1/*
2 * Copyright (C) 2008 Oracle.  All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/pagemap.h>
20#include <linux/spinlock.h>
21#include <linux/page-flags.h>
22#include <asm/bug.h>
23#include "ctree.h"
24#include "extent_io.h"
25#include "locking.h"
26
27static void btrfs_assert_tree_read_locked(struct extent_buffer *eb);
28
29/*
30 * if we currently have a spinning reader or writer lock
31 * (indicated by the rw flag) this will bump the count
32 * of blocking holders and drop the spinlock.
33 */
34void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
35{
36	/*
37	 * no lock is required.  The lock owner may change if
38	 * we have a read lock, but it won't change to or away
39	 * from us.  If we have the write lock, we are the owner
40	 * and it'll never change.
41	 */
42	if (eb->lock_nested && current->pid == eb->lock_owner)
43		return;
44	if (rw == BTRFS_WRITE_LOCK) {
45		if (atomic_read(&eb->blocking_writers) == 0) {
46			WARN_ON(atomic_read(&eb->spinning_writers) != 1);
47			atomic_dec(&eb->spinning_writers);
48			btrfs_assert_tree_locked(eb);
49			atomic_inc(&eb->blocking_writers);
50			write_unlock(&eb->lock);
51		}
52	} else if (rw == BTRFS_READ_LOCK) {
53		btrfs_assert_tree_read_locked(eb);
54		atomic_inc(&eb->blocking_readers);
55		WARN_ON(atomic_read(&eb->spinning_readers) == 0);
56		atomic_dec(&eb->spinning_readers);
57		read_unlock(&eb->lock);
58	}
59	return;
60}
61
62/*
63 * if we currently have a blocking lock, take the spinlock
64 * and drop our blocking count
65 */
66void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
67{
68	/*
69	 * no lock is required.  The lock owner may change if
70	 * we have a read lock, but it won't change to or away
71	 * from us.  If we have the write lock, we are the owner
72	 * and it'll never change.
73	 */
74	if (eb->lock_nested && current->pid == eb->lock_owner)
75		return;
76
77	if (rw == BTRFS_WRITE_LOCK_BLOCKING) {
78		BUG_ON(atomic_read(&eb->blocking_writers) != 1);
79		write_lock(&eb->lock);
80		WARN_ON(atomic_read(&eb->spinning_writers));
81		atomic_inc(&eb->spinning_writers);
82		if (atomic_dec_and_test(&eb->blocking_writers) &&
83		    waitqueue_active(&eb->write_lock_wq))
84			wake_up(&eb->write_lock_wq);
85	} else if (rw == BTRFS_READ_LOCK_BLOCKING) {
86		BUG_ON(atomic_read(&eb->blocking_readers) == 0);
87		read_lock(&eb->lock);
88		atomic_inc(&eb->spinning_readers);
89		if (atomic_dec_and_test(&eb->blocking_readers) &&
90		    waitqueue_active(&eb->read_lock_wq))
91			wake_up(&eb->read_lock_wq);
92	}
93	return;
94}
95
96/*
97 * take a spinning read lock.  This will wait for any blocking
98 * writers
99 */
100void btrfs_tree_read_lock(struct extent_buffer *eb)
101{
102again:
103	BUG_ON(!atomic_read(&eb->blocking_writers) &&
104	       current->pid == eb->lock_owner);
105
106	read_lock(&eb->lock);
107	if (atomic_read(&eb->blocking_writers) &&
108	    current->pid == eb->lock_owner) {
109		/*
110		 * This extent is already write-locked by our thread. We allow
111		 * an additional read lock to be added because it's for the same
112		 * thread. btrfs_find_all_roots() depends on this as it may be
113		 * called on a partly (write-)locked tree.
114		 */
115		BUG_ON(eb->lock_nested);
116		eb->lock_nested = 1;
117		read_unlock(&eb->lock);
118		return;
119	}
120	if (atomic_read(&eb->blocking_writers)) {
121		read_unlock(&eb->lock);
122		wait_event(eb->write_lock_wq,
123			   atomic_read(&eb->blocking_writers) == 0);
124		goto again;
125	}
126	atomic_inc(&eb->read_locks);
127	atomic_inc(&eb->spinning_readers);
128}
129
130/*
131 * take a spinning read lock.
132 * returns 1 if we get the read lock and 0 if we don't
133 * this won't wait for blocking writers
134 */
135int btrfs_tree_read_lock_atomic(struct extent_buffer *eb)
136{
137	if (atomic_read(&eb->blocking_writers))
138		return 0;
139
140	read_lock(&eb->lock);
141	if (atomic_read(&eb->blocking_writers)) {
142		read_unlock(&eb->lock);
143		return 0;
144	}
145	atomic_inc(&eb->read_locks);
146	atomic_inc(&eb->spinning_readers);
147	return 1;
148}
149
150/*
151 * returns 1 if we get the read lock and 0 if we don't
152 * this won't wait for blocking writers
153 */
154int btrfs_try_tree_read_lock(struct extent_buffer *eb)
155{
156	if (atomic_read(&eb->blocking_writers))
157		return 0;
158
159	if (!read_trylock(&eb->lock))
160		return 0;
161
162	if (atomic_read(&eb->blocking_writers)) {
163		read_unlock(&eb->lock);
164		return 0;
165	}
166	atomic_inc(&eb->read_locks);
167	atomic_inc(&eb->spinning_readers);
168	return 1;
169}
170
171/*
172 * returns 1 if we get the read lock and 0 if we don't
173 * this won't wait for blocking writers or readers
174 */
175int btrfs_try_tree_write_lock(struct extent_buffer *eb)
176{
177	if (atomic_read(&eb->blocking_writers) ||
178	    atomic_read(&eb->blocking_readers))
179		return 0;
180
181	write_lock(&eb->lock);
182	if (atomic_read(&eb->blocking_writers) ||
183	    atomic_read(&eb->blocking_readers)) {
184		write_unlock(&eb->lock);
185		return 0;
186	}
187	atomic_inc(&eb->write_locks);
188	atomic_inc(&eb->spinning_writers);
189	eb->lock_owner = current->pid;
190	return 1;
191}
192
193/*
194 * drop a spinning read lock
195 */
196void btrfs_tree_read_unlock(struct extent_buffer *eb)
197{
198	/*
199	 * if we're nested, we have the write lock.  No new locking
200	 * is needed as long as we are the lock owner.
201	 * The write unlock will do a barrier for us, and the lock_nested
202	 * field only matters to the lock owner.
203	 */
204	if (eb->lock_nested && current->pid == eb->lock_owner) {
205		eb->lock_nested = 0;
206		return;
207	}
208	btrfs_assert_tree_read_locked(eb);
209	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
210	atomic_dec(&eb->spinning_readers);
211	atomic_dec(&eb->read_locks);
212	read_unlock(&eb->lock);
213}
214
215/*
216 * drop a blocking read lock
217 */
218void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
219{
220	/*
221	 * if we're nested, we have the write lock.  No new locking
222	 * is needed as long as we are the lock owner.
223	 * The write unlock will do a barrier for us, and the lock_nested
224	 * field only matters to the lock owner.
225	 */
226	if (eb->lock_nested && current->pid == eb->lock_owner) {
227		eb->lock_nested = 0;
228		return;
229	}
230	btrfs_assert_tree_read_locked(eb);
231	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
232	if (atomic_dec_and_test(&eb->blocking_readers) &&
233	    waitqueue_active(&eb->read_lock_wq))
234		wake_up(&eb->read_lock_wq);
235	atomic_dec(&eb->read_locks);
236}
237
238/*
239 * take a spinning write lock.  This will wait for both
240 * blocking readers or writers
241 */
242void btrfs_tree_lock(struct extent_buffer *eb)
243{
244again:
245	wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
246	wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
247	write_lock(&eb->lock);
248	if (atomic_read(&eb->blocking_readers)) {
249		write_unlock(&eb->lock);
250		wait_event(eb->read_lock_wq,
251			   atomic_read(&eb->blocking_readers) == 0);
252		goto again;
253	}
254	if (atomic_read(&eb->blocking_writers)) {
255		write_unlock(&eb->lock);
256		wait_event(eb->write_lock_wq,
257			   atomic_read(&eb->blocking_writers) == 0);
258		goto again;
259	}
260	WARN_ON(atomic_read(&eb->spinning_writers));
261	atomic_inc(&eb->spinning_writers);
262	atomic_inc(&eb->write_locks);
263	eb->lock_owner = current->pid;
264}
265
266/*
267 * drop a spinning or a blocking write lock.
268 */
269void btrfs_tree_unlock(struct extent_buffer *eb)
270{
271	int blockers = atomic_read(&eb->blocking_writers);
272
273	BUG_ON(blockers > 1);
274
275	btrfs_assert_tree_locked(eb);
276	eb->lock_owner = 0;
277	atomic_dec(&eb->write_locks);
278
279	if (blockers) {
280		WARN_ON(atomic_read(&eb->spinning_writers));
281		atomic_dec(&eb->blocking_writers);
282		smp_mb();
283		if (waitqueue_active(&eb->write_lock_wq))
284			wake_up(&eb->write_lock_wq);
285	} else {
286		WARN_ON(atomic_read(&eb->spinning_writers) != 1);
287		atomic_dec(&eb->spinning_writers);
288		write_unlock(&eb->lock);
289	}
290}
291
292void btrfs_assert_tree_locked(struct extent_buffer *eb)
293{
294	BUG_ON(!atomic_read(&eb->write_locks));
295}
296
297static void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
298{
299	BUG_ON(!atomic_read(&eb->read_locks));
300}
301