Patrick Welche <prlw1@cam.ac.uk>
[netbsd-mini2440.git] / external / bsd / ntp / dist / lib / isc / rwlock.c
blob9983c2972c4bb5e09916427f7cafb3a10f6a5ada
1 /* $NetBSD$ */
3 /*
4 * Copyright (C) 2004, 2005, 2007, 2009 Internet Systems Consortium, Inc. ("ISC")
5 * Copyright (C) 1998-2001, 2003 Internet Software Consortium.
7 * Permission to use, copy, modify, and/or distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
11 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
12 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
13 * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
14 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
15 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
16 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
17 * PERFORMANCE OF THIS SOFTWARE.
20 /* Id: rwlock.c,v 1.44.332.2 2009/01/18 23:47:41 tbox Exp */
22 /*! \file */
24 #include <config.h>
26 #include <stddef.h>
28 #include <isc/atomic.h>
29 #include <isc/magic.h>
30 #include <isc/msgs.h>
31 #include <isc/platform.h>
32 #include <isc/rwlock.h>
33 #include <isc/util.h>
35 #define RWLOCK_MAGIC ISC_MAGIC('R', 'W', 'L', 'k')
36 #define VALID_RWLOCK(rwl) ISC_MAGIC_VALID(rwl, RWLOCK_MAGIC)
38 #ifdef ISC_PLATFORM_USETHREADS
40 #ifndef RWLOCK_DEFAULT_READ_QUOTA
41 #define RWLOCK_DEFAULT_READ_QUOTA 4
42 #endif
44 #ifndef RWLOCK_DEFAULT_WRITE_QUOTA
45 #define RWLOCK_DEFAULT_WRITE_QUOTA 4
46 #endif
48 #ifdef ISC_RWLOCK_TRACE
49 #include <stdio.h> /* Required for fprintf/stderr. */
50 #include <isc/thread.h> /* Required for isc_thread_self(). */
52 static void
53 print_lock(const char *operation, isc_rwlock_t *rwl, isc_rwlocktype_t type) {
54 fprintf(stderr,
55 isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
56 ISC_MSG_PRINTLOCK,
57 "rwlock %p thread %lu %s(%s): %s, %u active, "
58 "%u granted, %u rwaiting, %u wwaiting\n"),
59 rwl, isc_thread_self(), operation,
60 (type == isc_rwlocktype_read ?
61 isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
62 ISC_MSG_READ, "read") :
63 isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
64 ISC_MSG_WRITE, "write")),
65 (rwl->type == isc_rwlocktype_read ?
66 isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
67 ISC_MSG_READING, "reading") :
68 isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
69 ISC_MSG_WRITING, "writing")),
70 rwl->active, rwl->granted, rwl->readers_waiting,
71 rwl->writers_waiting);
73 #endif
75 isc_result_t
76 isc_rwlock_init(isc_rwlock_t *rwl, unsigned int read_quota,
77 unsigned int write_quota)
79 isc_result_t result;
81 REQUIRE(rwl != NULL);
84 * In case there's trouble initializing, we zero magic now. If all
85 * goes well, we'll set it to RWLOCK_MAGIC.
87 rwl->magic = 0;
89 #if defined(ISC_PLATFORM_HAVEXADD) && defined(ISC_PLATFORM_HAVECMPXCHG)
90 rwl->write_requests = 0;
91 rwl->write_completions = 0;
92 rwl->cnt_and_flag = 0;
93 rwl->readers_waiting = 0;
94 rwl->write_granted = 0;
95 if (read_quota != 0) {
96 UNEXPECTED_ERROR(__FILE__, __LINE__,
97 "read quota is not supported");
99 if (write_quota == 0)
100 write_quota = RWLOCK_DEFAULT_WRITE_QUOTA;
101 rwl->write_quota = write_quota;
102 #else
103 rwl->type = isc_rwlocktype_read;
104 rwl->original = isc_rwlocktype_none;
105 rwl->active = 0;
106 rwl->granted = 0;
107 rwl->readers_waiting = 0;
108 rwl->writers_waiting = 0;
109 if (read_quota == 0)
110 read_quota = RWLOCK_DEFAULT_READ_QUOTA;
111 rwl->read_quota = read_quota;
112 if (write_quota == 0)
113 write_quota = RWLOCK_DEFAULT_WRITE_QUOTA;
114 rwl->write_quota = write_quota;
115 #endif
117 result = isc_mutex_init(&rwl->lock);
118 if (result != ISC_R_SUCCESS)
119 return (result);
121 result = isc_condition_init(&rwl->readable);
122 if (result != ISC_R_SUCCESS) {
123 UNEXPECTED_ERROR(__FILE__, __LINE__,
124 "isc_condition_init(readable) %s: %s",
125 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
126 ISC_MSG_FAILED, "failed"),
127 isc_result_totext(result));
128 result = ISC_R_UNEXPECTED;
129 goto destroy_lock;
131 result = isc_condition_init(&rwl->writeable);
132 if (result != ISC_R_SUCCESS) {
133 UNEXPECTED_ERROR(__FILE__, __LINE__,
134 "isc_condition_init(writeable) %s: %s",
135 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
136 ISC_MSG_FAILED, "failed"),
137 isc_result_totext(result));
138 result = ISC_R_UNEXPECTED;
139 goto destroy_rcond;
142 rwl->magic = RWLOCK_MAGIC;
144 return (ISC_R_SUCCESS);
146 destroy_rcond:
147 (void)isc_condition_destroy(&rwl->readable);
148 destroy_lock:
149 DESTROYLOCK(&rwl->lock);
151 return (result);
154 void
155 isc_rwlock_destroy(isc_rwlock_t *rwl) {
156 REQUIRE(VALID_RWLOCK(rwl));
158 #if defined(ISC_PLATFORM_HAVEXADD) && defined(ISC_PLATFORM_HAVECMPXCHG)
159 REQUIRE(rwl->write_requests == rwl->write_completions &&
160 rwl->cnt_and_flag == 0 && rwl->readers_waiting == 0);
161 #else
162 LOCK(&rwl->lock);
163 REQUIRE(rwl->active == 0 &&
164 rwl->readers_waiting == 0 &&
165 rwl->writers_waiting == 0);
166 UNLOCK(&rwl->lock);
167 #endif
169 rwl->magic = 0;
170 (void)isc_condition_destroy(&rwl->readable);
171 (void)isc_condition_destroy(&rwl->writeable);
172 DESTROYLOCK(&rwl->lock);
175 #if defined(ISC_PLATFORM_HAVEXADD) && defined(ISC_PLATFORM_HAVECMPXCHG)
178 * When some architecture-dependent atomic operations are available,
179 * rwlock can be more efficient than the generic algorithm defined below.
180 * The basic algorithm is described in the following URL:
181 * http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/rw.html
183 * The key is to use the following integer variables modified atomically:
184 * write_requests, write_completions, and cnt_and_flag.
186 * write_requests and write_completions act as a waiting queue for writers
187 * in order to ensure the FIFO order. Both variables begin with the initial
188 * value of 0. When a new writer tries to get a write lock, it increments
189 * write_requests and gets the previous value of the variable as a "ticket".
190 * When write_completions reaches the ticket number, the new writer can start
191 * writing. When the writer completes its work, it increments
192 * write_completions so that another new writer can start working. If the
193 * write_requests is not equal to write_completions, it means a writer is now
194 * working or waiting. In this case, a new readers cannot start reading, or
195 * in other words, this algorithm basically prefers writers.
197 * cnt_and_flag is a "lock" shared by all readers and writers. This integer
198 * variable is a kind of structure with two members: writer_flag (1 bit) and
199 * reader_count (31 bits). The writer_flag shows whether a writer is working,
200 * and the reader_count shows the number of readers currently working or almost
201 * ready for working. A writer who has the current "ticket" tries to get the
202 * lock by exclusively setting the writer_flag to 1, provided that the whole
203 * 32-bit is 0 (meaning no readers or writers working). On the other hand,
204 * a new reader tries to increment the "reader_count" field provided that
205 * the writer_flag is 0 (meaning there is no writer working).
207 * If some of the above operations fail, the reader or the writer sleeps
208 * until the related condition changes. When a working reader or writer
209 * completes its work, some readers or writers are sleeping, and the condition
210 * that suspended the reader or writer has changed, it wakes up the sleeping
211 * readers or writers.
213 * As already noted, this algorithm basically prefers writers. In order to
214 * prevent readers from starving, however, the algorithm also introduces the
215 * "writer quota" (Q). When Q consecutive writers have completed their work,
216 * suspending readers, the last writer will wake up the readers, even if a new
217 * writer is waiting.
219 * Implementation specific note: due to the combination of atomic operations
220 * and a mutex lock, ordering between the atomic operation and locks can be
221 * very sensitive in some cases. In particular, it is generally very important
222 * to check the atomic variable that requires a reader or writer to sleep after
223 * locking the mutex and before actually sleeping; otherwise, it could be very
224 * likely to cause a deadlock. For example, assume "var" is a variable
225 * atomically modified, then the corresponding code would be:
226 * if (var == need_sleep) {
227 * LOCK(lock);
228 * if (var == need_sleep)
229 * WAIT(cond, lock);
230 * UNLOCK(lock);
232 * The second check is important, since "var" is protected by the atomic
233 * operation, not by the mutex, and can be changed just before sleeping.
234 * (The first "if" could be omitted, but this is also important in order to
235 * make the code efficient by avoiding the use of the mutex unless it is
236 * really necessary.)
239 #define WRITER_ACTIVE 0x1
240 #define READER_INCR 0x2
242 isc_result_t
243 isc_rwlock_lock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
244 isc_int32_t cntflag;
246 REQUIRE(VALID_RWLOCK(rwl));
248 #ifdef ISC_RWLOCK_TRACE
249 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
250 ISC_MSG_PRELOCK, "prelock"), rwl, type);
251 #endif
253 if (type == isc_rwlocktype_read) {
254 if (rwl->write_requests != rwl->write_completions) {
255 /* there is a waiting or active writer */
256 LOCK(&rwl->lock);
257 if (rwl->write_requests != rwl->write_completions) {
258 rwl->readers_waiting++;
259 WAIT(&rwl->readable, &rwl->lock);
260 rwl->readers_waiting--;
262 UNLOCK(&rwl->lock);
265 cntflag = isc_atomic_xadd(&rwl->cnt_and_flag, READER_INCR);
266 while (1) {
267 if ((rwl->cnt_and_flag & WRITER_ACTIVE) == 0)
268 break;
270 /* A writer is still working */
271 LOCK(&rwl->lock);
272 rwl->readers_waiting++;
273 if ((rwl->cnt_and_flag & WRITER_ACTIVE) != 0)
274 WAIT(&rwl->readable, &rwl->lock);
275 rwl->readers_waiting--;
276 UNLOCK(&rwl->lock);
279 * Typically, the reader should be able to get a lock
280 * at this stage:
281 * (1) there should have been no pending writer when
282 * the reader was trying to increment the
283 * counter; otherwise, the writer should be in
284 * the waiting queue, preventing the reader from
285 * proceeding to this point.
286 * (2) once the reader increments the counter, no
287 * more writer can get a lock.
288 * Still, it is possible another writer can work at
289 * this point, e.g. in the following scenario:
290 * A previous writer unlocks the writer lock.
291 * This reader proceeds to point (1).
292 * A new writer appears, and gets a new lock before
293 * the reader increments the counter.
294 * The reader then increments the counter.
295 * The previous writer notices there is a waiting
296 * reader who is almost ready, and wakes it up.
297 * So, the reader needs to confirm whether it can now
298 * read explicitly (thus we loop). Note that this is
299 * not an infinite process, since the reader has
300 * incremented the counter at this point.
305 * If we are temporarily preferred to writers due to the writer
306 * quota, reset the condition (race among readers doesn't
307 * matter).
309 rwl->write_granted = 0;
310 } else {
311 isc_int32_t prev_writer;
313 /* enter the waiting queue, and wait for our turn */
314 prev_writer = isc_atomic_xadd(&rwl->write_requests, 1);
315 while (rwl->write_completions != prev_writer) {
316 LOCK(&rwl->lock);
317 if (rwl->write_completions != prev_writer) {
318 WAIT(&rwl->writeable, &rwl->lock);
319 UNLOCK(&rwl->lock);
320 continue;
322 UNLOCK(&rwl->lock);
323 break;
326 while (1) {
327 cntflag = isc_atomic_cmpxchg(&rwl->cnt_and_flag, 0,
328 WRITER_ACTIVE);
329 if (cntflag == 0)
330 break;
332 /* Another active reader or writer is working. */
333 LOCK(&rwl->lock);
334 if (rwl->cnt_and_flag != 0)
335 WAIT(&rwl->writeable, &rwl->lock);
336 UNLOCK(&rwl->lock);
339 INSIST((rwl->cnt_and_flag & WRITER_ACTIVE) != 0);
340 rwl->write_granted++;
343 #ifdef ISC_RWLOCK_TRACE
344 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
345 ISC_MSG_POSTLOCK, "postlock"), rwl, type);
346 #endif
348 return (ISC_R_SUCCESS);
351 isc_result_t
352 isc_rwlock_trylock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
353 isc_int32_t cntflag;
355 REQUIRE(VALID_RWLOCK(rwl));
357 #ifdef ISC_RWLOCK_TRACE
358 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
359 ISC_MSG_PRELOCK, "prelock"), rwl, type);
360 #endif
362 if (type == isc_rwlocktype_read) {
363 /* If a writer is waiting or working, we fail. */
364 if (rwl->write_requests != rwl->write_completions)
365 return (ISC_R_LOCKBUSY);
367 /* Otherwise, be ready for reading. */
368 cntflag = isc_atomic_xadd(&rwl->cnt_and_flag, READER_INCR);
369 if ((cntflag & WRITER_ACTIVE) != 0) {
371 * A writer is working. We lose, and cancel the read
372 * request.
374 cntflag = isc_atomic_xadd(&rwl->cnt_and_flag,
375 -READER_INCR);
377 * If no other readers are waiting and we've suspended
378 * new writers in this short period, wake them up.
380 if (cntflag == READER_INCR &&
381 rwl->write_completions != rwl->write_requests) {
382 LOCK(&rwl->lock);
383 BROADCAST(&rwl->writeable);
384 UNLOCK(&rwl->lock);
387 return (ISC_R_LOCKBUSY);
389 } else {
390 /* Try locking without entering the waiting queue. */
391 cntflag = isc_atomic_cmpxchg(&rwl->cnt_and_flag, 0,
392 WRITER_ACTIVE);
393 if (cntflag != 0)
394 return (ISC_R_LOCKBUSY);
397 * XXXJT: jump into the queue, possibly breaking the writer
398 * order.
400 (void)isc_atomic_xadd(&rwl->write_completions, -1);
402 rwl->write_granted++;
405 #ifdef ISC_RWLOCK_TRACE
406 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
407 ISC_MSG_POSTLOCK, "postlock"), rwl, type);
408 #endif
410 return (ISC_R_SUCCESS);
413 isc_result_t
414 isc_rwlock_tryupgrade(isc_rwlock_t *rwl) {
415 isc_int32_t prevcnt;
417 REQUIRE(VALID_RWLOCK(rwl));
419 /* Try to acquire write access. */
420 prevcnt = isc_atomic_cmpxchg(&rwl->cnt_and_flag,
421 READER_INCR, WRITER_ACTIVE);
423 * There must have been no writer, and there must have been at least
424 * one reader.
426 INSIST((prevcnt & WRITER_ACTIVE) == 0 &&
427 (prevcnt & ~WRITER_ACTIVE) != 0);
429 if (prevcnt == READER_INCR) {
431 * We are the only reader and have been upgraded.
432 * Now jump into the head of the writer waiting queue.
434 (void)isc_atomic_xadd(&rwl->write_completions, -1);
435 } else
436 return (ISC_R_LOCKBUSY);
438 return (ISC_R_SUCCESS);
442 void
443 isc_rwlock_downgrade(isc_rwlock_t *rwl) {
444 isc_int32_t prev_readers;
446 REQUIRE(VALID_RWLOCK(rwl));
448 /* Become an active reader. */
449 prev_readers = isc_atomic_xadd(&rwl->cnt_and_flag, READER_INCR);
450 /* We must have been a writer. */
451 INSIST((prev_readers & WRITER_ACTIVE) != 0);
453 /* Complete write */
454 (void)isc_atomic_xadd(&rwl->cnt_and_flag, -WRITER_ACTIVE);
455 (void)isc_atomic_xadd(&rwl->write_completions, 1);
457 /* Resume other readers */
458 LOCK(&rwl->lock);
459 if (rwl->readers_waiting > 0)
460 BROADCAST(&rwl->readable);
461 UNLOCK(&rwl->lock);
464 isc_result_t
465 isc_rwlock_unlock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
466 isc_int32_t prev_cnt;
468 REQUIRE(VALID_RWLOCK(rwl));
470 #ifdef ISC_RWLOCK_TRACE
471 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
472 ISC_MSG_PREUNLOCK, "preunlock"), rwl, type);
473 #endif
475 if (type == isc_rwlocktype_read) {
476 prev_cnt = isc_atomic_xadd(&rwl->cnt_and_flag, -READER_INCR);
479 * If we're the last reader and any writers are waiting, wake
480 * them up. We need to wake up all of them to ensure the
481 * FIFO order.
483 if (prev_cnt == READER_INCR &&
484 rwl->write_completions != rwl->write_requests) {
485 LOCK(&rwl->lock);
486 BROADCAST(&rwl->writeable);
487 UNLOCK(&rwl->lock);
489 } else {
490 isc_boolean_t wakeup_writers = ISC_TRUE;
493 * Reset the flag, and (implicitly) tell other writers
494 * we are done.
496 (void)isc_atomic_xadd(&rwl->cnt_and_flag, -WRITER_ACTIVE);
497 (void)isc_atomic_xadd(&rwl->write_completions, 1);
499 if (rwl->write_granted >= rwl->write_quota ||
500 rwl->write_requests == rwl->write_completions ||
501 (rwl->cnt_and_flag & ~WRITER_ACTIVE) != 0) {
503 * We have passed the write quota, no writer is
504 * waiting, or some readers are almost ready, pending
505 * possible writers. Note that the last case can
506 * happen even if write_requests != write_completions
507 * (which means a new writer in the queue), so we need
508 * to catch the case explicitly.
510 LOCK(&rwl->lock);
511 if (rwl->readers_waiting > 0) {
512 wakeup_writers = ISC_FALSE;
513 BROADCAST(&rwl->readable);
515 UNLOCK(&rwl->lock);
518 if (rwl->write_requests != rwl->write_completions &&
519 wakeup_writers) {
520 LOCK(&rwl->lock);
521 BROADCAST(&rwl->writeable);
522 UNLOCK(&rwl->lock);
526 #ifdef ISC_RWLOCK_TRACE
527 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
528 ISC_MSG_POSTUNLOCK, "postunlock"),
529 rwl, type);
530 #endif
532 return (ISC_R_SUCCESS);
535 #else /* ISC_PLATFORM_HAVEXADD && ISC_PLATFORM_HAVECMPXCHG */
537 static isc_result_t
538 doit(isc_rwlock_t *rwl, isc_rwlocktype_t type, isc_boolean_t nonblock) {
539 isc_boolean_t skip = ISC_FALSE;
540 isc_boolean_t done = ISC_FALSE;
541 isc_result_t result = ISC_R_SUCCESS;
543 REQUIRE(VALID_RWLOCK(rwl));
545 LOCK(&rwl->lock);
547 #ifdef ISC_RWLOCK_TRACE
548 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
549 ISC_MSG_PRELOCK, "prelock"), rwl, type);
550 #endif
552 if (type == isc_rwlocktype_read) {
553 if (rwl->readers_waiting != 0)
554 skip = ISC_TRUE;
555 while (!done) {
556 if (!skip &&
557 ((rwl->active == 0 ||
558 (rwl->type == isc_rwlocktype_read &&
559 (rwl->writers_waiting == 0 ||
560 rwl->granted < rwl->read_quota)))))
562 rwl->type = isc_rwlocktype_read;
563 rwl->active++;
564 rwl->granted++;
565 done = ISC_TRUE;
566 } else if (nonblock) {
567 result = ISC_R_LOCKBUSY;
568 done = ISC_TRUE;
569 } else {
570 skip = ISC_FALSE;
571 rwl->readers_waiting++;
572 WAIT(&rwl->readable, &rwl->lock);
573 rwl->readers_waiting--;
576 } else {
577 if (rwl->writers_waiting != 0)
578 skip = ISC_TRUE;
579 while (!done) {
580 if (!skip && rwl->active == 0) {
581 rwl->type = isc_rwlocktype_write;
582 rwl->active = 1;
583 rwl->granted++;
584 done = ISC_TRUE;
585 } else if (nonblock) {
586 result = ISC_R_LOCKBUSY;
587 done = ISC_TRUE;
588 } else {
589 skip = ISC_FALSE;
590 rwl->writers_waiting++;
591 WAIT(&rwl->writeable, &rwl->lock);
592 rwl->writers_waiting--;
597 #ifdef ISC_RWLOCK_TRACE
598 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
599 ISC_MSG_POSTLOCK, "postlock"), rwl, type);
600 #endif
602 UNLOCK(&rwl->lock);
604 return (result);
607 isc_result_t
608 isc_rwlock_lock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
609 return (doit(rwl, type, ISC_FALSE));
612 isc_result_t
613 isc_rwlock_trylock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
614 return (doit(rwl, type, ISC_TRUE));
617 isc_result_t
618 isc_rwlock_tryupgrade(isc_rwlock_t *rwl) {
619 isc_result_t result = ISC_R_SUCCESS;
621 REQUIRE(VALID_RWLOCK(rwl));
622 LOCK(&rwl->lock);
623 REQUIRE(rwl->type == isc_rwlocktype_read);
624 REQUIRE(rwl->active != 0);
626 /* If we are the only reader then succeed. */
627 if (rwl->active == 1) {
628 rwl->original = (rwl->original == isc_rwlocktype_none) ?
629 isc_rwlocktype_read : isc_rwlocktype_none;
630 rwl->type = isc_rwlocktype_write;
631 } else
632 result = ISC_R_LOCKBUSY;
634 UNLOCK(&rwl->lock);
635 return (result);
638 void
639 isc_rwlock_downgrade(isc_rwlock_t *rwl) {
641 REQUIRE(VALID_RWLOCK(rwl));
642 LOCK(&rwl->lock);
643 REQUIRE(rwl->type == isc_rwlocktype_write);
644 REQUIRE(rwl->active == 1);
646 rwl->type = isc_rwlocktype_read;
647 rwl->original = (rwl->original == isc_rwlocktype_none) ?
648 isc_rwlocktype_write : isc_rwlocktype_none;
650 * Resume processing any read request that were blocked when
651 * we upgraded.
653 if (rwl->original == isc_rwlocktype_none &&
654 (rwl->writers_waiting == 0 || rwl->granted < rwl->read_quota) &&
655 rwl->readers_waiting > 0)
656 BROADCAST(&rwl->readable);
658 UNLOCK(&rwl->lock);
661 isc_result_t
662 isc_rwlock_unlock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
664 REQUIRE(VALID_RWLOCK(rwl));
665 LOCK(&rwl->lock);
666 REQUIRE(rwl->type == type);
668 UNUSED(type);
670 #ifdef ISC_RWLOCK_TRACE
671 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
672 ISC_MSG_PREUNLOCK, "preunlock"), rwl, type);
673 #endif
675 INSIST(rwl->active > 0);
676 rwl->active--;
677 if (rwl->active == 0) {
678 if (rwl->original != isc_rwlocktype_none) {
679 rwl->type = rwl->original;
680 rwl->original = isc_rwlocktype_none;
682 if (rwl->type == isc_rwlocktype_read) {
683 rwl->granted = 0;
684 if (rwl->writers_waiting > 0) {
685 rwl->type = isc_rwlocktype_write;
686 SIGNAL(&rwl->writeable);
687 } else if (rwl->readers_waiting > 0) {
688 /* Does this case ever happen? */
689 BROADCAST(&rwl->readable);
691 } else {
692 if (rwl->readers_waiting > 0) {
693 if (rwl->writers_waiting > 0 &&
694 rwl->granted < rwl->write_quota) {
695 SIGNAL(&rwl->writeable);
696 } else {
697 rwl->granted = 0;
698 rwl->type = isc_rwlocktype_read;
699 BROADCAST(&rwl->readable);
701 } else if (rwl->writers_waiting > 0) {
702 rwl->granted = 0;
703 SIGNAL(&rwl->writeable);
704 } else {
705 rwl->granted = 0;
709 INSIST(rwl->original == isc_rwlocktype_none);
711 #ifdef ISC_RWLOCK_TRACE
712 print_lock(isc_msgcat_get(isc_msgcat, ISC_MSGSET_RWLOCK,
713 ISC_MSG_POSTUNLOCK, "postunlock"),
714 rwl, type);
715 #endif
717 UNLOCK(&rwl->lock);
719 return (ISC_R_SUCCESS);
722 #endif /* ISC_PLATFORM_HAVEXADD && ISC_PLATFORM_HAVECMPXCHG */
723 #else /* ISC_PLATFORM_USETHREADS */
725 isc_result_t
726 isc_rwlock_init(isc_rwlock_t *rwl, unsigned int read_quota,
727 unsigned int write_quota)
729 REQUIRE(rwl != NULL);
731 UNUSED(read_quota);
732 UNUSED(write_quota);
734 rwl->type = isc_rwlocktype_read;
735 rwl->active = 0;
736 rwl->magic = RWLOCK_MAGIC;
738 return (ISC_R_SUCCESS);
741 isc_result_t
742 isc_rwlock_lock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
743 REQUIRE(VALID_RWLOCK(rwl));
745 if (type == isc_rwlocktype_read) {
746 if (rwl->type != isc_rwlocktype_read && rwl->active != 0)
747 return (ISC_R_LOCKBUSY);
748 rwl->type = isc_rwlocktype_read;
749 rwl->active++;
750 } else {
751 if (rwl->active != 0)
752 return (ISC_R_LOCKBUSY);
753 rwl->type = isc_rwlocktype_write;
754 rwl->active = 1;
756 return (ISC_R_SUCCESS);
759 isc_result_t
760 isc_rwlock_trylock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
761 return (isc_rwlock_lock(rwl, type));
764 isc_result_t
765 isc_rwlock_tryupgrade(isc_rwlock_t *rwl) {
766 isc_result_t result = ISC_R_SUCCESS;
768 REQUIRE(VALID_RWLOCK(rwl));
769 REQUIRE(rwl->type == isc_rwlocktype_read);
770 REQUIRE(rwl->active != 0);
772 /* If we are the only reader then succeed. */
773 if (rwl->active == 1)
774 rwl->type = isc_rwlocktype_write;
775 else
776 result = ISC_R_LOCKBUSY;
777 return (result);
780 void
781 isc_rwlock_downgrade(isc_rwlock_t *rwl) {
783 REQUIRE(VALID_RWLOCK(rwl));
784 REQUIRE(rwl->type == isc_rwlocktype_write);
785 REQUIRE(rwl->active == 1);
787 rwl->type = isc_rwlocktype_read;
790 isc_result_t
791 isc_rwlock_unlock(isc_rwlock_t *rwl, isc_rwlocktype_t type) {
792 REQUIRE(VALID_RWLOCK(rwl));
793 REQUIRE(rwl->type == type);
795 UNUSED(type);
797 INSIST(rwl->active > 0);
798 rwl->active--;
800 return (ISC_R_SUCCESS);
803 void
804 isc_rwlock_destroy(isc_rwlock_t *rwl) {
805 REQUIRE(rwl != NULL);
806 REQUIRE(rwl->active == 0);
807 rwl->magic = 0;
810 #endif /* ISC_PLATFORM_USETHREADS */