X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=udev%2Fudevd.c;h=d1f7cd8909f26b782f52e7529dbce4a90b72feff;hp=873634fe4519cf3149201b19727aee22e1b880a1;hb=aa29418a13dc1e1eedca702663e733cc8d9f9859;hpb=35df38c36a65bd0a9fd108f7e10ea6a593d5cdd7

diff --git a/udev/udevd.c b/udev/udevd.c
index 873634fe4..d1f7cd890 100644
--- a/udev/udevd.c
+++ b/udev/udevd.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2004-2009 Kay Sievers <kay.sievers@vrfy.org>
+ * Copyright (C) 2004-2011 Kay Sievers <kay.sievers@vrfy.org>
  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
  * Copyright (C) 2009 Canonical Ltd.
  * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
@@ -36,7 +36,7 @@
 #include <sys/socket.h>
 #include <sys/un.h>
 #include <sys/signalfd.h>
-#include <sys/select.h>
+#include <sys/epoll.h>
 #include <sys/poll.h>
 #include <sys/wait.h>
 #include <sys/stat.h>
@@ -45,9 +45,7 @@
 #include <sys/utsname.h>
 
 #include "udev.h"
-
-#define UDEVD_PRIORITY			-4
-#define UDEV_PRIORITY			-2
+#include "sd-daemon.h"
 
 static bool debug;
 
@@ -57,13 +55,12 @@ static void log_fn(struct udev *udev, int priority,
 {
 	if (debug) {
 		char buf[1024];
-		struct timeval tv;
-		struct timezone tz;
+		struct timespec ts;
 
 		vsnprintf(buf, sizeof(buf), format, args);
-		gettimeofday(&tv, &tz);
-		fprintf(stderr, "%llu.%06u [%u] %s: %s",
-			(unsigned long long) tv.tv_sec, (unsigned int) tv.tv_usec,
+		clock_gettime(CLOCK_MONOTONIC, &ts);
+		fprintf(stderr, "[%llu.%06u] [%u] %s: %s",
+			(unsigned long long) ts.tv_sec, (unsigned int) ts.tv_nsec/1000,
 			(int) getpid(), fn, buf);
 	} else {
 		vsyslog(priority, format, args);
@@ -74,34 +71,19 @@ static struct udev_rules *rules;
 static struct udev_queue_export *udev_queue_export;
 static struct udev_ctrl *udev_ctrl;
 static struct udev_monitor *monitor;
-static int worker_watch[2];
-static pid_t settle_pid;
+static int worker_watch[2] = { -1, -1 };
+static int fd_signal = -1;
+static int fd_ep = -1;
+static int fd_inotify = -1;
 static bool stop_exec_queue;
 static bool reload_config;
 static int children;
 static int children_max;
 static int exec_delay;
-static sigset_t orig_sigmask;
-static struct udev_list_node event_list;
-static struct udev_list_node worker_list;
+static sigset_t sigmask_orig;
+static UDEV_LIST(event_list);
+static UDEV_LIST(worker_list);
 static bool udev_exit;
-static volatile sig_atomic_t worker_exit;
-
-enum poll_fd {
-	FD_CONTROL,
-	FD_NETLINK,
-	FD_INOTIFY,
-	FD_SIGNAL,
-	FD_WORKER,
-};
-
-static struct pollfd pfd[] = {
-	[FD_NETLINK] = { .events = POLLIN },
-	[FD_WORKER] =  { .events = POLLIN },
-	[FD_SIGNAL] =  { .events = POLLIN },
-	[FD_INOTIFY] = { .events = POLLIN },
-	[FD_CONTROL] = { .events = POLLIN },
-};
 
 enum event_state {
 	EVENT_UNDEF,
@@ -122,6 +104,7 @@ struct event {
 	const char *devpath_old;
 	dev_t devnum;
 	bool is_block;
+	int ifindex;
 };
 
 static struct event *node_to_event(struct udev_list_node *node)
@@ -133,6 +116,8 @@ static struct event *node_to_event(struct udev_list_node *node)
 	return (struct event *)event;
 }
 
+static void event_queue_cleanup(struct udev *udev, enum event_state type);
+
 enum worker_state {
 	WORKER_UNDEF,
 	WORKER_RUNNING,
@@ -165,61 +150,61 @@ static struct worker *node_to_worker(struct udev_list_node *node)
 	return (struct worker *)worker;
 }
 
-static void event_queue_delete(struct event *event)
+static void event_queue_delete(struct event *event, bool export)
 {
 	udev_list_node_remove(&event->node);
 
-	/* mark as failed, if "add" event returns non-zero */
-	if (event->exitcode != 0 && strcmp(udev_device_get_action(event->dev), "remove") != 0)
-		udev_queue_export_device_failed(udev_queue_export, event->dev);
-	else
+	if (export) {
 		udev_queue_export_device_finished(udev_queue_export, event->dev);
-
-	info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
+		info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
+	}
 	udev_device_unref(event->dev);
 	free(event);
 }
 
-static void event_sig_handler(int signum)
-{
-	switch (signum) {
-	case SIGALRM:
-		_exit(1);
-		break;
-	case SIGTERM:
-		worker_exit = true;
-		break;
-	}
-}
-
 static struct worker *worker_ref(struct worker *worker)
 {
 	worker->refcount++;
 	return worker;
 }
 
+static void worker_cleanup(struct worker *worker)
+{
+	udev_list_node_remove(&worker->node);
+	udev_monitor_unref(worker->monitor);
+	children--;
+	free(worker);
+}
+
 static void worker_unref(struct worker *worker)
 {
 	worker->refcount--;
 	if (worker->refcount > 0)
 		return;
-
-	udev_list_node_remove(&worker->node);
-	udev_monitor_unref(worker->monitor);
-	children--;
 	info(worker->udev, "worker [%u] cleaned up\n", worker->pid);
-	free(worker);
+	worker_cleanup(worker);
+}
+
+static void worker_list_cleanup(struct udev *udev)
+{
+	struct udev_list_node *loop, *tmp;
+
+	udev_list_node_foreach_safe(loop, tmp, &worker_list) {
+		struct worker *worker = node_to_worker(loop);
+
+		worker_cleanup(worker);
+	}
 }
 
 static void worker_new(struct event *event)
 {
+	struct udev *udev = event->udev;
 	struct worker *worker;
 	struct udev_monitor *worker_monitor;
 	pid_t pid;
-	struct sigaction act;
 
 	/* listen for new events */
-	worker_monitor = udev_monitor_new_from_netlink(event->udev, NULL);
+	worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 	if (worker_monitor == NULL)
 		return;
 	/* allow the main daemon netlink address to send devices to the worker */
@@ -227,84 +212,98 @@ static void worker_new(struct event *event)
 	udev_monitor_enable_receiving(worker_monitor);
 
 	worker = calloc(1, sizeof(struct worker));
-	if (worker == NULL)
+	if (worker == NULL) {
+		udev_monitor_unref(worker_monitor);
 		return;
+	}
 	/* worker + event reference */
 	worker->refcount = 2;
-	worker->udev = event->udev;
+	worker->udev = udev;
 
 	pid = fork();
 	switch (pid) {
 	case 0: {
-		sigset_t sigmask;
-		struct udev_device *dev;
-		struct pollfd pmon = {
-			.fd = udev_monitor_get_fd(worker_monitor),
-			.events = POLLIN,
-		};
+		struct udev_device *dev = NULL;
+		int fd_monitor;
+		struct epoll_event ep_signal, ep_monitor;
+		sigset_t mask;
+		int rc = EXIT_SUCCESS;
 
+		/* move initial device from queue */
+		dev = event->dev;
+		event->dev = NULL;
+
+		free(worker);
+		worker_list_cleanup(udev);
+		event_queue_cleanup(udev, EVENT_UNDEF);
 		udev_queue_export_unref(udev_queue_export);
 		udev_monitor_unref(monitor);
 		udev_ctrl_unref(udev_ctrl);
-		close(pfd[FD_SIGNAL].fd);
+		close(fd_signal);
+		close(fd_ep);
 		close(worker_watch[READ_END]);
-		udev_log_close();
-		udev_log_init("udevd-work");
-		setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
-
-		/* set signal handlers */
-		memset(&act, 0x00, sizeof(act));
-		act.sa_handler = event_sig_handler;
-		sigemptyset (&act.sa_mask);
-		act.sa_flags = 0;
-		sigaction(SIGTERM, &act, NULL);
-		sigaction(SIGALRM, &act, NULL);
-
-		/* unblock SIGALRM */
-		sigfillset(&sigmask);
-		sigdelset(&sigmask, SIGALRM);
-		sigprocmask(SIG_SETMASK, &sigmask, NULL);
-		/* SIGTERM is unblocked in ppoll() */
-		sigdelset(&sigmask, SIGTERM);
+
+		sigfillset(&mask);
+		fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
+		if (fd_signal < 0) {
+			err(udev, "error creating signalfd %m\n");
+			rc = 2;
+			goto out;
+		}
+
+		fd_ep = epoll_create1(EPOLL_CLOEXEC);
+		if (fd_ep < 0) {
+			err(udev, "error creating epoll fd: %m\n");
+			rc = 3;
+			goto out;
+		}
+
+		memset(&ep_signal, 0, sizeof(struct epoll_event));
+		ep_signal.events = EPOLLIN;
+		ep_signal.data.fd = fd_signal;
+
+		fd_monitor = udev_monitor_get_fd(worker_monitor);
+		memset(&ep_monitor, 0, sizeof(struct epoll_event));
+		ep_monitor.events = EPOLLIN;
+		ep_monitor.data.fd = fd_monitor;
+
+		if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
+		    epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
+			err(udev, "fail to add fds to epoll: %m\n");
+			rc = 4;
+			goto out;
+		}
 
 		/* request TERM signal if parent exits */
 		prctl(PR_SET_PDEATHSIG, SIGTERM);
 
-		/* initial device */
-		dev = event->dev;
-
-		do {
+		for (;;) {
 			struct udev_event *udev_event;
-			struct worker_message msg = {};
+			struct worker_message msg;
 			int err;
-			int failed = 0;
 
-			info(event->udev, "seq %llu running\n", udev_device_get_seqnum(dev));
+			info(udev, "seq %llu running\n", udev_device_get_seqnum(dev));
 			udev_event = udev_event_new(dev);
-			if (udev_event == NULL)
-				_exit(3);
+			if (udev_event == NULL) {
+				rc = 5;
+				goto out;
+			}
 
-			/* set timeout to prevent hanging processes */
-			alarm(UDEV_EVENT_TIMEOUT);
+			/* needed for SIGCHLD/SIGTERM in spawn() */
+			udev_event->fd_signal = fd_signal;
 
 			if (exec_delay > 0)
 				udev_event->exec_delay = exec_delay;
 
 			/* apply rules, create node, symlinks */
-			err = udev_event_execute_rules(udev_event, rules);
-
-			/* rules may change/disable the timeout */
-			if (udev_device_get_event_timeout(dev) >= 0)
-				alarm(udev_device_get_event_timeout(dev));
+			err = udev_event_execute_rules(udev_event, rules, &sigmask_orig);
 
 			if (err == 0)
-				failed = udev_event_execute_run(udev_event, &orig_sigmask);
-
-			alarm(0);
+				udev_event_execute_run(udev_event, &sigmask_orig);
 
 			/* apply/restore inotify watch */
 			if (err == 0 && udev_event->inotify_watch) {
-				udev_watch_begin(udev_event->udev, dev);
+				udev_watch_begin(udev, dev);
 				udev_device_update_db(dev);
 			}
 
@@ -312,43 +311,77 @@ static void worker_new(struct event *event)
 			udev_monitor_send_device(worker_monitor, NULL, dev);
 
 			/* send udevd the result of the event execution */
+			memset(&msg, 0, sizeof(struct worker_message));
 			if (err != 0)
 				msg.exitcode = err;
-			else if (failed != 0)
-				msg.exitcode = failed;
 			msg.pid = getpid();
 			send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 
-			info(event->udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
-			udev_event_unref(udev_event);
+			info(udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
+
 			udev_device_unref(dev);
 			dev = NULL;
 
-			/* wait for more device messages or signal from udevd */
-			while (!worker_exit) {
-				int fdcount;
+			if (udev_event->sigterm) {
+				udev_event_unref(udev_event);
+				goto out;
+			}
 
-				fdcount = ppoll(&pmon, 1, NULL, &sigmask);
-				if (fdcount < 0)
-					continue;
+			udev_event_unref(udev_event);
+
+			/* wait for more device messages from main udevd, or term signal */
+			while (dev == NULL) {
+				struct epoll_event ev[4];
+				int fdcount;
+				int i;
+
+				fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), -1);
+				if (fdcount < 0) {
+					if (errno == EINTR)
+						continue;
+					err = -errno;
+					err(udev, "failed to poll: %m\n");
+					goto out;
+				}
 
-				if (pmon.revents & POLLIN) {
-					dev = udev_monitor_receive_device(worker_monitor);
-					if (dev != NULL)
+				for (i = 0; i < fdcount; i++) {
+					if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
+						dev = udev_monitor_receive_device(worker_monitor);
 						break;
+					} else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
+						struct signalfd_siginfo fdsi;
+						ssize_t size;
+
+						size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
+						if (size != sizeof(struct signalfd_siginfo))
+							continue;
+						switch (fdsi.ssi_signo) {
+						case SIGTERM:
+							goto out;
+						}
+					}
 				}
 			}
-		} while (dev != NULL);
-
+		}
+out:
+		udev_device_unref(dev);
+		if (fd_signal >= 0)
+			close(fd_signal);
+		if (fd_ep >= 0)
+			close(fd_ep);
+		close(fd_inotify);
+		close(worker_watch[WRITE_END]);
+		udev_rules_unref(rules);
 		udev_monitor_unref(worker_monitor);
+		udev_unref(udev);
 		udev_log_close();
-		exit(0);
+		exit(rc);
 	}
 	case -1:
 		udev_monitor_unref(worker_monitor);
 		event->state = EVENT_QUEUED;
 		free(worker);
-		err(event->udev, "fork of child failed: %m\n");
+		err(udev, "fork of child failed: %m\n");
 		break;
 	default:
 		/* close monitor, but keep address around */
@@ -360,7 +393,7 @@ static void worker_new(struct event *event)
 		event->state = EVENT_RUNNING;
 		udev_list_node_append(&worker->node, &worker_list);
 		children++;
-		info(event->udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
+		info(udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
 		break;
 	}
 }
@@ -400,13 +433,13 @@ static void event_run(struct event *event, bool force)
 	worker_new(event);
 }
 
-static void event_queue_insert(struct udev_device *dev)
+static int event_queue_insert(struct udev_device *dev)
 {
 	struct event *event;
 
 	event = calloc(1, sizeof(struct event));
 	if (event == NULL)
-		return;
+		return -1;
 
 	event->udev = udev_device_get_udev(dev);
 	event->dev = dev;
@@ -416,6 +449,7 @@ static void event_queue_insert(struct udev_device *dev)
 	event->devpath_old = udev_device_get_devpath_old(dev);
 	event->devnum = udev_device_get_devnum(dev);
 	event->is_block = (strcmp("block", udev_device_get_subsystem(dev)) == 0);
+	event->ifindex = udev_device_get_ifindex(dev);
 
 	udev_queue_export_device_queued(udev_queue_export, dev);
 	info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
@@ -427,8 +461,10 @@ static void event_queue_insert(struct udev_device *dev)
 	/* run all events with a timeout set immediately */
 	if (udev_device_get_timeout(dev) > 0) {
 		event_run(event, true);
-		return;
+		return 0;
 	}
+
+	return 0;
 }
 
 static void worker_kill(struct udev *udev, int retain)
@@ -481,6 +517,10 @@ static bool is_devpath_busy(struct event *event)
 		if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 			return true;
 
+		/* check network device ifindex */
+		if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
+			return true;
+
 		/* check our old name */
 		if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) {
 			event->delaying_seqnum = loop_event->seqnum;
@@ -496,6 +536,11 @@ static bool is_devpath_busy(struct event *event)
 
 		/* identical device event found */
 		if (loop_event->devpath_len == event->devpath_len) {
+			/* devices names might have changed/swapped in the meantime */
+			if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
+				continue;
+			if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
+				continue;
 			event->delaying_seqnum = loop_event->seqnum;
 			return true;
 		}
@@ -519,7 +564,7 @@ static bool is_devpath_busy(struct event *event)
 	return false;
 }
 
-static void events_start(struct udev *udev)
+static void event_queue_start(struct udev *udev)
 {
 	struct udev_list_node *loop;
 
@@ -539,14 +584,28 @@ static void events_start(struct udev *udev)
 	}
 }
 
-static void worker_returned(void)
+static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
+{
+	struct udev_list_node *loop, *tmp;
+
+	udev_list_node_foreach_safe(loop, tmp, &event_list) {
+		struct event *event = node_to_event(loop);
+
+		if (match_type != EVENT_UNDEF && match_type != event->state)
+			continue;
+
+		event_queue_delete(event, false);
+	}
+}
+
+static void worker_returned(int fd_worker)
 {
 	for (;;) {
 		struct worker_message msg;
 		ssize_t size;
 		struct udev_list_node *loop;
 
-		size = recv(pfd[FD_WORKER].fd, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
+		size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 		if (size != sizeof(struct worker_message))
 			break;
 
@@ -559,7 +618,7 @@ static void worker_returned(void)
 
 			/* worker returned */
 			worker->event->exitcode = msg.exitcode;
-			event_queue_delete(worker->event);
+			event_queue_delete(worker->event, true);
 			worker->event = NULL;
 			if (worker->state != WORKER_KILLED)
 				worker->state = WORKER_IDLE;
@@ -570,16 +629,21 @@ static void worker_returned(void)
 }
 
 /* receive the udevd message from userspace */
-static void handle_ctrl_msg(struct udev_ctrl *uctrl)
+static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
 {
 	struct udev *udev = udev_ctrl_get_udev(uctrl);
-	struct udev_ctrl_msg *ctrl_msg;
+	struct udev_ctrl_connection *ctrl_conn;
+	struct udev_ctrl_msg *ctrl_msg = NULL;
 	const char *str;
 	int i;
 
-	ctrl_msg = udev_ctrl_receive_msg(uctrl);
+	ctrl_conn = udev_ctrl_get_connection(uctrl);
+	if (ctrl_conn == NULL)
+		goto out;
+
+	ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 	if (ctrl_msg == NULL)
-		return;
+		goto out;
 
 	i = udev_ctrl_get_set_log_level(ctrl_msg);
 	if (i >= 0) {
@@ -636,13 +700,18 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl)
 		children_max = i;
 	}
 
-	settle_pid = udev_ctrl_get_settle(ctrl_msg);
-	if (settle_pid > 0) {
-		info(udev, "udevd message (SETTLE) received\n");
-		kill(settle_pid, SIGUSR1);
-		settle_pid = 0;
+	if (udev_ctrl_get_ping(ctrl_msg) > 0)
+		info(udev, "udevd message (SYNC) received\n");
+
+	if (udev_ctrl_get_exit(ctrl_msg) > 0) {
+		info(udev, "udevd message (EXIT) received\n");
+		udev_exit = true;
+		/* keep reference to block the client until we exit */
+		udev_ctrl_connection_ref(ctrl_conn);
 	}
+out:
 	udev_ctrl_msg_unref(ctrl_msg);
+	return udev_ctrl_connection_unref(ctrl_conn);
 }
 
 /* read inotify messages */
@@ -652,7 +721,7 @@ static int handle_inotify(struct udev *udev)
 	char *buf;
 	struct inotify_event *ev;
 
-	if ((ioctl(pfd[FD_INOTIFY].fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
+	if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
 		return 0;
 
 	buf = malloc(nbytes);
@@ -661,7 +730,7 @@ static int handle_inotify(struct udev *udev)
 		return -1;
 	}
 
-	nbytes = read(pfd[FD_INOTIFY].fd, buf, nbytes);
+	nbytes = read(fd_inotify, buf, nbytes);
 
 	for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 		struct udev_device *dev;
@@ -690,9 +759,11 @@ static int handle_inotify(struct udev *udev)
 				info(udev, "device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
 				util_strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 				fd = open(filename, O_WRONLY);
-				if (fd < 0 || write(fd, "change", 6) < 0)
-					info(udev, "error writing uevent: %m\n");
-				close(fd);
+				if (fd >= 0) {
+					if (write(fd, "change", 6) < 0)
+						info(udev, "error writing uevent: %m\n");
+					close(fd);
+				}
 			}
 			if (ev->mask & IN_IGNORED)
 				udev_watch_end(udev, dev);
@@ -728,15 +799,29 @@ static void handle_signal(struct udev *udev, int signo)
 
 				if (worker->pid != pid)
 					continue;
-
 				info(udev, "worker [%u] exit\n", pid);
+
+				if (WIFEXITED(status)) {
+					if (WEXITSTATUS(status) != 0)
+						err(udev, "worker [%u] exit with return code %i\n", pid, WEXITSTATUS(status));
+				} else if (WIFSIGNALED(status)) {
+					err(udev, "worker [%u] terminated by signal %i (%s)\n",
+					    pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
+				} else if (WIFSTOPPED(status)) {
+					err(udev, "worker [%u] stopped\n", pid);
+				} else if (WIFCONTINUED(status)) {
+					err(udev, "worker [%u] continued\n", pid);
+				} else {
+					err(udev, "worker [%u] exit with status 0x%04x\n", pid, status);
+				}
+
 				if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
-					err(udev, "worker [%u] unexpectedly returned with status 0x%04x\n", pid, status);
 					if (worker->event != NULL) {
-						err(udev, "worker [%u] failed while handling '%s'\n", pid, worker->event->devpath);
+						err(udev, "worker [%u] failed while handling '%s'\n",
+						    pid, worker->event->devpath);
 						worker->event->exitcode = -32;
-						event_queue_delete(worker->event);
-						/* drop reference from running event */
+						event_queue_delete(worker->event, true);
+						/* drop reference taken for state 'running' */
 						worker_unref(worker);
 					}
 				}
@@ -798,15 +883,16 @@ static void static_dev_create_from_modules(struct udev *udev)
 		if (sscanf(devno, "%c%u:%u", &type, &maj, &min) != 3)
 			continue;
 
+		/* set sticky bit, so we do not remove the node on module unload */
 		if (type == 'c')
-			mode = 0600 | S_IFCHR;
+			mode = 01600|S_IFCHR;
 		else if (type == 'b')
-			mode = 0600 | S_IFBLK;
+			mode = 01600|S_IFBLK;
 		else
 			continue;
 
 		util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/", devname, NULL);
-		util_create_path(udev, filename);
+		util_create_path_selinux(udev, filename);
 		udev_selinux_setfscreatecon(udev, filename, mode);
 		info(udev, "mknod '%s' %c%u:%u\n", filename, type, maj, min);
 		if (mknod(filename, mode, makedev(maj, min)) < 0 && errno == EEXIST)
@@ -817,7 +903,7 @@ static void static_dev_create_from_modules(struct udev *udev)
 	fclose(f);
 }
 
-static int copy_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
+static int copy_dev_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
 {
 	struct dirent *dent;
 
@@ -870,7 +956,7 @@ static int copy_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
 				continue;
 			}
 
-			copy_dir(udev, dir2_from, dir2_to, maxdepth-1);
+			copy_dev_dir(udev, dir2_from, dir2_to, maxdepth-1);
 
 			closedir(dir2_to);
 			closedir(dir2_from);
@@ -896,10 +982,14 @@ static void static_dev_create_links(struct udev *udev, DIR *dir)
 	unsigned int i;
 
 	for (i = 0; i < ARRAY_SIZE(stdlinks); i++) {
-		udev_selinux_setfscreateconat(udev, dirfd(dir), stdlinks[i].link, S_IFLNK);
-		if (symlinkat(stdlinks[i].target, dirfd(dir), stdlinks[i].link) < 0 && errno == EEXIST)
-			utimensat(dirfd(dir), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW);
-		udev_selinux_resetfscreatecon(udev);
+		struct stat sb;
+
+		if (stat(stdlinks[i].target, &sb) == 0) {
+			udev_selinux_setfscreateconat(udev, dirfd(dir), stdlinks[i].link, S_IFLNK);
+			if (symlinkat(stdlinks[i].target, dirfd(dir), stdlinks[i].link) < 0 && errno == EEXIST)
+				utimensat(dirfd(dir), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW);
+			udev_selinux_resetfscreatecon(udev);
+		}
 	}
 }
 
@@ -910,7 +1000,7 @@ static void static_dev_create_from_devices(struct udev *udev, DIR *dir)
 	dir_from = opendir(LIBEXECDIR "/devices");
 	if (dir_from == NULL)
 		return;
-	copy_dir(udev, dir_from, dir, 8);
+	copy_dev_dir(udev, dir_from, dir, 8);
 	closedir(dir_from);
 }
 
@@ -951,87 +1041,137 @@ static int mem_size_mb(void)
 	return memsize;
 }
 
-static int init_notify(const char *state)
+static int convert_db(struct udev *udev)
 {
-	int fd = -1, r;
-	struct msghdr msghdr;
-	struct iovec iovec;
-	struct ucred *ucred;
-	union {
-		struct sockaddr sa;
-		struct sockaddr_un un;
-	} sockaddr;
-	union {
-		struct cmsghdr cmsghdr;
-		uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
-	} control;
-	const char *e;
-
-	if (!(e = getenv("NOTIFY_SOCKET"))) {
-		r = 0;
-		goto finish;
-	}
+	char filename[UTIL_PATH_SIZE];
+	FILE *f;
+	struct udev_enumerate *udev_enumerate;
+	struct udev_list_entry *list_entry;
 
-	/* Must be an abstract socket, or an absolute path */
-	if ((e[0] != '@' && e[0] != '/') || e[1] == 0) {
-		r = -EINVAL;
-		goto finish;
-	}
+	/* current database */
+	util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/data", NULL);
+	if (access(filename, F_OK) >= 0)
+		return 0;
+
+	/* make sure we do not get here again */
+	util_create_path(udev, filename);
+	mkdir(filename, 0755);
 
-	if ((fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0)) < 0) {
-		r = -errno;
-		goto finish;
+	/* old database */
+	util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/db", NULL);
+	if (access(filename, F_OK) < 0)
+		return 0;
+
+	f = fopen("/dev/kmsg", "w");
+	if (f != NULL) {
+		fprintf(f, "<30>udevd[%u]: converting old udev database\n", getpid());
+		fclose(f);
 	}
 
-	memset(&sockaddr, 0, sizeof(sockaddr));
-	sockaddr.sa.sa_family = AF_UNIX;
-	strncpy(sockaddr.un.sun_path, e, sizeof(sockaddr.un.sun_path));
-
-	if (sockaddr.un.sun_path[0] == '@')
-		sockaddr.un.sun_path[0] = 0;
-
-	memset(&iovec, 0, sizeof(iovec));
-	iovec.iov_base = (char*) state;
-	iovec.iov_len = strlen(state);
-
-	memset(&control, 0, sizeof(control));
-	control.cmsghdr.cmsg_level = SOL_SOCKET;
-	control.cmsghdr.cmsg_type = SCM_CREDENTIALS;
-	control.cmsghdr.cmsg_len = CMSG_LEN(sizeof(struct ucred));
-
-	ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
-	ucred->pid = getpid();
-	ucred->uid = getuid();
-	ucred->gid = getgid();
-
-	memset(&msghdr, 0, sizeof(msghdr));
-	msghdr.msg_name = &sockaddr;
-	msghdr.msg_namelen = sizeof(sa_family_t) + strlen(e);
-	if (msghdr.msg_namelen > sizeof(struct sockaddr_un))
-		msghdr.msg_namelen = sizeof(struct sockaddr_un);
-	msghdr.msg_iov = &iovec;
-	msghdr.msg_iovlen = 1;
-	msghdr.msg_control = &control;
-	msghdr.msg_controllen = control.cmsghdr.cmsg_len;
-
-	if (sendmsg(fd, &msghdr, MSG_NOSIGNAL) < 0) {
-		r = -errno;
-		goto finish;
+	udev_enumerate = udev_enumerate_new(udev);
+	if (udev_enumerate == NULL)
+		return -1;
+	udev_enumerate_scan_devices(udev_enumerate);
+	udev_list_entry_foreach(list_entry, udev_enumerate_get_list_entry(udev_enumerate)) {
+		struct udev_device *device;
+
+		device = udev_device_new_from_syspath(udev, udev_list_entry_get_name(list_entry));
+		if (device == NULL)
+			continue;
+
+		/* try to find the old database for devices without a current one */
+		if (udev_device_read_db(device, NULL) < 0) {
+			bool have_db;
+			const char *id;
+			struct stat stats;
+			char devpath[UTIL_PATH_SIZE];
+			char from[UTIL_PATH_SIZE];
+
+			have_db = false;
+
+			/* find database in old location */
+			id = udev_device_get_id_filename(device);
+			util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", id, NULL);
+			if (lstat(from, &stats) == 0) {
+				if (!have_db) {
+					udev_device_read_db(device, from);
+					have_db = true;
+				}
+				unlink(from);
+			}
+
+			/* find old database with $subsys:$sysname name */
+			util_strscpyl(from, sizeof(from), udev_get_dev_path(udev),
+				     "/.udev/db/", udev_device_get_subsystem(device), ":",
+				     udev_device_get_sysname(device), NULL);
+			if (lstat(from, &stats) == 0) {
+				if (!have_db) {
+					udev_device_read_db(device, from);
+					have_db = true;
+				}
+				unlink(from);
+			}
+
+			/* find old database with the encoded devpath name */
+			util_path_encode(udev_device_get_devpath(device), devpath, sizeof(devpath));
+			util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", devpath, NULL);
+			if (lstat(from, &stats) == 0) {
+				if (!have_db) {
+					udev_device_read_db(device, from);
+					have_db = true;
+				}
+				unlink(from);
+			}
+
+			/* write out new database */
+			if (have_db)
+				udev_device_update_db(device);
+		}
+		udev_device_unref(device);
 	}
+	udev_enumerate_unref(udev_enumerate);
+	return 0;
+}
 
-	r = 0;
+static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
+{
+	int ctrl = -1, netlink = -1;
+	int fd, n;
+
+	n = sd_listen_fds(true);
+	if (n <= 0)
+		return -1;
+
+	for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
+		if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
+			if (ctrl >= 0)
+				return -1;
+			ctrl = fd;
+			continue;
+		}
 
-finish:
-	if (fd >= 0)
-		close(fd);
+		if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
+			if (netlink >= 0)
+				return -1;
+			netlink = fd;
+			continue;
+		}
 
-	return r;
+		return -1;
+	}
+
+	if (ctrl < 0 || netlink < 0)
+		return -1;
+
+	info(udev, "ctrl=%i netlink=%i\n", ctrl, netlink);
+	*rctrl = ctrl;
+	*rnetlink = netlink;
+	return 0;
 }
 
 int main(int argc, char *argv[])
 {
 	struct udev *udev;
-	int fd;
 	FILE *f;
 	sigset_t mask;
 	int daemonize = false;
@@ -1046,6 +1186,11 @@ int main(int argc, char *argv[])
 		{ "version", no_argument, NULL, 'V' },
 		{}
 	};
+	int fd_ctrl = -1;
+	int fd_netlink = -1;
+	int fd_worker = -1;
+	struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
+	struct udev_ctrl_connection *ctrl_conn = NULL;
 	int rc = 1;
 
 	udev = udev_new();
@@ -1057,10 +1202,35 @@ int main(int argc, char *argv[])
 	info(udev, "version %s\n", VERSION);
 	udev_selinux_init(udev);
 
+	/* make sure, that our runtime dir exists and is writable */
+	if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) < 0) {
+		/* try to create our own subdirectory, do not create parent directories */
+		mkdir(udev_get_run_config_path(udev), 0755);
+
+		if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) >= 0) {
+			/* directory seems writable now */
+			udev_set_run_path(udev, udev_get_run_config_path(udev));
+		} else {
+			/* fall back to /dev/.udev */
+			char filename[UTIL_PATH_SIZE];
+
+			util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev", NULL);
+			if (udev_set_run_path(udev, filename) == NULL)
+				goto exit;
+			mkdir(udev_get_run_path(udev), 0755);
+			err(udev, "error: runtime directory '%s' not writable, for now falling back to '%s'",
+			    udev_get_run_config_path(udev), udev_get_run_path(udev));
+		}
+	}
+	/* relabel runtime dir only if it resides below /dev */
+	if (strncmp(udev_get_run_path(udev), udev_get_dev_path(udev), strlen(udev_get_dev_path(udev))) == 0)
+		udev_selinux_lsetfilecon(udev, udev_get_run_path(udev), 0755);
+	info(udev, "runtime dir '%s'\n", udev_get_run_path(udev));
+
 	for (;;) {
 		int option;
 
-		option = getopt_long(argc, argv, "cdeDthV", options, NULL);
+		option = getopt_long(argc, argv, "c:deDtN:hV", options, NULL);
 		if (option == -1)
 			break;
 
@@ -1155,44 +1325,145 @@ int main(int argc, char *argv[])
 	chdir("/");
 	umask(022);
 
+	/* create standard links, copy static nodes, create nodes from modules */
+	static_dev_create(udev);
+	static_dev_create_from_modules(udev);
+
 	/* before opening new files, make sure std{in,out,err} fds are in a sane state */
-	fd = open("/dev/null", O_RDWR);
-	if (fd < 0) {
-		fprintf(stderr, "cannot open /dev/null\n");
-		err(udev, "cannot open /dev/null\n");
+	if (daemonize) {
+		int fd;
+
+		fd = open("/dev/null", O_RDWR);
+		if (fd >= 0) {
+			if (write(STDOUT_FILENO, 0, 0) < 0)
+				dup2(fd, STDOUT_FILENO);
+			if (write(STDERR_FILENO, 0, 0) < 0)
+				dup2(fd, STDERR_FILENO);
+			if (fd > STDERR_FILENO)
+				close(fd);
+		} else {
+			fprintf(stderr, "cannot open /dev/null\n");
+			err(udev, "cannot open /dev/null\n");
+		}
 	}
-	if (write(STDOUT_FILENO, 0, 0) < 0)
-		dup2(fd, STDOUT_FILENO);
-	if (write(STDERR_FILENO, 0, 0) < 0)
-		dup2(fd, STDERR_FILENO);
-
-	udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH);
-	if (udev_ctrl == NULL) {
-		fprintf(stderr, "error initializing control socket");
-		err(udev, "error initializing udevd socket");
-		rc = 1;
+
+	if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
+		/* get control and netlink socket from from systemd */
+		udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
+		if (udev_ctrl == NULL) {
+			err(udev, "error taking over udev control socket");
+			rc = 1;
+			goto exit;
+		}
+
+		monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
+		if (monitor == NULL) {
+			err(udev, "error taking over netlink socket\n");
+			rc = 3;
+			goto exit;
+		}
+	} else {
+		/* open control and netlink socket */
+		udev_ctrl = udev_ctrl_new(udev);
+		if (udev_ctrl == NULL) {
+			fprintf(stderr, "error initializing udev control socket");
+			err(udev, "error initializing udev control socket");
+			rc = 1;
+			goto exit;
+		}
+		fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
+
+		monitor = udev_monitor_new_from_netlink(udev, "kernel");
+		if (monitor == NULL) {
+			fprintf(stderr, "error initializing netlink socket\n");
+			err(udev, "error initializing netlink socket\n");
+			rc = 3;
+			goto exit;
+		}
+		fd_netlink = udev_monitor_get_fd(monitor);
+	}
+
+	if (udev_monitor_enable_receiving(monitor) < 0) {
+		fprintf(stderr, "error binding netlink socket\n");
+		err(udev, "error binding netlink socket\n");
+		rc = 3;
 		goto exit;
 	}
+
 	if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
-		fprintf(stderr, "error binding control socket, seems udevd is already running\n");
-		err(udev, "error binding control socket, seems udevd is already running\n");
+		fprintf(stderr, "error binding udev control socket\n");
+		err(udev, "error binding udev control socket\n");
 		rc = 1;
 		goto exit;
 	}
-	pfd[FD_CONTROL].fd = udev_ctrl_get_fd(udev_ctrl);
 
-	monitor = udev_monitor_new_from_netlink(udev, "kernel");
-	if (monitor == NULL || udev_monitor_enable_receiving(monitor) < 0) {
-		fprintf(stderr, "error initializing netlink socket\n");
-		err(udev, "error initializing netlink socket\n");
-		rc = 3;
+	udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
+
+	/* create queue file before signalling 'ready', to make sure we block 'settle' */
+	udev_queue_export = udev_queue_export_new(udev);
+	if (udev_queue_export == NULL) {
+		err(udev, "error creating queue file\n");
 		goto exit;
 	}
-	udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
-	pfd[FD_NETLINK].fd = udev_monitor_get_fd(monitor);
 
-	pfd[FD_INOTIFY].fd = udev_watch_init(udev);
-	if (pfd[FD_INOTIFY].fd < 0) {
+	if (daemonize) {
+		pid_t pid;
+		int fd;
+
+		pid = fork();
+		switch (pid) {
+		case 0:
+			break;
+		case -1:
+			err(udev, "fork of daemon failed: %m\n");
+			rc = 4;
+			goto exit;
+		default:
+			rc = EXIT_SUCCESS;
+			goto exit_daemonize;
+		}
+
+		setsid();
+
+		fd = open("/proc/self/oom_score_adj", O_RDWR);
+		if (fd < 0) {
+			/* Fallback to old interface */
+			fd = open("/proc/self/oom_adj", O_RDWR);
+			if (fd < 0) {
+				err(udev, "error disabling OOM: %m\n");
+			} else {
+				/* OOM_DISABLE == -17 */
+				write(fd, "-17", 3);
+				close(fd);
+			}
+		} else {
+			write(fd, "-1000", 5);
+			close(fd);
+		}
+	} else {
+		sd_notify(1, "READY=1");
+	}
+
+	f = fopen("/dev/kmsg", "w");
+	if (f != NULL) {
+		fprintf(f, "<30>udevd[%u]: starting version " VERSION "\n", getpid());
+		fclose(f);
+	}
+
+	if (!debug) {
+		int fd;
+
+		fd = open("/dev/null", O_RDWR);
+		if (fd >= 0) {
+			dup2(fd, STDIN_FILENO);
+			dup2(fd, STDOUT_FILENO);
+			dup2(fd, STDERR_FILENO);
+			close(fd);
+		}
+	}
+
+	fd_inotify = udev_watch_init(udev);
+	if (fd_inotify < 0) {
 		fprintf(stderr, "error initializing inotify\n");
 		err(udev, "error initializing inotify\n");
 		rc = 4;
@@ -1200,49 +1471,49 @@ int main(int argc, char *argv[])
 	}
 
 	if (udev_get_rules_path(udev) != NULL) {
-		inotify_add_watch(pfd[FD_INOTIFY].fd, udev_get_rules_path(udev),
+		inotify_add_watch(fd_inotify, udev_get_rules_path(udev),
 				  IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
 	} else {
 		char filename[UTIL_PATH_SIZE];
 		struct stat statbuf;
 
-		inotify_add_watch(pfd[FD_INOTIFY].fd, LIBEXECDIR "/rules.d",
+		inotify_add_watch(fd_inotify, LIBEXECDIR "/rules.d",
 				  IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
-		inotify_add_watch(pfd[FD_INOTIFY].fd, SYSCONFDIR "/udev/rules.d",
+		inotify_add_watch(fd_inotify, SYSCONFDIR "/udev/rules.d",
 				  IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
 
 		/* watch dynamic rules directory */
-		util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/rules.d", NULL);
+		util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/rules.d", NULL);
 		if (stat(filename, &statbuf) != 0) {
 			util_create_path(udev, filename);
-			udev_selinux_setfscreatecon(udev, filename, S_IFDIR|0755);
 			mkdir(filename, 0755);
-			udev_selinux_resetfscreatecon(udev);
 		}
-		inotify_add_watch(pfd[FD_INOTIFY].fd, filename,
+		inotify_add_watch(fd_inotify, filename,
 				  IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
 	}
 	udev_watch_restore(udev);
 
 	/* block and listen to all signals on signalfd */
 	sigfillset(&mask);
-	sigprocmask(SIG_SETMASK, &mask, &orig_sigmask);
-	pfd[FD_SIGNAL].fd = signalfd(-1, &mask, 0);
-	if (pfd[FD_SIGNAL].fd < 0) {
-		fprintf(stderr, "error getting signalfd\n");
-		err(udev, "error getting signalfd\n");
+	sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
+	fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
+	if (fd_signal < 0) {
+		fprintf(stderr, "error creating signalfd\n");
+		err(udev, "error creating signalfd\n");
 		rc = 5;
 		goto exit;
 	}
 
 	/* unnamed socket from workers to the main daemon */
 	if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
-		fprintf(stderr, "error getting socketpair\n");
-		err(udev, "error getting socketpair\n");
+		fprintf(stderr, "error creating socketpair\n");
+		err(udev, "error creating socketpair\n");
 		rc = 6;
 		goto exit;
 	}
-	pfd[FD_WORKER].fd = worker_watch[READ_END];
+	fd_worker = worker_watch[READ_END];
+
+	udev_builtin_load(udev);
 
 	rules = udev_rules_new(udev, resolve_names);
 	if (rules == NULL) {
@@ -1250,59 +1521,43 @@ int main(int argc, char *argv[])
 		goto exit;
 	}
 
-	udev_queue_export = udev_queue_export_new(udev);
-	if (udev_queue_export == NULL) {
-		err(udev, "error creating queue file\n");
-		goto exit;
-	}
-
-	if (!debug) {
-		dup2(fd, STDIN_FILENO);
-		dup2(fd, STDOUT_FILENO);
-		dup2(fd, STDERR_FILENO);
-	}
-	if (fd > STDERR_FILENO)
-		close(fd);
+	memset(&ep_ctrl, 0, sizeof(struct epoll_event));
+	ep_ctrl.events = EPOLLIN;
+	ep_ctrl.data.fd = fd_ctrl;
 
-	if (daemonize) {
-		pid_t pid;
+	memset(&ep_inotify, 0, sizeof(struct epoll_event));
+	ep_inotify.events = EPOLLIN;
+	ep_inotify.data.fd = fd_inotify;
 
-		pid = fork();
-		switch (pid) {
-		case 0:
-			break;
-		case -1:
-			err(udev, "fork of daemon failed: %m\n");
-			rc = 4;
-			goto exit;
-		default:
-			rc = 0;
-			goto exit;
-		}
-	} else {
-		init_notify("READY=1");
-	}
+	memset(&ep_signal, 0, sizeof(struct epoll_event));
+	ep_signal.events = EPOLLIN;
+	ep_signal.data.fd = fd_signal;
 
-	/* set scheduling priority for the main daemon process */
-	setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
+	memset(&ep_netlink, 0, sizeof(struct epoll_event));
+	ep_netlink.events = EPOLLIN;
+	ep_netlink.data.fd = fd_netlink;
 
-	setsid();
+	memset(&ep_worker, 0, sizeof(struct epoll_event));
+	ep_worker.events = EPOLLIN;
+	ep_worker.data.fd = fd_worker;
 
-	f = fopen("/dev/kmsg", "w");
-	if (f != NULL) {
-		fprintf(f, "<6>udev: starting version " VERSION "\n");
-		fclose(f);
+	fd_ep = epoll_create1(EPOLL_CLOEXEC);
+	if (fd_ep < 0) {
+		err(udev, "error creating epoll fd: %m\n");
+		goto exit;
 	}
-
-	/* OOM_DISABLE == -17 */
-	fd = open("/proc/self/oom_adj", O_RDWR);
-	if (fd < 0) {
-		err(udev, "error disabling OOM: %m\n");
-	} else {
-		write(fd, "-17", 3);
-		close(fd);
+	if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
+	    epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
+	    epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
+	    epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
+	    epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
+		err(udev, "fail to add fds to epoll: %m\n");
+		goto exit;
 	}
 
+	/* if needed, convert old database from earlier udev version */
+	convert_db(udev);
+
 	if (children_max <= 0) {
 		int memsize = mem_size_mb();
 
@@ -1314,73 +1569,126 @@ int main(int argc, char *argv[])
 	}
 	info(udev, "set children_max to %u\n", children_max);
 
-	static_dev_create(udev);
-	static_dev_create_from_modules(udev);
 	udev_rules_apply_static_dev_perms(rules);
 
-	udev_list_init(&event_list);
-	udev_list_init(&worker_list);
+	udev_list_node_init(&event_list);
+	udev_list_node_init(&worker_list);
 
-	while (!udev_exit) {
+	for (;;) {
+		struct epoll_event ev[8];
 		int fdcount;
 		int timeout;
+		bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
+		int i;
+
+		if (udev_exit) {
+			/* close sources of new events and discard buffered events */
+			if (fd_ctrl >= 0) {
+				epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
+				fd_ctrl = -1;
+			}
+			if (monitor != NULL) {
+				epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
+				udev_monitor_unref(monitor);
+				monitor = NULL;
+			}
+			if (fd_inotify >= 0) {
+				epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
+				close(fd_inotify);
+				fd_inotify = -1;
+			}
+
+			/* discard queued events and kill workers */
+			event_queue_cleanup(udev, EVENT_QUEUED);
+			worker_kill(udev, 0);
+
+			/* exit after all has cleaned up */
+			if (udev_list_node_is_empty(&event_list) && udev_list_node_is_empty(&worker_list))
+				break;
 
-		/* set timeout to kill idle workers */
-		if (udev_list_is_empty(&event_list) && children > 2)
+			/* timeout at exit for workers to finish */
+			timeout = 60 * 1000;
+		} else if (udev_list_node_is_empty(&event_list) && children > 2) {
+			/* set timeout to kill idle workers */
 			timeout = 3 * 1000;
-		else
+		} else {
 			timeout = -1;
-		/* wait for events */
-		fdcount = poll(pfd, ARRAY_SIZE(pfd), timeout);
+		}
+		fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), timeout);
 		if (fdcount < 0)
 			continue;
 
-		/* timeout - kill idle workers */
-		if (fdcount == 0)
+		if (fdcount == 0) {
+			if (udev_exit) {
+				info(udev, "timeout, giving up waiting for workers to finish\n");
+				break;
+			}
+
+			/* timeout - kill idle workers */
 			worker_kill(udev, 2);
+		}
+
+		is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
+		for (i = 0; i < fdcount; i++) {
+			if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
+				is_worker = true;
+			else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
+				is_netlink = true;
+			else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
+				is_signal = true;
+			else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
+				is_inotify = true;
+			else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
+				is_ctrl = true;
+		}
 
 		/* event has finished */
-		if (pfd[FD_WORKER].revents & POLLIN)
-			worker_returned();
+		if (is_worker)
+			worker_returned(fd_worker);
 
-		/* get kernel uevent */
-		if (pfd[FD_NETLINK].revents & POLLIN) {
+		if (is_netlink) {
 			struct udev_device *dev;
 
 			dev = udev_monitor_receive_device(monitor);
-			if (dev != NULL)
-				event_queue_insert(dev);
-			else
-				udev_device_unref(dev);
+			if (dev != NULL) {
+				udev_device_set_usec_initialized(dev, now_usec());
+				if (event_queue_insert(dev) < 0)
+					udev_device_unref(dev);
+			}
 		}
 
 		/* start new events */
-		if (!udev_list_is_empty(&event_list) && !stop_exec_queue)
-			events_start(udev);
+		if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue)
+			event_queue_start(udev);
 
-		/* get signal */
-		if (pfd[FD_SIGNAL].revents & POLLIN) {
+		if (is_signal) {
 			struct signalfd_siginfo fdsi;
 			ssize_t size;
 
-			size = read(pfd[FD_SIGNAL].fd, &fdsi, sizeof(struct signalfd_siginfo));
+			size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 			if (size == sizeof(struct signalfd_siginfo))
 				handle_signal(udev, fdsi.ssi_signo);
 		}
 
+		/* we are shutting down, the events below are not handled anymore */
+		if (udev_exit)
+			continue;
+
 		/* device node and rules directory inotify watch */
-		if (pfd[FD_INOTIFY].revents & POLLIN)
+		if (is_inotify)
 			handle_inotify(udev);
 
 		/*
-		 * get control message
-		 *
 		 * This needs to be after the inotify handling, to make sure,
-		 * that the settle signal is send back after the possibly generated
+		 * that the ping is send back after the possibly generated
 		 * "change" events by the inotify device node watch.
+		 *
+		 * A single time we may receive a client connection which we need to
+		 * keep open to block the client. It will be closed right before we
+		 * exit.
 		 */
-		if (pfd[FD_CONTROL].revents & POLLIN)
-			handle_ctrl_msg(udev_ctrl);
+		if (is_ctrl)
+			ctrl_conn = handle_ctrl_msg(udev_ctrl);
 
 		/* rules changed, set by inotify or a HUP signal */
 		if (reload_config) {
@@ -1396,19 +1704,27 @@ int main(int argc, char *argv[])
 		}
 	}
 
-	udev_queue_export_cleanup(udev_queue_export);
-	rc = 0;
+	rc = EXIT_SUCCESS;
 exit:
-	udev_queue_export_unref(udev_queue_export);
+	udev_queue_export_cleanup(udev_queue_export);
+	udev_ctrl_cleanup(udev_ctrl);
+exit_daemonize:
+	if (fd_ep >= 0)
+		close(fd_ep);
+	worker_list_cleanup(udev);
+	event_queue_cleanup(udev, EVENT_UNDEF);
 	udev_rules_unref(rules);
-	udev_ctrl_unref(udev_ctrl);
-	if (pfd[FD_SIGNAL].fd >= 0)
-		close(pfd[FD_SIGNAL].fd);
+	udev_builtin_unload(udev);
+	if (fd_signal >= 0)
+		close(fd_signal);
 	if (worker_watch[READ_END] >= 0)
 		close(worker_watch[READ_END]);
 	if (worker_watch[WRITE_END] >= 0)
 		close(worker_watch[WRITE_END]);
 	udev_monitor_unref(monitor);
+	udev_queue_export_unref(udev_queue_export);
+	udev_ctrl_connection_unref(ctrl_conn);
+	udev_ctrl_unref(udev_ctrl);
 	udev_selinux_exit(udev);
 	udev_unref(udev);
 	udev_log_close();