Commit | Line | Data |
---|---|---|
925baedd CM |
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> | |
925baedd CM |
19 | #include <linux/pagemap.h> |
20 | #include <linux/spinlock.h> | |
21 | #include <linux/page-flags.h> | |
4881ee5a | 22 | #include <asm/bug.h> |
925baedd CM |
23 | #include "ctree.h" |
24 | #include "extent_io.h" | |
25 | #include "locking.h" | |
26 | ||
48a3b636 | 27 | static void btrfs_assert_tree_read_locked(struct extent_buffer *eb); |
d397712b | 28 | |
b4ce94de | 29 | /* |
bd681513 CM |
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. | |
b4ce94de | 33 | */ |
bd681513 | 34 | void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw) |
925baedd | 35 | { |
ea4ebde0 CM |
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; | |
bd681513 CM |
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); | |
b4ce94de | 58 | } |
bd681513 | 59 | return; |
b4ce94de | 60 | } |
f9efa9c7 | 61 | |
b4ce94de | 62 | /* |
bd681513 CM |
63 | * if we currently have a blocking lock, take the spinlock |
64 | * and drop our blocking count | |
b4ce94de | 65 | */ |
bd681513 | 66 | void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw) |
b4ce94de | 67 | { |
ea4ebde0 CM |
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 | ||
bd681513 CM |
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); | |
cbea5ac1 CM |
82 | if (atomic_dec_and_test(&eb->blocking_writers) && |
83 | waitqueue_active(&eb->write_lock_wq)) | |
bd681513 CM |
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); | |
cbea5ac1 CM |
89 | if (atomic_dec_and_test(&eb->blocking_readers) && |
90 | waitqueue_active(&eb->read_lock_wq)) | |
bd681513 | 91 | wake_up(&eb->read_lock_wq); |
b4ce94de | 92 | } |
bd681513 | 93 | return; |
b4ce94de CM |
94 | } |
95 | ||
96 | /* | |
bd681513 CM |
97 | * take a spinning read lock. This will wait for any blocking |
98 | * writers | |
b4ce94de | 99 | */ |
bd681513 | 100 | void btrfs_tree_read_lock(struct extent_buffer *eb) |
b4ce94de | 101 | { |
bd681513 | 102 | again: |
ea4ebde0 CM |
103 | BUG_ON(!atomic_read(&eb->blocking_writers) && |
104 | current->pid == eb->lock_owner); | |
105 | ||
5b25f70f AJ |
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 | } | |
bd681513 CM |
120 | if (atomic_read(&eb->blocking_writers)) { |
121 | read_unlock(&eb->lock); | |
39f9d028 LB |
122 | wait_event(eb->write_lock_wq, |
123 | atomic_read(&eb->blocking_writers) == 0); | |
bd681513 | 124 | goto again; |
b4ce94de | 125 | } |
bd681513 CM |
126 | atomic_inc(&eb->read_locks); |
127 | atomic_inc(&eb->spinning_readers); | |
b4ce94de CM |
128 | } |
129 | ||
130 | /* | |
bd681513 CM |
131 | * returns 1 if we get the read lock and 0 if we don't |
132 | * this won't wait for blocking writers | |
b4ce94de | 133 | */ |
bd681513 | 134 | int btrfs_try_tree_read_lock(struct extent_buffer *eb) |
b4ce94de | 135 | { |
bd681513 CM |
136 | if (atomic_read(&eb->blocking_writers)) |
137 | return 0; | |
b4ce94de | 138 | |
ea4ebde0 CM |
139 | if (!read_trylock(&eb->lock)) |
140 | return 0; | |
141 | ||
bd681513 CM |
142 | if (atomic_read(&eb->blocking_writers)) { |
143 | read_unlock(&eb->lock); | |
144 | return 0; | |
b9473439 | 145 | } |
bd681513 CM |
146 | atomic_inc(&eb->read_locks); |
147 | atomic_inc(&eb->spinning_readers); | |
148 | return 1; | |
b4ce94de CM |
149 | } |
150 | ||
151 | /* | |
bd681513 CM |
152 | * returns 1 if we get the read lock and 0 if we don't |
153 | * this won't wait for blocking writers or readers | |
b4ce94de | 154 | */ |
bd681513 | 155 | int btrfs_try_tree_write_lock(struct extent_buffer *eb) |
b4ce94de | 156 | { |
bd681513 CM |
157 | if (atomic_read(&eb->blocking_writers) || |
158 | atomic_read(&eb->blocking_readers)) | |
159 | return 0; | |
ea4ebde0 CM |
160 | |
161 | if (!write_trylock(&eb->lock)) | |
162 | return 0; | |
163 | ||
bd681513 CM |
164 | if (atomic_read(&eb->blocking_writers) || |
165 | atomic_read(&eb->blocking_readers)) { | |
166 | write_unlock(&eb->lock); | |
167 | return 0; | |
168 | } | |
169 | atomic_inc(&eb->write_locks); | |
170 | atomic_inc(&eb->spinning_writers); | |
5b25f70f | 171 | eb->lock_owner = current->pid; |
b4ce94de CM |
172 | return 1; |
173 | } | |
174 | ||
175 | /* | |
bd681513 CM |
176 | * drop a spinning read lock |
177 | */ | |
178 | void btrfs_tree_read_unlock(struct extent_buffer *eb) | |
179 | { | |
ea4ebde0 CM |
180 | /* |
181 | * if we're nested, we have the write lock. No new locking | |
182 | * is needed as long as we are the lock owner. | |
183 | * The write unlock will do a barrier for us, and the lock_nested | |
184 | * field only matters to the lock owner. | |
185 | */ | |
186 | if (eb->lock_nested && current->pid == eb->lock_owner) { | |
187 | eb->lock_nested = 0; | |
188 | return; | |
5b25f70f | 189 | } |
bd681513 CM |
190 | btrfs_assert_tree_read_locked(eb); |
191 | WARN_ON(atomic_read(&eb->spinning_readers) == 0); | |
192 | atomic_dec(&eb->spinning_readers); | |
193 | atomic_dec(&eb->read_locks); | |
194 | read_unlock(&eb->lock); | |
195 | } | |
196 | ||
197 | /* | |
198 | * drop a blocking read lock | |
199 | */ | |
200 | void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb) | |
201 | { | |
ea4ebde0 CM |
202 | /* |
203 | * if we're nested, we have the write lock. No new locking | |
204 | * is needed as long as we are the lock owner. | |
205 | * The write unlock will do a barrier for us, and the lock_nested | |
206 | * field only matters to the lock owner. | |
207 | */ | |
208 | if (eb->lock_nested && current->pid == eb->lock_owner) { | |
209 | eb->lock_nested = 0; | |
210 | return; | |
5b25f70f | 211 | } |
bd681513 CM |
212 | btrfs_assert_tree_read_locked(eb); |
213 | WARN_ON(atomic_read(&eb->blocking_readers) == 0); | |
cbea5ac1 CM |
214 | if (atomic_dec_and_test(&eb->blocking_readers) && |
215 | waitqueue_active(&eb->read_lock_wq)) | |
bd681513 CM |
216 | wake_up(&eb->read_lock_wq); |
217 | atomic_dec(&eb->read_locks); | |
218 | } | |
219 | ||
220 | /* | |
221 | * take a spinning write lock. This will wait for both | |
222 | * blocking readers or writers | |
b4ce94de | 223 | */ |
143bede5 | 224 | void btrfs_tree_lock(struct extent_buffer *eb) |
b4ce94de | 225 | { |
bd681513 CM |
226 | again: |
227 | wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0); | |
228 | wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0); | |
229 | write_lock(&eb->lock); | |
230 | if (atomic_read(&eb->blocking_readers)) { | |
231 | write_unlock(&eb->lock); | |
232 | wait_event(eb->read_lock_wq, | |
233 | atomic_read(&eb->blocking_readers) == 0); | |
234 | goto again; | |
f9efa9c7 | 235 | } |
bd681513 CM |
236 | if (atomic_read(&eb->blocking_writers)) { |
237 | write_unlock(&eb->lock); | |
238 | wait_event(eb->write_lock_wq, | |
239 | atomic_read(&eb->blocking_writers) == 0); | |
240 | goto again; | |
241 | } | |
242 | WARN_ON(atomic_read(&eb->spinning_writers)); | |
243 | atomic_inc(&eb->spinning_writers); | |
244 | atomic_inc(&eb->write_locks); | |
5b25f70f | 245 | eb->lock_owner = current->pid; |
925baedd CM |
246 | } |
247 | ||
bd681513 CM |
248 | /* |
249 | * drop a spinning or a blocking write lock. | |
250 | */ | |
143bede5 | 251 | void btrfs_tree_unlock(struct extent_buffer *eb) |
925baedd | 252 | { |
bd681513 CM |
253 | int blockers = atomic_read(&eb->blocking_writers); |
254 | ||
255 | BUG_ON(blockers > 1); | |
256 | ||
257 | btrfs_assert_tree_locked(eb); | |
ea4ebde0 | 258 | eb->lock_owner = 0; |
bd681513 CM |
259 | atomic_dec(&eb->write_locks); |
260 | ||
261 | if (blockers) { | |
262 | WARN_ON(atomic_read(&eb->spinning_writers)); | |
263 | atomic_dec(&eb->blocking_writers); | |
cbea5ac1 CM |
264 | smp_mb(); |
265 | if (waitqueue_active(&eb->write_lock_wq)) | |
266 | wake_up(&eb->write_lock_wq); | |
bd681513 CM |
267 | } else { |
268 | WARN_ON(atomic_read(&eb->spinning_writers) != 1); | |
269 | atomic_dec(&eb->spinning_writers); | |
270 | write_unlock(&eb->lock); | |
271 | } | |
925baedd CM |
272 | } |
273 | ||
b9447ef8 | 274 | void btrfs_assert_tree_locked(struct extent_buffer *eb) |
925baedd | 275 | { |
bd681513 CM |
276 | BUG_ON(!atomic_read(&eb->write_locks)); |
277 | } | |
278 | ||
48a3b636 | 279 | static void btrfs_assert_tree_read_locked(struct extent_buffer *eb) |
bd681513 CM |
280 | { |
281 | BUG_ON(!atomic_read(&eb->read_locks)); | |
925baedd | 282 | } |