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>
44 #include <sys/ioctl.h>
45 #include <sys/inotify.h>
46 #include <sys/utsname.h>
49 #include "sd-daemon.h"
50 #include "cgroup-util.h"
51 #include "dev-setup.h"
56 void udev_main_log(struct udev *udev, int priority,
57 const char *file, int line, const char *fn,
58 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 int children_max;
74 static int exec_delay;
75 static sigset_t sigmask_orig;
76 static UDEV_LIST(event_list);
77 static UDEV_LIST(worker_list);
78 static char *udev_cgroup;
79 static bool udev_exit;
88 struct udev_list_node node;
90 struct udev_device *dev;
91 enum event_state state;
93 unsigned long long int delaying_seqnum;
94 unsigned long long int seqnum;
97 const char *devpath_old;
106 static inline struct event *node_to_event(struct udev_list_node *node)
108 return container_of(node, struct event, node);
111 static void event_queue_cleanup(struct udev *udev, enum event_state type);
121 struct udev_list_node node;
125 struct udev_monitor *monitor;
126 enum worker_state state;
128 usec_t event_start_usec;
131 /* passed from worker to main process */
132 struct worker_message {
137 static inline struct worker *node_to_worker(struct udev_list_node *node)
139 return container_of(node, struct worker, node);
142 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)
155 static void worker_cleanup(struct worker *worker)
157 udev_list_node_remove(&worker->node);
158 udev_monitor_unref(worker->monitor);
163 static void worker_unref(struct worker *worker)
166 if (worker->refcount > 0)
168 log_debug("worker [%u] cleaned up", worker->pid);
169 worker_cleanup(worker);
172 static void worker_list_cleanup(struct udev *udev)
174 struct udev_list_node *loop, *tmp;
176 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
177 struct worker *worker = node_to_worker(loop);
179 worker_cleanup(worker);
183 static void worker_new(struct event *event)
185 struct udev *udev = event->udev;
186 struct worker *worker;
187 struct udev_monitor *worker_monitor;
190 /* listen for new events */
191 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
192 if (worker_monitor == NULL)
194 /* allow the main daemon netlink address to send devices to the worker */
195 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
196 udev_monitor_enable_receiving(worker_monitor);
198 worker = new0(struct worker, 1);
199 if (worker == NULL) {
200 udev_monitor_unref(worker_monitor);
203 /* worker + event reference */
204 worker->refcount = 2;
210 struct udev_device *dev = NULL;
212 struct epoll_event ep_signal, ep_monitor;
214 int rc = EXIT_SUCCESS;
216 /* take initial device from queue */
221 worker_list_cleanup(udev);
222 event_queue_cleanup(udev, EVENT_UNDEF);
223 udev_monitor_unref(monitor);
224 udev_ctrl_unref(udev_ctrl);
227 close(worker_watch[READ_END]);
230 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
232 log_error("error creating signalfd %m");
237 fd_ep = epoll_create1(EPOLL_CLOEXEC);
239 log_error("error creating epoll fd: %m");
244 memzero(&ep_signal, sizeof(struct epoll_event));
245 ep_signal.events = EPOLLIN;
246 ep_signal.data.fd = fd_signal;
248 fd_monitor = udev_monitor_get_fd(worker_monitor);
249 memzero(&ep_monitor, sizeof(struct epoll_event));
250 ep_monitor.events = EPOLLIN;
251 ep_monitor.data.fd = fd_monitor;
253 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
254 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
255 log_error("fail to add fds to epoll: %m");
260 /* request TERM signal if parent exits */
261 prctl(PR_SET_PDEATHSIG, SIGTERM);
263 /* reset OOM score, we only protect the main daemon */
264 write_string_file("/proc/self/oom_score_adj", "0");
267 struct udev_event *udev_event;
268 struct worker_message msg;
272 log_debug("seq %llu running", udev_device_get_seqnum(dev));
273 udev_event = udev_event_new(dev);
274 if (udev_event == NULL) {
279 /* needed for SIGCHLD/SIGTERM in spawn() */
280 udev_event->fd_signal = fd_signal;
283 udev_event->exec_delay = exec_delay;
286 * Take a "read lock" on the device node; this establishes
287 * a concept of device "ownership" to serialize device
288 * access. External processes holding a "write lock" will
289 * cause udev to skip the event handling; in the case udev
290 * acquired the lock, the external process will block until
291 * udev has finished its event handling.
295 * <kabi_> since we make check - device seems unused - we try
296 * ioctl to deactivate - and device is found to be opened
297 * <kay> sure, you try to take a write lock
298 * <kay> if you get it udev is out
299 * <kay> if you can't get it, udev is busy
300 * <kabi_> we cannot deactivate openned device (as it is in-use)
301 * <kay> maybe we should just exclude dm from that thing entirely
302 * <kabi_> IMHO this sounds like a good plan for this moment
304 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
305 !startswith("dm-", udev_device_get_sysname(dev))) {
306 struct udev_device *d = dev;
308 if (streq_ptr("partition", udev_device_get_devtype(d)))
309 d = udev_device_get_parent(d);
312 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
313 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
314 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
316 fd_lock = safe_close(fd_lock);
322 /* apply rules, create node, symlinks */
323 udev_event_execute_rules(udev_event, rules, &sigmask_orig);
325 udev_event_execute_run(udev_event, &sigmask_orig);
327 /* apply/restore inotify watch */
328 if (udev_event->inotify_watch) {
329 udev_watch_begin(udev, dev);
330 udev_device_update_db(dev);
335 /* send processed event back to libudev listeners */
336 udev_monitor_send_device(worker_monitor, NULL, dev);
339 /* send udevd the result of the event execution */
340 memzero(&msg, sizeof(struct worker_message));
343 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
345 log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
347 udev_device_unref(dev);
350 if (udev_event->sigterm) {
351 udev_event_unref(udev_event);
355 udev_event_unref(udev_event);
357 /* wait for more device messages from main udevd, or term signal */
358 while (dev == NULL) {
359 struct epoll_event ev[4];
363 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
367 log_error("failed to poll: %m");
371 for (i = 0; i < fdcount; i++) {
372 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
373 dev = udev_monitor_receive_device(worker_monitor);
375 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
376 struct signalfd_siginfo fdsi;
379 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
380 if (size != sizeof(struct signalfd_siginfo))
382 switch (fdsi.ssi_signo) {
391 udev_device_unref(dev);
392 safe_close(fd_signal);
395 close(worker_watch[WRITE_END]);
396 udev_rules_unref(rules);
397 udev_builtin_exit(udev);
398 udev_monitor_unref(worker_monitor);
404 udev_monitor_unref(worker_monitor);
405 event->state = EVENT_QUEUED;
407 log_error("fork of child failed: %m");
410 /* close monitor, but keep address around */
411 udev_monitor_disconnect(worker_monitor);
412 worker->monitor = worker_monitor;
414 worker->state = WORKER_RUNNING;
415 worker->event_start_usec = now(CLOCK_MONOTONIC);
416 worker->event = event;
417 event->state = EVENT_RUNNING;
418 udev_list_node_append(&worker->node, &worker_list);
420 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
425 static void event_run(struct event *event)
427 struct udev_list_node *loop;
429 udev_list_node_foreach(loop, &worker_list) {
430 struct worker *worker = node_to_worker(loop);
433 if (worker->state != WORKER_IDLE)
436 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
438 log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
439 kill(worker->pid, SIGKILL);
440 worker->state = WORKER_KILLED;
444 worker->event = event;
445 worker->state = WORKER_RUNNING;
446 worker->event_start_usec = now(CLOCK_MONOTONIC);
447 event->state = EVENT_RUNNING;
451 if (children >= children_max) {
452 if (children_max > 1)
453 log_debug("maximum number (%i) of children reached", children);
457 /* start new worker and pass initial device */
461 static int event_queue_insert(struct udev_device *dev)
465 event = new0(struct event, 1);
469 event->udev = udev_device_get_udev(dev);
471 event->seqnum = udev_device_get_seqnum(dev);
472 event->devpath = udev_device_get_devpath(dev);
473 event->devpath_len = strlen(event->devpath);
474 event->devpath_old = udev_device_get_devpath_old(dev);
475 event->devnum = udev_device_get_devnum(dev);
476 event->is_block = streq("block", udev_device_get_subsystem(dev));
477 event->ifindex = udev_device_get_ifindex(dev);
479 if (streq(udev_device_get_subsystem(dev), "firmware"))
480 event->nodelay = true;
483 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
484 udev_device_get_action(dev), udev_device_get_subsystem(dev));
486 event->state = EVENT_QUEUED;
487 udev_list_node_append(&event->node, &event_list);
491 static void worker_kill(struct udev *udev)
493 struct udev_list_node *loop;
495 udev_list_node_foreach(loop, &worker_list) {
496 struct worker *worker = node_to_worker(loop);
498 if (worker->state == WORKER_KILLED)
501 worker->state = WORKER_KILLED;
502 kill(worker->pid, SIGTERM);
506 /* lookup event for identical, parent, child device */
507 static bool is_devpath_busy(struct event *event)
509 struct udev_list_node *loop;
512 /* check if queue contains events we depend on */
513 udev_list_node_foreach(loop, &event_list) {
514 struct event *loop_event = node_to_event(loop);
516 /* we already found a later event, earlier can not block us, no need to check again */
517 if (loop_event->seqnum < event->delaying_seqnum)
520 /* event we checked earlier still exists, no need to check again */
521 if (loop_event->seqnum == event->delaying_seqnum)
524 /* found ourself, no later event can block us */
525 if (loop_event->seqnum >= event->seqnum)
528 /* check major/minor */
529 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
532 /* check network device ifindex */
533 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
536 /* check our old name */
537 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
538 event->delaying_seqnum = loop_event->seqnum;
542 /* compare devpath */
543 common = MIN(loop_event->devpath_len, event->devpath_len);
545 /* one devpath is contained in the other? */
546 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
549 /* identical device event found */
550 if (loop_event->devpath_len == event->devpath_len) {
551 /* devices names might have changed/swapped in the meantime */
552 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
554 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
556 event->delaying_seqnum = loop_event->seqnum;
561 /* allow to bypass the dependency tracking */
566 /* parent device event found */
567 if (event->devpath[common] == '/') {
568 event->delaying_seqnum = loop_event->seqnum;
572 /* child device event found */
573 if (loop_event->devpath[common] == '/') {
574 event->delaying_seqnum = loop_event->seqnum;
578 /* no matching device */
585 static void event_queue_start(struct udev *udev)
587 struct udev_list_node *loop;
589 udev_list_node_foreach(loop, &event_list) {
590 struct event *event = node_to_event(loop);
592 if (event->state != EVENT_QUEUED)
595 /* do not start event if parent or child event is still running */
596 if (is_devpath_busy(event))
603 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
605 struct udev_list_node *loop, *tmp;
607 udev_list_node_foreach_safe(loop, tmp, &event_list) {
608 struct event *event = node_to_event(loop);
610 if (match_type != EVENT_UNDEF && match_type != event->state)
613 event_queue_delete(event);
617 static void worker_returned(int fd_worker)
620 struct worker_message msg;
622 struct udev_list_node *loop;
624 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
625 if (size != sizeof(struct worker_message))
628 /* lookup worker who sent the signal */
629 udev_list_node_foreach(loop, &worker_list) {
630 struct worker *worker = node_to_worker(loop);
632 if (worker->pid != msg.pid)
635 /* worker returned */
637 worker->event->exitcode = msg.exitcode;
638 event_queue_delete(worker->event);
639 worker->event = NULL;
641 if (worker->state != WORKER_KILLED)
642 worker->state = WORKER_IDLE;
643 worker_unref(worker);
649 /* receive the udevd message from userspace */
650 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
652 struct udev *udev = udev_ctrl_get_udev(uctrl);
653 struct udev_ctrl_connection *ctrl_conn;
654 struct udev_ctrl_msg *ctrl_msg = NULL;
658 ctrl_conn = udev_ctrl_get_connection(uctrl);
659 if (ctrl_conn == NULL)
662 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
663 if (ctrl_msg == NULL)
666 i = udev_ctrl_get_set_log_level(ctrl_msg);
668 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
669 log_set_max_level(i);
670 udev_set_log_priority(udev, i);
674 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
675 log_debug("udevd message (STOP_EXEC_QUEUE) received");
676 stop_exec_queue = true;
679 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
680 log_debug("udevd message (START_EXEC_QUEUE) received");
681 stop_exec_queue = false;
684 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
685 log_debug("udevd message (RELOAD) received");
689 str = udev_ctrl_get_set_env(ctrl_msg);
697 val = strchr(key, '=');
701 if (val[0] == '\0') {
702 log_debug("udevd message (ENV) received, unset '%s'", key);
703 udev_add_property(udev, key, NULL);
705 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
706 udev_add_property(udev, key, val);
709 log_error("wrong key format '%s'", key);
716 i = udev_ctrl_get_set_children_max(ctrl_msg);
718 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
722 if (udev_ctrl_get_ping(ctrl_msg) > 0)
723 log_debug("udevd message (SYNC) received");
725 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
726 log_debug("udevd message (EXIT) received");
728 /* keep reference to block the client until we exit */
729 udev_ctrl_connection_ref(ctrl_conn);
732 udev_ctrl_msg_unref(ctrl_msg);
733 return udev_ctrl_connection_unref(ctrl_conn);
736 static void synthesize_change(struct udev_device *dev) {
737 char filename[UTIL_PATH_SIZE];
739 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
740 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
741 write_string_file(filename, "change");
744 /* read inotify messages */
745 static int handle_inotify(struct udev *udev)
749 struct inotify_event *ev;
752 r = ioctl(fd_inotify, FIONREAD, &nbytes);
753 if (r < 0 || nbytes <= 0)
756 buf = malloc(nbytes);
758 log_error("error getting buffer for inotify");
762 nbytes = read(fd_inotify, buf, nbytes);
764 for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
765 struct udev_device *dev;
767 ev = (struct inotify_event *)(buf + pos);
768 dev = udev_watch_lookup(udev, ev->wd);
772 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
773 if (ev->mask & IN_CLOSE_WRITE)
774 synthesize_change(dev);
775 else if (ev->mask & IN_IGNORED)
776 udev_watch_end(udev, dev);
778 udev_device_unref(dev);
785 static void handle_signal(struct udev *udev, int signo)
796 struct udev_list_node *loop, *tmp;
798 pid = waitpid(-1, &status, WNOHANG);
802 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
803 struct worker *worker = node_to_worker(loop);
805 if (worker->pid != pid)
807 log_debug("worker [%u] exit", pid);
809 if (WIFEXITED(status)) {
810 if (WEXITSTATUS(status) != 0)
811 log_error("worker [%u] exit with return code %i",
812 pid, WEXITSTATUS(status));
813 } else if (WIFSIGNALED(status)) {
814 log_error("worker [%u] terminated by signal %i (%s)",
815 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
816 } else if (WIFSTOPPED(status)) {
817 log_error("worker [%u] stopped", pid);
818 } else if (WIFCONTINUED(status)) {
819 log_error("worker [%u] continued", pid);
821 log_error("worker [%u] exit with status 0x%04x", pid, status);
824 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
826 log_error("worker [%u] failed while handling '%s'",
827 pid, worker->event->devpath);
828 worker->event->exitcode = -32;
829 event_queue_delete(worker->event);
831 /* drop reference taken for state 'running' */
832 worker_unref(worker);
835 worker_unref(worker);
846 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
848 int ctrl = -1, netlink = -1;
851 n = sd_listen_fds(true);
855 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
856 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
863 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
873 if (ctrl < 0 || netlink < 0)
876 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
883 * read the kernel commandline, in case we need to get into debug mode
884 * udev.log-priority=<level> syslog priority
885 * udev.children-max=<number of workers> events are fully serialized if set to 1
886 * udev.exec-delay=<number of seconds> delay execution of every executed program
888 static void kernel_cmdline_options(struct udev *udev)
890 _cleanup_free_ char *line = NULL;
895 r = proc_cmdline(&line);
897 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
901 FOREACH_WORD_QUOTED(w, l, line, state) {
908 /* accept the same options for the initrd, prefixed with "rd." */
909 if (in_initrd() && startswith(s, "rd."))
914 if (startswith(opt, "udev.log-priority=")) {
917 prio = util_log_priority(opt + 18);
918 log_set_max_level(prio);
919 udev_set_log_priority(udev, prio);
920 } else if (startswith(opt, "udev.children-max=")) {
921 children_max = strtoul(opt + 18, NULL, 0);
922 } else if (startswith(opt, "udev.exec-delay=")) {
923 exec_delay = strtoul(opt + 16, NULL, 0);
930 int main(int argc, char *argv[])
934 int daemonize = false;
935 int resolve_names = 1;
936 static const struct option options[] = {
937 { "daemon", no_argument, NULL, 'd' },
938 { "debug", no_argument, NULL, 'D' },
939 { "children-max", required_argument, NULL, 'c' },
940 { "exec-delay", required_argument, NULL, 'e' },
941 { "resolve-names", required_argument, NULL, 'N' },
942 { "help", no_argument, NULL, 'h' },
943 { "version", no_argument, NULL, 'V' },
949 struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
950 struct udev_ctrl_connection *ctrl_conn = NULL;
957 log_set_target(LOG_TARGET_AUTO);
958 log_parse_environment();
961 udev_set_log_fn(udev, udev_main_log);
962 log_set_max_level(udev_get_log_priority(udev));
964 log_debug("version %s", VERSION);
970 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
979 children_max = strtoul(optarg, NULL, 0);
982 exec_delay = strtoul(optarg, NULL, 0);
986 log_set_max_level(LOG_DEBUG);
987 udev_set_log_priority(udev, LOG_DEBUG);
990 if (streq(optarg, "early")) {
992 } else if (streq(optarg, "late")) {
994 } else if (streq(optarg, "never")) {
997 fprintf(stderr, "resolve-names must be early, late or never\n");
998 log_error("resolve-names must be early, late or never");
1003 printf("Usage: udevd OPTIONS\n"
1006 " --children-max=<maximum number of workers>\n"
1007 " --exec-delay=<seconds to wait before executing RUN=>\n"
1008 " --resolve-names=early|late|never\n"
1014 printf("%s\n", VERSION);
1021 kernel_cmdline_options(udev);
1023 if (getuid() != 0) {
1024 fprintf(stderr, "root privileges required\n");
1025 log_error("root privileges required");
1029 /* set umask before creating any file/directory */
1033 mkdir("/run/udev", 0755);
1037 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1041 fd = open("/dev/null", O_RDWR);
1043 if (write(STDOUT_FILENO, 0, 0) < 0)
1044 dup2(fd, STDOUT_FILENO);
1045 if (write(STDERR_FILENO, 0, 0) < 0)
1046 dup2(fd, STDERR_FILENO);
1047 if (fd > STDERR_FILENO)
1050 fprintf(stderr, "cannot open /dev/null\n");
1051 log_error("cannot open /dev/null");
1055 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1056 /* get control and netlink socket from systemd */
1057 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1058 if (udev_ctrl == NULL) {
1059 log_error("error taking over udev control socket");
1064 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1065 if (monitor == NULL) {
1066 log_error("error taking over netlink socket");
1071 /* get our own cgroup, we regularly kill everything udev has left behind */
1072 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1075 /* open control and netlink socket */
1076 udev_ctrl = udev_ctrl_new(udev);
1077 if (udev_ctrl == NULL) {
1078 fprintf(stderr, "error initializing udev control socket");
1079 log_error("error initializing udev control socket");
1083 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1085 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1086 if (monitor == NULL) {
1087 fprintf(stderr, "error initializing netlink socket\n");
1088 log_error("error initializing netlink socket");
1092 fd_netlink = udev_monitor_get_fd(monitor);
1095 if (udev_monitor_enable_receiving(monitor) < 0) {
1096 fprintf(stderr, "error binding netlink socket\n");
1097 log_error("error binding netlink socket");
1102 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1103 fprintf(stderr, "error binding udev control socket\n");
1104 log_error("error binding udev control socket");
1109 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1119 log_error("fork of daemon failed: %m");
1124 goto exit_daemonize;
1129 write_string_file("/proc/self/oom_score_adj", "-1000");
1131 sd_notify(1, "READY=1");
1134 print_kmsg("starting version " VERSION "\n");
1139 fd = open("/dev/null", O_RDWR);
1141 dup2(fd, STDIN_FILENO);
1142 dup2(fd, STDOUT_FILENO);
1143 dup2(fd, STDERR_FILENO);
1148 fd_inotify = udev_watch_init(udev);
1149 if (fd_inotify < 0) {
1150 fprintf(stderr, "error initializing inotify\n");
1151 log_error("error initializing inotify");
1155 udev_watch_restore(udev);
1157 /* block and listen to all signals on signalfd */
1159 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1160 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1161 if (fd_signal < 0) {
1162 fprintf(stderr, "error creating signalfd\n");
1163 log_error("error creating signalfd");
1168 /* unnamed socket from workers to the main daemon */
1169 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1170 fprintf(stderr, "error creating socketpair\n");
1171 log_error("error creating socketpair");
1175 fd_worker = worker_watch[READ_END];
1177 udev_builtin_init(udev);
1179 rules = udev_rules_new(udev, resolve_names);
1180 if (rules == NULL) {
1181 log_error("error reading rules");
1185 memzero(&ep_ctrl, sizeof(struct epoll_event));
1186 ep_ctrl.events = EPOLLIN;
1187 ep_ctrl.data.fd = fd_ctrl;
1189 memzero(&ep_inotify, sizeof(struct epoll_event));
1190 ep_inotify.events = EPOLLIN;
1191 ep_inotify.data.fd = fd_inotify;
1193 memzero(&ep_signal, sizeof(struct epoll_event));
1194 ep_signal.events = EPOLLIN;
1195 ep_signal.data.fd = fd_signal;
1197 memzero(&ep_netlink, sizeof(struct epoll_event));
1198 ep_netlink.events = EPOLLIN;
1199 ep_netlink.data.fd = fd_netlink;
1201 memzero(&ep_worker, sizeof(struct epoll_event));
1202 ep_worker.events = EPOLLIN;
1203 ep_worker.data.fd = fd_worker;
1205 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1207 log_error("error creating epoll fd: %m");
1210 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1211 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1212 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1213 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1214 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1215 log_error("fail to add fds to epoll: %m");
1219 if (children_max <= 0) {
1224 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1225 children_max += CPU_COUNT(&cpu_set) * 2;
1228 log_debug("set children_max to %u", children_max);
1230 rc = udev_rules_apply_static_dev_perms(rules);
1232 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1234 udev_list_node_init(&event_list);
1235 udev_list_node_init(&worker_list);
1238 static usec_t last_usec;
1239 struct epoll_event ev[8];
1242 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1246 /* close sources of new events and discard buffered events */
1248 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1251 if (monitor != NULL) {
1252 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1253 udev_monitor_unref(monitor);
1256 if (fd_inotify >= 0) {
1257 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1262 /* discard queued events and kill workers */
1263 event_queue_cleanup(udev, EVENT_QUEUED);
1266 /* exit after all has cleaned up */
1267 if (udev_list_node_is_empty(&event_list) && children == 0)
1270 /* timeout at exit for workers to finish */
1271 timeout = 30 * MSEC_PER_SEC;
1272 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1276 /* cleanup possible left-over processes in our cgroup */
1278 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1280 /* kill idle or hanging workers */
1281 timeout = 3 * MSEC_PER_SEC;
1284 /* tell settle that we are busy or idle */
1285 if (!udev_list_node_is_empty(&event_list)) {
1288 fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1292 unlink("/run/udev/queue");
1295 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1300 struct udev_list_node *loop;
1304 log_error("timeout, giving up waiting for workers to finish");
1308 /* kill idle workers */
1309 if (udev_list_node_is_empty(&event_list)) {
1310 log_debug("cleanup idle workers");
1314 /* check for hanging events */
1315 udev_list_node_foreach(loop, &worker_list) {
1316 struct worker *worker = node_to_worker(loop);
1318 if (worker->state != WORKER_RUNNING)
1321 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * USEC_PER_SEC) {
1322 log_error("worker [%u] %s timeout; kill it", worker->pid,
1323 worker->event ? worker->event->devpath : "<idle>");
1324 kill(worker->pid, SIGKILL);
1325 worker->state = WORKER_KILLED;
1327 /* drop reference taken for state 'running' */
1328 worker_unref(worker);
1329 if (worker->event) {
1330 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1331 worker->event->exitcode = -64;
1332 event_queue_delete(worker->event);
1333 worker->event = NULL;
1340 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1341 for (i = 0; i < fdcount; i++) {
1342 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1344 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1346 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1348 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1350 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1354 /* check for changed config, every 3 seconds at most */
1355 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1356 if (udev_rules_check_timestamp(rules))
1358 if (udev_builtin_validate(udev))
1361 last_usec = now(CLOCK_MONOTONIC);
1364 /* reload requested, HUP signal received, rules changed, builtin changed */
1367 rules = udev_rules_unref(rules);
1368 udev_builtin_exit(udev);
1372 /* event has finished */
1374 worker_returned(fd_worker);
1377 struct udev_device *dev;
1379 dev = udev_monitor_receive_device(monitor);
1381 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1382 if (event_queue_insert(dev) < 0)
1383 udev_device_unref(dev);
1387 /* start new events */
1388 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1389 udev_builtin_init(udev);
1391 rules = udev_rules_new(udev, resolve_names);
1393 event_queue_start(udev);
1397 struct signalfd_siginfo fdsi;
1400 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1401 if (size == sizeof(struct signalfd_siginfo))
1402 handle_signal(udev, fdsi.ssi_signo);
1405 /* we are shutting down, the events below are not handled anymore */
1409 /* device node watch */
1411 handle_inotify(udev);
1414 * This needs to be after the inotify handling, to make sure,
1415 * that the ping is send back after the possibly generated
1416 * "change" events by the inotify device node watch.
1418 * A single time we may receive a client connection which we need to
1419 * keep open to block the client. It will be closed right before we
1423 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1428 udev_ctrl_cleanup(udev_ctrl);
1429 unlink("/run/udev/queue");
1433 worker_list_cleanup(udev);
1434 event_queue_cleanup(udev, EVENT_UNDEF);
1435 udev_rules_unref(rules);
1436 udev_builtin_exit(udev);
1439 if (worker_watch[READ_END] >= 0)
1440 close(worker_watch[READ_END]);
1441 if (worker_watch[WRITE_END] >= 0)
1442 close(worker_watch[WRITE_END]);
1443 udev_monitor_unref(monitor);
1444 udev_ctrl_connection_unref(ctrl_conn);
1445 udev_ctrl_unref(udev_ctrl);