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 #ifdef HAVE_SECCOMP
  39 #include <seccomp.h>
  40 #endif
  41
  42 #include "sd-daemon.h"
  43 #include "sd-messages.h"
  44 #include "sd-bus.h"
  45 #include "log.h"
  46 #include "fdset.h"
  47 #include "special.h"
  48 #include "conf-parser.h"
  49 #include "missing.h"
  50 #include "label.h"
  51 #include "pager.h"
  52 #include "build.h"
  53 #include "strv.h"
  54 #include "def.h"
  55 #include "virt.h"
  56 #include "architecture.h"
  57 #include "watchdog.h"
  58 #include "path-util.h"
  59 #include "switch-root.h"
  60 #include "capability.h"
  61 #include "killall.h"
  62 #include "env-util.h"
  63 #include "clock-util.h"
  64 #include "fileio.h"
  65 #include "bus-error.h"
  66 #include "bus-util.h"
  67 #include "selinux-util.h"
  68 #include "manager.h"
  69 #include "dbus-manager.h"
  70 #include "load-fragment.h"
  71
  72 #include "mount-setup.h"
  73 #include "loopback-setup.h"
  74 #include "hostname-setup.h"
  75 #include "machine-id-setup.h"
  76 #include "selinux-setup.h"
  77 #include "ima-setup.h"
  78 #include "smack-setup.h"
  79 #include "kmod-setup.h"
  80
  81 static enum {
  82         ACTION_RUN,
  83         ACTION_HELP,
  84         ACTION_VERSION,
  85         ACTION_TEST,
  86         ACTION_DUMP_CONFIGURATION_ITEMS,
  87         ACTION_DONE
  88 } arg_action = ACTION_RUN;
  89 static char *arg_default_unit = NULL;
  90 static SystemdRunningAs arg_running_as = _SYSTEMD_RUNNING_AS_INVALID;
  91 static bool arg_dump_core = true;
  92 static bool arg_crash_shell = false;
  93 static int arg_crash_chvt = -1;
  94 static bool arg_confirm_spawn = false;
  95 static ShowStatus arg_show_status = _SHOW_STATUS_UNSET;
  96 static bool arg_switched_root = false;
  97 static int arg_no_pager = -1;
  98 static char ***arg_join_controllers = NULL;
  99 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
 100 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
 101 static usec_t arg_default_restart_usec = DEFAULT_RESTART_USEC;
 102 static usec_t arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
 103 static usec_t arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
 104 static usec_t arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
 105 static unsigned arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
 106 static usec_t arg_runtime_watchdog = 0;
 107 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
 108 static char **arg_default_environment = NULL;
 109 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX] = {};
 110 static uint64_t arg_capability_bounding_set_drop = 0;
 111 static nsec_t arg_timer_slack_nsec = NSEC_INFINITY;
 112 static usec_t arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
 113 static Set* arg_syscall_archs = NULL;
 114 static FILE* arg_serialization = NULL;
 115 static bool arg_default_cpu_accounting = false;
 116 static bool arg_default_blockio_accounting = false;
 117 static bool arg_default_memory_accounting = false;
 118
 119 static void nop_handler(int sig) {}
 120
 121 static void pager_open_if_enabled(void) {
 122
 123         if (arg_no_pager <= 0)
 124                 return;
 125
 126         pager_open(false);
 127 }
 128
 129 noreturn static void crash(int sig) {
 130
 131         if (getpid() != 1)
 132                 /* Pass this on immediately, if this is not PID 1 */
 133                 raise(sig);
 134         else if (!arg_dump_core)
 135                 log_emergency("Caught <%s>, not dumping core.", signal_to_string(sig));
 136         else {
 137                 struct sigaction sa = {
 138                         .sa_handler = nop_handler,
 139                         .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 140                 };
 141                 pid_t pid;
 142
 143                 /* We want to wait for the core process, hence let's enable SIGCHLD */
 144                 sigaction(SIGCHLD, &sa, NULL);
 145
 146                 pid = raw_clone(SIGCHLD, NULL);
 147                 if (pid < 0)
 148                         log_emergency_errno(errno, "Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
 149
 150                 else if (pid == 0) {
 151                         struct rlimit rl = {};
 152
 153                         /* Enable default signal handler for core dump */
 154                         zero(sa);
 155                         sa.sa_handler = SIG_DFL;
 156                         sigaction(sig, &sa, NULL);
 157
 158                         /* Don't limit the core dump size */
 159                         rl.rlim_cur = RLIM_INFINITY;
 160                         rl.rlim_max = RLIM_INFINITY;
 161                         setrlimit(RLIMIT_CORE, &rl);
 162
 163                         /* Just to be sure... */
 164                         chdir("/");
 165
 166                         /* Raise the signal again */
 167                         pid = raw_getpid();
 168                         kill(pid, sig); /* raise() would kill the parent */
 169
 170                         assert_not_reached("We shouldn't be here...");
 171                         _exit(1);
 172                 } else {
 173                         siginfo_t status;
 174                         int r;
 175
 176                         /* Order things nicely. */
 177                         r = wait_for_terminate(pid, &status);
 178                         if (r < 0)
 179                                 log_emergency_errno(r, "Caught <%s>, waitpid() failed: %m", signal_to_string(sig));
 180                         else if (status.si_code != CLD_DUMPED)
 181                                 log_emergency("Caught <%s>, core dump failed (child "PID_FMT", code=%s, status=%i/%s).",
 182                                               signal_to_string(sig),
 183                                               pid, sigchld_code_to_string(status.si_code),
 184                                               status.si_status,
 185                                               strna(status.si_code == CLD_EXITED
 186                                                     ? exit_status_to_string(status.si_status, EXIT_STATUS_FULL)
 187                                                     : signal_to_string(status.si_status)));
 188                         else
 189                                 log_emergency("Caught <%s>, dumped core as pid "PID_FMT".", signal_to_string(sig), pid);
 190                 }
 191         }
 192
 193         if (arg_crash_chvt)
 194                 chvt(arg_crash_chvt);
 195
 196         if (arg_crash_shell) {
 197                 struct sigaction sa = {
 198                         .sa_handler = SIG_IGN,
 199                         .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
 200                 };
 201                 pid_t pid;
 202
 203                 log_info("Executing crash shell in 10s...");
 204                 sleep(10);
 205
 206                 /* Let the kernel reap children for us */
 207                 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 208
 209                 pid = raw_clone(SIGCHLD, NULL);
 210                 if (pid < 0)
 211                         log_emergency_errno(errno, "Failed to fork off crash shell: %m");
 212                 else if (pid == 0) {
 213                         make_console_stdio();
 214                         execle("/bin/sh", "/bin/sh", NULL, environ);
 215
 216                         log_emergency_errno(errno, "execle() failed: %m");
 217                         _exit(1);
 218                 } else
 219                         log_info("Successfully spawned crash shell as PID "PID_FMT".", pid);
 220         }
 221
 222         log_emergency("Freezing execution.");
 223         freeze();
 224 }
 225
 226 static void install_crash_handler(void) {
 227         static const struct sigaction sa = {
 228                 .sa_handler = crash,
 229                 .sa_flags = SA_NODEFER, /* So that we can raise the signal again from the signal handler */
 230         };
 231         int r;
 232
 233         /* We ignore the return value here, since, we don't mind if we
 234          * cannot set up a crash handler */
 235         r = sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
 236         if (r < 0)
 237                 log_debug_errno(r, "I had trouble setting up the crash handler, ignoring: %m");
 238 }
 239
 240 static int console_setup(void) {
 241         _cleanup_close_ int tty_fd = -1;
 242         int r;
 243
 244         tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
 245         if (tty_fd < 0)
 246                 return log_error_errno(tty_fd, "Failed to open /dev/console: %m");
 247
 248         /* We don't want to force text mode.  plymouth may be showing
 249          * pictures already from initrd. */
 250         r = reset_terminal_fd(tty_fd, false);
 251         if (r < 0)
 252                 return log_error_errno(r, "Failed to reset /dev/console: %m");
 253
 254         return 0;
 255 }
 256
 257 static int set_default_unit(const char *u) {
 258         char *c;
 259
 260         assert(u);
 261
 262         c = strdup(u);
 263         if (!c)
 264                 return -ENOMEM;
 265
 266         free(arg_default_unit);
 267         arg_default_unit = c;
 268
 269         return 0;
 270 }
 271
 272 static int parse_proc_cmdline_item(const char *key, const char *value) {
 273
 274         static const char * const rlmap[] = {
 275                 "emergency", SPECIAL_EMERGENCY_TARGET,
 276                 "-b",        SPECIAL_EMERGENCY_TARGET,
 277                 "rescue",    SPECIAL_RESCUE_TARGET,
 278                 "single",    SPECIAL_RESCUE_TARGET,
 279                 "-s",        SPECIAL_RESCUE_TARGET,
 280                 "s",         SPECIAL_RESCUE_TARGET,
 281                 "S",         SPECIAL_RESCUE_TARGET,
 282                 "1",         SPECIAL_RESCUE_TARGET,
 283                 "2",         SPECIAL_RUNLEVEL2_TARGET,
 284                 "3",         SPECIAL_RUNLEVEL3_TARGET,
 285                 "4",         SPECIAL_RUNLEVEL4_TARGET,
 286                 "5",         SPECIAL_RUNLEVEL5_TARGET,
 287         };
 288         int r;
 289
 290         assert(key);
 291
 292         if (streq(key, "systemd.unit") && value) {
 293
 294                 if (!in_initrd())
 295                         return set_default_unit(value);
 296
 297         } else if (streq(key, "rd.systemd.unit") && value) {
 298
 299                 if (in_initrd())
 300                         return set_default_unit(value);
 301
 302         } else if (streq(key, "systemd.dump_core") && value) {
 303
 304                 r = parse_boolean(value);
 305                 if (r < 0)
 306                         log_warning("Failed to parse dump core switch %s. Ignoring.", value);
 307                 else
 308                         arg_dump_core = r;
 309
 310         } else if (streq(key, "systemd.crash_shell") && value) {
 311
 312                 r = parse_boolean(value);
 313                 if (r < 0)
 314                         log_warning("Failed to parse crash shell switch %s. Ignoring.", value);
 315                 else
 316                         arg_crash_shell = r;
 317
 318         } else if (streq(key, "systemd.crash_chvt") && value) {
 319
 320                 if (safe_atoi(value, &r) < 0)
 321                         log_warning("Failed to parse crash chvt switch %s. Ignoring.", value);
 322                 else
 323                         arg_crash_chvt = r;
 324
 325         } else if (streq(key, "systemd.confirm_spawn") && value) {
 326
 327                 r = parse_boolean(value);
 328                 if (r < 0)
 329                         log_warning("Failed to parse confirm spawn switch %s. Ignoring.", value);
 330                 else
 331                         arg_confirm_spawn = r;
 332
 333         } else if (streq(key, "systemd.show_status") && value) {
 334
 335                 r = parse_show_status(value, &arg_show_status);
 336                 if (r < 0)
 337                         log_warning("Failed to parse show status switch %s. Ignoring.", value);
 338
 339         } else if (streq(key, "systemd.default_standard_output") && value) {
 340
 341                 r = exec_output_from_string(value);
 342                 if (r < 0)
 343                         log_warning("Failed to parse default standard output switch %s. Ignoring.", value);
 344                 else
 345                         arg_default_std_output = r;
 346
 347         } else if (streq(key, "systemd.default_standard_error") && value) {
 348
 349                 r = exec_output_from_string(value);
 350                 if (r < 0)
 351                         log_warning("Failed to parse default standard error switch %s. Ignoring.", value);
 352                 else
 353                         arg_default_std_error = r;
 354
 355         } else if (streq(key, "systemd.setenv") && value) {
 356
 357                 if (env_assignment_is_valid(value)) {
 358                         char **env;
 359
 360                         env = strv_env_set(arg_default_environment, value);
 361                         if (env)
 362                                 arg_default_environment = env;
 363                         else
 364                                 log_warning_errno(ENOMEM, "Setting environment variable '%s' failed, ignoring: %m", value);
 365                 } else
 366                         log_warning("Environment variable name '%s' is not valid. Ignoring.", value);
 367
 368         } else if (streq(key, "quiet") && !value) {
 369
 370                 log_set_max_level(LOG_NOTICE);
 371
 372                 if (arg_show_status == _SHOW_STATUS_UNSET)
 373                         arg_show_status = SHOW_STATUS_AUTO;
 374
 375         } else if (streq(key, "debug") && !value) {
 376
 377                 /* Note that log_parse_environment() handles 'debug'
 378                  * too, and sets the log level to LOG_DEBUG. */
 379
 380                 if (detect_container(NULL) > 0)
 381                         log_set_target(LOG_TARGET_CONSOLE);
 382
 383         } else if (!in_initrd() && !value) {
 384                 unsigned i;
 385
 386                 /* SysV compatibility */
 387                 for (i = 0; i < ELEMENTSOF(rlmap); i += 2)
 388                         if (streq(key, rlmap[i]))
 389                                 return set_default_unit(rlmap[i+1]);
 390         }
 391
 392         return 0;
 393 }
 394
 395 #define DEFINE_SETTER(name, func, descr)                              \
 396         static int name(const char *unit,                             \
 397                         const char *filename,                         \
 398                         unsigned line,                                \
 399                         const char *section,                          \
 400                         unsigned section_line,                        \
 401                         const char *lvalue,                           \
 402                         int ltype,                                    \
 403                         const char *rvalue,                           \
 404                         void *data,                                   \
 405                         void *userdata) {                             \
 406                                                                       \
 407                 int r;                                                \
 408                                                                       \
 409                 assert(filename);                                     \
 410                 assert(lvalue);                                       \
 411                 assert(rvalue);                                       \
 412                                                                       \
 413                 r = func(rvalue);                                     \
 414                 if (r < 0)                                            \
 415                         log_syntax(unit, LOG_ERR, filename, line, -r, \
 416                                    "Invalid " descr "'%s': %s",       \
 417                                    rvalue, strerror(-r));             \
 418                                                                       \
 419                 return 0;                                             \
 420         }
 421
 422 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level")
 423 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target")
 424 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" )
 425 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location")
 426
 427 static int config_parse_cpu_affinity2(
 428                 const char *unit,
 429                 const char *filename,
 430                 unsigned line,
 431                 const char *section,
 432                 unsigned section_line,
 433                 const char *lvalue,
 434                 int ltype,
 435                 const char *rvalue,
 436                 void *data,
 437                 void *userdata) {
 438
 439         const char *word, *state;
 440         size_t l;
 441         cpu_set_t *c = NULL;
 442         unsigned ncpus = 0;
 443
 444         assert(filename);
 445         assert(lvalue);
 446         assert(rvalue);
 447
 448         FOREACH_WORD_QUOTED(word, l, rvalue, state) {
 449                 char *t;
 450                 int r;
 451                 unsigned cpu;
 452
 453                 if (!(t = strndup(word, l)))
 454                         return log_oom();
 455
 456                 r = safe_atou(t, &cpu);
 457                 free(t);
 458
 459                 if (!c)
 460                         if (!(c = cpu_set_malloc(&ncpus)))
 461                                 return log_oom();
 462
 463                 if (r < 0 || cpu >= ncpus) {
 464                         log_syntax(unit, LOG_ERR, filename, line, -r,
 465                                    "Failed to parse CPU affinity '%s'", rvalue);
 466                         CPU_FREE(c);
 467                         return -EBADMSG;
 468                 }
 469
 470                 CPU_SET_S(cpu, CPU_ALLOC_SIZE(ncpus), c);
 471         }
 472         if (!isempty(state))
 473                 log_syntax(unit, LOG_ERR, filename, line, EINVAL,
 474                            "Trailing garbage, ignoring.");
 475
 476         if (c) {
 477                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
 478                         log_unit_warning(unit, "Failed to set CPU affinity: %m");
 479
 480                 CPU_FREE(c);
 481         }
 482
 483         return 0;
 484 }
 485
 486 static int config_parse_show_status(
 487                 const char* unit,
 488                 const char *filename,
 489                 unsigned line,
 490                 const char *section,
 491                 unsigned section_line,
 492                 const char *lvalue,
 493                 int ltype,
 494                 const char *rvalue,
 495                 void *data,
 496                 void *userdata) {
 497
 498         int k;
 499         ShowStatus *b = data;
 500
 501         assert(filename);
 502         assert(lvalue);
 503         assert(rvalue);
 504         assert(data);
 505
 506         k = parse_show_status(rvalue, b);
 507         if (k < 0) {
 508                 log_syntax(unit, LOG_ERR, filename, line, -k,
 509                            "Failed to parse show status setting, ignoring: %s", rvalue);
 510                 return 0;
 511         }
 512
 513         return 0;
 514 }
 515
 516 static void strv_free_free(char ***l) {
 517         char ***i;
 518
 519         if (!l)
 520                 return;
 521
 522         for (i = l; *i; i++)
 523                 strv_free(*i);
 524
 525         free(l);
 526 }
 527
 528 static void free_join_controllers(void) {
 529         strv_free_free(arg_join_controllers);
 530         arg_join_controllers = NULL;
 531 }
 532
 533 static int config_parse_join_controllers(const char *unit,
 534                                          const char *filename,
 535                                          unsigned line,
 536                                          const char *section,
 537                                          unsigned section_line,
 538                                          const char *lvalue,
 539                                          int ltype,
 540                                          const char *rvalue,
 541                                          void *data,
 542                                          void *userdata) {
 543
 544         unsigned n = 0;
 545         const char *word, *state;
 546         size_t length;
 547
 548         assert(filename);
 549         assert(lvalue);
 550         assert(rvalue);
 551
 552         free_join_controllers();
 553
 554         FOREACH_WORD_QUOTED(word, length, rvalue, state) {
 555                 char *s, **l;
 556
 557                 s = strndup(word, length);
 558                 if (!s)
 559                         return log_oom();
 560
 561                 l = strv_split(s, ",");
 562                 free(s);
 563
 564                 strv_uniq(l);
 565
 566                 if (strv_length(l) <= 1) {
 567                         strv_free(l);
 568                         continue;
 569                 }
 570
 571                 if (!arg_join_controllers) {
 572                         arg_join_controllers = new(char**, 2);
 573                         if (!arg_join_controllers) {
 574                                 strv_free(l);
 575                                 return log_oom();
 576                         }
 577
 578                         arg_join_controllers[0] = l;
 579                         arg_join_controllers[1] = NULL;
 580
 581                         n = 1;
 582                 } else {
 583                         char ***a;
 584                         char ***t;
 585
 586                         t = new0(char**, n+2);
 587                         if (!t) {
 588                                 strv_free(l);
 589                                 return log_oom();
 590                         }
 591
 592                         n = 0;
 593
 594                         for (a = arg_join_controllers; *a; a++) {
 595
 596                                 if (strv_overlap(*a, l)) {
 597                                         if (strv_extend_strv(&l, *a) < 0) {
 598                                                 strv_free(l);
 599                                                 strv_free_free(t);
 600                                                 return log_oom();
 601                                         }
 602
 603                                 } else {
 604                                         char **c;
 605
 606                                         c = strv_copy(*a);
 607                                         if (!c) {
 608                                                 strv_free(l);
 609                                                 strv_free_free(t);
 610                                                 return log_oom();
 611                                         }
 612
 613                                         t[n++] = c;
 614                                 }
 615                         }
 616
 617                         t[n++] = strv_uniq(l);
 618
 619                         strv_free_free(arg_join_controllers);
 620                         arg_join_controllers = t;
 621                 }
 622         }
 623         if (!isempty(state))
 624                 log_syntax(unit, LOG_ERR, filename, line, EINVAL,
 625                            "Trailing garbage, ignoring.");
 626
 627         return 0;
 628 }
 629
 630 static int parse_config_file(void) {
 631
 632         const ConfigTableItem items[] = {
 633                 { "Manager", "LogLevel",                  config_parse_level2,           0, NULL                                   },
 634                 { "Manager", "LogTarget",                 config_parse_target,           0, NULL                                   },
 635                 { "Manager", "LogColor",                  config_parse_color,            0, NULL                                   },
 636                 { "Manager", "LogLocation",               config_parse_location,         0, NULL                                   },
 637                 { "Manager", "DumpCore",                  config_parse_bool,             0, &arg_dump_core                         },
 638                 { "Manager", "CrashShell",                config_parse_bool,             0, &arg_crash_shell                       },
 639                 { "Manager", "ShowStatus",                config_parse_show_status,      0, &arg_show_status                       },
 640                 { "Manager", "CrashChVT",                 config_parse_int,              0, &arg_crash_chvt                        },
 641                 { "Manager", "CPUAffinity",               config_parse_cpu_affinity2,    0, NULL                                   },
 642                 { "Manager", "JoinControllers",           config_parse_join_controllers, 0, &arg_join_controllers                  },
 643                 { "Manager", "RuntimeWatchdogSec",        config_parse_sec,              0, &arg_runtime_watchdog                  },
 644                 { "Manager", "ShutdownWatchdogSec",       config_parse_sec,              0, &arg_shutdown_watchdog                 },
 645                 { "Manager", "CapabilityBoundingSet",     config_parse_bounding_set,     0, &arg_capability_bounding_set_drop      },
 646 #ifdef HAVE_SECCOMP
 647                 { "Manager", "SystemCallArchitectures",   config_parse_syscall_archs,    0, &arg_syscall_archs                     },
 648 #endif
 649                 { "Manager", "TimerSlackNSec",            config_parse_nsec,             0, &arg_timer_slack_nsec                  },
 650                 { "Manager", "DefaultTimerAccuracySec",   config_parse_sec,              0, &arg_default_timer_accuracy_usec       },
 651                 { "Manager", "DefaultStandardOutput",     config_parse_output,           0, &arg_default_std_output                },
 652                 { "Manager", "DefaultStandardError",      config_parse_output,           0, &arg_default_std_error                 },
 653                 { "Manager", "DefaultTimeoutStartSec",    config_parse_sec,              0, &arg_default_timeout_start_usec        },
 654                 { "Manager", "DefaultTimeoutStopSec",     config_parse_sec,              0, &arg_default_timeout_stop_usec         },
 655                 { "Manager", "DefaultRestartSec",         config_parse_sec,              0, &arg_default_restart_usec              },
 656                 { "Manager", "DefaultStartLimitInterval", config_parse_sec,              0, &arg_default_start_limit_interval      },
 657                 { "Manager", "DefaultStartLimitBurst",    config_parse_unsigned,         0, &arg_default_start_limit_burst         },
 658                 { "Manager", "DefaultEnvironment",        config_parse_environ,          0, &arg_default_environment               },
 659                 { "Manager", "DefaultLimitCPU",           config_parse_limit,            0, &arg_default_rlimit[RLIMIT_CPU]        },
 660                 { "Manager", "DefaultLimitFSIZE",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_FSIZE]      },
 661                 { "Manager", "DefaultLimitDATA",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_DATA]       },
 662                 { "Manager", "DefaultLimitSTACK",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_STACK]      },
 663                 { "Manager", "DefaultLimitCORE",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_CORE]       },
 664                 { "Manager", "DefaultLimitRSS",           config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RSS]        },
 665                 { "Manager", "DefaultLimitNOFILE",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NOFILE]     },
 666                 { "Manager", "DefaultLimitAS",            config_parse_limit,            0, &arg_default_rlimit[RLIMIT_AS]         },
 667                 { "Manager", "DefaultLimitNPROC",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NPROC]      },
 668                 { "Manager", "DefaultLimitMEMLOCK",       config_parse_limit,            0, &arg_default_rlimit[RLIMIT_MEMLOCK]    },
 669                 { "Manager", "DefaultLimitLOCKS",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_LOCKS]      },
 670                 { "Manager", "DefaultLimitSIGPENDING",    config_parse_limit,            0, &arg_default_rlimit[RLIMIT_SIGPENDING] },
 671                 { "Manager", "DefaultLimitMSGQUEUE",      config_parse_limit,            0, &arg_default_rlimit[RLIMIT_MSGQUEUE]   },
 672                 { "Manager", "DefaultLimitNICE",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NICE]       },
 673                 { "Manager", "DefaultLimitRTPRIO",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RTPRIO]     },
 674                 { "Manager", "DefaultLimitRTTIME",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RTTIME]     },
 675                 { "Manager", "DefaultCPUAccounting",      config_parse_bool,             0, &arg_default_cpu_accounting            },
 676                 { "Manager", "DefaultBlockIOAccounting",  config_parse_bool,             0, &arg_default_blockio_accounting        },
 677                 { "Manager", "DefaultMemoryAccounting",   config_parse_bool,             0, &arg_default_memory_accounting         },
 678                 {}
 679         };
 680
 681         const char *fn, *conf_dirs_nulstr;
 682
 683         fn = arg_running_as == SYSTEMD_SYSTEM ? PKGSYSCONFDIR "/system.conf" : PKGSYSCONFDIR "/user.conf";
 684         conf_dirs_nulstr = arg_running_as == SYSTEMD_SYSTEM ? CONF_DIRS_NULSTR("systemd/system.conf") : CONF_DIRS_NULSTR("systemd/user.conf");
 685         config_parse_many(fn, conf_dirs_nulstr, "Manager\0",
 686                           config_item_table_lookup, items, false, NULL);
 687
 688         return 0;
 689 }
 690
 691 static int parse_argv(int argc, char *argv[]) {
 692
 693         enum {
 694                 ARG_LOG_LEVEL = 0x100,
 695                 ARG_LOG_TARGET,
 696                 ARG_LOG_COLOR,
 697                 ARG_LOG_LOCATION,
 698                 ARG_UNIT,
 699                 ARG_SYSTEM,
 700                 ARG_USER,
 701                 ARG_TEST,
 702                 ARG_NO_PAGER,
 703                 ARG_VERSION,
 704                 ARG_DUMP_CONFIGURATION_ITEMS,
 705                 ARG_DUMP_CORE,
 706                 ARG_CRASH_SHELL,
 707                 ARG_CONFIRM_SPAWN,
 708                 ARG_SHOW_STATUS,
 709                 ARG_DESERIALIZE,
 710                 ARG_SWITCHED_ROOT,
 711                 ARG_DEFAULT_STD_OUTPUT,
 712                 ARG_DEFAULT_STD_ERROR
 713         };
 714
 715         static const struct option options[] = {
 716                 { "log-level",                required_argument, NULL, ARG_LOG_LEVEL                },
 717                 { "log-target",               required_argument, NULL, ARG_LOG_TARGET               },
 718                 { "log-color",                optional_argument, NULL, ARG_LOG_COLOR                },
 719                 { "log-location",             optional_argument, NULL, ARG_LOG_LOCATION             },
 720                 { "unit",                     required_argument, NULL, ARG_UNIT                     },
 721                 { "system",                   no_argument,       NULL, ARG_SYSTEM                   },
 722                 { "user",                     no_argument,       NULL, ARG_USER                     },
 723                 { "test",                     no_argument,       NULL, ARG_TEST                     },
 724                 { "no-pager",                 no_argument,       NULL, ARG_NO_PAGER                 },
 725                 { "help",                     no_argument,       NULL, 'h'                          },
 726                 { "version",                  no_argument,       NULL, ARG_VERSION                  },
 727                 { "dump-configuration-items", no_argument,       NULL, ARG_DUMP_CONFIGURATION_ITEMS },
 728                 { "dump-core",                optional_argument, NULL, ARG_DUMP_CORE                },
 729                 { "crash-shell",              optional_argument, NULL, ARG_CRASH_SHELL              },
 730                 { "confirm-spawn",            optional_argument, NULL, ARG_CONFIRM_SPAWN            },
 731                 { "show-status",              optional_argument, NULL, ARG_SHOW_STATUS              },
 732                 { "deserialize",              required_argument, NULL, ARG_DESERIALIZE              },
 733                 { "switched-root",            no_argument,       NULL, ARG_SWITCHED_ROOT            },
 734                 { "default-standard-output",  required_argument, NULL, ARG_DEFAULT_STD_OUTPUT,      },
 735                 { "default-standard-error",   required_argument, NULL, ARG_DEFAULT_STD_ERROR,       },
 736                 {}
 737         };
 738
 739         int c, r;
 740
 741         assert(argc >= 1);
 742         assert(argv);
 743
 744         if (getpid() == 1)
 745                 opterr = 0;
 746
 747         while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
 748
 749                 switch (c) {
 750
 751                 case ARG_LOG_LEVEL:
 752                         r = log_set_max_level_from_string(optarg);
 753                         if (r < 0) {
 754                                 log_error("Failed to parse log level %s.", optarg);
 755                                 return r;
 756                         }
 757
 758                         break;
 759
 760                 case ARG_LOG_TARGET:
 761                         r = log_set_target_from_string(optarg);
 762                         if (r < 0) {
 763                                 log_error("Failed to parse log target %s.", optarg);
 764                                 return r;
 765                         }
 766
 767                         break;
 768
 769                 case ARG_LOG_COLOR:
 770
 771                         if (optarg) {
 772                                 r = log_show_color_from_string(optarg);
 773                                 if (r < 0) {
 774                                         log_error("Failed to parse log color setting %s.", optarg);
 775                                         return r;
 776                                 }
 777                         } else
 778                                 log_show_color(true);
 779
 780                         break;
 781
 782                 case ARG_LOG_LOCATION:
 783                         if (optarg) {
 784                                 r = log_show_location_from_string(optarg);
 785                                 if (r < 0) {
 786                                         log_error("Failed to parse log location setting %s.", optarg);
 787                                         return r;
 788                                 }
 789                         } else
 790                                 log_show_location(true);
 791
 792                         break;
 793
 794                 case ARG_DEFAULT_STD_OUTPUT:
 795                         r = exec_output_from_string(optarg);
 796                         if (r < 0) {
 797                                 log_error("Failed to parse default standard output setting %s.", optarg);
 798                                 return r;
 799                         } else
 800                                 arg_default_std_output = r;
 801                         break;
 802
 803                 case ARG_DEFAULT_STD_ERROR:
 804                         r = exec_output_from_string(optarg);
 805                         if (r < 0) {
 806                                 log_error("Failed to parse default standard error output setting %s.", optarg);
 807                                 return r;
 808                         } else
 809                                 arg_default_std_error = r;
 810                         break;
 811
 812                 case ARG_UNIT:
 813
 814                         r = set_default_unit(optarg);
 815                         if (r < 0)
 816                                 return log_error_errno(r, "Failed to set default unit %s: %m", optarg);
 817
 818                         break;
 819
 820                 case ARG_SYSTEM:
 821                         arg_running_as = SYSTEMD_SYSTEM;
 822                         break;
 823
 824                 case ARG_USER:
 825                         arg_running_as = SYSTEMD_USER;
 826                         break;
 827
 828                 case ARG_TEST:
 829                         arg_action = ACTION_TEST;
 830                         if (arg_no_pager < 0)
 831                                 arg_no_pager = true;
 832                         break;
 833
 834                 case ARG_NO_PAGER:
 835                         arg_no_pager = true;
 836                         break;
 837
 838                 case ARG_VERSION:
 839                         arg_action = ACTION_VERSION;
 840                         break;
 841
 842                 case ARG_DUMP_CONFIGURATION_ITEMS:
 843                         arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
 844                         break;
 845
 846                 case ARG_DUMP_CORE:
 847                         r = optarg ? parse_boolean(optarg) : 1;
 848                         if (r < 0) {
 849                                 log_error("Failed to parse dump core boolean %s.", optarg);
 850                                 return r;
 851                         }
 852                         arg_dump_core = r;
 853                         break;
 854
 855                 case ARG_CRASH_SHELL:
 856                         r = optarg ? parse_boolean(optarg) : 1;
 857                         if (r < 0) {
 858                                 log_error("Failed to parse crash shell boolean %s.", optarg);
 859                                 return r;
 860                         }
 861                         arg_crash_shell = r;
 862                         break;
 863
 864                 case ARG_CONFIRM_SPAWN:
 865                         r = optarg ? parse_boolean(optarg) : 1;
 866                         if (r < 0) {
 867                                 log_error("Failed to parse confirm spawn boolean %s.", optarg);
 868                                 return r;
 869                         }
 870                         arg_confirm_spawn = r;
 871                         break;
 872
 873                 case ARG_SHOW_STATUS:
 874                         if (optarg) {
 875                                 r = parse_show_status(optarg, &arg_show_status);
 876                                 if (r < 0) {
 877                                         log_error("Failed to parse show status boolean %s.", optarg);
 878                                         return r;
 879                                 }
 880                         } else
 881                                 arg_show_status = SHOW_STATUS_YES;
 882                         break;
 883
 884                 case ARG_DESERIALIZE: {
 885                         int fd;
 886                         FILE *f;
 887
 888                         r = safe_atoi(optarg, &fd);
 889                         if (r < 0 || fd < 0) {
 890                                 log_error("Failed to parse deserialize option %s.", optarg);
 891                                 return r < 0 ? r : -EINVAL;
 892                         }
 893
 894                         fd_cloexec(fd, true);
 895
 896                         f = fdopen(fd, "r");
 897                         if (!f)
 898                                 return log_error_errno(errno, "Failed to open serialization fd: %m");
 899
 900                         if (arg_serialization)
 901                                 fclose(arg_serialization);
 902
 903                         arg_serialization = f;
 904
 905                         break;
 906                 }
 907
 908                 case ARG_SWITCHED_ROOT:
 909                         arg_switched_root = true;
 910                         break;
 911
 912                 case 'h':
 913                         arg_action = ACTION_HELP;
 914                         if (arg_no_pager < 0)
 915                                 arg_no_pager = true;
 916                         break;
 917
 918                 case 'D':
 919                         log_set_max_level(LOG_DEBUG);
 920                         break;
 921
 922                 case 'b':
 923                 case 's':
 924                 case 'z':
 925                         /* Just to eat away the sysvinit kernel
 926                          * cmdline args without getopt() error
 927                          * messages that we'll parse in
 928                          * parse_proc_cmdline_word() or ignore. */
 929
 930                 case '?':
 931                         if (getpid() != 1)
 932                                 return -EINVAL;
 933                         else
 934                                 return 0;
 935
 936                 default:
 937                         assert_not_reached("Unhandled option code.");
 938                 }
 939
 940         if (optind < argc && getpid() != 1) {
 941                 /* Hmm, when we aren't run as init system
 942                  * let's complain about excess arguments */
 943
 944                 log_error("Excess arguments.");
 945                 return -EINVAL;
 946         }
 947
 948         return 0;
 949 }
 950
 951 static int help(void) {
 952
 953         printf("%s [OPTIONS...]\n\n"
 954                "Starts up and maintains the system or user services.\n\n"
 955                "  -h --help                      Show this help\n"
 956                "     --test                      Determine startup sequence, dump it and exit\n"
 957                "     --no-pager                  Do not pipe output into a pager\n"
 958                "     --dump-configuration-items  Dump understood unit configuration items\n"
 959                "     --unit=UNIT                 Set default unit\n"
 960                "     --system                    Run a system instance, even if PID != 1\n"
 961                "     --user                      Run a user instance\n"
 962                "     --dump-core[=0|1]           Dump core on crash\n"
 963                "     --crash-shell[=0|1]         Run shell on crash\n"
 964                "     --confirm-spawn[=0|1]       Ask for confirmation when spawning processes\n"
 965                "     --show-status[=0|1]         Show status updates on the console during bootup\n"
 966                "     --log-target=TARGET         Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
 967                "     --log-level=LEVEL           Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
 968                "     --log-color[=0|1]           Highlight important log messages\n"
 969                "     --log-location[=0|1]        Include code location in log messages\n"
 970                "     --default-standard-output=  Set default standard output for services\n"
 971                "     --default-standard-error=   Set default standard error output for services\n",
 972                program_invocation_short_name);
 973
 974         return 0;
 975 }
 976
 977 static int version(void) {
 978         puts(PACKAGE_STRING);
 979         puts(SYSTEMD_FEATURES);
 980
 981         return 0;
 982 }
 983
 984 static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool switching_root) {
 985         FILE *f = NULL;
 986         FDSet *fds = NULL;
 987         int r;
 988
 989         assert(m);
 990         assert(_f);
 991         assert(_fds);
 992
 993         r = manager_open_serialization(m, &f);
 994         if (r < 0) {
 995                 log_error_errno(r, "Failed to create serialization file: %m");
 996                 goto fail;
 997         }
 998
 999         /* Make sure nothing is really destructed when we shut down */
1000         m->n_reloading ++;
1001         bus_manager_send_reloading(m, true);
1002
1003         fds = fdset_new();
1004         if (!fds) {
1005                 r = -ENOMEM;
1006                 log_error_errno(r, "Failed to allocate fd set: %m");
1007                 goto fail;
1008         }
1009
1010         r = manager_serialize(m, f, fds, switching_root);
1011         if (r < 0) {
1012                 log_error_errno(r, "Failed to serialize state: %m");
1013                 goto fail;
1014         }
1015
1016         if (fseeko(f, 0, SEEK_SET) < 0) {
1017                 log_error_errno(errno, "Failed to rewind serialization fd: %m");
1018                 goto fail;
1019         }
1020
1021         r = fd_cloexec(fileno(f), false);
1022         if (r < 0) {
1023                 log_error_errno(r, "Failed to disable O_CLOEXEC for serialization: %m");
1024                 goto fail;
1025         }
1026
1027         r = fdset_cloexec(fds, false);
1028         if (r < 0) {
1029                 log_error_errno(r, "Failed to disable O_CLOEXEC for serialization fds: %m");
1030                 goto fail;
1031         }
1032
1033         *_f = f;
1034         *_fds = fds;
1035
1036         return 0;
1037
1038 fail:
1039         fdset_free(fds);
1040
1041         if (f)
1042                 fclose(f);
1043
1044         return r;
1045 }
1046
1047 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1048         struct rlimit nl;
1049         int r;
1050
1051         assert(saved_rlimit);
1052
1053         /* Save the original RLIMIT_NOFILE so that we can reset it
1054          * later when transitioning from the initrd to the main
1055          * systemd or suchlike. */
1056         if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0)
1057                 return log_error_errno(errno, "Reading RLIMIT_NOFILE failed: %m");
1058
1059         /* Make sure forked processes get the default kernel setting */
1060         if (!arg_default_rlimit[RLIMIT_NOFILE]) {
1061                 struct rlimit *rl;
1062
1063                 rl = newdup(struct rlimit, saved_rlimit, 1);
1064                 if (!rl)
1065                         return log_oom();
1066
1067                 arg_default_rlimit[RLIMIT_NOFILE] = rl;
1068         }
1069
1070         /* Bump up the resource limit for ourselves substantially */
1071         nl.rlim_cur = nl.rlim_max = 64*1024;
1072         r = setrlimit_closest(RLIMIT_NOFILE, &nl);
1073         if (r < 0)
1074                 return log_error_errno(r, "Setting RLIMIT_NOFILE failed: %m");
1075
1076         return 0;
1077 }
1078
1079 static void test_mtab(void) {
1080
1081         static const char ok[] =
1082                 "/proc/self/mounts\0"
1083                 "/proc/mounts\0"
1084                 "../proc/self/mounts\0"
1085                 "../proc/mounts\0";
1086
1087         _cleanup_free_ char *p = NULL;
1088         int r;
1089
1090         /* Check that /etc/mtab is a symlink to the right place or
1091          * non-existing. But certainly not a file, or a symlink to
1092          * some weird place... */
1093
1094         r = readlink_malloc("/etc/mtab", &p);
1095         if (r == -ENOENT)
1096                 return;
1097         if (r >= 0 && nulstr_contains(ok, p))
1098                 return;
1099
1100         log_warning("/etc/mtab is not a symlink or not pointing to /proc/self/mounts. "
1101                     "This is not supported anymore. "
1102                     "Please make sure to replace this file by a symlink to avoid incorrect or misleading mount(8) output.");
1103 }
1104
1105 static void test_usr(void) {
1106
1107         /* Check that /usr is not a separate fs */
1108
1109         if (dir_is_empty("/usr") <= 0)
1110                 return;
1111
1112         log_warning("/usr appears to be on its own filesystem and is not already mounted. This is not a supported setup. "
1113                     "Some things will probably break (sometimes even silently) in mysterious ways. "
1114                     "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1115 }
1116
1117 static int initialize_join_controllers(void) {
1118         /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
1119          * + "net_prio". We'd like to add "cpuset" to the mix, but
1120          * "cpuset" doesn't really work for groups with no initialized
1121          * attributes. */
1122
1123         arg_join_controllers = new(char**, 3);
1124         if (!arg_join_controllers)
1125                 return -ENOMEM;
1126
1127         arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
1128         arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
1129         arg_join_controllers[2] = NULL;
1130
1131         if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
1132                 free_join_controllers();
1133                 return -ENOMEM;
1134         }
1135
1136         return 0;
1137 }
1138
1139 static int enforce_syscall_archs(Set *archs) {
1140 #ifdef HAVE_SECCOMP
1141         scmp_filter_ctx *seccomp;
1142         Iterator i;
1143         void *id;
1144         int r;
1145
1146         seccomp = seccomp_init(SCMP_ACT_ALLOW);
1147         if (!seccomp)
1148                 return log_oom();
1149
1150         SET_FOREACH(id, arg_syscall_archs, i) {
1151                 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
1152                 if (r == -EEXIST)
1153                         continue;
1154                 if (r < 0) {
1155                         log_error_errno(r, "Failed to add architecture to seccomp: %m");
1156                         goto finish;
1157                 }
1158         }
1159
1160         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1161         if (r < 0) {
1162                 log_error_errno(r, "Failed to unset NO_NEW_PRIVS: %m");
1163                 goto finish;
1164         }
1165
1166         r = seccomp_load(seccomp);
1167         if (r < 0)
1168                 log_error_errno(r, "Failed to add install architecture seccomp: %m");
1169
1170 finish:
1171         seccomp_release(seccomp);
1172         return r;
1173 #else
1174         return 0;
1175 #endif
1176 }
1177
1178 static int status_welcome(void) {
1179         _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1180         int r;
1181
1182         r = parse_env_file("/etc/os-release", NEWLINE,
1183                            "PRETTY_NAME", &pretty_name,
1184                            "ANSI_COLOR", &ansi_color,
1185                            NULL);
1186         if (r == -ENOENT) {
1187                 r = parse_env_file("/usr/lib/os-release", NEWLINE,
1188                                    "PRETTY_NAME", &pretty_name,
1189                                    "ANSI_COLOR", &ansi_color,
1190                                    NULL);
1191         }
1192
1193         if (r < 0 && r != -ENOENT)
1194                 log_warning_errno(r, "Failed to read os-release file: %m");
1195
1196         return status_printf(NULL, false, false,
1197                              "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1198                              isempty(ansi_color) ? "1" : ansi_color,
1199                              isempty(pretty_name) ? "Linux" : pretty_name);
1200 }
1201
1202 static int write_container_id(void) {
1203         const char *c;
1204
1205         c = getenv("container");
1206         if (isempty(c))
1207                 return 0;
1208
1209         return write_string_file("/run/systemd/container", c);
1210 }
1211
1212 int main(int argc, char *argv[]) {
1213         Manager *m = NULL;
1214         int r, retval = EXIT_FAILURE;
1215         usec_t before_startup, after_startup;
1216         char timespan[FORMAT_TIMESPAN_MAX];
1217         FDSet *fds = NULL;
1218         bool reexecute = false;
1219         const char *shutdown_verb = NULL;
1220         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
1221         dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
1222         dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
1223         dual_timestamp security_start_timestamp = { 0ULL, 0ULL };
1224         dual_timestamp security_finish_timestamp = { 0ULL, 0ULL };
1225         static char systemd[] = "systemd";
1226         bool skip_setup = false;
1227         unsigned j;
1228         bool loaded_policy = false;
1229         bool arm_reboot_watchdog = false;
1230         bool queue_default_job = false;
1231         bool empty_etc = false;
1232         char *switch_root_dir = NULL, *switch_root_init = NULL;
1233         static struct rlimit saved_rlimit_nofile = { 0, 0 };
1234         const char *error_message = NULL;
1235
1236 #ifdef HAVE_SYSV_COMPAT
1237         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
1238                 /* This is compatibility support for SysV, where
1239                  * calling init as a user is identical to telinit. */
1240
1241                 errno = -ENOENT;
1242                 execv(SYSTEMCTL_BINARY_PATH, argv);
1243                 log_error_errno(errno, "Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1244                 return 1;
1245         }
1246 #endif
1247
1248         dual_timestamp_from_monotonic(&kernel_timestamp, 0);
1249         dual_timestamp_get(&userspace_timestamp);
1250
1251         /* Determine if this is a reexecution or normal bootup. We do
1252          * the full command line parsing much later, so let's just
1253          * have a quick peek here. */
1254         if (strv_find(argv+1, "--deserialize"))
1255                 skip_setup = true;
1256
1257         /* If we have switched root, do all the special setup
1258          * things */
1259         if (strv_find(argv+1, "--switched-root"))
1260                 skip_setup = false;
1261
1262         /* If we get started via the /sbin/init symlink then we are
1263            called 'init'. After a subsequent reexecution we are then
1264            called 'systemd'. That is confusing, hence let's call us
1265            systemd right-away. */
1266         program_invocation_short_name = systemd;
1267         prctl(PR_SET_NAME, systemd);
1268
1269         saved_argv = argv;
1270         saved_argc = argc;
1271
1272         log_show_color(isatty(STDERR_FILENO) > 0);
1273         log_set_upgrade_syslog_to_journal(true);
1274
1275         /* Disable the umask logic */
1276         if (getpid() == 1)
1277                 umask(0);
1278
1279         if (getpid() == 1 && detect_container(NULL) <= 0) {
1280
1281                 /* Running outside of a container as PID 1 */
1282                 arg_running_as = SYSTEMD_SYSTEM;
1283                 make_null_stdio();
1284                 log_set_target(LOG_TARGET_KMSG);
1285                 log_open();
1286
1287                 if (in_initrd())
1288                         initrd_timestamp = userspace_timestamp;
1289
1290                 if (!skip_setup) {
1291                         mount_setup_early();
1292                         dual_timestamp_get(&security_start_timestamp);
1293                         if (mac_selinux_setup(&loaded_policy) < 0) {
1294                                 error_message = "Failed to load SELinux policy";
1295                                 goto finish;
1296                         } else if (ima_setup() < 0) {
1297                                 error_message = "Failed to load IMA policy";
1298                                 goto finish;
1299                         } else if (mac_smack_setup(&loaded_policy) < 0) {
1300                                 error_message = "Failed to load SMACK policy";
1301                                 goto finish;
1302                         }
1303                         dual_timestamp_get(&security_finish_timestamp);
1304                 }
1305
1306                 if (mac_selinux_init(NULL) < 0) {
1307                         error_message = "Failed to initialize SELinux policy";
1308                         goto finish;
1309                 }
1310
1311                 if (!skip_setup) {
1312                         if (clock_is_localtime() > 0) {
1313                                 int min;
1314
1315                                 /*
1316                                  * The very first call of settimeofday() also does a time warp in the kernel.
1317                                  *
1318                                  * In the rtc-in-local time mode, we set the kernel's timezone, and rely on
1319                                  * external tools to take care of maintaining the RTC and do all adjustments.
1320                                  * This matches the behavior of Windows, which leaves the RTC alone if the
1321                                  * registry tells that the RTC runs in UTC.
1322                                  */
1323                                 r = clock_set_timezone(&min);
1324                                 if (r < 0)
1325                                         log_error_errno(r, "Failed to apply local time delta, ignoring: %m");
1326                                 else
1327                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1328                         } else if (!in_initrd()) {
1329                                 /*
1330                                  * Do a dummy very first call to seal the kernel's time warp magic.
1331                                  *
1332                                  * Do not call this this from inside the initrd. The initrd might not
1333                                  * carry /etc/adjtime with LOCAL, but the real system could be set up
1334                                  * that way. In such case, we need to delay the time-warp or the sealing
1335                                  * until we reach the real system.
1336                                  *
1337                                  * Do no set the kernel's timezone. The concept of local time cannot
1338                                  * be supported reliably, the time will jump or be incorrect at every daylight
1339                                  * saving time change. All kernel local time concepts will be treated
1340                                  * as UTC that way.
1341                                  */
1342                                 clock_reset_timewarp();
1343                         }
1344                 }
1345
1346                 /* Set the default for later on, but don't actually
1347                  * open the logs like this for now. Note that if we
1348                  * are transitioning from the initrd there might still
1349                  * be journal fd open, and we shouldn't attempt
1350                  * opening that before we parsed /proc/cmdline which
1351                  * might redirect output elsewhere. */
1352                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1353
1354         } else if (getpid() == 1) {
1355                 /* Running inside a container, as PID 1 */
1356                 arg_running_as = SYSTEMD_SYSTEM;
1357                 log_set_target(LOG_TARGET_CONSOLE);
1358                 log_close_console(); /* force reopen of /dev/console */
1359                 log_open();
1360
1361                 /* For the later on, see above... */
1362                 log_set_target(LOG_TARGET_JOURNAL);
1363
1364                 /* clear the kernel timestamp,
1365                  * because we are in a container */
1366                 kernel_timestamp.monotonic = 0ULL;
1367                 kernel_timestamp.realtime = 0ULL;
1368
1369         } else {
1370                 /* Running as user instance */
1371                 arg_running_as = SYSTEMD_USER;
1372                 log_set_target(LOG_TARGET_AUTO);
1373                 log_open();
1374
1375                 /* clear the kernel timestamp,
1376                  * because we are not PID 1 */
1377                 kernel_timestamp.monotonic = 0ULL;
1378                 kernel_timestamp.realtime = 0ULL;
1379         }
1380
1381         /* Initialize default unit */
1382         r = set_default_unit(SPECIAL_DEFAULT_TARGET);
1383         if (r < 0) {
1384                 log_emergency_errno(r, "Failed to set default unit %s: %m", SPECIAL_DEFAULT_TARGET);
1385                 error_message = "Failed to set default unit";
1386                 goto finish;
1387         }
1388
1389         r = initialize_join_controllers();
1390         if (r < 0) {
1391                 error_message = "Failed to initalize cgroup controllers";
1392                 goto finish;
1393         }
1394
1395         /* Mount /proc, /sys and friends, so that /proc/cmdline and
1396          * /proc/$PID/fd is available. */
1397         if (getpid() == 1) {
1398
1399                 /* Load the kernel modules early, so that we kdbus.ko is loaded before kdbusfs shall be mounted */
1400                 if (!skip_setup)
1401                         kmod_setup();
1402
1403                 r = mount_setup(loaded_policy);
1404                 if (r < 0) {
1405                         error_message = "Failed to mount API filesystems";
1406                         goto finish;
1407                 }
1408         }
1409
1410         /* Reset all signal handlers. */
1411         assert_se(reset_all_signal_handlers() == 0);
1412
1413         ignore_signals(SIGNALS_IGNORE, -1);
1414
1415         if (parse_config_file() < 0) {
1416                 error_message = "Failed to parse config file";
1417                 goto finish;
1418         }
1419
1420         if (arg_running_as == SYSTEMD_SYSTEM) {
1421                 r = parse_proc_cmdline(parse_proc_cmdline_item);
1422                 if (r < 0)
1423                         log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
1424         }
1425
1426         /* Note that this also parses bits from the kernel command
1427          * line, including "debug". */
1428         log_parse_environment();
1429
1430         if (parse_argv(argc, argv) < 0) {
1431                 error_message = "Failed to parse commandline arguments";
1432                 goto finish;
1433         }
1434
1435         if (arg_action == ACTION_TEST &&
1436             geteuid() == 0) {
1437                 log_error("Don't run test mode as root.");
1438                 goto finish;
1439         }
1440
1441         if (arg_running_as == SYSTEMD_USER &&
1442             arg_action == ACTION_RUN &&
1443             sd_booted() <= 0) {
1444                 log_error("Trying to run as user instance, but the system has not been booted with systemd.");
1445                 goto finish;
1446         }
1447
1448         if (arg_running_as == SYSTEMD_SYSTEM &&
1449             arg_action == ACTION_RUN &&
1450             running_in_chroot() > 0) {
1451                 log_error("Cannot be run in a chroot() environment.");
1452                 goto finish;
1453         }
1454
1455         if (arg_action == ACTION_TEST)
1456                 skip_setup = true;
1457
1458         pager_open_if_enabled();
1459
1460         if (arg_action == ACTION_HELP) {
1461                 retval = help();
1462                 goto finish;
1463         } else if (arg_action == ACTION_VERSION) {
1464                 retval = version();
1465                 goto finish;
1466         } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
1467                 unit_dump_config_items(stdout);
1468                 retval = EXIT_SUCCESS;
1469                 goto finish;
1470         } else if (arg_action == ACTION_DONE) {
1471                 retval = EXIT_SUCCESS;
1472                 goto finish;
1473         }
1474
1475         if (arg_running_as == SYSTEMD_USER &&
1476             !getenv("XDG_RUNTIME_DIR")) {
1477                 log_error("Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
1478                 goto finish;
1479         }
1480
1481         assert_se(arg_action == ACTION_RUN || arg_action == ACTION_TEST);
1482
1483         /* Close logging fds, in order not to confuse fdset below */
1484         log_close();
1485
1486         /* Remember open file descriptors for later deserialization */
1487         r = fdset_new_fill(&fds);
1488         if (r < 0) {
1489                 log_emergency_errno(r, "Failed to allocate fd set: %m");
1490                 error_message = "Failed to allocate fd set";
1491                 goto finish;
1492         } else
1493                 fdset_cloexec(fds, true);
1494
1495         if (arg_serialization)
1496                 assert_se(fdset_remove(fds, fileno(arg_serialization)) >= 0);
1497
1498         if (arg_running_as == SYSTEMD_SYSTEM)
1499                 /* Become a session leader if we aren't one yet. */
1500                 setsid();
1501
1502         /* Move out of the way, so that we won't block unmounts */
1503         assert_se(chdir("/")  == 0);
1504
1505         /* Reset the console, but only if this is really init and we
1506          * are freshly booted */
1507         if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN) {
1508
1509                 /* If we are init, we connect stdin/stdout/stderr to
1510                  * /dev/null and make sure we don't have a controlling
1511                  * tty. */
1512                 release_terminal();
1513
1514                 if (getpid() == 1 && !skip_setup)
1515                         console_setup();
1516         }
1517
1518         /* Open the logging devices, if possible and necessary */
1519         log_open();
1520
1521         if (arg_show_status == _SHOW_STATUS_UNSET)
1522                 arg_show_status = SHOW_STATUS_YES;
1523
1524         /* Make sure we leave a core dump without panicing the
1525          * kernel. */
1526         if (getpid() == 1) {
1527                 install_crash_handler();
1528
1529                 r = mount_cgroup_controllers(arg_join_controllers);
1530                 if (r < 0)
1531                         goto finish;
1532         }
1533
1534         if (arg_running_as == SYSTEMD_SYSTEM) {
1535                 const char *virtualization = NULL;
1536
1537                 log_info(PACKAGE_STRING " running in %ssystem mode. (" SYSTEMD_FEATURES ")",
1538                          arg_action == ACTION_TEST ? "test " : "" );
1539
1540                 detect_virtualization(&virtualization);
1541                 if (virtualization)
1542                         log_info("Detected virtualization '%s'.", virtualization);
1543
1544                 write_container_id();
1545
1546                 log_info("Detected architecture '%s'.", architecture_to_string(uname_architecture()));
1547
1548                 if (in_initrd())
1549                         log_info("Running in initial RAM disk.");
1550
1551                 /* Let's check whether /etc is already populated. We
1552                  * don't actually really check for that, but use
1553                  * /etc/machine-id as flag file. This allows container
1554                  * managers and installers to provision a couple of
1555                  * files already. If the container manager wants to
1556                  * provision the machine ID itself it should pass
1557                  * $container_uuid to PID 1. */
1558
1559                 empty_etc = access("/etc/machine-id", F_OK) < 0;
1560                 if (empty_etc)
1561                         log_info("Running with unpopulated /etc.");
1562         } else {
1563                 _cleanup_free_ char *t;
1564
1565                 t = uid_to_name(getuid());
1566                 log_debug(PACKAGE_STRING " running in %suser mode for user "UID_FMT"/%s. (" SYSTEMD_FEATURES ")",
1567                           arg_action == ACTION_TEST ? " test" : "", getuid(), t);
1568         }
1569
1570         if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
1571                 if (arg_show_status > 0 || plymouth_running())
1572                         status_welcome();
1573
1574                 hostname_setup();
1575                 machine_id_setup(NULL);
1576                 loopback_setup();
1577
1578                 test_mtab();
1579                 test_usr();
1580         }
1581
1582         if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
1583                 watchdog_set_timeout(&arg_runtime_watchdog);
1584
1585         if (arg_timer_slack_nsec != NSEC_INFINITY)
1586                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1587                         log_error_errno(errno, "Failed to adjust timer slack: %m");
1588
1589         if (arg_capability_bounding_set_drop) {
1590                 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
1591                 if (r < 0) {
1592                         log_emergency_errno(r, "Failed to drop capability bounding set of usermode helpers: %m");
1593                         error_message = "Failed to drop capability bounding set of usermode helpers";
1594                         goto finish;
1595                 }
1596                 r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
1597                 if (r < 0) {
1598                         log_emergency_errno(r, "Failed to drop capability bounding set: %m");
1599                         error_message = "Failed to drop capability bounding set";
1600                         goto finish;
1601                 }
1602         }
1603
1604         if (arg_syscall_archs) {
1605                 r = enforce_syscall_archs(arg_syscall_archs);
1606                 if (r < 0) {
1607                         error_message = "Failed to set syscall architectures";
1608                         goto finish;
1609                 }
1610         }
1611
1612         if (arg_running_as == SYSTEMD_USER) {
1613                 /* Become reaper of our children */
1614                 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
1615                         log_warning_errno(errno, "Failed to make us a subreaper: %m");
1616                         if (errno == EINVAL)
1617                                 log_info("Perhaps the kernel version is too old (< 3.4?)");
1618                 }
1619         }
1620
1621         if (arg_running_as == SYSTEMD_SYSTEM) {
1622                 bump_rlimit_nofile(&saved_rlimit_nofile);
1623
1624                 if (empty_etc) {
1625                         r = unit_file_preset_all(UNIT_FILE_SYSTEM, false, NULL, UNIT_FILE_PRESET_FULL, false, NULL, 0);
1626                         if (r < 0)
1627                                 log_warning_errno(r, "Failed to populate /etc with preset unit settings, ignoring: %m");
1628                         else
1629                                 log_info("Populated /etc with preset unit settings.");
1630                 }
1631         }
1632
1633         r = manager_new(arg_running_as, arg_action == ACTION_TEST, &m);
1634         if (r < 0) {
1635                 log_emergency_errno(r, "Failed to allocate manager object: %m");
1636                 error_message = "Failed to allocate manager object";
1637                 goto finish;
1638         }
1639
1640         m->confirm_spawn = arg_confirm_spawn;
1641         m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
1642         m->default_std_output = arg_default_std_output;
1643         m->default_std_error = arg_default_std_error;
1644         m->default_restart_usec = arg_default_restart_usec;
1645         m->default_timeout_start_usec = arg_default_timeout_start_usec;
1646         m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
1647         m->default_start_limit_interval = arg_default_start_limit_interval;
1648         m->default_start_limit_burst = arg_default_start_limit_burst;
1649         m->default_cpu_accounting = arg_default_cpu_accounting;
1650         m->default_blockio_accounting = arg_default_blockio_accounting;
1651         m->default_memory_accounting = arg_default_memory_accounting;
1652         m->runtime_watchdog = arg_runtime_watchdog;
1653         m->shutdown_watchdog = arg_shutdown_watchdog;
1654
1655         m->userspace_timestamp = userspace_timestamp;
1656         m->kernel_timestamp = kernel_timestamp;
1657         m->initrd_timestamp = initrd_timestamp;
1658         m->security_start_timestamp = security_start_timestamp;
1659         m->security_finish_timestamp = security_finish_timestamp;
1660
1661         manager_set_default_rlimits(m, arg_default_rlimit);
1662         manager_environment_add(m, NULL, arg_default_environment);
1663         manager_set_show_status(m, arg_show_status);
1664         manager_set_first_boot(m, empty_etc);
1665
1666         /* Remember whether we should queue the default job */
1667         queue_default_job = !arg_serialization || arg_switched_root;
1668
1669         before_startup = now(CLOCK_MONOTONIC);
1670
1671         r = manager_startup(m, arg_serialization, fds);
1672         if (r < 0)
1673                 log_error_errno(r, "Failed to fully start up daemon: %m");
1674
1675         /* This will close all file descriptors that were opened, but
1676          * not claimed by any unit. */
1677         fdset_free(fds);
1678         fds = NULL;
1679
1680         if (arg_serialization) {
1681                 fclose(arg_serialization);
1682                 arg_serialization = NULL;
1683         }
1684
1685         if (queue_default_job) {
1686                 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1687                 Unit *target = NULL;
1688                 Job *default_unit_job;
1689
1690                 log_debug("Activating default unit: %s", arg_default_unit);
1691
1692                 r = manager_load_unit(m, arg_default_unit, NULL, &error, &target);
1693                 if (r < 0)
1694                         log_error("Failed to load default target: %s", bus_error_message(&error, r));
1695                 else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND)
1696                         log_error_errno(target->load_error, "Failed to load default target: %m");
1697                 else if (target->load_state == UNIT_MASKED)
1698                         log_error("Default target masked.");
1699
1700                 if (!target || target->load_state != UNIT_LOADED) {
1701                         log_info("Trying to load rescue target...");
1702
1703                         r = manager_load_unit(m, SPECIAL_RESCUE_TARGET, NULL, &error, &target);
1704                         if (r < 0) {
1705                                 log_emergency("Failed to load rescue target: %s", bus_error_message(&error, r));
1706                                 error_message = "Failed to load rescue target";
1707                                 goto finish;
1708                         } else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND) {
1709                                 log_emergency_errno(target->load_error, "Failed to load rescue target: %m");
1710                                 error_message = "Failed to load rescue target";
1711                                 goto finish;
1712                         } else if (target->load_state == UNIT_MASKED) {
1713                                 log_emergency("Rescue target masked.");
1714                                 error_message = "Rescue target masked";
1715                                 goto finish;
1716                         }
1717                 }
1718
1719                 assert(target->load_state == UNIT_LOADED);
1720
1721                 if (arg_action == ACTION_TEST) {
1722                         printf("-> By units:\n");
1723                         manager_dump_units(m, stdout, "\t");
1724                 }
1725
1726                 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
1727                 if (r == -EPERM) {
1728                         log_debug("Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1729
1730                         r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
1731                         if (r < 0) {
1732                                 log_emergency("Failed to start default target: %s", bus_error_message(&error, r));
1733                                 error_message = "Failed to start default target";
1734                                 goto finish;
1735                         }
1736                 } else if (r < 0) {
1737                         log_emergency("Failed to isolate default target: %s", bus_error_message(&error, r));
1738                         error_message = "Failed to isolate default target";
1739                         goto finish;
1740                 }
1741
1742                 m->default_unit_job_id = default_unit_job->id;
1743
1744                 after_startup = now(CLOCK_MONOTONIC);
1745                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
1746                          "Loaded units and determined initial transaction in %s.",
1747                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 100 * USEC_PER_MSEC));
1748
1749                 if (arg_action == ACTION_TEST) {
1750                         printf("-> By jobs:\n");
1751                         manager_dump_jobs(m, stdout, "\t");
1752                         retval = EXIT_SUCCESS;
1753                         goto finish;
1754                 }
1755         }
1756
1757         for (;;) {
1758                 r = manager_loop(m);
1759                 if (r < 0) {
1760                         log_emergency_errno(r, "Failed to run main loop: %m");
1761                         error_message = "Failed to run main loop";
1762                         goto finish;
1763                 }
1764
1765                 switch (m->exit_code) {
1766
1767                 case MANAGER_EXIT:
1768                         retval = EXIT_SUCCESS;
1769                         log_debug("Exit.");
1770                         goto finish;
1771
1772                 case MANAGER_RELOAD:
1773                         log_info("Reloading.");
1774                         r = manager_reload(m);
1775                         if (r < 0)
1776                                 log_error_errno(r, "Failed to reload: %m");
1777                         break;
1778
1779                 case MANAGER_REEXECUTE:
1780
1781                         if (prepare_reexecute(m, &arg_serialization, &fds, false) < 0) {
1782                                 error_message = "Failed to prepare for reexection";
1783                                 goto finish;
1784                         }
1785
1786                         reexecute = true;
1787                         log_notice("Reexecuting.");
1788                         goto finish;
1789
1790                 case MANAGER_SWITCH_ROOT:
1791                         /* Steal the switch root parameters */
1792                         switch_root_dir = m->switch_root;
1793                         switch_root_init = m->switch_root_init;
1794                         m->switch_root = m->switch_root_init = NULL;
1795
1796                         if (!switch_root_init)
1797                                 if (prepare_reexecute(m, &arg_serialization, &fds, true) < 0) {
1798                                         error_message = "Failed to prepare for reexection";
1799                                         goto finish;
1800                                 }
1801
1802                         reexecute = true;
1803                         log_notice("Switching root.");
1804                         goto finish;
1805
1806                 case MANAGER_REBOOT:
1807                 case MANAGER_POWEROFF:
1808                 case MANAGER_HALT:
1809                 case MANAGER_KEXEC: {
1810                         static const char * const table[_MANAGER_EXIT_CODE_MAX] = {
1811                                 [MANAGER_REBOOT] = "reboot",
1812                                 [MANAGER_POWEROFF] = "poweroff",
1813                                 [MANAGER_HALT] = "halt",
1814                                 [MANAGER_KEXEC] = "kexec"
1815                         };
1816
1817                         assert_se(shutdown_verb = table[m->exit_code]);
1818                         arm_reboot_watchdog = m->exit_code == MANAGER_REBOOT;
1819
1820                         log_notice("Shutting down.");
1821                         goto finish;
1822                 }
1823
1824                 default:
1825                         assert_not_reached("Unknown exit code.");
1826                 }
1827         }
1828
1829 finish:
1830         pager_close();
1831
1832         if (m)
1833                 arg_shutdown_watchdog = m->shutdown_watchdog;
1834         m = manager_free(m);
1835
1836         for (j = 0; j < ELEMENTSOF(arg_default_rlimit); j++) {
1837                 free(arg_default_rlimit[j]);
1838                 arg_default_rlimit[j] = NULL;
1839         }
1840
1841         free(arg_default_unit);
1842         arg_default_unit = NULL;
1843
1844         free_join_controllers();
1845
1846         strv_free(arg_default_environment);
1847         arg_default_environment = NULL;
1848
1849         set_free(arg_syscall_archs);
1850         arg_syscall_archs = NULL;
1851
1852         mac_selinux_finish();
1853
1854         if (reexecute) {
1855                 const char **args;
1856                 unsigned i, args_size;
1857
1858                 /* Close and disarm the watchdog, so that the new
1859                  * instance can reinitialize it, but doesn't get
1860                  * rebooted while we do that */
1861                 watchdog_close(true);
1862
1863                 /* Reset the RLIMIT_NOFILE to the kernel default, so
1864                  * that the new systemd can pass the kernel default to
1865                  * its child processes */
1866                 if (saved_rlimit_nofile.rlim_cur > 0)
1867                         setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
1868
1869                 if (switch_root_dir) {
1870                         /* Kill all remaining processes from the
1871                          * initrd, but don't wait for them, so that we
1872                          * can handle the SIGCHLD for them after
1873                          * deserializing. */
1874                         broadcast_signal(SIGTERM, false, true);
1875
1876                         /* And switch root with MS_MOVE, because we remove the old directory afterwards and detach it. */
1877                         r = switch_root(switch_root_dir, "/mnt", true, MS_MOVE);
1878                         if (r < 0)
1879                                 log_error_errno(r, "Failed to switch root, trying to continue: %m");
1880                 }
1881
1882                 args_size = MAX(6, argc+1);
1883                 args = newa(const char*, args_size);
1884
1885                 if (!switch_root_init) {
1886                         char sfd[16];
1887
1888                         /* First try to spawn ourselves with the right
1889                          * path, and with full serialization. We do
1890                          * this only if the user didn't specify an
1891                          * explicit init to spawn. */
1892
1893                         assert(arg_serialization);
1894                         assert(fds);
1895
1896                         snprintf(sfd, sizeof(sfd), "%i", fileno(arg_serialization));
1897                         char_array_0(sfd);
1898
1899                         i = 0;
1900                         args[i++] = SYSTEMD_BINARY_PATH;
1901                         if (switch_root_dir)
1902                                 args[i++] = "--switched-root";
1903                         args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
1904                         args[i++] = "--deserialize";
1905                         args[i++] = sfd;
1906                         args[i++] = NULL;
1907
1908                         /* do not pass along the environment we inherit from the kernel or initrd */
1909                         if (switch_root_dir)
1910                                 clearenv();
1911
1912                         assert(i <= args_size);
1913                         execv(args[0], (char* const*) args);
1914                 }
1915
1916                 /* Try the fallback, if there is any, without any
1917                  * serialization. We pass the original argv[] and
1918                  * envp[]. (Well, modulo the ordering changes due to
1919                  * getopt() in argv[], and some cleanups in envp[],
1920                  * but let's hope that doesn't matter.) */
1921
1922                 if (arg_serialization) {
1923                         fclose(arg_serialization);
1924                         arg_serialization = NULL;
1925                 }
1926
1927                 if (fds) {
1928                         fdset_free(fds);
1929                         fds = NULL;
1930                 }
1931
1932                 /* Reopen the console */
1933                 make_console_stdio();
1934
1935                 for (j = 1, i = 1; j < (unsigned) argc; j++)
1936                         args[i++] = argv[j];
1937                 args[i++] = NULL;
1938                 assert(i <= args_size);
1939
1940                 /* Reenable any blocked signals, especially important
1941                  * if we switch from initial ramdisk to init=... */
1942                 reset_all_signal_handlers();
1943                 reset_signal_mask();
1944
1945                 if (switch_root_init) {
1946                         args[0] = switch_root_init;
1947                         execv(args[0], (char* const*) args);
1948                         log_warning_errno(errno, "Failed to execute configured init, trying fallback: %m");
1949                 }
1950
1951                 args[0] = "/sbin/init";
1952                 execv(args[0], (char* const*) args);
1953
1954                 if (errno == ENOENT) {
1955                         log_warning("No /sbin/init, trying fallback");
1956
1957                         args[0] = "/bin/sh";
1958                         args[1] = NULL;
1959                         execv(args[0], (char* const*) args);
1960                         log_error_errno(errno, "Failed to execute /bin/sh, giving up: %m");
1961                 } else
1962                         log_warning_errno(errno, "Failed to execute /sbin/init, giving up: %m");
1963         }
1964
1965         if (arg_serialization) {
1966                 fclose(arg_serialization);
1967                 arg_serialization = NULL;
1968         }
1969
1970         if (fds) {
1971                 fdset_free(fds);
1972                 fds = NULL;
1973         }
1974
1975 #ifdef HAVE_VALGRIND_VALGRIND_H
1976         /* If we are PID 1 and running under valgrind, then let's exit
1977          * here explicitly. valgrind will only generate nice output on
1978          * exit(), not on exec(), hence let's do the former not the
1979          * latter here. */
1980         if (getpid() == 1 && RUNNING_ON_VALGRIND)
1981                 return 0;
1982 #endif
1983
1984         if (shutdown_verb) {
1985                 char log_level[DECIMAL_STR_MAX(int) + 1];
1986                 const char* command_line[9] = {
1987                         SYSTEMD_SHUTDOWN_BINARY_PATH,
1988                         shutdown_verb,
1989                         "--log-level", log_level,
1990                         "--log-target",
1991                 };
1992                 unsigned pos = 5;
1993                 _cleanup_strv_free_ char **env_block = NULL;
1994
1995                 assert(command_line[pos] == NULL);
1996                 env_block = strv_copy(environ);
1997
1998                 snprintf(log_level, sizeof(log_level), "%d", log_get_max_level());
1999
2000                 switch (log_get_target()) {
2001                 case LOG_TARGET_KMSG:
2002                 case LOG_TARGET_JOURNAL_OR_KMSG:
2003                 case LOG_TARGET_SYSLOG_OR_KMSG:
2004                         command_line[pos++] = "kmsg";
2005                         break;
2006
2007                 case LOG_TARGET_CONSOLE:
2008                 default:
2009                         command_line[pos++] = "console";
2010                         break;
2011                 };
2012
2013                 if (log_get_show_color())
2014                         command_line[pos++] = "--log-color";
2015
2016                 if (log_get_show_location())
2017                         command_line[pos++] = "--log-location";
2018
2019                 assert(pos < ELEMENTSOF(command_line));
2020
2021                 if (arm_reboot_watchdog && arg_shutdown_watchdog > 0) {
2022                         char *e;
2023
2024                         /* If we reboot let's set the shutdown
2025                          * watchdog and tell the shutdown binary to
2026                          * repeatedly ping it */
2027                         watchdog_set_timeout(&arg_shutdown_watchdog);
2028                         watchdog_close(false);
2029
2030                         /* Tell the binary how often to ping, ignore failure */
2031                         if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, arg_shutdown_watchdog) > 0)
2032                                 strv_push(&env_block, e);
2033                 } else
2034                         watchdog_close(true);
2035
2036                 /* Avoid the creation of new processes forked by the
2037                  * kernel; at this point, we will not listen to the
2038                  * signals anyway */
2039                 if (detect_container(NULL) <= 0)
2040                         cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
2041
2042                 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
2043                 log_error_errno(errno, "Failed to execute shutdown binary, %s: %m",
2044                           getpid() == 1 ? "freezing" : "quitting");
2045         }
2046
2047         if (getpid() == 1) {
2048                 if (error_message)
2049                         manager_status_printf(NULL, STATUS_TYPE_EMERGENCY,
2050                                               ANSI_HIGHLIGHT_RED_ON "!!!!!!" ANSI_HIGHLIGHT_OFF,
2051                                               "%s, freezing.", error_message);
2052                 freeze();
2053         }
2054
2055         return retval;
2056 }