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