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         DSS_DEBUG_LOG(("picking snapshot with earliest creation time\n"));
 655         num_complete = num_complete_snapshots(sl);
 656         if (num_complete <= OPT_UINT32_VAL(DSS, MIN_COMPLETE))
 657                 return NULL;
 658         FOR_EACH_SNAPSHOT(s, i, sl) {
 659                 if (snapshot_is_being_created(s))
 660                         continue;
 661                 if (is_reference_snapshot(s)) { /* avoid this one */
 662                         ref = s;
 663                         continue;
 664                 }
 665                 return s;
 666         }
 667         assert(ref);
 668         DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
 669         return ref;
 670 }
 671
 672 /* returns NULL <==> *reason is set to NULL */
 673 static struct snapshot *find_removable_snapshot(struct snapshot_list *sl,
 674                 bool try_hard, char **reason)
 675 {
 676         struct snapshot *victim;
 677
 678         /*
 679          * Don't remove anything if there is free space and we have fewer
 680          * snapshots than configured, plus one. This way there is always one
 681          * snapshot that can be recycled.
 682          */
 683         if (!try_hard && sl->num_snapshots <=
 684                         1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
 685                 goto nope;
 686         victim = find_orphaned_snapshot(sl);
 687         if (victim) {
 688                 *reason = make_message("orphaned");
 689                 return victim;
 690         }
 691         victim = find_outdated_snapshot(sl);
 692         if (victim) {
 693                 *reason = make_message("outdated");
 694                 return victim;
 695         }
 696         if (!OPT_GIVEN(DSS, KEEP_REDUNDANT)) {
 697                 victim = find_redundant_snapshot(sl);
 698                 if (victim) {
 699                         *reason = make_message("redundant");
 700                         return victim;
 701                 }
 702         }
 703         if (!try_hard)
 704                 goto nope;
 705         DSS_WARNING_LOG(("nothing obvious to remove\n"));
 706         victim = find_oldest_removable_snapshot(sl);
 707         if (victim) {
 708                 *reason = make_message("oldest");
 709                 return victim;
 710         }
 711 nope:
 712         *reason = NULL;
 713         return NULL;
 714 }
 715
 716 static int rename_incomplete_snapshot(int64_t start)
 717 {
 718         char *old_name;
 719         int ret;
 720         int64_t now;
 721
 722         /*
 723          * We don't want the dss_rename() below to fail with EEXIST because the
 724          * last complete snapshot was created (and completed) in the same
 725          * second as this one.
 726          */
 727         while ((now = get_current_time()) == start)
 728                 sleep(1);
 729         free(path_to_last_complete_snapshot);
 730         ret = complete_name(start, now, &path_to_last_complete_snapshot);
 731         if (ret < 0)
 732                 return ret;
 733         old_name = incomplete_name(start);
 734         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 735         if (ret >= 0)
 736                 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
 737                         path_to_last_complete_snapshot));
 738         free(old_name);
 739         return ret;
 740 }
 741
 742 static int try_to_free_disk_space(void)
 743 {
 744         int ret;
 745         struct snapshot_list sl;
 746         struct snapshot *victim;
 747         struct timeval now;
 748         char *why;
 749         int low_disk_space;
 750
 751         ret = disk_space_low(NULL);
 752         if (ret < 0)
 753                 return ret;
 754         low_disk_space = ret;
 755         gettimeofday(&now, NULL);
 756         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 757                 return 0;
 758         if (!low_disk_space) {
 759                 if (snapshot_creation_status != HS_READY)
 760                         return 0;
 761                 if (next_snapshot_is_due())
 762                         return 0;
 763         }
 764         /* Idle or low disk space, look at existing snapshots. */
 765         dss_get_snapshot_list(&sl);
 766         victim = find_removable_snapshot(&sl, low_disk_space, &why);
 767         if (victim) {
 768                 pre_remove_hook(victim, why);
 769                 free(why);
 770         }
 771         free_snapshot_list(&sl);
 772         if (victim)
 773                 return 1;
 774         if (!low_disk_space)
 775                 return 0;
 776         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 777         return -ERRNO_TO_DSS_ERROR(ENOSPC);
 778 }
 779
 780 static void post_create_hook(void)
 781 {
 782         char *cmd = make_message("%s %s/%s",
 783                 OPT_STRING_VAL(DSS, POST_CREATE_HOOK),
 784                 OPT_STRING_VAL(DSS, DEST_DIR), path_to_last_complete_snapshot);
 785         DSS_NOTICE_LOG(("executing %s\n", cmd));
 786         dss_exec_cmdline_pid(&create_pid, cmd);
 787         free(cmd);
 788         snapshot_creation_status = HS_POST_RUNNING;
 789 }
 790
 791 static void post_remove_hook(void)
 792 {
 793         char *cmd;
 794         struct snapshot *s = snapshot_currently_being_removed;
 795
 796         assert(s);
 797
 798         cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, POST_REMOVE_HOOK),
 799                 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
 800         DSS_NOTICE_LOG(("executing %s\n", cmd));
 801         dss_exec_cmdline_pid(&remove_pid, cmd);
 802         free(cmd);
 803         snapshot_removal_status = HS_POST_RUNNING;
 804 }
 805
 806 static void dss_kill(pid_t pid, int sig, const char *msg)
 807 {
 808         const char *signame, *process_name;
 809
 810         if (pid == 0)
 811                 return;
 812         switch (sig) {
 813         case SIGTERM: signame = "TERM"; break;
 814         case SIGSTOP: signame = "STOP"; break;
 815         case SIGCONT: signame = "CONT"; break;
 816         default: signame = "????";
 817         }
 818
 819         if (pid == create_pid)
 820                 process_name = "create";
 821         else if (pid == remove_pid)
 822                 process_name = "remove";
 823         else process_name = "??????";
 824
 825         if (msg)
 826                 DSS_INFO_LOG(("%s\n", msg));
 827         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 828                 sig, signame, (int)pid, process_name));
 829         if (kill(pid, sig) >= 0)
 830                 return;
 831         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 832                 sig, signame, (int)pid, process_name));
 833 }
 834
 835 static void stop_create_process(void)
 836 {
 837         if (create_process_stopped)
 838                 return;
 839         dss_kill(create_pid, SIGSTOP, "suspending create process");
 840         create_process_stopped = 1;
 841 }
 842
 843 static void restart_create_process(void)
 844 {
 845         if (!create_process_stopped)
 846                 return;
 847         dss_kill(create_pid, SIGCONT, "resuming create process");
 848         create_process_stopped = 0;
 849 }
 850
 851 /**
 852  * Print a log message about the exit status of a child.
 853  */
 854 static void log_termination_msg(pid_t pid, int status)
 855 {
 856         if (WIFEXITED(status))
 857                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 858                         WEXITSTATUS(status)));
 859         else if (WIFSIGNALED(status))
 860                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 861                         WTERMSIG(status)));
 862         else
 863                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 864 }
 865
 866 static int wait_for_process(pid_t pid, int *status)
 867 {
 868         int ret;
 869
 870         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 871         for (;;) {
 872                 fd_set rfds;
 873
 874                 FD_ZERO(&rfds);
 875                 FD_SET(signal_pipe, &rfds);
 876                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 877                 if (ret < 0)
 878                         break;
 879                 ret = next_signal();
 880                 if (!ret)
 881                         continue;
 882                 if (ret == SIGCHLD) {
 883                         ret = waitpid(pid, status, 0);
 884                         if (ret >= 0)
 885                                 break;
 886                         if (errno != EINTR) { /* error */
 887                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 888                                 break;
 889                         }
 890                 }
 891                 /* SIGINT or SIGTERM */
 892                 dss_kill(pid, SIGTERM, "killing child process");
 893         }
 894         if (ret < 0)
 895                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 896         else
 897                 log_termination_msg(pid, *status);
 898         return ret;
 899 }
 900
 901 static void handle_pre_remove_exit(int status)
 902 {
 903         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 904                 snapshot_removal_status = HS_READY;
 905                 gettimeofday(&next_removal_check, NULL);
 906                 next_removal_check.tv_sec += 60;
 907                 return;
 908         }
 909         snapshot_removal_status = HS_PRE_SUCCESS;
 910 }
 911
 912 static int handle_rm_exit(int status)
 913 {
 914         if (!WIFEXITED(status)) {
 915                 snapshot_removal_status = HS_READY;
 916                 return -E_INVOLUNTARY_EXIT;
 917         }
 918         if (WEXITSTATUS(status)) {
 919                 snapshot_removal_status = HS_READY;
 920                 return -E_BAD_EXIT_CODE;
 921         }
 922         snapshot_removal_status = HS_SUCCESS;
 923         return 1;
 924 }
 925
 926 static void handle_post_remove_exit(void)
 927 {
 928         snapshot_removal_status = HS_READY;
 929 }
 930
 931 static int handle_remove_exit(int status)
 932 {
 933         int ret;
 934         struct snapshot *s = snapshot_currently_being_removed;
 935
 936         assert(s);
 937         switch (snapshot_removal_status) {
 938         case HS_PRE_RUNNING:
 939                 handle_pre_remove_exit(status);
 940                 ret = 1;
 941                 break;
 942         case HS_RUNNING:
 943                 ret = handle_rm_exit(status);
 944                 break;
 945         case HS_POST_RUNNING:
 946                 handle_post_remove_exit();
 947                 ret = 1;
 948                 break;
 949         default:
 950                 ret = -E_BUG;
 951         }
 952         if (snapshot_removal_status == HS_READY) {
 953                 free(s->name);
 954                 free(s);
 955                 snapshot_currently_being_removed = NULL;
 956         }
 957         remove_pid = 0;
 958         return ret;
 959 }
 960
 961 static int wait_for_remove_process(void)
 962 {
 963         int status, ret;
 964
 965         assert(remove_pid);
 966         assert(
 967                 snapshot_removal_status == HS_PRE_RUNNING ||
 968                 snapshot_removal_status == HS_RUNNING ||
 969                 snapshot_removal_status == HS_POST_RUNNING
 970         );
 971         ret = wait_for_process(remove_pid, &status);
 972         if (ret < 0)
 973                 return ret;
 974         return handle_remove_exit(status);
 975 }
 976
 977 static int handle_rsync_exit(int status)
 978 {
 979         int es, ret;
 980
 981         if (!WIFEXITED(status)) {
 982                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 983                 ret = -E_INVOLUNTARY_EXIT;
 984                 snapshot_creation_status = HS_READY;
 985                 goto out;
 986         }
 987         es = WEXITSTATUS(status);
 988         /*
 989          * Restart rsync on non-fatal errors:
 990          * 24: Partial transfer due to vanished source files
 991          */
 992         if (es != 0 && es != 24) {
 993                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 994                         es, ++num_consecutive_rsync_errors));
 995                 if (!logfile) { /* called by com_run() */
 996                         ret = -E_BAD_EXIT_CODE;
 997                         goto out;
 998                 }
 999                 if (num_consecutive_rsync_errors >
1000                                 OPT_UINT32_VAL(RUN, MAX_RSYNC_ERRORS)) {
1001                         ret = -E_TOO_MANY_RSYNC_ERRORS;
1002                         snapshot_creation_status = HS_READY;
1003                         goto out;
1004                 }
1005                 DSS_WARNING_LOG(("restarting rsync process\n"));
1006                 snapshot_creation_status = HS_NEEDS_RESTART;
1007                 next_snapshot_time = get_current_time() + 60;
1008                 ret = 1;
1009                 goto out;
1010         }
1011         num_consecutive_rsync_errors = 0;
1012         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
1013         if (ret < 0)
1014                 goto out;
1015         snapshot_creation_status = HS_SUCCESS;
1016         free(name_of_reference_snapshot);
1017         name_of_reference_snapshot = NULL;
1018 out:
1019         create_process_stopped = 0;
1020         return ret;
1021 }
1022
1023 static int handle_pre_create_hook_exit(int status)
1024 {
1025         int es, ret;
1026         static int warn_count;
1027
1028         if (!WIFEXITED(status)) {
1029                 snapshot_creation_status = HS_READY;
1030                 ret = -E_INVOLUNTARY_EXIT;
1031                 goto out;
1032         }
1033         es = WEXITSTATUS(status);
1034         if (es) {
1035                 if (!warn_count--) {
1036                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
1037                                 OPT_STRING_VAL(DSS, PRE_CREATE_HOOK), es));
1038                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
1039                         warn_count = 60; /* warn only once per hour */
1040                 }
1041                 next_snapshot_time = get_current_time() + 60;
1042                 snapshot_creation_status = HS_READY;
1043                 ret = 0;
1044                 goto out;
1045         }
1046         warn_count = 0;
1047         snapshot_creation_status = HS_PRE_SUCCESS;
1048         ret = 1;
1049 out:
1050         return ret;
1051 }
1052
1053 static int handle_sigchld(void)
1054 {
1055         pid_t pid;
1056         int status, ret = reap_child(&pid, &status);
1057
1058         if (ret <= 0)
1059                 return ret;
1060
1061         if (pid == create_pid) {
1062                 switch (snapshot_creation_status) {
1063                 case HS_PRE_RUNNING:
1064                         ret = handle_pre_create_hook_exit(status);
1065                         break;
1066                 case HS_RUNNING:
1067                         ret = handle_rsync_exit(status);
1068                         break;
1069                 case HS_POST_RUNNING:
1070                         snapshot_creation_status = HS_READY;
1071                         ret = 1;
1072                         break;
1073                 default:
1074                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
1075                                 snapshot_creation_status));
1076                         return -E_BUG;
1077                 }
1078                 create_pid = 0;
1079                 return ret;
1080         }
1081         if (pid == remove_pid) {
1082                 ret = handle_remove_exit(status);
1083                 if (ret < 0)
1084                         return ret;
1085                 return ret;
1086         }
1087         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
1088         return -E_BUG;
1089 }
1090
1091 /* also checks if . is a mountpoint, if --mountpoint was given */
1092 static int change_to_dest_dir(void)
1093 {
1094         int ret;
1095         const char *dd = OPT_STRING_VAL(DSS, DEST_DIR);
1096         struct stat dot, dotdot;
1097
1098         DSS_INFO_LOG(("changing cwd to %s\n", dd));
1099         if (chdir(dd) < 0) {
1100                 ret = -ERRNO_TO_DSS_ERROR(errno);
1101                 DSS_ERROR_LOG(("could not change cwd to %s\n", dd));
1102                 return ret;
1103         }
1104         if (!OPT_GIVEN(DSS, MOUNTPOINT))
1105                 return 0;
1106         if (stat(".", &dot) < 0) {
1107                 ret = -ERRNO_TO_DSS_ERROR(errno);
1108                 DSS_ERROR_LOG(("could not stat .\n"));
1109                 return ret;
1110         }
1111         if (stat("..", &dotdot) < 0) {
1112                 ret = -ERRNO_TO_DSS_ERROR(errno);
1113                 DSS_ERROR_LOG(("could not stat ..\n"));
1114                 return ret;
1115         }
1116         if (dot.st_dev == dotdot.st_dev && dot.st_ino != dotdot.st_ino) {
1117                 DSS_ERROR_LOG(("mountpoint check failed for %s\n", dd));
1118                 return -E_MOUNTPOINT;
1119         }
1120         return 1;
1121 }
1122
1123 static int check_config(void)
1124 {
1125         int ret;
1126         uint32_t unit_interval = OPT_UINT32_VAL(DSS, UNIT_INTERVAL);
1127         uint32_t num_intervals = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
1128
1129         if (unit_interval == 0) {
1130                 DSS_ERROR_LOG(("bad unit interval: %i\n", unit_interval));
1131                 return -E_INVALID_NUMBER;
1132         }
1133         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", unit_interval));
1134
1135         if (num_intervals == 0 || num_intervals > 30) {
1136                 DSS_ERROR_LOG(("bad number of intervals: %i\n", num_intervals));
1137                 return -E_INVALID_NUMBER;
1138         }
1139         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE))
1140                 if (!OPT_GIVEN(DSS, SOURCE_DIR)) {
1141                         DSS_ERROR_LOG(("--source-dir required\n"));
1142                         return -E_SYNTAX;
1143                 }
1144         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE)
1145                         || subcmd == CMD_PTR(LS) || subcmd == CMD_PTR(PRUNE)) {
1146                 if (!OPT_GIVEN(DSS, DEST_DIR)) {
1147                         DSS_ERROR_LOG(("--dest-dir required\n"));
1148                         return -E_SYNTAX;
1149                 }
1150                 ret = change_to_dest_dir();
1151                 if (ret < 0)
1152                         return ret;
1153         }
1154         DSS_DEBUG_LOG(("number of intervals: %i\n", num_intervals));
1155         return 1;
1156 }
1157
1158 static int lopsub_error(int lopsub_ret, char **errctx)
1159 {
1160         const char *msg = lls_strerror(-lopsub_ret);
1161         if (*errctx)
1162                 DSS_ERROR_LOG(("%s: %s\n", *errctx, msg));
1163         else
1164                 DSS_ERROR_LOG(("%s\n", msg));
1165         free(*errctx);
1166         *errctx = NULL;
1167         return -E_LOPSUB;
1168 }
1169
1170 static int parse_config_file(bool sighup, const struct lls_command *cmd)
1171 {
1172         int ret, fd = -1;
1173         char *config_file = get_config_file_name();
1174         struct stat statbuf;
1175         void *map;
1176         size_t sz;
1177         int cf_argc;
1178         char **cf_argv, *errctx = NULL;
1179         struct lls_parse_result *cf_lpr, *merged_lpr, *clpr;
1180         const char *subcmd_name;
1181
1182         ret = open(config_file, O_RDONLY);
1183         if (ret < 0) {
1184                 if (errno != ENOENT || OPT_GIVEN(DSS, CONFIG_FILE)) {
1185                         ret = -ERRNO_TO_DSS_ERROR(errno);
1186                         DSS_ERROR_LOG(("config file %s can not be opened\n",
1187                                 config_file));
1188                         goto out;
1189                 }
1190                 /* no config file -- nothing to do */
1191                 ret = 0;
1192                 goto success;
1193         }
1194         fd = ret;
1195         ret = fstat(fd, &statbuf);
1196         if (ret < 0) {
1197                 ret = -ERRNO_TO_DSS_ERROR(errno);
1198                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
1199                 goto close_fd;
1200         }
1201         sz = statbuf.st_size;
1202         if (sz == 0) { /* config file is empty -- nothing to do */
1203                 ret = 0;
1204                 goto success;
1205         }
1206         map = mmap(NULL, sz, PROT_READ, MAP_PRIVATE, fd, 0);
1207         if (map == MAP_FAILED) {
1208                 ret = -ERRNO_TO_DSS_ERROR(errno);
1209                 DSS_ERROR_LOG(("failed to mmap config file %s\n",
1210                         config_file));
1211                 goto close_fd;
1212         }
1213         if (cmd == CMD_PTR(DSS))
1214                 subcmd_name = NULL;
1215         else
1216                 subcmd_name = lls_command_name(cmd);
1217         ret = lls_convert_config(map, sz, subcmd_name, &cf_argv, &errctx);
1218         munmap(map, sz);
1219         if (ret < 0) {
1220                 DSS_ERROR_LOG(("failed to convert config file %s\n",
1221                         config_file));
1222                 ret = lopsub_error(ret, &errctx);
1223                 goto close_fd;
1224         }
1225         cf_argc = ret;
1226         ret = lls_parse(cf_argc, cf_argv, cmd, &cf_lpr, &errctx);
1227         lls_free_argv(cf_argv);
1228         if (ret < 0) {
1229                 ret = lopsub_error(ret, &errctx);
1230                 goto close_fd;
1231         }
1232         clpr = cmd == CMD_PTR(DSS)? cmdline_lpr : cmdline_sublpr;
1233         if (sighup) /* config file overrides command line */
1234                 ret = lls_merge(cf_lpr, clpr, cmd, &merged_lpr, &errctx);
1235         else /* command line options overrride config file options */
1236                 ret = lls_merge(clpr, cf_lpr, cmd, &merged_lpr, &errctx);
1237         lls_free_parse_result(cf_lpr, cmd);
1238         if (ret < 0) {
1239                 ret = lopsub_error(ret, &errctx);
1240                 goto close_fd;
1241         }
1242         ret = 1;
1243 success:
1244         assert(ret >= 0);
1245         DSS_DEBUG_LOG(("loglevel: %d\n", OPT_UINT32_VAL(DSS, LOGLEVEL)));
1246         if (cmd != CMD_PTR(DSS)) {
1247                 if (ret > 0) {
1248                         if (sublpr != cmdline_sublpr)
1249                                 lls_free_parse_result(sublpr, cmd);
1250                         sublpr = merged_lpr;
1251                 } else
1252                         sublpr = cmdline_sublpr;
1253         } else {
1254                 if (ret > 0) {
1255                         if (lpr != cmdline_lpr)
1256                                 lls_free_parse_result(lpr, cmd);
1257                         lpr = merged_lpr;
1258                 } else
1259                         lpr = cmdline_lpr;
1260         }
1261 close_fd:
1262         if (fd >= 0)
1263                 close(fd);
1264 out:
1265         free(config_file);
1266         if (ret < 0)
1267                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1268         return ret;
1269 }
1270
1271 static int handle_sighup(void)
1272 {
1273         int ret;
1274
1275         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1276         dump_dss_config("old");
1277         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(DSS));
1278         if (ret < 0)
1279                 return ret;
1280         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(RUN));
1281         if (ret < 0)
1282                 return ret;
1283         ret = check_config();
1284         if (ret < 0)
1285                 return ret;
1286         close_log(logfile);
1287         logfile = NULL;
1288         if (OPT_GIVEN(RUN, DAEMON) || daemonized) {
1289                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1290                 log_welcome(OPT_UINT32_VAL(DSS, LOGLEVEL));
1291                 daemonized = true;
1292         }
1293         dump_dss_config("reloaded");
1294         invalidate_next_snapshot_time();
1295         return 1;
1296 }
1297
1298 static void kill_children(void)
1299 {
1300         restart_create_process();
1301         dss_kill(create_pid, SIGTERM, NULL);
1302         dss_kill(remove_pid, SIGTERM, NULL);
1303 }
1304
1305 static int handle_signal(void)
1306 {
1307         int sig, ret = next_signal();
1308
1309         if (ret <= 0)
1310                 goto out;
1311         sig = ret;
1312         switch (sig) {
1313         case SIGINT:
1314         case SIGTERM:
1315                 return -E_SIGNAL;
1316         case SIGHUP:
1317                 ret = handle_sighup();
1318                 break;
1319         case SIGCHLD:
1320                 ret = handle_sigchld();
1321                 break;
1322         }
1323 out:
1324         if (ret < 0)
1325                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1326         return ret;
1327 }
1328
1329 /*
1330  * We can not use rsync locally if the local user is different from the remote
1331  * user or if the src dir is not on the local host (or both).
1332  */
1333 static int use_rsync_locally(char *logname)
1334 {
1335         const char *h = OPT_STRING_VAL(DSS, REMOTE_HOST);
1336
1337         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1338                 return 0;
1339         if (OPT_GIVEN(DSS, REMOTE_USER) &&
1340                         strcmp(OPT_STRING_VAL(DSS, REMOTE_USER), logname))
1341                 return 0;
1342         return 1;
1343 }
1344
1345 static int rename_resume_snap(int64_t creation_time)
1346 {
1347         struct snapshot_list sl;
1348         struct snapshot *s = NULL;
1349         char *new_name = incomplete_name(creation_time);
1350         int ret;
1351         const char *why;
1352
1353         sl.num_snapshots = 0;
1354
1355         ret = 0;
1356         dss_get_snapshot_list(&sl);
1357         /*
1358          * Snapshot recycling: We first look at the newest snapshot. If this
1359          * snapshot happens to be incomplete, the last rsync process was
1360          * aborted and we reuse this one. Otherwise we look at snapshots which
1361          * could be removed (outdated and redundant snapshots) as candidates
1362          * for recycling. If no outdated/redundant snapshot exists, we check if
1363          * there is an orphaned snapshot, which likely is useless anyway.
1364          *
1365          * Only if no existing snapshot is suitable for recycling, we bite the
1366          * bullet and create a new one.
1367          */
1368         s = get_newest_snapshot(&sl);
1369         if (!s) /* no snapshots at all */
1370                 goto out;
1371         /* re-use last snapshot if it is incomplete */
1372         why = "aborted";
1373         if ((s->flags & SS_COMPLETE) == 0)
1374                 goto out;
1375         why = "outdated";
1376         s = find_outdated_snapshot(&sl);
1377         if (s)
1378                 goto out;
1379         why = "redundant";
1380         s = find_redundant_snapshot(&sl);
1381         if (s)
1382                 goto out;
1383         why = "orphaned";
1384         s = find_orphaned_snapshot(&sl);
1385 out:
1386         if (s) {
1387                 DSS_NOTICE_LOG(("recycling %s snapshot %s\n", why, s->name));
1388                 ret = dss_rename(s->name, new_name);
1389         }
1390         if (ret >= 0)
1391                 DSS_NOTICE_LOG(("creating %s\n", new_name));
1392         free(new_name);
1393         free_snapshot_list(&sl);
1394         return ret;
1395 }
1396
1397 static void create_rsync_argv(char ***argv, int64_t *num)
1398 {
1399         char *logname;
1400         int i = 0, j, N;
1401         struct snapshot_list sl;
1402         static bool seeded;
1403
1404         dss_get_snapshot_list(&sl);
1405         assert(!name_of_reference_snapshot);
1406         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1407         free_snapshot_list(&sl);
1408
1409         /*
1410          * We specify up to 6 arguments, one argument per given rsync option
1411          * and one argument per given source dir. We also need space for the
1412          * terminating NULL pointer.
1413          */
1414         N = OPT_GIVEN(DSS, RSYNC_OPTION) + OPT_GIVEN(DSS, SOURCE_DIR);
1415         *argv = dss_malloc((7 + N) * sizeof(char *));
1416         (*argv)[i++] = dss_strdup("rsync");
1417         (*argv)[i++] = dss_strdup("-a");
1418         (*argv)[i++] = dss_strdup("--delete");
1419         if (!seeded) {
1420                 srandom((unsigned)time(NULL)); /* no need to be fancy here */
1421                 seeded = true;
1422         }
1423         if (1000 * (random() / (RAND_MAX + 1.0)) < OPT_UINT32_VAL(DSS, CHECKSUM)) {
1424                 DSS_NOTICE_LOG(("adding --checksum to rsync options\n"));
1425                 (*argv)[i++] = dss_strdup("--checksum");
1426         }
1427         for (j = 0; j < OPT_GIVEN(DSS, RSYNC_OPTION); j++)
1428                 (*argv)[i++] = dss_strdup(lls_string_val(j,
1429                         OPT_RESULT(DSS, RSYNC_OPTION)));
1430         if (name_of_reference_snapshot) {
1431                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1432                 (*argv)[i++] = make_message("--link-dest=../%s",
1433                         name_of_reference_snapshot);
1434         } else
1435                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1436         logname = dss_logname();
1437         if (use_rsync_locally(logname)) {
1438                 for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++)
1439                         (*argv)[i++] = dss_strdup(lls_string_val(j,
1440                                 OPT_RESULT(DSS, SOURCE_DIR)));
1441         } else {
1442                 /*
1443                  * dss-1.0 and earlier did not support multiple source
1444                  * directories.  These versions appended a slash to the end of
1445                  * the source directory to make sure that only the contents of
1446                  * the single source directory, but not the directory itself,
1447                  * are copied to the destination. For multiple source
1448                  * directories, however, this is not a good idea because the
1449                  * source directories may well contain identical file names,
1450                  * which would then be copied to the same location on the
1451                  * destination, overwriting each other. Moreover, we want the
1452                  * directory on the destination match the source. To preserve
1453                  * the old behaviour, we thus have to special-case N=1.
1454                  */
1455                 for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++) {
1456                         (*argv)[i++] = make_message("%s@%s:%s%s",
1457                                 OPT_GIVEN(DSS, REMOTE_USER)?
1458                                         OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
1459                                 OPT_STRING_VAL(DSS, REMOTE_HOST),
1460                                 lls_string_val(j, OPT_RESULT(DSS, SOURCE_DIR)),
1461                                 OPT_GIVEN(DSS, SOURCE_DIR) == 1? "/" : ""
1462                         );
1463                 }
1464         }
1465         free(logname);
1466         *num = get_current_time();
1467         (*argv)[i++] = incomplete_name(*num);
1468         (*argv)[i++] = NULL;
1469         for (j = 0; j < i; j++)
1470                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1471 }
1472
1473 static void free_rsync_argv(char **argv)
1474 {
1475         int i;
1476
1477         if (!argv)
1478                 return;
1479         for (i = 0; argv[i]; i++)
1480                 free(argv[i]);
1481         free(argv);
1482 }
1483
1484 static int create_snapshot(char **argv)
1485 {
1486         int ret;
1487
1488         assert(argv);
1489         ret = rename_resume_snap(current_snapshot_creation_time);
1490         if (ret < 0)
1491                 return ret;
1492         dss_exec(&create_pid, argv[0], argv);
1493         snapshot_creation_status = HS_RUNNING;
1494         return ret;
1495 }
1496
1497 static int select_loop(void)
1498 {
1499         int ret;
1500         /* check every 60 seconds for free disk space */
1501         struct timeval tv;
1502         char **rsync_argv = NULL;
1503
1504         for (;;) {
1505                 fd_set rfds;
1506                 struct timeval *tvp;
1507
1508                 if (remove_pid)
1509                         tvp = NULL; /* sleep until rm hook/process dies */
1510                 else { /* sleep one minute */
1511                         tv.tv_sec = 60;
1512                         tv.tv_usec = 0;
1513                         tvp = &tv;
1514                 }
1515                 FD_ZERO(&rfds);
1516                 FD_SET(signal_pipe, &rfds);
1517                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1518                 if (ret < 0)
1519                         goto out;
1520                 if (FD_ISSET(signal_pipe, &rfds)) {
1521                         ret = handle_signal();
1522                         if (ret < 0)
1523                                 goto out;
1524                 }
1525                 if (remove_pid)
1526                         continue;
1527                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1528                         ret = exec_rm();
1529                         if (ret < 0)
1530                                 goto out;
1531                         continue;
1532                 }
1533                 if (snapshot_removal_status == HS_SUCCESS) {
1534                         post_remove_hook();
1535                         continue;
1536                 }
1537                 ret = try_to_free_disk_space();
1538                 if (ret < 0)
1539                         goto out;
1540                 if (snapshot_removal_status != HS_READY) {
1541                         stop_create_process();
1542                         continue;
1543                 }
1544                 restart_create_process();
1545                 switch (snapshot_creation_status) {
1546                 case HS_READY:
1547                         if (!next_snapshot_is_due())
1548                                 continue;
1549                         pre_create_hook();
1550                         continue;
1551                 case HS_PRE_RUNNING:
1552                 case HS_RUNNING:
1553                 case HS_POST_RUNNING:
1554                         continue;
1555                 case HS_PRE_SUCCESS:
1556                         if (!name_of_reference_snapshot) {
1557                                 free_rsync_argv(rsync_argv);
1558                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1559                         }
1560                         ret = create_snapshot(rsync_argv);
1561                         if (ret < 0)
1562                                 goto out;
1563                         continue;
1564                 case HS_NEEDS_RESTART:
1565                         if (!next_snapshot_is_due())
1566                                 continue;
1567                         ret = create_snapshot(rsync_argv);
1568                         if (ret < 0)
1569                                 goto out;
1570                         continue;
1571                 case HS_SUCCESS:
1572                         post_create_hook();
1573                         continue;
1574                 }
1575         }
1576 out:
1577         return ret;
1578 }
1579
1580 static void exit_hook(int exit_code)
1581 {
1582         pid_t pid;
1583         char **argv, *tmp = dss_strdup(OPT_STRING_VAL(DSS, EXIT_HOOK));
1584         unsigned n = split_args(tmp, &argv);
1585
1586         n++;
1587         argv = dss_realloc(argv, (n + 1) * sizeof(char *));
1588         argv[n - 1] = dss_strdup(dss_strerror(-exit_code));
1589         argv[n] = NULL;
1590         dss_exec(&pid, argv[0], argv);
1591         free(argv[n - 1]);
1592         free(argv);
1593         free(tmp);
1594 }
1595
1596 static void lock_dss_or_die(void)
1597 {
1598         char *config_file = get_config_file_name();
1599         int ret = lock_dss(config_file);
1600
1601         free(config_file);
1602         if (ret < 0) {
1603                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1604                 exit(EXIT_FAILURE);
1605         }
1606 }
1607
1608 static int com_run(void)
1609 {
1610         int ret, fd = -1;
1611         char *config_file;
1612         pid_t pid;
1613
1614         if (OPT_GIVEN(DSS, DRY_RUN)) {
1615                 DSS_ERROR_LOG(("dry run not supported by this command\n"));
1616                 return -E_SYNTAX;
1617         }
1618         config_file = get_config_file_name();
1619         ret = get_dss_pid(config_file, &pid);
1620         free(config_file);
1621         if (ret >= 0) {
1622                 DSS_ERROR_LOG(("pid %d\n", (int)pid));
1623                 return -E_ALREADY_RUNNING;
1624         }
1625         if (OPT_GIVEN(RUN, DAEMON)) {
1626                 fd = daemon_init();
1627                 daemonized = true;
1628                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1629         }
1630         lock_dss_or_die();
1631         dump_dss_config("startup");
1632         ret = install_sighandler(SIGHUP);
1633         if (ret < 0)
1634                 return ret;
1635         if (fd >= 0) {
1636                 ret = write(fd, "\0", 1);
1637                 if (ret != 1) {
1638                         DSS_ERROR_LOG(("write to daemon pipe returned %d\n",
1639                                 ret));
1640                         if (ret < 0)
1641                                 return -ERRNO_TO_DSS_ERROR(errno);
1642                         return -E_BUG;
1643                 }
1644         }
1645         ret = select_loop();
1646         if (ret >= 0) /* impossible */
1647                 ret = -E_BUG;
1648         kill_children();
1649         exit_hook(ret);
1650         while (wait(NULL) >= 0 || errno != ECHILD)
1651                 ; /* still have children to wait for */
1652         return ret;
1653 }
1654 EXPORT_CMD_HANDLER(run);
1655
1656 static int com_prune(void)
1657 {
1658         int ret;
1659         struct snapshot_list sl;
1660         struct snapshot *victim;
1661         struct disk_space ds;
1662         char *why;
1663         bool try_hard;
1664
1665         lock_dss_or_die();
1666         switch (OPT_UINT32_VAL(PRUNE, DISK_SPACE)) {
1667         case FDS_LOW: try_hard = true; break;
1668         case FDS_HIGH: try_hard = false; break;
1669         default:
1670                 ret = get_disk_space(".", &ds);
1671                 if (ret < 0)
1672                         return ret;
1673                 log_disk_space(&ds);
1674                 try_hard = disk_space_low(&ds);
1675         }
1676         dss_get_snapshot_list(&sl);
1677         victim = find_removable_snapshot(&sl, try_hard, &why);
1678         if (!victim) {
1679                 dss_msg("nothing to prune\n");
1680                 ret = 0;
1681                 goto free_sl;
1682         }
1683         if (OPT_GIVEN(DSS, DRY_RUN)) {
1684                 dss_msg("picking %s snapshot %s (interval = %i)\n",
1685                         why, victim->name, victim->interval);
1686                 ret = 0;
1687                 goto free_why;
1688         }
1689         pre_remove_hook(victim, why);
1690         if (snapshot_removal_status == HS_PRE_RUNNING) {
1691                 ret = wait_for_remove_process();
1692                 if (ret < 0)
1693                         goto free_why;
1694                 ret = -E_HOOK_FAILED;
1695                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1696                         goto free_why;
1697         }
1698         ret = exec_rm();
1699         if (ret < 0)
1700                 goto free_why;
1701         ret = wait_for_remove_process();
1702         if (ret < 0)
1703                 goto free_why;
1704         assert(snapshot_removal_status == HS_SUCCESS);
1705         post_remove_hook();
1706         assert(snapshot_removal_status == HS_POST_RUNNING);
1707         ret = wait_for_remove_process();
1708 free_why:
1709         free(why);
1710 free_sl:
1711         free_snapshot_list(&sl);
1712         return ret;
1713 }
1714 EXPORT_CMD_HANDLER(prune);
1715
1716 static int com_create(void)
1717 {
1718         int ret, status;
1719         char **rsync_argv;
1720
1721         lock_dss_or_die();
1722         if (OPT_GIVEN(DSS, DRY_RUN)) {
1723                 int i;
1724                 char *msg = NULL;
1725                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1726                 for (i = 0; rsync_argv[i]; i++) {
1727                         char *tmp = msg;
1728                         msg = make_message("%s%s%s", tmp? tmp : "",
1729                                 tmp? " " : "", rsync_argv[i]);
1730                         free(tmp);
1731                 }
1732                 free_rsync_argv(rsync_argv);
1733                 dss_msg("%s\n", msg);
1734                 free(msg);
1735                 return 1;
1736         }
1737         pre_create_hook();
1738         if (create_pid) {
1739                 ret = wait_for_process(create_pid, &status);
1740                 if (ret < 0)
1741                         return ret;
1742                 ret = handle_pre_create_hook_exit(status);
1743                 if (ret <= 0) /* error, or pre-create failed */
1744                         return ret;
1745         }
1746         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1747         ret = create_snapshot(rsync_argv);
1748         if (ret < 0)
1749                 goto out;
1750         ret = wait_for_process(create_pid, &status);
1751         if (ret < 0)
1752                 goto out;
1753         ret = handle_rsync_exit(status);
1754         if (ret < 0)
1755                 goto out;
1756         post_create_hook();
1757         if (create_pid)
1758                 ret = wait_for_process(create_pid, &status);
1759 out:
1760         free_rsync_argv(rsync_argv);
1761         return ret;
1762 }
1763 EXPORT_CMD_HANDLER(create);
1764
1765 static int com_ls(void)
1766 {
1767         int i;
1768         struct snapshot_list sl;
1769         struct snapshot *s;
1770         int64_t now = get_current_time();
1771
1772         dss_get_snapshot_list(&sl);
1773         FOR_EACH_SNAPSHOT(s, i, &sl) {
1774                 int64_t d;
1775                 if (s->flags & SS_COMPLETE)
1776                         d = (s->completion_time - s->creation_time) / 60;
1777                 else
1778                         d = (now - s->creation_time) / 60;
1779                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval,
1780                         s->name, d / 60, d % 60);
1781         }
1782         free_snapshot_list(&sl);
1783         return 1;
1784 }
1785 EXPORT_CMD_HANDLER(ls);
1786
1787 static int com_configtest(void)
1788 {
1789         printf("Syntax Ok\n");
1790         return 0;
1791 }
1792 EXPORT_CMD_HANDLER(configtest);
1793
1794 static int setup_signal_handling(void)
1795 {
1796         int ret;
1797
1798         DSS_INFO_LOG(("setting up signal handlers\n"));
1799         signal_pipe = signal_init(); /* always successful */
1800         ret = install_sighandler(SIGINT);
1801         if (ret < 0)
1802                 return ret;
1803         ret = install_sighandler(SIGTERM);
1804         if (ret < 0)
1805                 return ret;
1806         return install_sighandler(SIGCHLD);
1807 }
1808
1809 static void handle_version_and_help(void)
1810 {
1811         char *txt;
1812
1813         if (OPT_GIVEN(DSS, DETAILED_HELP))
1814                 txt = lls_long_help(CMD_PTR(DSS));
1815         else if (OPT_GIVEN(DSS, HELP))
1816                 txt = lls_short_help(CMD_PTR(DSS));
1817         else if (OPT_GIVEN(DSS, VERSION))
1818                 txt = make_message("%s\n", VERSION_STRING);
1819         else
1820                 return;
1821         printf("%s", txt);
1822         free(txt);
1823         exit(EXIT_SUCCESS);
1824 }
1825
1826 static void show_subcommand_summary(void)
1827 {
1828         const struct lls_command *cmd;
1829         int i;
1830
1831         printf("Available subcommands:\n");
1832         for (i = 1; (cmd = lls_cmd(i, dss_suite)); i++) {
1833                 const char *name = lls_command_name(cmd);
1834                 const char *purpose = lls_purpose(cmd);
1835                 printf("%-11s%s\n", name, purpose);
1836         }
1837         exit(EXIT_SUCCESS);
1838 }
1839
1840 int main(int argc, char **argv)
1841 {
1842         int ret;
1843         char *errctx = NULL;
1844         unsigned num_inputs;
1845         const struct dss_user_data *ud;
1846
1847         ret = lls_parse(argc, argv, CMD_PTR(DSS), &cmdline_lpr, &errctx);
1848         if (ret < 0) {
1849                 ret = lopsub_error(ret, &errctx);
1850                 goto out;
1851         }
1852         lpr = cmdline_lpr;
1853         ret = parse_config_file(false /* no SIGHUP */, CMD_PTR(DSS));
1854         if (ret < 0)
1855                 goto out;
1856         handle_version_and_help();
1857         num_inputs = lls_num_inputs(lpr);
1858         if (num_inputs == 0)
1859                 show_subcommand_summary();
1860         ret = lls_lookup_subcmd(argv[argc - num_inputs], dss_suite, &errctx);
1861         if (ret < 0) {
1862                 ret = lopsub_error(ret, &errctx);
1863                 goto out;
1864         }
1865         subcmd = lls_cmd(ret, dss_suite);
1866         ret = lls_parse(num_inputs, argv + argc - num_inputs, subcmd,
1867                 &cmdline_sublpr, &errctx);
1868         if (ret < 0) {
1869                 ret = lopsub_error(ret, &errctx);
1870                 goto out;
1871         }
1872         sublpr = cmdline_sublpr;
1873         ret = parse_config_file(false /* no SIGHUP */, subcmd);
1874         if (ret < 0)
1875                 goto out;
1876         ret = check_config();
1877         if (ret < 0)
1878                 goto out;
1879         ret = setup_signal_handling();
1880         if (ret < 0)
1881                 goto out;
1882         ud = lls_user_data(subcmd);
1883         ret = ud->handler();
1884         signal_shutdown();
1885 out:
1886         if (ret < 0) {
1887                 if (errctx)
1888                         DSS_ERROR_LOG(("%s\n", errctx));
1889                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1890         }
1891         free(errctx);
1892         lls_free_parse_result(lpr, CMD_PTR(DSS));
1893         if (lpr != cmdline_lpr)
1894                 lls_free_parse_result(cmdline_lpr, CMD_PTR(DSS));
1895         lls_free_parse_result(sublpr, subcmd);
1896         if (sublpr != cmdline_sublpr)
1897                 lls_free_parse_result(cmdline_sublpr, subcmd);
1898         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1899 }