2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/epoll.h>
27 #include <sys/inotify.h>
28 #include <sys/ioctl.h>
29 #include <sys/reboot.h>
30 #include <sys/timerfd.h>
38 #include "sd-daemon.h"
39 #include "sd-messages.h"
41 #include "alloc-util.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
49 #include "dbus-manager.h"
50 #include "dbus-unit.h"
52 #include "dirent-util.h"
55 #include "exit-status.h"
61 #include "locale-setup.h"
67 #include "parse-util.h"
68 #include "path-lookup.h"
69 #include "path-util.h"
70 #include "process-util.h"
71 #include "ratelimit.h"
73 #include "signal-util.h"
75 #include "stat-util.h"
76 #include "string-table.h"
77 #include "string-util.h"
79 #include "terminal-util.h"
80 #include "time-util.h"
81 #include "transaction.h"
82 #include "umask-util.h"
83 #include "unit-name.h"
88 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
89 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
91 /* Initial delay and the interval for printing status messages about running jobs */
92 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
93 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
94 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
96 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
97 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
98 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
102 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
103 static int manager_run_generators(Manager
*m
);
105 static void manager_watch_jobs_in_progress(Manager
*m
) {
111 if (m
->jobs_in_progress_event_source
)
114 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
115 r
= sd_event_add_time(
117 &m
->jobs_in_progress_event_source
,
120 manager_dispatch_jobs_in_progress
, m
);
124 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
127 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
129 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
132 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
133 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
137 p
= mempset(p
, ' ', pos
-2);
138 if (log_get_show_color())
139 p
= stpcpy(p
, ANSI_RED
);
143 if (pos
> 0 && pos
<= width
) {
144 if (log_get_show_color())
145 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
149 if (log_get_show_color())
150 p
= stpcpy(p
, ANSI_NORMAL
);
153 if (log_get_show_color())
154 p
= stpcpy(p
, ANSI_RED
);
157 p
= mempset(p
, ' ', width
-1-pos
);
158 if (log_get_show_color())
159 strcpy(p
, ANSI_NORMAL
);
163 void manager_flip_auto_status(Manager
*m
, bool enable
) {
167 if (m
->show_status
== SHOW_STATUS_AUTO
)
168 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
170 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
171 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
175 static void manager_print_jobs_in_progress(Manager
*m
) {
176 _cleanup_free_
char *job_of_n
= NULL
;
179 unsigned counter
= 0, print_nr
;
180 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
182 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
186 assert(m
->n_running_jobs
> 0);
188 manager_flip_auto_status(m
, true);
190 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
192 HASHMAP_FOREACH(j
, m
->jobs
, i
)
193 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
196 /* m->n_running_jobs must be consistent with the contents of m->jobs,
197 * so the above loop must have succeeded in finding j. */
198 assert(counter
== print_nr
+ 1);
201 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
203 cylon_pos
= 14 - cylon_pos
;
204 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
206 m
->jobs_in_progress_iteration
++;
208 if (m
->n_running_jobs
> 1) {
209 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
213 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
214 if (job_get_timeout(j
, &x
) > 0)
215 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
217 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
218 "%sA %s job is running for %s (%s / %s)",
220 job_type_to_string(j
->type
),
221 unit_description(j
->unit
),
225 static int have_ask_password(void) {
226 _cleanup_closedir_
DIR *dir
;
228 dir
= opendir("/run/systemd/ask-password");
241 if (!de
&& errno
> 0)
246 if (startswith(de
->d_name
, "ask."))
251 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
252 int fd
, uint32_t revents
, void *userdata
) {
253 Manager
*m
= userdata
;
259 m
->have_ask_password
= have_ask_password();
260 if (m
->have_ask_password
< 0)
261 /* Log error but continue. Negative have_ask_password
262 * is treated as unknown status. */
263 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
268 static void manager_close_ask_password(Manager
*m
) {
271 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
272 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
273 m
->have_ask_password
= -EINVAL
;
276 static int manager_check_ask_password(Manager
*m
) {
281 if (!m
->ask_password_event_source
) {
282 assert(m
->ask_password_inotify_fd
< 0);
284 mkdir_p_label("/run/systemd/ask-password", 0755);
286 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
287 if (m
->ask_password_inotify_fd
< 0)
288 return log_error_errno(errno
, "inotify_init1() failed: %m");
290 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
291 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
292 manager_close_ask_password(m
);
296 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
297 m
->ask_password_inotify_fd
, EPOLLIN
,
298 manager_dispatch_ask_password_fd
, m
);
300 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
301 manager_close_ask_password(m
);
305 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
307 /* Queries might have been added meanwhile... */
308 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
309 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
312 return m
->have_ask_password
;
315 static int manager_watch_idle_pipe(Manager
*m
) {
320 if (m
->idle_pipe_event_source
)
323 if (m
->idle_pipe
[2] < 0)
326 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
328 return log_error_errno(r
, "Failed to watch idle pipe: %m");
330 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
335 static void manager_close_idle_pipe(Manager
*m
) {
338 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
340 safe_close_pair(m
->idle_pipe
);
341 safe_close_pair(m
->idle_pipe
+ 2);
344 static int manager_setup_time_change(Manager
*m
) {
347 /* We only care for the cancellation event, hence we set the
348 * timeout to the latest possible value. */
349 struct itimerspec its
= {
350 .it_value
.tv_sec
= TIME_T_MAX
,
354 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
359 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
360 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
362 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
363 if (m
->time_change_fd
< 0)
364 return log_error_errno(errno
, "Failed to create timerfd: %m");
366 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
367 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
368 m
->time_change_fd
= safe_close(m
->time_change_fd
);
372 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
374 return log_error_errno(r
, "Failed to create time change event source: %m");
376 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
378 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
383 static int enable_special_signals(Manager
*m
) {
384 _cleanup_close_
int fd
= -1;
391 /* Enable that we get SIGINT on control-alt-del. In containers
392 * this will fail with EPERM (older) or EINVAL (newer), so
394 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
395 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
397 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
399 /* Support systems without virtual console */
401 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
403 /* Enable that we get SIGWINCH on kbrequest */
404 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
405 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
411 static int manager_setup_signals(Manager
*m
) {
412 struct sigaction sa
= {
413 .sa_handler
= SIG_DFL
,
414 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
421 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
423 /* We make liberal use of realtime signals here. On
424 * Linux/glibc we have 30 of them (with the exception of Linux
425 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
428 assert_se(sigemptyset(&mask
) == 0);
429 sigset_add_many(&mask
,
430 SIGCHLD
, /* Child died */
431 SIGTERM
, /* Reexecute daemon */
432 SIGHUP
, /* Reload configuration */
433 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
434 SIGUSR2
, /* systemd: dump status */
435 SIGINT
, /* Kernel sends us this on control-alt-del */
436 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
437 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
439 SIGRTMIN
+0, /* systemd: start default.target */
440 SIGRTMIN
+1, /* systemd: isolate rescue.target */
441 SIGRTMIN
+2, /* systemd: isolate emergency.target */
442 SIGRTMIN
+3, /* systemd: start halt.target */
443 SIGRTMIN
+4, /* systemd: start poweroff.target */
444 SIGRTMIN
+5, /* systemd: start reboot.target */
445 SIGRTMIN
+6, /* systemd: start kexec.target */
447 /* ... space for more special targets ... */
449 SIGRTMIN
+13, /* systemd: Immediate halt */
450 SIGRTMIN
+14, /* systemd: Immediate poweroff */
451 SIGRTMIN
+15, /* systemd: Immediate reboot */
452 SIGRTMIN
+16, /* systemd: Immediate kexec */
454 /* ... space for more immediate system state changes ... */
456 SIGRTMIN
+20, /* systemd: enable status messages */
457 SIGRTMIN
+21, /* systemd: disable status messages */
458 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
459 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
460 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
462 /* .. one free signal here ... */
464 #if !defined(__hppa64__) && !defined(__hppa__)
465 /* Apparently Linux on hppa has fewer RT
466 * signals (SIGRTMAX is SIGRTMIN+25 there),
467 * hence let's not try to make use of them
468 * here. Since these commands are accessible
469 * by different means and only really a safety
470 * net, the missing functionality on hppa
471 * shouldn't matter. */
473 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
474 SIGRTMIN
+27, /* systemd: set log target to console */
475 SIGRTMIN
+28, /* systemd: set log target to kmsg */
476 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
478 /* ... one free signal here SIGRTMIN+30 ... */
481 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
483 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
484 if (m
->signal_fd
< 0)
487 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
491 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
493 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
494 * notify processing can still figure out to which process/service a message belongs, before we reap the
495 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
496 * status information before detecting that there's no process in a cgroup. */
497 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
501 if (MANAGER_IS_SYSTEM(m
))
502 return enable_special_signals(m
);
507 static void manager_clean_environment(Manager
*m
) {
510 /* Let's remove some environment variables that we
511 * need ourselves to communicate with our clients */
525 static int manager_default_environment(Manager
*m
) {
528 if (MANAGER_IS_SYSTEM(m
)) {
529 /* The system manager always starts with a clean
530 * environment for its children. It does not import
531 * the kernel or the parents exported variables.
533 * The initial passed environ is untouched to keep
534 * /proc/self/environ valid; it is used for tagging
535 * the init process inside containers. */
536 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
539 /* Import locale variables LC_*= from configuration */
540 locale_setup(&m
->environment
);
542 /* The user manager passes its own environment
543 * along to its children. */
544 m
->environment
= strv_copy(environ
);
550 manager_clean_environment(m
);
551 strv_sort(m
->environment
);
556 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
561 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
563 m
= new0(Manager
, 1);
567 m
->unit_file_scope
= scope
;
568 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
569 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
570 m
->default_tasks_accounting
= true;
571 m
->default_tasks_max
= UINT64_MAX
;
574 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
575 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
578 /* Prepare log fields we can use for structured logging */
579 if (MANAGER_IS_SYSTEM(m
)) {
580 m
->unit_log_field
= "UNIT=";
581 m
->unit_log_format_string
= "UNIT=%s";
583 m
->unit_log_field
= "USER_UNIT=";
584 m
->unit_log_format_string
= "USER_UNIT=%s";
587 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
589 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
590 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->kdbus_fd
= m
->cgroup_inotify_fd
=
591 m
->ask_password_inotify_fd
= -1;
593 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
595 m
->have_ask_password
= -EINVAL
; /* we don't know */
598 m
->test_run
= test_run
;
600 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
601 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
603 r
= manager_default_environment(m
);
607 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
611 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
615 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
619 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
623 r
= sd_event_default(&m
->event
);
627 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
631 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
635 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
639 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
641 r
= manager_setup_signals(m
);
645 r
= manager_setup_cgroup(m
);
649 r
= manager_setup_time_change(m
);
653 m
->udev
= udev_new();
659 /* Note that we set up neither kdbus, nor the notify fd
660 * here. We do that after deserialization, since they might
661 * have gotten serialized across the reexec. */
663 m
->taint_usr
= dir_is_empty("/usr") > 0;
673 static int manager_setup_notify(Manager
*m
) {
679 if (m
->notify_fd
< 0) {
680 _cleanup_close_
int fd
= -1;
681 union sockaddr_union sa
= {
682 .sa
.sa_family
= AF_UNIX
,
684 static const int one
= 1;
687 /* First free all secondary fields */
688 m
->notify_socket
= mfree(m
->notify_socket
);
689 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
691 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
693 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
695 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
697 e
= manager_get_runtime_prefix(m
);
699 log_error("Failed to determine runtime prefix.");
703 m
->notify_socket
= strappend(e
, "/systemd/notify");
704 if (!m
->notify_socket
)
707 (void) mkdir_parents_label(m
->notify_socket
, 0755);
708 (void) unlink(m
->notify_socket
);
710 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
711 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
713 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
715 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
717 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
722 log_debug("Using notification socket %s", m
->notify_socket
);
725 if (!m
->notify_event_source
) {
726 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
728 return log_error_errno(r
, "Failed to allocate notify event source: %m");
730 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
731 * service an exit message belongs. */
732 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
734 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
736 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
742 static int manager_setup_cgroups_agent(Manager
*m
) {
744 static const union sockaddr_union sa
= {
745 .un
.sun_family
= AF_UNIX
,
746 .un
.sun_path
= "/run/systemd/cgroups-agent",
750 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
751 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
752 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
753 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
754 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
755 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
756 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
757 * we thus won't lose messages.
759 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
760 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
761 * bus for these messages. */
766 if (!MANAGER_IS_SYSTEM(m
))
769 if (cg_all_unified() > 0) /* We don't need this anymore on the unified hierarchy */
772 if (m
->cgroups_agent_fd
< 0) {
773 _cleanup_close_
int fd
= -1;
775 /* First free all secondary fields */
776 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
778 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
780 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
782 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
784 (void) unlink(sa
.un
.sun_path
);
786 /* Only allow root to connect to this socket */
788 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
790 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
792 m
->cgroups_agent_fd
= fd
;
796 if (!m
->cgroups_agent_event_source
) {
797 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
799 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
801 /* Process cgroups notifications early, but after having processed service notification messages or
802 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
803 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
804 * cgroup inotify for the unified cgroup stuff. */
805 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
807 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
809 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
815 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
816 bool try_bus_connect
;
826 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
828 /* Try to connect to the buses, if possible. */
829 return bus_init(m
, try_bus_connect
);
832 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
838 while ((u
= m
->cleanup_queue
)) {
839 assert(u
->in_cleanup_queue
);
849 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
850 GC_OFFSET_UNSURE
, /* No clue */
851 GC_OFFSET_GOOD
, /* We still need this unit */
852 GC_OFFSET_BAD
, /* We don't need this unit anymore */
856 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
)
861 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
863 /* Recursively mark referenced units as GOOD as well */
864 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
865 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
866 unit_gc_mark_good(other
, gc_marker
);
869 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
876 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
877 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
878 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
879 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
882 if (u
->in_cleanup_queue
)
885 if (unit_check_gc(u
))
888 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
892 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
893 unit_gc_sweep(other
, gc_marker
);
895 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
898 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
905 /* We were unable to find anything out about this entry, so
906 * let's investigate it later */
907 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
908 unit_add_to_gc_queue(u
);
912 /* We definitely know that this one is not useful anymore, so
913 * let's mark it for deletion */
914 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
915 unit_add_to_cleanup_queue(u
);
919 unit_gc_mark_good(u
, gc_marker
);
922 static unsigned manager_dispatch_gc_queue(Manager
*m
) {
929 /* log_debug("Running GC..."); */
931 m
->gc_marker
+= _GC_OFFSET_MAX
;
932 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
935 gc_marker
= m
->gc_marker
;
937 while ((u
= m
->gc_queue
)) {
938 assert(u
->in_gc_queue
);
940 unit_gc_sweep(u
, gc_marker
);
942 LIST_REMOVE(gc_queue
, m
->gc_queue
, u
);
943 u
->in_gc_queue
= false;
947 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
948 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
950 log_unit_debug(u
, "Collecting.");
951 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
952 unit_add_to_cleanup_queue(u
);
956 m
->n_in_gc_queue
= 0;
961 static void manager_clear_jobs_and_units(Manager
*m
) {
966 while ((u
= hashmap_first(m
->units
)))
969 manager_dispatch_cleanup_queue(m
);
971 assert(!m
->load_queue
);
972 assert(!m
->run_queue
);
973 assert(!m
->dbus_unit_queue
);
974 assert(!m
->dbus_job_queue
);
975 assert(!m
->cleanup_queue
);
976 assert(!m
->gc_queue
);
978 assert(hashmap_isempty(m
->jobs
));
979 assert(hashmap_isempty(m
->units
));
982 m
->n_running_jobs
= 0;
985 Manager
* manager_free(Manager
*m
) {
992 manager_clear_jobs_and_units(m
);
994 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
995 if (unit_vtable
[c
]->shutdown
)
996 unit_vtable
[c
]->shutdown(m
);
998 /* If we reexecute ourselves, we keep the root cgroup
1000 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1002 lookup_paths_flush_generator(&m
->lookup_paths
);
1006 dynamic_user_vacuum(m
, false);
1007 hashmap_free(m
->dynamic_users
);
1009 hashmap_free(m
->units
);
1010 hashmap_free(m
->jobs
);
1011 hashmap_free(m
->watch_pids1
);
1012 hashmap_free(m
->watch_pids2
);
1013 hashmap_free(m
->watch_bus
);
1015 set_free(m
->startup_units
);
1016 set_free(m
->failed_units
);
1018 sd_event_source_unref(m
->signal_event_source
);
1019 sd_event_source_unref(m
->notify_event_source
);
1020 sd_event_source_unref(m
->cgroups_agent_event_source
);
1021 sd_event_source_unref(m
->time_change_event_source
);
1022 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1023 sd_event_source_unref(m
->run_queue_event_source
);
1025 safe_close(m
->signal_fd
);
1026 safe_close(m
->notify_fd
);
1027 safe_close(m
->cgroups_agent_fd
);
1028 safe_close(m
->time_change_fd
);
1029 safe_close(m
->kdbus_fd
);
1031 manager_close_ask_password(m
);
1033 manager_close_idle_pipe(m
);
1035 udev_unref(m
->udev
);
1036 sd_event_unref(m
->event
);
1038 free(m
->notify_socket
);
1040 lookup_paths_free(&m
->lookup_paths
);
1041 strv_free(m
->environment
);
1043 hashmap_free(m
->cgroup_unit
);
1044 set_free_free(m
->unit_path_cache
);
1046 free(m
->switch_root
);
1047 free(m
->switch_root_init
);
1049 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1050 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1052 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1053 hashmap_free(m
->units_requiring_mounts_for
);
1059 void manager_enumerate(Manager
*m
) {
1064 /* Let's ask every type to load all units from disk/kernel
1065 * that it might know */
1066 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1067 if (!unit_type_supported(c
)) {
1068 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1072 if (!unit_vtable
[c
]->enumerate
)
1075 unit_vtable
[c
]->enumerate(m
);
1078 manager_dispatch_load_queue(m
);
1081 static void manager_coldplug(Manager
*m
) {
1089 /* Then, let's set up their initial state. */
1090 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1092 /* ignore aliases */
1096 r
= unit_coldplug(u
);
1098 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1102 static void manager_build_unit_path_cache(Manager
*m
) {
1108 set_free_free(m
->unit_path_cache
);
1110 m
->unit_path_cache
= set_new(&string_hash_ops
);
1111 if (!m
->unit_path_cache
) {
1116 /* This simply builds a list of files we know exist, so that
1117 * we don't always have to go to disk */
1119 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1120 _cleanup_closedir_
DIR *d
= NULL
;
1125 if (errno
!= ENOENT
)
1126 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1130 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1133 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
, NULL
);
1139 r
= set_consume(m
->unit_path_cache
, p
);
1148 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1149 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1152 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1158 HASHMAP_FOREACH(u
, m
->units
, i
) {
1160 if (fdset_size(fds
) <= 0)
1163 if (!UNIT_VTABLE(u
)->distribute_fds
)
1166 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1170 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1175 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1179 /* Make sure the transient directory always exists, so that it remains in the search path */
1180 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1184 dual_timestamp_get(&m
->generators_start_timestamp
);
1185 r
= manager_run_generators(m
);
1186 dual_timestamp_get(&m
->generators_finish_timestamp
);
1190 lookup_paths_reduce(&m
->lookup_paths
);
1191 manager_build_unit_path_cache(m
);
1193 /* If we will deserialize make sure that during enumeration
1194 * this is already known, so we increase the counter here
1199 /* First, enumerate what we can from all config files */
1200 dual_timestamp_get(&m
->units_load_start_timestamp
);
1201 manager_enumerate(m
);
1202 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1204 /* Second, deserialize if there is something to deserialize */
1206 r
= manager_deserialize(m
, serialization
, fds
);
1208 /* Any fds left? Find some unit which wants them. This is
1209 * useful to allow container managers to pass some file
1210 * descriptors to us pre-initialized. This enables
1211 * socket-based activation of entire containers. */
1212 manager_distribute_fds(m
, fds
);
1214 /* We might have deserialized the notify fd, but if we didn't
1215 * then let's create the bus now */
1216 q
= manager_setup_notify(m
);
1217 if (q
< 0 && r
== 0)
1220 q
= manager_setup_cgroups_agent(m
);
1221 if (q
< 0 && r
== 0)
1224 /* We might have deserialized the kdbus control fd, but if we
1225 * didn't, then let's create the bus now. */
1226 manager_connect_bus(m
, !!serialization
);
1227 bus_track_coldplug(m
, &m
->subscribed
, &m
->deserialized_subscribed
);
1229 /* Third, fire things up! */
1230 manager_coldplug(m
);
1232 /* Release any dynamic users no longer referenced */
1233 dynamic_user_vacuum(m
, true);
1235 if (serialization
) {
1236 assert(m
->n_reloading
> 0);
1239 /* Let's wait for the UnitNew/JobNew messages being
1240 * sent, before we notify that the reload is
1242 m
->send_reloading_done
= true;
1248 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1253 assert(type
< _JOB_TYPE_MAX
);
1255 assert(mode
< _JOB_MODE_MAX
);
1257 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1258 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1260 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1261 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1263 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1265 type
= job_type_collapse(type
, unit
);
1267 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1271 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1272 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1273 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1277 if (mode
== JOB_ISOLATE
) {
1278 r
= transaction_add_isolate_jobs(tr
, m
);
1283 r
= transaction_activate(tr
, m
, mode
, e
);
1287 log_unit_debug(unit
,
1288 "Enqueued job %s/%s as %u", unit
->id
,
1289 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1292 *_ret
= tr
->anchor_job
;
1294 transaction_free(tr
);
1298 transaction_abort(tr
);
1299 transaction_free(tr
);
1303 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1308 assert(type
< _JOB_TYPE_MAX
);
1310 assert(mode
< _JOB_MODE_MAX
);
1312 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1316 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1319 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1320 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1324 assert(type
< _JOB_TYPE_MAX
);
1326 assert(mode
< _JOB_MODE_MAX
);
1328 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1330 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1335 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1338 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1341 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1345 return hashmap_get(m
->units
, name
);
1348 unsigned manager_dispatch_load_queue(Manager
*m
) {
1354 /* Make sure we are not run recursively */
1355 if (m
->dispatching_load_queue
)
1358 m
->dispatching_load_queue
= true;
1360 /* Dispatches the load queue. Takes a unit from the queue and
1361 * tries to load its data until the queue is empty */
1363 while ((u
= m
->load_queue
)) {
1364 assert(u
->in_load_queue
);
1370 m
->dispatching_load_queue
= false;
1374 int manager_load_unit_prepare(
1386 assert(name
|| path
);
1388 /* This will prepare the unit for loading, but not actually
1389 * load anything from disk. */
1391 if (path
&& !is_path(path
))
1392 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1395 name
= basename(path
);
1397 t
= unit_name_to_type(name
);
1399 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1400 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1401 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1403 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1406 ret
= manager_get_unit(m
, name
);
1412 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1417 ret
->fragment_path
= strdup(path
);
1418 if (!ret
->fragment_path
) {
1424 r
= unit_add_name(ret
, name
);
1430 unit_add_to_load_queue(ret
);
1431 unit_add_to_dbus_queue(ret
);
1432 unit_add_to_gc_queue(ret
);
1440 int manager_load_unit(
1451 /* This will load the service information files, but not actually
1452 * start any services or anything. */
1454 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1458 manager_dispatch_load_queue(m
);
1461 *_ret
= unit_follow_merge(*_ret
);
1466 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1473 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1474 job_dump(j
, f
, prefix
);
1477 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1485 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1487 unit_dump(u
, f
, prefix
);
1490 void manager_clear_jobs(Manager
*m
) {
1495 while ((j
= hashmap_first(m
->jobs
)))
1496 /* No need to recurse. We're cancelling all jobs. */
1497 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1500 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1501 Manager
*m
= userdata
;
1507 while ((j
= m
->run_queue
)) {
1508 assert(j
->installed
);
1509 assert(j
->in_run_queue
);
1511 job_run_and_invalidate(j
);
1514 if (m
->n_running_jobs
> 0)
1515 manager_watch_jobs_in_progress(m
);
1517 if (m
->n_on_console
> 0)
1518 manager_watch_idle_pipe(m
);
1523 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1530 if (m
->dispatching_dbus_queue
)
1533 m
->dispatching_dbus_queue
= true;
1535 while ((u
= m
->dbus_unit_queue
)) {
1536 assert(u
->in_dbus_queue
);
1538 bus_unit_send_change_signal(u
);
1542 while ((j
= m
->dbus_job_queue
)) {
1543 assert(j
->in_dbus_queue
);
1545 bus_job_send_change_signal(j
);
1549 m
->dispatching_dbus_queue
= false;
1551 if (m
->send_reloading_done
) {
1552 m
->send_reloading_done
= false;
1554 bus_manager_send_reloading(m
, false);
1557 if (m
->queued_message
)
1558 bus_send_queued_message(m
);
1563 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1564 Manager
*m
= userdata
;
1565 char buf
[PATH_MAX
+1];
1568 n
= recv(fd
, buf
, sizeof(buf
), 0);
1570 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1572 log_error("Got zero-length cgroups agent message, ignoring.");
1575 if ((size_t) n
>= sizeof(buf
)) {
1576 log_error("Got overly long cgroups agent message, ignoring.");
1580 if (memchr(buf
, 0, n
)) {
1581 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1586 manager_notify_cgroup_empty(m
, buf
);
1587 bus_forward_agent_released(m
, buf
);
1592 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, size_t n
, FDSet
*fds
) {
1593 _cleanup_strv_free_
char **tags
= NULL
;
1600 tags
= strv_split(buf
, "\n\r");
1606 if (UNIT_VTABLE(u
)->notify_message
)
1607 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1609 log_unit_debug(u
, "Got notification message for unit. Ignoring.");
1612 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1614 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1615 Manager
*m
= userdata
;
1616 char buf
[NOTIFY_BUFFER_MAX
+1];
1617 struct iovec iovec
= {
1619 .iov_len
= sizeof(buf
)-1,
1622 struct cmsghdr cmsghdr
;
1623 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1624 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1626 struct msghdr msghdr
= {
1629 .msg_control
= &control
,
1630 .msg_controllen
= sizeof(control
),
1633 struct cmsghdr
*cmsg
;
1634 struct ucred
*ucred
= NULL
;
1637 int r
, *fd_array
= NULL
;
1642 assert(m
->notify_fd
== fd
);
1644 if (revents
!= EPOLLIN
) {
1645 log_warning("Got unexpected poll event for notify fd.");
1649 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
1651 if (errno
== EAGAIN
|| errno
== EINTR
)
1657 CMSG_FOREACH(cmsg
, &msghdr
) {
1658 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1660 fd_array
= (int*) CMSG_DATA(cmsg
);
1661 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1663 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1664 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1665 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1667 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1674 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1676 close_many(fd_array
, n_fds
);
1681 if (!ucred
|| ucred
->pid
<= 0) {
1682 log_warning("Received notify message without valid credentials. Ignoring.");
1686 if ((size_t) n
>= sizeof(buf
)) {
1687 log_warning("Received notify message exceeded maximum size. Ignoring.");
1693 /* Notify every unit that might be interested, but try
1694 * to avoid notifying the same one multiple times. */
1695 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1697 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, n
, fds
);
1701 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1702 if (u2
&& u2
!= u1
) {
1703 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, n
, fds
);
1707 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1708 if (u3
&& u3
!= u2
&& u3
!= u1
) {
1709 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, n
, fds
);
1714 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1716 if (fdset_size(fds
) > 0)
1717 log_warning("Got auxiliary fds with notification message, closing all.");
1722 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1729 sd_event_get_iteration(m
->event
, &iteration
);
1731 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1733 unit_unwatch_pid(u
, si
->si_pid
);
1735 if (UNIT_VTABLE(u
)->sigchld_event
) {
1736 if (set_size(u
->pids
) <= 1 ||
1737 iteration
!= u
->sigchldgen
||
1738 unit_main_pid(u
) == si
->si_pid
||
1739 unit_control_pid(u
) == si
->si_pid
) {
1740 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1741 u
->sigchldgen
= iteration
;
1743 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1747 static int manager_dispatch_sigchld(Manager
*m
) {
1753 /* First we call waitd() for a PID and do not reap the
1754 * zombie. That way we can still access /proc/$PID for
1755 * it while it is a zombie. */
1756 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1758 if (errno
== ECHILD
)
1770 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1771 _cleanup_free_
char *name
= NULL
;
1774 get_process_comm(si
.si_pid
, &name
);
1776 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1777 si
.si_pid
, strna(name
),
1778 sigchld_code_to_string(si
.si_code
),
1780 strna(si
.si_code
== CLD_EXITED
1781 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1782 : signal_to_string(si
.si_status
)));
1784 /* And now figure out the unit this belongs
1785 * to, it might be multiple... */
1786 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1788 invoke_sigchld_event(m
, u1
, &si
);
1789 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1791 invoke_sigchld_event(m
, u2
, &si
);
1792 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1793 if (u3
&& u3
!= u2
&& u3
!= u1
)
1794 invoke_sigchld_event(m
, u3
, &si
);
1797 /* And now, we actually reap the zombie. */
1798 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1809 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1810 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1813 log_debug("Activating special unit %s", name
);
1815 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1817 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1822 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1823 Manager
*m
= userdata
;
1825 struct signalfd_siginfo sfsi
;
1826 bool sigchld
= false;
1830 assert(m
->signal_fd
== fd
);
1832 if (revents
!= EPOLLIN
) {
1833 log_warning("Got unexpected events from signal file descriptor.");
1838 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1839 if (n
!= sizeof(sfsi
)) {
1844 if (errno
== EINTR
|| errno
== EAGAIN
)
1850 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1851 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1852 ? LOG_DEBUG
: LOG_INFO
,
1855 switch (sfsi
.ssi_signo
) {
1862 if (MANAGER_IS_SYSTEM(m
)) {
1863 /* This is for compatibility with the
1864 * original sysvinit */
1865 m
->exit_code
= MANAGER_REEXECUTE
;
1872 if (MANAGER_IS_SYSTEM(m
)) {
1874 /* If the user presses C-A-D more than
1875 * 7 times within 2s, we reboot
1878 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
))
1879 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1881 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1882 status_printf(NULL
, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1883 m
->exit_code
= MANAGER_REBOOT
;
1889 /* Run the exit target if there is one, if not, just exit. */
1890 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
1891 m
->exit_code
= MANAGER_EXIT
;
1898 if (MANAGER_IS_SYSTEM(m
))
1899 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
1901 /* This is a nop on non-init */
1905 if (MANAGER_IS_SYSTEM(m
))
1906 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
1908 /* This is a nop on non-init */
1914 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1916 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
1917 log_info("Trying to reconnect to bus...");
1921 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
1922 log_info("Loading D-Bus service...");
1923 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
1930 _cleanup_free_
char *dump
= NULL
;
1931 _cleanup_fclose_
FILE *f
= NULL
;
1934 f
= open_memstream(&dump
, &size
);
1936 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
1940 manager_dump_units(m
, f
, "\t");
1941 manager_dump_jobs(m
, f
, "\t");
1943 r
= fflush_and_check(f
);
1945 log_warning_errno(r
, "Failed to write status stream: %m");
1949 log_dump(LOG_INFO
, dump
);
1954 m
->exit_code
= MANAGER_RELOAD
;
1959 /* Starting SIGRTMIN+0 */
1960 static const char * const target_table
[] = {
1961 [0] = SPECIAL_DEFAULT_TARGET
,
1962 [1] = SPECIAL_RESCUE_TARGET
,
1963 [2] = SPECIAL_EMERGENCY_TARGET
,
1964 [3] = SPECIAL_HALT_TARGET
,
1965 [4] = SPECIAL_POWEROFF_TARGET
,
1966 [5] = SPECIAL_REBOOT_TARGET
,
1967 [6] = SPECIAL_KEXEC_TARGET
1970 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1971 static const ManagerExitCode code_table
[] = {
1973 [1] = MANAGER_POWEROFF
,
1974 [2] = MANAGER_REBOOT
,
1978 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
1979 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
1980 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
1981 manager_start_target(m
, target_table
[idx
],
1982 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
1986 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
1987 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
1988 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
1992 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
1995 manager_set_show_status(m
, SHOW_STATUS_YES
);
1999 manager_set_show_status(m
, SHOW_STATUS_NO
);
2003 log_set_max_level(LOG_DEBUG
);
2004 log_info("Setting log level to debug.");
2008 log_set_max_level(LOG_INFO
);
2009 log_info("Setting log level to info.");
2013 if (MANAGER_IS_USER(m
)) {
2014 m
->exit_code
= MANAGER_EXIT
;
2018 /* This is a nop on init */
2022 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2023 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2024 log_notice("Setting log target to journal-or-kmsg.");
2028 log_set_target(LOG_TARGET_CONSOLE
);
2029 log_notice("Setting log target to console.");
2033 log_set_target(LOG_TARGET_KMSG
);
2034 log_notice("Setting log target to kmsg.");
2038 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2045 manager_dispatch_sigchld(m
);
2050 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2051 Manager
*m
= userdata
;
2056 assert(m
->time_change_fd
== fd
);
2058 log_struct(LOG_INFO
,
2059 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE
),
2060 LOG_MESSAGE("Time has been changed"),
2063 /* Restart the watch */
2064 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2065 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2067 manager_setup_time_change(m
);
2069 HASHMAP_FOREACH(u
, m
->units
, i
)
2070 if (UNIT_VTABLE(u
)->time_change
)
2071 UNIT_VTABLE(u
)->time_change(u
);
2076 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2077 Manager
*m
= userdata
;
2080 assert(m
->idle_pipe
[2] == fd
);
2082 m
->no_console_output
= m
->n_on_console
> 0;
2084 manager_close_idle_pipe(m
);
2089 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2090 Manager
*m
= userdata
;
2097 manager_print_jobs_in_progress(m
);
2099 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2100 r
= sd_event_source_set_time(source
, next
);
2104 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2107 int manager_loop(Manager
*m
) {
2110 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2113 m
->exit_code
= MANAGER_OK
;
2115 /* Release the path cache */
2116 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2118 manager_check_finished(m
);
2120 /* There might still be some zombies hanging around from
2121 * before we were exec()'ed. Let's reap them. */
2122 r
= manager_dispatch_sigchld(m
);
2126 while (m
->exit_code
== MANAGER_OK
) {
2129 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2132 if (!ratelimit_test(&rl
)) {
2133 /* Yay, something is going seriously wrong, pause a little */
2134 log_warning("Looping too fast. Throttling execution a little.");
2138 if (manager_dispatch_load_queue(m
) > 0)
2141 if (manager_dispatch_gc_queue(m
) > 0)
2144 if (manager_dispatch_cleanup_queue(m
) > 0)
2147 if (manager_dispatch_cgroup_queue(m
) > 0)
2150 if (manager_dispatch_dbus_queue(m
) > 0)
2153 /* Sleep for half the watchdog time */
2154 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2155 wait_usec
= m
->runtime_watchdog
/ 2;
2159 wait_usec
= USEC_INFINITY
;
2161 r
= sd_event_run(m
->event
, wait_usec
);
2163 return log_error_errno(r
, "Failed to run event loop: %m");
2166 return m
->exit_code
;
2169 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2170 _cleanup_free_
char *n
= NULL
;
2178 r
= unit_name_from_dbus_path(s
, &n
);
2182 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2191 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2201 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2205 r
= safe_atou(p
, &id
);
2209 j
= manager_get_job(m
, id
);
2218 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2221 _cleanup_free_
char *p
= NULL
;
2225 if (!MANAGER_IS_SYSTEM(m
))
2228 audit_fd
= get_audit_fd();
2232 /* Don't generate audit events if the service was already
2233 * started and we're just deserializing */
2234 if (MANAGER_IS_RELOADING(m
))
2237 if (u
->type
!= UNIT_SERVICE
)
2240 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2242 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2246 msg
= strjoina("unit=", p
);
2247 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2249 /* We aren't allowed to send audit messages?
2250 * Then let's not retry again. */
2253 log_warning_errno(errno
, "Failed to send audit message: %m");
2259 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2260 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2261 _cleanup_free_
char *message
= NULL
;
2262 _cleanup_close_
int fd
= -1;
2265 /* Don't generate plymouth events if the service was already
2266 * started and we're just deserializing */
2267 if (MANAGER_IS_RELOADING(m
))
2270 if (!MANAGER_IS_SYSTEM(m
))
2273 if (detect_container() > 0)
2276 if (u
->type
!= UNIT_SERVICE
&&
2277 u
->type
!= UNIT_MOUNT
&&
2278 u
->type
!= UNIT_SWAP
)
2281 /* We set SOCK_NONBLOCK here so that we rather drop the
2282 * message then wait for plymouth */
2283 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2285 log_error_errno(errno
, "socket() failed: %m");
2289 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2291 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2292 log_error_errno(errno
, "connect() failed: %m");
2296 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2302 if (write(fd
, message
, n
+ 1) != n
+ 1)
2303 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2304 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2307 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2314 path
= MANAGER_IS_SYSTEM(m
) ? "/run/systemd" : "/tmp";
2315 fd
= open_tmpfile_unlinkable(path
, O_RDWR
|O_CLOEXEC
);
2319 log_debug("Serializing state to %s", path
);
2321 f
= fdopen(fd
, "w+");
2332 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2345 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2346 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2347 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2348 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2350 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2351 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2352 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2353 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2356 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2357 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2358 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2359 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2360 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2361 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2362 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2363 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2366 if (!switching_root
) {
2367 STRV_FOREACH(e
, m
->environment
) {
2368 _cleanup_free_
char *ce
;
2374 fprintf(f
, "env=%s\n", *e
);
2378 if (m
->notify_fd
>= 0) {
2381 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2385 fprintf(f
, "notify-fd=%i\n", copy
);
2386 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2389 if (m
->cgroups_agent_fd
>= 0) {
2392 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2396 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2399 if (m
->kdbus_fd
>= 0) {
2402 copy
= fdset_put_dup(fds
, m
->kdbus_fd
);
2406 fprintf(f
, "kdbus-fd=%i\n", copy
);
2409 bus_track_serialize(m
->subscribed
, f
);
2411 r
= dynamic_user_serialize(m
, f
, fds
);
2417 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2425 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2432 assert(m
->n_reloading
> 0);
2438 r
= bus_fdset_add_all(m
, fds
);
2445 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2451 log_debug("Deserializing state...");
2456 char line
[LINE_MAX
], *l
;
2458 if (!fgets(line
, sizeof(line
), f
)) {
2473 if (startswith(l
, "current-job-id=")) {
2476 if (safe_atou32(l
+15, &id
) < 0)
2477 log_debug("Failed to parse current job id value %s", l
+15);
2479 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2481 } else if (startswith(l
, "n-installed-jobs=")) {
2484 if (safe_atou32(l
+17, &n
) < 0)
2485 log_debug("Failed to parse installed jobs counter %s", l
+17);
2487 m
->n_installed_jobs
+= n
;
2489 } else if (startswith(l
, "n-failed-jobs=")) {
2492 if (safe_atou32(l
+14, &n
) < 0)
2493 log_debug("Failed to parse failed jobs counter %s", l
+14);
2495 m
->n_failed_jobs
+= n
;
2497 } else if (startswith(l
, "taint-usr=")) {
2500 b
= parse_boolean(l
+10);
2502 log_debug("Failed to parse taint /usr flag %s", l
+10);
2504 m
->taint_usr
= m
->taint_usr
|| b
;
2506 } else if (startswith(l
, "firmware-timestamp="))
2507 dual_timestamp_deserialize(l
+19, &m
->firmware_timestamp
);
2508 else if (startswith(l
, "loader-timestamp="))
2509 dual_timestamp_deserialize(l
+17, &m
->loader_timestamp
);
2510 else if (startswith(l
, "kernel-timestamp="))
2511 dual_timestamp_deserialize(l
+17, &m
->kernel_timestamp
);
2512 else if (startswith(l
, "initrd-timestamp="))
2513 dual_timestamp_deserialize(l
+17, &m
->initrd_timestamp
);
2514 else if (startswith(l
, "userspace-timestamp="))
2515 dual_timestamp_deserialize(l
+20, &m
->userspace_timestamp
);
2516 else if (startswith(l
, "finish-timestamp="))
2517 dual_timestamp_deserialize(l
+17, &m
->finish_timestamp
);
2518 else if (startswith(l
, "security-start-timestamp="))
2519 dual_timestamp_deserialize(l
+25, &m
->security_start_timestamp
);
2520 else if (startswith(l
, "security-finish-timestamp="))
2521 dual_timestamp_deserialize(l
+26, &m
->security_finish_timestamp
);
2522 else if (startswith(l
, "generators-start-timestamp="))
2523 dual_timestamp_deserialize(l
+27, &m
->generators_start_timestamp
);
2524 else if (startswith(l
, "generators-finish-timestamp="))
2525 dual_timestamp_deserialize(l
+28, &m
->generators_finish_timestamp
);
2526 else if (startswith(l
, "units-load-start-timestamp="))
2527 dual_timestamp_deserialize(l
+27, &m
->units_load_start_timestamp
);
2528 else if (startswith(l
, "units-load-finish-timestamp="))
2529 dual_timestamp_deserialize(l
+28, &m
->units_load_finish_timestamp
);
2530 else if (startswith(l
, "env=")) {
2531 _cleanup_free_
char *uce
= NULL
;
2534 r
= cunescape(l
+ 4, UNESCAPE_RELAX
, &uce
);
2538 e
= strv_env_set(m
->environment
, uce
);
2544 strv_free(m
->environment
);
2547 } else if (startswith(l
, "notify-fd=")) {
2550 if (safe_atoi(l
+ 10, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2551 log_debug("Failed to parse notify fd: %s", l
+ 10);
2553 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2554 safe_close(m
->notify_fd
);
2555 m
->notify_fd
= fdset_remove(fds
, fd
);
2558 } else if (startswith(l
, "notify-socket=")) {
2567 free(m
->notify_socket
);
2568 m
->notify_socket
= n
;
2570 } else if (startswith(l
, "cgroups-agent-fd=")) {
2573 if (safe_atoi(l
+ 17, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2574 log_debug("Failed to parse cgroups agent fd: %s", l
+ 10);
2576 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2577 safe_close(m
->cgroups_agent_fd
);
2578 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2581 } else if (startswith(l
, "kdbus-fd=")) {
2584 if (safe_atoi(l
+ 9, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2585 log_debug("Failed to parse kdbus fd: %s", l
+ 9);
2587 safe_close(m
->kdbus_fd
);
2588 m
->kdbus_fd
= fdset_remove(fds
, fd
);
2591 } else if (startswith(l
, "dynamic-user="))
2592 dynamic_user_deserialize_one(m
, l
+ 13, fds
);
2596 k
= bus_track_deserialize_item(&m
->deserialized_subscribed
, l
);
2598 log_debug_errno(k
, "Failed to deserialize bus tracker object: %m");
2600 log_debug("Unknown serialization item '%s'", l
);
2606 char name
[UNIT_NAME_MAX
+2];
2609 if (!fgets(name
, sizeof(name
), f
)) {
2620 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2624 r
= unit_deserialize(u
, f
, fds
);
2633 assert(m
->n_reloading
> 0);
2639 int manager_reload(Manager
*m
) {
2641 _cleanup_fclose_
FILE *f
= NULL
;
2642 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2646 r
= manager_open_serialization(m
, &f
);
2651 bus_manager_send_reloading(m
, true);
2659 r
= manager_serialize(m
, f
, fds
, false);
2665 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2670 /* From here on there is no way back. */
2671 manager_clear_jobs_and_units(m
);
2672 lookup_paths_flush_generator(&m
->lookup_paths
);
2673 lookup_paths_free(&m
->lookup_paths
);
2674 dynamic_user_vacuum(m
, false);
2676 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2677 if (q
< 0 && r
>= 0)
2680 /* Find new unit paths */
2681 q
= manager_run_generators(m
);
2682 if (q
< 0 && r
>= 0)
2685 lookup_paths_reduce(&m
->lookup_paths
);
2686 manager_build_unit_path_cache(m
);
2688 /* First, enumerate what we can from all config files */
2689 manager_enumerate(m
);
2691 /* Second, deserialize our stored data */
2692 q
= manager_deserialize(m
, f
, fds
);
2693 if (q
< 0 && r
>= 0)
2699 /* Re-register notify_fd as event source */
2700 q
= manager_setup_notify(m
);
2701 if (q
< 0 && r
>= 0)
2704 q
= manager_setup_cgroups_agent(m
);
2705 if (q
< 0 && r
>= 0)
2708 /* Third, fire things up! */
2709 manager_coldplug(m
);
2711 /* Release any dynamic users no longer referenced */
2712 dynamic_user_vacuum(m
, true);
2714 /* Sync current state of bus names with our set of listening units */
2716 manager_sync_bus_names(m
, m
->api_bus
);
2718 assert(m
->n_reloading
> 0);
2721 m
->send_reloading_done
= true;
2726 void manager_reset_failed(Manager
*m
) {
2732 HASHMAP_FOREACH(u
, m
->units
, i
)
2733 unit_reset_failed(u
);
2736 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2742 /* Returns true if the unit is inactive or going down */
2743 u
= manager_get_unit(m
, name
);
2747 return unit_inactive_or_pending(u
);
2750 static void manager_notify_finished(Manager
*m
) {
2751 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2752 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2757 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2759 /* Note that m->kernel_usec.monotonic is always at 0,
2760 * and m->firmware_usec.monotonic and
2761 * m->loader_usec.monotonic should be considered
2762 * negative values. */
2764 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2765 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2766 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2767 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2769 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2771 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2772 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2774 log_struct(LOG_INFO
,
2775 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2776 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2777 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2778 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2779 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2780 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2781 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2782 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2783 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2786 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2789 log_struct(LOG_INFO
,
2790 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2791 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2792 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2793 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2794 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2795 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2796 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2800 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2801 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2803 log_struct(LOG_INFO
,
2804 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2805 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2806 LOG_MESSAGE("Startup finished in %s.",
2807 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2811 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2815 "STATUS=Startup finished in %s.",
2816 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2819 void manager_check_finished(Manager
*m
) {
2822 if (MANAGER_IS_RELOADING(m
))
2825 /* Verify that we are actually running currently. Initially
2826 * the exit code is set to invalid, and during operation it is
2827 * then set to MANAGER_OK */
2828 if (m
->exit_code
!= MANAGER_OK
)
2831 if (hashmap_size(m
->jobs
) > 0) {
2832 if (m
->jobs_in_progress_event_source
)
2833 /* Ignore any failure, this is only for feedback */
2834 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2839 manager_flip_auto_status(m
, false);
2841 /* Notify Type=idle units that we are done now */
2842 manager_close_idle_pipe(m
);
2844 /* Turn off confirm spawn now */
2845 m
->confirm_spawn
= false;
2847 /* No need to update ask password status when we're going non-interactive */
2848 manager_close_ask_password(m
);
2850 /* This is no longer the first boot */
2851 manager_set_first_boot(m
, false);
2853 if (dual_timestamp_is_set(&m
->finish_timestamp
))
2856 dual_timestamp_get(&m
->finish_timestamp
);
2858 manager_notify_finished(m
);
2860 manager_invalidate_startup_units(m
);
2863 static int manager_run_generators(Manager
*m
) {
2864 _cleanup_strv_free_
char **paths
= NULL
;
2865 const char *argv
[5];
2874 paths
= generator_binary_paths(m
->unit_file_scope
);
2878 /* Optimize by skipping the whole process by not creating output directories
2879 * if no generators are found. */
2880 STRV_FOREACH(path
, paths
) {
2881 if (access(*path
, F_OK
) >= 0)
2883 if (errno
!= ENOENT
)
2884 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
2890 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
2894 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
2895 argv
[1] = m
->lookup_paths
.generator
;
2896 argv
[2] = m
->lookup_paths
.generator_early
;
2897 argv
[3] = m
->lookup_paths
.generator_late
;
2900 RUN_WITH_UMASK(0022)
2901 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, (char**) argv
);
2904 lookup_paths_trim_generator(&m
->lookup_paths
);
2908 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
2909 char **a
= NULL
, **b
= NULL
, **l
;
2914 if (!strv_isempty(minus
)) {
2915 a
= strv_env_delete(l
, 1, minus
);
2922 if (!strv_isempty(plus
)) {
2923 b
= strv_env_merge(2, l
, plus
);
2932 if (m
->environment
!= l
)
2933 strv_free(m
->environment
);
2940 manager_clean_environment(m
);
2941 strv_sort(m
->environment
);
2946 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
2951 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2952 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
2954 if (!default_rlimit
[i
])
2957 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
2965 void manager_recheck_journal(Manager
*m
) {
2970 if (!MANAGER_IS_SYSTEM(m
))
2973 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
2974 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
2975 log_close_journal();
2979 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
2980 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
2981 log_close_journal();
2985 /* Hmm, OK, so the socket is fully up and the service is up
2986 * too, then let's make use of the thing. */
2990 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
2992 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
2994 if (!MANAGER_IS_SYSTEM(m
))
2997 if (m
->show_status
!= mode
)
2998 log_debug("%s showing of status.",
2999 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3000 m
->show_status
= mode
;
3003 (void) touch("/run/systemd/show-status");
3005 (void) unlink("/run/systemd/show-status");
3008 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3011 if (!MANAGER_IS_SYSTEM(m
))
3014 if (m
->no_console_output
)
3017 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3020 /* If we cannot find out the status properly, just proceed. */
3021 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3024 if (m
->show_status
> 0)
3030 void manager_set_first_boot(Manager
*m
, bool b
) {
3033 if (!MANAGER_IS_SYSTEM(m
))
3036 if (m
->first_boot
!= (int) b
) {
3038 (void) touch("/run/systemd/first-boot");
3040 (void) unlink("/run/systemd/first-boot");
3046 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3049 /* If m is NULL, assume we're after shutdown and let the messages through. */
3051 if (m
&& !manager_get_show_status(m
, type
))
3054 /* XXX We should totally drop the check for ephemeral here
3055 * and thus effectively make 'Type=idle' pointless. */
3056 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3059 va_start(ap
, format
);
3060 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3064 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3065 char p
[strlen(path
)+1];
3071 path_kill_slashes(p
);
3073 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3076 const char *manager_get_runtime_prefix(Manager
*m
) {
3079 return MANAGER_IS_SYSTEM(m
) ?
3081 getenv("XDG_RUNTIME_DIR");
3084 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3089 assert(u
->manager
== m
);
3091 size
= set_size(m
->failed_units
);
3094 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3098 if (set_put(m
->failed_units
, u
) < 0)
3101 (void) set_remove(m
->failed_units
, u
);
3103 if (set_size(m
->failed_units
) != size
)
3104 bus_manager_send_change_signal(m
);
3109 ManagerState
manager_state(Manager
*m
) {
3114 /* Did we ever finish booting? If not then we are still starting up */
3115 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3117 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3118 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3119 return MANAGER_INITIALIZING
;
3121 return MANAGER_STARTING
;
3124 /* Is the special shutdown target queued? If so, we are in shutdown state */
3125 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3126 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3127 return MANAGER_STOPPING
;
3129 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3130 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3131 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3132 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3133 return MANAGER_MAINTENANCE
;
3135 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3136 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3137 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3138 return MANAGER_MAINTENANCE
;
3140 /* Are there any failed units? If so, we are in degraded mode */
3141 if (set_size(m
->failed_units
) > 0)
3142 return MANAGER_DEGRADED
;
3144 return MANAGER_RUNNING
;
3147 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3148 [MANAGER_INITIALIZING
] = "initializing",
3149 [MANAGER_STARTING
] = "starting",
3150 [MANAGER_RUNNING
] = "running",
3151 [MANAGER_DEGRADED
] = "degraded",
3152 [MANAGER_MAINTENANCE
] = "maintenance",
3153 [MANAGER_STOPPING
] = "stopping",
3156 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);