1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Copyright © 2004 Chris Friesen <chris_friesen@sympatico.ca>
4 * Copyright © 2009 Canonical Ltd.
5 * Copyright © 2009 Scott James Remnant <scott@netsplit.com>
15 #include <sys/epoll.h>
17 #include <sys/inotify.h>
18 #include <sys/ioctl.h>
19 #include <sys/mount.h>
20 #include <sys/prctl.h>
21 #include <sys/signalfd.h>
27 #include "sd-daemon.h"
30 #include "alloc-util.h"
31 #include "cgroup-setup.h"
32 #include "cgroup-util.h"
33 #include "cpu-set-util.h"
34 #include "dev-setup.h"
35 #include "device-monitor-private.h"
36 #include "device-private.h"
37 #include "device-util.h"
38 #include "errno-list.h"
39 #include "event-util.h"
42 #include "format-util.h"
45 #include "inotify-util.h"
47 #include "limits-util.h"
49 #include "main-func.h"
51 #include "netlink-util.h"
52 #include "parse-util.h"
53 #include "path-util.h"
54 #include "pretty-print.h"
55 #include "proc-cmdline.h"
56 #include "process-util.h"
57 #include "selinux-util.h"
58 #include "signal-util.h"
59 #include "socket-util.h"
60 #include "string-util.h"
63 #include "syslog-util.h"
65 #include "udev-builtin.h"
66 #include "udev-ctrl.h"
67 #include "udev-event.h"
68 #include "udev-util.h"
69 #include "udev-watch.h"
70 #include "user-util.h"
73 #define WORKER_NUM_MAX 2048U
74 #define EVENT_RETRY_INTERVAL_USEC (200 * USEC_PER_MSEC)
75 #define EVENT_RETRY_TIMEOUT_USEC (3 * USEC_PER_MINUTE)
77 static bool arg_debug
= false;
78 static int arg_daemonize
= false;
79 static ResolveNameTiming arg_resolve_name_timing
= RESOLVE_NAME_EARLY
;
80 static unsigned arg_children_max
= 0;
81 static usec_t arg_exec_delay_usec
= 0;
82 static usec_t arg_event_timeout_usec
= 180 * USEC_PER_SEC
;
83 static int arg_timeout_signal
= SIGKILL
;
84 static bool arg_blockdev_read_only
= false;
86 typedef struct Event Event
;
87 typedef struct Worker Worker
;
89 typedef struct Manager
{
92 LIST_HEAD(Event
, events
);
94 pid_t pid
; /* the process that originally allocated the manager object */
102 sd_device_monitor
*monitor
;
106 /* used by udev-watch */
108 sd_event_source
*inotify_event
;
110 sd_event_source
*kill_workers_event
;
114 bool stop_exec_queue
;
118 typedef enum EventState
{
124 typedef struct Event
{
131 sd_device_action_t action
;
133 uint64_t blocker_seqnum
;
136 const char *devpath_old
;
138 usec_t retry_again_next_usec
;
139 usec_t retry_again_timeout_usec
;
141 sd_event_source
*timeout_warning_event
;
142 sd_event_source
*timeout_event
;
144 LIST_FIELDS(Event
, event
);
147 typedef enum WorkerState
{
155 typedef struct Worker
{
158 sd_event_source
*child_event_source
;
159 sd_device_monitor
*monitor
;
164 /* passed from worker to main process */
165 typedef enum EventResult
{
166 EVENT_RESULT_NERRNO_MIN
= -ERRNO_MAX
,
167 EVENT_RESULT_NERRNO_MAX
= -1,
168 EVENT_RESULT_SUCCESS
= 0,
169 EVENT_RESULT_EXIT_STATUS_BASE
= 0,
170 EVENT_RESULT_EXIT_STATUS_MAX
= 255,
171 EVENT_RESULT_TRY_AGAIN
= 256, /* when the block device is locked by another process. */
172 EVENT_RESULT_SIGNAL_BASE
= 257,
173 EVENT_RESULT_SIGNAL_MAX
= EVENT_RESULT_SIGNAL_BASE
+ _NSIG
,
175 _EVENT_RESULT_INVALID
= -EINVAL
,
178 static Event
*event_free(Event
*event
) {
182 assert(event
->manager
);
184 LIST_REMOVE(event
, event
->manager
->events
, event
);
185 sd_device_unref(event
->dev
);
187 /* Do not use sd_event_source_disable_unref() here, as this is called by both workers and the
189 sd_event_source_unref(event
->timeout_warning_event
);
190 sd_event_source_unref(event
->timeout_event
);
193 event
->worker
->event
= NULL
;
198 static void event_queue_cleanup(Manager
*manager
, EventState match_state
) {
199 LIST_FOREACH(event
, event
, manager
->events
) {
200 if (match_state
!= EVENT_UNDEF
&& match_state
!= event
->state
)
207 static Worker
*worker_free(Worker
*worker
) {
212 hashmap_remove(worker
->manager
->workers
, PID_TO_PTR(worker
->pid
));
214 sd_event_source_unref(worker
->child_event_source
);
215 sd_device_monitor_unref(worker
->monitor
);
216 event_free(worker
->event
);
218 return mfree(worker
);
221 DEFINE_TRIVIAL_CLEANUP_FUNC(Worker
*, worker_free
);
222 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(worker_hash_op
, void, trivial_hash_func
, trivial_compare_func
, Worker
, worker_free
);
224 static void manager_clear_for_worker(Manager
*manager
) {
227 /* Do not use sd_event_source_disable_unref() here, as this is called by both workers and the
229 manager
->inotify_event
= sd_event_source_unref(manager
->inotify_event
);
230 manager
->kill_workers_event
= sd_event_source_unref(manager
->kill_workers_event
);
232 manager
->event
= sd_event_unref(manager
->event
);
234 manager
->workers
= hashmap_free(manager
->workers
);
235 event_queue_cleanup(manager
, EVENT_UNDEF
);
237 manager
->monitor
= sd_device_monitor_unref(manager
->monitor
);
238 manager
->ctrl
= udev_ctrl_unref(manager
->ctrl
);
240 manager
->worker_watch
[READ_END
] = safe_close(manager
->worker_watch
[READ_END
]);
243 static Manager
* manager_free(Manager
*manager
) {
249 manager_clear_for_worker(manager
);
251 sd_netlink_unref(manager
->rtnl
);
253 hashmap_free_free_free(manager
->properties
);
254 udev_rules_free(manager
->rules
);
256 safe_close(manager
->inotify_fd
);
257 safe_close_pair(manager
->worker_watch
);
259 free(manager
->cgroup
);
260 return mfree(manager
);
263 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager
*, manager_free
);
265 static int on_sigchld(sd_event_source
*s
, const siginfo_t
*si
, void *userdata
);
267 static int worker_new(Worker
**ret
, Manager
*manager
, sd_device_monitor
*worker_monitor
, pid_t pid
) {
268 _cleanup_(worker_freep
) Worker
*worker
= NULL
;
273 assert(worker_monitor
);
276 /* close monitor, but keep address around */
277 device_monitor_disconnect(worker_monitor
);
279 worker
= new(Worker
, 1);
284 .monitor
= sd_device_monitor_ref(worker_monitor
),
288 r
= sd_event_add_child(manager
->event
, &worker
->child_event_source
, pid
, WEXITED
, on_sigchld
, worker
);
292 r
= hashmap_ensure_put(&manager
->workers
, &worker_hash_op
, PID_TO_PTR(pid
), worker
);
296 worker
->manager
= manager
;
298 *ret
= TAKE_PTR(worker
);
302 static void manager_kill_workers(Manager
*manager
, bool force
) {
307 HASHMAP_FOREACH(worker
, manager
->workers
) {
308 if (worker
->state
== WORKER_KILLED
)
311 if (worker
->state
== WORKER_RUNNING
&& !force
) {
312 worker
->state
= WORKER_KILLING
;
316 worker
->state
= WORKER_KILLED
;
317 (void) kill(worker
->pid
, SIGTERM
);
321 static void manager_exit(Manager
*manager
) {
324 manager
->exit
= true;
328 "STATUS=Starting shutdown...");
330 /* close sources of new events and discard buffered events */
331 manager
->ctrl
= udev_ctrl_unref(manager
->ctrl
);
333 manager
->inotify_event
= sd_event_source_disable_unref(manager
->inotify_event
);
334 manager
->inotify_fd
= safe_close(manager
->inotify_fd
);
336 manager
->monitor
= sd_device_monitor_unref(manager
->monitor
);
338 /* discard queued events and kill workers */
339 event_queue_cleanup(manager
, EVENT_QUEUED
);
340 manager_kill_workers(manager
, true);
343 static void notify_ready(void) {
346 r
= sd_notifyf(false,
348 "STATUS=Processing with %u children at max", arg_children_max
);
350 log_warning_errno(r
, "Failed to send readiness notification, ignoring: %m");
353 /* reload requested, HUP signal received, rules changed, builtin changed */
354 static void manager_reload(Manager
*manager
, bool force
) {
355 _cleanup_(udev_rules_freep
) UdevRules
*rules
= NULL
;
361 assert_se(sd_event_now(manager
->event
, CLOCK_MONOTONIC
, &now_usec
) >= 0);
362 if (!force
&& now_usec
< usec_add(manager
->last_usec
, 3 * USEC_PER_SEC
))
363 /* check for changed config, every 3 seconds at most */
365 manager
->last_usec
= now_usec
;
367 /* Reload SELinux label database, to make the child inherit the up-to-date database. */
368 mac_selinux_maybe_reload();
370 /* Nothing changed. It is not necessary to reload. */
371 if (!udev_rules_should_reload(manager
->rules
) && !udev_builtin_validate())
376 "STATUS=Flushing configuration...");
378 manager_kill_workers(manager
, false);
383 r
= udev_rules_load(&rules
, arg_resolve_name_timing
);
385 log_warning_errno(r
, "Failed to read udev rules, using the previously loaded rules, ignoring: %m");
387 udev_rules_free_and_replace(manager
->rules
, rules
);
392 static int on_kill_workers_event(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
393 Manager
*manager
= userdata
;
397 log_debug("Cleanup idle workers");
398 manager_kill_workers(manager
, false);
403 static void device_broadcast(sd_device_monitor
*monitor
, sd_device
*dev
, EventResult result
) {
408 /* On exit, manager->monitor is already NULL. */
412 if (result
!= EVENT_RESULT_SUCCESS
) {
413 (void) device_add_property(dev
, "UDEV_WORKER_FAILED", "1");
416 case EVENT_RESULT_NERRNO_MIN
... EVENT_RESULT_NERRNO_MAX
: {
419 (void) device_add_propertyf(dev
, "UDEV_WORKER_ERRNO", "%i", -result
);
421 str
= errno_to_name(result
);
423 (void) device_add_property(dev
, "UDEV_WORKER_ERRNO_NAME", str
);
426 case EVENT_RESULT_EXIT_STATUS_BASE
... EVENT_RESULT_EXIT_STATUS_MAX
:
427 (void) device_add_propertyf(dev
, "UDEV_WORKER_EXIT_STATUS", "%i", result
- EVENT_RESULT_EXIT_STATUS_BASE
);
430 case EVENT_RESULT_TRY_AGAIN
:
431 assert_not_reached();
434 case EVENT_RESULT_SIGNAL_BASE
... EVENT_RESULT_SIGNAL_MAX
: {
437 (void) device_add_propertyf(dev
, "UDEV_WORKER_SIGNAL", "%i", result
- EVENT_RESULT_SIGNAL_BASE
);
439 str
= signal_to_string(result
- EVENT_RESULT_SIGNAL_BASE
);
441 (void) device_add_property(dev
, "UDEV_WORKER_SIGNAL_NAME", str
);
445 log_device_warning(dev
, "Unknown event result \"%i\", ignoring.", result
);
449 r
= device_monitor_send_device(monitor
, NULL
, dev
);
451 log_device_warning_errno(dev
, r
,
452 "Failed to broadcast event to libudev listeners, ignoring: %m");
455 static int worker_send_result(Manager
*manager
, EventResult result
) {
457 assert(manager
->worker_watch
[WRITE_END
] >= 0);
459 return loop_write(manager
->worker_watch
[WRITE_END
], &result
, sizeof(result
), false);
462 static int device_get_whole_disk(sd_device
*dev
, sd_device
**ret_device
, const char **ret_devname
) {
468 if (device_for_action(dev
, SD_DEVICE_REMOVE
))
471 r
= sd_device_get_subsystem(dev
, &val
);
473 return log_device_debug_errno(dev
, r
, "Failed to get subsystem: %m");
475 if (!streq(val
, "block"))
478 r
= sd_device_get_sysname(dev
, &val
);
480 return log_device_debug_errno(dev
, r
, "Failed to get sysname: %m");
482 /* Exclude the following devices:
483 * For "dm-", see the comment added by e918a1b5a94f270186dca59156354acd2a596494.
484 * For "md", see the commit message of 2e5b17d01347d3c3118be2b8ad63d20415dbb1f0,
485 * but not sure the assumption is still valid even when partitions are created on the md
486 * devices, surprisingly which seems to be possible, see PR #22973.
487 * For "drbd", see the commit message of fee854ee8ccde0cd28e0f925dea18cce35f3993d. */
488 if (STARTSWITH_SET(val
, "dm-", "md", "drbd"))
491 r
= sd_device_get_devtype(dev
, &val
);
492 if (r
< 0 && r
!= -ENOENT
)
493 return log_device_debug_errno(dev
, r
, "Failed to get devtype: %m");
494 if (r
>= 0 && streq(val
, "partition")) {
495 r
= sd_device_get_parent(dev
, &dev
);
496 if (r
== -ENOENT
) /* The device may be already removed. */
499 return log_device_debug_errno(dev
, r
, "Failed to get parent device: %m");
502 r
= sd_device_get_devname(dev
, &val
);
506 return log_device_debug_errno(dev
, r
, "Failed to get devname: %m");
522 static int worker_lock_whole_disk(sd_device
*dev
, int *ret_fd
) {
523 _cleanup_close_
int fd
= -1;
524 sd_device
*dev_whole_disk
;
531 /* Take a shared lock on the device node; this establishes a concept of device "ownership" to
532 * serialize device access. External processes holding an exclusive lock will cause udev to skip the
533 * event handling; in the case udev acquired the lock, the external process can block until udev has
534 * finished its event handling. */
536 r
= device_get_whole_disk(dev
, &dev_whole_disk
, &val
);
542 fd
= sd_device_open(dev_whole_disk
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
);
544 bool ignore
= ERRNO_IS_DEVICE_ABSENT(fd
);
546 log_device_debug_errno(dev
, fd
, "Failed to open '%s'%s: %m", val
, ignore
? ", ignoring" : "");
553 if (flock(fd
, LOCK_SH
|LOCK_NB
) < 0)
554 return log_device_debug_errno(dev
, errno
, "Failed to flock(%s): %m", val
);
556 *ret_fd
= TAKE_FD(fd
);
564 static int worker_mark_block_device_read_only(sd_device
*dev
) {
565 _cleanup_close_
int fd
= -1;
571 if (!arg_blockdev_read_only
)
574 /* Do this only once, when the block device is new. If the device is later retriggered let's not
575 * toggle the bit again, so that people can boot up with full read-only mode and then unset the bit
576 * for specific devices only. */
577 if (!device_for_action(dev
, SD_DEVICE_ADD
))
580 r
= sd_device_get_subsystem(dev
, &val
);
582 return log_device_debug_errno(dev
, r
, "Failed to get subsystem: %m");
584 if (!streq(val
, "block"))
587 r
= sd_device_get_sysname(dev
, &val
);
589 return log_device_debug_errno(dev
, r
, "Failed to get sysname: %m");
591 /* Exclude synthetic devices for now, this is supposed to be a safety feature to avoid modification
592 * of physical devices, and what sits on top of those doesn't really matter if we don't allow the
593 * underlying block devices to receive changes. */
594 if (STARTSWITH_SET(val
, "dm-", "md", "drbd", "loop", "nbd", "zram"))
597 fd
= sd_device_open(dev
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
);
599 return log_device_debug_errno(dev
, fd
, "Failed to open '%s', ignoring: %m", val
);
601 if (ioctl(fd
, BLKROSET
, &state
) < 0)
602 return log_device_warning_errno(dev
, errno
, "Failed to mark block device '%s' read-only: %m", val
);
604 log_device_info(dev
, "Successfully marked block device '%s' read-only.", val
);
608 static int worker_process_device(Manager
*manager
, sd_device
*dev
) {
609 _cleanup_(udev_event_freep
) UdevEvent
*udev_event
= NULL
;
610 _cleanup_close_
int fd_lock
= -1;
616 log_device_uevent(dev
, "Processing device");
618 udev_event
= udev_event_new(dev
, arg_exec_delay_usec
, manager
->rtnl
, manager
->log_level
);
622 /* If this is a block device and the device is locked currently via the BSD advisory locks,
623 * someone else is using it exclusively. We don't run our udev rules now to not interfere.
624 * Instead of processing the event, we requeue the event and will try again after a delay.
626 * The user-facing side of this: https://systemd.io/BLOCK_DEVICE_LOCKING */
627 r
= worker_lock_whole_disk(dev
, &fd_lock
);
629 return EVENT_RESULT_TRY_AGAIN
;
633 (void) worker_mark_block_device_read_only(dev
);
635 /* apply rules, create node, symlinks */
636 r
= udev_event_execute_rules(
639 arg_event_timeout_usec
,
646 udev_event_execute_run(udev_event
, arg_event_timeout_usec
, arg_timeout_signal
);
649 /* in case rtnl was initialized */
650 manager
->rtnl
= sd_netlink_ref(udev_event
->rtnl
);
652 r
= udev_event_process_inotify_watch(udev_event
, manager
->inotify_fd
);
656 log_device_uevent(dev
, "Device processed");
660 static int worker_device_monitor_handler(sd_device_monitor
*monitor
, sd_device
*dev
, void *userdata
) {
661 Manager
*manager
= userdata
;
667 r
= worker_process_device(manager
, dev
);
668 if (r
== EVENT_RESULT_TRY_AGAIN
)
669 /* if we couldn't acquire the flock(), then requeue the event */
670 log_device_debug(dev
, "Block device is currently locked, requeueing the event.");
673 log_device_warning_errno(dev
, r
, "Failed to process device, ignoring: %m");
675 /* send processed event back to libudev listeners */
676 device_broadcast(monitor
, dev
, r
);
679 /* send udevd the result of the event execution */
680 r
= worker_send_result(manager
, r
);
682 log_device_warning_errno(dev
, r
, "Failed to send signal to main daemon, ignoring: %m");
684 /* Reset the log level, as it might be changed by "OPTIONS=log_level=". */
685 log_set_max_level(manager
->log_level
);
690 static int worker_main(Manager
*_manager
, sd_device_monitor
*monitor
, sd_device
*first_device
) {
691 _cleanup_(sd_device_unrefp
) sd_device
*dev
= first_device
;
692 _cleanup_(manager_freep
) Manager
*manager
= _manager
;
699 assert_se(unsetenv("NOTIFY_SOCKET") == 0);
701 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGTERM
, -1) >= 0);
703 /* Reset OOM score, we only protect the main daemon. */
704 r
= set_oom_score_adjust(0);
706 log_debug_errno(r
, "Failed to reset OOM score, ignoring: %m");
708 /* Clear unnecessary data in Manager object. */
709 manager_clear_for_worker(manager
);
711 r
= sd_event_new(&manager
->event
);
713 return log_error_errno(r
, "Failed to allocate event loop: %m");
715 r
= sd_event_add_signal(manager
->event
, NULL
, SIGTERM
, NULL
, NULL
);
717 return log_error_errno(r
, "Failed to set SIGTERM event: %m");
719 r
= sd_device_monitor_attach_event(monitor
, manager
->event
);
721 return log_error_errno(r
, "Failed to attach event loop to device monitor: %m");
723 r
= sd_device_monitor_start(monitor
, worker_device_monitor_handler
, manager
);
725 return log_error_errno(r
, "Failed to start device monitor: %m");
727 (void) sd_event_source_set_description(sd_device_monitor_get_event_source(monitor
), "worker-device-monitor");
729 /* Process first device */
730 (void) worker_device_monitor_handler(monitor
, dev
, manager
);
732 r
= sd_event_loop(manager
->event
);
734 return log_error_errno(r
, "Event loop failed: %m");
739 static int on_event_timeout(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
740 Event
*event
= userdata
;
743 assert(event
->worker
);
745 kill_and_sigcont(event
->worker
->pid
, arg_timeout_signal
);
746 event
->worker
->state
= WORKER_KILLED
;
748 log_device_error(event
->dev
, "Worker ["PID_FMT
"] processing SEQNUM=%"PRIu64
" killed", event
->worker
->pid
, event
->seqnum
);
753 static int on_event_timeout_warning(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
754 Event
*event
= userdata
;
757 assert(event
->worker
);
759 log_device_warning(event
->dev
, "Worker ["PID_FMT
"] processing SEQNUM=%"PRIu64
" is taking a long time", event
->worker
->pid
, event
->seqnum
);
764 static void worker_attach_event(Worker
*worker
, Event
*event
) {
768 assert(worker
->manager
);
770 assert(!event
->worker
);
771 assert(!worker
->event
);
773 worker
->state
= WORKER_RUNNING
;
774 worker
->event
= event
;
775 event
->state
= EVENT_RUNNING
;
776 event
->worker
= worker
;
778 e
= worker
->manager
->event
;
780 (void) sd_event_add_time_relative(e
, &event
->timeout_warning_event
, CLOCK_MONOTONIC
,
781 udev_warn_timeout(arg_event_timeout_usec
), USEC_PER_SEC
,
782 on_event_timeout_warning
, event
);
784 (void) sd_event_add_time_relative(e
, &event
->timeout_event
, CLOCK_MONOTONIC
,
785 arg_event_timeout_usec
, USEC_PER_SEC
,
786 on_event_timeout
, event
);
789 static int worker_spawn(Manager
*manager
, Event
*event
) {
790 _cleanup_(sd_device_monitor_unrefp
) sd_device_monitor
*worker_monitor
= NULL
;
795 /* listen for new events */
796 r
= device_monitor_new_full(&worker_monitor
, MONITOR_GROUP_NONE
, -1);
800 /* allow the main daemon netlink address to send devices to the worker */
801 r
= device_monitor_allow_unicast_sender(worker_monitor
, manager
->monitor
);
803 return log_error_errno(r
, "Worker: Failed to set unicast sender: %m");
805 r
= device_monitor_enable_receiving(worker_monitor
);
807 return log_error_errno(r
, "Worker: Failed to enable receiving of device: %m");
809 r
= safe_fork(NULL
, FORK_DEATHSIG
, &pid
);
811 event
->state
= EVENT_QUEUED
;
812 return log_error_errno(r
, "Failed to fork() worker: %m");
815 DEVICE_TRACE_POINT(worker_spawned
, event
->dev
, getpid());
818 r
= worker_main(manager
, worker_monitor
, sd_device_ref(event
->dev
));
820 _exit(r
< 0 ? EXIT_FAILURE
: EXIT_SUCCESS
);
823 r
= worker_new(&worker
, manager
, worker_monitor
, pid
);
825 return log_error_errno(r
, "Failed to create worker object: %m");
827 worker_attach_event(worker
, event
);
829 log_device_debug(event
->dev
, "Worker ["PID_FMT
"] is forked for processing SEQNUM=%"PRIu64
".", pid
, event
->seqnum
);
833 static int event_run(Event
*event
) {
834 static bool log_children_max_reached
= true;
840 assert(event
->manager
);
842 log_device_uevent(event
->dev
, "Device ready for processing");
844 manager
= event
->manager
;
845 HASHMAP_FOREACH(worker
, manager
->workers
) {
846 if (worker
->state
!= WORKER_IDLE
)
849 r
= device_monitor_send_device(manager
->monitor
, worker
->monitor
, event
->dev
);
851 log_device_error_errno(event
->dev
, r
, "Worker ["PID_FMT
"] did not accept message, killing the worker: %m",
853 (void) kill(worker
->pid
, SIGKILL
);
854 worker
->state
= WORKER_KILLED
;
857 worker_attach_event(worker
, event
);
858 return 1; /* event is now processing. */
861 if (hashmap_size(manager
->workers
) >= arg_children_max
) {
862 /* Avoid spamming the debug logs if the limit is already reached and
863 * many events still need to be processed */
864 if (log_children_max_reached
&& arg_children_max
> 1) {
865 log_debug("Maximum number (%u) of children reached.", hashmap_size(manager
->workers
));
866 log_children_max_reached
= false;
868 return 0; /* no free worker */
871 /* Re-enable the debug message for the next batch of events */
872 log_children_max_reached
= true;
874 /* start new worker and pass initial device */
875 r
= worker_spawn(manager
, event
);
879 return 1; /* event is now processing. */
882 static int event_is_blocked(Event
*event
) {
883 Event
*loop_event
= NULL
;
886 /* lookup event for identical, parent, child device */
889 assert(event
->manager
);
890 assert(event
->blocker_seqnum
<= event
->seqnum
);
892 if (event
->retry_again_next_usec
> 0) {
895 r
= sd_event_now(event
->manager
->event
, CLOCK_BOOTTIME
, &now_usec
);
899 if (event
->retry_again_next_usec
<= now_usec
)
903 if (event
->blocker_seqnum
== event
->seqnum
)
904 /* we have checked previously and no blocker found */
907 LIST_FOREACH(event
, e
, event
->manager
->events
) {
910 /* we already found a later event, earlier cannot block us, no need to check again */
911 if (loop_event
->seqnum
< event
->blocker_seqnum
)
914 /* event we checked earlier still exists, no need to check again */
915 if (loop_event
->seqnum
== event
->blocker_seqnum
)
918 /* found ourself, no later event can block us */
919 if (loop_event
->seqnum
>= event
->seqnum
)
922 /* found event we have not checked */
927 assert(loop_event
->seqnum
> event
->blocker_seqnum
&&
928 loop_event
->seqnum
< event
->seqnum
);
930 /* check if queue contains events we depend on */
931 LIST_FOREACH(event
, e
, loop_event
) {
934 /* found ourself, no later event can block us */
935 if (loop_event
->seqnum
>= event
->seqnum
)
938 if (streq_ptr(loop_event
->id
, event
->id
))
941 if (devpath_conflict(event
->devpath
, loop_event
->devpath
) ||
942 devpath_conflict(event
->devpath
, loop_event
->devpath_old
) ||
943 devpath_conflict(event
->devpath_old
, loop_event
->devpath
))
946 if (event
->devnode
&& streq_ptr(event
->devnode
, loop_event
->devnode
))
952 log_device_debug(event
->dev
, "SEQNUM=%" PRIu64
" blocked by SEQNUM=%" PRIu64
,
953 event
->seqnum
, loop_event
->seqnum
);
955 event
->blocker_seqnum
= loop_event
->seqnum
;
959 event
->blocker_seqnum
= event
->seqnum
;
963 static int event_queue_start(Manager
*manager
) {
968 if (!manager
->events
|| manager
->exit
|| manager
->stop_exec_queue
)
971 r
= event_source_disable(manager
->kill_workers_event
);
973 log_warning_errno(r
, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
975 manager_reload(manager
, /* force = */ false);
977 LIST_FOREACH(event
, event
, manager
->events
) {
978 if (event
->state
!= EVENT_QUEUED
)
981 /* do not start event if parent or child event is still running or queued */
982 r
= event_is_blocked(event
);
986 log_device_warning_errno(event
->dev
, r
,
987 "Failed to check dependencies for event (SEQNUM=%"PRIu64
", ACTION=%s), "
988 "assuming there is no blocking event, ignoring: %m",
990 strna(device_action_to_string(event
->action
)));
992 r
= event_run(event
);
993 if (r
<= 0) /* 0 means there are no idle workers. Let's escape from the loop. */
1000 static int event_requeue(Event
*event
) {
1005 assert(event
->manager
);
1006 assert(event
->manager
->event
);
1008 event
->timeout_warning_event
= sd_event_source_disable_unref(event
->timeout_warning_event
);
1009 event
->timeout_event
= sd_event_source_disable_unref(event
->timeout_event
);
1011 /* add a short delay to suppress busy loop */
1012 r
= sd_event_now(event
->manager
->event
, CLOCK_BOOTTIME
, &now_usec
);
1014 return log_device_warning_errno(event
->dev
, r
,
1015 "Failed to get current time, "
1016 "skipping event (SEQNUM=%"PRIu64
", ACTION=%s): %m",
1017 event
->seqnum
, strna(device_action_to_string(event
->action
)));
1019 if (event
->retry_again_timeout_usec
> 0 && event
->retry_again_timeout_usec
<= now_usec
)
1020 return log_device_warning_errno(event
->dev
, SYNTHETIC_ERRNO(ETIMEDOUT
),
1021 "The underlying block device is locked by a process more than %s, "
1022 "skipping event (SEQNUM=%"PRIu64
", ACTION=%s).",
1023 FORMAT_TIMESPAN(EVENT_RETRY_TIMEOUT_USEC
, USEC_PER_MINUTE
),
1024 event
->seqnum
, strna(device_action_to_string(event
->action
)));
1026 event
->retry_again_next_usec
= usec_add(now_usec
, EVENT_RETRY_INTERVAL_USEC
);
1027 if (event
->retry_again_timeout_usec
== 0)
1028 event
->retry_again_timeout_usec
= usec_add(now_usec
, EVENT_RETRY_TIMEOUT_USEC
);
1030 if (event
->worker
&& event
->worker
->event
== event
)
1031 event
->worker
->event
= NULL
;
1032 event
->worker
= NULL
;
1034 event
->state
= EVENT_QUEUED
;
1038 static int event_queue_assume_block_device_unlocked(Manager
*manager
, sd_device
*dev
) {
1039 const char *devname
;
1042 /* When a new event for a block device is queued or we get an inotify event, assume that the
1043 * device is not locked anymore. The assumption may not be true, but that should not cause any
1044 * issues, as in that case events will be requeued soon. */
1046 r
= device_get_whole_disk(dev
, NULL
, &devname
);
1050 LIST_FOREACH(event
, event
, manager
->events
) {
1051 const char *event_devname
;
1053 if (event
->state
!= EVENT_QUEUED
)
1056 if (event
->retry_again_next_usec
== 0)
1059 if (device_get_whole_disk(event
->dev
, NULL
, &event_devname
) <= 0)
1062 if (!streq(devname
, event_devname
))
1065 event
->retry_again_next_usec
= 0;
1071 static int event_queue_insert(Manager
*manager
, sd_device
*dev
) {
1072 const char *devpath
, *devpath_old
= NULL
, *id
= NULL
, *devnode
= NULL
;
1073 sd_device_action_t action
;
1081 /* only one process can add events to the queue */
1082 assert(manager
->pid
== getpid_cached());
1084 /* We only accepts devices received by device monitor. */
1085 r
= sd_device_get_seqnum(dev
, &seqnum
);
1089 r
= sd_device_get_action(dev
, &action
);
1093 r
= sd_device_get_devpath(dev
, &devpath
);
1097 r
= sd_device_get_property_value(dev
, "DEVPATH_OLD", &devpath_old
);
1098 if (r
< 0 && r
!= -ENOENT
)
1101 r
= device_get_device_id(dev
, &id
);
1102 if (r
< 0 && r
!= -ENOENT
)
1105 r
= sd_device_get_devname(dev
, &devnode
);
1106 if (r
< 0 && r
!= -ENOENT
)
1109 event
= new(Event
, 1);
1115 .dev
= sd_device_ref(dev
),
1120 .devpath_old
= devpath_old
,
1122 .state
= EVENT_QUEUED
,
1125 if (!manager
->events
) {
1126 r
= touch("/run/udev/queue");
1128 log_warning_errno(r
, "Failed to touch /run/udev/queue, ignoring: %m");
1131 LIST_APPEND(event
, manager
->events
, event
);
1133 log_device_uevent(dev
, "Device is queued");
1138 static int on_uevent(sd_device_monitor
*monitor
, sd_device
*dev
, void *userdata
) {
1139 Manager
*manager
= userdata
;
1144 DEVICE_TRACE_POINT(kernel_uevent_received
, dev
);
1146 device_ensure_usec_initialized(dev
, NULL
);
1148 r
= event_queue_insert(manager
, dev
);
1150 log_device_error_errno(dev
, r
, "Failed to insert device into event queue: %m");
1154 (void) event_queue_assume_block_device_unlocked(manager
, dev
);
1156 /* we have fresh events, try to schedule them */
1157 event_queue_start(manager
);
1162 static int on_worker(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
1163 Manager
*manager
= userdata
;
1169 struct iovec iovec
= IOVEC_MAKE(&result
, sizeof(result
));
1170 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
))) control
;
1171 struct msghdr msghdr
= {
1174 .msg_control
= &control
,
1175 .msg_controllen
= sizeof(control
),
1178 struct ucred
*ucred
;
1181 size
= recvmsg_safe(fd
, &msghdr
, MSG_DONTWAIT
);
1184 if (size
== -EAGAIN
)
1185 /* nothing more to read */
1188 return log_error_errno(size
, "Failed to receive message: %m");
1190 cmsg_close_all(&msghdr
);
1192 if (size
!= sizeof(result
)) {
1193 log_warning("Ignoring worker message with invalid size %zi bytes", size
);
1197 ucred
= CMSG_FIND_DATA(&msghdr
, SOL_SOCKET
, SCM_CREDENTIALS
, struct ucred
);
1198 if (!ucred
|| ucred
->pid
<= 0) {
1199 log_warning("Ignoring worker message without valid PID");
1203 /* lookup worker who sent the signal */
1204 worker
= hashmap_get(manager
->workers
, PID_TO_PTR(ucred
->pid
));
1206 log_debug("Worker ["PID_FMT
"] returned, but is no longer tracked", ucred
->pid
);
1210 if (worker
->state
== WORKER_KILLING
) {
1211 worker
->state
= WORKER_KILLED
;
1212 (void) kill(worker
->pid
, SIGTERM
);
1213 } else if (worker
->state
!= WORKER_KILLED
)
1214 worker
->state
= WORKER_IDLE
;
1216 /* worker returned */
1217 if (result
== EVENT_RESULT_TRY_AGAIN
&&
1218 event_requeue(worker
->event
) < 0)
1219 device_broadcast(manager
->monitor
, worker
->event
->dev
, -ETIMEDOUT
);
1221 /* When event_requeue() succeeds, worker->event is NULL, and event_free() handles NULL gracefully. */
1222 event_free(worker
->event
);
1225 /* we have free workers, try to schedule events */
1226 event_queue_start(manager
);
1231 /* receive the udevd message from userspace */
1232 static int on_ctrl_msg(UdevCtrl
*uctrl
, UdevCtrlMessageType type
, const UdevCtrlMessageValue
*value
, void *userdata
) {
1233 Manager
*manager
= userdata
;
1240 case UDEV_CTRL_SET_LOG_LEVEL
:
1241 log_debug("Received udev control message (SET_LOG_LEVEL), setting log_level=%i", value
->intval
);
1242 log_set_max_level(value
->intval
);
1243 manager
->log_level
= value
->intval
;
1244 manager_kill_workers(manager
, false);
1246 case UDEV_CTRL_STOP_EXEC_QUEUE
:
1247 log_debug("Received udev control message (STOP_EXEC_QUEUE)");
1248 manager
->stop_exec_queue
= true;
1250 case UDEV_CTRL_START_EXEC_QUEUE
:
1251 log_debug("Received udev control message (START_EXEC_QUEUE)");
1252 manager
->stop_exec_queue
= false;
1253 /* It is not necessary to call event_queue_start() here, as it will be called in on_post() if necessary. */
1255 case UDEV_CTRL_RELOAD
:
1256 log_debug("Received udev control message (RELOAD)");
1257 manager_reload(manager
, /* force = */ true);
1259 case UDEV_CTRL_SET_ENV
: {
1260 _unused_ _cleanup_free_
char *old_val
= NULL
;
1261 _cleanup_free_
char *key
= NULL
, *val
= NULL
, *old_key
= NULL
;
1264 eq
= strchr(value
->buf
, '=');
1266 log_error("Invalid key format '%s'", value
->buf
);
1270 key
= strndup(value
->buf
, eq
- value
->buf
);
1276 old_val
= hashmap_remove2(manager
->properties
, key
, (void **) &old_key
);
1278 r
= hashmap_ensure_allocated(&manager
->properties
, &string_hash_ops
);
1286 log_debug("Received udev control message (ENV), unsetting '%s'", key
);
1288 r
= hashmap_put(manager
->properties
, key
, NULL
);
1300 log_debug("Received udev control message (ENV), setting '%s=%s'", key
, val
);
1302 r
= hashmap_put(manager
->properties
, key
, val
);
1310 manager_kill_workers(manager
, false);
1313 case UDEV_CTRL_SET_CHILDREN_MAX
:
1314 if (value
->intval
<= 0) {
1315 log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value
->intval
);
1319 log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value
->intval
);
1320 arg_children_max
= value
->intval
;
1324 case UDEV_CTRL_PING
:
1325 log_debug("Received udev control message (PING)");
1327 case UDEV_CTRL_EXIT
:
1328 log_debug("Received udev control message (EXIT)");
1329 manager_exit(manager
);
1332 log_debug("Received unknown udev control message, ignoring");
1338 static int synthesize_change_one(sd_device
*dev
, sd_device
*target
) {
1341 if (DEBUG_LOGGING
) {
1342 const char *syspath
= NULL
;
1343 (void) sd_device_get_syspath(target
, &syspath
);
1344 log_device_debug(dev
, "device is closed, synthesising 'change' on %s", strna(syspath
));
1347 r
= sd_device_trigger(target
, SD_DEVICE_CHANGE
);
1349 return log_device_debug_errno(target
, r
, "Failed to trigger 'change' uevent: %m");
1351 DEVICE_TRACE_POINT(synthetic_change_event
, dev
);
1356 static int synthesize_change(sd_device
*dev
) {
1357 const char *subsystem
, *sysname
, *devtype
;
1360 r
= sd_device_get_subsystem(dev
, &subsystem
);
1364 r
= sd_device_get_devtype(dev
, &devtype
);
1368 r
= sd_device_get_sysname(dev
, &sysname
);
1372 if (streq_ptr(subsystem
, "block") &&
1373 streq_ptr(devtype
, "disk") &&
1374 !startswith(sysname
, "dm-")) {
1375 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
1376 bool part_table_read
= false, has_partitions
= false;
1380 /* Try to re-read the partition table. This only succeeds if none of the devices is
1381 * busy. The kernel returns 0 if no partition table is found, and we will not get an
1382 * event for the disk. */
1383 fd
= sd_device_open(dev
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
);
1385 r
= flock(fd
, LOCK_EX
|LOCK_NB
);
1387 r
= ioctl(fd
, BLKRRPART
, 0);
1391 part_table_read
= true;
1394 /* search for partitions */
1395 r
= sd_device_enumerator_new(&e
);
1399 r
= sd_device_enumerator_allow_uninitialized(e
);
1403 r
= sd_device_enumerator_add_match_parent(e
, dev
);
1407 r
= sd_device_enumerator_add_match_subsystem(e
, "block", true);
1411 FOREACH_DEVICE(e
, d
) {
1414 if (sd_device_get_devtype(d
, &t
) < 0 || !streq(t
, "partition"))
1417 has_partitions
= true;
1421 /* We have partitions and re-read the table, the kernel already sent out a "change"
1422 * event for the disk, and "remove/add" for all partitions. */
1423 if (part_table_read
&& has_partitions
)
1426 /* We have partitions but re-reading the partition table did not work, synthesize
1427 * "change" for the disk and all partitions. */
1428 (void) synthesize_change_one(dev
, dev
);
1430 FOREACH_DEVICE(e
, d
) {
1433 if (sd_device_get_devtype(d
, &t
) < 0 || !streq(t
, "partition"))
1436 (void) synthesize_change_one(dev
, d
);
1440 (void) synthesize_change_one(dev
, dev
);
1445 static int on_inotify(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
1446 Manager
*manager
= userdata
;
1447 union inotify_event_buffer buffer
;
1453 r
= event_source_disable(manager
->kill_workers_event
);
1455 log_warning_errno(r
, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1457 l
= read(fd
, &buffer
, sizeof(buffer
));
1459 if (ERRNO_IS_TRANSIENT(errno
))
1462 return log_error_errno(errno
, "Failed to read inotify fd: %m");
1465 FOREACH_INOTIFY_EVENT_WARN(e
, buffer
, l
) {
1466 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
1467 const char *devnode
;
1469 r
= device_new_from_watch_handle(&dev
, e
->wd
);
1471 log_debug_errno(r
, "Failed to create sd_device object from watch handle, ignoring: %m");
1475 if (sd_device_get_devname(dev
, &devnode
) < 0)
1478 log_device_debug(dev
, "Inotify event: %x for %s", e
->mask
, devnode
);
1479 if (e
->mask
& IN_CLOSE_WRITE
) {
1480 (void) event_queue_assume_block_device_unlocked(manager
, dev
);
1481 (void) synthesize_change(dev
);
1484 /* Do not handle IN_IGNORED here. It should be handled by worker in 'remove' uevent;
1485 * udev_event_execute_rules() -> event_execute_rules_on_remove() -> udev_watch_end(). */
1491 static int on_sigterm(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
1492 Manager
*manager
= userdata
;
1496 manager_exit(manager
);
1501 static int on_sighup(sd_event_source
*s
, const struct signalfd_siginfo
*si
, void *userdata
) {
1502 Manager
*manager
= userdata
;
1506 manager_reload(manager
, /* force = */ true);
1511 static int on_sigchld(sd_event_source
*s
, const siginfo_t
*si
, void *userdata
) {
1512 Worker
*worker
= ASSERT_PTR(userdata
);
1513 Manager
*manager
= ASSERT_PTR(worker
->manager
);
1514 sd_device
*dev
= worker
->event
? ASSERT_PTR(worker
->event
->dev
) : NULL
;
1520 switch (si
->si_code
) {
1522 if (si
->si_status
== 0)
1523 log_device_debug(dev
, "Worker ["PID_FMT
"] exited.", si
->si_pid
);
1525 log_device_warning(dev
, "Worker ["PID_FMT
"] exited with return code %i.",
1526 si
->si_pid
, si
->si_status
);
1527 result
= EVENT_RESULT_EXIT_STATUS_BASE
+ si
->si_status
;
1532 log_device_warning(dev
, "Worker ["PID_FMT
"] terminated by signal %i (%s).",
1533 si
->si_pid
, si
->si_status
, signal_to_string(si
->si_status
));
1534 result
= EVENT_RESULT_SIGNAL_BASE
+ si
->si_status
;
1538 assert_not_reached();
1541 if (result
!= EVENT_RESULT_SUCCESS
&& dev
) {
1542 /* delete state from disk */
1543 device_delete_db(dev
);
1544 device_tag_index(dev
, NULL
, false);
1546 /* Forward kernel event to libudev listeners */
1547 device_broadcast(manager
->monitor
, dev
, result
);
1550 worker_free(worker
);
1552 /* we can start new workers, try to schedule events */
1553 event_queue_start(manager
);
1555 /* Disable unnecessary cleanup event */
1556 if (hashmap_isempty(manager
->workers
)) {
1557 r
= event_source_disable(manager
->kill_workers_event
);
1559 log_warning_errno(r
, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1565 static int on_post(sd_event_source
*s
, void *userdata
) {
1566 Manager
*manager
= userdata
;
1570 if (manager
->events
) {
1571 /* Try to process pending events if idle workers exist. Why is this necessary?
1572 * When a worker finished an event and became idle, even if there was a pending event,
1573 * the corresponding device might have been locked and the processing of the event
1574 * delayed for a while, preventing the worker from processing the event immediately.
1575 * Now, the device may be unlocked. Let's try again! */
1576 event_queue_start(manager
);
1580 /* There are no queued events. Let's remove /run/udev/queue and clean up the idle processes. */
1582 if (unlink("/run/udev/queue") < 0) {
1583 if (errno
!= ENOENT
)
1584 log_warning_errno(errno
, "Failed to unlink /run/udev/queue, ignoring: %m");
1586 log_debug("No events are queued, removing /run/udev/queue.");
1588 if (!hashmap_isempty(manager
->workers
)) {
1589 /* There are idle workers */
1590 (void) event_reset_time_relative(manager
->event
, &manager
->kill_workers_event
,
1591 CLOCK_MONOTONIC
, 3 * USEC_PER_SEC
, USEC_PER_SEC
,
1592 on_kill_workers_event
, manager
,
1593 0, "kill-workers-event", false);
1597 /* There are no idle workers. */
1600 return sd_event_exit(manager
->event
, 0);
1602 if (manager
->cgroup
)
1603 /* cleanup possible left-over processes in our cgroup */
1604 (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER
, manager
->cgroup
, SIGKILL
, CGROUP_IGNORE_SELF
, NULL
, NULL
, NULL
);
1609 static int listen_fds(int *ret_ctrl
, int *ret_netlink
) {
1610 int ctrl_fd
= -1, netlink_fd
= -1;
1614 assert(ret_netlink
);
1616 n
= sd_listen_fds(true);
1620 for (fd
= SD_LISTEN_FDS_START
; fd
< n
+ SD_LISTEN_FDS_START
; fd
++) {
1621 if (sd_is_socket(fd
, AF_UNIX
, SOCK_SEQPACKET
, -1) > 0) {
1628 if (sd_is_socket(fd
, AF_NETLINK
, SOCK_RAW
, -1) > 0) {
1629 if (netlink_fd
>= 0)
1638 *ret_ctrl
= ctrl_fd
;
1639 *ret_netlink
= netlink_fd
;
1645 * read the kernel command line, in case we need to get into debug mode
1646 * udev.log_level=<level> syslog priority
1647 * udev.children_max=<number of workers> events are fully serialized if set to 1
1648 * udev.exec_delay=<number of seconds> delay execution of every executed program
1649 * udev.event_timeout=<number of seconds> seconds to wait before terminating an event
1650 * udev.blockdev_read_only<=bool> mark all block devices read-only when they appear
1652 static int parse_proc_cmdline_item(const char *key
, const char *value
, void *data
) {
1657 if (proc_cmdline_key_streq(key
, "udev.log_level") ||
1658 proc_cmdline_key_streq(key
, "udev.log_priority")) { /* kept for backward compatibility */
1660 if (proc_cmdline_value_missing(key
, value
))
1663 r
= log_level_from_string(value
);
1665 log_set_max_level(r
);
1667 } else if (proc_cmdline_key_streq(key
, "udev.event_timeout")) {
1669 if (proc_cmdline_value_missing(key
, value
))
1672 r
= parse_sec(value
, &arg_event_timeout_usec
);
1674 } else if (proc_cmdline_key_streq(key
, "udev.children_max")) {
1676 if (proc_cmdline_value_missing(key
, value
))
1679 r
= safe_atou(value
, &arg_children_max
);
1681 } else if (proc_cmdline_key_streq(key
, "udev.exec_delay")) {
1683 if (proc_cmdline_value_missing(key
, value
))
1686 r
= parse_sec(value
, &arg_exec_delay_usec
);
1688 } else if (proc_cmdline_key_streq(key
, "udev.timeout_signal")) {
1690 if (proc_cmdline_value_missing(key
, value
))
1693 r
= signal_from_string(value
);
1695 arg_timeout_signal
= r
;
1697 } else if (proc_cmdline_key_streq(key
, "udev.blockdev_read_only")) {
1700 arg_blockdev_read_only
= true;
1702 r
= parse_boolean(value
);
1704 log_warning_errno(r
, "Failed to parse udev.blockdev-read-only argument, ignoring: %s", value
);
1706 arg_blockdev_read_only
= r
;
1709 if (arg_blockdev_read_only
)
1710 log_notice("All physical block devices will be marked read-only.");
1715 if (startswith(key
, "udev."))
1716 log_warning("Unknown udev kernel command line option \"%s\", ignoring.", key
);
1722 log_warning_errno(r
, "Failed to parse \"%s=%s\", ignoring: %m", key
, value
);
1727 static int help(void) {
1728 _cleanup_free_
char *link
= NULL
;
1731 r
= terminal_urlify_man("systemd-udevd.service", "8", &link
);
1735 printf("%s [OPTIONS...]\n\n"
1736 "Rule-based manager for device events and files.\n\n"
1737 " -h --help Print this message\n"
1738 " -V --version Print version of the program\n"
1739 " -d --daemon Detach and run in the background\n"
1740 " -D --debug Enable debug output\n"
1741 " -c --children-max=INT Set maximum number of workers\n"
1742 " -e --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
1743 " -t --event-timeout=SECONDS Seconds to wait before terminating an event\n"
1744 " -N --resolve-names=early|late|never\n"
1745 " When to resolve users and groups\n"
1746 "\nSee the %s for details.\n",
1747 program_invocation_short_name
,
1753 static int parse_argv(int argc
, char *argv
[]) {
1758 static const struct option options
[] = {
1759 { "daemon", no_argument
, NULL
, 'd' },
1760 { "debug", no_argument
, NULL
, 'D' },
1761 { "children-max", required_argument
, NULL
, 'c' },
1762 { "exec-delay", required_argument
, NULL
, 'e' },
1763 { "event-timeout", required_argument
, NULL
, 't' },
1764 { "resolve-names", required_argument
, NULL
, 'N' },
1765 { "help", no_argument
, NULL
, 'h' },
1766 { "version", no_argument
, NULL
, 'V' },
1767 { "timeout-signal", required_argument
, NULL
, ARG_TIMEOUT_SIGNAL
},
1776 while ((c
= getopt_long(argc
, argv
, "c:de:Dt:N:hV", options
, NULL
)) >= 0) {
1780 arg_daemonize
= true;
1783 r
= safe_atou(optarg
, &arg_children_max
);
1785 log_warning_errno(r
, "Failed to parse --children-max= value '%s', ignoring: %m", optarg
);
1788 r
= parse_sec(optarg
, &arg_exec_delay_usec
);
1790 log_warning_errno(r
, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg
);
1792 case ARG_TIMEOUT_SIGNAL
:
1793 r
= signal_from_string(optarg
);
1795 log_warning_errno(r
, "Failed to parse --timeout-signal= value '%s', ignoring: %m", optarg
);
1797 arg_timeout_signal
= r
;
1801 r
= parse_sec(optarg
, &arg_event_timeout_usec
);
1803 log_warning_errno(r
, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg
);
1809 ResolveNameTiming t
;
1811 t
= resolve_name_timing_from_string(optarg
);
1813 log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg
);
1815 arg_resolve_name_timing
= t
;
1821 printf("%s\n", GIT_VERSION
);
1826 assert_not_reached();
1834 static int create_subcgroup(char **ret
) {
1835 _cleanup_free_
char *cgroup
= NULL
, *subcgroup
= NULL
;
1839 return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "Not invoked by PID1.");
1843 return log_debug_errno(r
, "Failed to check if systemd is running: %m");
1845 return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "systemd is not running.");
1847 /* Get our own cgroup, we regularly kill everything udev has left behind.
1848 * We only do this on systemd systems, and only if we are directly spawned
1849 * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */
1851 r
= cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &cgroup
);
1853 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
1854 return log_debug_errno(r
, "Dedicated cgroup not found: %m");
1855 return log_debug_errno(r
, "Failed to get cgroup: %m");
1858 r
= cg_get_xattr_bool(SYSTEMD_CGROUP_CONTROLLER
, cgroup
, "trusted.delegate");
1859 if (IN_SET(r
, 0, -ENODATA
))
1860 return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "The cgroup %s is not delegated to us.", cgroup
);
1862 return log_debug_errno(r
, "Failed to read trusted.delegate attribute: %m");
1864 /* We are invoked with our own delegated cgroup tree, let's move us one level down, so that we
1865 * don't collide with the "no processes in inner nodes" rule of cgroups, when the service
1866 * manager invokes the ExecReload= job in the .control/ subcgroup. */
1868 subcgroup
= path_join(cgroup
, "/udev");
1870 return log_oom_debug();
1872 r
= cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER
, subcgroup
, 0);
1874 return log_debug_errno(r
, "Failed to create %s subcgroup: %m", subcgroup
);
1876 log_debug("Created %s subcgroup.", subcgroup
);
1878 *ret
= TAKE_PTR(subcgroup
);
1882 static int manager_new(Manager
**ret
, int fd_ctrl
, int fd_uevent
) {
1883 _cleanup_(manager_freep
) Manager
*manager
= NULL
;
1884 _cleanup_free_
char *cgroup
= NULL
;
1889 (void) create_subcgroup(&cgroup
);
1891 manager
= new(Manager
, 1);
1895 *manager
= (Manager
) {
1897 .worker_watch
= { -1, -1 },
1898 .cgroup
= TAKE_PTR(cgroup
),
1901 r
= udev_ctrl_new_from_fd(&manager
->ctrl
, fd_ctrl
);
1903 return log_error_errno(r
, "Failed to initialize udev control socket: %m");
1905 r
= udev_ctrl_enable_receiving(manager
->ctrl
);
1907 return log_error_errno(r
, "Failed to bind udev control socket: %m");
1909 r
= device_monitor_new_full(&manager
->monitor
, MONITOR_GROUP_KERNEL
, fd_uevent
);
1911 return log_error_errno(r
, "Failed to initialize device monitor: %m");
1913 /* Bump receiver buffer, but only if we are not called via socket activation, as in that
1914 * case systemd sets the receive buffer size for us, and the value in the .socket unit
1915 * should take full effect. */
1916 if (fd_uevent
< 0) {
1917 r
= sd_device_monitor_set_receive_buffer_size(manager
->monitor
, 128 * 1024 * 1024);
1919 log_warning_errno(r
, "Failed to set receive buffer size for device monitor, ignoring: %m");
1922 r
= device_monitor_enable_receiving(manager
->monitor
);
1924 return log_error_errno(r
, "Failed to bind netlink socket: %m");
1926 manager
->log_level
= log_get_max_level();
1928 *ret
= TAKE_PTR(manager
);
1933 static int main_loop(Manager
*manager
) {
1936 manager
->pid
= getpid_cached();
1938 /* unnamed socket from workers to the main daemon */
1939 r
= socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, manager
->worker_watch
);
1941 return log_error_errno(errno
, "Failed to create socketpair for communicating with workers: %m");
1943 fd_worker
= manager
->worker_watch
[READ_END
];
1945 r
= setsockopt_int(fd_worker
, SOL_SOCKET
, SO_PASSCRED
, true);
1947 return log_error_errno(r
, "Failed to enable SO_PASSCRED: %m");
1949 manager
->inotify_fd
= inotify_init1(IN_CLOEXEC
);
1950 if (manager
->inotify_fd
< 0)
1951 return log_error_errno(errno
, "Failed to create inotify descriptor: %m");
1953 udev_watch_restore(manager
->inotify_fd
);
1955 /* block and listen to all signals on signalfd */
1956 assert_se(sigprocmask_many(SIG_BLOCK
, NULL
, SIGTERM
, SIGINT
, SIGHUP
, SIGCHLD
, -1) >= 0);
1958 r
= sd_event_default(&manager
->event
);
1960 return log_error_errno(r
, "Failed to allocate event loop: %m");
1962 r
= sd_event_add_signal(manager
->event
, NULL
, SIGINT
, on_sigterm
, manager
);
1964 return log_error_errno(r
, "Failed to create SIGINT event source: %m");
1966 r
= sd_event_add_signal(manager
->event
, NULL
, SIGTERM
, on_sigterm
, manager
);
1968 return log_error_errno(r
, "Failed to create SIGTERM event source: %m");
1970 r
= sd_event_add_signal(manager
->event
, NULL
, SIGHUP
, on_sighup
, manager
);
1972 return log_error_errno(r
, "Failed to create SIGHUP event source: %m");
1974 r
= sd_event_set_watchdog(manager
->event
, true);
1976 return log_error_errno(r
, "Failed to create watchdog event source: %m");
1978 r
= udev_ctrl_attach_event(manager
->ctrl
, manager
->event
);
1980 return log_error_errno(r
, "Failed to attach event to udev control: %m");
1982 r
= udev_ctrl_start(manager
->ctrl
, on_ctrl_msg
, manager
);
1984 return log_error_errno(r
, "Failed to start device monitor: %m");
1986 /* This needs to be after the inotify and uevent handling, to make sure
1987 * that the ping is send back after fully processing the pending uevents
1988 * (including the synthetic ones we may create due to inotify events).
1990 r
= sd_event_source_set_priority(udev_ctrl_get_event_source(manager
->ctrl
), SD_EVENT_PRIORITY_IDLE
);
1992 return log_error_errno(r
, "Failed to set IDLE event priority for udev control event source: %m");
1994 r
= sd_event_add_io(manager
->event
, &manager
->inotify_event
, manager
->inotify_fd
, EPOLLIN
, on_inotify
, manager
);
1996 return log_error_errno(r
, "Failed to create inotify event source: %m");
1998 r
= sd_device_monitor_attach_event(manager
->monitor
, manager
->event
);
2000 return log_error_errno(r
, "Failed to attach event to device monitor: %m");
2002 r
= sd_device_monitor_start(manager
->monitor
, on_uevent
, manager
);
2004 return log_error_errno(r
, "Failed to start device monitor: %m");
2006 (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager
->monitor
), "device-monitor");
2008 r
= sd_event_add_io(manager
->event
, NULL
, fd_worker
, EPOLLIN
, on_worker
, manager
);
2010 return log_error_errno(r
, "Failed to create worker event source: %m");
2012 r
= sd_event_add_post(manager
->event
, NULL
, on_post
, manager
);
2014 return log_error_errno(r
, "Failed to create post event source: %m");
2016 manager
->last_usec
= now(CLOCK_MONOTONIC
);
2018 udev_builtin_init();
2020 r
= udev_rules_load(&manager
->rules
, arg_resolve_name_timing
);
2022 return log_error_errno(r
, "Failed to read udev rules: %m");
2024 r
= udev_rules_apply_static_dev_perms(manager
->rules
);
2026 log_warning_errno(r
, "Failed to apply permissions on static device nodes, ignoring: %m");
2030 r
= sd_event_loop(manager
->event
);
2032 log_error_errno(r
, "Event loop failed: %m");
2036 "STATUS=Shutting down...");
2040 int run_udevd(int argc
, char *argv
[]) {
2041 _cleanup_(manager_freep
) Manager
*manager
= NULL
;
2042 int fd_ctrl
= -1, fd_uevent
= -1;
2045 log_set_target(LOG_TARGET_AUTO
);
2047 udev_parse_config_full(&arg_children_max
, &arg_exec_delay_usec
, &arg_event_timeout_usec
, &arg_resolve_name_timing
, &arg_timeout_signal
);
2048 log_parse_environment();
2049 log_open(); /* Done again to update after reading configuration. */
2051 r
= parse_argv(argc
, argv
);
2055 r
= proc_cmdline_parse(parse_proc_cmdline_item
, NULL
, PROC_CMDLINE_STRIP_RD_PREFIX
);
2057 log_warning_errno(r
, "Failed to parse kernel command line, ignoring: %m");
2060 log_set_target(LOG_TARGET_CONSOLE
);
2061 log_set_max_level(LOG_DEBUG
);
2068 if (arg_children_max
== 0) {
2069 unsigned long cpu_limit
, mem_limit
, cpu_count
= 1;
2071 r
= cpus_in_affinity_mask();
2073 log_warning_errno(r
, "Failed to determine number of local CPUs, ignoring: %m");
2077 cpu_limit
= cpu_count
* 2 + 16;
2078 mem_limit
= MAX(physical_memory() / (128UL*1024*1024), 10U);
2080 arg_children_max
= MIN(cpu_limit
, mem_limit
);
2081 arg_children_max
= MIN(WORKER_NUM_MAX
, arg_children_max
);
2083 log_debug("Set children_max to %u", arg_children_max
);
2086 /* set umask before creating any file/directory */
2089 r
= mac_selinux_init();
2093 r
= RET_NERRNO(mkdir("/run/udev", 0755));
2094 if (r
< 0 && r
!= -EEXIST
)
2095 return log_error_errno(r
, "Failed to create /run/udev: %m");
2097 r
= listen_fds(&fd_ctrl
, &fd_uevent
);
2099 return log_error_errno(r
, "Failed to listen on fds: %m");
2101 r
= manager_new(&manager
, fd_ctrl
, fd_uevent
);
2103 return log_error_errno(r
, "Failed to create manager: %m");
2105 if (arg_daemonize
) {
2108 log_info("Starting systemd-udevd version " GIT_VERSION
);
2110 /* connect /dev/null to stdin, stdout, stderr */
2111 if (log_get_max_level() < LOG_DEBUG
) {
2112 r
= make_null_stdio();
2114 log_warning_errno(r
, "Failed to redirect standard streams to /dev/null: %m");
2119 return log_error_errno(errno
, "Failed to fork daemon: %m");
2128 return main_loop(manager
);