Commit | Line | Data |
---|---|---|
79154b1b CM |
1 | #include <linux/module.h> |
2 | #include <linux/fs.h> | |
3 | #include "ctree.h" | |
4 | #include "disk-io.h" | |
5 | #include "transaction.h" | |
6 | ||
78fae27e | 7 | static int total_trans = 0; |
2c90e5d6 CM |
8 | extern struct kmem_cache *btrfs_trans_handle_cachep; |
9 | extern struct kmem_cache *btrfs_transaction_cachep; | |
10 | ||
08607c1b CM |
11 | static struct workqueue_struct *trans_wq; |
12 | ||
0f7d52f4 CM |
13 | #define BTRFS_ROOT_TRANS_TAG 0 |
14 | ||
2c90e5d6 | 15 | #define TRANS_MAGIC 0xE1E10E |
79154b1b CM |
16 | static void put_transaction(struct btrfs_transaction *transaction) |
17 | { | |
2c90e5d6 | 18 | WARN_ON(transaction->use_count == 0); |
79154b1b | 19 | transaction->use_count--; |
2c90e5d6 | 20 | WARN_ON(transaction->magic != TRANS_MAGIC); |
78fae27e CM |
21 | if (transaction->use_count == 0) { |
22 | WARN_ON(total_trans == 0); | |
23 | total_trans--; | |
8fd17795 | 24 | list_del_init(&transaction->list); |
2c90e5d6 CM |
25 | memset(transaction, 0, sizeof(*transaction)); |
26 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 27 | } |
79154b1b CM |
28 | } |
29 | ||
30 | static int join_transaction(struct btrfs_root *root) | |
31 | { | |
32 | struct btrfs_transaction *cur_trans; | |
33 | cur_trans = root->fs_info->running_transaction; | |
34 | if (!cur_trans) { | |
2c90e5d6 CM |
35 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
36 | GFP_NOFS); | |
78fae27e | 37 | total_trans++; |
79154b1b | 38 | BUG_ON(!cur_trans); |
0f7d52f4 | 39 | root->fs_info->generation++; |
79154b1b CM |
40 | root->fs_info->running_transaction = cur_trans; |
41 | cur_trans->num_writers = 0; | |
0f7d52f4 | 42 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
43 | init_waitqueue_head(&cur_trans->writer_wait); |
44 | init_waitqueue_head(&cur_trans->commit_wait); | |
2c90e5d6 | 45 | cur_trans->magic = TRANS_MAGIC; |
79154b1b | 46 | cur_trans->in_commit = 0; |
d5719762 | 47 | cur_trans->use_count = 1; |
79154b1b | 48 | cur_trans->commit_done = 0; |
08607c1b | 49 | cur_trans->start_time = get_seconds(); |
8fd17795 | 50 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
7c4452b9 | 51 | init_bit_radix(&cur_trans->dirty_pages); |
79154b1b CM |
52 | } |
53 | cur_trans->num_writers++; | |
54 | return 0; | |
55 | } | |
56 | ||
57 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, | |
58 | int num_blocks) | |
59 | { | |
2c90e5d6 CM |
60 | struct btrfs_trans_handle *h = |
61 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
79154b1b | 62 | int ret; |
0f7d52f4 | 63 | u64 running_trans_id; |
79154b1b CM |
64 | |
65 | mutex_lock(&root->fs_info->trans_mutex); | |
66 | ret = join_transaction(root); | |
67 | BUG_ON(ret); | |
0f7d52f4 CM |
68 | running_trans_id = root->fs_info->running_transaction->transid; |
69 | ||
70 | if (root != root->fs_info->tree_root && root->last_trans < | |
71 | running_trans_id) { | |
72 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
2619ba1f CM |
73 | (unsigned long)root->root_key.objectid, |
74 | BTRFS_ROOT_TRANS_TAG); | |
0f7d52f4 CM |
75 | root->commit_root = root->node; |
76 | get_bh(root->node); | |
77 | } | |
78 | root->last_trans = running_trans_id; | |
79 | h->transid = running_trans_id; | |
79154b1b CM |
80 | h->transaction = root->fs_info->running_transaction; |
81 | h->blocks_reserved = num_blocks; | |
82 | h->blocks_used = 0; | |
31f3c99b | 83 | h->block_group = NULL; |
79154b1b CM |
84 | root->fs_info->running_transaction->use_count++; |
85 | mutex_unlock(&root->fs_info->trans_mutex); | |
2c90e5d6 | 86 | h->magic = h->magic2 = TRANS_MAGIC; |
79154b1b CM |
87 | return h; |
88 | } | |
89 | ||
90 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, | |
91 | struct btrfs_root *root) | |
92 | { | |
93 | struct btrfs_transaction *cur_trans; | |
d6e4a428 | 94 | |
2c90e5d6 CM |
95 | WARN_ON(trans->magic != TRANS_MAGIC); |
96 | WARN_ON(trans->magic2 != TRANS_MAGIC); | |
79154b1b CM |
97 | mutex_lock(&root->fs_info->trans_mutex); |
98 | cur_trans = root->fs_info->running_transaction; | |
d5719762 | 99 | WARN_ON(cur_trans->num_writers < 1); |
79154b1b CM |
100 | if (waitqueue_active(&cur_trans->writer_wait)) |
101 | wake_up(&cur_trans->writer_wait); | |
102 | cur_trans->num_writers--; | |
103 | put_transaction(cur_trans); | |
104 | mutex_unlock(&root->fs_info->trans_mutex); | |
d6025579 | 105 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 106 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b CM |
107 | return 0; |
108 | } | |
109 | ||
110 | ||
111 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, | |
112 | struct btrfs_root *root) | |
113 | { | |
7c4452b9 CM |
114 | unsigned long gang[16]; |
115 | int ret; | |
116 | int i; | |
117 | int err; | |
118 | int werr = 0; | |
119 | struct page *page; | |
120 | struct radix_tree_root *dirty_pages; | |
121 | struct inode *btree_inode = root->fs_info->btree_inode; | |
122 | ||
123 | if (!trans || !trans->transaction) { | |
124 | return filemap_write_and_wait(btree_inode->i_mapping); | |
125 | } | |
126 | dirty_pages = &trans->transaction->dirty_pages; | |
127 | while(1) { | |
e37c9e69 CM |
128 | ret = find_first_radix_bit(dirty_pages, gang, |
129 | 0, ARRAY_SIZE(gang)); | |
7c4452b9 CM |
130 | if (!ret) |
131 | break; | |
132 | for (i = 0; i < ret; i++) { | |
133 | /* FIXME EIO */ | |
134 | clear_radix_bit(dirty_pages, gang[i]); | |
135 | page = find_lock_page(btree_inode->i_mapping, | |
136 | gang[i]); | |
137 | if (!page) | |
138 | continue; | |
139 | err = write_one_page(page, 0); | |
140 | if (err) | |
141 | werr = err; | |
142 | page_cache_release(page); | |
143 | } | |
144 | } | |
145 | err = filemap_fdatawait(btree_inode->i_mapping); | |
146 | if (err) | |
147 | werr = err; | |
148 | return werr; | |
79154b1b CM |
149 | } |
150 | ||
151 | int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans, | |
152 | struct btrfs_root *root) | |
153 | { | |
154 | int ret; | |
155 | u64 old_extent_block; | |
156 | struct btrfs_fs_info *fs_info = root->fs_info; | |
157 | struct btrfs_root *tree_root = fs_info->tree_root; | |
158 | struct btrfs_root *extent_root = fs_info->extent_root; | |
8352d8a4 | 159 | struct btrfs_root *dev_root = fs_info->dev_root; |
79154b1b | 160 | |
8352d8a4 CM |
161 | if (btrfs_super_device_root(fs_info->disk_super) != |
162 | bh_blocknr(dev_root->node)) { | |
163 | btrfs_set_super_device_root(fs_info->disk_super, | |
164 | bh_blocknr(dev_root->node)); | |
165 | } | |
9078a3e1 | 166 | btrfs_write_dirty_block_groups(trans, extent_root); |
79154b1b CM |
167 | while(1) { |
168 | old_extent_block = btrfs_root_blocknr(&extent_root->root_item); | |
7eccb903 | 169 | if (old_extent_block == bh_blocknr(extent_root->node)) |
79154b1b CM |
170 | break; |
171 | btrfs_set_root_blocknr(&extent_root->root_item, | |
7eccb903 | 172 | bh_blocknr(extent_root->node)); |
79154b1b CM |
173 | ret = btrfs_update_root(trans, tree_root, |
174 | &extent_root->root_key, | |
175 | &extent_root->root_item); | |
176 | BUG_ON(ret); | |
9078a3e1 | 177 | btrfs_write_dirty_block_groups(trans, extent_root); |
79154b1b CM |
178 | } |
179 | return 0; | |
180 | } | |
181 | ||
182 | static int wait_for_commit(struct btrfs_root *root, | |
183 | struct btrfs_transaction *commit) | |
184 | { | |
185 | DEFINE_WAIT(wait); | |
79154b1b CM |
186 | while(!commit->commit_done) { |
187 | prepare_to_wait(&commit->commit_wait, &wait, | |
188 | TASK_UNINTERRUPTIBLE); | |
189 | if (commit->commit_done) | |
190 | break; | |
191 | mutex_unlock(&root->fs_info->trans_mutex); | |
192 | schedule(); | |
193 | mutex_lock(&root->fs_info->trans_mutex); | |
194 | } | |
195 | finish_wait(&commit->commit_wait, &wait); | |
196 | return 0; | |
197 | } | |
198 | ||
0f7d52f4 CM |
199 | struct dirty_root { |
200 | struct list_head list; | |
201 | struct btrfs_key snap_key; | |
202 | struct buffer_head *commit_root; | |
203 | struct btrfs_root *root; | |
204 | }; | |
205 | ||
35b7e476 CM |
206 | static int add_dirty_roots(struct btrfs_trans_handle *trans, |
207 | struct radix_tree_root *radix, | |
208 | struct list_head *list) | |
0f7d52f4 CM |
209 | { |
210 | struct dirty_root *dirty; | |
211 | struct btrfs_root *gang[8]; | |
212 | struct btrfs_root *root; | |
213 | int i; | |
214 | int ret; | |
215 | int err; | |
0f7d52f4 CM |
216 | while(1) { |
217 | ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0, | |
218 | ARRAY_SIZE(gang), | |
219 | BTRFS_ROOT_TRANS_TAG); | |
220 | if (ret == 0) | |
221 | break; | |
222 | for (i = 0; i < ret; i++) { | |
223 | root = gang[i]; | |
2619ba1f CM |
224 | radix_tree_tag_clear(radix, |
225 | (unsigned long)root->root_key.objectid, | |
226 | BTRFS_ROOT_TRANS_TAG); | |
0f7d52f4 | 227 | if (root->commit_root == root->node) { |
7eccb903 | 228 | WARN_ON(bh_blocknr(root->node) != |
0f7d52f4 CM |
229 | btrfs_root_blocknr(&root->root_item)); |
230 | brelse(root->commit_root); | |
231 | root->commit_root = NULL; | |
232 | continue; | |
233 | } | |
234 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
235 | BUG_ON(!dirty); | |
236 | memcpy(&dirty->snap_key, &root->root_key, | |
237 | sizeof(root->root_key)); | |
238 | dirty->commit_root = root->commit_root; | |
239 | root->commit_root = NULL; | |
240 | dirty->root = root; | |
0f7d52f4 CM |
241 | root->root_key.offset = root->fs_info->generation; |
242 | btrfs_set_root_blocknr(&root->root_item, | |
7eccb903 | 243 | bh_blocknr(root->node)); |
0f7d52f4 CM |
244 | err = btrfs_insert_root(trans, root->fs_info->tree_root, |
245 | &root->root_key, | |
246 | &root->root_item); | |
247 | BUG_ON(err); | |
248 | list_add(&dirty->list, list); | |
249 | } | |
250 | } | |
0f7d52f4 CM |
251 | return 0; |
252 | } | |
253 | ||
35b7e476 CM |
254 | static int drop_dirty_roots(struct btrfs_root *tree_root, |
255 | struct list_head *list) | |
0f7d52f4 CM |
256 | { |
257 | struct dirty_root *dirty; | |
258 | struct btrfs_trans_handle *trans; | |
259 | int ret; | |
260 | ||
261 | while(!list_empty(list)) { | |
262 | dirty = list_entry(list->next, struct dirty_root, list); | |
263 | list_del_init(&dirty->list); | |
264 | trans = btrfs_start_transaction(tree_root, 1); | |
0f7d52f4 CM |
265 | ret = btrfs_drop_snapshot(trans, dirty->root, |
266 | dirty->commit_root); | |
267 | BUG_ON(ret); | |
268 | ||
0f7d52f4 CM |
269 | ret = btrfs_del_root(trans, tree_root, &dirty->snap_key); |
270 | BUG_ON(ret); | |
271 | ret = btrfs_end_transaction(trans, tree_root); | |
272 | BUG_ON(ret); | |
273 | kfree(dirty); | |
274 | } | |
275 | return 0; | |
276 | } | |
277 | ||
79154b1b CM |
278 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
279 | struct btrfs_root *root) | |
280 | { | |
281 | int ret = 0; | |
79154b1b | 282 | struct btrfs_transaction *cur_trans; |
8fd17795 | 283 | struct btrfs_transaction *prev_trans = NULL; |
0f7d52f4 | 284 | struct list_head dirty_fs_roots; |
79154b1b CM |
285 | DEFINE_WAIT(wait); |
286 | ||
0f7d52f4 | 287 | INIT_LIST_HEAD(&dirty_fs_roots); |
d6e4a428 | 288 | |
79154b1b CM |
289 | mutex_lock(&root->fs_info->trans_mutex); |
290 | if (trans->transaction->in_commit) { | |
291 | cur_trans = trans->transaction; | |
292 | trans->transaction->use_count++; | |
293 | btrfs_end_transaction(trans, root); | |
294 | ret = wait_for_commit(root, cur_trans); | |
295 | BUG_ON(ret); | |
296 | put_transaction(cur_trans); | |
297 | mutex_unlock(&root->fs_info->trans_mutex); | |
298 | return 0; | |
299 | } | |
2c90e5d6 CM |
300 | cur_trans = trans->transaction; |
301 | trans->transaction->in_commit = 1; | |
79154b1b | 302 | while (trans->transaction->num_writers > 1) { |
2c90e5d6 | 303 | WARN_ON(cur_trans != trans->transaction); |
79154b1b CM |
304 | prepare_to_wait(&trans->transaction->writer_wait, &wait, |
305 | TASK_UNINTERRUPTIBLE); | |
306 | if (trans->transaction->num_writers <= 1) | |
307 | break; | |
308 | mutex_unlock(&root->fs_info->trans_mutex); | |
309 | schedule(); | |
310 | mutex_lock(&root->fs_info->trans_mutex); | |
2c90e5d6 | 311 | finish_wait(&trans->transaction->writer_wait, &wait); |
79154b1b CM |
312 | } |
313 | finish_wait(&trans->transaction->writer_wait, &wait); | |
2c90e5d6 | 314 | WARN_ON(cur_trans != trans->transaction); |
0f7d52f4 | 315 | add_dirty_roots(trans, &root->fs_info->fs_roots_radix, &dirty_fs_roots); |
79154b1b CM |
316 | ret = btrfs_commit_tree_roots(trans, root); |
317 | BUG_ON(ret); | |
78fae27e CM |
318 | cur_trans = root->fs_info->running_transaction; |
319 | root->fs_info->running_transaction = NULL; | |
8fd17795 CM |
320 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
321 | prev_trans = list_entry(cur_trans->list.prev, | |
322 | struct btrfs_transaction, list); | |
323 | if (prev_trans->commit_done) | |
324 | prev_trans = NULL; | |
325 | else | |
326 | prev_trans->use_count++; | |
327 | } | |
78fae27e | 328 | mutex_unlock(&root->fs_info->trans_mutex); |
8fd17795 | 329 | mutex_unlock(&root->fs_info->fs_mutex); |
79154b1b | 330 | ret = btrfs_write_and_wait_transaction(trans, root); |
8fd17795 CM |
331 | if (prev_trans) { |
332 | mutex_lock(&root->fs_info->trans_mutex); | |
333 | wait_for_commit(root, prev_trans); | |
334 | put_transaction(prev_trans); | |
335 | mutex_unlock(&root->fs_info->trans_mutex); | |
336 | } | |
337 | btrfs_set_super_generation(root->fs_info->disk_super, | |
338 | cur_trans->transid); | |
79154b1b | 339 | BUG_ON(ret); |
79154b1b | 340 | write_ctree_super(trans, root); |
8fd17795 CM |
341 | |
342 | mutex_lock(&root->fs_info->fs_mutex); | |
78fae27e CM |
343 | btrfs_finish_extent_commit(trans, root); |
344 | mutex_lock(&root->fs_info->trans_mutex); | |
2c90e5d6 CM |
345 | cur_trans->commit_done = 1; |
346 | wake_up(&cur_trans->commit_wait); | |
78fae27e | 347 | put_transaction(cur_trans); |
79154b1b | 348 | put_transaction(cur_trans); |
78fae27e | 349 | mutex_unlock(&root->fs_info->trans_mutex); |
2c90e5d6 | 350 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b | 351 | |
0f7d52f4 | 352 | drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots); |
79154b1b CM |
353 | return ret; |
354 | } | |
355 | ||
08607c1b CM |
356 | void btrfs_transaction_cleaner(struct work_struct *work) |
357 | { | |
358 | struct btrfs_fs_info *fs_info = container_of(work, | |
359 | struct btrfs_fs_info, | |
360 | trans_work.work); | |
361 | ||
362 | struct btrfs_root *root = fs_info->tree_root; | |
363 | struct btrfs_transaction *cur; | |
364 | struct btrfs_trans_handle *trans; | |
365 | unsigned long now; | |
366 | unsigned long delay = HZ * 30; | |
367 | int ret; | |
368 | ||
369 | printk("btrfs transaction cleaner\n"); | |
370 | mutex_lock(&root->fs_info->fs_mutex); | |
371 | mutex_lock(&root->fs_info->trans_mutex); | |
372 | cur = root->fs_info->running_transaction; | |
373 | if (!cur) { | |
374 | mutex_unlock(&root->fs_info->trans_mutex); | |
375 | goto out; | |
376 | } | |
377 | now = get_seconds(); | |
378 | if (now < cur->start_time || now - cur->start_time < 30) { | |
379 | mutex_unlock(&root->fs_info->trans_mutex); | |
380 | delay = HZ * 5; | |
381 | goto out; | |
382 | } | |
383 | mutex_unlock(&root->fs_info->trans_mutex); | |
384 | printk("forcing commit\n"); | |
385 | trans = btrfs_start_transaction(root, 1); | |
386 | ret = btrfs_commit_transaction(trans, root); | |
387 | out: | |
388 | mutex_unlock(&root->fs_info->fs_mutex); | |
389 | btrfs_transaction_queue_work(root, delay); | |
390 | } | |
391 | ||
392 | void btrfs_transaction_queue_work(struct btrfs_root *root, int delay) | |
393 | { | |
394 | queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay); | |
395 | } | |
396 | ||
397 | void btrfs_transaction_flush_work(struct btrfs_root *root) | |
398 | { | |
399 | cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work); | |
400 | flush_workqueue(trans_wq); | |
401 | } | |
402 | ||
403 | void __init btrfs_init_transaction_sys(void) | |
404 | { | |
405 | trans_wq = create_workqueue("btrfs"); | |
406 | } | |
407 | ||
408 | void __exit btrfs_exit_transaction_sys(void) | |
409 | { | |
410 | destroy_workqueue(trans_wq); | |
411 | } | |
412 |