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 /** The snapshot currently being removed. */
  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                 "logile: %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 /*
 578  * return: 0: no redundant snapshots, 1: rm process started, negative: error
 579  */
 580 static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 581 {
 582         int i, interval;
 583         struct snapshot *s;
 584         unsigned missing = 0;
 585         uint32_t N = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
 586
 587         DSS_DEBUG_LOG(("looking for intervals containing too many snapshots\n"));
 588         for (interval = N - 1; interval >= 0; interval--) {
 589                 unsigned keep = desired_number_of_snapshots(interval, N);
 590                 unsigned num = sl->interval_count[interval];
 591                 struct snapshot *victim = NULL, *prev = NULL;
 592                 int64_t score = LONG_MAX;
 593
 594                 if (keep >= num)
 595                         missing += keep - num;
 596                 if (keep + missing >= num)
 597                         continue;
 598                 /* redundant snapshot in this interval, pick snapshot with lowest score */
 599                 FOR_EACH_SNAPSHOT(s, i, sl) {
 600                         int64_t this_score;
 601
 602                         if (snapshot_is_being_created(s))
 603                                 continue;
 604                         if (is_reference_snapshot(s))
 605                                 continue;
 606                         if (s->interval > interval) {
 607                                 prev = s;
 608                                 continue;
 609                         }
 610                         if (s->interval < interval)
 611                                 break;
 612                         if (!victim) {
 613                                 victim = s;
 614                                 prev = s;
 615                                 continue;
 616                         }
 617                         assert(prev);
 618                         /* check if s is a better victim */
 619                         this_score = s->creation_time - prev->creation_time;
 620                         assert(this_score >= 0);
 621                         if (this_score < score) {
 622                                 score = this_score;
 623                                 victim = s;
 624                         }
 625                         prev = s;
 626                 }
 627                 assert(victim);
 628                 return victim;
 629         }
 630         return NULL;
 631 }
 632
 633 static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 634 {
 635         int i;
 636         struct snapshot *s;
 637
 638         DSS_DEBUG_LOG(("looking for snapshots belonging to intervals >= %d\n",
 639                 OPT_UINT32_VAL(DSS, NUM_INTERVALS)));
 640         FOR_EACH_SNAPSHOT(s, i, sl) {
 641                 if (snapshot_is_being_created(s))
 642                         continue;
 643                 if (is_reference_snapshot(s))
 644                         continue;
 645                 if (s->interval < OPT_UINT32_VAL(DSS, NUM_INTERVALS))
 646                         continue;
 647                 return s;
 648         }
 649         return NULL;
 650 }
 651
 652 static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 653 {
 654         int i, num_complete;
 655         struct snapshot *s, *ref = NULL;
 656
 657         num_complete = num_complete_snapshots(sl);
 658         if (num_complete <= OPT_UINT32_VAL(DSS, MIN_COMPLETE))
 659                 return NULL;
 660         FOR_EACH_SNAPSHOT(s, i, sl) {
 661                 if (snapshot_is_being_created(s))
 662                         continue;
 663                 if (is_reference_snapshot(s)) { /* avoid this one */
 664                         ref = s;
 665                         continue;
 666                 }
 667                 DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 668                 return s;
 669         }
 670         assert(ref);
 671         DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
 672         return ref;
 673 }
 674
 675 static int rename_incomplete_snapshot(int64_t start)
 676 {
 677         char *old_name;
 678         int ret;
 679         int64_t now;
 680
 681         /*
 682          * We don't want the dss_rename() below to fail with EEXIST because the
 683          * last complete snapshot was created (and completed) in the same
 684          * second as this one.
 685          */
 686         while ((now = get_current_time()) == start)
 687                 sleep(1);
 688         free(path_to_last_complete_snapshot);
 689         ret = complete_name(start, now, &path_to_last_complete_snapshot);
 690         if (ret < 0)
 691                 return ret;
 692         old_name = incomplete_name(start);
 693         ret = dss_rename(old_name, path_to_last_complete_snapshot);
 694         if (ret >= 0)
 695                 DSS_NOTICE_LOG(("%s -> %s\n", old_name,
 696                         path_to_last_complete_snapshot));
 697         free(old_name);
 698         return ret;
 699 }
 700
 701 static int try_to_free_disk_space(void)
 702 {
 703         int ret;
 704         struct snapshot_list sl;
 705         struct snapshot *victim;
 706         struct timeval now;
 707         const char *why;
 708         int low_disk_space;
 709
 710         ret = disk_space_low(NULL);
 711         if (ret < 0)
 712                 return ret;
 713         low_disk_space = ret;
 714         gettimeofday(&now, NULL);
 715         if (tv_diff(&next_removal_check, &now, NULL) > 0)
 716                 return 0;
 717         if (!low_disk_space) {
 718                 if (OPT_GIVEN(DSS, KEEP_REDUNDANT))
 719                         return 0;
 720                 if (snapshot_creation_status != HS_READY)
 721                         return 0;
 722                 if (next_snapshot_is_due())
 723                         return 0;
 724         }
 725         /*
 726          * Idle and --keep_redundant not given, or low disk space. Look at
 727          * existing snapshots.
 728          */
 729         dss_get_snapshot_list(&sl);
 730         ret = 0;
 731         /*
 732          * Don't remove anything if there is free space and we have fewer
 733          * snapshots than configured, plus one. This way there is always one
 734          * snapshot that can be recycled.
 735          */
 736         if (!low_disk_space && sl.num_snapshots <=
 737                         1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
 738                 goto out;
 739         why = "outdated";
 740         victim = find_outdated_snapshot(&sl);
 741         if (victim)
 742                 goto remove;
 743         why = "redundant";
 744         victim = find_redundant_snapshot(&sl);
 745         if (victim)
 746                 goto remove;
 747         why = "orphaned";
 748         victim = find_orphaned_snapshot(&sl);
 749         if (victim)
 750                 goto remove;
 751         /* try harder only if disk space is low */
 752         if (!low_disk_space)
 753                 goto out;
 754         DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
 755         why = "oldest";
 756         victim = find_oldest_removable_snapshot(&sl);
 757         if (victim)
 758                 goto remove;
 759         DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
 760         ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 761         goto out;
 762 remove:
 763         pre_remove_hook(victim, why);
 764 out:
 765         free_snapshot_list(&sl);
 766         return ret;
 767 }
 768
 769 static void post_create_hook(void)
 770 {
 771         char *cmd = make_message("%s %s/%s",
 772                 OPT_STRING_VAL(DSS, POST_CREATE_HOOK),
 773                 OPT_STRING_VAL(DSS, DEST_DIR), path_to_last_complete_snapshot);
 774         DSS_NOTICE_LOG(("executing %s\n", cmd));
 775         dss_exec_cmdline_pid(&create_pid, cmd);
 776         free(cmd);
 777         snapshot_creation_status = HS_POST_RUNNING;
 778 }
 779
 780 static void post_remove_hook(void)
 781 {
 782         char *cmd;
 783         struct snapshot *s = snapshot_currently_being_removed;
 784
 785         assert(s);
 786
 787         cmd = make_message("%s %s/%s", OPT_STRING_VAL(DSS, POST_REMOVE_HOOK),
 788                 OPT_STRING_VAL(DSS, DEST_DIR), s->name);
 789         DSS_NOTICE_LOG(("executing %s\n", cmd));
 790         dss_exec_cmdline_pid(&remove_pid, cmd);
 791         free(cmd);
 792         snapshot_removal_status = HS_POST_RUNNING;
 793 }
 794
 795 static void dss_kill(pid_t pid, int sig, const char *msg)
 796 {
 797         const char *signame, *process_name;
 798
 799         if (pid == 0)
 800                 return;
 801         switch (sig) {
 802         case SIGTERM: signame = "TERM"; break;
 803         case SIGSTOP: signame = "STOP"; break;
 804         case SIGCONT: signame = "CONT"; break;
 805         default: signame = "????";
 806         }
 807
 808         if (pid == create_pid)
 809                 process_name = "create";
 810         else if (pid == remove_pid)
 811                 process_name = "remove";
 812         else process_name = "??????";
 813
 814         if (msg)
 815                 DSS_INFO_LOG(("%s\n", msg));
 816         DSS_DEBUG_LOG(("sending signal %d (%s) to pid %d (%s process)\n",
 817                 sig, signame, (int)pid, process_name));
 818         if (kill(pid, sig) >= 0)
 819                 return;
 820         DSS_INFO_LOG(("failed to send signal %d (%s) to pid %d (%s process)\n",
 821                 sig, signame, (int)pid, process_name));
 822 }
 823
 824 static void stop_create_process(void)
 825 {
 826         if (create_process_stopped)
 827                 return;
 828         dss_kill(create_pid, SIGSTOP, "suspending create process");
 829         create_process_stopped = 1;
 830 }
 831
 832 static void restart_create_process(void)
 833 {
 834         if (!create_process_stopped)
 835                 return;
 836         dss_kill(create_pid, SIGCONT, "resuming create process");
 837         create_process_stopped = 0;
 838 }
 839
 840 /**
 841  * Print a log message about the exit status of a child.
 842  */
 843 static void log_termination_msg(pid_t pid, int status)
 844 {
 845         if (WIFEXITED(status))
 846                 DSS_INFO_LOG(("child %i exited. Exit status: %i\n", (int)pid,
 847                         WEXITSTATUS(status)));
 848         else if (WIFSIGNALED(status))
 849                 DSS_NOTICE_LOG(("child %i was killed by signal %i\n", (int)pid,
 850                         WTERMSIG(status)));
 851         else
 852                 DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
 853 }
 854
 855 static int wait_for_process(pid_t pid, int *status)
 856 {
 857         int ret;
 858
 859         DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
 860         for (;;) {
 861                 fd_set rfds;
 862
 863                 FD_ZERO(&rfds);
 864                 FD_SET(signal_pipe, &rfds);
 865                 ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
 866                 if (ret < 0)
 867                         break;
 868                 ret = next_signal();
 869                 if (!ret)
 870                         continue;
 871                 if (ret == SIGCHLD) {
 872                         ret = waitpid(pid, status, 0);
 873                         if (ret >= 0)
 874                                 break;
 875                         if (errno != EINTR) { /* error */
 876                                 ret = -ERRNO_TO_DSS_ERROR(errno);
 877                                 break;
 878                         }
 879                 }
 880                 /* SIGINT or SIGTERM */
 881                 dss_kill(pid, SIGTERM, "killing child process");
 882         }
 883         if (ret < 0)
 884                 DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
 885         else
 886                 log_termination_msg(pid, *status);
 887         return ret;
 888 }
 889
 890 static void handle_pre_remove_exit(int status)
 891 {
 892         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 893                 snapshot_removal_status = HS_READY;
 894                 gettimeofday(&next_removal_check, NULL);
 895                 next_removal_check.tv_sec += 60;
 896                 return;
 897         }
 898         snapshot_removal_status = HS_PRE_SUCCESS;
 899 }
 900
 901 static int handle_rm_exit(int status)
 902 {
 903         if (!WIFEXITED(status)) {
 904                 snapshot_removal_status = HS_READY;
 905                 return -E_INVOLUNTARY_EXIT;
 906         }
 907         if (WEXITSTATUS(status)) {
 908                 snapshot_removal_status = HS_READY;
 909                 return -E_BAD_EXIT_CODE;
 910         }
 911         snapshot_removal_status = HS_SUCCESS;
 912         return 1;
 913 }
 914
 915 static void handle_post_remove_exit(void)
 916 {
 917         snapshot_removal_status = HS_READY;
 918 }
 919
 920 static int handle_remove_exit(int status)
 921 {
 922         int ret;
 923         struct snapshot *s = snapshot_currently_being_removed;
 924
 925         assert(s);
 926         switch (snapshot_removal_status) {
 927         case HS_PRE_RUNNING:
 928                 handle_pre_remove_exit(status);
 929                 ret = 1;
 930                 break;
 931         case HS_RUNNING:
 932                 ret = handle_rm_exit(status);
 933                 break;
 934         case HS_POST_RUNNING:
 935                 handle_post_remove_exit();
 936                 ret = 1;
 937                 break;
 938         default:
 939                 ret = -E_BUG;
 940         }
 941         if (snapshot_removal_status == HS_READY) {
 942                 free(s->name);
 943                 free(s);
 944                 snapshot_currently_being_removed = NULL;
 945         }
 946         remove_pid = 0;
 947         return ret;
 948 }
 949
 950 static int wait_for_remove_process(void)
 951 {
 952         int status, ret;
 953
 954         assert(remove_pid);
 955         assert(
 956                 snapshot_removal_status == HS_PRE_RUNNING ||
 957                 snapshot_removal_status == HS_RUNNING ||
 958                 snapshot_removal_status == HS_POST_RUNNING
 959         );
 960         ret = wait_for_process(remove_pid, &status);
 961         if (ret < 0)
 962                 return ret;
 963         return handle_remove_exit(status);
 964 }
 965
 966 static int handle_rsync_exit(int status)
 967 {
 968         int es, ret;
 969
 970         if (!WIFEXITED(status)) {
 971                 DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
 972                 ret = -E_INVOLUNTARY_EXIT;
 973                 snapshot_creation_status = HS_READY;
 974                 goto out;
 975         }
 976         es = WEXITSTATUS(status);
 977         /*
 978          * Restart rsync on non-fatal errors:
 979          * 24: Partial transfer due to vanished source files
 980          */
 981         if (es != 0 && es != 24) {
 982                 DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
 983                         es, ++num_consecutive_rsync_errors));
 984                 if (!logfile) { /* called by com_run() */
 985                         ret = -E_BAD_EXIT_CODE;
 986                         goto out;
 987                 }
 988                 if (num_consecutive_rsync_errors >
 989                                 OPT_UINT32_VAL(RUN, MAX_RSYNC_ERRORS)) {
 990                         ret = -E_TOO_MANY_RSYNC_ERRORS;
 991                         snapshot_creation_status = HS_READY;
 992                         goto out;
 993                 }
 994                 DSS_WARNING_LOG(("restarting rsync process\n"));
 995                 snapshot_creation_status = HS_NEEDS_RESTART;
 996                 next_snapshot_time = get_current_time() + 60;
 997                 ret = 1;
 998                 goto out;
 999         }
1000         num_consecutive_rsync_errors = 0;
1001         ret = rename_incomplete_snapshot(current_snapshot_creation_time);
1002         if (ret < 0)
1003                 goto out;
1004         snapshot_creation_status = HS_SUCCESS;
1005         free(name_of_reference_snapshot);
1006         name_of_reference_snapshot = NULL;
1007 out:
1008         create_process_stopped = 0;
1009         return ret;
1010 }
1011
1012 static int handle_pre_create_hook_exit(int status)
1013 {
1014         int es, ret;
1015         static int warn_count;
1016
1017         if (!WIFEXITED(status)) {
1018                 snapshot_creation_status = HS_READY;
1019                 ret = -E_INVOLUNTARY_EXIT;
1020                 goto out;
1021         }
1022         es = WEXITSTATUS(status);
1023         if (es) {
1024                 if (!warn_count--) {
1025                         DSS_NOTICE_LOG(("pre_create_hook %s returned %d\n",
1026                                 OPT_STRING_VAL(DSS, PRE_CREATE_HOOK), es));
1027                         DSS_NOTICE_LOG(("deferring snapshot creation...\n"));
1028                         warn_count = 60; /* warn only once per hour */
1029                 }
1030                 next_snapshot_time = get_current_time() + 60;
1031                 snapshot_creation_status = HS_READY;
1032                 ret = 0;
1033                 goto out;
1034         }
1035         warn_count = 0;
1036         snapshot_creation_status = HS_PRE_SUCCESS;
1037         ret = 1;
1038 out:
1039         return ret;
1040 }
1041
1042 static int handle_sigchld(void)
1043 {
1044         pid_t pid;
1045         int status, ret = reap_child(&pid, &status);
1046
1047         if (ret <= 0)
1048                 return ret;
1049
1050         if (pid == create_pid) {
1051                 switch (snapshot_creation_status) {
1052                 case HS_PRE_RUNNING:
1053                         ret = handle_pre_create_hook_exit(status);
1054                         break;
1055                 case HS_RUNNING:
1056                         ret = handle_rsync_exit(status);
1057                         break;
1058                 case HS_POST_RUNNING:
1059                         snapshot_creation_status = HS_READY;
1060                         ret = 1;
1061                         break;
1062                 default:
1063                         DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
1064                                 snapshot_creation_status));
1065                         return -E_BUG;
1066                 }
1067                 create_pid = 0;
1068                 return ret;
1069         }
1070         if (pid == remove_pid) {
1071                 ret = handle_remove_exit(status);
1072                 if (ret < 0)
1073                         return ret;
1074                 return ret;
1075         }
1076         DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
1077         return -E_BUG;
1078 }
1079
1080 /* also checks if . is a mountpoint, if --mountpoint was given */
1081 static int change_to_dest_dir(void)
1082 {
1083         int ret;
1084         const char *dd = OPT_STRING_VAL(DSS, DEST_DIR);
1085         struct stat dot, dotdot;
1086
1087         DSS_INFO_LOG(("changing cwd to %s\n", dd));
1088         if (chdir(dd) < 0) {
1089                 ret = -ERRNO_TO_DSS_ERROR(errno);
1090                 DSS_ERROR_LOG(("could not change cwd to %s\n", dd));
1091                 return ret;
1092         }
1093         if (!OPT_GIVEN(DSS, MOUNTPOINT))
1094                 return 0;
1095         if (stat(".", &dot) < 0) {
1096                 ret = -ERRNO_TO_DSS_ERROR(errno);
1097                 DSS_ERROR_LOG(("could not stat .\n"));
1098                 return ret;
1099         }
1100         if (stat("..", &dotdot) < 0) {
1101                 ret = -ERRNO_TO_DSS_ERROR(errno);
1102                 DSS_ERROR_LOG(("could not stat ..\n"));
1103                 return ret;
1104         }
1105         if (dot.st_dev == dotdot.st_dev && dot.st_ino != dotdot.st_ino) {
1106                 DSS_ERROR_LOG(("mountpoint check failed for %s\n", dd));
1107                 return -E_MOUNTPOINT;
1108         }
1109         return 1;
1110 }
1111
1112 static int check_config(void)
1113 {
1114         int ret;
1115         uint32_t unit_interval = OPT_UINT32_VAL(DSS, UNIT_INTERVAL);
1116         uint32_t num_intervals = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
1117
1118         if (unit_interval == 0) {
1119                 DSS_ERROR_LOG(("bad unit interval: %i\n", unit_interval));
1120                 return -E_INVALID_NUMBER;
1121         }
1122         DSS_DEBUG_LOG(("unit interval: %i day(s)\n", unit_interval));
1123
1124         if (num_intervals == 0 || num_intervals > 30) {
1125                 DSS_ERROR_LOG(("bad number of intervals: %i\n", num_intervals));
1126                 return -E_INVALID_NUMBER;
1127         }
1128         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE))
1129                 if (!OPT_GIVEN(DSS, SOURCE_DIR)) {
1130                         DSS_ERROR_LOG(("--source-dir required\n"));
1131                         return -E_SYNTAX;
1132                 }
1133         if (subcmd == CMD_PTR(RUN) || subcmd == CMD_PTR(CREATE)
1134                         || subcmd == CMD_PTR(LS) || subcmd == CMD_PTR(PRUNE)) {
1135                 if (!OPT_GIVEN(DSS, DEST_DIR)) {
1136                         DSS_ERROR_LOG(("--dest-dir required\n"));
1137                         return -E_SYNTAX;
1138                 }
1139                 ret = change_to_dest_dir();
1140                 if (ret < 0)
1141                         return ret;
1142         }
1143         DSS_DEBUG_LOG(("number of intervals: %i\n", num_intervals));
1144         return 1;
1145 }
1146
1147 static int lopsub_error(int lopsub_ret, char **errctx)
1148 {
1149         const char *msg = lls_strerror(-lopsub_ret);
1150         if (*errctx)
1151                 DSS_ERROR_LOG(("%s: %s\n", *errctx, msg));
1152         else
1153                 DSS_ERROR_LOG(("%s\n", msg));
1154         free(*errctx);
1155         *errctx = NULL;
1156         return -E_LOPSUB;
1157 }
1158
1159 static int parse_config_file(bool sighup, const struct lls_command *cmd)
1160 {
1161         int ret, fd = -1;
1162         char *config_file = get_config_file_name();
1163         struct stat statbuf;
1164         void *map;
1165         size_t sz;
1166         int cf_argc;
1167         char **cf_argv, *errctx = NULL;
1168         struct lls_parse_result *cf_lpr, *merged_lpr, *clpr;
1169         const char *subcmd_name;
1170
1171         ret = open(config_file, O_RDONLY);
1172         if (ret < 0) {
1173                 if (errno != ENOENT || OPT_GIVEN(DSS, CONFIG_FILE)) {
1174                         ret = -ERRNO_TO_DSS_ERROR(errno);
1175                         DSS_ERROR_LOG(("config file %s can not be opened\n",
1176                                 config_file));
1177                         goto out;
1178                 }
1179                 /* no config file -- nothing to do */
1180                 ret = 0;
1181                 goto success;
1182         }
1183         fd = ret;
1184         ret = fstat(fd, &statbuf);
1185         if (ret < 0) {
1186                 ret = -ERRNO_TO_DSS_ERROR(errno);
1187                 DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
1188                 goto close_fd;
1189         }
1190         sz = statbuf.st_size;
1191         if (sz == 0) { /* config file is empty -- nothing to do */
1192                 ret = 0;
1193                 goto success;
1194         }
1195         map = mmap(NULL, sz, PROT_READ, MAP_PRIVATE, fd, 0);
1196         if (map == MAP_FAILED) {
1197                 ret = -ERRNO_TO_DSS_ERROR(errno);
1198                 DSS_ERROR_LOG(("failed to mmap config file %s\n",
1199                         config_file));
1200                 goto close_fd;
1201         }
1202         if (cmd == CMD_PTR(DSS))
1203                 subcmd_name = NULL;
1204         else
1205                 subcmd_name = lls_command_name(cmd);
1206         ret = lls_convert_config(map, sz, subcmd_name, &cf_argv, &errctx);
1207         munmap(map, sz);
1208         if (ret < 0) {
1209                 DSS_ERROR_LOG(("failed to convert config file %s\n",
1210                         config_file));
1211                 ret = lopsub_error(ret, &errctx);
1212                 goto close_fd;
1213         }
1214         cf_argc = ret;
1215         ret = lls_parse(cf_argc, cf_argv, cmd, &cf_lpr, &errctx);
1216         lls_free_argv(cf_argv);
1217         if (ret < 0) {
1218                 ret = lopsub_error(ret, &errctx);
1219                 goto close_fd;
1220         }
1221         clpr = cmd == CMD_PTR(DSS)? cmdline_lpr : cmdline_sublpr;
1222         if (sighup) /* config file overrides command line */
1223                 ret = lls_merge(cf_lpr, clpr, cmd, &merged_lpr, &errctx);
1224         else /* command line options overrride config file options */
1225                 ret = lls_merge(clpr, cf_lpr, cmd, &merged_lpr, &errctx);
1226         lls_free_parse_result(cf_lpr, cmd);
1227         if (ret < 0) {
1228                 ret = lopsub_error(ret, &errctx);
1229                 goto close_fd;
1230         }
1231         ret = 1;
1232 success:
1233         assert(ret >= 0);
1234         DSS_DEBUG_LOG(("loglevel: %d\n", OPT_UINT32_VAL(DSS, LOGLEVEL)));
1235         if (cmd != CMD_PTR(DSS)) {
1236                 if (ret > 0) {
1237                         if (sublpr != cmdline_sublpr)
1238                                 lls_free_parse_result(sublpr, cmd);
1239                         sublpr = merged_lpr;
1240                 } else
1241                         sublpr = cmdline_sublpr;
1242         } else {
1243                 if (ret > 0) {
1244                         if (lpr != cmdline_lpr)
1245                                 lls_free_parse_result(lpr, cmd);
1246                         lpr = merged_lpr;
1247                 } else
1248                         lpr = cmdline_lpr;
1249         }
1250 close_fd:
1251         if (fd >= 0)
1252                 close(fd);
1253 out:
1254         free(config_file);
1255         if (ret < 0)
1256                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1257         return ret;
1258 }
1259
1260 static int handle_sighup(void)
1261 {
1262         int ret;
1263
1264         DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
1265         dump_dss_config("old");
1266         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(DSS));
1267         if (ret < 0)
1268                 return ret;
1269         ret = parse_config_file(true /* SIGHUP */, CMD_PTR(RUN));
1270         if (ret < 0)
1271                 return ret;
1272         ret = check_config();
1273         if (ret < 0)
1274                 return ret;
1275         close_log(logfile);
1276         logfile = NULL;
1277         if (OPT_GIVEN(RUN, DAEMON) || daemonized) {
1278                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1279                 log_welcome(OPT_UINT32_VAL(DSS, LOGLEVEL));
1280                 daemonized = true;
1281         }
1282         dump_dss_config("reloaded");
1283         invalidate_next_snapshot_time();
1284         return 1;
1285 }
1286
1287 static void kill_children(void)
1288 {
1289         restart_create_process();
1290         dss_kill(create_pid, SIGTERM, NULL);
1291         dss_kill(remove_pid, SIGTERM, NULL);
1292 }
1293
1294 static int handle_signal(void)
1295 {
1296         int sig, ret = next_signal();
1297
1298         if (ret <= 0)
1299                 goto out;
1300         sig = ret;
1301         switch (sig) {
1302         case SIGINT:
1303         case SIGTERM:
1304                 kill_children();
1305                 ret = -E_SIGNAL;
1306                 break;
1307         case SIGHUP:
1308                 ret = handle_sighup();
1309                 break;
1310         case SIGCHLD:
1311                 ret = handle_sigchld();
1312                 break;
1313         }
1314 out:
1315         if (ret < 0)
1316                 DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
1317         return ret;
1318 }
1319
1320 /*
1321  * We can not use rsync locally if the local user is different from the remote
1322  * user or if the src dir is not on the local host (or both).
1323  */
1324 static int use_rsync_locally(char *logname)
1325 {
1326         const char *h = OPT_STRING_VAL(DSS, REMOTE_HOST);
1327
1328         if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
1329                 return 0;
1330         if (OPT_GIVEN(DSS, REMOTE_USER) &&
1331                         strcmp(OPT_STRING_VAL(DSS, REMOTE_USER), logname))
1332                 return 0;
1333         return 1;
1334 }
1335
1336 static int rename_resume_snap(int64_t creation_time)
1337 {
1338         struct snapshot_list sl;
1339         struct snapshot *s = NULL;
1340         char *new_name = incomplete_name(creation_time);
1341         int ret;
1342         const char *why;
1343
1344         sl.num_snapshots = 0;
1345
1346         ret = 0;
1347         dss_get_snapshot_list(&sl);
1348         /*
1349          * Snapshot recycling: We first look at the newest snapshot. If this
1350          * snapshot happens to be incomplete, the last rsync process was
1351          * aborted and we reuse this one. Otherwise we look at snapshots which
1352          * could be removed (outdated and redundant snapshots) as candidates
1353          * for recycling. If no outdated/redundant snapshot exists, we check if
1354          * there is an orphaned snapshot, which likely is useless anyway.
1355          *
1356          * Only if no existing snapshot is suitable for recycling, we bite the
1357          * bullet and create a new one.
1358          */
1359         s = get_newest_snapshot(&sl);
1360         if (!s) /* no snapshots at all */
1361                 goto out;
1362         /* re-use last snapshot if it is incomplete */
1363         why = "aborted";
1364         if ((s->flags & SS_COMPLETE) == 0)
1365                 goto out;
1366         why = "outdated";
1367         s = find_outdated_snapshot(&sl);
1368         if (s)
1369                 goto out;
1370         why = "redundant";
1371         s = find_redundant_snapshot(&sl);
1372         if (s)
1373                 goto out;
1374         why = "orphaned";
1375         s = find_orphaned_snapshot(&sl);
1376 out:
1377         if (s) {
1378                 DSS_NOTICE_LOG(("recycling %s snapshot %s\n", why, s->name));
1379                 ret = dss_rename(s->name, new_name);
1380         }
1381         if (ret >= 0)
1382                 DSS_NOTICE_LOG(("creating %s\n", new_name));
1383         free(new_name);
1384         free_snapshot_list(&sl);
1385         return ret;
1386 }
1387
1388 static void create_rsync_argv(char ***argv, int64_t *num)
1389 {
1390         char *logname;
1391         int i = 0, j, N = OPT_GIVEN(DSS, RSYNC_OPTION);
1392         struct snapshot_list sl;
1393         static bool seeded;
1394
1395         dss_get_snapshot_list(&sl);
1396         assert(!name_of_reference_snapshot);
1397         name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
1398         free_snapshot_list(&sl);
1399
1400         *argv = dss_malloc((15 + N) * sizeof(char *));
1401         (*argv)[i++] = dss_strdup("rsync");
1402         (*argv)[i++] = dss_strdup("-a");
1403         (*argv)[i++] = dss_strdup("--delete");
1404         if (!seeded) {
1405                 srandom((unsigned)time(NULL)); /* no need to be fancy here */
1406                 seeded = true;
1407         }
1408         if (1000 * (random() / (RAND_MAX + 1.0)) < OPT_UINT32_VAL(DSS, CHECKSUM)) {
1409                 DSS_NOTICE_LOG(("adding --checksum to rsync options\n"));
1410                 (*argv)[i++] = dss_strdup("--checksum");
1411         }
1412         for (j = 0; j < N; j++)
1413                 (*argv)[i++] = dss_strdup(lls_string_val(j,
1414                         OPT_RESULT(DSS, RSYNC_OPTION)));
1415         if (name_of_reference_snapshot) {
1416                 DSS_INFO_LOG(("using %s as reference\n", name_of_reference_snapshot));
1417                 (*argv)[i++] = make_message("--link-dest=../%s",
1418                         name_of_reference_snapshot);
1419         } else
1420                 DSS_INFO_LOG(("no suitable reference snapshot found\n"));
1421         logname = dss_logname();
1422         if (use_rsync_locally(logname))
1423                 (*argv)[i++] = dss_strdup(OPT_STRING_VAL(DSS, SOURCE_DIR));
1424         else
1425                 (*argv)[i++] = make_message("%s@%s:%s/",
1426                         OPT_GIVEN(DSS, REMOTE_USER)?
1427                                 OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
1428                         OPT_STRING_VAL(DSS, REMOTE_HOST),
1429                         OPT_STRING_VAL(DSS, SOURCE_DIR));
1430         free(logname);
1431         *num = get_current_time();
1432         (*argv)[i++] = incomplete_name(*num);
1433         (*argv)[i++] = NULL;
1434         for (j = 0; j < i; j++)
1435                 DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
1436 }
1437
1438 static void free_rsync_argv(char **argv)
1439 {
1440         int i;
1441
1442         if (!argv)
1443                 return;
1444         for (i = 0; argv[i]; i++)
1445                 free(argv[i]);
1446         free(argv);
1447 }
1448
1449 static int create_snapshot(char **argv)
1450 {
1451         int ret;
1452
1453         assert(argv);
1454         ret = rename_resume_snap(current_snapshot_creation_time);
1455         if (ret < 0)
1456                 return ret;
1457         dss_exec(&create_pid, argv[0], argv);
1458         snapshot_creation_status = HS_RUNNING;
1459         return ret;
1460 }
1461
1462 static int select_loop(void)
1463 {
1464         int ret;
1465         /* check every 60 seconds for free disk space */
1466         struct timeval tv;
1467         char **rsync_argv = NULL;
1468
1469         for (;;) {
1470                 fd_set rfds;
1471                 struct timeval *tvp;
1472
1473                 if (remove_pid)
1474                         tvp = NULL; /* sleep until rm hook/process dies */
1475                 else { /* sleep one minute */
1476                         tv.tv_sec = 60;
1477                         tv.tv_usec = 0;
1478                         tvp = &tv;
1479                 }
1480                 FD_ZERO(&rfds);
1481                 FD_SET(signal_pipe, &rfds);
1482                 ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
1483                 if (ret < 0)
1484                         goto out;
1485                 if (FD_ISSET(signal_pipe, &rfds)) {
1486                         ret = handle_signal();
1487                         if (ret < 0)
1488                                 goto out;
1489                 }
1490                 if (remove_pid)
1491                         continue;
1492                 if (snapshot_removal_status == HS_PRE_SUCCESS) {
1493                         ret = exec_rm();
1494                         if (ret < 0)
1495                                 goto out;
1496                         continue;
1497                 }
1498                 if (snapshot_removal_status == HS_SUCCESS) {
1499                         post_remove_hook();
1500                         continue;
1501                 }
1502                 ret = try_to_free_disk_space();
1503                 if (ret < 0)
1504                         goto out;
1505                 if (snapshot_removal_status != HS_READY) {
1506                         stop_create_process();
1507                         continue;
1508                 }
1509                 restart_create_process();
1510                 switch (snapshot_creation_status) {
1511                 case HS_READY:
1512                         if (!next_snapshot_is_due())
1513                                 continue;
1514                         pre_create_hook();
1515                         continue;
1516                 case HS_PRE_RUNNING:
1517                 case HS_RUNNING:
1518                 case HS_POST_RUNNING:
1519                         continue;
1520                 case HS_PRE_SUCCESS:
1521                         if (!name_of_reference_snapshot) {
1522                                 free_rsync_argv(rsync_argv);
1523                                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1524                         }
1525                         ret = create_snapshot(rsync_argv);
1526                         if (ret < 0)
1527                                 goto out;
1528                         continue;
1529                 case HS_NEEDS_RESTART:
1530                         if (!next_snapshot_is_due())
1531                                 continue;
1532                         ret = create_snapshot(rsync_argv);
1533                         if (ret < 0)
1534                                 goto out;
1535                         continue;
1536                 case HS_SUCCESS:
1537                         post_create_hook();
1538                         continue;
1539                 }
1540         }
1541 out:
1542         return ret;
1543 }
1544
1545 static void exit_hook(int exit_code)
1546 {
1547         pid_t pid;
1548         char **argv, *tmp = dss_strdup(OPT_STRING_VAL(DSS, EXIT_HOOK));
1549         unsigned n = split_args(tmp, &argv, " \t");
1550
1551         n++;
1552         argv = dss_realloc(argv, (n + 1) * sizeof(char *));
1553         argv[n - 1] = dss_strdup(dss_strerror(-exit_code));
1554         argv[n] = NULL;
1555         dss_exec(&pid, argv[0], argv);
1556         free(argv[n - 1]);
1557         free(argv);
1558         free(tmp);
1559 }
1560
1561 static void lock_dss_or_die(void)
1562 {
1563         char *config_file = get_config_file_name();
1564         int ret = lock_dss(config_file);
1565
1566         free(config_file);
1567         if (ret < 0) {
1568                 DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
1569                 exit(EXIT_FAILURE);
1570         }
1571 }
1572
1573 static int com_run(void)
1574 {
1575         int ret, fd = -1;
1576         char *config_file;
1577         pid_t pid;
1578
1579         if (OPT_GIVEN(DSS, DRY_RUN)) {
1580                 DSS_ERROR_LOG(("dry run not supported by this command\n"));
1581                 return -E_SYNTAX;
1582         }
1583         config_file = get_config_file_name();
1584         ret = get_dss_pid(config_file, &pid);
1585         free(config_file);
1586         if (ret >= 0) {
1587                 DSS_ERROR_LOG(("pid %d\n", (int)pid));
1588                 return -E_ALREADY_RUNNING;
1589         }
1590         if (OPT_GIVEN(RUN, DAEMON)) {
1591                 fd = daemon_init();
1592                 daemonized = true;
1593                 logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
1594         }
1595         lock_dss_or_die();
1596         dump_dss_config("startup");
1597         ret = install_sighandler(SIGHUP);
1598         if (ret < 0)
1599                 return ret;
1600         if (fd >= 0) {
1601                 ret = write(fd, "\0", 1);
1602                 if (ret != 1) {
1603                         DSS_ERROR_LOG(("write to daemon pipe returned %d\n",
1604                                 ret));
1605                         if (ret < 0)
1606                                 return -ERRNO_TO_DSS_ERROR(errno);
1607                         return -E_BUG;
1608                 }
1609         }
1610         ret = select_loop();
1611         if (ret >= 0) /* impossible */
1612                 ret = -E_BUG;
1613         kill_children();
1614         exit_hook(ret);
1615         while (wait(NULL) >= 0 || errno != ECHILD)
1616                 ; /* still have children to wait for */
1617         return ret;
1618 }
1619 EXPORT_CMD_HANDLER(run);
1620
1621 static int com_prune(void)
1622 {
1623         int ret;
1624         struct snapshot_list sl;
1625         struct snapshot *victim;
1626         struct disk_space ds;
1627         const char *why;
1628
1629         lock_dss_or_die();
1630         ret = get_disk_space(".", &ds);
1631         if (ret < 0)
1632                 return ret;
1633         log_disk_space(&ds);
1634         dss_get_snapshot_list(&sl);
1635         why = "outdated";
1636         victim = find_outdated_snapshot(&sl);
1637         if (victim)
1638                 goto rm;
1639         why = "redundant";
1640         victim = find_redundant_snapshot(&sl);
1641         if (victim)
1642                 goto rm;
1643         ret = 0;
1644         goto out;
1645 rm:
1646         if (OPT_GIVEN(DSS, DRY_RUN)) {
1647                 dss_msg("%s snapshot %s (interval = %i)\n",
1648                         why, victim->name, victim->interval);
1649                 ret = 0;
1650                 goto out;
1651         }
1652         pre_remove_hook(victim, why);
1653         if (snapshot_removal_status == HS_PRE_RUNNING) {
1654                 ret = wait_for_remove_process();
1655                 if (ret < 0)
1656                         goto out;
1657                 if (snapshot_removal_status != HS_PRE_SUCCESS)
1658                         goto out;
1659         }
1660         ret = exec_rm();
1661         if (ret < 0)
1662                 goto out;
1663         ret = wait_for_remove_process();
1664         if (ret < 0)
1665                 goto out;
1666         if (snapshot_removal_status != HS_SUCCESS)
1667                 goto out;
1668         post_remove_hook();
1669         if (snapshot_removal_status != HS_POST_RUNNING)
1670                 goto out;
1671         ret = wait_for_remove_process();
1672         if (ret < 0)
1673                 goto out;
1674         ret = 1;
1675 out:
1676         free_snapshot_list(&sl);
1677         return ret;
1678 }
1679 EXPORT_CMD_HANDLER(prune);
1680
1681 static int com_create(void)
1682 {
1683         int ret, status;
1684         char **rsync_argv;
1685
1686         lock_dss_or_die();
1687         if (OPT_GIVEN(DSS, DRY_RUN)) {
1688                 int i;
1689                 char *msg = NULL;
1690                 create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1691                 for (i = 0; rsync_argv[i]; i++) {
1692                         char *tmp = msg;
1693                         msg = make_message("%s%s%s", tmp? tmp : "",
1694                                 tmp? " " : "", rsync_argv[i]);
1695                         free(tmp);
1696                 }
1697                 free_rsync_argv(rsync_argv);
1698                 dss_msg("%s\n", msg);
1699                 free(msg);
1700                 return 1;
1701         }
1702         pre_create_hook();
1703         if (create_pid) {
1704                 ret = wait_for_process(create_pid, &status);
1705                 if (ret < 0)
1706                         return ret;
1707                 ret = handle_pre_create_hook_exit(status);
1708                 if (ret <= 0) /* error, or pre-create failed */
1709                         return ret;
1710         }
1711         create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
1712         ret = create_snapshot(rsync_argv);
1713         if (ret < 0)
1714                 goto out;
1715         ret = wait_for_process(create_pid, &status);
1716         if (ret < 0)
1717                 goto out;
1718         ret = handle_rsync_exit(status);
1719         if (ret < 0)
1720                 goto out;
1721         post_create_hook();
1722         if (create_pid)
1723                 ret = wait_for_process(create_pid, &status);
1724 out:
1725         free_rsync_argv(rsync_argv);
1726         return ret;
1727 }
1728 EXPORT_CMD_HANDLER(create);
1729
1730 static int com_ls(void)
1731 {
1732         int i;
1733         struct snapshot_list sl;
1734         struct snapshot *s;
1735         int64_t now = get_current_time();
1736
1737         dss_get_snapshot_list(&sl);
1738         FOR_EACH_SNAPSHOT(s, i, &sl) {
1739                 int64_t d;
1740                 if (s->flags & SS_COMPLETE)
1741                         d = (s->completion_time - s->creation_time) / 60;
1742                 else
1743                         d = (now - s->creation_time) / 60;
1744                 dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval,
1745                         s->name, d / 60, d % 60);
1746         }
1747         free_snapshot_list(&sl);
1748         return 1;
1749 }
1750 EXPORT_CMD_HANDLER(ls);
1751
1752 static int com_configtest(void)
1753 {
1754         printf("Syntax Ok\n");
1755         return 0;
1756 }
1757 EXPORT_CMD_HANDLER(configtest);
1758
1759 static int setup_signal_handling(void)
1760 {
1761         int ret;
1762
1763         DSS_INFO_LOG(("setting up signal handlers\n"));
1764         signal_pipe = signal_init(); /* always successful */
1765         ret = install_sighandler(SIGINT);
1766         if (ret < 0)
1767                 return ret;
1768         ret = install_sighandler(SIGTERM);
1769         if (ret < 0)
1770                 return ret;
1771         return install_sighandler(SIGCHLD);
1772 }
1773
1774 static void handle_version_and_help(void)
1775 {
1776         char *txt;
1777
1778         if (OPT_GIVEN(DSS, DETAILED_HELP))
1779                 txt = lls_long_help(CMD_PTR(DSS));
1780         else if (OPT_GIVEN(DSS, HELP))
1781                 txt = lls_short_help(CMD_PTR(DSS));
1782         else if (OPT_GIVEN(DSS, VERSION))
1783                 txt = dss_strdup(VERSION_STRING);
1784         else
1785                 return;
1786         printf("%s", txt);
1787         free(txt);
1788         exit(EXIT_SUCCESS);
1789 }
1790
1791 static void show_subcommand_summary(void)
1792 {
1793         const struct lls_command *cmd;
1794         int i;
1795
1796         printf("Available subcommands:\n");
1797         for (i = 1; (cmd = lls_cmd(i, dss_suite)); i++) {
1798                 const char *name = lls_command_name(cmd);
1799                 const char *purpose = lls_purpose(cmd);
1800                 printf("%-11s%s\n", name, purpose);
1801         }
1802         exit(EXIT_SUCCESS);
1803 }
1804
1805 int main(int argc, char **argv)
1806 {
1807         int ret;
1808         char *errctx = NULL;
1809         unsigned num_inputs;
1810         const struct dss_user_data *ud;
1811
1812         ret = lls_parse(argc, argv, CMD_PTR(DSS), &cmdline_lpr, &errctx);
1813         if (ret < 0) {
1814                 ret = lopsub_error(ret, &errctx);
1815                 goto out;
1816         }
1817         lpr = cmdline_lpr;
1818         ret = parse_config_file(false /* no SIGHUP */, CMD_PTR(DSS));
1819         if (ret < 0)
1820                 goto out;
1821         handle_version_and_help();
1822         num_inputs = lls_num_inputs(lpr);
1823         if (num_inputs == 0)
1824                 show_subcommand_summary();
1825         ret = lls_lookup_subcmd(argv[argc - num_inputs], dss_suite, &errctx);
1826         if (ret < 0) {
1827                 ret = lopsub_error(ret, &errctx);
1828                 goto out;
1829         }
1830         subcmd = lls_cmd(ret, dss_suite);
1831         ret = lls_parse(num_inputs, argv + argc - num_inputs, subcmd,
1832                 &cmdline_sublpr, &errctx);
1833         if (ret < 0) {
1834                 ret = lopsub_error(ret, &errctx);
1835                 goto out;
1836         }
1837         sublpr = cmdline_sublpr;
1838         ret = parse_config_file(false /* no SIGHUP */, subcmd);
1839         if (ret < 0)
1840                 goto out;
1841         ret = check_config();
1842         if (ret < 0)
1843                 goto out;
1844         ret = setup_signal_handling();
1845         if (ret < 0)
1846                 goto out;
1847         ud = lls_user_data(subcmd);
1848         ret = ud->handler();
1849         signal_shutdown();
1850 out:
1851         if (ret < 0) {
1852                 if (errctx)
1853                         DSS_ERROR_LOG(("%s\n", errctx));
1854                 DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
1855         }
1856         free(errctx);
1857         lls_free_parse_result(lpr, CMD_PTR(DSS));
1858         if (lpr != cmdline_lpr)
1859                 lls_free_parse_result(cmdline_lpr, CMD_PTR(DSS));
1860         lls_free_parse_result(sublpr, subcmd);
1861         if (sublpr != cmdline_sublpr)
1862                 lls_free_parse_result(cmdline_sublpr, subcmd);
1863         exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
1864 }