2 This file is part of systemd.
4 Copyright 2010 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/epoll.h>
27 #include <sys/inotify.h>
28 #include <sys/ioctl.h>
29 #include <sys/reboot.h>
30 #include <sys/timerfd.h>
38 #include "sd-daemon.h"
39 #include "sd-messages.h"
41 #include "alloc-util.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
48 #include "clean-ipc.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
53 #include "dirent-util.h"
56 #include "exit-status.h"
62 #include "locale-setup.h"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
74 #include "signal-util.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "user-util.h"
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
98 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
100 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
101 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
103 static int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
);
104 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
);
105 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
);
106 static int manager_run_generators(Manager
*m
);
108 static void manager_watch_jobs_in_progress(Manager
*m
) {
114 /* We do not want to show the cylon animation if the user
115 * needs to confirm service executions otherwise confirmation
116 * messages will be screwed by the cylon animation. */
117 if (!manager_is_confirm_spawn_disabled(m
))
120 if (m
->jobs_in_progress_event_source
)
123 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
124 r
= sd_event_add_time(
126 &m
->jobs_in_progress_event_source
,
129 manager_dispatch_jobs_in_progress
, m
);
133 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
136 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
138 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
141 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
142 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
146 p
= mempset(p
, ' ', pos
-2);
147 if (log_get_show_color())
148 p
= stpcpy(p
, ANSI_RED
);
152 if (pos
> 0 && pos
<= width
) {
153 if (log_get_show_color())
154 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
158 if (log_get_show_color())
159 p
= stpcpy(p
, ANSI_NORMAL
);
162 if (log_get_show_color())
163 p
= stpcpy(p
, ANSI_RED
);
166 p
= mempset(p
, ' ', width
-1-pos
);
167 if (log_get_show_color())
168 strcpy(p
, ANSI_NORMAL
);
172 void manager_flip_auto_status(Manager
*m
, bool enable
) {
176 if (m
->show_status
== SHOW_STATUS_AUTO
)
177 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
179 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
180 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
184 static void manager_print_jobs_in_progress(Manager
*m
) {
185 _cleanup_free_
char *job_of_n
= NULL
;
188 unsigned counter
= 0, print_nr
;
189 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
191 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
195 assert(m
->n_running_jobs
> 0);
197 manager_flip_auto_status(m
, true);
199 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
201 HASHMAP_FOREACH(j
, m
->jobs
, i
)
202 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
205 /* m->n_running_jobs must be consistent with the contents of m->jobs,
206 * so the above loop must have succeeded in finding j. */
207 assert(counter
== print_nr
+ 1);
210 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
212 cylon_pos
= 14 - cylon_pos
;
213 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
215 m
->jobs_in_progress_iteration
++;
217 if (m
->n_running_jobs
> 1) {
218 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
222 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
223 if (job_get_timeout(j
, &x
) > 0)
224 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
226 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
227 "%sA %s job is running for %s (%s / %s)",
229 job_type_to_string(j
->type
),
230 unit_description(j
->unit
),
234 static int have_ask_password(void) {
235 _cleanup_closedir_
DIR *dir
;
237 dir
= opendir("/run/systemd/ask-password");
250 if (!de
&& errno
> 0)
255 if (startswith(de
->d_name
, "ask."))
260 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
261 int fd
, uint32_t revents
, void *userdata
) {
262 Manager
*m
= userdata
;
268 m
->have_ask_password
= have_ask_password();
269 if (m
->have_ask_password
< 0)
270 /* Log error but continue. Negative have_ask_password
271 * is treated as unknown status. */
272 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
277 static void manager_close_ask_password(Manager
*m
) {
280 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
281 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
282 m
->have_ask_password
= -EINVAL
;
285 static int manager_check_ask_password(Manager
*m
) {
290 if (!m
->ask_password_event_source
) {
291 assert(m
->ask_password_inotify_fd
< 0);
293 mkdir_p_label("/run/systemd/ask-password", 0755);
295 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
296 if (m
->ask_password_inotify_fd
< 0)
297 return log_error_errno(errno
, "inotify_init1() failed: %m");
299 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
300 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
301 manager_close_ask_password(m
);
305 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
306 m
->ask_password_inotify_fd
, EPOLLIN
,
307 manager_dispatch_ask_password_fd
, m
);
309 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
310 manager_close_ask_password(m
);
314 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
316 /* Queries might have been added meanwhile... */
317 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
318 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
321 return m
->have_ask_password
;
324 static int manager_watch_idle_pipe(Manager
*m
) {
329 if (m
->idle_pipe_event_source
)
332 if (m
->idle_pipe
[2] < 0)
335 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
337 return log_error_errno(r
, "Failed to watch idle pipe: %m");
339 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
344 static void manager_close_idle_pipe(Manager
*m
) {
347 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
349 safe_close_pair(m
->idle_pipe
);
350 safe_close_pair(m
->idle_pipe
+ 2);
353 static int manager_setup_time_change(Manager
*m
) {
356 /* We only care for the cancellation event, hence we set the
357 * timeout to the latest possible value. */
358 struct itimerspec its
= {
359 .it_value
.tv_sec
= TIME_T_MAX
,
363 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
368 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
369 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
371 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
372 if (m
->time_change_fd
< 0)
373 return log_error_errno(errno
, "Failed to create timerfd: %m");
375 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
376 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
377 m
->time_change_fd
= safe_close(m
->time_change_fd
);
381 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
383 return log_error_errno(r
, "Failed to create time change event source: %m");
385 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
387 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
392 static int enable_special_signals(Manager
*m
) {
393 _cleanup_close_
int fd
= -1;
400 /* Enable that we get SIGINT on control-alt-del. In containers
401 * this will fail with EPERM (older) or EINVAL (newer), so
403 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
404 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
406 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
408 /* Support systems without virtual console */
410 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
412 /* Enable that we get SIGWINCH on kbrequest */
413 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
414 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
420 static int manager_setup_signals(Manager
*m
) {
421 struct sigaction sa
= {
422 .sa_handler
= SIG_DFL
,
423 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
430 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
432 /* We make liberal use of realtime signals here. On
433 * Linux/glibc we have 30 of them (with the exception of Linux
434 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
437 assert_se(sigemptyset(&mask
) == 0);
438 sigset_add_many(&mask
,
439 SIGCHLD
, /* Child died */
440 SIGTERM
, /* Reexecute daemon */
441 SIGHUP
, /* Reload configuration */
442 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
443 SIGUSR2
, /* systemd: dump status */
444 SIGINT
, /* Kernel sends us this on control-alt-del */
445 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
446 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
448 SIGRTMIN
+0, /* systemd: start default.target */
449 SIGRTMIN
+1, /* systemd: isolate rescue.target */
450 SIGRTMIN
+2, /* systemd: isolate emergency.target */
451 SIGRTMIN
+3, /* systemd: start halt.target */
452 SIGRTMIN
+4, /* systemd: start poweroff.target */
453 SIGRTMIN
+5, /* systemd: start reboot.target */
454 SIGRTMIN
+6, /* systemd: start kexec.target */
456 /* ... space for more special targets ... */
458 SIGRTMIN
+13, /* systemd: Immediate halt */
459 SIGRTMIN
+14, /* systemd: Immediate poweroff */
460 SIGRTMIN
+15, /* systemd: Immediate reboot */
461 SIGRTMIN
+16, /* systemd: Immediate kexec */
463 /* ... space for more immediate system state changes ... */
465 SIGRTMIN
+20, /* systemd: enable status messages */
466 SIGRTMIN
+21, /* systemd: disable status messages */
467 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
468 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
469 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
471 /* .. one free signal here ... */
473 #if !defined(__hppa64__) && !defined(__hppa__)
474 /* Apparently Linux on hppa has fewer RT
475 * signals (SIGRTMAX is SIGRTMIN+25 there),
476 * hence let's not try to make use of them
477 * here. Since these commands are accessible
478 * by different means and only really a safety
479 * net, the missing functionality on hppa
480 * shouldn't matter. */
482 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
483 SIGRTMIN
+27, /* systemd: set log target to console */
484 SIGRTMIN
+28, /* systemd: set log target to kmsg */
485 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
487 /* ... one free signal here SIGRTMIN+30 ... */
490 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
492 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
493 if (m
->signal_fd
< 0)
496 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
500 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
502 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
503 * notify processing can still figure out to which process/service a message belongs, before we reap the
504 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
505 * status information before detecting that there's no process in a cgroup. */
506 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
510 if (MANAGER_IS_SYSTEM(m
))
511 return enable_special_signals(m
);
516 static void manager_clean_environment(Manager
*m
) {
519 /* Let's remove some environment variables that we
520 * need ourselves to communicate with our clients */
535 static int manager_default_environment(Manager
*m
) {
538 if (MANAGER_IS_SYSTEM(m
)) {
539 /* The system manager always starts with a clean
540 * environment for its children. It does not import
541 * the kernel or the parents exported variables.
543 * The initial passed environ is untouched to keep
544 * /proc/self/environ valid; it is used for tagging
545 * the init process inside containers. */
546 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
549 /* Import locale variables LC_*= from configuration */
550 locale_setup(&m
->environment
);
552 /* The user manager passes its own environment
553 * along to its children. */
554 m
->environment
= strv_copy(environ
);
560 manager_clean_environment(m
);
561 strv_sort(m
->environment
);
566 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
571 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
573 m
= new0(Manager
, 1);
577 m
->unit_file_scope
= scope
;
578 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
579 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
580 m
->default_tasks_accounting
= true;
581 m
->default_tasks_max
= UINT64_MAX
;
584 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
585 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
588 /* Prepare log fields we can use for structured logging */
589 if (MANAGER_IS_SYSTEM(m
)) {
590 m
->unit_log_field
= "UNIT=";
591 m
->unit_log_format_string
= "UNIT=%s";
593 m
->invocation_log_field
= "INVOCATION_ID=";
594 m
->invocation_log_format_string
= "INVOCATION_ID=" SD_ID128_FORMAT_STR
;
596 m
->unit_log_field
= "USER_UNIT=";
597 m
->unit_log_format_string
= "USER_UNIT=%s";
599 m
->invocation_log_field
= "USER_INVOCATION_ID=";
600 m
->invocation_log_format_string
= "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR
;
603 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
605 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
606 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->cgroup_inotify_fd
=
607 m
->ask_password_inotify_fd
= -1;
609 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
611 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
613 m
->have_ask_password
= -EINVAL
; /* we don't know */
616 m
->test_run
= test_run
;
618 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
619 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
621 r
= manager_default_environment(m
);
625 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
629 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
633 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
637 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
641 r
= sd_event_default(&m
->event
);
645 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
649 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
653 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
657 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
659 r
= manager_setup_signals(m
);
663 r
= manager_setup_cgroup(m
);
667 r
= manager_setup_time_change(m
);
671 m
->udev
= udev_new();
677 /* Note that we do not set up the notify fd here. We do that after deserialization,
678 * since they might have gotten serialized across the reexec. */
680 m
->taint_usr
= dir_is_empty("/usr") > 0;
690 static int manager_setup_notify(Manager
*m
) {
696 if (m
->notify_fd
< 0) {
697 _cleanup_close_
int fd
= -1;
698 union sockaddr_union sa
= {
699 .sa
.sa_family
= AF_UNIX
,
701 static const int one
= 1;
704 /* First free all secondary fields */
705 m
->notify_socket
= mfree(m
->notify_socket
);
706 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
708 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
710 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
712 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
714 e
= manager_get_runtime_prefix(m
);
716 log_error("Failed to determine runtime prefix.");
720 m
->notify_socket
= strappend(e
, "/systemd/notify");
721 if (!m
->notify_socket
)
724 (void) mkdir_parents_label(m
->notify_socket
, 0755);
725 (void) unlink(m
->notify_socket
);
727 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
728 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
730 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
732 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
734 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
739 log_debug("Using notification socket %s", m
->notify_socket
);
742 if (!m
->notify_event_source
) {
743 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
745 return log_error_errno(r
, "Failed to allocate notify event source: %m");
747 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
748 * service an exit message belongs. */
749 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
751 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
753 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
759 static int manager_setup_cgroups_agent(Manager
*m
) {
761 static const union sockaddr_union sa
= {
762 .un
.sun_family
= AF_UNIX
,
763 .un
.sun_path
= "/run/systemd/cgroups-agent",
767 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
768 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
769 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
770 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
771 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
772 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
773 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
774 * we thus won't lose messages.
776 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
777 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
778 * bus for these messages. */
783 if (!MANAGER_IS_SYSTEM(m
))
786 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER
) > 0) /* We don't need this anymore on the unified hierarchy */
789 if (m
->cgroups_agent_fd
< 0) {
790 _cleanup_close_
int fd
= -1;
792 /* First free all secondary fields */
793 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
795 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
797 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
799 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
801 (void) unlink(sa
.un
.sun_path
);
803 /* Only allow root to connect to this socket */
805 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
807 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
809 m
->cgroups_agent_fd
= fd
;
813 if (!m
->cgroups_agent_event_source
) {
814 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
816 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
818 /* Process cgroups notifications early, but after having processed service notification messages or
819 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
820 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
821 * cgroup inotify for the unified cgroup stuff. */
822 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
824 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
826 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
832 static int manager_setup_user_lookup_fd(Manager
*m
) {
837 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
838 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
839 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
840 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
841 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
842 * hence we establish this communication channel so that forked off processes can pass their UID/GID
843 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
844 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
847 * You might wonder why we need a communication channel for this that is independent of the usual notification
848 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
849 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
850 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
852 * Note that this function is called under two circumstances: when we first initialize (in which case we
853 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
854 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
856 if (m
->user_lookup_fds
[0] < 0) {
858 /* Free all secondary fields */
859 safe_close_pair(m
->user_lookup_fds
);
860 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
862 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
863 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
865 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
868 if (!m
->user_lookup_event_source
) {
869 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
871 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
873 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
875 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
877 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
879 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
885 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
886 bool try_bus_connect
;
895 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
897 /* Try to connect to the buses, if possible. */
898 return bus_init(m
, try_bus_connect
);
901 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
907 while ((u
= m
->cleanup_queue
)) {
908 assert(u
->in_cleanup_queue
);
918 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
919 GC_OFFSET_UNSURE
, /* No clue */
920 GC_OFFSET_GOOD
, /* We still need this unit */
921 GC_OFFSET_BAD
, /* We don't need this unit anymore */
925 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
929 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
931 /* Recursively mark referenced units as GOOD as well */
932 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
933 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
934 unit_gc_mark_good(other
, gc_marker
);
937 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
944 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
945 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
946 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
947 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
950 if (u
->in_cleanup_queue
)
953 if (unit_check_gc(u
))
956 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
960 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
961 unit_gc_sweep(other
, gc_marker
);
963 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
966 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
973 /* We were unable to find anything out about this entry, so
974 * let's investigate it later */
975 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
976 unit_add_to_gc_queue(u
);
980 /* We definitely know that this one is not useful anymore, so
981 * let's mark it for deletion */
982 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
983 unit_add_to_cleanup_queue(u
);
987 unit_gc_mark_good(u
, gc_marker
);
990 static unsigned manager_dispatch_gc_unit_queue(Manager
*m
) {
991 unsigned n
= 0, gc_marker
;
996 /* log_debug("Running GC..."); */
998 m
->gc_marker
+= _GC_OFFSET_MAX
;
999 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
1002 gc_marker
= m
->gc_marker
;
1004 while ((u
= m
->gc_unit_queue
)) {
1005 assert(u
->in_gc_queue
);
1007 unit_gc_sweep(u
, gc_marker
);
1009 LIST_REMOVE(gc_queue
, m
->gc_unit_queue
, u
);
1010 u
->in_gc_queue
= false;
1014 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1015 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1017 log_unit_debug(u
, "Collecting.");
1018 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1019 unit_add_to_cleanup_queue(u
);
1026 static unsigned manager_dispatch_gc_job_queue(Manager
*m
) {
1032 while ((j
= m
->gc_job_queue
)) {
1033 assert(j
->in_gc_queue
);
1035 LIST_REMOVE(gc_queue
, m
->gc_job_queue
, j
);
1036 j
->in_gc_queue
= false;
1040 if (job_check_gc(j
))
1043 log_unit_debug(j
->unit
, "Collecting job.");
1044 (void) job_finish_and_invalidate(j
, JOB_COLLECTED
, false, false);
1050 static void manager_clear_jobs_and_units(Manager
*m
) {
1055 while ((u
= hashmap_first(m
->units
)))
1058 manager_dispatch_cleanup_queue(m
);
1060 assert(!m
->load_queue
);
1061 assert(!m
->run_queue
);
1062 assert(!m
->dbus_unit_queue
);
1063 assert(!m
->dbus_job_queue
);
1064 assert(!m
->cleanup_queue
);
1065 assert(!m
->gc_unit_queue
);
1066 assert(!m
->gc_job_queue
);
1068 assert(hashmap_isempty(m
->jobs
));
1069 assert(hashmap_isempty(m
->units
));
1071 m
->n_on_console
= 0;
1072 m
->n_running_jobs
= 0;
1075 Manager
* manager_free(Manager
*m
) {
1082 manager_clear_jobs_and_units(m
);
1084 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1085 if (unit_vtable
[c
]->shutdown
)
1086 unit_vtable
[c
]->shutdown(m
);
1088 /* If we reexecute ourselves, we keep the root cgroup
1090 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1092 lookup_paths_flush_generator(&m
->lookup_paths
);
1096 dynamic_user_vacuum(m
, false);
1097 hashmap_free(m
->dynamic_users
);
1099 hashmap_free(m
->units
);
1100 hashmap_free(m
->units_by_invocation_id
);
1101 hashmap_free(m
->jobs
);
1102 hashmap_free(m
->watch_pids1
);
1103 hashmap_free(m
->watch_pids2
);
1104 hashmap_free(m
->watch_bus
);
1106 set_free(m
->startup_units
);
1107 set_free(m
->failed_units
);
1109 sd_event_source_unref(m
->signal_event_source
);
1110 sd_event_source_unref(m
->notify_event_source
);
1111 sd_event_source_unref(m
->cgroups_agent_event_source
);
1112 sd_event_source_unref(m
->time_change_event_source
);
1113 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1114 sd_event_source_unref(m
->run_queue_event_source
);
1115 sd_event_source_unref(m
->user_lookup_event_source
);
1117 safe_close(m
->signal_fd
);
1118 safe_close(m
->notify_fd
);
1119 safe_close(m
->cgroups_agent_fd
);
1120 safe_close(m
->time_change_fd
);
1121 safe_close_pair(m
->user_lookup_fds
);
1123 manager_close_ask_password(m
);
1125 manager_close_idle_pipe(m
);
1127 udev_unref(m
->udev
);
1128 sd_event_unref(m
->event
);
1130 free(m
->notify_socket
);
1132 lookup_paths_free(&m
->lookup_paths
);
1133 strv_free(m
->environment
);
1135 hashmap_free(m
->cgroup_unit
);
1136 set_free_free(m
->unit_path_cache
);
1138 free(m
->switch_root
);
1139 free(m
->switch_root_init
);
1141 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1142 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1144 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1145 hashmap_free(m
->units_requiring_mounts_for
);
1147 hashmap_free(m
->uid_refs
);
1148 hashmap_free(m
->gid_refs
);
1153 void manager_enumerate(Manager
*m
) {
1158 /* Let's ask every type to load all units from disk/kernel
1159 * that it might know */
1160 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1161 if (!unit_type_supported(c
)) {
1162 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1166 if (!unit_vtable
[c
]->enumerate
)
1169 unit_vtable
[c
]->enumerate(m
);
1172 manager_dispatch_load_queue(m
);
1175 static void manager_coldplug(Manager
*m
) {
1183 /* Then, let's set up their initial state. */
1184 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1186 /* ignore aliases */
1190 r
= unit_coldplug(u
);
1192 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1196 static void manager_build_unit_path_cache(Manager
*m
) {
1202 set_free_free(m
->unit_path_cache
);
1204 m
->unit_path_cache
= set_new(&string_hash_ops
);
1205 if (!m
->unit_path_cache
) {
1210 /* This simply builds a list of files we know exist, so that
1211 * we don't always have to go to disk */
1213 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1214 _cleanup_closedir_
DIR *d
= NULL
;
1219 if (errno
!= ENOENT
)
1220 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1224 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1227 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
);
1233 r
= set_consume(m
->unit_path_cache
, p
);
1242 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1243 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1246 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1252 HASHMAP_FOREACH(u
, m
->units
, i
) {
1254 if (fdset_size(fds
) <= 0)
1257 if (!UNIT_VTABLE(u
)->distribute_fds
)
1260 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1264 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1269 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1273 /* Make sure the transient directory always exists, so that it remains in the search path */
1275 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1280 dual_timestamp_get(&m
->generators_start_timestamp
);
1281 r
= manager_run_generators(m
);
1282 dual_timestamp_get(&m
->generators_finish_timestamp
);
1286 lookup_paths_reduce(&m
->lookup_paths
);
1287 manager_build_unit_path_cache(m
);
1289 /* If we will deserialize make sure that during enumeration
1290 * this is already known, so we increase the counter here
1295 /* First, enumerate what we can from all config files */
1296 dual_timestamp_get(&m
->units_load_start_timestamp
);
1297 manager_enumerate(m
);
1298 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1300 /* Second, deserialize if there is something to deserialize */
1302 r
= manager_deserialize(m
, serialization
, fds
);
1304 /* Any fds left? Find some unit which wants them. This is
1305 * useful to allow container managers to pass some file
1306 * descriptors to us pre-initialized. This enables
1307 * socket-based activation of entire containers. */
1308 manager_distribute_fds(m
, fds
);
1310 /* We might have deserialized the notify fd, but if we didn't
1311 * then let's create the bus now */
1312 q
= manager_setup_notify(m
);
1313 if (q
< 0 && r
== 0)
1316 q
= manager_setup_cgroups_agent(m
);
1317 if (q
< 0 && r
== 0)
1320 q
= manager_setup_user_lookup_fd(m
);
1321 if (q
< 0 && r
== 0)
1324 /* Let's connect to the bus now. */
1325 (void) manager_connect_bus(m
, !!serialization
);
1327 (void) bus_track_coldplug(m
, &m
->subscribed
, false, m
->deserialized_subscribed
);
1328 m
->deserialized_subscribed
= strv_free(m
->deserialized_subscribed
);
1330 /* Third, fire things up! */
1331 manager_coldplug(m
);
1333 /* Release any dynamic users no longer referenced */
1334 dynamic_user_vacuum(m
, true);
1336 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1337 manager_vacuum_uid_refs(m
);
1338 manager_vacuum_gid_refs(m
);
1340 if (serialization
) {
1341 assert(m
->n_reloading
> 0);
1344 /* Let's wait for the UnitNew/JobNew messages being
1345 * sent, before we notify that the reload is
1347 m
->send_reloading_done
= true;
1353 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1358 assert(type
< _JOB_TYPE_MAX
);
1360 assert(mode
< _JOB_MODE_MAX
);
1362 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1363 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1365 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1366 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1368 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1370 type
= job_type_collapse(type
, unit
);
1372 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1376 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1377 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1378 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1382 if (mode
== JOB_ISOLATE
) {
1383 r
= transaction_add_isolate_jobs(tr
, m
);
1388 r
= transaction_activate(tr
, m
, mode
, e
);
1392 log_unit_debug(unit
,
1393 "Enqueued job %s/%s as %u", unit
->id
,
1394 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1397 *_ret
= tr
->anchor_job
;
1399 transaction_free(tr
);
1403 transaction_abort(tr
);
1404 transaction_free(tr
);
1408 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1413 assert(type
< _JOB_TYPE_MAX
);
1415 assert(mode
< _JOB_MODE_MAX
);
1417 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1421 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1424 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1425 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1429 assert(type
< _JOB_TYPE_MAX
);
1431 assert(mode
< _JOB_MODE_MAX
);
1433 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1435 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1440 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1443 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1446 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1450 return hashmap_get(m
->units
, name
);
1453 unsigned manager_dispatch_load_queue(Manager
*m
) {
1459 /* Make sure we are not run recursively */
1460 if (m
->dispatching_load_queue
)
1463 m
->dispatching_load_queue
= true;
1465 /* Dispatches the load queue. Takes a unit from the queue and
1466 * tries to load its data until the queue is empty */
1468 while ((u
= m
->load_queue
)) {
1469 assert(u
->in_load_queue
);
1475 m
->dispatching_load_queue
= false;
1479 int manager_load_unit_prepare(
1491 assert(name
|| path
);
1493 /* This will prepare the unit for loading, but not actually
1494 * load anything from disk. */
1496 if (path
&& !is_path(path
))
1497 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1500 name
= basename(path
);
1502 t
= unit_name_to_type(name
);
1504 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1505 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1506 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1508 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1511 ret
= manager_get_unit(m
, name
);
1517 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1522 ret
->fragment_path
= strdup(path
);
1523 if (!ret
->fragment_path
) {
1529 r
= unit_add_name(ret
, name
);
1535 unit_add_to_load_queue(ret
);
1536 unit_add_to_dbus_queue(ret
);
1537 unit_add_to_gc_queue(ret
);
1545 int manager_load_unit(
1556 /* This will load the service information files, but not actually
1557 * start any services or anything. */
1559 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1563 manager_dispatch_load_queue(m
);
1566 *_ret
= unit_follow_merge(*_ret
);
1571 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1578 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1579 job_dump(j
, f
, prefix
);
1582 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1590 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1592 unit_dump(u
, f
, prefix
);
1595 void manager_clear_jobs(Manager
*m
) {
1600 while ((j
= hashmap_first(m
->jobs
)))
1601 /* No need to recurse. We're cancelling all jobs. */
1602 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1605 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1606 Manager
*m
= userdata
;
1612 while ((j
= m
->run_queue
)) {
1613 assert(j
->installed
);
1614 assert(j
->in_run_queue
);
1616 job_run_and_invalidate(j
);
1619 if (m
->n_running_jobs
> 0)
1620 manager_watch_jobs_in_progress(m
);
1622 if (m
->n_on_console
> 0)
1623 manager_watch_idle_pipe(m
);
1628 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1635 if (m
->dispatching_dbus_queue
)
1638 m
->dispatching_dbus_queue
= true;
1640 while ((u
= m
->dbus_unit_queue
)) {
1641 assert(u
->in_dbus_queue
);
1643 bus_unit_send_change_signal(u
);
1647 while ((j
= m
->dbus_job_queue
)) {
1648 assert(j
->in_dbus_queue
);
1650 bus_job_send_change_signal(j
);
1654 m
->dispatching_dbus_queue
= false;
1656 if (m
->send_reloading_done
) {
1657 m
->send_reloading_done
= false;
1659 bus_manager_send_reloading(m
, false);
1662 if (m
->queued_message
)
1663 bus_send_queued_message(m
);
1668 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1669 Manager
*m
= userdata
;
1670 char buf
[PATH_MAX
+1];
1673 n
= recv(fd
, buf
, sizeof(buf
), 0);
1675 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1677 log_error("Got zero-length cgroups agent message, ignoring.");
1680 if ((size_t) n
>= sizeof(buf
)) {
1681 log_error("Got overly long cgroups agent message, ignoring.");
1685 if (memchr(buf
, 0, n
)) {
1686 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1691 manager_notify_cgroup_empty(m
, buf
);
1692 bus_forward_agent_released(m
, buf
);
1697 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, FDSet
*fds
) {
1698 _cleanup_strv_free_
char **tags
= NULL
;
1704 tags
= strv_split(buf
, "\n\r");
1710 if (UNIT_VTABLE(u
)->notify_message
)
1711 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1712 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG
)) {
1713 _cleanup_free_
char *x
= NULL
, *y
= NULL
;
1717 y
= ellipsize(x
, 20, 90);
1718 log_unit_debug(u
, "Got notification message \"%s\", ignoring.", strnull(y
));
1722 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1724 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1725 Manager
*m
= userdata
;
1726 char buf
[NOTIFY_BUFFER_MAX
+1];
1727 struct iovec iovec
= {
1729 .iov_len
= sizeof(buf
)-1,
1732 struct cmsghdr cmsghdr
;
1733 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1734 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1736 struct msghdr msghdr
= {
1739 .msg_control
= &control
,
1740 .msg_controllen
= sizeof(control
),
1743 struct cmsghdr
*cmsg
;
1744 struct ucred
*ucred
= NULL
;
1746 int r
, *fd_array
= NULL
;
1751 assert(m
->notify_fd
== fd
);
1753 if (revents
!= EPOLLIN
) {
1754 log_warning("Got unexpected poll event for notify fd.");
1758 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
|MSG_TRUNC
);
1760 if (IN_SET(errno
, EAGAIN
, EINTR
))
1761 return 0; /* Spurious wakeup, try again */
1763 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1764 * won't take notification messages anymore, but that's still better than busy looping around this:
1765 * being woken up over and over again but being unable to actually read the message off the socket. */
1766 return log_error_errno(errno
, "Failed to receive notification message: %m");
1769 CMSG_FOREACH(cmsg
, &msghdr
) {
1770 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1772 fd_array
= (int*) CMSG_DATA(cmsg
);
1773 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1775 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1776 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1777 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1779 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1786 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1788 close_many(fd_array
, n_fds
);
1794 if (!ucred
|| ucred
->pid
<= 0) {
1795 log_warning("Received notify message without valid credentials. Ignoring.");
1799 if ((size_t) n
>= sizeof(buf
) || (msghdr
.msg_flags
& MSG_TRUNC
)) {
1800 log_warning("Received notify message exceeded maximum size. Ignoring.");
1804 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1805 * trailing NUL byte in the message, but don't expect it. */
1806 if (n
> 1 && memchr(buf
, 0, n
-1)) {
1807 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1811 /* Make sure it's NUL-terminated. */
1814 /* Notify every unit that might be interested, but try
1815 * to avoid notifying the same one multiple times. */
1816 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1818 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, fds
);
1820 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1822 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, fds
);
1824 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1825 if (u3
&& u3
!= u2
&& u3
!= u1
)
1826 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, fds
);
1828 if (!u1
&& !u2
&& !u3
)
1829 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1831 if (fdset_size(fds
) > 0)
1832 log_warning("Got extra auxiliary fds with notification message, closing them.");
1837 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1844 sd_event_get_iteration(m
->event
, &iteration
);
1846 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1848 unit_unwatch_pid(u
, si
->si_pid
);
1850 if (UNIT_VTABLE(u
)->sigchld_event
) {
1851 if (set_size(u
->pids
) <= 1 ||
1852 iteration
!= u
->sigchldgen
||
1853 unit_main_pid(u
) == si
->si_pid
||
1854 unit_control_pid(u
) == si
->si_pid
) {
1855 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1856 u
->sigchldgen
= iteration
;
1858 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1862 static int manager_dispatch_sigchld(Manager
*m
) {
1868 /* First we call waitd() for a PID and do not reap the
1869 * zombie. That way we can still access /proc/$PID for
1870 * it while it is a zombie. */
1871 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1873 if (errno
== ECHILD
)
1885 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1886 _cleanup_free_
char *name
= NULL
;
1889 get_process_comm(si
.si_pid
, &name
);
1891 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1892 si
.si_pid
, strna(name
),
1893 sigchld_code_to_string(si
.si_code
),
1895 strna(si
.si_code
== CLD_EXITED
1896 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1897 : signal_to_string(si
.si_status
)));
1899 /* And now figure out the unit this belongs
1900 * to, it might be multiple... */
1901 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1903 invoke_sigchld_event(m
, u1
, &si
);
1904 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1906 invoke_sigchld_event(m
, u2
, &si
);
1907 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1908 if (u3
&& u3
!= u2
&& u3
!= u1
)
1909 invoke_sigchld_event(m
, u3
, &si
);
1912 /* And now, we actually reap the zombie. */
1913 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1924 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1925 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1928 log_debug("Activating special unit %s", name
);
1930 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1932 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1937 static void manager_handle_ctrl_alt_del(Manager
*m
) {
1938 /* If the user presses C-A-D more than
1939 * 7 times within 2s, we reboot/shutdown immediately,
1940 * unless it was disabled in system.conf */
1942 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
) || m
->cad_burst_action
== EMERGENCY_ACTION_NONE
)
1943 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1945 emergency_action(m
, m
->cad_burst_action
, NULL
,
1946 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1949 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1950 Manager
*m
= userdata
;
1952 struct signalfd_siginfo sfsi
;
1953 bool sigchld
= false;
1957 assert(m
->signal_fd
== fd
);
1959 if (revents
!= EPOLLIN
) {
1960 log_warning("Got unexpected events from signal file descriptor.");
1965 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1966 if (n
!= sizeof(sfsi
)) {
1968 log_warning("Truncated read from signal fd (%zu bytes)!", n
);
1972 if (IN_SET(errno
, EINTR
, EAGAIN
))
1975 /* We return an error here, which will kill this handler,
1976 * to avoid a busy loop on read error. */
1977 return log_error_errno(errno
, "Reading from signal fd failed: %m");
1980 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1981 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1982 ? LOG_DEBUG
: LOG_INFO
,
1985 switch (sfsi
.ssi_signo
) {
1992 if (MANAGER_IS_SYSTEM(m
)) {
1993 /* This is for compatibility with the
1994 * original sysvinit */
1995 m
->exit_code
= MANAGER_REEXECUTE
;
2002 if (MANAGER_IS_SYSTEM(m
)) {
2003 manager_handle_ctrl_alt_del(m
);
2007 /* Run the exit target if there is one, if not, just exit. */
2008 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
2009 m
->exit_code
= MANAGER_EXIT
;
2016 if (MANAGER_IS_SYSTEM(m
))
2017 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
2019 /* This is a nop on non-init */
2023 if (MANAGER_IS_SYSTEM(m
))
2024 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
2026 /* This is a nop on non-init */
2032 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
2034 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
2035 log_info("Trying to reconnect to bus...");
2039 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
2040 log_info("Loading D-Bus service...");
2041 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2048 _cleanup_free_
char *dump
= NULL
;
2049 _cleanup_fclose_
FILE *f
= NULL
;
2052 f
= open_memstream(&dump
, &size
);
2054 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2058 manager_dump_units(m
, f
, "\t");
2059 manager_dump_jobs(m
, f
, "\t");
2061 r
= fflush_and_check(f
);
2063 log_warning_errno(r
, "Failed to write status stream: %m");
2067 log_dump(LOG_INFO
, dump
);
2072 m
->exit_code
= MANAGER_RELOAD
;
2077 /* Starting SIGRTMIN+0 */
2078 static const char * const target_table
[] = {
2079 [0] = SPECIAL_DEFAULT_TARGET
,
2080 [1] = SPECIAL_RESCUE_TARGET
,
2081 [2] = SPECIAL_EMERGENCY_TARGET
,
2082 [3] = SPECIAL_HALT_TARGET
,
2083 [4] = SPECIAL_POWEROFF_TARGET
,
2084 [5] = SPECIAL_REBOOT_TARGET
,
2085 [6] = SPECIAL_KEXEC_TARGET
2088 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2089 static const ManagerExitCode code_table
[] = {
2091 [1] = MANAGER_POWEROFF
,
2092 [2] = MANAGER_REBOOT
,
2096 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2097 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2098 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2099 manager_start_target(m
, target_table
[idx
],
2100 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
2104 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2105 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2106 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2110 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2113 manager_set_show_status(m
, SHOW_STATUS_YES
);
2117 manager_set_show_status(m
, SHOW_STATUS_NO
);
2121 log_set_max_level(LOG_DEBUG
);
2122 log_info("Setting log level to debug.");
2126 log_set_max_level(LOG_INFO
);
2127 log_info("Setting log level to info.");
2131 if (MANAGER_IS_USER(m
)) {
2132 m
->exit_code
= MANAGER_EXIT
;
2136 /* This is a nop on init */
2140 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2141 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2142 log_notice("Setting log target to journal-or-kmsg.");
2146 log_set_target(LOG_TARGET_CONSOLE
);
2147 log_notice("Setting log target to console.");
2151 log_set_target(LOG_TARGET_KMSG
);
2152 log_notice("Setting log target to kmsg.");
2156 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2163 manager_dispatch_sigchld(m
);
2168 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2169 Manager
*m
= userdata
;
2174 assert(m
->time_change_fd
== fd
);
2176 log_struct(LOG_INFO
,
2177 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE
),
2178 LOG_MESSAGE("Time has been changed"),
2181 /* Restart the watch */
2182 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2183 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2185 manager_setup_time_change(m
);
2187 HASHMAP_FOREACH(u
, m
->units
, i
)
2188 if (UNIT_VTABLE(u
)->time_change
)
2189 UNIT_VTABLE(u
)->time_change(u
);
2194 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2195 Manager
*m
= userdata
;
2198 assert(m
->idle_pipe
[2] == fd
);
2200 m
->no_console_output
= m
->n_on_console
> 0;
2202 manager_close_idle_pipe(m
);
2207 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2208 Manager
*m
= userdata
;
2215 manager_print_jobs_in_progress(m
);
2217 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2218 r
= sd_event_source_set_time(source
, next
);
2222 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2225 int manager_loop(Manager
*m
) {
2228 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2231 m
->exit_code
= MANAGER_OK
;
2233 /* Release the path cache */
2234 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2236 manager_check_finished(m
);
2238 /* There might still be some zombies hanging around from
2239 * before we were exec()'ed. Let's reap them. */
2240 r
= manager_dispatch_sigchld(m
);
2244 while (m
->exit_code
== MANAGER_OK
) {
2247 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2250 if (!ratelimit_test(&rl
)) {
2251 /* Yay, something is going seriously wrong, pause a little */
2252 log_warning("Looping too fast. Throttling execution a little.");
2256 if (manager_dispatch_load_queue(m
) > 0)
2259 if (manager_dispatch_gc_job_queue(m
) > 0)
2262 if (manager_dispatch_gc_unit_queue(m
) > 0)
2265 if (manager_dispatch_cleanup_queue(m
) > 0)
2268 if (manager_dispatch_cgroup_queue(m
) > 0)
2271 if (manager_dispatch_dbus_queue(m
) > 0)
2274 /* Sleep for half the watchdog time */
2275 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2276 wait_usec
= m
->runtime_watchdog
/ 2;
2280 wait_usec
= USEC_INFINITY
;
2282 r
= sd_event_run(m
->event
, wait_usec
);
2284 return log_error_errno(r
, "Failed to run event loop: %m");
2287 return m
->exit_code
;
2290 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2291 _cleanup_free_
char *n
= NULL
;
2292 sd_id128_t invocation_id
;
2300 r
= unit_name_from_dbus_path(s
, &n
);
2304 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2305 * as invocation ID. */
2306 r
= sd_id128_from_string(n
, &invocation_id
);
2308 u
= hashmap_get(m
->units_by_invocation_id
, &invocation_id
);
2314 return sd_bus_error_setf(e
, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID
, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR
" known.", SD_ID128_FORMAT_VAL(invocation_id
));
2317 /* If this didn't work, we use the suffix as unit name. */
2318 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2326 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2336 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2340 r
= safe_atou(p
, &id
);
2344 j
= manager_get_job(m
, id
);
2353 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2356 _cleanup_free_
char *p
= NULL
;
2360 if (!MANAGER_IS_SYSTEM(m
))
2363 audit_fd
= get_audit_fd();
2367 /* Don't generate audit events if the service was already
2368 * started and we're just deserializing */
2369 if (MANAGER_IS_RELOADING(m
))
2372 if (u
->type
!= UNIT_SERVICE
)
2375 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2377 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2381 msg
= strjoina("unit=", p
);
2382 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2384 /* We aren't allowed to send audit messages?
2385 * Then let's not retry again. */
2388 log_warning_errno(errno
, "Failed to send audit message: %m");
2394 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2395 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2396 _cleanup_free_
char *message
= NULL
;
2397 _cleanup_close_
int fd
= -1;
2400 /* Don't generate plymouth events if the service was already
2401 * started and we're just deserializing */
2402 if (MANAGER_IS_RELOADING(m
))
2405 if (!MANAGER_IS_SYSTEM(m
))
2408 if (detect_container() > 0)
2411 if (u
->type
!= UNIT_SERVICE
&&
2412 u
->type
!= UNIT_MOUNT
&&
2413 u
->type
!= UNIT_SWAP
)
2416 /* We set SOCK_NONBLOCK here so that we rather drop the
2417 * message then wait for plymouth */
2418 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2420 log_error_errno(errno
, "socket() failed: %m");
2424 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2426 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2427 log_error_errno(errno
, "connect() failed: %m");
2431 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2437 if (write(fd
, message
, n
+ 1) != n
+ 1)
2438 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2439 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2442 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2449 path
= MANAGER_IS_SYSTEM(m
) ? "/run/systemd" : "/tmp";
2450 fd
= open_tmpfile_unlinkable(path
, O_RDWR
|O_CLOEXEC
);
2454 log_debug("Serializing state to %s", path
);
2456 f
= fdopen(fd
, "w+");
2467 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2480 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2481 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2482 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2483 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2485 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2486 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2487 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2488 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2491 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2492 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2493 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2494 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2495 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2496 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2497 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2498 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2501 if (!switching_root
) {
2502 STRV_FOREACH(e
, m
->environment
) {
2503 _cleanup_free_
char *ce
;
2509 fprintf(f
, "env=%s\n", *e
);
2513 if (m
->notify_fd
>= 0) {
2516 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2520 fprintf(f
, "notify-fd=%i\n", copy
);
2521 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2524 if (m
->cgroups_agent_fd
>= 0) {
2527 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2531 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2534 if (m
->user_lookup_fds
[0] >= 0) {
2537 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2541 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2545 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2548 bus_track_serialize(m
->subscribed
, f
, "subscribed");
2550 r
= dynamic_user_serialize(m
, f
, fds
);
2554 manager_serialize_uid_refs(m
, f
);
2555 manager_serialize_gid_refs(m
, f
);
2559 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2567 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2574 assert(m
->n_reloading
> 0);
2580 r
= bus_fdset_add_all(m
, fds
);
2587 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2593 log_debug("Deserializing state...");
2598 char line
[LINE_MAX
], *l
;
2601 if (!fgets(line
, sizeof(line
), f
)) {
2616 if ((val
= startswith(l
, "current-job-id="))) {
2619 if (safe_atou32(val
, &id
) < 0)
2620 log_debug("Failed to parse current job id value %s", val
);
2622 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2624 } else if ((val
= startswith(l
, "n-installed-jobs="))) {
2627 if (safe_atou32(val
, &n
) < 0)
2628 log_debug("Failed to parse installed jobs counter %s", val
);
2630 m
->n_installed_jobs
+= n
;
2632 } else if ((val
= startswith(l
, "n-failed-jobs="))) {
2635 if (safe_atou32(val
, &n
) < 0)
2636 log_debug("Failed to parse failed jobs counter %s", val
);
2638 m
->n_failed_jobs
+= n
;
2640 } else if ((val
= startswith(l
, "taint-usr="))) {
2643 b
= parse_boolean(val
);
2645 log_debug("Failed to parse taint /usr flag %s", val
);
2647 m
->taint_usr
= m
->taint_usr
|| b
;
2649 } else if ((val
= startswith(l
, "firmware-timestamp=")))
2650 dual_timestamp_deserialize(val
, &m
->firmware_timestamp
);
2651 else if ((val
= startswith(l
, "loader-timestamp=")))
2652 dual_timestamp_deserialize(val
, &m
->loader_timestamp
);
2653 else if ((val
= startswith(l
, "kernel-timestamp=")))
2654 dual_timestamp_deserialize(val
, &m
->kernel_timestamp
);
2655 else if ((val
= startswith(l
, "initrd-timestamp=")))
2656 dual_timestamp_deserialize(val
, &m
->initrd_timestamp
);
2657 else if ((val
= startswith(l
, "userspace-timestamp=")))
2658 dual_timestamp_deserialize(val
, &m
->userspace_timestamp
);
2659 else if ((val
= startswith(l
, "finish-timestamp=")))
2660 dual_timestamp_deserialize(val
, &m
->finish_timestamp
);
2661 else if ((val
= startswith(l
, "security-start-timestamp=")))
2662 dual_timestamp_deserialize(val
, &m
->security_start_timestamp
);
2663 else if ((val
= startswith(l
, "security-finish-timestamp=")))
2664 dual_timestamp_deserialize(val
, &m
->security_finish_timestamp
);
2665 else if ((val
= startswith(l
, "generators-start-timestamp=")))
2666 dual_timestamp_deserialize(val
, &m
->generators_start_timestamp
);
2667 else if ((val
= startswith(l
, "generators-finish-timestamp=")))
2668 dual_timestamp_deserialize(val
, &m
->generators_finish_timestamp
);
2669 else if ((val
= startswith(l
, "units-load-start-timestamp=")))
2670 dual_timestamp_deserialize(val
, &m
->units_load_start_timestamp
);
2671 else if ((val
= startswith(l
, "units-load-finish-timestamp=")))
2672 dual_timestamp_deserialize(val
, &m
->units_load_finish_timestamp
);
2673 else if (startswith(l
, "env=")) {
2674 _cleanup_free_
char *uce
= NULL
;
2677 r
= cunescape(l
+ 4, UNESCAPE_RELAX
, &uce
);
2681 e
= strv_env_set(m
->environment
, uce
);
2687 strv_free(m
->environment
);
2690 } else if ((val
= startswith(l
, "notify-fd="))) {
2693 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2694 log_debug("Failed to parse notify fd: %s", val
);
2696 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2697 safe_close(m
->notify_fd
);
2698 m
->notify_fd
= fdset_remove(fds
, fd
);
2701 } else if ((val
= startswith(l
, "notify-socket="))) {
2710 free(m
->notify_socket
);
2711 m
->notify_socket
= n
;
2713 } else if ((val
= startswith(l
, "cgroups-agent-fd="))) {
2716 if (safe_atoi(val
, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2717 log_debug("Failed to parse cgroups agent fd: %s", val
);
2719 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2720 safe_close(m
->cgroups_agent_fd
);
2721 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2724 } else if ((val
= startswith(l
, "user-lookup="))) {
2727 if (sscanf(val
, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2728 log_debug("Failed to parse user lookup fd: %s", val
);
2730 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2731 safe_close_pair(m
->user_lookup_fds
);
2732 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2733 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2736 } else if ((val
= startswith(l
, "dynamic-user=")))
2737 dynamic_user_deserialize_one(m
, val
, fds
);
2738 else if ((val
= startswith(l
, "destroy-ipc-uid=")))
2739 manager_deserialize_uid_refs_one(m
, val
);
2740 else if ((val
= startswith(l
, "destroy-ipc-gid=")))
2741 manager_deserialize_gid_refs_one(m
, val
);
2742 else if ((val
= startswith(l
, "subscribed="))) {
2744 if (strv_extend(&m
->deserialized_subscribed
, val
) < 0)
2747 } else if (!startswith(l
, "kdbus-fd=")) /* ignore this one */
2748 log_debug("Unknown serialization item '%s'", l
);
2753 char name
[UNIT_NAME_MAX
+2];
2756 if (!fgets(name
, sizeof(name
), f
)) {
2767 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2771 r
= unit_deserialize(u
, f
, fds
);
2780 assert(m
->n_reloading
> 0);
2786 int manager_reload(Manager
*m
) {
2788 _cleanup_fclose_
FILE *f
= NULL
;
2789 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2793 r
= manager_open_serialization(m
, &f
);
2798 bus_manager_send_reloading(m
, true);
2806 r
= manager_serialize(m
, f
, fds
, false);
2812 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2817 /* From here on there is no way back. */
2818 manager_clear_jobs_and_units(m
);
2819 lookup_paths_flush_generator(&m
->lookup_paths
);
2820 lookup_paths_free(&m
->lookup_paths
);
2821 dynamic_user_vacuum(m
, false);
2822 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2823 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2825 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2826 if (q
< 0 && r
>= 0)
2829 /* Find new unit paths */
2830 q
= manager_run_generators(m
);
2831 if (q
< 0 && r
>= 0)
2834 lookup_paths_reduce(&m
->lookup_paths
);
2835 manager_build_unit_path_cache(m
);
2837 /* First, enumerate what we can from all config files */
2838 manager_enumerate(m
);
2840 /* Second, deserialize our stored data */
2841 q
= manager_deserialize(m
, f
, fds
);
2842 if (q
< 0 && r
>= 0)
2848 /* Re-register notify_fd as event source */
2849 q
= manager_setup_notify(m
);
2850 if (q
< 0 && r
>= 0)
2853 q
= manager_setup_cgroups_agent(m
);
2854 if (q
< 0 && r
>= 0)
2857 q
= manager_setup_user_lookup_fd(m
);
2858 if (q
< 0 && r
>= 0)
2861 /* Third, fire things up! */
2862 manager_coldplug(m
);
2864 /* Release any dynamic users no longer referenced */
2865 dynamic_user_vacuum(m
, true);
2867 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2868 manager_vacuum_uid_refs(m
);
2869 manager_vacuum_gid_refs(m
);
2871 /* Sync current state of bus names with our set of listening units */
2873 manager_sync_bus_names(m
, m
->api_bus
);
2875 assert(m
->n_reloading
> 0);
2878 m
->send_reloading_done
= true;
2883 void manager_reset_failed(Manager
*m
) {
2889 HASHMAP_FOREACH(u
, m
->units
, i
)
2890 unit_reset_failed(u
);
2893 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2899 /* Returns true if the unit is inactive or going down */
2900 u
= manager_get_unit(m
, name
);
2904 return unit_inactive_or_pending(u
);
2907 static void manager_notify_finished(Manager
*m
) {
2908 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2909 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2914 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2916 /* Note that m->kernel_usec.monotonic is always at 0,
2917 * and m->firmware_usec.monotonic and
2918 * m->loader_usec.monotonic should be considered
2919 * negative values. */
2921 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2922 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2923 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2924 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2926 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2928 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2929 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2931 log_struct(LOG_INFO
,
2932 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2933 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2934 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2935 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2936 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2937 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2938 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2939 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2940 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2943 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2946 log_struct(LOG_INFO
,
2947 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2948 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2949 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2950 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2951 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2952 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2953 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2957 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2958 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2960 log_struct(LOG_INFO
,
2961 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2962 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2963 LOG_MESSAGE("Startup finished in %s.",
2964 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2968 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2972 "STATUS=Startup finished in %s.",
2973 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2976 void manager_check_finished(Manager
*m
) {
2979 if (MANAGER_IS_RELOADING(m
))
2982 /* Verify that we are actually running currently. Initially
2983 * the exit code is set to invalid, and during operation it is
2984 * then set to MANAGER_OK */
2985 if (m
->exit_code
!= MANAGER_OK
)
2988 if (hashmap_size(m
->jobs
) > 0) {
2989 if (m
->jobs_in_progress_event_source
)
2990 /* Ignore any failure, this is only for feedback */
2991 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2996 manager_flip_auto_status(m
, false);
2998 /* Notify Type=idle units that we are done now */
2999 manager_close_idle_pipe(m
);
3001 /* Turn off confirm spawn now */
3002 m
->confirm_spawn
= NULL
;
3004 /* No need to update ask password status when we're going non-interactive */
3005 manager_close_ask_password(m
);
3007 /* This is no longer the first boot */
3008 manager_set_first_boot(m
, false);
3010 if (dual_timestamp_is_set(&m
->finish_timestamp
))
3013 dual_timestamp_get(&m
->finish_timestamp
);
3015 manager_notify_finished(m
);
3017 manager_invalidate_startup_units(m
);
3020 static int manager_run_generators(Manager
*m
) {
3021 _cleanup_strv_free_
char **paths
= NULL
;
3022 const char *argv
[5];
3031 paths
= generator_binary_paths(m
->unit_file_scope
);
3035 /* Optimize by skipping the whole process by not creating output directories
3036 * if no generators are found. */
3037 STRV_FOREACH(path
, paths
) {
3038 if (access(*path
, F_OK
) >= 0)
3040 if (errno
!= ENOENT
)
3041 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3047 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3051 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3052 argv
[1] = m
->lookup_paths
.generator
;
3053 argv
[2] = m
->lookup_paths
.generator_early
;
3054 argv
[3] = m
->lookup_paths
.generator_late
;
3057 RUN_WITH_UMASK(0022)
3058 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, (char**) argv
);
3061 lookup_paths_trim_generator(&m
->lookup_paths
);
3065 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3066 char **a
= NULL
, **b
= NULL
, **l
;
3071 if (!strv_isempty(minus
)) {
3072 a
= strv_env_delete(l
, 1, minus
);
3079 if (!strv_isempty(plus
)) {
3080 b
= strv_env_merge(2, l
, plus
);
3089 if (m
->environment
!= l
)
3090 strv_free(m
->environment
);
3097 manager_clean_environment(m
);
3098 strv_sort(m
->environment
);
3103 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3108 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3109 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3111 if (!default_rlimit
[i
])
3114 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3122 void manager_recheck_journal(Manager
*m
) {
3127 if (!MANAGER_IS_SYSTEM(m
))
3130 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3131 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3132 log_close_journal();
3136 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3137 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3138 log_close_journal();
3142 /* Hmm, OK, so the socket is fully up and the service is up
3143 * too, then let's make use of the thing. */
3147 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3149 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3151 if (!MANAGER_IS_SYSTEM(m
))
3154 if (m
->show_status
!= mode
)
3155 log_debug("%s showing of status.",
3156 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3157 m
->show_status
= mode
;
3160 (void) touch("/run/systemd/show-status");
3162 (void) unlink("/run/systemd/show-status");
3165 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3168 if (!MANAGER_IS_SYSTEM(m
))
3171 if (m
->no_console_output
)
3174 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3177 /* If we cannot find out the status properly, just proceed. */
3178 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3181 if (m
->show_status
> 0)
3187 const char *manager_get_confirm_spawn(Manager
*m
) {
3188 static int last_errno
= 0;
3189 const char *vc
= m
->confirm_spawn
;
3193 /* Here's the deal: we want to test the validity of the console but don't want
3194 * PID1 to go through the whole console process which might block. But we also
3195 * want to warn the user only once if something is wrong with the console so we
3196 * cannot do the sanity checks after spawning our children. So here we simply do
3197 * really basic tests to hopefully trap common errors.
3199 * If the console suddenly disappear at the time our children will really it
3200 * then they will simply fail to acquire it and a positive answer will be
3201 * assumed. New children will fallback to /dev/console though.
3203 * Note: TTYs are devices that can come and go any time, and frequently aren't
3204 * available yet during early boot (consider a USB rs232 dongle...). If for any
3205 * reason the configured console is not ready, we fallback to the default
3208 if (!vc
|| path_equal(vc
, "/dev/console"))
3215 if (!S_ISCHR(st
.st_mode
)) {
3223 if (last_errno
!= errno
) {
3225 log_warning_errno(errno
, "Failed to open %s: %m, using default console", vc
);
3227 return "/dev/console";
3230 void manager_set_first_boot(Manager
*m
, bool b
) {
3233 if (!MANAGER_IS_SYSTEM(m
))
3236 if (m
->first_boot
!= (int) b
) {
3238 (void) touch("/run/systemd/first-boot");
3240 (void) unlink("/run/systemd/first-boot");
3246 void manager_disable_confirm_spawn(void) {
3247 (void) touch("/run/systemd/confirm_spawn_disabled");
3250 bool manager_is_confirm_spawn_disabled(Manager
*m
) {
3251 if (!m
->confirm_spawn
)
3254 return access("/run/systemd/confirm_spawn_disabled", F_OK
) >= 0;
3257 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3260 /* If m is NULL, assume we're after shutdown and let the messages through. */
3262 if (m
&& !manager_get_show_status(m
, type
))
3265 /* XXX We should totally drop the check for ephemeral here
3266 * and thus effectively make 'Type=idle' pointless. */
3267 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3270 va_start(ap
, format
);
3271 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3275 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3276 char p
[strlen(path
)+1];
3282 path_kill_slashes(p
);
3284 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3287 const char *manager_get_runtime_prefix(Manager
*m
) {
3290 return MANAGER_IS_SYSTEM(m
) ?
3292 getenv("XDG_RUNTIME_DIR");
3295 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3300 assert(u
->manager
== m
);
3302 size
= set_size(m
->failed_units
);
3305 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3309 if (set_put(m
->failed_units
, u
) < 0)
3312 (void) set_remove(m
->failed_units
, u
);
3314 if (set_size(m
->failed_units
) != size
)
3315 bus_manager_send_change_signal(m
);
3320 ManagerState
manager_state(Manager
*m
) {
3325 /* Did we ever finish booting? If not then we are still starting up */
3326 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3328 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3329 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3330 return MANAGER_INITIALIZING
;
3332 return MANAGER_STARTING
;
3335 /* Is the special shutdown target queued? If so, we are in shutdown state */
3336 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3337 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3338 return MANAGER_STOPPING
;
3340 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3341 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3342 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3343 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3344 return MANAGER_MAINTENANCE
;
3346 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3347 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3348 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3349 return MANAGER_MAINTENANCE
;
3351 /* Are there any failed units? If so, we are in degraded mode */
3352 if (set_size(m
->failed_units
) > 0)
3353 return MANAGER_DEGRADED
;
3355 return MANAGER_RUNNING
;
3358 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3360 static void manager_unref_uid_internal(
3365 int (*_clean_ipc
)(uid_t uid
)) {
3371 assert(uid_is_valid(uid
));
3374 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3375 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3377 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3378 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3379 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3380 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3382 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3383 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3385 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3388 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3390 n
= c
& ~DESTROY_IPC_FLAG
;
3394 if (destroy_now
&& n
== 0) {
3395 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3397 if (c
& DESTROY_IPC_FLAG
) {
3398 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3399 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3401 (void) _clean_ipc(uid
);
3404 c
= n
| (c
& DESTROY_IPC_FLAG
);
3405 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3409 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3410 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3413 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3414 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3417 static int manager_ref_uid_internal(
3428 assert(uid_is_valid(uid
));
3430 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3431 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3433 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3434 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3436 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3439 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3443 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3445 n
= c
& ~DESTROY_IPC_FLAG
;
3448 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3451 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3453 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3456 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3457 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3460 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3461 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3464 static void manager_vacuum_uid_refs_internal(
3467 int (*_clean_ipc
)(uid_t uid
)) {
3476 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3480 uid
= PTR_TO_UID(k
);
3481 c
= PTR_TO_UINT32(p
);
3483 n
= c
& ~DESTROY_IPC_FLAG
;
3487 if (c
& DESTROY_IPC_FLAG
) {
3488 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3489 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3491 (void) _clean_ipc(uid
);
3494 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3498 void manager_vacuum_uid_refs(Manager
*m
) {
3499 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3502 void manager_vacuum_gid_refs(Manager
*m
) {
3503 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3506 static void manager_serialize_uid_refs_internal(
3510 const char *field_name
) {
3520 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3521 * of it is better rebuild after a reload/reexec. */
3523 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3527 uid
= PTR_TO_UID(k
);
3528 c
= PTR_TO_UINT32(p
);
3530 if (!(c
& DESTROY_IPC_FLAG
))
3533 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3537 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3538 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3541 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3542 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3545 static void manager_deserialize_uid_refs_one_internal(
3548 const char *value
) {
3558 r
= parse_uid(value
, &uid
);
3559 if (r
< 0 || uid
== 0) {
3560 log_debug("Unable to parse UID reference serialization");
3564 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3570 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3571 if (c
& DESTROY_IPC_FLAG
)
3574 c
|= DESTROY_IPC_FLAG
;
3576 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3578 log_debug("Failed to add UID reference entry");
3583 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3584 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3587 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3588 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3591 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3595 char unit_name
[UNIT_NAME_MAX
+1];
3598 Manager
*m
= userdata
;
3606 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3607 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3608 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3610 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3612 if (errno
== EINTR
|| errno
== EAGAIN
)
3615 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3618 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3619 log_warning("Received too short user lookup message, ignoring.");
3623 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3624 log_warning("Received too long user lookup message, ignoring.");
3628 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3629 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3633 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3634 if (memchr(buffer
.unit_name
, 0, n
)) {
3635 log_warning("Received lookup message with embedded NUL character, ignoring.");
3639 buffer
.unit_name
[n
] = 0;
3640 u
= manager_get_unit(m
, buffer
.unit_name
);
3642 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3646 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3648 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3652 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3653 [MANAGER_INITIALIZING
] = "initializing",
3654 [MANAGER_STARTING
] = "starting",
3655 [MANAGER_RUNNING
] = "running",
3656 [MANAGER_DEGRADED
] = "degraded",
3657 [MANAGER_MAINTENANCE
] = "maintenance",
3658 [MANAGER_STOPPING
] = "stopping",
3661 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);