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         dss_get_snapshot_list(&sl);
 165         FOR_EACH_SNAPSHOT(s, i, &sl) {
 166                 if (!(s->flags & SS_COMPLETE))
 167                         continue;
 168                 num_complete_snapshots++;
 169                 x += s->completion_time - s->creation_time;
 170         }
 171         assert(x >= 0);
 172
 173         ret = now;
 174         if (num_complete_snapshots == 0)
 175                 goto out;
 176         x /= num_complete_snapshots; /* avg time to create one snapshot */
 177         if (unit_interval < x * wanted) /* oops, no sleep at all */
 178                 goto out;
 179         ret = s->completion_time + unit_interval / wanted - x;
 180 out:
 181         free_snapshot_list(&sl);
 182         return ret;
 183 }
 184
 185 static inline void invalidate_next_snapshot_time(void)
 186 {
 187         next_snapshot_time = 0;
 188 }
 189
 190 static inline int next_snapshot_time_is_valid(void)
 191 {
 192         return next_snapshot_time != 0;
 193 }
 194
 195 static int next_snapshot_is_due(void)
 196 {
 197         int64_t now = get_current_time();
 198
 199         if (!next_snapshot_time_is_valid())
 200                 next_snapshot_time = compute_next_snapshot_time();
 201         if (next_snapshot_time <= now) {
 202                 DSS_DEBUG_LOG("next snapshot: now\n");
 203                 return 1;
 204         }
 205         DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
 206                 next_snapshot_time - now);
 207         return 0;
 208 }
 209
 210 static int pre_create_hook(void)
 211 {
 212         int ret, fds[3] = {0, 0, 0};
 213
 214         assert(snapshot_creation_status == HS_READY);
 215         /* make sure that the next snapshot time will be recomputed */
 216         invalidate_next_snapshot_time();
 217         if (!conf.pre_create_hook_given) {
 218                 snapshot_creation_status = HS_PRE_SUCCESS;
 219                 return 0;
 220         }
 221         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 222         ret = dss_exec_cmdline_pid(&create_pid,
 223                 conf.pre_create_hook_arg, fds);
 224         if (ret < 0)
 225                 return ret;
 226         snapshot_creation_status = HS_PRE_RUNNING;
 227         return ret;
 228 }
 229
 230 static int pre_remove_hook(struct snapshot *s, const char *why)
 231 {
 232         int ret, fds[3] = {0, 0, 0};
 233         char *cmd;
 234
 235         if (!s)
 236                 return 0;
 237         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 238         assert(snapshot_removal_status == HS_READY);
 239         assert(remove_pid == 0);
 240         assert(!snapshot_currently_being_removed);
 241
 242         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 243         *snapshot_currently_being_removed = *s;
 244         snapshot_currently_being_removed->name = dss_strdup(s->name);
 245
 246         if (!conf.pre_remove_hook_given) {
 247                 snapshot_removal_status = HS_PRE_SUCCESS;
 248                 return 0;
 249         }
 250         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 251                 conf.dest_dir_arg, s->name);
 252         DSS_DEBUG_LOG("executing %s\n", cmd);
 253         ret = dss_exec_cmdline_pid(&remove_pid,
 254                 conf.pre_remove_hook_arg, fds);
 255         free(cmd);
 256         if (ret < 0)
 257                 return ret;
 258         snapshot_removal_status = HS_PRE_RUNNING;
 259         return ret;
 260 }
 261
 262 static int exec_rm(void)
 263 {
 264         struct snapshot *s = snapshot_currently_being_removed;
 265         int fds[3] = {0, 0, 0};
 266         char *new_name = being_deleted_name(s);
 267         char *argv[] = {"rm", "-rf", new_name, NULL};
 268         int ret;
 269
 270         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 271         assert(remove_pid == 0);
 272
 273         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 274         ret = dss_rename(s->name, new_name);
 275         if (ret < 0)
 276                 goto out;
 277         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 278         if (ret < 0)
 279                 goto out;
 280         snapshot_removal_status = HS_RUNNING;
 281 out:
 282         free(new_name);
 283         return ret;
 284 }
 285
 286 static int snapshot_is_being_created(struct snapshot *s)
 287 {
 288         return s->creation_time == current_snapshot_creation_time;
 289 }
 290
 291 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 292 {
 293         struct snapshot *s;
 294         int i;
 295
 296         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 297         FOR_EACH_SNAPSHOT(s, i, sl) {
 298                 if (snapshot_is_being_created(s))
 299                         continue;
 300                 /*
 301                  * We know that no rm is currently running, so if s is marked
 302                  * as being deleted, a previously started rm must have failed.
 303                  */
 304                 if (s->flags & SS_BEING_DELETED)
 305                         return s;
 306
 307                 if (s->flags & SS_COMPLETE) /* good snapshot */
 308                         continue;
 309                 /*
 310                  * This snapshot is incomplete and it is not the snapshot
 311                  * currently being created. However, we must not remove it if
 312                  * rsync is about to be restarted. As only the newest snapshot
 313                  * can be restarted, this snapshot is orphaned if it is not the
 314                  * newest snapshot or if we are not about to restart rsync.
 315                  */
 316                 if (get_newest_snapshot(sl) != s)
 317                         return s;
 318                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 319                         return s;
 320         }
 321         /* no orphaned snapshots */
 322         return NULL;
 323 }
 324
 325 static int is_reference_snapshot(struct snapshot *s)
 326 {
 327         if (!name_of_reference_snapshot)
 328                 return 0;
 329         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 330 }
 331
 332 /*
 333  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 334  */
 335 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 336 {
 337         int i, interval;
 338         struct snapshot *s;
 339         unsigned missing = 0;
 340
 341         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 342         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 343                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 344                 unsigned num = sl->interval_count[interval];
 345                 struct snapshot *victim = NULL, *prev = NULL;
 346                 int64_t score = LONG_MAX;
 347
 348                 if (keep >= num)
 349                         missing += keep - num;
 350 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 351 //                      interval, keep, num, missing);
 352                 if (keep + missing >= num)
 353                         continue;
 354                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 355                 FOR_EACH_SNAPSHOT(s, i, sl) {
 356                         int64_t this_score;
 357
 358                         if (snapshot_is_being_created(s))
 359                                 continue;
 360                         if (is_reference_snapshot(s))
 361                                 continue;
 362                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 363                         if (s->interval > interval) {
 364                                 prev = s;
 365                                 continue;
 366                         }
 367                         if (s->interval < interval)
 368                                 break;
 369                         if (!victim) {
 370                                 victim = s;
 371                                 prev = s;
 372                                 continue;
 373                         }
 374                         assert(prev);
 375                         /* check if s is a better victim */
 376                         this_score = s->creation_time - prev->creation_time;
 377                         assert(this_score >= 0);
 378                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 379                         if (this_score < score) {
 380                                 score = this_score;
 381                                 victim = s;
 382                         }
 383                         prev = s;
 384                 }
 385                 assert(victim);
 386                 return victim;
 387         }
 388         return NULL;
 389 }
 390
 391 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 392 {
 393         int i;
 394         struct snapshot *s;
 395
 396         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 397                 conf.num_intervals_arg);
 398         FOR_EACH_SNAPSHOT(s, i, sl) {
 399                 if (snapshot_is_being_created(s))
 400                         continue;
 401                 if (is_reference_snapshot(s))
 402                         continue;
 403                 if (s->interval < conf.num_intervals_arg)
 404                         continue;
 405                 return s;
 406         }
 407         return NULL;
 408 }
 409
 410 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 411 {
 412         int i;
 413         struct snapshot *s;
 414         FOR_EACH_SNAPSHOT(s, i, sl) {
 415                 if (snapshot_is_being_created(s))
 416                         continue;
 417                 if (is_reference_snapshot(s))
 418                         continue;
 419                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 420                 return s;
 421         }
 422         return NULL;
 423 }
 424
 425 static int rename_incomplete_snapshot(int64_t start)
 426 {
 427         char *old_name;
 428         int ret;
 429
 430         free(path_to_last_complete_snapshot);
 431         ret = complete_name(start, get_current_time(),
 432                 &path_to_last_complete_snapshot);
 433         if (ret < 0)
 434                 return ret;
 435         old_name = incomplete_name(start);
 436         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 437         if (ret >= 0)
 438                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 439                         path_to_last_complete_snapshot);
 440         free(old_name);
 441         return ret;
 442 }
 443
 444 static int try_to_free_disk_space(void)
 445 {
 446         int ret;
 447         struct snapshot_list sl;
 448         struct snapshot *victim;
 449         struct timeval now;
 450         const char *why;
 451         int low_disk_space;
 452
 453         ret = disk_space_low();
 454         if (ret < 0)
 455                 return ret;
 456         low_disk_space = ret;
 457         gettimeofday(&now, NULL);
 458         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 459                 return 0;
 460         if (!low_disk_space) {
 461                 if (conf.keep_redundant_given)
 462                         return 0;
 463                 if (snapshot_creation_status != HS_READY)
 464                         return 0;
 465                 if (next_snapshot_is_due())
 466                         return 0;
 467         }
 468         dss_get_snapshot_list(&sl);
 469         ret = 0;
 470         if (!low_disk_space && sl.num_snapshots <= 1)
 471                 goto out;
 472         why = "outdated";
 473         victim = find_outdated_snapshot(&sl);
 474         if (victim)
 475                 goto remove;
 476         why = "redundant";
 477         victim = find_redundant_snapshot(&sl);
 478         if (victim)
 479                 goto remove;
 480         /* try harder only if disk space is low */
 481         if (!low_disk_space)
 482                 goto out;
 483         why = "orphaned";
 484         victim = find_orphaned_snapshot(&sl);
 485         if (victim)
 486                 goto remove;
 487         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 488         victim = find_oldest_removable_snapshot(&sl);
 489         if (victim)
 490                 goto remove;
 491         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 492         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 493         goto out;
 494 remove:
 495         ret = pre_remove_hook(victim, why);
 496 out:
 497         free_snapshot_list(&sl);
 498         return ret;
 499 }
 500
 501 static int post_create_hook(void)
 502 {
 503         int ret, fds[3] = {0, 0, 0};
 504         char *cmd;
 505
 506         if (!conf.post_create_hook_given) {
 507                 create_pid = 0;
 508                 snapshot_creation_status = HS_READY;
 509                 return 0;
 510         }
 511         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 512                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 513         DSS_NOTICE_LOG("executing %s\n", cmd);
 514         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 515         free(cmd);
 516         if (ret < 0)
 517                 return ret;
 518         snapshot_creation_status = HS_POST_RUNNING;
 519         return ret;
 520 }
 521
 522 static int post_remove_hook(void)
 523 {
 524         int ret, fds[3] = {0, 0, 0};
 525         char *cmd;
 526         struct snapshot *s = snapshot_currently_being_removed;
 527
 528         assert(s);
 529
 530         if (!conf.post_remove_hook_given) {
 531                 snapshot_removal_status = HS_READY;
 532                 return 0;
 533         }
 534         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 535                 conf.dest_dir_arg, s->name);
 536         DSS_NOTICE_LOG("executing %s\n", cmd);
 537         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 538         free(cmd);
 539         if (ret < 0)
 540                 return ret;
 541         snapshot_removal_status = HS_POST_RUNNING;
 542         return ret;
 543 }
 544
 545 static void kill_process(pid_t pid)
 546 {
 547         if (!pid)
 548                 return;
 549         DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 550         kill(pid, SIGTERM);
 551 }
 552
 553 static void stop_create_process(void)
 554 {
 555         if (!create_pid || create_process_stopped)
 556                 return;
 557         kill(SIGSTOP, create_pid);
 558         create_process_stopped = 1;
 559 }
 560
 561 static void restart_create_process(void)
 562 {
 563         if (!create_pid || !create_process_stopped)
 564                 return;
 565         kill (SIGCONT, create_pid);
 566         create_process_stopped = 0;
 567 }
 568
 569 /**
 570  * Print a log message about the exit status of a child.
 571  */
 572 static void log_termination_msg(pid_t pid, int status)
 573 {
 574         if (WIFEXITED(status))
 575                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 576                         WEXITSTATUS(status));
 577         else if (WIFSIGNALED(status))
 578                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 579                         WTERMSIG(status));
 580         else
 581                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 582 }
 583
 584 static int wait_for_process(pid_t pid, int *status)
 585 {
 586         int ret;
 587
 588         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 589         for (;;) {
 590                 fd_set rfds;
 591
 592                 FD_ZERO(&rfds);
 593                 FD_SET(signal_pipe, &rfds);
 594                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 595                 if (ret < 0)
 596                         break;
 597                 ret = next_signal();
 598                 if (!ret)
 599                         continue;
 600                 if (ret == SIGCHLD) {
 601                         ret = waitpid(pid, status, 0);
 602                         if (ret >= 0)
 603                                 break;
 604                         if (errno != EINTR) { /* error */
 605                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 606                                 break;
 607                         }
 608                 }
 609                 /* SIGINT or SIGTERM */
 610                 DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
 611                 kill(pid, SIGTERM);
 612         }
 613         if (ret < 0)
 614                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 615         else
 616                 log_termination_msg(pid, *status);
 617         return ret;
 618 }
 619
 620 static void handle_pre_remove_exit(int status)
 621 {
 622         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 623                 snapshot_removal_status = HS_READY;
 624                 gettimeofday(&next_removal_check, NULL);
 625                 next_removal_check.tv_sec += 60;
 626                 return;
 627         }
 628         snapshot_removal_status = HS_PRE_SUCCESS;
 629 }
 630
 631 static int handle_rm_exit(int status)
 632 {
 633         if (!WIFEXITED(status)) {
 634                 snapshot_removal_status = HS_READY;
 635                 return -E_INVOLUNTARY_EXIT;
 636         }
 637         if (WEXITSTATUS(status)) {
 638                 snapshot_removal_status = HS_READY;
 639                 return -E_BAD_EXIT_CODE;
 640         }
 641         if (conf.post_remove_hook_given)
 642                 snapshot_removal_status = HS_SUCCESS;
 643         else
 644                 snapshot_removal_status = HS_READY;
 645         return 1;
 646 }
 647
 648 static void handle_post_remove_exit(void)
 649 {
 650         snapshot_removal_status = HS_READY;
 651 }
 652
 653 static int handle_remove_exit(int status)
 654 {
 655         int ret;
 656         struct snapshot *s = snapshot_currently_being_removed;
 657
 658         assert(s);
 659         switch (snapshot_removal_status) {
 660         case HS_PRE_RUNNING:
 661                 handle_pre_remove_exit(status);
 662                 ret = 1;
 663                 break;
 664         case HS_RUNNING:
 665                 ret = handle_rm_exit(status);
 666                 break;
 667         case HS_POST_RUNNING:
 668                 handle_post_remove_exit();
 669                 ret = 1;
 670                 break;
 671         default:
 672                 ret = -E_BUG;
 673         }
 674         if (snapshot_removal_status == HS_READY) {
 675                 free(s->name);
 676                 free(s);
 677                 snapshot_currently_being_removed = NULL;
 678         }
 679         remove_pid = 0;
 680         return ret;
 681 }
 682
 683 static int wait_for_remove_process(void)
 684 {
 685         int status, ret;
 686
 687         assert(remove_pid);
 688         assert(
 689                 snapshot_removal_status == HS_PRE_RUNNING ||
 690                 snapshot_removal_status == HS_RUNNING ||
 691                 snapshot_removal_status == HS_POST_RUNNING
 692         );
 693         ret = wait_for_process(remove_pid, &status);
 694         if (ret < 0)
 695                 return ret;
 696         return handle_remove_exit(status);
 697 }
 698
 699 static int handle_rsync_exit(int status)
 700 {
 701         int es, ret;
 702
 703         if (!WIFEXITED(status)) {
 704                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 705                 ret = -E_INVOLUNTARY_EXIT;
 706                 snapshot_creation_status = HS_READY;
 707                 goto out;
 708         }
 709         es = WEXITSTATUS(status);
 710         /*
 711          * Restart rsync on non-fatal errors:
 712          * 12: Error in rsync protocol data stream
 713          * 13: Errors with program diagnostics
 714          */
 715         if (es == 12 || es == 13) {
 716                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 717                         (int)create_pid, es);
 718                 snapshot_creation_status = HS_NEEDS_RESTART;
 719                 next_snapshot_time = get_current_time() + 60;
 720                 ret = 1;
 721                 goto out;
 722         }
 723         if (es != 0 && es != 23 && es != 24) {
 724                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 725                 ret = -E_BAD_EXIT_CODE;
 726                 snapshot_creation_status = HS_READY;
 727                 goto out;
 728         }
 729         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 730         if (ret < 0)
 731                 goto out;
 732         snapshot_creation_status = HS_SUCCESS;
 733         free(name_of_reference_snapshot);
 734         name_of_reference_snapshot = NULL;
 735 out:
 736         create_process_stopped = 0;
 737         return ret;
 738 }
 739
 740 static int handle_pre_create_hook_exit(int status)
 741 {
 742         int es, ret;
 743         static int warn_count;
 744
 745         if (!WIFEXITED(status)) {
 746                 snapshot_creation_status = HS_READY;
 747                 ret = -E_INVOLUNTARY_EXIT;
 748                 goto out;
 749         }
 750         es = WEXITSTATUS(status);
 751         if (es) {
 752                 if (!warn_count--) {
 753                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 754                                 conf.pre_create_hook_arg, es);
 755                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 756                         warn_count = 60; /* warn only once per hour */
 757                 }
 758                 next_snapshot_time = get_current_time() + 60;
 759                 snapshot_creation_status = HS_READY;
 760                 ret = 0;
 761                 goto out;
 762         }
 763         warn_count = 0;
 764         snapshot_creation_status = HS_PRE_SUCCESS;
 765         ret = 1;
 766 out:
 767         return ret;
 768 }
 769
 770 static int handle_sigchld(void)
 771 {
 772         pid_t pid;
 773         int status, ret = reap_child(&pid, &status);
 774
 775         if (ret <= 0)
 776                 return ret;
 777
 778         if (pid == create_pid) {
 779                 switch (snapshot_creation_status) {
 780                 case HS_PRE_RUNNING:
 781                         return handle_pre_create_hook_exit(status);
 782                 case HS_RUNNING:
 783                         return handle_rsync_exit(status);
 784                 case HS_POST_RUNNING:
 785                         snapshot_creation_status = HS_READY;
 786                         return 1;
 787                 default:
 788                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 789                                 snapshot_creation_status);
 790                         return -E_BUG;
 791                 }
 792                 create_pid = 0;
 793         }
 794         if (pid == remove_pid) {
 795                 ret = handle_remove_exit(status);
 796                 if (ret < 0)
 797                         return ret;
 798                 return ret;
 799         }
 800         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 801         return -E_BUG;
 802 }
 803
 804 static int check_config(void)
 805 {
 806         if (conf.unit_interval_arg <= 0) {
 807                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 808                 return -E_INVALID_NUMBER;
 809         }
 810         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 811         if (conf.num_intervals_arg <= 0) {
 812                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 813                 return -E_INVALID_NUMBER;
 814         }
 815         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 816         return 1;
 817 }
 818
 819 /*
 820  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 821  * file was read successfully.
 822  */
 823 static int parse_config_file(int override)
 824 {
 825         int ret, config_file_exists;
 826         char *config_file;
 827         struct stat statbuf;
 828         char *old_logfile_arg = NULL;
 829         int old_daemon_given = 0;
 830
 831         if (conf.config_file_given)
 832                 config_file = dss_strdup(conf.config_file_arg);
 833         else {
 834                 char *home = get_homedir();
 835                 config_file = make_message("%s/.dssrc", home);
 836                 free(home);
 837         }
 838         if (override) { /* SIGHUP */
 839                 if (conf.logfile_given)
 840                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 841                 old_daemon_given = conf.daemon_given;
 842         }
 843
 844         config_file_exists = !stat(config_file, &statbuf);
 845         if (!config_file_exists && conf.config_file_given) {
 846                 ret = -ERRNO_TO_DSS_ERROR(errno);
 847                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 848                 goto out;
 849         }
 850         if (config_file_exists) {
 851                 struct cmdline_parser_params params = {
 852                         .override = override,
 853                         .initialize = 0,
 854                         .check_required = 1,
 855                         .check_ambiguity = 0,
 856                         .print_errors = 1
 857                 };
 858                 if (override) { /* invalidate all rsync options */
 859                         int i;
 860
 861                         for (i = 0; i < conf.rsync_option_given; i++) {
 862                                 free(conf.rsync_option_arg[i]);
 863                                 conf.rsync_option_arg[i] = NULL;
 864                         }
 865                         conf.rsync_option_given = 0;
 866                 }
 867                 cmdline_parser_config_file(config_file, &conf, &params);
 868         }
 869         ret = check_config();
 870         if (ret < 0)
 871                 goto out;
 872         if (override) {
 873                 /* don't change daemon mode on SIGHUP */
 874                 conf.daemon_given = old_daemon_given;
 875                 close_log(logfile);
 876                 logfile = NULL;
 877                 if (conf.logfile_given)
 878                         free(old_logfile_arg);
 879                 else if (conf.daemon_given) { /* re-use old logfile */
 880                         conf.logfile_arg = old_logfile_arg;
 881                         conf.logfile_given = 1;
 882                 }
 883         }
 884         if (conf.logfile_given) {
 885                 logfile = open_log(conf.logfile_arg);
 886                 log_welcome(conf.loglevel_arg);
 887         }
 888         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 889         ret = config_file_exists;
 890 out:
 891         free(config_file);
 892         if (ret < 0)
 893                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 894         return ret;
 895 }
 896
 897 static int change_to_dest_dir(void)
 898 {
 899         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 900         return dss_chdir(conf.dest_dir_arg);
 901 }
 902
 903 static void dump_dss_config(const char *msg)
 904 {
 905         if (conf.loglevel_arg > INFO)
 906                 return;
 907         DSS_INFO_LOG("%s\n", msg);
 908         cmdline_parser_dump(logfile? logfile : stderr, &conf);
 909 }
 910
 911 static int handle_sighup(void)
 912 {
 913         int ret;
 914
 915         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
 916         dump_dss_config("current config");
 917         ret = parse_config_file(1);
 918         if (ret < 0)
 919                 return ret;
 920         dump_dss_config("new config");
 921         invalidate_next_snapshot_time();
 922         return change_to_dest_dir();
 923 }
 924
 925 static int handle_signal(void)
 926 {
 927         int sig, ret = next_signal();
 928
 929         if (ret <= 0)
 930                 goto out;
 931         sig = ret;
 932         switch (sig) {
 933         case SIGINT:
 934         case SIGTERM:
 935                 restart_create_process();
 936                 kill_process(create_pid);
 937                 kill_process(remove_pid);
 938                 ret = -E_SIGNAL;
 939                 break;
 940         case SIGHUP:
 941                 ret = handle_sighup();
 942                 break;
 943         case SIGCHLD:
 944                 ret = handle_sigchld();
 945                 break;
 946         }
 947 out:
 948         if (ret < 0)
 949                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 950         return ret;
 951 }
 952
 953 /*
 954  * We can not use rsync locally if the local user is different from the remote
 955  * user or if the src dir is not on the local host (or both).
 956  */
 957 static int use_rsync_locally(char *logname)
 958 {
 959         char *h = conf.remote_host_arg;
 960
 961         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 962                 return 0;
 963         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 964                 return 0;
 965         return 1;
 966 }
 967
 968 static int rename_resume_snap(int64_t creation_time)
 969 {
 970         struct snapshot_list sl = {.num_snapshots = 0};
 971         struct snapshot *s;
 972         char *new_name = incomplete_name(creation_time);
 973         int ret;
 974
 975         ret = 0;
 976         if (conf.no_resume_given)
 977                 goto out;
 978         dss_get_snapshot_list(&sl);
 979         s = get_newest_snapshot(&sl);
 980         if (!s)
 981                 goto out;
 982         if ((s->flags & SS_COMPLETE) != 0) /* complete */
 983                 goto out;
 984         DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
 985         ret = dss_rename(s->name, new_name);
 986 out:
 987         if (ret >= 0)
 988                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
 989         free(new_name);
 990         free_snapshot_list(&sl);
 991         return ret;
 992 }
 993
 994 static void create_rsync_argv(char ***argv, int64_t *num)
 995 {
 996         char *logname;
 997         int i = 0, j;
 998         struct snapshot_list sl;
 999
1000         dss_get_snapshot_list(&sl);
1001         assert(!name_of_reference_snapshot);
1002         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1003         free_snapshot_list(&sl);
1004
1005         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1006         (*argv)[i++] = dss_strdup("rsync");
1007         (*argv)[i++] = dss_strdup("-aq");
1008         (*argv)[i++] = dss_strdup("--delete");
1009         for (j = 0; j < conf.rsync_option_given; j++)
1010                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1011         if (name_of_reference_snapshot) {
1012                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1013                 (*argv)[i++] = make_message("--link-dest=../%s",
1014                         name_of_reference_snapshot);
1015         } else
1016                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1017         logname = dss_logname();
1018         if (use_rsync_locally(logname))
1019                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1020         else
1021                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1022                         conf.remote_user_arg : logname,
1023                         conf.remote_host_arg, conf.source_dir_arg);
1024         free(logname);
1025         *num = get_current_time();
1026         (*argv)[i++] = incomplete_name(*num);
1027         (*argv)[i++] = NULL;
1028         for (j = 0; j < i; j++)
1029                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1030 }
1031
1032 static void free_rsync_argv(char **argv)
1033 {
1034         int i;
1035
1036         if (!argv)
1037                 return;
1038         for (i = 0; argv[i]; i++)
1039                 free(argv[i]);
1040         free(argv);
1041 }
1042
1043 static int create_snapshot(char **argv)
1044 {
1045         int ret, fds[3] = {0, 0, 0};
1046
1047         ret = rename_resume_snap(current_snapshot_creation_time);
1048         if (ret < 0)
1049                 return ret;
1050         ret = dss_exec(&create_pid, argv[0], argv, fds);
1051         if (ret < 0)
1052                 return ret;
1053         snapshot_creation_status = HS_RUNNING;
1054         return ret;
1055 }
1056
1057 static int select_loop(void)
1058 {
1059         int ret;
1060         /* check every 60 seconds for free disk space */
1061         struct timeval tv;
1062         char **rsync_argv = NULL;
1063
1064         for (;;) {
1065                 fd_set rfds;
1066                 struct timeval *tvp;
1067
1068                 if (remove_pid)
1069                         tvp = NULL; /* sleep until rm hook/process dies */
1070                 else { /* sleep one minute */
1071                         tv.tv_sec = 60;
1072                         tv.tv_usec = 0;
1073                         tvp = &tv;
1074                 }
1075                 FD_ZERO(&rfds);
1076                 FD_SET(signal_pipe, &rfds);
1077                 DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
1078                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1079                 if (ret < 0)
1080                         goto out;
1081                 if (FD_ISSET(signal_pipe, &rfds)) {
1082                         ret = handle_signal();
1083                         if (ret < 0)
1084                                 goto out;
1085                 }
1086                 if (remove_pid)
1087                         continue;
1088                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1089                         ret = exec_rm();
1090                         if (ret < 0)
1091                                 goto out;
1092                         continue;
1093                 }
1094                 if (snapshot_removal_status == HS_SUCCESS) {
1095                         ret = post_remove_hook();
1096                         if (ret < 0)
1097                                 goto out;
1098                         continue;
1099                 }
1100                 ret = try_to_free_disk_space();
1101                 if (ret < 0)
1102                         goto out;
1103                 if (snapshot_removal_status != HS_READY) {
1104                         stop_create_process();
1105                         continue;
1106                 }
1107                 restart_create_process();
1108                 switch (snapshot_creation_status) {
1109                 case HS_READY:
1110                         if (!next_snapshot_is_due())
1111                                 continue;
1112                         ret = pre_create_hook();
1113                         if (ret < 0)
1114                                 goto out;
1115                         continue;
1116                 case HS_PRE_RUNNING:
1117                 case HS_RUNNING:
1118                 case HS_POST_RUNNING:
1119                         continue;
1120                 case HS_PRE_SUCCESS:
1121                         free_rsync_argv(rsync_argv);
1122                         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1123                         /* fall through */
1124                 case HS_NEEDS_RESTART:
1125                         if (!next_snapshot_is_due())
1126                                 continue;
1127                         ret = create_snapshot(rsync_argv);
1128                         if (ret < 0)
1129                                 goto out;
1130                         continue;
1131                 case HS_SUCCESS:
1132                         ret = post_create_hook();
1133                         if (ret < 0)
1134                                 goto out;
1135                         continue;
1136                 }
1137         }
1138 out:
1139         return ret;
1140 }
1141
1142 static void exit_hook(int exit_code)
1143 {
1144         int fds[3] = {0, 0, 0};
1145         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1146         pid_t pid;
1147
1148         if (!conf.exit_hook_given)
1149                 return;
1150         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1151         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1152 }
1153
1154 static int com_run(void)
1155 {
1156         int ret;
1157
1158         if (conf.dry_run_given) {
1159                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1160                 return -E_SYNTAX;
1161         }
1162         ret = install_sighandler(SIGHUP);
1163         if (ret < 0)
1164                 return ret;
1165         ret = select_loop();
1166         if (ret >= 0) /* impossible */
1167                 ret = -E_BUG;
1168         exit_hook(ret);
1169         return ret;
1170 }
1171
1172 static int com_prune(void)
1173 {
1174         int ret;
1175         struct snapshot_list sl;
1176         struct snapshot *victim;
1177         struct disk_space ds;
1178         const char *why;
1179
1180         ret = get_disk_space(".", &ds);
1181         if (ret < 0)
1182                 return ret;
1183         log_disk_space(&ds);
1184         dss_get_snapshot_list(&sl);
1185         why = "outdated";
1186         victim = find_outdated_snapshot(&sl);
1187         if (victim)
1188                 goto rm;
1189         why = "redundant";
1190         victim = find_redundant_snapshot(&sl);
1191         if (victim)
1192                 goto rm;
1193         ret = 0;
1194         goto out;
1195 rm:
1196         if (conf.dry_run_given) {
1197                 dss_msg("%s snapshot %s (interval = %i)\n",
1198                         why, victim->name, victim->interval);
1199                 ret = 0;
1200                 goto out;
1201         }
1202         ret = pre_remove_hook(victim, why);
1203         if (ret < 0)
1204                 goto out;
1205         if (snapshot_removal_status == HS_PRE_RUNNING) {
1206                 ret = wait_for_remove_process();
1207                 if (ret < 0)
1208                         goto out;
1209                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1210                         goto out;
1211         }
1212         ret = exec_rm();
1213         if (ret < 0)
1214                 goto out;
1215         ret = wait_for_remove_process();
1216         if (ret < 0)
1217                 goto out;
1218         if (snapshot_removal_status != HS_SUCCESS)
1219                 goto out;
1220         ret = post_remove_hook();
1221         if (ret < 0)
1222                 goto out;
1223         if (snapshot_removal_status != HS_POST_RUNNING)
1224                 goto out;
1225         ret = wait_for_remove_process();
1226         if (ret < 0)
1227                 goto out;
1228         ret = 1;
1229 out:
1230         free_snapshot_list(&sl);
1231         return ret;
1232 }
1233
1234 static int com_create(void)
1235 {
1236         int ret, status;
1237         char **rsync_argv;
1238
1239         if (conf.dry_run_given) {
1240                 int i;
1241                 char *msg = NULL;
1242                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1243                 for (i = 0; rsync_argv[i]; i++) {
1244                         char *tmp = msg;
1245                         msg = make_message("%s%s%s", tmp? tmp : "",
1246                                 tmp? " " : "", rsync_argv[i]);
1247                         free(tmp);
1248                 }
1249                 free_rsync_argv(rsync_argv);
1250                 dss_msg("%s\n", msg);
1251                 free(msg);
1252                 return 1;
1253         }
1254         ret = pre_create_hook();
1255         if (ret < 0)
1256                 return ret;
1257         if (create_pid) {
1258                 ret = wait_for_process(create_pid, &status);
1259                 if (ret < 0)
1260                         return ret;
1261                 ret = handle_pre_create_hook_exit(status);
1262                 if (ret <= 0) /* error, or pre-create failed */
1263                         return ret;
1264         }
1265         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1266         ret = create_snapshot(rsync_argv);
1267         if (ret < 0)
1268                 goto out;
1269         ret = wait_for_process(create_pid, &status);
1270         if (ret < 0)
1271                 goto out;
1272         ret = handle_rsync_exit(status);
1273         if (ret < 0)
1274                 goto out;
1275         post_create_hook();
1276         if (create_pid)
1277                 ret = wait_for_process(create_pid, &status);
1278 out:
1279         free_rsync_argv(rsync_argv);
1280         return ret;
1281 }
1282
1283 static int com_ls(void)
1284 {
1285         int i;
1286         struct snapshot_list sl;
1287         struct snapshot *s;
1288
1289         dss_get_snapshot_list(&sl);
1290         FOR_EACH_SNAPSHOT(s, i, &sl) {
1291                 int64_t d = 0;
1292                 if (s->flags & SS_COMPLETE)
1293                         d = (s->completion_time - s->creation_time) / 60;
1294                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1295         };
1296         free_snapshot_list(&sl);
1297         return 1;
1298 }
1299
1300 static int setup_signal_handling(void)
1301 {
1302         int ret;
1303
1304         DSS_INFO_LOG("setting up signal handlers\n");
1305         signal_pipe = signal_init(); /* always successful */
1306         ret = install_sighandler(SIGINT);
1307         if (ret < 0)
1308                 return ret;
1309         ret = install_sighandler(SIGTERM);
1310         if (ret < 0)
1311                 return ret;
1312         return install_sighandler(SIGCHLD);
1313 }
1314
1315 /**
1316  * The main function of dss.
1317  *
1318  * \param argc Usual argument count.
1319  * \param argv Usual argument vector.
1320  */
1321 int main(int argc, char **argv)
1322 {
1323         int ret;
1324         struct cmdline_parser_params params = {
1325                 .override = 0,
1326                 .initialize = 1,
1327                 .check_required = 0,
1328                 .check_ambiguity = 0,
1329                 .print_errors = 1
1330         };
1331
1332         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1333         ret = parse_config_file(0);
1334         if (ret < 0)
1335                 goto out;
1336         if (ret == 0) { /* no config file given */
1337                 /*
1338                  * Parse the command line options again, but this time check
1339                  * that all required options are given.
1340                  */
1341                 params = (struct cmdline_parser_params) {
1342                         .override = 1,
1343                         .initialize = 1,
1344                         .check_required = 1,
1345                         .check_ambiguity = 1,
1346                         .print_errors = 1
1347                 };
1348                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1349         }
1350         if (conf.daemon_given)
1351                 daemon_init();
1352         dump_dss_config("dss configuration");
1353         ret = change_to_dest_dir();
1354         if (ret < 0)
1355                 goto out;
1356         ret = setup_signal_handling();
1357         if (ret < 0)
1358                 goto out;
1359         ret = call_command_handler();
1360 out:
1361         if (ret < 0)
1362                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1363         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1364 }