1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <stdio_ext.h>
27 #include <sys/epoll.h>
28 #include <sys/inotify.h>
29 #include <sys/ioctl.h>
30 #include <sys/reboot.h>
31 #include <sys/timerfd.h>
39 #include "sd-daemon.h"
40 #include "sd-messages.h"
43 #include "alloc-util.h"
45 #include "boot-timestamps.h"
46 #include "bus-common-errors.h"
47 #include "bus-error.h"
48 #include "bus-kernel.h"
50 #include "clean-ipc.h"
51 #include "clock-util.h"
53 #include "dbus-manager.h"
54 #include "dbus-unit.h"
56 #include "dirent-util.h"
59 #include "exec-util.h"
61 #include "exit-status.h"
68 #include "locale-setup.h"
74 #include "parse-util.h"
75 #include "path-lookup.h"
76 #include "path-util.h"
77 #include "process-util.h"
78 #include "ratelimit.h"
80 #include "signal-util.h"
82 #include "stat-util.h"
83 #include "string-table.h"
84 #include "string-util.h"
87 #include "terminal-util.h"
88 #include "time-util.h"
89 #include "transaction.h"
90 #include "umask-util.h"
91 #include "unit-name.h"
92 #include "user-util.h"
97 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
98 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
100 /* Initial delay and the interval for printing status messages about running jobs */
101 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
102 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
103 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
105 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
106 * the queue gets more empty. */
107 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
109 /* How many units and jobs to process of the bus queue before returning to the event loop. */
110 #define MANAGER_BUS_MESSAGE_BUDGET 100U
112 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
113 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
114 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
115 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
116 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
117 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
118 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
119 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
120 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
121 static int manager_run_environment_generators(Manager
*m
);
122 static int manager_run_generators(Manager
*m
);
124 static void manager_watch_jobs_in_progress(Manager
*m
) {
130 /* We do not want to show the cylon animation if the user
131 * needs to confirm service executions otherwise confirmation
132 * messages will be screwed by the cylon animation. */
133 if (!manager_is_confirm_spawn_disabled(m
))
136 if (m
->jobs_in_progress_event_source
)
139 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
140 r
= sd_event_add_time(
142 &m
->jobs_in_progress_event_source
,
145 manager_dispatch_jobs_in_progress
, m
);
149 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
152 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
154 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
157 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
158 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
162 p
= mempset(p
, ' ', pos
-2);
163 if (log_get_show_color())
164 p
= stpcpy(p
, ANSI_RED
);
168 if (pos
> 0 && pos
<= width
) {
169 if (log_get_show_color())
170 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
174 if (log_get_show_color())
175 p
= stpcpy(p
, ANSI_NORMAL
);
178 if (log_get_show_color())
179 p
= stpcpy(p
, ANSI_RED
);
182 p
= mempset(p
, ' ', width
-1-pos
);
183 if (log_get_show_color())
184 strcpy(p
, ANSI_NORMAL
);
188 void manager_flip_auto_status(Manager
*m
, bool enable
) {
192 if (m
->show_status
== SHOW_STATUS_AUTO
)
193 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
195 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
196 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
200 static void manager_print_jobs_in_progress(Manager
*m
) {
201 _cleanup_free_
char *job_of_n
= NULL
;
204 unsigned counter
= 0, print_nr
;
205 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
207 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
211 assert(m
->n_running_jobs
> 0);
213 manager_flip_auto_status(m
, true);
215 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
217 HASHMAP_FOREACH(j
, m
->jobs
, i
)
218 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
221 /* m->n_running_jobs must be consistent with the contents of m->jobs,
222 * so the above loop must have succeeded in finding j. */
223 assert(counter
== print_nr
+ 1);
226 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
228 cylon_pos
= 14 - cylon_pos
;
229 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
231 m
->jobs_in_progress_iteration
++;
233 if (m
->n_running_jobs
> 1) {
234 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
238 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
239 if (job_get_timeout(j
, &x
) > 0)
240 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
242 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
243 "%sA %s job is running for %s (%s / %s)",
245 job_type_to_string(j
->type
),
246 unit_description(j
->unit
),
250 static int have_ask_password(void) {
251 _cleanup_closedir_
DIR *dir
;
254 dir
= opendir("/run/systemd/ask-password");
262 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
263 if (startswith(de
->d_name
, "ask."))
269 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
270 int fd
, uint32_t revents
, void *userdata
) {
271 Manager
*m
= userdata
;
277 m
->have_ask_password
= have_ask_password();
278 if (m
->have_ask_password
< 0)
279 /* Log error but continue. Negative have_ask_password
280 * is treated as unknown status. */
281 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
286 static void manager_close_ask_password(Manager
*m
) {
289 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
290 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
291 m
->have_ask_password
= -EINVAL
;
294 static int manager_check_ask_password(Manager
*m
) {
299 if (!m
->ask_password_event_source
) {
300 assert(m
->ask_password_inotify_fd
< 0);
302 mkdir_p_label("/run/systemd/ask-password", 0755);
304 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
305 if (m
->ask_password_inotify_fd
< 0)
306 return log_error_errno(errno
, "inotify_init1() failed: %m");
308 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
309 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
310 manager_close_ask_password(m
);
314 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
315 m
->ask_password_inotify_fd
, EPOLLIN
,
316 manager_dispatch_ask_password_fd
, m
);
318 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
319 manager_close_ask_password(m
);
323 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
325 /* Queries might have been added meanwhile... */
326 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
327 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
330 return m
->have_ask_password
;
333 static int manager_watch_idle_pipe(Manager
*m
) {
338 if (m
->idle_pipe_event_source
)
341 if (m
->idle_pipe
[2] < 0)
344 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
346 return log_error_errno(r
, "Failed to watch idle pipe: %m");
348 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
353 static void manager_close_idle_pipe(Manager
*m
) {
356 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
358 safe_close_pair(m
->idle_pipe
);
359 safe_close_pair(m
->idle_pipe
+ 2);
362 static int manager_setup_time_change(Manager
*m
) {
365 /* We only care for the cancellation event, hence we set the
366 * timeout to the latest possible value. */
367 struct itimerspec its
= {
368 .it_value
.tv_sec
= TIME_T_MAX
,
372 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
374 if (m
->test_run_flags
)
377 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
378 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
380 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
381 if (m
->time_change_fd
< 0)
382 return log_error_errno(errno
, "Failed to create timerfd: %m");
384 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
385 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
386 m
->time_change_fd
= safe_close(m
->time_change_fd
);
390 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
392 return log_error_errno(r
, "Failed to create time change event source: %m");
394 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
396 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
401 static int enable_special_signals(Manager
*m
) {
402 _cleanup_close_
int fd
= -1;
406 if (m
->test_run_flags
)
409 /* Enable that we get SIGINT on control-alt-del. In containers
410 * this will fail with EPERM (older) or EINVAL (newer), so
412 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
413 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
415 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
417 /* Support systems without virtual console */
419 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
421 /* Enable that we get SIGWINCH on kbrequest */
422 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
423 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
429 static int manager_setup_signals(Manager
*m
) {
430 struct sigaction sa
= {
431 .sa_handler
= SIG_DFL
,
432 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
439 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
441 /* We make liberal use of realtime signals here. On
442 * Linux/glibc we have 30 of them (with the exception of Linux
443 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
446 assert_se(sigemptyset(&mask
) == 0);
447 sigset_add_many(&mask
,
448 SIGCHLD
, /* Child died */
449 SIGTERM
, /* Reexecute daemon */
450 SIGHUP
, /* Reload configuration */
451 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
452 SIGUSR2
, /* systemd: dump status */
453 SIGINT
, /* Kernel sends us this on control-alt-del */
454 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
455 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
457 SIGRTMIN
+0, /* systemd: start default.target */
458 SIGRTMIN
+1, /* systemd: isolate rescue.target */
459 SIGRTMIN
+2, /* systemd: isolate emergency.target */
460 SIGRTMIN
+3, /* systemd: start halt.target */
461 SIGRTMIN
+4, /* systemd: start poweroff.target */
462 SIGRTMIN
+5, /* systemd: start reboot.target */
463 SIGRTMIN
+6, /* systemd: start kexec.target */
465 /* ... space for more special targets ... */
467 SIGRTMIN
+13, /* systemd: Immediate halt */
468 SIGRTMIN
+14, /* systemd: Immediate poweroff */
469 SIGRTMIN
+15, /* systemd: Immediate reboot */
470 SIGRTMIN
+16, /* systemd: Immediate kexec */
472 /* ... space for more immediate system state changes ... */
474 SIGRTMIN
+20, /* systemd: enable status messages */
475 SIGRTMIN
+21, /* systemd: disable status messages */
476 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
477 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
478 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
480 /* .. one free signal here ... */
482 #if !defined(__hppa64__) && !defined(__hppa__)
483 /* Apparently Linux on hppa has fewer RT
484 * signals (SIGRTMAX is SIGRTMIN+25 there),
485 * hence let's not try to make use of them
486 * here. Since these commands are accessible
487 * by different means and only really a safety
488 * net, the missing functionality on hppa
489 * shouldn't matter. */
491 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
492 SIGRTMIN
+27, /* systemd: set log target to console */
493 SIGRTMIN
+28, /* systemd: set log target to kmsg */
494 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
496 /* ... one free signal here SIGRTMIN+30 ... */
499 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
501 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
502 if (m
->signal_fd
< 0)
505 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
509 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
511 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
512 * notify processing can still figure out to which process/service a message belongs, before we reap the
513 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
514 * status information before detecting that there's no process in a cgroup. */
515 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
519 if (MANAGER_IS_SYSTEM(m
))
520 return enable_special_signals(m
);
525 static void manager_sanitize_environment(Manager
*m
) {
528 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
548 /* Let's order the environment alphabetically, just to make it pretty */
549 strv_sort(m
->environment
);
552 static int manager_default_environment(Manager
*m
) {
555 if (MANAGER_IS_SYSTEM(m
)) {
556 /* The system manager always starts with a clean
557 * environment for its children. It does not import
558 * the kernel's or the parents' exported variables.
560 * The initial passed environment is untouched to keep
561 * /proc/self/environ valid; it is used for tagging
562 * the init process inside containers. */
563 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
566 /* Import locale variables LC_*= from configuration */
567 locale_setup(&m
->environment
);
569 /* The user manager passes its own environment
570 * along to its children. */
571 m
->environment
= strv_copy(environ
);
576 manager_sanitize_environment(m
);
581 static int manager_setup_prefix(Manager
*m
) {
587 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
588 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
589 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
590 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
591 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
592 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
595 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
596 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
597 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
598 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
599 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
600 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
603 const struct table_entry
*p
;
609 if (MANAGER_IS_SYSTEM(m
))
614 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
615 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
623 static int manager_setup_run_queue(Manager
*m
) {
627 assert(!m
->run_queue_event_source
);
629 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
633 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
637 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
641 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
646 static int manager_setup_sigchld_event_source(Manager
*m
) {
650 assert(!m
->sigchld_event_source
);
652 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
656 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
660 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
664 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
669 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
674 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
676 m
= new0(Manager
, 1);
680 m
->unit_file_scope
= scope
;
681 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
682 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
683 m
->default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
;
684 m
->default_tasks_accounting
= true;
685 m
->default_tasks_max
= UINT64_MAX
;
686 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
687 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
688 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
691 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
692 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
693 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
694 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
697 /* Prepare log fields we can use for structured logging */
698 if (MANAGER_IS_SYSTEM(m
)) {
699 m
->unit_log_field
= "UNIT=";
700 m
->unit_log_format_string
= "UNIT=%s";
702 m
->invocation_log_field
= "INVOCATION_ID=";
703 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
705 m
->unit_log_field
= "USER_UNIT=";
706 m
->unit_log_format_string
= "USER_UNIT=%s";
708 m
->invocation_log_field
= "USER_INVOCATION_ID=";
709 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
712 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
714 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
715 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
716 m
->ask_password_inotify_fd
= -1;
718 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
720 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
722 m
->have_ask_password
= -EINVAL
; /* we don't know */
725 m
->test_run_flags
= test_run_flags
;
727 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
728 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
730 r
= manager_default_environment(m
);
734 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
738 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
742 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
746 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
750 r
= sd_event_default(&m
->event
);
754 r
= manager_setup_run_queue(m
);
758 r
= manager_setup_signals(m
);
762 r
= manager_setup_cgroup(m
);
766 r
= manager_setup_time_change(m
);
770 r
= manager_setup_sigchld_event_source(m
);
774 m
->udev
= udev_new();
780 r
= manager_setup_prefix(m
);
784 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
785 r
= mkdir_label("/run/systemd/units", 0755);
786 if (r
< 0 && r
!= -EEXIST
)
792 dir_is_empty("/usr") > 0;
794 /* Note that we do not set up the notify fd here. We do that after deserialization,
795 * since they might have gotten serialized across the reexec. */
805 static int manager_setup_notify(Manager
*m
) {
808 if (m
->test_run_flags
)
811 if (m
->notify_fd
< 0) {
812 _cleanup_close_
int fd
= -1;
813 union sockaddr_union sa
= {
814 .sa
.sa_family
= AF_UNIX
,
816 static const int one
= 1;
818 /* First free all secondary fields */
819 m
->notify_socket
= mfree(m
->notify_socket
);
820 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
822 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
824 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
826 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
828 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
829 if (!m
->notify_socket
)
832 (void) mkdir_parents_label(m
->notify_socket
, 0755);
833 (void) unlink(m
->notify_socket
);
835 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
836 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
838 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
840 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
842 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
847 log_debug("Using notification socket %s", m
->notify_socket
);
850 if (!m
->notify_event_source
) {
851 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
853 return log_error_errno(r
, "Failed to allocate notify event source: %m");
855 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
856 * service an exit message belongs. */
857 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
859 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
861 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
867 static int manager_setup_cgroups_agent(Manager
*m
) {
869 static const union sockaddr_union sa
= {
870 .un
.sun_family
= AF_UNIX
,
871 .un
.sun_path
= "/run/systemd/cgroups-agent",
875 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
876 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
877 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
878 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
879 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
880 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
881 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
882 * we thus won't lose messages.
884 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
885 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
886 * bus for these messages. */
888 if (m
->test_run_flags
)
891 if (!MANAGER_IS_SYSTEM(m
))
894 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
896 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
897 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
900 if (m
->cgroups_agent_fd
< 0) {
901 _cleanup_close_
int fd
= -1;
903 /* First free all secondary fields */
904 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
906 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
908 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
910 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
912 (void) unlink(sa
.un
.sun_path
);
914 /* Only allow root to connect to this socket */
916 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
918 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
920 m
->cgroups_agent_fd
= fd
;
924 if (!m
->cgroups_agent_event_source
) {
925 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
927 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
929 /* Process cgroups notifications early, but after having processed service notification messages or
930 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
931 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
932 * cgroup inotify for the unified cgroup stuff. */
933 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
935 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
937 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
943 static int manager_setup_user_lookup_fd(Manager
*m
) {
948 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
949 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
950 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
951 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
952 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
953 * hence we establish this communication channel so that forked off processes can pass their UID/GID
954 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
955 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
958 * You might wonder why we need a communication channel for this that is independent of the usual notification
959 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
960 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
961 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
963 * Note that this function is called under two circumstances: when we first initialize (in which case we
964 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
965 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
967 if (m
->user_lookup_fds
[0] < 0) {
969 /* Free all secondary fields */
970 safe_close_pair(m
->user_lookup_fds
);
971 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
973 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
974 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
976 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
979 if (!m
->user_lookup_event_source
) {
980 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
982 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
984 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
986 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
988 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
990 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
996 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1002 while ((u
= m
->cleanup_queue
)) {
1003 assert(u
->in_cleanup_queue
);
1013 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1014 GC_OFFSET_UNSURE
, /* No clue */
1015 GC_OFFSET_GOOD
, /* We still need this unit */
1016 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1020 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1025 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1027 /* Recursively mark referenced units as GOOD as well */
1028 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1029 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1030 unit_gc_mark_good(other
, gc_marker
);
1033 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1041 if (IN_SET(u
->gc_marker
- gc_marker
,
1042 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1045 if (u
->in_cleanup_queue
)
1048 if (!unit_may_gc(u
))
1051 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1055 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1056 unit_gc_sweep(other
, gc_marker
);
1058 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1061 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1065 if (u
->refs_by_target
) {
1068 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1069 unit_gc_sweep(ref
->source
, gc_marker
);
1071 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1074 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1082 /* We were unable to find anything out about this entry, so
1083 * let's investigate it later */
1084 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1085 unit_add_to_gc_queue(u
);
1089 /* We definitely know that this one is not useful anymore, so
1090 * let's mark it for deletion */
1091 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1092 unit_add_to_cleanup_queue(u
);
1096 unit_gc_mark_good(u
, gc_marker
);
1099 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1100 unsigned n
= 0, gc_marker
;
1105 /* log_debug("Running GC..."); */
1107 m
->gc_marker
+= _GC_OFFSET_MAX
;
1108 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1111 gc_marker
= m
->gc_marker
;
1113 while ((u
= m
->gc_unit_queue
)) {
1114 assert(u
->in_gc_queue
);
1116 unit_gc_sweep(u
, gc_marker
);
1118 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1119 u
->in_gc_queue
= false;
1123 if (IN_SET(u
->gc_marker
- gc_marker
,
1124 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1126 log_unit_debug(u
, "Collecting.");
1127 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1128 unit_add_to_cleanup_queue(u
);
1135 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1141 while ((j
= m
->gc_job_queue
)) {
1142 assert(j
->in_gc_queue
);
1144 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1145 j
->in_gc_queue
= false;
1152 log_unit_debug(j
->unit
, "Collecting job.");
1153 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1159 static void manager_clear_jobs_and_units(Manager
*m
) {
1164 while ((u
= hashmap_first(m
->units
)))
1167 manager_dispatch_cleanup_queue(m
);
1169 assert(!m
->load_queue
);
1170 assert(!m
->run_queue
);
1171 assert(!m
->dbus_unit_queue
);
1172 assert(!m
->dbus_job_queue
);
1173 assert(!m
->cleanup_queue
);
1174 assert(!m
->gc_unit_queue
);
1175 assert(!m
->gc_job_queue
);
1177 assert(hashmap_isempty(m
->jobs
));
1178 assert(hashmap_isempty(m
->units
));
1180 m
->n_on_console
= 0;
1181 m
->n_running_jobs
= 0;
1184 Manager
* manager_free(Manager
*m
) {
1187 ExecDirectoryType dt
;
1192 manager_clear_jobs_and_units(m
);
1194 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1195 if (unit_vtable
[c
]->shutdown
)
1196 unit_vtable
[c
]->shutdown(m
);
1198 /* If we reexecute ourselves, we keep the root cgroup around */
1199 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1201 lookup_paths_flush_generator(&m
->lookup_paths
);
1205 exec_runtime_vacuum(m
);
1206 hashmap_free(m
->exec_runtime_by_id
);
1208 dynamic_user_vacuum(m
, false);
1209 hashmap_free(m
->dynamic_users
);
1211 hashmap_free(m
->units
);
1212 hashmap_free(m
->units_by_invocation_id
);
1213 hashmap_free(m
->jobs
);
1214 hashmap_free(m
->watch_pids
);
1215 hashmap_free(m
->watch_bus
);
1217 set_free(m
->startup_units
);
1218 set_free(m
->failed_units
);
1220 sd_event_source_unref(m
->signal_event_source
);
1221 sd_event_source_unref(m
->sigchld_event_source
);
1222 sd_event_source_unref(m
->notify_event_source
);
1223 sd_event_source_unref(m
->cgroups_agent_event_source
);
1224 sd_event_source_unref(m
->time_change_event_source
);
1225 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1226 sd_event_source_unref(m
->run_queue_event_source
);
1227 sd_event_source_unref(m
->user_lookup_event_source
);
1228 sd_event_source_unref(m
->sync_bus_names_event_source
);
1230 safe_close(m
->signal_fd
);
1231 safe_close(m
->notify_fd
);
1232 safe_close(m
->cgroups_agent_fd
);
1233 safe_close(m
->time_change_fd
);
1234 safe_close_pair(m
->user_lookup_fds
);
1236 manager_close_ask_password(m
);
1238 manager_close_idle_pipe(m
);
1240 udev_unref(m
->udev
);
1241 sd_event_unref(m
->event
);
1243 free(m
->notify_socket
);
1245 lookup_paths_free(&m
->lookup_paths
);
1246 strv_free(m
->environment
);
1248 hashmap_free(m
->cgroup_unit
);
1249 set_free_free(m
->unit_path_cache
);
1251 free(m
->switch_root
);
1252 free(m
->switch_root_init
);
1254 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1255 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1257 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1258 hashmap_free(m
->units_requiring_mounts_for
);
1260 hashmap_free(m
->uid_refs
);
1261 hashmap_free(m
->gid_refs
);
1263 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1264 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1269 void manager_enumerate(Manager
*m
) {
1274 /* Let's ask every type to load all units from disk/kernel
1275 * that it might know */
1276 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1277 if (!unit_type_supported(c
)) {
1278 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1282 if (!unit_vtable
[c
]->enumerate
)
1285 unit_vtable
[c
]->enumerate(m
);
1288 manager_dispatch_load_queue(m
);
1291 static void manager_coldplug(Manager
*m
) {
1299 /* Then, let's set up their initial state. */
1300 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1302 /* ignore aliases */
1306 r
= unit_coldplug(u
);
1308 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1312 static void manager_build_unit_path_cache(Manager
*m
) {
1318 set_free_free(m
->unit_path_cache
);
1320 m
->unit_path_cache
= set_new(&path_hash_ops
);
1321 if (!m
->unit_path_cache
) {
1326 /* This simply builds a list of files we know exist, so that
1327 * we don't always have to go to disk */
1329 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1330 _cleanup_closedir_
DIR *d
= NULL
;
1335 if (errno
!= ENOENT
)
1336 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1340 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1343 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1349 r
= set_consume(m
->unit_path_cache
, p
);
1358 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1359 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1362 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1368 HASHMAP_FOREACH(u
, m
->units
, i
) {
1370 if (fdset_size(fds
) <= 0)
1373 if (!UNIT_VTABLE(u
)->distribute_fds
)
1376 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1380 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1385 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1386 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1387 * rather than the current one. */
1389 if (m
->test_run_flags
!= 0)
1392 /* If we are in the user instance, and the env var is already set for us, then this means D-Bus is ran
1393 * somewhere outside of our own logic. Let's use it */
1394 if (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"))
1397 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1400 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1403 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1406 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1412 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1417 /* If we are running in test mode, we still want to run the generators,
1418 * but we should not touch the real generator directories. */
1419 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1420 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1425 r
= manager_run_environment_generators(m
);
1429 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1430 r
= manager_run_generators(m
);
1431 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1435 /* If this is the first boot, and we are in the host system, then preset everything */
1436 if (m
->first_boot
> 0 &&
1437 MANAGER_IS_SYSTEM(m
) &&
1438 !m
->test_run_flags
) {
1440 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1442 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1443 "Failed to populate /etc with preset unit settings, ignoring: %m");
1445 log_info("Populated /etc with preset unit settings.");
1448 lookup_paths_reduce(&m
->lookup_paths
);
1449 manager_build_unit_path_cache(m
);
1451 /* If we will deserialize make sure that during enumeration
1452 * this is already known, so we increase the counter here
1457 /* First, enumerate what we can from all config files */
1458 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1459 manager_enumerate(m
);
1460 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1462 /* Second, deserialize if there is something to deserialize */
1463 if (serialization
) {
1464 r
= manager_deserialize(m
, serialization
, fds
);
1466 return log_error_errno(r
, "Deserialization failed: %m");
1469 /* Any fds left? Find some unit which wants them. This is
1470 * useful to allow container managers to pass some file
1471 * descriptors to us pre-initialized. This enables
1472 * socket-based activation of entire containers. */
1473 manager_distribute_fds(m
, fds
);
1475 /* We might have deserialized the notify fd, but if we didn't
1476 * then let's create the bus now */
1477 r
= manager_setup_notify(m
);
1479 /* No sense to continue without notifications, our children would fail anyway. */
1482 r
= manager_setup_cgroups_agent(m
);
1484 /* Likewise, no sense to continue without empty cgroup notifications. */
1487 r
= manager_setup_user_lookup_fd(m
);
1489 /* This shouldn't fail, except if things are really broken. */
1492 /* Let's set up our private bus connection now, unconditionally */
1493 (void) bus_init_private(m
);
1495 /* If we are in --user mode also connect to the system bus now */
1496 if (MANAGER_IS_USER(m
))
1497 (void) bus_init_system(m
);
1499 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1500 if (manager_dbus_is_running(m
, !!serialization
)) {
1501 (void) bus_init_api(m
);
1503 if (MANAGER_IS_SYSTEM(m
))
1504 (void) bus_init_system(m
);
1507 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1508 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1509 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1511 /* Third, fire things up! */
1512 manager_coldplug(m
);
1514 /* Release any dynamic users no longer referenced */
1515 dynamic_user_vacuum(m
, true);
1517 exec_runtime_vacuum(m
);
1519 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1520 manager_vacuum_uid_refs(m
);
1521 manager_vacuum_gid_refs(m
);
1523 if (serialization
) {
1524 assert(m
->n_reloading
> 0);
1527 /* Let's wait for the UnitNew/JobNew messages being
1528 * sent, before we notify that the reload is
1530 m
->send_reloading_done
= true;
1536 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1541 assert(type
< _JOB_TYPE_MAX
);
1543 assert(mode
< _JOB_MODE_MAX
);
1545 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1546 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1548 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1549 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1551 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1553 type
= job_type_collapse(type
, unit
);
1555 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1559 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1560 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1561 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1565 if (mode
== JOB_ISOLATE
) {
1566 r
= transaction_add_isolate_jobs(tr
, m
);
1571 r
= transaction_activate(tr
, m
, mode
, e
);
1575 log_unit_debug(unit
,
1576 "Enqueued job %s/%s as %u", unit
->id
,
1577 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1580 *_ret
= tr
->anchor_job
;
1582 transaction_free(tr
);
1586 transaction_abort(tr
);
1587 transaction_free(tr
);
1591 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1592 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1596 assert(type
< _JOB_TYPE_MAX
);
1598 assert(mode
< _JOB_MODE_MAX
);
1600 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1605 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1608 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1609 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1613 assert(type
< _JOB_TYPE_MAX
);
1615 assert(mode
< _JOB_MODE_MAX
);
1617 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1619 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1624 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1630 assert(mode
< _JOB_MODE_MAX
);
1631 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1633 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1637 /* We need an anchor job */
1638 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1642 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1643 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1645 r
= transaction_activate(tr
, m
, mode
, e
);
1649 transaction_free(tr
);
1653 transaction_abort(tr
);
1654 transaction_free(tr
);
1658 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1661 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1664 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1668 return hashmap_get(m
->units
, name
);
1671 unsigned manager_dispatch_load_queue(Manager
*m
) {
1677 /* Make sure we are not run recursively */
1678 if (m
->dispatching_load_queue
)
1681 m
->dispatching_load_queue
= true;
1683 /* Dispatches the load queue. Takes a unit from the queue and
1684 * tries to load its data until the queue is empty */
1686 while ((u
= m
->load_queue
)) {
1687 assert(u
->in_load_queue
);
1693 m
->dispatching_load_queue
= false;
1697 int manager_load_unit_prepare(
1709 assert(name
|| path
);
1712 /* This will prepare the unit for loading, but not actually
1713 * load anything from disk. */
1715 if (path
&& !is_path(path
))
1716 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1719 name
= basename(path
);
1721 t
= unit_name_to_type(name
);
1723 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1724 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1725 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1727 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1730 ret
= manager_get_unit(m
, name
);
1736 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1741 ret
->fragment_path
= strdup(path
);
1742 if (!ret
->fragment_path
) {
1748 r
= unit_add_name(ret
, name
);
1754 unit_add_to_load_queue(ret
);
1755 unit_add_to_dbus_queue(ret
);
1756 unit_add_to_gc_queue(ret
);
1763 int manager_load_unit(
1775 /* This will load the service information files, but not actually
1776 * start any services or anything. */
1778 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1782 manager_dispatch_load_queue(m
);
1784 *_ret
= unit_follow_merge(*_ret
);
1789 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1796 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1797 job_dump(j
, f
, prefix
);
1800 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1808 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1810 unit_dump(u
, f
, prefix
);
1813 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1819 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1820 char buf
[FORMAT_TIMESTAMP_MAX
];
1822 if (dual_timestamp_is_set(m
->timestamps
+ q
))
1823 fprintf(f
, "%sTimestamp %s: %s\n",
1825 manager_timestamp_to_string(q
),
1826 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
1829 manager_dump_units(m
, f
, prefix
);
1830 manager_dump_jobs(m
, f
, prefix
);
1833 int manager_get_dump_string(Manager
*m
, char **ret
) {
1834 _cleanup_free_
char *dump
= NULL
;
1835 _cleanup_fclose_
FILE *f
= NULL
;
1842 f
= open_memstream(&dump
, &size
);
1846 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1848 manager_dump(m
, f
, NULL
);
1850 r
= fflush_and_check(f
);
1862 void manager_clear_jobs(Manager
*m
) {
1867 while ((j
= hashmap_first(m
->jobs
)))
1868 /* No need to recurse. We're cancelling all jobs. */
1869 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1872 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1873 Manager
*m
= userdata
;
1879 while ((j
= m
->run_queue
)) {
1880 assert(j
->installed
);
1881 assert(j
->in_run_queue
);
1883 job_run_and_invalidate(j
);
1886 if (m
->n_running_jobs
> 0)
1887 manager_watch_jobs_in_progress(m
);
1889 if (m
->n_on_console
> 0)
1890 manager_watch_idle_pipe(m
);
1895 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1896 unsigned n
= 0, budget
;
1902 if (m
->dispatching_dbus_queue
)
1905 /* Anything to do at all? */
1906 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
1909 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
1910 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
1911 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
1914 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
1915 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
1916 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
1917 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
1918 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
1919 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
1920 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
1921 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
1922 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
1923 * the "threshold" is currently chosen much higher than the "budget". */
1924 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
1926 m
->dispatching_dbus_queue
= true;
1928 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
1930 assert(u
->in_dbus_queue
);
1932 bus_unit_send_change_signal(u
);
1936 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
1937 assert(j
->in_dbus_queue
);
1939 bus_job_send_change_signal(j
);
1943 m
->dispatching_dbus_queue
= false;
1945 if (budget
> 0 && m
->send_reloading_done
) {
1946 m
->send_reloading_done
= false;
1947 bus_manager_send_reloading(m
, false);
1951 if (budget
> 0 && m
->queued_message
) {
1952 bus_send_queued_message(m
);
1959 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1960 Manager
*m
= userdata
;
1961 char buf
[PATH_MAX
+1];
1964 n
= recv(fd
, buf
, sizeof(buf
), 0);
1966 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1968 log_error("Got zero-length cgroups agent message, ignoring.");
1971 if ((size_t) n
>= sizeof(buf
)) {
1972 log_error("Got overly long cgroups agent message, ignoring.");
1976 if (memchr(buf
, 0, n
)) {
1977 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1982 manager_notify_cgroup_empty(m
, buf
);
1983 (void) bus_forward_agent_released(m
, buf
);
1988 static void manager_invoke_notify_message(
1991 const struct ucred
*ucred
,
2000 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2002 u
->notifygen
= m
->notifygen
;
2004 if (UNIT_VTABLE(u
)->notify_message
) {
2005 _cleanup_strv_free_
char **tags
= NULL
;
2007 tags
= strv_split(buf
, NEWLINE
);
2013 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2015 } else if (DEBUG_LOGGING
) {
2016 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2018 x
= ellipsize(buf
, 20, 90);
2022 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2026 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2028 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2029 Manager
*m
= userdata
;
2030 char buf
[NOTIFY_BUFFER_MAX
+1];
2031 struct iovec iovec
= {
2033 .iov_len
= sizeof(buf
)-1,
2036 struct cmsghdr cmsghdr
;
2037 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2038 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2040 struct msghdr msghdr
= {
2043 .msg_control
= &control
,
2044 .msg_controllen
= sizeof(control
),
2047 struct cmsghdr
*cmsg
;
2048 struct ucred
*ucred
= NULL
;
2049 _cleanup_free_ Unit
**array_copy
= NULL
;
2050 Unit
*u1
, *u2
, **array
;
2051 int r
, *fd_array
= NULL
;
2057 assert(m
->notify_fd
== fd
);
2059 if (revents
!= EPOLLIN
) {
2060 log_warning("Got unexpected poll event for notify fd.");
2064 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2066 if (IN_SET(errno
, EAGAIN
, EINTR
))
2067 return 0; /* Spurious wakeup, try again */
2069 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2070 * won't take notification messages anymore, but that's still better than busy looping around this:
2071 * being woken up over and over again but being unable to actually read the message off the socket. */
2072 return log_error_errno(errno
, "Failed to receive notification message: %m");
2075 CMSG_FOREACH(cmsg
, &msghdr
) {
2076 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2078 fd_array
= (int*) CMSG_DATA(cmsg
);
2079 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2081 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2082 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2083 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2085 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2092 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2094 close_many(fd_array
, n_fds
);
2100 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2101 log_warning("Received notify message without valid credentials. Ignoring.");
2105 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2106 log_warning("Received notify message exceeded maximum size. Ignoring.");
2110 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2111 * trailing NUL byte in the message, but don't expect it. */
2112 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2113 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2117 /* Make sure it's NUL-terminated. */
2120 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2123 /* Notify every unit that might be interested, which might be multiple. */
2124 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2125 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2126 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2133 array_copy
= newdup(Unit
*, array
, k
+1);
2137 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2138 * make sure we only invoke each unit's handler once. */
2140 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2144 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2148 for (size_t i
= 0; array_copy
[i
]; i
++) {
2149 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2154 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2156 if (fdset_size(fds
) > 0)
2157 log_warning("Got extra auxiliary fds with notification message, closing them.");
2162 static void manager_invoke_sigchld_event(
2165 const siginfo_t
*si
) {
2171 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2172 if (u
->sigchldgen
== m
->sigchldgen
)
2174 u
->sigchldgen
= m
->sigchldgen
;
2176 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2177 unit_unwatch_pid(u
, si
->si_pid
);
2179 if (UNIT_VTABLE(u
)->sigchld_event
)
2180 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2183 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2184 Manager
*m
= userdata
;
2191 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2192 * while it is a zombie. */
2194 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2196 if (errno
!= ECHILD
)
2197 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2205 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2206 _cleanup_free_ Unit
**array_copy
= NULL
;
2207 _cleanup_free_
char *name
= NULL
;
2208 Unit
*u1
, *u2
, **array
;
2210 (void) get_process_comm(si
.si_pid
, &name
);
2212 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2213 si
.si_pid
, strna(name
),
2214 sigchld_code_to_string(si
.si_code
),
2216 strna(si
.si_code
== CLD_EXITED
2217 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2218 : signal_to_string(si
.si_status
)));
2220 /* Increase the generation counter used for filtering out duplicate unit invocations */
2223 /* And now figure out the unit this belongs to, it might be multiple... */
2224 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2225 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2226 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2230 /* Cound how many entries the array has */
2234 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2235 array_copy
= newdup(Unit
*, array
, n
+1);
2240 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2241 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2242 * each iteration. */
2244 manager_invoke_sigchld_event(m
, u1
, &si
);
2246 manager_invoke_sigchld_event(m
, u2
, &si
);
2248 for (size_t i
= 0; array_copy
[i
]; i
++)
2249 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2252 /* And now, we actually reap the zombie. */
2253 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2254 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2261 /* All children processed for now, turn off event source */
2263 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2265 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2270 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2271 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2274 log_debug("Activating special unit %s", name
);
2276 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2278 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2281 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2282 /* If the user presses C-A-D more than
2283 * 7 times within 2s, we reboot/shutdown immediately,
2284 * unless it was disabled in system.conf */
2286 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2287 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2289 emergency_action(m
, m
->cad_burst_action
, NULL
,
2290 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2293 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2294 Manager
*m
= userdata
;
2296 struct signalfd_siginfo sfsi
;
2300 assert(m
->signal_fd
== fd
);
2302 if (revents
!= EPOLLIN
) {
2303 log_warning("Got unexpected events from signal file descriptor.");
2307 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2308 if (n
!= sizeof(sfsi
)) {
2310 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2314 if (IN_SET(errno
, EINTR
, EAGAIN
))
2317 /* We return an error here, which will kill this handler,
2318 * to avoid a busy loop on read error. */
2319 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2322 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2323 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2324 ? LOG_DEBUG
: LOG_INFO
,
2327 switch (sfsi
.ssi_signo
) {
2330 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2332 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2337 if (MANAGER_IS_SYSTEM(m
)) {
2338 /* This is for compatibility with the
2339 * original sysvinit */
2340 r
= verify_run_space_and_log("Refusing to reexecute");
2342 m
->exit_code
= MANAGER_REEXECUTE
;
2348 if (MANAGER_IS_SYSTEM(m
))
2349 manager_handle_ctrl_alt_del(m
);
2351 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2352 JOB_REPLACE_IRREVERSIBLY
);
2356 if (MANAGER_IS_SYSTEM(m
))
2357 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2359 /* This is a nop on non-init */
2363 if (MANAGER_IS_SYSTEM(m
))
2364 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2366 /* This is a nop on non-init */
2371 if (manager_dbus_is_running(m
, false)) {
2372 log_info("Trying to reconnect to bus...");
2374 (void) bus_init_api(m
);
2376 if (MANAGER_IS_SYSTEM(m
))
2377 (void) bus_init_system(m
);
2379 log_info("Starting D-Bus service...");
2380 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2386 _cleanup_free_
char *dump
= NULL
;
2388 r
= manager_get_dump_string(m
, &dump
);
2390 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2394 log_dump(LOG_INFO
, dump
);
2399 r
= verify_run_space_and_log("Refusing to reload");
2401 m
->exit_code
= MANAGER_RELOAD
;
2406 /* Starting SIGRTMIN+0 */
2407 static const struct {
2410 } target_table
[] = {
2411 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2412 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2413 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2414 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2415 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2416 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2417 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2420 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2421 static const ManagerExitCode code_table
[] = {
2423 [1] = MANAGER_POWEROFF
,
2424 [2] = MANAGER_REBOOT
,
2425 [3] = MANAGER_KEXEC
,
2428 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2429 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2430 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2431 manager_start_target(m
, target_table
[idx
].target
,
2432 target_table
[idx
].mode
);
2436 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2437 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2438 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2442 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2445 manager_set_show_status(m
, SHOW_STATUS_YES
);
2449 manager_set_show_status(m
, SHOW_STATUS_NO
);
2453 log_set_max_level(LOG_DEBUG
);
2454 log_info("Setting log level to debug.");
2458 log_set_max_level(LOG_INFO
);
2459 log_info("Setting log level to info.");
2463 if (MANAGER_IS_USER(m
)) {
2464 m
->exit_code
= MANAGER_EXIT
;
2468 /* This is a nop on init */
2472 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2473 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2474 log_notice("Setting log target to journal-or-kmsg.");
2478 log_set_target(LOG_TARGET_CONSOLE
);
2479 log_notice("Setting log target to console.");
2483 log_set_target(LOG_TARGET_KMSG
);
2484 log_notice("Setting log target to kmsg.");
2488 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2495 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2496 Manager
*m
= userdata
;
2501 assert(m
->time_change_fd
== fd
);
2503 log_struct(LOG_DEBUG
,
2504 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2505 LOG_MESSAGE("Time has been changed"),
2508 /* Restart the watch */
2509 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2510 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2512 manager_setup_time_change(m
);
2514 HASHMAP_FOREACH(u
, m
->units
, i
)
2515 if (UNIT_VTABLE(u
)->time_change
)
2516 UNIT_VTABLE(u
)->time_change(u
);
2521 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2522 Manager
*m
= userdata
;
2525 assert(m
->idle_pipe
[2] == fd
);
2527 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2528 * now turn off any further console output if there's at least one service that needs console access, so that
2529 * from now on our own output should not spill into that service's output anymore. After all, we support
2530 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2531 * exclusively without our interference. */
2532 m
->no_console_output
= m
->n_on_console
> 0;
2534 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2535 * by closing the pipes towards them, which is what they are waiting for. */
2536 manager_close_idle_pipe(m
);
2541 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2542 Manager
*m
= userdata
;
2549 manager_print_jobs_in_progress(m
);
2551 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2552 r
= sd_event_source_set_time(source
, next
);
2556 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2559 int manager_loop(Manager
*m
) {
2562 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2565 m
->exit_code
= MANAGER_OK
;
2567 /* Release the path cache */
2568 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2570 manager_check_finished(m
);
2572 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2573 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2575 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2577 while (m
->exit_code
== MANAGER_OK
) {
2580 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2583 if (!ratelimit_test(&rl
)) {
2584 /* Yay, something is going seriously wrong, pause a little */
2585 log_warning("Looping too fast. Throttling execution a little.");
2589 if (manager_dispatch_load_queue(m
) > 0)
2592 if (manager_dispatch_gc_job_queue(m
) > 0)
2595 if (manager_dispatch_gc_unit_queue(m
) > 0)
2598 if (manager_dispatch_cleanup_queue(m
) > 0)
2601 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2604 if (manager_dispatch_dbus_queue(m
) > 0)
2607 /* Sleep for half the watchdog time */
2608 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2609 wait_usec
= m
->runtime_watchdog
/ 2;
2613 wait_usec
= USEC_INFINITY
;
2615 r
= sd_event_run(m
->event
, wait_usec
);
2617 return log_error_errno(r
, "Failed to run event loop: %m");
2620 return m
->exit_code
;
2623 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2624 _cleanup_free_
char *n
= NULL
;
2625 sd_id128_t invocation_id
;
2633 r
= unit_name_from_dbus_path(s
, &n
);
2637 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2638 * as invocation ID. */
2639 r
= sd_id128_from_string(n
, &invocation_id
);
2641 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2647 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
));
2650 /* If this didn't work, we check if this is a unit name */
2651 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2652 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2654 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2662 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2672 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2676 r
= safe_atou(p
, &id
);
2680 j
= manager_get_job(m
, id
);
2689 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2692 _cleanup_free_
char *p
= NULL
;
2696 if (!MANAGER_IS_SYSTEM(m
))
2699 audit_fd
= get_audit_fd();
2703 /* Don't generate audit events if the service was already
2704 * started and we're just deserializing */
2705 if (MANAGER_IS_RELOADING(m
))
2708 if (u
->type
!= UNIT_SERVICE
)
2711 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2713 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2717 msg
= strjoina("unit=", p
);
2718 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2720 /* We aren't allowed to send audit messages?
2721 * Then let's not retry again. */
2724 log_warning_errno(errno
, "Failed to send audit message: %m");
2730 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2731 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2732 _cleanup_free_
char *message
= NULL
;
2733 _cleanup_close_
int fd
= -1;
2736 /* Don't generate plymouth events if the service was already
2737 * started and we're just deserializing */
2738 if (MANAGER_IS_RELOADING(m
))
2741 if (!MANAGER_IS_SYSTEM(m
))
2744 if (detect_container() > 0)
2747 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2750 /* We set SOCK_NONBLOCK here so that we rather drop the
2751 * message then wait for plymouth */
2752 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2754 log_error_errno(errno
, "socket() failed: %m");
2758 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2759 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2760 log_error_errno(errno
, "connect() failed: %m");
2764 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2770 if (write(fd
, message
, n
+ 1) != n
+ 1)
2771 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2772 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2775 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2781 fd
= open_serialization_fd("systemd-state");
2785 f
= fdopen(fd
, "w+");
2795 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2808 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2809 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2810 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2811 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2812 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2813 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
2814 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
2816 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2817 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
2818 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
2821 t
= manager_timestamp_to_string(q
);
2823 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
2824 strcpy(stpcpy(field
, t
), "-timestamp");
2825 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
2829 if (!switching_root
)
2830 (void) serialize_environment(f
, m
->environment
);
2832 if (m
->notify_fd
>= 0) {
2835 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2839 fprintf(f
, "notify-fd=%i\n", copy
);
2840 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2843 if (m
->cgroups_agent_fd
>= 0) {
2846 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2850 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2853 if (m
->user_lookup_fds
[0] >= 0) {
2856 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2860 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2864 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2867 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2869 r
= dynamic_user_serialize(m
, f
, fds
);
2873 manager_serialize_uid_refs(m
, f
);
2874 manager_serialize_gid_refs(m
, f
);
2876 r
= exec_runtime_serialize(m
, f
, fds
);
2880 (void) fputc('\n', f
);
2882 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2890 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2897 assert(m
->n_reloading
> 0);
2903 r
= bus_fdset_add_all(m
, fds
);
2910 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2916 log_debug("Deserializing state...");
2921 char line
[LINE_MAX
];
2922 const char *val
, *l
;
2924 if (!fgets(line
, sizeof(line
), f
)) {
2939 if ((val
= startswith(l
, "current-job-id="))) {
2942 if (safe_atou32(val
, &id
) < 0)
2943 log_notice("Failed to parse current job id value %s", val
);
2945 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2947 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2950 if (safe_atou32(val
, &n
) < 0)
2951 log_notice("Failed to parse installed jobs counter %s", val
);
2953 m
->n_installed_jobs
+= n
;
2955 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2958 if (safe_atou32(val
, &n
) < 0)
2959 log_notice("Failed to parse failed jobs counter %s", val
);
2961 m
->n_failed_jobs
+= n
;
2963 } else if ((val
= startswith(l
, "taint-usr="))) {
2966 b
= parse_boolean(val
);
2968 log_notice("Failed to parse taint /usr flag %s", val
);
2970 m
->taint_usr
= m
->taint_usr
|| b
;
2972 } else if ((val
= startswith(l
, "ready-sent="))) {
2975 b
= parse_boolean(val
);
2977 log_notice("Failed to parse ready-sent flag %s", val
);
2979 m
->ready_sent
= m
->ready_sent
|| b
;
2981 } else if ((val
= startswith(l
, "taint-logged="))) {
2984 b
= parse_boolean(val
);
2986 log_notice("Failed to parse taint-logged flag %s", val
);
2988 m
->taint_logged
= m
->taint_logged
|| b
;
2990 } else if ((val
= startswith(l
, "service-watchdogs="))) {
2993 b
= parse_boolean(val
);
2995 log_notice("Failed to parse service-watchdogs flag %s", val
);
2997 m
->service_watchdogs
= b
;
2999 } else if (startswith(l
, "env=")) {
3000 r
= deserialize_environment(&m
->environment
, l
);
3004 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
3006 } else if ((val
= startswith(l
, "notify-fd="))) {
3009 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3010 log_notice("Failed to parse notify fd: \"%s\"", val
);
3012 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3013 safe_close(m
->notify_fd
);
3014 m
->notify_fd
= fdset_remove(fds
, fd
);
3017 } else if ((val
= startswith(l
, "notify-socket="))) {
3026 free(m
->notify_socket
);
3027 m
->notify_socket
= n
;
3029 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3032 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3033 log_notice("Failed to parse cgroups agent fd: %s", val
);
3035 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3036 safe_close(m
->cgroups_agent_fd
);
3037 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3040 } else if ((val
= startswith(l
, "user-lookup="))) {
3043 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3044 log_notice("Failed to parse user lookup fd: %s", val
);
3046 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3047 safe_close_pair(m
->user_lookup_fds
);
3048 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3049 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3052 } else if ((val
= startswith(l
, "dynamic-user=")))
3053 dynamic_user_deserialize_one(m
, val
, fds
);
3054 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3055 manager_deserialize_uid_refs_one(m
, val
);
3056 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3057 manager_deserialize_gid_refs_one(m
, val
);
3058 else if ((val
= startswith(l
, "exec-runtime=")))
3059 exec_runtime_deserialize_one(m
, val
, fds
);
3060 else if ((val
= startswith(l
, "subscribed="))) {
3062 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3067 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3068 val
= startswith(l
, manager_timestamp_to_string(q
));
3072 val
= startswith(val
, "-timestamp=");
3077 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3078 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3079 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3080 log_notice("Unknown serialization item '%s'", l
);
3086 char name
[UNIT_NAME_MAX
+2];
3087 const char* unit_name
;
3090 if (!fgets(name
, sizeof(name
), f
)) {
3100 unit_name
= strstrip(name
);
3102 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3104 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3107 unit_deserialize_skip(f
);
3111 r
= unit_deserialize(u
, f
, fds
);
3113 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3123 assert(m
->n_reloading
> 0);
3129 int manager_reload(Manager
*m
) {
3131 _cleanup_fclose_
FILE *f
= NULL
;
3132 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3136 r
= manager_open_serialization(m
, &f
);
3141 bus_manager_send_reloading(m
, true);
3149 r
= manager_serialize(m
, f
, fds
, false);
3155 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3160 /* From here on there is no way back. */
3161 manager_clear_jobs_and_units(m
);
3162 lookup_paths_flush_generator(&m
->lookup_paths
);
3163 lookup_paths_free(&m
->lookup_paths
);
3164 exec_runtime_vacuum(m
);
3165 dynamic_user_vacuum(m
, false);
3166 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3167 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3169 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3170 if (q
< 0 && r
>= 0)
3173 q
= manager_run_environment_generators(m
);
3174 if (q
< 0 && r
>= 0)
3177 /* Find new unit paths */
3178 q
= manager_run_generators(m
);
3179 if (q
< 0 && r
>= 0)
3182 lookup_paths_reduce(&m
->lookup_paths
);
3183 manager_build_unit_path_cache(m
);
3185 /* First, enumerate what we can from all config files */
3186 manager_enumerate(m
);
3188 /* Second, deserialize our stored data */
3189 q
= manager_deserialize(m
, f
, fds
);
3191 log_error_errno(q
, "Deserialization failed: %m");
3200 /* Re-register notify_fd as event source */
3201 q
= manager_setup_notify(m
);
3202 if (q
< 0 && r
>= 0)
3205 q
= manager_setup_cgroups_agent(m
);
3206 if (q
< 0 && r
>= 0)
3209 q
= manager_setup_user_lookup_fd(m
);
3210 if (q
< 0 && r
>= 0)
3213 /* Third, fire things up! */
3214 manager_coldplug(m
);
3216 /* Release any dynamic users no longer referenced */
3217 dynamic_user_vacuum(m
, true);
3219 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3220 manager_vacuum_uid_refs(m
);
3221 manager_vacuum_gid_refs(m
);
3223 exec_runtime_vacuum(m
);
3225 /* It might be safe to log to the journal now and connect to dbus */
3226 manager_recheck_journal(m
);
3227 manager_recheck_dbus(m
);
3229 /* Sync current state of bus names with our set of listening units */
3230 q
= manager_enqueue_sync_bus_names(m
);
3231 if (q
< 0 && r
>= 0)
3234 assert(m
->n_reloading
> 0);
3237 m
->send_reloading_done
= true;
3242 void manager_reset_failed(Manager
*m
) {
3248 HASHMAP_FOREACH(u
, m
->units
, i
)
3249 unit_reset_failed(u
);
3252 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3258 /* Returns true if the unit is inactive or going down */
3259 u
= manager_get_unit(m
, name
);
3263 return unit_inactive_or_pending(u
);
3266 static void log_taint_string(Manager
*m
) {
3267 _cleanup_free_
char *taint
= NULL
;
3271 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3274 m
->taint_logged
= true; /* only check for taint once */
3276 taint
= manager_taint_string(m
);
3280 log_struct(LOG_NOTICE
,
3281 LOG_MESSAGE("System is tainted: %s", taint
),
3283 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
,
3287 static void manager_notify_finished(Manager
*m
) {
3288 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3289 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3291 if (m
->test_run_flags
)
3294 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3295 char ts
[FORMAT_TIMESPAN_MAX
];
3296 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3299 size_t size
= sizeof buf
;
3301 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3302 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3303 * negative values. */
3305 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3306 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3307 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3308 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3310 if (firmware_usec
> 0)
3311 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3312 if (loader_usec
> 0)
3313 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3315 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3317 /* The initrd case on bare-metal*/
3318 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3319 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3321 log_struct(LOG_INFO
,
3322 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3323 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3324 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3325 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3326 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3328 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3329 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3330 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3331 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3334 /* The initrd-less case on bare-metal*/
3336 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3339 log_struct(LOG_INFO
,
3340 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3341 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3342 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3343 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3345 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3346 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3347 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3351 /* The container and --user case */
3352 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3353 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3355 log_struct(LOG_INFO
,
3356 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3357 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3358 LOG_MESSAGE("Startup finished in %s.",
3359 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3363 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3366 m
->ready_sent
? "STATUS=Startup finished in %s."
3368 "STATUS=Startup finished in %s.",
3369 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3370 m
->ready_sent
= true;
3372 log_taint_string(m
);
3375 static void manager_send_ready(Manager
*m
) {
3378 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3379 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3382 m
->ready_sent
= true;
3386 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3389 static void manager_check_basic_target(Manager
*m
) {
3394 /* Small shortcut */
3395 if (m
->ready_sent
&& m
->taint_logged
)
3398 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3399 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3402 /* For user managers, send out READY=1 as soon as we reach basic.target */
3403 manager_send_ready(m
);
3405 /* Log the taint string as soon as we reach basic.target */
3406 log_taint_string(m
);
3409 void manager_check_finished(Manager
*m
) {
3412 if (MANAGER_IS_RELOADING(m
))
3415 /* Verify that we have entered the event loop already, and not left it again. */
3416 if (!MANAGER_IS_RUNNING(m
))
3419 manager_check_basic_target(m
);
3421 if (hashmap_size(m
->jobs
) > 0) {
3422 if (m
->jobs_in_progress_event_source
)
3423 /* Ignore any failure, this is only for feedback */
3424 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3429 manager_flip_auto_status(m
, false);
3431 /* Notify Type=idle units that we are done now */
3432 manager_close_idle_pipe(m
);
3434 /* Turn off confirm spawn now */
3435 m
->confirm_spawn
= NULL
;
3437 /* No need to update ask password status when we're going non-interactive */
3438 manager_close_ask_password(m
);
3440 /* This is no longer the first boot */
3441 manager_set_first_boot(m
, false);
3443 if (MANAGER_IS_FINISHED(m
))
3446 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3448 manager_notify_finished(m
);
3450 manager_invalidate_startup_units(m
);
3453 static bool generator_path_any(const char* const* paths
) {
3457 /* Optimize by skipping the whole process by not creating output directories
3458 * if no generators are found. */
3459 STRV_FOREACH(path
, (char**) paths
)
3460 if (access(*path
, F_OK
) == 0)
3462 else if (errno
!= ENOENT
)
3463 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3468 static const char* system_env_generator_binary_paths
[] = {
3469 "/run/systemd/system-environment-generators",
3470 "/etc/systemd/system-environment-generators",
3471 "/usr/local/lib/systemd/system-environment-generators",
3472 SYSTEM_ENV_GENERATOR_PATH
,
3476 static const char* user_env_generator_binary_paths
[] = {
3477 "/run/systemd/user-environment-generators",
3478 "/etc/systemd/user-environment-generators",
3479 "/usr/local/lib/systemd/user-environment-generators",
3480 USER_ENV_GENERATOR_PATH
,
3484 static int manager_run_environment_generators(Manager
*m
) {
3485 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3487 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3489 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3492 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3494 if (!generator_path_any(paths
))
3497 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3500 static int manager_run_generators(Manager
*m
) {
3501 _cleanup_strv_free_
char **paths
= NULL
;
3502 const char *argv
[5];
3507 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3510 paths
= generator_binary_paths(m
->unit_file_scope
);
3514 if (!generator_path_any((const char* const*) paths
))
3517 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3521 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3522 argv
[1] = m
->lookup_paths
.generator
;
3523 argv
[2] = m
->lookup_paths
.generator_early
;
3524 argv
[3] = m
->lookup_paths
.generator_late
;
3527 RUN_WITH_UMASK(0022)
3528 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3529 NULL
, NULL
, (char**) argv
);
3532 lookup_paths_trim_generator(&m
->lookup_paths
);
3536 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3537 char **a
= NULL
, **b
= NULL
, **l
;
3542 if (!strv_isempty(minus
)) {
3543 a
= strv_env_delete(l
, 1, minus
);
3550 if (!strv_isempty(plus
)) {
3551 b
= strv_env_merge(2, l
, plus
);
3560 if (m
->environment
!= l
)
3561 strv_free(m
->environment
);
3568 manager_sanitize_environment(m
);
3573 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3578 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3579 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3581 if (!default_rlimit
[i
])
3584 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3592 void manager_recheck_dbus(Manager
*m
) {
3595 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3596 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3597 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3598 * while in the user instance we can assume it's already there. */
3600 if (manager_dbus_is_running(m
, false)) {
3601 (void) bus_init_api(m
);
3603 if (MANAGER_IS_SYSTEM(m
))
3604 (void) bus_init_system(m
);
3606 (void) bus_done_api(m
);
3608 if (MANAGER_IS_SYSTEM(m
))
3609 (void) bus_done_system(m
);
3613 static bool manager_journal_is_running(Manager
*m
) {
3618 if (m
->test_run_flags
!= 0)
3621 /* If we are the user manager we can safely assume that the journal is up */
3622 if (!MANAGER_IS_SYSTEM(m
))
3625 /* Check that the socket is not only up, but in RUNNING state */
3626 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3629 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3632 /* Similar, check if the daemon itself is fully up, too */
3633 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3636 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3642 void manager_recheck_journal(Manager
*m
) {
3646 /* Don't bother with this unless we are in the special situation of being PID 1 */
3647 if (getpid_cached() != 1)
3650 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3651 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3652 * an activation ourselves we can't fulfill. */
3653 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3657 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3659 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3661 if (!MANAGER_IS_SYSTEM(m
))
3664 if (m
->show_status
!= mode
)
3665 log_debug("%s showing of status.",
3666 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3667 m
->show_status
= mode
;
3670 (void) touch("/run/systemd/show-status");
3672 (void) unlink("/run/systemd/show-status");
3675 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3678 if (!MANAGER_IS_SYSTEM(m
))
3681 if (m
->no_console_output
)
3684 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3687 /* If we cannot find out the status properly, just proceed. */
3688 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3691 return m
->show_status
> 0;
3694 const char *manager_get_confirm_spawn(Manager
*m
) {
3695 static int last_errno
= 0;
3696 const char *vc
= m
->confirm_spawn
;
3700 /* Here's the deal: we want to test the validity of the console but don't want
3701 * PID1 to go through the whole console process which might block. But we also
3702 * want to warn the user only once if something is wrong with the console so we
3703 * cannot do the sanity checks after spawning our children. So here we simply do
3704 * really basic tests to hopefully trap common errors.
3706 * If the console suddenly disappear at the time our children will really it
3707 * then they will simply fail to acquire it and a positive answer will be
3708 * assumed. New children will fallback to /dev/console though.
3710 * Note: TTYs are devices that can come and go any time, and frequently aren't
3711 * available yet during early boot (consider a USB rs232 dongle...). If for any
3712 * reason the configured console is not ready, we fallback to the default
3715 if (!vc
|| path_equal(vc
, "/dev/console"))
3722 if (!S_ISCHR(st
.st_mode
)) {
3730 if (last_errno
!= errno
) {
3732 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3734 return "/dev/console";
3737 void manager_set_first_boot(Manager
*m
, bool b
) {
3740 if (!MANAGER_IS_SYSTEM(m
))
3743 if (m
->first_boot
!= (int) b
) {
3745 (void) touch("/run/systemd/first-boot");
3747 (void) unlink("/run/systemd/first-boot");
3753 void manager_disable_confirm_spawn(void) {
3754 (void) touch("/run/systemd/confirm_spawn_disabled");
3757 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3758 if (!m
->confirm_spawn
)
3761 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3764 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3767 /* If m is NULL, assume we're after shutdown and let the messages through. */
3769 if (m
&& !manager_get_show_status(m
, type
))
3772 /* XXX We should totally drop the check for ephemeral here
3773 * and thus effectively make 'Type=idle' pointless. */
3774 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3777 va_start(ap
, format
);
3778 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3782 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3783 char p
[strlen(path
)+1];
3789 path_kill_slashes(p
);
3791 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3794 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3799 assert(u
->manager
== m
);
3801 size
= set_size(m
->failed_units
);
3804 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3808 if (set_put(m
->failed_units
, u
) < 0)
3811 (void) set_remove(m
->failed_units
, u
);
3813 if (set_size(m
->failed_units
) != size
)
3814 bus_manager_send_change_signal(m
);
3819 ManagerState
manager_state(Manager
*m
) {
3824 /* Did we ever finish booting? If not then we are still starting up */
3825 if (!MANAGER_IS_FINISHED(m
)) {
3827 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3828 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3829 return MANAGER_INITIALIZING
;
3831 return MANAGER_STARTING
;
3834 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
3835 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3836 if (u
&& unit_active_or_pending(u
))
3837 return MANAGER_STOPPING
;
3839 if (MANAGER_IS_SYSTEM(m
)) {
3840 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3841 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3842 if (u
&& unit_active_or_pending(u
))
3843 return MANAGER_MAINTENANCE
;
3845 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3846 if (u
&& unit_active_or_pending(u
))
3847 return MANAGER_MAINTENANCE
;
3850 /* Are there any failed units? If so, we are in degraded mode */
3851 if (set_size(m
->failed_units
) > 0)
3852 return MANAGER_DEGRADED
;
3854 return MANAGER_RUNNING
;
3857 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3859 static void manager_unref_uid_internal(
3864 int (*_clean_ipc
)(uid_t uid
)) {
3870 assert(uid_is_valid(uid
));
3873 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3874 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3876 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3877 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3878 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3879 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3881 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3882 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3884 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3887 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3889 n
= c
& ~DESTROY_IPC_FLAG
;
3893 if (destroy_now
&& n
== 0) {
3894 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3896 if (c
& DESTROY_IPC_FLAG
) {
3897 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3898 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3900 (void) _clean_ipc(uid
);
3903 c
= n
| (c
& DESTROY_IPC_FLAG
);
3904 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3908 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3909 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3912 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3913 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3916 static int manager_ref_uid_internal(
3927 assert(uid_is_valid(uid
));
3929 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3930 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3932 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3933 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3935 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3938 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3942 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3944 n
= c
& ~DESTROY_IPC_FLAG
;
3947 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3950 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3952 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3955 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3956 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3959 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3960 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3963 static void manager_vacuum_uid_refs_internal(
3966 int (*_clean_ipc
)(uid_t uid
)) {
3975 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3979 uid
= PTR_TO_UID(k
);
3980 c
= PTR_TO_UINT32(p
);
3982 n
= c
& ~DESTROY_IPC_FLAG
;
3986 if (c
& DESTROY_IPC_FLAG
) {
3987 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3988 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3990 (void) _clean_ipc(uid
);
3993 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3997 void manager_vacuum_uid_refs(Manager
*m
) {
3998 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4001 void manager_vacuum_gid_refs(Manager
*m
) {
4002 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4005 static void manager_serialize_uid_refs_internal(
4009 const char *field_name
) {
4019 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4020 * of it is better rebuild after a reload/reexec. */
4022 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4026 uid
= PTR_TO_UID(k
);
4027 c
= PTR_TO_UINT32(p
);
4029 if (!(c
& DESTROY_IPC_FLAG
))
4032 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4036 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4037 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4040 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4041 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4044 static void manager_deserialize_uid_refs_one_internal(
4047 const char *value
) {
4057 r
= parse_uid(value
, &uid
);
4058 if (r
< 0 || uid
== 0) {
4059 log_debug("Unable to parse UID reference serialization");
4063 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4069 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4070 if (c
& DESTROY_IPC_FLAG
)
4073 c
|= DESTROY_IPC_FLAG
;
4075 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4077 log_debug("Failed to add UID reference entry");
4082 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4083 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4086 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4087 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4090 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4094 char unit_name
[UNIT_NAME_MAX
+1];
4097 Manager
*m
= userdata
;
4105 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4106 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4107 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4109 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4111 if (IN_SET(errno
, EINTR
, EAGAIN
))
4114 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4117 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4118 log_warning("Received too short user lookup message, ignoring.");
4122 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4123 log_warning("Received too long user lookup message, ignoring.");
4127 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4128 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4132 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4133 if (memchr(buffer
.unit_name
, 0, n
)) {
4134 log_warning("Received lookup message with embedded NUL character, ignoring.");
4138 buffer
.unit_name
[n
] = 0;
4139 u
= manager_get_unit(m
, buffer
.unit_name
);
4141 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4145 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4147 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4151 char *manager_taint_string(Manager
*m
) {
4152 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4156 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4157 * Only things that are detected at runtime should be tagged
4158 * here. For stuff that is set during compilation, emit a warning
4159 * in the configuration phase. */
4163 buf
= new(char, sizeof("split-usr:"
4167 "overflowuid-not-65534:"
4168 "overflowgid-not-65534:"));
4176 e
= stpcpy(e
, "split-usr:");
4178 if (access("/proc/cgroups", F_OK
) < 0)
4179 e
= stpcpy(e
, "cgroups-missing:");
4181 if (clock_is_localtime(NULL
) > 0)
4182 e
= stpcpy(e
, "local-hwclock:");
4184 r
= readlink_malloc("/var/run", &destination
);
4185 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4186 e
= stpcpy(e
, "var-run-bad:");
4188 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4189 if (r
>= 0 && !streq(overflowuid
, "65534"))
4190 e
= stpcpy(e
, "overflowuid-not-65534:");
4192 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4193 if (r
>= 0 && !streq(overflowgid
, "65534"))
4194 e
= stpcpy(e
, "overflowgid-not-65534:");
4196 /* remove the last ':' */
4203 void manager_ref_console(Manager
*m
) {
4209 void manager_unref_console(Manager
*m
) {
4211 assert(m
->n_on_console
> 0);
4214 if (m
->n_on_console
== 0)
4215 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4218 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4219 [MANAGER_INITIALIZING
] = "initializing",
4220 [MANAGER_STARTING
] = "starting",
4221 [MANAGER_RUNNING
] = "running",
4222 [MANAGER_DEGRADED
] = "degraded",
4223 [MANAGER_MAINTENANCE
] = "maintenance",
4224 [MANAGER_STOPPING
] = "stopping",
4227 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4229 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4230 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4231 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4232 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4233 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4234 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4235 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4236 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4237 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4238 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4239 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4240 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4241 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4244 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);