Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 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 | ||
79154b1b | 19 | #include <linux/fs.h> |
34088780 | 20 | #include <linux/sched.h> |
d3c2fdcf | 21 | #include <linux/writeback.h> |
5f39d397 | 22 | #include <linux/pagemap.h> |
5f2cc086 | 23 | #include <linux/blkdev.h> |
79154b1b CM |
24 | #include "ctree.h" |
25 | #include "disk-io.h" | |
26 | #include "transaction.h" | |
925baedd | 27 | #include "locking.h" |
31153d81 | 28 | #include "ref-cache.h" |
e02119d5 | 29 | #include "tree-log.h" |
79154b1b | 30 | |
2c90e5d6 CM |
31 | extern struct kmem_cache *btrfs_trans_handle_cachep; |
32 | extern struct kmem_cache *btrfs_transaction_cachep; | |
33 | ||
0f7d52f4 CM |
34 | #define BTRFS_ROOT_TRANS_TAG 0 |
35 | ||
80b6794d | 36 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 37 | { |
2c90e5d6 | 38 | WARN_ON(transaction->use_count == 0); |
79154b1b | 39 | transaction->use_count--; |
78fae27e | 40 | if (transaction->use_count == 0) { |
8fd17795 | 41 | list_del_init(&transaction->list); |
2c90e5d6 CM |
42 | memset(transaction, 0, sizeof(*transaction)); |
43 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 44 | } |
79154b1b CM |
45 | } |
46 | ||
d352ac68 CM |
47 | /* |
48 | * either allocate a new transaction or hop into the existing one | |
49 | */ | |
80b6794d | 50 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
51 | { |
52 | struct btrfs_transaction *cur_trans; | |
53 | cur_trans = root->fs_info->running_transaction; | |
54 | if (!cur_trans) { | |
2c90e5d6 CM |
55 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
56 | GFP_NOFS); | |
79154b1b | 57 | BUG_ON(!cur_trans); |
0f7d52f4 | 58 | root->fs_info->generation++; |
e18e4809 | 59 | root->fs_info->last_alloc = 0; |
4529ba49 | 60 | root->fs_info->last_data_alloc = 0; |
15ee9bc7 JB |
61 | cur_trans->num_writers = 1; |
62 | cur_trans->num_joined = 0; | |
0f7d52f4 | 63 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
64 | init_waitqueue_head(&cur_trans->writer_wait); |
65 | init_waitqueue_head(&cur_trans->commit_wait); | |
66 | cur_trans->in_commit = 0; | |
f9295749 | 67 | cur_trans->blocked = 0; |
d5719762 | 68 | cur_trans->use_count = 1; |
79154b1b | 69 | cur_trans->commit_done = 0; |
08607c1b | 70 | cur_trans->start_time = get_seconds(); |
3063d29f | 71 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 72 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 73 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
74 | root->fs_info->btree_inode->i_mapping, |
75 | GFP_NOFS); | |
48ec2cf8 CM |
76 | spin_lock(&root->fs_info->new_trans_lock); |
77 | root->fs_info->running_transaction = cur_trans; | |
78 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
79 | } else { |
80 | cur_trans->num_writers++; | |
81 | cur_trans->num_joined++; | |
79154b1b | 82 | } |
15ee9bc7 | 83 | |
79154b1b CM |
84 | return 0; |
85 | } | |
86 | ||
d352ac68 CM |
87 | /* |
88 | * this does all the record keeping required to make sure that a | |
89 | * reference counted root is properly recorded in a given transaction. | |
90 | * This is required to make sure the old root from before we joined the transaction | |
91 | * is deleted when the transaction commits | |
92 | */ | |
e02119d5 | 93 | noinline int btrfs_record_root_in_trans(struct btrfs_root *root) |
6702ed49 | 94 | { |
f321e491 | 95 | struct btrfs_dirty_root *dirty; |
6702ed49 CM |
96 | u64 running_trans_id = root->fs_info->running_transaction->transid; |
97 | if (root->ref_cows && root->last_trans < running_trans_id) { | |
98 | WARN_ON(root == root->fs_info->extent_root); | |
99 | if (root->root_item.refs != 0) { | |
100 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
101 | (unsigned long)root->root_key.objectid, | |
102 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
103 | |
104 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
105 | BUG_ON(!dirty); | |
106 | dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS); | |
107 | BUG_ON(!dirty->root); | |
31153d81 YZ |
108 | dirty->latest_root = root; |
109 | INIT_LIST_HEAD(&dirty->list); | |
31153d81 | 110 | |
925baedd | 111 | root->commit_root = btrfs_root_node(root); |
31153d81 YZ |
112 | |
113 | memcpy(dirty->root, root, sizeof(*root)); | |
114 | spin_lock_init(&dirty->root->node_lock); | |
bcc63abb | 115 | spin_lock_init(&dirty->root->list_lock); |
31153d81 | 116 | mutex_init(&dirty->root->objectid_mutex); |
5b21f2ed | 117 | mutex_init(&dirty->root->log_mutex); |
bcc63abb | 118 | INIT_LIST_HEAD(&dirty->root->dead_list); |
31153d81 YZ |
119 | dirty->root->node = root->commit_root; |
120 | dirty->root->commit_root = NULL; | |
bcc63abb Y |
121 | |
122 | spin_lock(&root->list_lock); | |
123 | list_add(&dirty->root->dead_list, &root->dead_list); | |
124 | spin_unlock(&root->list_lock); | |
125 | ||
126 | root->dirty_root = dirty; | |
6702ed49 CM |
127 | } else { |
128 | WARN_ON(1); | |
129 | } | |
130 | root->last_trans = running_trans_id; | |
131 | } | |
132 | return 0; | |
133 | } | |
134 | ||
d352ac68 CM |
135 | /* wait for commit against the current transaction to become unblocked |
136 | * when this is done, it is safe to start a new transaction, but the current | |
137 | * transaction might not be fully on disk. | |
138 | */ | |
37d1aeee | 139 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 140 | { |
f9295749 | 141 | struct btrfs_transaction *cur_trans; |
79154b1b | 142 | |
f9295749 | 143 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 144 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
145 | DEFINE_WAIT(wait); |
146 | cur_trans->use_count++; | |
147 | while(1) { | |
148 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
149 | TASK_UNINTERRUPTIBLE); | |
150 | if (cur_trans->blocked) { | |
151 | mutex_unlock(&root->fs_info->trans_mutex); | |
152 | schedule(); | |
153 | mutex_lock(&root->fs_info->trans_mutex); | |
154 | finish_wait(&root->fs_info->transaction_wait, | |
155 | &wait); | |
156 | } else { | |
157 | finish_wait(&root->fs_info->transaction_wait, | |
158 | &wait); | |
159 | break; | |
160 | } | |
161 | } | |
162 | put_transaction(cur_trans); | |
163 | } | |
37d1aeee CM |
164 | } |
165 | ||
e02119d5 | 166 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
9ca9ee09 | 167 | int num_blocks, int wait) |
37d1aeee CM |
168 | { |
169 | struct btrfs_trans_handle *h = | |
170 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
171 | int ret; | |
172 | ||
173 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 CM |
174 | if (!root->fs_info->log_root_recovering && |
175 | ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) | |
37d1aeee | 176 | wait_current_trans(root); |
79154b1b CM |
177 | ret = join_transaction(root); |
178 | BUG_ON(ret); | |
0f7d52f4 | 179 | |
e02119d5 | 180 | btrfs_record_root_in_trans(root); |
6702ed49 | 181 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
182 | h->transaction = root->fs_info->running_transaction; |
183 | h->blocks_reserved = num_blocks; | |
184 | h->blocks_used = 0; | |
31f3c99b | 185 | h->block_group = NULL; |
26b8003f CM |
186 | h->alloc_exclude_nr = 0; |
187 | h->alloc_exclude_start = 0; | |
79154b1b CM |
188 | root->fs_info->running_transaction->use_count++; |
189 | mutex_unlock(&root->fs_info->trans_mutex); | |
190 | return h; | |
191 | } | |
192 | ||
f9295749 CM |
193 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
194 | int num_blocks) | |
195 | { | |
9ca9ee09 | 196 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
197 | } |
198 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
199 | int num_blocks) | |
200 | { | |
9ca9ee09 | 201 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
202 | } |
203 | ||
9ca9ee09 SW |
204 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
205 | int num_blocks) | |
206 | { | |
207 | return start_transaction(r, num_blocks, 2); | |
208 | } | |
209 | ||
d352ac68 | 210 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
211 | static noinline int wait_for_commit(struct btrfs_root *root, |
212 | struct btrfs_transaction *commit) | |
213 | { | |
214 | DEFINE_WAIT(wait); | |
215 | mutex_lock(&root->fs_info->trans_mutex); | |
216 | while(!commit->commit_done) { | |
217 | prepare_to_wait(&commit->commit_wait, &wait, | |
218 | TASK_UNINTERRUPTIBLE); | |
219 | if (commit->commit_done) | |
220 | break; | |
221 | mutex_unlock(&root->fs_info->trans_mutex); | |
222 | schedule(); | |
223 | mutex_lock(&root->fs_info->trans_mutex); | |
224 | } | |
225 | mutex_unlock(&root->fs_info->trans_mutex); | |
226 | finish_wait(&commit->commit_wait, &wait); | |
227 | return 0; | |
228 | } | |
229 | ||
d352ac68 CM |
230 | /* |
231 | * rate limit against the drop_snapshot code. This helps to slow down new operations | |
232 | * if the drop_snapshot code isn't able to keep up. | |
233 | */ | |
37d1aeee | 234 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
235 | { |
236 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 237 | int harder_count = 0; |
ab78c84d | 238 | |
2dd3e67b | 239 | harder: |
ab78c84d CM |
240 | if (atomic_read(&info->throttles)) { |
241 | DEFINE_WAIT(wait); | |
242 | int thr; | |
ab78c84d CM |
243 | thr = atomic_read(&info->throttle_gen); |
244 | ||
245 | do { | |
246 | prepare_to_wait(&info->transaction_throttle, | |
247 | &wait, TASK_UNINTERRUPTIBLE); | |
248 | if (!atomic_read(&info->throttles)) { | |
249 | finish_wait(&info->transaction_throttle, &wait); | |
250 | break; | |
251 | } | |
252 | schedule(); | |
253 | finish_wait(&info->transaction_throttle, &wait); | |
254 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
255 | harder_count++; |
256 | ||
257 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
258 | harder_count < 2) | |
259 | goto harder; | |
260 | ||
261 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
262 | harder_count < 10) | |
263 | goto harder; | |
264 | ||
265 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
266 | harder_count < 20) | |
267 | goto harder; | |
ab78c84d CM |
268 | } |
269 | } | |
270 | ||
37d1aeee CM |
271 | void btrfs_throttle(struct btrfs_root *root) |
272 | { | |
273 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
274 | if (!root->fs_info->open_ioctl_trans) |
275 | wait_current_trans(root); | |
37d1aeee CM |
276 | mutex_unlock(&root->fs_info->trans_mutex); |
277 | ||
278 | throttle_on_drops(root); | |
279 | } | |
280 | ||
89ce8a63 CM |
281 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
282 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
283 | { |
284 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 285 | struct btrfs_fs_info *info = root->fs_info; |
d6e4a428 | 286 | |
ab78c84d CM |
287 | mutex_lock(&info->trans_mutex); |
288 | cur_trans = info->running_transaction; | |
ccd467d6 | 289 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 290 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 291 | cur_trans->num_writers--; |
89ce8a63 | 292 | |
79154b1b CM |
293 | if (waitqueue_active(&cur_trans->writer_wait)) |
294 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 295 | put_transaction(cur_trans); |
ab78c84d | 296 | mutex_unlock(&info->trans_mutex); |
d6025579 | 297 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 298 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d CM |
299 | |
300 | if (throttle) | |
37d1aeee | 301 | throttle_on_drops(root); |
ab78c84d | 302 | |
79154b1b CM |
303 | return 0; |
304 | } | |
305 | ||
89ce8a63 CM |
306 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
307 | struct btrfs_root *root) | |
308 | { | |
309 | return __btrfs_end_transaction(trans, root, 0); | |
310 | } | |
311 | ||
312 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
313 | struct btrfs_root *root) | |
314 | { | |
315 | return __btrfs_end_transaction(trans, root, 1); | |
316 | } | |
317 | ||
d352ac68 CM |
318 | /* |
319 | * when btree blocks are allocated, they have some corresponding bits set for | |
320 | * them in one of two extent_io trees. This is used to make sure all of | |
321 | * those extents are on disk for transaction or log commit | |
322 | */ | |
d0c803c4 CM |
323 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
324 | struct extent_io_tree *dirty_pages) | |
79154b1b | 325 | { |
7c4452b9 | 326 | int ret; |
777e6bd7 | 327 | int err = 0; |
7c4452b9 CM |
328 | int werr = 0; |
329 | struct page *page; | |
7c4452b9 | 330 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 331 | u64 start = 0; |
5f39d397 CM |
332 | u64 end; |
333 | unsigned long index; | |
7c4452b9 | 334 | |
7c4452b9 | 335 | while(1) { |
777e6bd7 | 336 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
337 | EXTENT_DIRTY); |
338 | if (ret) | |
7c4452b9 | 339 | break; |
5f39d397 | 340 | while(start <= end) { |
777e6bd7 CM |
341 | cond_resched(); |
342 | ||
5f39d397 | 343 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 344 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 345 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
346 | if (!page) |
347 | continue; | |
4bef0848 CM |
348 | |
349 | btree_lock_page_hook(page); | |
350 | if (!page->mapping) { | |
351 | unlock_page(page); | |
352 | page_cache_release(page); | |
353 | continue; | |
354 | } | |
355 | ||
6702ed49 CM |
356 | if (PageWriteback(page)) { |
357 | if (PageDirty(page)) | |
358 | wait_on_page_writeback(page); | |
359 | else { | |
360 | unlock_page(page); | |
361 | page_cache_release(page); | |
362 | continue; | |
363 | } | |
364 | } | |
7c4452b9 CM |
365 | err = write_one_page(page, 0); |
366 | if (err) | |
367 | werr = err; | |
368 | page_cache_release(page); | |
369 | } | |
370 | } | |
777e6bd7 CM |
371 | while(1) { |
372 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, | |
373 | EXTENT_DIRTY); | |
374 | if (ret) | |
375 | break; | |
376 | ||
377 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
378 | while(start <= end) { | |
379 | index = start >> PAGE_CACHE_SHIFT; | |
380 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
381 | page = find_get_page(btree_inode->i_mapping, index); | |
382 | if (!page) | |
383 | continue; | |
384 | if (PageDirty(page)) { | |
4bef0848 CM |
385 | btree_lock_page_hook(page); |
386 | wait_on_page_writeback(page); | |
777e6bd7 CM |
387 | err = write_one_page(page, 0); |
388 | if (err) | |
389 | werr = err; | |
390 | } | |
105d931d | 391 | wait_on_page_writeback(page); |
777e6bd7 CM |
392 | page_cache_release(page); |
393 | cond_resched(); | |
394 | } | |
395 | } | |
7c4452b9 CM |
396 | if (err) |
397 | werr = err; | |
398 | return werr; | |
79154b1b CM |
399 | } |
400 | ||
d0c803c4 CM |
401 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
402 | struct btrfs_root *root) | |
403 | { | |
404 | if (!trans || !trans->transaction) { | |
405 | struct inode *btree_inode; | |
406 | btree_inode = root->fs_info->btree_inode; | |
407 | return filemap_write_and_wait(btree_inode->i_mapping); | |
408 | } | |
409 | return btrfs_write_and_wait_marked_extents(root, | |
410 | &trans->transaction->dirty_pages); | |
411 | } | |
412 | ||
d352ac68 CM |
413 | /* |
414 | * this is used to update the root pointer in the tree of tree roots. | |
415 | * | |
416 | * But, in the case of the extent allocation tree, updating the root | |
417 | * pointer may allocate blocks which may change the root of the extent | |
418 | * allocation tree. | |
419 | * | |
420 | * So, this loops and repeats and makes sure the cowonly root didn't | |
421 | * change while the root pointer was being updated in the metadata. | |
422 | */ | |
0b86a832 CM |
423 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
424 | struct btrfs_root *root) | |
79154b1b CM |
425 | { |
426 | int ret; | |
0b86a832 CM |
427 | u64 old_root_bytenr; |
428 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 429 | |
87ef2bb4 | 430 | btrfs_extent_post_op(trans, root); |
0b86a832 | 431 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 CM |
432 | btrfs_extent_post_op(trans, root); |
433 | ||
79154b1b | 434 | while(1) { |
0b86a832 CM |
435 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
436 | if (old_root_bytenr == root->node->start) | |
79154b1b | 437 | break; |
0b86a832 CM |
438 | btrfs_set_root_bytenr(&root->root_item, |
439 | root->node->start); | |
440 | btrfs_set_root_level(&root->root_item, | |
441 | btrfs_header_level(root->node)); | |
84234f3a | 442 | btrfs_set_root_generation(&root->root_item, trans->transid); |
87ef2bb4 CM |
443 | |
444 | btrfs_extent_post_op(trans, root); | |
445 | ||
79154b1b | 446 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
447 | &root->root_key, |
448 | &root->root_item); | |
79154b1b | 449 | BUG_ON(ret); |
0b86a832 | 450 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 | 451 | btrfs_extent_post_op(trans, root); |
0b86a832 CM |
452 | } |
453 | return 0; | |
454 | } | |
455 | ||
d352ac68 CM |
456 | /* |
457 | * update all the cowonly tree roots on disk | |
458 | */ | |
0b86a832 CM |
459 | int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans, |
460 | struct btrfs_root *root) | |
461 | { | |
462 | struct btrfs_fs_info *fs_info = root->fs_info; | |
463 | struct list_head *next; | |
84234f3a YZ |
464 | struct extent_buffer *eb; |
465 | ||
87ef2bb4 CM |
466 | btrfs_extent_post_op(trans, fs_info->tree_root); |
467 | ||
84234f3a YZ |
468 | eb = btrfs_lock_root_node(fs_info->tree_root); |
469 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0); | |
470 | btrfs_tree_unlock(eb); | |
471 | free_extent_buffer(eb); | |
0b86a832 | 472 | |
87ef2bb4 CM |
473 | btrfs_extent_post_op(trans, fs_info->tree_root); |
474 | ||
0b86a832 CM |
475 | while(!list_empty(&fs_info->dirty_cowonly_roots)) { |
476 | next = fs_info->dirty_cowonly_roots.next; | |
477 | list_del_init(next); | |
478 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 479 | |
0b86a832 | 480 | update_cowonly_root(trans, root); |
79154b1b CM |
481 | } |
482 | return 0; | |
483 | } | |
484 | ||
d352ac68 CM |
485 | /* |
486 | * dead roots are old snapshots that need to be deleted. This allocates | |
487 | * a dirty root struct and adds it into the list of dead roots that need to | |
488 | * be deleted | |
489 | */ | |
b48652c1 | 490 | int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest) |
5eda7b5e | 491 | { |
f321e491 | 492 | struct btrfs_dirty_root *dirty; |
5eda7b5e CM |
493 | |
494 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
495 | if (!dirty) | |
496 | return -ENOMEM; | |
5eda7b5e | 497 | dirty->root = root; |
5ce14bbc | 498 | dirty->latest_root = latest; |
b48652c1 YZ |
499 | |
500 | mutex_lock(&root->fs_info->trans_mutex); | |
501 | list_add(&dirty->list, &latest->fs_info->dead_roots); | |
502 | mutex_unlock(&root->fs_info->trans_mutex); | |
5eda7b5e CM |
503 | return 0; |
504 | } | |
505 | ||
d352ac68 CM |
506 | /* |
507 | * at transaction commit time we need to schedule the old roots for | |
508 | * deletion via btrfs_drop_snapshot. This runs through all the | |
509 | * reference counted roots that were modified in the current | |
510 | * transaction and puts them into the drop list | |
511 | */ | |
80b6794d CM |
512 | static noinline int add_dirty_roots(struct btrfs_trans_handle *trans, |
513 | struct radix_tree_root *radix, | |
514 | struct list_head *list) | |
0f7d52f4 | 515 | { |
f321e491 | 516 | struct btrfs_dirty_root *dirty; |
0f7d52f4 CM |
517 | struct btrfs_root *gang[8]; |
518 | struct btrfs_root *root; | |
519 | int i; | |
520 | int ret; | |
54aa1f4d | 521 | int err = 0; |
5eda7b5e | 522 | u32 refs; |
54aa1f4d | 523 | |
0f7d52f4 CM |
524 | while(1) { |
525 | ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0, | |
526 | ARRAY_SIZE(gang), | |
527 | BTRFS_ROOT_TRANS_TAG); | |
528 | if (ret == 0) | |
529 | break; | |
530 | for (i = 0; i < ret; i++) { | |
531 | root = gang[i]; | |
2619ba1f CM |
532 | radix_tree_tag_clear(radix, |
533 | (unsigned long)root->root_key.objectid, | |
534 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
535 | |
536 | BUG_ON(!root->ref_tree); | |
017e5369 | 537 | dirty = root->dirty_root; |
31153d81 | 538 | |
e02119d5 | 539 | btrfs_free_log(trans, root); |
f82d02d9 | 540 | btrfs_free_reloc_root(trans, root); |
e02119d5 | 541 | |
0f7d52f4 | 542 | if (root->commit_root == root->node) { |
db94535d CM |
543 | WARN_ON(root->node->start != |
544 | btrfs_root_bytenr(&root->root_item)); | |
31153d81 | 545 | |
5f39d397 | 546 | free_extent_buffer(root->commit_root); |
0f7d52f4 | 547 | root->commit_root = NULL; |
7ea394f1 | 548 | root->dirty_root = NULL; |
bcc63abb Y |
549 | |
550 | spin_lock(&root->list_lock); | |
551 | list_del_init(&dirty->root->dead_list); | |
552 | spin_unlock(&root->list_lock); | |
553 | ||
31153d81 YZ |
554 | kfree(dirty->root); |
555 | kfree(dirty); | |
58176a96 JB |
556 | |
557 | /* make sure to update the root on disk | |
558 | * so we get any updates to the block used | |
559 | * counts | |
560 | */ | |
561 | err = btrfs_update_root(trans, | |
562 | root->fs_info->tree_root, | |
563 | &root->root_key, | |
564 | &root->root_item); | |
0f7d52f4 CM |
565 | continue; |
566 | } | |
9f3a7427 CM |
567 | |
568 | memset(&root->root_item.drop_progress, 0, | |
569 | sizeof(struct btrfs_disk_key)); | |
570 | root->root_item.drop_level = 0; | |
0f7d52f4 | 571 | root->commit_root = NULL; |
7ea394f1 | 572 | root->dirty_root = NULL; |
0f7d52f4 | 573 | root->root_key.offset = root->fs_info->generation; |
db94535d CM |
574 | btrfs_set_root_bytenr(&root->root_item, |
575 | root->node->start); | |
576 | btrfs_set_root_level(&root->root_item, | |
577 | btrfs_header_level(root->node)); | |
84234f3a YZ |
578 | btrfs_set_root_generation(&root->root_item, |
579 | root->root_key.offset); | |
580 | ||
0f7d52f4 CM |
581 | err = btrfs_insert_root(trans, root->fs_info->tree_root, |
582 | &root->root_key, | |
583 | &root->root_item); | |
54aa1f4d CM |
584 | if (err) |
585 | break; | |
9f3a7427 CM |
586 | |
587 | refs = btrfs_root_refs(&dirty->root->root_item); | |
588 | btrfs_set_root_refs(&dirty->root->root_item, refs - 1); | |
5eda7b5e | 589 | err = btrfs_update_root(trans, root->fs_info->tree_root, |
9f3a7427 CM |
590 | &dirty->root->root_key, |
591 | &dirty->root->root_item); | |
5eda7b5e CM |
592 | |
593 | BUG_ON(err); | |
9f3a7427 | 594 | if (refs == 1) { |
5eda7b5e | 595 | list_add(&dirty->list, list); |
9f3a7427 CM |
596 | } else { |
597 | WARN_ON(1); | |
31153d81 | 598 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 599 | kfree(dirty->root); |
5eda7b5e | 600 | kfree(dirty); |
9f3a7427 | 601 | } |
0f7d52f4 CM |
602 | } |
603 | } | |
54aa1f4d | 604 | return err; |
0f7d52f4 CM |
605 | } |
606 | ||
d352ac68 CM |
607 | /* |
608 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
609 | * otherwise every leaf in the btree is read and defragged. | |
610 | */ | |
e9d0b13b CM |
611 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
612 | { | |
613 | struct btrfs_fs_info *info = root->fs_info; | |
614 | int ret; | |
615 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 616 | unsigned long nr; |
e9d0b13b | 617 | |
a2135011 | 618 | smp_mb(); |
e9d0b13b CM |
619 | if (root->defrag_running) |
620 | return 0; | |
e9d0b13b | 621 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 622 | while (1) { |
e9d0b13b CM |
623 | root->defrag_running = 1; |
624 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 625 | nr = trans->blocks_used; |
e9d0b13b | 626 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 627 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
628 | cond_resched(); |
629 | ||
e9d0b13b | 630 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 631 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
632 | break; |
633 | } | |
634 | root->defrag_running = 0; | |
a2135011 | 635 | smp_mb(); |
e9d0b13b CM |
636 | btrfs_end_transaction(trans, root); |
637 | return 0; | |
638 | } | |
639 | ||
d352ac68 CM |
640 | /* |
641 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
642 | * all of them | |
643 | */ | |
80b6794d CM |
644 | static noinline int drop_dirty_roots(struct btrfs_root *tree_root, |
645 | struct list_head *list) | |
0f7d52f4 | 646 | { |
f321e491 | 647 | struct btrfs_dirty_root *dirty; |
0f7d52f4 | 648 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 649 | unsigned long nr; |
db94535d CM |
650 | u64 num_bytes; |
651 | u64 bytes_used; | |
bcc63abb | 652 | u64 max_useless; |
54aa1f4d | 653 | int ret = 0; |
9f3a7427 CM |
654 | int err; |
655 | ||
0f7d52f4 | 656 | while(!list_empty(list)) { |
58176a96 JB |
657 | struct btrfs_root *root; |
658 | ||
f321e491 | 659 | dirty = list_entry(list->prev, struct btrfs_dirty_root, list); |
0f7d52f4 | 660 | list_del_init(&dirty->list); |
5eda7b5e | 661 | |
db94535d | 662 | num_bytes = btrfs_root_used(&dirty->root->root_item); |
58176a96 | 663 | root = dirty->latest_root; |
a2135011 | 664 | atomic_inc(&root->fs_info->throttles); |
58176a96 | 665 | |
9f3a7427 CM |
666 | while(1) { |
667 | trans = btrfs_start_transaction(tree_root, 1); | |
5b21f2ed | 668 | mutex_lock(&root->fs_info->drop_mutex); |
9f3a7427 CM |
669 | ret = btrfs_drop_snapshot(trans, dirty->root); |
670 | if (ret != -EAGAIN) { | |
671 | break; | |
672 | } | |
5b21f2ed | 673 | mutex_unlock(&root->fs_info->drop_mutex); |
58176a96 | 674 | |
9f3a7427 CM |
675 | err = btrfs_update_root(trans, |
676 | tree_root, | |
677 | &dirty->root->root_key, | |
678 | &dirty->root->root_item); | |
679 | if (err) | |
680 | ret = err; | |
d3c2fdcf | 681 | nr = trans->blocks_used; |
017e5369 | 682 | ret = btrfs_end_transaction(trans, tree_root); |
9f3a7427 | 683 | BUG_ON(ret); |
a2135011 | 684 | |
d3c2fdcf | 685 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 686 | cond_resched(); |
9f3a7427 | 687 | } |
0f7d52f4 | 688 | BUG_ON(ret); |
a2135011 | 689 | atomic_dec(&root->fs_info->throttles); |
017e5369 | 690 | wake_up(&root->fs_info->transaction_throttle); |
58176a96 | 691 | |
db94535d CM |
692 | num_bytes -= btrfs_root_used(&dirty->root->root_item); |
693 | bytes_used = btrfs_root_used(&root->root_item); | |
694 | if (num_bytes) { | |
e02119d5 | 695 | btrfs_record_root_in_trans(root); |
5f39d397 | 696 | btrfs_set_root_used(&root->root_item, |
db94535d | 697 | bytes_used - num_bytes); |
58176a96 | 698 | } |
a2135011 | 699 | |
9f3a7427 | 700 | ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key); |
58176a96 JB |
701 | if (ret) { |
702 | BUG(); | |
54aa1f4d | 703 | break; |
58176a96 | 704 | } |
a2135011 CM |
705 | mutex_unlock(&root->fs_info->drop_mutex); |
706 | ||
bcc63abb Y |
707 | spin_lock(&root->list_lock); |
708 | list_del_init(&dirty->root->dead_list); | |
709 | if (!list_empty(&root->dead_list)) { | |
710 | struct btrfs_root *oldest; | |
711 | oldest = list_entry(root->dead_list.prev, | |
712 | struct btrfs_root, dead_list); | |
713 | max_useless = oldest->root_key.offset - 1; | |
714 | } else { | |
715 | max_useless = root->root_key.offset - 1; | |
716 | } | |
717 | spin_unlock(&root->list_lock); | |
718 | ||
d3c2fdcf | 719 | nr = trans->blocks_used; |
0f7d52f4 CM |
720 | ret = btrfs_end_transaction(trans, tree_root); |
721 | BUG_ON(ret); | |
5eda7b5e | 722 | |
e4657689 | 723 | ret = btrfs_remove_leaf_refs(root, max_useless, 0); |
bcc63abb Y |
724 | BUG_ON(ret); |
725 | ||
f510cfec | 726 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 727 | kfree(dirty->root); |
0f7d52f4 | 728 | kfree(dirty); |
d3c2fdcf CM |
729 | |
730 | btrfs_btree_balance_dirty(tree_root, nr); | |
4dc11904 | 731 | cond_resched(); |
0f7d52f4 | 732 | } |
54aa1f4d | 733 | return ret; |
0f7d52f4 CM |
734 | } |
735 | ||
d352ac68 CM |
736 | /* |
737 | * new snapshots need to be created at a very specific time in the | |
738 | * transaction commit. This does the actual creation | |
739 | */ | |
80b6794d | 740 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
741 | struct btrfs_fs_info *fs_info, |
742 | struct btrfs_pending_snapshot *pending) | |
743 | { | |
744 | struct btrfs_key key; | |
80b6794d | 745 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
746 | struct btrfs_root *tree_root = fs_info->tree_root; |
747 | struct btrfs_root *root = pending->root; | |
748 | struct extent_buffer *tmp; | |
925baedd | 749 | struct extent_buffer *old; |
3063d29f CM |
750 | int ret; |
751 | u64 objectid; | |
752 | ||
80b6794d CM |
753 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
754 | if (!new_root_item) { | |
755 | ret = -ENOMEM; | |
756 | goto fail; | |
757 | } | |
3063d29f CM |
758 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
759 | if (ret) | |
760 | goto fail; | |
761 | ||
80ff3856 YZ |
762 | btrfs_record_root_in_trans(root); |
763 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
80b6794d | 764 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
765 | |
766 | key.objectid = objectid; | |
5b21f2ed | 767 | key.offset = trans->transid; |
3063d29f CM |
768 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
769 | ||
925baedd | 770 | old = btrfs_lock_root_node(root); |
65b51a00 | 771 | btrfs_cow_block(trans, root, old, NULL, 0, &old, 0); |
3063d29f | 772 | |
925baedd CM |
773 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
774 | btrfs_tree_unlock(old); | |
775 | free_extent_buffer(old); | |
3063d29f | 776 | |
80b6794d CM |
777 | btrfs_set_root_bytenr(new_root_item, tmp->start); |
778 | btrfs_set_root_level(new_root_item, btrfs_header_level(tmp)); | |
84234f3a | 779 | btrfs_set_root_generation(new_root_item, trans->transid); |
3063d29f | 780 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 781 | new_root_item); |
925baedd | 782 | btrfs_tree_unlock(tmp); |
3063d29f CM |
783 | free_extent_buffer(tmp); |
784 | if (ret) | |
785 | goto fail; | |
786 | ||
3de4586c CM |
787 | key.offset = (u64)-1; |
788 | memcpy(&pending->root_key, &key, sizeof(key)); | |
789 | fail: | |
790 | kfree(new_root_item); | |
791 | return ret; | |
792 | } | |
793 | ||
794 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
795 | struct btrfs_pending_snapshot *pending) | |
796 | { | |
797 | int ret; | |
798 | int namelen; | |
799 | u64 index = 0; | |
800 | struct btrfs_trans_handle *trans; | |
801 | struct inode *parent_inode; | |
802 | struct inode *inode; | |
0660b5af | 803 | struct btrfs_root *parent_root; |
3de4586c | 804 | |
3394e160 | 805 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af CM |
806 | parent_root = BTRFS_I(parent_inode)->root; |
807 | trans = btrfs_start_transaction(parent_root, 1); | |
3de4586c | 808 | |
3063d29f CM |
809 | /* |
810 | * insert the directory item | |
811 | */ | |
3b96362c | 812 | namelen = strlen(pending->name); |
3de4586c | 813 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 814 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
815 | pending->name, namelen, |
816 | parent_inode->i_ino, | |
817 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
818 | |
819 | if (ret) | |
820 | goto fail; | |
0660b5af CM |
821 | |
822 | /* add the backref first */ | |
823 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
824 | pending->root_key.objectid, | |
825 | BTRFS_ROOT_BACKREF_KEY, | |
826 | parent_root->root_key.objectid, | |
827 | parent_inode->i_ino, index, pending->name, | |
828 | namelen); | |
829 | ||
830 | BUG_ON(ret); | |
831 | ||
832 | /* now add the forward ref */ | |
833 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
834 | parent_root->root_key.objectid, | |
835 | BTRFS_ROOT_REF_KEY, | |
836 | pending->root_key.objectid, | |
837 | parent_inode->i_ino, index, pending->name, | |
838 | namelen); | |
839 | ||
3de4586c CM |
840 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
841 | d_instantiate(pending->dentry, inode); | |
3063d29f | 842 | fail: |
3de4586c | 843 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
844 | return ret; |
845 | } | |
846 | ||
d352ac68 CM |
847 | /* |
848 | * create all the snapshots we've scheduled for creation | |
849 | */ | |
80b6794d CM |
850 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
851 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
852 | { |
853 | struct btrfs_pending_snapshot *pending; | |
854 | struct list_head *head = &trans->transaction->pending_snapshots; | |
855 | struct list_head *cur; | |
856 | int ret; | |
857 | ||
858 | list_for_each(cur, head) { | |
859 | pending = list_entry(cur, struct btrfs_pending_snapshot, list); | |
860 | ret = create_pending_snapshot(trans, fs_info, pending); | |
861 | BUG_ON(ret); | |
862 | } | |
863 | return 0; | |
864 | } | |
865 | ||
866 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
867 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
868 | { |
869 | struct btrfs_pending_snapshot *pending; | |
870 | struct list_head *head = &trans->transaction->pending_snapshots; | |
871 | int ret; | |
872 | ||
873 | while(!list_empty(head)) { | |
874 | pending = list_entry(head->next, | |
875 | struct btrfs_pending_snapshot, list); | |
3de4586c | 876 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
877 | BUG_ON(ret); |
878 | list_del(&pending->list); | |
879 | kfree(pending->name); | |
880 | kfree(pending); | |
881 | } | |
dc17ff8f CM |
882 | return 0; |
883 | } | |
884 | ||
79154b1b CM |
885 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
886 | struct btrfs_root *root) | |
887 | { | |
15ee9bc7 JB |
888 | unsigned long joined = 0; |
889 | unsigned long timeout = 1; | |
79154b1b | 890 | struct btrfs_transaction *cur_trans; |
8fd17795 | 891 | struct btrfs_transaction *prev_trans = NULL; |
0b86a832 | 892 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; |
0f7d52f4 | 893 | struct list_head dirty_fs_roots; |
d1310b2e | 894 | struct extent_io_tree *pinned_copy; |
79154b1b | 895 | DEFINE_WAIT(wait); |
15ee9bc7 | 896 | int ret; |
79154b1b | 897 | |
0f7d52f4 | 898 | INIT_LIST_HEAD(&dirty_fs_roots); |
79154b1b CM |
899 | mutex_lock(&root->fs_info->trans_mutex); |
900 | if (trans->transaction->in_commit) { | |
901 | cur_trans = trans->transaction; | |
902 | trans->transaction->use_count++; | |
ccd467d6 | 903 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 904 | btrfs_end_transaction(trans, root); |
ccd467d6 | 905 | |
79154b1b CM |
906 | ret = wait_for_commit(root, cur_trans); |
907 | BUG_ON(ret); | |
15ee9bc7 JB |
908 | |
909 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 910 | put_transaction(cur_trans); |
15ee9bc7 JB |
911 | mutex_unlock(&root->fs_info->trans_mutex); |
912 | ||
79154b1b CM |
913 | return 0; |
914 | } | |
4313b399 CM |
915 | |
916 | pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS); | |
917 | if (!pinned_copy) | |
918 | return -ENOMEM; | |
919 | ||
d1310b2e | 920 | extent_io_tree_init(pinned_copy, |
4313b399 CM |
921 | root->fs_info->btree_inode->i_mapping, GFP_NOFS); |
922 | ||
2c90e5d6 | 923 | trans->transaction->in_commit = 1; |
f9295749 | 924 | trans->transaction->blocked = 1; |
ccd467d6 CM |
925 | cur_trans = trans->transaction; |
926 | if (cur_trans->list.prev != &root->fs_info->trans_list) { | |
927 | prev_trans = list_entry(cur_trans->list.prev, | |
928 | struct btrfs_transaction, list); | |
929 | if (!prev_trans->commit_done) { | |
930 | prev_trans->use_count++; | |
ccd467d6 CM |
931 | mutex_unlock(&root->fs_info->trans_mutex); |
932 | ||
933 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 934 | |
ccd467d6 | 935 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 936 | put_transaction(prev_trans); |
ccd467d6 CM |
937 | } |
938 | } | |
15ee9bc7 JB |
939 | |
940 | do { | |
7ea394f1 | 941 | int snap_pending = 0; |
15ee9bc7 | 942 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
943 | if (!list_empty(&trans->transaction->pending_snapshots)) |
944 | snap_pending = 1; | |
945 | ||
2c90e5d6 | 946 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 947 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 948 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
949 | |
950 | if (cur_trans->num_writers > 1) | |
951 | timeout = MAX_SCHEDULE_TIMEOUT; | |
952 | else | |
953 | timeout = 1; | |
954 | ||
79154b1b | 955 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 956 | |
7ea394f1 YZ |
957 | if (snap_pending) { |
958 | ret = btrfs_wait_ordered_extents(root, 1); | |
959 | BUG_ON(ret); | |
960 | } | |
961 | ||
15ee9bc7 JB |
962 | schedule_timeout(timeout); |
963 | ||
79154b1b | 964 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
965 | finish_wait(&cur_trans->writer_wait, &wait); |
966 | } while (cur_trans->num_writers > 1 || | |
967 | (cur_trans->num_joined != joined)); | |
968 | ||
3063d29f CM |
969 | ret = create_pending_snapshots(trans, root->fs_info); |
970 | BUG_ON(ret); | |
971 | ||
2c90e5d6 | 972 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 973 | |
e02119d5 CM |
974 | /* btrfs_commit_tree_roots is responsible for getting the |
975 | * various roots consistent with each other. Every pointer | |
976 | * in the tree of tree roots has to point to the most up to date | |
977 | * root for every subvolume and other tree. So, we have to keep | |
978 | * the tree logging code from jumping in and changing any | |
979 | * of the trees. | |
980 | * | |
981 | * At this point in the commit, there can't be any tree-log | |
982 | * writers, but a little lower down we drop the trans mutex | |
983 | * and let new people in. By holding the tree_log_mutex | |
984 | * from now until after the super is written, we avoid races | |
985 | * with the tree-log code. | |
986 | */ | |
987 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1a40e23b ZY |
988 | /* |
989 | * keep tree reloc code from adding new reloc trees | |
990 | */ | |
991 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
992 | ||
e02119d5 | 993 | |
54aa1f4d CM |
994 | ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix, |
995 | &dirty_fs_roots); | |
996 | BUG_ON(ret); | |
997 | ||
e02119d5 CM |
998 | /* add_dirty_roots gets rid of all the tree log roots, it is now |
999 | * safe to free the root of tree log roots | |
1000 | */ | |
1001 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1002 | ||
79154b1b CM |
1003 | ret = btrfs_commit_tree_roots(trans, root); |
1004 | BUG_ON(ret); | |
54aa1f4d | 1005 | |
78fae27e | 1006 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1007 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1008 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1009 | spin_unlock(&root->fs_info->new_trans_lock); |
4b52dff6 CM |
1010 | btrfs_set_super_generation(&root->fs_info->super_copy, |
1011 | cur_trans->transid); | |
1012 | btrfs_set_super_root(&root->fs_info->super_copy, | |
db94535d CM |
1013 | root->fs_info->tree_root->node->start); |
1014 | btrfs_set_super_root_level(&root->fs_info->super_copy, | |
1015 | btrfs_header_level(root->fs_info->tree_root->node)); | |
5f39d397 | 1016 | |
0b86a832 CM |
1017 | btrfs_set_super_chunk_root(&root->fs_info->super_copy, |
1018 | chunk_root->node->start); | |
1019 | btrfs_set_super_chunk_root_level(&root->fs_info->super_copy, | |
1020 | btrfs_header_level(chunk_root->node)); | |
84234f3a YZ |
1021 | btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy, |
1022 | btrfs_header_generation(chunk_root->node)); | |
e02119d5 CM |
1023 | |
1024 | if (!root->fs_info->log_root_recovering) { | |
1025 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1026 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1027 | } | |
1028 | ||
a061fc8d CM |
1029 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1030 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1031 | |
4313b399 | 1032 | btrfs_copy_pinned(root, pinned_copy); |
ccd467d6 | 1033 | |
f9295749 | 1034 | trans->transaction->blocked = 0; |
e6dcd2dc | 1035 | wake_up(&root->fs_info->transaction_throttle); |
f9295749 | 1036 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1037 | |
78fae27e | 1038 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1039 | ret = btrfs_write_and_wait_transaction(trans, root); |
1040 | BUG_ON(ret); | |
a512bbf8 | 1041 | write_ctree_super(trans, root, 0); |
4313b399 | 1042 | |
e02119d5 CM |
1043 | /* |
1044 | * the super is written, we can safely allow the tree-loggers | |
1045 | * to go about their business | |
1046 | */ | |
1047 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1048 | ||
4313b399 | 1049 | btrfs_finish_extent_commit(trans, root, pinned_copy); |
4313b399 CM |
1050 | kfree(pinned_copy); |
1051 | ||
1a40e23b ZY |
1052 | btrfs_drop_dead_reloc_roots(root); |
1053 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
1054 | ||
3de4586c CM |
1055 | /* do the directory inserts of any pending snapshot creations */ |
1056 | finish_pending_snapshots(trans, root->fs_info); | |
1057 | ||
1a40e23b ZY |
1058 | mutex_lock(&root->fs_info->trans_mutex); |
1059 | ||
2c90e5d6 | 1060 | cur_trans->commit_done = 1; |
15ee9bc7 | 1061 | root->fs_info->last_trans_committed = cur_trans->transid; |
2c90e5d6 | 1062 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1063 | |
78fae27e | 1064 | put_transaction(cur_trans); |
79154b1b | 1065 | put_transaction(cur_trans); |
58176a96 | 1066 | |
bcc63abb | 1067 | list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots); |
facda1e7 CM |
1068 | if (root->fs_info->closing) |
1069 | list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots); | |
58176a96 | 1070 | |
78fae27e | 1071 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1072 | |
2c90e5d6 | 1073 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b | 1074 | |
facda1e7 | 1075 | if (root->fs_info->closing) { |
facda1e7 | 1076 | drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots); |
facda1e7 | 1077 | } |
79154b1b CM |
1078 | return ret; |
1079 | } | |
1080 | ||
d352ac68 CM |
1081 | /* |
1082 | * interface function to delete all the snapshots we have scheduled for deletion | |
1083 | */ | |
e9d0b13b CM |
1084 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1085 | { | |
1086 | struct list_head dirty_roots; | |
1087 | INIT_LIST_HEAD(&dirty_roots); | |
a74a4b97 | 1088 | again: |
e9d0b13b CM |
1089 | mutex_lock(&root->fs_info->trans_mutex); |
1090 | list_splice_init(&root->fs_info->dead_roots, &dirty_roots); | |
1091 | mutex_unlock(&root->fs_info->trans_mutex); | |
1092 | ||
1093 | if (!list_empty(&dirty_roots)) { | |
1094 | drop_dirty_roots(root, &dirty_roots); | |
a74a4b97 | 1095 | goto again; |
e9d0b13b CM |
1096 | } |
1097 | return 0; | |
1098 | } |