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> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
79154b1b CM |
25 | #include "ctree.h" |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
925baedd | 28 | #include "locking.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 | ||
817d52f8 JB |
44 | static noinline void switch_commit_root(struct btrfs_root *root) |
45 | { | |
817d52f8 JB |
46 | free_extent_buffer(root->commit_root); |
47 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
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); | |
79154b1b | 60 | BUG_ON(!cur_trans); |
0f7d52f4 | 61 | root->fs_info->generation++; |
15ee9bc7 JB |
62 | cur_trans->num_writers = 1; |
63 | cur_trans->num_joined = 0; | |
0f7d52f4 | 64 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
65 | init_waitqueue_head(&cur_trans->writer_wait); |
66 | init_waitqueue_head(&cur_trans->commit_wait); | |
67 | cur_trans->in_commit = 0; | |
f9295749 | 68 | cur_trans->blocked = 0; |
d5719762 | 69 | cur_trans->use_count = 1; |
79154b1b | 70 | cur_trans->commit_done = 0; |
08607c1b | 71 | cur_trans->start_time = get_seconds(); |
56bec294 | 72 | |
6bef4d31 | 73 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 74 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
75 | cur_trans->delayed_refs.num_heads_ready = 0; |
76 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 77 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 78 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
79 | spin_lock_init(&cur_trans->delayed_refs.lock); |
80 | ||
3063d29f | 81 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 82 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 83 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
84 | root->fs_info->btree_inode->i_mapping, |
85 | GFP_NOFS); | |
48ec2cf8 CM |
86 | spin_lock(&root->fs_info->new_trans_lock); |
87 | root->fs_info->running_transaction = cur_trans; | |
88 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
89 | } else { |
90 | cur_trans->num_writers++; | |
91 | cur_trans->num_joined++; | |
79154b1b | 92 | } |
15ee9bc7 | 93 | |
79154b1b CM |
94 | return 0; |
95 | } | |
96 | ||
d352ac68 | 97 | /* |
d397712b CM |
98 | * this does all the record keeping required to make sure that a reference |
99 | * counted root is properly recorded in a given transaction. This is required | |
100 | * to make sure the old root from before we joined the transaction is deleted | |
101 | * when the transaction commits | |
d352ac68 | 102 | */ |
5d4f98a2 YZ |
103 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root) | |
6702ed49 | 105 | { |
5d4f98a2 | 106 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 107 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
108 | WARN_ON(root->commit_root != root->node); |
109 | ||
110 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
111 | (unsigned long)root->root_key.objectid, | |
112 | BTRFS_ROOT_TRANS_TAG); | |
113 | root->last_trans = trans->transid; | |
114 | btrfs_init_reloc_root(trans, root); | |
115 | } | |
116 | return 0; | |
117 | } | |
bcc63abb | 118 | |
5d4f98a2 YZ |
119 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
120 | struct btrfs_root *root) | |
121 | { | |
122 | if (!root->ref_cows) | |
123 | return 0; | |
bcc63abb | 124 | |
5d4f98a2 YZ |
125 | mutex_lock(&root->fs_info->trans_mutex); |
126 | if (root->last_trans == trans->transid) { | |
127 | mutex_unlock(&root->fs_info->trans_mutex); | |
128 | return 0; | |
6702ed49 | 129 | } |
5d4f98a2 YZ |
130 | |
131 | record_root_in_trans(trans, root); | |
132 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
133 | return 0; |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* wait for commit against the current transaction to become unblocked |
137 | * when this is done, it is safe to start a new transaction, but the current | |
138 | * transaction might not be fully on disk. | |
139 | */ | |
37d1aeee | 140 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 141 | { |
f9295749 | 142 | struct btrfs_transaction *cur_trans; |
79154b1b | 143 | |
f9295749 | 144 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 145 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
146 | DEFINE_WAIT(wait); |
147 | cur_trans->use_count++; | |
d397712b | 148 | while (1) { |
f9295749 CM |
149 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
150 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 151 | if (!cur_trans->blocked) |
f9295749 | 152 | break; |
471fa17d ZL |
153 | mutex_unlock(&root->fs_info->trans_mutex); |
154 | schedule(); | |
155 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 156 | } |
471fa17d | 157 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
158 | put_transaction(cur_trans); |
159 | } | |
37d1aeee CM |
160 | } |
161 | ||
249ac1e5 JB |
162 | enum btrfs_trans_type { |
163 | TRANS_START, | |
164 | TRANS_JOIN, | |
165 | TRANS_USERSPACE, | |
166 | }; | |
167 | ||
e02119d5 | 168 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
249ac1e5 | 169 | int num_blocks, int type) |
37d1aeee CM |
170 | { |
171 | struct btrfs_trans_handle *h = | |
172 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
173 | int ret; | |
174 | ||
175 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 | 176 | if (!root->fs_info->log_root_recovering && |
249ac1e5 JB |
177 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || |
178 | type == TRANS_USERSPACE)) | |
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 | |
249ac1e5 | 192 | if (!current->journal_info && type != TRANS_USERSPACE) |
9ed74f2d JB |
193 | current->journal_info = h; |
194 | ||
79154b1b | 195 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 196 | record_root_in_trans(h, root); |
79154b1b CM |
197 | mutex_unlock(&root->fs_info->trans_mutex); |
198 | return h; | |
199 | } | |
200 | ||
f9295749 CM |
201 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
202 | int num_blocks) | |
203 | { | |
249ac1e5 | 204 | return start_transaction(root, num_blocks, TRANS_START); |
f9295749 CM |
205 | } |
206 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
207 | int num_blocks) | |
208 | { | |
249ac1e5 | 209 | return start_transaction(root, num_blocks, TRANS_JOIN); |
f9295749 CM |
210 | } |
211 | ||
9ca9ee09 SW |
212 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
213 | int num_blocks) | |
214 | { | |
249ac1e5 | 215 | return start_transaction(r, num_blocks, TRANS_USERSPACE); |
9ca9ee09 SW |
216 | } |
217 | ||
d352ac68 | 218 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
219 | static noinline int wait_for_commit(struct btrfs_root *root, |
220 | struct btrfs_transaction *commit) | |
221 | { | |
222 | DEFINE_WAIT(wait); | |
223 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 224 | while (!commit->commit_done) { |
89ce8a63 CM |
225 | prepare_to_wait(&commit->commit_wait, &wait, |
226 | TASK_UNINTERRUPTIBLE); | |
227 | if (commit->commit_done) | |
228 | break; | |
229 | mutex_unlock(&root->fs_info->trans_mutex); | |
230 | schedule(); | |
231 | mutex_lock(&root->fs_info->trans_mutex); | |
232 | } | |
233 | mutex_unlock(&root->fs_info->trans_mutex); | |
234 | finish_wait(&commit->commit_wait, &wait); | |
235 | return 0; | |
236 | } | |
237 | ||
5d4f98a2 | 238 | #if 0 |
d352ac68 | 239 | /* |
d397712b CM |
240 | * rate limit against the drop_snapshot code. This helps to slow down new |
241 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 242 | */ |
37d1aeee | 243 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
244 | { |
245 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 246 | int harder_count = 0; |
ab78c84d | 247 | |
2dd3e67b | 248 | harder: |
ab78c84d CM |
249 | if (atomic_read(&info->throttles)) { |
250 | DEFINE_WAIT(wait); | |
251 | int thr; | |
ab78c84d CM |
252 | thr = atomic_read(&info->throttle_gen); |
253 | ||
254 | do { | |
255 | prepare_to_wait(&info->transaction_throttle, | |
256 | &wait, TASK_UNINTERRUPTIBLE); | |
257 | if (!atomic_read(&info->throttles)) { | |
258 | finish_wait(&info->transaction_throttle, &wait); | |
259 | break; | |
260 | } | |
261 | schedule(); | |
262 | finish_wait(&info->transaction_throttle, &wait); | |
263 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
264 | harder_count++; |
265 | ||
266 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
267 | harder_count < 2) | |
268 | goto harder; | |
269 | ||
270 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
271 | harder_count < 10) | |
272 | goto harder; | |
273 | ||
274 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
275 | harder_count < 20) | |
276 | goto harder; | |
ab78c84d CM |
277 | } |
278 | } | |
5d4f98a2 | 279 | #endif |
ab78c84d | 280 | |
37d1aeee CM |
281 | void btrfs_throttle(struct btrfs_root *root) |
282 | { | |
283 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
284 | if (!root->fs_info->open_ioctl_trans) |
285 | wait_current_trans(root); | |
37d1aeee | 286 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
287 | } |
288 | ||
89ce8a63 CM |
289 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
290 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
291 | { |
292 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 293 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
294 | int count = 0; |
295 | ||
296 | while (count < 4) { | |
297 | unsigned long cur = trans->delayed_ref_updates; | |
298 | trans->delayed_ref_updates = 0; | |
299 | if (cur && | |
300 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
301 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
302 | |
303 | /* | |
304 | * do a full flush if the transaction is trying | |
305 | * to close | |
306 | */ | |
307 | if (trans->transaction->delayed_refs.flushing) | |
308 | cur = 0; | |
c3e69d58 CM |
309 | btrfs_run_delayed_refs(trans, root, cur); |
310 | } else { | |
311 | break; | |
312 | } | |
313 | count++; | |
56bec294 CM |
314 | } |
315 | ||
ab78c84d CM |
316 | mutex_lock(&info->trans_mutex); |
317 | cur_trans = info->running_transaction; | |
ccd467d6 | 318 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 319 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 320 | cur_trans->num_writers--; |
89ce8a63 | 321 | |
79154b1b CM |
322 | if (waitqueue_active(&cur_trans->writer_wait)) |
323 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 324 | put_transaction(cur_trans); |
ab78c84d | 325 | mutex_unlock(&info->trans_mutex); |
9ed74f2d JB |
326 | |
327 | if (current->journal_info == trans) | |
328 | current->journal_info = NULL; | |
d6025579 | 329 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 330 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 331 | |
24bbcf04 YZ |
332 | if (throttle) |
333 | btrfs_run_delayed_iputs(root); | |
334 | ||
79154b1b CM |
335 | return 0; |
336 | } | |
337 | ||
89ce8a63 CM |
338 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
339 | struct btrfs_root *root) | |
340 | { | |
341 | return __btrfs_end_transaction(trans, root, 0); | |
342 | } | |
343 | ||
344 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
345 | struct btrfs_root *root) | |
346 | { | |
347 | return __btrfs_end_transaction(trans, root, 1); | |
348 | } | |
349 | ||
d352ac68 CM |
350 | /* |
351 | * when btree blocks are allocated, they have some corresponding bits set for | |
352 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 353 | * those extents are sent to disk but does not wait on them |
d352ac68 | 354 | */ |
690587d1 | 355 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 356 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 357 | { |
7c4452b9 | 358 | int ret; |
777e6bd7 | 359 | int err = 0; |
7c4452b9 CM |
360 | int werr = 0; |
361 | struct page *page; | |
7c4452b9 | 362 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 363 | u64 start = 0; |
5f39d397 CM |
364 | u64 end; |
365 | unsigned long index; | |
7c4452b9 | 366 | |
d397712b | 367 | while (1) { |
777e6bd7 | 368 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 369 | mark); |
5f39d397 | 370 | if (ret) |
7c4452b9 | 371 | break; |
d397712b | 372 | while (start <= end) { |
777e6bd7 CM |
373 | cond_resched(); |
374 | ||
5f39d397 | 375 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 376 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 377 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
378 | if (!page) |
379 | continue; | |
4bef0848 CM |
380 | |
381 | btree_lock_page_hook(page); | |
382 | if (!page->mapping) { | |
383 | unlock_page(page); | |
384 | page_cache_release(page); | |
385 | continue; | |
386 | } | |
387 | ||
6702ed49 CM |
388 | if (PageWriteback(page)) { |
389 | if (PageDirty(page)) | |
390 | wait_on_page_writeback(page); | |
391 | else { | |
392 | unlock_page(page); | |
393 | page_cache_release(page); | |
394 | continue; | |
395 | } | |
396 | } | |
7c4452b9 CM |
397 | err = write_one_page(page, 0); |
398 | if (err) | |
399 | werr = err; | |
400 | page_cache_release(page); | |
401 | } | |
402 | } | |
690587d1 CM |
403 | if (err) |
404 | werr = err; | |
405 | return werr; | |
406 | } | |
407 | ||
408 | /* | |
409 | * when btree blocks are allocated, they have some corresponding bits set for | |
410 | * them in one of two extent_io trees. This is used to make sure all of | |
411 | * those extents are on disk for transaction or log commit. We wait | |
412 | * on all the pages and clear them from the dirty pages state tree | |
413 | */ | |
414 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 415 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
416 | { |
417 | int ret; | |
418 | int err = 0; | |
419 | int werr = 0; | |
420 | struct page *page; | |
421 | struct inode *btree_inode = root->fs_info->btree_inode; | |
422 | u64 start = 0; | |
423 | u64 end; | |
424 | unsigned long index; | |
425 | ||
d397712b | 426 | while (1) { |
8cef4e16 YZ |
427 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
428 | mark); | |
777e6bd7 CM |
429 | if (ret) |
430 | break; | |
431 | ||
8cef4e16 | 432 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 433 | while (start <= end) { |
777e6bd7 CM |
434 | index = start >> PAGE_CACHE_SHIFT; |
435 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
436 | page = find_get_page(btree_inode->i_mapping, index); | |
437 | if (!page) | |
438 | continue; | |
439 | if (PageDirty(page)) { | |
4bef0848 CM |
440 | btree_lock_page_hook(page); |
441 | wait_on_page_writeback(page); | |
777e6bd7 CM |
442 | err = write_one_page(page, 0); |
443 | if (err) | |
444 | werr = err; | |
445 | } | |
105d931d | 446 | wait_on_page_writeback(page); |
777e6bd7 CM |
447 | page_cache_release(page); |
448 | cond_resched(); | |
449 | } | |
450 | } | |
7c4452b9 CM |
451 | if (err) |
452 | werr = err; | |
453 | return werr; | |
79154b1b CM |
454 | } |
455 | ||
690587d1 CM |
456 | /* |
457 | * when btree blocks are allocated, they have some corresponding bits set for | |
458 | * them in one of two extent_io trees. This is used to make sure all of | |
459 | * those extents are on disk for transaction or log commit | |
460 | */ | |
461 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 462 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
463 | { |
464 | int ret; | |
465 | int ret2; | |
466 | ||
8cef4e16 YZ |
467 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
468 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
469 | return ret || ret2; |
470 | } | |
471 | ||
d0c803c4 CM |
472 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
473 | struct btrfs_root *root) | |
474 | { | |
475 | if (!trans || !trans->transaction) { | |
476 | struct inode *btree_inode; | |
477 | btree_inode = root->fs_info->btree_inode; | |
478 | return filemap_write_and_wait(btree_inode->i_mapping); | |
479 | } | |
480 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
481 | &trans->transaction->dirty_pages, |
482 | EXTENT_DIRTY); | |
d0c803c4 CM |
483 | } |
484 | ||
d352ac68 CM |
485 | /* |
486 | * this is used to update the root pointer in the tree of tree roots. | |
487 | * | |
488 | * But, in the case of the extent allocation tree, updating the root | |
489 | * pointer may allocate blocks which may change the root of the extent | |
490 | * allocation tree. | |
491 | * | |
492 | * So, this loops and repeats and makes sure the cowonly root didn't | |
493 | * change while the root pointer was being updated in the metadata. | |
494 | */ | |
0b86a832 CM |
495 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
496 | struct btrfs_root *root) | |
79154b1b CM |
497 | { |
498 | int ret; | |
0b86a832 | 499 | u64 old_root_bytenr; |
86b9f2ec | 500 | u64 old_root_used; |
0b86a832 | 501 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 502 | |
86b9f2ec | 503 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 504 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 505 | |
d397712b | 506 | while (1) { |
0b86a832 | 507 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
508 | if (old_root_bytenr == root->node->start && |
509 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 510 | break; |
87ef2bb4 | 511 | |
5d4f98a2 | 512 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 513 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
514 | &root->root_key, |
515 | &root->root_item); | |
79154b1b | 516 | BUG_ON(ret); |
56bec294 | 517 | |
86b9f2ec | 518 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 519 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 520 | BUG_ON(ret); |
0b86a832 | 521 | } |
276e680d YZ |
522 | |
523 | if (root != root->fs_info->extent_root) | |
524 | switch_commit_root(root); | |
525 | ||
0b86a832 CM |
526 | return 0; |
527 | } | |
528 | ||
d352ac68 CM |
529 | /* |
530 | * update all the cowonly tree roots on disk | |
531 | */ | |
5d4f98a2 YZ |
532 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
533 | struct btrfs_root *root) | |
0b86a832 CM |
534 | { |
535 | struct btrfs_fs_info *fs_info = root->fs_info; | |
536 | struct list_head *next; | |
84234f3a | 537 | struct extent_buffer *eb; |
56bec294 | 538 | int ret; |
84234f3a | 539 | |
56bec294 CM |
540 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
541 | BUG_ON(ret); | |
87ef2bb4 | 542 | |
84234f3a | 543 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 544 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
545 | btrfs_tree_unlock(eb); |
546 | free_extent_buffer(eb); | |
0b86a832 | 547 | |
56bec294 CM |
548 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
549 | BUG_ON(ret); | |
87ef2bb4 | 550 | |
d397712b | 551 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
552 | next = fs_info->dirty_cowonly_roots.next; |
553 | list_del_init(next); | |
554 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 555 | |
0b86a832 | 556 | update_cowonly_root(trans, root); |
79154b1b | 557 | } |
276e680d YZ |
558 | |
559 | down_write(&fs_info->extent_commit_sem); | |
560 | switch_commit_root(fs_info->extent_root); | |
561 | up_write(&fs_info->extent_commit_sem); | |
562 | ||
79154b1b CM |
563 | return 0; |
564 | } | |
565 | ||
d352ac68 CM |
566 | /* |
567 | * dead roots are old snapshots that need to be deleted. This allocates | |
568 | * a dirty root struct and adds it into the list of dead roots that need to | |
569 | * be deleted | |
570 | */ | |
5d4f98a2 | 571 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 572 | { |
b48652c1 | 573 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 574 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 575 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
576 | return 0; |
577 | } | |
578 | ||
d352ac68 | 579 | /* |
5d4f98a2 | 580 | * update all the cowonly tree roots on disk |
d352ac68 | 581 | */ |
5d4f98a2 YZ |
582 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
583 | struct btrfs_root *root) | |
0f7d52f4 | 584 | { |
0f7d52f4 | 585 | struct btrfs_root *gang[8]; |
5d4f98a2 | 586 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
587 | int i; |
588 | int ret; | |
54aa1f4d CM |
589 | int err = 0; |
590 | ||
d397712b | 591 | while (1) { |
5d4f98a2 YZ |
592 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
593 | (void **)gang, 0, | |
0f7d52f4 CM |
594 | ARRAY_SIZE(gang), |
595 | BTRFS_ROOT_TRANS_TAG); | |
596 | if (ret == 0) | |
597 | break; | |
598 | for (i = 0; i < ret; i++) { | |
599 | root = gang[i]; | |
5d4f98a2 YZ |
600 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
601 | (unsigned long)root->root_key.objectid, | |
602 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 603 | |
e02119d5 | 604 | btrfs_free_log(trans, root); |
5d4f98a2 | 605 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 606 | |
978d910d | 607 | if (root->commit_root != root->node) { |
817d52f8 | 608 | switch_commit_root(root); |
978d910d YZ |
609 | btrfs_set_root_node(&root->root_item, |
610 | root->node); | |
611 | } | |
5d4f98a2 | 612 | |
5d4f98a2 | 613 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
614 | &root->root_key, |
615 | &root->root_item); | |
54aa1f4d CM |
616 | if (err) |
617 | break; | |
0f7d52f4 CM |
618 | } |
619 | } | |
54aa1f4d | 620 | return err; |
0f7d52f4 CM |
621 | } |
622 | ||
d352ac68 CM |
623 | /* |
624 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
625 | * otherwise every leaf in the btree is read and defragged. | |
626 | */ | |
e9d0b13b CM |
627 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
628 | { | |
629 | struct btrfs_fs_info *info = root->fs_info; | |
630 | int ret; | |
631 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 632 | unsigned long nr; |
e9d0b13b | 633 | |
a2135011 | 634 | smp_mb(); |
e9d0b13b CM |
635 | if (root->defrag_running) |
636 | return 0; | |
e9d0b13b | 637 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 638 | while (1) { |
e9d0b13b CM |
639 | root->defrag_running = 1; |
640 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 641 | nr = trans->blocks_used; |
e9d0b13b | 642 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 643 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
644 | cond_resched(); |
645 | ||
e9d0b13b | 646 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 647 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
648 | break; |
649 | } | |
650 | root->defrag_running = 0; | |
a2135011 | 651 | smp_mb(); |
e9d0b13b CM |
652 | btrfs_end_transaction(trans, root); |
653 | return 0; | |
654 | } | |
655 | ||
2c47e605 | 656 | #if 0 |
b7ec40d7 CM |
657 | /* |
658 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
659 | * sense not to join the transaction while it is trying to flush the current | |
660 | * queue of delayed refs out. | |
661 | * | |
662 | * This is used by the drop snapshot code only | |
663 | */ | |
664 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
665 | { | |
666 | DEFINE_WAIT(wait); | |
667 | ||
668 | mutex_lock(&info->trans_mutex); | |
669 | while (info->running_transaction && | |
670 | info->running_transaction->delayed_refs.flushing) { | |
671 | prepare_to_wait(&info->transaction_wait, &wait, | |
672 | TASK_UNINTERRUPTIBLE); | |
673 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 674 | |
b7ec40d7 | 675 | schedule(); |
59bc5c75 | 676 | |
b7ec40d7 CM |
677 | mutex_lock(&info->trans_mutex); |
678 | finish_wait(&info->transaction_wait, &wait); | |
679 | } | |
680 | mutex_unlock(&info->trans_mutex); | |
681 | return 0; | |
682 | } | |
683 | ||
d352ac68 CM |
684 | /* |
685 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
686 | * all of them | |
687 | */ | |
5d4f98a2 | 688 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 689 | { |
0f7d52f4 | 690 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 691 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 692 | unsigned long nr; |
5d4f98a2 | 693 | int ret; |
58176a96 | 694 | |
5d4f98a2 YZ |
695 | while (1) { |
696 | /* | |
697 | * we don't want to jump in and create a bunch of | |
698 | * delayed refs if the transaction is starting to close | |
699 | */ | |
700 | wait_transaction_pre_flush(tree_root->fs_info); | |
701 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 702 | |
5d4f98a2 YZ |
703 | /* |
704 | * we've joined a transaction, make sure it isn't | |
705 | * closing right now | |
706 | */ | |
707 | if (trans->transaction->delayed_refs.flushing) { | |
708 | btrfs_end_transaction(trans, tree_root); | |
709 | continue; | |
9f3a7427 | 710 | } |
58176a96 | 711 | |
5d4f98a2 YZ |
712 | ret = btrfs_drop_snapshot(trans, root); |
713 | if (ret != -EAGAIN) | |
714 | break; | |
a2135011 | 715 | |
5d4f98a2 YZ |
716 | ret = btrfs_update_root(trans, tree_root, |
717 | &root->root_key, | |
718 | &root->root_item); | |
719 | if (ret) | |
54aa1f4d | 720 | break; |
bcc63abb | 721 | |
d3c2fdcf | 722 | nr = trans->blocks_used; |
0f7d52f4 CM |
723 | ret = btrfs_end_transaction(trans, tree_root); |
724 | BUG_ON(ret); | |
5eda7b5e | 725 | |
d3c2fdcf | 726 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 727 | cond_resched(); |
0f7d52f4 | 728 | } |
5d4f98a2 YZ |
729 | BUG_ON(ret); |
730 | ||
731 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
732 | BUG_ON(ret); | |
733 | ||
734 | nr = trans->blocks_used; | |
735 | ret = btrfs_end_transaction(trans, tree_root); | |
736 | BUG_ON(ret); | |
737 | ||
738 | free_extent_buffer(root->node); | |
739 | free_extent_buffer(root->commit_root); | |
740 | kfree(root); | |
741 | ||
742 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 743 | return ret; |
0f7d52f4 | 744 | } |
2c47e605 | 745 | #endif |
0f7d52f4 | 746 | |
d352ac68 CM |
747 | /* |
748 | * new snapshots need to be created at a very specific time in the | |
749 | * transaction commit. This does the actual creation | |
750 | */ | |
80b6794d | 751 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
752 | struct btrfs_fs_info *fs_info, |
753 | struct btrfs_pending_snapshot *pending) | |
754 | { | |
755 | struct btrfs_key key; | |
80b6794d | 756 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
757 | struct btrfs_root *tree_root = fs_info->tree_root; |
758 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
759 | struct btrfs_root *parent_root; |
760 | struct inode *parent_inode; | |
3063d29f | 761 | struct extent_buffer *tmp; |
925baedd | 762 | struct extent_buffer *old; |
3063d29f CM |
763 | int ret; |
764 | u64 objectid; | |
6bdb72de SW |
765 | int namelen; |
766 | u64 index = 0; | |
767 | ||
768 | parent_inode = pending->dentry->d_parent->d_inode; | |
769 | parent_root = BTRFS_I(parent_inode)->root; | |
3063d29f | 770 | |
80b6794d CM |
771 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
772 | if (!new_root_item) { | |
773 | ret = -ENOMEM; | |
774 | goto fail; | |
775 | } | |
3063d29f CM |
776 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
777 | if (ret) | |
778 | goto fail; | |
779 | ||
3063d29f | 780 | key.objectid = objectid; |
1c4850e2 YZ |
781 | /* record when the snapshot was created in key.offset */ |
782 | key.offset = trans->transid; | |
3063d29f CM |
783 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
784 | ||
3de4586c | 785 | memcpy(&pending->root_key, &key, sizeof(key)); |
6bdb72de | 786 | pending->root_key.offset = (u64)-1; |
3de4586c | 787 | |
6bdb72de | 788 | record_root_in_trans(trans, parent_root); |
3063d29f CM |
789 | /* |
790 | * insert the directory item | |
791 | */ | |
3b96362c | 792 | namelen = strlen(pending->name); |
3de4586c | 793 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 794 | BUG_ON(ret); |
0660b5af | 795 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
796 | pending->name, namelen, |
797 | parent_inode->i_ino, | |
798 | &pending->root_key, BTRFS_FT_DIR, index); | |
6bdb72de | 799 | BUG_ON(ret); |
0660b5af | 800 | |
52c26179 YZ |
801 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
802 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
803 | BUG_ON(ret); | |
804 | ||
6bdb72de SW |
805 | record_root_in_trans(trans, root); |
806 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
807 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
808 | ||
809 | old = btrfs_lock_root_node(root); | |
810 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
811 | btrfs_set_lock_blocking(old); | |
812 | ||
813 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
814 | btrfs_tree_unlock(old); | |
815 | free_extent_buffer(old); | |
816 | ||
817 | btrfs_set_root_node(new_root_item, tmp); | |
818 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, | |
819 | new_root_item); | |
820 | BUG_ON(ret); | |
821 | btrfs_tree_unlock(tmp); | |
822 | free_extent_buffer(tmp); | |
823 | ||
0660b5af CM |
824 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, |
825 | pending->root_key.objectid, | |
0660b5af CM |
826 | parent_root->root_key.objectid, |
827 | parent_inode->i_ino, index, pending->name, | |
828 | namelen); | |
0660b5af CM |
829 | BUG_ON(ret); |
830 | ||
3063d29f | 831 | fail: |
6bdb72de | 832 | kfree(new_root_item); |
3063d29f CM |
833 | return ret; |
834 | } | |
835 | ||
d352ac68 CM |
836 | /* |
837 | * create all the snapshots we've scheduled for creation | |
838 | */ | |
80b6794d CM |
839 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
840 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
841 | { |
842 | struct btrfs_pending_snapshot *pending; | |
843 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
844 | int ret; |
845 | ||
c6e30871 | 846 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
847 | ret = create_pending_snapshot(trans, fs_info, pending); |
848 | BUG_ON(ret); | |
849 | } | |
850 | return 0; | |
851 | } | |
852 | ||
5d4f98a2 YZ |
853 | static void update_super_roots(struct btrfs_root *root) |
854 | { | |
855 | struct btrfs_root_item *root_item; | |
856 | struct btrfs_super_block *super; | |
857 | ||
858 | super = &root->fs_info->super_copy; | |
859 | ||
860 | root_item = &root->fs_info->chunk_root->root_item; | |
861 | super->chunk_root = root_item->bytenr; | |
862 | super->chunk_root_generation = root_item->generation; | |
863 | super->chunk_root_level = root_item->level; | |
864 | ||
865 | root_item = &root->fs_info->tree_root->root_item; | |
866 | super->root = root_item->bytenr; | |
867 | super->generation = root_item->generation; | |
868 | super->root_level = root_item->level; | |
869 | } | |
870 | ||
f36f3042 CM |
871 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
872 | { | |
873 | int ret = 0; | |
874 | spin_lock(&info->new_trans_lock); | |
875 | if (info->running_transaction) | |
876 | ret = info->running_transaction->in_commit; | |
877 | spin_unlock(&info->new_trans_lock); | |
878 | return ret; | |
879 | } | |
880 | ||
79154b1b CM |
881 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
882 | struct btrfs_root *root) | |
883 | { | |
15ee9bc7 JB |
884 | unsigned long joined = 0; |
885 | unsigned long timeout = 1; | |
79154b1b | 886 | struct btrfs_transaction *cur_trans; |
8fd17795 | 887 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 888 | DEFINE_WAIT(wait); |
15ee9bc7 | 889 | int ret; |
89573b9c CM |
890 | int should_grow = 0; |
891 | unsigned long now = get_seconds(); | |
dccae999 | 892 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 893 | |
5a3f23d5 CM |
894 | btrfs_run_ordered_operations(root, 0); |
895 | ||
56bec294 CM |
896 | /* make a pass through all the delayed refs we have so far |
897 | * any runnings procs may add more while we are here | |
898 | */ | |
899 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
900 | BUG_ON(ret); | |
901 | ||
b7ec40d7 | 902 | cur_trans = trans->transaction; |
56bec294 CM |
903 | /* |
904 | * set the flushing flag so procs in this transaction have to | |
905 | * start sending their work down. | |
906 | */ | |
b7ec40d7 | 907 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 908 | |
c3e69d58 | 909 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
910 | BUG_ON(ret); |
911 | ||
79154b1b | 912 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
913 | if (cur_trans->in_commit) { |
914 | cur_trans->use_count++; | |
ccd467d6 | 915 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 916 | btrfs_end_transaction(trans, root); |
ccd467d6 | 917 | |
79154b1b CM |
918 | ret = wait_for_commit(root, cur_trans); |
919 | BUG_ON(ret); | |
15ee9bc7 JB |
920 | |
921 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 922 | put_transaction(cur_trans); |
15ee9bc7 JB |
923 | mutex_unlock(&root->fs_info->trans_mutex); |
924 | ||
79154b1b CM |
925 | return 0; |
926 | } | |
4313b399 | 927 | |
2c90e5d6 | 928 | trans->transaction->in_commit = 1; |
f9295749 | 929 | trans->transaction->blocked = 1; |
ccd467d6 CM |
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 | 943 | |
89573b9c CM |
944 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
945 | should_grow = 1; | |
946 | ||
15ee9bc7 | 947 | do { |
7ea394f1 | 948 | int snap_pending = 0; |
15ee9bc7 | 949 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
950 | if (!list_empty(&trans->transaction->pending_snapshots)) |
951 | snap_pending = 1; | |
952 | ||
2c90e5d6 | 953 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 954 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 955 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
956 | |
957 | if (cur_trans->num_writers > 1) | |
958 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 959 | else if (should_grow) |
15ee9bc7 JB |
960 | timeout = 1; |
961 | ||
79154b1b | 962 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 963 | |
0bdb1db2 | 964 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
965 | btrfs_start_delalloc_inodes(root, 1); |
966 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 967 | BUG_ON(ret); |
7ea394f1 YZ |
968 | } |
969 | ||
5a3f23d5 CM |
970 | /* |
971 | * rename don't use btrfs_join_transaction, so, once we | |
972 | * set the transaction to blocked above, we aren't going | |
973 | * to get any new ordered operations. We can safely run | |
974 | * it here and no for sure that nothing new will be added | |
975 | * to the list | |
976 | */ | |
977 | btrfs_run_ordered_operations(root, 1); | |
978 | ||
89573b9c CM |
979 | smp_mb(); |
980 | if (cur_trans->num_writers > 1 || should_grow) | |
981 | schedule_timeout(timeout); | |
15ee9bc7 | 982 | |
79154b1b | 983 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
984 | finish_wait(&cur_trans->writer_wait, &wait); |
985 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 986 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 987 | |
3063d29f CM |
988 | ret = create_pending_snapshots(trans, root->fs_info); |
989 | BUG_ON(ret); | |
990 | ||
56bec294 CM |
991 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
992 | BUG_ON(ret); | |
993 | ||
2c90e5d6 | 994 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 995 | |
e02119d5 CM |
996 | /* btrfs_commit_tree_roots is responsible for getting the |
997 | * various roots consistent with each other. Every pointer | |
998 | * in the tree of tree roots has to point to the most up to date | |
999 | * root for every subvolume and other tree. So, we have to keep | |
1000 | * the tree logging code from jumping in and changing any | |
1001 | * of the trees. | |
1002 | * | |
1003 | * At this point in the commit, there can't be any tree-log | |
1004 | * writers, but a little lower down we drop the trans mutex | |
1005 | * and let new people in. By holding the tree_log_mutex | |
1006 | * from now until after the super is written, we avoid races | |
1007 | * with the tree-log code. | |
1008 | */ | |
1009 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1010 | ||
5d4f98a2 | 1011 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1012 | BUG_ON(ret); |
1013 | ||
5d4f98a2 | 1014 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1015 | * safe to free the root of tree log roots |
1016 | */ | |
1017 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1018 | ||
5d4f98a2 | 1019 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1020 | BUG_ON(ret); |
54aa1f4d | 1021 | |
11833d66 YZ |
1022 | btrfs_prepare_extent_commit(trans, root); |
1023 | ||
78fae27e | 1024 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1025 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1026 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1027 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1028 | |
1029 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1030 | root->fs_info->tree_root->node); | |
817d52f8 | 1031 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1032 | |
1033 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1034 | root->fs_info->chunk_root->node); | |
817d52f8 | 1035 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1036 | |
1037 | update_super_roots(root); | |
e02119d5 CM |
1038 | |
1039 | if (!root->fs_info->log_root_recovering) { | |
1040 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1041 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1042 | } | |
1043 | ||
a061fc8d CM |
1044 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1045 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1046 | |
f9295749 | 1047 | trans->transaction->blocked = 0; |
b7ec40d7 | 1048 | |
f9295749 | 1049 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1050 | |
78fae27e | 1051 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1052 | ret = btrfs_write_and_wait_transaction(trans, root); |
1053 | BUG_ON(ret); | |
a512bbf8 | 1054 | write_ctree_super(trans, root, 0); |
4313b399 | 1055 | |
e02119d5 CM |
1056 | /* |
1057 | * the super is written, we can safely allow the tree-loggers | |
1058 | * to go about their business | |
1059 | */ | |
1060 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1061 | ||
11833d66 | 1062 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1063 | |
1a40e23b ZY |
1064 | mutex_lock(&root->fs_info->trans_mutex); |
1065 | ||
2c90e5d6 | 1066 | cur_trans->commit_done = 1; |
b7ec40d7 | 1067 | |
15ee9bc7 | 1068 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1069 | |
2c90e5d6 | 1070 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1071 | |
78fae27e | 1072 | put_transaction(cur_trans); |
79154b1b | 1073 | put_transaction(cur_trans); |
58176a96 | 1074 | |
78fae27e | 1075 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1076 | |
9ed74f2d JB |
1077 | if (current->journal_info == trans) |
1078 | current->journal_info = NULL; | |
1079 | ||
2c90e5d6 | 1080 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1081 | |
1082 | if (current != root->fs_info->transaction_kthread) | |
1083 | btrfs_run_delayed_iputs(root); | |
1084 | ||
79154b1b CM |
1085 | return ret; |
1086 | } | |
1087 | ||
d352ac68 CM |
1088 | /* |
1089 | * interface function to delete all the snapshots we have scheduled for deletion | |
1090 | */ | |
e9d0b13b CM |
1091 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1092 | { | |
5d4f98a2 YZ |
1093 | LIST_HEAD(list); |
1094 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1095 | ||
1096 | mutex_lock(&fs_info->trans_mutex); | |
1097 | list_splice_init(&fs_info->dead_roots, &list); | |
1098 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1099 | |
5d4f98a2 YZ |
1100 | while (!list_empty(&list)) { |
1101 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1102 | list_del(&root->root_list); |
1103 | ||
1104 | if (btrfs_header_backref_rev(root->node) < | |
1105 | BTRFS_MIXED_BACKREF_REV) | |
1106 | btrfs_drop_snapshot(root, 0); | |
1107 | else | |
1108 | btrfs_drop_snapshot(root, 1); | |
e9d0b13b CM |
1109 | } |
1110 | return 0; | |
1111 | } |