X-Git-Url: https://arthur.barton.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fngircd%2Fio.c;h=fb39ecdd99f0b07cb96c6afd61ef05a40468eaa3;hb=03628dbeaf40a9de34b3eb6d5bf6dd34eed8248c;hp=d759a5952462a0d6201e72bccce3632d59b9db7b;hpb=06bfb3adfb7ed6aef01c2fa892ea396405ee040a;p=ngircd.git

diff --git a/src/ngircd/io.c b/src/ngircd/io.c
index d759a595..fb39ecdd 100644
--- a/src/ngircd/io.c
+++ b/src/ngircd/io.c
@@ -5,14 +5,16 @@
  * (at your option) any later version.
  * Please read the file COPYING, README and AUTHORS for more information.
  *
- * I/O abstraction interface.
  * Copyright (c) 2005 Florian Westphal (westphal@foo.fh-furtwangen.de)
  *
  */
 
 #include "portab.h"
 
-static char UNUSED id[] = "$Id: io.c,v 1.26 2007/11/18 15:05:35 alex Exp $";
+/**
+ * @file
+ * I/O abstraction interface.
+ */
 
 #include <assert.h>
 #include <stdlib.h>
@@ -104,14 +106,18 @@ static bool io_event_change_devpoll(int fd, short what);
 
 #ifdef IO_USE_SELECT
 #include "defines.h"	/* for conn.h */
-#include "conn.h"	/* for CONN_IDX (needed by resolve.h) */
-#include "resolve.h"	/* for RES_STAT (needed by conf.h) */
+#include "proc.h"	/* for PROC_STAT (needed by conf.h) */
+#include "conn.h"	/* for CONN_ID (needed by conf.h) */
 #include "conf.h"	/* for Conf_MaxConnections */
 
 static fd_set readers;
 static fd_set writers;
-static int select_maxfd;		/* the select() interface sucks badly */
-static int io_dispatch_select(struct timeval *tv);
+/*
+ * this is the first argument for select(), i.e.
+ * the largest fd registered, plus one.
+ */
+static int select_maxfd;
+static int io_dispatch_select PARAMS((struct timeval *tv));
 
 #ifndef IO_USE_EPOLL
 #define io_masterfd -1
@@ -122,6 +128,18 @@ static array io_events;
 
 static void io_docallback PARAMS((int fd, short what));
 
+#ifdef DEBUG_IO
+static void
+io_debug(const char *s, int fd, int what)
+{
+	Log(LOG_DEBUG, "%s: %d, %d\n", s, fd, what);
+}
+#else
+static inline void
+io_debug(const char UNUSED *s,int UNUSED a, int UNUSED b)
+{ /* NOTHING */ }
+#endif
+
 static io_event *
 io_event_get(int fd)
 {
@@ -138,6 +156,72 @@ io_event_get(int fd)
 
 
 #ifdef IO_USE_DEVPOLL
+static int
+io_dispatch_devpoll(struct timeval *tv)
+{
+	struct dvpoll dvp;
+	time_t sec = tv->tv_sec * 1000;
+	int i, total, ret, timeout = tv->tv_usec + sec;
+	short what;
+	struct pollfd p[100];
+
+	if (timeout < 0)
+		timeout = 1000;
+
+	total = 0;
+	do {
+		dvp.dp_timeout = timeout;
+		dvp.dp_nfds = 100;
+		dvp.dp_fds = p;
+		ret = ioctl(io_masterfd, DP_POLL, &dvp);
+		total += ret;
+		if (ret <= 0)
+			return total;
+		for (i=0; i < ret ; i++) {
+			what = 0;
+			if (p[i].revents & (POLLIN|POLLPRI))
+				what = IO_WANTREAD;
+
+			if (p[i].revents & POLLOUT)
+				what |= IO_WANTWRITE;
+
+			if (p[i].revents && !what) {
+				/* other flag is set, probably POLLERR */
+				what = IO_ERROR;
+			}
+			io_docallback(p[i].fd, what);
+		}
+	} while (ret == 100);
+
+	return total;
+}
+
+
+static bool
+io_event_change_devpoll(int fd, short what)
+{
+	struct pollfd p;
+
+	p.events = 0;
+
+	if (what & IO_WANTREAD)
+		p.events = POLLIN | POLLPRI;
+	if (what & IO_WANTWRITE)
+		p.events |= POLLOUT;
+
+	p.fd = fd;
+	return write(io_masterfd, &p, sizeof p) == (ssize_t)sizeof p;
+}
+
+static void
+io_close_devpoll(int fd)
+{
+	struct pollfd p;
+	p.events = POLLREMOVE;
+	p.fd = fd;
+	write(io_masterfd, &p, sizeof p);
+}
+
 static void
 io_library_init_devpoll(unsigned int eventsize)
 {
@@ -147,10 +231,96 @@ io_library_init_devpoll(unsigned int eventsize)
 	Log(LOG_INFO, "IO subsystem: /dev/poll (initial maxfd %u, masterfd %d).",
 		eventsize, io_masterfd);
 }
+#else
+static inline void
+io_close_devpoll(int UNUSED x)
+{ /* NOTHING */ }
+static inline void
+io_library_init_devpoll(unsigned int UNUSED ev)
+{ /* NOTHING */ }
 #endif
 
 
 #ifdef IO_USE_POLL
+static int
+io_dispatch_poll(struct timeval *tv)
+{
+	time_t sec = tv->tv_sec * 1000;
+	int i, ret, timeout = tv->tv_usec + sec;
+	int fds_ready;
+	short what;
+	struct pollfd *p = array_start(&pollfds);
+
+	if (timeout < 0)
+		timeout = 1000;
+
+	ret = poll(p, poll_maxfd + 1, timeout);
+	if (ret <= 0)
+		return ret;
+
+	fds_ready = ret;
+	for (i=0; i <= poll_maxfd; i++) {
+		what = 0;
+		if (p[i].revents & (POLLIN|POLLPRI))
+			what = IO_WANTREAD;
+
+		if (p[i].revents & POLLOUT)
+			what |= IO_WANTWRITE;
+
+		if (p[i].revents && !what) {
+			/* other flag is set, probably POLLERR */
+			what = IO_ERROR;
+		}
+		if (what) {
+			fds_ready--;
+			io_docallback(i, what);
+		}
+		if (fds_ready <= 0)
+			break;
+	}
+
+	return ret;
+}
+
+static bool
+io_event_change_poll(int fd, short what)
+{
+	struct pollfd *p;
+	short events = 0;
+
+	if (what & IO_WANTREAD)
+		events = POLLIN | POLLPRI;
+	if (what & IO_WANTWRITE)
+		events |= POLLOUT;
+
+	p = array_alloc(&pollfds, sizeof *p, fd);
+	if (p) {
+		p->events = events;
+		p->fd = fd;
+		if (fd > poll_maxfd)
+			poll_maxfd = fd;
+	}
+	return p != NULL;
+}
+
+static void
+io_close_poll(int fd)
+{
+	struct pollfd *p;
+	p = array_get(&pollfds, sizeof *p, fd);
+	if (!p) return;
+
+	p->fd = -1;
+	if (fd == poll_maxfd) {
+		while (poll_maxfd > 0) {
+			--poll_maxfd;
+			p = array_get(&pollfds, sizeof *p, poll_maxfd);
+			if (p && p->fd >= 0)
+				break;
+		}
+	}
+}
+
 static void
 io_library_init_poll(unsigned int eventsize)
 {
@@ -169,13 +339,60 @@ io_library_init_poll(unsigned int eventsize)
 		library_initialized = true;
 	}
 }
+#else
+static inline void
+io_close_poll(int UNUSED x)
+{ /* NOTHING */ }
+static inline void
+io_library_init_poll(unsigned int UNUSED ev)
+{ /* NOTHING */ }
 #endif
 
 
 #ifdef IO_USE_SELECT
+static int
+io_dispatch_select(struct timeval *tv)
+{
+	fd_set readers_tmp;
+	fd_set writers_tmp;
+	short what;
+	int ret, i;
+	int fds_ready;
+
+	readers_tmp = readers;
+	writers_tmp = writers;
+
+	ret = select(select_maxfd + 1, &readers_tmp, &writers_tmp, NULL, tv);
+	if (ret <= 0)
+		return ret;
+
+	fds_ready = ret;
+
+	for (i = 0; i <= select_maxfd; i++) {
+		what = 0;
+		if (FD_ISSET(i, &readers_tmp)) {
+			what = IO_WANTREAD;
+			fds_ready--;
+		}
+
+		if (FD_ISSET(i, &writers_tmp)) {
+			what |= IO_WANTWRITE;
+			fds_ready--;
+		}
+		if (what)
+			io_docallback(i, what);
+		if (fds_ready <= 0)
+			break;
+	}
+
+	return ret;
+}
+
 static void
 io_library_init_select(unsigned int eventsize)
 {
+	if (library_initialized)
+		return;
 	Log(LOG_INFO, "IO subsystem: select (initial maxfd %u).",
 	    eventsize);
 	FD_ZERO(&readers);
@@ -188,222 +405,121 @@ io_library_init_select(unsigned int eventsize)
 
 		Conf_MaxConnections = FD_SETSIZE - 1;
 	}
+#else
+	Log(LOG_WARNING,
+	    "FD_SETSIZE undefined, don't know how many descriptors select() can handle on your platform ...");
 #endif /* FD_SETSIZE */
 	library_initialized = true;
 }
-#endif /* SELECT */
-
 
-#ifdef IO_USE_EPOLL
 static void
-io_library_init_epoll(unsigned int eventsize)
+io_close_select(int fd)
 {
-	int ecreate_hint = (int)eventsize;
-	if (ecreate_hint <= 0)
-		ecreate_hint = 128;
-	io_masterfd = epoll_create(ecreate_hint);
-	if (io_masterfd >= 0) {
-		library_initialized = true;
-		Log(LOG_INFO,
-		    "IO subsystem: epoll (hint size %d, initial maxfd %u, masterfd %d).",
-		    ecreate_hint, eventsize, io_masterfd);
-	}
-}
-#endif
+	io_event *i;
+
+	if (io_masterfd >= 0)	/* Are we using epoll()? */
+		return;
 
+	FD_CLR(fd, &writers);
+	FD_CLR(fd, &readers);
 
-#ifdef IO_USE_KQUEUE
-static void
-io_library_init_kqueue(unsigned int eventsize)
-{
-	io_masterfd = kqueue();
+	i = io_event_get(fd);
+	if (!i) return;
 
-	Log(LOG_INFO,
-	    "IO subsystem: kqueue (initial maxfd %u, masterfd %d)",
-	    eventsize, io_masterfd);
-	if (io_masterfd >= 0)
-		library_initialized = true;
+	if (fd == select_maxfd) {
+		while (select_maxfd>0) {
+			--select_maxfd; /* find largest fd */
+			i = io_event_get(select_maxfd);
+			if (i && i->callback) break;
+		}
+	}
 }
-#endif
+#else
+static inline void
+io_library_init_select(int UNUSED x)
+{ /* NOTHING */ }
+static inline void
+io_close_select(int UNUSED x)
+{ /* NOTHING */ }
+#endif /* SELECT */
 
 
-bool
-io_library_init(unsigned int eventsize)
-{
-	if (library_initialized)
-		return true;
-#ifdef IO_USE_SELECT
-#ifndef FD_SETSIZE
-	Log(LOG_WARNING,
-	    "FD_SETSIZE undefined, don't know how many descriptors select() can handle on your platform ...");
-#else
-	if (eventsize >= FD_SETSIZE)
-		eventsize = FD_SETSIZE - 1;
-#endif /* FD_SETSIZE */
-#endif /* IO_USE_SELECT */
-	if ((eventsize > 0) && !array_alloc(&io_events, sizeof(io_event), (size_t)eventsize))
-		eventsize = 0;
 #ifdef IO_USE_EPOLL
-	io_library_init_epoll(eventsize);
-#ifdef IO_USE_SELECT
-	if (io_masterfd < 0)
-		Log(LOG_INFO, "Can't initialize epoll() IO interface, falling back to select() ...");
-#endif
-#endif
-#ifdef IO_USE_KQUEUE
-	io_library_init_kqueue(eventsize);
-#endif
-#ifdef IO_USE_DEVPOLL
-	io_library_init_devpoll(eventsize);
-#endif
-#ifdef IO_USE_POLL
-	io_library_init_poll(eventsize);
-#endif
-#ifdef IO_USE_SELECT
-	if (! library_initialized)
-		io_library_init_select(eventsize);
-#endif
-	return library_initialized;
-}
-
-
-void
-io_library_shutdown(void)
+static bool
+io_event_change_epoll(int fd, short what, const int action)
 {
-#ifdef IO_USE_SELECT
-	FD_ZERO(&readers);
-	FD_ZERO(&writers);
-#endif
-#ifdef IO_USE_EPOLL
-	if (io_masterfd >= 0)
-		close(io_masterfd);
-	io_masterfd = -1;
-#endif
-#ifdef IO_USE_KQUEUE
-	close(io_masterfd);
-	io_masterfd = -1;
-	array_free(&io_evcache);
-#endif
-	library_initialized = false;
-}
-
+	struct epoll_event ev = { 0, {0} };
+	ev.data.fd = fd;
 
-bool
-io_event_setcb(int fd, void (*cbfunc) (int, short))
-{
-	io_event *i = io_event_get(fd);
-	if (!i)
-		return false;
+	if (what & IO_WANTREAD)
+		ev.events = EPOLLIN | EPOLLPRI;
+	if (what & IO_WANTWRITE)
+		ev.events |= EPOLLOUT;
 
-	i->callback = cbfunc;
-	return true;
+	return epoll_ctl(io_masterfd, action, fd, &ev) == 0;
 }
 
-
-bool
-io_event_create(int fd, short what, void (*cbfunc) (int, short))
+static int
+io_dispatch_epoll(struct timeval *tv)
 {
-	bool ret;
-	io_event *i;
-
-	assert(fd >= 0);
-#if defined(IO_USE_SELECT) && defined(FD_SETSIZE)
-	if (fd >= FD_SETSIZE) {
-		Log(LOG_ERR,
-		    "fd %d exceeds FD_SETSIZE (%u) (select can't handle more file descriptors)",
-		    fd, FD_SETSIZE);
-		return false;
-	}
-#endif
-	i = (io_event *) array_alloc(&io_events, sizeof(io_event), (size_t) fd);
-	if (!i) {
-		Log(LOG_WARNING,
-		    "array_alloc failed: could not allocate space for %d io_event structures",
-		    fd);
-		return false;
-	}
+	time_t sec = tv->tv_sec * 1000;
+	int i, total = 0, ret, timeout = tv->tv_usec + sec;
+	struct epoll_event epoll_ev[100];
+	short type;
 
-	i->callback = cbfunc;
-	i->what = 0;
-#ifdef IO_USE_DEVPOLL
-	ret = io_event_change_devpoll(fd, what);
-#endif
-#ifdef IO_USE_POLL
-	ret = io_event_change_poll(fd, what);
-#endif
-#ifdef IO_USE_EPOLL
-	ret = io_event_change_epoll(fd, what, EPOLL_CTL_ADD);
-#endif
-#ifdef IO_USE_KQUEUE
-	ret = io_event_change_kqueue(fd, what, EV_ADD|EV_ENABLE);
-#endif
-#ifdef IO_USE_SELECT
-	if (io_masterfd < 0)
-		ret = io_event_add(fd, what);
-#endif
-	if (ret) i->what = what;
-	return ret;
-}
+	if (timeout < 0)
+		timeout = 1000;
 
+	do {
+		ret = epoll_wait(io_masterfd, epoll_ev, 100, timeout);
+		total += ret;
+		if (ret <= 0)
+			return total;
 
-#ifdef IO_USE_DEVPOLL
-static bool
-io_event_change_devpoll(int fd, short what)
-{
-	struct pollfd p;
+		for (i = 0; i < ret; i++) {
+			type = 0;
+			if (epoll_ev[i].events & (EPOLLERR | EPOLLHUP))
+				type = IO_ERROR;
 
-	p.events = 0;
+			if (epoll_ev[i].events & (EPOLLIN | EPOLLPRI))
+				type |= IO_WANTREAD;
 
-	if (what & IO_WANTREAD)
-		p.events = POLLIN | POLLPRI;
-	if (what & IO_WANTWRITE)
-		p.events |= POLLOUT;
+			if (epoll_ev[i].events & EPOLLOUT)
+				type |= IO_WANTWRITE;
 
-	p.fd = fd;
-	return write(io_masterfd, &p, sizeof p) == (ssize_t)sizeof p;
-}
-#endif
+			io_docallback(epoll_ev[i].data.fd, type);
+		}
 
+		timeout = 0;
+	} while (ret == 100);
 
+	return total;
+}
 
-#ifdef IO_USE_POLL
-static bool
-io_event_change_poll(int fd, short what)
+static void
+io_library_init_epoll(unsigned int eventsize)
 {
-	struct pollfd *p;
-	short events = 0;
-
-	if (what & IO_WANTREAD)
-		events = POLLIN | POLLPRI;
-	if (what & IO_WANTWRITE)
-		events |= POLLOUT;
-
-	p = array_alloc(&pollfds, sizeof *p, fd);
-	if (p) {
-		p->events = events;
-		p->fd = fd;
-		if (fd > poll_maxfd)
-			poll_maxfd = fd;
+	int ecreate_hint = (int)eventsize;
+	if (ecreate_hint <= 0)
+		ecreate_hint = 128;
+	io_masterfd = epoll_create(ecreate_hint);
+	if (io_masterfd >= 0) {
+		library_initialized = true;
+		Log(LOG_INFO,
+		    "IO subsystem: epoll (hint size %d, initial maxfd %u, masterfd %d).",
+		    ecreate_hint, eventsize, io_masterfd);
+		return;
 	}
-	return p != NULL;
-}
+#ifdef IO_USE_SELECT
+	Log(LOG_INFO, "Can't initialize epoll() IO interface, falling back to select() ...");
 #endif
-
-#ifdef IO_USE_EPOLL
-static bool
-io_event_change_epoll(int fd, short what, const int action)
-{
-	struct epoll_event ev = { 0, {0} };
-	ev.data.fd = fd;
-
-	if (what & IO_WANTREAD)
-		ev.events = EPOLLIN | EPOLLPRI;
-	if (what & IO_WANTWRITE)
-		ev.events |= EPOLLOUT;
-
-	return epoll_ctl(io_masterfd, action, fd, &ev) == 0;
 }
-#endif
+#else
+static inline void
+io_library_init_epoll(unsigned int UNUSED ev)
+{ /* NOTHING */ }
+#endif /* IO_USE_EPOLL */
+
 
 #ifdef IO_USE_KQUEUE
 static bool
@@ -433,7 +549,6 @@ io_event_kqueue_commit_cache(void)
 	return ret;
 }
 
-
 static bool
 io_event_change_kqueue(int fd, short what, const int action)
 {
@@ -458,139 +573,255 @@ io_event_change_kqueue(int fd, short what, const int action)
 		io_event_kqueue_commit_cache();
 	return ret;
 }
+
+static int
+io_dispatch_kqueue(struct timeval *tv)
+{
+	int i, total = 0, ret;
+	struct kevent kev[100];
+	struct kevent *newevents;
+	struct timespec ts;
+	int newevents_len;
+	ts.tv_sec = tv->tv_sec;
+	ts.tv_nsec = tv->tv_usec * 1000;
+
+	do {
+		newevents_len = (int) array_length(&io_evcache, sizeof (struct kevent));
+		newevents = (newevents_len > 0) ? array_start(&io_evcache) : NULL;
+		assert(newevents_len >= 0);
+
+		ret = kevent(io_masterfd, newevents, newevents_len, kev, 100, &ts);
+		if (newevents && ret != -1)
+			array_trunc(&io_evcache);
+
+		total += ret;
+		if (ret <= 0)
+			return total;
+
+		for (i = 0; i < ret; i++) {
+			io_debug("dispatch_kqueue: fd, kev.flags", (int)kev[i].ident, kev[i].flags);
+			if (kev[i].flags & (EV_EOF|EV_ERROR)) {
+				if (kev[i].flags & EV_ERROR)
+					Log(LOG_ERR, "kevent fd %d: EV_ERROR (%s)",
+						(int)kev[i].ident, strerror((int)kev[i].data));
+				io_docallback((int)kev[i].ident, IO_ERROR);
+				continue;
+			}
+
+			switch (kev[i].filter) {
+			case EVFILT_READ:
+				io_docallback((int)kev[i].ident, IO_WANTREAD);
+				break;
+			case EVFILT_WRITE:
+				io_docallback((int)kev[i].ident, IO_WANTWRITE);
+				break;
+			default:
+				LogDebug("Unknown kev.filter number %d for fd %d",
+					kev[i].filter, kev[i].ident);
+				/* Fall through */
+			case EV_ERROR:
+				io_docallback((int)kev[i].ident, IO_ERROR);
+				break;
+			}
+		}
+		ts.tv_sec = 0;
+		ts.tv_nsec = 0;
+	} while (ret == 100);
+
+	return total;
+}
+
+static void
+io_library_init_kqueue(unsigned int eventsize)
+{
+	io_masterfd = kqueue();
+
+	Log(LOG_INFO,
+	    "IO subsystem: kqueue (initial maxfd %u, masterfd %d)",
+	    eventsize, io_masterfd);
+	if (io_masterfd >= 0)
+		library_initialized = true;
+}
+#else
+static inline void
+io_library_init_kqueue(unsigned int UNUSED ev)
+{ /* NOTHING */ }
 #endif
 
 
 bool
-io_event_add(int fd, short what)
+io_library_init(unsigned int eventsize)
 {
-	io_event *i = io_event_get(fd);
+	if (library_initialized)
+		return true;
 
-	if (!i) return false;
+	if ((eventsize > 0) && !array_alloc(&io_events, sizeof(io_event), (size_t)eventsize))
+		eventsize = 0;
 
-	if ((i->what & what) == what) /* event type is already registered */
-		return true;
-#ifdef DEBUG_IO
-	Log(LOG_DEBUG, "io_event_add(): fd %d, what %d.", fd, what);
+	io_library_init_epoll(eventsize);
+	io_library_init_kqueue(eventsize);
+	io_library_init_devpoll(eventsize);
+	io_library_init_poll(eventsize);
+	io_library_init_select(eventsize);
+
+	return library_initialized;
+}
+
+
+void
+io_library_shutdown(void)
+{
+#ifdef IO_USE_SELECT
+	FD_ZERO(&readers);
+	FD_ZERO(&writers);
 #endif
-	i->what |= what;
-#ifdef IO_USE_EPOLL
+#if defined(IO_USE_EPOLL) || defined(IO_USE_KQUEUE) || defined(IO_USE_DEVPOLL)
 	if (io_masterfd >= 0)
-		return io_event_change_epoll(fd, i->what, EPOLL_CTL_MOD);
+		close(io_masterfd);
+	io_masterfd = -1;
 #endif
-
 #ifdef IO_USE_KQUEUE
-	return io_event_change_kqueue(fd, what, EV_ADD | EV_ENABLE);
+	array_free(&io_evcache);
 #endif
+	library_initialized = false;
+}
+
+
+bool
+io_event_setcb(int fd, void (*cbfunc) (int, short))
+{
+	io_event *i = io_event_get(fd);
+	if (!i)
+		return false;
+
+	i->callback = cbfunc;
+	return true;
+}
+
+
+static bool
+backend_create_ev(int fd, short what)
+{
+	bool ret;
 #ifdef IO_USE_DEVPOLL
-	return io_event_change_devpoll(fd, i->what);
+	ret = io_event_change_devpoll(fd, what);
 #endif
 #ifdef IO_USE_POLL
-	return io_event_change_poll(fd, i->what);
+	ret = io_event_change_poll(fd, what);
+#endif
+#ifdef IO_USE_EPOLL
+	ret = io_event_change_epoll(fd, what, EPOLL_CTL_ADD);
+#endif
+#ifdef IO_USE_KQUEUE
+	ret = io_event_change_kqueue(fd, what, EV_ADD|EV_ENABLE);
 #endif
 #ifdef IO_USE_SELECT
-	if (fd > select_maxfd)
-		select_maxfd = fd;
-
-	if (what & IO_WANTREAD)
-		FD_SET(fd, &readers);
-	if (what & IO_WANTWRITE)
-		FD_SET(fd, &writers);
-
-	return true;
+	if (io_masterfd < 0)
+		ret = io_event_add(fd, what);
 #endif
+	return ret;
 }
 
 
 bool
-io_setnonblock(int fd)
+io_event_create(int fd, short what, void (*cbfunc) (int, short))
 {
-	int flags = fcntl(fd, F_GETFL);
-	if (flags == -1)
-		return false;
+	bool ret;
+	io_event *i;
 
-#ifndef O_NONBLOCK
-#define O_NONBLOCK O_NDELAY
+	assert(fd >= 0);
+#if defined(IO_USE_SELECT) && defined(FD_SETSIZE)
+	if (io_masterfd < 0 && fd >= FD_SETSIZE) {
+		Log(LOG_ERR,
+		    "fd %d exceeds FD_SETSIZE (%u) (select can't handle more file descriptors)",
+		    fd, FD_SETSIZE);
+		return false;
+	}
 #endif
-	flags |= O_NONBLOCK;
+	i = (io_event *) array_alloc(&io_events, sizeof(io_event), (size_t) fd);
+	if (!i) {
+		Log(LOG_WARNING,
+		    "array_alloc failed: could not allocate space for %d io_event structures",
+		    fd);
+		return false;
+	}
 
-	return fcntl(fd, F_SETFL, flags) == 0;
+	i->callback = cbfunc;
+	i->what = 0;
+	ret = backend_create_ev(fd, what);
+	if (ret)
+		i->what = what;
+	return ret;
 }
 
 
-#ifdef IO_USE_DEVPOLL
-static void
-io_close_devpoll(int fd)
+bool
+io_event_add(int fd, short what)
 {
-	struct pollfd p;
-	p.events = POLLREMOVE;
-	p.fd = fd;
-	write(io_masterfd, &p, sizeof p);
-}
-#else
-static inline void
-io_close_devpoll(int UNUSED x)
-{ 
-	/* NOTHING */
-}
-#endif
+	io_event *i = io_event_get(fd);
 
+	if (!i) return false;
 
+	if ((i->what & what) == what) /* event type is already registered */
+		return true;
 
-#ifdef IO_USE_POLL
-static void
-io_close_poll(int fd)
-{
-	struct pollfd *p;
-	p = array_get(&pollfds, sizeof *p, fd);
-	if (!p) return;
+	io_debug("io_event_add: fd, what", fd, what);
 
-	p->fd = -1;
-	if (fd == poll_maxfd) {
-		while (poll_maxfd > 0) {
-			--poll_maxfd;
-			p = array_get(&pollfds, sizeof *p, poll_maxfd);
-			if (p && p->fd >= 0)
-				break;
-		}
-	}
-}
-#else
-static inline void io_close_poll(int UNUSED x) { /* NOTHING */ }
+	i->what |= what;
+#ifdef IO_USE_EPOLL
+	if (io_masterfd >= 0)
+		return io_event_change_epoll(fd, i->what, EPOLL_CTL_MOD);
+#endif
+#ifdef IO_USE_KQUEUE
+	return io_event_change_kqueue(fd, what, EV_ADD | EV_ENABLE);
+#endif
+#ifdef IO_USE_DEVPOLL
+	return io_event_change_devpoll(fd, i->what);
+#endif
+#ifdef IO_USE_POLL
+	return io_event_change_poll(fd, i->what);
 #endif
-
-
 #ifdef IO_USE_SELECT
-static void
-io_close_select(int fd)
-{
-	io_event *i;
+	if (fd > select_maxfd)
+		select_maxfd = fd;
+
+	if (what & IO_WANTREAD)
+		FD_SET(fd, &readers);
+	if (what & IO_WANTWRITE)
+		FD_SET(fd, &writers);
 
-	if (io_masterfd >= 0)	/* Are we using epoll()? */
-		return;
+	return true;
+#endif
+	return false;
+}
 
-	FD_CLR(fd, &writers);
-	FD_CLR(fd, &readers);
 
-	i = io_event_get(fd);
-	if (!i) return;
+bool
+io_setnonblock(int fd)
+{
+	int flags = fcntl(fd, F_GETFL);
+	if (flags == -1)
+		return false;
+#ifndef O_NONBLOCK
+#define O_NONBLOCK O_NDELAY
+#endif
+	flags |= O_NONBLOCK;
 
-	if (fd == select_maxfd) {
-		while (select_maxfd>0) {
-			--select_maxfd; /* find largest fd */
-			i = io_event_get(select_maxfd);
-			if (i && i->callback) break;
-		}
-	}
-}
-#else
-static inline void
-io_close_select(int UNUSED x)
-{ 
-	/* NOTHING */
+	return fcntl(fd, F_SETFL, flags) == 0;
 }
+
+bool
+io_setcloexec(int fd)
+{
+	int flags = fcntl(fd, F_GETFD);
+	if (flags == -1)
+		return false;
+#ifdef FD_CLOEXEC
+	flags |= FD_CLOEXEC;
 #endif
 
+	return fcntl(fd, F_SETFD, flags) == 0;
+}
 
 bool
 io_close(int fd)
@@ -610,11 +841,9 @@ io_close(int fd)
 		io_event_kqueue_commit_cache();
 	}
 #endif
-
 	io_close_devpoll(fd);
 	io_close_poll(fd);
 	io_close_select(fd);
-
 #ifdef IO_USE_EPOLL
 	io_event_change_epoll(fd, 0, EPOLL_CTL_DEL);
 #endif
@@ -630,13 +859,14 @@ bool
 io_event_del(int fd, short what)
 {
 	io_event *i = io_event_get(fd);
-#ifdef DEBUG_IO
-	Log(LOG_DEBUG, "io_event_del(): trying to delete eventtype %d on fd %d", what, fd);
-#endif
+
+	io_debug("io_event_del: trying to delete eventtype; fd, what", fd, what);
 	if (!i) return false;
 
-	i->what &= ~what;
+	if (!(i->what & what)) /* event is already disabled */
+		return true;
 
+	i->what &= ~what;
 #ifdef IO_USE_DEVPOLL
 	return io_event_change_devpoll(fd, i->what);
 #endif
@@ -647,7 +877,6 @@ io_event_del(int fd, short what)
 	if (io_masterfd >= 0)
 		return io_event_change_epoll(fd, i->what, EPOLL_CTL_MOD);
 #endif
-
 #ifdef IO_USE_KQUEUE
 	return io_event_change_kqueue(fd, what, EV_DISABLE);
 #endif
@@ -657,242 +886,12 @@ io_event_del(int fd, short what)
 
 	if (what & IO_WANTREAD)
 		FD_CLR(fd, &readers);
-
 	return true;
 #endif
+	return false;
 }
 
 
-#ifdef IO_USE_SELECT
-static int
-io_dispatch_select(struct timeval *tv)
-{
-	fd_set readers_tmp = readers;
-	fd_set writers_tmp = writers;
-	short what;
-	int ret, i;
-	int fds_ready;
-	ret = select(select_maxfd + 1, &readers_tmp, &writers_tmp, NULL, tv);
-	if (ret <= 0)
-		return ret;
-
-	fds_ready = ret;
-
-	for (i = 0; i <= select_maxfd; i++) {
-		what = 0;
-		if (FD_ISSET(i, &readers_tmp)) {
-			what = IO_WANTREAD;
-			fds_ready--;
-		}
-
-		if (FD_ISSET(i, &writers_tmp)) {
-			what |= IO_WANTWRITE;
-			fds_ready--;
-		}
-		if (what)
-			io_docallback(i, what);
-		if (fds_ready <= 0)
-			break;
-	}
-
-	return ret;
-}
-#endif
-
-
-#ifdef IO_USE_DEVPOLL
-static int
-io_dispatch_devpoll(struct timeval *tv)
-{
-	struct dvpoll dvp;
-	time_t sec = tv->tv_sec * 1000;
-	int i, total, ret, timeout = tv->tv_usec + sec;
-	short what;
-	struct pollfd p[100];
-
-	if (timeout < 0)
-		timeout = 1000;
-
-	total = 0;
-	do {
-		dvp.dp_timeout = timeout;
-		dvp.dp_nfds = 100;
-		dvp.dp_fds = p;
-		ret = ioctl(io_masterfd, DP_POLL, &dvp);
-		total += ret;
-		if (ret <= 0)
-			return total;
-		for (i=0; i < ret ; i++) {
-			what = 0;
-			if (p[i].revents & (POLLIN|POLLPRI))
-				what = IO_WANTREAD;
-
-			if (p[i].revents & POLLOUT)
-				what |= IO_WANTWRITE;
-
-			if (p[i].revents && !what) {
-				/* other flag is set, probably POLLERR */
-				what = IO_ERROR;
-			}
-			io_docallback(p[i].fd, what);
-		}
-	} while (ret == 100);
-
-	return total;
-}
-#endif
-
-
-#ifdef IO_USE_POLL
-static int
-io_dispatch_poll(struct timeval *tv)
-{
-	time_t sec = tv->tv_sec * 1000;
-	int i, ret, timeout = tv->tv_usec + sec;
-	int fds_ready;
-	short what;
-	struct pollfd *p = array_start(&pollfds);
-
-	if (timeout < 0)
-		timeout = 1000;
-
-	ret = poll(p, poll_maxfd + 1, timeout);
-	if (ret <= 0)
-		return ret;
-
-	fds_ready = ret;
-	for (i=0; i <= poll_maxfd; i++) {
-		what = 0;
-		if (p[i].revents & (POLLIN|POLLPRI))
-			what = IO_WANTREAD;
-
-		if (p[i].revents & POLLOUT)
-			what |= IO_WANTWRITE;
-
-		if (p[i].revents && !what) {
-			/* other flag is set, probably POLLERR */
-			what = IO_ERROR;
-		}
-		if (what) {
-			fds_ready--;
-			io_docallback(i, what);
-		}
-		if (fds_ready <= 0)
-			break;
-	}
-
-	return ret;
-}
-#endif
-
-
-#ifdef IO_USE_EPOLL
-static int
-io_dispatch_epoll(struct timeval *tv)
-{
-	time_t sec = tv->tv_sec * 1000;
-	int i, total = 0, ret, timeout = tv->tv_usec + sec;
-	struct epoll_event epoll_ev[100];
-	short type;
-
-	if (timeout < 0)
-		timeout = 1000;
-
-	do {
-		ret = epoll_wait(io_masterfd, epoll_ev, 100, timeout);
-		total += ret;
-		if (ret <= 0)
-			return total;
-
-		for (i = 0; i < ret; i++) {
-			type = 0;
-			if (epoll_ev[i].events & (EPOLLERR | EPOLLHUP))
-				type = IO_ERROR;
-
-			if (epoll_ev[i].events & (EPOLLIN | EPOLLPRI))
-				type |= IO_WANTREAD;
-
-			if (epoll_ev[i].events & EPOLLOUT)
-				type |= IO_WANTWRITE;
-
-			io_docallback(epoll_ev[i].data.fd, type);
-		}
-
-		timeout = 0;
-	} while (ret == 100);
-
-	return total;
-}
-#endif
-
-
-#ifdef IO_USE_KQUEUE
-static int
-io_dispatch_kqueue(struct timeval *tv)
-{
-	int i, total = 0, ret;
-	struct kevent kev[100];
-	struct kevent *newevents;
-	struct timespec ts;
-	int newevents_len;
-	ts.tv_sec = tv->tv_sec;
-	ts.tv_nsec = tv->tv_usec * 1000;
-
-	do {
-		newevents_len = (int) array_length(&io_evcache, sizeof (struct kevent));
-		newevents = (newevents_len > 0) ? array_start(&io_evcache) : NULL;
-		assert(newevents_len >= 0);
-		if (newevents_len < 0)
-			newevents_len = 0;
-#ifdef DEBUG
-		if (newevents_len)
-			assert(newevents != NULL);
-#endif
-		ret = kevent(io_masterfd, newevents, newevents_len, kev,
-			     100, &ts);
-		if ((newevents_len>0) && ret != -1)
-			array_trunc(&io_evcache);
-
-		total += ret;
-		if (ret <= 0)
-			return total;
-
-		for (i = 0; i < ret; i++) {
-			if (kev[i].flags & EV_EOF) {
-#ifdef DEBUG
-				LogDebug("kev.flag has EV_EOF set, setting IO_ERROR",
-					kev[i].filter, kev[i].ident);
-#endif
-				io_docallback((int)kev[i].ident, IO_ERROR);
-				continue;
-			}
-
-			switch (kev[i].filter) {
-				case EVFILT_READ:
-					io_docallback((int)kev[i].ident, IO_WANTREAD);
-					break;
-				case EVFILT_WRITE:
-					io_docallback((int)kev[i].ident, IO_WANTWRITE);
-					break;
-				default:
-#ifdef DEBUG
-					LogDebug("Unknown kev.filter number %d for fd %d",
-						kev[i].filter, kev[i].ident); /* Fall through */
-#endif
-				case EV_ERROR:
-					io_docallback((int)kev[i].ident, IO_ERROR);
-					break;
-			}
-		}
-		ts.tv_sec = 0;
-		ts.tv_nsec = 0;
-	} while (ret == 100);
-
-	return total;
-}
-#endif
-
-
 int
 io_dispatch(struct timeval *tv)
 {
@@ -912,6 +911,7 @@ io_dispatch(struct timeval *tv)
 #ifdef IO_USE_POLL
 	return io_dispatch_poll(tv);
 #endif
+	return -1;
 }
 
 
@@ -919,11 +919,9 @@ io_dispatch(struct timeval *tv)
 static void
 io_docallback(int fd, short what)
 {
-	io_event *i;
-#ifdef DEBUG_IO
-	Log(LOG_DEBUG, "doing callback for fd %d, what %d", fd, what);
-#endif
-	i = io_event_get(fd);
+	io_event *i = io_event_get(fd);
+
+	io_debug("io_docallback; fd, what", fd, what);
 
 	if (i->callback) {	/* callback might be NULL if a previous callback function
 				   called io_close on this fd */