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"
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
;
292 dir
= opendir("/run/systemd/ask-password");
300 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
301 if (startswith(de
->d_name
, "ask."))
307 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
308 int fd
, uint32_t revents
, void *userdata
) {
309 Manager
*m
= userdata
;
315 m
->have_ask_password
= have_ask_password();
316 if (m
->have_ask_password
< 0)
317 /* Log error but continue. Negative have_ask_password
318 * is treated as unknown status. */
319 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
324 static void manager_close_ask_password(Manager
*m
) {
327 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
328 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
329 m
->have_ask_password
= -EINVAL
;
332 static int manager_check_ask_password(Manager
*m
) {
337 if (!m
->ask_password_event_source
) {
338 assert(m
->ask_password_inotify_fd
< 0);
340 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
342 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
343 if (m
->ask_password_inotify_fd
< 0)
344 return log_error_errno(errno
, "Failed to create inotify object: %m");
346 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
347 "/run/systemd/ask-password",
348 IN_CREATE
|IN_DELETE
|IN_MOVE
);
350 manager_close_ask_password(m
);
354 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
355 m
->ask_password_inotify_fd
, EPOLLIN
,
356 manager_dispatch_ask_password_fd
, m
);
358 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
359 manager_close_ask_password(m
);
363 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
365 /* Queries might have been added meanwhile... */
366 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
367 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
370 return m
->have_ask_password
;
373 static int manager_watch_idle_pipe(Manager
*m
) {
378 if (m
->idle_pipe_event_source
)
381 if (m
->idle_pipe
[2] < 0)
384 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
386 return log_error_errno(r
, "Failed to watch idle pipe: %m");
388 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
393 static void manager_close_idle_pipe(Manager
*m
) {
396 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
398 safe_close_pair(m
->idle_pipe
);
399 safe_close_pair(m
->idle_pipe
+ 2);
402 static int manager_setup_time_change(Manager
*m
) {
407 if (MANAGER_IS_TEST_RUN(m
))
410 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
411 m
->time_change_fd
= safe_close(m
->time_change_fd
);
413 m
->time_change_fd
= time_change_fd();
414 if (m
->time_change_fd
< 0)
415 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
417 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
419 return log_error_errno(r
, "Failed to create time change event source: %m");
421 /* Schedule this slightly earlier than the .timer event sources */
422 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
424 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
426 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
428 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
433 static int manager_read_timezone_stat(Manager
*m
) {
439 /* Read the current stat() data of /etc/localtime so that we detect changes */
440 if (lstat("/etc/localtime", &st
) < 0) {
441 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
442 changed
= m
->etc_localtime_accessible
;
443 m
->etc_localtime_accessible
= false;
447 k
= timespec_load(&st
.st_mtim
);
448 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
450 m
->etc_localtime_mtime
= k
;
451 m
->etc_localtime_accessible
= true;
457 static int manager_setup_timezone_change(Manager
*m
) {
458 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
463 if (MANAGER_IS_TEST_RUN(m
))
466 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
467 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
468 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
469 * went to zero and all fds to it are closed.
471 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
474 * Note that we create the new event source first here, before releasing the old one. This should optimize
475 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
477 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
478 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
480 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
481 * O_CREATE or by rename() */
483 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
484 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
485 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
488 return log_error_errno(r
, "Failed to create timezone change event source: %m");
490 /* Schedule this slightly earlier than the .timer event sources */
491 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
493 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
495 sd_event_source_unref(m
->timezone_change_event_source
);
496 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
501 static int enable_special_signals(Manager
*m
) {
502 _cleanup_close_
int fd
= -1;
506 if (MANAGER_IS_TEST_RUN(m
))
509 /* Enable that we get SIGINT on control-alt-del. In containers
510 * this will fail with EPERM (older) or EINVAL (newer), so
512 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
513 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
515 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
517 /* Support systems without virtual console */
519 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
521 /* Enable that we get SIGWINCH on kbrequest */
522 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
523 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
529 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
531 static int manager_setup_signals(Manager
*m
) {
532 struct sigaction sa
= {
533 .sa_handler
= SIG_DFL
,
534 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
541 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
543 /* We make liberal use of realtime signals here. On
544 * Linux/glibc we have 30 of them (with the exception of Linux
545 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
548 assert_se(sigemptyset(&mask
) == 0);
549 sigset_add_many(&mask
,
550 SIGCHLD
, /* Child died */
551 SIGTERM
, /* Reexecute daemon */
552 SIGHUP
, /* Reload configuration */
553 SIGUSR1
, /* systemd: reconnect to D-Bus */
554 SIGUSR2
, /* systemd: dump status */
555 SIGINT
, /* Kernel sends us this on control-alt-del */
556 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
557 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
559 SIGRTMIN
+0, /* systemd: start default.target */
560 SIGRTMIN
+1, /* systemd: isolate rescue.target */
561 SIGRTMIN
+2, /* systemd: isolate emergency.target */
562 SIGRTMIN
+3, /* systemd: start halt.target */
563 SIGRTMIN
+4, /* systemd: start poweroff.target */
564 SIGRTMIN
+5, /* systemd: start reboot.target */
565 SIGRTMIN
+6, /* systemd: start kexec.target */
567 /* ... space for more special targets ... */
569 SIGRTMIN
+13, /* systemd: Immediate halt */
570 SIGRTMIN
+14, /* systemd: Immediate poweroff */
571 SIGRTMIN
+15, /* systemd: Immediate reboot */
572 SIGRTMIN
+16, /* systemd: Immediate kexec */
574 /* ... space for more immediate system state changes ... */
576 SIGRTMIN
+20, /* systemd: enable status messages */
577 SIGRTMIN
+21, /* systemd: disable status messages */
578 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
579 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
580 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
581 SIGRTMIN
+25, /* systemd: reexecute manager */
583 /* Apparently Linux on hppa had fewer RT signals until v3.18,
584 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
585 * see commit v3.17-7614-g1f25df2eff.
587 * We cannot unconditionally make use of those signals here,
588 * so let's use a runtime check. Since these commands are
589 * accessible by different means and only really a safety
590 * net, the missing functionality on hppa shouldn't matter.
593 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
594 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
595 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
596 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
598 /* ... one free signal here SIGRTMIN+30 ... */
600 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
602 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
603 if (m
->signal_fd
< 0)
606 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
610 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
612 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
613 * notify processing can still figure out to which process/service a message belongs, before we reap the
614 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
615 * status information before detecting that there's no process in a cgroup. */
616 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
620 if (MANAGER_IS_SYSTEM(m
))
621 return enable_special_signals(m
);
626 static char** sanitize_environment(char **l
) {
628 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
632 "CONFIGURATION_DIRECTORY",
633 "CREDENTIALS_DIRECTORY",
655 /* Let's order the environment alphabetically, just to make it pretty */
659 int manager_default_environment(Manager
*m
) {
664 m
->transient_environment
= strv_free(m
->transient_environment
);
666 if (MANAGER_IS_SYSTEM(m
)) {
667 /* The system manager always starts with a clean
668 * environment for its children. It does not import
669 * the kernel's or the parents' exported variables.
671 * The initial passed environment is untouched to keep
672 * /proc/self/environ valid; it is used for tagging
673 * the init process inside containers. */
674 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
675 if (!m
->transient_environment
)
678 /* Import locale variables LC_*= from configuration */
679 (void) locale_setup(&m
->transient_environment
);
681 /* The user manager passes its own environment along to its children, except for $PATH. */
682 m
->transient_environment
= strv_copy(environ
);
683 if (!m
->transient_environment
)
686 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
691 sanitize_environment(m
->transient_environment
);
696 static int manager_setup_prefix(Manager
*m
) {
702 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
703 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
704 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
705 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
706 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
707 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
710 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
711 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
712 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
713 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
714 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
715 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
720 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
723 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
724 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
726 return log_warning_errno(r
, "Failed to lookup %s path: %m",
727 exec_directory_type_to_string(i
));
733 static void manager_free_unit_name_maps(Manager
*m
) {
734 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
735 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
736 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
737 m
->unit_cache_timestamp_hash
= 0;
740 static int manager_setup_run_queue(Manager
*m
) {
744 assert(!m
->run_queue_event_source
);
746 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
750 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
754 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
758 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
763 static int manager_setup_sigchld_event_source(Manager
*m
) {
767 assert(!m
->sigchld_event_source
);
769 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
773 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
777 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
781 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
786 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
787 _cleanup_(manager_freep
) Manager
*m
= NULL
;
792 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
799 .unit_file_scope
= scope
,
800 .objective
= _MANAGER_OBJECTIVE_INVALID
,
802 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
804 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
805 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
806 .default_tasks_accounting
= true,
807 .default_tasks_max
= TASKS_MAX_UNSET
,
808 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
809 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
810 .default_restart_usec
= DEFAULT_RESTART_USEC
,
812 .original_log_level
= -1,
813 .original_log_target
= _LOG_TARGET_INVALID
,
815 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
816 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
817 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
819 .show_status_overridden
= _SHOW_STATUS_INVALID
,
822 .cgroups_agent_fd
= -1,
824 .time_change_fd
= -1,
825 .user_lookup_fds
= { -1, -1 },
826 .private_listen_fd
= -1,
828 .cgroup_inotify_fd
= -1,
829 .pin_cgroupfs_fd
= -1,
830 .ask_password_inotify_fd
= -1,
831 .idle_pipe
= { -1, -1, -1, -1},
833 /* start as id #1, so that we can leave #0 around as "null-like" value */
836 .have_ask_password
= -EINVAL
, /* we don't know */
838 .test_run_flags
= test_run_flags
,
840 .default_oom_policy
= OOM_STOP
,
844 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
845 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
846 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
847 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
850 /* Prepare log fields we can use for structured logging */
851 if (MANAGER_IS_SYSTEM(m
)) {
852 m
->unit_log_field
= "UNIT=";
853 m
->unit_log_format_string
= "UNIT=%s";
855 m
->invocation_log_field
= "INVOCATION_ID=";
856 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
858 m
->unit_log_field
= "USER_UNIT=";
859 m
->unit_log_format_string
= "USER_UNIT=%s";
861 m
->invocation_log_field
= "USER_INVOCATION_ID=";
862 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
865 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
866 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
868 r
= manager_default_environment(m
);
872 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
876 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
880 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
884 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
888 r
= manager_setup_prefix(m
);
892 r
= get_credentials_dir(&e
);
894 m
->received_credentials
= strdup(e
);
895 if (!m
->received_credentials
)
899 r
= sd_event_default(&m
->event
);
903 r
= manager_setup_run_queue(m
);
907 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
908 m
->cgroup_root
= strdup("");
912 r
= manager_setup_signals(m
);
916 r
= manager_setup_cgroup(m
);
920 r
= manager_setup_time_change(m
);
924 r
= manager_read_timezone_stat(m
);
928 (void) manager_setup_timezone_change(m
);
930 r
= manager_setup_sigchld_event_source(m
);
935 if (MANAGER_IS_SYSTEM(m
) && lsm_bpf_supported()) {
936 r
= lsm_bpf_setup(m
);
943 if (test_run_flags
== 0) {
944 if (MANAGER_IS_SYSTEM(m
))
945 r
= mkdir_label("/run/systemd/units", 0755);
947 _cleanup_free_
char *units_path
= NULL
;
948 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
951 r
= mkdir_p_label(units_path
, 0755);
954 if (r
< 0 && r
!= -EEXIST
)
960 dir_is_empty("/usr") > 0;
962 /* Note that we do not set up the notify fd here. We do that after deserialization,
963 * since they might have gotten serialized across the reexec. */
970 static int manager_setup_notify(Manager
*m
) {
973 if (MANAGER_IS_TEST_RUN(m
))
976 if (m
->notify_fd
< 0) {
977 _cleanup_close_
int fd
= -1;
978 union sockaddr_union sa
;
981 /* First free all secondary fields */
982 m
->notify_socket
= mfree(m
->notify_socket
);
983 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
985 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
987 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
989 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
991 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
992 if (!m
->notify_socket
)
995 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
997 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1001 (void) mkdir_parents_label(m
->notify_socket
, 0755);
1002 (void) sockaddr_un_unlink(&sa
.un
);
1004 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
1006 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
1008 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1010 return log_error_errno(r
, "SO_PASSCRED failed: %m");
1012 m
->notify_fd
= TAKE_FD(fd
);
1014 log_debug("Using notification socket %s", m
->notify_socket
);
1017 if (!m
->notify_event_source
) {
1018 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
1020 return log_error_errno(r
, "Failed to allocate notify event source: %m");
1022 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1023 * service an exit message belongs. */
1024 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
1026 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
1028 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
1034 static int manager_setup_cgroups_agent(Manager
*m
) {
1036 static const union sockaddr_union sa
= {
1037 .un
.sun_family
= AF_UNIX
,
1038 .un
.sun_path
= "/run/systemd/cgroups-agent",
1042 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1043 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1044 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1045 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1046 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1047 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1048 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1049 * we thus won't lose messages.
1051 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1052 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1053 * bus for these messages. */
1055 if (MANAGER_IS_TEST_RUN(m
))
1058 if (!MANAGER_IS_SYSTEM(m
))
1061 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1063 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1064 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1067 if (m
->cgroups_agent_fd
< 0) {
1068 _cleanup_close_
int fd
= -1;
1070 /* First free all secondary fields */
1071 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1073 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1075 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1077 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1079 (void) sockaddr_un_unlink(&sa
.un
);
1081 /* Only allow root to connect to this socket */
1082 RUN_WITH_UMASK(0077)
1083 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1085 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1087 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1090 if (!m
->cgroups_agent_event_source
) {
1091 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1093 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1095 /* Process cgroups notifications early. Note that when the agent notification is received
1096 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1097 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1098 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1100 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1102 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1108 static int manager_setup_user_lookup_fd(Manager
*m
) {
1113 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1114 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1115 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1116 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1117 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1118 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1119 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1120 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1123 * You might wonder why we need a communication channel for this that is independent of the usual notification
1124 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1125 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1126 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1128 * Note that this function is called under two circumstances: when we first initialize (in which case we
1129 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1130 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1132 if (m
->user_lookup_fds
[0] < 0) {
1134 /* Free all secondary fields */
1135 safe_close_pair(m
->user_lookup_fds
);
1136 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1138 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1139 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1141 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1144 if (!m
->user_lookup_event_source
) {
1145 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1147 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1149 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1151 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1153 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1155 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1161 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1167 while ((u
= m
->cleanup_queue
)) {
1168 assert(u
->in_cleanup_queue
);
1178 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1179 GC_OFFSET_UNSURE
, /* No clue */
1180 GC_OFFSET_GOOD
, /* We still need this unit */
1181 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1185 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1188 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1190 /* Recursively mark referenced units as GOOD as well */
1191 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1192 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1193 unit_gc_mark_good(other
, gc_marker
);
1196 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1202 if (IN_SET(u
->gc_marker
- gc_marker
,
1203 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1206 if (u
->in_cleanup_queue
)
1209 if (!unit_may_gc(u
))
1212 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1216 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1217 unit_gc_sweep(other
, gc_marker
);
1219 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1222 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1227 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1228 unit_gc_sweep(ref
->source
, gc_marker
);
1230 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1233 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1240 /* We were unable to find anything out about this entry, so
1241 * let's investigate it later */
1242 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1243 unit_add_to_gc_queue(u
);
1247 /* We definitely know that this one is not useful anymore, so
1248 * let's mark it for deletion */
1249 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1250 unit_add_to_cleanup_queue(u
);
1254 unit_gc_mark_good(u
, gc_marker
);
1257 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1258 unsigned n
= 0, gc_marker
;
1263 /* log_debug("Running GC..."); */
1265 m
->gc_marker
+= _GC_OFFSET_MAX
;
1266 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1269 gc_marker
= m
->gc_marker
;
1271 while ((u
= m
->gc_unit_queue
)) {
1272 assert(u
->in_gc_queue
);
1274 unit_gc_sweep(u
, gc_marker
);
1276 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1277 u
->in_gc_queue
= false;
1281 if (IN_SET(u
->gc_marker
- gc_marker
,
1282 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1284 log_unit_debug(u
, "Collecting.");
1285 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1286 unit_add_to_cleanup_queue(u
);
1293 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1299 while ((j
= m
->gc_job_queue
)) {
1300 assert(j
->in_gc_queue
);
1302 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1303 j
->in_gc_queue
= false;
1310 log_unit_debug(j
->unit
, "Collecting job.");
1311 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1317 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1324 while ((u
= m
->stop_when_unneeded_queue
)) {
1325 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1327 assert(u
->in_stop_when_unneeded_queue
);
1328 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1329 u
->in_stop_when_unneeded_queue
= false;
1333 if (!unit_is_unneeded(u
))
1336 log_unit_debug(u
, "Unit is not needed anymore.");
1338 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1339 * service being unnecessary after a while. */
1341 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1342 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1346 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1347 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1349 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1355 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1362 while ((u
= m
->start_when_upheld_queue
)) {
1363 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1364 Unit
*culprit
= NULL
;
1366 assert(u
->in_start_when_upheld_queue
);
1367 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1368 u
->in_start_when_upheld_queue
= false;
1372 if (!unit_is_upheld_by_active(u
, &culprit
))
1375 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1377 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1378 * service being unnecessary after a while. */
1380 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1381 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
);
1385 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1387 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1393 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1400 while ((u
= m
->stop_when_bound_queue
)) {
1401 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1402 Unit
*culprit
= NULL
;
1404 assert(u
->in_stop_when_bound_queue
);
1405 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1406 u
->in_stop_when_bound_queue
= false;
1410 if (!unit_is_bound_by_inactive(u
, &culprit
))
1413 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1415 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1416 * service being unnecessary after a while. */
1418 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1419 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
);
1423 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1425 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1431 static void manager_clear_jobs_and_units(Manager
*m
) {
1436 while ((u
= hashmap_first(m
->units
)))
1439 manager_dispatch_cleanup_queue(m
);
1441 assert(!m
->load_queue
);
1442 assert(prioq_isempty(m
->run_queue
));
1443 assert(!m
->dbus_unit_queue
);
1444 assert(!m
->dbus_job_queue
);
1445 assert(!m
->cleanup_queue
);
1446 assert(!m
->gc_unit_queue
);
1447 assert(!m
->gc_job_queue
);
1448 assert(!m
->cgroup_realize_queue
);
1449 assert(!m
->cgroup_empty_queue
);
1450 assert(!m
->cgroup_oom_queue
);
1451 assert(!m
->target_deps_queue
);
1452 assert(!m
->stop_when_unneeded_queue
);
1453 assert(!m
->start_when_upheld_queue
);
1454 assert(!m
->stop_when_bound_queue
);
1456 assert(hashmap_isempty(m
->jobs
));
1457 assert(hashmap_isempty(m
->units
));
1459 m
->n_on_console
= 0;
1460 m
->n_running_jobs
= 0;
1461 m
->n_installed_jobs
= 0;
1462 m
->n_failed_jobs
= 0;
1465 Manager
* manager_free(Manager
*m
) {
1469 manager_clear_jobs_and_units(m
);
1471 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1472 if (unit_vtable
[c
]->shutdown
)
1473 unit_vtable
[c
]->shutdown(m
);
1475 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1476 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1478 lookup_paths_flush_generator(&m
->lookup_paths
);
1481 manager_varlink_done(m
);
1483 exec_runtime_vacuum(m
);
1484 hashmap_free(m
->exec_runtime_by_id
);
1486 dynamic_user_vacuum(m
, false);
1487 hashmap_free(m
->dynamic_users
);
1489 hashmap_free(m
->units
);
1490 hashmap_free(m
->units_by_invocation_id
);
1491 hashmap_free(m
->jobs
);
1492 hashmap_free(m
->watch_pids
);
1493 hashmap_free(m
->watch_bus
);
1495 prioq_free(m
->run_queue
);
1497 set_free(m
->startup_units
);
1498 set_free(m
->failed_units
);
1500 sd_event_source_unref(m
->signal_event_source
);
1501 sd_event_source_unref(m
->sigchld_event_source
);
1502 sd_event_source_unref(m
->notify_event_source
);
1503 sd_event_source_unref(m
->cgroups_agent_event_source
);
1504 sd_event_source_unref(m
->time_change_event_source
);
1505 sd_event_source_unref(m
->timezone_change_event_source
);
1506 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1507 sd_event_source_unref(m
->run_queue_event_source
);
1508 sd_event_source_unref(m
->user_lookup_event_source
);
1510 safe_close(m
->signal_fd
);
1511 safe_close(m
->notify_fd
);
1512 safe_close(m
->cgroups_agent_fd
);
1513 safe_close(m
->time_change_fd
);
1514 safe_close_pair(m
->user_lookup_fds
);
1516 manager_close_ask_password(m
);
1518 manager_close_idle_pipe(m
);
1520 sd_event_unref(m
->event
);
1522 free(m
->notify_socket
);
1524 lookup_paths_free(&m
->lookup_paths
);
1525 strv_free(m
->transient_environment
);
1526 strv_free(m
->client_environment
);
1528 hashmap_free(m
->cgroup_unit
);
1529 manager_free_unit_name_maps(m
);
1531 free(m
->switch_root
);
1532 free(m
->switch_root_init
);
1534 rlimit_free_all(m
->rlimit
);
1536 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1537 hashmap_free(m
->units_requiring_mounts_for
);
1539 hashmap_free(m
->uid_refs
);
1540 hashmap_free(m
->gid_refs
);
1542 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1543 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1544 free(m
->received_credentials
);
1547 lsm_bpf_destroy(m
->restrict_fs
);
1553 static void manager_enumerate_perpetual(Manager
*m
) {
1556 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1559 /* Let's ask every type to load all units from disk/kernel that it might know */
1560 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1561 if (!unit_type_supported(c
)) {
1562 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1566 if (unit_vtable
[c
]->enumerate_perpetual
)
1567 unit_vtable
[c
]->enumerate_perpetual(m
);
1571 static void manager_enumerate(Manager
*m
) {
1574 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1577 /* Let's ask every type to load all units from disk/kernel that it might know */
1578 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1579 if (!unit_type_supported(c
)) {
1580 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1584 if (unit_vtable
[c
]->enumerate
)
1585 unit_vtable
[c
]->enumerate(m
);
1588 manager_dispatch_load_queue(m
);
1591 static void manager_coldplug(Manager
*m
) {
1598 log_debug("Invoking unit coldplug() handlers…");
1600 /* Let's place the units back into their deserialized state */
1601 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1603 /* ignore aliases */
1607 r
= unit_coldplug(u
);
1609 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1613 static void manager_catchup(Manager
*m
) {
1619 log_debug("Invoking unit catchup() handlers…");
1621 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1622 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1624 /* ignore aliases */
1632 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1637 HASHMAP_FOREACH(u
, m
->units
) {
1639 if (fdset_size(fds
) <= 0)
1642 if (!UNIT_VTABLE(u
)->distribute_fds
)
1645 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1649 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1654 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1655 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1656 * rather than the current one. */
1658 if (MANAGER_IS_TEST_RUN(m
))
1661 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1664 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1667 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1670 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1676 static void manager_setup_bus(Manager
*m
) {
1679 /* Let's set up our private bus connection now, unconditionally */
1680 (void) bus_init_private(m
);
1682 /* If we are in --user mode also connect to the system bus now */
1683 if (MANAGER_IS_USER(m
))
1684 (void) bus_init_system(m
);
1686 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1687 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1688 (void) bus_init_api(m
);
1690 if (MANAGER_IS_SYSTEM(m
))
1691 (void) bus_init_system(m
);
1695 static void manager_preset_all(Manager
*m
) {
1700 if (m
->first_boot
<= 0)
1703 if (!MANAGER_IS_SYSTEM(m
))
1706 if (MANAGER_IS_TEST_RUN(m
))
1709 /* If this is the first boot, and we are in the host system, then preset everything */
1710 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1712 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1713 "Failed to populate /etc with preset unit settings, ignoring: %m");
1715 log_info("Populated /etc with preset unit settings.");
1718 static void manager_ready(Manager
*m
) {
1721 /* After having loaded everything, do the final round of catching up with what might have changed */
1723 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1725 /* It might be safe to log to the journal now and connect to dbus */
1726 manager_recheck_journal(m
);
1727 manager_recheck_dbus(m
);
1729 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1732 /* Create a file which will indicate when the manager started loading units the last time. */
1733 (void) touch_file("/run/systemd/systemd-units-load", false,
1734 m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].realtime
?: now(CLOCK_REALTIME
),
1735 UID_INVALID
, GID_INVALID
, 0444);
1737 m
->honor_device_enumeration
= true;
1740 Manager
* manager_reloading_start(Manager
*m
) {
1742 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD
);
1746 void manager_reloading_stopp(Manager
**m
) {
1748 assert((*m
)->n_reloading
> 0);
1749 (*m
)->n_reloading
--;
1753 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
, const char *root
) {
1758 /* If we are running in test mode, we still want to run the generators,
1759 * but we should not touch the real generator directories. */
1760 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1761 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1764 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1766 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1767 r
= manager_run_environment_generators(m
);
1769 r
= manager_run_generators(m
);
1770 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1774 manager_preset_all(m
);
1776 lookup_paths_log(&m
->lookup_paths
);
1779 /* This block is (optionally) done with the reloading counter bumped */
1780 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1782 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1783 * counter here already */
1785 reloading
= manager_reloading_start(m
);
1787 /* First, enumerate what we can from all config files */
1788 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1789 manager_enumerate_perpetual(m
);
1790 manager_enumerate(m
);
1791 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1793 /* Second, deserialize if there is something to deserialize */
1794 if (serialization
) {
1795 r
= manager_deserialize(m
, serialization
, fds
);
1797 return log_error_errno(r
, "Deserialization failed: %m");
1800 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1801 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1803 manager_distribute_fds(m
, fds
);
1805 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1806 r
= manager_setup_notify(m
);
1808 /* No sense to continue without notifications, our children would fail anyway. */
1811 r
= manager_setup_cgroups_agent(m
);
1813 /* Likewise, no sense to continue without empty cgroup notifications. */
1816 r
= manager_setup_user_lookup_fd(m
);
1818 /* This shouldn't fail, except if things are really broken. */
1821 /* Connect to the bus if we are good for it */
1822 manager_setup_bus(m
);
1824 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1825 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1827 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1828 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1830 r
= manager_varlink_init(m
);
1832 log_warning_errno(r
, "Failed to set up Varlink, ignoring: %m");
1834 /* Third, fire things up! */
1835 manager_coldplug(m
);
1837 /* Clean up runtime objects */
1841 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1842 * reload is finished */
1843 m
->send_reloading_done
= true;
1851 int manager_add_job(
1857 sd_bus_error
*error
,
1864 assert(type
< _JOB_TYPE_MAX
);
1866 assert(mode
< _JOB_MODE_MAX
);
1868 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1869 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1871 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1872 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1874 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1875 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1877 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1879 type
= job_type_collapse(type
, unit
);
1881 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1885 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1886 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1887 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1891 if (mode
== JOB_ISOLATE
) {
1892 r
= transaction_add_isolate_jobs(tr
, m
);
1897 if (mode
== JOB_TRIGGERING
) {
1898 r
= transaction_add_triggering_jobs(tr
, unit
);
1903 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1907 log_unit_debug(unit
,
1908 "Enqueued job %s/%s as %u", unit
->id
,
1909 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1912 *ret
= tr
->anchor_job
;
1914 transaction_free(tr
);
1918 transaction_abort(tr
);
1919 transaction_free(tr
);
1923 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1924 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1928 assert(type
< _JOB_TYPE_MAX
);
1930 assert(mode
< _JOB_MODE_MAX
);
1932 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1937 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1940 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1941 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1945 assert(type
< _JOB_TYPE_MAX
);
1947 assert(mode
< _JOB_MODE_MAX
);
1949 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1951 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1956 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1962 assert(mode
< _JOB_MODE_MAX
);
1963 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1965 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1969 /* We need an anchor job */
1970 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1974 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1975 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1977 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1981 transaction_free(tr
);
1985 transaction_abort(tr
);
1986 transaction_free(tr
);
1990 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1993 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1996 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
2000 return hashmap_get(m
->units
, name
);
2003 static int manager_dispatch_target_deps_queue(Manager
*m
) {
2009 while ((u
= m
->target_deps_queue
)) {
2010 _cleanup_free_ Unit
**targets
= NULL
;
2013 assert(u
->in_target_deps_queue
);
2015 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
2016 u
->in_target_deps_queue
= false;
2018 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2019 * dependencies, and we can't have it that hash tables we iterate through are modified while
2020 * we are iterating through them. */
2021 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
2025 for (int i
= 0; i
< n_targets
; i
++) {
2026 r
= unit_add_default_target_dependency(u
, targets
[i
]);
2035 unsigned manager_dispatch_load_queue(Manager
*m
) {
2041 /* Make sure we are not run recursively */
2042 if (m
->dispatching_load_queue
)
2045 m
->dispatching_load_queue
= true;
2047 /* Dispatches the load queue. Takes a unit from the queue and
2048 * tries to load its data until the queue is empty */
2050 while ((u
= m
->load_queue
)) {
2051 assert(u
->in_load_queue
);
2057 m
->dispatching_load_queue
= false;
2059 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2060 * should be loaded and have aliases resolved */
2061 (void) manager_dispatch_target_deps_queue(m
);
2066 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2069 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2070 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2071 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2072 if (u
->load_state
!= UNIT_NOT_FOUND
)
2075 /* The cache has been updated since the last time we tried to load the unit. There might be new
2076 * fragment paths to read. */
2077 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2080 /* The cache needs to be updated because there are modifications on disk. */
2081 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2084 int manager_load_unit_prepare(
2091 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
2099 /* This will prepare the unit for loading, but not actually load anything from disk. */
2101 if (path
&& !path_is_absolute(path
))
2102 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2105 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2106 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2107 * but this cannot be possible in any code path (See #6119). */
2109 name
= basename(path
);
2112 t
= unit_name_to_type(name
);
2114 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2115 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2116 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2118 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2121 ret
= manager_get_unit(m
, name
);
2123 /* The time-based cache allows to start new units without daemon-reload,
2124 * but if they are already referenced (because of dependencies or ordering)
2125 * then we have to force a load of the fragment. As an optimization, check
2126 * first if anything in the usual paths was modified since the last time
2127 * the cache was loaded. Also check if the last time an attempt to load the
2128 * unit was made was before the most recent cache refresh, so that we know
2129 * we need to try again — even if the cache is current, it might have been
2130 * updated in a different context before we had a chance to retry loading
2131 * this particular unit. */
2132 if (manager_unit_cache_should_retry_load(ret
))
2133 ret
->load_state
= UNIT_STUB
;
2139 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2145 r
= free_and_strdup(&ret
->fragment_path
, path
);
2150 r
= unit_add_name(ret
, name
);
2154 unit_add_to_load_queue(ret
);
2155 unit_add_to_dbus_queue(ret
);
2156 unit_add_to_gc_queue(ret
);
2164 int manager_load_unit(
2176 /* This will load the service information files, but not actually
2177 * start any services or anything. */
2179 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2183 manager_dispatch_load_queue(m
);
2185 *_ret
= unit_follow_merge(*_ret
);
2189 int manager_load_startable_unit_or_warn(
2195 /* Load a unit, make sure it loaded fully and is not masked. */
2197 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2201 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2203 return log_error_errno(r
, "Failed to load %s %s: %s",
2204 name
? "unit" : "unit file", name
?: path
,
2205 bus_error_message(&error
, r
));
2207 r
= bus_unit_validate_load_state(unit
, &error
);
2209 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2215 void manager_clear_jobs(Manager
*m
) {
2220 while ((j
= hashmap_first(m
->jobs
)))
2221 /* No need to recurse. We're cancelling all jobs. */
2222 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2225 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2228 /* First let's drop the unit keyed as "pid". */
2229 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2231 /* Then, let's also drop the array keyed by -pid. */
2232 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2235 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2236 Manager
*m
= userdata
;
2242 while ((j
= prioq_peek(m
->run_queue
))) {
2243 assert(j
->installed
);
2244 assert(j
->in_run_queue
);
2246 (void) job_run_and_invalidate(j
);
2249 if (m
->n_running_jobs
> 0)
2250 manager_watch_jobs_in_progress(m
);
2252 if (m
->n_on_console
> 0)
2253 manager_watch_idle_pipe(m
);
2258 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2259 unsigned n
= 0, budget
;
2265 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2266 * as we can. There's no point in throttling generation of signals in that case. */
2267 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2268 budget
= UINT_MAX
; /* infinite budget in this case */
2270 /* Anything to do at all? */
2271 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2274 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2275 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2276 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2279 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2280 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2281 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2282 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2283 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2284 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2285 * connections it will be counted five times. This difference in counting ("references"
2286 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2287 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2288 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2289 * currently chosen much higher than the "budget". */
2290 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2293 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2295 assert(u
->in_dbus_queue
);
2297 bus_unit_send_change_signal(u
);
2300 if (budget
!= UINT_MAX
)
2304 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2305 assert(j
->in_dbus_queue
);
2307 bus_job_send_change_signal(j
);
2310 if (budget
!= UINT_MAX
)
2314 if (m
->send_reloading_done
) {
2315 m
->send_reloading_done
= false;
2316 bus_manager_send_reloading(m
, false);
2320 if (m
->pending_reload_message
) {
2321 bus_send_pending_reload_message(m
);
2328 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2329 Manager
*m
= userdata
;
2333 n
= recv(fd
, buf
, sizeof(buf
), 0);
2335 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2337 log_error("Got zero-length cgroups agent message, ignoring.");
2340 if ((size_t) n
>= sizeof(buf
)) {
2341 log_error("Got overly long cgroups agent message, ignoring.");
2345 if (memchr(buf
, 0, n
)) {
2346 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2351 manager_notify_cgroup_empty(m
, buf
);
2352 (void) bus_forward_agent_released(m
, buf
);
2357 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2359 /* nothing else must be sent when using BARRIER=1 */
2360 if (strv_contains(tags
, "BARRIER=1")) {
2361 if (strv_length(tags
) == 1) {
2362 if (fdset_size(fds
) != 1)
2363 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2365 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2367 /* Drop the message if BARRIER=1 was found */
2374 static void manager_invoke_notify_message(
2377 const struct ucred
*ucred
,
2386 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2388 u
->notifygen
= m
->notifygen
;
2390 if (UNIT_VTABLE(u
)->notify_message
)
2391 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2393 else if (DEBUG_LOGGING
) {
2394 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2396 buf
= strv_join(tags
, ", ");
2398 x
= ellipsize(buf
, 20, 90);
2402 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2406 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2408 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2409 Manager
*m
= userdata
;
2410 char buf
[NOTIFY_BUFFER_MAX
+1];
2411 struct iovec iovec
= {
2413 .iov_len
= sizeof(buf
)-1,
2415 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2416 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2417 struct msghdr msghdr
= {
2420 .msg_control
= &control
,
2421 .msg_controllen
= sizeof(control
),
2424 struct cmsghdr
*cmsg
;
2425 struct ucred
*ucred
= NULL
;
2426 _cleanup_free_ Unit
**array_copy
= NULL
;
2427 _cleanup_strv_free_
char **tags
= NULL
;
2428 Unit
*u1
, *u2
, **array
;
2429 int r
, *fd_array
= NULL
;
2435 assert(m
->notify_fd
== fd
);
2437 if (revents
!= EPOLLIN
) {
2438 log_warning("Got unexpected poll event for notify fd.");
2442 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2443 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2444 return 0; /* Spurious wakeup, try again */
2446 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2450 /* If this is any other, real error, then let's stop processing this socket. This of course
2451 * means we won't take notification messages anymore, but that's still better than busy
2452 * looping around this: being woken up over and over again but being unable to actually read
2453 * the message off the socket. */
2454 return log_error_errno(n
, "Failed to receive notification message: %m");
2456 CMSG_FOREACH(cmsg
, &msghdr
) {
2457 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2460 fd_array
= (int*) CMSG_DATA(cmsg
);
2461 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2463 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2464 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2465 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2468 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2475 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2477 close_many(fd_array
, n_fds
);
2483 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2484 log_warning("Received notify message without valid credentials. Ignoring.");
2488 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2489 log_warning("Received notify message exceeded maximum size. Ignoring.");
2493 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2494 * trailing NUL byte in the message, but don't expect it. */
2495 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2496 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2500 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2502 tags
= strv_split_newlines(buf
);
2508 /* possibly a barrier fd, let's see */
2509 if (manager_process_barrier_fd(tags
, fds
))
2512 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2515 /* Notify every unit that might be interested, which might be multiple. */
2516 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2517 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2518 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2525 array_copy
= newdup(Unit
*, array
, k
+1);
2529 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2530 * make sure we only invoke each unit's handler once. */
2532 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2536 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2540 for (size_t i
= 0; array_copy
[i
]; i
++) {
2541 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2546 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2548 if (fdset_size(fds
) > 0)
2549 log_warning("Got extra auxiliary fds with notification message, closing them.");
2554 static void manager_invoke_sigchld_event(
2557 const siginfo_t
*si
) {
2563 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2564 if (u
->sigchldgen
== m
->sigchldgen
)
2566 u
->sigchldgen
= m
->sigchldgen
;
2568 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2569 unit_unwatch_pid(u
, si
->si_pid
);
2571 if (UNIT_VTABLE(u
)->sigchld_event
)
2572 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2575 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2576 Manager
*m
= userdata
;
2583 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2584 * while it is a zombie. */
2586 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2588 if (errno
!= ECHILD
)
2589 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2597 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2598 _cleanup_free_ Unit
**array_copy
= NULL
;
2599 _cleanup_free_
char *name
= NULL
;
2600 Unit
*u1
, *u2
, **array
;
2602 (void) get_process_comm(si
.si_pid
, &name
);
2604 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2605 si
.si_pid
, strna(name
),
2606 sigchld_code_to_string(si
.si_code
),
2608 strna(si
.si_code
== CLD_EXITED
2609 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2610 : signal_to_string(si
.si_status
)));
2612 /* Increase the generation counter used for filtering out duplicate unit invocations */
2615 /* And now figure out the unit this belongs to, it might be multiple... */
2616 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2617 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2618 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2622 /* Count how many entries the array has */
2626 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2627 array_copy
= newdup(Unit
*, array
, n
+1);
2632 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2633 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2634 * each iteration. */
2636 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2637 * We only do this for the cgroup the PID belonged to. */
2638 (void) unit_check_oom(u1
);
2640 /* This only logs for now. In the future when the interface for kills/notifications
2641 * is more stable we can extend service results table similar to how kernel oom kills
2643 (void) unit_check_oomd_kill(u1
);
2645 manager_invoke_sigchld_event(m
, u1
, &si
);
2648 manager_invoke_sigchld_event(m
, u2
, &si
);
2650 for (size_t i
= 0; array_copy
[i
]; i
++)
2651 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2654 /* And now, we actually reap the zombie. */
2655 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2656 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2663 /* All children processed for now, turn off event source */
2665 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2667 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2672 static void manager_start_special(Manager
*m
, const char *name
, JobMode mode
) {
2675 if (manager_add_job_by_name_and_warn(m
, JOB_START
, name
, mode
, NULL
, &job
) < 0)
2678 const char *s
= unit_status_string(job
->unit
, NULL
);
2680 log_info("Activating special unit %s...", s
);
2683 "STATUS=Activating special unit %s...", s
);
2684 m
->status_ready
= false;
2687 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2688 /* If the user presses C-A-D more than
2689 * 7 times within 2s, we reboot/shutdown immediately,
2690 * unless it was disabled in system.conf */
2692 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2693 manager_start_special(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2695 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2696 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2699 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2700 Manager
*m
= userdata
;
2702 struct signalfd_siginfo sfsi
;
2706 assert(m
->signal_fd
== fd
);
2708 if (revents
!= EPOLLIN
) {
2709 log_warning("Got unexpected events from signal file descriptor.");
2713 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2714 if (n
!= sizeof(sfsi
)) {
2716 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2720 if (IN_SET(errno
, EINTR
, EAGAIN
))
2723 /* We return an error here, which will kill this handler,
2724 * to avoid a busy loop on read error. */
2725 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2728 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2729 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2730 ? LOG_DEBUG
: LOG_INFO
,
2733 switch (sfsi
.ssi_signo
) {
2736 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2738 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2743 if (MANAGER_IS_SYSTEM(m
)) {
2744 /* This is for compatibility with the original sysvinit */
2745 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2748 m
->objective
= MANAGER_REEXECUTE
;
2754 if (MANAGER_IS_SYSTEM(m
))
2755 manager_handle_ctrl_alt_del(m
);
2757 manager_start_special(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2761 /* This is a nop on non-init */
2762 if (MANAGER_IS_SYSTEM(m
))
2763 manager_start_special(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2768 /* This is a nop on non-init */
2769 if (MANAGER_IS_SYSTEM(m
))
2770 manager_start_special(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2775 if (manager_dbus_is_running(m
, false)) {
2776 log_info("Trying to reconnect to bus...");
2778 (void) bus_init_api(m
);
2780 if (MANAGER_IS_SYSTEM(m
))
2781 (void) bus_init_system(m
);
2783 manager_start_special(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2788 _cleanup_free_
char *dump
= NULL
;
2790 r
= manager_get_dump_string(m
, &dump
);
2792 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2796 log_dump(LOG_INFO
, dump
);
2801 if (verify_run_space_and_log("Refusing to reload") < 0)
2804 m
->objective
= MANAGER_RELOAD
;
2809 /* Starting SIGRTMIN+0 */
2810 static const struct {
2813 } target_table
[] = {
2814 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2815 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2816 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2817 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2818 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2819 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2820 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2823 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2824 static const ManagerObjective objective_table
[] = {
2826 [1] = MANAGER_POWEROFF
,
2827 [2] = MANAGER_REBOOT
,
2828 [3] = MANAGER_KEXEC
,
2831 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2832 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2833 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2834 manager_start_special(m
, target_table
[idx
].target
, target_table
[idx
].mode
);
2838 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2839 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2840 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2844 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2847 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2851 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2855 manager_override_log_level(m
, LOG_DEBUG
);
2859 manager_restore_original_log_level(m
);
2863 if (MANAGER_IS_USER(m
)) {
2864 m
->objective
= MANAGER_EXIT
;
2868 /* This is a nop on init */
2872 m
->objective
= MANAGER_REEXECUTE
;
2876 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2877 manager_restore_original_log_target(m
);
2881 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2885 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2889 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2896 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2897 Manager
*m
= userdata
;
2901 assert(m
->time_change_fd
== fd
);
2903 log_struct(LOG_DEBUG
,
2904 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2905 LOG_MESSAGE("Time has been changed"));
2907 /* Restart the watch */
2908 (void) manager_setup_time_change(m
);
2910 HASHMAP_FOREACH(u
, m
->units
)
2911 if (UNIT_VTABLE(u
)->time_change
)
2912 UNIT_VTABLE(u
)->time_change(u
);
2917 static int manager_dispatch_timezone_change(
2918 sd_event_source
*source
,
2919 const struct inotify_event
*e
,
2922 Manager
*m
= userdata
;
2928 log_debug("inotify event for /etc/localtime");
2930 changed
= manager_read_timezone_stat(m
);
2934 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2935 (void) manager_setup_timezone_change(m
);
2937 /* Read the new timezone */
2940 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2942 HASHMAP_FOREACH(u
, m
->units
)
2943 if (UNIT_VTABLE(u
)->timezone_change
)
2944 UNIT_VTABLE(u
)->timezone_change(u
);
2949 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2950 Manager
*m
= userdata
;
2953 assert(m
->idle_pipe
[2] == fd
);
2955 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2956 * now turn off any further console output if there's at least one service that needs console access, so that
2957 * from now on our own output should not spill into that service's output anymore. After all, we support
2958 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2959 * exclusively without our interference. */
2960 m
->no_console_output
= m
->n_on_console
> 0;
2962 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2963 * by closing the pipes towards them, which is what they are waiting for. */
2964 manager_close_idle_pipe(m
);
2969 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2970 Manager
*m
= userdata
;
2976 manager_print_jobs_in_progress(m
);
2978 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2982 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2985 int manager_loop(Manager
*m
) {
2986 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2990 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2992 manager_check_finished(m
);
2994 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2995 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2997 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
2999 while (m
->objective
== MANAGER_OK
) {
3001 (void) watchdog_ping();
3003 if (!ratelimit_below(&rl
)) {
3004 /* Yay, something is going seriously wrong, pause a little */
3005 log_warning("Looping too fast. Throttling execution a little.");
3009 if (manager_dispatch_load_queue(m
) > 0)
3012 if (manager_dispatch_gc_job_queue(m
) > 0)
3015 if (manager_dispatch_gc_unit_queue(m
) > 0)
3018 if (manager_dispatch_cleanup_queue(m
) > 0)
3021 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3024 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3027 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3030 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3033 if (manager_dispatch_dbus_queue(m
) > 0)
3036 /* Sleep for watchdog runtime wait time */
3037 r
= sd_event_run(m
->event
, watchdog_runtime_wait());
3039 return log_error_errno(r
, "Failed to run event loop: %m");
3042 return m
->objective
;
3045 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3046 _cleanup_free_
char *n
= NULL
;
3047 sd_id128_t invocation_id
;
3055 r
= unit_name_from_dbus_path(s
, &n
);
3059 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3060 * as invocation ID. */
3061 r
= sd_id128_from_string(n
, &invocation_id
);
3063 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3069 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3070 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3071 SD_ID128_FORMAT_VAL(invocation_id
));
3074 /* If this didn't work, we check if this is a unit name */
3075 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3076 _cleanup_free_
char *nn
= NULL
;
3079 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3080 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3083 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3091 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3101 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3105 r
= safe_atou(p
, &id
);
3109 j
= manager_get_job(m
, id
);
3118 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3121 _cleanup_free_
char *p
= NULL
;
3125 if (!MANAGER_IS_SYSTEM(m
))
3128 audit_fd
= get_audit_fd();
3132 /* Don't generate audit events if the service was already
3133 * started and we're just deserializing */
3134 if (MANAGER_IS_RELOADING(m
))
3137 if (u
->type
!= UNIT_SERVICE
)
3140 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3142 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3146 msg
= strjoina("unit=", p
);
3147 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3149 /* We aren't allowed to send audit messages?
3150 * Then let's not retry again. */
3153 log_warning_errno(errno
, "Failed to send audit message: %m");
3159 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3160 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3161 _cleanup_free_
char *message
= NULL
;
3162 _cleanup_close_
int fd
= -1;
3165 /* Don't generate plymouth events if the service was already
3166 * started and we're just deserializing */
3167 if (MANAGER_IS_RELOADING(m
))
3170 if (!MANAGER_IS_SYSTEM(m
))
3173 if (detect_container() > 0)
3176 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3179 /* We set SOCK_NONBLOCK here so that we rather drop the
3180 * message then wait for plymouth */
3181 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3183 log_error_errno(errno
, "socket() failed: %m");
3187 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3188 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3189 log_error_errno(errno
, "connect() failed: %m");
3193 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0)
3194 return (void) log_oom();
3197 if (write(fd
, message
, n
+ 1) != n
+ 1)
3198 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3199 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3202 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3205 if (MANAGER_IS_USER(m
))
3206 return USEC_INFINITY
;
3208 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3209 return m
->watchdog_overridden
[t
];
3211 return m
->watchdog
[t
];
3214 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3218 if (MANAGER_IS_USER(m
))
3221 if (m
->watchdog
[t
] == timeout
)
3224 if (t
== WATCHDOG_RUNTIME
)
3225 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3226 (void) watchdog_setup(timeout
);
3228 m
->watchdog
[t
] = timeout
;
3231 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3235 if (MANAGER_IS_USER(m
))
3238 if (m
->watchdog_overridden
[t
] == timeout
)
3241 if (t
== WATCHDOG_RUNTIME
) {
3242 usec_t usec
= timestamp_is_set(timeout
) ? timeout
: m
->watchdog
[t
];
3244 (void) watchdog_setup(usec
);
3247 m
->watchdog_overridden
[t
] = timeout
;
3251 int manager_reload(Manager
*m
) {
3252 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3253 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3254 _cleanup_fclose_
FILE *f
= NULL
;
3259 r
= manager_open_serialization(m
, &f
);
3261 return log_error_errno(r
, "Failed to create serialization file: %m");
3267 /* We are officially in reload mode from here on. */
3268 reloading
= manager_reloading_start(m
);
3270 r
= manager_serialize(m
, f
, fds
, false);
3274 if (fseeko(f
, 0, SEEK_SET
) < 0)
3275 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3277 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3280 bus_manager_send_reloading(m
, true);
3282 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3283 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3286 manager_clear_jobs_and_units(m
);
3287 lookup_paths_flush_generator(&m
->lookup_paths
);
3288 lookup_paths_free(&m
->lookup_paths
);
3289 exec_runtime_vacuum(m
);
3290 dynamic_user_vacuum(m
, false);
3291 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3292 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3294 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3296 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3298 (void) manager_run_environment_generators(m
);
3299 (void) manager_run_generators(m
);
3301 lookup_paths_log(&m
->lookup_paths
);
3303 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3304 manager_free_unit_name_maps(m
);
3306 /* First, enumerate what we can from kernel and suchlike */
3307 manager_enumerate_perpetual(m
);
3308 manager_enumerate(m
);
3310 /* Second, deserialize our stored data */
3311 r
= manager_deserialize(m
, f
, fds
);
3313 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3315 /* We don't need the serialization anymore */
3318 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3319 (void) manager_setup_notify(m
);
3320 (void) manager_setup_cgroups_agent(m
);
3321 (void) manager_setup_user_lookup_fd(m
);
3323 /* Third, fire things up! */
3324 manager_coldplug(m
);
3326 /* Clean up runtime objects no longer referenced */
3329 /* Clean up deserialized tracked clients */
3330 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3332 /* Consider the reload process complete now. */
3333 assert(m
->n_reloading
> 0);
3336 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3337 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3338 * let's always set the flag here for safety. */
3339 m
->honor_device_enumeration
= true;
3343 m
->send_reloading_done
= true;
3347 void manager_reset_failed(Manager
*m
) {
3352 HASHMAP_FOREACH(u
, m
->units
)
3353 unit_reset_failed(u
);
3356 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3362 /* Returns true if the unit is inactive or going down */
3363 u
= manager_get_unit(m
, name
);
3367 return unit_inactive_or_pending(u
);
3370 static void log_taint_string(Manager
*m
) {
3371 _cleanup_free_
char *taint
= NULL
;
3375 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3378 m
->taint_logged
= true; /* only check for taint once */
3380 taint
= manager_taint_string(m
);
3384 log_struct(LOG_NOTICE
,
3385 LOG_MESSAGE("System is tainted: %s", taint
),
3387 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3390 static void manager_notify_finished(Manager
*m
) {
3391 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3393 if (MANAGER_IS_TEST_RUN(m
))
3396 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3397 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3400 size_t size
= sizeof buf
;
3402 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3403 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3404 * negative values. */
3406 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3407 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3408 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3409 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3411 if (firmware_usec
> 0)
3412 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3413 if (loader_usec
> 0)
3414 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3416 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3418 /* The initrd case on bare-metal */
3419 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3420 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3422 log_struct(LOG_INFO
,
3423 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3424 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3425 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3426 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3427 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3429 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3430 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3431 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3432 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3434 /* The initrd-less case on bare-metal */
3436 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3439 log_struct(LOG_INFO
,
3440 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3441 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3442 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3443 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3445 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3446 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3447 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3450 /* The container and --user case */
3451 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3452 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3454 log_struct(LOG_INFO
,
3455 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3456 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3457 LOG_MESSAGE("Startup finished in %s.",
3458 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3461 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3463 log_taint_string(m
);
3466 static void user_manager_send_ready(Manager
*m
) {
3469 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3470 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3475 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3476 m
->ready_sent
= true;
3477 m
->status_ready
= false;
3480 static void manager_send_ready(Manager
*m
) {
3481 if (m
->ready_sent
&& m
->status_ready
)
3482 /* Skip the notification if nothing changed. */
3487 m
->ready_sent
? "READY=1\n" : "");
3488 m
->ready_sent
= m
->status_ready
= true;
3491 static void manager_check_basic_target(Manager
*m
) {
3496 /* Small shortcut */
3497 if (m
->ready_sent
&& m
->taint_logged
)
3500 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3501 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3504 /* For user managers, send out READY=1 as soon as we reach basic.target */
3505 user_manager_send_ready(m
);
3507 /* Log the taint string as soon as we reach basic.target */
3508 log_taint_string(m
);
3511 void manager_check_finished(Manager
*m
) {
3514 if (MANAGER_IS_RELOADING(m
))
3517 /* Verify that we have entered the event loop already, and not left it again. */
3518 if (!MANAGER_IS_RUNNING(m
))
3521 manager_check_basic_target(m
);
3523 if (hashmap_size(m
->jobs
) > 0) {
3524 if (m
->jobs_in_progress_event_source
)
3525 /* Ignore any failure, this is only for feedback */
3526 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
3527 manager_watch_jobs_next_time(m
));
3531 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3532 kill the hashmap if it is relatively large. */
3533 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
3534 m
->jobs
= hashmap_free(m
->jobs
);
3536 manager_send_ready(m
);
3538 if (MANAGER_IS_FINISHED(m
))
3541 manager_flip_auto_status(m
, false, "boot finished");
3543 /* Notify Type=idle units that we are done now */
3544 manager_close_idle_pipe(m
);
3546 /* Turn off confirm spawn now */
3547 m
->confirm_spawn
= NULL
;
3549 /* No need to update ask password status when we're going non-interactive */
3550 manager_close_ask_password(m
);
3552 /* This is no longer the first boot */
3553 manager_set_first_boot(m
, false);
3555 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3557 manager_notify_finished(m
);
3559 manager_invalidate_startup_units(m
);
3562 static bool generator_path_any(const char* const* paths
) {
3566 /* Optimize by skipping the whole process by not creating output directories
3567 * if no generators are found. */
3568 STRV_FOREACH(path
, (char**) paths
)
3569 if (access(*path
, F_OK
) == 0)
3571 else if (errno
!= ENOENT
)
3572 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3577 static int manager_run_environment_generators(Manager
*m
) {
3578 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3579 _cleanup_strv_free_
char **paths
= NULL
;
3581 [STDOUT_GENERATE
] = &tmp
,
3582 [STDOUT_COLLECT
] = &tmp
,
3583 [STDOUT_CONSUME
] = &m
->transient_environment
,
3587 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3590 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
3594 if (!generator_path_any((const char* const*) paths
))
3597 RUN_WITH_UMASK(0022)
3598 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
3599 args
, NULL
, m
->transient_environment
,
3600 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3604 static int manager_run_generators(Manager
*m
) {
3605 _cleanup_strv_free_
char **paths
= NULL
;
3606 const char *argv
[5];
3611 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3614 paths
= generator_binary_paths(m
->unit_file_scope
);
3618 if (!generator_path_any((const char* const*) paths
))
3621 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3623 log_error_errno(r
, "Failed to create generator directories: %m");
3627 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3628 argv
[1] = m
->lookup_paths
.generator
;
3629 argv
[2] = m
->lookup_paths
.generator_early
;
3630 argv
[3] = m
->lookup_paths
.generator_late
;
3633 RUN_WITH_UMASK(0022)
3634 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
3635 (char**) argv
, m
->transient_environment
,
3636 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3641 lookup_paths_trim_generator(&m
->lookup_paths
);
3645 int manager_transient_environment_add(Manager
*m
, char **plus
) {
3650 if (strv_isempty(plus
))
3653 a
= strv_env_merge(m
->transient_environment
, plus
);
3657 sanitize_environment(a
);
3659 return strv_free_and_replace(m
->transient_environment
, a
);
3662 int manager_client_environment_modify(
3667 char **a
= NULL
, **b
= NULL
, **l
;
3671 if (strv_isempty(minus
) && strv_isempty(plus
))
3674 l
= m
->client_environment
;
3676 if (!strv_isempty(minus
)) {
3677 a
= strv_env_delete(l
, 1, minus
);
3684 if (!strv_isempty(plus
)) {
3685 b
= strv_env_merge(l
, plus
);
3694 if (m
->client_environment
!= l
)
3695 strv_free(m
->client_environment
);
3702 m
->client_environment
= sanitize_environment(l
);
3706 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
3712 l
= strv_env_merge(m
->transient_environment
, m
->client_environment
);
3720 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3723 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
3724 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3726 if (!default_rlimit
[i
])
3729 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3737 void manager_recheck_dbus(Manager
*m
) {
3740 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3741 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3742 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3743 * while in the user instance we can assume it's already there. */
3745 if (MANAGER_IS_RELOADING(m
))
3746 return; /* don't check while we are reloading… */
3748 if (manager_dbus_is_running(m
, false)) {
3749 (void) bus_init_api(m
);
3751 if (MANAGER_IS_SYSTEM(m
))
3752 (void) bus_init_system(m
);
3754 (void) bus_done_api(m
);
3756 if (MANAGER_IS_SYSTEM(m
))
3757 (void) bus_done_system(m
);
3761 static bool manager_journal_is_running(Manager
*m
) {
3766 if (MANAGER_IS_TEST_RUN(m
))
3769 /* If we are the user manager we can safely assume that the journal is up */
3770 if (!MANAGER_IS_SYSTEM(m
))
3773 /* Check that the socket is not only up, but in RUNNING state */
3774 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3777 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
3780 /* Similar, check if the daemon itself is fully up, too */
3781 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3784 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
3790 void disable_printk_ratelimit(void) {
3791 /* Disable kernel's printk ratelimit.
3793 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3794 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3795 * setting takes precedence. */
3798 r
= sysctl_write("kernel/printk_devkmsg", "on");
3800 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3803 void manager_recheck_journal(Manager
*m
) {
3807 /* Don't bother with this unless we are in the special situation of being PID 1 */
3808 if (getpid_cached() != 1)
3811 /* Don't check this while we are reloading, things might still change */
3812 if (MANAGER_IS_RELOADING(m
))
3815 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3816 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3817 * an activation ourselves we can't fulfill. */
3818 log_set_prohibit_ipc(!manager_journal_is_running(m
));
3822 static ShowStatus
manager_get_show_status(Manager
*m
) {
3825 if (MANAGER_IS_USER(m
))
3826 return _SHOW_STATUS_INVALID
;
3828 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3829 return m
->show_status_overridden
;
3831 return m
->show_status
;
3834 bool manager_get_show_status_on(Manager
*m
) {
3837 return show_status_on(manager_get_show_status(m
));
3840 static void set_show_status_marker(bool b
) {
3842 (void) touch("/run/systemd/show-status");
3844 (void) unlink("/run/systemd/show-status");
3847 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
3850 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
3852 if (MANAGER_IS_USER(m
))
3855 if (mode
== m
->show_status
)
3858 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
3861 enabled
= show_status_on(mode
);
3862 log_debug("%s (%s) showing of status (%s).",
3863 enabled
? "Enabling" : "Disabling",
3864 strna(show_status_to_string(mode
)),
3867 set_show_status_marker(enabled
);
3870 m
->show_status
= mode
;
3873 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
3875 assert(mode
< _SHOW_STATUS_MAX
);
3877 if (MANAGER_IS_USER(m
))
3880 if (mode
== m
->show_status_overridden
)
3883 m
->show_status_overridden
= mode
;
3885 if (mode
== _SHOW_STATUS_INVALID
)
3886 mode
= m
->show_status
;
3888 log_debug("%s (%s) showing of status (%s).",
3889 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
3890 strna(show_status_to_string(mode
)),
3893 set_show_status_marker(show_status_on(mode
));
3896 const char *manager_get_confirm_spawn(Manager
*m
) {
3897 static int last_errno
= 0;
3903 /* Here's the deal: we want to test the validity of the console but don't want
3904 * PID1 to go through the whole console process which might block. But we also
3905 * want to warn the user only once if something is wrong with the console so we
3906 * cannot do the sanity checks after spawning our children. So here we simply do
3907 * really basic tests to hopefully trap common errors.
3909 * If the console suddenly disappear at the time our children will really it
3910 * then they will simply fail to acquire it and a positive answer will be
3911 * assumed. New children will fall back to /dev/console though.
3913 * Note: TTYs are devices that can come and go any time, and frequently aren't
3914 * available yet during early boot (consider a USB rs232 dongle...). If for any
3915 * reason the configured console is not ready, we fall back to the default
3918 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
3919 return m
->confirm_spawn
;
3921 if (stat(m
->confirm_spawn
, &st
) < 0) {
3926 if (!S_ISCHR(st
.st_mode
)) {
3932 return m
->confirm_spawn
;
3935 if (last_errno
!= r
)
3936 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
3938 return "/dev/console";
3941 void manager_set_first_boot(Manager
*m
, bool b
) {
3944 if (!MANAGER_IS_SYSTEM(m
))
3947 if (m
->first_boot
!= (int) b
) {
3949 (void) touch("/run/systemd/first-boot");
3951 (void) unlink("/run/systemd/first-boot");
3957 void manager_disable_confirm_spawn(void) {
3958 (void) touch("/run/systemd/confirm_spawn_disabled");
3961 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3962 if (!m
->confirm_spawn
)
3965 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3968 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
3971 if (!MANAGER_IS_SYSTEM(m
))
3974 if (m
->no_console_output
)
3977 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3980 /* If we cannot find out the status properly, just proceed. */
3981 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3984 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
3987 return manager_get_show_status_on(m
);
3990 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3993 /* If m is NULL, assume we're after shutdown and let the messages through. */
3995 if (m
&& !manager_should_show_status(m
, type
))
3998 /* XXX We should totally drop the check for ephemeral here
3999 * and thus effectively make 'Type=idle' pointless. */
4000 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4003 va_start(ap
, format
);
4004 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4008 Set
* manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4012 if (path_equal(path
, "/"))
4015 return hashmap_get(m
->units_requiring_mounts_for
, path
);
4018 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4023 assert(u
->manager
== m
);
4025 size
= set_size(m
->failed_units
);
4028 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4032 (void) set_remove(m
->failed_units
, u
);
4034 if (set_size(m
->failed_units
) != size
)
4035 bus_manager_send_change_signal(m
);
4040 ManagerState
manager_state(Manager
*m
) {
4045 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4046 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4047 if (u
&& unit_active_or_pending(u
))
4048 return MANAGER_STOPPING
;
4050 /* Did we ever finish booting? If not then we are still starting up */
4051 if (!MANAGER_IS_FINISHED(m
)) {
4053 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4054 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4055 return MANAGER_INITIALIZING
;
4057 return MANAGER_STARTING
;
4060 if (MANAGER_IS_SYSTEM(m
)) {
4061 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4062 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4063 if (u
&& unit_active_or_pending(u
))
4064 return MANAGER_MAINTENANCE
;
4066 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4067 if (u
&& unit_active_or_pending(u
))
4068 return MANAGER_MAINTENANCE
;
4071 /* Are there any failed units? If so, we are in degraded mode */
4072 if (set_size(m
->failed_units
) > 0)
4073 return MANAGER_DEGRADED
;
4075 return MANAGER_RUNNING
;
4078 static void manager_unref_uid_internal(
4082 int (*_clean_ipc
)(uid_t uid
)) {
4086 assert(uid_is_valid(uid
));
4089 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4090 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4092 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose highest
4093 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4094 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4095 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4097 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4098 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4100 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4103 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4105 n
= c
& ~DESTROY_IPC_FLAG
;
4109 if (destroy_now
&& n
== 0) {
4110 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4112 if (c
& DESTROY_IPC_FLAG
) {
4113 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4114 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4116 (void) _clean_ipc(uid
);
4119 c
= n
| (c
& DESTROY_IPC_FLAG
);
4120 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4124 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4125 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4128 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4129 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4132 static int manager_ref_uid_internal(
4141 assert(uid_is_valid(uid
));
4143 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4144 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4146 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4147 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4149 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4152 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4156 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4158 n
= c
& ~DESTROY_IPC_FLAG
;
4161 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4164 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4166 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4169 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4170 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4173 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4174 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4177 static void manager_vacuum_uid_refs_internal(
4179 int (*_clean_ipc
)(uid_t uid
)) {
4185 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4189 uid
= PTR_TO_UID(k
);
4190 c
= PTR_TO_UINT32(p
);
4192 n
= c
& ~DESTROY_IPC_FLAG
;
4196 if (c
& DESTROY_IPC_FLAG
) {
4197 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4198 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4200 (void) _clean_ipc(uid
);
4203 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4207 static void manager_vacuum_uid_refs(Manager
*m
) {
4208 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4211 static void manager_vacuum_gid_refs(Manager
*m
) {
4212 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4215 static void manager_vacuum(Manager
*m
) {
4218 /* Release any dynamic users no longer referenced */
4219 dynamic_user_vacuum(m
, true);
4221 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4222 manager_vacuum_uid_refs(m
);
4223 manager_vacuum_gid_refs(m
);
4225 /* Release any runtimes no longer referenced */
4226 exec_runtime_vacuum(m
);
4229 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4233 char unit_name
[UNIT_NAME_MAX
+1];
4236 Manager
*m
= userdata
;
4244 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4245 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4246 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4248 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4250 if (IN_SET(errno
, EINTR
, EAGAIN
))
4253 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4256 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4257 log_warning("Received too short user lookup message, ignoring.");
4261 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4262 log_warning("Received too long user lookup message, ignoring.");
4266 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4267 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4271 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4272 if (memchr(buffer
.unit_name
, 0, n
)) {
4273 log_warning("Received lookup message with embedded NUL character, ignoring.");
4277 buffer
.unit_name
[n
] = 0;
4278 u
= manager_get_unit(m
, buffer
.unit_name
);
4280 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4284 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4286 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4290 char *manager_taint_string(Manager
*m
) {
4291 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4295 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4296 * Only things that are detected at runtime should be tagged
4297 * here. For stuff that is set during compilation, emit a warning
4298 * in the configuration phase. */
4302 buf
= new(char, sizeof("split-usr:"
4307 "overflowuid-not-65534:"
4308 "overflowgid-not-65534:"));
4316 e
= stpcpy(e
, "split-usr:");
4318 if (access("/proc/cgroups", F_OK
) < 0)
4319 e
= stpcpy(e
, "cgroups-missing:");
4321 if (cg_all_unified() == 0)
4322 e
= stpcpy(e
, "cgroupsv1:");
4324 if (clock_is_localtime(NULL
) > 0)
4325 e
= stpcpy(e
, "local-hwclock:");
4327 r
= readlink_malloc("/var/run", &destination
);
4328 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4329 e
= stpcpy(e
, "var-run-bad:");
4331 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4332 if (r
>= 0 && !streq(overflowuid
, "65534"))
4333 e
= stpcpy(e
, "overflowuid-not-65534:");
4335 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4336 if (r
>= 0 && !streq(overflowgid
, "65534"))
4337 e
= stpcpy(e
, "overflowgid-not-65534:");
4339 /* remove the last ':' */
4346 void manager_ref_console(Manager
*m
) {
4352 void manager_unref_console(Manager
*m
) {
4354 assert(m
->n_on_console
> 0);
4357 if (m
->n_on_console
== 0)
4358 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4361 void manager_override_log_level(Manager
*m
, int level
) {
4362 _cleanup_free_
char *s
= NULL
;
4365 if (!m
->log_level_overridden
) {
4366 m
->original_log_level
= log_get_max_level();
4367 m
->log_level_overridden
= true;
4370 (void) log_level_to_string_alloc(level
, &s
);
4371 log_info("Setting log level to %s.", strna(s
));
4373 log_set_max_level(level
);
4376 void manager_restore_original_log_level(Manager
*m
) {
4377 _cleanup_free_
char *s
= NULL
;
4380 if (!m
->log_level_overridden
)
4383 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4384 log_info("Restoring log level to original (%s).", strna(s
));
4386 log_set_max_level(m
->original_log_level
);
4387 m
->log_level_overridden
= false;
4390 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4393 if (!m
->log_target_overridden
) {
4394 m
->original_log_target
= log_get_target();
4395 m
->log_target_overridden
= true;
4398 log_info("Setting log target to %s.", log_target_to_string(target
));
4399 log_set_target(target
);
4402 void manager_restore_original_log_target(Manager
*m
) {
4405 if (!m
->log_target_overridden
)
4408 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4410 log_set_target(m
->original_log_target
);
4411 m
->log_target_overridden
= false;
4414 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4416 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4417 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4418 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4422 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4423 [MANAGER_INITIALIZING
] = "initializing",
4424 [MANAGER_STARTING
] = "starting",
4425 [MANAGER_RUNNING
] = "running",
4426 [MANAGER_DEGRADED
] = "degraded",
4427 [MANAGER_MAINTENANCE
] = "maintenance",
4428 [MANAGER_STOPPING
] = "stopping",
4431 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4433 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4434 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4435 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4436 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4437 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4438 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4439 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4440 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4441 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4442 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4443 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4444 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4445 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4446 [MANAGER_TIMESTAMP_UNITS_LOAD
] = "units-load",
4447 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4448 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4449 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4450 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4451 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4452 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4455 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4457 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4458 [OOM_CONTINUE
] = "continue",
4459 [OOM_STOP
] = "stop",
4460 [OOM_KILL
] = "kill",
4463 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);