* (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 1.4 2005/07/14 09:20:39 alex Exp $";
+/**
+ * @file
+ * I/O abstraction interface.
+ */
#include <assert.h>
#include <stdlib.h>
#include "io.h"
#include "log.h"
+/* Enables extra debug messages in event add/delete/callback code. */
+/* #define DEBUG_IO */
typedef struct {
+#ifdef PROTOTYPES
void (*callback)(int, short);
- int fd;
+#else
+ void (*callback)();
+#endif
short what;
} io_event;
-#define INIT_IOEVENT { NULL, -1, 0, NULL }
-#define IO_ERROR 4
+#define INIT_IOEVENT { NULL, -1, 0, NULL }
+#define IO_ERROR 4
#ifdef HAVE_EPOLL_CREATE
-#define IO_USE_EPOLL 1
+# define IO_USE_EPOLL 1
+# ifdef HAVE_SELECT
+# define IO_USE_SELECT 1
+# endif
#else
-# ifdef HAVE_KQUEUE
-#define IO_USE_KQUEUE 1
-# else
-#define IO_USE_SELECT 1
-#endif
-#endif
-
+# ifdef HAVE_KQUEUE
+# define IO_USE_KQUEUE 1
+# else
+# ifdef HAVE_SYS_DEVPOLL_H
+# define IO_USE_DEVPOLL 1
+# else
+# ifdef HAVE_POLL
+# define IO_USE_POLL 1
+# else
+# ifdef HAVE_SELECT
+# define IO_USE_SELECT 1
+# else
+# error "no IO API available!?"
+# endif /* HAVE_SELECT */
+# endif /* HAVE_POLL */
+# endif /* HAVE_SYS_DEVPOLL_H */
+# endif /* HAVE_KQUEUE */
+#endif /* HAVE_EPOLL_CREATE */
+
+static bool library_initialized = false;
#ifdef IO_USE_EPOLL
#include <sys/epoll.h>
-static int io_masterfd;
-static bool io_event_new_epoll(int fd, short what);
-static bool io_event_change_epoll(int fd, short what);
+static int io_masterfd = -1;
+static bool io_event_change_epoll(int fd, short what, const int action);
static int io_dispatch_epoll(struct timeval *tv);
#endif
static int io_masterfd;
static int io_dispatch_kqueue(struct timeval *tv);
-static bool io_event_add_kqueue(int, short);
+static bool io_event_change_kqueue(int, short, const int action);
+#endif
+
+#ifdef IO_USE_POLL
+#include <poll.h>
+
+static array pollfds;
+static int poll_maxfd;
+
+static bool io_event_change_poll PARAMS((int fd, short what));
+#endif
+
+#ifdef IO_USE_DEVPOLL
+#include <sys/devpoll.h>
+static int io_masterfd;
+
+static bool io_event_change_devpoll(int fd, short what);
#endif
#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
#endif
+#endif /* IO_USE_SELECT */
static array io_events;
static void io_docallback PARAMS((int fd, short what));
-static io_event *
-io_event_get(int fd)
+#ifdef DEBUG_IO
+static void
+io_debug(const char *s, int fd, int what)
{
- io_event *i;
- assert(fd >= 0);
- i = (io_event *) array_get(&io_events, sizeof(io_event), fd);
- assert(i);
-
- return i;
+ Log(LOG_DEBUG, "%s: %d, %d\n", s, fd, what);
}
-
-
-bool
-io_library_init(unsigned int eventsize)
-{
-#ifdef IO_USE_EPOLL
- int ecreate_hint = (int)eventsize;
- if (ecreate_hint <= 0)
- ecreate_hint = 128;
-#endif
-
-#ifdef IO_USE_SELECT
-#ifdef FD_SETSIZE
- if (eventsize >= FD_SETSIZE)
- eventsize = FD_SETSIZE - 1;
-#endif
-#endif
- if (eventsize && !array_alloc(&io_events, sizeof(io_event), eventsize))
- eventsize = 0;
-
-#ifdef IO_USE_EPOLL
- io_masterfd = epoll_create(ecreate_hint);
- Log(LOG_INFO,
- "IO subsystem: epoll (hint size %d, initial maxfd %u, masterfd %d).",
- ecreate_hint, eventsize, io_masterfd);
- return io_masterfd >= 0;
-#endif
-#ifdef IO_USE_SELECT
- Log(LOG_INFO, "IO subsystem: select (initial maxfd %u).",
- eventsize);
- FD_ZERO(&readers);
- FD_ZERO(&writers);
-#ifdef FD_SETSIZE
- if (Conf_MaxConnections >= FD_SETSIZE) {
- Log(LOG_WARNING,
- "MaxConnections (%d) exceeds limit (%u), changed MaxConnections to %u.",
- Conf_MaxConnections, FD_SETSIZE, FD_SETSIZE - 1);
-
- 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
- return true;
+static inline void
+io_debug(const char UNUSED *s,int UNUSED a, int UNUSED b)
+{ /* NOTHING */ }
#endif
-#ifdef IO_USE_KQUEUE
- io_masterfd = kqueue();
-
- Log(LOG_INFO,
- "IO subsystem: kqueue (initial maxfd %u, masterfd %d)",
- eventsize, io_masterfd);
- return io_masterfd >= 0;
-#endif
-}
-
-void
-io_library_shutdown(void)
+static io_event *
+io_event_get(int fd)
{
-#ifdef IO_USE_SELECT
- FD_ZERO(&readers);
- FD_ZERO(&writers);
-#else
- close(io_masterfd); /* kqueue, epoll */
- io_masterfd = -1;
-#endif
-#ifdef IO_USE_KQUEUE
- array_free(&io_evcache);
-#endif
-}
+ io_event *i;
+ assert(fd >= 0);
-bool
-io_event_setcb(int fd, void (*cbfunc) (int, short))
-{
- io_event *i = io_event_get(fd);
- if (!i)
- return false;
+ i = (io_event *) array_get(&io_events, sizeof(io_event), (size_t) fd);
- i->callback = cbfunc;
- return true;
+ assert(i != NULL);
+
+ return i;
}
-bool
-io_event_create(int fd, short what, void (*cbfunc) (int, short))
+#ifdef IO_USE_DEVPOLL
+static int
+io_dispatch_devpoll(struct timeval *tv)
{
- io_event *i;
-
- assert(fd >= 0);
+ 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];
-#ifdef IO_USE_SELECT
-#ifdef 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 /* FD_SETSIZE */
-#endif /* IO_USE_SELECT */
+ if (timeout < 0)
+ timeout = 1000;
- i = (io_event *) array_alloc(&io_events, sizeof(io_event), fd);
- if (!i) {
- Log(LOG_WARNING,
- "array_alloc failed: could not allocate space for %d io_event structures",
- fd);
- return false;
- }
+ 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);
- i->fd = fd;
- i->callback = cbfunc;
-#ifdef IO_USE_EPOLL
- i->what = what;
- return io_event_new_epoll(fd, what);
-#endif
-#ifdef IO_USE_KQUEUE
- i->what = what;
- return io_event_add_kqueue(fd, what);
-#endif
-#ifdef IO_USE_SELECT
- i->what = 0;
- return io_event_add(fd, what);
-#endif
+ return total;
}
-#ifdef IO_USE_EPOLL
static bool
-io_event_new_epoll(int fd, short what)
+io_event_change_devpoll(int fd, short what)
{
- struct epoll_event ev = { 0, {0} };
- ev.data.fd = fd;
+ struct pollfd p;
+
+ p.events = 0;
if (what & IO_WANTREAD)
- ev.events = EPOLLIN | EPOLLPRI;
+ p.events = POLLIN | POLLPRI;
if (what & IO_WANTWRITE)
- ev.events |= EPOLLOUT;
+ p.events |= POLLOUT;
- return epoll_ctl(io_masterfd, EPOLL_CTL_ADD, fd, &ev) == 0;
+ p.fd = fd;
+ return write(io_masterfd, &p, sizeof p) == (ssize_t)sizeof p;
}
-
-static bool
-io_event_change_epoll(int fd, short what)
+static void
+io_close_devpoll(int fd)
{
- 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;
+ struct pollfd p;
+ p.events = POLLREMOVE;
+ p.fd = fd;
+ write(io_masterfd, &p, sizeof p);
+}
- return epoll_ctl(io_masterfd, EPOLL_CTL_MOD, fd, &ev) == 0;
+static void
+io_library_init_devpoll(unsigned int eventsize)
+{
+ io_masterfd = open("/dev/poll", O_RDWR);
+ if (io_masterfd >= 0)
+ library_initialized = true;
+ 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_KQUEUE
-static bool
-io_event_kqueue_commit_cache(void)
+
+#ifdef IO_USE_POLL
+static int
+io_dispatch_poll(struct timeval *tv)
{
- struct kevent *events;
- bool ret;
- int len = (int) array_length(&io_evcache, sizeof (struct kevent));
-
- if (!len) /* nothing to do */
- return true;
+ 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);
- assert(len>0);
+ if (timeout < 0)
+ timeout = 1000;
- if (len < 0) {
- array_free(&io_evcache);
- return false;
- }
+ ret = poll(p, poll_maxfd + 1, timeout);
+ if (ret <= 0)
+ return ret;
- events = array_start(&io_evcache);
+ fds_ready = ret;
+ for (i=0; i <= poll_maxfd; i++) {
+ what = 0;
+ if (p[i].revents & (POLLIN|POLLPRI))
+ what = IO_WANTREAD;
- assert(events);
+ 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;
+ }
- ret = kevent(io_masterfd, events, len, NULL, 0, NULL) == 0;
- if (ret)
- array_trunc(&io_evcache);
return ret;
}
-
static bool
-io_event_add_kqueue(int fd, short what)
+io_event_change_poll(int fd, short what)
{
- struct kevent kev;
- short filter = 0;
- unsigned int len = array_length(&io_evcache, sizeof kev);
- bool ret;
+ struct pollfd *p;
+ short events = 0;
if (what & IO_WANTREAD)
- filter = EVFILT_READ;
-
+ events = POLLIN | POLLPRI;
if (what & IO_WANTWRITE)
- filter |= EVFILT_WRITE;
+ 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;
+}
- if (len >= 100) {
- ret = io_event_kqueue_commit_cache();
- if (ret)
- array_trunc(&io_evcache);
+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;
+ }
}
+}
- EV_SET(&kev, fd, filter, EV_ADD | EV_ENABLE, 0, 0, NULL);
- return array_catb(&io_evcache, (char*) &kev, sizeof (kev));
+static void
+io_library_init_poll(unsigned int eventsize)
+{
+ struct pollfd *p;
+ array_init(&pollfds);
+ poll_maxfd = 0;
+ Log(LOG_INFO, "IO subsystem: poll (initial maxfd %u).",
+ eventsize);
+ p = array_alloc(&pollfds, sizeof(struct pollfd), eventsize);
+ if (p) {
+ unsigned i;
+ p = array_start(&pollfds);
+ for (i = 0; i < eventsize; i++)
+ p[i].fd = -1;
+
+ 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
-bool
-io_event_add(int fd, short what)
+#ifdef IO_USE_SELECT
+static int
+io_dispatch_select(struct timeval *tv)
{
- io_event *i = io_event_get(fd);
+ fd_set readers_tmp;
+ fd_set writers_tmp;
+ short what;
+ int ret, i;
+ int fds_ready;
- assert(i);
+ readers_tmp = readers;
+ writers_tmp = writers;
- if (!i)
- return false;
- if (i->what == what)
- return true;
-#ifdef DEBUG
- Log(LOG_DEBUG, "io_event_add(): fd %d (arg: %d), what %d.", i->fd, fd, what);
-#endif
+ ret = select(select_maxfd + 1, &readers_tmp, &writers_tmp, NULL, tv);
+ if (ret <= 0)
+ return ret;
- i->what |= what;
+ fds_ready = ret;
-#ifdef IO_USE_EPOLL
- return io_event_change_epoll(fd, i->what);
-#endif
+ for (i = 0; i <= select_maxfd; i++) {
+ what = 0;
+ if (FD_ISSET(i, &readers_tmp)) {
+ what = IO_WANTREAD;
+ fds_ready--;
+ }
-#ifdef IO_USE_KQUEUE
- return io_event_add_kqueue(fd, what);
-#endif
+ if (FD_ISSET(i, &writers_tmp)) {
+ what |= IO_WANTWRITE;
+ fds_ready--;
+ }
+ if (what)
+ io_docallback(i, what);
+ if (fds_ready <= 0)
+ break;
+ }
-#ifdef IO_USE_SELECT
- if (fd > select_maxfd)
- select_maxfd = fd;
+ return ret;
+}
- if (what & IO_WANTREAD)
- FD_SET(fd, &readers);
- if (what & IO_WANTWRITE)
- FD_SET(fd, &writers);
+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);
+ FD_ZERO(&writers);
+#ifdef FD_SETSIZE
+ if (Conf_MaxConnections >= (int)FD_SETSIZE) {
+ Log(LOG_WARNING,
+ "MaxConnections (%d) exceeds limit (%u), changed MaxConnections to %u.",
+ Conf_MaxConnections, FD_SETSIZE, FD_SETSIZE - 1);
- return true;
-#endif
+ 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;
}
-
-bool
-io_setnonblock(int fd)
+static void
+io_close_select(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;
-
- return fcntl(fd, F_SETFL, flags) == 0;
-}
+ io_event *i;
+ if (io_masterfd >= 0) /* Are we using epoll()? */
+ return;
-bool
-io_close(int fd)
-{
- io_event *i = io_event_get(fd);
- if (i) {
- memset(i, 0, sizeof(io_event));
- i->fd = -1;
- }
-#ifdef IO_USE_SELECT
FD_CLR(fd, &writers);
FD_CLR(fd, &readers);
- if (fd == select_maxfd)
- select_maxfd--;
-#endif
-#ifdef IO_USE_KQUEUE
- if (array_length(&io_evcache, sizeof (struct kevent))) /* pending data in cache? */
- io_event_kqueue_commit_cache();
-#endif
- return close(fd) == 0; /* both epoll an kqueue will remove fd from all sets automatically */
-}
-
+ i = io_event_get(fd);
+ if (!i) return;
-bool
-io_event_del(int fd, short what)
-{
-#ifdef IO_USE_KQUEUE
- struct kevent kev;
- short filter = 0;
-#endif
- io_event *i = io_event_get(fd);
-#ifdef DEBUG
- Log(LOG_DEBUG, "io_event_del(): trying to delete eventtype %d on fd %d", what, fd);
-#endif
- assert(i);
- if (!i)
- return false;
+ 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_library_init_select(int UNUSED x)
+{ /* NOTHING */ }
+static inline void
+io_close_select(int UNUSED x)
+{ /* NOTHING */ }
+#endif /* SELECT */
- i->what &= ~what;
#ifdef IO_USE_EPOLL
- return io_event_change_epoll(fd, i->what);
-#endif
+static bool
+io_event_change_epoll(int fd, short what, const int action)
+{
+ struct epoll_event ev = { 0, {0} };
+ ev.data.fd = fd;
-#ifdef IO_USE_KQUEUE
if (what & IO_WANTREAD)
- filter = EVFILT_READ;
-
- if (what & IO_WANTWRITE)
- filter |= EVFILT_WRITE;
-
- EV_SET(&kev, fd, filter, EV_DELETE, 0, 0, NULL);
- return kevent(io_masterfd, &kev, 1, NULL, 0, NULL) == 0;
-#endif
-
-#ifdef IO_USE_SELECT
+ ev.events = EPOLLIN | EPOLLPRI;
if (what & IO_WANTWRITE)
- FD_CLR(i->fd, &writers);
-
- if (what & IO_WANTREAD)
- FD_CLR(i->fd, &readers);
-
- return true;
-#endif
-}
-
-
-#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;
- }
+ ev.events |= EPOLLOUT;
- return ret;
+ return epoll_ctl(io_masterfd, action, fd, &ev) == 0;
}
-#endif
-
-#ifdef IO_USE_EPOLL
static int
io_dispatch_epoll(struct timeval *tv)
{
return total;
}
+
+static void
+io_library_init_epoll(unsigned int eventsize)
+{
+ 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;
+ }
+#ifdef IO_USE_SELECT
+ Log(LOG_INFO, "Can't initialize epoll() IO interface, falling back to select() ...");
#endif
+}
+#else
+static inline void
+io_library_init_epoll(unsigned int UNUSED ev)
+{ /* NOTHING */ }
+#endif /* IO_USE_EPOLL */
#ifdef IO_USE_KQUEUE
+static bool
+io_event_kqueue_commit_cache(void)
+{
+ struct kevent *events;
+ bool ret;
+ int len = (int) array_length(&io_evcache, sizeof (struct kevent));
+
+ if (!len) /* nothing to do */
+ return true;
+
+ assert(len>0);
+
+ if (len < 0) {
+ array_free(&io_evcache);
+ return false;
+ }
+
+ events = array_start(&io_evcache);
+
+ assert(events != NULL);
+
+ ret = kevent(io_masterfd, events, len, NULL, 0, NULL) == 0;
+ if (ret)
+ array_trunc(&io_evcache);
+ return ret;
+}
+
+static bool
+io_event_change_kqueue(int fd, short what, const int action)
+{
+ struct kevent kev;
+ bool ret = true;
+
+ if (what & IO_WANTREAD) {
+ EV_SET(&kev, fd, EVFILT_READ, action, 0, 0, 0);
+ ret = array_catb(&io_evcache, (char*) &kev, sizeof (kev));
+ if (!ret)
+ ret = kevent(io_masterfd, &kev,1, NULL, 0, NULL) == 0;
+ }
+
+ if (ret && (what & IO_WANTWRITE)) {
+ EV_SET(&kev, fd, EVFILT_WRITE, action, 0, 0, 0);
+ ret = array_catb(&io_evcache, (char*) &kev, sizeof (kev));
+ if (!ret)
+ ret = kevent(io_masterfd, &kev, 1, NULL, 0, NULL) == 0;
+ }
+
+ if (array_length(&io_evcache, sizeof kev) >= 100)
+ 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 kevents *newevents;
+ struct kevent *newevents;
struct timespec ts;
int newevents_len;
- short type;
ts.tv_sec = tv->tv_sec;
ts.tv_nsec = tv->tv_usec * 1000;
-
+
do {
- newevents_len = array_length(&io_evcache, sizeof (struct kevent));
+ 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);
-#endif
- ret = kevent(io_masterfd, newevents, newevents_len, kev,
- 100, &ts);
- if ((newevents_len>0) && ret != -1)
+ ret = kevent(io_masterfd, newevents, newevents_len, kev, 100, &ts);
+ if (newevents && ret != -1)
array_trunc(&io_evcache);
total += ret;
return total;
for (i = 0; i < ret; i++) {
- type = 0;
- if (kev[i].flags & EV_EOF)
- type = IO_ERROR;
+ 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);
- if (kev[i].filter & EV_ERROR)
- type = IO_ERROR;
+ return total;
+}
- if (kev[i].filter & EVFILT_READ)
- type |= IO_WANTREAD;
+static void
+io_library_init_kqueue(unsigned int eventsize)
+{
+ io_masterfd = kqueue();
- if (kev[i].filter & EVFILT_WRITE)
- type |= IO_WANTWRITE;
+ 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
- io_docallback(kev[i].ident, type);
- }
- ts.tv_sec = 0;
- ts.tv_nsec = 0;
+bool
+io_library_init(unsigned int eventsize)
+{
+ if (library_initialized)
+ return true;
- } while (ret == 100);
+ if ((eventsize > 0) && !array_alloc(&io_events, sizeof(io_event), (size_t)eventsize))
+ eventsize = 0;
- return total;
+ 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
+#if defined(IO_USE_EPOLL) || defined(IO_USE_KQUEUE) || defined(IO_USE_DEVPOLL)
+ if (io_masterfd >= 0)
+ close(io_masterfd);
+ io_masterfd = -1;
+#endif
+#ifdef IO_USE_KQUEUE
+ 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
+ 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
+ return ret;
+}
+
+
+bool
+io_event_create(int fd, short what, void (*cbfunc) (int, short))
+{
+ bool ret;
+ io_event *i;
+
+ 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
+ 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;
+ }
+
+ i->callback = cbfunc;
+ i->what = 0;
+ ret = backend_create_ev(fd, what);
+ if (ret)
+ i->what = what;
+ return ret;
+}
+
+
+bool
+io_event_add(int fd, short what)
+{
+ io_event *i = io_event_get(fd);
+
+ if (!i) return false;
+
+ if ((i->what & what) == what) /* event type is already registered */
+ return true;
+
+ io_debug("io_event_add: fd, what", fd, what);
+
+ 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
+ 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;
+#endif
+ return false;
+}
+
+
+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;
+
+ 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)
+{
+ io_event *i;
+
+ i = io_event_get(fd);
+#ifdef IO_USE_KQUEUE
+ if (array_length(&io_evcache, sizeof (struct kevent))) /* pending data in cache? */
+ io_event_kqueue_commit_cache();
+
+ /* both kqueue and epoll remove fd from all sets automatically on the last close
+ * of the descriptor. since we don't know if this is the last close we'll have
+ * to remove the set explicitly. */
+ if (i) {
+ io_event_change_kqueue(fd, i->what, EV_DELETE);
+ 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
+ if (i) {
+ i->callback = NULL;
+ i->what = 0;
+ }
+ return close(fd) == 0;
}
+
+
+bool
+io_event_del(int fd, short what)
+{
+ io_event *i = io_event_get(fd);
+
+ io_debug("io_event_del: trying to delete eventtype; fd, what", fd, what);
+ if (!i) return false;
+
+ 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
+#ifdef IO_USE_POLL
+ return io_event_change_poll(fd, i->what);
+#endif
+#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_DISABLE);
#endif
+#ifdef IO_USE_SELECT
+ if (what & IO_WANTWRITE)
+ FD_CLR(fd, &writers);
+
+ if (what & IO_WANTREAD)
+ FD_CLR(fd, &readers);
+ return true;
+#endif
+ return false;
+}
int
io_dispatch(struct timeval *tv)
{
+#ifdef IO_USE_EPOLL
+ if (io_masterfd >= 0)
+ return io_dispatch_epoll(tv);
+#endif
#ifdef IO_USE_SELECT
return io_dispatch_select(tv);
#endif
#ifdef IO_USE_KQUEUE
return io_dispatch_kqueue(tv);
#endif
-#ifdef IO_USE_EPOLL
- return io_dispatch_epoll(tv);
+#ifdef IO_USE_DEVPOLL
+ return io_dispatch_devpoll(tv);
+#endif
+#ifdef IO_USE_POLL
+ return io_dispatch_poll(tv);
#endif
+ return -1;
}
static void
io_docallback(int fd, short what)
{
- io_event *i;
-#ifdef DEBUG
- Log(LOG_DEBUG, "doing callback for fd %d, what %d", fd, what);
-#endif
- i = io_event_get(fd);
- assert(i);
+ io_event *i = io_event_get(fd);
- if (i->callback) /* callback might be 0 if previous callback function called io_close on this 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 */
i->callback(fd, (what & IO_ERROR) ? i->what : what);
- /* if error indicator is set, we return the event(s) the app asked for */
+ }
+ /* if error indicator is set, we return the event(s) that were registered */
}