dss.c

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #include <string.h>
   3 #include <stdlib.h>
   4 #include <stdio.h>
   5 #include <stdarg.h>
   6 #include <assert.h>
   7 #include <errno.h>
   8 #include <sys/types.h>
   9 #include <signal.h>
  10 #include <ctype.h>
  11 #include <stdbool.h>
  12 #include <sys/stat.h>
  13 #include <unistd.h>
  14 #include <inttypes.h>
  15 #include <sys/time.h>
  16 #include <time.h>
  17 #include <sys/wait.h>
  18 #include <fnmatch.h>
  19 #include <limits.h>
  20 #include <fcntl.h>
  21 #include <lopsub.h>
  22 #include <sys/mman.h>
  23
  24 #include "gcc-compat.h"
  25 #include "log.h"
  26 #include "str.h"
  27 #include "err.h"
  28 #include "file.h"
  29 #include "exec.h"
  30 #include "daemon.h"
  31 #include "sig.h"
  32 #include "df.h"
  33 #include "tv.h"
  34 #include "snap.h"
  35 #include "ipc.h"
  36 #include "dss.lsg.h"
  37
  38 #define CMD_PTR(_cname) lls_cmd(LSG_DSS_CMD_ ## _cname, dss_suite)
  39 #define OPT_RESULT(_cname, _oname) (lls_opt_result(\
  40         LSG_DSS_ ## _cname ## _OPT_ ## _oname, (CMD_PTR(_cname) == CMD_PTR(DSS))? lpr : sublpr))
  41 #define OPT_GIVEN(_cname, _oname) (lls_opt_given(OPT_RESULT(_cname, _oname)))
  42 #define OPT_STRING_VAL(_cname, _oname) (lls_string_val(0, \
  43         OPT_RESULT(_cname, _oname)))
  44 #define OPT_UINT32_VAL(_cname, _oname) (lls_uint32_val(0, \
  45                 OPT_RESULT(_cname, _oname)))
  46
  47 struct dss_user_data {int (*handler)(void);};
  48 #define EXPORT_CMD_HANDLER(_cmd) const struct dss_user_data \
  49         lsg_dss_com_ ## _cmd ## _user_data = { \
  50                 .handler = com_ ## _cmd \
  51         };
  52
  53 /*
  54  * Command line and active options. We need to keep a copy of the parsed
  55  * command line options for the SIGHUP case where we merge the command line
  56  * options and the new config file options.
  57  */
  58 static struct lls_parse_result *cmdline_lpr, *lpr;
  59
  60 /** Parsed subcommand options. */
  61 static struct lls_parse_result *cmdline_sublpr, *sublpr;
  62 /* The executing subcommand (NULL at startup). */
  63 static const struct lls_command *subcmd;
  64 /** Wether daemon_init() was called. */
  65 static bool daemonized;
  66 /** Non-NULL if we log to a file. */
  67 static FILE *logfile;
  68 /** The read end of the signal pipe */
  69 static int signal_pipe;
  70 /** Process id of current pre-create-hook/rsync/post-create-hook process. */
  71 static pid_t create_pid;
  72 /** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
  73 static int create_process_stopped;
  74 /** How many times in a row the rsync command failed. */
  75 static int num_consecutive_rsync_errors;
  76 /** Process id of current pre-remove/rm/post-remove process. */
  77 static pid_t remove_pid;
  78 /** When the next snapshot is due. */
  79 static int64_t next_snapshot_time;
  80 /** When to try to remove something. */
  81 static struct timeval next_removal_check;
  82 /** Creation time of the snapshot currently being created. */
  83 static int64_t current_snapshot_creation_time;
  84 /* Set by the pre-rm hook, cleared by handle_remove_exit(). */
  85 struct snapshot *snapshot_currently_being_removed;
  86 /** Needed by the post-create hook. */
  87 static char *path_to_last_complete_snapshot;
  88 static char *name_of_reference_snapshot;
  89 /** \sa \ref snap.h for details. */
  90 enum hook_status snapshot_creation_status;
  91 /** \sa \ref snap.h for details. */
  92 enum hook_status snapshot_removal_status;
  93
  94
  95 DEFINE_DSS_ERRLIST;
  96 static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
  97
  98 /* may be called with ds == NULL. */
  99 static int disk_space_low(struct disk_space *ds)
 100 {
 101         struct disk_space ds_struct;
 102         uint32_t val;
 103
 104         if (!ds) {
 105                 int ret = get_disk_space(".", &ds_struct);
 106                 if (ret < 0)
 107                         return ret;
 108                 ds = &ds_struct;
 109         }
 110         val = OPT_UINT32_VAL(DSS, MIN_FREE_MB);
 111         if (val != 0)
 112                 if (ds->free_mb < val)
 113                         return 1;
 114         val = OPT_UINT32_VAL(DSS, MIN_FREE_PERCENT);
 115         if (val != 0)
 116                 if (ds->percent_free < val)
 117                         return 1;
 118         val = OPT_UINT32_VAL(DSS, MIN_FREE_PERCENT_INODES);
 119         if (val != 0)
 120                 if (ds->percent_free_inodes < val)
 121                         return 1;
 122         return 0;
 123 }
 124
 125 static void dump_dss_config(const char *msg)
 126 {
 127         const char dash[] = "-----------------------------";
 128         char *lopsub_dump;
 129         int ret;
 130         FILE *log = logfile? logfile : stderr;
 131         struct disk_space ds;
 132         int64_t now = get_current_time();
 133
 134         if (OPT_UINT32_VAL(DSS, LOGLEVEL) > INFO)
 135                 return;
 136
 137         fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
 138         fprintf(log, "\n*** disk space ***\n\n");
 139         ret = get_disk_space(".", &ds);
 140         if (ret >= 0) {
 141                 DSS_INFO_LOG(("disk space low: %s\n", disk_space_low(&ds)?
 142                         "yes" : "no"));
 143                 log_disk_space(&ds);
 144         } else
 145                 DSS_ERROR_LOG(("can not get free disk space: %s\n",
 146                         dss_strerror(-ret)));
 147
 148         /* we continue on errors from get_disk_space */
 149
 150         fprintf(log, "\n*** non-default options ***\n\n");
 151         lopsub_dump = lls_dump_parse_result(lpr, CMD_PTR(DSS), true);
 152         fprintf(log, "%s", lopsub_dump);
 153         free(lopsub_dump);
 154         fprintf(log, "\n*** non-default options for \"run\" ***\n\n");
 155         lopsub_dump = lls_dump_parse_result(lpr, CMD_PTR(RUN), true);
 156         fprintf(log, "%s", lopsub_dump);
 157         free(lopsub_dump);
 158         fprintf(log, "\n*** internal state ***\n\n");
 159         fprintf(log,
 160                 "pid: %d\n"
 161                 "logfile: %s\n"
 162                 "snapshot_currently_being_removed: %s\n"
 163                 "path_to_last_complete_snapshot: %s\n"
 164                 "reference_snapshot: %s\n"
 165                 "snapshot_creation_status: %s\n"
 166                 "snapshot_removal_status: %s\n"
 167                 "num_consecutive_rsync_errors: %d\n"
 168                 ,
 169                 (int) getpid(),
 170                 logfile? OPT_STRING_VAL(RUN, LOGFILE) : "stderr",
 171                 snapshot_currently_being_removed?
 172                         snapshot_currently_being_removed->name : "(none)",
 173                 path_to_last_complete_snapshot?
 174                         path_to_last_complete_snapshot : "(none)",
 175                 name_of_reference_snapshot?
 176                         name_of_reference_snapshot : "(none)",
 177                 hook_status_description[snapshot_creation_status],
 178                 hook_status_description[snapshot_removal_status],
 179                 num_consecutive_rsync_errors
 180         );
 181         if (create_pid != 0)
 182                 fprintf(log,
 183                         "create_pid: %" PRId32 "\n"
 184                         "create process is %sstopped\n"
 185                         ,
 186                         create_pid,
 187                         create_process_stopped? "" : "not "
 188                 );
 189         if (remove_pid != 0)
 190                 fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
 191         if (next_snapshot_time != 0)
 192                 fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
 193                         next_snapshot_time - now);
 194         if (current_snapshot_creation_time != 0)
 195                 fprintf(log, "current_snapshot_creation_time: %"
 196                         PRId64 " (%" PRId64 " seconds ago)\n",
 197                         current_snapshot_creation_time,
 198                         now - current_snapshot_creation_time
 199                 );
 200         if (next_removal_check.tv_sec != 0) {
 201                 fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
 202                         (long long unsigned)next_removal_check.tv_sec,
 203                         now - (long long unsigned)next_removal_check.tv_sec
 204                 );
 205
 206         }
 207         fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
 208 }
 209
 210 static int loglevel = -1;
 211 static const char *location_file = NULL;
 212 static int         location_line = -1;
 213 static const char *location_func = NULL;
 214
 215 void dss_log_set_params(int ll, const char *file, int line, const char *func)
 216 {
 217         loglevel = ll;
 218         location_file = file;
 219         location_line = line;
 220         location_func = func;
 221 }
 222
 223 /**
 224  * The log function of dss.
 225  *
 226  * \param ll Loglevel.
 227  * \param fml Usual format string.
 228  *
 229  * All DSS_XXX_LOG() macros use this function.
 230  */
 231 __printf_1_2 void dss_log(const char* fmt,...)
 232 {
 233         va_list argp;
 234         FILE *outfd;
 235         struct tm *tm;
 236         time_t t1;
 237         char str[255] = "";
 238         int lpr_ll = lpr? OPT_UINT32_VAL(DSS, LOGLEVEL) : WARNING;
 239
 240         if (loglevel < lpr_ll)
 241                 return;
 242         outfd = logfile? logfile : stderr;
 243         if (subcmd == CMD_PTR(RUN)) {
 244                 time(&t1);
 245                 tm = localtime(&t1);
 246                 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
 247                 fprintf(outfd, "%s ", str);
 248                 if (lpr_ll <= INFO)
 249                         fprintf(outfd, "%i: ", loglevel);
 250         }
 251         if (subcmd == CMD_PTR(RUN))
 252 #ifdef DSS_NO_FUNC_NAMES
 253                 fprintf(outfd, "%s:%d: ", location_file, location_line);
 254 #else
 255                 fprintf(outfd, "%s: ", location_func);
 256 #endif
 257         va_start(argp, fmt);
 258         vfprintf(outfd, fmt, argp);
 259         va_end(argp);
 260 }
 261
 262 /**
 263  * Print a message either to stdout or to the log file.
 264  */
 265 static __printf_1_2 void dss_msg(const char* fmt,...)
 266 {
 267         FILE *outfd = logfile? logfile : stdout;
 268         va_list argp;
 269         va_start(argp, fmt);
 270         vfprintf(outfd, fmt, argp);
 271         va_end(argp);
 272 }
 273
 274 static char *get_config_file_name(void)
 275 {
 276         char *home, *config_file;
 277
 278         if (OPT_GIVEN(DSS, CONFIG_FILE))
 279                 return dss_strdup(OPT_STRING_VAL(DSS, CONFIG_FILE));
 280         home = get_homedir();
 281         config_file = make_message("%s/.dssrc", home);
 282         free(home);
 283         return config_file;
 284 }
 285
 286 static int send_signal(int sig, bool wait)
 287 {
 288         pid_t pid;
 289         char *config_file = get_config_file_name();
 290         int ret = get_dss_pid(config_file, &pid);
 291         unsigned ms = 32;
 292         struct timespec ts;
 293
 294         free(config_file);
 295         if (ret < 0)
 296                 return ret;
 297         if (OPT_GIVEN(DSS, DRY_RUN)) {
 298                 dss_msg("%d\n", (int)pid);
 299                 return 0;
 300         }
 301         DSS_NOTICE_LOG(("sending signal %d to pid %d\n", sig, (int)pid));
 302         ret = kill(pid, sig);
 303         if (ret < 0)
 304                 return -ERRNO_TO_DSS_ERROR(errno);
 305         if (!wait)
 306                 return 1;
 307         while (ms < 5000) {
 308                 ts.tv_sec = ms / 1000;
 309                 ts.tv_nsec = (ms % 1000) * 1000 * 1000;
 310                 ret = nanosleep(&ts, NULL);
 311                 if (ret < 0)
 312                         return -ERRNO_TO_DSS_ERROR(errno);
 313                 ret = kill(pid, 0);
 314                 if (ret < 0) {
 315                         if (errno != ESRCH)
 316                                 return -ERRNO_TO_DSS_ERROR(errno);
 317                         return 1;
 318                 }
 319                 ms *= 2;
 320         }
 321         return -E_KILL_TIMEOUT;
 322 }
 323
 324 struct signal_info {
 325         const char * const name;
 326         int num;
 327 };
 328
 329 /*
 330  * The table below was taken 2016 from proc/sig.c of procps-3.2.8. Copyright
 331  * 1998-2003 by Albert Cahalan, GPLv2.
 332  */
 333 static const struct signal_info signal_table[] = {
 334         {"ABRT",   SIGABRT},  /* IOT */
 335         {"ALRM",   SIGALRM},
 336         {"BUS",    SIGBUS},
 337         {"CHLD",   SIGCHLD},  /* CLD */
 338         {"CONT",   SIGCONT},
 339         {"FPE",    SIGFPE},
 340         {"HUP",    SIGHUP},
 341         {"ILL",    SIGILL},
 342         {"INT",    SIGINT},
 343         {"KILL",   SIGKILL},
 344         {"PIPE",   SIGPIPE},
 345 #ifdef SIGPOLL
 346         {"POLL",   SIGPOLL},  /* IO */
 347 #endif
 348         {"PROF",   SIGPROF},
 349 #ifdef SIGPWR
 350         {"PWR",    SIGPWR},
 351 #endif
 352         {"QUIT",   SIGQUIT},
 353         {"SEGV",   SIGSEGV},
 354 #ifdef SIGSTKFLT
 355         {"STKFLT", SIGSTKFLT},
 356 #endif
 357         {"STOP",   SIGSTOP},
 358         {"SYS",    SIGSYS},   /* UNUSED */
 359         {"TERM",   SIGTERM},
 360         {"TRAP",   SIGTRAP},
 361         {"TSTP",   SIGTSTP},
 362         {"TTIN",   SIGTTIN},
 363         {"TTOU",   SIGTTOU},
 364         {"URG",    SIGURG},
 365         {"USR1",   SIGUSR1},
 366         {"USR2",   SIGUSR2},
 367         {"VTALRM", SIGVTALRM},
 368         {"WINCH",  SIGWINCH},
 369         {"XCPU",   SIGXCPU},
 370         {"XFSZ",   SIGXFSZ}
 371 };
 372
 373 #define SIGNAL_TABLE_SIZE (sizeof(signal_table) / sizeof(signal_table[0]))
 374 #ifndef SIGRTMAX
 375 #define SIGRTMAX 64
 376 #endif
 377
 378 static int com_kill(void)
 379 {
 380         bool w_given = OPT_GIVEN(KILL, WAIT);
 381         const char *arg = OPT_STRING_VAL(KILL, SIGNAL);
 382         int ret, i;
 383
 384         if (*arg >= '0' && *arg <= '9') {
 385                 int64_t val;
 386                 ret = dss_atoi64(arg, &val);
 387                 if (ret < 0)
 388                         return ret;
 389                 if (val < 0 || val > SIGRTMAX)
 390                         return -ERRNO_TO_DSS_ERROR(EINVAL);
 391                 return send_signal(val, w_given);
 392         }
 393         if (strncasecmp(arg, "sig", 3) == 0)
 394                 arg += 3;
 395         if (strcasecmp(arg, "CLD") == 0)
 396                 return send_signal(SIGCHLD, w_given);
 397         if (strcasecmp(arg, "IOT") == 0)
 398                 return send_signal(SIGABRT, w_given);
 399         for (i = 0; i < SIGNAL_TABLE_SIZE; i++)
 400                 if (strcasecmp(arg, signal_table[i].name) == 0)
 401                         return send_signal(signal_table[i].num, w_given);
 402         DSS_ERROR_LOG(("invalid sigspec: %s\n", arg));
 403         return -ERRNO_TO_DSS_ERROR(EINVAL);
 404 }
 405 EXPORT_CMD_HANDLER(kill);
 406
 407 static void dss_get_snapshot_list(struct snapshot_list *sl)
 408 {
 409         get_snapshot_list(sl, OPT_UINT32_VAL(DSS, UNIT_INTERVAL),
 410                 OPT_UINT32_VAL(DSS, NUM_INTERVALS));
 411 }
 412
 413 static int64_t compute_next_snapshot_time(void)
 414 {
 415         int64_t x = 0, now = get_current_time(), unit_interval
 416                 = 24 * 3600 * OPT_UINT32_VAL(DSS, UNIT_INTERVAL), ret,
 417                 last_completion_time;
 418         unsigned wanted = desired_number_of_snapshots(0,
 419                 OPT_UINT32_VAL(DSS, NUM_INTERVALS)),
 420                 num_complete = 0;
 421         int i;
 422         struct snapshot *s = NULL;
 423         struct snapshot_list sl;
 424
 425         dss_get_snapshot_list(&sl);
 426         FOR_EACH_SNAPSHOT(s, i, &sl) {
 427                 if (!(s->flags & SS_COMPLETE))
 428                         continue;
 429                 num_complete++;
 430                 x += s->completion_time - s->creation_time;
 431                 last_completion_time = s->completion_time;
 432         }
 433         assert(x >= 0);
 434
 435         ret = now;
 436         if (num_complete == 0)
 437                 goto out;
 438         x /= num_complete; /* avg time to create one snapshot */
 439         if (unit_interval < x * wanted) /* oops, no sleep at all */
 440                 goto out;
 441         ret = last_completion_time + unit_interval / wanted - x;
 442 out:
 443         free_snapshot_list(&sl);
 444         return ret;
 445 }
 446
 447 static inline void invalidate_next_snapshot_time(void)
 448 {
 449         next_snapshot_time = 0;
 450 }
 451
 452 static inline int next_snapshot_time_is_valid(void)
 453 {
 454         return next_snapshot_time != 0;
 455 }
 456
 457 static int next_snapshot_is_due(void)
 458 {
 459         int64_t now = get_current_time();
 460
 461         if (!next_snapshot_time_is_valid())
 462                 next_snapshot_time = compute_next_snapshot_time();
 463         if (next_snapshot_time <= now) {
 464                 DSS_DEBUG_LOG(("next snapshot: now\n"));
 465                 return 1;
 466         }
 467         DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
 468                 next_snapshot_time - now));
 469         return 0;
 470 }
 471
 472 static void pre_create_hook(void)
 473 {
 474         assert(snapshot_creation_status == HS_READY);
 475         /* make sure that the next snapshot time will be recomputed */
 476         invalidate_next_snapshot_time();
 477         DSS_DEBUG_LOG(("executing %s\n", OPT_STRING_VAL(DSS, PRE_CREATE_HOOK)));
 478         dss_exec_cmdline_pid(&create_pid, OPT_STRING_VAL(DSS, PRE_CREATE_HOOK));
 479         snapshot_creation_status = HS_PRE_RUNNING;
 480 }
 481
 482 static void pre_remove_hook(struct snapshot *s, const char *why)
 483 {
 484         char *cmd;
 485
 486         if (!s)
 487                 return;
 488         DSS_DEBUG_LOG(("%s snapshot %s\n", why, s->name));
 489         assert(snapshot_removal_status == HS_READY);
 490         assert(remove_pid == 0);
 491         assert(!snapshot_currently_being_removed);
 492
 493         snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
 494         *snapshot_currently_being_removed = *s;
 495         snapshot_currently_being_removed->name = dss_strdup(s->name);
 496
 497         cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, PRE_REMOVE_HOOK),
 498                 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
 499         DSS_DEBUG_LOG(("executing %s\n", cmd));
 500         dss_exec_cmdline_pid(&remove_pid, cmd);
 501         free(cmd);
 502         snapshot_removal_status = HS_PRE_RUNNING;
 503 }
 504
 505 static int exec_rm(void)
 506 {
 507         struct snapshot *s = snapshot_currently_being_removed;
 508         char *new_name = being_deleted_name(s);
 509         char *argv[4];
 510         int ret;
 511
 512         argv[0] = "rm";
 513         argv[1] = "-rf";
 514         argv[2] = new_name;
 515         argv[3] = NULL;
 516
 517         assert(snapshot_removal_status == HS_PRE_SUCCESS);
 518         assert(remove_pid == 0);
 519
 520         DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
 521         ret = dss_rename(s->name, new_name);
 522         if (ret < 0)
 523                 goto out;
 524         dss_exec(&remove_pid, argv[0], argv);
 525         snapshot_removal_status = HS_RUNNING;
 526 out:
 527         free(new_name);
 528         return ret;
 529 }
 530
 531 static int snapshot_is_being_created(struct snapshot *s)
 532 {
 533         return s->creation_time == current_snapshot_creation_time;
 534 }
 535
 536 static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 537 {
 538         struct snapshot *s;
 539         int i;
 540
 541         DSS_DEBUG_LOG(("looking for old incomplete snapshots\n"));
 542         FOR_EACH_SNAPSHOT(s, i, sl) {
 543                 if (snapshot_is_being_created(s))
 544                         continue;
 545                 /*
 546                  * We know that no rm is currently running, so if s is marked
 547                  * as being deleted, a previously started rm must have failed.
 548                  */
 549                 if (s->flags & SS_BEING_DELETED)
 550                         return s;
 551
 552                 if (s->flags & SS_COMPLETE) /* good snapshot */
 553                         continue;
 554                 /*
 555                  * This snapshot is incomplete and it is not the snapshot
 556                  * currently being created. However, we must not remove it if
 557                  * rsync is about to be restarted. As only the newest snapshot
 558                  * can be restarted, this snapshot is orphaned if it is not the
 559                  * newest snapshot or if we are not about to restart rsync.
 560                  */
 561                 if (get_newest_snapshot(sl) != s)
 562                         return s;
 563                 if (snapshot_creation_status != HS_NEEDS_RESTART)
 564                         return s;
 565         }
 566         /* no orphaned snapshots */
 567         return NULL;
 568 }
 569
 570 static int is_reference_snapshot(struct snapshot *s)
 571 {
 572         if (!name_of_reference_snapshot)
 573                 return 0;
 574         return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 575 }
 576
 577 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 578 {
 579         int i, interval;
 580         struct snapshot *s;
 581         unsigned missing = 0;
 582         uint32_t N = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
 583
 584         DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
 585         for (interval = N - 1; interval >= 0; interval--) {
 586                 unsigned keep = desired_number_of_snapshots(interval, N);
 587                 unsigned num = sl->interval_count[interval];
 588                 struct snapshot *victim = NULL, *prev = NULL;
 589                 int64_t score = LONG_MAX;
 590
 591                 if (keep >= num)
 592                         missing += keep - num;
 593                 if (keep + missing >= num)
 594                         continue;
 595                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 596                 FOR_EACH_SNAPSHOT(s, i, sl) {
 597                         int64_t this_score;
 598
 599                         if (snapshot_is_being_created(s))
 600                                 continue;
 601                         if (is_reference_snapshot(s))
 602                                 continue;
 603                         if (s->interval > interval) {
 604                                 prev = s;
 605                                 continue;
 606                         }
 607                         if (s->interval < interval)
 608                                 break;
 609                         if (!victim) {
 610                                 victim = s;
 611                                 prev = s;
 612                                 continue;
 613                         }
 614                         assert(prev);
 615                         /* check if s is a better victim */
 616                         this_score = s->creation_time - prev->creation_time;
 617                         assert(this_score >= 0);
 618                         if (this_score < score) {
 619                                 score = this_score;
 620                                 victim = s;
 621                         }
 622                         prev = s;
 623                 }
 624                 assert(victim);
 625                 return victim;
 626         }
 627         return NULL;
 628 }
 629
 630 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 631 {
 632         int i;
 633         struct snapshot *s;
 634
 635         DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
 636                 OPT_UINT32_VAL(DSS, NUM_INTERVALS)));
 637         FOR_EACH_SNAPSHOT(s, i, sl) {
 638                 if (snapshot_is_being_created(s))
 639                         continue;
 640                 if (is_reference_snapshot(s))
 641                         continue;
 642                 if (s->interval < OPT_UINT32_VAL(DSS, NUM_INTERVALS))
 643                         continue;
 644                 return s;
 645         }
 646         return NULL;
 647 }
 648
 649 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 650 {
 651         int i, num_complete;
 652         struct snapshot *s, *ref = NULL;
 653
 654         num_complete = num_complete_snapshots(sl);
 655         if (num_complete <= OPT_UINT32_VAL(DSS, MIN_COMPLETE))
 656                 return NULL;
 657         FOR_EACH_SNAPSHOT(s, i, sl) {
 658                 if (snapshot_is_being_created(s))
 659                         continue;
 660                 if (is_reference_snapshot(s)) { /* avoid this one */
 661                         ref = s;
 662                         continue;
 663                 }
 664                 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 665                 return s;
 666         }
 667         assert(ref);
 668         DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
 669         return ref;
 670 }
 671
 672 static int rename_incomplete_snapshot(int64_t start)
 673 {
 674         char *old_name;
 675         int ret;
 676         int64_t now;
 677
 678         /*
 679          * We don't want the dss_rename() below to fail with EEXIST because the
 680          * last complete snapshot was created (and completed) in the same
 681          * second as this one.
 682          */
 683         while ((now = get_current_time()) == start)
 684                 sleep(1);
 685         free(path_to_last_complete_snapshot);
 686         ret = complete_name(start, now, &path_to_last_complete_snapshot);
 687         if (ret < 0)
 688                 return ret;
 689         old_name = incomplete_name(start);
 690         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 691         if (ret >= 0)
 692                 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
 693                         path_to_last_complete_snapshot));
 694         free(old_name);
 695         return ret;
 696 }
 697
 698 static int try_to_free_disk_space(void)
 699 {
 700         int ret;
 701         struct snapshot_list sl;
 702         struct snapshot *victim;
 703         struct timeval now;
 704         const char *why;
 705         int low_disk_space;
 706
 707         ret = disk_space_low(NULL);
 708         if (ret < 0)
 709                 return ret;
 710         low_disk_space = ret;
 711         gettimeofday(&now, NULL);
 712         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 713                 return 0;
 714         if (!low_disk_space) {
 715                 if (OPT_GIVEN(DSS, KEEP_REDUNDANT))
 716                         return 0;
 717                 if (snapshot_creation_status != HS_READY)
 718                         return 0;
 719                 if (next_snapshot_is_due())
 720                         return 0;
 721         }
 722         /*
 723          * Idle and --keep_redundant not given, or low disk space. Look at
 724          * existing snapshots.
 725          */
 726         dss_get_snapshot_list(&sl);
 727         ret = 0;
 728         /*
 729          * Don't remove anything if there is free space and we have fewer
 730          * snapshots than configured, plus one. This way there is always one
 731          * snapshot that can be recycled.
 732          */
 733         if (!low_disk_space && sl.num_snapshots <=
 734                         1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
 735                 goto out;
 736         why = "outdated";
 737         victim = find_outdated_snapshot(&sl);
 738         if (victim)
 739                 goto remove;
 740         why = "redundant";
 741         victim = find_redundant_snapshot(&sl);
 742         if (victim)
 743                 goto remove;
 744         why = "orphaned";
 745         victim = find_orphaned_snapshot(&sl);
 746         if (victim)
 747                 goto remove;
 748         /* try harder only if disk space is low */
 749         if (!low_disk_space)
 750                 goto out;
 751         DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 752         why = "oldest";
 753         victim = find_oldest_removable_snapshot(&sl);
 754         if (victim)
 755                 goto remove;
 756         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 757         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 758         goto out;
 759 remove:
 760         pre_remove_hook(victim, why);
 761 out:
 762         free_snapshot_list(&sl);
 763         return ret;
 764 }
 765
 766 static void post_create_hook(void)
 767 {
 768         char *cmd = make_message("%s %s/%s",
 769                 OPT_STRING_VAL(DSS, POST_CREATE_HOOK),
 770                 OPT_STRING_VAL(DSS, DEST_DIR), path_to_last_complete_snapshot);
 771         DSS_NOTICE_LOG(("executing %s\n", cmd));
 772         dss_exec_cmdline_pid(&create_pid, cmd);
 773         free(cmd);
 774         snapshot_creation_status = HS_POST_RUNNING;
 775 }
 776
 777 static void post_remove_hook(void)
 778 {
 779         char *cmd;
 780         struct snapshot *s = snapshot_currently_being_removed;
 781
 782         assert(s);
 783
 784         cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, POST_REMOVE_HOOK),
 785                 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
 786         DSS_NOTICE_LOG(("executing %s\n", cmd));
 787         dss_exec_cmdline_pid(&remove_pid, cmd);
 788         free(cmd);
 789         snapshot_removal_status = HS_POST_RUNNING;
 790 }
 791
 792 static void dss_kill(pid_t pid, int sig, const char *msg)
 793 {
 794         const char *signame, *process_name;
 795
 796         if (pid == 0)
 797                 return;
 798         switch (sig) {
 799         case SIGTERM: signame = "TERM"; break;
 800         case SIGSTOP: signame = "STOP"; break;
 801         case SIGCONT: signame = "CONT"; break;
 802         default: signame = "????";
 803         }
 804
 805         if (pid == create_pid)
 806                 process_name = "create";
 807         else if (pid == remove_pid)
 808                 process_name = "remove";
 809         else process_name = "??????";
 810
 811         if (msg)
 812                 DSS_INFO_LOG(("%s\n", msg));
 813         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 814                 sig, signame, (int)pid, process_name));
 815         if (kill(pid, sig) >= 0)
 816                 return;
 817         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 818                 sig, signame, (int)pid, process_name));
 819 }
 820
 821 static void stop_create_process(void)
 822 {
 823         if (create_process_stopped)
 824                 return;
 825         dss_kill(create_pid, SIGSTOP, "suspending create process");
 826         create_process_stopped = 1;
 827 }
 828
 829 static void restart_create_process(void)
 830 {
 831         if (!create_process_stopped)
 832                 return;
 833         dss_kill(create_pid, SIGCONT, "resuming create process");
 834         create_process_stopped = 0;
 835 }
 836
 837 /**
 838  * Print a log message about the exit status of a child.
 839  */
 840 static void log_termination_msg(pid_t pid, int status)
 841 {
 842         if (WIFEXITED(status))
 843                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 844                         WEXITSTATUS(status)));
 845         else if (WIFSIGNALED(status))
 846                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 847                         WTERMSIG(status)));
 848         else
 849                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 850 }
 851
 852 static int wait_for_process(pid_t pid, int *status)
 853 {
 854         int ret;
 855
 856         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 857         for (;;) {
 858                 fd_set rfds;
 859
 860                 FD_ZERO(&rfds);
 861                 FD_SET(signal_pipe, &rfds);
 862                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 863                 if (ret < 0)
 864                         break;
 865                 ret = next_signal();
 866                 if (!ret)
 867                         continue;
 868                 if (ret == SIGCHLD) {
 869                         ret = waitpid(pid, status, 0);
 870                         if (ret >= 0)
 871                                 break;
 872                         if (errno != EINTR) { /* error */
 873                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 874                                 break;
 875                         }
 876                 }
 877                 /* SIGINT or SIGTERM */
 878                 dss_kill(pid, SIGTERM, "killing child process");
 879         }
 880         if (ret < 0)
 881                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 882         else
 883                 log_termination_msg(pid, *status);
 884         return ret;
 885 }
 886
 887 static void handle_pre_remove_exit(int status)
 888 {
 889         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 890                 snapshot_removal_status = HS_READY;
 891                 gettimeofday(&next_removal_check, NULL);
 892                 next_removal_check.tv_sec += 60;
 893                 return;
 894         }
 895         snapshot_removal_status = HS_PRE_SUCCESS;
 896 }
 897
 898 static int handle_rm_exit(int status)
 899 {
 900         if (!WIFEXITED(status)) {
 901                 snapshot_removal_status = HS_READY;
 902                 return -E_INVOLUNTARY_EXIT;
 903         }
 904         if (WEXITSTATUS(status)) {
 905                 snapshot_removal_status = HS_READY;
 906                 return -E_BAD_EXIT_CODE;
 907         }
 908         snapshot_removal_status = HS_SUCCESS;
 909         return 1;
 910 }
 911
 912 static void handle_post_remove_exit(void)
 913 {
 914         snapshot_removal_status = HS_READY;
 915 }
 916
 917 static int handle_remove_exit(int status)
 918 {
 919         int ret;
 920         struct snapshot *s = snapshot_currently_being_removed;
 921
 922         assert(s);
 923         switch (snapshot_removal_status) {
 924         case HS_PRE_RUNNING:
 925                 handle_pre_remove_exit(status);
 926                 ret = 1;
 927                 break;
 928         case HS_RUNNING:
 929                 ret = handle_rm_exit(status);
 930                 break;
 931         case HS_POST_RUNNING:
 932                 handle_post_remove_exit();
 933                 ret = 1;
 934                 break;
 935         default:
 936                 ret = -E_BUG;
 937         }
 938         if (snapshot_removal_status == HS_READY) {
 939                 free(s->name);
 940                 free(s);
 941                 snapshot_currently_being_removed = NULL;
 942         }
 943         remove_pid = 0;
 944         return ret;
 945 }
 946
 947 static int wait_for_remove_process(void)
 948 {
 949         int status, ret;
 950
 951         assert(remove_pid);
 952         assert(
 953                 snapshot_removal_status == HS_PRE_RUNNING ||
 954                 snapshot_removal_status == HS_RUNNING ||
 955                 snapshot_removal_status == HS_POST_RUNNING
 956         );
 957         ret = wait_for_process(remove_pid, &status);
 958         if (ret < 0)
 959                 return ret;
 960         return handle_remove_exit(status);
 961 }
 962
 963 static int handle_rsync_exit(int status)
 964 {
 965         int es, ret;
 966
 967         if (!WIFEXITED(status)) {
 968                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 969                 ret = -E_INVOLUNTARY_EXIT;
 970                 snapshot_creation_status = HS_READY;
 971                 goto out;
 972         }
 973         es = WEXITSTATUS(status);
 974         /*
 975          * Restart rsync on non-fatal errors:
 976          * 24: Partial transfer due to vanished source files
 977          */
 978         if (es != 0 && es != 24) {
 979                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 980                         es, ++num_consecutive_rsync_errors));
 981                 if (!logfile) { /* called by com_run() */
 982                         ret = -E_BAD_EXIT_CODE;
 983                         goto out;
 984                 }
 985                 if (num_consecutive_rsync_errors >
 986                                 OPT_UINT32_VAL(RUN, MAX_RSYNC_ERRORS)) {
 987                         ret = -E_TOO_MANY_RSYNC_ERRORS;
 988                         snapshot_creation_status = HS_READY;
 989                         goto out;
 990                 }
 991                 DSS_WARNING_LOG(("restarting rsync process\n"));
 992                 snapshot_creation_status = HS_NEEDS_RESTART;
 993                 next_snapshot_time = get_current_time() + 60;
 994                 ret = 1;
 995                 goto out;
 996         }
 997         num_consecutive_rsync_errors = 0;
 998         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 999         if (ret < 0)
1000                 goto out;
1001         snapshot_creation_status = HS_SUCCESS;
1002         free(name_of_reference_snapshot);
1003         name_of_reference_snapshot = NULL;
1004 out:
1005         create_process_stopped = 0;
1006         return ret;
1007 }
1008
1009 static int handle_pre_create_hook_exit(int status)
1010 {
1011         int es, ret;
1012         static int warn_count;
1013
1014         if (!WIFEXITED(status)) {
1015                 snapshot_creation_status = HS_READY;
1016                 ret = -E_INVOLUNTARY_EXIT;
1017                 goto out;
1018         }
1019         es = WEXITSTATUS(status);
1020         if (es) {
1021                 if (!warn_count--) {
1022                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
1023                                 OPT_STRING_VAL(DSS, PRE_CREATE_HOOK), es));
1024                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
1025                         warn_count = 60; /* warn only once per hour */
1026                 }
1027                 next_snapshot_time = get_current_time() + 60;
1028                 snapshot_creation_status = HS_READY;
1029                 ret = 0;
1030                 goto out;
1031         }
1032         warn_count = 0;
1033         snapshot_creation_status = HS_PRE_SUCCESS;
1034         ret = 1;
1035 out:
1036         return ret;
1037 }
1038
1039 static int handle_sigchld(void)
1040 {
1041         pid_t pid;
1042         int status, ret = reap_child(&pid, &status);
1043
1044         if (ret <= 0)
1045                 return ret;
1046
1047         if (pid == create_pid) {
1048                 switch (snapshot_creation_status) {
1049                 case HS_PRE_RUNNING:
1050                         ret = handle_pre_create_hook_exit(status);
1051                         break;
1052                 case HS_RUNNING:
1053                         ret = handle_rsync_exit(status);
1054                         break;
1055                 case HS_POST_RUNNING:
1056                         snapshot_creation_status = HS_READY;
1057                         ret = 1;
1058                         break;
1059                 default:
1060                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
1061                                 snapshot_creation_status));
1062                         return -E_BUG;
1063                 }
1064                 create_pid = 0;
1065                 return ret;
1066         }
1067         if (pid == remove_pid) {
1068                 ret = handle_remove_exit(status);
1069                 if (ret < 0)
1070                         return ret;
1071                 return ret;
1072         }
1073         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
1074         return -E_BUG;
1075 }
1076
1077 /* also checks if . is a mountpoint, if --mountpoint was given */
1078 static int change_to_dest_dir(void)
1079 {
1080         int ret;
1081         const char *dd = OPT_STRING_VAL(DSS, DEST_DIR);
1082         struct stat dot, dotdot;
1083
1084         DSS_INFO_LOG(("changing cwd to %s\n", dd));
1085         if (chdir(dd) < 0) {
1086                 ret = -ERRNO_TO_DSS_ERROR(errno);
1087                 DSS_ERROR_LOG(("could not change cwd to %s\n", dd));
1088                 return ret;
1089         }
1090         if (!OPT_GIVEN(DSS, MOUNTPOINT))
1091                 return 0;
1092         if (stat(".", &dot) < 0) {
1093                 ret = -ERRNO_TO_DSS_ERROR(errno);
1094                 DSS_ERROR_LOG(("could not stat .\n"));
1095                 return ret;
1096         }
1097         if (stat("..", &dotdot) < 0) {
1098                 ret = -ERRNO_TO_DSS_ERROR(errno);
1099                 DSS_ERROR_LOG(("could not stat ..\n"));
1100                 return ret;
1101         }
1102         if (dot.st_dev == dotdot.st_dev && dot.st_ino != dotdot.st_ino) {
1103                 DSS_ERROR_LOG(("mountpoint check failed for %s\n", dd));
1104                 return -E_MOUNTPOINT;
1105         }
1106         return 1;
1107 }
1108
1109 static int check_config(void)
1110 {
1111         int ret;
1112         uint32_t unit_interval = OPT_UINT32_VAL(DSS, UNIT_INTERVAL);
1113         uint32_t num_intervals = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
1114
1115         if (unit_interval == 0) {
1116                 DSS_ERROR_LOG(("bad unit interval: %i\n", unit_interval));
1117                 return -E_INVALID_NUMBER;
1118         }
1119         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", unit_interval));
1120
1121         if (num_intervals == 0 || num_intervals > 30) {
1122                 DSS_ERROR_LOG(("bad number of intervals: %i\n", num_intervals));
1123                 return -E_INVALID_NUMBER;
1124         }
1125         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE))
1126                 if (!OPT_GIVEN(DSS, SOURCE_DIR)) {
1127                         DSS_ERROR_LOG(("--source-dir required\n"));
1128                         return -E_SYNTAX;
1129                 }
1130         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE)
1131                         || subcmd == CMD_PTR(LS) || subcmd == CMD_PTR(PRUNE)) {
1132                 if (!OPT_GIVEN(DSS, DEST_DIR)) {
1133                         DSS_ERROR_LOG(("--dest-dir required\n"));
1134                         return -E_SYNTAX;
1135                 }
1136                 ret = change_to_dest_dir();
1137                 if (ret < 0)
1138                         return ret;
1139         }
1140         DSS_DEBUG_LOG(("number of intervals: %i\n", num_intervals));
1141         return 1;
1142 }
1143
1144 static int lopsub_error(int lopsub_ret, char **errctx)
1145 {
1146         const char *msg = lls_strerror(-lopsub_ret);
1147         if (*errctx)
1148                 DSS_ERROR_LOG(("%s: %s\n", *errctx, msg));
1149         else
1150                 DSS_ERROR_LOG(("%s\n", msg));
1151         free(*errctx);
1152         *errctx = NULL;
1153         return -E_LOPSUB;
1154 }
1155
1156 static int parse_config_file(bool sighup, const struct lls_command *cmd)
1157 {
1158         int ret, fd = -1;
1159         char *config_file = get_config_file_name();
1160         struct stat statbuf;
1161         void *map;
1162         size_t sz;
1163         int cf_argc;
1164         char **cf_argv, *errctx = NULL;
1165         struct lls_parse_result *cf_lpr, *merged_lpr, *clpr;
1166         const char *subcmd_name;
1167
1168         ret = open(config_file, O_RDONLY);
1169         if (ret < 0) {
1170                 if (errno != ENOENT || OPT_GIVEN(DSS, CONFIG_FILE)) {
1171                         ret = -ERRNO_TO_DSS_ERROR(errno);
1172                         DSS_ERROR_LOG(("config file %s can not be opened\n",
1173                                 config_file));
1174                         goto out;
1175                 }
1176                 /* no config file -- nothing to do */
1177                 ret = 0;
1178                 goto success;
1179         }
1180         fd = ret;
1181         ret = fstat(fd, &statbuf);
1182         if (ret < 0) {
1183                 ret = -ERRNO_TO_DSS_ERROR(errno);
1184                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
1185                 goto close_fd;
1186         }
1187         sz = statbuf.st_size;
1188         if (sz == 0) { /* config file is empty -- nothing to do */
1189                 ret = 0;
1190                 goto success;
1191         }
1192         map = mmap(NULL, sz, PROT_READ, MAP_PRIVATE, fd, 0);
1193         if (map == MAP_FAILED) {
1194                 ret = -ERRNO_TO_DSS_ERROR(errno);
1195                 DSS_ERROR_LOG(("failed to mmap config file %s\n",
1196                         config_file));
1197                 goto close_fd;
1198         }
1199         if (cmd == CMD_PTR(DSS))
1200                 subcmd_name = NULL;
1201         else
1202                 subcmd_name = lls_command_name(cmd);
1203         ret = lls_convert_config(map, sz, subcmd_name, &cf_argv, &errctx);
1204         munmap(map, sz);
1205         if (ret < 0) {
1206                 DSS_ERROR_LOG(("failed to convert config file %s\n",
1207                         config_file));
1208                 ret = lopsub_error(ret, &errctx);
1209                 goto close_fd;
1210         }
1211         cf_argc = ret;
1212         ret = lls_parse(cf_argc, cf_argv, cmd, &cf_lpr, &errctx);
1213         lls_free_argv(cf_argv);
1214         if (ret < 0) {
1215                 ret = lopsub_error(ret, &errctx);
1216                 goto close_fd;
1217         }
1218         clpr = cmd == CMD_PTR(DSS)? cmdline_lpr : cmdline_sublpr;
1219         if (sighup) /* config file overrides command line */
1220                 ret = lls_merge(cf_lpr, clpr, cmd, &merged_lpr, &errctx);
1221         else /* command line options overrride config file options */
1222                 ret = lls_merge(clpr, cf_lpr, cmd, &merged_lpr, &errctx);
1223         lls_free_parse_result(cf_lpr, cmd);
1224         if (ret < 0) {
1225                 ret = lopsub_error(ret, &errctx);
1226                 goto close_fd;
1227         }
1228         ret = 1;
1229 success:
1230         assert(ret >= 0);
1231         DSS_DEBUG_LOG(("loglevel: %d\n", OPT_UINT32_VAL(DSS, LOGLEVEL)));
1232         if (cmd != CMD_PTR(DSS)) {
1233                 if (ret > 0) {
1234                         if (sublpr != cmdline_sublpr)
1235                                 lls_free_parse_result(sublpr, cmd);
1236                         sublpr = merged_lpr;
1237                 } else
1238                         sublpr = cmdline_sublpr;
1239         } else {
1240                 if (ret > 0) {
1241                         if (lpr != cmdline_lpr)
1242                                 lls_free_parse_result(lpr, cmd);
1243                         lpr = merged_lpr;
1244                 } else
1245                         lpr = cmdline_lpr;
1246         }
1247 close_fd:
1248         if (fd >= 0)
1249                 close(fd);
1250 out:
1251         free(config_file);
1252         if (ret < 0)
1253                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1254         return ret;
1255 }
1256
1257 static int handle_sighup(void)
1258 {
1259         int ret;
1260
1261         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1262         dump_dss_config("old");
1263         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(DSS));
1264         if (ret < 0)
1265                 return ret;
1266         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(RUN));
1267         if (ret < 0)
1268                 return ret;
1269         ret = check_config();
1270         if (ret < 0)
1271                 return ret;
1272         close_log(logfile);
1273         logfile = NULL;
1274         if (OPT_GIVEN(RUN, DAEMON) || daemonized) {
1275                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1276                 log_welcome(OPT_UINT32_VAL(DSS, LOGLEVEL));
1277                 daemonized = true;
1278         }
1279         dump_dss_config("reloaded");
1280         invalidate_next_snapshot_time();
1281         return 1;
1282 }
1283
1284 static void kill_children(void)
1285 {
1286         restart_create_process();
1287         dss_kill(create_pid, SIGTERM, NULL);
1288         dss_kill(remove_pid, SIGTERM, NULL);
1289 }
1290
1291 static int handle_signal(void)
1292 {
1293         int sig, ret = next_signal();
1294
1295         if (ret <= 0)
1296                 goto out;
1297         sig = ret;
1298         switch (sig) {
1299         case SIGINT:
1300         case SIGTERM:
1301                 return -E_SIGNAL;
1302         case SIGHUP:
1303                 ret = handle_sighup();
1304                 break;
1305         case SIGCHLD:
1306                 ret = handle_sigchld();
1307                 break;
1308         }
1309 out:
1310         if (ret < 0)
1311                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1312         return ret;
1313 }
1314
1315 /*
1316  * We can not use rsync locally if the local user is different from the remote
1317  * user or if the src dir is not on the local host (or both).
1318  */
1319 static int use_rsync_locally(char *logname)
1320 {
1321         const char *h = OPT_STRING_VAL(DSS, REMOTE_HOST);
1322
1323         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1324                 return 0;
1325         if (OPT_GIVEN(DSS, REMOTE_USER) &&
1326                         strcmp(OPT_STRING_VAL(DSS, REMOTE_USER), logname))
1327                 return 0;
1328         return 1;
1329 }
1330
1331 static int rename_resume_snap(int64_t creation_time)
1332 {
1333         struct snapshot_list sl;
1334         struct snapshot *s = NULL;
1335         char *new_name = incomplete_name(creation_time);
1336         int ret;
1337         const char *why;
1338
1339         sl.num_snapshots = 0;
1340
1341         ret = 0;
1342         dss_get_snapshot_list(&sl);
1343         /*
1344          * Snapshot recycling: We first look at the newest snapshot. If this
1345          * snapshot happens to be incomplete, the last rsync process was
1346          * aborted and we reuse this one. Otherwise we look at snapshots which
1347          * could be removed (outdated and redundant snapshots) as candidates
1348          * for recycling. If no outdated/redundant snapshot exists, we check if
1349          * there is an orphaned snapshot, which likely is useless anyway.
1350          *
1351          * Only if no existing snapshot is suitable for recycling, we bite the
1352          * bullet and create a new one.
1353          */
1354         s = get_newest_snapshot(&sl);
1355         if (!s) /* no snapshots at all */
1356                 goto out;
1357         /* re-use last snapshot if it is incomplete */
1358         why = "aborted";
1359         if ((s->flags & SS_COMPLETE) == 0)
1360                 goto out;
1361         why = "outdated";
1362         s = find_outdated_snapshot(&sl);
1363         if (s)
1364                 goto out;
1365         why = "redundant";
1366         s = find_redundant_snapshot(&sl);
1367         if (s)
1368                 goto out;
1369         why = "orphaned";
1370         s = find_orphaned_snapshot(&sl);
1371 out:
1372         if (s) {
1373                 DSS_NOTICE_LOG(("recycling %s snapshot %s\n", why, s->name));
1374                 ret = dss_rename(s->name, new_name);
1375         }
1376         if (ret >= 0)
1377                 DSS_NOTICE_LOG(("creating %s\n", new_name));
1378         free(new_name);
1379         free_snapshot_list(&sl);
1380         return ret;
1381 }
1382
1383 static void create_rsync_argv(char ***argv, int64_t *num)
1384 {
1385         char *logname;
1386         int i = 0, j, N;
1387         struct snapshot_list sl;
1388         static bool seeded;
1389
1390         dss_get_snapshot_list(&sl);
1391         assert(!name_of_reference_snapshot);
1392         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1393         free_snapshot_list(&sl);
1394
1395         /*
1396          * We specify up to 6 arguments, one argument per given rsync option
1397          * and one argument per given source dir. We also need space for the
1398          * terminating NULL pointer.
1399          */
1400         N = OPT_GIVEN(DSS, RSYNC_OPTION) + OPT_GIVEN(DSS, SOURCE_DIR);
1401         *argv = dss_malloc((7 + N) * sizeof(char *));
1402         (*argv)[i++] = dss_strdup("rsync");
1403         (*argv)[i++] = dss_strdup("-a");
1404         (*argv)[i++] = dss_strdup("--delete");
1405         if (!seeded) {
1406                 srandom((unsigned)time(NULL)); /* no need to be fancy here */
1407                 seeded = true;
1408         }
1409         if (1000 * (random() / (RAND_MAX + 1.0)) < OPT_UINT32_VAL(DSS, CHECKSUM)) {
1410                 DSS_NOTICE_LOG(("adding --checksum to rsync options\n"));
1411                 (*argv)[i++] = dss_strdup("--checksum");
1412         }
1413         for (j = 0; j < OPT_GIVEN(DSS, RSYNC_OPTION); j++)
1414                 (*argv)[i++] = dss_strdup(lls_string_val(j,
1415                         OPT_RESULT(DSS, RSYNC_OPTION)));
1416         if (name_of_reference_snapshot) {
1417                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1418                 (*argv)[i++] = make_message("--link-dest=../%s",
1419                         name_of_reference_snapshot);
1420         } else
1421                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1422         logname = dss_logname();
1423         if (use_rsync_locally(logname)) {
1424                 for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++)
1425                         (*argv)[i++] = dss_strdup(lls_string_val(j,
1426                                 OPT_RESULT(DSS, SOURCE_DIR)));
1427         } else {
1428                 /*
1429                  * dss-1.0 and earlier did not support multiple source
1430                  * directories.  These versions appended a slash to the end of
1431                  * the source directory to make sure that only the contents of
1432                  * the single source directory, but not the directory itself,
1433                  * are copied to the destination. For multiple source
1434                  * directories, however, this is not a good idea because the
1435                  * source directories may well contain identical file names,
1436                  * which would then be copied to the same location on the
1437                  * destination, overwriting each other. Moreover, we want the
1438                  * directory on the destination match the source. To preserve
1439                  * the old behaviour, we thus have to special-case N=1.
1440                  */
1441                 for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++) {
1442                         (*argv)[i++] = make_message("%s@%s:%s%s",
1443                                 OPT_GIVEN(DSS, REMOTE_USER)?
1444                                         OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
1445                                 OPT_STRING_VAL(DSS, REMOTE_HOST),
1446                                 lls_string_val(j, OPT_RESULT(DSS, SOURCE_DIR)),
1447                                 N == 1? "/" : ""
1448                         );
1449                 }
1450         }
1451         free(logname);
1452         *num = get_current_time();
1453         (*argv)[i++] = incomplete_name(*num);
1454         (*argv)[i++] = NULL;
1455         for (j = 0; j < i; j++)
1456                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1457 }
1458
1459 static void free_rsync_argv(char **argv)
1460 {
1461         int i;
1462
1463         if (!argv)
1464                 return;
1465         for (i = 0; argv[i]; i++)
1466                 free(argv[i]);
1467         free(argv);
1468 }
1469
1470 static int create_snapshot(char **argv)
1471 {
1472         int ret;
1473
1474         assert(argv);
1475         ret = rename_resume_snap(current_snapshot_creation_time);
1476         if (ret < 0)
1477                 return ret;
1478         dss_exec(&create_pid, argv[0], argv);
1479         snapshot_creation_status = HS_RUNNING;
1480         return ret;
1481 }
1482
1483 static int select_loop(void)
1484 {
1485         int ret;
1486         /* check every 60 seconds for free disk space */
1487         struct timeval tv;
1488         char **rsync_argv = NULL;
1489
1490         for (;;) {
1491                 fd_set rfds;
1492                 struct timeval *tvp;
1493
1494                 if (remove_pid)
1495                         tvp = NULL; /* sleep until rm hook/process dies */
1496                 else { /* sleep one minute */
1497                         tv.tv_sec = 60;
1498                         tv.tv_usec = 0;
1499                         tvp = &tv;
1500                 }
1501                 FD_ZERO(&rfds);
1502                 FD_SET(signal_pipe, &rfds);
1503                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1504                 if (ret < 0)
1505                         goto out;
1506                 if (FD_ISSET(signal_pipe, &rfds)) {
1507                         ret = handle_signal();
1508                         if (ret < 0)
1509                                 goto out;
1510                 }
1511                 if (remove_pid)
1512                         continue;
1513                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1514                         ret = exec_rm();
1515                         if (ret < 0)
1516                                 goto out;
1517                         continue;
1518                 }
1519                 if (snapshot_removal_status == HS_SUCCESS) {
1520                         post_remove_hook();
1521                         continue;
1522                 }
1523                 ret = try_to_free_disk_space();
1524                 if (ret < 0)
1525                         goto out;
1526                 if (snapshot_removal_status != HS_READY) {
1527                         stop_create_process();
1528                         continue;
1529                 }
1530                 restart_create_process();
1531                 switch (snapshot_creation_status) {
1532                 case HS_READY:
1533                         if (!next_snapshot_is_due())
1534                                 continue;
1535                         pre_create_hook();
1536                         continue;
1537                 case HS_PRE_RUNNING:
1538                 case HS_RUNNING:
1539                 case HS_POST_RUNNING:
1540                         continue;
1541                 case HS_PRE_SUCCESS:
1542                         if (!name_of_reference_snapshot) {
1543                                 free_rsync_argv(rsync_argv);
1544                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1545                         }
1546                         ret = create_snapshot(rsync_argv);
1547                         if (ret < 0)
1548                                 goto out;
1549                         continue;
1550                 case HS_NEEDS_RESTART:
1551                         if (!next_snapshot_is_due())
1552                                 continue;
1553                         ret = create_snapshot(rsync_argv);
1554                         if (ret < 0)
1555                                 goto out;
1556                         continue;
1557                 case HS_SUCCESS:
1558                         post_create_hook();
1559                         continue;
1560                 }
1561         }
1562 out:
1563         return ret;
1564 }
1565
1566 static void exit_hook(int exit_code)
1567 {
1568         pid_t pid;
1569         char **argv, *tmp = dss_strdup(OPT_STRING_VAL(DSS, EXIT_HOOK));
1570         unsigned n = split_args(tmp, &argv, " \t");
1571
1572         n++;
1573         argv = dss_realloc(argv, (n + 1) * sizeof(char *));
1574         argv[n - 1] = dss_strdup(dss_strerror(-exit_code));
1575         argv[n] = NULL;
1576         dss_exec(&pid, argv[0], argv);
1577         free(argv[n - 1]);
1578         free(argv);
1579         free(tmp);
1580 }
1581
1582 static void lock_dss_or_die(void)
1583 {
1584         char *config_file = get_config_file_name();
1585         int ret = lock_dss(config_file);
1586
1587         free(config_file);
1588         if (ret < 0) {
1589                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1590                 exit(EXIT_FAILURE);
1591         }
1592 }
1593
1594 static int com_run(void)
1595 {
1596         int ret, fd = -1;
1597         char *config_file;
1598         pid_t pid;
1599
1600         if (OPT_GIVEN(DSS, DRY_RUN)) {
1601                 DSS_ERROR_LOG(("dry run not supported by this command\n"));
1602                 return -E_SYNTAX;
1603         }
1604         config_file = get_config_file_name();
1605         ret = get_dss_pid(config_file, &pid);
1606         free(config_file);
1607         if (ret >= 0) {
1608                 DSS_ERROR_LOG(("pid %d\n", (int)pid));
1609                 return -E_ALREADY_RUNNING;
1610         }
1611         if (OPT_GIVEN(RUN, DAEMON)) {
1612                 fd = daemon_init();
1613                 daemonized = true;
1614                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1615         }
1616         lock_dss_or_die();
1617         dump_dss_config("startup");
1618         ret = install_sighandler(SIGHUP);
1619         if (ret < 0)
1620                 return ret;
1621         if (fd >= 0) {
1622                 ret = write(fd, "\0", 1);
1623                 if (ret != 1) {
1624                         DSS_ERROR_LOG(("write to daemon pipe returned %d\n",
1625                                 ret));
1626                         if (ret < 0)
1627                                 return -ERRNO_TO_DSS_ERROR(errno);
1628                         return -E_BUG;
1629                 }
1630         }
1631         ret = select_loop();
1632         if (ret >= 0) /* impossible */
1633                 ret = -E_BUG;
1634         kill_children();
1635         exit_hook(ret);
1636         while (wait(NULL) >= 0 || errno != ECHILD)
1637                 ; /* still have children to wait for */
1638         return ret;
1639 }
1640 EXPORT_CMD_HANDLER(run);
1641
1642 static int com_prune(void)
1643 {
1644         int ret;
1645         struct snapshot_list sl;
1646         struct snapshot *victim;
1647         struct disk_space ds;
1648         const char *why;
1649
1650         lock_dss_or_die();
1651         ret = get_disk_space(".", &ds);
1652         if (ret < 0)
1653                 return ret;
1654         log_disk_space(&ds);
1655         dss_get_snapshot_list(&sl);
1656         why = "outdated";
1657         victim = find_outdated_snapshot(&sl);
1658         if (victim)
1659                 goto rm;
1660         why = "redundant";
1661         victim = find_redundant_snapshot(&sl);
1662         if (victim)
1663                 goto rm;
1664         ret = 0;
1665         goto out;
1666 rm:
1667         if (OPT_GIVEN(DSS, DRY_RUN)) {
1668                 dss_msg("%s snapshot %s (interval = %i)\n",
1669                         why, victim->name, victim->interval);
1670                 ret = 0;
1671                 goto out;
1672         }
1673         pre_remove_hook(victim, why);
1674         if (snapshot_removal_status == HS_PRE_RUNNING) {
1675                 ret = wait_for_remove_process();
1676                 if (ret < 0)
1677                         goto out;
1678                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1679                         goto out;
1680         }
1681         ret = exec_rm();
1682         if (ret < 0)
1683                 goto out;
1684         ret = wait_for_remove_process();
1685         if (ret < 0)
1686                 goto out;
1687         if (snapshot_removal_status != HS_SUCCESS)
1688                 goto out;
1689         post_remove_hook();
1690         if (snapshot_removal_status != HS_POST_RUNNING)
1691                 goto out;
1692         ret = wait_for_remove_process();
1693         if (ret < 0)
1694                 goto out;
1695         ret = 1;
1696 out:
1697         free_snapshot_list(&sl);
1698         return ret;
1699 }
1700 EXPORT_CMD_HANDLER(prune);
1701
1702 static int com_create(void)
1703 {
1704         int ret, status;
1705         char **rsync_argv;
1706
1707         lock_dss_or_die();
1708         if (OPT_GIVEN(DSS, DRY_RUN)) {
1709                 int i;
1710                 char *msg = NULL;
1711                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1712                 for (i = 0; rsync_argv[i]; i++) {
1713                         char *tmp = msg;
1714                         msg = make_message("%s%s%s", tmp? tmp : "",
1715                                 tmp? " " : "", rsync_argv[i]);
1716                         free(tmp);
1717                 }
1718                 free_rsync_argv(rsync_argv);
1719                 dss_msg("%s\n", msg);
1720                 free(msg);
1721                 return 1;
1722         }
1723         pre_create_hook();
1724         if (create_pid) {
1725                 ret = wait_for_process(create_pid, &status);
1726                 if (ret < 0)
1727                         return ret;
1728                 ret = handle_pre_create_hook_exit(status);
1729                 if (ret <= 0) /* error, or pre-create failed */
1730                         return ret;
1731         }
1732         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1733         ret = create_snapshot(rsync_argv);
1734         if (ret < 0)
1735                 goto out;
1736         ret = wait_for_process(create_pid, &status);
1737         if (ret < 0)
1738                 goto out;
1739         ret = handle_rsync_exit(status);
1740         if (ret < 0)
1741                 goto out;
1742         post_create_hook();
1743         if (create_pid)
1744                 ret = wait_for_process(create_pid, &status);
1745 out:
1746         free_rsync_argv(rsync_argv);
1747         return ret;
1748 }
1749 EXPORT_CMD_HANDLER(create);
1750
1751 static int com_ls(void)
1752 {
1753         int i;
1754         struct snapshot_list sl;
1755         struct snapshot *s;
1756         int64_t now = get_current_time();
1757
1758         dss_get_snapshot_list(&sl);
1759         FOR_EACH_SNAPSHOT(s, i, &sl) {
1760                 int64_t d;
1761                 if (s->flags & SS_COMPLETE)
1762                         d = (s->completion_time - s->creation_time) / 60;
1763                 else
1764                         d = (now - s->creation_time) / 60;
1765                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval,
1766                         s->name, d / 60, d % 60);
1767         }
1768         free_snapshot_list(&sl);
1769         return 1;
1770 }
1771 EXPORT_CMD_HANDLER(ls);
1772
1773 static int com_configtest(void)
1774 {
1775         printf("Syntax Ok\n");
1776         return 0;
1777 }
1778 EXPORT_CMD_HANDLER(configtest);
1779
1780 static int setup_signal_handling(void)
1781 {
1782         int ret;
1783
1784         DSS_INFO_LOG(("setting up signal handlers\n"));
1785         signal_pipe = signal_init(); /* always successful */
1786         ret = install_sighandler(SIGINT);
1787         if (ret < 0)
1788                 return ret;
1789         ret = install_sighandler(SIGTERM);
1790         if (ret < 0)
1791                 return ret;
1792         return install_sighandler(SIGCHLD);
1793 }
1794
1795 static void handle_version_and_help(void)
1796 {
1797         char *txt;
1798
1799         if (OPT_GIVEN(DSS, DETAILED_HELP))
1800                 txt = lls_long_help(CMD_PTR(DSS));
1801         else if (OPT_GIVEN(DSS, HELP))
1802                 txt = lls_short_help(CMD_PTR(DSS));
1803         else if (OPT_GIVEN(DSS, VERSION))
1804                 txt = make_message("%s\n", VERSION_STRING);
1805         else
1806                 return;
1807         printf("%s", txt);
1808         free(txt);
1809         exit(EXIT_SUCCESS);
1810 }
1811
1812 static void show_subcommand_summary(void)
1813 {
1814         const struct lls_command *cmd;
1815         int i;
1816
1817         printf("Available subcommands:\n");
1818         for (i = 1; (cmd = lls_cmd(i, dss_suite)); i++) {
1819                 const char *name = lls_command_name(cmd);
1820                 const char *purpose = lls_purpose(cmd);
1821                 printf("%-11s%s\n", name, purpose);
1822         }
1823         exit(EXIT_SUCCESS);
1824 }
1825
1826 int main(int argc, char **argv)
1827 {
1828         int ret;
1829         char *errctx = NULL;
1830         unsigned num_inputs;
1831         const struct dss_user_data *ud;
1832
1833         ret = lls_parse(argc, argv, CMD_PTR(DSS), &cmdline_lpr, &errctx);
1834         if (ret < 0) {
1835                 ret = lopsub_error(ret, &errctx);
1836                 goto out;
1837         }
1838         lpr = cmdline_lpr;
1839         ret = parse_config_file(false /* no SIGHUP */, CMD_PTR(DSS));
1840         if (ret < 0)
1841                 goto out;
1842         handle_version_and_help();
1843         num_inputs = lls_num_inputs(lpr);
1844         if (num_inputs == 0)
1845                 show_subcommand_summary();
1846         ret = lls_lookup_subcmd(argv[argc - num_inputs], dss_suite, &errctx);
1847         if (ret < 0) {
1848                 ret = lopsub_error(ret, &errctx);
1849                 goto out;
1850         }
1851         subcmd = lls_cmd(ret, dss_suite);
1852         ret = lls_parse(num_inputs, argv + argc - num_inputs, subcmd,
1853                 &cmdline_sublpr, &errctx);
1854         if (ret < 0) {
1855                 ret = lopsub_error(ret, &errctx);
1856                 goto out;
1857         }
1858         sublpr = cmdline_sublpr;
1859         ret = parse_config_file(false /* no SIGHUP */, subcmd);
1860         if (ret < 0)
1861                 goto out;
1862         ret = check_config();
1863         if (ret < 0)
1864                 goto out;
1865         ret = setup_signal_handling();
1866         if (ret < 0)
1867                 goto out;
1868         ud = lls_user_data(subcmd);
1869         ret = ud->handler();
1870         signal_shutdown();
1871 out:
1872         if (ret < 0) {
1873                 if (errctx)
1874                         DSS_ERROR_LOG(("%s\n", errctx));
1875                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1876         }
1877         free(errctx);
1878         lls_free_parse_result(lpr, CMD_PTR(DSS));
1879         if (lpr != cmdline_lpr)
1880                 lls_free_parse_result(cmdline_lpr, CMD_PTR(DSS));
1881         lls_free_parse_result(sublpr, subcmd);
1882         if (sublpr != cmdline_sublpr)
1883                 lls_free_parse_result(cmdline_sublpr, subcmd);
1884         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1885 }