Btrfs: async-thread: fix possible memory leak
[linux-2.6-block.git] / fs / btrfs / transaction.c
CommitLineData
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>
79154b1b
CM
23#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
925baedd 26#include "locking.h"
31153d81 27#include "ref-cache.h"
79154b1b 28
78fae27e 29static int total_trans = 0;
2c90e5d6
CM
30extern struct kmem_cache *btrfs_trans_handle_cachep;
31extern struct kmem_cache *btrfs_transaction_cachep;
32
0f7d52f4
CM
33#define BTRFS_ROOT_TRANS_TAG 0
34
31153d81
YZ
35struct dirty_root {
36 struct list_head list;
37 struct btrfs_root *root;
38 struct btrfs_root *latest_root;
31153d81
YZ
39};
40
80b6794d 41static noinline void put_transaction(struct btrfs_transaction *transaction)
79154b1b 42{
2c90e5d6 43 WARN_ON(transaction->use_count == 0);
79154b1b 44 transaction->use_count--;
78fae27e
CM
45 if (transaction->use_count == 0) {
46 WARN_ON(total_trans == 0);
47 total_trans--;
8fd17795 48 list_del_init(&transaction->list);
2c90e5d6
CM
49 memset(transaction, 0, sizeof(*transaction));
50 kmem_cache_free(btrfs_transaction_cachep, transaction);
78fae27e 51 }
79154b1b
CM
52}
53
80b6794d 54static noinline int join_transaction(struct btrfs_root *root)
79154b1b
CM
55{
56 struct btrfs_transaction *cur_trans;
57 cur_trans = root->fs_info->running_transaction;
58 if (!cur_trans) {
2c90e5d6
CM
59 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
60 GFP_NOFS);
78fae27e 61 total_trans++;
79154b1b 62 BUG_ON(!cur_trans);
0f7d52f4 63 root->fs_info->generation++;
e18e4809 64 root->fs_info->last_alloc = 0;
4529ba49 65 root->fs_info->last_data_alloc = 0;
15ee9bc7
JB
66 cur_trans->num_writers = 1;
67 cur_trans->num_joined = 0;
0f7d52f4 68 cur_trans->transid = root->fs_info->generation;
79154b1b
CM
69 init_waitqueue_head(&cur_trans->writer_wait);
70 init_waitqueue_head(&cur_trans->commit_wait);
71 cur_trans->in_commit = 0;
f9295749 72 cur_trans->blocked = 0;
d5719762 73 cur_trans->use_count = 1;
79154b1b 74 cur_trans->commit_done = 0;
08607c1b 75 cur_trans->start_time = get_seconds();
3063d29f 76 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
8fd17795 77 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
d1310b2e 78 extent_io_tree_init(&cur_trans->dirty_pages,
5f39d397
CM
79 root->fs_info->btree_inode->i_mapping,
80 GFP_NOFS);
48ec2cf8
CM
81 spin_lock(&root->fs_info->new_trans_lock);
82 root->fs_info->running_transaction = cur_trans;
83 spin_unlock(&root->fs_info->new_trans_lock);
15ee9bc7
JB
84 } else {
85 cur_trans->num_writers++;
86 cur_trans->num_joined++;
79154b1b 87 }
15ee9bc7 88
79154b1b
CM
89 return 0;
90}
91
80b6794d 92static noinline int record_root_in_trans(struct btrfs_root *root)
6702ed49 93{
31153d81 94 struct dirty_root *dirty;
6702ed49
CM
95 u64 running_trans_id = root->fs_info->running_transaction->transid;
96 if (root->ref_cows && root->last_trans < running_trans_id) {
97 WARN_ON(root == root->fs_info->extent_root);
98 if (root->root_item.refs != 0) {
99 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
100 (unsigned long)root->root_key.objectid,
101 BTRFS_ROOT_TRANS_TAG);
31153d81
YZ
102
103 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
104 BUG_ON(!dirty);
105 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
106 BUG_ON(!dirty->root);
107
108 dirty->latest_root = root;
109 INIT_LIST_HEAD(&dirty->list);
31153d81 110
925baedd 111 root->commit_root = btrfs_root_node(root);
017e5369 112 root->dirty_root = dirty;
31153d81
YZ
113
114 memcpy(dirty->root, root, sizeof(*root));
017e5369
CM
115 dirty->root->ref_tree = &root->ref_tree_struct;
116
31153d81
YZ
117 spin_lock_init(&dirty->root->node_lock);
118 mutex_init(&dirty->root->objectid_mutex);
119 dirty->root->node = root->commit_root;
120 dirty->root->commit_root = NULL;
6702ed49
CM
121 } else {
122 WARN_ON(1);
123 }
124 root->last_trans = running_trans_id;
125 }
126 return 0;
127}
128
f9295749
CM
129struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
130 int num_blocks, int join)
79154b1b 131{
2c90e5d6
CM
132 struct btrfs_trans_handle *h =
133 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
f9295749 134 struct btrfs_transaction *cur_trans;
79154b1b
CM
135 int ret;
136
137 mutex_lock(&root->fs_info->trans_mutex);
f9295749
CM
138 cur_trans = root->fs_info->running_transaction;
139 if (cur_trans && cur_trans->blocked && !join) {
140 DEFINE_WAIT(wait);
141 cur_trans->use_count++;
142 while(1) {
143 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
144 TASK_UNINTERRUPTIBLE);
145 if (cur_trans->blocked) {
146 mutex_unlock(&root->fs_info->trans_mutex);
147 schedule();
148 mutex_lock(&root->fs_info->trans_mutex);
149 finish_wait(&root->fs_info->transaction_wait,
150 &wait);
151 } else {
152 finish_wait(&root->fs_info->transaction_wait,
153 &wait);
154 break;
155 }
156 }
157 put_transaction(cur_trans);
158 }
79154b1b
CM
159 ret = join_transaction(root);
160 BUG_ON(ret);
0f7d52f4 161
6702ed49
CM
162 record_root_in_trans(root);
163 h->transid = root->fs_info->running_transaction->transid;
79154b1b
CM
164 h->transaction = root->fs_info->running_transaction;
165 h->blocks_reserved = num_blocks;
166 h->blocks_used = 0;
31f3c99b 167 h->block_group = NULL;
26b8003f
CM
168 h->alloc_exclude_nr = 0;
169 h->alloc_exclude_start = 0;
79154b1b
CM
170 root->fs_info->running_transaction->use_count++;
171 mutex_unlock(&root->fs_info->trans_mutex);
172 return h;
173}
174
f9295749
CM
175struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
176 int num_blocks)
177{
178 return start_transaction(root, num_blocks, 0);
179}
180struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
181 int num_blocks)
182{
183 return start_transaction(root, num_blocks, 1);
184}
185
89ce8a63
CM
186static noinline int wait_for_commit(struct btrfs_root *root,
187 struct btrfs_transaction *commit)
188{
189 DEFINE_WAIT(wait);
190 mutex_lock(&root->fs_info->trans_mutex);
191 while(!commit->commit_done) {
192 prepare_to_wait(&commit->commit_wait, &wait,
193 TASK_UNINTERRUPTIBLE);
194 if (commit->commit_done)
195 break;
196 mutex_unlock(&root->fs_info->trans_mutex);
197 schedule();
198 mutex_lock(&root->fs_info->trans_mutex);
199 }
200 mutex_unlock(&root->fs_info->trans_mutex);
201 finish_wait(&commit->commit_wait, &wait);
202 return 0;
203}
204
ab78c84d
CM
205void btrfs_throttle(struct btrfs_root *root)
206{
207 struct btrfs_fs_info *info = root->fs_info;
208
209harder:
210 if (atomic_read(&info->throttles)) {
211 DEFINE_WAIT(wait);
212 int thr;
213 int harder_count = 0;
214 thr = atomic_read(&info->throttle_gen);
215
216 do {
217 prepare_to_wait(&info->transaction_throttle,
218 &wait, TASK_UNINTERRUPTIBLE);
219 if (!atomic_read(&info->throttles)) {
220 finish_wait(&info->transaction_throttle, &wait);
221 break;
222 }
223 schedule();
224 finish_wait(&info->transaction_throttle, &wait);
225 } while (thr == atomic_read(&info->throttle_gen));
226
227 if (harder_count < 5 &&
228 info->total_ref_cache_size > 5 * 1024 * 1024) {
229 harder_count++;
230 goto harder;
231 }
232
233 if (harder_count < 10 &&
234 info->total_ref_cache_size > 10 * 1024 * 1024) {
235 harder_count++;
236 goto harder;
237 }
238 }
239}
240
89ce8a63
CM
241static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
242 struct btrfs_root *root, int throttle)
79154b1b
CM
243{
244 struct btrfs_transaction *cur_trans;
ab78c84d 245 struct btrfs_fs_info *info = root->fs_info;
d6e4a428 246
ab78c84d
CM
247 mutex_lock(&info->trans_mutex);
248 cur_trans = info->running_transaction;
ccd467d6 249 WARN_ON(cur_trans != trans->transaction);
d5719762 250 WARN_ON(cur_trans->num_writers < 1);
ccd467d6 251 cur_trans->num_writers--;
89ce8a63 252
79154b1b
CM
253 if (waitqueue_active(&cur_trans->writer_wait))
254 wake_up(&cur_trans->writer_wait);
79154b1b 255 put_transaction(cur_trans);
ab78c84d 256 mutex_unlock(&info->trans_mutex);
d6025579 257 memset(trans, 0, sizeof(*trans));
2c90e5d6 258 kmem_cache_free(btrfs_trans_handle_cachep, trans);
ab78c84d
CM
259
260 if (throttle)
261 btrfs_throttle(root);
262
79154b1b
CM
263 return 0;
264}
265
89ce8a63
CM
266int btrfs_end_transaction(struct btrfs_trans_handle *trans,
267 struct btrfs_root *root)
268{
269 return __btrfs_end_transaction(trans, root, 0);
270}
271
272int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
273 struct btrfs_root *root)
274{
275 return __btrfs_end_transaction(trans, root, 1);
276}
277
79154b1b
CM
278
279int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
280 struct btrfs_root *root)
281{
7c4452b9 282 int ret;
7c4452b9
CM
283 int err;
284 int werr = 0;
d1310b2e 285 struct extent_io_tree *dirty_pages;
7c4452b9 286 struct page *page;
7c4452b9 287 struct inode *btree_inode = root->fs_info->btree_inode;
5f39d397
CM
288 u64 start;
289 u64 end;
290 unsigned long index;
7c4452b9
CM
291
292 if (!trans || !trans->transaction) {
293 return filemap_write_and_wait(btree_inode->i_mapping);
294 }
295 dirty_pages = &trans->transaction->dirty_pages;
296 while(1) {
5f39d397
CM
297 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
298 EXTENT_DIRTY);
299 if (ret)
7c4452b9 300 break;
5f39d397
CM
301 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
302 while(start <= end) {
303 index = start >> PAGE_CACHE_SHIFT;
35ebb934 304 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
5f39d397 305 page = find_lock_page(btree_inode->i_mapping, index);
7c4452b9
CM
306 if (!page)
307 continue;
6702ed49
CM
308 if (PageWriteback(page)) {
309 if (PageDirty(page))
310 wait_on_page_writeback(page);
311 else {
312 unlock_page(page);
313 page_cache_release(page);
314 continue;
315 }
316 }
7c4452b9
CM
317 err = write_one_page(page, 0);
318 if (err)
319 werr = err;
320 page_cache_release(page);
321 }
322 }
323 err = filemap_fdatawait(btree_inode->i_mapping);
324 if (err)
325 werr = err;
326 return werr;
79154b1b
CM
327}
328
0b86a832
CM
329static int update_cowonly_root(struct btrfs_trans_handle *trans,
330 struct btrfs_root *root)
79154b1b
CM
331{
332 int ret;
0b86a832
CM
333 u64 old_root_bytenr;
334 struct btrfs_root *tree_root = root->fs_info->tree_root;
79154b1b 335
0b86a832 336 btrfs_write_dirty_block_groups(trans, root);
79154b1b 337 while(1) {
0b86a832
CM
338 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
339 if (old_root_bytenr == root->node->start)
79154b1b 340 break;
0b86a832
CM
341 btrfs_set_root_bytenr(&root->root_item,
342 root->node->start);
343 btrfs_set_root_level(&root->root_item,
344 btrfs_header_level(root->node));
79154b1b 345 ret = btrfs_update_root(trans, tree_root,
0b86a832
CM
346 &root->root_key,
347 &root->root_item);
79154b1b 348 BUG_ON(ret);
0b86a832
CM
349 btrfs_write_dirty_block_groups(trans, root);
350 }
351 return 0;
352}
353
354int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
355 struct btrfs_root *root)
356{
357 struct btrfs_fs_info *fs_info = root->fs_info;
358 struct list_head *next;
359
360 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
361 next = fs_info->dirty_cowonly_roots.next;
362 list_del_init(next);
363 root = list_entry(next, struct btrfs_root, dirty_list);
364 update_cowonly_root(trans, root);
017e5369
CM
365 if (root->fs_info->closing)
366 btrfs_remove_leaf_refs(root);
79154b1b
CM
367 }
368 return 0;
369}
370
5ce14bbc
CM
371int btrfs_add_dead_root(struct btrfs_root *root,
372 struct btrfs_root *latest,
373 struct list_head *dead_list)
5eda7b5e
CM
374{
375 struct dirty_root *dirty;
376
377 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
378 if (!dirty)
379 return -ENOMEM;
5eda7b5e 380 dirty->root = root;
5ce14bbc 381 dirty->latest_root = latest;
5eda7b5e
CM
382 list_add(&dirty->list, dead_list);
383 return 0;
384}
385
80b6794d
CM
386static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
387 struct radix_tree_root *radix,
388 struct list_head *list)
0f7d52f4
CM
389{
390 struct dirty_root *dirty;
391 struct btrfs_root *gang[8];
392 struct btrfs_root *root;
393 int i;
394 int ret;
54aa1f4d 395 int err = 0;
5eda7b5e 396 u32 refs;
54aa1f4d 397
0f7d52f4
CM
398 while(1) {
399 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
400 ARRAY_SIZE(gang),
401 BTRFS_ROOT_TRANS_TAG);
402 if (ret == 0)
403 break;
404 for (i = 0; i < ret; i++) {
405 root = gang[i];
2619ba1f
CM
406 radix_tree_tag_clear(radix,
407 (unsigned long)root->root_key.objectid,
408 BTRFS_ROOT_TRANS_TAG);
31153d81
YZ
409
410 BUG_ON(!root->ref_tree);
017e5369 411 dirty = root->dirty_root;
31153d81 412
0f7d52f4 413 if (root->commit_root == root->node) {
db94535d
CM
414 WARN_ON(root->node->start !=
415 btrfs_root_bytenr(&root->root_item));
31153d81 416
5f39d397 417 free_extent_buffer(root->commit_root);
0f7d52f4 418 root->commit_root = NULL;
31153d81
YZ
419
420 kfree(dirty->root);
421 kfree(dirty);
58176a96
JB
422
423 /* make sure to update the root on disk
424 * so we get any updates to the block used
425 * counts
426 */
427 err = btrfs_update_root(trans,
428 root->fs_info->tree_root,
429 &root->root_key,
430 &root->root_item);
0f7d52f4
CM
431 continue;
432 }
9f3a7427
CM
433
434 memset(&root->root_item.drop_progress, 0,
435 sizeof(struct btrfs_disk_key));
436 root->root_item.drop_level = 0;
0f7d52f4 437 root->commit_root = NULL;
0f7d52f4 438 root->root_key.offset = root->fs_info->generation;
db94535d
CM
439 btrfs_set_root_bytenr(&root->root_item,
440 root->node->start);
441 btrfs_set_root_level(&root->root_item,
442 btrfs_header_level(root->node));
0f7d52f4
CM
443 err = btrfs_insert_root(trans, root->fs_info->tree_root,
444 &root->root_key,
445 &root->root_item);
54aa1f4d
CM
446 if (err)
447 break;
9f3a7427
CM
448
449 refs = btrfs_root_refs(&dirty->root->root_item);
450 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
5eda7b5e 451 err = btrfs_update_root(trans, root->fs_info->tree_root,
9f3a7427
CM
452 &dirty->root->root_key,
453 &dirty->root->root_item);
5eda7b5e
CM
454
455 BUG_ON(err);
9f3a7427 456 if (refs == 1) {
5eda7b5e 457 list_add(&dirty->list, list);
9f3a7427
CM
458 } else {
459 WARN_ON(1);
31153d81 460 free_extent_buffer(dirty->root->node);
9f3a7427 461 kfree(dirty->root);
5eda7b5e 462 kfree(dirty);
9f3a7427 463 }
0f7d52f4
CM
464 }
465 }
54aa1f4d 466 return err;
0f7d52f4
CM
467}
468
e9d0b13b
CM
469int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
470{
471 struct btrfs_fs_info *info = root->fs_info;
472 int ret;
473 struct btrfs_trans_handle *trans;
d3c2fdcf 474 unsigned long nr;
e9d0b13b 475
a2135011 476 smp_mb();
e9d0b13b
CM
477 if (root->defrag_running)
478 return 0;
e9d0b13b 479 trans = btrfs_start_transaction(root, 1);
6b80053d 480 while (1) {
e9d0b13b
CM
481 root->defrag_running = 1;
482 ret = btrfs_defrag_leaves(trans, root, cacheonly);
d3c2fdcf 483 nr = trans->blocks_used;
e9d0b13b 484 btrfs_end_transaction(trans, root);
d3c2fdcf 485 btrfs_btree_balance_dirty(info->tree_root, nr);
e9d0b13b
CM
486 cond_resched();
487
e9d0b13b 488 trans = btrfs_start_transaction(root, 1);
3f157a2f 489 if (root->fs_info->closing || ret != -EAGAIN)
e9d0b13b
CM
490 break;
491 }
492 root->defrag_running = 0;
a2135011 493 smp_mb();
e9d0b13b
CM
494 btrfs_end_transaction(trans, root);
495 return 0;
496}
497
80b6794d
CM
498static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
499 struct list_head *list)
0f7d52f4
CM
500{
501 struct dirty_root *dirty;
502 struct btrfs_trans_handle *trans;
d3c2fdcf 503 unsigned long nr;
db94535d
CM
504 u64 num_bytes;
505 u64 bytes_used;
54aa1f4d 506 int ret = 0;
9f3a7427
CM
507 int err;
508
0f7d52f4 509 while(!list_empty(list)) {
58176a96
JB
510 struct btrfs_root *root;
511
017e5369 512 dirty = list_entry(list->prev, struct dirty_root, list);
0f7d52f4 513 list_del_init(&dirty->list);
5eda7b5e 514
db94535d 515 num_bytes = btrfs_root_used(&dirty->root->root_item);
58176a96 516 root = dirty->latest_root;
a2135011 517 atomic_inc(&root->fs_info->throttles);
58176a96 518
a2135011 519 mutex_lock(&root->fs_info->drop_mutex);
9f3a7427
CM
520 while(1) {
521 trans = btrfs_start_transaction(tree_root, 1);
522 ret = btrfs_drop_snapshot(trans, dirty->root);
523 if (ret != -EAGAIN) {
524 break;
525 }
58176a96 526
9f3a7427
CM
527 err = btrfs_update_root(trans,
528 tree_root,
529 &dirty->root->root_key,
530 &dirty->root->root_item);
531 if (err)
532 ret = err;
d3c2fdcf 533 nr = trans->blocks_used;
017e5369 534 ret = btrfs_end_transaction(trans, tree_root);
9f3a7427 535 BUG_ON(ret);
a2135011
CM
536
537 mutex_unlock(&root->fs_info->drop_mutex);
d3c2fdcf 538 btrfs_btree_balance_dirty(tree_root, nr);
4dc11904 539 cond_resched();
a2135011 540 mutex_lock(&root->fs_info->drop_mutex);
9f3a7427 541 }
0f7d52f4 542 BUG_ON(ret);
a2135011 543 atomic_dec(&root->fs_info->throttles);
017e5369 544 wake_up(&root->fs_info->transaction_throttle);
58176a96 545
a2135011 546 mutex_lock(&root->fs_info->alloc_mutex);
db94535d
CM
547 num_bytes -= btrfs_root_used(&dirty->root->root_item);
548 bytes_used = btrfs_root_used(&root->root_item);
549 if (num_bytes) {
58176a96 550 record_root_in_trans(root);
5f39d397 551 btrfs_set_root_used(&root->root_item,
db94535d 552 bytes_used - num_bytes);
58176a96 553 }
a2135011
CM
554 mutex_unlock(&root->fs_info->alloc_mutex);
555
9f3a7427 556 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
58176a96
JB
557 if (ret) {
558 BUG();
54aa1f4d 559 break;
58176a96 560 }
a2135011
CM
561 mutex_unlock(&root->fs_info->drop_mutex);
562
d3c2fdcf 563 nr = trans->blocks_used;
0f7d52f4
CM
564 ret = btrfs_end_transaction(trans, tree_root);
565 BUG_ON(ret);
5eda7b5e 566
f510cfec 567 free_extent_buffer(dirty->root->node);
9f3a7427 568 kfree(dirty->root);
0f7d52f4 569 kfree(dirty);
d3c2fdcf
CM
570
571 btrfs_btree_balance_dirty(tree_root, nr);
4dc11904 572 cond_resched();
0f7d52f4 573 }
54aa1f4d 574 return ret;
0f7d52f4
CM
575}
576
80b6794d 577static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
3063d29f
CM
578 struct btrfs_fs_info *fs_info,
579 struct btrfs_pending_snapshot *pending)
580{
581 struct btrfs_key key;
80b6794d 582 struct btrfs_root_item *new_root_item;
3063d29f
CM
583 struct btrfs_root *tree_root = fs_info->tree_root;
584 struct btrfs_root *root = pending->root;
585 struct extent_buffer *tmp;
925baedd 586 struct extent_buffer *old;
3063d29f 587 int ret;
3b96362c 588 int namelen;
3063d29f
CM
589 u64 objectid;
590
80b6794d
CM
591 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
592 if (!new_root_item) {
593 ret = -ENOMEM;
594 goto fail;
595 }
3063d29f
CM
596 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
597 if (ret)
598 goto fail;
599
80b6794d 600 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
3063d29f
CM
601
602 key.objectid = objectid;
603 key.offset = 1;
604 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
605
925baedd
CM
606 old = btrfs_lock_root_node(root);
607 btrfs_cow_block(trans, root, old, NULL, 0, &old);
3063d29f 608
925baedd
CM
609 btrfs_copy_root(trans, root, old, &tmp, objectid);
610 btrfs_tree_unlock(old);
611 free_extent_buffer(old);
3063d29f 612
80b6794d
CM
613 btrfs_set_root_bytenr(new_root_item, tmp->start);
614 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
3063d29f 615 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
80b6794d 616 new_root_item);
925baedd 617 btrfs_tree_unlock(tmp);
3063d29f
CM
618 free_extent_buffer(tmp);
619 if (ret)
620 goto fail;
621
622 /*
623 * insert the directory item
624 */
625 key.offset = (u64)-1;
3b96362c 626 namelen = strlen(pending->name);
3063d29f 627 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
3b96362c 628 pending->name, namelen,
3063d29f 629 root->fs_info->sb->s_root->d_inode->i_ino,
aec7477b 630 &key, BTRFS_FT_DIR, 0);
3063d29f
CM
631
632 if (ret)
633 goto fail;
634
635 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
636 pending->name, strlen(pending->name), objectid,
aec7477b 637 root->fs_info->sb->s_root->d_inode->i_ino, 0);
3b96362c
SW
638
639 /* Invalidate existing dcache entry for new snapshot. */
640 btrfs_invalidate_dcache_root(root, pending->name, namelen);
641
3063d29f 642fail:
80b6794d 643 kfree(new_root_item);
3063d29f
CM
644 return ret;
645}
646
80b6794d
CM
647static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
648 struct btrfs_fs_info *fs_info)
3063d29f
CM
649{
650 struct btrfs_pending_snapshot *pending;
651 struct list_head *head = &trans->transaction->pending_snapshots;
652 int ret;
653
654 while(!list_empty(head)) {
655 pending = list_entry(head->next,
656 struct btrfs_pending_snapshot, list);
657 ret = create_pending_snapshot(trans, fs_info, pending);
658 BUG_ON(ret);
659 list_del(&pending->list);
660 kfree(pending->name);
661 kfree(pending);
662 }
dc17ff8f
CM
663 return 0;
664}
665
79154b1b
CM
666int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
667 struct btrfs_root *root)
668{
15ee9bc7
JB
669 unsigned long joined = 0;
670 unsigned long timeout = 1;
79154b1b 671 struct btrfs_transaction *cur_trans;
8fd17795 672 struct btrfs_transaction *prev_trans = NULL;
0b86a832 673 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
0f7d52f4 674 struct list_head dirty_fs_roots;
d1310b2e 675 struct extent_io_tree *pinned_copy;
79154b1b 676 DEFINE_WAIT(wait);
15ee9bc7 677 int ret;
79154b1b 678
0f7d52f4 679 INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428 680
79154b1b
CM
681 mutex_lock(&root->fs_info->trans_mutex);
682 if (trans->transaction->in_commit) {
683 cur_trans = trans->transaction;
684 trans->transaction->use_count++;
ccd467d6 685 mutex_unlock(&root->fs_info->trans_mutex);
79154b1b 686 btrfs_end_transaction(trans, root);
ccd467d6 687
79154b1b
CM
688 ret = wait_for_commit(root, cur_trans);
689 BUG_ON(ret);
15ee9bc7
JB
690
691 mutex_lock(&root->fs_info->trans_mutex);
79154b1b 692 put_transaction(cur_trans);
15ee9bc7
JB
693 mutex_unlock(&root->fs_info->trans_mutex);
694
79154b1b
CM
695 return 0;
696 }
4313b399
CM
697
698 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
699 if (!pinned_copy)
700 return -ENOMEM;
701
d1310b2e 702 extent_io_tree_init(pinned_copy,
4313b399
CM
703 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
704
2c90e5d6 705 trans->transaction->in_commit = 1;
f9295749 706 trans->transaction->blocked = 1;
ccd467d6
CM
707 cur_trans = trans->transaction;
708 if (cur_trans->list.prev != &root->fs_info->trans_list) {
709 prev_trans = list_entry(cur_trans->list.prev,
710 struct btrfs_transaction, list);
711 if (!prev_trans->commit_done) {
712 prev_trans->use_count++;
ccd467d6
CM
713 mutex_unlock(&root->fs_info->trans_mutex);
714
715 wait_for_commit(root, prev_trans);
ccd467d6 716
ccd467d6 717 mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7 718 put_transaction(prev_trans);
ccd467d6
CM
719 }
720 }
15ee9bc7
JB
721
722 do {
723 joined = cur_trans->num_joined;
2c90e5d6 724 WARN_ON(cur_trans != trans->transaction);
15ee9bc7 725 prepare_to_wait(&cur_trans->writer_wait, &wait,
79154b1b 726 TASK_UNINTERRUPTIBLE);
15ee9bc7
JB
727
728 if (cur_trans->num_writers > 1)
729 timeout = MAX_SCHEDULE_TIMEOUT;
730 else
731 timeout = 1;
732
79154b1b 733 mutex_unlock(&root->fs_info->trans_mutex);
15ee9bc7
JB
734
735 schedule_timeout(timeout);
736
79154b1b 737 mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7
JB
738 finish_wait(&cur_trans->writer_wait, &wait);
739 } while (cur_trans->num_writers > 1 ||
740 (cur_trans->num_joined != joined));
741
3063d29f
CM
742 ret = create_pending_snapshots(trans, root->fs_info);
743 BUG_ON(ret);
744
2c90e5d6 745 WARN_ON(cur_trans != trans->transaction);
dc17ff8f 746
54aa1f4d
CM
747 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
748 &dirty_fs_roots);
749 BUG_ON(ret);
750
79154b1b
CM
751 ret = btrfs_commit_tree_roots(trans, root);
752 BUG_ON(ret);
54aa1f4d 753
78fae27e 754 cur_trans = root->fs_info->running_transaction;
cee36a03 755 spin_lock(&root->fs_info->new_trans_lock);
78fae27e 756 root->fs_info->running_transaction = NULL;
cee36a03 757 spin_unlock(&root->fs_info->new_trans_lock);
4b52dff6
CM
758 btrfs_set_super_generation(&root->fs_info->super_copy,
759 cur_trans->transid);
760 btrfs_set_super_root(&root->fs_info->super_copy,
db94535d
CM
761 root->fs_info->tree_root->node->start);
762 btrfs_set_super_root_level(&root->fs_info->super_copy,
763 btrfs_header_level(root->fs_info->tree_root->node));
5f39d397 764
0b86a832
CM
765 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
766 chunk_root->node->start);
767 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
768 btrfs_header_level(chunk_root->node));
a061fc8d
CM
769 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
770 sizeof(root->fs_info->super_copy));
ccd467d6 771
4313b399 772 btrfs_copy_pinned(root, pinned_copy);
ccd467d6 773
f9295749 774 trans->transaction->blocked = 0;
e6dcd2dc 775 wake_up(&root->fs_info->transaction_throttle);
f9295749 776 wake_up(&root->fs_info->transaction_wait);
e6dcd2dc 777
78fae27e 778 mutex_unlock(&root->fs_info->trans_mutex);
79154b1b
CM
779 ret = btrfs_write_and_wait_transaction(trans, root);
780 BUG_ON(ret);
79154b1b 781 write_ctree_super(trans, root);
4313b399 782
4313b399 783 btrfs_finish_extent_commit(trans, root, pinned_copy);
78fae27e 784 mutex_lock(&root->fs_info->trans_mutex);
4313b399
CM
785
786 kfree(pinned_copy);
787
2c90e5d6 788 cur_trans->commit_done = 1;
15ee9bc7 789 root->fs_info->last_trans_committed = cur_trans->transid;
2c90e5d6 790 wake_up(&cur_trans->commit_wait);
78fae27e 791 put_transaction(cur_trans);
79154b1b 792 put_transaction(cur_trans);
58176a96 793
facda1e7
CM
794 if (root->fs_info->closing)
795 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
796 else
797 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
58176a96 798
78fae27e 799 mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6 800 kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b 801
facda1e7 802 if (root->fs_info->closing) {
facda1e7 803 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
facda1e7 804 }
79154b1b
CM
805 return ret;
806}
807
e9d0b13b
CM
808int btrfs_clean_old_snapshots(struct btrfs_root *root)
809{
810 struct list_head dirty_roots;
811 INIT_LIST_HEAD(&dirty_roots);
a74a4b97 812again:
e9d0b13b
CM
813 mutex_lock(&root->fs_info->trans_mutex);
814 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
815 mutex_unlock(&root->fs_info->trans_mutex);
816
817 if (!list_empty(&dirty_roots)) {
818 drop_dirty_roots(root, &dirty_roots);
a74a4b97 819 goto again;
e9d0b13b
CM
820 }
821 return 0;
822}
08607c1b 823