Btrfs: Add missing hunk from Yan Zheng's cache reclaim patch
[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
80b6794d 35static noinline void put_transaction(struct btrfs_transaction *transaction)
79154b1b 36{
2c90e5d6 37 WARN_ON(transaction->use_count == 0);
79154b1b 38 transaction->use_count--;
78fae27e
CM
39 if (transaction->use_count == 0) {
40 WARN_ON(total_trans == 0);
41 total_trans--;
8fd17795 42 list_del_init(&transaction->list);
2c90e5d6
CM
43 memset(transaction, 0, sizeof(*transaction));
44 kmem_cache_free(btrfs_transaction_cachep, transaction);
78fae27e 45 }
79154b1b
CM
46}
47
80b6794d 48static noinline int join_transaction(struct btrfs_root *root)
79154b1b
CM
49{
50 struct btrfs_transaction *cur_trans;
51 cur_trans = root->fs_info->running_transaction;
52 if (!cur_trans) {
2c90e5d6
CM
53 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
54 GFP_NOFS);
78fae27e 55 total_trans++;
79154b1b 56 BUG_ON(!cur_trans);
0f7d52f4 57 root->fs_info->generation++;
e18e4809 58 root->fs_info->last_alloc = 0;
4529ba49 59 root->fs_info->last_data_alloc = 0;
15ee9bc7
JB
60 cur_trans->num_writers = 1;
61 cur_trans->num_joined = 0;
0f7d52f4 62 cur_trans->transid = root->fs_info->generation;
79154b1b
CM
63 init_waitqueue_head(&cur_trans->writer_wait);
64 init_waitqueue_head(&cur_trans->commit_wait);
65 cur_trans->in_commit = 0;
f9295749 66 cur_trans->blocked = 0;
d5719762 67 cur_trans->use_count = 1;
79154b1b 68 cur_trans->commit_done = 0;
08607c1b 69 cur_trans->start_time = get_seconds();
3063d29f 70 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
8fd17795 71 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
d1310b2e 72 extent_io_tree_init(&cur_trans->dirty_pages,
5f39d397
CM
73 root->fs_info->btree_inode->i_mapping,
74 GFP_NOFS);
48ec2cf8
CM
75 spin_lock(&root->fs_info->new_trans_lock);
76 root->fs_info->running_transaction = cur_trans;
77 spin_unlock(&root->fs_info->new_trans_lock);
15ee9bc7
JB
78 } else {
79 cur_trans->num_writers++;
80 cur_trans->num_joined++;
79154b1b 81 }
15ee9bc7 82
79154b1b
CM
83 return 0;
84}
85
80b6794d 86static noinline int record_root_in_trans(struct btrfs_root *root)
6702ed49 87{
f321e491 88 struct btrfs_dirty_root *dirty;
6702ed49
CM
89 u64 running_trans_id = root->fs_info->running_transaction->transid;
90 if (root->ref_cows && root->last_trans < running_trans_id) {
91 WARN_ON(root == root->fs_info->extent_root);
92 if (root->root_item.refs != 0) {
93 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
94 (unsigned long)root->root_key.objectid,
95 BTRFS_ROOT_TRANS_TAG);
31153d81
YZ
96
97 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
98 BUG_ON(!dirty);
99 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
100 BUG_ON(!dirty->root);
31153d81
YZ
101 dirty->latest_root = root;
102 INIT_LIST_HEAD(&dirty->list);
31153d81 103
925baedd 104 root->commit_root = btrfs_root_node(root);
31153d81
YZ
105
106 memcpy(dirty->root, root, sizeof(*root));
107 spin_lock_init(&dirty->root->node_lock);
bcc63abb 108 spin_lock_init(&dirty->root->list_lock);
31153d81 109 mutex_init(&dirty->root->objectid_mutex);
bcc63abb 110 INIT_LIST_HEAD(&dirty->root->dead_list);
31153d81
YZ
111 dirty->root->node = root->commit_root;
112 dirty->root->commit_root = NULL;
bcc63abb
Y
113
114 spin_lock(&root->list_lock);
115 list_add(&dirty->root->dead_list, &root->dead_list);
116 spin_unlock(&root->list_lock);
117
118 root->dirty_root = dirty;
6702ed49
CM
119 } else {
120 WARN_ON(1);
121 }
122 root->last_trans = running_trans_id;
123 }
124 return 0;
125}
126
f9295749
CM
127struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
128 int num_blocks, int join)
79154b1b 129{
2c90e5d6
CM
130 struct btrfs_trans_handle *h =
131 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
f9295749 132 struct btrfs_transaction *cur_trans;
79154b1b
CM
133 int ret;
134
135 mutex_lock(&root->fs_info->trans_mutex);
f9295749
CM
136 cur_trans = root->fs_info->running_transaction;
137 if (cur_trans && cur_trans->blocked && !join) {
138 DEFINE_WAIT(wait);
139 cur_trans->use_count++;
140 while(1) {
141 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
142 TASK_UNINTERRUPTIBLE);
143 if (cur_trans->blocked) {
144 mutex_unlock(&root->fs_info->trans_mutex);
145 schedule();
146 mutex_lock(&root->fs_info->trans_mutex);
147 finish_wait(&root->fs_info->transaction_wait,
148 &wait);
149 } else {
150 finish_wait(&root->fs_info->transaction_wait,
151 &wait);
152 break;
153 }
154 }
155 put_transaction(cur_trans);
156 }
79154b1b
CM
157 ret = join_transaction(root);
158 BUG_ON(ret);
0f7d52f4 159
6702ed49
CM
160 record_root_in_trans(root);
161 h->transid = root->fs_info->running_transaction->transid;
79154b1b
CM
162 h->transaction = root->fs_info->running_transaction;
163 h->blocks_reserved = num_blocks;
164 h->blocks_used = 0;
31f3c99b 165 h->block_group = NULL;
26b8003f
CM
166 h->alloc_exclude_nr = 0;
167 h->alloc_exclude_start = 0;
79154b1b
CM
168 root->fs_info->running_transaction->use_count++;
169 mutex_unlock(&root->fs_info->trans_mutex);
170 return h;
171}
172
f9295749
CM
173struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
174 int num_blocks)
175{
176 return start_transaction(root, num_blocks, 0);
177}
178struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
179 int num_blocks)
180{
181 return start_transaction(root, num_blocks, 1);
182}
183
89ce8a63
CM
184static noinline int wait_for_commit(struct btrfs_root *root,
185 struct btrfs_transaction *commit)
186{
187 DEFINE_WAIT(wait);
188 mutex_lock(&root->fs_info->trans_mutex);
189 while(!commit->commit_done) {
190 prepare_to_wait(&commit->commit_wait, &wait,
191 TASK_UNINTERRUPTIBLE);
192 if (commit->commit_done)
193 break;
194 mutex_unlock(&root->fs_info->trans_mutex);
195 schedule();
196 mutex_lock(&root->fs_info->trans_mutex);
197 }
198 mutex_unlock(&root->fs_info->trans_mutex);
199 finish_wait(&commit->commit_wait, &wait);
200 return 0;
201}
202
ab78c84d
CM
203void btrfs_throttle(struct btrfs_root *root)
204{
205 struct btrfs_fs_info *info = root->fs_info;
206
207harder:
208 if (atomic_read(&info->throttles)) {
209 DEFINE_WAIT(wait);
210 int thr;
211 int harder_count = 0;
212 thr = atomic_read(&info->throttle_gen);
213
214 do {
215 prepare_to_wait(&info->transaction_throttle,
216 &wait, TASK_UNINTERRUPTIBLE);
217 if (!atomic_read(&info->throttles)) {
218 finish_wait(&info->transaction_throttle, &wait);
219 break;
220 }
221 schedule();
222 finish_wait(&info->transaction_throttle, &wait);
223 } while (thr == atomic_read(&info->throttle_gen));
224
225 if (harder_count < 5 &&
226 info->total_ref_cache_size > 5 * 1024 * 1024) {
227 harder_count++;
228 goto harder;
229 }
230
231 if (harder_count < 10 &&
232 info->total_ref_cache_size > 10 * 1024 * 1024) {
233 harder_count++;
234 goto harder;
235 }
236 }
237}
238
89ce8a63
CM
239static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
240 struct btrfs_root *root, int throttle)
79154b1b
CM
241{
242 struct btrfs_transaction *cur_trans;
ab78c84d 243 struct btrfs_fs_info *info = root->fs_info;
d6e4a428 244
ab78c84d
CM
245 mutex_lock(&info->trans_mutex);
246 cur_trans = info->running_transaction;
ccd467d6 247 WARN_ON(cur_trans != trans->transaction);
d5719762 248 WARN_ON(cur_trans->num_writers < 1);
ccd467d6 249 cur_trans->num_writers--;
89ce8a63 250
79154b1b
CM
251 if (waitqueue_active(&cur_trans->writer_wait))
252 wake_up(&cur_trans->writer_wait);
79154b1b 253 put_transaction(cur_trans);
ab78c84d 254 mutex_unlock(&info->trans_mutex);
d6025579 255 memset(trans, 0, sizeof(*trans));
2c90e5d6 256 kmem_cache_free(btrfs_trans_handle_cachep, trans);
ab78c84d
CM
257
258 if (throttle)
259 btrfs_throttle(root);
260
79154b1b
CM
261 return 0;
262}
263
89ce8a63
CM
264int btrfs_end_transaction(struct btrfs_trans_handle *trans,
265 struct btrfs_root *root)
266{
267 return __btrfs_end_transaction(trans, root, 0);
268}
269
270int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
271 struct btrfs_root *root)
272{
273 return __btrfs_end_transaction(trans, root, 1);
274}
275
79154b1b
CM
276
277int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root)
279{
7c4452b9 280 int ret;
7c4452b9
CM
281 int err;
282 int werr = 0;
d1310b2e 283 struct extent_io_tree *dirty_pages;
7c4452b9 284 struct page *page;
7c4452b9 285 struct inode *btree_inode = root->fs_info->btree_inode;
5f39d397
CM
286 u64 start;
287 u64 end;
288 unsigned long index;
7c4452b9
CM
289
290 if (!trans || !trans->transaction) {
291 return filemap_write_and_wait(btree_inode->i_mapping);
292 }
293 dirty_pages = &trans->transaction->dirty_pages;
294 while(1) {
5f39d397
CM
295 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
296 EXTENT_DIRTY);
297 if (ret)
7c4452b9 298 break;
5f39d397
CM
299 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
300 while(start <= end) {
301 index = start >> PAGE_CACHE_SHIFT;
35ebb934 302 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
5f39d397 303 page = find_lock_page(btree_inode->i_mapping, index);
7c4452b9
CM
304 if (!page)
305 continue;
6702ed49
CM
306 if (PageWriteback(page)) {
307 if (PageDirty(page))
308 wait_on_page_writeback(page);
309 else {
310 unlock_page(page);
311 page_cache_release(page);
312 continue;
313 }
314 }
7c4452b9
CM
315 err = write_one_page(page, 0);
316 if (err)
317 werr = err;
318 page_cache_release(page);
319 }
320 }
321 err = filemap_fdatawait(btree_inode->i_mapping);
322 if (err)
323 werr = err;
324 return werr;
79154b1b
CM
325}
326
0b86a832
CM
327static int update_cowonly_root(struct btrfs_trans_handle *trans,
328 struct btrfs_root *root)
79154b1b
CM
329{
330 int ret;
0b86a832
CM
331 u64 old_root_bytenr;
332 struct btrfs_root *tree_root = root->fs_info->tree_root;
79154b1b 333
0b86a832 334 btrfs_write_dirty_block_groups(trans, root);
79154b1b 335 while(1) {
0b86a832
CM
336 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
337 if (old_root_bytenr == root->node->start)
79154b1b 338 break;
0b86a832
CM
339 btrfs_set_root_bytenr(&root->root_item,
340 root->node->start);
341 btrfs_set_root_level(&root->root_item,
342 btrfs_header_level(root->node));
79154b1b 343 ret = btrfs_update_root(trans, tree_root,
0b86a832
CM
344 &root->root_key,
345 &root->root_item);
79154b1b 346 BUG_ON(ret);
0b86a832
CM
347 btrfs_write_dirty_block_groups(trans, root);
348 }
349 return 0;
350}
351
352int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
353 struct btrfs_root *root)
354{
355 struct btrfs_fs_info *fs_info = root->fs_info;
356 struct list_head *next;
357
358 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
359 next = fs_info->dirty_cowonly_roots.next;
360 list_del_init(next);
361 root = list_entry(next, struct btrfs_root, dirty_list);
362 update_cowonly_root(trans, root);
79154b1b
CM
363 }
364 return 0;
365}
366
5ce14bbc
CM
367int btrfs_add_dead_root(struct btrfs_root *root,
368 struct btrfs_root *latest,
369 struct list_head *dead_list)
5eda7b5e 370{
f321e491 371 struct btrfs_dirty_root *dirty;
5eda7b5e
CM
372
373 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
374 if (!dirty)
375 return -ENOMEM;
5eda7b5e 376 dirty->root = root;
5ce14bbc 377 dirty->latest_root = latest;
5eda7b5e
CM
378 list_add(&dirty->list, dead_list);
379 return 0;
380}
381
80b6794d
CM
382static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
383 struct radix_tree_root *radix,
384 struct list_head *list)
0f7d52f4 385{
f321e491 386 struct btrfs_dirty_root *dirty;
0f7d52f4
CM
387 struct btrfs_root *gang[8];
388 struct btrfs_root *root;
389 int i;
390 int ret;
54aa1f4d 391 int err = 0;
5eda7b5e 392 u32 refs;
54aa1f4d 393
0f7d52f4
CM
394 while(1) {
395 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
396 ARRAY_SIZE(gang),
397 BTRFS_ROOT_TRANS_TAG);
398 if (ret == 0)
399 break;
400 for (i = 0; i < ret; i++) {
401 root = gang[i];
2619ba1f
CM
402 radix_tree_tag_clear(radix,
403 (unsigned long)root->root_key.objectid,
404 BTRFS_ROOT_TRANS_TAG);
31153d81
YZ
405
406 BUG_ON(!root->ref_tree);
017e5369 407 dirty = root->dirty_root;
31153d81 408
0f7d52f4 409 if (root->commit_root == root->node) {
db94535d
CM
410 WARN_ON(root->node->start !=
411 btrfs_root_bytenr(&root->root_item));
31153d81 412
5f39d397 413 free_extent_buffer(root->commit_root);
0f7d52f4 414 root->commit_root = NULL;
bcc63abb
Y
415
416 spin_lock(&root->list_lock);
417 list_del_init(&dirty->root->dead_list);
418 spin_unlock(&root->list_lock);
419
31153d81
YZ
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 500{
f321e491 501 struct btrfs_dirty_root *dirty;
0f7d52f4 502 struct btrfs_trans_handle *trans;
d3c2fdcf 503 unsigned long nr;
db94535d
CM
504 u64 num_bytes;
505 u64 bytes_used;
bcc63abb 506 u64 max_useless;
54aa1f4d 507 int ret = 0;
9f3a7427
CM
508 int err;
509
0f7d52f4 510 while(!list_empty(list)) {
58176a96
JB
511 struct btrfs_root *root;
512
f321e491 513 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
0f7d52f4 514 list_del_init(&dirty->list);
5eda7b5e 515
db94535d 516 num_bytes = btrfs_root_used(&dirty->root->root_item);
58176a96 517 root = dirty->latest_root;
a2135011 518 atomic_inc(&root->fs_info->throttles);
58176a96 519
a2135011 520 mutex_lock(&root->fs_info->drop_mutex);
9f3a7427
CM
521 while(1) {
522 trans = btrfs_start_transaction(tree_root, 1);
523 ret = btrfs_drop_snapshot(trans, dirty->root);
524 if (ret != -EAGAIN) {
525 break;
526 }
58176a96 527
9f3a7427
CM
528 err = btrfs_update_root(trans,
529 tree_root,
530 &dirty->root->root_key,
531 &dirty->root->root_item);
532 if (err)
533 ret = err;
d3c2fdcf 534 nr = trans->blocks_used;
017e5369 535 ret = btrfs_end_transaction(trans, tree_root);
9f3a7427 536 BUG_ON(ret);
a2135011
CM
537
538 mutex_unlock(&root->fs_info->drop_mutex);
d3c2fdcf 539 btrfs_btree_balance_dirty(tree_root, nr);
4dc11904 540 cond_resched();
a2135011 541 mutex_lock(&root->fs_info->drop_mutex);
9f3a7427 542 }
0f7d52f4 543 BUG_ON(ret);
a2135011 544 atomic_dec(&root->fs_info->throttles);
017e5369 545 wake_up(&root->fs_info->transaction_throttle);
58176a96 546
a2135011 547 mutex_lock(&root->fs_info->alloc_mutex);
db94535d
CM
548 num_bytes -= btrfs_root_used(&dirty->root->root_item);
549 bytes_used = btrfs_root_used(&root->root_item);
550 if (num_bytes) {
58176a96 551 record_root_in_trans(root);
5f39d397 552 btrfs_set_root_used(&root->root_item,
db94535d 553 bytes_used - num_bytes);
58176a96 554 }
a2135011
CM
555 mutex_unlock(&root->fs_info->alloc_mutex);
556
9f3a7427 557 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
58176a96
JB
558 if (ret) {
559 BUG();
54aa1f4d 560 break;
58176a96 561 }
a2135011
CM
562 mutex_unlock(&root->fs_info->drop_mutex);
563
bcc63abb
Y
564 spin_lock(&root->list_lock);
565 list_del_init(&dirty->root->dead_list);
566 if (!list_empty(&root->dead_list)) {
567 struct btrfs_root *oldest;
568 oldest = list_entry(root->dead_list.prev,
569 struct btrfs_root, dead_list);
570 max_useless = oldest->root_key.offset - 1;
571 } else {
572 max_useless = root->root_key.offset - 1;
573 }
574 spin_unlock(&root->list_lock);
575
d3c2fdcf 576 nr = trans->blocks_used;
0f7d52f4
CM
577 ret = btrfs_end_transaction(trans, tree_root);
578 BUG_ON(ret);
5eda7b5e 579
bcc63abb
Y
580 ret = btrfs_remove_leaf_refs(root, max_useless);
581 BUG_ON(ret);
582
f510cfec 583 free_extent_buffer(dirty->root->node);
9f3a7427 584 kfree(dirty->root);
0f7d52f4 585 kfree(dirty);
d3c2fdcf
CM
586
587 btrfs_btree_balance_dirty(tree_root, nr);
4dc11904 588 cond_resched();
0f7d52f4 589 }
54aa1f4d 590 return ret;
0f7d52f4
CM
591}
592
80b6794d 593static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
3063d29f
CM
594 struct btrfs_fs_info *fs_info,
595 struct btrfs_pending_snapshot *pending)
596{
597 struct btrfs_key key;
80b6794d 598 struct btrfs_root_item *new_root_item;
3063d29f
CM
599 struct btrfs_root *tree_root = fs_info->tree_root;
600 struct btrfs_root *root = pending->root;
601 struct extent_buffer *tmp;
925baedd 602 struct extent_buffer *old;
3063d29f 603 int ret;
3b96362c 604 int namelen;
3063d29f
CM
605 u64 objectid;
606
80b6794d
CM
607 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
608 if (!new_root_item) {
609 ret = -ENOMEM;
610 goto fail;
611 }
3063d29f
CM
612 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
613 if (ret)
614 goto fail;
615
80b6794d 616 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
3063d29f
CM
617
618 key.objectid = objectid;
619 key.offset = 1;
620 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
621
925baedd
CM
622 old = btrfs_lock_root_node(root);
623 btrfs_cow_block(trans, root, old, NULL, 0, &old);
3063d29f 624
925baedd
CM
625 btrfs_copy_root(trans, root, old, &tmp, objectid);
626 btrfs_tree_unlock(old);
627 free_extent_buffer(old);
3063d29f 628
80b6794d
CM
629 btrfs_set_root_bytenr(new_root_item, tmp->start);
630 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
3063d29f 631 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
80b6794d 632 new_root_item);
925baedd 633 btrfs_tree_unlock(tmp);
3063d29f
CM
634 free_extent_buffer(tmp);
635 if (ret)
636 goto fail;
637
638 /*
639 * insert the directory item
640 */
641 key.offset = (u64)-1;
3b96362c 642 namelen = strlen(pending->name);
3063d29f 643 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
3b96362c 644 pending->name, namelen,
3063d29f 645 root->fs_info->sb->s_root->d_inode->i_ino,
aec7477b 646 &key, BTRFS_FT_DIR, 0);
3063d29f
CM
647
648 if (ret)
649 goto fail;
650
651 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
652 pending->name, strlen(pending->name), objectid,
aec7477b 653 root->fs_info->sb->s_root->d_inode->i_ino, 0);
3b96362c
SW
654
655 /* Invalidate existing dcache entry for new snapshot. */
656 btrfs_invalidate_dcache_root(root, pending->name, namelen);
657
3063d29f 658fail:
80b6794d 659 kfree(new_root_item);
3063d29f
CM
660 return ret;
661}
662
80b6794d
CM
663static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
664 struct btrfs_fs_info *fs_info)
3063d29f
CM
665{
666 struct btrfs_pending_snapshot *pending;
667 struct list_head *head = &trans->transaction->pending_snapshots;
668 int ret;
669
670 while(!list_empty(head)) {
671 pending = list_entry(head->next,
672 struct btrfs_pending_snapshot, list);
673 ret = create_pending_snapshot(trans, fs_info, pending);
674 BUG_ON(ret);
675 list_del(&pending->list);
676 kfree(pending->name);
677 kfree(pending);
678 }
dc17ff8f
CM
679 return 0;
680}
681
79154b1b
CM
682int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
683 struct btrfs_root *root)
684{
15ee9bc7
JB
685 unsigned long joined = 0;
686 unsigned long timeout = 1;
79154b1b 687 struct btrfs_transaction *cur_trans;
8fd17795 688 struct btrfs_transaction *prev_trans = NULL;
0b86a832 689 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
0f7d52f4 690 struct list_head dirty_fs_roots;
d1310b2e 691 struct extent_io_tree *pinned_copy;
79154b1b 692 DEFINE_WAIT(wait);
15ee9bc7 693 int ret;
79154b1b 694
0f7d52f4 695 INIT_LIST_HEAD(&dirty_fs_roots);
d6e4a428 696
79154b1b
CM
697 mutex_lock(&root->fs_info->trans_mutex);
698 if (trans->transaction->in_commit) {
699 cur_trans = trans->transaction;
700 trans->transaction->use_count++;
ccd467d6 701 mutex_unlock(&root->fs_info->trans_mutex);
79154b1b 702 btrfs_end_transaction(trans, root);
ccd467d6 703
79154b1b
CM
704 ret = wait_for_commit(root, cur_trans);
705 BUG_ON(ret);
15ee9bc7
JB
706
707 mutex_lock(&root->fs_info->trans_mutex);
79154b1b 708 put_transaction(cur_trans);
15ee9bc7
JB
709 mutex_unlock(&root->fs_info->trans_mutex);
710
79154b1b
CM
711 return 0;
712 }
4313b399
CM
713
714 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
715 if (!pinned_copy)
716 return -ENOMEM;
717
d1310b2e 718 extent_io_tree_init(pinned_copy,
4313b399
CM
719 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
720
2c90e5d6 721 trans->transaction->in_commit = 1;
f9295749 722 trans->transaction->blocked = 1;
ccd467d6
CM
723 cur_trans = trans->transaction;
724 if (cur_trans->list.prev != &root->fs_info->trans_list) {
725 prev_trans = list_entry(cur_trans->list.prev,
726 struct btrfs_transaction, list);
727 if (!prev_trans->commit_done) {
728 prev_trans->use_count++;
ccd467d6
CM
729 mutex_unlock(&root->fs_info->trans_mutex);
730
731 wait_for_commit(root, prev_trans);
ccd467d6 732
ccd467d6 733 mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7 734 put_transaction(prev_trans);
ccd467d6
CM
735 }
736 }
15ee9bc7
JB
737
738 do {
739 joined = cur_trans->num_joined;
2c90e5d6 740 WARN_ON(cur_trans != trans->transaction);
15ee9bc7 741 prepare_to_wait(&cur_trans->writer_wait, &wait,
79154b1b 742 TASK_UNINTERRUPTIBLE);
15ee9bc7
JB
743
744 if (cur_trans->num_writers > 1)
745 timeout = MAX_SCHEDULE_TIMEOUT;
746 else
747 timeout = 1;
748
79154b1b 749 mutex_unlock(&root->fs_info->trans_mutex);
15ee9bc7
JB
750
751 schedule_timeout(timeout);
752
79154b1b 753 mutex_lock(&root->fs_info->trans_mutex);
15ee9bc7
JB
754 finish_wait(&cur_trans->writer_wait, &wait);
755 } while (cur_trans->num_writers > 1 ||
756 (cur_trans->num_joined != joined));
757
3063d29f
CM
758 ret = create_pending_snapshots(trans, root->fs_info);
759 BUG_ON(ret);
760
2c90e5d6 761 WARN_ON(cur_trans != trans->transaction);
dc17ff8f 762
54aa1f4d
CM
763 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
764 &dirty_fs_roots);
765 BUG_ON(ret);
766
79154b1b
CM
767 ret = btrfs_commit_tree_roots(trans, root);
768 BUG_ON(ret);
54aa1f4d 769
78fae27e 770 cur_trans = root->fs_info->running_transaction;
cee36a03 771 spin_lock(&root->fs_info->new_trans_lock);
78fae27e 772 root->fs_info->running_transaction = NULL;
cee36a03 773 spin_unlock(&root->fs_info->new_trans_lock);
4b52dff6
CM
774 btrfs_set_super_generation(&root->fs_info->super_copy,
775 cur_trans->transid);
776 btrfs_set_super_root(&root->fs_info->super_copy,
db94535d
CM
777 root->fs_info->tree_root->node->start);
778 btrfs_set_super_root_level(&root->fs_info->super_copy,
779 btrfs_header_level(root->fs_info->tree_root->node));
5f39d397 780
0b86a832
CM
781 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
782 chunk_root->node->start);
783 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
784 btrfs_header_level(chunk_root->node));
a061fc8d
CM
785 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
786 sizeof(root->fs_info->super_copy));
ccd467d6 787
4313b399 788 btrfs_copy_pinned(root, pinned_copy);
ccd467d6 789
f9295749 790 trans->transaction->blocked = 0;
e6dcd2dc 791 wake_up(&root->fs_info->transaction_throttle);
f9295749 792 wake_up(&root->fs_info->transaction_wait);
e6dcd2dc 793
78fae27e 794 mutex_unlock(&root->fs_info->trans_mutex);
79154b1b
CM
795 ret = btrfs_write_and_wait_transaction(trans, root);
796 BUG_ON(ret);
79154b1b 797 write_ctree_super(trans, root);
4313b399 798
4313b399 799 btrfs_finish_extent_commit(trans, root, pinned_copy);
78fae27e 800 mutex_lock(&root->fs_info->trans_mutex);
4313b399
CM
801
802 kfree(pinned_copy);
803
2c90e5d6 804 cur_trans->commit_done = 1;
15ee9bc7 805 root->fs_info->last_trans_committed = cur_trans->transid;
2c90e5d6 806 wake_up(&cur_trans->commit_wait);
78fae27e 807 put_transaction(cur_trans);
79154b1b 808 put_transaction(cur_trans);
58176a96 809
bcc63abb 810 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
facda1e7
CM
811 if (root->fs_info->closing)
812 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
58176a96 813
78fae27e 814 mutex_unlock(&root->fs_info->trans_mutex);
2c90e5d6 815 kmem_cache_free(btrfs_trans_handle_cachep, trans);
79154b1b 816
facda1e7 817 if (root->fs_info->closing) {
facda1e7 818 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
facda1e7 819 }
79154b1b
CM
820 return ret;
821}
822
e9d0b13b
CM
823int btrfs_clean_old_snapshots(struct btrfs_root *root)
824{
825 struct list_head dirty_roots;
826 INIT_LIST_HEAD(&dirty_roots);
a74a4b97 827again:
e9d0b13b
CM
828 mutex_lock(&root->fs_info->trans_mutex);
829 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
830 mutex_unlock(&root->fs_info->trans_mutex);
831
832 if (!list_empty(&dirty_roots)) {
833 drop_dirty_roots(root, &dirty_roots);
a74a4b97 834 goto again;
e9d0b13b
CM
835 }
836 return 0;
837}