Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
34088780 | 20 | #include <linux/sched.h> |
d3c2fdcf | 21 | #include <linux/writeback.h> |
5f39d397 | 22 | #include <linux/pagemap.h> |
5f2cc086 | 23 | #include <linux/blkdev.h> |
79154b1b CM |
24 | #include "ctree.h" |
25 | #include "disk-io.h" | |
26 | #include "transaction.h" | |
925baedd | 27 | #include "locking.h" |
e02119d5 | 28 | #include "tree-log.h" |
79154b1b | 29 | |
0f7d52f4 CM |
30 | #define BTRFS_ROOT_TRANS_TAG 0 |
31 | ||
80b6794d | 32 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 33 | { |
2c90e5d6 | 34 | WARN_ON(transaction->use_count == 0); |
79154b1b | 35 | transaction->use_count--; |
78fae27e | 36 | if (transaction->use_count == 0) { |
8fd17795 | 37 | list_del_init(&transaction->list); |
2c90e5d6 CM |
38 | memset(transaction, 0, sizeof(*transaction)); |
39 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 40 | } |
79154b1b CM |
41 | } |
42 | ||
817d52f8 JB |
43 | static noinline void switch_commit_root(struct btrfs_root *root) |
44 | { | |
817d52f8 JB |
45 | free_extent_buffer(root->commit_root); |
46 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
47 | } |
48 | ||
d352ac68 CM |
49 | /* |
50 | * either allocate a new transaction or hop into the existing one | |
51 | */ | |
80b6794d | 52 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
53 | { |
54 | struct btrfs_transaction *cur_trans; | |
55 | cur_trans = root->fs_info->running_transaction; | |
56 | if (!cur_trans) { | |
2c90e5d6 CM |
57 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
58 | GFP_NOFS); | |
79154b1b | 59 | BUG_ON(!cur_trans); |
0f7d52f4 | 60 | root->fs_info->generation++; |
15ee9bc7 JB |
61 | cur_trans->num_writers = 1; |
62 | cur_trans->num_joined = 0; | |
0f7d52f4 | 63 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
64 | init_waitqueue_head(&cur_trans->writer_wait); |
65 | init_waitqueue_head(&cur_trans->commit_wait); | |
66 | cur_trans->in_commit = 0; | |
f9295749 | 67 | cur_trans->blocked = 0; |
d5719762 | 68 | cur_trans->use_count = 1; |
79154b1b | 69 | cur_trans->commit_done = 0; |
08607c1b | 70 | cur_trans->start_time = get_seconds(); |
56bec294 | 71 | |
6bef4d31 | 72 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 73 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
74 | cur_trans->delayed_refs.num_heads_ready = 0; |
75 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 76 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 77 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
78 | spin_lock_init(&cur_trans->delayed_refs.lock); |
79 | ||
3063d29f | 80 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 81 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 82 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
83 | root->fs_info->btree_inode->i_mapping, |
84 | GFP_NOFS); | |
48ec2cf8 CM |
85 | spin_lock(&root->fs_info->new_trans_lock); |
86 | root->fs_info->running_transaction = cur_trans; | |
87 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
88 | } else { |
89 | cur_trans->num_writers++; | |
90 | cur_trans->num_joined++; | |
79154b1b | 91 | } |
15ee9bc7 | 92 | |
79154b1b CM |
93 | return 0; |
94 | } | |
95 | ||
d352ac68 | 96 | /* |
d397712b CM |
97 | * this does all the record keeping required to make sure that a reference |
98 | * counted root is properly recorded in a given transaction. This is required | |
99 | * to make sure the old root from before we joined the transaction is deleted | |
100 | * when the transaction commits | |
d352ac68 | 101 | */ |
5d4f98a2 YZ |
102 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root) | |
6702ed49 | 104 | { |
5d4f98a2 | 105 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 106 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
107 | WARN_ON(root->commit_root != root->node); |
108 | ||
109 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
110 | (unsigned long)root->root_key.objectid, | |
111 | BTRFS_ROOT_TRANS_TAG); | |
112 | root->last_trans = trans->transid; | |
113 | btrfs_init_reloc_root(trans, root); | |
114 | } | |
115 | return 0; | |
116 | } | |
bcc63abb | 117 | |
5d4f98a2 YZ |
118 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
119 | struct btrfs_root *root) | |
120 | { | |
121 | if (!root->ref_cows) | |
122 | return 0; | |
bcc63abb | 123 | |
5d4f98a2 YZ |
124 | mutex_lock(&root->fs_info->trans_mutex); |
125 | if (root->last_trans == trans->transid) { | |
126 | mutex_unlock(&root->fs_info->trans_mutex); | |
127 | return 0; | |
6702ed49 | 128 | } |
5d4f98a2 YZ |
129 | |
130 | record_root_in_trans(trans, root); | |
131 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
132 | return 0; |
133 | } | |
134 | ||
d352ac68 CM |
135 | /* wait for commit against the current transaction to become unblocked |
136 | * when this is done, it is safe to start a new transaction, but the current | |
137 | * transaction might not be fully on disk. | |
138 | */ | |
37d1aeee | 139 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 140 | { |
f9295749 | 141 | struct btrfs_transaction *cur_trans; |
79154b1b | 142 | |
f9295749 | 143 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 144 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
145 | DEFINE_WAIT(wait); |
146 | cur_trans->use_count++; | |
d397712b | 147 | while (1) { |
f9295749 CM |
148 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
149 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 150 | if (!cur_trans->blocked) |
f9295749 | 151 | break; |
471fa17d ZL |
152 | mutex_unlock(&root->fs_info->trans_mutex); |
153 | schedule(); | |
154 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 155 | } |
471fa17d | 156 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
157 | put_transaction(cur_trans); |
158 | } | |
37d1aeee CM |
159 | } |
160 | ||
249ac1e5 JB |
161 | enum btrfs_trans_type { |
162 | TRANS_START, | |
163 | TRANS_JOIN, | |
164 | TRANS_USERSPACE, | |
165 | }; | |
166 | ||
e02119d5 | 167 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
249ac1e5 | 168 | int num_blocks, int type) |
37d1aeee CM |
169 | { |
170 | struct btrfs_trans_handle *h = | |
171 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
172 | int ret; | |
173 | ||
174 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 | 175 | if (!root->fs_info->log_root_recovering && |
249ac1e5 JB |
176 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || |
177 | type == TRANS_USERSPACE)) | |
37d1aeee | 178 | wait_current_trans(root); |
79154b1b CM |
179 | ret = join_transaction(root); |
180 | BUG_ON(ret); | |
0f7d52f4 | 181 | |
6702ed49 | 182 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
183 | h->transaction = root->fs_info->running_transaction; |
184 | h->blocks_reserved = num_blocks; | |
185 | h->blocks_used = 0; | |
d2fb3437 | 186 | h->block_group = 0; |
26b8003f CM |
187 | h->alloc_exclude_nr = 0; |
188 | h->alloc_exclude_start = 0; | |
56bec294 | 189 | h->delayed_ref_updates = 0; |
b7ec40d7 | 190 | |
249ac1e5 | 191 | if (!current->journal_info && type != TRANS_USERSPACE) |
9ed74f2d JB |
192 | current->journal_info = h; |
193 | ||
79154b1b | 194 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 195 | record_root_in_trans(h, root); |
79154b1b CM |
196 | mutex_unlock(&root->fs_info->trans_mutex); |
197 | return h; | |
198 | } | |
199 | ||
f9295749 CM |
200 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
201 | int num_blocks) | |
202 | { | |
249ac1e5 | 203 | return start_transaction(root, num_blocks, TRANS_START); |
f9295749 CM |
204 | } |
205 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
206 | int num_blocks) | |
207 | { | |
249ac1e5 | 208 | return start_transaction(root, num_blocks, TRANS_JOIN); |
f9295749 CM |
209 | } |
210 | ||
9ca9ee09 SW |
211 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
212 | int num_blocks) | |
213 | { | |
249ac1e5 | 214 | return start_transaction(r, num_blocks, TRANS_USERSPACE); |
9ca9ee09 SW |
215 | } |
216 | ||
d352ac68 | 217 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
218 | static noinline int wait_for_commit(struct btrfs_root *root, |
219 | struct btrfs_transaction *commit) | |
220 | { | |
221 | DEFINE_WAIT(wait); | |
222 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 223 | while (!commit->commit_done) { |
89ce8a63 CM |
224 | prepare_to_wait(&commit->commit_wait, &wait, |
225 | TASK_UNINTERRUPTIBLE); | |
226 | if (commit->commit_done) | |
227 | break; | |
228 | mutex_unlock(&root->fs_info->trans_mutex); | |
229 | schedule(); | |
230 | mutex_lock(&root->fs_info->trans_mutex); | |
231 | } | |
232 | mutex_unlock(&root->fs_info->trans_mutex); | |
233 | finish_wait(&commit->commit_wait, &wait); | |
234 | return 0; | |
235 | } | |
236 | ||
5d4f98a2 | 237 | #if 0 |
d352ac68 | 238 | /* |
d397712b CM |
239 | * rate limit against the drop_snapshot code. This helps to slow down new |
240 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 241 | */ |
37d1aeee | 242 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
243 | { |
244 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 245 | int harder_count = 0; |
ab78c84d | 246 | |
2dd3e67b | 247 | harder: |
ab78c84d CM |
248 | if (atomic_read(&info->throttles)) { |
249 | DEFINE_WAIT(wait); | |
250 | int thr; | |
ab78c84d CM |
251 | thr = atomic_read(&info->throttle_gen); |
252 | ||
253 | do { | |
254 | prepare_to_wait(&info->transaction_throttle, | |
255 | &wait, TASK_UNINTERRUPTIBLE); | |
256 | if (!atomic_read(&info->throttles)) { | |
257 | finish_wait(&info->transaction_throttle, &wait); | |
258 | break; | |
259 | } | |
260 | schedule(); | |
261 | finish_wait(&info->transaction_throttle, &wait); | |
262 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
263 | harder_count++; |
264 | ||
265 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
266 | harder_count < 2) | |
267 | goto harder; | |
268 | ||
269 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
270 | harder_count < 10) | |
271 | goto harder; | |
272 | ||
273 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
274 | harder_count < 20) | |
275 | goto harder; | |
ab78c84d CM |
276 | } |
277 | } | |
5d4f98a2 | 278 | #endif |
ab78c84d | 279 | |
37d1aeee CM |
280 | void btrfs_throttle(struct btrfs_root *root) |
281 | { | |
282 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
283 | if (!root->fs_info->open_ioctl_trans) |
284 | wait_current_trans(root); | |
37d1aeee | 285 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
286 | } |
287 | ||
89ce8a63 CM |
288 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
289 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
290 | { |
291 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 292 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
293 | int count = 0; |
294 | ||
295 | while (count < 4) { | |
296 | unsigned long cur = trans->delayed_ref_updates; | |
297 | trans->delayed_ref_updates = 0; | |
298 | if (cur && | |
299 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
300 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
301 | |
302 | /* | |
303 | * do a full flush if the transaction is trying | |
304 | * to close | |
305 | */ | |
306 | if (trans->transaction->delayed_refs.flushing) | |
307 | cur = 0; | |
c3e69d58 CM |
308 | btrfs_run_delayed_refs(trans, root, cur); |
309 | } else { | |
310 | break; | |
311 | } | |
312 | count++; | |
56bec294 CM |
313 | } |
314 | ||
ab78c84d CM |
315 | mutex_lock(&info->trans_mutex); |
316 | cur_trans = info->running_transaction; | |
ccd467d6 | 317 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 318 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 319 | cur_trans->num_writers--; |
89ce8a63 | 320 | |
79154b1b CM |
321 | if (waitqueue_active(&cur_trans->writer_wait)) |
322 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 323 | put_transaction(cur_trans); |
ab78c84d | 324 | mutex_unlock(&info->trans_mutex); |
9ed74f2d JB |
325 | |
326 | if (current->journal_info == trans) | |
327 | current->journal_info = NULL; | |
d6025579 | 328 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 329 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 330 | |
24bbcf04 YZ |
331 | if (throttle) |
332 | btrfs_run_delayed_iputs(root); | |
333 | ||
79154b1b CM |
334 | return 0; |
335 | } | |
336 | ||
89ce8a63 CM |
337 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
338 | struct btrfs_root *root) | |
339 | { | |
340 | return __btrfs_end_transaction(trans, root, 0); | |
341 | } | |
342 | ||
343 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
344 | struct btrfs_root *root) | |
345 | { | |
346 | return __btrfs_end_transaction(trans, root, 1); | |
347 | } | |
348 | ||
d352ac68 CM |
349 | /* |
350 | * when btree blocks are allocated, they have some corresponding bits set for | |
351 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 352 | * those extents are sent to disk but does not wait on them |
d352ac68 | 353 | */ |
690587d1 | 354 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 355 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 356 | { |
7c4452b9 | 357 | int ret; |
777e6bd7 | 358 | int err = 0; |
7c4452b9 CM |
359 | int werr = 0; |
360 | struct page *page; | |
7c4452b9 | 361 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 362 | u64 start = 0; |
5f39d397 CM |
363 | u64 end; |
364 | unsigned long index; | |
7c4452b9 | 365 | |
d397712b | 366 | while (1) { |
777e6bd7 | 367 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 368 | mark); |
5f39d397 | 369 | if (ret) |
7c4452b9 | 370 | break; |
d397712b | 371 | while (start <= end) { |
777e6bd7 CM |
372 | cond_resched(); |
373 | ||
5f39d397 | 374 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 375 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 376 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
377 | if (!page) |
378 | continue; | |
4bef0848 CM |
379 | |
380 | btree_lock_page_hook(page); | |
381 | if (!page->mapping) { | |
382 | unlock_page(page); | |
383 | page_cache_release(page); | |
384 | continue; | |
385 | } | |
386 | ||
6702ed49 CM |
387 | if (PageWriteback(page)) { |
388 | if (PageDirty(page)) | |
389 | wait_on_page_writeback(page); | |
390 | else { | |
391 | unlock_page(page); | |
392 | page_cache_release(page); | |
393 | continue; | |
394 | } | |
395 | } | |
7c4452b9 CM |
396 | err = write_one_page(page, 0); |
397 | if (err) | |
398 | werr = err; | |
399 | page_cache_release(page); | |
400 | } | |
401 | } | |
690587d1 CM |
402 | if (err) |
403 | werr = err; | |
404 | return werr; | |
405 | } | |
406 | ||
407 | /* | |
408 | * when btree blocks are allocated, they have some corresponding bits set for | |
409 | * them in one of two extent_io trees. This is used to make sure all of | |
410 | * those extents are on disk for transaction or log commit. We wait | |
411 | * on all the pages and clear them from the dirty pages state tree | |
412 | */ | |
413 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 414 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
415 | { |
416 | int ret; | |
417 | int err = 0; | |
418 | int werr = 0; | |
419 | struct page *page; | |
420 | struct inode *btree_inode = root->fs_info->btree_inode; | |
421 | u64 start = 0; | |
422 | u64 end; | |
423 | unsigned long index; | |
424 | ||
d397712b | 425 | while (1) { |
8cef4e16 YZ |
426 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
427 | mark); | |
777e6bd7 CM |
428 | if (ret) |
429 | break; | |
430 | ||
8cef4e16 | 431 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 432 | while (start <= end) { |
777e6bd7 CM |
433 | index = start >> PAGE_CACHE_SHIFT; |
434 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
435 | page = find_get_page(btree_inode->i_mapping, index); | |
436 | if (!page) | |
437 | continue; | |
438 | if (PageDirty(page)) { | |
4bef0848 CM |
439 | btree_lock_page_hook(page); |
440 | wait_on_page_writeback(page); | |
777e6bd7 CM |
441 | err = write_one_page(page, 0); |
442 | if (err) | |
443 | werr = err; | |
444 | } | |
105d931d | 445 | wait_on_page_writeback(page); |
777e6bd7 CM |
446 | page_cache_release(page); |
447 | cond_resched(); | |
448 | } | |
449 | } | |
7c4452b9 CM |
450 | if (err) |
451 | werr = err; | |
452 | return werr; | |
79154b1b CM |
453 | } |
454 | ||
690587d1 CM |
455 | /* |
456 | * when btree blocks are allocated, they have some corresponding bits set for | |
457 | * them in one of two extent_io trees. This is used to make sure all of | |
458 | * those extents are on disk for transaction or log commit | |
459 | */ | |
460 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 461 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
462 | { |
463 | int ret; | |
464 | int ret2; | |
465 | ||
8cef4e16 YZ |
466 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
467 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
468 | return ret || ret2; |
469 | } | |
470 | ||
d0c803c4 CM |
471 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
472 | struct btrfs_root *root) | |
473 | { | |
474 | if (!trans || !trans->transaction) { | |
475 | struct inode *btree_inode; | |
476 | btree_inode = root->fs_info->btree_inode; | |
477 | return filemap_write_and_wait(btree_inode->i_mapping); | |
478 | } | |
479 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
480 | &trans->transaction->dirty_pages, |
481 | EXTENT_DIRTY); | |
d0c803c4 CM |
482 | } |
483 | ||
d352ac68 CM |
484 | /* |
485 | * this is used to update the root pointer in the tree of tree roots. | |
486 | * | |
487 | * But, in the case of the extent allocation tree, updating the root | |
488 | * pointer may allocate blocks which may change the root of the extent | |
489 | * allocation tree. | |
490 | * | |
491 | * So, this loops and repeats and makes sure the cowonly root didn't | |
492 | * change while the root pointer was being updated in the metadata. | |
493 | */ | |
0b86a832 CM |
494 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
495 | struct btrfs_root *root) | |
79154b1b CM |
496 | { |
497 | int ret; | |
0b86a832 | 498 | u64 old_root_bytenr; |
86b9f2ec | 499 | u64 old_root_used; |
0b86a832 | 500 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 501 | |
86b9f2ec | 502 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 503 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 504 | |
d397712b | 505 | while (1) { |
0b86a832 | 506 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
507 | if (old_root_bytenr == root->node->start && |
508 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 509 | break; |
87ef2bb4 | 510 | |
5d4f98a2 | 511 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 512 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
513 | &root->root_key, |
514 | &root->root_item); | |
79154b1b | 515 | BUG_ON(ret); |
56bec294 | 516 | |
86b9f2ec | 517 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 518 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 519 | BUG_ON(ret); |
0b86a832 | 520 | } |
276e680d YZ |
521 | |
522 | if (root != root->fs_info->extent_root) | |
523 | switch_commit_root(root); | |
524 | ||
0b86a832 CM |
525 | return 0; |
526 | } | |
527 | ||
d352ac68 CM |
528 | /* |
529 | * update all the cowonly tree roots on disk | |
530 | */ | |
5d4f98a2 YZ |
531 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
532 | struct btrfs_root *root) | |
0b86a832 CM |
533 | { |
534 | struct btrfs_fs_info *fs_info = root->fs_info; | |
535 | struct list_head *next; | |
84234f3a | 536 | struct extent_buffer *eb; |
56bec294 | 537 | int ret; |
84234f3a | 538 | |
56bec294 CM |
539 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
540 | BUG_ON(ret); | |
87ef2bb4 | 541 | |
84234f3a | 542 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 543 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
544 | btrfs_tree_unlock(eb); |
545 | free_extent_buffer(eb); | |
0b86a832 | 546 | |
56bec294 CM |
547 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
548 | BUG_ON(ret); | |
87ef2bb4 | 549 | |
d397712b | 550 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
551 | next = fs_info->dirty_cowonly_roots.next; |
552 | list_del_init(next); | |
553 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 554 | |
0b86a832 | 555 | update_cowonly_root(trans, root); |
79154b1b | 556 | } |
276e680d YZ |
557 | |
558 | down_write(&fs_info->extent_commit_sem); | |
559 | switch_commit_root(fs_info->extent_root); | |
560 | up_write(&fs_info->extent_commit_sem); | |
561 | ||
79154b1b CM |
562 | return 0; |
563 | } | |
564 | ||
d352ac68 CM |
565 | /* |
566 | * dead roots are old snapshots that need to be deleted. This allocates | |
567 | * a dirty root struct and adds it into the list of dead roots that need to | |
568 | * be deleted | |
569 | */ | |
5d4f98a2 | 570 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 571 | { |
b48652c1 | 572 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 573 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 574 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
575 | return 0; |
576 | } | |
577 | ||
d352ac68 | 578 | /* |
5d4f98a2 | 579 | * update all the cowonly tree roots on disk |
d352ac68 | 580 | */ |
5d4f98a2 YZ |
581 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
582 | struct btrfs_root *root) | |
0f7d52f4 | 583 | { |
0f7d52f4 | 584 | struct btrfs_root *gang[8]; |
5d4f98a2 | 585 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
586 | int i; |
587 | int ret; | |
54aa1f4d CM |
588 | int err = 0; |
589 | ||
d397712b | 590 | while (1) { |
5d4f98a2 YZ |
591 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
592 | (void **)gang, 0, | |
0f7d52f4 CM |
593 | ARRAY_SIZE(gang), |
594 | BTRFS_ROOT_TRANS_TAG); | |
595 | if (ret == 0) | |
596 | break; | |
597 | for (i = 0; i < ret; i++) { | |
598 | root = gang[i]; | |
5d4f98a2 YZ |
599 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
600 | (unsigned long)root->root_key.objectid, | |
601 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 602 | |
e02119d5 | 603 | btrfs_free_log(trans, root); |
5d4f98a2 | 604 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 605 | |
978d910d | 606 | if (root->commit_root != root->node) { |
817d52f8 | 607 | switch_commit_root(root); |
978d910d YZ |
608 | btrfs_set_root_node(&root->root_item, |
609 | root->node); | |
610 | } | |
5d4f98a2 | 611 | |
5d4f98a2 | 612 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
613 | &root->root_key, |
614 | &root->root_item); | |
54aa1f4d CM |
615 | if (err) |
616 | break; | |
0f7d52f4 CM |
617 | } |
618 | } | |
54aa1f4d | 619 | return err; |
0f7d52f4 CM |
620 | } |
621 | ||
d352ac68 CM |
622 | /* |
623 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
624 | * otherwise every leaf in the btree is read and defragged. | |
625 | */ | |
e9d0b13b CM |
626 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
627 | { | |
628 | struct btrfs_fs_info *info = root->fs_info; | |
629 | int ret; | |
630 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 631 | unsigned long nr; |
e9d0b13b | 632 | |
a2135011 | 633 | smp_mb(); |
e9d0b13b CM |
634 | if (root->defrag_running) |
635 | return 0; | |
e9d0b13b | 636 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 637 | while (1) { |
e9d0b13b CM |
638 | root->defrag_running = 1; |
639 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 640 | nr = trans->blocks_used; |
e9d0b13b | 641 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 642 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
643 | cond_resched(); |
644 | ||
e9d0b13b | 645 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 646 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
647 | break; |
648 | } | |
649 | root->defrag_running = 0; | |
a2135011 | 650 | smp_mb(); |
e9d0b13b CM |
651 | btrfs_end_transaction(trans, root); |
652 | return 0; | |
653 | } | |
654 | ||
2c47e605 | 655 | #if 0 |
b7ec40d7 CM |
656 | /* |
657 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
658 | * sense not to join the transaction while it is trying to flush the current | |
659 | * queue of delayed refs out. | |
660 | * | |
661 | * This is used by the drop snapshot code only | |
662 | */ | |
663 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
664 | { | |
665 | DEFINE_WAIT(wait); | |
666 | ||
667 | mutex_lock(&info->trans_mutex); | |
668 | while (info->running_transaction && | |
669 | info->running_transaction->delayed_refs.flushing) { | |
670 | prepare_to_wait(&info->transaction_wait, &wait, | |
671 | TASK_UNINTERRUPTIBLE); | |
672 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 673 | |
b7ec40d7 | 674 | schedule(); |
59bc5c75 | 675 | |
b7ec40d7 CM |
676 | mutex_lock(&info->trans_mutex); |
677 | finish_wait(&info->transaction_wait, &wait); | |
678 | } | |
679 | mutex_unlock(&info->trans_mutex); | |
680 | return 0; | |
681 | } | |
682 | ||
d352ac68 CM |
683 | /* |
684 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
685 | * all of them | |
686 | */ | |
5d4f98a2 | 687 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 688 | { |
0f7d52f4 | 689 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 690 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 691 | unsigned long nr; |
5d4f98a2 | 692 | int ret; |
58176a96 | 693 | |
5d4f98a2 YZ |
694 | while (1) { |
695 | /* | |
696 | * we don't want to jump in and create a bunch of | |
697 | * delayed refs if the transaction is starting to close | |
698 | */ | |
699 | wait_transaction_pre_flush(tree_root->fs_info); | |
700 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 701 | |
5d4f98a2 YZ |
702 | /* |
703 | * we've joined a transaction, make sure it isn't | |
704 | * closing right now | |
705 | */ | |
706 | if (trans->transaction->delayed_refs.flushing) { | |
707 | btrfs_end_transaction(trans, tree_root); | |
708 | continue; | |
9f3a7427 | 709 | } |
58176a96 | 710 | |
5d4f98a2 YZ |
711 | ret = btrfs_drop_snapshot(trans, root); |
712 | if (ret != -EAGAIN) | |
713 | break; | |
a2135011 | 714 | |
5d4f98a2 YZ |
715 | ret = btrfs_update_root(trans, tree_root, |
716 | &root->root_key, | |
717 | &root->root_item); | |
718 | if (ret) | |
54aa1f4d | 719 | break; |
bcc63abb | 720 | |
d3c2fdcf | 721 | nr = trans->blocks_used; |
0f7d52f4 CM |
722 | ret = btrfs_end_transaction(trans, tree_root); |
723 | BUG_ON(ret); | |
5eda7b5e | 724 | |
d3c2fdcf | 725 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 726 | cond_resched(); |
0f7d52f4 | 727 | } |
5d4f98a2 YZ |
728 | BUG_ON(ret); |
729 | ||
730 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
731 | BUG_ON(ret); | |
732 | ||
733 | nr = trans->blocks_used; | |
734 | ret = btrfs_end_transaction(trans, tree_root); | |
735 | BUG_ON(ret); | |
736 | ||
737 | free_extent_buffer(root->node); | |
738 | free_extent_buffer(root->commit_root); | |
739 | kfree(root); | |
740 | ||
741 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 742 | return ret; |
0f7d52f4 | 743 | } |
2c47e605 | 744 | #endif |
0f7d52f4 | 745 | |
d352ac68 CM |
746 | /* |
747 | * new snapshots need to be created at a very specific time in the | |
748 | * transaction commit. This does the actual creation | |
749 | */ | |
80b6794d | 750 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
751 | struct btrfs_fs_info *fs_info, |
752 | struct btrfs_pending_snapshot *pending) | |
753 | { | |
754 | struct btrfs_key key; | |
80b6794d | 755 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
756 | struct btrfs_root *tree_root = fs_info->tree_root; |
757 | struct btrfs_root *root = pending->root; | |
758 | struct extent_buffer *tmp; | |
925baedd | 759 | struct extent_buffer *old; |
3063d29f CM |
760 | int ret; |
761 | u64 objectid; | |
762 | ||
80b6794d CM |
763 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
764 | if (!new_root_item) { | |
765 | ret = -ENOMEM; | |
766 | goto fail; | |
767 | } | |
3063d29f CM |
768 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
769 | if (ret) | |
770 | goto fail; | |
771 | ||
5d4f98a2 | 772 | record_root_in_trans(trans, root); |
80ff3856 | 773 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
80b6794d | 774 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
775 | |
776 | key.objectid = objectid; | |
1c4850e2 YZ |
777 | /* record when the snapshot was created in key.offset */ |
778 | key.offset = trans->transid; | |
3063d29f CM |
779 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
780 | ||
925baedd | 781 | old = btrfs_lock_root_node(root); |
9fa8cfe7 | 782 | btrfs_cow_block(trans, root, old, NULL, 0, &old); |
5d4f98a2 | 783 | btrfs_set_lock_blocking(old); |
3063d29f | 784 | |
925baedd CM |
785 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
786 | btrfs_tree_unlock(old); | |
787 | free_extent_buffer(old); | |
3063d29f | 788 | |
5d4f98a2 | 789 | btrfs_set_root_node(new_root_item, tmp); |
3063d29f | 790 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 791 | new_root_item); |
925baedd | 792 | btrfs_tree_unlock(tmp); |
3063d29f CM |
793 | free_extent_buffer(tmp); |
794 | if (ret) | |
795 | goto fail; | |
796 | ||
3de4586c CM |
797 | key.offset = (u64)-1; |
798 | memcpy(&pending->root_key, &key, sizeof(key)); | |
799 | fail: | |
800 | kfree(new_root_item); | |
801 | return ret; | |
802 | } | |
803 | ||
804 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
805 | struct btrfs_pending_snapshot *pending) | |
806 | { | |
807 | int ret; | |
808 | int namelen; | |
809 | u64 index = 0; | |
810 | struct btrfs_trans_handle *trans; | |
811 | struct inode *parent_inode; | |
0660b5af | 812 | struct btrfs_root *parent_root; |
3de4586c | 813 | |
3394e160 | 814 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 815 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 816 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 817 | |
3063d29f CM |
818 | /* |
819 | * insert the directory item | |
820 | */ | |
3b96362c | 821 | namelen = strlen(pending->name); |
3de4586c | 822 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 823 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
824 | pending->name, namelen, |
825 | parent_inode->i_ino, | |
826 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
827 | |
828 | if (ret) | |
829 | goto fail; | |
0660b5af | 830 | |
52c26179 YZ |
831 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
832 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
833 | BUG_ON(ret); | |
834 | ||
0660b5af CM |
835 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, |
836 | pending->root_key.objectid, | |
0660b5af CM |
837 | parent_root->root_key.objectid, |
838 | parent_inode->i_ino, index, pending->name, | |
839 | namelen); | |
840 | ||
841 | BUG_ON(ret); | |
842 | ||
3063d29f | 843 | fail: |
3de4586c | 844 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
845 | return ret; |
846 | } | |
847 | ||
d352ac68 CM |
848 | /* |
849 | * create all the snapshots we've scheduled for creation | |
850 | */ | |
80b6794d CM |
851 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
852 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
853 | { |
854 | struct btrfs_pending_snapshot *pending; | |
855 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
856 | int ret; |
857 | ||
c6e30871 | 858 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
859 | ret = create_pending_snapshot(trans, fs_info, pending); |
860 | BUG_ON(ret); | |
861 | } | |
862 | return 0; | |
863 | } | |
864 | ||
865 | static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans, | |
866 | struct btrfs_fs_info *fs_info) | |
3063d29f CM |
867 | { |
868 | struct btrfs_pending_snapshot *pending; | |
869 | struct list_head *head = &trans->transaction->pending_snapshots; | |
870 | int ret; | |
871 | ||
d397712b | 872 | while (!list_empty(head)) { |
3063d29f CM |
873 | pending = list_entry(head->next, |
874 | struct btrfs_pending_snapshot, list); | |
3de4586c | 875 | ret = finish_pending_snapshot(fs_info, pending); |
3063d29f CM |
876 | BUG_ON(ret); |
877 | list_del(&pending->list); | |
878 | kfree(pending->name); | |
879 | kfree(pending); | |
880 | } | |
dc17ff8f CM |
881 | return 0; |
882 | } | |
883 | ||
5d4f98a2 YZ |
884 | static void update_super_roots(struct btrfs_root *root) |
885 | { | |
886 | struct btrfs_root_item *root_item; | |
887 | struct btrfs_super_block *super; | |
888 | ||
889 | super = &root->fs_info->super_copy; | |
890 | ||
891 | root_item = &root->fs_info->chunk_root->root_item; | |
892 | super->chunk_root = root_item->bytenr; | |
893 | super->chunk_root_generation = root_item->generation; | |
894 | super->chunk_root_level = root_item->level; | |
895 | ||
896 | root_item = &root->fs_info->tree_root->root_item; | |
897 | super->root = root_item->bytenr; | |
898 | super->generation = root_item->generation; | |
899 | super->root_level = root_item->level; | |
900 | } | |
901 | ||
f36f3042 CM |
902 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
903 | { | |
904 | int ret = 0; | |
905 | spin_lock(&info->new_trans_lock); | |
906 | if (info->running_transaction) | |
907 | ret = info->running_transaction->in_commit; | |
908 | spin_unlock(&info->new_trans_lock); | |
909 | return ret; | |
910 | } | |
911 | ||
79154b1b CM |
912 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
913 | struct btrfs_root *root) | |
914 | { | |
15ee9bc7 JB |
915 | unsigned long joined = 0; |
916 | unsigned long timeout = 1; | |
79154b1b | 917 | struct btrfs_transaction *cur_trans; |
8fd17795 | 918 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 919 | DEFINE_WAIT(wait); |
15ee9bc7 | 920 | int ret; |
89573b9c CM |
921 | int should_grow = 0; |
922 | unsigned long now = get_seconds(); | |
dccae999 | 923 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 924 | |
5a3f23d5 CM |
925 | btrfs_run_ordered_operations(root, 0); |
926 | ||
56bec294 CM |
927 | /* make a pass through all the delayed refs we have so far |
928 | * any runnings procs may add more while we are here | |
929 | */ | |
930 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
931 | BUG_ON(ret); | |
932 | ||
b7ec40d7 | 933 | cur_trans = trans->transaction; |
56bec294 CM |
934 | /* |
935 | * set the flushing flag so procs in this transaction have to | |
936 | * start sending their work down. | |
937 | */ | |
b7ec40d7 | 938 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 939 | |
c3e69d58 | 940 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
941 | BUG_ON(ret); |
942 | ||
79154b1b | 943 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
944 | if (cur_trans->in_commit) { |
945 | cur_trans->use_count++; | |
ccd467d6 | 946 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 947 | btrfs_end_transaction(trans, root); |
ccd467d6 | 948 | |
79154b1b CM |
949 | ret = wait_for_commit(root, cur_trans); |
950 | BUG_ON(ret); | |
15ee9bc7 JB |
951 | |
952 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 953 | put_transaction(cur_trans); |
15ee9bc7 JB |
954 | mutex_unlock(&root->fs_info->trans_mutex); |
955 | ||
79154b1b CM |
956 | return 0; |
957 | } | |
4313b399 | 958 | |
2c90e5d6 | 959 | trans->transaction->in_commit = 1; |
f9295749 | 960 | trans->transaction->blocked = 1; |
ccd467d6 CM |
961 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
962 | prev_trans = list_entry(cur_trans->list.prev, | |
963 | struct btrfs_transaction, list); | |
964 | if (!prev_trans->commit_done) { | |
965 | prev_trans->use_count++; | |
ccd467d6 CM |
966 | mutex_unlock(&root->fs_info->trans_mutex); |
967 | ||
968 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 969 | |
ccd467d6 | 970 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 971 | put_transaction(prev_trans); |
ccd467d6 CM |
972 | } |
973 | } | |
15ee9bc7 | 974 | |
89573b9c CM |
975 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
976 | should_grow = 1; | |
977 | ||
15ee9bc7 | 978 | do { |
7ea394f1 | 979 | int snap_pending = 0; |
15ee9bc7 | 980 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
981 | if (!list_empty(&trans->transaction->pending_snapshots)) |
982 | snap_pending = 1; | |
983 | ||
2c90e5d6 | 984 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 985 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 986 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
987 | |
988 | if (cur_trans->num_writers > 1) | |
989 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 990 | else if (should_grow) |
15ee9bc7 JB |
991 | timeout = 1; |
992 | ||
79154b1b | 993 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 994 | |
0bdb1db2 | 995 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
996 | btrfs_start_delalloc_inodes(root, 1); |
997 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 998 | BUG_ON(ret); |
7ea394f1 YZ |
999 | } |
1000 | ||
5a3f23d5 CM |
1001 | /* |
1002 | * rename don't use btrfs_join_transaction, so, once we | |
1003 | * set the transaction to blocked above, we aren't going | |
1004 | * to get any new ordered operations. We can safely run | |
1005 | * it here and no for sure that nothing new will be added | |
1006 | * to the list | |
1007 | */ | |
1008 | btrfs_run_ordered_operations(root, 1); | |
1009 | ||
89573b9c CM |
1010 | smp_mb(); |
1011 | if (cur_trans->num_writers > 1 || should_grow) | |
1012 | schedule_timeout(timeout); | |
15ee9bc7 | 1013 | |
79154b1b | 1014 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1015 | finish_wait(&cur_trans->writer_wait, &wait); |
1016 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1017 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1018 | |
3063d29f CM |
1019 | ret = create_pending_snapshots(trans, root->fs_info); |
1020 | BUG_ON(ret); | |
1021 | ||
56bec294 CM |
1022 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1023 | BUG_ON(ret); | |
1024 | ||
2c90e5d6 | 1025 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1026 | |
e02119d5 CM |
1027 | /* btrfs_commit_tree_roots is responsible for getting the |
1028 | * various roots consistent with each other. Every pointer | |
1029 | * in the tree of tree roots has to point to the most up to date | |
1030 | * root for every subvolume and other tree. So, we have to keep | |
1031 | * the tree logging code from jumping in and changing any | |
1032 | * of the trees. | |
1033 | * | |
1034 | * At this point in the commit, there can't be any tree-log | |
1035 | * writers, but a little lower down we drop the trans mutex | |
1036 | * and let new people in. By holding the tree_log_mutex | |
1037 | * from now until after the super is written, we avoid races | |
1038 | * with the tree-log code. | |
1039 | */ | |
1040 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1041 | ||
5d4f98a2 | 1042 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1043 | BUG_ON(ret); |
1044 | ||
5d4f98a2 | 1045 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1046 | * safe to free the root of tree log roots |
1047 | */ | |
1048 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1049 | ||
5d4f98a2 | 1050 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1051 | BUG_ON(ret); |
54aa1f4d | 1052 | |
11833d66 YZ |
1053 | btrfs_prepare_extent_commit(trans, root); |
1054 | ||
78fae27e | 1055 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1056 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1057 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1058 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1059 | |
1060 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1061 | root->fs_info->tree_root->node); | |
817d52f8 | 1062 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1063 | |
1064 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1065 | root->fs_info->chunk_root->node); | |
817d52f8 | 1066 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1067 | |
1068 | update_super_roots(root); | |
e02119d5 CM |
1069 | |
1070 | if (!root->fs_info->log_root_recovering) { | |
1071 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1072 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1073 | } | |
1074 | ||
a061fc8d CM |
1075 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1076 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1077 | |
f9295749 | 1078 | trans->transaction->blocked = 0; |
b7ec40d7 | 1079 | |
f9295749 | 1080 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1081 | |
78fae27e | 1082 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1083 | ret = btrfs_write_and_wait_transaction(trans, root); |
1084 | BUG_ON(ret); | |
a512bbf8 | 1085 | write_ctree_super(trans, root, 0); |
4313b399 | 1086 | |
e02119d5 CM |
1087 | /* |
1088 | * the super is written, we can safely allow the tree-loggers | |
1089 | * to go about their business | |
1090 | */ | |
1091 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1092 | ||
11833d66 | 1093 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1094 | |
3de4586c CM |
1095 | /* do the directory inserts of any pending snapshot creations */ |
1096 | finish_pending_snapshots(trans, root->fs_info); | |
1097 | ||
1a40e23b ZY |
1098 | mutex_lock(&root->fs_info->trans_mutex); |
1099 | ||
2c90e5d6 | 1100 | cur_trans->commit_done = 1; |
b7ec40d7 | 1101 | |
15ee9bc7 | 1102 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1103 | |
2c90e5d6 | 1104 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1105 | |
78fae27e | 1106 | put_transaction(cur_trans); |
79154b1b | 1107 | put_transaction(cur_trans); |
58176a96 | 1108 | |
78fae27e | 1109 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1110 | |
9ed74f2d JB |
1111 | if (current->journal_info == trans) |
1112 | current->journal_info = NULL; | |
1113 | ||
2c90e5d6 | 1114 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1115 | |
1116 | if (current != root->fs_info->transaction_kthread) | |
1117 | btrfs_run_delayed_iputs(root); | |
1118 | ||
79154b1b CM |
1119 | return ret; |
1120 | } | |
1121 | ||
d352ac68 CM |
1122 | /* |
1123 | * interface function to delete all the snapshots we have scheduled for deletion | |
1124 | */ | |
e9d0b13b CM |
1125 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1126 | { | |
5d4f98a2 YZ |
1127 | LIST_HEAD(list); |
1128 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1129 | ||
1130 | mutex_lock(&fs_info->trans_mutex); | |
1131 | list_splice_init(&fs_info->dead_roots, &list); | |
1132 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1133 | |
5d4f98a2 YZ |
1134 | while (!list_empty(&list)) { |
1135 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1136 | list_del(&root->root_list); |
1137 | ||
1138 | if (btrfs_header_backref_rev(root->node) < | |
1139 | BTRFS_MIXED_BACKREF_REV) | |
1140 | btrfs_drop_snapshot(root, 0); | |
1141 | else | |
1142 | btrfs_drop_snapshot(root, 1); | |
e9d0b13b CM |
1143 | } |
1144 | return 0; | |
1145 | } |