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