static int
devpoll_dispatch(struct event_base *base, struct timeval *tv)
{
struct devpollop *devpollop = base->evbase;
struct pollfd *events = devpollop->events;
struct dvpoll dvp;
int i, res, timeout = -1;
if (devpollop->nchanges)
devpoll_commit(devpollop);
if (tv != NULL)
timeout = tv->tv_sec * 1000 + (tv->tv_usec + 999) / 1000;
dvp.dp_fds = devpollop->events;
dvp.dp_nfds = devpollop->nevents;
dvp.dp_timeout = timeout;
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = ioctl(devpollop->dpfd, DP_POLL, &dvp);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
if (res == -1) {
if (errno != EINTR) {
event_warn("ioctl: DP_POLL");
return (-1);
}
return (0);
}
event_debug(("%s: devpoll_wait reports %d", __func__, res));
for (i = 0; i < res; i++) {
int which = 0;
int what = events[i].revents;
if (what & POLLHUP)
what |= POLLIN | POLLOUT;
else if (what & POLLERR)
what |= POLLIN | POLLOUT;
if (what & POLLIN)
which |= EV_READ;
if (what & POLLOUT)
which |= EV_WRITE;
if (!which)
continue;
/* XXX(niels): not sure if this works for devpoll */
evmap_io_active_(base, events[i].fd, which);
}
return (0);
}
static int
kq_dispatch(struct event_base *base, struct timeval *tv)
{
struct kqop *kqop = base->evbase;
struct kevent *events = kqop->events;
struct kevent *changes;
struct timespec ts, *ts_p = NULL;
int i, n_changes, res;
if (tv != NULL) {
TIMEVAL_TO_TIMESPEC(tv, &ts);
ts_p = &ts;
}
/* Build "changes" from "base->changes" */
EVUTIL_ASSERT(kqop->changes);
n_changes = kq_build_changes_list(&base->changelist, kqop);
if (n_changes < 0)
return -1;
event_changelist_remove_all_(&base->changelist, base);
/* steal the changes array in case some broken code tries to call
* dispatch twice at once. */
changes = kqop->changes;
kqop->changes = NULL;
/* Make sure that 'events' is at least as long as the list of changes:
* otherwise errors in the changes can get reported as a -1 return
* value from kevent() rather than as EV_ERROR events in the events
* array.
*
* (We could instead handle -1 return values from kevent() by
* retrying with a smaller changes array or a larger events array,
* but this approach seems less risky for now.)
*/
if (kqop->events_size < n_changes) {
int new_size = kqop->events_size;
do {
new_size *= 2;
} while (new_size < n_changes);
kq_grow_events(kqop, new_size);
events = kqop->events;
}
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = kevent(kqop->kq, changes, n_changes,
events, kqop->events_size, ts_p);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
EVUTIL_ASSERT(kqop->changes == NULL);
kqop->changes = changes;
if (res == -1) {
if (errno != EINTR) {
event_warn("kevent");
return (-1);
}
return (0);
}
event_debug(("%s: kevent reports %d", __func__, res));
for (i = 0; i < res; i++) {
int which = 0;
if (events[i].flags & EV_ERROR) {
switch (events[i].data) {
/* Can occur on delete if we are not currently
* watching any events on this fd. That can
* happen when the fd was closed and another
* file was opened with that fd. */
case ENOENT:
/* Can occur for reasons not fully understood
* on FreeBSD. */
case EINVAL:
continue;
/* Can occur on a delete if the fd is closed. */
case EBADF:
/* XXXX On NetBSD, we can also get EBADF if we
* try to add the write side of a pipe, but
* the read side has already been closed.
* Other BSDs call this situation 'EPIPE'. It
* would be good if we had a way to report
* this situation. */
continue;
/* These two can occur on an add if the fd was one side
* of a pipe, and the other side was closed. */
case EPERM:
case EPIPE:
/* Report read events, if we're listening for
* them, so that the user can learn about any
* add errors. (If the operation was a
* delete, then udata should be cleared.) */
if (events[i].udata) {
/* The operation was an add:
* report the error as a read. */
which |= EV_READ;
break;
} else {
/* The operation was a del:
* report nothing. */
continue;
}
/* Other errors shouldn't occur. */
default:
errno = events[i].data;
return (-1);
}
} else if (events[i].filter == EVFILT_READ) {
which |= EV_READ;
} else if (events[i].filter == EVFILT_WRITE) {
which |= EV_WRITE;
} else if (events[i].filter == EVFILT_SIGNAL) {
which |= EV_SIGNAL;
#ifdef EVFILT_USER
} else if (events[i].filter == EVFILT_USER) {
base->is_notify_pending = 0;
#endif
}
if (!which)
continue;
if (events[i].filter == EVFILT_SIGNAL) {
evmap_signal_active_(base, events[i].ident, 1);
} else {
evmap_io_active_(base, events[i].ident, which | EV_ET);
}
}
if (res == kqop->events_size) {
/* We used all the events space that we have. Maybe we should
make it bigger. */
kq_grow_events(kqop, kqop->events_size * 2);
}
return (0);
}
int
win32_dispatch(struct event_base *base, struct timeval *tv)
{
struct win32op *win32op = base->evbase;
int res = 0;
unsigned j, i;
int fd_count;
SOCKET s;
if (win32op->resize_out_sets) {
size_t size = FD_SET_ALLOC_SIZE(win32op->num_fds_in_fd_sets);
if (!(win32op->readset_out = mm_realloc(win32op->readset_out, size)))
return (-1);
if (!(win32op->exset_out = mm_realloc(win32op->exset_out, size)))
return (-1);
if (!(win32op->writeset_out = mm_realloc(win32op->writeset_out, size)))
return (-1);
win32op->resize_out_sets = 0;
}
fd_set_copy(win32op->readset_out, win32op->readset_in);
fd_set_copy(win32op->exset_out, win32op->writeset_in);
fd_set_copy(win32op->writeset_out, win32op->writeset_in);
fd_count =
(win32op->readset_out->fd_count > win32op->writeset_out->fd_count) ?
win32op->readset_out->fd_count : win32op->writeset_out->fd_count;
if (!fd_count) {
long msec = tv ? evutil_tv_to_msec_(tv) : LONG_MAX;
/* Sleep's DWORD argument is unsigned long */
if (msec < 0)
msec = LONG_MAX;
/* Windows doesn't like you to call select() with no sockets */
Sleep(msec);
return (0);
}
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = select(fd_count,
(struct fd_set*)win32op->readset_out,
(struct fd_set*)win32op->writeset_out,
(struct fd_set*)win32op->exset_out, tv);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
event_debug(("%s: select returned %d", __func__, res));
if (res <= 0) {
return res;
}
if (win32op->readset_out->fd_count) {
i = evutil_weakrand_range_(&base->weakrand_seed,
win32op->readset_out->fd_count);
for (j=0; j<win32op->readset_out->fd_count; ++j) {
if (++i >= win32op->readset_out->fd_count)
i = 0;
s = win32op->readset_out->fd_array[i];
evmap_io_active_(base, s, EV_READ);
}
}
if (win32op->exset_out->fd_count) {
i = evutil_weakrand_range_(&base->weakrand_seed,
win32op->exset_out->fd_count);
for (j=0; j<win32op->exset_out->fd_count; ++j) {
if (++i >= win32op->exset_out->fd_count)
i = 0;
s = win32op->exset_out->fd_array[i];
evmap_io_active_(base, s, EV_WRITE);
}
}
if (win32op->writeset_out->fd_count) {
SOCKET s;
i = evutil_weakrand_range_(&base->weakrand_seed,
win32op->writeset_out->fd_count);
for (j=0; j<win32op->writeset_out->fd_count; ++j) {
if (++i >= win32op->writeset_out->fd_count)
i = 0;
s = win32op->writeset_out->fd_array[i];
evmap_io_active_(base, s, EV_WRITE);
}
}
return (0);
}
static int
epoll_dispatch(struct event_base *base, struct timeval *tv)
{
struct epollop *epollop = base->evbase;
struct epoll_event *events = epollop->events;
int i, res;
long timeout = -1;
#ifdef USING_TIMERFD
if (epollop->timerfd >= 0) {
struct itimerspec is;
is.it_interval.tv_sec = 0;
is.it_interval.tv_nsec = 0;
if (tv == NULL) {
/* No timeout; disarm the timer. */
is.it_value.tv_sec = 0;
is.it_value.tv_nsec = 0;
} else {
if (tv->tv_sec == 0 && tv->tv_usec == 0) {
/* we need to exit immediately; timerfd can't
* do that. */
timeout = 0;
}
is.it_value.tv_sec = tv->tv_sec;
is.it_value.tv_nsec = tv->tv_usec * 1000;
}
/* TODO: we could avoid unnecessary syscalls here by only
calling timerfd_settime when the top timeout changes, or
when we're called with a different timeval.
*/
if (timerfd_settime(epollop->timerfd, 0, &is, NULL) < 0) {
event_warn("timerfd_settime");
}
} else
#endif
if (tv != NULL) {
timeout = evutil_tv_to_msec_(tv);
if (timeout < 0 || timeout > MAX_EPOLL_TIMEOUT_MSEC) {
/* Linux kernels can wait forever if the timeout is
* too big; see comment on MAX_EPOLL_TIMEOUT_MSEC. */
timeout = MAX_EPOLL_TIMEOUT_MSEC;
}
}
epoll_apply_changes(base);
event_changelist_remove_all_(&base->changelist, base);
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = epoll_wait(epollop->epfd, events, epollop->nevents, timeout);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
if (res == -1) {
if (errno != EINTR) {
event_warn("epoll_wait");
return (-1);
}
return (0);
}
event_debug(("%s: epoll_wait reports %d", __func__, res));
EVUTIL_ASSERT(res <= epollop->nevents);
for (i = 0; i < res; i++) {
int what = events[i].events;
short ev = 0;
#ifdef USING_TIMERFD
if (events[i].data.fd == epollop->timerfd)
continue;
#endif
if (what & (EPOLLHUP|EPOLLERR)) {
ev = EV_READ | EV_WRITE;
} else {
if (what & EPOLLIN)
ev |= EV_READ;
if (what & EPOLLOUT)
ev |= EV_WRITE;
if (what & EPOLLRDHUP)
ev |= EV_CLOSED;
}
if (!ev)
continue;
evmap_io_active_(base, events[i].data.fd, ev | EV_ET);
}
if (res == epollop->nevents && epollop->nevents < MAX_NEVENT) {
/* We used all of the event space this time. We should
be ready for more events next time. */
int new_nevents = epollop->nevents * 2;
struct epoll_event *new_events;
new_events = mm_realloc(epollop->events,
new_nevents * sizeof(struct epoll_event));
if (new_events) {
epollop->events = new_events;
epollop->nevents = new_nevents;
}
}
return (0);
}
static int
evport_dispatch(struct event_base *base, struct timeval *tv)
{
int i, res;
struct evport_data *epdp = base->evbase;
port_event_t *pevtlist = epdp->ed_pevtlist;
/*
* port_getn will block until it has at least nevents events. It will
* also return how many it's given us (which may be more than we asked
* for, as long as it's less than our maximum (ed_maxevents)) in
* nevents.
*/
int nevents = 1;
/*
* We have to convert a struct timeval to a struct timespec
* (only difference is nanoseconds vs. microseconds). If no time-based
* events are active, we should wait for I/O (and tv == NULL).
*/
struct timespec ts;
struct timespec *ts_p = NULL;
if (tv != NULL) {
ts.tv_sec = tv->tv_sec;
ts.tv_nsec = tv->tv_usec * 1000;
ts_p = &ts;
}
/*
* Before doing anything else, we need to reassociate the events we hit
* last time which need reassociation. See comment at the end of the
* loop below.
*/
for (i = 0; i < epdp->ed_npending; ++i) {
struct fd_info *fdi = NULL;
const int fd = epdp->ed_pending[i];
if (fd != -1) {
/* We might have cleared out this event; we need
* to be sure that it's still set. */
fdi = evmap_io_get_fdinfo_(&base->io, fd);
}
if (fdi != NULL && FDI_HAS_EVENTS(fdi)) {
reassociate(epdp, fdi, fd);
/* epdp->ed_pending[i] = -1; */
fdi->pending_idx_plus_1 = 0;
}
}
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = port_getn(epdp->ed_port, pevtlist, epdp->ed_maxevents,
(unsigned int *) &nevents, ts_p);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
if (res == -1) {
if (errno == EINTR || errno == EAGAIN) {
return (0);
} else if (errno == ETIME) {
if (nevents == 0)
return (0);
} else {
event_warn("port_getn");
return (-1);
}
}
event_debug(("%s: port_getn reports %d events", __func__, nevents));
for (i = 0; i < nevents; ++i) {
port_event_t *pevt = &pevtlist[i];
int fd = (int) pevt->portev_object;
struct fd_info *fdi = pevt->portev_user;
/*EVUTIL_ASSERT(evmap_io_get_fdinfo_(&base->io, fd) == fdi);*/
check_evportop(epdp);
check_event(pevt);
epdp->ed_pending[i] = fd;
fdi->pending_idx_plus_1 = i + 1;
/*
* Figure out what kind of event it was
* (because we have to pass this to the callback)
*/
res = 0;
if (pevt->portev_events & (POLLERR|POLLHUP)) {
res = EV_READ | EV_WRITE;
} else {
if (pevt->portev_events & POLLIN)
res |= EV_READ;
if (pevt->portev_events & POLLOUT)
res |= EV_WRITE;
}
/*
* Check for the error situations or a hangup situation
*/
if (pevt->portev_events & (POLLERR|POLLHUP|POLLNVAL))
res |= EV_READ|EV_WRITE;
evmap_io_active_(base, fd, res);
} /* end of all events gotten */
epdp->ed_npending = nevents;
if (nevents == epdp->ed_maxevents &&
epdp->ed_maxevents < MAX_EVENTS_PER_GETN) {
/* we used all the space this time. We should be ready
* for more events next time around. */
grow(epdp, epdp->ed_maxevents * 2);
}
check_evportop(epdp);
return (0);
}
static int
poll_dispatch(struct event_base *base, struct timeval *tv)
{
int res, i, j, nfds;
long msec = -1;
struct pollop *pop = base->evbase;
struct pollfd *event_set;
poll_check_ok(pop);
nfds = pop->nfds;
#ifndef EVENT__DISABLE_THREAD_SUPPORT
if (base->th_base_lock) {
/* If we're using this backend in a multithreaded setting,
* then we need to work on a copy of event_set, so that we can
* let other threads modify the main event_set while we're
* polling. If we're not multithreaded, then we'll skip the
* copy step here to save memory and time. */
if (pop->realloc_copy) {
struct pollfd *tmp = mm_realloc(pop->event_set_copy,
pop->event_count * sizeof(struct pollfd));
if (tmp == NULL) {
event_warn("realloc");
return -1;
}
pop->event_set_copy = tmp;
pop->realloc_copy = 0;
}
memcpy(pop->event_set_copy, pop->event_set,
sizeof(struct pollfd)*nfds);
event_set = pop->event_set_copy;
} else {
event_set = pop->event_set;
}
#else
event_set = pop->event_set;
#endif
if (tv != NULL) {
msec = evutil_tv_to_msec_(tv);
if (msec < 0 || msec > INT_MAX)
msec = INT_MAX;
}
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = poll(event_set, nfds, msec);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
if (res == -1) {
if (errno != EINTR) {
event_warn("poll");
return (-1);
}
return (0);
}
event_debug(("%s: poll reports %d", __func__, res));
if (res == 0 || nfds == 0)
return (0);
i = evutil_weakrand_range_(&base->weakrand_seed, nfds);
for (j = 0; j < nfds; j++) {
int what;
if (++i == nfds)
i = 0;
what = event_set[i].revents;
if (!what)
continue;
res = 0;
/* If the file gets closed notify */
if (what & (POLLHUP|POLLERR|POLLNVAL))
what |= POLLIN|POLLOUT;
if (what & POLLIN)
res |= EV_READ;
if (what & POLLOUT)
res |= EV_WRITE;
if (res == 0)
continue;
evmap_io_active_(base, event_set[i].fd, res);
}
return (0);
}
static int
select_dispatch(struct event_base *base, struct timeval *tv)
{
int res=0, i, j, nfds;
struct selectop *sop = base->evbase;
check_selectop(sop);
if (sop->resize_out_sets) {
fd_set *readset_out=NULL, *writeset_out=NULL;
size_t sz = sop->event_fdsz;
if (!(readset_out = mm_realloc(sop->event_readset_out, sz)))
return (-1);
sop->event_readset_out = readset_out;
if (!(writeset_out = mm_realloc(sop->event_writeset_out, sz))) {
/* We don't free readset_out here, since it was
* already successfully reallocated. The next time
* we call select_dispatch, the realloc will be a
* no-op. */
return (-1);
}
sop->event_writeset_out = writeset_out;
sop->resize_out_sets = 0;
}
memcpy(sop->event_readset_out, sop->event_readset_in,
sop->event_fdsz);
memcpy(sop->event_writeset_out, sop->event_writeset_in,
sop->event_fdsz);
nfds = sop->event_fds+1;
EVBASE_RELEASE_LOCK(base, th_base_lock);
res = select(nfds, sop->event_readset_out,
sop->event_writeset_out, NULL, tv);
EVBASE_ACQUIRE_LOCK(base, th_base_lock);
check_selectop(sop);
if (res == -1) {
if (errno != EINTR) {
event_warn("select");
return (-1);
}
return (0);
}
event_debug(("%s: select reports %d", __func__, res));
check_selectop(sop);
i = evutil_weakrand_range_(&base->weakrand_seed, nfds);
for (j = 0; j < nfds; ++j) {
if (++i >= nfds)
i = 0;
res = 0;
if (FD_ISSET(i, sop->event_readset_out))
res |= EV_READ;
if (FD_ISSET(i, sop->event_writeset_out))
res |= EV_WRITE;
if (res == 0)
continue;
evmap_io_active_(base, i, res);
}
check_selectop(sop);
return (0);
}