chiark / gitweb /
3c6fccf527d4adf7f0875b1a75cebce61182efac
[elogind.git] / 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 <dbus/dbus.h>
23
24 #include <stdio.h>
25 #include <errno.h>
26 #include <string.h>
27 #include <unistd.h>
28 #include <sys/types.h>
29 #include <sys/stat.h>
30 #include <getopt.h>
31 #include <signal.h>
32 #include <sys/wait.h>
33 #include <fcntl.h>
34 #include <sys/prctl.h>
35 #include <sys/mount.h>
36
37 #include "manager.h"
38 #include "log.h"
39 #include "load-fragment.h"
40 #include "fdset.h"
41 #include "special.h"
42 #include "conf-parser.h"
43 #include "dbus-common.h"
44 #include "missing.h"
45 #include "label.h"
46 #include "build.h"
47 #include "strv.h"
48 #include "def.h"
49 #include "virt.h"
50 #include "watchdog.h"
51 #include "path-util.h"
52 #include "switch-root.h"
53 #include "capability.h"
54 #include "killall.h"
55 #include "env-util.h"
56 #include "hwclock.h"
57 #include "sd-daemon.h"
58 #include "sd-messages.h"
59
60 #include "mount-setup.h"
61 #include "loopback-setup.h"
62 #ifdef HAVE_KMOD
63 #include "kmod-setup.h"
64 #endif
65 #include "hostname-setup.h"
66 #include "machine-id-setup.h"
67 #include "locale-setup.h"
68 #include "selinux-setup.h"
69 #include "ima-setup.h"
70 #include "fileio.h"
71 #include "smack-setup.h"
72
73 static enum {
74         ACTION_RUN,
75         ACTION_HELP,
76         ACTION_VERSION,
77         ACTION_TEST,
78         ACTION_DUMP_CONFIGURATION_ITEMS,
79         ACTION_DONE
80 } arg_action = ACTION_RUN;
81
82 static char *arg_default_unit = NULL;
83 static SystemdRunningAs arg_running_as = _SYSTEMD_RUNNING_AS_INVALID;
84
85 static bool arg_dump_core = true;
86 static bool arg_crash_shell = false;
87 static int arg_crash_chvt = -1;
88 static bool arg_confirm_spawn = false;
89 static bool arg_show_status = true;
90 static bool arg_switched_root = false;
91 static char ***arg_join_controllers = NULL;
92 static ExecOutput arg_default_std_output = EXEC_OUTPUT_JOURNAL;
93 static ExecOutput arg_default_std_error = EXEC_OUTPUT_INHERIT;
94 static usec_t arg_runtime_watchdog = 0;
95 static usec_t arg_shutdown_watchdog = 10 * USEC_PER_MINUTE;
96 static char **arg_default_environment = NULL;
97 static struct rlimit *arg_default_rlimit[RLIMIT_NLIMITS] = {};
98 static uint64_t arg_capability_bounding_set_drop = 0;
99 static nsec_t arg_timer_slack_nsec = (nsec_t) -1;
100
101 static FILE* serialization = NULL;
102
103 static void nop_handler(int sig) {
104 }
105
106 _noreturn_ static void crash(int sig) {
107
108         if (getpid() != 1)
109                 /* Pass this on immediately, if this is not PID 1 */
110                 raise(sig);
111         else if (!arg_dump_core)
112                 log_error("Caught <%s>, not dumping core.", signal_to_string(sig));
113         else {
114                 struct sigaction sa = {
115                         .sa_handler = nop_handler,
116                         .sa_flags = SA_NOCLDSTOP|SA_RESTART,
117                 };
118                 pid_t pid;
119
120                 /* We want to wait for the core process, hence let's enable SIGCHLD */
121                 sigaction(SIGCHLD, &sa, NULL);
122
123                 pid = fork();
124                 if (pid < 0)
125                         log_error("Caught <%s>, cannot fork for core dump: %s", signal_to_string(sig), strerror(errno));
126
127                 else if (pid == 0) {
128                         struct rlimit rl = {};
129
130                         /* Enable default signal handler for core dump */
131                         zero(sa);
132                         sa.sa_handler = SIG_DFL;
133                         sigaction(sig, &sa, NULL);
134
135                         /* Don't limit the core dump size */
136                         rl.rlim_cur = RLIM_INFINITY;
137                         rl.rlim_max = RLIM_INFINITY;
138                         setrlimit(RLIMIT_CORE, &rl);
139
140                         /* Just to be sure... */
141                         chdir("/");
142
143                         /* Raise the signal again */
144                         raise(sig);
145
146                         assert_not_reached("We shouldn't be here...");
147                         _exit(1);
148
149                 } else {
150                         siginfo_t status;
151                         int r;
152
153                         /* Order things nicely. */
154                         r = wait_for_terminate(pid, &status);
155                         if (r < 0)
156                                 log_error("Caught <%s>, waitpid() failed: %s", signal_to_string(sig), strerror(-r));
157                         else if (status.si_code != CLD_DUMPED)
158                                 log_error("Caught <%s>, core dump failed.", signal_to_string(sig));
159                         else
160                                 log_error("Caught <%s>, dumped core as pid %lu.", signal_to_string(sig), (unsigned long) pid);
161                 }
162         }
163
164         if (arg_crash_chvt)
165                 chvt(arg_crash_chvt);
166
167         if (arg_crash_shell) {
168                 struct sigaction sa = {
169                         .sa_handler = SIG_IGN,
170                         .sa_flags = SA_NOCLDSTOP|SA_NOCLDWAIT|SA_RESTART,
171                 };
172                 pid_t pid;
173
174                 log_info("Executing crash shell in 10s...");
175                 sleep(10);
176
177                 /* Let the kernel reap children for us */
178                 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
179
180                 pid = fork();
181                 if (pid < 0)
182                         log_error("Failed to fork off crash shell: %m");
183                 else if (pid == 0) {
184                         make_console_stdio();
185                         execl("/bin/sh", "/bin/sh", NULL);
186
187                         log_error("execl() failed: %m");
188                         _exit(1);
189                 }
190
191                 log_info("Successfully spawned crash shell as pid %lu.", (unsigned long) pid);
192         }
193
194         log_info("Freezing execution.");
195         freeze();
196 }
197
198 static void install_crash_handler(void) {
199         struct sigaction sa = {
200                 .sa_handler = crash,
201                 .sa_flags = SA_NODEFER,
202         };
203
204         sigaction_many(&sa, SIGNALS_CRASH_HANDLER, -1);
205 }
206
207 static int console_setup(bool do_reset) {
208         int tty_fd, r;
209
210         /* If we are init, we connect stdin/stdout/stderr to /dev/null
211          * and make sure we don't have a controlling tty. */
212
213         release_terminal();
214
215         if (!do_reset)
216                 return 0;
217
218         tty_fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
219         if (tty_fd < 0) {
220                 log_error("Failed to open /dev/console: %s", strerror(-tty_fd));
221                 return -tty_fd;
222         }
223
224         /* We don't want to force text mode.
225          * plymouth may be showing pictures already from initrd. */
226         r = reset_terminal_fd(tty_fd, false);
227         if (r < 0)
228                 log_error("Failed to reset /dev/console: %s", strerror(-r));
229
230         close_nointr_nofail(tty_fd);
231         return r;
232 }
233
234 static int set_default_unit(const char *u) {
235         char *c;
236
237         assert(u);
238
239         c = strdup(u);
240         if (!c)
241                 return -ENOMEM;
242
243         free(arg_default_unit);
244         arg_default_unit = c;
245
246         return 0;
247 }
248
249 static int parse_proc_cmdline_word(const char *word) {
250
251         static const char * const rlmap[] = {
252                 "emergency", SPECIAL_EMERGENCY_TARGET,
253                 "-b",        SPECIAL_EMERGENCY_TARGET,
254                 "single",    SPECIAL_RESCUE_TARGET,
255                 "-s",        SPECIAL_RESCUE_TARGET,
256                 "s",         SPECIAL_RESCUE_TARGET,
257                 "S",         SPECIAL_RESCUE_TARGET,
258                 "1",         SPECIAL_RESCUE_TARGET,
259                 "2",         SPECIAL_RUNLEVEL2_TARGET,
260                 "3",         SPECIAL_RUNLEVEL3_TARGET,
261                 "4",         SPECIAL_RUNLEVEL4_TARGET,
262                 "5",         SPECIAL_RUNLEVEL5_TARGET,
263         };
264
265         assert(word);
266
267         if (startswith(word, "systemd.unit=")) {
268
269                 if (!in_initrd())
270                         return set_default_unit(word + 13);
271
272         } else if (startswith(word, "rd.systemd.unit=")) {
273
274                 if (in_initrd())
275                         return set_default_unit(word + 16);
276
277         } else if (startswith(word, "systemd.log_target=")) {
278
279                 if (log_set_target_from_string(word + 19) < 0)
280                         log_warning("Failed to parse log target %s. Ignoring.", word + 19);
281
282         } else if (startswith(word, "systemd.log_level=")) {
283
284                 if (log_set_max_level_from_string(word + 18) < 0)
285                         log_warning("Failed to parse log level %s. Ignoring.", word + 18);
286
287         } else if (startswith(word, "systemd.log_color=")) {
288
289                 if (log_show_color_from_string(word + 18) < 0)
290                         log_warning("Failed to parse log color setting %s. Ignoring.", word + 18);
291
292         } else if (startswith(word, "systemd.log_location=")) {
293
294                 if (log_show_location_from_string(word + 21) < 0)
295                         log_warning("Failed to parse log location setting %s. Ignoring.", word + 21);
296
297         } else if (startswith(word, "systemd.dump_core=")) {
298                 int r;
299
300                 if ((r = parse_boolean(word + 18)) < 0)
301                         log_warning("Failed to parse dump core switch %s. Ignoring.", word + 18);
302                 else
303                         arg_dump_core = r;
304
305         } else if (startswith(word, "systemd.crash_shell=")) {
306                 int r;
307
308                 if ((r = parse_boolean(word + 20)) < 0)
309                         log_warning("Failed to parse crash shell switch %s. Ignoring.", word + 20);
310                 else
311                         arg_crash_shell = r;
312
313         } else if (startswith(word, "systemd.confirm_spawn=")) {
314                 int r;
315
316                 if ((r = parse_boolean(word + 22)) < 0)
317                         log_warning("Failed to parse confirm spawn switch %s. Ignoring.", word + 22);
318                 else
319                         arg_confirm_spawn = r;
320
321         } else if (startswith(word, "systemd.crash_chvt=")) {
322                 int k;
323
324                 if (safe_atoi(word + 19, &k) < 0)
325                         log_warning("Failed to parse crash chvt switch %s. Ignoring.", word + 19);
326                 else
327                         arg_crash_chvt = k;
328
329         } else if (startswith(word, "systemd.show_status=")) {
330                 int r;
331
332                 if ((r = parse_boolean(word + 20)) < 0)
333                         log_warning("Failed to parse show status switch %s. Ignoring.", word + 20);
334                 else
335                         arg_show_status = r;
336         } else if (startswith(word, "systemd.default_standard_output=")) {
337                 int r;
338
339                 if ((r = exec_output_from_string(word + 32)) < 0)
340                         log_warning("Failed to parse default standard output switch %s. Ignoring.", word + 32);
341                 else
342                         arg_default_std_output = r;
343         } else if (startswith(word, "systemd.default_standard_error=")) {
344                 int r;
345
346                 if ((r = exec_output_from_string(word + 31)) < 0)
347                         log_warning("Failed to parse default standard error switch %s. Ignoring.", word + 31);
348                 else
349                         arg_default_std_error = r;
350         } else if (startswith(word, "systemd.setenv=")) {
351                 _cleanup_free_ char *cenv = NULL;
352                 char *eq;
353                 int r;
354
355                 cenv = strdup(word + 15);
356                 if (!cenv)
357                         return -ENOMEM;
358
359                 eq = strchr(cenv, '=');
360                 if (!eq) {
361                         if (!env_name_is_valid(cenv))
362                                 log_warning("Environment variable name '%s' is not valid. Ignoring.", cenv);
363                         else  {
364                                 r = unsetenv(cenv);
365                                 if (r < 0)
366                                         log_warning("Unsetting environment variable '%s' failed, ignoring: %m", cenv);
367                         }
368                 } else {
369                         if (!env_assignment_is_valid(cenv))
370                                 log_warning("Environment variable assignment '%s' is not valid. Ignoring.", cenv);
371                         else {
372                                 *eq = 0;
373                                 r = setenv(cenv, eq + 1, 1);
374                                 if (r < 0)
375                                         log_warning("Setting environment variable '%s=%s' failed, ignoring: %m", cenv, eq + 1);
376                         }
377                 }
378
379         } else if (startswith(word, "systemd.") ||
380                    (in_initrd() && startswith(word, "rd.systemd."))) {
381
382                 const char *c;
383
384                 /* Ignore systemd.journald.xyz and friends */
385                 c = word;
386                 if (startswith(c, "rd."))
387                         c += 3;
388                 if (startswith(c, "systemd."))
389                         c += 8;
390                 if (c[strcspn(c, ".=")] != '.')  {
391
392                         log_warning("Unknown kernel switch %s. Ignoring.", word);
393
394                         log_info("Supported kernel switches:\n"
395                                  "systemd.unit=UNIT                        Default unit to start\n"
396                                  "rd.systemd.unit=UNIT                     Default unit to start when run in initrd\n"
397                                  "systemd.dump_core=0|1                    Dump core on crash\n"
398                                  "systemd.crash_shell=0|1                  Run shell on crash\n"
399                                  "systemd.crash_chvt=N                     Change to VT #N on crash\n"
400                                  "systemd.confirm_spawn=0|1                Confirm every process spawn\n"
401                                  "systemd.show_status=0|1                  Show status updates on the console during bootup\n"
402                                  "systemd.log_target=console|kmsg|journal|journal-or-kmsg|syslog|syslog-or-kmsg|null\n"
403                                  "                                         Log target\n"
404                                  "systemd.log_level=LEVEL                  Log level\n"
405                                  "systemd.log_color=0|1                    Highlight important log messages\n"
406                                  "systemd.log_location=0|1                 Include code location in log messages\n"
407                                  "systemd.default_standard_output=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
408                                  "                                         Set default log output for services\n"
409                                  "systemd.default_standard_error=null|tty|syslog|syslog+console|kmsg|kmsg+console|journal|journal+console\n"
410                                  "                                         Set default log error output for services\n"
411                                  "systemd.setenv=ASSIGNMENT                Set an environment variable for all spawned processes\n");
412                 }
413
414         } else if (streq(word, "quiet"))
415                 arg_show_status = false;
416         else if (streq(word, "debug"))
417                 log_set_max_level(LOG_DEBUG);
418         else if (!in_initrd()) {
419                 unsigned i;
420
421                 /* SysV compatibility */
422                 for (i = 0; i < ELEMENTSOF(rlmap); i += 2)
423                         if (streq(word, rlmap[i]))
424                                 return set_default_unit(rlmap[i+1]);
425         }
426
427         return 0;
428 }
429
430 #define DEFINE_SETTER(name, func, descr)                              \
431         static int name(const char *unit,                             \
432                         const char *filename,                         \
433                         unsigned line,                                \
434                         const char *section,                          \
435                         const char *lvalue,                           \
436                         int ltype,                                    \
437                         const char *rvalue,                           \
438                         void *data,                                   \
439                         void *userdata) {                             \
440                                                                       \
441                 int r;                                                \
442                                                                       \
443                 assert(filename);                                     \
444                 assert(lvalue);                                       \
445                 assert(rvalue);                                       \
446                                                                       \
447                 r = func(rvalue);                                     \
448                 if (r < 0)                                            \
449                         log_syntax(unit, LOG_ERR, filename, line, -r, \
450                                    "Invalid " descr "'%s': %s",       \
451                                    rvalue, strerror(-r));             \
452                                                                       \
453                 return 0;                                             \
454         }
455
456 DEFINE_SETTER(config_parse_level2, log_set_max_level_from_string, "log level")
457 DEFINE_SETTER(config_parse_target, log_set_target_from_string, "target")
458 DEFINE_SETTER(config_parse_color, log_show_color_from_string, "color" )
459 DEFINE_SETTER(config_parse_location, log_show_location_from_string, "location")
460
461
462 static int config_parse_cpu_affinity2(const char *unit,
463                                       const char *filename,
464                                       unsigned line,
465                                       const char *section,
466                                       const char *lvalue,
467                                       int ltype,
468                                       const char *rvalue,
469                                       void *data,
470                                       void *userdata) {
471
472         char *w;
473         size_t l;
474         char *state;
475         cpu_set_t *c = NULL;
476         unsigned ncpus = 0;
477
478         assert(filename);
479         assert(lvalue);
480         assert(rvalue);
481
482         FOREACH_WORD_QUOTED(w, l, rvalue, state) {
483                 char *t;
484                 int r;
485                 unsigned cpu;
486
487                 if (!(t = strndup(w, l)))
488                         return log_oom();
489
490                 r = safe_atou(t, &cpu);
491                 free(t);
492
493                 if (!c)
494                         if (!(c = cpu_set_malloc(&ncpus)))
495                                 return log_oom();
496
497                 if (r < 0 || cpu >= ncpus) {
498                         log_syntax(unit, LOG_ERR, filename, line, -r,
499                                    "Failed to parse CPU affinity '%s'", rvalue);
500                         CPU_FREE(c);
501                         return -EBADMSG;
502                 }
503
504                 CPU_SET_S(cpu, CPU_ALLOC_SIZE(ncpus), c);
505         }
506
507         if (c) {
508                 if (sched_setaffinity(0, CPU_ALLOC_SIZE(ncpus), c) < 0)
509                         log_warning_unit(unit, "Failed to set CPU affinity: %m");
510
511                 CPU_FREE(c);
512         }
513
514         return 0;
515 }
516
517 static void strv_free_free(char ***l) {
518         char ***i;
519
520         if (!l)
521                 return;
522
523         for (i = l; *i; i++)
524                 strv_free(*i);
525
526         free(l);
527 }
528
529 static void free_join_controllers(void) {
530         strv_free_free(arg_join_controllers);
531         arg_join_controllers = NULL;
532 }
533
534 static int config_parse_join_controllers(const char *unit,
535                                          const char *filename,
536                                          unsigned line,
537                                          const char *section,
538                                          const char *lvalue,
539                                          int ltype,
540                                          const char *rvalue,
541                                          void *data,
542                                          void *userdata) {
543
544         unsigned n = 0;
545         char *state, *w;
546         size_t length;
547
548         assert(filename);
549         assert(lvalue);
550         assert(rvalue);
551
552         free_join_controllers();
553
554         FOREACH_WORD_QUOTED(w, length, rvalue, state) {
555                 char *s, **l;
556
557                 s = strndup(w, length);
558                 if (!s)
559                         return log_oom();
560
561                 l = strv_split(s, ",");
562                 free(s);
563
564                 strv_uniq(l);
565
566                 if (strv_length(l) <= 1) {
567                         strv_free(l);
568                         continue;
569                 }
570
571                 if (!arg_join_controllers) {
572                         arg_join_controllers = new(char**, 2);
573                         if (!arg_join_controllers) {
574                                 strv_free(l);
575                                 return log_oom();
576                         }
577
578                         arg_join_controllers[0] = l;
579                         arg_join_controllers[1] = NULL;
580
581                         n = 1;
582                 } else {
583                         char ***a;
584                         char ***t;
585
586                         t = new0(char**, n+2);
587                         if (!t) {
588                                 strv_free(l);
589                                 return log_oom();
590                         }
591
592                         n = 0;
593
594                         for (a = arg_join_controllers; *a; a++) {
595
596                                 if (strv_overlap(*a, l)) {
597                                         char **c;
598
599                                         c = strv_merge(*a, l);
600                                         if (!c) {
601                                                 strv_free(l);
602                                                 strv_free_free(t);
603                                                 return log_oom();
604                                         }
605
606                                         strv_free(l);
607                                         l = c;
608                                 } else {
609                                         char **c;
610
611                                         c = strv_copy(*a);
612                                         if (!c) {
613                                                 strv_free(l);
614                                                 strv_free_free(t);
615                                                 return log_oom();
616                                         }
617
618                                         t[n++] = c;
619                                 }
620                         }
621
622                         t[n++] = strv_uniq(l);
623
624                         strv_free_free(arg_join_controllers);
625                         arg_join_controllers = t;
626                 }
627         }
628
629         return 0;
630 }
631
632 static int parse_config_file(void) {
633
634         const ConfigTableItem items[] = {
635                 { "Manager", "LogLevel",              config_parse_level2,       0, NULL                     },
636                 { "Manager", "LogTarget",             config_parse_target,       0, NULL                     },
637                 { "Manager", "LogColor",              config_parse_color,        0, NULL                     },
638                 { "Manager", "LogLocation",           config_parse_location,     0, NULL                     },
639                 { "Manager", "DumpCore",              config_parse_bool,         0, &arg_dump_core           },
640                 { "Manager", "CrashShell",            config_parse_bool,         0, &arg_crash_shell         },
641                 { "Manager", "ShowStatus",            config_parse_bool,         0, &arg_show_status         },
642                 { "Manager", "CrashChVT",             config_parse_int,          0, &arg_crash_chvt          },
643                 { "Manager", "CPUAffinity",           config_parse_cpu_affinity2, 0, NULL                    },
644                 { "Manager", "DefaultStandardOutput", config_parse_output,       0, &arg_default_std_output  },
645                 { "Manager", "DefaultStandardError",  config_parse_output,       0, &arg_default_std_error   },
646                 { "Manager", "JoinControllers",       config_parse_join_controllers, 0, &arg_join_controllers },
647                 { "Manager", "RuntimeWatchdogSec",    config_parse_sec,          0, &arg_runtime_watchdog    },
648                 { "Manager", "ShutdownWatchdogSec",   config_parse_sec,          0, &arg_shutdown_watchdog   },
649                 { "Manager", "CapabilityBoundingSet", config_parse_bounding_set, 0, &arg_capability_bounding_set_drop },
650                 { "Manager", "TimerSlackNSec",        config_parse_nsec,         0, &arg_timer_slack_nsec    },
651                 { "Manager", "DefaultEnvironment",    config_parse_environ,      0, &arg_default_environment },
652                 { "Manager", "DefaultLimitCPU",       config_parse_limit,        0, &arg_default_rlimit[RLIMIT_CPU]},
653                 { "Manager", "DefaultLimitFSIZE",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_FSIZE]},
654                 { "Manager", "DefaultLimitDATA",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_DATA]},
655                 { "Manager", "DefaultLimitSTACK",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_STACK]},
656                 { "Manager", "DefaultLimitCORE",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_CORE]},
657                 { "Manager", "DefaultLimitRSS",       config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RSS]},
658                 { "Manager", "DefaultLimitNOFILE",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NOFILE]},
659                 { "Manager", "DefaultLimitAS",        config_parse_limit,        0, &arg_default_rlimit[RLIMIT_AS]},
660                 { "Manager", "DefaultLimitNPROC",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NPROC]},
661                 { "Manager", "DefaultLimitMEMLOCK",   config_parse_limit,        0, &arg_default_rlimit[RLIMIT_MEMLOCK]},
662                 { "Manager", "DefaultLimitLOCKS",     config_parse_limit,        0, &arg_default_rlimit[RLIMIT_LOCKS]},
663                 { "Manager", "DefaultLimitSIGPENDING",config_parse_limit,        0, &arg_default_rlimit[RLIMIT_SIGPENDING]},
664                 { "Manager", "DefaultLimitMSGQUEUE",  config_parse_limit,        0, &arg_default_rlimit[RLIMIT_MSGQUEUE]},
665                 { "Manager", "DefaultLimitNICE",      config_parse_limit,        0, &arg_default_rlimit[RLIMIT_NICE]},
666                 { "Manager", "DefaultLimitRTPRIO",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RTPRIO]},
667                 { "Manager", "DefaultLimitRTTIME",    config_parse_limit,        0, &arg_default_rlimit[RLIMIT_RTTIME]},
668                 { NULL, NULL, NULL, 0, NULL }
669         };
670
671         _cleanup_fclose_ FILE *f;
672         const char *fn;
673         int r;
674
675         fn = arg_running_as == SYSTEMD_SYSTEM ? PKGSYSCONFDIR "/system.conf" : PKGSYSCONFDIR "/user.conf";
676         f = fopen(fn, "re");
677         if (!f) {
678                 if (errno == ENOENT)
679                         return 0;
680
681                 log_warning("Failed to open configuration file '%s': %m", fn);
682                 return 0;
683         }
684
685         r = config_parse(NULL, fn, f, "Manager\0", config_item_table_lookup, (void*) items, false, false, NULL);
686         if (r < 0)
687                 log_warning("Failed to parse configuration file: %s", strerror(-r));
688
689         return 0;
690 }
691
692 static int parse_proc_cmdline(void) {
693         _cleanup_free_ char *line = NULL;
694         char *w, *state;
695         int r;
696         size_t l;
697
698         /* Don't read /proc/cmdline if we are in a container, since
699          * that is only relevant for the host system */
700         if (detect_container(NULL) > 0)
701                 return 0;
702
703         r = read_one_line_file("/proc/cmdline", &line);
704         if (r < 0) {
705                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
706                 return 0;
707         }
708
709         FOREACH_WORD_QUOTED(w, l, line, state) {
710                 _cleanup_free_ char *word;
711
712                 word = strndup(w, l);
713                 if (!word)
714                         return log_oom();
715
716                 r = parse_proc_cmdline_word(word);
717                 if (r < 0) {
718                         log_error("Failed on cmdline argument %s: %s", word, strerror(-r));
719                         return r;
720                 }
721         }
722
723         return 0;
724 }
725
726 static int parse_argv(int argc, char *argv[]) {
727
728         enum {
729                 ARG_LOG_LEVEL = 0x100,
730                 ARG_LOG_TARGET,
731                 ARG_LOG_COLOR,
732                 ARG_LOG_LOCATION,
733                 ARG_UNIT,
734                 ARG_SYSTEM,
735                 ARG_USER,
736                 ARG_TEST,
737                 ARG_VERSION,
738                 ARG_DUMP_CONFIGURATION_ITEMS,
739                 ARG_DUMP_CORE,
740                 ARG_CRASH_SHELL,
741                 ARG_CONFIRM_SPAWN,
742                 ARG_SHOW_STATUS,
743                 ARG_DESERIALIZE,
744                 ARG_SWITCHED_ROOT,
745                 ARG_INTROSPECT,
746                 ARG_DEFAULT_STD_OUTPUT,
747                 ARG_DEFAULT_STD_ERROR
748         };
749
750         static const struct option options[] = {
751                 { "log-level",                required_argument, NULL, ARG_LOG_LEVEL                },
752                 { "log-target",               required_argument, NULL, ARG_LOG_TARGET               },
753                 { "log-color",                optional_argument, NULL, ARG_LOG_COLOR                },
754                 { "log-location",             optional_argument, NULL, ARG_LOG_LOCATION             },
755                 { "unit",                     required_argument, NULL, ARG_UNIT                     },
756                 { "system",                   no_argument,       NULL, ARG_SYSTEM                   },
757                 { "user",                     no_argument,       NULL, ARG_USER                     },
758                 { "test",                     no_argument,       NULL, ARG_TEST                     },
759                 { "help",                     no_argument,       NULL, 'h'                          },
760                 { "version",                  no_argument,       NULL, ARG_VERSION                  },
761                 { "dump-configuration-items", no_argument,       NULL, ARG_DUMP_CONFIGURATION_ITEMS },
762                 { "dump-core",                optional_argument, NULL, ARG_DUMP_CORE                },
763                 { "crash-shell",              optional_argument, NULL, ARG_CRASH_SHELL              },
764                 { "confirm-spawn",            optional_argument, NULL, ARG_CONFIRM_SPAWN            },
765                 { "show-status",              optional_argument, NULL, ARG_SHOW_STATUS              },
766                 { "deserialize",              required_argument, NULL, ARG_DESERIALIZE              },
767                 { "switched-root",            no_argument,       NULL, ARG_SWITCHED_ROOT            },
768                 { "introspect",               optional_argument, NULL, ARG_INTROSPECT               },
769                 { "default-standard-output",  required_argument, NULL, ARG_DEFAULT_STD_OUTPUT,      },
770                 { "default-standard-error",   required_argument, NULL, ARG_DEFAULT_STD_ERROR,       },
771                 { NULL,                       0,                 NULL, 0                            }
772         };
773
774         int c, r;
775
776         assert(argc >= 1);
777         assert(argv);
778
779         if (getpid() == 1)
780                 opterr = 0;
781
782         while ((c = getopt_long(argc, argv, "hDbsz:", options, NULL)) >= 0)
783
784                 switch (c) {
785
786                 case ARG_LOG_LEVEL:
787                         if ((r = log_set_max_level_from_string(optarg)) < 0) {
788                                 log_error("Failed to parse log level %s.", optarg);
789                                 return r;
790                         }
791
792                         break;
793
794                 case ARG_LOG_TARGET:
795
796                         if ((r = log_set_target_from_string(optarg)) < 0) {
797                                 log_error("Failed to parse log target %s.", optarg);
798                                 return r;
799                         }
800
801                         break;
802
803                 case ARG_LOG_COLOR:
804
805                         if (optarg) {
806                                 if ((r = log_show_color_from_string(optarg)) < 0) {
807                                         log_error("Failed to parse log color setting %s.", optarg);
808                                         return r;
809                                 }
810                         } else
811                                 log_show_color(true);
812
813                         break;
814
815                 case ARG_LOG_LOCATION:
816
817                         if (optarg) {
818                                 if ((r = log_show_location_from_string(optarg)) < 0) {
819                                         log_error("Failed to parse log location setting %s.", optarg);
820                                         return r;
821                                 }
822                         } else
823                                 log_show_location(true);
824
825                         break;
826
827                 case ARG_DEFAULT_STD_OUTPUT:
828
829                         if ((r = exec_output_from_string(optarg)) < 0) {
830                                 log_error("Failed to parse default standard output setting %s.", optarg);
831                                 return r;
832                         } else
833                                 arg_default_std_output = r;
834                         break;
835
836                 case ARG_DEFAULT_STD_ERROR:
837
838                         if ((r = exec_output_from_string(optarg)) < 0) {
839                                 log_error("Failed to parse default standard error output setting %s.", optarg);
840                                 return r;
841                         } else
842                                 arg_default_std_error = r;
843                         break;
844
845                 case ARG_UNIT:
846
847                         if ((r = set_default_unit(optarg)) < 0) {
848                                 log_error("Failed to set default unit %s: %s", optarg, strerror(-r));
849                                 return r;
850                         }
851
852                         break;
853
854                 case ARG_SYSTEM:
855                         arg_running_as = SYSTEMD_SYSTEM;
856                         break;
857
858                 case ARG_USER:
859                         arg_running_as = SYSTEMD_USER;
860                         break;
861
862                 case ARG_TEST:
863                         arg_action = ACTION_TEST;
864                         break;
865
866                 case ARG_VERSION:
867                         arg_action = ACTION_VERSION;
868                         break;
869
870                 case ARG_DUMP_CONFIGURATION_ITEMS:
871                         arg_action = ACTION_DUMP_CONFIGURATION_ITEMS;
872                         break;
873
874                 case ARG_DUMP_CORE:
875                         r = optarg ? parse_boolean(optarg) : 1;
876                         if (r < 0) {
877                                 log_error("Failed to parse dump core boolean %s.", optarg);
878                                 return r;
879                         }
880                         arg_dump_core = r;
881                         break;
882
883                 case ARG_CRASH_SHELL:
884                         r = optarg ? parse_boolean(optarg) : 1;
885                         if (r < 0) {
886                                 log_error("Failed to parse crash shell boolean %s.", optarg);
887                                 return r;
888                         }
889                         arg_crash_shell = r;
890                         break;
891
892                 case ARG_CONFIRM_SPAWN:
893                         r = optarg ? parse_boolean(optarg) : 1;
894                         if (r < 0) {
895                                 log_error("Failed to parse confirm spawn boolean %s.", optarg);
896                                 return r;
897                         }
898                         arg_confirm_spawn = r;
899                         break;
900
901                 case ARG_SHOW_STATUS:
902                         r = optarg ? parse_boolean(optarg) : 1;
903                         if (r < 0) {
904                                 log_error("Failed to parse show status boolean %s.", optarg);
905                                 return r;
906                         }
907                         arg_show_status = r;
908                         break;
909
910                 case ARG_DESERIALIZE: {
911                         int fd;
912                         FILE *f;
913
914                         r = safe_atoi(optarg, &fd);
915                         if (r < 0 || fd < 0) {
916                                 log_error("Failed to parse deserialize option %s.", optarg);
917                                 return r < 0 ? r : -EINVAL;
918                         }
919
920                         fd_cloexec(fd, true);
921
922                         f = fdopen(fd, "r");
923                         if (!f) {
924                                 log_error("Failed to open serialization fd: %m");
925                                 return -errno;
926                         }
927
928                         if (serialization)
929                                 fclose(serialization);
930
931                         serialization = f;
932
933                         break;
934                 }
935
936                 case ARG_SWITCHED_ROOT:
937                         arg_switched_root = true;
938                         break;
939
940                 case ARG_INTROSPECT: {
941                         const char * const * i = NULL;
942
943                         for (i = bus_interface_table; *i; i += 2)
944                                 if (!optarg || streq(i[0], optarg)) {
945                                         fputs(DBUS_INTROSPECT_1_0_XML_DOCTYPE_DECL_NODE
946                                               "<node>\n", stdout);
947                                         fputs(i[1], stdout);
948                                         fputs("</node>\n", stdout);
949
950                                         if (optarg)
951                                                 break;
952                                 }
953
954                         if (!i[0] && optarg)
955                                 log_error("Unknown interface %s.", optarg);
956
957                         arg_action = ACTION_DONE;
958                         break;
959                 }
960
961                 case 'h':
962                         arg_action = ACTION_HELP;
963                         break;
964
965                 case 'D':
966                         log_set_max_level(LOG_DEBUG);
967                         break;
968
969                 case 'b':
970                 case 's':
971                 case 'z':
972                         /* Just to eat away the sysvinit kernel
973                          * cmdline args without getopt() error
974                          * messages that we'll parse in
975                          * parse_proc_cmdline_word() or ignore. */
976
977                 case '?':
978                 default:
979                         if (getpid() != 1) {
980                                 log_error("Unknown option code %c", c);
981                                 return -EINVAL;
982                         }
983
984                         break;
985                 }
986
987         if (optind < argc && getpid() != 1) {
988                 /* Hmm, when we aren't run as init system
989                  * let's complain about excess arguments */
990
991                 log_error("Excess arguments.");
992                 return -EINVAL;
993         }
994
995         if (detect_container(NULL) > 0) {
996                 char **a;
997
998                 /* All /proc/cmdline arguments the kernel didn't
999                  * understand it passed to us. We're not really
1000                  * interested in that usually since /proc/cmdline is
1001                  * more interesting and complete. With one exception:
1002                  * if we are run in a container /proc/cmdline is not
1003                  * relevant for the container, hence we rely on argv[]
1004                  * instead. */
1005
1006                 for (a = argv; a < argv + argc; a++)
1007                         if ((r = parse_proc_cmdline_word(*a)) < 0) {
1008                                 log_error("Failed on cmdline argument %s: %s", *a, strerror(-r));
1009                                 return r;
1010                         }
1011         }
1012
1013         return 0;
1014 }
1015
1016 static int help(void) {
1017
1018         printf("%s [OPTIONS...]\n\n"
1019                "Starts up and maintains the system or user services.\n\n"
1020                "  -h --help                      Show this help\n"
1021                "     --test                      Determine startup sequence, dump it and exit\n"
1022                "     --dump-configuration-items  Dump understood unit configuration items\n"
1023                "     --introspect[=INTERFACE]    Extract D-Bus interface data\n"
1024                "     --unit=UNIT                 Set default unit\n"
1025                "     --system                    Run a system instance, even if PID != 1\n"
1026                "     --user                      Run a user instance\n"
1027                "     --dump-core[=0|1]           Dump core on crash\n"
1028                "     --crash-shell[=0|1]         Run shell on crash\n"
1029                "     --confirm-spawn[=0|1]       Ask for confirmation when spawning processes\n"
1030                "     --show-status[=0|1]         Show status updates on the console during bootup\n"
1031                "     --log-target=TARGET         Set log target (console, journal, syslog, kmsg, journal-or-kmsg, syslog-or-kmsg, null)\n"
1032                "     --log-level=LEVEL           Set log level (debug, info, notice, warning, err, crit, alert, emerg)\n"
1033                "     --log-color[=0|1]           Highlight important log messages\n"
1034                "     --log-location[=0|1]        Include code location in log messages\n"
1035                "     --default-standard-output=  Set default standard output for services\n"
1036                "     --default-standard-error=   Set default standard error output for services\n",
1037                program_invocation_short_name);
1038
1039         return 0;
1040 }
1041
1042 static int version(void) {
1043         puts(PACKAGE_STRING);
1044         puts(SYSTEMD_FEATURES);
1045
1046         return 0;
1047 }
1048
1049 static int prepare_reexecute(Manager *m, FILE **_f, FDSet **_fds, bool switching_root) {
1050         FILE *f = NULL;
1051         FDSet *fds = NULL;
1052         int r;
1053
1054         assert(m);
1055         assert(_f);
1056         assert(_fds);
1057
1058         /* Make sure nothing is really destructed when we shut down */
1059         m->n_reloading ++;
1060
1061         r = manager_open_serialization(m, &f);
1062         if (r < 0) {
1063                 log_error("Failed to create serialization file: %s", strerror(-r));
1064                 goto fail;
1065         }
1066
1067         fds = fdset_new();
1068         if (!fds) {
1069                 r = -ENOMEM;
1070                 log_error("Failed to allocate fd set: %s", strerror(-r));
1071                 goto fail;
1072         }
1073
1074         r = manager_serialize(m, f, fds, switching_root);
1075         if (r < 0) {
1076                 log_error("Failed to serialize state: %s", strerror(-r));
1077                 goto fail;
1078         }
1079
1080         if (fseeko(f, 0, SEEK_SET) < 0) {
1081                 log_error("Failed to rewind serialization fd: %m");
1082                 goto fail;
1083         }
1084
1085         r = fd_cloexec(fileno(f), false);
1086         if (r < 0) {
1087                 log_error("Failed to disable O_CLOEXEC for serialization: %s", strerror(-r));
1088                 goto fail;
1089         }
1090
1091         r = fdset_cloexec(fds, false);
1092         if (r < 0) {
1093                 log_error("Failed to disable O_CLOEXEC for serialization fds: %s", strerror(-r));
1094                 goto fail;
1095         }
1096
1097         *_f = f;
1098         *_fds = fds;
1099
1100         return 0;
1101
1102 fail:
1103         fdset_free(fds);
1104
1105         if (f)
1106                 fclose(f);
1107
1108         return r;
1109 }
1110
1111 static int bump_rlimit_nofile(struct rlimit *saved_rlimit) {
1112         struct rlimit nl;
1113         int r;
1114
1115         assert(saved_rlimit);
1116
1117         /* Save the original RLIMIT_NOFILE so that we can reset it
1118          * later when transitioning from the initrd to the main
1119          * systemd or suchlike. */
1120         if (getrlimit(RLIMIT_NOFILE, saved_rlimit) < 0) {
1121                 log_error("Reading RLIMIT_NOFILE failed: %m");
1122                 return -errno;
1123         }
1124
1125         /* Make sure forked processes get the default kernel setting */
1126         if (!arg_default_rlimit[RLIMIT_NOFILE]) {
1127                 struct rlimit *rl;
1128
1129                 rl = newdup(struct rlimit, saved_rlimit, 1);
1130                 if (!rl)
1131                         return log_oom();
1132
1133                 arg_default_rlimit[RLIMIT_NOFILE] = rl;
1134         }
1135
1136         /* Bump up the resource limit for ourselves substantially */
1137         nl.rlim_cur = nl.rlim_max = 64*1024;
1138         r = setrlimit_closest(RLIMIT_NOFILE, &nl);
1139         if (r < 0) {
1140                 log_error("Setting RLIMIT_NOFILE failed: %s", strerror(-r));
1141                 return r;
1142         }
1143
1144         return 0;
1145 }
1146
1147 static struct dual_timestamp* parse_initrd_timestamp(struct dual_timestamp *t) {
1148         const char *e;
1149         unsigned long long a, b;
1150
1151         assert(t);
1152
1153         e = getenv("RD_TIMESTAMP");
1154         if (!e)
1155                 return NULL;
1156
1157         if (sscanf(e, "%llu %llu", &a, &b) != 2)
1158                 return NULL;
1159
1160         t->realtime = (usec_t) a;
1161         t->monotonic = (usec_t) b;
1162
1163         return t;
1164 }
1165
1166 static void test_mtab(void) {
1167         char *p;
1168
1169         /* Check that /etc/mtab is a symlink */
1170
1171         if (readlink_malloc("/etc/mtab", &p) >= 0) {
1172                 bool b;
1173
1174                 b = streq(p, "/proc/self/mounts") || streq(p, "/proc/mounts");
1175                 free(p);
1176
1177                 if (b)
1178                         return;
1179         }
1180
1181         log_warning("/etc/mtab is not a symlink or not pointing to /proc/self/mounts. "
1182                     "This is not supported anymore. "
1183                     "Please make sure to replace this file by a symlink to avoid incorrect or misleading mount(8) output.");
1184 }
1185
1186 static void test_usr(void) {
1187
1188         /* Check that /usr is not a separate fs */
1189
1190         if (dir_is_empty("/usr") <= 0)
1191                 return;
1192
1193         log_warning("/usr appears to be on its own filesytem and is not already mounted. This is not a supported setup. "
1194                     "Some things will probably break (sometimes even silently) in mysterious ways. "
1195                     "Consult http://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken for more information.");
1196 }
1197
1198 static void test_cgroups(void) {
1199
1200         if (access("/proc/cgroups", F_OK) >= 0)
1201                 return;
1202
1203         log_warning("CONFIG_CGROUPS was not set when your kernel was compiled. "
1204                     "Systems without control groups are not supported. "
1205                     "We will now sleep for 10s, and then continue boot-up. "
1206                     "Expect breakage and please do not file bugs. "
1207                     "Instead fix your kernel and enable CONFIG_CGROUPS. "
1208                     "Consult http://0pointer.de/blog/projects/cgroups-vs-cgroups.html for more information.");
1209
1210         sleep(10);
1211 }
1212
1213 static int initialize_join_controllers(void) {
1214         /* By default, mount "cpu" + "cpuacct" together, and "net_cls"
1215          * + "net_prio". We'd like to add "cpuset" to the mix, but
1216          * "cpuset" does't really work for groups with no initialized
1217          * attributes. */
1218
1219         arg_join_controllers = new(char**, 3);
1220         if (!arg_join_controllers)
1221                 return -ENOMEM;
1222
1223         arg_join_controllers[0] = strv_new("cpu", "cpuacct", NULL);
1224         arg_join_controllers[1] = strv_new("net_cls", "net_prio", NULL);
1225         arg_join_controllers[2] = NULL;
1226
1227         if (!arg_join_controllers[0] || !arg_join_controllers[1]) {
1228                 free_join_controllers();
1229                 return -ENOMEM;
1230         }
1231
1232         return 0;
1233 }
1234
1235 int main(int argc, char *argv[]) {
1236         Manager *m = NULL;
1237         int r, retval = EXIT_FAILURE;
1238         usec_t before_startup, after_startup;
1239         char timespan[FORMAT_TIMESPAN_MAX];
1240         FDSet *fds = NULL;
1241         bool reexecute = false;
1242         const char *shutdown_verb = NULL;
1243         dual_timestamp initrd_timestamp = { 0ULL, 0ULL };
1244         dual_timestamp userspace_timestamp = { 0ULL, 0ULL };
1245         dual_timestamp kernel_timestamp = { 0ULL, 0ULL };
1246         static char systemd[] = "systemd";
1247         bool skip_setup = false;
1248         int j;
1249         bool loaded_policy = false;
1250         bool arm_reboot_watchdog = false;
1251         bool queue_default_job = false;
1252         char *switch_root_dir = NULL, *switch_root_init = NULL;
1253         static struct rlimit saved_rlimit_nofile = { 0, 0 };
1254
1255 #ifdef HAVE_SYSV_COMPAT
1256         if (getpid() != 1 && strstr(program_invocation_short_name, "init")) {
1257                 /* This is compatibility support for SysV, where
1258                  * calling init as a user is identical to telinit. */
1259
1260                 errno = -ENOENT;
1261                 execv(SYSTEMCTL_BINARY_PATH, argv);
1262                 log_error("Failed to exec " SYSTEMCTL_BINARY_PATH ": %m");
1263                 return 1;
1264         }
1265 #endif
1266
1267         dual_timestamp_from_monotonic(&kernel_timestamp, 0);
1268         dual_timestamp_get(&userspace_timestamp);
1269
1270         /* Determine if this is a reexecution or normal bootup. We do
1271          * the full command line parsing much later, so let's just
1272          * have a quick peek here. */
1273         if (strv_find(argv+1, "--deserialize"))
1274                 skip_setup = true;
1275
1276         /* If we have switched root, do all the special setup
1277          * things */
1278         if (strv_find(argv+1, "--switched-root"))
1279                 skip_setup = false;
1280
1281         /* If we get started via the /sbin/init symlink then we are
1282            called 'init'. After a subsequent reexecution we are then
1283            called 'systemd'. That is confusing, hence let's call us
1284            systemd right-away. */
1285         program_invocation_short_name = systemd;
1286         prctl(PR_SET_NAME, systemd);
1287
1288         saved_argv = argv;
1289         saved_argc = argc;
1290
1291         log_show_color(isatty(STDERR_FILENO) > 0);
1292
1293         if (getpid() == 1 && detect_container(NULL) <= 0) {
1294
1295                 /* Running outside of a container as PID 1 */
1296                 arg_running_as = SYSTEMD_SYSTEM;
1297                 make_null_stdio();
1298                 log_set_target(LOG_TARGET_KMSG);
1299                 log_open();
1300
1301                 if (in_initrd()) {
1302                         char *rd_timestamp = NULL;
1303
1304                         initrd_timestamp = userspace_timestamp;
1305                         asprintf(&rd_timestamp, "%llu %llu",
1306                                  (unsigned long long) initrd_timestamp.realtime,
1307                                  (unsigned long long) initrd_timestamp.monotonic);
1308                         if (rd_timestamp) {
1309                                 setenv("RD_TIMESTAMP", rd_timestamp, 1);
1310                                 free(rd_timestamp);
1311                         }
1312                 }
1313
1314                 if (!skip_setup) {
1315                         mount_setup_early();
1316                         if (selinux_setup(&loaded_policy) < 0)
1317                                 goto finish;
1318                         if (ima_setup() < 0)
1319                                 goto finish;
1320                         if (smack_setup() < 0)
1321                                 goto finish;
1322                 }
1323
1324                 if (label_init(NULL) < 0)
1325                         goto finish;
1326
1327                 if (!skip_setup) {
1328                         if (hwclock_is_localtime() > 0) {
1329                                 int min;
1330
1331                                 /* The first-time call to settimeofday() does a time warp in the kernel */
1332                                 r = hwclock_set_timezone(&min);
1333                                 if (r < 0)
1334                                         log_error("Failed to apply local time delta, ignoring: %s", strerror(-r));
1335                                 else
1336                                         log_info("RTC configured in localtime, applying delta of %i minutes to system time.", min);
1337                         } else if (!in_initrd()) {
1338                                 /*
1339                                  * Do dummy first-time call to seal the kernel's time warp magic
1340                                  *
1341                                  * Do not call this this from inside the initrd. The initrd might not
1342                                  * carry /etc/adjtime with LOCAL, but the real system could be set up
1343                                  * that way. In such case, we need to delay the time-warp or the sealing
1344                                  * until we reach the real system.
1345                                  */
1346                                 hwclock_reset_timezone();
1347
1348                                 /* Tell the kernel our time zone */
1349                                 r = hwclock_set_timezone(NULL);
1350                                 if (r < 0)
1351                                         log_error("Failed to set the kernel's time zone, ignoring: %s", strerror(-r));
1352                         }
1353                 }
1354
1355                 /* Set the default for later on, but don't actually
1356                  * open the logs like this for now. Note that if we
1357                  * are transitioning from the initrd there might still
1358                  * be journal fd open, and we shouldn't attempt
1359                  * opening that before we parsed /proc/cmdline which
1360                  * might redirect output elsewhere. */
1361                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1362
1363         } else if (getpid() == 1) {
1364                 /* Running inside a container, as PID 1 */
1365                 arg_running_as = SYSTEMD_SYSTEM;
1366                 log_set_target(LOG_TARGET_CONSOLE);
1367                 log_open();
1368
1369                 /* For the later on, see above... */
1370                 log_set_target(LOG_TARGET_JOURNAL);
1371
1372                 /* clear the kernel timestamp,
1373                  * because we are in a container */
1374                 kernel_timestamp.monotonic = 0ULL;
1375                 kernel_timestamp.realtime = 0ULL;
1376
1377         } else {
1378                 /* Running as user instance */
1379                 arg_running_as = SYSTEMD_USER;
1380                 log_set_target(LOG_TARGET_AUTO);
1381                 log_open();
1382
1383                 /* clear the kernel timestamp,
1384                  * because we are not PID 1 */
1385                 kernel_timestamp.monotonic = 0ULL;
1386                 kernel_timestamp.realtime = 0ULL;
1387         }
1388
1389         /* Initialize default unit */
1390         r = set_default_unit(SPECIAL_DEFAULT_TARGET);
1391         if (r < 0) {
1392                 log_error("Failed to set default unit %s: %s", SPECIAL_DEFAULT_TARGET, strerror(-r));
1393                 goto finish;
1394         }
1395
1396         r = initialize_join_controllers();
1397         if (r < 0)
1398                 goto finish;
1399
1400         /* Mount /proc, /sys and friends, so that /proc/cmdline and
1401          * /proc/$PID/fd is available. */
1402         if (getpid() == 1) {
1403                 r = mount_setup(loaded_policy);
1404                 if (r < 0)
1405                         goto finish;
1406         }
1407
1408         /* Reset all signal handlers. */
1409         assert_se(reset_all_signal_handlers() == 0);
1410
1411         ignore_signals(SIGNALS_IGNORE, -1);
1412
1413         if (parse_config_file() < 0)
1414                 goto finish;
1415
1416         if (arg_running_as == SYSTEMD_SYSTEM)
1417                 if (parse_proc_cmdline() < 0)
1418                         goto finish;
1419
1420         log_parse_environment();
1421
1422         if (parse_argv(argc, argv) < 0)
1423                 goto finish;
1424
1425         if (arg_action == ACTION_TEST &&
1426             geteuid() == 0) {
1427                 log_error("Don't run test mode as root.");
1428                 goto finish;
1429         }
1430
1431         if (arg_running_as == SYSTEMD_USER &&
1432             arg_action == ACTION_RUN &&
1433             sd_booted() <= 0) {
1434                 log_error("Trying to run as user instance, but the system has not been booted with systemd.");
1435                 goto finish;
1436         }
1437
1438         if (arg_running_as == SYSTEMD_SYSTEM &&
1439             arg_action == ACTION_RUN &&
1440             running_in_chroot() > 0) {
1441                 log_error("Cannot be run in a chroot() environment.");
1442                 goto finish;
1443         }
1444
1445         if (arg_action == ACTION_HELP) {
1446                 retval = help();
1447                 goto finish;
1448         } else if (arg_action == ACTION_VERSION) {
1449                 retval = version();
1450                 goto finish;
1451         } else if (arg_action == ACTION_DUMP_CONFIGURATION_ITEMS) {
1452                 unit_dump_config_items(stdout);
1453                 retval = EXIT_SUCCESS;
1454                 goto finish;
1455         } else if (arg_action == ACTION_DONE) {
1456                 retval = EXIT_SUCCESS;
1457                 goto finish;
1458         }
1459
1460         assert_se(arg_action == ACTION_RUN || arg_action == ACTION_TEST);
1461
1462         /* Close logging fds, in order not to confuse fdset below */
1463         log_close();
1464
1465         /* Remember open file descriptors for later deserialization */
1466         r = fdset_new_fill(&fds);
1467         if (r < 0) {
1468                 log_error("Failed to allocate fd set: %s", strerror(-r));
1469                 goto finish;
1470         } else
1471                 fdset_cloexec(fds, true);
1472
1473         if (serialization)
1474                 assert_se(fdset_remove(fds, fileno(serialization)) >= 0);
1475
1476         /* Set up PATH unless it is already set */
1477         setenv("PATH",
1478 #ifdef HAVE_SPLIT_USR
1479                "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
1480 #else
1481                "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin",
1482 #endif
1483                arg_running_as == SYSTEMD_SYSTEM);
1484
1485         if (arg_running_as == SYSTEMD_SYSTEM) {
1486                 /* Parse the data passed to us. We leave this
1487                  * variables set, but the manager later on will not
1488                  * pass them on to our children. */
1489                 if (!in_initrd())
1490                         parse_initrd_timestamp(&initrd_timestamp);
1491
1492                 /* Unset some environment variables passed in from the
1493                  * kernel that don't really make sense for us. */
1494                 unsetenv("HOME");
1495                 unsetenv("TERM");
1496
1497                 /* When we are invoked by a shell, these might be set,
1498                  * but make little sense to pass on */
1499                 unsetenv("PWD");
1500                 unsetenv("SHLVL");
1501                 unsetenv("_");
1502
1503                 /* When we are invoked by a chroot-like tool such as
1504                  * nspawn, these might be set, but make little sense
1505                  * to pass on */
1506                 unsetenv("USER");
1507                 unsetenv("LOGNAME");
1508
1509                 /* We suppress the socket activation env vars, as
1510                  * we'll try to match *any* open fd to units if
1511                  * possible. */
1512                 unsetenv("LISTEN_FDS");
1513                 unsetenv("LISTEN_PID");
1514
1515                 /* All other variables are left as is, so that clients
1516                  * can still read them via /proc/1/environ */
1517         }
1518
1519         /* Move out of the way, so that we won't block unmounts */
1520         assert_se(chdir("/")  == 0);
1521
1522         if (arg_running_as == SYSTEMD_SYSTEM) {
1523                 /* Become a session leader if we aren't one yet. */
1524                 setsid();
1525
1526                 /* Disable the umask logic */
1527                 umask(0);
1528         }
1529
1530         /* Make sure D-Bus doesn't fiddle with the SIGPIPE handlers */
1531         dbus_connection_set_change_sigpipe(FALSE);
1532
1533         /* Reset the console, but only if this is really init and we
1534          * are freshly booted */
1535         if (arg_running_as == SYSTEMD_SYSTEM && arg_action == ACTION_RUN)
1536                 console_setup(getpid() == 1 && !skip_setup);
1537
1538         /* Open the logging devices, if possible and necessary */
1539         log_open();
1540
1541         /* Make sure we leave a core dump without panicing the
1542          * kernel. */
1543         if (getpid() == 1) {
1544                 install_crash_handler();
1545
1546                 r = mount_cgroup_controllers(arg_join_controllers);
1547                 if (r < 0)
1548                         goto finish;
1549         }
1550
1551         if (arg_running_as == SYSTEMD_SYSTEM) {
1552                 const char *virtualization = NULL;
1553
1554                 log_info(PACKAGE_STRING " running in system mode. (" SYSTEMD_FEATURES ")");
1555
1556                 detect_virtualization(&virtualization);
1557                 if (virtualization)
1558                         log_info("Detected virtualization '%s'.", virtualization);
1559
1560                 if (in_initrd())
1561                         log_info("Running in initial RAM disk.");
1562
1563         } else
1564                 log_debug(PACKAGE_STRING " running in user mode. (" SYSTEMD_FEATURES ")");
1565
1566         if (arg_running_as == SYSTEMD_SYSTEM && !skip_setup) {
1567                 locale_setup();
1568
1569                 if (arg_show_status || plymouth_running())
1570                         status_welcome();
1571
1572 #ifdef HAVE_KMOD
1573                 kmod_setup();
1574 #endif
1575                 hostname_setup();
1576                 machine_id_setup();
1577                 loopback_setup();
1578
1579                 test_mtab();
1580                 test_usr();
1581                 test_cgroups();
1582         }
1583
1584         if (arg_running_as == SYSTEMD_SYSTEM && arg_runtime_watchdog > 0)
1585                 watchdog_set_timeout(&arg_runtime_watchdog);
1586
1587         if (arg_timer_slack_nsec != (nsec_t) -1)
1588                 if (prctl(PR_SET_TIMERSLACK, arg_timer_slack_nsec) < 0)
1589                         log_error("Failed to adjust timer slack: %m");
1590
1591         if (arg_capability_bounding_set_drop) {
1592                 r = capability_bounding_set_drop(arg_capability_bounding_set_drop, true);
1593                 if (r < 0) {
1594                         log_error("Failed to drop capability bounding set: %s", strerror(-r));
1595                         goto finish;
1596                 }
1597                 r = capability_bounding_set_drop_usermode(arg_capability_bounding_set_drop);
1598                 if (r < 0) {
1599                         log_error("Failed to drop capability bounding set of usermode helpers: %s", strerror(-r));
1600                         goto finish;
1601                 }
1602         }
1603
1604         if (arg_running_as == SYSTEMD_USER) {
1605                 /* Become reaper of our children */
1606                 if (prctl(PR_SET_CHILD_SUBREAPER, 1) < 0) {
1607                         log_warning("Failed to make us a subreaper: %m");
1608                         if (errno == EINVAL)
1609                                 log_info("Perhaps the kernel version is too old (< 3.4?)");
1610                 }
1611         }
1612
1613         if (arg_running_as == SYSTEMD_SYSTEM)
1614                 bump_rlimit_nofile(&saved_rlimit_nofile);
1615
1616         r = manager_new(arg_running_as, &m);
1617         if (r < 0) {
1618                 log_error("Failed to allocate manager object: %s", strerror(-r));
1619                 goto finish;
1620         }
1621
1622         m->confirm_spawn = arg_confirm_spawn;
1623         m->default_std_output = arg_default_std_output;
1624         m->default_std_error = arg_default_std_error;
1625         m->runtime_watchdog = arg_runtime_watchdog;
1626         m->shutdown_watchdog = arg_shutdown_watchdog;
1627         m->userspace_timestamp = userspace_timestamp;
1628         m->kernel_timestamp = kernel_timestamp;
1629         m->initrd_timestamp = initrd_timestamp;
1630
1631         manager_set_default_rlimits(m, arg_default_rlimit);
1632
1633         if (arg_default_environment)
1634                 manager_set_default_environment(m, arg_default_environment);
1635
1636         manager_set_show_status(m, arg_show_status);
1637
1638         /* Remember whether we should queue the default job */
1639         queue_default_job = !serialization || arg_switched_root;
1640
1641         before_startup = now(CLOCK_MONOTONIC);
1642
1643         r = manager_startup(m, serialization, fds);
1644         if (r < 0)
1645                 log_error("Failed to fully start up daemon: %s", strerror(-r));
1646
1647         /* This will close all file descriptors that were opened, but
1648          * not claimed by any unit. */
1649         fdset_free(fds);
1650
1651         if (serialization) {
1652                 fclose(serialization);
1653                 serialization = NULL;
1654         }
1655
1656         if (queue_default_job) {
1657                 DBusError error;
1658                 Unit *target = NULL;
1659                 Job *default_unit_job;
1660
1661                 dbus_error_init(&error);
1662
1663                 log_debug("Activating default unit: %s", arg_default_unit);
1664
1665                 r = manager_load_unit(m, arg_default_unit, NULL, &error, &target);
1666                 if (r < 0) {
1667                         log_error("Failed to load default target: %s", bus_error(&error, r));
1668                         dbus_error_free(&error);
1669                 } else if (target->load_state == UNIT_ERROR)
1670                         log_error("Failed to load default target: %s", strerror(-target->load_error));
1671                 else if (target->load_state == UNIT_MASKED)
1672                         log_error("Default target masked.");
1673
1674                 if (!target || target->load_state != UNIT_LOADED) {
1675                         log_info("Trying to load rescue target...");
1676
1677                         r = manager_load_unit(m, SPECIAL_RESCUE_TARGET, NULL, &error, &target);
1678                         if (r < 0) {
1679                                 log_error("Failed to load rescue target: %s", bus_error(&error, r));
1680                                 dbus_error_free(&error);
1681                                 goto finish;
1682                         } else if (target->load_state == UNIT_ERROR) {
1683                                 log_error("Failed to load rescue target: %s", strerror(-target->load_error));
1684                                 goto finish;
1685                         } else if (target->load_state == UNIT_MASKED) {
1686                                 log_error("Rescue target masked.");
1687                                 goto finish;
1688                         }
1689                 }
1690
1691                 assert(target->load_state == UNIT_LOADED);
1692
1693                 if (arg_action == ACTION_TEST) {
1694                         printf("-> By units:\n");
1695                         manager_dump_units(m, stdout, "\t");
1696                 }
1697
1698                 r = manager_add_job(m, JOB_START, target, JOB_ISOLATE, false, &error, &default_unit_job);
1699                 if (r == -EPERM) {
1700                         log_debug("Default target could not be isolated, starting instead: %s", bus_error(&error, r));
1701                         dbus_error_free(&error);
1702
1703                         r = manager_add_job(m, JOB_START, target, JOB_REPLACE, false, &error, &default_unit_job);
1704                         if (r < 0) {
1705                                 log_error("Failed to start default target: %s", bus_error(&error, r));
1706                                 dbus_error_free(&error);
1707                                 goto finish;
1708                         }
1709                 } else if (r < 0) {
1710                         log_error("Failed to isolate default target: %s", bus_error(&error, r));
1711                         dbus_error_free(&error);
1712                         goto finish;
1713                 }
1714
1715                 m->default_unit_job_id = default_unit_job->id;
1716
1717                 after_startup = now(CLOCK_MONOTONIC);
1718                 log_full(arg_action == ACTION_TEST ? LOG_INFO : LOG_DEBUG,
1719                          "Loaded units and determined initial transaction in %s.",
1720                          format_timespan(timespan, sizeof(timespan), after_startup - before_startup, 0));
1721
1722                 if (arg_action == ACTION_TEST) {
1723                         printf("-> By jobs:\n");
1724                         manager_dump_jobs(m, stdout, "\t");
1725                         retval = EXIT_SUCCESS;
1726                         goto finish;
1727                 }
1728         }
1729
1730         for (;;) {
1731                 r = manager_loop(m);
1732                 if (r < 0) {
1733                         log_error("Failed to run mainloop: %s", strerror(-r));
1734                         goto finish;
1735                 }
1736
1737                 switch (m->exit_code) {
1738
1739                 case MANAGER_EXIT:
1740                         retval = EXIT_SUCCESS;
1741                         log_debug("Exit.");
1742                         goto finish;
1743
1744                 case MANAGER_RELOAD:
1745                         log_info("Reloading.");
1746                         r = manager_reload(m);
1747                         if (r < 0)
1748                                 log_error("Failed to reload: %s", strerror(-r));
1749                         break;
1750
1751                 case MANAGER_REEXECUTE:
1752
1753                         if (prepare_reexecute(m, &serialization, &fds, false) < 0)
1754                                 goto finish;
1755
1756                         reexecute = true;
1757                         log_notice("Reexecuting.");
1758                         goto finish;
1759
1760                 case MANAGER_SWITCH_ROOT:
1761                         /* Steal the switch root parameters */
1762                         switch_root_dir = m->switch_root;
1763                         switch_root_init = m->switch_root_init;
1764                         m->switch_root = m->switch_root_init = NULL;
1765
1766                         if (!switch_root_init)
1767                                 if (prepare_reexecute(m, &serialization, &fds, true) < 0)
1768                                         goto finish;
1769
1770                         reexecute = true;
1771                         log_notice("Switching root.");
1772                         goto finish;
1773
1774                 case MANAGER_REBOOT:
1775                 case MANAGER_POWEROFF:
1776                 case MANAGER_HALT:
1777                 case MANAGER_KEXEC: {
1778                         static const char * const table[_MANAGER_EXIT_CODE_MAX] = {
1779                                 [MANAGER_REBOOT] = "reboot",
1780                                 [MANAGER_POWEROFF] = "poweroff",
1781                                 [MANAGER_HALT] = "halt",
1782                                 [MANAGER_KEXEC] = "kexec"
1783                         };
1784
1785                         assert_se(shutdown_verb = table[m->exit_code]);
1786                         arm_reboot_watchdog = m->exit_code == MANAGER_REBOOT;
1787
1788                         log_notice("Shutting down.");
1789                         goto finish;
1790                 }
1791
1792                 default:
1793                         assert_not_reached("Unknown exit code.");
1794                 }
1795         }
1796
1797 finish:
1798         if (m)
1799                 manager_free(m);
1800
1801         for (j = 0; j < RLIMIT_NLIMITS; j++)
1802                 free(arg_default_rlimit[j]);
1803
1804         free(arg_default_unit);
1805         free_join_controllers();
1806
1807         dbus_shutdown();
1808         label_finish();
1809
1810         if (reexecute) {
1811                 const char **args;
1812                 unsigned i, args_size;
1813
1814                 /* Close and disarm the watchdog, so that the new
1815                  * instance can reinitialize it, but doesn't get
1816                  * rebooted while we do that */
1817                 watchdog_close(true);
1818
1819                 /* Reset the RLIMIT_NOFILE to the kernel default, so
1820                  * that the new systemd can pass the kernel default to
1821                  * its child processes */
1822                 if (saved_rlimit_nofile.rlim_cur > 0)
1823                         setrlimit(RLIMIT_NOFILE, &saved_rlimit_nofile);
1824
1825                 if (switch_root_dir) {
1826                         /* Kill all remaining processes from the
1827                          * initrd, but don't wait for them, so that we
1828                          * can handle the SIGCHLD for them after
1829                          * deserializing. */
1830                         broadcast_signal(SIGTERM, false);
1831
1832                         /* And switch root */
1833                         r = switch_root(switch_root_dir);
1834                         if (r < 0)
1835                                 log_error("Failed to switch root, ignoring: %s", strerror(-r));
1836                 }
1837
1838                 args_size = MAX(6, argc+1);
1839                 args = newa(const char*, args_size);
1840
1841                 if (!switch_root_init) {
1842                         char sfd[16];
1843
1844                         /* First try to spawn ourselves with the right
1845                          * path, and with full serialization. We do
1846                          * this only if the user didn't specify an
1847                          * explicit init to spawn. */
1848
1849                         assert(serialization);
1850                         assert(fds);
1851
1852                         snprintf(sfd, sizeof(sfd), "%i", fileno(serialization));
1853                         char_array_0(sfd);
1854
1855                         i = 0;
1856                         args[i++] = SYSTEMD_BINARY_PATH;
1857                         if (switch_root_dir)
1858                                 args[i++] = "--switched-root";
1859                         args[i++] = arg_running_as == SYSTEMD_SYSTEM ? "--system" : "--user";
1860                         args[i++] = "--deserialize";
1861                         args[i++] = sfd;
1862                         args[i++] = NULL;
1863
1864                         assert(i <= args_size);
1865                         execv(args[0], (char* const*) args);
1866                 }
1867
1868                 /* Try the fallback, if there is any, without any
1869                  * serialization. We pass the original argv[] and
1870                  * envp[]. (Well, modulo the ordering changes due to
1871                  * getopt() in argv[], and some cleanups in envp[],
1872                  * but let's hope that doesn't matter.) */
1873
1874                 if (serialization) {
1875                         fclose(serialization);
1876                         serialization = NULL;
1877                 }
1878
1879                 if (fds) {
1880                         fdset_free(fds);
1881                         fds = NULL;
1882                 }
1883
1884                 /* Reopen the console */
1885                 make_console_stdio();
1886
1887                 for (j = 1, i = 1; j < argc; j++)
1888                         args[i++] = argv[j];
1889                 args[i++] = NULL;
1890                 assert(i <= args_size);
1891
1892                 if (switch_root_init) {
1893                         args[0] = switch_root_init;
1894                         execv(args[0], (char* const*) args);
1895                         log_warning("Failed to execute configured init, trying fallback: %m");
1896                 }
1897
1898                 args[0] = "/sbin/init";
1899                 execv(args[0], (char* const*) args);
1900
1901                 if (errno == ENOENT) {
1902                         log_warning("No /sbin/init, trying fallback");
1903
1904                         args[0] = "/bin/sh";
1905                         args[1] = NULL;
1906                         execv(args[0], (char* const*) args);
1907                         log_error("Failed to execute /bin/sh, giving up: %m");
1908                 } else
1909                         log_warning("Failed to execute /sbin/init, giving up: %m");
1910         }
1911
1912         if (serialization)
1913                 fclose(serialization);
1914
1915         if (fds)
1916                 fdset_free(fds);
1917
1918         if (shutdown_verb) {
1919                 const char * command_line[] = {
1920                         SYSTEMD_SHUTDOWN_BINARY_PATH,
1921                         shutdown_verb,
1922                         NULL
1923                 };
1924                 char **env_block;
1925
1926                 if (arm_reboot_watchdog && arg_shutdown_watchdog > 0) {
1927                         char e[32];
1928
1929                         /* If we reboot let's set the shutdown
1930                          * watchdog and tell the shutdown binary to
1931                          * repeatedly ping it */
1932                         watchdog_set_timeout(&arg_shutdown_watchdog);
1933                         watchdog_close(false);
1934
1935                         /* Tell the binary how often to ping */
1936                         snprintf(e, sizeof(e), "WATCHDOG_USEC=%llu", (unsigned long long) arg_shutdown_watchdog);
1937                         char_array_0(e);
1938
1939                         env_block = strv_append(environ, e);
1940                 } else {
1941                         env_block = strv_copy(environ);
1942                         watchdog_close(true);
1943                 }
1944
1945                 execve(SYSTEMD_SHUTDOWN_BINARY_PATH, (char **) command_line, env_block);
1946                 free(env_block);
1947                 log_error("Failed to execute shutdown binary, freezing: %m");
1948         }
1949
1950         if (getpid() == 1)
1951                 freeze();
1952
1953         return retval;
1954 }