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/>.
33 #include <sys/prctl.h>
34 #include <sys/socket.h>
35 #include <sys/signalfd.h>
36 #include <sys/epoll.h>
37 #include <sys/mount.h>
40 #include <sys/ioctl.h>
41 #include <sys/inotify.h>
43 #include "sd-daemon.h"
44 #include "rtnl-util.h"
45 #include "cgroup-util.h"
46 #include "dev-setup.h"
48 #include "selinux-util.h"
50 #include "udev-util.h"
52 static struct udev_rules *rules;
53 static struct udev_ctrl *udev_ctrl;
54 static struct udev_monitor *monitor;
55 static int worker_watch[2] = { -1, -1 };
56 static int fd_signal = -1;
57 static int fd_ep = -1;
58 static int fd_inotify = -1;
59 static bool stop_exec_queue;
62 static bool arg_debug = false;
63 static int arg_daemonize = false;
64 static int arg_resolve_names = 1;
65 static int arg_children_max;
66 static int arg_exec_delay;
67 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
68 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
69 static sigset_t sigmask_orig;
70 static UDEV_LIST(event_list);
71 static UDEV_LIST(worker_list);
72 static char *udev_cgroup;
73 static struct udev_list properties_list;
74 static bool udev_exit;
83 struct udev_list_node node;
85 struct udev_device *dev;
86 enum event_state state;
88 unsigned long long int delaying_seqnum;
89 unsigned long long int seqnum;
92 const char *devpath_old;
98 static inline struct event *node_to_event(struct udev_list_node *node) {
99 return container_of(node, struct event, node);
102 static void event_queue_cleanup(struct udev *udev, enum event_state type);
112 struct udev_list_node node;
116 struct udev_monitor *monitor;
117 enum worker_state state;
119 usec_t event_start_usec;
123 /* passed from worker to main process */
124 struct worker_message {
129 static inline struct worker *node_to_worker(struct udev_list_node *node) {
130 return container_of(node, struct worker, node);
133 static void event_queue_delete(struct event *event) {
134 udev_list_node_remove(&event->node);
135 udev_device_unref(event->dev);
139 static struct worker *worker_ref(struct worker *worker) {
144 static void worker_cleanup(struct worker *worker) {
145 udev_list_node_remove(&worker->node);
146 udev_monitor_unref(worker->monitor);
151 static void worker_unref(struct worker *worker) {
153 if (worker->refcount > 0)
155 log_debug("worker ["PID_FMT"] cleaned up", worker->pid);
156 worker_cleanup(worker);
159 static void worker_list_cleanup(struct udev *udev) {
160 struct udev_list_node *loop, *tmp;
162 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
163 struct worker *worker = node_to_worker(loop);
165 worker_cleanup(worker);
169 static void worker_new(struct event *event) {
170 struct udev *udev = event->udev;
171 struct worker *worker;
172 struct udev_monitor *worker_monitor;
175 /* listen for new events */
176 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
177 if (worker_monitor == NULL)
179 /* allow the main daemon netlink address to send devices to the worker */
180 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
181 udev_monitor_enable_receiving(worker_monitor);
183 worker = new0(struct worker, 1);
184 if (worker == NULL) {
185 udev_monitor_unref(worker_monitor);
188 /* worker + event reference */
189 worker->refcount = 2;
195 struct udev_device *dev = NULL;
197 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
198 struct epoll_event ep_signal, ep_monitor;
200 int rc = EXIT_SUCCESS;
202 /* take initial device from queue */
207 worker_list_cleanup(udev);
208 event_queue_cleanup(udev, EVENT_UNDEF);
209 udev_monitor_unref(monitor);
210 udev_ctrl_unref(udev_ctrl);
213 close(worker_watch[READ_END]);
216 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
218 log_error_errno(errno, "error creating signalfd %m");
223 fd_ep = epoll_create1(EPOLL_CLOEXEC);
225 log_error_errno(errno, "error creating epoll fd: %m");
230 memzero(&ep_signal, sizeof(struct epoll_event));
231 ep_signal.events = EPOLLIN;
232 ep_signal.data.fd = fd_signal;
234 fd_monitor = udev_monitor_get_fd(worker_monitor);
235 memzero(&ep_monitor, sizeof(struct epoll_event));
236 ep_monitor.events = EPOLLIN;
237 ep_monitor.data.fd = fd_monitor;
239 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
240 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
241 log_error_errno(errno, "fail to add fds to epoll: %m");
246 /* request TERM signal if parent exits */
247 prctl(PR_SET_PDEATHSIG, SIGTERM);
249 /* reset OOM score, we only protect the main daemon */
250 write_string_file("/proc/self/oom_score_adj", "0");
253 struct udev_event *udev_event;
254 struct worker_message msg;
258 log_debug("seq %llu running", udev_device_get_seqnum(dev));
259 udev_event = udev_event_new(dev);
260 if (udev_event == NULL) {
265 /* needed for SIGCHLD/SIGTERM in spawn() */
266 udev_event->fd_signal = fd_signal;
268 if (arg_exec_delay > 0)
269 udev_event->exec_delay = arg_exec_delay;
272 * Take a shared lock on the device node; this establishes
273 * a concept of device "ownership" to serialize device
274 * access. External processes holding an exclusive lock will
275 * cause udev to skip the event handling; in the case udev
276 * acquired the lock, the external process can block until
277 * udev has finished its event handling.
279 if (!streq_ptr(udev_device_get_action(dev), "remove") &&
280 streq_ptr("block", udev_device_get_subsystem(dev)) &&
281 !startswith(udev_device_get_sysname(dev), "dm-") &&
282 !startswith(udev_device_get_sysname(dev), "md")) {
283 struct udev_device *d = dev;
285 if (streq_ptr("partition", udev_device_get_devtype(d)))
286 d = udev_device_get_parent(d);
289 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
290 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
291 log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
293 fd_lock = safe_close(fd_lock);
299 /* needed for renaming netifs */
300 udev_event->rtnl = rtnl;
302 /* apply rules, create node, symlinks */
303 udev_event_execute_rules(udev_event,
304 arg_event_timeout_usec, arg_event_timeout_warn_usec,
309 udev_event_execute_run(udev_event,
310 arg_event_timeout_usec, arg_event_timeout_warn_usec,
313 if (udev_event->rtnl)
314 /* in case rtnl was initialized */
315 rtnl = sd_rtnl_ref(udev_event->rtnl);
317 /* apply/restore inotify watch */
318 if (udev_event->inotify_watch) {
319 udev_watch_begin(udev, dev);
320 udev_device_update_db(dev);
325 /* send processed event back to libudev listeners */
326 udev_monitor_send_device(worker_monitor, NULL, dev);
329 /* send udevd the result of the event execution */
330 memzero(&msg, sizeof(struct worker_message));
333 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
335 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
337 udev_device_unref(dev);
340 if (udev_event->sigterm) {
341 udev_event_unref(udev_event);
345 udev_event_unref(udev_event);
347 /* wait for more device messages from main udevd, or term signal */
348 while (dev == NULL) {
349 struct epoll_event ev[4];
353 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
357 log_error_errno(errno, "failed to poll: %m");
361 for (i = 0; i < fdcount; i++) {
362 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
363 dev = udev_monitor_receive_device(worker_monitor);
365 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
366 struct signalfd_siginfo fdsi;
369 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
370 if (size != sizeof(struct signalfd_siginfo))
372 switch (fdsi.ssi_signo) {
381 udev_device_unref(dev);
382 safe_close(fd_signal);
385 close(worker_watch[WRITE_END]);
386 udev_rules_unref(rules);
387 udev_builtin_exit(udev);
388 udev_monitor_unref(worker_monitor);
394 udev_monitor_unref(worker_monitor);
395 event->state = EVENT_QUEUED;
397 log_error_errno(errno, "fork of child failed: %m");
400 /* close monitor, but keep address around */
401 udev_monitor_disconnect(worker_monitor);
402 worker->monitor = worker_monitor;
404 worker->state = WORKER_RUNNING;
405 worker->event_start_usec = now(CLOCK_MONOTONIC);
406 worker->event_warned = false;
407 worker->event = event;
408 event->state = EVENT_RUNNING;
409 udev_list_node_append(&worker->node, &worker_list);
411 log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
416 static void event_run(struct event *event) {
417 struct udev_list_node *loop;
419 udev_list_node_foreach(loop, &worker_list) {
420 struct worker *worker = node_to_worker(loop);
423 if (worker->state != WORKER_IDLE)
426 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
428 log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
430 kill(worker->pid, SIGKILL);
431 worker->state = WORKER_KILLED;
435 worker->event = event;
436 worker->state = WORKER_RUNNING;
437 worker->event_start_usec = now(CLOCK_MONOTONIC);
438 worker->event_warned = false;
439 event->state = EVENT_RUNNING;
443 if (children >= arg_children_max) {
444 if (arg_children_max > 1)
445 log_debug("maximum number (%i) of children reached", children);
449 /* start new worker and pass initial device */
453 static int event_queue_insert(struct udev_device *dev) {
456 event = new0(struct event, 1);
460 event->udev = udev_device_get_udev(dev);
462 event->seqnum = udev_device_get_seqnum(dev);
463 event->devpath = udev_device_get_devpath(dev);
464 event->devpath_len = strlen(event->devpath);
465 event->devpath_old = udev_device_get_devpath_old(dev);
466 event->devnum = udev_device_get_devnum(dev);
467 event->is_block = streq("block", udev_device_get_subsystem(dev));
468 event->ifindex = udev_device_get_ifindex(dev);
470 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
471 udev_device_get_action(dev), udev_device_get_subsystem(dev));
473 event->state = EVENT_QUEUED;
474 udev_list_node_append(&event->node, &event_list);
478 static void worker_kill(struct udev *udev) {
479 struct udev_list_node *loop;
481 udev_list_node_foreach(loop, &worker_list) {
482 struct worker *worker = node_to_worker(loop);
484 if (worker->state == WORKER_KILLED)
487 worker->state = WORKER_KILLED;
488 kill(worker->pid, SIGTERM);
492 /* lookup event for identical, parent, child device */
493 static bool is_devpath_busy(struct event *event) {
494 struct udev_list_node *loop;
497 /* check if queue contains events we depend on */
498 udev_list_node_foreach(loop, &event_list) {
499 struct event *loop_event = node_to_event(loop);
501 /* we already found a later event, earlier can not block us, no need to check again */
502 if (loop_event->seqnum < event->delaying_seqnum)
505 /* event we checked earlier still exists, no need to check again */
506 if (loop_event->seqnum == event->delaying_seqnum)
509 /* found ourself, no later event can block us */
510 if (loop_event->seqnum >= event->seqnum)
513 /* check major/minor */
514 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
517 /* check network device ifindex */
518 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
521 /* check our old name */
522 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
523 event->delaying_seqnum = loop_event->seqnum;
527 /* compare devpath */
528 common = MIN(loop_event->devpath_len, event->devpath_len);
530 /* one devpath is contained in the other? */
531 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
534 /* identical device event found */
535 if (loop_event->devpath_len == event->devpath_len) {
536 /* devices names might have changed/swapped in the meantime */
537 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
539 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
541 event->delaying_seqnum = loop_event->seqnum;
545 /* parent device event found */
546 if (event->devpath[common] == '/') {
547 event->delaying_seqnum = loop_event->seqnum;
551 /* child device event found */
552 if (loop_event->devpath[common] == '/') {
553 event->delaying_seqnum = loop_event->seqnum;
557 /* no matching device */
564 static void event_queue_start(struct udev *udev) {
565 struct udev_list_node *loop;
567 udev_list_node_foreach(loop, &event_list) {
568 struct event *event = node_to_event(loop);
570 if (event->state != EVENT_QUEUED)
573 /* do not start event if parent or child event is still running */
574 if (is_devpath_busy(event))
581 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
582 struct udev_list_node *loop, *tmp;
584 udev_list_node_foreach_safe(loop, tmp, &event_list) {
585 struct event *event = node_to_event(loop);
587 if (match_type != EVENT_UNDEF && match_type != event->state)
590 event_queue_delete(event);
594 static void worker_returned(int fd_worker) {
596 struct worker_message msg;
598 struct udev_list_node *loop;
600 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
601 if (size != sizeof(struct worker_message))
604 /* lookup worker who sent the signal */
605 udev_list_node_foreach(loop, &worker_list) {
606 struct worker *worker = node_to_worker(loop);
608 if (worker->pid != msg.pid)
611 /* worker returned */
613 worker->event->exitcode = msg.exitcode;
614 event_queue_delete(worker->event);
615 worker->event = NULL;
617 if (worker->state != WORKER_KILLED)
618 worker->state = WORKER_IDLE;
619 worker_unref(worker);
625 /* receive the udevd message from userspace */
626 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
627 struct udev *udev = udev_ctrl_get_udev(uctrl);
628 struct udev_ctrl_connection *ctrl_conn;
629 struct udev_ctrl_msg *ctrl_msg = NULL;
633 ctrl_conn = udev_ctrl_get_connection(uctrl);
634 if (ctrl_conn == NULL)
637 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
638 if (ctrl_msg == NULL)
641 i = udev_ctrl_get_set_log_level(ctrl_msg);
643 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
644 log_set_max_level(i);
648 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
649 log_debug("udevd message (STOP_EXEC_QUEUE) received");
650 stop_exec_queue = true;
653 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
654 log_debug("udevd message (START_EXEC_QUEUE) received");
655 stop_exec_queue = false;
658 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
659 log_debug("udevd message (RELOAD) received");
663 str = udev_ctrl_get_set_env(ctrl_msg);
671 val = strchr(key, '=');
675 if (val[0] == '\0') {
676 log_debug("udevd message (ENV) received, unset '%s'", key);
677 udev_list_entry_add(&properties_list, key, NULL);
679 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
680 udev_list_entry_add(&properties_list, key, val);
683 log_error("wrong key format '%s'", key);
690 i = udev_ctrl_get_set_children_max(ctrl_msg);
692 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
693 arg_children_max = i;
696 if (udev_ctrl_get_ping(ctrl_msg) > 0)
697 log_debug("udevd message (SYNC) received");
699 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
700 log_debug("udevd message (EXIT) received");
702 /* keep reference to block the client until we exit */
703 udev_ctrl_connection_ref(ctrl_conn);
706 udev_ctrl_msg_unref(ctrl_msg);
707 return udev_ctrl_connection_unref(ctrl_conn);
710 static int synthesize_change(struct udev_device *dev) {
711 char filename[UTIL_PATH_SIZE];
714 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
715 streq_ptr("disk", udev_device_get_devtype(dev)) &&
716 !startswith(udev_device_get_sysname(dev), "dm-")) {
717 bool part_table_read = false;
718 bool has_partitions = false;
720 struct udev *udev = udev_device_get_udev(dev);
721 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
722 struct udev_list_entry *item;
725 * Try to re-read the partition table. This only succeeds if
726 * none of the devices is busy. The kernel returns 0 if no
727 * partition table is found, and we will not get an event for
730 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
732 r = flock(fd, LOCK_EX|LOCK_NB);
734 r = ioctl(fd, BLKRRPART, 0);
738 part_table_read = true;
741 /* search for partitions */
742 e = udev_enumerate_new(udev);
746 r = udev_enumerate_add_match_parent(e, dev);
750 r = udev_enumerate_add_match_subsystem(e, "block");
754 r = udev_enumerate_scan_devices(e);
758 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
759 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
761 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
765 if (!streq_ptr("partition", udev_device_get_devtype(d)))
768 has_partitions = true;
773 * We have partitions and re-read the table, the kernel already sent
774 * out a "change" event for the disk, and "remove/add" for all
777 if (part_table_read && has_partitions)
781 * We have partitions but re-reading the partition table did not
782 * work, synthesize "change" for the disk and all partitions.
784 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
785 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
786 write_string_file(filename, "change");
788 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
789 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
791 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
795 if (!streq_ptr("partition", udev_device_get_devtype(d)))
798 log_debug("device %s closed, synthesising partition '%s' 'change'",
799 udev_device_get_devnode(dev), udev_device_get_devnode(d));
800 strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
801 write_string_file(filename, "change");
807 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
808 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
809 write_string_file(filename, "change");
814 static int handle_inotify(struct udev *udev) {
815 union inotify_event_buffer buffer;
816 struct inotify_event *e;
819 l = read(fd_inotify, &buffer, sizeof(buffer));
821 if (errno == EAGAIN || errno == EINTR)
824 return log_error_errno(errno, "Failed to read inotify fd: %m");
827 FOREACH_INOTIFY_EVENT(e, buffer, l) {
828 struct udev_device *dev;
830 dev = udev_watch_lookup(udev, e->wd);
834 log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
835 if (e->mask & IN_CLOSE_WRITE)
836 synthesize_change(dev);
837 else if (e->mask & IN_IGNORED)
838 udev_watch_end(udev, dev);
840 udev_device_unref(dev);
846 static void handle_signal(struct udev *udev, int signo) {
856 struct udev_list_node *loop, *tmp;
858 pid = waitpid(-1, &status, WNOHANG);
862 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
863 struct worker *worker = node_to_worker(loop);
865 if (worker->pid != pid)
867 log_debug("worker ["PID_FMT"] exit", pid);
869 if (WIFEXITED(status)) {
870 if (WEXITSTATUS(status) != 0)
871 log_error("worker ["PID_FMT"] exit with return code %i",
872 pid, WEXITSTATUS(status));
873 } else if (WIFSIGNALED(status)) {
874 log_error("worker ["PID_FMT"] terminated by signal %i (%s)",
875 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
876 } else if (WIFSTOPPED(status)) {
877 log_error("worker ["PID_FMT"] stopped", pid);
878 } else if (WIFCONTINUED(status)) {
879 log_error("worker ["PID_FMT"] continued", pid);
881 log_error("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
884 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
886 log_error("worker ["PID_FMT"] failed while handling '%s'",
887 pid, worker->event->devpath);
888 worker->event->exitcode = -32;
889 event_queue_delete(worker->event);
891 /* drop reference taken for state 'running' */
892 worker_unref(worker);
895 worker_unref(worker);
906 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
907 int ctrl = -1, netlink = -1;
910 n = sd_listen_fds(true);
914 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
915 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
922 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
932 if (ctrl < 0 || netlink < 0)
935 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
942 * read the kernel command line, in case we need to get into debug mode
943 * udev.log-priority=<level> syslog priority
944 * udev.children-max=<number of workers> events are fully serialized if set to 1
945 * udev.exec-delay=<number of seconds> delay execution of every executed program
947 static void kernel_cmdline_options(struct udev *udev) {
948 _cleanup_free_ char *line = NULL;
949 const char *word, *state;
953 r = proc_cmdline(&line);
955 log_warning_errno(r, "Failed to read /proc/cmdline, ignoring: %m");
959 FOREACH_WORD_QUOTED(word, l, line, state) {
960 char *s, *opt, *value;
962 s = strndup(word, l);
966 /* accept the same options for the initrd, prefixed with "rd." */
967 if (in_initrd() && startswith(s, "rd."))
972 if ((value = startswith(opt, "udev.log-priority="))) {
975 prio = util_log_priority(value);
976 log_set_max_level(prio);
977 } else if ((value = startswith(opt, "udev.children-max="))) {
978 r = safe_atoi(value, &arg_children_max);
980 log_warning("Invalid udev.children-max ignored: %s", value);
981 } else if ((value = startswith(opt, "udev.exec-delay="))) {
982 r = safe_atoi(value, &arg_exec_delay);
984 log_warning("Invalid udev.exec-delay ignored: %s", value);
985 } else if ((value = startswith(opt, "udev.event-timeout="))) {
986 r = safe_atou64(value, &arg_event_timeout_usec);
988 log_warning("Invalid udev.event-timeout ignored: %s", value);
991 arg_event_timeout_usec *= USEC_PER_SEC;
992 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
999 static void help(void) {
1000 printf("%s [OPTIONS...]\n\n"
1001 "Manages devices.\n\n"
1002 " -h --help Print this message\n"
1003 " --version Print version of the program\n"
1004 " --daemon Detach and run in the background\n"
1005 " --debug Enable debug output\n"
1006 " --children-max=INT Set maximum number of workers\n"
1007 " --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
1008 " --event-timeout=SECONDS Seconds to wait before terminating an event\n"
1009 " --resolve-names=early|late|never\n"
1010 " When to resolve users and groups\n"
1011 , program_invocation_short_name);
1014 static int parse_argv(int argc, char *argv[]) {
1015 static const struct option options[] = {
1016 { "daemon", no_argument, NULL, 'd' },
1017 { "debug", no_argument, NULL, 'D' },
1018 { "children-max", required_argument, NULL, 'c' },
1019 { "exec-delay", required_argument, NULL, 'e' },
1020 { "event-timeout", required_argument, NULL, 't' },
1021 { "resolve-names", required_argument, NULL, 'N' },
1022 { "help", no_argument, NULL, 'h' },
1023 { "version", no_argument, NULL, 'V' },
1032 while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1038 arg_daemonize = true;
1041 r = safe_atoi(optarg, &arg_children_max);
1043 log_warning("Invalid --children-max ignored: %s", optarg);
1046 r = safe_atoi(optarg, &arg_exec_delay);
1048 log_warning("Invalid --exec-delay ignored: %s", optarg);
1051 r = safe_atou64(optarg, &arg_event_timeout_usec);
1053 log_warning("Invalid --event-timeout ignored: %s", optarg);
1055 arg_event_timeout_usec *= USEC_PER_SEC;
1056 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1063 if (streq(optarg, "early")) {
1064 arg_resolve_names = 1;
1065 } else if (streq(optarg, "late")) {
1066 arg_resolve_names = 0;
1067 } else if (streq(optarg, "never")) {
1068 arg_resolve_names = -1;
1070 log_error("resolve-names must be early, late or never");
1078 printf("%s\n", VERSION);
1083 assert_not_reached("Unhandled option");
1091 int main(int argc, char *argv[]) {
1095 int fd_netlink = -1;
1097 struct epoll_event ep_ctrl = { .events = EPOLLIN };
1098 struct epoll_event ep_inotify = { .events = EPOLLIN };
1099 struct epoll_event ep_signal = { .events = EPOLLIN };
1100 struct epoll_event ep_netlink = { .events = EPOLLIN };
1101 struct epoll_event ep_worker = { .events = EPOLLIN };
1102 struct udev_ctrl_connection *ctrl_conn = NULL;
1109 log_set_target(LOG_TARGET_AUTO);
1110 log_parse_environment();
1113 r = parse_argv(argc, argv);
1117 kernel_cmdline_options(udev);
1120 log_set_max_level(LOG_DEBUG);
1122 if (getuid() != 0) {
1123 log_error("root privileges required");
1127 r = mac_selinux_init("/dev");
1129 log_error_errno(r, "could not initialize labelling: %m");
1133 /* set umask before creating any file/directory */
1136 log_error_errno(errno, "could not change dir to /: %m");
1142 udev_list_init(udev, &properties_list, true);
1144 r = mkdir("/run/udev", 0755);
1145 if (r < 0 && errno != EEXIST) {
1146 log_error_errno(errno, "could not create /run/udev: %m");
1152 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1153 if (arg_daemonize) {
1156 fd = open("/dev/null", O_RDWR);
1158 if (write(STDOUT_FILENO, 0, 0) < 0)
1159 dup2(fd, STDOUT_FILENO);
1160 if (write(STDERR_FILENO, 0, 0) < 0)
1161 dup2(fd, STDERR_FILENO);
1162 if (fd > STDERR_FILENO)
1165 log_error("cannot open /dev/null");
1169 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1170 /* get control and netlink socket from systemd */
1171 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1172 if (udev_ctrl == NULL) {
1173 log_error("error taking over udev control socket");
1178 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1179 if (monitor == NULL) {
1180 log_error("error taking over netlink socket");
1185 /* get our own cgroup, we regularly kill everything udev has left behind */
1186 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1189 /* open control and netlink socket */
1190 udev_ctrl = udev_ctrl_new(udev);
1191 if (udev_ctrl == NULL) {
1192 log_error("error initializing udev control socket");
1196 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1198 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1199 if (monitor == NULL) {
1200 log_error("error initializing netlink socket");
1204 fd_netlink = udev_monitor_get_fd(monitor);
1206 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1209 if (udev_monitor_enable_receiving(monitor) < 0) {
1210 log_error("error binding netlink socket");
1215 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1216 log_error("error binding udev control socket");
1221 log_info("starting version " VERSION);
1223 udev_builtin_init(udev);
1225 rules = udev_rules_new(udev, arg_resolve_names);
1226 if (rules == NULL) {
1227 log_error("error reading rules");
1231 rc = udev_rules_apply_static_dev_perms(rules);
1233 log_error_errno(rc, "failed to apply permissions on static device nodes - %m");
1235 if (arg_daemonize) {
1243 log_error_errno(errno, "fork of daemon failed: %m");
1248 goto exit_daemonize;
1253 write_string_file("/proc/self/oom_score_adj", "-1000");
1255 sd_notify(1, "READY=1");
1258 if (arg_children_max <= 0) {
1261 arg_children_max = 8;
1263 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1264 arg_children_max += CPU_COUNT(&cpu_set) * 2;
1267 log_debug("set children_max to %u", arg_children_max);
1269 udev_list_node_init(&event_list);
1270 udev_list_node_init(&worker_list);
1272 fd_inotify = udev_watch_init(udev);
1273 if (fd_inotify < 0) {
1274 log_error("error initializing inotify");
1278 udev_watch_restore(udev);
1280 /* block and listen to all signals on signalfd */
1282 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1283 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1284 if (fd_signal < 0) {
1285 log_error("error creating signalfd");
1290 /* unnamed socket from workers to the main daemon */
1291 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1292 log_error("error creating socketpair");
1296 fd_worker = worker_watch[READ_END];
1298 ep_ctrl.data.fd = fd_ctrl;
1299 ep_inotify.data.fd = fd_inotify;
1300 ep_signal.data.fd = fd_signal;
1301 ep_netlink.data.fd = fd_netlink;
1302 ep_worker.data.fd = fd_worker;
1304 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1306 log_error_errno(errno, "error creating epoll fd: %m");
1309 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1310 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1311 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1312 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1313 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1314 log_error_errno(errno, "fail to add fds to epoll: %m");
1319 static usec_t last_usec;
1320 struct epoll_event ev[8];
1323 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1327 /* close sources of new events and discard buffered events */
1329 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1332 if (monitor != NULL) {
1333 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1334 udev_monitor_unref(monitor);
1337 if (fd_inotify >= 0) {
1338 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1343 /* discard queued events and kill workers */
1344 event_queue_cleanup(udev, EVENT_QUEUED);
1347 /* exit after all has cleaned up */
1348 if (udev_list_node_is_empty(&event_list) && children == 0)
1351 /* timeout at exit for workers to finish */
1352 timeout = 30 * MSEC_PER_SEC;
1353 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1357 /* cleanup possible left-over processes in our cgroup */
1359 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1361 /* kill idle or hanging workers */
1362 timeout = 3 * MSEC_PER_SEC;
1365 /* tell settle that we are busy or idle */
1366 if (!udev_list_node_is_empty(&event_list)) {
1369 fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1373 unlink("/run/udev/queue");
1376 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1381 struct udev_list_node *loop;
1385 log_error("timeout, giving up waiting for workers to finish");
1389 /* kill idle workers */
1390 if (udev_list_node_is_empty(&event_list)) {
1391 log_debug("cleanup idle workers");
1395 /* check for hanging events */
1396 udev_list_node_foreach(loop, &worker_list) {
1397 struct worker *worker = node_to_worker(loop);
1400 if (worker->state != WORKER_RUNNING)
1403 ts = now(CLOCK_MONOTONIC);
1405 if ((ts - worker->event_start_usec) > arg_event_timeout_warn_usec) {
1406 if ((ts - worker->event_start_usec) > arg_event_timeout_usec) {
1407 log_error("worker ["PID_FMT"] %s timeout; kill it", worker->pid, worker->event->devpath);
1408 kill(worker->pid, SIGKILL);
1409 worker->state = WORKER_KILLED;
1411 /* drop reference taken for state 'running' */
1412 worker_unref(worker);
1413 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1414 worker->event->exitcode = -64;
1415 event_queue_delete(worker->event);
1416 worker->event = NULL;
1417 } else if (!worker->event_warned) {
1418 log_warning("worker ["PID_FMT"] %s is taking a long time", worker->pid, worker->event->devpath);
1419 worker->event_warned = true;
1426 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1427 for (i = 0; i < fdcount; i++) {
1428 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1430 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1432 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1434 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1436 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1440 /* check for changed config, every 3 seconds at most */
1441 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1442 if (udev_rules_check_timestamp(rules))
1444 if (udev_builtin_validate(udev))
1447 last_usec = now(CLOCK_MONOTONIC);
1450 /* reload requested, HUP signal received, rules changed, builtin changed */
1453 rules = udev_rules_unref(rules);
1454 udev_builtin_exit(udev);
1458 /* event has finished */
1460 worker_returned(fd_worker);
1463 struct udev_device *dev;
1465 dev = udev_monitor_receive_device(monitor);
1467 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1468 if (event_queue_insert(dev) < 0)
1469 udev_device_unref(dev);
1473 /* start new events */
1474 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1475 udev_builtin_init(udev);
1477 rules = udev_rules_new(udev, arg_resolve_names);
1479 event_queue_start(udev);
1483 struct signalfd_siginfo fdsi;
1486 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1487 if (size == sizeof(struct signalfd_siginfo))
1488 handle_signal(udev, fdsi.ssi_signo);
1491 /* we are shutting down, the events below are not handled anymore */
1495 /* device node watch */
1497 handle_inotify(udev);
1500 * This needs to be after the inotify handling, to make sure,
1501 * that the ping is send back after the possibly generated
1502 * "change" events by the inotify device node watch.
1504 * A single time we may receive a client connection which we need to
1505 * keep open to block the client. It will be closed right before we
1509 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1514 udev_ctrl_cleanup(udev_ctrl);
1515 unlink("/run/udev/queue");
1519 worker_list_cleanup(udev);
1520 event_queue_cleanup(udev, EVENT_UNDEF);
1521 udev_rules_unref(rules);
1522 udev_builtin_exit(udev);
1525 if (worker_watch[READ_END] >= 0)
1526 close(worker_watch[READ_END]);
1527 if (worker_watch[WRITE_END] >= 0)
1528 close(worker_watch[WRITE_END]);
1529 udev_monitor_unref(monitor);
1530 udev_ctrl_connection_unref(ctrl_conn);
1531 udev_ctrl_unref(udev_ctrl);
1532 udev_list_cleanup(&properties_list);
1533 mac_selinux_finish();