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/>.
35 #include <sys/prctl.h>
36 #include <sys/socket.h>
38 #include <sys/signalfd.h>
39 #include <sys/epoll.h>
43 #include <sys/ioctl.h>
44 #include <sys/inotify.h>
45 #include <sys/utsname.h>
48 #include "sd-daemon.h"
49 #include "cgroup-util.h"
50 #include "dev-setup.h"
55 void udev_main_log(struct udev *udev, int priority,
56 const char *file, int line, const char *fn,
57 const char *format, va_list args)
59 log_metav(priority, file, line, fn, format, args);
62 static struct udev_rules *rules;
63 static struct udev_queue_export *udev_queue_export;
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);
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;
104 static inline struct event *node_to_event(struct udev_list_node *node)
106 return container_of(node, struct event, node);
109 static void event_queue_cleanup(struct udev *udev, enum event_state type);
119 struct udev_list_node node;
123 struct udev_monitor *monitor;
124 enum worker_state state;
126 usec_t event_start_usec;
129 /* passed from worker to main process */
130 struct worker_message {
135 static inline struct worker *node_to_worker(struct udev_list_node *node)
137 return container_of(node, struct worker, node);
140 static void event_queue_delete(struct event *event, bool export)
142 udev_list_node_remove(&event->node);
145 udev_queue_export_device_finished(udev_queue_export, event->dev);
146 log_debug("seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
148 udev_device_unref(event->dev);
152 static struct worker *worker_ref(struct worker *worker)
158 static void worker_cleanup(struct worker *worker)
160 udev_list_node_remove(&worker->node);
161 udev_monitor_unref(worker->monitor);
166 static void worker_unref(struct worker *worker)
169 if (worker->refcount > 0)
171 log_debug("worker [%u] cleaned up\n", worker->pid);
172 worker_cleanup(worker);
175 static void worker_list_cleanup(struct udev *udev)
177 struct udev_list_node *loop, *tmp;
179 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
180 struct worker *worker = node_to_worker(loop);
182 worker_cleanup(worker);
186 static void worker_new(struct event *event)
188 struct udev *udev = event->udev;
189 struct worker *worker;
190 struct udev_monitor *worker_monitor;
193 /* listen for new events */
194 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
195 if (worker_monitor == NULL)
197 /* allow the main daemon netlink address to send devices to the worker */
198 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
199 udev_monitor_enable_receiving(worker_monitor);
201 worker = calloc(1, sizeof(struct worker));
202 if (worker == NULL) {
203 udev_monitor_unref(worker_monitor);
206 /* worker + event reference */
207 worker->refcount = 2;
213 struct udev_device *dev = NULL;
215 struct epoll_event ep_signal, ep_monitor;
217 int rc = EXIT_SUCCESS;
219 /* take initial device from queue */
224 worker_list_cleanup(udev);
225 event_queue_cleanup(udev, EVENT_UNDEF);
226 udev_queue_export_unref(udev_queue_export);
227 udev_monitor_unref(monitor);
228 udev_ctrl_unref(udev_ctrl);
231 close(worker_watch[READ_END]);
234 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
236 log_error("error creating signalfd %m\n");
241 fd_ep = epoll_create1(EPOLL_CLOEXEC);
243 log_error("error creating epoll fd: %m\n");
248 memset(&ep_signal, 0, sizeof(struct epoll_event));
249 ep_signal.events = EPOLLIN;
250 ep_signal.data.fd = fd_signal;
252 fd_monitor = udev_monitor_get_fd(worker_monitor);
253 memset(&ep_monitor, 0, sizeof(struct epoll_event));
254 ep_monitor.events = EPOLLIN;
255 ep_monitor.data.fd = fd_monitor;
257 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
258 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
259 log_error("fail to add fds to epoll: %m\n");
264 /* request TERM signal if parent exits */
265 prctl(PR_SET_PDEATHSIG, SIGTERM);
267 /* reset OOM score, we only protect the main daemon */
268 write_one_line_file("/proc/self/oom_score_adj", "0");
271 struct udev_event *udev_event;
272 struct worker_message msg;
275 log_debug("seq %llu running\n", udev_device_get_seqnum(dev));
276 udev_event = udev_event_new(dev);
277 if (udev_event == NULL) {
282 /* needed for SIGCHLD/SIGTERM in spawn() */
283 udev_event->fd_signal = fd_signal;
286 udev_event->exec_delay = exec_delay;
288 /* apply rules, create node, symlinks */
289 err = udev_event_execute_rules(udev_event, rules, &sigmask_orig);
292 udev_event_execute_run(udev_event, &sigmask_orig);
294 /* apply/restore inotify watch */
295 if (err == 0 && udev_event->inotify_watch) {
296 udev_watch_begin(udev, dev);
297 udev_device_update_db(dev);
300 /* send processed event back to libudev listeners */
301 udev_monitor_send_device(worker_monitor, NULL, dev);
303 /* send udevd the result of the event execution */
304 memset(&msg, 0, sizeof(struct worker_message));
308 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
310 log_debug("seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
312 udev_device_unref(dev);
315 if (udev_event->sigterm) {
316 udev_event_unref(udev_event);
320 udev_event_unref(udev_event);
322 /* wait for more device messages from main udevd, or term signal */
323 while (dev == NULL) {
324 struct epoll_event ev[4];
328 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
332 log_error("failed to poll: %m\n");
336 for (i = 0; i < fdcount; i++) {
337 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
338 dev = udev_monitor_receive_device(worker_monitor);
340 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
341 struct signalfd_siginfo fdsi;
344 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
345 if (size != sizeof(struct signalfd_siginfo))
347 switch (fdsi.ssi_signo) {
356 udev_device_unref(dev);
362 close(worker_watch[WRITE_END]);
363 udev_rules_unref(rules);
364 udev_builtin_exit(udev);
365 udev_monitor_unref(worker_monitor);
371 udev_monitor_unref(worker_monitor);
372 event->state = EVENT_QUEUED;
374 log_error("fork of child failed: %m\n");
377 /* close monitor, but keep address around */
378 udev_monitor_disconnect(worker_monitor);
379 worker->monitor = worker_monitor;
381 worker->state = WORKER_RUNNING;
382 worker->event_start_usec = now(CLOCK_MONOTONIC);
383 worker->event = event;
384 event->state = EVENT_RUNNING;
385 udev_list_node_append(&worker->node, &worker_list);
387 log_debug("seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
392 static void event_run(struct event *event)
394 struct udev_list_node *loop;
396 udev_list_node_foreach(loop, &worker_list) {
397 struct worker *worker = node_to_worker(loop);
400 if (worker->state != WORKER_IDLE)
403 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
405 log_error("worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count);
406 kill(worker->pid, SIGKILL);
407 worker->state = WORKER_KILLED;
411 worker->event = event;
412 worker->state = WORKER_RUNNING;
413 worker->event_start_usec = now(CLOCK_MONOTONIC);
414 event->state = EVENT_RUNNING;
418 if (children >= children_max) {
419 if (children_max > 1)
420 log_debug("maximum number (%i) of children reached\n", children);
424 /* start new worker and pass initial device */
428 static int event_queue_insert(struct udev_device *dev)
432 event = calloc(1, sizeof(struct event));
436 event->udev = udev_device_get_udev(dev);
438 event->seqnum = udev_device_get_seqnum(dev);
439 event->devpath = udev_device_get_devpath(dev);
440 event->devpath_len = strlen(event->devpath);
441 event->devpath_old = udev_device_get_devpath_old(dev);
442 event->devnum = udev_device_get_devnum(dev);
443 event->is_block = streq("block", udev_device_get_subsystem(dev));
444 event->ifindex = udev_device_get_ifindex(dev);
445 if (streq(udev_device_get_subsystem(dev), "firmware"))
446 event->nodelay = true;
448 udev_queue_export_device_queued(udev_queue_export, dev);
449 log_debug("seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
450 udev_device_get_action(dev), udev_device_get_subsystem(dev));
452 event->state = EVENT_QUEUED;
453 udev_list_node_append(&event->node, &event_list);
457 static void worker_kill(struct udev *udev)
459 struct udev_list_node *loop;
461 udev_list_node_foreach(loop, &worker_list) {
462 struct worker *worker = node_to_worker(loop);
464 if (worker->state == WORKER_KILLED)
467 worker->state = WORKER_KILLED;
468 kill(worker->pid, SIGTERM);
472 /* lookup event for identical, parent, child device */
473 static bool is_devpath_busy(struct event *event)
475 struct udev_list_node *loop;
478 /* check if queue contains events we depend on */
479 udev_list_node_foreach(loop, &event_list) {
480 struct event *loop_event = node_to_event(loop);
482 /* we already found a later event, earlier can not block us, no need to check again */
483 if (loop_event->seqnum < event->delaying_seqnum)
486 /* event we checked earlier still exists, no need to check again */
487 if (loop_event->seqnum == event->delaying_seqnum)
490 /* found ourself, no later event can block us */
491 if (loop_event->seqnum >= event->seqnum)
494 /* check major/minor */
495 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
498 /* check network device ifindex */
499 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
502 /* check our old name */
503 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
504 event->delaying_seqnum = loop_event->seqnum;
508 /* compare devpath */
509 common = MIN(loop_event->devpath_len, event->devpath_len);
511 /* one devpath is contained in the other? */
512 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
515 /* identical device event found */
516 if (loop_event->devpath_len == event->devpath_len) {
517 /* devices names might have changed/swapped in the meantime */
518 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
520 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
522 event->delaying_seqnum = loop_event->seqnum;
526 /* allow to bypass the dependency tracking */
530 /* parent device event found */
531 if (event->devpath[common] == '/') {
532 event->delaying_seqnum = loop_event->seqnum;
536 /* child device event found */
537 if (loop_event->devpath[common] == '/') {
538 event->delaying_seqnum = loop_event->seqnum;
542 /* no matching device */
549 static void event_queue_start(struct udev *udev)
551 struct udev_list_node *loop;
553 udev_list_node_foreach(loop, &event_list) {
554 struct event *event = node_to_event(loop);
556 if (event->state != EVENT_QUEUED)
559 /* do not start event if parent or child event is still running */
560 if (is_devpath_busy(event))
567 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
569 struct udev_list_node *loop, *tmp;
571 udev_list_node_foreach_safe(loop, tmp, &event_list) {
572 struct event *event = node_to_event(loop);
574 if (match_type != EVENT_UNDEF && match_type != event->state)
577 event_queue_delete(event, false);
581 static void worker_returned(int fd_worker)
584 struct worker_message msg;
586 struct udev_list_node *loop;
588 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
589 if (size != sizeof(struct worker_message))
592 /* lookup worker who sent the signal */
593 udev_list_node_foreach(loop, &worker_list) {
594 struct worker *worker = node_to_worker(loop);
596 if (worker->pid != msg.pid)
599 /* worker returned */
601 worker->event->exitcode = msg.exitcode;
602 event_queue_delete(worker->event, true);
603 worker->event = NULL;
605 if (worker->state != WORKER_KILLED)
606 worker->state = WORKER_IDLE;
607 worker_unref(worker);
613 /* receive the udevd message from userspace */
614 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
616 struct udev *udev = udev_ctrl_get_udev(uctrl);
617 struct udev_ctrl_connection *ctrl_conn;
618 struct udev_ctrl_msg *ctrl_msg = NULL;
622 ctrl_conn = udev_ctrl_get_connection(uctrl);
623 if (ctrl_conn == NULL)
626 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
627 if (ctrl_msg == NULL)
630 i = udev_ctrl_get_set_log_level(ctrl_msg);
632 log_debug("udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i);
633 log_set_max_level(i);
634 udev_set_log_priority(udev, i);
638 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
639 log_debug("udevd message (STOP_EXEC_QUEUE) received\n");
640 stop_exec_queue = true;
643 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
644 log_debug("udevd message (START_EXEC_QUEUE) received\n");
645 stop_exec_queue = false;
648 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
649 log_debug("udevd message (RELOAD) received\n");
653 str = udev_ctrl_get_set_env(ctrl_msg);
661 val = strchr(key, '=');
665 if (val[0] == '\0') {
666 log_debug("udevd message (ENV) received, unset '%s'\n", key);
667 udev_add_property(udev, key, NULL);
669 log_debug("udevd message (ENV) received, set '%s=%s'\n", key, val);
670 udev_add_property(udev, key, val);
673 log_error("wrong key format '%s'\n", key);
680 i = udev_ctrl_get_set_children_max(ctrl_msg);
682 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i);
686 if (udev_ctrl_get_ping(ctrl_msg) > 0)
687 log_debug("udevd message (SYNC) received\n");
689 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
690 log_debug("udevd message (EXIT) received\n");
692 /* keep reference to block the client until we exit */
693 udev_ctrl_connection_ref(ctrl_conn);
696 udev_ctrl_msg_unref(ctrl_msg);
697 return udev_ctrl_connection_unref(ctrl_conn);
700 /* read inotify messages */
701 static int handle_inotify(struct udev *udev)
705 struct inotify_event *ev;
707 if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
710 buf = malloc(nbytes);
712 log_error("error getting buffer for inotify\n");
716 nbytes = read(fd_inotify, buf, nbytes);
718 for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
719 struct udev_device *dev;
721 ev = (struct inotify_event *)(buf + pos);
722 dev = udev_watch_lookup(udev, ev->wd);
724 log_debug("inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev));
725 if (ev->mask & IN_CLOSE_WRITE) {
726 char filename[UTIL_PATH_SIZE];
729 log_debug("device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
730 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
731 fd = open(filename, O_WRONLY);
733 if (write(fd, "change", 6) < 0)
734 log_debug("error writing uevent: %m\n");
738 if (ev->mask & IN_IGNORED)
739 udev_watch_end(udev, dev);
741 udev_device_unref(dev);
750 static void handle_signal(struct udev *udev, int signo)
761 struct udev_list_node *loop, *tmp;
763 pid = waitpid(-1, &status, WNOHANG);
767 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
768 struct worker *worker = node_to_worker(loop);
770 if (worker->pid != pid)
772 log_debug("worker [%u] exit\n", pid);
774 if (WIFEXITED(status)) {
775 if (WEXITSTATUS(status) != 0)
776 log_error("worker [%u] exit with return code %i\n", pid, WEXITSTATUS(status));
777 } else if (WIFSIGNALED(status)) {
778 log_error("worker [%u] terminated by signal %i (%s)\n",
779 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
780 } else if (WIFSTOPPED(status)) {
781 log_error("worker [%u] stopped\n", pid);
782 } else if (WIFCONTINUED(status)) {
783 log_error("worker [%u] continued\n", pid);
785 log_error("worker [%u] exit with status 0x%04x\n", pid, status);
788 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
790 log_error("worker [%u] failed while handling '%s'\n",
791 pid, worker->event->devpath);
792 worker->event->exitcode = -32;
793 event_queue_delete(worker->event, true);
794 /* drop reference taken for state 'running' */
795 worker_unref(worker);
798 worker_unref(worker);
809 static void static_dev_create_from_modules(struct udev *udev)
811 struct utsname kernel;
812 char modules[UTIL_PATH_SIZE];
817 strscpyl(modules, sizeof(modules), ROOTPREFIX "/lib/modules/", kernel.release, "/modules.devname", NULL);
818 f = fopen(modules, "re");
822 while (fgets(buf, sizeof(buf), f) != NULL) {
830 char filename[UTIL_PATH_SIZE];
836 s = strchr(modname, ' ');
842 s = strchr(devname, ' ');
848 s = strchr(devno, ' ');
850 s = strchr(devno, '\n');
853 if (sscanf(devno, "%c%u:%u", &type, &maj, &min) != 3)
859 else if (type == 'b')
864 strscpyl(filename, sizeof(filename), "/dev/", devname, NULL);
865 mkdir_parents_label(filename, 0755);
866 label_context_set(filename, mode);
867 log_debug("mknod '%s' %c%u:%u\n", filename, type, maj, min);
868 if (mknod(filename, mode, makedev(maj, min)) < 0 && errno == EEXIST)
869 utimensat(AT_FDCWD, filename, NULL, 0);
870 label_context_clear();
876 static int mem_size_mb(void)
880 long int memsize = -1;
882 f = fopen("/proc/meminfo", "re");
886 while (fgets(buf, sizeof(buf), f) != NULL) {
889 if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) {
890 memsize = value / 1024;
899 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
901 int ctrl = -1, netlink = -1;
904 n = sd_listen_fds(true);
908 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
909 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
916 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
926 if (ctrl < 0 || netlink < 0)
929 log_debug("ctrl=%i netlink=%i\n", ctrl, netlink);
936 * read the kernel commandline, in case we need to get into debug mode
937 * udev.log-priority=<level> syslog priority
938 * udev.children-max=<number of workers> events are fully serialized if set to 1
939 * udev.exec-delay=<number of seconds> delay execution of every executed program
941 static void kernel_cmdline_options(struct udev *udev)
943 char *line, *w, *state;
946 if (read_one_line_file("/proc/cmdline", &line) < 0)
949 FOREACH_WORD_QUOTED(w, l, line, state) {
956 /* accept the same options for the initrd, prefixed with "rd." */
957 if (in_initrd() && startswith(s, "rd."))
962 if (startswith(opt, "udev.log-priority=")) {
965 prio = util_log_priority(opt + 18);
966 log_set_max_level(prio);
967 udev_set_log_priority(udev, prio);
968 } else if (startswith(opt, "udev.children-max=")) {
969 children_max = strtoul(opt + 18, NULL, 0);
970 } else if (startswith(opt, "udev.exec-delay=")) {
971 exec_delay = strtoul(opt + 16, NULL, 0);
980 int main(int argc, char *argv[])
984 int daemonize = false;
985 int resolve_names = 1;
986 static const struct option options[] = {
987 { "daemon", no_argument, NULL, 'd' },
988 { "debug", no_argument, NULL, 'D' },
989 { "children-max", required_argument, NULL, 'c' },
990 { "exec-delay", required_argument, NULL, 'e' },
991 { "resolve-names", required_argument, NULL, 'N' },
992 { "help", no_argument, NULL, 'h' },
993 { "version", no_argument, NULL, 'V' },
999 struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1000 struct udev_ctrl_connection *ctrl_conn = NULL;
1007 log_set_target(LOG_TARGET_AUTO);
1008 log_parse_environment();
1010 udev_set_log_fn(udev, udev_main_log);
1011 log_debug("version %s\n", VERSION);
1017 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
1026 children_max = strtoul(optarg, NULL, 0);
1029 exec_delay = strtoul(optarg, NULL, 0);
1033 log_set_max_level(LOG_DEBUG);
1034 udev_set_log_priority(udev, LOG_DEBUG);
1037 if (streq(optarg, "early")) {
1039 } else if (streq(optarg, "late")) {
1041 } else if (streq(optarg, "never")) {
1044 fprintf(stderr, "resolve-names must be early, late or never\n");
1045 log_error("resolve-names must be early, late or never\n");
1050 printf("Usage: udevd OPTIONS\n"
1053 " --children-max=<maximum number of workers>\n"
1054 " --exec-delay=<seconds to wait before executing RUN=>\n"
1055 " --resolve-names=early|late|never\n"
1061 printf("%s\n", VERSION);
1068 kernel_cmdline_options(udev);
1070 if (getuid() != 0) {
1071 fprintf(stderr, "root privileges required\n");
1072 log_error("root privileges required\n");
1076 /* set umask before creating any file/directory */
1080 mkdir("/run/udev", 0755);
1083 static_dev_create_from_modules(udev);
1085 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1089 fd = open("/dev/null", O_RDWR);
1091 if (write(STDOUT_FILENO, 0, 0) < 0)
1092 dup2(fd, STDOUT_FILENO);
1093 if (write(STDERR_FILENO, 0, 0) < 0)
1094 dup2(fd, STDERR_FILENO);
1095 if (fd > STDERR_FILENO)
1098 fprintf(stderr, "cannot open /dev/null\n");
1099 log_error("cannot open /dev/null\n");
1103 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1104 /* get control and netlink socket from systemd */
1105 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1106 if (udev_ctrl == NULL) {
1107 log_error("error taking over udev control socket");
1112 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1113 if (monitor == NULL) {
1114 log_error("error taking over netlink socket\n");
1119 /* get our own cgroup, we regularly kill everything udev has left behind */
1120 if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1123 /* open control and netlink socket */
1124 udev_ctrl = udev_ctrl_new(udev);
1125 if (udev_ctrl == NULL) {
1126 fprintf(stderr, "error initializing udev control socket");
1127 log_error("error initializing udev control socket");
1131 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1133 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1134 if (monitor == NULL) {
1135 fprintf(stderr, "error initializing netlink socket\n");
1136 log_error("error initializing netlink socket\n");
1140 fd_netlink = udev_monitor_get_fd(monitor);
1143 if (udev_monitor_enable_receiving(monitor) < 0) {
1144 fprintf(stderr, "error binding netlink socket\n");
1145 log_error("error binding netlink socket\n");
1150 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1151 fprintf(stderr, "error binding udev control socket\n");
1152 log_error("error binding udev control socket\n");
1157 udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
1159 /* create queue file before signalling 'ready', to make sure we block 'settle' */
1160 udev_queue_export = udev_queue_export_new(udev);
1161 if (udev_queue_export == NULL) {
1162 log_error("error creating queue file\n");
1174 log_error("fork of daemon failed: %m\n");
1179 goto exit_daemonize;
1184 write_one_line_file("/proc/self/oom_score_adj", "-1000");
1186 sd_notify(1, "READY=1");
1189 print_kmsg("starting version " VERSION "\n");
1194 fd = open("/dev/null", O_RDWR);
1196 dup2(fd, STDIN_FILENO);
1197 dup2(fd, STDOUT_FILENO);
1198 dup2(fd, STDERR_FILENO);
1203 fd_inotify = udev_watch_init(udev);
1204 if (fd_inotify < 0) {
1205 fprintf(stderr, "error initializing inotify\n");
1206 log_error("error initializing inotify\n");
1210 udev_watch_restore(udev);
1212 /* block and listen to all signals on signalfd */
1214 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1215 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1216 if (fd_signal < 0) {
1217 fprintf(stderr, "error creating signalfd\n");
1218 log_error("error creating signalfd\n");
1223 /* unnamed socket from workers to the main daemon */
1224 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1225 fprintf(stderr, "error creating socketpair\n");
1226 log_error("error creating socketpair\n");
1230 fd_worker = worker_watch[READ_END];
1232 udev_builtin_init(udev);
1234 rules = udev_rules_new(udev, resolve_names);
1235 if (rules == NULL) {
1236 log_error("error reading rules\n");
1240 memset(&ep_ctrl, 0, sizeof(struct epoll_event));
1241 ep_ctrl.events = EPOLLIN;
1242 ep_ctrl.data.fd = fd_ctrl;
1244 memset(&ep_inotify, 0, sizeof(struct epoll_event));
1245 ep_inotify.events = EPOLLIN;
1246 ep_inotify.data.fd = fd_inotify;
1248 memset(&ep_signal, 0, sizeof(struct epoll_event));
1249 ep_signal.events = EPOLLIN;
1250 ep_signal.data.fd = fd_signal;
1252 memset(&ep_netlink, 0, sizeof(struct epoll_event));
1253 ep_netlink.events = EPOLLIN;
1254 ep_netlink.data.fd = fd_netlink;
1256 memset(&ep_worker, 0, sizeof(struct epoll_event));
1257 ep_worker.events = EPOLLIN;
1258 ep_worker.data.fd = fd_worker;
1260 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1262 log_error("error creating epoll fd: %m\n");
1265 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1266 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1267 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1268 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1269 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1270 log_error("fail to add fds to epoll: %m\n");
1274 if (children_max <= 0) {
1275 int memsize = mem_size_mb();
1277 /* set value depending on the amount of RAM */
1279 children_max = 16 + (memsize / 8);
1283 log_debug("set children_max to %u\n", children_max);
1285 udev_rules_apply_static_dev_perms(rules);
1287 udev_list_node_init(&event_list);
1288 udev_list_node_init(&worker_list);
1291 static usec_t last_usec;
1292 struct epoll_event ev[8];
1295 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1299 /* close sources of new events and discard buffered events */
1301 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1304 if (monitor != NULL) {
1305 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1306 udev_monitor_unref(monitor);
1309 if (fd_inotify >= 0) {
1310 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1315 /* discard queued events and kill workers */
1316 event_queue_cleanup(udev, EVENT_QUEUED);
1319 /* exit after all has cleaned up */
1320 if (udev_list_node_is_empty(&event_list) && udev_list_node_is_empty(&worker_list))
1323 /* timeout at exit for workers to finish */
1324 timeout = 30 * 1000;
1325 } else if (udev_list_node_is_empty(&event_list) && !children) {
1329 /* cleanup possible left-over processes in our cgroup */
1331 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1333 /* kill idle or hanging workers */
1336 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1341 struct udev_list_node *loop;
1345 log_error("timeout, giving up waiting for workers to finish\n");
1349 /* kill idle workers */
1350 if (udev_list_node_is_empty(&event_list)) {
1351 log_debug("cleanup idle workers\n");
1355 /* check for hanging events */
1356 udev_list_node_foreach(loop, &worker_list) {
1357 struct worker *worker = node_to_worker(loop);
1359 if (worker->state != WORKER_RUNNING)
1362 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * 1000 * 1000) {
1363 log_error("worker [%u] %s timeout; kill it\n", worker->pid,
1364 worker->event ? worker->event->devpath : "<idle>");
1365 kill(worker->pid, SIGKILL);
1366 worker->state = WORKER_KILLED;
1367 /* drop reference taken for state 'running' */
1368 worker_unref(worker);
1369 if (worker->event) {
1370 log_error("seq %llu '%s' killed\n",
1371 udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1372 worker->event->exitcode = -64;
1373 event_queue_delete(worker->event, true);
1374 worker->event = NULL;
1381 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1382 for (i = 0; i < fdcount; i++) {
1383 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1385 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1387 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1389 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1391 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1395 /* check for changed config, every 3 seconds at most */
1396 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * 1000 * 1000) {
1397 if (udev_rules_check_timestamp(rules))
1399 if (udev_builtin_validate(udev))
1402 last_usec = now(CLOCK_MONOTONIC);
1405 /* reload requested, HUP signal received, rules changed, builtin changed */
1408 rules = udev_rules_unref(rules);
1409 udev_builtin_exit(udev);
1413 /* event has finished */
1415 worker_returned(fd_worker);
1418 struct udev_device *dev;
1420 dev = udev_monitor_receive_device(monitor);
1422 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1423 if (event_queue_insert(dev) < 0)
1424 udev_device_unref(dev);
1428 /* start new events */
1429 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1430 udev_builtin_init(udev);
1432 rules = udev_rules_new(udev, resolve_names);
1434 event_queue_start(udev);
1438 struct signalfd_siginfo fdsi;
1441 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1442 if (size == sizeof(struct signalfd_siginfo))
1443 handle_signal(udev, fdsi.ssi_signo);
1446 /* we are shutting down, the events below are not handled anymore */
1450 /* device node watch */
1452 handle_inotify(udev);
1455 * This needs to be after the inotify handling, to make sure,
1456 * that the ping is send back after the possibly generated
1457 * "change" events by the inotify device node watch.
1459 * A single time we may receive a client connection which we need to
1460 * keep open to block the client. It will be closed right before we
1464 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1469 udev_queue_export_cleanup(udev_queue_export);
1470 udev_ctrl_cleanup(udev_ctrl);
1474 worker_list_cleanup(udev);
1475 event_queue_cleanup(udev, EVENT_UNDEF);
1476 udev_rules_unref(rules);
1477 udev_builtin_exit(udev);
1480 if (worker_watch[READ_END] >= 0)
1481 close(worker_watch[READ_END]);
1482 if (worker_watch[WRITE_END] >= 0)
1483 close(worker_watch[WRITE_END]);
1484 udev_monitor_unref(monitor);
1485 udev_queue_export_unref(udev_queue_export);
1486 udev_ctrl_connection_unref(ctrl_conn);
1487 udev_ctrl_unref(udev_ctrl);