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"
78 #include "time-util.h"
80 /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */
81 #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC)
83 /* Initial delay and the interval for printing status messages about running jobs */
84 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
85 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
86 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
88 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
89 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
90 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
91 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
92 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
93 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
95 static int manager_watch_jobs_in_progress(Manager *m) {
100 if (m->jobs_in_progress_event_source)
103 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
104 return sd_event_add_time(
106 &m->jobs_in_progress_event_source,
109 manager_dispatch_jobs_in_progress, m);
112 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED_ON)-1) + sizeof(ANSI_HIGHLIGHT_RED_ON)-1 + 2*(sizeof(ANSI_HIGHLIGHT_OFF)-1))
114 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
117 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
118 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
122 p = mempset(p, ' ', pos-2);
123 p = stpcpy(p, ANSI_RED_ON);
127 if (pos > 0 && pos <= width) {
128 p = stpcpy(p, ANSI_HIGHLIGHT_RED_ON);
132 p = stpcpy(p, ANSI_HIGHLIGHT_OFF);
135 p = stpcpy(p, ANSI_RED_ON);
138 p = mempset(p, ' ', width-1-pos);
139 strcpy(p, ANSI_HIGHLIGHT_OFF);
143 void manager_flip_auto_status(Manager *m, bool enable) {
147 if (m->show_status == SHOW_STATUS_AUTO)
148 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
150 if (m->show_status == SHOW_STATUS_TEMPORARY)
151 manager_set_show_status(m, SHOW_STATUS_AUTO);
155 static void manager_print_jobs_in_progress(Manager *m) {
156 _cleanup_free_ char *job_of_n = NULL;
159 unsigned counter = 0, print_nr;
160 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
162 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
167 manager_flip_auto_status(m, true);
169 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
171 HASHMAP_FOREACH(j, m->jobs, i)
172 if (j->state == JOB_RUNNING && counter++ == print_nr)
175 /* m->n_running_jobs must be consistent with the contents of m->jobs,
176 * so the above loop must have succeeded in finding j. */
177 assert(counter == print_nr + 1);
180 cylon_pos = m->jobs_in_progress_iteration % 14;
182 cylon_pos = 14 - cylon_pos;
183 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
185 m->jobs_in_progress_iteration++;
187 if (m->n_running_jobs > 1)
188 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
191 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
192 if (job_get_timeout(j, &x) > 0)
193 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
195 manager_status_printf(m, true, cylon,
196 "%sA %s job is running for %s (%s / %s)",
198 job_type_to_string(j->type),
199 unit_description(j->unit),
204 static int manager_watch_idle_pipe(Manager *m) {
209 if (m->idle_pipe_event_source)
212 if (m->idle_pipe[2] < 0)
215 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
217 log_error("Failed to watch idle pipe: %s", strerror(-r));
224 static void manager_close_idle_pipe(Manager *m) {
227 safe_close_pair(m->idle_pipe);
228 safe_close_pair(m->idle_pipe + 2);
231 static int manager_setup_time_change(Manager *m) {
234 /* We only care for the cancellation event, hence we set the
235 * timeout to the latest possible value. */
236 struct itimerspec its = {
237 .it_value.tv_sec = TIME_T_MAX,
241 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
246 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
247 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
249 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
250 if (m->time_change_fd < 0) {
251 log_error("Failed to create timerfd: %m");
255 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
256 log_debug("Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
257 m->time_change_fd = safe_close(m->time_change_fd);
261 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
263 log_error("Failed to create time change event source: %s", strerror(-r));
267 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
272 static int enable_special_signals(Manager *m) {
273 _cleanup_close_ int fd = -1;
277 /* Enable that we get SIGINT on control-alt-del. In containers
278 * this will fail with EPERM (older) or EINVAL (newer), so
280 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
281 log_warning("Failed to enable ctrl-alt-del handling: %m");
283 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
285 /* Support systems without virtual console */
287 log_warning("Failed to open /dev/tty0: %m");
289 /* Enable that we get SIGWINCH on kbrequest */
290 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
291 log_warning("Failed to enable kbrequest handling: %m");
297 static int manager_setup_signals(Manager *m) {
298 struct sigaction sa = {
299 .sa_handler = SIG_DFL,
300 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
310 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
312 /* We make liberal use of realtime signals here. On
313 * Linux/glibc we have 30 of them (with the exception of Linux
314 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
317 assert_se(sigemptyset(&mask) == 0);
318 sigset_add_many(&mask,
319 SIGCHLD, /* Child died */
320 SIGTERM, /* Reexecute daemon */
321 SIGHUP, /* Reload configuration */
322 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
323 SIGUSR2, /* systemd: dump status */
324 SIGINT, /* Kernel sends us this on control-alt-del */
325 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
326 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
328 SIGRTMIN+0, /* systemd: start default.target */
329 SIGRTMIN+1, /* systemd: isolate rescue.target */
330 SIGRTMIN+2, /* systemd: isolate emergency.target */
331 SIGRTMIN+3, /* systemd: start halt.target */
332 SIGRTMIN+4, /* systemd: start poweroff.target */
333 SIGRTMIN+5, /* systemd: start reboot.target */
334 SIGRTMIN+6, /* systemd: start kexec.target */
336 /* ... space for more special targets ... */
338 SIGRTMIN+13, /* systemd: Immediate halt */
339 SIGRTMIN+14, /* systemd: Immediate poweroff */
340 SIGRTMIN+15, /* systemd: Immediate reboot */
341 SIGRTMIN+16, /* systemd: Immediate kexec */
343 /* ... space for more immediate system state changes ... */
345 SIGRTMIN+20, /* systemd: enable status messages */
346 SIGRTMIN+21, /* systemd: disable status messages */
347 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
348 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
349 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
351 /* .. one free signal here ... */
353 #if !defined(__hppa64__) && !defined(__hppa__)
354 /* Apparently Linux on hppa has fewer RT
355 * signals (SIGRTMAX is SIGRTMIN+25 there),
356 * hence let's not try to make use of them
357 * here. Since these commands are accessible
358 * by different means and only really a safety
359 * net, the missing functionality on hppa
360 * shouldn't matter. */
362 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
363 SIGRTMIN+27, /* systemd: set log target to console */
364 SIGRTMIN+28, /* systemd: set log target to kmsg */
365 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete)*/
367 /* ... one free signal here SIGRTMIN+30 ... */
370 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
372 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
373 if (m->signal_fd < 0)
376 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
380 /* Process signals a bit earlier than the rest of things, but
381 * later than notify_fd processing, so that the notify
382 * processing can still figure out to which process/service a
383 * message belongs, before we reap the process. */
384 r = sd_event_source_set_priority(m->signal_event_source, -5);
388 if (m->running_as == SYSTEMD_SYSTEM)
389 return enable_special_signals(m);
394 static void manager_clean_environment(Manager *m) {
397 /* Let's remove some environment variables that we
398 * need ourselves to communicate with our clients */
411 static int manager_default_environment(Manager *m) {
414 if (m->running_as == SYSTEMD_SYSTEM) {
415 /* The system manager always starts with a clean
416 * environment for its children. It does not import
417 * the kernel or the parents exported variables.
419 * The initial passed environ is untouched to keep
420 * /proc/self/environ valid; it is used for tagging
421 * the init process inside containers. */
422 m->environment = strv_new("PATH=" DEFAULT_PATH,
425 /* Import locale variables LC_*= from configuration */
426 locale_setup(&m->environment);
428 /* The user manager passes its own environment
429 * along to its children. */
430 m->environment = strv_copy(environ);
436 manager_clean_environment(m);
437 strv_sort(m->environment);
442 int manager_new(SystemdRunningAs running_as, bool test_run, Manager **_m) {
447 assert(running_as >= 0);
448 assert(running_as < _SYSTEMD_RUNNING_AS_MAX);
450 m = new0(Manager, 1);
455 if (running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0)
456 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
459 m->running_as = running_as;
460 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
461 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
463 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
465 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;
466 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
468 m->test_run = test_run;
470 r = manager_default_environment(m);
474 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
478 r = hashmap_ensure_allocated(&m->jobs, NULL);
482 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
486 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
490 r = set_ensure_allocated(&m->startup_units, NULL);
494 r = set_ensure_allocated(&m->failed_units, NULL);
498 r = sd_event_default(&m->event);
502 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
506 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
510 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
514 r = manager_setup_signals(m);
518 r = manager_setup_cgroup(m);
522 r = manager_setup_time_change(m);
526 m->udev = udev_new();
532 /* Note that we set up neither kdbus, nor the notify fd
533 * here. We do that after deserialization, since they might
534 * have gotten serialized across the reexec. */
536 m->taint_usr = dir_is_empty("/usr") > 0;
546 static int manager_setup_notify(Manager *m) {
552 if (m->notify_fd < 0) {
553 _cleanup_close_ int fd = -1;
554 union sockaddr_union sa = {
555 .sa.sa_family = AF_UNIX,
557 static const int one = 1;
559 /* First free all secondary fields */
560 free(m->notify_socket);
561 m->notify_socket = NULL;
562 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
564 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
566 log_error("Failed to allocate notification socket: %m");
570 if (m->running_as == SYSTEMD_SYSTEM)
571 m->notify_socket = strdup("/run/systemd/notify");
575 e = getenv("XDG_RUNTIME_DIR");
577 log_error("XDG_RUNTIME_DIR is not set: %m");
581 m->notify_socket = strappend(e, "/systemd/notify");
583 if (!m->notify_socket)
586 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
587 r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
589 log_error("bind(%s) failed: %m", sa.un.sun_path);
590 if (errno == EADDRINUSE) {
591 log_notice("Removing %s socket and trying again.", m->notify_socket);
592 r = unlink(m->notify_socket);
594 log_error("Failed to remove %s: %m", m->notify_socket);
598 r = bind(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(sa.un.sun_path));
600 log_error("bind(%s) failed: %m", sa.un.sun_path);
607 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
609 log_error("SO_PASSCRED failed: %m");
616 log_debug("Using notification socket %s", m->notify_socket);
619 if (!m->notify_event_source) {
620 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
622 log_error("Failed to allocate notify event source: %s", strerror(-r));
626 /* Process signals a bit earlier than SIGCHLD, so that we can
627 * still identify to which service an exit message belongs */
628 r = sd_event_source_set_priority(m->notify_event_source, -7);
630 log_error("Failed to set priority of notify event source: %s", strerror(-r));
638 static int manager_setup_kdbus(Manager *m) {
640 _cleanup_free_ char *p = NULL;
644 if (m->test_run || m->kdbus_fd >= 0)
647 m->kdbus_fd = bus_kernel_create_bus(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", m->running_as == SYSTEMD_SYSTEM, &p);
648 if (m->kdbus_fd < 0) {
649 log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd));
653 log_debug("Successfully set up kdbus on %s", p);
655 /* Create the namespace directory here, so that the contents
656 * of that directory is not visible to non-root users. This is
657 * necessary to ensure that users cannot get access to busses
658 * of virtualized users when no UID namespacing is used. */
659 if (m->running_as == SYSTEMD_SYSTEM)
660 mkdir_p_label("/dev/kdbus/domain", 0700);
666 static int manager_connect_bus(Manager *m, bool reexecuting) {
667 bool try_bus_connect;
677 (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"));
679 /* Try to connect to the busses, if possible. */
680 return bus_init(m, try_bus_connect);
683 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
689 while ((u = m->cleanup_queue)) {
690 assert(u->in_cleanup_queue);
700 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
701 GC_OFFSET_UNSURE, /* No clue */
702 GC_OFFSET_GOOD, /* We still need this unit */
703 GC_OFFSET_BAD, /* We don't need this unit anymore */
707 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
714 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
715 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
716 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
719 if (u->in_cleanup_queue)
722 if (unit_check_gc(u))
725 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
729 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
730 unit_gc_sweep(other, gc_marker);
732 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
735 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
742 /* We were unable to find anything out about this entry, so
743 * let's investigate it later */
744 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
745 unit_add_to_gc_queue(u);
749 /* We definitely know that this one is not useful anymore, so
750 * let's mark it for deletion */
751 u->gc_marker = gc_marker + GC_OFFSET_BAD;
752 unit_add_to_cleanup_queue(u);
756 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
759 static unsigned manager_dispatch_gc_queue(Manager *m) {
766 /* log_debug("Running GC..."); */
768 m->gc_marker += _GC_OFFSET_MAX;
769 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
772 gc_marker = m->gc_marker;
774 while ((u = m->gc_queue)) {
775 assert(u->in_gc_queue);
777 unit_gc_sweep(u, gc_marker);
779 LIST_REMOVE(gc_queue, m->gc_queue, u);
780 u->in_gc_queue = false;
784 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
785 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
786 log_debug_unit(u->id, "Collecting %s", u->id);
787 u->gc_marker = gc_marker + GC_OFFSET_BAD;
788 unit_add_to_cleanup_queue(u);
792 m->n_in_gc_queue = 0;
797 static void manager_clear_jobs_and_units(Manager *m) {
802 while ((u = hashmap_first(m->units)))
805 manager_dispatch_cleanup_queue(m);
807 assert(!m->load_queue);
808 assert(!m->run_queue);
809 assert(!m->dbus_unit_queue);
810 assert(!m->dbus_job_queue);
811 assert(!m->cleanup_queue);
812 assert(!m->gc_queue);
814 assert(hashmap_isempty(m->jobs));
815 assert(hashmap_isempty(m->units));
818 m->n_running_jobs = 0;
821 void manager_free(Manager *m) {
827 manager_clear_jobs_and_units(m);
829 for (c = 0; c < _UNIT_TYPE_MAX; c++)
830 if (unit_vtable[c]->shutdown)
831 unit_vtable[c]->shutdown(m);
833 /* If we reexecute ourselves, we keep the root cgroup
835 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
837 manager_undo_generators(m);
841 hashmap_free(m->units);
842 hashmap_free(m->jobs);
843 hashmap_free(m->watch_pids1);
844 hashmap_free(m->watch_pids2);
845 hashmap_free(m->watch_bus);
847 set_free(m->startup_units);
848 set_free(m->failed_units);
850 sd_event_source_unref(m->signal_event_source);
851 sd_event_source_unref(m->notify_event_source);
852 sd_event_source_unref(m->time_change_event_source);
853 sd_event_source_unref(m->jobs_in_progress_event_source);
854 sd_event_source_unref(m->idle_pipe_event_source);
855 sd_event_source_unref(m->run_queue_event_source);
857 safe_close(m->signal_fd);
858 safe_close(m->notify_fd);
859 safe_close(m->time_change_fd);
860 safe_close(m->kdbus_fd);
862 manager_close_idle_pipe(m);
865 sd_event_unref(m->event);
867 free(m->notify_socket);
869 lookup_paths_free(&m->lookup_paths);
870 strv_free(m->environment);
872 hashmap_free(m->cgroup_unit);
873 set_free_free(m->unit_path_cache);
875 free(m->switch_root);
876 free(m->switch_root_init);
878 for (i = 0; i < _RLIMIT_MAX; i++)
881 assert(hashmap_isempty(m->units_requiring_mounts_for));
882 hashmap_free(m->units_requiring_mounts_for);
887 int manager_enumerate(Manager *m) {
893 /* Let's ask every type to load all units from disk/kernel
894 * that it might know */
895 for (c = 0; c < _UNIT_TYPE_MAX; c++)
896 if (unit_vtable[c]->enumerate) {
897 q = unit_vtable[c]->enumerate(m);
902 manager_dispatch_load_queue(m);
906 static int manager_coldplug(Manager *m) {
914 /* Then, let's set up their initial state. */
915 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
922 q = unit_coldplug(u);
930 static void manager_build_unit_path_cache(Manager *m) {
932 _cleanup_closedir_ DIR *d = NULL;
937 set_free_free(m->unit_path_cache);
939 m->unit_path_cache = set_new(&string_hash_ops);
940 if (!m->unit_path_cache) {
941 log_error("Failed to allocate unit path cache.");
945 /* This simply builds a list of files we know exist, so that
946 * we don't always have to go to disk */
948 STRV_FOREACH(i, m->lookup_paths.unit_path) {
954 log_error("Failed to open directory %s: %m", *i);
958 while ((de = readdir(d))) {
961 if (ignore_file(de->d_name))
964 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
970 r = set_consume(m->unit_path_cache, p);
982 log_error("Failed to build unit path cache: %s", strerror(-r));
984 set_free_free(m->unit_path_cache);
985 m->unit_path_cache = NULL;
989 static int manager_distribute_fds(Manager *m, FDSet *fds) {
996 HASHMAP_FOREACH(u, m->units, i) {
998 if (fdset_size(fds) <= 0)
1001 if (UNIT_VTABLE(u)->distribute_fds) {
1002 r = UNIT_VTABLE(u)->distribute_fds(u, fds);
1011 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1016 dual_timestamp_get(&m->generators_start_timestamp);
1017 manager_run_generators(m);
1018 dual_timestamp_get(&m->generators_finish_timestamp);
1020 r = lookup_paths_init(
1021 &m->lookup_paths, m->running_as, true,
1023 m->generator_unit_path,
1024 m->generator_unit_path_early,
1025 m->generator_unit_path_late);
1029 manager_build_unit_path_cache(m);
1031 /* If we will deserialize make sure that during enumeration
1032 * this is already known, so we increase the counter here
1037 /* First, enumerate what we can from all config files */
1038 dual_timestamp_get(&m->units_load_start_timestamp);
1039 r = manager_enumerate(m);
1040 dual_timestamp_get(&m->units_load_finish_timestamp);
1042 /* Second, deserialize if there is something to deserialize */
1044 r = manager_deserialize(m, serialization, fds);
1046 /* Any fds left? Find some unit which wants them. This is
1047 * useful to allow container managers to pass some file
1048 * descriptors to us pre-initialized. This enables
1049 * socket-based activation of entire containers. */
1050 if (fdset_size(fds) > 0) {
1051 q = manager_distribute_fds(m, fds);
1052 if (q < 0 && r == 0)
1056 /* We might have deserialized the notify fd, but if we didn't
1057 * then let's create the bus now */
1058 q = manager_setup_notify(m);
1059 if (q < 0 && r == 0)
1062 /* We might have deserialized the kdbus control fd, but if we
1063 * didn't, then let's create the bus now. */
1064 manager_setup_kdbus(m);
1065 manager_connect_bus(m, !!serialization);
1066 bus_track_coldplug(m, &m->subscribed, &m->deserialized_subscribed);
1068 /* Third, fire things up! */
1069 q = manager_coldplug(m);
1070 if (q < 0 && r == 0)
1073 if (serialization) {
1074 assert(m->n_reloading > 0);
1077 /* Let's wait for the UnitNew/JobNew messages being
1078 * sent, before we notify that the reload is
1080 m->send_reloading_done = true;
1086 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1091 assert(type < _JOB_TYPE_MAX);
1093 assert(mode < _JOB_MODE_MAX);
1095 if (mode == JOB_ISOLATE && type != JOB_START) {
1096 sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1100 if (mode == JOB_ISOLATE && !unit->allow_isolate) {
1101 sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1105 log_debug_unit(unit->id,
1106 "Trying to enqueue job %s/%s/%s", unit->id,
1107 job_type_to_string(type), job_mode_to_string(mode));
1109 job_type_collapse(&type, unit);
1111 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1115 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, override, false,
1116 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1117 mode == JOB_IGNORE_DEPENDENCIES, e);
1121 if (mode == JOB_ISOLATE) {
1122 r = transaction_add_isolate_jobs(tr, m);
1127 r = transaction_activate(tr, m, mode, e);
1131 log_debug_unit(unit->id,
1132 "Enqueued job %s/%s as %u", unit->id,
1133 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1136 *_ret = tr->anchor_job;
1138 transaction_free(tr);
1142 transaction_abort(tr);
1143 transaction_free(tr);
1147 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, bool override, sd_bus_error *e, Job **_ret) {
1152 assert(type < _JOB_TYPE_MAX);
1154 assert(mode < _JOB_MODE_MAX);
1156 r = manager_load_unit(m, name, NULL, NULL, &unit);
1160 return manager_add_job(m, type, unit, mode, override, e, _ret);
1163 Job *manager_get_job(Manager *m, uint32_t id) {
1166 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1169 Unit *manager_get_unit(Manager *m, const char *name) {
1173 return hashmap_get(m->units, name);
1176 unsigned manager_dispatch_load_queue(Manager *m) {
1182 /* Make sure we are not run recursively */
1183 if (m->dispatching_load_queue)
1186 m->dispatching_load_queue = true;
1188 /* Dispatches the load queue. Takes a unit from the queue and
1189 * tries to load its data until the queue is empty */
1191 while ((u = m->load_queue)) {
1192 assert(u->in_load_queue);
1198 m->dispatching_load_queue = false;
1202 int manager_load_unit_prepare(
1214 assert(name || path);
1216 /* This will prepare the unit for loading, but not actually
1217 * load anything from disk. */
1219 if (path && !is_path(path))
1220 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1223 name = basename(path);
1225 t = unit_name_to_type(name);
1227 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, TEMPLATE_INVALID))
1228 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1230 ret = manager_get_unit(m, name);
1236 ret = unit_new(m, unit_vtable[t]->object_size);
1241 ret->fragment_path = strdup(path);
1242 if (!ret->fragment_path) {
1248 r = unit_add_name(ret, name);
1254 unit_add_to_load_queue(ret);
1255 unit_add_to_dbus_queue(ret);
1256 unit_add_to_gc_queue(ret);
1264 int manager_load_unit(
1275 /* This will load the service information files, but not actually
1276 * start any services or anything. */
1278 r = manager_load_unit_prepare(m, name, path, e, _ret);
1282 manager_dispatch_load_queue(m);
1285 *_ret = unit_follow_merge(*_ret);
1290 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1297 HASHMAP_FOREACH(j, s->jobs, i)
1298 job_dump(j, f, prefix);
1301 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1309 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1311 unit_dump(u, f, prefix);
1314 void manager_clear_jobs(Manager *m) {
1319 while ((j = hashmap_first(m->jobs)))
1320 /* No need to recurse. We're cancelling all jobs. */
1321 job_finish_and_invalidate(j, JOB_CANCELED, false);
1324 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1325 Manager *m = userdata;
1331 while ((j = m->run_queue)) {
1332 assert(j->installed);
1333 assert(j->in_run_queue);
1335 job_run_and_invalidate(j);
1338 if (m->n_running_jobs > 0)
1339 manager_watch_jobs_in_progress(m);
1341 if (m->n_on_console > 0)
1342 manager_watch_idle_pipe(m);
1347 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1354 if (m->dispatching_dbus_queue)
1357 m->dispatching_dbus_queue = true;
1359 while ((u = m->dbus_unit_queue)) {
1360 assert(u->in_dbus_queue);
1362 bus_unit_send_change_signal(u);
1366 while ((j = m->dbus_job_queue)) {
1367 assert(j->in_dbus_queue);
1369 bus_job_send_change_signal(j);
1373 m->dispatching_dbus_queue = false;
1375 if (m->send_reloading_done) {
1376 m->send_reloading_done = false;
1378 bus_manager_send_reloading(m, false);
1381 if (m->queued_message)
1382 bus_send_queued_message(m);
1387 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n) {
1388 _cleanup_strv_free_ char **tags = NULL;
1395 tags = strv_split(buf, "\n\r");
1401 log_debug_unit(u->id, "Got notification message for unit %s", u->id);
1403 if (UNIT_VTABLE(u)->notify_message)
1404 UNIT_VTABLE(u)->notify_message(u, pid, tags);
1407 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1408 Manager *m = userdata;
1412 assert(m->notify_fd == fd);
1414 if (revents != EPOLLIN) {
1415 log_warning("Got unexpected poll event for notify fd.");
1421 struct iovec iovec = {
1423 .iov_len = sizeof(buf)-1,
1428 struct cmsghdr cmsghdr;
1429 uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
1432 struct msghdr msghdr = {
1435 .msg_control = &control,
1436 .msg_controllen = sizeof(control),
1438 struct ucred *ucred;
1441 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT);
1446 if (errno == EAGAIN || errno == EINTR)
1452 if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
1453 control.cmsghdr.cmsg_level != SOL_SOCKET ||
1454 control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
1455 control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
1456 log_warning("Received notify message without credentials. Ignoring.");
1460 ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
1462 assert((size_t) n < sizeof(buf));
1465 /* Notify every unit that might be interested, but try
1466 * to avoid notifying the same one multiple times. */
1467 u1 = manager_get_unit_by_pid(m, ucred->pid);
1469 manager_invoke_notify_message(m, u1, ucred->pid, buf, n);
1473 u2 = hashmap_get(m->watch_pids1, LONG_TO_PTR(ucred->pid));
1474 if (u2 && u2 != u1) {
1475 manager_invoke_notify_message(m, u2, ucred->pid, buf, n);
1479 u3 = hashmap_get(m->watch_pids2, LONG_TO_PTR(ucred->pid));
1480 if (u3 && u3 != u2 && u3 != u1) {
1481 manager_invoke_notify_message(m, u3, ucred->pid, buf, n);
1486 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1492 static void invoke_sigchld_event(Manager *m, Unit *u, siginfo_t *si) {
1497 log_debug_unit(u->id, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1499 unit_unwatch_pid(u, si->si_pid);
1500 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1503 static int manager_dispatch_sigchld(Manager *m) {
1509 /* First we call waitd() for a PID and do not reap the
1510 * zombie. That way we can still access /proc/$PID for
1511 * it while it is a zombie. */
1512 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1514 if (errno == ECHILD)
1526 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1527 _cleanup_free_ char *name = NULL;
1530 get_process_comm(si.si_pid, &name);
1532 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1533 si.si_pid, strna(name),
1534 sigchld_code_to_string(si.si_code),
1536 strna(si.si_code == CLD_EXITED
1537 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1538 : signal_to_string(si.si_status)));
1540 /* And now figure out the unit this belongs
1541 * to, it might be multiple... */
1542 u1 = manager_get_unit_by_pid(m, si.si_pid);
1544 invoke_sigchld_event(m, u1, &si);
1545 u2 = hashmap_get(m->watch_pids1, LONG_TO_PTR(si.si_pid));
1547 invoke_sigchld_event(m, u2, &si);
1548 u3 = hashmap_get(m->watch_pids2, LONG_TO_PTR(si.si_pid));
1549 if (u3 && u3 != u2 && u3 != u1)
1550 invoke_sigchld_event(m, u3, &si);
1553 /* And now, we actually reap the zombie. */
1554 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1565 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1566 _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
1569 log_debug_unit(name, "Activating special unit %s", name);
1571 r = manager_add_job_by_name(m, JOB_START, name, mode, true, &error, NULL);
1573 log_error_unit(name, "Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1578 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1579 Manager *m = userdata;
1581 struct signalfd_siginfo sfsi;
1582 bool sigchld = false;
1585 assert(m->signal_fd == fd);
1587 if (revents != EPOLLIN) {
1588 log_warning("Got unexpected events from signal file descriptor.");
1593 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1594 if (n != sizeof(sfsi)) {
1599 if (errno == EINTR || errno == EAGAIN)
1605 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1606 (sfsi.ssi_signo == SIGTERM && m->running_as == SYSTEMD_USER)
1607 ? LOG_DEBUG : LOG_INFO,
1610 switch (sfsi.ssi_signo) {
1617 if (m->running_as == SYSTEMD_SYSTEM) {
1618 /* This is for compatibility with the
1619 * original sysvinit */
1620 m->exit_code = MANAGER_REEXECUTE;
1627 if (m->running_as == SYSTEMD_SYSTEM) {
1628 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1632 /* Run the exit target if there is one, if not, just exit. */
1633 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1634 m->exit_code = MANAGER_EXIT;
1641 if (m->running_as == SYSTEMD_SYSTEM)
1642 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
1644 /* This is a nop on non-init */
1648 if (m->running_as == SYSTEMD_SYSTEM)
1649 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
1651 /* This is a nop on non-init */
1657 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1659 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
1660 log_info("Trying to reconnect to bus...");
1664 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
1665 log_info("Loading D-Bus service...");
1666 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
1673 _cleanup_free_ char *dump = NULL;
1674 _cleanup_fclose_ FILE *f = NULL;
1677 f = open_memstream(&dump, &size);
1679 log_warning("Failed to allocate memory stream.");
1683 manager_dump_units(m, f, "\t");
1684 manager_dump_jobs(m, f, "\t");
1687 log_warning("Failed to write status stream");
1692 log_warning("Failed to flush status stream");
1696 log_dump(LOG_INFO, dump);
1701 m->exit_code = MANAGER_RELOAD;
1706 /* Starting SIGRTMIN+0 */
1707 static const char * const target_table[] = {
1708 [0] = SPECIAL_DEFAULT_TARGET,
1709 [1] = SPECIAL_RESCUE_TARGET,
1710 [2] = SPECIAL_EMERGENCY_TARGET,
1711 [3] = SPECIAL_HALT_TARGET,
1712 [4] = SPECIAL_POWEROFF_TARGET,
1713 [5] = SPECIAL_REBOOT_TARGET,
1714 [6] = SPECIAL_KEXEC_TARGET
1717 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1718 static const ManagerExitCode code_table[] = {
1720 [1] = MANAGER_POWEROFF,
1721 [2] = MANAGER_REBOOT,
1725 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
1726 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
1727 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
1728 manager_start_target(m, target_table[idx],
1729 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
1733 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
1734 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
1735 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
1739 switch (sfsi.ssi_signo - SIGRTMIN) {
1742 log_debug("Enabling showing of status.");
1743 manager_set_show_status(m, SHOW_STATUS_YES);
1747 log_debug("Disabling showing of status.");
1748 manager_set_show_status(m, SHOW_STATUS_NO);
1752 log_set_max_level(LOG_DEBUG);
1753 log_notice("Setting log level to debug.");
1757 log_set_max_level(LOG_INFO);
1758 log_notice("Setting log level to info.");
1762 if (m->running_as == SYSTEMD_USER) {
1763 m->exit_code = MANAGER_EXIT;
1767 /* This is a nop on init */
1771 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
1772 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
1773 log_notice("Setting log target to journal-or-kmsg.");
1777 log_set_target(LOG_TARGET_CONSOLE);
1778 log_notice("Setting log target to console.");
1782 log_set_target(LOG_TARGET_KMSG);
1783 log_notice("Setting log target to kmsg.");
1787 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
1794 manager_dispatch_sigchld(m);
1799 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1800 Manager *m = userdata;
1805 assert(m->time_change_fd == fd);
1807 log_struct(LOG_INFO,
1808 MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
1809 "MESSAGE=Time has been changed",
1812 /* Restart the watch */
1813 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
1814 m->time_change_fd = safe_close(m->time_change_fd);
1816 manager_setup_time_change(m);
1818 HASHMAP_FOREACH(u, m->units, i)
1819 if (UNIT_VTABLE(u)->time_change)
1820 UNIT_VTABLE(u)->time_change(u);
1825 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1826 Manager *m = userdata;
1829 assert(m->idle_pipe[2] == fd);
1831 m->no_console_output = m->n_on_console > 0;
1833 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
1834 manager_close_idle_pipe(m);
1839 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
1840 Manager *m = userdata;
1847 manager_print_jobs_in_progress(m);
1849 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
1850 r = sd_event_source_set_time(source, next);
1854 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
1857 int manager_loop(Manager *m) {
1860 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
1863 m->exit_code = MANAGER_OK;
1865 /* Release the path cache */
1866 set_free_free(m->unit_path_cache);
1867 m->unit_path_cache = NULL;
1869 manager_check_finished(m);
1871 /* There might still be some zombies hanging around from
1872 * before we were exec()'ed. Let's reap them. */
1873 r = manager_dispatch_sigchld(m);
1877 while (m->exit_code == MANAGER_OK) {
1880 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM)
1883 if (!ratelimit_test(&rl)) {
1884 /* Yay, something is going seriously wrong, pause a little */
1885 log_warning("Looping too fast. Throttling execution a little.");
1890 if (manager_dispatch_load_queue(m) > 0)
1893 if (manager_dispatch_gc_queue(m) > 0)
1896 if (manager_dispatch_cleanup_queue(m) > 0)
1899 if (manager_dispatch_cgroup_queue(m) > 0)
1902 if (manager_dispatch_dbus_queue(m) > 0)
1905 /* Sleep for half the watchdog time */
1906 if (m->runtime_watchdog > 0 && m->running_as == SYSTEMD_SYSTEM) {
1907 wait_usec = m->runtime_watchdog / 2;
1911 wait_usec = USEC_INFINITY;
1913 r = sd_event_run(m->event, wait_usec);
1915 log_error("Failed to run event loop: %s", strerror(-r));
1920 return m->exit_code;
1923 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
1924 _cleanup_free_ char *n = NULL;
1932 r = unit_name_from_dbus_path(s, &n);
1936 r = manager_load_unit(m, n, NULL, e, &u);
1945 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
1955 p = startswith(s, "/org/freedesktop/systemd1/job/");
1959 r = safe_atou(p, &id);
1963 j = manager_get_job(m, id);
1972 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
1975 _cleanup_free_ char *p = NULL;
1978 audit_fd = get_audit_fd();
1982 /* Don't generate audit events if the service was already
1983 * started and we're just deserializing */
1984 if (m->n_reloading > 0)
1987 if (m->running_as != SYSTEMD_SYSTEM)
1990 if (u->type != UNIT_SERVICE)
1993 p = unit_name_to_prefix_and_instance(u->id);
1995 log_error_unit(u->id,
1996 "Failed to allocate unit name for audit message: %s", strerror(ENOMEM));
2000 if (audit_log_user_comm_message(audit_fd, type, "", p, NULL, NULL, NULL, success) < 0) {
2001 if (errno == EPERM) {
2002 /* We aren't allowed to send audit messages?
2003 * Then let's not retry again. */
2006 log_warning("Failed to send audit message: %m");
2012 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2013 union sockaddr_union sa = PLYMOUTH_SOCKET;
2016 _cleanup_free_ char *message = NULL;
2017 _cleanup_close_ int fd = -1;
2019 /* Don't generate plymouth events if the service was already
2020 * started and we're just deserializing */
2021 if (m->n_reloading > 0)
2024 if (m->running_as != SYSTEMD_SYSTEM)
2027 if (detect_container(NULL) > 0)
2030 if (u->type != UNIT_SERVICE &&
2031 u->type != UNIT_MOUNT &&
2032 u->type != UNIT_SWAP)
2035 /* We set SOCK_NONBLOCK here so that we rather drop the
2036 * message then wait for plymouth */
2037 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2039 log_error("socket() failed: %m");
2043 if (connect(fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + 1 + strlen(sa.un.sun_path+1)) < 0) {
2045 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2046 log_error("connect() failed: %m");
2050 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2056 if (write(fd, message, n + 1) != n + 1)
2057 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2058 log_error("Failed to write Plymouth message: %m");
2061 void manager_dispatch_bus_name_owner_changed(
2064 const char* old_owner,
2065 const char *new_owner) {
2072 u = hashmap_get(m->watch_bus, name);
2076 UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
2079 int manager_open_serialization(Manager *m, FILE **_f) {
2086 path = m->running_as == SYSTEMD_SYSTEM ? "/run/systemd" : "/tmp";
2087 fd = open_tmpfile(path, O_RDWR|O_CLOEXEC);
2091 log_debug("Serializing state to %s", path);
2093 f = fdopen(fd, "w+");
2104 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2117 fprintf(f, "current-job-id=%i\n", m->current_job_id);
2118 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2119 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2120 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2122 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2123 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2124 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2125 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2128 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2129 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2130 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2131 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2132 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2133 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2134 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2135 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2138 if (!switching_root) {
2139 STRV_FOREACH(e, m->environment) {
2140 _cleanup_free_ char *ce;
2146 fprintf(f, "env=%s\n", *e);
2150 if (m->notify_fd >= 0) {
2153 copy = fdset_put_dup(fds, m->notify_fd);
2157 fprintf(f, "notify-fd=%i\n", copy);
2158 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2161 if (m->kdbus_fd >= 0) {
2164 copy = fdset_put_dup(fds, m->kdbus_fd);
2168 fprintf(f, "kdbus-fd=%i\n", copy);
2171 bus_track_serialize(m->subscribed, f);
2175 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2183 r = unit_serialize(u, f, fds, !switching_root);
2190 assert(m->n_reloading > 0);
2196 r = bus_fdset_add_all(m, fds);
2203 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2209 log_debug("Deserializing state...");
2214 char line[LINE_MAX], *l;
2216 if (!fgets(line, sizeof(line), f)) {
2231 if (startswith(l, "current-job-id=")) {
2234 if (safe_atou32(l+15, &id) < 0)
2235 log_warning("Failed to parse current job id value %s", l+15);
2237 m->current_job_id = MAX(m->current_job_id, id);
2239 } else if (startswith(l, "n-installed-jobs=")) {
2242 if (safe_atou32(l+17, &n) < 0)
2243 log_warning("Failed to parse installed jobs counter %s", l+17);
2245 m->n_installed_jobs += n;
2247 } else if (startswith(l, "n-failed-jobs=")) {
2250 if (safe_atou32(l+14, &n) < 0)
2251 log_warning("Failed to parse failed jobs counter %s", l+14);
2253 m->n_failed_jobs += n;
2255 } else if (startswith(l, "taint-usr=")) {
2258 b = parse_boolean(l+10);
2260 log_warning("Failed to parse taint /usr flag %s", l+10);
2262 m->taint_usr = m->taint_usr || b;
2264 } else if (startswith(l, "firmware-timestamp="))
2265 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2266 else if (startswith(l, "loader-timestamp="))
2267 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2268 else if (startswith(l, "kernel-timestamp="))
2269 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2270 else if (startswith(l, "initrd-timestamp="))
2271 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2272 else if (startswith(l, "userspace-timestamp="))
2273 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2274 else if (startswith(l, "finish-timestamp="))
2275 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2276 else if (startswith(l, "security-start-timestamp="))
2277 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2278 else if (startswith(l, "security-finish-timestamp="))
2279 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2280 else if (startswith(l, "generators-start-timestamp="))
2281 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2282 else if (startswith(l, "generators-finish-timestamp="))
2283 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2284 else if (startswith(l, "units-load-start-timestamp="))
2285 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2286 else if (startswith(l, "units-load-finish-timestamp="))
2287 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2288 else if (startswith(l, "env=")) {
2289 _cleanup_free_ char *uce = NULL;
2292 uce = cunescape(l+4);
2298 e = strv_env_set(m->environment, uce);
2304 strv_free(m->environment);
2307 } else if (startswith(l, "notify-fd=")) {
2310 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2311 log_warning("Failed to parse notify fd: %s", l + 10);
2313 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2314 safe_close(m->notify_fd);
2315 m->notify_fd = fdset_remove(fds, fd);
2318 } else if (startswith(l, "notify-socket=")) {
2327 free(m->notify_socket);
2328 m->notify_socket = n;
2330 } else if (startswith(l, "kdbus-fd=")) {
2333 if (safe_atoi(l + 9, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2334 log_warning("Failed to parse kdbus fd: %s", l + 9);
2336 safe_close(m->kdbus_fd);
2337 m->kdbus_fd = fdset_remove(fds, fd);
2340 } else if (bus_track_deserialize_item(&m->deserialized_subscribed, l) == 0)
2341 log_warning("Unknown serialization item '%s'", l);
2346 char name[UNIT_NAME_MAX+2];
2349 if (!fgets(name, sizeof(name), f)) {
2360 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2364 r = unit_deserialize(u, f, fds);
2373 assert(m->n_reloading > 0);
2379 int manager_reload(Manager *m) {
2381 _cleanup_fclose_ FILE *f = NULL;
2382 _cleanup_fdset_free_ FDSet *fds = NULL;
2386 r = manager_open_serialization(m, &f);
2391 bus_manager_send_reloading(m, true);
2399 r = manager_serialize(m, f, fds, false);
2405 if (fseeko(f, 0, SEEK_SET) < 0) {
2410 /* From here on there is no way back. */
2411 manager_clear_jobs_and_units(m);
2412 manager_undo_generators(m);
2413 lookup_paths_free(&m->lookup_paths);
2415 /* Find new unit paths */
2416 manager_run_generators(m);
2418 q = lookup_paths_init(
2419 &m->lookup_paths, m->running_as, true,
2421 m->generator_unit_path,
2422 m->generator_unit_path_early,
2423 m->generator_unit_path_late);
2427 manager_build_unit_path_cache(m);
2429 /* First, enumerate what we can from all config files */
2430 q = manager_enumerate(m);
2434 /* Second, deserialize our stored data */
2435 q = manager_deserialize(m, f, fds);
2442 /* Re-register notify_fd as event source */
2443 q = manager_setup_notify(m);
2447 /* Third, fire things up! */
2448 q = manager_coldplug(m);
2452 assert(m->n_reloading > 0);
2455 m->send_reloading_done = true;
2460 bool manager_is_reloading_or_reexecuting(Manager *m) {
2463 return m->n_reloading != 0;
2466 void manager_reset_failed(Manager *m) {
2472 HASHMAP_FOREACH(u, m->units, i)
2473 unit_reset_failed(u);
2476 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2482 /* Returns true if the unit is inactive or going down */
2483 u = manager_get_unit(m, name);
2487 return unit_inactive_or_pending(u);
2490 void manager_check_finished(Manager *m) {
2491 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2492 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2498 if (m->n_running_jobs == 0)
2499 m->jobs_in_progress_event_source = sd_event_source_unref(m->jobs_in_progress_event_source);
2501 if (hashmap_size(m->jobs) > 0) {
2503 if (m->jobs_in_progress_event_source)
2504 sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2509 manager_flip_auto_status(m, false);
2511 /* Notify Type=idle units that we are done now */
2512 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
2513 manager_close_idle_pipe(m);
2515 /* Turn off confirm spawn now */
2516 m->confirm_spawn = false;
2518 /* This is no longer the first boot */
2519 manager_set_first_boot(m, false);
2521 if (dual_timestamp_is_set(&m->finish_timestamp))
2524 dual_timestamp_get(&m->finish_timestamp);
2526 if (m->running_as == SYSTEMD_SYSTEM && detect_container(NULL) <= 0) {
2528 /* Note that m->kernel_usec.monotonic is always at 0,
2529 * and m->firmware_usec.monotonic and
2530 * m->loader_usec.monotonic should be considered
2531 * negative values. */
2533 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2534 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2535 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2536 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2538 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2540 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2541 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2543 log_struct(LOG_INFO,
2544 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2545 "KERNEL_USEC="USEC_FMT, kernel_usec,
2546 "INITRD_USEC="USEC_FMT, initrd_usec,
2547 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2548 "MESSAGE=Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2549 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2550 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2551 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2552 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2555 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2558 log_struct(LOG_INFO,
2559 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2560 "KERNEL_USEC="USEC_FMT, kernel_usec,
2561 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2562 "MESSAGE=Startup finished in %s (kernel) + %s (userspace) = %s.",
2563 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2564 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2565 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2569 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2570 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2572 log_struct(LOG_INFO,
2573 MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2574 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2575 "MESSAGE=Startup finished in %s.",
2576 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC),
2580 SET_FOREACH(u, m->startup_units, i)
2582 cgroup_context_apply(unit_get_cgroup_context(u), unit_get_cgroup_mask(u), u->cgroup_path, manager_state(m));
2584 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2588 "STATUS=Startup finished in %s.",
2589 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2592 static int create_generator_dir(Manager *m, char **generator, const char *name) {
2603 if (m->running_as == SYSTEMD_SYSTEM && getpid() == 1) {
2604 /* systemd --system, not running --test */
2606 p = strappend("/run/systemd/", name);
2610 r = mkdir_p_label(p, 0755);
2612 log_error("Failed to create generator directory %s: %s",
2617 } else if (m->running_as == SYSTEMD_USER) {
2618 const char *s = NULL;
2620 s = getenv("XDG_RUNTIME_DIR");
2623 p = strjoin(s, "/systemd/", name, NULL);
2627 r = mkdir_p_label(p, 0755);
2629 log_error("Failed to create generator directory %s: %s",
2635 /* systemd --system --test */
2637 p = strjoin("/tmp/systemd-", name, ".XXXXXX", NULL);
2642 log_error("Failed to create generator directory %s: %m",
2653 static void trim_generator_dir(Manager *m, char **generator) {
2660 if (rmdir(*generator) >= 0) {
2668 void manager_run_generators(Manager *m) {
2669 _cleanup_closedir_ DIR *d = NULL;
2670 const char *generator_path;
2671 const char *argv[5];
2679 generator_path = m->running_as == SYSTEMD_SYSTEM ? SYSTEM_GENERATOR_PATH : USER_GENERATOR_PATH;
2680 d = opendir(generator_path);
2682 if (errno == ENOENT)
2685 log_error("Failed to enumerate generator directory %s: %m",
2690 r = create_generator_dir(m, &m->generator_unit_path, "generator");
2694 r = create_generator_dir(m, &m->generator_unit_path_early, "generator.early");
2698 r = create_generator_dir(m, &m->generator_unit_path_late, "generator.late");
2702 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
2703 argv[1] = m->generator_unit_path;
2704 argv[2] = m->generator_unit_path_early;
2705 argv[3] = m->generator_unit_path_late;
2708 RUN_WITH_UMASK(0022)
2709 execute_directory(generator_path, d, DEFAULT_TIMEOUT_USEC, (char**) argv);
2712 trim_generator_dir(m, &m->generator_unit_path);
2713 trim_generator_dir(m, &m->generator_unit_path_early);
2714 trim_generator_dir(m, &m->generator_unit_path_late);
2717 static void remove_generator_dir(Manager *m, char **generator) {
2724 strv_remove(m->lookup_paths.unit_path, *generator);
2725 rm_rf(*generator, false, true, false);
2731 void manager_undo_generators(Manager *m) {
2734 remove_generator_dir(m, &m->generator_unit_path);
2735 remove_generator_dir(m, &m->generator_unit_path_early);
2736 remove_generator_dir(m, &m->generator_unit_path_late);
2739 int manager_environment_add(Manager *m, char **minus, char **plus) {
2740 char **a = NULL, **b = NULL, **l;
2745 if (!strv_isempty(minus)) {
2746 a = strv_env_delete(l, 1, minus);
2753 if (!strv_isempty(plus)) {
2754 b = strv_env_merge(2, l, plus);
2763 if (m->environment != l)
2764 strv_free(m->environment);
2771 manager_clean_environment(m);
2772 strv_sort(m->environment);
2777 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
2782 for (i = 0; i < _RLIMIT_MAX; i++) {
2783 if (!default_rlimit[i])
2786 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
2794 void manager_recheck_journal(Manager *m) {
2799 if (m->running_as != SYSTEMD_SYSTEM)
2802 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
2803 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
2804 log_close_journal();
2808 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
2809 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
2810 log_close_journal();
2814 /* Hmm, OK, so the socket is fully up and the service is up
2815 * too, then let's make use of the thing. */
2819 void manager_set_show_status(Manager *m, ShowStatus mode) {
2821 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
2823 if (m->running_as != SYSTEMD_SYSTEM)
2826 m->show_status = mode;
2829 touch("/run/systemd/show-status");
2831 unlink("/run/systemd/show-status");
2834 static bool manager_get_show_status(Manager *m) {
2837 if (m->running_as != SYSTEMD_SYSTEM)
2840 if (m->no_console_output)
2843 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
2846 if (m->show_status > 0)
2849 /* If Plymouth is running make sure we show the status, so
2850 * that there's something nice to see when people press Esc */
2852 return plymouth_running();
2855 void manager_set_first_boot(Manager *m, bool b) {
2858 if (m->running_as != SYSTEMD_SYSTEM)
2864 touch("/run/systemd/first-boot");
2866 unlink("/run/systemd/first-boot");
2869 void manager_status_printf(Manager *m, bool ephemeral, const char *status, const char *format, ...) {
2872 if (!manager_get_show_status(m))
2875 /* XXX We should totally drop the check for ephemeral here
2876 * and thus effectively make 'Type=idle' pointless. */
2877 if (ephemeral && m->n_on_console > 0)
2880 va_start(ap, format);
2881 status_vprintf(status, true, ephemeral, format, ap);
2885 int manager_get_unit_by_path(Manager *m, const char *path, const char *suffix, Unit **_found) {
2886 _cleanup_free_ char *p = NULL;
2894 p = unit_name_from_path(path, suffix);
2898 found = manager_get_unit(m, p);
2908 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
2909 char p[strlen(path)+1];
2915 path_kill_slashes(p);
2917 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
2920 const char *manager_get_runtime_prefix(Manager *m) {
2923 return m->running_as == SYSTEMD_SYSTEM ?
2925 getenv("XDG_RUNTIME_DIR");
2928 ManagerState manager_state(Manager *m) {
2933 /* Did we ever finish booting? If not then we are still starting up */
2934 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
2936 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
2937 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
2938 return MANAGER_INITIALIZING;
2940 return MANAGER_STARTING;
2943 /* Is the special shutdown target queued? If so, we are in shutdown state */
2944 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
2945 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))
2946 return MANAGER_STOPPING;
2948 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
2949 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
2950 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2951 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2952 return MANAGER_MAINTENANCE;
2954 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
2955 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
2956 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_TRY_RESTART, JOB_RELOAD_OR_START))))
2957 return MANAGER_MAINTENANCE;
2959 /* Are there any failed units? If so, we are in degraded mode */
2960 if (set_size(m->failed_units) > 0)
2961 return MANAGER_DEGRADED;
2963 return MANAGER_RUNNING;
2966 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
2967 [MANAGER_INITIALIZING] = "initializing",
2968 [MANAGER_STARTING] = "starting",
2969 [MANAGER_RUNNING] = "running",
2970 [MANAGER_DEGRADED] = "degraded",
2971 [MANAGER_MAINTENANCE] = "maintenance",
2972 [MANAGER_STOPPING] = "stopping",
2975 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);