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"
52 #include "dbus-manager.h"
53 #include "dbus-unit.h"
55 #include "dirent-util.h"
58 #include "exec-util.h"
60 #include "exit-status.h"
67 #include "locale-setup.h"
73 #include "parse-util.h"
74 #include "path-lookup.h"
75 #include "path-util.h"
76 #include "process-util.h"
77 #include "ratelimit.h"
79 #include "signal-util.h"
81 #include "stat-util.h"
82 #include "string-table.h"
83 #include "string-util.h"
85 #include "terminal-util.h"
86 #include "time-util.h"
87 #include "transaction.h"
88 #include "umask-util.h"
89 #include "unit-name.h"
90 #include "user-util.h"
95 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
96 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
98 /* Initial delay and the interval for printing status messages about running jobs */
99 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
100 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
101 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
103 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
110 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
111 static int manager_run_environment_generators(Manager
*m
);
112 static int manager_run_generators(Manager
*m
);
114 static void manager_watch_jobs_in_progress(Manager
*m
) {
120 /* We do not want to show the cylon animation if the user
121 * needs to confirm service executions otherwise confirmation
122 * messages will be screwed by the cylon animation. */
123 if (!manager_is_confirm_spawn_disabled(m
))
126 if (m
->jobs_in_progress_event_source
)
129 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
130 r
= sd_event_add_time(
132 &m
->jobs_in_progress_event_source
,
135 manager_dispatch_jobs_in_progress
, m
);
139 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
142 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
144 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
147 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
148 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
152 p
= mempset(p
, ' ', pos
-2);
153 if (log_get_show_color())
154 p
= stpcpy(p
, ANSI_RED
);
158 if (pos
> 0 && pos
<= width
) {
159 if (log_get_show_color())
160 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
164 if (log_get_show_color())
165 p
= stpcpy(p
, ANSI_NORMAL
);
168 if (log_get_show_color())
169 p
= stpcpy(p
, ANSI_RED
);
172 p
= mempset(p
, ' ', width
-1-pos
);
173 if (log_get_show_color())
174 strcpy(p
, ANSI_NORMAL
);
178 void manager_flip_auto_status(Manager
*m
, bool enable
) {
182 if (m
->show_status
== SHOW_STATUS_AUTO
)
183 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
185 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
186 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
190 static void manager_print_jobs_in_progress(Manager
*m
) {
191 _cleanup_free_
char *job_of_n
= NULL
;
194 unsigned counter
= 0, print_nr
;
195 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
197 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
201 assert(m
->n_running_jobs
> 0);
203 manager_flip_auto_status(m
, true);
205 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
207 HASHMAP_FOREACH(j
, m
->jobs
, i
)
208 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
211 /* m->n_running_jobs must be consistent with the contents of m->jobs,
212 * so the above loop must have succeeded in finding j. */
213 assert(counter
== print_nr
+ 1);
216 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
218 cylon_pos
= 14 - cylon_pos
;
219 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
221 m
->jobs_in_progress_iteration
++;
223 if (m
->n_running_jobs
> 1) {
224 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
228 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
229 if (job_get_timeout(j
, &x
) > 0)
230 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
232 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
233 "%sA %s job is running for %s (%s / %s)",
235 job_type_to_string(j
->type
),
236 unit_description(j
->unit
),
240 static int have_ask_password(void) {
241 _cleanup_closedir_
DIR *dir
;
244 dir
= opendir("/run/systemd/ask-password");
252 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
253 if (startswith(de
->d_name
, "ask."))
259 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
260 int fd
, uint32_t revents
, void *userdata
) {
261 Manager
*m
= userdata
;
267 m
->have_ask_password
= have_ask_password();
268 if (m
->have_ask_password
< 0)
269 /* Log error but continue. Negative have_ask_password
270 * is treated as unknown status. */
271 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
276 static void manager_close_ask_password(Manager
*m
) {
279 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
280 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
281 m
->have_ask_password
= -EINVAL
;
284 static int manager_check_ask_password(Manager
*m
) {
289 if (!m
->ask_password_event_source
) {
290 assert(m
->ask_password_inotify_fd
< 0);
292 mkdir_p_label("/run/systemd/ask-password", 0755);
294 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
295 if (m
->ask_password_inotify_fd
< 0)
296 return log_error_errno(errno
, "inotify_init1() failed: %m");
298 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
299 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
300 manager_close_ask_password(m
);
304 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
305 m
->ask_password_inotify_fd
, EPOLLIN
,
306 manager_dispatch_ask_password_fd
, m
);
308 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
309 manager_close_ask_password(m
);
313 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
315 /* Queries might have been added meanwhile... */
316 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
317 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
320 return m
->have_ask_password
;
323 static int manager_watch_idle_pipe(Manager
*m
) {
328 if (m
->idle_pipe_event_source
)
331 if (m
->idle_pipe
[2] < 0)
334 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
336 return log_error_errno(r
, "Failed to watch idle pipe: %m");
338 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
343 static void manager_close_idle_pipe(Manager
*m
) {
346 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
348 safe_close_pair(m
->idle_pipe
);
349 safe_close_pair(m
->idle_pipe
+ 2);
352 static int manager_setup_time_change(Manager
*m
) {
355 /* We only care for the cancellation event, hence we set the
356 * timeout to the latest possible value. */
357 struct itimerspec its
= {
358 .it_value
.tv_sec
= TIME_T_MAX
,
362 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
364 if (m
->test_run_flags
)
367 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
368 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
370 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
371 if (m
->time_change_fd
< 0)
372 return log_error_errno(errno
, "Failed to create timerfd: %m");
374 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
375 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
376 m
->time_change_fd
= safe_close(m
->time_change_fd
);
380 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
382 return log_error_errno(r
, "Failed to create time change event source: %m");
384 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
386 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
391 static int enable_special_signals(Manager
*m
) {
392 _cleanup_close_
int fd
= -1;
396 if (m
->test_run_flags
)
399 /* Enable that we get SIGINT on control-alt-del. In containers
400 * this will fail with EPERM (older) or EINVAL (newer), so
402 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
403 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
405 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
407 /* Support systems without virtual console */
409 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
411 /* Enable that we get SIGWINCH on kbrequest */
412 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
413 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
419 static int manager_setup_signals(Manager
*m
) {
420 struct sigaction sa
= {
421 .sa_handler
= SIG_DFL
,
422 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
429 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
431 /* We make liberal use of realtime signals here. On
432 * Linux/glibc we have 30 of them (with the exception of Linux
433 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
436 assert_se(sigemptyset(&mask
) == 0);
437 sigset_add_many(&mask
,
438 SIGCHLD
, /* Child died */
439 SIGTERM
, /* Reexecute daemon */
440 SIGHUP
, /* Reload configuration */
441 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
442 SIGUSR2
, /* systemd: dump status */
443 SIGINT
, /* Kernel sends us this on control-alt-del */
444 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
445 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
447 SIGRTMIN
+0, /* systemd: start default.target */
448 SIGRTMIN
+1, /* systemd: isolate rescue.target */
449 SIGRTMIN
+2, /* systemd: isolate emergency.target */
450 SIGRTMIN
+3, /* systemd: start halt.target */
451 SIGRTMIN
+4, /* systemd: start poweroff.target */
452 SIGRTMIN
+5, /* systemd: start reboot.target */
453 SIGRTMIN
+6, /* systemd: start kexec.target */
455 /* ... space for more special targets ... */
457 SIGRTMIN
+13, /* systemd: Immediate halt */
458 SIGRTMIN
+14, /* systemd: Immediate poweroff */
459 SIGRTMIN
+15, /* systemd: Immediate reboot */
460 SIGRTMIN
+16, /* systemd: Immediate kexec */
462 /* ... space for more immediate system state changes ... */
464 SIGRTMIN
+20, /* systemd: enable status messages */
465 SIGRTMIN
+21, /* systemd: disable status messages */
466 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
467 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
468 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
470 /* .. one free signal here ... */
472 #if !defined(__hppa64__) && !defined(__hppa__)
473 /* Apparently Linux on hppa has fewer RT
474 * signals (SIGRTMAX is SIGRTMIN+25 there),
475 * hence let's not try to make use of them
476 * here. Since these commands are accessible
477 * by different means and only really a safety
478 * net, the missing functionality on hppa
479 * shouldn't matter. */
481 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
482 SIGRTMIN
+27, /* systemd: set log target to console */
483 SIGRTMIN
+28, /* systemd: set log target to kmsg */
484 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
486 /* ... one free signal here SIGRTMIN+30 ... */
489 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
491 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
492 if (m
->signal_fd
< 0)
495 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
499 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
501 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
502 * notify processing can still figure out to which process/service a message belongs, before we reap the
503 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
504 * status information before detecting that there's no process in a cgroup. */
505 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
509 if (MANAGER_IS_SYSTEM(m
))
510 return enable_special_signals(m
);
515 static void manager_clean_environment(Manager
*m
) {
518 /* Let's remove some environment variables that we
519 * need ourselves to communicate with our clients */
534 static int manager_default_environment(Manager
*m
) {
537 if (MANAGER_IS_SYSTEM(m
)) {
538 /* The system manager always starts with a clean
539 * environment for its children. It does not import
540 * the kernel's or the parents' exported variables.
542 * The initial passed environment is untouched to keep
543 * /proc/self/environ valid; it is used for tagging
544 * the init process inside containers. */
545 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
548 /* Import locale variables LC_*= from configuration */
549 locale_setup(&m
->environment
);
551 /* The user manager passes its own environment
552 * along to its children. */
553 m
->environment
= strv_copy(environ
);
558 manager_clean_environment(m
);
559 strv_sort(m
->environment
);
564 static int manager_setup_prefix(Manager
*m
) {
570 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
571 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
572 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
573 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
574 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
575 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
578 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
579 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
580 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
581 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
582 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
583 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
586 const struct table_entry
*p
;
592 if (MANAGER_IS_SYSTEM(m
))
597 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
598 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
606 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
611 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
613 m
= new0(Manager
, 1);
617 m
->unit_file_scope
= scope
;
618 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
619 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
620 m
->default_tasks_accounting
= true;
621 m
->default_tasks_max
= UINT64_MAX
;
622 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
623 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
624 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
627 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
628 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
629 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
630 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
633 /* Prepare log fields we can use for structured logging */
634 if (MANAGER_IS_SYSTEM(m
)) {
635 m
->unit_log_field
= "UNIT=";
636 m
->unit_log_format_string
= "UNIT=%s";
638 m
->invocation_log_field
= "INVOCATION_ID=";
639 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
641 m
->unit_log_field
= "USER_UNIT=";
642 m
->unit_log_format_string
= "USER_UNIT=%s";
644 m
->invocation_log_field
= "USER_INVOCATION_ID=";
645 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
648 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
650 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
651 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
652 m
->ask_password_inotify_fd
= -1;
654 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
656 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
658 m
->have_ask_password
= -EINVAL
; /* we don't know */
661 m
->test_run_flags
= test_run_flags
;
663 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
664 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
666 r
= manager_default_environment(m
);
670 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
674 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
678 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
682 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
686 r
= sd_event_default(&m
->event
);
690 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
694 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
698 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
702 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
704 r
= manager_setup_signals(m
);
708 r
= manager_setup_cgroup(m
);
712 r
= manager_setup_time_change(m
);
716 m
->udev
= udev_new();
722 if (MANAGER_IS_SYSTEM(m
)) {
723 r
= mkdir_label("/run/systemd/units", 0755);
724 if (r
< 0 && r
!= -EEXIST
)
728 /* Note that we do not set up the notify fd here. We do that after deserialization,
729 * since they might have gotten serialized across the reexec. */
731 m
->taint_usr
= dir_is_empty("/usr") > 0;
733 r
= manager_setup_prefix(m
);
745 static int manager_setup_notify(Manager
*m
) {
748 if (m
->test_run_flags
)
751 if (m
->notify_fd
< 0) {
752 _cleanup_close_
int fd
= -1;
753 union sockaddr_union sa
= {
754 .sa
.sa_family
= AF_UNIX
,
756 static const int one
= 1;
758 /* First free all secondary fields */
759 m
->notify_socket
= mfree(m
->notify_socket
);
760 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
762 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
764 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
766 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
768 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
769 if (!m
->notify_socket
)
772 (void) mkdir_parents_label(m
->notify_socket
, 0755);
773 (void) unlink(m
->notify_socket
);
775 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
776 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
778 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
780 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
782 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
787 log_debug("Using notification socket %s", m
->notify_socket
);
790 if (!m
->notify_event_source
) {
791 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
793 return log_error_errno(r
, "Failed to allocate notify event source: %m");
795 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
796 * service an exit message belongs. */
797 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
799 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
801 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
807 static int manager_setup_cgroups_agent(Manager
*m
) {
809 static const union sockaddr_union sa
= {
810 .un
.sun_family
= AF_UNIX
,
811 .un
.sun_path
= "/run/systemd/cgroups-agent",
815 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
816 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
817 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
818 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
819 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
820 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
821 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
822 * we thus won't lose messages.
824 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
825 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
826 * bus for these messages. */
828 if (m
->test_run_flags
)
831 if (!MANAGER_IS_SYSTEM(m
))
834 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
836 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
837 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
840 if (m
->cgroups_agent_fd
< 0) {
841 _cleanup_close_
int fd
= -1;
843 /* First free all secondary fields */
844 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
846 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
848 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
850 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
852 (void) unlink(sa
.un
.sun_path
);
854 /* Only allow root to connect to this socket */
856 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
858 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
860 m
->cgroups_agent_fd
= fd
;
864 if (!m
->cgroups_agent_event_source
) {
865 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
867 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
869 /* Process cgroups notifications early, but after having processed service notification messages or
870 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
871 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
872 * cgroup inotify for the unified cgroup stuff. */
873 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
875 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
877 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
883 static int manager_setup_user_lookup_fd(Manager
*m
) {
888 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
889 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
890 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
891 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
892 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
893 * hence we establish this communication channel so that forked off processes can pass their UID/GID
894 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
895 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
898 * You might wonder why we need a communication channel for this that is independent of the usual notification
899 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
900 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
901 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
903 * Note that this function is called under two circumstances: when we first initialize (in which case we
904 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
905 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
907 if (m
->user_lookup_fds
[0] < 0) {
909 /* Free all secondary fields */
910 safe_close_pair(m
->user_lookup_fds
);
911 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
913 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
914 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
916 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
919 if (!m
->user_lookup_event_source
) {
920 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
922 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
924 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
926 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
928 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
930 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
936 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
937 bool try_bus_connect
;
942 if (m
->test_run_flags
)
945 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
948 (u
&& SERVICE(u
)->deserialized_state
== SERVICE_RUNNING
) &&
950 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS")));
952 /* Try to connect to the buses, if possible. */
953 return bus_init(m
, try_bus_connect
);
956 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
962 while ((u
= m
->cleanup_queue
)) {
963 assert(u
->in_cleanup_queue
);
973 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
974 GC_OFFSET_UNSURE
, /* No clue */
975 GC_OFFSET_GOOD
, /* We still need this unit */
976 GC_OFFSET_BAD
, /* We don't need this unit anymore */
980 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
985 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
987 /* Recursively mark referenced units as GOOD as well */
988 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
989 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
990 unit_gc_mark_good(other
, gc_marker
);
993 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1001 if (IN_SET(u
->gc_marker
- gc_marker
,
1002 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1005 if (u
->in_cleanup_queue
)
1008 if (unit_check_gc(u
))
1011 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1015 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1016 unit_gc_sweep(other
, gc_marker
);
1018 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1021 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1028 /* We were unable to find anything out about this entry, so
1029 * let's investigate it later */
1030 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1031 unit_add_to_gc_queue(u
);
1035 /* We definitely know that this one is not useful anymore, so
1036 * let's mark it for deletion */
1037 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1038 unit_add_to_cleanup_queue(u
);
1042 unit_gc_mark_good(u
, gc_marker
);
1045 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1046 unsigned n
= 0, gc_marker
;
1051 /* log_debug("Running GC..."); */
1053 m
->gc_marker
+= _GC_OFFSET_MAX
;
1054 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1057 gc_marker
= m
->gc_marker
;
1059 while ((u
= m
->gc_unit_queue
)) {
1060 assert(u
->in_gc_queue
);
1062 unit_gc_sweep(u
, gc_marker
);
1064 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1065 u
->in_gc_queue
= false;
1069 if (IN_SET(u
->gc_marker
- gc_marker
,
1070 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1072 log_unit_debug(u
, "Collecting.");
1073 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1074 unit_add_to_cleanup_queue(u
);
1081 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1087 while ((j
= m
->gc_job_queue
)) {
1088 assert(j
->in_gc_queue
);
1090 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1091 j
->in_gc_queue
= false;
1095 if (job_check_gc(j
))
1098 log_unit_debug(j
->unit
, "Collecting job.");
1099 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1105 static void manager_clear_jobs_and_units(Manager
*m
) {
1110 while ((u
= hashmap_first(m
->units
)))
1113 manager_dispatch_cleanup_queue(m
);
1115 assert(!m
->load_queue
);
1116 assert(!m
->run_queue
);
1117 assert(!m
->dbus_unit_queue
);
1118 assert(!m
->dbus_job_queue
);
1119 assert(!m
->cleanup_queue
);
1120 assert(!m
->gc_unit_queue
);
1121 assert(!m
->gc_job_queue
);
1123 assert(hashmap_isempty(m
->jobs
));
1124 assert(hashmap_isempty(m
->units
));
1126 m
->n_on_console
= 0;
1127 m
->n_running_jobs
= 0;
1130 Manager
* manager_free(Manager
*m
) {
1133 ExecDirectoryType dt
;
1138 manager_clear_jobs_and_units(m
);
1140 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1141 if (unit_vtable
[c
]->shutdown
)
1142 unit_vtable
[c
]->shutdown(m
);
1144 /* If we reexecute ourselves, we keep the root cgroup around */
1145 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1147 lookup_paths_flush_generator(&m
->lookup_paths
);
1151 dynamic_user_vacuum(m
, false);
1152 hashmap_free(m
->dynamic_users
);
1154 hashmap_free(m
->units
);
1155 hashmap_free(m
->units_by_invocation_id
);
1156 hashmap_free(m
->jobs
);
1157 hashmap_free(m
->watch_pids1
);
1158 hashmap_free(m
->watch_pids2
);
1159 hashmap_free(m
->watch_bus
);
1161 set_free(m
->startup_units
);
1162 set_free(m
->failed_units
);
1164 sd_event_source_unref(m
->signal_event_source
);
1165 sd_event_source_unref(m
->notify_event_source
);
1166 sd_event_source_unref(m
->cgroups_agent_event_source
);
1167 sd_event_source_unref(m
->time_change_event_source
);
1168 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1169 sd_event_source_unref(m
->run_queue_event_source
);
1170 sd_event_source_unref(m
->user_lookup_event_source
);
1172 safe_close(m
->signal_fd
);
1173 safe_close(m
->notify_fd
);
1174 safe_close(m
->cgroups_agent_fd
);
1175 safe_close(m
->time_change_fd
);
1176 safe_close_pair(m
->user_lookup_fds
);
1178 manager_close_ask_password(m
);
1180 manager_close_idle_pipe(m
);
1182 udev_unref(m
->udev
);
1183 sd_event_unref(m
->event
);
1185 free(m
->notify_socket
);
1187 lookup_paths_free(&m
->lookup_paths
);
1188 strv_free(m
->environment
);
1190 hashmap_free(m
->cgroup_unit
);
1191 set_free_free(m
->unit_path_cache
);
1193 free(m
->switch_root
);
1194 free(m
->switch_root_init
);
1196 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1197 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1199 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1200 hashmap_free(m
->units_requiring_mounts_for
);
1202 hashmap_free(m
->uid_refs
);
1203 hashmap_free(m
->gid_refs
);
1205 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1206 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1211 void manager_enumerate(Manager
*m
) {
1216 /* Let's ask every type to load all units from disk/kernel
1217 * that it might know */
1218 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1219 if (!unit_type_supported(c
)) {
1220 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1224 if (!unit_vtable
[c
]->enumerate
)
1227 unit_vtable
[c
]->enumerate(m
);
1230 manager_dispatch_load_queue(m
);
1233 static void manager_coldplug(Manager
*m
) {
1241 /* Then, let's set up their initial state. */
1242 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1244 /* ignore aliases */
1248 r
= unit_coldplug(u
);
1250 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1254 static void manager_build_unit_path_cache(Manager
*m
) {
1260 set_free_free(m
->unit_path_cache
);
1262 m
->unit_path_cache
= set_new(&string_hash_ops
);
1263 if (!m
->unit_path_cache
) {
1268 /* This simply builds a list of files we know exist, so that
1269 * we don't always have to go to disk */
1271 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1272 _cleanup_closedir_
DIR *d
= NULL
;
1277 if (errno
!= ENOENT
)
1278 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1282 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1285 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1291 r
= set_consume(m
->unit_path_cache
, p
);
1300 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1301 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1304 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1310 HASHMAP_FOREACH(u
, m
->units
, i
) {
1312 if (fdset_size(fds
) <= 0)
1315 if (!UNIT_VTABLE(u
)->distribute_fds
)
1318 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1322 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1327 /* If we are running in test mode, we still want to run the generators,
1328 * but we should not touch the real generator directories. */
1329 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1330 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1335 r
= manager_run_environment_generators(m
);
1339 /* Make sure the transient directory always exists, so that it remains
1340 * in the search path */
1341 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1343 return log_error_errno(r
, "Failed to create transient generator directory \"%s\": %m",
1344 m
->lookup_paths
.transient
);
1346 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1347 r
= manager_run_generators(m
);
1348 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1352 /* If this is the first boot, and we are in the host system, then preset everything */
1353 if (m
->first_boot
> 0 &&
1354 MANAGER_IS_SYSTEM(m
) &&
1355 !m
->test_run_flags
) {
1357 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1359 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1360 "Failed to populate /etc with preset unit settings, ignoring: %m");
1362 log_info("Populated /etc with preset unit settings.");
1365 lookup_paths_reduce(&m
->lookup_paths
);
1366 manager_build_unit_path_cache(m
);
1368 /* If we will deserialize make sure that during enumeration
1369 * this is already known, so we increase the counter here
1374 /* First, enumerate what we can from all config files */
1375 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1376 manager_enumerate(m
);
1377 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1379 /* Second, deserialize if there is something to deserialize */
1380 if (serialization
) {
1381 r
= manager_deserialize(m
, serialization
, fds
);
1383 return log_error_errno(r
, "Deserialization failed: %m");
1386 /* Any fds left? Find some unit which wants them. This is
1387 * useful to allow container managers to pass some file
1388 * descriptors to us pre-initialized. This enables
1389 * socket-based activation of entire containers. */
1390 manager_distribute_fds(m
, fds
);
1392 /* We might have deserialized the notify fd, but if we didn't
1393 * then let's create the bus now */
1394 r
= manager_setup_notify(m
);
1396 /* No sense to continue without notifications, our children would fail anyway. */
1399 r
= manager_setup_cgroups_agent(m
);
1401 /* Likewise, no sense to continue without empty cgroup notifications. */
1404 r
= manager_setup_user_lookup_fd(m
);
1406 /* This shouldn't fail, except if things are really broken. */
1409 /* Let's connect to the bus now. */
1410 (void) manager_connect_bus(m
, !!serialization
);
1412 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1413 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1415 /* Third, fire things up! */
1416 manager_coldplug(m
);
1418 /* Release any dynamic users no longer referenced */
1419 dynamic_user_vacuum(m
, true);
1421 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1422 manager_vacuum_uid_refs(m
);
1423 manager_vacuum_gid_refs(m
);
1425 if (serialization
) {
1426 assert(m
->n_reloading
> 0);
1429 /* Let's wait for the UnitNew/JobNew messages being
1430 * sent, before we notify that the reload is
1432 m
->send_reloading_done
= true;
1438 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1443 assert(type
< _JOB_TYPE_MAX
);
1445 assert(mode
< _JOB_MODE_MAX
);
1447 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1448 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1450 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1451 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1453 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1455 type
= job_type_collapse(type
, unit
);
1457 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1461 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1462 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1463 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1467 if (mode
== JOB_ISOLATE
) {
1468 r
= transaction_add_isolate_jobs(tr
, m
);
1473 r
= transaction_activate(tr
, m
, mode
, e
);
1477 log_unit_debug(unit
,
1478 "Enqueued job %s/%s as %u", unit
->id
,
1479 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1482 *_ret
= tr
->anchor_job
;
1484 transaction_free(tr
);
1488 transaction_abort(tr
);
1489 transaction_free(tr
);
1493 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1494 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1498 assert(type
< _JOB_TYPE_MAX
);
1500 assert(mode
< _JOB_MODE_MAX
);
1502 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1507 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1510 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1511 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1515 assert(type
< _JOB_TYPE_MAX
);
1517 assert(mode
< _JOB_MODE_MAX
);
1519 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1521 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1526 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1532 assert(mode
< _JOB_MODE_MAX
);
1533 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1535 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1539 /* We need an anchor job */
1540 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1544 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1545 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1547 r
= transaction_activate(tr
, m
, mode
, e
);
1551 transaction_free(tr
);
1555 transaction_abort(tr
);
1556 transaction_free(tr
);
1560 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1563 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1566 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1570 return hashmap_get(m
->units
, name
);
1573 unsigned manager_dispatch_load_queue(Manager
*m
) {
1579 /* Make sure we are not run recursively */
1580 if (m
->dispatching_load_queue
)
1583 m
->dispatching_load_queue
= true;
1585 /* Dispatches the load queue. Takes a unit from the queue and
1586 * tries to load its data until the queue is empty */
1588 while ((u
= m
->load_queue
)) {
1589 assert(u
->in_load_queue
);
1595 m
->dispatching_load_queue
= false;
1599 int manager_load_unit_prepare(
1611 assert(name
|| path
);
1614 /* This will prepare the unit for loading, but not actually
1615 * load anything from disk. */
1617 if (path
&& !is_path(path
))
1618 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1621 name
= basename(path
);
1623 t
= unit_name_to_type(name
);
1625 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1626 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1627 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1629 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1632 ret
= manager_get_unit(m
, name
);
1638 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1643 ret
->fragment_path
= strdup(path
);
1644 if (!ret
->fragment_path
) {
1650 r
= unit_add_name(ret
, name
);
1656 unit_add_to_load_queue(ret
);
1657 unit_add_to_dbus_queue(ret
);
1658 unit_add_to_gc_queue(ret
);
1665 int manager_load_unit(
1677 /* This will load the service information files, but not actually
1678 * start any services or anything. */
1680 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1684 manager_dispatch_load_queue(m
);
1686 *_ret
= unit_follow_merge(*_ret
);
1691 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1698 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1699 job_dump(j
, f
, prefix
);
1702 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1710 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1712 unit_dump(u
, f
, prefix
);
1715 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1721 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1722 char buf
[FORMAT_TIMESTAMP_MAX
];
1724 if (dual_timestamp_is_set(m
->timestamps
+ q
))
1725 fprintf(f
, "%sTimestamp %s: %s\n",
1727 manager_timestamp_to_string(q
),
1728 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
1731 manager_dump_units(m
, f
, prefix
);
1732 manager_dump_jobs(m
, f
, prefix
);
1735 int manager_get_dump_string(Manager
*m
, char **ret
) {
1736 _cleanup_free_
char *dump
= NULL
;
1737 _cleanup_fclose_
FILE *f
= NULL
;
1744 f
= open_memstream(&dump
, &size
);
1748 __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1750 manager_dump(m
, f
, NULL
);
1752 r
= fflush_and_check(f
);
1764 void manager_clear_jobs(Manager
*m
) {
1769 while ((j
= hashmap_first(m
->jobs
)))
1770 /* No need to recurse. We're cancelling all jobs. */
1771 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1774 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1775 Manager
*m
= userdata
;
1781 while ((j
= m
->run_queue
)) {
1782 assert(j
->installed
);
1783 assert(j
->in_run_queue
);
1785 job_run_and_invalidate(j
);
1788 if (m
->n_running_jobs
> 0)
1789 manager_watch_jobs_in_progress(m
);
1791 if (m
->n_on_console
> 0)
1792 manager_watch_idle_pipe(m
);
1797 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1804 if (m
->dispatching_dbus_queue
)
1807 m
->dispatching_dbus_queue
= true;
1809 while ((u
= m
->dbus_unit_queue
)) {
1810 assert(u
->in_dbus_queue
);
1812 bus_unit_send_change_signal(u
);
1816 while ((j
= m
->dbus_job_queue
)) {
1817 assert(j
->in_dbus_queue
);
1819 bus_job_send_change_signal(j
);
1823 m
->dispatching_dbus_queue
= false;
1825 if (m
->send_reloading_done
) {
1826 m
->send_reloading_done
= false;
1828 bus_manager_send_reloading(m
, false);
1831 if (m
->queued_message
)
1832 bus_send_queued_message(m
);
1837 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1838 Manager
*m
= userdata
;
1839 char buf
[PATH_MAX
+1];
1842 n
= recv(fd
, buf
, sizeof(buf
), 0);
1844 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1846 log_error("Got zero-length cgroups agent message, ignoring.");
1849 if ((size_t) n
>= sizeof(buf
)) {
1850 log_error("Got overly long cgroups agent message, ignoring.");
1854 if (memchr(buf
, 0, n
)) {
1855 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1860 manager_notify_cgroup_empty(m
, buf
);
1861 (void) bus_forward_agent_released(m
, buf
);
1866 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1867 _cleanup_strv_free_
char **tags
= NULL
;
1873 tags
= strv_split(buf
, "\n\r");
1879 if (UNIT_VTABLE(u
)->notify_message
)
1880 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1881 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1882 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1886 y
= ellipsize(x
, 20, 90);
1887 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1891 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1893 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1894 Manager
*m
= userdata
;
1895 char buf
[NOTIFY_BUFFER_MAX
+1];
1896 struct iovec iovec
= {
1898 .iov_len
= sizeof(buf
)-1,
1901 struct cmsghdr cmsghdr
;
1902 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1903 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1905 struct msghdr msghdr
= {
1908 .msg_control
= &control
,
1909 .msg_controllen
= sizeof(control
),
1912 struct cmsghdr
*cmsg
;
1913 struct ucred
*ucred
= NULL
;
1915 int r
, *fd_array
= NULL
;
1920 assert(m
->notify_fd
== fd
);
1922 if (revents
!= EPOLLIN
) {
1923 log_warning("Got unexpected poll event for notify fd.");
1927 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1929 if (IN_SET(errno
, EAGAIN
, EINTR
))
1930 return 0; /* Spurious wakeup, try again */
1932 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1933 * won't take notification messages anymore, but that's still better than busy looping around this:
1934 * being woken up over and over again but being unable to actually read the message off the socket. */
1935 return log_error_errno(errno
, "Failed to receive notification message: %m");
1938 CMSG_FOREACH(cmsg
, &msghdr
) {
1939 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1941 fd_array
= (int*) CMSG_DATA(cmsg
);
1942 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1944 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1945 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1946 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1948 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1955 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1957 close_many(fd_array
, n_fds
);
1963 if (!ucred
|| ucred
->pid
<= 0) {
1964 log_warning("Received notify message without valid credentials. Ignoring.");
1968 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1969 log_warning("Received notify message exceeded maximum size. Ignoring.");
1973 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1974 * trailing NUL byte in the message, but don't expect it. */
1975 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1976 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1980 /* Make sure it's NUL-terminated. */
1983 /* Notify every unit that might be interested, but try
1984 * to avoid notifying the same one multiple times. */
1985 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1987 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1989 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1991 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1993 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1994 if (u3
&& u3
!= u2
&& u3
!= u1
)
1995 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1997 if (!u1
&& !u2
&& !u3
)
1998 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
2000 if (fdset_size(fds
) > 0)
2001 log_warning("Got extra auxiliary fds with notification message, closing them.");
2006 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
2013 sd_event_get_iteration(m
->event
, &iteration
);
2015 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
2017 unit_unwatch_pid(u
, si
->si_pid
);
2019 if (UNIT_VTABLE(u
)->sigchld_event
) {
2020 if (set_size(u
->pids
) <= 1 ||
2021 iteration
!= u
->sigchldgen
||
2022 unit_main_pid(u
) == si
->si_pid
||
2023 unit_control_pid(u
) == si
->si_pid
) {
2024 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2025 u
->sigchldgen
= iteration
;
2027 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
2031 static int manager_dispatch_sigchld(Manager
*m
) {
2037 /* First we call waitd() for a PID and do not reap the
2038 * zombie. That way we can still access /proc/$PID for
2039 * it while it is a zombie. */
2040 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2042 if (errno
== ECHILD
)
2054 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2055 _cleanup_free_
char *name
= NULL
;
2058 get_process_comm(si
.si_pid
, &name
);
2060 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2061 si
.si_pid
, strna(name
),
2062 sigchld_code_to_string(si
.si_code
),
2064 strna(si
.si_code
== CLD_EXITED
2065 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2066 : signal_to_string(si
.si_status
)));
2068 /* And now figure out the unit this belongs
2069 * to, it might be multiple... */
2070 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2072 invoke_sigchld_event(m
, u1
, &si
);
2073 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
2075 invoke_sigchld_event(m
, u2
, &si
);
2076 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
2077 if (u3
&& u3
!= u2
&& u3
!= u1
)
2078 invoke_sigchld_event(m
, u3
, &si
);
2081 /* And now, we actually reap the zombie. */
2082 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2093 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2094 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2097 log_debug("Activating special unit %s", name
);
2099 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2101 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2104 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2105 /* If the user presses C-A-D more than
2106 * 7 times within 2s, we reboot/shutdown immediately,
2107 * unless it was disabled in system.conf */
2109 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2110 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2112 emergency_action(m
, m
->cad_burst_action
, NULL
,
2113 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2116 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2117 Manager
*m
= userdata
;
2119 struct signalfd_siginfo sfsi
;
2120 bool sigchld
= false;
2124 assert(m
->signal_fd
== fd
);
2126 if (revents
!= EPOLLIN
) {
2127 log_warning("Got unexpected events from signal file descriptor.");
2132 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2133 if (n
!= sizeof(sfsi
)) {
2135 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
2139 if (IN_SET(errno
, EINTR
, EAGAIN
))
2142 /* We return an error here, which will kill this handler,
2143 * to avoid a busy loop on read error. */
2144 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2147 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2148 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2149 ? LOG_DEBUG
: LOG_INFO
,
2152 switch (sfsi
.ssi_signo
) {
2159 if (MANAGER_IS_SYSTEM(m
)) {
2160 /* This is for compatibility with the
2161 * original sysvinit */
2162 r
= verify_run_space_and_log("Refusing to reexecute");
2164 m
->exit_code
= MANAGER_REEXECUTE
;
2170 if (MANAGER_IS_SYSTEM(m
))
2171 manager_handle_ctrl_alt_del(m
);
2173 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2174 JOB_REPLACE_IRREVERSIBLY
);
2178 if (MANAGER_IS_SYSTEM(m
))
2179 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2181 /* This is a nop on non-init */
2185 if (MANAGER_IS_SYSTEM(m
))
2186 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2188 /* This is a nop on non-init */
2194 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2196 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2197 log_info("Trying to reconnect to bus...");
2201 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2202 log_info("Loading D-Bus service...");
2203 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2210 _cleanup_free_
char *dump
= NULL
;
2212 r
= manager_get_dump_string(m
, &dump
);
2214 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2218 log_dump(LOG_INFO
, dump
);
2223 r
= verify_run_space_and_log("Refusing to reload");
2225 m
->exit_code
= MANAGER_RELOAD
;
2230 /* Starting SIGRTMIN+0 */
2231 static const struct {
2234 } target_table
[] = {
2235 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2236 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2237 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2238 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2239 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2240 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2241 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
}
2244 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2245 static const ManagerExitCode code_table
[] = {
2247 [1] = MANAGER_POWEROFF
,
2248 [2] = MANAGER_REBOOT
,
2252 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2253 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2254 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2255 manager_start_target(m
, target_table
[idx
].target
,
2256 target_table
[idx
].mode
);
2260 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2261 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2262 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2266 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2269 manager_set_show_status(m
, SHOW_STATUS_YES
);
2273 manager_set_show_status(m
, SHOW_STATUS_NO
);
2277 log_set_max_level(LOG_DEBUG
);
2278 log_info("Setting log level to debug.");
2282 log_set_max_level(LOG_INFO
);
2283 log_info("Setting log level to info.");
2287 if (MANAGER_IS_USER(m
)) {
2288 m
->exit_code
= MANAGER_EXIT
;
2292 /* This is a nop on init */
2296 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2297 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2298 log_notice("Setting log target to journal-or-kmsg.");
2302 log_set_target(LOG_TARGET_CONSOLE
);
2303 log_notice("Setting log target to console.");
2307 log_set_target(LOG_TARGET_KMSG
);
2308 log_notice("Setting log target to kmsg.");
2312 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2319 manager_dispatch_sigchld(m
);
2324 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2325 Manager
*m
= userdata
;
2330 assert(m
->time_change_fd
== fd
);
2332 log_struct(LOG_DEBUG
,
2333 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2334 LOG_MESSAGE("Time has been changed"),
2337 /* Restart the watch */
2338 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2339 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2341 manager_setup_time_change(m
);
2343 HASHMAP_FOREACH(u
, m
->units
, i
)
2344 if (UNIT_VTABLE(u
)->time_change
)
2345 UNIT_VTABLE(u
)->time_change(u
);
2350 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2351 Manager
*m
= userdata
;
2354 assert(m
->idle_pipe
[2] == fd
);
2356 m
->no_console_output
= m
->n_on_console
> 0;
2358 manager_close_idle_pipe(m
);
2363 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2364 Manager
*m
= userdata
;
2371 manager_print_jobs_in_progress(m
);
2373 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2374 r
= sd_event_source_set_time(source
, next
);
2378 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2381 int manager_loop(Manager
*m
) {
2384 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2387 m
->exit_code
= MANAGER_OK
;
2389 /* Release the path cache */
2390 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2392 manager_check_finished(m
);
2394 /* There might still be some zombies hanging around from
2395 * before we were exec()'ed. Let's reap them. */
2396 r
= manager_dispatch_sigchld(m
);
2400 while (m
->exit_code
== MANAGER_OK
) {
2403 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2406 if (!ratelimit_test(&rl
)) {
2407 /* Yay, something is going seriously wrong, pause a little */
2408 log_warning("Looping too fast. Throttling execution a little.");
2412 if (manager_dispatch_load_queue(m
) > 0)
2415 if (manager_dispatch_gc_job_queue(m
) > 0)
2418 if (manager_dispatch_gc_unit_queue(m
) > 0)
2421 if (manager_dispatch_cleanup_queue(m
) > 0)
2424 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2427 if (manager_dispatch_dbus_queue(m
) > 0)
2430 /* Sleep for half the watchdog time */
2431 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2432 wait_usec
= m
->runtime_watchdog
/ 2;
2436 wait_usec
= USEC_INFINITY
;
2438 r
= sd_event_run(m
->event
, wait_usec
);
2440 return log_error_errno(r
, "Failed to run event loop: %m");
2443 return m
->exit_code
;
2446 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2447 _cleanup_free_
char *n
= NULL
;
2448 sd_id128_t invocation_id
;
2456 r
= unit_name_from_dbus_path(s
, &n
);
2460 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2461 * as invocation ID. */
2462 r
= sd_id128_from_string(n
, &invocation_id
);
2464 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2470 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
));
2473 /* If this didn't work, we check if this is a unit name */
2474 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2475 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2477 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2485 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2495 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2499 r
= safe_atou(p
, &id
);
2503 j
= manager_get_job(m
, id
);
2512 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2515 _cleanup_free_
char *p
= NULL
;
2519 if (!MANAGER_IS_SYSTEM(m
))
2522 audit_fd
= get_audit_fd();
2526 /* Don't generate audit events if the service was already
2527 * started and we're just deserializing */
2528 if (MANAGER_IS_RELOADING(m
))
2531 if (u
->type
!= UNIT_SERVICE
)
2534 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2536 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2540 msg
= strjoina("unit=", p
);
2541 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2543 /* We aren't allowed to send audit messages?
2544 * Then let's not retry again. */
2547 log_warning_errno(errno
, "Failed to send audit message: %m");
2553 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2554 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2555 _cleanup_free_
char *message
= NULL
;
2556 _cleanup_close_
int fd
= -1;
2559 /* Don't generate plymouth events if the service was already
2560 * started and we're just deserializing */
2561 if (MANAGER_IS_RELOADING(m
))
2564 if (!MANAGER_IS_SYSTEM(m
))
2567 if (detect_container() > 0)
2570 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2573 /* We set SOCK_NONBLOCK here so that we rather drop the
2574 * message then wait for plymouth */
2575 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2577 log_error_errno(errno
, "socket() failed: %m");
2581 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2583 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2584 log_error_errno(errno
, "connect() failed: %m");
2588 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2594 if (write(fd
, message
, n
+ 1) != n
+ 1)
2595 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2596 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2599 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2605 fd
= open_serialization_fd("systemd-state");
2609 f
= fdopen(fd
, "w+");
2619 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2632 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2633 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2634 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2635 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2636 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2638 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2639 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
2640 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
2643 t
= manager_timestamp_to_string(q
);
2645 char field
[strlen(t
) + strlen("-timestamp") + 1];
2646 strcpy(stpcpy(field
, t
), "-timestamp");
2647 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
2651 if (!switching_root
)
2652 (void) serialize_environment(f
, m
->environment
);
2654 if (m
->notify_fd
>= 0) {
2657 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2661 fprintf(f
, "notify-fd=%i\n", copy
);
2662 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2665 if (m
->cgroups_agent_fd
>= 0) {
2668 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2672 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2675 if (m
->user_lookup_fds
[0] >= 0) {
2678 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2682 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2686 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2689 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2691 r
= dynamic_user_serialize(m
, f
, fds
);
2695 manager_serialize_uid_refs(m
, f
);
2696 manager_serialize_gid_refs(m
, f
);
2698 fputc_unlocked('\n', f
);
2700 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2705 fputs_unlocked(u
->id
, f
);
2706 fputc_unlocked('\n', f
);
2708 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2715 assert(m
->n_reloading
> 0);
2721 r
= bus_fdset_add_all(m
, fds
);
2728 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2734 log_debug("Deserializing state...");
2739 char line
[LINE_MAX
];
2740 const char *val
, *l
;
2742 if (!fgets(line
, sizeof(line
), f
)) {
2757 if ((val
= startswith(l
, "current-job-id="))) {
2760 if (safe_atou32(val
, &id
) < 0)
2761 log_notice("Failed to parse current job id value %s", val
);
2763 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2765 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2768 if (safe_atou32(val
, &n
) < 0)
2769 log_notice("Failed to parse installed jobs counter %s", val
);
2771 m
->n_installed_jobs
+= n
;
2773 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2776 if (safe_atou32(val
, &n
) < 0)
2777 log_notice("Failed to parse failed jobs counter %s", val
);
2779 m
->n_failed_jobs
+= n
;
2781 } else if ((val
= startswith(l
, "taint-usr="))) {
2784 b
= parse_boolean(val
);
2786 log_notice("Failed to parse taint /usr flag %s", val
);
2788 m
->taint_usr
= m
->taint_usr
|| b
;
2790 } else if ((val
= startswith(l
, "ready-sent="))) {
2793 b
= parse_boolean(val
);
2795 log_notice("Failed to parse ready-sent flag %s", val
);
2797 m
->ready_sent
= m
->ready_sent
|| b
;
2799 } else if (startswith(l
, "env=")) {
2800 r
= deserialize_environment(&m
->environment
, l
);
2804 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2806 } else if ((val
= startswith(l
, "notify-fd="))) {
2809 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2810 log_notice("Failed to parse notify fd: \"%s\"", val
);
2812 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2813 safe_close(m
->notify_fd
);
2814 m
->notify_fd
= fdset_remove(fds
, fd
);
2817 } else if ((val
= startswith(l
, "notify-socket="))) {
2826 free(m
->notify_socket
);
2827 m
->notify_socket
= n
;
2829 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2832 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2833 log_notice("Failed to parse cgroups agent fd: %s", val
);
2835 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2836 safe_close(m
->cgroups_agent_fd
);
2837 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2840 } else if ((val
= startswith(l
, "user-lookup="))) {
2843 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2844 log_notice("Failed to parse user lookup fd: %s", val
);
2846 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2847 safe_close_pair(m
->user_lookup_fds
);
2848 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2849 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2852 } else if ((val
= startswith(l
, "dynamic-user=")))
2853 dynamic_user_deserialize_one(m
, val
, fds
);
2854 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2855 manager_deserialize_uid_refs_one(m
, val
);
2856 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2857 manager_deserialize_gid_refs_one(m
, val
);
2858 else if ((val
= startswith(l
, "subscribed="))) {
2860 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2865 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2866 val
= startswith(l
, manager_timestamp_to_string(q
));
2870 val
= startswith(val
, "-timestamp=");
2875 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
2876 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
2877 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
2878 log_notice("Unknown serialization item '%s'", l
);
2884 char name
[UNIT_NAME_MAX
+2];
2885 const char* unit_name
;
2888 if (!fgets(name
, sizeof(name
), f
)) {
2898 unit_name
= strstrip(name
);
2900 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
2902 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
2905 unit_deserialize_skip(f
);
2909 r
= unit_deserialize(u
, f
, fds
);
2911 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
2921 assert(m
->n_reloading
> 0);
2927 int manager_reload(Manager
*m
) {
2929 _cleanup_fclose_
FILE *f
= NULL
;
2930 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2934 r
= manager_open_serialization(m
, &f
);
2939 bus_manager_send_reloading(m
, true);
2947 r
= manager_serialize(m
, f
, fds
, false);
2953 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2958 /* From here on there is no way back. */
2959 manager_clear_jobs_and_units(m
);
2960 lookup_paths_flush_generator(&m
->lookup_paths
);
2961 lookup_paths_free(&m
->lookup_paths
);
2962 dynamic_user_vacuum(m
, false);
2963 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2964 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2966 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2967 if (q
< 0 && r
>= 0)
2970 q
= manager_run_environment_generators(m
);
2971 if (q
< 0 && r
>= 0)
2974 /* Find new unit paths */
2975 q
= manager_run_generators(m
);
2976 if (q
< 0 && r
>= 0)
2979 lookup_paths_reduce(&m
->lookup_paths
);
2980 manager_build_unit_path_cache(m
);
2982 /* First, enumerate what we can from all config files */
2983 manager_enumerate(m
);
2985 /* Second, deserialize our stored data */
2986 q
= manager_deserialize(m
, f
, fds
);
2988 log_error_errno(q
, "Deserialization failed: %m");
2997 /* Re-register notify_fd as event source */
2998 q
= manager_setup_notify(m
);
2999 if (q
< 0 && r
>= 0)
3002 q
= manager_setup_cgroups_agent(m
);
3003 if (q
< 0 && r
>= 0)
3006 q
= manager_setup_user_lookup_fd(m
);
3007 if (q
< 0 && r
>= 0)
3010 /* Third, fire things up! */
3011 manager_coldplug(m
);
3013 /* Release any dynamic users no longer referenced */
3014 dynamic_user_vacuum(m
, true);
3016 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3017 manager_vacuum_uid_refs(m
);
3018 manager_vacuum_gid_refs(m
);
3020 /* Sync current state of bus names with our set of listening units */
3022 manager_sync_bus_names(m
, m
->api_bus
);
3024 assert(m
->n_reloading
> 0);
3027 m
->send_reloading_done
= true;
3032 void manager_reset_failed(Manager
*m
) {
3038 HASHMAP_FOREACH(u
, m
->units
, i
)
3039 unit_reset_failed(u
);
3042 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3048 /* Returns true if the unit is inactive or going down */
3049 u
= manager_get_unit(m
, name
);
3053 return unit_inactive_or_pending(u
);
3056 static void manager_notify_finished(Manager
*m
) {
3057 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3058 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3060 if (m
->test_run_flags
)
3063 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3065 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3066 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3067 * negative values. */
3069 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3070 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3071 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3072 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3074 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3076 /* The initrd case on bare-metal*/
3077 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3078 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3080 log_struct(LOG_INFO
,
3081 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3082 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3083 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3084 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3085 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3086 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3087 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3088 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3089 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3092 /* The initrd-less case on bare-metal*/
3094 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3097 log_struct(LOG_INFO
,
3098 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3099 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3100 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3101 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3102 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3103 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3104 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3108 /* The container case */
3109 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3110 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3112 log_struct(LOG_INFO
,
3113 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3114 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3115 LOG_MESSAGE("Startup finished in %s.",
3116 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3120 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3123 m
->ready_sent
? "STATUS=Startup finished in %s."
3125 "STATUS=Startup finished in %s.",
3126 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3127 m
->ready_sent
= true;
3130 void manager_check_finished(Manager
*m
) {
3133 if (MANAGER_IS_RELOADING(m
))
3136 /* Verify that we are actually running currently. Initially
3137 * the exit code is set to invalid, and during operation it is
3138 * then set to MANAGER_OK */
3139 if (m
->exit_code
!= MANAGER_OK
)
3142 /* For user managers, send out READY=1 as soon as we reach basic.target */
3143 if (MANAGER_IS_USER(m
) && !m
->ready_sent
) {
3146 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3150 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3151 m
->ready_sent
= true;
3155 if (hashmap_size(m
->jobs
) > 0) {
3156 if (m
->jobs_in_progress_event_source
)
3157 /* Ignore any failure, this is only for feedback */
3158 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3163 manager_flip_auto_status(m
, false);
3165 /* Notify Type=idle units that we are done now */
3166 manager_close_idle_pipe(m
);
3168 /* Turn off confirm spawn now */
3169 m
->confirm_spawn
= NULL
;
3171 /* No need to update ask password status when we're going non-interactive */
3172 manager_close_ask_password(m
);
3174 /* This is no longer the first boot */
3175 manager_set_first_boot(m
, false);
3177 if (MANAGER_IS_FINISHED(m
))
3180 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3182 manager_notify_finished(m
);
3184 manager_invalidate_startup_units(m
);
3187 static bool generator_path_any(const char* const* paths
) {
3191 /* Optimize by skipping the whole process by not creating output directories
3192 * if no generators are found. */
3193 STRV_FOREACH(path
, (char**) paths
)
3194 if (access(*path
, F_OK
) == 0)
3196 else if (errno
!= ENOENT
)
3197 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3202 static const char* system_env_generator_binary_paths
[] = {
3203 "/run/systemd/system-environment-generators",
3204 "/etc/systemd/system-environment-generators",
3205 "/usr/local/lib/systemd/system-environment-generators",
3206 SYSTEM_ENV_GENERATOR_PATH
,
3210 static const char* user_env_generator_binary_paths
[] = {
3211 "/run/systemd/user-environment-generators",
3212 "/etc/systemd/user-environment-generators",
3213 "/usr/local/lib/systemd/user-environment-generators",
3214 USER_ENV_GENERATOR_PATH
,
3218 static int manager_run_environment_generators(Manager
*m
) {
3219 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3221 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3223 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3226 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3228 if (!generator_path_any(paths
))
3231 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3234 static int manager_run_generators(Manager
*m
) {
3235 _cleanup_strv_free_
char **paths
= NULL
;
3236 const char *argv
[5];
3241 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3244 paths
= generator_binary_paths(m
->unit_file_scope
);
3248 if (!generator_path_any((const char* const*) paths
))
3251 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3255 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3256 argv
[1] = m
->lookup_paths
.generator
;
3257 argv
[2] = m
->lookup_paths
.generator_early
;
3258 argv
[3] = m
->lookup_paths
.generator_late
;
3261 RUN_WITH_UMASK(0022)
3262 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3263 NULL
, NULL
, (char**) argv
);
3266 lookup_paths_trim_generator(&m
->lookup_paths
);
3270 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3271 char **a
= NULL
, **b
= NULL
, **l
;
3276 if (!strv_isempty(minus
)) {
3277 a
= strv_env_delete(l
, 1, minus
);
3284 if (!strv_isempty(plus
)) {
3285 b
= strv_env_merge(2, l
, plus
);
3294 if (m
->environment
!= l
)
3295 strv_free(m
->environment
);
3302 manager_clean_environment(m
);
3303 strv_sort(m
->environment
);
3308 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3313 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3314 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3316 if (!default_rlimit
[i
])
3319 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3327 void manager_recheck_journal(Manager
*m
) {
3332 if (!MANAGER_IS_SYSTEM(m
))
3335 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3336 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3337 log_close_journal();
3341 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3342 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3343 log_close_journal();
3347 /* Hmm, OK, so the socket is fully up and the service is up
3348 * too, then let's make use of the thing. */
3352 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3354 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3356 if (!MANAGER_IS_SYSTEM(m
))
3359 if (m
->show_status
!= mode
)
3360 log_debug("%s showing of status.",
3361 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3362 m
->show_status
= mode
;
3365 (void) touch("/run/systemd/show-status");
3367 (void) unlink("/run/systemd/show-status");
3370 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3373 if (!MANAGER_IS_SYSTEM(m
))
3376 if (m
->no_console_output
)
3379 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3382 /* If we cannot find out the status properly, just proceed. */
3383 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3386 if (m
->show_status
> 0)
3392 const char *manager_get_confirm_spawn(Manager
*m
) {
3393 static int last_errno
= 0;
3394 const char *vc
= m
->confirm_spawn
;
3398 /* Here's the deal: we want to test the validity of the console but don't want
3399 * PID1 to go through the whole console process which might block. But we also
3400 * want to warn the user only once if something is wrong with the console so we
3401 * cannot do the sanity checks after spawning our children. So here we simply do
3402 * really basic tests to hopefully trap common errors.
3404 * If the console suddenly disappear at the time our children will really it
3405 * then they will simply fail to acquire it and a positive answer will be
3406 * assumed. New children will fallback to /dev/console though.
3408 * Note: TTYs are devices that can come and go any time, and frequently aren't
3409 * available yet during early boot (consider a USB rs232 dongle...). If for any
3410 * reason the configured console is not ready, we fallback to the default
3413 if (!vc
|| path_equal(vc
, "/dev/console"))
3420 if (!S_ISCHR(st
.st_mode
)) {
3428 if (last_errno
!= errno
) {
3430 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3432 return "/dev/console";
3435 void manager_set_first_boot(Manager
*m
, bool b
) {
3438 if (!MANAGER_IS_SYSTEM(m
))
3441 if (m
->first_boot
!= (int) b
) {
3443 (void) touch("/run/systemd/first-boot");
3445 (void) unlink("/run/systemd/first-boot");
3451 void manager_disable_confirm_spawn(void) {
3452 (void) touch("/run/systemd/confirm_spawn_disabled");
3455 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3456 if (!m
->confirm_spawn
)
3459 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3462 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3465 /* If m is NULL, assume we're after shutdown and let the messages through. */
3467 if (m
&& !manager_get_show_status(m
, type
))
3470 /* XXX We should totally drop the check for ephemeral here
3471 * and thus effectively make 'Type=idle' pointless. */
3472 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3475 va_start(ap
, format
);
3476 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3480 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3481 char p
[strlen(path
)+1];
3487 path_kill_slashes(p
);
3489 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3492 void manager_set_exec_params(Manager
*m
, ExecParameters
*p
) {
3496 p
->environment
= m
->environment
;
3497 p
->confirm_spawn
= manager_get_confirm_spawn(m
);
3498 p
->cgroup_supported
= m
->cgroup_supported
;
3499 p
->prefix
= m
->prefix
;
3501 SET_FLAG(p
->flags
, EXEC_PASS_LOG_UNIT
|EXEC_CHOWN_DIRECTORIES
, MANAGER_IS_SYSTEM(m
));
3504 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3509 assert(u
->manager
== m
);
3511 size
= set_size(m
->failed_units
);
3514 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3518 if (set_put(m
->failed_units
, u
) < 0)
3521 (void) set_remove(m
->failed_units
, u
);
3523 if (set_size(m
->failed_units
) != size
)
3524 bus_manager_send_change_signal(m
);
3529 ManagerState
manager_state(Manager
*m
) {
3534 /* Did we ever finish booting? If not then we are still starting up */
3535 if (!MANAGER_IS_FINISHED(m
)) {
3537 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3538 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3539 return MANAGER_INITIALIZING
;
3541 return MANAGER_STARTING
;
3544 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
3545 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3546 if (u
&& unit_active_or_pending(u
))
3547 return MANAGER_STOPPING
;
3549 if (MANAGER_IS_SYSTEM(m
)) {
3550 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3551 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3552 if (u
&& unit_active_or_pending(u
))
3553 return MANAGER_MAINTENANCE
;
3555 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3556 if (u
&& unit_active_or_pending(u
))
3557 return MANAGER_MAINTENANCE
;
3560 /* Are there any failed units? If so, we are in degraded mode */
3561 if (set_size(m
->failed_units
) > 0)
3562 return MANAGER_DEGRADED
;
3564 return MANAGER_RUNNING
;
3567 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3569 static void manager_unref_uid_internal(
3574 int (*_clean_ipc
)(uid_t uid
)) {
3580 assert(uid_is_valid(uid
));
3583 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3584 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3586 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3587 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3588 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3589 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3591 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3592 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3594 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3597 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3599 n
= c
& ~DESTROY_IPC_FLAG
;
3603 if (destroy_now
&& n
== 0) {
3604 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3606 if (c
& DESTROY_IPC_FLAG
) {
3607 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3608 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3610 (void) _clean_ipc(uid
);
3613 c
= n
| (c
& DESTROY_IPC_FLAG
);
3614 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3618 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3619 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3622 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3623 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3626 static int manager_ref_uid_internal(
3637 assert(uid_is_valid(uid
));
3639 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3640 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3642 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3643 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3645 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3648 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3652 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3654 n
= c
& ~DESTROY_IPC_FLAG
;
3657 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3660 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3662 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3665 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3666 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3669 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3670 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3673 static void manager_vacuum_uid_refs_internal(
3676 int (*_clean_ipc
)(uid_t uid
)) {
3685 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3689 uid
= PTR_TO_UID(k
);
3690 c
= PTR_TO_UINT32(p
);
3692 n
= c
& ~DESTROY_IPC_FLAG
;
3696 if (c
& DESTROY_IPC_FLAG
) {
3697 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3698 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3700 (void) _clean_ipc(uid
);
3703 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3707 void manager_vacuum_uid_refs(Manager
*m
) {
3708 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3711 void manager_vacuum_gid_refs(Manager
*m
) {
3712 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3715 static void manager_serialize_uid_refs_internal(
3719 const char *field_name
) {
3729 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3730 * of it is better rebuild after a reload/reexec. */
3732 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3736 uid
= PTR_TO_UID(k
);
3737 c
= PTR_TO_UINT32(p
);
3739 if (!(c
& DESTROY_IPC_FLAG
))
3742 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3746 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3747 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3750 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3751 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3754 static void manager_deserialize_uid_refs_one_internal(
3757 const char *value
) {
3767 r
= parse_uid(value
, &uid
);
3768 if (r
< 0 || uid
== 0) {
3769 log_debug("Unable to parse UID reference serialization");
3773 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3779 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3780 if (c
& DESTROY_IPC_FLAG
)
3783 c
|= DESTROY_IPC_FLAG
;
3785 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3787 log_debug("Failed to add UID reference entry");
3792 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3793 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3796 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3797 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3800 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3804 char unit_name
[UNIT_NAME_MAX
+1];
3807 Manager
*m
= userdata
;
3815 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3816 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3817 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3819 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3821 if (IN_SET(errno
, EINTR
, EAGAIN
))
3824 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3827 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3828 log_warning("Received too short user lookup message, ignoring.");
3832 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3833 log_warning("Received too long user lookup message, ignoring.");
3837 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3838 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3842 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3843 if (memchr(buffer
.unit_name
, 0, n
)) {
3844 log_warning("Received lookup message with embedded NUL character, ignoring.");
3848 buffer
.unit_name
[n
] = 0;
3849 u
= manager_get_unit(m
, buffer
.unit_name
);
3851 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3855 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3857 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3861 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3862 [MANAGER_INITIALIZING
] = "initializing",
3863 [MANAGER_STARTING
] = "starting",
3864 [MANAGER_RUNNING
] = "running",
3865 [MANAGER_DEGRADED
] = "degraded",
3866 [MANAGER_MAINTENANCE
] = "maintenance",
3867 [MANAGER_STOPPING
] = "stopping",
3870 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
3872 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
3873 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
3874 [MANAGER_TIMESTAMP_LOADER
] = "loader",
3875 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
3876 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
3877 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
3878 [MANAGER_TIMESTAMP_FINISH
] = "finish",
3879 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
3880 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
3881 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
3882 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
3883 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
3884 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
3887 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);