1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
7 #include <sys/inotify.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
22 #include "all-units.h"
23 #include "alloc-util.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
33 #include "creds-util.h"
35 #include "dbus-manager.h"
36 #include "dbus-unit.h"
39 #include "dirent-util.h"
42 #include "exec-util.h"
44 #include "exit-status.h"
48 #include "generator-setup.h"
53 #include "load-fragment.h"
54 #include "locale-setup.h"
58 #include "memory-util.h"
60 #include "parse-util.h"
61 #include "path-lookup.h"
62 #include "path-util.h"
63 #include "process-util.h"
64 #include "ratelimit.h"
65 #include "rlimit-util.h"
67 #include "selinux-util.h"
68 #include "serialize.h"
69 #include "signal-util.h"
70 #include "socket-util.h"
72 #include "stat-util.h"
73 #include "string-table.h"
74 #include "string-util.h"
77 #include "sysctl-util.h"
78 #include "syslog-util.h"
79 #include "terminal-util.h"
80 #include "time-util.h"
81 #include "transaction.h"
82 #include "umask-util.h"
83 #include "unit-name.h"
84 #include "unit-serialize.h"
85 #include "user-util.h"
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
98 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
99 * the queue gets more empty. */
100 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
102 /* How many units and jobs to process of the bus queue before returning to the event loop. */
103 #define MANAGER_BUS_MESSAGE_BUDGET 100U
105 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
107 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
108 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
110 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
111 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
112 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
113 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
);
114 static int manager_dispatch_timezone_change(sd_event_source
*source
, const struct inotify_event
*event
, void *userdata
);
115 static int manager_run_environment_generators(Manager
*m
);
116 static int manager_run_generators(Manager
*m
);
117 static void manager_vacuum(Manager
*m
);
119 static usec_t
manager_watch_jobs_next_time(Manager
*m
) {
120 return usec_add(now(CLOCK_MONOTONIC
),
121 show_status_on(m
->show_status
) ? JOBS_IN_PROGRESS_WAIT_USEC
:
122 JOBS_IN_PROGRESS_QUIET_WAIT_USEC
);
125 static void manager_watch_jobs_in_progress(Manager
*m
) {
131 /* We do not want to show the cylon animation if the user
132 * needs to confirm service executions otherwise confirmation
133 * messages will be screwed by the cylon animation. */
134 if (!manager_is_confirm_spawn_disabled(m
))
137 if (m
->jobs_in_progress_event_source
)
140 next
= manager_watch_jobs_next_time(m
);
141 r
= sd_event_add_time(
143 &m
->jobs_in_progress_event_source
,
146 manager_dispatch_jobs_in_progress
, m
);
150 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
153 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
155 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
158 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
159 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
163 p
= mempset(p
, ' ', pos
-2);
164 if (log_get_show_color())
165 p
= stpcpy(p
, ANSI_RED
);
169 if (pos
> 0 && pos
<= width
) {
170 if (log_get_show_color())
171 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
175 if (log_get_show_color())
176 p
= stpcpy(p
, ANSI_NORMAL
);
179 if (log_get_show_color())
180 p
= stpcpy(p
, ANSI_RED
);
183 p
= mempset(p
, ' ', width
-1-pos
);
184 if (log_get_show_color())
185 strcpy(p
, ANSI_NORMAL
);
189 static void manager_flip_auto_status(Manager
*m
, bool enable
, const char *reason
) {
193 if (m
->show_status
== SHOW_STATUS_AUTO
)
194 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
, reason
);
196 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
197 manager_set_show_status(m
, SHOW_STATUS_AUTO
, reason
);
201 static void manager_print_jobs_in_progress(Manager
*m
) {
202 _cleanup_free_
char *job_of_n
= NULL
;
204 unsigned counter
= 0, print_nr
;
205 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
207 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
211 assert(m
->n_running_jobs
> 0);
213 manager_flip_auto_status(m
, true, "delay");
215 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
217 HASHMAP_FOREACH(j
, m
->jobs
)
218 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
221 /* m->n_running_jobs must be consistent with the contents of m->jobs,
222 * so the above loop must have succeeded in finding j. */
223 assert(counter
== print_nr
+ 1);
226 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
228 cylon_pos
= 14 - cylon_pos
;
229 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
231 m
->jobs_in_progress_iteration
++;
233 if (m
->n_running_jobs
> 1) {
234 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
238 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
239 if (job_get_timeout(j
, &x
) > 0)
240 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
242 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
243 "%sA %s job is running for %s (%s / %s)",
245 job_type_to_string(j
->type
),
246 unit_status_string(j
->unit
),
250 static int have_ask_password(void) {
251 _cleanup_closedir_
DIR *dir
= NULL
;
254 dir
= opendir("/run/systemd/ask-password");
262 FOREACH_DIRENT_ALL(de
, dir
, return -errno
) {
263 if (startswith(de
->d_name
, "ask."))
269 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
270 int fd
, uint32_t revents
, void *userdata
) {
271 Manager
*m
= userdata
;
277 m
->have_ask_password
= have_ask_password();
278 if (m
->have_ask_password
< 0)
279 /* Log error but continue. Negative have_ask_password
280 * is treated as unknown status. */
281 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
286 static void manager_close_ask_password(Manager
*m
) {
289 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
290 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
291 m
->have_ask_password
= -EINVAL
;
294 static int manager_check_ask_password(Manager
*m
) {
299 if (!m
->ask_password_event_source
) {
300 assert(m
->ask_password_inotify_fd
< 0);
302 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
304 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
305 if (m
->ask_password_inotify_fd
< 0)
306 return log_error_errno(errno
, "Failed to create inotify object: %m");
308 r
= inotify_add_watch_and_warn(m
->ask_password_inotify_fd
,
309 "/run/systemd/ask-password",
310 IN_CREATE
|IN_DELETE
|IN_MOVE
);
312 manager_close_ask_password(m
);
316 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
317 m
->ask_password_inotify_fd
, EPOLLIN
,
318 manager_dispatch_ask_password_fd
, m
);
320 log_error_errno(r
, "Failed to add event source for /run/systemd/ask-password: %m");
321 manager_close_ask_password(m
);
325 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
327 /* Queries might have been added meanwhile... */
328 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
329 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
332 return m
->have_ask_password
;
335 static int manager_watch_idle_pipe(Manager
*m
) {
340 if (m
->idle_pipe_event_source
)
343 if (m
->idle_pipe
[2] < 0)
346 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
348 return log_error_errno(r
, "Failed to watch idle pipe: %m");
350 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
355 static void manager_close_idle_pipe(Manager
*m
) {
358 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
360 safe_close_pair(m
->idle_pipe
);
361 safe_close_pair(m
->idle_pipe
+ 2);
364 static int manager_setup_time_change(Manager
*m
) {
369 if (MANAGER_IS_TEST_RUN(m
))
372 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
373 m
->time_change_fd
= safe_close(m
->time_change_fd
);
375 m
->time_change_fd
= time_change_fd();
376 if (m
->time_change_fd
< 0)
377 return log_error_errno(m
->time_change_fd
, "Failed to create timer change timer fd: %m");
379 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
381 return log_error_errno(r
, "Failed to create time change event source: %m");
383 /* Schedule this slightly earlier than the .timer event sources */
384 r
= sd_event_source_set_priority(m
->time_change_event_source
, SD_EVENT_PRIORITY_NORMAL
-1);
386 return log_error_errno(r
, "Failed to set priority of time change event sources: %m");
388 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
390 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
395 static int manager_read_timezone_stat(Manager
*m
) {
401 /* Read the current stat() data of /etc/localtime so that we detect changes */
402 if (lstat("/etc/localtime", &st
) < 0) {
403 log_debug_errno(errno
, "Failed to stat /etc/localtime, ignoring: %m");
404 changed
= m
->etc_localtime_accessible
;
405 m
->etc_localtime_accessible
= false;
409 k
= timespec_load(&st
.st_mtim
);
410 changed
= !m
->etc_localtime_accessible
|| k
!= m
->etc_localtime_mtime
;
412 m
->etc_localtime_mtime
= k
;
413 m
->etc_localtime_accessible
= true;
419 static int manager_setup_timezone_change(Manager
*m
) {
420 _cleanup_(sd_event_source_unrefp
) sd_event_source
*new_event
= NULL
;
425 if (MANAGER_IS_TEST_RUN(m
))
428 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
429 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
430 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
431 * went to zero and all fds to it are closed.
433 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
436 * Note that we create the new event source first here, before releasing the old one. This should optimize
437 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
439 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc/localtime",
440 IN_ATTRIB
|IN_MOVE_SELF
|IN_CLOSE_WRITE
|IN_DONT_FOLLOW
, manager_dispatch_timezone_change
, m
);
442 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
443 * O_CREATE or by rename() */
445 log_debug_errno(r
, "/etc/localtime doesn't exist yet, watching /etc instead.");
446 r
= sd_event_add_inotify(m
->event
, &new_event
, "/etc",
447 IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
, manager_dispatch_timezone_change
, m
);
450 return log_error_errno(r
, "Failed to create timezone change event source: %m");
452 /* Schedule this slightly earlier than the .timer event sources */
453 r
= sd_event_source_set_priority(new_event
, SD_EVENT_PRIORITY_NORMAL
-1);
455 return log_error_errno(r
, "Failed to set priority of timezone change event sources: %m");
457 sd_event_source_unref(m
->timezone_change_event_source
);
458 m
->timezone_change_event_source
= TAKE_PTR(new_event
);
463 static int enable_special_signals(Manager
*m
) {
464 _cleanup_close_
int fd
= -1;
468 if (MANAGER_IS_TEST_RUN(m
))
471 /* Enable that we get SIGINT on control-alt-del. In containers
472 * this will fail with EPERM (older) or EINVAL (newer), so
474 if (reboot(RB_DISABLE_CAD
) < 0 && !IN_SET(errno
, EPERM
, EINVAL
))
475 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
477 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
479 /* Support systems without virtual console */
481 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
483 /* Enable that we get SIGWINCH on kbrequest */
484 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
485 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
491 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
493 static int manager_setup_signals(Manager
*m
) {
494 struct sigaction sa
= {
495 .sa_handler
= SIG_DFL
,
496 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
503 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
505 /* We make liberal use of realtime signals here. On
506 * Linux/glibc we have 30 of them (with the exception of Linux
507 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
510 assert_se(sigemptyset(&mask
) == 0);
511 sigset_add_many(&mask
,
512 SIGCHLD
, /* Child died */
513 SIGTERM
, /* Reexecute daemon */
514 SIGHUP
, /* Reload configuration */
515 SIGUSR1
, /* systemd: reconnect to D-Bus */
516 SIGUSR2
, /* systemd: dump status */
517 SIGINT
, /* Kernel sends us this on control-alt-del */
518 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
519 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
521 SIGRTMIN
+0, /* systemd: start default.target */
522 SIGRTMIN
+1, /* systemd: isolate rescue.target */
523 SIGRTMIN
+2, /* systemd: isolate emergency.target */
524 SIGRTMIN
+3, /* systemd: start halt.target */
525 SIGRTMIN
+4, /* systemd: start poweroff.target */
526 SIGRTMIN
+5, /* systemd: start reboot.target */
527 SIGRTMIN
+6, /* systemd: start kexec.target */
529 /* ... space for more special targets ... */
531 SIGRTMIN
+13, /* systemd: Immediate halt */
532 SIGRTMIN
+14, /* systemd: Immediate poweroff */
533 SIGRTMIN
+15, /* systemd: Immediate reboot */
534 SIGRTMIN
+16, /* systemd: Immediate kexec */
536 /* ... space for more immediate system state changes ... */
538 SIGRTMIN
+20, /* systemd: enable status messages */
539 SIGRTMIN
+21, /* systemd: disable status messages */
540 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
541 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
542 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
544 /* .. one free signal here ... */
546 /* Apparently Linux on hppa had fewer RT signals until v3.18,
547 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
548 * see commit v3.17-7614-g1f25df2eff.
550 * We cannot unconditionally make use of those signals here,
551 * so let's use a runtime check. Since these commands are
552 * accessible by different means and only really a safety
553 * net, the missing functionality on hppa shouldn't matter.
556 RTSIG_IF_AVAILABLE(SIGRTMIN
+26), /* systemd: set log target to journal-or-kmsg */
557 RTSIG_IF_AVAILABLE(SIGRTMIN
+27), /* systemd: set log target to console */
558 RTSIG_IF_AVAILABLE(SIGRTMIN
+28), /* systemd: set log target to kmsg */
559 RTSIG_IF_AVAILABLE(SIGRTMIN
+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
561 /* ... one free signal here SIGRTMIN+30 ... */
563 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
565 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
566 if (m
->signal_fd
< 0)
569 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
573 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
575 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
576 * notify processing can still figure out to which process/service a message belongs, before we reap the
577 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
578 * status information before detecting that there's no process in a cgroup. */
579 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
583 if (MANAGER_IS_SYSTEM(m
))
584 return enable_special_signals(m
);
589 static char** sanitize_environment(char **l
) {
591 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
595 "CONFIGURATION_DIRECTORY",
596 "CREDENTIALS_DIRECTORY",
618 /* Let's order the environment alphabetically, just to make it pretty */
624 int manager_default_environment(Manager
*m
) {
629 m
->transient_environment
= strv_free(m
->transient_environment
);
631 if (MANAGER_IS_SYSTEM(m
)) {
632 /* The system manager always starts with a clean
633 * environment for its children. It does not import
634 * the kernel's or the parents' exported variables.
636 * The initial passed environment is untouched to keep
637 * /proc/self/environ valid; it is used for tagging
638 * the init process inside containers. */
639 m
->transient_environment
= strv_new("PATH=" DEFAULT_PATH
);
640 if (!m
->transient_environment
)
643 /* Import locale variables LC_*= from configuration */
644 (void) locale_setup(&m
->transient_environment
);
646 /* The user manager passes its own environment along to its children, except for $PATH. */
647 m
->transient_environment
= strv_copy(environ
);
648 if (!m
->transient_environment
)
651 r
= strv_env_replace_strdup(&m
->transient_environment
, "PATH=" DEFAULT_USER_PATH
);
656 sanitize_environment(m
->transient_environment
);
661 static int manager_setup_prefix(Manager
*m
) {
667 static const struct table_entry paths_system
[_EXEC_DIRECTORY_TYPE_MAX
] = {
668 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_SYSTEM_RUNTIME
, NULL
},
669 [EXEC_DIRECTORY_STATE
] = { SD_PATH_SYSTEM_STATE_PRIVATE
, NULL
},
670 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_SYSTEM_STATE_CACHE
, NULL
},
671 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_SYSTEM_STATE_LOGS
, NULL
},
672 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_SYSTEM_CONFIGURATION
, NULL
},
675 static const struct table_entry paths_user
[_EXEC_DIRECTORY_TYPE_MAX
] = {
676 [EXEC_DIRECTORY_RUNTIME
] = { SD_PATH_USER_RUNTIME
, NULL
},
677 [EXEC_DIRECTORY_STATE
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
678 [EXEC_DIRECTORY_CACHE
] = { SD_PATH_USER_STATE_CACHE
, NULL
},
679 [EXEC_DIRECTORY_LOGS
] = { SD_PATH_USER_CONFIGURATION
, "log" },
680 [EXEC_DIRECTORY_CONFIGURATION
] = { SD_PATH_USER_CONFIGURATION
, NULL
},
685 const struct table_entry
*p
= MANAGER_IS_SYSTEM(m
) ? paths_system
: paths_user
;
688 for (ExecDirectoryType i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++) {
689 r
= sd_path_lookup(p
[i
].type
, p
[i
].suffix
, &m
->prefix
[i
]);
691 return log_warning_errno(r
, "Failed to lookup %s path: %m",
692 exec_directory_type_to_string(i
));
698 static void manager_free_unit_name_maps(Manager
*m
) {
699 m
->unit_id_map
= hashmap_free(m
->unit_id_map
);
700 m
->unit_name_map
= hashmap_free(m
->unit_name_map
);
701 m
->unit_path_cache
= set_free(m
->unit_path_cache
);
702 m
->unit_cache_timestamp_hash
= 0;
705 static int manager_setup_run_queue(Manager
*m
) {
709 assert(!m
->run_queue_event_source
);
711 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
715 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
719 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
723 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
728 static int manager_setup_sigchld_event_source(Manager
*m
) {
732 assert(!m
->sigchld_event_source
);
734 r
= sd_event_add_defer(m
->event
, &m
->sigchld_event_source
, manager_dispatch_sigchld
, m
);
738 r
= sd_event_source_set_priority(m
->sigchld_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
742 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
746 (void) sd_event_source_set_description(m
->sigchld_event_source
, "manager-sigchld");
751 int manager_new(UnitFileScope scope
, ManagerTestRunFlags test_run_flags
, Manager
**_m
) {
752 _cleanup_(manager_freep
) Manager
*m
= NULL
;
757 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
764 .unit_file_scope
= scope
,
765 .objective
= _MANAGER_OBJECTIVE_INVALID
,
767 .status_unit_format
= STATUS_UNIT_FORMAT_DEFAULT
,
769 .default_timer_accuracy_usec
= USEC_PER_MINUTE
,
770 .default_memory_accounting
= MEMORY_ACCOUNTING_DEFAULT
,
771 .default_tasks_accounting
= true,
772 .default_tasks_max
= TASKS_MAX_UNSET
,
773 .default_timeout_start_usec
= DEFAULT_TIMEOUT_USEC
,
774 .default_timeout_stop_usec
= DEFAULT_TIMEOUT_USEC
,
775 .default_restart_usec
= DEFAULT_RESTART_USEC
,
777 .original_log_level
= -1,
778 .original_log_target
= _LOG_TARGET_INVALID
,
780 .watchdog_overridden
[WATCHDOG_RUNTIME
] = USEC_INFINITY
,
781 .watchdog_overridden
[WATCHDOG_REBOOT
] = USEC_INFINITY
,
782 .watchdog_overridden
[WATCHDOG_KEXEC
] = USEC_INFINITY
,
784 .show_status_overridden
= _SHOW_STATUS_INVALID
,
787 .cgroups_agent_fd
= -1,
789 .time_change_fd
= -1,
790 .user_lookup_fds
= { -1, -1 },
791 .private_listen_fd
= -1,
793 .cgroup_inotify_fd
= -1,
794 .pin_cgroupfs_fd
= -1,
795 .ask_password_inotify_fd
= -1,
796 .idle_pipe
= { -1, -1, -1, -1},
798 /* start as id #1, so that we can leave #0 around as "null-like" value */
801 .have_ask_password
= -EINVAL
, /* we don't know */
803 .test_run_flags
= test_run_flags
,
805 .default_oom_policy
= OOM_STOP
,
809 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
810 boot_timestamps(m
->timestamps
+ MANAGER_TIMESTAMP_USERSPACE
,
811 m
->timestamps
+ MANAGER_TIMESTAMP_FIRMWARE
,
812 m
->timestamps
+ MANAGER_TIMESTAMP_LOADER
);
815 /* Prepare log fields we can use for structured logging */
816 if (MANAGER_IS_SYSTEM(m
)) {
817 m
->unit_log_field
= "UNIT=";
818 m
->unit_log_format_string
= "UNIT=%s";
820 m
->invocation_log_field
= "INVOCATION_ID=";
821 m
->invocation_log_format_string
= "INVOCATION_ID=%s";
823 m
->unit_log_field
= "USER_UNIT=";
824 m
->unit_log_format_string
= "USER_UNIT=%s";
826 m
->invocation_log_field
= "USER_INVOCATION_ID=";
827 m
->invocation_log_format_string
= "USER_INVOCATION_ID=%s";
830 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
831 m
->ctrl_alt_del_ratelimit
= (RateLimit
) { .interval
= 2 * USEC_PER_SEC
, .burst
= 7 };
833 r
= manager_default_environment(m
);
837 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
841 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &path_hash_ops
);
845 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
849 r
= prioq_ensure_allocated(&m
->run_queue
, compare_job_priority
);
853 r
= manager_setup_prefix(m
);
857 r
= get_credentials_dir(&e
);
859 m
->received_credentials
= strdup(e
);
860 if (!m
->received_credentials
)
864 r
= sd_event_default(&m
->event
);
868 r
= manager_setup_run_queue(m
);
872 if (test_run_flags
== MANAGER_TEST_RUN_MINIMAL
) {
873 m
->cgroup_root
= strdup("");
877 r
= manager_setup_signals(m
);
881 r
= manager_setup_cgroup(m
);
885 r
= manager_setup_time_change(m
);
889 r
= manager_read_timezone_stat(m
);
893 (void) manager_setup_timezone_change(m
);
895 r
= manager_setup_sigchld_event_source(m
);
900 if (test_run_flags
== 0) {
901 if (MANAGER_IS_SYSTEM(m
))
902 r
= mkdir_label("/run/systemd/units", 0755);
904 _cleanup_free_
char *units_path
= NULL
;
905 r
= xdg_user_runtime_dir(&units_path
, "/systemd/units");
908 r
= mkdir_p_label(units_path
, 0755);
911 if (r
< 0 && r
!= -EEXIST
)
917 dir_is_empty("/usr") > 0;
919 /* Note that we do not set up the notify fd here. We do that after deserialization,
920 * since they might have gotten serialized across the reexec. */
927 static int manager_setup_notify(Manager
*m
) {
930 if (MANAGER_IS_TEST_RUN(m
))
933 if (m
->notify_fd
< 0) {
934 _cleanup_close_
int fd
= -1;
935 union sockaddr_union sa
;
938 /* First free all secondary fields */
939 m
->notify_socket
= mfree(m
->notify_socket
);
940 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
942 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
944 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
946 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
948 m
->notify_socket
= path_join(m
->prefix
[EXEC_DIRECTORY_RUNTIME
], "systemd/notify");
949 if (!m
->notify_socket
)
952 r
= sockaddr_un_set_path(&sa
.un
, m
->notify_socket
);
954 return log_error_errno(r
, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
958 (void) mkdir_parents_label(m
->notify_socket
, 0755);
959 (void) sockaddr_un_unlink(&sa
.un
);
961 r
= mac_selinux_bind(fd
, &sa
.sa
, sa_len
);
963 return log_error_errno(r
, "bind(%s) failed: %m", m
->notify_socket
);
965 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
967 return log_error_errno(r
, "SO_PASSCRED failed: %m");
969 m
->notify_fd
= TAKE_FD(fd
);
971 log_debug("Using notification socket %s", m
->notify_socket
);
974 if (!m
->notify_event_source
) {
975 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
977 return log_error_errno(r
, "Failed to allocate notify event source: %m");
979 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
980 * service an exit message belongs. */
981 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
983 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
985 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
991 static int manager_setup_cgroups_agent(Manager
*m
) {
993 static const union sockaddr_union sa
= {
994 .un
.sun_family
= AF_UNIX
,
995 .un
.sun_path
= "/run/systemd/cgroups-agent",
999 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1000 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1001 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1002 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1003 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1004 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1005 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1006 * we thus won't lose messages.
1008 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1009 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1010 * bus for these messages. */
1012 if (MANAGER_IS_TEST_RUN(m
))
1015 if (!MANAGER_IS_SYSTEM(m
))
1018 r
= cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER
);
1020 return log_error_errno(r
, "Failed to determine whether unified cgroups hierarchy is used: %m");
1021 if (r
> 0) /* We don't need this anymore on the unified hierarchy */
1024 if (m
->cgroups_agent_fd
< 0) {
1025 _cleanup_close_
int fd
= -1;
1027 /* First free all secondary fields */
1028 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
1030 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1032 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
1034 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
1036 (void) sockaddr_un_unlink(&sa
.un
);
1038 /* Only allow root to connect to this socket */
1039 RUN_WITH_UMASK(0077)
1040 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
1042 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
1044 m
->cgroups_agent_fd
= TAKE_FD(fd
);
1047 if (!m
->cgroups_agent_event_source
) {
1048 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
1050 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
1052 /* Process cgroups notifications early. Note that when the agent notification is received
1053 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1054 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1055 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-9);
1057 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
1059 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
1065 static int manager_setup_user_lookup_fd(Manager
*m
) {
1070 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1071 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1072 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1073 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1074 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1075 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1076 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1077 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1080 * You might wonder why we need a communication channel for this that is independent of the usual notification
1081 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1082 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1083 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1085 * Note that this function is called under two circumstances: when we first initialize (in which case we
1086 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1087 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1089 if (m
->user_lookup_fds
[0] < 0) {
1091 /* Free all secondary fields */
1092 safe_close_pair(m
->user_lookup_fds
);
1093 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
1095 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
1096 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
1098 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
1101 if (!m
->user_lookup_event_source
) {
1102 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
1104 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
1106 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1108 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-11);
1110 return log_error_errno(errno
, "Failed to set priority of user lookup event source: %m");
1112 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
1118 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
1124 while ((u
= m
->cleanup_queue
)) {
1125 assert(u
->in_cleanup_queue
);
1135 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
1136 GC_OFFSET_UNSURE
, /* No clue */
1137 GC_OFFSET_GOOD
, /* We still need this unit */
1138 GC_OFFSET_BAD
, /* We don't need this unit anymore */
1142 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
1145 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
1147 /* Recursively mark referenced units as GOOD as well */
1148 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCES
)
1149 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
1150 unit_gc_mark_good(other
, gc_marker
);
1153 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
1159 if (IN_SET(u
->gc_marker
- gc_marker
,
1160 GC_OFFSET_GOOD
, GC_OFFSET_BAD
, GC_OFFSET_UNSURE
, GC_OFFSET_IN_PATH
))
1163 if (u
->in_cleanup_queue
)
1166 if (!unit_may_gc(u
))
1169 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
1173 UNIT_FOREACH_DEPENDENCY(other
, u
, UNIT_ATOM_REFERENCED_BY
) {
1174 unit_gc_sweep(other
, gc_marker
);
1176 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1179 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1184 LIST_FOREACH(refs_by_target
, ref
, u
->refs_by_target
) {
1185 unit_gc_sweep(ref
->source
, gc_marker
);
1187 if (ref
->source
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
1190 if (ref
->source
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
1197 /* We were unable to find anything out about this entry, so
1198 * let's investigate it later */
1199 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
1200 unit_add_to_gc_queue(u
);
1204 /* We definitely know that this one is not useful anymore, so
1205 * let's mark it for deletion */
1206 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1207 unit_add_to_cleanup_queue(u
);
1211 unit_gc_mark_good(u
, gc_marker
);
1214 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
1215 unsigned n
= 0, gc_marker
;
1220 /* log_debug("Running GC..."); */
1222 m
->gc_marker
+= _GC_OFFSET_MAX
;
1223 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1226 gc_marker
= m
->gc_marker
;
1228 while ((u
= m
->gc_unit_queue
)) {
1229 assert(u
->in_gc_queue
);
1231 unit_gc_sweep(u
, gc_marker
);
1233 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1234 u
->in_gc_queue
= false;
1238 if (IN_SET(u
->gc_marker
- gc_marker
,
1239 GC_OFFSET_BAD
, GC_OFFSET_UNSURE
)) {
1241 log_unit_debug(u
, "Collecting.");
1242 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1243 unit_add_to_cleanup_queue(u
);
1250 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1256 while ((j
= m
->gc_job_queue
)) {
1257 assert(j
->in_gc_queue
);
1259 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1260 j
->in_gc_queue
= false;
1267 log_unit_debug(j
->unit
, "Collecting job.");
1268 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1274 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager
*m
) {
1281 while ((u
= m
->stop_when_unneeded_queue
)) {
1282 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1284 assert(u
->in_stop_when_unneeded_queue
);
1285 LIST_REMOVE(stop_when_unneeded_queue
, m
->stop_when_unneeded_queue
, u
);
1286 u
->in_stop_when_unneeded_queue
= false;
1290 if (!unit_is_unneeded(u
))
1293 log_unit_debug(u
, "Unit is not needed anymore.");
1295 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1296 * service being unnecessary after a while. */
1298 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1299 log_unit_warning(u
, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1303 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1304 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1306 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1312 static unsigned manager_dispatch_start_when_upheld_queue(Manager
*m
) {
1319 while ((u
= m
->start_when_upheld_queue
)) {
1320 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1321 Unit
*culprit
= NULL
;
1323 assert(u
->in_start_when_upheld_queue
);
1324 LIST_REMOVE(start_when_upheld_queue
, m
->start_when_upheld_queue
, u
);
1325 u
->in_start_when_upheld_queue
= false;
1329 if (!unit_is_upheld_by_active(u
, &culprit
))
1332 log_unit_debug(u
, "Unit is started because upheld by active unit %s.", culprit
->id
);
1334 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1335 * service being unnecessary after a while. */
1337 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1338 log_unit_warning(u
, "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.", culprit
->id
);
1342 r
= manager_add_job(u
->manager
, JOB_START
, u
, JOB_FAIL
, NULL
, &error
, NULL
);
1344 log_unit_warning_errno(u
, r
, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error
, r
));
1350 static unsigned manager_dispatch_stop_when_bound_queue(Manager
*m
) {
1357 while ((u
= m
->stop_when_bound_queue
)) {
1358 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1359 Unit
*culprit
= NULL
;
1361 assert(u
->in_stop_when_bound_queue
);
1362 LIST_REMOVE(stop_when_bound_queue
, m
->stop_when_bound_queue
, u
);
1363 u
->in_stop_when_bound_queue
= false;
1367 if (!unit_is_bound_by_inactive(u
, &culprit
))
1370 log_unit_debug(u
, "Unit is stopped because bound to inactive unit %s.", culprit
->id
);
1372 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1373 * service being unnecessary after a while. */
1375 if (!ratelimit_below(&u
->auto_start_stop_ratelimit
)) {
1376 log_unit_warning(u
, "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.", culprit
->id
);
1380 r
= manager_add_job(u
->manager
, JOB_STOP
, u
, JOB_REPLACE
, NULL
, &error
, NULL
);
1382 log_unit_warning_errno(u
, r
, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error
, r
));
1388 static void manager_clear_jobs_and_units(Manager
*m
) {
1393 while ((u
= hashmap_first(m
->units
)))
1396 manager_dispatch_cleanup_queue(m
);
1398 assert(!m
->load_queue
);
1399 assert(prioq_isempty(m
->run_queue
));
1400 assert(!m
->dbus_unit_queue
);
1401 assert(!m
->dbus_job_queue
);
1402 assert(!m
->cleanup_queue
);
1403 assert(!m
->gc_unit_queue
);
1404 assert(!m
->gc_job_queue
);
1405 assert(!m
->stop_when_unneeded_queue
);
1406 assert(!m
->start_when_upheld_queue
);
1407 assert(!m
->stop_when_bound_queue
);
1409 assert(hashmap_isempty(m
->jobs
));
1410 assert(hashmap_isempty(m
->units
));
1412 m
->n_on_console
= 0;
1413 m
->n_running_jobs
= 0;
1414 m
->n_installed_jobs
= 0;
1415 m
->n_failed_jobs
= 0;
1418 Manager
* manager_free(Manager
*m
) {
1422 manager_clear_jobs_and_units(m
);
1424 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1425 if (unit_vtable
[c
]->shutdown
)
1426 unit_vtable
[c
]->shutdown(m
);
1428 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1429 manager_shutdown_cgroup(m
, IN_SET(m
->objective
, MANAGER_EXIT
, MANAGER_REBOOT
, MANAGER_POWEROFF
, MANAGER_HALT
, MANAGER_KEXEC
));
1431 lookup_paths_flush_generator(&m
->lookup_paths
);
1434 manager_varlink_done(m
);
1436 exec_runtime_vacuum(m
);
1437 hashmap_free(m
->exec_runtime_by_id
);
1439 dynamic_user_vacuum(m
, false);
1440 hashmap_free(m
->dynamic_users
);
1442 hashmap_free(m
->units
);
1443 hashmap_free(m
->units_by_invocation_id
);
1444 hashmap_free(m
->jobs
);
1445 hashmap_free(m
->watch_pids
);
1446 hashmap_free(m
->watch_bus
);
1448 prioq_free(m
->run_queue
);
1450 set_free(m
->startup_units
);
1451 set_free(m
->failed_units
);
1453 sd_event_source_unref(m
->signal_event_source
);
1454 sd_event_source_unref(m
->sigchld_event_source
);
1455 sd_event_source_unref(m
->notify_event_source
);
1456 sd_event_source_unref(m
->cgroups_agent_event_source
);
1457 sd_event_source_unref(m
->time_change_event_source
);
1458 sd_event_source_unref(m
->timezone_change_event_source
);
1459 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1460 sd_event_source_unref(m
->run_queue_event_source
);
1461 sd_event_source_unref(m
->user_lookup_event_source
);
1463 safe_close(m
->signal_fd
);
1464 safe_close(m
->notify_fd
);
1465 safe_close(m
->cgroups_agent_fd
);
1466 safe_close(m
->time_change_fd
);
1467 safe_close_pair(m
->user_lookup_fds
);
1469 manager_close_ask_password(m
);
1471 manager_close_idle_pipe(m
);
1473 sd_event_unref(m
->event
);
1475 free(m
->notify_socket
);
1477 lookup_paths_free(&m
->lookup_paths
);
1478 strv_free(m
->transient_environment
);
1479 strv_free(m
->client_environment
);
1481 hashmap_free(m
->cgroup_unit
);
1482 manager_free_unit_name_maps(m
);
1484 free(m
->switch_root
);
1485 free(m
->switch_root_init
);
1487 rlimit_free_all(m
->rlimit
);
1489 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1490 hashmap_free(m
->units_requiring_mounts_for
);
1492 hashmap_free(m
->uid_refs
);
1493 hashmap_free(m
->gid_refs
);
1495 for (ExecDirectoryType dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++)
1496 m
->prefix
[dt
] = mfree(m
->prefix
[dt
]);
1497 free(m
->received_credentials
);
1502 static void manager_enumerate_perpetual(Manager
*m
) {
1505 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1508 /* Let's ask every type to load all units from disk/kernel that it might know */
1509 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1510 if (!unit_type_supported(c
)) {
1511 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1515 if (unit_vtable
[c
]->enumerate_perpetual
)
1516 unit_vtable
[c
]->enumerate_perpetual(m
);
1520 static void manager_enumerate(Manager
*m
) {
1523 if (m
->test_run_flags
== MANAGER_TEST_RUN_MINIMAL
)
1526 /* Let's ask every type to load all units from disk/kernel that it might know */
1527 for (UnitType c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1528 if (!unit_type_supported(c
)) {
1529 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1533 if (unit_vtable
[c
]->enumerate
)
1534 unit_vtable
[c
]->enumerate(m
);
1537 manager_dispatch_load_queue(m
);
1540 static void manager_coldplug(Manager
*m
) {
1547 log_debug("Invoking unit coldplug() handlers…");
1549 /* Let's place the units back into their deserialized state */
1550 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1552 /* ignore aliases */
1556 r
= unit_coldplug(u
);
1558 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1562 static void manager_catchup(Manager
*m
) {
1568 log_debug("Invoking unit catchup() handlers…");
1570 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1571 HASHMAP_FOREACH_KEY(u
, k
, m
->units
) {
1573 /* ignore aliases */
1581 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1586 HASHMAP_FOREACH(u
, m
->units
) {
1588 if (fdset_size(fds
) <= 0)
1591 if (!UNIT_VTABLE(u
)->distribute_fds
)
1594 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1598 static bool manager_dbus_is_running(Manager
*m
, bool deserialized
) {
1603 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1604 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1605 * rather than the current one. */
1607 if (MANAGER_IS_TEST_RUN(m
))
1610 u
= manager_get_unit(m
, SPECIAL_DBUS_SOCKET
);
1613 if ((deserialized
? SOCKET(u
)->deserialized_state
: SOCKET(u
)->state
) != SOCKET_RUNNING
)
1616 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1619 if (!IN_SET((deserialized
? SERVICE(u
)->deserialized_state
: SERVICE(u
)->state
), SERVICE_RUNNING
, SERVICE_RELOAD
))
1625 static void manager_setup_bus(Manager
*m
) {
1628 /* Let's set up our private bus connection now, unconditionally */
1629 (void) bus_init_private(m
);
1631 /* If we are in --user mode also connect to the system bus now */
1632 if (MANAGER_IS_USER(m
))
1633 (void) bus_init_system(m
);
1635 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1636 if (manager_dbus_is_running(m
, MANAGER_IS_RELOADING(m
))) {
1637 (void) bus_init_api(m
);
1639 if (MANAGER_IS_SYSTEM(m
))
1640 (void) bus_init_system(m
);
1644 static void manager_preset_all(Manager
*m
) {
1649 if (m
->first_boot
<= 0)
1652 if (!MANAGER_IS_SYSTEM(m
))
1655 if (MANAGER_IS_TEST_RUN(m
))
1658 /* If this is the first boot, and we are in the host system, then preset everything */
1659 r
= unit_file_preset_all(UNIT_FILE_SYSTEM
, 0, NULL
, UNIT_FILE_PRESET_ENABLE_ONLY
, NULL
, 0);
1661 log_full_errno(r
== -EEXIST
? LOG_NOTICE
: LOG_WARNING
, r
,
1662 "Failed to populate /etc with preset unit settings, ignoring: %m");
1664 log_info("Populated /etc with preset unit settings.");
1667 static void manager_ready(Manager
*m
) {
1670 /* After having loaded everything, do the final round of catching up with what might have changed */
1672 m
->objective
= MANAGER_OK
; /* Tell everyone we are up now */
1674 /* It might be safe to log to the journal now and connect to dbus */
1675 manager_recheck_journal(m
);
1676 manager_recheck_dbus(m
);
1678 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1681 m
->honor_device_enumeration
= true;
1684 static Manager
* manager_reloading_start(Manager
*m
) {
1688 static void manager_reloading_stopp(Manager
**m
) {
1690 assert((*m
)->n_reloading
> 0);
1691 (*m
)->n_reloading
--;
1695 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1700 /* If we are running in test mode, we still want to run the generators,
1701 * but we should not touch the real generator directories. */
1702 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
,
1703 MANAGER_IS_TEST_RUN(m
) ? LOOKUP_PATHS_TEMPORARY_GENERATED
: 0,
1706 return log_error_errno(r
, "Failed to initialize path lookup table: %m");
1708 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START
));
1709 r
= manager_run_environment_generators(m
);
1711 r
= manager_run_generators(m
);
1712 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH
));
1716 manager_preset_all(m
);
1718 lookup_paths_log(&m
->lookup_paths
);
1721 /* This block is (optionally) done with the reloading counter bumped */
1722 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
1724 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1725 * counter here already */
1727 reloading
= manager_reloading_start(m
);
1729 /* First, enumerate what we can from all config files */
1730 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START
));
1731 manager_enumerate_perpetual(m
);
1732 manager_enumerate(m
);
1733 dual_timestamp_get(m
->timestamps
+ manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
));
1735 /* Second, deserialize if there is something to deserialize */
1736 if (serialization
) {
1737 r
= manager_deserialize(m
, serialization
, fds
);
1739 return log_error_errno(r
, "Deserialization failed: %m");
1742 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1743 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1745 manager_distribute_fds(m
, fds
);
1747 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1748 r
= manager_setup_notify(m
);
1750 /* No sense to continue without notifications, our children would fail anyway. */
1753 r
= manager_setup_cgroups_agent(m
);
1755 /* Likewise, no sense to continue without empty cgroup notifications. */
1758 r
= manager_setup_user_lookup_fd(m
);
1760 /* This shouldn't fail, except if things are really broken. */
1763 /* Connect to the bus if we are good for it */
1764 manager_setup_bus(m
);
1766 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1767 r
= bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1769 log_warning_errno(r
, "Failed to deserialized tracked clients, ignoring: %m");
1770 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1772 r
= manager_varlink_init(m
);
1774 log_warning_errno(r
, "Failed to set up Varlink server, ignoring: %m");
1776 /* Third, fire things up! */
1777 manager_coldplug(m
);
1779 /* Clean up runtime objects */
1783 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1784 * reload is finished */
1785 m
->send_reloading_done
= true;
1793 int manager_add_job(
1799 sd_bus_error
*error
,
1806 assert(type
< _JOB_TYPE_MAX
);
1808 assert(mode
< _JOB_MODE_MAX
);
1810 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1811 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1813 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1814 return sd_bus_error_set(error
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1816 if (mode
== JOB_TRIGGERING
&& type
!= JOB_STOP
)
1817 return sd_bus_error_set(error
, SD_BUS_ERROR_INVALID_ARGS
, "--job-mode=triggering is only valid for stop.");
1819 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1821 type
= job_type_collapse(type
, unit
);
1823 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1827 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1828 IN_SET(mode
, JOB_IGNORE_DEPENDENCIES
, JOB_IGNORE_REQUIREMENTS
),
1829 mode
== JOB_IGNORE_DEPENDENCIES
, error
);
1833 if (mode
== JOB_ISOLATE
) {
1834 r
= transaction_add_isolate_jobs(tr
, m
);
1839 if (mode
== JOB_TRIGGERING
) {
1840 r
= transaction_add_triggering_jobs(tr
, unit
);
1845 r
= transaction_activate(tr
, m
, mode
, affected_jobs
, error
);
1849 log_unit_debug(unit
,
1850 "Enqueued job %s/%s as %u", unit
->id
,
1851 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1854 *ret
= tr
->anchor_job
;
1856 transaction_free(tr
);
1860 transaction_abort(tr
);
1861 transaction_free(tr
);
1865 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, sd_bus_error
*e
, Job
**ret
) {
1866 Unit
*unit
= NULL
; /* just to appease gcc, initialization is not really necessary */
1870 assert(type
< _JOB_TYPE_MAX
);
1872 assert(mode
< _JOB_MODE_MAX
);
1874 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1879 return manager_add_job(m
, type
, unit
, mode
, affected_jobs
, e
, ret
);
1882 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Set
*affected_jobs
, Job
**ret
) {
1883 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1887 assert(type
< _JOB_TYPE_MAX
);
1889 assert(mode
< _JOB_MODE_MAX
);
1891 r
= manager_add_job_by_name(m
, type
, name
, mode
, affected_jobs
, &error
, ret
);
1893 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1898 int manager_propagate_reload(Manager
*m
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
) {
1904 assert(mode
< _JOB_MODE_MAX
);
1905 assert(mode
!= JOB_ISOLATE
); /* Isolate is only valid for start */
1907 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1911 /* We need an anchor job */
1912 r
= transaction_add_job_and_dependencies(tr
, JOB_NOP
, unit
, NULL
, false, false, true, true, e
);
1916 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1917 transaction_add_propagate_reload_jobs(tr
, unit
, tr
->anchor_job
, mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1919 r
= transaction_activate(tr
, m
, mode
, NULL
, e
);
1923 transaction_free(tr
);
1927 transaction_abort(tr
);
1928 transaction_free(tr
);
1932 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1935 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1938 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1942 return hashmap_get(m
->units
, name
);
1945 static int manager_dispatch_target_deps_queue(Manager
*m
) {
1951 while ((u
= m
->target_deps_queue
)) {
1952 _cleanup_free_ Unit
**targets
= NULL
;
1955 assert(u
->in_target_deps_queue
);
1957 LIST_REMOVE(target_deps_queue
, u
->manager
->target_deps_queue
, u
);
1958 u
->in_target_deps_queue
= false;
1960 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
1961 * dependencies, and we can't have it that hash tables we iterate through are modified while
1962 * we are iterating through them. */
1963 n_targets
= unit_get_dependency_array(u
, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES
, &targets
);
1967 for (int i
= 0; i
< n_targets
; i
++) {
1968 r
= unit_add_default_target_dependency(u
, targets
[i
]);
1977 unsigned manager_dispatch_load_queue(Manager
*m
) {
1983 /* Make sure we are not run recursively */
1984 if (m
->dispatching_load_queue
)
1987 m
->dispatching_load_queue
= true;
1989 /* Dispatches the load queue. Takes a unit from the queue and
1990 * tries to load its data until the queue is empty */
1992 while ((u
= m
->load_queue
)) {
1993 assert(u
->in_load_queue
);
1999 m
->dispatching_load_queue
= false;
2001 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2002 * should be loaded and have aliases resolved */
2003 (void) manager_dispatch_target_deps_queue(m
);
2008 bool manager_unit_cache_should_retry_load(Unit
*u
) {
2011 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2012 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2013 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2014 if (u
->load_state
!= UNIT_NOT_FOUND
)
2017 /* The cache has been updated since the last time we tried to load the unit. There might be new
2018 * fragment paths to read. */
2019 if (u
->manager
->unit_cache_timestamp_hash
!= u
->fragment_not_found_timestamp_hash
)
2022 /* The cache needs to be updated because there are modifications on disk. */
2023 return !lookup_paths_timestamp_hash_same(&u
->manager
->lookup_paths
, u
->manager
->unit_cache_timestamp_hash
, NULL
);
2026 int manager_load_unit_prepare(
2033 _cleanup_(unit_freep
) Unit
*cleanup_ret
= NULL
;
2041 /* This will prepare the unit for loading, but not actually load anything from disk. */
2043 if (path
&& !path_is_absolute(path
))
2044 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
2047 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2048 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2049 * but this cannot be possible in any code path (See #6119). */
2051 name
= basename(path
);
2054 t
= unit_name_to_type(name
);
2056 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
2057 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
2058 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
2060 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
2063 ret
= manager_get_unit(m
, name
);
2065 /* The time-based cache allows to start new units without daemon-reload,
2066 * but if they are already referenced (because of dependencies or ordering)
2067 * then we have to force a load of the fragment. As an optimization, check
2068 * first if anything in the usual paths was modified since the last time
2069 * the cache was loaded. Also check if the last time an attempt to load the
2070 * unit was made was before the most recent cache refresh, so that we know
2071 * we need to try again — even if the cache is current, it might have been
2072 * updated in a different context before we had a chance to retry loading
2073 * this particular unit. */
2074 if (manager_unit_cache_should_retry_load(ret
))
2075 ret
->load_state
= UNIT_STUB
;
2081 ret
= cleanup_ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
2087 r
= free_and_strdup(&ret
->fragment_path
, path
);
2092 r
= unit_add_name(ret
, name
);
2096 unit_add_to_load_queue(ret
);
2097 unit_add_to_dbus_queue(ret
);
2098 unit_add_to_gc_queue(ret
);
2106 int manager_load_unit(
2118 /* This will load the service information files, but not actually
2119 * start any services or anything. */
2121 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
2125 manager_dispatch_load_queue(m
);
2127 *_ret
= unit_follow_merge(*_ret
);
2131 int manager_load_startable_unit_or_warn(
2137 /* Load a unit, make sure it loaded fully and is not masked. */
2139 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2143 r
= manager_load_unit(m
, name
, path
, &error
, &unit
);
2145 return log_error_errno(r
, "Failed to load %s %s: %s",
2146 name
? "unit" : "unit file", name
?: path
,
2147 bus_error_message(&error
, r
));
2149 r
= bus_unit_validate_load_state(unit
, &error
);
2151 return log_error_errno(r
, "%s", bus_error_message(&error
, r
));
2157 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
2163 HASHMAP_FOREACH(j
, s
->jobs
)
2164 job_dump(j
, f
, prefix
);
2167 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
2174 HASHMAP_FOREACH_KEY(u
, t
, s
->units
)
2176 unit_dump(u
, f
, prefix
);
2179 void manager_dump(Manager
*m
, FILE *f
, const char *prefix
) {
2183 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
2184 const dual_timestamp
*t
= m
->timestamps
+ q
;
2185 char buf
[CONST_MAX(FORMAT_TIMESPAN_MAX
, FORMAT_TIMESTAMP_MAX
)];
2187 if (dual_timestamp_is_set(t
))
2188 fprintf(f
, "%sTimestamp %s: %s\n",
2190 manager_timestamp_to_string(q
),
2191 timestamp_is_set(t
->realtime
) ? format_timestamp(buf
, sizeof buf
, t
->realtime
) :
2192 format_timespan(buf
, sizeof buf
, t
->monotonic
, 1));
2195 manager_dump_units(m
, f
, prefix
);
2196 manager_dump_jobs(m
, f
, prefix
);
2199 int manager_get_dump_string(Manager
*m
, char **ret
) {
2200 _cleanup_free_
char *dump
= NULL
;
2201 _cleanup_fclose_
FILE *f
= NULL
;
2208 f
= open_memstream_unlocked(&dump
, &size
);
2212 manager_dump(m
, f
, NULL
);
2214 r
= fflush_and_check(f
);
2220 *ret
= TAKE_PTR(dump
);
2225 void manager_clear_jobs(Manager
*m
) {
2230 while ((j
= hashmap_first(m
->jobs
)))
2231 /* No need to recurse. We're cancelling all jobs. */
2232 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
2235 void manager_unwatch_pid(Manager
*m
, pid_t pid
) {
2238 /* First let's drop the unit keyed as "pid". */
2239 (void) hashmap_remove(m
->watch_pids
, PID_TO_PTR(pid
));
2241 /* Then, let's also drop the array keyed by -pid. */
2242 free(hashmap_remove(m
->watch_pids
, PID_TO_PTR(-pid
)));
2245 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
2246 Manager
*m
= userdata
;
2252 while ((j
= prioq_peek(m
->run_queue
))) {
2253 assert(j
->installed
);
2254 assert(j
->in_run_queue
);
2256 (void) job_run_and_invalidate(j
);
2259 if (m
->n_running_jobs
> 0)
2260 manager_watch_jobs_in_progress(m
);
2262 if (m
->n_on_console
> 0)
2263 manager_watch_idle_pipe(m
);
2268 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
2269 unsigned n
= 0, budget
;
2275 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2276 * as we can. There's no point in throttling generation of signals in that case. */
2277 if (MANAGER_IS_RELOADING(m
) || m
->send_reloading_done
|| m
->pending_reload_message
)
2278 budget
= UINT_MAX
; /* infinite budget in this case */
2280 /* Anything to do at all? */
2281 if (!m
->dbus_unit_queue
&& !m
->dbus_job_queue
)
2284 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2285 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2286 if (manager_bus_n_queued_write(m
) > MANAGER_BUS_BUSY_THRESHOLD
)
2289 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2290 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2291 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2292 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2293 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2294 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2295 * connections it will be counted five times. This difference in counting ("references"
2296 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2297 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2298 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2299 * currently chosen much higher than the "budget". */
2300 budget
= MANAGER_BUS_MESSAGE_BUDGET
;
2303 while (budget
!= 0 && (u
= m
->dbus_unit_queue
)) {
2305 assert(u
->in_dbus_queue
);
2307 bus_unit_send_change_signal(u
);
2310 if (budget
!= UINT_MAX
)
2314 while (budget
!= 0 && (j
= m
->dbus_job_queue
)) {
2315 assert(j
->in_dbus_queue
);
2317 bus_job_send_change_signal(j
);
2320 if (budget
!= UINT_MAX
)
2324 if (m
->send_reloading_done
) {
2325 m
->send_reloading_done
= false;
2326 bus_manager_send_reloading(m
, false);
2330 if (m
->pending_reload_message
) {
2331 bus_send_pending_reload_message(m
);
2338 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2339 Manager
*m
= userdata
;
2343 n
= recv(fd
, buf
, sizeof(buf
), 0);
2345 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
2347 log_error("Got zero-length cgroups agent message, ignoring.");
2350 if ((size_t) n
>= sizeof(buf
)) {
2351 log_error("Got overly long cgroups agent message, ignoring.");
2355 if (memchr(buf
, 0, n
)) {
2356 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2361 manager_notify_cgroup_empty(m
, buf
);
2362 (void) bus_forward_agent_released(m
, buf
);
2367 static bool manager_process_barrier_fd(char * const *tags
, FDSet
*fds
) {
2369 /* nothing else must be sent when using BARRIER=1 */
2370 if (strv_contains(tags
, "BARRIER=1")) {
2371 if (strv_length(tags
) == 1) {
2372 if (fdset_size(fds
) != 1)
2373 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2375 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2377 /* Drop the message if BARRIER=1 was found */
2384 static void manager_invoke_notify_message(
2387 const struct ucred
*ucred
,
2396 if (u
->notifygen
== m
->notifygen
) /* Already invoked on this same unit in this same iteration? */
2398 u
->notifygen
= m
->notifygen
;
2400 if (UNIT_VTABLE(u
)->notify_message
)
2401 UNIT_VTABLE(u
)->notify_message(u
, ucred
, tags
, fds
);
2403 else if (DEBUG_LOGGING
) {
2404 _cleanup_free_
char *buf
= NULL
, *x
= NULL
, *y
= NULL
;
2406 buf
= strv_join(tags
, ", ");
2408 x
= ellipsize(buf
, 20, 90);
2412 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
2416 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2418 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2419 Manager
*m
= userdata
;
2420 char buf
[NOTIFY_BUFFER_MAX
+1];
2421 struct iovec iovec
= {
2423 .iov_len
= sizeof(buf
)-1,
2425 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) +
2426 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)) control
;
2427 struct msghdr msghdr
= {
2430 .msg_control
= &control
,
2431 .msg_controllen
= sizeof(control
),
2434 struct cmsghdr
*cmsg
;
2435 struct ucred
*ucred
= NULL
;
2436 _cleanup_free_ Unit
**array_copy
= NULL
;
2437 _cleanup_strv_free_
char **tags
= NULL
;
2438 Unit
*u1
, *u2
, **array
;
2439 int r
, *fd_array
= NULL
;
2445 assert(m
->notify_fd
== fd
);
2447 if (revents
!= EPOLLIN
) {
2448 log_warning("Got unexpected poll event for notify fd.");
2452 n
= recvmsg_safe(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
2453 if (IN_SET(n
, -EAGAIN
, -EINTR
))
2454 return 0; /* Spurious wakeup, try again */
2456 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2460 /* If this is any other, real error, then let's stop processing this socket. This of course
2461 * means we won't take notification messages anymore, but that's still better than busy
2462 * looping around this: being woken up over and over again but being unable to actually read
2463 * the message off the socket. */
2464 return log_error_errno(n
, "Failed to receive notification message: %m");
2466 CMSG_FOREACH(cmsg
, &msghdr
) {
2467 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
2470 fd_array
= (int*) CMSG_DATA(cmsg
);
2471 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
2473 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
2474 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
2475 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
2478 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
2485 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
2487 close_many(fd_array
, n_fds
);
2493 if (!ucred
|| !pid_is_valid(ucred
->pid
)) {
2494 log_warning("Received notify message without valid credentials. Ignoring.");
2498 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
2499 log_warning("Received notify message exceeded maximum size. Ignoring.");
2503 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2504 * trailing NUL byte in the message, but don't expect it. */
2505 if (n
> 1 && memchr(buf
, 0, n
-1)) {
2506 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2510 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2512 tags
= strv_split_newlines(buf
);
2518 /* possibly a barrier fd, let's see */
2519 if (manager_process_barrier_fd(tags
, fds
))
2522 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2525 /* Notify every unit that might be interested, which might be multiple. */
2526 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
2527 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(ucred
->pid
));
2528 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-ucred
->pid
));
2535 array_copy
= newdup(Unit
*, array
, k
+1);
2539 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2540 * make sure we only invoke each unit's handler once. */
2542 manager_invoke_notify_message(m
, u1
, ucred
, tags
, fds
);
2546 manager_invoke_notify_message(m
, u2
, ucred
, tags
, fds
);
2550 for (size_t i
= 0; array_copy
[i
]; i
++) {
2551 manager_invoke_notify_message(m
, array_copy
[i
], ucred
, tags
, fds
);
2556 log_warning("Cannot find unit for notify message of PID "PID_FMT
", ignoring.", ucred
->pid
);
2558 if (fdset_size(fds
) > 0)
2559 log_warning("Got extra auxiliary fds with notification message, closing them.");
2564 static void manager_invoke_sigchld_event(
2567 const siginfo_t
*si
) {
2573 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2574 if (u
->sigchldgen
== m
->sigchldgen
)
2576 u
->sigchldgen
= m
->sigchldgen
;
2578 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s.", si
->si_pid
, u
->id
);
2579 unit_unwatch_pid(u
, si
->si_pid
);
2581 if (UNIT_VTABLE(u
)->sigchld_event
)
2582 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
2585 static int manager_dispatch_sigchld(sd_event_source
*source
, void *userdata
) {
2586 Manager
*m
= userdata
;
2593 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2594 * while it is a zombie. */
2596 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
2598 if (errno
!= ECHILD
)
2599 log_error_errno(errno
, "Failed to peek for child with waitid(), ignoring: %m");
2607 if (IN_SET(si
.si_code
, CLD_EXITED
, CLD_KILLED
, CLD_DUMPED
)) {
2608 _cleanup_free_ Unit
**array_copy
= NULL
;
2609 _cleanup_free_
char *name
= NULL
;
2610 Unit
*u1
, *u2
, **array
;
2612 (void) get_process_comm(si
.si_pid
, &name
);
2614 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
2615 si
.si_pid
, strna(name
),
2616 sigchld_code_to_string(si
.si_code
),
2618 strna(si
.si_code
== CLD_EXITED
2619 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
2620 : signal_to_string(si
.si_status
)));
2622 /* Increase the generation counter used for filtering out duplicate unit invocations */
2625 /* And now figure out the unit this belongs to, it might be multiple... */
2626 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
2627 u2
= hashmap_get(m
->watch_pids
, PID_TO_PTR(si
.si_pid
));
2628 array
= hashmap_get(m
->watch_pids
, PID_TO_PTR(-si
.si_pid
));
2632 /* Count how many entries the array has */
2636 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2637 array_copy
= newdup(Unit
*, array
, n
+1);
2642 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2643 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2644 * each iteration. */
2646 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2647 * We only do this for the cgroup the PID belonged to. */
2648 (void) unit_check_oom(u1
);
2650 /* This only logs for now. In the future when the interface for kills/notifications
2651 * is more stable we can extend service results table similar to how kernel oom kills
2653 (void) unit_check_oomd_kill(u1
);
2655 manager_invoke_sigchld_event(m
, u1
, &si
);
2658 manager_invoke_sigchld_event(m
, u2
, &si
);
2660 for (size_t i
= 0; array_copy
[i
]; i
++)
2661 manager_invoke_sigchld_event(m
, array_copy
[i
], &si
);
2664 /* And now, we actually reap the zombie. */
2665 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
2666 log_error_errno(errno
, "Failed to dequeue child, ignoring: %m");
2673 /* All children processed for now, turn off event source */
2675 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_OFF
);
2677 return log_error_errno(r
, "Failed to disable SIGCHLD event source: %m");
2682 static void manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
2683 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
2686 log_debug("Activating special unit %s", name
);
2688 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, NULL
, &error
, NULL
);
2690 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
2693 static void manager_handle_ctrl_alt_del(Manager
*m
) {
2694 /* If the user presses C-A-D more than
2695 * 7 times within 2s, we reboot/shutdown immediately,
2696 * unless it was disabled in system.conf */
2698 if (ratelimit_below(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
2699 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
2701 emergency_action(m
, m
->cad_burst_action
, EMERGENCY_ACTION_WARN
, NULL
, -1,
2702 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2705 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2706 Manager
*m
= userdata
;
2708 struct signalfd_siginfo sfsi
;
2712 assert(m
->signal_fd
== fd
);
2714 if (revents
!= EPOLLIN
) {
2715 log_warning("Got unexpected events from signal file descriptor.");
2719 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
2720 if (n
!= sizeof(sfsi
)) {
2722 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n
);
2726 if (IN_SET(errno
, EINTR
, EAGAIN
))
2729 /* We return an error here, which will kill this handler,
2730 * to avoid a busy loop on read error. */
2731 return log_error_errno(errno
, "Reading from signal fd failed: %m");
2734 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
2735 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
2736 ? LOG_DEBUG
: LOG_INFO
,
2739 switch (sfsi
.ssi_signo
) {
2742 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
2744 log_warning_errno(r
, "Failed to enable SIGCHLD event source, ignoring: %m");
2749 if (MANAGER_IS_SYSTEM(m
)) {
2750 /* This is for compatibility with the original sysvinit */
2751 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2754 m
->objective
= MANAGER_REEXECUTE
;
2760 if (MANAGER_IS_SYSTEM(m
))
2761 manager_handle_ctrl_alt_del(m
);
2763 manager_start_target(m
, SPECIAL_EXIT_TARGET
,
2764 JOB_REPLACE_IRREVERSIBLY
);
2768 /* This is a nop on non-init */
2769 if (MANAGER_IS_SYSTEM(m
))
2770 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2775 /* This is a nop on non-init */
2776 if (MANAGER_IS_SYSTEM(m
))
2777 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2782 if (manager_dbus_is_running(m
, false)) {
2783 log_info("Trying to reconnect to bus...");
2785 (void) bus_init_api(m
);
2787 if (MANAGER_IS_SYSTEM(m
))
2788 (void) bus_init_system(m
);
2790 log_info("Starting D-Bus service...");
2791 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2797 _cleanup_free_
char *dump
= NULL
;
2799 r
= manager_get_dump_string(m
, &dump
);
2801 log_warning_errno(errno
, "Failed to acquire manager dump: %m");
2805 log_dump(LOG_INFO
, dump
);
2810 if (verify_run_space_and_log("Refusing to reload") < 0)
2813 m
->objective
= MANAGER_RELOAD
;
2818 /* Starting SIGRTMIN+0 */
2819 static const struct {
2822 } target_table
[] = {
2823 [0] = { SPECIAL_DEFAULT_TARGET
, JOB_ISOLATE
},
2824 [1] = { SPECIAL_RESCUE_TARGET
, JOB_ISOLATE
},
2825 [2] = { SPECIAL_EMERGENCY_TARGET
, JOB_ISOLATE
},
2826 [3] = { SPECIAL_HALT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2827 [4] = { SPECIAL_POWEROFF_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2828 [5] = { SPECIAL_REBOOT_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2829 [6] = { SPECIAL_KEXEC_TARGET
, JOB_REPLACE_IRREVERSIBLY
},
2832 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2833 static const ManagerObjective objective_table
[] = {
2835 [1] = MANAGER_POWEROFF
,
2836 [2] = MANAGER_REBOOT
,
2837 [3] = MANAGER_KEXEC
,
2840 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2841 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2842 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2843 manager_start_target(m
, target_table
[idx
].target
,
2844 target_table
[idx
].mode
);
2848 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2849 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(objective_table
)) {
2850 m
->objective
= objective_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2854 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2857 manager_override_show_status(m
, SHOW_STATUS_YES
, "signal");
2861 manager_override_show_status(m
, SHOW_STATUS_NO
, "signal");
2865 manager_override_log_level(m
, LOG_DEBUG
);
2869 manager_restore_original_log_level(m
);
2873 if (MANAGER_IS_USER(m
)) {
2874 m
->objective
= MANAGER_EXIT
;
2878 /* This is a nop on init */
2882 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2883 manager_restore_original_log_target(m
);
2887 manager_override_log_target(m
, LOG_TARGET_CONSOLE
);
2891 manager_override_log_target(m
, LOG_TARGET_KMSG
);
2895 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2902 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2903 Manager
*m
= userdata
;
2907 assert(m
->time_change_fd
== fd
);
2909 log_struct(LOG_DEBUG
,
2910 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR
,
2911 LOG_MESSAGE("Time has been changed"));
2913 /* Restart the watch */
2914 (void) manager_setup_time_change(m
);
2916 HASHMAP_FOREACH(u
, m
->units
)
2917 if (UNIT_VTABLE(u
)->time_change
)
2918 UNIT_VTABLE(u
)->time_change(u
);
2923 static int manager_dispatch_timezone_change(
2924 sd_event_source
*source
,
2925 const struct inotify_event
*e
,
2928 Manager
*m
= userdata
;
2934 log_debug("inotify event for /etc/localtime");
2936 changed
= manager_read_timezone_stat(m
);
2940 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2941 (void) manager_setup_timezone_change(m
);
2943 /* Read the new timezone */
2946 log_debug("Timezone has been changed (now: %s).", tzname
[daylight
]);
2948 HASHMAP_FOREACH(u
, m
->units
)
2949 if (UNIT_VTABLE(u
)->timezone_change
)
2950 UNIT_VTABLE(u
)->timezone_change(u
);
2955 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2956 Manager
*m
= userdata
;
2959 assert(m
->idle_pipe
[2] == fd
);
2961 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2962 * now turn off any further console output if there's at least one service that needs console access, so that
2963 * from now on our own output should not spill into that service's output anymore. After all, we support
2964 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2965 * exclusively without our interference. */
2966 m
->no_console_output
= m
->n_on_console
> 0;
2968 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2969 * by closing the pipes towards them, which is what they are waiting for. */
2970 manager_close_idle_pipe(m
);
2975 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2976 Manager
*m
= userdata
;
2982 manager_print_jobs_in_progress(m
);
2984 r
= sd_event_source_set_time_relative(source
, JOBS_IN_PROGRESS_PERIOD_USEC
);
2988 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2991 int manager_loop(Manager
*m
) {
2992 RateLimit rl
= { .interval
= 1*USEC_PER_SEC
, .burst
= 50000 };
2996 assert(m
->objective
== MANAGER_OK
); /* Ensure manager_startup() has been called */
2998 manager_check_finished(m
);
3000 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3001 r
= sd_event_source_set_enabled(m
->sigchld_event_source
, SD_EVENT_ON
);
3003 return log_error_errno(r
, "Failed to enable SIGCHLD event source: %m");
3005 while (m
->objective
== MANAGER_OK
) {
3006 usec_t wait_usec
, watchdog_usec
;
3008 watchdog_usec
= manager_get_watchdog(m
, WATCHDOG_RUNTIME
);
3009 if (m
->runtime_watchdog_running
)
3010 (void) watchdog_ping();
3011 else if (timestamp_is_set(watchdog_usec
))
3012 manager_retry_runtime_watchdog(m
);
3014 if (!ratelimit_below(&rl
)) {
3015 /* Yay, something is going seriously wrong, pause a little */
3016 log_warning("Looping too fast. Throttling execution a little.");
3020 if (manager_dispatch_load_queue(m
) > 0)
3023 if (manager_dispatch_gc_job_queue(m
) > 0)
3026 if (manager_dispatch_gc_unit_queue(m
) > 0)
3029 if (manager_dispatch_cleanup_queue(m
) > 0)
3032 if (manager_dispatch_cgroup_realize_queue(m
) > 0)
3035 if (manager_dispatch_start_when_upheld_queue(m
) > 0)
3038 if (manager_dispatch_stop_when_bound_queue(m
) > 0)
3041 if (manager_dispatch_stop_when_unneeded_queue(m
) > 0)
3044 if (manager_dispatch_dbus_queue(m
) > 0)
3047 /* Sleep for watchdog runtime wait time */
3048 if (timestamp_is_set(watchdog_usec
))
3049 wait_usec
= watchdog_runtime_wait();
3051 wait_usec
= USEC_INFINITY
;
3053 r
= sd_event_run(m
->event
, wait_usec
);
3055 return log_error_errno(r
, "Failed to run event loop: %m");
3058 return m
->objective
;
3061 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
3062 _cleanup_free_
char *n
= NULL
;
3063 sd_id128_t invocation_id
;
3071 r
= unit_name_from_dbus_path(s
, &n
);
3075 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3076 * as invocation ID. */
3077 r
= sd_id128_from_string(n
, &invocation_id
);
3079 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
3085 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
,
3086 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.",
3087 SD_ID128_FORMAT_VAL(invocation_id
));
3090 /* If this didn't work, we check if this is a unit name */
3091 if (!unit_name_is_valid(n
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
3092 _cleanup_free_
char *nn
= NULL
;
3095 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
,
3096 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn
));
3099 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
3107 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
3117 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
3121 r
= safe_atou(p
, &id
);
3125 j
= manager_get_job(m
, id
);
3134 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
3137 _cleanup_free_
char *p
= NULL
;
3141 if (!MANAGER_IS_SYSTEM(m
))
3144 audit_fd
= get_audit_fd();
3148 /* Don't generate audit events if the service was already
3149 * started and we're just deserializing */
3150 if (MANAGER_IS_RELOADING(m
))
3153 if (u
->type
!= UNIT_SERVICE
)
3156 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
3158 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
3162 msg
= strjoina("unit=", p
);
3163 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
3165 /* We aren't allowed to send audit messages?
3166 * Then let's not retry again. */
3169 log_warning_errno(errno
, "Failed to send audit message: %m");
3175 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
3176 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
3177 _cleanup_free_
char *message
= NULL
;
3178 _cleanup_close_
int fd
= -1;
3181 /* Don't generate plymouth events if the service was already
3182 * started and we're just deserializing */
3183 if (MANAGER_IS_RELOADING(m
))
3186 if (!MANAGER_IS_SYSTEM(m
))
3189 if (detect_container() > 0)
3192 if (!IN_SET(u
->type
, UNIT_SERVICE
, UNIT_MOUNT
, UNIT_SWAP
))
3195 /* We set SOCK_NONBLOCK here so that we rather drop the
3196 * message then wait for plymouth */
3197 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
3199 log_error_errno(errno
, "socket() failed: %m");
3203 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
3204 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3205 log_error_errno(errno
, "connect() failed: %m");
3209 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
3215 if (write(fd
, message
, n
+ 1) != n
+ 1)
3216 if (!IN_SET(errno
, EAGAIN
, ENOENT
) && !ERRNO_IS_DISCONNECT(errno
))
3217 log_error_errno(errno
, "Failed to write Plymouth message: %m");
3220 int manager_open_serialization(Manager
*m
, FILE **_f
) {
3221 _cleanup_close_
int fd
= -1;
3226 fd
= open_serialization_fd("systemd-state");
3230 f
= take_fdopen(&fd
, "w+");
3238 static bool manager_timestamp_shall_serialize(ManagerTimestamp t
) {
3243 /* The following timestamps only apply to the host system, hence only serialize them there */
3245 MANAGER_TIMESTAMP_USERSPACE
, MANAGER_TIMESTAMP_FINISH
,
3246 MANAGER_TIMESTAMP_SECURITY_START
, MANAGER_TIMESTAMP_SECURITY_FINISH
,
3247 MANAGER_TIMESTAMP_GENERATORS_START
, MANAGER_TIMESTAMP_GENERATORS_FINISH
,
3248 MANAGER_TIMESTAMP_UNITS_LOAD_START
, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
);
3251 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3253 static void manager_serialize_uid_refs_internal(
3256 const char *field_name
) {
3263 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as
3264 * the actual counter of it is better rebuild after a reload/reexec. */
3266 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
3270 uid
= PTR_TO_UID(k
);
3271 c
= PTR_TO_UINT32(p
);
3273 if (!(c
& DESTROY_IPC_FLAG
))
3276 (void) serialize_item_format(f
, field_name
, UID_FMT
, uid
);
3280 static void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3281 manager_serialize_uid_refs_internal(f
, m
->uid_refs
, "destroy-ipc-uid");
3284 static void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3285 manager_serialize_uid_refs_internal(f
, m
->gid_refs
, "destroy-ipc-gid");
3288 int manager_serialize(
3292 bool switching_root
) {
3302 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3304 (void) serialize_item_format(f
, "current-job-id", "%" PRIu32
, m
->current_job_id
);
3305 (void) serialize_item_format(f
, "n-installed-jobs", "%u", m
->n_installed_jobs
);
3306 (void) serialize_item_format(f
, "n-failed-jobs", "%u", m
->n_failed_jobs
);
3307 (void) serialize_bool(f
, "taint-usr", m
->taint_usr
);
3308 (void) serialize_bool(f
, "ready-sent", m
->ready_sent
);
3309 (void) serialize_bool(f
, "taint-logged", m
->taint_logged
);
3310 (void) serialize_bool(f
, "service-watchdogs", m
->service_watchdogs
);
3312 /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */
3313 (void) serialize_bool(f
, "honor-device-enumeration", !switching_root
);
3315 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
3316 (void) serialize_item(f
, "show-status-overridden",
3317 show_status_to_string(m
->show_status_overridden
));
3319 if (m
->log_level_overridden
)
3320 (void) serialize_item_format(f
, "log-level-override", "%i", log_get_max_level());
3321 if (m
->log_target_overridden
)
3322 (void) serialize_item(f
, "log-target-override", log_target_to_string(log_get_target()));
3324 (void) serialize_usec(f
, "runtime-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3325 (void) serialize_usec(f
, "reboot-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_REBOOT
]);
3326 (void) serialize_usec(f
, "kexec-watchdog-overridden", m
->watchdog_overridden
[WATCHDOG_KEXEC
]);
3328 for (ManagerTimestamp q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3329 _cleanup_free_
char *joined
= NULL
;
3331 if (!manager_timestamp_shall_serialize(q
))
3334 joined
= strjoin(manager_timestamp_to_string(q
), "-timestamp");
3338 (void) serialize_dual_timestamp(f
, joined
, m
->timestamps
+ q
);
3341 if (!switching_root
)
3342 (void) serialize_strv(f
, "env", m
->client_environment
);
3344 if (m
->notify_fd
>= 0) {
3345 r
= serialize_fd(f
, fds
, "notify-fd", m
->notify_fd
);
3349 (void) serialize_item(f
, "notify-socket", m
->notify_socket
);
3352 if (m
->cgroups_agent_fd
>= 0) {
3353 r
= serialize_fd(f
, fds
, "cgroups-agent-fd", m
->cgroups_agent_fd
);
3358 if (m
->user_lookup_fds
[0] >= 0) {
3361 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
3363 return log_error_errno(copy0
, "Failed to add user lookup fd to serialization: %m");
3365 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
3367 return log_error_errno(copy1
, "Failed to add user lookup fd to serialization: %m");
3369 (void) serialize_item_format(f
, "user-lookup", "%i %i", copy0
, copy1
);
3372 bus_track_serialize(m
->subscribed
, f
, "subscribed");
3374 r
= dynamic_user_serialize(m
, f
, fds
);
3378 manager_serialize_uid_refs(m
, f
);
3379 manager_serialize_gid_refs(m
, f
);
3381 r
= exec_runtime_serialize(m
, f
, fds
);
3385 (void) fputc('\n', f
);
3387 HASHMAP_FOREACH_KEY(u
, t
, m
->units
) {
3395 r
= unit_serialize(u
, f
, fds
, !switching_root
);
3400 r
= fflush_and_check(f
);
3402 return log_error_errno(r
, "Failed to flush serialization: %m");
3404 r
= bus_fdset_add_all(m
, fds
);
3406 return log_error_errno(r
, "Failed to add bus sockets to serialization: %m");
3411 static int manager_deserialize_one_unit(Manager
*m
, const char *name
, FILE *f
, FDSet
*fds
) {
3415 r
= manager_load_unit(m
, name
, NULL
, NULL
, &u
);
3419 return log_notice_errno(r
, "Failed to load unit \"%s\", skipping deserialization: %m", name
);
3422 r
= unit_deserialize(u
, f
, fds
);
3426 return log_notice_errno(r
, "Failed to deserialize unit \"%s\", skipping: %m", name
);
3432 static int manager_deserialize_units(Manager
*m
, FILE *f
, FDSet
*fds
) {
3433 const char *unit_name
;
3437 _cleanup_free_
char *line
= NULL
;
3439 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3441 return log_error_errno(r
, "Failed to read serialization line: %m");
3445 unit_name
= strstrip(line
);
3447 r
= manager_deserialize_one_unit(m
, unit_name
, f
, fds
);
3451 r
= unit_deserialize_skip(f
);
3460 usec_t
manager_get_watchdog(Manager
*m
, WatchdogType t
) {
3463 if (MANAGER_IS_USER(m
))
3464 return USEC_INFINITY
;
3466 if (timestamp_is_set(m
->watchdog_overridden
[t
]))
3467 return m
->watchdog_overridden
[t
];
3469 return m
->watchdog
[t
];
3472 void manager_set_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3477 if (MANAGER_IS_USER(m
))
3480 if (m
->watchdog
[t
] == timeout
)
3483 if (t
== WATCHDOG_RUNTIME
)
3484 if (!timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
])) {
3485 if (timestamp_is_set(timeout
)) {
3486 r
= watchdog_set_timeout(&timeout
);
3489 m
->runtime_watchdog_running
= true;
3491 watchdog_close(true);
3492 m
->runtime_watchdog_running
= false;
3496 m
->watchdog
[t
] = timeout
;
3499 int manager_override_watchdog(Manager
*m
, WatchdogType t
, usec_t timeout
) {
3504 if (MANAGER_IS_USER(m
))
3507 if (m
->watchdog_overridden
[t
] == timeout
)
3510 if (t
== WATCHDOG_RUNTIME
) {
3513 p
= timestamp_is_set(timeout
) ? &timeout
: &m
->watchdog
[t
];
3514 if (timestamp_is_set(*p
)) {
3515 r
= watchdog_set_timeout(p
);
3518 m
->runtime_watchdog_running
= true;
3520 watchdog_close(true);
3521 m
->runtime_watchdog_running
= false;
3525 m
->watchdog_overridden
[t
] = timeout
;
3530 void manager_retry_runtime_watchdog(Manager
*m
) {
3535 if (timestamp_is_set(m
->watchdog_overridden
[WATCHDOG_RUNTIME
]))
3536 r
= watchdog_set_timeout(&m
->watchdog_overridden
[WATCHDOG_RUNTIME
]);
3538 r
= watchdog_set_timeout(&m
->watchdog
[WATCHDOG_RUNTIME
]);
3541 m
->runtime_watchdog_running
= true;
3544 static void manager_deserialize_uid_refs_one_internal(
3546 const char *value
) {
3555 r
= parse_uid(value
, &uid
);
3556 if (r
< 0 || uid
== 0) {
3557 log_debug("Unable to parse UID/GID reference serialization: " UID_FMT
, uid
);
3561 if (hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
) < 0) {
3566 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3567 if (c
& DESTROY_IPC_FLAG
)
3570 c
|= DESTROY_IPC_FLAG
;
3572 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3574 log_debug_errno(r
, "Failed to add UID/GID reference entry: %m");
3579 static void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3580 manager_deserialize_uid_refs_one_internal(&m
->uid_refs
, value
);
3583 static void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3584 manager_deserialize_uid_refs_one_internal(&m
->gid_refs
, value
);
3587 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
3593 if (DEBUG_LOGGING
) {
3594 if (fdset_isempty(fds
))
3595 log_debug("No file descriptors passed");
3599 FDSET_FOREACH(fd
, fds
) {
3600 _cleanup_free_
char *fn
= NULL
;
3602 r
= fd_get_path(fd
, &fn
);
3604 log_debug_errno(r
, "Received serialized fd %i → %m", fd
);
3606 log_debug("Received serialized fd %i → %s", fd
, strna(fn
));
3611 log_debug("Deserializing state...");
3613 /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have
3614 * increased it to non-zero, which is why we just increase it by one here and down again at the end of this
3616 _cleanup_(manager_reloading_stopp
) _unused_ Manager
*reloading
= manager_reloading_start(m
);
3619 _cleanup_free_
char *line
= NULL
;
3620 const char *val
, *l
;
3622 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3624 return log_error_errno(r
, "Failed to read serialization line: %m");
3629 if (isempty(l
)) /* end marker */
3632 if ((val
= startswith(l
, "current-job-id="))) {
3635 if (safe_atou32(val
, &id
) < 0)
3636 log_notice("Failed to parse current job id value '%s', ignoring.", val
);
3638 m
->current_job_id
= MAX(m
->current_job_id
, id
);
3640 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
3643 if (safe_atou32(val
, &n
) < 0)
3644 log_notice("Failed to parse installed jobs counter '%s', ignoring.", val
);
3646 m
->n_installed_jobs
+= n
;
3648 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
3651 if (safe_atou32(val
, &n
) < 0)
3652 log_notice("Failed to parse failed jobs counter '%s', ignoring.", val
);
3654 m
->n_failed_jobs
+= n
;
3656 } else if ((val
= startswith(l
, "taint-usr="))) {
3659 b
= parse_boolean(val
);
3661 log_notice("Failed to parse taint /usr flag '%s', ignoring.", val
);
3663 m
->taint_usr
= m
->taint_usr
|| b
;
3665 } else if ((val
= startswith(l
, "ready-sent="))) {
3668 b
= parse_boolean(val
);
3670 log_notice("Failed to parse ready-sent flag '%s', ignoring.", val
);
3672 m
->ready_sent
= m
->ready_sent
|| b
;
3674 } else if ((val
= startswith(l
, "taint-logged="))) {
3677 b
= parse_boolean(val
);
3679 log_notice("Failed to parse taint-logged flag '%s', ignoring.", val
);
3681 m
->taint_logged
= m
->taint_logged
|| b
;
3683 } else if ((val
= startswith(l
, "service-watchdogs="))) {
3686 b
= parse_boolean(val
);
3688 log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val
);
3690 m
->service_watchdogs
= b
;
3692 } else if ((val
= startswith(l
, "honor-device-enumeration="))) {
3695 b
= parse_boolean(val
);
3697 log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val
);
3699 m
->honor_device_enumeration
= b
;
3701 } else if ((val
= startswith(l
, "show-status-overridden="))) {
3704 s
= show_status_from_string(val
);
3706 log_notice("Failed to parse show-status-overridden flag '%s', ignoring.", val
);
3708 manager_override_show_status(m
, s
, "deserialize");
3710 } else if ((val
= startswith(l
, "log-level-override="))) {
3713 level
= log_level_from_string(val
);
3715 log_notice("Failed to parse log-level-override value '%s', ignoring.", val
);
3717 manager_override_log_level(m
, level
);
3719 } else if ((val
= startswith(l
, "log-target-override="))) {
3722 target
= log_target_from_string(val
);
3724 log_notice("Failed to parse log-target-override value '%s', ignoring.", val
);
3726 manager_override_log_target(m
, target
);
3728 } else if ((val
= startswith(l
, "runtime-watchdog-overridden="))) {
3731 if (deserialize_usec(val
, &t
) < 0)
3732 log_notice("Failed to parse runtime-watchdog-overridden value '%s', ignoring.", val
);
3734 manager_override_watchdog(m
, WATCHDOG_RUNTIME
, t
);
3736 } else if ((val
= startswith(l
, "reboot-watchdog-overridden="))) {
3739 if (deserialize_usec(val
, &t
) < 0)
3740 log_notice("Failed to parse reboot-watchdog-overridden value '%s', ignoring.", val
);
3742 manager_override_watchdog(m
, WATCHDOG_REBOOT
, t
);
3744 } else if ((val
= startswith(l
, "kexec-watchdog-overridden="))) {
3747 if (deserialize_usec(val
, &t
) < 0)
3748 log_notice("Failed to parse kexec-watchdog-overridden value '%s', ignoring.", val
);
3750 manager_override_watchdog(m
, WATCHDOG_KEXEC
, t
);
3752 } else if (startswith(l
, "env=")) {
3753 r
= deserialize_environment(l
+ 4, &m
->client_environment
);
3755 log_notice_errno(r
, "Failed to parse environment entry: \"%s\", ignoring: %m", l
);
3757 } else if ((val
= startswith(l
, "notify-fd="))) {
3760 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3761 log_notice("Failed to parse notify fd, ignoring: \"%s\"", val
);
3763 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
3764 safe_close(m
->notify_fd
);
3765 m
->notify_fd
= fdset_remove(fds
, fd
);
3768 } else if ((val
= startswith(l
, "notify-socket="))) {
3769 r
= free_and_strdup(&m
->notify_socket
, val
);
3773 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
3776 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
3777 log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val
);
3779 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
3780 safe_close(m
->cgroups_agent_fd
);
3781 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
3784 } else if ((val
= startswith(l
, "user-lookup="))) {
3787 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
3788 log_notice("Failed to parse user lookup fd, ignoring: %s", val
);
3790 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
3791 safe_close_pair(m
->user_lookup_fds
);
3792 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
3793 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
3796 } else if ((val
= startswith(l
, "dynamic-user=")))
3797 dynamic_user_deserialize_one(m
, val
, fds
);
3798 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
3799 manager_deserialize_uid_refs_one(m
, val
);
3800 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
3801 manager_deserialize_gid_refs_one(m
, val
);
3802 else if ((val
= startswith(l
, "exec-runtime=")))
3803 (void) exec_runtime_deserialize_one(m
, val
, fds
);
3804 else if ((val
= startswith(l
, "subscribed="))) {
3806 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
3812 for (q
= 0; q
< _MANAGER_TIMESTAMP_MAX
; q
++) {
3813 val
= startswith(l
, manager_timestamp_to_string(q
));
3817 val
= startswith(val
, "-timestamp=");
3822 if (q
< _MANAGER_TIMESTAMP_MAX
) /* found it */
3823 (void) deserialize_dual_timestamp(val
, m
->timestamps
+ q
);
3824 else if (!startswith(l
, "kdbus-fd=")) /* ignore kdbus */
3825 log_notice("Unknown serialization item '%s', ignoring.", l
);
3829 return manager_deserialize_units(m
, f
, fds
);
3832 int manager_reload(Manager
*m
) {
3833 _cleanup_(manager_reloading_stopp
) Manager
*reloading
= NULL
;
3834 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
3835 _cleanup_fclose_
FILE *f
= NULL
;
3840 r
= manager_open_serialization(m
, &f
);
3842 return log_error_errno(r
, "Failed to create serialization file: %m");
3848 /* We are officially in reload mode from here on. */
3849 reloading
= manager_reloading_start(m
);
3851 r
= manager_serialize(m
, f
, fds
, false);
3855 if (fseeko(f
, 0, SEEK_SET
) < 0)
3856 return log_error_errno(errno
, "Failed to seek to beginning of serialization: %m");
3858 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3861 bus_manager_send_reloading(m
, true);
3863 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3864 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3867 manager_clear_jobs_and_units(m
);
3868 lookup_paths_flush_generator(&m
->lookup_paths
);
3869 lookup_paths_free(&m
->lookup_paths
);
3870 exec_runtime_vacuum(m
);
3871 dynamic_user_vacuum(m
, false);
3872 m
->uid_refs
= hashmap_free(m
->uid_refs
);
3873 m
->gid_refs
= hashmap_free(m
->gid_refs
);
3875 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
3877 log_warning_errno(r
, "Failed to initialize path lookup table, ignoring: %m");
3879 (void) manager_run_environment_generators(m
);
3880 (void) manager_run_generators(m
);
3882 lookup_paths_log(&m
->lookup_paths
);
3884 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3885 manager_free_unit_name_maps(m
);
3887 /* First, enumerate what we can from kernel and suchlike */
3888 manager_enumerate_perpetual(m
);
3889 manager_enumerate(m
);
3891 /* Second, deserialize our stored data */
3892 r
= manager_deserialize(m
, f
, fds
);
3894 log_warning_errno(r
, "Deserialization failed, proceeding anyway: %m");
3896 /* We don't need the serialization anymore */
3899 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3900 (void) manager_setup_notify(m
);
3901 (void) manager_setup_cgroups_agent(m
);
3902 (void) manager_setup_user_lookup_fd(m
);
3904 /* Third, fire things up! */
3905 manager_coldplug(m
);
3907 /* Clean up runtime objects no longer referenced */
3910 /* Clean up deserialized tracked clients */
3911 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
3913 /* Consider the reload process complete now. */
3914 assert(m
->n_reloading
> 0);
3917 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3918 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3919 * let's always set the flag here for safety. */
3920 m
->honor_device_enumeration
= true;
3924 m
->send_reloading_done
= true;
3928 void manager_reset_failed(Manager
*m
) {
3933 HASHMAP_FOREACH(u
, m
->units
)
3934 unit_reset_failed(u
);
3937 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
3943 /* Returns true if the unit is inactive or going down */
3944 u
= manager_get_unit(m
, name
);
3948 return unit_inactive_or_pending(u
);
3951 static void log_taint_string(Manager
*m
) {
3952 _cleanup_free_
char *taint
= NULL
;
3956 if (MANAGER_IS_USER(m
) || m
->taint_logged
)
3959 m
->taint_logged
= true; /* only check for taint once */
3961 taint
= manager_taint_string(m
);
3965 log_struct(LOG_NOTICE
,
3966 LOG_MESSAGE("System is tainted: %s", taint
),
3968 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR
);
3971 static void manager_notify_finished(Manager
*m
) {
3972 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
3973 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
3975 if (MANAGER_IS_TEST_RUN(m
))
3978 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
3979 char ts
[FORMAT_TIMESPAN_MAX
];
3980 char buf
[FORMAT_TIMESPAN_MAX
+ STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX
+ STRLEN(" (loader) + ")]
3983 size_t size
= sizeof buf
;
3985 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3986 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3987 * negative values. */
3989 firmware_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
;
3990 loader_usec
= m
->timestamps
[MANAGER_TIMESTAMP_LOADER
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
3991 userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
3992 total_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FIRMWARE
].monotonic
+ m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
;
3994 if (firmware_usec
> 0)
3995 size
= strpcpyf(&p
, size
, "%s (firmware) + ", format_timespan(ts
, sizeof(ts
), firmware_usec
, USEC_PER_MSEC
));
3996 if (loader_usec
> 0)
3997 size
= strpcpyf(&p
, size
, "%s (loader) + ", format_timespan(ts
, sizeof(ts
), loader_usec
, USEC_PER_MSEC
));
3999 if (dual_timestamp_is_set(&m
->timestamps
[MANAGER_TIMESTAMP_INITRD
])) {
4001 /* The initrd case on bare-metal*/
4002 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
4003 initrd_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_INITRD
].monotonic
;
4005 log_struct(LOG_INFO
,
4006 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
4007 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
4008 "INITRD_USEC="USEC_FMT
, initrd_usec
,
4009 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
4010 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
4012 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
4013 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
4014 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
4015 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
4017 /* The initrd-less case on bare-metal*/
4019 kernel_usec
= m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_KERNEL
].monotonic
;
4022 log_struct(LOG_INFO
,
4023 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR
,
4024 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
4025 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
4026 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
4028 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
4029 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
4030 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
4033 /* The container and --user case */
4034 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
4035 total_usec
= userspace_usec
= m
->timestamps
[MANAGER_TIMESTAMP_FINISH
].monotonic
- m
->timestamps
[MANAGER_TIMESTAMP_USERSPACE
].monotonic
;
4037 log_struct(LOG_INFO
,
4038 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR
,
4039 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
4040 LOG_MESSAGE("Startup finished in %s.",
4041 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)));
4044 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
4047 m
->ready_sent
? "STATUS=Startup finished in %s."
4049 "STATUS=Startup finished in %s.",
4050 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
4051 m
->ready_sent
= true;
4053 log_taint_string(m
);
4056 static void manager_send_ready(Manager
*m
) {
4059 /* We send READY=1 on reaching basic.target only when running in --user mode. */
4060 if (!MANAGER_IS_USER(m
) || m
->ready_sent
)
4063 m
->ready_sent
= true;
4067 "STATUS=Reached " SPECIAL_BASIC_TARGET
".");
4070 static void manager_check_basic_target(Manager
*m
) {
4075 /* Small shortcut */
4076 if (m
->ready_sent
&& m
->taint_logged
)
4079 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4080 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4083 /* For user managers, send out READY=1 as soon as we reach basic.target */
4084 manager_send_ready(m
);
4086 /* Log the taint string as soon as we reach basic.target */
4087 log_taint_string(m
);
4090 void manager_check_finished(Manager
*m
) {
4093 if (MANAGER_IS_RELOADING(m
))
4096 /* Verify that we have entered the event loop already, and not left it again. */
4097 if (!MANAGER_IS_RUNNING(m
))
4100 manager_check_basic_target(m
);
4102 if (hashmap_size(m
->jobs
) > 0) {
4103 if (m
->jobs_in_progress_event_source
)
4104 /* Ignore any failure, this is only for feedback */
4105 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
,
4106 manager_watch_jobs_next_time(m
));
4110 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
4111 kill the hashmap if it is relatively large. */
4112 if (hashmap_buckets(m
->jobs
) > hashmap_size(m
->units
) / 10)
4113 m
->jobs
= hashmap_free(m
->jobs
);
4115 manager_flip_auto_status(m
, false, "boot finished");
4117 /* Notify Type=idle units that we are done now */
4118 manager_close_idle_pipe(m
);
4120 /* Turn off confirm spawn now */
4121 m
->confirm_spawn
= NULL
;
4123 /* No need to update ask password status when we're going non-interactive */
4124 manager_close_ask_password(m
);
4126 /* This is no longer the first boot */
4127 manager_set_first_boot(m
, false);
4129 if (MANAGER_IS_FINISHED(m
))
4132 dual_timestamp_get(m
->timestamps
+ MANAGER_TIMESTAMP_FINISH
);
4134 manager_notify_finished(m
);
4136 manager_invalidate_startup_units(m
);
4139 static bool generator_path_any(const char* const* paths
) {
4143 /* Optimize by skipping the whole process by not creating output directories
4144 * if no generators are found. */
4145 STRV_FOREACH(path
, (char**) paths
)
4146 if (access(*path
, F_OK
) == 0)
4148 else if (errno
!= ENOENT
)
4149 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
4154 static int manager_run_environment_generators(Manager
*m
) {
4155 char **tmp
= NULL
; /* this is only used in the forked process, no cleanup here */
4156 _cleanup_strv_free_
char **paths
= NULL
;
4158 [STDOUT_GENERATE
] = &tmp
,
4159 [STDOUT_COLLECT
] = &tmp
,
4160 [STDOUT_CONSUME
] = &m
->transient_environment
,
4164 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_ENV_GENERATORS
))
4167 paths
= env_generator_binary_paths(MANAGER_IS_SYSTEM(m
));
4171 if (!generator_path_any((const char* const*) paths
))
4174 RUN_WITH_UMASK(0022)
4175 r
= execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, gather_environment
,
4176 args
, NULL
, m
->transient_environment
,
4177 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4181 static int manager_run_generators(Manager
*m
) {
4182 _cleanup_strv_free_
char **paths
= NULL
;
4183 const char *argv
[5];
4188 if (MANAGER_IS_TEST_RUN(m
) && !(m
->test_run_flags
& MANAGER_TEST_RUN_GENERATORS
))
4191 paths
= generator_binary_paths(m
->unit_file_scope
);
4195 if (!generator_path_any((const char* const*) paths
))
4198 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
4200 log_error_errno(r
, "Failed to create generator directories: %m");
4204 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
4205 argv
[1] = m
->lookup_paths
.generator
;
4206 argv
[2] = m
->lookup_paths
.generator_early
;
4207 argv
[3] = m
->lookup_paths
.generator_late
;
4210 RUN_WITH_UMASK(0022)
4211 (void) execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, NULL
, NULL
,
4212 (char**) argv
, m
->transient_environment
,
4213 EXEC_DIR_PARALLEL
| EXEC_DIR_IGNORE_ERRORS
| EXEC_DIR_SET_SYSTEMD_EXEC_PID
);
4218 lookup_paths_trim_generator(&m
->lookup_paths
);
4222 int manager_transient_environment_add(Manager
*m
, char **plus
) {
4227 if (strv_isempty(plus
))
4230 a
= strv_env_merge(2, m
->transient_environment
, plus
);
4234 sanitize_environment(a
);
4236 return strv_free_and_replace(m
->transient_environment
, a
);
4239 int manager_client_environment_modify(
4244 char **a
= NULL
, **b
= NULL
, **l
;
4248 if (strv_isempty(minus
) && strv_isempty(plus
))
4251 l
= m
->client_environment
;
4253 if (!strv_isempty(minus
)) {
4254 a
= strv_env_delete(l
, 1, minus
);
4261 if (!strv_isempty(plus
)) {
4262 b
= strv_env_merge(2, l
, plus
);
4271 if (m
->client_environment
!= l
)
4272 strv_free(m
->client_environment
);
4279 m
->client_environment
= sanitize_environment(l
);
4283 int manager_get_effective_environment(Manager
*m
, char ***ret
) {
4289 l
= strv_env_merge(2, m
->transient_environment
, m
->client_environment
);
4297 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
4300 for (unsigned i
= 0; i
< _RLIMIT_MAX
; i
++) {
4301 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
4303 if (!default_rlimit
[i
])
4306 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
4314 void manager_recheck_dbus(Manager
*m
) {
4317 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
4318 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
4319 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
4320 * while in the user instance we can assume it's already there. */
4322 if (MANAGER_IS_RELOADING(m
))
4323 return; /* don't check while we are reloading… */
4325 if (manager_dbus_is_running(m
, false)) {
4326 (void) bus_init_api(m
);
4328 if (MANAGER_IS_SYSTEM(m
))
4329 (void) bus_init_system(m
);
4331 (void) bus_done_api(m
);
4333 if (MANAGER_IS_SYSTEM(m
))
4334 (void) bus_done_system(m
);
4338 static bool manager_journal_is_running(Manager
*m
) {
4343 if (MANAGER_IS_TEST_RUN(m
))
4346 /* If we are the user manager we can safely assume that the journal is up */
4347 if (!MANAGER_IS_SYSTEM(m
))
4350 /* Check that the socket is not only up, but in RUNNING state */
4351 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
4354 if (SOCKET(u
)->state
!= SOCKET_RUNNING
)
4357 /* Similar, check if the daemon itself is fully up, too */
4358 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
4361 if (!IN_SET(SERVICE(u
)->state
, SERVICE_RELOAD
, SERVICE_RUNNING
))
4367 void disable_printk_ratelimit(void) {
4368 /* Disable kernel's printk ratelimit.
4370 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
4371 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
4372 * setting takes precedence. */
4375 r
= sysctl_write("kernel/printk_devkmsg", "on");
4377 log_debug_errno(r
, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
4380 void manager_recheck_journal(Manager
*m
) {
4384 /* Don't bother with this unless we are in the special situation of being PID 1 */
4385 if (getpid_cached() != 1)
4388 /* Don't check this while we are reloading, things might still change */
4389 if (MANAGER_IS_RELOADING(m
))
4392 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
4393 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
4394 * an activation ourselves we can't fulfill. */
4395 log_set_prohibit_ipc(!manager_journal_is_running(m
));
4399 static ShowStatus
manager_get_show_status(Manager
*m
) {
4402 if (MANAGER_IS_USER(m
))
4403 return _SHOW_STATUS_INVALID
;
4405 if (m
->show_status_overridden
!= _SHOW_STATUS_INVALID
)
4406 return m
->show_status_overridden
;
4408 return m
->show_status
;
4411 bool manager_get_show_status_on(Manager
*m
) {
4414 return show_status_on(manager_get_show_status(m
));
4417 static void set_show_status_marker(bool b
) {
4419 (void) touch("/run/systemd/show-status");
4421 (void) unlink("/run/systemd/show-status");
4424 void manager_set_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4427 assert(mode
>= 0 && mode
< _SHOW_STATUS_MAX
);
4429 if (MANAGER_IS_USER(m
))
4432 if (mode
== m
->show_status
)
4435 if (m
->show_status_overridden
== _SHOW_STATUS_INVALID
) {
4438 enabled
= show_status_on(mode
);
4439 log_debug("%s (%s) showing of status (%s).",
4440 enabled
? "Enabling" : "Disabling",
4441 strna(show_status_to_string(mode
)),
4444 set_show_status_marker(enabled
);
4447 m
->show_status
= mode
;
4450 void manager_override_show_status(Manager
*m
, ShowStatus mode
, const char *reason
) {
4452 assert(mode
< _SHOW_STATUS_MAX
);
4454 if (MANAGER_IS_USER(m
))
4457 if (mode
== m
->show_status_overridden
)
4460 m
->show_status_overridden
= mode
;
4462 if (mode
== _SHOW_STATUS_INVALID
)
4463 mode
= m
->show_status
;
4465 log_debug("%s (%s) showing of status (%s).",
4466 m
->show_status_overridden
!= _SHOW_STATUS_INVALID
? "Overriding" : "Restoring",
4467 strna(show_status_to_string(mode
)),
4470 set_show_status_marker(show_status_on(mode
));
4473 const char *manager_get_confirm_spawn(Manager
*m
) {
4474 static int last_errno
= 0;
4480 /* Here's the deal: we want to test the validity of the console but don't want
4481 * PID1 to go through the whole console process which might block. But we also
4482 * want to warn the user only once if something is wrong with the console so we
4483 * cannot do the sanity checks after spawning our children. So here we simply do
4484 * really basic tests to hopefully trap common errors.
4486 * If the console suddenly disappear at the time our children will really it
4487 * then they will simply fail to acquire it and a positive answer will be
4488 * assumed. New children will fall back to /dev/console though.
4490 * Note: TTYs are devices that can come and go any time, and frequently aren't
4491 * available yet during early boot (consider a USB rs232 dongle...). If for any
4492 * reason the configured console is not ready, we fall back to the default
4495 if (!m
->confirm_spawn
|| path_equal(m
->confirm_spawn
, "/dev/console"))
4496 return m
->confirm_spawn
;
4498 if (stat(m
->confirm_spawn
, &st
) < 0) {
4503 if (!S_ISCHR(st
.st_mode
)) {
4509 return m
->confirm_spawn
;
4512 if (last_errno
!= r
)
4513 last_errno
= log_warning_errno(r
, "Failed to open %s, using default console: %m", m
->confirm_spawn
);
4515 return "/dev/console";
4518 void manager_set_first_boot(Manager
*m
, bool b
) {
4521 if (!MANAGER_IS_SYSTEM(m
))
4524 if (m
->first_boot
!= (int) b
) {
4526 (void) touch("/run/systemd/first-boot");
4528 (void) unlink("/run/systemd/first-boot");
4534 void manager_disable_confirm_spawn(void) {
4535 (void) touch("/run/systemd/confirm_spawn_disabled");
4538 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
4539 if (!m
->confirm_spawn
)
4542 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
4545 static bool manager_should_show_status(Manager
*m
, StatusType type
) {
4548 if (!MANAGER_IS_SYSTEM(m
))
4551 if (m
->no_console_output
)
4554 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
4557 /* If we cannot find out the status properly, just proceed. */
4558 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
4561 if (type
== STATUS_TYPE_NOTICE
&& m
->show_status
!= SHOW_STATUS_NO
)
4564 return manager_get_show_status_on(m
);
4567 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
4570 /* If m is NULL, assume we're after shutdown and let the messages through. */
4572 if (m
&& !manager_should_show_status(m
, type
))
4575 /* XXX We should totally drop the check for ephemeral here
4576 * and thus effectively make 'Type=idle' pointless. */
4577 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
4580 va_start(ap
, format
);
4581 status_vprintf(status
, SHOW_STATUS_ELLIPSIZE
|(type
== STATUS_TYPE_EPHEMERAL
? SHOW_STATUS_EPHEMERAL
: 0), format
, ap
);
4585 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
4586 char p
[strlen(path
)+1];
4594 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
4597 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
4602 assert(u
->manager
== m
);
4604 size
= set_size(m
->failed_units
);
4607 r
= set_ensure_put(&m
->failed_units
, NULL
, u
);
4611 (void) set_remove(m
->failed_units
, u
);
4613 if (set_size(m
->failed_units
) != size
)
4614 bus_manager_send_change_signal(m
);
4619 ManagerState
manager_state(Manager
*m
) {
4624 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4625 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
4626 if (u
&& unit_active_or_pending(u
))
4627 return MANAGER_STOPPING
;
4629 /* Did we ever finish booting? If not then we are still starting up */
4630 if (!MANAGER_IS_FINISHED(m
)) {
4632 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
4633 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
4634 return MANAGER_INITIALIZING
;
4636 return MANAGER_STARTING
;
4639 if (MANAGER_IS_SYSTEM(m
)) {
4640 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4641 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
4642 if (u
&& unit_active_or_pending(u
))
4643 return MANAGER_MAINTENANCE
;
4645 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
4646 if (u
&& unit_active_or_pending(u
))
4647 return MANAGER_MAINTENANCE
;
4650 /* Are there any failed units? If so, we are in degraded mode */
4651 if (set_size(m
->failed_units
) > 0)
4652 return MANAGER_DEGRADED
;
4654 return MANAGER_RUNNING
;
4657 static void manager_unref_uid_internal(
4661 int (*_clean_ipc
)(uid_t uid
)) {
4665 assert(uid_is_valid(uid
));
4668 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4669 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4671 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4672 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4673 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4674 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4676 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4677 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4679 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4682 c
= PTR_TO_UINT32(hashmap_get(uid_refs
, UID_TO_PTR(uid
)));
4684 n
= c
& ~DESTROY_IPC_FLAG
;
4688 if (destroy_now
&& n
== 0) {
4689 hashmap_remove(uid_refs
, UID_TO_PTR(uid
));
4691 if (c
& DESTROY_IPC_FLAG
) {
4692 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
4693 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4695 (void) _clean_ipc(uid
);
4698 c
= n
| (c
& DESTROY_IPC_FLAG
);
4699 assert_se(hashmap_update(uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
4703 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
4704 manager_unref_uid_internal(m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
4707 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
4708 manager_unref_uid_internal(m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
4711 static int manager_ref_uid_internal(
4720 assert(uid_is_valid(uid
));
4722 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4723 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4725 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
4726 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
4728 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
4731 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
4735 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
4737 n
= c
& ~DESTROY_IPC_FLAG
;
4740 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
4743 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
4745 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
4748 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
4749 return manager_ref_uid_internal(&m
->uid_refs
, uid
, clean_ipc
);
4752 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
4753 return manager_ref_uid_internal(&m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
4756 static void manager_vacuum_uid_refs_internal(
4758 int (*_clean_ipc
)(uid_t uid
)) {
4764 HASHMAP_FOREACH_KEY(p
, k
, uid_refs
) {
4768 uid
= PTR_TO_UID(k
);
4769 c
= PTR_TO_UINT32(p
);
4771 n
= c
& ~DESTROY_IPC_FLAG
;
4775 if (c
& DESTROY_IPC_FLAG
) {
4776 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
4777 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
4779 (void) _clean_ipc(uid
);
4782 assert_se(hashmap_remove(uid_refs
, k
) == p
);
4786 static void manager_vacuum_uid_refs(Manager
*m
) {
4787 manager_vacuum_uid_refs_internal(m
->uid_refs
, clean_ipc_by_uid
);
4790 static void manager_vacuum_gid_refs(Manager
*m
) {
4791 manager_vacuum_uid_refs_internal(m
->gid_refs
, clean_ipc_by_gid
);
4794 static void manager_vacuum(Manager
*m
) {
4797 /* Release any dynamic users no longer referenced */
4798 dynamic_user_vacuum(m
, true);
4800 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4801 manager_vacuum_uid_refs(m
);
4802 manager_vacuum_gid_refs(m
);
4804 /* Release any runtimes no longer referenced */
4805 exec_runtime_vacuum(m
);
4808 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
4812 char unit_name
[UNIT_NAME_MAX
+1];
4815 Manager
*m
= userdata
;
4823 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4824 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4825 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4827 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
4829 if (IN_SET(errno
, EINTR
, EAGAIN
))
4832 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
4835 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
4836 log_warning("Received too short user lookup message, ignoring.");
4840 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
4841 log_warning("Received too long user lookup message, ignoring.");
4845 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
4846 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4850 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
4851 if (memchr(buffer
.unit_name
, 0, n
)) {
4852 log_warning("Received lookup message with embedded NUL character, ignoring.");
4856 buffer
.unit_name
[n
] = 0;
4857 u
= manager_get_unit(m
, buffer
.unit_name
);
4859 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4863 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
4865 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
4869 char *manager_taint_string(Manager
*m
) {
4870 _cleanup_free_
char *destination
= NULL
, *overflowuid
= NULL
, *overflowgid
= NULL
;
4874 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4875 * Only things that are detected at runtime should be tagged
4876 * here. For stuff that is set during compilation, emit a warning
4877 * in the configuration phase. */
4881 buf
= new(char, sizeof("split-usr:"
4886 "overflowuid-not-65534:"
4887 "overflowgid-not-65534:"));
4895 e
= stpcpy(e
, "split-usr:");
4897 if (access("/proc/cgroups", F_OK
) < 0)
4898 e
= stpcpy(e
, "cgroups-missing:");
4900 if (cg_all_unified() == 0)
4901 e
= stpcpy(e
, "cgroupsv1:");
4903 if (clock_is_localtime(NULL
) > 0)
4904 e
= stpcpy(e
, "local-hwclock:");
4906 r
= readlink_malloc("/var/run", &destination
);
4907 if (r
< 0 || !PATH_IN_SET(destination
, "../run", "/run"))
4908 e
= stpcpy(e
, "var-run-bad:");
4910 r
= read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid
);
4911 if (r
>= 0 && !streq(overflowuid
, "65534"))
4912 e
= stpcpy(e
, "overflowuid-not-65534:");
4914 r
= read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid
);
4915 if (r
>= 0 && !streq(overflowgid
, "65534"))
4916 e
= stpcpy(e
, "overflowgid-not-65534:");
4918 /* remove the last ':' */
4925 void manager_ref_console(Manager
*m
) {
4931 void manager_unref_console(Manager
*m
) {
4933 assert(m
->n_on_console
> 0);
4936 if (m
->n_on_console
== 0)
4937 m
->no_console_output
= false; /* unset no_console_output flag, since the console is definitely free now */
4940 void manager_override_log_level(Manager
*m
, int level
) {
4941 _cleanup_free_
char *s
= NULL
;
4944 if (!m
->log_level_overridden
) {
4945 m
->original_log_level
= log_get_max_level();
4946 m
->log_level_overridden
= true;
4949 (void) log_level_to_string_alloc(level
, &s
);
4950 log_info("Setting log level to %s.", strna(s
));
4952 log_set_max_level(level
);
4955 void manager_restore_original_log_level(Manager
*m
) {
4956 _cleanup_free_
char *s
= NULL
;
4959 if (!m
->log_level_overridden
)
4962 (void) log_level_to_string_alloc(m
->original_log_level
, &s
);
4963 log_info("Restoring log level to original (%s).", strna(s
));
4965 log_set_max_level(m
->original_log_level
);
4966 m
->log_level_overridden
= false;
4969 void manager_override_log_target(Manager
*m
, LogTarget target
) {
4972 if (!m
->log_target_overridden
) {
4973 m
->original_log_target
= log_get_target();
4974 m
->log_target_overridden
= true;
4977 log_info("Setting log target to %s.", log_target_to_string(target
));
4978 log_set_target(target
);
4981 void manager_restore_original_log_target(Manager
*m
) {
4984 if (!m
->log_target_overridden
)
4987 log_info("Restoring log target to original %s.", log_target_to_string(m
->original_log_target
));
4989 log_set_target(m
->original_log_target
);
4990 m
->log_target_overridden
= false;
4993 ManagerTimestamp
manager_timestamp_initrd_mangle(ManagerTimestamp s
) {
4995 s
>= MANAGER_TIMESTAMP_SECURITY_START
&&
4996 s
<= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
)
4997 return s
- MANAGER_TIMESTAMP_SECURITY_START
+ MANAGER_TIMESTAMP_INITRD_SECURITY_START
;
5001 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
5002 [MANAGER_INITIALIZING
] = "initializing",
5003 [MANAGER_STARTING
] = "starting",
5004 [MANAGER_RUNNING
] = "running",
5005 [MANAGER_DEGRADED
] = "degraded",
5006 [MANAGER_MAINTENANCE
] = "maintenance",
5007 [MANAGER_STOPPING
] = "stopping",
5010 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);
5012 static const char *const manager_timestamp_table
[_MANAGER_TIMESTAMP_MAX
] = {
5013 [MANAGER_TIMESTAMP_FIRMWARE
] = "firmware",
5014 [MANAGER_TIMESTAMP_LOADER
] = "loader",
5015 [MANAGER_TIMESTAMP_KERNEL
] = "kernel",
5016 [MANAGER_TIMESTAMP_INITRD
] = "initrd",
5017 [MANAGER_TIMESTAMP_USERSPACE
] = "userspace",
5018 [MANAGER_TIMESTAMP_FINISH
] = "finish",
5019 [MANAGER_TIMESTAMP_SECURITY_START
] = "security-start",
5020 [MANAGER_TIMESTAMP_SECURITY_FINISH
] = "security-finish",
5021 [MANAGER_TIMESTAMP_GENERATORS_START
] = "generators-start",
5022 [MANAGER_TIMESTAMP_GENERATORS_FINISH
] = "generators-finish",
5023 [MANAGER_TIMESTAMP_UNITS_LOAD_START
] = "units-load-start",
5024 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH
] = "units-load-finish",
5025 [MANAGER_TIMESTAMP_INITRD_SECURITY_START
] = "initrd-security-start",
5026 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH
] = "initrd-security-finish",
5027 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START
] = "initrd-generators-start",
5028 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH
] = "initrd-generators-finish",
5029 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START
] = "initrd-units-load-start",
5030 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH
] = "initrd-units-load-finish",
5033 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp
, ManagerTimestamp
);
5035 static const char* const oom_policy_table
[_OOM_POLICY_MAX
] = {
5036 [OOM_CONTINUE
] = "continue",
5037 [OOM_STOP
] = "stop",
5038 [OOM_KILL
] = "kill",
5041 DEFINE_STRING_TABLE_LOOKUP(oom_policy
, OOMPolicy
);