xref: /illumos-gate/usr/src/uts/common/fs/zfs/rrwlock.c (revision 843c398e8904ed9d833d2af3103894f909fb4b52)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 2012 by Delphix. All rights reserved.
27  */
28 
29 #include <sys/refcount.h>
30 #include <sys/rrwlock.h>
31 
32 /*
33  * This file contains the implementation of a re-entrant read
34  * reader/writer lock (aka "rrwlock").
35  *
36  * This is a normal reader/writer lock with the additional feature
37  * of allowing threads who have already obtained a read lock to
38  * re-enter another read lock (re-entrant read) - even if there are
39  * waiting writers.
40  *
41  * Callers who have not obtained a read lock give waiting writers priority.
42  *
43  * The rrwlock_t lock does not allow re-entrant writers, nor does it
44  * allow a re-entrant mix of reads and writes (that is, it does not
45  * allow a caller who has already obtained a read lock to be able to
46  * then grab a write lock without first dropping all read locks, and
47  * vice versa).
48  *
49  * The rrwlock_t uses tsd (thread specific data) to keep a list of
50  * nodes (rrw_node_t), where each node keeps track of which specific
51  * lock (rrw_node_t::rn_rrl) the thread has grabbed.  Since re-entering
52  * should be rare, a thread that grabs multiple reads on the same rrwlock_t
53  * will store multiple rrw_node_ts of the same 'rrn_rrl'. Nodes on the
54  * tsd list can represent a different rrwlock_t.  This allows a thread
55  * to enter multiple and unique rrwlock_ts for read locks at the same time.
56  *
57  * Since using tsd exposes some overhead, the rrwlock_t only needs to
58  * keep tsd data when writers are waiting.  If no writers are waiting, then
59  * a reader just bumps the anonymous read count (rr_anon_rcount) - no tsd
60  * is needed.  Once a writer attempts to grab the lock, readers then
61  * keep tsd data and bump the linked readers count (rr_linked_rcount).
62  *
63  * If there are waiting writers and there are anonymous readers, then a
64  * reader doesn't know if it is a re-entrant lock. But since it may be one,
65  * we allow the read to proceed (otherwise it could deadlock).  Since once
66  * waiting writers are active, readers no longer bump the anonymous count,
67  * the anonymous readers will eventually flush themselves out.  At this point,
68  * readers will be able to tell if they are a re-entrant lock (have a
69  * rrw_node_t entry for the lock) or not. If they are a re-entrant lock, then
70  * we must let the proceed.  If they are not, then the reader blocks for the
71  * waiting writers.  Hence, we do not starve writers.
72  */
73 
74 /* global key for TSD */
75 uint_t rrw_tsd_key;
76 
77 typedef struct rrw_node {
78 	struct rrw_node *rn_next;
79 	rrwlock_t *rn_rrl;
80 	void *rn_tag;
81 } rrw_node_t;
82 
83 static rrw_node_t *
84 rrn_find(rrwlock_t *rrl)
85 {
86 	rrw_node_t *rn;
87 
88 	if (refcount_count(&rrl->rr_linked_rcount) == 0)
89 		return (NULL);
90 
91 	for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
92 		if (rn->rn_rrl == rrl)
93 			return (rn);
94 	}
95 	return (NULL);
96 }
97 
98 /*
99  * Add a node to the head of the singly linked list.
100  */
101 static void
102 rrn_add(rrwlock_t *rrl, void *tag)
103 {
104 	rrw_node_t *rn;
105 
106 	rn = kmem_alloc(sizeof (*rn), KM_SLEEP);
107 	rn->rn_rrl = rrl;
108 	rn->rn_next = tsd_get(rrw_tsd_key);
109 	rn->rn_tag = tag;
110 	VERIFY(tsd_set(rrw_tsd_key, rn) == 0);
111 }
112 
113 /*
114  * If a node is found for 'rrl', then remove the node from this
115  * thread's list and return TRUE; otherwise return FALSE.
116  */
117 static boolean_t
118 rrn_find_and_remove(rrwlock_t *rrl, void *tag)
119 {
120 	rrw_node_t *rn;
121 	rrw_node_t *prev = NULL;
122 
123 	if (refcount_count(&rrl->rr_linked_rcount) == 0)
124 		return (B_FALSE);
125 
126 	for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
127 		if (rn->rn_rrl == rrl && rn->rn_tag == tag) {
128 			if (prev)
129 				prev->rn_next = rn->rn_next;
130 			else
131 				VERIFY(tsd_set(rrw_tsd_key, rn->rn_next) == 0);
132 			kmem_free(rn, sizeof (*rn));
133 			return (B_TRUE);
134 		}
135 		prev = rn;
136 	}
137 	return (B_FALSE);
138 }
139 
140 void
141 rrw_init(rrwlock_t *rrl, boolean_t track_all)
142 {
143 	mutex_init(&rrl->rr_lock, NULL, MUTEX_DEFAULT, NULL);
144 	cv_init(&rrl->rr_cv, NULL, CV_DEFAULT, NULL);
145 	rrl->rr_writer = NULL;
146 	refcount_create(&rrl->rr_anon_rcount);
147 	refcount_create(&rrl->rr_linked_rcount);
148 	rrl->rr_writer_wanted = B_FALSE;
149 	rrl->rr_track_all = track_all;
150 }
151 
152 void
153 rrw_destroy(rrwlock_t *rrl)
154 {
155 	mutex_destroy(&rrl->rr_lock);
156 	cv_destroy(&rrl->rr_cv);
157 	ASSERT(rrl->rr_writer == NULL);
158 	refcount_destroy(&rrl->rr_anon_rcount);
159 	refcount_destroy(&rrl->rr_linked_rcount);
160 }
161 
162 static void
163 rrw_enter_read_impl(rrwlock_t *rrl, boolean_t prio, void *tag)
164 {
165 	mutex_enter(&rrl->rr_lock);
166 #if !defined(DEBUG) && defined(_KERNEL)
167 	if (rrl->rr_writer == NULL && !rrl->rr_writer_wanted &&
168 	    !rrl->rr_track_all) {
169 		rrl->rr_anon_rcount.rc_count++;
170 		mutex_exit(&rrl->rr_lock);
171 		return;
172 	}
173 	DTRACE_PROBE(zfs__rrwfastpath__rdmiss);
174 #endif
175 	ASSERT(rrl->rr_writer != curthread);
176 	ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);
177 
178 	while (rrl->rr_writer != NULL || (rrl->rr_writer_wanted &&
179 	    refcount_is_zero(&rrl->rr_anon_rcount) && !prio &&
180 	    rrn_find(rrl) == NULL))
181 		cv_wait(&rrl->rr_cv, &rrl->rr_lock);
182 
183 	if (rrl->rr_writer_wanted || rrl->rr_track_all) {
184 		/* may or may not be a re-entrant enter */
185 		rrn_add(rrl, tag);
186 		(void) refcount_add(&rrl->rr_linked_rcount, tag);
187 	} else {
188 		(void) refcount_add(&rrl->rr_anon_rcount, tag);
189 	}
190 	ASSERT(rrl->rr_writer == NULL);
191 	mutex_exit(&rrl->rr_lock);
192 }
193 
194 void
195 rrw_enter_read(rrwlock_t *rrl, void *tag)
196 {
197 	rrw_enter_read_impl(rrl, B_FALSE, tag);
198 }
199 
200 /*
201  * take a read lock even if there are pending write lock requests. if we want
202  * to take a lock reentrantly, but from different threads (that have a
203  * relationship to each other), the normal detection mechanism to overrule
204  * the pending writer does not work, so we have to give an explicit hint here.
205  */
206 void
207 rrw_enter_read_prio(rrwlock_t *rrl, void *tag)
208 {
209 	rrw_enter_read_impl(rrl, B_TRUE, tag);
210 }
211 
212 
213 void
214 rrw_enter_write(rrwlock_t *rrl)
215 {
216 	mutex_enter(&rrl->rr_lock);
217 	ASSERT(rrl->rr_writer != curthread);
218 
219 	while (refcount_count(&rrl->rr_anon_rcount) > 0 ||
220 	    refcount_count(&rrl->rr_linked_rcount) > 0 ||
221 	    rrl->rr_writer != NULL) {
222 		rrl->rr_writer_wanted = B_TRUE;
223 		cv_wait(&rrl->rr_cv, &rrl->rr_lock);
224 	}
225 	rrl->rr_writer_wanted = B_FALSE;
226 	rrl->rr_writer = curthread;
227 	mutex_exit(&rrl->rr_lock);
228 }
229 
230 void
231 rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag)
232 {
233 	if (rw == RW_READER)
234 		rrw_enter_read(rrl, tag);
235 	else
236 		rrw_enter_write(rrl);
237 }
238 
239 void
240 rrw_exit(rrwlock_t *rrl, void *tag)
241 {
242 	mutex_enter(&rrl->rr_lock);
243 #if !defined(DEBUG) && defined(_KERNEL)
244 	if (!rrl->rr_writer && rrl->rr_linked_rcount.rc_count == 0) {
245 		rrl->rr_anon_rcount.rc_count--;
246 		if (rrl->rr_anon_rcount.rc_count == 0)
247 			cv_broadcast(&rrl->rr_cv);
248 		mutex_exit(&rrl->rr_lock);
249 		return;
250 	}
251 	DTRACE_PROBE(zfs__rrwfastpath__exitmiss);
252 #endif
253 	ASSERT(!refcount_is_zero(&rrl->rr_anon_rcount) ||
254 	    !refcount_is_zero(&rrl->rr_linked_rcount) ||
255 	    rrl->rr_writer != NULL);
256 
257 	if (rrl->rr_writer == NULL) {
258 		int64_t count;
259 		if (rrn_find_and_remove(rrl, tag)) {
260 			count = refcount_remove(&rrl->rr_linked_rcount, tag);
261 		} else {
262 			ASSERT(!rrl->rr_track_all);
263 			count = refcount_remove(&rrl->rr_anon_rcount, tag);
264 		}
265 		if (count == 0)
266 			cv_broadcast(&rrl->rr_cv);
267 	} else {
268 		ASSERT(rrl->rr_writer == curthread);
269 		ASSERT(refcount_is_zero(&rrl->rr_anon_rcount) &&
270 		    refcount_is_zero(&rrl->rr_linked_rcount));
271 		rrl->rr_writer = NULL;
272 		cv_broadcast(&rrl->rr_cv);
273 	}
274 	mutex_exit(&rrl->rr_lock);
275 }
276 
277 /*
278  * If the lock was created with track_all, rrw_held(RW_READER) will return
279  * B_TRUE iff the current thread has the lock for reader.  Otherwise it may
280  * return B_TRUE if any thread has the lock for reader.
281  */
282 boolean_t
283 rrw_held(rrwlock_t *rrl, krw_t rw)
284 {
285 	boolean_t held;
286 
287 	mutex_enter(&rrl->rr_lock);
288 	if (rw == RW_WRITER) {
289 		held = (rrl->rr_writer == curthread);
290 	} else {
291 		held = (!refcount_is_zero(&rrl->rr_anon_rcount) ||
292 		    rrn_find(rrl) != NULL);
293 	}
294 	mutex_exit(&rrl->rr_lock);
295 
296 	return (held);
297 }
298 
299 void
300 rrw_tsd_destroy(void *arg)
301 {
302 	rrw_node_t *rn = arg;
303 	if (rn != NULL) {
304 		panic("thread %p terminating with rrw lock %p held",
305 		    (void *)curthread, (void *)rn->rn_rrl);
306 	}
307 }
308 
309 /*
310  * A reader-mostly lock implementation, tuning above reader-writer locks
311  * for hightly parallel read acquisitions, while pessimizing writes.
312  *
313  * The idea is to split single busy lock into array of locks, so that
314  * each reader can lock only one of them for read, depending on result
315  * of simple hash function.  That proportionally reduces lock congestion.
316  * Writer same time has to sequentially aquire write on all the locks.
317  * That makes write aquisition proportionally slower, but in places where
318  * it is used (filesystem unmount) performance is not critical.
319  *
320  * All the functions below are direct wrappers around functions above.
321  */
322 void
323 rrm_init(rrmlock_t *rrl, boolean_t track_all)
324 {
325 	int i;
326 
327 	for (i = 0; i < RRM_NUM_LOCKS; i++)
328 		rrw_init(&rrl->locks[i], track_all);
329 }
330 
331 void
332 rrm_destroy(rrmlock_t *rrl)
333 {
334 	int i;
335 
336 	for (i = 0; i < RRM_NUM_LOCKS; i++)
337 		rrw_destroy(&rrl->locks[i]);
338 }
339 
340 void
341 rrm_enter(rrmlock_t *rrl, krw_t rw, void *tag)
342 {
343 	if (rw == RW_READER)
344 		rrm_enter_read(rrl, tag);
345 	else
346 		rrm_enter_write(rrl);
347 }
348 
349 /*
350  * This maps the current thread to a specific lock.  Note that the lock
351  * must be released by the same thread that acquired it.  We do this
352  * mapping by taking the thread pointer mod a prime number.  We examine
353  * only the low 32 bits of the thread pointer, because 32-bit division
354  * is faster than 64-bit division, and the high 32 bits have little
355  * entropy anyway.
356  */
357 #define	RRM_TD_LOCK()	(((uint32_t)(uintptr_t)(curthread)) % RRM_NUM_LOCKS)
358 
359 void
360 rrm_enter_read(rrmlock_t *rrl, void *tag)
361 {
362 	rrw_enter_read(&rrl->locks[RRM_TD_LOCK()], tag);
363 }
364 
365 void
366 rrm_enter_write(rrmlock_t *rrl)
367 {
368 	int i;
369 
370 	for (i = 0; i < RRM_NUM_LOCKS; i++)
371 		rrw_enter_write(&rrl->locks[i]);
372 }
373 
374 void
375 rrm_exit(rrmlock_t *rrl, void *tag)
376 {
377 	int i;
378 
379 	if (rrl->locks[0].rr_writer == curthread) {
380 		for (i = 0; i < RRM_NUM_LOCKS; i++)
381 			rrw_exit(&rrl->locks[i], tag);
382 	} else {
383 		rrw_exit(&rrl->locks[RRM_TD_LOCK()], tag);
384 	}
385 }
386 
387 boolean_t
388 rrm_held(rrmlock_t *rrl, krw_t rw)
389 {
390 	if (rw == RW_WRITER) {
391 		return (rrw_held(&rrl->locks[0], rw));
392 	} else {
393 		return (rrw_held(&rrl->locks[RRM_TD_LOCK()], rw));
394 	}
395 }
396