aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/locking.c
blob: 68fd9ccf1805e63adf0973612f9822b9585034f8 (plain) (blame)
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
/*
 * Copyright (C) 2008 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */
#include <linux/sched.h>
#include <linux/gfp.h>
#include <linux/pagemap.h>
#include <linux/spinlock.h>
#include <linux/page-flags.h>
#include <asm/bug.h>
#include "ctree.h"
#include "extent_io.h"
#include "locking.h"

/*
 * btrfs_header_level() isn't free, so don't call it when lockdep isn't
 * on
 */
#ifdef CONFIG_DEBUG_LOCK_ALLOC
static inline void spin_nested(struct extent_buffer *eb)
{
	spin_lock_nested(&eb->lock, BTRFS_MAX_LEVEL - btrfs_header_level(eb));
}
#else
static inline void spin_nested(struct extent_buffer *eb)
{
	spin_lock(&eb->lock);
}
#endif

/*
 * Setting a lock to blocking will drop the spinlock and set the
 * flag that forces other procs who want the lock to wait.  After
 * this you can safely schedule with the lock held.
 */
void btrfs_set_lock_blocking(struct extent_buffer *eb)
{
	if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) {
		set_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags);
		spin_unlock(&eb->lock);
	}
	/* exit with the spin lock released and the bit set */
}

/*
 * clearing the blocking flag will take the spinlock again.
 * After this you can't safely schedule
 */
void btrfs_clear_lock_blocking(struct extent_buffer *eb)
{
	if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) {
		spin_nested(eb);
		clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags);
		smp_mb__after_clear_bit();
	}
	/* exit with the spin lock held */
}

/*
 * unfortunately, many of the places that currently set a lock to blocking
 * don't end up blocking for every long, and often they don't block
 * at all.  For a dbench 50 run, if we don't spin one the blocking bit
 * at all, the context switch rate can jump up to 400,000/sec or more.
 *
 * So, we're still stuck with this crummy spin on the blocking bit,
 * at least until the most common causes of the short blocks
 * can be dealt with.
 */
static int btrfs_spin_on_block(struct extent_buffer *eb)
{
	int i;
	for (i = 0; i < 512; i++) {
		cpu_relax();
		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
			return 1;
		if (need_resched())
			break;
	}
	return 0;
}

/*
 * This is somewhat different from trylock.  It will take the
 * spinlock but if it finds the lock is set to blocking, it will
 * return without the lock held.
 *
 * returns 1 if it was able to take the lock and zero otherwise
 *
 * After this call, scheduling is not safe without first calling
 * btrfs_set_lock_blocking()
 */
int btrfs_try_spin_lock(struct extent_buffer *eb)
{
	int i;

	spin_nested(eb);
	if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
		return 1;
	spin_unlock(&eb->lock);

	/* spin for a bit on the BLOCKING flag */
	for (i = 0; i < 2; i++) {
		if (!btrfs_spin_on_block(eb))
			break;

		spin_nested(eb);
		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
			return 1;
		spin_unlock(&eb->lock);
	}
	return 0;
}

/*
 * the autoremove wake function will return 0 if it tried to wake up
 * a process that was already awake, which means that process won't
 * count as an exclusive wakeup.  The waitq code will continue waking
 * procs until it finds one that was actually sleeping.
 *
 * For btrfs, this isn't quite what we want.  We want a single proc
 * to be notified that the lock is ready for taking.  If that proc
 * already happen to be awake, great, it will loop around and try for
 * the lock.
 *
 * So, btrfs_wake_function always returns 1, even when the proc that we
 * tried to wake up was already awake.
 */
static int btrfs_wake_function(wait_queue_t *wait, unsigned mode,
			       int sync, void *key)
{
	autoremove_wake_function(wait, mode, sync, key);
	return 1;
}

/*
 * returns with the extent buffer spinlocked.
 *
 * This will spin and/or wait as required to take the lock, and then
 * return with the spinlock held.
 *
 * After this call, scheduling is not safe without first calling
 * btrfs_set_lock_blocking()
 */
int btrfs_tree_lock(struct extent_buffer *eb)
{
	DEFINE_WAIT(wait);
	wait.func = btrfs_wake_function;

	while(1) {
		spin_nested(eb);

		/* nobody is blocking, exit with the spinlock held */
		if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
			return 0;

		/*
		 * we have the spinlock, but the real owner is blocking.
		 * wait for them
		 */
		spin_unlock(&eb->lock);

		/*
		 * spin for a bit, and if the blocking flag goes away,
		 * loop around
		 */
		if (btrfs_spin_on_block(eb))
			continue;

		prepare_to_wait_exclusive(&eb->lock_wq, &wait,
					  TASK_UNINTERRUPTIBLE);

		if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
			schedule();

		finish_wait(&eb->lock_wq, &wait);
	}
	return 0;
}

/*
 * Very quick trylock, this does not spin or schedule.  It returns
 * 1 with the spinlock held if it was able to take the lock, or it
 * returns zero if it was unable to take the lock.
 *
 * After this call, scheduling is not safe without first calling
 * btrfs_set_lock_blocking()
 */
int btrfs_try_tree_lock(struct extent_buffer *eb)
{
	if (spin_trylock(&eb->lock)) {
		if (test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) {
			/*
			 * we've got the spinlock, but the real owner is
			 * blocking.  Drop the spinlock and return failure
			 */
			spin_unlock(&eb->lock);
			return 0;
		}
		return 1;
	}
	/* someone else has the spinlock giveup */
	return 0;
}

int btrfs_tree_unlock(struct extent_buffer *eb)
{
	/*
	 * if we were a blocking owner, we don't have the spinlock held
	 * just clear the bit and look for waiters
	 */
	if (test_and_clear_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
		smp_mb__after_clear_bit();
	else
		spin_unlock(&eb->lock);

	if (waitqueue_active(&eb->lock_wq))
		wake_up(&eb->lock_wq);
	return 0;
}

int btrfs_tree_locked(struct extent_buffer *eb)
{
	return test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags) ||
			spin_is_locked(&eb->lock);
}

/*
 * btrfs_search_slot uses this to decide if it should drop its locks
 * before doing something expensive like allocating free blocks for cow.
 */
int btrfs_path_lock_waiting(struct btrfs_path *path, int level)
{
	int i;
	struct extent_buffer *eb;

	for (i = level; i <= level + 1 && i < BTRFS_MAX_LEVEL; i++) {
		eb = path->nodes[i];
		if (!eb)
			break;
		smp_mb();
		if (spin_is_contended(&eb->lock) ||
		    waitqueue_active(&eb->lock_wq))
			return 1;
	}
	return 0;
}