2 * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
3 * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
4 * Copyright (C) 2009 Canonical Ltd.
5 * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
7 * This program is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program. If not, see <http://www.gnu.org/licenses/>.
36 #include <sys/prctl.h>
37 #include <sys/socket.h>
39 #include <sys/signalfd.h>
40 #include <sys/epoll.h>
41 #include <sys/mount.h>
45 #include <sys/ioctl.h>
46 #include <sys/inotify.h>
47 #include <sys/utsname.h>
50 #include "udev-util.h"
51 #include "sd-daemon.h"
52 #include "cgroup-util.h"
53 #include "dev-setup.h"
58 void udev_main_log(struct udev
*udev
, int priority
,
59 const char *file
, int line
, const char *fn
,
60 const char *format
, va_list args
) {
61 log_metav(priority
, file
, line
, fn
, format
, args
);
64 static struct udev_rules
*rules
;
65 static struct udev_ctrl
*udev_ctrl
;
66 static struct udev_monitor
*monitor
;
67 static int worker_watch
[2] = { -1, -1 };
68 static int fd_signal
= -1;
69 static int fd_ep
= -1;
70 static int fd_inotify
= -1;
71 static bool stop_exec_queue
;
74 static int children_max
;
75 static int exec_delay
;
76 static usec_t event_timeout_usec
= 60 * USEC_PER_SEC
;
77 static sigset_t sigmask_orig
;
78 static UDEV_LIST(event_list
);
79 static UDEV_LIST(worker_list
);
80 static char *udev_cgroup
;
81 static bool udev_exit
;
90 struct udev_list_node node
;
92 struct udev_device
*dev
;
93 enum event_state state
;
95 unsigned long long int delaying_seqnum
;
96 unsigned long long int seqnum
;
99 const char *devpath_old
;
105 static inline struct event
*node_to_event(struct udev_list_node
*node
) {
106 return container_of(node
, struct event
, node
);
109 static void event_queue_cleanup(struct udev
*udev
, enum event_state type
);
119 struct udev_list_node node
;
123 struct udev_monitor
*monitor
;
124 enum worker_state state
;
126 usec_t event_start_usec
;
129 /* passed from worker to main process */
130 struct worker_message
{
135 static inline struct worker
*node_to_worker(struct udev_list_node
*node
) {
136 return container_of(node
, struct worker
, node
);
139 static void event_queue_delete(struct event
*event
) {
140 udev_list_node_remove(&event
->node
);
141 udev_device_unref(event
->dev
);
145 static struct worker
*worker_ref(struct worker
*worker
) {
150 static void worker_cleanup(struct worker
*worker
) {
151 udev_list_node_remove(&worker
->node
);
152 udev_monitor_unref(worker
->monitor
);
157 static void worker_unref(struct worker
*worker
) {
159 if (worker
->refcount
> 0)
161 log_debug("worker [%u] cleaned up", worker
->pid
);
162 worker_cleanup(worker
);
165 static void worker_list_cleanup(struct udev
*udev
) {
166 struct udev_list_node
*loop
, *tmp
;
168 udev_list_node_foreach_safe(loop
, tmp
, &worker_list
) {
169 struct worker
*worker
= node_to_worker(loop
);
171 worker_cleanup(worker
);
175 static void worker_new(struct event
*event
) {
176 struct udev
*udev
= event
->udev
;
177 struct worker
*worker
;
178 struct udev_monitor
*worker_monitor
;
181 /* listen for new events */
182 worker_monitor
= udev_monitor_new_from_netlink(udev
, NULL
);
183 if (worker_monitor
== NULL
)
185 /* allow the main daemon netlink address to send devices to the worker */
186 udev_monitor_allow_unicast_sender(worker_monitor
, monitor
);
187 udev_monitor_enable_receiving(worker_monitor
);
189 worker
= new0(struct worker
, 1);
190 if (worker
== NULL
) {
191 udev_monitor_unref(worker_monitor
);
194 /* worker + event reference */
195 worker
->refcount
= 2;
201 struct udev_device
*dev
= NULL
;
203 struct epoll_event ep_signal
, ep_monitor
;
205 int rc
= EXIT_SUCCESS
;
207 /* take initial device from queue */
212 worker_list_cleanup(udev
);
213 event_queue_cleanup(udev
, EVENT_UNDEF
);
214 udev_monitor_unref(monitor
);
215 udev_ctrl_unref(udev_ctrl
);
218 close(worker_watch
[READ_END
]);
221 fd_signal
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
223 log_error("error creating signalfd %m");
228 fd_ep
= epoll_create1(EPOLL_CLOEXEC
);
230 log_error("error creating epoll fd: %m");
235 memzero(&ep_signal
, sizeof(struct epoll_event
));
236 ep_signal
.events
= EPOLLIN
;
237 ep_signal
.data
.fd
= fd_signal
;
239 fd_monitor
= udev_monitor_get_fd(worker_monitor
);
240 memzero(&ep_monitor
, sizeof(struct epoll_event
));
241 ep_monitor
.events
= EPOLLIN
;
242 ep_monitor
.data
.fd
= fd_monitor
;
244 if (epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_signal
, &ep_signal
) < 0 ||
245 epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_monitor
, &ep_monitor
) < 0) {
246 log_error("fail to add fds to epoll: %m");
251 /* request TERM signal if parent exits */
252 prctl(PR_SET_PDEATHSIG
, SIGTERM
);
254 /* reset OOM score, we only protect the main daemon */
255 write_string_file("/proc/self/oom_score_adj", "0");
258 struct udev_event
*udev_event
;
259 struct worker_message msg
;
263 log_debug("seq %llu running", udev_device_get_seqnum(dev
));
264 udev_event
= udev_event_new(dev
);
265 if (udev_event
== NULL
) {
270 /* needed for SIGCHLD/SIGTERM in spawn() */
271 udev_event
->fd_signal
= fd_signal
;
274 udev_event
->exec_delay
= exec_delay
;
277 * Take a shared lock on the device node; this establishes
278 * a concept of device "ownership" to serialize device
279 * access. External processes holding an exclusive lock will
280 * cause udev to skip the event handling; in the case udev
281 * acquired the lock, the external process can block until
282 * udev has finished its event handling.
284 if (!streq_ptr(udev_device_get_action(dev
), "remove") &&
285 streq_ptr("block", udev_device_get_subsystem(dev
)) &&
286 !startswith(udev_device_get_sysname(dev
), "dm-") &&
287 !startswith(udev_device_get_sysname(dev
), "md")) {
288 struct udev_device
*d
= dev
;
290 if (streq_ptr("partition", udev_device_get_devtype(d
)))
291 d
= udev_device_get_parent(d
);
294 fd_lock
= open(udev_device_get_devnode(d
), O_RDONLY
|O_CLOEXEC
|O_NOFOLLOW
|O_NONBLOCK
);
295 if (fd_lock
>= 0 && flock(fd_lock
, LOCK_SH
|LOCK_NB
) < 0) {
296 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d
));
298 fd_lock
= safe_close(fd_lock
);
304 /* apply rules, create node, symlinks */
305 udev_event_execute_rules(udev_event
, event_timeout_usec
, rules
, &sigmask_orig
);
307 udev_event_execute_run(udev_event
, event_timeout_usec
, &sigmask_orig
);
309 /* apply/restore inotify watch */
310 if (udev_event
->inotify_watch
) {
311 udev_watch_begin(udev
, dev
);
312 udev_device_update_db(dev
);
317 /* send processed event back to libudev listeners */
318 udev_monitor_send_device(worker_monitor
, NULL
, dev
);
321 /* send udevd the result of the event execution */
322 memzero(&msg
, sizeof(struct worker_message
));
325 send(worker_watch
[WRITE_END
], &msg
, sizeof(struct worker_message
), 0);
327 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev
), err
);
329 udev_device_unref(dev
);
332 if (udev_event
->sigterm
) {
333 udev_event_unref(udev_event
);
337 udev_event_unref(udev_event
);
339 /* wait for more device messages from main udevd, or term signal */
340 while (dev
== NULL
) {
341 struct epoll_event ev
[4];
345 fdcount
= epoll_wait(fd_ep
, ev
, ELEMENTSOF(ev
), -1);
349 log_error("failed to poll: %m");
353 for (i
= 0; i
< fdcount
; i
++) {
354 if (ev
[i
].data
.fd
== fd_monitor
&& ev
[i
].events
& EPOLLIN
) {
355 dev
= udev_monitor_receive_device(worker_monitor
);
357 } else if (ev
[i
].data
.fd
== fd_signal
&& ev
[i
].events
& EPOLLIN
) {
358 struct signalfd_siginfo fdsi
;
361 size
= read(fd_signal
, &fdsi
, sizeof(struct signalfd_siginfo
));
362 if (size
!= sizeof(struct signalfd_siginfo
))
364 switch (fdsi
.ssi_signo
) {
373 udev_device_unref(dev
);
374 safe_close(fd_signal
);
377 close(worker_watch
[WRITE_END
]);
378 udev_rules_unref(rules
);
379 udev_builtin_exit(udev
);
380 udev_monitor_unref(worker_monitor
);
386 udev_monitor_unref(worker_monitor
);
387 event
->state
= EVENT_QUEUED
;
389 log_error("fork of child failed: %m");
392 /* close monitor, but keep address around */
393 udev_monitor_disconnect(worker_monitor
);
394 worker
->monitor
= worker_monitor
;
396 worker
->state
= WORKER_RUNNING
;
397 worker
->event_start_usec
= now(CLOCK_MONOTONIC
);
398 worker
->event
= event
;
399 event
->state
= EVENT_RUNNING
;
400 udev_list_node_append(&worker
->node
, &worker_list
);
402 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event
->dev
), pid
);
407 static void event_run(struct event
*event
) {
408 struct udev_list_node
*loop
;
410 udev_list_node_foreach(loop
, &worker_list
) {
411 struct worker
*worker
= node_to_worker(loop
);
414 if (worker
->state
!= WORKER_IDLE
)
417 count
= udev_monitor_send_device(monitor
, worker
->monitor
, event
->dev
);
419 log_error("worker [%u] did not accept message %zi (%m), kill it", worker
->pid
, count
);
420 kill(worker
->pid
, SIGKILL
);
421 worker
->state
= WORKER_KILLED
;
425 worker
->event
= event
;
426 worker
->state
= WORKER_RUNNING
;
427 worker
->event_start_usec
= now(CLOCK_MONOTONIC
);
428 event
->state
= EVENT_RUNNING
;
432 if (children
>= children_max
) {
433 if (children_max
> 1)
434 log_debug("maximum number (%i) of children reached", children
);
438 /* start new worker and pass initial device */
442 static int event_queue_insert(struct udev_device
*dev
) {
445 event
= new0(struct event
, 1);
449 event
->udev
= udev_device_get_udev(dev
);
451 event
->seqnum
= udev_device_get_seqnum(dev
);
452 event
->devpath
= udev_device_get_devpath(dev
);
453 event
->devpath_len
= strlen(event
->devpath
);
454 event
->devpath_old
= udev_device_get_devpath_old(dev
);
455 event
->devnum
= udev_device_get_devnum(dev
);
456 event
->is_block
= streq("block", udev_device_get_subsystem(dev
));
457 event
->ifindex
= udev_device_get_ifindex(dev
);
459 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev
),
460 udev_device_get_action(dev
), udev_device_get_subsystem(dev
));
462 event
->state
= EVENT_QUEUED
;
463 udev_list_node_append(&event
->node
, &event_list
);
467 static void worker_kill(struct udev
*udev
) {
468 struct udev_list_node
*loop
;
470 udev_list_node_foreach(loop
, &worker_list
) {
471 struct worker
*worker
= node_to_worker(loop
);
473 if (worker
->state
== WORKER_KILLED
)
476 worker
->state
= WORKER_KILLED
;
477 kill(worker
->pid
, SIGTERM
);
481 /* lookup event for identical, parent, child device */
482 static bool is_devpath_busy(struct event
*event
) {
483 struct udev_list_node
*loop
;
486 /* check if queue contains events we depend on */
487 udev_list_node_foreach(loop
, &event_list
) {
488 struct event
*loop_event
= node_to_event(loop
);
490 /* we already found a later event, earlier can not block us, no need to check again */
491 if (loop_event
->seqnum
< event
->delaying_seqnum
)
494 /* event we checked earlier still exists, no need to check again */
495 if (loop_event
->seqnum
== event
->delaying_seqnum
)
498 /* found ourself, no later event can block us */
499 if (loop_event
->seqnum
>= event
->seqnum
)
502 /* check major/minor */
503 if (major(event
->devnum
) != 0 && event
->devnum
== loop_event
->devnum
&& event
->is_block
== loop_event
->is_block
)
506 /* check network device ifindex */
507 if (event
->ifindex
!= 0 && event
->ifindex
== loop_event
->ifindex
)
510 /* check our old name */
511 if (event
->devpath_old
!= NULL
&& streq(loop_event
->devpath
, event
->devpath_old
)) {
512 event
->delaying_seqnum
= loop_event
->seqnum
;
516 /* compare devpath */
517 common
= MIN(loop_event
->devpath_len
, event
->devpath_len
);
519 /* one devpath is contained in the other? */
520 if (memcmp(loop_event
->devpath
, event
->devpath
, common
) != 0)
523 /* identical device event found */
524 if (loop_event
->devpath_len
== event
->devpath_len
) {
525 /* devices names might have changed/swapped in the meantime */
526 if (major(event
->devnum
) != 0 && (event
->devnum
!= loop_event
->devnum
|| event
->is_block
!= loop_event
->is_block
))
528 if (event
->ifindex
!= 0 && event
->ifindex
!= loop_event
->ifindex
)
530 event
->delaying_seqnum
= loop_event
->seqnum
;
534 /* parent device event found */
535 if (event
->devpath
[common
] == '/') {
536 event
->delaying_seqnum
= loop_event
->seqnum
;
540 /* child device event found */
541 if (loop_event
->devpath
[common
] == '/') {
542 event
->delaying_seqnum
= loop_event
->seqnum
;
546 /* no matching device */
553 static void event_queue_start(struct udev
*udev
) {
554 struct udev_list_node
*loop
;
556 udev_list_node_foreach(loop
, &event_list
) {
557 struct event
*event
= node_to_event(loop
);
559 if (event
->state
!= EVENT_QUEUED
)
562 /* do not start event if parent or child event is still running */
563 if (is_devpath_busy(event
))
570 static void event_queue_cleanup(struct udev
*udev
, enum event_state match_type
) {
571 struct udev_list_node
*loop
, *tmp
;
573 udev_list_node_foreach_safe(loop
, tmp
, &event_list
) {
574 struct event
*event
= node_to_event(loop
);
576 if (match_type
!= EVENT_UNDEF
&& match_type
!= event
->state
)
579 event_queue_delete(event
);
583 static void worker_returned(int fd_worker
) {
585 struct worker_message msg
;
587 struct udev_list_node
*loop
;
589 size
= recv(fd_worker
, &msg
, sizeof(struct worker_message
), MSG_DONTWAIT
);
590 if (size
!= sizeof(struct worker_message
))
593 /* lookup worker who sent the signal */
594 udev_list_node_foreach(loop
, &worker_list
) {
595 struct worker
*worker
= node_to_worker(loop
);
597 if (worker
->pid
!= msg
.pid
)
600 /* worker returned */
602 worker
->event
->exitcode
= msg
.exitcode
;
603 event_queue_delete(worker
->event
);
604 worker
->event
= NULL
;
606 if (worker
->state
!= WORKER_KILLED
)
607 worker
->state
= WORKER_IDLE
;
608 worker_unref(worker
);
614 /* receive the udevd message from userspace */
615 static struct udev_ctrl_connection
*handle_ctrl_msg(struct udev_ctrl
*uctrl
) {
616 struct udev
*udev
= udev_ctrl_get_udev(uctrl
);
617 struct udev_ctrl_connection
*ctrl_conn
;
618 struct udev_ctrl_msg
*ctrl_msg
= NULL
;
622 ctrl_conn
= udev_ctrl_get_connection(uctrl
);
623 if (ctrl_conn
== NULL
)
626 ctrl_msg
= udev_ctrl_receive_msg(ctrl_conn
);
627 if (ctrl_msg
== NULL
)
630 i
= udev_ctrl_get_set_log_level(ctrl_msg
);
632 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i
);
633 log_set_max_level(i
);
634 udev_set_log_priority(udev
, i
);
638 if (udev_ctrl_get_stop_exec_queue(ctrl_msg
) > 0) {
639 log_debug("udevd message (STOP_EXEC_QUEUE) received");
640 stop_exec_queue
= true;
643 if (udev_ctrl_get_start_exec_queue(ctrl_msg
) > 0) {
644 log_debug("udevd message (START_EXEC_QUEUE) received");
645 stop_exec_queue
= false;
648 if (udev_ctrl_get_reload(ctrl_msg
) > 0) {
649 log_debug("udevd message (RELOAD) received");
653 str
= udev_ctrl_get_set_env(ctrl_msg
);
661 val
= strchr(key
, '=');
665 if (val
[0] == '\0') {
666 log_debug("udevd message (ENV) received, unset '%s'", key
);
667 udev_add_property(udev
, key
, NULL
);
669 log_debug("udevd message (ENV) received, set '%s=%s'", key
, val
);
670 udev_add_property(udev
, key
, val
);
673 log_error("wrong key format '%s'", key
);
680 i
= udev_ctrl_get_set_children_max(ctrl_msg
);
682 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i
);
686 if (udev_ctrl_get_ping(ctrl_msg
) > 0)
687 log_debug("udevd message (SYNC) received");
689 if (udev_ctrl_get_exit(ctrl_msg
) > 0) {
690 log_debug("udevd message (EXIT) received");
692 /* keep reference to block the client until we exit */
693 udev_ctrl_connection_ref(ctrl_conn
);
696 udev_ctrl_msg_unref(ctrl_msg
);
697 return udev_ctrl_connection_unref(ctrl_conn
);
700 static int synthesize_change(struct udev_device
*dev
) {
701 char filename
[UTIL_PATH_SIZE
];
704 if (streq_ptr("block", udev_device_get_subsystem(dev
)) &&
705 streq_ptr("disk", udev_device_get_devtype(dev
)) &&
706 !startswith(udev_device_get_sysname(dev
), "dm-")) {
707 bool part_table_read
= false;
708 bool has_partitions
= false;
710 struct udev
*udev
= udev_device_get_udev(dev
);
711 _cleanup_udev_enumerate_unref_
struct udev_enumerate
*e
= NULL
;
712 struct udev_list_entry
*item
;
715 * Try to re-read the partition table. This only succeeds if
716 * none of the devices is busy. The kernel returns 0 if no
717 * partition table is found, and we will not get an event for
720 fd
= open(udev_device_get_devnode(dev
), O_RDONLY
|O_CLOEXEC
|O_NOFOLLOW
|O_NONBLOCK
);
722 r
= flock(fd
, LOCK_EX
|LOCK_NB
);
724 r
= ioctl(fd
, BLKRRPART
, 0);
728 part_table_read
= true;
731 /* search for partitions */
732 e
= udev_enumerate_new(udev
);
736 r
= udev_enumerate_add_match_parent(e
, dev
);
740 r
= udev_enumerate_add_match_subsystem(e
, "block");
744 r
= udev_enumerate_scan_devices(e
);
748 udev_list_entry_foreach(item
, udev_enumerate_get_list_entry(e
)) {
749 _cleanup_udev_device_unref_
struct udev_device
*d
= NULL
;
751 d
= udev_device_new_from_syspath(udev
, udev_list_entry_get_name(item
));
755 if (!streq_ptr("partition", udev_device_get_devtype(d
)))
758 has_partitions
= true;
763 * We have partitions and re-read the table, the kernel already sent
764 * out a "change" event for the disk, and "remove/add" for all
767 if (part_table_read
&& has_partitions
)
771 * We have partitions but re-reading the partition table did not
772 * work, synthesize "change" for the disk and all partitions.
774 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev
));
775 strscpyl(filename
, sizeof(filename
), udev_device_get_syspath(dev
), "/uevent", NULL
);
776 write_string_file(filename
, "change");
778 udev_list_entry_foreach(item
, udev_enumerate_get_list_entry(e
)) {
779 _cleanup_udev_device_unref_
struct udev_device
*d
= NULL
;
781 d
= udev_device_new_from_syspath(udev
, udev_list_entry_get_name(item
));
785 if (!streq_ptr("partition", udev_device_get_devtype(d
)))
788 log_debug("device %s closed, synthesising partition '%s' 'change'",
789 udev_device_get_devnode(dev
), udev_device_get_devnode(d
));
790 strscpyl(filename
, sizeof(filename
), udev_device_get_syspath(d
), "/uevent", NULL
);
791 write_string_file(filename
, "change");
797 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev
));
798 strscpyl(filename
, sizeof(filename
), udev_device_get_syspath(dev
), "/uevent", NULL
);
799 write_string_file(filename
, "change");
804 static int handle_inotify(struct udev
*udev
) {
807 struct inotify_event
*ev
;
810 r
= ioctl(fd_inotify
, FIONREAD
, &nbytes
);
811 if (r
< 0 || nbytes
<= 0)
814 buf
= malloc(nbytes
);
816 log_error("error getting buffer for inotify");
820 nbytes
= read(fd_inotify
, buf
, nbytes
);
822 for (pos
= 0; pos
< nbytes
; pos
+= sizeof(struct inotify_event
) + ev
->len
) {
823 struct udev_device
*dev
;
825 ev
= (struct inotify_event
*)(buf
+ pos
);
826 dev
= udev_watch_lookup(udev
, ev
->wd
);
830 log_debug("inotify event: %x for %s", ev
->mask
, udev_device_get_devnode(dev
));
831 if (ev
->mask
& IN_CLOSE_WRITE
)
832 synthesize_change(dev
);
833 else if (ev
->mask
& IN_IGNORED
)
834 udev_watch_end(udev
, dev
);
836 udev_device_unref(dev
);
843 static void handle_signal(struct udev
*udev
, int signo
) {
853 struct udev_list_node
*loop
, *tmp
;
855 pid
= waitpid(-1, &status
, WNOHANG
);
859 udev_list_node_foreach_safe(loop
, tmp
, &worker_list
) {
860 struct worker
*worker
= node_to_worker(loop
);
862 if (worker
->pid
!= pid
)
864 log_debug("worker [%u] exit", pid
);
866 if (WIFEXITED(status
)) {
867 if (WEXITSTATUS(status
) != 0)
868 log_error("worker [%u] exit with return code %i",
869 pid
, WEXITSTATUS(status
));
870 } else if (WIFSIGNALED(status
)) {
871 log_error("worker [%u] terminated by signal %i (%s)",
872 pid
, WTERMSIG(status
), strsignal(WTERMSIG(status
)));
873 } else if (WIFSTOPPED(status
)) {
874 log_error("worker [%u] stopped", pid
);
875 } else if (WIFCONTINUED(status
)) {
876 log_error("worker [%u] continued", pid
);
878 log_error("worker [%u] exit with status 0x%04x", pid
, status
);
881 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
883 log_error("worker [%u] failed while handling '%s'",
884 pid
, worker
->event
->devpath
);
885 worker
->event
->exitcode
= -32;
886 event_queue_delete(worker
->event
);
888 /* drop reference taken for state 'running' */
889 worker_unref(worker
);
892 worker_unref(worker
);
903 static int systemd_fds(struct udev
*udev
, int *rctrl
, int *rnetlink
) {
904 int ctrl
= -1, netlink
= -1;
907 n
= sd_listen_fds(true);
911 for (fd
= SD_LISTEN_FDS_START
; fd
< n
+ SD_LISTEN_FDS_START
; fd
++) {
912 if (sd_is_socket(fd
, AF_LOCAL
, SOCK_SEQPACKET
, -1)) {
919 if (sd_is_socket(fd
, AF_NETLINK
, SOCK_RAW
, -1)) {
929 if (ctrl
< 0 || netlink
< 0)
932 log_debug("ctrl=%i netlink=%i", ctrl
, netlink
);
939 * read the kernel commandline, in case we need to get into debug mode
940 * udev.log-priority=<level> syslog priority
941 * udev.children-max=<number of workers> events are fully serialized if set to 1
942 * udev.exec-delay=<number of seconds> delay execution of every executed program
944 static void kernel_cmdline_options(struct udev
*udev
) {
945 _cleanup_free_
char *line
= NULL
;
946 const char *word
, *state
;
950 r
= proc_cmdline(&line
);
952 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r
));
956 FOREACH_WORD_QUOTED(word
, l
, line
, state
) {
959 s
= strndup(word
, l
);
963 /* accept the same options for the initrd, prefixed with "rd." */
964 if (in_initrd() && startswith(s
, "rd."))
969 if (startswith(opt
, "udev.log-priority=")) {
972 prio
= util_log_priority(opt
+ 18);
973 log_set_max_level(prio
);
974 udev_set_log_priority(udev
, prio
);
975 } else if (startswith(opt
, "udev.children-max=")) {
976 children_max
= strtoul(opt
+ 18, NULL
, 0);
977 } else if (startswith(opt
, "udev.exec-delay=")) {
978 exec_delay
= strtoul(opt
+ 16, NULL
, 0);
979 } else if (startswith(opt
, "udev.event-timeout=")) {
980 event_timeout_usec
= strtoul(opt
+ 16, NULL
, 0) * USEC_PER_SEC
;
987 int main(int argc
, char *argv
[]) {
990 int daemonize
= false;
991 int resolve_names
= 1;
992 static const struct option options
[] = {
993 { "daemon", no_argument
, NULL
, 'd' },
994 { "debug", no_argument
, NULL
, 'D' },
995 { "children-max", required_argument
, NULL
, 'c' },
996 { "exec-delay", required_argument
, NULL
, 'e' },
997 { "event-timeout", required_argument
, NULL
, 't' },
998 { "resolve-names", required_argument
, NULL
, 'N' },
999 { "help", no_argument
, NULL
, 'h' },
1000 { "version", no_argument
, NULL
, 'V' },
1004 int fd_netlink
= -1;
1006 struct epoll_event ep_ctrl
, ep_inotify
, ep_signal
, ep_netlink
, ep_worker
;
1007 struct udev_ctrl_connection
*ctrl_conn
= NULL
;
1014 log_set_target(LOG_TARGET_AUTO
);
1015 log_parse_environment();
1018 udev_set_log_fn(udev
, udev_main_log
);
1019 log_set_max_level(udev_get_log_priority(udev
));
1021 log_debug("version %s", VERSION
);
1027 option
= getopt_long(argc
, argv
, "c:de:DtN:hV", options
, NULL
);
1036 children_max
= strtoul(optarg
, NULL
, 0);
1039 exec_delay
= strtoul(optarg
, NULL
, 0);
1042 event_timeout_usec
= strtoul(optarg
, NULL
, 0) * USEC_PER_SEC
;
1046 log_set_max_level(LOG_DEBUG
);
1047 udev_set_log_priority(udev
, LOG_DEBUG
);
1050 if (streq(optarg
, "early")) {
1052 } else if (streq(optarg
, "late")) {
1054 } else if (streq(optarg
, "never")) {
1057 fprintf(stderr
, "resolve-names must be early, late or never\n");
1058 log_error("resolve-names must be early, late or never");
1063 printf("Usage: udevd OPTIONS\n"
1066 " --children-max=<maximum number of workers>\n"
1067 " --exec-delay=<seconds to wait before executing RUN=>\n"
1068 " --event-timeout=<seconds to wait before terminating an event>\n"
1069 " --resolve-names=early|late|never\n"
1075 printf("%s\n", VERSION
);
1082 kernel_cmdline_options(udev
);
1084 if (getuid() != 0) {
1085 fprintf(stderr
, "root privileges required\n");
1086 log_error("root privileges required");
1090 /* set umask before creating any file/directory */
1094 mkdir("/run/udev", 0755);
1098 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1102 fd
= open("/dev/null", O_RDWR
);
1104 if (write(STDOUT_FILENO
, 0, 0) < 0)
1105 dup2(fd
, STDOUT_FILENO
);
1106 if (write(STDERR_FILENO
, 0, 0) < 0)
1107 dup2(fd
, STDERR_FILENO
);
1108 if (fd
> STDERR_FILENO
)
1111 fprintf(stderr
, "cannot open /dev/null\n");
1112 log_error("cannot open /dev/null");
1116 if (systemd_fds(udev
, &fd_ctrl
, &fd_netlink
) >= 0) {
1117 /* get control and netlink socket from systemd */
1118 udev_ctrl
= udev_ctrl_new_from_fd(udev
, fd_ctrl
);
1119 if (udev_ctrl
== NULL
) {
1120 log_error("error taking over udev control socket");
1125 monitor
= udev_monitor_new_from_netlink_fd(udev
, "kernel", fd_netlink
);
1126 if (monitor
== NULL
) {
1127 log_error("error taking over netlink socket");
1132 /* get our own cgroup, we regularly kill everything udev has left behind */
1133 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER
, 0, &udev_cgroup
) < 0)
1136 /* open control and netlink socket */
1137 udev_ctrl
= udev_ctrl_new(udev
);
1138 if (udev_ctrl
== NULL
) {
1139 fprintf(stderr
, "error initializing udev control socket");
1140 log_error("error initializing udev control socket");
1144 fd_ctrl
= udev_ctrl_get_fd(udev_ctrl
);
1146 monitor
= udev_monitor_new_from_netlink(udev
, "kernel");
1147 if (monitor
== NULL
) {
1148 fprintf(stderr
, "error initializing netlink socket\n");
1149 log_error("error initializing netlink socket");
1153 fd_netlink
= udev_monitor_get_fd(monitor
);
1156 if (udev_monitor_enable_receiving(monitor
) < 0) {
1157 fprintf(stderr
, "error binding netlink socket\n");
1158 log_error("error binding netlink socket");
1163 if (udev_ctrl_enable_receiving(udev_ctrl
) < 0) {
1164 fprintf(stderr
, "error binding udev control socket\n");
1165 log_error("error binding udev control socket");
1170 udev_monitor_set_receive_buffer_size(monitor
, 128 * 1024 * 1024);
1180 log_error("fork of daemon failed: %m");
1185 goto exit_daemonize
;
1190 write_string_file("/proc/self/oom_score_adj", "-1000");
1192 sd_notify(1, "READY=1");
1195 print_kmsg("starting version " VERSION
"\n");
1200 fd
= open("/dev/null", O_RDWR
);
1202 dup2(fd
, STDIN_FILENO
);
1203 dup2(fd
, STDOUT_FILENO
);
1204 dup2(fd
, STDERR_FILENO
);
1209 fd_inotify
= udev_watch_init(udev
);
1210 if (fd_inotify
< 0) {
1211 fprintf(stderr
, "error initializing inotify\n");
1212 log_error("error initializing inotify");
1216 udev_watch_restore(udev
);
1218 /* block and listen to all signals on signalfd */
1220 sigprocmask(SIG_SETMASK
, &mask
, &sigmask_orig
);
1221 fd_signal
= signalfd(-1, &mask
, SFD_NONBLOCK
|SFD_CLOEXEC
);
1222 if (fd_signal
< 0) {
1223 fprintf(stderr
, "error creating signalfd\n");
1224 log_error("error creating signalfd");
1229 /* unnamed socket from workers to the main daemon */
1230 if (socketpair(AF_LOCAL
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, worker_watch
) < 0) {
1231 fprintf(stderr
, "error creating socketpair\n");
1232 log_error("error creating socketpair");
1236 fd_worker
= worker_watch
[READ_END
];
1238 udev_builtin_init(udev
);
1240 rules
= udev_rules_new(udev
, resolve_names
);
1241 if (rules
== NULL
) {
1242 log_error("error reading rules");
1246 memzero(&ep_ctrl
, sizeof(struct epoll_event
));
1247 ep_ctrl
.events
= EPOLLIN
;
1248 ep_ctrl
.data
.fd
= fd_ctrl
;
1250 memzero(&ep_inotify
, sizeof(struct epoll_event
));
1251 ep_inotify
.events
= EPOLLIN
;
1252 ep_inotify
.data
.fd
= fd_inotify
;
1254 memzero(&ep_signal
, sizeof(struct epoll_event
));
1255 ep_signal
.events
= EPOLLIN
;
1256 ep_signal
.data
.fd
= fd_signal
;
1258 memzero(&ep_netlink
, sizeof(struct epoll_event
));
1259 ep_netlink
.events
= EPOLLIN
;
1260 ep_netlink
.data
.fd
= fd_netlink
;
1262 memzero(&ep_worker
, sizeof(struct epoll_event
));
1263 ep_worker
.events
= EPOLLIN
;
1264 ep_worker
.data
.fd
= fd_worker
;
1266 fd_ep
= epoll_create1(EPOLL_CLOEXEC
);
1268 log_error("error creating epoll fd: %m");
1271 if (epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_ctrl
, &ep_ctrl
) < 0 ||
1272 epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_inotify
, &ep_inotify
) < 0 ||
1273 epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_signal
, &ep_signal
) < 0 ||
1274 epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_netlink
, &ep_netlink
) < 0 ||
1275 epoll_ctl(fd_ep
, EPOLL_CTL_ADD
, fd_worker
, &ep_worker
) < 0) {
1276 log_error("fail to add fds to epoll: %m");
1280 if (children_max
<= 0) {
1285 if (sched_getaffinity(0, sizeof (cpu_set
), &cpu_set
) == 0) {
1286 children_max
+= CPU_COUNT(&cpu_set
) * 2;
1289 log_debug("set children_max to %u", children_max
);
1291 rc
= udev_rules_apply_static_dev_perms(rules
);
1293 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc
));
1295 udev_list_node_init(&event_list
);
1296 udev_list_node_init(&worker_list
);
1299 static usec_t last_usec
;
1300 struct epoll_event ev
[8];
1303 bool is_worker
, is_signal
, is_inotify
, is_netlink
, is_ctrl
;
1307 /* close sources of new events and discard buffered events */
1309 epoll_ctl(fd_ep
, EPOLL_CTL_DEL
, fd_ctrl
, NULL
);
1312 if (monitor
!= NULL
) {
1313 epoll_ctl(fd_ep
, EPOLL_CTL_DEL
, fd_netlink
, NULL
);
1314 udev_monitor_unref(monitor
);
1317 if (fd_inotify
>= 0) {
1318 epoll_ctl(fd_ep
, EPOLL_CTL_DEL
, fd_inotify
, NULL
);
1323 /* discard queued events and kill workers */
1324 event_queue_cleanup(udev
, EVENT_QUEUED
);
1327 /* exit after all has cleaned up */
1328 if (udev_list_node_is_empty(&event_list
) && children
== 0)
1331 /* timeout at exit for workers to finish */
1332 timeout
= 30 * MSEC_PER_SEC
;
1333 } else if (udev_list_node_is_empty(&event_list
) && children
== 0) {
1337 /* cleanup possible left-over processes in our cgroup */
1339 cg_kill(SYSTEMD_CGROUP_CONTROLLER
, udev_cgroup
, SIGKILL
, false, true, NULL
);
1341 /* kill idle or hanging workers */
1342 timeout
= 3 * MSEC_PER_SEC
;
1345 /* tell settle that we are busy or idle */
1346 if (!udev_list_node_is_empty(&event_list
)) {
1349 fd
= open("/run/udev/queue", O_WRONLY
|O_CREAT
|O_CLOEXEC
|O_TRUNC
|O_NOFOLLOW
, 0444);
1353 unlink("/run/udev/queue");
1356 fdcount
= epoll_wait(fd_ep
, ev
, ELEMENTSOF(ev
), timeout
);
1361 struct udev_list_node
*loop
;
1365 log_error("timeout, giving up waiting for workers to finish");
1369 /* kill idle workers */
1370 if (udev_list_node_is_empty(&event_list
)) {
1371 log_debug("cleanup idle workers");
1375 /* check for hanging events */
1376 udev_list_node_foreach(loop
, &worker_list
) {
1377 struct worker
*worker
= node_to_worker(loop
);
1379 if (worker
->state
!= WORKER_RUNNING
)
1382 if ((now(CLOCK_MONOTONIC
) - worker
->event_start_usec
) > event_timeout_usec
) {
1383 log_error("worker [%u] %s timeout; kill it", worker
->pid
, worker
->event
->devpath
);
1384 kill(worker
->pid
, SIGKILL
);
1385 worker
->state
= WORKER_KILLED
;
1387 /* drop reference taken for state 'running' */
1388 worker_unref(worker
);
1389 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker
->event
->dev
), worker
->event
->devpath
);
1390 worker
->event
->exitcode
= -64;
1391 event_queue_delete(worker
->event
);
1392 worker
->event
= NULL
;
1398 is_worker
= is_signal
= is_inotify
= is_netlink
= is_ctrl
= false;
1399 for (i
= 0; i
< fdcount
; i
++) {
1400 if (ev
[i
].data
.fd
== fd_worker
&& ev
[i
].events
& EPOLLIN
)
1402 else if (ev
[i
].data
.fd
== fd_netlink
&& ev
[i
].events
& EPOLLIN
)
1404 else if (ev
[i
].data
.fd
== fd_signal
&& ev
[i
].events
& EPOLLIN
)
1406 else if (ev
[i
].data
.fd
== fd_inotify
&& ev
[i
].events
& EPOLLIN
)
1408 else if (ev
[i
].data
.fd
== fd_ctrl
&& ev
[i
].events
& EPOLLIN
)
1412 /* check for changed config, every 3 seconds at most */
1413 if ((now(CLOCK_MONOTONIC
) - last_usec
) > 3 * USEC_PER_SEC
) {
1414 if (udev_rules_check_timestamp(rules
))
1416 if (udev_builtin_validate(udev
))
1419 last_usec
= now(CLOCK_MONOTONIC
);
1422 /* reload requested, HUP signal received, rules changed, builtin changed */
1425 rules
= udev_rules_unref(rules
);
1426 udev_builtin_exit(udev
);
1430 /* event has finished */
1432 worker_returned(fd_worker
);
1435 struct udev_device
*dev
;
1437 dev
= udev_monitor_receive_device(monitor
);
1439 udev_device_set_usec_initialized(dev
, now(CLOCK_MONOTONIC
));
1440 if (event_queue_insert(dev
) < 0)
1441 udev_device_unref(dev
);
1445 /* start new events */
1446 if (!udev_list_node_is_empty(&event_list
) && !udev_exit
&& !stop_exec_queue
) {
1447 udev_builtin_init(udev
);
1449 rules
= udev_rules_new(udev
, resolve_names
);
1451 event_queue_start(udev
);
1455 struct signalfd_siginfo fdsi
;
1458 size
= read(fd_signal
, &fdsi
, sizeof(struct signalfd_siginfo
));
1459 if (size
== sizeof(struct signalfd_siginfo
))
1460 handle_signal(udev
, fdsi
.ssi_signo
);
1463 /* we are shutting down, the events below are not handled anymore */
1467 /* device node watch */
1469 handle_inotify(udev
);
1472 * This needs to be after the inotify handling, to make sure,
1473 * that the ping is send back after the possibly generated
1474 * "change" events by the inotify device node watch.
1476 * A single time we may receive a client connection which we need to
1477 * keep open to block the client. It will be closed right before we
1481 ctrl_conn
= handle_ctrl_msg(udev_ctrl
);
1486 udev_ctrl_cleanup(udev_ctrl
);
1487 unlink("/run/udev/queue");
1491 worker_list_cleanup(udev
);
1492 event_queue_cleanup(udev
, EVENT_UNDEF
);
1493 udev_rules_unref(rules
);
1494 udev_builtin_exit(udev
);
1497 if (worker_watch
[READ_END
] >= 0)
1498 close(worker_watch
[READ_END
]);
1499 if (worker_watch
[WRITE_END
] >= 0)
1500 close(worker_watch
[WRITE_END
]);
1501 udev_monitor_unref(monitor
);
1502 udev_ctrl_connection_unref(ctrl_conn
);
1503 udev_ctrl_unref(udev_ctrl
);