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 | { | |
45 | down_write(&root->commit_root_sem); | |
46 | free_extent_buffer(root->commit_root); | |
47 | root->commit_root = btrfs_root_node(root); | |
48 | up_write(&root->commit_root_sem); | |
49 | } | |
50 | ||
d352ac68 CM |
51 | /* |
52 | * either allocate a new transaction or hop into the existing one | |
53 | */ | |
80b6794d | 54 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
55 | { |
56 | struct btrfs_transaction *cur_trans; | |
57 | cur_trans = root->fs_info->running_transaction; | |
58 | if (!cur_trans) { | |
2c90e5d6 CM |
59 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
60 | GFP_NOFS); | |
79154b1b | 61 | BUG_ON(!cur_trans); |
0f7d52f4 | 62 | root->fs_info->generation++; |
15ee9bc7 JB |
63 | cur_trans->num_writers = 1; |
64 | cur_trans->num_joined = 0; | |
0f7d52f4 | 65 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
66 | init_waitqueue_head(&cur_trans->writer_wait); |
67 | init_waitqueue_head(&cur_trans->commit_wait); | |
68 | cur_trans->in_commit = 0; | |
f9295749 | 69 | cur_trans->blocked = 0; |
d5719762 | 70 | cur_trans->use_count = 1; |
79154b1b | 71 | cur_trans->commit_done = 0; |
08607c1b | 72 | cur_trans->start_time = get_seconds(); |
56bec294 CM |
73 | |
74 | cur_trans->delayed_refs.root.rb_node = NULL; | |
75 | cur_trans->delayed_refs.num_entries = 0; | |
c3e69d58 CM |
76 | cur_trans->delayed_refs.num_heads_ready = 0; |
77 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 78 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 79 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
80 | spin_lock_init(&cur_trans->delayed_refs.lock); |
81 | ||
3063d29f | 82 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 83 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 84 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
85 | root->fs_info->btree_inode->i_mapping, |
86 | GFP_NOFS); | |
48ec2cf8 CM |
87 | spin_lock(&root->fs_info->new_trans_lock); |
88 | root->fs_info->running_transaction = cur_trans; | |
89 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
90 | } else { |
91 | cur_trans->num_writers++; | |
92 | cur_trans->num_joined++; | |
79154b1b | 93 | } |
15ee9bc7 | 94 | |
79154b1b CM |
95 | return 0; |
96 | } | |
97 | ||
d352ac68 | 98 | /* |
d397712b CM |
99 | * this does all the record keeping required to make sure that a reference |
100 | * counted root is properly recorded in a given transaction. This is required | |
101 | * to make sure the old root from before we joined the transaction is deleted | |
102 | * when the transaction commits | |
d352ac68 | 103 | */ |
5d4f98a2 YZ |
104 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
105 | struct btrfs_root *root) | |
6702ed49 | 106 | { |
5d4f98a2 | 107 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 108 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
109 | WARN_ON(root->root_item.refs == 0); |
110 | WARN_ON(root->commit_root != root->node); | |
111 | ||
112 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
113 | (unsigned long)root->root_key.objectid, | |
114 | BTRFS_ROOT_TRANS_TAG); | |
115 | root->last_trans = trans->transid; | |
116 | btrfs_init_reloc_root(trans, root); | |
117 | } | |
118 | return 0; | |
119 | } | |
bcc63abb | 120 | |
5d4f98a2 YZ |
121 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
122 | struct btrfs_root *root) | |
123 | { | |
124 | if (!root->ref_cows) | |
125 | return 0; | |
bcc63abb | 126 | |
5d4f98a2 YZ |
127 | mutex_lock(&root->fs_info->trans_mutex); |
128 | if (root->last_trans == trans->transid) { | |
129 | mutex_unlock(&root->fs_info->trans_mutex); | |
130 | return 0; | |
6702ed49 | 131 | } |
5d4f98a2 YZ |
132 | |
133 | record_root_in_trans(trans, root); | |
134 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
135 | return 0; |
136 | } | |
137 | ||
d352ac68 CM |
138 | /* wait for commit against the current transaction to become unblocked |
139 | * when this is done, it is safe to start a new transaction, but the current | |
140 | * transaction might not be fully on disk. | |
141 | */ | |
37d1aeee | 142 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 143 | { |
f9295749 | 144 | struct btrfs_transaction *cur_trans; |
79154b1b | 145 | |
f9295749 | 146 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 147 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
148 | DEFINE_WAIT(wait); |
149 | cur_trans->use_count++; | |
d397712b | 150 | while (1) { |
f9295749 CM |
151 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
152 | TASK_UNINTERRUPTIBLE); | |
153 | if (cur_trans->blocked) { | |
154 | mutex_unlock(&root->fs_info->trans_mutex); | |
155 | schedule(); | |
156 | mutex_lock(&root->fs_info->trans_mutex); | |
157 | finish_wait(&root->fs_info->transaction_wait, | |
158 | &wait); | |
159 | } else { | |
160 | finish_wait(&root->fs_info->transaction_wait, | |
161 | &wait); | |
162 | break; | |
163 | } | |
164 | } | |
165 | put_transaction(cur_trans); | |
166 | } | |
37d1aeee CM |
167 | } |
168 | ||
e02119d5 | 169 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
9ca9ee09 | 170 | int num_blocks, int wait) |
37d1aeee CM |
171 | { |
172 | struct btrfs_trans_handle *h = | |
173 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
174 | int ret; | |
175 | ||
176 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 CM |
177 | if (!root->fs_info->log_root_recovering && |
178 | ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) | |
37d1aeee | 179 | wait_current_trans(root); |
79154b1b CM |
180 | ret = join_transaction(root); |
181 | BUG_ON(ret); | |
0f7d52f4 | 182 | |
6702ed49 | 183 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
184 | h->transaction = root->fs_info->running_transaction; |
185 | h->blocks_reserved = num_blocks; | |
186 | h->blocks_used = 0; | |
d2fb3437 | 187 | h->block_group = 0; |
26b8003f CM |
188 | h->alloc_exclude_nr = 0; |
189 | h->alloc_exclude_start = 0; | |
56bec294 | 190 | h->delayed_ref_updates = 0; |
b7ec40d7 | 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); |
d6025579 | 323 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 324 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 325 | |
79154b1b CM |
326 | return 0; |
327 | } | |
328 | ||
89ce8a63 CM |
329 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
330 | struct btrfs_root *root) | |
331 | { | |
332 | return __btrfs_end_transaction(trans, root, 0); | |
333 | } | |
334 | ||
335 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
336 | struct btrfs_root *root) | |
337 | { | |
338 | return __btrfs_end_transaction(trans, root, 1); | |
339 | } | |
340 | ||
d352ac68 CM |
341 | /* |
342 | * when btree blocks are allocated, they have some corresponding bits set for | |
343 | * them in one of two extent_io trees. This is used to make sure all of | |
344 | * those extents are on disk for transaction or log commit | |
345 | */ | |
d0c803c4 CM |
346 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
347 | struct extent_io_tree *dirty_pages) | |
79154b1b | 348 | { |
7c4452b9 | 349 | int ret; |
777e6bd7 | 350 | int err = 0; |
7c4452b9 CM |
351 | int werr = 0; |
352 | struct page *page; | |
7c4452b9 | 353 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 354 | u64 start = 0; |
5f39d397 CM |
355 | u64 end; |
356 | unsigned long index; | |
7c4452b9 | 357 | |
d397712b | 358 | while (1) { |
777e6bd7 | 359 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
360 | EXTENT_DIRTY); |
361 | if (ret) | |
7c4452b9 | 362 | break; |
d397712b | 363 | while (start <= end) { |
777e6bd7 CM |
364 | cond_resched(); |
365 | ||
5f39d397 | 366 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 367 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 368 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
369 | if (!page) |
370 | continue; | |
4bef0848 CM |
371 | |
372 | btree_lock_page_hook(page); | |
373 | if (!page->mapping) { | |
374 | unlock_page(page); | |
375 | page_cache_release(page); | |
376 | continue; | |
377 | } | |
378 | ||
6702ed49 CM |
379 | if (PageWriteback(page)) { |
380 | if (PageDirty(page)) | |
381 | wait_on_page_writeback(page); | |
382 | else { | |
383 | unlock_page(page); | |
384 | page_cache_release(page); | |
385 | continue; | |
386 | } | |
387 | } | |
7c4452b9 CM |
388 | err = write_one_page(page, 0); |
389 | if (err) | |
390 | werr = err; | |
391 | page_cache_release(page); | |
392 | } | |
393 | } | |
d397712b | 394 | while (1) { |
777e6bd7 CM |
395 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, |
396 | EXTENT_DIRTY); | |
397 | if (ret) | |
398 | break; | |
399 | ||
400 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
d397712b | 401 | while (start <= end) { |
777e6bd7 CM |
402 | index = start >> PAGE_CACHE_SHIFT; |
403 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
404 | page = find_get_page(btree_inode->i_mapping, index); | |
405 | if (!page) | |
406 | continue; | |
407 | if (PageDirty(page)) { | |
4bef0848 CM |
408 | btree_lock_page_hook(page); |
409 | wait_on_page_writeback(page); | |
777e6bd7 CM |
410 | err = write_one_page(page, 0); |
411 | if (err) | |
412 | werr = err; | |
413 | } | |
105d931d | 414 | wait_on_page_writeback(page); |
777e6bd7 CM |
415 | page_cache_release(page); |
416 | cond_resched(); | |
417 | } | |
418 | } | |
7c4452b9 CM |
419 | if (err) |
420 | werr = err; | |
421 | return werr; | |
79154b1b CM |
422 | } |
423 | ||
d0c803c4 CM |
424 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
425 | struct btrfs_root *root) | |
426 | { | |
427 | if (!trans || !trans->transaction) { | |
428 | struct inode *btree_inode; | |
429 | btree_inode = root->fs_info->btree_inode; | |
430 | return filemap_write_and_wait(btree_inode->i_mapping); | |
431 | } | |
432 | return btrfs_write_and_wait_marked_extents(root, | |
433 | &trans->transaction->dirty_pages); | |
434 | } | |
435 | ||
d352ac68 CM |
436 | /* |
437 | * this is used to update the root pointer in the tree of tree roots. | |
438 | * | |
439 | * But, in the case of the extent allocation tree, updating the root | |
440 | * pointer may allocate blocks which may change the root of the extent | |
441 | * allocation tree. | |
442 | * | |
443 | * So, this loops and repeats and makes sure the cowonly root didn't | |
444 | * change while the root pointer was being updated in the metadata. | |
445 | */ | |
0b86a832 CM |
446 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
447 | struct btrfs_root *root) | |
79154b1b CM |
448 | { |
449 | int ret; | |
0b86a832 CM |
450 | u64 old_root_bytenr; |
451 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 452 | |
0b86a832 | 453 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 454 | |
d397712b | 455 | while (1) { |
0b86a832 CM |
456 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
457 | if (old_root_bytenr == root->node->start) | |
79154b1b | 458 | break; |
87ef2bb4 | 459 | |
5d4f98a2 | 460 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 461 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
462 | &root->root_key, |
463 | &root->root_item); | |
79154b1b | 464 | BUG_ON(ret); |
56bec294 | 465 | |
4a8c9a62 | 466 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 467 | BUG_ON(ret); |
0b86a832 | 468 | } |
817d52f8 | 469 | switch_commit_root(root); |
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 CM |
501 | } |
502 | return 0; | |
503 | } | |
504 | ||
d352ac68 CM |
505 | /* |
506 | * dead roots are old snapshots that need to be deleted. This allocates | |
507 | * a dirty root struct and adds it into the list of dead roots that need to | |
508 | * be deleted | |
509 | */ | |
5d4f98a2 | 510 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 511 | { |
b48652c1 | 512 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 513 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 514 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
515 | return 0; |
516 | } | |
517 | ||
d352ac68 | 518 | /* |
5d4f98a2 | 519 | * update all the cowonly tree roots on disk |
d352ac68 | 520 | */ |
5d4f98a2 YZ |
521 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
522 | struct btrfs_root *root) | |
0f7d52f4 | 523 | { |
0f7d52f4 | 524 | struct btrfs_root *gang[8]; |
5d4f98a2 | 525 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
526 | int i; |
527 | int ret; | |
54aa1f4d CM |
528 | int err = 0; |
529 | ||
d397712b | 530 | while (1) { |
5d4f98a2 YZ |
531 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
532 | (void **)gang, 0, | |
0f7d52f4 CM |
533 | ARRAY_SIZE(gang), |
534 | BTRFS_ROOT_TRANS_TAG); | |
535 | if (ret == 0) | |
536 | break; | |
537 | for (i = 0; i < ret; i++) { | |
538 | root = gang[i]; | |
5d4f98a2 YZ |
539 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
540 | (unsigned long)root->root_key.objectid, | |
541 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 542 | |
e02119d5 | 543 | btrfs_free_log(trans, root); |
5d4f98a2 | 544 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 545 | |
978d910d | 546 | if (root->commit_root != root->node) { |
817d52f8 | 547 | switch_commit_root(root); |
978d910d YZ |
548 | btrfs_set_root_node(&root->root_item, |
549 | root->node); | |
550 | } | |
5d4f98a2 | 551 | |
5d4f98a2 | 552 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
553 | &root->root_key, |
554 | &root->root_item); | |
54aa1f4d CM |
555 | if (err) |
556 | break; | |
0f7d52f4 CM |
557 | } |
558 | } | |
54aa1f4d | 559 | return err; |
0f7d52f4 CM |
560 | } |
561 | ||
d352ac68 CM |
562 | /* |
563 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
564 | * otherwise every leaf in the btree is read and defragged. | |
565 | */ | |
e9d0b13b CM |
566 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
567 | { | |
568 | struct btrfs_fs_info *info = root->fs_info; | |
569 | int ret; | |
570 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 571 | unsigned long nr; |
e9d0b13b | 572 | |
a2135011 | 573 | smp_mb(); |
e9d0b13b CM |
574 | if (root->defrag_running) |
575 | return 0; | |
e9d0b13b | 576 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 577 | while (1) { |
e9d0b13b CM |
578 | root->defrag_running = 1; |
579 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 580 | nr = trans->blocks_used; |
e9d0b13b | 581 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 582 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
583 | cond_resched(); |
584 | ||
e9d0b13b | 585 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 586 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
587 | break; |
588 | } | |
589 | root->defrag_running = 0; | |
a2135011 | 590 | smp_mb(); |
e9d0b13b CM |
591 | btrfs_end_transaction(trans, root); |
592 | return 0; | |
593 | } | |
594 | ||
2c47e605 | 595 | #if 0 |
b7ec40d7 CM |
596 | /* |
597 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
598 | * sense not to join the transaction while it is trying to flush the current | |
599 | * queue of delayed refs out. | |
600 | * | |
601 | * This is used by the drop snapshot code only | |
602 | */ | |
603 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
604 | { | |
605 | DEFINE_WAIT(wait); | |
606 | ||
607 | mutex_lock(&info->trans_mutex); | |
608 | while (info->running_transaction && | |
609 | info->running_transaction->delayed_refs.flushing) { | |
610 | prepare_to_wait(&info->transaction_wait, &wait, | |
611 | TASK_UNINTERRUPTIBLE); | |
612 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 613 | |
b7ec40d7 | 614 | schedule(); |
59bc5c75 | 615 | |
b7ec40d7 CM |
616 | mutex_lock(&info->trans_mutex); |
617 | finish_wait(&info->transaction_wait, &wait); | |
618 | } | |
619 | mutex_unlock(&info->trans_mutex); | |
620 | return 0; | |
621 | } | |
622 | ||
d352ac68 CM |
623 | /* |
624 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
625 | * all of them | |
626 | */ | |
5d4f98a2 | 627 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 628 | { |
0f7d52f4 | 629 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 630 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 631 | unsigned long nr; |
5d4f98a2 | 632 | int ret; |
58176a96 | 633 | |
5d4f98a2 YZ |
634 | while (1) { |
635 | /* | |
636 | * we don't want to jump in and create a bunch of | |
637 | * delayed refs if the transaction is starting to close | |
638 | */ | |
639 | wait_transaction_pre_flush(tree_root->fs_info); | |
640 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 641 | |
5d4f98a2 YZ |
642 | /* |
643 | * we've joined a transaction, make sure it isn't | |
644 | * closing right now | |
645 | */ | |
646 | if (trans->transaction->delayed_refs.flushing) { | |
647 | btrfs_end_transaction(trans, tree_root); | |
648 | continue; | |
9f3a7427 | 649 | } |
58176a96 | 650 | |
5d4f98a2 YZ |
651 | ret = btrfs_drop_snapshot(trans, root); |
652 | if (ret != -EAGAIN) | |
653 | break; | |
a2135011 | 654 | |
5d4f98a2 YZ |
655 | ret = btrfs_update_root(trans, tree_root, |
656 | &root->root_key, | |
657 | &root->root_item); | |
658 | if (ret) | |
54aa1f4d | 659 | break; |
bcc63abb | 660 | |
d3c2fdcf | 661 | nr = trans->blocks_used; |
0f7d52f4 CM |
662 | ret = btrfs_end_transaction(trans, tree_root); |
663 | BUG_ON(ret); | |
5eda7b5e | 664 | |
d3c2fdcf | 665 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 666 | cond_resched(); |
0f7d52f4 | 667 | } |
5d4f98a2 YZ |
668 | BUG_ON(ret); |
669 | ||
670 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
671 | BUG_ON(ret); | |
672 | ||
673 | nr = trans->blocks_used; | |
674 | ret = btrfs_end_transaction(trans, tree_root); | |
675 | BUG_ON(ret); | |
676 | ||
677 | free_extent_buffer(root->node); | |
678 | free_extent_buffer(root->commit_root); | |
679 | kfree(root); | |
680 | ||
681 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 682 | return ret; |
0f7d52f4 | 683 | } |
2c47e605 | 684 | #endif |
0f7d52f4 | 685 | |
d352ac68 CM |
686 | /* |
687 | * new snapshots need to be created at a very specific time in the | |
688 | * transaction commit. This does the actual creation | |
689 | */ | |
80b6794d | 690 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
691 | struct btrfs_fs_info *fs_info, |
692 | struct btrfs_pending_snapshot *pending) | |
693 | { | |
694 | struct btrfs_key key; | |
80b6794d | 695 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
696 | struct btrfs_root *tree_root = fs_info->tree_root; |
697 | struct btrfs_root *root = pending->root; | |
698 | struct extent_buffer *tmp; | |
925baedd | 699 | struct extent_buffer *old; |
3063d29f CM |
700 | int ret; |
701 | u64 objectid; | |
702 | ||
80b6794d CM |
703 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
704 | if (!new_root_item) { | |
705 | ret = -ENOMEM; | |
706 | goto fail; | |
707 | } | |
3063d29f CM |
708 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
709 | if (ret) | |
710 | goto fail; | |
711 | ||
5d4f98a2 | 712 | record_root_in_trans(trans, root); |
80ff3856 | 713 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
80b6794d | 714 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
715 | |
716 | key.objectid = objectid; | |
5d4f98a2 | 717 | key.offset = 0; |
3063d29f CM |
718 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
719 | ||
925baedd | 720 | old = btrfs_lock_root_node(root); |
9fa8cfe7 | 721 | btrfs_cow_block(trans, root, old, NULL, 0, &old); |
5d4f98a2 | 722 | btrfs_set_lock_blocking(old); |
3063d29f | 723 | |
925baedd CM |
724 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
725 | btrfs_tree_unlock(old); | |
726 | free_extent_buffer(old); | |
3063d29f | 727 | |
5d4f98a2 | 728 | btrfs_set_root_node(new_root_item, tmp); |
3063d29f | 729 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 730 | new_root_item); |
925baedd | 731 | btrfs_tree_unlock(tmp); |
3063d29f CM |
732 | free_extent_buffer(tmp); |
733 | if (ret) | |
734 | goto fail; | |
735 | ||
3de4586c CM |
736 | key.offset = (u64)-1; |
737 | memcpy(&pending->root_key, &key, sizeof(key)); | |
738 | fail: | |
739 | kfree(new_root_item); | |
740 | return ret; | |
741 | } | |
742 | ||
743 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
744 | struct btrfs_pending_snapshot *pending) | |
745 | { | |
746 | int ret; | |
747 | int namelen; | |
748 | u64 index = 0; | |
749 | struct btrfs_trans_handle *trans; | |
750 | struct inode *parent_inode; | |
751 | struct inode *inode; | |
0660b5af | 752 | struct btrfs_root *parent_root; |
3de4586c | 753 | |
3394e160 | 754 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 755 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 756 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 757 | |
3063d29f CM |
758 | /* |
759 | * insert the directory item | |
760 | */ | |
3b96362c | 761 | namelen = strlen(pending->name); |
3de4586c | 762 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 763 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
764 | pending->name, namelen, |
765 | parent_inode->i_ino, | |
766 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
767 | |
768 | if (ret) | |
769 | goto fail; | |
0660b5af | 770 | |
52c26179 YZ |
771 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
772 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
773 | BUG_ON(ret); | |
774 | ||
0660b5af CM |
775 | /* add the backref first */ |
776 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
777 | pending->root_key.objectid, | |
778 | BTRFS_ROOT_BACKREF_KEY, | |
779 | parent_root->root_key.objectid, | |
780 | parent_inode->i_ino, index, pending->name, | |
781 | namelen); | |
782 | ||
783 | BUG_ON(ret); | |
784 | ||
785 | /* now add the forward ref */ | |
786 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
787 | parent_root->root_key.objectid, | |
788 | BTRFS_ROOT_REF_KEY, | |
789 | pending->root_key.objectid, | |
790 | parent_inode->i_ino, index, pending->name, | |
791 | namelen); | |
792 | ||
3de4586c CM |
793 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
794 | d_instantiate(pending->dentry, inode); | |
3063d29f | 795 | fail: |
3de4586c | 796 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
797 | return ret; |
798 | } | |
799 | ||
d352ac68 CM |
800 | /* |
801 | * create all the snapshots we've scheduled for creation | |
802 | */ | |
80b6794d CM |
803 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
804 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
805 | { |
806 | struct btrfs_pending_snapshot *pending; | |
807 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
808 | int ret; |
809 | ||
c6e30871 | 810 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
811 | ret = create_pending_snapshot(trans, fs_info, pending); |
812 | BUG_ON(ret); | |
813 | } | |
814 | return 0; | |
815 | } | |
816 | ||
817 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
818 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
819 | { |
820 | struct btrfs_pending_snapshot *pending; | |
821 | struct list_head *head = &trans->transaction->pending_snapshots; | |
822 | int ret; | |
823 | ||
d397712b | 824 | while (!list_empty(head)) { |
3063d29f CM |
825 | pending = list_entry(head->next, |
826 | struct btrfs_pending_snapshot, list); | |
3de4586c | 827 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
828 | BUG_ON(ret); |
829 | list_del(&pending->list); | |
830 | kfree(pending->name); | |
831 | kfree(pending); | |
832 | } | |
dc17ff8f CM |
833 | return 0; |
834 | } | |
835 | ||
5d4f98a2 YZ |
836 | static void update_super_roots(struct btrfs_root *root) |
837 | { | |
838 | struct btrfs_root_item *root_item; | |
839 | struct btrfs_super_block *super; | |
840 | ||
841 | super = &root->fs_info->super_copy; | |
842 | ||
843 | root_item = &root->fs_info->chunk_root->root_item; | |
844 | super->chunk_root = root_item->bytenr; | |
845 | super->chunk_root_generation = root_item->generation; | |
846 | super->chunk_root_level = root_item->level; | |
847 | ||
848 | root_item = &root->fs_info->tree_root->root_item; | |
849 | super->root = root_item->bytenr; | |
850 | super->generation = root_item->generation; | |
851 | super->root_level = root_item->level; | |
852 | } | |
853 | ||
79154b1b CM |
854 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
855 | struct btrfs_root *root) | |
856 | { | |
15ee9bc7 JB |
857 | unsigned long joined = 0; |
858 | unsigned long timeout = 1; | |
79154b1b | 859 | struct btrfs_transaction *cur_trans; |
8fd17795 | 860 | struct btrfs_transaction *prev_trans = NULL; |
d1310b2e | 861 | struct extent_io_tree *pinned_copy; |
79154b1b | 862 | DEFINE_WAIT(wait); |
15ee9bc7 | 863 | int ret; |
89573b9c CM |
864 | int should_grow = 0; |
865 | unsigned long now = get_seconds(); | |
dccae999 | 866 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 867 | |
5a3f23d5 CM |
868 | btrfs_run_ordered_operations(root, 0); |
869 | ||
56bec294 CM |
870 | /* make a pass through all the delayed refs we have so far |
871 | * any runnings procs may add more while we are here | |
872 | */ | |
873 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
874 | BUG_ON(ret); | |
875 | ||
b7ec40d7 | 876 | cur_trans = trans->transaction; |
56bec294 CM |
877 | /* |
878 | * set the flushing flag so procs in this transaction have to | |
879 | * start sending their work down. | |
880 | */ | |
b7ec40d7 | 881 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 882 | |
c3e69d58 | 883 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
884 | BUG_ON(ret); |
885 | ||
79154b1b | 886 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
887 | if (cur_trans->in_commit) { |
888 | cur_trans->use_count++; | |
ccd467d6 | 889 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 890 | btrfs_end_transaction(trans, root); |
ccd467d6 | 891 | |
79154b1b CM |
892 | ret = wait_for_commit(root, cur_trans); |
893 | BUG_ON(ret); | |
15ee9bc7 JB |
894 | |
895 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 896 | put_transaction(cur_trans); |
15ee9bc7 JB |
897 | mutex_unlock(&root->fs_info->trans_mutex); |
898 | ||
79154b1b CM |
899 | return 0; |
900 | } | |
4313b399 CM |
901 | |
902 | pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS); | |
903 | if (!pinned_copy) | |
904 | return -ENOMEM; | |
905 | ||
d1310b2e | 906 | extent_io_tree_init(pinned_copy, |
4313b399 CM |
907 | root->fs_info->btree_inode->i_mapping, GFP_NOFS); |
908 | ||
2c90e5d6 | 909 | trans->transaction->in_commit = 1; |
f9295749 | 910 | trans->transaction->blocked = 1; |
ccd467d6 CM |
911 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
912 | prev_trans = list_entry(cur_trans->list.prev, | |
913 | struct btrfs_transaction, list); | |
914 | if (!prev_trans->commit_done) { | |
915 | prev_trans->use_count++; | |
ccd467d6 CM |
916 | mutex_unlock(&root->fs_info->trans_mutex); |
917 | ||
918 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 919 | |
ccd467d6 | 920 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 921 | put_transaction(prev_trans); |
ccd467d6 CM |
922 | } |
923 | } | |
15ee9bc7 | 924 | |
89573b9c CM |
925 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
926 | should_grow = 1; | |
927 | ||
15ee9bc7 | 928 | do { |
7ea394f1 | 929 | int snap_pending = 0; |
15ee9bc7 | 930 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
931 | if (!list_empty(&trans->transaction->pending_snapshots)) |
932 | snap_pending = 1; | |
933 | ||
2c90e5d6 | 934 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 935 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 936 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
937 | |
938 | if (cur_trans->num_writers > 1) | |
939 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 940 | else if (should_grow) |
15ee9bc7 JB |
941 | timeout = 1; |
942 | ||
79154b1b | 943 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 944 | |
dccae999 SW |
945 | if (flush_on_commit || snap_pending) { |
946 | if (flush_on_commit) | |
947 | btrfs_start_delalloc_inodes(root); | |
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 | |
78fae27e | 1004 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1005 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1006 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1007 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1008 | |
1009 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1010 | root->fs_info->tree_root->node); | |
817d52f8 | 1011 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1012 | |
1013 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1014 | root->fs_info->chunk_root->node); | |
817d52f8 | 1015 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1016 | |
1017 | update_super_roots(root); | |
e02119d5 CM |
1018 | |
1019 | if (!root->fs_info->log_root_recovering) { | |
1020 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1021 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1022 | } | |
1023 | ||
a061fc8d CM |
1024 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1025 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1026 | |
4313b399 | 1027 | btrfs_copy_pinned(root, pinned_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 | ||
4313b399 | 1044 | btrfs_finish_extent_commit(trans, root, pinned_copy); |
4313b399 CM |
1045 | kfree(pinned_copy); |
1046 | ||
3de4586c CM |
1047 | /* do the directory inserts of any pending snapshot creations */ |
1048 | finish_pending_snapshots(trans, root->fs_info); | |
1049 | ||
1a40e23b ZY |
1050 | mutex_lock(&root->fs_info->trans_mutex); |
1051 | ||
2c90e5d6 | 1052 | cur_trans->commit_done = 1; |
b7ec40d7 | 1053 | |
15ee9bc7 | 1054 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1055 | |
2c90e5d6 | 1056 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1057 | |
78fae27e | 1058 | put_transaction(cur_trans); |
79154b1b | 1059 | put_transaction(cur_trans); |
58176a96 | 1060 | |
78fae27e | 1061 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1062 | |
2c90e5d6 | 1063 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b CM |
1064 | return ret; |
1065 | } | |
1066 | ||
d352ac68 CM |
1067 | /* |
1068 | * interface function to delete all the snapshots we have scheduled for deletion | |
1069 | */ | |
e9d0b13b CM |
1070 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1071 | { | |
5d4f98a2 YZ |
1072 | LIST_HEAD(list); |
1073 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1074 | ||
1075 | mutex_lock(&fs_info->trans_mutex); | |
1076 | list_splice_init(&fs_info->dead_roots, &list); | |
1077 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1078 | |
5d4f98a2 YZ |
1079 | while (!list_empty(&list)) { |
1080 | root = list_entry(list.next, struct btrfs_root, root_list); | |
1081 | list_del_init(&root->root_list); | |
2c47e605 | 1082 | btrfs_drop_snapshot(root, 0); |
e9d0b13b CM |
1083 | } |
1084 | return 0; | |
1085 | } |