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         assert(snapshot_creation_status == HS_READY);
 201         if (!next_snapshot_time_is_valid())
 202                 next_snapshot_time = compute_next_snapshot_time();
 203         if (next_snapshot_time <= now) {
 204                 DSS_DEBUG_LOG("next snapshot: now\n");
 205                 return 1;
 206         }
 207         DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
 208                 next_snapshot_time - now);
 209         return 0;
 210 }
 211
 212 static int pre_create_hook(void)
 213 {
 214         int ret, fds[3] = {0, 0, 0};
 215
 216         assert(snapshot_creation_status == HS_READY);
 217         /* make sure that the next snapshot time will be recomputed */
 218         invalidate_next_snapshot_time();
 219         if (!conf.pre_create_hook_given) {
 220                 snapshot_creation_status = HS_PRE_SUCCESS;
 221                 return 0;
 222         }
 223         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 224         ret = dss_exec_cmdline_pid(&create_pid,
 225                 conf.pre_create_hook_arg, fds);
 226         if (ret < 0)
 227                 return ret;
 228         snapshot_creation_status = HS_PRE_RUNNING;
 229         return ret;
 230 }
 231
 232 static int pre_remove_hook(struct snapshot *s, const char *why)
 233 {
 234         int ret, fds[3] = {0, 0, 0};
 235         char *cmd;
 236
 237         if (!s)
 238                 return 0;
 239         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 240         assert(snapshot_removal_status == HS_READY);
 241         assert(remove_pid == 0);
 242         assert(!snapshot_currently_being_removed);
 243
 244         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 245         *snapshot_currently_being_removed = *s;
 246         snapshot_currently_being_removed->name = dss_strdup(s->name);
 247
 248         if (!conf.pre_remove_hook_given) {
 249                 snapshot_removal_status = HS_PRE_SUCCESS;
 250                 return 0;
 251         }
 252         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 253                 conf.dest_dir_arg, s->name);
 254         DSS_DEBUG_LOG("executing %s\n", cmd);
 255         ret = dss_exec_cmdline_pid(&remove_pid,
 256                 conf.pre_remove_hook_arg, fds);
 257         free(cmd);
 258         if (ret < 0)
 259                 return ret;
 260         snapshot_removal_status = HS_PRE_RUNNING;
 261         return ret;
 262 }
 263
 264 static int exec_rm(void)
 265 {
 266         struct snapshot *s = snapshot_currently_being_removed;
 267         int fds[3] = {0, 0, 0};
 268         char *new_name = being_deleted_name(s);
 269         char *argv[] = {"rm", "-rf", new_name, NULL};
 270         int ret;
 271
 272         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 273         assert(remove_pid == 0);
 274
 275         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 276         ret = dss_rename(s->name, new_name);
 277         if (ret < 0)
 278                 goto out;
 279         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 280         if (ret < 0)
 281                 goto out;
 282         snapshot_removal_status = HS_RUNNING;
 283 out:
 284         free(new_name);
 285         return ret;
 286 }
 287
 288 static int snapshot_is_being_created(struct snapshot *s)
 289 {
 290         return s->creation_time == current_snapshot_creation_time;
 291 }
 292
 293 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 294 {
 295         struct snapshot *s;
 296         int i;
 297
 298         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 299         FOR_EACH_SNAPSHOT(s, i, sl) {
 300                 if (snapshot_is_being_created(s))
 301                         continue;
 302                 /*
 303                  * We know that no rm is currently running, so if s is marked
 304                  * as being deleted, a previously started rm must have failed.
 305                  */
 306                 if (s->flags & SS_BEING_DELETED)
 307                         return s;
 308
 309                 if (s->flags & SS_COMPLETE) /* good snapshot */
 310                         continue;
 311                 /*
 312                  * This snapshot is incomplete and it is not the snapshot
 313                  * currently being created. However, we must not remove it if
 314                  * rsync is about to be restarted. As only the newest snapshot
 315                  * can be restarted, this snapshot is orphaned if it is not the
 316                  * newest snapshot or if we are not about to restart rsync.
 317                  */
 318                 if (get_newest_snapshot(sl) != s)
 319                         return s;
 320                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 321                         return s;
 322         }
 323         /* no orphaned snapshots */
 324         return NULL;
 325 }
 326
 327 static int is_reference_snapshot(struct snapshot *s)
 328 {
 329         if (!name_of_reference_snapshot)
 330                 return 0;
 331         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 332 }
 333
 334 /*
 335  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 336  */
 337 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 338 {
 339         int i, interval;
 340         struct snapshot *s;
 341         unsigned missing = 0;
 342
 343         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 344         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 345                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 346                 unsigned num = sl->interval_count[interval];
 347                 struct snapshot *victim = NULL, *prev = NULL;
 348                 int64_t score = LONG_MAX;
 349
 350                 if (keep >= num)
 351                         missing += keep - num;
 352 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 353 //                      interval, keep, num, missing);
 354                 if (keep + missing >= num)
 355                         continue;
 356                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 357                 FOR_EACH_SNAPSHOT(s, i, sl) {
 358                         int64_t this_score;
 359
 360                         if (snapshot_is_being_created(s))
 361                                 continue;
 362                         if (is_reference_snapshot(s))
 363                                 continue;
 364                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 365                         if (s->interval > interval) {
 366                                 prev = s;
 367                                 continue;
 368                         }
 369                         if (s->interval < interval)
 370                                 break;
 371                         if (!victim) {
 372                                 victim = s;
 373                                 prev = s;
 374                                 continue;
 375                         }
 376                         assert(prev);
 377                         /* check if s is a better victim */
 378                         this_score = s->creation_time - prev->creation_time;
 379                         assert(this_score >= 0);
 380                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 381                         if (this_score < score) {
 382                                 score = this_score;
 383                                 victim = s;
 384                         }
 385                         prev = s;
 386                 }
 387                 assert(victim);
 388                 return victim;
 389         }
 390         return NULL;
 391 }
 392
 393 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 394 {
 395         int i;
 396         struct snapshot *s;
 397
 398         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 399                 conf.num_intervals_arg);
 400         FOR_EACH_SNAPSHOT(s, i, sl) {
 401                 if (snapshot_is_being_created(s))
 402                         continue;
 403                 if (is_reference_snapshot(s))
 404                         continue;
 405                 if (s->interval < conf.num_intervals_arg)
 406                         continue;
 407                 return s;
 408         }
 409         return NULL;
 410 }
 411
 412 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 413 {
 414         int i;
 415         struct snapshot *s;
 416         FOR_EACH_SNAPSHOT(s, i, sl) {
 417                 if (snapshot_is_being_created(s))
 418                         continue;
 419                 if (is_reference_snapshot(s))
 420                         continue;
 421                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 422                 return s;
 423         }
 424         return NULL;
 425 }
 426
 427 static int rename_incomplete_snapshot(int64_t start)
 428 {
 429         char *old_name;
 430         int ret;
 431
 432         free(path_to_last_complete_snapshot);
 433         ret = complete_name(start, get_current_time(),
 434                 &path_to_last_complete_snapshot);
 435         if (ret < 0)
 436                 return ret;
 437         old_name = incomplete_name(start);
 438         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 439         if (ret >= 0)
 440                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 441                         path_to_last_complete_snapshot);
 442         free(old_name);
 443         return ret;
 444 }
 445
 446 static int try_to_free_disk_space(void)
 447 {
 448         int ret;
 449         struct snapshot_list sl;
 450         struct snapshot *victim;
 451         struct timeval now;
 452         const char *why;
 453         int low_disk_space;
 454
 455         ret = disk_space_low();
 456         if (ret < 0)
 457                 return ret;
 458         low_disk_space = ret;
 459         gettimeofday(&now, NULL);
 460         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 461                 return 0;
 462         if (!low_disk_space) {
 463                 if (conf.keep_redundant_given)
 464                         return 0;
 465                 if (snapshot_creation_status != HS_READY)
 466                         return 0;
 467                 if (next_snapshot_is_due())
 468                         return 0;
 469         }
 470         dss_get_snapshot_list(&sl);
 471         ret = 0;
 472         if (!low_disk_space && sl.num_snapshots <= 1)
 473                 goto out;
 474         why = "outdated";
 475         victim = find_outdated_snapshot(&sl);
 476         if (victim)
 477                 goto remove;
 478         why = "redundant";
 479         victim = find_redundant_snapshot(&sl);
 480         if (victim)
 481                 goto remove;
 482         /* try harder only if disk space is low */
 483         if (!low_disk_space)
 484                 goto out;
 485         why = "orphaned";
 486         victim = find_orphaned_snapshot(&sl);
 487         if (victim)
 488                 goto remove;
 489         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 490         victim = find_oldest_removable_snapshot(&sl);
 491         if (victim)
 492                 goto remove;
 493         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 494         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 495         goto out;
 496 remove:
 497         ret = pre_remove_hook(victim, why);
 498 out:
 499         free_snapshot_list(&sl);
 500         return ret;
 501 }
 502
 503 static int post_create_hook(void)
 504 {
 505         int ret, fds[3] = {0, 0, 0};
 506         char *cmd;
 507
 508         if (!conf.post_create_hook_given) {
 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_pid = 0;
 738         create_process_stopped = 0;
 739         return ret;
 740 }
 741
 742 static int handle_pre_create_hook_exit(int status)
 743 {
 744         int es, ret;
 745         static int warn_count;
 746
 747         if (!WIFEXITED(status)) {
 748                 snapshot_creation_status = HS_READY;
 749                 ret = -E_INVOLUNTARY_EXIT;
 750                 goto out;
 751         }
 752         es = WEXITSTATUS(status);
 753         if (es) {
 754                 if (!warn_count--) {
 755                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 756                                 conf.pre_create_hook_arg, es);
 757                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 758                         warn_count = 60; /* warn only once per hour */
 759                 }
 760                 next_snapshot_time = get_current_time() + 60;
 761                 snapshot_creation_status = HS_READY;
 762                 ret = 0;
 763                 goto out;
 764         }
 765         warn_count = 0;
 766         snapshot_creation_status = HS_PRE_SUCCESS;
 767         ret = 1;
 768 out:
 769         create_pid = 0;
 770         return ret;
 771 }
 772
 773 static int handle_sigchld(void)
 774 {
 775         pid_t pid;
 776         int status, ret = reap_child(&pid, &status);
 777
 778         if (ret <= 0)
 779                 return ret;
 780
 781         if (pid == create_pid) {
 782                 switch (snapshot_creation_status) {
 783                 case HS_PRE_RUNNING:
 784                         return handle_pre_create_hook_exit(status);
 785                 case HS_RUNNING:
 786                         return handle_rsync_exit(status);
 787                 case HS_POST_RUNNING:
 788                         snapshot_creation_status = HS_READY;
 789                         return 1;
 790                 default:
 791                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 792                                 snapshot_creation_status);
 793                         return -E_BUG;
 794                 }
 795         }
 796         if (pid == remove_pid) {
 797                 ret = handle_remove_exit(status);
 798                 if (ret < 0)
 799                         return ret;
 800                 return ret;
 801         }
 802         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 803         return -E_BUG;
 804 }
 805
 806 static int check_config(void)
 807 {
 808         if (conf.unit_interval_arg <= 0) {
 809                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 810                 return -E_INVALID_NUMBER;
 811         }
 812         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 813         if (conf.num_intervals_arg <= 0) {
 814                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 815                 return -E_INVALID_NUMBER;
 816         }
 817         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 818         return 1;
 819 }
 820
 821 static int parse_config_file(int override)
 822 {
 823         int ret;
 824         char *config_file;
 825         struct stat statbuf;
 826         char *old_logfile_arg = NULL;
 827         int old_daemon_given = 0;
 828
 829         if (conf.config_file_given)
 830                 config_file = dss_strdup(conf.config_file_arg);
 831         else {
 832                 char *home = get_homedir();
 833                 config_file = make_message("%s/.dssrc", home);
 834                 free(home);
 835         }
 836         if (override) { /* SIGHUP */
 837                 if (conf.logfile_given)
 838                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 839                 old_daemon_given = conf.daemon_given;
 840         }
 841
 842         ret = stat(config_file, &statbuf);
 843         if (ret && conf.config_file_given) {
 844                 ret = -ERRNO_TO_DSS_ERROR(errno);
 845                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 846                 goto out;
 847         }
 848         if (!ret) {
 849                 struct cmdline_parser_params params = {
 850                         .override = override,
 851                         .initialize = 0,
 852                         .check_required = 1,
 853                         .check_ambiguity = 0,
 854                         .print_errors = 1
 855                 };
 856                 if (override) { /* invalidate all rsync options */
 857                         int i;
 858
 859                         for (i = 0; i < conf.rsync_option_given; i++) {
 860                                 free(conf.rsync_option_arg[i]);
 861                                 conf.rsync_option_arg[i] = NULL;
 862                         }
 863                         conf.rsync_option_given = 0;
 864                 }
 865                 cmdline_parser_config_file(config_file, &conf, &params);
 866         }
 867         ret = check_config();
 868         if (ret < 0)
 869                 goto out;
 870         if (override) {
 871                 /* don't change daemon mode on SIGHUP */
 872                 conf.daemon_given = old_daemon_given;
 873                 close_log(logfile);
 874                 logfile = NULL;
 875                 if (conf.logfile_given)
 876                         free(old_logfile_arg);
 877                 else if (conf.daemon_given) { /* re-use old logfile */
 878                         conf.logfile_arg = old_logfile_arg;
 879                         conf.logfile_given = 1;
 880                 }
 881         }
 882         if (conf.logfile_given) {
 883                 logfile = open_log(conf.logfile_arg);
 884                 log_welcome(conf.loglevel_arg);
 885         }
 886         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 887 //      cmdline_parser_dump(logfile? logfile : stdout, &conf);
 888 out:
 889         free(config_file);
 890         if (ret < 0)
 891                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 892         return ret;
 893 }
 894
 895 static int change_to_dest_dir(void)
 896 {
 897         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 898         return dss_chdir(conf.dest_dir_arg);
 899 }
 900
 901 static int handle_sighup(void)
 902 {
 903         int ret;
 904
 905         DSS_NOTICE_LOG("SIGHUP\n");
 906         ret = parse_config_file(1);
 907         if (ret < 0)
 908                 return ret;
 909         invalidate_next_snapshot_time();
 910         return change_to_dest_dir();
 911 }
 912
 913 static int handle_signal(void)
 914 {
 915         int sig, ret = next_signal();
 916
 917         if (ret <= 0)
 918                 goto out;
 919         sig = ret;
 920         switch (sig) {
 921         case SIGINT:
 922         case SIGTERM:
 923                 restart_create_process();
 924                 kill_process(create_pid);
 925                 kill_process(remove_pid);
 926                 ret = -E_SIGNAL;
 927                 break;
 928         case SIGHUP:
 929                 ret = handle_sighup();
 930                 break;
 931         case SIGCHLD:
 932                 ret = handle_sigchld();
 933                 break;
 934         }
 935 out:
 936         if (ret < 0)
 937                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 938         return ret;
 939 }
 940
 941 /*
 942  * We can not use rsync locally if the local user is different from the remote
 943  * user or if the src dir is not on the local host (or both).
 944  */
 945 static int use_rsync_locally(char *logname)
 946 {
 947         char *h = conf.remote_host_arg;
 948
 949         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 950                 return 0;
 951         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 952                 return 0;
 953         return 1;
 954 }
 955
 956 static void create_rsync_argv(char ***argv, int64_t *num)
 957 {
 958         char *logname;
 959         int i = 0, j;
 960         struct snapshot_list sl;
 961
 962         dss_get_snapshot_list(&sl);
 963         assert(!name_of_reference_snapshot);
 964         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
 965         free_snapshot_list(&sl);
 966
 967         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 968         (*argv)[i++] = dss_strdup("rsync");
 969         (*argv)[i++] = dss_strdup("-aq");
 970         (*argv)[i++] = dss_strdup("--delete");
 971         for (j = 0; j < conf.rsync_option_given; j++)
 972                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
 973         if (name_of_reference_snapshot) {
 974                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
 975                 (*argv)[i++] = make_message("--link-dest=../%s",
 976                         name_of_reference_snapshot);
 977         } else
 978                 DSS_INFO_LOG("no suitable reference snapshot found\n");
 979         logname = dss_logname();
 980         if (use_rsync_locally(logname))
 981                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
 982         else
 983                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
 984                         conf.remote_user_arg : logname,
 985                         conf.remote_host_arg, conf.source_dir_arg);
 986         free(logname);
 987         *num = get_current_time();
 988         (*argv)[i++] = incomplete_name(*num);
 989         (*argv)[i++] = NULL;
 990         for (j = 0; j < i; j++)
 991                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 992 }
 993
 994 static void free_rsync_argv(char **argv)
 995 {
 996         int i;
 997
 998         if (!argv)
 999                 return;
1000         for (i = 0; argv[i]; i++)
1001                 free(argv[i]);
1002         free(argv);
1003 }
1004
1005 static int create_snapshot(char **argv)
1006 {
1007         int ret, fds[3] = {0, 0, 0};
1008         char *name;
1009
1010         name = incomplete_name(current_snapshot_creation_time);
1011         DSS_NOTICE_LOG("creating new snapshot %s\n", name);
1012         free(name);
1013         ret = dss_exec(&create_pid, argv[0], argv, fds);
1014         if (ret < 0)
1015                 return ret;
1016         snapshot_creation_status = HS_RUNNING;
1017         return ret;
1018 }
1019
1020 static int select_loop(void)
1021 {
1022         int ret;
1023         /* check every 60 seconds for free disk space */
1024         struct timeval tv;
1025         char **rsync_argv = NULL;
1026
1027         for (;;) {
1028                 fd_set rfds;
1029                 struct timeval *tvp;
1030
1031                 if (remove_pid)
1032                         tvp = NULL; /* sleep until rm hook/process dies */
1033                 else { /* sleep one minute */
1034                         tv.tv_sec = 60;
1035                         tv.tv_usec = 0;
1036                         tvp = &tv;
1037                 }
1038                 FD_ZERO(&rfds);
1039                 FD_SET(signal_pipe, &rfds);
1040                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1041                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1042                 if (ret < 0)
1043                         goto out;
1044                 if (FD_ISSET(signal_pipe, &rfds)) {
1045                         ret = handle_signal();
1046                         if (ret < 0)
1047                                 goto out;
1048                 }
1049                 if (remove_pid)
1050                         continue;
1051                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1052                         ret = exec_rm();
1053                         if (ret < 0)
1054                                 goto out;
1055                         continue;
1056                 }
1057                 if (snapshot_removal_status == HS_SUCCESS) {
1058                         ret = post_remove_hook();
1059                         if (ret < 0)
1060                                 goto out;
1061                         continue;
1062                 }
1063                 ret = try_to_free_disk_space();
1064                 if (ret < 0)
1065                         goto out;
1066                 if (snapshot_removal_status != HS_READY) {
1067                         stop_create_process();
1068                         continue;
1069                 }
1070                 restart_create_process();
1071                 switch (snapshot_creation_status) {
1072                 case HS_READY:
1073                         if (!next_snapshot_is_due())
1074                                 continue;
1075                         ret = pre_create_hook();
1076                         if (ret < 0)
1077                                 goto out;
1078                         continue;
1079                 case HS_PRE_RUNNING:
1080                 case HS_RUNNING:
1081                 case HS_POST_RUNNING:
1082                         continue;
1083                 case HS_PRE_SUCCESS:
1084                         free_rsync_argv(rsync_argv);
1085                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1086                         /* fall through */
1087                 case HS_NEEDS_RESTART:
1088                         ret = create_snapshot(rsync_argv);
1089                         if (ret < 0)
1090                                 goto out;
1091                         continue;
1092                 case HS_SUCCESS:
1093                         ret = post_create_hook();
1094                         if (ret < 0)
1095                                 goto out;
1096                         continue;
1097                 }
1098         }
1099 out:
1100         return ret;
1101 }
1102
1103 static void exit_hook(int exit_code)
1104 {
1105         int fds[3] = {0, 0, 0};
1106         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1107         pid_t pid;
1108
1109         if (!conf.exit_hook_given)
1110                 return;
1111         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1112         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1113 }
1114
1115 static int com_run(void)
1116 {
1117         int ret;
1118
1119         if (conf.dry_run_given) {
1120                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1121                 return -E_SYNTAX;
1122         }
1123         ret = install_sighandler(SIGHUP);
1124         if (ret < 0)
1125                 return ret;
1126         ret = select_loop();
1127         if (ret >= 0) /* impossible */
1128                 ret = -E_BUG;
1129         exit_hook(ret);
1130         return ret;
1131 }
1132
1133 static int com_prune(void)
1134 {
1135         int ret;
1136         struct snapshot_list sl;
1137         struct snapshot *victim;
1138         struct disk_space ds;
1139         const char *why;
1140
1141         ret = get_disk_space(".", &ds);
1142         if (ret < 0)
1143                 return ret;
1144         log_disk_space(&ds);
1145         dss_get_snapshot_list(&sl);
1146         why = "outdated";
1147         victim = find_outdated_snapshot(&sl);
1148         if (victim)
1149                 goto rm;
1150         why = "redundant";
1151         victim = find_redundant_snapshot(&sl);
1152         if (victim)
1153                 goto rm;
1154         ret = 0;
1155         goto out;
1156 rm:
1157         if (conf.dry_run_given) {
1158                 dss_msg("%s snapshot %s (interval = %i)\n",
1159                         why, victim->name, victim->interval);
1160                 ret = 0;
1161                 goto out;
1162         }
1163         ret = pre_remove_hook(victim, why);
1164         if (ret < 0)
1165                 goto out;
1166         if (snapshot_removal_status == HS_PRE_RUNNING) {
1167                 ret = wait_for_remove_process();
1168                 if (ret < 0)
1169                         goto out;
1170                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1171                         goto out;
1172         }
1173         ret = exec_rm();
1174         if (ret < 0)
1175                 goto out;
1176         ret = wait_for_remove_process();
1177         if (ret < 0)
1178                 goto out;
1179         if (snapshot_removal_status != HS_SUCCESS)
1180                 goto out;
1181         ret = post_remove_hook();
1182         if (ret < 0)
1183                 goto out;
1184         if (snapshot_removal_status != HS_POST_RUNNING)
1185                 goto out;
1186         ret = wait_for_remove_process();
1187         if (ret < 0)
1188                 goto out;
1189         ret = 1;
1190 out:
1191         free_snapshot_list(&sl);
1192         return ret;
1193 }
1194
1195 static int com_create(void)
1196 {
1197         int ret, status;
1198         char **rsync_argv;
1199
1200         if (conf.dry_run_given) {
1201                 int i;
1202                 char *msg = NULL;
1203                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1204                 for (i = 0; rsync_argv[i]; i++) {
1205                         char *tmp = msg;
1206                         msg = make_message("%s%s%s", tmp? tmp : "",
1207                                 tmp? " " : "", rsync_argv[i]);
1208                         free(tmp);
1209                 }
1210                 free_rsync_argv(rsync_argv);
1211                 dss_msg("%s\n", msg);
1212                 free(msg);
1213                 return 1;
1214         }
1215         ret = pre_create_hook();
1216         if (ret < 0)
1217                 return ret;
1218         if (create_pid) {
1219                 ret = wait_for_process(create_pid, &status);
1220                 if (ret < 0)
1221                         return ret;
1222                 ret = handle_pre_create_hook_exit(status);
1223                 if (ret <= 0) /* error, or pre-create failed */
1224                         return ret;
1225         }
1226         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1227         ret = create_snapshot(rsync_argv);
1228         if (ret < 0)
1229                 goto out;
1230         ret = wait_for_process(create_pid, &status);
1231         if (ret < 0)
1232                 goto out;
1233         ret = handle_rsync_exit(status);
1234         if (ret < 0)
1235                 goto out;
1236         post_create_hook();
1237         if (create_pid)
1238                 ret = wait_for_process(create_pid, &status);
1239 out:
1240         free_rsync_argv(rsync_argv);
1241         return ret;
1242 }
1243
1244 static int com_ls(void)
1245 {
1246         int i;
1247         struct snapshot_list sl;
1248         struct snapshot *s;
1249
1250         dss_get_snapshot_list(&sl);
1251         FOR_EACH_SNAPSHOT(s, i, &sl) {
1252                 int64_t d = 0;
1253                 if (s->flags & SS_COMPLETE)
1254                         d = (s->completion_time - s->creation_time) / 60;
1255                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1256         };
1257         free_snapshot_list(&sl);
1258         return 1;
1259 }
1260
1261 static int setup_signal_handling(void)
1262 {
1263         int ret;
1264
1265         DSS_INFO_LOG("setting up signal handlers\n");
1266         signal_pipe = signal_init(); /* always successful */
1267         ret = install_sighandler(SIGINT);
1268         if (ret < 0)
1269                 return ret;
1270         ret = install_sighandler(SIGTERM);
1271         if (ret < 0)
1272                 return ret;
1273         return install_sighandler(SIGCHLD);
1274 }
1275
1276 /**
1277  * The main function of dss.
1278  *
1279  * \param argc Usual argument count.
1280  * \param argv Usual argument vector.
1281  */
1282 int main(int argc, char **argv)
1283 {
1284         int ret;
1285         struct cmdline_parser_params params = {
1286                 .override = 0,
1287                 .initialize = 1,
1288                 .check_required = 0,
1289                 .check_ambiguity = 0,
1290                 .print_errors = 1
1291         };
1292
1293         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1294         ret = parse_config_file(0);
1295         if (ret < 0)
1296                 goto out;
1297         if (conf.daemon_given)
1298                 daemon_init();
1299         ret = change_to_dest_dir();
1300         if (ret < 0)
1301                 goto out;
1302         ret = setup_signal_handling();
1303         if (ret < 0)
1304                 goto out;
1305         ret = call_command_handler();
1306 out:
1307         if (ret < 0)
1308                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1309         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1310 }