summaryrefslogtreecommitdiffstats
path: root/module/zfs/rrwlock.c
blob: 704f76067bf09b6d48219a260dc260c0bc27699a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */
/*
 * Copyright (c) 2012 by Delphix. All rights reserved.
 */

#include <sys/refcount.h>
#include <sys/rrwlock.h>

/*
 * This file contains the implementation of a re-entrant read
 * reader/writer lock (aka "rrwlock").
 *
 * This is a normal reader/writer lock with the additional feature
 * of allowing threads who have already obtained a read lock to
 * re-enter another read lock (re-entrant read) - even if there are
 * waiting writers.
 *
 * Callers who have not obtained a read lock give waiting writers priority.
 *
 * The rrwlock_t lock does not allow re-entrant writers, nor does it
 * allow a re-entrant mix of reads and writes (that is, it does not
 * allow a caller who has already obtained a read lock to be able to
 * then grab a write lock without first dropping all read locks, and
 * vice versa).
 *
 * The rrwlock_t uses tsd (thread specific data) to keep a list of
 * nodes (rrw_node_t), where each node keeps track of which specific
 * lock (rrw_node_t::rn_rrl) the thread has grabbed.  Since re-entering
 * should be rare, a thread that grabs multiple reads on the same rrwlock_t
 * will store multiple rrw_node_ts of the same 'rrn_rrl'. Nodes on the
 * tsd list can represent a different rrwlock_t.  This allows a thread
 * to enter multiple and unique rrwlock_ts for read locks at the same time.
 *
 * Since using tsd exposes some overhead, the rrwlock_t only needs to
 * keep tsd data when writers are waiting.  If no writers are waiting, then
 * a reader just bumps the anonymous read count (rr_anon_rcount) - no tsd
 * is needed.  Once a writer attempts to grab the lock, readers then
 * keep tsd data and bump the linked readers count (rr_linked_rcount).
 *
 * If there are waiting writers and there are anonymous readers, then a
 * reader doesn't know if it is a re-entrant lock. But since it may be one,
 * we allow the read to proceed (otherwise it could deadlock).  Since once
 * waiting writers are active, readers no longer bump the anonymous count,
 * the anonymous readers will eventually flush themselves out.  At this point,
 * readers will be able to tell if they are a re-entrant lock (have a
 * rrw_node_t entry for the lock) or not. If they are a re-entrant lock, then
 * we must let the proceed.  If they are not, then the reader blocks for the
 * waiting writers.  Hence, we do not starve writers.
 */

/* global key for TSD */
uint_t rrw_tsd_key;

typedef struct rrw_node {
	struct rrw_node *rn_next;
	rrwlock_t *rn_rrl;
	void *rn_tag;
} rrw_node_t;

static rrw_node_t *
rrn_find(rrwlock_t *rrl)
{
	rrw_node_t *rn;

	if (refcount_count(&rrl->rr_linked_rcount) == 0)
		return (NULL);

	for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
		if (rn->rn_rrl == rrl)
			return (rn);
	}
	return (NULL);
}

/*
 * Add a node to the head of the singly linked list.
 */
static void
rrn_add(rrwlock_t *rrl, void *tag)
{
	rrw_node_t *rn;

	rn = kmem_alloc(sizeof (*rn), KM_SLEEP);
	rn->rn_rrl = rrl;
	rn->rn_next = tsd_get(rrw_tsd_key);
	rn->rn_tag = tag;
	VERIFY(tsd_set(rrw_tsd_key, rn) == 0);
}

/*
 * If a node is found for 'rrl', then remove the node from this
 * thread's list and return TRUE; otherwise return FALSE.
 */
static boolean_t
rrn_find_and_remove(rrwlock_t *rrl, void *tag)
{
	rrw_node_t *rn;
	rrw_node_t *prev = NULL;

	if (refcount_count(&rrl->rr_linked_rcount) == 0)
		return (B_FALSE);

	for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
		if (rn->rn_rrl == rrl && rn->rn_tag == tag) {
			if (prev)
				prev->rn_next = rn->rn_next;
			else
				VERIFY(tsd_set(rrw_tsd_key, rn->rn_next) == 0);
			kmem_free(rn, sizeof (*rn));
			return (B_TRUE);
		}
		prev = rn;
	}
	return (B_FALSE);
}

void
rrw_init(rrwlock_t *rrl, boolean_t track_all)
{
	mutex_init(&rrl->rr_lock, NULL, MUTEX_DEFAULT, NULL);
	cv_init(&rrl->rr_cv, NULL, CV_DEFAULT, NULL);
	rrl->rr_writer = NULL;
	refcount_create(&rrl->rr_anon_rcount);
	refcount_create(&rrl->rr_linked_rcount);
	rrl->rr_writer_wanted = B_FALSE;
	rrl->rr_track_all = track_all;
}

void
rrw_destroy(rrwlock_t *rrl)
{
	mutex_destroy(&rrl->rr_lock);
	cv_destroy(&rrl->rr_cv);
	ASSERT(rrl->rr_writer == NULL);
	refcount_destroy(&rrl->rr_anon_rcount);
	refcount_destroy(&rrl->rr_linked_rcount);
}

static void
rrw_enter_read_impl(rrwlock_t *rrl, boolean_t prio, void *tag)
{
	mutex_enter(&rrl->rr_lock);
#if !defined(DEBUG) && defined(_KERNEL)
	if (rrl->rr_writer == NULL && !rrl->rr_writer_wanted &&
	    !rrl->rr_track_all) {
		rrl->rr_anon_rcount.rc_count++;
		mutex_exit(&rrl->rr_lock);
		return;
	}
	DTRACE_PROBE(zfs__rrwfastpath__rdmiss);
#endif
	ASSERT(rrl->rr_writer != curthread);
	ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);

	while (rrl->rr_writer != NULL || (rrl->rr_writer_wanted &&
	    refcount_is_zero(&rrl->rr_anon_rcount) && !prio &&
	    rrn_find(rrl) == NULL))
		cv_wait(&rrl->rr_cv, &rrl->rr_lock);

	if (rrl->rr_writer_wanted || rrl->rr_track_all) {
		/* may or may not be a re-entrant enter */
		rrn_add(rrl, tag);
		(void) refcount_add(&rrl->rr_linked_rcount, tag);
	} else {
		(void) refcount_add(&rrl->rr_anon_rcount, tag);
	}
	ASSERT(rrl->rr_writer == NULL);
	mutex_exit(&rrl->rr_lock);
}

void
rrw_enter_read(rrwlock_t *rrl, void *tag)
{
	rrw_enter_read_impl(rrl, B_FALSE, tag);
}

/*
 * take a read lock even if there are pending write lock requests. if we want
 * to take a lock reentrantly, but from different threads (that have a
 * relationship to each other), the normal detection mechanism to overrule
 * the pending writer does not work, so we have to give an explicit hint here.
 */
void
rrw_enter_read_prio(rrwlock_t *rrl, void *tag)
{
	rrw_enter_read_impl(rrl, B_TRUE, tag);
}


void
rrw_enter_write(rrwlock_t *rrl)
{
	mutex_enter(&rrl->rr_lock);
	ASSERT(rrl->rr_writer != curthread);

	while (refcount_count(&rrl->rr_anon_rcount) > 0 ||
	    refcount_count(&rrl->rr_linked_rcount) > 0 ||
	    rrl->rr_writer != NULL) {
		rrl->rr_writer_wanted = B_TRUE;
		cv_wait(&rrl->rr_cv, &rrl->rr_lock);
	}
	rrl->rr_writer_wanted = B_FALSE;
	rrl->rr_writer = curthread;
	mutex_exit(&rrl->rr_lock);
}

void
rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag)
{
	if (rw == RW_READER)
		rrw_enter_read(rrl, tag);
	else
		rrw_enter_write(rrl);
}

void
rrw_exit(rrwlock_t *rrl, void *tag)
{
	mutex_enter(&rrl->rr_lock);
#if !defined(DEBUG) && defined(_KERNEL)
	if (!rrl->rr_writer && rrl->rr_linked_rcount.rc_count == 0) {
		rrl->rr_anon_rcount.rc_count--;
		if (rrl->rr_anon_rcount.rc_count == 0)
			cv_broadcast(&rrl->rr_cv);
		mutex_exit(&rrl->rr_lock);
		return;
	}
	DTRACE_PROBE(zfs__rrwfastpath__exitmiss);
#endif
	ASSERT(!refcount_is_zero(&rrl->rr_anon_rcount) ||
	    !refcount_is_zero(&rrl->rr_linked_rcount) ||
	    rrl->rr_writer != NULL);

	if (rrl->rr_writer == NULL) {
		int64_t count;
		if (rrn_find_and_remove(rrl, tag)) {
			count = refcount_remove(&rrl->rr_linked_rcount, tag);
		} else {
			ASSERT(!rrl->rr_track_all);
			count = refcount_remove(&rrl->rr_anon_rcount, tag);
		}
		if (count == 0)
			cv_broadcast(&rrl->rr_cv);
	} else {
		ASSERT(rrl->rr_writer == curthread);
		ASSERT(refcount_is_zero(&rrl->rr_anon_rcount) &&
		    refcount_is_zero(&rrl->rr_linked_rcount));
		rrl->rr_writer = NULL;
		cv_broadcast(&rrl->rr_cv);
	}
	mutex_exit(&rrl->rr_lock);
}

/*
 * If the lock was created with track_all, rrw_held(RW_READER) will return
 * B_TRUE iff the current thread has the lock for reader.  Otherwise it may
 * return B_TRUE if any thread has the lock for reader.
 */
boolean_t
rrw_held(rrwlock_t *rrl, krw_t rw)
{
	boolean_t held;

	mutex_enter(&rrl->rr_lock);
	if (rw == RW_WRITER) {
		held = (rrl->rr_writer == curthread);
	} else {
		held = (!refcount_is_zero(&rrl->rr_anon_rcount) ||
		    rrn_find(rrl) != NULL);
	}
	mutex_exit(&rrl->rr_lock);

	return (held);
}

void
rrw_tsd_destroy(void *arg)
{
	rrw_node_t *rn = arg;
	if (rn != NULL) {
		panic("thread %p terminating with rrw lock %p held",
		    (void *)curthread, (void *)rn->rn_rrl);
	}
}

/*
 * A reader-mostly lock implementation, tuning above reader-writer locks
 * for hightly parallel read acquisitions, while pessimizing writes.
 *
 * The idea is to split single busy lock into array of locks, so that
 * each reader can lock only one of them for read, depending on result
 * of simple hash function.  That proportionally reduces lock congestion.
 * Writer at the same time has to sequentially acquire write on all the locks.
 * That makes write acquisition proportionally slower, but in places where
 * it is used (filesystem unmount) performance is not critical.
 *
 * All the functions below are direct wrappers around functions above.
 */
void
rrm_init(rrmlock_t *rrl, boolean_t track_all)
{
	int i;

	for (i = 0; i < RRM_NUM_LOCKS; i++)
		rrw_init(&rrl->locks[i], track_all);
}

void
rrm_destroy(rrmlock_t *rrl)
{
	int i;

	for (i = 0; i < RRM_NUM_LOCKS; i++)
		rrw_destroy(&rrl->locks[i]);
}

void
rrm_enter(rrmlock_t *rrl, krw_t rw, void *tag)
{
	if (rw == RW_READER)
		rrm_enter_read(rrl, tag);
	else
		rrm_enter_write(rrl);
}

/*
 * This maps the current thread to a specific lock.  Note that the lock
 * must be released by the same thread that acquired it.  We do this
 * mapping by taking the thread pointer mod a prime number.  We examine
 * only the low 32 bits of the thread pointer, because 32-bit division
 * is faster than 64-bit division, and the high 32 bits have little
 * entropy anyway.
 */
#define	RRM_TD_LOCK()	(((uint32_t)(uintptr_t)(curthread)) % RRM_NUM_LOCKS)

void
rrm_enter_read(rrmlock_t *rrl, void *tag)
{
	rrw_enter_read(&rrl->locks[RRM_TD_LOCK()], tag);
}

void
rrm_enter_write(rrmlock_t *rrl)
{
	int i;

	for (i = 0; i < RRM_NUM_LOCKS; i++)
		rrw_enter_write(&rrl->locks[i]);
}

void
rrm_exit(rrmlock_t *rrl, void *tag)
{
	int i;

	if (rrl->locks[0].rr_writer == curthread) {
		for (i = 0; i < RRM_NUM_LOCKS; i++)
			rrw_exit(&rrl->locks[i], tag);
	} else {
		rrw_exit(&rrl->locks[RRM_TD_LOCK()], tag);
	}
}

boolean_t
rrm_held(rrmlock_t *rrl, krw_t rw)
{
	if (rw == RW_WRITER) {
		return (rrw_held(&rrl->locks[0], rw));
	} else {
		return (rrw_held(&rrl->locks[RRM_TD_LOCK()], rw));
	}
}