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         DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
 218         ret = dss_exec_cmdline_pid(&create_pid,
 219                 conf.pre_create_hook_arg, fds);
 220         if (ret < 0)
 221                 return ret;
 222         snapshot_creation_status = HS_PRE_RUNNING;
 223         return ret;
 224 }
 225
 226 static int pre_remove_hook(struct snapshot *s, const char *why)
 227 {
 228         int ret, fds[3] = {0, 0, 0};
 229         char *cmd;
 230
 231         if (!s)
 232                 return 0;
 233         DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 234         assert(snapshot_removal_status == HS_READY);
 235         assert(remove_pid == 0);
 236         assert(!snapshot_currently_being_removed);
 237
 238         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 239         *snapshot_currently_being_removed = *s;
 240         snapshot_currently_being_removed->name = dss_strdup(s->name);
 241
 242         cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
 243                 conf.dest_dir_arg, s->name);
 244         DSS_DEBUG_LOG("executing %s\n", cmd);
 245         ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 246         free(cmd);
 247         if (ret < 0)
 248                 return ret;
 249         snapshot_removal_status = HS_PRE_RUNNING;
 250         return ret;
 251 }
 252
 253 static int exec_rm(void)
 254 {
 255         struct snapshot *s = snapshot_currently_being_removed;
 256         int fds[3] = {0, 0, 0};
 257         char *new_name = being_deleted_name(s);
 258         char *argv[] = {"rm", "-rf", new_name, NULL};
 259         int ret;
 260
 261         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 262         assert(remove_pid == 0);
 263
 264         DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
 265         ret = dss_rename(s->name, new_name);
 266         if (ret < 0)
 267                 goto out;
 268         ret = dss_exec(&remove_pid, argv[0], argv, fds);
 269         if (ret < 0)
 270                 goto out;
 271         snapshot_removal_status = HS_RUNNING;
 272 out:
 273         free(new_name);
 274         return ret;
 275 }
 276
 277 static int snapshot_is_being_created(struct snapshot *s)
 278 {
 279         return s->creation_time == current_snapshot_creation_time;
 280 }
 281
 282 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 283 {
 284         struct snapshot *s;
 285         int i;
 286
 287         DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 288         FOR_EACH_SNAPSHOT(s, i, sl) {
 289                 if (snapshot_is_being_created(s))
 290                         continue;
 291                 /*
 292                  * We know that no rm is currently running, so if s is marked
 293                  * as being deleted, a previously started rm must have failed.
 294                  */
 295                 if (s->flags & SS_BEING_DELETED)
 296                         return s;
 297
 298                 if (s->flags & SS_COMPLETE) /* good snapshot */
 299                         continue;
 300                 /*
 301                  * This snapshot is incomplete and it is not the snapshot
 302                  * currently being created. However, we must not remove it if
 303                  * rsync is about to be restarted. As only the newest snapshot
 304                  * can be restarted, this snapshot is orphaned if it is not the
 305                  * newest snapshot or if we are not about to restart rsync.
 306                  */
 307                 if (get_newest_snapshot(sl) != s)
 308                         return s;
 309                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 310                         return s;
 311         }
 312         /* no orphaned snapshots */
 313         return NULL;
 314 }
 315
 316 static int is_reference_snapshot(struct snapshot *s)
 317 {
 318         if (!name_of_reference_snapshot)
 319                 return 0;
 320         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 321 }
 322
 323 /*
 324  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 325  */
 326 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 327 {
 328         int i, interval;
 329         struct snapshot *s;
 330         unsigned missing = 0;
 331
 332         DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 333         for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 334                 unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 335                 unsigned num = sl->interval_count[interval];
 336                 struct snapshot *victim = NULL, *prev = NULL;
 337                 int64_t score = LONG_MAX;
 338
 339                 if (keep >= num)
 340                         missing += keep - num;
 341 //              DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
 342 //                      interval, keep, num, missing);
 343                 if (keep + missing >= num)
 344                         continue;
 345                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 346                 FOR_EACH_SNAPSHOT(s, i, sl) {
 347                         int64_t this_score;
 348
 349                         if (snapshot_is_being_created(s))
 350                                 continue;
 351                         if (is_reference_snapshot(s))
 352                                 continue;
 353                         //DSS_DEBUG_LOG("checking %s\n", s->name);
 354                         if (s->interval > interval) {
 355                                 prev = s;
 356                                 continue;
 357                         }
 358                         if (s->interval < interval)
 359                                 break;
 360                         if (!victim) {
 361                                 victim = s;
 362                                 prev = s;
 363                                 continue;
 364                         }
 365                         assert(prev);
 366                         /* check if s is a better victim */
 367                         this_score = s->creation_time - prev->creation_time;
 368                         assert(this_score >= 0);
 369                         //DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 370                         if (this_score < score) {
 371                                 score = this_score;
 372                                 victim = s;
 373                         }
 374                         prev = s;
 375                 }
 376                 assert(victim);
 377                 return victim;
 378         }
 379         return NULL;
 380 }
 381
 382 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 383 {
 384         int i;
 385         struct snapshot *s;
 386
 387         DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
 388                 conf.num_intervals_arg);
 389         FOR_EACH_SNAPSHOT(s, i, sl) {
 390                 if (snapshot_is_being_created(s))
 391                         continue;
 392                 if (is_reference_snapshot(s))
 393                         continue;
 394                 if (s->interval < conf.num_intervals_arg)
 395                         continue;
 396                 return s;
 397         }
 398         return NULL;
 399 }
 400
 401 struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 402 {
 403         int i;
 404         struct snapshot *s;
 405         FOR_EACH_SNAPSHOT(s, i, sl) {
 406                 if (snapshot_is_being_created(s))
 407                         continue;
 408                 if (is_reference_snapshot(s))
 409                         continue;
 410                 DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
 411                 return s;
 412         }
 413         return NULL;
 414 }
 415
 416 static int rename_incomplete_snapshot(int64_t start)
 417 {
 418         char *old_name;
 419         int ret;
 420
 421         free(path_to_last_complete_snapshot);
 422         ret = complete_name(start, get_current_time(),
 423                 &path_to_last_complete_snapshot);
 424         if (ret < 0)
 425                 return ret;
 426         old_name = incomplete_name(start);
 427         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 428         if (ret >= 0)
 429                 DSS_NOTICE_LOG("%s -> %s\n", old_name,
 430                         path_to_last_complete_snapshot);
 431         free(old_name);
 432         return ret;
 433 }
 434
 435 static int try_to_free_disk_space(void)
 436 {
 437         int ret;
 438         struct snapshot_list sl;
 439         struct snapshot *victim;
 440         struct timeval now;
 441         const char *why;
 442         int low_disk_space;
 443
 444         ret = disk_space_low();
 445         if (ret < 0)
 446                 return ret;
 447         low_disk_space = ret;
 448         gettimeofday(&now, NULL);
 449         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 450                 return 0;
 451         if (!low_disk_space) {
 452                 if (conf.keep_redundant_given)
 453                         return 0;
 454                 if (snapshot_creation_status != HS_READY)
 455                         return 0;
 456                 if (next_snapshot_is_due())
 457                         return 0;
 458         }
 459         dss_get_snapshot_list(&sl);
 460         ret = 0;
 461         if (!low_disk_space && sl.num_snapshots <= 1)
 462                 goto out;
 463         why = "outdated";
 464         victim = find_outdated_snapshot(&sl);
 465         if (victim)
 466                 goto remove;
 467         why = "redundant";
 468         victim = find_redundant_snapshot(&sl);
 469         if (victim)
 470                 goto remove;
 471         /* try harder only if disk space is low */
 472         if (!low_disk_space)
 473                 goto out;
 474         why = "orphaned";
 475         victim = find_orphaned_snapshot(&sl);
 476         if (victim)
 477                 goto remove;
 478         DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
 479         victim = find_oldest_removable_snapshot(&sl);
 480         if (victim)
 481                 goto remove;
 482         DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
 483         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 484         goto out;
 485 remove:
 486         ret = pre_remove_hook(victim, why);
 487 out:
 488         free_snapshot_list(&sl);
 489         return ret;
 490 }
 491
 492 static int post_create_hook(void)
 493 {
 494         int ret, fds[3] = {0, 0, 0};
 495         char *cmd;
 496
 497         cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
 498                 conf.dest_dir_arg, path_to_last_complete_snapshot);
 499         DSS_NOTICE_LOG("executing %s\n", cmd);
 500         ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
 501         free(cmd);
 502         if (ret < 0)
 503                 return ret;
 504         snapshot_creation_status = HS_POST_RUNNING;
 505         return ret;
 506 }
 507
 508 static int post_remove_hook(void)
 509 {
 510         int ret, fds[3] = {0, 0, 0};
 511         char *cmd;
 512         struct snapshot *s = snapshot_currently_being_removed;
 513
 514         assert(s);
 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 dss_kill(pid_t pid, int sig, const char *msg)
 528 {
 529         const char *signame, *process_name;
 530
 531         if (pid == 0)
 532                 return;
 533         switch (sig) {
 534         case SIGTERM: signame = "TERM"; break;
 535         case SIGSTOP: signame = "STOP"; break;
 536         case SIGCONT: signame = "CONT"; break;
 537         default: signame = "????";
 538         }
 539
 540         if (pid == create_pid)
 541                 process_name = "create";
 542         else if (pid == remove_pid)
 543                 process_name = "remove";
 544         else process_name = "??????";
 545
 546         if (msg)
 547                 DSS_INFO_LOG("%s\n", msg);
 548         DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n",
 549                 sig, signame, (int)pid, process_name);
 550         if (kill(pid, sig) >= 0)
 551                 return;
 552         DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
 553                 sig, signame, (int)pid, process_name);
 554 }
 555
 556 static void stop_create_process(void)
 557 {
 558         if (create_process_stopped)
 559                 return;
 560         dss_kill(create_pid, SIGSTOP, "suspending create process");
 561         create_process_stopped = 1;
 562 }
 563
 564 static void restart_create_process(void)
 565 {
 566         if (!create_process_stopped)
 567                 return;
 568         dss_kill(create_pid, SIGCONT, "resuming create process");
 569         create_process_stopped = 0;
 570 }
 571
 572 /**
 573  * Print a log message about the exit status of a child.
 574  */
 575 static void log_termination_msg(pid_t pid, int status)
 576 {
 577         if (WIFEXITED(status))
 578                 DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
 579                         WEXITSTATUS(status));
 580         else if (WIFSIGNALED(status))
 581                 DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
 582                         WTERMSIG(status));
 583         else
 584                 DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
 585 }
 586
 587 static int wait_for_process(pid_t pid, int *status)
 588 {
 589         int ret;
 590
 591         DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
 592         for (;;) {
 593                 fd_set rfds;
 594
 595                 FD_ZERO(&rfds);
 596                 FD_SET(signal_pipe, &rfds);
 597                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 598                 if (ret < 0)
 599                         break;
 600                 ret = next_signal();
 601                 if (!ret)
 602                         continue;
 603                 if (ret == SIGCHLD) {
 604                         ret = waitpid(pid, status, 0);
 605                         if (ret >= 0)
 606                                 break;
 607                         if (errno != EINTR) { /* error */
 608                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 609                                 break;
 610                         }
 611                 }
 612                 /* SIGINT or SIGTERM */
 613                 dss_kill(pid, SIGTERM, "killing child process");
 614         }
 615         if (ret < 0)
 616                 DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
 617         else
 618                 log_termination_msg(pid, *status);
 619         return ret;
 620 }
 621
 622 static void handle_pre_remove_exit(int status)
 623 {
 624         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 625                 snapshot_removal_status = HS_READY;
 626                 gettimeofday(&next_removal_check, NULL);
 627                 next_removal_check.tv_sec += 60;
 628                 return;
 629         }
 630         snapshot_removal_status = HS_PRE_SUCCESS;
 631 }
 632
 633 static int handle_rm_exit(int status)
 634 {
 635         if (!WIFEXITED(status)) {
 636                 snapshot_removal_status = HS_READY;
 637                 return -E_INVOLUNTARY_EXIT;
 638         }
 639         if (WEXITSTATUS(status)) {
 640                 snapshot_removal_status = HS_READY;
 641                 return -E_BAD_EXIT_CODE;
 642         }
 643         snapshot_removal_status = HS_SUCCESS;
 644         return 1;
 645 }
 646
 647 static void handle_post_remove_exit(void)
 648 {
 649         snapshot_removal_status = HS_READY;
 650 }
 651
 652 static int handle_remove_exit(int status)
 653 {
 654         int ret;
 655         struct snapshot *s = snapshot_currently_being_removed;
 656
 657         assert(s);
 658         switch (snapshot_removal_status) {
 659         case HS_PRE_RUNNING:
 660                 handle_pre_remove_exit(status);
 661                 ret = 1;
 662                 break;
 663         case HS_RUNNING:
 664                 ret = handle_rm_exit(status);
 665                 break;
 666         case HS_POST_RUNNING:
 667                 handle_post_remove_exit();
 668                 ret = 1;
 669                 break;
 670         default:
 671                 ret = -E_BUG;
 672         }
 673         if (snapshot_removal_status == HS_READY) {
 674                 free(s->name);
 675                 free(s);
 676                 snapshot_currently_being_removed = NULL;
 677         }
 678         remove_pid = 0;
 679         return ret;
 680 }
 681
 682 static int wait_for_remove_process(void)
 683 {
 684         int status, ret;
 685
 686         assert(remove_pid);
 687         assert(
 688                 snapshot_removal_status == HS_PRE_RUNNING ||
 689                 snapshot_removal_status == HS_RUNNING ||
 690                 snapshot_removal_status == HS_POST_RUNNING
 691         );
 692         ret = wait_for_process(remove_pid, &status);
 693         if (ret < 0)
 694                 return ret;
 695         return handle_remove_exit(status);
 696 }
 697
 698 static int handle_rsync_exit(int status)
 699 {
 700         int es, ret;
 701
 702         if (!WIFEXITED(status)) {
 703                 DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 704                 ret = -E_INVOLUNTARY_EXIT;
 705                 snapshot_creation_status = HS_READY;
 706                 goto out;
 707         }
 708         es = WEXITSTATUS(status);
 709         /*
 710          * Restart rsync on non-fatal errors:
 711          * 12: Error in rsync protocol data stream
 712          * 13: Errors with program diagnostics
 713          */
 714         if (es == 12 || es == 13) {
 715                 DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
 716                         (int)create_pid, es);
 717                 snapshot_creation_status = HS_NEEDS_RESTART;
 718                 next_snapshot_time = get_current_time() + 60;
 719                 ret = 1;
 720                 goto out;
 721         }
 722         if (es != 0 && es != 23 && es != 24) {
 723                 DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 724                 ret = -E_BAD_EXIT_CODE;
 725                 snapshot_creation_status = HS_READY;
 726                 goto out;
 727         }
 728         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 729         if (ret < 0)
 730                 goto out;
 731         snapshot_creation_status = HS_SUCCESS;
 732         free(name_of_reference_snapshot);
 733         name_of_reference_snapshot = NULL;
 734 out:
 735         create_process_stopped = 0;
 736         return ret;
 737 }
 738
 739 static int handle_pre_create_hook_exit(int status)
 740 {
 741         int es, ret;
 742         static int warn_count;
 743
 744         if (!WIFEXITED(status)) {
 745                 snapshot_creation_status = HS_READY;
 746                 ret = -E_INVOLUNTARY_EXIT;
 747                 goto out;
 748         }
 749         es = WEXITSTATUS(status);
 750         if (es) {
 751                 if (!warn_count--) {
 752                         DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
 753                                 conf.pre_create_hook_arg, es);
 754                         DSS_NOTICE_LOG("deferring snapshot creation...\n");
 755                         warn_count = 60; /* warn only once per hour */
 756                 }
 757                 next_snapshot_time = get_current_time() + 60;
 758                 snapshot_creation_status = HS_READY;
 759                 ret = 0;
 760                 goto out;
 761         }
 762         warn_count = 0;
 763         snapshot_creation_status = HS_PRE_SUCCESS;
 764         ret = 1;
 765 out:
 766         return ret;
 767 }
 768
 769 static int handle_sigchld(void)
 770 {
 771         pid_t pid;
 772         int status, ret = reap_child(&pid, &status);
 773
 774         if (ret <= 0)
 775                 return ret;
 776
 777         if (pid == create_pid) {
 778                 switch (snapshot_creation_status) {
 779                 case HS_PRE_RUNNING:
 780                         ret = handle_pre_create_hook_exit(status);
 781                         break;
 782                 case HS_RUNNING:
 783                         ret = handle_rsync_exit(status);
 784                         break;
 785                 case HS_POST_RUNNING:
 786                         snapshot_creation_status = HS_READY;
 787                         ret = 1;
 788                         break;
 789                 default:
 790                         DSS_EMERG_LOG("BUG: create can't die in status %d\n",
 791                                 snapshot_creation_status);
 792                         return -E_BUG;
 793                 }
 794                 create_pid = 0;
 795                 return ret;
 796         }
 797         if (pid == remove_pid) {
 798                 ret = handle_remove_exit(status);
 799                 if (ret < 0)
 800                         return ret;
 801                 return ret;
 802         }
 803         DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
 804         return -E_BUG;
 805 }
 806
 807 static int check_config(void)
 808 {
 809         if (conf.unit_interval_arg <= 0) {
 810                 DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
 811                 return -E_INVALID_NUMBER;
 812         }
 813         DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
 814         if (conf.num_intervals_arg <= 0) {
 815                 DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
 816                 return -E_INVALID_NUMBER;
 817         }
 818         DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
 819         return 1;
 820 }
 821
 822 /*
 823  * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
 824  * file was read successfully.
 825  */
 826 static int parse_config_file(int override)
 827 {
 828         int ret, config_file_exists;
 829         char *config_file;
 830         struct stat statbuf;
 831         char *old_logfile_arg = NULL;
 832         int old_daemon_given = 0;
 833
 834         if (conf.config_file_given)
 835                 config_file = dss_strdup(conf.config_file_arg);
 836         else {
 837                 char *home = get_homedir();
 838                 config_file = make_message("%s/.dssrc", home);
 839                 free(home);
 840         }
 841         if (override) { /* SIGHUP */
 842                 if (conf.logfile_given)
 843                         old_logfile_arg = dss_strdup(conf.logfile_arg);
 844                 old_daemon_given = conf.daemon_given;
 845         }
 846
 847         config_file_exists = !stat(config_file, &statbuf);
 848         if (!config_file_exists && conf.config_file_given) {
 849                 ret = -ERRNO_TO_DSS_ERROR(errno);
 850                 DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 851                 goto out;
 852         }
 853         if (config_file_exists) {
 854                 struct cmdline_parser_params params = {
 855                         .override = override,
 856                         .initialize = 0,
 857                         .check_required = 1,
 858                         .check_ambiguity = 0,
 859                         .print_errors = 1
 860                 };
 861                 if (override) { /* invalidate all rsync options */
 862                         int i;
 863
 864                         for (i = 0; i < conf.rsync_option_given; i++) {
 865                                 free(conf.rsync_option_arg[i]);
 866                                 conf.rsync_option_arg[i] = NULL;
 867                         }
 868                         conf.rsync_option_given = 0;
 869                 }
 870                 cmdline_parser_config_file(config_file, &conf, &params);
 871         }
 872         ret = check_config();
 873         if (ret < 0)
 874                 goto out;
 875         if (override) {
 876                 /* don't change daemon mode on SIGHUP */
 877                 conf.daemon_given = old_daemon_given;
 878                 close_log(logfile);
 879                 logfile = NULL;
 880                 if (conf.logfile_given)
 881                         free(old_logfile_arg);
 882                 else if (conf.daemon_given) { /* re-use old logfile */
 883                         conf.logfile_arg = old_logfile_arg;
 884                         conf.logfile_given = 1;
 885                 }
 886         }
 887         if (conf.logfile_given) {
 888                 logfile = open_log(conf.logfile_arg);
 889                 log_welcome(conf.loglevel_arg);
 890         }
 891         DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 892         ret = config_file_exists;
 893 out:
 894         free(config_file);
 895         if (ret < 0)
 896                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 897         return ret;
 898 }
 899
 900 static int change_to_dest_dir(void)
 901 {
 902         DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
 903         return dss_chdir(conf.dest_dir_arg);
 904 }
 905
 906 static void dump_dss_config(const char *msg)
 907 {
 908         if (conf.loglevel_arg > INFO)
 909                 return;
 910         DSS_INFO_LOG("%s\n", msg);
 911         cmdline_parser_dump(logfile? logfile : stderr, &conf);
 912 }
 913
 914 static int handle_sighup(void)
 915 {
 916         int ret;
 917
 918         DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
 919         dump_dss_config("current config");
 920         ret = parse_config_file(1);
 921         if (ret < 0)
 922                 return ret;
 923         dump_dss_config("new config");
 924         invalidate_next_snapshot_time();
 925         return change_to_dest_dir();
 926 }
 927
 928 static int handle_signal(void)
 929 {
 930         int sig, ret = next_signal();
 931
 932         if (ret <= 0)
 933                 goto out;
 934         sig = ret;
 935         switch (sig) {
 936         case SIGINT:
 937         case SIGTERM:
 938                 restart_create_process();
 939                 dss_kill(create_pid, SIGTERM, NULL);
 940                 dss_kill(remove_pid, SIGTERM, NULL);
 941                 ret = -E_SIGNAL;
 942                 break;
 943         case SIGHUP:
 944                 ret = handle_sighup();
 945                 break;
 946         case SIGCHLD:
 947                 ret = handle_sigchld();
 948                 break;
 949         }
 950 out:
 951         if (ret < 0)
 952                 DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 953         return ret;
 954 }
 955
 956 /*
 957  * We can not use rsync locally if the local user is different from the remote
 958  * user or if the src dir is not on the local host (or both).
 959  */
 960 static int use_rsync_locally(char *logname)
 961 {
 962         char *h = conf.remote_host_arg;
 963
 964         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
 965                 return 0;
 966         if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
 967                 return 0;
 968         return 1;
 969 }
 970
 971 static int rename_resume_snap(int64_t creation_time)
 972 {
 973         struct snapshot_list sl = {.num_snapshots = 0};
 974         struct snapshot *s;
 975         char *new_name = incomplete_name(creation_time);
 976         int ret;
 977
 978         ret = 0;
 979         if (conf.no_resume_given)
 980                 goto out;
 981         dss_get_snapshot_list(&sl);
 982         s = get_newest_snapshot(&sl);
 983         if (!s)
 984                 goto out;
 985         if ((s->flags & SS_COMPLETE) != 0) /* complete */
 986                 goto out;
 987         DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
 988         ret = dss_rename(s->name, new_name);
 989 out:
 990         if (ret >= 0)
 991                 DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
 992         free(new_name);
 993         free_snapshot_list(&sl);
 994         return ret;
 995 }
 996
 997 static void create_rsync_argv(char ***argv, int64_t *num)
 998 {
 999         char *logname;
1000         int i = 0, j;
1001         struct snapshot_list sl;
1002
1003         dss_get_snapshot_list(&sl);
1004         assert(!name_of_reference_snapshot);
1005         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1006         free_snapshot_list(&sl);
1007
1008         *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
1009         (*argv)[i++] = dss_strdup("rsync");
1010         (*argv)[i++] = dss_strdup("-aq");
1011         (*argv)[i++] = dss_strdup("--delete");
1012         for (j = 0; j < conf.rsync_option_given; j++)
1013                 (*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
1014         if (name_of_reference_snapshot) {
1015                 DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
1016                 (*argv)[i++] = make_message("--link-dest=../%s",
1017                         name_of_reference_snapshot);
1018         } else
1019                 DSS_INFO_LOG("no suitable reference snapshot found\n");
1020         logname = dss_logname();
1021         if (use_rsync_locally(logname))
1022                 (*argv)[i++] = dss_strdup(conf.source_dir_arg);
1023         else
1024                 (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
1025                         conf.remote_user_arg : logname,
1026                         conf.remote_host_arg, conf.source_dir_arg);
1027         free(logname);
1028         *num = get_current_time();
1029         (*argv)[i++] = incomplete_name(*num);
1030         (*argv)[i++] = NULL;
1031         for (j = 0; j < i; j++)
1032                 DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
1033 }
1034
1035 static void free_rsync_argv(char **argv)
1036 {
1037         int i;
1038
1039         if (!argv)
1040                 return;
1041         for (i = 0; argv[i]; i++)
1042                 free(argv[i]);
1043         free(argv);
1044 }
1045
1046 static int create_snapshot(char **argv)
1047 {
1048         int ret, fds[3] = {0, 0, 0};
1049
1050         ret = rename_resume_snap(current_snapshot_creation_time);
1051         if (ret < 0)
1052                 return ret;
1053         ret = dss_exec(&create_pid, argv[0], argv, fds);
1054         if (ret < 0)
1055                 return ret;
1056         snapshot_creation_status = HS_RUNNING;
1057         return ret;
1058 }
1059
1060 static int select_loop(void)
1061 {
1062         int ret;
1063         /* check every 60 seconds for free disk space */
1064         struct timeval tv;
1065         char **rsync_argv = NULL;
1066
1067         for (;;) {
1068                 fd_set rfds;
1069                 struct timeval *tvp;
1070
1071                 if (remove_pid)
1072                         tvp = NULL; /* sleep until rm hook/process dies */
1073                 else { /* sleep one minute */
1074                         tv.tv_sec = 60;
1075                         tv.tv_usec = 0;
1076                         tvp = &tv;
1077                 }
1078                 FD_ZERO(&rfds);
1079                 FD_SET(signal_pipe, &rfds);
1080                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1081                 if (ret < 0)
1082                         goto out;
1083                 if (FD_ISSET(signal_pipe, &rfds)) {
1084                         ret = handle_signal();
1085                         if (ret < 0)
1086                                 goto out;
1087                 }
1088                 if (remove_pid)
1089                         continue;
1090                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1091                         ret = exec_rm();
1092                         if (ret < 0)
1093                                 goto out;
1094                         continue;
1095                 }
1096                 if (snapshot_removal_status == HS_SUCCESS) {
1097                         ret = post_remove_hook();
1098                         if (ret < 0)
1099                                 goto out;
1100                         continue;
1101                 }
1102                 ret = try_to_free_disk_space();
1103                 if (ret < 0)
1104                         goto out;
1105                 if (snapshot_removal_status != HS_READY) {
1106                         stop_create_process();
1107                         continue;
1108                 }
1109                 restart_create_process();
1110                 switch (snapshot_creation_status) {
1111                 case HS_READY:
1112                         if (!next_snapshot_is_due())
1113                                 continue;
1114                         ret = pre_create_hook();
1115                         if (ret < 0)
1116                                 goto out;
1117                         continue;
1118                 case HS_PRE_RUNNING:
1119                 case HS_RUNNING:
1120                 case HS_POST_RUNNING:
1121                         continue;
1122                 case HS_PRE_SUCCESS:
1123                         if (!name_of_reference_snapshot) {
1124                                 free_rsync_argv(rsync_argv);
1125                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1126                         }
1127                         ret = create_snapshot(rsync_argv);
1128                         if (ret < 0)
1129                                 goto out;
1130                         continue;
1131                 case HS_NEEDS_RESTART:
1132                         if (!next_snapshot_is_due())
1133                                 continue;
1134                         ret = create_snapshot(rsync_argv);
1135                         if (ret < 0)
1136                                 goto out;
1137                         continue;
1138                 case HS_SUCCESS:
1139                         ret = post_create_hook();
1140                         if (ret < 0)
1141                                 goto out;
1142                         continue;
1143                 }
1144         }
1145 out:
1146         return ret;
1147 }
1148
1149 static void exit_hook(int exit_code)
1150 {
1151         int fds[3] = {0, 0, 0};
1152         char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
1153         pid_t pid;
1154
1155         DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
1156         dss_exec(&pid, conf.exit_hook_arg, argv, fds);
1157 }
1158
1159 static int com_run(void)
1160 {
1161         int ret;
1162
1163         if (conf.dry_run_given) {
1164                 DSS_ERROR_LOG("dry_run not supported by this command\n");
1165                 return -E_SYNTAX;
1166         }
1167         ret = install_sighandler(SIGHUP);
1168         if (ret < 0)
1169                 return ret;
1170         ret = select_loop();
1171         if (ret >= 0) /* impossible */
1172                 ret = -E_BUG;
1173         exit_hook(ret);
1174         return ret;
1175 }
1176
1177 static int com_prune(void)
1178 {
1179         int ret;
1180         struct snapshot_list sl;
1181         struct snapshot *victim;
1182         struct disk_space ds;
1183         const char *why;
1184
1185         ret = get_disk_space(".", &ds);
1186         if (ret < 0)
1187                 return ret;
1188         log_disk_space(&ds);
1189         dss_get_snapshot_list(&sl);
1190         why = "outdated";
1191         victim = find_outdated_snapshot(&sl);
1192         if (victim)
1193                 goto rm;
1194         why = "redundant";
1195         victim = find_redundant_snapshot(&sl);
1196         if (victim)
1197                 goto rm;
1198         ret = 0;
1199         goto out;
1200 rm:
1201         if (conf.dry_run_given) {
1202                 dss_msg("%s snapshot %s (interval = %i)\n",
1203                         why, victim->name, victim->interval);
1204                 ret = 0;
1205                 goto out;
1206         }
1207         ret = pre_remove_hook(victim, why);
1208         if (ret < 0)
1209                 goto out;
1210         if (snapshot_removal_status == HS_PRE_RUNNING) {
1211                 ret = wait_for_remove_process();
1212                 if (ret < 0)
1213                         goto out;
1214                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1215                         goto out;
1216         }
1217         ret = exec_rm();
1218         if (ret < 0)
1219                 goto out;
1220         ret = wait_for_remove_process();
1221         if (ret < 0)
1222                 goto out;
1223         if (snapshot_removal_status != HS_SUCCESS)
1224                 goto out;
1225         ret = post_remove_hook();
1226         if (ret < 0)
1227                 goto out;
1228         if (snapshot_removal_status != HS_POST_RUNNING)
1229                 goto out;
1230         ret = wait_for_remove_process();
1231         if (ret < 0)
1232                 goto out;
1233         ret = 1;
1234 out:
1235         free_snapshot_list(&sl);
1236         return ret;
1237 }
1238
1239 static int com_create(void)
1240 {
1241         int ret, status;
1242         char **rsync_argv;
1243
1244         if (conf.dry_run_given) {
1245                 int i;
1246                 char *msg = NULL;
1247                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1248                 for (i = 0; rsync_argv[i]; i++) {
1249                         char *tmp = msg;
1250                         msg = make_message("%s%s%s", tmp? tmp : "",
1251                                 tmp? " " : "", rsync_argv[i]);
1252                         free(tmp);
1253                 }
1254                 free_rsync_argv(rsync_argv);
1255                 dss_msg("%s\n", msg);
1256                 free(msg);
1257                 return 1;
1258         }
1259         ret = pre_create_hook();
1260         if (ret < 0)
1261                 return ret;
1262         if (create_pid) {
1263                 ret = wait_for_process(create_pid, &status);
1264                 if (ret < 0)
1265                         return ret;
1266                 ret = handle_pre_create_hook_exit(status);
1267                 if (ret <= 0) /* error, or pre-create failed */
1268                         return ret;
1269         }
1270         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1271         ret = create_snapshot(rsync_argv);
1272         if (ret < 0)
1273                 goto out;
1274         ret = wait_for_process(create_pid, &status);
1275         if (ret < 0)
1276                 goto out;
1277         ret = handle_rsync_exit(status);
1278         if (ret < 0)
1279                 goto out;
1280         post_create_hook();
1281         if (create_pid)
1282                 ret = wait_for_process(create_pid, &status);
1283 out:
1284         free_rsync_argv(rsync_argv);
1285         return ret;
1286 }
1287
1288 static int com_ls(void)
1289 {
1290         int i;
1291         struct snapshot_list sl;
1292         struct snapshot *s;
1293
1294         dss_get_snapshot_list(&sl);
1295         FOR_EACH_SNAPSHOT(s, i, &sl) {
1296                 int64_t d = 0;
1297                 if (s->flags & SS_COMPLETE)
1298                         d = (s->completion_time - s->creation_time) / 60;
1299                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
1300         };
1301         free_snapshot_list(&sl);
1302         return 1;
1303 }
1304
1305 static int setup_signal_handling(void)
1306 {
1307         int ret;
1308
1309         DSS_INFO_LOG("setting up signal handlers\n");
1310         signal_pipe = signal_init(); /* always successful */
1311         ret = install_sighandler(SIGINT);
1312         if (ret < 0)
1313                 return ret;
1314         ret = install_sighandler(SIGTERM);
1315         if (ret < 0)
1316                 return ret;
1317         return install_sighandler(SIGCHLD);
1318 }
1319
1320 /**
1321  * The main function of dss.
1322  *
1323  * \param argc Usual argument count.
1324  * \param argv Usual argument vector.
1325  */
1326 int main(int argc, char **argv)
1327 {
1328         int ret;
1329         struct cmdline_parser_params params = {
1330                 .override = 0,
1331                 .initialize = 1,
1332                 .check_required = 0,
1333                 .check_ambiguity = 0,
1334                 .print_errors = 1
1335         };
1336
1337         cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1338         ret = parse_config_file(0);
1339         if (ret < 0)
1340                 goto out;
1341         if (ret == 0) { /* no config file given */
1342                 /*
1343                  * Parse the command line options again, but this time check
1344                  * that all required options are given.
1345                  */
1346                 params = (struct cmdline_parser_params) {
1347                         .override = 1,
1348                         .initialize = 1,
1349                         .check_required = 1,
1350                         .check_ambiguity = 1,
1351                         .print_errors = 1
1352                 };
1353                 cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
1354         }
1355         if (conf.daemon_given)
1356                 daemon_init();
1357         dump_dss_config("dss configuration");
1358         ret = change_to_dest_dir();
1359         if (ret < 0)
1360                 goto out;
1361         ret = setup_signal_handling();
1362         if (ret < 0)
1363                 goto out;
1364         ret = call_command_handler();
1365 out:
1366         if (ret < 0)
1367                 DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
1368         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1369 }