1/*
2 * Copyright (C) 2004-2012  Internet Systems Consortium, Inc. ("ISC")
3 * Copyright (C) 1998-2003  Internet Software Consortium.
4 *
5 * Permission to use, copy, modify, and/or distribute this software for any
6 * purpose with or without fee is hereby granted, provided that the above
7 * copyright notice and this permission notice appear in all copies.
8 *
9 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
10 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
11 * AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
12 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
13 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
14 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
15 * PERFORMANCE OF THIS SOFTWARE.
16 */
17
18/* $Id$ */
19
20/*! \file
21 * \author Principal Author: Bob Halley
22 */
23
24/*
25 * XXXRTH  Need to document the states a task can be in, and the rules
26 * for changing states.
27 */
28
29#include <config.h>
30
31#include <isc/condition.h>
32#include <isc/event.h>
33#include <isc/magic.h>
34#include <isc/mem.h>
35#include <isc/msgs.h>
36#include <isc/platform.h>
37#include <isc/string.h>
38#include <isc/task.h>
39#include <isc/thread.h>
40#include <isc/util.h>
41#include <isc/xml.h>
42
43#ifdef OPENSSL_LEAKS
44#include <openssl/err.h>
45#endif
46
47/*%
48 * For BIND9 internal applications:
49 * when built with threads we use multiple worker threads shared by the whole
50 * application.
51 * when built without threads we share a single global task manager and use
52 * an integrated event loop for socket, timer, and other generic task events.
53 * For generic library:
54 * we don't use either of them: an application can have multiple task managers
55 * whether or not it's threaded, and if the application is threaded each thread
56 * is expected to have a separate manager; no "worker threads" are shared by
57 * the application threads.
58 */
59#ifdef BIND9
60#ifdef ISC_PLATFORM_USETHREADS
61#define USE_WORKER_THREADS
62#else
63#define USE_SHARED_MANAGER
64#endif	/* ISC_PLATFORM_USETHREADS */
65#endif	/* BIND9 */
66
67#include "task_p.h"
68
69#ifdef ISC_TASK_TRACE
70#define XTRACE(m)		fprintf(stderr, "task %p thread %lu: %s\n", \
71				       task, isc_thread_self(), (m))
72#define XTTRACE(t, m)		fprintf(stderr, "task %p thread %lu: %s\n", \
73				       (t), isc_thread_self(), (m))
74#define XTHREADTRACE(m)		fprintf(stderr, "thread %lu: %s\n", \
75				       isc_thread_self(), (m))
76#else
77#define XTRACE(m)
78#define XTTRACE(t, m)
79#define XTHREADTRACE(m)
80#endif
81
82/***
83 *** Types.
84 ***/
85
86typedef enum {
87	task_state_idle, task_state_ready, task_state_running,
88	task_state_done
89} task_state_t;
90
91#if defined(HAVE_LIBXML2) && defined(BIND9)
92static const char *statenames[] = {
93	"idle", "ready", "running", "done",
94};
95#endif
96
97#define TASK_MAGIC			ISC_MAGIC('T', 'A', 'S', 'K')
98#define VALID_TASK(t)			ISC_MAGIC_VALID(t, TASK_MAGIC)
99
100typedef struct isc__task isc__task_t;
101typedef struct isc__taskmgr isc__taskmgr_t;
102
103struct isc__task {
104	/* Not locked. */
105	isc_task_t			common;
106	isc__taskmgr_t *		manager;
107	isc_mutex_t			lock;
108	/* Locked by task lock. */
109	task_state_t			state;
110	unsigned int			references;
111	isc_eventlist_t			events;
112	isc_eventlist_t			on_shutdown;
113	unsigned int			quantum;
114	unsigned int			flags;
115	isc_stdtime_t			now;
116	char				name[16];
117	void *				tag;
118	/* Locked by task manager lock. */
119	LINK(isc__task_t)		link;
120	LINK(isc__task_t)		ready_link;
121	LINK(isc__task_t)		ready_priority_link;
122};
123
124#define TASK_F_SHUTTINGDOWN		0x01
125#define TASK_F_PRIVILEGED		0x02
126
127#define TASK_SHUTTINGDOWN(t)		(((t)->flags & TASK_F_SHUTTINGDOWN) \
128					 != 0)
129
130#define TASK_MANAGER_MAGIC		ISC_MAGIC('T', 'S', 'K', 'M')
131#define VALID_MANAGER(m)		ISC_MAGIC_VALID(m, TASK_MANAGER_MAGIC)
132
133typedef ISC_LIST(isc__task_t)	isc__tasklist_t;
134
135struct isc__taskmgr {
136	/* Not locked. */
137	isc_taskmgr_t			common;
138	isc_mem_t *			mctx;
139	isc_mutex_t			lock;
140#ifdef ISC_PLATFORM_USETHREADS
141	unsigned int			workers;
142	isc_thread_t *			threads;
143#endif /* ISC_PLATFORM_USETHREADS */
144	/* Locked by task manager lock. */
145	unsigned int			default_quantum;
146	LIST(isc__task_t)		tasks;
147	isc__tasklist_t			ready_tasks;
148	isc__tasklist_t			ready_priority_tasks;
149	isc_taskmgrmode_t		mode;
150#ifdef ISC_PLATFORM_USETHREADS
151	isc_condition_t			work_available;
152	isc_condition_t			exclusive_granted;
153	isc_condition_t			paused;
154#endif /* ISC_PLATFORM_USETHREADS */
155	unsigned int			tasks_running;
156	isc_boolean_t			pause_requested;
157	isc_boolean_t			exclusive_requested;
158	isc_boolean_t			exiting;
159#ifdef USE_SHARED_MANAGER
160	unsigned int			refs;
161#endif /* ISC_PLATFORM_USETHREADS */
162};
163
164#define DEFAULT_TASKMGR_QUANTUM		10
165#define DEFAULT_DEFAULT_QUANTUM		5
166#define FINISHED(m)			((m)->exiting && EMPTY((m)->tasks))
167
168#ifdef USE_SHARED_MANAGER
169static isc__taskmgr_t *taskmgr = NULL;
170#endif /* USE_SHARED_MANAGER */
171
172/*%
173 * The following can be either static or public, depending on build environment.
174 */
175
176#ifdef BIND9
177#define ISC_TASKFUNC_SCOPE
178#else
179#define ISC_TASKFUNC_SCOPE static
180#endif
181
182ISC_TASKFUNC_SCOPE isc_result_t
183isc__task_create(isc_taskmgr_t *manager0, unsigned int quantum,
184		 isc_task_t **taskp);
185ISC_TASKFUNC_SCOPE void
186isc__task_attach(isc_task_t *source0, isc_task_t **targetp);
187ISC_TASKFUNC_SCOPE void
188isc__task_detach(isc_task_t **taskp);
189ISC_TASKFUNC_SCOPE void
190isc__task_send(isc_task_t *task0, isc_event_t **eventp);
191ISC_TASKFUNC_SCOPE void
192isc__task_sendanddetach(isc_task_t **taskp, isc_event_t **eventp);
193ISC_TASKFUNC_SCOPE unsigned int
194isc__task_purgerange(isc_task_t *task0, void *sender, isc_eventtype_t first,
195		     isc_eventtype_t last, void *tag);
196ISC_TASKFUNC_SCOPE unsigned int
197isc__task_purge(isc_task_t *task, void *sender, isc_eventtype_t type,
198		void *tag);
199ISC_TASKFUNC_SCOPE isc_boolean_t
200isc__task_purgeevent(isc_task_t *task0, isc_event_t *event);
201ISC_TASKFUNC_SCOPE unsigned int
202isc__task_unsendrange(isc_task_t *task, void *sender, isc_eventtype_t first,
203		      isc_eventtype_t last, void *tag,
204		      isc_eventlist_t *events);
205ISC_TASKFUNC_SCOPE unsigned int
206isc__task_unsend(isc_task_t *task, void *sender, isc_eventtype_t type,
207		 void *tag, isc_eventlist_t *events);
208ISC_TASKFUNC_SCOPE isc_result_t
209isc__task_onshutdown(isc_task_t *task0, isc_taskaction_t action,
210		     const void *arg);
211ISC_TASKFUNC_SCOPE void
212isc__task_shutdown(isc_task_t *task0);
213ISC_TASKFUNC_SCOPE void
214isc__task_destroy(isc_task_t **taskp);
215ISC_TASKFUNC_SCOPE void
216isc__task_setname(isc_task_t *task0, const char *name, void *tag);
217ISC_TASKFUNC_SCOPE const char *
218isc__task_getname(isc_task_t *task0);
219ISC_TASKFUNC_SCOPE void *
220isc__task_gettag(isc_task_t *task0);
221ISC_TASKFUNC_SCOPE void
222isc__task_getcurrenttime(isc_task_t *task0, isc_stdtime_t *t);
223ISC_TASKFUNC_SCOPE isc_result_t
224isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
225		    unsigned int default_quantum, isc_taskmgr_t **managerp);
226ISC_TASKFUNC_SCOPE void
227isc__taskmgr_destroy(isc_taskmgr_t **managerp);
228ISC_TASKFUNC_SCOPE isc_result_t
229isc__task_beginexclusive(isc_task_t *task);
230ISC_TASKFUNC_SCOPE void
231isc__task_endexclusive(isc_task_t *task0);
232ISC_TASKFUNC_SCOPE void
233isc__task_setprivilege(isc_task_t *task0, isc_boolean_t priv);
234ISC_TASKFUNC_SCOPE isc_boolean_t
235isc__task_privilege(isc_task_t *task0);
236ISC_TASKFUNC_SCOPE void
237isc__taskmgr_setmode(isc_taskmgr_t *manager0, isc_taskmgrmode_t mode);
238ISC_TASKFUNC_SCOPE isc_taskmgrmode_t
239isc__taskmgr_mode(isc_taskmgr_t *manager0);
240
241static inline isc_boolean_t
242empty_readyq(isc__taskmgr_t *manager);
243
244static inline isc__task_t *
245pop_readyq(isc__taskmgr_t *manager);
246
247static inline void
248push_readyq(isc__taskmgr_t *manager, isc__task_t *task);
249
250static struct isc__taskmethods {
251	isc_taskmethods_t methods;
252
253	/*%
254	 * The following are defined just for avoiding unused static functions.
255	 */
256#ifndef BIND9
257	void *purgeevent, *unsendrange, *getname, *gettag, *getcurrenttime;
258#endif
259} taskmethods = {
260	{
261		isc__task_attach,
262		isc__task_detach,
263		isc__task_destroy,
264		isc__task_send,
265		isc__task_sendanddetach,
266		isc__task_unsend,
267		isc__task_onshutdown,
268		isc__task_shutdown,
269		isc__task_setname,
270		isc__task_purge,
271		isc__task_purgerange,
272		isc__task_beginexclusive,
273		isc__task_endexclusive,
274		isc__task_setprivilege,
275		isc__task_privilege
276	}
277#ifndef BIND9
278	,
279	(void *)isc__task_purgeevent, (void *)isc__task_unsendrange,
280	(void *)isc__task_getname, (void *)isc__task_gettag,
281	(void *)isc__task_getcurrenttime
282#endif
283};
284
285static isc_taskmgrmethods_t taskmgrmethods = {
286	isc__taskmgr_destroy,
287	isc__taskmgr_setmode,
288	isc__taskmgr_mode,
289	isc__task_create
290};
291
292/***
293 *** Tasks.
294 ***/
295
296static void
297task_finished(isc__task_t *task) {
298	isc__taskmgr_t *manager = task->manager;
299
300	REQUIRE(EMPTY(task->events));
301	REQUIRE(EMPTY(task->on_shutdown));
302	REQUIRE(task->references == 0);
303	REQUIRE(task->state == task_state_done);
304
305	XTRACE("task_finished");
306
307	LOCK(&manager->lock);
308	UNLINK(manager->tasks, task, link);
309#ifdef USE_WORKER_THREADS
310	if (FINISHED(manager)) {
311		/*
312		 * All tasks have completed and the
313		 * task manager is exiting.  Wake up
314		 * any idle worker threads so they
315		 * can exit.
316		 */
317		BROADCAST(&manager->work_available);
318	}
319#endif /* USE_WORKER_THREADS */
320	UNLOCK(&manager->lock);
321
322	DESTROYLOCK(&task->lock);
323	task->common.impmagic = 0;
324	task->common.magic = 0;
325	isc_mem_put(manager->mctx, task, sizeof(*task));
326}
327
328ISC_TASKFUNC_SCOPE isc_result_t
329isc__task_create(isc_taskmgr_t *manager0, unsigned int quantum,
330		 isc_task_t **taskp)
331{
332	isc__taskmgr_t *manager = (void*)manager0;
333	isc__task_t *task;
334	isc_boolean_t exiting;
335	isc_result_t result;
336
337	REQUIRE(VALID_MANAGER(manager));
338	REQUIRE(taskp != NULL && *taskp == NULL);
339
340	task = isc_mem_get(manager->mctx, sizeof(*task));
341	if (task == NULL)
342		return (ISC_R_NOMEMORY);
343	XTRACE("isc_task_create");
344	result = isc_mutex_init(&task->lock);
345	if (result != ISC_R_SUCCESS) {
346		isc_mem_put(manager->mctx, task, sizeof(*task));
347		return (result);
348	}
349	LOCK(&manager->lock);
350	LOCK(&task->lock);	/* helps coverity analysis noise ratio */
351	task->manager = manager;
352	task->state = task_state_idle;
353	task->references = 1;
354	INIT_LIST(task->events);
355	INIT_LIST(task->on_shutdown);
356	task->quantum = quantum;
357	task->flags = 0;
358	task->now = 0;
359	memset(task->name, 0, sizeof(task->name));
360	task->tag = NULL;
361	INIT_LINK(task, link);
362	INIT_LINK(task, ready_link);
363	INIT_LINK(task, ready_priority_link);
364	UNLOCK(&task->lock);
365	UNLOCK(&manager->lock);
366
367	exiting = ISC_FALSE;
368	LOCK(&manager->lock);
369	if (!manager->exiting) {
370		if (task->quantum == 0)
371			task->quantum = manager->default_quantum;
372		APPEND(manager->tasks, task, link);
373	} else
374		exiting = ISC_TRUE;
375	UNLOCK(&manager->lock);
376
377	if (exiting) {
378		DESTROYLOCK(&task->lock);
379		isc_mem_put(manager->mctx, task, sizeof(*task));
380		return (ISC_R_SHUTTINGDOWN);
381	}
382
383	task->common.methods = (isc_taskmethods_t *)&taskmethods;
384	task->common.magic = ISCAPI_TASK_MAGIC;
385	task->common.impmagic = TASK_MAGIC;
386	*taskp = (isc_task_t *)task;
387
388	return (ISC_R_SUCCESS);
389}
390
391ISC_TASKFUNC_SCOPE void
392isc__task_attach(isc_task_t *source0, isc_task_t **targetp) {
393	isc__task_t *source = (isc__task_t *)source0;
394
395	/*
396	 * Attach *targetp to source.
397	 */
398
399	REQUIRE(VALID_TASK(source));
400	REQUIRE(targetp != NULL && *targetp == NULL);
401
402	XTTRACE(source, "isc_task_attach");
403
404	LOCK(&source->lock);
405	source->references++;
406	UNLOCK(&source->lock);
407
408	*targetp = (isc_task_t *)source;
409}
410
411static inline isc_boolean_t
412task_shutdown(isc__task_t *task) {
413	isc_boolean_t was_idle = ISC_FALSE;
414	isc_event_t *event, *prev;
415
416	/*
417	 * Caller must be holding the task's lock.
418	 */
419
420	XTRACE("task_shutdown");
421
422	if (! TASK_SHUTTINGDOWN(task)) {
423		XTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
424				      ISC_MSG_SHUTTINGDOWN, "shutting down"));
425		task->flags |= TASK_F_SHUTTINGDOWN;
426		if (task->state == task_state_idle) {
427			INSIST(EMPTY(task->events));
428			task->state = task_state_ready;
429			was_idle = ISC_TRUE;
430		}
431		INSIST(task->state == task_state_ready ||
432		       task->state == task_state_running);
433
434		/*
435		 * Note that we post shutdown events LIFO.
436		 */
437		for (event = TAIL(task->on_shutdown);
438		     event != NULL;
439		     event = prev) {
440			prev = PREV(event, ev_link);
441			DEQUEUE(task->on_shutdown, event, ev_link);
442			ENQUEUE(task->events, event, ev_link);
443		}
444	}
445
446	return (was_idle);
447}
448
449/*
450 * Moves a task onto the appropriate run queue.
451 *
452 * Caller must NOT hold manager lock.
453 */
454static inline void
455task_ready(isc__task_t *task) {
456	isc__taskmgr_t *manager = task->manager;
457#ifdef USE_WORKER_THREADS
458	isc_boolean_t has_privilege = isc__task_privilege((isc_task_t *) task);
459#endif /* USE_WORKER_THREADS */
460
461	REQUIRE(VALID_MANAGER(manager));
462	REQUIRE(task->state == task_state_ready);
463
464	XTRACE("task_ready");
465
466	LOCK(&manager->lock);
467	push_readyq(manager, task);
468#ifdef USE_WORKER_THREADS
469	if (manager->mode == isc_taskmgrmode_normal || has_privilege)
470		SIGNAL(&manager->work_available);
471#endif /* USE_WORKER_THREADS */
472	UNLOCK(&manager->lock);
473}
474
475static inline isc_boolean_t
476task_detach(isc__task_t *task) {
477
478	/*
479	 * Caller must be holding the task lock.
480	 */
481
482	REQUIRE(task->references > 0);
483
484	XTRACE("detach");
485
486	task->references--;
487	if (task->references == 0 && task->state == task_state_idle) {
488		INSIST(EMPTY(task->events));
489		/*
490		 * There are no references to this task, and no
491		 * pending events.  We could try to optimize and
492		 * either initiate shutdown or clean up the task,
493		 * depending on its state, but it's easier to just
494		 * make the task ready and allow run() or the event
495		 * loop to deal with shutting down and termination.
496		 */
497		task->state = task_state_ready;
498		return (ISC_TRUE);
499	}
500
501	return (ISC_FALSE);
502}
503
504ISC_TASKFUNC_SCOPE void
505isc__task_detach(isc_task_t **taskp) {
506	isc__task_t *task;
507	isc_boolean_t was_idle;
508
509	/*
510	 * Detach *taskp from its task.
511	 */
512
513	REQUIRE(taskp != NULL);
514	task = (isc__task_t *)*taskp;
515	REQUIRE(VALID_TASK(task));
516
517	XTRACE("isc_task_detach");
518
519	LOCK(&task->lock);
520	was_idle = task_detach(task);
521	UNLOCK(&task->lock);
522
523	if (was_idle)
524		task_ready(task);
525
526	*taskp = NULL;
527}
528
529static inline isc_boolean_t
530task_send(isc__task_t *task, isc_event_t **eventp) {
531	isc_boolean_t was_idle = ISC_FALSE;
532	isc_event_t *event;
533
534	/*
535	 * Caller must be holding the task lock.
536	 */
537
538	REQUIRE(eventp != NULL);
539	event = *eventp;
540	REQUIRE(event != NULL);
541	REQUIRE(event->ev_type > 0);
542	REQUIRE(task->state != task_state_done);
543
544	XTRACE("task_send");
545
546	if (task->state == task_state_idle) {
547		was_idle = ISC_TRUE;
548		INSIST(EMPTY(task->events));
549		task->state = task_state_ready;
550	}
551	INSIST(task->state == task_state_ready ||
552	       task->state == task_state_running);
553	ENQUEUE(task->events, event, ev_link);
554	*eventp = NULL;
555
556	return (was_idle);
557}
558
559ISC_TASKFUNC_SCOPE void
560isc__task_send(isc_task_t *task0, isc_event_t **eventp) {
561	isc__task_t *task = (isc__task_t *)task0;
562	isc_boolean_t was_idle;
563
564	/*
565	 * Send '*event' to 'task'.
566	 */
567
568	REQUIRE(VALID_TASK(task));
569
570	XTRACE("isc_task_send");
571
572	/*
573	 * We're trying hard to hold locks for as short a time as possible.
574	 * We're also trying to hold as few locks as possible.  This is why
575	 * some processing is deferred until after the lock is released.
576	 */
577	LOCK(&task->lock);
578	was_idle = task_send(task, eventp);
579	UNLOCK(&task->lock);
580
581	if (was_idle) {
582		/*
583		 * We need to add this task to the ready queue.
584		 *
585		 * We've waited until now to do it because making a task
586		 * ready requires locking the manager.  If we tried to do
587		 * this while holding the task lock, we could deadlock.
588		 *
589		 * We've changed the state to ready, so no one else will
590		 * be trying to add this task to the ready queue.  The
591		 * only way to leave the ready state is by executing the
592		 * task.  It thus doesn't matter if events are added,
593		 * removed, or a shutdown is started in the interval
594		 * between the time we released the task lock, and the time
595		 * we add the task to the ready queue.
596		 */
597		task_ready(task);
598	}
599}
600
601ISC_TASKFUNC_SCOPE void
602isc__task_sendanddetach(isc_task_t **taskp, isc_event_t **eventp) {
603	isc_boolean_t idle1, idle2;
604	isc__task_t *task;
605
606	/*
607	 * Send '*event' to '*taskp' and then detach '*taskp' from its
608	 * task.
609	 */
610
611	REQUIRE(taskp != NULL);
612	task = (isc__task_t *)*taskp;
613	REQUIRE(VALID_TASK(task));
614
615	XTRACE("isc_task_sendanddetach");
616
617	LOCK(&task->lock);
618	idle1 = task_send(task, eventp);
619	idle2 = task_detach(task);
620	UNLOCK(&task->lock);
621
622	/*
623	 * If idle1, then idle2 shouldn't be true as well since we're holding
624	 * the task lock, and thus the task cannot switch from ready back to
625	 * idle.
626	 */
627	INSIST(!(idle1 && idle2));
628
629	if (idle1 || idle2)
630		task_ready(task);
631
632	*taskp = NULL;
633}
634
635#define PURGE_OK(event)	(((event)->ev_attributes & ISC_EVENTATTR_NOPURGE) == 0)
636
637static unsigned int
638dequeue_events(isc__task_t *task, void *sender, isc_eventtype_t first,
639	       isc_eventtype_t last, void *tag,
640	       isc_eventlist_t *events, isc_boolean_t purging)
641{
642	isc_event_t *event, *next_event;
643	unsigned int count = 0;
644
645	REQUIRE(VALID_TASK(task));
646	REQUIRE(last >= first);
647
648	XTRACE("dequeue_events");
649
650	/*
651	 * Events matching 'sender', whose type is >= first and <= last, and
652	 * whose tag is 'tag' will be dequeued.  If 'purging', matching events
653	 * which are marked as unpurgable will not be dequeued.
654	 *
655	 * sender == NULL means "any sender", and tag == NULL means "any tag".
656	 */
657
658	LOCK(&task->lock);
659
660	for (event = HEAD(task->events); event != NULL; event = next_event) {
661		next_event = NEXT(event, ev_link);
662		if (event->ev_type >= first && event->ev_type <= last &&
663		    (sender == NULL || event->ev_sender == sender) &&
664		    (tag == NULL || event->ev_tag == tag) &&
665		    (!purging || PURGE_OK(event))) {
666			DEQUEUE(task->events, event, ev_link);
667			ENQUEUE(*events, event, ev_link);
668			count++;
669		}
670	}
671
672	UNLOCK(&task->lock);
673
674	return (count);
675}
676
677ISC_TASKFUNC_SCOPE unsigned int
678isc__task_purgerange(isc_task_t *task0, void *sender, isc_eventtype_t first,
679		     isc_eventtype_t last, void *tag)
680{
681	isc__task_t *task = (isc__task_t *)task0;
682	unsigned int count;
683	isc_eventlist_t events;
684	isc_event_t *event, *next_event;
685
686	/*
687	 * Purge events from a task's event queue.
688	 */
689
690	XTRACE("isc_task_purgerange");
691
692	ISC_LIST_INIT(events);
693
694	count = dequeue_events(task, sender, first, last, tag, &events,
695			       ISC_TRUE);
696
697	for (event = HEAD(events); event != NULL; event = next_event) {
698		next_event = NEXT(event, ev_link);
699		isc_event_free(&event);
700	}
701
702	/*
703	 * Note that purging never changes the state of the task.
704	 */
705
706	return (count);
707}
708
709ISC_TASKFUNC_SCOPE unsigned int
710isc__task_purge(isc_task_t *task, void *sender, isc_eventtype_t type,
711		void *tag)
712{
713	/*
714	 * Purge events from a task's event queue.
715	 */
716
717	XTRACE("isc_task_purge");
718
719	return (isc__task_purgerange(task, sender, type, type, tag));
720}
721
722ISC_TASKFUNC_SCOPE isc_boolean_t
723isc__task_purgeevent(isc_task_t *task0, isc_event_t *event) {
724	isc__task_t *task = (isc__task_t *)task0;
725	isc_event_t *curr_event, *next_event;
726
727	/*
728	 * Purge 'event' from a task's event queue.
729	 *
730	 * XXXRTH:  WARNING:  This method may be removed before beta.
731	 */
732
733	REQUIRE(VALID_TASK(task));
734
735	/*
736	 * If 'event' is on the task's event queue, it will be purged,
737	 * unless it is marked as unpurgeable.  'event' does not have to be
738	 * on the task's event queue; in fact, it can even be an invalid
739	 * pointer.  Purging only occurs if the event is actually on the task's
740	 * event queue.
741	 *
742	 * Purging never changes the state of the task.
743	 */
744
745	LOCK(&task->lock);
746	for (curr_event = HEAD(task->events);
747	     curr_event != NULL;
748	     curr_event = next_event) {
749		next_event = NEXT(curr_event, ev_link);
750		if (curr_event == event && PURGE_OK(event)) {
751			DEQUEUE(task->events, curr_event, ev_link);
752			break;
753		}
754	}
755	UNLOCK(&task->lock);
756
757	if (curr_event == NULL)
758		return (ISC_FALSE);
759
760	isc_event_free(&curr_event);
761
762	return (ISC_TRUE);
763}
764
765ISC_TASKFUNC_SCOPE unsigned int
766isc__task_unsendrange(isc_task_t *task, void *sender, isc_eventtype_t first,
767		      isc_eventtype_t last, void *tag,
768		      isc_eventlist_t *events)
769{
770	/*
771	 * Remove events from a task's event queue.
772	 */
773
774	XTRACE("isc_task_unsendrange");
775
776	return (dequeue_events((isc__task_t *)task, sender, first,
777			       last, tag, events, ISC_FALSE));
778}
779
780ISC_TASKFUNC_SCOPE unsigned int
781isc__task_unsend(isc_task_t *task, void *sender, isc_eventtype_t type,
782		 void *tag, isc_eventlist_t *events)
783{
784	/*
785	 * Remove events from a task's event queue.
786	 */
787
788	XTRACE("isc_task_unsend");
789
790	return (dequeue_events((isc__task_t *)task, sender, type,
791			       type, tag, events, ISC_FALSE));
792}
793
794ISC_TASKFUNC_SCOPE isc_result_t
795isc__task_onshutdown(isc_task_t *task0, isc_taskaction_t action,
796		     const void *arg)
797{
798	isc__task_t *task = (isc__task_t *)task0;
799	isc_boolean_t disallowed = ISC_FALSE;
800	isc_result_t result = ISC_R_SUCCESS;
801	isc_event_t *event;
802
803	/*
804	 * Send a shutdown event with action 'action' and argument 'arg' when
805	 * 'task' is shutdown.
806	 */
807
808	REQUIRE(VALID_TASK(task));
809	REQUIRE(action != NULL);
810
811	event = isc_event_allocate(task->manager->mctx,
812				   NULL,
813				   ISC_TASKEVENT_SHUTDOWN,
814				   action,
815				   arg,
816				   sizeof(*event));
817	if (event == NULL)
818		return (ISC_R_NOMEMORY);
819
820	LOCK(&task->lock);
821	if (TASK_SHUTTINGDOWN(task)) {
822		disallowed = ISC_TRUE;
823		result = ISC_R_SHUTTINGDOWN;
824	} else
825		ENQUEUE(task->on_shutdown, event, ev_link);
826	UNLOCK(&task->lock);
827
828	if (disallowed)
829		isc_mem_put(task->manager->mctx, event, sizeof(*event));
830
831	return (result);
832}
833
834ISC_TASKFUNC_SCOPE void
835isc__task_shutdown(isc_task_t *task0) {
836	isc__task_t *task = (isc__task_t *)task0;
837	isc_boolean_t was_idle;
838
839	/*
840	 * Shutdown 'task'.
841	 */
842
843	REQUIRE(VALID_TASK(task));
844
845	LOCK(&task->lock);
846	was_idle = task_shutdown(task);
847	UNLOCK(&task->lock);
848
849	if (was_idle)
850		task_ready(task);
851}
852
853ISC_TASKFUNC_SCOPE void
854isc__task_destroy(isc_task_t **taskp) {
855
856	/*
857	 * Destroy '*taskp'.
858	 */
859
860	REQUIRE(taskp != NULL);
861
862	isc_task_shutdown(*taskp);
863	isc_task_detach(taskp);
864}
865
866ISC_TASKFUNC_SCOPE void
867isc__task_setname(isc_task_t *task0, const char *name, void *tag) {
868	isc__task_t *task = (isc__task_t *)task0;
869
870	/*
871	 * Name 'task'.
872	 */
873
874	REQUIRE(VALID_TASK(task));
875
876	LOCK(&task->lock);
877	memset(task->name, 0, sizeof(task->name));
878	strncpy(task->name, name, sizeof(task->name) - 1);
879	task->tag = tag;
880	UNLOCK(&task->lock);
881}
882
883ISC_TASKFUNC_SCOPE const char *
884isc__task_getname(isc_task_t *task0) {
885	isc__task_t *task = (isc__task_t *)task0;
886
887	REQUIRE(VALID_TASK(task));
888
889	return (task->name);
890}
891
892ISC_TASKFUNC_SCOPE void *
893isc__task_gettag(isc_task_t *task0) {
894	isc__task_t *task = (isc__task_t *)task0;
895
896	REQUIRE(VALID_TASK(task));
897
898	return (task->tag);
899}
900
901ISC_TASKFUNC_SCOPE void
902isc__task_getcurrenttime(isc_task_t *task0, isc_stdtime_t *t) {
903	isc__task_t *task = (isc__task_t *)task0;
904
905	REQUIRE(VALID_TASK(task));
906	REQUIRE(t != NULL);
907
908	LOCK(&task->lock);
909	*t = task->now;
910	UNLOCK(&task->lock);
911}
912
913/***
914 *** Task Manager.
915 ***/
916
917/*
918 * Return ISC_TRUE if the current ready list for the manager, which is
919 * either ready_tasks or the ready_priority_tasks, depending on whether
920 * the manager is currently in normal or privileged execution mode.
921 *
922 * Caller must hold the task manager lock.
923 */
924static inline isc_boolean_t
925empty_readyq(isc__taskmgr_t *manager) {
926	isc__tasklist_t queue;
927
928	if (manager->mode == isc_taskmgrmode_normal)
929		queue = manager->ready_tasks;
930	else
931		queue = manager->ready_priority_tasks;
932
933	return (ISC_TF(EMPTY(queue)));
934}
935
936/*
937 * Dequeue and return a pointer to the first task on the current ready
938 * list for the manager.
939 * If the task is privileged, dequeue it from the other ready list
940 * as well.
941 *
942 * Caller must hold the task manager lock.
943 */
944static inline isc__task_t *
945pop_readyq(isc__taskmgr_t *manager) {
946	isc__task_t *task;
947
948	if (manager->mode == isc_taskmgrmode_normal)
949		task = HEAD(manager->ready_tasks);
950	else
951		task = HEAD(manager->ready_priority_tasks);
952
953	if (task != NULL) {
954		DEQUEUE(manager->ready_tasks, task, ready_link);
955		if (ISC_LINK_LINKED(task, ready_priority_link))
956			DEQUEUE(manager->ready_priority_tasks, task,
957				ready_priority_link);
958	}
959
960	return (task);
961}
962
963/*
964 * Push 'task' onto the ready_tasks queue.  If 'task' has the privilege
965 * flag set, then also push it onto the ready_priority_tasks queue.
966 *
967 * Caller must hold the task manager lock.
968 */
969static inline void
970push_readyq(isc__taskmgr_t *manager, isc__task_t *task) {
971	ENQUEUE(manager->ready_tasks, task, ready_link);
972	if ((task->flags & TASK_F_PRIVILEGED) != 0)
973		ENQUEUE(manager->ready_priority_tasks, task,
974			ready_priority_link);
975}
976
977static void
978dispatch(isc__taskmgr_t *manager) {
979	isc__task_t *task;
980#ifndef USE_WORKER_THREADS
981	unsigned int total_dispatch_count = 0;
982	isc__tasklist_t new_ready_tasks;
983	isc__tasklist_t new_priority_tasks;
984#endif /* USE_WORKER_THREADS */
985
986	REQUIRE(VALID_MANAGER(manager));
987
988	/*
989	 * Again we're trying to hold the lock for as short a time as possible
990	 * and to do as little locking and unlocking as possible.
991	 *
992	 * In both while loops, the appropriate lock must be held before the
993	 * while body starts.  Code which acquired the lock at the top of
994	 * the loop would be more readable, but would result in a lot of
995	 * extra locking.  Compare:
996	 *
997	 * Straightforward:
998	 *
999	 *	LOCK();
1000	 *	...
1001	 *	UNLOCK();
1002	 *	while (expression) {
1003	 *		LOCK();
1004	 *		...
1005	 *		UNLOCK();
1006	 *
1007	 *	       	Unlocked part here...
1008	 *
1009	 *		LOCK();
1010	 *		...
1011	 *		UNLOCK();
1012	 *	}
1013	 *
1014	 * Note how if the loop continues we unlock and then immediately lock.
1015	 * For N iterations of the loop, this code does 2N+1 locks and 2N+1
1016	 * unlocks.  Also note that the lock is not held when the while
1017	 * condition is tested, which may or may not be important, depending
1018	 * on the expression.
1019	 *
1020	 * As written:
1021	 *
1022	 *	LOCK();
1023	 *	while (expression) {
1024	 *		...
1025	 *		UNLOCK();
1026	 *
1027	 *	       	Unlocked part here...
1028	 *
1029	 *		LOCK();
1030	 *		...
1031	 *	}
1032	 *	UNLOCK();
1033	 *
1034	 * For N iterations of the loop, this code does N+1 locks and N+1
1035	 * unlocks.  The while expression is always protected by the lock.
1036	 */
1037
1038#ifndef USE_WORKER_THREADS
1039	ISC_LIST_INIT(new_ready_tasks);
1040	ISC_LIST_INIT(new_priority_tasks);
1041#endif
1042	LOCK(&manager->lock);
1043
1044	while (!FINISHED(manager)) {
1045#ifdef USE_WORKER_THREADS
1046		/*
1047		 * For reasons similar to those given in the comment in
1048		 * isc_task_send() above, it is safe for us to dequeue
1049		 * the task while only holding the manager lock, and then
1050		 * change the task to running state while only holding the
1051		 * task lock.
1052		 *
1053		 * If a pause has been requested, don't do any work
1054		 * until it's been released.
1055		 */
1056		while ((empty_readyq(manager) || manager->pause_requested ||
1057			manager->exclusive_requested) && !FINISHED(manager))
1058		{
1059			XTHREADTRACE(isc_msgcat_get(isc_msgcat,
1060						    ISC_MSGSET_GENERAL,
1061						    ISC_MSG_WAIT, "wait"));
1062			WAIT(&manager->work_available, &manager->lock);
1063			XTHREADTRACE(isc_msgcat_get(isc_msgcat,
1064						    ISC_MSGSET_TASK,
1065						    ISC_MSG_AWAKE, "awake"));
1066		}
1067#else /* USE_WORKER_THREADS */
1068		if (total_dispatch_count >= DEFAULT_TASKMGR_QUANTUM ||
1069		    empty_readyq(manager))
1070			break;
1071#endif /* USE_WORKER_THREADS */
1072		XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_TASK,
1073					    ISC_MSG_WORKING, "working"));
1074
1075		task = pop_readyq(manager);
1076		if (task != NULL) {
1077			unsigned int dispatch_count = 0;
1078			isc_boolean_t done = ISC_FALSE;
1079			isc_boolean_t requeue = ISC_FALSE;
1080			isc_boolean_t finished = ISC_FALSE;
1081			isc_event_t *event;
1082
1083			INSIST(VALID_TASK(task));
1084
1085			/*
1086			 * Note we only unlock the manager lock if we actually
1087			 * have a task to do.  We must reacquire the manager
1088			 * lock before exiting the 'if (task != NULL)' block.
1089			 */
1090			manager->tasks_running++;
1091			UNLOCK(&manager->lock);
1092
1093			LOCK(&task->lock);
1094			INSIST(task->state == task_state_ready);
1095			task->state = task_state_running;
1096			XTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1097					      ISC_MSG_RUNNING, "running"));
1098			isc_stdtime_get(&task->now);
1099			do {
1100				if (!EMPTY(task->events)) {
1101					event = HEAD(task->events);
1102					DEQUEUE(task->events, event, ev_link);
1103
1104					/*
1105					 * Execute the event action.
1106					 */
1107					XTRACE(isc_msgcat_get(isc_msgcat,
1108							    ISC_MSGSET_TASK,
1109							    ISC_MSG_EXECUTE,
1110							    "execute action"));
1111					if (event->ev_action != NULL) {
1112						UNLOCK(&task->lock);
1113						(event->ev_action)(
1114							(isc_task_t *)task,
1115							event);
1116						LOCK(&task->lock);
1117					}
1118					dispatch_count++;
1119#ifndef USE_WORKER_THREADS
1120					total_dispatch_count++;
1121#endif /* USE_WORKER_THREADS */
1122				}
1123
1124				if (task->references == 0 &&
1125				    EMPTY(task->events) &&
1126				    !TASK_SHUTTINGDOWN(task)) {
1127					isc_boolean_t was_idle;
1128
1129					/*
1130					 * There are no references and no
1131					 * pending events for this task,
1132					 * which means it will not become
1133					 * runnable again via an external
1134					 * action (such as sending an event
1135					 * or detaching).
1136					 *
1137					 * We initiate shutdown to prevent
1138					 * it from becoming a zombie.
1139					 *
1140					 * We do this here instead of in
1141					 * the "if EMPTY(task->events)" block
1142					 * below because:
1143					 *
1144					 *	If we post no shutdown events,
1145					 *	we want the task to finish.
1146					 *
1147					 *	If we did post shutdown events,
1148					 *	will still want the task's
1149					 *	quantum to be applied.
1150					 */
1151					was_idle = task_shutdown(task);
1152					INSIST(!was_idle);
1153				}
1154
1155				if (EMPTY(task->events)) {
1156					/*
1157					 * Nothing else to do for this task
1158					 * right now.
1159					 */
1160					XTRACE(isc_msgcat_get(isc_msgcat,
1161							      ISC_MSGSET_TASK,
1162							      ISC_MSG_EMPTY,
1163							      "empty"));
1164					if (task->references == 0 &&
1165					    TASK_SHUTTINGDOWN(task)) {
1166						/*
1167						 * The task is done.
1168						 */
1169						XTRACE(isc_msgcat_get(
1170							       isc_msgcat,
1171							       ISC_MSGSET_TASK,
1172							       ISC_MSG_DONE,
1173							       "done"));
1174						finished = ISC_TRUE;
1175						task->state = task_state_done;
1176					} else
1177						task->state = task_state_idle;
1178					done = ISC_TRUE;
1179				} else if (dispatch_count >= task->quantum) {
1180					/*
1181					 * Our quantum has expired, but
1182					 * there is more work to be done.
1183					 * We'll requeue it to the ready
1184					 * queue later.
1185					 *
1186					 * We don't check quantum until
1187					 * dispatching at least one event,
1188					 * so the minimum quantum is one.
1189					 */
1190					XTRACE(isc_msgcat_get(isc_msgcat,
1191							      ISC_MSGSET_TASK,
1192							      ISC_MSG_QUANTUM,
1193							      "quantum"));
1194					task->state = task_state_ready;
1195					requeue = ISC_TRUE;
1196					done = ISC_TRUE;
1197				}
1198			} while (!done);
1199			UNLOCK(&task->lock);
1200
1201			if (finished)
1202				task_finished(task);
1203
1204			LOCK(&manager->lock);
1205			manager->tasks_running--;
1206#ifdef USE_WORKER_THREADS
1207			if (manager->exclusive_requested &&
1208			    manager->tasks_running == 1) {
1209				SIGNAL(&manager->exclusive_granted);
1210			} else if (manager->pause_requested &&
1211				   manager->tasks_running == 0) {
1212				SIGNAL(&manager->paused);
1213			}
1214#endif /* USE_WORKER_THREADS */
1215			if (requeue) {
1216				/*
1217				 * We know we're awake, so we don't have
1218				 * to wakeup any sleeping threads if the
1219				 * ready queue is empty before we requeue.
1220				 *
1221				 * A possible optimization if the queue is
1222				 * empty is to 'goto' the 'if (task != NULL)'
1223				 * block, avoiding the ENQUEUE of the task
1224				 * and the subsequent immediate DEQUEUE
1225				 * (since it is the only executable task).
1226				 * We don't do this because then we'd be
1227				 * skipping the exit_requested check.  The
1228				 * cost of ENQUEUE is low anyway, especially
1229				 * when you consider that we'd have to do
1230				 * an extra EMPTY check to see if we could
1231				 * do the optimization.  If the ready queue
1232				 * were usually nonempty, the 'optimization'
1233				 * might even hurt rather than help.
1234				 */
1235#ifdef USE_WORKER_THREADS
1236				push_readyq(manager, task);
1237#else
1238				ENQUEUE(new_ready_tasks, task, ready_link);
1239				if ((task->flags & TASK_F_PRIVILEGED) != 0)
1240					ENQUEUE(new_priority_tasks, task,
1241						ready_priority_link);
1242#endif
1243			}
1244		}
1245
1246#ifdef USE_WORKER_THREADS
1247		/*
1248		 * If we are in privileged execution mode and there are no
1249		 * tasks remaining on the current ready queue, then
1250		 * we're stuck.  Automatically drop privileges at that
1251		 * point and continue with the regular ready queue.
1252		 */
1253		if (manager->tasks_running == 0 && empty_readyq(manager)) {
1254			manager->mode = isc_taskmgrmode_normal;
1255			if (!empty_readyq(manager))
1256				BROADCAST(&manager->work_available);
1257		}
1258#endif
1259	}
1260
1261#ifndef USE_WORKER_THREADS
1262	ISC_LIST_APPENDLIST(manager->ready_tasks, new_ready_tasks, ready_link);
1263	ISC_LIST_APPENDLIST(manager->ready_priority_tasks, new_priority_tasks,
1264			    ready_priority_link);
1265	if (empty_readyq(manager))
1266		manager->mode = isc_taskmgrmode_normal;
1267#endif
1268
1269	UNLOCK(&manager->lock);
1270}
1271
1272#ifdef USE_WORKER_THREADS
1273static isc_threadresult_t
1274#ifdef _WIN32
1275WINAPI
1276#endif
1277run(void *uap) {
1278	isc__taskmgr_t *manager = uap;
1279
1280	XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1281				    ISC_MSG_STARTING, "starting"));
1282
1283	dispatch(manager);
1284
1285	XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1286				    ISC_MSG_EXITING, "exiting"));
1287
1288#ifdef OPENSSL_LEAKS
1289	ERR_remove_state(0);
1290#endif
1291
1292	return ((isc_threadresult_t)0);
1293}
1294#endif /* USE_WORKER_THREADS */
1295
1296static void
1297manager_free(isc__taskmgr_t *manager) {
1298	isc_mem_t *mctx;
1299
1300	LOCK(&manager->lock);
1301#ifdef USE_WORKER_THREADS
1302	(void)isc_condition_destroy(&manager->exclusive_granted);
1303	(void)isc_condition_destroy(&manager->work_available);
1304	(void)isc_condition_destroy(&manager->paused);
1305	isc_mem_free(manager->mctx, manager->threads);
1306#endif /* USE_WORKER_THREADS */
1307	manager->common.impmagic = 0;
1308	manager->common.magic = 0;
1309	mctx = manager->mctx;
1310	UNLOCK(&manager->lock);
1311	DESTROYLOCK(&manager->lock);
1312	isc_mem_put(mctx, manager, sizeof(*manager));
1313	isc_mem_detach(&mctx);
1314
1315#ifdef USE_SHARED_MANAGER
1316	taskmgr = NULL;
1317#endif	/* USE_SHARED_MANAGER */
1318}
1319
1320ISC_TASKFUNC_SCOPE isc_result_t
1321isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
1322		    unsigned int default_quantum, isc_taskmgr_t **managerp)
1323{
1324	isc_result_t result;
1325	unsigned int i, started = 0;
1326	isc__taskmgr_t *manager;
1327
1328	/*
1329	 * Create a new task manager.
1330	 */
1331
1332	REQUIRE(workers > 0);
1333	REQUIRE(managerp != NULL && *managerp == NULL);
1334
1335#ifndef USE_WORKER_THREADS
1336	UNUSED(i);
1337	UNUSED(started);
1338#endif
1339
1340#ifdef USE_SHARED_MANAGER
1341	if (taskmgr != NULL) {
1342		if (taskmgr->refs == 0)
1343			return (ISC_R_SHUTTINGDOWN);
1344		taskmgr->refs++;
1345		*managerp = (isc_taskmgr_t *)taskmgr;
1346		return (ISC_R_SUCCESS);
1347	}
1348#endif /* USE_SHARED_MANAGER */
1349
1350	manager = isc_mem_get(mctx, sizeof(*manager));
1351	if (manager == NULL)
1352		return (ISC_R_NOMEMORY);
1353	manager->common.methods = &taskmgrmethods;
1354	manager->common.impmagic = TASK_MANAGER_MAGIC;
1355	manager->common.magic = ISCAPI_TASKMGR_MAGIC;
1356	manager->mode = isc_taskmgrmode_normal;
1357	manager->mctx = NULL;
1358	result = isc_mutex_init(&manager->lock);
1359	if (result != ISC_R_SUCCESS)
1360		goto cleanup_mgr;
1361	LOCK(&manager->lock);
1362
1363#ifdef USE_WORKER_THREADS
1364	manager->workers = 0;
1365	manager->threads = isc_mem_allocate(mctx,
1366					    workers * sizeof(isc_thread_t));
1367	if (manager->threads == NULL) {
1368		result = ISC_R_NOMEMORY;
1369		goto cleanup_lock;
1370	}
1371	if (isc_condition_init(&manager->work_available) != ISC_R_SUCCESS) {
1372		UNEXPECTED_ERROR(__FILE__, __LINE__,
1373				 "isc_condition_init() %s",
1374				 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1375						ISC_MSG_FAILED, "failed"));
1376		result = ISC_R_UNEXPECTED;
1377		goto cleanup_threads;
1378	}
1379	if (isc_condition_init(&manager->exclusive_granted) != ISC_R_SUCCESS) {
1380		UNEXPECTED_ERROR(__FILE__, __LINE__,
1381				 "isc_condition_init() %s",
1382				 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1383						ISC_MSG_FAILED, "failed"));
1384		result = ISC_R_UNEXPECTED;
1385		goto cleanup_workavailable;
1386	}
1387	if (isc_condition_init(&manager->paused) != ISC_R_SUCCESS) {
1388		UNEXPECTED_ERROR(__FILE__, __LINE__,
1389				 "isc_condition_init() %s",
1390				 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1391						ISC_MSG_FAILED, "failed"));
1392		result = ISC_R_UNEXPECTED;
1393		goto cleanup_exclusivegranted;
1394	}
1395#endif /* USE_WORKER_THREADS */
1396	if (default_quantum == 0)
1397		default_quantum = DEFAULT_DEFAULT_QUANTUM;
1398	manager->default_quantum = default_quantum;
1399	INIT_LIST(manager->tasks);
1400	INIT_LIST(manager->ready_tasks);
1401	INIT_LIST(manager->ready_priority_tasks);
1402	manager->tasks_running = 0;
1403	manager->exclusive_requested = ISC_FALSE;
1404	manager->pause_requested = ISC_FALSE;
1405	manager->exiting = ISC_FALSE;
1406
1407	isc_mem_attach(mctx, &manager->mctx);
1408
1409#ifdef USE_WORKER_THREADS
1410	/*
1411	 * Start workers.
1412	 */
1413	for (i = 0; i < workers; i++) {
1414		if (isc_thread_create(run, manager,
1415				      &manager->threads[manager->workers]) ==
1416		    ISC_R_SUCCESS) {
1417			manager->workers++;
1418			started++;
1419		}
1420	}
1421	UNLOCK(&manager->lock);
1422
1423	if (started == 0) {
1424		manager_free(manager);
1425		return (ISC_R_NOTHREADS);
1426	}
1427	isc_thread_setconcurrency(workers);
1428#endif /* USE_WORKER_THREADS */
1429#ifdef USE_SHARED_MANAGER
1430	manager->refs = 1;
1431	UNLOCK(&manager->lock);
1432	taskmgr = manager;
1433#endif /* USE_SHARED_MANAGER */
1434
1435	*managerp = (isc_taskmgr_t *)manager;
1436
1437	return (ISC_R_SUCCESS);
1438
1439#ifdef USE_WORKER_THREADS
1440 cleanup_exclusivegranted:
1441	(void)isc_condition_destroy(&manager->exclusive_granted);
1442 cleanup_workavailable:
1443	(void)isc_condition_destroy(&manager->work_available);
1444 cleanup_threads:
1445	isc_mem_free(mctx, manager->threads);
1446 cleanup_lock:
1447	UNLOCK(&manager->lock);
1448	DESTROYLOCK(&manager->lock);
1449#endif
1450 cleanup_mgr:
1451	isc_mem_put(mctx, manager, sizeof(*manager));
1452	return (result);
1453}
1454
1455ISC_TASKFUNC_SCOPE void
1456isc__taskmgr_destroy(isc_taskmgr_t **managerp) {
1457	isc__taskmgr_t *manager;
1458	isc__task_t *task;
1459	unsigned int i;
1460
1461	/*
1462	 * Destroy '*managerp'.
1463	 */
1464
1465	REQUIRE(managerp != NULL);
1466	manager = (void*)(*managerp);
1467	REQUIRE(VALID_MANAGER(manager));
1468
1469#ifndef USE_WORKER_THREADS
1470	UNUSED(i);
1471#endif /* USE_WORKER_THREADS */
1472
1473#ifdef USE_SHARED_MANAGER
1474	manager->refs--;
1475	if (manager->refs > 0) {
1476		*managerp = NULL;
1477		return;
1478	}
1479#endif
1480
1481	XTHREADTRACE("isc_taskmgr_destroy");
1482	/*
1483	 * Only one non-worker thread may ever call this routine.
1484	 * If a worker thread wants to initiate shutdown of the
1485	 * task manager, it should ask some non-worker thread to call
1486	 * isc_taskmgr_destroy(), e.g. by signalling a condition variable
1487	 * that the startup thread is sleeping on.
1488	 */
1489
1490	/*
1491	 * Unlike elsewhere, we're going to hold this lock a long time.
1492	 * We need to do so, because otherwise the list of tasks could
1493	 * change while we were traversing it.
1494	 *
1495	 * This is also the only function where we will hold both the
1496	 * task manager lock and a task lock at the same time.
1497	 */
1498
1499	LOCK(&manager->lock);
1500
1501	/*
1502	 * Make sure we only get called once.
1503	 */
1504	INSIST(!manager->exiting);
1505	manager->exiting = ISC_TRUE;
1506
1507	/*
1508	 * If privileged mode was on, turn it off.
1509	 */
1510	manager->mode = isc_taskmgrmode_normal;
1511
1512	/*
1513	 * Post shutdown event(s) to every task (if they haven't already been
1514	 * posted).
1515	 */
1516	for (task = HEAD(manager->tasks);
1517	     task != NULL;
1518	     task = NEXT(task, link)) {
1519		LOCK(&task->lock);
1520		if (task_shutdown(task))
1521			push_readyq(manager, task);
1522		UNLOCK(&task->lock);
1523	}
1524#ifdef USE_WORKER_THREADS
1525	/*
1526	 * Wake up any sleeping workers.  This ensures we get work done if
1527	 * there's work left to do, and if there are already no tasks left
1528	 * it will cause the workers to see manager->exiting.
1529	 */
1530	BROADCAST(&manager->work_available);
1531	UNLOCK(&manager->lock);
1532
1533	/*
1534	 * Wait for all the worker threads to exit.
1535	 */
1536	for (i = 0; i < manager->workers; i++)
1537		(void)isc_thread_join(manager->threads[i], NULL);
1538#else /* USE_WORKER_THREADS */
1539	/*
1540	 * Dispatch the shutdown events.
1541	 */
1542	UNLOCK(&manager->lock);
1543	while (isc__taskmgr_ready((isc_taskmgr_t *)manager))
1544		(void)isc__taskmgr_dispatch((isc_taskmgr_t *)manager);
1545#ifdef BIND9
1546	if (!ISC_LIST_EMPTY(manager->tasks))
1547		isc_mem_printallactive(stderr);
1548#endif
1549	INSIST(ISC_LIST_EMPTY(manager->tasks));
1550#ifdef USE_SHARED_MANAGER
1551	taskmgr = NULL;
1552#endif
1553#endif /* USE_WORKER_THREADS */
1554
1555	manager_free(manager);
1556
1557	*managerp = NULL;
1558}
1559
1560ISC_TASKFUNC_SCOPE void
1561isc__taskmgr_setmode(isc_taskmgr_t *manager0, isc_taskmgrmode_t mode) {
1562	isc__taskmgr_t *manager = (void*)manager0;
1563
1564	LOCK(&manager->lock);
1565	manager->mode = mode;
1566	UNLOCK(&manager->lock);
1567}
1568
1569ISC_TASKFUNC_SCOPE isc_taskmgrmode_t
1570isc__taskmgr_mode(isc_taskmgr_t *manager0) {
1571	isc__taskmgr_t *manager = (void*)manager0;
1572	isc_taskmgrmode_t mode;
1573	LOCK(&manager->lock);
1574	mode = manager->mode;
1575	UNLOCK(&manager->lock);
1576	return (mode);
1577}
1578
1579#ifndef USE_WORKER_THREADS
1580isc_boolean_t
1581isc__taskmgr_ready(isc_taskmgr_t *manager0) {
1582	isc__taskmgr_t *manager = (void*)manager0;
1583	isc_boolean_t is_ready;
1584
1585#ifdef USE_SHARED_MANAGER
1586	if (manager == NULL)
1587		manager = taskmgr;
1588#endif
1589	if (manager == NULL)
1590		return (ISC_FALSE);
1591
1592	LOCK(&manager->lock);
1593	is_ready = !empty_readyq(manager);
1594	UNLOCK(&manager->lock);
1595
1596	return (is_ready);
1597}
1598
1599isc_result_t
1600isc__taskmgr_dispatch(isc_taskmgr_t *manager0) {
1601	isc__taskmgr_t *manager = (void*)manager0;
1602
1603#ifdef USE_SHARED_MANAGER
1604	if (manager == NULL)
1605		manager = taskmgr;
1606#endif
1607	if (manager == NULL)
1608		return (ISC_R_NOTFOUND);
1609
1610	dispatch(manager);
1611
1612	return (ISC_R_SUCCESS);
1613}
1614
1615#else
1616ISC_TASKFUNC_SCOPE void
1617isc__taskmgr_pause(isc_taskmgr_t *manager0) {
1618	isc__taskmgr_t *manager = (void*)manager0;
1619	LOCK(&manager->lock);
1620	while (manager->tasks_running > 0) {
1621		WAIT(&manager->paused, &manager->lock);
1622	}
1623	manager->pause_requested = ISC_TRUE;
1624	UNLOCK(&manager->lock);
1625}
1626
1627ISC_TASKFUNC_SCOPE void
1628isc__taskmgr_resume(isc_taskmgr_t *manager0) {
1629	isc__taskmgr_t *manager = (void*)manager0;
1630
1631	LOCK(&manager->lock);
1632	if (manager->pause_requested) {
1633		manager->pause_requested = ISC_FALSE;
1634		BROADCAST(&manager->work_available);
1635	}
1636	UNLOCK(&manager->lock);
1637}
1638#endif /* USE_WORKER_THREADS */
1639
1640ISC_TASKFUNC_SCOPE isc_result_t
1641isc__task_beginexclusive(isc_task_t *task0) {
1642#ifdef USE_WORKER_THREADS
1643	isc__task_t *task = (isc__task_t *)task0;
1644	isc__taskmgr_t *manager = task->manager;
1645	REQUIRE(task->state == task_state_running);
1646	LOCK(&manager->lock);
1647	if (manager->exclusive_requested) {
1648		UNLOCK(&manager->lock);
1649		return (ISC_R_LOCKBUSY);
1650	}
1651	manager->exclusive_requested = ISC_TRUE;
1652	while (manager->tasks_running > 1) {
1653		WAIT(&manager->exclusive_granted, &manager->lock);
1654	}
1655	UNLOCK(&manager->lock);
1656#else
1657	UNUSED(task0);
1658#endif
1659	return (ISC_R_SUCCESS);
1660}
1661
1662ISC_TASKFUNC_SCOPE void
1663isc__task_endexclusive(isc_task_t *task0) {
1664#ifdef USE_WORKER_THREADS
1665	isc__task_t *task = (isc__task_t *)task0;
1666	isc__taskmgr_t *manager = task->manager;
1667
1668	REQUIRE(task->state == task_state_running);
1669	LOCK(&manager->lock);
1670	REQUIRE(manager->exclusive_requested);
1671	manager->exclusive_requested = ISC_FALSE;
1672	BROADCAST(&manager->work_available);
1673	UNLOCK(&manager->lock);
1674#else
1675	UNUSED(task0);
1676#endif
1677}
1678
1679ISC_TASKFUNC_SCOPE void
1680isc__task_setprivilege(isc_task_t *task0, isc_boolean_t priv) {
1681	isc__task_t *task = (isc__task_t *)task0;
1682	isc__taskmgr_t *manager = task->manager;
1683	isc_boolean_t oldpriv;
1684
1685	LOCK(&task->lock);
1686	oldpriv = ISC_TF((task->flags & TASK_F_PRIVILEGED) != 0);
1687	if (priv)
1688		task->flags |= TASK_F_PRIVILEGED;
1689	else
1690		task->flags &= ~TASK_F_PRIVILEGED;
1691	UNLOCK(&task->lock);
1692
1693	if (priv == oldpriv)
1694		return;
1695
1696	LOCK(&manager->lock);
1697	if (priv && ISC_LINK_LINKED(task, ready_link))
1698		ENQUEUE(manager->ready_priority_tasks, task,
1699			ready_priority_link);
1700	else if (!priv && ISC_LINK_LINKED(task, ready_priority_link))
1701		DEQUEUE(manager->ready_priority_tasks, task,
1702			ready_priority_link);
1703	UNLOCK(&manager->lock);
1704}
1705
1706ISC_TASKFUNC_SCOPE isc_boolean_t
1707isc__task_privilege(isc_task_t *task0) {
1708	isc__task_t *task = (isc__task_t *)task0;
1709	isc_boolean_t priv;
1710
1711	LOCK(&task->lock);
1712	priv = ISC_TF((task->flags & TASK_F_PRIVILEGED) != 0);
1713	UNLOCK(&task->lock);
1714	return (priv);
1715}
1716
1717#ifdef USE_SOCKETIMPREGISTER
1718isc_result_t
1719isc__task_register() {
1720	return (isc_task_register(isc__taskmgr_create));
1721}
1722#endif
1723
1724isc_boolean_t
1725isc_task_exiting(isc_task_t *t) {
1726	isc__task_t *task = (isc__task_t *)t;
1727
1728	REQUIRE(VALID_TASK(task));
1729	return (TASK_SHUTTINGDOWN(task));
1730}
1731
1732
1733#if defined(HAVE_LIBXML2) && defined(BIND9)
1734void
1735isc_taskmgr_renderxml(isc_taskmgr_t *mgr0, xmlTextWriterPtr writer) {
1736	isc__taskmgr_t *mgr = (isc__taskmgr_t *)mgr0;
1737	isc__task_t *task;
1738
1739	LOCK(&mgr->lock);
1740
1741	/*
1742	 * Write out the thread-model, and some details about each depending
1743	 * on which type is enabled.
1744	 */
1745	xmlTextWriterStartElement(writer, ISC_XMLCHAR "thread-model");
1746#ifdef ISC_PLATFORM_USETHREADS
1747	xmlTextWriterStartElement(writer, ISC_XMLCHAR "type");
1748	xmlTextWriterWriteString(writer, ISC_XMLCHAR "threaded");
1749	xmlTextWriterEndElement(writer); /* type */
1750
1751	xmlTextWriterStartElement(writer, ISC_XMLCHAR "worker-threads");
1752	xmlTextWriterWriteFormatString(writer, "%d", mgr->workers);
1753	xmlTextWriterEndElement(writer); /* worker-threads */
1754#else /* ISC_PLATFORM_USETHREADS */
1755	xmlTextWriterStartElement(writer, ISC_XMLCHAR "type");
1756	xmlTextWriterWriteString(writer, ISC_XMLCHAR "non-threaded");
1757	xmlTextWriterEndElement(writer); /* type */
1758
1759	xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
1760	xmlTextWriterWriteFormatString(writer, "%d", mgr->refs);
1761	xmlTextWriterEndElement(writer); /* references */
1762#endif /* ISC_PLATFORM_USETHREADS */
1763
1764	xmlTextWriterStartElement(writer, ISC_XMLCHAR "default-quantum");
1765	xmlTextWriterWriteFormatString(writer, "%d", mgr->default_quantum);
1766	xmlTextWriterEndElement(writer); /* default-quantum */
1767
1768	xmlTextWriterStartElement(writer, ISC_XMLCHAR "tasks-running");
1769	xmlTextWriterWriteFormatString(writer, "%d", mgr->tasks_running);
1770	xmlTextWriterEndElement(writer); /* tasks-running */
1771
1772	xmlTextWriterEndElement(writer); /* thread-model */
1773
1774	xmlTextWriterStartElement(writer, ISC_XMLCHAR "tasks");
1775	task = ISC_LIST_HEAD(mgr->tasks);
1776	while (task != NULL) {
1777		LOCK(&task->lock);
1778		xmlTextWriterStartElement(writer, ISC_XMLCHAR "task");
1779
1780		if (task->name[0] != 0) {
1781			xmlTextWriterStartElement(writer, ISC_XMLCHAR "name");
1782			xmlTextWriterWriteFormatString(writer, "%s",
1783						       task->name);
1784			xmlTextWriterEndElement(writer); /* name */
1785		}
1786
1787		xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
1788		xmlTextWriterWriteFormatString(writer, "%d", task->references);
1789		xmlTextWriterEndElement(writer); /* references */
1790
1791		xmlTextWriterStartElement(writer, ISC_XMLCHAR "id");
1792		xmlTextWriterWriteFormatString(writer, "%p", task);
1793		xmlTextWriterEndElement(writer); /* id */
1794
1795		xmlTextWriterStartElement(writer, ISC_XMLCHAR "state");
1796		xmlTextWriterWriteFormatString(writer, "%s",
1797					       statenames[task->state]);
1798		xmlTextWriterEndElement(writer); /* state */
1799
1800		xmlTextWriterStartElement(writer, ISC_XMLCHAR "quantum");
1801		xmlTextWriterWriteFormatString(writer, "%d", task->quantum);
1802		xmlTextWriterEndElement(writer); /* quantum */
1803
1804		xmlTextWriterEndElement(writer);
1805
1806		UNLOCK(&task->lock);
1807		task = ISC_LIST_NEXT(task, link);
1808	}
1809	xmlTextWriterEndElement(writer); /* tasks */
1810
1811	UNLOCK(&mgr->lock);
1812}
1813#endif /* HAVE_LIBXML2 && BIND9 */
1814