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 "hwclock.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                 if (arg_show_status == _SHOW_STATUS_UNSET)
 417                         arg_show_status = SHOW_STATUS_AUTO;
 418
 419         } else if (streq(key, "debug") && !value) {
 420
 421                 /* Log to kmsg, the journal socket will fill up before the
 422                  * journal is started and tools running during that time
 423                  * will block with every log message for for 60 seconds,
 424                  * before they give up. */
 425                 log_set_max_level(LOG_DEBUG);
 426                 log_set_target(detect_container(NULL) > 0 ? LOG_TARGET_CONSOLE : LOG_TARGET_KMSG);
 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         char *p;
1160
1161         /* Check that /etc/mtab is a symlink */
1162
1163         if (readlink_malloc("/etc/mtab", &p) >= 0) {
1164                 bool b;
1165
1166                 b = streq(p, "/proc/self/mounts") || streq(p, "/proc/mounts");
1167                 free(p);
1168
1169                 if (b)
1170                         return;
1171         }
1172
1173         log_warning("/etc/mtab is not a symlink or not pointing to /proc/self/mounts. "
1174                     "This is not supported anymore. "
1175                     "Please make sure to replace this file by a symlink to avoid incorrect or misleading mount(8) output.");
1176 }
1177
1178 static void test_usr(void) {
1179
1180         /* Check that /usr is not a separate fs */
1181
1182         if (dir_is_empty("/usr") <= 0)
1183                 return;
1184
1185         log_warning("/usr appears to be on its own filesytem and is not already mounted. This is not a supported setup. "
1186                     "Some things will probably break (sometimes even silently) in mysterious ways. "
1187                     "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1188 }
1189
1190 static void test_cgroups(void) {
1191
1192         if (access("/proc/cgroups", F_OK) >= 0)
1193                 return;
1194
1195         log_warning("CONFIG_CGROUPS was not set when your kernel was compiled. "
1196                     "Systems without control groups are not supported. "
1197                     "We will now sleep for 10s, and then continue boot-up. "
1198                     "Expect breakage and please do not file bugs. "
1199                     "Instead fix your kernel and enable CONFIG_CGROUPS. "
1200                     "Consult http://0pointer.de/blog/projects/cgroups-vs-cgroups.html for more information.");
1201
1202         sleep(10);
1203 }
1204
1205 static int initialize_join_controllers(void) {
1206         /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
1207          * + "net_prio". We'd like to add "cpuset" to the mix, but
1208          * "cpuset" does't really work for groups with no initialized
1209          * attributes. */
1210
1211         arg_join_controllers = new(char**, 3);
1212         if (!arg_join_controllers)
1213                 return -ENOMEM;
1214
1215         arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
1216         arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
1217         arg_join_controllers[2] = NULL;
1218
1219         if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
1220                 free_join_controllers();
1221                 return -ENOMEM;
1222         }
1223
1224         return 0;
1225 }
1226
1227 static int enforce_syscall_archs(Set *archs) {
1228 #ifdef HAVE_SECCOMP
1229         scmp_filter_ctx *seccomp;
1230         Iterator i;
1231         void *id;
1232         int r;
1233
1234         seccomp = seccomp_init(SCMP_ACT_ALLOW);
1235         if (!seccomp)
1236                 return log_oom();
1237
1238         SET_FOREACH(id, arg_syscall_archs, i) {
1239                 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
1240                 if (r == -EEXIST)
1241                         continue;
1242                 if (r < 0) {
1243                         log_error("Failed to add architecture to seccomp: %s", strerror(-r));
1244                         goto finish;
1245                 }
1246         }
1247
1248         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1249         if (r < 0) {
1250                 log_error("Failed to unset NO_NEW_PRIVS: %s", strerror(-r));
1251                 goto finish;
1252         }
1253
1254         r = seccomp_load(seccomp);
1255         if (r < 0)
1256                 log_error("Failed to add install architecture seccomp: %s", strerror(-r));
1257
1258 finish:
1259         seccomp_release(seccomp);
1260         return r;
1261 #else
1262         return 0;
1263 #endif
1264 }
1265
1266 static int status_welcome(void) {
1267         _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1268         int r;
1269
1270         r = parse_env_file("/etc/os-release", NEWLINE,
1271                            "PRETTY_NAME", &pretty_name,
1272                            "ANSI_COLOR", &ansi_color,
1273                            NULL);
1274
1275         if (r < 0 && r != -ENOENT)
1276                 log_warning("Failed to read /etc/os-release: %s", strerror(-r));
1277
1278         return status_printf(NULL, false, false,
1279                              "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1280                              isempty(ansi_color) ? "1" : ansi_color,
1281                              isempty(pretty_name) ? "Linux" : pretty_name);
1282 }
1283
1284 int main(int argc, char *argv[]) {
1285         Manager *m = NULL;
1286         int r, retval = EXIT_FAILURE;
1287         usec_t before_startup, after_startup;
1288         char timespan[FORMAT_TIMESPAN_MAX];
1289         FDSet *fds = NULL;
1290         bool reexecute = false;
1291         const char *shutdown_verb = NULL;
1292         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
1293         dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
1294         dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
1295         dual_timestamp security_start_timestamp = { 0ULL, 0ULL };
1296         dual_timestamp security_finish_timestamp = { 0ULL, 0ULL };
1297         static char systemd[] = "systemd";
1298         bool skip_setup = false;
1299         unsigned j;
1300         bool loaded_policy = false;
1301         bool arm_reboot_watchdog = false;
1302         bool queue_default_job = false;
1303         char *switch_root_dir = NULL, *switch_root_init = NULL;
1304         static struct rlimit saved_rlimit_nofile = { 0, 0 };
1305
1306 #ifdef HAVE_SYSV_COMPAT
1307         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
1308                 /* This is compatibility support for SysV, where
1309                  * calling init as a user is identical to telinit. */
1310
1311                 errno = -ENOENT;
1312                 execv(SYSTEMCTL_BINARY_PATH, argv);
1313                 log_error("Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1314                 return 1;
1315         }
1316 #endif
1317
1318         dual_timestamp_from_monotonic(&kernel_timestamp, 0);
1319         dual_timestamp_get(&userspace_timestamp);
1320
1321         /* Determine if this is a reexecution or normal bootup. We do
1322          * the full command line parsing much later, so let's just
1323          * have a quick peek here. */
1324         if (strv_find(argv+1, "--deserialize"))
1325                 skip_setup = true;
1326
1327         /* If we have switched root, do all the special setup
1328          * things */
1329         if (strv_find(argv+1, "--switched-root"))
1330                 skip_setup = false;
1331
1332         /* If we get started via the /sbin/init symlink then we are
1333            called 'init'. After a subsequent reexecution we are then
1334            called 'systemd'. That is confusing, hence let's call us
1335            systemd right-away. */
1336         program_invocation_short_name = systemd;
1337         prctl(PR_SET_NAME, systemd);
1338
1339         saved_argv = argv;
1340         saved_argc = argc;
1341
1342         log_show_color(isatty(STDERR_FILENO) > 0);
1343
1344         /* Disable the umask logic */
1345         if (getpid() == 1)
1346                 umask(0);
1347
1348         if (getpid() == 1 && detect_container(NULL) <= 0) {
1349
1350                 /* Running outside of a container as PID 1 */
1351                 arg_running_as = SYSTEMD_SYSTEM;
1352                 make_null_stdio();
1353                 log_set_target(LOG_TARGET_KMSG);
1354                 log_open();
1355
1356                 if (in_initrd())
1357                         initrd_timestamp = userspace_timestamp;
1358
1359                 if (!skip_setup) {
1360                         mount_setup_early();
1361                         dual_timestamp_get(&security_start_timestamp);
1362                         if (selinux_setup(&loaded_policy) < 0)
1363                                 goto finish;
1364                         if (ima_setup() < 0)
1365                                 goto finish;
1366                         if (smack_setup(&loaded_policy) < 0)
1367                                 goto finish;
1368                         dual_timestamp_get(&security_finish_timestamp);
1369                 }
1370
1371                 if (label_init(NULL) < 0)
1372                         goto finish;
1373
1374                 if (!skip_setup) {
1375                         if (hwclock_is_localtime() > 0) {
1376                                 int min;
1377
1378                                 /* The first-time call to settimeofday() does a time warp in the kernel */
1379                                 r = hwclock_set_timezone(&min);
1380                                 if (r < 0)
1381                                         log_error("Failed to apply local time delta, ignoring: %s", strerror(-r));
1382                                 else
1383                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1384                         } else if (!in_initrd()) {
1385                                 /*
1386                                  * Do dummy first-time call to seal the kernel's time warp magic
1387                                  *
1388                                  * Do not call this this from inside the initrd. The initrd might not
1389                                  * carry /etc/adjtime with LOCAL, but the real system could be set up
1390                                  * that way. In such case, we need to delay the time-warp or the sealing
1391                                  * until we reach the real system.
1392                                  */
1393                                 hwclock_reset_timezone();
1394
1395                                 /* Tell the kernel our timezone */
1396                                 r = hwclock_set_timezone(NULL);
1397                                 if (r < 0)
1398                                         log_error("Failed to set the kernel's timezone, ignoring: %s", strerror(-r));
1399                         }
1400                 }
1401
1402                 /* Set the default for later on, but don't actually
1403                  * open the logs like this for now. Note that if we
1404                  * are transitioning from the initrd there might still
1405                  * be journal fd open, and we shouldn't attempt
1406                  * opening that before we parsed /proc/cmdline which
1407                  * might redirect output elsewhere. */
1408                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1409
1410         } else if (getpid() == 1) {
1411                 /* Running inside a container, as PID 1 */
1412                 arg_running_as = SYSTEMD_SYSTEM;
1413                 log_set_target(LOG_TARGET_CONSOLE);
1414                 log_close_console(); /* force reopen of /dev/console */
1415                 log_open();
1416
1417                 /* For the later on, see above... */
1418                 log_set_target(LOG_TARGET_JOURNAL);
1419
1420                 /* clear the kernel timestamp,
1421                  * because we are in a container */
1422                 kernel_timestamp.monotonic = 0ULL;
1423                 kernel_timestamp.realtime = 0ULL;
1424
1425         } else {
1426                 /* Running as user instance */
1427                 arg_running_as = SYSTEMD_USER;
1428                 log_set_target(LOG_TARGET_AUTO);
1429                 log_open();
1430
1431                 /* clear the kernel timestamp,
1432                  * because we are not PID 1 */
1433                 kernel_timestamp.monotonic = 0ULL;
1434                 kernel_timestamp.realtime = 0ULL;
1435         }
1436
1437         /* Initialize default unit */
1438         r = set_default_unit(SPECIAL_DEFAULT_TARGET);
1439         if (r < 0) {
1440                 log_error("Failed to set default unit %s: %s", SPECIAL_DEFAULT_TARGET, strerror(-r));
1441                 goto finish;
1442         }
1443
1444         r = initialize_join_controllers();
1445         if (r < 0)
1446                 goto finish;
1447
1448         /* Mount /proc, /sys and friends, so that /proc/cmdline and
1449          * /proc/$PID/fd is available. */
1450         if (getpid() == 1) {
1451                 r = mount_setup(loaded_policy);
1452                 if (r < 0)
1453                         goto finish;
1454         }
1455
1456         /* Reset all signal handlers. */
1457         assert_se(reset_all_signal_handlers() == 0);
1458
1459         ignore_signals(SIGNALS_IGNORE, -1);
1460
1461         if (parse_config_file() < 0)
1462                 goto finish;
1463
1464         if (arg_running_as == SYSTEMD_SYSTEM)
1465                 if (parse_proc_cmdline(parse_proc_cmdline_item) < 0)
1466                         goto finish;
1467
1468         log_parse_environment();
1469
1470         if (parse_argv(argc, argv) < 0)
1471                 goto finish;
1472
1473         if (arg_action == ACTION_TEST &&
1474             geteuid() == 0) {
1475                 log_error("Don't run test mode as root.");
1476                 goto finish;
1477         }
1478
1479         if (arg_running_as == SYSTEMD_USER &&
1480             arg_action == ACTION_RUN &&
1481             sd_booted() <= 0) {
1482                 log_error("Trying to run as user instance, but the system has not been booted with systemd.");
1483                 goto finish;
1484         }
1485
1486         if (arg_running_as == SYSTEMD_SYSTEM &&
1487             arg_action == ACTION_RUN &&
1488             running_in_chroot() > 0) {
1489                 log_error("Cannot be run in a chroot() environment.");
1490                 goto finish;
1491         }
1492
1493         if (arg_action == ACTION_HELP) {
1494                 retval = help();
1495                 goto finish;
1496         } else if (arg_action == ACTION_VERSION) {
1497                 retval = version();
1498                 goto finish;
1499         } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
1500                 unit_dump_config_items(stdout);
1501                 retval = EXIT_SUCCESS;
1502                 goto finish;
1503         } else if (arg_action == ACTION_DONE) {
1504                 retval = EXIT_SUCCESS;
1505                 goto finish;
1506         }
1507
1508         if (arg_running_as == SYSTEMD_USER &&
1509             !getenv("XDG_RUNTIME_DIR")) {
1510                 log_error("Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
1511                 goto finish;
1512         }
1513
1514         assert_se(arg_action == ACTION_RUN || arg_action == ACTION_TEST);
1515
1516         /* Close logging fds, in order not to confuse fdset below */
1517         log_close();
1518
1519         /* Remember open file descriptors for later deserialization */
1520         r = fdset_new_fill(&fds);
1521         if (r < 0) {
1522                 log_error("Failed to allocate fd set: %s", strerror(-r));
1523                 goto finish;
1524         } else
1525                 fdset_cloexec(fds, true);
1526
1527         if (arg_serialization)
1528                 assert_se(fdset_remove(fds, fileno(arg_serialization)) >= 0);
1529
1530         if (arg_running_as == SYSTEMD_SYSTEM)
1531                 /* Become a session leader if we aren't one yet. */
1532                 setsid();
1533
1534         /* Move out of the way, so that we won't block unmounts */
1535         assert_se(chdir("/")  == 0);
1536
1537         /* Reset the console, but only if this is really init and we
1538          * are freshly booted */
1539         if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN)
1540                 console_setup(getpid() == 1 && !skip_setup);
1541
1542         /* Open the logging devices, if possible and necessary */
1543         log_open();
1544
1545         if (arg_show_status == _SHOW_STATUS_UNSET)
1546                 arg_show_status = SHOW_STATUS_YES;
1547
1548         /* Make sure we leave a core dump without panicing the
1549          * kernel. */
1550         if (getpid() == 1) {
1551                 install_crash_handler();
1552
1553                 r = mount_cgroup_controllers(arg_join_controllers);
1554                 if (r < 0)
1555                         goto finish;
1556         }
1557
1558         if (arg_running_as == SYSTEMD_SYSTEM) {
1559                 const char *virtualization = NULL;
1560
1561                 log_info(PACKAGE_STRING " running in system mode. (" SYSTEMD_FEATURES ")");
1562
1563                 detect_virtualization(&virtualization);
1564                 if (virtualization)
1565                         log_info("Detected virtualization '%s'.", virtualization);
1566
1567                 log_info("Detected architecture '%s'.", architecture_to_string(uname_architecture()));
1568
1569                 if (in_initrd())
1570                         log_info("Running in initial RAM disk.");
1571
1572         } else {
1573                 _cleanup_free_ char *t = uid_to_name(getuid());
1574                 log_debug(PACKAGE_STRING " running in user mode for user "PID_FMT"/%s. (" SYSTEMD_FEATURES ")",
1575                           getuid(), t);
1576         }
1577
1578         if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
1579                 if (arg_show_status > 0 || plymouth_running())
1580                         status_welcome();
1581
1582 #ifdef HAVE_KMOD
1583                 if (detect_container(NULL) <= 0)
1584                         kmod_setup();
1585 #endif
1586                 hostname_setup();
1587                 machine_id_setup("");
1588                 loopback_setup();
1589
1590                 test_mtab();
1591                 test_usr();
1592                 test_cgroups();
1593         }
1594
1595         if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
1596                 watchdog_set_timeout(&arg_runtime_watchdog);
1597
1598         if (arg_timer_slack_nsec != (nsec_t) -1)
1599                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1600                         log_error("Failed to adjust timer slack: %m");
1601
1602         if (arg_capability_bounding_set_drop) {
1603                 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
1604                 if (r < 0) {
1605                         log_error("Failed to drop capability bounding set of usermode helpers: %s", strerror(-r));
1606                         goto finish;
1607                 }
1608                 r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
1609                 if (r < 0) {
1610                         log_error("Failed to drop capability bounding set: %s", strerror(-r));
1611                         goto finish;
1612                 }
1613         }
1614
1615         if (arg_syscall_archs) {
1616                 r = enforce_syscall_archs(arg_syscall_archs);
1617                 if (r < 0)
1618                         goto finish;
1619         }
1620
1621         if (arg_running_as == SYSTEMD_USER) {
1622                 /* Become reaper of our children */
1623                 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
1624                         log_warning("Failed to make us a subreaper: %m");
1625                         if (errno == EINVAL)
1626                                 log_info("Perhaps the kernel version is too old (< 3.4?)");
1627                 }
1628         }
1629
1630         if (arg_running_as == SYSTEMD_SYSTEM)
1631                 bump_rlimit_nofile(&saved_rlimit_nofile);
1632
1633         r = manager_new(arg_running_as, &m);
1634         if (r < 0) {
1635                 log_error("Failed to allocate manager object: %s", strerror(-r));
1636                 goto finish;
1637         }
1638
1639         m->confirm_spawn = arg_confirm_spawn;
1640         m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
1641         m->default_std_output = arg_default_std_output;
1642         m->default_std_error = arg_default_std_error;
1643         m->default_restart_usec = arg_default_restart_usec;
1644         m->default_timeout_start_usec = arg_default_timeout_start_usec;
1645         m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
1646         m->default_start_limit_interval = arg_default_start_limit_interval;
1647         m->default_start_limit_burst = arg_default_start_limit_burst;
1648         m->default_cpu_accounting = arg_default_cpu_accounting;
1649         m->default_blockio_accounting = arg_default_blockio_accounting;
1650         m->default_memory_accounting = arg_default_memory_accounting;
1651         m->runtime_watchdog = arg_runtime_watchdog;
1652         m->shutdown_watchdog = arg_shutdown_watchdog;
1653         m->userspace_timestamp = userspace_timestamp;
1654         m->kernel_timestamp = kernel_timestamp;
1655         m->initrd_timestamp = initrd_timestamp;
1656         m->security_start_timestamp = security_start_timestamp;
1657         m->security_finish_timestamp = security_finish_timestamp;
1658
1659         manager_set_default_rlimits(m, arg_default_rlimit);
1660         manager_environment_add(m, NULL, arg_default_environment);
1661         manager_set_show_status(m, arg_show_status);
1662
1663         /* Remember whether we should queue the default job */
1664         queue_default_job = !arg_serialization || arg_switched_root;
1665
1666         before_startup = now(CLOCK_MONOTONIC);
1667
1668         r = manager_startup(m, arg_serialization, fds);
1669         if (r < 0)
1670                 log_error("Failed to fully start up daemon: %s", strerror(-r));
1671
1672         /* This will close all file descriptors that were opened, but
1673          * not claimed by any unit. */
1674         fdset_free(fds);
1675         fds = NULL;
1676
1677         if (arg_serialization) {
1678                 fclose(arg_serialization);
1679                 arg_serialization = NULL;
1680         }
1681
1682         if (queue_default_job) {
1683                 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1684                 Unit *target = NULL;
1685                 Job *default_unit_job;
1686
1687                 log_debug("Activating default unit: %s", arg_default_unit);
1688
1689                 r = manager_load_unit(m, arg_default_unit, NULL, &error, &target);
1690                 if (r < 0)
1691                         log_error("Failed to load default target: %s", bus_error_message(&error, r));
1692                 else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND)
1693                         log_error("Failed to load default target: %s", strerror(-target->load_error));
1694                 else if (target->load_state == UNIT_MASKED)
1695                         log_error("Default target masked.");
1696
1697                 if (!target || target->load_state != UNIT_LOADED) {
1698                         log_info("Trying to load rescue target...");
1699
1700                         r = manager_load_unit(m, SPECIAL_RESCUE_TARGET, NULL, &error, &target);
1701                         if (r < 0) {
1702                                 log_error("Failed to load rescue target: %s", bus_error_message(&error, r));
1703                                 goto finish;
1704                         } else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND) {
1705                                 log_error("Failed to load rescue target: %s", strerror(-target->load_error));
1706                                 goto finish;
1707                         } else if (target->load_state == UNIT_MASKED) {
1708                                 log_error("Rescue target masked.");
1709                                 goto finish;
1710                         }
1711                 }
1712
1713                 assert(target->load_state == UNIT_LOADED);
1714
1715                 if (arg_action == ACTION_TEST) {
1716                         printf("-> By units:\n");
1717                         manager_dump_units(m, stdout, "\t");
1718                 }
1719
1720                 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
1721                 if (r == -EPERM) {
1722                         log_debug("Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1723
1724                         r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
1725                         if (r < 0) {
1726                                 log_error("Failed to start default target: %s", bus_error_message(&error, r));
1727                                 goto finish;
1728                         }
1729                 } else if (r < 0) {
1730                         log_error("Failed to isolate default target: %s", bus_error_message(&error, r));
1731                         goto finish;
1732                 }
1733
1734                 m->default_unit_job_id = default_unit_job->id;
1735
1736                 after_startup = now(CLOCK_MONOTONIC);
1737                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
1738                          "Loaded units and determined initial transaction in %s.",
1739                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 0));
1740
1741                 if (arg_action == ACTION_TEST) {
1742                         printf("-> By jobs:\n");
1743                         manager_dump_jobs(m, stdout, "\t");
1744                         retval = EXIT_SUCCESS;
1745                         goto finish;
1746                 }
1747         }
1748
1749         for (;;) {
1750                 r = manager_loop(m);
1751                 if (r < 0) {
1752                         log_error("Failed to run mainloop: %s", strerror(-r));
1753                         goto finish;
1754                 }
1755
1756                 switch (m->exit_code) {
1757
1758                 case MANAGER_EXIT:
1759                         retval = EXIT_SUCCESS;
1760                         log_debug("Exit.");
1761                         goto finish;
1762
1763                 case MANAGER_RELOAD:
1764                         log_info("Reloading.");
1765                         r = manager_reload(m);
1766                         if (r < 0)
1767                                 log_error("Failed to reload: %s", strerror(-r));
1768                         break;
1769
1770                 case MANAGER_REEXECUTE:
1771
1772                         if (prepare_reexecute(m, &arg_serialization, &fds, false) < 0)
1773                                 goto finish;
1774
1775                         reexecute = true;
1776                         log_notice("Reexecuting.");
1777                         goto finish;
1778
1779                 case MANAGER_SWITCH_ROOT:
1780                         /* Steal the switch root parameters */
1781                         switch_root_dir = m->switch_root;
1782                         switch_root_init = m->switch_root_init;
1783                         m->switch_root = m->switch_root_init = NULL;
1784
1785                         if (!switch_root_init)
1786                                 if (prepare_reexecute(m, &arg_serialization, &fds, true) < 0)
1787                                         goto finish;
1788
1789                         reexecute = true;
1790                         log_notice("Switching root.");
1791                         goto finish;
1792
1793                 case MANAGER_REBOOT:
1794                 case MANAGER_POWEROFF:
1795                 case MANAGER_HALT:
1796                 case MANAGER_KEXEC: {
1797                         static const char * const table[_MANAGER_EXIT_CODE_MAX] = {
1798                                 [MANAGER_REBOOT] = "reboot",
1799                                 [MANAGER_POWEROFF] = "poweroff",
1800                                 [MANAGER_HALT] = "halt",
1801                                 [MANAGER_KEXEC] = "kexec"
1802                         };
1803
1804                         assert_se(shutdown_verb = table[m->exit_code]);
1805                         arm_reboot_watchdog = m->exit_code == MANAGER_REBOOT;
1806
1807                         log_notice("Shutting down.");
1808                         goto finish;
1809                 }
1810
1811                 default:
1812                         assert_not_reached("Unknown exit code.");
1813                 }
1814         }
1815
1816 finish:
1817         if (m) {
1818                 manager_free(m);
1819                 m = NULL;
1820         }
1821
1822         for (j = 0; j < ELEMENTSOF(arg_default_rlimit); j++) {
1823                 free(arg_default_rlimit[j]);
1824                 arg_default_rlimit[j] = NULL;
1825         }
1826
1827         free(arg_default_unit);
1828         arg_default_unit = NULL;
1829
1830         free_join_controllers();
1831
1832         strv_free(arg_default_environment);
1833         arg_default_environment = NULL;
1834
1835         set_free(arg_syscall_archs);
1836         arg_syscall_archs = NULL;
1837
1838         label_finish();
1839
1840         if (reexecute) {
1841                 const char **args;
1842                 unsigned i, args_size;
1843
1844                 /* Close and disarm the watchdog, so that the new
1845                  * instance can reinitialize it, but doesn't get
1846                  * rebooted while we do that */
1847                 watchdog_close(true);
1848
1849                 /* Reset the RLIMIT_NOFILE to the kernel default, so
1850                  * that the new systemd can pass the kernel default to
1851                  * its child processes */
1852                 if (saved_rlimit_nofile.rlim_cur > 0)
1853                         setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
1854
1855                 if (switch_root_dir) {
1856                         /* Kill all remaining processes from the
1857                          * initrd, but don't wait for them, so that we
1858                          * can handle the SIGCHLD for them after
1859                          * deserializing. */
1860                         broadcast_signal(SIGTERM, false, true);
1861
1862                         /* And switch root */
1863                         r = switch_root(switch_root_dir);
1864                         if (r < 0)
1865                                 log_error("Failed to switch root, ignoring: %s", strerror(-r));
1866                 }
1867
1868                 args_size = MAX(6, argc+1);
1869                 args = newa(const char*, args_size);
1870
1871                 if (!switch_root_init) {
1872                         char sfd[16];
1873
1874                         /* First try to spawn ourselves with the right
1875                          * path, and with full serialization. We do
1876                          * this only if the user didn't specify an
1877                          * explicit init to spawn. */
1878
1879                         assert(arg_serialization);
1880                         assert(fds);
1881
1882                         snprintf(sfd, sizeof(sfd), "%i", fileno(arg_serialization));
1883                         char_array_0(sfd);
1884
1885                         i = 0;
1886                         args[i++] = SYSTEMD_BINARY_PATH;
1887                         if (switch_root_dir)
1888                                 args[i++] = "--switched-root";
1889                         args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
1890                         args[i++] = "--deserialize";
1891                         args[i++] = sfd;
1892                         args[i++] = NULL;
1893
1894                         /* do not pass along the environment we inherit from the kernel or initrd */
1895                         if (switch_root_dir)
1896                                 clearenv();
1897
1898                         assert(i <= args_size);
1899                         execv(args[0], (char* const*) args);
1900                 }
1901
1902                 /* Try the fallback, if there is any, without any
1903                  * serialization. We pass the original argv[] and
1904                  * envp[]. (Well, modulo the ordering changes due to
1905                  * getopt() in argv[], and some cleanups in envp[],
1906                  * but let's hope that doesn't matter.) */
1907
1908                 if (arg_serialization) {
1909                         fclose(arg_serialization);
1910                         arg_serialization = NULL;
1911                 }
1912
1913                 if (fds) {
1914                         fdset_free(fds);
1915                         fds = NULL;
1916                 }
1917
1918                 /* Reopen the console */
1919                 make_console_stdio();
1920
1921                 for (j = 1, i = 1; j < (unsigned) argc; j++)
1922                         args[i++] = argv[j];
1923                 args[i++] = NULL;
1924                 assert(i <= args_size);
1925
1926                 if (switch_root_init) {
1927                         args[0] = switch_root_init;
1928                         execv(args[0], (char* const*) args);
1929                         log_warning("Failed to execute configured init, trying fallback: %m");
1930                 }
1931
1932                 args[0] = "/sbin/init";
1933                 execv(args[0], (char* const*) args);
1934
1935                 if (errno == ENOENT) {
1936                         log_warning("No /sbin/init, trying fallback");
1937
1938                         args[0] = "/bin/sh";
1939                         args[1] = NULL;
1940                         execv(args[0], (char* const*) args);
1941                         log_error("Failed to execute /bin/sh, giving up: %m");
1942                 } else
1943                         log_warning("Failed to execute /sbin/init, giving up: %m");
1944         }
1945
1946         if (arg_serialization) {
1947                 fclose(arg_serialization);
1948                 arg_serialization = NULL;
1949         }
1950
1951         if (fds) {
1952                 fdset_free(fds);
1953                 fds = NULL;
1954         }
1955
1956 #ifdef HAVE_VALGRIND_VALGRIND_H
1957         /* If we are PID 1 and running under valgrind, then let's exit
1958          * here explicitly. valgrind will only generate nice output on
1959          * exit(), not on exec(), hence let's do the former not the
1960          * latter here. */
1961         if (getpid() == 1 && RUNNING_ON_VALGRIND)
1962                 return 0;
1963 #endif
1964
1965         if (shutdown_verb) {
1966                 char log_level[DECIMAL_STR_MAX(int) + 1];
1967                 const char* command_line[9] = {
1968                         SYSTEMD_SHUTDOWN_BINARY_PATH,
1969                         shutdown_verb,
1970                         "--log-level", log_level,
1971                         "--log-target",
1972                 };
1973                 unsigned pos = 5;
1974                 _cleanup_strv_free_ char **env_block = NULL;
1975
1976                 assert(command_line[pos] == NULL);
1977                 env_block = strv_copy(environ);
1978
1979                 snprintf(log_level, sizeof(log_level), "%d", log_get_max_level());
1980
1981                 switch (log_get_target()) {
1982                 case LOG_TARGET_KMSG:
1983                 case LOG_TARGET_JOURNAL_OR_KMSG:
1984                 case LOG_TARGET_SYSLOG_OR_KMSG:
1985                         command_line[pos++] = "kmsg";
1986                         break;
1987
1988                 case LOG_TARGET_CONSOLE:
1989                 default:
1990                         command_line[pos++] = "console";
1991                         break;
1992                 };
1993
1994                 if (log_get_show_color())
1995                         command_line[pos++] = "--log-color";
1996
1997                 if (log_get_show_location())
1998                         command_line[pos++] = "--log-location";
1999
2000                 assert(pos < ELEMENTSOF(command_line));
2001
2002                 if (arm_reboot_watchdog && arg_shutdown_watchdog > 0) {
2003                         char *e;
2004
2005                         /* If we reboot let's set the shutdown
2006                          * watchdog and tell the shutdown binary to
2007                          * repeatedly ping it */
2008                         watchdog_set_timeout(&arg_shutdown_watchdog);
2009                         watchdog_close(false);
2010
2011                         /* Tell the binary how often to ping, ignore failure */
2012                         if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, arg_shutdown_watchdog) > 0)
2013                                 strv_push(&env_block, e);
2014                 } else
2015                         watchdog_close(true);
2016
2017                 /* Avoid the creation of new processes forked by the
2018                  * kernel; at this point, we will not listen to the
2019                  * signals anyway */
2020                 if (detect_container(NULL) <= 0)
2021                         cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
2022
2023                 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
2024                 log_error("Failed to execute shutdown binary, %s: %m",
2025                           getpid() == 1 ? "freezing" : "quitting");
2026         }
2027
2028         if (getpid() == 1)
2029                 freeze();
2030
2031         return retval;
2032 }