src/core/main.c

   1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
   2
   3 /***
   4   This file is part of systemd.
   5
   6   Copyright 2010 Lennart Poettering
   7
   8   systemd is free software; you can redistribute it and/or modify it
   9   under the terms of the GNU Lesser General Public License as published by
  10   the Free Software Foundation; either version 2.1 of the License, or
  11   (at your option) any later version.
  12
  13   systemd is distributed in the hope that it will be useful, but
  14   WITHOUT ANY WARRANTY; without even the implied warranty of
  15   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16   Lesser General Public License for more details.
  17
  18   You should have received a copy of the GNU Lesser General Public License
  19   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  20 ***/
  21
  22 #include <stdio.h>
  23 #include <errno.h>
  24 #include <string.h>
  25 #include <unistd.h>
  26 #include <sys/types.h>
  27 #include <sys/stat.h>
  28 #include <getopt.h>
  29 #include <signal.h>
  30 #include <sys/wait.h>
  31 #include <fcntl.h>
  32 #include <sys/prctl.h>
  33 #include <sys/mount.h>
  34
  35 #ifdef HAVE_VALGRIND_VALGRIND_H
  36 #include <valgrind/valgrind.h>
  37 #endif
  38 #ifdef HAVE_SECCOMP
  39 #include <seccomp.h>
  40 #endif
  41
  42 #include "sd-daemon.h"
  43 #include "sd-messages.h"
  44 #include "sd-bus.h"
  45 #include "manager.h"
  46 #include "log.h"
  47 #include "load-fragment.h"
  48 #include "fdset.h"
  49 #include "special.h"
  50 #include "conf-parser.h"
  51 #include "missing.h"
  52 #include "label.h"
  53 #include "pager.h"
  54 #include "build.h"
  55 #include "strv.h"
  56 #include "def.h"
  57 #include "virt.h"
  58 #include "architecture.h"
  59 #include "watchdog.h"
  60 #include "path-util.h"
  61 #include "switch-root.h"
  62 #include "capability.h"
  63 #include "killall.h"
  64 #include "env-util.h"
  65 #include "clock-util.h"
  66 #include "fileio.h"
  67 #include "dbus-manager.h"
  68 #include "bus-error.h"
  69 #include "bus-util.h"
  70
  71 #include "mount-setup.h"
  72 #include "loopback-setup.h"
  73 #include "hostname-setup.h"
  74 #include "machine-id-setup.h"
  75 #include "selinux-setup.h"
  76 #include "ima-setup.h"
  77 #include "smack-setup.h"
  78 #ifdef HAVE_KMOD
  79 #include "kmod-setup.h"
  80 #endif
  81
  82 static enum {
  83         ACTION_RUN,
  84         ACTION_HELP,
  85         ACTION_VERSION,
  86         ACTION_TEST,
  87         ACTION_DUMP_CONFIGURATION_ITEMS,
  88         ACTION_DONE
  89 } arg_action = ACTION_RUN;
  90 static char *arg_default_unit = NULL;
  91 static SystemdRunningAs arg_running_as = _SYSTEMD_RUNNING_AS_INVALID;
  92 static bool arg_dump_core = true;
  93 static bool arg_crash_shell = false;
  94 static int arg_crash_chvt = -1;
  95 static bool arg_confirm_spawn = false;
  96 static ShowStatus arg_show_status = _SHOW_STATUS_UNSET;
  97 static bool arg_switched_root = false;
  98 static int arg_no_pager = -1;
  99 static char ***arg_join_controllers = NULL;
 100 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
 101 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
 102 static usec_t arg_default_restart_usec = DEFAULT_RESTART_USEC;
 103 static usec_t arg_default_timeout_start_usec = DEFAULT_TIMEOUT_USEC;
 104 static usec_t arg_default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC;
 105 static usec_t arg_default_start_limit_interval = DEFAULT_START_LIMIT_INTERVAL;
 106 static unsigned arg_default_start_limit_burst = DEFAULT_START_LIMIT_BURST;
 107 static usec_t arg_runtime_watchdog = 0;
 108 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
 109 static char **arg_default_environment = NULL;
 110 static struct rlimit *arg_default_rlimit[_RLIMIT_MAX] = {};
 111 static uint64_t arg_capability_bounding_set_drop = 0;
 112 static nsec_t arg_timer_slack_nsec = NSEC_INFINITY;
 113 static usec_t arg_default_timer_accuracy_usec = 1 * USEC_PER_MINUTE;
 114 static Set* arg_syscall_archs = NULL;
 115 static FILE* arg_serialization = NULL;
 116 static bool arg_default_cpu_accounting = false;
 117 static bool arg_default_blockio_accounting = false;
 118 static bool arg_default_memory_accounting = false;
 119
 120 static void nop_handler(int sig) {}
 121
 122 static void pager_open_if_enabled(void) {
 123
 124         if (arg_no_pager <= 0)
 125                 return;
 126
 127         pager_open(false);
 128 }
 129
 130 noreturn static void crash(int sig) {
 131
 132         if (getpid() != 1)
 133                 /* Pass this on immediately, if this is not PID 1 */
 134                 raise(sig);
 135         else if (!arg_dump_core)
 136                 log_error("Caught <%s>, not dumping core.", signal_to_string(sig));
 137         else {
 138                 struct sigaction sa = {
 139                         .sa_handler = nop_handler,
 140                         .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 141                 };
 142                 pid_t pid;
 143
 144                 /* We want to wait for the core process, hence let's enable SIGCHLD */
 145                 sigaction(SIGCHLD, &sa, NULL);
 146
 147                 pid = fork();
 148                 if (pid < 0)
 149                         log_error("Caught <%s>, cannot fork for core dump: %m", signal_to_string(sig));
 150
 151                 else if (pid == 0) {
 152                         struct rlimit rl = {};
 153
 154                         /* Enable default signal handler for core dump */
 155                         zero(sa);
 156                         sa.sa_handler = SIG_DFL;
 157                         sigaction(sig, &sa, NULL);
 158
 159                         /* Don't limit the core dump size */
 160                         rl.rlim_cur = RLIM_INFINITY;
 161                         rl.rlim_max = RLIM_INFINITY;
 162                         setrlimit(RLIMIT_CORE, &rl);
 163
 164                         /* Just to be sure... */
 165                         chdir("/");
 166
 167                         /* Raise the signal again */
 168                         raise(sig);
 169
 170                         assert_not_reached("We shouldn't be here...");
 171                         _exit(1);
 172
 173                 } else {
 174                         siginfo_t status;
 175                         int r;
 176
 177                         /* Order things nicely. */
 178                         r = wait_for_terminate(pid, &status);
 179                         if (r < 0)
 180                                 log_error("Caught <%s>, waitpid() failed: %s", signal_to_string(sig), strerror(-r));
 181                         else if (status.si_code != CLD_DUMPED)
 182                                 log_error("Caught <%s>, core dump failed.", signal_to_string(sig));
 183                         else
 184                                 log_error("Caught <%s>, dumped core as pid "PID_FMT".", signal_to_string(sig), pid);
 185                 }
 186         }
 187
 188         if (arg_crash_chvt)
 189                 chvt(arg_crash_chvt);
 190
 191         if (arg_crash_shell) {
 192                 struct sigaction sa = {
 193                         .sa_handler = SIG_IGN,
 194                         .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
 195                 };
 196                 pid_t pid;
 197
 198                 log_info("Executing crash shell in 10s...");
 199                 sleep(10);
 200
 201                 /* Let the kernel reap children for us */
 202                 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 203
 204                 pid = fork();
 205                 if (pid < 0)
 206                         log_error("Failed to fork off crash shell: %m");
 207                 else if (pid == 0) {
 208                         make_console_stdio();
 209                         execl("/bin/sh", "/bin/sh", NULL);
 210
 211                         log_error("execl() failed: %m");
 212                         _exit(1);
 213                 }
 214
 215                 log_info("Successfully spawned crash shell as pid "PID_FMT".", pid);
 216         }
 217
 218         log_info("Freezing execution.");
 219         freeze();
 220 }
 221
 222 static void install_crash_handler(void) {
 223         struct sigaction sa = {
 224                 .sa_handler = crash,
 225                 .sa_flags = SA_NODEFER,
 226         };
 227
 228         sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
 229 }
 230
 231 static int console_setup(void) {
 232         _cleanup_close_ int tty_fd = -1;
 233         int r;
 234
 235         tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
 236         if (tty_fd < 0) {
 237                 log_error("Failed to open /dev/console: %s", strerror(-tty_fd));
 238                 return tty_fd;
 239         }
 240
 241         /* We don't want to force text mode.  plymouth may be showing
 242          * pictures already from initrd. */
 243         r = reset_terminal_fd(tty_fd, false);
 244         if (r < 0) {
 245                 log_error("Failed to reset /dev/console: %s", strerror(-r));
 246                 return r;
 247         }
 248
 249         return 0;
 250 }
 251
 252 static int set_default_unit(const char *u) {
 253         char *c;
 254
 255         assert(u);
 256
 257         c = strdup(u);
 258         if (!c)
 259                 return -ENOMEM;
 260
 261         free(arg_default_unit);
 262         arg_default_unit = c;
 263
 264         return 0;
 265 }
 266
 267 static int parse_proc_cmdline_item(const char *key, const char *value) {
 268
 269         static const char * const rlmap[] = {
 270                 "emergency", SPECIAL_EMERGENCY_TARGET,
 271                 "-b",        SPECIAL_EMERGENCY_TARGET,
 272                 "single",    SPECIAL_RESCUE_TARGET,
 273                 "-s",        SPECIAL_RESCUE_TARGET,
 274                 "s",         SPECIAL_RESCUE_TARGET,
 275                 "S",         SPECIAL_RESCUE_TARGET,
 276                 "1",         SPECIAL_RESCUE_TARGET,
 277                 "2",         SPECIAL_RUNLEVEL2_TARGET,
 278                 "3",         SPECIAL_RUNLEVEL3_TARGET,
 279                 "4",         SPECIAL_RUNLEVEL4_TARGET,
 280                 "5",         SPECIAL_RUNLEVEL5_TARGET,
 281         };
 282         int r;
 283
 284         assert(key);
 285
 286         if (streq(key, "systemd.unit") && value) {
 287
 288                 if (!in_initrd())
 289                         return set_default_unit(value);
 290
 291         } else if (streq(key, "rd.systemd.unit") && value) {
 292
 293                 if (in_initrd())
 294                         return set_default_unit(value);
 295
 296         } else if (streq(key, "systemd.log_target") && value) {
 297
 298                 if (log_set_target_from_string(value) < 0)
 299                         log_warning("Failed to parse log target %s. Ignoring.", value);
 300
 301         } else if (streq(key, "systemd.log_level") && value) {
 302
 303                 if (log_set_max_level_from_string(value) < 0)
 304                         log_warning("Failed to parse log level %s. Ignoring.", value);
 305
 306         } else if (streq(key, "systemd.log_color") && value) {
 307
 308                 if (log_show_color_from_string(value) < 0)
 309                         log_warning("Failed to parse log color setting %s. Ignoring.", value);
 310
 311         } else if (streq(key, "systemd.log_location") && value) {
 312
 313                 if (log_show_location_from_string(value) < 0)
 314                         log_warning("Failed to parse log location setting %s. Ignoring.", value);
 315
 316         } else if (streq(key, "systemd.dump_core") && value) {
 317
 318                 r = parse_boolean(value);
 319                 if (r < 0)
 320                         log_warning("Failed to parse dump core switch %s. Ignoring.", value);
 321                 else
 322                         arg_dump_core = r;
 323
 324         } else if (streq(key, "systemd.crash_shell") && value) {
 325
 326                 r = parse_boolean(value);
 327                 if (r < 0)
 328                         log_warning("Failed to parse crash shell switch %s. Ignoring.", value);
 329                 else
 330                         arg_crash_shell = r;
 331
 332         } else if (streq(key, "systemd.crash_chvt") && value) {
 333
 334                 if (safe_atoi(value, &r) < 0)
 335                         log_warning("Failed to parse crash chvt switch %s. Ignoring.", value);
 336                 else
 337                         arg_crash_chvt = r;
 338
 339         } else if (streq(key, "systemd.confirm_spawn") && value) {
 340
 341                 r = parse_boolean(value);
 342                 if (r < 0)
 343                         log_warning("Failed to parse confirm spawn switch %s. Ignoring.", value);
 344                 else
 345                         arg_confirm_spawn = r;
 346
 347         } else if (streq(key, "systemd.show_status") && value) {
 348
 349                 r = parse_show_status(value, &arg_show_status);
 350                 if (r < 0)
 351                         log_warning("Failed to parse show status switch %s. Ignoring.", value);
 352
 353         } else if (streq(key, "systemd.default_standard_output") && value) {
 354
 355                 r = exec_output_from_string(value);
 356                 if (r < 0)
 357                         log_warning("Failed to parse default standard output switch %s. Ignoring.", value);
 358                 else
 359                         arg_default_std_output = r;
 360
 361         } else if (streq(key, "systemd.default_standard_error") && value) {
 362
 363                 r = exec_output_from_string(value);
 364                 if (r < 0)
 365                         log_warning("Failed to parse default standard error switch %s. Ignoring.", value);
 366                 else
 367                         arg_default_std_error = r;
 368
 369         } else if (streq(key, "systemd.setenv") && value) {
 370
 371                 if (env_assignment_is_valid(value)) {
 372                         char **env;
 373
 374                         env = strv_env_set(arg_default_environment, value);
 375                         if (env)
 376                                 arg_default_environment = env;
 377                         else
 378                                 log_warning("Setting environment variable '%s' failed, ignoring: %s", value, strerror(ENOMEM));
 379                 } else
 380                         log_warning("Environment variable name '%s' is not valid. Ignoring.", value);
 381
 382         } else if (streq(key, "quiet") && !value) {
 383
 384                 log_set_max_level(LOG_NOTICE);
 385
 386                 if (arg_show_status == _SHOW_STATUS_UNSET)
 387                         arg_show_status = SHOW_STATUS_AUTO;
 388
 389         } else if (streq(key, "debug") && !value) {
 390
 391                 log_set_max_level(LOG_DEBUG);
 392
 393                 if (detect_container(NULL) > 0)
 394                         log_set_target(LOG_TARGET_CONSOLE);
 395
 396         } else if (!in_initrd() && !value) {
 397                 unsigned i;
 398
 399                 /* SysV compatibility */
 400                 for (i = 0; i < ELEMENTSOF(rlmap); i += 2)
 401                         if (streq(key, rlmap[i]))
 402                                 return set_default_unit(rlmap[i+1]);
 403         }
 404
 405         return 0;
 406 }
 407
 408 #define DEFINE_SETTER(name, func, descr)                              \
 409         static int name(const char *unit,                             \
 410                         const char *filename,                         \
 411                         unsigned line,                                \
 412                         const char *section,                          \
 413                         unsigned section_line,                        \
 414                         const char *lvalue,                           \
 415                         int ltype,                                    \
 416                         const char *rvalue,                           \
 417                         void *data,                                   \
 418                         void *userdata) {                             \
 419                                                                       \
 420                 int r;                                                \
 421                                                                       \
 422                 assert(filename);                                     \
 423                 assert(lvalue);                                       \
 424                 assert(rvalue);                                       \
 425                                                                       \
 426                 r = func(rvalue);                                     \
 427                 if (r < 0)                                            \
 428                         log_syntax(unit, LOG_ERR, filename, line, -r, \
 429                                    "Invalid " descr "'%s': %s",       \
 430                                    rvalue, strerror(-r));             \
 431                                                                       \
 432                 return 0;                                             \
 433         }
 434
 435 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level")
 436 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target")
 437 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" )
 438 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location")
 439
 440 static int config_parse_cpu_affinity2(
 441                 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         const char *word, *state;
 453         size_t l;
 454         cpu_set_t *c = NULL;
 455         unsigned ncpus = 0;
 456
 457         assert(filename);
 458         assert(lvalue);
 459         assert(rvalue);
 460
 461         FOREACH_WORD_QUOTED(word, l, rvalue, state) {
 462                 char *t;
 463                 int r;
 464                 unsigned cpu;
 465
 466                 if (!(t = strndup(word, l)))
 467                         return log_oom();
 468
 469                 r = safe_atou(t, &cpu);
 470                 free(t);
 471
 472                 if (!c)
 473                         if (!(c = cpu_set_malloc(&ncpus)))
 474                                 return log_oom();
 475
 476                 if (r < 0 || cpu >= ncpus) {
 477                         log_syntax(unit, LOG_ERR, filename, line, -r,
 478                                    "Failed to parse CPU affinity '%s'", rvalue);
 479                         CPU_FREE(c);
 480                         return -EBADMSG;
 481                 }
 482
 483                 CPU_SET_S(cpu, CPU_ALLOC_SIZE(ncpus), c);
 484         }
 485         if (!isempty(state))
 486                 log_syntax(unit, LOG_ERR, filename, line, EINVAL,
 487                            "Trailing garbage, ignoring.");
 488
 489         if (c) {
 490                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
 491                         log_warning_unit(unit, "Failed to set CPU affinity: %m");
 492
 493                 CPU_FREE(c);
 494         }
 495
 496         return 0;
 497 }
 498
 499 static int config_parse_show_status(
 500                 const char* unit,
 501                 const char *filename,
 502                 unsigned line,
 503                 const char *section,
 504                 unsigned section_line,
 505                 const char *lvalue,
 506                 int ltype,
 507                 const char *rvalue,
 508                 void *data,
 509                 void *userdata) {
 510
 511         int k;
 512         ShowStatus *b = data;
 513
 514         assert(filename);
 515         assert(lvalue);
 516         assert(rvalue);
 517         assert(data);
 518
 519         k = parse_show_status(rvalue, b);
 520         if (k < 0) {
 521                 log_syntax(unit, LOG_ERR, filename, line, -k,
 522                            "Failed to parse show status setting, ignoring: %s", rvalue);
 523                 return 0;
 524         }
 525
 526         return 0;
 527 }
 528
 529 static void strv_free_free(char ***l) {
 530         char ***i;
 531
 532         if (!l)
 533                 return;
 534
 535         for (i = l; *i; i++)
 536                 strv_free(*i);
 537
 538         free(l);
 539 }
 540
 541 static void free_join_controllers(void) {
 542         strv_free_free(arg_join_controllers);
 543         arg_join_controllers = NULL;
 544 }
 545
 546 static int config_parse_join_controllers(const char *unit,
 547                                          const char *filename,
 548                                          unsigned line,
 549                                          const char *section,
 550                                          unsigned section_line,
 551                                          const char *lvalue,
 552                                          int ltype,
 553                                          const char *rvalue,
 554                                          void *data,
 555                                          void *userdata) {
 556
 557         unsigned n = 0;
 558         const char *word, *state;
 559         size_t length;
 560
 561         assert(filename);
 562         assert(lvalue);
 563         assert(rvalue);
 564
 565         free_join_controllers();
 566
 567         FOREACH_WORD_QUOTED(word, length, rvalue, state) {
 568                 char *s, **l;
 569
 570                 s = strndup(word, length);
 571                 if (!s)
 572                         return log_oom();
 573
 574                 l = strv_split(s, ",");
 575                 free(s);
 576
 577                 strv_uniq(l);
 578
 579                 if (strv_length(l) <= 1) {
 580                         strv_free(l);
 581                         continue;
 582                 }
 583
 584                 if (!arg_join_controllers) {
 585                         arg_join_controllers = new(char**, 2);
 586                         if (!arg_join_controllers) {
 587                                 strv_free(l);
 588                                 return log_oom();
 589                         }
 590
 591                         arg_join_controllers[0] = l;
 592                         arg_join_controllers[1] = NULL;
 593
 594                         n = 1;
 595                 } else {
 596                         char ***a;
 597                         char ***t;
 598
 599                         t = new0(char**, n+2);
 600                         if (!t) {
 601                                 strv_free(l);
 602                                 return log_oom();
 603                         }
 604
 605                         n = 0;
 606
 607                         for (a = arg_join_controllers; *a; a++) {
 608
 609                                 if (strv_overlap(*a, l)) {
 610                                         if (strv_extend_strv(&l, *a) < 0) {
 611                                                 strv_free(l);
 612                                                 strv_free_free(t);
 613                                                 return log_oom();
 614                                         }
 615
 616                                 } else {
 617                                         char **c;
 618
 619                                         c = strv_copy(*a);
 620                                         if (!c) {
 621                                                 strv_free(l);
 622                                                 strv_free_free(t);
 623                                                 return log_oom();
 624                                         }
 625
 626                                         t[n++] = c;
 627                                 }
 628                         }
 629
 630                         t[n++] = strv_uniq(l);
 631
 632                         strv_free_free(arg_join_controllers);
 633                         arg_join_controllers = t;
 634                 }
 635         }
 636         if (!isempty(state))
 637                 log_syntax(unit, LOG_ERR, filename, line, EINVAL,
 638                            "Trailing garbage, ignoring.");
 639
 640         return 0;
 641 }
 642
 643 static int parse_config_file(void) {
 644
 645         const ConfigTableItem items[] = {
 646                 { "Manager", "LogLevel",                  config_parse_level2,           0, NULL                                   },
 647                 { "Manager", "LogTarget",                 config_parse_target,           0, NULL                                   },
 648                 { "Manager", "LogColor",                  config_parse_color,            0, NULL                                   },
 649                 { "Manager", "LogLocation",               config_parse_location,         0, NULL                                   },
 650                 { "Manager", "DumpCore",                  config_parse_bool,             0, &arg_dump_core                         },
 651                 { "Manager", "CrashShell",                config_parse_bool,             0, &arg_crash_shell                       },
 652                 { "Manager", "ShowStatus",                config_parse_show_status,      0, &arg_show_status                       },
 653                 { "Manager", "CrashChVT",                 config_parse_int,              0, &arg_crash_chvt                        },
 654                 { "Manager", "CPUAffinity",               config_parse_cpu_affinity2,    0, NULL                                   },
 655                 { "Manager", "JoinControllers",           config_parse_join_controllers, 0, &arg_join_controllers                  },
 656                 { "Manager", "RuntimeWatchdogSec",        config_parse_sec,              0, &arg_runtime_watchdog                  },
 657                 { "Manager", "ShutdownWatchdogSec",       config_parse_sec,              0, &arg_shutdown_watchdog                 },
 658                 { "Manager", "CapabilityBoundingSet",     config_parse_bounding_set,     0, &arg_capability_bounding_set_drop      },
 659 #ifdef HAVE_SECCOMP
 660                 { "Manager", "SystemCallArchitectures",   config_parse_syscall_archs,    0, &arg_syscall_archs                     },
 661 #endif
 662                 { "Manager", "TimerSlackNSec",            config_parse_nsec,             0, &arg_timer_slack_nsec                  },
 663                 { "Manager", "DefaultTimerAccuracySec",   config_parse_sec,              0, &arg_default_timer_accuracy_usec       },
 664                 { "Manager", "DefaultStandardOutput",     config_parse_output,           0, &arg_default_std_output                },
 665                 { "Manager", "DefaultStandardError",      config_parse_output,           0, &arg_default_std_error                 },
 666                 { "Manager", "DefaultTimeoutStartSec",    config_parse_sec,              0, &arg_default_timeout_start_usec        },
 667                 { "Manager", "DefaultTimeoutStopSec",     config_parse_sec,              0, &arg_default_timeout_stop_usec         },
 668                 { "Manager", "DefaultRestartSec",         config_parse_sec,              0, &arg_default_restart_usec              },
 669                 { "Manager", "DefaultStartLimitInterval", config_parse_sec,              0, &arg_default_start_limit_interval      },
 670                 { "Manager", "DefaultStartLimitBurst",    config_parse_unsigned,         0, &arg_default_start_limit_burst         },
 671                 { "Manager", "DefaultEnvironment",        config_parse_environ,          0, &arg_default_environment               },
 672                 { "Manager", "DefaultLimitCPU",           config_parse_limit,            0, &arg_default_rlimit[RLIMIT_CPU]        },
 673                 { "Manager", "DefaultLimitFSIZE",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_FSIZE]      },
 674                 { "Manager", "DefaultLimitDATA",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_DATA]       },
 675                 { "Manager", "DefaultLimitSTACK",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_STACK]      },
 676                 { "Manager", "DefaultLimitCORE",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_CORE]       },
 677                 { "Manager", "DefaultLimitRSS",           config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RSS]        },
 678                 { "Manager", "DefaultLimitNOFILE",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NOFILE]     },
 679                 { "Manager", "DefaultLimitAS",            config_parse_limit,            0, &arg_default_rlimit[RLIMIT_AS]         },
 680                 { "Manager", "DefaultLimitNPROC",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NPROC]      },
 681                 { "Manager", "DefaultLimitMEMLOCK",       config_parse_limit,            0, &arg_default_rlimit[RLIMIT_MEMLOCK]    },
 682                 { "Manager", "DefaultLimitLOCKS",         config_parse_limit,            0, &arg_default_rlimit[RLIMIT_LOCKS]      },
 683                 { "Manager", "DefaultLimitSIGPENDING",    config_parse_limit,            0, &arg_default_rlimit[RLIMIT_SIGPENDING] },
 684                 { "Manager", "DefaultLimitMSGQUEUE",      config_parse_limit,            0, &arg_default_rlimit[RLIMIT_MSGQUEUE]   },
 685                 { "Manager", "DefaultLimitNICE",          config_parse_limit,            0, &arg_default_rlimit[RLIMIT_NICE]       },
 686                 { "Manager", "DefaultLimitRTPRIO",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RTPRIO]     },
 687                 { "Manager", "DefaultLimitRTTIME",        config_parse_limit,            0, &arg_default_rlimit[RLIMIT_RTTIME]     },
 688                 { "Manager", "DefaultCPUAccounting",      config_parse_bool,             0, &arg_default_cpu_accounting            },
 689                 { "Manager", "DefaultBlockIOAccounting",  config_parse_bool,             0, &arg_default_blockio_accounting        },
 690                 { "Manager", "DefaultMemoryAccounting",   config_parse_bool,             0, &arg_default_memory_accounting         },
 691                 {}
 692         };
 693
 694         const char *fn;
 695
 696         fn = arg_running_as == SYSTEMD_SYSTEM ? PKGSYSCONFDIR "/system.conf" : PKGSYSCONFDIR "/user.conf";
 697         config_parse(NULL, fn, NULL,
 698                      "Manager\0",
 699                      config_item_table_lookup, items,
 700                      false, false, true, NULL);
 701
 702         return 0;
 703 }
 704
 705 static int parse_argv(int argc, char *argv[]) {
 706
 707         enum {
 708                 ARG_LOG_LEVEL = 0x100,
 709                 ARG_LOG_TARGET,
 710                 ARG_LOG_COLOR,
 711                 ARG_LOG_LOCATION,
 712                 ARG_UNIT,
 713                 ARG_SYSTEM,
 714                 ARG_USER,
 715                 ARG_TEST,
 716                 ARG_NO_PAGER,
 717                 ARG_VERSION,
 718                 ARG_DUMP_CONFIGURATION_ITEMS,
 719                 ARG_DUMP_CORE,
 720                 ARG_CRASH_SHELL,
 721                 ARG_CONFIRM_SPAWN,
 722                 ARG_SHOW_STATUS,
 723                 ARG_DESERIALIZE,
 724                 ARG_SWITCHED_ROOT,
 725                 ARG_DEFAULT_STD_OUTPUT,
 726                 ARG_DEFAULT_STD_ERROR
 727         };
 728
 729         static const struct option options[] = {
 730                 { "log-level",                required_argument, NULL, ARG_LOG_LEVEL                },
 731                 { "log-target",               required_argument, NULL, ARG_LOG_TARGET               },
 732                 { "log-color",                optional_argument, NULL, ARG_LOG_COLOR                },
 733                 { "log-location",             optional_argument, NULL, ARG_LOG_LOCATION             },
 734                 { "unit",                     required_argument, NULL, ARG_UNIT                     },
 735                 { "system",                   no_argument,       NULL, ARG_SYSTEM                   },
 736                 { "user",                     no_argument,       NULL, ARG_USER                     },
 737                 { "test",                     no_argument,       NULL, ARG_TEST                     },
 738                 { "no-pager",                 no_argument,       NULL, ARG_NO_PAGER                 },
 739                 { "help",                     no_argument,       NULL, 'h'                          },
 740                 { "version",                  no_argument,       NULL, ARG_VERSION                  },
 741                 { "dump-configuration-items", no_argument,       NULL, ARG_DUMP_CONFIGURATION_ITEMS },
 742                 { "dump-core",                optional_argument, NULL, ARG_DUMP_CORE                },
 743                 { "crash-shell",              optional_argument, NULL, ARG_CRASH_SHELL              },
 744                 { "confirm-spawn",            optional_argument, NULL, ARG_CONFIRM_SPAWN            },
 745                 { "show-status",              optional_argument, NULL, ARG_SHOW_STATUS              },
 746                 { "deserialize",              required_argument, NULL, ARG_DESERIALIZE              },
 747                 { "switched-root",            no_argument,       NULL, ARG_SWITCHED_ROOT            },
 748                 { "default-standard-output",  required_argument, NULL, ARG_DEFAULT_STD_OUTPUT,      },
 749                 { "default-standard-error",   required_argument, NULL, ARG_DEFAULT_STD_ERROR,       },
 750                 {}
 751         };
 752
 753         int c, r;
 754
 755         assert(argc >= 1);
 756         assert(argv);
 757
 758         if (getpid() == 1)
 759                 opterr = 0;
 760
 761         while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
 762
 763                 switch (c) {
 764
 765                 case ARG_LOG_LEVEL:
 766                         r = log_set_max_level_from_string(optarg);
 767                         if (r < 0) {
 768                                 log_error("Failed to parse log level %s.", optarg);
 769                                 return r;
 770                         }
 771
 772                         break;
 773
 774                 case ARG_LOG_TARGET:
 775                         r = log_set_target_from_string(optarg);
 776                         if (r < 0) {
 777                                 log_error("Failed to parse log target %s.", optarg);
 778                                 return r;
 779                         }
 780
 781                         break;
 782
 783                 case ARG_LOG_COLOR:
 784
 785                         if (optarg) {
 786                                 r = log_show_color_from_string(optarg);
 787                                 if (r < 0) {
 788                                         log_error("Failed to parse log color setting %s.", optarg);
 789                                         return r;
 790                                 }
 791                         } else
 792                                 log_show_color(true);
 793
 794                         break;
 795
 796                 case ARG_LOG_LOCATION:
 797                         if (optarg) {
 798                                 r = log_show_location_from_string(optarg);
 799                                 if (r < 0) {
 800                                         log_error("Failed to parse log location setting %s.", optarg);
 801                                         return r;
 802                                 }
 803                         } else
 804                                 log_show_location(true);
 805
 806                         break;
 807
 808                 case ARG_DEFAULT_STD_OUTPUT:
 809                         r = exec_output_from_string(optarg);
 810                         if (r < 0) {
 811                                 log_error("Failed to parse default standard output setting %s.", optarg);
 812                                 return r;
 813                         } else
 814                                 arg_default_std_output = r;
 815                         break;
 816
 817                 case ARG_DEFAULT_STD_ERROR:
 818                         r = exec_output_from_string(optarg);
 819                         if (r < 0) {
 820                                 log_error("Failed to parse default standard error output setting %s.", optarg);
 821                                 return r;
 822                         } else
 823                                 arg_default_std_error = r;
 824                         break;
 825
 826                 case ARG_UNIT:
 827
 828                         r = set_default_unit(optarg);
 829                         if (r < 0) {
 830                                 log_error("Failed to set default unit %s: %s", optarg, strerror(-r));
 831                                 return r;
 832                         }
 833
 834                         break;
 835
 836                 case ARG_SYSTEM:
 837                         arg_running_as = SYSTEMD_SYSTEM;
 838                         break;
 839
 840                 case ARG_USER:
 841                         arg_running_as = SYSTEMD_USER;
 842                         break;
 843
 844                 case ARG_TEST:
 845                         arg_action = ACTION_TEST;
 846                         if (arg_no_pager < 0)
 847                                 arg_no_pager = true;
 848                         break;
 849
 850                 case ARG_NO_PAGER:
 851                         arg_no_pager = true;
 852                         break;
 853
 854                 case ARG_VERSION:
 855                         arg_action = ACTION_VERSION;
 856                         break;
 857
 858                 case ARG_DUMP_CONFIGURATION_ITEMS:
 859                         arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
 860                         break;
 861
 862                 case ARG_DUMP_CORE:
 863                         r = optarg ? parse_boolean(optarg) : 1;
 864                         if (r < 0) {
 865                                 log_error("Failed to parse dump core boolean %s.", optarg);
 866                                 return r;
 867                         }
 868                         arg_dump_core = r;
 869                         break;
 870
 871                 case ARG_CRASH_SHELL:
 872                         r = optarg ? parse_boolean(optarg) : 1;
 873                         if (r < 0) {
 874                                 log_error("Failed to parse crash shell boolean %s.", optarg);
 875                                 return r;
 876                         }
 877                         arg_crash_shell = r;
 878                         break;
 879
 880                 case ARG_CONFIRM_SPAWN:
 881                         r = optarg ? parse_boolean(optarg) : 1;
 882                         if (r < 0) {
 883                                 log_error("Failed to parse confirm spawn boolean %s.", optarg);
 884                                 return r;
 885                         }
 886                         arg_confirm_spawn = r;
 887                         break;
 888
 889                 case ARG_SHOW_STATUS:
 890                         if (optarg) {
 891                                 r = parse_show_status(optarg, &arg_show_status);
 892                                 if (r < 0) {
 893                                         log_error("Failed to parse show status boolean %s.", optarg);
 894                                         return r;
 895                                 }
 896                         } else
 897                                 arg_show_status = SHOW_STATUS_YES;
 898                         break;
 899
 900                 case ARG_DESERIALIZE: {
 901                         int fd;
 902                         FILE *f;
 903
 904                         r = safe_atoi(optarg, &fd);
 905                         if (r < 0 || fd < 0) {
 906                                 log_error("Failed to parse deserialize option %s.", optarg);
 907                                 return r < 0 ? r : -EINVAL;
 908                         }
 909
 910                         fd_cloexec(fd, true);
 911
 912                         f = fdopen(fd, "r");
 913                         if (!f) {
 914                                 log_error("Failed to open serialization fd: %m");
 915                                 return -errno;
 916                         }
 917
 918                         if (arg_serialization)
 919                                 fclose(arg_serialization);
 920
 921                         arg_serialization = f;
 922
 923                         break;
 924                 }
 925
 926                 case ARG_SWITCHED_ROOT:
 927                         arg_switched_root = true;
 928                         break;
 929
 930                 case 'h':
 931                         arg_action = ACTION_HELP;
 932                         if (arg_no_pager < 0)
 933                                 arg_no_pager = true;
 934                         break;
 935
 936                 case 'D':
 937                         log_set_max_level(LOG_DEBUG);
 938                         break;
 939
 940                 case 'b':
 941                 case 's':
 942                 case 'z':
 943                         /* Just to eat away the sysvinit kernel
 944                          * cmdline args without getopt() error
 945                          * messages that we'll parse in
 946                          * parse_proc_cmdline_word() or ignore. */
 947
 948                 case '?':
 949                         if (getpid() != 1)
 950                                 return -EINVAL;
 951                         else
 952                                 return 0;
 953
 954                 default:
 955                         assert_not_reached("Unhandled option code.");
 956                 }
 957
 958         if (optind < argc && getpid() != 1) {
 959                 /* Hmm, when we aren't run as init system
 960                  * let's complain about excess arguments */
 961
 962                 log_error("Excess arguments.");
 963                 return -EINVAL;
 964         }
 965
 966         if (detect_container(NULL) > 0) {
 967                 char **a;
 968
 969                 /* All /proc/cmdline arguments the kernel didn't
 970                  * understand it passed to us. We're not really
 971                  * interested in that usually since /proc/cmdline is
 972                  * more interesting and complete. With one exception:
 973                  * if we are run in a container /proc/cmdline is not
 974                  * relevant for the container, hence we rely on argv[]
 975                  * instead. */
 976
 977                 for (a = argv; a < argv + argc; a++) {
 978                         _cleanup_free_ char *w;
 979                         char *value;
 980
 981                         w = strdup(*a);
 982                         if (!w)
 983                                 return log_oom();
 984
 985                         value = strchr(w, '=');
 986                         if (value)
 987                                 *(value++) = 0;
 988
 989                         r = parse_proc_cmdline_item(w, value);
 990                         if (r < 0) {
 991                                 log_error("Failed on cmdline argument %s: %s", *a, strerror(-r));
 992                                 return r;
 993                         }
 994                 }
 995         }
 996
 997         return 0;
 998 }
 999
1000 static int help(void) {
1001
1002         printf("%s [OPTIONS...]\n\n"
1003                "Starts up and maintains the system or user services.\n\n"
1004                "  -h --help                      Show this help\n"
1005                "     --test                      Determine startup sequence, dump it and exit\n"
1006                "     --no-pager                  Do not pipe output into a pager\n"
1007                "     --dump-configuration-items  Dump understood unit configuration items\n"
1008                "     --unit=UNIT                 Set default unit\n"
1009                "     --system                    Run a system instance, even if PID != 1\n"
1010                "     --user                      Run a user instance\n"
1011                "     --dump-core[=0|1]           Dump core on crash\n"
1012                "     --crash-shell[=0|1]         Run shell on crash\n"
1013                "     --confirm-spawn[=0|1]       Ask for confirmation when spawning processes\n"
1014                "     --show-status[=0|1]         Show status updates on the console during bootup\n"
1015                "     --log-target=TARGET         Set log target (console, journal, kmsg, journal-or-kmsg, null)\n"
1016                "     --log-level=LEVEL           Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1017                "     --log-color[=0|1]           Highlight important log messages\n"
1018                "     --log-location[=0|1]        Include code location in log messages\n"
1019                "     --default-standard-output=  Set default standard output for services\n"
1020                "     --default-standard-error=   Set default standard error output for services\n",
1021                program_invocation_short_name);
1022
1023         return 0;
1024 }
1025
1026 static int version(void) {
1027         puts(PACKAGE_STRING);
1028         puts(SYSTEMD_FEATURES);
1029
1030         return 0;
1031 }
1032
1033 static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool switching_root) {
1034         FILE *f = NULL;
1035         FDSet *fds = NULL;
1036         int r;
1037
1038         assert(m);
1039         assert(_f);
1040         assert(_fds);
1041
1042         r = manager_open_serialization(m, &f);
1043         if (r < 0) {
1044                 log_error("Failed to create serialization file: %s", strerror(-r));
1045                 goto fail;
1046         }
1047
1048         /* Make sure nothing is really destructed when we shut down */
1049         m->n_reloading ++;
1050         bus_manager_send_reloading(m, true);
1051
1052         fds = fdset_new();
1053         if (!fds) {
1054                 r = -ENOMEM;
1055                 log_error("Failed to allocate fd set: %s", strerror(-r));
1056                 goto fail;
1057         }
1058
1059         r = manager_serialize(m, f, fds, switching_root);
1060         if (r < 0) {
1061                 log_error("Failed to serialize state: %s", strerror(-r));
1062                 goto fail;
1063         }
1064
1065         if (fseeko(f, 0, SEEK_SET) < 0) {
1066                 log_error("Failed to rewind serialization fd: %m");
1067                 goto fail;
1068         }
1069
1070         r = fd_cloexec(fileno(f), false);
1071         if (r < 0) {
1072                 log_error("Failed to disable O_CLOEXEC for serialization: %s", strerror(-r));
1073                 goto fail;
1074         }
1075
1076         r = fdset_cloexec(fds, false);
1077         if (r < 0) {
1078                 log_error("Failed to disable O_CLOEXEC for serialization fds: %s", strerror(-r));
1079                 goto fail;
1080         }
1081
1082         *_f = f;
1083         *_fds = fds;
1084
1085         return 0;
1086
1087 fail:
1088         fdset_free(fds);
1089
1090         if (f)
1091                 fclose(f);
1092
1093         return r;
1094 }
1095
1096 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1097         struct rlimit nl;
1098         int r;
1099
1100         assert(saved_rlimit);
1101
1102         /* Save the original RLIMIT_NOFILE so that we can reset it
1103          * later when transitioning from the initrd to the main
1104          * systemd or suchlike. */
1105         if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0) {
1106                 log_error("Reading RLIMIT_NOFILE failed: %m");
1107                 return -errno;
1108         }
1109
1110         /* Make sure forked processes get the default kernel setting */
1111         if (!arg_default_rlimit[RLIMIT_NOFILE]) {
1112                 struct rlimit *rl;
1113
1114                 rl = newdup(struct rlimit, saved_rlimit, 1);
1115                 if (!rl)
1116                         return log_oom();
1117
1118                 arg_default_rlimit[RLIMIT_NOFILE] = rl;
1119         }
1120
1121         /* Bump up the resource limit for ourselves substantially */
1122         nl.rlim_cur = nl.rlim_max = 64*1024;
1123         r = setrlimit_closest(RLIMIT_NOFILE, &nl);
1124         if (r < 0) {
1125                 log_error("Setting RLIMIT_NOFILE failed: %s", strerror(-r));
1126                 return r;
1127         }
1128
1129         return 0;
1130 }
1131
1132 static void test_mtab(void) {
1133
1134         static const char ok[] =
1135                 "/proc/self/mounts\0"
1136                 "/proc/mounts\0"
1137                 "../proc/self/mounts\0"
1138                 "../proc/mounts\0";
1139
1140         _cleanup_free_ char *p = NULL;
1141         int r;
1142
1143         /* Check that /etc/mtab is a symlink to the right place or
1144          * non-existing. But certainly not a file, or a symlink to
1145          * some weird place... */
1146
1147         r = readlink_malloc("/etc/mtab", &p);
1148         if (r == -ENOENT)
1149                 return;
1150         if (r >= 0 && nulstr_contains(ok, p))
1151                 return;
1152
1153         log_warning("/etc/mtab is not a symlink or not pointing to /proc/self/mounts. "
1154                     "This is not supported anymore. "
1155                     "Please make sure to replace this file by a symlink to avoid incorrect or misleading mount(8) output.");
1156 }
1157
1158 static void test_usr(void) {
1159
1160         /* Check that /usr is not a separate fs */
1161
1162         if (dir_is_empty("/usr") <= 0)
1163                 return;
1164
1165         log_warning("/usr appears to be on its own filesytem and is not already mounted. This is not a supported setup. "
1166                     "Some things will probably break (sometimes even silently) in mysterious ways. "
1167                     "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1168 }
1169
1170 static int initialize_join_controllers(void) {
1171         /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
1172          * + "net_prio". We'd like to add "cpuset" to the mix, but
1173          * "cpuset" does't really work for groups with no initialized
1174          * attributes. */
1175
1176         arg_join_controllers = new(char**, 3);
1177         if (!arg_join_controllers)
1178                 return -ENOMEM;
1179
1180         arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
1181         arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
1182         arg_join_controllers[2] = NULL;
1183
1184         if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
1185                 free_join_controllers();
1186                 return -ENOMEM;
1187         }
1188
1189         return 0;
1190 }
1191
1192 static int enforce_syscall_archs(Set *archs) {
1193 #ifdef HAVE_SECCOMP
1194         scmp_filter_ctx *seccomp;
1195         Iterator i;
1196         void *id;
1197         int r;
1198
1199         seccomp = seccomp_init(SCMP_ACT_ALLOW);
1200         if (!seccomp)
1201                 return log_oom();
1202
1203         SET_FOREACH(id, arg_syscall_archs, i) {
1204                 r = seccomp_arch_add(seccomp, PTR_TO_UINT32(id) - 1);
1205                 if (r == -EEXIST)
1206                         continue;
1207                 if (r < 0) {
1208                         log_error("Failed to add architecture to seccomp: %s", strerror(-r));
1209                         goto finish;
1210                 }
1211         }
1212
1213         r = seccomp_attr_set(seccomp, SCMP_FLTATR_CTL_NNP, 0);
1214         if (r < 0) {
1215                 log_error("Failed to unset NO_NEW_PRIVS: %s", strerror(-r));
1216                 goto finish;
1217         }
1218
1219         r = seccomp_load(seccomp);
1220         if (r < 0)
1221                 log_error("Failed to add install architecture seccomp: %s", strerror(-r));
1222
1223 finish:
1224         seccomp_release(seccomp);
1225         return r;
1226 #else
1227         return 0;
1228 #endif
1229 }
1230
1231 static int status_welcome(void) {
1232         _cleanup_free_ char *pretty_name = NULL, *ansi_color = NULL;
1233         int r;
1234
1235         r = parse_env_file("/etc/os-release", NEWLINE,
1236                            "PRETTY_NAME", &pretty_name,
1237                            "ANSI_COLOR", &ansi_color,
1238                            NULL);
1239         if (r == -ENOENT) {
1240                 r = parse_env_file("/usr/lib/os-release", NEWLINE,
1241                                    "PRETTY_NAME", &pretty_name,
1242                                    "ANSI_COLOR", &ansi_color,
1243                                    NULL);
1244         }
1245
1246         if (r < 0 && r != -ENOENT)
1247                 log_warning("Failed to read os-release file: %s", strerror(-r));
1248
1249         return status_printf(NULL, false, false,
1250                              "\nWelcome to \x1B[%sm%s\x1B[0m!\n",
1251                              isempty(ansi_color) ? "1" : ansi_color,
1252                              isempty(pretty_name) ? "Linux" : pretty_name);
1253 }
1254
1255 static int write_container_id(void) {
1256         const char *c;
1257
1258         c = getenv("container");
1259         if (isempty(c))
1260                 return 0;
1261
1262         return write_string_file("/run/systemd/container", c);
1263 }
1264
1265 int main(int argc, char *argv[]) {
1266         Manager *m = NULL;
1267         int r, retval = EXIT_FAILURE;
1268         usec_t before_startup, after_startup;
1269         char timespan[FORMAT_TIMESPAN_MAX];
1270         FDSet *fds = NULL;
1271         bool reexecute = false;
1272         const char *shutdown_verb = NULL;
1273         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
1274         dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
1275         dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
1276         dual_timestamp security_start_timestamp = { 0ULL, 0ULL };
1277         dual_timestamp security_finish_timestamp = { 0ULL, 0ULL };
1278         static char systemd[] = "systemd";
1279         bool skip_setup = false;
1280         unsigned j;
1281         bool loaded_policy = false;
1282         bool arm_reboot_watchdog = false;
1283         bool queue_default_job = false;
1284         bool empty_etc = false;
1285         char *switch_root_dir = NULL, *switch_root_init = NULL;
1286         static struct rlimit saved_rlimit_nofile = { 0, 0 };
1287
1288 #ifdef HAVE_SYSV_COMPAT
1289         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
1290                 /* This is compatibility support for SysV, where
1291                  * calling init as a user is identical to telinit. */
1292
1293                 errno = -ENOENT;
1294                 execv(SYSTEMCTL_BINARY_PATH, argv);
1295                 log_error("Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1296                 return 1;
1297         }
1298 #endif
1299
1300         dual_timestamp_from_monotonic(&kernel_timestamp, 0);
1301         dual_timestamp_get(&userspace_timestamp);
1302
1303         /* Determine if this is a reexecution or normal bootup. We do
1304          * the full command line parsing much later, so let's just
1305          * have a quick peek here. */
1306         if (strv_find(argv+1, "--deserialize"))
1307                 skip_setup = true;
1308
1309         /* If we have switched root, do all the special setup
1310          * things */
1311         if (strv_find(argv+1, "--switched-root"))
1312                 skip_setup = false;
1313
1314         /* If we get started via the /sbin/init symlink then we are
1315            called 'init'. After a subsequent reexecution we are then
1316            called 'systemd'. That is confusing, hence let's call us
1317            systemd right-away. */
1318         program_invocation_short_name = systemd;
1319         prctl(PR_SET_NAME, systemd);
1320
1321         saved_argv = argv;
1322         saved_argc = argc;
1323
1324         log_show_color(isatty(STDERR_FILENO) > 0);
1325         log_set_upgrade_syslog_to_journal(true);
1326
1327         /* Disable the umask logic */
1328         if (getpid() == 1)
1329                 umask(0);
1330
1331         if (getpid() == 1 && detect_container(NULL) <= 0) {
1332
1333                 /* Running outside of a container as PID 1 */
1334                 arg_running_as = SYSTEMD_SYSTEM;
1335                 make_null_stdio();
1336                 log_set_target(LOG_TARGET_KMSG);
1337                 log_open();
1338
1339                 if (in_initrd())
1340                         initrd_timestamp = userspace_timestamp;
1341
1342                 if (!skip_setup) {
1343                         mount_setup_early();
1344                         dual_timestamp_get(&security_start_timestamp);
1345                         if (selinux_setup(&loaded_policy) < 0)
1346                                 goto finish;
1347                         if (ima_setup() < 0)
1348                                 goto finish;
1349                         if (smack_setup(&loaded_policy) < 0)
1350                                 goto finish;
1351                         dual_timestamp_get(&security_finish_timestamp);
1352                 }
1353
1354                 if (label_init(NULL) < 0)
1355                         goto finish;
1356
1357                 if (!skip_setup) {
1358                         if (clock_is_localtime() > 0) {
1359                                 int min;
1360
1361                                 /*
1362                                  * The very first call of settimeofday() also does a time warp in the kernel.
1363                                  *
1364                                  * In the rtc-in-local time mode, we set the kernel's timezone, and rely on
1365                                  * external tools to take care of maintaining the RTC and do all adjustments.
1366                                  * This matches the behavior of Windows, which leaves the RTC alone if the
1367                                  * registry tells that the RTC runs in UTC.
1368                                  */
1369                                 r = clock_set_timezone(&min);
1370                                 if (r < 0)
1371                                         log_error("Failed to apply local time delta, ignoring: %s", strerror(-r));
1372                                 else
1373                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1374                         } else if (!in_initrd()) {
1375                                 /*
1376                                  * Do a dummy very first call to seal the kernel's time warp magic.
1377                                  *
1378                                  * Do not call this this from inside the initrd. The initrd might not
1379                                  * carry /etc/adjtime with LOCAL, but the real system could be set up
1380                                  * that way. In such case, we need to delay the time-warp or the sealing
1381                                  * until we reach the real system.
1382                                  *
1383                                  * Do no set the kernel's timezone. The concept of local time cannot
1384                                  * be supported reliably, the time will jump or be incorrect at every daylight
1385                                  * saving time change. All kernel local time concepts will be treated
1386                                  * as UTC that way.
1387                                  */
1388                                 clock_reset_timewarp();
1389                         }
1390                 }
1391
1392                 /* Set the default for later on, but don't actually
1393                  * open the logs like this for now. Note that if we
1394                  * are transitioning from the initrd there might still
1395                  * be journal fd open, and we shouldn't attempt
1396                  * opening that before we parsed /proc/cmdline which
1397                  * might redirect output elsewhere. */
1398                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1399
1400         } else if (getpid() == 1) {
1401                 /* Running inside a container, as PID 1 */
1402                 arg_running_as = SYSTEMD_SYSTEM;
1403                 log_set_target(LOG_TARGET_CONSOLE);
1404                 log_close_console(); /* force reopen of /dev/console */
1405                 log_open();
1406
1407                 /* For the later on, see above... */
1408                 log_set_target(LOG_TARGET_JOURNAL);
1409
1410                 /* clear the kernel timestamp,
1411                  * because we are in a container */
1412                 kernel_timestamp.monotonic = 0ULL;
1413                 kernel_timestamp.realtime = 0ULL;
1414
1415         } else {
1416                 /* Running as user instance */
1417                 arg_running_as = SYSTEMD_USER;
1418                 log_set_target(LOG_TARGET_AUTO);
1419                 log_open();
1420
1421                 /* clear the kernel timestamp,
1422                  * because we are not PID 1 */
1423                 kernel_timestamp.monotonic = 0ULL;
1424                 kernel_timestamp.realtime = 0ULL;
1425         }
1426
1427         /* Initialize default unit */
1428         r = set_default_unit(SPECIAL_DEFAULT_TARGET);
1429         if (r < 0) {
1430                 log_error("Failed to set default unit %s: %s", SPECIAL_DEFAULT_TARGET, strerror(-r));
1431                 goto finish;
1432         }
1433
1434         r = initialize_join_controllers();
1435         if (r < 0)
1436                 goto finish;
1437
1438         /* Mount /proc, /sys and friends, so that /proc/cmdline and
1439          * /proc/$PID/fd is available. */
1440         if (getpid() == 1) {
1441                 r = mount_setup(loaded_policy);
1442                 if (r < 0)
1443                         goto finish;
1444         }
1445
1446         /* Reset all signal handlers. */
1447         assert_se(reset_all_signal_handlers() == 0);
1448
1449         ignore_signals(SIGNALS_IGNORE, -1);
1450
1451         if (parse_config_file() < 0)
1452                 goto finish;
1453
1454         if (arg_running_as == SYSTEMD_SYSTEM)
1455                 if (parse_proc_cmdline(parse_proc_cmdline_item) < 0)
1456                         goto finish;
1457
1458         log_parse_environment();
1459
1460         if (parse_argv(argc, argv) < 0)
1461                 goto finish;
1462
1463         if (arg_action == ACTION_TEST &&
1464             geteuid() == 0) {
1465                 log_error("Don't run test mode as root.");
1466                 goto finish;
1467         }
1468
1469         if (arg_running_as == SYSTEMD_USER &&
1470             arg_action == ACTION_RUN &&
1471             sd_booted() <= 0) {
1472                 log_error("Trying to run as user instance, but the system has not been booted with systemd.");
1473                 goto finish;
1474         }
1475
1476         if (arg_running_as == SYSTEMD_SYSTEM &&
1477             arg_action == ACTION_RUN &&
1478             running_in_chroot() > 0) {
1479                 log_error("Cannot be run in a chroot() environment.");
1480                 goto finish;
1481         }
1482
1483         if (arg_action == ACTION_TEST)
1484                 skip_setup = true;
1485
1486         pager_open_if_enabled();
1487
1488         if (arg_action == ACTION_HELP) {
1489                 retval = help();
1490                 goto finish;
1491         } else if (arg_action == ACTION_VERSION) {
1492                 retval = version();
1493                 goto finish;
1494         } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
1495                 unit_dump_config_items(stdout);
1496                 retval = EXIT_SUCCESS;
1497                 goto finish;
1498         } else if (arg_action == ACTION_DONE) {
1499                 retval = EXIT_SUCCESS;
1500                 goto finish;
1501         }
1502
1503         if (arg_running_as == SYSTEMD_USER &&
1504             !getenv("XDG_RUNTIME_DIR")) {
1505                 log_error("Trying to run as user instance, but $XDG_RUNTIME_DIR is not set.");
1506                 goto finish;
1507         }
1508
1509         assert_se(arg_action == ACTION_RUN || arg_action == ACTION_TEST);
1510
1511         /* Close logging fds, in order not to confuse fdset below */
1512         log_close();
1513
1514         /* Remember open file descriptors for later deserialization */
1515         r = fdset_new_fill(&fds);
1516         if (r < 0) {
1517                 log_error("Failed to allocate fd set: %s", strerror(-r));
1518                 goto finish;
1519         } else
1520                 fdset_cloexec(fds, true);
1521
1522         if (arg_serialization)
1523                 assert_se(fdset_remove(fds, fileno(arg_serialization)) >= 0);
1524
1525         if (arg_running_as == SYSTEMD_SYSTEM)
1526                 /* Become a session leader if we aren't one yet. */
1527                 setsid();
1528
1529         /* Move out of the way, so that we won't block unmounts */
1530         assert_se(chdir("/")  == 0);
1531
1532         /* Reset the console, but only if this is really init and we
1533          * are freshly booted */
1534         if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN) {
1535
1536                 /* If we are init, we connect stdin/stdout/stderr to
1537                  * /dev/null and make sure we don't have a controlling
1538                  * tty. */
1539                 release_terminal();
1540
1541                 if (getpid() == 1 && !skip_setup)
1542                         console_setup();
1543         }
1544
1545         /* Open the logging devices, if possible and necessary */
1546         log_open();
1547
1548         if (arg_show_status == _SHOW_STATUS_UNSET)
1549                 arg_show_status = SHOW_STATUS_YES;
1550
1551         /* Make sure we leave a core dump without panicing the
1552          * kernel. */
1553         if (getpid() == 1) {
1554                 install_crash_handler();
1555
1556                 r = mount_cgroup_controllers(arg_join_controllers);
1557                 if (r < 0)
1558                         goto finish;
1559         }
1560
1561         if (arg_running_as == SYSTEMD_SYSTEM) {
1562                 const char *virtualization = NULL;
1563
1564                 log_info(PACKAGE_STRING " running in %ssystem mode. (" SYSTEMD_FEATURES ")",
1565                          arg_action == ACTION_TEST ? "test " : "" );
1566
1567                 detect_virtualization(&virtualization);
1568                 if (virtualization)
1569                         log_info("Detected virtualization '%s'.", virtualization);
1570
1571                 write_container_id();
1572
1573                 log_info("Detected architecture '%s'.", architecture_to_string(uname_architecture()));
1574
1575                 if (in_initrd())
1576                         log_info("Running in initial RAM disk.");
1577
1578                 /* Let's check whether /etc is already populated. We
1579                  * don't actually really check for that, but use
1580                  * /etc/machine-id as flag file. This allows container
1581                  * managers and installers to provision a couple of
1582                  * files already. If the container manager wants to
1583                  * provision the machine ID itself it should pass
1584                  * $container_uuid to PID 1.*/
1585
1586                 empty_etc = access("/etc/machine-id", F_OK) < 0;
1587                 if (empty_etc)
1588                         log_info("Running with unpopulated /etc.");
1589         } else {
1590                 _cleanup_free_ char *t;
1591
1592                 t = uid_to_name(getuid());
1593                 log_debug(PACKAGE_STRING " running in %suser mode for user "UID_FMT"/%s. (" SYSTEMD_FEATURES ")",
1594                           arg_action == ACTION_TEST ? " test" : "", getuid(), t);
1595         }
1596
1597         if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
1598                 if (arg_show_status > 0 || plymouth_running())
1599                         status_welcome();
1600
1601 #ifdef HAVE_KMOD
1602                 kmod_setup();
1603 #endif
1604                 hostname_setup();
1605                 machine_id_setup(NULL);
1606                 loopback_setup();
1607
1608                 test_mtab();
1609                 test_usr();
1610         }
1611
1612         if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
1613                 watchdog_set_timeout(&arg_runtime_watchdog);
1614
1615         if (arg_timer_slack_nsec != NSEC_INFINITY)
1616                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1617                         log_error("Failed to adjust timer slack: %m");
1618
1619         if (arg_capability_bounding_set_drop) {
1620                 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
1621                 if (r < 0) {
1622                         log_error("Failed to drop capability bounding set of usermode helpers: %s", strerror(-r));
1623                         goto finish;
1624                 }
1625                 r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
1626                 if (r < 0) {
1627                         log_error("Failed to drop capability bounding set: %s", strerror(-r));
1628                         goto finish;
1629                 }
1630         }
1631
1632         if (arg_syscall_archs) {
1633                 r = enforce_syscall_archs(arg_syscall_archs);
1634                 if (r < 0)
1635                         goto finish;
1636         }
1637
1638         if (arg_running_as == SYSTEMD_USER) {
1639                 /* Become reaper of our children */
1640                 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
1641                         log_warning("Failed to make us a subreaper: %m");
1642                         if (errno == EINVAL)
1643                                 log_info("Perhaps the kernel version is too old (< 3.4?)");
1644                 }
1645         }
1646
1647         if (arg_running_as == SYSTEMD_SYSTEM) {
1648                 bump_rlimit_nofile(&saved_rlimit_nofile);
1649
1650                 if (empty_etc) {
1651                         r = unit_file_preset_all(UNIT_FILE_SYSTEM, false, NULL, UNIT_FILE_PRESET_FULL, false, NULL, 0);
1652                         if (r < 0)
1653                                 log_warning("Failed to populate /etc with preset unit settings, ignoring: %s", strerror(-r));
1654                         else
1655                                 log_info("Populated /etc with preset unit settings.");
1656                 }
1657         }
1658
1659         r = manager_new(arg_running_as, arg_action == ACTION_TEST, &m);
1660         if (r < 0) {
1661                 log_error("Failed to allocate manager object: %s", strerror(-r));
1662                 goto finish;
1663         }
1664
1665         m->confirm_spawn = arg_confirm_spawn;
1666         m->default_timer_accuracy_usec = arg_default_timer_accuracy_usec;
1667         m->default_std_output = arg_default_std_output;
1668         m->default_std_error = arg_default_std_error;
1669         m->default_restart_usec = arg_default_restart_usec;
1670         m->default_timeout_start_usec = arg_default_timeout_start_usec;
1671         m->default_timeout_stop_usec = arg_default_timeout_stop_usec;
1672         m->default_start_limit_interval = arg_default_start_limit_interval;
1673         m->default_start_limit_burst = arg_default_start_limit_burst;
1674         m->default_cpu_accounting = arg_default_cpu_accounting;
1675         m->default_blockio_accounting = arg_default_blockio_accounting;
1676         m->default_memory_accounting = arg_default_memory_accounting;
1677         m->runtime_watchdog = arg_runtime_watchdog;
1678         m->shutdown_watchdog = arg_shutdown_watchdog;
1679         m->userspace_timestamp = userspace_timestamp;
1680         m->kernel_timestamp = kernel_timestamp;
1681         m->initrd_timestamp = initrd_timestamp;
1682         m->security_start_timestamp = security_start_timestamp;
1683         m->security_finish_timestamp = security_finish_timestamp;
1684
1685         manager_set_default_rlimits(m, arg_default_rlimit);
1686         manager_environment_add(m, NULL, arg_default_environment);
1687         manager_set_show_status(m, arg_show_status);
1688         manager_set_first_boot(m, empty_etc);
1689
1690         /* Remember whether we should queue the default job */
1691         queue_default_job = !arg_serialization || arg_switched_root;
1692
1693         before_startup = now(CLOCK_MONOTONIC);
1694
1695         r = manager_startup(m, arg_serialization, fds);
1696         if (r < 0)
1697                 log_error("Failed to fully start up daemon: %s", strerror(-r));
1698
1699         /* This will close all file descriptors that were opened, but
1700          * not claimed by any unit. */
1701         fdset_free(fds);
1702         fds = NULL;
1703
1704         if (arg_serialization) {
1705                 fclose(arg_serialization);
1706                 arg_serialization = NULL;
1707         }
1708
1709         if (queue_default_job) {
1710                 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1711                 Unit *target = NULL;
1712                 Job *default_unit_job;
1713
1714                 log_debug("Activating default unit: %s", arg_default_unit);
1715
1716                 r = manager_load_unit(m, arg_default_unit, NULL, &error, &target);
1717                 if (r < 0)
1718                         log_error("Failed to load default target: %s", bus_error_message(&error, r));
1719                 else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND)
1720                         log_error("Failed to load default target: %s", strerror(-target->load_error));
1721                 else if (target->load_state == UNIT_MASKED)
1722                         log_error("Default target masked.");
1723
1724                 if (!target || target->load_state != UNIT_LOADED) {
1725                         log_info("Trying to load rescue target...");
1726
1727                         r = manager_load_unit(m, SPECIAL_RESCUE_TARGET, NULL, &error, &target);
1728                         if (r < 0) {
1729                                 log_error("Failed to load rescue target: %s", bus_error_message(&error, r));
1730                                 goto finish;
1731                         } else if (target->load_state == UNIT_ERROR || target->load_state == UNIT_NOT_FOUND) {
1732                                 log_error("Failed to load rescue target: %s", strerror(-target->load_error));
1733                                 goto finish;
1734                         } else if (target->load_state == UNIT_MASKED) {
1735                                 log_error("Rescue target masked.");
1736                                 goto finish;
1737                         }
1738                 }
1739
1740                 assert(target->load_state == UNIT_LOADED);
1741
1742                 if (arg_action == ACTION_TEST) {
1743                         printf("-> By units:\n");
1744                         manager_dump_units(m, stdout, "\t");
1745                 }
1746
1747                 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
1748                 if (r == -EPERM) {
1749                         log_debug("Default target could not be isolated, starting instead: %s", bus_error_message(&error, r));
1750
1751                         r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
1752                         if (r < 0) {
1753                                 log_error("Failed to start default target: %s", bus_error_message(&error, r));
1754                                 goto finish;
1755                         }
1756                 } else if (r < 0) {
1757                         log_error("Failed to isolate default target: %s", bus_error_message(&error, r));
1758                         goto finish;
1759                 }
1760
1761                 m->default_unit_job_id = default_unit_job->id;
1762
1763                 after_startup = now(CLOCK_MONOTONIC);
1764                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
1765                          "Loaded units and determined initial transaction in %s.",
1766                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 0));
1767
1768                 if (arg_action == ACTION_TEST) {
1769                         printf("-> By jobs:\n");
1770                         manager_dump_jobs(m, stdout, "\t");
1771                         retval = EXIT_SUCCESS;
1772                         goto finish;
1773                 }
1774         }
1775
1776         for (;;) {
1777                 r = manager_loop(m);
1778                 if (r < 0) {
1779                         log_error("Failed to run mainloop: %s", strerror(-r));
1780                         goto finish;
1781                 }
1782
1783                 switch (m->exit_code) {
1784
1785                 case MANAGER_EXIT:
1786                         retval = EXIT_SUCCESS;
1787                         log_debug("Exit.");
1788                         goto finish;
1789
1790                 case MANAGER_RELOAD:
1791                         log_info("Reloading.");
1792                         r = manager_reload(m);
1793                         if (r < 0)
1794                                 log_error("Failed to reload: %s", strerror(-r));
1795                         break;
1796
1797                 case MANAGER_REEXECUTE:
1798
1799                         if (prepare_reexecute(m, &arg_serialization, &fds, false) < 0)
1800                                 goto finish;
1801
1802                         reexecute = true;
1803                         log_notice("Reexecuting.");
1804                         goto finish;
1805
1806                 case MANAGER_SWITCH_ROOT:
1807                         /* Steal the switch root parameters */
1808                         switch_root_dir = m->switch_root;
1809                         switch_root_init = m->switch_root_init;
1810                         m->switch_root = m->switch_root_init = NULL;
1811
1812                         if (!switch_root_init)
1813                                 if (prepare_reexecute(m, &arg_serialization, &fds, true) < 0)
1814                                         goto finish;
1815
1816                         reexecute = true;
1817                         log_notice("Switching root.");
1818                         goto finish;
1819
1820                 case MANAGER_REBOOT:
1821                 case MANAGER_POWEROFF:
1822                 case MANAGER_HALT:
1823                 case MANAGER_KEXEC: {
1824                         static const char * const table[_MANAGER_EXIT_CODE_MAX] = {
1825                                 [MANAGER_REBOOT] = "reboot",
1826                                 [MANAGER_POWEROFF] = "poweroff",
1827                                 [MANAGER_HALT] = "halt",
1828                                 [MANAGER_KEXEC] = "kexec"
1829                         };
1830
1831                         assert_se(shutdown_verb = table[m->exit_code]);
1832                         arm_reboot_watchdog = m->exit_code == MANAGER_REBOOT;
1833
1834                         log_notice("Shutting down.");
1835                         goto finish;
1836                 }
1837
1838                 default:
1839                         assert_not_reached("Unknown exit code.");
1840                 }
1841         }
1842
1843 finish:
1844         pager_close();
1845
1846         if (m) {
1847                 manager_free(m);
1848                 m = NULL;
1849         }
1850
1851         for (j = 0; j < ELEMENTSOF(arg_default_rlimit); j++) {
1852                 free(arg_default_rlimit[j]);
1853                 arg_default_rlimit[j] = NULL;
1854         }
1855
1856         free(arg_default_unit);
1857         arg_default_unit = NULL;
1858
1859         free_join_controllers();
1860
1861         strv_free(arg_default_environment);
1862         arg_default_environment = NULL;
1863
1864         set_free(arg_syscall_archs);
1865         arg_syscall_archs = NULL;
1866
1867         label_finish();
1868
1869         if (reexecute) {
1870                 const char **args;
1871                 unsigned i, args_size;
1872                 sigset_t ss;
1873
1874                 /* Close and disarm the watchdog, so that the new
1875                  * instance can reinitialize it, but doesn't get
1876                  * rebooted while we do that */
1877                 watchdog_close(true);
1878
1879                 /* Reset the RLIMIT_NOFILE to the kernel default, so
1880                  * that the new systemd can pass the kernel default to
1881                  * its child processes */
1882                 if (saved_rlimit_nofile.rlim_cur > 0)
1883                         setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
1884
1885                 if (switch_root_dir) {
1886                         /* Kill all remaining processes from the
1887                          * initrd, but don't wait for them, so that we
1888                          * can handle the SIGCHLD for them after
1889                          * deserializing. */
1890                         broadcast_signal(SIGTERM, false, true);
1891
1892                         /* And switch root */
1893                         r = switch_root(switch_root_dir);
1894                         if (r < 0)
1895                                 log_error("Failed to switch root, ignoring: %s", strerror(-r));
1896                 }
1897
1898                 args_size = MAX(6, argc+1);
1899                 args = newa(const char*, args_size);
1900
1901                 if (!switch_root_init) {
1902                         char sfd[16];
1903
1904                         /* First try to spawn ourselves with the right
1905                          * path, and with full serialization. We do
1906                          * this only if the user didn't specify an
1907                          * explicit init to spawn. */
1908
1909                         assert(arg_serialization);
1910                         assert(fds);
1911
1912                         snprintf(sfd, sizeof(sfd), "%i", fileno(arg_serialization));
1913                         char_array_0(sfd);
1914
1915                         i = 0;
1916                         args[i++] = SYSTEMD_BINARY_PATH;
1917                         if (switch_root_dir)
1918                                 args[i++] = "--switched-root";
1919                         args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
1920                         args[i++] = "--deserialize";
1921                         args[i++] = sfd;
1922                         args[i++] = NULL;
1923
1924                         /* do not pass along the environment we inherit from the kernel or initrd */
1925                         if (switch_root_dir)
1926                                 clearenv();
1927
1928                         assert(i <= args_size);
1929                         execv(args[0], (char* const*) args);
1930                 }
1931
1932                 /* Try the fallback, if there is any, without any
1933                  * serialization. We pass the original argv[] and
1934                  * envp[]. (Well, modulo the ordering changes due to
1935                  * getopt() in argv[], and some cleanups in envp[],
1936                  * but let's hope that doesn't matter.) */
1937
1938                 if (arg_serialization) {
1939                         fclose(arg_serialization);
1940                         arg_serialization = NULL;
1941                 }
1942
1943                 if (fds) {
1944                         fdset_free(fds);
1945                         fds = NULL;
1946                 }
1947
1948                 /* Reopen the console */
1949                 make_console_stdio();
1950
1951                 for (j = 1, i = 1; j < (unsigned) argc; j++)
1952                         args[i++] = argv[j];
1953                 args[i++] = NULL;
1954                 assert(i <= args_size);
1955
1956                 /* reenable any blocked signals, especially important
1957                  * if we switch from initial ramdisk to init=... */
1958                 reset_all_signal_handlers();
1959
1960                 assert_se(sigemptyset(&ss) == 0);
1961                 assert_se(sigprocmask(SIG_SETMASK, &ss, NULL) == 0);
1962
1963                 if (switch_root_init) {
1964                         args[0] = switch_root_init;
1965                         execv(args[0], (char* const*) args);
1966                         log_warning("Failed to execute configured init, trying fallback: %m");
1967                 }
1968
1969                 args[0] = "/sbin/init";
1970                 execv(args[0], (char* const*) args);
1971
1972                 if (errno == ENOENT) {
1973                         log_warning("No /sbin/init, trying fallback");
1974
1975                         args[0] = "/bin/sh";
1976                         args[1] = NULL;
1977                         execv(args[0], (char* const*) args);
1978                         log_error("Failed to execute /bin/sh, giving up: %m");
1979                 } else
1980                         log_warning("Failed to execute /sbin/init, giving up: %m");
1981         }
1982
1983         if (arg_serialization) {
1984                 fclose(arg_serialization);
1985                 arg_serialization = NULL;
1986         }
1987
1988         if (fds) {
1989                 fdset_free(fds);
1990                 fds = NULL;
1991         }
1992
1993 #ifdef HAVE_VALGRIND_VALGRIND_H
1994         /* If we are PID 1 and running under valgrind, then let's exit
1995          * here explicitly. valgrind will only generate nice output on
1996          * exit(), not on exec(), hence let's do the former not the
1997          * latter here. */
1998         if (getpid() == 1 && RUNNING_ON_VALGRIND)
1999                 return 0;
2000 #endif
2001
2002         if (shutdown_verb) {
2003                 char log_level[DECIMAL_STR_MAX(int) + 1];
2004                 const char* command_line[9] = {
2005                         SYSTEMD_SHUTDOWN_BINARY_PATH,
2006                         shutdown_verb,
2007                         "--log-level", log_level,
2008                         "--log-target",
2009                 };
2010                 unsigned pos = 5;
2011                 _cleanup_strv_free_ char **env_block = NULL;
2012
2013                 assert(command_line[pos] == NULL);
2014                 env_block = strv_copy(environ);
2015
2016                 snprintf(log_level, sizeof(log_level), "%d", log_get_max_level());
2017
2018                 switch (log_get_target()) {
2019                 case LOG_TARGET_KMSG:
2020                 case LOG_TARGET_JOURNAL_OR_KMSG:
2021                 case LOG_TARGET_SYSLOG_OR_KMSG:
2022                         command_line[pos++] = "kmsg";
2023                         break;
2024
2025                 case LOG_TARGET_CONSOLE:
2026                 default:
2027                         command_line[pos++] = "console";
2028                         break;
2029                 };
2030
2031                 if (log_get_show_color())
2032                         command_line[pos++] = "--log-color";
2033
2034                 if (log_get_show_location())
2035                         command_line[pos++] = "--log-location";
2036
2037                 assert(pos < ELEMENTSOF(command_line));
2038
2039                 if (arm_reboot_watchdog && arg_shutdown_watchdog > 0) {
2040                         char *e;
2041
2042                         /* If we reboot let's set the shutdown
2043                          * watchdog and tell the shutdown binary to
2044                          * repeatedly ping it */
2045                         watchdog_set_timeout(&arg_shutdown_watchdog);
2046                         watchdog_close(false);
2047
2048                         /* Tell the binary how often to ping, ignore failure */
2049                         if (asprintf(&e, "WATCHDOG_USEC="USEC_FMT, arg_shutdown_watchdog) > 0)
2050                                 strv_push(&env_block, e);
2051                 } else
2052                         watchdog_close(true);
2053
2054                 /* Avoid the creation of new processes forked by the
2055                  * kernel; at this point, we will not listen to the
2056                  * signals anyway */
2057                 if (detect_container(NULL) <= 0)
2058                         cg_uninstall_release_agent(SYSTEMD_CGROUP_CONTROLLER);
2059
2060                 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
2061                 log_error("Failed to execute shutdown binary, %s: %m",
2062                           getpid() == 1 ? "freezing" : "quitting");
2063         }
2064
2065         if (getpid() == 1)
2066                 freeze();
2067
2068         return retval;
2069 }