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 | |
78fae27e | 31 | static int total_trans = 0; |
2c90e5d6 CM |
32 | extern struct kmem_cache *btrfs_trans_handle_cachep; |
33 | extern struct kmem_cache *btrfs_transaction_cachep; | |
34 | ||
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
80b6794d | 37 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 38 | { |
2c90e5d6 | 39 | WARN_ON(transaction->use_count == 0); |
79154b1b | 40 | transaction->use_count--; |
78fae27e CM |
41 | if (transaction->use_count == 0) { |
42 | WARN_ON(total_trans == 0); | |
43 | total_trans--; | |
8fd17795 | 44 | list_del_init(&transaction->list); |
2c90e5d6 CM |
45 | memset(transaction, 0, sizeof(*transaction)); |
46 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 47 | } |
79154b1b CM |
48 | } |
49 | ||
d352ac68 CM |
50 | /* |
51 | * either allocate a new transaction or hop into the existing one | |
52 | */ | |
80b6794d | 53 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
54 | { |
55 | struct btrfs_transaction *cur_trans; | |
56 | cur_trans = root->fs_info->running_transaction; | |
57 | if (!cur_trans) { | |
2c90e5d6 CM |
58 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
59 | GFP_NOFS); | |
78fae27e | 60 | total_trans++; |
79154b1b | 61 | BUG_ON(!cur_trans); |
0f7d52f4 | 62 | root->fs_info->generation++; |
e18e4809 | 63 | root->fs_info->last_alloc = 0; |
4529ba49 | 64 | root->fs_info->last_data_alloc = 0; |
15ee9bc7 JB |
65 | cur_trans->num_writers = 1; |
66 | cur_trans->num_joined = 0; | |
0f7d52f4 | 67 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
68 | init_waitqueue_head(&cur_trans->writer_wait); |
69 | init_waitqueue_head(&cur_trans->commit_wait); | |
70 | cur_trans->in_commit = 0; | |
f9295749 | 71 | cur_trans->blocked = 0; |
d5719762 | 72 | cur_trans->use_count = 1; |
79154b1b | 73 | cur_trans->commit_done = 0; |
08607c1b | 74 | cur_trans->start_time = get_seconds(); |
3063d29f | 75 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 76 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 77 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
78 | root->fs_info->btree_inode->i_mapping, |
79 | GFP_NOFS); | |
48ec2cf8 CM |
80 | spin_lock(&root->fs_info->new_trans_lock); |
81 | root->fs_info->running_transaction = cur_trans; | |
82 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
83 | } else { |
84 | cur_trans->num_writers++; | |
85 | cur_trans->num_joined++; | |
79154b1b | 86 | } |
15ee9bc7 | 87 | |
79154b1b CM |
88 | return 0; |
89 | } | |
90 | ||
d352ac68 CM |
91 | /* |
92 | * this does all the record keeping required to make sure that a | |
93 | * reference counted root is properly recorded in a given transaction. | |
94 | * This is required to make sure the old root from before we joined the transaction | |
95 | * is deleted when the transaction commits | |
96 | */ | |
e02119d5 | 97 | noinline int btrfs_record_root_in_trans(struct btrfs_root *root) |
6702ed49 | 98 | { |
f321e491 | 99 | struct btrfs_dirty_root *dirty; |
6702ed49 CM |
100 | u64 running_trans_id = root->fs_info->running_transaction->transid; |
101 | if (root->ref_cows && root->last_trans < running_trans_id) { | |
102 | WARN_ON(root == root->fs_info->extent_root); | |
103 | if (root->root_item.refs != 0) { | |
104 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
105 | (unsigned long)root->root_key.objectid, | |
106 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
107 | |
108 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
109 | BUG_ON(!dirty); | |
110 | dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS); | |
111 | BUG_ON(!dirty->root); | |
31153d81 YZ |
112 | dirty->latest_root = root; |
113 | INIT_LIST_HEAD(&dirty->list); | |
31153d81 | 114 | |
925baedd | 115 | root->commit_root = btrfs_root_node(root); |
31153d81 YZ |
116 | |
117 | memcpy(dirty->root, root, sizeof(*root)); | |
118 | spin_lock_init(&dirty->root->node_lock); | |
bcc63abb | 119 | spin_lock_init(&dirty->root->list_lock); |
31153d81 | 120 | mutex_init(&dirty->root->objectid_mutex); |
5b21f2ed | 121 | mutex_init(&dirty->root->log_mutex); |
bcc63abb | 122 | INIT_LIST_HEAD(&dirty->root->dead_list); |
31153d81 YZ |
123 | dirty->root->node = root->commit_root; |
124 | dirty->root->commit_root = NULL; | |
bcc63abb Y |
125 | |
126 | spin_lock(&root->list_lock); | |
127 | list_add(&dirty->root->dead_list, &root->dead_list); | |
128 | spin_unlock(&root->list_lock); | |
129 | ||
130 | root->dirty_root = dirty; | |
6702ed49 CM |
131 | } else { |
132 | WARN_ON(1); | |
133 | } | |
134 | root->last_trans = running_trans_id; | |
135 | } | |
136 | return 0; | |
137 | } | |
138 | ||
d352ac68 CM |
139 | /* wait for commit against the current transaction to become unblocked |
140 | * when this is done, it is safe to start a new transaction, but the current | |
141 | * transaction might not be fully on disk. | |
142 | */ | |
37d1aeee | 143 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 144 | { |
f9295749 | 145 | struct btrfs_transaction *cur_trans; |
79154b1b | 146 | |
f9295749 | 147 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 148 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
149 | DEFINE_WAIT(wait); |
150 | cur_trans->use_count++; | |
151 | while(1) { | |
152 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
153 | TASK_UNINTERRUPTIBLE); | |
154 | if (cur_trans->blocked) { | |
155 | mutex_unlock(&root->fs_info->trans_mutex); | |
156 | schedule(); | |
157 | mutex_lock(&root->fs_info->trans_mutex); | |
158 | finish_wait(&root->fs_info->transaction_wait, | |
159 | &wait); | |
160 | } else { | |
161 | finish_wait(&root->fs_info->transaction_wait, | |
162 | &wait); | |
163 | break; | |
164 | } | |
165 | } | |
166 | put_transaction(cur_trans); | |
167 | } | |
37d1aeee CM |
168 | } |
169 | ||
e02119d5 | 170 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
9ca9ee09 | 171 | int num_blocks, int wait) |
37d1aeee CM |
172 | { |
173 | struct btrfs_trans_handle *h = | |
174 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
175 | int ret; | |
176 | ||
177 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 CM |
178 | if (!root->fs_info->log_root_recovering && |
179 | ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2)) | |
37d1aeee | 180 | wait_current_trans(root); |
79154b1b CM |
181 | ret = join_transaction(root); |
182 | BUG_ON(ret); | |
0f7d52f4 | 183 | |
e02119d5 | 184 | btrfs_record_root_in_trans(root); |
6702ed49 | 185 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
186 | h->transaction = root->fs_info->running_transaction; |
187 | h->blocks_reserved = num_blocks; | |
188 | h->blocks_used = 0; | |
31f3c99b | 189 | h->block_group = NULL; |
26b8003f CM |
190 | h->alloc_exclude_nr = 0; |
191 | h->alloc_exclude_start = 0; | |
79154b1b CM |
192 | root->fs_info->running_transaction->use_count++; |
193 | mutex_unlock(&root->fs_info->trans_mutex); | |
194 | return h; | |
195 | } | |
196 | ||
f9295749 CM |
197 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
198 | int num_blocks) | |
199 | { | |
9ca9ee09 | 200 | return start_transaction(root, num_blocks, 1); |
f9295749 CM |
201 | } |
202 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
203 | int num_blocks) | |
204 | { | |
9ca9ee09 | 205 | return start_transaction(root, num_blocks, 0); |
f9295749 CM |
206 | } |
207 | ||
9ca9ee09 SW |
208 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
209 | int num_blocks) | |
210 | { | |
211 | return start_transaction(r, num_blocks, 2); | |
212 | } | |
213 | ||
d352ac68 | 214 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
215 | static noinline int wait_for_commit(struct btrfs_root *root, |
216 | struct btrfs_transaction *commit) | |
217 | { | |
218 | DEFINE_WAIT(wait); | |
219 | mutex_lock(&root->fs_info->trans_mutex); | |
220 | while(!commit->commit_done) { | |
221 | prepare_to_wait(&commit->commit_wait, &wait, | |
222 | TASK_UNINTERRUPTIBLE); | |
223 | if (commit->commit_done) | |
224 | break; | |
225 | mutex_unlock(&root->fs_info->trans_mutex); | |
226 | schedule(); | |
227 | mutex_lock(&root->fs_info->trans_mutex); | |
228 | } | |
229 | mutex_unlock(&root->fs_info->trans_mutex); | |
230 | finish_wait(&commit->commit_wait, &wait); | |
231 | return 0; | |
232 | } | |
233 | ||
d352ac68 CM |
234 | /* |
235 | * rate limit against the drop_snapshot code. This helps to slow down new operations | |
236 | * if the drop_snapshot code isn't able to keep up. | |
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 | } | |
274 | ||
37d1aeee CM |
275 | void btrfs_throttle(struct btrfs_root *root) |
276 | { | |
277 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
278 | if (!root->fs_info->open_ioctl_trans) |
279 | wait_current_trans(root); | |
37d1aeee CM |
280 | mutex_unlock(&root->fs_info->trans_mutex); |
281 | ||
282 | throttle_on_drops(root); | |
283 | } | |
284 | ||
89ce8a63 CM |
285 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
286 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
287 | { |
288 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 289 | struct btrfs_fs_info *info = root->fs_info; |
d6e4a428 | 290 | |
ab78c84d CM |
291 | mutex_lock(&info->trans_mutex); |
292 | cur_trans = info->running_transaction; | |
ccd467d6 | 293 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 294 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 295 | cur_trans->num_writers--; |
89ce8a63 | 296 | |
79154b1b CM |
297 | if (waitqueue_active(&cur_trans->writer_wait)) |
298 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 299 | put_transaction(cur_trans); |
ab78c84d | 300 | mutex_unlock(&info->trans_mutex); |
d6025579 | 301 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 302 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d CM |
303 | |
304 | if (throttle) | |
37d1aeee | 305 | throttle_on_drops(root); |
ab78c84d | 306 | |
79154b1b CM |
307 | return 0; |
308 | } | |
309 | ||
89ce8a63 CM |
310 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
311 | struct btrfs_root *root) | |
312 | { | |
313 | return __btrfs_end_transaction(trans, root, 0); | |
314 | } | |
315 | ||
316 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
317 | struct btrfs_root *root) | |
318 | { | |
319 | return __btrfs_end_transaction(trans, root, 1); | |
320 | } | |
321 | ||
d352ac68 CM |
322 | /* |
323 | * when btree blocks are allocated, they have some corresponding bits set for | |
324 | * them in one of two extent_io trees. This is used to make sure all of | |
325 | * those extents are on disk for transaction or log commit | |
326 | */ | |
d0c803c4 CM |
327 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
328 | struct extent_io_tree *dirty_pages) | |
79154b1b | 329 | { |
7c4452b9 | 330 | int ret; |
777e6bd7 | 331 | int err = 0; |
7c4452b9 CM |
332 | int werr = 0; |
333 | struct page *page; | |
7c4452b9 | 334 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 335 | u64 start = 0; |
5f39d397 CM |
336 | u64 end; |
337 | unsigned long index; | |
7c4452b9 | 338 | |
7c4452b9 | 339 | while(1) { |
777e6bd7 | 340 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
5f39d397 CM |
341 | EXTENT_DIRTY); |
342 | if (ret) | |
7c4452b9 | 343 | break; |
5f39d397 | 344 | while(start <= end) { |
777e6bd7 CM |
345 | cond_resched(); |
346 | ||
5f39d397 | 347 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 348 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 349 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
350 | if (!page) |
351 | continue; | |
4bef0848 CM |
352 | |
353 | btree_lock_page_hook(page); | |
354 | if (!page->mapping) { | |
355 | unlock_page(page); | |
356 | page_cache_release(page); | |
357 | continue; | |
358 | } | |
359 | ||
6702ed49 CM |
360 | if (PageWriteback(page)) { |
361 | if (PageDirty(page)) | |
362 | wait_on_page_writeback(page); | |
363 | else { | |
364 | unlock_page(page); | |
365 | page_cache_release(page); | |
366 | continue; | |
367 | } | |
368 | } | |
7c4452b9 CM |
369 | err = write_one_page(page, 0); |
370 | if (err) | |
371 | werr = err; | |
372 | page_cache_release(page); | |
373 | } | |
374 | } | |
777e6bd7 CM |
375 | while(1) { |
376 | ret = find_first_extent_bit(dirty_pages, 0, &start, &end, | |
377 | EXTENT_DIRTY); | |
378 | if (ret) | |
379 | break; | |
380 | ||
381 | clear_extent_dirty(dirty_pages, start, end, GFP_NOFS); | |
382 | while(start <= end) { | |
383 | index = start >> PAGE_CACHE_SHIFT; | |
384 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
385 | page = find_get_page(btree_inode->i_mapping, index); | |
386 | if (!page) | |
387 | continue; | |
388 | if (PageDirty(page)) { | |
4bef0848 CM |
389 | btree_lock_page_hook(page); |
390 | wait_on_page_writeback(page); | |
777e6bd7 CM |
391 | err = write_one_page(page, 0); |
392 | if (err) | |
393 | werr = err; | |
394 | } | |
105d931d | 395 | wait_on_page_writeback(page); |
777e6bd7 CM |
396 | page_cache_release(page); |
397 | cond_resched(); | |
398 | } | |
399 | } | |
7c4452b9 CM |
400 | if (err) |
401 | werr = err; | |
402 | return werr; | |
79154b1b CM |
403 | } |
404 | ||
d0c803c4 CM |
405 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
406 | struct btrfs_root *root) | |
407 | { | |
408 | if (!trans || !trans->transaction) { | |
409 | struct inode *btree_inode; | |
410 | btree_inode = root->fs_info->btree_inode; | |
411 | return filemap_write_and_wait(btree_inode->i_mapping); | |
412 | } | |
413 | return btrfs_write_and_wait_marked_extents(root, | |
414 | &trans->transaction->dirty_pages); | |
415 | } | |
416 | ||
d352ac68 CM |
417 | /* |
418 | * this is used to update the root pointer in the tree of tree roots. | |
419 | * | |
420 | * But, in the case of the extent allocation tree, updating the root | |
421 | * pointer may allocate blocks which may change the root of the extent | |
422 | * allocation tree. | |
423 | * | |
424 | * So, this loops and repeats and makes sure the cowonly root didn't | |
425 | * change while the root pointer was being updated in the metadata. | |
426 | */ | |
0b86a832 CM |
427 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
428 | struct btrfs_root *root) | |
79154b1b CM |
429 | { |
430 | int ret; | |
0b86a832 CM |
431 | u64 old_root_bytenr; |
432 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
79154b1b | 433 | |
87ef2bb4 | 434 | btrfs_extent_post_op(trans, root); |
0b86a832 | 435 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 CM |
436 | btrfs_extent_post_op(trans, root); |
437 | ||
79154b1b | 438 | while(1) { |
0b86a832 CM |
439 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
440 | if (old_root_bytenr == root->node->start) | |
79154b1b | 441 | break; |
0b86a832 CM |
442 | btrfs_set_root_bytenr(&root->root_item, |
443 | root->node->start); | |
444 | btrfs_set_root_level(&root->root_item, | |
445 | btrfs_header_level(root->node)); | |
84234f3a | 446 | btrfs_set_root_generation(&root->root_item, trans->transid); |
87ef2bb4 CM |
447 | |
448 | btrfs_extent_post_op(trans, root); | |
449 | ||
79154b1b | 450 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
451 | &root->root_key, |
452 | &root->root_item); | |
79154b1b | 453 | BUG_ON(ret); |
0b86a832 | 454 | btrfs_write_dirty_block_groups(trans, root); |
87ef2bb4 | 455 | btrfs_extent_post_op(trans, root); |
0b86a832 CM |
456 | } |
457 | return 0; | |
458 | } | |
459 | ||
d352ac68 CM |
460 | /* |
461 | * update all the cowonly tree roots on disk | |
462 | */ | |
0b86a832 CM |
463 | int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans, |
464 | struct btrfs_root *root) | |
465 | { | |
466 | struct btrfs_fs_info *fs_info = root->fs_info; | |
467 | struct list_head *next; | |
84234f3a YZ |
468 | struct extent_buffer *eb; |
469 | ||
87ef2bb4 CM |
470 | btrfs_extent_post_op(trans, fs_info->tree_root); |
471 | ||
84234f3a YZ |
472 | eb = btrfs_lock_root_node(fs_info->tree_root); |
473 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0); | |
474 | btrfs_tree_unlock(eb); | |
475 | free_extent_buffer(eb); | |
0b86a832 | 476 | |
87ef2bb4 CM |
477 | btrfs_extent_post_op(trans, fs_info->tree_root); |
478 | ||
0b86a832 CM |
479 | while(!list_empty(&fs_info->dirty_cowonly_roots)) { |
480 | next = fs_info->dirty_cowonly_roots.next; | |
481 | list_del_init(next); | |
482 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 483 | |
0b86a832 | 484 | update_cowonly_root(trans, root); |
79154b1b CM |
485 | } |
486 | return 0; | |
487 | } | |
488 | ||
d352ac68 CM |
489 | /* |
490 | * dead roots are old snapshots that need to be deleted. This allocates | |
491 | * a dirty root struct and adds it into the list of dead roots that need to | |
492 | * be deleted | |
493 | */ | |
b48652c1 | 494 | int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest) |
5eda7b5e | 495 | { |
f321e491 | 496 | struct btrfs_dirty_root *dirty; |
5eda7b5e CM |
497 | |
498 | dirty = kmalloc(sizeof(*dirty), GFP_NOFS); | |
499 | if (!dirty) | |
500 | return -ENOMEM; | |
5eda7b5e | 501 | dirty->root = root; |
5ce14bbc | 502 | dirty->latest_root = latest; |
b48652c1 YZ |
503 | |
504 | mutex_lock(&root->fs_info->trans_mutex); | |
505 | list_add(&dirty->list, &latest->fs_info->dead_roots); | |
506 | mutex_unlock(&root->fs_info->trans_mutex); | |
5eda7b5e CM |
507 | return 0; |
508 | } | |
509 | ||
d352ac68 CM |
510 | /* |
511 | * at transaction commit time we need to schedule the old roots for | |
512 | * deletion via btrfs_drop_snapshot. This runs through all the | |
513 | * reference counted roots that were modified in the current | |
514 | * transaction and puts them into the drop list | |
515 | */ | |
80b6794d CM |
516 | static noinline int add_dirty_roots(struct btrfs_trans_handle *trans, |
517 | struct radix_tree_root *radix, | |
518 | struct list_head *list) | |
0f7d52f4 | 519 | { |
f321e491 | 520 | struct btrfs_dirty_root *dirty; |
0f7d52f4 CM |
521 | struct btrfs_root *gang[8]; |
522 | struct btrfs_root *root; | |
523 | int i; | |
524 | int ret; | |
54aa1f4d | 525 | int err = 0; |
5eda7b5e | 526 | u32 refs; |
54aa1f4d | 527 | |
0f7d52f4 CM |
528 | while(1) { |
529 | ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0, | |
530 | ARRAY_SIZE(gang), | |
531 | BTRFS_ROOT_TRANS_TAG); | |
532 | if (ret == 0) | |
533 | break; | |
534 | for (i = 0; i < ret; i++) { | |
535 | root = gang[i]; | |
2619ba1f CM |
536 | radix_tree_tag_clear(radix, |
537 | (unsigned long)root->root_key.objectid, | |
538 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 YZ |
539 | |
540 | BUG_ON(!root->ref_tree); | |
017e5369 | 541 | dirty = root->dirty_root; |
31153d81 | 542 | |
e02119d5 | 543 | btrfs_free_log(trans, root); |
f82d02d9 | 544 | btrfs_free_reloc_root(trans, root); |
e02119d5 | 545 | |
0f7d52f4 | 546 | if (root->commit_root == root->node) { |
db94535d CM |
547 | WARN_ON(root->node->start != |
548 | btrfs_root_bytenr(&root->root_item)); | |
31153d81 | 549 | |
5f39d397 | 550 | free_extent_buffer(root->commit_root); |
0f7d52f4 | 551 | root->commit_root = NULL; |
7ea394f1 | 552 | root->dirty_root = NULL; |
bcc63abb Y |
553 | |
554 | spin_lock(&root->list_lock); | |
555 | list_del_init(&dirty->root->dead_list); | |
556 | spin_unlock(&root->list_lock); | |
557 | ||
31153d81 YZ |
558 | kfree(dirty->root); |
559 | kfree(dirty); | |
58176a96 JB |
560 | |
561 | /* make sure to update the root on disk | |
562 | * so we get any updates to the block used | |
563 | * counts | |
564 | */ | |
565 | err = btrfs_update_root(trans, | |
566 | root->fs_info->tree_root, | |
567 | &root->root_key, | |
568 | &root->root_item); | |
0f7d52f4 CM |
569 | continue; |
570 | } | |
9f3a7427 CM |
571 | |
572 | memset(&root->root_item.drop_progress, 0, | |
573 | sizeof(struct btrfs_disk_key)); | |
574 | root->root_item.drop_level = 0; | |
0f7d52f4 | 575 | root->commit_root = NULL; |
7ea394f1 | 576 | root->dirty_root = NULL; |
0f7d52f4 | 577 | root->root_key.offset = root->fs_info->generation; |
db94535d CM |
578 | btrfs_set_root_bytenr(&root->root_item, |
579 | root->node->start); | |
580 | btrfs_set_root_level(&root->root_item, | |
581 | btrfs_header_level(root->node)); | |
84234f3a YZ |
582 | btrfs_set_root_generation(&root->root_item, |
583 | root->root_key.offset); | |
584 | ||
0f7d52f4 CM |
585 | err = btrfs_insert_root(trans, root->fs_info->tree_root, |
586 | &root->root_key, | |
587 | &root->root_item); | |
54aa1f4d CM |
588 | if (err) |
589 | break; | |
9f3a7427 CM |
590 | |
591 | refs = btrfs_root_refs(&dirty->root->root_item); | |
592 | btrfs_set_root_refs(&dirty->root->root_item, refs - 1); | |
5eda7b5e | 593 | err = btrfs_update_root(trans, root->fs_info->tree_root, |
9f3a7427 CM |
594 | &dirty->root->root_key, |
595 | &dirty->root->root_item); | |
5eda7b5e CM |
596 | |
597 | BUG_ON(err); | |
9f3a7427 | 598 | if (refs == 1) { |
5eda7b5e | 599 | list_add(&dirty->list, list); |
9f3a7427 CM |
600 | } else { |
601 | WARN_ON(1); | |
31153d81 | 602 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 603 | kfree(dirty->root); |
5eda7b5e | 604 | kfree(dirty); |
9f3a7427 | 605 | } |
0f7d52f4 CM |
606 | } |
607 | } | |
54aa1f4d | 608 | return err; |
0f7d52f4 CM |
609 | } |
610 | ||
d352ac68 CM |
611 | /* |
612 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
613 | * otherwise every leaf in the btree is read and defragged. | |
614 | */ | |
e9d0b13b CM |
615 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
616 | { | |
617 | struct btrfs_fs_info *info = root->fs_info; | |
618 | int ret; | |
619 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 620 | unsigned long nr; |
e9d0b13b | 621 | |
a2135011 | 622 | smp_mb(); |
e9d0b13b CM |
623 | if (root->defrag_running) |
624 | return 0; | |
e9d0b13b | 625 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 626 | while (1) { |
e9d0b13b CM |
627 | root->defrag_running = 1; |
628 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 629 | nr = trans->blocks_used; |
e9d0b13b | 630 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 631 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
632 | cond_resched(); |
633 | ||
e9d0b13b | 634 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 635 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
636 | break; |
637 | } | |
638 | root->defrag_running = 0; | |
a2135011 | 639 | smp_mb(); |
e9d0b13b CM |
640 | btrfs_end_transaction(trans, root); |
641 | return 0; | |
642 | } | |
643 | ||
d352ac68 CM |
644 | /* |
645 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
646 | * all of them | |
647 | */ | |
80b6794d CM |
648 | static noinline int drop_dirty_roots(struct btrfs_root *tree_root, |
649 | struct list_head *list) | |
0f7d52f4 | 650 | { |
f321e491 | 651 | struct btrfs_dirty_root *dirty; |
0f7d52f4 | 652 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 653 | unsigned long nr; |
db94535d CM |
654 | u64 num_bytes; |
655 | u64 bytes_used; | |
bcc63abb | 656 | u64 max_useless; |
54aa1f4d | 657 | int ret = 0; |
9f3a7427 CM |
658 | int err; |
659 | ||
0f7d52f4 | 660 | while(!list_empty(list)) { |
58176a96 JB |
661 | struct btrfs_root *root; |
662 | ||
f321e491 | 663 | dirty = list_entry(list->prev, struct btrfs_dirty_root, list); |
0f7d52f4 | 664 | list_del_init(&dirty->list); |
5eda7b5e | 665 | |
db94535d | 666 | num_bytes = btrfs_root_used(&dirty->root->root_item); |
58176a96 | 667 | root = dirty->latest_root; |
a2135011 | 668 | atomic_inc(&root->fs_info->throttles); |
58176a96 | 669 | |
9f3a7427 CM |
670 | while(1) { |
671 | trans = btrfs_start_transaction(tree_root, 1); | |
5b21f2ed | 672 | mutex_lock(&root->fs_info->drop_mutex); |
9f3a7427 CM |
673 | ret = btrfs_drop_snapshot(trans, dirty->root); |
674 | if (ret != -EAGAIN) { | |
675 | break; | |
676 | } | |
5b21f2ed | 677 | mutex_unlock(&root->fs_info->drop_mutex); |
58176a96 | 678 | |
9f3a7427 CM |
679 | err = btrfs_update_root(trans, |
680 | tree_root, | |
681 | &dirty->root->root_key, | |
682 | &dirty->root->root_item); | |
683 | if (err) | |
684 | ret = err; | |
d3c2fdcf | 685 | nr = trans->blocks_used; |
017e5369 | 686 | ret = btrfs_end_transaction(trans, tree_root); |
9f3a7427 | 687 | BUG_ON(ret); |
a2135011 | 688 | |
d3c2fdcf | 689 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 690 | cond_resched(); |
9f3a7427 | 691 | } |
0f7d52f4 | 692 | BUG_ON(ret); |
a2135011 | 693 | atomic_dec(&root->fs_info->throttles); |
017e5369 | 694 | wake_up(&root->fs_info->transaction_throttle); |
58176a96 | 695 | |
db94535d CM |
696 | num_bytes -= btrfs_root_used(&dirty->root->root_item); |
697 | bytes_used = btrfs_root_used(&root->root_item); | |
698 | if (num_bytes) { | |
e02119d5 | 699 | btrfs_record_root_in_trans(root); |
5f39d397 | 700 | btrfs_set_root_used(&root->root_item, |
db94535d | 701 | bytes_used - num_bytes); |
58176a96 | 702 | } |
a2135011 | 703 | |
9f3a7427 | 704 | ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key); |
58176a96 JB |
705 | if (ret) { |
706 | BUG(); | |
54aa1f4d | 707 | break; |
58176a96 | 708 | } |
a2135011 CM |
709 | mutex_unlock(&root->fs_info->drop_mutex); |
710 | ||
bcc63abb Y |
711 | spin_lock(&root->list_lock); |
712 | list_del_init(&dirty->root->dead_list); | |
713 | if (!list_empty(&root->dead_list)) { | |
714 | struct btrfs_root *oldest; | |
715 | oldest = list_entry(root->dead_list.prev, | |
716 | struct btrfs_root, dead_list); | |
717 | max_useless = oldest->root_key.offset - 1; | |
718 | } else { | |
719 | max_useless = root->root_key.offset - 1; | |
720 | } | |
721 | spin_unlock(&root->list_lock); | |
722 | ||
d3c2fdcf | 723 | nr = trans->blocks_used; |
0f7d52f4 CM |
724 | ret = btrfs_end_transaction(trans, tree_root); |
725 | BUG_ON(ret); | |
5eda7b5e | 726 | |
e4657689 | 727 | ret = btrfs_remove_leaf_refs(root, max_useless, 0); |
bcc63abb Y |
728 | BUG_ON(ret); |
729 | ||
f510cfec | 730 | free_extent_buffer(dirty->root->node); |
9f3a7427 | 731 | kfree(dirty->root); |
0f7d52f4 | 732 | kfree(dirty); |
d3c2fdcf CM |
733 | |
734 | btrfs_btree_balance_dirty(tree_root, nr); | |
4dc11904 | 735 | cond_resched(); |
0f7d52f4 | 736 | } |
54aa1f4d | 737 | return ret; |
0f7d52f4 CM |
738 | } |
739 | ||
d352ac68 CM |
740 | /* |
741 | * new snapshots need to be created at a very specific time in the | |
742 | * transaction commit. This does the actual creation | |
743 | */ | |
80b6794d | 744 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
745 | struct btrfs_fs_info *fs_info, |
746 | struct btrfs_pending_snapshot *pending) | |
747 | { | |
748 | struct btrfs_key key; | |
80b6794d | 749 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
750 | struct btrfs_root *tree_root = fs_info->tree_root; |
751 | struct btrfs_root *root = pending->root; | |
752 | struct extent_buffer *tmp; | |
925baedd | 753 | struct extent_buffer *old; |
3063d29f CM |
754 | int ret; |
755 | u64 objectid; | |
756 | ||
80b6794d CM |
757 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
758 | if (!new_root_item) { | |
759 | ret = -ENOMEM; | |
760 | goto fail; | |
761 | } | |
3063d29f CM |
762 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
763 | if (ret) | |
764 | goto fail; | |
765 | ||
80ff3856 YZ |
766 | btrfs_record_root_in_trans(root); |
767 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
80b6794d | 768 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
769 | |
770 | key.objectid = objectid; | |
5b21f2ed | 771 | key.offset = trans->transid; |
3063d29f CM |
772 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
773 | ||
925baedd | 774 | old = btrfs_lock_root_node(root); |
65b51a00 | 775 | btrfs_cow_block(trans, root, old, NULL, 0, &old, 0); |
3063d29f | 776 | |
925baedd CM |
777 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
778 | btrfs_tree_unlock(old); | |
779 | free_extent_buffer(old); | |
3063d29f | 780 | |
80b6794d CM |
781 | btrfs_set_root_bytenr(new_root_item, tmp->start); |
782 | btrfs_set_root_level(new_root_item, btrfs_header_level(tmp)); | |
84234f3a | 783 | btrfs_set_root_generation(new_root_item, trans->transid); |
3063d29f | 784 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 785 | new_root_item); |
925baedd | 786 | btrfs_tree_unlock(tmp); |
3063d29f CM |
787 | free_extent_buffer(tmp); |
788 | if (ret) | |
789 | goto fail; | |
790 | ||
3de4586c CM |
791 | key.offset = (u64)-1; |
792 | memcpy(&pending->root_key, &key, sizeof(key)); | |
793 | fail: | |
794 | kfree(new_root_item); | |
795 | return ret; | |
796 | } | |
797 | ||
798 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
799 | struct btrfs_pending_snapshot *pending) | |
800 | { | |
801 | int ret; | |
802 | int namelen; | |
803 | u64 index = 0; | |
804 | struct btrfs_trans_handle *trans; | |
805 | struct inode *parent_inode; | |
806 | struct inode *inode; | |
0660b5af | 807 | struct btrfs_root *parent_root; |
3de4586c | 808 | |
3394e160 | 809 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af CM |
810 | parent_root = BTRFS_I(parent_inode)->root; |
811 | trans = btrfs_start_transaction(parent_root, 1); | |
3de4586c | 812 | |
3063d29f CM |
813 | /* |
814 | * insert the directory item | |
815 | */ | |
3b96362c | 816 | namelen = strlen(pending->name); |
3de4586c | 817 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 818 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
819 | pending->name, namelen, |
820 | parent_inode->i_ino, | |
821 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
822 | |
823 | if (ret) | |
824 | goto fail; | |
0660b5af CM |
825 | |
826 | /* add the backref first */ | |
827 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
828 | pending->root_key.objectid, | |
829 | BTRFS_ROOT_BACKREF_KEY, | |
830 | parent_root->root_key.objectid, | |
831 | parent_inode->i_ino, index, pending->name, | |
832 | namelen); | |
833 | ||
834 | BUG_ON(ret); | |
835 | ||
836 | /* now add the forward ref */ | |
837 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, | |
838 | parent_root->root_key.objectid, | |
839 | BTRFS_ROOT_REF_KEY, | |
840 | pending->root_key.objectid, | |
841 | parent_inode->i_ino, index, pending->name, | |
842 | namelen); | |
843 | ||
3de4586c CM |
844 | inode = btrfs_lookup_dentry(parent_inode, pending->dentry); |
845 | d_instantiate(pending->dentry, inode); | |
3063d29f | 846 | fail: |
3de4586c | 847 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
848 | return ret; |
849 | } | |
850 | ||
d352ac68 CM |
851 | /* |
852 | * create all the snapshots we've scheduled for creation | |
853 | */ | |
80b6794d CM |
854 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
855 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
856 | { |
857 | struct btrfs_pending_snapshot *pending; | |
858 | struct list_head *head = &trans->transaction->pending_snapshots; | |
859 | struct list_head *cur; | |
860 | int ret; | |
861 | ||
862 | list_for_each(cur, head) { | |
863 | pending = list_entry(cur, struct btrfs_pending_snapshot, list); | |
864 | ret = create_pending_snapshot(trans, fs_info, pending); | |
865 | BUG_ON(ret); | |
866 | } | |
867 | return 0; | |
868 | } | |
869 | ||
870 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
871 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
872 | { |
873 | struct btrfs_pending_snapshot *pending; | |
874 | struct list_head *head = &trans->transaction->pending_snapshots; | |
875 | int ret; | |
876 | ||
877 | while(!list_empty(head)) { | |
878 | pending = list_entry(head->next, | |
879 | struct btrfs_pending_snapshot, list); | |
3de4586c | 880 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
881 | BUG_ON(ret); |
882 | list_del(&pending->list); | |
883 | kfree(pending->name); | |
884 | kfree(pending); | |
885 | } | |
dc17ff8f CM |
886 | return 0; |
887 | } | |
888 | ||
79154b1b CM |
889 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
890 | struct btrfs_root *root) | |
891 | { | |
15ee9bc7 JB |
892 | unsigned long joined = 0; |
893 | unsigned long timeout = 1; | |
79154b1b | 894 | struct btrfs_transaction *cur_trans; |
8fd17795 | 895 | struct btrfs_transaction *prev_trans = NULL; |
0b86a832 | 896 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; |
0f7d52f4 | 897 | struct list_head dirty_fs_roots; |
d1310b2e | 898 | struct extent_io_tree *pinned_copy; |
79154b1b | 899 | DEFINE_WAIT(wait); |
15ee9bc7 | 900 | int ret; |
79154b1b | 901 | |
0f7d52f4 | 902 | INIT_LIST_HEAD(&dirty_fs_roots); |
79154b1b CM |
903 | mutex_lock(&root->fs_info->trans_mutex); |
904 | if (trans->transaction->in_commit) { | |
905 | cur_trans = trans->transaction; | |
906 | trans->transaction->use_count++; | |
ccd467d6 | 907 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 908 | btrfs_end_transaction(trans, root); |
ccd467d6 | 909 | |
79154b1b CM |
910 | ret = wait_for_commit(root, cur_trans); |
911 | BUG_ON(ret); | |
15ee9bc7 JB |
912 | |
913 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 914 | put_transaction(cur_trans); |
15ee9bc7 JB |
915 | mutex_unlock(&root->fs_info->trans_mutex); |
916 | ||
79154b1b CM |
917 | return 0; |
918 | } | |
4313b399 CM |
919 | |
920 | pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS); | |
921 | if (!pinned_copy) | |
922 | return -ENOMEM; | |
923 | ||
d1310b2e | 924 | extent_io_tree_init(pinned_copy, |
4313b399 CM |
925 | root->fs_info->btree_inode->i_mapping, GFP_NOFS); |
926 | ||
2c90e5d6 | 927 | trans->transaction->in_commit = 1; |
f9295749 | 928 | trans->transaction->blocked = 1; |
ccd467d6 CM |
929 | cur_trans = trans->transaction; |
930 | if (cur_trans->list.prev != &root->fs_info->trans_list) { | |
931 | prev_trans = list_entry(cur_trans->list.prev, | |
932 | struct btrfs_transaction, list); | |
933 | if (!prev_trans->commit_done) { | |
934 | prev_trans->use_count++; | |
ccd467d6 CM |
935 | mutex_unlock(&root->fs_info->trans_mutex); |
936 | ||
937 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 938 | |
ccd467d6 | 939 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 940 | put_transaction(prev_trans); |
ccd467d6 CM |
941 | } |
942 | } | |
15ee9bc7 JB |
943 | |
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; | |
956 | else | |
957 | timeout = 1; | |
958 | ||
79154b1b | 959 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 960 | |
7ea394f1 YZ |
961 | if (snap_pending) { |
962 | ret = btrfs_wait_ordered_extents(root, 1); | |
963 | BUG_ON(ret); | |
964 | } | |
965 | ||
15ee9bc7 JB |
966 | schedule_timeout(timeout); |
967 | ||
79154b1b | 968 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
969 | finish_wait(&cur_trans->writer_wait, &wait); |
970 | } while (cur_trans->num_writers > 1 || | |
971 | (cur_trans->num_joined != joined)); | |
972 | ||
3063d29f CM |
973 | ret = create_pending_snapshots(trans, root->fs_info); |
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); | |
1a40e23b ZY |
992 | /* |
993 | * keep tree reloc code from adding new reloc trees | |
994 | */ | |
995 | mutex_lock(&root->fs_info->tree_reloc_mutex); | |
996 | ||
e02119d5 | 997 | |
54aa1f4d CM |
998 | ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix, |
999 | &dirty_fs_roots); | |
1000 | BUG_ON(ret); | |
1001 | ||
e02119d5 CM |
1002 | /* add_dirty_roots gets rid of all the tree log roots, it is now |
1003 | * safe to free the root of tree log roots | |
1004 | */ | |
1005 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1006 | ||
79154b1b CM |
1007 | ret = btrfs_commit_tree_roots(trans, root); |
1008 | BUG_ON(ret); | |
54aa1f4d | 1009 | |
78fae27e | 1010 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1011 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1012 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1013 | spin_unlock(&root->fs_info->new_trans_lock); |
4b52dff6 CM |
1014 | btrfs_set_super_generation(&root->fs_info->super_copy, |
1015 | cur_trans->transid); | |
1016 | btrfs_set_super_root(&root->fs_info->super_copy, | |
db94535d CM |
1017 | root->fs_info->tree_root->node->start); |
1018 | btrfs_set_super_root_level(&root->fs_info->super_copy, | |
1019 | btrfs_header_level(root->fs_info->tree_root->node)); | |
5f39d397 | 1020 | |
0b86a832 CM |
1021 | btrfs_set_super_chunk_root(&root->fs_info->super_copy, |
1022 | chunk_root->node->start); | |
1023 | btrfs_set_super_chunk_root_level(&root->fs_info->super_copy, | |
1024 | btrfs_header_level(chunk_root->node)); | |
84234f3a YZ |
1025 | btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy, |
1026 | btrfs_header_generation(chunk_root->node)); | |
e02119d5 CM |
1027 | |
1028 | if (!root->fs_info->log_root_recovering) { | |
1029 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1030 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1031 | } | |
1032 | ||
a061fc8d CM |
1033 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1034 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1035 | |
4313b399 | 1036 | btrfs_copy_pinned(root, pinned_copy); |
ccd467d6 | 1037 | |
f9295749 | 1038 | trans->transaction->blocked = 0; |
e6dcd2dc | 1039 | wake_up(&root->fs_info->transaction_throttle); |
f9295749 | 1040 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1041 | |
78fae27e | 1042 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1043 | ret = btrfs_write_and_wait_transaction(trans, root); |
1044 | BUG_ON(ret); | |
79154b1b | 1045 | write_ctree_super(trans, root); |
4313b399 | 1046 | |
e02119d5 CM |
1047 | /* |
1048 | * the super is written, we can safely allow the tree-loggers | |
1049 | * to go about their business | |
1050 | */ | |
1051 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1052 | ||
4313b399 | 1053 | btrfs_finish_extent_commit(trans, root, pinned_copy); |
4313b399 CM |
1054 | kfree(pinned_copy); |
1055 | ||
1a40e23b ZY |
1056 | btrfs_drop_dead_reloc_roots(root); |
1057 | mutex_unlock(&root->fs_info->tree_reloc_mutex); | |
1058 | ||
3de4586c CM |
1059 | /* do the directory inserts of any pending snapshot creations */ |
1060 | finish_pending_snapshots(trans, root->fs_info); | |
1061 | ||
1a40e23b ZY |
1062 | mutex_lock(&root->fs_info->trans_mutex); |
1063 | ||
2c90e5d6 | 1064 | cur_trans->commit_done = 1; |
15ee9bc7 | 1065 | root->fs_info->last_trans_committed = cur_trans->transid; |
2c90e5d6 | 1066 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1067 | |
78fae27e | 1068 | put_transaction(cur_trans); |
79154b1b | 1069 | put_transaction(cur_trans); |
58176a96 | 1070 | |
bcc63abb | 1071 | list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots); |
facda1e7 CM |
1072 | if (root->fs_info->closing) |
1073 | list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots); | |
58176a96 | 1074 | |
78fae27e | 1075 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1076 | |
2c90e5d6 | 1077 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
79154b1b | 1078 | |
facda1e7 | 1079 | if (root->fs_info->closing) { |
facda1e7 | 1080 | drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots); |
facda1e7 | 1081 | } |
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 | { | |
1090 | struct list_head dirty_roots; | |
1091 | INIT_LIST_HEAD(&dirty_roots); | |
a74a4b97 | 1092 | again: |
e9d0b13b CM |
1093 | mutex_lock(&root->fs_info->trans_mutex); |
1094 | list_splice_init(&root->fs_info->dead_roots, &dirty_roots); | |
1095 | mutex_unlock(&root->fs_info->trans_mutex); | |
1096 | ||
1097 | if (!list_empty(&dirty_roots)) { | |
1098 | drop_dirty_roots(root, &dirty_roots); | |
a74a4b97 | 1099 | goto again; |
e9d0b13b CM |
1100 | } |
1101 | return 0; | |
1102 | } |