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 */
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;
532 struct sockaddr_un un;
534 .sa.sa_family = AF_UNIX,
538 /* First free all secondary fields */
539 free(m->notify_socket);
540 m->notify_socket = NULL;
541 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
543 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
545 log_error("Failed to allocate notification socket: %m");
549 if (m->running_as == SYSTEMD_SYSTEM)
550 m->notify_socket = strdup("/run/systemd/notify");
554 e = getenv("XDG_RUNTIME_DIR");
556 log_error("XDG_RUNTIME_DIR is not set: %m");
560 m->notify_socket = strappend(e, "/systemd/notify");
562 if (!m->notify_socket)
565 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
566 r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
568 log_error("bind(@%s) failed: %m", sa.un.sun_path+1);
572 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
574 log_error("SO_PASSCRED failed: %m");
581 log_debug("Using notification socket %s", m->notify_socket);
584 if (!m->notify_event_source) {
585 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
587 log_error("Failed to allocate notify event source: %s", strerror(-r));
591 /* Process signals a bit earlier than SIGCHLD, so that we can
592 * still identify to which service an exit message belongs */
593 r = sd_event_source_set_priority(m->notify_event_source, -7);
595 log_error("Failed to set priority of notify event source: %s", strerror(-r));
603 static int manager_setup_kdbus(Manager *m) {
605 _cleanup_free_ char *p = NULL;
609 if (m->test_run || m->kdbus_fd >= 0)
612 m->kdbus_fd = bus_kernel_create_bus(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", m->running_as == SYSTEMD_SYSTEM, &p);
613 if (m->kdbus_fd < 0) {
614 log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd));
618 log_debug("Successfully set up kdbus on %s", p);
620 /* Create the namespace directory here, so that the contents
621 * of that directory is not visible to non-root users. This is
622 * necessary to ensure that users cannot get access to busses
623 * of virtualized users when no UID namespacing is used. */
624 if (m->running_as == SYSTEMD_SYSTEM)
625 mkdir_p_label("/dev/kdbus/domain", 0700);
631 static int manager_connect_bus(Manager *m, bool reexecuting) {
632 bool try_bus_connect;
642 (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"));
644 /* Try to connect to the busses, if possible. */
645 return bus_init(m, try_bus_connect);
648 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
654 while ((u = m->cleanup_queue)) {
655 assert(u->in_cleanup_queue);
665 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
666 GC_OFFSET_UNSURE, /* No clue */
667 GC_OFFSET_GOOD, /* We still need this unit */
668 GC_OFFSET_BAD, /* We don't need this unit anymore */
672 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
679 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
680 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
681 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
684 if (u->in_cleanup_queue)
687 if (unit_check_gc(u))
690 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
694 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
695 unit_gc_sweep(other, gc_marker);
697 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
700 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
707 /* We were unable to find anything out about this entry, so
708 * let's investigate it later */
709 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
710 unit_add_to_gc_queue(u);
714 /* We definitely know that this one is not useful anymore, so
715 * let's mark it for deletion */
716 u->gc_marker = gc_marker + GC_OFFSET_BAD;
717 unit_add_to_cleanup_queue(u);
721 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
724 static unsigned manager_dispatch_gc_queue(Manager *m) {
731 /* log_debug("Running GC..."); */
733 m->gc_marker += _GC_OFFSET_MAX;
734 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
737 gc_marker = m->gc_marker;
739 while ((u = m->gc_queue)) {
740 assert(u->in_gc_queue);
742 unit_gc_sweep(u, gc_marker);
744 LIST_REMOVE(gc_queue, m->gc_queue, u);
745 u->in_gc_queue = false;
749 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
750 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
751 log_debug_unit(u->id, "Collecting %s", u->id);
752 u->gc_marker = gc_marker + GC_OFFSET_BAD;
753 unit_add_to_cleanup_queue(u);
757 m->n_in_gc_queue = 0;
762 static void manager_clear_jobs_and_units(Manager *m) {
767 while ((u = hashmap_first(m->units)))
770 manager_dispatch_cleanup_queue(m);
772 assert(!m->load_queue);
773 assert(!m->run_queue);
774 assert(!m->dbus_unit_queue);
775 assert(!m->dbus_job_queue);
776 assert(!m->cleanup_queue);
777 assert(!m->gc_queue);
779 assert(hashmap_isempty(m->jobs));
780 assert(hashmap_isempty(m->units));
783 m->n_running_jobs = 0;
786 void manager_free(Manager *m) {
792 manager_clear_jobs_and_units(m);
794 for (c = 0; c < _UNIT_TYPE_MAX; c++)
795 if (unit_vtable[c]->shutdown)
796 unit_vtable[c]->shutdown(m);
798 /* If we reexecute ourselves, we keep the root cgroup
800 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
802 manager_undo_generators(m);
806 hashmap_free(m->units);
807 hashmap_free(m->jobs);
808 hashmap_free(m->watch_pids1);
809 hashmap_free(m->watch_pids2);
810 hashmap_free(m->watch_bus);
812 set_free(m->startup_units);
813 set_free(m->failed_units);
815 sd_event_source_unref(m->signal_event_source);
816 sd_event_source_unref(m->notify_event_source);
817 sd_event_source_unref(m->time_change_event_source);
818 sd_event_source_unref(m->jobs_in_progress_event_source);
819 sd_event_source_unref(m->idle_pipe_event_source);
820 sd_event_source_unref(m->run_queue_event_source);
822 safe_close(m->signal_fd);
823 safe_close(m->notify_fd);
824 safe_close(m->time_change_fd);
825 safe_close(m->kdbus_fd);
827 manager_close_idle_pipe(m);
830 sd_event_unref(m->event);
832 free(m->notify_socket);
834 lookup_paths_free(&m->lookup_paths);
835 strv_free(m->environment);
837 hashmap_free(m->cgroup_unit);
838 set_free_free(m->unit_path_cache);
840 free(m->switch_root);
841 free(m->switch_root_init);
843 for (i = 0; i < _RLIMIT_MAX; i++)
846 assert(hashmap_isempty(m->units_requiring_mounts_for));
847 hashmap_free(m->units_requiring_mounts_for);
852 int manager_enumerate(Manager *m) {
858 /* Let's ask every type to load all units from disk/kernel
859 * that it might know */
860 for (c = 0; c < _UNIT_TYPE_MAX; c++)
861 if (unit_vtable[c]->enumerate) {
862 q = unit_vtable[c]->enumerate(m);
867 manager_dispatch_load_queue(m);
871 static int manager_coldplug(Manager *m) {
879 /* Then, let's set up their initial state. */
880 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
887 q = unit_coldplug(u);
895 static void manager_build_unit_path_cache(Manager *m) {
897 _cleanup_free_ DIR *d = NULL;
902 set_free_free(m->unit_path_cache);
904 m->unit_path_cache = set_new(string_hash_func, string_compare_func);
905 if (!m->unit_path_cache) {
906 log_error("Failed to allocate unit path cache.");
910 /* This simply builds a list of files we know exist, so that
911 * we don't always have to go to disk */
913 STRV_FOREACH(i, m->lookup_paths.unit_path) {
919 log_error("Failed to open directory %s: %m", *i);
923 while ((de = readdir(d))) {
926 if (ignore_file(de->d_name))
929 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
935 r = set_consume(m->unit_path_cache, p);
947 log_error("Failed to build unit path cache: %s", strerror(-r));
949 set_free_free(m->unit_path_cache);
950 m->unit_path_cache = NULL;
954 static int manager_distribute_fds(Manager *m, FDSet *fds) {
961 HASHMAP_FOREACH(u, m->units, i) {
963 if (fdset_size(fds) <= 0)
966 if (UNIT_VTABLE(u)->distribute_fds) {
967 r = UNIT_VTABLE(u)->distribute_fds(u, fds);
976 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
981 dual_timestamp_get(&m->generators_start_timestamp);
982 manager_run_generators(m);
983 dual_timestamp_get(&m->generators_finish_timestamp);
985 r = lookup_paths_init(
986 &m->lookup_paths, m->running_as, true,
988 m->generator_unit_path,
989 m->generator_unit_path_early,
990 m->generator_unit_path_late);
994 manager_build_unit_path_cache(m);
996 /* If we will deserialize make sure that during enumeration
997 * this is already known, so we increase the counter here
1002 /* First, enumerate what we can from all config files */
1003 dual_timestamp_get(&m->units_load_start_timestamp);
1004 r = manager_enumerate(m);
1005 dual_timestamp_get(&m->units_load_finish_timestamp);
1007 /* Second, deserialize if there is something to deserialize */
1008 if (serialization) {
1009 q = manager_deserialize(m, serialization, fds);
1014 /* Any fds left? Find some unit which wants them. This is
1015 * useful to allow container managers to pass some file
1016 * descriptors to us pre-initialized. This enables
1017 * socket-based activation of entire containers. */
1018 if (fdset_size(fds) > 0) {
1019 q = manager_distribute_fds(m, fds);
1024 /* We might have deserialized the notify fd, but if we didn't
1025 * then let's create the bus now */
1026 manager_setup_notify(m);
1028 /* We might have deserialized the kdbus control fd, but if we
1029 * didn't, then let's create the bus now. */
1030 manager_setup_kdbus(m);
1031 manager_connect_bus(m, !!serialization);
1032 bus_track_coldplug(m, &m->subscribed, &m->deserialized_subscribed);
1034 /* Third, fire things up! */
1035 q = manager_coldplug(m);
1039 if (serialization) {
1040 assert(m->n_reloading > 0);
1043 /* Let's wait for the UnitNew/JobNew messages being
1044 * sent, before we notify that the reload is
1046 m->send_reloading_done = true;
1052 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1057 assert(type < _JOB_TYPE_MAX);
1059 assert(mode < _JOB_MODE_MAX);
1061 if (mode == JOB_ISOLATE && type != JOB_START) {
1062 sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1066 if (mode == JOB_ISOLATE && !unit->allow_isolate) {
1067 sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1071 log_debug_unit(unit->id,
1072 "Trying to enqueue job %s/%s/%s", unit->id,
1073 job_type_to_string(type), job_mode_to_string(mode));
1075 job_type_collapse(&type, unit);
1077 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1081 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, override, false,
1082 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1083 mode == JOB_IGNORE_DEPENDENCIES, e);
1087 if (mode == JOB_ISOLATE) {
1088 r = transaction_add_isolate_jobs(tr, m);
1093 r = transaction_activate(tr, m, mode, e);
1097 log_debug_unit(unit->id,
1098 "Enqueued job %s/%s as %u", unit->id,
1099 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1102 *_ret = tr->anchor_job;
1104 transaction_free(tr);
1108 transaction_abort(tr);
1109 transaction_free(tr);
1113 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1118 assert(type < _JOB_TYPE_MAX);
1120 assert(mode < _JOB_MODE_MAX);
1122 r = manager_load_unit(m, name, NULL, NULL, &unit);
1126 return manager_add_job(m, type, unit, mode, override, e, _ret);
1129 Job *manager_get_job(Manager *m, uint32_t id) {
1132 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1135 Unit *manager_get_unit(Manager *m, const char *name) {
1139 return hashmap_get(m->units, name);
1142 unsigned manager_dispatch_load_queue(Manager *m) {
1148 /* Make sure we are not run recursively */
1149 if (m->dispatching_load_queue)
1152 m->dispatching_load_queue = true;
1154 /* Dispatches the load queue. Takes a unit from the queue and
1155 * tries to load its data until the queue is empty */
1157 while ((u = m->load_queue)) {
1158 assert(u->in_load_queue);
1164 m->dispatching_load_queue = false;
1168 int manager_load_unit_prepare(
1180 assert(name || path);
1182 /* This will prepare the unit for loading, but not actually
1183 * load anything from disk. */
1185 if (path && !is_path(path))
1186 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1189 name = basename(path);
1191 t = unit_name_to_type(name);
1193 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, TEMPLATE_INVALID))
1194 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1196 ret = manager_get_unit(m, name);
1202 ret = unit_new(m, unit_vtable[t]->object_size);
1207 ret->fragment_path = strdup(path);
1208 if (!ret->fragment_path) {
1214 r = unit_add_name(ret, name);
1220 unit_add_to_load_queue(ret);
1221 unit_add_to_dbus_queue(ret);
1222 unit_add_to_gc_queue(ret);
1230 int manager_load_unit(
1241 /* This will load the service information files, but not actually
1242 * start any services or anything. */
1244 r = manager_load_unit_prepare(m, name, path, e, _ret);
1248 manager_dispatch_load_queue(m);
1251 *_ret = unit_follow_merge(*_ret);
1256 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1263 HASHMAP_FOREACH(j, s->jobs, i)
1264 job_dump(j, f, prefix);
1267 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1275 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1277 unit_dump(u, f, prefix);
1280 void manager_clear_jobs(Manager *m) {
1285 while ((j = hashmap_first(m->jobs)))
1286 /* No need to recurse. We're cancelling all jobs. */
1287 job_finish_and_invalidate(j, JOB_CANCELED, false);
1290 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1291 Manager *m = userdata;
1297 while ((j = m->run_queue)) {
1298 assert(j->installed);
1299 assert(j->in_run_queue);
1301 job_run_and_invalidate(j);
1304 if (m->n_running_jobs > 0)
1305 manager_watch_jobs_in_progress(m);
1307 if (m->n_on_console > 0)
1308 manager_watch_idle_pipe(m);
1313 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1320 if (m->dispatching_dbus_queue)
1323 m->dispatching_dbus_queue = true;
1325 while ((u = m->dbus_unit_queue)) {
1326 assert(u->in_dbus_queue);
1328 bus_unit_send_change_signal(u);
1332 while ((j = m->dbus_job_queue)) {
1333 assert(j->in_dbus_queue);
1335 bus_job_send_change_signal(j);
1339 m->dispatching_dbus_queue = false;
1341 if (m->send_reloading_done) {
1342 m->send_reloading_done = false;
1344 bus_manager_send_reloading(m, false);
1347 if (m->queued_message)
1348 bus_send_queued_message(m);
1353 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n) {
1354 _cleanup_strv_free_ char **tags = NULL;
1361 tags = strv_split(buf, "\n\r");
1367 log_debug_unit(u->id, "Got notification message for unit %s", u->id);
1369 if (UNIT_VTABLE(u)->notify_message)
1370 UNIT_VTABLE(u)->notify_message(u, pid, tags);
1373 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1374 Manager *m = userdata;
1378 assert(m->notify_fd == fd);
1380 if (revents != EPOLLIN) {
1381 log_warning("Got unexpected poll event for notify fd.");
1387 struct iovec iovec = {
1389 .iov_len = sizeof(buf)-1,
1394 struct cmsghdr cmsghdr;
1395 uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
1398 struct msghdr msghdr = {
1401 .msg_control = &control,
1402 .msg_controllen = sizeof(control),
1404 struct ucred *ucred;
1407 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT);
1412 if (errno == EAGAIN || errno == EINTR)
1418 if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
1419 control.cmsghdr.cmsg_level != SOL_SOCKET ||
1420 control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
1421 control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
1422 log_warning("Received notify message without credentials. Ignoring.");
1426 ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
1428 assert((size_t) n < sizeof(buf));
1431 u = manager_get_unit_by_pid(m, ucred->pid);
1433 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1437 u = hashmap_get(m->watch_pids1, LONG_TO_PTR(ucred->pid));
1439 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1443 u = hashmap_get(m->watch_pids2, LONG_TO_PTR(ucred->pid));
1445 manager_invoke_notify_message(m, u, ucred->pid, buf, n);
1450 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1456 static void invoke_sigchld_event(Manager *m, Unit *u, siginfo_t *si) {
1461 log_debug_unit(u->id, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1463 unit_unwatch_pid(u, si->si_pid);
1464 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1467 static int manager_dispatch_sigchld(Manager *m) {
1473 /* First we call waitd() for a PID and do not reap the
1474 * zombie. That way we can still access /proc/$PID for
1475 * it while it is a zombie. */
1476 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1478 if (errno == ECHILD)
1490 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1491 _cleanup_free_ char *name = NULL;
1494 get_process_comm(si.si_pid, &name);
1496 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1497 si.si_pid, strna(name),
1498 sigchld_code_to_string(si.si_code),
1500 strna(si.si_code == CLD_EXITED
1501 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1502 : signal_to_string(si.si_status)));
1504 /* And now figure out the unit this belongs
1505 * to, it might be multiple... */
1506 u = manager_get_unit_by_pid(m, si.si_pid);
1508 invoke_sigchld_event(m, u, &si);
1509 u = hashmap_get(m->watch_pids1, LONG_TO_PTR(si.si_pid));
1511 invoke_sigchld_event(m, u, &si);
1512 u = hashmap_get(m->watch_pids2, LONG_TO_PTR(si.si_pid));
1514 invoke_sigchld_event(m, u, &si);
1517 /* And now, we actually reap the zombie. */
1518 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1529 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1530 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1533 log_debug_unit(name, "Activating special unit %s", name);
1535 r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL);
1537 log_error_unit(name, "Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1542 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1543 Manager *m = userdata;
1545 struct signalfd_siginfo sfsi;
1546 bool sigchld = false;
1549 assert(m->signal_fd == fd);
1551 if (revents != EPOLLIN) {
1552 log_warning("Got unexpected events from signal file descriptor.");
1557 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1558 if (n != sizeof(sfsi)) {
1563 if (errno == EINTR || errno == EAGAIN)
1569 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1570 (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER)
1571 ? LOG_DEBUG : LOG_INFO,
1574 switch (sfsi.ssi_signo) {
1581 if (m->running_as == SYSTEMD_SYSTEM) {
1582 /* This is for compatibility with the
1583 * original sysvinit */
1584 m->exit_code = MANAGER_REEXECUTE;
1591 if (m->running_as == SYSTEMD_SYSTEM) {
1592 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1596 /* Run the exit target if there is one, if not, just exit. */
1597 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1598 m->exit_code = MANAGER_EXIT;
1605 if (m->running_as == SYSTEMD_SYSTEM)
1606 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1608 /* This is a nop on non-init */
1612 if (m->running_as == SYSTEMD_SYSTEM)
1613 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1615 /* This is a nop on non-init */
1621 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1623 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1624 log_info("Trying to reconnect to bus...");
1628 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1629 log_info("Loading D-Bus service...");
1630 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1637 _cleanup_free_ char *dump = NULL;
1638 _cleanup_fclose_ FILE *f = NULL;
1641 f = open_memstream(&dump, &size);
1643 log_warning("Failed to allocate memory stream.");
1647 manager_dump_units(m, f, "\t");
1648 manager_dump_jobs(m, f, "\t");
1651 log_warning("Failed to write status stream");
1656 log_warning("Failed to flush status stream");
1660 log_dump(LOG_INFO, dump);
1665 m->exit_code = MANAGER_RELOAD;
1670 /* Starting SIGRTMIN+0 */
1671 static const char * const target_table[] = {
1672 [0] = SPECIAL_DEFAULT_TARGET,
1673 [1] = SPECIAL_RESCUE_TARGET,
1674 [2] = SPECIAL_EMERGENCY_TARGET,
1675 [3] = SPECIAL_HALT_TARGET,
1676 [4] = SPECIAL_POWEROFF_TARGET,
1677 [5] = SPECIAL_REBOOT_TARGET,
1678 [6] = SPECIAL_KEXEC_TARGET
1681 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1682 static const ManagerExitCode code_table[] = {
1684 [1] = MANAGER_POWEROFF,
1685 [2] = MANAGER_REBOOT,
1689 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1690 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1691 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1692 manager_start_target(m, target_table[idx],
1693 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1697 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1698 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1699 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1703 switch (sfsi.ssi_signo - SIGRTMIN) {
1706 log_debug("Enabling showing of status.");
1707 manager_set_show_status(m, SHOW_STATUS_YES);
1711 log_debug("Disabling showing of status.");
1712 manager_set_show_status(m, SHOW_STATUS_NO);
1716 log_set_max_level(LOG_DEBUG);
1717 log_notice("Setting log level to debug.");
1721 log_set_max_level(LOG_INFO);
1722 log_notice("Setting log level to info.");
1726 if (m->running_as == SYSTEMD_USER) {
1727 m->exit_code = MANAGER_EXIT;
1731 /* This is a nop on init */
1735 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1736 log_notice("Setting log target to journal-or-kmsg.");
1740 log_set_target(LOG_TARGET_CONSOLE);
1741 log_notice("Setting log target to console.");
1745 log_set_target(LOG_TARGET_KMSG);
1746 log_notice("Setting log target to kmsg.");
1750 log_set_target(LOG_TARGET_SYSLOG_OR_KMSG);
1751 log_notice("Setting log target to syslog-or-kmsg.");
1755 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
1762 manager_dispatch_sigchld(m);
1767 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1768 Manager *m = userdata;
1773 assert(m->time_change_fd == fd);
1775 log_struct(LOG_INFO,
1776 MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
1777 "MESSAGE=Time has been changed",
1780 /* Restart the watch */
1781 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
1782 m->time_change_fd = safe_close(m->time_change_fd);
1784 manager_setup_time_change(m);
1786 HASHMAP_FOREACH(u, m->units, i)
1787 if (UNIT_VTABLE(u)->time_change)
1788 UNIT_VTABLE(u)->time_change(u);
1793 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1794 Manager *m = userdata;
1797 assert(m->idle_pipe[2] == fd);
1799 m->no_console_output = m->n_on_console > 0;
1801 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
1802 manager_close_idle_pipe(m);
1807 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
1808 Manager *m = userdata;
1815 manager_print_jobs_in_progress(m);
1817 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
1818 r = sd_event_source_set_time(source, next);
1822 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
1825 int manager_loop(Manager *m) {
1828 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
1831 m->exit_code = MANAGER_OK;
1833 /* Release the path cache */
1834 set_free_free(m->unit_path_cache);
1835 m->unit_path_cache = NULL;
1837 manager_check_finished(m);
1839 /* There might still be some zombies hanging around from
1840 * before we were exec()'ed. Let's reap them. */
1841 r = manager_dispatch_sigchld(m);
1845 while (m->exit_code == MANAGER_OK) {
1848 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM)
1851 if (!ratelimit_test(&rl)) {
1852 /* Yay, something is going seriously wrong, pause a little */
1853 log_warning("Looping too fast. Throttling execution a little.");
1858 if (manager_dispatch_load_queue(m) > 0)
1861 if (manager_dispatch_gc_queue(m) > 0)
1864 if (manager_dispatch_cleanup_queue(m) > 0)
1867 if (manager_dispatch_cgroup_queue(m) > 0)
1870 if (manager_dispatch_dbus_queue(m) > 0)
1873 /* Sleep for half the watchdog time */
1874 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) {
1875 wait_usec = m->runtime_watchdog / 2;
1879 wait_usec = (usec_t) -1;
1881 r = sd_event_run(m->event, wait_usec);
1883 log_error("Failed to run event loop: %s", strerror(-r));
1888 return m->exit_code;
1891 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
1892 _cleanup_free_ char *n = NULL;
1900 r = unit_name_from_dbus_path(s, &n);
1904 r = manager_load_unit(m, n, NULL, e, &u);
1913 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
1923 p = startswith(s, "/org/freedesktop/systemd1/job/");
1927 r = safe_atou(p, &id);
1931 j = manager_get_job(m, id);
1940 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
1943 _cleanup_free_ char *p = NULL;
1946 audit_fd = get_audit_fd();
1950 /* Don't generate audit events if the service was already
1951 * started and we're just deserializing */
1952 if (m->n_reloading > 0)
1955 if (m->running_as != SYSTEMD_SYSTEM)
1958 if (u->type != UNIT_SERVICE)
1961 p = unit_name_to_prefix_and_instance(u->id);
1963 log_error_unit(u->id,
1964 "Failed to allocate unit name for audit message: %s", strerror(ENOMEM));
1968 if (audit_log_user_comm_message(audit_fd, type, "", p, NULL, NULL, NULL, success) < 0) {
1969 if (errno == EPERM) {
1970 /* We aren't allowed to send audit messages?
1971 * Then let's not retry again. */
1974 log_warning("Failed to send audit message: %m");
1980 void manager_send_unit_plymouth(Manager *m, Unit *u) {
1981 union sockaddr_union sa = {
1982 .un.sun_family = AF_UNIX,
1983 .un.sun_path = "\0/org/freedesktop/plymouthd",
1987 _cleanup_free_ char *message = NULL;
1988 _cleanup_close_ int fd = -1;
1990 /* Don't generate plymouth events if the service was already
1991 * started and we're just deserializing */
1992 if (m->n_reloading > 0)
1995 if (m->running_as != SYSTEMD_SYSTEM)
1998 if (detect_container(NULL) > 0)
2001 if (u->type != UNIT_SERVICE &&
2002 u->type != UNIT_MOUNT &&
2003 u->type != UNIT_SWAP)
2006 /* We set SOCK_NONBLOCK here so that we rather drop the
2007 * message then wait for plymouth */
2008 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2010 log_error("socket() failed: %m");
2014 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
2016 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2017 log_error("connect() failed: %m");
2021 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2027 if (write(fd, message, n + 1) != n + 1)
2028 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2029 log_error("Failed to write Plymouth message: %m");
2032 void manager_dispatch_bus_name_owner_changed(
2035 const char* old_owner,
2036 const char *new_owner) {
2043 u = hashmap_get(m->watch_bus, name);
2047 UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
2050 int manager_open_serialization(Manager *m, FILE **_f) {
2057 path = m->running_as == SYSTEMD_SYSTEM ? "/run/systemd" : "/tmp";
2058 fd = open_tmpfile(path, O_RDWR|O_CLOEXEC);
2062 log_debug("Serializing state to %s", path);
2064 f = fdopen(fd, "w+");
2075 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2088 fprintf(f, "current-job-id=%i\n", m->current_job_id);
2089 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2090 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2091 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2093 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2094 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2095 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2096 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2099 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2100 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2101 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2102 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2103 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2104 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2105 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2106 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2109 if (!switching_root) {
2110 STRV_FOREACH(e, m->environment) {
2111 _cleanup_free_ char *ce;
2117 fprintf(f, "env=%s\n", *e);
2121 if (m->notify_fd >= 0) {
2124 copy = fdset_put_dup(fds, m->notify_fd);
2128 fprintf(f, "notify-fd=%i\n", copy);
2129 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2132 if (m->kdbus_fd >= 0) {
2135 copy = fdset_put_dup(fds, m->kdbus_fd);
2139 fprintf(f, "kdbus-fd=%i\n", copy);
2142 bus_track_serialize(m->subscribed, f);
2146 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2154 r = unit_serialize(u, f, fds, !switching_root);
2161 assert(m->n_reloading > 0);
2167 r = bus_fdset_add_all(m, fds);
2174 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2180 log_debug("Deserializing state...");
2185 char line[LINE_MAX], *l;
2187 if (!fgets(line, sizeof(line), f)) {
2202 if (startswith(l, "current-job-id=")) {
2205 if (safe_atou32(l+15, &id) < 0)
2206 log_debug("Failed to parse current job id value %s", l+15);
2208 m->current_job_id = MAX(m->current_job_id, id);
2210 } else if (startswith(l, "n-installed-jobs=")) {
2213 if (safe_atou32(l+17, &n) < 0)
2214 log_debug("Failed to parse installed jobs counter %s", l+17);
2216 m->n_installed_jobs += n;
2218 } else if (startswith(l, "n-failed-jobs=")) {
2221 if (safe_atou32(l+14, &n) < 0)
2222 log_debug("Failed to parse failed jobs counter %s", l+14);
2224 m->n_failed_jobs += n;
2226 } else if (startswith(l, "taint-usr=")) {
2229 b = parse_boolean(l+10);
2231 log_debug("Failed to parse taint /usr flag %s", l+10);
2233 m->taint_usr = m->taint_usr || b;
2235 } else if (startswith(l, "firmware-timestamp="))
2236 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2237 else if (startswith(l, "loader-timestamp="))
2238 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2239 else if (startswith(l, "kernel-timestamp="))
2240 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2241 else if (startswith(l, "initrd-timestamp="))
2242 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2243 else if (startswith(l, "userspace-timestamp="))
2244 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2245 else if (startswith(l, "finish-timestamp="))
2246 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2247 else if (startswith(l, "security-start-timestamp="))
2248 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2249 else if (startswith(l, "security-finish-timestamp="))
2250 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2251 else if (startswith(l, "generators-start-timestamp="))
2252 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2253 else if (startswith(l, "generators-finish-timestamp="))
2254 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2255 else if (startswith(l, "units-load-start-timestamp="))
2256 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2257 else if (startswith(l, "units-load-finish-timestamp="))
2258 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2259 else if (startswith(l, "env=")) {
2260 _cleanup_free_ char *uce = NULL;
2263 uce = cunescape(l+4);
2269 e = strv_env_set(m->environment, uce);
2275 strv_free(m->environment);
2278 } else if (startswith(l, "notify-fd=")) {
2281 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2282 log_debug("Failed to parse notify fd: %s", l + 10);
2284 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2285 safe_close(m->notify_fd);
2286 m->notify_fd = fdset_remove(fds, fd);
2289 } else if (startswith(l, "notify-socket=")) {
2298 free(m->notify_socket);
2299 m->notify_socket = n;
2301 } else if (startswith(l, "kdbus-fd=")) {
2304 if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2305 log_debug("Failed to parse kdbus fd: %s", l + 9);
2307 safe_close(m->kdbus_fd);
2308 m->kdbus_fd = fdset_remove(fds, fd);
2311 } else if (bus_track_deserialize_item(&m->deserialized_subscribed, l) == 0)
2312 log_debug("Unknown serialization item '%s'", l);
2317 char name[UNIT_NAME_MAX+2];
2320 if (!fgets(name, sizeof(name), f)) {
2331 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2335 r = unit_deserialize(u, f, fds);
2344 assert(m->n_reloading > 0);
2350 int manager_reload(Manager *m) {
2352 _cleanup_fclose_ FILE *f = NULL;
2353 _cleanup_fdset_free_ FDSet *fds = NULL;
2357 r = manager_open_serialization(m, &f);
2362 bus_manager_send_reloading(m, true);
2370 r = manager_serialize(m, f, fds, false);
2376 if (fseeko(f, 0, SEEK_SET) < 0) {
2381 /* From here on there is no way back. */
2382 manager_clear_jobs_and_units(m);
2383 manager_undo_generators(m);
2384 lookup_paths_free(&m->lookup_paths);
2386 /* Find new unit paths */
2387 manager_run_generators(m);
2389 q = lookup_paths_init(
2390 &m->lookup_paths, m->running_as, true,
2392 m->generator_unit_path,
2393 m->generator_unit_path_early,
2394 m->generator_unit_path_late);
2398 manager_build_unit_path_cache(m);
2400 /* First, enumerate what we can from all config files */
2401 q = manager_enumerate(m);
2405 /* Second, deserialize our stored data */
2406 q = manager_deserialize(m, f, fds);
2413 /* Re-register notify_fd as event source */
2414 q = manager_setup_notify(m);
2418 /* Third, fire things up! */
2419 q = manager_coldplug(m);
2423 assert(m->n_reloading > 0);
2426 m->send_reloading_done = true;
2431 bool manager_is_reloading_or_reexecuting(Manager *m) {
2434 return m->n_reloading != 0;
2437 void manager_reset_failed(Manager *m) {
2443 HASHMAP_FOREACH(u, m->units, i)
2444 unit_reset_failed(u);
2447 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2453 /* Returns true if the unit is inactive or going down */
2454 u = manager_get_unit(m, name);
2458 return unit_inactive_or_pending(u);
2461 void manager_check_finished(Manager *m) {
2462 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2463 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2469 if (m->n_running_jobs == 0)
2470 m->jobs_in_progress_event_source = sd_event_source_unref(m->jobs_in_progress_event_source);
2472 if (hashmap_size(m->jobs) > 0) {
2474 if (m->jobs_in_progress_event_source) {
2475 sd_event_source_set_time(m->jobs_in_progress_event_source,
2476 now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2482 manager_flip_auto_status(m, false);
2484 /* Notify Type=idle units that we are done now */
2485 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
2486 manager_close_idle_pipe(m);
2488 /* Turn off confirm spawn now */
2489 m->confirm_spawn = false;
2491 /* This is no longer the first boot */
2492 manager_set_first_boot(m, false);
2494 if (dual_timestamp_is_set(&m->finish_timestamp))
2497 dual_timestamp_get(&m->finish_timestamp);
2499 if (m->running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0) {
2501 /* Note that m->kernel_usec.monotonic is always at 0,
2502 * and m->firmware_usec.monotonic and
2503 * m->loader_usec.monotonic should be considered
2504 * negative values. */
2506 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2507 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2508 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2509 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2511 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2513 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2514 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2516 if (!log_on_console())
2517 log_struct(LOG_INFO,
2518 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2519 "KERNEL_USEC="USEC_FMT, kernel_usec,
2520 "INITRD_USEC="USEC_FMT, initrd_usec,
2521 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2522 "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2523 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2524 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2525 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2526 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2529 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2532 if (!log_on_console())
2533 log_struct(LOG_INFO,
2534 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2535 "KERNEL_USEC="USEC_FMT, kernel_usec,
2536 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2537 "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.",
2538 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2539 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2540 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2544 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2545 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2547 if (!log_on_console())
2548 log_struct(LOG_INFO,
2549 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2550 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2551 "MESSAGE=Startup finished in %s.",
2552 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2556 SET_FOREACH(u, m->startup_units, i)
2558 cgroup_context_apply(unit_get_cgroup_context(u), unit_get_cgroup_mask(u), u->cgroup_path, manager_state(m));
2560 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2563 "READY=1\nSTATUS=Startup finished in %s.",
2564 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2567 static int create_generator_dir(Manager *m, char **generator, const char *name) {
2578 if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) {
2579 /* systemd --system, not running --test */
2581 p = strappend("/run/systemd/", name);
2585 r = mkdir_p_label(p, 0755);
2587 log_error("Failed to create generator directory %s: %s",
2592 } else if (m->running_as == SYSTEMD_USER) {
2593 const char *s = NULL;
2595 s = getenv("XDG_RUNTIME_DIR");
2598 p = strjoin(s, "/systemd/", name, NULL);
2602 r = mkdir_p_label(p, 0755);
2604 log_error("Failed to create generator directory %s: %s",
2610 /* systemd --system --test */
2612 p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL);
2617 log_error("Failed to create generator directory %s: %m",
2628 static void trim_generator_dir(Manager *m, char **generator) {
2635 if (rmdir(*generator) >= 0) {
2643 void manager_run_generators(Manager *m) {
2644 _cleanup_closedir_ DIR *d = NULL;
2645 const char *generator_path;
2646 const char *argv[5];
2654 generator_path = m->running_as == SYSTEMD_SYSTEM ? SYSTEM_GENERATOR_PATH : USER_GENERATOR_PATH;
2655 d = opendir(generator_path);
2657 if (errno == ENOENT)
2660 log_error("Failed to enumerate generator directory %s: %m",
2665 r = create_generator_dir(m, &m->generator_unit_path, "generator");
2669 r = create_generator_dir(m, &m->generator_unit_path_early, "generator.early");
2673 r = create_generator_dir(m, &m->generator_unit_path_late, "generator.late");
2677 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2678 argv[1] = m->generator_unit_path;
2679 argv[2] = m->generator_unit_path_early;
2680 argv[3] = m->generator_unit_path_late;
2683 RUN_WITH_UMASK(0022)
2684 execute_directory(generator_path, d, DEFAULT_TIMEOUT_USEC, (char**) argv);
2687 trim_generator_dir(m, &m->generator_unit_path);
2688 trim_generator_dir(m, &m->generator_unit_path_early);
2689 trim_generator_dir(m, &m->generator_unit_path_late);
2692 static void remove_generator_dir(Manager *m, char **generator) {
2699 strv_remove(m->lookup_paths.unit_path, *generator);
2700 rm_rf(*generator, false, true, false);
2706 void manager_undo_generators(Manager *m) {
2709 remove_generator_dir(m, &m->generator_unit_path);
2710 remove_generator_dir(m, &m->generator_unit_path_early);
2711 remove_generator_dir(m, &m->generator_unit_path_late);
2714 int manager_environment_add(Manager *m, char **minus, char **plus) {
2715 char **a = NULL, **b = NULL, **l;
2720 if (!strv_isempty(minus)) {
2721 a = strv_env_delete(l, 1, minus);
2728 if (!strv_isempty(plus)) {
2729 b = strv_env_merge(2, l, plus);
2736 if (m->environment != l)
2737 strv_free(m->environment);
2744 manager_clean_environment(m);
2745 strv_sort(m->environment);
2750 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2755 for (i = 0; i < _RLIMIT_MAX; i++) {
2756 if (!default_rlimit[i])
2759 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2767 void manager_recheck_journal(Manager *m) {
2772 if (m->running_as != SYSTEMD_SYSTEM)
2775 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2776 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2777 log_close_journal();
2781 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2782 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2783 log_close_journal();
2787 /* Hmm, OK, so the socket is fully up and the service is up
2788 * too, then let's make use of the thing. */
2792 void manager_set_show_status(Manager *m, ShowStatus mode) {
2794 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
2796 if (m->running_as != SYSTEMD_SYSTEM)
2799 m->show_status = mode;
2802 touch("/run/systemd/show-status");
2804 unlink("/run/systemd/show-status");
2807 static bool manager_get_show_status(Manager *m) {
2810 if (m->running_as != SYSTEMD_SYSTEM)
2813 if (m->no_console_output)
2816 if (!IN_SET(manager_state(m), MANAGER_STARTING, MANAGER_STOPPING))
2819 if (m->show_status > 0)
2822 /* If Plymouth is running make sure we show the status, so
2823 * that there's something nice to see when people press Esc */
2825 return plymouth_running();
2828 void manager_set_first_boot(Manager *m, bool b) {
2831 if (m->running_as != SYSTEMD_SYSTEM)
2837 touch("/run/systemd/first-boot");
2839 unlink("/run/systemd/first-boot");
2842 void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) {
2845 if (!manager_get_show_status(m))
2848 /* XXX We should totally drop the check for ephemeral here
2849 * and thus effectively make 'Type=idle' pointless. */
2850 if (ephemeral && m->n_on_console > 0)
2853 va_start(ap, format);
2854 status_vprintf(status, true, ephemeral, format, ap);
2858 int manager_get_unit_by_path(Manager *m, const char *path, const char *suffix, Unit **_found) {
2859 _cleanup_free_ char *p = NULL;
2867 p = unit_name_from_path(path, suffix);
2871 found = manager_get_unit(m, p);
2881 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
2882 char p[strlen(path)+1];
2888 path_kill_slashes(p);
2890 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
2893 const char *manager_get_runtime_prefix(Manager *m) {
2896 return m->running_as == SYSTEMD_SYSTEM ?
2898 getenv("XDG_RUNTIME_DIR");
2901 ManagerState manager_state(Manager *m) {
2906 /* Did we ever finish booting? If not then we are still starting up */
2907 if (!dual_timestamp_is_set(&m->finish_timestamp))
2908 return MANAGER_STARTING;
2910 /* Is the special shutdown target queued? If so, we are in shutdown state */
2911 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
2912 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))
2913 return MANAGER_STOPPING;
2915 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
2916 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
2917 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2918 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2919 return MANAGER_MAINTENANCE;
2921 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
2922 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2923 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2924 return MANAGER_MAINTENANCE;
2926 /* Are there any failed units? If so, we are in degraded mode */
2927 if (set_size(m->failed_units) > 0)
2928 return MANAGER_DEGRADED;
2930 return MANAGER_RUNNING;
2933 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
2934 [MANAGER_STARTING] = "starting",
2935 [MANAGER_RUNNING] = "running",
2936 [MANAGER_DEGRADED] = "degraded",
2937 [MANAGER_MAINTENANCE] = "maintenance",
2938 [MANAGER_STOPPING] = "stopping",
2941 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);