT(s, i, &sl) {
1243                 int64_t d = 0;
1244                 if (s->flags & SS_COMPLETE)
1245                         d = (s->completion_time - s->creation_time) / 60;
1246                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1247         };
1248         free_snapshot_list(&sl);
1249         return 1;
1250 }
1251
1252 static int setup_signal_handling(void)
1253 {
1254         int ret;
1255
1256         DSS_INFO_LOG("setting up signal handlers\n");
1257         signal_pipe = signal_init(); /* always successful */
1258         ret = install_sighandler(SIGINT);
1259         if (ret < 0)
1260                 return ret;
1261         ret = install_sighandler(SIGTERM);
1262         if (ret < 0)
1263                 return ret;
1264         return install_sighandler(SIGCHLD);
1265 }
1266
1267 /**
1268  * The main function of dss.
1269  *
1270  * \param argc Usual argument count.
1271  * \param argv Usual argument vector.
1272  */
1273 int main(int argc, char **argv)
1274 {
1275         int ret;
1276         struct cmdline_parser_params params = {
1277                 .override = 0,
1278                 .initialize = 1,
1279                 .check_required = 0,
1280                 .check_ambiguity = 0,
1281                 .print_errors = 1
1282         };
1283
1284         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1285         ret = parse_config_file(0);
1286         if (ret < 0)
1287                 goto out;
1288         if (conf.daemon_given)
1289                 daemon_init();
1290         ret = change_to_dest_dir();
1291         if (ret < 0)
1292                 goto out;
1293         ret = setup_signal_handling();
1294         if (ret < 0)
1295                 goto out;
1296         ret = call_command_handler();
1297 out:
1298         if (ret < 0)
1299                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1300         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1301 }