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)
62 log_metav(priority, file, line, fn, format, args);
65 static struct udev_rules *rules;
66 static struct udev_ctrl *udev_ctrl;
67 static struct udev_monitor *monitor;
68 static int worker_watch[2] = { -1, -1 };
69 static int fd_signal = -1;
70 static int fd_ep = -1;
71 static int fd_inotify = -1;
72 static bool stop_exec_queue;
75 static int children_max;
76 static int exec_delay;
77 static usec_t event_timeout_usec = 30 * USEC_PER_SEC;
78 static sigset_t sigmask_orig;
79 static UDEV_LIST(event_list);
80 static UDEV_LIST(worker_list);
81 static char *udev_cgroup;
82 static bool udev_exit;
91 struct udev_list_node node;
93 struct udev_device *dev;
94 enum event_state state;
96 unsigned long long int delaying_seqnum;
97 unsigned long long int seqnum;
100 const char *devpath_old;
109 static inline struct event *node_to_event(struct udev_list_node *node)
111 return container_of(node, struct event, node);
114 static void event_queue_cleanup(struct udev *udev, enum event_state type);
124 struct udev_list_node node;
128 struct udev_monitor *monitor;
129 enum worker_state state;
131 usec_t event_start_usec;
134 /* passed from worker to main process */
135 struct worker_message {
140 static inline struct worker *node_to_worker(struct udev_list_node *node)
142 return container_of(node, struct worker, node);
145 static void event_queue_delete(struct event *event)
147 udev_list_node_remove(&event->node);
148 udev_device_unref(event->dev);
152 static struct worker *worker_ref(struct worker *worker)
158 static void worker_cleanup(struct worker *worker)
160 udev_list_node_remove(&worker->node);
161 udev_monitor_unref(worker->monitor);
166 static void worker_unref(struct worker *worker)
169 if (worker->refcount > 0)
171 log_debug("worker [%u] cleaned up", worker->pid);
172 worker_cleanup(worker);
175 static void worker_list_cleanup(struct udev *udev)
177 struct udev_list_node *loop, *tmp;
179 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
180 struct worker *worker = node_to_worker(loop);
182 worker_cleanup(worker);
186 static void worker_new(struct event *event)
188 struct udev *udev = event->udev;
189 struct worker *worker;
190 struct udev_monitor *worker_monitor;
193 /* listen for new events */
194 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
195 if (worker_monitor == NULL)
197 /* allow the main daemon netlink address to send devices to the worker */
198 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
199 udev_monitor_enable_receiving(worker_monitor);
201 worker = new0(struct worker, 1);
202 if (worker == NULL) {
203 udev_monitor_unref(worker_monitor);
206 /* worker + event reference */
207 worker->refcount = 2;
213 struct udev_device *dev = NULL;
215 struct epoll_event ep_signal, ep_monitor;
217 int rc = EXIT_SUCCESS;
219 /* take initial device from queue */
224 worker_list_cleanup(udev);
225 event_queue_cleanup(udev, EVENT_UNDEF);
226 udev_monitor_unref(monitor);
227 udev_ctrl_unref(udev_ctrl);
230 close(worker_watch[READ_END]);
233 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
235 log_error("error creating signalfd %m");
240 fd_ep = epoll_create1(EPOLL_CLOEXEC);
242 log_error("error creating epoll fd: %m");
247 memzero(&ep_signal, sizeof(struct epoll_event));
248 ep_signal.events = EPOLLIN;
249 ep_signal.data.fd = fd_signal;
251 fd_monitor = udev_monitor_get_fd(worker_monitor);
252 memzero(&ep_monitor, sizeof(struct epoll_event));
253 ep_monitor.events = EPOLLIN;
254 ep_monitor.data.fd = fd_monitor;
256 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
257 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
258 log_error("fail to add fds to epoll: %m");
263 /* request TERM signal if parent exits */
264 prctl(PR_SET_PDEATHSIG, SIGTERM);
266 /* reset OOM score, we only protect the main daemon */
267 write_string_file("/proc/self/oom_score_adj", "0");
270 struct udev_event *udev_event;
271 struct worker_message msg;
275 log_debug("seq %llu running", udev_device_get_seqnum(dev));
276 udev_event = udev_event_new(dev);
277 if (udev_event == NULL) {
282 /* needed for SIGCHLD/SIGTERM in spawn() */
283 udev_event->fd_signal = fd_signal;
286 udev_event->exec_delay = exec_delay;
289 * Take a shared lock on the device node; this establishes
290 * a concept of device "ownership" to serialize device
291 * access. External processes holding an exclusive lock will
292 * cause udev to skip the event handling; in the case udev
293 * acquired the lock, the external process can block until
294 * udev has finished its event handling.
296 if (!streq_ptr(udev_device_get_action(dev), "remove") &&
297 streq_ptr("block", udev_device_get_subsystem(dev)) &&
298 !startswith(udev_device_get_sysname(dev), "dm-") &&
299 !startswith(udev_device_get_sysname(dev), "md")) {
300 struct udev_device *d = dev;
302 if (streq_ptr("partition", udev_device_get_devtype(d)))
303 d = udev_device_get_parent(d);
306 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
307 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
308 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
310 fd_lock = safe_close(fd_lock);
316 /* apply rules, create node, symlinks */
317 udev_event_execute_rules(udev_event, event_timeout_usec, rules, &sigmask_orig);
319 udev_event_execute_run(udev_event, event_timeout_usec, &sigmask_orig);
321 /* apply/restore inotify watch */
322 if (udev_event->inotify_watch) {
323 udev_watch_begin(udev, dev);
324 udev_device_update_db(dev);
329 /* send processed event back to libudev listeners */
330 udev_monitor_send_device(worker_monitor, NULL, dev);
333 /* send udevd the result of the event execution */
334 memzero(&msg, sizeof(struct worker_message));
337 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
339 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
341 udev_device_unref(dev);
344 if (udev_event->sigterm) {
345 udev_event_unref(udev_event);
349 udev_event_unref(udev_event);
351 /* wait for more device messages from main udevd, or term signal */
352 while (dev == NULL) {
353 struct epoll_event ev[4];
357 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
361 log_error("failed to poll: %m");
365 for (i = 0; i < fdcount; i++) {
366 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
367 dev = udev_monitor_receive_device(worker_monitor);
369 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
370 struct signalfd_siginfo fdsi;
373 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
374 if (size != sizeof(struct signalfd_siginfo))
376 switch (fdsi.ssi_signo) {
385 udev_device_unref(dev);
386 safe_close(fd_signal);
389 close(worker_watch[WRITE_END]);
390 udev_rules_unref(rules);
391 udev_builtin_exit(udev);
392 udev_monitor_unref(worker_monitor);
398 udev_monitor_unref(worker_monitor);
399 event->state = EVENT_QUEUED;
401 log_error("fork of child failed: %m");
404 /* close monitor, but keep address around */
405 udev_monitor_disconnect(worker_monitor);
406 worker->monitor = worker_monitor;
408 worker->state = WORKER_RUNNING;
409 worker->event_start_usec = now(CLOCK_MONOTONIC);
410 worker->event = event;
411 event->state = EVENT_RUNNING;
412 udev_list_node_append(&worker->node, &worker_list);
414 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
419 static void event_run(struct event *event)
421 struct udev_list_node *loop;
423 udev_list_node_foreach(loop, &worker_list) {
424 struct worker *worker = node_to_worker(loop);
427 if (worker->state != WORKER_IDLE)
430 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
432 log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
433 kill(worker->pid, SIGKILL);
434 worker->state = WORKER_KILLED;
438 worker->event = event;
439 worker->state = WORKER_RUNNING;
440 worker->event_start_usec = now(CLOCK_MONOTONIC);
441 event->state = EVENT_RUNNING;
445 if (children >= children_max) {
446 if (children_max > 1)
447 log_debug("maximum number (%i) of children reached", children);
451 /* start new worker and pass initial device */
455 static int event_queue_insert(struct udev_device *dev)
459 event = new0(struct event, 1);
463 event->udev = udev_device_get_udev(dev);
465 event->seqnum = udev_device_get_seqnum(dev);
466 event->devpath = udev_device_get_devpath(dev);
467 event->devpath_len = strlen(event->devpath);
468 event->devpath_old = udev_device_get_devpath_old(dev);
469 event->devnum = udev_device_get_devnum(dev);
470 event->is_block = streq("block", udev_device_get_subsystem(dev));
471 event->ifindex = udev_device_get_ifindex(dev);
473 if (streq(udev_device_get_subsystem(dev), "firmware"))
474 event->nodelay = true;
477 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
478 udev_device_get_action(dev), udev_device_get_subsystem(dev));
480 event->state = EVENT_QUEUED;
481 udev_list_node_append(&event->node, &event_list);
485 static void worker_kill(struct udev *udev)
487 struct udev_list_node *loop;
489 udev_list_node_foreach(loop, &worker_list) {
490 struct worker *worker = node_to_worker(loop);
492 if (worker->state == WORKER_KILLED)
495 worker->state = WORKER_KILLED;
496 kill(worker->pid, SIGTERM);
500 /* lookup event for identical, parent, child device */
501 static bool is_devpath_busy(struct event *event)
503 struct udev_list_node *loop;
506 /* check if queue contains events we depend on */
507 udev_list_node_foreach(loop, &event_list) {
508 struct event *loop_event = node_to_event(loop);
510 /* we already found a later event, earlier can not block us, no need to check again */
511 if (loop_event->seqnum < event->delaying_seqnum)
514 /* event we checked earlier still exists, no need to check again */
515 if (loop_event->seqnum == event->delaying_seqnum)
518 /* found ourself, no later event can block us */
519 if (loop_event->seqnum >= event->seqnum)
522 /* check major/minor */
523 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
526 /* check network device ifindex */
527 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
530 /* check our old name */
531 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
532 event->delaying_seqnum = loop_event->seqnum;
536 /* compare devpath */
537 common = MIN(loop_event->devpath_len, event->devpath_len);
539 /* one devpath is contained in the other? */
540 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
543 /* identical device event found */
544 if (loop_event->devpath_len == event->devpath_len) {
545 /* devices names might have changed/swapped in the meantime */
546 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
548 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
550 event->delaying_seqnum = loop_event->seqnum;
555 /* allow to bypass the dependency tracking */
560 /* parent device event found */
561 if (event->devpath[common] == '/') {
562 event->delaying_seqnum = loop_event->seqnum;
566 /* child device event found */
567 if (loop_event->devpath[common] == '/') {
568 event->delaying_seqnum = loop_event->seqnum;
572 /* no matching device */
579 static void event_queue_start(struct udev *udev)
581 struct udev_list_node *loop;
583 udev_list_node_foreach(loop, &event_list) {
584 struct event *event = node_to_event(loop);
586 if (event->state != EVENT_QUEUED)
589 /* do not start event if parent or child event is still running */
590 if (is_devpath_busy(event))
597 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
599 struct udev_list_node *loop, *tmp;
601 udev_list_node_foreach_safe(loop, tmp, &event_list) {
602 struct event *event = node_to_event(loop);
604 if (match_type != EVENT_UNDEF && match_type != event->state)
607 event_queue_delete(event);
611 static void worker_returned(int fd_worker)
614 struct worker_message msg;
616 struct udev_list_node *loop;
618 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
619 if (size != sizeof(struct worker_message))
622 /* lookup worker who sent the signal */
623 udev_list_node_foreach(loop, &worker_list) {
624 struct worker *worker = node_to_worker(loop);
626 if (worker->pid != msg.pid)
629 /* worker returned */
631 worker->event->exitcode = msg.exitcode;
632 event_queue_delete(worker->event);
633 worker->event = NULL;
635 if (worker->state != WORKER_KILLED)
636 worker->state = WORKER_IDLE;
637 worker_unref(worker);
643 /* receive the udevd message from userspace */
644 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
646 struct udev *udev = udev_ctrl_get_udev(uctrl);
647 struct udev_ctrl_connection *ctrl_conn;
648 struct udev_ctrl_msg *ctrl_msg = NULL;
652 ctrl_conn = udev_ctrl_get_connection(uctrl);
653 if (ctrl_conn == NULL)
656 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
657 if (ctrl_msg == NULL)
660 i = udev_ctrl_get_set_log_level(ctrl_msg);
662 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
663 log_set_max_level(i);
664 udev_set_log_priority(udev, i);
668 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
669 log_debug("udevd message (STOP_EXEC_QUEUE) received");
670 stop_exec_queue = true;
673 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
674 log_debug("udevd message (START_EXEC_QUEUE) received");
675 stop_exec_queue = false;
678 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
679 log_debug("udevd message (RELOAD) received");
683 str = udev_ctrl_get_set_env(ctrl_msg);
691 val = strchr(key, '=');
695 if (val[0] == '\0') {
696 log_debug("udevd message (ENV) received, unset '%s'", key);
697 udev_add_property(udev, key, NULL);
699 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
700 udev_add_property(udev, key, val);
703 log_error("wrong key format '%s'", key);
710 i = udev_ctrl_get_set_children_max(ctrl_msg);
712 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
716 if (udev_ctrl_get_ping(ctrl_msg) > 0)
717 log_debug("udevd message (SYNC) received");
719 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
720 log_debug("udevd message (EXIT) received");
722 /* keep reference to block the client until we exit */
723 udev_ctrl_connection_ref(ctrl_conn);
726 udev_ctrl_msg_unref(ctrl_msg);
727 return udev_ctrl_connection_unref(ctrl_conn);
730 static int synthesize_change(struct udev_device *dev) {
731 char filename[UTIL_PATH_SIZE];
734 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
735 streq_ptr("disk", udev_device_get_devtype(dev)) &&
736 !startswith(udev_device_get_sysname(dev), "dm-")) {
737 bool part_table_read = false;
738 bool has_partitions = false;
740 struct udev *udev = udev_device_get_udev(dev);
741 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
742 struct udev_list_entry *item;
745 * Try to re-read the partition table. This only succeeds if
746 * none of the devices is busy. The kernel returns 0 if no
747 * partition table is found, and we will not get an event for
750 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
752 r = flock(fd, LOCK_EX|LOCK_NB);
754 r = ioctl(fd, BLKRRPART, 0);
758 part_table_read = true;
761 /* search for partitions */
762 e = udev_enumerate_new(udev);
766 r = udev_enumerate_add_match_parent(e, dev);
770 r = udev_enumerate_add_match_subsystem(e, "block");
774 r = udev_enumerate_scan_devices(e);
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 has_partitions = true;
793 * We have partitions and re-read the table, the kernel already sent
794 * out a "change" event for the disk, and "remove/add" for all
797 if (part_table_read && has_partitions)
801 * We have partitions but re-reading the partition table did not
802 * work, synthesize "change" for the disk and all partitions.
804 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
805 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
806 write_string_file(filename, "change");
808 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
809 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
811 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
815 if (!streq_ptr("partition", udev_device_get_devtype(d)))
818 log_debug("device %s closed, synthesising partition '%s' 'change'",
819 udev_device_get_devnode(dev), udev_device_get_devnode(d));
820 strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
821 write_string_file(filename, "change");
827 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
828 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
829 write_string_file(filename, "change");
834 static int handle_inotify(struct udev *udev)
838 struct inotify_event *ev;
841 r = ioctl(fd_inotify, FIONREAD, &nbytes);
842 if (r < 0 || nbytes <= 0)
845 buf = malloc(nbytes);
847 log_error("error getting buffer for inotify");
851 nbytes = read(fd_inotify, buf, nbytes);
853 for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
854 struct udev_device *dev;
856 ev = (struct inotify_event *)(buf + pos);
857 dev = udev_watch_lookup(udev, ev->wd);
861 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
862 if (ev->mask & IN_CLOSE_WRITE)
863 synthesize_change(dev);
864 else if (ev->mask & IN_IGNORED)
865 udev_watch_end(udev, dev);
867 udev_device_unref(dev);
874 static void handle_signal(struct udev *udev, int signo)
885 struct udev_list_node *loop, *tmp;
887 pid = waitpid(-1, &status, WNOHANG);
891 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
892 struct worker *worker = node_to_worker(loop);
894 if (worker->pid != pid)
896 log_debug("worker [%u] exit", pid);
898 if (WIFEXITED(status)) {
899 if (WEXITSTATUS(status) != 0)
900 log_error("worker [%u] exit with return code %i",
901 pid, WEXITSTATUS(status));
902 } else if (WIFSIGNALED(status)) {
903 log_error("worker [%u] terminated by signal %i (%s)",
904 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
905 } else if (WIFSTOPPED(status)) {
906 log_error("worker [%u] stopped", pid);
907 } else if (WIFCONTINUED(status)) {
908 log_error("worker [%u] continued", pid);
910 log_error("worker [%u] exit with status 0x%04x", pid, status);
913 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
915 log_error("worker [%u] failed while handling '%s'",
916 pid, worker->event->devpath);
917 worker->event->exitcode = -32;
918 event_queue_delete(worker->event);
920 /* drop reference taken for state 'running' */
921 worker_unref(worker);
924 worker_unref(worker);
935 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
937 int ctrl = -1, netlink = -1;
940 n = sd_listen_fds(true);
944 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
945 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
952 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
962 if (ctrl < 0 || netlink < 0)
965 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
972 * read the kernel commandline, in case we need to get into debug mode
973 * udev.log-priority=<level> syslog priority
974 * udev.children-max=<number of workers> events are fully serialized if set to 1
975 * udev.exec-delay=<number of seconds> delay execution of every executed program
977 static void kernel_cmdline_options(struct udev *udev)
979 _cleanup_free_ char *line = NULL;
984 r = proc_cmdline(&line);
986 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
990 FOREACH_WORD_QUOTED(w, l, line, state) {
997 /* accept the same options for the initrd, prefixed with "rd." */
998 if (in_initrd() && startswith(s, "rd."))
1003 if (startswith(opt, "udev.log-priority=")) {
1006 prio = util_log_priority(opt + 18);
1007 log_set_max_level(prio);
1008 udev_set_log_priority(udev, prio);
1009 } else if (startswith(opt, "udev.children-max=")) {
1010 children_max = strtoul(opt + 18, NULL, 0);
1011 } else if (startswith(opt, "udev.exec-delay=")) {
1012 exec_delay = strtoul(opt + 16, NULL, 0);
1013 } else if (startswith(opt, "udev.event-timeout=")) {
1014 event_timeout_usec = strtoul(opt + 16, NULL, 0) * USEC_PER_SEC;
1021 int main(int argc, char *argv[]) {
1024 int daemonize = false;
1025 int resolve_names = 1;
1026 static const struct option options[] = {
1027 { "daemon", no_argument, NULL, 'd' },
1028 { "debug", no_argument, NULL, 'D' },
1029 { "children-max", required_argument, NULL, 'c' },
1030 { "exec-delay", required_argument, NULL, 'e' },
1031 { "event-timeout", required_argument, NULL, 't' },
1032 { "resolve-names", required_argument, NULL, 'N' },
1033 { "help", no_argument, NULL, 'h' },
1034 { "version", no_argument, NULL, 'V' },
1038 int fd_netlink = -1;
1040 struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1041 struct udev_ctrl_connection *ctrl_conn = NULL;
1048 log_set_target(LOG_TARGET_AUTO);
1049 log_parse_environment();
1052 udev_set_log_fn(udev, udev_main_log);
1053 log_set_max_level(udev_get_log_priority(udev));
1055 log_debug("version %s", VERSION);
1061 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
1070 children_max = strtoul(optarg, NULL, 0);
1073 exec_delay = strtoul(optarg, NULL, 0);
1076 event_timeout_usec = strtoul(optarg, NULL, 0) * USEC_PER_SEC;
1080 log_set_max_level(LOG_DEBUG);
1081 udev_set_log_priority(udev, LOG_DEBUG);
1084 if (streq(optarg, "early")) {
1086 } else if (streq(optarg, "late")) {
1088 } else if (streq(optarg, "never")) {
1091 fprintf(stderr, "resolve-names must be early, late or never\n");
1092 log_error("resolve-names must be early, late or never");
1097 printf("Usage: udevd OPTIONS\n"
1100 " --children-max=<maximum number of workers>\n"
1101 " --exec-delay=<seconds to wait before executing RUN=>\n"
1102 " --event-timeout=<seconds to wait before terminating an event>\n"
1103 " --resolve-names=early|late|never\n"
1109 printf("%s\n", VERSION);
1116 kernel_cmdline_options(udev);
1118 if (getuid() != 0) {
1119 fprintf(stderr, "root privileges required\n");
1120 log_error("root privileges required");
1124 /* set umask before creating any file/directory */
1128 mkdir("/run/udev", 0755);
1132 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1136 fd = open("/dev/null", O_RDWR);
1138 if (write(STDOUT_FILENO, 0, 0) < 0)
1139 dup2(fd, STDOUT_FILENO);
1140 if (write(STDERR_FILENO, 0, 0) < 0)
1141 dup2(fd, STDERR_FILENO);
1142 if (fd > STDERR_FILENO)
1145 fprintf(stderr, "cannot open /dev/null\n");
1146 log_error("cannot open /dev/null");
1150 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1151 /* get control and netlink socket from systemd */
1152 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1153 if (udev_ctrl == NULL) {
1154 log_error("error taking over udev control socket");
1159 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1160 if (monitor == NULL) {
1161 log_error("error taking over netlink socket");
1166 /* get our own cgroup, we regularly kill everything udev has left behind */
1167 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1170 /* open control and netlink socket */
1171 udev_ctrl = udev_ctrl_new(udev);
1172 if (udev_ctrl == NULL) {
1173 fprintf(stderr, "error initializing udev control socket");
1174 log_error("error initializing udev control socket");
1178 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1180 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1181 if (monitor == NULL) {
1182 fprintf(stderr, "error initializing netlink socket\n");
1183 log_error("error initializing netlink socket");
1187 fd_netlink = udev_monitor_get_fd(monitor);
1190 if (udev_monitor_enable_receiving(monitor) < 0) {
1191 fprintf(stderr, "error binding netlink socket\n");
1192 log_error("error binding netlink socket");
1197 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1198 fprintf(stderr, "error binding udev control socket\n");
1199 log_error("error binding udev control socket");
1204 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1214 log_error("fork of daemon failed: %m");
1219 goto exit_daemonize;
1224 write_string_file("/proc/self/oom_score_adj", "-1000");
1226 sd_notify(1, "READY=1");
1229 print_kmsg("starting version " VERSION "\n");
1234 fd = open("/dev/null", O_RDWR);
1236 dup2(fd, STDIN_FILENO);
1237 dup2(fd, STDOUT_FILENO);
1238 dup2(fd, STDERR_FILENO);
1243 fd_inotify = udev_watch_init(udev);
1244 if (fd_inotify < 0) {
1245 fprintf(stderr, "error initializing inotify\n");
1246 log_error("error initializing inotify");
1250 udev_watch_restore(udev);
1252 /* block and listen to all signals on signalfd */
1254 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1255 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1256 if (fd_signal < 0) {
1257 fprintf(stderr, "error creating signalfd\n");
1258 log_error("error creating signalfd");
1263 /* unnamed socket from workers to the main daemon */
1264 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1265 fprintf(stderr, "error creating socketpair\n");
1266 log_error("error creating socketpair");
1270 fd_worker = worker_watch[READ_END];
1272 udev_builtin_init(udev);
1274 rules = udev_rules_new(udev, resolve_names);
1275 if (rules == NULL) {
1276 log_error("error reading rules");
1280 memzero(&ep_ctrl, sizeof(struct epoll_event));
1281 ep_ctrl.events = EPOLLIN;
1282 ep_ctrl.data.fd = fd_ctrl;
1284 memzero(&ep_inotify, sizeof(struct epoll_event));
1285 ep_inotify.events = EPOLLIN;
1286 ep_inotify.data.fd = fd_inotify;
1288 memzero(&ep_signal, sizeof(struct epoll_event));
1289 ep_signal.events = EPOLLIN;
1290 ep_signal.data.fd = fd_signal;
1292 memzero(&ep_netlink, sizeof(struct epoll_event));
1293 ep_netlink.events = EPOLLIN;
1294 ep_netlink.data.fd = fd_netlink;
1296 memzero(&ep_worker, sizeof(struct epoll_event));
1297 ep_worker.events = EPOLLIN;
1298 ep_worker.data.fd = fd_worker;
1300 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1302 log_error("error creating epoll fd: %m");
1305 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1306 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1307 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1308 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1309 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1310 log_error("fail to add fds to epoll: %m");
1314 if (children_max <= 0) {
1319 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1320 children_max += CPU_COUNT(&cpu_set) * 2;
1323 log_debug("set children_max to %u", children_max);
1325 rc = udev_rules_apply_static_dev_perms(rules);
1327 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1329 udev_list_node_init(&event_list);
1330 udev_list_node_init(&worker_list);
1333 static usec_t last_usec;
1334 struct epoll_event ev[8];
1337 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1341 /* close sources of new events and discard buffered events */
1343 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1346 if (monitor != NULL) {
1347 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1348 udev_monitor_unref(monitor);
1351 if (fd_inotify >= 0) {
1352 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1357 /* discard queued events and kill workers */
1358 event_queue_cleanup(udev, EVENT_QUEUED);
1361 /* exit after all has cleaned up */
1362 if (udev_list_node_is_empty(&event_list) && children == 0)
1365 /* timeout at exit for workers to finish */
1366 timeout = 30 * MSEC_PER_SEC;
1367 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1371 /* cleanup possible left-over processes in our cgroup */
1373 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1375 /* kill idle or hanging workers */
1376 timeout = 3 * MSEC_PER_SEC;
1379 /* tell settle that we are busy or idle */
1380 if (!udev_list_node_is_empty(&event_list)) {
1383 fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1387 unlink("/run/udev/queue");
1390 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1395 struct udev_list_node *loop;
1399 log_error("timeout, giving up waiting for workers to finish");
1403 /* kill idle workers */
1404 if (udev_list_node_is_empty(&event_list)) {
1405 log_debug("cleanup idle workers");
1409 /* check for hanging events */
1410 udev_list_node_foreach(loop, &worker_list) {
1411 struct worker *worker = node_to_worker(loop);
1413 if (worker->state != WORKER_RUNNING)
1416 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > event_timeout_usec) {
1417 log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
1418 kill(worker->pid, SIGKILL);
1419 worker->state = WORKER_KILLED;
1421 /* drop reference taken for state 'running' */
1422 worker_unref(worker);
1423 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1424 worker->event->exitcode = -64;
1425 event_queue_delete(worker->event);
1426 worker->event = NULL;
1432 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1433 for (i = 0; i < fdcount; i++) {
1434 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1436 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1438 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1440 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1442 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1446 /* check for changed config, every 3 seconds at most */
1447 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1448 if (udev_rules_check_timestamp(rules))
1450 if (udev_builtin_validate(udev))
1453 last_usec = now(CLOCK_MONOTONIC);
1456 /* reload requested, HUP signal received, rules changed, builtin changed */
1459 rules = udev_rules_unref(rules);
1460 udev_builtin_exit(udev);
1464 /* event has finished */
1466 worker_returned(fd_worker);
1469 struct udev_device *dev;
1471 dev = udev_monitor_receive_device(monitor);
1473 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1474 if (event_queue_insert(dev) < 0)
1475 udev_device_unref(dev);
1479 /* start new events */
1480 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1481 udev_builtin_init(udev);
1483 rules = udev_rules_new(udev, resolve_names);
1485 event_queue_start(udev);
1489 struct signalfd_siginfo fdsi;
1492 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1493 if (size == sizeof(struct signalfd_siginfo))
1494 handle_signal(udev, fdsi.ssi_signo);
1497 /* we are shutting down, the events below are not handled anymore */
1501 /* device node watch */
1503 handle_inotify(udev);
1506 * This needs to be after the inotify handling, to make sure,
1507 * that the ping is send back after the possibly generated
1508 * "change" events by the inotify device node watch.
1510 * A single time we may receive a client connection which we need to
1511 * keep open to block the client. It will be closed right before we
1515 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1520 udev_ctrl_cleanup(udev_ctrl);
1521 unlink("/run/udev/queue");
1525 worker_list_cleanup(udev);
1526 event_queue_cleanup(udev, EVENT_UNDEF);
1527 udev_rules_unref(rules);
1528 udev_builtin_exit(udev);
1531 if (worker_watch[READ_END] >= 0)
1532 close(worker_watch[READ_END]);
1533 if (worker_watch[WRITE_END] >= 0)
1534 close(worker_watch[WRITE_END]);
1535 udev_monitor_unref(monitor);
1536 udev_ctrl_connection_unref(ctrl_conn);
1537 udev_ctrl_unref(udev_ctrl);