src/core/main.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <stdio.h>
  23 #include <errno.h>
  24 #include <string.h>
  25 #include <unistd.h>
  26 #include <sys/types.h>
  27 #include <sys/stat.h>
  28 #include <getopt.h>
  29 #include <signal.h>
  30 #include <sys/wait.h>
  31 #include <fcntl.h>
  32 #include <sys/prctl.h>
  33 #include <sys/mount.h>
  34
  35 #ifdef HAVE_VALGRIND_VALGRIND_H
  36 #include <valgrind/valgrind.h>
  37 #endif
  38
  39 #include "sd-daemon.h"
  40 #include "sd-messages.h"
  41 #include "sd-bus.h"
  42 #include "manager.h"
  43 #include "log.h"
  44 #include "load-fragment.h"
  45 #include "fdset.h"
  46 #include "special.h"
  47 #include "conf-parser.h"
  48 #include "missing.h"
  49 #include "label.h"
  50 #include "build.h"
  51 #include "strv.h"
  52 #include "def.h"
  53 #include "virt.h"
  54 #include "watchdog.h"
  55 #include "path-util.h"
  56 #include "switch-root.h"
  57 #include "capability.h"
  58 #include "killall.h"
  59 #include "env-util.h"
  60 #include "hwclock.h"
  61 #include "fileio.h"
  62 #include "dbus-manager.h"
  63 #include "bus-error.h"
  64 #include "bus-util.h"
  65
  66 #include "mount-setup.h"
  67 #include "loopback-setup.h"
  68 #include "hostname-setup.h"
  69 #include "machine-id-setup.h"
  70 #include "selinux-setup.h"
  71 #include "ima-setup.h"
  72 #include "smack-setup.h"
  73 #ifdef HAVE_KMOD
  74 #include "kmod-setup.h"
  75 #endif
  76
  77 static enum {
  78         ACTION_RUN,
  79         ACTION_HELP,
  80         ACTION_VERSION,
  81         ACTION_TEST,
  82         ACTION_DUMP_CONFIGURATION_ITEMS,
  83         ACTION_DONE
  84 } arg_action = ACTION_RUN;
  85
  86 static char *arg_default_unit = NULL;
  87 static SystemdRunningAs arg_running_as = _SYSTEMD_RUNNING_AS_INVALID;
  88
  89 static bool arg_dump_core = true;
  90 static bool arg_crash_shell = false;
  91 static int arg_crash_chvt = -1;
  92 static bool arg_confirm_spawn = false;
  93 static bool arg_show_status = true;
  94 static bool arg_switched_root = false;
  95 static char ***arg_join_controllers = NULL;
  96 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
  97 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
  98 static usec_t arg_default_restart_usec = DEFAULT_RESTART_USEC;
  99 static usec_t arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
 100 static usec_t arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
 101 static usec_t arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
 102 static unsigned arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
 103 static usec_t arg_runtime_watchdog = 0;
 104 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
 105 static char **arg_default_environment = NULL;
 106 static struct rlimit *arg_default_rlimit[RLIMIT_NLIMITS] = {};
 107 static uint64_t arg_capability_bounding_set_drop = 0;
 108 static nsec_t arg_timer_slack_nsec = (nsec_t) -1;
 109
 110 static FILE* serialization = NULL;
 111
 112 static void nop_handler(int sig) {
 113 }
 114
 115 noreturn static void crash(int sig) {
 116
 117         if (getpid() != 1)
 118                 /* Pass this on immediately, if this is not PID 1 */
 119                 raise(sig);
 120         else if (!arg_dump_core)
 121                 log_error("Caught <%s>, not dumping core.", signal_to_string(sig));
 122         else {
 123                 struct sigaction sa = {
 124                         .sa_handler = nop_handler,
 125                         .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 126                 };
 127                 pid_t pid;
 128
 129                 /* We want to wait for the core process, hence let's enable SIGCHLD */
 130                 sigaction(SIGCHLD, &sa, NULL);
 131
 132                 pid = fork();
 133                 if (pid < 0)
 134                         log_error("Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
 135
 136                 else if (pid == 0) {
 137                         struct rlimit rl = {};
 138
 139                         /* Enable default signal handler for core dump */
 140                         zero(sa);
 141                         sa.sa_handler = SIG_DFL;
 142                         sigaction(sig, &sa, NULL);
 143
 144                         /* Don't limit the core dump size */
 145                         rl.rlim_cur = RLIM_INFINITY;
 146                         rl.rlim_max = RLIM_INFINITY;
 147                         setrlimit(RLIMIT_CORE, &rl);
 148
 149                         /* Just to be sure... */
 150                         chdir("/");
 151
 152                         /* Raise the signal again */
 153                         raise(sig);
 154
 155                         assert_not_reached("We shouldn't be here...");
 156                         _exit(1);
 157
 158                 } else {
 159                         siginfo_t status;
 160                         int r;
 161
 162                         /* Order things nicely. */
 163                         r = wait_for_terminate(pid, &status);
 164                         if (r < 0)
 165                                 log_error("Caught <%s>, waitpid() failed: %s", signal_to_string(sig), strerror(-r));
 166                         else if (status.si_code != CLD_DUMPED)
 167                                 log_error("Caught <%s>, core dump failed.", signal_to_string(sig));
 168                         else
 169                                 log_error("Caught <%s>, dumped core as pid %lu.", signal_to_string(sig), (unsigned long) pid);
 170                 }
 171         }
 172
 173         if (arg_crash_chvt)
 174                 chvt(arg_crash_chvt);
 175
 176         if (arg_crash_shell) {
 177                 struct sigaction sa = {
 178                         .sa_handler = SIG_IGN,
 179                         .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
 180                 };
 181                 pid_t pid;
 182
 183                 log_info("Executing crash shell in 10s...");
 184                 sleep(10);
 185
 186                 /* Let the kernel reap children for us */
 187                 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 188
 189                 pid = fork();
 190                 if (pid < 0)
 191                         log_error("Failed to fork off crash shell: %m");
 192                 else if (pid == 0) {
 193                         make_console_stdio();
 194                         execl("/bin/sh", "/bin/sh", NULL);
 195
 196                         log_error("execl() failed: %m");
 197                         _exit(1);
 198                 }
 199
 200                 log_info("Successfully spawned crash shell as pid %lu.", (unsigned long) pid);
 201         }
 202
 203         log_info("Freezing execution.");
 204         freeze();
 205 }
 206
 207 static void install_crash_handler(void) {
 208         struct sigaction sa = {
 209                 .sa_handler = crash,
 210                 .sa_flags = SA_NODEFER,
 211         };
 212
 213         sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
 214 }
 215
 216 static int console_setup(bool do_reset) {
 217         int tty_fd, r;
 218
 219         /* If we are init, we connect stdin/stdout/stderr to /dev/null
 220          * and make sure we don't have a controlling tty. */
 221
 222         release_terminal();
 223
 224         if (!do_reset)
 225                 return 0;
 226
 227         tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
 228         if (tty_fd < 0) {
 229                 log_error("Failed to open /dev/console: %s", strerror(-tty_fd));
 230                 return -tty_fd;
 231         }
 232
 233         /* We don't want to force text mode.
 234          * plymouth may be showing pictures already from initrd. */
 235         r = reset_terminal_fd(tty_fd, false);
 236         if (r < 0)
 237                 log_error("Failed to reset /dev/console: %s", strerror(-r));
 238
 239         close_nointr_nofail(tty_fd);
 240         return r;
 241 }
 242
 243 static int set_default_unit(const char *u) {
 244         char *c;
 245
 246         assert(u);
 247
 248         c = strdup(u);
 249         if (!c)
 250                 return -ENOMEM;
 251
 252         free(arg_default_unit);
 253         arg_default_unit = c;
 254
 255         return 0;
 256 }
 257
 258 static int parse_proc_cmdline_word(const char *word) {
 259
 260         static const char * const rlmap[] = {
 261                 "emergency", SPECIAL_EMERGENCY_TARGET,
 262                 "-b",        SPECIAL_EMERGENCY_TARGET,
 263                 "single",    SPECIAL_RESCUE_TARGET,
 264                 "-s",        SPECIAL_RESCUE_TARGET,
 265                 "s",         SPECIAL_RESCUE_TARGET,
 266                 "S",         SPECIAL_RESCUE_TARGET,
 267                 "1",         SPECIAL_RESCUE_TARGET,
 268                 "2",         SPECIAL_RUNLEVEL2_TARGET,
 269                 "3",         SPECIAL_RUNLEVEL3_TARGET,
 270                 "4",         SPECIAL_RUNLEVEL4_TARGET,
 271                 "5",         SPECIAL_RUNLEVEL5_TARGET,
 272         };
 273
 274         assert(word);
 275
 276         if (startswith(word, "systemd.unit=")) {
 277
 278                 if (!in_initrd())
 279                         return set_default_unit(word + 13);
 280
 281         } else if (startswith(word, "rd.systemd.unit=")) {
 282
 283                 if (in_initrd())
 284                         return set_default_unit(word + 16);
 285
 286         } else if (startswith(word, "systemd.log_target=")) {
 287
 288                 if (log_set_target_from_string(word + 19) < 0)
 289                         log_warning("Failed to parse log target %s. Ignoring.", word + 19);
 290
 291         } else if (startswith(word, "systemd.log_level=")) {
 292
 293                 if (log_set_max_level_from_string(word + 18) < 0)
 294                         log_warning("Failed to parse log level %s. Ignoring.", word + 18);
 295
 296         } else if (startswith(word, "systemd.log_color=")) {
 297
 298                 if (log_show_color_from_string(word + 18) < 0)
 299                         log_warning("Failed to parse log color setting %s. Ignoring.", word + 18);
 300
 301         } else if (startswith(word, "systemd.log_location=")) {
 302
 303                 if (log_show_location_from_string(word + 21) < 0)
 304                         log_warning("Failed to parse log location setting %s. Ignoring.", word + 21);
 305
 306         } else if (startswith(word, "systemd.dump_core=")) {
 307                 int r;
 308
 309                 if ((r = parse_boolean(word + 18)) < 0)
 310                         log_warning("Failed to parse dump core switch %s. Ignoring.", word + 18);
 311                 else
 312                         arg_dump_core = r;
 313
 314         } else if (startswith(word, "systemd.crash_shell=")) {
 315                 int r;
 316
 317                 if ((r = parse_boolean(word + 20)) < 0)
 318                         log_warning("Failed to parse crash shell switch %s. Ignoring.", word + 20);
 319                 else
 320                         arg_crash_shell = r;
 321
 322         } else if (startswith(word, "systemd.confirm_spawn=")) {
 323                 int r;
 324
 325                 if ((r = parse_boolean(word + 22)) < 0)
 326                         log_warning("Failed to parse confirm spawn switch %s. Ignoring.", word + 22);
 327                 else
 328                         arg_confirm_spawn = r;
 329
 330         } else if (startswith(word, "systemd.crash_chvt=")) {
 331                 int k;
 332
 333                 if (safe_atoi(word + 19, &k) < 0)
 334                         log_warning("Failed to parse crash chvt switch %s. Ignoring.", word + 19);
 335                 else
 336                         arg_crash_chvt = k;
 337
 338         } else if (startswith(word, "systemd.show_status=")) {
 339                 int r;
 340
 341                 if ((r = parse_boolean(word + 20)) < 0)
 342                         log_warning("Failed to parse show status switch %s. Ignoring.", word + 20);
 343                 else
 344                         arg_show_status = r;
 345         } else if (startswith(word, "systemd.default_standard_output=")) {
 346                 int r;
 347
 348                 if ((r = exec_output_from_string(word + 32)) < 0)
 349                         log_warning("Failed to parse default standard output switch %s. Ignoring.", word + 32);
 350                 else
 351                         arg_default_std_output = r;
 352         } else if (startswith(word, "systemd.default_standard_error=")) {
 353                 int r;
 354
 355                 if ((r = exec_output_from_string(word + 31)) < 0)
 356                         log_warning("Failed to parse default standard error switch %s. Ignoring.", word + 31);
 357                 else
 358                         arg_default_std_error = r;
 359         } else if (startswith(word, "systemd.setenv=")) {
 360                 _cleanup_free_ char *cenv = NULL;
 361
 362                 cenv = strdup(word + 15);
 363                 if (!cenv)
 364                         return -ENOMEM;
 365
 366                 if (env_assignment_is_valid(cenv)) {
 367                         char **env;
 368
 369                         env = strv_env_set(arg_default_environment, cenv);
 370                         if (env)
 371                                 arg_default_environment = env;
 372                         else
 373                                 log_warning("Setting environment variable '%s' failed, ignoring: %m", cenv);
 374                 } else
 375                         log_warning("Environment variable name '%s' is not valid. Ignoring.", cenv);
 376
 377         } else if (startswith(word, "systemd.") ||
 378                    (in_initrd() && startswith(word, "rd.systemd."))) {
 379
 380                 const char *c;
 381
 382                 /* Ignore systemd.journald.xyz and friends */
 383                 c = word;
 384                 if (startswith(c, "rd."))
 385                         c += 3;
 386                 if (startswith(c, "systemd."))
 387                         c += 8;
 388                 if (c[strcspn(c, ".=")] != '.')  {
 389
 390                         log_warning("Unknown kernel switch %s. Ignoring.", word);
 391
 392                         log_info("Supported kernel switches:\n"
 393                                  "systemd.unit=UNIT                        Default unit to start\n"
 394                                  "rd.systemd.unit=UNIT                     Default unit to start when run in initrd\n"
 395                                  "systemd.dump_core=0|1                    Dump core on crash\n"
 396                                  "systemd.crash_shell=0|1                  Run shell on crash\n"
 397                                  "systemd.crash_chvt=N                     Change to VT #N on crash\n"
 398                                  "systemd.confirm_spawn=0|1                Confirm every process spawn\n"
 399                                  "systemd.show_status=0|1                  Show status updates on the console during bootup\n"
 400                                  "systemd.log_target=console|kmsg|journal|journal-or-kmsg|syslog|syslog-or-kmsg|null\n"
 401                                  "                                         Log target\n"
 402                                  "systemd.log_level=LEVEL                  Log level\n"
 403                                  "systemd.log_color=0|1                    Highlight important log messages\n"
 404                                  "systemd.log_location=0|1                 Include code location in log messages\n"
 405                                  "systemd.default_standard_output=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
 406                                  "                                         Set default log output for services\n"
 407                                  "systemd.default_standard_error=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
 408                                  "                                         Set default log error output for services\n"
 409                                  "systemd.setenv=ASSIGNMENT                Set an environment variable for all spawned processes\n");
 410                 }
 411
 412         } else if (streq(word, "quiet"))
 413                 arg_show_status = false;
 414         else if (streq(word, "debug")) {
 415                 /* Log to kmsg, the journal socket will fill up before the
 416                  * journal is started and tools running during that time
 417                  * will block with every log message for for 60 seconds,
 418                  * before they give up. */
 419                 log_set_max_level(LOG_DEBUG);
 420                 log_set_target(detect_container(NULL) > 0 ? LOG_TARGET_CONSOLE : LOG_TARGET_KMSG);
 421         } else if (!in_initrd()) {
 422                 unsigned i;
 423
 424                 /* SysV compatibility */
 425                 for (i = 0; i < ELEMENTSOF(rlmap); i += 2)
 426                         if (streq(word, rlmap[i]))
 427                                 return set_default_unit(rlmap[i+1]);
 428         }
 429
 430         return 0;
 431 }
 432
 433 #define DEFINE_SETTER(name, func, descr)                              \
 434         static int name(const char *unit,                             \
 435                         const char *filename,                         \
 436                         unsigned line,                                \
 437                         const char *section,                          \
 438                         unsigned section_line,                        \
 439                         const char *lvalue,                           \
 440                         int ltype,                                    \
 441                         const char *rvalue,                           \
 442                         void *data,                                   \
 443                         void *userdata) {                             \
 444                                                                       \
 445                 int r;                                                \
 446                                                                       \
 447                 assert(filename);                                     \
 448                 assert(lvalue);                                       \
 449                 assert(rvalue);                                       \
 450                                                                       \
 451                 r = func(rvalue);                                     \
 452                 if (r < 0)                                            \
 453                         log_syntax(unit, LOG_ERR, filename, line, -r, \
 454                                    "Invalid " descr "'%s': %s",       \
 455                                    rvalue, strerror(-r));             \
 456                                                                       \
 457                 return 0;                                             \
 458         }
 459
 460 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level")
 461 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target")
 462 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" )
 463 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location")
 464
 465
 466 static int config_parse_cpu_affinity2(const char *unit,
 467                                       const char *filename,
 468                                       unsigned line,
 469                                       const char *section,
 470                                       unsigned section_line,
 471                                       const char *lvalue,
 472                                       int ltype,
 473                                       const char *rvalue,
 474                                       void *data,
 475                                       void *userdata) {
 476
 477         char *w;
 478         size_t l;
 479         char *state;
 480         cpu_set_t *c = NULL;
 481         unsigned ncpus = 0;
 482
 483         assert(filename);
 484         assert(lvalue);
 485         assert(rvalue);
 486
 487         FOREACH_WORD_QUOTED(w, l, rvalue, state) {
 488                 char *t;
 489                 int r;
 490                 unsigned cpu;
 491
 492                 if (!(t = strndup(w, l)))
 493                         return log_oom();
 494
 495                 r = safe_atou(t, &cpu);
 496                 free(t);
 497
 498                 if (!c)
 499                         if (!(c = cpu_set_malloc(&ncpus)))
 500                                 return log_oom();
 501
 502                 if (r < 0 || cpu >= ncpus) {
 503                         log_syntax(unit, LOG_ERR, filename, line, -r,
 504                                    "Failed to parse CPU affinity '%s'", rvalue);
 505                         CPU_FREE(c);
 506                         return -EBADMSG;
 507                 }
 508
 509                 CPU_SET_S(cpu, CPU_ALLOC_SIZE(ncpus), c);
 510         }
 511
 512         if (c) {
 513                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
 514                         log_warning_unit(unit, "Failed to set CPU affinity: %m");
 515
 516                 CPU_FREE(c);
 517         }
 518
 519         return 0;
 520 }
 521
 522 static void strv_free_free(char ***l) {
 523         char ***i;
 524
 525         if (!l)
 526                 return;
 527
 528         for (i = l; *i; i++)
 529                 strv_free(*i);
 530
 531         free(l);
 532 }
 533
 534 static void free_join_controllers(void) {
 535         strv_free_free(arg_join_controllers);
 536         arg_join_controllers = NULL;
 537 }
 538
 539 static int config_parse_join_controllers(const char *unit,
 540                                          const char *filename,
 541                                          unsigned line,
 542                                          const char *section,
 543                                          unsigned section_line,
 544                                          const char *lvalue,
 545                                          int ltype,
 546                                          const char *rvalue,
 547                                          void *data,
 548                                          void *userdata) {
 549
 550         unsigned n = 0;
 551         char *state, *w;
 552         size_t length;
 553
 554         assert(filename);
 555         assert(lvalue);
 556         assert(rvalue);
 557
 558         free_join_controllers();
 559
 560         FOREACH_WORD_QUOTED(w, length, rvalue, state) {
 561                 char *s, **l;
 562
 563                 s = strndup(w, length);
 564                 if (!s)
 565                         return log_oom();
 566
 567                 l = strv_split(s, ",");
 568                 free(s);
 569
 570                 strv_uniq(l);
 571
 572                 if (strv_length(l) <= 1) {
 573                         strv_free(l);
 574                         continue;
 575                 }
 576
 577                 if (!arg_join_controllers) {
 578                         arg_join_controllers = new(char**, 2);
 579                         if (!arg_join_controllers) {
 580                                 strv_free(l);
 581                                 return log_oom();
 582                         }
 583
 584                         arg_join_controllers[0] = l;
 585                         arg_join_controllers[1] = NULL;
 586
 587                         n = 1;
 588                 } else {
 589                         char ***a;
 590                         char ***t;
 591
 592                         t = new0(char**, n+2);
 593                         if (!t) {
 594                                 strv_free(l);
 595                                 return log_oom();
 596                         }
 597
 598                         n = 0;
 599
 600                         for (a = arg_join_controllers; *a; a++) {
 601
 602                                 if (strv_overlap(*a, l)) {
 603                                         char **c;
 604
 605                                         c = strv_merge(*a, l);
 606                                         if (!c) {
 607                                                 strv_free(l);
 608                                                 strv_free_free(t);
 609                                                 return log_oom();
 610                                         }
 611
 612                                         strv_free(l);
 613                                         l = c;
 614                                 } else {
 615                                         char **c;
 616
 617                                         c = strv_copy(*a);
 618                                         if (!c) {
 619                                                 strv_free(l);
 620                                                 strv_free_free(t);
 621                                                 return log_oom();
 622                                         }
 623
 624                                         t[n++] = c;
 625                                 }
 626                         }
 627
 628                         t[n++] = strv_uniq(l);
 629
 630                         strv_free_free(arg_join_controllers);
 631                         arg_join_controllers = t;
 632                 }
 633         }
 634
 635         return 0;
 636 }
 637
 638 static int parse_config_file(void) {
 639
 640         const ConfigTableItem items[] = {
 641                 { "Manager", "LogLevel",              config_parse_level2,       0, NULL                     },
 642                 { "Manager", "LogTarget",             config_parse_target,       0, NULL                     },
 643                 { "Manager", "LogColor",              config_parse_color,        0, NULL                     },
 644                 { "Manager", "LogLocation",           config_parse_location,     0, NULL                     },
 645                 { "Manager", "DumpCore",              config_parse_bool,         0, &arg_dump_core           },
 646                 { "Manager", "CrashShell",            config_parse_bool,         0, &arg_crash_shell         },
 647                 { "Manager", "ShowStatus",            config_parse_bool,         0, &arg_show_status         },
 648                 { "Manager", "CrashChVT",             config_parse_int,          0, &arg_crash_chvt          },
 649                 { "Manager", "CPUAffinity",           config_parse_cpu_affinity2, 0, NULL                    },
 650                 { "Manager", "DefaultStandardOutput", config_parse_output,       0, &arg_default_std_output  },
 651                 { "Manager", "DefaultStandardError",  config_parse_output,       0, &arg_default_std_error   },
 652                 { "Manager", "DefaultTimeoutStartSec", config_parse_sec,         0, &arg_default_timeout_start_usec },
 653                 { "Manager", "DefaultTimeoutStopSec", config_parse_sec,          0, &arg_default_timeout_stop_usec  },
 654                 { "Manager", "DefaultRestartSec",     config_parse_sec,          0, &arg_default_restart_usec  },
 655                 { "Manager", "DefaultStartLimitInterval", config_parse_sec,      0, &arg_default_start_limit_interval },
 656                 { "Manager", "DefaultStartLimitBurst", config_parse_unsigned,    0, &arg_default_start_limit_burst },
 657                 { "Manager", "JoinControllers",       config_parse_join_controllers, 0, &arg_join_controllers },
 658                 { "Manager", "RuntimeWatchdogSec",    config_parse_sec,          0, &arg_runtime_watchdog    },
 659                 { "Manager", "ShutdownWatchdogSec",   config_parse_sec,          0, &arg_shutdown_watchdog   },
 660                 { "Manager", "CapabilityBoundingSet", config_parse_bounding_set, 0, &arg_capability_bounding_set_drop },
 661                 { "Manager", "TimerSlackNSec",        config_parse_nsec,         0, &arg_timer_slack_nsec    },
 662                 { "Manager", "DefaultEnvironment",    config_parse_environ,      0, &arg_default_environment },
 663                 { "Manager", "DefaultLimitCPU",       config_parse_limit,        0, &arg_default_rlimit[RLIMIT_CPU]},
 664                 { "Manager", "DefaultLimitFSIZE",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_FSIZE]},
 665                 { "Manager", "DefaultLimitDATA",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_DATA]},
 666                 { "Manager", "DefaultLimitSTACK",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_STACK]},
 667                 { "Manager", "DefaultLimitCORE",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_CORE]},
 668                 { "Manager", "DefaultLimitRSS",       config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RSS]},
 669                 { "Manager", "DefaultLimitNOFILE",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NOFILE]},
 670                 { "Manager", "DefaultLimitAS",        config_parse_limit,        0, &arg_default_rlimit[RLIMIT_AS]},
 671                 { "Manager", "DefaultLimitNPROC",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NPROC]},
 672                 { "Manager", "DefaultLimitMEMLOCK",   config_parse_limit,        0, &arg_default_rlimit[RLIMIT_MEMLOCK]},
 673                 { "Manager", "DefaultLimitLOCKS",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_LOCKS]},
 674                 { "Manager", "DefaultLimitSIGPENDING",config_parse_limit,        0, &arg_default_rlimit[RLIMIT_SIGPENDING]},
 675                 { "Manager", "DefaultLimitMSGQUEUE",  config_parse_limit,        0, &arg_default_rlimit[RLIMIT_MSGQUEUE]},
 676                 { "Manager", "DefaultLimitNICE",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NICE]},
 677                 { "Manager", "DefaultLimitRTPRIO",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RTPRIO]},
 678                 { "Manager", "DefaultLimitRTTIME",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RTTIME]},
 679                 { NULL, NULL, NULL, 0, NULL }
 680         };
 681
 682         _cleanup_fclose_ FILE *f;
 683         const char *fn;
 684         int r;
 685
 686         fn = arg_running_as == SYSTEMD_SYSTEM ? PKGSYSCONFDIR "/system.conf" : PKGSYSCONFDIR "/user.conf";
 687         f = fopen(fn, "re");
 688         if (!f) {
 689                 if (errno == ENOENT)
 690                         return 0;
 691
 692                 log_warning("Failed to open configuration file '%s': %m", fn);
 693                 return 0;
 694         }
 695
 696         r = config_parse(NULL, fn, f, "Manager\0", config_item_table_lookup, (void*) items, false, false, NULL);
 697         if (r < 0)
 698                 log_warning("Failed to parse configuration file: %s", strerror(-r));
 699
 700         return 0;
 701 }
 702
 703 static int parse_proc_cmdline(void) {
 704         _cleanup_free_ char *line = NULL;
 705         char *w, *state;
 706         size_t l;
 707         int r;
 708
 709         r = proc_cmdline(&line);
 710         if (r < 0)
 711                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
 712         if (r <= 0)
 713                 return 0;
 714
 715         FOREACH_WORD_QUOTED(w, l, line, state) {
 716                 _cleanup_free_ char *word;
 717
 718                 word = strndup(w, l);
 719                 if (!word)
 720                         return log_oom();
 721
 722                 r = parse_proc_cmdline_word(word);
 723                 if (r < 0) {
 724                         log_error("Failed on cmdline argument %s: %s", word, strerror(-r));
 725                         return r;
 726                 }
 727         }
 728
 729         return 0;
 730 }
 731
 732 static int parse_argv(int argc, char *argv[]) {
 733
 734         enum {
 735                 ARG_LOG_LEVEL = 0x100,
 736                 ARG_LOG_TARGET,
 737                 ARG_LOG_COLOR,
 738                 ARG_LOG_LOCATION,
 739                 ARG_UNIT,
 740                 ARG_SYSTEM,
 741                 ARG_USER,
 742                 ARG_TEST,
 743                 ARG_VERSION,
 744                 ARG_DUMP_CONFIGURATION_ITEMS,
 745                 ARG_DUMP_CORE,
 746                 ARG_CRASH_SHELL,
 747                 ARG_CONFIRM_SPAWN,
 748                 ARG_SHOW_STATUS,
 749                 ARG_DESERIALIZE,
 750                 ARG_SWITCHED_ROOT,
 751                 ARG_DEFAULT_STD_OUTPUT,
 752                 ARG_DEFAULT_STD_ERROR
 753         };
 754
 755         static const struct option options[] = {
 756                 { "log-level",                required_argument, NULL, ARG_LOG_LEVEL                },
 757                 { "log-target",               required_argument, NULL, ARG_LOG_TARGET               },
 758                 { "log-color",                optional_argument, NULL, ARG_LOG_COLOR                },
 759                 { "log-location",             optional_argument, NULL, ARG_LOG_LOCATION             },
 760                 { "unit",                     required_argument, NULL, ARG_UNIT                     },
 761                 { "system",                   no_argument,       NULL, ARG_SYSTEM                   },
 762                 { "user",                     no_argument,       NULL, ARG_USER                     },
 763                 { "test",                     no_argument,       NULL, ARG_TEST                     },
 764                 { "help",                     no_argument,       NULL, 'h'                          },
 765                 { "version",                  no_argument,       NULL, ARG_VERSION                  },
 766                 { "dump-configuration-items", no_argument,       NULL, ARG_DUMP_CONFIGURATION_ITEMS },
 767                 { "dump-core",                optional_argument, NULL, ARG_DUMP_CORE                },
 768                 { "crash-shell",              optional_argument, NULL, ARG_CRASH_SHELL              },
 769                 { "confirm-spawn",            optional_argument, NULL, ARG_CONFIRM_SPAWN            },
 770                 { "show-status",              optional_argument, NULL, ARG_SHOW_STATUS              },
 771                 { "deserialize",              required_argument, NULL, ARG_DESERIALIZE              },
 772                 { "switched-root",            no_argument,       NULL, ARG_SWITCHED_ROOT            },
 773                 { "default-standard-output",  required_argument, NULL, ARG_DEFAULT_STD_OUTPUT,      },
 774                 { "default-standard-error",   required_argument, NULL, ARG_DEFAULT_STD_ERROR,       },
 775                 { NULL,                       0,                 NULL, 0                            }
 776         };
 777
 778         int c, r;
 779
 780         assert(argc >= 1);
 781         assert(argv);
 782
 783         if (getpid() == 1)
 784                 opterr = 0;
 785
 786         while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
 787
 788                 switch (c) {
 789
 790                 case ARG_LOG_LEVEL:
 791                         if ((r = log_set_max_level_from_string(optarg)) < 0) {
 792                                 log_error("Failed to parse log level %s.", optarg);
 793                                 return r;
 794                         }
 795
 796                         break;
 797
 798                 case ARG_LOG_TARGET:
 799
 800                         if ((r = log_set_target_from_string(optarg)) < 0) {
 801                                 log_error("Failed to parse log target %s.", optarg);
 802                                 return r;
 803                         }
 804
 805                         break;
 806
 807                 case ARG_LOG_COLOR:
 808
 809                         if (optarg) {
 810                                 if ((r = log_show_color_from_string(optarg)) < 0) {
 811                                         log_error("Failed to parse log color setting %s.", optarg);
 812                                         return r;
 813                                 }
 814                         } else
 815                                 log_show_color(true);
 816
 817                         break;
 818
 819                 case ARG_LOG_LOCATION:
 820
 821                         if (optarg) {
 822                                 if ((r = log_show_location_from_string(optarg)) < 0) {
 823                                         log_error("Failed to parse log location setting %s.", optarg);
 824                                         return r;
 825                                 }
 826                         } else
 827                                 log_show_location(true);
 828
 829                         break;
 830
 831                 case ARG_DEFAULT_STD_OUTPUT:
 832
 833                         if ((r = exec_output_from_string(optarg)) < 0) {
 834                                 log_error("Failed to parse default standard output setting %s.", optarg);
 835                                 return r;
 836                         } else
 837                                 arg_default_std_output = r;
 838                         break;
 839
 840                 case ARG_DEFAULT_STD_ERROR:
 841
 842                         if ((r = exec_output_from_string(optarg)) < 0) {
 843                                 log_error("Failed to parse default standard error output setting %s.", optarg);
 844                                 return r;
 845                         } else
 846                                 arg_default_std_error = r;
 847                         break;
 848
 849                 case ARG_UNIT:
 850
 851                         if ((r = set_default_unit(optarg)) < 0) {
 852                                 log_error("Failed to set default unit %s: %s", optarg, strerror(-r));
 853                                 return r;
 854                         }
 855
 856                         break;
 857
 858                 case ARG_SYSTEM:
 859                         arg_running_as = SYSTEMD_SYSTEM;
 860                         break;
 861
 862                 case ARG_USER:
 863                         arg_running_as = SYSTEMD_USER;
 864                         break;
 865
 866                 case ARG_TEST:
 867                         arg_action = ACTION_TEST;
 868                         break;
 869
 870                 case ARG_VERSION:
 871                         arg_action = ACTION_VERSION;
 872                         break;
 873
 874                 case ARG_DUMP_CONFIGURATION_ITEMS:
 875                         arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
 876                         break;
 877
 878                 case ARG_DUMP_CORE:
 879                         r = optarg ? parse_boolean(optarg) : 1;
 880                         if (r < 0) {
 881                                 log_error("Failed to parse dump core boolean %s.", optarg);
 882                                 return r;
 883                         }
 884                         arg_dump_core = r;
 885                         break;
 886
 887                 case ARG_CRASH_SHELL:
 888                         r = optarg ? parse_boolean(optarg) : 1;
 889                         if (r < 0) {
 890                                 log_error("Failed to parse crash shell boolean %s.", optarg);
 891                                 return r;
 892                         }
 893                         arg_crash_shell = r;
 894                         break;
 895
 896                 case ARG_CONFIRM_SPAWN:
 897                         r = optarg ? parse_boolean(optarg) : 1;
 898                         if (r < 0) {
 899                                 log_error("Failed to parse confirm spawn boolean %s.", optarg);
 900                                 return r;
 901                         }
 902                         arg_confirm_spawn = r;
 903                         break;
 904
 905                 case ARG_SHOW_STATUS:
 906                         r = optarg ? parse_boolean(optarg) : 1;
 907                         if (r < 0) {
 908                                 log_error("Failed to parse show status boolean %s.", optarg);
 909                                 return r;
 910                         }
 911                         arg_show_status = r;
 912                         break;
 913
 914                 case ARG_DESERIALIZE: {
 915                         int fd;
 916                         FILE *f;
 917
 918                         r = safe_atoi(optarg, &fd);
 919                         if (r < 0 || fd < 0) {
 920                                 log_error("Failed to parse deserialize option %s.", optarg);
 921                                 return r < 0 ? r : -EINVAL;
 922                         }
 923
 924                         fd_cloexec(fd, true);
 925
 926                         f = fdopen(fd, "r");
 927                         if (!f) {
 928                                 log_error("Failed to open serialization fd: %m");
 929                                 return -errno;
 930                         }
 931
 932                         if (serialization)
 933                                 fclose(serialization);
 934
 935                         serialization = f;
 936
 937                         break;
 938                 }
 939
 940                 case ARG_SWITCHED_ROOT:
 941                         arg_switched_root = true;
 942                         break;
 943
 944                 case 'h':
 945                         arg_action = ACTION_HELP;
 946                         break;
 947
 948                 case 'D':
 949                         log_set_max_level(LOG_DEBUG);
 950                         break;
 951
 952                 case 'b':
 953                 case 's':
 954                 case 'z':
 955                         /* Just to eat away the sysvinit kernel
 956                          * cmdline args without getopt() error
 957                          * messages that we'll parse in
 958                          * parse_proc_cmdline_word() or ignore. */
 959
 960                 case '?':
 961                 default:
 962                         if (getpid() != 1) {
 963                                 log_error("Unknown option code %c", c);
 964                                 return -EINVAL;
 965                         }
 966
 967                         break;
 968                 }
 969
 970         if (optind < argc && getpid() != 1) {
 971                 /* Hmm, when we aren't run as init system
 972                  * let's complain about excess arguments */
 973
 974                 log_error("Excess arguments.");
 975                 return -EINVAL;
 976         }
 977
 978         if (detect_container(NULL) > 0) {
 979                 char **a;
 980
 981                 /* All /proc/cmdline arguments the kernel didn't
 982                  * understand it passed to us. We're not really
 983                  * interested in that usually since /proc/cmdline is
 984                  * more interesting and complete. With one exception:
 985                  * if we are run in a container /proc/cmdline is not
 986                  * relevant for the container, hence we rely on argv[]
 987                  * instead. */
 988
 989                 for (a = argv; a < argv + argc; a++) {
 990                         r = parse_proc_cmdline_word(*a);
 991                         if (r < 0) {
 992                                 log_error("Failed on cmdline argument %s: %s", *a, strerror(-r));
 993                                 return r;
 994                         }
 995                 }
 996         }
 997
 998         return 0;
 999 }
1000
1001 static int help(void) {
1002
1003         printf("%s [OPTIONS...]\n\n"
1004                "Starts up and maintains the system or user services.\n\n"
1005                "  -h --help                      Show this help\n"
1006                "     --test                      Determine startup sequence, dump it and exit\n"
1007                "     --dump-configuration-items  Dump understood unit configuration items\n"
1008                "     --unit=UNIT                 Set default unit\n"
1009                "     --system                    Run a system instance, even if PID != 1\n"
1010                "     --user                      Run a user instance\n"
1011                "     --dump-core[=0|1]           Dump core on crash\n"
1012                "     --crash-shell[=0|1]         Run shell on crash\n"
1013                "     --confirm-spawn[=0|1]       Ask for confirmation when spawning processes\n"
1014                "     --show-status[=0|1]         Show status updates on the console during bootup\n"
1015                "     --log-target=TARGET         Set log target (console, journal, syslog, kmsg, journal-or-kmsg, syslog-or-kmsg, null)\n"
1016                "     --log-level=LEVEL           Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1017                "     --log-color[=0|1]           Highlight important log messages\n"
1018                "     --log-location[=0|1]        Include code location in log messages\n"
1019                "     --default-standard-output=  Set default standard output for services\n"
1020                "     --default-standard-error=   Set default standard error output for services\n",
1021                program_invocation_short_name);
1022
1023         return 0;
1024 }
1025
1026 static int version(void) {
1027         puts(PACKAGE_STRING);
1028         puts(SYSTEMD_FEATURES);
1029
1030         return 0;
1031 }
1032
1033 static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool switching_root) {
1034         FILE *f = NULL;
1035         FDSet *fds = NULL;
1036         int r;
1037
1038         assert(m);
1039         assert(_f);
1040         assert(_fds);
1041
1042         r = manager_open_serialization(m, &f);
1043         if (r < 0) {
1044                 log_error("Failed to create serialization file: %s", strerror(-r));
1045                 goto fail;
1046         }
1047
1048         /* Make sure nothing is really destructed when we shut down */
1049         m->n_reloading ++;
1050         bus_manager_send_reloading(m, true);
1051
1052         fds = fdset_new();
1053         if (!fds) {
1054                 r = -ENOMEM;
1055                 log_error("Failed to allocate fd set: %s", strerror(-r));
1056                 goto fail;
1057         }
1058
1059         r = manager_serialize(m, f, fds, switching_root);
1060         if (r < 0) {
1061                 log_error("Failed to serialize state: %s", strerror(-r));
1062                 goto fail;
1063         }
1064
1065         if (fseeko(f, 0, SEEK_SET) < 0) {
1066                 log_error("Failed to rewind serialization fd: %m");
1067                 goto fail;
1068         }
1069
1070         r = fd_cloexec(fileno(f), false);
1071         if (r < 0) {
1072                 log_error("Failed to disable O_CLOEXEC for serialization: %s", strerror(-r));
1073                 goto fail;
1074         }
1075
1076         r = fdset_cloexec(fds, false);
1077         if (r < 0) {
1078                 log_error("Failed to disable O_CLOEXEC for serialization fds: %s", strerror(-r));
1079                 goto fail;
1080         }
1081
1082         *_f = f;
1083         *_fds = fds;
1084
1085         return 0;
1086
1087 fail:
1088         fdset_free(fds);
1089
1090         if (f)
1091                 fclose(f);
1092
1093         return r;
1094 }
1095
1096 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1097         struct rlimit nl;
1098         int r;
1099
1100         assert(saved_rlimit);
1101
1102         /* Save the original RLIMIT_NOFILE so that we can reset it
1103          * later when transitioning from the initrd to the main
1104          * systemd or suchlike. */
1105         if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0) {
1106                 log_error("Reading RLIMIT_NOFILE failed: %m");
1107                 return -errno;
1108         }
1109
1110         /* Make sure forked processes get the default kernel setting */
1111         if (!arg_default_rlimit[RLIMIT_NOFILE]) {
1112                 struct rlimit *rl;
1113
1114                 rl = newdup(struct rlimit, saved_rlimit, 1);
1115                 if (!rl)
1116                         return log_oom();
1117
1118                 arg_default_rlimit[RLIMIT_NOFILE] = rl;
1119         }
1120
1121         /* Bump up the resource limit for ourselves substantially */
1122         nl.rlim_cur = nl.rlim_max = 64*1024;
1123         r = setrlimit_closest(RLIMIT_NOFILE, &nl);
1124         if (r < 0) {
1125                 log_error("Setting RLIMIT_NOFILE failed: %s", strerror(-r));
1126                 return r;
1127         }
1128
1129         return 0;
1130 }
1131
1132 static void test_mtab(void) {
1133         char *p;
1134
1135         /* Check that /etc/mtab is a symlink */
1136
1137         if (readlink_malloc("/etc/mtab", &p) >= 0) {
1138                 bool b;
1139
1140                 b = streq(p, "/proc/self/mounts") || streq(p, "/proc/mounts");
1141                 free(p);
1142
1143                 if (b)
1144                         return;
1145         }
1146
1147         log_warning("/etc/mtab is not a symlink or not pointing to /proc/self/mounts. "
1148                     "This is not supported anymore. "
1149                     "Please make sure to replace this file by a symlink to avoid incorrect or misleading mount(8) output.");
1150 }
1151
1152 static void test_usr(void) {
1153
1154         /* Check that /usr is not a separate fs */
1155
1156         if (dir_is_empty("/usr") <= 0)
1157                 return;
1158
1159         log_warning("/usr appears to be on its own filesytem and is not already mounted. This is not a supported setup. "
1160                     "Some things will probably break (sometimes even silently) in mysterious ways. "
1161                     "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1162 }
1163
1164 static void test_cgroups(void) {
1165
1166         if (access("/proc/cgroups", F_OK) >= 0)
1167                 return;
1168
1169         log_warning("CONFIG_CGROUPS was not set when your kernel was compiled. "
1170                     "Systems without control groups are not supported. "
1171                     "We will now sleep for 10s, and then continue boot-up. "
1172                     "Expect breakage and please do not file bugs. "
1173                     "Instead fix your kernel and enable CONFIG_CGROUPS. "
1174                     "Consult http://0pointer.de/blog/projects/cgroups-vs-cgroups.html for more information.");
1175
1176         sleep(10);
1177 }
1178
1179 static int initialize_join_controllers(void) {
1180         /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
1181          * + "net_prio". We'd like to add "cpuset" to the mix, but
1182          * "cpuset" does't really work for groups with no initialized
1183          * attributes. */
1184
1185         arg_join_controllers = new(char**, 3);
1186         if (!arg_join_controllers)
1187                 return -ENOMEM;
1188
1189         arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
1190         arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
1191         arg_join_controllers[2] = NULL;
1192
1193         if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
1194                 free_join_controllers();
1195                 return -ENOMEM;
1196         }
1197
1198         return 0;
1199 }
1200
1201 int main(int argc, char *argv[]) {
1202         Manager *m = NULL;
1203         int r, retval = EXIT_FAILURE;
1204         usec_t before_startup, after_startup;
1205         char timespan[FORMAT_TIMESPAN_MAX];
1206         FDSet *fds = NULL;
1207         bool reexecute = false;
1208         const char *shutdown_verb = NULL;
1209         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
1210         dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
1211         dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
1212         dual_timestamp security_start_timestamp = { 0ULL, 0ULL };
1213         dual_timestamp security_finish_timestamp = { 0ULL, 0ULL };
1214         static char systemd[] = "systemd";
1215         bool skip_setup = false;
1216         int j;
1217         bool loaded_policy = false;
1218         bool arm_reboot_watchdog = false;
1219         bool queue_default_job = false;
1220         char *switch_root_dir = NULL, *switch_root_init = NULL;
1221         static struct rlimit saved_rlimit_nofile = { 0, 0 };
1222
1223 #ifdef HAVE_SYSV_COMPAT
1224         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
1225                 /* This is compatibility support for SysV, where
1226                  * calling init as a user is identical to telinit. */
1227
1228                 errno = -ENOENT;
1229                 execv(SYSTEMCTL_BINARY_PATH, argv);
1230                 log_error("Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1231                 return 1;
1232         }
1233 #endif
1234
1235         dual_timestamp_from_monotonic(&kernel_timestamp, 0);
1236         dual_timestamp_get(&userspace_timestamp);
1237
1238         /* Determine if this is a reexecution or normal bootup. We do
1239          * the full command line parsing much later, so let's just
1240          * have a quick peek here. */
1241         if (strv_find(argv+1, "--deserialize"))
1242                 skip_setup = true;
1243
1244         /* If we have switched root, do all the special setup
1245          * things */
1246         if (strv_find(argv+1, "--switched-root"))
1247                 skip_setup = false;
1248
1249         /* If we get started via the /sbin/init symlink then we are
1250            called 'init'. After a subsequent reexecution we are then
1251            called 'systemd'. That is confusing, hence let's call us
1252            systemd right-away. */
1253         program_invocation_short_name = systemd;
1254         prctl(PR_SET_NAME, systemd);
1255
1256         saved_argv = argv;
1257         saved_argc = argc;
1258
1259         log_show_color(isatty(STDERR_FILENO) > 0);
1260
1261         /* Disable the umask logic */
1262         if (getpid() == 1)
1263                 umask(0);
1264
1265         if (getpid() == 1 && detect_container(NULL) <= 0) {
1266
1267                 /* Running outside of a container as PID 1 */
1268                 arg_running_as = SYSTEMD_SYSTEM;
1269                 make_null_stdio();
1270                 log_set_target(LOG_TARGET_KMSG);
1271                 log_open();
1272
1273                 if (in_initrd())
1274                         initrd_timestamp = userspace_timestamp;
1275
1276                 if (!skip_setup) {
1277                         mount_setup_early();
1278                         dual_timestamp_get(&security_start_timestamp);
1279                         if (selinux_setup(&loaded_policy) < 0)
1280                                 goto finish;
1281                         if (ima_setup() < 0)
1282                                 goto finish;
1283                         if (smack_setup() < 0)
1284                                 goto finish;
1285                         dual_timestamp_get(&security_finish_timestamp);
1286                 }
1287
1288                 if (label_init(NULL) < 0)
1289                         goto finish;
1290
1291                 if (!skip_setup) {
1292                         if (hwclock_is_localtime() > 0) {
1293                                 int min;
1294
1295                                 /* The first-time call to settimeofday() does a time warp in the kernel */
1296                                 r = hwclock_set_timezone(&min);
1297                                 if (r < 0)
1298                                         log_error("Failed to apply local time delta, ignoring: %s", strerror(-r));
1299                                 else
1300                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1301                         } else if (!in_initrd()) {
1302                                 /*
1303                                  * Do dummy first-time call to seal the kernel's time warp magic
1304                                  *
1305                                  * Do not call this this from inside the initrd. The initrd might not
1306                                  * carry /etc/adjtime with LOCAL, but the real system could be set up
1307                                  * that way. In such case, we need to delay the time-warp or the sealing
1308                                  * until we reach the real system.
1309                                  */
1310                                 hwclock_reset_timezone();
1311
1312                                 /* Tell the kernel our timezone */
1313                                 r = hwclock_set_timezone(NULL);
1314                                 if (r < 0)
1315                                         log_error("Failed to set the kernel's timezone, ignoring: %s", strerror(-r));
1316                         }
1317                 }
1318
1319                 /* Set the default for later on, but don't actually
1320                  * open the logs like this for now. Note that if we
1321                  * are transitioning from the initrd there might still
1322                  * be journal fd open, and we shouldn't attempt
1323                  * opening that before we parsed /proc/cmdline which
1324                  * might redirect output elsewhere. */
1325                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1326
1327         } else if (getpid() == 1) {
1328                 /* Running inside a container, as PID 1 */
1329                 arg_running_as = SYSTEMD_SYSTEM;
1330                 log_set_target(LOG_TARGET_CONSOLE);
1331                 log_close_console(); /* force reopen of /dev/console */
1332                 log_open();
1333
1334                 /* For the later on, see above... */
1335                 log_set_target(LOG_TARGET_JOURNAL);
1336
1337                 /* clear the kernel timestamp,
1338                  * because we are in a container */
1339                 kernel_timestamp.monotonic = 0ULL;
1340                 kernel_timestamp.realtime = 0ULL;
1341
1342         } else {
1343                 /* Running as user instance */
1344                 arg_running_as = SYSTEMD_USER;
1345                 log_set_target(LOG_TARGET_AUTO);
1346                 log_open();
1347
1348                 /* clear the kernel timestamp,
1349                  * because we are not PID 1 */
1350                 kernel_timestamp.monotonic = 0ULL;
1351                 kernel_timestamp.realtime = 0ULL;
1352         }
1353
1354         /* Initialize default unit */
1355         r = set_default_unit(SPECIAL_DEFAULT_TARGET);
1356         if (r < 0) {
1357                 log_error("Failed to set default unit %s: %s", SPECIAL_DEFAULT_TARGET, strerror(-r));
1358                 goto finish;
1359         }
1360
1361         r = initialize_join_controllers();
1362         if (r < 0)
1363                 goto finish;
1364
1365         /* Mount /proc, /sys and friends, so that /proc/cmdline and
1366          * /proc/$PID/fd is available. */
1367         if (getpid() == 1) {
1368                 r = mount_setup(loaded_policy);
1369                 if (r < 0)
1370                         goto finish;
1371         }
1372
1373         /* Reset all signal handlers. */
1374         assert_se(reset_all_signal_handlers() == 0);
1375
1376         ignore_signals(SIGNALS_IGNORE, -1);
1377
1378         if (parse_config_file() < 0)
1379                 goto finish;
1380
1381         if (arg_running_as == SYSTEMD_SYSTEM)
1382                 if (parse_proc_cmdline() < 0)
1383                         goto finish;
1384
1385         log_parse_environment();
1386
1387         if (parse_argv(argc, argv) < 0)
1388                 goto finish;
1389
1390         if (arg_action == ACTION_TEST &&
1391             geteuid() == 0) {
1392                 log_error("Don't run test mode as root.");
1393                 goto finish;
1394         }
1395
1396         if (arg_running_as == SYSTEMD_USER &&
1397             arg_action == ACTION_RUN &&
1398             sd_booted() <= 0) {
1399                 log_error("Trying to run as user instance, but the system has not been booted with systemd.");
1400                 goto finish;
1401         }
1402
1403         if (arg_running_as == SYSTEMD_SYSTEM &&
1404             arg_action == ACTION_RUN &&
1405             running_in_chroot() > 0) {
1406                 log_error("Cannot be run in a chroot() environment.");
1407                 goto finish;
1408         }
1409
1410         if (arg_action == ACTION_HELP) {
1411                 retval = help();
1412                 goto finish;
1413         } else if (arg_action == ACTION_VERSION) {
1414                 retval = version();
1415                 goto finish;
1416         } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
1417                 unit_dump_config_items(stdout);
1418                 retval = EXIT_SUCCESS;
1419                 goto finish;
1420         } else if (arg_action == ACTION_DONE) {
1421                 retval = EXIT_SUCCESS;
1422                 goto finish;
1423         }
1424
1425         if (arg_running_as == SYSTEMD_USER &&
1426             !getenv("XDG_RUNTIME_DIR")) {
1427                 log_error("Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
1428                 goto finish;
1429         }
1430
1431         assert_se(arg_action == ACTION_RUN || arg_action == ACTION_TEST);
1432
1433         /* Close logging fds, in order not to confuse fdset below */
1434         log_close();
1435
1436         /* Remember open file descriptors for later deserialization */
1437         r = fdset_new_fill(&fds);
1438         if (r < 0) {
1439                 log_error("Failed to allocate fd set: %s", strerror(-r));
1440                 goto finish;
1441         } else
1442                 fdset_cloexec(fds, true);
1443
1444         if (serialization)
1445                 assert_se(fdset_remove(fds, fileno(serialization)) >= 0);
1446
1447         if (arg_running_as == SYSTEMD_SYSTEM)
1448                 /* Become a session leader if we aren't one yet. */
1449                 setsid();
1450
1451         /* Move out of the way, so that we won't block unmounts */
1452         assert_se(chdir("/")  == 0);
1453
1454         /* Reset the console, but only if this is really init and we
1455          * are freshly booted */
1456         if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN)
1457                 console_setup(getpid() == 1 && !skip_setup);
1458
1459         /* Open the logging devices, if possible and necessary */
1460         log_open();
1461
1462         /* Make sure we leave a core dump without panicing the
1463          * kernel. */
1464         if (getpid() == 1) {
1465                 install_crash_handler();
1466
1467                 r = mount_cgroup_controllers(arg_join_controllers);
1468                 if (r < 0)
1469                         goto finish;
1470         }
1471
1472         if (arg_running_as == SYSTEMD_SYSTEM) {
1473                 const char *virtualization = NULL;
1474
1475                 log_info(PACKAGE_STRING " running in system mode. (" SYSTEMD_FEATURES ")");
1476
1477                 detect_virtualization(&virtualization);
1478                 if (virtualization)
1479                         log_info("Detected virtualization '%s'.", virtualization);
1480
1481                 if (in_initrd())
1482                         log_info("Running in initial RAM disk.");
1483
1484         } else
1485                 log_debug(PACKAGE_STRING " running in user mode. (" SYSTEMD_FEATURES ")");
1486
1487         if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
1488                 if (arg_show_status || plymouth_running())
1489                         status_welcome();
1490
1491 #ifdef HAVE_KMOD
1492                 kmod_setup();
1493 #endif
1494                 hostname_setup();
1495                 machine_id_setup();
1496                 loopback_setup();
1497
1498                 test_mtab();
1499                 test_usr();
1500                 test_cgroups();
1501         }
1502
1503         if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
1504                 watchdog_set_timeout(&arg_runtime_watchdog);
1505
1506         if (arg_timer_slack_nsec != (nsec_t) -1)
1507                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1508                         log_error("Failed to adjust timer slack: %m");
1509
1510         if (arg_capability_bounding_set_drop) {
1511                 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
1512                 if (r < 0) {
1513                         log_error("Failed to drop capability bounding set of usermode helpers: %s", strerror(-r));
1514                         goto finish;
1515                 }
1516                 r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
1517                 if (r < 0) {
1518                         log_error("Failed to drop capability bounding set: %s", strerror(-r));
1519                         goto finish;
1520                 }
1521         }
1522
1523         if (arg_running_as == SYSTEMD_USER) {
1524                 /* Become reaper of our children */
1525                 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
1526                         log_warning("Failed to make us a subreaper: %m");
1527                         if (errno == EINVAL)
1528                                 log_info("Perhaps the kernel version is too old (< 3.4?)");
1529                 }
1530         }
1531
1532         if (arg_running_as == SYSTEMD_SYSTEM)
1533                 bump_rlimit_nofile(&saved_rlimit_nofile);
1534
1535         r = manager_new(arg_running_as, &m);
1536         if (r < 0) {
1537                 log_error("Failed to allocate manager object: %s", strerror(-r));
1538                 goto finish;
1539         }
1540
1541         m->confirm_spawn = arg_confirm_spawn;
1542         m->default_std_output = arg_default_std_output;
1543         m->default_std_error = arg_default_std_error;
1544         m->default_restart_usec = arg_default_restart_usec;
1545         m->default_timeout_start_usec = arg_default_timeout_start_usec;
1546         m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
1547         m->default_start_limit_interval = arg_default_start_limit_interval;
1548         m->default_start_limit_burst = arg_default_start_limit_burst;
1549         m->runtime_watchdog = arg_runtime_watchdog;
1550         m->shutdown_watchdog = arg_shutdown_watchdog;
1551         m->userspace_timestamp = userspace_timestamp;
1552         m->kernel_timestamp = kernel_timestamp;
1553         m->initrd_timestamp = initrd_timestamp;
1554         m->security_start_timestamp = security_start_timestamp;
1555         m->security_finish_timestamp = security_finish_timestamp;
1556
1557         manager_set_default_rlimits(m, arg_default_rlimit);
1558
1559         if (arg_default_environment)
1560                 manager_environment_add(m, NULL, arg_default_environment);
1561
1562         manager_set_show_status(m, arg_show_status);
1563
1564         /* Remember whether we should queue the default job */
1565         queue_default_job = !serialization || arg_switched_root;
1566
1567         before_startup = now(CLOCK_MONOTONIC);
1568
1569         r = manager_startup(m, serialization, fds);
1570         if (r < 0)
1571                 log_error("Failed to fully start up daemon: %s", strerror(-r));
1572
1573         /* This will close all file descriptors that were opened, but
1574          * not claimed by any unit. */
1575         fdset_free(fds);
1576         fds = NULL;
1577
1578         if (serialization) {
1579                 fclose(serialization);
1580                 serialization = NULL;
1581         }
1582
1583         if (queue_default_job) {
1584                 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1585                 Unit *target = NULL;
1586                 Job *default_unit_job;
1587
1588                 log_debug("Activating default unit: %s", arg_default_unit);
1589
1590                 r = manager_load_unit(m, arg_default_unit, NULL, &error, &target);
1591                 if (r < 0)
1592                         log_error("Failed to load default target: %s", bus_error_message(&error, r));
1593                 else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND)
1594                         log_error("Failed to load default target: %s", strerror(-target->load_error));
1595                 else if (target->load_state == UNIT_MASKED)
1596                         log_error("Default target masked.");
1597
1598                 if (!target || target->load_state != UNIT_LOADED) {
1599                         log_info("Trying to load rescue target...");
1600
1601                         r = manager_load_unit(m, SPECIAL_RESCUE_TARGET, NULL, &error, &target);
1602                         if (r < 0) {
1603                                 log_error("Failed to load rescue target: %s", bus_error_message(&error, r));
1604                                 goto finish;
1605                         } else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND) {
1606                                 log_error("Failed to load rescue target: %s", strerror(-target->load_error));
1607                                 goto finish;
1608                         } else if (target->load_state == UNIT_MASKED) {
1609                                 log_error("Rescue target masked.");
1610                                 goto finish;
1611                         }
1612                 }
1613
1614                 assert(target->load_state == UNIT_LOADED);
1615
1616                 if (arg_action == ACTION_TEST) {
1617                         printf("-> By units:\n");
1618                         manager_dump_units(m, stdout, "\t");
1619                 }
1620
1621                 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
1622                 if (r == -EPERM) {
1623                         log_debug("Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1624
1625                         r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
1626                         if (r < 0) {
1627                                 log_error("Failed to start default target: %s", bus_error_message(&error, r));
1628                                 goto finish;
1629                         }
1630                 } else if (r < 0) {
1631                         log_error("Failed to isolate default target: %s", bus_error_message(&error, r));
1632                         goto finish;
1633                 }
1634
1635                 m->default_unit_job_id = default_unit_job->id;
1636
1637                 after_startup = now(CLOCK_MONOTONIC);
1638                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
1639                          "Loaded units and determined initial transaction in %s.",
1640                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 0));
1641
1642                 if (arg_action == ACTION_TEST) {
1643                         printf("-> By jobs:\n");
1644                         manager_dump_jobs(m, stdout, "\t");
1645                         retval = EXIT_SUCCESS;
1646                         goto finish;
1647                 }
1648         }
1649
1650         for (;;) {
1651                 r = manager_loop(m);
1652                 if (r < 0) {
1653                         log_error("Failed to run mainloop: %s", strerror(-r));
1654                         goto finish;
1655                 }
1656
1657                 switch (m->exit_code) {
1658
1659                 case MANAGER_EXIT:
1660                         retval = EXIT_SUCCESS;
1661                         log_debug("Exit.");
1662                         goto finish;
1663
1664                 case MANAGER_RELOAD:
1665                         log_info("Reloading.");
1666                         r = manager_reload(m);
1667                         if (r < 0)
1668                                 log_error("Failed to reload: %s", strerror(-r));
1669                         break;
1670
1671                 case MANAGER_REEXECUTE:
1672
1673                         if (prepare_reexecute(m, &serialization, &fds, false) < 0)
1674                                 goto finish;
1675
1676                         reexecute = true;
1677                         log_notice("Reexecuting.");
1678                         goto finish;
1679
1680                 case MANAGER_SWITCH_ROOT:
1681                         /* Steal the switch root parameters */
1682                         switch_root_dir = m->switch_root;
1683                         switch_root_init = m->switch_root_init;
1684                         m->switch_root = m->switch_root_init = NULL;
1685
1686                         if (!switch_root_init)
1687                                 if (prepare_reexecute(m, &serialization, &fds, true) < 0)
1688                                         goto finish;
1689
1690                         reexecute = true;
1691                         log_notice("Switching root.");
1692                         goto finish;
1693
1694                 case MANAGER_REBOOT:
1695                 case MANAGER_POWEROFF:
1696                 case MANAGER_HALT:
1697                 case MANAGER_KEXEC: {
1698                         static const char * const table[_MANAGER_EXIT_CODE_MAX] = {
1699                                 [MANAGER_REBOOT] = "reboot",
1700                                 [MANAGER_POWEROFF] = "poweroff",
1701                                 [MANAGER_HALT] = "halt",
1702                                 [MANAGER_KEXEC] = "kexec"
1703                         };
1704
1705                         assert_se(shutdown_verb = table[m->exit_code]);
1706                         arm_reboot_watchdog = m->exit_code == MANAGER_REBOOT;
1707
1708                         log_notice("Shutting down.");
1709                         goto finish;
1710                 }
1711
1712                 default:
1713                         assert_not_reached("Unknown exit code.");
1714                 }
1715         }
1716
1717 finish:
1718         if (m)
1719                 manager_free(m);
1720
1721         for (j = 0; j < RLIMIT_NLIMITS; j++)
1722                 free(arg_default_rlimit[j]);
1723
1724         free(arg_default_unit);
1725         free_join_controllers();
1726
1727         label_finish();
1728
1729         if (reexecute) {
1730                 const char **args;
1731                 unsigned i, args_size;
1732
1733                 /* Close and disarm the watchdog, so that the new
1734                  * instance can reinitialize it, but doesn't get
1735                  * rebooted while we do that */
1736                 watchdog_close(true);
1737
1738                 /* Reset the RLIMIT_NOFILE to the kernel default, so
1739                  * that the new systemd can pass the kernel default to
1740                  * its child processes */
1741                 if (saved_rlimit_nofile.rlim_cur > 0)
1742                         setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
1743
1744                 if (switch_root_dir) {
1745                         /* Kill all remaining processes from the
1746                          * initrd, but don't wait for them, so that we
1747                          * can handle the SIGCHLD for them after
1748                          * deserializing. */
1749                         broadcast_signal(SIGTERM, false, true);
1750
1751                         /* And switch root */
1752                         r = switch_root(switch_root_dir);
1753                         if (r < 0)
1754                                 log_error("Failed to switch root, ignoring: %s", strerror(-r));
1755                 }
1756
1757                 args_size = MAX(6, argc+1);
1758                 args = newa(const char*, args_size);
1759
1760                 if (!switch_root_init) {
1761                         char sfd[16];
1762
1763                         /* First try to spawn ourselves with the right
1764                          * path, and with full serialization. We do
1765                          * this only if the user didn't specify an
1766                          * explicit init to spawn. */
1767
1768                         assert(serialization);
1769                         assert(fds);
1770
1771                         snprintf(sfd, sizeof(sfd), "%i", fileno(serialization));
1772                         char_array_0(sfd);
1773
1774                         i = 0;
1775                         args[i++] = SYSTEMD_BINARY_PATH;
1776                         if (switch_root_dir)
1777                                 args[i++] = "--switched-root";
1778                         args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
1779                         args[i++] = "--deserialize";
1780                         args[i++] = sfd;
1781                         args[i++] = NULL;
1782
1783                         /* do not pass along the environment we inherit from the kernel or initrd */
1784                         if (switch_root_dir)
1785                                 clearenv();
1786
1787                         assert(i <= args_size);
1788                         execv(args[0], (char* const*) args);
1789                 }
1790
1791                 /* Try the fallback, if there is any, without any
1792                  * serialization. We pass the original argv[] and
1793                  * envp[]. (Well, modulo the ordering changes due to
1794                  * getopt() in argv[], and some cleanups in envp[],
1795                  * but let's hope that doesn't matter.) */
1796
1797                 if (serialization) {
1798                         fclose(serialization);
1799                         serialization = NULL;
1800                 }
1801
1802                 if (fds) {
1803                         fdset_free(fds);
1804                         fds = NULL;
1805                 }
1806
1807                 /* Reopen the console */
1808                 make_console_stdio();
1809
1810                 for (j = 1, i = 1; j < argc; j++)
1811                         args[i++] = argv[j];
1812                 args[i++] = NULL;
1813                 assert(i <= args_size);
1814
1815                 if (switch_root_init) {
1816                         args[0] = switch_root_init;
1817                         execv(args[0], (char* const*) args);
1818                         log_warning("Failed to execute configured init, trying fallback: %m");
1819                 }
1820
1821                 args[0] = "/sbin/init";
1822                 execv(args[0], (char* const*) args);
1823
1824                 if (errno == ENOENT) {
1825                         log_warning("No /sbin/init, trying fallback");
1826
1827                         args[0] = "/bin/sh";
1828                         args[1] = NULL;
1829                         execv(args[0], (char* const*) args);
1830                         log_error("Failed to execute /bin/sh, giving up: %m");
1831                 } else
1832                         log_warning("Failed to execute /sbin/init, giving up: %m");
1833         }
1834
1835         if (serialization)
1836                 fclose(serialization);
1837
1838         if (fds)
1839                 fdset_free(fds);
1840
1841 #ifdef HAVE_VALGRIND_VALGRIND_H
1842         /* If we are PID 1 and running under valgrind, then let's exit
1843          * here explicitly. valgrind will only generate nice output on
1844          * exit(), not on exec(), hence let's do the former not the
1845          * latter here. */
1846         if (getpid() == 1 && RUNNING_ON_VALGRIND)
1847                 return 0;
1848 #endif
1849
1850         if (shutdown_verb) {
1851                 const char * command_line[] = {
1852                         SYSTEMD_SHUTDOWN_BINARY_PATH,
1853                         shutdown_verb,
1854                         NULL
1855                 };
1856                 char **env_block;
1857
1858                 if (arm_reboot_watchdog && arg_shutdown_watchdog > 0) {
1859                         char e[32];
1860
1861                         /* If we reboot let's set the shutdown
1862                          * watchdog and tell the shutdown binary to
1863                          * repeatedly ping it */
1864                         watchdog_set_timeout(&arg_shutdown_watchdog);
1865                         watchdog_close(false);
1866
1867                         /* Tell the binary how often to ping */
1868                         snprintf(e, sizeof(e), "WATCHDOG_USEC=%llu", (unsigned long long) arg_shutdown_watchdog);
1869                         char_array_0(e);
1870
1871                         env_block = strv_append(environ, e);
1872                 } else {
1873                         env_block = strv_copy(environ);
1874                         watchdog_close(true);
1875                 }
1876
1877                 /* Avoid the creation of new processes forked by the
1878                  * kernel; at this point, we will not listen to the
1879                  * signals anyway */
1880                 if (detect_container(NULL) <= 0)
1881                         cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
1882
1883                 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1884                 free(env_block);
1885                 log_error("Failed to execute shutdown binary, freezing: %m");
1886         }
1887
1888         if (getpid() == 1)
1889                 freeze();
1890
1891         return retval;
1892 }