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"
48 #include "clean-ipc.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
53 #include "dirent-util.h"
56 #include "exit-status.h"
62 #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"
85 #include "user-util.h"
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
98 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
105 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
106 static int manager_run_generators(Manager
*m
);
108 static void manager_watch_jobs_in_progress(Manager
*m
) {
114 if (m
->jobs_in_progress_event_source
)
117 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
118 r
= sd_event_add_time(
120 &m
->jobs_in_progress_event_source
,
123 manager_dispatch_jobs_in_progress
, m
);
127 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
130 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
132 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
135 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
136 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
140 p
= mempset(p
, ' ', pos
-2);
141 if (log_get_show_color())
142 p
= stpcpy(p
, ANSI_RED
);
146 if (pos
> 0 && pos
<= width
) {
147 if (log_get_show_color())
148 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
152 if (log_get_show_color())
153 p
= stpcpy(p
, ANSI_NORMAL
);
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_RED
);
160 p
= mempset(p
, ' ', width
-1-pos
);
161 if (log_get_show_color())
162 strcpy(p
, ANSI_NORMAL
);
166 void manager_flip_auto_status(Manager
*m
, bool enable
) {
170 if (m
->show_status
== SHOW_STATUS_AUTO
)
171 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
173 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
174 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
178 static void manager_print_jobs_in_progress(Manager
*m
) {
179 _cleanup_free_
char *job_of_n
= NULL
;
182 unsigned counter
= 0, print_nr
;
183 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
185 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
189 assert(m
->n_running_jobs
> 0);
191 manager_flip_auto_status(m
, true);
193 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
195 HASHMAP_FOREACH(j
, m
->jobs
, i
)
196 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
199 /* m->n_running_jobs must be consistent with the contents of m->jobs,
200 * so the above loop must have succeeded in finding j. */
201 assert(counter
== print_nr
+ 1);
204 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
206 cylon_pos
= 14 - cylon_pos
;
207 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
209 m
->jobs_in_progress_iteration
++;
211 if (m
->n_running_jobs
> 1) {
212 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
216 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
217 if (job_get_timeout(j
, &x
) > 0)
218 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
220 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
221 "%sA %s job is running for %s (%s / %s)",
223 job_type_to_string(j
->type
),
224 unit_description(j
->unit
),
228 static int have_ask_password(void) {
229 _cleanup_closedir_
DIR *dir
;
231 dir
= opendir("/run/systemd/ask-password");
244 if (!de
&& errno
> 0)
249 if (startswith(de
->d_name
, "ask."))
254 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
255 int fd
, uint32_t revents
, void *userdata
) {
256 Manager
*m
= userdata
;
262 m
->have_ask_password
= have_ask_password();
263 if (m
->have_ask_password
< 0)
264 /* Log error but continue. Negative have_ask_password
265 * is treated as unknown status. */
266 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
271 static void manager_close_ask_password(Manager
*m
) {
274 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
275 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
276 m
->have_ask_password
= -EINVAL
;
279 static int manager_check_ask_password(Manager
*m
) {
284 if (!m
->ask_password_event_source
) {
285 assert(m
->ask_password_inotify_fd
< 0);
287 mkdir_p_label("/run/systemd/ask-password", 0755);
289 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
290 if (m
->ask_password_inotify_fd
< 0)
291 return log_error_errno(errno
, "inotify_init1() failed: %m");
293 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
294 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
295 manager_close_ask_password(m
);
299 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
300 m
->ask_password_inotify_fd
, EPOLLIN
,
301 manager_dispatch_ask_password_fd
, m
);
303 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
304 manager_close_ask_password(m
);
308 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
310 /* Queries might have been added meanwhile... */
311 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
312 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
315 return m
->have_ask_password
;
318 static int manager_watch_idle_pipe(Manager
*m
) {
323 if (m
->idle_pipe_event_source
)
326 if (m
->idle_pipe
[2] < 0)
329 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
331 return log_error_errno(r
, "Failed to watch idle pipe: %m");
333 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
338 static void manager_close_idle_pipe(Manager
*m
) {
341 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
343 safe_close_pair(m
->idle_pipe
);
344 safe_close_pair(m
->idle_pipe
+ 2);
347 static int manager_setup_time_change(Manager
*m
) {
350 /* We only care for the cancellation event, hence we set the
351 * timeout to the latest possible value. */
352 struct itimerspec its
= {
353 .it_value
.tv_sec
= TIME_T_MAX
,
357 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
362 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
363 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
365 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
366 if (m
->time_change_fd
< 0)
367 return log_error_errno(errno
, "Failed to create timerfd: %m");
369 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
370 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
371 m
->time_change_fd
= safe_close(m
->time_change_fd
);
375 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
377 return log_error_errno(r
, "Failed to create time change event source: %m");
379 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
381 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
386 static int enable_special_signals(Manager
*m
) {
387 _cleanup_close_
int fd
= -1;
394 /* Enable that we get SIGINT on control-alt-del. In containers
395 * this will fail with EPERM (older) or EINVAL (newer), so
397 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
398 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
400 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
402 /* Support systems without virtual console */
404 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
406 /* Enable that we get SIGWINCH on kbrequest */
407 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
408 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
414 static int manager_setup_signals(Manager
*m
) {
415 struct sigaction sa
= {
416 .sa_handler
= SIG_DFL
,
417 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
424 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
426 /* We make liberal use of realtime signals here. On
427 * Linux/glibc we have 30 of them (with the exception of Linux
428 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
431 assert_se(sigemptyset(&mask
) == 0);
432 sigset_add_many(&mask
,
433 SIGCHLD
, /* Child died */
434 SIGTERM
, /* Reexecute daemon */
435 SIGHUP
, /* Reload configuration */
436 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
437 SIGUSR2
, /* systemd: dump status */
438 SIGINT
, /* Kernel sends us this on control-alt-del */
439 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
440 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
442 SIGRTMIN
+0, /* systemd: start default.target */
443 SIGRTMIN
+1, /* systemd: isolate rescue.target */
444 SIGRTMIN
+2, /* systemd: isolate emergency.target */
445 SIGRTMIN
+3, /* systemd: start halt.target */
446 SIGRTMIN
+4, /* systemd: start poweroff.target */
447 SIGRTMIN
+5, /* systemd: start reboot.target */
448 SIGRTMIN
+6, /* systemd: start kexec.target */
450 /* ... space for more special targets ... */
452 SIGRTMIN
+13, /* systemd: Immediate halt */
453 SIGRTMIN
+14, /* systemd: Immediate poweroff */
454 SIGRTMIN
+15, /* systemd: Immediate reboot */
455 SIGRTMIN
+16, /* systemd: Immediate kexec */
457 /* ... space for more immediate system state changes ... */
459 SIGRTMIN
+20, /* systemd: enable status messages */
460 SIGRTMIN
+21, /* systemd: disable status messages */
461 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
462 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
463 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
465 /* .. one free signal here ... */
467 #if !defined(__hppa64__) && !defined(__hppa__)
468 /* Apparently Linux on hppa has fewer RT
469 * signals (SIGRTMAX is SIGRTMIN+25 there),
470 * hence let's not try to make use of them
471 * here. Since these commands are accessible
472 * by different means and only really a safety
473 * net, the missing functionality on hppa
474 * shouldn't matter. */
476 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
477 SIGRTMIN
+27, /* systemd: set log target to console */
478 SIGRTMIN
+28, /* systemd: set log target to kmsg */
479 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
481 /* ... one free signal here SIGRTMIN+30 ... */
484 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
486 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
487 if (m
->signal_fd
< 0)
490 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
494 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
496 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
497 * notify processing can still figure out to which process/service a message belongs, before we reap the
498 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
499 * status information before detecting that there's no process in a cgroup. */
500 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
504 if (MANAGER_IS_SYSTEM(m
))
505 return enable_special_signals(m
);
510 static void manager_clean_environment(Manager
*m
) {
513 /* Let's remove some environment variables that we
514 * need ourselves to communicate with our clients */
529 static int manager_default_environment(Manager
*m
) {
532 if (MANAGER_IS_SYSTEM(m
)) {
533 /* The system manager always starts with a clean
534 * environment for its children. It does not import
535 * the kernel or the parents exported variables.
537 * The initial passed environ is untouched to keep
538 * /proc/self/environ valid; it is used for tagging
539 * the init process inside containers. */
540 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
543 /* Import locale variables LC_*= from configuration */
544 locale_setup(&m
->environment
);
546 /* The user manager passes its own environment
547 * along to its children. */
548 m
->environment
= strv_copy(environ
);
554 manager_clean_environment(m
);
555 strv_sort(m
->environment
);
560 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
565 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
567 m
= new0(Manager
, 1);
571 m
->unit_file_scope
= scope
;
572 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
573 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
574 m
->default_tasks_accounting
= true;
575 m
->default_tasks_max
= UINT64_MAX
;
578 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
579 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
582 /* Prepare log fields we can use for structured logging */
583 if (MANAGER_IS_SYSTEM(m
)) {
584 m
->unit_log_field
= "UNIT=";
585 m
->unit_log_format_string
= "UNIT=%s";
587 m
->invocation_log_field
= "INVOCATION_ID=";
588 m
->invocation_log_format_string
= "INVOCATION_ID=" SD_ID128_FORMAT_STR
;
590 m
->unit_log_field
= "USER_UNIT=";
591 m
->unit_log_format_string
= "USER_UNIT=%s";
593 m
->invocation_log_field
= "USER_INVOCATION_ID=";
594 m
->invocation_log_format_string
= "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR
;
597 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
599 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
600 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
601 m
->ask_password_inotify_fd
= -1;
603 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
605 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
607 m
->have_ask_password
= -EINVAL
; /* we don't know */
610 m
->test_run
= test_run
;
612 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
613 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
615 r
= manager_default_environment(m
);
619 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
623 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
627 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
631 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
635 r
= sd_event_default(&m
->event
);
639 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
643 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
647 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
651 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
653 r
= manager_setup_signals(m
);
657 r
= manager_setup_cgroup(m
);
661 r
= manager_setup_time_change(m
);
665 m
->udev
= udev_new();
671 /* Note that we do not set up the notify fd here. We do that after deserialization,
672 * since they might have gotten serialized across the reexec. */
674 m
->taint_usr
= dir_is_empty("/usr") > 0;
684 static int manager_setup_notify(Manager
*m
) {
690 if (m
->notify_fd
< 0) {
691 _cleanup_close_
int fd
= -1;
692 union sockaddr_union sa
= {
693 .sa
.sa_family
= AF_UNIX
,
695 static const int one
= 1;
698 /* First free all secondary fields */
699 m
->notify_socket
= mfree(m
->notify_socket
);
700 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
702 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
704 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
706 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
708 e
= manager_get_runtime_prefix(m
);
710 log_error("Failed to determine runtime prefix.");
714 m
->notify_socket
= strappend(e
, "/systemd/notify");
715 if (!m
->notify_socket
)
718 (void) mkdir_parents_label(m
->notify_socket
, 0755);
719 (void) unlink(m
->notify_socket
);
721 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
722 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
724 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
726 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
728 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
733 log_debug("Using notification socket %s", m
->notify_socket
);
736 if (!m
->notify_event_source
) {
737 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
739 return log_error_errno(r
, "Failed to allocate notify event source: %m");
741 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
742 * service an exit message belongs. */
743 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
745 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
747 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
753 static int manager_setup_cgroups_agent(Manager
*m
) {
755 static const union sockaddr_union sa
= {
756 .un
.sun_family
= AF_UNIX
,
757 .un
.sun_path
= "/run/systemd/cgroups-agent",
761 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
762 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
763 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
764 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
765 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
766 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
767 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
768 * we thus won't lose messages.
770 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
771 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
772 * bus for these messages. */
777 if (!MANAGER_IS_SYSTEM(m
))
780 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER
) > 0) /* We don't need this anymore on the unified hierarchy */
783 if (m
->cgroups_agent_fd
< 0) {
784 _cleanup_close_
int fd
= -1;
786 /* First free all secondary fields */
787 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
789 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
791 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
793 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
795 (void) unlink(sa
.un
.sun_path
);
797 /* Only allow root to connect to this socket */
799 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
801 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
803 m
->cgroups_agent_fd
= fd
;
807 if (!m
->cgroups_agent_event_source
) {
808 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
810 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
812 /* Process cgroups notifications early, but after having processed service notification messages or
813 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
814 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
815 * cgroup inotify for the unified cgroup stuff. */
816 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
818 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
820 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
826 static int manager_setup_user_lookup_fd(Manager
*m
) {
831 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
832 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
833 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
834 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
835 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
836 * hence we establish this communication channel so that forked off processes can pass their UID/GID
837 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
838 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
841 * You might wonder why we need a communication channel for this that is independent of the usual notification
842 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
843 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
844 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
846 * Note that this function is called under two circumstances: when we first initialize (in which case we
847 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
848 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
850 if (m
->user_lookup_fds
[0] < 0) {
852 /* Free all secondary fields */
853 safe_close_pair(m
->user_lookup_fds
);
854 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
856 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
857 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
859 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
862 if (!m
->user_lookup_event_source
) {
863 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
865 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
867 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
869 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
871 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
873 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
879 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
880 bool try_bus_connect
;
889 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
891 /* Try to connect to the buses, if possible. */
892 return bus_init(m
, try_bus_connect
);
895 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
901 while ((u
= m
->cleanup_queue
)) {
902 assert(u
->in_cleanup_queue
);
912 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
913 GC_OFFSET_UNSURE
, /* No clue */
914 GC_OFFSET_GOOD
, /* We still need this unit */
915 GC_OFFSET_BAD
, /* We don't need this unit anymore */
919 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
923 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
925 /* Recursively mark referenced units as GOOD as well */
926 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
927 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
928 unit_gc_mark_good(other
, gc_marker
);
931 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
938 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
939 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
940 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
941 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
944 if (u
->in_cleanup_queue
)
947 if (unit_check_gc(u
))
950 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
954 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
955 unit_gc_sweep(other
, gc_marker
);
957 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
960 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
967 /* We were unable to find anything out about this entry, so
968 * let's investigate it later */
969 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
970 unit_add_to_gc_queue(u
);
974 /* We definitely know that this one is not useful anymore, so
975 * let's mark it for deletion */
976 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
977 unit_add_to_cleanup_queue(u
);
981 unit_gc_mark_good(u
, gc_marker
);
984 static unsigned manager_dispatch_gc_queue(Manager
*m
) {
991 /* log_debug("Running GC..."); */
993 m
->gc_marker
+= _GC_OFFSET_MAX
;
994 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
997 gc_marker
= m
->gc_marker
;
999 while ((u
= m
->gc_queue
)) {
1000 assert(u
->in_gc_queue
);
1002 unit_gc_sweep(u
, gc_marker
);
1004 LIST_REMOVE(gc_queue
, m
->gc_queue
, u
);
1005 u
->in_gc_queue
= false;
1009 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1010 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1012 log_unit_debug(u
, "Collecting.");
1013 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1014 unit_add_to_cleanup_queue(u
);
1018 m
->n_in_gc_queue
= 0;
1023 static void manager_clear_jobs_and_units(Manager
*m
) {
1028 while ((u
= hashmap_first(m
->units
)))
1031 manager_dispatch_cleanup_queue(m
);
1033 assert(!m
->load_queue
);
1034 assert(!m
->run_queue
);
1035 assert(!m
->dbus_unit_queue
);
1036 assert(!m
->dbus_job_queue
);
1037 assert(!m
->cleanup_queue
);
1038 assert(!m
->gc_queue
);
1040 assert(hashmap_isempty(m
->jobs
));
1041 assert(hashmap_isempty(m
->units
));
1043 m
->n_on_console
= 0;
1044 m
->n_running_jobs
= 0;
1047 Manager
* manager_free(Manager
*m
) {
1054 manager_clear_jobs_and_units(m
);
1056 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1057 if (unit_vtable
[c
]->shutdown
)
1058 unit_vtable
[c
]->shutdown(m
);
1060 /* If we reexecute ourselves, we keep the root cgroup
1062 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1064 lookup_paths_flush_generator(&m
->lookup_paths
);
1068 dynamic_user_vacuum(m
, false);
1069 hashmap_free(m
->dynamic_users
);
1071 hashmap_free(m
->units
);
1072 hashmap_free(m
->units_by_invocation_id
);
1073 hashmap_free(m
->jobs
);
1074 hashmap_free(m
->watch_pids1
);
1075 hashmap_free(m
->watch_pids2
);
1076 hashmap_free(m
->watch_bus
);
1078 set_free(m
->startup_units
);
1079 set_free(m
->failed_units
);
1081 sd_event_source_unref(m
->signal_event_source
);
1082 sd_event_source_unref(m
->notify_event_source
);
1083 sd_event_source_unref(m
->cgroups_agent_event_source
);
1084 sd_event_source_unref(m
->time_change_event_source
);
1085 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1086 sd_event_source_unref(m
->run_queue_event_source
);
1087 sd_event_source_unref(m
->user_lookup_event_source
);
1089 safe_close(m
->signal_fd
);
1090 safe_close(m
->notify_fd
);
1091 safe_close(m
->cgroups_agent_fd
);
1092 safe_close(m
->time_change_fd
);
1093 safe_close_pair(m
->user_lookup_fds
);
1095 manager_close_ask_password(m
);
1097 manager_close_idle_pipe(m
);
1099 udev_unref(m
->udev
);
1100 sd_event_unref(m
->event
);
1102 free(m
->notify_socket
);
1104 lookup_paths_free(&m
->lookup_paths
);
1105 strv_free(m
->environment
);
1107 hashmap_free(m
->cgroup_unit
);
1108 set_free_free(m
->unit_path_cache
);
1110 free(m
->switch_root
);
1111 free(m
->switch_root_init
);
1113 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1114 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1116 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1117 hashmap_free(m
->units_requiring_mounts_for
);
1119 hashmap_free(m
->uid_refs
);
1120 hashmap_free(m
->gid_refs
);
1125 void manager_enumerate(Manager
*m
) {
1130 /* Let's ask every type to load all units from disk/kernel
1131 * that it might know */
1132 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1133 if (!unit_type_supported(c
)) {
1134 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1138 if (!unit_vtable
[c
]->enumerate
)
1141 unit_vtable
[c
]->enumerate(m
);
1144 manager_dispatch_load_queue(m
);
1147 static void manager_coldplug(Manager
*m
) {
1155 /* Then, let's set up their initial state. */
1156 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1158 /* ignore aliases */
1162 r
= unit_coldplug(u
);
1164 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1168 static void manager_build_unit_path_cache(Manager
*m
) {
1174 set_free_free(m
->unit_path_cache
);
1176 m
->unit_path_cache
= set_new(&string_hash_ops
);
1177 if (!m
->unit_path_cache
) {
1182 /* This simply builds a list of files we know exist, so that
1183 * we don't always have to go to disk */
1185 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1186 _cleanup_closedir_
DIR *d
= NULL
;
1191 if (errno
!= ENOENT
)
1192 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1196 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1199 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
, NULL
);
1205 r
= set_consume(m
->unit_path_cache
, p
);
1214 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1215 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1218 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1224 HASHMAP_FOREACH(u
, m
->units
, i
) {
1226 if (fdset_size(fds
) <= 0)
1229 if (!UNIT_VTABLE(u
)->distribute_fds
)
1232 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1236 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1241 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1245 /* Make sure the transient directory always exists, so that it remains in the search path */
1247 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1252 dual_timestamp_get(&m
->generators_start_timestamp
);
1253 r
= manager_run_generators(m
);
1254 dual_timestamp_get(&m
->generators_finish_timestamp
);
1258 lookup_paths_reduce(&m
->lookup_paths
);
1259 manager_build_unit_path_cache(m
);
1261 /* If we will deserialize make sure that during enumeration
1262 * this is already known, so we increase the counter here
1267 /* First, enumerate what we can from all config files */
1268 dual_timestamp_get(&m
->units_load_start_timestamp
);
1269 manager_enumerate(m
);
1270 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1272 /* Second, deserialize if there is something to deserialize */
1274 r
= manager_deserialize(m
, serialization
, fds
);
1276 /* Any fds left? Find some unit which wants them. This is
1277 * useful to allow container managers to pass some file
1278 * descriptors to us pre-initialized. This enables
1279 * socket-based activation of entire containers. */
1280 manager_distribute_fds(m
, fds
);
1282 /* We might have deserialized the notify fd, but if we didn't
1283 * then let's create the bus now */
1284 q
= manager_setup_notify(m
);
1285 if (q
< 0 && r
== 0)
1288 q
= manager_setup_cgroups_agent(m
);
1289 if (q
< 0 && r
== 0)
1292 q
= manager_setup_user_lookup_fd(m
);
1293 if (q
< 0 && r
== 0)
1296 /* Let's connect to the bus now. */
1297 (void) manager_connect_bus(m
, !!serialization
);
1299 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1300 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1302 /* Third, fire things up! */
1303 manager_coldplug(m
);
1305 /* Release any dynamic users no longer referenced */
1306 dynamic_user_vacuum(m
, true);
1308 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1309 manager_vacuum_uid_refs(m
);
1310 manager_vacuum_gid_refs(m
);
1312 if (serialization
) {
1313 assert(m
->n_reloading
> 0);
1316 /* Let's wait for the UnitNew/JobNew messages being
1317 * sent, before we notify that the reload is
1319 m
->send_reloading_done
= true;
1325 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1330 assert(type
< _JOB_TYPE_MAX
);
1332 assert(mode
< _JOB_MODE_MAX
);
1334 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1335 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1337 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1338 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1340 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1342 type
= job_type_collapse(type
, unit
);
1344 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1348 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1349 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1350 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1354 if (mode
== JOB_ISOLATE
) {
1355 r
= transaction_add_isolate_jobs(tr
, m
);
1360 r
= transaction_activate(tr
, m
, mode
, e
);
1364 log_unit_debug(unit
,
1365 "Enqueued job %s/%s as %u", unit
->id
,
1366 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1369 *_ret
= tr
->anchor_job
;
1371 transaction_free(tr
);
1375 transaction_abort(tr
);
1376 transaction_free(tr
);
1380 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1385 assert(type
< _JOB_TYPE_MAX
);
1387 assert(mode
< _JOB_MODE_MAX
);
1389 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1393 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1396 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1397 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1401 assert(type
< _JOB_TYPE_MAX
);
1403 assert(mode
< _JOB_MODE_MAX
);
1405 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1407 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1412 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1415 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1418 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1422 return hashmap_get(m
->units
, name
);
1425 unsigned manager_dispatch_load_queue(Manager
*m
) {
1431 /* Make sure we are not run recursively */
1432 if (m
->dispatching_load_queue
)
1435 m
->dispatching_load_queue
= true;
1437 /* Dispatches the load queue. Takes a unit from the queue and
1438 * tries to load its data until the queue is empty */
1440 while ((u
= m
->load_queue
)) {
1441 assert(u
->in_load_queue
);
1447 m
->dispatching_load_queue
= false;
1451 int manager_load_unit_prepare(
1463 assert(name
|| path
);
1465 /* This will prepare the unit for loading, but not actually
1466 * load anything from disk. */
1468 if (path
&& !is_path(path
))
1469 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1472 name
= basename(path
);
1474 t
= unit_name_to_type(name
);
1476 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1477 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1478 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1480 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1483 ret
= manager_get_unit(m
, name
);
1489 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1494 ret
->fragment_path
= strdup(path
);
1495 if (!ret
->fragment_path
) {
1501 r
= unit_add_name(ret
, name
);
1507 unit_add_to_load_queue(ret
);
1508 unit_add_to_dbus_queue(ret
);
1509 unit_add_to_gc_queue(ret
);
1517 int manager_load_unit(
1528 /* This will load the service information files, but not actually
1529 * start any services or anything. */
1531 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1535 manager_dispatch_load_queue(m
);
1538 *_ret
= unit_follow_merge(*_ret
);
1543 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1550 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1551 job_dump(j
, f
, prefix
);
1554 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1562 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1564 unit_dump(u
, f
, prefix
);
1567 void manager_clear_jobs(Manager
*m
) {
1572 while ((j
= hashmap_first(m
->jobs
)))
1573 /* No need to recurse. We're cancelling all jobs. */
1574 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1577 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1578 Manager
*m
= userdata
;
1584 while ((j
= m
->run_queue
)) {
1585 assert(j
->installed
);
1586 assert(j
->in_run_queue
);
1588 job_run_and_invalidate(j
);
1591 if (m
->n_running_jobs
> 0)
1592 manager_watch_jobs_in_progress(m
);
1594 if (m
->n_on_console
> 0)
1595 manager_watch_idle_pipe(m
);
1600 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1607 if (m
->dispatching_dbus_queue
)
1610 m
->dispatching_dbus_queue
= true;
1612 while ((u
= m
->dbus_unit_queue
)) {
1613 assert(u
->in_dbus_queue
);
1615 bus_unit_send_change_signal(u
);
1619 while ((j
= m
->dbus_job_queue
)) {
1620 assert(j
->in_dbus_queue
);
1622 bus_job_send_change_signal(j
);
1626 m
->dispatching_dbus_queue
= false;
1628 if (m
->send_reloading_done
) {
1629 m
->send_reloading_done
= false;
1631 bus_manager_send_reloading(m
, false);
1634 if (m
->queued_message
)
1635 bus_send_queued_message(m
);
1640 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1641 Manager
*m
= userdata
;
1642 char buf
[PATH_MAX
+1];
1645 n
= recv(fd
, buf
, sizeof(buf
), 0);
1647 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1649 log_error("Got zero-length cgroups agent message, ignoring.");
1652 if ((size_t) n
>= sizeof(buf
)) {
1653 log_error("Got overly long cgroups agent message, ignoring.");
1657 if (memchr(buf
, 0, n
)) {
1658 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1663 manager_notify_cgroup_empty(m
, buf
);
1664 bus_forward_agent_released(m
, buf
);
1669 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1670 _cleanup_strv_free_
char **tags
= NULL
;
1676 tags
= strv_split(buf
, "\n\r");
1682 if (UNIT_VTABLE(u
)->notify_message
)
1683 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1684 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1685 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1689 y
= ellipsize(x
, 20, 90);
1690 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1694 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1696 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1697 Manager
*m
= userdata
;
1698 char buf
[NOTIFY_BUFFER_MAX
+1];
1699 struct iovec iovec
= {
1701 .iov_len
= sizeof(buf
)-1,
1704 struct cmsghdr cmsghdr
;
1705 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1706 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1708 struct msghdr msghdr
= {
1711 .msg_control
= &control
,
1712 .msg_controllen
= sizeof(control
),
1715 struct cmsghdr
*cmsg
;
1716 struct ucred
*ucred
= NULL
;
1718 int r
, *fd_array
= NULL
;
1723 assert(m
->notify_fd
== fd
);
1725 if (revents
!= EPOLLIN
) {
1726 log_warning("Got unexpected poll event for notify fd.");
1730 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1732 if (IN_SET(errno
, EAGAIN
, EINTR
))
1733 return 0; /* Spurious wakeup, try again */
1735 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1736 * won't take notification messages anymore, but that's still better than busy looping around this:
1737 * being woken up over and over again but being unable to actually read the message off the socket. */
1738 return log_error_errno(errno
, "Failed to receive notification message: %m");
1741 CMSG_FOREACH(cmsg
, &msghdr
) {
1742 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1744 fd_array
= (int*) CMSG_DATA(cmsg
);
1745 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1747 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1748 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1749 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1751 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1758 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1760 close_many(fd_array
, n_fds
);
1766 if (!ucred
|| ucred
->pid
<= 0) {
1767 log_warning("Received notify message without valid credentials. Ignoring.");
1771 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1772 log_warning("Received notify message exceeded maximum size. Ignoring.");
1776 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1777 * trailing NUL byte in the message, but don't expect it. */
1778 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1779 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1783 /* Make sure it's NUL-terminated. */
1786 /* Notify every unit that might be interested, but try
1787 * to avoid notifying the same one multiple times. */
1788 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1790 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1792 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1794 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1796 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1797 if (u3
&& u3
!= u2
&& u3
!= u1
)
1798 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1800 if (!u1
&& !u2
&& !u3
)
1801 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1803 if (fdset_size(fds
) > 0)
1804 log_warning("Got extra auxiliary fds with notification message, closing them.");
1809 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1816 sd_event_get_iteration(m
->event
, &iteration
);
1818 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1820 unit_unwatch_pid(u
, si
->si_pid
);
1822 if (UNIT_VTABLE(u
)->sigchld_event
) {
1823 if (set_size(u
->pids
) <= 1 ||
1824 iteration
!= u
->sigchldgen
||
1825 unit_main_pid(u
) == si
->si_pid
||
1826 unit_control_pid(u
) == si
->si_pid
) {
1827 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1828 u
->sigchldgen
= iteration
;
1830 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1834 static int manager_dispatch_sigchld(Manager
*m
) {
1840 /* First we call waitd() for a PID and do not reap the
1841 * zombie. That way we can still access /proc/$PID for
1842 * it while it is a zombie. */
1843 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1845 if (errno
== ECHILD
)
1857 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1858 _cleanup_free_
char *name
= NULL
;
1861 get_process_comm(si
.si_pid
, &name
);
1863 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1864 si
.si_pid
, strna(name
),
1865 sigchld_code_to_string(si
.si_code
),
1867 strna(si
.si_code
== CLD_EXITED
1868 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1869 : signal_to_string(si
.si_status
)));
1871 /* And now figure out the unit this belongs
1872 * to, it might be multiple... */
1873 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1875 invoke_sigchld_event(m
, u1
, &si
);
1876 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1878 invoke_sigchld_event(m
, u2
, &si
);
1879 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1880 if (u3
&& u3
!= u2
&& u3
!= u1
)
1881 invoke_sigchld_event(m
, u3
, &si
);
1884 /* And now, we actually reap the zombie. */
1885 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1896 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1897 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1900 log_debug("Activating special unit %s", name
);
1902 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1904 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1909 static void manager_handle_ctrl_alt_del(Manager
*m
) {
1910 /* If the user presses C-A-D more than
1911 * 7 times within 2s, we reboot/shutdown immediately,
1912 * unless it was disabled in system.conf */
1914 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== CAD_BURST_ACTION_IGNORE
)
1915 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1917 switch (m
->cad_burst_action
) {
1919 case CAD_BURST_ACTION_REBOOT
:
1920 m
->exit_code
= MANAGER_REBOOT
;
1923 case CAD_BURST_ACTION_POWEROFF
:
1924 m
->exit_code
= MANAGER_POWEROFF
;
1928 assert_not_reached("Unknown action.");
1931 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1932 cad_burst_action_to_string(m
->cad_burst_action
));
1933 status_printf(NULL
, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1934 cad_burst_action_to_string(m
->cad_burst_action
));
1938 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1939 Manager
*m
= userdata
;
1941 struct signalfd_siginfo sfsi
;
1942 bool sigchld
= false;
1946 assert(m
->signal_fd
== fd
);
1948 if (revents
!= EPOLLIN
) {
1949 log_warning("Got unexpected events from signal file descriptor.");
1954 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1955 if (n
!= sizeof(sfsi
)) {
1957 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
1961 if (IN_SET(errno
, EINTR
, EAGAIN
))
1964 /* We return an error here, which will kill this handler,
1965 * to avoid a busy loop on read error. */
1966 return log_error_errno(errno
, "Reading from signal fd failed: %m");
1969 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1970 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1971 ? LOG_DEBUG
: LOG_INFO
,
1974 switch (sfsi
.ssi_signo
) {
1981 if (MANAGER_IS_SYSTEM(m
)) {
1982 /* This is for compatibility with the
1983 * original sysvinit */
1984 m
->exit_code
= MANAGER_REEXECUTE
;
1991 if (MANAGER_IS_SYSTEM(m
)) {
1992 manager_handle_ctrl_alt_del(m
);
1996 /* Run the exit target if there is one, if not, just exit. */
1997 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
1998 m
->exit_code
= MANAGER_EXIT
;
2005 if (MANAGER_IS_SYSTEM(m
))
2006 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2008 /* This is a nop on non-init */
2012 if (MANAGER_IS_SYSTEM(m
))
2013 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2015 /* This is a nop on non-init */
2021 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2023 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2024 log_info("Trying to reconnect to bus...");
2028 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2029 log_info("Loading D-Bus service...");
2030 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2037 _cleanup_free_
char *dump
= NULL
;
2038 _cleanup_fclose_
FILE *f
= NULL
;
2041 f
= open_memstream(&dump
, &size
);
2043 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2047 manager_dump_units(m
, f
, "\t");
2048 manager_dump_jobs(m
, f
, "\t");
2050 r
= fflush_and_check(f
);
2052 log_warning_errno(r
, "Failed to write status stream: %m");
2056 log_dump(LOG_INFO
, dump
);
2061 m
->exit_code
= MANAGER_RELOAD
;
2066 /* Starting SIGRTMIN+0 */
2067 static const char * const target_table
[] = {
2068 [0] = SPECIAL_DEFAULT_TARGET
,
2069 [1] = SPECIAL_RESCUE_TARGET
,
2070 [2] = SPECIAL_EMERGENCY_TARGET
,
2071 [3] = SPECIAL_HALT_TARGET
,
2072 [4] = SPECIAL_POWEROFF_TARGET
,
2073 [5] = SPECIAL_REBOOT_TARGET
,
2074 [6] = SPECIAL_KEXEC_TARGET
2077 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2078 static const ManagerExitCode code_table
[] = {
2080 [1] = MANAGER_POWEROFF
,
2081 [2] = MANAGER_REBOOT
,
2085 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2086 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2087 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2088 manager_start_target(m
, target_table
[idx
],
2089 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
2093 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2094 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2095 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2099 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2102 manager_set_show_status(m
, SHOW_STATUS_YES
);
2106 manager_set_show_status(m
, SHOW_STATUS_NO
);
2110 log_set_max_level(LOG_DEBUG
);
2111 log_info("Setting log level to debug.");
2115 log_set_max_level(LOG_INFO
);
2116 log_info("Setting log level to info.");
2120 if (MANAGER_IS_USER(m
)) {
2121 m
->exit_code
= MANAGER_EXIT
;
2125 /* This is a nop on init */
2129 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2130 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2131 log_notice("Setting log target to journal-or-kmsg.");
2135 log_set_target(LOG_TARGET_CONSOLE
);
2136 log_notice("Setting log target to console.");
2140 log_set_target(LOG_TARGET_KMSG
);
2141 log_notice("Setting log target to kmsg.");
2145 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2152 manager_dispatch_sigchld(m
);
2157 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2158 Manager
*m
= userdata
;
2163 assert(m
->time_change_fd
== fd
);
2165 log_struct(LOG_INFO
,
2166 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE
),
2167 LOG_MESSAGE("Time has been changed"),
2170 /* Restart the watch */
2171 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2172 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2174 manager_setup_time_change(m
);
2176 HASHMAP_FOREACH(u
, m
->units
, i
)
2177 if (UNIT_VTABLE(u
)->time_change
)
2178 UNIT_VTABLE(u
)->time_change(u
);
2183 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2184 Manager
*m
= userdata
;
2187 assert(m
->idle_pipe
[2] == fd
);
2189 m
->no_console_output
= m
->n_on_console
> 0;
2191 manager_close_idle_pipe(m
);
2196 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2197 Manager
*m
= userdata
;
2204 manager_print_jobs_in_progress(m
);
2206 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2207 r
= sd_event_source_set_time(source
, next
);
2211 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2214 int manager_loop(Manager
*m
) {
2217 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2220 m
->exit_code
= MANAGER_OK
;
2222 /* Release the path cache */
2223 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2225 manager_check_finished(m
);
2227 /* There might still be some zombies hanging around from
2228 * before we were exec()'ed. Let's reap them. */
2229 r
= manager_dispatch_sigchld(m
);
2233 while (m
->exit_code
== MANAGER_OK
) {
2236 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2239 if (!ratelimit_test(&rl
)) {
2240 /* Yay, something is going seriously wrong, pause a little */
2241 log_warning("Looping too fast. Throttling execution a little.");
2245 if (manager_dispatch_load_queue(m
) > 0)
2248 if (manager_dispatch_gc_queue(m
) > 0)
2251 if (manager_dispatch_cleanup_queue(m
) > 0)
2254 if (manager_dispatch_cgroup_queue(m
) > 0)
2257 if (manager_dispatch_dbus_queue(m
) > 0)
2260 /* Sleep for half the watchdog time */
2261 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2262 wait_usec
= m
->runtime_watchdog
/ 2;
2266 wait_usec
= USEC_INFINITY
;
2268 r
= sd_event_run(m
->event
, wait_usec
);
2270 return log_error_errno(r
, "Failed to run event loop: %m");
2273 return m
->exit_code
;
2276 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2277 _cleanup_free_
char *n
= NULL
;
2278 sd_id128_t invocation_id
;
2286 r
= unit_name_from_dbus_path(s
, &n
);
2290 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2291 * as invocation ID. */
2292 r
= sd_id128_from_string(n
, &invocation_id
);
2294 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2300 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.", SD_ID128_FORMAT_VAL(invocation_id
));
2303 /* If this didn't work, we use the suffix as unit name. */
2304 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2312 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2322 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2326 r
= safe_atou(p
, &id
);
2330 j
= manager_get_job(m
, id
);
2339 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2342 _cleanup_free_
char *p
= NULL
;
2346 if (!MANAGER_IS_SYSTEM(m
))
2349 audit_fd
= get_audit_fd();
2353 /* Don't generate audit events if the service was already
2354 * started and we're just deserializing */
2355 if (MANAGER_IS_RELOADING(m
))
2358 if (u
->type
!= UNIT_SERVICE
)
2361 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2363 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2367 msg
= strjoina("unit=", p
);
2368 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2370 /* We aren't allowed to send audit messages?
2371 * Then let's not retry again. */
2374 log_warning_errno(errno
, "Failed to send audit message: %m");
2380 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2381 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2382 _cleanup_free_
char *message
= NULL
;
2383 _cleanup_close_
int fd
= -1;
2386 /* Don't generate plymouth events if the service was already
2387 * started and we're just deserializing */
2388 if (MANAGER_IS_RELOADING(m
))
2391 if (!MANAGER_IS_SYSTEM(m
))
2394 if (detect_container() > 0)
2397 if (u
->type
!= UNIT_SERVICE
&&
2398 u
->type
!= UNIT_MOUNT
&&
2399 u
->type
!= UNIT_SWAP
)
2402 /* We set SOCK_NONBLOCK here so that we rather drop the
2403 * message then wait for plymouth */
2404 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2406 log_error_errno(errno
, "socket() failed: %m");
2410 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2412 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2413 log_error_errno(errno
, "connect() failed: %m");
2417 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2423 if (write(fd
, message
, n
+ 1) != n
+ 1)
2424 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2425 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2428 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2435 path
= MANAGER_IS_SYSTEM(m
) ? "/run/systemd" : "/tmp";
2436 fd
= open_tmpfile_unlinkable(path
, O_RDWR
|O_CLOEXEC
);
2440 log_debug("Serializing state to %s", path
);
2442 f
= fdopen(fd
, "w+");
2453 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2466 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2467 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2468 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2469 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2471 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2472 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2473 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2474 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2477 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2478 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2479 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2480 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2481 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2482 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2483 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2484 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2487 if (!switching_root
) {
2488 STRV_FOREACH(e
, m
->environment
) {
2489 _cleanup_free_
char *ce
;
2495 fprintf(f
, "env=%s\n", *e
);
2499 if (m
->notify_fd
>= 0) {
2502 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2506 fprintf(f
, "notify-fd=%i\n", copy
);
2507 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2510 if (m
->cgroups_agent_fd
>= 0) {
2513 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2517 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2520 if (m
->user_lookup_fds
[0] >= 0) {
2523 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2527 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2531 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2534 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2536 r
= dynamic_user_serialize(m
, f
, fds
);
2540 manager_serialize_uid_refs(m
, f
);
2541 manager_serialize_gid_refs(m
, f
);
2545 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2553 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2560 assert(m
->n_reloading
> 0);
2566 r
= bus_fdset_add_all(m
, fds
);
2573 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2579 log_debug("Deserializing state...");
2584 char line
[LINE_MAX
], *l
;
2586 if (!fgets(line
, sizeof(line
), f
)) {
2601 if (startswith(l
, "current-job-id=")) {
2604 if (safe_atou32(l
+15, &id
) < 0)
2605 log_debug("Failed to parse current job id value %s", l
+15);
2607 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2609 } else if (startswith(l
, "n-installed-jobs=")) {
2612 if (safe_atou32(l
+17, &n
) < 0)
2613 log_debug("Failed to parse installed jobs counter %s", l
+17);
2615 m
->n_installed_jobs
+= n
;
2617 } else if (startswith(l
, "n-failed-jobs=")) {
2620 if (safe_atou32(l
+14, &n
) < 0)
2621 log_debug("Failed to parse failed jobs counter %s", l
+14);
2623 m
->n_failed_jobs
+= n
;
2625 } else if (startswith(l
, "taint-usr=")) {
2628 b
= parse_boolean(l
+10);
2630 log_debug("Failed to parse taint /usr flag %s", l
+10);
2632 m
->taint_usr
= m
->taint_usr
|| b
;
2634 } else if (startswith(l
, "firmware-timestamp="))
2635 dual_timestamp_deserialize(l
+19, &m
->firmware_timestamp
);
2636 else if (startswith(l
, "loader-timestamp="))
2637 dual_timestamp_deserialize(l
+17, &m
->loader_timestamp
);
2638 else if (startswith(l
, "kernel-timestamp="))
2639 dual_timestamp_deserialize(l
+17, &m
->kernel_timestamp
);
2640 else if (startswith(l
, "initrd-timestamp="))
2641 dual_timestamp_deserialize(l
+17, &m
->initrd_timestamp
);
2642 else if (startswith(l
, "userspace-timestamp="))
2643 dual_timestamp_deserialize(l
+20, &m
->userspace_timestamp
);
2644 else if (startswith(l
, "finish-timestamp="))
2645 dual_timestamp_deserialize(l
+17, &m
->finish_timestamp
);
2646 else if (startswith(l
, "security-start-timestamp="))
2647 dual_timestamp_deserialize(l
+25, &m
->security_start_timestamp
);
2648 else if (startswith(l
, "security-finish-timestamp="))
2649 dual_timestamp_deserialize(l
+26, &m
->security_finish_timestamp
);
2650 else if (startswith(l
, "generators-start-timestamp="))
2651 dual_timestamp_deserialize(l
+27, &m
->generators_start_timestamp
);
2652 else if (startswith(l
, "generators-finish-timestamp="))
2653 dual_timestamp_deserialize(l
+28, &m
->generators_finish_timestamp
);
2654 else if (startswith(l
, "units-load-start-timestamp="))
2655 dual_timestamp_deserialize(l
+27, &m
->units_load_start_timestamp
);
2656 else if (startswith(l
, "units-load-finish-timestamp="))
2657 dual_timestamp_deserialize(l
+28, &m
->units_load_finish_timestamp
);
2658 else if (startswith(l
, "env=")) {
2659 _cleanup_free_
char *uce
= NULL
;
2662 r
= cunescape(l
+ 4, UNESCAPE_RELAX
, &uce
);
2666 e
= strv_env_set(m
->environment
, uce
);
2672 strv_free(m
->environment
);
2675 } else if (startswith(l
, "notify-fd=")) {
2678 if (safe_atoi(l
+ 10, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2679 log_debug("Failed to parse notify fd: %s", l
+ 10);
2681 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2682 safe_close(m
->notify_fd
);
2683 m
->notify_fd
= fdset_remove(fds
, fd
);
2686 } else if (startswith(l
, "notify-socket=")) {
2695 free(m
->notify_socket
);
2696 m
->notify_socket
= n
;
2698 } else if (startswith(l
, "cgroups-agent-fd=")) {
2701 if (safe_atoi(l
+ 17, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2702 log_debug("Failed to parse cgroups agent fd: %s", l
+ 10);
2704 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2705 safe_close(m
->cgroups_agent_fd
);
2706 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2709 } else if (startswith(l
, "user-lookup=")) {
2712 if (sscanf(l
+ 12, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2713 log_debug("Failed to parse user lookup fd: %s", l
+ 12);
2715 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2716 safe_close_pair(m
->user_lookup_fds
);
2717 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2718 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2721 } else if (startswith(l
, "dynamic-user="))
2722 dynamic_user_deserialize_one(m
, l
+ 13, fds
);
2723 else if (startswith(l
, "destroy-ipc-uid="))
2724 manager_deserialize_uid_refs_one(m
, l
+ 16);
2725 else if (startswith(l
, "destroy-ipc-gid="))
2726 manager_deserialize_gid_refs_one(m
, l
+ 16);
2727 else if (startswith(l
, "subscribed=")) {
2729 if (strv_extend(&m
->deserialized_subscribed
, l
+11) < 0)
2732 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2733 log_debug("Unknown serialization item '%s'", l
);
2738 char name
[UNIT_NAME_MAX
+2];
2741 if (!fgets(name
, sizeof(name
), f
)) {
2752 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2756 r
= unit_deserialize(u
, f
, fds
);
2765 assert(m
->n_reloading
> 0);
2771 int manager_reload(Manager
*m
) {
2773 _cleanup_fclose_
FILE *f
= NULL
;
2774 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2778 r
= manager_open_serialization(m
, &f
);
2783 bus_manager_send_reloading(m
, true);
2791 r
= manager_serialize(m
, f
, fds
, false);
2797 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2802 /* From here on there is no way back. */
2803 manager_clear_jobs_and_units(m
);
2804 lookup_paths_flush_generator(&m
->lookup_paths
);
2805 lookup_paths_free(&m
->lookup_paths
);
2806 dynamic_user_vacuum(m
, false);
2807 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2808 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2810 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2811 if (q
< 0 && r
>= 0)
2814 /* Find new unit paths */
2815 q
= manager_run_generators(m
);
2816 if (q
< 0 && r
>= 0)
2819 lookup_paths_reduce(&m
->lookup_paths
);
2820 manager_build_unit_path_cache(m
);
2822 /* First, enumerate what we can from all config files */
2823 manager_enumerate(m
);
2825 /* Second, deserialize our stored data */
2826 q
= manager_deserialize(m
, f
, fds
);
2827 if (q
< 0 && r
>= 0)
2833 /* Re-register notify_fd as event source */
2834 q
= manager_setup_notify(m
);
2835 if (q
< 0 && r
>= 0)
2838 q
= manager_setup_cgroups_agent(m
);
2839 if (q
< 0 && r
>= 0)
2842 q
= manager_setup_user_lookup_fd(m
);
2843 if (q
< 0 && r
>= 0)
2846 /* Third, fire things up! */
2847 manager_coldplug(m
);
2849 /* Release any dynamic users no longer referenced */
2850 dynamic_user_vacuum(m
, true);
2852 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2853 manager_vacuum_uid_refs(m
);
2854 manager_vacuum_gid_refs(m
);
2856 /* Sync current state of bus names with our set of listening units */
2858 manager_sync_bus_names(m
, m
->api_bus
);
2860 assert(m
->n_reloading
> 0);
2863 m
->send_reloading_done
= true;
2868 void manager_reset_failed(Manager
*m
) {
2874 HASHMAP_FOREACH(u
, m
->units
, i
)
2875 unit_reset_failed(u
);
2878 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2884 /* Returns true if the unit is inactive or going down */
2885 u
= manager_get_unit(m
, name
);
2889 return unit_inactive_or_pending(u
);
2892 static void manager_notify_finished(Manager
*m
) {
2893 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2894 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2899 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2901 /* Note that m->kernel_usec.monotonic is always at 0,
2902 * and m->firmware_usec.monotonic and
2903 * m->loader_usec.monotonic should be considered
2904 * negative values. */
2906 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2907 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2908 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2909 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2911 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2913 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2914 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2916 log_struct(LOG_INFO
,
2917 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2918 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2919 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2920 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2921 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2922 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2923 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2924 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2925 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2928 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2931 log_struct(LOG_INFO
,
2932 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2933 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2934 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2935 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2936 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2937 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2938 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2942 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2943 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2945 log_struct(LOG_INFO
,
2946 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2947 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2948 LOG_MESSAGE("Startup finished in %s.",
2949 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2953 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2957 "STATUS=Startup finished in %s.",
2958 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2961 void manager_check_finished(Manager
*m
) {
2964 if (MANAGER_IS_RELOADING(m
))
2967 /* Verify that we are actually running currently. Initially
2968 * the exit code is set to invalid, and during operation it is
2969 * then set to MANAGER_OK */
2970 if (m
->exit_code
!= MANAGER_OK
)
2973 if (hashmap_size(m
->jobs
) > 0) {
2974 if (m
->jobs_in_progress_event_source
)
2975 /* Ignore any failure, this is only for feedback */
2976 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2981 manager_flip_auto_status(m
, false);
2983 /* Notify Type=idle units that we are done now */
2984 manager_close_idle_pipe(m
);
2986 /* Turn off confirm spawn now */
2987 m
->confirm_spawn
= false;
2989 /* No need to update ask password status when we're going non-interactive */
2990 manager_close_ask_password(m
);
2992 /* This is no longer the first boot */
2993 manager_set_first_boot(m
, false);
2995 if (dual_timestamp_is_set(&m
->finish_timestamp
))
2998 dual_timestamp_get(&m
->finish_timestamp
);
3000 manager_notify_finished(m
);
3002 manager_invalidate_startup_units(m
);
3005 static int manager_run_generators(Manager
*m
) {
3006 _cleanup_strv_free_
char **paths
= NULL
;
3007 const char *argv
[5];
3016 paths
= generator_binary_paths(m
->unit_file_scope
);
3020 /* Optimize by skipping the whole process by not creating output directories
3021 * if no generators are found. */
3022 STRV_FOREACH(path
, paths
) {
3023 if (access(*path
, F_OK
) >= 0)
3025 if (errno
!= ENOENT
)
3026 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3032 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3036 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3037 argv
[1] = m
->lookup_paths
.generator
;
3038 argv
[2] = m
->lookup_paths
.generator_early
;
3039 argv
[3] = m
->lookup_paths
.generator_late
;
3042 RUN_WITH_UMASK(0022)
3043 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, (char**) argv
);
3046 lookup_paths_trim_generator(&m
->lookup_paths
);
3050 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3051 char **a
= NULL
, **b
= NULL
, **l
;
3056 if (!strv_isempty(minus
)) {
3057 a
= strv_env_delete(l
, 1, minus
);
3064 if (!strv_isempty(plus
)) {
3065 b
= strv_env_merge(2, l
, plus
);
3074 if (m
->environment
!= l
)
3075 strv_free(m
->environment
);
3082 manager_clean_environment(m
);
3083 strv_sort(m
->environment
);
3088 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3093 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3094 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3096 if (!default_rlimit
[i
])
3099 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3107 void manager_recheck_journal(Manager
*m
) {
3112 if (!MANAGER_IS_SYSTEM(m
))
3115 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3116 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3117 log_close_journal();
3121 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3122 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3123 log_close_journal();
3127 /* Hmm, OK, so the socket is fully up and the service is up
3128 * too, then let's make use of the thing. */
3132 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3134 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3136 if (!MANAGER_IS_SYSTEM(m
))
3139 if (m
->show_status
!= mode
)
3140 log_debug("%s showing of status.",
3141 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3142 m
->show_status
= mode
;
3145 (void) touch("/run/systemd/show-status");
3147 (void) unlink("/run/systemd/show-status");
3150 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3153 if (!MANAGER_IS_SYSTEM(m
))
3156 if (m
->no_console_output
)
3159 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3162 /* If we cannot find out the status properly, just proceed. */
3163 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3166 if (m
->show_status
> 0)
3172 void manager_set_first_boot(Manager
*m
, bool b
) {
3175 if (!MANAGER_IS_SYSTEM(m
))
3178 if (m
->first_boot
!= (int) b
) {
3180 (void) touch("/run/systemd/first-boot");
3182 (void) unlink("/run/systemd/first-boot");
3188 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3191 /* If m is NULL, assume we're after shutdown and let the messages through. */
3193 if (m
&& !manager_get_show_status(m
, type
))
3196 /* XXX We should totally drop the check for ephemeral here
3197 * and thus effectively make 'Type=idle' pointless. */
3198 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3201 va_start(ap
, format
);
3202 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3206 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3207 char p
[strlen(path
)+1];
3213 path_kill_slashes(p
);
3215 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3218 const char *manager_get_runtime_prefix(Manager
*m
) {
3221 return MANAGER_IS_SYSTEM(m
) ?
3223 getenv("XDG_RUNTIME_DIR");
3226 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3231 assert(u
->manager
== m
);
3233 size
= set_size(m
->failed_units
);
3236 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3240 if (set_put(m
->failed_units
, u
) < 0)
3243 (void) set_remove(m
->failed_units
, u
);
3245 if (set_size(m
->failed_units
) != size
)
3246 bus_manager_send_change_signal(m
);
3251 ManagerState
manager_state(Manager
*m
) {
3256 /* Did we ever finish booting? If not then we are still starting up */
3257 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3259 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3260 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3261 return MANAGER_INITIALIZING
;
3263 return MANAGER_STARTING
;
3266 /* Is the special shutdown target queued? If so, we are in shutdown state */
3267 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3268 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3269 return MANAGER_STOPPING
;
3271 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3272 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3273 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3274 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3275 return MANAGER_MAINTENANCE
;
3277 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3278 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3279 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3280 return MANAGER_MAINTENANCE
;
3282 /* Are there any failed units? If so, we are in degraded mode */
3283 if (set_size(m
->failed_units
) > 0)
3284 return MANAGER_DEGRADED
;
3286 return MANAGER_RUNNING
;
3289 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3291 static void manager_unref_uid_internal(
3296 int (*_clean_ipc
)(uid_t uid
)) {
3302 assert(uid_is_valid(uid
));
3305 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3306 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3308 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3309 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3310 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3311 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3313 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3314 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3316 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3319 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3321 n
= c
& ~DESTROY_IPC_FLAG
;
3325 if (destroy_now
&& n
== 0) {
3326 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3328 if (c
& DESTROY_IPC_FLAG
) {
3329 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3330 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3332 (void) _clean_ipc(uid
);
3335 c
= n
| (c
& DESTROY_IPC_FLAG
);
3336 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3340 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3341 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3344 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3345 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3348 static int manager_ref_uid_internal(
3359 assert(uid_is_valid(uid
));
3361 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3362 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3364 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3365 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3367 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3370 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3374 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3376 n
= c
& ~DESTROY_IPC_FLAG
;
3379 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3382 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3384 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3387 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3388 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3391 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3392 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3395 static void manager_vacuum_uid_refs_internal(
3398 int (*_clean_ipc
)(uid_t uid
)) {
3407 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3411 uid
= PTR_TO_UID(k
);
3412 c
= PTR_TO_UINT32(p
);
3414 n
= c
& ~DESTROY_IPC_FLAG
;
3418 if (c
& DESTROY_IPC_FLAG
) {
3419 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3420 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3422 (void) _clean_ipc(uid
);
3425 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3429 void manager_vacuum_uid_refs(Manager
*m
) {
3430 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3433 void manager_vacuum_gid_refs(Manager
*m
) {
3434 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3437 static void manager_serialize_uid_refs_internal(
3441 const char *field_name
) {
3451 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3452 * of it is better rebuild after a reload/reexec. */
3454 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3458 uid
= PTR_TO_UID(k
);
3459 c
= PTR_TO_UINT32(p
);
3461 if (!(c
& DESTROY_IPC_FLAG
))
3464 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3468 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3469 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3472 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3473 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3476 static void manager_deserialize_uid_refs_one_internal(
3479 const char *value
) {
3489 r
= parse_uid(value
, &uid
);
3490 if (r
< 0 || uid
== 0) {
3491 log_debug("Unable to parse UID reference serialization");
3495 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3501 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3502 if (c
& DESTROY_IPC_FLAG
)
3505 c
|= DESTROY_IPC_FLAG
;
3507 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3509 log_debug("Failed to add UID reference entry");
3514 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3515 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3518 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3519 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3522 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3526 char unit_name
[UNIT_NAME_MAX
+1];
3529 Manager
*m
= userdata
;
3537 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3538 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3539 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3541 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3543 if (errno
== EINTR
|| errno
== EAGAIN
)
3546 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3549 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3550 log_warning("Received too short user lookup message, ignoring.");
3554 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3555 log_warning("Received too long user lookup message, ignoring.");
3559 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3560 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3564 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3565 if (memchr(buffer
.unit_name
, 0, n
)) {
3566 log_warning("Received lookup message with embedded NUL character, ignoring.");
3570 buffer
.unit_name
[n
] = 0;
3571 u
= manager_get_unit(m
, buffer
.unit_name
);
3573 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3577 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3579 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3583 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3584 [MANAGER_INITIALIZING
] = "initializing",
3585 [MANAGER_STARTING
] = "starting",
3586 [MANAGER_RUNNING
] = "running",
3587 [MANAGER_DEGRADED
] = "degraded",
3588 [MANAGER_MAINTENANCE
] = "maintenance",
3589 [MANAGER_STOPPING
] = "stopping",
3592 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
3594 static const char *const cad_burst_action_table
[_CAD_BURST_ACTION_MAX
] = {
3595 [CAD_BURST_ACTION_IGNORE
] = "ignore",
3596 [CAD_BURST_ACTION_REBOOT
] = "reboot-force",
3597 [CAD_BURST_ACTION_POWEROFF
] = "poweroff-force",
3600 DEFINE_STRING_TABLE_LOOKUP(cad_burst_action
, CADBurstAction
);