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 "rtnl-util.h"
52 #include "sd-daemon.h"
53 #include "cgroup-util.h"
54 #include "dev-setup.h"
57 void udev_main_log(struct udev *udev, int priority,
58 const char *file, int line, const char *fn,
59 const char *format, va_list args) {
60 log_metav(priority, file, line, fn, format, args);
63 static struct udev_rules *rules;
64 static struct udev_ctrl *udev_ctrl;
65 static struct udev_monitor *monitor;
66 static int worker_watch[2] = { -1, -1 };
67 static int fd_signal = -1;
68 static int fd_ep = -1;
69 static int fd_inotify = -1;
70 static bool stop_exec_queue;
73 static bool arg_debug = false;
74 static int arg_daemonize = false;
75 static int arg_resolve_names = 1;
76 static int arg_children_max;
77 static int arg_exec_delay;
78 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
79 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
80 static sigset_t sigmask_orig;
81 static UDEV_LIST(event_list);
82 static UDEV_LIST(worker_list);
83 static char *udev_cgroup;
84 static bool udev_exit;
93 struct udev_list_node node;
95 struct udev_device *dev;
96 enum event_state state;
98 unsigned long long int delaying_seqnum;
99 unsigned long long int seqnum;
102 const char *devpath_old;
108 static inline struct event *node_to_event(struct udev_list_node *node) {
109 return container_of(node, struct event, node);
112 static void event_queue_cleanup(struct udev *udev, enum event_state type);
122 struct udev_list_node node;
126 struct udev_monitor *monitor;
127 enum worker_state state;
129 usec_t event_start_usec;
133 /* passed from worker to main process */
134 struct worker_message {
139 static inline struct worker *node_to_worker(struct udev_list_node *node) {
140 return container_of(node, struct worker, node);
143 static void event_queue_delete(struct event *event) {
144 udev_list_node_remove(&event->node);
145 udev_device_unref(event->dev);
149 static struct worker *worker_ref(struct worker *worker) {
154 static void worker_cleanup(struct worker *worker) {
155 udev_list_node_remove(&worker->node);
156 udev_monitor_unref(worker->monitor);
161 static void worker_unref(struct worker *worker) {
163 if (worker->refcount > 0)
165 log_debug("worker [%u] cleaned up", worker->pid);
166 worker_cleanup(worker);
169 static void worker_list_cleanup(struct udev *udev) {
170 struct udev_list_node *loop, *tmp;
172 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
173 struct worker *worker = node_to_worker(loop);
175 worker_cleanup(worker);
179 static void worker_new(struct event *event) {
180 struct udev *udev = event->udev;
181 struct worker *worker;
182 struct udev_monitor *worker_monitor;
185 /* listen for new events */
186 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
187 if (worker_monitor == NULL)
189 /* allow the main daemon netlink address to send devices to the worker */
190 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
191 udev_monitor_enable_receiving(worker_monitor);
193 worker = new0(struct worker, 1);
194 if (worker == NULL) {
195 udev_monitor_unref(worker_monitor);
198 /* worker + event reference */
199 worker->refcount = 2;
205 struct udev_device *dev = NULL;
207 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
208 struct epoll_event ep_signal, ep_monitor;
210 int rc = EXIT_SUCCESS;
212 /* take initial device from queue */
217 worker_list_cleanup(udev);
218 event_queue_cleanup(udev, EVENT_UNDEF);
219 udev_monitor_unref(monitor);
220 udev_ctrl_unref(udev_ctrl);
223 close(worker_watch[READ_END]);
226 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
228 log_error("error creating signalfd %m");
233 fd_ep = epoll_create1(EPOLL_CLOEXEC);
235 log_error("error creating epoll fd: %m");
240 memzero(&ep_signal, sizeof(struct epoll_event));
241 ep_signal.events = EPOLLIN;
242 ep_signal.data.fd = fd_signal;
244 fd_monitor = udev_monitor_get_fd(worker_monitor);
245 memzero(&ep_monitor, sizeof(struct epoll_event));
246 ep_monitor.events = EPOLLIN;
247 ep_monitor.data.fd = fd_monitor;
249 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
250 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
251 log_error("fail to add fds to epoll: %m");
256 /* request TERM signal if parent exits */
257 prctl(PR_SET_PDEATHSIG, SIGTERM);
259 /* reset OOM score, we only protect the main daemon */
260 write_string_file("/proc/self/oom_score_adj", "0");
263 struct udev_event *udev_event;
264 struct worker_message msg;
268 log_debug("seq %llu running", udev_device_get_seqnum(dev));
269 udev_event = udev_event_new(dev);
270 if (udev_event == NULL) {
275 /* needed for SIGCHLD/SIGTERM in spawn() */
276 udev_event->fd_signal = fd_signal;
278 if (arg_exec_delay > 0)
279 udev_event->exec_delay = arg_exec_delay;
282 * Take a shared lock on the device node; this establishes
283 * a concept of device "ownership" to serialize device
284 * access. External processes holding an exclusive lock will
285 * cause udev to skip the event handling; in the case udev
286 * acquired the lock, the external process can block until
287 * udev has finished its event handling.
289 if (!streq_ptr(udev_device_get_action(dev), "remove") &&
290 streq_ptr("block", udev_device_get_subsystem(dev)) &&
291 !startswith(udev_device_get_sysname(dev), "dm-") &&
292 !startswith(udev_device_get_sysname(dev), "md")) {
293 struct udev_device *d = dev;
295 if (streq_ptr("partition", udev_device_get_devtype(d)))
296 d = udev_device_get_parent(d);
299 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
300 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
301 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
303 fd_lock = safe_close(fd_lock);
309 /* needed for renaming netifs */
310 udev_event->rtnl = rtnl;
312 /* apply rules, create node, symlinks */
313 udev_event_execute_rules(udev_event, arg_event_timeout_usec, arg_event_timeout_warn_usec, rules, &sigmask_orig);
315 udev_event_execute_run(udev_event, arg_event_timeout_usec, arg_event_timeout_warn_usec, &sigmask_orig);
317 /* in case rtnl was initialized */
318 rtnl = sd_rtnl_ref(udev_event->rtnl);
320 /* apply/restore inotify watch */
321 if (udev_event->inotify_watch) {
322 udev_watch_begin(udev, dev);
323 udev_device_update_db(dev);
328 /* send processed event back to libudev listeners */
329 udev_monitor_send_device(worker_monitor, NULL, dev);
332 /* send udevd the result of the event execution */
333 memzero(&msg, sizeof(struct worker_message));
336 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
338 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
340 udev_device_unref(dev);
343 if (udev_event->sigterm) {
344 udev_event_unref(udev_event);
348 udev_event_unref(udev_event);
350 /* wait for more device messages from main udevd, or term signal */
351 while (dev == NULL) {
352 struct epoll_event ev[4];
356 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
360 log_error("failed to poll: %m");
364 for (i = 0; i < fdcount; i++) {
365 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
366 dev = udev_monitor_receive_device(worker_monitor);
368 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
369 struct signalfd_siginfo fdsi;
372 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
373 if (size != sizeof(struct signalfd_siginfo))
375 switch (fdsi.ssi_signo) {
384 udev_device_unref(dev);
385 safe_close(fd_signal);
388 close(worker_watch[WRITE_END]);
389 udev_rules_unref(rules);
390 udev_builtin_exit(udev);
391 udev_monitor_unref(worker_monitor);
397 udev_monitor_unref(worker_monitor);
398 event->state = EVENT_QUEUED;
400 log_error("fork of child failed: %m");
403 /* close monitor, but keep address around */
404 udev_monitor_disconnect(worker_monitor);
405 worker->monitor = worker_monitor;
407 worker->state = WORKER_RUNNING;
408 worker->event_start_usec = now(CLOCK_MONOTONIC);
409 worker->event_warned = false;
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) {
420 struct udev_list_node *loop;
422 udev_list_node_foreach(loop, &worker_list) {
423 struct worker *worker = node_to_worker(loop);
426 if (worker->state != WORKER_IDLE)
429 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
431 log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
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->seqnum = udev_device_get_seqnum(dev);
465 event->devpath = udev_device_get_devpath(dev);
466 event->devpath_len = strlen(event->devpath);
467 event->devpath_old = udev_device_get_devpath_old(dev);
468 event->devnum = udev_device_get_devnum(dev);
469 event->is_block = streq("block", udev_device_get_subsystem(dev));
470 event->ifindex = udev_device_get_ifindex(dev);
472 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
473 udev_device_get_action(dev), udev_device_get_subsystem(dev));
475 event->state = EVENT_QUEUED;
476 udev_list_node_append(&event->node, &event_list);
480 static void worker_kill(struct udev *udev) {
481 struct udev_list_node *loop;
483 udev_list_node_foreach(loop, &worker_list) {
484 struct worker *worker = node_to_worker(loop);
486 if (worker->state == WORKER_KILLED)
489 worker->state = WORKER_KILLED;
490 kill(worker->pid, SIGTERM);
494 /* lookup event for identical, parent, child device */
495 static bool is_devpath_busy(struct event *event) {
496 struct udev_list_node *loop;
499 /* check if queue contains events we depend on */
500 udev_list_node_foreach(loop, &event_list) {
501 struct event *loop_event = node_to_event(loop);
503 /* we already found a later event, earlier can not block us, no need to check again */
504 if (loop_event->seqnum < event->delaying_seqnum)
507 /* event we checked earlier still exists, no need to check again */
508 if (loop_event->seqnum == event->delaying_seqnum)
511 /* found ourself, no later event can block us */
512 if (loop_event->seqnum >= event->seqnum)
515 /* check major/minor */
516 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
519 /* check network device ifindex */
520 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
523 /* check our old name */
524 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
525 event->delaying_seqnum = loop_event->seqnum;
529 /* compare devpath */
530 common = MIN(loop_event->devpath_len, event->devpath_len);
532 /* one devpath is contained in the other? */
533 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
536 /* identical device event found */
537 if (loop_event->devpath_len == event->devpath_len) {
538 /* devices names might have changed/swapped in the meantime */
539 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
541 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
543 event->delaying_seqnum = loop_event->seqnum;
547 /* parent device event found */
548 if (event->devpath[common] == '/') {
549 event->delaying_seqnum = loop_event->seqnum;
553 /* child device event found */
554 if (loop_event->devpath[common] == '/') {
555 event->delaying_seqnum = loop_event->seqnum;
559 /* no matching device */
566 static void event_queue_start(struct udev *udev) {
567 struct udev_list_node *loop;
569 udev_list_node_foreach(loop, &event_list) {
570 struct event *event = node_to_event(loop);
572 if (event->state != EVENT_QUEUED)
575 /* do not start event if parent or child event is still running */
576 if (is_devpath_busy(event))
583 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
584 struct udev_list_node *loop, *tmp;
586 udev_list_node_foreach_safe(loop, tmp, &event_list) {
587 struct event *event = node_to_event(loop);
589 if (match_type != EVENT_UNDEF && match_type != event->state)
592 event_queue_delete(event);
596 static void worker_returned(int fd_worker) {
598 struct worker_message msg;
600 struct udev_list_node *loop;
602 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
603 if (size != sizeof(struct worker_message))
606 /* lookup worker who sent the signal */
607 udev_list_node_foreach(loop, &worker_list) {
608 struct worker *worker = node_to_worker(loop);
610 if (worker->pid != msg.pid)
613 /* worker returned */
615 worker->event->exitcode = msg.exitcode;
616 event_queue_delete(worker->event);
617 worker->event = NULL;
619 if (worker->state != WORKER_KILLED)
620 worker->state = WORKER_IDLE;
621 worker_unref(worker);
627 /* receive the udevd message from userspace */
628 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
629 struct udev *udev = udev_ctrl_get_udev(uctrl);
630 struct udev_ctrl_connection *ctrl_conn;
631 struct udev_ctrl_msg *ctrl_msg = NULL;
635 ctrl_conn = udev_ctrl_get_connection(uctrl);
636 if (ctrl_conn == NULL)
639 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
640 if (ctrl_msg == NULL)
643 i = udev_ctrl_get_set_log_level(ctrl_msg);
645 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
646 log_set_max_level(i);
647 udev_set_log_priority(udev, i);
651 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
652 log_debug("udevd message (STOP_EXEC_QUEUE) received");
653 stop_exec_queue = true;
656 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
657 log_debug("udevd message (START_EXEC_QUEUE) received");
658 stop_exec_queue = false;
661 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
662 log_debug("udevd message (RELOAD) received");
666 str = udev_ctrl_get_set_env(ctrl_msg);
674 val = strchr(key, '=');
678 if (val[0] == '\0') {
679 log_debug("udevd message (ENV) received, unset '%s'", key);
680 udev_add_property(udev, key, NULL);
682 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
683 udev_add_property(udev, key, val);
686 log_error("wrong key format '%s'", key);
693 i = udev_ctrl_get_set_children_max(ctrl_msg);
695 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
696 arg_children_max = i;
699 if (udev_ctrl_get_ping(ctrl_msg) > 0)
700 log_debug("udevd message (SYNC) received");
702 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
703 log_debug("udevd message (EXIT) received");
705 /* keep reference to block the client until we exit */
706 udev_ctrl_connection_ref(ctrl_conn);
709 udev_ctrl_msg_unref(ctrl_msg);
710 return udev_ctrl_connection_unref(ctrl_conn);
713 static int synthesize_change(struct udev_device *dev) {
714 char filename[UTIL_PATH_SIZE];
717 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
718 streq_ptr("disk", udev_device_get_devtype(dev)) &&
719 !startswith(udev_device_get_sysname(dev), "dm-")) {
720 bool part_table_read = false;
721 bool has_partitions = false;
723 struct udev *udev = udev_device_get_udev(dev);
724 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
725 struct udev_list_entry *item;
728 * Try to re-read the partition table. This only succeeds if
729 * none of the devices is busy. The kernel returns 0 if no
730 * partition table is found, and we will not get an event for
733 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
735 r = flock(fd, LOCK_EX|LOCK_NB);
737 r = ioctl(fd, BLKRRPART, 0);
741 part_table_read = true;
744 /* search for partitions */
745 e = udev_enumerate_new(udev);
749 r = udev_enumerate_add_match_parent(e, dev);
753 r = udev_enumerate_add_match_subsystem(e, "block");
757 r = udev_enumerate_scan_devices(e);
761 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
762 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
764 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
768 if (!streq_ptr("partition", udev_device_get_devtype(d)))
771 has_partitions = true;
776 * We have partitions and re-read the table, the kernel already sent
777 * out a "change" event for the disk, and "remove/add" for all
780 if (part_table_read && has_partitions)
784 * We have partitions but re-reading the partition table did not
785 * work, synthesize "change" for the disk and all partitions.
787 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
788 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
789 write_string_file(filename, "change");
791 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
792 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
794 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
798 if (!streq_ptr("partition", udev_device_get_devtype(d)))
801 log_debug("device %s closed, synthesising partition '%s' 'change'",
802 udev_device_get_devnode(dev), udev_device_get_devnode(d));
803 strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
804 write_string_file(filename, "change");
810 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
811 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
812 write_string_file(filename, "change");
817 static int handle_inotify(struct udev *udev) {
820 struct inotify_event *ev;
823 r = ioctl(fd_inotify, FIONREAD, &nbytes);
824 if (r < 0 || nbytes <= 0)
827 buf = malloc(nbytes);
829 log_error("error getting buffer for inotify");
833 nbytes = read(fd_inotify, buf, nbytes);
835 for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
836 struct udev_device *dev;
838 ev = (struct inotify_event *)(buf + pos);
839 dev = udev_watch_lookup(udev, ev->wd);
843 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
844 if (ev->mask & IN_CLOSE_WRITE)
845 synthesize_change(dev);
846 else if (ev->mask & IN_IGNORED)
847 udev_watch_end(udev, dev);
849 udev_device_unref(dev);
856 static void handle_signal(struct udev *udev, int signo) {
866 struct udev_list_node *loop, *tmp;
868 pid = waitpid(-1, &status, WNOHANG);
872 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
873 struct worker *worker = node_to_worker(loop);
875 if (worker->pid != pid)
877 log_debug("worker [%u] exit", pid);
879 if (WIFEXITED(status)) {
880 if (WEXITSTATUS(status) != 0)
881 log_error("worker [%u] exit with return code %i",
882 pid, WEXITSTATUS(status));
883 } else if (WIFSIGNALED(status)) {
884 log_error("worker [%u] terminated by signal %i (%s)",
885 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
886 } else if (WIFSTOPPED(status)) {
887 log_error("worker [%u] stopped", pid);
888 } else if (WIFCONTINUED(status)) {
889 log_error("worker [%u] continued", pid);
891 log_error("worker [%u] exit with status 0x%04x", pid, status);
894 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
896 log_error("worker [%u] failed while handling '%s'",
897 pid, worker->event->devpath);
898 worker->event->exitcode = -32;
899 event_queue_delete(worker->event);
901 /* drop reference taken for state 'running' */
902 worker_unref(worker);
905 worker_unref(worker);
916 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
917 int ctrl = -1, netlink = -1;
920 n = sd_listen_fds(true);
924 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
925 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
932 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
942 if (ctrl < 0 || netlink < 0)
945 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
952 * read the kernel commandline, in case we need to get into debug mode
953 * udev.log-priority=<level> syslog priority
954 * udev.children-max=<number of workers> events are fully serialized if set to 1
955 * udev.exec-delay=<number of seconds> delay execution of every executed program
957 static void kernel_cmdline_options(struct udev *udev) {
958 _cleanup_free_ char *line = NULL;
959 const char *word, *state;
963 r = proc_cmdline(&line);
965 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
969 FOREACH_WORD_QUOTED(word, l, line, state) {
972 s = strndup(word, l);
976 /* accept the same options for the initrd, prefixed with "rd." */
977 if (in_initrd() && startswith(s, "rd."))
982 if (startswith(opt, "udev.log-priority=")) {
985 prio = util_log_priority(opt + 18);
986 log_set_max_level(prio);
987 udev_set_log_priority(udev, prio);
988 } else if (startswith(opt, "udev.children-max=")) {
989 r = safe_atoi(opt + 18, &arg_children_max);
991 log_warning("Invalid udev.children-max ignored: %s", opt + 18);
992 } else if (startswith(opt, "udev.exec-delay=")) {
993 r = safe_atoi(opt + 16, &arg_exec_delay);
995 log_warning("Invalid udev.exec-delay ignored: %s", opt + 16);
996 } else if (startswith(opt, "udev.event-timeout=")) {
997 r = safe_atou64(opt + 16, &arg_event_timeout_usec);
999 log_warning("Invalid udev.event-timeout ignored: %s", opt + 16);
1002 arg_event_timeout_usec *= USEC_PER_SEC;
1003 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1010 static void help(void) {
1011 printf("%s [OPTIONS...]\n\n"
1012 "Manages devices.\n\n"
1015 " --children-max=<maximum number of workers>\n"
1016 " --exec-delay=<seconds to wait before executing RUN=>\n"
1017 " --event-timeout=<seconds to wait before terminating an event>\n"
1018 " --resolve-names=early|late|never\n"
1021 , program_invocation_short_name);
1024 static int parse_argv(int argc, char *argv[]) {
1025 static const struct option options[] = {
1026 { "daemon", no_argument, NULL, 'd' },
1027 { "debug", no_argument, NULL, 'D' },
1028 { "children-max", required_argument, NULL, 'c' },
1029 { "exec-delay", required_argument, NULL, 'e' },
1030 { "event-timeout", required_argument, NULL, 't' },
1031 { "resolve-names", required_argument, NULL, 'N' },
1032 { "help", no_argument, NULL, 'h' },
1033 { "version", no_argument, NULL, 'V' },
1042 while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1048 arg_daemonize = true;
1051 r = safe_atoi(optarg, &arg_children_max);
1053 log_warning("Invalid --children-max ignored: %s", optarg);
1056 r = safe_atoi(optarg, &arg_exec_delay);
1058 log_warning("Invalid --exec-delay ignored: %s", optarg);
1061 r = safe_atou64(optarg, &arg_event_timeout_usec);
1063 log_warning("Invalid --event-timeout ignored: %s", optarg);
1065 arg_event_timeout_usec *= USEC_PER_SEC;
1066 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1073 if (streq(optarg, "early")) {
1074 arg_resolve_names = 1;
1075 } else if (streq(optarg, "late")) {
1076 arg_resolve_names = 0;
1077 } else if (streq(optarg, "never")) {
1078 arg_resolve_names = -1;
1080 log_error("resolve-names must be early, late or never");
1088 printf("%s\n", VERSION);
1093 assert_not_reached("Unhandled option");
1101 int main(int argc, char *argv[]) {
1105 int fd_netlink = -1;
1107 struct epoll_event ep_ctrl = { .events = EPOLLIN };
1108 struct epoll_event ep_inotify = { .events = EPOLLIN };
1109 struct epoll_event ep_signal = { .events = EPOLLIN };
1110 struct epoll_event ep_netlink = { .events = EPOLLIN };
1111 struct epoll_event ep_worker = { .events = EPOLLIN };
1112 struct udev_ctrl_connection *ctrl_conn = NULL;
1119 log_set_target(LOG_TARGET_AUTO);
1120 log_parse_environment();
1123 udev_set_log_fn(udev, udev_main_log);
1124 log_set_max_level(udev_get_log_priority(udev));
1126 r = parse_argv(argc, argv);
1130 kernel_cmdline_options(udev);
1133 log_set_max_level(LOG_DEBUG);
1134 udev_set_log_priority(udev, LOG_DEBUG);
1137 if (getuid() != 0) {
1138 log_error("root privileges required");
1142 r = mac_selinux_init("/dev");
1144 log_error("could not initialize labelling: %s", strerror(-r));
1148 /* set umask before creating any file/directory */
1151 log_error("could not change dir to /: %m");
1157 r = mkdir("/run/udev", 0755);
1158 if (r < 0 && errno != EEXIST) {
1159 log_error("could not create /run/udev: %m");
1165 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1166 if (arg_daemonize) {
1169 fd = open("/dev/null", O_RDWR);
1171 if (write(STDOUT_FILENO, 0, 0) < 0)
1172 dup2(fd, STDOUT_FILENO);
1173 if (write(STDERR_FILENO, 0, 0) < 0)
1174 dup2(fd, STDERR_FILENO);
1175 if (fd > STDERR_FILENO)
1178 log_error("cannot open /dev/null");
1182 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1183 /* get control and netlink socket from systemd */
1184 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1185 if (udev_ctrl == NULL) {
1186 log_error("error taking over udev control socket");
1191 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1192 if (monitor == NULL) {
1193 log_error("error taking over netlink socket");
1198 /* get our own cgroup, we regularly kill everything udev has left behind */
1199 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1202 /* open control and netlink socket */
1203 udev_ctrl = udev_ctrl_new(udev);
1204 if (udev_ctrl == NULL) {
1205 log_error("error initializing udev control socket");
1209 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1211 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1212 if (monitor == NULL) {
1213 log_error("error initializing netlink socket");
1217 fd_netlink = udev_monitor_get_fd(monitor);
1220 if (udev_monitor_enable_receiving(monitor) < 0) {
1221 log_error("error binding netlink socket");
1226 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1227 log_error("error binding udev control socket");
1232 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1234 log_info("starting version " VERSION "\n");
1236 udev_builtin_init(udev);
1238 rules = udev_rules_new(udev, arg_resolve_names);
1239 if (rules == NULL) {
1240 log_error("error reading rules");
1244 rc = udev_rules_apply_static_dev_perms(rules);
1246 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1248 if (arg_daemonize) {
1256 log_error("fork of daemon failed: %m");
1261 goto exit_daemonize;
1266 write_string_file("/proc/self/oom_score_adj", "-1000");
1268 sd_notify(1, "READY=1");
1271 if (arg_children_max <= 0) {
1274 arg_children_max = 8;
1276 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1277 arg_children_max += CPU_COUNT(&cpu_set) * 2;
1280 log_debug("set children_max to %u", arg_children_max);
1282 udev_list_node_init(&event_list);
1283 udev_list_node_init(&worker_list);
1285 fd_inotify = udev_watch_init(udev);
1286 if (fd_inotify < 0) {
1287 log_error("error initializing inotify");
1291 udev_watch_restore(udev);
1293 /* block and listen to all signals on signalfd */
1295 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1296 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1297 if (fd_signal < 0) {
1298 log_error("error creating signalfd");
1303 /* unnamed socket from workers to the main daemon */
1304 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1305 log_error("error creating socketpair");
1309 fd_worker = worker_watch[READ_END];
1311 ep_ctrl.data.fd = fd_ctrl;
1312 ep_inotify.data.fd = fd_inotify;
1313 ep_signal.data.fd = fd_signal;
1314 ep_netlink.data.fd = fd_netlink;
1315 ep_worker.data.fd = fd_worker;
1317 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1319 log_error("error creating epoll fd: %m");
1322 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1323 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1324 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1325 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1326 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1327 log_error("fail to add fds to epoll: %m");
1332 static usec_t last_usec;
1333 struct epoll_event ev[8];
1336 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1340 /* close sources of new events and discard buffered events */
1342 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1345 if (monitor != NULL) {
1346 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1347 udev_monitor_unref(monitor);
1350 if (fd_inotify >= 0) {
1351 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1356 /* discard queued events and kill workers */
1357 event_queue_cleanup(udev, EVENT_QUEUED);
1360 /* exit after all has cleaned up */
1361 if (udev_list_node_is_empty(&event_list) && children == 0)
1364 /* timeout at exit for workers to finish */
1365 timeout = 30 * MSEC_PER_SEC;
1366 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1370 /* cleanup possible left-over processes in our cgroup */
1372 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1374 /* kill idle or hanging workers */
1375 timeout = 3 * MSEC_PER_SEC;
1378 /* tell settle that we are busy or idle */
1379 if (!udev_list_node_is_empty(&event_list)) {
1382 fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1386 unlink("/run/udev/queue");
1389 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1394 struct udev_list_node *loop;
1398 log_error("timeout, giving up waiting for workers to finish");
1402 /* kill idle workers */
1403 if (udev_list_node_is_empty(&event_list)) {
1404 log_debug("cleanup idle workers");
1408 /* check for hanging events */
1409 udev_list_node_foreach(loop, &worker_list) {
1410 struct worker *worker = node_to_worker(loop);
1413 if (worker->state != WORKER_RUNNING)
1416 ts = now(CLOCK_MONOTONIC);
1418 if ((ts - worker->event_start_usec) > arg_event_timeout_warn_usec) {
1419 if ((ts - worker->event_start_usec) > arg_event_timeout_usec) {
1420 log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
1421 kill(worker->pid, SIGKILL);
1422 worker->state = WORKER_KILLED;
1424 /* drop reference taken for state 'running' */
1425 worker_unref(worker);
1426 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1427 worker->event->exitcode = -64;
1428 event_queue_delete(worker->event);
1429 worker->event = NULL;
1430 } else if (!worker->event_warned) {
1431 log_warning("worker [%u] %s is taking a long time", worker->pid, worker->event->devpath);
1432 worker->event_warned = true;
1439 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1440 for (i = 0; i < fdcount; i++) {
1441 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1443 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1445 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1447 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1449 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1453 /* check for changed config, every 3 seconds at most */
1454 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1455 if (udev_rules_check_timestamp(rules))
1457 if (udev_builtin_validate(udev))
1460 last_usec = now(CLOCK_MONOTONIC);
1463 /* reload requested, HUP signal received, rules changed, builtin changed */
1466 rules = udev_rules_unref(rules);
1467 udev_builtin_exit(udev);
1471 /* event has finished */
1473 worker_returned(fd_worker);
1476 struct udev_device *dev;
1478 dev = udev_monitor_receive_device(monitor);
1480 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1481 if (event_queue_insert(dev) < 0)
1482 udev_device_unref(dev);
1486 /* start new events */
1487 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1488 udev_builtin_init(udev);
1490 rules = udev_rules_new(udev, arg_resolve_names);
1492 event_queue_start(udev);
1496 struct signalfd_siginfo fdsi;
1499 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1500 if (size == sizeof(struct signalfd_siginfo))
1501 handle_signal(udev, fdsi.ssi_signo);
1504 /* we are shutting down, the events below are not handled anymore */
1508 /* device node watch */
1510 handle_inotify(udev);
1513 * This needs to be after the inotify handling, to make sure,
1514 * that the ping is send back after the possibly generated
1515 * "change" events by the inotify device node watch.
1517 * A single time we may receive a client connection which we need to
1518 * keep open to block the client. It will be closed right before we
1522 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1527 udev_ctrl_cleanup(udev_ctrl);
1528 unlink("/run/udev/queue");
1532 worker_list_cleanup(udev);
1533 event_queue_cleanup(udev, EVENT_UNDEF);
1534 udev_rules_unref(rules);
1535 udev_builtin_exit(udev);
1538 if (worker_watch[READ_END] >= 0)
1539 close(worker_watch[READ_END]);
1540 if (worker_watch[WRITE_END] >= 0)
1541 close(worker_watch[WRITE_END]);
1542 udev_monitor_unref(monitor);
1543 udev_ctrl_connection_unref(ctrl_conn);
1544 udev_ctrl_unref(udev_ctrl);
1545 mac_selinux_finish();