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"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
74 #include "signal-util.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
95 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
98 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
103 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
104 static int manager_run_generators(Manager
*m
);
106 static void manager_watch_jobs_in_progress(Manager
*m
) {
112 if (m
->jobs_in_progress_event_source
)
115 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
116 r
= sd_event_add_time(
118 &m
->jobs_in_progress_event_source
,
121 manager_dispatch_jobs_in_progress
, m
);
125 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
128 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
130 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
133 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
134 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
138 p
= mempset(p
, ' ', pos
-2);
139 p
= stpcpy(p
, ANSI_RED
);
143 if (pos
> 0 && pos
<= width
) {
144 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
148 p
= stpcpy(p
, ANSI_NORMAL
);
151 p
= stpcpy(p
, ANSI_RED
);
154 p
= mempset(p
, ' ', width
-1-pos
);
155 strcpy(p
, ANSI_NORMAL
);
159 void manager_flip_auto_status(Manager
*m
, bool enable
) {
163 if (m
->show_status
== SHOW_STATUS_AUTO
)
164 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
166 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
167 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
171 static void manager_print_jobs_in_progress(Manager
*m
) {
172 _cleanup_free_
char *job_of_n
= NULL
;
175 unsigned counter
= 0, print_nr
;
176 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
178 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
182 assert(m
->n_running_jobs
> 0);
184 manager_flip_auto_status(m
, true);
186 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
188 HASHMAP_FOREACH(j
, m
->jobs
, i
)
189 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
192 /* m->n_running_jobs must be consistent with the contents of m->jobs,
193 * so the above loop must have succeeded in finding j. */
194 assert(counter
== print_nr
+ 1);
197 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
199 cylon_pos
= 14 - cylon_pos
;
200 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
202 m
->jobs_in_progress_iteration
++;
204 if (m
->n_running_jobs
> 1) {
205 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
209 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
210 if (job_get_timeout(j
, &x
) > 0)
211 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
213 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
214 "%sA %s job is running for %s (%s / %s)",
216 job_type_to_string(j
->type
),
217 unit_description(j
->unit
),
221 static int have_ask_password(void) {
222 _cleanup_closedir_
DIR *dir
;
224 dir
= opendir("/run/systemd/ask-password");
237 if (!de
&& errno
> 0)
242 if (startswith(de
->d_name
, "ask."))
247 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
248 int fd
, uint32_t revents
, void *userdata
) {
249 Manager
*m
= userdata
;
255 m
->have_ask_password
= have_ask_password();
256 if (m
->have_ask_password
< 0)
257 /* Log error but continue. Negative have_ask_password
258 * is treated as unknown status. */
259 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
264 static void manager_close_ask_password(Manager
*m
) {
267 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
268 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
269 m
->have_ask_password
= -EINVAL
;
272 static int manager_check_ask_password(Manager
*m
) {
277 if (!m
->ask_password_event_source
) {
278 assert(m
->ask_password_inotify_fd
< 0);
280 mkdir_p_label("/run/systemd/ask-password", 0755);
282 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
283 if (m
->ask_password_inotify_fd
< 0)
284 return log_error_errno(errno
, "inotify_init1() failed: %m");
286 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
287 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
288 manager_close_ask_password(m
);
292 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
293 m
->ask_password_inotify_fd
, EPOLLIN
,
294 manager_dispatch_ask_password_fd
, m
);
296 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
297 manager_close_ask_password(m
);
301 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
303 /* Queries might have been added meanwhile... */
304 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
305 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
308 return m
->have_ask_password
;
311 static int manager_watch_idle_pipe(Manager
*m
) {
316 if (m
->idle_pipe_event_source
)
319 if (m
->idle_pipe
[2] < 0)
322 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
324 return log_error_errno(r
, "Failed to watch idle pipe: %m");
326 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
331 static void manager_close_idle_pipe(Manager
*m
) {
334 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
336 safe_close_pair(m
->idle_pipe
);
337 safe_close_pair(m
->idle_pipe
+ 2);
340 static int manager_setup_time_change(Manager
*m
) {
343 /* We only care for the cancellation event, hence we set the
344 * timeout to the latest possible value. */
345 struct itimerspec its
= {
346 .it_value
.tv_sec
= TIME_T_MAX
,
350 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
355 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
356 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
358 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
359 if (m
->time_change_fd
< 0)
360 return log_error_errno(errno
, "Failed to create timerfd: %m");
362 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
363 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
364 m
->time_change_fd
= safe_close(m
->time_change_fd
);
368 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
370 return log_error_errno(r
, "Failed to create time change event source: %m");
372 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
374 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
379 static int enable_special_signals(Manager
*m
) {
380 _cleanup_close_
int fd
= -1;
387 /* Enable that we get SIGINT on control-alt-del. In containers
388 * this will fail with EPERM (older) or EINVAL (newer), so
390 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
391 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
393 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
395 /* Support systems without virtual console */
397 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
399 /* Enable that we get SIGWINCH on kbrequest */
400 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
401 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
407 static int manager_setup_signals(Manager
*m
) {
408 struct sigaction sa
= {
409 .sa_handler
= SIG_DFL
,
410 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
417 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
419 /* We make liberal use of realtime signals here. On
420 * Linux/glibc we have 30 of them (with the exception of Linux
421 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
424 assert_se(sigemptyset(&mask
) == 0);
425 sigset_add_many(&mask
,
426 SIGCHLD
, /* Child died */
427 SIGTERM
, /* Reexecute daemon */
428 SIGHUP
, /* Reload configuration */
429 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
430 SIGUSR2
, /* systemd: dump status */
431 SIGINT
, /* Kernel sends us this on control-alt-del */
432 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
433 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
435 SIGRTMIN
+0, /* systemd: start default.target */
436 SIGRTMIN
+1, /* systemd: isolate rescue.target */
437 SIGRTMIN
+2, /* systemd: isolate emergency.target */
438 SIGRTMIN
+3, /* systemd: start halt.target */
439 SIGRTMIN
+4, /* systemd: start poweroff.target */
440 SIGRTMIN
+5, /* systemd: start reboot.target */
441 SIGRTMIN
+6, /* systemd: start kexec.target */
443 /* ... space for more special targets ... */
445 SIGRTMIN
+13, /* systemd: Immediate halt */
446 SIGRTMIN
+14, /* systemd: Immediate poweroff */
447 SIGRTMIN
+15, /* systemd: Immediate reboot */
448 SIGRTMIN
+16, /* systemd: Immediate kexec */
450 /* ... space for more immediate system state changes ... */
452 SIGRTMIN
+20, /* systemd: enable status messages */
453 SIGRTMIN
+21, /* systemd: disable status messages */
454 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
455 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
456 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
458 /* .. one free signal here ... */
460 #if !defined(__hppa64__) && !defined(__hppa__)
461 /* Apparently Linux on hppa has fewer RT
462 * signals (SIGRTMAX is SIGRTMIN+25 there),
463 * hence let's not try to make use of them
464 * here. Since these commands are accessible
465 * by different means and only really a safety
466 * net, the missing functionality on hppa
467 * shouldn't matter. */
469 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
470 SIGRTMIN
+27, /* systemd: set log target to console */
471 SIGRTMIN
+28, /* systemd: set log target to kmsg */
472 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
474 /* ... one free signal here SIGRTMIN+30 ... */
477 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
479 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
480 if (m
->signal_fd
< 0)
483 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
487 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
489 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
490 * notify processing can still figure out to which process/service a message belongs, before we reap the
491 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
492 * status information before detecting that there's no process in a cgroup. */
493 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
497 if (MANAGER_IS_SYSTEM(m
))
498 return enable_special_signals(m
);
503 static void manager_clean_environment(Manager
*m
) {
506 /* Let's remove some environment variables that we
507 * need ourselves to communicate with our clients */
521 static int manager_default_environment(Manager
*m
) {
524 if (MANAGER_IS_SYSTEM(m
)) {
525 /* The system manager always starts with a clean
526 * environment for its children. It does not import
527 * the kernel or the parents exported variables.
529 * The initial passed environ is untouched to keep
530 * /proc/self/environ valid; it is used for tagging
531 * the init process inside containers. */
532 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
535 /* Import locale variables LC_*= from configuration */
536 locale_setup(&m
->environment
);
538 /* The user manager passes its own environment
539 * along to its children. */
540 m
->environment
= strv_copy(environ
);
546 manager_clean_environment(m
);
547 strv_sort(m
->environment
);
553 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
558 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
560 m
= new0(Manager
, 1);
564 m
->unit_file_scope
= scope
;
565 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
566 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
567 m
->default_tasks_accounting
= true;
568 m
->default_tasks_max
= UINT64_C(512);
571 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
572 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
575 /* Prepare log fields we can use for structured logging */
576 if (MANAGER_IS_SYSTEM(m
)) {
577 m
->unit_log_field
= "UNIT=";
578 m
->unit_log_format_string
= "UNIT=%s";
580 m
->unit_log_field
= "USER_UNIT=";
581 m
->unit_log_format_string
= "USER_UNIT=%s";
584 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
586 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
587 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->kdbus_fd
= m
->cgroup_inotify_fd
=
588 m
->ask_password_inotify_fd
= -1;
590 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
592 m
->have_ask_password
= -EINVAL
; /* we don't know */
595 m
->test_run
= test_run
;
597 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
598 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
600 r
= manager_default_environment(m
);
604 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
608 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
612 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
616 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
620 r
= sd_event_default(&m
->event
);
624 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
628 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
632 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
636 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
638 r
= manager_setup_signals(m
);
642 r
= manager_setup_cgroup(m
);
646 r
= manager_setup_time_change(m
);
650 m
->udev
= udev_new();
656 /* Note that we set up neither kdbus, nor the notify fd
657 * here. We do that after deserialization, since they might
658 * have gotten serialized across the reexec. */
660 m
->taint_usr
= dir_is_empty("/usr") > 0;
670 static int manager_setup_notify(Manager
*m
) {
676 if (m
->notify_fd
< 0) {
677 _cleanup_close_
int fd
= -1;
678 union sockaddr_union sa
= {
679 .sa
.sa_family
= AF_UNIX
,
681 static const int one
= 1;
684 /* First free all secondary fields */
685 m
->notify_socket
= mfree(m
->notify_socket
);
686 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
688 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
690 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
692 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
694 e
= manager_get_runtime_prefix(m
);
696 log_error("Failed to determine runtime prefix.");
700 m
->notify_socket
= strappend(e
, "/systemd/notify");
701 if (!m
->notify_socket
)
704 (void) mkdir_parents_label(m
->notify_socket
, 0755);
705 (void) unlink(m
->notify_socket
);
707 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
708 r
= bind(fd
, &sa
.sa
, offsetof(struct sockaddr_un
, sun_path
) + strlen(sa
.un
.sun_path
));
710 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
712 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
714 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
719 log_debug("Using notification socket %s", m
->notify_socket
);
722 if (!m
->notify_event_source
) {
723 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
725 return log_error_errno(r
, "Failed to allocate notify event source: %m");
727 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
728 * service an exit message belongs. */
729 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
731 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
733 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
739 static int manager_setup_cgroups_agent(Manager
*m
) {
741 static const union sockaddr_union sa
= {
742 .un
.sun_family
= AF_UNIX
,
743 .un
.sun_path
= "/run/systemd/cgroups-agent",
747 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
748 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
749 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
750 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
751 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
752 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
753 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
754 * we thus won't lose messages.
756 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
757 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
758 * bus for these messages. */
763 if (!MANAGER_IS_SYSTEM(m
))
766 if (cg_unified() > 0) /* We don't need this anymore on the unified hierarchy */
769 if (m
->cgroups_agent_fd
< 0) {
770 _cleanup_close_
int fd
= -1;
772 /* First free all secondary fields */
773 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
775 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
777 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
779 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
781 (void) unlink(sa
.un
.sun_path
);
783 /* Only allow root to connect to this socket */
785 r
= bind(fd
, &sa
.sa
, offsetof(struct sockaddr_un
, sun_path
) + strlen(sa
.un
.sun_path
));
787 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
789 m
->cgroups_agent_fd
= fd
;
793 if (!m
->cgroups_agent_event_source
) {
794 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
796 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
798 /* Process cgroups notifications early, but after having processed service notification messages or
799 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
800 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
801 * cgroup inotify for the unified cgroup stuff. */
802 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
804 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
806 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
812 static int manager_setup_kdbus(Manager
*m
) {
813 _cleanup_free_
char *p
= NULL
;
817 if (m
->test_run
|| m
->kdbus_fd
>= 0)
819 if (!is_kdbus_available())
820 return -ESOCKTNOSUPPORT
;
822 m
->kdbus_fd
= bus_kernel_create_bus(
823 MANAGER_IS_SYSTEM(m
) ? "system" : "user",
824 MANAGER_IS_SYSTEM(m
), &p
);
827 return log_debug_errno(m
->kdbus_fd
, "Failed to set up kdbus: %m");
829 log_debug("Successfully set up kdbus on %s", p
);
834 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
835 bool try_bus_connect
;
845 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
847 /* Try to connect to the buses, if possible. */
848 return bus_init(m
, try_bus_connect
);
851 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
857 while ((u
= m
->cleanup_queue
)) {
858 assert(u
->in_cleanup_queue
);
868 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
869 GC_OFFSET_UNSURE
, /* No clue */
870 GC_OFFSET_GOOD
, /* We still need this unit */
871 GC_OFFSET_BAD
, /* We don't need this unit anymore */
875 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
882 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
883 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
884 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
887 if (u
->in_cleanup_queue
)
890 if (unit_check_gc(u
))
893 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
897 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
898 unit_gc_sweep(other
, gc_marker
);
900 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
903 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
910 /* We were unable to find anything out about this entry, so
911 * let's investigate it later */
912 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
913 unit_add_to_gc_queue(u
);
917 /* We definitely know that this one is not useful anymore, so
918 * let's mark it for deletion */
919 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
920 unit_add_to_cleanup_queue(u
);
924 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
927 static unsigned manager_dispatch_gc_queue(Manager
*m
) {
934 /* log_debug("Running GC..."); */
936 m
->gc_marker
+= _GC_OFFSET_MAX
;
937 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
940 gc_marker
= m
->gc_marker
;
942 while ((u
= m
->gc_queue
)) {
943 assert(u
->in_gc_queue
);
945 unit_gc_sweep(u
, gc_marker
);
947 LIST_REMOVE(gc_queue
, m
->gc_queue
, u
);
948 u
->in_gc_queue
= false;
952 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
953 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
955 log_unit_debug(u
, "Collecting.");
956 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
957 unit_add_to_cleanup_queue(u
);
961 m
->n_in_gc_queue
= 0;
966 static void manager_clear_jobs_and_units(Manager
*m
) {
971 while ((u
= hashmap_first(m
->units
)))
974 manager_dispatch_cleanup_queue(m
);
976 assert(!m
->load_queue
);
977 assert(!m
->run_queue
);
978 assert(!m
->dbus_unit_queue
);
979 assert(!m
->dbus_job_queue
);
980 assert(!m
->cleanup_queue
);
981 assert(!m
->gc_queue
);
983 assert(hashmap_isempty(m
->jobs
));
984 assert(hashmap_isempty(m
->units
));
987 m
->n_running_jobs
= 0;
990 Manager
* manager_free(Manager
*m
) {
997 manager_clear_jobs_and_units(m
);
999 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1000 if (unit_vtable
[c
]->shutdown
)
1001 unit_vtable
[c
]->shutdown(m
);
1003 /* If we reexecute ourselves, we keep the root cgroup
1005 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1007 lookup_paths_flush_generator(&m
->lookup_paths
);
1011 hashmap_free(m
->units
);
1012 hashmap_free(m
->jobs
);
1013 hashmap_free(m
->watch_pids1
);
1014 hashmap_free(m
->watch_pids2
);
1015 hashmap_free(m
->watch_bus
);
1017 set_free(m
->startup_units
);
1018 set_free(m
->failed_units
);
1020 sd_event_source_unref(m
->signal_event_source
);
1021 sd_event_source_unref(m
->notify_event_source
);
1022 sd_event_source_unref(m
->cgroups_agent_event_source
);
1023 sd_event_source_unref(m
->time_change_event_source
);
1024 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1025 sd_event_source_unref(m
->run_queue_event_source
);
1027 safe_close(m
->signal_fd
);
1028 safe_close(m
->notify_fd
);
1029 safe_close(m
->cgroups_agent_fd
);
1030 safe_close(m
->time_change_fd
);
1031 safe_close(m
->kdbus_fd
);
1033 manager_close_ask_password(m
);
1035 manager_close_idle_pipe(m
);
1037 udev_unref(m
->udev
);
1038 sd_event_unref(m
->event
);
1040 free(m
->notify_socket
);
1042 lookup_paths_free(&m
->lookup_paths
);
1043 strv_free(m
->environment
);
1045 hashmap_free(m
->cgroup_unit
);
1046 set_free_free(m
->unit_path_cache
);
1048 free(m
->switch_root
);
1049 free(m
->switch_root_init
);
1051 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1052 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1054 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1055 hashmap_free(m
->units_requiring_mounts_for
);
1061 void manager_enumerate(Manager
*m
) {
1066 /* Let's ask every type to load all units from disk/kernel
1067 * that it might know */
1068 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1069 if (!unit_type_supported(c
)) {
1070 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1074 if (!unit_vtable
[c
]->enumerate
)
1077 unit_vtable
[c
]->enumerate(m
);
1080 manager_dispatch_load_queue(m
);
1083 static void manager_coldplug(Manager
*m
) {
1091 /* Then, let's set up their initial state. */
1092 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1094 /* ignore aliases */
1098 r
= unit_coldplug(u
);
1100 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1104 static void manager_build_unit_path_cache(Manager
*m
) {
1110 set_free_free(m
->unit_path_cache
);
1112 m
->unit_path_cache
= set_new(&string_hash_ops
);
1113 if (!m
->unit_path_cache
) {
1118 /* This simply builds a list of files we know exist, so that
1119 * we don't always have to go to disk */
1121 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1122 _cleanup_closedir_
DIR *d
= NULL
;
1127 if (errno
!= ENOENT
)
1128 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1132 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1135 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
, NULL
);
1141 r
= set_consume(m
->unit_path_cache
, p
);
1150 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1151 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1154 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1160 HASHMAP_FOREACH(u
, m
->units
, i
) {
1162 if (fdset_size(fds
) <= 0)
1165 if (!UNIT_VTABLE(u
)->distribute_fds
)
1168 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1172 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1177 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1181 /* Make sure the transient directory always exists, so that it remains in the search path */
1182 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1186 dual_timestamp_get(&m
->generators_start_timestamp
);
1187 r
= manager_run_generators(m
);
1188 dual_timestamp_get(&m
->generators_finish_timestamp
);
1192 lookup_paths_reduce(&m
->lookup_paths
);
1193 manager_build_unit_path_cache(m
);
1195 /* If we will deserialize make sure that during enumeration
1196 * this is already known, so we increase the counter here
1201 /* First, enumerate what we can from all config files */
1202 dual_timestamp_get(&m
->units_load_start_timestamp
);
1203 manager_enumerate(m
);
1204 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1206 /* Second, deserialize if there is something to deserialize */
1208 r
= manager_deserialize(m
, serialization
, fds
);
1210 /* Any fds left? Find some unit which wants them. This is
1211 * useful to allow container managers to pass some file
1212 * descriptors to us pre-initialized. This enables
1213 * socket-based activation of entire containers. */
1214 manager_distribute_fds(m
, fds
);
1216 /* We might have deserialized the notify fd, but if we didn't
1217 * then let's create the bus now */
1218 q
= manager_setup_notify(m
);
1219 if (q
< 0 && r
== 0)
1222 q
= manager_setup_cgroups_agent(m
);
1223 if (q
< 0 && r
== 0)
1226 /* We might have deserialized the kdbus control fd, but if we
1227 * didn't, then let's create the bus now. */
1228 manager_setup_kdbus(m
);
1229 manager_connect_bus(m
, !!serialization
);
1230 bus_track_coldplug(m
, &m
->subscribed
, &m
->deserialized_subscribed
);
1232 /* Third, fire things up! */
1233 manager_coldplug(m
);
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);
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
) {
1613 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1614 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
) {
1727 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1729 unit_unwatch_pid(u
, si
->si_pid
);
1731 if (UNIT_VTABLE(u
)->sigchld_event
)
1732 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1735 static int manager_dispatch_sigchld(Manager
*m
) {
1741 /* First we call waitd() for a PID and do not reap the
1742 * zombie. That way we can still access /proc/$PID for
1743 * it while it is a zombie. */
1744 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1746 if (errno
== ECHILD
)
1758 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1759 _cleanup_free_
char *name
= NULL
;
1762 get_process_comm(si
.si_pid
, &name
);
1764 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1765 si
.si_pid
, strna(name
),
1766 sigchld_code_to_string(si
.si_code
),
1768 strna(si
.si_code
== CLD_EXITED
1769 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1770 : signal_to_string(si
.si_status
)));
1772 /* And now figure out the unit this belongs
1773 * to, it might be multiple... */
1774 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1776 invoke_sigchld_event(m
, u1
, &si
);
1777 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1779 invoke_sigchld_event(m
, u2
, &si
);
1780 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1781 if (u3
&& u3
!= u2
&& u3
!= u1
)
1782 invoke_sigchld_event(m
, u3
, &si
);
1785 /* And now, we actually reap the zombie. */
1786 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1797 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1798 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1801 log_debug("Activating special unit %s", name
);
1803 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1805 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1810 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1811 Manager
*m
= userdata
;
1813 struct signalfd_siginfo sfsi
;
1814 bool sigchld
= false;
1818 assert(m
->signal_fd
== fd
);
1820 if (revents
!= EPOLLIN
) {
1821 log_warning("Got unexpected events from signal file descriptor.");
1826 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1827 if (n
!= sizeof(sfsi
)) {
1832 if (errno
== EINTR
|| errno
== EAGAIN
)
1838 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1839 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1840 ? LOG_DEBUG
: LOG_INFO
,
1843 switch (sfsi
.ssi_signo
) {
1850 if (MANAGER_IS_SYSTEM(m
)) {
1851 /* This is for compatibility with the
1852 * original sysvinit */
1853 m
->exit_code
= MANAGER_REEXECUTE
;
1860 if (MANAGER_IS_SYSTEM(m
)) {
1862 /* If the user presses C-A-D more than
1863 * 7 times within 2s, we reboot
1866 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
))
1867 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1869 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1870 status_printf(NULL
, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1871 m
->exit_code
= MANAGER_REBOOT
;
1877 /* Run the exit target if there is one, if not, just exit. */
1878 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
1879 m
->exit_code
= MANAGER_EXIT
;
1886 if (MANAGER_IS_SYSTEM(m
))
1887 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
1889 /* This is a nop on non-init */
1893 if (MANAGER_IS_SYSTEM(m
))
1894 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
1896 /* This is a nop on non-init */
1902 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1904 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
1905 log_info("Trying to reconnect to bus...");
1909 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
1910 log_info("Loading D-Bus service...");
1911 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
1918 _cleanup_free_
char *dump
= NULL
;
1919 _cleanup_fclose_
FILE *f
= NULL
;
1922 f
= open_memstream(&dump
, &size
);
1924 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
1928 manager_dump_units(m
, f
, "\t");
1929 manager_dump_jobs(m
, f
, "\t");
1931 r
= fflush_and_check(f
);
1933 log_warning_errno(r
, "Failed to write status stream: %m");
1937 log_dump(LOG_INFO
, dump
);
1942 m
->exit_code
= MANAGER_RELOAD
;
1947 /* Starting SIGRTMIN+0 */
1948 static const char * const target_table
[] = {
1949 [0] = SPECIAL_DEFAULT_TARGET
,
1950 [1] = SPECIAL_RESCUE_TARGET
,
1951 [2] = SPECIAL_EMERGENCY_TARGET
,
1952 [3] = SPECIAL_HALT_TARGET
,
1953 [4] = SPECIAL_POWEROFF_TARGET
,
1954 [5] = SPECIAL_REBOOT_TARGET
,
1955 [6] = SPECIAL_KEXEC_TARGET
1958 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
1959 static const ManagerExitCode code_table
[] = {
1961 [1] = MANAGER_POWEROFF
,
1962 [2] = MANAGER_REBOOT
,
1966 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
1967 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
1968 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
1969 manager_start_target(m
, target_table
[idx
],
1970 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
1974 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
1975 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
1976 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
1980 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
1983 manager_set_show_status(m
, SHOW_STATUS_YES
);
1987 manager_set_show_status(m
, SHOW_STATUS_NO
);
1991 log_set_max_level(LOG_DEBUG
);
1992 log_info("Setting log level to debug.");
1996 log_set_max_level(LOG_INFO
);
1997 log_info("Setting log level to info.");
2001 if (MANAGER_IS_USER(m
)) {
2002 m
->exit_code
= MANAGER_EXIT
;
2006 /* This is a nop on init */
2010 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2011 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2012 log_notice("Setting log target to journal-or-kmsg.");
2016 log_set_target(LOG_TARGET_CONSOLE
);
2017 log_notice("Setting log target to console.");
2021 log_set_target(LOG_TARGET_KMSG
);
2022 log_notice("Setting log target to kmsg.");
2026 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2033 manager_dispatch_sigchld(m
);
2038 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2039 Manager
*m
= userdata
;
2044 assert(m
->time_change_fd
== fd
);
2046 log_struct(LOG_INFO
,
2047 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE
),
2048 LOG_MESSAGE("Time has been changed"),
2051 /* Restart the watch */
2052 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2053 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2055 manager_setup_time_change(m
);
2057 HASHMAP_FOREACH(u
, m
->units
, i
)
2058 if (UNIT_VTABLE(u
)->time_change
)
2059 UNIT_VTABLE(u
)->time_change(u
);
2064 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2065 Manager
*m
= userdata
;
2068 assert(m
->idle_pipe
[2] == fd
);
2070 m
->no_console_output
= m
->n_on_console
> 0;
2072 manager_close_idle_pipe(m
);
2077 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2078 Manager
*m
= userdata
;
2085 manager_print_jobs_in_progress(m
);
2087 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2088 r
= sd_event_source_set_time(source
, next
);
2092 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2095 int manager_loop(Manager
*m
) {
2098 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2101 m
->exit_code
= MANAGER_OK
;
2103 /* Release the path cache */
2104 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2106 manager_check_finished(m
);
2108 /* There might still be some zombies hanging around from
2109 * before we were exec()'ed. Let's reap them. */
2110 r
= manager_dispatch_sigchld(m
);
2114 while (m
->exit_code
== MANAGER_OK
) {
2117 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2120 if (!ratelimit_test(&rl
)) {
2121 /* Yay, something is going seriously wrong, pause a little */
2122 log_warning("Looping too fast. Throttling execution a little.");
2126 if (manager_dispatch_load_queue(m
) > 0)
2129 if (manager_dispatch_gc_queue(m
) > 0)
2132 if (manager_dispatch_cleanup_queue(m
) > 0)
2135 if (manager_dispatch_cgroup_queue(m
) > 0)
2138 if (manager_dispatch_dbus_queue(m
) > 0)
2141 /* Sleep for half the watchdog time */
2142 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2143 wait_usec
= m
->runtime_watchdog
/ 2;
2147 wait_usec
= USEC_INFINITY
;
2149 r
= sd_event_run(m
->event
, wait_usec
);
2151 return log_error_errno(r
, "Failed to run event loop: %m");
2154 return m
->exit_code
;
2157 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2158 _cleanup_free_
char *n
= NULL
;
2166 r
= unit_name_from_dbus_path(s
, &n
);
2170 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2179 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2189 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2193 r
= safe_atou(p
, &id
);
2197 j
= manager_get_job(m
, id
);
2206 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2209 _cleanup_free_
char *p
= NULL
;
2213 if (!MANAGER_IS_SYSTEM(m
))
2216 audit_fd
= get_audit_fd();
2220 /* Don't generate audit events if the service was already
2221 * started and we're just deserializing */
2222 if (MANAGER_IS_RELOADING(m
))
2225 if (u
->type
!= UNIT_SERVICE
)
2228 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2230 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2234 msg
= strjoina("unit=", p
);
2235 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2237 /* We aren't allowed to send audit messages?
2238 * Then let's not retry again. */
2241 log_warning_errno(errno
, "Failed to send audit message: %m");
2247 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2248 union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2251 _cleanup_free_
char *message
= NULL
;
2252 _cleanup_close_
int fd
= -1;
2254 /* Don't generate plymouth events if the service was already
2255 * started and we're just deserializing */
2256 if (MANAGER_IS_RELOADING(m
))
2259 if (!MANAGER_IS_SYSTEM(m
))
2262 if (detect_container() > 0)
2265 if (u
->type
!= UNIT_SERVICE
&&
2266 u
->type
!= UNIT_MOUNT
&&
2267 u
->type
!= UNIT_SWAP
)
2270 /* We set SOCK_NONBLOCK here so that we rather drop the
2271 * message then wait for plymouth */
2272 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2274 log_error_errno(errno
, "socket() failed: %m");
2278 if (connect(fd
, &sa
.sa
, offsetof(struct sockaddr_un
, sun_path
) + 1 + strlen(sa
.un
.sun_path
+1)) < 0) {
2280 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2281 log_error_errno(errno
, "connect() failed: %m");
2285 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2291 if (write(fd
, message
, n
+ 1) != n
+ 1)
2292 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2293 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2296 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2303 path
= MANAGER_IS_SYSTEM(m
) ? "/run/systemd" : "/tmp";
2304 fd
= open_tmpfile_unlinkable(path
, O_RDWR
|O_CLOEXEC
);
2308 log_debug("Serializing state to %s", path
);
2310 f
= fdopen(fd
, "w+");
2321 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2334 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2335 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2336 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2337 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2339 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2340 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2341 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2342 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2345 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2346 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2347 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2348 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2349 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2350 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2351 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2352 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2355 if (!switching_root
) {
2356 STRV_FOREACH(e
, m
->environment
) {
2357 _cleanup_free_
char *ce
;
2363 fprintf(f
, "env=%s\n", *e
);
2367 if (m
->notify_fd
>= 0) {
2370 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2374 fprintf(f
, "notify-fd=%i\n", copy
);
2375 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2378 if (m
->cgroups_agent_fd
>= 0) {
2381 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2385 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2388 if (m
->kdbus_fd
>= 0) {
2391 copy
= fdset_put_dup(fds
, m
->kdbus_fd
);
2395 fprintf(f
, "kdbus-fd=%i\n", copy
);
2398 bus_track_serialize(m
->subscribed
, f
);
2402 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2410 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2417 assert(m
->n_reloading
> 0);
2423 r
= bus_fdset_add_all(m
, fds
);
2430 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2436 log_debug("Deserializing state...");
2441 char line
[LINE_MAX
], *l
;
2443 if (!fgets(line
, sizeof(line
), f
)) {
2458 if (startswith(l
, "current-job-id=")) {
2461 if (safe_atou32(l
+15, &id
) < 0)
2462 log_debug("Failed to parse current job id value %s", l
+15);
2464 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2466 } else if (startswith(l
, "n-installed-jobs=")) {
2469 if (safe_atou32(l
+17, &n
) < 0)
2470 log_debug("Failed to parse installed jobs counter %s", l
+17);
2472 m
->n_installed_jobs
+= n
;
2474 } else if (startswith(l
, "n-failed-jobs=")) {
2477 if (safe_atou32(l
+14, &n
) < 0)
2478 log_debug("Failed to parse failed jobs counter %s", l
+14);
2480 m
->n_failed_jobs
+= n
;
2482 } else if (startswith(l
, "taint-usr=")) {
2485 b
= parse_boolean(l
+10);
2487 log_debug("Failed to parse taint /usr flag %s", l
+10);
2489 m
->taint_usr
= m
->taint_usr
|| b
;
2491 } else if (startswith(l
, "firmware-timestamp="))
2492 dual_timestamp_deserialize(l
+19, &m
->firmware_timestamp
);
2493 else if (startswith(l
, "loader-timestamp="))
2494 dual_timestamp_deserialize(l
+17, &m
->loader_timestamp
);
2495 else if (startswith(l
, "kernel-timestamp="))
2496 dual_timestamp_deserialize(l
+17, &m
->kernel_timestamp
);
2497 else if (startswith(l
, "initrd-timestamp="))
2498 dual_timestamp_deserialize(l
+17, &m
->initrd_timestamp
);
2499 else if (startswith(l
, "userspace-timestamp="))
2500 dual_timestamp_deserialize(l
+20, &m
->userspace_timestamp
);
2501 else if (startswith(l
, "finish-timestamp="))
2502 dual_timestamp_deserialize(l
+17, &m
->finish_timestamp
);
2503 else if (startswith(l
, "security-start-timestamp="))
2504 dual_timestamp_deserialize(l
+25, &m
->security_start_timestamp
);
2505 else if (startswith(l
, "security-finish-timestamp="))
2506 dual_timestamp_deserialize(l
+26, &m
->security_finish_timestamp
);
2507 else if (startswith(l
, "generators-start-timestamp="))
2508 dual_timestamp_deserialize(l
+27, &m
->generators_start_timestamp
);
2509 else if (startswith(l
, "generators-finish-timestamp="))
2510 dual_timestamp_deserialize(l
+28, &m
->generators_finish_timestamp
);
2511 else if (startswith(l
, "units-load-start-timestamp="))
2512 dual_timestamp_deserialize(l
+27, &m
->units_load_start_timestamp
);
2513 else if (startswith(l
, "units-load-finish-timestamp="))
2514 dual_timestamp_deserialize(l
+28, &m
->units_load_finish_timestamp
);
2515 else if (startswith(l
, "env=")) {
2516 _cleanup_free_
char *uce
= NULL
;
2519 r
= cunescape(l
+ 4, UNESCAPE_RELAX
, &uce
);
2523 e
= strv_env_set(m
->environment
, uce
);
2529 strv_free(m
->environment
);
2532 } else if (startswith(l
, "notify-fd=")) {
2535 if (safe_atoi(l
+ 10, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2536 log_debug("Failed to parse notify fd: %s", l
+ 10);
2538 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2539 safe_close(m
->notify_fd
);
2540 m
->notify_fd
= fdset_remove(fds
, fd
);
2543 } else if (startswith(l
, "notify-socket=")) {
2552 free(m
->notify_socket
);
2553 m
->notify_socket
= n
;
2555 } else if (startswith(l
, "cgroups-agent-fd=")) {
2558 if (safe_atoi(l
+ 17, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2559 log_debug("Failed to parse cgroups agent fd: %s", l
+ 10);
2561 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2562 safe_close(m
->cgroups_agent_fd
);
2563 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2566 } else if (startswith(l
, "kdbus-fd=")) {
2569 if (safe_atoi(l
+ 9, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2570 log_debug("Failed to parse kdbus fd: %s", l
+ 9);
2572 safe_close(m
->kdbus_fd
);
2573 m
->kdbus_fd
= fdset_remove(fds
, fd
);
2579 k
= bus_track_deserialize_item(&m
->deserialized_subscribed
, l
);
2581 log_debug_errno(k
, "Failed to deserialize bus tracker object: %m");
2583 log_debug("Unknown serialization item '%s'", l
);
2589 char name
[UNIT_NAME_MAX
+2];
2592 if (!fgets(name
, sizeof(name
), f
)) {
2603 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2607 r
= unit_deserialize(u
, f
, fds
);
2616 assert(m
->n_reloading
> 0);
2622 int manager_reload(Manager
*m
) {
2624 _cleanup_fclose_
FILE *f
= NULL
;
2625 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2629 r
= manager_open_serialization(m
, &f
);
2634 bus_manager_send_reloading(m
, true);
2642 r
= manager_serialize(m
, f
, fds
, false);
2648 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2653 /* From here on there is no way back. */
2654 manager_clear_jobs_and_units(m
);
2655 lookup_paths_flush_generator(&m
->lookup_paths
);
2656 lookup_paths_free(&m
->lookup_paths
);
2658 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2659 if (q
< 0 && r
>= 0)
2662 /* Find new unit paths */
2663 q
= manager_run_generators(m
);
2664 if (q
< 0 && r
>= 0)
2667 lookup_paths_reduce(&m
->lookup_paths
);
2668 manager_build_unit_path_cache(m
);
2670 /* First, enumerate what we can from all config files */
2671 manager_enumerate(m
);
2673 /* Second, deserialize our stored data */
2674 q
= manager_deserialize(m
, f
, fds
);
2675 if (q
< 0 && r
>= 0)
2681 /* Re-register notify_fd as event source */
2682 q
= manager_setup_notify(m
);
2683 if (q
< 0 && r
>= 0)
2686 q
= manager_setup_cgroups_agent(m
);
2687 if (q
< 0 && r
>= 0)
2690 /* Third, fire things up! */
2691 manager_coldplug(m
);
2693 /* Sync current state of bus names with our set of listening units */
2695 manager_sync_bus_names(m
, m
->api_bus
);
2697 assert(m
->n_reloading
> 0);
2700 m
->send_reloading_done
= true;
2705 void manager_reset_failed(Manager
*m
) {
2711 HASHMAP_FOREACH(u
, m
->units
, i
)
2712 unit_reset_failed(u
);
2715 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2721 /* Returns true if the unit is inactive or going down */
2722 u
= manager_get_unit(m
, name
);
2726 return unit_inactive_or_pending(u
);
2729 static void manager_notify_finished(Manager
*m
) {
2730 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2731 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2736 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2738 /* Note that m->kernel_usec.monotonic is always at 0,
2739 * and m->firmware_usec.monotonic and
2740 * m->loader_usec.monotonic should be considered
2741 * negative values. */
2743 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2744 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2745 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2746 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2748 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2750 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2751 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2753 log_struct(LOG_INFO
,
2754 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2755 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2756 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2757 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2758 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2759 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2760 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2761 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2762 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2765 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2768 log_struct(LOG_INFO
,
2769 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2770 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2771 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2772 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2773 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2774 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2775 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2779 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2780 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2782 log_struct(LOG_INFO
,
2783 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2784 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2785 LOG_MESSAGE("Startup finished in %s.",
2786 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2790 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2794 "STATUS=Startup finished in %s.",
2795 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2798 void manager_check_finished(Manager
*m
) {
2801 if (MANAGER_IS_RELOADING(m
))
2804 /* Verify that we are actually running currently. Initially
2805 * the exit code is set to invalid, and during operation it is
2806 * then set to MANAGER_OK */
2807 if (m
->exit_code
!= MANAGER_OK
)
2810 if (hashmap_size(m
->jobs
) > 0) {
2811 if (m
->jobs_in_progress_event_source
)
2812 /* Ignore any failure, this is only for feedback */
2813 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2818 manager_flip_auto_status(m
, false);
2820 /* Notify Type=idle units that we are done now */
2821 manager_close_idle_pipe(m
);
2823 /* Turn off confirm spawn now */
2824 m
->confirm_spawn
= false;
2826 /* No need to update ask password status when we're going non-interactive */
2827 manager_close_ask_password(m
);
2829 /* This is no longer the first boot */
2830 manager_set_first_boot(m
, false);
2832 if (dual_timestamp_is_set(&m
->finish_timestamp
))
2835 dual_timestamp_get(&m
->finish_timestamp
);
2837 manager_notify_finished(m
);
2839 manager_invalidate_startup_units(m
);
2842 static int manager_run_generators(Manager
*m
) {
2843 _cleanup_strv_free_
char **paths
= NULL
;
2844 const char *argv
[5];
2853 paths
= generator_binary_paths(m
->unit_file_scope
);
2857 /* Optimize by skipping the whole process by not creating output directories
2858 * if no generators are found. */
2859 STRV_FOREACH(path
, paths
) {
2860 if (access(*path
, F_OK
) >= 0)
2862 if (errno
!= ENOENT
)
2863 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
2869 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
2873 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
2874 argv
[1] = m
->lookup_paths
.generator
;
2875 argv
[2] = m
->lookup_paths
.generator_early
;
2876 argv
[3] = m
->lookup_paths
.generator_late
;
2879 RUN_WITH_UMASK(0022)
2880 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, (char**) argv
);
2883 lookup_paths_trim_generator(&m
->lookup_paths
);
2887 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
2888 char **a
= NULL
, **b
= NULL
, **l
;
2893 if (!strv_isempty(minus
)) {
2894 a
= strv_env_delete(l
, 1, minus
);
2901 if (!strv_isempty(plus
)) {
2902 b
= strv_env_merge(2, l
, plus
);
2911 if (m
->environment
!= l
)
2912 strv_free(m
->environment
);
2919 manager_clean_environment(m
);
2920 strv_sort(m
->environment
);
2925 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
2930 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
2931 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
2933 if (!default_rlimit
[i
])
2936 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
2944 void manager_recheck_journal(Manager
*m
) {
2949 if (!MANAGER_IS_SYSTEM(m
))
2952 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
2953 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
2954 log_close_journal();
2958 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
2959 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
2960 log_close_journal();
2964 /* Hmm, OK, so the socket is fully up and the service is up
2965 * too, then let's make use of the thing. */
2969 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
2971 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
2973 if (!MANAGER_IS_SYSTEM(m
))
2976 if (m
->show_status
!= mode
)
2977 log_debug("%s showing of status.",
2978 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
2979 m
->show_status
= mode
;
2982 (void) touch("/run/systemd/show-status");
2984 (void) unlink("/run/systemd/show-status");
2987 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
2990 if (!MANAGER_IS_SYSTEM(m
))
2993 if (m
->no_console_output
)
2996 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
2999 /* If we cannot find out the status properly, just proceed. */
3000 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3003 if (m
->show_status
> 0)
3009 void manager_set_first_boot(Manager
*m
, bool b
) {
3012 if (!MANAGER_IS_SYSTEM(m
))
3015 if (m
->first_boot
!= (int) b
) {
3017 (void) touch("/run/systemd/first-boot");
3019 (void) unlink("/run/systemd/first-boot");
3025 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3028 /* If m is NULL, assume we're after shutdown and let the messages through. */
3030 if (m
&& !manager_get_show_status(m
, type
))
3033 /* XXX We should totally drop the check for ephemeral here
3034 * and thus effectively make 'Type=idle' pointless. */
3035 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3038 va_start(ap
, format
);
3039 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3043 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3044 char p
[strlen(path
)+1];
3050 path_kill_slashes(p
);
3052 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3055 const char *manager_get_runtime_prefix(Manager
*m
) {
3058 return MANAGER_IS_SYSTEM(m
) ?
3060 getenv("XDG_RUNTIME_DIR");
3063 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3068 assert(u
->manager
== m
);
3070 size
= set_size(m
->failed_units
);
3073 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3077 if (set_put(m
->failed_units
, u
) < 0)
3080 (void) set_remove(m
->failed_units
, u
);
3082 if (set_size(m
->failed_units
) != size
)
3083 bus_manager_send_change_signal(m
);
3088 ManagerState
manager_state(Manager
*m
) {
3093 /* Did we ever finish booting? If not then we are still starting up */
3094 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3096 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3097 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3098 return MANAGER_INITIALIZING
;
3100 return MANAGER_STARTING
;
3103 /* Is the special shutdown target queued? If so, we are in shutdown state */
3104 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3105 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3106 return MANAGER_STOPPING
;
3108 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3109 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3110 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3111 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3112 return MANAGER_MAINTENANCE
;
3114 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3115 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3116 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3117 return MANAGER_MAINTENANCE
;
3119 /* Are there any failed units? If so, we are in degraded mode */
3120 if (set_size(m
->failed_units
) > 0)
3121 return MANAGER_DEGRADED
;
3123 return MANAGER_RUNNING
;
3126 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3127 [MANAGER_INITIALIZING
] = "initializing",
3128 [MANAGER_STARTING
] = "starting",
3129 [MANAGER_RUNNING
] = "running",
3130 [MANAGER_DEGRADED
] = "degraded",
3131 [MANAGER_MAINTENANCE
] = "maintenance",
3132 [MANAGER_STOPPING
] = "stopping",
3135 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);