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->root_item.refs == 0); |
108 | WARN_ON(root->commit_root != root->node); | |
109 | ||
110 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
111 | (unsigned long)root->root_key.objectid, | |
112 | BTRFS_ROOT_TRANS_TAG); | |
113 | root->last_trans = trans->transid; | |
114 | btrfs_init_reloc_root(trans, root); | |
115 | } | |
116 | return 0; | |
117 | } | |
bcc63abb | 118 | |
5d4f98a2 YZ |
119 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
120 | struct btrfs_root *root) | |
121 | { | |
122 | if (!root->ref_cows) | |
123 | return 0; | |
bcc63abb | 124 | |
5d4f98a2 YZ |
125 | mutex_lock(&root->fs_info->trans_mutex); |
126 | if (root->last_trans == trans->transid) { | |
127 | mutex_unlock(&root->fs_info->trans_mutex); | |
128 | return 0; | |
6702ed49 | 129 | } |
5d4f98a2 YZ |
130 | |
131 | record_root_in_trans(trans, root); | |
132 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
133 | return 0; |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* wait for commit against the current transaction to become unblocked |
137 | * when this is done, it is safe to start a new transaction, but the current | |
138 | * transaction might not be fully on disk. | |
139 | */ | |
37d1aeee | 140 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 141 | { |
f9295749 | 142 | struct btrfs_transaction *cur_trans; |
79154b1b | 143 | |
f9295749 | 144 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 145 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
146 | DEFINE_WAIT(wait); |
147 | cur_trans->use_count++; | |
d397712b | 148 | while (1) { |
f9295749 CM |
149 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
150 | TASK_UNINTERRUPTIBLE); | |
151 | if (cur_trans->blocked) { | |
152 | mutex_unlock(&root->fs_info->trans_mutex); | |
153 | schedule(); | |
154 | mutex_lock(&root->fs_info->trans_mutex); | |
155 | finish_wait(&root->fs_info->transaction_wait, | |
156 | &wait); | |
157 | } else { | |
158 | finish_wait(&root->fs_info->transaction_wait, | |
159 | &wait); | |
160 | break; | |
161 | } | |
162 | } | |
163 | put_transaction(cur_trans); | |
164 | } | |
37d1aeee CM |
165 | } |
166 | ||
e02119d5 | 167 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
9ca9ee09 | 168 | int num_blocks, int wait) |
37d1aeee CM |
169 | { |
170 | struct btrfs_trans_handle *h = | |
171 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
172 | int ret; | |
173 | ||
174 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 CM |
175 | if (!root->fs_info->log_root_recovering && |
176 | ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) | |
37d1aeee | 177 | wait_current_trans(root); |
79154b1b CM |
178 | ret = join_transaction(root); |
179 | BUG_ON(ret); | |
0f7d52f4 | 180 | |
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; | |
d2fb3437 | 185 | h->block_group = 0; |
26b8003f CM |
186 | h->alloc_exclude_nr = 0; |
187 | h->alloc_exclude_start = 0; | |
56bec294 | 188 | h->delayed_ref_updates = 0; |
b7ec40d7 | 189 | |
79154b1b | 190 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 191 | record_root_in_trans(h, root); |
79154b1b CM |
192 | mutex_unlock(&root->fs_info->trans_mutex); |
193 | return h; | |
194 | } | |
195 | ||
f9295749 CM |
196 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
197 | int num_blocks) | |
198 | { | |
9ca9ee09 | 199 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
200 | } |
201 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
202 | int num_blocks) | |
203 | { | |
9ca9ee09 | 204 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
205 | } |
206 | ||
9ca9ee09 SW |
207 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
208 | int num_blocks) | |
209 | { | |
210 | return start_transaction(r, num_blocks, 2); | |
211 | } | |
212 | ||
d352ac68 | 213 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
214 | static noinline int wait_for_commit(struct btrfs_root *root, |
215 | struct btrfs_transaction *commit) | |
216 | { | |
217 | DEFINE_WAIT(wait); | |
218 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 219 | while (!commit->commit_done) { |
89ce8a63 CM |
220 | prepare_to_wait(&commit->commit_wait, &wait, |
221 | TASK_UNINTERRUPTIBLE); | |
222 | if (commit->commit_done) | |
223 | break; | |
224 | mutex_unlock(&root->fs_info->trans_mutex); | |
225 | schedule(); | |
226 | mutex_lock(&root->fs_info->trans_mutex); | |
227 | } | |
228 | mutex_unlock(&root->fs_info->trans_mutex); | |
229 | finish_wait(&commit->commit_wait, &wait); | |
230 | return 0; | |
231 | } | |
232 | ||
5d4f98a2 | 233 | #if 0 |
d352ac68 | 234 | /* |
d397712b CM |
235 | * rate limit against the drop_snapshot code. This helps to slow down new |
236 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 237 | */ |
37d1aeee | 238 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
239 | { |
240 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 241 | int harder_count = 0; |
ab78c84d | 242 | |
2dd3e67b | 243 | harder: |
ab78c84d CM |
244 | if (atomic_read(&info->throttles)) { |
245 | DEFINE_WAIT(wait); | |
246 | int thr; | |
ab78c84d CM |
247 | thr = atomic_read(&info->throttle_gen); |
248 | ||
249 | do { | |
250 | prepare_to_wait(&info->transaction_throttle, | |
251 | &wait, TASK_UNINTERRUPTIBLE); | |
252 | if (!atomic_read(&info->throttles)) { | |
253 | finish_wait(&info->transaction_throttle, &wait); | |
254 | break; | |
255 | } | |
256 | schedule(); | |
257 | finish_wait(&info->transaction_throttle, &wait); | |
258 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
259 | harder_count++; |
260 | ||
261 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
262 | harder_count < 2) | |
263 | goto harder; | |
264 | ||
265 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
266 | harder_count < 10) | |
267 | goto harder; | |
268 | ||
269 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
270 | harder_count < 20) | |
271 | goto harder; | |
ab78c84d CM |
272 | } |
273 | } | |
5d4f98a2 | 274 | #endif |
ab78c84d | 275 | |
37d1aeee CM |
276 | void btrfs_throttle(struct btrfs_root *root) |
277 | { | |
278 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
279 | if (!root->fs_info->open_ioctl_trans) |
280 | wait_current_trans(root); | |
37d1aeee | 281 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
282 | } |
283 | ||
89ce8a63 CM |
284 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
285 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
286 | { |
287 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 288 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
289 | int count = 0; |
290 | ||
291 | while (count < 4) { | |
292 | unsigned long cur = trans->delayed_ref_updates; | |
293 | trans->delayed_ref_updates = 0; | |
294 | if (cur && | |
295 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
296 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
297 | |
298 | /* | |
299 | * do a full flush if the transaction is trying | |
300 | * to close | |
301 | */ | |
302 | if (trans->transaction->delayed_refs.flushing) | |
303 | cur = 0; | |
c3e69d58 CM |
304 | btrfs_run_delayed_refs(trans, root, cur); |
305 | } else { | |
306 | break; | |
307 | } | |
308 | count++; | |
56bec294 CM |
309 | } |
310 | ||
ab78c84d CM |
311 | mutex_lock(&info->trans_mutex); |
312 | cur_trans = info->running_transaction; | |
ccd467d6 | 313 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 314 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 315 | cur_trans->num_writers--; |
89ce8a63 | 316 | |
79154b1b CM |
317 | if (waitqueue_active(&cur_trans->writer_wait)) |
318 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 319 | put_transaction(cur_trans); |
ab78c84d | 320 | mutex_unlock(&info->trans_mutex); |
d6025579 | 321 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 322 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 323 | |
79154b1b CM |
324 | return 0; |
325 | } | |
326 | ||
89ce8a63 CM |
327 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
328 | struct btrfs_root *root) | |
329 | { | |
330 | return __btrfs_end_transaction(trans, root, 0); | |
331 | } | |
332 | ||
333 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
334 | struct btrfs_root *root) | |
335 | { | |
336 | return __btrfs_end_transaction(trans, root, 1); | |
337 | } | |
338 | ||
d352ac68 CM |
339 | /* |
340 | * when btree blocks are allocated, they have some corresponding bits set for | |
341 | * them in one of two extent_io trees. This is used to make sure all of | |
342 | * those extents are on disk for transaction or log commit | |
343 | */ | |
d0c803c4 CM |
344 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
345 | struct extent_io_tree *dirty_pages) | |
79154b1b | 346 | { |
7c4452b9 | 347 | int ret; |
777e6bd7 | 348 | int err = 0; |
7c4452b9 CM |
349 | int werr = 0; |
350 | struct page *page; | |
7c4452b9 | 351 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 352 | u64 start = 0; |
5f39d397 CM |
353 | u64 end; |
354 | unsigned long index; | |
7c4452b9 | 355 | |
d397712b | 356 | while (1) { |
777e6bd7 | 357 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
358 | EXTENT_DIRTY); |
359 | if (ret) | |
7c4452b9 | 360 | break; |
d397712b | 361 | while (start <= end) { |
777e6bd7 CM |
362 | cond_resched(); |
363 | ||
5f39d397 | 364 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 365 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 366 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
367 | if (!page) |
368 | continue; | |
4bef0848 CM |
369 | |
370 | btree_lock_page_hook(page); | |
371 | if (!page->mapping) { | |
372 | unlock_page(page); | |
373 | page_cache_release(page); | |
374 | continue; | |
375 | } | |
376 | ||
6702ed49 CM |
377 | if (PageWriteback(page)) { |
378 | if (PageDirty(page)) | |
379 | wait_on_page_writeback(page); | |
380 | else { | |
381 | unlock_page(page); | |
382 | page_cache_release(page); | |
383 | continue; | |
384 | } | |
385 | } | |
7c4452b9 CM |
386 | err = write_one_page(page, 0); |
387 | if (err) | |
388 | werr = err; | |
389 | page_cache_release(page); | |
390 | } | |
391 | } | |
d397712b | 392 | while (1) { |
777e6bd7 CM |
393 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, |
394 | EXTENT_DIRTY); | |
395 | if (ret) | |
396 | break; | |
397 | ||
398 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
d397712b | 399 | while (start <= end) { |
777e6bd7 CM |
400 | index = start >> PAGE_CACHE_SHIFT; |
401 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
402 | page = find_get_page(btree_inode->i_mapping, index); | |
403 | if (!page) | |
404 | continue; | |
405 | if (PageDirty(page)) { | |
4bef0848 CM |
406 | btree_lock_page_hook(page); |
407 | wait_on_page_writeback(page); | |
777e6bd7 CM |
408 | err = write_one_page(page, 0); |
409 | if (err) | |
410 | werr = err; | |
411 | } | |
105d931d | 412 | wait_on_page_writeback(page); |
777e6bd7 CM |
413 | page_cache_release(page); |
414 | cond_resched(); | |
415 | } | |
416 | } | |
7c4452b9 CM |
417 | if (err) |
418 | werr = err; | |
419 | return werr; | |
79154b1b CM |
420 | } |
421 | ||
d0c803c4 CM |
422 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
423 | struct btrfs_root *root) | |
424 | { | |
425 | if (!trans || !trans->transaction) { | |
426 | struct inode *btree_inode; | |
427 | btree_inode = root->fs_info->btree_inode; | |
428 | return filemap_write_and_wait(btree_inode->i_mapping); | |
429 | } | |
430 | return btrfs_write_and_wait_marked_extents(root, | |
431 | &trans->transaction->dirty_pages); | |
432 | } | |
433 | ||
d352ac68 CM |
434 | /* |
435 | * this is used to update the root pointer in the tree of tree roots. | |
436 | * | |
437 | * But, in the case of the extent allocation tree, updating the root | |
438 | * pointer may allocate blocks which may change the root of the extent | |
439 | * allocation tree. | |
440 | * | |
441 | * So, this loops and repeats and makes sure the cowonly root didn't | |
442 | * change while the root pointer was being updated in the metadata. | |
443 | */ | |
0b86a832 CM |
444 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
445 | struct btrfs_root *root) | |
79154b1b CM |
446 | { |
447 | int ret; | |
0b86a832 CM |
448 | u64 old_root_bytenr; |
449 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 450 | |
0b86a832 | 451 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 452 | |
d397712b | 453 | while (1) { |
0b86a832 CM |
454 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
455 | if (old_root_bytenr == root->node->start) | |
79154b1b | 456 | break; |
87ef2bb4 | 457 | |
5d4f98a2 | 458 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 459 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
460 | &root->root_key, |
461 | &root->root_item); | |
79154b1b | 462 | BUG_ON(ret); |
56bec294 | 463 | |
4a8c9a62 | 464 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 465 | BUG_ON(ret); |
0b86a832 | 466 | } |
276e680d YZ |
467 | |
468 | if (root != root->fs_info->extent_root) | |
469 | switch_commit_root(root); | |
470 | ||
0b86a832 CM |
471 | return 0; |
472 | } | |
473 | ||
d352ac68 CM |
474 | /* |
475 | * update all the cowonly tree roots on disk | |
476 | */ | |
5d4f98a2 YZ |
477 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
478 | struct btrfs_root *root) | |
0b86a832 CM |
479 | { |
480 | struct btrfs_fs_info *fs_info = root->fs_info; | |
481 | struct list_head *next; | |
84234f3a | 482 | struct extent_buffer *eb; |
56bec294 | 483 | int ret; |
84234f3a | 484 | |
56bec294 CM |
485 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
486 | BUG_ON(ret); | |
87ef2bb4 | 487 | |
84234f3a | 488 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 489 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
490 | btrfs_tree_unlock(eb); |
491 | free_extent_buffer(eb); | |
0b86a832 | 492 | |
56bec294 CM |
493 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
494 | BUG_ON(ret); | |
87ef2bb4 | 495 | |
d397712b | 496 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
497 | next = fs_info->dirty_cowonly_roots.next; |
498 | list_del_init(next); | |
499 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 500 | |
0b86a832 | 501 | update_cowonly_root(trans, root); |
79154b1b | 502 | } |
276e680d YZ |
503 | |
504 | down_write(&fs_info->extent_commit_sem); | |
505 | switch_commit_root(fs_info->extent_root); | |
506 | up_write(&fs_info->extent_commit_sem); | |
507 | ||
79154b1b CM |
508 | return 0; |
509 | } | |
510 | ||
d352ac68 CM |
511 | /* |
512 | * dead roots are old snapshots that need to be deleted. This allocates | |
513 | * a dirty root struct and adds it into the list of dead roots that need to | |
514 | * be deleted | |
515 | */ | |
5d4f98a2 | 516 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 517 | { |
b48652c1 | 518 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 519 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 520 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
521 | return 0; |
522 | } | |
523 | ||
d352ac68 | 524 | /* |
5d4f98a2 | 525 | * update all the cowonly tree roots on disk |
d352ac68 | 526 | */ |
5d4f98a2 YZ |
527 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
528 | struct btrfs_root *root) | |
0f7d52f4 | 529 | { |
0f7d52f4 | 530 | struct btrfs_root *gang[8]; |
5d4f98a2 | 531 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
532 | int i; |
533 | int ret; | |
54aa1f4d CM |
534 | int err = 0; |
535 | ||
d397712b | 536 | while (1) { |
5d4f98a2 YZ |
537 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
538 | (void **)gang, 0, | |
0f7d52f4 CM |
539 | ARRAY_SIZE(gang), |
540 | BTRFS_ROOT_TRANS_TAG); | |
541 | if (ret == 0) | |
542 | break; | |
543 | for (i = 0; i < ret; i++) { | |
544 | root = gang[i]; | |
5d4f98a2 YZ |
545 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
546 | (unsigned long)root->root_key.objectid, | |
547 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 548 | |
e02119d5 | 549 | btrfs_free_log(trans, root); |
5d4f98a2 | 550 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 551 | |
978d910d | 552 | if (root->commit_root != root->node) { |
817d52f8 | 553 | switch_commit_root(root); |
978d910d YZ |
554 | btrfs_set_root_node(&root->root_item, |
555 | root->node); | |
556 | } | |
5d4f98a2 | 557 | |
5d4f98a2 | 558 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
559 | &root->root_key, |
560 | &root->root_item); | |
54aa1f4d CM |
561 | if (err) |
562 | break; | |
0f7d52f4 CM |
563 | } |
564 | } | |
54aa1f4d | 565 | return err; |
0f7d52f4 CM |
566 | } |
567 | ||
d352ac68 CM |
568 | /* |
569 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
570 | * otherwise every leaf in the btree is read and defragged. | |
571 | */ | |
e9d0b13b CM |
572 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
573 | { | |
574 | struct btrfs_fs_info *info = root->fs_info; | |
575 | int ret; | |
576 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 577 | unsigned long nr; |
e9d0b13b | 578 | |
a2135011 | 579 | smp_mb(); |
e9d0b13b CM |
580 | if (root->defrag_running) |
581 | return 0; | |
e9d0b13b | 582 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 583 | while (1) { |
e9d0b13b CM |
584 | root->defrag_running = 1; |
585 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 586 | nr = trans->blocks_used; |
e9d0b13b | 587 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 588 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
589 | cond_resched(); |
590 | ||
e9d0b13b | 591 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 592 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
593 | break; |
594 | } | |
595 | root->defrag_running = 0; | |
a2135011 | 596 | smp_mb(); |
e9d0b13b CM |
597 | btrfs_end_transaction(trans, root); |
598 | return 0; | |
599 | } | |
600 | ||
2c47e605 | 601 | #if 0 |
b7ec40d7 CM |
602 | /* |
603 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
604 | * sense not to join the transaction while it is trying to flush the current | |
605 | * queue of delayed refs out. | |
606 | * | |
607 | * This is used by the drop snapshot code only | |
608 | */ | |
609 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
610 | { | |
611 | DEFINE_WAIT(wait); | |
612 | ||
613 | mutex_lock(&info->trans_mutex); | |
614 | while (info->running_transaction && | |
615 | info->running_transaction->delayed_refs.flushing) { | |
616 | prepare_to_wait(&info->transaction_wait, &wait, | |
617 | TASK_UNINTERRUPTIBLE); | |
618 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 619 | |
b7ec40d7 | 620 | schedule(); |
59bc5c75 | 621 | |
b7ec40d7 CM |
622 | mutex_lock(&info->trans_mutex); |
623 | finish_wait(&info->transaction_wait, &wait); | |
624 | } | |
625 | mutex_unlock(&info->trans_mutex); | |
626 | return 0; | |
627 | } | |
628 | ||
d352ac68 CM |
629 | /* |
630 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
631 | * all of them | |
632 | */ | |
5d4f98a2 | 633 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 634 | { |
0f7d52f4 | 635 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 636 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 637 | unsigned long nr; |
5d4f98a2 | 638 | int ret; |
58176a96 | 639 | |
5d4f98a2 YZ |
640 | while (1) { |
641 | /* | |
642 | * we don't want to jump in and create a bunch of | |
643 | * delayed refs if the transaction is starting to close | |
644 | */ | |
645 | wait_transaction_pre_flush(tree_root->fs_info); | |
646 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 647 | |
5d4f98a2 YZ |
648 | /* |
649 | * we've joined a transaction, make sure it isn't | |
650 | * closing right now | |
651 | */ | |
652 | if (trans->transaction->delayed_refs.flushing) { | |
653 | btrfs_end_transaction(trans, tree_root); | |
654 | continue; | |
9f3a7427 | 655 | } |
58176a96 | 656 | |
5d4f98a2 YZ |
657 | ret = btrfs_drop_snapshot(trans, root); |
658 | if (ret != -EAGAIN) | |
659 | break; | |
a2135011 | 660 | |
5d4f98a2 YZ |
661 | ret = btrfs_update_root(trans, tree_root, |
662 | &root->root_key, | |
663 | &root->root_item); | |
664 | if (ret) | |
54aa1f4d | 665 | break; |
bcc63abb | 666 | |
d3c2fdcf | 667 | nr = trans->blocks_used; |
0f7d52f4 CM |
668 | ret = btrfs_end_transaction(trans, tree_root); |
669 | BUG_ON(ret); | |
5eda7b5e | 670 | |
d3c2fdcf | 671 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 672 | cond_resched(); |
0f7d52f4 | 673 | } |
5d4f98a2 YZ |
674 | BUG_ON(ret); |
675 | ||
676 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
677 | BUG_ON(ret); | |
678 | ||
679 | nr = trans->blocks_used; | |
680 | ret = btrfs_end_transaction(trans, tree_root); | |
681 | BUG_ON(ret); | |
682 | ||
683 | free_extent_buffer(root->node); | |
684 | free_extent_buffer(root->commit_root); | |
685 | kfree(root); | |
686 | ||
687 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 688 | return ret; |
0f7d52f4 | 689 | } |
2c47e605 | 690 | #endif |
0f7d52f4 | 691 | |
d352ac68 CM |
692 | /* |
693 | * new snapshots need to be created at a very specific time in the | |
694 | * transaction commit. This does the actual creation | |
695 | */ | |
80b6794d | 696 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
697 | struct btrfs_fs_info *fs_info, |
698 | struct btrfs_pending_snapshot *pending) | |
699 | { | |
700 | struct btrfs_key key; | |
80b6794d | 701 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
702 | struct btrfs_root *tree_root = fs_info->tree_root; |
703 | struct btrfs_root *root = pending->root; | |
704 | struct extent_buffer *tmp; | |
925baedd | 705 | struct extent_buffer *old; |
3063d29f CM |
706 | int ret; |
707 | u64 objectid; | |
708 | ||
80b6794d CM |
709 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
710 | if (!new_root_item) { | |
711 | ret = -ENOMEM; | |
712 | goto fail; | |
713 | } | |
3063d29f CM |
714 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
715 | if (ret) | |
716 | goto fail; | |
717 | ||
5d4f98a2 | 718 | record_root_in_trans(trans, root); |
80ff3856 | 719 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
80b6794d | 720 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
721 | |
722 | key.objectid = objectid; | |
5d4f98a2 | 723 | key.offset = 0; |
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 | /* add the backref first */ |
782 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
783 | pending->root_key.objectid, | |
784 | BTRFS_ROOT_BACKREF_KEY, | |
785 | parent_root->root_key.objectid, | |
786 | parent_inode->i_ino, index, pending->name, | |
787 | namelen); | |
788 | ||
789 | BUG_ON(ret); | |
790 | ||
791 | /* now add the forward ref */ | |
792 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
793 | parent_root->root_key.objectid, | |
794 | BTRFS_ROOT_REF_KEY, | |
795 | pending->root_key.objectid, | |
796 | parent_inode->i_ino, index, pending->name, | |
797 | namelen); | |
798 | ||
3de4586c CM |
799 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
800 | d_instantiate(pending->dentry, inode); | |
3063d29f | 801 | fail: |
3de4586c | 802 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
803 | return ret; |
804 | } | |
805 | ||
d352ac68 CM |
806 | /* |
807 | * create all the snapshots we've scheduled for creation | |
808 | */ | |
80b6794d CM |
809 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
810 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
811 | { |
812 | struct btrfs_pending_snapshot *pending; | |
813 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
814 | int ret; |
815 | ||
c6e30871 | 816 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
817 | ret = create_pending_snapshot(trans, fs_info, pending); |
818 | BUG_ON(ret); | |
819 | } | |
820 | return 0; | |
821 | } | |
822 | ||
823 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
824 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
825 | { |
826 | struct btrfs_pending_snapshot *pending; | |
827 | struct list_head *head = &trans->transaction->pending_snapshots; | |
828 | int ret; | |
829 | ||
d397712b | 830 | while (!list_empty(head)) { |
3063d29f CM |
831 | pending = list_entry(head->next, |
832 | struct btrfs_pending_snapshot, list); | |
3de4586c | 833 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
834 | BUG_ON(ret); |
835 | list_del(&pending->list); | |
836 | kfree(pending->name); | |
837 | kfree(pending); | |
838 | } | |
dc17ff8f CM |
839 | return 0; |
840 | } | |
841 | ||
5d4f98a2 YZ |
842 | static void update_super_roots(struct btrfs_root *root) |
843 | { | |
844 | struct btrfs_root_item *root_item; | |
845 | struct btrfs_super_block *super; | |
846 | ||
847 | super = &root->fs_info->super_copy; | |
848 | ||
849 | root_item = &root->fs_info->chunk_root->root_item; | |
850 | super->chunk_root = root_item->bytenr; | |
851 | super->chunk_root_generation = root_item->generation; | |
852 | super->chunk_root_level = root_item->level; | |
853 | ||
854 | root_item = &root->fs_info->tree_root->root_item; | |
855 | super->root = root_item->bytenr; | |
856 | super->generation = root_item->generation; | |
857 | super->root_level = root_item->level; | |
858 | } | |
859 | ||
f36f3042 CM |
860 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
861 | { | |
862 | int ret = 0; | |
863 | spin_lock(&info->new_trans_lock); | |
864 | if (info->running_transaction) | |
865 | ret = info->running_transaction->in_commit; | |
866 | spin_unlock(&info->new_trans_lock); | |
867 | return ret; | |
868 | } | |
869 | ||
79154b1b CM |
870 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
871 | struct btrfs_root *root) | |
872 | { | |
15ee9bc7 JB |
873 | unsigned long joined = 0; |
874 | unsigned long timeout = 1; | |
79154b1b | 875 | struct btrfs_transaction *cur_trans; |
8fd17795 | 876 | struct btrfs_transaction *prev_trans = NULL; |
d1310b2e | 877 | struct extent_io_tree *pinned_copy; |
79154b1b | 878 | DEFINE_WAIT(wait); |
15ee9bc7 | 879 | int ret; |
89573b9c CM |
880 | int should_grow = 0; |
881 | unsigned long now = get_seconds(); | |
dccae999 | 882 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 883 | |
5a3f23d5 CM |
884 | btrfs_run_ordered_operations(root, 0); |
885 | ||
56bec294 CM |
886 | /* make a pass through all the delayed refs we have so far |
887 | * any runnings procs may add more while we are here | |
888 | */ | |
889 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
890 | BUG_ON(ret); | |
891 | ||
b7ec40d7 | 892 | cur_trans = trans->transaction; |
56bec294 CM |
893 | /* |
894 | * set the flushing flag so procs in this transaction have to | |
895 | * start sending their work down. | |
896 | */ | |
b7ec40d7 | 897 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 898 | |
c3e69d58 | 899 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
900 | BUG_ON(ret); |
901 | ||
79154b1b | 902 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
903 | if (cur_trans->in_commit) { |
904 | cur_trans->use_count++; | |
ccd467d6 | 905 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 906 | btrfs_end_transaction(trans, root); |
ccd467d6 | 907 | |
79154b1b CM |
908 | ret = wait_for_commit(root, cur_trans); |
909 | BUG_ON(ret); | |
15ee9bc7 JB |
910 | |
911 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 912 | put_transaction(cur_trans); |
15ee9bc7 JB |
913 | mutex_unlock(&root->fs_info->trans_mutex); |
914 | ||
79154b1b CM |
915 | return 0; |
916 | } | |
4313b399 CM |
917 | |
918 | pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS); | |
919 | if (!pinned_copy) | |
920 | return -ENOMEM; | |
921 | ||
d1310b2e | 922 | extent_io_tree_init(pinned_copy, |
4313b399 CM |
923 | root->fs_info->btree_inode->i_mapping, GFP_NOFS); |
924 | ||
2c90e5d6 | 925 | trans->transaction->in_commit = 1; |
f9295749 | 926 | trans->transaction->blocked = 1; |
ccd467d6 CM |
927 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
928 | prev_trans = list_entry(cur_trans->list.prev, | |
929 | struct btrfs_transaction, list); | |
930 | if (!prev_trans->commit_done) { | |
931 | prev_trans->use_count++; | |
ccd467d6 CM |
932 | mutex_unlock(&root->fs_info->trans_mutex); |
933 | ||
934 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 935 | |
ccd467d6 | 936 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 937 | put_transaction(prev_trans); |
ccd467d6 CM |
938 | } |
939 | } | |
15ee9bc7 | 940 | |
89573b9c CM |
941 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
942 | should_grow = 1; | |
943 | ||
15ee9bc7 | 944 | do { |
7ea394f1 | 945 | int snap_pending = 0; |
15ee9bc7 | 946 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
947 | if (!list_empty(&trans->transaction->pending_snapshots)) |
948 | snap_pending = 1; | |
949 | ||
2c90e5d6 | 950 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 951 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 952 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
953 | |
954 | if (cur_trans->num_writers > 1) | |
955 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 956 | else if (should_grow) |
15ee9bc7 JB |
957 | timeout = 1; |
958 | ||
79154b1b | 959 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 960 | |
ebecd3d9 SW |
961 | if (flush_on_commit) { |
962 | btrfs_start_delalloc_inodes(root); | |
963 | ret = btrfs_wait_ordered_extents(root, 0); | |
964 | BUG_ON(ret); | |
965 | } else if (snap_pending) { | |
7ea394f1 YZ |
966 | ret = btrfs_wait_ordered_extents(root, 1); |
967 | BUG_ON(ret); | |
968 | } | |
969 | ||
5a3f23d5 CM |
970 | /* |
971 | * rename don't use btrfs_join_transaction, so, once we | |
972 | * set the transaction to blocked above, we aren't going | |
973 | * to get any new ordered operations. We can safely run | |
974 | * it here and no for sure that nothing new will be added | |
975 | * to the list | |
976 | */ | |
977 | btrfs_run_ordered_operations(root, 1); | |
978 | ||
89573b9c CM |
979 | smp_mb(); |
980 | if (cur_trans->num_writers > 1 || should_grow) | |
981 | schedule_timeout(timeout); | |
15ee9bc7 | 982 | |
79154b1b | 983 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
984 | finish_wait(&cur_trans->writer_wait, &wait); |
985 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 986 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 987 | |
3063d29f CM |
988 | ret = create_pending_snapshots(trans, root->fs_info); |
989 | BUG_ON(ret); | |
990 | ||
56bec294 CM |
991 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
992 | BUG_ON(ret); | |
993 | ||
2c90e5d6 | 994 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 995 | |
e02119d5 CM |
996 | /* btrfs_commit_tree_roots is responsible for getting the |
997 | * various roots consistent with each other. Every pointer | |
998 | * in the tree of tree roots has to point to the most up to date | |
999 | * root for every subvolume and other tree. So, we have to keep | |
1000 | * the tree logging code from jumping in and changing any | |
1001 | * of the trees. | |
1002 | * | |
1003 | * At this point in the commit, there can't be any tree-log | |
1004 | * writers, but a little lower down we drop the trans mutex | |
1005 | * and let new people in. By holding the tree_log_mutex | |
1006 | * from now until after the super is written, we avoid races | |
1007 | * with the tree-log code. | |
1008 | */ | |
1009 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1010 | ||
5d4f98a2 | 1011 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1012 | BUG_ON(ret); |
1013 | ||
5d4f98a2 | 1014 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1015 | * safe to free the root of tree log roots |
1016 | */ | |
1017 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1018 | ||
5d4f98a2 | 1019 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1020 | BUG_ON(ret); |
54aa1f4d | 1021 | |
78fae27e | 1022 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1023 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1024 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1025 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1026 | |
1027 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1028 | root->fs_info->tree_root->node); | |
817d52f8 | 1029 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1030 | |
1031 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1032 | root->fs_info->chunk_root->node); | |
817d52f8 | 1033 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1034 | |
1035 | update_super_roots(root); | |
e02119d5 CM |
1036 | |
1037 | if (!root->fs_info->log_root_recovering) { | |
1038 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1039 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1040 | } | |
1041 | ||
a061fc8d CM |
1042 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1043 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1044 | |
4313b399 | 1045 | btrfs_copy_pinned(root, pinned_copy); |
ccd467d6 | 1046 | |
f9295749 | 1047 | trans->transaction->blocked = 0; |
b7ec40d7 | 1048 | |
f9295749 | 1049 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1050 | |
78fae27e | 1051 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1052 | ret = btrfs_write_and_wait_transaction(trans, root); |
1053 | BUG_ON(ret); | |
a512bbf8 | 1054 | write_ctree_super(trans, root, 0); |
4313b399 | 1055 | |
e02119d5 CM |
1056 | /* |
1057 | * the super is written, we can safely allow the tree-loggers | |
1058 | * to go about their business | |
1059 | */ | |
1060 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1061 | ||
4313b399 | 1062 | btrfs_finish_extent_commit(trans, root, pinned_copy); |
4313b399 CM |
1063 | kfree(pinned_copy); |
1064 | ||
3de4586c CM |
1065 | /* do the directory inserts of any pending snapshot creations */ |
1066 | finish_pending_snapshots(trans, root->fs_info); | |
1067 | ||
1a40e23b ZY |
1068 | mutex_lock(&root->fs_info->trans_mutex); |
1069 | ||
2c90e5d6 | 1070 | cur_trans->commit_done = 1; |
b7ec40d7 | 1071 | |
15ee9bc7 | 1072 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1073 | |
2c90e5d6 | 1074 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1075 | |
78fae27e | 1076 | put_transaction(cur_trans); |
79154b1b | 1077 | put_transaction(cur_trans); |
58176a96 | 1078 | |
78fae27e | 1079 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1080 | |
2c90e5d6 | 1081 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b CM |
1082 | return ret; |
1083 | } | |
1084 | ||
d352ac68 CM |
1085 | /* |
1086 | * interface function to delete all the snapshots we have scheduled for deletion | |
1087 | */ | |
e9d0b13b CM |
1088 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1089 | { | |
5d4f98a2 YZ |
1090 | LIST_HEAD(list); |
1091 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1092 | ||
1093 | mutex_lock(&fs_info->trans_mutex); | |
1094 | list_splice_init(&fs_info->dead_roots, &list); | |
1095 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1096 | |
5d4f98a2 YZ |
1097 | while (!list_empty(&list)) { |
1098 | root = list_entry(list.next, struct btrfs_root, root_list); | |
1099 | list_del_init(&root->root_list); | |
2c47e605 | 1100 | btrfs_drop_snapshot(root, 0); |
e9d0b13b CM |
1101 | } |
1102 | return 0; | |
1103 | } |