Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
34088780 | 20 | #include <linux/sched.h> |
d3c2fdcf | 21 | #include <linux/writeback.h> |
5f39d397 | 22 | #include <linux/pagemap.h> |
5f2cc086 | 23 | #include <linux/blkdev.h> |
79154b1b CM |
24 | #include "ctree.h" |
25 | #include "disk-io.h" | |
26 | #include "transaction.h" | |
925baedd | 27 | #include "locking.h" |
31153d81 | 28 | #include "ref-cache.h" |
e02119d5 | 29 | #include "tree-log.h" |
79154b1b | 30 | |
0f7d52f4 CM |
31 | #define BTRFS_ROOT_TRANS_TAG 0 |
32 | ||
80b6794d | 33 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 34 | { |
2c90e5d6 | 35 | WARN_ON(transaction->use_count == 0); |
79154b1b | 36 | transaction->use_count--; |
78fae27e | 37 | if (transaction->use_count == 0) { |
8fd17795 | 38 | list_del_init(&transaction->list); |
2c90e5d6 CM |
39 | memset(transaction, 0, sizeof(*transaction)); |
40 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 41 | } |
79154b1b CM |
42 | } |
43 | ||
d352ac68 CM |
44 | /* |
45 | * either allocate a new transaction or hop into the existing one | |
46 | */ | |
80b6794d | 47 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
48 | { |
49 | struct btrfs_transaction *cur_trans; | |
50 | cur_trans = root->fs_info->running_transaction; | |
51 | if (!cur_trans) { | |
2c90e5d6 CM |
52 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
53 | GFP_NOFS); | |
79154b1b | 54 | BUG_ON(!cur_trans); |
0f7d52f4 | 55 | root->fs_info->generation++; |
e18e4809 | 56 | root->fs_info->last_alloc = 0; |
4529ba49 | 57 | root->fs_info->last_data_alloc = 0; |
15ee9bc7 JB |
58 | cur_trans->num_writers = 1; |
59 | cur_trans->num_joined = 0; | |
0f7d52f4 | 60 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
61 | init_waitqueue_head(&cur_trans->writer_wait); |
62 | init_waitqueue_head(&cur_trans->commit_wait); | |
63 | cur_trans->in_commit = 0; | |
f9295749 | 64 | cur_trans->blocked = 0; |
d5719762 | 65 | cur_trans->use_count = 1; |
79154b1b | 66 | cur_trans->commit_done = 0; |
08607c1b | 67 | cur_trans->start_time = get_seconds(); |
3063d29f | 68 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 69 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 70 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
71 | root->fs_info->btree_inode->i_mapping, |
72 | GFP_NOFS); | |
48ec2cf8 CM |
73 | spin_lock(&root->fs_info->new_trans_lock); |
74 | root->fs_info->running_transaction = cur_trans; | |
75 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
76 | } else { |
77 | cur_trans->num_writers++; | |
78 | cur_trans->num_joined++; | |
79154b1b | 79 | } |
15ee9bc7 | 80 | |
79154b1b CM |
81 | return 0; |
82 | } | |
83 | ||
d352ac68 | 84 | /* |
d397712b CM |
85 | * this does all the record keeping required to make sure that a reference |
86 | * counted root is properly recorded in a given transaction. This is required | |
87 | * to make sure the old root from before we joined the transaction is deleted | |
88 | * when the transaction commits | |
d352ac68 | 89 | */ |
e02119d5 | 90 | noinline int btrfs_record_root_in_trans(struct btrfs_root *root) |
6702ed49 | 91 | { |
f321e491 | 92 | struct btrfs_dirty_root *dirty; |
6702ed49 CM |
93 | u64 running_trans_id = root->fs_info->running_transaction->transid; |
94 | if (root->ref_cows && root->last_trans < running_trans_id) { | |
95 | WARN_ON(root == root->fs_info->extent_root); | |
96 | if (root->root_item.refs != 0) { | |
97 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
98 | (unsigned long)root->root_key.objectid, | |
99 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
100 | |
101 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
102 | BUG_ON(!dirty); | |
103 | dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS); | |
104 | BUG_ON(!dirty->root); | |
31153d81 YZ |
105 | dirty->latest_root = root; |
106 | INIT_LIST_HEAD(&dirty->list); | |
31153d81 | 107 | |
925baedd | 108 | root->commit_root = btrfs_root_node(root); |
31153d81 YZ |
109 | |
110 | memcpy(dirty->root, root, sizeof(*root)); | |
111 | spin_lock_init(&dirty->root->node_lock); | |
bcc63abb | 112 | spin_lock_init(&dirty->root->list_lock); |
31153d81 | 113 | mutex_init(&dirty->root->objectid_mutex); |
5b21f2ed | 114 | mutex_init(&dirty->root->log_mutex); |
bcc63abb | 115 | INIT_LIST_HEAD(&dirty->root->dead_list); |
31153d81 YZ |
116 | dirty->root->node = root->commit_root; |
117 | dirty->root->commit_root = NULL; | |
bcc63abb Y |
118 | |
119 | spin_lock(&root->list_lock); | |
120 | list_add(&dirty->root->dead_list, &root->dead_list); | |
121 | spin_unlock(&root->list_lock); | |
122 | ||
123 | root->dirty_root = dirty; | |
6702ed49 CM |
124 | } else { |
125 | WARN_ON(1); | |
126 | } | |
127 | root->last_trans = running_trans_id; | |
128 | } | |
129 | return 0; | |
130 | } | |
131 | ||
d352ac68 CM |
132 | /* wait for commit against the current transaction to become unblocked |
133 | * when this is done, it is safe to start a new transaction, but the current | |
134 | * transaction might not be fully on disk. | |
135 | */ | |
37d1aeee | 136 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 137 | { |
f9295749 | 138 | struct btrfs_transaction *cur_trans; |
79154b1b | 139 | |
f9295749 | 140 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 141 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
142 | DEFINE_WAIT(wait); |
143 | cur_trans->use_count++; | |
d397712b | 144 | while (1) { |
f9295749 CM |
145 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
146 | TASK_UNINTERRUPTIBLE); | |
147 | if (cur_trans->blocked) { | |
148 | mutex_unlock(&root->fs_info->trans_mutex); | |
149 | schedule(); | |
150 | mutex_lock(&root->fs_info->trans_mutex); | |
151 | finish_wait(&root->fs_info->transaction_wait, | |
152 | &wait); | |
153 | } else { | |
154 | finish_wait(&root->fs_info->transaction_wait, | |
155 | &wait); | |
156 | break; | |
157 | } | |
158 | } | |
159 | put_transaction(cur_trans); | |
160 | } | |
37d1aeee CM |
161 | } |
162 | ||
e02119d5 | 163 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
9ca9ee09 | 164 | int num_blocks, int wait) |
37d1aeee CM |
165 | { |
166 | struct btrfs_trans_handle *h = | |
167 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
168 | int ret; | |
169 | ||
170 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 CM |
171 | if (!root->fs_info->log_root_recovering && |
172 | ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) | |
37d1aeee | 173 | wait_current_trans(root); |
79154b1b CM |
174 | ret = join_transaction(root); |
175 | BUG_ON(ret); | |
0f7d52f4 | 176 | |
e02119d5 | 177 | btrfs_record_root_in_trans(root); |
6702ed49 | 178 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
179 | h->transaction = root->fs_info->running_transaction; |
180 | h->blocks_reserved = num_blocks; | |
181 | h->blocks_used = 0; | |
d2fb3437 | 182 | h->block_group = 0; |
26b8003f CM |
183 | h->alloc_exclude_nr = 0; |
184 | h->alloc_exclude_start = 0; | |
79154b1b CM |
185 | root->fs_info->running_transaction->use_count++; |
186 | mutex_unlock(&root->fs_info->trans_mutex); | |
187 | return h; | |
188 | } | |
189 | ||
f9295749 CM |
190 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
191 | int num_blocks) | |
192 | { | |
9ca9ee09 | 193 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
194 | } |
195 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
196 | int num_blocks) | |
197 | { | |
9ca9ee09 | 198 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
199 | } |
200 | ||
9ca9ee09 SW |
201 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
202 | int num_blocks) | |
203 | { | |
204 | return start_transaction(r, num_blocks, 2); | |
205 | } | |
206 | ||
d352ac68 | 207 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
208 | static noinline int wait_for_commit(struct btrfs_root *root, |
209 | struct btrfs_transaction *commit) | |
210 | { | |
211 | DEFINE_WAIT(wait); | |
212 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 213 | while (!commit->commit_done) { |
89ce8a63 CM |
214 | prepare_to_wait(&commit->commit_wait, &wait, |
215 | TASK_UNINTERRUPTIBLE); | |
216 | if (commit->commit_done) | |
217 | break; | |
218 | mutex_unlock(&root->fs_info->trans_mutex); | |
219 | schedule(); | |
220 | mutex_lock(&root->fs_info->trans_mutex); | |
221 | } | |
222 | mutex_unlock(&root->fs_info->trans_mutex); | |
223 | finish_wait(&commit->commit_wait, &wait); | |
224 | return 0; | |
225 | } | |
226 | ||
d352ac68 | 227 | /* |
d397712b CM |
228 | * rate limit against the drop_snapshot code. This helps to slow down new |
229 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 230 | */ |
37d1aeee | 231 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
232 | { |
233 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 234 | int harder_count = 0; |
ab78c84d | 235 | |
2dd3e67b | 236 | harder: |
ab78c84d CM |
237 | if (atomic_read(&info->throttles)) { |
238 | DEFINE_WAIT(wait); | |
239 | int thr; | |
ab78c84d CM |
240 | thr = atomic_read(&info->throttle_gen); |
241 | ||
242 | do { | |
243 | prepare_to_wait(&info->transaction_throttle, | |
244 | &wait, TASK_UNINTERRUPTIBLE); | |
245 | if (!atomic_read(&info->throttles)) { | |
246 | finish_wait(&info->transaction_throttle, &wait); | |
247 | break; | |
248 | } | |
249 | schedule(); | |
250 | finish_wait(&info->transaction_throttle, &wait); | |
251 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
252 | harder_count++; |
253 | ||
254 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
255 | harder_count < 2) | |
256 | goto harder; | |
257 | ||
258 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
259 | harder_count < 10) | |
260 | goto harder; | |
261 | ||
262 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
263 | harder_count < 20) | |
264 | goto harder; | |
ab78c84d CM |
265 | } |
266 | } | |
267 | ||
37d1aeee CM |
268 | void btrfs_throttle(struct btrfs_root *root) |
269 | { | |
270 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
271 | if (!root->fs_info->open_ioctl_trans) |
272 | wait_current_trans(root); | |
37d1aeee CM |
273 | mutex_unlock(&root->fs_info->trans_mutex); |
274 | ||
275 | throttle_on_drops(root); | |
276 | } | |
277 | ||
89ce8a63 CM |
278 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
279 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
280 | { |
281 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 282 | struct btrfs_fs_info *info = root->fs_info; |
d6e4a428 | 283 | |
ab78c84d CM |
284 | mutex_lock(&info->trans_mutex); |
285 | cur_trans = info->running_transaction; | |
ccd467d6 | 286 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 287 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 288 | cur_trans->num_writers--; |
89ce8a63 | 289 | |
79154b1b CM |
290 | if (waitqueue_active(&cur_trans->writer_wait)) |
291 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 292 | put_transaction(cur_trans); |
ab78c84d | 293 | mutex_unlock(&info->trans_mutex); |
d6025579 | 294 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 295 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d CM |
296 | |
297 | if (throttle) | |
37d1aeee | 298 | throttle_on_drops(root); |
ab78c84d | 299 | |
79154b1b CM |
300 | return 0; |
301 | } | |
302 | ||
89ce8a63 CM |
303 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
304 | struct btrfs_root *root) | |
305 | { | |
306 | return __btrfs_end_transaction(trans, root, 0); | |
307 | } | |
308 | ||
309 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
310 | struct btrfs_root *root) | |
311 | { | |
312 | return __btrfs_end_transaction(trans, root, 1); | |
313 | } | |
314 | ||
d352ac68 CM |
315 | /* |
316 | * when btree blocks are allocated, they have some corresponding bits set for | |
317 | * them in one of two extent_io trees. This is used to make sure all of | |
318 | * those extents are on disk for transaction or log commit | |
319 | */ | |
d0c803c4 CM |
320 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
321 | struct extent_io_tree *dirty_pages) | |
79154b1b | 322 | { |
7c4452b9 | 323 | int ret; |
777e6bd7 | 324 | int err = 0; |
7c4452b9 CM |
325 | int werr = 0; |
326 | struct page *page; | |
7c4452b9 | 327 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 328 | u64 start = 0; |
5f39d397 CM |
329 | u64 end; |
330 | unsigned long index; | |
7c4452b9 | 331 | |
d397712b | 332 | while (1) { |
777e6bd7 | 333 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
334 | EXTENT_DIRTY); |
335 | if (ret) | |
7c4452b9 | 336 | break; |
d397712b | 337 | while (start <= end) { |
777e6bd7 CM |
338 | cond_resched(); |
339 | ||
5f39d397 | 340 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 341 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 342 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
343 | if (!page) |
344 | continue; | |
4bef0848 CM |
345 | |
346 | btree_lock_page_hook(page); | |
347 | if (!page->mapping) { | |
348 | unlock_page(page); | |
349 | page_cache_release(page); | |
350 | continue; | |
351 | } | |
352 | ||
6702ed49 CM |
353 | if (PageWriteback(page)) { |
354 | if (PageDirty(page)) | |
355 | wait_on_page_writeback(page); | |
356 | else { | |
357 | unlock_page(page); | |
358 | page_cache_release(page); | |
359 | continue; | |
360 | } | |
361 | } | |
7c4452b9 CM |
362 | err = write_one_page(page, 0); |
363 | if (err) | |
364 | werr = err; | |
365 | page_cache_release(page); | |
366 | } | |
367 | } | |
d397712b | 368 | while (1) { |
777e6bd7 CM |
369 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, |
370 | EXTENT_DIRTY); | |
371 | if (ret) | |
372 | break; | |
373 | ||
374 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
d397712b | 375 | while (start <= end) { |
777e6bd7 CM |
376 | index = start >> PAGE_CACHE_SHIFT; |
377 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
378 | page = find_get_page(btree_inode->i_mapping, index); | |
379 | if (!page) | |
380 | continue; | |
381 | if (PageDirty(page)) { | |
4bef0848 CM |
382 | btree_lock_page_hook(page); |
383 | wait_on_page_writeback(page); | |
777e6bd7 CM |
384 | err = write_one_page(page, 0); |
385 | if (err) | |
386 | werr = err; | |
387 | } | |
105d931d | 388 | wait_on_page_writeback(page); |
777e6bd7 CM |
389 | page_cache_release(page); |
390 | cond_resched(); | |
391 | } | |
392 | } | |
7c4452b9 CM |
393 | if (err) |
394 | werr = err; | |
395 | return werr; | |
79154b1b CM |
396 | } |
397 | ||
d0c803c4 CM |
398 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
399 | struct btrfs_root *root) | |
400 | { | |
401 | if (!trans || !trans->transaction) { | |
402 | struct inode *btree_inode; | |
403 | btree_inode = root->fs_info->btree_inode; | |
404 | return filemap_write_and_wait(btree_inode->i_mapping); | |
405 | } | |
406 | return btrfs_write_and_wait_marked_extents(root, | |
407 | &trans->transaction->dirty_pages); | |
408 | } | |
409 | ||
d352ac68 CM |
410 | /* |
411 | * this is used to update the root pointer in the tree of tree roots. | |
412 | * | |
413 | * But, in the case of the extent allocation tree, updating the root | |
414 | * pointer may allocate blocks which may change the root of the extent | |
415 | * allocation tree. | |
416 | * | |
417 | * So, this loops and repeats and makes sure the cowonly root didn't | |
418 | * change while the root pointer was being updated in the metadata. | |
419 | */ | |
0b86a832 CM |
420 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
421 | struct btrfs_root *root) | |
79154b1b CM |
422 | { |
423 | int ret; | |
0b86a832 CM |
424 | u64 old_root_bytenr; |
425 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 426 | |
87ef2bb4 | 427 | btrfs_extent_post_op(trans, root); |
0b86a832 | 428 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 CM |
429 | btrfs_extent_post_op(trans, root); |
430 | ||
d397712b | 431 | while (1) { |
0b86a832 CM |
432 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
433 | if (old_root_bytenr == root->node->start) | |
79154b1b | 434 | break; |
0b86a832 CM |
435 | btrfs_set_root_bytenr(&root->root_item, |
436 | root->node->start); | |
437 | btrfs_set_root_level(&root->root_item, | |
438 | btrfs_header_level(root->node)); | |
84234f3a | 439 | btrfs_set_root_generation(&root->root_item, trans->transid); |
87ef2bb4 CM |
440 | |
441 | btrfs_extent_post_op(trans, root); | |
442 | ||
79154b1b | 443 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
444 | &root->root_key, |
445 | &root->root_item); | |
79154b1b | 446 | BUG_ON(ret); |
0b86a832 | 447 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 | 448 | btrfs_extent_post_op(trans, root); |
0b86a832 CM |
449 | } |
450 | return 0; | |
451 | } | |
452 | ||
d352ac68 CM |
453 | /* |
454 | * update all the cowonly tree roots on disk | |
455 | */ | |
0b86a832 CM |
456 | int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans, |
457 | struct btrfs_root *root) | |
458 | { | |
459 | struct btrfs_fs_info *fs_info = root->fs_info; | |
460 | struct list_head *next; | |
84234f3a YZ |
461 | struct extent_buffer *eb; |
462 | ||
87ef2bb4 CM |
463 | btrfs_extent_post_op(trans, fs_info->tree_root); |
464 | ||
84234f3a YZ |
465 | eb = btrfs_lock_root_node(fs_info->tree_root); |
466 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0); | |
467 | btrfs_tree_unlock(eb); | |
468 | free_extent_buffer(eb); | |
0b86a832 | 469 | |
87ef2bb4 CM |
470 | btrfs_extent_post_op(trans, fs_info->tree_root); |
471 | ||
d397712b | 472 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
473 | next = fs_info->dirty_cowonly_roots.next; |
474 | list_del_init(next); | |
475 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 476 | |
0b86a832 | 477 | update_cowonly_root(trans, root); |
79154b1b CM |
478 | } |
479 | return 0; | |
480 | } | |
481 | ||
d352ac68 CM |
482 | /* |
483 | * dead roots are old snapshots that need to be deleted. This allocates | |
484 | * a dirty root struct and adds it into the list of dead roots that need to | |
485 | * be deleted | |
486 | */ | |
b48652c1 | 487 | int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest) |
5eda7b5e | 488 | { |
f321e491 | 489 | struct btrfs_dirty_root *dirty; |
5eda7b5e CM |
490 | |
491 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
492 | if (!dirty) | |
493 | return -ENOMEM; | |
5eda7b5e | 494 | dirty->root = root; |
5ce14bbc | 495 | dirty->latest_root = latest; |
b48652c1 YZ |
496 | |
497 | mutex_lock(&root->fs_info->trans_mutex); | |
498 | list_add(&dirty->list, &latest->fs_info->dead_roots); | |
499 | mutex_unlock(&root->fs_info->trans_mutex); | |
5eda7b5e CM |
500 | return 0; |
501 | } | |
502 | ||
d352ac68 CM |
503 | /* |
504 | * at transaction commit time we need to schedule the old roots for | |
505 | * deletion via btrfs_drop_snapshot. This runs through all the | |
506 | * reference counted roots that were modified in the current | |
507 | * transaction and puts them into the drop list | |
508 | */ | |
80b6794d CM |
509 | static noinline int add_dirty_roots(struct btrfs_trans_handle *trans, |
510 | struct radix_tree_root *radix, | |
511 | struct list_head *list) | |
0f7d52f4 | 512 | { |
f321e491 | 513 | struct btrfs_dirty_root *dirty; |
0f7d52f4 CM |
514 | struct btrfs_root *gang[8]; |
515 | struct btrfs_root *root; | |
516 | int i; | |
517 | int ret; | |
54aa1f4d | 518 | int err = 0; |
5eda7b5e | 519 | u32 refs; |
54aa1f4d | 520 | |
d397712b | 521 | while (1) { |
0f7d52f4 CM |
522 | ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0, |
523 | ARRAY_SIZE(gang), | |
524 | BTRFS_ROOT_TRANS_TAG); | |
525 | if (ret == 0) | |
526 | break; | |
527 | for (i = 0; i < ret; i++) { | |
528 | root = gang[i]; | |
2619ba1f CM |
529 | radix_tree_tag_clear(radix, |
530 | (unsigned long)root->root_key.objectid, | |
531 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
532 | |
533 | BUG_ON(!root->ref_tree); | |
017e5369 | 534 | dirty = root->dirty_root; |
31153d81 | 535 | |
e02119d5 | 536 | btrfs_free_log(trans, root); |
f82d02d9 | 537 | btrfs_free_reloc_root(trans, root); |
e02119d5 | 538 | |
0f7d52f4 | 539 | if (root->commit_root == root->node) { |
db94535d CM |
540 | WARN_ON(root->node->start != |
541 | btrfs_root_bytenr(&root->root_item)); | |
31153d81 | 542 | |
5f39d397 | 543 | free_extent_buffer(root->commit_root); |
0f7d52f4 | 544 | root->commit_root = NULL; |
7ea394f1 | 545 | root->dirty_root = NULL; |
bcc63abb Y |
546 | |
547 | spin_lock(&root->list_lock); | |
548 | list_del_init(&dirty->root->dead_list); | |
549 | spin_unlock(&root->list_lock); | |
550 | ||
31153d81 YZ |
551 | kfree(dirty->root); |
552 | kfree(dirty); | |
58176a96 JB |
553 | |
554 | /* make sure to update the root on disk | |
555 | * so we get any updates to the block used | |
556 | * counts | |
557 | */ | |
558 | err = btrfs_update_root(trans, | |
559 | root->fs_info->tree_root, | |
560 | &root->root_key, | |
561 | &root->root_item); | |
0f7d52f4 CM |
562 | continue; |
563 | } | |
9f3a7427 CM |
564 | |
565 | memset(&root->root_item.drop_progress, 0, | |
566 | sizeof(struct btrfs_disk_key)); | |
567 | root->root_item.drop_level = 0; | |
0f7d52f4 | 568 | root->commit_root = NULL; |
7ea394f1 | 569 | root->dirty_root = NULL; |
0f7d52f4 | 570 | root->root_key.offset = root->fs_info->generation; |
db94535d CM |
571 | btrfs_set_root_bytenr(&root->root_item, |
572 | root->node->start); | |
573 | btrfs_set_root_level(&root->root_item, | |
574 | btrfs_header_level(root->node)); | |
84234f3a YZ |
575 | btrfs_set_root_generation(&root->root_item, |
576 | root->root_key.offset); | |
577 | ||
0f7d52f4 CM |
578 | err = btrfs_insert_root(trans, root->fs_info->tree_root, |
579 | &root->root_key, | |
580 | &root->root_item); | |
54aa1f4d CM |
581 | if (err) |
582 | break; | |
9f3a7427 CM |
583 | |
584 | refs = btrfs_root_refs(&dirty->root->root_item); | |
585 | btrfs_set_root_refs(&dirty->root->root_item, refs - 1); | |
5eda7b5e | 586 | err = btrfs_update_root(trans, root->fs_info->tree_root, |
9f3a7427 CM |
587 | &dirty->root->root_key, |
588 | &dirty->root->root_item); | |
5eda7b5e CM |
589 | |
590 | BUG_ON(err); | |
9f3a7427 | 591 | if (refs == 1) { |
5eda7b5e | 592 | list_add(&dirty->list, list); |
9f3a7427 CM |
593 | } else { |
594 | WARN_ON(1); | |
31153d81 | 595 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 596 | kfree(dirty->root); |
5eda7b5e | 597 | kfree(dirty); |
9f3a7427 | 598 | } |
0f7d52f4 CM |
599 | } |
600 | } | |
54aa1f4d | 601 | return err; |
0f7d52f4 CM |
602 | } |
603 | ||
d352ac68 CM |
604 | /* |
605 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
606 | * otherwise every leaf in the btree is read and defragged. | |
607 | */ | |
e9d0b13b CM |
608 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
609 | { | |
610 | struct btrfs_fs_info *info = root->fs_info; | |
611 | int ret; | |
612 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 613 | unsigned long nr; |
e9d0b13b | 614 | |
a2135011 | 615 | smp_mb(); |
e9d0b13b CM |
616 | if (root->defrag_running) |
617 | return 0; | |
e9d0b13b | 618 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 619 | while (1) { |
e9d0b13b CM |
620 | root->defrag_running = 1; |
621 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 622 | nr = trans->blocks_used; |
e9d0b13b | 623 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 624 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
625 | cond_resched(); |
626 | ||
e9d0b13b | 627 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 628 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
629 | break; |
630 | } | |
631 | root->defrag_running = 0; | |
a2135011 | 632 | smp_mb(); |
e9d0b13b CM |
633 | btrfs_end_transaction(trans, root); |
634 | return 0; | |
635 | } | |
636 | ||
d352ac68 CM |
637 | /* |
638 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
639 | * all of them | |
640 | */ | |
80b6794d CM |
641 | static noinline int drop_dirty_roots(struct btrfs_root *tree_root, |
642 | struct list_head *list) | |
0f7d52f4 | 643 | { |
f321e491 | 644 | struct btrfs_dirty_root *dirty; |
0f7d52f4 | 645 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 646 | unsigned long nr; |
db94535d CM |
647 | u64 num_bytes; |
648 | u64 bytes_used; | |
bcc63abb | 649 | u64 max_useless; |
54aa1f4d | 650 | int ret = 0; |
9f3a7427 CM |
651 | int err; |
652 | ||
d397712b | 653 | while (!list_empty(list)) { |
58176a96 JB |
654 | struct btrfs_root *root; |
655 | ||
f321e491 | 656 | dirty = list_entry(list->prev, struct btrfs_dirty_root, list); |
0f7d52f4 | 657 | list_del_init(&dirty->list); |
5eda7b5e | 658 | |
db94535d | 659 | num_bytes = btrfs_root_used(&dirty->root->root_item); |
58176a96 | 660 | root = dirty->latest_root; |
a2135011 | 661 | atomic_inc(&root->fs_info->throttles); |
58176a96 | 662 | |
d397712b | 663 | while (1) { |
9f3a7427 | 664 | trans = btrfs_start_transaction(tree_root, 1); |
5b21f2ed | 665 | mutex_lock(&root->fs_info->drop_mutex); |
9f3a7427 | 666 | ret = btrfs_drop_snapshot(trans, dirty->root); |
d397712b | 667 | if (ret != -EAGAIN) |
9f3a7427 | 668 | break; |
5b21f2ed | 669 | mutex_unlock(&root->fs_info->drop_mutex); |
58176a96 | 670 | |
9f3a7427 CM |
671 | err = btrfs_update_root(trans, |
672 | tree_root, | |
673 | &dirty->root->root_key, | |
674 | &dirty->root->root_item); | |
675 | if (err) | |
676 | ret = err; | |
d3c2fdcf | 677 | nr = trans->blocks_used; |
017e5369 | 678 | ret = btrfs_end_transaction(trans, tree_root); |
9f3a7427 | 679 | BUG_ON(ret); |
a2135011 | 680 | |
d3c2fdcf | 681 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 682 | cond_resched(); |
9f3a7427 | 683 | } |
0f7d52f4 | 684 | BUG_ON(ret); |
a2135011 | 685 | atomic_dec(&root->fs_info->throttles); |
017e5369 | 686 | wake_up(&root->fs_info->transaction_throttle); |
58176a96 | 687 | |
db94535d CM |
688 | num_bytes -= btrfs_root_used(&dirty->root->root_item); |
689 | bytes_used = btrfs_root_used(&root->root_item); | |
690 | if (num_bytes) { | |
24562425 | 691 | mutex_lock(&root->fs_info->trans_mutex); |
e02119d5 | 692 | btrfs_record_root_in_trans(root); |
24562425 | 693 | mutex_unlock(&root->fs_info->trans_mutex); |
5f39d397 | 694 | btrfs_set_root_used(&root->root_item, |
db94535d | 695 | bytes_used - num_bytes); |
58176a96 | 696 | } |
a2135011 | 697 | |
9f3a7427 | 698 | ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key); |
58176a96 JB |
699 | if (ret) { |
700 | BUG(); | |
54aa1f4d | 701 | break; |
58176a96 | 702 | } |
a2135011 CM |
703 | mutex_unlock(&root->fs_info->drop_mutex); |
704 | ||
bcc63abb Y |
705 | spin_lock(&root->list_lock); |
706 | list_del_init(&dirty->root->dead_list); | |
707 | if (!list_empty(&root->dead_list)) { | |
708 | struct btrfs_root *oldest; | |
709 | oldest = list_entry(root->dead_list.prev, | |
710 | struct btrfs_root, dead_list); | |
711 | max_useless = oldest->root_key.offset - 1; | |
712 | } else { | |
713 | max_useless = root->root_key.offset - 1; | |
714 | } | |
715 | spin_unlock(&root->list_lock); | |
716 | ||
d3c2fdcf | 717 | nr = trans->blocks_used; |
0f7d52f4 CM |
718 | ret = btrfs_end_transaction(trans, tree_root); |
719 | BUG_ON(ret); | |
5eda7b5e | 720 | |
e4657689 | 721 | ret = btrfs_remove_leaf_refs(root, max_useless, 0); |
bcc63abb Y |
722 | BUG_ON(ret); |
723 | ||
f510cfec | 724 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 725 | kfree(dirty->root); |
0f7d52f4 | 726 | kfree(dirty); |
d3c2fdcf CM |
727 | |
728 | btrfs_btree_balance_dirty(tree_root, nr); | |
4dc11904 | 729 | cond_resched(); |
0f7d52f4 | 730 | } |
54aa1f4d | 731 | return ret; |
0f7d52f4 CM |
732 | } |
733 | ||
d352ac68 CM |
734 | /* |
735 | * new snapshots need to be created at a very specific time in the | |
736 | * transaction commit. This does the actual creation | |
737 | */ | |
80b6794d | 738 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
739 | struct btrfs_fs_info *fs_info, |
740 | struct btrfs_pending_snapshot *pending) | |
741 | { | |
742 | struct btrfs_key key; | |
80b6794d | 743 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
744 | struct btrfs_root *tree_root = fs_info->tree_root; |
745 | struct btrfs_root *root = pending->root; | |
746 | struct extent_buffer *tmp; | |
925baedd | 747 | struct extent_buffer *old; |
3063d29f CM |
748 | int ret; |
749 | u64 objectid; | |
750 | ||
80b6794d CM |
751 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
752 | if (!new_root_item) { | |
753 | ret = -ENOMEM; | |
754 | goto fail; | |
755 | } | |
3063d29f CM |
756 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
757 | if (ret) | |
758 | goto fail; | |
759 | ||
80ff3856 YZ |
760 | btrfs_record_root_in_trans(root); |
761 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
80b6794d | 762 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
763 | |
764 | key.objectid = objectid; | |
5b21f2ed | 765 | key.offset = trans->transid; |
3063d29f CM |
766 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
767 | ||
925baedd | 768 | old = btrfs_lock_root_node(root); |
65b51a00 | 769 | btrfs_cow_block(trans, root, old, NULL, 0, &old, 0); |
3063d29f | 770 | |
925baedd CM |
771 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
772 | btrfs_tree_unlock(old); | |
773 | free_extent_buffer(old); | |
3063d29f | 774 | |
80b6794d CM |
775 | btrfs_set_root_bytenr(new_root_item, tmp->start); |
776 | btrfs_set_root_level(new_root_item, btrfs_header_level(tmp)); | |
84234f3a | 777 | btrfs_set_root_generation(new_root_item, trans->transid); |
3063d29f | 778 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 779 | new_root_item); |
925baedd | 780 | btrfs_tree_unlock(tmp); |
3063d29f CM |
781 | free_extent_buffer(tmp); |
782 | if (ret) | |
783 | goto fail; | |
784 | ||
3de4586c CM |
785 | key.offset = (u64)-1; |
786 | memcpy(&pending->root_key, &key, sizeof(key)); | |
787 | fail: | |
788 | kfree(new_root_item); | |
789 | return ret; | |
790 | } | |
791 | ||
792 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
793 | struct btrfs_pending_snapshot *pending) | |
794 | { | |
795 | int ret; | |
796 | int namelen; | |
797 | u64 index = 0; | |
798 | struct btrfs_trans_handle *trans; | |
799 | struct inode *parent_inode; | |
800 | struct inode *inode; | |
0660b5af | 801 | struct btrfs_root *parent_root; |
3de4586c | 802 | |
3394e160 | 803 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 804 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 805 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 806 | |
3063d29f CM |
807 | /* |
808 | * insert the directory item | |
809 | */ | |
3b96362c | 810 | namelen = strlen(pending->name); |
3de4586c | 811 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 812 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
813 | pending->name, namelen, |
814 | parent_inode->i_ino, | |
815 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
816 | |
817 | if (ret) | |
818 | goto fail; | |
0660b5af | 819 | |
52c26179 YZ |
820 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
821 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
822 | BUG_ON(ret); | |
823 | ||
0660b5af CM |
824 | /* add the backref first */ |
825 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
826 | pending->root_key.objectid, | |
827 | BTRFS_ROOT_BACKREF_KEY, | |
828 | parent_root->root_key.objectid, | |
829 | parent_inode->i_ino, index, pending->name, | |
830 | namelen); | |
831 | ||
832 | BUG_ON(ret); | |
833 | ||
834 | /* now add the forward ref */ | |
835 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
836 | parent_root->root_key.objectid, | |
837 | BTRFS_ROOT_REF_KEY, | |
838 | pending->root_key.objectid, | |
839 | parent_inode->i_ino, index, pending->name, | |
840 | namelen); | |
841 | ||
3de4586c CM |
842 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
843 | d_instantiate(pending->dentry, inode); | |
3063d29f | 844 | fail: |
3de4586c | 845 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
846 | return ret; |
847 | } | |
848 | ||
d352ac68 CM |
849 | /* |
850 | * create all the snapshots we've scheduled for creation | |
851 | */ | |
80b6794d CM |
852 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
853 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
854 | { |
855 | struct btrfs_pending_snapshot *pending; | |
856 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
857 | int ret; |
858 | ||
c6e30871 | 859 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
860 | ret = create_pending_snapshot(trans, fs_info, pending); |
861 | BUG_ON(ret); | |
862 | } | |
863 | return 0; | |
864 | } | |
865 | ||
866 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
867 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
868 | { |
869 | struct btrfs_pending_snapshot *pending; | |
870 | struct list_head *head = &trans->transaction->pending_snapshots; | |
871 | int ret; | |
872 | ||
d397712b | 873 | while (!list_empty(head)) { |
3063d29f CM |
874 | pending = list_entry(head->next, |
875 | struct btrfs_pending_snapshot, list); | |
3de4586c | 876 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
877 | BUG_ON(ret); |
878 | list_del(&pending->list); | |
879 | kfree(pending->name); | |
880 | kfree(pending); | |
881 | } | |
dc17ff8f CM |
882 | return 0; |
883 | } | |
884 | ||
79154b1b CM |
885 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
886 | struct btrfs_root *root) | |
887 | { | |
15ee9bc7 JB |
888 | unsigned long joined = 0; |
889 | unsigned long timeout = 1; | |
79154b1b | 890 | struct btrfs_transaction *cur_trans; |
8fd17795 | 891 | struct btrfs_transaction *prev_trans = NULL; |
0b86a832 | 892 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; |
0f7d52f4 | 893 | struct list_head dirty_fs_roots; |
d1310b2e | 894 | struct extent_io_tree *pinned_copy; |
79154b1b | 895 | DEFINE_WAIT(wait); |
15ee9bc7 | 896 | int ret; |
79154b1b | 897 | |
0f7d52f4 | 898 | INIT_LIST_HEAD(&dirty_fs_roots); |
79154b1b CM |
899 | mutex_lock(&root->fs_info->trans_mutex); |
900 | if (trans->transaction->in_commit) { | |
901 | cur_trans = trans->transaction; | |
902 | trans->transaction->use_count++; | |
ccd467d6 | 903 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 904 | btrfs_end_transaction(trans, root); |
ccd467d6 | 905 | |
79154b1b CM |
906 | ret = wait_for_commit(root, cur_trans); |
907 | BUG_ON(ret); | |
15ee9bc7 JB |
908 | |
909 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 910 | put_transaction(cur_trans); |
15ee9bc7 JB |
911 | mutex_unlock(&root->fs_info->trans_mutex); |
912 | ||
79154b1b CM |
913 | return 0; |
914 | } | |
4313b399 CM |
915 | |
916 | pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS); | |
917 | if (!pinned_copy) | |
918 | return -ENOMEM; | |
919 | ||
d1310b2e | 920 | extent_io_tree_init(pinned_copy, |
4313b399 CM |
921 | root->fs_info->btree_inode->i_mapping, GFP_NOFS); |
922 | ||
2c90e5d6 | 923 | trans->transaction->in_commit = 1; |
f9295749 | 924 | trans->transaction->blocked = 1; |
ccd467d6 CM |
925 | cur_trans = trans->transaction; |
926 | if (cur_trans->list.prev != &root->fs_info->trans_list) { | |
927 | prev_trans = list_entry(cur_trans->list.prev, | |
928 | struct btrfs_transaction, list); | |
929 | if (!prev_trans->commit_done) { | |
930 | prev_trans->use_count++; | |
ccd467d6 CM |
931 | mutex_unlock(&root->fs_info->trans_mutex); |
932 | ||
933 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 934 | |
ccd467d6 | 935 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 936 | put_transaction(prev_trans); |
ccd467d6 CM |
937 | } |
938 | } | |
15ee9bc7 JB |
939 | |
940 | do { | |
7ea394f1 | 941 | int snap_pending = 0; |
15ee9bc7 | 942 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
943 | if (!list_empty(&trans->transaction->pending_snapshots)) |
944 | snap_pending = 1; | |
945 | ||
2c90e5d6 | 946 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 947 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 948 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
949 | |
950 | if (cur_trans->num_writers > 1) | |
951 | timeout = MAX_SCHEDULE_TIMEOUT; | |
952 | else | |
953 | timeout = 1; | |
954 | ||
79154b1b | 955 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 956 | |
7ea394f1 YZ |
957 | if (snap_pending) { |
958 | ret = btrfs_wait_ordered_extents(root, 1); | |
959 | BUG_ON(ret); | |
960 | } | |
961 | ||
15ee9bc7 JB |
962 | schedule_timeout(timeout); |
963 | ||
79154b1b | 964 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
965 | finish_wait(&cur_trans->writer_wait, &wait); |
966 | } while (cur_trans->num_writers > 1 || | |
967 | (cur_trans->num_joined != joined)); | |
968 | ||
3063d29f CM |
969 | ret = create_pending_snapshots(trans, root->fs_info); |
970 | BUG_ON(ret); | |
971 | ||
2c90e5d6 | 972 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 973 | |
e02119d5 CM |
974 | /* btrfs_commit_tree_roots is responsible for getting the |
975 | * various roots consistent with each other. Every pointer | |
976 | * in the tree of tree roots has to point to the most up to date | |
977 | * root for every subvolume and other tree. So, we have to keep | |
978 | * the tree logging code from jumping in and changing any | |
979 | * of the trees. | |
980 | * | |
981 | * At this point in the commit, there can't be any tree-log | |
982 | * writers, but a little lower down we drop the trans mutex | |
983 | * and let new people in. By holding the tree_log_mutex | |
984 | * from now until after the super is written, we avoid races | |
985 | * with the tree-log code. | |
986 | */ | |
987 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1a40e23b ZY |
988 | /* |
989 | * keep tree reloc code from adding new reloc trees | |
990 | */ | |
991 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
992 | ||
e02119d5 | 993 | |
54aa1f4d CM |
994 | ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix, |
995 | &dirty_fs_roots); | |
996 | BUG_ON(ret); | |
997 | ||
e02119d5 CM |
998 | /* add_dirty_roots gets rid of all the tree log roots, it is now |
999 | * safe to free the root of tree log roots | |
1000 | */ | |
1001 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1002 | ||
79154b1b CM |
1003 | ret = btrfs_commit_tree_roots(trans, root); |
1004 | BUG_ON(ret); | |
54aa1f4d | 1005 | |
78fae27e | 1006 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1007 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1008 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1009 | spin_unlock(&root->fs_info->new_trans_lock); |
4b52dff6 CM |
1010 | btrfs_set_super_generation(&root->fs_info->super_copy, |
1011 | cur_trans->transid); | |
1012 | btrfs_set_super_root(&root->fs_info->super_copy, | |
db94535d CM |
1013 | root->fs_info->tree_root->node->start); |
1014 | btrfs_set_super_root_level(&root->fs_info->super_copy, | |
1015 | btrfs_header_level(root->fs_info->tree_root->node)); | |
5f39d397 | 1016 | |
0b86a832 CM |
1017 | btrfs_set_super_chunk_root(&root->fs_info->super_copy, |
1018 | chunk_root->node->start); | |
1019 | btrfs_set_super_chunk_root_level(&root->fs_info->super_copy, | |
1020 | btrfs_header_level(chunk_root->node)); | |
84234f3a YZ |
1021 | btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy, |
1022 | btrfs_header_generation(chunk_root->node)); | |
e02119d5 CM |
1023 | |
1024 | if (!root->fs_info->log_root_recovering) { | |
1025 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1026 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1027 | } | |
1028 | ||
a061fc8d CM |
1029 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1030 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1031 | |
4313b399 | 1032 | btrfs_copy_pinned(root, pinned_copy); |
ccd467d6 | 1033 | |
f9295749 | 1034 | trans->transaction->blocked = 0; |
e6dcd2dc | 1035 | wake_up(&root->fs_info->transaction_throttle); |
f9295749 | 1036 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1037 | |
78fae27e | 1038 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1039 | ret = btrfs_write_and_wait_transaction(trans, root); |
1040 | BUG_ON(ret); | |
a512bbf8 | 1041 | write_ctree_super(trans, root, 0); |
4313b399 | 1042 | |
e02119d5 CM |
1043 | /* |
1044 | * the super is written, we can safely allow the tree-loggers | |
1045 | * to go about their business | |
1046 | */ | |
1047 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1048 | ||
4313b399 | 1049 | btrfs_finish_extent_commit(trans, root, pinned_copy); |
4313b399 CM |
1050 | kfree(pinned_copy); |
1051 | ||
1a40e23b ZY |
1052 | btrfs_drop_dead_reloc_roots(root); |
1053 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
1054 | ||
3de4586c CM |
1055 | /* do the directory inserts of any pending snapshot creations */ |
1056 | finish_pending_snapshots(trans, root->fs_info); | |
1057 | ||
1a40e23b ZY |
1058 | mutex_lock(&root->fs_info->trans_mutex); |
1059 | ||
2c90e5d6 | 1060 | cur_trans->commit_done = 1; |
15ee9bc7 | 1061 | root->fs_info->last_trans_committed = cur_trans->transid; |
2c90e5d6 | 1062 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1063 | |
78fae27e | 1064 | put_transaction(cur_trans); |
79154b1b | 1065 | put_transaction(cur_trans); |
58176a96 | 1066 | |
bcc63abb | 1067 | list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots); |
facda1e7 CM |
1068 | if (root->fs_info->closing) |
1069 | list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots); | |
58176a96 | 1070 | |
78fae27e | 1071 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1072 | |
2c90e5d6 | 1073 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b | 1074 | |
d397712b | 1075 | if (root->fs_info->closing) |
facda1e7 | 1076 | drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots); |
79154b1b CM |
1077 | return ret; |
1078 | } | |
1079 | ||
d352ac68 CM |
1080 | /* |
1081 | * interface function to delete all the snapshots we have scheduled for deletion | |
1082 | */ | |
e9d0b13b CM |
1083 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1084 | { | |
1085 | struct list_head dirty_roots; | |
1086 | INIT_LIST_HEAD(&dirty_roots); | |
a74a4b97 | 1087 | again: |
e9d0b13b CM |
1088 | mutex_lock(&root->fs_info->trans_mutex); |
1089 | list_splice_init(&root->fs_info->dead_roots, &dirty_roots); | |
1090 | mutex_unlock(&root->fs_info->trans_mutex); | |
1091 | ||
1092 | if (!list_empty(&dirty_roots)) { | |
1093 | drop_dirty_roots(root, &dirty_roots); | |
a74a4b97 | 1094 | goto again; |
e9d0b13b CM |
1095 | } |
1096 | return 0; | |
1097 | } |