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"
28 static int total_trans = 0;
29 extern struct kmem_cache *btrfs_trans_handle_cachep;
30 extern struct kmem_cache *btrfs_transaction_cachep;
32 #define BTRFS_ROOT_TRANS_TAG 0
34 static noinline void put_transaction(struct btrfs_transaction *transaction)
36 WARN_ON(transaction->use_count == 0);
37 transaction->use_count--;
38 if (transaction->use_count == 0) {
39 WARN_ON(total_trans == 0);
41 list_del_init(&transaction->list);
42 memset(transaction, 0, sizeof(*transaction));
43 kmem_cache_free(btrfs_transaction_cachep, transaction);
47 static noinline int join_transaction(struct btrfs_root *root)
49 struct btrfs_transaction *cur_trans;
50 cur_trans = root->fs_info->running_transaction;
52 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
56 root->fs_info->generation++;
57 root->fs_info->last_alloc = 0;
58 root->fs_info->last_data_alloc = 0;
59 cur_trans->num_writers = 1;
60 cur_trans->num_joined = 0;
61 cur_trans->transid = root->fs_info->generation;
62 init_waitqueue_head(&cur_trans->writer_wait);
63 init_waitqueue_head(&cur_trans->commit_wait);
64 cur_trans->in_commit = 0;
65 cur_trans->blocked = 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 extent_io_tree_init(&cur_trans->dirty_pages,
72 root->fs_info->btree_inode->i_mapping,
74 spin_lock(&root->fs_info->new_trans_lock);
75 root->fs_info->running_transaction = cur_trans;
76 spin_unlock(&root->fs_info->new_trans_lock);
78 cur_trans->num_writers++;
79 cur_trans->num_joined++;
85 static noinline int record_root_in_trans(struct btrfs_root *root)
87 u64 running_trans_id = root->fs_info->running_transaction->transid;
88 if (root->ref_cows && root->last_trans < running_trans_id) {
89 WARN_ON(root == root->fs_info->extent_root);
90 if (root->root_item.refs != 0) {
91 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
92 (unsigned long)root->root_key.objectid,
93 BTRFS_ROOT_TRANS_TAG);
94 root->commit_root = btrfs_root_node(root);
98 root->last_trans = running_trans_id;
103 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
104 int num_blocks, int join)
106 struct btrfs_trans_handle *h =
107 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
108 struct btrfs_transaction *cur_trans;
111 mutex_lock(&root->fs_info->trans_mutex);
112 cur_trans = root->fs_info->running_transaction;
113 if (cur_trans && cur_trans->blocked && !join) {
115 cur_trans->use_count++;
117 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
118 TASK_UNINTERRUPTIBLE);
119 if (cur_trans->blocked) {
120 mutex_unlock(&root->fs_info->trans_mutex);
122 mutex_lock(&root->fs_info->trans_mutex);
123 finish_wait(&root->fs_info->transaction_wait,
126 finish_wait(&root->fs_info->transaction_wait,
131 put_transaction(cur_trans);
133 ret = join_transaction(root);
136 record_root_in_trans(root);
137 h->transid = root->fs_info->running_transaction->transid;
138 h->transaction = root->fs_info->running_transaction;
139 h->blocks_reserved = num_blocks;
141 h->block_group = NULL;
142 h->alloc_exclude_nr = 0;
143 h->alloc_exclude_start = 0;
144 root->fs_info->running_transaction->use_count++;
145 mutex_unlock(&root->fs_info->trans_mutex);
149 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
152 return start_transaction(root, num_blocks, 0);
154 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
157 return start_transaction(root, num_blocks, 1);
160 static noinline int wait_for_commit(struct btrfs_root *root,
161 struct btrfs_transaction *commit)
164 mutex_lock(&root->fs_info->trans_mutex);
165 while(!commit->commit_done) {
166 prepare_to_wait(&commit->commit_wait, &wait,
167 TASK_UNINTERRUPTIBLE);
168 if (commit->commit_done)
170 mutex_unlock(&root->fs_info->trans_mutex);
172 mutex_lock(&root->fs_info->trans_mutex);
174 mutex_unlock(&root->fs_info->trans_mutex);
175 finish_wait(&commit->commit_wait, &wait);
179 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
180 struct btrfs_root *root, int throttle)
182 struct btrfs_transaction *cur_trans;
184 mutex_lock(&root->fs_info->trans_mutex);
185 cur_trans = root->fs_info->running_transaction;
186 WARN_ON(cur_trans != trans->transaction);
187 WARN_ON(cur_trans->num_writers < 1);
188 cur_trans->num_writers--;
190 if (waitqueue_active(&cur_trans->writer_wait))
191 wake_up(&cur_trans->writer_wait);
193 if (0 && cur_trans->in_commit && throttle) {
195 mutex_unlock(&root->fs_info->trans_mutex);
196 prepare_to_wait(&root->fs_info->transaction_throttle, &wait,
197 TASK_UNINTERRUPTIBLE);
199 finish_wait(&root->fs_info->transaction_throttle, &wait);
200 mutex_lock(&root->fs_info->trans_mutex);
203 put_transaction(cur_trans);
204 mutex_unlock(&root->fs_info->trans_mutex);
205 memset(trans, 0, sizeof(*trans));
206 kmem_cache_free(btrfs_trans_handle_cachep, trans);
210 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
211 struct btrfs_root *root)
213 return __btrfs_end_transaction(trans, root, 0);
216 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
217 struct btrfs_root *root)
219 return __btrfs_end_transaction(trans, root, 1);
223 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
224 struct btrfs_root *root)
229 struct extent_io_tree *dirty_pages;
231 struct inode *btree_inode = root->fs_info->btree_inode;
236 if (!trans || !trans->transaction) {
237 return filemap_write_and_wait(btree_inode->i_mapping);
239 dirty_pages = &trans->transaction->dirty_pages;
241 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
245 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
246 while(start <= end) {
247 index = start >> PAGE_CACHE_SHIFT;
248 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
249 page = find_lock_page(btree_inode->i_mapping, index);
252 if (PageWriteback(page)) {
254 wait_on_page_writeback(page);
257 page_cache_release(page);
261 err = write_one_page(page, 0);
264 page_cache_release(page);
267 err = filemap_fdatawait(btree_inode->i_mapping);
273 static int update_cowonly_root(struct btrfs_trans_handle *trans,
274 struct btrfs_root *root)
278 struct btrfs_root *tree_root = root->fs_info->tree_root;
280 btrfs_write_dirty_block_groups(trans, root);
282 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
283 if (old_root_bytenr == root->node->start)
285 btrfs_set_root_bytenr(&root->root_item,
287 btrfs_set_root_level(&root->root_item,
288 btrfs_header_level(root->node));
289 ret = btrfs_update_root(trans, tree_root,
293 btrfs_write_dirty_block_groups(trans, root);
298 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
299 struct btrfs_root *root)
301 struct btrfs_fs_info *fs_info = root->fs_info;
302 struct list_head *next;
304 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
305 next = fs_info->dirty_cowonly_roots.next;
307 root = list_entry(next, struct btrfs_root, dirty_list);
308 update_cowonly_root(trans, root);
314 struct list_head list;
315 struct btrfs_root *root;
316 struct btrfs_root *latest_root;
319 int btrfs_add_dead_root(struct btrfs_root *root,
320 struct btrfs_root *latest,
321 struct list_head *dead_list)
323 struct dirty_root *dirty;
325 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
329 dirty->latest_root = latest;
330 list_add(&dirty->list, dead_list);
334 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
335 struct radix_tree_root *radix,
336 struct list_head *list)
338 struct dirty_root *dirty;
339 struct btrfs_root *gang[8];
340 struct btrfs_root *root;
347 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
349 BTRFS_ROOT_TRANS_TAG);
352 for (i = 0; i < ret; i++) {
354 radix_tree_tag_clear(radix,
355 (unsigned long)root->root_key.objectid,
356 BTRFS_ROOT_TRANS_TAG);
357 if (root->commit_root == root->node) {
358 WARN_ON(root->node->start !=
359 btrfs_root_bytenr(&root->root_item));
360 free_extent_buffer(root->commit_root);
361 root->commit_root = NULL;
363 /* make sure to update the root on disk
364 * so we get any updates to the block used
367 err = btrfs_update_root(trans,
368 root->fs_info->tree_root,
373 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
375 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
376 BUG_ON(!dirty->root);
378 memset(&root->root_item.drop_progress, 0,
379 sizeof(struct btrfs_disk_key));
380 root->root_item.drop_level = 0;
382 memcpy(dirty->root, root, sizeof(*root));
383 dirty->root->node = root->commit_root;
384 dirty->latest_root = root;
385 spin_lock_init(&dirty->root->node_lock);
386 mutex_init(&dirty->root->objectid_mutex);
388 root->commit_root = NULL;
390 root->root_key.offset = root->fs_info->generation;
391 btrfs_set_root_bytenr(&root->root_item,
393 btrfs_set_root_level(&root->root_item,
394 btrfs_header_level(root->node));
395 err = btrfs_insert_root(trans, root->fs_info->tree_root,
401 refs = btrfs_root_refs(&dirty->root->root_item);
402 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
403 err = btrfs_update_root(trans, root->fs_info->tree_root,
404 &dirty->root->root_key,
405 &dirty->root->root_item);
409 list_add(&dirty->list, list);
420 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
422 struct btrfs_fs_info *info = root->fs_info;
424 struct btrfs_trans_handle *trans;
428 if (root->defrag_running)
430 trans = btrfs_start_transaction(root, 1);
432 root->defrag_running = 1;
433 ret = btrfs_defrag_leaves(trans, root, cacheonly);
434 nr = trans->blocks_used;
435 btrfs_end_transaction(trans, root);
436 btrfs_btree_balance_dirty(info->tree_root, nr);
439 trans = btrfs_start_transaction(root, 1);
440 if (root->fs_info->closing || ret != -EAGAIN)
443 root->defrag_running = 0;
445 btrfs_end_transaction(trans, root);
449 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
450 struct list_head *list)
452 struct dirty_root *dirty;
453 struct btrfs_trans_handle *trans;
460 while(!list_empty(list)) {
461 struct btrfs_root *root;
463 dirty = list_entry(list->next, struct dirty_root, list);
464 list_del_init(&dirty->list);
466 num_bytes = btrfs_root_used(&dirty->root->root_item);
467 root = dirty->latest_root;
468 atomic_inc(&root->fs_info->throttles);
470 mutex_lock(&root->fs_info->drop_mutex);
472 trans = btrfs_start_transaction(tree_root, 1);
473 ret = btrfs_drop_snapshot(trans, dirty->root);
474 if (ret != -EAGAIN) {
478 err = btrfs_update_root(trans,
480 &dirty->root->root_key,
481 &dirty->root->root_item);
484 nr = trans->blocks_used;
485 ret = btrfs_end_transaction_throttle(trans, tree_root);
488 mutex_unlock(&root->fs_info->drop_mutex);
489 btrfs_btree_balance_dirty(tree_root, nr);
491 mutex_lock(&root->fs_info->drop_mutex);
494 atomic_dec(&root->fs_info->throttles);
496 mutex_lock(&root->fs_info->alloc_mutex);
497 num_bytes -= btrfs_root_used(&dirty->root->root_item);
498 bytes_used = btrfs_root_used(&root->root_item);
500 record_root_in_trans(root);
501 btrfs_set_root_used(&root->root_item,
502 bytes_used - num_bytes);
504 mutex_unlock(&root->fs_info->alloc_mutex);
506 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
511 mutex_unlock(&root->fs_info->drop_mutex);
513 nr = trans->blocks_used;
514 ret = btrfs_end_transaction(trans, tree_root);
517 free_extent_buffer(dirty->root->node);
521 btrfs_btree_balance_dirty(tree_root, nr);
527 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
528 struct btrfs_fs_info *fs_info,
529 struct btrfs_pending_snapshot *pending)
531 struct btrfs_key key;
532 struct btrfs_root_item *new_root_item;
533 struct btrfs_root *tree_root = fs_info->tree_root;
534 struct btrfs_root *root = pending->root;
535 struct extent_buffer *tmp;
536 struct extent_buffer *old;
541 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
542 if (!new_root_item) {
546 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
550 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
552 key.objectid = objectid;
554 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
556 old = btrfs_lock_root_node(root);
557 btrfs_cow_block(trans, root, old, NULL, 0, &old);
559 btrfs_copy_root(trans, root, old, &tmp, objectid);
560 btrfs_tree_unlock(old);
561 free_extent_buffer(old);
563 btrfs_set_root_bytenr(new_root_item, tmp->start);
564 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
565 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
567 btrfs_tree_unlock(tmp);
568 free_extent_buffer(tmp);
573 * insert the directory item
575 key.offset = (u64)-1;
576 namelen = strlen(pending->name);
577 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
578 pending->name, namelen,
579 root->fs_info->sb->s_root->d_inode->i_ino,
580 &key, BTRFS_FT_DIR, 0);
585 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
586 pending->name, strlen(pending->name), objectid,
587 root->fs_info->sb->s_root->d_inode->i_ino, 0);
589 /* Invalidate existing dcache entry for new snapshot. */
590 btrfs_invalidate_dcache_root(root, pending->name, namelen);
593 kfree(new_root_item);
597 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
598 struct btrfs_fs_info *fs_info)
600 struct btrfs_pending_snapshot *pending;
601 struct list_head *head = &trans->transaction->pending_snapshots;
604 while(!list_empty(head)) {
605 pending = list_entry(head->next,
606 struct btrfs_pending_snapshot, list);
607 ret = create_pending_snapshot(trans, fs_info, pending);
609 list_del(&pending->list);
610 kfree(pending->name);
616 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
617 struct btrfs_root *root)
619 unsigned long joined = 0;
620 unsigned long timeout = 1;
621 struct btrfs_transaction *cur_trans;
622 struct btrfs_transaction *prev_trans = NULL;
623 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
624 struct list_head dirty_fs_roots;
625 struct extent_io_tree *pinned_copy;
629 INIT_LIST_HEAD(&dirty_fs_roots);
631 mutex_lock(&root->fs_info->trans_mutex);
632 if (trans->transaction->in_commit) {
633 cur_trans = trans->transaction;
634 trans->transaction->use_count++;
635 mutex_unlock(&root->fs_info->trans_mutex);
636 btrfs_end_transaction(trans, root);
638 ret = wait_for_commit(root, cur_trans);
641 mutex_lock(&root->fs_info->trans_mutex);
642 put_transaction(cur_trans);
643 mutex_unlock(&root->fs_info->trans_mutex);
648 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
652 extent_io_tree_init(pinned_copy,
653 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
655 trans->transaction->in_commit = 1;
656 trans->transaction->blocked = 1;
657 cur_trans = trans->transaction;
658 if (cur_trans->list.prev != &root->fs_info->trans_list) {
659 prev_trans = list_entry(cur_trans->list.prev,
660 struct btrfs_transaction, list);
661 if (!prev_trans->commit_done) {
662 prev_trans->use_count++;
663 mutex_unlock(&root->fs_info->trans_mutex);
665 wait_for_commit(root, prev_trans);
667 mutex_lock(&root->fs_info->trans_mutex);
668 put_transaction(prev_trans);
673 joined = cur_trans->num_joined;
674 WARN_ON(cur_trans != trans->transaction);
675 prepare_to_wait(&cur_trans->writer_wait, &wait,
676 TASK_UNINTERRUPTIBLE);
678 if (cur_trans->num_writers > 1)
679 timeout = MAX_SCHEDULE_TIMEOUT;
683 mutex_unlock(&root->fs_info->trans_mutex);
685 schedule_timeout(timeout);
687 mutex_lock(&root->fs_info->trans_mutex);
688 finish_wait(&cur_trans->writer_wait, &wait);
689 } while (cur_trans->num_writers > 1 ||
690 (cur_trans->num_joined != joined));
692 ret = create_pending_snapshots(trans, root->fs_info);
695 WARN_ON(cur_trans != trans->transaction);
697 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
701 ret = btrfs_commit_tree_roots(trans, root);
704 cur_trans = root->fs_info->running_transaction;
705 spin_lock(&root->fs_info->new_trans_lock);
706 root->fs_info->running_transaction = NULL;
707 spin_unlock(&root->fs_info->new_trans_lock);
708 btrfs_set_super_generation(&root->fs_info->super_copy,
710 btrfs_set_super_root(&root->fs_info->super_copy,
711 root->fs_info->tree_root->node->start);
712 btrfs_set_super_root_level(&root->fs_info->super_copy,
713 btrfs_header_level(root->fs_info->tree_root->node));
715 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
716 chunk_root->node->start);
717 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
718 btrfs_header_level(chunk_root->node));
719 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
720 sizeof(root->fs_info->super_copy));
722 btrfs_copy_pinned(root, pinned_copy);
724 trans->transaction->blocked = 0;
725 wake_up(&root->fs_info->transaction_throttle);
726 wake_up(&root->fs_info->transaction_wait);
728 mutex_unlock(&root->fs_info->trans_mutex);
729 ret = btrfs_write_and_wait_transaction(trans, root);
731 write_ctree_super(trans, root);
733 btrfs_finish_extent_commit(trans, root, pinned_copy);
734 mutex_lock(&root->fs_info->trans_mutex);
738 cur_trans->commit_done = 1;
739 root->fs_info->last_trans_committed = cur_trans->transid;
740 wake_up(&cur_trans->commit_wait);
741 put_transaction(cur_trans);
742 put_transaction(cur_trans);
744 if (root->fs_info->closing)
745 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
747 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
749 mutex_unlock(&root->fs_info->trans_mutex);
750 kmem_cache_free(btrfs_trans_handle_cachep, trans);
752 if (root->fs_info->closing) {
753 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
758 int btrfs_clean_old_snapshots(struct btrfs_root *root)
760 struct list_head dirty_roots;
761 INIT_LIST_HEAD(&dirty_roots);
763 mutex_lock(&root->fs_info->trans_mutex);
764 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
765 mutex_unlock(&root->fs_info->trans_mutex);
767 if (!list_empty(&dirty_roots)) {
768 drop_dirty_roots(root, &dirty_roots);