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 #include <sys/refcount.h>
27 #include <sys/rrwlock.h>
28
29 /*
30 * This file contains the implementation of a re-entrant read
31 * reader/writer lock (aka "rrwlock").
32 *
33 * This is a normal reader/writer lock with the additional feature
34 * of allowing threads who have already obtained a read lock to
35 * re-enter another read lock (re-entrant read) - even if there are
36 * waiting writers.
37 *
38 * Callers who have not obtained a read lock give waiting writers priority.
39 *
40 * The rrwlock_t lock does not allow re-entrant writers, nor does it
41 * allow a re-entrant mix of reads and writes (that is, it does not
42 * allow a caller who has already obtained a read lock to be able to
43 * then grab a write lock without first dropping all read locks, and
44 * vice versa).
45 *
46 * The rrwlock_t uses tsd (thread specific data) to keep a list of
47 * nodes (rrw_node_t), where each node keeps track of which specific
48 * lock (rrw_node_t::rn_rrl) the thread has grabbed. Since re-entering
49 * should be rare, a thread that grabs multiple reads on the same rrwlock_t
50 * will store multiple rrw_node_ts of the same 'rrn_rrl'. Nodes on the
51 * tsd list can represent a different rrwlock_t. This allows a thread
52 * to enter multiple and unique rrwlock_ts for read locks at the same time.
53 *
54 * Since using tsd exposes some overhead, the rrwlock_t only needs to
55 * keep tsd data when writers are waiting. If no writers are waiting, then
56 * a reader just bumps the anonymous read count (rr_anon_rcount) - no tsd
57 * is needed. Once a writer attempts to grab the lock, readers then
58 * keep tsd data and bump the linked readers count (rr_linked_rcount).
59 *
60 * If there are waiting writers and there are anonymous readers, then a
61 * reader doesn't know if it is a re-entrant lock. But since it may be one,
62 * we allow the read to proceed (otherwise it could deadlock). Since once
63 * waiting writers are active, readers no longer bump the anonymous count,
64 * the anonymous readers will eventually flush themselves out. At this point,
65 * readers will be able to tell if they are a re-entrant lock (have a
66 * rrw_node_t entry for the lock) or not. If they are a re-entrant lock, then
67 * we must let the proceed. If they are not, then the reader blocks for the
68 * waiting writers. Hence, we do not starve writers.
69 */
70
71 /* global key for TSD */
72 uint_t rrw_tsd_key;
73
74 typedef struct rrw_node {
75 struct rrw_node *rn_next;
76 rrwlock_t *rn_rrl;
77 } rrw_node_t;
78
79 static rrw_node_t *
80 rrn_find(rrwlock_t *rrl)
81 {
82 rrw_node_t *rn;
83
84 if (refcount_count(&rrl->rr_linked_rcount) == 0)
85 return (NULL);
86
87 for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
88 if (rn->rn_rrl == rrl)
89 return (rn);
90 }
91 return (NULL);
92 }
93
94 /*
95 * Add a node to the head of the singly linked list.
96 */
97 static void
98 rrn_add(rrwlock_t *rrl)
99 {
100 rrw_node_t *rn;
101
102 rn = kmem_alloc(sizeof (*rn), KM_SLEEP);
103 rn->rn_rrl = rrl;
104 rn->rn_next = tsd_get(rrw_tsd_key);
105 VERIFY(tsd_set(rrw_tsd_key, rn) == 0);
106 }
107
108 /*
109 * If a node is found for 'rrl', then remove the node from this
110 * thread's list and return TRUE; otherwise return FALSE.
111 */
112 static boolean_t
113 rrn_find_and_remove(rrwlock_t *rrl)
114 {
115 rrw_node_t *rn;
116 rrw_node_t *prev = NULL;
117
118 if (refcount_count(&rrl->rr_linked_rcount) == 0)
119 return (B_FALSE);
120
121 for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
122 if (rn->rn_rrl == rrl) {
123 if (prev)
124 prev->rn_next = rn->rn_next;
125 else
126 VERIFY(tsd_set(rrw_tsd_key, rn->rn_next) == 0);
127 kmem_free(rn, sizeof (*rn));
128 return (B_TRUE);
129 }
130 prev = rn;
131 }
132 return (B_FALSE);
133 }
134
135 void
136 rrw_init(rrwlock_t *rrl)
137 {
138 mutex_init(&rrl->rr_lock, NULL, MUTEX_DEFAULT, NULL);
139 cv_init(&rrl->rr_cv, NULL, CV_DEFAULT, NULL);
140 rrl->rr_writer = NULL;
141 refcount_create(&rrl->rr_anon_rcount);
142 refcount_create(&rrl->rr_linked_rcount);
143 rrl->rr_writer_wanted = B_FALSE;
144 }
145
146 void
147 rrw_destroy(rrwlock_t *rrl)
148 {
149 mutex_destroy(&rrl->rr_lock);
150 cv_destroy(&rrl->rr_cv);
151 ASSERT(rrl->rr_writer == NULL);
152 refcount_destroy(&rrl->rr_anon_rcount);
153 refcount_destroy(&rrl->rr_linked_rcount);
154 }
155
156 static void
157 rrw_enter_read(rrwlock_t *rrl, void *tag)
158 {
159 mutex_enter(&rrl->rr_lock);
160 #if !defined(DEBUG) && defined(_KERNEL)
161 if (!rrl->rr_writer && !rrl->rr_writer_wanted) {
162 rrl->rr_anon_rcount.rc_count++;
163 mutex_exit(&rrl->rr_lock);
164 return;
165 }
166 DTRACE_PROBE(zfs__rrwfastpath__rdmiss);
167 #endif
168 ASSERT(rrl->rr_writer != curthread);
169 ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);
170
171 while (rrl->rr_writer || (rrl->rr_writer_wanted &&
172 refcount_is_zero(&rrl->rr_anon_rcount) &&
173 rrn_find(rrl) == NULL))
174 cv_wait(&rrl->rr_cv, &rrl->rr_lock);
175
176 if (rrl->rr_writer_wanted) {
177 /* may or may not be a re-entrant enter */
178 rrn_add(rrl);
179 (void) refcount_add(&rrl->rr_linked_rcount, tag);
180 } else {
181 (void) refcount_add(&rrl->rr_anon_rcount, tag);
182 }
183 ASSERT(rrl->rr_writer == NULL);
184 mutex_exit(&rrl->rr_lock);
185 }
186
187 static void
188 rrw_enter_write(rrwlock_t *rrl)
189 {
190 mutex_enter(&rrl->rr_lock);
191 ASSERT(rrl->rr_writer != curthread);
192
193 while (refcount_count(&rrl->rr_anon_rcount) > 0 ||
194 refcount_count(&rrl->rr_linked_rcount) > 0 ||
195 rrl->rr_writer != NULL) {
196 rrl->rr_writer_wanted = B_TRUE;
197 cv_wait(&rrl->rr_cv, &rrl->rr_lock);
198 }
199 rrl->rr_writer_wanted = B_FALSE;
200 rrl->rr_writer = curthread;
201 mutex_exit(&rrl->rr_lock);
202 }
203
204 void
205 rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag)
206 {
207 if (rw == RW_READER)
208 rrw_enter_read(rrl, tag);
209 else
210 rrw_enter_write(rrl);
211 }
212
213 void
214 rrw_exit(rrwlock_t *rrl, void *tag)
215 {
216 mutex_enter(&rrl->rr_lock);
217 #if !defined(DEBUG) && defined(_KERNEL)
218 if (!rrl->rr_writer && rrl->rr_linked_rcount.rc_count == 0) {
219 rrl->rr_anon_rcount.rc_count--;
220 if (rrl->rr_anon_rcount.rc_count == 0)
221 cv_broadcast(&rrl->rr_cv);
222 mutex_exit(&rrl->rr_lock);
223 return;
224 }
225 DTRACE_PROBE(zfs__rrwfastpath__exitmiss);
226 #endif
227 ASSERT(!refcount_is_zero(&rrl->rr_anon_rcount) ||
228 !refcount_is_zero(&rrl->rr_linked_rcount) ||
229 rrl->rr_writer != NULL);
230
231 if (rrl->rr_writer == NULL) {
232 int64_t count;
233 if (rrn_find_and_remove(rrl))
234 count = refcount_remove(&rrl->rr_linked_rcount, tag);
235 else
236 count = refcount_remove(&rrl->rr_anon_rcount, tag);
237 if (count == 0)
238 cv_broadcast(&rrl->rr_cv);
239 } else {
240 ASSERT(rrl->rr_writer == curthread);
241 ASSERT(refcount_is_zero(&rrl->rr_anon_rcount) &&
242 refcount_is_zero(&rrl->rr_linked_rcount));
243 rrl->rr_writer = NULL;
244 cv_broadcast(&rrl->rr_cv);
245 }
246 mutex_exit(&rrl->rr_lock);
247 }
248
249 boolean_t
250 rrw_held(rrwlock_t *rrl, krw_t rw)
251 {
252 boolean_t held;
253
254 mutex_enter(&rrl->rr_lock);
255 if (rw == RW_WRITER) {
256 held = (rrl->rr_writer == curthread);
257 } else {
258 held = (!refcount_is_zero(&rrl->rr_anon_rcount) ||
259 !refcount_is_zero(&rrl->rr_linked_rcount));
260 }
261 mutex_exit(&rrl->rr_lock);
262
263 return (held);
264 }