1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
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.
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.
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/>.
29 #include <sys/reboot.h>
30 #include <sys/ioctl.h>
34 #include <sys/types.h>
37 #include <sys/timerfd.h>
43 #include "sd-daemon.h"
45 #include "sd-messages.h"
48 #include "transaction.h"
55 #include "ratelimit.h"
56 #include "locale-setup.h"
57 #include "mount-setup.h"
58 #include "unit-name.h"
60 #include "path-lookup.h"
62 #include "exit-status.h"
65 #include "cgroup-util.h"
66 #include "path-util.h"
68 #include "boot-timestamps.h"
70 #include "bus-errors.h"
71 #include "bus-error.h"
74 #include "dbus-unit.h"
76 #include "dbus-manager.h"
77 #include "bus-kernel.h"
79 /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */
80 #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC)
82 /* Initial delay and the interval for printing status messages about running jobs */
83 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
84 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
85 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
87 #define TIME_T_MAX (time_t)((1UL << ((sizeof(time_t) << 3) - 1)) - 1)
89 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
90 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
91 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
92 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
93 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
94 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
96 static int manager_watch_jobs_in_progress(Manager *m) {
101 if (m->jobs_in_progress_event_source)
104 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
105 return sd_event_add_time(
107 &m->jobs_in_progress_event_source,
110 manager_dispatch_jobs_in_progress, m);
113 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED_ON)-1) + sizeof(ANSI_HIGHLIGHT_RED_ON)-1 + 2*(sizeof(ANSI_HIGHLIGHT_OFF)-1))
115 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
118 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
119 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
123 p = mempset(p, ' ', pos-2);
124 p = stpcpy(p, ANSI_RED_ON);
128 if (pos > 0 && pos <= width) {
129 p = stpcpy(p, ANSI_HIGHLIGHT_RED_ON);
133 p = stpcpy(p, ANSI_HIGHLIGHT_OFF);
136 p = stpcpy(p, ANSI_RED_ON);
139 p = mempset(p, ' ', width-1-pos);
140 strcpy(p, ANSI_HIGHLIGHT_OFF);
144 void manager_flip_auto_status(Manager *m, bool enable) {
148 if (m->show_status == SHOW_STATUS_AUTO)
149 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
151 if (m->show_status == SHOW_STATUS_TEMPORARY)
152 manager_set_show_status(m, SHOW_STATUS_AUTO);
156 static void manager_print_jobs_in_progress(Manager *m) {
157 _cleanup_free_ char *job_of_n = NULL;
160 unsigned counter = 0, print_nr;
161 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
163 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
168 manager_flip_auto_status(m, true);
170 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
172 HASHMAP_FOREACH(j, m->jobs, i)
173 if (j->state == JOB_RUNNING && counter++ == print_nr)
176 /* m->n_running_jobs must be consistent with the contents of m->jobs,
177 * so the above loop must have succeeded in finding j. */
178 assert(counter == print_nr + 1);
181 cylon_pos = m->jobs_in_progress_iteration % 14;
183 cylon_pos = 14 - cylon_pos;
184 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
186 m->jobs_in_progress_iteration++;
188 if (m->n_running_jobs > 1)
189 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
192 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
193 if (job_get_timeout(j, &x) > 0)
194 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
196 manager_status_printf(m, true, cylon,
197 "%sA %s job is running for %s (%s / %s)",
199 job_type_to_string(j->type),
200 unit_description(j->unit),
205 static int manager_watch_idle_pipe(Manager *m) {
210 if (m->idle_pipe_event_source)
213 if (m->idle_pipe[2] < 0)
216 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
218 log_error("Failed to watch idle pipe: %s", strerror(-r));
225 static void manager_close_idle_pipe(Manager *m) {
228 safe_close_pair(m->idle_pipe);
229 safe_close_pair(m->idle_pipe + 2);
232 static int manager_setup_time_change(Manager *m) {
235 /* We only care for the cancellation event, hence we set the
236 * timeout to the latest possible value. */
237 struct itimerspec its = {
238 .it_value.tv_sec = TIME_T_MAX,
242 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
247 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
248 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
250 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
251 if (m->time_change_fd < 0) {
252 log_error("Failed to create timerfd: %m");
256 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
257 log_debug("Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
258 m->time_change_fd = safe_close(m->time_change_fd);
262 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
264 log_error("Failed to create time change event source: %s", strerror(-r));
268 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
273 static int enable_special_signals(Manager *m) {
274 _cleanup_close_ int fd = -1;
278 /* Enable that we get SIGINT on control-alt-del. In containers
279 * this will fail with EPERM (older) or EINVAL (newer), so
281 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
282 log_warning("Failed to enable ctrl-alt-del handling: %m");
284 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
286 /* Support systems without virtual console */
288 log_warning("Failed to open /dev/tty0: %m");
290 /* Enable that we get SIGWINCH on kbrequest */
291 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
292 log_warning("Failed to enable kbrequest handling: %m");
298 static int manager_setup_signals(Manager *m) {
299 struct sigaction sa = {
300 .sa_handler = SIG_DFL,
301 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
311 /* We are not interested in SIGSTOP and friends. */
312 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
314 assert_se(sigemptyset(&mask) == 0);
316 sigset_add_many(&mask,
317 SIGCHLD, /* Child died */
318 SIGTERM, /* Reexecute daemon */
319 SIGHUP, /* Reload configuration */
320 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
321 SIGUSR2, /* systemd: dump status */
322 SIGINT, /* Kernel sends us this on control-alt-del */
323 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
324 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
325 SIGRTMIN+0, /* systemd: start default.target */
326 SIGRTMIN+1, /* systemd: isolate rescue.target */
327 SIGRTMIN+2, /* systemd: isolate emergency.target */
328 SIGRTMIN+3, /* systemd: start halt.target */
329 SIGRTMIN+4, /* systemd: start poweroff.target */
330 SIGRTMIN+5, /* systemd: start reboot.target */
331 SIGRTMIN+6, /* systemd: start kexec.target */
332 SIGRTMIN+13, /* systemd: Immediate halt */
333 SIGRTMIN+14, /* systemd: Immediate poweroff */
334 SIGRTMIN+15, /* systemd: Immediate reboot */
335 SIGRTMIN+16, /* systemd: Immediate kexec */
336 SIGRTMIN+20, /* systemd: enable status messages */
337 SIGRTMIN+21, /* systemd: disable status messages */
338 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
339 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
340 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
341 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
342 SIGRTMIN+27, /* systemd: set log target to console */
343 SIGRTMIN+28, /* systemd: set log target to kmsg */
344 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete)*/
346 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
348 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
349 if (m->signal_fd < 0)
352 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
356 /* Process signals a bit earlier than the rest of things, but
357 * later that notify_fd processing, so that the notify
358 * processing can still figure out to which process/service a
359 * message belongs, before we reap the process. */
360 r = sd_event_source_set_priority(m->signal_event_source, -5);
364 if (m->running_as == SYSTEMD_SYSTEM)
365 return enable_special_signals(m);
370 static void manager_clean_environment(Manager *m) {
373 /* Let's remove some environment variables that we
374 * need ourselves to communicate with our clients */
387 static int manager_default_environment(Manager *m) {
390 if (m->running_as == SYSTEMD_SYSTEM) {
391 /* The system manager always starts with a clean
392 * environment for its children. It does not import
393 * the kernel or the parents exported variables.
395 * The initial passed environ is untouched to keep
396 * /proc/self/environ valid; it is used for tagging
397 * the init process inside containers. */
398 m->environment = strv_new("PATH=" DEFAULT_PATH,
401 /* Import locale variables LC_*= from configuration */
402 locale_setup(&m->environment);
404 /* The user manager passes its own environment
405 * along to its children. */
406 m->environment = strv_copy(environ);
412 manager_clean_environment(m);
413 strv_sort(m->environment);
418 int manager_new(SystemdRunningAs running_as, bool test_run, Manager **_m) {
423 assert(running_as >= 0);
424 assert(running_as < _SYSTEMD_RUNNING_AS_MAX);
426 m = new0(Manager, 1);
431 if (running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0)
432 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
435 m->running_as = running_as;
436 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
437 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
439 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
441 m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd = m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = -1;
442 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
444 m->test_run = test_run;
446 r = manager_default_environment(m);
450 r = hashmap_ensure_allocated(&m->units, string_hash_func, string_compare_func);
454 r = hashmap_ensure_allocated(&m->jobs, trivial_hash_func, trivial_compare_func);
458 r = hashmap_ensure_allocated(&m->cgroup_unit, string_hash_func, string_compare_func);
462 r = hashmap_ensure_allocated(&m->watch_bus, string_hash_func, string_compare_func);
466 r = set_ensure_allocated(&m->startup_units, trivial_hash_func, trivial_compare_func);
470 r = set_ensure_allocated(&m->failed_units, trivial_hash_func, trivial_compare_func);
474 r = sd_event_default(&m->event);
478 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
482 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
486 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
490 r = manager_setup_signals(m);
494 r = manager_setup_cgroup(m);
498 r = manager_setup_time_change(m);
502 m->udev = udev_new();
508 /* Note that we set up neither kdbus, nor the notify fd
509 * here. We do that after deserialization, since they might
510 * have gotten serialized across the reexec. */
512 m->taint_usr = dir_is_empty("/usr") > 0;
522 static int manager_setup_notify(Manager *m) {
528 if (m->notify_fd < 0) {
529 _cleanup_close_ int fd = -1;
530 union sockaddr_union sa = {
531 .sa.sa_family = AF_UNIX,
533 static const int one = 1;
535 /* First free all secondary fields */
536 free(m->notify_socket);
537 m->notify_socket = NULL;
538 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
540 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
542 log_error("Failed to allocate notification socket: %m");
546 if (m->running_as == SYSTEMD_SYSTEM)
547 m->notify_socket = strdup("/run/systemd/notify");
551 e = getenv("XDG_RUNTIME_DIR");
553 log_error("XDG_RUNTIME_DIR is not set: %m");
557 m->notify_socket = strappend(e, "/systemd/notify");
559 if (!m->notify_socket)
562 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
563 r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
565 log_error("bind(@%s) failed: %m", sa.un.sun_path+1);
569 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
571 log_error("SO_PASSCRED failed: %m");
578 log_debug("Using notification socket %s", m->notify_socket);
581 if (!m->notify_event_source) {
582 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
584 log_error("Failed to allocate notify event source: %s", strerror(-r));
588 /* Process signals a bit earlier than SIGCHLD, so that we can
589 * still identify to which service an exit message belongs */
590 r = sd_event_source_set_priority(m->notify_event_source, -7);
592 log_error("Failed to set priority of notify event source: %s", strerror(-r));
600 static int manager_setup_kdbus(Manager *m) {
602 _cleanup_free_ char *p = NULL;
606 if (m->test_run || m->kdbus_fd >= 0)
609 m->kdbus_fd = bus_kernel_create_bus(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", m->running_as == SYSTEMD_SYSTEM, &p);
610 if (m->kdbus_fd < 0) {
611 log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd));
615 log_debug("Successfully set up kdbus on %s", p);
617 /* Create the namespace directory here, so that the contents
618 * of that directory is not visible to non-root users. This is
619 * necessary to ensure that users cannot get access to busses
620 * of virtualized users when no UID namespacing is used. */
621 if (m->running_as == SYSTEMD_SYSTEM)
622 mkdir_p_label("/dev/kdbus/domain", 0700);
628 static int manager_connect_bus(Manager *m, bool reexecuting) {
629 bool try_bus_connect;
639 (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"));
641 /* Try to connect to the busses, if possible. */
642 return bus_init(m, try_bus_connect);
645 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
651 while ((u = m->cleanup_queue)) {
652 assert(u->in_cleanup_queue);
662 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
663 GC_OFFSET_UNSURE, /* No clue */
664 GC_OFFSET_GOOD, /* We still need this unit */
665 GC_OFFSET_BAD, /* We don't need this unit anymore */
669 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
676 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
677 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
678 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
681 if (u->in_cleanup_queue)
684 if (unit_check_gc(u))
687 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
691 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
692 unit_gc_sweep(other, gc_marker);
694 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
697 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
704 /* We were unable to find anything out about this entry, so
705 * let's investigate it later */
706 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
707 unit_add_to_gc_queue(u);
711 /* We definitely know that this one is not useful anymore, so
712 * let's mark it for deletion */
713 u->gc_marker = gc_marker + GC_OFFSET_BAD;
714 unit_add_to_cleanup_queue(u);
718 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
721 static unsigned manager_dispatch_gc_queue(Manager *m) {
728 /* log_debug("Running GC..."); */
730 m->gc_marker += _GC_OFFSET_MAX;
731 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
734 gc_marker = m->gc_marker;
736 while ((u = m->gc_queue)) {
737 assert(u->in_gc_queue);
739 unit_gc_sweep(u, gc_marker);
741 LIST_REMOVE(gc_queue, m->gc_queue, u);
742 u->in_gc_queue = false;
746 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
747 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
748 log_debug_unit(u->id, "Collecting %s", u->id);
749 u->gc_marker = gc_marker + GC_OFFSET_BAD;
750 unit_add_to_cleanup_queue(u);
754 m->n_in_gc_queue = 0;
759 static void manager_clear_jobs_and_units(Manager *m) {
764 while ((u = hashmap_first(m->units)))
767 manager_dispatch_cleanup_queue(m);
769 assert(!m->load_queue);
770 assert(!m->run_queue);
771 assert(!m->dbus_unit_queue);
772 assert(!m->dbus_job_queue);
773 assert(!m->cleanup_queue);
774 assert(!m->gc_queue);
776 assert(hashmap_isempty(m->jobs));
777 assert(hashmap_isempty(m->units));
780 m->n_running_jobs = 0;
783 void manager_free(Manager *m) {
789 manager_clear_jobs_and_units(m);
791 for (c = 0; c < _UNIT_TYPE_MAX; c++)
792 if (unit_vtable[c]->shutdown)
793 unit_vtable[c]->shutdown(m);
795 /* If we reexecute ourselves, we keep the root cgroup
797 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
799 manager_undo_generators(m);
803 hashmap_free(m->units);
804 hashmap_free(m->jobs);
805 hashmap_free(m->watch_pids1);
806 hashmap_free(m->watch_pids2);
807 hashmap_free(m->watch_bus);
809 set_free(m->startup_units);
810 set_free(m->failed_units);
812 sd_event_source_unref(m->signal_event_source);
813 sd_event_source_unref(m->notify_event_source);
814 sd_event_source_unref(m->time_change_event_source);
815 sd_event_source_unref(m->jobs_in_progress_event_source);
816 sd_event_source_unref(m->idle_pipe_event_source);
817 sd_event_source_unref(m->run_queue_event_source);
819 safe_close(m->signal_fd);
820 safe_close(m->notify_fd);
821 safe_close(m->time_change_fd);
822 safe_close(m->kdbus_fd);
824 manager_close_idle_pipe(m);
827 sd_event_unref(m->event);
829 free(m->notify_socket);
831 lookup_paths_free(&m->lookup_paths);
832 strv_free(m->environment);
834 hashmap_free(m->cgroup_unit);
835 set_free_free(m->unit_path_cache);
837 free(m->switch_root);
838 free(m->switch_root_init);
840 for (i = 0; i < _RLIMIT_MAX; i++)
843 assert(hashmap_isempty(m->units_requiring_mounts_for));
844 hashmap_free(m->units_requiring_mounts_for);
849 int manager_enumerate(Manager *m) {
855 /* Let's ask every type to load all units from disk/kernel
856 * that it might know */
857 for (c = 0; c < _UNIT_TYPE_MAX; c++)
858 if (unit_vtable[c]->enumerate) {
859 q = unit_vtable[c]->enumerate(m);
864 manager_dispatch_load_queue(m);
868 static int manager_coldplug(Manager *m) {
876 /* Then, let's set up their initial state. */
877 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
884 q = unit_coldplug(u);
892 static void manager_build_unit_path_cache(Manager *m) {
894 _cleanup_free_ DIR *d = NULL;
899 set_free_free(m->unit_path_cache);
901 m->unit_path_cache = set_new(string_hash_func, string_compare_func);
902 if (!m->unit_path_cache) {
903 log_error("Failed to allocate unit path cache.");
907 /* This simply builds a list of files we know exist, so that
908 * we don't always have to go to disk */
910 STRV_FOREACH(i, m->lookup_paths.unit_path) {
916 log_error("Failed to open directory %s: %m", *i);
920 while ((de = readdir(d))) {
923 if (ignore_file(de->d_name))
926 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
932 r = set_consume(m->unit_path_cache, p);
944 log_error("Failed to build unit path cache: %s", strerror(-r));
946 set_free_free(m->unit_path_cache);
947 m->unit_path_cache = NULL;
951 static int manager_distribute_fds(Manager *m, FDSet *fds) {
958 HASHMAP_FOREACH(u, m->units, i) {
960 if (fdset_size(fds) <= 0)
963 if (UNIT_VTABLE(u)->distribute_fds) {
964 r = UNIT_VTABLE(u)->distribute_fds(u, fds);
973 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
978 dual_timestamp_get(&m->generators_start_timestamp);
979 manager_run_generators(m);
980 dual_timestamp_get(&m->generators_finish_timestamp);
982 r = lookup_paths_init(
983 &m->lookup_paths, m->running_as, true,
985 m->generator_unit_path,
986 m->generator_unit_path_early,
987 m->generator_unit_path_late);
991 manager_build_unit_path_cache(m);
993 /* If we will deserialize make sure that during enumeration
994 * this is already known, so we increase the counter here
999 /* First, enumerate what we can from all config files */
1000 dual_timestamp_get(&m->units_load_start_timestamp);
1001 r = manager_enumerate(m);
1002 dual_timestamp_get(&m->units_load_finish_timestamp);
1004 /* Second, deserialize if there is something to deserialize */
1006 r = manager_deserialize(m, serialization, fds);
1008 /* Any fds left? Find some unit which wants them. This is
1009 * useful to allow container managers to pass some file
1010 * descriptors to us pre-initialized. This enables
1011 * socket-based activation of entire containers. */
1012 if (fdset_size(fds) > 0) {
1013 q = manager_distribute_fds(m, fds);
1014 if (q < 0 && r == 0)
1018 /* We might have deserialized the notify fd, but if we didn't
1019 * then let's create the bus now */
1020 q = manager_setup_notify(m);
1021 if (q < 0 && r == 0)
1024 /* We might have deserialized the kdbus control fd, but if we
1025 * didn't, then let's create the bus now. */
1026 manager_setup_kdbus(m);
1027 manager_connect_bus(m, !!serialization);
1028 bus_track_coldplug(m, &m->subscribed, &m->deserialized_subscribed);
1030 /* Third, fire things up! */
1031 q = manager_coldplug(m);
1032 if (q < 0 && r == 0)
1035 if (serialization) {
1036 assert(m->n_reloading > 0);
1039 /* Let's wait for the UnitNew/JobNew messages being
1040 * sent, before we notify that the reload is
1042 m->send_reloading_done = true;
1048 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1053 assert(type < _JOB_TYPE_MAX);
1055 assert(mode < _JOB_MODE_MAX);
1057 if (mode == JOB_ISOLATE && type != JOB_START) {
1058 sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1062 if (mode == JOB_ISOLATE && !unit->allow_isolate) {
1063 sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1067 log_debug_unit(unit->id,
1068 "Trying to enqueue job %s/%s/%s", unit->id,
1069 job_type_to_string(type), job_mode_to_string(mode));
1071 job_type_collapse(&type, unit);
1073 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1077 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, override, false,
1078 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1079 mode == JOB_IGNORE_DEPENDENCIES, e);
1083 if (mode == JOB_ISOLATE) {
1084 r = transaction_add_isolate_jobs(tr, m);
1089 r = transaction_activate(tr, m, mode, e);
1093 log_debug_unit(unit->id,
1094 "Enqueued job %s/%s as %u", unit->id,
1095 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1098 *_ret = tr->anchor_job;
1100 transaction_free(tr);
1104 transaction_abort(tr);
1105 transaction_free(tr);
1109 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1114 assert(type < _JOB_TYPE_MAX);
1116 assert(mode < _JOB_MODE_MAX);
1118 r = manager_load_unit(m, name, NULL, NULL, &unit);
1122 return manager_add_job(m, type, unit, mode, override, e, _ret);
1125 Job *manager_get_job(Manager *m, uint32_t id) {
1128 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1131 Unit *manager_get_unit(Manager *m, const char *name) {
1135 return hashmap_get(m->units, name);
1138 unsigned manager_dispatch_load_queue(Manager *m) {
1144 /* Make sure we are not run recursively */
1145 if (m->dispatching_load_queue)
1148 m->dispatching_load_queue = true;
1150 /* Dispatches the load queue. Takes a unit from the queue and
1151 * tries to load its data until the queue is empty */
1153 while ((u = m->load_queue)) {
1154 assert(u->in_load_queue);
1160 m->dispatching_load_queue = false;
1164 int manager_load_unit_prepare(
1176 assert(name || path);
1178 /* This will prepare the unit for loading, but not actually
1179 * load anything from disk. */
1181 if (path && !is_path(path))
1182 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1185 name = basename(path);
1187 t = unit_name_to_type(name);
1189 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, TEMPLATE_INVALID))
1190 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1192 ret = manager_get_unit(m, name);
1198 ret = unit_new(m, unit_vtable[t]->object_size);
1203 ret->fragment_path = strdup(path);
1204 if (!ret->fragment_path) {
1210 r = unit_add_name(ret, name);
1216 unit_add_to_load_queue(ret);
1217 unit_add_to_dbus_queue(ret);
1218 unit_add_to_gc_queue(ret);
1226 int manager_load_unit(
1237 /* This will load the service information files, but not actually
1238 * start any services or anything. */
1240 r = manager_load_unit_prepare(m, name, path, e, _ret);
1244 manager_dispatch_load_queue(m);
1247 *_ret = unit_follow_merge(*_ret);
1252 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1259 HASHMAP_FOREACH(j, s->jobs, i)
1260 job_dump(j, f, prefix);
1263 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1271 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1273 unit_dump(u, f, prefix);
1276 void manager_clear_jobs(Manager *m) {
1281 while ((j = hashmap_first(m->jobs)))
1282 /* No need to recurse. We're cancelling all jobs. */
1283 job_finish_and_invalidate(j, JOB_CANCELED, false);
1286 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1287 Manager *m = userdata;
1293 while ((j = m->run_queue)) {
1294 assert(j->installed);
1295 assert(j->in_run_queue);
1297 job_run_and_invalidate(j);
1300 if (m->n_running_jobs > 0)
1301 manager_watch_jobs_in_progress(m);
1303 if (m->n_on_console > 0)
1304 manager_watch_idle_pipe(m);
1309 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1316 if (m->dispatching_dbus_queue)
1319 m->dispatching_dbus_queue = true;
1321 while ((u = m->dbus_unit_queue)) {
1322 assert(u->in_dbus_queue);
1324 bus_unit_send_change_signal(u);
1328 while ((j = m->dbus_job_queue)) {
1329 assert(j->in_dbus_queue);
1331 bus_job_send_change_signal(j);
1335 m->dispatching_dbus_queue = false;
1337 if (m->send_reloading_done) {
1338 m->send_reloading_done = false;
1340 bus_manager_send_reloading(m, false);
1343 if (m->queued_message)
1344 bus_send_queued_message(m);
1349 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n) {
1350 _cleanup_strv_free_ char **tags = NULL;
1357 tags = strv_split(buf, "\n\r");
1363 log_debug_unit(u->id, "Got notification message for unit %s", u->id);
1365 if (UNIT_VTABLE(u)->notify_message)
1366 UNIT_VTABLE(u)->notify_message(u, pid, tags);
1369 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1370 Manager *m = userdata;
1374 assert(m->notify_fd == fd);
1376 if (revents != EPOLLIN) {
1377 log_warning("Got unexpected poll event for notify fd.");
1383 struct iovec iovec = {
1385 .iov_len = sizeof(buf)-1,
1390 struct cmsghdr cmsghdr;
1391 uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
1394 struct msghdr msghdr = {
1397 .msg_control = &control,
1398 .msg_controllen = sizeof(control),
1400 struct ucred *ucred;
1403 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT);
1408 if (errno == EAGAIN || errno == EINTR)
1414 if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
1415 control.cmsghdr.cmsg_level != SOL_SOCKET ||
1416 control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
1417 control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
1418 log_warning("Received notify message without credentials. Ignoring.");
1422 ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
1424 assert((size_t) n < sizeof(buf));
1427 u = manager_get_unit_by_pid(m, ucred->pid);
1429 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1433 u = hashmap_get(m->watch_pids1, LONG_TO_PTR(ucred->pid));
1435 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1439 u = hashmap_get(m->watch_pids2, LONG_TO_PTR(ucred->pid));
1441 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1446 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1452 static void invoke_sigchld_event(Manager *m, Unit *u, siginfo_t *si) {
1457 log_debug_unit(u->id, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1459 unit_unwatch_pid(u, si->si_pid);
1460 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1463 static int manager_dispatch_sigchld(Manager *m) {
1469 /* First we call waitd() for a PID and do not reap the
1470 * zombie. That way we can still access /proc/$PID for
1471 * it while it is a zombie. */
1472 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1474 if (errno == ECHILD)
1486 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1487 _cleanup_free_ char *name = NULL;
1490 get_process_comm(si.si_pid, &name);
1492 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1493 si.si_pid, strna(name),
1494 sigchld_code_to_string(si.si_code),
1496 strna(si.si_code == CLD_EXITED
1497 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1498 : signal_to_string(si.si_status)));
1500 /* And now figure out the unit this belongs
1501 * to, it might be multiple... */
1502 u = manager_get_unit_by_pid(m, si.si_pid);
1504 invoke_sigchld_event(m, u, &si);
1505 u = hashmap_get(m->watch_pids1, LONG_TO_PTR(si.si_pid));
1507 invoke_sigchld_event(m, u, &si);
1508 u = hashmap_get(m->watch_pids2, LONG_TO_PTR(si.si_pid));
1510 invoke_sigchld_event(m, u, &si);
1513 /* And now, we actually reap the zombie. */
1514 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1525 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1526 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1529 log_debug_unit(name, "Activating special unit %s", name);
1531 r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL);
1533 log_error_unit(name, "Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1538 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1539 Manager *m = userdata;
1541 struct signalfd_siginfo sfsi;
1542 bool sigchld = false;
1545 assert(m->signal_fd == fd);
1547 if (revents != EPOLLIN) {
1548 log_warning("Got unexpected events from signal file descriptor.");
1553 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1554 if (n != sizeof(sfsi)) {
1559 if (errno == EINTR || errno == EAGAIN)
1565 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1566 (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER)
1567 ? LOG_DEBUG : LOG_INFO,
1570 switch (sfsi.ssi_signo) {
1577 if (m->running_as == SYSTEMD_SYSTEM) {
1578 /* This is for compatibility with the
1579 * original sysvinit */
1580 m->exit_code = MANAGER_REEXECUTE;
1587 if (m->running_as == SYSTEMD_SYSTEM) {
1588 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1592 /* Run the exit target if there is one, if not, just exit. */
1593 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1594 m->exit_code = MANAGER_EXIT;
1601 if (m->running_as == SYSTEMD_SYSTEM)
1602 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1604 /* This is a nop on non-init */
1608 if (m->running_as == SYSTEMD_SYSTEM)
1609 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1611 /* This is a nop on non-init */
1617 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1619 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1620 log_info("Trying to reconnect to bus...");
1624 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1625 log_info("Loading D-Bus service...");
1626 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1633 _cleanup_free_ char *dump = NULL;
1634 _cleanup_fclose_ FILE *f = NULL;
1637 f = open_memstream(&dump, &size);
1639 log_warning("Failed to allocate memory stream.");
1643 manager_dump_units(m, f, "\t");
1644 manager_dump_jobs(m, f, "\t");
1647 log_warning("Failed to write status stream");
1652 log_warning("Failed to flush status stream");
1656 log_dump(LOG_INFO, dump);
1661 m->exit_code = MANAGER_RELOAD;
1666 /* Starting SIGRTMIN+0 */
1667 static const char * const target_table[] = {
1668 [0] = SPECIAL_DEFAULT_TARGET,
1669 [1] = SPECIAL_RESCUE_TARGET,
1670 [2] = SPECIAL_EMERGENCY_TARGET,
1671 [3] = SPECIAL_HALT_TARGET,
1672 [4] = SPECIAL_POWEROFF_TARGET,
1673 [5] = SPECIAL_REBOOT_TARGET,
1674 [6] = SPECIAL_KEXEC_TARGET
1677 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1678 static const ManagerExitCode code_table[] = {
1680 [1] = MANAGER_POWEROFF,
1681 [2] = MANAGER_REBOOT,
1685 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1686 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1687 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1688 manager_start_target(m, target_table[idx],
1689 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1693 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1694 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1695 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1699 switch (sfsi.ssi_signo - SIGRTMIN) {
1702 log_debug("Enabling showing of status.");
1703 manager_set_show_status(m, SHOW_STATUS_YES);
1707 log_debug("Disabling showing of status.");
1708 manager_set_show_status(m, SHOW_STATUS_NO);
1712 log_set_max_level(LOG_DEBUG);
1713 log_notice("Setting log level to debug.");
1717 log_set_max_level(LOG_INFO);
1718 log_notice("Setting log level to info.");
1722 if (m->running_as == SYSTEMD_USER) {
1723 m->exit_code = MANAGER_EXIT;
1727 /* This is a nop on init */
1731 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
1732 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1733 log_notice("Setting log target to journal-or-kmsg.");
1737 log_set_target(LOG_TARGET_CONSOLE);
1738 log_notice("Setting log target to console.");
1742 log_set_target(LOG_TARGET_KMSG);
1743 log_notice("Setting log target to kmsg.");
1747 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
1754 manager_dispatch_sigchld(m);
1759 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1760 Manager *m = userdata;
1765 assert(m->time_change_fd == fd);
1767 log_struct(LOG_INFO,
1768 MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
1769 "MESSAGE=Time has been changed",
1772 /* Restart the watch */
1773 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
1774 m->time_change_fd = safe_close(m->time_change_fd);
1776 manager_setup_time_change(m);
1778 HASHMAP_FOREACH(u, m->units, i)
1779 if (UNIT_VTABLE(u)->time_change)
1780 UNIT_VTABLE(u)->time_change(u);
1785 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1786 Manager *m = userdata;
1789 assert(m->idle_pipe[2] == fd);
1791 m->no_console_output = m->n_on_console > 0;
1793 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
1794 manager_close_idle_pipe(m);
1799 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
1800 Manager *m = userdata;
1807 manager_print_jobs_in_progress(m);
1809 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
1810 r = sd_event_source_set_time(source, next);
1814 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
1817 int manager_loop(Manager *m) {
1820 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
1823 m->exit_code = MANAGER_OK;
1825 /* Release the path cache */
1826 set_free_free(m->unit_path_cache);
1827 m->unit_path_cache = NULL;
1829 manager_check_finished(m);
1831 /* There might still be some zombies hanging around from
1832 * before we were exec()'ed. Let's reap them. */
1833 r = manager_dispatch_sigchld(m);
1837 while (m->exit_code == MANAGER_OK) {
1840 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM)
1843 if (!ratelimit_test(&rl)) {
1844 /* Yay, something is going seriously wrong, pause a little */
1845 log_warning("Looping too fast. Throttling execution a little.");
1850 if (manager_dispatch_load_queue(m) > 0)
1853 if (manager_dispatch_gc_queue(m) > 0)
1856 if (manager_dispatch_cleanup_queue(m) > 0)
1859 if (manager_dispatch_cgroup_queue(m) > 0)
1862 if (manager_dispatch_dbus_queue(m) > 0)
1865 /* Sleep for half the watchdog time */
1866 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) {
1867 wait_usec = m->runtime_watchdog / 2;
1871 wait_usec = USEC_INFINITY;
1873 r = sd_event_run(m->event, wait_usec);
1875 log_error("Failed to run event loop: %s", strerror(-r));
1880 return m->exit_code;
1883 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
1884 _cleanup_free_ char *n = NULL;
1892 r = unit_name_from_dbus_path(s, &n);
1896 r = manager_load_unit(m, n, NULL, e, &u);
1905 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
1915 p = startswith(s, "/org/freedesktop/systemd1/job/");
1919 r = safe_atou(p, &id);
1923 j = manager_get_job(m, id);
1932 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
1935 _cleanup_free_ char *p = NULL;
1938 audit_fd = get_audit_fd();
1942 /* Don't generate audit events if the service was already
1943 * started and we're just deserializing */
1944 if (m->n_reloading > 0)
1947 if (m->running_as != SYSTEMD_SYSTEM)
1950 if (u->type != UNIT_SERVICE)
1953 p = unit_name_to_prefix_and_instance(u->id);
1955 log_error_unit(u->id,
1956 "Failed to allocate unit name for audit message: %s", strerror(ENOMEM));
1960 if (audit_log_user_comm_message(audit_fd, type, "", p, NULL, NULL, NULL, success) < 0) {
1961 if (errno == EPERM) {
1962 /* We aren't allowed to send audit messages?
1963 * Then let's not retry again. */
1966 log_warning("Failed to send audit message: %m");
1972 void manager_send_unit_plymouth(Manager *m, Unit *u) {
1973 union sockaddr_union sa = PLYMOUTH_SOCKET;
1976 _cleanup_free_ char *message = NULL;
1977 _cleanup_close_ int fd = -1;
1979 /* Don't generate plymouth events if the service was already
1980 * started and we're just deserializing */
1981 if (m->n_reloading > 0)
1984 if (m->running_as != SYSTEMD_SYSTEM)
1987 if (detect_container(NULL) > 0)
1990 if (u->type != UNIT_SERVICE &&
1991 u->type != UNIT_MOUNT &&
1992 u->type != UNIT_SWAP)
1995 /* We set SOCK_NONBLOCK here so that we rather drop the
1996 * message then wait for plymouth */
1997 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1999 log_error("socket() failed: %m");
2003 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
2005 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2006 log_error("connect() failed: %m");
2010 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2016 if (write(fd, message, n + 1) != n + 1)
2017 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2018 log_error("Failed to write Plymouth message: %m");
2021 void manager_dispatch_bus_name_owner_changed(
2024 const char* old_owner,
2025 const char *new_owner) {
2032 u = hashmap_get(m->watch_bus, name);
2036 UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
2039 int manager_open_serialization(Manager *m, FILE **_f) {
2046 path = m->running_as == SYSTEMD_SYSTEM ? "/run/systemd" : "/tmp";
2047 fd = open_tmpfile(path, O_RDWR|O_CLOEXEC);
2051 log_debug("Serializing state to %s", path);
2053 f = fdopen(fd, "w+");
2064 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2077 fprintf(f, "current-job-id=%i\n", m->current_job_id);
2078 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2079 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2080 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2082 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2083 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2084 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2085 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2088 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2089 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2090 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2091 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2092 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2093 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2094 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2095 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2098 if (!switching_root) {
2099 STRV_FOREACH(e, m->environment) {
2100 _cleanup_free_ char *ce;
2106 fprintf(f, "env=%s\n", *e);
2110 if (m->notify_fd >= 0) {
2113 copy = fdset_put_dup(fds, m->notify_fd);
2117 fprintf(f, "notify-fd=%i\n", copy);
2118 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2121 if (m->kdbus_fd >= 0) {
2124 copy = fdset_put_dup(fds, m->kdbus_fd);
2128 fprintf(f, "kdbus-fd=%i\n", copy);
2131 bus_track_serialize(m->subscribed, f);
2135 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2143 r = unit_serialize(u, f, fds, !switching_root);
2150 assert(m->n_reloading > 0);
2156 r = bus_fdset_add_all(m, fds);
2163 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2169 log_debug("Deserializing state...");
2174 char line[LINE_MAX], *l;
2176 if (!fgets(line, sizeof(line), f)) {
2191 if (startswith(l, "current-job-id=")) {
2194 if (safe_atou32(l+15, &id) < 0)
2195 log_debug("Failed to parse current job id value %s", l+15);
2197 m->current_job_id = MAX(m->current_job_id, id);
2199 } else if (startswith(l, "n-installed-jobs=")) {
2202 if (safe_atou32(l+17, &n) < 0)
2203 log_debug("Failed to parse installed jobs counter %s", l+17);
2205 m->n_installed_jobs += n;
2207 } else if (startswith(l, "n-failed-jobs=")) {
2210 if (safe_atou32(l+14, &n) < 0)
2211 log_debug("Failed to parse failed jobs counter %s", l+14);
2213 m->n_failed_jobs += n;
2215 } else if (startswith(l, "taint-usr=")) {
2218 b = parse_boolean(l+10);
2220 log_debug("Failed to parse taint /usr flag %s", l+10);
2222 m->taint_usr = m->taint_usr || b;
2224 } else if (startswith(l, "firmware-timestamp="))
2225 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2226 else if (startswith(l, "loader-timestamp="))
2227 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2228 else if (startswith(l, "kernel-timestamp="))
2229 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2230 else if (startswith(l, "initrd-timestamp="))
2231 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2232 else if (startswith(l, "userspace-timestamp="))
2233 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2234 else if (startswith(l, "finish-timestamp="))
2235 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2236 else if (startswith(l, "security-start-timestamp="))
2237 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2238 else if (startswith(l, "security-finish-timestamp="))
2239 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2240 else if (startswith(l, "generators-start-timestamp="))
2241 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2242 else if (startswith(l, "generators-finish-timestamp="))
2243 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2244 else if (startswith(l, "units-load-start-timestamp="))
2245 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2246 else if (startswith(l, "units-load-finish-timestamp="))
2247 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2248 else if (startswith(l, "env=")) {
2249 _cleanup_free_ char *uce = NULL;
2252 uce = cunescape(l+4);
2258 e = strv_env_set(m->environment, uce);
2264 strv_free(m->environment);
2267 } else if (startswith(l, "notify-fd=")) {
2270 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2271 log_debug("Failed to parse notify fd: %s", l + 10);
2273 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2274 safe_close(m->notify_fd);
2275 m->notify_fd = fdset_remove(fds, fd);
2278 } else if (startswith(l, "notify-socket=")) {
2287 free(m->notify_socket);
2288 m->notify_socket = n;
2290 } else if (startswith(l, "kdbus-fd=")) {
2293 if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2294 log_debug("Failed to parse kdbus fd: %s", l + 9);
2296 safe_close(m->kdbus_fd);
2297 m->kdbus_fd = fdset_remove(fds, fd);
2300 } else if (bus_track_deserialize_item(&m->deserialized_subscribed, l) == 0)
2301 log_debug("Unknown serialization item '%s'", l);
2306 char name[UNIT_NAME_MAX+2];
2309 if (!fgets(name, sizeof(name), f)) {
2320 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2324 r = unit_deserialize(u, f, fds);
2333 assert(m->n_reloading > 0);
2339 int manager_reload(Manager *m) {
2341 _cleanup_fclose_ FILE *f = NULL;
2342 _cleanup_fdset_free_ FDSet *fds = NULL;
2346 r = manager_open_serialization(m, &f);
2351 bus_manager_send_reloading(m, true);
2359 r = manager_serialize(m, f, fds, false);
2365 if (fseeko(f, 0, SEEK_SET) < 0) {
2370 /* From here on there is no way back. */
2371 manager_clear_jobs_and_units(m);
2372 manager_undo_generators(m);
2373 lookup_paths_free(&m->lookup_paths);
2375 /* Find new unit paths */
2376 manager_run_generators(m);
2378 q = lookup_paths_init(
2379 &m->lookup_paths, m->running_as, true,
2381 m->generator_unit_path,
2382 m->generator_unit_path_early,
2383 m->generator_unit_path_late);
2387 manager_build_unit_path_cache(m);
2389 /* First, enumerate what we can from all config files */
2390 q = manager_enumerate(m);
2394 /* Second, deserialize our stored data */
2395 q = manager_deserialize(m, f, fds);
2402 /* Re-register notify_fd as event source */
2403 q = manager_setup_notify(m);
2407 /* Third, fire things up! */
2408 q = manager_coldplug(m);
2412 assert(m->n_reloading > 0);
2415 m->send_reloading_done = true;
2420 bool manager_is_reloading_or_reexecuting(Manager *m) {
2423 return m->n_reloading != 0;
2426 void manager_reset_failed(Manager *m) {
2432 HASHMAP_FOREACH(u, m->units, i)
2433 unit_reset_failed(u);
2436 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2442 /* Returns true if the unit is inactive or going down */
2443 u = manager_get_unit(m, name);
2447 return unit_inactive_or_pending(u);
2450 void manager_check_finished(Manager *m) {
2451 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2452 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2458 if (m->n_running_jobs == 0)
2459 m->jobs_in_progress_event_source = sd_event_source_unref(m->jobs_in_progress_event_source);
2461 if (hashmap_size(m->jobs) > 0) {
2463 if (m->jobs_in_progress_event_source) {
2464 sd_event_source_set_time(m->jobs_in_progress_event_source,
2465 now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2471 manager_flip_auto_status(m, false);
2473 /* Notify Type=idle units that we are done now */
2474 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
2475 manager_close_idle_pipe(m);
2477 /* Turn off confirm spawn now */
2478 m->confirm_spawn = false;
2480 /* This is no longer the first boot */
2481 manager_set_first_boot(m, false);
2483 if (dual_timestamp_is_set(&m->finish_timestamp))
2486 dual_timestamp_get(&m->finish_timestamp);
2488 if (m->running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0) {
2490 /* Note that m->kernel_usec.monotonic is always at 0,
2491 * and m->firmware_usec.monotonic and
2492 * m->loader_usec.monotonic should be considered
2493 * negative values. */
2495 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2496 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2497 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2498 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2500 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2502 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2503 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2505 if (!log_on_console())
2506 log_struct(LOG_INFO,
2507 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2508 "KERNEL_USEC="USEC_FMT, kernel_usec,
2509 "INITRD_USEC="USEC_FMT, initrd_usec,
2510 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2511 "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2512 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2513 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2514 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2515 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2518 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2521 if (!log_on_console())
2522 log_struct(LOG_INFO,
2523 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2524 "KERNEL_USEC="USEC_FMT, kernel_usec,
2525 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2526 "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.",
2527 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2528 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2529 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2533 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2534 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2536 if (!log_on_console())
2537 log_struct(LOG_INFO,
2538 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2539 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2540 "MESSAGE=Startup finished in %s.",
2541 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2545 SET_FOREACH(u, m->startup_units, i)
2547 cgroup_context_apply(unit_get_cgroup_context(u), unit_get_cgroup_mask(u), u->cgroup_path, manager_state(m));
2549 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2552 "READY=1\nSTATUS=Startup finished in %s.",
2553 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2556 static int create_generator_dir(Manager *m, char **generator, const char *name) {
2567 if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) {
2568 /* systemd --system, not running --test */
2570 p = strappend("/run/systemd/", name);
2574 r = mkdir_p_label(p, 0755);
2576 log_error("Failed to create generator directory %s: %s",
2581 } else if (m->running_as == SYSTEMD_USER) {
2582 const char *s = NULL;
2584 s = getenv("XDG_RUNTIME_DIR");
2587 p = strjoin(s, "/systemd/", name, NULL);
2591 r = mkdir_p_label(p, 0755);
2593 log_error("Failed to create generator directory %s: %s",
2599 /* systemd --system --test */
2601 p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL);
2606 log_error("Failed to create generator directory %s: %m",
2617 static void trim_generator_dir(Manager *m, char **generator) {
2624 if (rmdir(*generator) >= 0) {
2632 void manager_run_generators(Manager *m) {
2633 _cleanup_closedir_ DIR *d = NULL;
2634 const char *generator_path;
2635 const char *argv[5];
2643 generator_path = m->running_as == SYSTEMD_SYSTEM ? SYSTEM_GENERATOR_PATH : USER_GENERATOR_PATH;
2644 d = opendir(generator_path);
2646 if (errno == ENOENT)
2649 log_error("Failed to enumerate generator directory %s: %m",
2654 r = create_generator_dir(m, &m->generator_unit_path, "generator");
2658 r = create_generator_dir(m, &m->generator_unit_path_early, "generator.early");
2662 r = create_generator_dir(m, &m->generator_unit_path_late, "generator.late");
2666 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2667 argv[1] = m->generator_unit_path;
2668 argv[2] = m->generator_unit_path_early;
2669 argv[3] = m->generator_unit_path_late;
2672 RUN_WITH_UMASK(0022)
2673 execute_directory(generator_path, d, DEFAULT_TIMEOUT_USEC, (char**) argv);
2676 trim_generator_dir(m, &m->generator_unit_path);
2677 trim_generator_dir(m, &m->generator_unit_path_early);
2678 trim_generator_dir(m, &m->generator_unit_path_late);
2681 static void remove_generator_dir(Manager *m, char **generator) {
2688 strv_remove(m->lookup_paths.unit_path, *generator);
2689 rm_rf(*generator, false, true, false);
2695 void manager_undo_generators(Manager *m) {
2698 remove_generator_dir(m, &m->generator_unit_path);
2699 remove_generator_dir(m, &m->generator_unit_path_early);
2700 remove_generator_dir(m, &m->generator_unit_path_late);
2703 int manager_environment_add(Manager *m, char **minus, char **plus) {
2704 char **a = NULL, **b = NULL, **l;
2709 if (!strv_isempty(minus)) {
2710 a = strv_env_delete(l, 1, minus);
2717 if (!strv_isempty(plus)) {
2718 b = strv_env_merge(2, l, plus);
2725 if (m->environment != l)
2726 strv_free(m->environment);
2733 manager_clean_environment(m);
2734 strv_sort(m->environment);
2739 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2744 for (i = 0; i < _RLIMIT_MAX; i++) {
2745 if (!default_rlimit[i])
2748 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2756 void manager_recheck_journal(Manager *m) {
2761 if (m->running_as != SYSTEMD_SYSTEM)
2764 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2765 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2766 log_close_journal();
2770 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2771 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2772 log_close_journal();
2776 /* Hmm, OK, so the socket is fully up and the service is up
2777 * too, then let's make use of the thing. */
2781 void manager_set_show_status(Manager *m, ShowStatus mode) {
2783 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
2785 if (m->running_as != SYSTEMD_SYSTEM)
2788 m->show_status = mode;
2791 touch("/run/systemd/show-status");
2793 unlink("/run/systemd/show-status");
2796 static bool manager_get_show_status(Manager *m) {
2799 if (m->running_as != SYSTEMD_SYSTEM)
2802 if (m->no_console_output)
2805 if (!IN_SET(manager_state(m), MANAGER_STARTING, MANAGER_STOPPING))
2808 if (m->show_status > 0)
2811 /* If Plymouth is running make sure we show the status, so
2812 * that there's something nice to see when people press Esc */
2814 return plymouth_running();
2817 void manager_set_first_boot(Manager *m, bool b) {
2820 if (m->running_as != SYSTEMD_SYSTEM)
2826 touch("/run/systemd/first-boot");
2828 unlink("/run/systemd/first-boot");
2831 void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) {
2834 if (!manager_get_show_status(m))
2837 /* XXX We should totally drop the check for ephemeral here
2838 * and thus effectively make 'Type=idle' pointless. */
2839 if (ephemeral && m->n_on_console > 0)
2842 va_start(ap, format);
2843 status_vprintf(status, true, ephemeral, format, ap);
2847 int manager_get_unit_by_path(Manager *m, const char *path, const char *suffix, Unit **_found) {
2848 _cleanup_free_ char *p = NULL;
2856 p = unit_name_from_path(path, suffix);
2860 found = manager_get_unit(m, p);
2870 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
2871 char p[strlen(path)+1];
2877 path_kill_slashes(p);
2879 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
2882 const char *manager_get_runtime_prefix(Manager *m) {
2885 return m->running_as == SYSTEMD_SYSTEM ?
2887 getenv("XDG_RUNTIME_DIR");
2890 ManagerState manager_state(Manager *m) {
2895 /* Did we ever finish booting? If not then we are still starting up */
2896 if (!dual_timestamp_is_set(&m->finish_timestamp))
2897 return MANAGER_STARTING;
2899 /* Is the special shutdown target queued? If so, we are in shutdown state */
2900 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
2901 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))
2902 return MANAGER_STOPPING;
2904 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
2905 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
2906 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2907 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2908 return MANAGER_MAINTENANCE;
2910 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
2911 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2912 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2913 return MANAGER_MAINTENANCE;
2915 /* Are there any failed units? If so, we are in degraded mode */
2916 if (set_size(m->failed_units) > 0)
2917 return MANAGER_DEGRADED;
2919 return MANAGER_RUNNING;
2922 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
2923 [MANAGER_STARTING] = "starting",
2924 [MANAGER_RUNNING] = "running",
2925 [MANAGER_DEGRADED] = "degraded",
2926 [MANAGER_MAINTENANCE] = "maintenance",
2927 [MANAGER_STOPPING] = "stopping",
2930 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);