1/*
2 * Copyright (c) 2000-2007 Niels Provos <provos@citi.umich.edu>
3 * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. The name of the author may not be used to endorse or promote products
14 *    derived from this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27#include "event2/event-config.h"
28#include "evconfig-private.h"
29
30#ifdef _WIN32
31#include <winsock2.h>
32#define WIN32_LEAN_AND_MEAN
33#include <windows.h>
34#undef WIN32_LEAN_AND_MEAN
35#endif
36#include <sys/types.h>
37#if !defined(_WIN32) && defined(EVENT__HAVE_SYS_TIME_H)
38#include <sys/time.h>
39#endif
40#include <sys/queue.h>
41#ifdef EVENT__HAVE_SYS_SOCKET_H
42#include <sys/socket.h>
43#endif
44#include <stdio.h>
45#include <stdlib.h>
46#ifdef EVENT__HAVE_UNISTD_H
47#include <unistd.h>
48#endif
49#include <ctype.h>
50#include <errno.h>
51#include <signal.h>
52#include <string.h>
53#include <time.h>
54#include <limits.h>
55#ifdef EVENT__HAVE_FCNTL_H
56#include <fcntl.h>
57#endif
58
59#include "event2/event.h"
60#include "event2/event_struct.h"
61#include "event2/event_compat.h"
62#include "event-internal.h"
63#include "defer-internal.h"
64#include "evthread-internal.h"
65#include "event2/thread.h"
66#include "event2/util.h"
67#include "log-internal.h"
68#include "evmap-internal.h"
69#include "iocp-internal.h"
70#include "changelist-internal.h"
71#define HT_NO_CACHE_HASH_VALUES
72#include "ht-internal.h"
73#include "util-internal.h"
74
75
76#ifdef EVENT__HAVE_WORKING_KQUEUE
77#include "kqueue-internal.h"
78#endif
79
80#ifdef EVENT__HAVE_EVENT_PORTS
81extern const struct eventop evportops;
82#endif
83#ifdef EVENT__HAVE_SELECT
84extern const struct eventop selectops;
85#endif
86#ifdef EVENT__HAVE_POLL
87extern const struct eventop pollops;
88#endif
89#ifdef EVENT__HAVE_EPOLL
90extern const struct eventop epollops;
91#endif
92#ifdef EVENT__HAVE_WORKING_KQUEUE
93extern const struct eventop kqops;
94#endif
95#ifdef EVENT__HAVE_DEVPOLL
96extern const struct eventop devpollops;
97#endif
98#ifdef _WIN32
99extern const struct eventop win32ops;
100#endif
101
102/* Array of backends in order of preference. */
103static const struct eventop *eventops[] = {
104#ifdef EVENT__HAVE_EVENT_PORTS
105	&evportops,
106#endif
107#ifdef EVENT__HAVE_WORKING_KQUEUE
108	&kqops,
109#endif
110#ifdef EVENT__HAVE_EPOLL
111	&epollops,
112#endif
113#ifdef EVENT__HAVE_DEVPOLL
114	&devpollops,
115#endif
116#ifdef EVENT__HAVE_POLL
117	&pollops,
118#endif
119#ifdef EVENT__HAVE_SELECT
120	&selectops,
121#endif
122#ifdef _WIN32
123	&win32ops,
124#endif
125	NULL
126};
127
128/* Global state; deprecated */
129EVENT2_EXPORT_SYMBOL
130struct event_base *event_global_current_base_ = NULL;
131#define current_base event_global_current_base_
132
133/* Global state */
134
135static void *event_self_cbarg_ptr_ = NULL;
136
137/* Prototypes */
138static void	event_queue_insert_active(struct event_base *, struct event_callback *);
139static void	event_queue_insert_active_later(struct event_base *, struct event_callback *);
140static void	event_queue_insert_timeout(struct event_base *, struct event *);
141static void	event_queue_insert_inserted(struct event_base *, struct event *);
142static void	event_queue_remove_active(struct event_base *, struct event_callback *);
143static void	event_queue_remove_active_later(struct event_base *, struct event_callback *);
144static void	event_queue_remove_timeout(struct event_base *, struct event *);
145static void	event_queue_remove_inserted(struct event_base *, struct event *);
146static void event_queue_make_later_events_active(struct event_base *base);
147
148static int evthread_make_base_notifiable_nolock_(struct event_base *base);
149static int event_del_(struct event *ev, int blocking);
150
151#ifdef USE_REINSERT_TIMEOUT
152/* This code seems buggy; only turn it on if we find out what the trouble is. */
153static void	event_queue_reinsert_timeout(struct event_base *,struct event *, int was_common, int is_common, int old_timeout_idx);
154#endif
155
156static int	event_haveevents(struct event_base *);
157
158static int	event_process_active(struct event_base *);
159
160static int	timeout_next(struct event_base *, struct timeval **);
161static void	timeout_process(struct event_base *);
162
163static inline void	event_signal_closure(struct event_base *, struct event *ev);
164static inline void	event_persist_closure(struct event_base *, struct event *ev);
165
166static int	evthread_notify_base(struct event_base *base);
167
168static void insert_common_timeout_inorder(struct common_timeout_list *ctl,
169    struct event *ev);
170
171#ifndef EVENT__DISABLE_DEBUG_MODE
172/* These functions implement a hashtable of which 'struct event *' structures
173 * have been setup or added.  We don't want to trust the content of the struct
174 * event itself, since we're trying to work through cases where an event gets
175 * clobbered or freed.  Instead, we keep a hashtable indexed by the pointer.
176 */
177
178struct event_debug_entry {
179	HT_ENTRY(event_debug_entry) node;
180	const struct event *ptr;
181	unsigned added : 1;
182};
183
184static inline unsigned
185hash_debug_entry(const struct event_debug_entry *e)
186{
187	/* We need to do this silliness to convince compilers that we
188	 * honestly mean to cast e->ptr to an integer, and discard any
189	 * part of it that doesn't fit in an unsigned.
190	 */
191	unsigned u = (unsigned) ((ev_uintptr_t) e->ptr);
192	/* Our hashtable implementation is pretty sensitive to low bits,
193	 * and every struct event is over 64 bytes in size, so we can
194	 * just say >>6. */
195	return (u >> 6);
196}
197
198static inline int
199eq_debug_entry(const struct event_debug_entry *a,
200    const struct event_debug_entry *b)
201{
202	return a->ptr == b->ptr;
203}
204
205int event_debug_mode_on_ = 0;
206
207
208#if !defined(EVENT__DISABLE_THREAD_SUPPORT) && !defined(EVENT__DISABLE_DEBUG_MODE)
209/**
210 * @brief debug mode variable which is set for any function/structure that needs
211 *        to be shared across threads (if thread support is enabled).
212 *
213 *        When and if evthreads are initialized, this variable will be evaluated,
214 *        and if set to something other than zero, this means the evthread setup
215 *        functions were called out of order.
216 *
217 *        See: "Locks and threading" in the documentation.
218 */
219int event_debug_created_threadable_ctx_ = 0;
220#endif
221
222/* Set if it's too late to enable event_debug_mode. */
223static int event_debug_mode_too_late = 0;
224#ifndef EVENT__DISABLE_THREAD_SUPPORT
225static void *event_debug_map_lock_ = NULL;
226#endif
227static HT_HEAD(event_debug_map, event_debug_entry) global_debug_map =
228	HT_INITIALIZER();
229
230HT_PROTOTYPE(event_debug_map, event_debug_entry, node, hash_debug_entry,
231    eq_debug_entry)
232HT_GENERATE(event_debug_map, event_debug_entry, node, hash_debug_entry,
233    eq_debug_entry, 0.5, mm_malloc, mm_realloc, mm_free)
234
235/* record that ev is now setup (that is, ready for an add) */
236static void event_debug_note_setup_(const struct event *ev)
237{
238	struct event_debug_entry *dent, find;
239
240	if (!event_debug_mode_on_)
241		goto out;
242
243	find.ptr = ev;
244	EVLOCK_LOCK(event_debug_map_lock_, 0);
245	dent = HT_FIND(event_debug_map, &global_debug_map, &find);
246	if (dent) {
247		dent->added = 0;
248	} else {
249		dent = mm_malloc(sizeof(*dent));
250		if (!dent)
251			event_err(1,
252			    "Out of memory in debugging code");
253		dent->ptr = ev;
254		dent->added = 0;
255		HT_INSERT(event_debug_map, &global_debug_map, dent);
256	}
257	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
258
259out:
260	event_debug_mode_too_late = 1;
261}
262/* record that ev is no longer setup */
263static void event_debug_note_teardown_(const struct event *ev)
264{
265	struct event_debug_entry *dent, find;
266
267	if (!event_debug_mode_on_)
268		goto out;
269
270	find.ptr = ev;
271	EVLOCK_LOCK(event_debug_map_lock_, 0);
272	dent = HT_REMOVE(event_debug_map, &global_debug_map, &find);
273	if (dent)
274		mm_free(dent);
275	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
276
277out:
278	event_debug_mode_too_late = 1;
279}
280/* Macro: record that ev is now added */
281static void event_debug_note_add_(const struct event *ev)
282{
283	struct event_debug_entry *dent,find;
284
285	if (!event_debug_mode_on_)
286		goto out;
287
288	find.ptr = ev;
289	EVLOCK_LOCK(event_debug_map_lock_, 0);
290	dent = HT_FIND(event_debug_map, &global_debug_map, &find);
291	if (dent) {
292		dent->added = 1;
293	} else {
294		event_errx(EVENT_ERR_ABORT_,
295		    "%s: noting an add on a non-setup event %p"
296		    " (events: 0x%x, fd: "EV_SOCK_FMT
297		    ", flags: 0x%x)",
298		    __func__, ev, ev->ev_events,
299		    EV_SOCK_ARG(ev->ev_fd), ev->ev_flags);
300	}
301	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
302
303out:
304	event_debug_mode_too_late = 1;
305}
306/* record that ev is no longer added */
307static void event_debug_note_del_(const struct event *ev)
308{
309	struct event_debug_entry *dent, find;
310
311	if (!event_debug_mode_on_)
312		goto out;
313
314	find.ptr = ev;
315	EVLOCK_LOCK(event_debug_map_lock_, 0);
316	dent = HT_FIND(event_debug_map, &global_debug_map, &find);
317	if (dent) {
318		dent->added = 0;
319	} else {
320		event_errx(EVENT_ERR_ABORT_,
321		    "%s: noting a del on a non-setup event %p"
322		    " (events: 0x%x, fd: "EV_SOCK_FMT
323		    ", flags: 0x%x)",
324		    __func__, ev, ev->ev_events,
325		    EV_SOCK_ARG(ev->ev_fd), ev->ev_flags);
326	}
327	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
328
329out:
330	event_debug_mode_too_late = 1;
331}
332/* assert that ev is setup (i.e., okay to add or inspect) */
333static void event_debug_assert_is_setup_(const struct event *ev)
334{
335	struct event_debug_entry *dent, find;
336
337	if (!event_debug_mode_on_)
338		return;
339
340	find.ptr = ev;
341	EVLOCK_LOCK(event_debug_map_lock_, 0);
342	dent = HT_FIND(event_debug_map, &global_debug_map, &find);
343	if (!dent) {
344		event_errx(EVENT_ERR_ABORT_,
345		    "%s called on a non-initialized event %p"
346		    " (events: 0x%x, fd: "EV_SOCK_FMT
347		    ", flags: 0x%x)",
348		    __func__, ev, ev->ev_events,
349		    EV_SOCK_ARG(ev->ev_fd), ev->ev_flags);
350	}
351	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
352}
353/* assert that ev is not added (i.e., okay to tear down or set up again) */
354static void event_debug_assert_not_added_(const struct event *ev)
355{
356	struct event_debug_entry *dent, find;
357
358	if (!event_debug_mode_on_)
359		return;
360
361	find.ptr = ev;
362	EVLOCK_LOCK(event_debug_map_lock_, 0);
363	dent = HT_FIND(event_debug_map, &global_debug_map, &find);
364	if (dent && dent->added) {
365		event_errx(EVENT_ERR_ABORT_,
366		    "%s called on an already added event %p"
367		    " (events: 0x%x, fd: "EV_SOCK_FMT", "
368		    "flags: 0x%x)",
369		    __func__, ev, ev->ev_events,
370		    EV_SOCK_ARG(ev->ev_fd), ev->ev_flags);
371	}
372	EVLOCK_UNLOCK(event_debug_map_lock_, 0);
373}
374static void event_debug_assert_socket_nonblocking_(evutil_socket_t fd)
375{
376	if (!event_debug_mode_on_)
377		return;
378	if (fd < 0)
379		return;
380
381#ifndef _WIN32
382	{
383		int flags;
384		if ((flags = fcntl(fd, F_GETFL, NULL)) >= 0) {
385			EVUTIL_ASSERT(flags & O_NONBLOCK);
386		}
387	}
388#endif
389}
390#else
391static void event_debug_note_setup_(const struct event *ev) { (void)ev; }
392static void event_debug_note_teardown_(const struct event *ev) { (void)ev; }
393static void event_debug_note_add_(const struct event *ev) { (void)ev; }
394static void event_debug_note_del_(const struct event *ev) { (void)ev; }
395static void event_debug_assert_is_setup_(const struct event *ev) { (void)ev; }
396static void event_debug_assert_not_added_(const struct event *ev) { (void)ev; }
397static void event_debug_assert_socket_nonblocking_(evutil_socket_t fd) { (void)fd; }
398#endif
399
400#define EVENT_BASE_ASSERT_LOCKED(base)		\
401	EVLOCK_ASSERT_LOCKED((base)->th_base_lock)
402
403/* How often (in seconds) do we check for changes in wall clock time relative
404 * to monotonic time?  Set this to -1 for 'never.' */
405#define CLOCK_SYNC_INTERVAL 5
406
407/** Set 'tp' to the current time according to 'base'.  We must hold the lock
408 * on 'base'.  If there is a cached time, return it.  Otherwise, use
409 * clock_gettime or gettimeofday as appropriate to find out the right time.
410 * Return 0 on success, -1 on failure.
411 */
412static int
413gettime(struct event_base *base, struct timeval *tp)
414{
415	EVENT_BASE_ASSERT_LOCKED(base);
416
417	if (base->tv_cache.tv_sec) {
418		*tp = base->tv_cache;
419		return (0);
420	}
421
422	if (evutil_gettime_monotonic_(&base->monotonic_timer, tp) == -1) {
423		return -1;
424	}
425
426	if (base->last_updated_clock_diff + CLOCK_SYNC_INTERVAL
427	    < tp->tv_sec) {
428		struct timeval tv;
429		evutil_gettimeofday(&tv,NULL);
430		evutil_timersub(&tv, tp, &base->tv_clock_diff);
431		base->last_updated_clock_diff = tp->tv_sec;
432	}
433
434	return 0;
435}
436
437int
438event_base_gettimeofday_cached(struct event_base *base, struct timeval *tv)
439{
440	int r;
441	if (!base) {
442		base = current_base;
443		if (!current_base)
444			return evutil_gettimeofday(tv, NULL);
445	}
446
447	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
448	if (base->tv_cache.tv_sec == 0) {
449		r = evutil_gettimeofday(tv, NULL);
450	} else {
451		evutil_timeradd(&base->tv_cache, &base->tv_clock_diff, tv);
452		r = 0;
453	}
454	EVBASE_RELEASE_LOCK(base, th_base_lock);
455	return r;
456}
457
458/** Make 'base' have no current cached time. */
459static inline void
460clear_time_cache(struct event_base *base)
461{
462	base->tv_cache.tv_sec = 0;
463}
464
465/** Replace the cached time in 'base' with the current time. */
466static inline void
467update_time_cache(struct event_base *base)
468{
469	base->tv_cache.tv_sec = 0;
470	if (!(base->flags & EVENT_BASE_FLAG_NO_CACHE_TIME))
471	    gettime(base, &base->tv_cache);
472}
473
474int
475event_base_update_cache_time(struct event_base *base)
476{
477
478	if (!base) {
479		base = current_base;
480		if (!current_base)
481			return -1;
482	}
483
484	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
485	if (base->running_loop)
486		update_time_cache(base);
487	EVBASE_RELEASE_LOCK(base, th_base_lock);
488	return 0;
489}
490
491static inline struct event *
492event_callback_to_event(struct event_callback *evcb)
493{
494	EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_INIT));
495	return EVUTIL_UPCAST(evcb, struct event, ev_evcallback);
496}
497
498static inline struct event_callback *
499event_to_event_callback(struct event *ev)
500{
501	return &ev->ev_evcallback;
502}
503
504struct event_base *
505event_init(void)
506{
507	struct event_base *base = event_base_new_with_config(NULL);
508
509	if (base == NULL) {
510		event_errx(1, "%s: Unable to construct event_base", __func__);
511		return NULL;
512	}
513
514	current_base = base;
515
516	return (base);
517}
518
519struct event_base *
520event_base_new(void)
521{
522	struct event_base *base = NULL;
523	struct event_config *cfg = event_config_new();
524	if (cfg) {
525		base = event_base_new_with_config(cfg);
526		event_config_free(cfg);
527	}
528	return base;
529}
530
531/** Return true iff 'method' is the name of a method that 'cfg' tells us to
532 * avoid. */
533static int
534event_config_is_avoided_method(const struct event_config *cfg,
535    const char *method)
536{
537	struct event_config_entry *entry;
538
539	TAILQ_FOREACH(entry, &cfg->entries, next) {
540		if (entry->avoid_method != NULL &&
541		    strcmp(entry->avoid_method, method) == 0)
542			return (1);
543	}
544
545	return (0);
546}
547
548/** Return true iff 'method' is disabled according to the environment. */
549static int
550event_is_method_disabled(const char *name)
551{
552	char environment[64];
553	int i;
554
555	evutil_snprintf(environment, sizeof(environment), "EVENT_NO%s", name);
556	for (i = 8; environment[i] != '\0'; ++i)
557		environment[i] = EVUTIL_TOUPPER_(environment[i]);
558	/* Note that evutil_getenv_() ignores the environment entirely if
559	 * we're setuid */
560	return (evutil_getenv_(environment) != NULL);
561}
562
563int
564event_base_get_features(const struct event_base *base)
565{
566	return base->evsel->features;
567}
568
569void
570event_enable_debug_mode(void)
571{
572#ifndef EVENT__DISABLE_DEBUG_MODE
573	if (event_debug_mode_on_)
574		event_errx(1, "%s was called twice!", __func__);
575	if (event_debug_mode_too_late)
576		event_errx(1, "%s must be called *before* creating any events "
577		    "or event_bases",__func__);
578
579	event_debug_mode_on_ = 1;
580
581	HT_INIT(event_debug_map, &global_debug_map);
582#endif
583}
584
585void
586event_disable_debug_mode(void)
587{
588#ifndef EVENT__DISABLE_DEBUG_MODE
589	struct event_debug_entry **ent, *victim;
590
591	EVLOCK_LOCK(event_debug_map_lock_, 0);
592	for (ent = HT_START(event_debug_map, &global_debug_map); ent; ) {
593		victim = *ent;
594		ent = HT_NEXT_RMV(event_debug_map, &global_debug_map, ent);
595		mm_free(victim);
596	}
597	HT_CLEAR(event_debug_map, &global_debug_map);
598	EVLOCK_UNLOCK(event_debug_map_lock_ , 0);
599
600	event_debug_mode_on_  = 0;
601#endif
602}
603
604struct event_base *
605event_base_new_with_config(const struct event_config *cfg)
606{
607	int i;
608	struct event_base *base;
609	int should_check_environment;
610
611#ifndef EVENT__DISABLE_DEBUG_MODE
612	event_debug_mode_too_late = 1;
613#endif
614
615	if ((base = mm_calloc(1, sizeof(struct event_base))) == NULL) {
616		event_warn("%s: calloc", __func__);
617		return NULL;
618	}
619
620	if (cfg)
621		base->flags = cfg->flags;
622
623	should_check_environment =
624	    !(cfg && (cfg->flags & EVENT_BASE_FLAG_IGNORE_ENV));
625
626	{
627		struct timeval tmp;
628		int precise_time =
629		    cfg && (cfg->flags & EVENT_BASE_FLAG_PRECISE_TIMER);
630		int flags;
631		if (should_check_environment && !precise_time) {
632			precise_time = evutil_getenv_("EVENT_PRECISE_TIMER") != NULL;
633			if (precise_time) {
634				base->flags |= EVENT_BASE_FLAG_PRECISE_TIMER;
635			}
636		}
637		flags = precise_time ? EV_MONOT_PRECISE : 0;
638		evutil_configure_monotonic_time_(&base->monotonic_timer, flags);
639
640		gettime(base, &tmp);
641	}
642
643	min_heap_ctor_(&base->timeheap);
644
645	base->sig.ev_signal_pair[0] = -1;
646	base->sig.ev_signal_pair[1] = -1;
647	base->th_notify_fd[0] = -1;
648	base->th_notify_fd[1] = -1;
649
650	TAILQ_INIT(&base->active_later_queue);
651
652	evmap_io_initmap_(&base->io);
653	evmap_signal_initmap_(&base->sigmap);
654	event_changelist_init_(&base->changelist);
655
656	base->evbase = NULL;
657
658	if (cfg) {
659		memcpy(&base->max_dispatch_time,
660		    &cfg->max_dispatch_interval, sizeof(struct timeval));
661		base->limit_callbacks_after_prio =
662		    cfg->limit_callbacks_after_prio;
663	} else {
664		base->max_dispatch_time.tv_sec = -1;
665		base->limit_callbacks_after_prio = 1;
666	}
667	if (cfg && cfg->max_dispatch_callbacks >= 0) {
668		base->max_dispatch_callbacks = cfg->max_dispatch_callbacks;
669	} else {
670		base->max_dispatch_callbacks = INT_MAX;
671	}
672	if (base->max_dispatch_callbacks == INT_MAX &&
673	    base->max_dispatch_time.tv_sec == -1)
674		base->limit_callbacks_after_prio = INT_MAX;
675
676	for (i = 0; eventops[i] && !base->evbase; i++) {
677		if (cfg != NULL) {
678			/* determine if this backend should be avoided */
679			if (event_config_is_avoided_method(cfg,
680				eventops[i]->name))
681				continue;
682			if ((eventops[i]->features & cfg->require_features)
683			    != cfg->require_features)
684				continue;
685		}
686
687		/* also obey the environment variables */
688		if (should_check_environment &&
689		    event_is_method_disabled(eventops[i]->name))
690			continue;
691
692		base->evsel = eventops[i];
693
694		base->evbase = base->evsel->init(base);
695	}
696
697	if (base->evbase == NULL) {
698		event_warnx("%s: no event mechanism available",
699		    __func__);
700		base->evsel = NULL;
701		event_base_free(base);
702		return NULL;
703	}
704
705	if (evutil_getenv_("EVENT_SHOW_METHOD"))
706		event_msgx("libevent using: %s", base->evsel->name);
707
708	/* allocate a single active event queue */
709	if (event_base_priority_init(base, 1) < 0) {
710		event_base_free(base);
711		return NULL;
712	}
713
714	/* prepare for threading */
715
716#if !defined(EVENT__DISABLE_THREAD_SUPPORT) && !defined(EVENT__DISABLE_DEBUG_MODE)
717	event_debug_created_threadable_ctx_ = 1;
718#endif
719
720#ifndef EVENT__DISABLE_THREAD_SUPPORT
721	if (EVTHREAD_LOCKING_ENABLED() &&
722	    (!cfg || !(cfg->flags & EVENT_BASE_FLAG_NOLOCK))) {
723		int r;
724		EVTHREAD_ALLOC_LOCK(base->th_base_lock, 0);
725		EVTHREAD_ALLOC_COND(base->current_event_cond);
726		r = evthread_make_base_notifiable(base);
727		if (r<0) {
728			event_warnx("%s: Unable to make base notifiable.", __func__);
729			event_base_free(base);
730			return NULL;
731		}
732	}
733#endif
734
735#ifdef _WIN32
736	if (cfg && (cfg->flags & EVENT_BASE_FLAG_STARTUP_IOCP))
737		event_base_start_iocp_(base, cfg->n_cpus_hint);
738#endif
739
740	return (base);
741}
742
743int
744event_base_start_iocp_(struct event_base *base, int n_cpus)
745{
746#ifdef _WIN32
747	if (base->iocp)
748		return 0;
749	base->iocp = event_iocp_port_launch_(n_cpus);
750	if (!base->iocp) {
751		event_warnx("%s: Couldn't launch IOCP", __func__);
752		return -1;
753	}
754	return 0;
755#else
756	return -1;
757#endif
758}
759
760void
761event_base_stop_iocp_(struct event_base *base)
762{
763#ifdef _WIN32
764	int rv;
765
766	if (!base->iocp)
767		return;
768	rv = event_iocp_shutdown_(base->iocp, -1);
769	EVUTIL_ASSERT(rv >= 0);
770	base->iocp = NULL;
771#endif
772}
773
774static int
775event_base_cancel_single_callback_(struct event_base *base,
776    struct event_callback *evcb,
777    int run_finalizers)
778{
779	int result = 0;
780
781	if (evcb->evcb_flags & EVLIST_INIT) {
782		struct event *ev = event_callback_to_event(evcb);
783		if (!(ev->ev_flags & EVLIST_INTERNAL)) {
784			event_del_(ev, EVENT_DEL_EVEN_IF_FINALIZING);
785			result = 1;
786		}
787	} else {
788		EVBASE_ACQUIRE_LOCK(base, th_base_lock);
789		event_callback_cancel_nolock_(base, evcb, 1);
790		EVBASE_RELEASE_LOCK(base, th_base_lock);
791		result = 1;
792	}
793
794	if (run_finalizers && (evcb->evcb_flags & EVLIST_FINALIZING)) {
795		switch (evcb->evcb_closure) {
796		case EV_CLOSURE_EVENT_FINALIZE:
797		case EV_CLOSURE_EVENT_FINALIZE_FREE: {
798			struct event *ev = event_callback_to_event(evcb);
799			ev->ev_evcallback.evcb_cb_union.evcb_evfinalize(ev, ev->ev_arg);
800			if (evcb->evcb_closure == EV_CLOSURE_EVENT_FINALIZE_FREE)
801				mm_free(ev);
802			break;
803		}
804		case EV_CLOSURE_CB_FINALIZE:
805			evcb->evcb_cb_union.evcb_cbfinalize(evcb, evcb->evcb_arg);
806			break;
807		default:
808			break;
809		}
810	}
811	return result;
812}
813
814static int event_base_free_queues_(struct event_base *base, int run_finalizers)
815{
816	int deleted = 0, i;
817
818	for (i = 0; i < base->nactivequeues; ++i) {
819		struct event_callback *evcb, *next;
820		for (evcb = TAILQ_FIRST(&base->activequeues[i]); evcb; ) {
821			next = TAILQ_NEXT(evcb, evcb_active_next);
822			deleted += event_base_cancel_single_callback_(base, evcb, run_finalizers);
823			evcb = next;
824		}
825	}
826
827	{
828		struct event_callback *evcb;
829		while ((evcb = TAILQ_FIRST(&base->active_later_queue))) {
830			deleted += event_base_cancel_single_callback_(base, evcb, run_finalizers);
831		}
832	}
833
834	return deleted;
835}
836
837static void
838event_base_free_(struct event_base *base, int run_finalizers)
839{
840	int i, n_deleted=0;
841	struct event *ev;
842	/* XXXX grab the lock? If there is contention when one thread frees
843	 * the base, then the contending thread will be very sad soon. */
844
845	/* event_base_free(NULL) is how to free the current_base if we
846	 * made it with event_init and forgot to hold a reference to it. */
847	if (base == NULL && current_base)
848		base = current_base;
849	/* Don't actually free NULL. */
850	if (base == NULL) {
851		event_warnx("%s: no base to free", __func__);
852		return;
853	}
854	/* XXX(niels) - check for internal events first */
855
856#ifdef _WIN32
857	event_base_stop_iocp_(base);
858#endif
859
860	/* threading fds if we have them */
861	if (base->th_notify_fd[0] != -1) {
862		event_del(&base->th_notify);
863		EVUTIL_CLOSESOCKET(base->th_notify_fd[0]);
864		if (base->th_notify_fd[1] != -1)
865			EVUTIL_CLOSESOCKET(base->th_notify_fd[1]);
866		base->th_notify_fd[0] = -1;
867		base->th_notify_fd[1] = -1;
868		event_debug_unassign(&base->th_notify);
869	}
870
871	/* Delete all non-internal events. */
872	evmap_delete_all_(base);
873
874	while ((ev = min_heap_top_(&base->timeheap)) != NULL) {
875		event_del(ev);
876		++n_deleted;
877	}
878	for (i = 0; i < base->n_common_timeouts; ++i) {
879		struct common_timeout_list *ctl =
880		    base->common_timeout_queues[i];
881		event_del(&ctl->timeout_event); /* Internal; doesn't count */
882		event_debug_unassign(&ctl->timeout_event);
883		for (ev = TAILQ_FIRST(&ctl->events); ev; ) {
884			struct event *next = TAILQ_NEXT(ev,
885			    ev_timeout_pos.ev_next_with_common_timeout);
886			if (!(ev->ev_flags & EVLIST_INTERNAL)) {
887				event_del(ev);
888				++n_deleted;
889			}
890			ev = next;
891		}
892		mm_free(ctl);
893	}
894	if (base->common_timeout_queues)
895		mm_free(base->common_timeout_queues);
896
897	for (;;) {
898		/* For finalizers we can register yet another finalizer out from
899		 * finalizer, and iff finalizer will be in active_later_queue we can
900		 * add finalizer to activequeues, and we will have events in
901		 * activequeues after this function returns, which is not what we want
902		 * (we even have an assertion for this).
903		 *
904		 * A simple case is bufferevent with underlying (i.e. filters).
905		 */
906		int i = event_base_free_queues_(base, run_finalizers);
907		event_debug(("%s: %d events freed", __func__, i));
908		if (!i) {
909			break;
910		}
911		n_deleted += i;
912	}
913
914	if (n_deleted)
915		event_debug(("%s: %d events were still set in base",
916			__func__, n_deleted));
917
918	while (LIST_FIRST(&base->once_events)) {
919		struct event_once *eonce = LIST_FIRST(&base->once_events);
920		LIST_REMOVE(eonce, next_once);
921		mm_free(eonce);
922	}
923
924	if (base->evsel != NULL && base->evsel->dealloc != NULL)
925		base->evsel->dealloc(base);
926
927	for (i = 0; i < base->nactivequeues; ++i)
928		EVUTIL_ASSERT(TAILQ_EMPTY(&base->activequeues[i]));
929
930	EVUTIL_ASSERT(min_heap_empty_(&base->timeheap));
931	min_heap_dtor_(&base->timeheap);
932
933	mm_free(base->activequeues);
934
935	evmap_io_clear_(&base->io);
936	evmap_signal_clear_(&base->sigmap);
937	event_changelist_freemem_(&base->changelist);
938
939	EVTHREAD_FREE_LOCK(base->th_base_lock, 0);
940	EVTHREAD_FREE_COND(base->current_event_cond);
941
942	/* If we're freeing current_base, there won't be a current_base. */
943	if (base == current_base)
944		current_base = NULL;
945	mm_free(base);
946}
947
948void
949event_base_free_nofinalize(struct event_base *base)
950{
951	event_base_free_(base, 0);
952}
953
954void
955event_base_free(struct event_base *base)
956{
957	event_base_free_(base, 1);
958}
959
960/* Fake eventop; used to disable the backend temporarily inside event_reinit
961 * so that we can call event_del() on an event without telling the backend.
962 */
963static int
964nil_backend_del(struct event_base *b, evutil_socket_t fd, short old,
965    short events, void *fdinfo)
966{
967	return 0;
968}
969const struct eventop nil_eventop = {
970	"nil",
971	NULL, /* init: unused. */
972	NULL, /* add: unused. */
973	nil_backend_del, /* del: used, so needs to be killed. */
974	NULL, /* dispatch: unused. */
975	NULL, /* dealloc: unused. */
976	0, 0, 0
977};
978
979/* reinitialize the event base after a fork */
980int
981event_reinit(struct event_base *base)
982{
983	const struct eventop *evsel;
984	int res = 0;
985	int was_notifiable = 0;
986	int had_signal_added = 0;
987
988	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
989
990	evsel = base->evsel;
991
992	/* check if this event mechanism requires reinit on the backend */
993	if (evsel->need_reinit) {
994		/* We're going to call event_del() on our notify events (the
995		 * ones that tell about signals and wakeup events).  But we
996		 * don't actually want to tell the backend to change its
997		 * state, since it might still share some resource (a kqueue,
998		 * an epoll fd) with the parent process, and we don't want to
999		 * delete the fds from _that_ backend, we temporarily stub out
1000		 * the evsel with a replacement.
1001		 */
1002		base->evsel = &nil_eventop;
1003	}
1004
1005	/* We need to re-create a new signal-notification fd and a new
1006	 * thread-notification fd.  Otherwise, we'll still share those with
1007	 * the parent process, which would make any notification sent to them
1008	 * get received by one or both of the event loops, more or less at
1009	 * random.
1010	 */
1011	if (base->sig.ev_signal_added) {
1012		event_del_nolock_(&base->sig.ev_signal, EVENT_DEL_AUTOBLOCK);
1013		event_debug_unassign(&base->sig.ev_signal);
1014		memset(&base->sig.ev_signal, 0, sizeof(base->sig.ev_signal));
1015		had_signal_added = 1;
1016		base->sig.ev_signal_added = 0;
1017	}
1018	if (base->sig.ev_signal_pair[0] != -1)
1019		EVUTIL_CLOSESOCKET(base->sig.ev_signal_pair[0]);
1020	if (base->sig.ev_signal_pair[1] != -1)
1021		EVUTIL_CLOSESOCKET(base->sig.ev_signal_pair[1]);
1022	if (base->th_notify_fn != NULL) {
1023		was_notifiable = 1;
1024		base->th_notify_fn = NULL;
1025	}
1026	if (base->th_notify_fd[0] != -1) {
1027		event_del_nolock_(&base->th_notify, EVENT_DEL_AUTOBLOCK);
1028		EVUTIL_CLOSESOCKET(base->th_notify_fd[0]);
1029		if (base->th_notify_fd[1] != -1)
1030			EVUTIL_CLOSESOCKET(base->th_notify_fd[1]);
1031		base->th_notify_fd[0] = -1;
1032		base->th_notify_fd[1] = -1;
1033		event_debug_unassign(&base->th_notify);
1034	}
1035
1036	/* Replace the original evsel. */
1037        base->evsel = evsel;
1038
1039	if (evsel->need_reinit) {
1040		/* Reconstruct the backend through brute-force, so that we do
1041		 * not share any structures with the parent process. For some
1042		 * backends, this is necessary: epoll and kqueue, for
1043		 * instance, have events associated with a kernel
1044		 * structure. If didn't reinitialize, we'd share that
1045		 * structure with the parent process, and any changes made by
1046		 * the parent would affect our backend's behavior (and vice
1047		 * versa).
1048		 */
1049		if (base->evsel->dealloc != NULL)
1050			base->evsel->dealloc(base);
1051		base->evbase = evsel->init(base);
1052		if (base->evbase == NULL) {
1053			event_errx(1,
1054			   "%s: could not reinitialize event mechanism",
1055			   __func__);
1056			res = -1;
1057			goto done;
1058		}
1059
1060		/* Empty out the changelist (if any): we are starting from a
1061		 * blank slate. */
1062		event_changelist_freemem_(&base->changelist);
1063
1064		/* Tell the event maps to re-inform the backend about all
1065		 * pending events. This will make the signal notification
1066		 * event get re-created if necessary. */
1067		if (evmap_reinit_(base) < 0)
1068			res = -1;
1069	} else {
1070		res = evsig_init_(base);
1071		if (res == 0 && had_signal_added) {
1072			res = event_add_nolock_(&base->sig.ev_signal, NULL, 0);
1073			if (res == 0)
1074				base->sig.ev_signal_added = 1;
1075		}
1076	}
1077
1078	/* If we were notifiable before, and nothing just exploded, become
1079	 * notifiable again. */
1080	if (was_notifiable && res == 0)
1081		res = evthread_make_base_notifiable_nolock_(base);
1082
1083done:
1084	EVBASE_RELEASE_LOCK(base, th_base_lock);
1085	return (res);
1086}
1087
1088/* Get the monotonic time for this event_base' timer */
1089int
1090event_gettime_monotonic(struct event_base *base, struct timeval *tv)
1091{
1092  int rv = -1;
1093
1094  if (base && tv) {
1095    EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1096    rv = evutil_gettime_monotonic_(&(base->monotonic_timer), tv);
1097    EVBASE_RELEASE_LOCK(base, th_base_lock);
1098  }
1099
1100  return rv;
1101}
1102
1103const char **
1104event_get_supported_methods(void)
1105{
1106	static const char **methods = NULL;
1107	const struct eventop **method;
1108	const char **tmp;
1109	int i = 0, k;
1110
1111	/* count all methods */
1112	for (method = &eventops[0]; *method != NULL; ++method) {
1113		++i;
1114	}
1115
1116	/* allocate one more than we need for the NULL pointer */
1117	tmp = mm_calloc((i + 1), sizeof(char *));
1118	if (tmp == NULL)
1119		return (NULL);
1120
1121	/* populate the array with the supported methods */
1122	for (k = 0, i = 0; eventops[k] != NULL; ++k) {
1123		tmp[i++] = eventops[k]->name;
1124	}
1125	tmp[i] = NULL;
1126
1127	if (methods != NULL)
1128		mm_free((char**)methods);
1129
1130	methods = tmp;
1131
1132	return (methods);
1133}
1134
1135struct event_config *
1136event_config_new(void)
1137{
1138	struct event_config *cfg = mm_calloc(1, sizeof(*cfg));
1139
1140	if (cfg == NULL)
1141		return (NULL);
1142
1143	TAILQ_INIT(&cfg->entries);
1144	cfg->max_dispatch_interval.tv_sec = -1;
1145	cfg->max_dispatch_callbacks = INT_MAX;
1146	cfg->limit_callbacks_after_prio = 1;
1147
1148	return (cfg);
1149}
1150
1151static void
1152event_config_entry_free(struct event_config_entry *entry)
1153{
1154	if (entry->avoid_method != NULL)
1155		mm_free((char *)entry->avoid_method);
1156	mm_free(entry);
1157}
1158
1159void
1160event_config_free(struct event_config *cfg)
1161{
1162	struct event_config_entry *entry;
1163
1164	while ((entry = TAILQ_FIRST(&cfg->entries)) != NULL) {
1165		TAILQ_REMOVE(&cfg->entries, entry, next);
1166		event_config_entry_free(entry);
1167	}
1168	mm_free(cfg);
1169}
1170
1171int
1172event_config_set_flag(struct event_config *cfg, int flag)
1173{
1174	if (!cfg)
1175		return -1;
1176	cfg->flags |= flag;
1177	return 0;
1178}
1179
1180int
1181event_config_avoid_method(struct event_config *cfg, const char *method)
1182{
1183	struct event_config_entry *entry = mm_malloc(sizeof(*entry));
1184	if (entry == NULL)
1185		return (-1);
1186
1187	if ((entry->avoid_method = mm_strdup(method)) == NULL) {
1188		mm_free(entry);
1189		return (-1);
1190	}
1191
1192	TAILQ_INSERT_TAIL(&cfg->entries, entry, next);
1193
1194	return (0);
1195}
1196
1197int
1198event_config_require_features(struct event_config *cfg,
1199    int features)
1200{
1201	if (!cfg)
1202		return (-1);
1203	cfg->require_features = features;
1204	return (0);
1205}
1206
1207int
1208event_config_set_num_cpus_hint(struct event_config *cfg, int cpus)
1209{
1210	if (!cfg)
1211		return (-1);
1212	cfg->n_cpus_hint = cpus;
1213	return (0);
1214}
1215
1216int
1217event_config_set_max_dispatch_interval(struct event_config *cfg,
1218    const struct timeval *max_interval, int max_callbacks, int min_priority)
1219{
1220	if (max_interval)
1221		memcpy(&cfg->max_dispatch_interval, max_interval,
1222		    sizeof(struct timeval));
1223	else
1224		cfg->max_dispatch_interval.tv_sec = -1;
1225	cfg->max_dispatch_callbacks =
1226	    max_callbacks >= 0 ? max_callbacks : INT_MAX;
1227	if (min_priority < 0)
1228		min_priority = 0;
1229	cfg->limit_callbacks_after_prio = min_priority;
1230	return (0);
1231}
1232
1233int
1234event_priority_init(int npriorities)
1235{
1236	return event_base_priority_init(current_base, npriorities);
1237}
1238
1239int
1240event_base_priority_init(struct event_base *base, int npriorities)
1241{
1242	int i, r;
1243	r = -1;
1244
1245	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1246
1247	if (N_ACTIVE_CALLBACKS(base) || npriorities < 1
1248	    || npriorities >= EVENT_MAX_PRIORITIES)
1249		goto err;
1250
1251	if (npriorities == base->nactivequeues)
1252		goto ok;
1253
1254	if (base->nactivequeues) {
1255		mm_free(base->activequeues);
1256		base->nactivequeues = 0;
1257	}
1258
1259	/* Allocate our priority queues */
1260	base->activequeues = (struct evcallback_list *)
1261	  mm_calloc(npriorities, sizeof(struct evcallback_list));
1262	if (base->activequeues == NULL) {
1263		event_warn("%s: calloc", __func__);
1264		goto err;
1265	}
1266	base->nactivequeues = npriorities;
1267
1268	for (i = 0; i < base->nactivequeues; ++i) {
1269		TAILQ_INIT(&base->activequeues[i]);
1270	}
1271
1272ok:
1273	r = 0;
1274err:
1275	EVBASE_RELEASE_LOCK(base, th_base_lock);
1276	return (r);
1277}
1278
1279int
1280event_base_get_npriorities(struct event_base *base)
1281{
1282
1283	int n;
1284	if (base == NULL)
1285		base = current_base;
1286
1287	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1288	n = base->nactivequeues;
1289	EVBASE_RELEASE_LOCK(base, th_base_lock);
1290	return (n);
1291}
1292
1293int
1294event_base_get_num_events(struct event_base *base, unsigned int type)
1295{
1296	int r = 0;
1297
1298	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1299
1300	if (type & EVENT_BASE_COUNT_ACTIVE)
1301		r += base->event_count_active;
1302
1303	if (type & EVENT_BASE_COUNT_VIRTUAL)
1304		r += base->virtual_event_count;
1305
1306	if (type & EVENT_BASE_COUNT_ADDED)
1307		r += base->event_count;
1308
1309	EVBASE_RELEASE_LOCK(base, th_base_lock);
1310
1311	return r;
1312}
1313
1314int
1315event_base_get_max_events(struct event_base *base, unsigned int type, int clear)
1316{
1317	int r = 0;
1318
1319	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1320
1321	if (type & EVENT_BASE_COUNT_ACTIVE) {
1322		r += base->event_count_active_max;
1323		if (clear)
1324			base->event_count_active_max = 0;
1325	}
1326
1327	if (type & EVENT_BASE_COUNT_VIRTUAL) {
1328		r += base->virtual_event_count_max;
1329		if (clear)
1330			base->virtual_event_count_max = 0;
1331	}
1332
1333	if (type & EVENT_BASE_COUNT_ADDED) {
1334		r += base->event_count_max;
1335		if (clear)
1336			base->event_count_max = 0;
1337	}
1338
1339	EVBASE_RELEASE_LOCK(base, th_base_lock);
1340
1341	return r;
1342}
1343
1344/* Returns true iff we're currently watching any events. */
1345static int
1346event_haveevents(struct event_base *base)
1347{
1348	/* Caller must hold th_base_lock */
1349	return (base->virtual_event_count > 0 || base->event_count > 0);
1350}
1351
1352/* "closure" function called when processing active signal events */
1353static inline void
1354event_signal_closure(struct event_base *base, struct event *ev)
1355{
1356	short ncalls;
1357	int should_break;
1358
1359	/* Allows deletes to work */
1360	ncalls = ev->ev_ncalls;
1361	if (ncalls != 0)
1362		ev->ev_pncalls = &ncalls;
1363	EVBASE_RELEASE_LOCK(base, th_base_lock);
1364	while (ncalls) {
1365		ncalls--;
1366		ev->ev_ncalls = ncalls;
1367		if (ncalls == 0)
1368			ev->ev_pncalls = NULL;
1369		(*ev->ev_callback)(ev->ev_fd, ev->ev_res, ev->ev_arg);
1370
1371		EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1372		should_break = base->event_break;
1373		EVBASE_RELEASE_LOCK(base, th_base_lock);
1374
1375		if (should_break) {
1376			if (ncalls != 0)
1377				ev->ev_pncalls = NULL;
1378			return;
1379		}
1380	}
1381}
1382
1383/* Common timeouts are special timeouts that are handled as queues rather than
1384 * in the minheap.  This is more efficient than the minheap if we happen to
1385 * know that we're going to get several thousands of timeout events all with
1386 * the same timeout value.
1387 *
1388 * Since all our timeout handling code assumes timevals can be copied,
1389 * assigned, etc, we can't use "magic pointer" to encode these common
1390 * timeouts.  Searching through a list to see if every timeout is common could
1391 * also get inefficient.  Instead, we take advantage of the fact that tv_usec
1392 * is 32 bits long, but only uses 20 of those bits (since it can never be over
1393 * 999999.)  We use the top bits to encode 4 bites of magic number, and 8 bits
1394 * of index into the event_base's aray of common timeouts.
1395 */
1396
1397#define MICROSECONDS_MASK       COMMON_TIMEOUT_MICROSECONDS_MASK
1398#define COMMON_TIMEOUT_IDX_MASK 0x0ff00000
1399#define COMMON_TIMEOUT_IDX_SHIFT 20
1400#define COMMON_TIMEOUT_MASK     0xf0000000
1401#define COMMON_TIMEOUT_MAGIC    0x50000000
1402
1403#define COMMON_TIMEOUT_IDX(tv) \
1404	(((tv)->tv_usec & COMMON_TIMEOUT_IDX_MASK)>>COMMON_TIMEOUT_IDX_SHIFT)
1405
1406/** Return true iff if 'tv' is a common timeout in 'base' */
1407static inline int
1408is_common_timeout(const struct timeval *tv,
1409    const struct event_base *base)
1410{
1411	int idx;
1412	if ((tv->tv_usec & COMMON_TIMEOUT_MASK) != COMMON_TIMEOUT_MAGIC)
1413		return 0;
1414	idx = COMMON_TIMEOUT_IDX(tv);
1415	return idx < base->n_common_timeouts;
1416}
1417
1418/* True iff tv1 and tv2 have the same common-timeout index, or if neither
1419 * one is a common timeout. */
1420static inline int
1421is_same_common_timeout(const struct timeval *tv1, const struct timeval *tv2)
1422{
1423	return (tv1->tv_usec & ~MICROSECONDS_MASK) ==
1424	    (tv2->tv_usec & ~MICROSECONDS_MASK);
1425}
1426
1427/** Requires that 'tv' is a common timeout.  Return the corresponding
1428 * common_timeout_list. */
1429static inline struct common_timeout_list *
1430get_common_timeout_list(struct event_base *base, const struct timeval *tv)
1431{
1432	return base->common_timeout_queues[COMMON_TIMEOUT_IDX(tv)];
1433}
1434
1435#if 0
1436static inline int
1437common_timeout_ok(const struct timeval *tv,
1438    struct event_base *base)
1439{
1440	const struct timeval *expect =
1441	    &get_common_timeout_list(base, tv)->duration;
1442	return tv->tv_sec == expect->tv_sec &&
1443	    tv->tv_usec == expect->tv_usec;
1444}
1445#endif
1446
1447/* Add the timeout for the first event in given common timeout list to the
1448 * event_base's minheap. */
1449static void
1450common_timeout_schedule(struct common_timeout_list *ctl,
1451    const struct timeval *now, struct event *head)
1452{
1453	struct timeval timeout = head->ev_timeout;
1454	timeout.tv_usec &= MICROSECONDS_MASK;
1455	event_add_nolock_(&ctl->timeout_event, &timeout, 1);
1456}
1457
1458/* Callback: invoked when the timeout for a common timeout queue triggers.
1459 * This means that (at least) the first event in that queue should be run,
1460 * and the timeout should be rescheduled if there are more events. */
1461static void
1462common_timeout_callback(evutil_socket_t fd, short what, void *arg)
1463{
1464	struct timeval now;
1465	struct common_timeout_list *ctl = arg;
1466	struct event_base *base = ctl->base;
1467	struct event *ev = NULL;
1468	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1469	gettime(base, &now);
1470	while (1) {
1471		ev = TAILQ_FIRST(&ctl->events);
1472		if (!ev || ev->ev_timeout.tv_sec > now.tv_sec ||
1473		    (ev->ev_timeout.tv_sec == now.tv_sec &&
1474			(ev->ev_timeout.tv_usec&MICROSECONDS_MASK) > now.tv_usec))
1475			break;
1476		event_del_nolock_(ev, EVENT_DEL_NOBLOCK);
1477		event_active_nolock_(ev, EV_TIMEOUT, 1);
1478	}
1479	if (ev)
1480		common_timeout_schedule(ctl, &now, ev);
1481	EVBASE_RELEASE_LOCK(base, th_base_lock);
1482}
1483
1484#define MAX_COMMON_TIMEOUTS 256
1485
1486const struct timeval *
1487event_base_init_common_timeout(struct event_base *base,
1488    const struct timeval *duration)
1489{
1490	int i;
1491	struct timeval tv;
1492	const struct timeval *result=NULL;
1493	struct common_timeout_list *new_ctl;
1494
1495	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1496	if (duration->tv_usec > 1000000) {
1497		memcpy(&tv, duration, sizeof(struct timeval));
1498		if (is_common_timeout(duration, base))
1499			tv.tv_usec &= MICROSECONDS_MASK;
1500		tv.tv_sec += tv.tv_usec / 1000000;
1501		tv.tv_usec %= 1000000;
1502		duration = &tv;
1503	}
1504	for (i = 0; i < base->n_common_timeouts; ++i) {
1505		const struct common_timeout_list *ctl =
1506		    base->common_timeout_queues[i];
1507		if (duration->tv_sec == ctl->duration.tv_sec &&
1508		    duration->tv_usec ==
1509		    (ctl->duration.tv_usec & MICROSECONDS_MASK)) {
1510			EVUTIL_ASSERT(is_common_timeout(&ctl->duration, base));
1511			result = &ctl->duration;
1512			goto done;
1513		}
1514	}
1515	if (base->n_common_timeouts == MAX_COMMON_TIMEOUTS) {
1516		event_warnx("%s: Too many common timeouts already in use; "
1517		    "we only support %d per event_base", __func__,
1518		    MAX_COMMON_TIMEOUTS);
1519		goto done;
1520	}
1521	if (base->n_common_timeouts_allocated == base->n_common_timeouts) {
1522		int n = base->n_common_timeouts < 16 ? 16 :
1523		    base->n_common_timeouts*2;
1524		struct common_timeout_list **newqueues =
1525		    mm_realloc(base->common_timeout_queues,
1526			n*sizeof(struct common_timeout_queue *));
1527		if (!newqueues) {
1528			event_warn("%s: realloc",__func__);
1529			goto done;
1530		}
1531		base->n_common_timeouts_allocated = n;
1532		base->common_timeout_queues = newqueues;
1533	}
1534	new_ctl = mm_calloc(1, sizeof(struct common_timeout_list));
1535	if (!new_ctl) {
1536		event_warn("%s: calloc",__func__);
1537		goto done;
1538	}
1539	TAILQ_INIT(&new_ctl->events);
1540	new_ctl->duration.tv_sec = duration->tv_sec;
1541	new_ctl->duration.tv_usec =
1542	    duration->tv_usec | COMMON_TIMEOUT_MAGIC |
1543	    (base->n_common_timeouts << COMMON_TIMEOUT_IDX_SHIFT);
1544	evtimer_assign(&new_ctl->timeout_event, base,
1545	    common_timeout_callback, new_ctl);
1546	new_ctl->timeout_event.ev_flags |= EVLIST_INTERNAL;
1547	event_priority_set(&new_ctl->timeout_event, 0);
1548	new_ctl->base = base;
1549	base->common_timeout_queues[base->n_common_timeouts++] = new_ctl;
1550	result = &new_ctl->duration;
1551
1552done:
1553	if (result)
1554		EVUTIL_ASSERT(is_common_timeout(result, base));
1555
1556	EVBASE_RELEASE_LOCK(base, th_base_lock);
1557	return result;
1558}
1559
1560/* Closure function invoked when we're activating a persistent event. */
1561static inline void
1562event_persist_closure(struct event_base *base, struct event *ev)
1563{
1564	void (*evcb_callback)(evutil_socket_t, short, void *);
1565
1566        // Other fields of *ev that must be stored before executing
1567        evutil_socket_t evcb_fd;
1568        short evcb_res;
1569        void *evcb_arg;
1570
1571	/* reschedule the persistent event if we have a timeout. */
1572	if (ev->ev_io_timeout.tv_sec || ev->ev_io_timeout.tv_usec) {
1573		/* If there was a timeout, we want it to run at an interval of
1574		 * ev_io_timeout after the last time it was _scheduled_ for,
1575		 * not ev_io_timeout after _now_.  If it fired for another
1576		 * reason, though, the timeout ought to start ticking _now_. */
1577		struct timeval run_at, relative_to, delay, now;
1578		ev_uint32_t usec_mask = 0;
1579		EVUTIL_ASSERT(is_same_common_timeout(&ev->ev_timeout,
1580			&ev->ev_io_timeout));
1581		gettime(base, &now);
1582		if (is_common_timeout(&ev->ev_timeout, base)) {
1583			delay = ev->ev_io_timeout;
1584			usec_mask = delay.tv_usec & ~MICROSECONDS_MASK;
1585			delay.tv_usec &= MICROSECONDS_MASK;
1586			if (ev->ev_res & EV_TIMEOUT) {
1587				relative_to = ev->ev_timeout;
1588				relative_to.tv_usec &= MICROSECONDS_MASK;
1589			} else {
1590				relative_to = now;
1591			}
1592		} else {
1593			delay = ev->ev_io_timeout;
1594			if (ev->ev_res & EV_TIMEOUT) {
1595				relative_to = ev->ev_timeout;
1596			} else {
1597				relative_to = now;
1598			}
1599		}
1600		evutil_timeradd(&relative_to, &delay, &run_at);
1601		if (evutil_timercmp(&run_at, &now, <)) {
1602			/* Looks like we missed at least one invocation due to
1603			 * a clock jump, not running the event loop for a
1604			 * while, really slow callbacks, or
1605			 * something. Reschedule relative to now.
1606			 */
1607			evutil_timeradd(&now, &delay, &run_at);
1608		}
1609		run_at.tv_usec |= usec_mask;
1610		event_add_nolock_(ev, &run_at, 1);
1611	}
1612
1613	// Save our callback before we release the lock
1614	evcb_callback = ev->ev_callback;
1615        evcb_fd = ev->ev_fd;
1616        evcb_res = ev->ev_res;
1617        evcb_arg = ev->ev_arg;
1618
1619	// Release the lock
1620 	EVBASE_RELEASE_LOCK(base, th_base_lock);
1621
1622	// Execute the callback
1623        (evcb_callback)(evcb_fd, evcb_res, evcb_arg);
1624}
1625
1626/*
1627  Helper for event_process_active to process all the events in a single queue,
1628  releasing the lock as we go.  This function requires that the lock be held
1629  when it's invoked.  Returns -1 if we get a signal or an event_break that
1630  means we should stop processing any active events now.  Otherwise returns
1631  the number of non-internal event_callbacks that we processed.
1632*/
1633static int
1634event_process_active_single_queue(struct event_base *base,
1635    struct evcallback_list *activeq,
1636    int max_to_process, const struct timeval *endtime)
1637{
1638	struct event_callback *evcb;
1639	int count = 0;
1640
1641	EVUTIL_ASSERT(activeq != NULL);
1642
1643	for (evcb = TAILQ_FIRST(activeq); evcb; evcb = TAILQ_FIRST(activeq)) {
1644		struct event *ev=NULL;
1645		if (evcb->evcb_flags & EVLIST_INIT) {
1646			ev = event_callback_to_event(evcb);
1647
1648			if (ev->ev_events & EV_PERSIST || ev->ev_flags & EVLIST_FINALIZING)
1649				event_queue_remove_active(base, evcb);
1650			else
1651				event_del_nolock_(ev, EVENT_DEL_NOBLOCK);
1652			event_debug((
1653			    "event_process_active: event: %p, %s%s%scall %p",
1654			    ev,
1655			    ev->ev_res & EV_READ ? "EV_READ " : " ",
1656			    ev->ev_res & EV_WRITE ? "EV_WRITE " : " ",
1657			    ev->ev_res & EV_CLOSED ? "EV_CLOSED " : " ",
1658			    ev->ev_callback));
1659		} else {
1660			event_queue_remove_active(base, evcb);
1661			event_debug(("event_process_active: event_callback %p, "
1662				"closure %d, call %p",
1663				evcb, evcb->evcb_closure, evcb->evcb_cb_union.evcb_callback));
1664		}
1665
1666		if (!(evcb->evcb_flags & EVLIST_INTERNAL))
1667			++count;
1668
1669
1670		base->current_event = evcb;
1671#ifndef EVENT__DISABLE_THREAD_SUPPORT
1672		base->current_event_waiters = 0;
1673#endif
1674
1675		switch (evcb->evcb_closure) {
1676		case EV_CLOSURE_EVENT_SIGNAL:
1677			EVUTIL_ASSERT(ev != NULL);
1678			event_signal_closure(base, ev);
1679			break;
1680		case EV_CLOSURE_EVENT_PERSIST:
1681			EVUTIL_ASSERT(ev != NULL);
1682			event_persist_closure(base, ev);
1683			break;
1684		case EV_CLOSURE_EVENT: {
1685			void (*evcb_callback)(evutil_socket_t, short, void *);
1686			short res;
1687			EVUTIL_ASSERT(ev != NULL);
1688			evcb_callback = *ev->ev_callback;
1689			res = ev->ev_res;
1690			EVBASE_RELEASE_LOCK(base, th_base_lock);
1691			evcb_callback(ev->ev_fd, res, ev->ev_arg);
1692		}
1693		break;
1694		case EV_CLOSURE_CB_SELF: {
1695			void (*evcb_selfcb)(struct event_callback *, void *) = evcb->evcb_cb_union.evcb_selfcb;
1696			EVBASE_RELEASE_LOCK(base, th_base_lock);
1697			evcb_selfcb(evcb, evcb->evcb_arg);
1698		}
1699		break;
1700		case EV_CLOSURE_EVENT_FINALIZE:
1701		case EV_CLOSURE_EVENT_FINALIZE_FREE: {
1702			void (*evcb_evfinalize)(struct event *, void *);
1703			int evcb_closure = evcb->evcb_closure;
1704			EVUTIL_ASSERT(ev != NULL);
1705			base->current_event = NULL;
1706			evcb_evfinalize = ev->ev_evcallback.evcb_cb_union.evcb_evfinalize;
1707			EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_FINALIZING));
1708			EVBASE_RELEASE_LOCK(base, th_base_lock);
1709			event_debug_note_teardown_(ev);
1710			evcb_evfinalize(ev, ev->ev_arg);
1711			if (evcb_closure == EV_CLOSURE_EVENT_FINALIZE_FREE)
1712				mm_free(ev);
1713		}
1714		break;
1715		case EV_CLOSURE_CB_FINALIZE: {
1716			void (*evcb_cbfinalize)(struct event_callback *, void *) = evcb->evcb_cb_union.evcb_cbfinalize;
1717			base->current_event = NULL;
1718			EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_FINALIZING));
1719			EVBASE_RELEASE_LOCK(base, th_base_lock);
1720			evcb_cbfinalize(evcb, evcb->evcb_arg);
1721		}
1722		break;
1723		default:
1724			EVUTIL_ASSERT(0);
1725		}
1726
1727		EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1728		base->current_event = NULL;
1729#ifndef EVENT__DISABLE_THREAD_SUPPORT
1730		if (base->current_event_waiters) {
1731			base->current_event_waiters = 0;
1732			EVTHREAD_COND_BROADCAST(base->current_event_cond);
1733		}
1734#endif
1735
1736		if (base->event_break)
1737			return -1;
1738		if (count >= max_to_process)
1739			return count;
1740		if (count && endtime) {
1741			struct timeval now;
1742			update_time_cache(base);
1743			gettime(base, &now);
1744			if (evutil_timercmp(&now, endtime, >=))
1745				return count;
1746		}
1747		if (base->event_continue)
1748			break;
1749	}
1750	return count;
1751}
1752
1753/*
1754 * Active events are stored in priority queues.  Lower priorities are always
1755 * process before higher priorities.  Low priority events can starve high
1756 * priority ones.
1757 */
1758
1759static int
1760event_process_active(struct event_base *base)
1761{
1762	/* Caller must hold th_base_lock */
1763	struct evcallback_list *activeq = NULL;
1764	int i, c = 0;
1765	const struct timeval *endtime;
1766	struct timeval tv;
1767	const int maxcb = base->max_dispatch_callbacks;
1768	const int limit_after_prio = base->limit_callbacks_after_prio;
1769	if (base->max_dispatch_time.tv_sec >= 0) {
1770		update_time_cache(base);
1771		gettime(base, &tv);
1772		evutil_timeradd(&base->max_dispatch_time, &tv, &tv);
1773		endtime = &tv;
1774	} else {
1775		endtime = NULL;
1776	}
1777
1778	for (i = 0; i < base->nactivequeues; ++i) {
1779		if (TAILQ_FIRST(&base->activequeues[i]) != NULL) {
1780			base->event_running_priority = i;
1781			activeq = &base->activequeues[i];
1782			if (i < limit_after_prio)
1783				c = event_process_active_single_queue(base, activeq,
1784				    INT_MAX, NULL);
1785			else
1786				c = event_process_active_single_queue(base, activeq,
1787				    maxcb, endtime);
1788			if (c < 0) {
1789				goto done;
1790			} else if (c > 0)
1791				break; /* Processed a real event; do not
1792					* consider lower-priority events */
1793			/* If we get here, all of the events we processed
1794			 * were internal.  Continue. */
1795		}
1796	}
1797
1798done:
1799	base->event_running_priority = -1;
1800
1801	return c;
1802}
1803
1804/*
1805 * Wait continuously for events.  We exit only if no events are left.
1806 */
1807
1808int
1809event_dispatch(void)
1810{
1811	return (event_loop(0));
1812}
1813
1814int
1815event_base_dispatch(struct event_base *event_base)
1816{
1817	return (event_base_loop(event_base, 0));
1818}
1819
1820const char *
1821event_base_get_method(const struct event_base *base)
1822{
1823	EVUTIL_ASSERT(base);
1824	return (base->evsel->name);
1825}
1826
1827/** Callback: used to implement event_base_loopexit by telling the event_base
1828 * that it's time to exit its loop. */
1829static void
1830event_loopexit_cb(evutil_socket_t fd, short what, void *arg)
1831{
1832	struct event_base *base = arg;
1833	base->event_gotterm = 1;
1834}
1835
1836int
1837event_loopexit(const struct timeval *tv)
1838{
1839	return (event_once(-1, EV_TIMEOUT, event_loopexit_cb,
1840		    current_base, tv));
1841}
1842
1843int
1844event_base_loopexit(struct event_base *event_base, const struct timeval *tv)
1845{
1846	return (event_base_once(event_base, -1, EV_TIMEOUT, event_loopexit_cb,
1847		    event_base, tv));
1848}
1849
1850int
1851event_loopbreak(void)
1852{
1853	return (event_base_loopbreak(current_base));
1854}
1855
1856int
1857event_base_loopbreak(struct event_base *event_base)
1858{
1859	int r = 0;
1860	if (event_base == NULL)
1861		return (-1);
1862
1863	EVBASE_ACQUIRE_LOCK(event_base, th_base_lock);
1864	event_base->event_break = 1;
1865
1866	if (EVBASE_NEED_NOTIFY(event_base)) {
1867		r = evthread_notify_base(event_base);
1868	} else {
1869		r = (0);
1870	}
1871	EVBASE_RELEASE_LOCK(event_base, th_base_lock);
1872	return r;
1873}
1874
1875int
1876event_base_loopcontinue(struct event_base *event_base)
1877{
1878	int r = 0;
1879	if (event_base == NULL)
1880		return (-1);
1881
1882	EVBASE_ACQUIRE_LOCK(event_base, th_base_lock);
1883	event_base->event_continue = 1;
1884
1885	if (EVBASE_NEED_NOTIFY(event_base)) {
1886		r = evthread_notify_base(event_base);
1887	} else {
1888		r = (0);
1889	}
1890	EVBASE_RELEASE_LOCK(event_base, th_base_lock);
1891	return r;
1892}
1893
1894int
1895event_base_got_break(struct event_base *event_base)
1896{
1897	int res;
1898	EVBASE_ACQUIRE_LOCK(event_base, th_base_lock);
1899	res = event_base->event_break;
1900	EVBASE_RELEASE_LOCK(event_base, th_base_lock);
1901	return res;
1902}
1903
1904int
1905event_base_got_exit(struct event_base *event_base)
1906{
1907	int res;
1908	EVBASE_ACQUIRE_LOCK(event_base, th_base_lock);
1909	res = event_base->event_gotterm;
1910	EVBASE_RELEASE_LOCK(event_base, th_base_lock);
1911	return res;
1912}
1913
1914/* not thread safe */
1915
1916int
1917event_loop(int flags)
1918{
1919	return event_base_loop(current_base, flags);
1920}
1921
1922int
1923event_base_loop(struct event_base *base, int flags)
1924{
1925	const struct eventop *evsel = base->evsel;
1926	struct timeval tv;
1927	struct timeval *tv_p;
1928	int res, done, retval = 0;
1929
1930	/* Grab the lock.  We will release it inside evsel.dispatch, and again
1931	 * as we invoke user callbacks. */
1932	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1933
1934	if (base->running_loop) {
1935		event_warnx("%s: reentrant invocation.  Only one event_base_loop"
1936		    " can run on each event_base at once.", __func__);
1937		EVBASE_RELEASE_LOCK(base, th_base_lock);
1938		return -1;
1939	}
1940
1941	base->running_loop = 1;
1942
1943	clear_time_cache(base);
1944
1945	if (base->sig.ev_signal_added && base->sig.ev_n_signals_added)
1946		evsig_set_base_(base);
1947
1948	done = 0;
1949
1950#ifndef EVENT__DISABLE_THREAD_SUPPORT
1951	base->th_owner_id = EVTHREAD_GET_ID();
1952#endif
1953
1954	base->event_gotterm = base->event_break = 0;
1955
1956	while (!done) {
1957		base->event_continue = 0;
1958		base->n_deferreds_queued = 0;
1959
1960		/* Terminate the loop if we have been asked to */
1961		if (base->event_gotterm) {
1962			break;
1963		}
1964
1965		if (base->event_break) {
1966			break;
1967		}
1968
1969		tv_p = &tv;
1970		if (!N_ACTIVE_CALLBACKS(base) && !(flags & EVLOOP_NONBLOCK)) {
1971			timeout_next(base, &tv_p);
1972		} else {
1973			/*
1974			 * if we have active events, we just poll new events
1975			 * without waiting.
1976			 */
1977			evutil_timerclear(&tv);
1978		}
1979
1980		/* If we have no events, we just exit */
1981		if (0==(flags&EVLOOP_NO_EXIT_ON_EMPTY) &&
1982		    !event_haveevents(base) && !N_ACTIVE_CALLBACKS(base)) {
1983			event_debug(("%s: no events registered.", __func__));
1984			retval = 1;
1985			goto done;
1986		}
1987
1988		event_queue_make_later_events_active(base);
1989
1990		clear_time_cache(base);
1991
1992		res = evsel->dispatch(base, tv_p);
1993
1994		if (res == -1) {
1995			event_debug(("%s: dispatch returned unsuccessfully.",
1996				__func__));
1997			retval = -1;
1998			goto done;
1999		}
2000
2001		update_time_cache(base);
2002
2003		timeout_process(base);
2004
2005		if (N_ACTIVE_CALLBACKS(base)) {
2006			int n = event_process_active(base);
2007			if ((flags & EVLOOP_ONCE)
2008			    && N_ACTIVE_CALLBACKS(base) == 0
2009			    && n != 0)
2010				done = 1;
2011		} else if (flags & EVLOOP_NONBLOCK)
2012			done = 1;
2013	}
2014	event_debug(("%s: asked to terminate loop.", __func__));
2015
2016done:
2017	clear_time_cache(base);
2018	base->running_loop = 0;
2019
2020	EVBASE_RELEASE_LOCK(base, th_base_lock);
2021
2022	return (retval);
2023}
2024
2025/* One-time callback to implement event_base_once: invokes the user callback,
2026 * then deletes the allocated storage */
2027static void
2028event_once_cb(evutil_socket_t fd, short events, void *arg)
2029{
2030	struct event_once *eonce = arg;
2031
2032	(*eonce->cb)(fd, events, eonce->arg);
2033	EVBASE_ACQUIRE_LOCK(eonce->ev.ev_base, th_base_lock);
2034	LIST_REMOVE(eonce, next_once);
2035	EVBASE_RELEASE_LOCK(eonce->ev.ev_base, th_base_lock);
2036	event_debug_unassign(&eonce->ev);
2037	mm_free(eonce);
2038}
2039
2040/* not threadsafe, event scheduled once. */
2041int
2042event_once(evutil_socket_t fd, short events,
2043    void (*callback)(evutil_socket_t, short, void *),
2044    void *arg, const struct timeval *tv)
2045{
2046	return event_base_once(current_base, fd, events, callback, arg, tv);
2047}
2048
2049/* Schedules an event once */
2050int
2051event_base_once(struct event_base *base, evutil_socket_t fd, short events,
2052    void (*callback)(evutil_socket_t, short, void *),
2053    void *arg, const struct timeval *tv)
2054{
2055	struct event_once *eonce;
2056	int res = 0;
2057	int activate = 0;
2058
2059	if (!base)
2060		return (-1);
2061
2062	/* We cannot support signals that just fire once, or persistent
2063	 * events. */
2064	if (events & (EV_SIGNAL|EV_PERSIST))
2065		return (-1);
2066
2067	if ((eonce = mm_calloc(1, sizeof(struct event_once))) == NULL)
2068		return (-1);
2069
2070	eonce->cb = callback;
2071	eonce->arg = arg;
2072
2073	if ((events & (EV_TIMEOUT|EV_SIGNAL|EV_READ|EV_WRITE|EV_CLOSED)) == EV_TIMEOUT) {
2074		evtimer_assign(&eonce->ev, base, event_once_cb, eonce);
2075
2076		if (tv == NULL || ! evutil_timerisset(tv)) {
2077			/* If the event is going to become active immediately,
2078			 * don't put it on the timeout queue.  This is one
2079			 * idiom for scheduling a callback, so let's make
2080			 * it fast (and order-preserving). */
2081			activate = 1;
2082		}
2083	} else if (events & (EV_READ|EV_WRITE|EV_CLOSED)) {
2084		events &= EV_READ|EV_WRITE|EV_CLOSED;
2085
2086		event_assign(&eonce->ev, base, fd, events, event_once_cb, eonce);
2087	} else {
2088		/* Bad event combination */
2089		mm_free(eonce);
2090		return (-1);
2091	}
2092
2093	if (res == 0) {
2094		EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2095		if (activate)
2096			event_active_nolock_(&eonce->ev, EV_TIMEOUT, 1);
2097		else
2098			res = event_add_nolock_(&eonce->ev, tv, 0);
2099
2100		if (res != 0) {
2101			mm_free(eonce);
2102			return (res);
2103		} else {
2104			LIST_INSERT_HEAD(&base->once_events, eonce, next_once);
2105		}
2106		EVBASE_RELEASE_LOCK(base, th_base_lock);
2107	}
2108
2109	return (0);
2110}
2111
2112int
2113event_assign(struct event *ev, struct event_base *base, evutil_socket_t fd, short events, void (*callback)(evutil_socket_t, short, void *), void *arg)
2114{
2115	if (!base)
2116		base = current_base;
2117	if (arg == &event_self_cbarg_ptr_)
2118		arg = ev;
2119
2120	if (!(events & EV_SIGNAL))
2121		event_debug_assert_socket_nonblocking_(fd);
2122	event_debug_assert_not_added_(ev);
2123
2124	ev->ev_base = base;
2125
2126	ev->ev_callback = callback;
2127	ev->ev_arg = arg;
2128	ev->ev_fd = fd;
2129	ev->ev_events = events;
2130	ev->ev_res = 0;
2131	ev->ev_flags = EVLIST_INIT;
2132	ev->ev_ncalls = 0;
2133	ev->ev_pncalls = NULL;
2134
2135	if (events & EV_SIGNAL) {
2136		if ((events & (EV_READ|EV_WRITE|EV_CLOSED)) != 0) {
2137			event_warnx("%s: EV_SIGNAL is not compatible with "
2138			    "EV_READ, EV_WRITE or EV_CLOSED", __func__);
2139			return -1;
2140		}
2141		ev->ev_closure = EV_CLOSURE_EVENT_SIGNAL;
2142	} else {
2143		if (events & EV_PERSIST) {
2144			evutil_timerclear(&ev->ev_io_timeout);
2145			ev->ev_closure = EV_CLOSURE_EVENT_PERSIST;
2146		} else {
2147			ev->ev_closure = EV_CLOSURE_EVENT;
2148		}
2149	}
2150
2151	min_heap_elem_init_(ev);
2152
2153	if (base != NULL) {
2154		/* by default, we put new events into the middle priority */
2155		ev->ev_pri = base->nactivequeues / 2;
2156	}
2157
2158	event_debug_note_setup_(ev);
2159
2160	return 0;
2161}
2162
2163int
2164event_base_set(struct event_base *base, struct event *ev)
2165{
2166	/* Only innocent events may be assigned to a different base */
2167	if (ev->ev_flags != EVLIST_INIT)
2168		return (-1);
2169
2170	event_debug_assert_is_setup_(ev);
2171
2172	ev->ev_base = base;
2173	ev->ev_pri = base->nactivequeues/2;
2174
2175	return (0);
2176}
2177
2178void
2179event_set(struct event *ev, evutil_socket_t fd, short events,
2180	  void (*callback)(evutil_socket_t, short, void *), void *arg)
2181{
2182	int r;
2183	r = event_assign(ev, current_base, fd, events, callback, arg);
2184	EVUTIL_ASSERT(r == 0);
2185}
2186
2187void *
2188event_self_cbarg(void)
2189{
2190	return &event_self_cbarg_ptr_;
2191}
2192
2193struct event *
2194event_base_get_running_event(struct event_base *base)
2195{
2196	struct event *ev = NULL;
2197	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2198	if (EVBASE_IN_THREAD(base)) {
2199		struct event_callback *evcb = base->current_event;
2200		if (evcb->evcb_flags & EVLIST_INIT)
2201			ev = event_callback_to_event(evcb);
2202	}
2203	EVBASE_RELEASE_LOCK(base, th_base_lock);
2204	return ev;
2205}
2206
2207struct event *
2208event_new(struct event_base *base, evutil_socket_t fd, short events, void (*cb)(evutil_socket_t, short, void *), void *arg)
2209{
2210	struct event *ev;
2211	ev = mm_malloc(sizeof(struct event));
2212	if (ev == NULL)
2213		return (NULL);
2214	if (event_assign(ev, base, fd, events, cb, arg) < 0) {
2215		mm_free(ev);
2216		return (NULL);
2217	}
2218
2219	return (ev);
2220}
2221
2222void
2223event_free(struct event *ev)
2224{
2225	/* This is disabled, so that events which have been finalized be a
2226	 * valid target for event_free(). That's */
2227	// event_debug_assert_is_setup_(ev);
2228
2229	/* make sure that this event won't be coming back to haunt us. */
2230	event_del(ev);
2231	event_debug_note_teardown_(ev);
2232	mm_free(ev);
2233
2234}
2235
2236void
2237event_debug_unassign(struct event *ev)
2238{
2239	event_debug_assert_not_added_(ev);
2240	event_debug_note_teardown_(ev);
2241
2242	ev->ev_flags &= ~EVLIST_INIT;
2243}
2244
2245#define EVENT_FINALIZE_FREE_ 0x10000
2246static int
2247event_finalize_nolock_(struct event_base *base, unsigned flags, struct event *ev, event_finalize_callback_fn cb)
2248{
2249	ev_uint8_t closure = (flags & EVENT_FINALIZE_FREE_) ?
2250	    EV_CLOSURE_EVENT_FINALIZE_FREE : EV_CLOSURE_EVENT_FINALIZE;
2251
2252	event_del_nolock_(ev, EVENT_DEL_NOBLOCK);
2253	ev->ev_closure = closure;
2254	ev->ev_evcallback.evcb_cb_union.evcb_evfinalize = cb;
2255	event_active_nolock_(ev, EV_FINALIZE, 1);
2256	ev->ev_flags |= EVLIST_FINALIZING;
2257	return 0;
2258}
2259
2260static int
2261event_finalize_impl_(unsigned flags, struct event *ev, event_finalize_callback_fn cb)
2262{
2263	int r;
2264	struct event_base *base = ev->ev_base;
2265	if (EVUTIL_FAILURE_CHECK(!base)) {
2266		event_warnx("%s: event has no event_base set.", __func__);
2267		return -1;
2268	}
2269
2270	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2271	r = event_finalize_nolock_(base, flags, ev, cb);
2272	EVBASE_RELEASE_LOCK(base, th_base_lock);
2273	return r;
2274}
2275
2276int
2277event_finalize(unsigned flags, struct event *ev, event_finalize_callback_fn cb)
2278{
2279	return event_finalize_impl_(flags, ev, cb);
2280}
2281
2282int
2283event_free_finalize(unsigned flags, struct event *ev, event_finalize_callback_fn cb)
2284{
2285	return event_finalize_impl_(flags|EVENT_FINALIZE_FREE_, ev, cb);
2286}
2287
2288void
2289event_callback_finalize_nolock_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *))
2290{
2291	struct event *ev = NULL;
2292	if (evcb->evcb_flags & EVLIST_INIT) {
2293		ev = event_callback_to_event(evcb);
2294		event_del_nolock_(ev, EVENT_DEL_NOBLOCK);
2295	} else {
2296		event_callback_cancel_nolock_(base, evcb, 0); /*XXX can this fail?*/
2297	}
2298
2299	evcb->evcb_closure = EV_CLOSURE_CB_FINALIZE;
2300	evcb->evcb_cb_union.evcb_cbfinalize = cb;
2301	event_callback_activate_nolock_(base, evcb); /* XXX can this really fail?*/
2302	evcb->evcb_flags |= EVLIST_FINALIZING;
2303}
2304
2305void
2306event_callback_finalize_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *))
2307{
2308	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2309	event_callback_finalize_nolock_(base, flags, evcb, cb);
2310	EVBASE_RELEASE_LOCK(base, th_base_lock);
2311}
2312
2313/** Internal: Finalize all of the n_cbs callbacks in evcbs.  The provided
2314 * callback will be invoked on *one of them*, after they have *all* been
2315 * finalized. */
2316int
2317event_callback_finalize_many_(struct event_base *base, int n_cbs, struct event_callback **evcbs, void (*cb)(struct event_callback *, void *))
2318{
2319	int n_pending = 0, i;
2320
2321	if (base == NULL)
2322		base = current_base;
2323
2324	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2325
2326	event_debug(("%s: %d events finalizing", __func__, n_cbs));
2327
2328	/* At most one can be currently executing; the rest we just
2329	 * cancel... But we always make sure that the finalize callback
2330	 * runs. */
2331	for (i = 0; i < n_cbs; ++i) {
2332		struct event_callback *evcb = evcbs[i];
2333		if (evcb == base->current_event) {
2334			event_callback_finalize_nolock_(base, 0, evcb, cb);
2335			++n_pending;
2336		} else {
2337			event_callback_cancel_nolock_(base, evcb, 0);
2338		}
2339	}
2340
2341	if (n_pending == 0) {
2342		/* Just do the first one. */
2343		event_callback_finalize_nolock_(base, 0, evcbs[0], cb);
2344	}
2345
2346	EVBASE_RELEASE_LOCK(base, th_base_lock);
2347	return 0;
2348}
2349
2350/*
2351 * Set's the priority of an event - if an event is already scheduled
2352 * changing the priority is going to fail.
2353 */
2354
2355int
2356event_priority_set(struct event *ev, int pri)
2357{
2358	event_debug_assert_is_setup_(ev);
2359
2360	if (ev->ev_flags & EVLIST_ACTIVE)
2361		return (-1);
2362	if (pri < 0 || pri >= ev->ev_base->nactivequeues)
2363		return (-1);
2364
2365	ev->ev_pri = pri;
2366
2367	return (0);
2368}
2369
2370/*
2371 * Checks if a specific event is pending or scheduled.
2372 */
2373
2374int
2375event_pending(const struct event *ev, short event, struct timeval *tv)
2376{
2377	int flags = 0;
2378
2379	if (EVUTIL_FAILURE_CHECK(ev->ev_base == NULL)) {
2380		event_warnx("%s: event has no event_base set.", __func__);
2381		return 0;
2382	}
2383
2384	EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock);
2385	event_debug_assert_is_setup_(ev);
2386
2387	if (ev->ev_flags & EVLIST_INSERTED)
2388		flags |= (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL));
2389	if (ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))
2390		flags |= ev->ev_res;
2391	if (ev->ev_flags & EVLIST_TIMEOUT)
2392		flags |= EV_TIMEOUT;
2393
2394	event &= (EV_TIMEOUT|EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL);
2395
2396	/* See if there is a timeout that we should report */
2397	if (tv != NULL && (flags & event & EV_TIMEOUT)) {
2398		struct timeval tmp = ev->ev_timeout;
2399		tmp.tv_usec &= MICROSECONDS_MASK;
2400		/* correctly remamp to real time */
2401		evutil_timeradd(&ev->ev_base->tv_clock_diff, &tmp, tv);
2402	}
2403
2404	EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock);
2405
2406	return (flags & event);
2407}
2408
2409int
2410event_initialized(const struct event *ev)
2411{
2412	if (!(ev->ev_flags & EVLIST_INIT))
2413		return 0;
2414
2415	return 1;
2416}
2417
2418void
2419event_get_assignment(const struct event *event, struct event_base **base_out, evutil_socket_t *fd_out, short *events_out, event_callback_fn *callback_out, void **arg_out)
2420{
2421	event_debug_assert_is_setup_(event);
2422
2423	if (base_out)
2424		*base_out = event->ev_base;
2425	if (fd_out)
2426		*fd_out = event->ev_fd;
2427	if (events_out)
2428		*events_out = event->ev_events;
2429	if (callback_out)
2430		*callback_out = event->ev_callback;
2431	if (arg_out)
2432		*arg_out = event->ev_arg;
2433}
2434
2435size_t
2436event_get_struct_event_size(void)
2437{
2438	return sizeof(struct event);
2439}
2440
2441evutil_socket_t
2442event_get_fd(const struct event *ev)
2443{
2444	event_debug_assert_is_setup_(ev);
2445	return ev->ev_fd;
2446}
2447
2448struct event_base *
2449event_get_base(const struct event *ev)
2450{
2451	event_debug_assert_is_setup_(ev);
2452	return ev->ev_base;
2453}
2454
2455short
2456event_get_events(const struct event *ev)
2457{
2458	event_debug_assert_is_setup_(ev);
2459	return ev->ev_events;
2460}
2461
2462event_callback_fn
2463event_get_callback(const struct event *ev)
2464{
2465	event_debug_assert_is_setup_(ev);
2466	return ev->ev_callback;
2467}
2468
2469void *
2470event_get_callback_arg(const struct event *ev)
2471{
2472	event_debug_assert_is_setup_(ev);
2473	return ev->ev_arg;
2474}
2475
2476int
2477event_get_priority(const struct event *ev)
2478{
2479	event_debug_assert_is_setup_(ev);
2480	return ev->ev_pri;
2481}
2482
2483int
2484event_add(struct event *ev, const struct timeval *tv)
2485{
2486	int res;
2487
2488	if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) {
2489		event_warnx("%s: event has no event_base set.", __func__);
2490		return -1;
2491	}
2492
2493	EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock);
2494
2495	res = event_add_nolock_(ev, tv, 0);
2496
2497	EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock);
2498
2499	return (res);
2500}
2501
2502/* Helper callback: wake an event_base from another thread.  This version
2503 * works by writing a byte to one end of a socketpair, so that the event_base
2504 * listening on the other end will wake up as the corresponding event
2505 * triggers */
2506static int
2507evthread_notify_base_default(struct event_base *base)
2508{
2509	char buf[1];
2510	int r;
2511	buf[0] = (char) 0;
2512#ifdef _WIN32
2513	r = send(base->th_notify_fd[1], buf, 1, 0);
2514#else
2515	r = write(base->th_notify_fd[1], buf, 1);
2516#endif
2517	return (r < 0 && ! EVUTIL_ERR_IS_EAGAIN(errno)) ? -1 : 0;
2518}
2519
2520#ifdef EVENT__HAVE_EVENTFD
2521/* Helper callback: wake an event_base from another thread.  This version
2522 * assumes that you have a working eventfd() implementation. */
2523static int
2524evthread_notify_base_eventfd(struct event_base *base)
2525{
2526	ev_uint64_t msg = 1;
2527	int r;
2528	do {
2529		r = write(base->th_notify_fd[0], (void*) &msg, sizeof(msg));
2530	} while (r < 0 && errno == EAGAIN);
2531
2532	return (r < 0) ? -1 : 0;
2533}
2534#endif
2535
2536
2537/** Tell the thread currently running the event_loop for base (if any) that it
2538 * needs to stop waiting in its dispatch function (if it is) and process all
2539 * active callbacks. */
2540static int
2541evthread_notify_base(struct event_base *base)
2542{
2543	EVENT_BASE_ASSERT_LOCKED(base);
2544	if (!base->th_notify_fn)
2545		return -1;
2546	if (base->is_notify_pending)
2547		return 0;
2548	base->is_notify_pending = 1;
2549	return base->th_notify_fn(base);
2550}
2551
2552/* Implementation function to remove a timeout on a currently pending event.
2553 */
2554int
2555event_remove_timer_nolock_(struct event *ev)
2556{
2557	struct event_base *base = ev->ev_base;
2558
2559	EVENT_BASE_ASSERT_LOCKED(base);
2560	event_debug_assert_is_setup_(ev);
2561
2562	event_debug(("event_remove_timer_nolock: event: %p", ev));
2563
2564	/* If it's not pending on a timeout, we don't need to do anything. */
2565	if (ev->ev_flags & EVLIST_TIMEOUT) {
2566		event_queue_remove_timeout(base, ev);
2567		evutil_timerclear(&ev->ev_.ev_io.ev_timeout);
2568	}
2569
2570	return (0);
2571}
2572
2573int
2574event_remove_timer(struct event *ev)
2575{
2576	int res;
2577
2578	if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) {
2579		event_warnx("%s: event has no event_base set.", __func__);
2580		return -1;
2581	}
2582
2583	EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock);
2584
2585	res = event_remove_timer_nolock_(ev);
2586
2587	EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock);
2588
2589	return (res);
2590}
2591
2592/* Implementation function to add an event.  Works just like event_add,
2593 * except: 1) it requires that we have the lock.  2) if tv_is_absolute is set,
2594 * we treat tv as an absolute time, not as an interval to add to the current
2595 * time */
2596int
2597event_add_nolock_(struct event *ev, const struct timeval *tv,
2598    int tv_is_absolute)
2599{
2600	struct event_base *base = ev->ev_base;
2601	int res = 0;
2602	int notify = 0;
2603
2604	EVENT_BASE_ASSERT_LOCKED(base);
2605	event_debug_assert_is_setup_(ev);
2606
2607	event_debug((
2608		 "event_add: event: %p (fd "EV_SOCK_FMT"), %s%s%s%scall %p",
2609		 ev,
2610		 EV_SOCK_ARG(ev->ev_fd),
2611		 ev->ev_events & EV_READ ? "EV_READ " : " ",
2612		 ev->ev_events & EV_WRITE ? "EV_WRITE " : " ",
2613		 ev->ev_events & EV_CLOSED ? "EV_CLOSED " : " ",
2614		 tv ? "EV_TIMEOUT " : " ",
2615		 ev->ev_callback));
2616
2617	EVUTIL_ASSERT(!(ev->ev_flags & ~EVLIST_ALL));
2618
2619	if (ev->ev_flags & EVLIST_FINALIZING) {
2620		/* XXXX debug */
2621		return (-1);
2622	}
2623
2624	/*
2625	 * prepare for timeout insertion further below, if we get a
2626	 * failure on any step, we should not change any state.
2627	 */
2628	if (tv != NULL && !(ev->ev_flags & EVLIST_TIMEOUT)) {
2629		if (min_heap_reserve_(&base->timeheap,
2630			1 + min_heap_size_(&base->timeheap)) == -1)
2631			return (-1);  /* ENOMEM == errno */
2632	}
2633
2634	/* If the main thread is currently executing a signal event's
2635	 * callback, and we are not the main thread, then we want to wait
2636	 * until the callback is done before we mess with the event, or else
2637	 * we can race on ev_ncalls and ev_pncalls below. */
2638#ifndef EVENT__DISABLE_THREAD_SUPPORT
2639	if (base->current_event == event_to_event_callback(ev) &&
2640	    (ev->ev_events & EV_SIGNAL)
2641	    && !EVBASE_IN_THREAD(base)) {
2642		++base->current_event_waiters;
2643		EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock);
2644	}
2645#endif
2646
2647	if ((ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL)) &&
2648	    !(ev->ev_flags & (EVLIST_INSERTED|EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) {
2649		if (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED))
2650			res = evmap_io_add_(base, ev->ev_fd, ev);
2651		else if (ev->ev_events & EV_SIGNAL)
2652			res = evmap_signal_add_(base, (int)ev->ev_fd, ev);
2653		if (res != -1)
2654			event_queue_insert_inserted(base, ev);
2655		if (res == 1) {
2656			/* evmap says we need to notify the main thread. */
2657			notify = 1;
2658			res = 0;
2659		}
2660	}
2661
2662	/*
2663	 * we should change the timeout state only if the previous event
2664	 * addition succeeded.
2665	 */
2666	if (res != -1 && tv != NULL) {
2667		struct timeval now;
2668		int common_timeout;
2669#ifdef USE_REINSERT_TIMEOUT
2670		int was_common;
2671		int old_timeout_idx;
2672#endif
2673
2674		/*
2675		 * for persistent timeout events, we remember the
2676		 * timeout value and re-add the event.
2677		 *
2678		 * If tv_is_absolute, this was already set.
2679		 */
2680		if (ev->ev_closure == EV_CLOSURE_EVENT_PERSIST && !tv_is_absolute)
2681			ev->ev_io_timeout = *tv;
2682
2683#ifndef USE_REINSERT_TIMEOUT
2684		if (ev->ev_flags & EVLIST_TIMEOUT) {
2685			event_queue_remove_timeout(base, ev);
2686		}
2687#endif
2688
2689		/* Check if it is active due to a timeout.  Rescheduling
2690		 * this timeout before the callback can be executed
2691		 * removes it from the active list. */
2692		if ((ev->ev_flags & EVLIST_ACTIVE) &&
2693		    (ev->ev_res & EV_TIMEOUT)) {
2694			if (ev->ev_events & EV_SIGNAL) {
2695				/* See if we are just active executing
2696				 * this event in a loop
2697				 */
2698				if (ev->ev_ncalls && ev->ev_pncalls) {
2699					/* Abort loop */
2700					*ev->ev_pncalls = 0;
2701				}
2702			}
2703
2704			event_queue_remove_active(base, event_to_event_callback(ev));
2705		}
2706
2707		gettime(base, &now);
2708
2709		common_timeout = is_common_timeout(tv, base);
2710#ifdef USE_REINSERT_TIMEOUT
2711		was_common = is_common_timeout(&ev->ev_timeout, base);
2712		old_timeout_idx = COMMON_TIMEOUT_IDX(&ev->ev_timeout);
2713#endif
2714
2715		if (tv_is_absolute) {
2716			ev->ev_timeout = *tv;
2717		} else if (common_timeout) {
2718			struct timeval tmp = *tv;
2719			tmp.tv_usec &= MICROSECONDS_MASK;
2720			evutil_timeradd(&now, &tmp, &ev->ev_timeout);
2721			ev->ev_timeout.tv_usec |=
2722			    (tv->tv_usec & ~MICROSECONDS_MASK);
2723		} else {
2724			evutil_timeradd(&now, tv, &ev->ev_timeout);
2725		}
2726
2727		event_debug((
2728			 "event_add: event %p, timeout in %d seconds %d useconds, call %p",
2729			 ev, (int)tv->tv_sec, (int)tv->tv_usec, ev->ev_callback));
2730
2731#ifdef USE_REINSERT_TIMEOUT
2732		event_queue_reinsert_timeout(base, ev, was_common, common_timeout, old_timeout_idx);
2733#else
2734		event_queue_insert_timeout(base, ev);
2735#endif
2736
2737		if (common_timeout) {
2738			struct common_timeout_list *ctl =
2739			    get_common_timeout_list(base, &ev->ev_timeout);
2740			if (ev == TAILQ_FIRST(&ctl->events)) {
2741				common_timeout_schedule(ctl, &now, ev);
2742			}
2743		} else {
2744			struct event* top = NULL;
2745			/* See if the earliest timeout is now earlier than it
2746			 * was before: if so, we will need to tell the main
2747			 * thread to wake up earlier than it would otherwise.
2748			 * We double check the timeout of the top element to
2749			 * handle time distortions due to system suspension.
2750			 */
2751			if (min_heap_elt_is_top_(ev))
2752				notify = 1;
2753			else if ((top = min_heap_top_(&base->timeheap)) != NULL &&
2754					 evutil_timercmp(&top->ev_timeout, &now, <))
2755				notify = 1;
2756		}
2757	}
2758
2759	/* if we are not in the right thread, we need to wake up the loop */
2760	if (res != -1 && notify && EVBASE_NEED_NOTIFY(base))
2761		evthread_notify_base(base);
2762
2763	event_debug_note_add_(ev);
2764
2765	return (res);
2766}
2767
2768static int
2769event_del_(struct event *ev, int blocking)
2770{
2771	int res;
2772	struct event_base *base = ev->ev_base;
2773
2774	if (EVUTIL_FAILURE_CHECK(!base)) {
2775		event_warnx("%s: event has no event_base set.", __func__);
2776		return -1;
2777	}
2778
2779	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
2780	res = event_del_nolock_(ev, blocking);
2781	EVBASE_RELEASE_LOCK(base, th_base_lock);
2782
2783	return (res);
2784}
2785
2786int
2787event_del(struct event *ev)
2788{
2789	return event_del_(ev, EVENT_DEL_AUTOBLOCK);
2790}
2791
2792int
2793event_del_block(struct event *ev)
2794{
2795	return event_del_(ev, EVENT_DEL_BLOCK);
2796}
2797
2798int
2799event_del_noblock(struct event *ev)
2800{
2801	return event_del_(ev, EVENT_DEL_NOBLOCK);
2802}
2803
2804/** Helper for event_del: always called with th_base_lock held.
2805 *
2806 * "blocking" must be one of the EVENT_DEL_{BLOCK, NOBLOCK, AUTOBLOCK,
2807 * EVEN_IF_FINALIZING} values. See those for more information.
2808 */
2809int
2810event_del_nolock_(struct event *ev, int blocking)
2811{
2812	struct event_base *base;
2813	int res = 0, notify = 0;
2814
2815	event_debug(("event_del: %p (fd "EV_SOCK_FMT"), callback %p",
2816		ev, EV_SOCK_ARG(ev->ev_fd), ev->ev_callback));
2817
2818	/* An event without a base has not been added */
2819	if (ev->ev_base == NULL)
2820		return (-1);
2821
2822	EVENT_BASE_ASSERT_LOCKED(ev->ev_base);
2823
2824	if (blocking != EVENT_DEL_EVEN_IF_FINALIZING) {
2825		if (ev->ev_flags & EVLIST_FINALIZING) {
2826			/* XXXX Debug */
2827			return 0;
2828		}
2829	}
2830
2831	base = ev->ev_base;
2832
2833	EVUTIL_ASSERT(!(ev->ev_flags & ~EVLIST_ALL));
2834
2835	/* See if we are just active executing this event in a loop */
2836	if (ev->ev_events & EV_SIGNAL) {
2837		if (ev->ev_ncalls && ev->ev_pncalls) {
2838			/* Abort loop */
2839			*ev->ev_pncalls = 0;
2840		}
2841	}
2842
2843	if (ev->ev_flags & EVLIST_TIMEOUT) {
2844		/* NOTE: We never need to notify the main thread because of a
2845		 * deleted timeout event: all that could happen if we don't is
2846		 * that the dispatch loop might wake up too early.  But the
2847		 * point of notifying the main thread _is_ to wake up the
2848		 * dispatch loop early anyway, so we wouldn't gain anything by
2849		 * doing it.
2850		 */
2851		event_queue_remove_timeout(base, ev);
2852	}
2853
2854	if (ev->ev_flags & EVLIST_ACTIVE)
2855		event_queue_remove_active(base, event_to_event_callback(ev));
2856	else if (ev->ev_flags & EVLIST_ACTIVE_LATER)
2857		event_queue_remove_active_later(base, event_to_event_callback(ev));
2858
2859	if (ev->ev_flags & EVLIST_INSERTED) {
2860		event_queue_remove_inserted(base, ev);
2861		if (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED))
2862			res = evmap_io_del_(base, ev->ev_fd, ev);
2863		else
2864			res = evmap_signal_del_(base, (int)ev->ev_fd, ev);
2865		if (res == 1) {
2866			/* evmap says we need to notify the main thread. */
2867			notify = 1;
2868			res = 0;
2869		}
2870		/* If we do not have events, let's notify event base so it can
2871		 * exit without waiting */
2872		if (!event_haveevents(base) && !N_ACTIVE_CALLBACKS(base))
2873			notify = 1;
2874	}
2875
2876	/* if we are not in the right thread, we need to wake up the loop */
2877	if (res != -1 && notify && EVBASE_NEED_NOTIFY(base))
2878		evthread_notify_base(base);
2879
2880	event_debug_note_del_(ev);
2881
2882	/* If the main thread is currently executing this event's callback,
2883	 * and we are not the main thread, then we want to wait until the
2884	 * callback is done before returning. That way, when this function
2885	 * returns, it will be safe to free the user-supplied argument.
2886	 */
2887#ifndef EVENT__DISABLE_THREAD_SUPPORT
2888	if (blocking != EVENT_DEL_NOBLOCK &&
2889	    base->current_event == event_to_event_callback(ev) &&
2890	    !EVBASE_IN_THREAD(base) &&
2891	    (blocking == EVENT_DEL_BLOCK || !(ev->ev_events & EV_FINALIZE))) {
2892		++base->current_event_waiters;
2893		EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock);
2894	}
2895#endif
2896
2897	return (res);
2898}
2899
2900void
2901event_active(struct event *ev, int res, short ncalls)
2902{
2903	if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) {
2904		event_warnx("%s: event has no event_base set.", __func__);
2905		return;
2906	}
2907
2908	EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock);
2909
2910	event_debug_assert_is_setup_(ev);
2911
2912	event_active_nolock_(ev, res, ncalls);
2913
2914	EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock);
2915}
2916
2917
2918void
2919event_active_nolock_(struct event *ev, int res, short ncalls)
2920{
2921	struct event_base *base;
2922
2923	event_debug(("event_active: %p (fd "EV_SOCK_FMT"), res %d, callback %p",
2924		ev, EV_SOCK_ARG(ev->ev_fd), (int)res, ev->ev_callback));
2925
2926	base = ev->ev_base;
2927	EVENT_BASE_ASSERT_LOCKED(base);
2928
2929	if (ev->ev_flags & EVLIST_FINALIZING) {
2930		/* XXXX debug */
2931		return;
2932	}
2933
2934	switch ((ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) {
2935	default:
2936	case EVLIST_ACTIVE|EVLIST_ACTIVE_LATER:
2937		EVUTIL_ASSERT(0);
2938		break;
2939	case EVLIST_ACTIVE:
2940		/* We get different kinds of events, add them together */
2941		ev->ev_res |= res;
2942		return;
2943	case EVLIST_ACTIVE_LATER:
2944		ev->ev_res |= res;
2945		break;
2946	case 0:
2947		ev->ev_res = res;
2948		break;
2949	}
2950
2951	if (ev->ev_pri < base->event_running_priority)
2952		base->event_continue = 1;
2953
2954	if (ev->ev_events & EV_SIGNAL) {
2955#ifndef EVENT__DISABLE_THREAD_SUPPORT
2956		if (base->current_event == event_to_event_callback(ev) &&
2957		    !EVBASE_IN_THREAD(base)) {
2958			++base->current_event_waiters;
2959			EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock);
2960		}
2961#endif
2962		ev->ev_ncalls = ncalls;
2963		ev->ev_pncalls = NULL;
2964	}
2965
2966	event_callback_activate_nolock_(base, event_to_event_callback(ev));
2967}
2968
2969void
2970event_active_later_(struct event *ev, int res)
2971{
2972	EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock);
2973	event_active_later_nolock_(ev, res);
2974	EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock);
2975}
2976
2977void
2978event_active_later_nolock_(struct event *ev, int res)
2979{
2980	struct event_base *base = ev->ev_base;
2981	EVENT_BASE_ASSERT_LOCKED(base);
2982
2983	if (ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) {
2984		/* We get different kinds of events, add them together */
2985		ev->ev_res |= res;
2986		return;
2987	}
2988
2989	ev->ev_res = res;
2990
2991	event_callback_activate_later_nolock_(base, event_to_event_callback(ev));
2992}
2993
2994int
2995event_callback_activate_(struct event_base *base,
2996    struct event_callback *evcb)
2997{
2998	int r;
2999	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3000	r = event_callback_activate_nolock_(base, evcb);
3001	EVBASE_RELEASE_LOCK(base, th_base_lock);
3002	return r;
3003}
3004
3005int
3006event_callback_activate_nolock_(struct event_base *base,
3007    struct event_callback *evcb)
3008{
3009	int r = 1;
3010
3011	if (evcb->evcb_flags & EVLIST_FINALIZING)
3012		return 0;
3013
3014	switch (evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) {
3015	default:
3016		EVUTIL_ASSERT(0);
3017		EVUTIL_FALLTHROUGH;
3018	case EVLIST_ACTIVE_LATER:
3019		event_queue_remove_active_later(base, evcb);
3020		r = 0;
3021		break;
3022	case EVLIST_ACTIVE:
3023		return 0;
3024	case 0:
3025		break;
3026	}
3027
3028	event_queue_insert_active(base, evcb);
3029
3030	if (EVBASE_NEED_NOTIFY(base))
3031		evthread_notify_base(base);
3032
3033	return r;
3034}
3035
3036int
3037event_callback_activate_later_nolock_(struct event_base *base,
3038    struct event_callback *evcb)
3039{
3040	if (evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))
3041		return 0;
3042
3043	event_queue_insert_active_later(base, evcb);
3044	if (EVBASE_NEED_NOTIFY(base))
3045		evthread_notify_base(base);
3046	return 1;
3047}
3048
3049void
3050event_callback_init_(struct event_base *base,
3051    struct event_callback *cb)
3052{
3053	memset(cb, 0, sizeof(*cb));
3054	cb->evcb_pri = base->nactivequeues - 1;
3055}
3056
3057int
3058event_callback_cancel_(struct event_base *base,
3059    struct event_callback *evcb)
3060{
3061	int r;
3062	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3063	r = event_callback_cancel_nolock_(base, evcb, 0);
3064	EVBASE_RELEASE_LOCK(base, th_base_lock);
3065	return r;
3066}
3067
3068int
3069event_callback_cancel_nolock_(struct event_base *base,
3070    struct event_callback *evcb, int even_if_finalizing)
3071{
3072	if ((evcb->evcb_flags & EVLIST_FINALIZING) && !even_if_finalizing)
3073		return 0;
3074
3075	if (evcb->evcb_flags & EVLIST_INIT)
3076		return event_del_nolock_(event_callback_to_event(evcb),
3077		    even_if_finalizing ? EVENT_DEL_EVEN_IF_FINALIZING : EVENT_DEL_AUTOBLOCK);
3078
3079	switch ((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) {
3080	default:
3081	case EVLIST_ACTIVE|EVLIST_ACTIVE_LATER:
3082		EVUTIL_ASSERT(0);
3083		break;
3084	case EVLIST_ACTIVE:
3085		/* We get different kinds of events, add them together */
3086		event_queue_remove_active(base, evcb);
3087		return 0;
3088	case EVLIST_ACTIVE_LATER:
3089		event_queue_remove_active_later(base, evcb);
3090		break;
3091	case 0:
3092		break;
3093	}
3094
3095	return 0;
3096}
3097
3098void
3099event_deferred_cb_init_(struct event_callback *cb, ev_uint8_t priority, deferred_cb_fn fn, void *arg)
3100{
3101	memset(cb, 0, sizeof(*cb));
3102	cb->evcb_cb_union.evcb_selfcb = fn;
3103	cb->evcb_arg = arg;
3104	cb->evcb_pri = priority;
3105	cb->evcb_closure = EV_CLOSURE_CB_SELF;
3106}
3107
3108void
3109event_deferred_cb_set_priority_(struct event_callback *cb, ev_uint8_t priority)
3110{
3111	cb->evcb_pri = priority;
3112}
3113
3114void
3115event_deferred_cb_cancel_(struct event_base *base, struct event_callback *cb)
3116{
3117	if (!base)
3118		base = current_base;
3119	event_callback_cancel_(base, cb);
3120}
3121
3122#define MAX_DEFERREDS_QUEUED 32
3123int
3124event_deferred_cb_schedule_(struct event_base *base, struct event_callback *cb)
3125{
3126	int r = 1;
3127	if (!base)
3128		base = current_base;
3129	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3130	if (base->n_deferreds_queued > MAX_DEFERREDS_QUEUED) {
3131		r = event_callback_activate_later_nolock_(base, cb);
3132	} else {
3133		r = event_callback_activate_nolock_(base, cb);
3134		if (r) {
3135			++base->n_deferreds_queued;
3136		}
3137	}
3138	EVBASE_RELEASE_LOCK(base, th_base_lock);
3139	return r;
3140}
3141
3142static int
3143timeout_next(struct event_base *base, struct timeval **tv_p)
3144{
3145	/* Caller must hold th_base_lock */
3146	struct timeval now;
3147	struct event *ev;
3148	struct timeval *tv = *tv_p;
3149	int res = 0;
3150
3151	ev = min_heap_top_(&base->timeheap);
3152
3153	if (ev == NULL) {
3154		/* if no time-based events are active wait for I/O */
3155		*tv_p = NULL;
3156		goto out;
3157	}
3158
3159	if (gettime(base, &now) == -1) {
3160		res = -1;
3161		goto out;
3162	}
3163
3164	if (evutil_timercmp(&ev->ev_timeout, &now, <=)) {
3165		evutil_timerclear(tv);
3166		goto out;
3167	}
3168
3169	evutil_timersub(&ev->ev_timeout, &now, tv);
3170
3171	EVUTIL_ASSERT(tv->tv_sec >= 0);
3172	EVUTIL_ASSERT(tv->tv_usec >= 0);
3173	event_debug(("timeout_next: event: %p, in %d seconds, %d useconds", ev, (int)tv->tv_sec, (int)tv->tv_usec));
3174
3175out:
3176	return (res);
3177}
3178
3179/* Activate every event whose timeout has elapsed. */
3180static void
3181timeout_process(struct event_base *base)
3182{
3183	/* Caller must hold lock. */
3184	struct timeval now;
3185	struct event *ev;
3186
3187	if (min_heap_empty_(&base->timeheap)) {
3188		return;
3189	}
3190
3191	gettime(base, &now);
3192
3193	while ((ev = min_heap_top_(&base->timeheap))) {
3194		if (evutil_timercmp(&ev->ev_timeout, &now, >))
3195			break;
3196
3197		/* delete this event from the I/O queues */
3198		event_del_nolock_(ev, EVENT_DEL_NOBLOCK);
3199
3200		event_debug(("timeout_process: event: %p, call %p",
3201			 ev, ev->ev_callback));
3202		event_active_nolock_(ev, EV_TIMEOUT, 1);
3203	}
3204}
3205
3206#ifndef MAX
3207#define MAX(a,b) (((a)>(b))?(a):(b))
3208#endif
3209
3210#define MAX_EVENT_COUNT(var, v) var = MAX(var, v)
3211
3212/* These are a fancy way to spell
3213     if (~flags & EVLIST_INTERNAL)
3214         base->event_count--/++;
3215*/
3216#define DECR_EVENT_COUNT(base,flags) \
3217	((base)->event_count -= !((flags) & EVLIST_INTERNAL))
3218#define INCR_EVENT_COUNT(base,flags) do {					\
3219	((base)->event_count += !((flags) & EVLIST_INTERNAL));			\
3220	MAX_EVENT_COUNT((base)->event_count_max, (base)->event_count);		\
3221} while (0)
3222
3223static void
3224event_queue_remove_inserted(struct event_base *base, struct event *ev)
3225{
3226	EVENT_BASE_ASSERT_LOCKED(base);
3227	if (EVUTIL_FAILURE_CHECK(!(ev->ev_flags & EVLIST_INSERTED))) {
3228		event_errx(1, "%s: %p(fd "EV_SOCK_FMT") not on queue %x", __func__,
3229		    ev, EV_SOCK_ARG(ev->ev_fd), EVLIST_INSERTED);
3230		return;
3231	}
3232	DECR_EVENT_COUNT(base, ev->ev_flags);
3233	ev->ev_flags &= ~EVLIST_INSERTED;
3234}
3235static void
3236event_queue_remove_active(struct event_base *base, struct event_callback *evcb)
3237{
3238	EVENT_BASE_ASSERT_LOCKED(base);
3239	if (EVUTIL_FAILURE_CHECK(!(evcb->evcb_flags & EVLIST_ACTIVE))) {
3240		event_errx(1, "%s: %p not on queue %x", __func__,
3241			   evcb, EVLIST_ACTIVE);
3242		return;
3243	}
3244	DECR_EVENT_COUNT(base, evcb->evcb_flags);
3245	evcb->evcb_flags &= ~EVLIST_ACTIVE;
3246	base->event_count_active--;
3247
3248	TAILQ_REMOVE(&base->activequeues[evcb->evcb_pri],
3249	    evcb, evcb_active_next);
3250}
3251static void
3252event_queue_remove_active_later(struct event_base *base, struct event_callback *evcb)
3253{
3254	EVENT_BASE_ASSERT_LOCKED(base);
3255	if (EVUTIL_FAILURE_CHECK(!(evcb->evcb_flags & EVLIST_ACTIVE_LATER))) {
3256		event_errx(1, "%s: %p not on queue %x", __func__,
3257			   evcb, EVLIST_ACTIVE_LATER);
3258		return;
3259	}
3260	DECR_EVENT_COUNT(base, evcb->evcb_flags);
3261	evcb->evcb_flags &= ~EVLIST_ACTIVE_LATER;
3262	base->event_count_active--;
3263
3264	TAILQ_REMOVE(&base->active_later_queue, evcb, evcb_active_next);
3265}
3266static void
3267event_queue_remove_timeout(struct event_base *base, struct event *ev)
3268{
3269	EVENT_BASE_ASSERT_LOCKED(base);
3270	if (EVUTIL_FAILURE_CHECK(!(ev->ev_flags & EVLIST_TIMEOUT))) {
3271		event_errx(1, "%s: %p(fd "EV_SOCK_FMT") not on queue %x", __func__,
3272		    ev, EV_SOCK_ARG(ev->ev_fd), EVLIST_TIMEOUT);
3273		return;
3274	}
3275	DECR_EVENT_COUNT(base, ev->ev_flags);
3276	ev->ev_flags &= ~EVLIST_TIMEOUT;
3277
3278	if (is_common_timeout(&ev->ev_timeout, base)) {
3279		struct common_timeout_list *ctl =
3280		    get_common_timeout_list(base, &ev->ev_timeout);
3281		TAILQ_REMOVE(&ctl->events, ev,
3282		    ev_timeout_pos.ev_next_with_common_timeout);
3283	} else {
3284		min_heap_erase_(&base->timeheap, ev);
3285	}
3286}
3287
3288#ifdef USE_REINSERT_TIMEOUT
3289/* Remove and reinsert 'ev' into the timeout queue. */
3290static void
3291event_queue_reinsert_timeout(struct event_base *base, struct event *ev,
3292    int was_common, int is_common, int old_timeout_idx)
3293{
3294	struct common_timeout_list *ctl;
3295	if (!(ev->ev_flags & EVLIST_TIMEOUT)) {
3296		event_queue_insert_timeout(base, ev);
3297		return;
3298	}
3299
3300	switch ((was_common<<1) | is_common) {
3301	case 3: /* Changing from one common timeout to another */
3302		ctl = base->common_timeout_queues[old_timeout_idx];
3303		TAILQ_REMOVE(&ctl->events, ev,
3304		    ev_timeout_pos.ev_next_with_common_timeout);
3305		ctl = get_common_timeout_list(base, &ev->ev_timeout);
3306		insert_common_timeout_inorder(ctl, ev);
3307		break;
3308	case 2: /* Was common; is no longer common */
3309		ctl = base->common_timeout_queues[old_timeout_idx];
3310		TAILQ_REMOVE(&ctl->events, ev,
3311		    ev_timeout_pos.ev_next_with_common_timeout);
3312		min_heap_push_(&base->timeheap, ev);
3313		break;
3314	case 1: /* Wasn't common; has become common. */
3315		min_heap_erase_(&base->timeheap, ev);
3316		ctl = get_common_timeout_list(base, &ev->ev_timeout);
3317		insert_common_timeout_inorder(ctl, ev);
3318		break;
3319	case 0: /* was in heap; is still on heap. */
3320		min_heap_adjust_(&base->timeheap, ev);
3321		break;
3322	default:
3323		EVUTIL_ASSERT(0); /* unreachable */
3324		break;
3325	}
3326}
3327#endif
3328
3329/* Add 'ev' to the common timeout list in 'ev'. */
3330static void
3331insert_common_timeout_inorder(struct common_timeout_list *ctl,
3332    struct event *ev)
3333{
3334	struct event *e;
3335	/* By all logic, we should just be able to append 'ev' to the end of
3336	 * ctl->events, since the timeout on each 'ev' is set to {the common
3337	 * timeout} + {the time when we add the event}, and so the events
3338	 * should arrive in order of their timeeouts.  But just in case
3339	 * there's some wacky threading issue going on, we do a search from
3340	 * the end of 'ev' to find the right insertion point.
3341	 */
3342	TAILQ_FOREACH_REVERSE(e, &ctl->events,
3343	    event_list, ev_timeout_pos.ev_next_with_common_timeout) {
3344		/* This timercmp is a little sneaky, since both ev and e have
3345		 * magic values in tv_usec.  Fortunately, they ought to have
3346		 * the _same_ magic values in tv_usec.  Let's assert for that.
3347		 */
3348		EVUTIL_ASSERT(
3349			is_same_common_timeout(&e->ev_timeout, &ev->ev_timeout));
3350		if (evutil_timercmp(&ev->ev_timeout, &e->ev_timeout, >=)) {
3351			TAILQ_INSERT_AFTER(&ctl->events, e, ev,
3352			    ev_timeout_pos.ev_next_with_common_timeout);
3353			return;
3354		}
3355	}
3356	TAILQ_INSERT_HEAD(&ctl->events, ev,
3357	    ev_timeout_pos.ev_next_with_common_timeout);
3358}
3359
3360static void
3361event_queue_insert_inserted(struct event_base *base, struct event *ev)
3362{
3363	EVENT_BASE_ASSERT_LOCKED(base);
3364
3365	if (EVUTIL_FAILURE_CHECK(ev->ev_flags & EVLIST_INSERTED)) {
3366		event_errx(1, "%s: %p(fd "EV_SOCK_FMT") already inserted", __func__,
3367		    ev, EV_SOCK_ARG(ev->ev_fd));
3368		return;
3369	}
3370
3371	INCR_EVENT_COUNT(base, ev->ev_flags);
3372
3373	ev->ev_flags |= EVLIST_INSERTED;
3374}
3375
3376static void
3377event_queue_insert_active(struct event_base *base, struct event_callback *evcb)
3378{
3379	EVENT_BASE_ASSERT_LOCKED(base);
3380
3381	if (evcb->evcb_flags & EVLIST_ACTIVE) {
3382		/* Double insertion is possible for active events */
3383		return;
3384	}
3385
3386	INCR_EVENT_COUNT(base, evcb->evcb_flags);
3387
3388	evcb->evcb_flags |= EVLIST_ACTIVE;
3389
3390	base->event_count_active++;
3391	MAX_EVENT_COUNT(base->event_count_active_max, base->event_count_active);
3392	EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues);
3393	TAILQ_INSERT_TAIL(&base->activequeues[evcb->evcb_pri],
3394	    evcb, evcb_active_next);
3395}
3396
3397static void
3398event_queue_insert_active_later(struct event_base *base, struct event_callback *evcb)
3399{
3400	EVENT_BASE_ASSERT_LOCKED(base);
3401	if (evcb->evcb_flags & (EVLIST_ACTIVE_LATER|EVLIST_ACTIVE)) {
3402		/* Double insertion is possible */
3403		return;
3404	}
3405
3406	INCR_EVENT_COUNT(base, evcb->evcb_flags);
3407	evcb->evcb_flags |= EVLIST_ACTIVE_LATER;
3408	base->event_count_active++;
3409	MAX_EVENT_COUNT(base->event_count_active_max, base->event_count_active);
3410	EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues);
3411	TAILQ_INSERT_TAIL(&base->active_later_queue, evcb, evcb_active_next);
3412}
3413
3414static void
3415event_queue_insert_timeout(struct event_base *base, struct event *ev)
3416{
3417	EVENT_BASE_ASSERT_LOCKED(base);
3418
3419	if (EVUTIL_FAILURE_CHECK(ev->ev_flags & EVLIST_TIMEOUT)) {
3420		event_errx(1, "%s: %p(fd "EV_SOCK_FMT") already on timeout", __func__,
3421		    ev, EV_SOCK_ARG(ev->ev_fd));
3422		return;
3423	}
3424
3425	INCR_EVENT_COUNT(base, ev->ev_flags);
3426
3427	ev->ev_flags |= EVLIST_TIMEOUT;
3428
3429	if (is_common_timeout(&ev->ev_timeout, base)) {
3430		struct common_timeout_list *ctl =
3431		    get_common_timeout_list(base, &ev->ev_timeout);
3432		insert_common_timeout_inorder(ctl, ev);
3433	} else {
3434		min_heap_push_(&base->timeheap, ev);
3435	}
3436}
3437
3438static void
3439event_queue_make_later_events_active(struct event_base *base)
3440{
3441	struct event_callback *evcb;
3442	EVENT_BASE_ASSERT_LOCKED(base);
3443
3444	while ((evcb = TAILQ_FIRST(&base->active_later_queue))) {
3445		TAILQ_REMOVE(&base->active_later_queue, evcb, evcb_active_next);
3446		evcb->evcb_flags = (evcb->evcb_flags & ~EVLIST_ACTIVE_LATER) | EVLIST_ACTIVE;
3447		EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues);
3448		TAILQ_INSERT_TAIL(&base->activequeues[evcb->evcb_pri], evcb, evcb_active_next);
3449		base->n_deferreds_queued += (evcb->evcb_closure == EV_CLOSURE_CB_SELF);
3450	}
3451}
3452
3453/* Functions for debugging */
3454
3455const char *
3456event_get_version(void)
3457{
3458	return (EVENT__VERSION);
3459}
3460
3461ev_uint32_t
3462event_get_version_number(void)
3463{
3464	return (EVENT__NUMERIC_VERSION);
3465}
3466
3467/*
3468 * No thread-safe interface needed - the information should be the same
3469 * for all threads.
3470 */
3471
3472const char *
3473event_get_method(void)
3474{
3475	return (current_base->evsel->name);
3476}
3477
3478#ifndef EVENT__DISABLE_MM_REPLACEMENT
3479static void *(*mm_malloc_fn_)(size_t sz) = NULL;
3480static void *(*mm_realloc_fn_)(void *p, size_t sz) = NULL;
3481static void (*mm_free_fn_)(void *p) = NULL;
3482
3483void *
3484event_mm_malloc_(size_t sz)
3485{
3486	if (sz == 0)
3487		return NULL;
3488
3489	if (mm_malloc_fn_)
3490		return mm_malloc_fn_(sz);
3491	else
3492		return malloc(sz);
3493}
3494
3495void *
3496event_mm_calloc_(size_t count, size_t size)
3497{
3498	if (count == 0 || size == 0)
3499		return NULL;
3500
3501	if (mm_malloc_fn_) {
3502		size_t sz = count * size;
3503		void *p = NULL;
3504		if (count > EV_SIZE_MAX / size)
3505			goto error;
3506		p = mm_malloc_fn_(sz);
3507		if (p)
3508			return memset(p, 0, sz);
3509	} else {
3510		void *p = calloc(count, size);
3511#ifdef _WIN32
3512		/* Windows calloc doesn't reliably set ENOMEM */
3513		if (p == NULL)
3514			goto error;
3515#endif
3516		return p;
3517	}
3518
3519error:
3520	errno = ENOMEM;
3521	return NULL;
3522}
3523
3524char *
3525event_mm_strdup_(const char *str)
3526{
3527	if (!str) {
3528		errno = EINVAL;
3529		return NULL;
3530	}
3531
3532	if (mm_malloc_fn_) {
3533		size_t ln = strlen(str);
3534		void *p = NULL;
3535		if (ln == EV_SIZE_MAX)
3536			goto error;
3537		p = mm_malloc_fn_(ln+1);
3538		if (p)
3539			return memcpy(p, str, ln+1);
3540	} else
3541#ifdef _WIN32
3542		return _strdup(str);
3543#else
3544		return strdup(str);
3545#endif
3546
3547error:
3548	errno = ENOMEM;
3549	return NULL;
3550}
3551
3552void *
3553event_mm_realloc_(void *ptr, size_t sz)
3554{
3555	if (mm_realloc_fn_)
3556		return mm_realloc_fn_(ptr, sz);
3557	else
3558		return realloc(ptr, sz);
3559}
3560
3561void
3562event_mm_free_(void *ptr)
3563{
3564	if (mm_free_fn_)
3565		mm_free_fn_(ptr);
3566	else
3567		free(ptr);
3568}
3569
3570void
3571event_set_mem_functions(void *(*malloc_fn)(size_t sz),
3572			void *(*realloc_fn)(void *ptr, size_t sz),
3573			void (*free_fn)(void *ptr))
3574{
3575	mm_malloc_fn_ = malloc_fn;
3576	mm_realloc_fn_ = realloc_fn;
3577	mm_free_fn_ = free_fn;
3578}
3579#endif
3580
3581#ifdef EVENT__HAVE_EVENTFD
3582static void
3583evthread_notify_drain_eventfd(evutil_socket_t fd, short what, void *arg)
3584{
3585	ev_uint64_t msg;
3586	ev_ssize_t r;
3587	struct event_base *base = arg;
3588
3589	r = read(fd, (void*) &msg, sizeof(msg));
3590	if (r<0 && errno != EAGAIN) {
3591		event_sock_warn(fd, "Error reading from eventfd");
3592	}
3593	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3594	base->is_notify_pending = 0;
3595	EVBASE_RELEASE_LOCK(base, th_base_lock);
3596}
3597#endif
3598
3599static void
3600evthread_notify_drain_default(evutil_socket_t fd, short what, void *arg)
3601{
3602	unsigned char buf[1024];
3603	struct event_base *base = arg;
3604#ifdef _WIN32
3605	while (recv(fd, (char*)buf, sizeof(buf), 0) > 0)
3606		;
3607#else
3608	while (read(fd, (char*)buf, sizeof(buf)) > 0)
3609		;
3610#endif
3611
3612	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3613	base->is_notify_pending = 0;
3614	EVBASE_RELEASE_LOCK(base, th_base_lock);
3615}
3616
3617int
3618evthread_make_base_notifiable(struct event_base *base)
3619{
3620	int r;
3621	if (!base)
3622		return -1;
3623
3624	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3625	r = evthread_make_base_notifiable_nolock_(base);
3626	EVBASE_RELEASE_LOCK(base, th_base_lock);
3627	return r;
3628}
3629
3630static int
3631evthread_make_base_notifiable_nolock_(struct event_base *base)
3632{
3633	void (*cb)(evutil_socket_t, short, void *);
3634	int (*notify)(struct event_base *);
3635
3636	if (base->th_notify_fn != NULL) {
3637		/* The base is already notifiable: we're doing fine. */
3638		return 0;
3639	}
3640
3641#if defined(EVENT__HAVE_WORKING_KQUEUE)
3642	if (base->evsel == &kqops && event_kq_add_notify_event_(base) == 0) {
3643		base->th_notify_fn = event_kq_notify_base_;
3644		/* No need to add an event here; the backend can wake
3645		 * itself up just fine. */
3646		return 0;
3647	}
3648#endif
3649
3650#ifdef EVENT__HAVE_EVENTFD
3651	base->th_notify_fd[0] = evutil_eventfd_(0,
3652	    EVUTIL_EFD_CLOEXEC|EVUTIL_EFD_NONBLOCK);
3653	if (base->th_notify_fd[0] >= 0) {
3654		base->th_notify_fd[1] = -1;
3655		notify = evthread_notify_base_eventfd;
3656		cb = evthread_notify_drain_eventfd;
3657	} else
3658#endif
3659	if (evutil_make_internal_pipe_(base->th_notify_fd) == 0) {
3660		notify = evthread_notify_base_default;
3661		cb = evthread_notify_drain_default;
3662	} else {
3663		return -1;
3664	}
3665
3666	base->th_notify_fn = notify;
3667
3668	/* prepare an event that we can use for wakeup */
3669	event_assign(&base->th_notify, base, base->th_notify_fd[0],
3670				 EV_READ|EV_PERSIST, cb, base);
3671
3672	/* we need to mark this as internal event */
3673	base->th_notify.ev_flags |= EVLIST_INTERNAL;
3674	event_priority_set(&base->th_notify, 0);
3675
3676	return event_add_nolock_(&base->th_notify, NULL, 0);
3677}
3678
3679int
3680event_base_foreach_event_nolock_(struct event_base *base,
3681    event_base_foreach_event_cb fn, void *arg)
3682{
3683	int r, i;
3684	unsigned u;
3685	struct event *ev;
3686
3687	/* Start out with all the EVLIST_INSERTED events. */
3688	if ((r = evmap_foreach_event_(base, fn, arg)))
3689		return r;
3690
3691	/* Okay, now we deal with those events that have timeouts and are in
3692	 * the min-heap. */
3693	for (u = 0; u < base->timeheap.n; ++u) {
3694		ev = base->timeheap.p[u];
3695		if (ev->ev_flags & EVLIST_INSERTED) {
3696			/* we already processed this one */
3697			continue;
3698		}
3699		if ((r = fn(base, ev, arg)))
3700			return r;
3701	}
3702
3703	/* Now for the events in one of the timeout queues.
3704	 * the min-heap. */
3705	for (i = 0; i < base->n_common_timeouts; ++i) {
3706		struct common_timeout_list *ctl =
3707		    base->common_timeout_queues[i];
3708		TAILQ_FOREACH(ev, &ctl->events,
3709		    ev_timeout_pos.ev_next_with_common_timeout) {
3710			if (ev->ev_flags & EVLIST_INSERTED) {
3711				/* we already processed this one */
3712				continue;
3713			}
3714			if ((r = fn(base, ev, arg)))
3715				return r;
3716		}
3717	}
3718
3719	/* Finally, we deal wit all the active events that we haven't touched
3720	 * yet. */
3721	for (i = 0; i < base->nactivequeues; ++i) {
3722		struct event_callback *evcb;
3723		TAILQ_FOREACH(evcb, &base->activequeues[i], evcb_active_next) {
3724			if ((evcb->evcb_flags & (EVLIST_INIT|EVLIST_INSERTED|EVLIST_TIMEOUT)) != EVLIST_INIT) {
3725				/* This isn't an event (evlist_init clear), or
3726				 * we already processed it. (inserted or
3727				 * timeout set */
3728				continue;
3729			}
3730			ev = event_callback_to_event(evcb);
3731			if ((r = fn(base, ev, arg)))
3732				return r;
3733		}
3734	}
3735
3736	return 0;
3737}
3738
3739/* Helper for event_base_dump_events: called on each event in the event base;
3740 * dumps only the inserted events. */
3741static int
3742dump_inserted_event_fn(const struct event_base *base, const struct event *e, void *arg)
3743{
3744	FILE *output = arg;
3745	const char *gloss = (e->ev_events & EV_SIGNAL) ?
3746	    "sig" : "fd ";
3747
3748	if (! (e->ev_flags & (EVLIST_INSERTED|EVLIST_TIMEOUT)))
3749		return 0;
3750
3751	fprintf(output, "  %p [%s "EV_SOCK_FMT"]%s%s%s%s%s%s%s",
3752	    (void*)e, gloss, EV_SOCK_ARG(e->ev_fd),
3753	    (e->ev_events&EV_READ)?" Read":"",
3754	    (e->ev_events&EV_WRITE)?" Write":"",
3755	    (e->ev_events&EV_CLOSED)?" EOF":"",
3756	    (e->ev_events&EV_SIGNAL)?" Signal":"",
3757	    (e->ev_events&EV_PERSIST)?" Persist":"",
3758	    (e->ev_events&EV_ET)?" ET":"",
3759	    (e->ev_flags&EVLIST_INTERNAL)?" Internal":"");
3760	if (e->ev_flags & EVLIST_TIMEOUT) {
3761		struct timeval tv;
3762		tv.tv_sec = e->ev_timeout.tv_sec;
3763		tv.tv_usec = e->ev_timeout.tv_usec & MICROSECONDS_MASK;
3764		evutil_timeradd(&tv, &base->tv_clock_diff, &tv);
3765		fprintf(output, " Timeout=%ld.%06d",
3766		    (long)tv.tv_sec, (int)(tv.tv_usec & MICROSECONDS_MASK));
3767	}
3768	fputc('\n', output);
3769
3770	return 0;
3771}
3772
3773/* Helper for event_base_dump_events: called on each event in the event base;
3774 * dumps only the active events. */
3775static int
3776dump_active_event_fn(const struct event_base *base, const struct event *e, void *arg)
3777{
3778	FILE *output = arg;
3779	const char *gloss = (e->ev_events & EV_SIGNAL) ?
3780	    "sig" : "fd ";
3781
3782	if (! (e->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)))
3783		return 0;
3784
3785	fprintf(output, "  %p [%s "EV_SOCK_FMT", priority=%d]%s%s%s%s%s active%s%s\n",
3786	    (void*)e, gloss, EV_SOCK_ARG(e->ev_fd), e->ev_pri,
3787	    (e->ev_res&EV_READ)?" Read":"",
3788	    (e->ev_res&EV_WRITE)?" Write":"",
3789	    (e->ev_res&EV_CLOSED)?" EOF":"",
3790	    (e->ev_res&EV_SIGNAL)?" Signal":"",
3791	    (e->ev_res&EV_TIMEOUT)?" Timeout":"",
3792	    (e->ev_flags&EVLIST_INTERNAL)?" [Internal]":"",
3793	    (e->ev_flags&EVLIST_ACTIVE_LATER)?" [NextTime]":"");
3794
3795	return 0;
3796}
3797
3798int
3799event_base_foreach_event(struct event_base *base,
3800    event_base_foreach_event_cb fn, void *arg)
3801{
3802	int r;
3803	if ((!fn) || (!base)) {
3804		return -1;
3805	}
3806	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3807	r = event_base_foreach_event_nolock_(base, fn, arg);
3808	EVBASE_RELEASE_LOCK(base, th_base_lock);
3809	return r;
3810}
3811
3812
3813void
3814event_base_dump_events(struct event_base *base, FILE *output)
3815{
3816	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3817	fprintf(output, "Inserted events:\n");
3818	event_base_foreach_event_nolock_(base, dump_inserted_event_fn, output);
3819
3820	fprintf(output, "Active events:\n");
3821	event_base_foreach_event_nolock_(base, dump_active_event_fn, output);
3822	EVBASE_RELEASE_LOCK(base, th_base_lock);
3823}
3824
3825void
3826event_base_active_by_fd(struct event_base *base, evutil_socket_t fd, short events)
3827{
3828	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3829
3830	/* Activate any non timer events */
3831	if (!(events & EV_TIMEOUT)) {
3832		evmap_io_active_(base, fd, events & (EV_READ|EV_WRITE|EV_CLOSED));
3833	} else {
3834		/* If we want to activate timer events, loop and activate each event with
3835		 * the same fd in both the timeheap and common timeouts list */
3836		int i;
3837		unsigned u;
3838		struct event *ev;
3839
3840		for (u = 0; u < base->timeheap.n; ++u) {
3841			ev = base->timeheap.p[u];
3842			if (ev->ev_fd == fd) {
3843				event_active_nolock_(ev, EV_TIMEOUT, 1);
3844			}
3845		}
3846
3847		for (i = 0; i < base->n_common_timeouts; ++i) {
3848			struct common_timeout_list *ctl = base->common_timeout_queues[i];
3849			TAILQ_FOREACH(ev, &ctl->events,
3850				ev_timeout_pos.ev_next_with_common_timeout) {
3851				if (ev->ev_fd == fd) {
3852					event_active_nolock_(ev, EV_TIMEOUT, 1);
3853				}
3854			}
3855		}
3856	}
3857
3858	EVBASE_RELEASE_LOCK(base, th_base_lock);
3859}
3860
3861void
3862event_base_active_by_signal(struct event_base *base, int sig)
3863{
3864	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3865	evmap_signal_active_(base, sig, 1);
3866	EVBASE_RELEASE_LOCK(base, th_base_lock);
3867}
3868
3869
3870void
3871event_base_add_virtual_(struct event_base *base)
3872{
3873	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3874	base->virtual_event_count++;
3875	MAX_EVENT_COUNT(base->virtual_event_count_max, base->virtual_event_count);
3876	EVBASE_RELEASE_LOCK(base, th_base_lock);
3877}
3878
3879void
3880event_base_del_virtual_(struct event_base *base)
3881{
3882	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3883	EVUTIL_ASSERT(base->virtual_event_count > 0);
3884	base->virtual_event_count--;
3885	if (base->virtual_event_count == 0 && EVBASE_NEED_NOTIFY(base))
3886		evthread_notify_base(base);
3887	EVBASE_RELEASE_LOCK(base, th_base_lock);
3888}
3889
3890static void
3891event_free_debug_globals_locks(void)
3892{
3893#ifndef EVENT__DISABLE_THREAD_SUPPORT
3894#ifndef EVENT__DISABLE_DEBUG_MODE
3895	if (event_debug_map_lock_ != NULL) {
3896		EVTHREAD_FREE_LOCK(event_debug_map_lock_, 0);
3897		event_debug_map_lock_ = NULL;
3898		evthreadimpl_disable_lock_debugging_();
3899	}
3900#endif /* EVENT__DISABLE_DEBUG_MODE */
3901#endif /* EVENT__DISABLE_THREAD_SUPPORT */
3902	return;
3903}
3904
3905static void
3906event_free_debug_globals(void)
3907{
3908	event_free_debug_globals_locks();
3909}
3910
3911static void
3912event_free_evsig_globals(void)
3913{
3914	evsig_free_globals_();
3915}
3916
3917static void
3918event_free_evutil_globals(void)
3919{
3920	evutil_free_globals_();
3921}
3922
3923static void
3924event_free_globals(void)
3925{
3926	event_free_debug_globals();
3927	event_free_evsig_globals();
3928	event_free_evutil_globals();
3929}
3930
3931void
3932libevent_global_shutdown(void)
3933{
3934	event_disable_debug_mode();
3935	event_free_globals();
3936}
3937
3938#ifndef EVENT__DISABLE_THREAD_SUPPORT
3939int
3940event_global_setup_locks_(const int enable_locks)
3941{
3942#ifndef EVENT__DISABLE_DEBUG_MODE
3943	EVTHREAD_SETUP_GLOBAL_LOCK(event_debug_map_lock_, 0);
3944#endif
3945	if (evsig_global_setup_locks_(enable_locks) < 0)
3946		return -1;
3947	if (evutil_global_setup_locks_(enable_locks) < 0)
3948		return -1;
3949	if (evutil_secure_rng_global_setup_locks_(enable_locks) < 0)
3950		return -1;
3951	return 0;
3952}
3953#endif
3954
3955void
3956event_base_assert_ok_(struct event_base *base)
3957{
3958	EVBASE_ACQUIRE_LOCK(base, th_base_lock);
3959	event_base_assert_ok_nolock_(base);
3960	EVBASE_RELEASE_LOCK(base, th_base_lock);
3961}
3962
3963void
3964event_base_assert_ok_nolock_(struct event_base *base)
3965{
3966	int i;
3967	int count;
3968
3969	/* First do checks on the per-fd and per-signal lists */
3970	evmap_check_integrity_(base);
3971
3972	/* Check the heap property */
3973	for (i = 1; i < (int)base->timeheap.n; ++i) {
3974		int parent = (i - 1) / 2;
3975		struct event *ev, *p_ev;
3976		ev = base->timeheap.p[i];
3977		p_ev = base->timeheap.p[parent];
3978		EVUTIL_ASSERT(ev->ev_flags & EVLIST_TIMEOUT);
3979		EVUTIL_ASSERT(evutil_timercmp(&p_ev->ev_timeout, &ev->ev_timeout, <=));
3980		EVUTIL_ASSERT(ev->ev_timeout_pos.min_heap_idx == i);
3981	}
3982
3983	/* Check that the common timeouts are fine */
3984	for (i = 0; i < base->n_common_timeouts; ++i) {
3985		struct common_timeout_list *ctl = base->common_timeout_queues[i];
3986		struct event *last=NULL, *ev;
3987
3988		EVUTIL_ASSERT_TAILQ_OK(&ctl->events, event, ev_timeout_pos.ev_next_with_common_timeout);
3989
3990		TAILQ_FOREACH(ev, &ctl->events, ev_timeout_pos.ev_next_with_common_timeout) {
3991			if (last)
3992				EVUTIL_ASSERT(evutil_timercmp(&last->ev_timeout, &ev->ev_timeout, <=));
3993			EVUTIL_ASSERT(ev->ev_flags & EVLIST_TIMEOUT);
3994			EVUTIL_ASSERT(is_common_timeout(&ev->ev_timeout,base));
3995			EVUTIL_ASSERT(COMMON_TIMEOUT_IDX(&ev->ev_timeout) == i);
3996			last = ev;
3997		}
3998	}
3999
4000	/* Check the active queues. */
4001	count = 0;
4002	for (i = 0; i < base->nactivequeues; ++i) {
4003		struct event_callback *evcb;
4004		EVUTIL_ASSERT_TAILQ_OK(&base->activequeues[i], event_callback, evcb_active_next);
4005		TAILQ_FOREACH(evcb, &base->activequeues[i], evcb_active_next) {
4006			EVUTIL_ASSERT((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) == EVLIST_ACTIVE);
4007			EVUTIL_ASSERT(evcb->evcb_pri == i);
4008			++count;
4009		}
4010	}
4011
4012	{
4013		struct event_callback *evcb;
4014		TAILQ_FOREACH(evcb, &base->active_later_queue, evcb_active_next) {
4015			EVUTIL_ASSERT((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) == EVLIST_ACTIVE_LATER);
4016			++count;
4017		}
4018	}
4019	EVUTIL_ASSERT(count == base->event_count_active);
4020}
4021