1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2010 Lennart Poettering
10 #include <stdio_ext.h>
12 #include <sys/epoll.h>
13 #include <sys/inotify.h>
14 #include <sys/ioctl.h>
15 #include <sys/reboot.h>
16 #include <sys/timerfd.h>
24 #include "sd-daemon.h"
25 #include "sd-messages.h"
28 #include "alloc-util.h"
29 #include "all-units.h"
31 #include "boot-timestamps.h"
32 #include "bus-common-errors.h"
33 #include "bus-error.h"
34 #include "bus-kernel.h"
36 #include "clean-ipc.h"
37 #include "clock-util.h"
39 #include "dbus-manager.h"
40 #include "dbus-unit.h"
42 #include "dirent-util.h"
45 #include "exec-util.h"
47 #include "exit-status.h"
54 #include "locale-setup.h"
60 #include "parse-util.h"
61 #include "path-lookup.h"
62 #include "path-util.h"
63 #include "process-util.h"
64 #include "ratelimit.h"
65 #include "rlimit-util.h"
67 #include "signal-util.h"
68 #include "socket-util.h"
70 #include "stat-util.h"
71 #include "string-table.h"
72 #include "string-util.h"
75 #include "syslog-util.h"
76 #include "terminal-util.h"
77 #include "time-util.h"
78 #include "transaction.h"
79 #include "umask-util.h"
80 #include "unit-name.h"
81 #include "user-util.h"
86 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
87 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
89 /* Initial delay and the interval for printing status messages about running jobs */
90 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
91 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
92 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
94 /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until
95 * the queue gets more empty. */
96 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
98 /* How many units and jobs to process of the bus queue before returning to the event loop. */
99 #define MANAGER_BUS_MESSAGE_BUDGET 100U
101 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
105 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
108 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
109 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
110 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
111 static int manager_run_environment_generators(Manager
*m
);
112 static int manager_run_generators(Manager
*m
);
114 static void manager_watch_jobs_in_progress(Manager
*m
) {
120 /* We do not want to show the cylon animation if the user
121 * needs to confirm service executions otherwise confirmation
122 * messages will be screwed by the cylon animation. */
123 if (!manager_is_confirm_spawn_disabled(m
))
126 if (m
->jobs_in_progress_event_source
)
129 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
130 r
= sd_event_add_time(
132 &m
->jobs_in_progress_event_source
,
135 manager_dispatch_jobs_in_progress
, m
);
139 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
142 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
144 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
147 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
148 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
152 p
= mempset(p
, ' ', pos
-2);
153 if (log_get_show_color())
154 p
= stpcpy(p
, ANSI_RED
);
158 if (pos
> 0 && pos
<= width
) {
159 if (log_get_show_color())
160 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
164 if (log_get_show_color())
165 p
= stpcpy(p
, ANSI_NORMAL
);
168 if (log_get_show_color())
169 p
= stpcpy(p
, ANSI_RED
);
172 p
= mempset(p
, ' ', width
-1-pos
);
173 if (log_get_show_color())
174 strcpy(p
, ANSI_NORMAL
);
178 void manager_flip_auto_status(Manager
*m
, bool enable
) {
182 if (m
->show_status
== SHOW_STATUS_AUTO
)
183 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
185 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
186 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
190 static void manager_print_jobs_in_progress(Manager
*m
) {
191 _cleanup_free_
char *job_of_n
= NULL
;
194 unsigned counter
= 0, print_nr
;
195 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
197 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
201 assert(m
->n_running_jobs
> 0);
203 manager_flip_auto_status(m
, true);
205 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
207 HASHMAP_FOREACH(j
, m
->jobs
, i
)
208 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
211 /* m->n_running_jobs must be consistent with the contents of m->jobs,
212 * so the above loop must have succeeded in finding j. */
213 assert(counter
== print_nr
+ 1);
216 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
218 cylon_pos
= 14 - cylon_pos
;
219 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
221 m
->jobs_in_progress_iteration
++;
223 if (m
->n_running_jobs
> 1) {
224 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
228 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
229 if (job_get_timeout(j
, &x
) > 0)
230 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
232 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
233 "%sA %s job is running for %s (%s / %s)",
235 job_type_to_string(j
->type
),
236 unit_description(j
->unit
),
240 static int have_ask_password(void) {
241 _cleanup_closedir_
DIR *dir
;
244 dir
= opendir("/run/systemd/ask-password");
252 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
253 if (startswith(de
->d_name
, "ask."))
259 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
260 int fd
, uint32_t revents
, void *userdata
) {
261 Manager
*m
= userdata
;
267 m
->have_ask_password
= have_ask_password();
268 if (m
->have_ask_password
< 0)
269 /* Log error but continue. Negative have_ask_password
270 * is treated as unknown status. */
271 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
276 static void manager_close_ask_password(Manager
*m
) {
279 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
280 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
281 m
->have_ask_password
= -EINVAL
;
284 static int manager_check_ask_password(Manager
*m
) {
289 if (!m
->ask_password_event_source
) {
290 assert(m
->ask_password_inotify_fd
< 0);
292 mkdir_p_label("/run/systemd/ask-password", 0755);
294 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
295 if (m
->ask_password_inotify_fd
< 0)
296 return log_error_errno(errno
, "inotify_init1() failed: %m");
298 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
299 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
300 manager_close_ask_password(m
);
304 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
305 m
->ask_password_inotify_fd
, EPOLLIN
,
306 manager_dispatch_ask_password_fd
, m
);
308 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
309 manager_close_ask_password(m
);
313 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
315 /* Queries might have been added meanwhile... */
316 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
317 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
320 return m
->have_ask_password
;
323 static int manager_watch_idle_pipe(Manager
*m
) {
328 if (m
->idle_pipe_event_source
)
331 if (m
->idle_pipe
[2] < 0)
334 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
336 return log_error_errno(r
, "Failed to watch idle pipe: %m");
338 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
343 static void manager_close_idle_pipe(Manager
*m
) {
346 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
348 safe_close_pair(m
->idle_pipe
);
349 safe_close_pair(m
->idle_pipe
+ 2);
352 static int manager_setup_time_change(Manager
*m
) {
357 if (m
->test_run_flags
)
360 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
361 m
->time_change_fd
= safe_close(m
->time_change_fd
);
363 m
->time_change_fd
= time_change_fd();
364 if (m
->time_change_fd
< 0)
365 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
367 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
369 return log_error_errno(r
, "Failed to create time change event source: %m");
371 /* Schedule this slightly earlier than the .timer event sources */
372 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
374 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
376 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
378 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
383 static int manager_read_timezone_stat(Manager
*m
) {
389 /* Read the current stat() data of /etc/localtime so that we detect changes */
390 if (lstat("/etc/localtime", &st
) < 0) {
391 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
392 changed
= m
->etc_localtime_accessible
;
393 m
->etc_localtime_accessible
= false;
397 k
= timespec_load(&st
.st_mtim
);
398 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
400 m
->etc_localtime_mtime
= k
;
401 m
->etc_localtime_accessible
= true;
407 static int manager_setup_timezone_change(Manager
*m
) {
408 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
413 if (m
->test_run_flags
!= 0)
416 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
417 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
418 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
419 * went to zero and all fds to it are closed.
421 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
424 * Note that we create the new event source first here, before releasing the old one. This should optimize
425 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
427 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
428 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
429 if (r
== -ENOENT
) /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created
430 * either by O_CREATE or by rename() */
431 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
432 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
434 return log_error_errno(r
, "Failed to create timezone change event source: %m");
436 /* Schedule this slightly earlier than the .timer event sources */
437 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
439 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
441 sd_event_source_unref(m
->timezone_change_event_source
);
442 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
447 static int enable_special_signals(Manager
*m
) {
448 _cleanup_close_
int fd
= -1;
452 if (m
->test_run_flags
)
455 /* Enable that we get SIGINT on control-alt-del. In containers
456 * this will fail with EPERM (older) or EINVAL (newer), so
458 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
459 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
461 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
463 /* Support systems without virtual console */
465 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
467 /* Enable that we get SIGWINCH on kbrequest */
468 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
469 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
475 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
477 static int manager_setup_signals(Manager
*m
) {
478 struct sigaction sa
= {
479 .sa_handler
= SIG_DFL
,
480 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
487 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
489 /* We make liberal use of realtime signals here. On
490 * Linux/glibc we have 30 of them (with the exception of Linux
491 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
494 assert_se(sigemptyset(&mask
) == 0);
495 sigset_add_many(&mask
,
496 SIGCHLD
, /* Child died */
497 SIGTERM
, /* Reexecute daemon */
498 SIGHUP
, /* Reload configuration */
499 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
500 SIGUSR2
, /* systemd: dump status */
501 SIGINT
, /* Kernel sends us this on control-alt-del */
502 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
503 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
505 SIGRTMIN
+0, /* systemd: start default.target */
506 SIGRTMIN
+1, /* systemd: isolate rescue.target */
507 SIGRTMIN
+2, /* systemd: isolate emergency.target */
508 SIGRTMIN
+3, /* systemd: start halt.target */
509 SIGRTMIN
+4, /* systemd: start poweroff.target */
510 SIGRTMIN
+5, /* systemd: start reboot.target */
511 SIGRTMIN
+6, /* systemd: start kexec.target */
513 /* ... space for more special targets ... */
515 SIGRTMIN
+13, /* systemd: Immediate halt */
516 SIGRTMIN
+14, /* systemd: Immediate poweroff */
517 SIGRTMIN
+15, /* systemd: Immediate reboot */
518 SIGRTMIN
+16, /* systemd: Immediate kexec */
520 /* ... space for more immediate system state changes ... */
522 SIGRTMIN
+20, /* systemd: enable status messages */
523 SIGRTMIN
+21, /* systemd: disable status messages */
524 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
525 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
526 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
528 /* .. one free signal here ... */
530 /* Apparently Linux on hppa had fewer RT signals until v3.18,
531 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
532 * see commit v3.17-7614-g1f25df2eff.
534 * We cannot unconditionally make use of those signals here,
535 * so let's use a runtime check. Since these commands are
536 * accessible by different means and only really a safety
537 * net, the missing functionality on hppa shouldn't matter.
540 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
541 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
542 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
543 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
545 /* ... one free signal here SIGRTMIN+30 ... */
547 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
549 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
550 if (m
->signal_fd
< 0)
553 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
557 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
559 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
560 * notify processing can still figure out to which process/service a message belongs, before we reap the
561 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
562 * status information before detecting that there's no process in a cgroup. */
563 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
567 if (MANAGER_IS_SYSTEM(m
))
568 return enable_special_signals(m
);
573 static void manager_sanitize_environment(Manager
*m
) {
576 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
596 /* Let's order the environment alphabetically, just to make it pretty */
597 strv_sort(m
->environment
);
600 static int manager_default_environment(Manager
*m
) {
603 if (MANAGER_IS_SYSTEM(m
)) {
604 /* The system manager always starts with a clean
605 * environment for its children. It does not import
606 * the kernel's or the parents' exported variables.
608 * The initial passed environment is untouched to keep
609 * /proc/self/environ valid; it is used for tagging
610 * the init process inside containers. */
611 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
614 /* Import locale variables LC_*= from configuration */
615 locale_setup(&m
->environment
);
617 /* The user manager passes its own environment
618 * along to its children. */
619 m
->environment
= strv_copy(environ
);
624 manager_sanitize_environment(m
);
629 static int manager_setup_prefix(Manager
*m
) {
635 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
636 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
637 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
638 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
639 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
640 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
643 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
644 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
645 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
646 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
647 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
648 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
651 const struct table_entry
*p
;
657 if (MANAGER_IS_SYSTEM(m
))
662 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
663 r
= sd_path_home(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
671 static int manager_setup_run_queue(Manager
*m
) {
675 assert(!m
->run_queue_event_source
);
677 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
681 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
685 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
689 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
694 static int manager_setup_sigchld_event_source(Manager
*m
) {
698 assert(!m
->sigchld_event_source
);
700 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
704 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
708 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
712 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
717 int manager_new(UnitFileScope scope
, unsigned test_run_flags
, Manager
**_m
) {
718 _cleanup_(manager_freep
) Manager
*m
= NULL
;
722 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
724 m
= new0(Manager
, 1);
728 m
->unit_file_scope
= scope
;
729 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
730 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
731 m
->default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
;
732 m
->default_tasks_accounting
= true;
733 m
->default_tasks_max
= UINT64_MAX
;
734 m
->default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
;
735 m
->default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
;
736 m
->default_restart_usec
= DEFAULT_RESTART_USEC
;
737 m
->original_log_level
= -1;
738 m
->original_log_target
= _LOG_TARGET_INVALID
;
741 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
742 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
743 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
744 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
747 /* Prepare log fields we can use for structured logging */
748 if (MANAGER_IS_SYSTEM(m
)) {
749 m
->unit_log_field
= "UNIT=";
750 m
->unit_log_format_string
= "UNIT=%s";
752 m
->invocation_log_field
= "INVOCATION_ID=";
753 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
755 m
->unit_log_field
= "USER_UNIT=";
756 m
->unit_log_format_string
= "USER_UNIT=%s";
758 m
->invocation_log_field
= "USER_INVOCATION_ID=";
759 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
762 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
764 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
765 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
766 m
->ask_password_inotify_fd
= -1;
768 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
770 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
772 m
->have_ask_password
= -EINVAL
; /* we don't know */
775 m
->test_run_flags
= test_run_flags
;
777 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
778 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
780 r
= manager_default_environment(m
);
784 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
788 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
792 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
796 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
800 r
= manager_setup_prefix(m
);
804 m
->udev
= udev_new();
808 r
= sd_event_default(&m
->event
);
812 r
= manager_setup_run_queue(m
);
816 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
817 m
->cgroup_root
= strdup("");
821 r
= manager_setup_signals(m
);
825 r
= manager_setup_cgroup(m
);
829 r
= manager_setup_time_change(m
);
833 r
= manager_read_timezone_stat(m
);
837 r
= manager_setup_timezone_change(m
);
841 r
= manager_setup_sigchld_event_source(m
);
846 if (MANAGER_IS_SYSTEM(m
) && test_run_flags
== 0) {
847 r
= mkdir_label("/run/systemd/units", 0755);
848 if (r
< 0 && r
!= -EEXIST
)
854 dir_is_empty("/usr") > 0;
856 /* Note that we do not set up the notify fd here. We do that after deserialization,
857 * since they might have gotten serialized across the reexec. */
864 static int manager_setup_notify(Manager
*m
) {
867 if (m
->test_run_flags
)
870 if (m
->notify_fd
< 0) {
871 _cleanup_close_
int fd
= -1;
872 union sockaddr_union sa
= {
873 .sa
.sa_family
= AF_UNIX
,
875 static const int one
= 1;
877 /* First free all secondary fields */
878 m
->notify_socket
= mfree(m
->notify_socket
);
879 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
881 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
883 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
885 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
887 m
->notify_socket
= strappend(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "/systemd/notify");
888 if (!m
->notify_socket
)
891 (void) mkdir_parents_label(m
->notify_socket
, 0755);
892 (void) unlink(m
->notify_socket
);
894 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
895 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
897 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
899 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
901 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
903 m
->notify_fd
= TAKE_FD(fd
);
905 log_debug("Using notification socket %s", m
->notify_socket
);
908 if (!m
->notify_event_source
) {
909 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
911 return log_error_errno(r
, "Failed to allocate notify event source: %m");
913 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
914 * service an exit message belongs. */
915 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
917 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
919 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
925 static int manager_setup_cgroups_agent(Manager
*m
) {
927 static const union sockaddr_union sa
= {
928 .un
.sun_family
= AF_UNIX
,
929 .un
.sun_path
= "/run/systemd/cgroups-agent",
933 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
934 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
935 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
936 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
937 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
938 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
939 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
940 * we thus won't lose messages.
942 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
943 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
944 * bus for these messages. */
946 if (m
->test_run_flags
)
949 if (!MANAGER_IS_SYSTEM(m
))
952 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
954 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
955 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
958 if (m
->cgroups_agent_fd
< 0) {
959 _cleanup_close_
int fd
= -1;
961 /* First free all secondary fields */
962 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
964 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
966 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
968 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
970 (void) unlink(sa
.un
.sun_path
);
972 /* Only allow root to connect to this socket */
974 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
976 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
978 m
->cgroups_agent_fd
= fd
;
982 if (!m
->cgroups_agent_event_source
) {
983 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
985 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
987 /* Process cgroups notifications early, but after having processed service notification messages or
988 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
989 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
990 * cgroup inotify for the unified cgroup stuff. */
991 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-4);
993 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
995 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1001 static int manager_setup_user_lookup_fd(Manager
*m
) {
1006 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1007 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1008 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1009 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1010 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1011 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1012 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1013 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1016 * You might wonder why we need a communication channel for this that is independent of the usual notification
1017 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1018 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1019 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1021 * Note that this function is called under two circumstances: when we first initialize (in which case we
1022 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1023 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1025 if (m
->user_lookup_fds
[0] < 0) {
1027 /* Free all secondary fields */
1028 safe_close_pair(m
->user_lookup_fds
);
1029 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1031 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1032 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1034 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1037 if (!m
->user_lookup_event_source
) {
1038 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1040 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1042 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1044 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1046 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
1048 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1054 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1060 while ((u
= m
->cleanup_queue
)) {
1061 assert(u
->in_cleanup_queue
);
1071 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1072 GC_OFFSET_UNSURE
, /* No clue */
1073 GC_OFFSET_GOOD
, /* We still need this unit */
1074 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1078 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1083 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1085 /* Recursively mark referenced units as GOOD as well */
1086 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCES
], i
)
1087 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1088 unit_gc_mark_good(other
, gc_marker
);
1091 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1099 if (IN_SET(u
->gc_marker
- gc_marker
,
1100 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1103 if (u
->in_cleanup_queue
)
1106 if (!unit_may_gc(u
))
1109 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1113 HASHMAP_FOREACH_KEY(v
, other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
1114 unit_gc_sweep(other
, gc_marker
);
1116 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1119 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1123 if (u
->refs_by_target
) {
1126 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1127 unit_gc_sweep(ref
->source
, gc_marker
);
1129 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1132 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1140 /* We were unable to find anything out about this entry, so
1141 * let's investigate it later */
1142 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1143 unit_add_to_gc_queue(u
);
1147 /* We definitely know that this one is not useful anymore, so
1148 * let's mark it for deletion */
1149 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1150 unit_add_to_cleanup_queue(u
);
1154 unit_gc_mark_good(u
, gc_marker
);
1157 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1158 unsigned n
= 0, gc_marker
;
1163 /* log_debug("Running GC..."); */
1165 m
->gc_marker
+= _GC_OFFSET_MAX
;
1166 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1169 gc_marker
= m
->gc_marker
;
1171 while ((u
= m
->gc_unit_queue
)) {
1172 assert(u
->in_gc_queue
);
1174 unit_gc_sweep(u
, gc_marker
);
1176 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1177 u
->in_gc_queue
= false;
1181 if (IN_SET(u
->gc_marker
- gc_marker
,
1182 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1184 log_unit_debug(u
, "Collecting.");
1185 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1186 unit_add_to_cleanup_queue(u
);
1193 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1199 while ((j
= m
->gc_job_queue
)) {
1200 assert(j
->in_gc_queue
);
1202 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1203 j
->in_gc_queue
= false;
1210 log_unit_debug(j
->unit
, "Collecting job.");
1211 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1217 static void manager_clear_jobs_and_units(Manager
*m
) {
1222 while ((u
= hashmap_first(m
->units
)))
1225 manager_dispatch_cleanup_queue(m
);
1227 assert(!m
->load_queue
);
1228 assert(!m
->run_queue
);
1229 assert(!m
->dbus_unit_queue
);
1230 assert(!m
->dbus_job_queue
);
1231 assert(!m
->cleanup_queue
);
1232 assert(!m
->gc_unit_queue
);
1233 assert(!m
->gc_job_queue
);
1235 assert(hashmap_isempty(m
->jobs
));
1236 assert(hashmap_isempty(m
->units
));
1238 m
->n_on_console
= 0;
1239 m
->n_running_jobs
= 0;
1242 Manager
* manager_free(Manager
*m
) {
1244 ExecDirectoryType dt
;
1249 manager_clear_jobs_and_units(m
);
1251 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1252 if (unit_vtable
[c
]->shutdown
)
1253 unit_vtable
[c
]->shutdown(m
);
1255 /* If we reexecute ourselves, we keep the root cgroup around */
1256 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1258 lookup_paths_flush_generator(&m
->lookup_paths
);
1262 exec_runtime_vacuum(m
);
1263 hashmap_free(m
->exec_runtime_by_id
);
1265 dynamic_user_vacuum(m
, false);
1266 hashmap_free(m
->dynamic_users
);
1268 hashmap_free(m
->units
);
1269 hashmap_free(m
->units_by_invocation_id
);
1270 hashmap_free(m
->jobs
);
1271 hashmap_free(m
->watch_pids
);
1272 hashmap_free(m
->watch_bus
);
1274 set_free(m
->startup_units
);
1275 set_free(m
->failed_units
);
1277 sd_event_source_unref(m
->signal_event_source
);
1278 sd_event_source_unref(m
->sigchld_event_source
);
1279 sd_event_source_unref(m
->notify_event_source
);
1280 sd_event_source_unref(m
->cgroups_agent_event_source
);
1281 sd_event_source_unref(m
->time_change_event_source
);
1282 sd_event_source_unref(m
->timezone_change_event_source
);
1283 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1284 sd_event_source_unref(m
->run_queue_event_source
);
1285 sd_event_source_unref(m
->user_lookup_event_source
);
1286 sd_event_source_unref(m
->sync_bus_names_event_source
);
1288 safe_close(m
->signal_fd
);
1289 safe_close(m
->notify_fd
);
1290 safe_close(m
->cgroups_agent_fd
);
1291 safe_close(m
->time_change_fd
);
1292 safe_close_pair(m
->user_lookup_fds
);
1294 manager_close_ask_password(m
);
1296 manager_close_idle_pipe(m
);
1298 udev_unref(m
->udev
);
1299 sd_event_unref(m
->event
);
1301 free(m
->notify_socket
);
1303 lookup_paths_free(&m
->lookup_paths
);
1304 strv_free(m
->environment
);
1306 hashmap_free(m
->cgroup_unit
);
1307 set_free_free(m
->unit_path_cache
);
1309 free(m
->switch_root
);
1310 free(m
->switch_root_init
);
1312 rlimit_free_all(m
->rlimit
);
1314 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1315 hashmap_free(m
->units_requiring_mounts_for
);
1317 hashmap_free(m
->uid_refs
);
1318 hashmap_free(m
->gid_refs
);
1320 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1321 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1326 static void manager_enumerate_perpetual(Manager
*m
) {
1331 /* Let's ask every type to load all units from disk/kernel that it might know */
1332 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1333 if (!unit_type_supported(c
)) {
1334 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1338 if (unit_vtable
[c
]->enumerate_perpetual
)
1339 unit_vtable
[c
]->enumerate_perpetual(m
);
1344 static void manager_enumerate(Manager
*m
) {
1349 /* Let's ask every type to load all units from disk/kernel that it might know */
1350 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1351 if (!unit_type_supported(c
)) {
1352 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1356 if (unit_vtable
[c
]->enumerate
)
1357 unit_vtable
[c
]->enumerate(m
);
1360 manager_dispatch_load_queue(m
);
1363 static void manager_coldplug(Manager
*m
) {
1371 log_debug("Invoking unit coldplug() handlers…");
1373 /* Let's place the units back into their deserialized state */
1374 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1376 /* ignore aliases */
1380 r
= unit_coldplug(u
);
1382 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1386 static void manager_catchup(Manager
*m
) {
1393 log_debug("Invoking unit catchup() handlers…");
1395 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1396 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1398 /* ignore aliases */
1406 static void manager_build_unit_path_cache(Manager
*m
) {
1412 set_free_free(m
->unit_path_cache
);
1414 m
->unit_path_cache
= set_new(&path_hash_ops
);
1415 if (!m
->unit_path_cache
) {
1420 /* This simply builds a list of files we know exist, so that
1421 * we don't always have to go to disk */
1423 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1424 _cleanup_closedir_
DIR *d
= NULL
;
1429 if (errno
!= ENOENT
)
1430 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1434 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1437 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1443 r
= set_consume(m
->unit_path_cache
, p
);
1452 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1453 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1456 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1462 HASHMAP_FOREACH(u
, m
->units
, i
) {
1464 if (fdset_size(fds
) <= 0)
1467 if (!UNIT_VTABLE(u
)->distribute_fds
)
1470 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1474 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1479 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1480 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1481 * rather than the current one. */
1483 if (m
->test_run_flags
!= 0)
1486 /* If we are in the user instance, and the env var is already set for us, then this means D-Bus is ran
1487 * somewhere outside of our own logic. Let's use it */
1488 if (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"))
1491 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1494 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1497 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1500 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1506 static void manager_setup_bus(Manager
*m
) {
1509 /* Let's set up our private bus connection now, unconditionally */
1510 (void) bus_init_private(m
);
1512 /* If we are in --user mode also connect to the system bus now */
1513 if (MANAGER_IS_USER(m
))
1514 (void) bus_init_system(m
);
1516 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1517 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1518 (void) bus_init_api(m
);
1520 if (MANAGER_IS_SYSTEM(m
))
1521 (void) bus_init_system(m
);
1525 static void manager_preset_all(Manager
*m
) {
1530 if (m
->first_boot
<= 0)
1533 if (!MANAGER_IS_SYSTEM(m
))
1536 if (m
->test_run_flags
!= 0)
1539 /* If this is the first boot, and we are in the host system, then preset everything */
1540 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1542 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1543 "Failed to populate /etc with preset unit settings, ignoring: %m");
1545 log_info("Populated /etc with preset unit settings.");
1548 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1553 /* If we are running in test mode, we still want to run the generators,
1554 * but we should not touch the real generator directories. */
1555 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1556 m
->test_run_flags
? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1561 r
= manager_run_environment_generators(m
);
1565 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_START
);
1566 r
= manager_run_generators(m
);
1567 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_GENERATORS_FINISH
);
1571 manager_preset_all(m
);
1572 lookup_paths_reduce(&m
->lookup_paths
);
1573 manager_build_unit_path_cache(m
);
1575 /* If we will deserialize make sure that during enumeration
1576 * this is already known, so we increase the counter here
1581 /* First, enumerate what we can from all config files */
1582 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_START
);
1583 manager_enumerate_perpetual(m
);
1584 manager_enumerate(m
);
1585 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
1587 /* Second, deserialize if there is something to deserialize */
1588 if (serialization
) {
1589 r
= manager_deserialize(m
, serialization
, fds
);
1591 return log_error_errno(r
, "Deserialization failed: %m");
1594 /* Any fds left? Find some unit which wants them. This is
1595 * useful to allow container managers to pass some file
1596 * descriptors to us pre-initialized. This enables
1597 * socket-based activation of entire containers. */
1598 manager_distribute_fds(m
, fds
);
1600 /* We might have deserialized the notify fd, but if we didn't
1601 * then let's create the bus now */
1602 r
= manager_setup_notify(m
);
1604 /* No sense to continue without notifications, our children would fail anyway. */
1607 r
= manager_setup_cgroups_agent(m
);
1609 /* Likewise, no sense to continue without empty cgroup notifications. */
1612 r
= manager_setup_user_lookup_fd(m
);
1614 /* This shouldn't fail, except if things are really broken. */
1617 /* Connect to the bus if we are good for it */
1618 manager_setup_bus(m
);
1620 /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */
1621 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1622 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1624 /* Third, fire things up! */
1625 manager_coldplug(m
);
1627 /* Release any dynamic users no longer referenced */
1628 dynamic_user_vacuum(m
, true);
1630 exec_runtime_vacuum(m
);
1632 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1633 manager_vacuum_uid_refs(m
);
1634 manager_vacuum_gid_refs(m
);
1636 if (serialization
) {
1637 assert(m
->n_reloading
> 0);
1640 /* Let's wait for the UnitNew/JobNew messages being
1641 * sent, before we notify that the reload is
1643 m
->send_reloading_done
= true;
1646 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1652 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1657 assert(type
< _JOB_TYPE_MAX
);
1659 assert(mode
< _JOB_MODE_MAX
);
1661 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1662 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1664 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1665 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1667 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1669 type
= job_type_collapse(type
, unit
);
1671 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1675 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1676 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1677 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1681 if (mode
== JOB_ISOLATE
) {
1682 r
= transaction_add_isolate_jobs(tr
, m
);
1687 r
= transaction_activate(tr
, m
, mode
, e
);
1691 log_unit_debug(unit
,
1692 "Enqueued job %s/%s as %u", unit
->id
,
1693 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1696 *_ret
= tr
->anchor_job
;
1698 transaction_free(tr
);
1702 transaction_abort(tr
);
1703 transaction_free(tr
);
1707 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1708 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1712 assert(type
< _JOB_TYPE_MAX
);
1714 assert(mode
< _JOB_MODE_MAX
);
1716 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1721 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1724 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1725 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1729 assert(type
< _JOB_TYPE_MAX
);
1731 assert(mode
< _JOB_MODE_MAX
);
1733 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1735 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1740 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1746 assert(mode
< _JOB_MODE_MAX
);
1747 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1749 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1753 /* We need an anchor job */
1754 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1758 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1759 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1761 r
= transaction_activate(tr
, m
, mode
, e
);
1765 transaction_free(tr
);
1769 transaction_abort(tr
);
1770 transaction_free(tr
);
1774 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1777 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1780 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1784 return hashmap_get(m
->units
, name
);
1787 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1792 static const UnitDependency deps
[] = {
1801 while ((u
= m
->target_deps_queue
)) {
1802 assert(u
->in_target_deps_queue
);
1804 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1805 u
->in_target_deps_queue
= false;
1807 for (k
= 0; k
< ELEMENTSOF(deps
); k
++) {
1812 HASHMAP_FOREACH_KEY(v
, target
, u
->dependencies
[deps
[k
]], i
) {
1813 r
= unit_add_default_target_dependency(u
, target
);
1823 unsigned manager_dispatch_load_queue(Manager
*m
) {
1829 /* Make sure we are not run recursively */
1830 if (m
->dispatching_load_queue
)
1833 m
->dispatching_load_queue
= true;
1835 /* Dispatches the load queue. Takes a unit from the queue and
1836 * tries to load its data until the queue is empty */
1838 while ((u
= m
->load_queue
)) {
1839 assert(u
->in_load_queue
);
1845 m
->dispatching_load_queue
= false;
1847 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
1848 * should be loaded and have aliases resolved */
1849 (void) manager_dispatch_target_deps_queue(m
);
1854 int manager_load_unit_prepare(
1861 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
1867 assert(name
|| path
);
1870 /* This will prepare the unit for loading, but not actually
1871 * load anything from disk. */
1873 if (path
&& !is_path(path
))
1874 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1877 name
= basename(path
);
1879 t
= unit_name_to_type(name
);
1881 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1882 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1883 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1885 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1888 ret
= manager_get_unit(m
, name
);
1894 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1899 ret
->fragment_path
= strdup(path
);
1900 if (!ret
->fragment_path
)
1904 r
= unit_add_name(ret
, name
);
1908 unit_add_to_load_queue(ret
);
1909 unit_add_to_dbus_queue(ret
);
1910 unit_add_to_gc_queue(ret
);
1918 int manager_load_unit(
1930 /* This will load the service information files, but not actually
1931 * start any services or anything. */
1933 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1937 manager_dispatch_load_queue(m
);
1939 *_ret
= unit_follow_merge(*_ret
);
1943 int manager_load_startable_unit_or_warn(
1949 /* Load a unit, make sure it loaded fully and is not masked. */
1951 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1955 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
1957 return log_error_errno(r
, "Failed to load %s %s: %s",
1958 name
? "unit" : "unit file", name
?: path
,
1959 bus_error_message(&error
, r
));
1961 r
= bus_unit_validate_load_state(unit
, &error
);
1963 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
1969 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1976 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1977 job_dump(j
, f
, prefix
);
1980 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1988 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1990 unit_dump(u
, f
, prefix
);
1993 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
1999 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2000 char buf
[FORMAT_TIMESTAMP_MAX
];
2002 if (dual_timestamp_is_set(m
->timestamps
+ q
))
2003 fprintf(f
, "%sTimestamp %s: %s\n",
2005 manager_timestamp_to_string(q
),
2006 format_timestamp(buf
, sizeof(buf
), m
->timestamps
[q
].realtime
));
2009 manager_dump_units(m
, f
, prefix
);
2010 manager_dump_jobs(m
, f
, prefix
);
2013 int manager_get_dump_string(Manager
*m
, char **ret
) {
2014 _cleanup_free_
char *dump
= NULL
;
2015 _cleanup_fclose_
FILE *f
= NULL
;
2022 f
= open_memstream(&dump
, &size
);
2026 (void) __fsetlocking(f
, FSETLOCKING_BYCALLER
);
2028 manager_dump(m
, f
, NULL
);
2030 r
= fflush_and_check(f
);
2036 *ret
= TAKE_PTR(dump
);
2041 void manager_clear_jobs(Manager
*m
) {
2046 while ((j
= hashmap_first(m
->jobs
)))
2047 /* No need to recurse. We're cancelling all jobs. */
2048 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2051 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2052 Manager
*m
= userdata
;
2058 while ((j
= m
->run_queue
)) {
2059 assert(j
->installed
);
2060 assert(j
->in_run_queue
);
2062 job_run_and_invalidate(j
);
2065 if (m
->n_running_jobs
> 0)
2066 manager_watch_jobs_in_progress(m
);
2068 if (m
->n_on_console
> 0)
2069 manager_watch_idle_pipe(m
);
2074 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2075 unsigned n
= 0, budget
;
2081 if (m
->dispatching_dbus_queue
)
2084 /* Anything to do at all? */
2085 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
&& !m
->send_reloading_done
&& !m
->queued_message
)
2088 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's sit this
2089 * cycle out, and process things in a later cycle when the queues got a bit emptier. */
2090 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2093 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't overly
2094 * full before this call we shouldn't increase it in size too wildly in one step, and we shouldn't monopolize
2095 * CPU time with generating these messages. Note the difference in counting of this "budget" and the
2096 * "threshold" above: the "budget" is decreased only once per generated message, regardless how many
2097 * busses/direct connections it is enqueued on, while the "threshold" is applied to each queued instance of bus
2098 * message, i.e. if the same message is enqueued to five busses/direct connections it will be counted five
2099 * times. This difference in counting ("references" vs. "instances") is primarily a result of the fact that
2100 * it's easier to implement it this way, however it also reflects the thinking that the "threshold" should put
2101 * a limit on used queue memory, i.e. space, while the "budget" should put a limit on time. Also note that
2102 * the "threshold" is currently chosen much higher than the "budget". */
2103 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2105 m
->dispatching_dbus_queue
= true;
2107 while (budget
> 0 && (u
= m
->dbus_unit_queue
)) {
2109 assert(u
->in_dbus_queue
);
2111 bus_unit_send_change_signal(u
);
2115 while (budget
> 0 && (j
= m
->dbus_job_queue
)) {
2116 assert(j
->in_dbus_queue
);
2118 bus_job_send_change_signal(j
);
2122 m
->dispatching_dbus_queue
= false;
2124 if (budget
> 0 && m
->send_reloading_done
) {
2125 m
->send_reloading_done
= false;
2126 bus_manager_send_reloading(m
, false);
2130 if (budget
> 0 && m
->queued_message
) {
2131 bus_send_queued_message(m
);
2138 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2139 Manager
*m
= userdata
;
2140 char buf
[PATH_MAX
+1];
2143 n
= recv(fd
, buf
, sizeof(buf
), 0);
2145 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2147 log_error("Got zero-length cgroups agent message, ignoring.");
2150 if ((size_t) n
>= sizeof(buf
)) {
2151 log_error("Got overly long cgroups agent message, ignoring.");
2155 if (memchr(buf
, 0, n
)) {
2156 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2161 manager_notify_cgroup_empty(m
, buf
);
2162 (void) bus_forward_agent_released(m
, buf
);
2167 static void manager_invoke_notify_message(
2170 const struct ucred
*ucred
,
2179 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2181 u
->notifygen
= m
->notifygen
;
2183 if (UNIT_VTABLE(u
)->notify_message
) {
2184 _cleanup_strv_free_
char **tags
= NULL
;
2186 tags
= strv_split(buf
, NEWLINE
);
2192 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2194 } else if (DEBUG_LOGGING
) {
2195 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
2197 x
= ellipsize(buf
, 20, 90);
2201 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2205 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2207 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2208 Manager
*m
= userdata
;
2209 char buf
[NOTIFY_BUFFER_MAX
+1];
2210 struct iovec iovec
= {
2212 .iov_len
= sizeof(buf
)-1,
2215 struct cmsghdr cmsghdr
;
2216 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
2217 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
2219 struct msghdr msghdr
= {
2222 .msg_control
= &control
,
2223 .msg_controllen
= sizeof(control
),
2226 struct cmsghdr
*cmsg
;
2227 struct ucred
*ucred
= NULL
;
2228 _cleanup_free_ Unit
**array_copy
= NULL
;
2229 Unit
*u1
, *u2
, **array
;
2230 int r
, *fd_array
= NULL
;
2236 assert(m
->notify_fd
== fd
);
2238 if (revents
!= EPOLLIN
) {
2239 log_warning("Got unexpected poll event for notify fd.");
2243 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2245 if (IN_SET(errno
, EAGAIN
, EINTR
))
2246 return 0; /* Spurious wakeup, try again */
2248 /* If this is any other, real error, then let's stop processing this socket. This of course means we
2249 * won't take notification messages anymore, but that's still better than busy looping around this:
2250 * being woken up over and over again but being unable to actually read the message off the socket. */
2251 return log_error_errno(errno
, "Failed to receive notification message: %m");
2254 CMSG_FOREACH(cmsg
, &msghdr
) {
2255 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2257 fd_array
= (int*) CMSG_DATA(cmsg
);
2258 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2260 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2261 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2262 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2264 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2271 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2273 close_many(fd_array
, n_fds
);
2279 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2280 log_warning("Received notify message without valid credentials. Ignoring.");
2284 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2285 log_warning("Received notify message exceeded maximum size. Ignoring.");
2289 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2290 * trailing NUL byte in the message, but don't expect it. */
2291 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2292 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2296 /* Make sure it's NUL-terminated. */
2299 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2302 /* Notify every unit that might be interested, which might be multiple. */
2303 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2304 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2305 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2312 array_copy
= newdup(Unit
*, array
, k
+1);
2316 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2317 * make sure we only invoke each unit's handler once. */
2319 manager_invoke_notify_message(m
, u1
, ucred
, buf
, fds
);
2323 manager_invoke_notify_message(m
, u2
, ucred
, buf
, fds
);
2327 for (size_t i
= 0; array_copy
[i
]; i
++) {
2328 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, buf
, fds
);
2333 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2335 if (fdset_size(fds
) > 0)
2336 log_warning("Got extra auxiliary fds with notification message, closing them.");
2341 static void manager_invoke_sigchld_event(
2344 const siginfo_t
*si
) {
2350 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2351 if (u
->sigchldgen
== m
->sigchldgen
)
2353 u
->sigchldgen
= m
->sigchldgen
;
2355 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2356 unit_unwatch_pid(u
, si
->si_pid
);
2358 if (UNIT_VTABLE(u
)->sigchld_event
)
2359 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2362 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2363 Manager
*m
= userdata
;
2370 /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2371 * while it is a zombie. */
2373 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2375 if (errno
!= ECHILD
)
2376 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2384 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2385 _cleanup_free_ Unit
**array_copy
= NULL
;
2386 _cleanup_free_
char *name
= NULL
;
2387 Unit
*u1
, *u2
, **array
;
2389 (void) get_process_comm(si
.si_pid
, &name
);
2391 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2392 si
.si_pid
, strna(name
),
2393 sigchld_code_to_string(si
.si_code
),
2395 strna(si
.si_code
== CLD_EXITED
2396 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2397 : signal_to_string(si
.si_status
)));
2399 /* Increase the generation counter used for filtering out duplicate unit invocations */
2402 /* And now figure out the unit this belongs to, it might be multiple... */
2403 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2404 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2405 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2409 /* Cound how many entries the array has */
2413 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2414 array_copy
= newdup(Unit
*, array
, n
+1);
2419 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2420 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2421 * each iteration. */
2423 manager_invoke_sigchld_event(m
, u1
, &si
);
2425 manager_invoke_sigchld_event(m
, u2
, &si
);
2427 for (size_t i
= 0; array_copy
[i
]; i
++)
2428 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2431 /* And now, we actually reap the zombie. */
2432 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2433 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2440 /* All children processed for now, turn off event source */
2442 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2444 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2449 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2450 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2453 log_debug("Activating special unit %s", name
);
2455 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
2457 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2460 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2461 /* If the user presses C-A-D more than
2462 * 7 times within 2s, we reboot/shutdown immediately,
2463 * unless it was disabled in system.conf */
2465 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2466 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2468 emergency_action(m
, m
->cad_burst_action
, NULL
,
2469 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2472 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2473 Manager
*m
= userdata
;
2475 struct signalfd_siginfo sfsi
;
2479 assert(m
->signal_fd
== fd
);
2481 if (revents
!= EPOLLIN
) {
2482 log_warning("Got unexpected events from signal file descriptor.");
2486 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2487 if (n
!= sizeof(sfsi
)) {
2489 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2493 if (IN_SET(errno
, EINTR
, EAGAIN
))
2496 /* We return an error here, which will kill this handler,
2497 * to avoid a busy loop on read error. */
2498 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2501 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2502 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2503 ? LOG_DEBUG
: LOG_INFO
,
2506 switch (sfsi
.ssi_signo
) {
2509 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2511 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2516 if (MANAGER_IS_SYSTEM(m
)) {
2517 /* This is for compatibility with the original sysvinit */
2518 r
= verify_run_space_and_log("Refusing to reexecute");
2520 m
->exit_code
= MANAGER_REEXECUTE
;
2526 if (MANAGER_IS_SYSTEM(m
))
2527 manager_handle_ctrl_alt_del(m
);
2529 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2530 JOB_REPLACE_IRREVERSIBLY
);
2534 /* This is a nop on non-init */
2535 if (MANAGER_IS_SYSTEM(m
))
2536 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2541 /* This is a nop on non-init */
2542 if (MANAGER_IS_SYSTEM(m
))
2543 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2548 if (manager_dbus_is_running(m
, false)) {
2549 log_info("Trying to reconnect to bus...");
2551 (void) bus_init_api(m
);
2553 if (MANAGER_IS_SYSTEM(m
))
2554 (void) bus_init_system(m
);
2556 log_info("Starting D-Bus service...");
2557 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2563 _cleanup_free_
char *dump
= NULL
;
2565 r
= manager_get_dump_string(m
, &dump
);
2567 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2571 log_dump(LOG_INFO
, dump
);
2576 r
= verify_run_space_and_log("Refusing to reload");
2578 m
->exit_code
= MANAGER_RELOAD
;
2583 /* Starting SIGRTMIN+0 */
2584 static const struct {
2587 } target_table
[] = {
2588 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2589 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2590 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2591 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2592 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2593 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2594 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2597 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2598 static const ManagerExitCode code_table
[] = {
2600 [1] = MANAGER_POWEROFF
,
2601 [2] = MANAGER_REBOOT
,
2602 [3] = MANAGER_KEXEC
,
2605 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2606 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2607 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2608 manager_start_target(m
, target_table
[idx
].target
,
2609 target_table
[idx
].mode
);
2613 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2614 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2615 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2619 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2622 manager_set_show_status(m
, SHOW_STATUS_YES
);
2626 manager_set_show_status(m
, SHOW_STATUS_NO
);
2630 manager_override_log_level(m
, LOG_DEBUG
);
2634 manager_restore_original_log_level(m
);
2638 if (MANAGER_IS_USER(m
)) {
2639 m
->exit_code
= MANAGER_EXIT
;
2643 /* This is a nop on init */
2647 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2648 manager_restore_original_log_target(m
);
2652 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2656 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2660 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2667 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2668 Manager
*m
= userdata
;
2673 assert(m
->time_change_fd
== fd
);
2675 log_struct(LOG_DEBUG
,
2676 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2677 LOG_MESSAGE("Time has been changed"));
2679 /* Restart the watch */
2680 (void) manager_setup_time_change(m
);
2682 HASHMAP_FOREACH(u
, m
->units
, i
)
2683 if (UNIT_VTABLE(u
)->time_change
)
2684 UNIT_VTABLE(u
)->time_change(u
);
2689 static int manager_dispatch_timezone_change(
2690 sd_event_source
*source
,
2691 const struct inotify_event
*e
,
2694 Manager
*m
= userdata
;
2701 log_debug("inotify event for /etc/localtime");
2703 changed
= manager_read_timezone_stat(m
);
2709 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2710 (void) manager_setup_timezone_change(m
);
2712 /* Read the new timezone */
2715 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2717 HASHMAP_FOREACH(u
, m
->units
, i
)
2718 if (UNIT_VTABLE(u
)->timezone_change
)
2719 UNIT_VTABLE(u
)->timezone_change(u
);
2724 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2725 Manager
*m
= userdata
;
2728 assert(m
->idle_pipe
[2] == fd
);
2730 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2731 * now turn off any further console output if there's at least one service that needs console access, so that
2732 * from now on our own output should not spill into that service's output anymore. After all, we support
2733 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2734 * exclusively without our interference. */
2735 m
->no_console_output
= m
->n_on_console
> 0;
2737 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2738 * by closing the pipes towards them, which is what they are waiting for. */
2739 manager_close_idle_pipe(m
);
2744 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2745 Manager
*m
= userdata
;
2752 manager_print_jobs_in_progress(m
);
2754 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2755 r
= sd_event_source_set_time(source
, next
);
2759 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2762 int manager_loop(Manager
*m
) {
2765 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2768 m
->exit_code
= MANAGER_OK
;
2770 /* Release the path cache */
2771 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2773 manager_check_finished(m
);
2775 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2776 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2778 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2780 while (m
->exit_code
== MANAGER_OK
) {
2783 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2786 if (!ratelimit_below(&rl
)) {
2787 /* Yay, something is going seriously wrong, pause a little */
2788 log_warning("Looping too fast. Throttling execution a little.");
2792 if (manager_dispatch_load_queue(m
) > 0)
2795 if (manager_dispatch_gc_job_queue(m
) > 0)
2798 if (manager_dispatch_gc_unit_queue(m
) > 0)
2801 if (manager_dispatch_cleanup_queue(m
) > 0)
2804 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
2807 if (manager_dispatch_dbus_queue(m
) > 0)
2810 /* Sleep for half the watchdog time */
2811 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2812 wait_usec
= m
->runtime_watchdog
/ 2;
2816 wait_usec
= USEC_INFINITY
;
2818 r
= sd_event_run(m
->event
, wait_usec
);
2820 return log_error_errno(r
, "Failed to run event loop: %m");
2823 return m
->exit_code
;
2826 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2827 _cleanup_free_
char *n
= NULL
;
2828 sd_id128_t invocation_id
;
2836 r
= unit_name_from_dbus_path(s
, &n
);
2840 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2841 * as invocation ID. */
2842 r
= sd_id128_from_string(n
, &invocation_id
);
2844 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2850 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
2851 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
2852 SD_ID128_FORMAT_VAL(invocation_id
));
2855 /* If this didn't work, we check if this is a unit name */
2856 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2857 _cleanup_free_
char *nn
= NULL
;
2860 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
2861 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
2864 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2872 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2882 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2886 r
= safe_atou(p
, &id
);
2890 j
= manager_get_job(m
, id
);
2899 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2902 _cleanup_free_
char *p
= NULL
;
2906 if (!MANAGER_IS_SYSTEM(m
))
2909 audit_fd
= get_audit_fd();
2913 /* Don't generate audit events if the service was already
2914 * started and we're just deserializing */
2915 if (MANAGER_IS_RELOADING(m
))
2918 if (u
->type
!= UNIT_SERVICE
)
2921 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2923 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2927 msg
= strjoina("unit=", p
);
2928 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2930 /* We aren't allowed to send audit messages?
2931 * Then let's not retry again. */
2934 log_warning_errno(errno
, "Failed to send audit message: %m");
2940 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2941 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2942 _cleanup_free_
char *message
= NULL
;
2943 _cleanup_close_
int fd
= -1;
2946 /* Don't generate plymouth events if the service was already
2947 * started and we're just deserializing */
2948 if (MANAGER_IS_RELOADING(m
))
2951 if (!MANAGER_IS_SYSTEM(m
))
2954 if (detect_container() > 0)
2957 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
2960 /* We set SOCK_NONBLOCK here so that we rather drop the
2961 * message then wait for plymouth */
2962 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2964 log_error_errno(errno
, "socket() failed: %m");
2968 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2969 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2970 log_error_errno(errno
, "connect() failed: %m");
2974 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2980 if (write(fd
, message
, n
+ 1) != n
+ 1)
2981 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2982 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2985 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2991 fd
= open_serialization_fd("systemd-state");
2995 f
= fdopen(fd
, "w+");
3005 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
3018 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
3019 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
3020 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
3021 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
3022 fprintf(f
, "ready-sent=%s\n", yes_no(m
->ready_sent
));
3023 fprintf(f
, "taint-logged=%s\n", yes_no(m
->taint_logged
));
3024 fprintf(f
, "service-watchdogs=%s\n", yes_no(m
->service_watchdogs
));
3026 if (m
->log_level_overridden
)
3027 fprintf(f
, "log-level-override=%i\n", log_get_max_level());
3028 if (m
->log_target_overridden
)
3029 fprintf(f
, "log-target-override=%s\n", log_target_to_string(log_get_target()));
3031 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3032 /* The userspace and finish timestamps only apply to the host system, hence only serialize them there */
3033 if (in_initrd() && IN_SET(q
, MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
))
3036 t
= manager_timestamp_to_string(q
);
3038 char field
[strlen(t
) + STRLEN("-timestamp") + 1];
3039 strcpy(stpcpy(field
, t
), "-timestamp");
3040 dual_timestamp_serialize(f
, field
, m
->timestamps
+ q
);
3044 if (!switching_root
)
3045 (void) serialize_environment(f
, m
->environment
);
3047 if (m
->notify_fd
>= 0) {
3050 copy
= fdset_put_dup(fds
, m
->notify_fd
);
3054 fprintf(f
, "notify-fd=%i\n", copy
);
3055 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
3058 if (m
->cgroups_agent_fd
>= 0) {
3061 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
3065 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
3068 if (m
->user_lookup_fds
[0] >= 0) {
3071 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3075 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3079 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
3082 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3084 r
= dynamic_user_serialize(m
, f
, fds
);
3088 manager_serialize_uid_refs(m
, f
);
3089 manager_serialize_gid_refs(m
, f
);
3091 r
= exec_runtime_serialize(m
, f
, fds
);
3095 (void) fputc('\n', f
);
3097 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
3105 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3112 assert(m
->n_reloading
> 0);
3118 r
= bus_fdset_add_all(m
, fds
);
3125 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3131 log_debug("Deserializing state...");
3136 char line
[LINE_MAX
];
3137 const char *val
, *l
;
3139 if (!fgets(line
, sizeof(line
), f
)) {
3154 if ((val
= startswith(l
, "current-job-id="))) {
3157 if (safe_atou32(val
, &id
) < 0)
3158 log_notice("Failed to parse current job id value %s", val
);
3160 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3162 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3165 if (safe_atou32(val
, &n
) < 0)
3166 log_notice("Failed to parse installed jobs counter %s", val
);
3168 m
->n_installed_jobs
+= n
;
3170 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3173 if (safe_atou32(val
, &n
) < 0)
3174 log_notice("Failed to parse failed jobs counter %s", val
);
3176 m
->n_failed_jobs
+= n
;
3178 } else if ((val
= startswith(l
, "taint-usr="))) {
3181 b
= parse_boolean(val
);
3183 log_notice("Failed to parse taint /usr flag %s", val
);
3185 m
->taint_usr
= m
->taint_usr
|| b
;
3187 } else if ((val
= startswith(l
, "ready-sent="))) {
3190 b
= parse_boolean(val
);
3192 log_notice("Failed to parse ready-sent flag %s", val
);
3194 m
->ready_sent
= m
->ready_sent
|| b
;
3196 } else if ((val
= startswith(l
, "taint-logged="))) {
3199 b
= parse_boolean(val
);
3201 log_notice("Failed to parse taint-logged flag %s", val
);
3203 m
->taint_logged
= m
->taint_logged
|| b
;
3205 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3208 b
= parse_boolean(val
);
3210 log_notice("Failed to parse service-watchdogs flag %s", val
);
3212 m
->service_watchdogs
= b
;
3214 } else if ((val
= startswith(l
, "log-level-override="))) {
3217 level
= log_level_from_string(val
);
3219 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3221 manager_override_log_level(m
, level
);
3223 } else if ((val
= startswith(l
, "log-target-override="))) {
3226 target
= log_target_from_string(val
);
3228 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3230 manager_override_log_target(m
, target
);
3232 } else if (startswith(l
, "env=")) {
3233 r
= deserialize_environment(&m
->environment
, l
);
3237 log_notice_errno(r
, "Failed to parse environment entry: \"%s\": %m", l
);
3239 } else if ((val
= startswith(l
, "notify-fd="))) {
3242 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3243 log_notice("Failed to parse notify fd: \"%s\"", val
);
3245 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3246 safe_close(m
->notify_fd
);
3247 m
->notify_fd
= fdset_remove(fds
, fd
);
3250 } else if ((val
= startswith(l
, "notify-socket="))) {
3259 free(m
->notify_socket
);
3260 m
->notify_socket
= n
;
3262 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3265 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3266 log_notice("Failed to parse cgroups agent fd: %s", val
);
3268 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3269 safe_close(m
->cgroups_agent_fd
);
3270 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3273 } else if ((val
= startswith(l
, "user-lookup="))) {
3276 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3277 log_notice("Failed to parse user lookup fd: %s", val
);
3279 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3280 safe_close_pair(m
->user_lookup_fds
);
3281 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3282 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3285 } else if ((val
= startswith(l
, "dynamic-user=")))
3286 dynamic_user_deserialize_one(m
, val
, fds
);
3287 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3288 manager_deserialize_uid_refs_one(m
, val
);
3289 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3290 manager_deserialize_gid_refs_one(m
, val
);
3291 else if ((val
= startswith(l
, "exec-runtime=")))
3292 exec_runtime_deserialize_one(m
, val
, fds
);
3293 else if ((val
= startswith(l
, "subscribed="))) {
3295 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3300 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3301 val
= startswith(l
, manager_timestamp_to_string(q
));
3305 val
= startswith(val
, "-timestamp=");
3310 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3311 dual_timestamp_deserialize(val
, m
->timestamps
+ q
);
3312 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3313 log_notice("Unknown serialization item '%s'", l
);
3319 char name
[UNIT_NAME_MAX
+2];
3320 const char* unit_name
;
3323 if (!fgets(name
, sizeof(name
), f
)) {
3333 unit_name
= strstrip(name
);
3335 r
= manager_load_unit(m
, unit_name
, NULL
, NULL
, &u
);
3337 log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", unit_name
);
3340 unit_deserialize_skip(f
);
3344 r
= unit_deserialize(u
, f
, fds
);
3346 log_notice_errno(r
, "Failed to deserialize unit \"%s\": %m", unit_name
);
3356 assert(m
->n_reloading
> 0);
3362 static void manager_flush_finished_jobs(Manager
*m
) {
3365 while ((j
= set_steal_first(m
->pending_finished_jobs
))) {
3366 bus_job_send_removed_signal(j
);
3370 m
->pending_finished_jobs
= set_free(m
->pending_finished_jobs
);
3373 int manager_reload(Manager
*m
) {
3375 _cleanup_fclose_
FILE *f
= NULL
;
3376 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3380 r
= manager_open_serialization(m
, &f
);
3385 bus_manager_send_reloading(m
, true);
3393 r
= manager_serialize(m
, f
, fds
, false);
3399 if (fseeko(f
, 0, SEEK_SET
) < 0) {
3404 /* From here on there is no way back. */
3405 manager_clear_jobs_and_units(m
);
3406 lookup_paths_flush_generator(&m
->lookup_paths
);
3407 lookup_paths_free(&m
->lookup_paths
);
3408 exec_runtime_vacuum(m
);
3409 dynamic_user_vacuum(m
, false);
3410 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3411 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3413 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3414 if (q
< 0 && r
>= 0)
3417 q
= manager_run_environment_generators(m
);
3418 if (q
< 0 && r
>= 0)
3421 /* Find new unit paths */
3422 q
= manager_run_generators(m
);
3423 if (q
< 0 && r
>= 0)
3426 lookup_paths_reduce(&m
->lookup_paths
);
3427 manager_build_unit_path_cache(m
);
3429 /* First, enumerate what we can from all config files */
3430 manager_enumerate(m
);
3432 /* Second, deserialize our stored data */
3433 q
= manager_deserialize(m
, f
, fds
);
3435 log_error_errno(q
, "Deserialization failed: %m");
3443 /* Re-register notify_fd as event source */
3444 q
= manager_setup_notify(m
);
3445 if (q
< 0 && r
>= 0)
3448 q
= manager_setup_cgroups_agent(m
);
3449 if (q
< 0 && r
>= 0)
3452 q
= manager_setup_user_lookup_fd(m
);
3453 if (q
< 0 && r
>= 0)
3456 /* Third, fire things up! */
3457 manager_coldplug(m
);
3459 /* Release any dynamic users no longer referenced */
3460 dynamic_user_vacuum(m
, true);
3462 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
3463 manager_vacuum_uid_refs(m
);
3464 manager_vacuum_gid_refs(m
);
3466 exec_runtime_vacuum(m
);
3468 assert(m
->n_reloading
> 0);
3471 /* It might be safe to log to the journal now and connect to dbus */
3472 manager_recheck_journal(m
);
3473 manager_recheck_dbus(m
);
3475 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
3478 /* Sync current state of bus names with our set of listening units */
3479 q
= manager_enqueue_sync_bus_names(m
);
3480 if (q
< 0 && r
>= 0)
3483 if (!MANAGER_IS_RELOADING(m
))
3484 manager_flush_finished_jobs(m
);
3486 m
->send_reloading_done
= true;
3491 void manager_reset_failed(Manager
*m
) {
3497 HASHMAP_FOREACH(u
, m
->units
, i
)
3498 unit_reset_failed(u
);
3501 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3507 /* Returns true if the unit is inactive or going down */
3508 u
= manager_get_unit(m
, name
);
3512 return unit_inactive_or_pending(u
);
3515 static void log_taint_string(Manager
*m
) {
3516 _cleanup_free_
char *taint
= NULL
;
3520 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3523 m
->taint_logged
= true; /* only check for taint once */
3525 taint
= manager_taint_string(m
);
3529 log_struct(LOG_NOTICE
,
3530 LOG_MESSAGE("System is tainted: %s", taint
),
3532 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3535 static void manager_notify_finished(Manager
*m
) {
3536 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3537 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3539 if (m
->test_run_flags
)
3542 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3543 char ts
[FORMAT_TIMESPAN_MAX
];
3544 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3547 size_t size
= sizeof buf
;
3549 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3550 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3551 * negative values. */
3553 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3554 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3555 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3556 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3558 if (firmware_usec
> 0)
3559 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3560 if (loader_usec
> 0)
3561 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3563 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3565 /* The initrd case on bare-metal*/
3566 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3567 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3569 log_struct(LOG_INFO
,
3570 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3571 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3572 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3573 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3574 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3576 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3577 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
3578 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3579 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3581 /* The initrd-less case on bare-metal*/
3583 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3586 log_struct(LOG_INFO
,
3587 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3588 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3589 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3590 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3592 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
3593 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
3594 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3597 /* The container and --user case */
3598 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3599 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3601 log_struct(LOG_INFO
,
3602 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3603 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3604 LOG_MESSAGE("Startup finished in %s.",
3605 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
3608 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3611 m
->ready_sent
? "STATUS=Startup finished in %s."
3613 "STATUS=Startup finished in %s.",
3614 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
3615 m
->ready_sent
= true;
3617 log_taint_string(m
);
3620 static void manager_send_ready(Manager
*m
) {
3623 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3624 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3627 m
->ready_sent
= true;
3631 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3634 static void manager_check_basic_target(Manager
*m
) {
3639 /* Small shortcut */
3640 if (m
->ready_sent
&& m
->taint_logged
)
3643 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3644 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3647 /* For user managers, send out READY=1 as soon as we reach basic.target */
3648 manager_send_ready(m
);
3650 /* Log the taint string as soon as we reach basic.target */
3651 log_taint_string(m
);
3654 void manager_check_finished(Manager
*m
) {
3657 if (MANAGER_IS_RELOADING(m
))
3660 /* Verify that we have entered the event loop already, and not left it again. */
3661 if (!MANAGER_IS_RUNNING(m
))
3664 manager_check_basic_target(m
);
3666 if (hashmap_size(m
->jobs
) > 0) {
3667 if (m
->jobs_in_progress_event_source
)
3668 /* Ignore any failure, this is only for feedback */
3669 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
3674 manager_flip_auto_status(m
, false);
3676 /* Notify Type=idle units that we are done now */
3677 manager_close_idle_pipe(m
);
3679 /* Turn off confirm spawn now */
3680 m
->confirm_spawn
= NULL
;
3682 /* No need to update ask password status when we're going non-interactive */
3683 manager_close_ask_password(m
);
3685 /* This is no longer the first boot */
3686 manager_set_first_boot(m
, false);
3688 if (MANAGER_IS_FINISHED(m
))
3691 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3693 manager_notify_finished(m
);
3695 manager_invalidate_startup_units(m
);
3698 static bool generator_path_any(const char* const* paths
) {
3702 /* Optimize by skipping the whole process by not creating output directories
3703 * if no generators are found. */
3704 STRV_FOREACH(path
, (char**) paths
)
3705 if (access(*path
, F_OK
) == 0)
3707 else if (errno
!= ENOENT
)
3708 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3713 static const char* system_env_generator_binary_paths
[] = {
3714 "/run/systemd/system-environment-generators",
3715 "/etc/systemd/system-environment-generators",
3716 "/usr/local/lib/systemd/system-environment-generators",
3717 SYSTEM_ENV_GENERATOR_PATH
,
3721 static const char* user_env_generator_binary_paths
[] = {
3722 "/run/systemd/user-environment-generators",
3723 "/etc/systemd/user-environment-generators",
3724 "/usr/local/lib/systemd/user-environment-generators",
3725 USER_ENV_GENERATOR_PATH
,
3729 static int manager_run_environment_generators(Manager
*m
) {
3730 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3732 void* args
[] = {&tmp
, &tmp
, &m
->environment
};
3734 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3737 paths
= MANAGER_IS_SYSTEM(m
) ? system_env_generator_binary_paths
: user_env_generator_binary_paths
;
3739 if (!generator_path_any(paths
))
3742 return execute_directories(paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
, args
, NULL
);
3745 static int manager_run_generators(Manager
*m
) {
3746 _cleanup_strv_free_
char **paths
= NULL
;
3747 const char *argv
[5];
3752 if (m
->test_run_flags
&& !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3755 paths
= generator_binary_paths(m
->unit_file_scope
);
3759 if (!generator_path_any((const char* const*) paths
))
3762 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3766 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3767 argv
[1] = m
->lookup_paths
.generator
;
3768 argv
[2] = m
->lookup_paths
.generator_early
;
3769 argv
[3] = m
->lookup_paths
.generator_late
;
3772 RUN_WITH_UMASK(0022)
3773 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
,
3774 NULL
, NULL
, (char**) argv
);
3777 lookup_paths_trim_generator(&m
->lookup_paths
);
3781 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3782 char **a
= NULL
, **b
= NULL
, **l
;
3787 if (!strv_isempty(minus
)) {
3788 a
= strv_env_delete(l
, 1, minus
);
3795 if (!strv_isempty(plus
)) {
3796 b
= strv_env_merge(2, l
, plus
);
3805 if (m
->environment
!= l
)
3806 strv_free(m
->environment
);
3813 manager_sanitize_environment(m
);
3818 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3823 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3824 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3826 if (!default_rlimit
[i
])
3829 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3837 void manager_recheck_dbus(Manager
*m
) {
3840 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3841 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3842 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3843 * while in the user instance we can assume it's already there. */
3845 if (MANAGER_IS_RELOADING(m
))
3846 return; /* don't check while we are reloading… */
3848 if (manager_dbus_is_running(m
, false)) {
3849 (void) bus_init_api(m
);
3851 if (MANAGER_IS_SYSTEM(m
))
3852 (void) bus_init_system(m
);
3854 (void) bus_done_api(m
);
3856 if (MANAGER_IS_SYSTEM(m
))
3857 (void) bus_done_system(m
);
3861 static bool manager_journal_is_running(Manager
*m
) {
3866 if (m
->test_run_flags
!= 0)
3869 /* If we are the user manager we can safely assume that the journal is up */
3870 if (!MANAGER_IS_SYSTEM(m
))
3873 /* Check that the socket is not only up, but in RUNNING state */
3874 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3877 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3880 /* Similar, check if the daemon itself is fully up, too */
3881 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3884 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3890 void manager_recheck_journal(Manager
*m
) {
3894 /* Don't bother with this unless we are in the special situation of being PID 1 */
3895 if (getpid_cached() != 1)
3898 /* Don't check this while we are reloading, things might still change */
3899 if (MANAGER_IS_RELOADING(m
))
3902 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3903 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3904 * an activation ourselves we can't fulfill. */
3905 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3909 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3911 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3913 if (!MANAGER_IS_SYSTEM(m
))
3916 if (m
->show_status
!= mode
)
3917 log_debug("%s showing of status.",
3918 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3919 m
->show_status
= mode
;
3922 (void) touch("/run/systemd/show-status");
3924 (void) unlink("/run/systemd/show-status");
3927 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3930 if (!MANAGER_IS_SYSTEM(m
))
3933 if (m
->no_console_output
)
3936 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3939 /* If we cannot find out the status properly, just proceed. */
3940 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3943 return m
->show_status
> 0;
3946 const char *manager_get_confirm_spawn(Manager
*m
) {
3947 static int last_errno
= 0;
3948 const char *vc
= m
->confirm_spawn
;
3952 /* Here's the deal: we want to test the validity of the console but don't want
3953 * PID1 to go through the whole console process which might block. But we also
3954 * want to warn the user only once if something is wrong with the console so we
3955 * cannot do the sanity checks after spawning our children. So here we simply do
3956 * really basic tests to hopefully trap common errors.
3958 * If the console suddenly disappear at the time our children will really it
3959 * then they will simply fail to acquire it and a positive answer will be
3960 * assumed. New children will fallback to /dev/console though.
3962 * Note: TTYs are devices that can come and go any time, and frequently aren't
3963 * available yet during early boot (consider a USB rs232 dongle...). If for any
3964 * reason the configured console is not ready, we fallback to the default
3967 if (!vc
|| path_equal(vc
, "/dev/console"))
3974 if (!S_ISCHR(st
.st_mode
)) {
3982 if (last_errno
!= errno
) {
3984 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3986 return "/dev/console";
3989 void manager_set_first_boot(Manager
*m
, bool b
) {
3992 if (!MANAGER_IS_SYSTEM(m
))
3995 if (m
->first_boot
!= (int) b
) {
3997 (void) touch("/run/systemd/first-boot");
3999 (void) unlink("/run/systemd/first-boot");
4005 void manager_disable_confirm_spawn(void) {
4006 (void) touch("/run/systemd/confirm_spawn_disabled");
4009 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4010 if (!m
->confirm_spawn
)
4013 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4016 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4019 /* If m is NULL, assume we're after shutdown and let the messages through. */
4021 if (m
&& !manager_get_show_status(m
, type
))
4024 /* XXX We should totally drop the check for ephemeral here
4025 * and thus effectively make 'Type=idle' pointless. */
4026 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4029 va_start(ap
, format
);
4030 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
4034 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4035 char p
[strlen(path
)+1];
4041 path_simplify(p
, false);
4043 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4046 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4051 assert(u
->manager
== m
);
4053 size
= set_size(m
->failed_units
);
4056 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
4060 if (set_put(m
->failed_units
, u
) < 0)
4063 (void) set_remove(m
->failed_units
, u
);
4065 if (set_size(m
->failed_units
) != size
)
4066 bus_manager_send_change_signal(m
);
4071 ManagerState
manager_state(Manager
*m
) {
4076 /* Did we ever finish booting? If not then we are still starting up */
4077 if (!MANAGER_IS_FINISHED(m
)) {
4079 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4080 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4081 return MANAGER_INITIALIZING
;
4083 return MANAGER_STARTING
;
4086 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4087 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4088 if (u
&& unit_active_or_pending(u
))
4089 return MANAGER_STOPPING
;
4091 if (MANAGER_IS_SYSTEM(m
)) {
4092 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4093 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4094 if (u
&& unit_active_or_pending(u
))
4095 return MANAGER_MAINTENANCE
;
4097 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4098 if (u
&& unit_active_or_pending(u
))
4099 return MANAGER_MAINTENANCE
;
4102 /* Are there any failed units? If so, we are in degraded mode */
4103 if (set_size(m
->failed_units
) > 0)
4104 return MANAGER_DEGRADED
;
4106 return MANAGER_RUNNING
;
4109 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
4111 static void manager_unref_uid_internal(
4116 int (*_clean_ipc
)(uid_t uid
)) {
4122 assert(uid_is_valid(uid
));
4125 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4126 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4128 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4129 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4130 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4131 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4133 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4134 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4136 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4139 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4141 n
= c
& ~DESTROY_IPC_FLAG
;
4145 if (destroy_now
&& n
== 0) {
4146 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
4148 if (c
& DESTROY_IPC_FLAG
) {
4149 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4150 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4152 (void) _clean_ipc(uid
);
4155 c
= n
| (c
& DESTROY_IPC_FLAG
);
4156 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4160 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4161 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4164 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4165 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4168 static int manager_ref_uid_internal(
4179 assert(uid_is_valid(uid
));
4181 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4182 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4184 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4185 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4187 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4190 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4194 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4196 n
= c
& ~DESTROY_IPC_FLAG
;
4199 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4202 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4204 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4207 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4208 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
4211 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4212 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4215 static void manager_vacuum_uid_refs_internal(
4218 int (*_clean_ipc
)(uid_t uid
)) {
4227 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4231 uid
= PTR_TO_UID(k
);
4232 c
= PTR_TO_UINT32(p
);
4234 n
= c
& ~DESTROY_IPC_FLAG
;
4238 if (c
& DESTROY_IPC_FLAG
) {
4239 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4240 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4242 (void) _clean_ipc(uid
);
4245 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
4249 void manager_vacuum_uid_refs(Manager
*m
) {
4250 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
4253 void manager_vacuum_gid_refs(Manager
*m
) {
4254 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
4257 static void manager_serialize_uid_refs_internal(
4261 const char *field_name
) {
4271 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
4272 * of it is better rebuild after a reload/reexec. */
4274 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
4278 uid
= PTR_TO_UID(k
);
4279 c
= PTR_TO_UINT32(p
);
4281 if (!(c
& DESTROY_IPC_FLAG
))
4284 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
4288 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
4289 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
4292 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
4293 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
4296 static void manager_deserialize_uid_refs_one_internal(
4299 const char *value
) {
4309 r
= parse_uid(value
, &uid
);
4310 if (r
< 0 || uid
== 0) {
4311 log_debug("Unable to parse UID reference serialization");
4315 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4321 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4322 if (c
& DESTROY_IPC_FLAG
)
4325 c
|= DESTROY_IPC_FLAG
;
4327 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4329 log_debug("Failed to add UID reference entry");
4334 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
4335 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
4338 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
4339 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
4342 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4346 char unit_name
[UNIT_NAME_MAX
+1];
4349 Manager
*m
= userdata
;
4357 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4358 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4359 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4361 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4363 if (IN_SET(errno
, EINTR
, EAGAIN
))
4366 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4369 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4370 log_warning("Received too short user lookup message, ignoring.");
4374 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4375 log_warning("Received too long user lookup message, ignoring.");
4379 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4380 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4384 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4385 if (memchr(buffer
.unit_name
, 0, n
)) {
4386 log_warning("Received lookup message with embedded NUL character, ignoring.");
4390 buffer
.unit_name
[n
] = 0;
4391 u
= manager_get_unit(m
, buffer
.unit_name
);
4393 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4397 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4399 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4403 char *manager_taint_string(Manager
*m
) {
4404 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4408 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4409 * Only things that are detected at runtime should be tagged
4410 * here. For stuff that is set during compilation, emit a warning
4411 * in the configuration phase. */
4415 buf
= new(char, sizeof("split-usr:"
4419 "overflowuid-not-65534:"
4420 "overflowgid-not-65534:"));
4428 e
= stpcpy(e
, "split-usr:");
4430 if (access("/proc/cgroups", F_OK
) < 0)
4431 e
= stpcpy(e
, "cgroups-missing:");
4433 if (clock_is_localtime(NULL
) > 0)
4434 e
= stpcpy(e
, "local-hwclock:");
4436 r
= readlink_malloc("/var/run", &destination
);
4437 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4438 e
= stpcpy(e
, "var-run-bad:");
4440 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4441 if (r
>= 0 && !streq(overflowuid
, "65534"))
4442 e
= stpcpy(e
, "overflowuid-not-65534:");
4444 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4445 if (r
>= 0 && !streq(overflowgid
, "65534"))
4446 e
= stpcpy(e
, "overflowgid-not-65534:");
4448 /* remove the last ':' */
4455 void manager_ref_console(Manager
*m
) {
4461 void manager_unref_console(Manager
*m
) {
4463 assert(m
->n_on_console
> 0);
4466 if (m
->n_on_console
== 0)
4467 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4470 void manager_override_log_level(Manager
*m
, int level
) {
4471 _cleanup_free_
char *s
= NULL
;
4474 if (!m
->log_level_overridden
) {
4475 m
->original_log_level
= log_get_max_level();
4476 m
->log_level_overridden
= true;
4479 (void) log_level_to_string_alloc(level
, &s
);
4480 log_info("Setting log level to %s.", strna(s
));
4482 log_set_max_level(level
);
4485 void manager_restore_original_log_level(Manager
*m
) {
4486 _cleanup_free_
char *s
= NULL
;
4489 if (!m
->log_level_overridden
)
4492 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4493 log_info("Restoring log level to original (%s).", strna(s
));
4495 log_set_max_level(m
->original_log_level
);
4496 m
->log_level_overridden
= false;
4499 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4502 if (!m
->log_target_overridden
) {
4503 m
->original_log_target
= log_get_target();
4504 m
->log_target_overridden
= true;
4507 log_info("Setting log target to %s.", log_target_to_string(target
));
4508 log_set_target(target
);
4511 void manager_restore_original_log_target(Manager
*m
) {
4514 if (!m
->log_target_overridden
)
4517 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4519 log_set_target(m
->original_log_target
);
4520 m
->log_target_overridden
= false;
4523 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4524 [MANAGER_INITIALIZING
] = "initializing",
4525 [MANAGER_STARTING
] = "starting",
4526 [MANAGER_RUNNING
] = "running",
4527 [MANAGER_DEGRADED
] = "degraded",
4528 [MANAGER_MAINTENANCE
] = "maintenance",
4529 [MANAGER_STOPPING
] = "stopping",
4532 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4534 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4535 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4536 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4537 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4538 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4539 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4540 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4541 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4542 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4543 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4544 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4545 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4546 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4549 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);