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" |
e02119d5 | 28 | #include "tree-log.h" |
79154b1b | 29 | |
0f7d52f4 CM |
30 | #define BTRFS_ROOT_TRANS_TAG 0 |
31 | ||
80b6794d | 32 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 33 | { |
2c90e5d6 | 34 | WARN_ON(transaction->use_count == 0); |
79154b1b | 35 | transaction->use_count--; |
78fae27e | 36 | if (transaction->use_count == 0) { |
8fd17795 | 37 | list_del_init(&transaction->list); |
2c90e5d6 CM |
38 | memset(transaction, 0, sizeof(*transaction)); |
39 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 40 | } |
79154b1b CM |
41 | } |
42 | ||
817d52f8 JB |
43 | static noinline void switch_commit_root(struct btrfs_root *root) |
44 | { | |
817d52f8 JB |
45 | free_extent_buffer(root->commit_root); |
46 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
47 | } |
48 | ||
d352ac68 CM |
49 | /* |
50 | * either allocate a new transaction or hop into the existing one | |
51 | */ | |
80b6794d | 52 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
53 | { |
54 | struct btrfs_transaction *cur_trans; | |
55 | cur_trans = root->fs_info->running_transaction; | |
56 | if (!cur_trans) { | |
2c90e5d6 CM |
57 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
58 | GFP_NOFS); | |
79154b1b | 59 | BUG_ON(!cur_trans); |
0f7d52f4 | 60 | root->fs_info->generation++; |
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(); |
56bec294 CM |
71 | |
72 | cur_trans->delayed_refs.root.rb_node = NULL; | |
73 | cur_trans->delayed_refs.num_entries = 0; | |
c3e69d58 CM |
74 | cur_trans->delayed_refs.num_heads_ready = 0; |
75 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 76 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 77 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
78 | spin_lock_init(&cur_trans->delayed_refs.lock); |
79 | ||
3063d29f | 80 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 81 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 82 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
83 | root->fs_info->btree_inode->i_mapping, |
84 | GFP_NOFS); | |
48ec2cf8 CM |
85 | spin_lock(&root->fs_info->new_trans_lock); |
86 | root->fs_info->running_transaction = cur_trans; | |
87 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
88 | } else { |
89 | cur_trans->num_writers++; | |
90 | cur_trans->num_joined++; | |
79154b1b | 91 | } |
15ee9bc7 | 92 | |
79154b1b CM |
93 | return 0; |
94 | } | |
95 | ||
d352ac68 | 96 | /* |
d397712b CM |
97 | * this does all the record keeping required to make sure that a reference |
98 | * counted root is properly recorded in a given transaction. This is required | |
99 | * to make sure the old root from before we joined the transaction is deleted | |
100 | * when the transaction commits | |
d352ac68 | 101 | */ |
5d4f98a2 YZ |
102 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root) | |
6702ed49 | 104 | { |
5d4f98a2 | 105 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 106 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
107 | WARN_ON(root->commit_root != root->node); |
108 | ||
109 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
110 | (unsigned long)root->root_key.objectid, | |
111 | BTRFS_ROOT_TRANS_TAG); | |
112 | root->last_trans = trans->transid; | |
113 | btrfs_init_reloc_root(trans, root); | |
114 | } | |
115 | return 0; | |
116 | } | |
bcc63abb | 117 | |
5d4f98a2 YZ |
118 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
119 | struct btrfs_root *root) | |
120 | { | |
121 | if (!root->ref_cows) | |
122 | return 0; | |
bcc63abb | 123 | |
5d4f98a2 YZ |
124 | mutex_lock(&root->fs_info->trans_mutex); |
125 | if (root->last_trans == trans->transid) { | |
126 | mutex_unlock(&root->fs_info->trans_mutex); | |
127 | return 0; | |
6702ed49 | 128 | } |
5d4f98a2 YZ |
129 | |
130 | record_root_in_trans(trans, root); | |
131 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
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++; | |
d397712b | 147 | while (1) { |
f9295749 CM |
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 | |
6702ed49 | 180 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
181 | h->transaction = root->fs_info->running_transaction; |
182 | h->blocks_reserved = num_blocks; | |
183 | h->blocks_used = 0; | |
d2fb3437 | 184 | h->block_group = 0; |
26b8003f CM |
185 | h->alloc_exclude_nr = 0; |
186 | h->alloc_exclude_start = 0; | |
56bec294 | 187 | h->delayed_ref_updates = 0; |
b7ec40d7 | 188 | |
79154b1b | 189 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 190 | record_root_in_trans(h, root); |
79154b1b CM |
191 | mutex_unlock(&root->fs_info->trans_mutex); |
192 | return h; | |
193 | } | |
194 | ||
f9295749 CM |
195 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
196 | int num_blocks) | |
197 | { | |
9ca9ee09 | 198 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
199 | } |
200 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
201 | int num_blocks) | |
202 | { | |
9ca9ee09 | 203 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
204 | } |
205 | ||
9ca9ee09 SW |
206 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
207 | int num_blocks) | |
208 | { | |
209 | return start_transaction(r, num_blocks, 2); | |
210 | } | |
211 | ||
d352ac68 | 212 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
213 | static noinline int wait_for_commit(struct btrfs_root *root, |
214 | struct btrfs_transaction *commit) | |
215 | { | |
216 | DEFINE_WAIT(wait); | |
217 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 218 | while (!commit->commit_done) { |
89ce8a63 CM |
219 | prepare_to_wait(&commit->commit_wait, &wait, |
220 | TASK_UNINTERRUPTIBLE); | |
221 | if (commit->commit_done) | |
222 | break; | |
223 | mutex_unlock(&root->fs_info->trans_mutex); | |
224 | schedule(); | |
225 | mutex_lock(&root->fs_info->trans_mutex); | |
226 | } | |
227 | mutex_unlock(&root->fs_info->trans_mutex); | |
228 | finish_wait(&commit->commit_wait, &wait); | |
229 | return 0; | |
230 | } | |
231 | ||
5d4f98a2 | 232 | #if 0 |
d352ac68 | 233 | /* |
d397712b CM |
234 | * rate limit against the drop_snapshot code. This helps to slow down new |
235 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 236 | */ |
37d1aeee | 237 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
238 | { |
239 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 240 | int harder_count = 0; |
ab78c84d | 241 | |
2dd3e67b | 242 | harder: |
ab78c84d CM |
243 | if (atomic_read(&info->throttles)) { |
244 | DEFINE_WAIT(wait); | |
245 | int thr; | |
ab78c84d CM |
246 | thr = atomic_read(&info->throttle_gen); |
247 | ||
248 | do { | |
249 | prepare_to_wait(&info->transaction_throttle, | |
250 | &wait, TASK_UNINTERRUPTIBLE); | |
251 | if (!atomic_read(&info->throttles)) { | |
252 | finish_wait(&info->transaction_throttle, &wait); | |
253 | break; | |
254 | } | |
255 | schedule(); | |
256 | finish_wait(&info->transaction_throttle, &wait); | |
257 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
258 | harder_count++; |
259 | ||
260 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
261 | harder_count < 2) | |
262 | goto harder; | |
263 | ||
264 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
265 | harder_count < 10) | |
266 | goto harder; | |
267 | ||
268 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
269 | harder_count < 20) | |
270 | goto harder; | |
ab78c84d CM |
271 | } |
272 | } | |
5d4f98a2 | 273 | #endif |
ab78c84d | 274 | |
37d1aeee CM |
275 | void btrfs_throttle(struct btrfs_root *root) |
276 | { | |
277 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
278 | if (!root->fs_info->open_ioctl_trans) |
279 | wait_current_trans(root); | |
37d1aeee | 280 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
281 | } |
282 | ||
89ce8a63 CM |
283 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
284 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
285 | { |
286 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 287 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
288 | int count = 0; |
289 | ||
290 | while (count < 4) { | |
291 | unsigned long cur = trans->delayed_ref_updates; | |
292 | trans->delayed_ref_updates = 0; | |
293 | if (cur && | |
294 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
295 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
296 | |
297 | /* | |
298 | * do a full flush if the transaction is trying | |
299 | * to close | |
300 | */ | |
301 | if (trans->transaction->delayed_refs.flushing) | |
302 | cur = 0; | |
c3e69d58 CM |
303 | btrfs_run_delayed_refs(trans, root, cur); |
304 | } else { | |
305 | break; | |
306 | } | |
307 | count++; | |
56bec294 CM |
308 | } |
309 | ||
ab78c84d CM |
310 | mutex_lock(&info->trans_mutex); |
311 | cur_trans = info->running_transaction; | |
ccd467d6 | 312 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 313 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 314 | cur_trans->num_writers--; |
89ce8a63 | 315 | |
79154b1b CM |
316 | if (waitqueue_active(&cur_trans->writer_wait)) |
317 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 318 | put_transaction(cur_trans); |
ab78c84d | 319 | mutex_unlock(&info->trans_mutex); |
d6025579 | 320 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 321 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 322 | |
79154b1b CM |
323 | return 0; |
324 | } | |
325 | ||
89ce8a63 CM |
326 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
327 | struct btrfs_root *root) | |
328 | { | |
329 | return __btrfs_end_transaction(trans, root, 0); | |
330 | } | |
331 | ||
332 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
333 | struct btrfs_root *root) | |
334 | { | |
335 | return __btrfs_end_transaction(trans, root, 1); | |
336 | } | |
337 | ||
d352ac68 CM |
338 | /* |
339 | * when btree blocks are allocated, they have some corresponding bits set for | |
340 | * them in one of two extent_io trees. This is used to make sure all of | |
341 | * those extents are on disk for transaction or log commit | |
342 | */ | |
d0c803c4 CM |
343 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
344 | struct extent_io_tree *dirty_pages) | |
79154b1b | 345 | { |
7c4452b9 | 346 | int ret; |
777e6bd7 | 347 | int err = 0; |
7c4452b9 CM |
348 | int werr = 0; |
349 | struct page *page; | |
7c4452b9 | 350 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 351 | u64 start = 0; |
5f39d397 CM |
352 | u64 end; |
353 | unsigned long index; | |
7c4452b9 | 354 | |
d397712b | 355 | while (1) { |
777e6bd7 | 356 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
357 | EXTENT_DIRTY); |
358 | if (ret) | |
7c4452b9 | 359 | break; |
d397712b | 360 | while (start <= end) { |
777e6bd7 CM |
361 | cond_resched(); |
362 | ||
5f39d397 | 363 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 364 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 365 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
366 | if (!page) |
367 | continue; | |
4bef0848 CM |
368 | |
369 | btree_lock_page_hook(page); | |
370 | if (!page->mapping) { | |
371 | unlock_page(page); | |
372 | page_cache_release(page); | |
373 | continue; | |
374 | } | |
375 | ||
6702ed49 CM |
376 | if (PageWriteback(page)) { |
377 | if (PageDirty(page)) | |
378 | wait_on_page_writeback(page); | |
379 | else { | |
380 | unlock_page(page); | |
381 | page_cache_release(page); | |
382 | continue; | |
383 | } | |
384 | } | |
7c4452b9 CM |
385 | err = write_one_page(page, 0); |
386 | if (err) | |
387 | werr = err; | |
388 | page_cache_release(page); | |
389 | } | |
390 | } | |
d397712b | 391 | while (1) { |
777e6bd7 CM |
392 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, |
393 | EXTENT_DIRTY); | |
394 | if (ret) | |
395 | break; | |
396 | ||
397 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
d397712b | 398 | while (start <= end) { |
777e6bd7 CM |
399 | index = start >> PAGE_CACHE_SHIFT; |
400 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
401 | page = find_get_page(btree_inode->i_mapping, index); | |
402 | if (!page) | |
403 | continue; | |
404 | if (PageDirty(page)) { | |
4bef0848 CM |
405 | btree_lock_page_hook(page); |
406 | wait_on_page_writeback(page); | |
777e6bd7 CM |
407 | err = write_one_page(page, 0); |
408 | if (err) | |
409 | werr = err; | |
410 | } | |
105d931d | 411 | wait_on_page_writeback(page); |
777e6bd7 CM |
412 | page_cache_release(page); |
413 | cond_resched(); | |
414 | } | |
415 | } | |
7c4452b9 CM |
416 | if (err) |
417 | werr = err; | |
418 | return werr; | |
79154b1b CM |
419 | } |
420 | ||
d0c803c4 CM |
421 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
422 | struct btrfs_root *root) | |
423 | { | |
424 | if (!trans || !trans->transaction) { | |
425 | struct inode *btree_inode; | |
426 | btree_inode = root->fs_info->btree_inode; | |
427 | return filemap_write_and_wait(btree_inode->i_mapping); | |
428 | } | |
429 | return btrfs_write_and_wait_marked_extents(root, | |
430 | &trans->transaction->dirty_pages); | |
431 | } | |
432 | ||
d352ac68 CM |
433 | /* |
434 | * this is used to update the root pointer in the tree of tree roots. | |
435 | * | |
436 | * But, in the case of the extent allocation tree, updating the root | |
437 | * pointer may allocate blocks which may change the root of the extent | |
438 | * allocation tree. | |
439 | * | |
440 | * So, this loops and repeats and makes sure the cowonly root didn't | |
441 | * change while the root pointer was being updated in the metadata. | |
442 | */ | |
0b86a832 CM |
443 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
444 | struct btrfs_root *root) | |
79154b1b CM |
445 | { |
446 | int ret; | |
0b86a832 CM |
447 | u64 old_root_bytenr; |
448 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 449 | |
0b86a832 | 450 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 451 | |
d397712b | 452 | while (1) { |
0b86a832 CM |
453 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
454 | if (old_root_bytenr == root->node->start) | |
79154b1b | 455 | break; |
87ef2bb4 | 456 | |
5d4f98a2 | 457 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 458 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
459 | &root->root_key, |
460 | &root->root_item); | |
79154b1b | 461 | BUG_ON(ret); |
56bec294 | 462 | |
4a8c9a62 | 463 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 464 | BUG_ON(ret); |
0b86a832 | 465 | } |
276e680d YZ |
466 | |
467 | if (root != root->fs_info->extent_root) | |
468 | switch_commit_root(root); | |
469 | ||
0b86a832 CM |
470 | return 0; |
471 | } | |
472 | ||
d352ac68 CM |
473 | /* |
474 | * update all the cowonly tree roots on disk | |
475 | */ | |
5d4f98a2 YZ |
476 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
477 | struct btrfs_root *root) | |
0b86a832 CM |
478 | { |
479 | struct btrfs_fs_info *fs_info = root->fs_info; | |
480 | struct list_head *next; | |
84234f3a | 481 | struct extent_buffer *eb; |
56bec294 | 482 | int ret; |
84234f3a | 483 | |
56bec294 CM |
484 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
485 | BUG_ON(ret); | |
87ef2bb4 | 486 | |
84234f3a | 487 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 488 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
489 | btrfs_tree_unlock(eb); |
490 | free_extent_buffer(eb); | |
0b86a832 | 491 | |
56bec294 CM |
492 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
493 | BUG_ON(ret); | |
87ef2bb4 | 494 | |
d397712b | 495 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
496 | next = fs_info->dirty_cowonly_roots.next; |
497 | list_del_init(next); | |
498 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 499 | |
0b86a832 | 500 | update_cowonly_root(trans, root); |
79154b1b | 501 | } |
276e680d YZ |
502 | |
503 | down_write(&fs_info->extent_commit_sem); | |
504 | switch_commit_root(fs_info->extent_root); | |
505 | up_write(&fs_info->extent_commit_sem); | |
506 | ||
79154b1b CM |
507 | return 0; |
508 | } | |
509 | ||
d352ac68 CM |
510 | /* |
511 | * dead roots are old snapshots that need to be deleted. This allocates | |
512 | * a dirty root struct and adds it into the list of dead roots that need to | |
513 | * be deleted | |
514 | */ | |
5d4f98a2 | 515 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 516 | { |
b48652c1 | 517 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 518 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 519 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
520 | return 0; |
521 | } | |
522 | ||
d352ac68 | 523 | /* |
5d4f98a2 | 524 | * update all the cowonly tree roots on disk |
d352ac68 | 525 | */ |
5d4f98a2 YZ |
526 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
527 | struct btrfs_root *root) | |
0f7d52f4 | 528 | { |
0f7d52f4 | 529 | struct btrfs_root *gang[8]; |
5d4f98a2 | 530 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
531 | int i; |
532 | int ret; | |
54aa1f4d CM |
533 | int err = 0; |
534 | ||
d397712b | 535 | while (1) { |
5d4f98a2 YZ |
536 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
537 | (void **)gang, 0, | |
0f7d52f4 CM |
538 | ARRAY_SIZE(gang), |
539 | BTRFS_ROOT_TRANS_TAG); | |
540 | if (ret == 0) | |
541 | break; | |
542 | for (i = 0; i < ret; i++) { | |
543 | root = gang[i]; | |
5d4f98a2 YZ |
544 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
545 | (unsigned long)root->root_key.objectid, | |
546 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 547 | |
e02119d5 | 548 | btrfs_free_log(trans, root); |
5d4f98a2 | 549 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 550 | |
978d910d | 551 | if (root->commit_root != root->node) { |
817d52f8 | 552 | switch_commit_root(root); |
978d910d YZ |
553 | btrfs_set_root_node(&root->root_item, |
554 | root->node); | |
555 | } | |
5d4f98a2 | 556 | |
5d4f98a2 | 557 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
558 | &root->root_key, |
559 | &root->root_item); | |
54aa1f4d CM |
560 | if (err) |
561 | break; | |
0f7d52f4 CM |
562 | } |
563 | } | |
54aa1f4d | 564 | return err; |
0f7d52f4 CM |
565 | } |
566 | ||
d352ac68 CM |
567 | /* |
568 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
569 | * otherwise every leaf in the btree is read and defragged. | |
570 | */ | |
e9d0b13b CM |
571 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
572 | { | |
573 | struct btrfs_fs_info *info = root->fs_info; | |
574 | int ret; | |
575 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 576 | unsigned long nr; |
e9d0b13b | 577 | |
a2135011 | 578 | smp_mb(); |
e9d0b13b CM |
579 | if (root->defrag_running) |
580 | return 0; | |
e9d0b13b | 581 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 582 | while (1) { |
e9d0b13b CM |
583 | root->defrag_running = 1; |
584 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 585 | nr = trans->blocks_used; |
e9d0b13b | 586 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 587 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
588 | cond_resched(); |
589 | ||
e9d0b13b | 590 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 591 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
592 | break; |
593 | } | |
594 | root->defrag_running = 0; | |
a2135011 | 595 | smp_mb(); |
e9d0b13b CM |
596 | btrfs_end_transaction(trans, root); |
597 | return 0; | |
598 | } | |
599 | ||
2c47e605 | 600 | #if 0 |
b7ec40d7 CM |
601 | /* |
602 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
603 | * sense not to join the transaction while it is trying to flush the current | |
604 | * queue of delayed refs out. | |
605 | * | |
606 | * This is used by the drop snapshot code only | |
607 | */ | |
608 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
609 | { | |
610 | DEFINE_WAIT(wait); | |
611 | ||
612 | mutex_lock(&info->trans_mutex); | |
613 | while (info->running_transaction && | |
614 | info->running_transaction->delayed_refs.flushing) { | |
615 | prepare_to_wait(&info->transaction_wait, &wait, | |
616 | TASK_UNINTERRUPTIBLE); | |
617 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 618 | |
b7ec40d7 | 619 | schedule(); |
59bc5c75 | 620 | |
b7ec40d7 CM |
621 | mutex_lock(&info->trans_mutex); |
622 | finish_wait(&info->transaction_wait, &wait); | |
623 | } | |
624 | mutex_unlock(&info->trans_mutex); | |
625 | return 0; | |
626 | } | |
627 | ||
d352ac68 CM |
628 | /* |
629 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
630 | * all of them | |
631 | */ | |
5d4f98a2 | 632 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 633 | { |
0f7d52f4 | 634 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 635 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 636 | unsigned long nr; |
5d4f98a2 | 637 | int ret; |
58176a96 | 638 | |
5d4f98a2 YZ |
639 | while (1) { |
640 | /* | |
641 | * we don't want to jump in and create a bunch of | |
642 | * delayed refs if the transaction is starting to close | |
643 | */ | |
644 | wait_transaction_pre_flush(tree_root->fs_info); | |
645 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 646 | |
5d4f98a2 YZ |
647 | /* |
648 | * we've joined a transaction, make sure it isn't | |
649 | * closing right now | |
650 | */ | |
651 | if (trans->transaction->delayed_refs.flushing) { | |
652 | btrfs_end_transaction(trans, tree_root); | |
653 | continue; | |
9f3a7427 | 654 | } |
58176a96 | 655 | |
5d4f98a2 YZ |
656 | ret = btrfs_drop_snapshot(trans, root); |
657 | if (ret != -EAGAIN) | |
658 | break; | |
a2135011 | 659 | |
5d4f98a2 YZ |
660 | ret = btrfs_update_root(trans, tree_root, |
661 | &root->root_key, | |
662 | &root->root_item); | |
663 | if (ret) | |
54aa1f4d | 664 | break; |
bcc63abb | 665 | |
d3c2fdcf | 666 | nr = trans->blocks_used; |
0f7d52f4 CM |
667 | ret = btrfs_end_transaction(trans, tree_root); |
668 | BUG_ON(ret); | |
5eda7b5e | 669 | |
d3c2fdcf | 670 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 671 | cond_resched(); |
0f7d52f4 | 672 | } |
5d4f98a2 YZ |
673 | BUG_ON(ret); |
674 | ||
675 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
676 | BUG_ON(ret); | |
677 | ||
678 | nr = trans->blocks_used; | |
679 | ret = btrfs_end_transaction(trans, tree_root); | |
680 | BUG_ON(ret); | |
681 | ||
682 | free_extent_buffer(root->node); | |
683 | free_extent_buffer(root->commit_root); | |
684 | kfree(root); | |
685 | ||
686 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 687 | return ret; |
0f7d52f4 | 688 | } |
2c47e605 | 689 | #endif |
0f7d52f4 | 690 | |
d352ac68 CM |
691 | /* |
692 | * new snapshots need to be created at a very specific time in the | |
693 | * transaction commit. This does the actual creation | |
694 | */ | |
80b6794d | 695 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
696 | struct btrfs_fs_info *fs_info, |
697 | struct btrfs_pending_snapshot *pending) | |
698 | { | |
699 | struct btrfs_key key; | |
80b6794d | 700 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
701 | struct btrfs_root *tree_root = fs_info->tree_root; |
702 | struct btrfs_root *root = pending->root; | |
703 | struct extent_buffer *tmp; | |
925baedd | 704 | struct extent_buffer *old; |
3063d29f CM |
705 | int ret; |
706 | u64 objectid; | |
707 | ||
80b6794d CM |
708 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
709 | if (!new_root_item) { | |
710 | ret = -ENOMEM; | |
711 | goto fail; | |
712 | } | |
3063d29f CM |
713 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
714 | if (ret) | |
715 | goto fail; | |
716 | ||
5d4f98a2 | 717 | record_root_in_trans(trans, root); |
80ff3856 | 718 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
80b6794d | 719 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
720 | |
721 | key.objectid = objectid; | |
1c4850e2 YZ |
722 | /* record when the snapshot was created in key.offset */ |
723 | key.offset = trans->transid; | |
3063d29f CM |
724 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
725 | ||
925baedd | 726 | old = btrfs_lock_root_node(root); |
9fa8cfe7 | 727 | btrfs_cow_block(trans, root, old, NULL, 0, &old); |
5d4f98a2 | 728 | btrfs_set_lock_blocking(old); |
3063d29f | 729 | |
925baedd CM |
730 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
731 | btrfs_tree_unlock(old); | |
732 | free_extent_buffer(old); | |
3063d29f | 733 | |
5d4f98a2 | 734 | btrfs_set_root_node(new_root_item, tmp); |
3063d29f | 735 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 736 | new_root_item); |
925baedd | 737 | btrfs_tree_unlock(tmp); |
3063d29f CM |
738 | free_extent_buffer(tmp); |
739 | if (ret) | |
740 | goto fail; | |
741 | ||
3de4586c CM |
742 | key.offset = (u64)-1; |
743 | memcpy(&pending->root_key, &key, sizeof(key)); | |
744 | fail: | |
745 | kfree(new_root_item); | |
746 | return ret; | |
747 | } | |
748 | ||
749 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
750 | struct btrfs_pending_snapshot *pending) | |
751 | { | |
752 | int ret; | |
753 | int namelen; | |
754 | u64 index = 0; | |
755 | struct btrfs_trans_handle *trans; | |
756 | struct inode *parent_inode; | |
757 | struct inode *inode; | |
0660b5af | 758 | struct btrfs_root *parent_root; |
3de4586c | 759 | |
3394e160 | 760 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 761 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 762 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 763 | |
3063d29f CM |
764 | /* |
765 | * insert the directory item | |
766 | */ | |
3b96362c | 767 | namelen = strlen(pending->name); |
3de4586c | 768 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 769 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
770 | pending->name, namelen, |
771 | parent_inode->i_ino, | |
772 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
773 | |
774 | if (ret) | |
775 | goto fail; | |
0660b5af | 776 | |
52c26179 YZ |
777 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
778 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
779 | BUG_ON(ret); | |
780 | ||
0660b5af CM |
781 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, |
782 | pending->root_key.objectid, | |
0660b5af CM |
783 | parent_root->root_key.objectid, |
784 | parent_inode->i_ino, index, pending->name, | |
785 | namelen); | |
786 | ||
787 | BUG_ON(ret); | |
788 | ||
3de4586c CM |
789 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
790 | d_instantiate(pending->dentry, inode); | |
3063d29f | 791 | fail: |
3de4586c | 792 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
793 | return ret; |
794 | } | |
795 | ||
d352ac68 CM |
796 | /* |
797 | * create all the snapshots we've scheduled for creation | |
798 | */ | |
80b6794d CM |
799 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
800 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
801 | { |
802 | struct btrfs_pending_snapshot *pending; | |
803 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
804 | int ret; |
805 | ||
c6e30871 | 806 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
807 | ret = create_pending_snapshot(trans, fs_info, pending); |
808 | BUG_ON(ret); | |
809 | } | |
810 | return 0; | |
811 | } | |
812 | ||
813 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
814 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
815 | { |
816 | struct btrfs_pending_snapshot *pending; | |
817 | struct list_head *head = &trans->transaction->pending_snapshots; | |
818 | int ret; | |
819 | ||
d397712b | 820 | while (!list_empty(head)) { |
3063d29f CM |
821 | pending = list_entry(head->next, |
822 | struct btrfs_pending_snapshot, list); | |
3de4586c | 823 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
824 | BUG_ON(ret); |
825 | list_del(&pending->list); | |
826 | kfree(pending->name); | |
827 | kfree(pending); | |
828 | } | |
dc17ff8f CM |
829 | return 0; |
830 | } | |
831 | ||
5d4f98a2 YZ |
832 | static void update_super_roots(struct btrfs_root *root) |
833 | { | |
834 | struct btrfs_root_item *root_item; | |
835 | struct btrfs_super_block *super; | |
836 | ||
837 | super = &root->fs_info->super_copy; | |
838 | ||
839 | root_item = &root->fs_info->chunk_root->root_item; | |
840 | super->chunk_root = root_item->bytenr; | |
841 | super->chunk_root_generation = root_item->generation; | |
842 | super->chunk_root_level = root_item->level; | |
843 | ||
844 | root_item = &root->fs_info->tree_root->root_item; | |
845 | super->root = root_item->bytenr; | |
846 | super->generation = root_item->generation; | |
847 | super->root_level = root_item->level; | |
848 | } | |
849 | ||
f36f3042 CM |
850 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
851 | { | |
852 | int ret = 0; | |
853 | spin_lock(&info->new_trans_lock); | |
854 | if (info->running_transaction) | |
855 | ret = info->running_transaction->in_commit; | |
856 | spin_unlock(&info->new_trans_lock); | |
857 | return ret; | |
858 | } | |
859 | ||
79154b1b CM |
860 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
861 | struct btrfs_root *root) | |
862 | { | |
15ee9bc7 JB |
863 | unsigned long joined = 0; |
864 | unsigned long timeout = 1; | |
79154b1b | 865 | struct btrfs_transaction *cur_trans; |
8fd17795 | 866 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 867 | DEFINE_WAIT(wait); |
15ee9bc7 | 868 | int ret; |
89573b9c CM |
869 | int should_grow = 0; |
870 | unsigned long now = get_seconds(); | |
dccae999 | 871 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 872 | |
5a3f23d5 CM |
873 | btrfs_run_ordered_operations(root, 0); |
874 | ||
56bec294 CM |
875 | /* make a pass through all the delayed refs we have so far |
876 | * any runnings procs may add more while we are here | |
877 | */ | |
878 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
879 | BUG_ON(ret); | |
880 | ||
b7ec40d7 | 881 | cur_trans = trans->transaction; |
56bec294 CM |
882 | /* |
883 | * set the flushing flag so procs in this transaction have to | |
884 | * start sending their work down. | |
885 | */ | |
b7ec40d7 | 886 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 887 | |
c3e69d58 | 888 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
889 | BUG_ON(ret); |
890 | ||
79154b1b | 891 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
892 | if (cur_trans->in_commit) { |
893 | cur_trans->use_count++; | |
ccd467d6 | 894 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 895 | btrfs_end_transaction(trans, root); |
ccd467d6 | 896 | |
79154b1b CM |
897 | ret = wait_for_commit(root, cur_trans); |
898 | BUG_ON(ret); | |
15ee9bc7 JB |
899 | |
900 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 901 | put_transaction(cur_trans); |
15ee9bc7 JB |
902 | mutex_unlock(&root->fs_info->trans_mutex); |
903 | ||
79154b1b CM |
904 | return 0; |
905 | } | |
4313b399 | 906 | |
2c90e5d6 | 907 | trans->transaction->in_commit = 1; |
f9295749 | 908 | trans->transaction->blocked = 1; |
ccd467d6 CM |
909 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
910 | prev_trans = list_entry(cur_trans->list.prev, | |
911 | struct btrfs_transaction, list); | |
912 | if (!prev_trans->commit_done) { | |
913 | prev_trans->use_count++; | |
ccd467d6 CM |
914 | mutex_unlock(&root->fs_info->trans_mutex); |
915 | ||
916 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 917 | |
ccd467d6 | 918 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 919 | put_transaction(prev_trans); |
ccd467d6 CM |
920 | } |
921 | } | |
15ee9bc7 | 922 | |
89573b9c CM |
923 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
924 | should_grow = 1; | |
925 | ||
15ee9bc7 | 926 | do { |
7ea394f1 | 927 | int snap_pending = 0; |
15ee9bc7 | 928 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
929 | if (!list_empty(&trans->transaction->pending_snapshots)) |
930 | snap_pending = 1; | |
931 | ||
2c90e5d6 | 932 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 933 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 934 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
935 | |
936 | if (cur_trans->num_writers > 1) | |
937 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 938 | else if (should_grow) |
15ee9bc7 JB |
939 | timeout = 1; |
940 | ||
79154b1b | 941 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 942 | |
ebecd3d9 SW |
943 | if (flush_on_commit) { |
944 | btrfs_start_delalloc_inodes(root); | |
945 | ret = btrfs_wait_ordered_extents(root, 0); | |
946 | BUG_ON(ret); | |
947 | } else if (snap_pending) { | |
7ea394f1 YZ |
948 | ret = btrfs_wait_ordered_extents(root, 1); |
949 | BUG_ON(ret); | |
950 | } | |
951 | ||
5a3f23d5 CM |
952 | /* |
953 | * rename don't use btrfs_join_transaction, so, once we | |
954 | * set the transaction to blocked above, we aren't going | |
955 | * to get any new ordered operations. We can safely run | |
956 | * it here and no for sure that nothing new will be added | |
957 | * to the list | |
958 | */ | |
959 | btrfs_run_ordered_operations(root, 1); | |
960 | ||
89573b9c CM |
961 | smp_mb(); |
962 | if (cur_trans->num_writers > 1 || should_grow) | |
963 | schedule_timeout(timeout); | |
15ee9bc7 | 964 | |
79154b1b | 965 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
966 | finish_wait(&cur_trans->writer_wait, &wait); |
967 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 968 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 969 | |
3063d29f CM |
970 | ret = create_pending_snapshots(trans, root->fs_info); |
971 | BUG_ON(ret); | |
972 | ||
56bec294 CM |
973 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
974 | BUG_ON(ret); | |
975 | ||
2c90e5d6 | 976 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 977 | |
e02119d5 CM |
978 | /* btrfs_commit_tree_roots is responsible for getting the |
979 | * various roots consistent with each other. Every pointer | |
980 | * in the tree of tree roots has to point to the most up to date | |
981 | * root for every subvolume and other tree. So, we have to keep | |
982 | * the tree logging code from jumping in and changing any | |
983 | * of the trees. | |
984 | * | |
985 | * At this point in the commit, there can't be any tree-log | |
986 | * writers, but a little lower down we drop the trans mutex | |
987 | * and let new people in. By holding the tree_log_mutex | |
988 | * from now until after the super is written, we avoid races | |
989 | * with the tree-log code. | |
990 | */ | |
991 | mutex_lock(&root->fs_info->tree_log_mutex); | |
992 | ||
5d4f98a2 | 993 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
994 | BUG_ON(ret); |
995 | ||
5d4f98a2 | 996 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
997 | * safe to free the root of tree log roots |
998 | */ | |
999 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1000 | ||
5d4f98a2 | 1001 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1002 | BUG_ON(ret); |
54aa1f4d | 1003 | |
11833d66 YZ |
1004 | btrfs_prepare_extent_commit(trans, root); |
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); |
5d4f98a2 YZ |
1010 | |
1011 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1012 | root->fs_info->tree_root->node); | |
817d52f8 | 1013 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1014 | |
1015 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1016 | root->fs_info->chunk_root->node); | |
817d52f8 | 1017 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1018 | |
1019 | update_super_roots(root); | |
e02119d5 CM |
1020 | |
1021 | if (!root->fs_info->log_root_recovering) { | |
1022 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1023 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1024 | } | |
1025 | ||
a061fc8d CM |
1026 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1027 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1028 | |
f9295749 | 1029 | trans->transaction->blocked = 0; |
b7ec40d7 | 1030 | |
f9295749 | 1031 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1032 | |
78fae27e | 1033 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1034 | ret = btrfs_write_and_wait_transaction(trans, root); |
1035 | BUG_ON(ret); | |
a512bbf8 | 1036 | write_ctree_super(trans, root, 0); |
4313b399 | 1037 | |
e02119d5 CM |
1038 | /* |
1039 | * the super is written, we can safely allow the tree-loggers | |
1040 | * to go about their business | |
1041 | */ | |
1042 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1043 | ||
11833d66 | 1044 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1045 | |
3de4586c CM |
1046 | /* do the directory inserts of any pending snapshot creations */ |
1047 | finish_pending_snapshots(trans, root->fs_info); | |
1048 | ||
1a40e23b ZY |
1049 | mutex_lock(&root->fs_info->trans_mutex); |
1050 | ||
2c90e5d6 | 1051 | cur_trans->commit_done = 1; |
b7ec40d7 | 1052 | |
15ee9bc7 | 1053 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1054 | |
2c90e5d6 | 1055 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1056 | |
78fae27e | 1057 | put_transaction(cur_trans); |
79154b1b | 1058 | put_transaction(cur_trans); |
58176a96 | 1059 | |
78fae27e | 1060 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1061 | |
2c90e5d6 | 1062 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b CM |
1063 | return ret; |
1064 | } | |
1065 | ||
d352ac68 CM |
1066 | /* |
1067 | * interface function to delete all the snapshots we have scheduled for deletion | |
1068 | */ | |
e9d0b13b CM |
1069 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1070 | { | |
5d4f98a2 YZ |
1071 | LIST_HEAD(list); |
1072 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1073 | ||
1074 | mutex_lock(&fs_info->trans_mutex); | |
1075 | list_splice_init(&fs_info->dead_roots, &list); | |
1076 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1077 | |
5d4f98a2 YZ |
1078 | while (!list_empty(&list)) { |
1079 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1080 | list_del(&root->root_list); |
1081 | ||
1082 | if (btrfs_header_backref_rev(root->node) < | |
1083 | BTRFS_MIXED_BACKREF_REV) | |
1084 | btrfs_drop_snapshot(root, 0); | |
1085 | else | |
1086 | btrfs_drop_snapshot(root, 1); | |
e9d0b13b CM |
1087 | } |
1088 | return 0; | |
1089 | } |