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 struct udev_device *dev_kernel;
87 enum event_state state;
89 unsigned long long int delaying_seqnum;
90 unsigned long long int seqnum;
93 const char *devpath_old;
99 static inline struct event *node_to_event(struct udev_list_node *node) {
100 return container_of(node, struct event, node);
103 static void event_queue_cleanup(struct udev *udev, enum event_state type);
113 struct udev_list_node node;
117 struct udev_monitor *monitor;
118 enum worker_state state;
120 usec_t event_start_usec;
124 /* passed from worker to main process */
125 struct worker_message {
130 static inline struct worker *node_to_worker(struct udev_list_node *node) {
131 return container_of(node, struct worker, node);
134 static void event_queue_delete(struct event *event) {
135 udev_list_node_remove(&event->node);
136 udev_device_unref(event->dev);
137 udev_device_unref(event->dev_kernel);
141 static struct worker *worker_ref(struct worker *worker) {
146 static void worker_cleanup(struct worker *worker) {
147 udev_list_node_remove(&worker->node);
148 udev_monitor_unref(worker->monitor);
153 static void worker_unref(struct worker *worker) {
155 if (worker->refcount > 0)
157 log_debug("worker ["PID_FMT"] cleaned up", worker->pid);
158 worker_cleanup(worker);
161 static void worker_list_cleanup(struct udev *udev) {
162 struct udev_list_node *loop, *tmp;
164 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
165 struct worker *worker = node_to_worker(loop);
167 worker_cleanup(worker);
171 static void worker_new(struct event *event) {
172 struct udev *udev = event->udev;
173 struct worker *worker;
174 struct udev_monitor *worker_monitor;
177 /* listen for new events */
178 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
179 if (worker_monitor == NULL)
181 /* allow the main daemon netlink address to send devices to the worker */
182 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
183 udev_monitor_enable_receiving(worker_monitor);
185 worker = new0(struct worker, 1);
186 if (worker == NULL) {
187 udev_monitor_unref(worker_monitor);
190 /* worker + event reference */
191 worker->refcount = 2;
197 struct udev_device *dev = NULL;
199 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
200 struct epoll_event ep_signal, ep_monitor;
202 int rc = EXIT_SUCCESS;
204 /* take initial device from queue */
209 worker_list_cleanup(udev);
210 event_queue_cleanup(udev, EVENT_UNDEF);
211 udev_monitor_unref(monitor);
212 udev_ctrl_unref(udev_ctrl);
215 close(worker_watch[READ_END]);
218 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
220 log_error_errno(errno, "error creating signalfd %m");
225 fd_ep = epoll_create1(EPOLL_CLOEXEC);
227 log_error_errno(errno, "error creating epoll fd: %m");
232 memzero(&ep_signal, sizeof(struct epoll_event));
233 ep_signal.events = EPOLLIN;
234 ep_signal.data.fd = fd_signal;
236 fd_monitor = udev_monitor_get_fd(worker_monitor);
237 memzero(&ep_monitor, sizeof(struct epoll_event));
238 ep_monitor.events = EPOLLIN;
239 ep_monitor.data.fd = fd_monitor;
241 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
242 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
243 log_error_errno(errno, "fail to add fds to epoll: %m");
248 /* request TERM signal if parent exits */
249 prctl(PR_SET_PDEATHSIG, SIGTERM);
251 /* reset OOM score, we only protect the main daemon */
252 write_string_file("/proc/self/oom_score_adj", "0");
255 struct udev_event *udev_event;
256 struct worker_message msg;
260 log_debug("seq %llu running", udev_device_get_seqnum(dev));
261 udev_event = udev_event_new(dev);
262 if (udev_event == NULL) {
267 /* needed for SIGCHLD/SIGTERM in spawn() */
268 udev_event->fd_signal = fd_signal;
270 if (arg_exec_delay > 0)
271 udev_event->exec_delay = arg_exec_delay;
274 * Take a shared lock on the device node; this establishes
275 * a concept of device "ownership" to serialize device
276 * access. External processes holding an exclusive lock will
277 * cause udev to skip the event handling; in the case udev
278 * acquired the lock, the external process can block until
279 * udev has finished its event handling.
281 if (!streq_ptr(udev_device_get_action(dev), "remove") &&
282 streq_ptr("block", udev_device_get_subsystem(dev)) &&
283 !startswith(udev_device_get_sysname(dev), "dm-") &&
284 !startswith(udev_device_get_sysname(dev), "md")) {
285 struct udev_device *d = dev;
287 if (streq_ptr("partition", udev_device_get_devtype(d)))
288 d = udev_device_get_parent(d);
291 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
292 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
293 log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
295 fd_lock = safe_close(fd_lock);
301 /* needed for renaming netifs */
302 udev_event->rtnl = rtnl;
304 /* apply rules, create node, symlinks */
305 udev_event_execute_rules(udev_event,
306 arg_event_timeout_usec, arg_event_timeout_warn_usec,
311 udev_event_execute_run(udev_event,
312 arg_event_timeout_usec, arg_event_timeout_warn_usec,
315 if (udev_event->rtnl)
316 /* in case rtnl was initialized */
317 rtnl = sd_rtnl_ref(udev_event->rtnl);
319 /* apply/restore inotify watch */
320 if (udev_event->inotify_watch) {
321 udev_watch_begin(udev, dev);
322 udev_device_update_db(dev);
327 /* send processed event back to libudev listeners */
328 udev_monitor_send_device(worker_monitor, NULL, dev);
331 /* send udevd the result of the event execution */
332 memzero(&msg, sizeof(struct worker_message));
335 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
337 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
339 udev_device_unref(dev);
342 if (udev_event->sigterm) {
343 udev_event_unref(udev_event);
347 udev_event_unref(udev_event);
349 /* wait for more device messages from main udevd, or term signal */
350 while (dev == NULL) {
351 struct epoll_event ev[4];
355 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
359 log_error_errno(errno, "failed to poll: %m");
363 for (i = 0; i < fdcount; i++) {
364 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
365 dev = udev_monitor_receive_device(worker_monitor);
367 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
368 struct signalfd_siginfo fdsi;
371 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
372 if (size != sizeof(struct signalfd_siginfo))
374 switch (fdsi.ssi_signo) {
383 udev_device_unref(dev);
384 safe_close(fd_signal);
387 close(worker_watch[WRITE_END]);
388 udev_rules_unref(rules);
389 udev_builtin_exit(udev);
390 udev_monitor_unref(worker_monitor);
396 udev_monitor_unref(worker_monitor);
397 event->state = EVENT_QUEUED;
399 log_error_errno(errno, "fork of child failed: %m");
402 /* close monitor, but keep address around */
403 udev_monitor_disconnect(worker_monitor);
404 worker->monitor = worker_monitor;
406 worker->state = WORKER_RUNNING;
407 worker->event_start_usec = now(CLOCK_MONOTONIC);
408 worker->event_warned = false;
409 worker->event = event;
410 event->state = EVENT_RUNNING;
411 udev_list_node_append(&worker->node, &worker_list);
413 log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
418 static void event_run(struct event *event) {
419 struct udev_list_node *loop;
421 udev_list_node_foreach(loop, &worker_list) {
422 struct worker *worker = node_to_worker(loop);
425 if (worker->state != WORKER_IDLE)
428 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
430 log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
432 kill(worker->pid, SIGKILL);
433 worker->state = WORKER_KILLED;
437 worker->event = event;
438 worker->state = WORKER_RUNNING;
439 worker->event_start_usec = now(CLOCK_MONOTONIC);
440 worker->event_warned = false;
441 event->state = EVENT_RUNNING;
445 if (children >= arg_children_max) {
446 if (arg_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) {
458 event = new0(struct event, 1);
462 event->udev = udev_device_get_udev(dev);
464 event->dev_kernel = udev_device_shallow_clone(dev);
465 udev_device_copy_properties(event->dev_kernel, dev);
466 event->seqnum = udev_device_get_seqnum(dev);
467 event->devpath = udev_device_get_devpath(dev);
468 event->devpath_len = strlen(event->devpath);
469 event->devpath_old = udev_device_get_devpath_old(dev);
470 event->devnum = udev_device_get_devnum(dev);
471 event->is_block = streq("block", udev_device_get_subsystem(dev));
472 event->ifindex = udev_device_get_ifindex(dev);
474 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
475 udev_device_get_action(dev), udev_device_get_subsystem(dev));
477 event->state = EVENT_QUEUED;
478 udev_list_node_append(&event->node, &event_list);
482 static void worker_kill(struct udev *udev) {
483 struct udev_list_node *loop;
485 udev_list_node_foreach(loop, &worker_list) {
486 struct worker *worker = node_to_worker(loop);
488 if (worker->state == WORKER_KILLED)
491 worker->state = WORKER_KILLED;
492 kill(worker->pid, SIGTERM);
496 /* lookup event for identical, parent, child device */
497 static bool is_devpath_busy(struct event *event) {
498 struct udev_list_node *loop;
501 /* check if queue contains events we depend on */
502 udev_list_node_foreach(loop, &event_list) {
503 struct event *loop_event = node_to_event(loop);
505 /* we already found a later event, earlier can not block us, no need to check again */
506 if (loop_event->seqnum < event->delaying_seqnum)
509 /* event we checked earlier still exists, no need to check again */
510 if (loop_event->seqnum == event->delaying_seqnum)
513 /* found ourself, no later event can block us */
514 if (loop_event->seqnum >= event->seqnum)
517 /* check major/minor */
518 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
521 /* check network device ifindex */
522 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
525 /* check our old name */
526 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
527 event->delaying_seqnum = loop_event->seqnum;
531 /* compare devpath */
532 common = MIN(loop_event->devpath_len, event->devpath_len);
534 /* one devpath is contained in the other? */
535 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
538 /* identical device event found */
539 if (loop_event->devpath_len == event->devpath_len) {
540 /* devices names might have changed/swapped in the meantime */
541 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
543 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
545 event->delaying_seqnum = loop_event->seqnum;
549 /* parent device event found */
550 if (event->devpath[common] == '/') {
551 event->delaying_seqnum = loop_event->seqnum;
555 /* child device event found */
556 if (loop_event->devpath[common] == '/') {
557 event->delaying_seqnum = loop_event->seqnum;
561 /* no matching device */
568 static void event_queue_start(struct udev *udev) {
569 struct udev_list_node *loop;
571 udev_list_node_foreach(loop, &event_list) {
572 struct event *event = node_to_event(loop);
574 if (event->state != EVENT_QUEUED)
577 /* do not start event if parent or child event is still running */
578 if (is_devpath_busy(event))
585 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
586 struct udev_list_node *loop, *tmp;
588 udev_list_node_foreach_safe(loop, tmp, &event_list) {
589 struct event *event = node_to_event(loop);
591 if (match_type != EVENT_UNDEF && match_type != event->state)
594 event_queue_delete(event);
598 static void worker_returned(int fd_worker) {
600 struct worker_message msg;
602 struct udev_list_node *loop;
604 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
605 if (size != sizeof(struct worker_message))
608 /* lookup worker who sent the signal */
609 udev_list_node_foreach(loop, &worker_list) {
610 struct worker *worker = node_to_worker(loop);
612 if (worker->pid != msg.pid)
615 /* worker returned */
617 worker->event->exitcode = msg.exitcode;
618 event_queue_delete(worker->event);
619 worker->event = NULL;
621 if (worker->state != WORKER_KILLED)
622 worker->state = WORKER_IDLE;
623 worker_unref(worker);
629 /* receive the udevd message from userspace */
630 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
631 struct udev *udev = udev_ctrl_get_udev(uctrl);
632 struct udev_ctrl_connection *ctrl_conn;
633 struct udev_ctrl_msg *ctrl_msg = NULL;
637 ctrl_conn = udev_ctrl_get_connection(uctrl);
638 if (ctrl_conn == NULL)
641 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
642 if (ctrl_msg == NULL)
645 i = udev_ctrl_get_set_log_level(ctrl_msg);
647 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
648 log_set_max_level(i);
652 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
653 log_debug("udevd message (STOP_EXEC_QUEUE) received");
654 stop_exec_queue = true;
657 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
658 log_debug("udevd message (START_EXEC_QUEUE) received");
659 stop_exec_queue = false;
662 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
663 log_debug("udevd message (RELOAD) received");
667 str = udev_ctrl_get_set_env(ctrl_msg);
675 val = strchr(key, '=');
679 if (val[0] == '\0') {
680 log_debug("udevd message (ENV) received, unset '%s'", key);
681 udev_list_entry_add(&properties_list, key, NULL);
683 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
684 udev_list_entry_add(&properties_list, key, val);
687 log_error("wrong key format '%s'", key);
694 i = udev_ctrl_get_set_children_max(ctrl_msg);
696 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
697 arg_children_max = i;
700 if (udev_ctrl_get_ping(ctrl_msg) > 0)
701 log_debug("udevd message (SYNC) received");
703 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
704 log_debug("udevd message (EXIT) received");
706 /* keep reference to block the client until we exit */
707 udev_ctrl_connection_ref(ctrl_conn);
710 udev_ctrl_msg_unref(ctrl_msg);
711 return udev_ctrl_connection_unref(ctrl_conn);
714 static int synthesize_change(struct udev_device *dev) {
715 char filename[UTIL_PATH_SIZE];
718 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
719 streq_ptr("disk", udev_device_get_devtype(dev)) &&
720 !startswith(udev_device_get_sysname(dev), "dm-")) {
721 bool part_table_read = false;
722 bool has_partitions = false;
724 struct udev *udev = udev_device_get_udev(dev);
725 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
726 struct udev_list_entry *item;
729 * Try to re-read the partition table. This only succeeds if
730 * none of the devices is busy. The kernel returns 0 if no
731 * partition table is found, and we will not get an event for
734 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
736 r = flock(fd, LOCK_EX|LOCK_NB);
738 r = ioctl(fd, BLKRRPART, 0);
742 part_table_read = true;
745 /* search for partitions */
746 e = udev_enumerate_new(udev);
750 r = udev_enumerate_add_match_parent(e, dev);
754 r = udev_enumerate_add_match_subsystem(e, "block");
758 r = udev_enumerate_scan_devices(e);
762 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
763 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
765 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
769 if (!streq_ptr("partition", udev_device_get_devtype(d)))
772 has_partitions = true;
777 * We have partitions and re-read the table, the kernel already sent
778 * out a "change" event for the disk, and "remove/add" for all
781 if (part_table_read && has_partitions)
785 * We have partitions but re-reading the partition table did not
786 * work, synthesize "change" for the disk and all partitions.
788 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
789 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
790 write_string_file(filename, "change");
792 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
793 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
795 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
799 if (!streq_ptr("partition", udev_device_get_devtype(d)))
802 log_debug("device %s closed, synthesising partition '%s' 'change'",
803 udev_device_get_devnode(dev), udev_device_get_devnode(d));
804 strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
805 write_string_file(filename, "change");
811 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
812 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
813 write_string_file(filename, "change");
818 static int handle_inotify(struct udev *udev) {
819 union inotify_event_buffer buffer;
820 struct inotify_event *e;
823 l = read(fd_inotify, &buffer, sizeof(buffer));
825 if (errno == EAGAIN || errno == EINTR)
828 return log_error_errno(errno, "Failed to read inotify fd: %m");
831 FOREACH_INOTIFY_EVENT(e, buffer, l) {
832 struct udev_device *dev;
834 dev = udev_watch_lookup(udev, e->wd);
838 log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
839 if (e->mask & IN_CLOSE_WRITE)
840 synthesize_change(dev);
841 else if (e->mask & IN_IGNORED)
842 udev_watch_end(udev, dev);
844 udev_device_unref(dev);
850 static void handle_signal(struct udev *udev, int signo) {
860 struct udev_list_node *loop, *tmp;
862 pid = waitpid(-1, &status, WNOHANG);
866 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
867 struct worker *worker = node_to_worker(loop);
869 if (worker->pid != pid)
871 log_debug("worker ["PID_FMT"] exit", pid);
873 if (WIFEXITED(status)) {
874 if (WEXITSTATUS(status) != 0)
875 log_error("worker ["PID_FMT"] exit with return code %i",
876 pid, WEXITSTATUS(status));
877 } else if (WIFSIGNALED(status)) {
878 log_error("worker ["PID_FMT"] terminated by signal %i (%s)",
879 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
880 } else if (WIFSTOPPED(status)) {
881 log_error("worker ["PID_FMT"] stopped", pid);
882 } else if (WIFCONTINUED(status)) {
883 log_error("worker ["PID_FMT"] continued", pid);
885 log_error("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
888 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
890 log_error("worker ["PID_FMT"] failed while handling '%s'",
891 pid, worker->event->devpath);
892 worker->event->exitcode = -32;
893 /* delete state from disk */
894 udev_device_delete_db(worker->event->dev);
895 udev_device_tag_index(worker->event->dev, NULL, false);
896 /* forward kernel event without ammending it */
897 udev_monitor_send_device(monitor, NULL, worker->event->dev_kernel);
898 event_queue_delete(worker->event);
900 /* drop reference taken for state 'running' */
901 worker_unref(worker);
904 worker_unref(worker);
915 static void event_queue_update(void) {
918 if (!udev_list_node_is_empty(&event_list)) {
919 r = touch("/run/udev/queue");
921 log_warning_errno(r, "could not touch /run/udev/queue: %m");
923 r = unlink("/run/udev/queue");
924 if (r < 0 && errno != ENOENT)
925 log_warning("could not unlink /run/udev/queue: %m");
929 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
930 int ctrl = -1, netlink = -1;
933 n = sd_listen_fds(true);
937 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
938 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
945 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
955 if (ctrl < 0 || netlink < 0)
958 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
965 * read the kernel command line, in case we need to get into debug mode
966 * udev.log-priority=<level> syslog priority
967 * udev.children-max=<number of workers> events are fully serialized if set to 1
968 * udev.exec-delay=<number of seconds> delay execution of every executed program
970 static void kernel_cmdline_options(struct udev *udev) {
971 _cleanup_free_ char *line = NULL;
972 const char *word, *state;
976 r = proc_cmdline(&line);
978 log_warning_errno(r, "Failed to read /proc/cmdline, ignoring: %m");
982 FOREACH_WORD_QUOTED(word, l, line, state) {
983 char *s, *opt, *value;
985 s = strndup(word, l);
989 /* accept the same options for the initrd, prefixed with "rd." */
990 if (in_initrd() && startswith(s, "rd."))
995 if ((value = startswith(opt, "udev.log-priority="))) {
998 prio = util_log_priority(value);
999 log_set_max_level(prio);
1000 } else if ((value = startswith(opt, "udev.children-max="))) {
1001 r = safe_atoi(value, &arg_children_max);
1003 log_warning("Invalid udev.children-max ignored: %s", value);
1004 } else if ((value = startswith(opt, "udev.exec-delay="))) {
1005 r = safe_atoi(value, &arg_exec_delay);
1007 log_warning("Invalid udev.exec-delay ignored: %s", value);
1008 } else if ((value = startswith(opt, "udev.event-timeout="))) {
1009 r = safe_atou64(value, &arg_event_timeout_usec);
1011 log_warning("Invalid udev.event-timeout ignored: %s", value);
1014 arg_event_timeout_usec *= USEC_PER_SEC;
1015 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1022 static void help(void) {
1023 printf("%s [OPTIONS...]\n\n"
1024 "Manages devices.\n\n"
1025 " -h --help Print this message\n"
1026 " --version Print version of the program\n"
1027 " --daemon Detach and run in the background\n"
1028 " --debug Enable debug output\n"
1029 " --children-max=INT Set maximum number of workers\n"
1030 " --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
1031 " --event-timeout=SECONDS Seconds to wait before terminating an event\n"
1032 " --resolve-names=early|late|never\n"
1033 " When to resolve users and groups\n"
1034 , program_invocation_short_name);
1037 static int parse_argv(int argc, char *argv[]) {
1038 static const struct option options[] = {
1039 { "daemon", no_argument, NULL, 'd' },
1040 { "debug", no_argument, NULL, 'D' },
1041 { "children-max", required_argument, NULL, 'c' },
1042 { "exec-delay", required_argument, NULL, 'e' },
1043 { "event-timeout", required_argument, NULL, 't' },
1044 { "resolve-names", required_argument, NULL, 'N' },
1045 { "help", no_argument, NULL, 'h' },
1046 { "version", no_argument, NULL, 'V' },
1055 while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1061 arg_daemonize = true;
1064 r = safe_atoi(optarg, &arg_children_max);
1066 log_warning("Invalid --children-max ignored: %s", optarg);
1069 r = safe_atoi(optarg, &arg_exec_delay);
1071 log_warning("Invalid --exec-delay ignored: %s", optarg);
1074 r = safe_atou64(optarg, &arg_event_timeout_usec);
1076 log_warning("Invalid --event-timeout ignored: %s", optarg);
1078 arg_event_timeout_usec *= USEC_PER_SEC;
1079 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1086 if (streq(optarg, "early")) {
1087 arg_resolve_names = 1;
1088 } else if (streq(optarg, "late")) {
1089 arg_resolve_names = 0;
1090 } else if (streq(optarg, "never")) {
1091 arg_resolve_names = -1;
1093 log_error("resolve-names must be early, late or never");
1101 printf("%s\n", VERSION);
1106 assert_not_reached("Unhandled option");
1114 int main(int argc, char *argv[]) {
1118 int fd_netlink = -1;
1120 struct epoll_event ep_ctrl = { .events = EPOLLIN };
1121 struct epoll_event ep_inotify = { .events = EPOLLIN };
1122 struct epoll_event ep_signal = { .events = EPOLLIN };
1123 struct epoll_event ep_netlink = { .events = EPOLLIN };
1124 struct epoll_event ep_worker = { .events = EPOLLIN };
1125 struct udev_ctrl_connection *ctrl_conn = NULL;
1132 log_set_target(LOG_TARGET_AUTO);
1133 log_parse_environment();
1136 r = parse_argv(argc, argv);
1140 kernel_cmdline_options(udev);
1143 log_set_max_level(LOG_DEBUG);
1145 if (getuid() != 0) {
1146 log_error("root privileges required");
1150 r = mac_selinux_init("/dev");
1152 log_error_errno(r, "could not initialize labelling: %m");
1156 /* set umask before creating any file/directory */
1159 log_error_errno(errno, "could not change dir to /: %m");
1165 udev_list_init(udev, &properties_list, true);
1167 r = mkdir("/run/udev", 0755);
1168 if (r < 0 && errno != EEXIST) {
1169 log_error_errno(errno, "could not create /run/udev: %m");
1175 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1176 if (arg_daemonize) {
1179 fd = open("/dev/null", O_RDWR);
1181 if (write(STDOUT_FILENO, 0, 0) < 0)
1182 dup2(fd, STDOUT_FILENO);
1183 if (write(STDERR_FILENO, 0, 0) < 0)
1184 dup2(fd, STDERR_FILENO);
1185 if (fd > STDERR_FILENO)
1188 log_error("cannot open /dev/null");
1192 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1193 /* get control and netlink socket from systemd */
1194 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1195 if (udev_ctrl == NULL) {
1196 log_error("error taking over udev control socket");
1201 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1202 if (monitor == NULL) {
1203 log_error("error taking over netlink socket");
1208 /* get our own cgroup, we regularly kill everything udev has left behind */
1209 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1212 /* open control and netlink socket */
1213 udev_ctrl = udev_ctrl_new(udev);
1214 if (udev_ctrl == NULL) {
1215 log_error("error initializing udev control socket");
1219 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1221 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1222 if (monitor == NULL) {
1223 log_error("error initializing netlink socket");
1227 fd_netlink = udev_monitor_get_fd(monitor);
1229 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1232 if (udev_monitor_enable_receiving(monitor) < 0) {
1233 log_error("error binding netlink socket");
1238 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1239 log_error("error binding udev control socket");
1244 log_info("starting version " VERSION);
1246 udev_builtin_init(udev);
1248 rules = udev_rules_new(udev, arg_resolve_names);
1249 if (rules == NULL) {
1250 log_error("error reading rules");
1254 rc = udev_rules_apply_static_dev_perms(rules);
1256 log_error_errno(rc, "failed to apply permissions on static device nodes - %m");
1258 if (arg_daemonize) {
1266 log_error_errno(errno, "fork of daemon failed: %m");
1271 goto exit_daemonize;
1276 write_string_file("/proc/self/oom_score_adj", "-1000");
1278 sd_notify(1, "READY=1");
1281 if (arg_children_max <= 0) {
1284 arg_children_max = 8;
1286 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1287 arg_children_max += CPU_COUNT(&cpu_set) * 2;
1290 log_debug("set children_max to %u", arg_children_max);
1292 udev_list_node_init(&event_list);
1293 udev_list_node_init(&worker_list);
1295 fd_inotify = udev_watch_init(udev);
1296 if (fd_inotify < 0) {
1297 log_error("error initializing inotify");
1301 udev_watch_restore(udev);
1303 /* block and listen to all signals on signalfd */
1305 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1306 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1307 if (fd_signal < 0) {
1308 log_error("error creating signalfd");
1313 /* unnamed socket from workers to the main daemon */
1314 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1315 log_error("error creating socketpair");
1319 fd_worker = worker_watch[READ_END];
1321 ep_ctrl.data.fd = fd_ctrl;
1322 ep_inotify.data.fd = fd_inotify;
1323 ep_signal.data.fd = fd_signal;
1324 ep_netlink.data.fd = fd_netlink;
1325 ep_worker.data.fd = fd_worker;
1327 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1329 log_error_errno(errno, "error creating epoll fd: %m");
1332 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1333 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1334 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1335 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1336 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1337 log_error_errno(errno, "fail to add fds to epoll: %m");
1342 static usec_t last_usec;
1343 struct epoll_event ev[8];
1346 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1350 /* close sources of new events and discard buffered events */
1352 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1355 if (monitor != NULL) {
1356 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1357 udev_monitor_unref(monitor);
1360 if (fd_inotify >= 0) {
1361 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1366 /* discard queued events and kill workers */
1367 event_queue_cleanup(udev, EVENT_QUEUED);
1370 /* exit after all has cleaned up */
1371 if (udev_list_node_is_empty(&event_list) && children == 0)
1374 /* timeout at exit for workers to finish */
1375 timeout = 30 * MSEC_PER_SEC;
1376 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1380 /* cleanup possible left-over processes in our cgroup */
1382 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1384 /* kill idle or hanging workers */
1385 timeout = 3 * MSEC_PER_SEC;
1388 /* tell settle that we are busy or idle */
1389 event_queue_update();
1391 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1396 struct udev_list_node *loop;
1400 log_error("timeout, giving up waiting for workers to finish");
1404 /* kill idle workers */
1405 if (udev_list_node_is_empty(&event_list)) {
1406 log_debug("cleanup idle workers");
1410 /* check for hanging events */
1411 udev_list_node_foreach(loop, &worker_list) {
1412 struct worker *worker = node_to_worker(loop);
1415 if (worker->state != WORKER_RUNNING)
1418 ts = now(CLOCK_MONOTONIC);
1420 if ((ts - worker->event_start_usec) > arg_event_timeout_warn_usec) {
1421 if ((ts - worker->event_start_usec) > arg_event_timeout_usec) {
1422 log_error("worker ["PID_FMT"] %s timeout; kill it", worker->pid, worker->event->devpath);
1423 kill(worker->pid, SIGKILL);
1424 worker->state = WORKER_KILLED;
1426 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1427 worker->event->exitcode = -64;
1428 } else if (!worker->event_warned) {
1429 log_warning("worker ["PID_FMT"] %s is taking a long time", worker->pid, worker->event->devpath);
1430 worker->event_warned = true;
1437 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1438 for (i = 0; i < fdcount; i++) {
1439 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1441 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1443 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1445 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1447 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1451 /* check for changed config, every 3 seconds at most */
1452 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1453 if (udev_rules_check_timestamp(rules))
1455 if (udev_builtin_validate(udev))
1458 last_usec = now(CLOCK_MONOTONIC);
1461 /* reload requested, HUP signal received, rules changed, builtin changed */
1464 rules = udev_rules_unref(rules);
1465 udev_builtin_exit(udev);
1469 /* event has finished */
1471 worker_returned(fd_worker);
1474 struct udev_device *dev;
1476 dev = udev_monitor_receive_device(monitor);
1478 udev_device_ensure_usec_initialized(dev, NULL);
1479 if (event_queue_insert(dev) < 0)
1480 udev_device_unref(dev);
1484 /* start new events */
1485 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1486 udev_builtin_init(udev);
1488 rules = udev_rules_new(udev, arg_resolve_names);
1490 event_queue_start(udev);
1494 struct signalfd_siginfo fdsi;
1497 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1498 if (size == sizeof(struct signalfd_siginfo))
1499 handle_signal(udev, fdsi.ssi_signo);
1502 /* we are shutting down, the events below are not handled anymore */
1506 /* device node watch */
1508 handle_inotify(udev);
1510 /* tell settle that we are busy or idle, this needs to be before the
1513 event_queue_update();
1516 * This needs to be after the inotify handling, to make sure,
1517 * that the ping is send back after the possibly generated
1518 * "change" events by the inotify device node watch.
1520 * A single time we may receive a client connection which we need to
1521 * keep open to block the client. It will be closed right before we
1525 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1530 udev_ctrl_cleanup(udev_ctrl);
1531 unlink("/run/udev/queue");
1535 worker_list_cleanup(udev);
1536 event_queue_cleanup(udev, EVENT_UNDEF);
1537 udev_rules_unref(rules);
1538 udev_builtin_exit(udev);
1541 if (worker_watch[READ_END] >= 0)
1542 close(worker_watch[READ_END]);
1543 if (worker_watch[WRITE_END] >= 0)
1544 close(worker_watch[WRITE_END]);
1545 udev_monitor_unref(monitor);
1546 udev_ctrl_connection_unref(ctrl_conn);
1547 udev_ctrl_unref(udev_ctrl);
1548 udev_list_cleanup(&properties_list);
1549 mac_selinux_finish();