1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <stdio_ext.h>
27 #include <sys/epoll.h>
28 #include <sys/inotify.h>
29 #include <sys/ioctl.h>
30 #include <sys/reboot.h>
31 #include <sys/timerfd.h>
39 #include "sd-daemon.h"
40 #include "sd-messages.h"
43 #include "alloc-util.h"
45 #include "boot-timestamps.h"
46 #include "bus-common-errors.h"
47 #include "bus-error.h"
48 #include "bus-kernel.h"
50 #include "clean-ipc.h"
51 #include "clock-util.h"
53 #include "dbus-manager.h"
54 #include "dbus-unit.h"
56 #include "dirent-util.h"
59 #include "exec-util.h"
61 #include "exit-status.h"
68 #include "locale-setup.h"
74 #include "parse-util.h"
75 #include "path-lookup.h"
76 #include "path-util.h"
77 #include "process-util.h"
78 #include "ratelimit.h"
80 #include "signal-util.h"
82 #include "stat-util.h"
83 #include "string-table.h"
84 #include "string-util.h"
86 #include "terminal-util.h"
87 #include "time-util.h"
88 #include "transaction.h"
89 #include "umask-util.h"
90 #include "unit-name.h"
91 #include "user-util.h"
96 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
97 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
99 /* Initial delay and the interval for printing status messages about running jobs */
100 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
101 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
102 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
104 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
111 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
112 static int manager_run_environment_generators(Manager
*m
);
113 static int manager_run_generators(Manager
*m
);
115 static void manager_watch_jobs_in_progress(Manager
*m
) {
121 /* We do not want to show the cylon animation if the user
122 * needs to confirm service executions otherwise confirmation
123 * messages will be screwed by the cylon animation. */
124 if (!manager_is_confirm_spawn_disabled(m
))
127 if (m
->jobs_in_progress_event_source
)
130 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
131 r
= sd_event_add_time(
133 &m
->jobs_in_progress_event_source
,
136 manager_dispatch_jobs_in_progress
, m
);
140 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
143 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
145 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
148 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
149 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
153 p
= mempset(p
, ' ', pos
-2);
154 if (log_get_show_color())
155 p
= stpcpy(p
, ANSI_RED
);
159 if (pos
> 0 && pos
<= width
) {
160 if (log_get_show_color())
161 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
165 if (log_get_show_color())
166 p
= stpcpy(p
, ANSI_NORMAL
);
169 if (log_get_show_color())
170 p
= stpcpy(p
, ANSI_RED
);
173 p
= mempset(p
, ' ', width
-1-pos
);
174 if (log_get_show_color())
175 strcpy(p
, ANSI_NORMAL
);
179 void manager_flip_auto_status(Manager
*m
, bool enable
) {
183 if (m
->show_status
== SHOW_STATUS_AUTO
)
184 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
186 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
187 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
191 static void manager_print_jobs_in_progress(Manager
*m
) {
192 _cleanup_free_
char *job_of_n
= NULL
;
195 unsigned counter
= 0, print_nr
;
196 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
198 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
202 assert(m
->n_running_jobs
> 0);
204 manager_flip_auto_status(m
, true);
206 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
208 HASHMAP_FOREACH(j
, m
->jobs
, i
)
209 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
212 /* m->n_running_jobs must be consistent with the contents of m->jobs,
213 * so the above loop must have succeeded in finding j. */
214 assert(counter
== print_nr
+ 1);
217 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
219 cylon_pos
= 14 - cylon_pos
;
220 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
222 m
->jobs_in_progress_iteration
++;
224 if (m
->n_running_jobs
> 1) {
225 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
229 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
230 if (job_get_timeout(j
, &x
) > 0)
231 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
233 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
234 "%sA %s job is running for %s (%s / %s)",
236 job_type_to_string(j
->type
),
237 unit_description(j
->unit
),
241 static int have_ask_password(void) {
242 _cleanup_closedir_
DIR *dir
;
245 dir
= opendir("/run/systemd/ask-password");
253 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
254 if (startswith(de
->d_name
, "ask."))
260 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
261 int fd
, uint32_t revents
, void *userdata
) {
262 Manager
*m
= userdata
;
268 m
->have_ask_password
= have_ask_password();
269 if (m
->have_ask_password
< 0)
270 /* Log error but continue. Negative have_ask_password
271 * is treated as unknown status. */
272 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
277 static void manager_close_ask_password(Manager
*m
) {
280 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
281 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
282 m
->have_ask_password
= -EINVAL
;
285 static int manager_check_ask_password(Manager
*m
) {
290 if (!m
->ask_password_event_source
) {
291 assert(m
->ask_password_inotify_fd
< 0);
293 mkdir_p_label("/run/systemd/ask-password", 0755);
295 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
296 if (m
->ask_password_inotify_fd
< 0)
297 return log_error_errno(errno
, "inotify_init1() failed: %m");
299 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
300 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
301 manager_close_ask_password(m
);
305 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
306 m
->ask_password_inotify_fd
, EPOLLIN
,
307 manager_dispatch_ask_password_fd
, m
);
309 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
310 manager_close_ask_password(m
);
314 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
316 /* Queries might have been added meanwhile... */
317 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
318 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
321 return m
->have_ask_password
;
324 static int manager_watch_idle_pipe(Manager
*m
) {
329 if (m
->idle_pipe_event_source
)
332 if (m
->idle_pipe
[2] < 0)
335 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
337 return log_error_errno(r
, "Failed to watch idle pipe: %m");
339 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
344 static void manager_close_idle_pipe(Manager
*m
) {
347 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
349 safe_close_pair(m
->idle_pipe
);
350 safe_close_pair(m
->idle_pipe
+ 2);
353 static int manager_setup_time_change(Manager
*m
) {
356 /* We only care for the cancellation event, hence we set the
357 * timeout to the latest possible value. */
358 struct itimerspec its
= {
359 .it_value
.tv_sec
= TIME_T_MAX
,
363 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
365 if (m
->test_run_flags
)
368 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
369 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
371 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
372 if (m
->time_change_fd
< 0)
373 return log_error_errno(errno
, "Failed to create timerfd: %m");
375 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
376 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
377 m
->time_change_fd
= safe_close(m
->time_change_fd
);
381 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
383 return log_error_errno(r
, "Failed to create time change event source: %m");
385 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
387 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
392 static int enable_special_signals(Manager
*m
) {
393 _cleanup_close_
int fd
= -1;
397 if (m
->test_run_flags
)
400 /* Enable that we get SIGINT on control-alt-del. In containers
401 * this will fail with EPERM (older) or EINVAL (newer), so
403 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
404 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
406 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
408 /* Support systems without virtual console */
410 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
412 /* Enable that we get SIGWINCH on kbrequest */
413 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
414 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
420 static int manager_setup_signals(Manager
*m
) {
421 struct sigaction sa
= {
422 .sa_handler
= SIG_DFL
,
423 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
430 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
432 /* We make liberal use of realtime signals here. On
433 * Linux/glibc we have 30 of them (with the exception of Linux
434 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
437 assert_se(sigemptyset(&mask
) == 0);
438 sigset_add_many(&mask
,
439 SIGCHLD
, /* Child died */
440 SIGTERM
, /* Reexecute daemon */
441 SIGHUP
, /* Reload configuration */
442 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
443 SIGUSR2
, /* systemd: dump status */
444 SIGINT
, /* Kernel sends us this on control-alt-del */
445 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
446 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
448 SIGRTMIN
+0, /* systemd: start default.target */
449 SIGRTMIN
+1, /* systemd: isolate rescue.target */
450 SIGRTMIN
+2, /* systemd: isolate emergency.target */
451 SIGRTMIN
+3, /* systemd: start halt.target */
452 SIGRTMIN
+4, /* systemd: start poweroff.target */
453 SIGRTMIN
+5, /* systemd: start reboot.target */
454 SIGRTMIN
+6, /* systemd: start kexec.target */
456 /* ... space for more special targets ... */
458 SIGRTMIN
+13, /* systemd: Immediate halt */
459 SIGRTMIN
+14, /* systemd: Immediate poweroff */
460 SIGRTMIN
+15, /* systemd: Immediate reboot */
461 SIGRTMIN
+16, /* systemd: Immediate kexec */
463 /* ... space for more immediate system state changes ... */
465 SIGRTMIN
+20, /* systemd: enable status messages */
466 SIGRTMIN
+21, /* systemd: disable status messages */
467 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
468 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
469 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
471 /* .. one free signal here ... */
473 #if !defined(__hppa64__) && !defined(__hppa__)
474 /* Apparently Linux on hppa has fewer RT
475 * signals (SIGRTMAX is SIGRTMIN+25 there),
476 * hence let's not try to make use of them
477 * here. Since these commands are accessible
478 * by different means and only really a safety
479 * net, the missing functionality on hppa
480 * shouldn't matter. */
482 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
483 SIGRTMIN
+27, /* systemd: set log target to console */
484 SIGRTMIN
+28, /* systemd: set log target to kmsg */
485 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
487 /* ... one free signal here SIGRTMIN+30 ... */
490 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
492 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
493 if (m
->signal_fd
< 0)
496 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
500 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
502 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
503 * notify processing can still figure out to which process/service a message belongs, before we reap the
504 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
505 * status information before detecting that there's no process in a cgroup. */
506 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
510 if (MANAGER_IS_SYSTEM(m
))
511 return enable_special_signals(m
);
516 static void manager_clean_environment(Manager
*m
) {
519 /* Let's remove some environment variables that we
520 * need ourselves to communicate with our clients */
535 static int manager_default_environment(Manager
*m
) {
538 if (MANAGER_IS_SYSTEM(m
)) {
539 /* The system manager always starts with a clean
540 * environment for its children. It does not import
541 * the kernel's or the parents' exported variables.
543 * The initial passed environment is untouched to keep
544 * /proc/self/environ valid; it is used for tagging
545 * the init process inside containers. */
546 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
549 /* Import locale variables LC_*= from configuration */
550 locale_setup(&m
->environment
);
552 /* The user manager passes its own environment
553 * along to its children. */
554 m
->environment
= strv_copy(environ
);
559 manager_clean_environment(m
);
560 strv_sort(m
->environment
);
565 static int manager_setup_prefix(Manager
*m
) {
571 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
572 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
573 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
574 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
575 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
576 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
579 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
580 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
581 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
582 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
583 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
584 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
587 const struct table_entry
*p
;
593 if (MANAGER_IS_SYSTEM(m
))
598 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
599 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
607 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
612 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
614 m
= new0(Manager
, 1);
618 m
->unit_file_scope
= scope
;
619 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
620 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
621 m
->default_tasks_accounting
= true;
622 m
->default_tasks_max
= UINT64_MAX
;
623 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
624 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
625 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
628 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
629 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
630 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
631 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
634 /* Prepare log fields we can use for structured logging */
635 if (MANAGER_IS_SYSTEM(m
)) {
636 m
->unit_log_field
= "UNIT=";
637 m
->unit_log_format_string
= "UNIT=%s";
639 m
->invocation_log_field
= "INVOCATION_ID=";
640 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
642 m
->unit_log_field
= "USER_UNIT=";
643 m
->unit_log_format_string
= "USER_UNIT=%s";
645 m
->invocation_log_field
= "USER_INVOCATION_ID=";
646 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
649 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
651 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
652 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
653 m
->ask_password_inotify_fd
= -1;
655 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
657 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
659 m
->have_ask_password
= -EINVAL
; /* we don't know */
662 m
->test_run_flags
= test_run_flags
;
664 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
665 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
667 r
= manager_default_environment(m
);
671 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
675 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
679 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
683 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
687 r
= sd_event_default(&m
->event
);
691 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
695 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
699 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
703 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
705 r
= manager_setup_signals(m
);
709 r
= manager_setup_cgroup(m
);
713 r
= manager_setup_time_change(m
);
717 m
->udev
= udev_new();
723 if (MANAGER_IS_SYSTEM(m
)) {
724 r
= mkdir_label("/run/systemd/units", 0755);
725 if (r
< 0 && r
!= -EEXIST
)
729 /* Note that we do not set up the notify fd here. We do that after deserialization,
730 * since they might have gotten serialized across the reexec. */
732 m
->taint_usr
= dir_is_empty("/usr") > 0;
734 r
= manager_setup_prefix(m
);
746 static int manager_setup_notify(Manager
*m
) {
749 if (m
->test_run_flags
)
752 if (m
->notify_fd
< 0) {
753 _cleanup_close_
int fd
= -1;
754 union sockaddr_union sa
= {
755 .sa
.sa_family
= AF_UNIX
,
757 static const int one
= 1;
759 /* First free all secondary fields */
760 m
->notify_socket
= mfree(m
->notify_socket
);
761 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
763 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
765 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
767 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
769 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
770 if (!m
->notify_socket
)
773 (void) mkdir_parents_label(m
->notify_socket
, 0755);
774 (void) unlink(m
->notify_socket
);
776 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
777 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
779 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
781 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
783 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
788 log_debug("Using notification socket %s", m
->notify_socket
);
791 if (!m
->notify_event_source
) {
792 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
794 return log_error_errno(r
, "Failed to allocate notify event source: %m");
796 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
797 * service an exit message belongs. */
798 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
800 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
802 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
808 static int manager_setup_cgroups_agent(Manager
*m
) {
810 static const union sockaddr_union sa
= {
811 .un
.sun_family
= AF_UNIX
,
812 .un
.sun_path
= "/run/systemd/cgroups-agent",
816 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
817 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
818 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
819 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
820 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
821 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
822 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
823 * we thus won't lose messages.
825 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
826 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
827 * bus for these messages. */
829 if (m
->test_run_flags
)
832 if (!MANAGER_IS_SYSTEM(m
))
835 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
837 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
838 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
841 if (m
->cgroups_agent_fd
< 0) {
842 _cleanup_close_
int fd
= -1;
844 /* First free all secondary fields */
845 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
847 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
849 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
851 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
853 (void) unlink(sa
.un
.sun_path
);
855 /* Only allow root to connect to this socket */
857 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
859 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
861 m
->cgroups_agent_fd
= fd
;
865 if (!m
->cgroups_agent_event_source
) {
866 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
868 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
870 /* Process cgroups notifications early, but after having processed service notification messages or
871 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
872 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
873 * cgroup inotify for the unified cgroup stuff. */
874 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
876 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
878 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
884 static int manager_setup_user_lookup_fd(Manager
*m
) {
889 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
890 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
891 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
892 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
893 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
894 * hence we establish this communication channel so that forked off processes can pass their UID/GID
895 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
896 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
899 * You might wonder why we need a communication channel for this that is independent of the usual notification
900 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
901 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
902 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
904 * Note that this function is called under two circumstances: when we first initialize (in which case we
905 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
906 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
908 if (m
->user_lookup_fds
[0] < 0) {
910 /* Free all secondary fields */
911 safe_close_pair(m
->user_lookup_fds
);
912 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
914 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
915 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
917 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
920 if (!m
->user_lookup_event_source
) {
921 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
923 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
925 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
927 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
929 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
931 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
937 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
938 bool try_bus_connect
;
943 if (m
->test_run_flags
)
946 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
949 (u
&& SERVICE(u
)->deserialized_state
== SERVICE_RUNNING
) &&
951 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS")));
953 /* Try to connect to the buses, if possible. */
954 return bus_init(m
, try_bus_connect
);
957 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
963 while ((u
= m
->cleanup_queue
)) {
964 assert(u
->in_cleanup_queue
);
974 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
975 GC_OFFSET_UNSURE
, /* No clue */
976 GC_OFFSET_GOOD
, /* We still need this unit */
977 GC_OFFSET_BAD
, /* We don't need this unit anymore */
981 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
986 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
988 /* Recursively mark referenced units as GOOD as well */
989 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
990 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
991 unit_gc_mark_good(other
, gc_marker
);
994 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1002 if (IN_SET(u
->gc_marker
- gc_marker
,
1003 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1006 if (u
->in_cleanup_queue
)
1009 if (unit_check_gc(u
))
1012 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1016 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1017 unit_gc_sweep(other
, gc_marker
);
1019 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1022 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1029 /* We were unable to find anything out about this entry, so
1030 * let's investigate it later */
1031 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1032 unit_add_to_gc_queue(u
);
1036 /* We definitely know that this one is not useful anymore, so
1037 * let's mark it for deletion */
1038 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1039 unit_add_to_cleanup_queue(u
);
1043 unit_gc_mark_good(u
, gc_marker
);
1046 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1047 unsigned n
= 0, gc_marker
;
1052 /* log_debug("Running GC..."); */
1054 m
->gc_marker
+= _GC_OFFSET_MAX
;
1055 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1058 gc_marker
= m
->gc_marker
;
1060 while ((u
= m
->gc_unit_queue
)) {
1061 assert(u
->in_gc_queue
);
1063 unit_gc_sweep(u
, gc_marker
);
1065 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1066 u
->in_gc_queue
= false;
1070 if (IN_SET(u
->gc_marker
- gc_marker
,
1071 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1073 log_unit_debug(u
, "Collecting.");
1074 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1075 unit_add_to_cleanup_queue(u
);
1082 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1088 while ((j
= m
->gc_job_queue
)) {
1089 assert(j
->in_gc_queue
);
1091 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1092 j
->in_gc_queue
= false;
1096 if (job_check_gc(j
))
1099 log_unit_debug(j
->unit
, "Collecting job.");
1100 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1106 static void manager_clear_jobs_and_units(Manager
*m
) {
1111 while ((u
= hashmap_first(m
->units
)))
1114 manager_dispatch_cleanup_queue(m
);
1116 assert(!m
->load_queue
);
1117 assert(!m
->run_queue
);
1118 assert(!m
->dbus_unit_queue
);
1119 assert(!m
->dbus_job_queue
);
1120 assert(!m
->cleanup_queue
);
1121 assert(!m
->gc_unit_queue
);
1122 assert(!m
->gc_job_queue
);
1124 assert(hashmap_isempty(m
->jobs
));
1125 assert(hashmap_isempty(m
->units
));
1127 m
->n_on_console
= 0;
1128 m
->n_running_jobs
= 0;
1131 Manager
* manager_free(Manager
*m
) {
1134 ExecDirectoryType dt
;
1139 manager_clear_jobs_and_units(m
);
1141 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1142 if (unit_vtable
[c
]->shutdown
)
1143 unit_vtable
[c
]->shutdown(m
);
1145 /* If we reexecute ourselves, we keep the root cgroup around */
1146 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1148 lookup_paths_flush_generator(&m
->lookup_paths
);
1152 dynamic_user_vacuum(m
, false);
1153 hashmap_free(m
->dynamic_users
);
1155 hashmap_free(m
->units
);
1156 hashmap_free(m
->units_by_invocation_id
);
1157 hashmap_free(m
->jobs
);
1158 hashmap_free(m
->watch_pids1
);
1159 hashmap_free(m
->watch_pids2
);
1160 hashmap_free(m
->watch_bus
);
1162 set_free(m
->startup_units
);
1163 set_free(m
->failed_units
);
1165 sd_event_source_unref(m
->signal_event_source
);
1166 sd_event_source_unref(m
->notify_event_source
);
1167 sd_event_source_unref(m
->cgroups_agent_event_source
);
1168 sd_event_source_unref(m
->time_change_event_source
);
1169 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1170 sd_event_source_unref(m
->run_queue_event_source
);
1171 sd_event_source_unref(m
->user_lookup_event_source
);
1173 safe_close(m
->signal_fd
);
1174 safe_close(m
->notify_fd
);
1175 safe_close(m
->cgroups_agent_fd
);
1176 safe_close(m
->time_change_fd
);
1177 safe_close_pair(m
->user_lookup_fds
);
1179 manager_close_ask_password(m
);
1181 manager_close_idle_pipe(m
);
1183 udev_unref(m
->udev
);
1184 sd_event_unref(m
->event
);
1186 free(m
->notify_socket
);
1188 lookup_paths_free(&m
->lookup_paths
);
1189 strv_free(m
->environment
);
1191 hashmap_free(m
->cgroup_unit
);
1192 set_free_free(m
->unit_path_cache
);
1194 free(m
->switch_root
);
1195 free(m
->switch_root_init
);
1197 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1198 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1200 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1201 hashmap_free(m
->units_requiring_mounts_for
);
1203 hashmap_free(m
->uid_refs
);
1204 hashmap_free(m
->gid_refs
);
1206 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1207 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1212 void manager_enumerate(Manager
*m
) {
1217 /* Let's ask every type to load all units from disk/kernel
1218 * that it might know */
1219 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1220 if (!unit_type_supported(c
)) {
1221 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1225 if (!unit_vtable
[c
]->enumerate
)
1228 unit_vtable
[c
]->enumerate(m
);
1231 manager_dispatch_load_queue(m
);
1234 static void manager_coldplug(Manager
*m
) {
1242 /* Then, let's set up their initial state. */
1243 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1245 /* ignore aliases */
1249 r
= unit_coldplug(u
);
1251 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1255 static void manager_build_unit_path_cache(Manager
*m
) {
1261 set_free_free(m
->unit_path_cache
);
1263 m
->unit_path_cache
= set_new(&string_hash_ops
);
1264 if (!m
->unit_path_cache
) {
1269 /* This simply builds a list of files we know exist, so that
1270 * we don't always have to go to disk */
1272 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1273 _cleanup_closedir_
DIR *d
= NULL
;
1278 if (errno
!= ENOENT
)
1279 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1283 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1286 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1292 r
= set_consume(m
->unit_path_cache
, p
);
1301 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1302 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1305 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1311 HASHMAP_FOREACH(u
, m
->units
, i
) {
1313 if (fdset_size(fds
) <= 0)
1316 if (!UNIT_VTABLE(u
)->distribute_fds
)
1319 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1323 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1328 /* If we are running in test mode, we still want to run the generators,
1329 * but we should not touch the real generator directories. */
1330 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1331 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1336 r
= manager_run_environment_generators(m
);
1340 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1341 r
= manager_run_generators(m
);
1342 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1346 /* If this is the first boot, and we are in the host system, then preset everything */
1347 if (m
->first_boot
> 0 &&
1348 MANAGER_IS_SYSTEM(m
) &&
1349 !m
->test_run_flags
) {
1351 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1353 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1354 "Failed to populate /etc with preset unit settings, ignoring: %m");
1356 log_info("Populated /etc with preset unit settings.");
1359 lookup_paths_reduce(&m
->lookup_paths
);
1360 manager_build_unit_path_cache(m
);
1362 /* If we will deserialize make sure that during enumeration
1363 * this is already known, so we increase the counter here
1368 /* First, enumerate what we can from all config files */
1369 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1370 manager_enumerate(m
);
1371 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1373 /* Second, deserialize if there is something to deserialize */
1374 if (serialization
) {
1375 r
= manager_deserialize(m
, serialization
, fds
);
1377 return log_error_errno(r
, "Deserialization failed: %m");
1380 /* Any fds left? Find some unit which wants them. This is
1381 * useful to allow container managers to pass some file
1382 * descriptors to us pre-initialized. This enables
1383 * socket-based activation of entire containers. */
1384 manager_distribute_fds(m
, fds
);
1386 /* We might have deserialized the notify fd, but if we didn't
1387 * then let's create the bus now */
1388 r
= manager_setup_notify(m
);
1390 /* No sense to continue without notifications, our children would fail anyway. */
1393 r
= manager_setup_cgroups_agent(m
);
1395 /* Likewise, no sense to continue without empty cgroup notifications. */
1398 r
= manager_setup_user_lookup_fd(m
);
1400 /* This shouldn't fail, except if things are really broken. */
1403 /* Let's connect to the bus now. */
1404 (void) manager_connect_bus(m
, !!serialization
);
1406 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1407 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1409 /* Third, fire things up! */
1410 manager_coldplug(m
);
1412 /* Release any dynamic users no longer referenced */
1413 dynamic_user_vacuum(m
, true);
1415 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1416 manager_vacuum_uid_refs(m
);
1417 manager_vacuum_gid_refs(m
);
1419 if (serialization
) {
1420 assert(m
->n_reloading
> 0);
1423 /* Let's wait for the UnitNew/JobNew messages being
1424 * sent, before we notify that the reload is
1426 m
->send_reloading_done
= true;
1432 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1437 assert(type
< _JOB_TYPE_MAX
);
1439 assert(mode
< _JOB_MODE_MAX
);
1441 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1442 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1444 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1445 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1447 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1449 type
= job_type_collapse(type
, unit
);
1451 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1455 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1456 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1457 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1461 if (mode
== JOB_ISOLATE
) {
1462 r
= transaction_add_isolate_jobs(tr
, m
);
1467 r
= transaction_activate(tr
, m
, mode
, e
);
1471 log_unit_debug(unit
,
1472 "Enqueued job %s/%s as %u", unit
->id
,
1473 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1476 *_ret
= tr
->anchor_job
;
1478 transaction_free(tr
);
1482 transaction_abort(tr
);
1483 transaction_free(tr
);
1487 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1488 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1492 assert(type
< _JOB_TYPE_MAX
);
1494 assert(mode
< _JOB_MODE_MAX
);
1496 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1501 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1504 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1505 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1509 assert(type
< _JOB_TYPE_MAX
);
1511 assert(mode
< _JOB_MODE_MAX
);
1513 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1515 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1520 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1526 assert(mode
< _JOB_MODE_MAX
);
1527 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1529 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1533 /* We need an anchor job */
1534 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1538 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1539 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1541 r
= transaction_activate(tr
, m
, mode
, e
);
1545 transaction_free(tr
);
1549 transaction_abort(tr
);
1550 transaction_free(tr
);
1554 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1557 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1560 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1564 return hashmap_get(m
->units
, name
);
1567 unsigned manager_dispatch_load_queue(Manager
*m
) {
1573 /* Make sure we are not run recursively */
1574 if (m
->dispatching_load_queue
)
1577 m
->dispatching_load_queue
= true;
1579 /* Dispatches the load queue. Takes a unit from the queue and
1580 * tries to load its data until the queue is empty */
1582 while ((u
= m
->load_queue
)) {
1583 assert(u
->in_load_queue
);
1589 m
->dispatching_load_queue
= false;
1593 int manager_load_unit_prepare(
1605 assert(name
|| path
);
1608 /* This will prepare the unit for loading, but not actually
1609 * load anything from disk. */
1611 if (path
&& !is_path(path
))
1612 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1615 name
= basename(path
);
1617 t
= unit_name_to_type(name
);
1619 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1620 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1621 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1623 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1626 ret
= manager_get_unit(m
, name
);
1632 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1637 ret
->fragment_path
= strdup(path
);
1638 if (!ret
->fragment_path
) {
1644 r
= unit_add_name(ret
, name
);
1650 unit_add_to_load_queue(ret
);
1651 unit_add_to_dbus_queue(ret
);
1652 unit_add_to_gc_queue(ret
);
1659 int manager_load_unit(
1671 /* This will load the service information files, but not actually
1672 * start any services or anything. */
1674 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1678 manager_dispatch_load_queue(m
);
1680 *_ret
= unit_follow_merge(*_ret
);
1685 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1692 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1693 job_dump(j
, f
, prefix
);
1696 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1704 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1706 unit_dump(u
, f
, prefix
);
1709 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1715 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
1716 char buf
[FORMAT_TIMESTAMP_MAX
];
1718 if (dual_timestamp_is_set(m
->timestamps
+ q
))
1719 fprintf(f
, "%sTimestamp %s: %s\n",
1721 manager_timestamp_to_string(q
),
1722 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
1725 manager_dump_units(m
, f
, prefix
);
1726 manager_dump_jobs(m
, f
, prefix
);
1729 int manager_get_dump_string(Manager
*m
, char **ret
) {
1730 _cleanup_free_
char *dump
= NULL
;
1731 _cleanup_fclose_
FILE *f
= NULL
;
1738 f
= open_memstream(&dump
, &size
);
1742 __fsetlocking(f
, FSETLOCKING_BYCALLER
);
1744 manager_dump(m
, f
, NULL
);
1746 r
= fflush_and_check(f
);
1758 void manager_clear_jobs(Manager
*m
) {
1763 while ((j
= hashmap_first(m
->jobs
)))
1764 /* No need to recurse. We're cancelling all jobs. */
1765 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1768 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1769 Manager
*m
= userdata
;
1775 while ((j
= m
->run_queue
)) {
1776 assert(j
->installed
);
1777 assert(j
->in_run_queue
);
1779 job_run_and_invalidate(j
);
1782 if (m
->n_running_jobs
> 0)
1783 manager_watch_jobs_in_progress(m
);
1785 if (m
->n_on_console
> 0)
1786 manager_watch_idle_pipe(m
);
1791 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1798 if (m
->dispatching_dbus_queue
)
1801 m
->dispatching_dbus_queue
= true;
1803 while ((u
= m
->dbus_unit_queue
)) {
1804 assert(u
->in_dbus_queue
);
1806 bus_unit_send_change_signal(u
);
1810 while ((j
= m
->dbus_job_queue
)) {
1811 assert(j
->in_dbus_queue
);
1813 bus_job_send_change_signal(j
);
1817 m
->dispatching_dbus_queue
= false;
1819 if (m
->send_reloading_done
) {
1820 m
->send_reloading_done
= false;
1822 bus_manager_send_reloading(m
, false);
1825 if (m
->queued_message
)
1826 bus_send_queued_message(m
);
1831 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1832 Manager
*m
= userdata
;
1833 char buf
[PATH_MAX
+1];
1836 n
= recv(fd
, buf
, sizeof(buf
), 0);
1838 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1840 log_error("Got zero-length cgroups agent message, ignoring.");
1843 if ((size_t) n
>= sizeof(buf
)) {
1844 log_error("Got overly long cgroups agent message, ignoring.");
1848 if (memchr(buf
, 0, n
)) {
1849 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1854 manager_notify_cgroup_empty(m
, buf
);
1855 (void) bus_forward_agent_released(m
, buf
);
1860 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1861 _cleanup_strv_free_
char **tags
= NULL
;
1867 tags
= strv_split(buf
, "\n\r");
1873 if (UNIT_VTABLE(u
)->notify_message
)
1874 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1875 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1876 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1880 y
= ellipsize(x
, 20, 90);
1881 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1885 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1887 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1888 Manager
*m
= userdata
;
1889 char buf
[NOTIFY_BUFFER_MAX
+1];
1890 struct iovec iovec
= {
1892 .iov_len
= sizeof(buf
)-1,
1895 struct cmsghdr cmsghdr
;
1896 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1897 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1899 struct msghdr msghdr
= {
1902 .msg_control
= &control
,
1903 .msg_controllen
= sizeof(control
),
1906 struct cmsghdr
*cmsg
;
1907 struct ucred
*ucred
= NULL
;
1909 int r
, *fd_array
= NULL
;
1914 assert(m
->notify_fd
== fd
);
1916 if (revents
!= EPOLLIN
) {
1917 log_warning("Got unexpected poll event for notify fd.");
1921 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1923 if (IN_SET(errno
, EAGAIN
, EINTR
))
1924 return 0; /* Spurious wakeup, try again */
1926 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1927 * won't take notification messages anymore, but that's still better than busy looping around this:
1928 * being woken up over and over again but being unable to actually read the message off the socket. */
1929 return log_error_errno(errno
, "Failed to receive notification message: %m");
1932 CMSG_FOREACH(cmsg
, &msghdr
) {
1933 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1935 fd_array
= (int*) CMSG_DATA(cmsg
);
1936 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1938 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1939 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1940 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1942 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1949 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1951 close_many(fd_array
, n_fds
);
1957 if (!ucred
|| ucred
->pid
<= 0) {
1958 log_warning("Received notify message without valid credentials. Ignoring.");
1962 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1963 log_warning("Received notify message exceeded maximum size. Ignoring.");
1967 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1968 * trailing NUL byte in the message, but don't expect it. */
1969 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1970 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1974 /* Make sure it's NUL-terminated. */
1977 /* Notify every unit that might be interested, but try
1978 * to avoid notifying the same one multiple times. */
1979 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1981 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1983 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1985 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1987 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1988 if (u3
&& u3
!= u2
&& u3
!= u1
)
1989 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1991 if (!u1
&& !u2
&& !u3
)
1992 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1994 if (fdset_size(fds
) > 0)
1995 log_warning("Got extra auxiliary fds with notification message, closing them.");
2000 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
2007 sd_event_get_iteration(m
->event
, &iteration
);
2009 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
2011 unit_unwatch_pid(u
, si
->si_pid
);
2013 if (UNIT_VTABLE(u
)->sigchld_event
) {
2014 if (set_size(u
->pids
) <= 1 ||
2015 iteration
!= u
->sigchldgen
||
2016 unit_main_pid(u
) == si
->si_pid
||
2017 unit_control_pid(u
) == si
->si_pid
) {
2018 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2019 u
->sigchldgen
= iteration
;
2021 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
2025 static int manager_dispatch_sigchld(Manager
*m
) {
2031 /* First we call waitd() for a PID and do not reap the
2032 * zombie. That way we can still access /proc/$PID for
2033 * it while it is a zombie. */
2034 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2036 if (errno
== ECHILD
)
2048 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2049 _cleanup_free_
char *name
= NULL
;
2052 get_process_comm(si
.si_pid
, &name
);
2054 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2055 si
.si_pid
, strna(name
),
2056 sigchld_code_to_string(si
.si_code
),
2058 strna(si
.si_code
== CLD_EXITED
2059 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2060 : signal_to_string(si
.si_status
)));
2062 /* And now figure out the unit this belongs
2063 * to, it might be multiple... */
2064 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2066 invoke_sigchld_event(m
, u1
, &si
);
2067 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
2069 invoke_sigchld_event(m
, u2
, &si
);
2070 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
2071 if (u3
&& u3
!= u2
&& u3
!= u1
)
2072 invoke_sigchld_event(m
, u3
, &si
);
2075 /* And now, we actually reap the zombie. */
2076 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2087 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2088 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2091 log_debug("Activating special unit %s", name
);
2093 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2095 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2098 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2099 /* If the user presses C-A-D more than
2100 * 7 times within 2s, we reboot/shutdown immediately,
2101 * unless it was disabled in system.conf */
2103 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2104 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2106 emergency_action(m
, m
->cad_burst_action
, NULL
,
2107 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2110 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2111 Manager
*m
= userdata
;
2113 struct signalfd_siginfo sfsi
;
2114 bool sigchld
= false;
2118 assert(m
->signal_fd
== fd
);
2120 if (revents
!= EPOLLIN
) {
2121 log_warning("Got unexpected events from signal file descriptor.");
2126 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2127 if (n
!= sizeof(sfsi
)) {
2129 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
2133 if (IN_SET(errno
, EINTR
, EAGAIN
))
2136 /* We return an error here, which will kill this handler,
2137 * to avoid a busy loop on read error. */
2138 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2141 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2142 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2143 ? LOG_DEBUG
: LOG_INFO
,
2146 switch (sfsi
.ssi_signo
) {
2153 if (MANAGER_IS_SYSTEM(m
)) {
2154 /* This is for compatibility with the
2155 * original sysvinit */
2156 r
= verify_run_space_and_log("Refusing to reexecute");
2158 m
->exit_code
= MANAGER_REEXECUTE
;
2164 if (MANAGER_IS_SYSTEM(m
))
2165 manager_handle_ctrl_alt_del(m
);
2167 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2168 JOB_REPLACE_IRREVERSIBLY
);
2172 if (MANAGER_IS_SYSTEM(m
))
2173 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2175 /* This is a nop on non-init */
2179 if (MANAGER_IS_SYSTEM(m
))
2180 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2182 /* This is a nop on non-init */
2188 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2190 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2191 log_info("Trying to reconnect to bus...");
2195 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2196 log_info("Loading D-Bus service...");
2197 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2204 _cleanup_free_
char *dump
= NULL
;
2206 r
= manager_get_dump_string(m
, &dump
);
2208 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2212 log_dump(LOG_INFO
, dump
);
2217 r
= verify_run_space_and_log("Refusing to reload");
2219 m
->exit_code
= MANAGER_RELOAD
;
2224 /* Starting SIGRTMIN+0 */
2225 static const struct {
2228 } target_table
[] = {
2229 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2230 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2231 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2232 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2233 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2234 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2235 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
}
2238 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2239 static const ManagerExitCode code_table
[] = {
2241 [1] = MANAGER_POWEROFF
,
2242 [2] = MANAGER_REBOOT
,
2246 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2247 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2248 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2249 manager_start_target(m
, target_table
[idx
].target
,
2250 target_table
[idx
].mode
);
2254 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2255 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2256 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2260 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2263 manager_set_show_status(m
, SHOW_STATUS_YES
);
2267 manager_set_show_status(m
, SHOW_STATUS_NO
);
2271 log_set_max_level(LOG_DEBUG
);
2272 log_info("Setting log level to debug.");
2276 log_set_max_level(LOG_INFO
);
2277 log_info("Setting log level to info.");
2281 if (MANAGER_IS_USER(m
)) {
2282 m
->exit_code
= MANAGER_EXIT
;
2286 /* This is a nop on init */
2290 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2291 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2292 log_notice("Setting log target to journal-or-kmsg.");
2296 log_set_target(LOG_TARGET_CONSOLE
);
2297 log_notice("Setting log target to console.");
2301 log_set_target(LOG_TARGET_KMSG
);
2302 log_notice("Setting log target to kmsg.");
2306 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2313 manager_dispatch_sigchld(m
);
2318 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2319 Manager
*m
= userdata
;
2324 assert(m
->time_change_fd
== fd
);
2326 log_struct(LOG_DEBUG
,
2327 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2328 LOG_MESSAGE("Time has been changed"),
2331 /* Restart the watch */
2332 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2333 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2335 manager_setup_time_change(m
);
2337 HASHMAP_FOREACH(u
, m
->units
, i
)
2338 if (UNIT_VTABLE(u
)->time_change
)
2339 UNIT_VTABLE(u
)->time_change(u
);
2344 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2345 Manager
*m
= userdata
;
2348 assert(m
->idle_pipe
[2] == fd
);
2350 m
->no_console_output
= m
->n_on_console
> 0;
2352 manager_close_idle_pipe(m
);
2357 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2358 Manager
*m
= userdata
;
2365 manager_print_jobs_in_progress(m
);
2367 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2368 r
= sd_event_source_set_time(source
, next
);
2372 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2375 int manager_loop(Manager
*m
) {
2378 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2381 m
->exit_code
= MANAGER_OK
;
2383 /* Release the path cache */
2384 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2386 manager_check_finished(m
);
2388 /* There might still be some zombies hanging around from
2389 * before we were exec()'ed. Let's reap them. */
2390 r
= manager_dispatch_sigchld(m
);
2394 while (m
->exit_code
== MANAGER_OK
) {
2397 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2400 if (!ratelimit_test(&rl
)) {
2401 /* Yay, something is going seriously wrong, pause a little */
2402 log_warning("Looping too fast. Throttling execution a little.");
2406 if (manager_dispatch_load_queue(m
) > 0)
2409 if (manager_dispatch_gc_job_queue(m
) > 0)
2412 if (manager_dispatch_gc_unit_queue(m
) > 0)
2415 if (manager_dispatch_cleanup_queue(m
) > 0)
2418 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2421 if (manager_dispatch_dbus_queue(m
) > 0)
2424 /* Sleep for half the watchdog time */
2425 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2426 wait_usec
= m
->runtime_watchdog
/ 2;
2430 wait_usec
= USEC_INFINITY
;
2432 r
= sd_event_run(m
->event
, wait_usec
);
2434 return log_error_errno(r
, "Failed to run event loop: %m");
2437 return m
->exit_code
;
2440 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2441 _cleanup_free_
char *n
= NULL
;
2442 sd_id128_t invocation_id
;
2450 r
= unit_name_from_dbus_path(s
, &n
);
2454 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2455 * as invocation ID. */
2456 r
= sd_id128_from_string(n
, &invocation_id
);
2458 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2464 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
));
2467 /* If this didn't work, we check if this is a unit name */
2468 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
))
2469 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is neither a valid invocation ID nor unit name.", n
);
2471 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2479 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2489 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2493 r
= safe_atou(p
, &id
);
2497 j
= manager_get_job(m
, id
);
2506 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2509 _cleanup_free_
char *p
= NULL
;
2513 if (!MANAGER_IS_SYSTEM(m
))
2516 audit_fd
= get_audit_fd();
2520 /* Don't generate audit events if the service was already
2521 * started and we're just deserializing */
2522 if (MANAGER_IS_RELOADING(m
))
2525 if (u
->type
!= UNIT_SERVICE
)
2528 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2530 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2534 msg
= strjoina("unit=", p
);
2535 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2537 /* We aren't allowed to send audit messages?
2538 * Then let's not retry again. */
2541 log_warning_errno(errno
, "Failed to send audit message: %m");
2547 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2548 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2549 _cleanup_free_
char *message
= NULL
;
2550 _cleanup_close_
int fd
= -1;
2553 /* Don't generate plymouth events if the service was already
2554 * started and we're just deserializing */
2555 if (MANAGER_IS_RELOADING(m
))
2558 if (!MANAGER_IS_SYSTEM(m
))
2561 if (detect_container() > 0)
2564 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2567 /* We set SOCK_NONBLOCK here so that we rather drop the
2568 * message then wait for plymouth */
2569 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2571 log_error_errno(errno
, "socket() failed: %m");
2575 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2577 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2578 log_error_errno(errno
, "connect() failed: %m");
2582 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2588 if (write(fd
, message
, n
+ 1) != n
+ 1)
2589 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2590 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2593 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2599 fd
= open_serialization_fd("systemd-state");
2603 f
= fdopen(fd
, "w+");
2613 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2626 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2627 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2628 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2629 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2630 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
2632 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2633 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
2634 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
2637 t
= manager_timestamp_to_string(q
);
2639 char field
[strlen(t
) + strlen("-timestamp") + 1];
2640 strcpy(stpcpy(field
, t
), "-timestamp");
2641 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
2645 if (!switching_root
)
2646 (void) serialize_environment(f
, m
->environment
);
2648 if (m
->notify_fd
>= 0) {
2651 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2655 fprintf(f
, "notify-fd=%i\n", copy
);
2656 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2659 if (m
->cgroups_agent_fd
>= 0) {
2662 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2666 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2669 if (m
->user_lookup_fds
[0] >= 0) {
2672 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2676 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2680 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2683 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2685 r
= dynamic_user_serialize(m
, f
, fds
);
2689 manager_serialize_uid_refs(m
, f
);
2690 manager_serialize_gid_refs(m
, f
);
2692 fputc_unlocked('\n', f
);
2694 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2699 fputs_unlocked(u
->id
, f
);
2700 fputc_unlocked('\n', f
);
2702 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2709 assert(m
->n_reloading
> 0);
2715 r
= bus_fdset_add_all(m
, fds
);
2722 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2728 log_debug("Deserializing state...");
2733 char line
[LINE_MAX
];
2734 const char *val
, *l
;
2736 if (!fgets(line
, sizeof(line
), f
)) {
2751 if ((val
= startswith(l
, "current-job-id="))) {
2754 if (safe_atou32(val
, &id
) < 0)
2755 log_notice("Failed to parse current job id value %s", val
);
2757 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2759 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2762 if (safe_atou32(val
, &n
) < 0)
2763 log_notice("Failed to parse installed jobs counter %s", val
);
2765 m
->n_installed_jobs
+= n
;
2767 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2770 if (safe_atou32(val
, &n
) < 0)
2771 log_notice("Failed to parse failed jobs counter %s", val
);
2773 m
->n_failed_jobs
+= n
;
2775 } else if ((val
= startswith(l
, "taint-usr="))) {
2778 b
= parse_boolean(val
);
2780 log_notice("Failed to parse taint /usr flag %s", val
);
2782 m
->taint_usr
= m
->taint_usr
|| b
;
2784 } else if ((val
= startswith(l
, "ready-sent="))) {
2787 b
= parse_boolean(val
);
2789 log_notice("Failed to parse ready-sent flag %s", val
);
2791 m
->ready_sent
= m
->ready_sent
|| b
;
2793 } else if (startswith(l
, "env=")) {
2794 r
= deserialize_environment(&m
->environment
, l
);
2798 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
2800 } else if ((val
= startswith(l
, "notify-fd="))) {
2803 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2804 log_notice("Failed to parse notify fd: \"%s\"", val
);
2806 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2807 safe_close(m
->notify_fd
);
2808 m
->notify_fd
= fdset_remove(fds
, fd
);
2811 } else if ((val
= startswith(l
, "notify-socket="))) {
2820 free(m
->notify_socket
);
2821 m
->notify_socket
= n
;
2823 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2826 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2827 log_notice("Failed to parse cgroups agent fd: %s", val
);
2829 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2830 safe_close(m
->cgroups_agent_fd
);
2831 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2834 } else if ((val
= startswith(l
, "user-lookup="))) {
2837 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2838 log_notice("Failed to parse user lookup fd: %s", val
);
2840 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2841 safe_close_pair(m
->user_lookup_fds
);
2842 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2843 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2846 } else if ((val
= startswith(l
, "dynamic-user=")))
2847 dynamic_user_deserialize_one(m
, val
, fds
);
2848 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2849 manager_deserialize_uid_refs_one(m
, val
);
2850 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2851 manager_deserialize_gid_refs_one(m
, val
);
2852 else if ((val
= startswith(l
, "subscribed="))) {
2854 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2859 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2860 val
= startswith(l
, manager_timestamp_to_string(q
));
2864 val
= startswith(val
, "-timestamp=");
2869 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
2870 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
2871 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
2872 log_notice("Unknown serialization item '%s'", l
);
2878 char name
[UNIT_NAME_MAX
+2];
2879 const char* unit_name
;
2882 if (!fgets(name
, sizeof(name
), f
)) {
2892 unit_name
= strstrip(name
);
2894 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
2896 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
2899 unit_deserialize_skip(f
);
2903 r
= unit_deserialize(u
, f
, fds
);
2905 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
2915 assert(m
->n_reloading
> 0);
2921 int manager_reload(Manager
*m
) {
2923 _cleanup_fclose_
FILE *f
= NULL
;
2924 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2928 r
= manager_open_serialization(m
, &f
);
2933 bus_manager_send_reloading(m
, true);
2941 r
= manager_serialize(m
, f
, fds
, false);
2947 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2952 /* From here on there is no way back. */
2953 manager_clear_jobs_and_units(m
);
2954 lookup_paths_flush_generator(&m
->lookup_paths
);
2955 lookup_paths_free(&m
->lookup_paths
);
2956 dynamic_user_vacuum(m
, false);
2957 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2958 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2960 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2961 if (q
< 0 && r
>= 0)
2964 q
= manager_run_environment_generators(m
);
2965 if (q
< 0 && r
>= 0)
2968 /* Find new unit paths */
2969 q
= manager_run_generators(m
);
2970 if (q
< 0 && r
>= 0)
2973 lookup_paths_reduce(&m
->lookup_paths
);
2974 manager_build_unit_path_cache(m
);
2976 /* First, enumerate what we can from all config files */
2977 manager_enumerate(m
);
2979 /* Second, deserialize our stored data */
2980 q
= manager_deserialize(m
, f
, fds
);
2982 log_error_errno(q
, "Deserialization failed: %m");
2991 /* Re-register notify_fd as event source */
2992 q
= manager_setup_notify(m
);
2993 if (q
< 0 && r
>= 0)
2996 q
= manager_setup_cgroups_agent(m
);
2997 if (q
< 0 && r
>= 0)
3000 q
= manager_setup_user_lookup_fd(m
);
3001 if (q
< 0 && r
>= 0)
3004 /* Third, fire things up! */
3005 manager_coldplug(m
);
3007 /* Release any dynamic users no longer referenced */
3008 dynamic_user_vacuum(m
, true);
3010 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3011 manager_vacuum_uid_refs(m
);
3012 manager_vacuum_gid_refs(m
);
3014 /* Sync current state of bus names with our set of listening units */
3016 manager_sync_bus_names(m
, m
->api_bus
);
3018 assert(m
->n_reloading
> 0);
3021 m
->send_reloading_done
= true;
3026 void manager_reset_failed(Manager
*m
) {
3032 HASHMAP_FOREACH(u
, m
->units
, i
)
3033 unit_reset_failed(u
);
3036 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3042 /* Returns true if the unit is inactive or going down */
3043 u
= manager_get_unit(m
, name
);
3047 return unit_inactive_or_pending(u
);
3050 static void manager_notify_finished(Manager
*m
) {
3051 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3052 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3054 if (m
->test_run_flags
)
3057 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3059 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3060 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3061 * negative values. */
3063 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3064 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3065 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3066 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3068 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3070 /* The initrd case on bare-metal*/
3071 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3072 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3074 log_struct(LOG_INFO
,
3075 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3076 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3077 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3078 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3079 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
3080 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3081 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3082 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3083 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3086 /* The initrd-less case on bare-metal*/
3088 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3091 log_struct(LOG_INFO
,
3092 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3093 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3094 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3095 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
3096 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3097 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3098 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3102 /* The container case */
3103 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3104 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3106 log_struct(LOG_INFO
,
3107 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3108 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3109 LOG_MESSAGE("Startup finished in %s.",
3110 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
3114 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3117 m
->ready_sent
? "STATUS=Startup finished in %s."
3119 "STATUS=Startup finished in %s.",
3120 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3121 m
->ready_sent
= true;
3124 void manager_check_finished(Manager
*m
) {
3127 if (MANAGER_IS_RELOADING(m
))
3130 /* Verify that we are actually running currently. Initially
3131 * the exit code is set to invalid, and during operation it is
3132 * then set to MANAGER_OK */
3133 if (m
->exit_code
!= MANAGER_OK
)
3136 /* For user managers, send out READY=1 as soon as we reach basic.target */
3137 if (MANAGER_IS_USER(m
) && !m
->ready_sent
) {
3140 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3144 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3145 m
->ready_sent
= true;
3149 if (hashmap_size(m
->jobs
) > 0) {
3150 if (m
->jobs_in_progress_event_source
)
3151 /* Ignore any failure, this is only for feedback */
3152 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3157 manager_flip_auto_status(m
, false);
3159 /* Notify Type=idle units that we are done now */
3160 manager_close_idle_pipe(m
);
3162 /* Turn off confirm spawn now */
3163 m
->confirm_spawn
= NULL
;
3165 /* No need to update ask password status when we're going non-interactive */
3166 manager_close_ask_password(m
);
3168 /* This is no longer the first boot */
3169 manager_set_first_boot(m
, false);
3171 if (MANAGER_IS_FINISHED(m
))
3174 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3176 manager_notify_finished(m
);
3178 manager_invalidate_startup_units(m
);
3181 static bool generator_path_any(const char* const* paths
) {
3185 /* Optimize by skipping the whole process by not creating output directories
3186 * if no generators are found. */
3187 STRV_FOREACH(path
, (char**) paths
)
3188 if (access(*path
, F_OK
) == 0)
3190 else if (errno
!= ENOENT
)
3191 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3196 static const char* system_env_generator_binary_paths
[] = {
3197 "/run/systemd/system-environment-generators",
3198 "/etc/systemd/system-environment-generators",
3199 "/usr/local/lib/systemd/system-environment-generators",
3200 SYSTEM_ENV_GENERATOR_PATH
,
3204 static const char* user_env_generator_binary_paths
[] = {
3205 "/run/systemd/user-environment-generators",
3206 "/etc/systemd/user-environment-generators",
3207 "/usr/local/lib/systemd/user-environment-generators",
3208 USER_ENV_GENERATOR_PATH
,
3212 static int manager_run_environment_generators(Manager
*m
) {
3213 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3215 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3217 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3220 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3222 if (!generator_path_any(paths
))
3225 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3228 static int manager_run_generators(Manager
*m
) {
3229 _cleanup_strv_free_
char **paths
= NULL
;
3230 const char *argv
[5];
3235 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3238 paths
= generator_binary_paths(m
->unit_file_scope
);
3242 if (!generator_path_any((const char* const*) paths
))
3245 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3249 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3250 argv
[1] = m
->lookup_paths
.generator
;
3251 argv
[2] = m
->lookup_paths
.generator_early
;
3252 argv
[3] = m
->lookup_paths
.generator_late
;
3255 RUN_WITH_UMASK(0022)
3256 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3257 NULL
, NULL
, (char**) argv
);
3260 lookup_paths_trim_generator(&m
->lookup_paths
);
3264 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3265 char **a
= NULL
, **b
= NULL
, **l
;
3270 if (!strv_isempty(minus
)) {
3271 a
= strv_env_delete(l
, 1, minus
);
3278 if (!strv_isempty(plus
)) {
3279 b
= strv_env_merge(2, l
, plus
);
3288 if (m
->environment
!= l
)
3289 strv_free(m
->environment
);
3296 manager_clean_environment(m
);
3297 strv_sort(m
->environment
);
3302 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3307 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3308 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3310 if (!default_rlimit
[i
])
3313 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3321 void manager_recheck_journal(Manager
*m
) {
3326 if (!MANAGER_IS_SYSTEM(m
))
3329 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3330 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3331 log_close_journal();
3335 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3336 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3337 log_close_journal();
3341 /* Hmm, OK, so the socket is fully up and the service is up
3342 * too, then let's make use of the thing. */
3346 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3348 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3350 if (!MANAGER_IS_SYSTEM(m
))
3353 if (m
->show_status
!= mode
)
3354 log_debug("%s showing of status.",
3355 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3356 m
->show_status
= mode
;
3359 (void) touch("/run/systemd/show-status");
3361 (void) unlink("/run/systemd/show-status");
3364 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3367 if (!MANAGER_IS_SYSTEM(m
))
3370 if (m
->no_console_output
)
3373 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3376 /* If we cannot find out the status properly, just proceed. */
3377 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3380 if (m
->show_status
> 0)
3386 const char *manager_get_confirm_spawn(Manager
*m
) {
3387 static int last_errno
= 0;
3388 const char *vc
= m
->confirm_spawn
;
3392 /* Here's the deal: we want to test the validity of the console but don't want
3393 * PID1 to go through the whole console process which might block. But we also
3394 * want to warn the user only once if something is wrong with the console so we
3395 * cannot do the sanity checks after spawning our children. So here we simply do
3396 * really basic tests to hopefully trap common errors.
3398 * If the console suddenly disappear at the time our children will really it
3399 * then they will simply fail to acquire it and a positive answer will be
3400 * assumed. New children will fallback to /dev/console though.
3402 * Note: TTYs are devices that can come and go any time, and frequently aren't
3403 * available yet during early boot (consider a USB rs232 dongle...). If for any
3404 * reason the configured console is not ready, we fallback to the default
3407 if (!vc
|| path_equal(vc
, "/dev/console"))
3414 if (!S_ISCHR(st
.st_mode
)) {
3422 if (last_errno
!= errno
) {
3424 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3426 return "/dev/console";
3429 void manager_set_first_boot(Manager
*m
, bool b
) {
3432 if (!MANAGER_IS_SYSTEM(m
))
3435 if (m
->first_boot
!= (int) b
) {
3437 (void) touch("/run/systemd/first-boot");
3439 (void) unlink("/run/systemd/first-boot");
3445 void manager_disable_confirm_spawn(void) {
3446 (void) touch("/run/systemd/confirm_spawn_disabled");
3449 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3450 if (!m
->confirm_spawn
)
3453 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3456 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3459 /* If m is NULL, assume we're after shutdown and let the messages through. */
3461 if (m
&& !manager_get_show_status(m
, type
))
3464 /* XXX We should totally drop the check for ephemeral here
3465 * and thus effectively make 'Type=idle' pointless. */
3466 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3469 va_start(ap
, format
);
3470 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3474 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3475 char p
[strlen(path
)+1];
3481 path_kill_slashes(p
);
3483 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3486 void manager_set_exec_params(Manager
*m
, ExecParameters
*p
) {
3490 p
->environment
= m
->environment
;
3491 p
->confirm_spawn
= manager_get_confirm_spawn(m
);
3492 p
->cgroup_supported
= m
->cgroup_supported
;
3493 p
->prefix
= m
->prefix
;
3495 SET_FLAG(p
->flags
, EXEC_PASS_LOG_UNIT
|EXEC_CHOWN_DIRECTORIES
, MANAGER_IS_SYSTEM(m
));
3498 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3503 assert(u
->manager
== m
);
3505 size
= set_size(m
->failed_units
);
3508 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3512 if (set_put(m
->failed_units
, u
) < 0)
3515 (void) set_remove(m
->failed_units
, u
);
3517 if (set_size(m
->failed_units
) != size
)
3518 bus_manager_send_change_signal(m
);
3523 ManagerState
manager_state(Manager
*m
) {
3528 /* Did we ever finish booting? If not then we are still starting up */
3529 if (!MANAGER_IS_FINISHED(m
)) {
3531 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3532 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3533 return MANAGER_INITIALIZING
;
3535 return MANAGER_STARTING
;
3538 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
3539 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3540 if (u
&& unit_active_or_pending(u
))
3541 return MANAGER_STOPPING
;
3543 if (MANAGER_IS_SYSTEM(m
)) {
3544 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3545 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3546 if (u
&& unit_active_or_pending(u
))
3547 return MANAGER_MAINTENANCE
;
3549 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3550 if (u
&& unit_active_or_pending(u
))
3551 return MANAGER_MAINTENANCE
;
3554 /* Are there any failed units? If so, we are in degraded mode */
3555 if (set_size(m
->failed_units
) > 0)
3556 return MANAGER_DEGRADED
;
3558 return MANAGER_RUNNING
;
3561 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3563 static void manager_unref_uid_internal(
3568 int (*_clean_ipc
)(uid_t uid
)) {
3574 assert(uid_is_valid(uid
));
3577 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3578 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3580 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3581 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3582 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3583 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3585 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3586 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3588 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3591 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3593 n
= c
& ~DESTROY_IPC_FLAG
;
3597 if (destroy_now
&& n
== 0) {
3598 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3600 if (c
& DESTROY_IPC_FLAG
) {
3601 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3602 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3604 (void) _clean_ipc(uid
);
3607 c
= n
| (c
& DESTROY_IPC_FLAG
);
3608 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3612 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3613 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3616 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3617 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3620 static int manager_ref_uid_internal(
3631 assert(uid_is_valid(uid
));
3633 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3634 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3636 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3637 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3639 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3642 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3646 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3648 n
= c
& ~DESTROY_IPC_FLAG
;
3651 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3654 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3656 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3659 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3660 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3663 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3664 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3667 static void manager_vacuum_uid_refs_internal(
3670 int (*_clean_ipc
)(uid_t uid
)) {
3679 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3683 uid
= PTR_TO_UID(k
);
3684 c
= PTR_TO_UINT32(p
);
3686 n
= c
& ~DESTROY_IPC_FLAG
;
3690 if (c
& DESTROY_IPC_FLAG
) {
3691 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3692 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3694 (void) _clean_ipc(uid
);
3697 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3701 void manager_vacuum_uid_refs(Manager
*m
) {
3702 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3705 void manager_vacuum_gid_refs(Manager
*m
) {
3706 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3709 static void manager_serialize_uid_refs_internal(
3713 const char *field_name
) {
3723 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3724 * of it is better rebuild after a reload/reexec. */
3726 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3730 uid
= PTR_TO_UID(k
);
3731 c
= PTR_TO_UINT32(p
);
3733 if (!(c
& DESTROY_IPC_FLAG
))
3736 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3740 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3741 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3744 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3745 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3748 static void manager_deserialize_uid_refs_one_internal(
3751 const char *value
) {
3761 r
= parse_uid(value
, &uid
);
3762 if (r
< 0 || uid
== 0) {
3763 log_debug("Unable to parse UID reference serialization");
3767 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3773 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3774 if (c
& DESTROY_IPC_FLAG
)
3777 c
|= DESTROY_IPC_FLAG
;
3779 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3781 log_debug("Failed to add UID reference entry");
3786 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3787 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3790 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3791 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3794 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3798 char unit_name
[UNIT_NAME_MAX
+1];
3801 Manager
*m
= userdata
;
3809 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3810 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3811 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3813 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3815 if (IN_SET(errno
, EINTR
, EAGAIN
))
3818 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3821 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3822 log_warning("Received too short user lookup message, ignoring.");
3826 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3827 log_warning("Received too long user lookup message, ignoring.");
3831 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3832 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3836 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3837 if (memchr(buffer
.unit_name
, 0, n
)) {
3838 log_warning("Received lookup message with embedded NUL character, ignoring.");
3842 buffer
.unit_name
[n
] = 0;
3843 u
= manager_get_unit(m
, buffer
.unit_name
);
3845 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3849 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3851 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3855 char *manager_taint_string(Manager
*m
) {
3856 _cleanup_free_
char *destination
= NULL
;
3862 buf
= new(char, sizeof("split-usr:"
3866 "weird-nobody-user:"
3867 "weird-nobody-group:"));
3874 e
= stpcpy(e
, "split-usr:");
3876 if (access("/proc/cgroups", F_OK
) < 0)
3877 e
= stpcpy(e
, "cgroups-missing:");
3879 if (clock_is_localtime(NULL
) > 0)
3880 e
= stpcpy(e
, "local-hwclock:");
3882 r
= readlink_malloc("/var/run", &destination
);
3883 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
3884 e
= stpcpy(e
, "var-run-bad:");
3886 if (!streq(NOBODY_USER_NAME
, "nobody"))
3887 e
= stpcpy(e
, "weird-nobody-user:");
3889 if (!streq(NOBODY_GROUP_NAME
, "nobody"))
3890 e
= stpcpy(e
, "weird-nobody-group:");
3892 /* remove the last ':' */
3899 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3900 [MANAGER_INITIALIZING
] = "initializing",
3901 [MANAGER_STARTING
] = "starting",
3902 [MANAGER_RUNNING
] = "running",
3903 [MANAGER_DEGRADED
] = "degraded",
3904 [MANAGER_MAINTENANCE
] = "maintenance",
3905 [MANAGER_STOPPING
] = "stopping",
3908 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
3910 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
3911 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
3912 [MANAGER_TIMESTAMP_LOADER
] = "loader",
3913 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
3914 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
3915 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
3916 [MANAGER_TIMESTAMP_FINISH
] = "finish",
3917 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
3918 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
3919 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
3920 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
3921 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
3922 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
3925 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);