1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
22 #include "all-units.h"
23 #include "alloc-util.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
33 #include "creds-util.h"
35 #include "dbus-manager.h"
36 #include "dbus-unit.h"
39 #include "dirent-util.h"
42 #include "exec-util.h"
44 #include "exit-status.h"
47 #include "generator-setup.h"
49 #include "inotify-util.h"
53 #include "load-fragment.h"
54 #include "locale-setup.h"
58 #include "manager-dump.h"
59 #include "manager-serialize.h"
60 #include "memory-util.h"
61 #include "mkdir-label.h"
62 #include "parse-util.h"
63 #include "path-lookup.h"
64 #include "path-util.h"
65 #include "process-util.h"
66 #include "ratelimit.h"
67 #include "rlimit-util.h"
69 #include "selinux-util.h"
70 #include "signal-util.h"
71 #include "socket-util.h"
73 #include "stat-util.h"
74 #include "string-table.h"
75 #include "string-util.h"
78 #include "sysctl-util.h"
79 #include "syslog-util.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "user-util.h"
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
98 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
99 * the queue gets more empty. */
100 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
102 /* How many units and jobs to process of the bus queue before returning to the event loop. */
103 #define MANAGER_BUS_MESSAGE_BUDGET 100U
105 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
111 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
112 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
113 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
114 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
115 static int manager_run_environment_generators(Manager
*m
);
116 static int manager_run_generators(Manager
*m
);
117 static void manager_vacuum(Manager
*m
);
119 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
122 if (MANAGER_IS_USER(m
))
123 /* Let the user manager without a timeout show status quickly, so the system manager can make
124 * use of it, if it wants to. */
125 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
* 2 / 3;
126 else if (show_status_on(m
->show_status
))
127 /* When status is on, just use the usual timeout. */
128 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
;
130 timeout
= JOBS_IN_PROGRESS_QUIET_WAIT_USEC
;
132 return usec_add(now(CLOCK_MONOTONIC
), timeout
);
135 static void manager_watch_jobs_in_progress(Manager
*m
) {
141 /* We do not want to show the cylon animation if the user
142 * needs to confirm service executions otherwise confirmation
143 * messages will be screwed by the cylon animation. */
144 if (!manager_is_confirm_spawn_disabled(m
))
147 if (m
->jobs_in_progress_event_source
)
150 next
= manager_watch_jobs_next_time(m
);
151 r
= sd_event_add_time(
153 &m
->jobs_in_progress_event_source
,
156 manager_dispatch_jobs_in_progress
, m
);
160 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
163 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
165 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
168 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
169 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
173 p
= mempset(p
, ' ', pos
-2);
174 if (log_get_show_color())
175 p
= stpcpy(p
, ANSI_RED
);
179 if (pos
> 0 && pos
<= width
) {
180 if (log_get_show_color())
181 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
185 if (log_get_show_color())
186 p
= stpcpy(p
, ANSI_NORMAL
);
189 if (log_get_show_color())
190 p
= stpcpy(p
, ANSI_RED
);
193 p
= mempset(p
, ' ', width
-1-pos
);
194 if (log_get_show_color())
195 strcpy(p
, ANSI_NORMAL
);
199 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
203 if (m
->show_status
== SHOW_STATUS_AUTO
)
204 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
206 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
207 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
211 static void manager_print_jobs_in_progress(Manager
*m
) {
213 unsigned counter
= 0, print_nr
;
214 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
219 assert(m
->n_running_jobs
> 0);
221 manager_flip_auto_status(m
, true, "delay");
223 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
225 HASHMAP_FOREACH(j
, m
->jobs
)
226 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
229 /* m->n_running_jobs must be consistent with the contents of m->jobs,
230 * so the above loop must have succeeded in finding j. */
231 assert(counter
== print_nr
+ 1);
234 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
236 cylon_pos
= 14 - cylon_pos
;
237 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
239 m
->jobs_in_progress_iteration
++;
241 char job_of_n
[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
242 if (m
->n_running_jobs
> 1)
243 xsprintf(job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
);
245 bool have_timeout
= job_get_timeout(j
, &x
) > 0;
247 /* We want to use enough information for the user to identify previous lines talking about the same
248 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
249 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
250 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
251 * second argument to unit_status_string(). */
252 const char *ident
= unit_status_string(j
->unit
, NULL
);
254 const char *time
= FORMAT_TIMESPAN(now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
255 const char *limit
= have_timeout
? FORMAT_TIMESPAN(x
- j
->begin_usec
, 1*USEC_PER_SEC
) : "no limit";
257 if (m
->status_unit_format
== STATUS_UNIT_FORMAT_DESCRIPTION
)
258 /* When using 'Description', we effectively don't have enough space to show the nested status
259 * without ellipsization, so let's not even try. */
260 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
261 "%sA %s job is running for %s (%s / %s)",
263 job_type_to_string(j
->type
),
267 const char *status_text
= unit_status_text(j
->unit
);
269 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
270 "%sJob %s/%s running (%s / %s)%s%s",
273 job_type_to_string(j
->type
),
275 status_text
? ": " : "",
276 strempty(status_text
));
280 "STATUS=%sUser job %s/%s running (%s / %s)...",
283 job_type_to_string(j
->type
),
285 m
->status_ready
= false;
288 static int have_ask_password(void) {
289 _cleanup_closedir_
DIR *dir
= NULL
;
291 dir
= opendir("/run/systemd/ask-password");
299 FOREACH_DIRENT_ALL(de
, dir
, return -errno
)
300 if (startswith(de
->d_name
, "ask."))
305 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
306 int fd
, uint32_t revents
, void *userdata
) {
307 Manager
*m
= userdata
;
313 m
->have_ask_password
= have_ask_password();
314 if (m
->have_ask_password
< 0)
315 /* Log error but continue. Negative have_ask_password
316 * is treated as unknown status. */
317 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
322 static void manager_close_ask_password(Manager
*m
) {
325 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
326 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
327 m
->have_ask_password
= -EINVAL
;
330 static int manager_check_ask_password(Manager
*m
) {
335 if (!m
->ask_password_event_source
) {
336 assert(m
->ask_password_inotify_fd
< 0);
338 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
340 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
341 if (m
->ask_password_inotify_fd
< 0)
342 return log_error_errno(errno
, "Failed to create inotify object: %m");
344 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
345 "/run/systemd/ask-password",
346 IN_CREATE
|IN_DELETE
|IN_MOVE
);
348 manager_close_ask_password(m
);
352 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
353 m
->ask_password_inotify_fd
, EPOLLIN
,
354 manager_dispatch_ask_password_fd
, m
);
356 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
357 manager_close_ask_password(m
);
361 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
363 /* Queries might have been added meanwhile... */
364 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
365 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
368 return m
->have_ask_password
;
371 static int manager_watch_idle_pipe(Manager
*m
) {
376 if (m
->idle_pipe_event_source
)
379 if (m
->idle_pipe
[2] < 0)
382 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
384 return log_error_errno(r
, "Failed to watch idle pipe: %m");
386 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
391 static void manager_close_idle_pipe(Manager
*m
) {
394 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
396 safe_close_pair(m
->idle_pipe
);
397 safe_close_pair(m
->idle_pipe
+ 2);
400 static int manager_setup_time_change(Manager
*m
) {
405 if (MANAGER_IS_TEST_RUN(m
))
408 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
409 m
->time_change_fd
= safe_close(m
->time_change_fd
);
411 m
->time_change_fd
= time_change_fd();
412 if (m
->time_change_fd
< 0)
413 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
415 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
417 return log_error_errno(r
, "Failed to create time change event source: %m");
419 /* Schedule this slightly earlier than the .timer event sources */
420 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
422 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
424 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
426 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
431 static int manager_read_timezone_stat(Manager
*m
) {
437 /* Read the current stat() data of /etc/localtime so that we detect changes */
438 if (lstat("/etc/localtime", &st
) < 0) {
439 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
440 changed
= m
->etc_localtime_accessible
;
441 m
->etc_localtime_accessible
= false;
445 k
= timespec_load(&st
.st_mtim
);
446 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
448 m
->etc_localtime_mtime
= k
;
449 m
->etc_localtime_accessible
= true;
455 static int manager_setup_timezone_change(Manager
*m
) {
456 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
461 if (MANAGER_IS_TEST_RUN(m
))
464 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
465 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
466 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
467 * went to zero and all fds to it are closed.
469 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
472 * Note that we create the new event source first here, before releasing the old one. This should optimize
473 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
475 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
476 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
478 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
479 * O_CREATE or by rename() */
481 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
482 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
483 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
486 return log_error_errno(r
, "Failed to create timezone change event source: %m");
488 /* Schedule this slightly earlier than the .timer event sources */
489 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
491 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
493 sd_event_source_unref(m
->timezone_change_event_source
);
494 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
499 static int enable_special_signals(Manager
*m
) {
500 _cleanup_close_
int fd
= -1;
504 if (MANAGER_IS_TEST_RUN(m
))
507 /* Enable that we get SIGINT on control-alt-del. In containers
508 * this will fail with EPERM (older) or EINVAL (newer), so
510 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
511 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
513 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
515 /* Support systems without virtual console */
517 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
519 /* Enable that we get SIGWINCH on kbrequest */
520 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
521 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
527 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
529 static int manager_setup_signals(Manager
*m
) {
530 struct sigaction sa
= {
531 .sa_handler
= SIG_DFL
,
532 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
539 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
541 /* We make liberal use of realtime signals here. On
542 * Linux/glibc we have 30 of them (with the exception of Linux
543 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
546 assert_se(sigemptyset(&mask
) == 0);
547 sigset_add_many(&mask
,
548 SIGCHLD
, /* Child died */
549 SIGTERM
, /* Reexecute daemon */
550 SIGHUP
, /* Reload configuration */
551 SIGUSR1
, /* systemd: reconnect to D-Bus */
552 SIGUSR2
, /* systemd: dump status */
553 SIGINT
, /* Kernel sends us this on control-alt-del */
554 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
555 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
557 SIGRTMIN
+0, /* systemd: start default.target */
558 SIGRTMIN
+1, /* systemd: isolate rescue.target */
559 SIGRTMIN
+2, /* systemd: isolate emergency.target */
560 SIGRTMIN
+3, /* systemd: start halt.target */
561 SIGRTMIN
+4, /* systemd: start poweroff.target */
562 SIGRTMIN
+5, /* systemd: start reboot.target */
563 SIGRTMIN
+6, /* systemd: start kexec.target */
565 /* ... space for more special targets ... */
567 SIGRTMIN
+13, /* systemd: Immediate halt */
568 SIGRTMIN
+14, /* systemd: Immediate poweroff */
569 SIGRTMIN
+15, /* systemd: Immediate reboot */
570 SIGRTMIN
+16, /* systemd: Immediate kexec */
572 /* ... space for more immediate system state changes ... */
574 SIGRTMIN
+20, /* systemd: enable status messages */
575 SIGRTMIN
+21, /* systemd: disable status messages */
576 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
577 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
578 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
579 SIGRTMIN
+25, /* systemd: reexecute manager */
581 /* Apparently Linux on hppa had fewer RT signals until v3.18,
582 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
583 * see commit v3.17-7614-g1f25df2eff.
585 * We cannot unconditionally make use of those signals here,
586 * so let's use a runtime check. Since these commands are
587 * accessible by different means and only really a safety
588 * net, the missing functionality on hppa shouldn't matter.
591 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
592 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
593 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
594 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
596 /* ... one free signal here SIGRTMIN+30 ... */
598 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
600 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
601 if (m
->signal_fd
< 0)
604 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
608 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
610 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
611 * notify processing can still figure out to which process/service a message belongs, before we reap the
612 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
613 * status information before detecting that there's no process in a cgroup. */
614 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
618 if (MANAGER_IS_SYSTEM(m
))
619 return enable_special_signals(m
);
624 static char** sanitize_environment(char **l
) {
626 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
630 "CONFIGURATION_DIRECTORY",
631 "CREDENTIALS_DIRECTORY",
653 /* Let's order the environment alphabetically, just to make it pretty */
657 int manager_default_environment(Manager
*m
) {
662 m
->transient_environment
= strv_free(m
->transient_environment
);
664 if (MANAGER_IS_SYSTEM(m
)) {
665 /* The system manager always starts with a clean
666 * environment for its children. It does not import
667 * the kernel's or the parents' exported variables.
669 * The initial passed environment is untouched to keep
670 * /proc/self/environ valid; it is used for tagging
671 * the init process inside containers. */
672 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
673 if (!m
->transient_environment
)
676 /* Import locale variables LC_*= from configuration */
677 (void) locale_setup(&m
->transient_environment
);
679 /* The user manager passes its own environment along to its children, except for $PATH. */
680 m
->transient_environment
= strv_copy(environ
);
681 if (!m
->transient_environment
)
684 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
689 sanitize_environment(m
->transient_environment
);
694 static int manager_setup_prefix(Manager
*m
) {
700 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
701 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
702 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
703 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
704 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
705 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
708 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
709 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
710 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
711 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
712 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
713 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
718 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
721 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
722 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
724 return log_warning_errno(r
, "Failed to lookup %s path: %m",
725 exec_directory_type_to_string(i
));
731 static void manager_free_unit_name_maps(Manager
*m
) {
732 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
733 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
734 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
735 m
->unit_cache_timestamp_hash
= 0;
738 static int manager_setup_run_queue(Manager
*m
) {
742 assert(!m
->run_queue_event_source
);
744 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
748 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
752 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
756 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
761 static int manager_setup_sigchld_event_source(Manager
*m
) {
765 assert(!m
->sigchld_event_source
);
767 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
771 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
775 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
779 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
784 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
785 _cleanup_(manager_freep
) Manager
*m
= NULL
;
790 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
797 .unit_file_scope
= scope
,
798 .objective
= _MANAGER_OBJECTIVE_INVALID
,
800 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
802 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
803 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
804 .default_tasks_accounting
= true,
805 .default_tasks_max
= TASKS_MAX_UNSET
,
806 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
807 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
808 .default_restart_usec
= DEFAULT_RESTART_USEC
,
810 .original_log_level
= -1,
811 .original_log_target
= _LOG_TARGET_INVALID
,
813 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
814 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
815 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
816 .watchdog_overridden
[WATCHDOG_PRETIMEOUT
] = USEC_INFINITY
,
818 .show_status_overridden
= _SHOW_STATUS_INVALID
,
821 .cgroups_agent_fd
= -1,
823 .time_change_fd
= -1,
824 .user_lookup_fds
= { -1, -1 },
825 .private_listen_fd
= -1,
827 .cgroup_inotify_fd
= -1,
828 .pin_cgroupfs_fd
= -1,
829 .ask_password_inotify_fd
= -1,
830 .idle_pipe
= { -1, -1, -1, -1},
832 /* start as id #1, so that we can leave #0 around as "null-like" value */
835 .have_ask_password
= -EINVAL
, /* we don't know */
837 .test_run_flags
= test_run_flags
,
839 .default_oom_policy
= OOM_STOP
,
843 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
844 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
845 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
846 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
849 /* Prepare log fields we can use for structured logging */
850 if (MANAGER_IS_SYSTEM(m
)) {
851 m
->unit_log_field
= "UNIT=";
852 m
->unit_log_format_string
= "UNIT=%s";
854 m
->invocation_log_field
= "INVOCATION_ID=";
855 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
857 m
->unit_log_field
= "USER_UNIT=";
858 m
->unit_log_format_string
= "USER_UNIT=%s";
860 m
->invocation_log_field
= "USER_INVOCATION_ID=";
861 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
864 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
865 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
867 r
= manager_default_environment(m
);
871 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
875 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
879 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
883 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
887 r
= manager_setup_prefix(m
);
891 r
= get_credentials_dir(&e
);
893 m
->received_credentials
= strdup(e
);
894 if (!m
->received_credentials
)
898 r
= sd_event_default(&m
->event
);
902 r
= manager_setup_run_queue(m
);
906 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
907 m
->cgroup_root
= strdup("");
911 r
= manager_setup_signals(m
);
915 r
= manager_setup_cgroup(m
);
919 r
= manager_setup_time_change(m
);
923 r
= manager_read_timezone_stat(m
);
927 (void) manager_setup_timezone_change(m
);
929 r
= manager_setup_sigchld_event_source(m
);
934 if (MANAGER_IS_SYSTEM(m
) && lsm_bpf_supported()) {
935 r
= lsm_bpf_setup(m
);
937 log_warning_errno(r
, "Failed to setup LSM BPF, ignoring: %m");
942 if (test_run_flags
== 0) {
943 if (MANAGER_IS_SYSTEM(m
))
944 r
= mkdir_label("/run/systemd/units", 0755);
946 _cleanup_free_
char *units_path
= NULL
;
947 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
950 r
= mkdir_p_label(units_path
, 0755);
953 if (r
< 0 && r
!= -EEXIST
)
959 dir_is_empty("/usr") > 0;
961 /* Note that we do not set up the notify fd here. We do that after deserialization,
962 * since they might have gotten serialized across the reexec. */
969 static int manager_setup_notify(Manager
*m
) {
972 if (MANAGER_IS_TEST_RUN(m
))
975 if (m
->notify_fd
< 0) {
976 _cleanup_close_
int fd
= -1;
977 union sockaddr_union sa
;
980 /* First free all secondary fields */
981 m
->notify_socket
= mfree(m
->notify_socket
);
982 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
984 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
986 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
988 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
990 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
991 if (!m
->notify_socket
)
994 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
996 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1000 (void) mkdir_parents_label(m
->notify_socket
, 0755);
1001 (void) sockaddr_un_unlink(&sa
.un
);
1003 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
1005 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
1007 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1009 return log_error_errno(r
, "SO_PASSCRED failed: %m");
1011 m
->notify_fd
= TAKE_FD(fd
);
1013 log_debug("Using notification socket %s", m
->notify_socket
);
1016 if (!m
->notify_event_source
) {
1017 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
1019 return log_error_errno(r
, "Failed to allocate notify event source: %m");
1021 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1022 * service an exit message belongs. */
1023 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
1025 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
1027 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
1033 static int manager_setup_cgroups_agent(Manager
*m
) {
1035 static const union sockaddr_union sa
= {
1036 .un
.sun_family
= AF_UNIX
,
1037 .un
.sun_path
= "/run/systemd/cgroups-agent",
1041 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1042 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1043 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1044 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1045 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1046 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1047 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1048 * we thus won't lose messages.
1050 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1051 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1052 * bus for these messages. */
1054 if (MANAGER_IS_TEST_RUN(m
))
1057 if (!MANAGER_IS_SYSTEM(m
))
1060 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1062 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1063 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1066 if (m
->cgroups_agent_fd
< 0) {
1067 _cleanup_close_
int fd
= -1;
1069 /* First free all secondary fields */
1070 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1072 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1074 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1076 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1078 (void) sockaddr_un_unlink(&sa
.un
);
1080 /* Only allow root to connect to this socket */
1081 RUN_WITH_UMASK(0077)
1082 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1084 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1086 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1089 if (!m
->cgroups_agent_event_source
) {
1090 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1092 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1094 /* Process cgroups notifications early. Note that when the agent notification is received
1095 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1096 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1097 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1099 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1101 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1107 static int manager_setup_user_lookup_fd(Manager
*m
) {
1112 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1113 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1114 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1115 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1116 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1117 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1118 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1119 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1122 * You might wonder why we need a communication channel for this that is independent of the usual notification
1123 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1124 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1125 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1127 * Note that this function is called under two circumstances: when we first initialize (in which case we
1128 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1129 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1131 if (m
->user_lookup_fds
[0] < 0) {
1133 /* Free all secondary fields */
1134 safe_close_pair(m
->user_lookup_fds
);
1135 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1137 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1138 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1140 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1143 if (!m
->user_lookup_event_source
) {
1144 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1146 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1148 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1150 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1152 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1154 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1160 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1166 while ((u
= m
->cleanup_queue
)) {
1167 assert(u
->in_cleanup_queue
);
1177 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1178 GC_OFFSET_UNSURE
, /* No clue */
1179 GC_OFFSET_GOOD
, /* We still need this unit */
1180 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1184 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1187 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1189 /* Recursively mark referenced units as GOOD as well */
1190 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1191 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1192 unit_gc_mark_good(other
, gc_marker
);
1195 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1201 if (IN_SET(u
->gc_marker
- gc_marker
,
1202 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1205 if (u
->in_cleanup_queue
)
1208 if (!unit_may_gc(u
))
1211 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1215 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1216 unit_gc_sweep(other
, gc_marker
);
1218 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1221 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1225 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1226 unit_gc_sweep(ref
->source
, gc_marker
);
1228 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1231 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1238 /* We were unable to find anything out about this entry, so
1239 * let's investigate it later */
1240 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1241 unit_add_to_gc_queue(u
);
1245 /* We definitely know that this one is not useful anymore, so
1246 * let's mark it for deletion */
1247 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1248 unit_add_to_cleanup_queue(u
);
1252 unit_gc_mark_good(u
, gc_marker
);
1255 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1256 unsigned n
= 0, gc_marker
;
1261 /* log_debug("Running GC..."); */
1263 m
->gc_marker
+= _GC_OFFSET_MAX
;
1264 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1267 gc_marker
= m
->gc_marker
;
1269 while ((u
= m
->gc_unit_queue
)) {
1270 assert(u
->in_gc_queue
);
1272 unit_gc_sweep(u
, gc_marker
);
1274 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1275 u
->in_gc_queue
= false;
1279 if (IN_SET(u
->gc_marker
- gc_marker
,
1280 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1282 log_unit_debug(u
, "Collecting.");
1283 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1284 unit_add_to_cleanup_queue(u
);
1291 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1297 while ((j
= m
->gc_job_queue
)) {
1298 assert(j
->in_gc_queue
);
1300 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1301 j
->in_gc_queue
= false;
1308 log_unit_debug(j
->unit
, "Collecting job.");
1309 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1315 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1322 while ((u
= m
->stop_when_unneeded_queue
)) {
1323 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1325 assert(u
->in_stop_when_unneeded_queue
);
1326 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1327 u
->in_stop_when_unneeded_queue
= false;
1331 if (!unit_is_unneeded(u
))
1334 log_unit_debug(u
, "Unit is not needed anymore.");
1336 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1337 * service being unnecessary after a while. */
1339 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1340 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1344 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1345 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1347 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1353 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1360 while ((u
= m
->start_when_upheld_queue
)) {
1361 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1362 Unit
*culprit
= NULL
;
1364 assert(u
->in_start_when_upheld_queue
);
1365 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1366 u
->in_start_when_upheld_queue
= false;
1370 if (!unit_is_upheld_by_active(u
, &culprit
))
1373 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1375 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1376 * service being unnecessary after a while. */
1378 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1379 log_unit_warning(u
, "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.", culprit
->id
);
1383 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1385 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1391 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1398 while ((u
= m
->stop_when_bound_queue
)) {
1399 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1400 Unit
*culprit
= NULL
;
1402 assert(u
->in_stop_when_bound_queue
);
1403 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1404 u
->in_stop_when_bound_queue
= false;
1408 if (!unit_is_bound_by_inactive(u
, &culprit
))
1411 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1413 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1414 * service being unnecessary after a while. */
1416 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1417 log_unit_warning(u
, "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.", culprit
->id
);
1421 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1423 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1429 static void manager_clear_jobs_and_units(Manager
*m
) {
1434 while ((u
= hashmap_first(m
->units
)))
1437 manager_dispatch_cleanup_queue(m
);
1439 assert(!m
->load_queue
);
1440 assert(prioq_isempty(m
->run_queue
));
1441 assert(!m
->dbus_unit_queue
);
1442 assert(!m
->dbus_job_queue
);
1443 assert(!m
->cleanup_queue
);
1444 assert(!m
->gc_unit_queue
);
1445 assert(!m
->gc_job_queue
);
1446 assert(!m
->cgroup_realize_queue
);
1447 assert(!m
->cgroup_empty_queue
);
1448 assert(!m
->cgroup_oom_queue
);
1449 assert(!m
->target_deps_queue
);
1450 assert(!m
->stop_when_unneeded_queue
);
1451 assert(!m
->start_when_upheld_queue
);
1452 assert(!m
->stop_when_bound_queue
);
1454 assert(hashmap_isempty(m
->jobs
));
1455 assert(hashmap_isempty(m
->units
));
1457 m
->n_on_console
= 0;
1458 m
->n_running_jobs
= 0;
1459 m
->n_installed_jobs
= 0;
1460 m
->n_failed_jobs
= 0;
1463 Manager
* manager_free(Manager
*m
) {
1467 manager_clear_jobs_and_units(m
);
1469 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1470 if (unit_vtable
[c
]->shutdown
)
1471 unit_vtable
[c
]->shutdown(m
);
1473 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1474 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1476 lookup_paths_flush_generator(&m
->lookup_paths
);
1479 manager_varlink_done(m
);
1481 exec_runtime_vacuum(m
);
1482 hashmap_free(m
->exec_runtime_by_id
);
1484 dynamic_user_vacuum(m
, false);
1485 hashmap_free(m
->dynamic_users
);
1487 hashmap_free(m
->units
);
1488 hashmap_free(m
->units_by_invocation_id
);
1489 hashmap_free(m
->jobs
);
1490 hashmap_free(m
->watch_pids
);
1491 hashmap_free(m
->watch_bus
);
1493 prioq_free(m
->run_queue
);
1495 set_free(m
->startup_units
);
1496 set_free(m
->failed_units
);
1498 sd_event_source_unref(m
->signal_event_source
);
1499 sd_event_source_unref(m
->sigchld_event_source
);
1500 sd_event_source_unref(m
->notify_event_source
);
1501 sd_event_source_unref(m
->cgroups_agent_event_source
);
1502 sd_event_source_unref(m
->time_change_event_source
);
1503 sd_event_source_unref(m
->timezone_change_event_source
);
1504 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1505 sd_event_source_unref(m
->run_queue_event_source
);
1506 sd_event_source_unref(m
->user_lookup_event_source
);
1508 safe_close(m
->signal_fd
);
1509 safe_close(m
->notify_fd
);
1510 safe_close(m
->cgroups_agent_fd
);
1511 safe_close(m
->time_change_fd
);
1512 safe_close_pair(m
->user_lookup_fds
);
1514 manager_close_ask_password(m
);
1516 manager_close_idle_pipe(m
);
1518 sd_event_unref(m
->event
);
1520 free(m
->notify_socket
);
1522 lookup_paths_free(&m
->lookup_paths
);
1523 strv_free(m
->transient_environment
);
1524 strv_free(m
->client_environment
);
1526 hashmap_free(m
->cgroup_unit
);
1527 manager_free_unit_name_maps(m
);
1529 free(m
->switch_root
);
1530 free(m
->switch_root_init
);
1532 rlimit_free_all(m
->rlimit
);
1534 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1535 hashmap_free(m
->units_requiring_mounts_for
);
1537 hashmap_free(m
->uid_refs
);
1538 hashmap_free(m
->gid_refs
);
1540 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1541 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1542 free(m
->received_credentials
);
1544 free(m
->watchdog_pretimeout_governor
);
1545 free(m
->watchdog_pretimeout_governor_overridden
);
1548 lsm_bpf_destroy(m
->restrict_fs
);
1554 static void manager_enumerate_perpetual(Manager
*m
) {
1557 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1560 /* Let's ask every type to load all units from disk/kernel that it might know */
1561 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1562 if (!unit_type_supported(c
)) {
1563 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1567 if (unit_vtable
[c
]->enumerate_perpetual
)
1568 unit_vtable
[c
]->enumerate_perpetual(m
);
1572 static void manager_enumerate(Manager
*m
) {
1575 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1578 /* Let's ask every type to load all units from disk/kernel that it might know */
1579 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1580 if (!unit_type_supported(c
)) {
1581 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1585 if (unit_vtable
[c
]->enumerate
)
1586 unit_vtable
[c
]->enumerate(m
);
1589 manager_dispatch_load_queue(m
);
1592 static void manager_coldplug(Manager
*m
) {
1599 log_debug("Invoking unit coldplug() handlers…");
1601 /* Let's place the units back into their deserialized state */
1602 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1604 /* ignore aliases */
1608 r
= unit_coldplug(u
);
1610 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1614 static void manager_catchup(Manager
*m
) {
1620 log_debug("Invoking unit catchup() handlers…");
1622 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1623 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1625 /* ignore aliases */
1633 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1638 HASHMAP_FOREACH(u
, m
->units
) {
1640 if (fdset_size(fds
) <= 0)
1643 if (!UNIT_VTABLE(u
)->distribute_fds
)
1646 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1650 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1655 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1656 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1657 * rather than the current one. */
1659 if (MANAGER_IS_TEST_RUN(m
))
1662 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1665 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1668 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1671 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1677 static void manager_setup_bus(Manager
*m
) {
1680 /* Let's set up our private bus connection now, unconditionally */
1681 (void) bus_init_private(m
);
1683 /* If we are in --user mode also connect to the system bus now */
1684 if (MANAGER_IS_USER(m
))
1685 (void) bus_init_system(m
);
1687 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1688 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1689 (void) bus_init_api(m
);
1691 if (MANAGER_IS_SYSTEM(m
))
1692 (void) bus_init_system(m
);
1696 static void manager_preset_all(Manager
*m
) {
1701 if (m
->first_boot
<= 0)
1704 if (!MANAGER_IS_SYSTEM(m
))
1707 if (MANAGER_IS_TEST_RUN(m
))
1710 /* If this is the first boot, and we are in the host system, then preset everything */
1711 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1713 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1714 "Failed to populate /etc with preset unit settings, ignoring: %m");
1716 log_info("Populated /etc with preset unit settings.");
1719 static void manager_ready(Manager
*m
) {
1722 /* After having loaded everything, do the final round of catching up with what might have changed */
1724 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1726 /* It might be safe to log to the journal now and connect to dbus */
1727 manager_recheck_journal(m
);
1728 manager_recheck_dbus(m
);
1730 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1733 /* Create a file which will indicate when the manager started loading units the last time. */
1734 if (MANAGER_IS_SYSTEM(m
))
1735 (void) touch_file("/run/systemd/systemd-units-load", false,
1736 m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].realtime
?: now(CLOCK_REALTIME
),
1737 UID_INVALID
, GID_INVALID
, 0444);
1739 m
->honor_device_enumeration
= true;
1742 Manager
* manager_reloading_start(Manager
*m
) {
1744 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD
);
1748 void manager_reloading_stopp(Manager
**m
) {
1750 assert((*m
)->n_reloading
> 0);
1751 (*m
)->n_reloading
--;
1755 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
, const char *root
) {
1760 /* If we are running in test mode, we still want to run the generators,
1761 * but we should not touch the real generator directories. */
1762 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1763 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1766 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1768 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1769 r
= manager_run_environment_generators(m
);
1771 r
= manager_run_generators(m
);
1772 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1776 manager_preset_all(m
);
1778 lookup_paths_log(&m
->lookup_paths
);
1781 /* This block is (optionally) done with the reloading counter bumped */
1782 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1784 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1785 * counter here already */
1787 reloading
= manager_reloading_start(m
);
1789 /* First, enumerate what we can from all config files */
1790 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1791 manager_enumerate_perpetual(m
);
1792 manager_enumerate(m
);
1793 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1795 /* Second, deserialize if there is something to deserialize */
1796 if (serialization
) {
1797 r
= manager_deserialize(m
, serialization
, fds
);
1799 return log_error_errno(r
, "Deserialization failed: %m");
1802 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1803 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1805 manager_distribute_fds(m
, fds
);
1807 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1808 r
= manager_setup_notify(m
);
1810 /* No sense to continue without notifications, our children would fail anyway. */
1813 r
= manager_setup_cgroups_agent(m
);
1815 /* Likewise, no sense to continue without empty cgroup notifications. */
1818 r
= manager_setup_user_lookup_fd(m
);
1820 /* This shouldn't fail, except if things are really broken. */
1823 /* Connect to the bus if we are good for it */
1824 manager_setup_bus(m
);
1826 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1827 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1829 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1830 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1832 r
= manager_varlink_init(m
);
1834 log_warning_errno(r
, "Failed to set up Varlink, ignoring: %m");
1836 /* Third, fire things up! */
1837 manager_coldplug(m
);
1839 /* Clean up runtime objects */
1843 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1844 * reload is finished */
1845 m
->send_reloading_done
= true;
1853 int manager_add_job(
1859 sd_bus_error
*error
,
1866 assert(type
< _JOB_TYPE_MAX
);
1868 assert(mode
< _JOB_MODE_MAX
);
1870 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1871 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1873 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1874 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1876 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1877 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1879 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1881 type
= job_type_collapse(type
, unit
);
1883 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1887 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1888 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1889 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1893 if (mode
== JOB_ISOLATE
) {
1894 r
= transaction_add_isolate_jobs(tr
, m
);
1899 if (mode
== JOB_TRIGGERING
) {
1900 r
= transaction_add_triggering_jobs(tr
, unit
);
1905 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1909 log_unit_debug(unit
,
1910 "Enqueued job %s/%s as %u", unit
->id
,
1911 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1914 *ret
= tr
->anchor_job
;
1916 transaction_free(tr
);
1920 transaction_abort(tr
);
1921 transaction_free(tr
);
1925 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1926 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1930 assert(type
< _JOB_TYPE_MAX
);
1932 assert(mode
< _JOB_MODE_MAX
);
1934 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1939 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1942 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1943 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1947 assert(type
< _JOB_TYPE_MAX
);
1949 assert(mode
< _JOB_MODE_MAX
);
1951 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1953 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1958 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1964 assert(mode
< _JOB_MODE_MAX
);
1965 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1967 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1971 /* We need an anchor job */
1972 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1976 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1977 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1979 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1983 transaction_free(tr
);
1987 transaction_abort(tr
);
1988 transaction_free(tr
);
1992 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1995 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1998 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
2002 return hashmap_get(m
->units
, name
);
2005 static int manager_dispatch_target_deps_queue(Manager
*m
) {
2011 while ((u
= m
->target_deps_queue
)) {
2012 _cleanup_free_ Unit
**targets
= NULL
;
2015 assert(u
->in_target_deps_queue
);
2017 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
2018 u
->in_target_deps_queue
= false;
2020 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2021 * dependencies, and we can't have it that hash tables we iterate through are modified while
2022 * we are iterating through them. */
2023 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
2027 for (int i
= 0; i
< n_targets
; i
++) {
2028 r
= unit_add_default_target_dependency(u
, targets
[i
]);
2037 unsigned manager_dispatch_load_queue(Manager
*m
) {
2043 /* Make sure we are not run recursively */
2044 if (m
->dispatching_load_queue
)
2047 m
->dispatching_load_queue
= true;
2049 /* Dispatches the load queue. Takes a unit from the queue and
2050 * tries to load its data until the queue is empty */
2052 while ((u
= m
->load_queue
)) {
2053 assert(u
->in_load_queue
);
2059 m
->dispatching_load_queue
= false;
2061 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2062 * should be loaded and have aliases resolved */
2063 (void) manager_dispatch_target_deps_queue(m
);
2068 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2071 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2072 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2073 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2074 if (u
->load_state
!= UNIT_NOT_FOUND
)
2077 /* The cache has been updated since the last time we tried to load the unit. There might be new
2078 * fragment paths to read. */
2079 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2082 /* The cache needs to be updated because there are modifications on disk. */
2083 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2086 int manager_load_unit_prepare(
2093 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
2101 /* This will prepare the unit for loading, but not actually load anything from disk. */
2103 if (path
&& !path_is_absolute(path
))
2104 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2107 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2108 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2109 * but this cannot be possible in any code path (See #6119). */
2111 name
= basename(path
);
2114 t
= unit_name_to_type(name
);
2116 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2117 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2118 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2120 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2123 ret
= manager_get_unit(m
, name
);
2125 /* The time-based cache allows to start new units without daemon-reload,
2126 * but if they are already referenced (because of dependencies or ordering)
2127 * then we have to force a load of the fragment. As an optimization, check
2128 * first if anything in the usual paths was modified since the last time
2129 * the cache was loaded. Also check if the last time an attempt to load the
2130 * unit was made was before the most recent cache refresh, so that we know
2131 * we need to try again — even if the cache is current, it might have been
2132 * updated in a different context before we had a chance to retry loading
2133 * this particular unit. */
2134 if (manager_unit_cache_should_retry_load(ret
))
2135 ret
->load_state
= UNIT_STUB
;
2141 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2147 r
= free_and_strdup(&ret
->fragment_path
, path
);
2152 r
= unit_add_name(ret
, name
);
2156 unit_add_to_load_queue(ret
);
2157 unit_add_to_dbus_queue(ret
);
2158 unit_add_to_gc_queue(ret
);
2166 int manager_load_unit(
2178 /* This will load the service information files, but not actually
2179 * start any services or anything. */
2181 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2185 manager_dispatch_load_queue(m
);
2187 *_ret
= unit_follow_merge(*_ret
);
2191 int manager_load_startable_unit_or_warn(
2197 /* Load a unit, make sure it loaded fully and is not masked. */
2199 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2203 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2205 return log_error_errno(r
, "Failed to load %s %s: %s",
2206 name
? "unit" : "unit file", name
?: path
,
2207 bus_error_message(&error
, r
));
2209 r
= bus_unit_validate_load_state(unit
, &error
);
2211 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2217 void manager_clear_jobs(Manager
*m
) {
2222 while ((j
= hashmap_first(m
->jobs
)))
2223 /* No need to recurse. We're cancelling all jobs. */
2224 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2227 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2230 /* First let's drop the unit keyed as "pid". */
2231 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2233 /* Then, let's also drop the array keyed by -pid. */
2234 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2237 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2238 Manager
*m
= userdata
;
2244 while ((j
= prioq_peek(m
->run_queue
))) {
2245 assert(j
->installed
);
2246 assert(j
->in_run_queue
);
2248 (void) job_run_and_invalidate(j
);
2251 if (m
->n_running_jobs
> 0)
2252 manager_watch_jobs_in_progress(m
);
2254 if (m
->n_on_console
> 0)
2255 manager_watch_idle_pipe(m
);
2260 void manager_trigger_run_queue(Manager
*m
) {
2265 r
= sd_event_source_set_enabled(
2266 m
->run_queue_event_source
,
2267 prioq_isempty(m
->run_queue
) ? SD_EVENT_OFF
: SD_EVENT_ONESHOT
);
2269 log_warning_errno(r
, "Failed to enable job run queue event source, ignoring: %m");
2272 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2273 unsigned n
= 0, budget
;
2279 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2280 * as we can. There's no point in throttling generation of signals in that case. */
2281 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2282 budget
= UINT_MAX
; /* infinite budget in this case */
2284 /* Anything to do at all? */
2285 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2288 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2289 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2290 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2293 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2294 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2295 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2296 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2297 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2298 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2299 * connections it will be counted five times. This difference in counting ("references"
2300 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2301 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2302 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2303 * currently chosen much higher than the "budget". */
2304 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2307 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2309 assert(u
->in_dbus_queue
);
2311 bus_unit_send_change_signal(u
);
2314 if (budget
!= UINT_MAX
)
2318 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2319 assert(j
->in_dbus_queue
);
2321 bus_job_send_change_signal(j
);
2324 if (budget
!= UINT_MAX
)
2328 if (m
->send_reloading_done
) {
2329 m
->send_reloading_done
= false;
2330 bus_manager_send_reloading(m
, false);
2334 if (m
->pending_reload_message
) {
2335 bus_send_pending_reload_message(m
);
2342 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2343 Manager
*m
= userdata
;
2347 n
= recv(fd
, buf
, sizeof(buf
), 0);
2349 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2351 log_error("Got zero-length cgroups agent message, ignoring.");
2354 if ((size_t) n
>= sizeof(buf
)) {
2355 log_error("Got overly long cgroups agent message, ignoring.");
2359 if (memchr(buf
, 0, n
)) {
2360 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2365 manager_notify_cgroup_empty(m
, buf
);
2366 (void) bus_forward_agent_released(m
, buf
);
2371 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2373 /* nothing else must be sent when using BARRIER=1 */
2374 if (strv_contains(tags
, "BARRIER=1")) {
2375 if (strv_length(tags
) == 1) {
2376 if (fdset_size(fds
) != 1)
2377 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2379 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2381 /* Drop the message if BARRIER=1 was found */
2388 static void manager_invoke_notify_message(
2391 const struct ucred
*ucred
,
2400 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2402 u
->notifygen
= m
->notifygen
;
2404 if (UNIT_VTABLE(u
)->notify_message
)
2405 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2407 else if (DEBUG_LOGGING
) {
2408 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2410 buf
= strv_join(tags
, ", ");
2412 x
= ellipsize(buf
, 20, 90);
2416 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2420 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2422 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2423 Manager
*m
= userdata
;
2424 char buf
[NOTIFY_BUFFER_MAX
+1];
2425 struct iovec iovec
= {
2427 .iov_len
= sizeof(buf
)-1,
2429 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2430 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2431 struct msghdr msghdr
= {
2434 .msg_control
= &control
,
2435 .msg_controllen
= sizeof(control
),
2438 struct cmsghdr
*cmsg
;
2439 struct ucred
*ucred
= NULL
;
2440 _cleanup_free_ Unit
**array_copy
= NULL
;
2441 _cleanup_strv_free_
char **tags
= NULL
;
2442 Unit
*u1
, *u2
, **array
;
2443 int r
, *fd_array
= NULL
;
2449 assert(m
->notify_fd
== fd
);
2451 if (revents
!= EPOLLIN
) {
2452 log_warning("Got unexpected poll event for notify fd.");
2456 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2458 if (ERRNO_IS_TRANSIENT(n
))
2459 return 0; /* Spurious wakeup, try again */
2461 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2464 /* If this is any other, real error, then let's stop processing this socket. This of course
2465 * means we won't take notification messages anymore, but that's still better than busy
2466 * looping around this: being woken up over and over again but being unable to actually read
2467 * the message off the socket. */
2468 return log_error_errno(n
, "Failed to receive notification message: %m");
2471 CMSG_FOREACH(cmsg
, &msghdr
) {
2472 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2475 fd_array
= (int*) CMSG_DATA(cmsg
);
2476 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2478 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2479 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2480 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2483 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2490 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2492 close_many(fd_array
, n_fds
);
2498 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2499 log_warning("Received notify message without valid credentials. Ignoring.");
2503 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2504 log_warning("Received notify message exceeded maximum size. Ignoring.");
2508 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2509 * trailing NUL byte in the message, but don't expect it. */
2510 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2511 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2515 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2517 tags
= strv_split_newlines(buf
);
2523 /* possibly a barrier fd, let's see */
2524 if (manager_process_barrier_fd(tags
, fds
))
2527 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2530 /* Notify every unit that might be interested, which might be multiple. */
2531 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2532 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2533 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2540 array_copy
= newdup(Unit
*, array
, k
+1);
2544 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2545 * make sure we only invoke each unit's handler once. */
2547 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2551 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2555 for (size_t i
= 0; array_copy
[i
]; i
++) {
2556 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2561 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2563 if (fdset_size(fds
) > 0)
2564 log_warning("Got extra auxiliary fds with notification message, closing them.");
2569 static void manager_invoke_sigchld_event(
2572 const siginfo_t
*si
) {
2578 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2579 if (u
->sigchldgen
== m
->sigchldgen
)
2581 u
->sigchldgen
= m
->sigchldgen
;
2583 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2584 unit_unwatch_pid(u
, si
->si_pid
);
2586 if (UNIT_VTABLE(u
)->sigchld_event
)
2587 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2590 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2591 Manager
*m
= userdata
;
2598 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2599 * while it is a zombie. */
2601 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2603 if (errno
!= ECHILD
)
2604 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2612 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2613 _cleanup_free_ Unit
**array_copy
= NULL
;
2614 _cleanup_free_
char *name
= NULL
;
2615 Unit
*u1
, *u2
, **array
;
2617 (void) get_process_comm(si
.si_pid
, &name
);
2619 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2620 si
.si_pid
, strna(name
),
2621 sigchld_code_to_string(si
.si_code
),
2623 strna(si
.si_code
== CLD_EXITED
2624 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2625 : signal_to_string(si
.si_status
)));
2627 /* Increase the generation counter used for filtering out duplicate unit invocations */
2630 /* And now figure out the unit this belongs to, it might be multiple... */
2631 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2632 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2633 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2637 /* Count how many entries the array has */
2641 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2642 array_copy
= newdup(Unit
*, array
, n
+1);
2647 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2648 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2649 * each iteration. */
2651 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2652 * We only do this for the cgroup the PID belonged to. */
2653 (void) unit_check_oom(u1
);
2655 /* This only logs for now. In the future when the interface for kills/notifications
2656 * is more stable we can extend service results table similar to how kernel oom kills
2658 (void) unit_check_oomd_kill(u1
);
2660 manager_invoke_sigchld_event(m
, u1
, &si
);
2663 manager_invoke_sigchld_event(m
, u2
, &si
);
2665 for (size_t i
= 0; array_copy
[i
]; i
++)
2666 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2669 /* And now, we actually reap the zombie. */
2670 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2671 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2678 /* All children processed for now, turn off event source */
2680 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2682 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2687 static void manager_start_special(Manager
*m
, const char *name
, JobMode mode
) {
2690 if (manager_add_job_by_name_and_warn(m
, JOB_START
, name
, mode
, NULL
, &job
) < 0)
2693 const char *s
= unit_status_string(job
->unit
, NULL
);
2695 log_info("Activating special unit %s...", s
);
2698 "STATUS=Activating special unit %s...", s
);
2699 m
->status_ready
= false;
2702 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2703 /* If the user presses C-A-D more than
2704 * 7 times within 2s, we reboot/shutdown immediately,
2705 * unless it was disabled in system.conf */
2707 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2708 manager_start_special(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2710 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2711 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2714 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2715 Manager
*m
= userdata
;
2717 struct signalfd_siginfo sfsi
;
2721 assert(m
->signal_fd
== fd
);
2723 if (revents
!= EPOLLIN
) {
2724 log_warning("Got unexpected events from signal file descriptor.");
2728 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2730 if (ERRNO_IS_TRANSIENT(errno
))
2733 /* We return an error here, which will kill this handler,
2734 * to avoid a busy loop on read error. */
2735 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2737 if (n
!= sizeof(sfsi
)) {
2738 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2742 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2743 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2744 ? LOG_DEBUG
: LOG_INFO
,
2747 switch (sfsi
.ssi_signo
) {
2750 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2752 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2757 if (MANAGER_IS_SYSTEM(m
)) {
2758 /* This is for compatibility with the original sysvinit */
2759 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2762 m
->objective
= MANAGER_REEXECUTE
;
2768 if (MANAGER_IS_SYSTEM(m
))
2769 manager_handle_ctrl_alt_del(m
);
2771 manager_start_special(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2775 /* This is a nop on non-init */
2776 if (MANAGER_IS_SYSTEM(m
))
2777 manager_start_special(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2782 /* This is a nop on non-init */
2783 if (MANAGER_IS_SYSTEM(m
))
2784 manager_start_special(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2789 if (manager_dbus_is_running(m
, false)) {
2790 log_info("Trying to reconnect to bus...");
2792 (void) bus_init_api(m
);
2794 if (MANAGER_IS_SYSTEM(m
))
2795 (void) bus_init_system(m
);
2797 manager_start_special(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2802 _cleanup_free_
char *dump
= NULL
;
2804 r
= manager_get_dump_string(m
, &dump
);
2806 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2810 log_dump(LOG_INFO
, dump
);
2815 if (verify_run_space_and_log("Refusing to reload") < 0)
2818 m
->objective
= MANAGER_RELOAD
;
2823 /* Starting SIGRTMIN+0 */
2824 static const struct {
2827 } target_table
[] = {
2828 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2829 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2830 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2831 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2832 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2833 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2834 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2837 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2838 static const ManagerObjective objective_table
[] = {
2840 [1] = MANAGER_POWEROFF
,
2841 [2] = MANAGER_REBOOT
,
2842 [3] = MANAGER_KEXEC
,
2845 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2846 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2847 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2848 manager_start_special(m
, target_table
[idx
].target
, target_table
[idx
].mode
);
2852 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2853 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2854 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2858 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2861 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2865 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2869 manager_override_log_level(m
, LOG_DEBUG
);
2873 manager_restore_original_log_level(m
);
2877 if (MANAGER_IS_USER(m
)) {
2878 m
->objective
= MANAGER_EXIT
;
2882 /* This is a nop on init */
2886 m
->objective
= MANAGER_REEXECUTE
;
2890 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2891 manager_restore_original_log_target(m
);
2895 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2899 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2903 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2910 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2911 Manager
*m
= userdata
;
2915 assert(m
->time_change_fd
== fd
);
2917 log_struct(LOG_DEBUG
,
2918 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2919 LOG_MESSAGE("Time has been changed"));
2921 /* Restart the watch */
2922 (void) manager_setup_time_change(m
);
2924 HASHMAP_FOREACH(u
, m
->units
)
2925 if (UNIT_VTABLE(u
)->time_change
)
2926 UNIT_VTABLE(u
)->time_change(u
);
2931 static int manager_dispatch_timezone_change(
2932 sd_event_source
*source
,
2933 const struct inotify_event
*e
,
2936 Manager
*m
= userdata
;
2942 log_debug("inotify event for /etc/localtime");
2944 changed
= manager_read_timezone_stat(m
);
2948 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2949 (void) manager_setup_timezone_change(m
);
2951 /* Read the new timezone */
2954 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2956 HASHMAP_FOREACH(u
, m
->units
)
2957 if (UNIT_VTABLE(u
)->timezone_change
)
2958 UNIT_VTABLE(u
)->timezone_change(u
);
2963 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2964 Manager
*m
= userdata
;
2967 assert(m
->idle_pipe
[2] == fd
);
2969 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2970 * now turn off any further console output if there's at least one service that needs console access, so that
2971 * from now on our own output should not spill into that service's output anymore. After all, we support
2972 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2973 * exclusively without our interference. */
2974 m
->no_console_output
= m
->n_on_console
> 0;
2976 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2977 * by closing the pipes towards them, which is what they are waiting for. */
2978 manager_close_idle_pipe(m
);
2983 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2984 Manager
*m
= userdata
;
2990 manager_print_jobs_in_progress(m
);
2992 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2996 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2999 int manager_loop(Manager
*m
) {
3000 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
3004 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
3006 manager_check_finished(m
);
3008 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3009 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
3011 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
3013 while (m
->objective
== MANAGER_OK
) {
3015 (void) watchdog_ping();
3017 if (!ratelimit_below(&rl
)) {
3018 /* Yay, something is going seriously wrong, pause a little */
3019 log_warning("Looping too fast. Throttling execution a little.");
3023 if (manager_dispatch_load_queue(m
) > 0)
3026 if (manager_dispatch_gc_job_queue(m
) > 0)
3029 if (manager_dispatch_gc_unit_queue(m
) > 0)
3032 if (manager_dispatch_cleanup_queue(m
) > 0)
3035 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3038 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3041 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3044 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3047 if (manager_dispatch_dbus_queue(m
) > 0)
3050 /* Sleep for watchdog runtime wait time */
3051 r
= sd_event_run(m
->event
, watchdog_runtime_wait());
3053 return log_error_errno(r
, "Failed to run event loop: %m");
3056 return m
->objective
;
3059 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3060 _cleanup_free_
char *n
= NULL
;
3061 sd_id128_t invocation_id
;
3069 r
= unit_name_from_dbus_path(s
, &n
);
3073 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3074 * as invocation ID. */
3075 r
= sd_id128_from_string(n
, &invocation_id
);
3077 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3083 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3084 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3085 SD_ID128_FORMAT_VAL(invocation_id
));
3088 /* If this didn't work, we check if this is a unit name */
3089 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3090 _cleanup_free_
char *nn
= NULL
;
3093 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3094 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3097 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3105 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3115 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3119 r
= safe_atou(p
, &id
);
3123 j
= manager_get_job(m
, id
);
3132 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3135 _cleanup_free_
char *p
= NULL
;
3139 if (!MANAGER_IS_SYSTEM(m
))
3142 audit_fd
= get_audit_fd();
3146 /* Don't generate audit events if the service was already
3147 * started and we're just deserializing */
3148 if (MANAGER_IS_RELOADING(m
))
3151 if (u
->type
!= UNIT_SERVICE
)
3154 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3156 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3160 msg
= strjoina("unit=", p
);
3161 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3163 /* We aren't allowed to send audit messages?
3164 * Then let's not retry again. */
3167 log_warning_errno(errno
, "Failed to send audit message: %m");
3173 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3174 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3175 _cleanup_free_
char *message
= NULL
;
3176 _cleanup_close_
int fd
= -1;
3179 /* Don't generate plymouth events if the service was already
3180 * started and we're just deserializing */
3181 if (MANAGER_IS_RELOADING(m
))
3184 if (!MANAGER_IS_SYSTEM(m
))
3187 if (detect_container() > 0)
3190 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3193 /* We set SOCK_NONBLOCK here so that we rather drop the
3194 * message then wait for plymouth */
3195 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3197 log_error_errno(errno
, "socket() failed: %m");
3201 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3202 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3203 log_error_errno(errno
, "connect() failed: %m");
3207 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0)
3208 return (void) log_oom();
3211 if (write(fd
, message
, n
+ 1) != n
+ 1)
3212 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3213 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3216 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3219 if (MANAGER_IS_USER(m
))
3220 return USEC_INFINITY
;
3222 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3223 return m
->watchdog_overridden
[t
];
3225 return m
->watchdog
[t
];
3228 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3232 if (MANAGER_IS_USER(m
))
3235 if (m
->watchdog
[t
] == timeout
)
3238 if (t
== WATCHDOG_RUNTIME
) {
3239 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3240 (void) watchdog_setup(timeout
);
3241 } else if (t
== WATCHDOG_PRETIMEOUT
)
3242 if (m
->watchdog_overridden
[WATCHDOG_PRETIMEOUT
] == USEC_INFINITY
)
3243 (void) watchdog_setup_pretimeout(timeout
);
3245 m
->watchdog
[t
] = timeout
;
3248 void manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3252 if (MANAGER_IS_USER(m
))
3255 if (m
->watchdog_overridden
[t
] == timeout
)
3258 if (t
== WATCHDOG_RUNTIME
) {
3259 usec_t usec
= timestamp_is_set(timeout
) ? timeout
: m
->watchdog
[t
];
3261 (void) watchdog_setup(usec
);
3262 } else if (t
== WATCHDOG_PRETIMEOUT
)
3263 (void) watchdog_setup_pretimeout(timeout
);
3265 m
->watchdog_overridden
[t
] = timeout
;
3268 int manager_set_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3269 _cleanup_free_
char *p
= NULL
;
3274 if (MANAGER_IS_USER(m
))
3277 if (streq_ptr(m
->watchdog_pretimeout_governor
, governor
))
3280 p
= strdup(governor
);
3284 r
= watchdog_setup_pretimeout_governor(governor
);
3288 return free_and_replace(m
->watchdog_pretimeout_governor
, p
);
3291 int manager_override_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3292 _cleanup_free_
char *p
= NULL
;
3297 if (MANAGER_IS_USER(m
))
3300 if (streq_ptr(m
->watchdog_pretimeout_governor_overridden
, governor
))
3303 p
= strdup(governor
);
3307 r
= watchdog_setup_pretimeout_governor(governor
);
3311 return free_and_replace(m
->watchdog_pretimeout_governor_overridden
, p
);
3314 int manager_reload(Manager
*m
) {
3315 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3316 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3317 _cleanup_fclose_
FILE *f
= NULL
;
3322 r
= manager_open_serialization(m
, &f
);
3324 return log_error_errno(r
, "Failed to create serialization file: %m");
3330 /* We are officially in reload mode from here on. */
3331 reloading
= manager_reloading_start(m
);
3333 r
= manager_serialize(m
, f
, fds
, false);
3337 if (fseeko(f
, 0, SEEK_SET
) < 0)
3338 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3340 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3343 bus_manager_send_reloading(m
, true);
3345 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3346 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3349 manager_clear_jobs_and_units(m
);
3350 lookup_paths_flush_generator(&m
->lookup_paths
);
3351 lookup_paths_free(&m
->lookup_paths
);
3352 exec_runtime_vacuum(m
);
3353 dynamic_user_vacuum(m
, false);
3354 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3355 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3357 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3359 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3361 (void) manager_run_environment_generators(m
);
3362 (void) manager_run_generators(m
);
3364 lookup_paths_log(&m
->lookup_paths
);
3366 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3367 manager_free_unit_name_maps(m
);
3369 /* First, enumerate what we can from kernel and suchlike */
3370 manager_enumerate_perpetual(m
);
3371 manager_enumerate(m
);
3373 /* Second, deserialize our stored data */
3374 r
= manager_deserialize(m
, f
, fds
);
3376 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3378 /* We don't need the serialization anymore */
3381 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3382 (void) manager_setup_notify(m
);
3383 (void) manager_setup_cgroups_agent(m
);
3384 (void) manager_setup_user_lookup_fd(m
);
3386 /* Third, fire things up! */
3387 manager_coldplug(m
);
3389 /* Clean up runtime objects no longer referenced */
3392 /* Clean up deserialized tracked clients */
3393 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3395 /* Consider the reload process complete now. */
3396 assert(m
->n_reloading
> 0);
3399 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3400 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3401 * let's always set the flag here for safety. */
3402 m
->honor_device_enumeration
= true;
3406 m
->send_reloading_done
= true;
3410 void manager_reset_failed(Manager
*m
) {
3415 HASHMAP_FOREACH(u
, m
->units
)
3416 unit_reset_failed(u
);
3419 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3425 /* Returns true if the unit is inactive or going down */
3426 u
= manager_get_unit(m
, name
);
3430 return unit_inactive_or_pending(u
);
3433 static void log_taint_string(Manager
*m
) {
3434 _cleanup_free_
char *taint
= NULL
;
3438 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3441 m
->taint_logged
= true; /* only check for taint once */
3443 taint
= manager_taint_string(m
);
3447 log_struct(LOG_NOTICE
,
3448 LOG_MESSAGE("System is tainted: %s", taint
),
3450 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3453 static void manager_notify_finished(Manager
*m
) {
3454 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3456 if (MANAGER_IS_TEST_RUN(m
))
3459 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3460 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3463 size_t size
= sizeof buf
;
3465 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3466 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3467 * negative values. */
3469 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3470 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3471 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3472 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3474 if (firmware_usec
> 0)
3475 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3476 if (loader_usec
> 0)
3477 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3479 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3481 /* The initrd case on bare-metal */
3482 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3483 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3485 log_struct(LOG_INFO
,
3486 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3487 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3488 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3489 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3490 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3492 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3493 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3494 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3495 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3497 /* The initrd-less case on bare-metal */
3499 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3502 log_struct(LOG_INFO
,
3503 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3504 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3505 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3506 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3508 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3509 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3510 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3513 /* The container and --user case */
3514 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3515 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3517 log_struct(LOG_INFO
,
3518 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3519 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3520 LOG_MESSAGE("Startup finished in %s.",
3521 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3524 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3526 log_taint_string(m
);
3529 static void user_manager_send_ready(Manager
*m
) {
3534 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3535 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3538 r
= sd_notify(false,
3540 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3542 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
3544 m
->ready_sent
= true;
3545 m
->status_ready
= false;
3548 static void manager_send_ready(Manager
*m
) {
3551 if (m
->ready_sent
&& m
->status_ready
)
3552 /* Skip the notification if nothing changed. */
3555 r
= sd_notify(false,
3559 log_full_errno(m
->ready_sent
? LOG_DEBUG
: LOG_WARNING
, r
,
3560 "Failed to send readiness notification, ignoring: %m");
3562 m
->ready_sent
= m
->status_ready
= true;
3565 static void manager_check_basic_target(Manager
*m
) {
3570 /* Small shortcut */
3571 if (m
->ready_sent
&& m
->taint_logged
)
3574 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3575 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3578 /* For user managers, send out READY=1 as soon as we reach basic.target */
3579 user_manager_send_ready(m
);
3581 /* Log the taint string as soon as we reach basic.target */
3582 log_taint_string(m
);
3585 void manager_check_finished(Manager
*m
) {
3588 if (MANAGER_IS_RELOADING(m
))
3591 /* Verify that we have entered the event loop already, and not left it again. */
3592 if (!MANAGER_IS_RUNNING(m
))
3595 manager_check_basic_target(m
);
3597 if (hashmap_size(m
->jobs
) > 0) {
3598 if (m
->jobs_in_progress_event_source
)
3599 /* Ignore any failure, this is only for feedback */
3600 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
3601 manager_watch_jobs_next_time(m
));
3605 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3606 kill the hashmap if it is relatively large. */
3607 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
3608 m
->jobs
= hashmap_free(m
->jobs
);
3610 manager_send_ready(m
);
3612 /* Notify Type=idle units that we are done now */
3613 manager_close_idle_pipe(m
);
3615 if (MANAGER_IS_FINISHED(m
))
3618 manager_flip_auto_status(m
, false, "boot finished");
3620 /* Turn off confirm spawn now */
3621 m
->confirm_spawn
= NULL
;
3623 /* No need to update ask password status when we're going non-interactive */
3624 manager_close_ask_password(m
);
3626 /* This is no longer the first boot */
3627 manager_set_first_boot(m
, false);
3629 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3631 manager_notify_finished(m
);
3633 manager_invalidate_startup_units(m
);
3636 static bool generator_path_any(const char* const* paths
) {
3639 /* Optimize by skipping the whole process by not creating output directories
3640 * if no generators are found. */
3641 STRV_FOREACH(path
, (char**) paths
)
3642 if (access(*path
, F_OK
) == 0)
3644 else if (errno
!= ENOENT
)
3645 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3650 static int manager_run_environment_generators(Manager
*m
) {
3651 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3652 _cleanup_strv_free_
char **paths
= NULL
;
3654 [STDOUT_GENERATE
] = &tmp
,
3655 [STDOUT_COLLECT
] = &tmp
,
3656 [STDOUT_CONSUME
] = &m
->transient_environment
,
3660 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3663 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
3667 if (!generator_path_any((const char* const*) paths
))
3670 RUN_WITH_UMASK(0022)
3671 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
3672 args
, NULL
, m
->transient_environment
,
3673 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3677 static int manager_run_generators(Manager
*m
) {
3678 _cleanup_strv_free_
char **paths
= NULL
;
3679 const char *argv
[5];
3684 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3687 paths
= generator_binary_paths(m
->unit_file_scope
);
3691 if (!generator_path_any((const char* const*) paths
))
3694 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3696 log_error_errno(r
, "Failed to create generator directories: %m");
3700 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3701 argv
[1] = m
->lookup_paths
.generator
;
3702 argv
[2] = m
->lookup_paths
.generator_early
;
3703 argv
[3] = m
->lookup_paths
.generator_late
;
3706 RUN_WITH_UMASK(0022)
3707 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
3708 (char**) argv
, m
->transient_environment
,
3709 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3714 lookup_paths_trim_generator(&m
->lookup_paths
);
3718 int manager_transient_environment_add(Manager
*m
, char **plus
) {
3723 if (strv_isempty(plus
))
3726 a
= strv_env_merge(m
->transient_environment
, plus
);
3730 sanitize_environment(a
);
3732 return strv_free_and_replace(m
->transient_environment
, a
);
3735 int manager_client_environment_modify(
3740 char **a
= NULL
, **b
= NULL
, **l
;
3744 if (strv_isempty(minus
) && strv_isempty(plus
))
3747 l
= m
->client_environment
;
3749 if (!strv_isempty(minus
)) {
3750 a
= strv_env_delete(l
, 1, minus
);
3757 if (!strv_isempty(plus
)) {
3758 b
= strv_env_merge(l
, plus
);
3767 if (m
->client_environment
!= l
)
3768 strv_free(m
->client_environment
);
3775 m
->client_environment
= sanitize_environment(l
);
3779 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
3785 l
= strv_env_merge(m
->transient_environment
, m
->client_environment
);
3793 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3796 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
3797 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3799 if (!default_rlimit
[i
])
3802 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3810 void manager_recheck_dbus(Manager
*m
) {
3813 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3814 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3815 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3816 * while in the user instance we can assume it's already there. */
3818 if (MANAGER_IS_RELOADING(m
))
3819 return; /* don't check while we are reloading… */
3821 if (manager_dbus_is_running(m
, false)) {
3822 (void) bus_init_api(m
);
3824 if (MANAGER_IS_SYSTEM(m
))
3825 (void) bus_init_system(m
);
3827 (void) bus_done_api(m
);
3829 if (MANAGER_IS_SYSTEM(m
))
3830 (void) bus_done_system(m
);
3834 static bool manager_journal_is_running(Manager
*m
) {
3839 if (MANAGER_IS_TEST_RUN(m
))
3842 /* If we are the user manager we can safely assume that the journal is up */
3843 if (!MANAGER_IS_SYSTEM(m
))
3846 /* Check that the socket is not only up, but in RUNNING state */
3847 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3850 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3853 /* Similar, check if the daemon itself is fully up, too */
3854 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3857 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3863 void disable_printk_ratelimit(void) {
3864 /* Disable kernel's printk ratelimit.
3866 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3867 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3868 * setting takes precedence. */
3871 r
= sysctl_write("kernel/printk_devkmsg", "on");
3873 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3876 void manager_recheck_journal(Manager
*m
) {
3880 /* Don't bother with this unless we are in the special situation of being PID 1 */
3881 if (getpid_cached() != 1)
3884 /* Don't check this while we are reloading, things might still change */
3885 if (MANAGER_IS_RELOADING(m
))
3888 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3889 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3890 * an activation ourselves we can't fulfill. */
3891 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3895 static ShowStatus
manager_get_show_status(Manager
*m
) {
3898 if (MANAGER_IS_USER(m
))
3899 return _SHOW_STATUS_INVALID
;
3901 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3902 return m
->show_status_overridden
;
3904 return m
->show_status
;
3907 bool manager_get_show_status_on(Manager
*m
) {
3910 return show_status_on(manager_get_show_status(m
));
3913 static void set_show_status_marker(bool b
) {
3915 (void) touch("/run/systemd/show-status");
3917 (void) unlink("/run/systemd/show-status");
3920 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
3923 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
3925 if (MANAGER_IS_USER(m
))
3928 if (mode
== m
->show_status
)
3931 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
3934 enabled
= show_status_on(mode
);
3935 log_debug("%s (%s) showing of status (%s).",
3936 enabled
? "Enabling" : "Disabling",
3937 strna(show_status_to_string(mode
)),
3940 set_show_status_marker(enabled
);
3943 m
->show_status
= mode
;
3946 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
3948 assert(mode
< _SHOW_STATUS_MAX
);
3950 if (MANAGER_IS_USER(m
))
3953 if (mode
== m
->show_status_overridden
)
3956 m
->show_status_overridden
= mode
;
3958 if (mode
== _SHOW_STATUS_INVALID
)
3959 mode
= m
->show_status
;
3961 log_debug("%s (%s) showing of status (%s).",
3962 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
3963 strna(show_status_to_string(mode
)),
3966 set_show_status_marker(show_status_on(mode
));
3969 const char *manager_get_confirm_spawn(Manager
*m
) {
3970 static int last_errno
= 0;
3976 /* Here's the deal: we want to test the validity of the console but don't want
3977 * PID1 to go through the whole console process which might block. But we also
3978 * want to warn the user only once if something is wrong with the console so we
3979 * cannot do the sanity checks after spawning our children. So here we simply do
3980 * really basic tests to hopefully trap common errors.
3982 * If the console suddenly disappear at the time our children will really it
3983 * then they will simply fail to acquire it and a positive answer will be
3984 * assumed. New children will fall back to /dev/console though.
3986 * Note: TTYs are devices that can come and go any time, and frequently aren't
3987 * available yet during early boot (consider a USB rs232 dongle...). If for any
3988 * reason the configured console is not ready, we fall back to the default
3991 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
3992 return m
->confirm_spawn
;
3994 if (stat(m
->confirm_spawn
, &st
) < 0) {
3999 if (!S_ISCHR(st
.st_mode
)) {
4005 return m
->confirm_spawn
;
4008 if (last_errno
!= r
)
4009 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4011 return "/dev/console";
4014 void manager_set_first_boot(Manager
*m
, bool b
) {
4017 if (!MANAGER_IS_SYSTEM(m
))
4020 if (m
->first_boot
!= (int) b
) {
4022 (void) touch("/run/systemd/first-boot");
4024 (void) unlink("/run/systemd/first-boot");
4030 void manager_disable_confirm_spawn(void) {
4031 (void) touch("/run/systemd/confirm_spawn_disabled");
4034 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4035 if (!m
->confirm_spawn
)
4038 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4041 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4044 if (!MANAGER_IS_SYSTEM(m
))
4047 if (m
->no_console_output
)
4050 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4053 /* If we cannot find out the status properly, just proceed. */
4054 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4057 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4060 return manager_get_show_status_on(m
);
4063 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4066 /* If m is NULL, assume we're after shutdown and let the messages through. */
4068 if (m
&& !manager_should_show_status(m
, type
))
4071 /* XXX We should totally drop the check for ephemeral here
4072 * and thus effectively make 'Type=idle' pointless. */
4073 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4076 va_start(ap
, format
);
4077 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4081 Set
* manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4085 if (path_equal(path
, "/"))
4088 return hashmap_get(m
->units_requiring_mounts_for
, path
);
4091 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4096 assert(u
->manager
== m
);
4098 size
= set_size(m
->failed_units
);
4101 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4105 (void) set_remove(m
->failed_units
, u
);
4107 if (set_size(m
->failed_units
) != size
)
4108 bus_manager_send_change_signal(m
);
4113 ManagerState
manager_state(Manager
*m
) {
4118 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4119 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4120 if (u
&& unit_active_or_pending(u
))
4121 return MANAGER_STOPPING
;
4123 /* Did we ever finish booting? If not then we are still starting up */
4124 if (!MANAGER_IS_FINISHED(m
)) {
4126 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4127 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4128 return MANAGER_INITIALIZING
;
4130 return MANAGER_STARTING
;
4133 if (MANAGER_IS_SYSTEM(m
)) {
4134 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4135 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4136 if (u
&& unit_active_or_pending(u
))
4137 return MANAGER_MAINTENANCE
;
4139 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4140 if (u
&& unit_active_or_pending(u
))
4141 return MANAGER_MAINTENANCE
;
4144 /* Are there any failed units? If so, we are in degraded mode */
4145 if (set_size(m
->failed_units
) > 0)
4146 return MANAGER_DEGRADED
;
4148 return MANAGER_RUNNING
;
4151 static void manager_unref_uid_internal(
4155 int (*_clean_ipc
)(uid_t uid
)) {
4159 assert(uid_is_valid(uid
));
4162 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4163 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4165 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose highest
4166 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4167 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4168 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4170 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4171 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4173 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4176 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4178 n
= c
& ~DESTROY_IPC_FLAG
;
4182 if (destroy_now
&& n
== 0) {
4183 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4185 if (c
& DESTROY_IPC_FLAG
) {
4186 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4187 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4189 (void) _clean_ipc(uid
);
4192 c
= n
| (c
& DESTROY_IPC_FLAG
);
4193 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4197 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4198 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4201 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4202 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4205 static int manager_ref_uid_internal(
4214 assert(uid_is_valid(uid
));
4216 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4217 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4219 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4220 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4222 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4225 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4229 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4231 n
= c
& ~DESTROY_IPC_FLAG
;
4234 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4237 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4239 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4242 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4243 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4246 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4247 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4250 static void manager_vacuum_uid_refs_internal(
4252 int (*_clean_ipc
)(uid_t uid
)) {
4258 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4262 uid
= PTR_TO_UID(k
);
4263 c
= PTR_TO_UINT32(p
);
4265 n
= c
& ~DESTROY_IPC_FLAG
;
4269 if (c
& DESTROY_IPC_FLAG
) {
4270 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4271 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4273 (void) _clean_ipc(uid
);
4276 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4280 static void manager_vacuum_uid_refs(Manager
*m
) {
4281 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4284 static void manager_vacuum_gid_refs(Manager
*m
) {
4285 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4288 static void manager_vacuum(Manager
*m
) {
4291 /* Release any dynamic users no longer referenced */
4292 dynamic_user_vacuum(m
, true);
4294 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4295 manager_vacuum_uid_refs(m
);
4296 manager_vacuum_gid_refs(m
);
4298 /* Release any runtimes no longer referenced */
4299 exec_runtime_vacuum(m
);
4302 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4306 char unit_name
[UNIT_NAME_MAX
+1];
4309 Manager
*m
= userdata
;
4317 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4318 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4319 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4321 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4323 if (ERRNO_IS_TRANSIENT(errno
))
4326 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4329 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4330 log_warning("Received too short user lookup message, ignoring.");
4334 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4335 log_warning("Received too long user lookup message, ignoring.");
4339 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4340 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4344 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4345 if (memchr(buffer
.unit_name
, 0, n
)) {
4346 log_warning("Received lookup message with embedded NUL character, ignoring.");
4350 buffer
.unit_name
[n
] = 0;
4351 u
= manager_get_unit(m
, buffer
.unit_name
);
4353 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4357 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4359 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4363 char *manager_taint_string(Manager
*m
) {
4364 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4368 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4369 * Only things that are detected at runtime should be tagged
4370 * here. For stuff that is set during compilation, emit a warning
4371 * in the configuration phase. */
4375 buf
= new(char, sizeof("split-usr:"
4380 "overflowuid-not-65534:"
4381 "overflowgid-not-65534:"));
4389 e
= stpcpy(e
, "split-usr:");
4391 if (access("/proc/cgroups", F_OK
) < 0)
4392 e
= stpcpy(e
, "cgroups-missing:");
4394 if (cg_all_unified() == 0)
4395 e
= stpcpy(e
, "cgroupsv1:");
4397 if (clock_is_localtime(NULL
) > 0)
4398 e
= stpcpy(e
, "local-hwclock:");
4400 r
= readlink_malloc("/var/run", &destination
);
4401 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4402 e
= stpcpy(e
, "var-run-bad:");
4404 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4405 if (r
>= 0 && !streq(overflowuid
, "65534"))
4406 e
= stpcpy(e
, "overflowuid-not-65534:");
4408 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4409 if (r
>= 0 && !streq(overflowgid
, "65534"))
4410 e
= stpcpy(e
, "overflowgid-not-65534:");
4412 /* remove the last ':' */
4419 void manager_ref_console(Manager
*m
) {
4425 void manager_unref_console(Manager
*m
) {
4427 assert(m
->n_on_console
> 0);
4430 if (m
->n_on_console
== 0)
4431 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4434 void manager_override_log_level(Manager
*m
, int level
) {
4435 _cleanup_free_
char *s
= NULL
;
4438 if (!m
->log_level_overridden
) {
4439 m
->original_log_level
= log_get_max_level();
4440 m
->log_level_overridden
= true;
4443 (void) log_level_to_string_alloc(level
, &s
);
4444 log_info("Setting log level to %s.", strna(s
));
4446 log_set_max_level(level
);
4449 void manager_restore_original_log_level(Manager
*m
) {
4450 _cleanup_free_
char *s
= NULL
;
4453 if (!m
->log_level_overridden
)
4456 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4457 log_info("Restoring log level to original (%s).", strna(s
));
4459 log_set_max_level(m
->original_log_level
);
4460 m
->log_level_overridden
= false;
4463 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4466 if (!m
->log_target_overridden
) {
4467 m
->original_log_target
= log_get_target();
4468 m
->log_target_overridden
= true;
4471 log_info("Setting log target to %s.", log_target_to_string(target
));
4472 log_set_target(target
);
4475 void manager_restore_original_log_target(Manager
*m
) {
4478 if (!m
->log_target_overridden
)
4481 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4483 log_set_target(m
->original_log_target
);
4484 m
->log_target_overridden
= false;
4487 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4489 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4490 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4491 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4495 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4496 [MANAGER_INITIALIZING
] = "initializing",
4497 [MANAGER_STARTING
] = "starting",
4498 [MANAGER_RUNNING
] = "running",
4499 [MANAGER_DEGRADED
] = "degraded",
4500 [MANAGER_MAINTENANCE
] = "maintenance",
4501 [MANAGER_STOPPING
] = "stopping",
4504 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4506 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4507 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4508 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4509 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4510 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4511 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4512 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4513 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4514 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4515 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4516 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4517 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4518 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4519 [MANAGER_TIMESTAMP_UNITS_LOAD
] = "units-load",
4520 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4521 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4522 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4523 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4524 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4525 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4528 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4530 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4531 [OOM_CONTINUE
] = "continue",
4532 [OOM_STOP
] = "stop",
4533 [OOM_KILL
] = "kill",
4536 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);