]>
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); | |
150 | if (cur_trans->blocked) { | |
151 | mutex_unlock(&root->fs_info->trans_mutex); | |
152 | schedule(); | |
153 | mutex_lock(&root->fs_info->trans_mutex); | |
154 | finish_wait(&root->fs_info->transaction_wait, | |
155 | &wait); | |
156 | } else { | |
157 | finish_wait(&root->fs_info->transaction_wait, | |
158 | &wait); | |
159 | break; | |
160 | } | |
161 | } | |
162 | put_transaction(cur_trans); | |
163 | } | |
37d1aeee CM |
164 | } |
165 | ||
249ac1e5 JB |
166 | enum btrfs_trans_type { |
167 | TRANS_START, | |
168 | TRANS_JOIN, | |
169 | TRANS_USERSPACE, | |
170 | }; | |
171 | ||
e02119d5 | 172 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
249ac1e5 | 173 | int num_blocks, int type) |
37d1aeee CM |
174 | { |
175 | struct btrfs_trans_handle *h = | |
176 | kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
177 | int ret; | |
178 | ||
179 | mutex_lock(&root->fs_info->trans_mutex); | |
4bef0848 | 180 | if (!root->fs_info->log_root_recovering && |
249ac1e5 JB |
181 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || |
182 | type == TRANS_USERSPACE)) | |
37d1aeee | 183 | wait_current_trans(root); |
79154b1b CM |
184 | ret = join_transaction(root); |
185 | BUG_ON(ret); | |
0f7d52f4 | 186 | |
6702ed49 | 187 | h->transid = root->fs_info->running_transaction->transid; |
79154b1b CM |
188 | h->transaction = root->fs_info->running_transaction; |
189 | h->blocks_reserved = num_blocks; | |
190 | h->blocks_used = 0; | |
d2fb3437 | 191 | h->block_group = 0; |
26b8003f CM |
192 | h->alloc_exclude_nr = 0; |
193 | h->alloc_exclude_start = 0; | |
56bec294 | 194 | h->delayed_ref_updates = 0; |
b7ec40d7 | 195 | |
249ac1e5 | 196 | if (!current->journal_info && type != TRANS_USERSPACE) |
9ed74f2d JB |
197 | current->journal_info = h; |
198 | ||
79154b1b | 199 | root->fs_info->running_transaction->use_count++; |
5d4f98a2 | 200 | record_root_in_trans(h, root); |
79154b1b CM |
201 | mutex_unlock(&root->fs_info->trans_mutex); |
202 | return h; | |
203 | } | |
204 | ||
f9295749 CM |
205 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
206 | int num_blocks) | |
207 | { | |
249ac1e5 | 208 | return start_transaction(root, num_blocks, TRANS_START); |
f9295749 CM |
209 | } |
210 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
211 | int num_blocks) | |
212 | { | |
249ac1e5 | 213 | return start_transaction(root, num_blocks, TRANS_JOIN); |
f9295749 CM |
214 | } |
215 | ||
9ca9ee09 SW |
216 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
217 | int num_blocks) | |
218 | { | |
249ac1e5 | 219 | return start_transaction(r, num_blocks, TRANS_USERSPACE); |
9ca9ee09 SW |
220 | } |
221 | ||
d352ac68 | 222 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
223 | static noinline int wait_for_commit(struct btrfs_root *root, |
224 | struct btrfs_transaction *commit) | |
225 | { | |
226 | DEFINE_WAIT(wait); | |
227 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 228 | while (!commit->commit_done) { |
89ce8a63 CM |
229 | prepare_to_wait(&commit->commit_wait, &wait, |
230 | TASK_UNINTERRUPTIBLE); | |
231 | if (commit->commit_done) | |
232 | break; | |
233 | mutex_unlock(&root->fs_info->trans_mutex); | |
234 | schedule(); | |
235 | mutex_lock(&root->fs_info->trans_mutex); | |
236 | } | |
237 | mutex_unlock(&root->fs_info->trans_mutex); | |
238 | finish_wait(&commit->commit_wait, &wait); | |
239 | return 0; | |
240 | } | |
241 | ||
5d4f98a2 | 242 | #if 0 |
d352ac68 | 243 | /* |
d397712b CM |
244 | * rate limit against the drop_snapshot code. This helps to slow down new |
245 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 246 | */ |
37d1aeee | 247 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
248 | { |
249 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 250 | int harder_count = 0; |
ab78c84d | 251 | |
2dd3e67b | 252 | harder: |
ab78c84d CM |
253 | if (atomic_read(&info->throttles)) { |
254 | DEFINE_WAIT(wait); | |
255 | int thr; | |
ab78c84d CM |
256 | thr = atomic_read(&info->throttle_gen); |
257 | ||
258 | do { | |
259 | prepare_to_wait(&info->transaction_throttle, | |
260 | &wait, TASK_UNINTERRUPTIBLE); | |
261 | if (!atomic_read(&info->throttles)) { | |
262 | finish_wait(&info->transaction_throttle, &wait); | |
263 | break; | |
264 | } | |
265 | schedule(); | |
266 | finish_wait(&info->transaction_throttle, &wait); | |
267 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
268 | harder_count++; |
269 | ||
270 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
271 | harder_count < 2) | |
272 | goto harder; | |
273 | ||
274 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
275 | harder_count < 10) | |
276 | goto harder; | |
277 | ||
278 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
279 | harder_count < 20) | |
280 | goto harder; | |
ab78c84d CM |
281 | } |
282 | } | |
5d4f98a2 | 283 | #endif |
ab78c84d | 284 | |
37d1aeee CM |
285 | void btrfs_throttle(struct btrfs_root *root) |
286 | { | |
287 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
288 | if (!root->fs_info->open_ioctl_trans) |
289 | wait_current_trans(root); | |
37d1aeee | 290 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
291 | } |
292 | ||
89ce8a63 CM |
293 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
294 | struct btrfs_root *root, int throttle) | |
79154b1b CM |
295 | { |
296 | struct btrfs_transaction *cur_trans; | |
ab78c84d | 297 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
298 | int count = 0; |
299 | ||
300 | while (count < 4) { | |
301 | unsigned long cur = trans->delayed_ref_updates; | |
302 | trans->delayed_ref_updates = 0; | |
303 | if (cur && | |
304 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
305 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
306 | |
307 | /* | |
308 | * do a full flush if the transaction is trying | |
309 | * to close | |
310 | */ | |
311 | if (trans->transaction->delayed_refs.flushing) | |
312 | cur = 0; | |
c3e69d58 CM |
313 | btrfs_run_delayed_refs(trans, root, cur); |
314 | } else { | |
315 | break; | |
316 | } | |
317 | count++; | |
56bec294 CM |
318 | } |
319 | ||
ab78c84d CM |
320 | mutex_lock(&info->trans_mutex); |
321 | cur_trans = info->running_transaction; | |
ccd467d6 | 322 | WARN_ON(cur_trans != trans->transaction); |
d5719762 | 323 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 324 | cur_trans->num_writers--; |
89ce8a63 | 325 | |
79154b1b CM |
326 | if (waitqueue_active(&cur_trans->writer_wait)) |
327 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 328 | put_transaction(cur_trans); |
ab78c84d | 329 | mutex_unlock(&info->trans_mutex); |
9ed74f2d JB |
330 | |
331 | if (current->journal_info == trans) | |
332 | current->journal_info = NULL; | |
d6025579 | 333 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 334 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 335 | |
24bbcf04 YZ |
336 | if (throttle) |
337 | btrfs_run_delayed_iputs(root); | |
338 | ||
79154b1b CM |
339 | return 0; |
340 | } | |
341 | ||
89ce8a63 CM |
342 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
343 | struct btrfs_root *root) | |
344 | { | |
345 | return __btrfs_end_transaction(trans, root, 0); | |
346 | } | |
347 | ||
348 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
349 | struct btrfs_root *root) | |
350 | { | |
351 | return __btrfs_end_transaction(trans, root, 1); | |
352 | } | |
353 | ||
d352ac68 CM |
354 | /* |
355 | * when btree blocks are allocated, they have some corresponding bits set for | |
356 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 357 | * those extents are sent to disk but does not wait on them |
d352ac68 | 358 | */ |
690587d1 | 359 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 360 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 361 | { |
7c4452b9 | 362 | int ret; |
777e6bd7 | 363 | int err = 0; |
7c4452b9 CM |
364 | int werr = 0; |
365 | struct page *page; | |
7c4452b9 | 366 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 367 | u64 start = 0; |
5f39d397 CM |
368 | u64 end; |
369 | unsigned long index; | |
7c4452b9 | 370 | |
d397712b | 371 | while (1) { |
777e6bd7 | 372 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 373 | mark); |
5f39d397 | 374 | if (ret) |
7c4452b9 | 375 | break; |
d397712b | 376 | while (start <= end) { |
777e6bd7 CM |
377 | cond_resched(); |
378 | ||
5f39d397 | 379 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 380 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 381 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
382 | if (!page) |
383 | continue; | |
4bef0848 CM |
384 | |
385 | btree_lock_page_hook(page); | |
386 | if (!page->mapping) { | |
387 | unlock_page(page); | |
388 | page_cache_release(page); | |
389 | continue; | |
390 | } | |
391 | ||
6702ed49 CM |
392 | if (PageWriteback(page)) { |
393 | if (PageDirty(page)) | |
394 | wait_on_page_writeback(page); | |
395 | else { | |
396 | unlock_page(page); | |
397 | page_cache_release(page); | |
398 | continue; | |
399 | } | |
400 | } | |
7c4452b9 CM |
401 | err = write_one_page(page, 0); |
402 | if (err) | |
403 | werr = err; | |
404 | page_cache_release(page); | |
405 | } | |
406 | } | |
690587d1 CM |
407 | if (err) |
408 | werr = err; | |
409 | return werr; | |
410 | } | |
411 | ||
412 | /* | |
413 | * when btree blocks are allocated, they have some corresponding bits set for | |
414 | * them in one of two extent_io trees. This is used to make sure all of | |
415 | * those extents are on disk for transaction or log commit. We wait | |
416 | * on all the pages and clear them from the dirty pages state tree | |
417 | */ | |
418 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 419 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
420 | { |
421 | int ret; | |
422 | int err = 0; | |
423 | int werr = 0; | |
424 | struct page *page; | |
425 | struct inode *btree_inode = root->fs_info->btree_inode; | |
426 | u64 start = 0; | |
427 | u64 end; | |
428 | unsigned long index; | |
429 | ||
d397712b | 430 | while (1) { |
8cef4e16 YZ |
431 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
432 | mark); | |
777e6bd7 CM |
433 | if (ret) |
434 | break; | |
435 | ||
8cef4e16 | 436 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 437 | while (start <= end) { |
777e6bd7 CM |
438 | index = start >> PAGE_CACHE_SHIFT; |
439 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
440 | page = find_get_page(btree_inode->i_mapping, index); | |
441 | if (!page) | |
442 | continue; | |
443 | if (PageDirty(page)) { | |
4bef0848 CM |
444 | btree_lock_page_hook(page); |
445 | wait_on_page_writeback(page); | |
777e6bd7 CM |
446 | err = write_one_page(page, 0); |
447 | if (err) | |
448 | werr = err; | |
449 | } | |
105d931d | 450 | wait_on_page_writeback(page); |
777e6bd7 CM |
451 | page_cache_release(page); |
452 | cond_resched(); | |
453 | } | |
454 | } | |
7c4452b9 CM |
455 | if (err) |
456 | werr = err; | |
457 | return werr; | |
79154b1b CM |
458 | } |
459 | ||
690587d1 CM |
460 | /* |
461 | * when btree blocks are allocated, they have some corresponding bits set for | |
462 | * them in one of two extent_io trees. This is used to make sure all of | |
463 | * those extents are on disk for transaction or log commit | |
464 | */ | |
465 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 466 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
467 | { |
468 | int ret; | |
469 | int ret2; | |
470 | ||
8cef4e16 YZ |
471 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
472 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
473 | return ret || ret2; |
474 | } | |
475 | ||
d0c803c4 CM |
476 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
477 | struct btrfs_root *root) | |
478 | { | |
479 | if (!trans || !trans->transaction) { | |
480 | struct inode *btree_inode; | |
481 | btree_inode = root->fs_info->btree_inode; | |
482 | return filemap_write_and_wait(btree_inode->i_mapping); | |
483 | } | |
484 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
485 | &trans->transaction->dirty_pages, |
486 | EXTENT_DIRTY); | |
d0c803c4 CM |
487 | } |
488 | ||
d352ac68 CM |
489 | /* |
490 | * this is used to update the root pointer in the tree of tree roots. | |
491 | * | |
492 | * But, in the case of the extent allocation tree, updating the root | |
493 | * pointer may allocate blocks which may change the root of the extent | |
494 | * allocation tree. | |
495 | * | |
496 | * So, this loops and repeats and makes sure the cowonly root didn't | |
497 | * change while the root pointer was being updated in the metadata. | |
498 | */ | |
0b86a832 CM |
499 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
500 | struct btrfs_root *root) | |
79154b1b CM |
501 | { |
502 | int ret; | |
0b86a832 | 503 | u64 old_root_bytenr; |
86b9f2ec | 504 | u64 old_root_used; |
0b86a832 | 505 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 506 | |
86b9f2ec | 507 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 508 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 509 | |
d397712b | 510 | while (1) { |
0b86a832 | 511 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
512 | if (old_root_bytenr == root->node->start && |
513 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 514 | break; |
87ef2bb4 | 515 | |
5d4f98a2 | 516 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 517 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
518 | &root->root_key, |
519 | &root->root_item); | |
79154b1b | 520 | BUG_ON(ret); |
56bec294 | 521 | |
86b9f2ec | 522 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 523 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 524 | BUG_ON(ret); |
0b86a832 | 525 | } |
276e680d YZ |
526 | |
527 | if (root != root->fs_info->extent_root) | |
528 | switch_commit_root(root); | |
529 | ||
0b86a832 CM |
530 | return 0; |
531 | } | |
532 | ||
d352ac68 CM |
533 | /* |
534 | * update all the cowonly tree roots on disk | |
535 | */ | |
5d4f98a2 YZ |
536 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
537 | struct btrfs_root *root) | |
0b86a832 CM |
538 | { |
539 | struct btrfs_fs_info *fs_info = root->fs_info; | |
540 | struct list_head *next; | |
84234f3a | 541 | struct extent_buffer *eb; |
56bec294 | 542 | int ret; |
84234f3a | 543 | |
56bec294 CM |
544 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
545 | BUG_ON(ret); | |
87ef2bb4 | 546 | |
84234f3a | 547 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 548 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
549 | btrfs_tree_unlock(eb); |
550 | free_extent_buffer(eb); | |
0b86a832 | 551 | |
56bec294 CM |
552 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
553 | BUG_ON(ret); | |
87ef2bb4 | 554 | |
d397712b | 555 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
556 | next = fs_info->dirty_cowonly_roots.next; |
557 | list_del_init(next); | |
558 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 559 | |
0b86a832 | 560 | update_cowonly_root(trans, root); |
79154b1b | 561 | } |
276e680d YZ |
562 | |
563 | down_write(&fs_info->extent_commit_sem); | |
564 | switch_commit_root(fs_info->extent_root); | |
565 | up_write(&fs_info->extent_commit_sem); | |
566 | ||
79154b1b CM |
567 | return 0; |
568 | } | |
569 | ||
d352ac68 CM |
570 | /* |
571 | * dead roots are old snapshots that need to be deleted. This allocates | |
572 | * a dirty root struct and adds it into the list of dead roots that need to | |
573 | * be deleted | |
574 | */ | |
5d4f98a2 | 575 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 576 | { |
b48652c1 | 577 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 578 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 579 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
580 | return 0; |
581 | } | |
582 | ||
d352ac68 | 583 | /* |
5d4f98a2 | 584 | * update all the cowonly tree roots on disk |
d352ac68 | 585 | */ |
5d4f98a2 YZ |
586 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
587 | struct btrfs_root *root) | |
0f7d52f4 | 588 | { |
0f7d52f4 | 589 | struct btrfs_root *gang[8]; |
5d4f98a2 | 590 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
591 | int i; |
592 | int ret; | |
54aa1f4d CM |
593 | int err = 0; |
594 | ||
d397712b | 595 | while (1) { |
5d4f98a2 YZ |
596 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
597 | (void **)gang, 0, | |
0f7d52f4 CM |
598 | ARRAY_SIZE(gang), |
599 | BTRFS_ROOT_TRANS_TAG); | |
600 | if (ret == 0) | |
601 | break; | |
602 | for (i = 0; i < ret; i++) { | |
603 | root = gang[i]; | |
5d4f98a2 YZ |
604 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
605 | (unsigned long)root->root_key.objectid, | |
606 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 607 | |
e02119d5 | 608 | btrfs_free_log(trans, root); |
5d4f98a2 | 609 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 610 | |
978d910d | 611 | if (root->commit_root != root->node) { |
817d52f8 | 612 | switch_commit_root(root); |
978d910d YZ |
613 | btrfs_set_root_node(&root->root_item, |
614 | root->node); | |
615 | } | |
5d4f98a2 | 616 | |
5d4f98a2 | 617 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
618 | &root->root_key, |
619 | &root->root_item); | |
54aa1f4d CM |
620 | if (err) |
621 | break; | |
0f7d52f4 CM |
622 | } |
623 | } | |
54aa1f4d | 624 | return err; |
0f7d52f4 CM |
625 | } |
626 | ||
d352ac68 CM |
627 | /* |
628 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
629 | * otherwise every leaf in the btree is read and defragged. | |
630 | */ | |
e9d0b13b CM |
631 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
632 | { | |
633 | struct btrfs_fs_info *info = root->fs_info; | |
634 | int ret; | |
635 | struct btrfs_trans_handle *trans; | |
d3c2fdcf | 636 | unsigned long nr; |
e9d0b13b | 637 | |
a2135011 | 638 | smp_mb(); |
e9d0b13b CM |
639 | if (root->defrag_running) |
640 | return 0; | |
e9d0b13b | 641 | trans = btrfs_start_transaction(root, 1); |
6b80053d | 642 | while (1) { |
e9d0b13b CM |
643 | root->defrag_running = 1; |
644 | ret = btrfs_defrag_leaves(trans, root, cacheonly); | |
d3c2fdcf | 645 | nr = trans->blocks_used; |
e9d0b13b | 646 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 647 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
648 | cond_resched(); |
649 | ||
e9d0b13b | 650 | trans = btrfs_start_transaction(root, 1); |
3f157a2f | 651 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
652 | break; |
653 | } | |
654 | root->defrag_running = 0; | |
a2135011 | 655 | smp_mb(); |
e9d0b13b CM |
656 | btrfs_end_transaction(trans, root); |
657 | return 0; | |
658 | } | |
659 | ||
2c47e605 | 660 | #if 0 |
b7ec40d7 CM |
661 | /* |
662 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
663 | * sense not to join the transaction while it is trying to flush the current | |
664 | * queue of delayed refs out. | |
665 | * | |
666 | * This is used by the drop snapshot code only | |
667 | */ | |
668 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
669 | { | |
670 | DEFINE_WAIT(wait); | |
671 | ||
672 | mutex_lock(&info->trans_mutex); | |
673 | while (info->running_transaction && | |
674 | info->running_transaction->delayed_refs.flushing) { | |
675 | prepare_to_wait(&info->transaction_wait, &wait, | |
676 | TASK_UNINTERRUPTIBLE); | |
677 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 678 | |
b7ec40d7 | 679 | schedule(); |
59bc5c75 | 680 | |
b7ec40d7 CM |
681 | mutex_lock(&info->trans_mutex); |
682 | finish_wait(&info->transaction_wait, &wait); | |
683 | } | |
684 | mutex_unlock(&info->trans_mutex); | |
685 | return 0; | |
686 | } | |
687 | ||
d352ac68 CM |
688 | /* |
689 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
690 | * all of them | |
691 | */ | |
5d4f98a2 | 692 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 693 | { |
0f7d52f4 | 694 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 695 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 696 | unsigned long nr; |
5d4f98a2 | 697 | int ret; |
58176a96 | 698 | |
5d4f98a2 YZ |
699 | while (1) { |
700 | /* | |
701 | * we don't want to jump in and create a bunch of | |
702 | * delayed refs if the transaction is starting to close | |
703 | */ | |
704 | wait_transaction_pre_flush(tree_root->fs_info); | |
705 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 706 | |
5d4f98a2 YZ |
707 | /* |
708 | * we've joined a transaction, make sure it isn't | |
709 | * closing right now | |
710 | */ | |
711 | if (trans->transaction->delayed_refs.flushing) { | |
712 | btrfs_end_transaction(trans, tree_root); | |
713 | continue; | |
9f3a7427 | 714 | } |
58176a96 | 715 | |
5d4f98a2 YZ |
716 | ret = btrfs_drop_snapshot(trans, root); |
717 | if (ret != -EAGAIN) | |
718 | break; | |
a2135011 | 719 | |
5d4f98a2 YZ |
720 | ret = btrfs_update_root(trans, tree_root, |
721 | &root->root_key, | |
722 | &root->root_item); | |
723 | if (ret) | |
54aa1f4d | 724 | break; |
bcc63abb | 725 | |
d3c2fdcf | 726 | nr = trans->blocks_used; |
0f7d52f4 CM |
727 | ret = btrfs_end_transaction(trans, tree_root); |
728 | BUG_ON(ret); | |
5eda7b5e | 729 | |
d3c2fdcf | 730 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 731 | cond_resched(); |
0f7d52f4 | 732 | } |
5d4f98a2 YZ |
733 | BUG_ON(ret); |
734 | ||
735 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
736 | BUG_ON(ret); | |
737 | ||
738 | nr = trans->blocks_used; | |
739 | ret = btrfs_end_transaction(trans, tree_root); | |
740 | BUG_ON(ret); | |
741 | ||
742 | free_extent_buffer(root->node); | |
743 | free_extent_buffer(root->commit_root); | |
744 | kfree(root); | |
745 | ||
746 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 747 | return ret; |
0f7d52f4 | 748 | } |
2c47e605 | 749 | #endif |
0f7d52f4 | 750 | |
d352ac68 CM |
751 | /* |
752 | * new snapshots need to be created at a very specific time in the | |
753 | * transaction commit. This does the actual creation | |
754 | */ | |
80b6794d | 755 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
756 | struct btrfs_fs_info *fs_info, |
757 | struct btrfs_pending_snapshot *pending) | |
758 | { | |
759 | struct btrfs_key key; | |
80b6794d | 760 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
761 | struct btrfs_root *tree_root = fs_info->tree_root; |
762 | struct btrfs_root *root = pending->root; | |
763 | struct extent_buffer *tmp; | |
925baedd | 764 | struct extent_buffer *old; |
3063d29f CM |
765 | int ret; |
766 | u64 objectid; | |
767 | ||
80b6794d CM |
768 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
769 | if (!new_root_item) { | |
770 | ret = -ENOMEM; | |
771 | goto fail; | |
772 | } | |
3063d29f CM |
773 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
774 | if (ret) | |
775 | goto fail; | |
776 | ||
5d4f98a2 | 777 | record_root_in_trans(trans, root); |
80ff3856 | 778 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
80b6794d | 779 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
3063d29f CM |
780 | |
781 | key.objectid = objectid; | |
1c4850e2 YZ |
782 | /* record when the snapshot was created in key.offset */ |
783 | key.offset = trans->transid; | |
3063d29f CM |
784 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
785 | ||
925baedd | 786 | old = btrfs_lock_root_node(root); |
9fa8cfe7 | 787 | btrfs_cow_block(trans, root, old, NULL, 0, &old); |
5d4f98a2 | 788 | btrfs_set_lock_blocking(old); |
3063d29f | 789 | |
925baedd CM |
790 | btrfs_copy_root(trans, root, old, &tmp, objectid); |
791 | btrfs_tree_unlock(old); | |
792 | free_extent_buffer(old); | |
3063d29f | 793 | |
5d4f98a2 | 794 | btrfs_set_root_node(new_root_item, tmp); |
3063d29f | 795 | ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, |
80b6794d | 796 | new_root_item); |
925baedd | 797 | btrfs_tree_unlock(tmp); |
3063d29f CM |
798 | free_extent_buffer(tmp); |
799 | if (ret) | |
800 | goto fail; | |
801 | ||
3de4586c CM |
802 | key.offset = (u64)-1; |
803 | memcpy(&pending->root_key, &key, sizeof(key)); | |
804 | fail: | |
805 | kfree(new_root_item); | |
806 | return ret; | |
807 | } | |
808 | ||
809 | static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info, | |
810 | struct btrfs_pending_snapshot *pending) | |
811 | { | |
812 | int ret; | |
813 | int namelen; | |
814 | u64 index = 0; | |
815 | struct btrfs_trans_handle *trans; | |
816 | struct inode *parent_inode; | |
0660b5af | 817 | struct btrfs_root *parent_root; |
3de4586c | 818 | |
3394e160 | 819 | parent_inode = pending->dentry->d_parent->d_inode; |
0660b5af | 820 | parent_root = BTRFS_I(parent_inode)->root; |
180591bc | 821 | trans = btrfs_join_transaction(parent_root, 1); |
3de4586c | 822 | |
3063d29f CM |
823 | /* |
824 | * insert the directory item | |
825 | */ | |
3b96362c | 826 | namelen = strlen(pending->name); |
3de4586c | 827 | ret = btrfs_set_inode_index(parent_inode, &index); |
0660b5af | 828 | ret = btrfs_insert_dir_item(trans, parent_root, |
3de4586c CM |
829 | pending->name, namelen, |
830 | parent_inode->i_ino, | |
831 | &pending->root_key, BTRFS_FT_DIR, index); | |
3063d29f CM |
832 | |
833 | if (ret) | |
834 | goto fail; | |
0660b5af | 835 | |
52c26179 YZ |
836 | btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2); |
837 | ret = btrfs_update_inode(trans, parent_root, parent_inode); | |
838 | BUG_ON(ret); | |
839 | ||
0660b5af CM |
840 | ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root, |
841 | pending->root_key.objectid, | |
0660b5af CM |
842 | parent_root->root_key.objectid, |
843 | parent_inode->i_ino, index, pending->name, | |
844 | namelen); | |
845 | ||
846 | BUG_ON(ret); | |
847 | ||
3063d29f | 848 | fail: |
3de4586c | 849 | btrfs_end_transaction(trans, fs_info->fs_root); |
3063d29f CM |
850 | return ret; |
851 | } | |
852 | ||
d352ac68 CM |
853 | /* |
854 | * create all the snapshots we've scheduled for creation | |
855 | */ | |
80b6794d CM |
856 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
857 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
858 | { |
859 | struct btrfs_pending_snapshot *pending; | |
860 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
861 | int ret; |
862 | ||
c6e30871 | 863 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
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 | ||
d397712b | 877 | while (!list_empty(head)) { |
3063d29f CM |
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 | ||
5d4f98a2 YZ |
889 | static void update_super_roots(struct btrfs_root *root) |
890 | { | |
891 | struct btrfs_root_item *root_item; | |
892 | struct btrfs_super_block *super; | |
893 | ||
894 | super = &root->fs_info->super_copy; | |
895 | ||
896 | root_item = &root->fs_info->chunk_root->root_item; | |
897 | super->chunk_root = root_item->bytenr; | |
898 | super->chunk_root_generation = root_item->generation; | |
899 | super->chunk_root_level = root_item->level; | |
900 | ||
901 | root_item = &root->fs_info->tree_root->root_item; | |
902 | super->root = root_item->bytenr; | |
903 | super->generation = root_item->generation; | |
904 | super->root_level = root_item->level; | |
905 | } | |
906 | ||
f36f3042 CM |
907 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
908 | { | |
909 | int ret = 0; | |
910 | spin_lock(&info->new_trans_lock); | |
911 | if (info->running_transaction) | |
912 | ret = info->running_transaction->in_commit; | |
913 | spin_unlock(&info->new_trans_lock); | |
914 | return ret; | |
915 | } | |
916 | ||
79154b1b CM |
917 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
918 | struct btrfs_root *root) | |
919 | { | |
15ee9bc7 JB |
920 | unsigned long joined = 0; |
921 | unsigned long timeout = 1; | |
79154b1b | 922 | struct btrfs_transaction *cur_trans; |
8fd17795 | 923 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 924 | DEFINE_WAIT(wait); |
15ee9bc7 | 925 | int ret; |
89573b9c CM |
926 | int should_grow = 0; |
927 | unsigned long now = get_seconds(); | |
dccae999 | 928 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 929 | |
5a3f23d5 CM |
930 | btrfs_run_ordered_operations(root, 0); |
931 | ||
56bec294 CM |
932 | /* make a pass through all the delayed refs we have so far |
933 | * any runnings procs may add more while we are here | |
934 | */ | |
935 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
936 | BUG_ON(ret); | |
937 | ||
b7ec40d7 | 938 | cur_trans = trans->transaction; |
56bec294 CM |
939 | /* |
940 | * set the flushing flag so procs in this transaction have to | |
941 | * start sending their work down. | |
942 | */ | |
b7ec40d7 | 943 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 944 | |
c3e69d58 | 945 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
946 | BUG_ON(ret); |
947 | ||
79154b1b | 948 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
949 | if (cur_trans->in_commit) { |
950 | cur_trans->use_count++; | |
ccd467d6 | 951 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 952 | btrfs_end_transaction(trans, root); |
ccd467d6 | 953 | |
79154b1b CM |
954 | ret = wait_for_commit(root, cur_trans); |
955 | BUG_ON(ret); | |
15ee9bc7 JB |
956 | |
957 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 958 | put_transaction(cur_trans); |
15ee9bc7 JB |
959 | mutex_unlock(&root->fs_info->trans_mutex); |
960 | ||
79154b1b CM |
961 | return 0; |
962 | } | |
4313b399 | 963 | |
2c90e5d6 | 964 | trans->transaction->in_commit = 1; |
f9295749 | 965 | trans->transaction->blocked = 1; |
ccd467d6 CM |
966 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
967 | prev_trans = list_entry(cur_trans->list.prev, | |
968 | struct btrfs_transaction, list); | |
969 | if (!prev_trans->commit_done) { | |
970 | prev_trans->use_count++; | |
ccd467d6 CM |
971 | mutex_unlock(&root->fs_info->trans_mutex); |
972 | ||
973 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 974 | |
ccd467d6 | 975 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 976 | put_transaction(prev_trans); |
ccd467d6 CM |
977 | } |
978 | } | |
15ee9bc7 | 979 | |
89573b9c CM |
980 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
981 | should_grow = 1; | |
982 | ||
15ee9bc7 | 983 | do { |
7ea394f1 | 984 | int snap_pending = 0; |
15ee9bc7 | 985 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
986 | if (!list_empty(&trans->transaction->pending_snapshots)) |
987 | snap_pending = 1; | |
988 | ||
2c90e5d6 | 989 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 990 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
79154b1b | 991 | TASK_UNINTERRUPTIBLE); |
15ee9bc7 JB |
992 | |
993 | if (cur_trans->num_writers > 1) | |
994 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 995 | else if (should_grow) |
15ee9bc7 JB |
996 | timeout = 1; |
997 | ||
79154b1b | 998 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 999 | |
0bdb1db2 | 1000 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1001 | btrfs_start_delalloc_inodes(root, 1); |
1002 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1003 | BUG_ON(ret); |
7ea394f1 YZ |
1004 | } |
1005 | ||
5a3f23d5 CM |
1006 | /* |
1007 | * rename don't use btrfs_join_transaction, so, once we | |
1008 | * set the transaction to blocked above, we aren't going | |
1009 | * to get any new ordered operations. We can safely run | |
1010 | * it here and no for sure that nothing new will be added | |
1011 | * to the list | |
1012 | */ | |
1013 | btrfs_run_ordered_operations(root, 1); | |
1014 | ||
89573b9c CM |
1015 | smp_mb(); |
1016 | if (cur_trans->num_writers > 1 || should_grow) | |
1017 | schedule_timeout(timeout); | |
15ee9bc7 | 1018 | |
79154b1b | 1019 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1020 | finish_wait(&cur_trans->writer_wait, &wait); |
1021 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1022 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1023 | |
3063d29f CM |
1024 | ret = create_pending_snapshots(trans, root->fs_info); |
1025 | BUG_ON(ret); | |
1026 | ||
56bec294 CM |
1027 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1028 | BUG_ON(ret); | |
1029 | ||
2c90e5d6 | 1030 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1031 | |
e02119d5 CM |
1032 | /* btrfs_commit_tree_roots is responsible for getting the |
1033 | * various roots consistent with each other. Every pointer | |
1034 | * in the tree of tree roots has to point to the most up to date | |
1035 | * root for every subvolume and other tree. So, we have to keep | |
1036 | * the tree logging code from jumping in and changing any | |
1037 | * of the trees. | |
1038 | * | |
1039 | * At this point in the commit, there can't be any tree-log | |
1040 | * writers, but a little lower down we drop the trans mutex | |
1041 | * and let new people in. By holding the tree_log_mutex | |
1042 | * from now until after the super is written, we avoid races | |
1043 | * with the tree-log code. | |
1044 | */ | |
1045 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1046 | ||
5d4f98a2 | 1047 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1048 | BUG_ON(ret); |
1049 | ||
5d4f98a2 | 1050 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1051 | * safe to free the root of tree log roots |
1052 | */ | |
1053 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1054 | ||
5d4f98a2 | 1055 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1056 | BUG_ON(ret); |
54aa1f4d | 1057 | |
11833d66 YZ |
1058 | btrfs_prepare_extent_commit(trans, root); |
1059 | ||
78fae27e | 1060 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1061 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1062 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1063 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1064 | |
1065 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1066 | root->fs_info->tree_root->node); | |
817d52f8 | 1067 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1068 | |
1069 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1070 | root->fs_info->chunk_root->node); | |
817d52f8 | 1071 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1072 | |
1073 | update_super_roots(root); | |
e02119d5 CM |
1074 | |
1075 | if (!root->fs_info->log_root_recovering) { | |
1076 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1077 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1078 | } | |
1079 | ||
a061fc8d CM |
1080 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1081 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1082 | |
f9295749 | 1083 | trans->transaction->blocked = 0; |
b7ec40d7 | 1084 | |
f9295749 | 1085 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1086 | |
78fae27e | 1087 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1088 | ret = btrfs_write_and_wait_transaction(trans, root); |
1089 | BUG_ON(ret); | |
a512bbf8 | 1090 | write_ctree_super(trans, root, 0); |
4313b399 | 1091 | |
e02119d5 CM |
1092 | /* |
1093 | * the super is written, we can safely allow the tree-loggers | |
1094 | * to go about their business | |
1095 | */ | |
1096 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1097 | ||
11833d66 | 1098 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1099 | |
3de4586c CM |
1100 | /* do the directory inserts of any pending snapshot creations */ |
1101 | finish_pending_snapshots(trans, root->fs_info); | |
1102 | ||
1a40e23b ZY |
1103 | mutex_lock(&root->fs_info->trans_mutex); |
1104 | ||
2c90e5d6 | 1105 | cur_trans->commit_done = 1; |
b7ec40d7 | 1106 | |
15ee9bc7 | 1107 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1108 | |
2c90e5d6 | 1109 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1110 | |
78fae27e | 1111 | put_transaction(cur_trans); |
79154b1b | 1112 | put_transaction(cur_trans); |
58176a96 | 1113 | |
78fae27e | 1114 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1115 | |
9ed74f2d JB |
1116 | if (current->journal_info == trans) |
1117 | current->journal_info = NULL; | |
1118 | ||
2c90e5d6 | 1119 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1120 | |
1121 | if (current != root->fs_info->transaction_kthread) | |
1122 | btrfs_run_delayed_iputs(root); | |
1123 | ||
79154b1b CM |
1124 | return ret; |
1125 | } | |
1126 | ||
d352ac68 CM |
1127 | /* |
1128 | * interface function to delete all the snapshots we have scheduled for deletion | |
1129 | */ | |
e9d0b13b CM |
1130 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1131 | { | |
5d4f98a2 YZ |
1132 | LIST_HEAD(list); |
1133 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1134 | ||
1135 | mutex_lock(&fs_info->trans_mutex); | |
1136 | list_splice_init(&fs_info->dead_roots, &list); | |
1137 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1138 | |
5d4f98a2 YZ |
1139 | while (!list_empty(&list)) { |
1140 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1141 | list_del(&root->root_list); |
1142 | ||
1143 | if (btrfs_header_backref_rev(root->node) < | |
1144 | BTRFS_MIXED_BACKREF_REV) | |
1145 | btrfs_drop_snapshot(root, 0); | |
1146 | else | |
1147 | btrfs_drop_snapshot(root, 1); | |
e9d0b13b CM |
1148 | } |
1149 | return 0; | |
1150 | } |