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 void btrfs_throttle(struct btrfs_root *root)
207 struct btrfs_fs_info *info = root->fs_info;
210 if (atomic_read(&info->throttles)) {
213 int harder_count = 0;
214 thr = atomic_read(&info->throttle_gen);
217 prepare_to_wait(&info->transaction_throttle,
218 &wait, TASK_UNINTERRUPTIBLE);
219 if (!atomic_read(&info->throttles)) {
220 finish_wait(&info->transaction_throttle, &wait);
224 finish_wait(&info->transaction_throttle, &wait);
225 } while (thr == atomic_read(&info->throttle_gen));
227 if (harder_count < 5 &&
228 info->total_ref_cache_size > 5 * 1024 * 1024) {
233 if (harder_count < 10 &&
234 info->total_ref_cache_size > 10 * 1024 * 1024) {
241 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
242 struct btrfs_root *root, int throttle)
244 struct btrfs_transaction *cur_trans;
245 struct btrfs_fs_info *info = root->fs_info;
247 mutex_lock(&info->trans_mutex);
248 cur_trans = info->running_transaction;
249 WARN_ON(cur_trans != trans->transaction);
250 WARN_ON(cur_trans->num_writers < 1);
251 cur_trans->num_writers--;
253 if (waitqueue_active(&cur_trans->writer_wait))
254 wake_up(&cur_trans->writer_wait);
255 put_transaction(cur_trans);
256 mutex_unlock(&info->trans_mutex);
257 memset(trans, 0, sizeof(*trans));
258 kmem_cache_free(btrfs_trans_handle_cachep, trans);
261 btrfs_throttle(root);
266 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
267 struct btrfs_root *root)
269 return __btrfs_end_transaction(trans, root, 0);
272 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
273 struct btrfs_root *root)
275 return __btrfs_end_transaction(trans, root, 1);
279 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
280 struct btrfs_root *root)
285 struct extent_io_tree *dirty_pages;
287 struct inode *btree_inode = root->fs_info->btree_inode;
292 if (!trans || !trans->transaction) {
293 return filemap_write_and_wait(btree_inode->i_mapping);
295 dirty_pages = &trans->transaction->dirty_pages;
297 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
301 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
302 while(start <= end) {
303 index = start >> PAGE_CACHE_SHIFT;
304 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
305 page = find_lock_page(btree_inode->i_mapping, index);
308 if (PageWriteback(page)) {
310 wait_on_page_writeback(page);
313 page_cache_release(page);
317 err = write_one_page(page, 0);
320 page_cache_release(page);
323 err = filemap_fdatawait(btree_inode->i_mapping);
329 static int update_cowonly_root(struct btrfs_trans_handle *trans,
330 struct btrfs_root *root)
334 struct btrfs_root *tree_root = root->fs_info->tree_root;
336 btrfs_write_dirty_block_groups(trans, root);
338 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
339 if (old_root_bytenr == root->node->start)
341 btrfs_set_root_bytenr(&root->root_item,
343 btrfs_set_root_level(&root->root_item,
344 btrfs_header_level(root->node));
345 ret = btrfs_update_root(trans, tree_root,
349 btrfs_write_dirty_block_groups(trans, root);
354 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
355 struct btrfs_root *root)
357 struct btrfs_fs_info *fs_info = root->fs_info;
358 struct list_head *next;
360 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
361 next = fs_info->dirty_cowonly_roots.next;
363 root = list_entry(next, struct btrfs_root, dirty_list);
364 update_cowonly_root(trans, root);
365 if (root->fs_info->closing)
366 btrfs_remove_leaf_refs(root);
371 int btrfs_add_dead_root(struct btrfs_root *root,
372 struct btrfs_root *latest,
373 struct list_head *dead_list)
375 struct dirty_root *dirty;
377 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
381 dirty->latest_root = latest;
382 list_add(&dirty->list, dead_list);
386 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
387 struct radix_tree_root *radix,
388 struct list_head *list)
390 struct dirty_root *dirty;
391 struct btrfs_root *gang[8];
392 struct btrfs_root *root;
399 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
401 BTRFS_ROOT_TRANS_TAG);
404 for (i = 0; i < ret; i++) {
406 radix_tree_tag_clear(radix,
407 (unsigned long)root->root_key.objectid,
408 BTRFS_ROOT_TRANS_TAG);
410 BUG_ON(!root->ref_tree);
411 dirty = root->dirty_root;
413 if (root->commit_root == root->node) {
414 WARN_ON(root->node->start !=
415 btrfs_root_bytenr(&root->root_item));
417 free_extent_buffer(root->commit_root);
418 root->commit_root = NULL;
423 /* make sure to update the root on disk
424 * so we get any updates to the block used
427 err = btrfs_update_root(trans,
428 root->fs_info->tree_root,
434 memset(&root->root_item.drop_progress, 0,
435 sizeof(struct btrfs_disk_key));
436 root->root_item.drop_level = 0;
437 root->commit_root = NULL;
438 root->root_key.offset = root->fs_info->generation;
439 btrfs_set_root_bytenr(&root->root_item,
441 btrfs_set_root_level(&root->root_item,
442 btrfs_header_level(root->node));
443 err = btrfs_insert_root(trans, root->fs_info->tree_root,
449 refs = btrfs_root_refs(&dirty->root->root_item);
450 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
451 err = btrfs_update_root(trans, root->fs_info->tree_root,
452 &dirty->root->root_key,
453 &dirty->root->root_item);
457 list_add(&dirty->list, list);
460 free_extent_buffer(dirty->root->node);
469 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
471 struct btrfs_fs_info *info = root->fs_info;
473 struct btrfs_trans_handle *trans;
477 if (root->defrag_running)
479 trans = btrfs_start_transaction(root, 1);
481 root->defrag_running = 1;
482 ret = btrfs_defrag_leaves(trans, root, cacheonly);
483 nr = trans->blocks_used;
484 btrfs_end_transaction(trans, root);
485 btrfs_btree_balance_dirty(info->tree_root, nr);
488 trans = btrfs_start_transaction(root, 1);
489 if (root->fs_info->closing || ret != -EAGAIN)
492 root->defrag_running = 0;
494 btrfs_end_transaction(trans, root);
498 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
499 struct list_head *list)
501 struct dirty_root *dirty;
502 struct btrfs_trans_handle *trans;
509 while(!list_empty(list)) {
510 struct btrfs_root *root;
512 dirty = list_entry(list->prev, struct dirty_root, list);
513 list_del_init(&dirty->list);
515 num_bytes = btrfs_root_used(&dirty->root->root_item);
516 root = dirty->latest_root;
517 atomic_inc(&root->fs_info->throttles);
519 mutex_lock(&root->fs_info->drop_mutex);
521 trans = btrfs_start_transaction(tree_root, 1);
522 ret = btrfs_drop_snapshot(trans, dirty->root);
523 if (ret != -EAGAIN) {
527 err = btrfs_update_root(trans,
529 &dirty->root->root_key,
530 &dirty->root->root_item);
533 nr = trans->blocks_used;
534 ret = btrfs_end_transaction(trans, tree_root);
537 mutex_unlock(&root->fs_info->drop_mutex);
538 btrfs_btree_balance_dirty(tree_root, nr);
540 mutex_lock(&root->fs_info->drop_mutex);
543 atomic_dec(&root->fs_info->throttles);
544 wake_up(&root->fs_info->transaction_throttle);
546 mutex_lock(&root->fs_info->alloc_mutex);
547 num_bytes -= btrfs_root_used(&dirty->root->root_item);
548 bytes_used = btrfs_root_used(&root->root_item);
550 record_root_in_trans(root);
551 btrfs_set_root_used(&root->root_item,
552 bytes_used - num_bytes);
554 mutex_unlock(&root->fs_info->alloc_mutex);
556 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
561 mutex_unlock(&root->fs_info->drop_mutex);
563 nr = trans->blocks_used;
564 ret = btrfs_end_transaction(trans, tree_root);
567 free_extent_buffer(dirty->root->node);
571 btrfs_btree_balance_dirty(tree_root, nr);
577 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
578 struct btrfs_fs_info *fs_info,
579 struct btrfs_pending_snapshot *pending)
581 struct btrfs_key key;
582 struct btrfs_root_item *new_root_item;
583 struct btrfs_root *tree_root = fs_info->tree_root;
584 struct btrfs_root *root = pending->root;
585 struct extent_buffer *tmp;
586 struct extent_buffer *old;
591 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
592 if (!new_root_item) {
596 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
600 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
602 key.objectid = objectid;
604 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
606 old = btrfs_lock_root_node(root);
607 btrfs_cow_block(trans, root, old, NULL, 0, &old);
609 btrfs_copy_root(trans, root, old, &tmp, objectid);
610 btrfs_tree_unlock(old);
611 free_extent_buffer(old);
613 btrfs_set_root_bytenr(new_root_item, tmp->start);
614 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
615 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
617 btrfs_tree_unlock(tmp);
618 free_extent_buffer(tmp);
623 * insert the directory item
625 key.offset = (u64)-1;
626 namelen = strlen(pending->name);
627 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
628 pending->name, namelen,
629 root->fs_info->sb->s_root->d_inode->i_ino,
630 &key, BTRFS_FT_DIR, 0);
635 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
636 pending->name, strlen(pending->name), objectid,
637 root->fs_info->sb->s_root->d_inode->i_ino, 0);
639 /* Invalidate existing dcache entry for new snapshot. */
640 btrfs_invalidate_dcache_root(root, pending->name, namelen);
643 kfree(new_root_item);
647 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
648 struct btrfs_fs_info *fs_info)
650 struct btrfs_pending_snapshot *pending;
651 struct list_head *head = &trans->transaction->pending_snapshots;
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);
659 list_del(&pending->list);
660 kfree(pending->name);
666 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
667 struct btrfs_root *root)
669 unsigned long joined = 0;
670 unsigned long timeout = 1;
671 struct btrfs_transaction *cur_trans;
672 struct btrfs_transaction *prev_trans = NULL;
673 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
674 struct list_head dirty_fs_roots;
675 struct extent_io_tree *pinned_copy;
679 INIT_LIST_HEAD(&dirty_fs_roots);
681 mutex_lock(&root->fs_info->trans_mutex);
682 if (trans->transaction->in_commit) {
683 cur_trans = trans->transaction;
684 trans->transaction->use_count++;
685 mutex_unlock(&root->fs_info->trans_mutex);
686 btrfs_end_transaction(trans, root);
688 ret = wait_for_commit(root, cur_trans);
691 mutex_lock(&root->fs_info->trans_mutex);
692 put_transaction(cur_trans);
693 mutex_unlock(&root->fs_info->trans_mutex);
698 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
702 extent_io_tree_init(pinned_copy,
703 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
705 trans->transaction->in_commit = 1;
706 trans->transaction->blocked = 1;
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++;
713 mutex_unlock(&root->fs_info->trans_mutex);
715 wait_for_commit(root, prev_trans);
717 mutex_lock(&root->fs_info->trans_mutex);
718 put_transaction(prev_trans);
723 joined = cur_trans->num_joined;
724 WARN_ON(cur_trans != trans->transaction);
725 prepare_to_wait(&cur_trans->writer_wait, &wait,
726 TASK_UNINTERRUPTIBLE);
728 if (cur_trans->num_writers > 1)
729 timeout = MAX_SCHEDULE_TIMEOUT;
733 mutex_unlock(&root->fs_info->trans_mutex);
735 schedule_timeout(timeout);
737 mutex_lock(&root->fs_info->trans_mutex);
738 finish_wait(&cur_trans->writer_wait, &wait);
739 } while (cur_trans->num_writers > 1 ||
740 (cur_trans->num_joined != joined));
742 ret = create_pending_snapshots(trans, root->fs_info);
745 WARN_ON(cur_trans != trans->transaction);
747 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
751 ret = btrfs_commit_tree_roots(trans, root);
754 cur_trans = root->fs_info->running_transaction;
755 spin_lock(&root->fs_info->new_trans_lock);
756 root->fs_info->running_transaction = NULL;
757 spin_unlock(&root->fs_info->new_trans_lock);
758 btrfs_set_super_generation(&root->fs_info->super_copy,
760 btrfs_set_super_root(&root->fs_info->super_copy,
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));
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));
769 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
770 sizeof(root->fs_info->super_copy));
772 btrfs_copy_pinned(root, pinned_copy);
774 trans->transaction->blocked = 0;
775 wake_up(&root->fs_info->transaction_throttle);
776 wake_up(&root->fs_info->transaction_wait);
778 mutex_unlock(&root->fs_info->trans_mutex);
779 ret = btrfs_write_and_wait_transaction(trans, root);
781 write_ctree_super(trans, root);
783 btrfs_finish_extent_commit(trans, root, pinned_copy);
784 mutex_lock(&root->fs_info->trans_mutex);
788 cur_trans->commit_done = 1;
789 root->fs_info->last_trans_committed = cur_trans->transid;
790 wake_up(&cur_trans->commit_wait);
791 put_transaction(cur_trans);
792 put_transaction(cur_trans);
794 if (root->fs_info->closing)
795 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
797 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
799 mutex_unlock(&root->fs_info->trans_mutex);
800 kmem_cache_free(btrfs_trans_handle_cachep, trans);
802 if (root->fs_info->closing) {
803 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
808 int btrfs_clean_old_snapshots(struct btrfs_root *root)
810 struct list_head dirty_roots;
811 INIT_LIST_HEAD(&dirty_roots);
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);
817 if (!list_empty(&dirty_roots)) {
818 drop_dirty_roots(root, &dirty_roots);