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, cmd, fds);
 254         free(cmd);
 255         if (ret < 0)
 256                 return ret;
 257         snapshot_removal_status = HS_PRE_RUNNING;
 258         return ret;
 259 }
 260
 261 static int exec_rm(void)
 262 {
 263         struct snapshot *s = snapshot_currently_being_removed;
 264         int fds[3] = {0, 0, 0};
 265         char *new_name = being_deleted_name(s);
 266         char *argv[] = {"rm", "-rf", new_name, NULL};
 267         int ret;
 268
 269         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 270         assert(remove_pid == 0);
 271
 272         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 273         ret = dss_rename(s->name, new_name);
 274         if (ret < 0)
 275                 goto out;
 276         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 277         if (ret < 0)
 278                 goto out;
 279         snapshot_removal_status = HS_RUNNING;
 280 out:
 281         free(new_name);
 282         return ret;
 283 }
 284
 285 static int snapshot_is_being_created(struct snapshot *s)
 286 {
 287         return s->creation_time == current_snapshot_creation_time;
 288 }
 289
 290 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 291 {
 292         struct snapshot *s;
 293         int i;
 294
 295         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 296         FOR_EACH_SNAPSHOT(s, i, sl) {
 297                 if (snapshot_is_being_created(s))
 298                         continue;
 299                 /*
 300                  * We know that no rm is currently running, so if s is marked
 301                  * as being deleted, a previously started rm must have failed.
 302                  */
 303                 if (s->flags & SS_BEING_DELETED)
 304                         return s;
 305
 306                 if (s->flags & SS_COMPLETE) /* good snapshot */
 307                         continue;
 308                 /*
 309                  * This snapshot is incomplete and it is not the snapshot
 310                  * currently being created. However, we must not remove it if
 311                  * rsync is about to be restarted. As only the newest snapshot
 312                  * can be restarted, this snapshot is orphaned if it is not the
 313                  * newest snapshot or if we are not about to restart rsync.
 314                  */
 315                 if (get_newest_snapshot(sl) != s)
 316                         return s;
 317                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 318                         return s;
 319         }
 320         /* no orphaned snapshots */
 321         return NULL;
 322 }
 323
 324 static int is_reference_snapshot(struct snapshot *s)
 325 {
 326         if (!name_of_reference_snapshot)
 327                 return 0;
 328         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 329 }
 330
 331 /*
 332  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 333  */
 334 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 335 {
 336         int i, interval;
 337         struct snapshot *s;
 338         unsigned missing = 0;
 339
 340         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 341         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 342                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 343                 unsigned num = sl->interval_count[interval];
 344                 struct snapshot *victim = NULL, *prev = NULL;
 345                 int64_t score = LONG_MAX;
 346
 347                 if (keep >= num)
 348                         missing += keep - num;
 349 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 350 //                      interval, keep, num, missing);
 351                 if (keep + missing >= num)
 352                         continue;
 353                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 354                 FOR_EACH_SNAPSHOT(s, i, sl) {
 355                         int64_t this_score;
 356
 357                         if (snapshot_is_being_created(s))
 358                                 continue;
 359                         if (is_reference_snapshot(s))
 360                                 continue;
 361                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 362                         if (s->interval > interval) {
 363                                 prev = s;
 364                                 continue;
 365                         }
 366                         if (s->interval < interval)
 367                                 break;
 368                         if (!victim) {
 369                                 victim = s;
 370                                 prev = s;
 371                                 continue;
 372                         }
 373                         assert(prev);
 374                         /* check if s is a better victim */
 375                         this_score = s->creation_time - prev->creation_time;
 376                         assert(this_score >= 0);
 377                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 378                         if (this_score < score) {
 379                                 score = this_score;
 380                                 victim = s;
 381                         }
 382                         prev = s;
 383                 }
 384                 assert(victim);
 385                 return victim;
 386         }
 387         return NULL;
 388 }
 389
 390 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 391 {
 392         int i;
 393         struct snapshot *s;
 394
 395         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 396                 conf.num_intervals_arg);
 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                 if (s->interval < conf.num_intervals_arg)
 403                         continue;
 404                 return s;
 405         }
 406         return NULL;
 407 }
 408
 409 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 410 {
 411         int i;
 412         struct snapshot *s;
 413         FOR_EACH_SNAPSHOT(s, i, sl) {
 414                 if (snapshot_is_being_created(s))
 415                         continue;
 416                 if (is_reference_snapshot(s))
 417                         continue;
 418                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 419                 return s;
 420         }
 421         return NULL;
 422 }
 423
 424 static int rename_incomplete_snapshot(int64_t start)
 425 {
 426         char *old_name;
 427         int ret;
 428
 429         free(path_to_last_complete_snapshot);
 430         ret = complete_name(start, get_current_time(),
 431                 &path_to_last_complete_snapshot);
 432         if (ret < 0)
 433                 return ret;
 434         old_name = incomplete_name(start);
 435         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 436         if (ret >= 0)
 437                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 438                         path_to_last_complete_snapshot);
 439         free(old_name);
 440         return ret;
 441 }
 442
 443 static int try_to_free_disk_space(void)
 444 {
 445         int ret;
 446         struct snapshot_list sl;
 447         struct snapshot *victim;
 448         struct timeval now;
 449         const char *why;
 450         int low_disk_space;
 451
 452         ret = disk_space_low();
 453         if (ret < 0)
 454                 return ret;
 455         low_disk_space = ret;
 456         gettimeofday(&now, NULL);
 457         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 458                 return 0;
 459         if (!low_disk_space) {
 460                 if (conf.keep_redundant_given)
 461                         return 0;
 462                 if (snapshot_creation_status != HS_READY)
 463                         return 0;
 464                 if (next_snapshot_is_due())
 465                         return 0;
 466         }
 467         dss_get_snapshot_list(&sl);
 468         ret = 0;
 469         if (!low_disk_space && sl.num_snapshots <= 1)
 470                 goto out;
 471         why = "outdated";
 472         victim = find_outdated_snapshot(&sl);
 473         if (victim)
 474                 goto remove;
 475         why = "redundant";
 476         victim = find_redundant_snapshot(&sl);
 477         if (victim)
 478                 goto remove;
 479         /* try harder only if disk space is low */
 480         if (!low_disk_space)
 481                 goto out;
 482         why = "orphaned";
 483         victim = find_orphaned_snapshot(&sl);
 484         if (victim)
 485                 goto remove;
 486         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 487         victim = find_oldest_removable_snapshot(&sl);
 488         if (victim)
 489                 goto remove;
 490         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 491         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 492         goto out;
 493 remove:
 494         ret = pre_remove_hook(victim, why);
 495 out:
 496         free_snapshot_list(&sl);
 497         return ret;
 498 }
 499
 500 static int post_create_hook(void)
 501 {
 502         int ret, fds[3] = {0, 0, 0};
 503         char *cmd;
 504
 505         if (!conf.post_create_hook_given) {
 506                 create_pid = 0;
 507                 snapshot_creation_status = HS_READY;
 508                 return 0;
 509         }
 510         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 511                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 512         DSS_NOTICE_LOG("executing %s\n", cmd);
 513         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 514         free(cmd);
 515         if (ret < 0)
 516                 return ret;
 517         snapshot_creation_status = HS_POST_RUNNING;
 518         return ret;
 519 }
 520
 521 static int post_remove_hook(void)
 522 {
 523         int ret, fds[3] = {0, 0, 0};
 524         char *cmd;
 525         struct snapshot *s = snapshot_currently_being_removed;
 526
 527         assert(s);
 528
 529         if (!conf.post_remove_hook_given) {
 530                 snapshot_removal_status = HS_READY;
 531                 return 0;
 532         }
 533         cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
 534                 conf.dest_dir_arg, s->name);
 535         DSS_NOTICE_LOG("executing %s\n", cmd);
 536         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 537         free(cmd);
 538         if (ret < 0)
 539                 return ret;
 540         snapshot_removal_status = HS_POST_RUNNING;
 541         return ret;
 542 }
 543
 544 static void dss_kill(pid_t pid, int sig, const char *msg)
 545 {
 546         const char *signame, *process_name;
 547
 548         if (pid == 0)
 549                 return;
 550         switch (sig) {
 551         case SIGTERM: signame = "TERM"; break;
 552         case SIGSTOP: signame = "STOP"; break;
 553         case SIGCONT: signame = "CONT"; break;
 554         default: signame = "????";
 555         }
 556
 557         if (pid == create_pid)
 558                 process_name = "create";
 559         else if (pid == remove_pid)
 560                 process_name = "remove";
 561         else process_name = "??????";
 562
 563         if (msg)
 564                 DSS_INFO_LOG("%s\n", msg);
 565         DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n",
 566                 sig, signame, (int)pid, process_name);
 567         if (kill(pid, sig) >= 0)
 568                 return;
 569         DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
 570                 sig, signame, (int)pid, process_name);
 571 }
 572
 573 static void stop_create_process(void)
 574 {
 575         if (create_process_stopped)
 576                 return;
 577         dss_kill(create_pid, SIGSTOP, "suspending create process");
 578         create_process_stopped = 1;
 579 }
 580
 581 static void restart_create_process(void)
 582 {
 583         if (!create_process_stopped)
 584                 return;
 585         dss_kill(create_pid, SIGCONT, "resuming create process");
 586         create_process_stopped = 0;
 587 }
 588
 589 /**
 590  * Print a log message about the exit status of a child.
 591  */
 592 static void log_termination_msg(pid_t pid, int status)
 593 {
 594         if (WIFEXITED(status))
 595                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 596                         WEXITSTATUS(status));
 597         else if (WIFSIGNALED(status))
 598                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 599                         WTERMSIG(status));
 600         else
 601                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 602 }
 603
 604 static int wait_for_process(pid_t pid, int *status)
 605 {
 606         int ret;
 607
 608         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 609         for (;;) {
 610                 fd_set rfds;
 611
 612                 FD_ZERO(&rfds);
 613                 FD_SET(signal_pipe, &rfds);
 614                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 615                 if (ret < 0)
 616                         break;
 617                 ret = next_signal();
 618                 if (!ret)
 619                         continue;
 620                 if (ret == SIGCHLD) {
 621                         ret = waitpid(pid, status, 0);
 622                         if (ret >= 0)
 623                                 break;
 624                         if (errno != EINTR) { /* error */
 625                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 626                                 break;
 627                         }
 628                 }
 629                 /* SIGINT or SIGTERM */
 630                 dss_kill(pid, SIGTERM, "killing child process");
 631         }
 632         if (ret < 0)
 633                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 634         else
 635                 log_termination_msg(pid, *status);
 636         return ret;
 637 }
 638
 639 static void handle_pre_remove_exit(int status)
 640 {
 641         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 642                 snapshot_removal_status = HS_READY;
 643                 gettimeofday(&next_removal_check, NULL);
 644                 next_removal_check.tv_sec += 60;
 645                 return;
 646         }
 647         snapshot_removal_status = HS_PRE_SUCCESS;
 648 }
 649
 650 static int handle_rm_exit(int status)
 651 {
 652         if (!WIFEXITED(status)) {
 653                 snapshot_removal_status = HS_READY;
 654                 return -E_INVOLUNTARY_EXIT;
 655         }
 656         if (WEXITSTATUS(status)) {
 657                 snapshot_removal_status = HS_READY;
 658                 return -E_BAD_EXIT_CODE;
 659         }
 660         if (conf.post_remove_hook_given)
 661                 snapshot_removal_status = HS_SUCCESS;
 662         else
 663                 snapshot_removal_status = HS_READY;
 664         return 1;
 665 }
 666
 667 static void handle_post_remove_exit(void)
 668 {
 669         snapshot_removal_status = HS_READY;
 670 }
 671
 672 static int handle_remove_exit(int status)
 673 {
 674         int ret;
 675         struct snapshot *s = snapshot_currently_being_removed;
 676
 677         assert(s);
 678         switch (snapshot_removal_status) {
 679         case HS_PRE_RUNNING:
 680                 handle_pre_remove_exit(status);
 681                 ret = 1;
 682                 break;
 683         case HS_RUNNING:
 684                 ret = handle_rm_exit(status);
 685                 break;
 686         case HS_POST_RUNNING:
 687                 handle_post_remove_exit();
 688                 ret = 1;
 689                 break;
 690         default:
 691                 ret = -E_BUG;
 692         }
 693         if (snapshot_removal_status == HS_READY) {
 694                 free(s->name);
 695                 free(s);
 696                 snapshot_currently_being_removed = NULL;
 697         }
 698         remove_pid = 0;
 699         return ret;
 700 }
 701
 702 static int wait_for_remove_process(void)
 703 {
 704         int status, ret;
 705
 706         assert(remove_pid);
 707         assert(
 708                 snapshot_removal_status == HS_PRE_RUNNING ||
 709                 snapshot_removal_status == HS_RUNNING ||
 710                 snapshot_removal_status == HS_POST_RUNNING
 711         );
 712         ret = wait_for_process(remove_pid, &status);
 713         if (ret < 0)
 714                 return ret;
 715         return handle_remove_exit(status);
 716 }
 717
 718 static int handle_rsync_exit(int status)
 719 {
 720         int es, ret;
 721
 722         if (!WIFEXITED(status)) {
 723                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 724                 ret = -E_INVOLUNTARY_EXIT;
 725                 snapshot_creation_status = HS_READY;
 726                 goto out;
 727         }
 728         es = WEXITSTATUS(status);
 729         /*
 730          * Restart rsync on non-fatal errors:
 731          * 12: Error in rsync protocol data stream
 732          * 13: Errors with program diagnostics
 733          */
 734         if (es == 12 || es == 13) {
 735                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 736                         (int)create_pid, es);
 737                 snapshot_creation_status = HS_NEEDS_RESTART;
 738                 next_snapshot_time = get_current_time() + 60;
 739                 ret = 1;
 740                 goto out;
 741         }
 742         if (es != 0 && es != 23 && es != 24) {
 743                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 744                 ret = -E_BAD_EXIT_CODE;
 745                 snapshot_creation_status = HS_READY;
 746                 goto out;
 747         }
 748         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 749         if (ret < 0)
 750                 goto out;
 751         snapshot_creation_status = HS_SUCCESS;
 752         free(name_of_reference_snapshot);
 753         name_of_reference_snapshot = NULL;
 754 out:
 755         create_process_stopped = 0;
 756         return ret;
 757 }
 758
 759 static int handle_pre_create_hook_exit(int status)
 760 {
 761         int es, ret;
 762         static int warn_count;
 763
 764         if (!WIFEXITED(status)) {
 765                 snapshot_creation_status = HS_READY;
 766                 ret = -E_INVOLUNTARY_EXIT;
 767                 goto out;
 768         }
 769         es = WEXITSTATUS(status);
 770         if (es) {
 771                 if (!warn_count--) {
 772                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 773                                 conf.pre_create_hook_arg, es);
 774                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 775                         warn_count = 60; /* warn only once per hour */
 776                 }
 777                 next_snapshot_time = get_current_time() + 60;
 778                 snapshot_creation_status = HS_READY;
 779                 ret = 0;
 780                 goto out;
 781         }
 782         warn_count = 0;
 783         snapshot_creation_status = HS_PRE_SUCCESS;
 784         ret = 1;
 785 out:
 786         return ret;
 787 }
 788
 789 static int handle_sigchld(void)
 790 {
 791         pid_t pid;
 792         int status, ret = reap_child(&pid, &status);
 793
 794         if (ret <= 0)
 795                 return ret;
 796
 797         if (pid == create_pid) {
 798                 switch (snapshot_creation_status) {
 799                 case HS_PRE_RUNNING:
 800                         ret = handle_pre_create_hook_exit(status);
 801                         break;
 802                 case HS_RUNNING:
 803                         ret = handle_rsync_exit(status);
 804                         break;
 805                 case HS_POST_RUNNING:
 806                         snapshot_creation_status = HS_READY;
 807                         ret = 1;
 808                         break;
 809                 default:
 810                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 811                                 snapshot_creation_status);
 812                         return -E_BUG;
 813                 }
 814                 create_pid = 0;
 815                 return ret;
 816         }
 817         if (pid == remove_pid) {
 818                 ret = handle_remove_exit(status);
 819                 if (ret < 0)
 820                         return ret;
 821                 return ret;
 822         }
 823         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 824         return -E_BUG;
 825 }
 826
 827 static int check_config(void)
 828 {
 829         if (conf.unit_interval_arg <= 0) {
 830                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 831                 return -E_INVALID_NUMBER;
 832         }
 833         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 834         if (conf.num_intervals_arg <= 0) {
 835                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 836                 return -E_INVALID_NUMBER;
 837         }
 838         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 839         return 1;
 840 }
 841
 842 /*
 843  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 844  * file was read successfully.
 845  */
 846 static int parse_config_file(int override)
 847 {
 848         int ret, config_file_exists;
 849         char *config_file;
 850         struct stat statbuf;
 851         char *old_logfile_arg = NULL;
 852         int old_daemon_given = 0;
 853
 854         if (conf.config_file_given)
 855                 config_file = dss_strdup(conf.config_file_arg);
 856         else {
 857                 char *home = get_homedir();
 858                 config_file = make_message("%s/.dssrc", home);
 859                 free(home);
 860         }
 861         if (override) { /* SIGHUP */
 862                 if (conf.logfile_given)
 863                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 864                 old_daemon_given = conf.daemon_given;
 865         }
 866
 867         config_file_exists = !stat(config_file, &statbuf);
 868         if (!config_file_exists && conf.config_file_given) {
 869                 ret = -ERRNO_TO_DSS_ERROR(errno);
 870                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 871                 goto out;
 872         }
 873         if (config_file_exists) {
 874                 struct cmdline_parser_params params = {
 875                         .override = override,
 876                         .initialize = 0,
 877                         .check_required = 1,
 878                         .check_ambiguity = 0,
 879                         .print_errors = 1
 880                 };
 881                 if (override) { /* invalidate all rsync options */
 882                         int i;
 883
 884                         for (i = 0; i < conf.rsync_option_given; i++) {
 885                                 free(conf.rsync_option_arg[i]);
 886                                 conf.rsync_option_arg[i] = NULL;
 887                         }
 888                         conf.rsync_option_given = 0;
 889                 }
 890                 cmdline_parser_config_file(config_file, &conf, &params);
 891         }
 892         ret = check_config();
 893         if (ret < 0)
 894                 goto out;
 895         if (override) {
 896                 /* don't change daemon mode on SIGHUP */
 897                 conf.daemon_given = old_daemon_given;
 898                 close_log(logfile);
 899                 logfile = NULL;
 900                 if (conf.logfile_given)
 901                         free(old_logfile_arg);
 902                 else if (conf.daemon_given) { /* re-use old logfile */
 903                         conf.logfile_arg = old_logfile_arg;
 904                         conf.logfile_given = 1;
 905                 }
 906         }
 907         if (conf.logfile_given) {
 908                 logfile = open_log(conf.logfile_arg);
 909                 log_welcome(conf.loglevel_arg);
 910         }
 911         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 912         ret = config_file_exists;
 913 out:
 914         free(config_file);
 915         if (ret < 0)
 916                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 917         return ret;
 918 }
 919
 920 static int change_to_dest_dir(void)
 921 {
 922         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 923         return dss_chdir(conf.dest_dir_arg);
 924 }
 925
 926 static void dump_dss_config(const char *msg)
 927 {
 928         if (conf.loglevel_arg > INFO)
 929                 return;
 930         DSS_INFO_LOG("%s\n", msg);
 931         cmdline_parser_dump(logfile? logfile : stderr, &conf);
 932 }
 933
 934 static int handle_sighup(void)
 935 {
 936         int ret;
 937
 938         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
 939         dump_dss_config("current config");
 940         ret = parse_config_file(1);
 941         if (ret < 0)
 942                 return ret;
 943         dump_dss_config("new config");
 944         invalidate_next_snapshot_time();
 945         return change_to_dest_dir();
 946 }
 947
 948 static int handle_signal(void)
 949 {
 950         int sig, ret = next_signal();
 951
 952         if (ret <= 0)
 953                 goto out;
 954         sig = ret;
 955         switch (sig) {
 956         case SIGINT:
 957         case SIGTERM:
 958                 restart_create_process();
 959                 dss_kill(create_pid, SIGTERM, NULL);
 960                 dss_kill(remove_pid, SIGTERM, NULL);
 961                 ret = -E_SIGNAL;
 962                 break;
 963         case SIGHUP:
 964                 ret = handle_sighup();
 965                 break;
 966         case SIGCHLD:
 967                 ret = handle_sigchld();
 968                 break;
 969         }
 970 out:
 971         if (ret < 0)
 972                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 973         return ret;
 974 }
 975
 976 /*
 977  * We can not use rsync locally if the local user is different from the remote
 978  * user or if the src dir is not on the local host (or both).
 979  */
 980 static int use_rsync_locally(char *logname)
 981 {
 982         char *h = conf.remote_host_arg;
 983
 984         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 985                 return 0;
 986         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 987                 return 0;
 988         return 1;
 989 }
 990
 991 static int rename_resume_snap(int64_t creation_time)
 992 {
 993         struct snapshot_list sl = {.num_snapshots = 0};
 994         struct snapshot *s;
 995         char *new_name = incomplete_name(creation_time);
 996         int ret;
 997
 998         ret = 0;
 999         if (conf.no_resume_given)
1000                 goto out;
1001         dss_get_snapshot_list(&sl);
1002         s = get_newest_snapshot(&sl);
1003         if (!s)
1004                 goto out;
1005         if ((s->flags & SS_COMPLETE) != 0) /* complete */
1006                 goto out;
1007         DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
1008         ret = dss_rename(s->name, new_name);
1009 out:
1010         if (ret >= 0)
1011                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
1012         free(new_name);
1013         free_snapshot_list(&sl);
1014         return ret;
1015 }
1016
1017 static void create_rsync_argv(char ***argv, int64_t *num)
1018 {
1019         char *logname;
1020         int i = 0, j;
1021         struct snapshot_list sl;
1022
1023         dss_get_snapshot_list(&sl);
1024         assert(!name_of_reference_snapshot);
1025         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1026         free_snapshot_list(&sl);
1027
1028         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1029         (*argv)[i++] = dss_strdup("rsync");
1030         (*argv)[i++] = dss_strdup("-aq");
1031         (*argv)[i++] = dss_strdup("--delete");
1032         for (j = 0; j < conf.rsync_option_given; j++)
1033                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1034         if (name_of_reference_snapshot) {
1035                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1036                 (*argv)[i++] = make_message("--link-dest=../%s",
1037                         name_of_reference_snapshot);
1038         } else
1039                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1040         logname = dss_logname();
1041         if (use_rsync_locally(logname))
1042                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1043         else
1044                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1045                         conf.remote_user_arg : logname,
1046                         conf.remote_host_arg, conf.source_dir_arg);
1047         free(logname);
1048         *num = get_current_time();
1049         (*argv)[i++] = incomplete_name(*num);
1050         (*argv)[i++] = NULL;
1051         for (j = 0; j < i; j++)
1052                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1053 }
1054
1055 static void free_rsync_argv(char **argv)
1056 {
1057         int i;
1058
1059         if (!argv)
1060                 return;
1061         for (i = 0; argv[i]; i++)
1062                 free(argv[i]);
1063         free(argv);
1064 }
1065
1066 static int create_snapshot(char **argv)
1067 {
1068         int ret, fds[3] = {0, 0, 0};
1069
1070         ret = rename_resume_snap(current_snapshot_creation_time);
1071         if (ret < 0)
1072                 return ret;
1073         ret = dss_exec(&create_pid, argv[0], argv, fds);
1074         if (ret < 0)
1075                 return ret;
1076         snapshot_creation_status = HS_RUNNING;
1077         return ret;
1078 }
1079
1080 static int select_loop(void)
1081 {
1082         int ret;
1083         /* check every 60 seconds for free disk space */
1084         struct timeval tv;
1085         char **rsync_argv = NULL;
1086
1087         for (;;) {
1088                 fd_set rfds;
1089                 struct timeval *tvp;
1090
1091                 if (remove_pid)
1092                         tvp = NULL; /* sleep until rm hook/process dies */
1093                 else { /* sleep one minute */
1094                         tv.tv_sec = 60;
1095                         tv.tv_usec = 0;
1096                         tvp = &tv;
1097                 }
1098                 FD_ZERO(&rfds);
1099                 FD_SET(signal_pipe, &rfds);
1100                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1101                 if (ret < 0)
1102                         goto out;
1103                 if (FD_ISSET(signal_pipe, &rfds)) {
1104                         ret = handle_signal();
1105                         if (ret < 0)
1106                                 goto out;
1107                 }
1108                 if (remove_pid)
1109                         continue;
1110                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1111                         ret = exec_rm();
1112                         if (ret < 0)
1113                                 goto out;
1114                         continue;
1115                 }
1116                 if (snapshot_removal_status == HS_SUCCESS) {
1117                         ret = post_remove_hook();
1118                         if (ret < 0)
1119                                 goto out;
1120                         continue;
1121                 }
1122                 ret = try_to_free_disk_space();
1123                 if (ret < 0)
1124                         goto out;
1125                 if (snapshot_removal_status != HS_READY) {
1126                         stop_create_process();
1127                         continue;
1128                 }
1129                 restart_create_process();
1130                 switch (snapshot_creation_status) {
1131                 case HS_READY:
1132                         if (!next_snapshot_is_due())
1133                                 continue;
1134                         ret = pre_create_hook();
1135                         if (ret < 0)
1136                                 goto out;
1137                         continue;
1138                 case HS_PRE_RUNNING:
1139                 case HS_RUNNING:
1140                 case HS_POST_RUNNING:
1141                         continue;
1142                 case HS_PRE_SUCCESS:
1143                         if (!name_of_reference_snapshot) {
1144                                 free_rsync_argv(rsync_argv);
1145                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1146                         }
1147                         ret = create_snapshot(rsync_argv);
1148                         if (ret < 0)
1149                                 goto out;
1150                         continue;
1151                 case HS_NEEDS_RESTART:
1152                         if (!next_snapshot_is_due())
1153                                 continue;
1154                         ret = create_snapshot(rsync_argv);
1155                         if (ret < 0)
1156                                 goto out;
1157                         continue;
1158                 case HS_SUCCESS:
1159                         ret = post_create_hook();
1160                         if (ret < 0)
1161                                 goto out;
1162                         continue;
1163                 }
1164         }
1165 out:
1166         return ret;
1167 }
1168
1169 static void exit_hook(int exit_code)
1170 {
1171         int fds[3] = {0, 0, 0};
1172         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1173         pid_t pid;
1174
1175         if (!conf.exit_hook_given)
1176                 return;
1177         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1178         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1179 }
1180
1181 static int com_run(void)
1182 {
1183         int ret;
1184
1185         if (conf.dry_run_given) {
1186                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1187                 return -E_SYNTAX;
1188         }
1189         ret = install_sighandler(SIGHUP);
1190         if (ret < 0)
1191                 return ret;
1192         ret = select_loop();
1193         if (ret >= 0) /* impossible */
1194                 ret = -E_BUG;
1195         exit_hook(ret);
1196         return ret;
1197 }
1198
1199 static int com_prune(void)
1200 {
1201         int ret;
1202         struct snapshot_list sl;
1203         struct snapshot *victim;
1204         struct disk_space ds;
1205         const char *why;
1206
1207         ret = get_disk_space(".", &ds);
1208         if (ret < 0)
1209                 return ret;
1210         log_disk_space(&ds);
1211         dss_get_snapshot_list(&sl);
1212         why = "outdated";
1213         victim = find_outdated_snapshot(&sl);
1214         if (victim)
1215                 goto rm;
1216         why = "redundant";
1217         victim = find_redundant_snapshot(&sl);
1218         if (victim)
1219                 goto rm;
1220         ret = 0;
1221         goto out;
1222 rm:
1223         if (conf.dry_run_given) {
1224                 dss_msg("%s snapshot %s (interval = %i)\n",
1225                         why, victim->name, victim->interval);
1226                 ret = 0;
1227                 goto out;
1228         }
1229         ret = pre_remove_hook(victim, why);
1230         if (ret < 0)
1231                 goto out;
1232         if (snapshot_removal_status == HS_PRE_RUNNING) {
1233                 ret = wait_for_remove_process();
1234                 if (ret < 0)
1235                         goto out;
1236                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1237                         goto out;
1238         }
1239         ret = exec_rm();
1240         if (ret < 0)
1241                 goto out;
1242         ret = wait_for_remove_process();
1243         if (ret < 0)
1244                 goto out;
1245         if (snapshot_removal_status != HS_SUCCESS)
1246                 goto out;
1247         ret = post_remove_hook();
1248         if (ret < 0)
1249                 goto out;
1250         if (snapshot_removal_status != HS_POST_RUNNING)
1251                 goto out;
1252         ret = wait_for_remove_process();
1253         if (ret < 0)
1254                 goto out;
1255         ret = 1;
1256 out:
1257         free_snapshot_list(&sl);
1258         return ret;
1259 }
1260
1261 static int com_create(void)
1262 {
1263         int ret, status;
1264         char **rsync_argv;
1265
1266         if (conf.dry_run_given) {
1267                 int i;
1268                 char *msg = NULL;
1269                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1270                 for (i = 0; rsync_argv[i]; i++) {
1271                         char *tmp = msg;
1272                         msg = make_message("%s%s%s", tmp? tmp : "",
1273                                 tmp? " " : "", rsync_argv[i]);
1274                         free(tmp);
1275                 }
1276                 free_rsync_argv(rsync_argv);
1277                 dss_msg("%s\n", msg);
1278                 free(msg);
1279                 return 1;
1280         }
1281         ret = pre_create_hook();
1282         if (ret < 0)
1283                 return ret;
1284         if (create_pid) {
1285                 ret = wait_for_process(create_pid, &status);
1286                 if (ret < 0)
1287                         return ret;
1288                 ret = handle_pre_create_hook_exit(status);
1289                 if (ret <= 0) /* error, or pre-create failed */
1290                         return ret;
1291         }
1292         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1293         ret = create_snapshot(rsync_argv);
1294         if (ret < 0)
1295                 goto out;
1296         ret = wait_for_process(create_pid, &status);
1297         if (ret < 0)
1298                 goto out;
1299         ret = handle_rsync_exit(status);
1300         if (ret < 0)
1301                 goto out;
1302         post_create_hook();
1303         if (create_pid)
1304                 ret = wait_for_process(create_pid, &status);
1305 out:
1306         free_rsync_argv(rsync_argv);
1307         return ret;
1308 }
1309
1310 static int com_ls(void)
1311 {
1312         int i;
1313         struct snapshot_list sl;
1314         struct snapshot *s;
1315
1316         dss_get_snapshot_list(&sl);
1317         FOR_EACH_SNAPSHOT(s, i, &sl) {
1318                 int64_t d = 0;
1319                 if (s->flags & SS_COMPLETE)
1320                         d = (s->completion_time - s->creation_time) / 60;
1321                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1322         };
1323         free_snapshot_list(&sl);
1324         return 1;
1325 }
1326
1327 static int setup_signal_handling(void)
1328 {
1329         int ret;
1330
1331         DSS_INFO_LOG("setting up signal handlers\n");
1332         signal_pipe = signal_init(); /* always successful */
1333         ret = install_sighandler(SIGINT);
1334         if (ret < 0)
1335                 return ret;
1336         ret = install_sighandler(SIGTERM);
1337         if (ret < 0)
1338                 return ret;
1339         return install_sighandler(SIGCHLD);
1340 }
1341
1342 /**
1343  * The main function of dss.
1344  *
1345  * \param argc Usual argument count.
1346  * \param argv Usual argument vector.
1347  */
1348 int main(int argc, char **argv)
1349 {
1350         int ret;
1351         struct cmdline_parser_params params = {
1352                 .override = 0,
1353                 .initialize = 1,
1354                 .check_required = 0,
1355                 .check_ambiguity = 0,
1356                 .print_errors = 1
1357         };
1358
1359         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1360         ret = parse_config_file(0);
1361         if (ret < 0)
1362                 goto out;
1363         if (ret == 0) { /* no config file given */
1364                 /*
1365                  * Parse the command line options again, but this time check
1366                  * that all required options are given.
1367                  */
1368                 params = (struct cmdline_parser_params) {
1369                         .override = 1,
1370                         .initialize = 1,
1371                         .check_required = 1,
1372                         .check_ambiguity = 1,
1373                         .print_errors = 1
1374                 };
1375                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1376         }
1377         if (conf.daemon_given)
1378                 daemon_init();
1379         dump_dss_config("dss configuration");
1380         ret = change_to_dest_dir();
1381         if (ret < 0)
1382                 goto out;
1383         ret = setup_signal_handling();
1384         if (ret < 0)
1385                 goto out;
1386         ret = call_command_handler();
1387 out:
1388         if (ret < 0)
1389                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1390         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1391 }