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 | |
9ed74f2d JB |
189 | if (!current->journal_info) |
190 | current->journal_info = h; | |
191 | ||
79154b1b | 192 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 193 | record_root_in_trans(h, root); |
79154b1b CM |
194 | mutex_unlock(&root->fs_info->trans_mutex); |
195 | return h; | |
196 | } | |
197 | ||
f9295749 CM |
198 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
199 | int num_blocks) | |
200 | { | |
9ca9ee09 | 201 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
202 | } |
203 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
204 | int num_blocks) | |
205 | { | |
9ca9ee09 | 206 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
207 | } |
208 | ||
9ca9ee09 SW |
209 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
210 | int num_blocks) | |
211 | { | |
212 | return start_transaction(r, num_blocks, 2); | |
213 | } | |
214 | ||
d352ac68 | 215 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
216 | static noinline int wait_for_commit(struct btrfs_root *root, |
217 | struct btrfs_transaction *commit) | |
218 | { | |
219 | DEFINE_WAIT(wait); | |
220 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 221 | while (!commit->commit_done) { |
89ce8a63 CM |
222 | prepare_to_wait(&commit->commit_wait, &wait, |
223 | TASK_UNINTERRUPTIBLE); | |
224 | if (commit->commit_done) | |
225 | break; | |
226 | mutex_unlock(&root->fs_info->trans_mutex); | |
227 | schedule(); | |
228 | mutex_lock(&root->fs_info->trans_mutex); | |
229 | } | |
230 | mutex_unlock(&root->fs_info->trans_mutex); | |
231 | finish_wait(&commit->commit_wait, &wait); | |
232 | return 0; | |
233 | } | |
234 | ||
5d4f98a2 | 235 | #if 0 |
d352ac68 | 236 | /* |
d397712b CM |
237 | * rate limit against the drop_snapshot code. This helps to slow down new |
238 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 239 | */ |
37d1aeee | 240 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
241 | { |
242 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 243 | int harder_count = 0; |
ab78c84d | 244 | |
2dd3e67b | 245 | harder: |
ab78c84d CM |
246 | if (atomic_read(&info->throttles)) { |
247 | DEFINE_WAIT(wait); | |
248 | int thr; | |
ab78c84d CM |
249 | thr = atomic_read(&info->throttle_gen); |
250 | ||
251 | do { | |
252 | prepare_to_wait(&info->transaction_throttle, | |
253 | &wait, TASK_UNINTERRUPTIBLE); | |
254 | if (!atomic_read(&info->throttles)) { | |
255 | finish_wait(&info->transaction_throttle, &wait); | |
256 | break; | |
257 | } | |
258 | schedule(); | |
259 | finish_wait(&info->transaction_throttle, &wait); | |
260 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
261 | harder_count++; |
262 | ||
263 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
264 | harder_count < 2) | |
265 | goto harder; | |
266 | ||
267 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
268 | harder_count < 10) | |
269 | goto harder; | |
270 | ||
271 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
272 | harder_count < 20) | |
273 | goto harder; | |
ab78c84d CM |
274 | } |
275 | } | |
5d4f98a2 | 276 | #endif |
ab78c84d | 277 | |
37d1aeee CM |
278 | void btrfs_throttle(struct btrfs_root *root) |
279 | { | |
280 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
281 | if (!root->fs_info->open_ioctl_trans) |
282 | wait_current_trans(root); | |
37d1aeee | 283 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
284 | } |
285 | ||
89ce8a63 CM |
286 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
287 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
288 | { |
289 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 290 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
291 | int count = 0; |
292 | ||
293 | while (count < 4) { | |
294 | unsigned long cur = trans->delayed_ref_updates; | |
295 | trans->delayed_ref_updates = 0; | |
296 | if (cur && | |
297 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
298 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
299 | |
300 | /* | |
301 | * do a full flush if the transaction is trying | |
302 | * to close | |
303 | */ | |
304 | if (trans->transaction->delayed_refs.flushing) | |
305 | cur = 0; | |
c3e69d58 CM |
306 | btrfs_run_delayed_refs(trans, root, cur); |
307 | } else { | |
308 | break; | |
309 | } | |
310 | count++; | |
56bec294 CM |
311 | } |
312 | ||
ab78c84d CM |
313 | mutex_lock(&info->trans_mutex); |
314 | cur_trans = info->running_transaction; | |
ccd467d6 | 315 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 316 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 317 | cur_trans->num_writers--; |
89ce8a63 | 318 | |
79154b1b CM |
319 | if (waitqueue_active(&cur_trans->writer_wait)) |
320 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 321 | put_transaction(cur_trans); |
ab78c84d | 322 | mutex_unlock(&info->trans_mutex); |
9ed74f2d JB |
323 | |
324 | if (current->journal_info == trans) | |
325 | current->journal_info = NULL; | |
d6025579 | 326 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 327 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 328 | |
79154b1b CM |
329 | return 0; |
330 | } | |
331 | ||
89ce8a63 CM |
332 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
333 | struct btrfs_root *root) | |
334 | { | |
335 | return __btrfs_end_transaction(trans, root, 0); | |
336 | } | |
337 | ||
338 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
339 | struct btrfs_root *root) | |
340 | { | |
341 | return __btrfs_end_transaction(trans, root, 1); | |
342 | } | |
343 | ||
d352ac68 CM |
344 | /* |
345 | * when btree blocks are allocated, they have some corresponding bits set for | |
346 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 347 | * those extents are sent to disk but does not wait on them |
d352ac68 | 348 | */ |
690587d1 CM |
349 | int btrfs_write_marked_extents(struct btrfs_root *root, |
350 | struct extent_io_tree *dirty_pages) | |
79154b1b | 351 | { |
7c4452b9 | 352 | int ret; |
777e6bd7 | 353 | int err = 0; |
7c4452b9 CM |
354 | int werr = 0; |
355 | struct page *page; | |
7c4452b9 | 356 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 357 | u64 start = 0; |
5f39d397 CM |
358 | u64 end; |
359 | unsigned long index; | |
7c4452b9 | 360 | |
d397712b | 361 | while (1) { |
777e6bd7 | 362 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
363 | EXTENT_DIRTY); |
364 | if (ret) | |
7c4452b9 | 365 | break; |
d397712b | 366 | while (start <= end) { |
777e6bd7 CM |
367 | cond_resched(); |
368 | ||
5f39d397 | 369 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 370 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 371 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
372 | if (!page) |
373 | continue; | |
4bef0848 CM |
374 | |
375 | btree_lock_page_hook(page); | |
376 | if (!page->mapping) { | |
377 | unlock_page(page); | |
378 | page_cache_release(page); | |
379 | continue; | |
380 | } | |
381 | ||
6702ed49 CM |
382 | if (PageWriteback(page)) { |
383 | if (PageDirty(page)) | |
384 | wait_on_page_writeback(page); | |
385 | else { | |
386 | unlock_page(page); | |
387 | page_cache_release(page); | |
388 | continue; | |
389 | } | |
390 | } | |
7c4452b9 CM |
391 | err = write_one_page(page, 0); |
392 | if (err) | |
393 | werr = err; | |
394 | page_cache_release(page); | |
395 | } | |
396 | } | |
690587d1 CM |
397 | if (err) |
398 | werr = err; | |
399 | return werr; | |
400 | } | |
401 | ||
402 | /* | |
403 | * when btree blocks are allocated, they have some corresponding bits set for | |
404 | * them in one of two extent_io trees. This is used to make sure all of | |
405 | * those extents are on disk for transaction or log commit. We wait | |
406 | * on all the pages and clear them from the dirty pages state tree | |
407 | */ | |
408 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
409 | struct extent_io_tree *dirty_pages) | |
410 | { | |
411 | int ret; | |
412 | int err = 0; | |
413 | int werr = 0; | |
414 | struct page *page; | |
415 | struct inode *btree_inode = root->fs_info->btree_inode; | |
416 | u64 start = 0; | |
417 | u64 end; | |
418 | unsigned long index; | |
419 | ||
d397712b | 420 | while (1) { |
777e6bd7 CM |
421 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, |
422 | EXTENT_DIRTY); | |
423 | if (ret) | |
424 | break; | |
425 | ||
426 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
d397712b | 427 | while (start <= end) { |
777e6bd7 CM |
428 | index = start >> PAGE_CACHE_SHIFT; |
429 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
430 | page = find_get_page(btree_inode->i_mapping, index); | |
431 | if (!page) | |
432 | continue; | |
433 | if (PageDirty(page)) { | |
4bef0848 CM |
434 | btree_lock_page_hook(page); |
435 | wait_on_page_writeback(page); | |
777e6bd7 CM |
436 | err = write_one_page(page, 0); |
437 | if (err) | |
438 | werr = err; | |
439 | } | |
105d931d | 440 | wait_on_page_writeback(page); |
777e6bd7 CM |
441 | page_cache_release(page); |
442 | cond_resched(); | |
443 | } | |
444 | } | |
7c4452b9 CM |
445 | if (err) |
446 | werr = err; | |
447 | return werr; | |
79154b1b CM |
448 | } |
449 | ||
690587d1 CM |
450 | /* |
451 | * when btree blocks are allocated, they have some corresponding bits set for | |
452 | * them in one of two extent_io trees. This is used to make sure all of | |
453 | * those extents are on disk for transaction or log commit | |
454 | */ | |
455 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
456 | struct extent_io_tree *dirty_pages) | |
457 | { | |
458 | int ret; | |
459 | int ret2; | |
460 | ||
461 | ret = btrfs_write_marked_extents(root, dirty_pages); | |
462 | ret2 = btrfs_wait_marked_extents(root, dirty_pages); | |
463 | return ret || ret2; | |
464 | } | |
465 | ||
d0c803c4 CM |
466 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
467 | struct btrfs_root *root) | |
468 | { | |
469 | if (!trans || !trans->transaction) { | |
470 | struct inode *btree_inode; | |
471 | btree_inode = root->fs_info->btree_inode; | |
472 | return filemap_write_and_wait(btree_inode->i_mapping); | |
473 | } | |
474 | return btrfs_write_and_wait_marked_extents(root, | |
475 | &trans->transaction->dirty_pages); | |
476 | } | |
477 | ||
d352ac68 CM |
478 | /* |
479 | * this is used to update the root pointer in the tree of tree roots. | |
480 | * | |
481 | * But, in the case of the extent allocation tree, updating the root | |
482 | * pointer may allocate blocks which may change the root of the extent | |
483 | * allocation tree. | |
484 | * | |
485 | * So, this loops and repeats and makes sure the cowonly root didn't | |
486 | * change while the root pointer was being updated in the metadata. | |
487 | */ | |
0b86a832 CM |
488 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
489 | struct btrfs_root *root) | |
79154b1b CM |
490 | { |
491 | int ret; | |
0b86a832 CM |
492 | u64 old_root_bytenr; |
493 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 494 | |
0b86a832 | 495 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 496 | |
d397712b | 497 | while (1) { |
0b86a832 CM |
498 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
499 | if (old_root_bytenr == root->node->start) | |
79154b1b | 500 | break; |
87ef2bb4 | 501 | |
5d4f98a2 | 502 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 503 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
504 | &root->root_key, |
505 | &root->root_item); | |
79154b1b | 506 | BUG_ON(ret); |
56bec294 | 507 | |
4a8c9a62 | 508 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 509 | BUG_ON(ret); |
0b86a832 | 510 | } |
276e680d YZ |
511 | |
512 | if (root != root->fs_info->extent_root) | |
513 | switch_commit_root(root); | |
514 | ||
0b86a832 CM |
515 | return 0; |
516 | } | |
517 | ||
d352ac68 CM |
518 | /* |
519 | * update all the cowonly tree roots on disk | |
520 | */ | |
5d4f98a2 YZ |
521 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
522 | struct btrfs_root *root) | |
0b86a832 CM |
523 | { |
524 | struct btrfs_fs_info *fs_info = root->fs_info; | |
525 | struct list_head *next; | |
84234f3a | 526 | struct extent_buffer *eb; |
56bec294 | 527 | int ret; |
84234f3a | 528 | |
56bec294 CM |
529 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
530 | BUG_ON(ret); | |
87ef2bb4 | 531 | |
84234f3a | 532 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 533 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
534 | btrfs_tree_unlock(eb); |
535 | free_extent_buffer(eb); | |
0b86a832 | 536 | |
56bec294 CM |
537 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
538 | BUG_ON(ret); | |
87ef2bb4 | 539 | |
d397712b | 540 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
541 | next = fs_info->dirty_cowonly_roots.next; |
542 | list_del_init(next); | |
543 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 544 | |
0b86a832 | 545 | update_cowonly_root(trans, root); |
79154b1b | 546 | } |
276e680d YZ |
547 | |
548 | down_write(&fs_info->extent_commit_sem); | |
549 | switch_commit_root(fs_info->extent_root); | |
550 | up_write(&fs_info->extent_commit_sem); | |
551 | ||
79154b1b CM |
552 | return 0; |
553 | } | |
554 | ||
d352ac68 CM |
555 | /* |
556 | * dead roots are old snapshots that need to be deleted. This allocates | |
557 | * a dirty root struct and adds it into the list of dead roots that need to | |
558 | * be deleted | |
559 | */ | |
5d4f98a2 | 560 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 561 | { |
b48652c1 | 562 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 563 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 564 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
565 | return 0; |
566 | } | |
567 | ||
d352ac68 | 568 | /* |
5d4f98a2 | 569 | * update all the cowonly tree roots on disk |
d352ac68 | 570 | */ |
5d4f98a2 YZ |
571 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
572 | struct btrfs_root *root) | |
0f7d52f4 | 573 | { |
0f7d52f4 | 574 | struct btrfs_root *gang[8]; |
5d4f98a2 | 575 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
576 | int i; |
577 | int ret; | |
54aa1f4d CM |
578 | int err = 0; |
579 | ||
d397712b | 580 | while (1) { |
5d4f98a2 YZ |
581 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
582 | (void **)gang, 0, | |
0f7d52f4 CM |
583 | ARRAY_SIZE(gang), |
584 | BTRFS_ROOT_TRANS_TAG); | |
585 | if (ret == 0) | |
586 | break; | |
587 | for (i = 0; i < ret; i++) { | |
588 | root = gang[i]; | |
5d4f98a2 YZ |
589 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
590 | (unsigned long)root->root_key.objectid, | |
591 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 592 | |
e02119d5 | 593 | btrfs_free_log(trans, root); |
5d4f98a2 | 594 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 595 | |
978d910d | 596 | if (root->commit_root != root->node) { |
817d52f8 | 597 | switch_commit_root(root); |
978d910d YZ |
598 | btrfs_set_root_node(&root->root_item, |
599 | root->node); | |
600 | } | |
5d4f98a2 | 601 | |
5d4f98a2 | 602 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
603 | &root->root_key, |
604 | &root->root_item); | |
54aa1f4d CM |
605 | if (err) |
606 | break; | |
0f7d52f4 CM |
607 | } |
608 | } | |
54aa1f4d | 609 | return err; |
0f7d52f4 CM |
610 | } |
611 | ||
d352ac68 CM |
612 | /* |
613 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
614 | * otherwise every leaf in the btree is read and defragged. | |
615 | */ | |
e9d0b13b CM |
616 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
617 | { | |
618 | struct btrfs_fs_info *info = root->fs_info; | |
619 | int ret; | |
620 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 621 | unsigned long nr; |
e9d0b13b | 622 | |
a2135011 | 623 | smp_mb(); |
e9d0b13b CM |
624 | if (root->defrag_running) |
625 | return 0; | |
e9d0b13b | 626 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 627 | while (1) { |
e9d0b13b CM |
628 | root->defrag_running = 1; |
629 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 630 | nr = trans->blocks_used; |
e9d0b13b | 631 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 632 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
633 | cond_resched(); |
634 | ||
e9d0b13b | 635 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 636 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
637 | break; |
638 | } | |
639 | root->defrag_running = 0; | |
a2135011 | 640 | smp_mb(); |
e9d0b13b CM |
641 | btrfs_end_transaction(trans, root); |
642 | return 0; | |
643 | } | |
644 | ||
2c47e605 | 645 | #if 0 |
b7ec40d7 CM |
646 | /* |
647 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
648 | * sense not to join the transaction while it is trying to flush the current | |
649 | * queue of delayed refs out. | |
650 | * | |
651 | * This is used by the drop snapshot code only | |
652 | */ | |
653 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
654 | { | |
655 | DEFINE_WAIT(wait); | |
656 | ||
657 | mutex_lock(&info->trans_mutex); | |
658 | while (info->running_transaction && | |
659 | info->running_transaction->delayed_refs.flushing) { | |
660 | prepare_to_wait(&info->transaction_wait, &wait, | |
661 | TASK_UNINTERRUPTIBLE); | |
662 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 663 | |
b7ec40d7 | 664 | schedule(); |
59bc5c75 | 665 | |
b7ec40d7 CM |
666 | mutex_lock(&info->trans_mutex); |
667 | finish_wait(&info->transaction_wait, &wait); | |
668 | } | |
669 | mutex_unlock(&info->trans_mutex); | |
670 | return 0; | |
671 | } | |
672 | ||
d352ac68 CM |
673 | /* |
674 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
675 | * all of them | |
676 | */ | |
5d4f98a2 | 677 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 678 | { |
0f7d52f4 | 679 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 680 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 681 | unsigned long nr; |
5d4f98a2 | 682 | int ret; |
58176a96 | 683 | |
5d4f98a2 YZ |
684 | while (1) { |
685 | /* | |
686 | * we don't want to jump in and create a bunch of | |
687 | * delayed refs if the transaction is starting to close | |
688 | */ | |
689 | wait_transaction_pre_flush(tree_root->fs_info); | |
690 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 691 | |
5d4f98a2 YZ |
692 | /* |
693 | * we've joined a transaction, make sure it isn't | |
694 | * closing right now | |
695 | */ | |
696 | if (trans->transaction->delayed_refs.flushing) { | |
697 | btrfs_end_transaction(trans, tree_root); | |
698 | continue; | |
9f3a7427 | 699 | } |
58176a96 | 700 | |
5d4f98a2 YZ |
701 | ret = btrfs_drop_snapshot(trans, root); |
702 | if (ret != -EAGAIN) | |
703 | break; | |
a2135011 | 704 | |
5d4f98a2 YZ |
705 | ret = btrfs_update_root(trans, tree_root, |
706 | &root->root_key, | |
707 | &root->root_item); | |
708 | if (ret) | |
54aa1f4d | 709 | break; |
bcc63abb | 710 | |
d3c2fdcf | 711 | nr = trans->blocks_used; |
0f7d52f4 CM |
712 | ret = btrfs_end_transaction(trans, tree_root); |
713 | BUG_ON(ret); | |
5eda7b5e | 714 | |
d3c2fdcf | 715 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 716 | cond_resched(); |
0f7d52f4 | 717 | } |
5d4f98a2 YZ |
718 | BUG_ON(ret); |
719 | ||
720 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
721 | BUG_ON(ret); | |
722 | ||
723 | nr = trans->blocks_used; | |
724 | ret = btrfs_end_transaction(trans, tree_root); | |
725 | BUG_ON(ret); | |
726 | ||
727 | free_extent_buffer(root->node); | |
728 | free_extent_buffer(root->commit_root); | |
729 | kfree(root); | |
730 | ||
731 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 732 | return ret; |
0f7d52f4 | 733 | } |
2c47e605 | 734 | #endif |
0f7d52f4 | 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 | ||
5d4f98a2 | 762 | record_root_in_trans(trans, root); |
80ff3856 | 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; | |
1c4850e2 YZ |
767 | /* record when the snapshot was created in key.offset */ |
768 | key.offset = trans->transid; | |
3063d29f CM |
769 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
770 | ||
925baedd | 771 | old = btrfs_lock_root_node(root); |
9fa8cfe7 | 772 | btrfs_cow_block(trans, root, old, NULL, 0, &old); |
5d4f98a2 | 773 | btrfs_set_lock_blocking(old); |
3063d29f | 774 | |
925baedd CM |
775 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
776 | btrfs_tree_unlock(old); | |
777 | free_extent_buffer(old); | |
3063d29f | 778 | |
5d4f98a2 | 779 | btrfs_set_root_node(new_root_item, tmp); |
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); | |
9ed74f2d | 791 | btrfs_unreserve_metadata_space(root, 6); |
3de4586c CM |
792 | return ret; |
793 | } | |
794 | ||
795 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
796 | struct btrfs_pending_snapshot *pending) | |
797 | { | |
798 | int ret; | |
799 | int namelen; | |
800 | u64 index = 0; | |
801 | struct btrfs_trans_handle *trans; | |
802 | struct inode *parent_inode; | |
803 | struct inode *inode; | |
0660b5af | 804 | struct btrfs_root *parent_root; |
3de4586c | 805 | |
3394e160 | 806 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 807 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 808 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 809 | |
3063d29f CM |
810 | /* |
811 | * insert the directory item | |
812 | */ | |
3b96362c | 813 | namelen = strlen(pending->name); |
3de4586c | 814 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 815 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
816 | pending->name, namelen, |
817 | parent_inode->i_ino, | |
818 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
819 | |
820 | if (ret) | |
821 | goto fail; | |
0660b5af | 822 | |
52c26179 YZ |
823 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
824 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
825 | BUG_ON(ret); | |
826 | ||
0660b5af CM |
827 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, |
828 | pending->root_key.objectid, | |
0660b5af CM |
829 | parent_root->root_key.objectid, |
830 | parent_inode->i_ino, index, pending->name, | |
831 | namelen); | |
832 | ||
833 | BUG_ON(ret); | |
834 | ||
3de4586c CM |
835 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
836 | d_instantiate(pending->dentry, inode); | |
3063d29f | 837 | fail: |
3de4586c | 838 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
839 | return ret; |
840 | } | |
841 | ||
d352ac68 CM |
842 | /* |
843 | * create all the snapshots we've scheduled for creation | |
844 | */ | |
80b6794d CM |
845 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
846 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
847 | { |
848 | struct btrfs_pending_snapshot *pending; | |
849 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
850 | int ret; |
851 | ||
c6e30871 | 852 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
853 | ret = create_pending_snapshot(trans, fs_info, pending); |
854 | BUG_ON(ret); | |
855 | } | |
856 | return 0; | |
857 | } | |
858 | ||
859 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
860 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
861 | { |
862 | struct btrfs_pending_snapshot *pending; | |
863 | struct list_head *head = &trans->transaction->pending_snapshots; | |
864 | int ret; | |
865 | ||
d397712b | 866 | while (!list_empty(head)) { |
3063d29f CM |
867 | pending = list_entry(head->next, |
868 | struct btrfs_pending_snapshot, list); | |
3de4586c | 869 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
870 | BUG_ON(ret); |
871 | list_del(&pending->list); | |
872 | kfree(pending->name); | |
873 | kfree(pending); | |
874 | } | |
dc17ff8f CM |
875 | return 0; |
876 | } | |
877 | ||
5d4f98a2 YZ |
878 | static void update_super_roots(struct btrfs_root *root) |
879 | { | |
880 | struct btrfs_root_item *root_item; | |
881 | struct btrfs_super_block *super; | |
882 | ||
883 | super = &root->fs_info->super_copy; | |
884 | ||
885 | root_item = &root->fs_info->chunk_root->root_item; | |
886 | super->chunk_root = root_item->bytenr; | |
887 | super->chunk_root_generation = root_item->generation; | |
888 | super->chunk_root_level = root_item->level; | |
889 | ||
890 | root_item = &root->fs_info->tree_root->root_item; | |
891 | super->root = root_item->bytenr; | |
892 | super->generation = root_item->generation; | |
893 | super->root_level = root_item->level; | |
894 | } | |
895 | ||
f36f3042 CM |
896 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
897 | { | |
898 | int ret = 0; | |
899 | spin_lock(&info->new_trans_lock); | |
900 | if (info->running_transaction) | |
901 | ret = info->running_transaction->in_commit; | |
902 | spin_unlock(&info->new_trans_lock); | |
903 | return ret; | |
904 | } | |
905 | ||
79154b1b CM |
906 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
907 | struct btrfs_root *root) | |
908 | { | |
15ee9bc7 JB |
909 | unsigned long joined = 0; |
910 | unsigned long timeout = 1; | |
79154b1b | 911 | struct btrfs_transaction *cur_trans; |
8fd17795 | 912 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 913 | DEFINE_WAIT(wait); |
15ee9bc7 | 914 | int ret; |
89573b9c CM |
915 | int should_grow = 0; |
916 | unsigned long now = get_seconds(); | |
dccae999 | 917 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 918 | |
5a3f23d5 CM |
919 | btrfs_run_ordered_operations(root, 0); |
920 | ||
56bec294 CM |
921 | /* make a pass through all the delayed refs we have so far |
922 | * any runnings procs may add more while we are here | |
923 | */ | |
924 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
925 | BUG_ON(ret); | |
926 | ||
b7ec40d7 | 927 | cur_trans = trans->transaction; |
56bec294 CM |
928 | /* |
929 | * set the flushing flag so procs in this transaction have to | |
930 | * start sending their work down. | |
931 | */ | |
b7ec40d7 | 932 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 933 | |
c3e69d58 | 934 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
935 | BUG_ON(ret); |
936 | ||
79154b1b | 937 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
938 | if (cur_trans->in_commit) { |
939 | cur_trans->use_count++; | |
ccd467d6 | 940 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 941 | btrfs_end_transaction(trans, root); |
ccd467d6 | 942 | |
79154b1b CM |
943 | ret = wait_for_commit(root, cur_trans); |
944 | BUG_ON(ret); | |
15ee9bc7 JB |
945 | |
946 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 947 | put_transaction(cur_trans); |
15ee9bc7 JB |
948 | mutex_unlock(&root->fs_info->trans_mutex); |
949 | ||
79154b1b CM |
950 | return 0; |
951 | } | |
4313b399 | 952 | |
2c90e5d6 | 953 | trans->transaction->in_commit = 1; |
f9295749 | 954 | trans->transaction->blocked = 1; |
ccd467d6 CM |
955 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
956 | prev_trans = list_entry(cur_trans->list.prev, | |
957 | struct btrfs_transaction, list); | |
958 | if (!prev_trans->commit_done) { | |
959 | prev_trans->use_count++; | |
ccd467d6 CM |
960 | mutex_unlock(&root->fs_info->trans_mutex); |
961 | ||
962 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 963 | |
ccd467d6 | 964 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 965 | put_transaction(prev_trans); |
ccd467d6 CM |
966 | } |
967 | } | |
15ee9bc7 | 968 | |
89573b9c CM |
969 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
970 | should_grow = 1; | |
971 | ||
15ee9bc7 | 972 | do { |
7ea394f1 | 973 | int snap_pending = 0; |
15ee9bc7 | 974 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
975 | if (!list_empty(&trans->transaction->pending_snapshots)) |
976 | snap_pending = 1; | |
977 | ||
2c90e5d6 | 978 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 979 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 980 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
981 | |
982 | if (cur_trans->num_writers > 1) | |
983 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 984 | else if (should_grow) |
15ee9bc7 JB |
985 | timeout = 1; |
986 | ||
79154b1b | 987 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 988 | |
ebecd3d9 SW |
989 | if (flush_on_commit) { |
990 | btrfs_start_delalloc_inodes(root); | |
991 | ret = btrfs_wait_ordered_extents(root, 0); | |
992 | BUG_ON(ret); | |
993 | } else if (snap_pending) { | |
7ea394f1 YZ |
994 | ret = btrfs_wait_ordered_extents(root, 1); |
995 | BUG_ON(ret); | |
996 | } | |
997 | ||
5a3f23d5 CM |
998 | /* |
999 | * rename don't use btrfs_join_transaction, so, once we | |
1000 | * set the transaction to blocked above, we aren't going | |
1001 | * to get any new ordered operations. We can safely run | |
1002 | * it here and no for sure that nothing new will be added | |
1003 | * to the list | |
1004 | */ | |
1005 | btrfs_run_ordered_operations(root, 1); | |
1006 | ||
89573b9c CM |
1007 | smp_mb(); |
1008 | if (cur_trans->num_writers > 1 || should_grow) | |
1009 | schedule_timeout(timeout); | |
15ee9bc7 | 1010 | |
79154b1b | 1011 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1012 | finish_wait(&cur_trans->writer_wait, &wait); |
1013 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1014 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1015 | |
3063d29f CM |
1016 | ret = create_pending_snapshots(trans, root->fs_info); |
1017 | BUG_ON(ret); | |
1018 | ||
56bec294 CM |
1019 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1020 | BUG_ON(ret); | |
1021 | ||
2c90e5d6 | 1022 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1023 | |
e02119d5 CM |
1024 | /* btrfs_commit_tree_roots is responsible for getting the |
1025 | * various roots consistent with each other. Every pointer | |
1026 | * in the tree of tree roots has to point to the most up to date | |
1027 | * root for every subvolume and other tree. So, we have to keep | |
1028 | * the tree logging code from jumping in and changing any | |
1029 | * of the trees. | |
1030 | * | |
1031 | * At this point in the commit, there can't be any tree-log | |
1032 | * writers, but a little lower down we drop the trans mutex | |
1033 | * and let new people in. By holding the tree_log_mutex | |
1034 | * from now until after the super is written, we avoid races | |
1035 | * with the tree-log code. | |
1036 | */ | |
1037 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1038 | ||
5d4f98a2 | 1039 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1040 | BUG_ON(ret); |
1041 | ||
5d4f98a2 | 1042 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1043 | * safe to free the root of tree log roots |
1044 | */ | |
1045 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1046 | ||
5d4f98a2 | 1047 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1048 | BUG_ON(ret); |
54aa1f4d | 1049 | |
11833d66 YZ |
1050 | btrfs_prepare_extent_commit(trans, root); |
1051 | ||
78fae27e | 1052 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1053 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1054 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1055 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1056 | |
1057 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1058 | root->fs_info->tree_root->node); | |
817d52f8 | 1059 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1060 | |
1061 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1062 | root->fs_info->chunk_root->node); | |
817d52f8 | 1063 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1064 | |
1065 | update_super_roots(root); | |
e02119d5 CM |
1066 | |
1067 | if (!root->fs_info->log_root_recovering) { | |
1068 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1069 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1070 | } | |
1071 | ||
a061fc8d CM |
1072 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1073 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1074 | |
f9295749 | 1075 | trans->transaction->blocked = 0; |
b7ec40d7 | 1076 | |
f9295749 | 1077 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1078 | |
78fae27e | 1079 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1080 | ret = btrfs_write_and_wait_transaction(trans, root); |
1081 | BUG_ON(ret); | |
a512bbf8 | 1082 | write_ctree_super(trans, root, 0); |
4313b399 | 1083 | |
e02119d5 CM |
1084 | /* |
1085 | * the super is written, we can safely allow the tree-loggers | |
1086 | * to go about their business | |
1087 | */ | |
1088 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1089 | ||
11833d66 | 1090 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1091 | |
3de4586c CM |
1092 | /* do the directory inserts of any pending snapshot creations */ |
1093 | finish_pending_snapshots(trans, root->fs_info); | |
1094 | ||
1a40e23b ZY |
1095 | mutex_lock(&root->fs_info->trans_mutex); |
1096 | ||
2c90e5d6 | 1097 | cur_trans->commit_done = 1; |
b7ec40d7 | 1098 | |
15ee9bc7 | 1099 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1100 | |
2c90e5d6 | 1101 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1102 | |
78fae27e | 1103 | put_transaction(cur_trans); |
79154b1b | 1104 | put_transaction(cur_trans); |
58176a96 | 1105 | |
78fae27e | 1106 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1107 | |
9ed74f2d JB |
1108 | if (current->journal_info == trans) |
1109 | current->journal_info = NULL; | |
1110 | ||
2c90e5d6 | 1111 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b CM |
1112 | return ret; |
1113 | } | |
1114 | ||
d352ac68 CM |
1115 | /* |
1116 | * interface function to delete all the snapshots we have scheduled for deletion | |
1117 | */ | |
e9d0b13b CM |
1118 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1119 | { | |
5d4f98a2 YZ |
1120 | LIST_HEAD(list); |
1121 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1122 | ||
1123 | mutex_lock(&fs_info->trans_mutex); | |
1124 | list_splice_init(&fs_info->dead_roots, &list); | |
1125 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1126 | |
5d4f98a2 YZ |
1127 | while (!list_empty(&list)) { |
1128 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1129 | list_del(&root->root_list); |
1130 | ||
1131 | if (btrfs_header_backref_rev(root->node) < | |
1132 | BTRFS_MIXED_BACKREF_REV) | |
1133 | btrfs_drop_snapshot(root, 0); | |
1134 | else | |
1135 | btrfs_drop_snapshot(root, 1); | |
e9d0b13b CM |
1136 | } |
1137 | return 0; | |
1138 | } |