2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
20 #include <linux/sched.h>
21 #include <linux/writeback.h>
22 #include <linux/pagemap.h>
25 #include "transaction.h"
27 static int total_trans = 0;
28 extern struct kmem_cache *btrfs_trans_handle_cachep;
29 extern struct kmem_cache *btrfs_transaction_cachep;
31 static struct workqueue_struct *trans_wq;
33 #define BTRFS_ROOT_TRANS_TAG 0
34 #define BTRFS_ROOT_DEFRAG_TAG 1
36 static void put_transaction(struct btrfs_transaction *transaction)
38 WARN_ON(transaction->use_count == 0);
39 transaction->use_count--;
40 if (transaction->use_count == 0) {
41 WARN_ON(total_trans == 0);
43 list_del_init(&transaction->list);
44 memset(transaction, 0, sizeof(*transaction));
45 kmem_cache_free(btrfs_transaction_cachep, transaction);
49 static int join_transaction(struct btrfs_root *root)
51 struct btrfs_transaction *cur_trans;
52 cur_trans = root->fs_info->running_transaction;
54 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
58 root->fs_info->generation++;
59 root->fs_info->running_transaction = cur_trans;
60 cur_trans->num_writers = 1;
61 cur_trans->num_joined = 0;
62 cur_trans->transid = root->fs_info->generation;
63 init_waitqueue_head(&cur_trans->writer_wait);
64 init_waitqueue_head(&cur_trans->commit_wait);
65 cur_trans->in_commit = 0;
66 cur_trans->use_count = 1;
67 cur_trans->commit_done = 0;
68 cur_trans->start_time = get_seconds();
69 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
70 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
71 btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
72 extent_map_tree_init(&cur_trans->dirty_pages,
73 root->fs_info->btree_inode->i_mapping,
76 cur_trans->num_writers++;
77 cur_trans->num_joined++;
83 static int record_root_in_trans(struct btrfs_root *root)
85 u64 running_trans_id = root->fs_info->running_transaction->transid;
86 if (root->ref_cows && root->last_trans < running_trans_id) {
87 WARN_ON(root == root->fs_info->extent_root);
88 if (root->root_item.refs != 0) {
89 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
90 (unsigned long)root->root_key.objectid,
91 BTRFS_ROOT_TRANS_TAG);
92 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
93 (unsigned long)root->root_key.objectid,
94 BTRFS_ROOT_DEFRAG_TAG);
95 root->commit_root = root->node;
96 extent_buffer_get(root->node);
100 root->last_trans = running_trans_id;
105 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
108 struct btrfs_trans_handle *h =
109 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
112 mutex_lock(&root->fs_info->trans_mutex);
113 ret = join_transaction(root);
116 record_root_in_trans(root);
117 h->transid = root->fs_info->running_transaction->transid;
118 h->transaction = root->fs_info->running_transaction;
119 h->blocks_reserved = num_blocks;
121 h->block_group = NULL;
122 h->alloc_exclude_nr = 0;
123 h->alloc_exclude_start = 0;
124 root->fs_info->running_transaction->use_count++;
125 mutex_unlock(&root->fs_info->trans_mutex);
129 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
130 struct btrfs_root *root)
132 struct btrfs_transaction *cur_trans;
134 mutex_lock(&root->fs_info->trans_mutex);
135 cur_trans = root->fs_info->running_transaction;
136 WARN_ON(cur_trans != trans->transaction);
137 WARN_ON(cur_trans->num_writers < 1);
138 cur_trans->num_writers--;
139 if (waitqueue_active(&cur_trans->writer_wait))
140 wake_up(&cur_trans->writer_wait);
141 put_transaction(cur_trans);
142 mutex_unlock(&root->fs_info->trans_mutex);
143 memset(trans, 0, sizeof(*trans));
144 kmem_cache_free(btrfs_trans_handle_cachep, trans);
149 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
150 struct btrfs_root *root)
155 struct extent_map_tree *dirty_pages;
157 struct inode *btree_inode = root->fs_info->btree_inode;
162 if (!trans || !trans->transaction) {
163 return filemap_write_and_wait(btree_inode->i_mapping);
165 dirty_pages = &trans->transaction->dirty_pages;
167 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
171 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
172 while(start <= end) {
173 index = start >> PAGE_CACHE_SHIFT;
174 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
175 page = find_lock_page(btree_inode->i_mapping, index);
178 if (PageWriteback(page)) {
180 wait_on_page_writeback(page);
183 page_cache_release(page);
187 err = write_one_page(page, 0);
190 page_cache_release(page);
193 err = filemap_fdatawait(btree_inode->i_mapping);
199 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
200 struct btrfs_root *root)
203 u64 old_extent_block;
204 struct btrfs_fs_info *fs_info = root->fs_info;
205 struct btrfs_root *tree_root = fs_info->tree_root;
206 struct btrfs_root *extent_root = fs_info->extent_root;
208 btrfs_write_dirty_block_groups(trans, extent_root);
210 old_extent_block = btrfs_root_bytenr(&extent_root->root_item);
211 if (old_extent_block == extent_root->node->start)
213 btrfs_set_root_bytenr(&extent_root->root_item,
214 extent_root->node->start);
215 btrfs_set_root_level(&extent_root->root_item,
216 btrfs_header_level(extent_root->node));
217 ret = btrfs_update_root(trans, tree_root,
218 &extent_root->root_key,
219 &extent_root->root_item);
221 btrfs_write_dirty_block_groups(trans, extent_root);
226 static int wait_for_commit(struct btrfs_root *root,
227 struct btrfs_transaction *commit)
230 mutex_lock(&root->fs_info->trans_mutex);
231 while(!commit->commit_done) {
232 prepare_to_wait(&commit->commit_wait, &wait,
233 TASK_UNINTERRUPTIBLE);
234 if (commit->commit_done)
236 mutex_unlock(&root->fs_info->trans_mutex);
238 mutex_lock(&root->fs_info->trans_mutex);
240 mutex_unlock(&root->fs_info->trans_mutex);
241 finish_wait(&commit->commit_wait, &wait);
246 struct list_head list;
247 struct btrfs_root *root;
248 struct btrfs_root *latest_root;
251 int btrfs_add_dead_root(struct btrfs_root *root,
252 struct btrfs_root *latest,
253 struct list_head *dead_list)
255 struct dirty_root *dirty;
257 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
261 dirty->latest_root = latest;
262 list_add(&dirty->list, dead_list);
266 static int add_dirty_roots(struct btrfs_trans_handle *trans,
267 struct radix_tree_root *radix,
268 struct list_head *list)
270 struct dirty_root *dirty;
271 struct btrfs_root *gang[8];
272 struct btrfs_root *root;
279 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
281 BTRFS_ROOT_TRANS_TAG);
284 for (i = 0; i < ret; i++) {
286 radix_tree_tag_clear(radix,
287 (unsigned long)root->root_key.objectid,
288 BTRFS_ROOT_TRANS_TAG);
289 if (root->commit_root == root->node) {
290 WARN_ON(root->node->start !=
291 btrfs_root_bytenr(&root->root_item));
292 free_extent_buffer(root->commit_root);
293 root->commit_root = NULL;
295 /* make sure to update the root on disk
296 * so we get any updates to the block used
299 err = btrfs_update_root(trans,
300 root->fs_info->tree_root,
305 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
307 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
308 BUG_ON(!dirty->root);
310 memset(&root->root_item.drop_progress, 0,
311 sizeof(struct btrfs_disk_key));
312 root->root_item.drop_level = 0;
314 memcpy(dirty->root, root, sizeof(*root));
315 dirty->root->node = root->commit_root;
316 dirty->latest_root = root;
317 root->commit_root = NULL;
319 root->root_key.offset = root->fs_info->generation;
320 btrfs_set_root_bytenr(&root->root_item,
322 btrfs_set_root_level(&root->root_item,
323 btrfs_header_level(root->node));
324 err = btrfs_insert_root(trans, root->fs_info->tree_root,
330 refs = btrfs_root_refs(&dirty->root->root_item);
331 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
332 err = btrfs_update_root(trans, root->fs_info->tree_root,
333 &dirty->root->root_key,
334 &dirty->root->root_item);
338 list_add(&dirty->list, list);
349 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
351 struct btrfs_fs_info *info = root->fs_info;
353 struct btrfs_trans_handle *trans;
356 if (root->defrag_running)
358 trans = btrfs_start_transaction(root, 1);
360 root->defrag_running = 1;
361 ret = btrfs_defrag_leaves(trans, root, cacheonly);
362 nr = trans->blocks_used;
363 btrfs_end_transaction(trans, root);
364 mutex_unlock(&info->fs_mutex);
365 btrfs_btree_balance_dirty(info->tree_root, nr);
368 mutex_lock(&info->fs_mutex);
369 trans = btrfs_start_transaction(root, 1);
373 root->defrag_running = 0;
374 radix_tree_tag_clear(&info->fs_roots_radix,
375 (unsigned long)root->root_key.objectid,
376 BTRFS_ROOT_DEFRAG_TAG);
377 btrfs_end_transaction(trans, root);
381 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
383 struct btrfs_root *gang[1];
384 struct btrfs_root *root;
391 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
394 BTRFS_ROOT_DEFRAG_TAG);
397 for (i = 0; i < ret; i++) {
399 last = root->root_key.objectid + 1;
400 btrfs_defrag_root(root, 1);
403 btrfs_defrag_root(info->extent_root, 1);
407 static int drop_dirty_roots(struct btrfs_root *tree_root,
408 struct list_head *list)
410 struct dirty_root *dirty;
411 struct btrfs_trans_handle *trans;
418 while(!list_empty(list)) {
419 struct btrfs_root *root;
421 mutex_lock(&tree_root->fs_info->fs_mutex);
422 dirty = list_entry(list->next, struct dirty_root, list);
423 list_del_init(&dirty->list);
425 num_bytes = btrfs_root_used(&dirty->root->root_item);
426 root = dirty->latest_root;
429 trans = btrfs_start_transaction(tree_root, 1);
430 ret = btrfs_drop_snapshot(trans, dirty->root);
431 if (ret != -EAGAIN) {
435 err = btrfs_update_root(trans,
437 &dirty->root->root_key,
438 &dirty->root->root_item);
441 nr = trans->blocks_used;
442 ret = btrfs_end_transaction(trans, tree_root);
444 mutex_unlock(&tree_root->fs_info->fs_mutex);
445 btrfs_btree_balance_dirty(tree_root, nr);
447 mutex_lock(&tree_root->fs_info->fs_mutex);
451 num_bytes -= btrfs_root_used(&dirty->root->root_item);
452 bytes_used = btrfs_root_used(&root->root_item);
454 record_root_in_trans(root);
455 btrfs_set_root_used(&root->root_item,
456 bytes_used - num_bytes);
458 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
463 nr = trans->blocks_used;
464 ret = btrfs_end_transaction(trans, tree_root);
467 free_extent_buffer(dirty->root->node);
470 mutex_unlock(&tree_root->fs_info->fs_mutex);
472 btrfs_btree_balance_dirty(tree_root, nr);
478 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
479 struct btrfs_root *root)
481 struct btrfs_transaction *cur_trans = trans->transaction;
483 u64 root_objectid = 0;
488 ret = btrfs_find_first_ordered_inode(
489 &cur_trans->ordered_inode_tree,
490 &root_objectid, &objectid);
494 mutex_unlock(&root->fs_info->trans_mutex);
495 mutex_unlock(&root->fs_info->fs_mutex);
496 inode = btrfs_ilookup(root->fs_info->sb, objectid,
499 if (S_ISREG(inode->i_mode))
500 filemap_fdatawrite(inode->i_mapping);
503 mutex_lock(&root->fs_info->fs_mutex);
504 mutex_lock(&root->fs_info->trans_mutex);
509 ret = btrfs_find_del_first_ordered_inode(
510 &cur_trans->ordered_inode_tree,
511 &root_objectid, &objectid);
514 mutex_unlock(&root->fs_info->trans_mutex);
515 mutex_unlock(&root->fs_info->fs_mutex);
516 inode = btrfs_ilookup(root->fs_info->sb, objectid,
519 if (S_ISREG(inode->i_mode))
520 filemap_write_and_wait(inode->i_mapping);
523 mutex_lock(&root->fs_info->fs_mutex);
524 mutex_lock(&root->fs_info->trans_mutex);
529 static int create_pending_snapshot(struct btrfs_trans_handle *trans,
530 struct btrfs_fs_info *fs_info,
531 struct btrfs_pending_snapshot *pending)
533 struct btrfs_key key;
534 struct btrfs_root_item new_root_item;
535 struct btrfs_root *tree_root = fs_info->tree_root;
536 struct btrfs_root *root = pending->root;
537 struct extent_buffer *tmp;
541 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
545 memcpy(&new_root_item, &root->root_item, sizeof(new_root_item));
547 key.objectid = objectid;
549 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
551 extent_buffer_get(root->node);
552 btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
553 free_extent_buffer(tmp);
555 btrfs_copy_root(trans, root, root->node, &tmp, objectid);
557 btrfs_set_root_bytenr(&new_root_item, tmp->start);
558 btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
559 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
561 free_extent_buffer(tmp);
566 * insert the directory item
568 key.offset = (u64)-1;
569 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
570 pending->name, strlen(pending->name),
571 root->fs_info->sb->s_root->d_inode->i_ino,
577 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
578 pending->name, strlen(pending->name), objectid,
579 root->fs_info->sb->s_root->d_inode->i_ino);
584 static int create_pending_snapshots(struct btrfs_trans_handle *trans,
585 struct btrfs_fs_info *fs_info)
587 struct btrfs_pending_snapshot *pending;
588 struct list_head *head = &trans->transaction->pending_snapshots;
591 while(!list_empty(head)) {
592 pending = list_entry(head->next,
593 struct btrfs_pending_snapshot, list);
594 ret = create_pending_snapshot(trans, fs_info, pending);
596 list_del(&pending->list);
597 kfree(pending->name);
603 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
604 struct btrfs_root *root)
606 unsigned long joined = 0;
607 unsigned long timeout = 1;
608 struct btrfs_transaction *cur_trans;
609 struct btrfs_transaction *prev_trans = NULL;
610 struct list_head dirty_fs_roots;
611 struct extent_map_tree *pinned_copy;
615 INIT_LIST_HEAD(&dirty_fs_roots);
617 mutex_lock(&root->fs_info->trans_mutex);
618 if (trans->transaction->in_commit) {
619 cur_trans = trans->transaction;
620 trans->transaction->use_count++;
621 mutex_unlock(&root->fs_info->trans_mutex);
622 btrfs_end_transaction(trans, root);
624 mutex_unlock(&root->fs_info->fs_mutex);
625 ret = wait_for_commit(root, cur_trans);
628 mutex_lock(&root->fs_info->trans_mutex);
629 put_transaction(cur_trans);
630 mutex_unlock(&root->fs_info->trans_mutex);
632 mutex_lock(&root->fs_info->fs_mutex);
636 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
640 extent_map_tree_init(pinned_copy,
641 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
643 trans->transaction->in_commit = 1;
644 cur_trans = trans->transaction;
645 if (cur_trans->list.prev != &root->fs_info->trans_list) {
646 prev_trans = list_entry(cur_trans->list.prev,
647 struct btrfs_transaction, list);
648 if (!prev_trans->commit_done) {
649 prev_trans->use_count++;
650 mutex_unlock(&root->fs_info->fs_mutex);
651 mutex_unlock(&root->fs_info->trans_mutex);
653 wait_for_commit(root, prev_trans);
655 mutex_lock(&root->fs_info->fs_mutex);
656 mutex_lock(&root->fs_info->trans_mutex);
657 put_transaction(prev_trans);
662 joined = cur_trans->num_joined;
663 WARN_ON(cur_trans != trans->transaction);
664 prepare_to_wait(&cur_trans->writer_wait, &wait,
665 TASK_UNINTERRUPTIBLE);
667 if (cur_trans->num_writers > 1)
668 timeout = MAX_SCHEDULE_TIMEOUT;
672 mutex_unlock(&root->fs_info->fs_mutex);
673 mutex_unlock(&root->fs_info->trans_mutex);
675 schedule_timeout(timeout);
677 mutex_lock(&root->fs_info->fs_mutex);
678 mutex_lock(&root->fs_info->trans_mutex);
679 finish_wait(&cur_trans->writer_wait, &wait);
680 ret = btrfs_write_ordered_inodes(trans, root);
682 } while (cur_trans->num_writers > 1 ||
683 (cur_trans->num_joined != joined));
685 ret = create_pending_snapshots(trans, root->fs_info);
688 WARN_ON(cur_trans != trans->transaction);
690 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
694 ret = btrfs_commit_tree_roots(trans, root);
697 cur_trans = root->fs_info->running_transaction;
698 root->fs_info->running_transaction = NULL;
699 btrfs_set_super_generation(&root->fs_info->super_copy,
701 btrfs_set_super_root(&root->fs_info->super_copy,
702 root->fs_info->tree_root->node->start);
703 btrfs_set_super_root_level(&root->fs_info->super_copy,
704 btrfs_header_level(root->fs_info->tree_root->node));
706 write_extent_buffer(root->fs_info->sb_buffer,
707 &root->fs_info->super_copy, 0,
708 sizeof(root->fs_info->super_copy));
710 btrfs_copy_pinned(root, pinned_copy);
712 mutex_unlock(&root->fs_info->trans_mutex);
713 mutex_unlock(&root->fs_info->fs_mutex);
714 ret = btrfs_write_and_wait_transaction(trans, root);
716 write_ctree_super(trans, root);
718 mutex_lock(&root->fs_info->fs_mutex);
719 btrfs_finish_extent_commit(trans, root, pinned_copy);
720 mutex_lock(&root->fs_info->trans_mutex);
724 cur_trans->commit_done = 1;
725 root->fs_info->last_trans_committed = cur_trans->transid;
726 wake_up(&cur_trans->commit_wait);
727 put_transaction(cur_trans);
728 put_transaction(cur_trans);
730 if (root->fs_info->closing)
731 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
733 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
735 mutex_unlock(&root->fs_info->trans_mutex);
736 kmem_cache_free(btrfs_trans_handle_cachep, trans);
738 if (root->fs_info->closing) {
739 mutex_unlock(&root->fs_info->fs_mutex);
740 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
741 mutex_lock(&root->fs_info->fs_mutex);
746 int btrfs_clean_old_snapshots(struct btrfs_root *root)
748 struct list_head dirty_roots;
749 INIT_LIST_HEAD(&dirty_roots);
751 mutex_lock(&root->fs_info->trans_mutex);
752 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
753 mutex_unlock(&root->fs_info->trans_mutex);
755 if (!list_empty(&dirty_roots)) {
756 drop_dirty_roots(root, &dirty_roots);
760 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
761 void btrfs_transaction_cleaner(void *p)
763 void btrfs_transaction_cleaner(struct work_struct *work)
766 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
767 struct btrfs_fs_info *fs_info = p;
769 struct btrfs_fs_info *fs_info = container_of(work,
770 struct btrfs_fs_info,
774 struct btrfs_root *root = fs_info->tree_root;
775 struct btrfs_transaction *cur;
776 struct btrfs_trans_handle *trans;
778 unsigned long delay = HZ * 30;
781 mutex_lock(&root->fs_info->fs_mutex);
782 mutex_lock(&root->fs_info->trans_mutex);
783 cur = root->fs_info->running_transaction;
785 mutex_unlock(&root->fs_info->trans_mutex);
789 if (now < cur->start_time || now - cur->start_time < 30) {
790 mutex_unlock(&root->fs_info->trans_mutex);
794 mutex_unlock(&root->fs_info->trans_mutex);
795 btrfs_defrag_dirty_roots(root->fs_info);
796 trans = btrfs_start_transaction(root, 1);
797 ret = btrfs_commit_transaction(trans, root);
799 mutex_unlock(&root->fs_info->fs_mutex);
800 btrfs_clean_old_snapshots(root);
801 btrfs_transaction_queue_work(root, delay);
804 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
806 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
809 void btrfs_transaction_flush_work(struct btrfs_root *root)
811 cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
812 flush_workqueue(trans_wq);
815 void __init btrfs_init_transaction_sys(void)
817 trans_wq = create_workqueue("btrfs");
820 void btrfs_exit_transaction_sys(void)
822 destroy_workqueue(trans_wq);