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 #include "ref-cache.h"
29 static int total_trans = 0;
30 extern struct kmem_cache *btrfs_trans_handle_cachep;
31 extern struct kmem_cache *btrfs_transaction_cachep;
33 #define BTRFS_ROOT_TRANS_TAG 0
36 struct list_head list;
37 struct btrfs_root *root;
38 struct btrfs_root *latest_root;
41 static noinline void put_transaction(struct btrfs_transaction *transaction)
43 WARN_ON(transaction->use_count == 0);
44 transaction->use_count--;
45 if (transaction->use_count == 0) {
46 WARN_ON(total_trans == 0);
48 list_del_init(&transaction->list);
49 memset(transaction, 0, sizeof(*transaction));
50 kmem_cache_free(btrfs_transaction_cachep, transaction);
54 static noinline int join_transaction(struct btrfs_root *root)
56 struct btrfs_transaction *cur_trans;
57 cur_trans = root->fs_info->running_transaction;
59 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
63 root->fs_info->generation++;
64 root->fs_info->last_alloc = 0;
65 root->fs_info->last_data_alloc = 0;
66 cur_trans->num_writers = 1;
67 cur_trans->num_joined = 0;
68 cur_trans->transid = root->fs_info->generation;
69 init_waitqueue_head(&cur_trans->writer_wait);
70 init_waitqueue_head(&cur_trans->commit_wait);
71 cur_trans->in_commit = 0;
72 cur_trans->blocked = 0;
73 cur_trans->use_count = 1;
74 cur_trans->commit_done = 0;
75 cur_trans->start_time = get_seconds();
76 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
77 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
78 extent_io_tree_init(&cur_trans->dirty_pages,
79 root->fs_info->btree_inode->i_mapping,
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);
85 cur_trans->num_writers++;
86 cur_trans->num_joined++;
92 static noinline int record_root_in_trans(struct btrfs_root *root)
94 struct dirty_root *dirty;
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);
103 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
105 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
106 BUG_ON(!dirty->root);
108 dirty->latest_root = root;
109 INIT_LIST_HEAD(&dirty->list);
111 root->commit_root = btrfs_root_node(root);
112 root->dirty_root = dirty;
114 memcpy(dirty->root, root, sizeof(*root));
115 dirty->root->ref_tree = &root->ref_tree_struct;
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;
124 root->last_trans = running_trans_id;
129 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
130 int num_blocks, int join)
132 struct btrfs_trans_handle *h =
133 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
134 struct btrfs_transaction *cur_trans;
137 mutex_lock(&root->fs_info->trans_mutex);
138 cur_trans = root->fs_info->running_transaction;
139 if (cur_trans && cur_trans->blocked && !join) {
141 cur_trans->use_count++;
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);
148 mutex_lock(&root->fs_info->trans_mutex);
149 finish_wait(&root->fs_info->transaction_wait,
152 finish_wait(&root->fs_info->transaction_wait,
157 put_transaction(cur_trans);
159 ret = join_transaction(root);
162 record_root_in_trans(root);
163 h->transid = root->fs_info->running_transaction->transid;
164 h->transaction = root->fs_info->running_transaction;
165 h->blocks_reserved = num_blocks;
167 h->block_group = NULL;
168 h->alloc_exclude_nr = 0;
169 h->alloc_exclude_start = 0;
170 root->fs_info->running_transaction->use_count++;
171 mutex_unlock(&root->fs_info->trans_mutex);
175 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
178 return start_transaction(root, num_blocks, 0);
180 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
183 return start_transaction(root, num_blocks, 1);
186 static noinline int wait_for_commit(struct btrfs_root *root,
187 struct btrfs_transaction *commit)
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)
196 mutex_unlock(&root->fs_info->trans_mutex);
198 mutex_lock(&root->fs_info->trans_mutex);
200 mutex_unlock(&root->fs_info->trans_mutex);
201 finish_wait(&commit->commit_wait, &wait);
205 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
206 struct btrfs_root *root, int throttle)
208 struct btrfs_transaction *cur_trans;
210 mutex_lock(&root->fs_info->trans_mutex);
211 cur_trans = root->fs_info->running_transaction;
212 WARN_ON(cur_trans != trans->transaction);
213 WARN_ON(cur_trans->num_writers < 1);
214 cur_trans->num_writers--;
216 if (waitqueue_active(&cur_trans->writer_wait))
217 wake_up(&cur_trans->writer_wait);
219 if (throttle && atomic_read(&root->fs_info->throttles)) {
221 mutex_unlock(&root->fs_info->trans_mutex);
222 prepare_to_wait(&root->fs_info->transaction_throttle, &wait,
223 TASK_UNINTERRUPTIBLE);
224 if (atomic_read(&root->fs_info->throttles))
226 finish_wait(&root->fs_info->transaction_throttle, &wait);
227 mutex_lock(&root->fs_info->trans_mutex);
230 put_transaction(cur_trans);
231 mutex_unlock(&root->fs_info->trans_mutex);
232 memset(trans, 0, sizeof(*trans));
233 kmem_cache_free(btrfs_trans_handle_cachep, trans);
237 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
238 struct btrfs_root *root)
240 return __btrfs_end_transaction(trans, root, 0);
243 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
244 struct btrfs_root *root)
246 return __btrfs_end_transaction(trans, root, 1);
250 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
251 struct btrfs_root *root)
256 struct extent_io_tree *dirty_pages;
258 struct inode *btree_inode = root->fs_info->btree_inode;
263 if (!trans || !trans->transaction) {
264 return filemap_write_and_wait(btree_inode->i_mapping);
266 dirty_pages = &trans->transaction->dirty_pages;
268 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
272 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
273 while(start <= end) {
274 index = start >> PAGE_CACHE_SHIFT;
275 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
276 page = find_lock_page(btree_inode->i_mapping, index);
279 if (PageWriteback(page)) {
281 wait_on_page_writeback(page);
284 page_cache_release(page);
288 err = write_one_page(page, 0);
291 page_cache_release(page);
294 err = filemap_fdatawait(btree_inode->i_mapping);
300 static int update_cowonly_root(struct btrfs_trans_handle *trans,
301 struct btrfs_root *root)
305 struct btrfs_root *tree_root = root->fs_info->tree_root;
307 btrfs_write_dirty_block_groups(trans, root);
309 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
310 if (old_root_bytenr == root->node->start)
312 btrfs_set_root_bytenr(&root->root_item,
314 btrfs_set_root_level(&root->root_item,
315 btrfs_header_level(root->node));
316 ret = btrfs_update_root(trans, tree_root,
320 btrfs_write_dirty_block_groups(trans, root);
325 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
326 struct btrfs_root *root)
328 struct btrfs_fs_info *fs_info = root->fs_info;
329 struct list_head *next;
331 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
332 next = fs_info->dirty_cowonly_roots.next;
334 root = list_entry(next, struct btrfs_root, dirty_list);
335 update_cowonly_root(trans, root);
336 if (root->fs_info->closing)
337 btrfs_remove_leaf_refs(root);
342 int btrfs_add_dead_root(struct btrfs_root *root,
343 struct btrfs_root *latest,
344 struct list_head *dead_list)
346 struct dirty_root *dirty;
348 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
352 dirty->latest_root = latest;
353 list_add(&dirty->list, dead_list);
357 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
358 struct radix_tree_root *radix,
359 struct list_head *list)
361 struct dirty_root *dirty;
362 struct btrfs_root *gang[8];
363 struct btrfs_root *root;
370 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
372 BTRFS_ROOT_TRANS_TAG);
375 for (i = 0; i < ret; i++) {
377 radix_tree_tag_clear(radix,
378 (unsigned long)root->root_key.objectid,
379 BTRFS_ROOT_TRANS_TAG);
381 BUG_ON(!root->ref_tree);
382 dirty = root->dirty_root;
384 if (root->commit_root == root->node) {
385 WARN_ON(root->node->start !=
386 btrfs_root_bytenr(&root->root_item));
388 free_extent_buffer(root->commit_root);
389 root->commit_root = NULL;
394 /* make sure to update the root on disk
395 * so we get any updates to the block used
398 err = btrfs_update_root(trans,
399 root->fs_info->tree_root,
405 memset(&root->root_item.drop_progress, 0,
406 sizeof(struct btrfs_disk_key));
407 root->root_item.drop_level = 0;
408 root->commit_root = NULL;
409 root->root_key.offset = root->fs_info->generation;
410 btrfs_set_root_bytenr(&root->root_item,
412 btrfs_set_root_level(&root->root_item,
413 btrfs_header_level(root->node));
414 err = btrfs_insert_root(trans, root->fs_info->tree_root,
420 refs = btrfs_root_refs(&dirty->root->root_item);
421 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
422 err = btrfs_update_root(trans, root->fs_info->tree_root,
423 &dirty->root->root_key,
424 &dirty->root->root_item);
428 list_add(&dirty->list, list);
431 free_extent_buffer(dirty->root->node);
440 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
442 struct btrfs_fs_info *info = root->fs_info;
444 struct btrfs_trans_handle *trans;
448 if (root->defrag_running)
450 trans = btrfs_start_transaction(root, 1);
452 root->defrag_running = 1;
453 ret = btrfs_defrag_leaves(trans, root, cacheonly);
454 nr = trans->blocks_used;
455 btrfs_end_transaction(trans, root);
456 btrfs_btree_balance_dirty(info->tree_root, nr);
459 trans = btrfs_start_transaction(root, 1);
460 if (root->fs_info->closing || ret != -EAGAIN)
463 root->defrag_running = 0;
465 btrfs_end_transaction(trans, root);
469 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
470 struct list_head *list)
472 struct dirty_root *dirty;
473 struct btrfs_trans_handle *trans;
480 while(!list_empty(list)) {
481 struct btrfs_root *root;
483 dirty = list_entry(list->prev, struct dirty_root, list);
484 list_del_init(&dirty->list);
486 num_bytes = btrfs_root_used(&dirty->root->root_item);
487 root = dirty->latest_root;
488 atomic_inc(&root->fs_info->throttles);
490 mutex_lock(&root->fs_info->drop_mutex);
492 trans = btrfs_start_transaction(tree_root, 1);
493 ret = btrfs_drop_snapshot(trans, dirty->root);
494 if (ret != -EAGAIN) {
498 err = btrfs_update_root(trans,
500 &dirty->root->root_key,
501 &dirty->root->root_item);
504 nr = trans->blocks_used;
505 ret = btrfs_end_transaction(trans, tree_root);
508 mutex_unlock(&root->fs_info->drop_mutex);
509 btrfs_btree_balance_dirty(tree_root, nr);
511 mutex_lock(&root->fs_info->drop_mutex);
514 atomic_dec(&root->fs_info->throttles);
515 wake_up(&root->fs_info->transaction_throttle);
517 mutex_lock(&root->fs_info->alloc_mutex);
518 num_bytes -= btrfs_root_used(&dirty->root->root_item);
519 bytes_used = btrfs_root_used(&root->root_item);
521 record_root_in_trans(root);
522 btrfs_set_root_used(&root->root_item,
523 bytes_used - num_bytes);
525 mutex_unlock(&root->fs_info->alloc_mutex);
527 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
532 mutex_unlock(&root->fs_info->drop_mutex);
534 nr = trans->blocks_used;
535 ret = btrfs_end_transaction(trans, tree_root);
538 free_extent_buffer(dirty->root->node);
542 btrfs_btree_balance_dirty(tree_root, nr);
548 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
549 struct btrfs_fs_info *fs_info,
550 struct btrfs_pending_snapshot *pending)
552 struct btrfs_key key;
553 struct btrfs_root_item *new_root_item;
554 struct btrfs_root *tree_root = fs_info->tree_root;
555 struct btrfs_root *root = pending->root;
556 struct extent_buffer *tmp;
557 struct extent_buffer *old;
562 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
563 if (!new_root_item) {
567 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
571 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
573 key.objectid = objectid;
575 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
577 old = btrfs_lock_root_node(root);
578 btrfs_cow_block(trans, root, old, NULL, 0, &old);
580 btrfs_copy_root(trans, root, old, &tmp, objectid);
581 btrfs_tree_unlock(old);
582 free_extent_buffer(old);
584 btrfs_set_root_bytenr(new_root_item, tmp->start);
585 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
586 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
588 btrfs_tree_unlock(tmp);
589 free_extent_buffer(tmp);
594 * insert the directory item
596 key.offset = (u64)-1;
597 namelen = strlen(pending->name);
598 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
599 pending->name, namelen,
600 root->fs_info->sb->s_root->d_inode->i_ino,
601 &key, BTRFS_FT_DIR, 0);
606 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
607 pending->name, strlen(pending->name), objectid,
608 root->fs_info->sb->s_root->d_inode->i_ino, 0);
610 /* Invalidate existing dcache entry for new snapshot. */
611 btrfs_invalidate_dcache_root(root, pending->name, namelen);
614 kfree(new_root_item);
618 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
619 struct btrfs_fs_info *fs_info)
621 struct btrfs_pending_snapshot *pending;
622 struct list_head *head = &trans->transaction->pending_snapshots;
625 while(!list_empty(head)) {
626 pending = list_entry(head->next,
627 struct btrfs_pending_snapshot, list);
628 ret = create_pending_snapshot(trans, fs_info, pending);
630 list_del(&pending->list);
631 kfree(pending->name);
637 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root)
640 unsigned long joined = 0;
641 unsigned long timeout = 1;
642 struct btrfs_transaction *cur_trans;
643 struct btrfs_transaction *prev_trans = NULL;
644 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
645 struct list_head dirty_fs_roots;
646 struct extent_io_tree *pinned_copy;
650 INIT_LIST_HEAD(&dirty_fs_roots);
652 mutex_lock(&root->fs_info->trans_mutex);
653 if (trans->transaction->in_commit) {
654 cur_trans = trans->transaction;
655 trans->transaction->use_count++;
656 mutex_unlock(&root->fs_info->trans_mutex);
657 btrfs_end_transaction(trans, root);
659 ret = wait_for_commit(root, cur_trans);
662 mutex_lock(&root->fs_info->trans_mutex);
663 put_transaction(cur_trans);
664 mutex_unlock(&root->fs_info->trans_mutex);
669 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
673 extent_io_tree_init(pinned_copy,
674 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
676 trans->transaction->in_commit = 1;
677 trans->transaction->blocked = 1;
678 cur_trans = trans->transaction;
679 if (cur_trans->list.prev != &root->fs_info->trans_list) {
680 prev_trans = list_entry(cur_trans->list.prev,
681 struct btrfs_transaction, list);
682 if (!prev_trans->commit_done) {
683 prev_trans->use_count++;
684 mutex_unlock(&root->fs_info->trans_mutex);
686 wait_for_commit(root, prev_trans);
688 mutex_lock(&root->fs_info->trans_mutex);
689 put_transaction(prev_trans);
694 joined = cur_trans->num_joined;
695 WARN_ON(cur_trans != trans->transaction);
696 prepare_to_wait(&cur_trans->writer_wait, &wait,
697 TASK_UNINTERRUPTIBLE);
699 if (cur_trans->num_writers > 1)
700 timeout = MAX_SCHEDULE_TIMEOUT;
704 mutex_unlock(&root->fs_info->trans_mutex);
706 schedule_timeout(timeout);
708 mutex_lock(&root->fs_info->trans_mutex);
709 finish_wait(&cur_trans->writer_wait, &wait);
710 } while (cur_trans->num_writers > 1 ||
711 (cur_trans->num_joined != joined));
713 ret = create_pending_snapshots(trans, root->fs_info);
716 WARN_ON(cur_trans != trans->transaction);
718 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
722 ret = btrfs_commit_tree_roots(trans, root);
725 cur_trans = root->fs_info->running_transaction;
726 spin_lock(&root->fs_info->new_trans_lock);
727 root->fs_info->running_transaction = NULL;
728 spin_unlock(&root->fs_info->new_trans_lock);
729 btrfs_set_super_generation(&root->fs_info->super_copy,
731 btrfs_set_super_root(&root->fs_info->super_copy,
732 root->fs_info->tree_root->node->start);
733 btrfs_set_super_root_level(&root->fs_info->super_copy,
734 btrfs_header_level(root->fs_info->tree_root->node));
736 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
737 chunk_root->node->start);
738 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
739 btrfs_header_level(chunk_root->node));
740 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
741 sizeof(root->fs_info->super_copy));
743 btrfs_copy_pinned(root, pinned_copy);
745 trans->transaction->blocked = 0;
746 wake_up(&root->fs_info->transaction_throttle);
747 wake_up(&root->fs_info->transaction_wait);
749 mutex_unlock(&root->fs_info->trans_mutex);
750 ret = btrfs_write_and_wait_transaction(trans, root);
752 write_ctree_super(trans, root);
754 btrfs_finish_extent_commit(trans, root, pinned_copy);
755 mutex_lock(&root->fs_info->trans_mutex);
759 cur_trans->commit_done = 1;
760 root->fs_info->last_trans_committed = cur_trans->transid;
761 wake_up(&cur_trans->commit_wait);
762 put_transaction(cur_trans);
763 put_transaction(cur_trans);
765 if (root->fs_info->closing)
766 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
768 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
770 mutex_unlock(&root->fs_info->trans_mutex);
771 kmem_cache_free(btrfs_trans_handle_cachep, trans);
773 if (root->fs_info->closing) {
774 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
779 int btrfs_clean_old_snapshots(struct btrfs_root *root)
781 struct list_head dirty_roots;
782 INIT_LIST_HEAD(&dirty_roots);
784 mutex_lock(&root->fs_info->trans_mutex);
785 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
786 mutex_unlock(&root->fs_info->trans_mutex);
788 if (!list_empty(&dirty_roots)) {
789 drop_dirty_roots(root, &dirty_roots);