*/
#include "includes.h"
-#include <tevent_internal.h>
+#include "lib/tevent/tevent_internal.h"
+#include "../lib/util/select.h"
+#include "system/select.h"
+
+struct tevent_poll_private {
+ /*
+ * Index from file descriptor into the pollfd array
+ */
+ int *pollfd_idx;
+
+ /*
+ * Cache for s3_event_loop_once to avoid reallocs
+ */
+ struct pollfd *pfds;
+};
-void event_fd_set_writeable(struct tevent_fd *fde)
+static struct tevent_poll_private *tevent_get_poll_private(
+ struct tevent_context *ev)
{
- TEVENT_FD_WRITEABLE(fde);
+ struct tevent_poll_private *state;
+
+ state = (struct tevent_poll_private *)ev->additional_data;
+ if (state == NULL) {
+ state = talloc_zero(ev, struct tevent_poll_private);
+ ev->additional_data = (void *)state;
+ if (state == NULL) {
+ DEBUG(10, ("talloc failed\n"));
+ }
+ }
+ return state;
}
-void event_fd_set_not_writeable(struct tevent_fd *fde)
+static void count_fds(struct tevent_context *ev,
+ int *pnum_fds, int *pmax_fd)
{
- TEVENT_FD_NOT_WRITEABLE(fde);
+ struct tevent_fd *fde;
+ int num_fds = 0;
+ int max_fd = 0;
+
+ for (fde = ev->fd_events; fde != NULL; fde = fde->next) {
+ if (fde->flags & (EVENT_FD_READ|EVENT_FD_WRITE)) {
+ num_fds += 1;
+ if (fde->fd > max_fd) {
+ max_fd = fde->fd;
+ }
+ }
+ }
+ *pnum_fds = num_fds;
+ *pmax_fd = max_fd;
}
-void event_fd_set_readable(struct tevent_fd *fde)
+bool event_add_to_poll_args(struct tevent_context *ev, TALLOC_CTX *mem_ctx,
+ struct pollfd **pfds, int *pnum_pfds,
+ int *ptimeout)
{
- TEVENT_FD_READABLE(fde);
-}
+ struct tevent_poll_private *state;
+ struct tevent_fd *fde;
+ int i, num_fds, max_fd, num_pollfds, idx_len;
+ struct pollfd *fds;
+ struct timeval now, diff;
+ int timeout;
-void event_fd_set_not_readable(struct tevent_fd *fde)
-{
- TEVENT_FD_NOT_READABLE(fde);
-}
+ state = tevent_get_poll_private(ev);
+ if (state == NULL) {
+ return false;
+ }
+ count_fds(ev, &num_fds, &max_fd);
-/*
- * Return if there's something in the queue
- */
+ idx_len = max_fd+1;
-bool event_add_to_select_args(struct tevent_context *ev,
- const struct timeval *now,
- fd_set *read_fds, fd_set *write_fds,
- struct timeval *timeout, int *maxfd)
-{
- struct tevent_fd *fde;
- struct timeval diff;
- bool ret = false;
+ if (talloc_array_length(state->pollfd_idx) < idx_len) {
+ state->pollfd_idx = talloc_realloc(
+ state, state->pollfd_idx, int, idx_len);
+ if (state->pollfd_idx == NULL) {
+ DEBUG(10, ("talloc_realloc failed\n"));
+ return false;
+ }
+ }
+
+ fds = *pfds;
+ num_pollfds = *pnum_pfds;
+
+ /*
+ * The +1 is for the sys_poll calling convention. It expects
+ * an array 1 longer for the signal pipe
+ */
+
+ if (talloc_array_length(fds) < num_pollfds + num_fds + 1) {
+ fds = talloc_realloc(mem_ctx, fds, struct pollfd,
+ num_pollfds + num_fds + 1);
+ if (fds == NULL) {
+ DEBUG(10, ("talloc_realloc failed\n"));
+ return false;
+ }
+ }
+
+ memset(&fds[num_pollfds], 0, sizeof(struct pollfd) * num_fds);
+
+ /*
+ * This needs tuning. We need to cope with multiple fde's for a file
+ * descriptor. The problem is that we need to re-use pollfd_idx across
+ * calls for efficiency. One way would be a direct bitmask that might
+ * be initialized quicker, but our bitmap_init implementation is
+ * pretty heavy-weight as well.
+ */
+ for (i=0; i<idx_len; i++) {
+ state->pollfd_idx[i] = -1;
+ }
for (fde = ev->fd_events; fde; fde = fde->next) {
- if (fde->flags & EVENT_FD_READ) {
- FD_SET(fde->fd, read_fds);
- ret = true;
+ struct pollfd *pfd;
+
+ if ((fde->flags & (EVENT_FD_READ|EVENT_FD_WRITE)) == 0) {
+ continue;
}
- if (fde->flags & EVENT_FD_WRITE) {
- FD_SET(fde->fd, write_fds);
- ret = true;
+
+ if (state->pollfd_idx[fde->fd] == -1) {
+ /*
+ * We haven't seen this fd yet. Allocate a new pollfd.
+ */
+ state->pollfd_idx[fde->fd] = num_pollfds;
+ pfd = &fds[num_pollfds];
+ num_pollfds += 1;
+ } else {
+ /*
+ * We have already seen this fd. OR in the flags.
+ */
+ pfd = &fds[state->pollfd_idx[fde->fd]];
}
- if ((fde->flags & (EVENT_FD_READ|EVENT_FD_WRITE))
- && (fde->fd > *maxfd)) {
- *maxfd = fde->fd;
+ pfd->fd = fde->fd;
+
+ if (fde->flags & EVENT_FD_READ) {
+ pfd->events |= (POLLIN|POLLHUP);
+ }
+ if (fde->flags & EVENT_FD_WRITE) {
+ pfd->events |= POLLOUT;
}
}
+ *pfds = fds;
+ *pnum_pfds = num_pollfds;
+ if (ev->immediate_events != NULL) {
+ *ptimeout = 0;
+ return true;
+ }
if (ev->timer_events == NULL) {
- return ret;
+ *ptimeout = MIN(*ptimeout, INT_MAX);
+ return true;
}
- diff = timeval_until(now, &ev->timer_events->next_event);
- *timeout = timeval_min(timeout, &diff);
+ now = timeval_current();
+ diff = timeval_until(&now, &ev->timer_events->next_event);
+ timeout = timeval_to_msec(diff);
+
+ if (timeout < *ptimeout) {
+ *ptimeout = timeout;
+ }
return true;
}
-bool run_events(struct tevent_context *ev,
- int selrtn, fd_set *read_fds, fd_set *write_fds)
+bool run_events_poll(struct tevent_context *ev, int pollrtn,
+ struct pollfd *pfds, int num_pfds)
{
+ struct tevent_poll_private *state;
+ int *pollfd_idx;
struct tevent_fd *fde;
struct timeval now;
return true;
}
+ if (ev->immediate_events &&
+ tevent_common_loop_immediate(ev)) {
+ return true;
+ }
+
GetTimeOfDay(&now);
if ((ev->timer_events != NULL)
&& (timeval_compare(&now, &ev->timer_events->next_event) >= 0)) {
+ /* this older events system did not auto-free timed
+ events on running them, and had a race condition
+ where the event could be called twice if the
+ talloc_free of the te happened after the callback
+ made a call which invoked the event loop. To avoid
+ this while still allowing old code which frees the
+ te, we need to create a temporary context which
+ will be used to ensure the te is freed. We also
+ remove the te from the timed event list before we
+ call the handler, to ensure we can't loop */
+
+ struct tevent_timer *te = ev->timer_events;
+ TALLOC_CTX *tmp_ctx = talloc_new(ev);
DEBUG(10, ("Running timed event \"%s\" %p\n",
ev->timer_events->handler_name, ev->timer_events));
- ev->timer_events->handler(ev, ev->timer_events, now,
- ev->timer_events->private_data);
+ DLIST_REMOVE(ev->timer_events, te);
+ talloc_steal(tmp_ctx, te);
+
+ te->handler(ev, te, now, te->private_data);
+
+ talloc_free(tmp_ctx);
return true;
}
- if (selrtn == 0) {
+ if (pollrtn <= 0) {
/*
* No fd ready
*/
return false;
}
+ state = (struct tevent_poll_private *)ev->additional_data;
+ pollfd_idx = state->pollfd_idx;
+
for (fde = ev->fd_events; fde; fde = fde->next) {
+ struct pollfd *pfd;
uint16 flags = 0;
- if (FD_ISSET(fde->fd, read_fds)) flags |= EVENT_FD_READ;
- if (FD_ISSET(fde->fd, write_fds)) flags |= EVENT_FD_WRITE;
+ if (pollfd_idx[fde->fd] >= num_pfds) {
+ DEBUG(1, ("internal error: pollfd_idx[fde->fd] (%d) "
+ ">= num_pfds (%d)\n", pollfd_idx[fde->fd],
+ num_pfds));
+ return false;
+ }
+ pfd = &pfds[pollfd_idx[fde->fd]];
+
+ if (pfd->fd != fde->fd) {
+ DEBUG(1, ("internal error: pfd->fd (%d) "
+ "!= fde->fd (%d)\n", pollfd_idx[fde->fd],
+ num_pfds));
+ return false;
+ }
+
+ if (pfd->revents & (POLLHUP|POLLERR)) {
+ /* If we only wait for EVENT_FD_WRITE, we
+ should not tell the event handler about it,
+ and remove the writable flag, as we only
+ report errors when waiting for read events
+ to match the select behavior. */
+ if (!(fde->flags & EVENT_FD_READ)) {
+ EVENT_FD_NOT_WRITEABLE(fde);
+ continue;
+ }
+ flags |= EVENT_FD_READ;
+ }
+ if (pfd->revents & POLLIN) {
+ flags |= EVENT_FD_READ;
+ }
+ if (pfd->revents & POLLOUT) {
+ flags |= EVENT_FD_WRITE;
+ }
if (flags & fde->flags) {
+ DLIST_DEMOTE(ev->fd_events, fde, struct tevent_fd);
fde->handler(ev, fde, flags, fde->private_data);
return true;
}
return false;
}
-
struct timeval *get_timed_events_timeout(struct tevent_context *ev,
struct timeval *to_ret)
{
struct timeval now;
- if (ev->timer_events == NULL) {
+ if ((ev->timer_events == NULL) && (ev->immediate_events == NULL)) {
return NULL;
}
+ if (ev->immediate_events != NULL) {
+ *to_ret = timeval_zero();
+ return to_ret;
+ }
now = timeval_current();
*to_ret = timeval_until(&now, &ev->timer_events->next_event);
return to_ret;
}
-static int s3_event_loop_once(struct tevent_context *ev)
+static int s3_event_loop_once(struct tevent_context *ev, const char *location)
{
- struct timeval now, to;
- fd_set r_fds, w_fds;
- int maxfd = 0;
+ struct tevent_poll_private *state;
+ int timeout;
+ int num_pfds;
int ret;
- FD_ZERO(&r_fds);
- FD_ZERO(&w_fds);
+ timeout = INT_MAX;
- to.tv_sec = 9999; /* Max timeout */
- to.tv_usec = 0;
+ state = tevent_get_poll_private(ev);
+ if (state == NULL) {
+ errno = ENOMEM;
+ return -1;
+ }
- if (run_events(ev, 0, NULL, NULL)) {
+ if (run_events_poll(ev, 0, NULL, 0)) {
return 0;
}
- GetTimeOfDay(&now);
-
- if (!event_add_to_select_args(ev, &now, &r_fds, &w_fds, &to, &maxfd)) {
+ num_pfds = 0;
+ if (!event_add_to_poll_args(ev, state,
+ &state->pfds, &num_pfds, &timeout)) {
return -1;
}
- ret = sys_select(maxfd+1, &r_fds, &w_fds, NULL, &to);
-
+ ret = sys_poll(state->pfds, num_pfds, timeout);
if (ret == -1 && errno != EINTR) {
tevent_debug(ev, TEVENT_DEBUG_FATAL,
- "sys_select() failed: %d:%s\n",
+ "poll() failed: %d:%s\n",
errno, strerror(errno));
return -1;
}
- run_events(ev, ret, &r_fds, &w_fds);
+ run_events_poll(ev, ret, state->pfds, num_pfds);
return 0;
}
-static int s3_event_loop_wait(struct tevent_context *ev)
-{
- int ret = 0;
-
- while (ret == 0) {
- ret = s3_event_loop_once(ev);
- }
-
- return ret;
-}
-
-void event_context_reinit(struct tevent_context *ev)
-{
- tevent_common_context_destructor(ev);
- return;
-}
-
static int s3_event_context_init(struct tevent_context *ev)
{
return 0;
}
static const struct tevent_ops s3_event_ops = {
- .context_init = s3_event_context_init,
- .add_fd = tevent_common_add_fd,
- .set_fd_close_fn= tevent_common_fd_set_close_fn,
- .get_fd_flags = tevent_common_fd_get_flags,
- .set_fd_flags = tevent_common_fd_set_flags,
- .add_timer = tevent_common_add_timer,
- .add_signal = tevent_common_add_signal,
- .loop_once = s3_event_loop_once,
- .loop_wait = s3_event_loop_wait,
+ .context_init = s3_event_context_init,
+ .add_fd = tevent_common_add_fd,
+ .set_fd_close_fn = tevent_common_fd_set_close_fn,
+ .get_fd_flags = tevent_common_fd_get_flags,
+ .set_fd_flags = tevent_common_fd_set_flags,
+ .add_timer = tevent_common_add_timer,
+ .schedule_immediate = tevent_common_schedule_immediate,
+ .add_signal = tevent_common_add_signal,
+ .loop_once = s3_event_loop_once,
+ .loop_wait = tevent_common_loop_wait,
};
static bool s3_tevent_init(void)
samba_level = 2;
break;
case TEVENT_DEBUG_TRACE:
- samba_level = 10;
+ samba_level = 11;
break;
};
- vasprintf(&s, fmt, ap);
- if (!s) return;
+ if (vasprintf(&s, fmt, ap) == -1) {
+ return;
+ }
DEBUG(samba_level, ("s3_event: %s", s));
free(s);
}