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 if (m
->jobs_in_progress_event_source
)
117 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
;
118 r
= sd_event_add_time(
120 &m
->jobs_in_progress_event_source
,
123 manager_dispatch_jobs_in_progress
, m
);
127 (void) sd_event_source_set_description(m
->jobs_in_progress_event_source
, "manager-jobs-in-progress");
130 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
132 static void draw_cylon(char buffer
[], size_t buflen
, unsigned width
, unsigned pos
) {
135 assert(buflen
>= CYLON_BUFFER_EXTRA
+ width
+ 1);
136 assert(pos
<= width
+1); /* 0 or width+1 mean that the center light is behind the corner */
140 p
= mempset(p
, ' ', pos
-2);
141 if (log_get_show_color())
142 p
= stpcpy(p
, ANSI_RED
);
146 if (pos
> 0 && pos
<= width
) {
147 if (log_get_show_color())
148 p
= stpcpy(p
, ANSI_HIGHLIGHT_RED
);
152 if (log_get_show_color())
153 p
= stpcpy(p
, ANSI_NORMAL
);
156 if (log_get_show_color())
157 p
= stpcpy(p
, ANSI_RED
);
160 p
= mempset(p
, ' ', width
-1-pos
);
161 if (log_get_show_color())
162 strcpy(p
, ANSI_NORMAL
);
166 void manager_flip_auto_status(Manager
*m
, bool enable
) {
170 if (m
->show_status
== SHOW_STATUS_AUTO
)
171 manager_set_show_status(m
, SHOW_STATUS_TEMPORARY
);
173 if (m
->show_status
== SHOW_STATUS_TEMPORARY
)
174 manager_set_show_status(m
, SHOW_STATUS_AUTO
);
178 static void manager_print_jobs_in_progress(Manager
*m
) {
179 _cleanup_free_
char *job_of_n
= NULL
;
182 unsigned counter
= 0, print_nr
;
183 char cylon
[6 + CYLON_BUFFER_EXTRA
+ 1];
185 char time
[FORMAT_TIMESPAN_MAX
], limit
[FORMAT_TIMESPAN_MAX
] = "no limit";
189 assert(m
->n_running_jobs
> 0);
191 manager_flip_auto_status(m
, true);
193 print_nr
= (m
->jobs_in_progress_iteration
/ JOBS_IN_PROGRESS_PERIOD_DIVISOR
) % m
->n_running_jobs
;
195 HASHMAP_FOREACH(j
, m
->jobs
, i
)
196 if (j
->state
== JOB_RUNNING
&& counter
++ == print_nr
)
199 /* m->n_running_jobs must be consistent with the contents of m->jobs,
200 * so the above loop must have succeeded in finding j. */
201 assert(counter
== print_nr
+ 1);
204 cylon_pos
= m
->jobs_in_progress_iteration
% 14;
206 cylon_pos
= 14 - cylon_pos
;
207 draw_cylon(cylon
, sizeof(cylon
), 6, cylon_pos
);
209 m
->jobs_in_progress_iteration
++;
211 if (m
->n_running_jobs
> 1) {
212 if (asprintf(&job_of_n
, "(%u of %u) ", counter
, m
->n_running_jobs
) < 0)
216 format_timespan(time
, sizeof(time
), now(CLOCK_MONOTONIC
) - j
->begin_usec
, 1*USEC_PER_SEC
);
217 if (job_get_timeout(j
, &x
) > 0)
218 format_timespan(limit
, sizeof(limit
), x
- j
->begin_usec
, 1*USEC_PER_SEC
);
220 manager_status_printf(m
, STATUS_TYPE_EPHEMERAL
, cylon
,
221 "%sA %s job is running for %s (%s / %s)",
223 job_type_to_string(j
->type
),
224 unit_description(j
->unit
),
228 static int have_ask_password(void) {
229 _cleanup_closedir_
DIR *dir
;
231 dir
= opendir("/run/systemd/ask-password");
244 if (!de
&& errno
> 0)
249 if (startswith(de
->d_name
, "ask."))
254 static int manager_dispatch_ask_password_fd(sd_event_source
*source
,
255 int fd
, uint32_t revents
, void *userdata
) {
256 Manager
*m
= userdata
;
262 m
->have_ask_password
= have_ask_password();
263 if (m
->have_ask_password
< 0)
264 /* Log error but continue. Negative have_ask_password
265 * is treated as unknown status. */
266 log_error_errno(m
->have_ask_password
, "Failed to list /run/systemd/ask-password: %m");
271 static void manager_close_ask_password(Manager
*m
) {
274 m
->ask_password_event_source
= sd_event_source_unref(m
->ask_password_event_source
);
275 m
->ask_password_inotify_fd
= safe_close(m
->ask_password_inotify_fd
);
276 m
->have_ask_password
= -EINVAL
;
279 static int manager_check_ask_password(Manager
*m
) {
284 if (!m
->ask_password_event_source
) {
285 assert(m
->ask_password_inotify_fd
< 0);
287 mkdir_p_label("/run/systemd/ask-password", 0755);
289 m
->ask_password_inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
290 if (m
->ask_password_inotify_fd
< 0)
291 return log_error_errno(errno
, "inotify_init1() failed: %m");
293 if (inotify_add_watch(m
->ask_password_inotify_fd
, "/run/systemd/ask-password", IN_CREATE
|IN_DELETE
|IN_MOVE
) < 0) {
294 log_error_errno(errno
, "Failed to add watch on /run/systemd/ask-password: %m");
295 manager_close_ask_password(m
);
299 r
= sd_event_add_io(m
->event
, &m
->ask_password_event_source
,
300 m
->ask_password_inotify_fd
, EPOLLIN
,
301 manager_dispatch_ask_password_fd
, m
);
303 log_error_errno(errno
, "Failed to add event source for /run/systemd/ask-password: %m");
304 manager_close_ask_password(m
);
308 (void) sd_event_source_set_description(m
->ask_password_event_source
, "manager-ask-password");
310 /* Queries might have been added meanwhile... */
311 manager_dispatch_ask_password_fd(m
->ask_password_event_source
,
312 m
->ask_password_inotify_fd
, EPOLLIN
, m
);
315 return m
->have_ask_password
;
318 static int manager_watch_idle_pipe(Manager
*m
) {
323 if (m
->idle_pipe_event_source
)
326 if (m
->idle_pipe
[2] < 0)
329 r
= sd_event_add_io(m
->event
, &m
->idle_pipe_event_source
, m
->idle_pipe
[2], EPOLLIN
, manager_dispatch_idle_pipe_fd
, m
);
331 return log_error_errno(r
, "Failed to watch idle pipe: %m");
333 (void) sd_event_source_set_description(m
->idle_pipe_event_source
, "manager-idle-pipe");
338 static void manager_close_idle_pipe(Manager
*m
) {
341 m
->idle_pipe_event_source
= sd_event_source_unref(m
->idle_pipe_event_source
);
343 safe_close_pair(m
->idle_pipe
);
344 safe_close_pair(m
->idle_pipe
+ 2);
347 static int manager_setup_time_change(Manager
*m
) {
350 /* We only care for the cancellation event, hence we set the
351 * timeout to the latest possible value. */
352 struct itimerspec its
= {
353 .it_value
.tv_sec
= TIME_T_MAX
,
357 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX
));
362 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
363 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
365 m
->time_change_fd
= timerfd_create(CLOCK_REALTIME
, TFD_NONBLOCK
|TFD_CLOEXEC
);
366 if (m
->time_change_fd
< 0)
367 return log_error_errno(errno
, "Failed to create timerfd: %m");
369 if (timerfd_settime(m
->time_change_fd
, TFD_TIMER_ABSTIME
|TFD_TIMER_CANCEL_ON_SET
, &its
, NULL
) < 0) {
370 log_debug_errno(errno
, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
371 m
->time_change_fd
= safe_close(m
->time_change_fd
);
375 r
= sd_event_add_io(m
->event
, &m
->time_change_event_source
, m
->time_change_fd
, EPOLLIN
, manager_dispatch_time_change_fd
, m
);
377 return log_error_errno(r
, "Failed to create time change event source: %m");
379 (void) sd_event_source_set_description(m
->time_change_event_source
, "manager-time-change");
381 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
386 static int enable_special_signals(Manager
*m
) {
387 _cleanup_close_
int fd
= -1;
394 /* Enable that we get SIGINT on control-alt-del. In containers
395 * this will fail with EPERM (older) or EINVAL (newer), so
397 if (reboot(RB_DISABLE_CAD
) < 0 && errno
!= EPERM
&& errno
!= EINVAL
)
398 log_warning_errno(errno
, "Failed to enable ctrl-alt-del handling: %m");
400 fd
= open_terminal("/dev/tty0", O_RDWR
|O_NOCTTY
|O_CLOEXEC
);
402 /* Support systems without virtual console */
404 log_warning_errno(errno
, "Failed to open /dev/tty0: %m");
406 /* Enable that we get SIGWINCH on kbrequest */
407 if (ioctl(fd
, KDSIGACCEPT
, SIGWINCH
) < 0)
408 log_warning_errno(errno
, "Failed to enable kbrequest handling: %m");
414 static int manager_setup_signals(Manager
*m
) {
415 struct sigaction sa
= {
416 .sa_handler
= SIG_DFL
,
417 .sa_flags
= SA_NOCLDSTOP
|SA_RESTART
,
424 assert_se(sigaction(SIGCHLD
, &sa
, NULL
) == 0);
426 /* We make liberal use of realtime signals here. On
427 * Linux/glibc we have 30 of them (with the exception of Linux
428 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
431 assert_se(sigemptyset(&mask
) == 0);
432 sigset_add_many(&mask
,
433 SIGCHLD
, /* Child died */
434 SIGTERM
, /* Reexecute daemon */
435 SIGHUP
, /* Reload configuration */
436 SIGUSR1
, /* systemd/upstart: reconnect to D-Bus */
437 SIGUSR2
, /* systemd: dump status */
438 SIGINT
, /* Kernel sends us this on control-alt-del */
439 SIGWINCH
, /* Kernel sends us this on kbrequest (alt-arrowup) */
440 SIGPWR
, /* Some kernel drivers and upsd send us this on power failure */
442 SIGRTMIN
+0, /* systemd: start default.target */
443 SIGRTMIN
+1, /* systemd: isolate rescue.target */
444 SIGRTMIN
+2, /* systemd: isolate emergency.target */
445 SIGRTMIN
+3, /* systemd: start halt.target */
446 SIGRTMIN
+4, /* systemd: start poweroff.target */
447 SIGRTMIN
+5, /* systemd: start reboot.target */
448 SIGRTMIN
+6, /* systemd: start kexec.target */
450 /* ... space for more special targets ... */
452 SIGRTMIN
+13, /* systemd: Immediate halt */
453 SIGRTMIN
+14, /* systemd: Immediate poweroff */
454 SIGRTMIN
+15, /* systemd: Immediate reboot */
455 SIGRTMIN
+16, /* systemd: Immediate kexec */
457 /* ... space for more immediate system state changes ... */
459 SIGRTMIN
+20, /* systemd: enable status messages */
460 SIGRTMIN
+21, /* systemd: disable status messages */
461 SIGRTMIN
+22, /* systemd: set log level to LOG_DEBUG */
462 SIGRTMIN
+23, /* systemd: set log level to LOG_INFO */
463 SIGRTMIN
+24, /* systemd: Immediate exit (--user only) */
465 /* .. one free signal here ... */
467 #if !defined(__hppa64__) && !defined(__hppa__)
468 /* Apparently Linux on hppa has fewer RT
469 * signals (SIGRTMAX is SIGRTMIN+25 there),
470 * hence let's not try to make use of them
471 * here. Since these commands are accessible
472 * by different means and only really a safety
473 * net, the missing functionality on hppa
474 * shouldn't matter. */
476 SIGRTMIN
+26, /* systemd: set log target to journal-or-kmsg */
477 SIGRTMIN
+27, /* systemd: set log target to console */
478 SIGRTMIN
+28, /* systemd: set log target to kmsg */
479 SIGRTMIN
+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
481 /* ... one free signal here SIGRTMIN+30 ... */
484 assert_se(sigprocmask(SIG_SETMASK
, &mask
, NULL
) == 0);
486 m
->signal_fd
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
487 if (m
->signal_fd
< 0)
490 r
= sd_event_add_io(m
->event
, &m
->signal_event_source
, m
->signal_fd
, EPOLLIN
, manager_dispatch_signal_fd
, m
);
494 (void) sd_event_source_set_description(m
->signal_event_source
, "manager-signal");
496 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
497 * notify processing can still figure out to which process/service a message belongs, before we reap the
498 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
499 * status information before detecting that there's no process in a cgroup. */
500 r
= sd_event_source_set_priority(m
->signal_event_source
, SD_EVENT_PRIORITY_NORMAL
-6);
504 if (MANAGER_IS_SYSTEM(m
))
505 return enable_special_signals(m
);
510 static void manager_clean_environment(Manager
*m
) {
513 /* Let's remove some environment variables that we
514 * need ourselves to communicate with our clients */
528 static int manager_default_environment(Manager
*m
) {
531 if (MANAGER_IS_SYSTEM(m
)) {
532 /* The system manager always starts with a clean
533 * environment for its children. It does not import
534 * the kernel or the parents exported variables.
536 * The initial passed environ is untouched to keep
537 * /proc/self/environ valid; it is used for tagging
538 * the init process inside containers. */
539 m
->environment
= strv_new("PATH=" DEFAULT_PATH
,
542 /* Import locale variables LC_*= from configuration */
543 locale_setup(&m
->environment
);
545 /* The user manager passes its own environment
546 * along to its children. */
547 m
->environment
= strv_copy(environ
);
553 manager_clean_environment(m
);
554 strv_sort(m
->environment
);
559 int manager_new(UnitFileScope scope
, bool test_run
, Manager
**_m
) {
564 assert(IN_SET(scope
, UNIT_FILE_SYSTEM
, UNIT_FILE_USER
));
566 m
= new0(Manager
, 1);
570 m
->unit_file_scope
= scope
;
571 m
->exit_code
= _MANAGER_EXIT_CODE_INVALID
;
572 m
->default_timer_accuracy_usec
= USEC_PER_MINUTE
;
573 m
->default_tasks_accounting
= true;
574 m
->default_tasks_max
= UINT64_MAX
;
577 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0)
578 boot_timestamps(&m
->userspace_timestamp
, &m
->firmware_timestamp
, &m
->loader_timestamp
);
581 /* Prepare log fields we can use for structured logging */
582 if (MANAGER_IS_SYSTEM(m
)) {
583 m
->unit_log_field
= "UNIT=";
584 m
->unit_log_format_string
= "UNIT=%s";
586 m
->unit_log_field
= "USER_UNIT=";
587 m
->unit_log_format_string
= "USER_UNIT=%s";
590 m
->idle_pipe
[0] = m
->idle_pipe
[1] = m
->idle_pipe
[2] = m
->idle_pipe
[3] = -1;
592 m
->pin_cgroupfs_fd
= m
->notify_fd
= m
->cgroups_agent_fd
= m
->signal_fd
= m
->time_change_fd
=
593 m
->dev_autofs_fd
= m
->private_listen_fd
= m
->kdbus_fd
= m
->cgroup_inotify_fd
=
594 m
->ask_password_inotify_fd
= -1;
596 m
->user_lookup_fds
[0] = m
->user_lookup_fds
[1] = -1;
598 m
->current_job_id
= 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
600 m
->have_ask_password
= -EINVAL
; /* we don't know */
603 m
->test_run
= test_run
;
605 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
606 RATELIMIT_INIT(m
->ctrl_alt_del_ratelimit
, 2 * USEC_PER_SEC
, 7);
608 r
= manager_default_environment(m
);
612 r
= hashmap_ensure_allocated(&m
->units
, &string_hash_ops
);
616 r
= hashmap_ensure_allocated(&m
->jobs
, NULL
);
620 r
= hashmap_ensure_allocated(&m
->cgroup_unit
, &string_hash_ops
);
624 r
= hashmap_ensure_allocated(&m
->watch_bus
, &string_hash_ops
);
628 r
= sd_event_default(&m
->event
);
632 r
= sd_event_add_defer(m
->event
, &m
->run_queue_event_source
, manager_dispatch_run_queue
, m
);
636 r
= sd_event_source_set_priority(m
->run_queue_event_source
, SD_EVENT_PRIORITY_IDLE
);
640 r
= sd_event_source_set_enabled(m
->run_queue_event_source
, SD_EVENT_OFF
);
644 (void) sd_event_source_set_description(m
->run_queue_event_source
, "manager-run-queue");
646 r
= manager_setup_signals(m
);
650 r
= manager_setup_cgroup(m
);
654 r
= manager_setup_time_change(m
);
658 m
->udev
= udev_new();
664 /* Note that we set up neither kdbus, nor the notify fd
665 * here. We do that after deserialization, since they might
666 * have gotten serialized across the reexec. */
668 m
->taint_usr
= dir_is_empty("/usr") > 0;
678 static int manager_setup_notify(Manager
*m
) {
684 if (m
->notify_fd
< 0) {
685 _cleanup_close_
int fd
= -1;
686 union sockaddr_union sa
= {
687 .sa
.sa_family
= AF_UNIX
,
689 static const int one
= 1;
692 /* First free all secondary fields */
693 m
->notify_socket
= mfree(m
->notify_socket
);
694 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
696 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
698 return log_error_errno(errno
, "Failed to allocate notification socket: %m");
700 fd_inc_rcvbuf(fd
, NOTIFY_RCVBUF_SIZE
);
702 e
= manager_get_runtime_prefix(m
);
704 log_error("Failed to determine runtime prefix.");
708 m
->notify_socket
= strappend(e
, "/systemd/notify");
709 if (!m
->notify_socket
)
712 (void) mkdir_parents_label(m
->notify_socket
, 0755);
713 (void) unlink(m
->notify_socket
);
715 strncpy(sa
.un
.sun_path
, m
->notify_socket
, sizeof(sa
.un
.sun_path
)-1);
716 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
718 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
720 r
= setsockopt(fd
, SOL_SOCKET
, SO_PASSCRED
, &one
, sizeof(one
));
722 return log_error_errno(errno
, "SO_PASSCRED failed: %m");
727 log_debug("Using notification socket %s", m
->notify_socket
);
730 if (!m
->notify_event_source
) {
731 r
= sd_event_add_io(m
->event
, &m
->notify_event_source
, m
->notify_fd
, EPOLLIN
, manager_dispatch_notify_fd
, m
);
733 return log_error_errno(r
, "Failed to allocate notify event source: %m");
735 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
736 * service an exit message belongs. */
737 r
= sd_event_source_set_priority(m
->notify_event_source
, SD_EVENT_PRIORITY_NORMAL
-7);
739 return log_error_errno(r
, "Failed to set priority of notify event source: %m");
741 (void) sd_event_source_set_description(m
->notify_event_source
, "manager-notify");
747 static int manager_setup_cgroups_agent(Manager
*m
) {
749 static const union sockaddr_union sa
= {
750 .un
.sun_family
= AF_UNIX
,
751 .un
.sun_path
= "/run/systemd/cgroups-agent",
755 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
756 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
757 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
758 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
759 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
760 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
761 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
762 * we thus won't lose messages.
764 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
765 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
766 * bus for these messages. */
771 if (!MANAGER_IS_SYSTEM(m
))
774 if (cg_unified() > 0) /* We don't need this anymore on the unified hierarchy */
777 if (m
->cgroups_agent_fd
< 0) {
778 _cleanup_close_
int fd
= -1;
780 /* First free all secondary fields */
781 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
783 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
785 return log_error_errno(errno
, "Failed to allocate cgroups agent socket: %m");
787 fd_inc_rcvbuf(fd
, CGROUPS_AGENT_RCVBUF_SIZE
);
789 (void) unlink(sa
.un
.sun_path
);
791 /* Only allow root to connect to this socket */
793 r
= bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
));
795 return log_error_errno(errno
, "bind(%s) failed: %m", sa
.un
.sun_path
);
797 m
->cgroups_agent_fd
= fd
;
801 if (!m
->cgroups_agent_event_source
) {
802 r
= sd_event_add_io(m
->event
, &m
->cgroups_agent_event_source
, m
->cgroups_agent_fd
, EPOLLIN
, manager_dispatch_cgroups_agent_fd
, m
);
804 return log_error_errno(r
, "Failed to allocate cgroups agent event source: %m");
806 /* Process cgroups notifications early, but after having processed service notification messages or
807 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
808 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
809 * cgroup inotify for the unified cgroup stuff. */
810 r
= sd_event_source_set_priority(m
->cgroups_agent_event_source
, SD_EVENT_PRIORITY_NORMAL
-5);
812 return log_error_errno(r
, "Failed to set priority of cgroups agent event source: %m");
814 (void) sd_event_source_set_description(m
->cgroups_agent_event_source
, "manager-cgroups-agent");
820 static int manager_setup_user_lookup_fd(Manager
*m
) {
825 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
826 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
827 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
828 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
829 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
830 * hence we establish this communication channel so that forked off processes can pass their UID/GID
831 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
832 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
835 * You might wonder why we need a communication channel for this that is independent of the usual notification
836 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
837 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
838 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
840 * Note that this function is called under two circumstances: when we first initialize (in which case we
841 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
842 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
844 if (m
->user_lookup_fds
[0] < 0) {
846 /* Free all secondary fields */
847 safe_close_pair(m
->user_lookup_fds
);
848 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
850 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, m
->user_lookup_fds
) < 0)
851 return log_error_errno(errno
, "Failed to allocate user lookup socket: %m");
853 (void) fd_inc_rcvbuf(m
->user_lookup_fds
[0], NOTIFY_RCVBUF_SIZE
);
856 if (!m
->user_lookup_event_source
) {
857 r
= sd_event_add_io(m
->event
, &m
->user_lookup_event_source
, m
->user_lookup_fds
[0], EPOLLIN
, manager_dispatch_user_lookup_fd
, m
);
859 return log_error_errno(errno
, "Failed to allocate user lookup event source: %m");
861 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
863 r
= sd_event_source_set_priority(m
->user_lookup_event_source
, SD_EVENT_PRIORITY_NORMAL
-8);
865 return log_error_errno(errno
, "Failed to set priority ot user lookup event source: %m");
867 (void) sd_event_source_set_description(m
->user_lookup_event_source
, "user-lookup");
873 static int manager_connect_bus(Manager
*m
, bool reexecuting
) {
874 bool try_bus_connect
;
884 (MANAGER_IS_USER(m
) && getenv("DBUS_SESSION_BUS_ADDRESS"));
886 /* Try to connect to the buses, if possible. */
887 return bus_init(m
, try_bus_connect
);
890 static unsigned manager_dispatch_cleanup_queue(Manager
*m
) {
896 while ((u
= m
->cleanup_queue
)) {
897 assert(u
->in_cleanup_queue
);
907 GC_OFFSET_IN_PATH
, /* This one is on the path we were traveling */
908 GC_OFFSET_UNSURE
, /* No clue */
909 GC_OFFSET_GOOD
, /* We still need this unit */
910 GC_OFFSET_BAD
, /* We don't need this unit anymore */
914 static void unit_gc_mark_good(Unit
*u
, unsigned gc_marker
) {
918 u
->gc_marker
= gc_marker
+ GC_OFFSET_GOOD
;
920 /* Recursively mark referenced units as GOOD as well */
921 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCES
], i
)
922 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
)
923 unit_gc_mark_good(other
, gc_marker
);
926 static void unit_gc_sweep(Unit
*u
, unsigned gc_marker
) {
933 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
||
934 u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
935 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
||
936 u
->gc_marker
== gc_marker
+ GC_OFFSET_IN_PATH
)
939 if (u
->in_cleanup_queue
)
942 if (unit_check_gc(u
))
945 u
->gc_marker
= gc_marker
+ GC_OFFSET_IN_PATH
;
949 SET_FOREACH(other
, u
->dependencies
[UNIT_REFERENCED_BY
], i
) {
950 unit_gc_sweep(other
, gc_marker
);
952 if (other
->gc_marker
== gc_marker
+ GC_OFFSET_GOOD
)
955 if (other
->gc_marker
!= gc_marker
+ GC_OFFSET_BAD
)
962 /* We were unable to find anything out about this entry, so
963 * let's investigate it later */
964 u
->gc_marker
= gc_marker
+ GC_OFFSET_UNSURE
;
965 unit_add_to_gc_queue(u
);
969 /* We definitely know that this one is not useful anymore, so
970 * let's mark it for deletion */
971 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
972 unit_add_to_cleanup_queue(u
);
976 unit_gc_mark_good(u
, gc_marker
);
979 static unsigned manager_dispatch_gc_queue(Manager
*m
) {
986 /* log_debug("Running GC..."); */
988 m
->gc_marker
+= _GC_OFFSET_MAX
;
989 if (m
->gc_marker
+ _GC_OFFSET_MAX
<= _GC_OFFSET_MAX
)
992 gc_marker
= m
->gc_marker
;
994 while ((u
= m
->gc_queue
)) {
995 assert(u
->in_gc_queue
);
997 unit_gc_sweep(u
, gc_marker
);
999 LIST_REMOVE(gc_queue
, m
->gc_queue
, u
);
1000 u
->in_gc_queue
= false;
1004 if (u
->gc_marker
== gc_marker
+ GC_OFFSET_BAD
||
1005 u
->gc_marker
== gc_marker
+ GC_OFFSET_UNSURE
) {
1007 log_unit_debug(u
, "Collecting.");
1008 u
->gc_marker
= gc_marker
+ GC_OFFSET_BAD
;
1009 unit_add_to_cleanup_queue(u
);
1013 m
->n_in_gc_queue
= 0;
1018 static void manager_clear_jobs_and_units(Manager
*m
) {
1023 while ((u
= hashmap_first(m
->units
)))
1026 manager_dispatch_cleanup_queue(m
);
1028 assert(!m
->load_queue
);
1029 assert(!m
->run_queue
);
1030 assert(!m
->dbus_unit_queue
);
1031 assert(!m
->dbus_job_queue
);
1032 assert(!m
->cleanup_queue
);
1033 assert(!m
->gc_queue
);
1035 assert(hashmap_isempty(m
->jobs
));
1036 assert(hashmap_isempty(m
->units
));
1038 m
->n_on_console
= 0;
1039 m
->n_running_jobs
= 0;
1042 Manager
* manager_free(Manager
*m
) {
1049 manager_clear_jobs_and_units(m
);
1051 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++)
1052 if (unit_vtable
[c
]->shutdown
)
1053 unit_vtable
[c
]->shutdown(m
);
1055 /* If we reexecute ourselves, we keep the root cgroup
1057 manager_shutdown_cgroup(m
, m
->exit_code
!= MANAGER_REEXECUTE
);
1059 lookup_paths_flush_generator(&m
->lookup_paths
);
1063 dynamic_user_vacuum(m
, false);
1064 hashmap_free(m
->dynamic_users
);
1066 hashmap_free(m
->units
);
1067 hashmap_free(m
->jobs
);
1068 hashmap_free(m
->watch_pids1
);
1069 hashmap_free(m
->watch_pids2
);
1070 hashmap_free(m
->watch_bus
);
1072 set_free(m
->startup_units
);
1073 set_free(m
->failed_units
);
1075 sd_event_source_unref(m
->signal_event_source
);
1076 sd_event_source_unref(m
->notify_event_source
);
1077 sd_event_source_unref(m
->cgroups_agent_event_source
);
1078 sd_event_source_unref(m
->time_change_event_source
);
1079 sd_event_source_unref(m
->jobs_in_progress_event_source
);
1080 sd_event_source_unref(m
->run_queue_event_source
);
1081 sd_event_source_unref(m
->user_lookup_event_source
);
1083 safe_close(m
->signal_fd
);
1084 safe_close(m
->notify_fd
);
1085 safe_close(m
->cgroups_agent_fd
);
1086 safe_close(m
->time_change_fd
);
1087 safe_close(m
->kdbus_fd
);
1088 safe_close_pair(m
->user_lookup_fds
);
1090 manager_close_ask_password(m
);
1092 manager_close_idle_pipe(m
);
1094 udev_unref(m
->udev
);
1095 sd_event_unref(m
->event
);
1097 free(m
->notify_socket
);
1099 lookup_paths_free(&m
->lookup_paths
);
1100 strv_free(m
->environment
);
1102 hashmap_free(m
->cgroup_unit
);
1103 set_free_free(m
->unit_path_cache
);
1105 free(m
->switch_root
);
1106 free(m
->switch_root_init
);
1108 for (i
= 0; i
< _RLIMIT_MAX
; i
++)
1109 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
1111 assert(hashmap_isempty(m
->units_requiring_mounts_for
));
1112 hashmap_free(m
->units_requiring_mounts_for
);
1114 hashmap_free(m
->uid_refs
);
1115 hashmap_free(m
->gid_refs
);
1121 void manager_enumerate(Manager
*m
) {
1126 /* Let's ask every type to load all units from disk/kernel
1127 * that it might know */
1128 for (c
= 0; c
< _UNIT_TYPE_MAX
; c
++) {
1129 if (!unit_type_supported(c
)) {
1130 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c
));
1134 if (!unit_vtable
[c
]->enumerate
)
1137 unit_vtable
[c
]->enumerate(m
);
1140 manager_dispatch_load_queue(m
);
1143 static void manager_coldplug(Manager
*m
) {
1151 /* Then, let's set up their initial state. */
1152 HASHMAP_FOREACH_KEY(u
, k
, m
->units
, i
) {
1154 /* ignore aliases */
1158 r
= unit_coldplug(u
);
1160 log_warning_errno(r
, "We couldn't coldplug %s, proceeding anyway: %m", u
->id
);
1164 static void manager_build_unit_path_cache(Manager
*m
) {
1170 set_free_free(m
->unit_path_cache
);
1172 m
->unit_path_cache
= set_new(&string_hash_ops
);
1173 if (!m
->unit_path_cache
) {
1178 /* This simply builds a list of files we know exist, so that
1179 * we don't always have to go to disk */
1181 STRV_FOREACH(i
, m
->lookup_paths
.search_path
) {
1182 _cleanup_closedir_
DIR *d
= NULL
;
1187 if (errno
!= ENOENT
)
1188 log_warning_errno(errno
, "Failed to open directory %s, ignoring: %m", *i
);
1192 FOREACH_DIRENT(de
, d
, r
= -errno
; goto fail
) {
1195 p
= strjoin(streq(*i
, "/") ? "" : *i
, "/", de
->d_name
, NULL
);
1201 r
= set_consume(m
->unit_path_cache
, p
);
1210 log_warning_errno(r
, "Failed to build unit path cache, proceeding without: %m");
1211 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
1214 static void manager_distribute_fds(Manager
*m
, FDSet
*fds
) {
1220 HASHMAP_FOREACH(u
, m
->units
, i
) {
1222 if (fdset_size(fds
) <= 0)
1225 if (!UNIT_VTABLE(u
)->distribute_fds
)
1228 UNIT_VTABLE(u
)->distribute_fds(u
, fds
);
1232 int manager_startup(Manager
*m
, FILE *serialization
, FDSet
*fds
) {
1237 r
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
1241 /* Make sure the transient directory always exists, so that it remains in the search path */
1242 r
= mkdir_p_label(m
->lookup_paths
.transient
, 0755);
1246 dual_timestamp_get(&m
->generators_start_timestamp
);
1247 r
= manager_run_generators(m
);
1248 dual_timestamp_get(&m
->generators_finish_timestamp
);
1252 lookup_paths_reduce(&m
->lookup_paths
);
1253 manager_build_unit_path_cache(m
);
1255 /* If we will deserialize make sure that during enumeration
1256 * this is already known, so we increase the counter here
1261 /* First, enumerate what we can from all config files */
1262 dual_timestamp_get(&m
->units_load_start_timestamp
);
1263 manager_enumerate(m
);
1264 dual_timestamp_get(&m
->units_load_finish_timestamp
);
1266 /* Second, deserialize if there is something to deserialize */
1268 r
= manager_deserialize(m
, serialization
, fds
);
1270 /* Any fds left? Find some unit which wants them. This is
1271 * useful to allow container managers to pass some file
1272 * descriptors to us pre-initialized. This enables
1273 * socket-based activation of entire containers. */
1274 manager_distribute_fds(m
, fds
);
1276 /* We might have deserialized the notify fd, but if we didn't
1277 * then let's create the bus now */
1278 q
= manager_setup_notify(m
);
1279 if (q
< 0 && r
== 0)
1282 q
= manager_setup_cgroups_agent(m
);
1283 if (q
< 0 && r
== 0)
1286 q
= manager_setup_user_lookup_fd(m
);
1287 if (q
< 0 && r
== 0)
1290 /* We might have deserialized the kdbus control fd, but if we
1291 * didn't, then let's create the bus now. */
1292 manager_connect_bus(m
, !!serialization
);
1293 bus_track_coldplug(m
, &m
->subscribed
, &m
->deserialized_subscribed
);
1295 /* Third, fire things up! */
1296 manager_coldplug(m
);
1298 /* Release any dynamic users no longer referenced */
1299 dynamic_user_vacuum(m
, true);
1301 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1302 manager_vacuum_uid_refs(m
);
1303 manager_vacuum_gid_refs(m
);
1305 if (serialization
) {
1306 assert(m
->n_reloading
> 0);
1309 /* Let's wait for the UnitNew/JobNew messages being
1310 * sent, before we notify that the reload is
1312 m
->send_reloading_done
= true;
1318 int manager_add_job(Manager
*m
, JobType type
, Unit
*unit
, JobMode mode
, sd_bus_error
*e
, Job
**_ret
) {
1323 assert(type
< _JOB_TYPE_MAX
);
1325 assert(mode
< _JOB_MODE_MAX
);
1327 if (mode
== JOB_ISOLATE
&& type
!= JOB_START
)
1328 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Isolate is only valid for start.");
1330 if (mode
== JOB_ISOLATE
&& !unit
->allow_isolate
)
1331 return sd_bus_error_setf(e
, BUS_ERROR_NO_ISOLATION
, "Operation refused, unit may not be isolated.");
1333 log_unit_debug(unit
, "Trying to enqueue job %s/%s/%s", unit
->id
, job_type_to_string(type
), job_mode_to_string(mode
));
1335 type
= job_type_collapse(type
, unit
);
1337 tr
= transaction_new(mode
== JOB_REPLACE_IRREVERSIBLY
);
1341 r
= transaction_add_job_and_dependencies(tr
, type
, unit
, NULL
, true, false,
1342 mode
== JOB_IGNORE_DEPENDENCIES
|| mode
== JOB_IGNORE_REQUIREMENTS
,
1343 mode
== JOB_IGNORE_DEPENDENCIES
, e
);
1347 if (mode
== JOB_ISOLATE
) {
1348 r
= transaction_add_isolate_jobs(tr
, m
);
1353 r
= transaction_activate(tr
, m
, mode
, e
);
1357 log_unit_debug(unit
,
1358 "Enqueued job %s/%s as %u", unit
->id
,
1359 job_type_to_string(type
), (unsigned) tr
->anchor_job
->id
);
1362 *_ret
= tr
->anchor_job
;
1364 transaction_free(tr
);
1368 transaction_abort(tr
);
1369 transaction_free(tr
);
1373 int manager_add_job_by_name(Manager
*m
, JobType type
, const char *name
, JobMode mode
, sd_bus_error
*e
, Job
**ret
) {
1378 assert(type
< _JOB_TYPE_MAX
);
1380 assert(mode
< _JOB_MODE_MAX
);
1382 r
= manager_load_unit(m
, name
, NULL
, NULL
, &unit
);
1386 return manager_add_job(m
, type
, unit
, mode
, e
, ret
);
1389 int manager_add_job_by_name_and_warn(Manager
*m
, JobType type
, const char *name
, JobMode mode
, Job
**ret
) {
1390 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1394 assert(type
< _JOB_TYPE_MAX
);
1396 assert(mode
< _JOB_MODE_MAX
);
1398 r
= manager_add_job_by_name(m
, type
, name
, mode
, &error
, ret
);
1400 return log_warning_errno(r
, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode
), name
, bus_error_message(&error
, r
));
1405 Job
*manager_get_job(Manager
*m
, uint32_t id
) {
1408 return hashmap_get(m
->jobs
, UINT32_TO_PTR(id
));
1411 Unit
*manager_get_unit(Manager
*m
, const char *name
) {
1415 return hashmap_get(m
->units
, name
);
1418 unsigned manager_dispatch_load_queue(Manager
*m
) {
1424 /* Make sure we are not run recursively */
1425 if (m
->dispatching_load_queue
)
1428 m
->dispatching_load_queue
= true;
1430 /* Dispatches the load queue. Takes a unit from the queue and
1431 * tries to load its data until the queue is empty */
1433 while ((u
= m
->load_queue
)) {
1434 assert(u
->in_load_queue
);
1440 m
->dispatching_load_queue
= false;
1444 int manager_load_unit_prepare(
1456 assert(name
|| path
);
1458 /* This will prepare the unit for loading, but not actually
1459 * load anything from disk. */
1461 if (path
&& !is_path(path
))
1462 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Path %s is not absolute.", path
);
1465 name
= basename(path
);
1467 t
= unit_name_to_type(name
);
1469 if (t
== _UNIT_TYPE_INVALID
|| !unit_name_is_valid(name
, UNIT_NAME_PLAIN
|UNIT_NAME_INSTANCE
)) {
1470 if (unit_name_is_valid(name
, UNIT_NAME_TEMPLATE
))
1471 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is missing the instance name.", name
);
1473 return sd_bus_error_setf(e
, SD_BUS_ERROR_INVALID_ARGS
, "Unit name %s is not valid.", name
);
1476 ret
= manager_get_unit(m
, name
);
1482 ret
= unit_new(m
, unit_vtable
[t
]->object_size
);
1487 ret
->fragment_path
= strdup(path
);
1488 if (!ret
->fragment_path
) {
1494 r
= unit_add_name(ret
, name
);
1500 unit_add_to_load_queue(ret
);
1501 unit_add_to_dbus_queue(ret
);
1502 unit_add_to_gc_queue(ret
);
1510 int manager_load_unit(
1521 /* This will load the service information files, but not actually
1522 * start any services or anything. */
1524 r
= manager_load_unit_prepare(m
, name
, path
, e
, _ret
);
1528 manager_dispatch_load_queue(m
);
1531 *_ret
= unit_follow_merge(*_ret
);
1536 void manager_dump_jobs(Manager
*s
, FILE *f
, const char *prefix
) {
1543 HASHMAP_FOREACH(j
, s
->jobs
, i
)
1544 job_dump(j
, f
, prefix
);
1547 void manager_dump_units(Manager
*s
, FILE *f
, const char *prefix
) {
1555 HASHMAP_FOREACH_KEY(u
, t
, s
->units
, i
)
1557 unit_dump(u
, f
, prefix
);
1560 void manager_clear_jobs(Manager
*m
) {
1565 while ((j
= hashmap_first(m
->jobs
)))
1566 /* No need to recurse. We're cancelling all jobs. */
1567 job_finish_and_invalidate(j
, JOB_CANCELED
, false, false);
1570 static int manager_dispatch_run_queue(sd_event_source
*source
, void *userdata
) {
1571 Manager
*m
= userdata
;
1577 while ((j
= m
->run_queue
)) {
1578 assert(j
->installed
);
1579 assert(j
->in_run_queue
);
1581 job_run_and_invalidate(j
);
1584 if (m
->n_running_jobs
> 0)
1585 manager_watch_jobs_in_progress(m
);
1587 if (m
->n_on_console
> 0)
1588 manager_watch_idle_pipe(m
);
1593 static unsigned manager_dispatch_dbus_queue(Manager
*m
) {
1600 if (m
->dispatching_dbus_queue
)
1603 m
->dispatching_dbus_queue
= true;
1605 while ((u
= m
->dbus_unit_queue
)) {
1606 assert(u
->in_dbus_queue
);
1608 bus_unit_send_change_signal(u
);
1612 while ((j
= m
->dbus_job_queue
)) {
1613 assert(j
->in_dbus_queue
);
1615 bus_job_send_change_signal(j
);
1619 m
->dispatching_dbus_queue
= false;
1621 if (m
->send_reloading_done
) {
1622 m
->send_reloading_done
= false;
1624 bus_manager_send_reloading(m
, false);
1627 if (m
->queued_message
)
1628 bus_send_queued_message(m
);
1633 static int manager_dispatch_cgroups_agent_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1634 Manager
*m
= userdata
;
1635 char buf
[PATH_MAX
+1];
1638 n
= recv(fd
, buf
, sizeof(buf
), 0);
1640 return log_error_errno(errno
, "Failed to read cgroups agent message: %m");
1642 log_error("Got zero-length cgroups agent message, ignoring.");
1645 if ((size_t) n
>= sizeof(buf
)) {
1646 log_error("Got overly long cgroups agent message, ignoring.");
1650 if (memchr(buf
, 0, n
)) {
1651 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1656 manager_notify_cgroup_empty(m
, buf
);
1657 bus_forward_agent_released(m
, buf
);
1662 static void manager_invoke_notify_message(Manager
*m
, Unit
*u
, pid_t pid
, const char *buf
, size_t n
, FDSet
*fds
) {
1663 _cleanup_strv_free_
char **tags
= NULL
;
1670 tags
= strv_split(buf
, "\n\r");
1676 if (UNIT_VTABLE(u
)->notify_message
)
1677 UNIT_VTABLE(u
)->notify_message(u
, pid
, tags
, fds
);
1679 log_unit_debug(u
, "Got notification message for unit. Ignoring.");
1682 static int manager_dispatch_notify_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1684 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
1685 Manager
*m
= userdata
;
1686 char buf
[NOTIFY_BUFFER_MAX
+1];
1687 struct iovec iovec
= {
1689 .iov_len
= sizeof(buf
)-1,
1692 struct cmsghdr cmsghdr
;
1693 uint8_t buf
[CMSG_SPACE(sizeof(struct ucred
)) +
1694 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX
)];
1696 struct msghdr msghdr
= {
1699 .msg_control
= &control
,
1700 .msg_controllen
= sizeof(control
),
1703 struct cmsghdr
*cmsg
;
1704 struct ucred
*ucred
= NULL
;
1707 int r
, *fd_array
= NULL
;
1712 assert(m
->notify_fd
== fd
);
1714 if (revents
!= EPOLLIN
) {
1715 log_warning("Got unexpected poll event for notify fd.");
1719 n
= recvmsg(m
->notify_fd
, &msghdr
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
1721 if (errno
== EAGAIN
|| errno
== EINTR
)
1727 CMSG_FOREACH(cmsg
, &msghdr
) {
1728 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
1730 fd_array
= (int*) CMSG_DATA(cmsg
);
1731 n_fds
= (cmsg
->cmsg_len
- CMSG_LEN(0)) / sizeof(int);
1733 } else if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1734 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1735 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1737 ucred
= (struct ucred
*) CMSG_DATA(cmsg
);
1744 r
= fdset_new_array(&fds
, fd_array
, n_fds
);
1746 close_many(fd_array
, n_fds
);
1751 if (!ucred
|| ucred
->pid
<= 0) {
1752 log_warning("Received notify message without valid credentials. Ignoring.");
1756 if ((size_t) n
>= sizeof(buf
)) {
1757 log_warning("Received notify message exceeded maximum size. Ignoring.");
1763 /* Notify every unit that might be interested, but try
1764 * to avoid notifying the same one multiple times. */
1765 u1
= manager_get_unit_by_pid_cgroup(m
, ucred
->pid
);
1767 manager_invoke_notify_message(m
, u1
, ucred
->pid
, buf
, n
, fds
);
1771 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(ucred
->pid
));
1772 if (u2
&& u2
!= u1
) {
1773 manager_invoke_notify_message(m
, u2
, ucred
->pid
, buf
, n
, fds
);
1777 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(ucred
->pid
));
1778 if (u3
&& u3
!= u2
&& u3
!= u1
) {
1779 manager_invoke_notify_message(m
, u3
, ucred
->pid
, buf
, n
, fds
);
1784 log_warning("Cannot find unit for notify message of PID "PID_FMT
".", ucred
->pid
);
1786 if (fdset_size(fds
) > 0)
1787 log_warning("Got auxiliary fds with notification message, closing all.");
1792 static void invoke_sigchld_event(Manager
*m
, Unit
*u
, const siginfo_t
*si
) {
1799 sd_event_get_iteration(m
->event
, &iteration
);
1801 log_unit_debug(u
, "Child "PID_FMT
" belongs to %s", si
->si_pid
, u
->id
);
1803 unit_unwatch_pid(u
, si
->si_pid
);
1805 if (UNIT_VTABLE(u
)->sigchld_event
) {
1806 if (set_size(u
->pids
) <= 1 ||
1807 iteration
!= u
->sigchldgen
||
1808 unit_main_pid(u
) == si
->si_pid
||
1809 unit_control_pid(u
) == si
->si_pid
) {
1810 UNIT_VTABLE(u
)->sigchld_event(u
, si
->si_pid
, si
->si_code
, si
->si_status
);
1811 u
->sigchldgen
= iteration
;
1813 log_debug("%s already issued a sigchld this iteration %" PRIu64
", skipping. Pids still being watched %d", u
->id
, iteration
, set_size(u
->pids
));
1817 static int manager_dispatch_sigchld(Manager
*m
) {
1823 /* First we call waitd() for a PID and do not reap the
1824 * zombie. That way we can still access /proc/$PID for
1825 * it while it is a zombie. */
1826 if (waitid(P_ALL
, 0, &si
, WEXITED
|WNOHANG
|WNOWAIT
) < 0) {
1828 if (errno
== ECHILD
)
1840 if (si
.si_code
== CLD_EXITED
|| si
.si_code
== CLD_KILLED
|| si
.si_code
== CLD_DUMPED
) {
1841 _cleanup_free_
char *name
= NULL
;
1844 get_process_comm(si
.si_pid
, &name
);
1846 log_debug("Child "PID_FMT
" (%s) died (code=%s, status=%i/%s)",
1847 si
.si_pid
, strna(name
),
1848 sigchld_code_to_string(si
.si_code
),
1850 strna(si
.si_code
== CLD_EXITED
1851 ? exit_status_to_string(si
.si_status
, EXIT_STATUS_FULL
)
1852 : signal_to_string(si
.si_status
)));
1854 /* And now figure out the unit this belongs
1855 * to, it might be multiple... */
1856 u1
= manager_get_unit_by_pid_cgroup(m
, si
.si_pid
);
1858 invoke_sigchld_event(m
, u1
, &si
);
1859 u2
= hashmap_get(m
->watch_pids1
, PID_TO_PTR(si
.si_pid
));
1861 invoke_sigchld_event(m
, u2
, &si
);
1862 u3
= hashmap_get(m
->watch_pids2
, PID_TO_PTR(si
.si_pid
));
1863 if (u3
&& u3
!= u2
&& u3
!= u1
)
1864 invoke_sigchld_event(m
, u3
, &si
);
1867 /* And now, we actually reap the zombie. */
1868 if (waitid(P_PID
, si
.si_pid
, &si
, WEXITED
) < 0) {
1879 static int manager_start_target(Manager
*m
, const char *name
, JobMode mode
) {
1880 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1883 log_debug("Activating special unit %s", name
);
1885 r
= manager_add_job_by_name(m
, JOB_START
, name
, mode
, &error
, NULL
);
1887 log_error("Failed to enqueue %s job: %s", name
, bus_error_message(&error
, r
));
1892 static int manager_dispatch_signal_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
1893 Manager
*m
= userdata
;
1895 struct signalfd_siginfo sfsi
;
1896 bool sigchld
= false;
1900 assert(m
->signal_fd
== fd
);
1902 if (revents
!= EPOLLIN
) {
1903 log_warning("Got unexpected events from signal file descriptor.");
1908 n
= read(m
->signal_fd
, &sfsi
, sizeof(sfsi
));
1909 if (n
!= sizeof(sfsi
)) {
1914 if (errno
== EINTR
|| errno
== EAGAIN
)
1920 log_received_signal(sfsi
.ssi_signo
== SIGCHLD
||
1921 (sfsi
.ssi_signo
== SIGTERM
&& MANAGER_IS_USER(m
))
1922 ? LOG_DEBUG
: LOG_INFO
,
1925 switch (sfsi
.ssi_signo
) {
1932 if (MANAGER_IS_SYSTEM(m
)) {
1933 /* This is for compatibility with the
1934 * original sysvinit */
1935 m
->exit_code
= MANAGER_REEXECUTE
;
1942 if (MANAGER_IS_SYSTEM(m
)) {
1944 /* If the user presses C-A-D more than
1945 * 7 times within 2s, we reboot
1948 if (ratelimit_test(&m
->ctrl_alt_del_ratelimit
))
1949 manager_start_target(m
, SPECIAL_CTRL_ALT_DEL_TARGET
, JOB_REPLACE_IRREVERSIBLY
);
1951 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1952 status_printf(NULL
, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, rebooting immediately.");
1953 m
->exit_code
= MANAGER_REBOOT
;
1959 /* Run the exit target if there is one, if not, just exit. */
1960 if (manager_start_target(m
, SPECIAL_EXIT_TARGET
, JOB_REPLACE
) < 0) {
1961 m
->exit_code
= MANAGER_EXIT
;
1968 if (MANAGER_IS_SYSTEM(m
))
1969 manager_start_target(m
, SPECIAL_KBREQUEST_TARGET
, JOB_REPLACE
);
1971 /* This is a nop on non-init */
1975 if (MANAGER_IS_SYSTEM(m
))
1976 manager_start_target(m
, SPECIAL_SIGPWR_TARGET
, JOB_REPLACE
);
1978 /* This is a nop on non-init */
1984 u
= manager_get_unit(m
, SPECIAL_DBUS_SERVICE
);
1986 if (!u
|| UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
))) {
1987 log_info("Trying to reconnect to bus...");
1991 if (!u
|| !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
))) {
1992 log_info("Loading D-Bus service...");
1993 manager_start_target(m
, SPECIAL_DBUS_SERVICE
, JOB_REPLACE
);
2000 _cleanup_free_
char *dump
= NULL
;
2001 _cleanup_fclose_
FILE *f
= NULL
;
2004 f
= open_memstream(&dump
, &size
);
2006 log_warning_errno(errno
, "Failed to allocate memory stream: %m");
2010 manager_dump_units(m
, f
, "\t");
2011 manager_dump_jobs(m
, f
, "\t");
2013 r
= fflush_and_check(f
);
2015 log_warning_errno(r
, "Failed to write status stream: %m");
2019 log_dump(LOG_INFO
, dump
);
2024 m
->exit_code
= MANAGER_RELOAD
;
2029 /* Starting SIGRTMIN+0 */
2030 static const char * const target_table
[] = {
2031 [0] = SPECIAL_DEFAULT_TARGET
,
2032 [1] = SPECIAL_RESCUE_TARGET
,
2033 [2] = SPECIAL_EMERGENCY_TARGET
,
2034 [3] = SPECIAL_HALT_TARGET
,
2035 [4] = SPECIAL_POWEROFF_TARGET
,
2036 [5] = SPECIAL_REBOOT_TARGET
,
2037 [6] = SPECIAL_KEXEC_TARGET
2040 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2041 static const ManagerExitCode code_table
[] = {
2043 [1] = MANAGER_POWEROFF
,
2044 [2] = MANAGER_REBOOT
,
2048 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+0 &&
2049 (int) sfsi
.ssi_signo
< SIGRTMIN
+(int) ELEMENTSOF(target_table
)) {
2050 int idx
= (int) sfsi
.ssi_signo
- SIGRTMIN
;
2051 manager_start_target(m
, target_table
[idx
],
2052 (idx
== 1 || idx
== 2) ? JOB_ISOLATE
: JOB_REPLACE
);
2056 if ((int) sfsi
.ssi_signo
>= SIGRTMIN
+13 &&
2057 (int) sfsi
.ssi_signo
< SIGRTMIN
+13+(int) ELEMENTSOF(code_table
)) {
2058 m
->exit_code
= code_table
[sfsi
.ssi_signo
- SIGRTMIN
- 13];
2062 switch (sfsi
.ssi_signo
- SIGRTMIN
) {
2065 manager_set_show_status(m
, SHOW_STATUS_YES
);
2069 manager_set_show_status(m
, SHOW_STATUS_NO
);
2073 log_set_max_level(LOG_DEBUG
);
2074 log_info("Setting log level to debug.");
2078 log_set_max_level(LOG_INFO
);
2079 log_info("Setting log level to info.");
2083 if (MANAGER_IS_USER(m
)) {
2084 m
->exit_code
= MANAGER_EXIT
;
2088 /* This is a nop on init */
2092 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2093 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG
);
2094 log_notice("Setting log target to journal-or-kmsg.");
2098 log_set_target(LOG_TARGET_CONSOLE
);
2099 log_notice("Setting log target to console.");
2103 log_set_target(LOG_TARGET_KMSG
);
2104 log_notice("Setting log target to kmsg.");
2108 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi
.ssi_signo
));
2115 manager_dispatch_sigchld(m
);
2120 static int manager_dispatch_time_change_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2121 Manager
*m
= userdata
;
2126 assert(m
->time_change_fd
== fd
);
2128 log_struct(LOG_INFO
,
2129 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE
),
2130 LOG_MESSAGE("Time has been changed"),
2133 /* Restart the watch */
2134 m
->time_change_event_source
= sd_event_source_unref(m
->time_change_event_source
);
2135 m
->time_change_fd
= safe_close(m
->time_change_fd
);
2137 manager_setup_time_change(m
);
2139 HASHMAP_FOREACH(u
, m
->units
, i
)
2140 if (UNIT_VTABLE(u
)->time_change
)
2141 UNIT_VTABLE(u
)->time_change(u
);
2146 static int manager_dispatch_idle_pipe_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
2147 Manager
*m
= userdata
;
2150 assert(m
->idle_pipe
[2] == fd
);
2152 m
->no_console_output
= m
->n_on_console
> 0;
2154 manager_close_idle_pipe(m
);
2159 static int manager_dispatch_jobs_in_progress(sd_event_source
*source
, usec_t usec
, void *userdata
) {
2160 Manager
*m
= userdata
;
2167 manager_print_jobs_in_progress(m
);
2169 next
= now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_PERIOD_USEC
;
2170 r
= sd_event_source_set_time(source
, next
);
2174 return sd_event_source_set_enabled(source
, SD_EVENT_ONESHOT
);
2177 int manager_loop(Manager
*m
) {
2180 RATELIMIT_DEFINE(rl
, 1*USEC_PER_SEC
, 50000);
2183 m
->exit_code
= MANAGER_OK
;
2185 /* Release the path cache */
2186 m
->unit_path_cache
= set_free_free(m
->unit_path_cache
);
2188 manager_check_finished(m
);
2190 /* There might still be some zombies hanging around from
2191 * before we were exec()'ed. Let's reap them. */
2192 r
= manager_dispatch_sigchld(m
);
2196 while (m
->exit_code
== MANAGER_OK
) {
2199 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
))
2202 if (!ratelimit_test(&rl
)) {
2203 /* Yay, something is going seriously wrong, pause a little */
2204 log_warning("Looping too fast. Throttling execution a little.");
2208 if (manager_dispatch_load_queue(m
) > 0)
2211 if (manager_dispatch_gc_queue(m
) > 0)
2214 if (manager_dispatch_cleanup_queue(m
) > 0)
2217 if (manager_dispatch_cgroup_queue(m
) > 0)
2220 if (manager_dispatch_dbus_queue(m
) > 0)
2223 /* Sleep for half the watchdog time */
2224 if (m
->runtime_watchdog
> 0 && m
->runtime_watchdog
!= USEC_INFINITY
&& MANAGER_IS_SYSTEM(m
)) {
2225 wait_usec
= m
->runtime_watchdog
/ 2;
2229 wait_usec
= USEC_INFINITY
;
2231 r
= sd_event_run(m
->event
, wait_usec
);
2233 return log_error_errno(r
, "Failed to run event loop: %m");
2236 return m
->exit_code
;
2239 int manager_load_unit_from_dbus_path(Manager
*m
, const char *s
, sd_bus_error
*e
, Unit
**_u
) {
2240 _cleanup_free_
char *n
= NULL
;
2248 r
= unit_name_from_dbus_path(s
, &n
);
2252 r
= manager_load_unit(m
, n
, NULL
, e
, &u
);
2261 int manager_get_job_from_dbus_path(Manager
*m
, const char *s
, Job
**_j
) {
2271 p
= startswith(s
, "/org/freedesktop/systemd1/job/");
2275 r
= safe_atou(p
, &id
);
2279 j
= manager_get_job(m
, id
);
2288 void manager_send_unit_audit(Manager
*m
, Unit
*u
, int type
, bool success
) {
2291 _cleanup_free_
char *p
= NULL
;
2295 if (!MANAGER_IS_SYSTEM(m
))
2298 audit_fd
= get_audit_fd();
2302 /* Don't generate audit events if the service was already
2303 * started and we're just deserializing */
2304 if (MANAGER_IS_RELOADING(m
))
2307 if (u
->type
!= UNIT_SERVICE
)
2310 r
= unit_name_to_prefix_and_instance(u
->id
, &p
);
2312 log_error_errno(r
, "Failed to extract prefix and instance of unit name: %m");
2316 msg
= strjoina("unit=", p
);
2317 if (audit_log_user_comm_message(audit_fd
, type
, msg
, "systemd", NULL
, NULL
, NULL
, success
) < 0) {
2319 /* We aren't allowed to send audit messages?
2320 * Then let's not retry again. */
2323 log_warning_errno(errno
, "Failed to send audit message: %m");
2329 void manager_send_unit_plymouth(Manager
*m
, Unit
*u
) {
2330 static const union sockaddr_union sa
= PLYMOUTH_SOCKET
;
2331 _cleanup_free_
char *message
= NULL
;
2332 _cleanup_close_
int fd
= -1;
2335 /* Don't generate plymouth events if the service was already
2336 * started and we're just deserializing */
2337 if (MANAGER_IS_RELOADING(m
))
2340 if (!MANAGER_IS_SYSTEM(m
))
2343 if (detect_container() > 0)
2346 if (u
->type
!= UNIT_SERVICE
&&
2347 u
->type
!= UNIT_MOUNT
&&
2348 u
->type
!= UNIT_SWAP
)
2351 /* We set SOCK_NONBLOCK here so that we rather drop the
2352 * message then wait for plymouth */
2353 fd
= socket(AF_UNIX
, SOCK_STREAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
2355 log_error_errno(errno
, "socket() failed: %m");
2359 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
2361 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2362 log_error_errno(errno
, "connect() failed: %m");
2366 if (asprintf(&message
, "U\002%c%s%n", (int) (strlen(u
->id
) + 1), u
->id
, &n
) < 0) {
2372 if (write(fd
, message
, n
+ 1) != n
+ 1)
2373 if (!IN_SET(errno
, EPIPE
, EAGAIN
, ENOENT
, ECONNREFUSED
, ECONNRESET
, ECONNABORTED
))
2374 log_error_errno(errno
, "Failed to write Plymouth message: %m");
2377 int manager_open_serialization(Manager
*m
, FILE **_f
) {
2384 path
= MANAGER_IS_SYSTEM(m
) ? "/run/systemd" : "/tmp";
2385 fd
= open_tmpfile_unlinkable(path
, O_RDWR
|O_CLOEXEC
);
2389 log_debug("Serializing state to %s", path
);
2391 f
= fdopen(fd
, "w+");
2402 int manager_serialize(Manager
*m
, FILE *f
, FDSet
*fds
, bool switching_root
) {
2415 fprintf(f
, "current-job-id=%"PRIu32
"\n", m
->current_job_id
);
2416 fprintf(f
, "taint-usr=%s\n", yes_no(m
->taint_usr
));
2417 fprintf(f
, "n-installed-jobs=%u\n", m
->n_installed_jobs
);
2418 fprintf(f
, "n-failed-jobs=%u\n", m
->n_failed_jobs
);
2420 dual_timestamp_serialize(f
, "firmware-timestamp", &m
->firmware_timestamp
);
2421 dual_timestamp_serialize(f
, "loader-timestamp", &m
->loader_timestamp
);
2422 dual_timestamp_serialize(f
, "kernel-timestamp", &m
->kernel_timestamp
);
2423 dual_timestamp_serialize(f
, "initrd-timestamp", &m
->initrd_timestamp
);
2426 dual_timestamp_serialize(f
, "userspace-timestamp", &m
->userspace_timestamp
);
2427 dual_timestamp_serialize(f
, "finish-timestamp", &m
->finish_timestamp
);
2428 dual_timestamp_serialize(f
, "security-start-timestamp", &m
->security_start_timestamp
);
2429 dual_timestamp_serialize(f
, "security-finish-timestamp", &m
->security_finish_timestamp
);
2430 dual_timestamp_serialize(f
, "generators-start-timestamp", &m
->generators_start_timestamp
);
2431 dual_timestamp_serialize(f
, "generators-finish-timestamp", &m
->generators_finish_timestamp
);
2432 dual_timestamp_serialize(f
, "units-load-start-timestamp", &m
->units_load_start_timestamp
);
2433 dual_timestamp_serialize(f
, "units-load-finish-timestamp", &m
->units_load_finish_timestamp
);
2436 if (!switching_root
) {
2437 STRV_FOREACH(e
, m
->environment
) {
2438 _cleanup_free_
char *ce
;
2444 fprintf(f
, "env=%s\n", *e
);
2448 if (m
->notify_fd
>= 0) {
2451 copy
= fdset_put_dup(fds
, m
->notify_fd
);
2455 fprintf(f
, "notify-fd=%i\n", copy
);
2456 fprintf(f
, "notify-socket=%s\n", m
->notify_socket
);
2459 if (m
->cgroups_agent_fd
>= 0) {
2462 copy
= fdset_put_dup(fds
, m
->cgroups_agent_fd
);
2466 fprintf(f
, "cgroups-agent-fd=%i\n", copy
);
2469 if (m
->user_lookup_fds
[0] >= 0) {
2472 copy0
= fdset_put_dup(fds
, m
->user_lookup_fds
[0]);
2476 copy1
= fdset_put_dup(fds
, m
->user_lookup_fds
[1]);
2480 fprintf(f
, "user-lookup=%i %i\n", copy0
, copy1
);
2483 if (m
->kdbus_fd
>= 0) {
2486 copy
= fdset_put_dup(fds
, m
->kdbus_fd
);
2490 fprintf(f
, "kdbus-fd=%i\n", copy
);
2493 bus_track_serialize(m
->subscribed
, f
);
2495 r
= dynamic_user_serialize(m
, f
, fds
);
2499 manager_serialize_uid_refs(m
, f
);
2500 manager_serialize_gid_refs(m
, f
);
2504 HASHMAP_FOREACH_KEY(u
, t
, m
->units
, i
) {
2512 r
= unit_serialize(u
, f
, fds
, !switching_root
);
2519 assert(m
->n_reloading
> 0);
2525 r
= bus_fdset_add_all(m
, fds
);
2532 int manager_deserialize(Manager
*m
, FILE *f
, FDSet
*fds
) {
2538 log_debug("Deserializing state...");
2543 char line
[LINE_MAX
], *l
;
2545 if (!fgets(line
, sizeof(line
), f
)) {
2560 if (startswith(l
, "current-job-id=")) {
2563 if (safe_atou32(l
+15, &id
) < 0)
2564 log_debug("Failed to parse current job id value %s", l
+15);
2566 m
->current_job_id
= MAX(m
->current_job_id
, id
);
2568 } else if (startswith(l
, "n-installed-jobs=")) {
2571 if (safe_atou32(l
+17, &n
) < 0)
2572 log_debug("Failed to parse installed jobs counter %s", l
+17);
2574 m
->n_installed_jobs
+= n
;
2576 } else if (startswith(l
, "n-failed-jobs=")) {
2579 if (safe_atou32(l
+14, &n
) < 0)
2580 log_debug("Failed to parse failed jobs counter %s", l
+14);
2582 m
->n_failed_jobs
+= n
;
2584 } else if (startswith(l
, "taint-usr=")) {
2587 b
= parse_boolean(l
+10);
2589 log_debug("Failed to parse taint /usr flag %s", l
+10);
2591 m
->taint_usr
= m
->taint_usr
|| b
;
2593 } else if (startswith(l
, "firmware-timestamp="))
2594 dual_timestamp_deserialize(l
+19, &m
->firmware_timestamp
);
2595 else if (startswith(l
, "loader-timestamp="))
2596 dual_timestamp_deserialize(l
+17, &m
->loader_timestamp
);
2597 else if (startswith(l
, "kernel-timestamp="))
2598 dual_timestamp_deserialize(l
+17, &m
->kernel_timestamp
);
2599 else if (startswith(l
, "initrd-timestamp="))
2600 dual_timestamp_deserialize(l
+17, &m
->initrd_timestamp
);
2601 else if (startswith(l
, "userspace-timestamp="))
2602 dual_timestamp_deserialize(l
+20, &m
->userspace_timestamp
);
2603 else if (startswith(l
, "finish-timestamp="))
2604 dual_timestamp_deserialize(l
+17, &m
->finish_timestamp
);
2605 else if (startswith(l
, "security-start-timestamp="))
2606 dual_timestamp_deserialize(l
+25, &m
->security_start_timestamp
);
2607 else if (startswith(l
, "security-finish-timestamp="))
2608 dual_timestamp_deserialize(l
+26, &m
->security_finish_timestamp
);
2609 else if (startswith(l
, "generators-start-timestamp="))
2610 dual_timestamp_deserialize(l
+27, &m
->generators_start_timestamp
);
2611 else if (startswith(l
, "generators-finish-timestamp="))
2612 dual_timestamp_deserialize(l
+28, &m
->generators_finish_timestamp
);
2613 else if (startswith(l
, "units-load-start-timestamp="))
2614 dual_timestamp_deserialize(l
+27, &m
->units_load_start_timestamp
);
2615 else if (startswith(l
, "units-load-finish-timestamp="))
2616 dual_timestamp_deserialize(l
+28, &m
->units_load_finish_timestamp
);
2617 else if (startswith(l
, "env=")) {
2618 _cleanup_free_
char *uce
= NULL
;
2621 r
= cunescape(l
+ 4, UNESCAPE_RELAX
, &uce
);
2625 e
= strv_env_set(m
->environment
, uce
);
2631 strv_free(m
->environment
);
2634 } else if (startswith(l
, "notify-fd=")) {
2637 if (safe_atoi(l
+ 10, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2638 log_debug("Failed to parse notify fd: %s", l
+ 10);
2640 m
->notify_event_source
= sd_event_source_unref(m
->notify_event_source
);
2641 safe_close(m
->notify_fd
);
2642 m
->notify_fd
= fdset_remove(fds
, fd
);
2645 } else if (startswith(l
, "notify-socket=")) {
2654 free(m
->notify_socket
);
2655 m
->notify_socket
= n
;
2657 } else if (startswith(l
, "cgroups-agent-fd=")) {
2660 if (safe_atoi(l
+ 17, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2661 log_debug("Failed to parse cgroups agent fd: %s", l
+ 10);
2663 m
->cgroups_agent_event_source
= sd_event_source_unref(m
->cgroups_agent_event_source
);
2664 safe_close(m
->cgroups_agent_fd
);
2665 m
->cgroups_agent_fd
= fdset_remove(fds
, fd
);
2668 } else if (startswith(l
, "user-lookup=")) {
2671 if (sscanf(l
+ 12, "%i %i", &fd0
, &fd1
) != 2 || fd0
< 0 || fd1
< 0 || fd0
== fd1
|| !fdset_contains(fds
, fd0
) || !fdset_contains(fds
, fd1
))
2672 log_debug("Failed to parse user lookup fd: %s", l
+ 12);
2674 m
->user_lookup_event_source
= sd_event_source_unref(m
->user_lookup_event_source
);
2675 safe_close_pair(m
->user_lookup_fds
);
2676 m
->user_lookup_fds
[0] = fdset_remove(fds
, fd0
);
2677 m
->user_lookup_fds
[1] = fdset_remove(fds
, fd1
);
2680 } else if (startswith(l
, "kdbus-fd=")) {
2683 if (safe_atoi(l
+ 9, &fd
) < 0 || fd
< 0 || !fdset_contains(fds
, fd
))
2684 log_debug("Failed to parse kdbus fd: %s", l
+ 9);
2686 safe_close(m
->kdbus_fd
);
2687 m
->kdbus_fd
= fdset_remove(fds
, fd
);
2690 } else if (startswith(l
, "dynamic-user="))
2691 dynamic_user_deserialize_one(m
, l
+ 13, fds
);
2692 else if (startswith(l
, "destroy-ipc-uid="))
2693 manager_deserialize_uid_refs_one(m
, l
+ 16);
2694 else if (startswith(l
, "destroy-ipc-gid="))
2695 manager_deserialize_gid_refs_one(m
, l
+ 16);
2699 k
= bus_track_deserialize_item(&m
->deserialized_subscribed
, l
);
2701 log_debug_errno(k
, "Failed to deserialize bus tracker object: %m");
2703 log_debug("Unknown serialization item '%s'", l
);
2709 char name
[UNIT_NAME_MAX
+2];
2712 if (!fgets(name
, sizeof(name
), f
)) {
2723 r
= manager_load_unit(m
, strstrip(name
), NULL
, NULL
, &u
);
2727 r
= unit_deserialize(u
, f
, fds
);
2736 assert(m
->n_reloading
> 0);
2742 int manager_reload(Manager
*m
) {
2744 _cleanup_fclose_
FILE *f
= NULL
;
2745 _cleanup_fdset_free_ FDSet
*fds
= NULL
;
2749 r
= manager_open_serialization(m
, &f
);
2754 bus_manager_send_reloading(m
, true);
2762 r
= manager_serialize(m
, f
, fds
, false);
2768 if (fseeko(f
, 0, SEEK_SET
) < 0) {
2773 /* From here on there is no way back. */
2774 manager_clear_jobs_and_units(m
);
2775 lookup_paths_flush_generator(&m
->lookup_paths
);
2776 lookup_paths_free(&m
->lookup_paths
);
2777 dynamic_user_vacuum(m
, false);
2778 m
->uid_refs
= hashmap_free(m
->uid_refs
);
2779 m
->gid_refs
= hashmap_free(m
->gid_refs
);
2781 q
= lookup_paths_init(&m
->lookup_paths
, m
->unit_file_scope
, 0, NULL
);
2782 if (q
< 0 && r
>= 0)
2785 /* Find new unit paths */
2786 q
= manager_run_generators(m
);
2787 if (q
< 0 && r
>= 0)
2790 lookup_paths_reduce(&m
->lookup_paths
);
2791 manager_build_unit_path_cache(m
);
2793 /* First, enumerate what we can from all config files */
2794 manager_enumerate(m
);
2796 /* Second, deserialize our stored data */
2797 q
= manager_deserialize(m
, f
, fds
);
2798 if (q
< 0 && r
>= 0)
2804 /* Re-register notify_fd as event source */
2805 q
= manager_setup_notify(m
);
2806 if (q
< 0 && r
>= 0)
2809 q
= manager_setup_cgroups_agent(m
);
2810 if (q
< 0 && r
>= 0)
2813 q
= manager_setup_user_lookup_fd(m
);
2814 if (q
< 0 && r
>= 0)
2817 /* Third, fire things up! */
2818 manager_coldplug(m
);
2820 /* Release any dynamic users no longer referenced */
2821 dynamic_user_vacuum(m
, true);
2823 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2824 manager_vacuum_uid_refs(m
);
2825 manager_vacuum_gid_refs(m
);
2827 /* Sync current state of bus names with our set of listening units */
2829 manager_sync_bus_names(m
, m
->api_bus
);
2831 assert(m
->n_reloading
> 0);
2834 m
->send_reloading_done
= true;
2839 void manager_reset_failed(Manager
*m
) {
2845 HASHMAP_FOREACH(u
, m
->units
, i
)
2846 unit_reset_failed(u
);
2849 bool manager_unit_inactive_or_pending(Manager
*m
, const char *name
) {
2855 /* Returns true if the unit is inactive or going down */
2856 u
= manager_get_unit(m
, name
);
2860 return unit_inactive_or_pending(u
);
2863 static void manager_notify_finished(Manager
*m
) {
2864 char userspace
[FORMAT_TIMESPAN_MAX
], initrd
[FORMAT_TIMESPAN_MAX
], kernel
[FORMAT_TIMESPAN_MAX
], sum
[FORMAT_TIMESPAN_MAX
];
2865 usec_t firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
;
2870 if (MANAGER_IS_SYSTEM(m
) && detect_container() <= 0) {
2872 /* Note that m->kernel_usec.monotonic is always at 0,
2873 * and m->firmware_usec.monotonic and
2874 * m->loader_usec.monotonic should be considered
2875 * negative values. */
2877 firmware_usec
= m
->firmware_timestamp
.monotonic
- m
->loader_timestamp
.monotonic
;
2878 loader_usec
= m
->loader_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2879 userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2880 total_usec
= m
->firmware_timestamp
.monotonic
+ m
->finish_timestamp
.monotonic
;
2882 if (dual_timestamp_is_set(&m
->initrd_timestamp
)) {
2884 kernel_usec
= m
->initrd_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2885 initrd_usec
= m
->userspace_timestamp
.monotonic
- m
->initrd_timestamp
.monotonic
;
2887 log_struct(LOG_INFO
,
2888 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2889 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2890 "INITRD_USEC="USEC_FMT
, initrd_usec
,
2891 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2892 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2893 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2894 format_timespan(initrd
, sizeof(initrd
), initrd_usec
, USEC_PER_MSEC
),
2895 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2896 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2899 kernel_usec
= m
->userspace_timestamp
.monotonic
- m
->kernel_timestamp
.monotonic
;
2902 log_struct(LOG_INFO
,
2903 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2904 "KERNEL_USEC="USEC_FMT
, kernel_usec
,
2905 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2906 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2907 format_timespan(kernel
, sizeof(kernel
), kernel_usec
, USEC_PER_MSEC
),
2908 format_timespan(userspace
, sizeof(userspace
), userspace_usec
, USEC_PER_MSEC
),
2909 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2913 firmware_usec
= loader_usec
= initrd_usec
= kernel_usec
= 0;
2914 total_usec
= userspace_usec
= m
->finish_timestamp
.monotonic
- m
->userspace_timestamp
.monotonic
;
2916 log_struct(LOG_INFO
,
2917 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED
),
2918 "USERSPACE_USEC="USEC_FMT
, userspace_usec
,
2919 LOG_MESSAGE("Startup finished in %s.",
2920 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
)),
2924 bus_manager_send_finished(m
, firmware_usec
, loader_usec
, kernel_usec
, initrd_usec
, userspace_usec
, total_usec
);
2928 "STATUS=Startup finished in %s.",
2929 format_timespan(sum
, sizeof(sum
), total_usec
, USEC_PER_MSEC
));
2932 void manager_check_finished(Manager
*m
) {
2935 if (MANAGER_IS_RELOADING(m
))
2938 /* Verify that we are actually running currently. Initially
2939 * the exit code is set to invalid, and during operation it is
2940 * then set to MANAGER_OK */
2941 if (m
->exit_code
!= MANAGER_OK
)
2944 if (hashmap_size(m
->jobs
) > 0) {
2945 if (m
->jobs_in_progress_event_source
)
2946 /* Ignore any failure, this is only for feedback */
2947 (void) sd_event_source_set_time(m
->jobs_in_progress_event_source
, now(CLOCK_MONOTONIC
) + JOBS_IN_PROGRESS_WAIT_USEC
);
2952 manager_flip_auto_status(m
, false);
2954 /* Notify Type=idle units that we are done now */
2955 manager_close_idle_pipe(m
);
2957 /* Turn off confirm spawn now */
2958 m
->confirm_spawn
= false;
2960 /* No need to update ask password status when we're going non-interactive */
2961 manager_close_ask_password(m
);
2963 /* This is no longer the first boot */
2964 manager_set_first_boot(m
, false);
2966 if (dual_timestamp_is_set(&m
->finish_timestamp
))
2969 dual_timestamp_get(&m
->finish_timestamp
);
2971 manager_notify_finished(m
);
2973 manager_invalidate_startup_units(m
);
2976 static int manager_run_generators(Manager
*m
) {
2977 _cleanup_strv_free_
char **paths
= NULL
;
2978 const char *argv
[5];
2987 paths
= generator_binary_paths(m
->unit_file_scope
);
2991 /* Optimize by skipping the whole process by not creating output directories
2992 * if no generators are found. */
2993 STRV_FOREACH(path
, paths
) {
2994 if (access(*path
, F_OK
) >= 0)
2996 if (errno
!= ENOENT
)
2997 log_warning_errno(errno
, "Failed to open generator directory %s: %m", *path
);
3003 r
= lookup_paths_mkdir_generator(&m
->lookup_paths
);
3007 argv
[0] = NULL
; /* Leave this empty, execute_directory() will fill something in */
3008 argv
[1] = m
->lookup_paths
.generator
;
3009 argv
[2] = m
->lookup_paths
.generator_early
;
3010 argv
[3] = m
->lookup_paths
.generator_late
;
3013 RUN_WITH_UMASK(0022)
3014 execute_directories((const char* const*) paths
, DEFAULT_TIMEOUT_USEC
, (char**) argv
);
3017 lookup_paths_trim_generator(&m
->lookup_paths
);
3021 int manager_environment_add(Manager
*m
, char **minus
, char **plus
) {
3022 char **a
= NULL
, **b
= NULL
, **l
;
3027 if (!strv_isempty(minus
)) {
3028 a
= strv_env_delete(l
, 1, minus
);
3035 if (!strv_isempty(plus
)) {
3036 b
= strv_env_merge(2, l
, plus
);
3045 if (m
->environment
!= l
)
3046 strv_free(m
->environment
);
3053 manager_clean_environment(m
);
3054 strv_sort(m
->environment
);
3059 int manager_set_default_rlimits(Manager
*m
, struct rlimit
**default_rlimit
) {
3064 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3065 m
->rlimit
[i
] = mfree(m
->rlimit
[i
]);
3067 if (!default_rlimit
[i
])
3070 m
->rlimit
[i
] = newdup(struct rlimit
, default_rlimit
[i
], 1);
3078 void manager_recheck_journal(Manager
*m
) {
3083 if (!MANAGER_IS_SYSTEM(m
))
3086 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SOCKET
);
3087 if (u
&& SOCKET(u
)->state
!= SOCKET_RUNNING
) {
3088 log_close_journal();
3092 u
= manager_get_unit(m
, SPECIAL_JOURNALD_SERVICE
);
3093 if (u
&& SERVICE(u
)->state
!= SERVICE_RUNNING
) {
3094 log_close_journal();
3098 /* Hmm, OK, so the socket is fully up and the service is up
3099 * too, then let's make use of the thing. */
3103 void manager_set_show_status(Manager
*m
, ShowStatus mode
) {
3105 assert(IN_SET(mode
, SHOW_STATUS_AUTO
, SHOW_STATUS_NO
, SHOW_STATUS_YES
, SHOW_STATUS_TEMPORARY
));
3107 if (!MANAGER_IS_SYSTEM(m
))
3110 if (m
->show_status
!= mode
)
3111 log_debug("%s showing of status.",
3112 mode
== SHOW_STATUS_NO
? "Disabling" : "Enabling");
3113 m
->show_status
= mode
;
3116 (void) touch("/run/systemd/show-status");
3118 (void) unlink("/run/systemd/show-status");
3121 static bool manager_get_show_status(Manager
*m
, StatusType type
) {
3124 if (!MANAGER_IS_SYSTEM(m
))
3127 if (m
->no_console_output
)
3130 if (!IN_SET(manager_state(m
), MANAGER_INITIALIZING
, MANAGER_STARTING
, MANAGER_STOPPING
))
3133 /* If we cannot find out the status properly, just proceed. */
3134 if (type
!= STATUS_TYPE_EMERGENCY
&& manager_check_ask_password(m
) > 0)
3137 if (m
->show_status
> 0)
3143 void manager_set_first_boot(Manager
*m
, bool b
) {
3146 if (!MANAGER_IS_SYSTEM(m
))
3149 if (m
->first_boot
!= (int) b
) {
3151 (void) touch("/run/systemd/first-boot");
3153 (void) unlink("/run/systemd/first-boot");
3159 void manager_status_printf(Manager
*m
, StatusType type
, const char *status
, const char *format
, ...) {
3162 /* If m is NULL, assume we're after shutdown and let the messages through. */
3164 if (m
&& !manager_get_show_status(m
, type
))
3167 /* XXX We should totally drop the check for ephemeral here
3168 * and thus effectively make 'Type=idle' pointless. */
3169 if (type
== STATUS_TYPE_EPHEMERAL
&& m
&& m
->n_on_console
> 0)
3172 va_start(ap
, format
);
3173 status_vprintf(status
, true, type
== STATUS_TYPE_EPHEMERAL
, format
, ap
);
3177 Set
*manager_get_units_requiring_mounts_for(Manager
*m
, const char *path
) {
3178 char p
[strlen(path
)+1];
3184 path_kill_slashes(p
);
3186 return hashmap_get(m
->units_requiring_mounts_for
, streq(p
, "/") ? "" : p
);
3189 const char *manager_get_runtime_prefix(Manager
*m
) {
3192 return MANAGER_IS_SYSTEM(m
) ?
3194 getenv("XDG_RUNTIME_DIR");
3197 int manager_update_failed_units(Manager
*m
, Unit
*u
, bool failed
) {
3202 assert(u
->manager
== m
);
3204 size
= set_size(m
->failed_units
);
3207 r
= set_ensure_allocated(&m
->failed_units
, NULL
);
3211 if (set_put(m
->failed_units
, u
) < 0)
3214 (void) set_remove(m
->failed_units
, u
);
3216 if (set_size(m
->failed_units
) != size
)
3217 bus_manager_send_change_signal(m
);
3222 ManagerState
manager_state(Manager
*m
) {
3227 /* Did we ever finish booting? If not then we are still starting up */
3228 if (!dual_timestamp_is_set(&m
->finish_timestamp
)) {
3230 u
= manager_get_unit(m
, SPECIAL_BASIC_TARGET
);
3231 if (!u
|| !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u
)))
3232 return MANAGER_INITIALIZING
;
3234 return MANAGER_STARTING
;
3237 /* Is the special shutdown target queued? If so, we are in shutdown state */
3238 u
= manager_get_unit(m
, SPECIAL_SHUTDOWN_TARGET
);
3239 if (u
&& u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))
3240 return MANAGER_STOPPING
;
3242 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3243 u
= manager_get_unit(m
, SPECIAL_RESCUE_TARGET
);
3244 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3245 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3246 return MANAGER_MAINTENANCE
;
3248 u
= manager_get_unit(m
, SPECIAL_EMERGENCY_TARGET
);
3249 if (u
&& (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u
)) ||
3250 (u
->job
&& IN_SET(u
->job
->type
, JOB_START
, JOB_RESTART
, JOB_RELOAD_OR_START
))))
3251 return MANAGER_MAINTENANCE
;
3253 /* Are there any failed units? If so, we are in degraded mode */
3254 if (set_size(m
->failed_units
) > 0)
3255 return MANAGER_DEGRADED
;
3257 return MANAGER_RUNNING
;
3260 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3262 static void manager_unref_uid_internal(
3267 int (*_clean_ipc
)(uid_t uid
)) {
3273 assert(uid_is_valid(uid
));
3276 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3277 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3279 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3280 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3281 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3282 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3284 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3285 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3287 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3290 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3292 n
= c
& ~DESTROY_IPC_FLAG
;
3296 if (destroy_now
&& n
== 0) {
3297 hashmap_remove(*uid_refs
, UID_TO_PTR(uid
));
3299 if (c
& DESTROY_IPC_FLAG
) {
3300 log_debug("%s " UID_FMT
" is no longer referenced, cleaning up its IPC.",
3301 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3303 (void) _clean_ipc(uid
);
3306 c
= n
| (c
& DESTROY_IPC_FLAG
);
3307 assert_se(hashmap_update(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
)) >= 0);
3311 void manager_unref_uid(Manager
*m
, uid_t uid
, bool destroy_now
) {
3312 manager_unref_uid_internal(m
, &m
->uid_refs
, uid
, destroy_now
, clean_ipc_by_uid
);
3315 void manager_unref_gid(Manager
*m
, gid_t gid
, bool destroy_now
) {
3316 manager_unref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, destroy_now
, clean_ipc_by_gid
);
3319 static int manager_ref_uid_internal(
3330 assert(uid_is_valid(uid
));
3332 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3333 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3335 assert_cc(sizeof(uid_t
) == sizeof(gid_t
));
3336 assert_cc(UID_INVALID
== (uid_t
) GID_INVALID
);
3338 if (uid
== 0) /* We don't keep track of root, and will never destroy it */
3341 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3345 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3347 n
= c
& ~DESTROY_IPC_FLAG
;
3350 if (n
& DESTROY_IPC_FLAG
) /* check for overflow */
3353 c
= n
| (c
& DESTROY_IPC_FLAG
) | (clean_ipc
? DESTROY_IPC_FLAG
: 0);
3355 return hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3358 int manager_ref_uid(Manager
*m
, uid_t uid
, bool clean_ipc
) {
3359 return manager_ref_uid_internal(m
, &m
->uid_refs
, uid
, clean_ipc
);
3362 int manager_ref_gid(Manager
*m
, gid_t gid
, bool clean_ipc
) {
3363 return manager_ref_uid_internal(m
, &m
->gid_refs
, (uid_t
) gid
, clean_ipc
);
3366 static void manager_vacuum_uid_refs_internal(
3369 int (*_clean_ipc
)(uid_t uid
)) {
3378 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3382 uid
= PTR_TO_UID(k
);
3383 c
= PTR_TO_UINT32(p
);
3385 n
= c
& ~DESTROY_IPC_FLAG
;
3389 if (c
& DESTROY_IPC_FLAG
) {
3390 log_debug("Found unreferenced %s " UID_FMT
" after reload/reexec. Cleaning up.",
3391 _clean_ipc
== clean_ipc_by_uid
? "UID" : "GID",
3393 (void) _clean_ipc(uid
);
3396 assert_se(hashmap_remove(*uid_refs
, k
) == p
);
3400 void manager_vacuum_uid_refs(Manager
*m
) {
3401 manager_vacuum_uid_refs_internal(m
, &m
->uid_refs
, clean_ipc_by_uid
);
3404 void manager_vacuum_gid_refs(Manager
*m
) {
3405 manager_vacuum_uid_refs_internal(m
, &m
->gid_refs
, clean_ipc_by_gid
);
3408 static void manager_serialize_uid_refs_internal(
3412 const char *field_name
) {
3422 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3423 * of it is better rebuild after a reload/reexec. */
3425 HASHMAP_FOREACH_KEY(p
, k
, *uid_refs
, i
) {
3429 uid
= PTR_TO_UID(k
);
3430 c
= PTR_TO_UINT32(p
);
3432 if (!(c
& DESTROY_IPC_FLAG
))
3435 fprintf(f
, "%s=" UID_FMT
"\n", field_name
, uid
);
3439 void manager_serialize_uid_refs(Manager
*m
, FILE *f
) {
3440 manager_serialize_uid_refs_internal(m
, f
, &m
->uid_refs
, "destroy-ipc-uid");
3443 void manager_serialize_gid_refs(Manager
*m
, FILE *f
) {
3444 manager_serialize_uid_refs_internal(m
, f
, &m
->gid_refs
, "destroy-ipc-gid");
3447 static void manager_deserialize_uid_refs_one_internal(
3450 const char *value
) {
3460 r
= parse_uid(value
, &uid
);
3461 if (r
< 0 || uid
== 0) {
3462 log_debug("Unable to parse UID reference serialization");
3466 r
= hashmap_ensure_allocated(uid_refs
, &trivial_hash_ops
);
3472 c
= PTR_TO_UINT32(hashmap_get(*uid_refs
, UID_TO_PTR(uid
)));
3473 if (c
& DESTROY_IPC_FLAG
)
3476 c
|= DESTROY_IPC_FLAG
;
3478 r
= hashmap_replace(*uid_refs
, UID_TO_PTR(uid
), UINT32_TO_PTR(c
));
3480 log_debug("Failed to add UID reference entry");
3485 void manager_deserialize_uid_refs_one(Manager
*m
, const char *value
) {
3486 manager_deserialize_uid_refs_one_internal(m
, &m
->uid_refs
, value
);
3489 void manager_deserialize_gid_refs_one(Manager
*m
, const char *value
) {
3490 manager_deserialize_uid_refs_one_internal(m
, &m
->gid_refs
, value
);
3493 int manager_dispatch_user_lookup_fd(sd_event_source
*source
, int fd
, uint32_t revents
, void *userdata
) {
3497 char unit_name
[UNIT_NAME_MAX
+1];
3500 Manager
*m
= userdata
;
3508 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3509 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3510 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3512 l
= recv(fd
, &buffer
, sizeof(buffer
), MSG_DONTWAIT
);
3514 if (errno
== EINTR
|| errno
== EAGAIN
)
3517 return log_error_errno(errno
, "Failed to read from user lookup fd: %m");
3520 if ((size_t) l
<= offsetof(struct buffer
, unit_name
)) {
3521 log_warning("Received too short user lookup message, ignoring.");
3525 if ((size_t) l
> offsetof(struct buffer
, unit_name
) + UNIT_NAME_MAX
) {
3526 log_warning("Received too long user lookup message, ignoring.");
3530 if (!uid_is_valid(buffer
.uid
) && !gid_is_valid(buffer
.gid
)) {
3531 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3535 n
= (size_t) l
- offsetof(struct buffer
, unit_name
);
3536 if (memchr(buffer
.unit_name
, 0, n
)) {
3537 log_warning("Received lookup message with embedded NUL character, ignoring.");
3541 buffer
.unit_name
[n
] = 0;
3542 u
= manager_get_unit(m
, buffer
.unit_name
);
3544 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3548 log_unit_debug(u
, "User lookup succeeded: uid=" UID_FMT
" gid=" GID_FMT
, buffer
.uid
, buffer
.gid
);
3550 unit_notify_user_lookup(u
, buffer
.uid
, buffer
.gid
);
3554 static const char *const manager_state_table
[_MANAGER_STATE_MAX
] = {
3555 [MANAGER_INITIALIZING
] = "initializing",
3556 [MANAGER_STARTING
] = "starting",
3557 [MANAGER_RUNNING
] = "running",
3558 [MANAGER_DEGRADED
] = "degraded",
3559 [MANAGER_MAINTENANCE
] = "maintenance",
3560 [MANAGER_STOPPING
] = "stopping",
3563 DEFINE_STRING_TABLE_LOOKUP(manager_state
, ManagerState
);