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