1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
10 #include <sys/reboot.h>
11 #include <sys/timerfd.h>
12 #include <sys/utsname.h>
20 #include "sd-daemon.h"
21 #include "sd-messages.h"
24 #include "all-units.h"
25 #include "alloc-util.h"
27 #include "boot-timestamps.h"
28 #include "bus-common-errors.h"
29 #include "bus-error.h"
30 #include "bus-kernel.h"
32 #include "clean-ipc.h"
33 #include "clock-util.h"
34 #include "common-signal.h"
35 #include "constants.h"
36 #include "core-varlink.h"
37 #include "creds-util.h"
39 #include "dbus-manager.h"
40 #include "dbus-unit.h"
42 #include "dirent-util.h"
45 #include "event-util.h"
46 #include "exec-util.h"
48 #include "exit-status.h"
51 #include "generator-setup.h"
53 #include "initrd-util.h"
54 #include "inotify-util.h"
57 #include "label-util.h"
58 #include "load-fragment.h"
59 #include "locale-setup.h"
63 #include "manager-dump.h"
64 #include "manager-serialize.h"
65 #include "memory-util.h"
66 #include "mkdir-label.h"
67 #include "mount-util.h"
69 #include "parse-util.h"
70 #include "path-lookup.h"
71 #include "path-util.h"
72 #include "process-util.h"
74 #include "ratelimit.h"
75 #include "rlimit-util.h"
77 #include "selinux-util.h"
78 #include "signal-util.h"
79 #include "socket-util.h"
81 #include "stat-util.h"
82 #include "string-table.h"
83 #include "string-util.h"
86 #include "sysctl-util.h"
87 #include "syslog-util.h"
88 #include "terminal-util.h"
89 #include "time-util.h"
90 #include "transaction.h"
91 #include "uid-range.h"
92 #include "umask-util.h"
93 #include "unit-name.h"
94 #include "user-util.h"
98 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
99 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
101 /* Initial delay and the interval for printing status messages about running jobs */
102 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
103 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
104 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
105 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
107 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
108 * the queue gets more empty. */
109 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
111 /* How many units and jobs to process of the bus queue before returning to the event loop. */
112 #define MANAGER_BUS_MESSAGE_BUDGET 100U
114 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
115 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
116 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
117 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
118 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
119 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
120 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
121 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
122 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
123 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
124 static int manager_run_environment_generators(Manager
*m
);
125 static int manager_run_generators(Manager
*m
);
126 static void manager_vacuum(Manager
*m
);
128 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
131 if (MANAGER_IS_USER(m
))
132 /* Let the user manager without a timeout show status quickly, so the system manager can make
133 * use of it, if it wants to. */
134 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
* 2 / 3;
135 else if (show_status_on(m
->show_status
))
136 /* When status is on, just use the usual timeout. */
137 timeout
= JOBS_IN_PROGRESS_WAIT_USEC
;
139 timeout
= JOBS_IN_PROGRESS_QUIET_WAIT_USEC
;
141 return usec_add(now(CLOCK_MONOTONIC
), timeout
);
144 static void manager_watch_jobs_in_progress(Manager
*m
) {
150 /* We do not want to show the cylon animation if the user
151 * needs to confirm service executions otherwise confirmation
152 * messages will be screwed by the cylon animation. */
153 if (!manager_is_confirm_spawn_disabled(m
))
156 if (m
->jobs_in_progress_event_source
)
159 next
= manager_watch_jobs_next_time(m
);
160 r
= sd_event_add_time(
162 &m
->jobs_in_progress_event_source
,
165 manager_dispatch_jobs_in_progress
, m
);
169 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
172 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
174 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
177 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
178 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
182 p
= mempset(p
, ' ', pos
-2);
183 if (log_get_show_color())
184 p
= stpcpy(p
, ANSI_RED
);
188 if (pos
> 0 && pos
<= width
) {
189 if (log_get_show_color())
190 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
194 if (log_get_show_color())
195 p
= stpcpy(p
, ANSI_NORMAL
);
198 if (log_get_show_color())
199 p
= stpcpy(p
, ANSI_RED
);
202 p
= mempset(p
, ' ', width
-1-pos
);
203 if (log_get_show_color())
204 strcpy(p
, ANSI_NORMAL
);
208 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
212 if (m
->show_status
== SHOW_STATUS_AUTO
)
213 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
215 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
216 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
220 static void manager_print_jobs_in_progress(Manager
*m
) {
222 unsigned counter
= 0, print_nr
;
223 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
225 uint64_t timeout
= 0;
228 assert(m
->n_running_jobs
> 0);
230 manager_flip_auto_status(m
, true, "delay");
232 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
234 HASHMAP_FOREACH(j
, m
->jobs
)
235 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
238 /* m->n_running_jobs must be consistent with the contents of m->jobs,
239 * so the above loop must have succeeded in finding j. */
240 assert(counter
== print_nr
+ 1);
243 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
245 cylon_pos
= 14 - cylon_pos
;
246 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
248 m
->jobs_in_progress_iteration
++;
250 char job_of_n
[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
251 if (m
->n_running_jobs
> 1)
252 xsprintf(job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
);
254 (void) job_get_timeout(j
, &timeout
);
256 /* We want to use enough information for the user to identify previous lines talking about the same
257 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
258 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
259 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
260 * second argument to unit_status_string(). */
261 const char *ident
= unit_status_string(j
->unit
, NULL
);
263 const char *time
= FORMAT_TIMESPAN(now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
264 const char *limit
= timeout
> 0 ? FORMAT_TIMESPAN(timeout
- j
->begin_usec
, 1*USEC_PER_SEC
) : "no limit";
266 if (m
->status_unit_format
== STATUS_UNIT_FORMAT_DESCRIPTION
)
267 /* When using 'Description', we effectively don't have enough space to show the nested status
268 * without ellipsization, so let's not even try. */
269 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
270 "%sA %s job is running for %s (%s / %s)",
272 job_type_to_string(j
->type
),
276 const char *status_text
= unit_status_text(j
->unit
);
278 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
279 "%sJob %s/%s running (%s / %s)%s%s",
282 job_type_to_string(j
->type
),
284 status_text
? ": " : "",
285 strempty(status_text
));
289 "STATUS=%sUser job %s/%s running (%s / %s)...",
292 job_type_to_string(j
->type
),
294 m
->status_ready
= false;
297 static int have_ask_password(void) {
298 _cleanup_closedir_
DIR *dir
= NULL
;
300 dir
= opendir("/run/systemd/ask-password");
308 FOREACH_DIRENT_ALL(de
, dir
, return -errno
)
309 if (startswith(de
->d_name
, "ask."))
314 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
315 int fd
, uint32_t revents
, void *userdata
) {
316 Manager
*m
= ASSERT_PTR(userdata
);
320 m
->have_ask_password
= have_ask_password();
321 if (m
->have_ask_password
< 0)
322 /* Log error but continue. Negative have_ask_password
323 * is treated as unknown status. */
324 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
329 static void manager_close_ask_password(Manager
*m
) {
332 m
->ask_password_event_source
= sd_event_source_disable_unref(m
->ask_password_event_source
);
333 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
334 m
->have_ask_password
= -EINVAL
;
337 static int manager_check_ask_password(Manager
*m
) {
342 if (!m
->ask_password_event_source
) {
343 assert(m
->ask_password_inotify_fd
< 0);
345 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
347 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
348 if (m
->ask_password_inotify_fd
< 0)
349 return log_error_errno(errno
, "Failed to create inotify object: %m");
351 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
352 "/run/systemd/ask-password",
353 IN_CREATE
|IN_DELETE
|IN_MOVE
);
355 manager_close_ask_password(m
);
359 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
360 m
->ask_password_inotify_fd
, EPOLLIN
,
361 manager_dispatch_ask_password_fd
, m
);
363 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
364 manager_close_ask_password(m
);
368 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
370 /* Queries might have been added meanwhile... */
371 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
372 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
375 return m
->have_ask_password
;
378 static int manager_watch_idle_pipe(Manager
*m
) {
383 if (m
->idle_pipe_event_source
)
386 if (m
->idle_pipe
[2] < 0)
389 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
391 return log_error_errno(r
, "Failed to watch idle pipe: %m");
393 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
398 static void manager_close_idle_pipe(Manager
*m
) {
401 m
->idle_pipe_event_source
= sd_event_source_disable_unref(m
->idle_pipe_event_source
);
403 safe_close_pair(m
->idle_pipe
);
404 safe_close_pair(m
->idle_pipe
+ 2);
407 static int manager_setup_time_change(Manager
*m
) {
412 if (MANAGER_IS_TEST_RUN(m
))
415 m
->time_change_event_source
= sd_event_source_disable_unref(m
->time_change_event_source
);
417 r
= event_add_time_change(m
->event
, &m
->time_change_event_source
, 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 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
431 static int manager_read_timezone_stat(Manager
*m
) {
437 /* Read the current stat() data of /etc/localtime so that we detect changes */
438 if (lstat("/etc/localtime", &st
) < 0) {
439 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
440 changed
= m
->etc_localtime_accessible
;
441 m
->etc_localtime_accessible
= false;
445 k
= timespec_load(&st
.st_mtim
);
446 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
448 m
->etc_localtime_mtime
= k
;
449 m
->etc_localtime_accessible
= true;
455 static int manager_setup_timezone_change(Manager
*m
) {
456 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
461 if (MANAGER_IS_TEST_RUN(m
))
464 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
465 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
466 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
467 * went to zero and all fds to it are closed.
469 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
472 * Note that we create the new event source first here, before releasing the old one. This should optimize
473 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
475 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
476 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
478 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
479 * O_CREATE or by rename() */
481 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
482 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
483 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
486 return log_error_errno(r
, "Failed to create timezone change event source: %m");
488 /* Schedule this slightly earlier than the .timer event sources */
489 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
491 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
493 sd_event_source_unref(m
->timezone_change_event_source
);
494 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
499 static int enable_special_signals(Manager
*m
) {
500 _cleanup_close_
int fd
= -EBADF
;
504 if (MANAGER_IS_TEST_RUN(m
))
507 /* Enable that we get SIGINT on control-alt-del. In containers
508 * this will fail with EPERM (older) or EINVAL (newer), so
510 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
511 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
513 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
515 /* Support systems without virtual console */
517 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
519 /* Enable that we get SIGWINCH on kbrequest */
520 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
521 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
527 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
529 static int manager_setup_signals(Manager
*m
) {
530 struct sigaction sa
= {
531 .sa_handler
= SIG_DFL
,
532 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
539 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
541 /* We make liberal use of realtime signals here. On
542 * Linux/glibc we have 30 of them (with the exception of Linux
543 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
546 assert_se(sigemptyset(&mask
) == 0);
547 sigset_add_many(&mask
,
548 SIGCHLD
, /* Child died */
549 SIGTERM
, /* Reexecute daemon */
550 SIGHUP
, /* Reload configuration */
551 SIGUSR1
, /* systemd: reconnect to D-Bus */
552 SIGUSR2
, /* systemd: dump status */
553 SIGINT
, /* Kernel sends us this on control-alt-del */
554 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
555 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
557 SIGRTMIN
+0, /* systemd: start default.target */
558 SIGRTMIN
+1, /* systemd: isolate rescue.target */
559 SIGRTMIN
+2, /* systemd: isolate emergency.target */
560 SIGRTMIN
+3, /* systemd: start halt.target */
561 SIGRTMIN
+4, /* systemd: start poweroff.target */
562 SIGRTMIN
+5, /* systemd: start reboot.target */
563 SIGRTMIN
+6, /* systemd: start kexec.target */
564 SIGRTMIN
+7, /* systemd: start soft-reboot.target */
566 /* ... space for more special targets ... */
568 SIGRTMIN
+13, /* systemd: Immediate halt */
569 SIGRTMIN
+14, /* systemd: Immediate poweroff */
570 SIGRTMIN
+15, /* systemd: Immediate reboot */
571 SIGRTMIN
+16, /* systemd: Immediate kexec */
572 SIGRTMIN
+17, /* systemd: Immediate soft-reboot */
573 SIGRTMIN
+18, /* systemd: control command */
577 SIGRTMIN
+20, /* systemd: enable status messages */
578 SIGRTMIN
+21, /* systemd: disable status messages */
579 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
580 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
581 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
582 SIGRTMIN
+25, /* systemd: reexecute manager */
584 /* Apparently Linux on hppa had fewer RT signals until v3.18,
585 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
586 * see commit v3.17-7614-g1f25df2eff.
588 * We cannot unconditionally make use of those signals here,
589 * so let's use a runtime check. Since these commands are
590 * accessible by different means and only really a safety
591 * net, the missing functionality on hppa shouldn't matter.
594 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
595 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
596 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
597 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
599 /* ... one free signal here SIGRTMIN+30 ... */
601 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
603 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
604 if (m
->signal_fd
< 0)
607 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
611 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
613 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
614 * notify processing can still figure out to which process/service a message belongs, before we reap the
615 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
616 * status information before detecting that there's no process in a cgroup. */
617 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
621 if (MANAGER_IS_SYSTEM(m
))
622 return enable_special_signals(m
);
627 static char** sanitize_environment(char **l
) {
629 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
633 "CONFIGURATION_DIRECTORY",
634 "CREDENTIALS_DIRECTORY",
646 "MEMORY_PRESSURE_WATCH",
647 "MEMORY_PRESSURE_WRITE",
649 "MONITOR_EXIT_STATUS",
650 "MONITOR_INVOCATION_ID",
651 "MONITOR_SERVICE_RESULT",
662 "TRIGGER_TIMER_MONOTONIC_USEC",
663 "TRIGGER_TIMER_REALTIME_USEC",
669 /* Let's order the environment alphabetically, just to make it pretty */
673 int manager_default_environment(Manager
*m
) {
678 m
->transient_environment
= strv_free(m
->transient_environment
);
680 if (MANAGER_IS_SYSTEM(m
)) {
681 /* The system manager always starts with a clean environment for its children. It does not
682 * import the kernel's or the parents' exported variables.
684 * The initial passed environment is untouched to keep /proc/self/environ valid; it is used
685 * for tagging the init process inside containers. */
686 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
687 if (!m
->transient_environment
)
690 /* Import locale variables LC_*= from configuration */
691 (void) locale_setup(&m
->transient_environment
);
693 /* The user manager passes its own environment along to its children, except for $PATH. */
694 m
->transient_environment
= strv_copy(environ
);
695 if (!m
->transient_environment
)
698 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
703 sanitize_environment(m
->transient_environment
);
707 static int manager_setup_prefix(Manager
*m
) {
713 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
714 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
715 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
716 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
717 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
718 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
721 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
722 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
723 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_STATE_PRIVATE
, NULL
},
724 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
725 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_STATE_PRIVATE
, "log" },
726 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
731 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
734 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
735 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
737 return log_warning_errno(r
, "Failed to lookup %s path: %m",
738 exec_directory_type_to_string(i
));
744 static void manager_free_unit_name_maps(Manager
*m
) {
745 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
746 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
747 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
748 m
->unit_cache_timestamp_hash
= 0;
751 static int manager_setup_run_queue(Manager
*m
) {
755 assert(!m
->run_queue_event_source
);
757 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
761 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
765 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
769 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
774 static int manager_setup_sigchld_event_source(Manager
*m
) {
778 assert(!m
->sigchld_event_source
);
780 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
784 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
788 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
792 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
797 int manager_setup_memory_pressure_event_source(Manager
*m
) {
802 m
->memory_pressure_event_source
= sd_event_source_disable_unref(m
->memory_pressure_event_source
);
804 r
= sd_event_add_memory_pressure(m
->event
, &m
->memory_pressure_event_source
, NULL
, NULL
);
806 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r
) || ERRNO_IS_PRIVILEGE(r
) || (r
== -EHOSTDOWN
) ? LOG_DEBUG
: LOG_NOTICE
, r
,
807 "Failed to establish memory pressure event source, ignoring: %m");
808 else if (m
->default_memory_pressure_threshold_usec
!= USEC_INFINITY
) {
810 /* If there's a default memory pressure threshold set, also apply it to the service manager itself */
811 r
= sd_event_source_set_memory_pressure_period(
812 m
->memory_pressure_event_source
,
813 m
->default_memory_pressure_threshold_usec
,
814 MEMORY_PRESSURE_DEFAULT_WINDOW_USEC
);
816 log_warning_errno(r
, "Failed to adjust memory pressure threshold, ignoring: %m");
822 static int manager_find_credentials_dirs(Manager
*m
) {
828 r
= get_credentials_dir(&e
);
831 log_debug_errno(r
, "Failed to determine credentials directory, ignoring: %m");
833 m
->received_credentials_directory
= strdup(e
);
834 if (!m
->received_credentials_directory
)
838 r
= get_encrypted_credentials_dir(&e
);
841 log_debug_errno(r
, "Failed to determine encrypted credentials directory, ignoring: %m");
843 m
->received_encrypted_credentials_directory
= strdup(e
);
844 if (!m
->received_encrypted_credentials_directory
)
851 void manager_set_switching_root(Manager
*m
, bool switching_root
) {
854 m
->switching_root
= MANAGER_IS_SYSTEM(m
) && switching_root
;
857 int manager_new(RuntimeScope runtime_scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
858 _cleanup_(manager_freep
) Manager
*m
= NULL
;
862 assert(IN_SET(runtime_scope
, RUNTIME_SCOPE_SYSTEM
, RUNTIME_SCOPE_USER
));
869 .runtime_scope
= runtime_scope
,
870 .objective
= _MANAGER_OBJECTIVE_INVALID
,
872 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
874 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
875 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
876 .default_tasks_accounting
= true,
877 .default_tasks_max
= TASKS_MAX_UNSET
,
878 .default_timeout_start_usec
= manager_default_timeout(runtime_scope
),
879 .default_timeout_stop_usec
= manager_default_timeout(runtime_scope
),
880 .default_restart_usec
= DEFAULT_RESTART_USEC
,
881 .default_device_timeout_usec
= manager_default_timeout(runtime_scope
),
883 .original_log_level
= -1,
884 .original_log_target
= _LOG_TARGET_INVALID
,
886 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
887 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
888 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
889 .watchdog_overridden
[WATCHDOG_PRETIMEOUT
] = USEC_INFINITY
,
891 .show_status_overridden
= _SHOW_STATUS_INVALID
,
894 .cgroups_agent_fd
= -EBADF
,
896 .user_lookup_fds
= PIPE_EBADF
,
897 .private_listen_fd
= -EBADF
,
898 .dev_autofs_fd
= -EBADF
,
899 .cgroup_inotify_fd
= -EBADF
,
900 .pin_cgroupfs_fd
= -EBADF
,
901 .ask_password_inotify_fd
= -EBADF
,
902 .idle_pipe
= { -EBADF
, -EBADF
, -EBADF
, -EBADF
},
904 /* start as id #1, so that we can leave #0 around as "null-like" value */
907 .have_ask_password
= -EINVAL
, /* we don't know */
909 .test_run_flags
= test_run_flags
,
911 .default_oom_policy
= OOM_STOP
,
913 .default_memory_pressure_watch
= CGROUP_PRESSURE_WATCH_AUTO
,
914 .default_memory_pressure_threshold_usec
= USEC_INFINITY
,
917 .interval
= 10 * USEC_PER_MINUTE
,
923 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
924 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
925 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
926 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
929 /* Prepare log fields we can use for structured logging */
930 if (MANAGER_IS_SYSTEM(m
)) {
931 m
->unit_log_field
= "UNIT=";
932 m
->unit_log_format_string
= "UNIT=%s";
934 m
->invocation_log_field
= "INVOCATION_ID=";
935 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
937 m
->unit_log_field
= "USER_UNIT=";
938 m
->unit_log_format_string
= "USER_UNIT=%s";
940 m
->invocation_log_field
= "USER_INVOCATION_ID=";
941 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
944 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
945 m
->ctrl_alt_del_ratelimit
= (const RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
947 r
= manager_default_environment(m
);
951 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
955 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
959 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
963 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
967 r
= manager_setup_prefix(m
);
971 r
= manager_find_credentials_dirs(m
);
975 r
= sd_event_default(&m
->event
);
979 r
= manager_setup_run_queue(m
);
983 if (FLAGS_SET(test_run_flags
, MANAGER_TEST_RUN_MINIMAL
)) {
984 m
->cgroup_root
= strdup("");
988 r
= manager_setup_signals(m
);
992 r
= manager_setup_cgroup(m
);
996 r
= manager_setup_time_change(m
);
1000 r
= manager_read_timezone_stat(m
);
1004 (void) manager_setup_timezone_change(m
);
1006 r
= manager_setup_sigchld_event_source(m
);
1010 r
= manager_setup_memory_pressure_event_source(m
);
1015 if (MANAGER_IS_SYSTEM(m
) && lsm_bpf_supported(/* initialize = */ true)) {
1016 r
= lsm_bpf_setup(m
);
1018 log_warning_errno(r
, "Failed to setup LSM BPF, ignoring: %m");
1023 if (test_run_flags
== 0) {
1024 if (MANAGER_IS_SYSTEM(m
))
1025 r
= mkdir_label("/run/systemd/units", 0755);
1027 _cleanup_free_
char *units_path
= NULL
;
1028 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
1031 r
= mkdir_p_label(units_path
, 0755);
1034 if (r
< 0 && r
!= -EEXIST
)
1040 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
1042 /* Note that we do not set up the notify fd here. We do that after deserialization,
1043 * since they might have gotten serialized across the reexec. */
1050 static int manager_setup_notify(Manager
*m
) {
1053 if (MANAGER_IS_TEST_RUN(m
))
1056 if (m
->notify_fd
< 0) {
1057 _cleanup_close_
int fd
= -EBADF
;
1058 union sockaddr_union sa
;
1061 /* First free all secondary fields */
1062 m
->notify_socket
= mfree(m
->notify_socket
);
1063 m
->notify_event_source
= sd_event_source_disable_unref(m
->notify_event_source
);
1065 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1067 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
1069 fd_increase_rxbuf(fd
, NOTIFY_RCVBUF_SIZE
);
1071 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
1072 if (!m
->notify_socket
)
1075 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
1077 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1081 (void) mkdir_parents_label(m
->notify_socket
, 0755);
1082 (void) sockaddr_un_unlink(&sa
.un
);
1084 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
1086 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
1088 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1090 return log_error_errno(r
, "SO_PASSCRED failed: %m");
1092 m
->notify_fd
= TAKE_FD(fd
);
1094 log_debug("Using notification socket %s", m
->notify_socket
);
1097 if (!m
->notify_event_source
) {
1098 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
1100 return log_error_errno(r
, "Failed to allocate notify event source: %m");
1102 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1103 * service an exit message belongs. */
1104 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
1106 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
1108 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
1114 static int manager_setup_cgroups_agent(Manager
*m
) {
1116 static const union sockaddr_union sa
= {
1117 .un
.sun_family
= AF_UNIX
,
1118 .un
.sun_path
= "/run/systemd/cgroups-agent",
1122 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1123 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1124 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1125 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1126 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1127 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1128 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1129 * we thus won't lose messages.
1131 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1132 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1133 * bus for these messages. */
1135 if (MANAGER_IS_TEST_RUN(m
))
1138 if (!MANAGER_IS_SYSTEM(m
))
1141 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1143 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1144 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1147 if (m
->cgroups_agent_fd
< 0) {
1148 _cleanup_close_
int fd
= -EBADF
;
1150 /* First free all secondary fields */
1151 m
->cgroups_agent_event_source
= sd_event_source_disable_unref(m
->cgroups_agent_event_source
);
1153 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1155 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1157 fd_increase_rxbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1159 (void) sockaddr_un_unlink(&sa
.un
);
1161 /* Only allow root to connect to this socket */
1163 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1165 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1167 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1170 if (!m
->cgroups_agent_event_source
) {
1171 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1173 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1175 /* Process cgroups notifications early. Note that when the agent notification is received
1176 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1177 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1178 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1180 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1182 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1188 static int manager_setup_user_lookup_fd(Manager
*m
) {
1193 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1194 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1195 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1196 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1197 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1198 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1199 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1200 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1203 * You might wonder why we need a communication channel for this that is independent of the usual notification
1204 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1205 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1206 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1208 * Note that this function is called under two circumstances: when we first initialize (in which case we
1209 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1210 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1212 if (m
->user_lookup_fds
[0] < 0) {
1214 /* Free all secondary fields */
1215 safe_close_pair(m
->user_lookup_fds
);
1216 m
->user_lookup_event_source
= sd_event_source_disable_unref(m
->user_lookup_event_source
);
1218 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1219 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1221 (void) fd_increase_rxbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1224 if (!m
->user_lookup_event_source
) {
1225 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1227 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1229 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1231 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1233 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1235 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1241 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1247 while ((u
= m
->cleanup_queue
)) {
1248 assert(u
->in_cleanup_queue
);
1257 static unsigned manager_dispatch_release_resources_queue(Manager
*m
) {
1263 while ((u
= m
->release_resources_queue
)) {
1264 assert(u
->in_release_resources_queue
);
1266 LIST_REMOVE(release_resources_queue
, m
->release_resources_queue
, u
);
1267 u
->in_release_resources_queue
= false;
1271 unit_release_resources(u
);
1278 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1279 GC_OFFSET_UNSURE
, /* No clue */
1280 GC_OFFSET_GOOD
, /* We still need this unit */
1281 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1285 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1288 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1290 /* Recursively mark referenced units as GOOD as well */
1291 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1292 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1293 unit_gc_mark_good(other
, gc_marker
);
1296 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1302 if (IN_SET(u
->gc_marker
- gc_marker
,
1303 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1306 if (u
->in_cleanup_queue
)
1309 if (!unit_may_gc(u
))
1312 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1316 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1317 unit_gc_sweep(other
, gc_marker
);
1319 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1322 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1326 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1327 unit_gc_sweep(ref
->source
, gc_marker
);
1329 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1332 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1339 /* We were unable to find anything out about this entry, so
1340 * let's investigate it later */
1341 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1342 unit_add_to_gc_queue(u
);
1346 /* We definitely know that this one is not useful anymore, so
1347 * let's mark it for deletion */
1348 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1349 unit_add_to_cleanup_queue(u
);
1353 unit_gc_mark_good(u
, gc_marker
);
1356 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1357 unsigned n
= 0, gc_marker
;
1362 /* log_debug("Running GC..."); */
1364 m
->gc_marker
+= _GC_OFFSET_MAX
;
1365 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1368 gc_marker
= m
->gc_marker
;
1370 while ((u
= m
->gc_unit_queue
)) {
1371 assert(u
->in_gc_queue
);
1373 unit_gc_sweep(u
, gc_marker
);
1375 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1376 u
->in_gc_queue
= false;
1380 if (IN_SET(u
->gc_marker
- gc_marker
,
1381 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1383 log_unit_debug(u
, "Collecting.");
1384 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1385 unit_add_to_cleanup_queue(u
);
1392 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1398 while ((j
= m
->gc_job_queue
)) {
1399 assert(j
->in_gc_queue
);
1401 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1402 j
->in_gc_queue
= false;
1409 log_unit_debug(j
->unit
, "Collecting job.");
1410 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1416 static int manager_ratelimit_requeue(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
1420 assert(s
== u
->auto_start_stop_event_source
);
1422 u
->auto_start_stop_event_source
= sd_event_source_unref(u
->auto_start_stop_event_source
);
1424 /* Re-queue to all queues, if the rate limit hit we might have been throttled on any of them. */
1425 unit_submit_to_stop_when_unneeded_queue(u
);
1426 unit_submit_to_start_when_upheld_queue(u
);
1427 unit_submit_to_stop_when_bound_queue(u
);
1432 static int manager_ratelimit_check_and_queue(Unit
*u
) {
1437 if (ratelimit_below(&u
->auto_start_stop_ratelimit
))
1440 /* Already queued, no need to requeue */
1441 if (u
->auto_start_stop_event_source
)
1444 r
= sd_event_add_time(
1446 &u
->auto_start_stop_event_source
,
1448 ratelimit_end(&u
->auto_start_stop_ratelimit
),
1450 manager_ratelimit_requeue
,
1453 return log_unit_error_errno(u
, r
, "Failed to queue timer on event loop: %m");
1458 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1465 while ((u
= m
->stop_when_unneeded_queue
)) {
1466 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1468 assert(u
->in_stop_when_unneeded_queue
);
1469 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1470 u
->in_stop_when_unneeded_queue
= false;
1474 if (!unit_is_unneeded(u
))
1477 log_unit_debug(u
, "Unit is not needed anymore.");
1479 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1480 * service being unnecessary after a while. */
1482 r
= manager_ratelimit_check_and_queue(u
);
1485 "Unit not needed anymore, but not stopping since we tried this too often recently.%s",
1486 r
== 0 ? " Will retry later." : "");
1490 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1491 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1493 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1499 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1506 while ((u
= m
->start_when_upheld_queue
)) {
1507 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1508 Unit
*culprit
= NULL
;
1510 assert(u
->in_start_when_upheld_queue
);
1511 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1512 u
->in_start_when_upheld_queue
= false;
1516 if (!unit_is_upheld_by_active(u
, &culprit
))
1519 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1521 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1522 * service being unnecessary after a while. */
1524 r
= manager_ratelimit_check_and_queue(u
);
1527 "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.%s",
1529 r
== 0 ? " Will retry later." : "");
1533 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1535 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1541 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1548 while ((u
= m
->stop_when_bound_queue
)) {
1549 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1550 Unit
*culprit
= NULL
;
1552 assert(u
->in_stop_when_bound_queue
);
1553 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1554 u
->in_stop_when_bound_queue
= false;
1558 if (!unit_is_bound_by_inactive(u
, &culprit
))
1561 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1563 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1564 * service being unnecessary after a while. */
1566 r
= manager_ratelimit_check_and_queue(u
);
1569 "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.%s",
1571 r
== 0 ? " Will retry later." : "");
1575 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1577 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1583 static void manager_clear_jobs_and_units(Manager
*m
) {
1588 while ((u
= hashmap_first(m
->units
)))
1591 manager_dispatch_cleanup_queue(m
);
1593 assert(!m
->load_queue
);
1594 assert(prioq_isempty(m
->run_queue
));
1595 assert(!m
->dbus_unit_queue
);
1596 assert(!m
->dbus_job_queue
);
1597 assert(!m
->cleanup_queue
);
1598 assert(!m
->gc_unit_queue
);
1599 assert(!m
->gc_job_queue
);
1600 assert(!m
->cgroup_realize_queue
);
1601 assert(!m
->cgroup_empty_queue
);
1602 assert(!m
->cgroup_oom_queue
);
1603 assert(!m
->target_deps_queue
);
1604 assert(!m
->stop_when_unneeded_queue
);
1605 assert(!m
->start_when_upheld_queue
);
1606 assert(!m
->stop_when_bound_queue
);
1607 assert(!m
->release_resources_queue
);
1609 assert(hashmap_isempty(m
->jobs
));
1610 assert(hashmap_isempty(m
->units
));
1612 m
->n_on_console
= 0;
1613 m
->n_running_jobs
= 0;
1614 m
->n_installed_jobs
= 0;
1615 m
->n_failed_jobs
= 0;
1618 Manager
* manager_free(Manager
*m
) {
1622 manager_clear_jobs_and_units(m
);
1624 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1625 if (unit_vtable
[c
]->shutdown
)
1626 unit_vtable
[c
]->shutdown(m
);
1628 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1629 manager_shutdown_cgroup(m
, /* delete= */ IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1631 lookup_paths_flush_generator(&m
->lookup_paths
);
1634 manager_varlink_done(m
);
1636 exec_shared_runtime_vacuum(m
);
1637 hashmap_free(m
->exec_shared_runtime_by_id
);
1639 dynamic_user_vacuum(m
, false);
1640 hashmap_free(m
->dynamic_users
);
1642 hashmap_free(m
->units
);
1643 hashmap_free(m
->units_by_invocation_id
);
1644 hashmap_free(m
->jobs
);
1645 hashmap_free(m
->watch_pids
);
1646 hashmap_free(m
->watch_bus
);
1648 prioq_free(m
->run_queue
);
1650 set_free(m
->startup_units
);
1651 set_free(m
->failed_units
);
1653 sd_event_source_unref(m
->signal_event_source
);
1654 sd_event_source_unref(m
->sigchld_event_source
);
1655 sd_event_source_unref(m
->notify_event_source
);
1656 sd_event_source_unref(m
->cgroups_agent_event_source
);
1657 sd_event_source_unref(m
->time_change_event_source
);
1658 sd_event_source_unref(m
->timezone_change_event_source
);
1659 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1660 sd_event_source_unref(m
->run_queue_event_source
);
1661 sd_event_source_unref(m
->user_lookup_event_source
);
1662 sd_event_source_unref(m
->memory_pressure_event_source
);
1664 safe_close(m
->signal_fd
);
1665 safe_close(m
->notify_fd
);
1666 safe_close(m
->cgroups_agent_fd
);
1667 safe_close_pair(m
->user_lookup_fds
);
1669 manager_close_ask_password(m
);
1671 manager_close_idle_pipe(m
);
1673 sd_event_unref(m
->event
);
1675 free(m
->notify_socket
);
1677 lookup_paths_free(&m
->lookup_paths
);
1678 strv_free(m
->transient_environment
);
1679 strv_free(m
->client_environment
);
1681 hashmap_free(m
->cgroup_unit
);
1682 manager_free_unit_name_maps(m
);
1684 free(m
->switch_root
);
1685 free(m
->switch_root_init
);
1687 free(m
->default_smack_process_label
);
1689 rlimit_free_all(m
->rlimit
);
1691 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1692 hashmap_free(m
->units_requiring_mounts_for
);
1694 hashmap_free(m
->uid_refs
);
1695 hashmap_free(m
->gid_refs
);
1697 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1698 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1699 free(m
->received_credentials_directory
);
1700 free(m
->received_encrypted_credentials_directory
);
1702 free(m
->watchdog_pretimeout_governor
);
1703 free(m
->watchdog_pretimeout_governor_overridden
);
1706 lsm_bpf_destroy(m
->restrict_fs
);
1712 static void manager_enumerate_perpetual(Manager
*m
) {
1715 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1718 /* Let's ask every type to load all units from disk/kernel that it might know */
1719 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1720 if (!unit_type_supported(c
)) {
1721 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1725 if (unit_vtable
[c
]->enumerate_perpetual
)
1726 unit_vtable
[c
]->enumerate_perpetual(m
);
1730 static void manager_enumerate(Manager
*m
) {
1733 if (FLAGS_SET(m
->test_run_flags
, MANAGER_TEST_RUN_MINIMAL
))
1736 /* Let's ask every type to load all units from disk/kernel that it might know */
1737 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1738 if (!unit_type_supported(c
)) {
1739 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1743 if (unit_vtable
[c
]->enumerate
)
1744 unit_vtable
[c
]->enumerate(m
);
1747 manager_dispatch_load_queue(m
);
1750 static void manager_coldplug(Manager
*m
) {
1757 log_debug("Invoking unit coldplug() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1759 /* Let's place the units back into their deserialized state */
1760 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1762 /* ignore aliases */
1766 r
= unit_coldplug(u
);
1768 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1772 static void manager_catchup(Manager
*m
) {
1778 log_debug("Invoking unit catchup() handlers%s", special_glyph(SPECIAL_GLYPH_ELLIPSIS
));
1780 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1781 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1783 /* ignore aliases */
1791 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1796 HASHMAP_FOREACH(u
, m
->units
) {
1798 if (fdset_size(fds
) <= 0)
1801 if (!UNIT_VTABLE(u
)->distribute_fds
)
1804 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1808 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1813 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1814 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1815 * rather than the current one. */
1817 if (MANAGER_IS_TEST_RUN(m
))
1820 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1823 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1826 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1829 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
),
1832 SERVICE_RELOAD_NOTIFY
,
1833 SERVICE_RELOAD_SIGNAL
))
1839 static void manager_setup_bus(Manager
*m
) {
1842 /* Let's set up our private bus connection now, unconditionally */
1843 (void) bus_init_private(m
);
1845 /* If we are in --user mode also connect to the system bus now */
1846 if (MANAGER_IS_USER(m
))
1847 (void) bus_init_system(m
);
1849 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1850 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1851 (void) bus_init_api(m
);
1853 if (MANAGER_IS_SYSTEM(m
))
1854 (void) bus_init_system(m
);
1858 static void manager_preset_all(Manager
*m
) {
1863 if (m
->first_boot
<= 0)
1866 if (!MANAGER_IS_SYSTEM(m
))
1869 if (MANAGER_IS_TEST_RUN(m
))
1872 /* If this is the first boot, and we are in the host system, then preset everything */
1873 UnitFilePresetMode mode
= FIRST_BOOT_FULL_PRESET
? UNIT_FILE_PRESET_FULL
: UNIT_FILE_PRESET_ENABLE_ONLY
;
1875 r
= unit_file_preset_all(RUNTIME_SCOPE_SYSTEM
, 0, NULL
, mode
, NULL
, 0);
1877 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1878 "Failed to populate /etc with preset unit settings, ignoring: %m");
1880 log_info("Populated /etc with preset unit settings.");
1883 static void manager_ready(Manager
*m
) {
1886 /* After having loaded everything, do the final round of catching up with what might have changed */
1888 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1890 /* It might be safe to log to the journal now and connect to dbus */
1891 manager_recheck_journal(m
);
1892 manager_recheck_dbus(m
);
1894 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1897 /* Create a file which will indicate when the manager started loading units the last time. */
1898 if (MANAGER_IS_SYSTEM(m
))
1899 (void) touch_file("/run/systemd/systemd-units-load", false,
1900 m
->timestamps
[MANAGER_TIMESTAMP_UNITS_LOAD
].realtime
?: now(CLOCK_REALTIME
),
1901 UID_INVALID
, GID_INVALID
, 0444);
1904 Manager
* manager_reloading_start(Manager
*m
) {
1906 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_UNITS_LOAD
);
1910 void manager_reloading_stopp(Manager
**m
) {
1912 assert((*m
)->n_reloading
> 0);
1913 (*m
)->n_reloading
--;
1917 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
, const char *root
) {
1922 /* If we are running in test mode, we still want to run the generators,
1923 * but we should not touch the real generator directories. */
1924 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
,
1925 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1930 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1931 r
= manager_run_environment_generators(m
);
1933 r
= manager_run_generators(m
);
1934 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1938 manager_preset_all(m
);
1940 lookup_paths_log(&m
->lookup_paths
);
1943 /* This block is (optionally) done with the reloading counter bumped */
1944 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1946 /* Make sure we don't have a left-over from a previous run */
1948 (void) rm_rf(m
->lookup_paths
.transient
, 0);
1950 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1951 * counter here already */
1953 reloading
= manager_reloading_start(m
);
1955 /* First, enumerate what we can from all config files */
1956 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1957 manager_enumerate_perpetual(m
);
1958 manager_enumerate(m
);
1959 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1961 /* Second, deserialize if there is something to deserialize */
1962 if (serialization
) {
1963 r
= manager_deserialize(m
, serialization
, fds
);
1965 return log_error_errno(r
, "Deserialization failed: %m");
1968 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1969 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1971 manager_distribute_fds(m
, fds
);
1973 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1974 r
= manager_setup_notify(m
);
1976 /* No sense to continue without notifications, our children would fail anyway. */
1979 r
= manager_setup_cgroups_agent(m
);
1981 /* Likewise, no sense to continue without empty cgroup notifications. */
1984 r
= manager_setup_user_lookup_fd(m
);
1986 /* This shouldn't fail, except if things are really broken. */
1989 /* Connect to the bus if we are good for it */
1990 manager_setup_bus(m
);
1992 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1993 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1995 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1996 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1998 r
= manager_varlink_init(m
);
2000 log_warning_errno(r
, "Failed to set up Varlink, ignoring: %m");
2002 /* Third, fire things up! */
2003 manager_coldplug(m
);
2005 /* Clean up runtime objects */
2009 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
2010 * reload is finished */
2011 m
->send_reloading_done
= true;
2016 manager_set_switching_root(m
, false);
2021 int manager_add_job(
2027 sd_bus_error
*error
,
2030 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2034 assert(type
< _JOB_TYPE_MAX
);
2036 assert(mode
< _JOB_MODE_MAX
);
2038 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
2039 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
2041 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
2042 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
2044 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
2045 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
2047 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
2049 type
= job_type_collapse(type
, unit
);
2051 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2055 r
= transaction_add_job_and_dependencies(
2060 TRANSACTION_MATTERS
|
2061 (IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
) ? TRANSACTION_IGNORE_REQUIREMENTS
: 0) |
2062 (mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0),
2067 if (mode
== JOB_ISOLATE
) {
2068 r
= transaction_add_isolate_jobs(tr
, m
);
2073 if (mode
== JOB_TRIGGERING
) {
2074 r
= transaction_add_triggering_jobs(tr
, unit
);
2079 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
2083 log_unit_debug(unit
,
2084 "Enqueued job %s/%s as %u", unit
->id
,
2085 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
2088 *ret
= tr
->anchor_job
;
2090 tr
= transaction_free(tr
);
2094 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
2095 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
2099 assert(type
< _JOB_TYPE_MAX
);
2101 assert(mode
< _JOB_MODE_MAX
);
2103 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
2108 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
2111 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
2112 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2116 assert(type
< _JOB_TYPE_MAX
);
2118 assert(mode
< _JOB_MODE_MAX
);
2120 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
2122 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
2127 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
2129 _cleanup_(transaction_abort_and_freep
) Transaction
*tr
= NULL
;
2133 assert(mode
< _JOB_MODE_MAX
);
2134 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
2136 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
2140 /* We need an anchor job */
2141 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, TRANSACTION_IGNORE_REQUIREMENTS
|TRANSACTION_IGNORE_ORDER
, e
);
2145 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
2146 transaction_add_propagate_reload_jobs(
2150 mode
== JOB_IGNORE_DEPENDENCIES
? TRANSACTION_IGNORE_ORDER
: 0);
2152 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
2156 tr
= transaction_free(tr
);
2160 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
2163 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
2166 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
2170 return hashmap_get(m
->units
, name
);
2173 static int manager_dispatch_target_deps_queue(Manager
*m
) {
2179 while ((u
= m
->target_deps_queue
)) {
2180 _cleanup_free_ Unit
**targets
= NULL
;
2183 assert(u
->in_target_deps_queue
);
2185 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
2186 u
->in_target_deps_queue
= false;
2188 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2189 * dependencies, and we can't have it that hash tables we iterate through are modified while
2190 * we are iterating through them. */
2191 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
2195 for (int i
= 0; i
< n_targets
; i
++) {
2196 r
= unit_add_default_target_dependency(u
, targets
[i
]);
2205 unsigned manager_dispatch_load_queue(Manager
*m
) {
2211 /* Make sure we are not run recursively */
2212 if (m
->dispatching_load_queue
)
2215 m
->dispatching_load_queue
= true;
2217 /* Dispatches the load queue. Takes a unit from the queue and
2218 * tries to load its data until the queue is empty */
2220 while ((u
= m
->load_queue
)) {
2221 assert(u
->in_load_queue
);
2227 m
->dispatching_load_queue
= false;
2229 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2230 * should be loaded and have aliases resolved */
2231 (void) manager_dispatch_target_deps_queue(m
);
2236 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2239 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2240 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2241 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2242 if (u
->load_state
!= UNIT_NOT_FOUND
)
2245 /* The cache has been updated since the last time we tried to load the unit. There might be new
2246 * fragment paths to read. */
2247 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2250 /* The cache needs to be updated because there are modifications on disk. */
2251 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2254 int manager_load_unit_prepare(
2261 _cleanup_(unit_freep
) Unit
*cleanup_unit
= NULL
;
2262 _cleanup_free_
char *nbuf
= NULL
;
2267 assert(name
|| path
);
2269 /* This will prepare the unit for loading, but not actually load anything from disk. */
2271 if (path
&& !path_is_absolute(path
))
2272 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2275 r
= path_extract_filename(path
, &nbuf
);
2278 if (r
== O_DIRECTORY
)
2279 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path '%s' refers to directory, refusing.", path
);
2284 UnitType t
= unit_name_to_type(name
);
2286 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2287 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2288 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2290 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2293 Unit
*unit
= manager_get_unit(m
, name
);
2295 /* The time-based cache allows to start new units without daemon-reload,
2296 * but if they are already referenced (because of dependencies or ordering)
2297 * then we have to force a load of the fragment. As an optimization, check
2298 * first if anything in the usual paths was modified since the last time
2299 * the cache was loaded. Also check if the last time an attempt to load the
2300 * unit was made was before the most recent cache refresh, so that we know
2301 * we need to try again — even if the cache is current, it might have been
2302 * updated in a different context before we had a chance to retry loading
2303 * this particular unit. */
2304 if (manager_unit_cache_should_retry_load(unit
))
2305 unit
->load_state
= UNIT_STUB
;
2308 return 0; /* The unit was already loaded */
2311 unit
= cleanup_unit
= unit_new(m
, unit_vtable
[t
]->object_size
);
2317 r
= free_and_strdup(&unit
->fragment_path
, path
);
2322 r
= unit_add_name(unit
, name
);
2326 unit_add_to_load_queue(unit
);
2327 unit_add_to_dbus_queue(unit
);
2328 unit_add_to_gc_queue(unit
);
2331 TAKE_PTR(cleanup_unit
);
2333 return 1; /* The unit was added the load queue */
2336 int manager_load_unit(
2347 /* This will load the unit config, but not actually start any services or anything. */
2349 r
= manager_load_unit_prepare(m
, name
, path
, e
, ret
);
2353 /* Unit was newly loaded */
2354 manager_dispatch_load_queue(m
);
2355 *ret
= unit_follow_merge(*ret
);
2359 int manager_load_startable_unit_or_warn(
2365 /* Load a unit, make sure it loaded fully and is not masked. */
2367 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2371 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2373 return log_error_errno(r
, "Failed to load %s %s: %s",
2374 name
? "unit" : "unit file", name
?: path
,
2375 bus_error_message(&error
, r
));
2377 r
= bus_unit_validate_load_state(unit
, &error
);
2379 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2385 void manager_clear_jobs(Manager
*m
) {
2390 while ((j
= hashmap_first(m
->jobs
)))
2391 /* No need to recurse. We're cancelling all jobs. */
2392 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2395 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2398 /* First let's drop the unit keyed as "pid". */
2399 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2401 /* Then, let's also drop the array keyed by -pid. */
2402 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2405 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2406 Manager
*m
= ASSERT_PTR(userdata
);
2411 while ((j
= prioq_peek(m
->run_queue
))) {
2412 assert(j
->installed
);
2413 assert(j
->in_run_queue
);
2415 (void) job_run_and_invalidate(j
);
2418 if (m
->n_running_jobs
> 0)
2419 manager_watch_jobs_in_progress(m
);
2421 if (m
->n_on_console
> 0)
2422 manager_watch_idle_pipe(m
);
2427 void manager_trigger_run_queue(Manager
*m
) {
2432 r
= sd_event_source_set_enabled(
2433 m
->run_queue_event_source
,
2434 prioq_isempty(m
->run_queue
) ? SD_EVENT_OFF
: SD_EVENT_ONESHOT
);
2436 log_warning_errno(r
, "Failed to enable job run queue event source, ignoring: %m");
2439 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2440 unsigned n
= 0, budget
;
2446 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2447 * as we can. There's no point in throttling generation of signals in that case. */
2448 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2449 budget
= UINT_MAX
; /* infinite budget in this case */
2451 /* Anything to do at all? */
2452 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2455 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2456 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2457 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2460 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2461 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2462 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2463 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2464 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2465 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2466 * connections it will be counted five times. This difference in counting ("references"
2467 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2468 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2469 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2470 * currently chosen much higher than the "budget". */
2471 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2474 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2476 assert(u
->in_dbus_queue
);
2478 bus_unit_send_change_signal(u
);
2481 if (budget
!= UINT_MAX
)
2485 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2486 assert(j
->in_dbus_queue
);
2488 bus_job_send_change_signal(j
);
2491 if (budget
!= UINT_MAX
)
2495 if (m
->send_reloading_done
) {
2496 m
->send_reloading_done
= false;
2497 bus_manager_send_reloading(m
, false);
2501 if (m
->pending_reload_message
) {
2502 bus_send_pending_reload_message(m
);
2509 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2510 Manager
*m
= userdata
;
2514 n
= recv(fd
, buf
, sizeof(buf
), 0);
2516 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2518 log_error("Got zero-length cgroups agent message, ignoring.");
2521 if ((size_t) n
>= sizeof(buf
)) {
2522 log_error("Got overly long cgroups agent message, ignoring.");
2526 if (memchr(buf
, 0, n
)) {
2527 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2532 manager_notify_cgroup_empty(m
, buf
);
2533 (void) bus_forward_agent_released(m
, buf
);
2538 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2540 /* nothing else must be sent when using BARRIER=1 */
2541 if (strv_contains(tags
, "BARRIER=1")) {
2542 if (strv_length(tags
) != 1)
2543 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2544 else if (fdset_size(fds
) != 1)
2545 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2547 /* Drop the message if BARRIER=1 was found */
2554 static void manager_invoke_notify_message(
2557 const struct ucred
*ucred
,
2566 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2568 u
->notifygen
= m
->notifygen
;
2570 if (UNIT_VTABLE(u
)->notify_message
)
2571 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2573 else if (DEBUG_LOGGING
) {
2574 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2576 buf
= strv_join(tags
, ", ");
2578 x
= ellipsize(buf
, 20, 90);
2582 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2586 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2588 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2589 Manager
*m
= ASSERT_PTR(userdata
);
2590 char buf
[NOTIFY_BUFFER_MAX
+1];
2591 struct iovec iovec
= {
2593 .iov_len
= sizeof(buf
)-1,
2595 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2596 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2597 struct msghdr msghdr
= {
2600 .msg_control
= &control
,
2601 .msg_controllen
= sizeof(control
),
2604 struct cmsghdr
*cmsg
;
2605 struct ucred
*ucred
= NULL
;
2606 _cleanup_free_ Unit
**array_copy
= NULL
;
2607 _cleanup_strv_free_
char **tags
= NULL
;
2608 Unit
*u1
, *u2
, **array
;
2609 int r
, *fd_array
= NULL
;
2614 assert(m
->notify_fd
== fd
);
2616 if (revents
!= EPOLLIN
) {
2617 log_warning("Got unexpected poll event for notify fd.");
2621 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2623 if (ERRNO_IS_TRANSIENT(n
))
2624 return 0; /* Spurious wakeup, try again */
2626 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2629 /* If this is any other, real error, then let's stop processing this socket. This of course
2630 * means we won't take notification messages anymore, but that's still better than busy
2631 * looping around this: being woken up over and over again but being unable to actually read
2632 * the message off the socket. */
2633 return log_error_errno(n
, "Failed to receive notification message: %m");
2636 CMSG_FOREACH(cmsg
, &msghdr
) {
2637 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2640 fd_array
= CMSG_TYPED_DATA(cmsg
, int);
2641 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2643 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2644 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2645 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2648 ucred
= CMSG_TYPED_DATA(cmsg
, struct ucred
);
2655 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2657 close_many(fd_array
, n_fds
);
2663 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2664 log_warning("Received notify message without valid credentials. Ignoring.");
2668 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2669 log_warning("Received notify message exceeded maximum size. Ignoring.");
2673 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes.
2674 * We permit one trailing NUL byte in the message, but don't expect it. */
2675 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2676 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2680 /* Make sure it's NUL-terminated, then parse it to obtain the tags list. */
2682 tags
= strv_split_newlines(buf
);
2688 /* Possibly a barrier fd, let's see. */
2689 if (manager_process_barrier_fd(tags
, fds
)) {
2690 log_debug("Received barrier notification message from PID " PID_FMT
".", ucred
->pid
);
2694 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2697 /* Notify every unit that might be interested, which might be multiple. */
2698 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2699 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2700 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2707 array_copy
= newdup(Unit
*, array
, k
+1);
2711 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle
2712 * duplicate units make sure we only invoke each unit's handler once. */
2714 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2718 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2722 for (size_t i
= 0; array_copy
[i
]; i
++) {
2723 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2728 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2730 if (fdset_size(fds
) > 0)
2731 log_warning("Got extra auxiliary fds with notification message, closing them.");
2736 static void manager_invoke_sigchld_event(
2739 const siginfo_t
*si
) {
2745 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2746 if (u
->sigchldgen
== m
->sigchldgen
)
2748 u
->sigchldgen
= m
->sigchldgen
;
2750 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2751 unit_unwatch_pid(u
, si
->si_pid
);
2753 if (UNIT_VTABLE(u
)->sigchld_event
)
2754 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2757 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2758 Manager
*m
= ASSERT_PTR(userdata
);
2764 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access
2765 * /proc/$PID for it while it is a zombie. */
2767 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2769 if (errno
!= ECHILD
)
2770 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2778 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2779 _cleanup_free_ Unit
**array_copy
= NULL
;
2780 _cleanup_free_
char *name
= NULL
;
2781 Unit
*u1
, *u2
, **array
;
2783 (void) get_process_comm(si
.si_pid
, &name
);
2785 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2786 si
.si_pid
, strna(name
),
2787 sigchld_code_to_string(si
.si_code
),
2789 strna(si
.si_code
== CLD_EXITED
2790 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2791 : signal_to_string(si
.si_status
)));
2793 /* Increase the generation counter used for filtering out duplicate unit invocations */
2796 /* And now figure out the unit this belongs to, it might be multiple... */
2797 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2798 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2799 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2803 /* Count how many entries the array has */
2807 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2808 array_copy
= newdup(Unit
*, array
, n
+1);
2813 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2814 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2815 * each iteration. */
2817 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2818 * We only do this for the cgroup the PID belonged to. */
2819 (void) unit_check_oom(u1
);
2821 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2822 (void) unit_check_oomd_kill(u1
);
2824 manager_invoke_sigchld_event(m
, u1
, &si
);
2827 manager_invoke_sigchld_event(m
, u2
, &si
);
2829 for (size_t i
= 0; array_copy
[i
]; i
++)
2830 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2833 /* And now, we actually reap the zombie. */
2834 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2835 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2842 /* All children processed for now, turn off event source */
2844 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2846 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2851 static void manager_start_special(Manager
*m
, const char *name
, JobMode mode
) {
2854 if (manager_add_job_by_name_and_warn(m
, JOB_START
, name
, mode
, NULL
, &job
) < 0)
2857 const char *s
= unit_status_string(job
->unit
, NULL
);
2859 log_info("Activating special unit %s...", s
);
2862 "STATUS=Activating special unit %s...", s
);
2863 m
->status_ready
= false;
2866 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2867 /* If the user presses C-A-D more than
2868 * 7 times within 2s, we reboot/shutdown immediately,
2869 * unless it was disabled in system.conf */
2871 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2872 manager_start_special(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2874 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2875 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2878 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2879 Manager
*m
= ASSERT_PTR(userdata
);
2881 struct signalfd_siginfo sfsi
;
2884 assert(m
->signal_fd
== fd
);
2886 if (revents
!= EPOLLIN
) {
2887 log_warning("Got unexpected events from signal file descriptor.");
2891 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2893 if (ERRNO_IS_TRANSIENT(errno
))
2896 /* We return an error here, which will kill this handler,
2897 * to avoid a busy loop on read error. */
2898 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2900 if (n
!= sizeof(sfsi
)) {
2901 log_warning("Truncated read from signal fd (%zi bytes), ignoring!", n
);
2905 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2906 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2907 ? LOG_DEBUG
: LOG_INFO
,
2910 switch (sfsi
.ssi_signo
) {
2913 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2915 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2920 if (MANAGER_IS_SYSTEM(m
)) {
2921 /* This is for compatibility with the original sysvinit */
2922 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2925 m
->objective
= MANAGER_REEXECUTE
;
2931 if (MANAGER_IS_SYSTEM(m
))
2932 manager_handle_ctrl_alt_del(m
);
2934 manager_start_special(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2938 /* This is a nop on non-init */
2939 if (MANAGER_IS_SYSTEM(m
))
2940 manager_start_special(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2945 /* This is a nop on non-init */
2946 if (MANAGER_IS_SYSTEM(m
))
2947 manager_start_special(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2952 if (manager_dbus_is_running(m
, false)) {
2953 log_info("Trying to reconnect to bus...");
2955 (void) bus_init_api(m
);
2957 if (MANAGER_IS_SYSTEM(m
))
2958 (void) bus_init_system(m
);
2960 manager_start_special(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2965 _cleanup_free_
char *dump
= NULL
;
2967 r
= manager_get_dump_string(m
, /* patterns= */ NULL
, &dump
);
2969 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2973 log_dump(LOG_INFO
, dump
);
2978 if (verify_run_space_and_log("Refusing to reload") < 0)
2981 m
->objective
= MANAGER_RELOAD
;
2986 /* Starting SIGRTMIN+0 */
2987 static const struct {
2990 } target_table
[] = {
2991 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2992 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2993 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2994 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2995 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2996 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2997 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2998 [7] = { SPECIAL_SOFT_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
3001 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
3002 static const ManagerObjective objective_table
[] = {
3004 [1] = MANAGER_POWEROFF
,
3005 [2] = MANAGER_REBOOT
,
3006 [3] = MANAGER_KEXEC
,
3007 [4] = MANAGER_SOFT_REBOOT
,
3010 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
3011 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
3012 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
3013 manager_start_special(m
, target_table
[idx
].target
, target_table
[idx
].mode
);
3017 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
3018 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
3019 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
3023 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
3026 bool generic
= false;
3028 if (sfsi
.ssi_code
!= SI_QUEUE
)
3031 /* Override a few select commands by our own PID1-specific logic */
3033 switch (sfsi
.ssi_int
) {
3035 case _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
..._COMMON_SIGNAL_COMMAND_LOG_LEVEL_END
:
3036 manager_override_log_level(m
, sfsi
.ssi_int
- _COMMON_SIGNAL_COMMAND_LOG_LEVEL_BASE
);
3039 case COMMON_SIGNAL_COMMAND_CONSOLE
:
3040 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3043 case COMMON_SIGNAL_COMMAND_JOURNAL
:
3044 manager_override_log_target(m
, LOG_TARGET_JOURNAL
);
3047 case COMMON_SIGNAL_COMMAND_KMSG
:
3048 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3051 case COMMON_SIGNAL_COMMAND_NULL
:
3052 manager_override_log_target(m
, LOG_TARGET_NULL
);
3061 return sigrtmin18_handler(source
, &sfsi
, NULL
);
3067 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
3071 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
3075 manager_override_log_level(m
, LOG_DEBUG
);
3079 manager_restore_original_log_level(m
);
3083 if (MANAGER_IS_USER(m
)) {
3084 m
->objective
= MANAGER_EXIT
;
3088 /* This is a nop on init */
3092 m
->objective
= MANAGER_REEXECUTE
;
3096 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
3097 manager_restore_original_log_target(m
);
3101 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
3105 manager_override_log_target(m
, LOG_TARGET_KMSG
);
3109 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
3116 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3117 Manager
*m
= ASSERT_PTR(userdata
);
3120 log_struct(LOG_DEBUG
,
3121 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
3122 LOG_MESSAGE("Time has been changed"));
3124 /* Restart the watch */
3125 (void) manager_setup_time_change(m
);
3127 HASHMAP_FOREACH(u
, m
->units
)
3128 if (UNIT_VTABLE(u
)->time_change
)
3129 UNIT_VTABLE(u
)->time_change(u
);
3134 static int manager_dispatch_timezone_change(
3135 sd_event_source
*source
,
3136 const struct inotify_event
*e
,
3139 Manager
*m
= ASSERT_PTR(userdata
);
3143 log_debug("inotify event for /etc/localtime");
3145 changed
= manager_read_timezone_stat(m
);
3149 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
3150 (void) manager_setup_timezone_change(m
);
3152 /* Read the new timezone */
3155 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
3157 HASHMAP_FOREACH(u
, m
->units
)
3158 if (UNIT_VTABLE(u
)->timezone_change
)
3159 UNIT_VTABLE(u
)->timezone_change(u
);
3164 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3165 Manager
*m
= ASSERT_PTR(userdata
);
3167 assert(m
->idle_pipe
[2] == fd
);
3169 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to
3170 * complete. Let's now turn off any further console output if there's at least one service that needs
3171 * console access, so that from now on our own output should not spill into that service's output
3172 * anymore. After all, we support Type=idle only to beautify console output and it generally is set
3173 * on services that want to own the console exclusively without our interference. */
3174 m
->no_console_output
= m
->n_on_console
> 0;
3176 /* Acknowledge the child's request, and let all other children know too that they shouldn't wait
3177 * any longer by closing the pipes towards them, which is what they are waiting for. */
3178 manager_close_idle_pipe(m
);
3183 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
3184 Manager
*m
= ASSERT_PTR(userdata
);
3189 manager_print_jobs_in_progress(m
);
3191 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
3195 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
3198 int manager_loop(Manager
*m
) {
3199 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
3203 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
3205 manager_check_finished(m
);
3207 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3208 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
3210 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
3212 while (m
->objective
== MANAGER_OK
) {
3214 (void) watchdog_ping();
3216 if (!ratelimit_below(&rl
)) {
3217 /* Yay, something is going seriously wrong, pause a little */
3218 log_warning("Looping too fast. Throttling execution a little.");
3222 if (manager_dispatch_load_queue(m
) > 0)
3225 if (manager_dispatch_gc_job_queue(m
) > 0)
3228 if (manager_dispatch_gc_unit_queue(m
) > 0)
3231 if (manager_dispatch_cleanup_queue(m
) > 0)
3234 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3237 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3240 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3243 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3246 if (manager_dispatch_release_resources_queue(m
) > 0)
3249 if (manager_dispatch_dbus_queue(m
) > 0)
3252 /* Sleep for watchdog runtime wait time */
3253 r
= sd_event_run(m
->event
, watchdog_runtime_wait());
3255 return log_error_errno(r
, "Failed to run event loop: %m");
3258 return m
->objective
;
3261 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3262 _cleanup_free_
char *n
= NULL
;
3263 sd_id128_t invocation_id
;
3271 r
= unit_name_from_dbus_path(s
, &n
);
3275 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128-bit ID then
3276 * we use it as invocation ID. */
3277 r
= sd_id128_from_string(n
, &invocation_id
);
3279 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3285 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3286 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3287 SD_ID128_FORMAT_VAL(invocation_id
));
3290 /* If this didn't work, we check if this is a unit name */
3291 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3292 _cleanup_free_
char *nn
= NULL
;
3295 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3296 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3299 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3307 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3317 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3321 r
= safe_atou(p
, &id
);
3325 j
= manager_get_job(m
, id
);
3334 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3337 _cleanup_free_
char *p
= NULL
;
3341 if (!MANAGER_IS_SYSTEM(m
))
3344 audit_fd
= get_audit_fd();
3348 /* Don't generate audit events if the service was already
3349 * started and we're just deserializing */
3350 if (MANAGER_IS_RELOADING(m
))
3353 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3355 log_warning_errno(r
, "Failed to extract prefix and instance of unit name, ignoring: %m");
3359 msg
= strjoina("unit=", p
);
3360 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3361 if (ERRNO_IS_PRIVILEGE(errno
)) {
3362 /* We aren't allowed to send audit messages? Then let's not retry again. */
3363 log_debug_errno(errno
, "Failed to send audit message, closing audit socket: %m");
3366 log_warning_errno(errno
, "Failed to send audit message, ignoring: %m");
3372 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3373 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3374 _cleanup_free_
char *message
= NULL
;
3375 _cleanup_close_
int fd
= -EBADF
;
3378 /* Don't generate plymouth events if the service was already
3379 * started and we're just deserializing */
3380 if (MANAGER_IS_RELOADING(m
))
3383 if (!MANAGER_IS_SYSTEM(m
))
3386 if (detect_container() > 0)
3389 if (!UNIT_VTABLE(u
)->notify_plymouth
)
3392 /* We set SOCK_NONBLOCK here so that we rather drop the
3393 * message then wait for plymouth */
3394 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3396 log_error_errno(errno
, "socket() failed: %m");
3400 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3401 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3402 log_error_errno(errno
, "connect() failed: %m");
3406 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0)
3407 return (void) log_oom();
3410 if (write(fd
, message
, n
+ 1) != n
+ 1)
3411 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3412 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3415 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3418 if (MANAGER_IS_USER(m
))
3419 return USEC_INFINITY
;
3421 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3422 return m
->watchdog_overridden
[t
];
3424 return m
->watchdog
[t
];
3427 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3431 if (MANAGER_IS_USER(m
))
3434 if (m
->watchdog
[t
] == timeout
)
3437 if (t
== WATCHDOG_RUNTIME
) {
3438 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3439 (void) watchdog_setup(timeout
);
3440 } else if (t
== WATCHDOG_PRETIMEOUT
)
3441 if (m
->watchdog_overridden
[WATCHDOG_PRETIMEOUT
] == USEC_INFINITY
)
3442 (void) watchdog_setup_pretimeout(timeout
);
3444 m
->watchdog
[t
] = timeout
;
3447 void manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3451 if (MANAGER_IS_USER(m
))
3454 if (m
->watchdog_overridden
[t
] == timeout
)
3457 if (t
== WATCHDOG_RUNTIME
) {
3458 usec_t usec
= timestamp_is_set(timeout
) ? timeout
: m
->watchdog
[t
];
3460 (void) watchdog_setup(usec
);
3461 } else if (t
== WATCHDOG_PRETIMEOUT
)
3462 (void) watchdog_setup_pretimeout(timeout
);
3464 m
->watchdog_overridden
[t
] = timeout
;
3467 int manager_set_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3468 _cleanup_free_
char *p
= NULL
;
3473 if (MANAGER_IS_USER(m
))
3476 if (streq_ptr(m
->watchdog_pretimeout_governor
, governor
))
3479 p
= strdup(governor
);
3483 r
= watchdog_setup_pretimeout_governor(governor
);
3487 return free_and_replace(m
->watchdog_pretimeout_governor
, p
);
3490 int manager_override_watchdog_pretimeout_governor(Manager
*m
, const char *governor
) {
3491 _cleanup_free_
char *p
= NULL
;
3496 if (MANAGER_IS_USER(m
))
3499 if (streq_ptr(m
->watchdog_pretimeout_governor_overridden
, governor
))
3502 p
= strdup(governor
);
3506 r
= watchdog_setup_pretimeout_governor(governor
);
3510 return free_and_replace(m
->watchdog_pretimeout_governor_overridden
, p
);
3513 int manager_reload(Manager
*m
) {
3514 _unused_
_cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3515 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3516 _cleanup_fclose_
FILE *f
= NULL
;
3521 r
= manager_open_serialization(m
, &f
);
3523 return log_error_errno(r
, "Failed to create serialization file: %m");
3529 /* We are officially in reload mode from here on. */
3530 reloading
= manager_reloading_start(m
);
3532 r
= manager_serialize(m
, f
, fds
, false);
3536 if (fseeko(f
, 0, SEEK_SET
) < 0)
3537 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3539 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3542 bus_manager_send_reloading(m
, true);
3544 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3545 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3548 manager_clear_jobs_and_units(m
);
3549 lookup_paths_flush_generator(&m
->lookup_paths
);
3550 lookup_paths_free(&m
->lookup_paths
);
3551 exec_shared_runtime_vacuum(m
);
3552 dynamic_user_vacuum(m
, false);
3553 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3554 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3556 r
= lookup_paths_init_or_warn(&m
->lookup_paths
, m
->runtime_scope
, 0, NULL
);
3560 (void) manager_run_environment_generators(m
);
3561 (void) manager_run_generators(m
);
3563 lookup_paths_log(&m
->lookup_paths
);
3565 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3566 manager_free_unit_name_maps(m
);
3568 /* First, enumerate what we can from kernel and suchlike */
3569 manager_enumerate_perpetual(m
);
3570 manager_enumerate(m
);
3572 /* Second, deserialize our stored data */
3573 r
= manager_deserialize(m
, f
, fds
);
3575 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3577 /* We don't need the serialization anymore */
3580 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3581 (void) manager_setup_notify(m
);
3582 (void) manager_setup_cgroups_agent(m
);
3583 (void) manager_setup_user_lookup_fd(m
);
3585 /* Third, fire things up! */
3586 manager_coldplug(m
);
3588 /* Clean up runtime objects no longer referenced */
3591 /* Clean up deserialized tracked clients */
3592 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3594 /* Consider the reload process complete now. */
3595 assert(m
->n_reloading
> 0);
3600 m
->send_reloading_done
= true;
3604 void manager_reset_failed(Manager
*m
) {
3609 HASHMAP_FOREACH(u
, m
->units
)
3610 unit_reset_failed(u
);
3613 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3619 /* Returns true if the unit is inactive or going down */
3620 u
= manager_get_unit(m
, name
);
3624 return unit_inactive_or_pending(u
);
3627 static void log_taint_string(Manager
*m
) {
3628 _cleanup_free_
char *taint
= NULL
;
3632 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3635 m
->taint_logged
= true; /* only check for taint once */
3637 taint
= manager_taint_string(m
);
3641 log_struct(LOG_NOTICE
,
3642 LOG_MESSAGE("System is tainted: %s", taint
),
3644 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3647 static void manager_notify_finished(Manager
*m
) {
3648 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3650 if (MANAGER_IS_TEST_RUN(m
))
3653 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3654 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3657 size_t size
= sizeof buf
;
3659 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3660 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3661 * negative values. */
3663 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3664 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3665 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3666 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3668 if (firmware_usec
> 0)
3669 size
= strpcpyf(&p
, size
, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec
, USEC_PER_MSEC
));
3670 if (loader_usec
> 0)
3671 size
= strpcpyf(&p
, size
, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec
, USEC_PER_MSEC
));
3673 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
3675 /* The initrd case on bare-metal */
3676 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3677 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
3679 log_struct(LOG_INFO
,
3680 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3681 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3682 "INITRD_USEC="USEC_FMT
, initrd_usec
,
3683 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3684 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3686 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3687 FORMAT_TIMESPAN(initrd_usec
, USEC_PER_MSEC
),
3688 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3689 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3691 /* The initrd-less case on bare-metal */
3693 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3696 log_struct(LOG_INFO
,
3697 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
3698 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
3699 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3700 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3702 FORMAT_TIMESPAN(kernel_usec
, USEC_PER_MSEC
),
3703 FORMAT_TIMESPAN(userspace_usec
, USEC_PER_MSEC
),
3704 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3707 /* The container and --user case */
3708 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
3709 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3711 log_struct(LOG_INFO
,
3712 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
3713 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
3714 LOG_MESSAGE("Startup finished in %s.",
3715 FORMAT_TIMESPAN(total_usec
, USEC_PER_MSEC
)));
3718 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
3720 log_taint_string(m
);
3723 static void user_manager_send_ready(Manager
*m
) {
3728 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3729 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
3732 r
= sd_notify(false,
3734 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
3736 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
3738 m
->ready_sent
= true;
3739 m
->status_ready
= false;
3742 static void manager_send_ready(Manager
*m
) {
3745 if (m
->ready_sent
&& m
->status_ready
)
3746 /* Skip the notification if nothing changed. */
3749 r
= sd_notify(false,
3753 log_full_errno(m
->ready_sent
? LOG_DEBUG
: LOG_WARNING
, r
,
3754 "Failed to send readiness notification, ignoring: %m");
3756 m
->ready_sent
= m
->status_ready
= true;
3759 static void manager_check_basic_target(Manager
*m
) {
3764 /* Small shortcut */
3765 if (m
->ready_sent
&& m
->taint_logged
)
3768 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3769 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3772 /* For user managers, send out READY=1 as soon as we reach basic.target */
3773 user_manager_send_ready(m
);
3775 /* Log the taint string as soon as we reach basic.target */
3776 log_taint_string(m
);
3779 void manager_check_finished(Manager
*m
) {
3782 if (MANAGER_IS_RELOADING(m
))
3785 /* Verify that we have entered the event loop already, and not left it again. */
3786 if (!MANAGER_IS_RUNNING(m
))
3789 manager_check_basic_target(m
);
3791 if (hashmap_size(m
->jobs
) > 0) {
3792 if (m
->jobs_in_progress_event_source
)
3793 /* Ignore any failure, this is only for feedback */
3794 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
3795 manager_watch_jobs_next_time(m
));
3799 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3800 kill the hashmap if it is relatively large. */
3801 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
3802 m
->jobs
= hashmap_free(m
->jobs
);
3804 manager_send_ready(m
);
3806 /* Notify Type=idle units that we are done now */
3807 manager_close_idle_pipe(m
);
3809 if (MANAGER_IS_FINISHED(m
))
3812 manager_flip_auto_status(m
, false, "boot finished");
3814 /* Turn off confirm spawn now */
3815 m
->confirm_spawn
= NULL
;
3817 /* No need to update ask password status when we're going non-interactive */
3818 manager_close_ask_password(m
);
3820 /* This is no longer the first boot */
3821 manager_set_first_boot(m
, false);
3823 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
3825 manager_notify_finished(m
);
3827 manager_invalidate_startup_units(m
);
3830 void manager_send_reloading(Manager
*m
) {
3833 /* Let whoever invoked us know that we are now reloading */
3834 (void) sd_notifyf(/* unset= */ false,
3836 "MONOTONIC_USEC=" USEC_FMT
"\n", now(CLOCK_MONOTONIC
));
3838 /* And ensure that we'll send READY=1 again as soon as we are ready again */
3839 m
->ready_sent
= false;
3842 static bool generator_path_any(const char* const* paths
) {
3845 /* Optimize by skipping the whole process by not creating output directories
3846 * if no generators are found. */
3847 STRV_FOREACH(path
, paths
)
3848 if (access(*path
, F_OK
) == 0)
3850 else if (errno
!= ENOENT
)
3851 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3856 static int manager_run_environment_generators(Manager
*m
) {
3857 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
3858 _cleanup_strv_free_
char **paths
= NULL
;
3860 [STDOUT_GENERATE
] = &tmp
,
3861 [STDOUT_COLLECT
] = &tmp
,
3862 [STDOUT_CONSUME
] = &m
->transient_environment
,
3866 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
3869 paths
= env_generator_binary_paths(m
->runtime_scope
);
3873 if (!generator_path_any((const char* const*) paths
))
3877 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
3878 args
, NULL
, m
->transient_environment
,
3879 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3883 static int build_generator_environment(Manager
*m
, char ***ret
) {
3884 _cleanup_strv_free_
char **nl
= NULL
;
3891 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3892 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3893 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3895 nl
= strv_copy(m
->transient_environment
);
3899 r
= strv_env_assign(&nl
, "SYSTEMD_SCOPE", runtime_scope_to_string(m
->runtime_scope
));
3903 if (MANAGER_IS_SYSTEM(m
)) {
3904 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3905 * codebase. This is hence more than purely informational. It will shortcut detection of the
3906 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3907 * same results (hopefully). */
3908 r
= strv_env_assign(&nl
, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3912 if (m
->first_boot
>= 0) {
3913 r
= strv_env_assign(&nl
, "SYSTEMD_FIRST_BOOT", one_zero(m
->first_boot
));
3919 v
= detect_virtualization();
3921 log_debug_errno(v
, "Failed to detect virtualization, ignoring: %m");
3925 s
= strjoina(VIRTUALIZATION_IS_VM(v
) ? "vm:" :
3926 VIRTUALIZATION_IS_CONTAINER(v
) ? "container:" : ":",
3927 virtualization_to_string(v
));
3929 r
= strv_env_assign(&nl
, "SYSTEMD_VIRTUALIZATION", s
);
3934 r
= strv_env_assign(&nl
, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3938 *ret
= TAKE_PTR(nl
);
3942 static int manager_execute_generators(Manager
*m
, char **paths
, bool remount_ro
) {
3943 _cleanup_strv_free_
char **ge
= NULL
;
3944 const char *argv
[] = {
3945 NULL
, /* Leave this empty, execute_directory() will fill something in */
3946 m
->lookup_paths
.generator
,
3947 m
->lookup_paths
.generator_early
,
3948 m
->lookup_paths
.generator_late
,
3953 r
= build_generator_environment(m
, &ge
);
3955 return log_error_errno(r
, "Failed to build generator environment: %m");
3958 /* Remount most of the filesystem tree read-only. We leave /sys/ as-is, because our code
3959 * checks whether it is read-only to detect containerized execution environments. We leave
3960 * /run/ as-is too, because that's where our output goes. We also leave /proc/ and /dev/shm/
3961 * because they're API, and /tmp/ that safe_fork() mounted for us.
3963 r
= bind_remount_recursive("/", MS_RDONLY
, MS_RDONLY
,
3964 STRV_MAKE("/sys", "/run", "/proc", "/dev/shm", "/tmp"));
3966 log_warning_errno(r
, "Read-only bind remount failed, ignoring: %m");
3969 BLOCK_WITH_UMASK(0022);
3970 return execute_directories(
3971 (const char* const*) paths
,
3972 DEFAULT_TIMEOUT_USEC
,
3973 /* callbacks= */ NULL
, /* callback_args= */ NULL
,
3976 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
3979 static int manager_run_generators(Manager
*m
) {
3980 ForkFlags flags
= FORK_RESET_SIGNALS
| FORK_WAIT
| FORK_NEW_MOUNTNS
| FORK_MOUNTNS_SLAVE
;
3981 _cleanup_strv_free_
char **paths
= NULL
;
3986 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
3989 paths
= generator_binary_paths(m
->runtime_scope
);
3993 if (!generator_path_any((const char* const*) paths
))
3996 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3998 log_error_errno(r
, "Failed to create generator directories: %m");
4002 /* If we are the system manager, we fork and invoke the generators in a sanitized mount namespace. If
4003 * we are the user manager, let's just execute the generators directly. We might not have the
4004 * necessary privileges, and the system manager has already mounted /tmp/ and everything else for us.
4006 if (MANAGER_IS_USER(m
)) {
4007 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4011 /* On some systems /tmp/ doesn't exist, and on some other systems we cannot create it at all. Avoid
4012 * trying to mount a private tmpfs on it as there's no one size fits all. */
4013 if (is_dir("/tmp", /* follow= */ false) > 0)
4014 flags
|= FORK_PRIVATE_TMP
;
4016 r
= safe_fork("(sd-gens)", flags
, NULL
);
4018 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ true);
4019 _exit(r
>= 0 ? EXIT_SUCCESS
: EXIT_FAILURE
);
4022 if (!ERRNO_IS_PRIVILEGE(r
)) {
4023 log_error_errno(r
, "Failed to fork off sandboxing environment for executing generators: %m");
4027 /* Failed to fork with new mount namespace? Maybe, running in a container environment with
4028 * seccomp or without capability. */
4030 "Failed to fork off sandboxing environment for executing generators. "
4031 "Falling back to execute generators without sandboxing: %m");
4032 r
= manager_execute_generators(m
, paths
, /* remount_ro= */ false);
4036 lookup_paths_trim_generator(&m
->lookup_paths
);
4040 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4045 if (strv_isempty(plus
))
4048 a
= strv_env_merge(m
->transient_environment
, plus
);
4052 sanitize_environment(a
);
4054 return strv_free_and_replace(m
->transient_environment
, a
);
4057 int manager_client_environment_modify(
4062 char **a
= NULL
, **b
= NULL
, **l
;
4066 if (strv_isempty(minus
) && strv_isempty(plus
))
4069 l
= m
->client_environment
;
4071 if (!strv_isempty(minus
)) {
4072 a
= strv_env_delete(l
, 1, minus
);
4079 if (!strv_isempty(plus
)) {
4080 b
= strv_env_merge(l
, plus
);
4089 if (m
->client_environment
!= l
)
4090 strv_free(m
->client_environment
);
4097 m
->client_environment
= sanitize_environment(l
);
4101 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4107 l
= strv_env_merge(m
->transient_environment
, m
->client_environment
);
4115 int manager_set_default_smack_process_label(Manager
*m
, const char *label
) {
4118 #ifdef SMACK_DEFAULT_PROCESS_LABEL
4120 return free_and_strdup(&m
->default_smack_process_label
, SMACK_DEFAULT_PROCESS_LABEL
);
4122 if (streq_ptr(label
, "/"))
4123 return free_and_strdup(&m
->default_smack_process_label
, NULL
);
4125 return free_and_strdup(&m
->default_smack_process_label
, label
);
4128 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4131 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4132 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4134 if (!default_rlimit
[i
])
4137 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4145 void manager_recheck_dbus(Manager
*m
) {
4148 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode
4149 * this is all it does. In system mode we'll also connect to the system bus (which will most likely
4150 * just reuse the connection of the API bus). That's because the system bus after all runs as service
4151 * of the system instance, while in the user instance we can assume it's already there. */
4153 if (MANAGER_IS_RELOADING(m
))
4154 return; /* don't check while we are reloading… */
4156 if (manager_dbus_is_running(m
, false)) {
4157 (void) bus_init_api(m
);
4159 if (MANAGER_IS_SYSTEM(m
))
4160 (void) bus_init_system(m
);
4162 (void) bus_done_api(m
);
4164 if (MANAGER_IS_SYSTEM(m
))
4165 (void) bus_done_system(m
);
4169 static bool manager_journal_is_running(Manager
*m
) {
4174 if (MANAGER_IS_TEST_RUN(m
))
4177 /* If we are the user manager we can safely assume that the journal is up */
4178 if (!MANAGER_IS_SYSTEM(m
))
4181 /* Check that the socket is not only up, but in RUNNING state */
4182 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4185 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4188 /* Similar, check if the daemon itself is fully up, too */
4189 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4192 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4198 void disable_printk_ratelimit(void) {
4199 /* Disable kernel's printk ratelimit.
4201 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4202 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4203 * setting takes precedence. */
4206 r
= sysctl_write("kernel/printk_devkmsg", "on");
4208 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4211 void manager_recheck_journal(Manager
*m
) {
4215 /* Don't bother with this unless we are in the special situation of being PID 1 */
4216 if (getpid_cached() != 1)
4219 /* Don't check this while we are reloading, things might still change */
4220 if (MANAGER_IS_RELOADING(m
))
4223 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If
4224 * the journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we
4225 * might trigger an activation ourselves we can't fulfill. */
4226 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4230 static ShowStatus
manager_get_show_status(Manager
*m
) {
4233 if (MANAGER_IS_USER(m
))
4234 return _SHOW_STATUS_INVALID
;
4236 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4237 return m
->show_status_overridden
;
4239 return m
->show_status
;
4242 bool manager_get_show_status_on(Manager
*m
) {
4245 return show_status_on(manager_get_show_status(m
));
4248 static void set_show_status_marker(bool b
) {
4250 (void) touch("/run/systemd/show-status");
4252 (void) unlink("/run/systemd/show-status");
4255 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4258 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4260 if (MANAGER_IS_USER(m
))
4263 if (mode
== m
->show_status
)
4266 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4269 enabled
= show_status_on(mode
);
4270 log_debug("%s (%s) showing of status (%s).",
4271 enabled
? "Enabling" : "Disabling",
4272 strna(show_status_to_string(mode
)),
4275 set_show_status_marker(enabled
);
4278 m
->show_status
= mode
;
4281 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4283 assert(mode
< _SHOW_STATUS_MAX
);
4285 if (MANAGER_IS_USER(m
))
4288 if (mode
== m
->show_status_overridden
)
4291 m
->show_status_overridden
= mode
;
4293 if (mode
== _SHOW_STATUS_INVALID
)
4294 mode
= m
->show_status
;
4296 log_debug("%s (%s) showing of status (%s).",
4297 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4298 strna(show_status_to_string(mode
)),
4301 set_show_status_marker(show_status_on(mode
));
4304 const char *manager_get_confirm_spawn(Manager
*m
) {
4305 static int last_errno
= 0;
4311 /* Here's the deal: we want to test the validity of the console but don't want
4312 * PID1 to go through the whole console process which might block. But we also
4313 * want to warn the user only once if something is wrong with the console so we
4314 * cannot do the sanity checks after spawning our children. So here we simply do
4315 * really basic tests to hopefully trap common errors.
4317 * If the console suddenly disappear at the time our children will really it
4318 * then they will simply fail to acquire it and a positive answer will be
4319 * assumed. New children will fall back to /dev/console though.
4321 * Note: TTYs are devices that can come and go any time, and frequently aren't
4322 * available yet during early boot (consider a USB rs232 dongle...). If for any
4323 * reason the configured console is not ready, we fall back to the default
4326 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4327 return m
->confirm_spawn
;
4329 if (stat(m
->confirm_spawn
, &st
) < 0) {
4334 if (!S_ISCHR(st
.st_mode
)) {
4340 return m
->confirm_spawn
;
4343 if (last_errno
!= r
)
4344 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4346 return "/dev/console";
4349 void manager_set_first_boot(Manager
*m
, bool b
) {
4352 if (!MANAGER_IS_SYSTEM(m
))
4355 if (m
->first_boot
!= (int) b
) {
4357 (void) touch("/run/systemd/first-boot");
4359 (void) unlink("/run/systemd/first-boot");
4365 void manager_disable_confirm_spawn(void) {
4366 (void) touch("/run/systemd/confirm_spawn_disabled");
4369 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4370 if (!m
->confirm_spawn
)
4373 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4376 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4379 if (!MANAGER_IS_SYSTEM(m
))
4382 if (m
->no_console_output
)
4385 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4388 /* If we cannot find out the status properly, just proceed. */
4389 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4392 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4395 return manager_get_show_status_on(m
);
4398 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4401 /* If m is NULL, assume we're after shutdown and let the messages through. */
4403 if (m
&& !manager_should_show_status(m
, type
))
4406 /* XXX We should totally drop the check for ephemeral here
4407 * and thus effectively make 'Type=idle' pointless. */
4408 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4411 va_start(ap
, format
);
4412 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4416 Set
* manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4420 if (path_equal(path
, "/"))
4423 return hashmap_get(m
->units_requiring_mounts_for
, path
);
4426 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4431 assert(u
->manager
== m
);
4433 size
= set_size(m
->failed_units
);
4436 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4440 (void) set_remove(m
->failed_units
, u
);
4442 if (set_size(m
->failed_units
) != size
)
4443 bus_manager_send_change_signal(m
);
4448 ManagerState
manager_state(Manager
*m
) {
4453 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4454 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4455 if (u
&& unit_active_or_pending(u
))
4456 return MANAGER_STOPPING
;
4458 /* Did we ever finish booting? If not then we are still starting up */
4459 if (!MANAGER_IS_FINISHED(m
)) {
4461 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4462 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4463 return MANAGER_INITIALIZING
;
4465 return MANAGER_STARTING
;
4468 if (MANAGER_IS_SYSTEM(m
)) {
4469 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4470 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4471 if (u
&& unit_active_or_pending(u
))
4472 return MANAGER_MAINTENANCE
;
4474 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4475 if (u
&& unit_active_or_pending(u
))
4476 return MANAGER_MAINTENANCE
;
4479 /* Are there any failed units? If so, we are in degraded mode */
4480 if (set_size(m
->failed_units
) > 0)
4481 return MANAGER_DEGRADED
;
4483 return MANAGER_RUNNING
;
4486 static void manager_unref_uid_internal(
4490 int (*_clean_ipc
)(uid_t uid
)) {
4494 assert(uid_is_valid(uid
));
4497 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the
4498 * assumption that uid_t and gid_t are actually defined the same way, with the same validity rules.
4500 * We store a hashmap where the key is the UID/GID and the value is a 32-bit reference counter, whose
4501 * highest bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last
4502 * reference to the UID/GID is dropped. The flag is set to on, once at least one reference from a
4503 * unit where RemoveIPC= is set is added on a UID/GID. It is reset when the UID's/GID's reference
4504 * counter drops to 0 again. */
4506 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4507 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4509 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4512 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4514 n
= c
& ~DESTROY_IPC_FLAG
;
4518 if (destroy_now
&& n
== 0) {
4519 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4521 if (c
& DESTROY_IPC_FLAG
) {
4522 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4523 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4525 (void) _clean_ipc(uid
);
4528 c
= n
| (c
& DESTROY_IPC_FLAG
);
4529 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4533 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4534 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4537 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4538 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4541 static int manager_ref_uid_internal(
4550 assert(uid_is_valid(uid
));
4552 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the
4553 * assumption that uid_t and gid_t are actually defined the same way, with the same validity
4556 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4557 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4559 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4562 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4566 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4568 n
= c
& ~DESTROY_IPC_FLAG
;
4571 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4574 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4576 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4579 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4580 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4583 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4584 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4587 static void manager_vacuum_uid_refs_internal(
4589 int (*_clean_ipc
)(uid_t uid
)) {
4595 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4599 uid
= PTR_TO_UID(k
);
4600 c
= PTR_TO_UINT32(p
);
4602 n
= c
& ~DESTROY_IPC_FLAG
;
4606 if (c
& DESTROY_IPC_FLAG
) {
4607 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4608 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4610 (void) _clean_ipc(uid
);
4613 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4617 static void manager_vacuum_uid_refs(Manager
*m
) {
4618 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4621 static void manager_vacuum_gid_refs(Manager
*m
) {
4622 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4625 static void manager_vacuum(Manager
*m
) {
4628 /* Release any dynamic users no longer referenced */
4629 dynamic_user_vacuum(m
, true);
4631 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4632 manager_vacuum_uid_refs(m
);
4633 manager_vacuum_gid_refs(m
);
4635 /* Release any runtimes no longer referenced */
4636 exec_shared_runtime_vacuum(m
);
4639 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4643 char unit_name
[UNIT_NAME_MAX
+1];
4646 Manager
*m
= userdata
;
4654 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the
4655 * resulting UID/GID in a datagram. We parse the datagram here and pass it off to the unit, so that
4656 * it can add a reference to the UID/GID so that it can destroy the UID/GID's IPC objects when the
4657 * reference counter drops to 0. */
4659 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4661 if (ERRNO_IS_TRANSIENT(errno
))
4664 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4667 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4668 log_warning("Received too short user lookup message, ignoring.");
4672 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4673 log_warning("Received too long user lookup message, ignoring.");
4677 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4678 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4682 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4683 if (memchr(buffer
.unit_name
, 0, n
)) {
4684 log_warning("Received lookup message with embedded NUL character, ignoring.");
4688 buffer
.unit_name
[n
] = 0;
4689 u
= manager_get_unit(m
, buffer
.unit_name
);
4691 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4695 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4697 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4701 static int short_uid_range(const char *path
) {
4702 _cleanup_(uid_range_freep
) UidRange
*p
= NULL
;
4707 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4708 * i.e. from root to nobody. */
4710 r
= uid_range_load_userns(&p
, path
);
4712 if (ERRNO_IS_NOT_SUPPORTED(r
))
4714 return log_debug_errno(r
, "Failed to load %s: %m", path
);
4717 return !uid_range_covers(p
, 0, 65535);
4720 char* manager_taint_string(const Manager
*m
) {
4721 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4722 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4723 * configuration phase. */
4727 const char* stage
[13] = {};
4731 stage
[n
++] = "split-usr";
4733 _cleanup_free_
char *usrbin
= NULL
;
4734 if (readlink_malloc("/bin", &usrbin
) < 0 || !PATH_IN_SET(usrbin
, "usr/bin", "/usr/bin"))
4735 stage
[n
++] = "unmerged-usr";
4737 if (access("/proc/cgroups", F_OK
) < 0)
4738 stage
[n
++] = "cgroups-missing";
4740 if (cg_all_unified() == 0)
4741 stage
[n
++] = "cgroupsv1";
4743 if (clock_is_localtime(NULL
) > 0)
4744 stage
[n
++] = "local-hwclock";
4746 if (os_release_support_ended(NULL
, /* quiet= */ true, NULL
) > 0)
4747 stage
[n
++] = "support-ended";
4749 _cleanup_free_
char *destination
= NULL
;
4750 if (readlink_malloc("/var/run", &destination
) < 0 ||
4751 !PATH_IN_SET(destination
, "../run", "/run"))
4752 stage
[n
++] = "var-run-bad";
4754 _cleanup_free_
char *overflowuid
= NULL
, *overflowgid
= NULL
;
4755 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
) >= 0 &&
4756 !streq(overflowuid
, "65534"))
4757 stage
[n
++] = "overflowuid-not-65534";
4758 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
) >= 0 &&
4759 !streq(overflowgid
, "65534"))
4760 stage
[n
++] = "overflowgid-not-65534";
4763 assert_se(uname(&uts
) >= 0);
4764 if (strverscmp_improved(uts
.release
, KERNEL_BASELINE_VERSION
) < 0)
4765 stage
[n
++] = "old-kernel";
4767 if (short_uid_range("/proc/self/uid_map") > 0)
4768 stage
[n
++] = "short-uid-range";
4769 if (short_uid_range("/proc/self/gid_map") > 0)
4770 stage
[n
++] = "short-gid-range";
4772 assert(n
< ELEMENTSOF(stage
) - 1); /* One extra for NULL terminator */
4774 return strv_join((char**) stage
, ":");
4777 void manager_ref_console(Manager
*m
) {
4783 void manager_unref_console(Manager
*m
) {
4785 assert(m
->n_on_console
> 0);
4788 if (m
->n_on_console
== 0)
4789 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4792 void manager_override_log_level(Manager
*m
, int level
) {
4793 _cleanup_free_
char *s
= NULL
;
4796 if (!m
->log_level_overridden
) {
4797 m
->original_log_level
= log_get_max_level();
4798 m
->log_level_overridden
= true;
4801 (void) log_level_to_string_alloc(level
, &s
);
4802 log_info("Setting log level to %s.", strna(s
));
4804 log_set_max_level(level
);
4807 void manager_restore_original_log_level(Manager
*m
) {
4808 _cleanup_free_
char *s
= NULL
;
4811 if (!m
->log_level_overridden
)
4814 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4815 log_info("Restoring log level to original (%s).", strna(s
));
4817 log_set_max_level(m
->original_log_level
);
4818 m
->log_level_overridden
= false;
4821 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4824 if (!m
->log_target_overridden
) {
4825 m
->original_log_target
= log_get_target();
4826 m
->log_target_overridden
= true;
4829 log_info("Setting log target to %s.", log_target_to_string(target
));
4830 log_set_target(target
);
4833 void manager_restore_original_log_target(Manager
*m
) {
4836 if (!m
->log_target_overridden
)
4839 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4841 log_set_target(m
->original_log_target
);
4842 m
->log_target_overridden
= false;
4845 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4847 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4848 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4849 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
4853 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
4854 [MANAGER_INITIALIZING
] = "initializing",
4855 [MANAGER_STARTING
] = "starting",
4856 [MANAGER_RUNNING
] = "running",
4857 [MANAGER_DEGRADED
] = "degraded",
4858 [MANAGER_MAINTENANCE
] = "maintenance",
4859 [MANAGER_STOPPING
] = "stopping",
4862 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
4864 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
4865 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
4866 [MANAGER_TIMESTAMP_LOADER
] = "loader",
4867 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
4868 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
4869 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
4870 [MANAGER_TIMESTAMP_FINISH
] = "finish",
4871 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
4872 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
4873 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
4874 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
4875 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
4876 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
4877 [MANAGER_TIMESTAMP_UNITS_LOAD
] = "units-load",
4878 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
4879 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
4880 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
4881 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
4882 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
4883 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
4886 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
4888 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
4889 [OOM_CONTINUE
] = "continue",
4890 [OOM_STOP
] = "stop",
4891 [OOM_KILL
] = "kill",
4894 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);