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/>.
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
40 #include "alloc-util.h"
42 #include "boot-timestamps.h"
43 #include "bus-common-errors.h"
44 #include "bus-error.h"
45 #include "bus-kernel.h"
47 #include "clean-ipc.h"
49 #include "dbus-manager.h"
50 #include "dbus-unit.h"
52 #include "dirent-util.h"
55 #include "exec-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_environment_generators(Manager
*m
);
107 static int manager_run_generators(Manager
*m
);
109 static void manager_watch_jobs_in_progress(Manager
*m
) {
115 /* We do not want to show the cylon animation if the user
116 * needs to confirm service executions otherwise confirmation
117 * messages will be screwed by the cylon animation. */
118 if (!manager_is_confirm_spawn_disabled(m
))
121 if (m
->jobs_in_progress_event_source
)
124 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
125 r
= sd_event_add_time(
127 &m
->jobs_in_progress_event_source
,
130 manager_dispatch_jobs_in_progress
, m
);
134 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
137 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
139 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
142 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
143 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
147 p
= mempset(p
, ' ', pos
-2);
148 if (log_get_show_color())
149 p
= stpcpy(p
, ANSI_RED
);
153 if (pos
> 0 && pos
<= width
) {
154 if (log_get_show_color())
155 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
159 if (log_get_show_color())
160 p
= stpcpy(p
, ANSI_NORMAL
);
163 if (log_get_show_color())
164 p
= stpcpy(p
, ANSI_RED
);
167 p
= mempset(p
, ' ', width
-1-pos
);
168 if (log_get_show_color())
169 strcpy(p
, ANSI_NORMAL
);
173 void manager_flip_auto_status(Manager
*m
, bool enable
) {
177 if (m
->show_status
== SHOW_STATUS_AUTO
)
178 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
180 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
181 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
185 static void manager_print_jobs_in_progress(Manager
*m
) {
186 _cleanup_free_
char *job_of_n
= NULL
;
189 unsigned counter
= 0, print_nr
;
190 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
192 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
196 assert(m
->n_running_jobs
> 0);
198 manager_flip_auto_status(m
, true);
200 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
202 HASHMAP_FOREACH(j
, m
->jobs
, i
)
203 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
206 /* m->n_running_jobs must be consistent with the contents of m->jobs,
207 * so the above loop must have succeeded in finding j. */
208 assert(counter
== print_nr
+ 1);
211 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
213 cylon_pos
= 14 - cylon_pos
;
214 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
216 m
->jobs_in_progress_iteration
++;
218 if (m
->n_running_jobs
> 1) {
219 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
223 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
224 if (job_get_timeout(j
, &x
) > 0)
225 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
227 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
228 "%sA %s job is running for %s (%s / %s)",
230 job_type_to_string(j
->type
),
231 unit_description(j
->unit
),
235 static int have_ask_password(void) {
236 _cleanup_closedir_
DIR *dir
;
239 dir
= opendir("/run/systemd/ask-password");
247 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
248 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's or the parents' exported variables.
537 * The initial passed environment 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
);
553 manager_clean_environment(m
);
554 strv_sort(m
->environment
);
559 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
564 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
566 m
= new0(Manager
, 1);
570 m
->unit_file_scope
= scope
;
571 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
572 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
573 m
->default_tasks_accounting
= true;
574 m
->default_tasks_max
= UINT64_MAX
;
577 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
578 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
581 /* Prepare log fields we can use for structured logging */
582 if (MANAGER_IS_SYSTEM(m
)) {
583 m
->unit_log_field
= "UNIT=";
584 m
->unit_log_format_string
= "UNIT=%s";
586 m
->invocation_log_field
= "INVOCATION_ID=";
587 m
->invocation_log_format_string
= "INVOCATION_ID=" SD_ID128_FORMAT_STR
;
589 m
->unit_log_field
= "USER_UNIT=";
590 m
->unit_log_format_string
= "USER_UNIT=%s";
592 m
->invocation_log_field
= "USER_INVOCATION_ID=";
593 m
->invocation_log_format_string
= "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR
;
596 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
598 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
599 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
600 m
->ask_password_inotify_fd
= -1;
602 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
604 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
606 m
->have_ask_password
= -EINVAL
; /* we don't know */
609 m
->test_run
= test_run
;
611 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
612 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
614 r
= manager_default_environment(m
);
618 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
622 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
626 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
630 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
634 r
= sd_event_default(&m
->event
);
638 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
642 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
646 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
650 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
652 r
= manager_setup_signals(m
);
656 r
= manager_setup_cgroup(m
);
660 r
= manager_setup_time_change(m
);
664 m
->udev
= udev_new();
670 /* Note that we do not set up the notify fd here. We do that after deserialization,
671 * since they might have gotten serialized across the reexec. */
673 m
->taint_usr
= dir_is_empty("/usr") > 0;
683 static int manager_setup_notify(Manager
*m
) {
689 if (m
->notify_fd
< 0) {
690 _cleanup_close_
int fd
= -1;
691 union sockaddr_union sa
= {
692 .sa
.sa_family
= AF_UNIX
,
694 static const int one
= 1;
697 /* First free all secondary fields */
698 m
->notify_socket
= mfree(m
->notify_socket
);
699 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
701 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
703 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
705 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
707 e
= manager_get_runtime_prefix(m
);
709 log_error("Failed to determine runtime prefix.");
713 m
->notify_socket
= strappend(e
, "/systemd/notify");
714 if (!m
->notify_socket
)
717 (void) mkdir_parents_label(m
->notify_socket
, 0755);
718 (void) unlink(m
->notify_socket
);
720 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
721 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
723 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
725 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
727 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
732 log_debug("Using notification socket %s", m
->notify_socket
);
735 if (!m
->notify_event_source
) {
736 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
738 return log_error_errno(r
, "Failed to allocate notify event source: %m");
740 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
741 * service an exit message belongs. */
742 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
744 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
746 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
752 static int manager_setup_cgroups_agent(Manager
*m
) {
754 static const union sockaddr_union sa
= {
755 .un
.sun_family
= AF_UNIX
,
756 .un
.sun_path
= "/run/systemd/cgroups-agent",
760 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
761 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
762 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
763 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
764 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
765 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
766 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
767 * we thus won't lose messages.
769 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
770 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
771 * bus for these messages. */
776 if (!MANAGER_IS_SYSTEM(m
))
779 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
781 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
782 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
785 if (m
->cgroups_agent_fd
< 0) {
786 _cleanup_close_
int fd
= -1;
788 /* First free all secondary fields */
789 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
791 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
793 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
795 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
797 (void) unlink(sa
.un
.sun_path
);
799 /* Only allow root to connect to this socket */
801 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
803 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
805 m
->cgroups_agent_fd
= fd
;
809 if (!m
->cgroups_agent_event_source
) {
810 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
812 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
814 /* Process cgroups notifications early, but after having processed service notification messages or
815 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
816 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
817 * cgroup inotify for the unified cgroup stuff. */
818 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
820 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
822 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
828 static int manager_setup_user_lookup_fd(Manager
*m
) {
833 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
834 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
835 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
836 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
837 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
838 * hence we establish this communication channel so that forked off processes can pass their UID/GID
839 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
840 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
843 * You might wonder why we need a communication channel for this that is independent of the usual notification
844 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
845 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
846 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
848 * Note that this function is called under two circumstances: when we first initialize (in which case we
849 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
850 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
852 if (m
->user_lookup_fds
[0] < 0) {
854 /* Free all secondary fields */
855 safe_close_pair(m
->user_lookup_fds
);
856 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
858 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
859 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
861 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
864 if (!m
->user_lookup_event_source
) {
865 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
867 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
869 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
871 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
873 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
875 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
881 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
882 bool try_bus_connect
;
891 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
893 /* Try to connect to the buses, if possible. */
894 return bus_init(m
, try_bus_connect
);
897 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
903 while ((u
= m
->cleanup_queue
)) {
904 assert(u
->in_cleanup_queue
);
914 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
915 GC_OFFSET_UNSURE
, /* No clue */
916 GC_OFFSET_GOOD
, /* We still need this unit */
917 GC_OFFSET_BAD
, /* We don't need this unit anymore */
921 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
925 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
927 /* Recursively mark referenced units as GOOD as well */
928 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
929 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
930 unit_gc_mark_good(other
, gc_marker
);
933 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
940 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
941 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
942 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
943 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
946 if (u
->in_cleanup_queue
)
949 if (unit_check_gc(u
))
952 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
956 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
957 unit_gc_sweep(other
, gc_marker
);
959 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
962 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
969 /* We were unable to find anything out about this entry, so
970 * let's investigate it later */
971 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
972 unit_add_to_gc_queue(u
);
976 /* We definitely know that this one is not useful anymore, so
977 * let's mark it for deletion */
978 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
979 unit_add_to_cleanup_queue(u
);
983 unit_gc_mark_good(u
, gc_marker
);
986 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
987 unsigned n
= 0, gc_marker
;
992 /* log_debug("Running GC..."); */
994 m
->gc_marker
+= _GC_OFFSET_MAX
;
995 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
998 gc_marker
= m
->gc_marker
;
1000 while ((u
= m
->gc_unit_queue
)) {
1001 assert(u
->in_gc_queue
);
1003 unit_gc_sweep(u
, gc_marker
);
1005 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1006 u
->in_gc_queue
= false;
1010 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1011 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1013 log_unit_debug(u
, "Collecting.");
1014 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1015 unit_add_to_cleanup_queue(u
);
1022 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1028 while ((j
= m
->gc_job_queue
)) {
1029 assert(j
->in_gc_queue
);
1031 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1032 j
->in_gc_queue
= false;
1036 if (job_check_gc(j
))
1039 log_unit_debug(j
->unit
, "Collecting job.");
1040 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1046 static void manager_clear_jobs_and_units(Manager
*m
) {
1051 while ((u
= hashmap_first(m
->units
)))
1054 manager_dispatch_cleanup_queue(m
);
1056 assert(!m
->load_queue
);
1057 assert(!m
->run_queue
);
1058 assert(!m
->dbus_unit_queue
);
1059 assert(!m
->dbus_job_queue
);
1060 assert(!m
->cleanup_queue
);
1061 assert(!m
->gc_unit_queue
);
1062 assert(!m
->gc_job_queue
);
1064 assert(hashmap_isempty(m
->jobs
));
1065 assert(hashmap_isempty(m
->units
));
1067 m
->n_on_console
= 0;
1068 m
->n_running_jobs
= 0;
1071 Manager
* manager_free(Manager
*m
) {
1078 manager_clear_jobs_and_units(m
);
1080 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1081 if (unit_vtable
[c
]->shutdown
)
1082 unit_vtable
[c
]->shutdown(m
);
1084 /* If we reexecute ourselves, we keep the root cgroup
1086 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1088 lookup_paths_flush_generator(&m
->lookup_paths
);
1092 dynamic_user_vacuum(m
, false);
1093 hashmap_free(m
->dynamic_users
);
1095 hashmap_free(m
->units
);
1096 hashmap_free(m
->units_by_invocation_id
);
1097 hashmap_free(m
->jobs
);
1098 hashmap_free(m
->watch_pids1
);
1099 hashmap_free(m
->watch_pids2
);
1100 hashmap_free(m
->watch_bus
);
1102 set_free(m
->startup_units
);
1103 set_free(m
->failed_units
);
1105 sd_event_source_unref(m
->signal_event_source
);
1106 sd_event_source_unref(m
->notify_event_source
);
1107 sd_event_source_unref(m
->cgroups_agent_event_source
);
1108 sd_event_source_unref(m
->time_change_event_source
);
1109 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1110 sd_event_source_unref(m
->run_queue_event_source
);
1111 sd_event_source_unref(m
->user_lookup_event_source
);
1113 safe_close(m
->signal_fd
);
1114 safe_close(m
->notify_fd
);
1115 safe_close(m
->cgroups_agent_fd
);
1116 safe_close(m
->time_change_fd
);
1117 safe_close_pair(m
->user_lookup_fds
);
1119 manager_close_ask_password(m
);
1121 manager_close_idle_pipe(m
);
1123 udev_unref(m
->udev
);
1124 sd_event_unref(m
->event
);
1126 free(m
->notify_socket
);
1128 lookup_paths_free(&m
->lookup_paths
);
1129 strv_free(m
->environment
);
1131 hashmap_free(m
->cgroup_unit
);
1132 set_free_free(m
->unit_path_cache
);
1134 free(m
->switch_root
);
1135 free(m
->switch_root_init
);
1137 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1138 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1140 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1141 hashmap_free(m
->units_requiring_mounts_for
);
1143 hashmap_free(m
->uid_refs
);
1144 hashmap_free(m
->gid_refs
);
1149 void manager_enumerate(Manager
*m
) {
1154 /* Let's ask every type to load all units from disk/kernel
1155 * that it might know */
1156 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1157 if (!unit_type_supported(c
)) {
1158 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1162 if (!unit_vtable
[c
]->enumerate
)
1165 unit_vtable
[c
]->enumerate(m
);
1168 manager_dispatch_load_queue(m
);
1171 static void manager_coldplug(Manager
*m
) {
1179 /* Then, let's set up their initial state. */
1180 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1182 /* ignore aliases */
1186 r
= unit_coldplug(u
);
1188 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1192 static void manager_build_unit_path_cache(Manager
*m
) {
1198 set_free_free(m
->unit_path_cache
);
1200 m
->unit_path_cache
= set_new(&string_hash_ops
);
1201 if (!m
->unit_path_cache
) {
1206 /* This simply builds a list of files we know exist, so that
1207 * we don't always have to go to disk */
1209 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1210 _cleanup_closedir_
DIR *d
= NULL
;
1215 if (errno
!= ENOENT
)
1216 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1220 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1223 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1229 r
= set_consume(m
->unit_path_cache
, p
);
1238 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1239 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1242 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1248 HASHMAP_FOREACH(u
, m
->units
, i
) {
1250 if (fdset_size(fds
) <= 0)
1253 if (!UNIT_VTABLE(u
)->distribute_fds
)
1256 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1260 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1265 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1269 r
= manager_run_environment_generators(m
);
1273 /* Make sure the transient directory always exists, so that it remains in the search path */
1275 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1280 dual_timestamp_get(&m
->generators_start_timestamp
);
1281 r
= manager_run_generators(m
);
1282 dual_timestamp_get(&m
->generators_finish_timestamp
);
1286 lookup_paths_reduce(&m
->lookup_paths
);
1287 manager_build_unit_path_cache(m
);
1289 /* If we will deserialize make sure that during enumeration
1290 * this is already known, so we increase the counter here
1295 /* First, enumerate what we can from all config files */
1296 dual_timestamp_get(&m
->units_load_start_timestamp
);
1297 manager_enumerate(m
);
1298 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1300 /* Second, deserialize if there is something to deserialize */
1302 r
= manager_deserialize(m
, serialization
, fds
);
1304 /* Any fds left? Find some unit which wants them. This is
1305 * useful to allow container managers to pass some file
1306 * descriptors to us pre-initialized. This enables
1307 * socket-based activation of entire containers. */
1308 manager_distribute_fds(m
, fds
);
1310 /* We might have deserialized the notify fd, but if we didn't
1311 * then let's create the bus now */
1312 q
= manager_setup_notify(m
);
1313 if (q
< 0 && r
== 0)
1316 q
= manager_setup_cgroups_agent(m
);
1317 if (q
< 0 && r
== 0)
1320 q
= manager_setup_user_lookup_fd(m
);
1321 if (q
< 0 && r
== 0)
1324 /* Let's connect to the bus now. */
1325 (void) manager_connect_bus(m
, !!serialization
);
1327 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1328 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1330 /* Third, fire things up! */
1331 manager_coldplug(m
);
1333 /* Release any dynamic users no longer referenced */
1334 dynamic_user_vacuum(m
, true);
1336 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1337 manager_vacuum_uid_refs(m
);
1338 manager_vacuum_gid_refs(m
);
1340 if (serialization
) {
1341 assert(m
->n_reloading
> 0);
1344 /* Let's wait for the UnitNew/JobNew messages being
1345 * sent, before we notify that the reload is
1347 m
->send_reloading_done
= true;
1353 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1358 assert(type
< _JOB_TYPE_MAX
);
1360 assert(mode
< _JOB_MODE_MAX
);
1362 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1363 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1365 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1366 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1368 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1370 type
= job_type_collapse(type
, unit
);
1372 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1376 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1377 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1378 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1382 if (mode
== JOB_ISOLATE
) {
1383 r
= transaction_add_isolate_jobs(tr
, m
);
1388 r
= transaction_activate(tr
, m
, mode
, e
);
1392 log_unit_debug(unit
,
1393 "Enqueued job %s/%s as %u", unit
->id
,
1394 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1397 *_ret
= tr
->anchor_job
;
1399 transaction_free(tr
);
1403 transaction_abort(tr
);
1404 transaction_free(tr
);
1408 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1409 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1413 assert(type
< _JOB_TYPE_MAX
);
1415 assert(mode
< _JOB_MODE_MAX
);
1417 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1422 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1425 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1426 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1430 assert(type
< _JOB_TYPE_MAX
);
1432 assert(mode
< _JOB_MODE_MAX
);
1434 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1436 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1441 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1444 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1447 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1451 return hashmap_get(m
->units
, name
);
1454 unsigned manager_dispatch_load_queue(Manager
*m
) {
1460 /* Make sure we are not run recursively */
1461 if (m
->dispatching_load_queue
)
1464 m
->dispatching_load_queue
= true;
1466 /* Dispatches the load queue. Takes a unit from the queue and
1467 * tries to load its data until the queue is empty */
1469 while ((u
= m
->load_queue
)) {
1470 assert(u
->in_load_queue
);
1476 m
->dispatching_load_queue
= false;
1480 int manager_load_unit_prepare(
1492 assert(name
|| path
);
1495 /* This will prepare the unit for loading, but not actually
1496 * load anything from disk. */
1498 if (path
&& !is_path(path
))
1499 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1502 name
= basename(path
);
1504 t
= unit_name_to_type(name
);
1506 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1507 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1508 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1510 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1513 ret
= manager_get_unit(m
, name
);
1519 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1524 ret
->fragment_path
= strdup(path
);
1525 if (!ret
->fragment_path
) {
1531 r
= unit_add_name(ret
, name
);
1537 unit_add_to_load_queue(ret
);
1538 unit_add_to_dbus_queue(ret
);
1539 unit_add_to_gc_queue(ret
);
1546 int manager_load_unit(
1558 /* This will load the service information files, but not actually
1559 * start any services or anything. */
1561 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1565 manager_dispatch_load_queue(m
);
1567 *_ret
= unit_follow_merge(*_ret
);
1572 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1579 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1580 job_dump(j
, f
, prefix
);
1583 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1591 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1593 unit_dump(u
, f
, prefix
);
1596 void manager_clear_jobs(Manager
*m
) {
1601 while ((j
= hashmap_first(m
->jobs
)))
1602 /* No need to recurse. We're cancelling all jobs. */
1603 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1606 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1607 Manager
*m
= userdata
;
1613 while ((j
= m
->run_queue
)) {
1614 assert(j
->installed
);
1615 assert(j
->in_run_queue
);
1617 job_run_and_invalidate(j
);
1620 if (m
->n_running_jobs
> 0)
1621 manager_watch_jobs_in_progress(m
);
1623 if (m
->n_on_console
> 0)
1624 manager_watch_idle_pipe(m
);
1629 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1636 if (m
->dispatching_dbus_queue
)
1639 m
->dispatching_dbus_queue
= true;
1641 while ((u
= m
->dbus_unit_queue
)) {
1642 assert(u
->in_dbus_queue
);
1644 bus_unit_send_change_signal(u
);
1648 while ((j
= m
->dbus_job_queue
)) {
1649 assert(j
->in_dbus_queue
);
1651 bus_job_send_change_signal(j
);
1655 m
->dispatching_dbus_queue
= false;
1657 if (m
->send_reloading_done
) {
1658 m
->send_reloading_done
= false;
1660 bus_manager_send_reloading(m
, false);
1663 if (m
->queued_message
)
1664 bus_send_queued_message(m
);
1669 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1670 Manager
*m
= userdata
;
1671 char buf
[PATH_MAX
+1];
1674 n
= recv(fd
, buf
, sizeof(buf
), 0);
1676 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1678 log_error("Got zero-length cgroups agent message, ignoring.");
1681 if ((size_t) n
>= sizeof(buf
)) {
1682 log_error("Got overly long cgroups agent message, ignoring.");
1686 if (memchr(buf
, 0, n
)) {
1687 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1692 manager_notify_cgroup_empty(m
, buf
);
1693 bus_forward_agent_released(m
, buf
);
1698 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1699 _cleanup_strv_free_
char **tags
= NULL
;
1705 tags
= strv_split(buf
, "\n\r");
1711 if (UNIT_VTABLE(u
)->notify_message
)
1712 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1713 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1714 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1718 y
= ellipsize(x
, 20, 90);
1719 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1723 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1725 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1726 Manager
*m
= userdata
;
1727 char buf
[NOTIFY_BUFFER_MAX
+1];
1728 struct iovec iovec
= {
1730 .iov_len
= sizeof(buf
)-1,
1733 struct cmsghdr cmsghdr
;
1734 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1735 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1737 struct msghdr msghdr
= {
1740 .msg_control
= &control
,
1741 .msg_controllen
= sizeof(control
),
1744 struct cmsghdr
*cmsg
;
1745 struct ucred
*ucred
= NULL
;
1747 int r
, *fd_array
= NULL
;
1752 assert(m
->notify_fd
== fd
);
1754 if (revents
!= EPOLLIN
) {
1755 log_warning("Got unexpected poll event for notify fd.");
1759 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1761 if (IN_SET(errno
, EAGAIN
, EINTR
))
1762 return 0; /* Spurious wakeup, try again */
1764 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1765 * won't take notification messages anymore, but that's still better than busy looping around this:
1766 * being woken up over and over again but being unable to actually read the message off the socket. */
1767 return log_error_errno(errno
, "Failed to receive notification message: %m");
1770 CMSG_FOREACH(cmsg
, &msghdr
) {
1771 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1773 fd_array
= (int*) CMSG_DATA(cmsg
);
1774 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1776 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1777 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1778 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1780 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1787 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1789 close_many(fd_array
, n_fds
);
1795 if (!ucred
|| ucred
->pid
<= 0) {
1796 log_warning("Received notify message without valid credentials. Ignoring.");
1800 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1801 log_warning("Received notify message exceeded maximum size. Ignoring.");
1805 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1806 * trailing NUL byte in the message, but don't expect it. */
1807 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1808 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1812 /* Make sure it's NUL-terminated. */
1815 /* Notify every unit that might be interested, but try
1816 * to avoid notifying the same one multiple times. */
1817 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1819 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1821 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1823 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1825 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1826 if (u3
&& u3
!= u2
&& u3
!= u1
)
1827 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1829 if (!u1
&& !u2
&& !u3
)
1830 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1832 if (fdset_size(fds
) > 0)
1833 log_warning("Got extra auxiliary fds with notification message, closing them.");
1838 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1845 sd_event_get_iteration(m
->event
, &iteration
);
1847 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1849 unit_unwatch_pid(u
, si
->si_pid
);
1851 if (UNIT_VTABLE(u
)->sigchld_event
) {
1852 if (set_size(u
->pids
) <= 1 ||
1853 iteration
!= u
->sigchldgen
||
1854 unit_main_pid(u
) == si
->si_pid
||
1855 unit_control_pid(u
) == si
->si_pid
) {
1856 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1857 u
->sigchldgen
= iteration
;
1859 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1863 static int manager_dispatch_sigchld(Manager
*m
) {
1869 /* First we call waitd() for a PID and do not reap the
1870 * zombie. That way we can still access /proc/$PID for
1871 * it while it is a zombie. */
1872 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1874 if (errno
== ECHILD
)
1886 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1887 _cleanup_free_
char *name
= NULL
;
1890 get_process_comm(si
.si_pid
, &name
);
1892 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1893 si
.si_pid
, strna(name
),
1894 sigchld_code_to_string(si
.si_code
),
1896 strna(si
.si_code
== CLD_EXITED
1897 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1898 : signal_to_string(si
.si_status
)));
1900 /* And now figure out the unit this belongs
1901 * to, it might be multiple... */
1902 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1904 invoke_sigchld_event(m
, u1
, &si
);
1905 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1907 invoke_sigchld_event(m
, u2
, &si
);
1908 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1909 if (u3
&& u3
!= u2
&& u3
!= u1
)
1910 invoke_sigchld_event(m
, u3
, &si
);
1913 /* And now, we actually reap the zombie. */
1914 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1925 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1926 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1929 log_debug("Activating special unit %s", name
);
1931 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1933 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1938 static void manager_handle_ctrl_alt_del(Manager
*m
) {
1939 /* If the user presses C-A-D more than
1940 * 7 times within 2s, we reboot/shutdown immediately,
1941 * unless it was disabled in system.conf */
1943 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
1944 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1946 emergency_action(m
, m
->cad_burst_action
, NULL
,
1947 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1950 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1951 Manager
*m
= userdata
;
1953 struct signalfd_siginfo sfsi
;
1954 bool sigchld
= false;
1958 assert(m
->signal_fd
== fd
);
1960 if (revents
!= EPOLLIN
) {
1961 log_warning("Got unexpected events from signal file descriptor.");
1966 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1967 if (n
!= sizeof(sfsi
)) {
1969 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
1973 if (IN_SET(errno
, EINTR
, EAGAIN
))
1976 /* We return an error here, which will kill this handler,
1977 * to avoid a busy loop on read error. */
1978 return log_error_errno(errno
, "Reading from signal fd failed: %m");
1981 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1982 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1983 ? LOG_DEBUG
: LOG_INFO
,
1986 switch (sfsi
.ssi_signo
) {
1993 if (MANAGER_IS_SYSTEM(m
)) {
1994 /* This is for compatibility with the
1995 * original sysvinit */
1996 r
= verify_run_space_and_log("Refusing to reexecute");
1998 m
->exit_code
= MANAGER_REEXECUTE
;
2005 if (MANAGER_IS_SYSTEM(m
)) {
2006 manager_handle_ctrl_alt_del(m
);
2010 /* Run the exit target if there is one, if not, just exit. */
2011 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
2012 m
->exit_code
= MANAGER_EXIT
;
2019 if (MANAGER_IS_SYSTEM(m
))
2020 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2022 /* This is a nop on non-init */
2026 if (MANAGER_IS_SYSTEM(m
))
2027 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2029 /* This is a nop on non-init */
2035 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2037 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2038 log_info("Trying to reconnect to bus...");
2042 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2043 log_info("Loading D-Bus service...");
2044 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2051 _cleanup_free_
char *dump
= NULL
;
2052 _cleanup_fclose_
FILE *f
= NULL
;
2055 f
= open_memstream(&dump
, &size
);
2057 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2061 manager_dump_units(m
, f
, "\t");
2062 manager_dump_jobs(m
, f
, "\t");
2064 r
= fflush_and_check(f
);
2066 log_warning_errno(r
, "Failed to write status stream: %m");
2070 log_dump(LOG_INFO
, dump
);
2075 r
= verify_run_space_and_log("Refusing to reload");
2077 m
->exit_code
= MANAGER_RELOAD
;
2082 /* Starting SIGRTMIN+0 */
2083 static const char * const target_table
[] = {
2084 [0] = SPECIAL_DEFAULT_TARGET
,
2085 [1] = SPECIAL_RESCUE_TARGET
,
2086 [2] = SPECIAL_EMERGENCY_TARGET
,
2087 [3] = SPECIAL_HALT_TARGET
,
2088 [4] = SPECIAL_POWEROFF_TARGET
,
2089 [5] = SPECIAL_REBOOT_TARGET
,
2090 [6] = SPECIAL_KEXEC_TARGET
2093 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2094 static const ManagerExitCode code_table
[] = {
2096 [1] = MANAGER_POWEROFF
,
2097 [2] = MANAGER_REBOOT
,
2101 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2102 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2103 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2104 manager_start_target(m
, target_table
[idx
],
2105 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
2109 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2110 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2111 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2115 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2118 manager_set_show_status(m
, SHOW_STATUS_YES
);
2122 manager_set_show_status(m
, SHOW_STATUS_NO
);
2126 log_set_max_level(LOG_DEBUG
);
2127 log_info("Setting log level to debug.");
2131 log_set_max_level(LOG_INFO
);
2132 log_info("Setting log level to info.");
2136 if (MANAGER_IS_USER(m
)) {
2137 m
->exit_code
= MANAGER_EXIT
;
2141 /* This is a nop on init */
2145 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2146 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2147 log_notice("Setting log target to journal-or-kmsg.");
2151 log_set_target(LOG_TARGET_CONSOLE
);
2152 log_notice("Setting log target to console.");
2156 log_set_target(LOG_TARGET_KMSG
);
2157 log_notice("Setting log target to kmsg.");
2161 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2168 manager_dispatch_sigchld(m
);
2173 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2174 Manager
*m
= userdata
;
2179 assert(m
->time_change_fd
== fd
);
2181 log_struct(LOG_DEBUG
,
2182 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2183 LOG_MESSAGE("Time has been changed"),
2186 /* Restart the watch */
2187 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2188 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2190 manager_setup_time_change(m
);
2192 HASHMAP_FOREACH(u
, m
->units
, i
)
2193 if (UNIT_VTABLE(u
)->time_change
)
2194 UNIT_VTABLE(u
)->time_change(u
);
2199 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2200 Manager
*m
= userdata
;
2203 assert(m
->idle_pipe
[2] == fd
);
2205 m
->no_console_output
= m
->n_on_console
> 0;
2207 manager_close_idle_pipe(m
);
2212 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2213 Manager
*m
= userdata
;
2220 manager_print_jobs_in_progress(m
);
2222 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2223 r
= sd_event_source_set_time(source
, next
);
2227 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2230 int manager_loop(Manager
*m
) {
2233 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2236 m
->exit_code
= MANAGER_OK
;
2238 /* Release the path cache */
2239 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2241 manager_check_finished(m
);
2243 /* There might still be some zombies hanging around from
2244 * before we were exec()'ed. Let's reap them. */
2245 r
= manager_dispatch_sigchld(m
);
2249 while (m
->exit_code
== MANAGER_OK
) {
2252 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2255 if (!ratelimit_test(&rl
)) {
2256 /* Yay, something is going seriously wrong, pause a little */
2257 log_warning("Looping too fast. Throttling execution a little.");
2261 if (manager_dispatch_load_queue(m
) > 0)
2264 if (manager_dispatch_gc_job_queue(m
) > 0)
2267 if (manager_dispatch_gc_unit_queue(m
) > 0)
2270 if (manager_dispatch_cleanup_queue(m
) > 0)
2273 if (manager_dispatch_cgroup_queue(m
) > 0)
2276 if (manager_dispatch_dbus_queue(m
) > 0)
2279 /* Sleep for half the watchdog time */
2280 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2281 wait_usec
= m
->runtime_watchdog
/ 2;
2285 wait_usec
= USEC_INFINITY
;
2287 r
= sd_event_run(m
->event
, wait_usec
);
2289 return log_error_errno(r
, "Failed to run event loop: %m");
2292 return m
->exit_code
;
2295 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2296 _cleanup_free_
char *n
= NULL
;
2297 sd_id128_t invocation_id
;
2305 r
= unit_name_from_dbus_path(s
, &n
);
2309 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2310 * as invocation ID. */
2311 r
= sd_id128_from_string(n
, &invocation_id
);
2313 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2319 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
));
2322 /* If this didn't work, we use the suffix as unit name. */
2323 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2331 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2341 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2345 r
= safe_atou(p
, &id
);
2349 j
= manager_get_job(m
, id
);
2358 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2361 _cleanup_free_
char *p
= NULL
;
2365 if (!MANAGER_IS_SYSTEM(m
))
2368 audit_fd
= get_audit_fd();
2372 /* Don't generate audit events if the service was already
2373 * started and we're just deserializing */
2374 if (MANAGER_IS_RELOADING(m
))
2377 if (u
->type
!= UNIT_SERVICE
)
2380 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2382 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2386 msg
= strjoina("unit=", p
);
2387 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2389 /* We aren't allowed to send audit messages?
2390 * Then let's not retry again. */
2393 log_warning_errno(errno
, "Failed to send audit message: %m");
2399 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2400 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2401 _cleanup_free_
char *message
= NULL
;
2402 _cleanup_close_
int fd
= -1;
2405 /* Don't generate plymouth events if the service was already
2406 * started and we're just deserializing */
2407 if (MANAGER_IS_RELOADING(m
))
2410 if (!MANAGER_IS_SYSTEM(m
))
2413 if (detect_container() > 0)
2416 if (u
->type
!= UNIT_SERVICE
&&
2417 u
->type
!= UNIT_MOUNT
&&
2418 u
->type
!= UNIT_SWAP
)
2421 /* We set SOCK_NONBLOCK here so that we rather drop the
2422 * message then wait for plymouth */
2423 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2425 log_error_errno(errno
, "socket() failed: %m");
2429 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2431 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2432 log_error_errno(errno
, "connect() failed: %m");
2436 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2442 if (write(fd
, message
, n
+ 1) != n
+ 1)
2443 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2444 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2447 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2453 fd
= open_serialization_fd("systemd-state");
2457 f
= fdopen(fd
, "w+");
2467 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2479 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2480 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2481 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2482 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2484 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2485 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2486 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2487 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2490 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2491 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2492 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2493 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2494 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2495 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2496 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2497 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2500 if (!switching_root
)
2501 (void) serialize_environment(f
, m
->environment
);
2503 if (m
->notify_fd
>= 0) {
2506 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2510 fprintf(f
, "notify-fd=%i\n", copy
);
2511 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2514 if (m
->cgroups_agent_fd
>= 0) {
2517 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2521 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2524 if (m
->user_lookup_fds
[0] >= 0) {
2527 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2531 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2535 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2538 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2540 r
= dynamic_user_serialize(m
, f
, fds
);
2544 manager_serialize_uid_refs(m
, f
);
2545 manager_serialize_gid_refs(m
, f
);
2549 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2557 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2564 assert(m
->n_reloading
> 0);
2570 r
= bus_fdset_add_all(m
, fds
);
2577 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2583 log_debug("Deserializing state...");
2588 char line
[LINE_MAX
], *l
;
2591 if (!fgets(line
, sizeof(line
), f
)) {
2606 if ((val
= startswith(l
, "current-job-id="))) {
2609 if (safe_atou32(val
, &id
) < 0)
2610 log_debug("Failed to parse current job id value %s", val
);
2612 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2614 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2617 if (safe_atou32(val
, &n
) < 0)
2618 log_debug("Failed to parse installed jobs counter %s", val
);
2620 m
->n_installed_jobs
+= n
;
2622 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2625 if (safe_atou32(val
, &n
) < 0)
2626 log_debug("Failed to parse failed jobs counter %s", val
);
2628 m
->n_failed_jobs
+= n
;
2630 } else if ((val
= startswith(l
, "taint-usr="))) {
2633 b
= parse_boolean(val
);
2635 log_debug("Failed to parse taint /usr flag %s", val
);
2637 m
->taint_usr
= m
->taint_usr
|| b
;
2639 } else if ((val
= startswith(l
, "firmware-timestamp=")))
2640 dual_timestamp_deserialize(val
, &m
->firmware_timestamp
);
2641 else if ((val
= startswith(l
, "loader-timestamp=")))
2642 dual_timestamp_deserialize(val
, &m
->loader_timestamp
);
2643 else if ((val
= startswith(l
, "kernel-timestamp=")))
2644 dual_timestamp_deserialize(val
, &m
->kernel_timestamp
);
2645 else if ((val
= startswith(l
, "initrd-timestamp=")))
2646 dual_timestamp_deserialize(val
, &m
->initrd_timestamp
);
2647 else if ((val
= startswith(l
, "userspace-timestamp=")))
2648 dual_timestamp_deserialize(val
, &m
->userspace_timestamp
);
2649 else if ((val
= startswith(l
, "finish-timestamp=")))
2650 dual_timestamp_deserialize(val
, &m
->finish_timestamp
);
2651 else if ((val
= startswith(l
, "security-start-timestamp=")))
2652 dual_timestamp_deserialize(val
, &m
->security_start_timestamp
);
2653 else if ((val
= startswith(l
, "security-finish-timestamp=")))
2654 dual_timestamp_deserialize(val
, &m
->security_finish_timestamp
);
2655 else if ((val
= startswith(l
, "generators-start-timestamp=")))
2656 dual_timestamp_deserialize(val
, &m
->generators_start_timestamp
);
2657 else if ((val
= startswith(l
, "generators-finish-timestamp=")))
2658 dual_timestamp_deserialize(val
, &m
->generators_finish_timestamp
);
2659 else if ((val
= startswith(l
, "units-load-start-timestamp=")))
2660 dual_timestamp_deserialize(val
, &m
->units_load_start_timestamp
);
2661 else if ((val
= startswith(l
, "units-load-finish-timestamp=")))
2662 dual_timestamp_deserialize(val
, &m
->units_load_finish_timestamp
);
2663 else if (startswith(l
, "env=")) {
2664 r
= deserialize_environment(&m
->environment
, l
);
2668 } else if ((val
= startswith(l
, "notify-fd="))) {
2671 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2672 log_debug("Failed to parse notify fd: %s", val
);
2674 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2675 safe_close(m
->notify_fd
);
2676 m
->notify_fd
= fdset_remove(fds
, fd
);
2679 } else if ((val
= startswith(l
, "notify-socket="))) {
2688 free(m
->notify_socket
);
2689 m
->notify_socket
= n
;
2691 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2694 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2695 log_debug("Failed to parse cgroups agent fd: %s", val
);
2697 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2698 safe_close(m
->cgroups_agent_fd
);
2699 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2702 } else if ((val
= startswith(l
, "user-lookup="))) {
2705 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2706 log_debug("Failed to parse user lookup fd: %s", val
);
2708 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2709 safe_close_pair(m
->user_lookup_fds
);
2710 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2711 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2714 } else if ((val
= startswith(l
, "dynamic-user=")))
2715 dynamic_user_deserialize_one(m
, val
, fds
);
2716 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2717 manager_deserialize_uid_refs_one(m
, val
);
2718 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2719 manager_deserialize_gid_refs_one(m
, val
);
2720 else if ((val
= startswith(l
, "subscribed="))) {
2722 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2725 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2726 log_debug("Unknown serialization item '%s'", l
);
2731 char name
[UNIT_NAME_MAX
+2];
2734 if (!fgets(name
, sizeof(name
), f
)) {
2745 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2749 r
= unit_deserialize(u
, f
, fds
);
2758 assert(m
->n_reloading
> 0);
2764 int manager_reload(Manager
*m
) {
2766 _cleanup_fclose_
FILE *f
= NULL
;
2767 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2771 r
= manager_open_serialization(m
, &f
);
2776 bus_manager_send_reloading(m
, true);
2784 r
= manager_serialize(m
, f
, fds
, false);
2790 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2795 /* From here on there is no way back. */
2796 manager_clear_jobs_and_units(m
);
2797 lookup_paths_flush_generator(&m
->lookup_paths
);
2798 lookup_paths_free(&m
->lookup_paths
);
2799 dynamic_user_vacuum(m
, false);
2800 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2801 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2803 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2804 if (q
< 0 && r
>= 0)
2807 q
= manager_run_environment_generators(m
);
2808 if (q
< 0 && r
>= 0)
2811 /* Find new unit paths */
2812 q
= manager_run_generators(m
);
2813 if (q
< 0 && r
>= 0)
2816 lookup_paths_reduce(&m
->lookup_paths
);
2817 manager_build_unit_path_cache(m
);
2819 /* First, enumerate what we can from all config files */
2820 manager_enumerate(m
);
2822 /* Second, deserialize our stored data */
2823 q
= manager_deserialize(m
, f
, fds
);
2824 if (q
< 0 && r
>= 0)
2830 /* Re-register notify_fd as event source */
2831 q
= manager_setup_notify(m
);
2832 if (q
< 0 && r
>= 0)
2835 q
= manager_setup_cgroups_agent(m
);
2836 if (q
< 0 && r
>= 0)
2839 q
= manager_setup_user_lookup_fd(m
);
2840 if (q
< 0 && r
>= 0)
2843 /* Third, fire things up! */
2844 manager_coldplug(m
);
2846 /* Release any dynamic users no longer referenced */
2847 dynamic_user_vacuum(m
, true);
2849 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2850 manager_vacuum_uid_refs(m
);
2851 manager_vacuum_gid_refs(m
);
2853 /* Sync current state of bus names with our set of listening units */
2855 manager_sync_bus_names(m
, m
->api_bus
);
2857 assert(m
->n_reloading
> 0);
2860 m
->send_reloading_done
= true;
2865 void manager_reset_failed(Manager
*m
) {
2871 HASHMAP_FOREACH(u
, m
->units
, i
)
2872 unit_reset_failed(u
);
2875 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2881 /* Returns true if the unit is inactive or going down */
2882 u
= manager_get_unit(m
, name
);
2886 return unit_inactive_or_pending(u
);
2889 static void manager_notify_finished(Manager
*m
) {
2890 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2891 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2896 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2898 /* Note that m->kernel_usec.monotonic is always at 0,
2899 * and m->firmware_usec.monotonic and
2900 * m->loader_usec.monotonic should be considered
2901 * negative values. */
2903 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2904 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2905 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2906 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2908 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2910 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2911 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2913 log_struct(LOG_INFO
,
2914 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
2915 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2916 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2917 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2918 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2919 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2920 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2921 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2922 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2925 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2928 log_struct(LOG_INFO
,
2929 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
2930 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2931 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2932 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2933 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2934 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2935 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2939 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2940 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2942 log_struct(LOG_INFO
,
2943 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
2944 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2945 LOG_MESSAGE("Startup finished in %s.",
2946 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2950 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2954 "STATUS=Startup finished in %s.",
2955 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2958 void manager_check_finished(Manager
*m
) {
2961 if (MANAGER_IS_RELOADING(m
))
2964 /* Verify that we are actually running currently. Initially
2965 * the exit code is set to invalid, and during operation it is
2966 * then set to MANAGER_OK */
2967 if (m
->exit_code
!= MANAGER_OK
)
2970 if (hashmap_size(m
->jobs
) > 0) {
2971 if (m
->jobs_in_progress_event_source
)
2972 /* Ignore any failure, this is only for feedback */
2973 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2978 manager_flip_auto_status(m
, false);
2980 /* Notify Type=idle units that we are done now */
2981 manager_close_idle_pipe(m
);
2983 /* Turn off confirm spawn now */
2984 m
->confirm_spawn
= NULL
;
2986 /* No need to update ask password status when we're going non-interactive */
2987 manager_close_ask_password(m
);
2989 /* This is no longer the first boot */
2990 manager_set_first_boot(m
, false);
2992 if (dual_timestamp_is_set(&m
->finish_timestamp
))
2995 dual_timestamp_get(&m
->finish_timestamp
);
2997 manager_notify_finished(m
);
2999 manager_invalidate_startup_units(m
);
3002 static bool generator_path_any(const char* const* paths
) {
3006 /* Optimize by skipping the whole process by not creating output directories
3007 * if no generators are found. */
3008 STRV_FOREACH(path
, (char**) paths
)
3009 if (access(*path
, F_OK
) == 0)
3011 else if (errno
!= ENOENT
)
3012 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3017 static const char* system_env_generator_binary_paths
[] = {
3018 "/run/systemd/system-environment-generators",
3019 "/etc/systemd/system-environment-generators",
3020 "/usr/local/lib/systemd/system-environment-generators",
3021 SYSTEM_ENV_GENERATOR_PATH
,
3025 static const char* user_env_generator_binary_paths
[] = {
3026 "/run/systemd/user-environment-generators",
3027 "/etc/systemd/user-environment-generators",
3028 "/usr/local/lib/systemd/user-environment-generators",
3029 USER_ENV_GENERATOR_PATH
,
3033 static int manager_run_environment_generators(Manager
*m
) {
3034 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3036 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3041 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3043 if (!generator_path_any(paths
))
3046 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3049 static int manager_run_generators(Manager
*m
) {
3050 _cleanup_strv_free_
char **paths
= NULL
;
3051 const char *argv
[5];
3059 paths
= generator_binary_paths(m
->unit_file_scope
);
3063 if (!generator_path_any((const char* const*) paths
))
3066 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3070 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3071 argv
[1] = m
->lookup_paths
.generator
;
3072 argv
[2] = m
->lookup_paths
.generator_early
;
3073 argv
[3] = m
->lookup_paths
.generator_late
;
3076 RUN_WITH_UMASK(0022)
3077 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3078 NULL
, NULL
, (char**) argv
);
3081 lookup_paths_trim_generator(&m
->lookup_paths
);
3085 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3086 char **a
= NULL
, **b
= NULL
, **l
;
3091 if (!strv_isempty(minus
)) {
3092 a
= strv_env_delete(l
, 1, minus
);
3099 if (!strv_isempty(plus
)) {
3100 b
= strv_env_merge(2, l
, plus
);
3109 if (m
->environment
!= l
)
3110 strv_free(m
->environment
);
3117 manager_clean_environment(m
);
3118 strv_sort(m
->environment
);
3123 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3128 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3129 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3131 if (!default_rlimit
[i
])
3134 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3142 void manager_recheck_journal(Manager
*m
) {
3147 if (!MANAGER_IS_SYSTEM(m
))
3150 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3151 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3152 log_close_journal();
3156 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3157 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3158 log_close_journal();
3162 /* Hmm, OK, so the socket is fully up and the service is up
3163 * too, then let's make use of the thing. */
3167 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3169 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3171 if (!MANAGER_IS_SYSTEM(m
))
3174 if (m
->show_status
!= mode
)
3175 log_debug("%s showing of status.",
3176 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3177 m
->show_status
= mode
;
3180 (void) touch("/run/systemd/show-status");
3182 (void) unlink("/run/systemd/show-status");
3185 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3188 if (!MANAGER_IS_SYSTEM(m
))
3191 if (m
->no_console_output
)
3194 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3197 /* If we cannot find out the status properly, just proceed. */
3198 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3201 if (m
->show_status
> 0)
3207 const char *manager_get_confirm_spawn(Manager
*m
) {
3208 static int last_errno
= 0;
3209 const char *vc
= m
->confirm_spawn
;
3213 /* Here's the deal: we want to test the validity of the console but don't want
3214 * PID1 to go through the whole console process which might block. But we also
3215 * want to warn the user only once if something is wrong with the console so we
3216 * cannot do the sanity checks after spawning our children. So here we simply do
3217 * really basic tests to hopefully trap common errors.
3219 * If the console suddenly disappear at the time our children will really it
3220 * then they will simply fail to acquire it and a positive answer will be
3221 * assumed. New children will fallback to /dev/console though.
3223 * Note: TTYs are devices that can come and go any time, and frequently aren't
3224 * available yet during early boot (consider a USB rs232 dongle...). If for any
3225 * reason the configured console is not ready, we fallback to the default
3228 if (!vc
|| path_equal(vc
, "/dev/console"))
3235 if (!S_ISCHR(st
.st_mode
)) {
3243 if (last_errno
!= errno
) {
3245 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3247 return "/dev/console";
3250 void manager_set_first_boot(Manager
*m
, bool b
) {
3253 if (!MANAGER_IS_SYSTEM(m
))
3256 if (m
->first_boot
!= (int) b
) {
3258 (void) touch("/run/systemd/first-boot");
3260 (void) unlink("/run/systemd/first-boot");
3266 void manager_disable_confirm_spawn(void) {
3267 (void) touch("/run/systemd/confirm_spawn_disabled");
3270 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3271 if (!m
->confirm_spawn
)
3274 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3277 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3280 /* If m is NULL, assume we're after shutdown and let the messages through. */
3282 if (m
&& !manager_get_show_status(m
, type
))
3285 /* XXX We should totally drop the check for ephemeral here
3286 * and thus effectively make 'Type=idle' pointless. */
3287 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3290 va_start(ap
, format
);
3291 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3295 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3296 char p
[strlen(path
)+1];
3302 path_kill_slashes(p
);
3304 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3307 const char *manager_get_runtime_prefix(Manager
*m
) {
3310 return MANAGER_IS_SYSTEM(m
) ?
3312 getenv("XDG_RUNTIME_DIR");
3315 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3320 assert(u
->manager
== m
);
3322 size
= set_size(m
->failed_units
);
3325 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3329 if (set_put(m
->failed_units
, u
) < 0)
3332 (void) set_remove(m
->failed_units
, u
);
3334 if (set_size(m
->failed_units
) != size
)
3335 bus_manager_send_change_signal(m
);
3340 ManagerState
manager_state(Manager
*m
) {
3345 /* Did we ever finish booting? If not then we are still starting up */
3346 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3348 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3349 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3350 return MANAGER_INITIALIZING
;
3352 return MANAGER_STARTING
;
3355 /* Is the special shutdown target queued? If so, we are in shutdown state */
3356 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3357 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3358 return MANAGER_STOPPING
;
3360 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3361 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3362 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3363 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3364 return MANAGER_MAINTENANCE
;
3366 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3367 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3368 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3369 return MANAGER_MAINTENANCE
;
3371 /* Are there any failed units? If so, we are in degraded mode */
3372 if (set_size(m
->failed_units
) > 0)
3373 return MANAGER_DEGRADED
;
3375 return MANAGER_RUNNING
;
3378 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3380 static void manager_unref_uid_internal(
3385 int (*_clean_ipc
)(uid_t uid
)) {
3391 assert(uid_is_valid(uid
));
3394 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3395 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3397 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3398 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3399 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3400 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3402 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3403 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3405 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3408 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3410 n
= c
& ~DESTROY_IPC_FLAG
;
3414 if (destroy_now
&& n
== 0) {
3415 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3417 if (c
& DESTROY_IPC_FLAG
) {
3418 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3419 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3421 (void) _clean_ipc(uid
);
3424 c
= n
| (c
& DESTROY_IPC_FLAG
);
3425 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3429 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3430 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3433 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3434 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3437 static int manager_ref_uid_internal(
3448 assert(uid_is_valid(uid
));
3450 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3451 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3453 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3454 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3456 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3459 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3463 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3465 n
= c
& ~DESTROY_IPC_FLAG
;
3468 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3471 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3473 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3476 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3477 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3480 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3481 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3484 static void manager_vacuum_uid_refs_internal(
3487 int (*_clean_ipc
)(uid_t uid
)) {
3496 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3500 uid
= PTR_TO_UID(k
);
3501 c
= PTR_TO_UINT32(p
);
3503 n
= c
& ~DESTROY_IPC_FLAG
;
3507 if (c
& DESTROY_IPC_FLAG
) {
3508 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3509 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3511 (void) _clean_ipc(uid
);
3514 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3518 void manager_vacuum_uid_refs(Manager
*m
) {
3519 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3522 void manager_vacuum_gid_refs(Manager
*m
) {
3523 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3526 static void manager_serialize_uid_refs_internal(
3530 const char *field_name
) {
3540 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3541 * of it is better rebuild after a reload/reexec. */
3543 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3547 uid
= PTR_TO_UID(k
);
3548 c
= PTR_TO_UINT32(p
);
3550 if (!(c
& DESTROY_IPC_FLAG
))
3553 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3557 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3558 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3561 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3562 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3565 static void manager_deserialize_uid_refs_one_internal(
3568 const char *value
) {
3578 r
= parse_uid(value
, &uid
);
3579 if (r
< 0 || uid
== 0) {
3580 log_debug("Unable to parse UID reference serialization");
3584 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3590 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3591 if (c
& DESTROY_IPC_FLAG
)
3594 c
|= DESTROY_IPC_FLAG
;
3596 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3598 log_debug("Failed to add UID reference entry");
3603 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3604 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3607 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3608 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3611 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3615 char unit_name
[UNIT_NAME_MAX
+1];
3618 Manager
*m
= userdata
;
3626 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3627 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3628 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3630 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3632 if (errno
== EINTR
|| errno
== EAGAIN
)
3635 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3638 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3639 log_warning("Received too short user lookup message, ignoring.");
3643 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3644 log_warning("Received too long user lookup message, ignoring.");
3648 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3649 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3653 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3654 if (memchr(buffer
.unit_name
, 0, n
)) {
3655 log_warning("Received lookup message with embedded NUL character, ignoring.");
3659 buffer
.unit_name
[n
] = 0;
3660 u
= manager_get_unit(m
, buffer
.unit_name
);
3662 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3666 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3668 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3672 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3673 [MANAGER_INITIALIZING
] = "initializing",
3674 [MANAGER_STARTING
] = "starting",
3675 [MANAGER_RUNNING
] = "running",
3676 [MANAGER_DEGRADED
] = "degraded",
3677 [MANAGER_MAINTENANCE
] = "maintenance",
3678 [MANAGER_STOPPING
] = "stopping",
3681 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);