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
35 static noinline void put_transaction(struct btrfs_transaction *transaction)
37 WARN_ON(transaction->use_count == 0);
38 transaction->use_count--;
39 if (transaction->use_count == 0) {
40 WARN_ON(total_trans == 0);
42 list_del_init(&transaction->list);
43 memset(transaction, 0, sizeof(*transaction));
44 kmem_cache_free(btrfs_transaction_cachep, transaction);
48 static noinline int join_transaction(struct btrfs_root *root)
50 struct btrfs_transaction *cur_trans;
51 cur_trans = root->fs_info->running_transaction;
53 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
57 root->fs_info->generation++;
58 root->fs_info->last_alloc = 0;
59 root->fs_info->last_data_alloc = 0;
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->blocked = 0;
67 cur_trans->use_count = 1;
68 cur_trans->commit_done = 0;
69 cur_trans->start_time = get_seconds();
70 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
71 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
72 extent_io_tree_init(&cur_trans->dirty_pages,
73 root->fs_info->btree_inode->i_mapping,
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);
79 cur_trans->num_writers++;
80 cur_trans->num_joined++;
86 static noinline int record_root_in_trans(struct btrfs_root *root)
88 struct btrfs_dirty_root *dirty;
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);
97 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
99 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
100 BUG_ON(!dirty->root);
101 dirty->latest_root = root;
102 INIT_LIST_HEAD(&dirty->list);
104 root->commit_root = btrfs_root_node(root);
106 memcpy(dirty->root, root, sizeof(*root));
107 spin_lock_init(&dirty->root->node_lock);
108 spin_lock_init(&dirty->root->list_lock);
109 mutex_init(&dirty->root->objectid_mutex);
110 INIT_LIST_HEAD(&dirty->root->dead_list);
111 dirty->root->node = root->commit_root;
112 dirty->root->commit_root = NULL;
114 spin_lock(&root->list_lock);
115 list_add(&dirty->root->dead_list, &root->dead_list);
116 spin_unlock(&root->list_lock);
118 root->dirty_root = dirty;
122 root->last_trans = running_trans_id;
127 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
128 int num_blocks, int join)
130 struct btrfs_trans_handle *h =
131 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
132 struct btrfs_transaction *cur_trans;
135 mutex_lock(&root->fs_info->trans_mutex);
136 cur_trans = root->fs_info->running_transaction;
137 if (cur_trans && cur_trans->blocked && !join) {
139 cur_trans->use_count++;
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);
146 mutex_lock(&root->fs_info->trans_mutex);
147 finish_wait(&root->fs_info->transaction_wait,
150 finish_wait(&root->fs_info->transaction_wait,
155 put_transaction(cur_trans);
157 ret = join_transaction(root);
160 record_root_in_trans(root);
161 h->transid = root->fs_info->running_transaction->transid;
162 h->transaction = root->fs_info->running_transaction;
163 h->blocks_reserved = num_blocks;
165 h->block_group = NULL;
166 h->alloc_exclude_nr = 0;
167 h->alloc_exclude_start = 0;
168 root->fs_info->running_transaction->use_count++;
169 mutex_unlock(&root->fs_info->trans_mutex);
173 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
176 return start_transaction(root, num_blocks, 0);
178 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
181 return start_transaction(root, num_blocks, 1);
184 static noinline int wait_for_commit(struct btrfs_root *root,
185 struct btrfs_transaction *commit)
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)
194 mutex_unlock(&root->fs_info->trans_mutex);
196 mutex_lock(&root->fs_info->trans_mutex);
198 mutex_unlock(&root->fs_info->trans_mutex);
199 finish_wait(&commit->commit_wait, &wait);
203 void btrfs_throttle(struct btrfs_root *root)
205 struct btrfs_fs_info *info = root->fs_info;
208 if (atomic_read(&info->throttles)) {
211 int harder_count = 0;
212 thr = atomic_read(&info->throttle_gen);
215 prepare_to_wait(&info->transaction_throttle,
216 &wait, TASK_UNINTERRUPTIBLE);
217 if (!atomic_read(&info->throttles)) {
218 finish_wait(&info->transaction_throttle, &wait);
222 finish_wait(&info->transaction_throttle, &wait);
223 } while (thr == atomic_read(&info->throttle_gen));
225 if (harder_count < 5 &&
226 info->total_ref_cache_size > 5 * 1024 * 1024) {
231 if (harder_count < 10 &&
232 info->total_ref_cache_size > 10 * 1024 * 1024) {
239 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
240 struct btrfs_root *root, int throttle)
242 struct btrfs_transaction *cur_trans;
243 struct btrfs_fs_info *info = root->fs_info;
245 mutex_lock(&info->trans_mutex);
246 cur_trans = info->running_transaction;
247 WARN_ON(cur_trans != trans->transaction);
248 WARN_ON(cur_trans->num_writers < 1);
249 cur_trans->num_writers--;
251 if (waitqueue_active(&cur_trans->writer_wait))
252 wake_up(&cur_trans->writer_wait);
253 put_transaction(cur_trans);
254 mutex_unlock(&info->trans_mutex);
255 memset(trans, 0, sizeof(*trans));
256 kmem_cache_free(btrfs_trans_handle_cachep, trans);
259 btrfs_throttle(root);
264 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
265 struct btrfs_root *root)
267 return __btrfs_end_transaction(trans, root, 0);
270 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
271 struct btrfs_root *root)
273 return __btrfs_end_transaction(trans, root, 1);
277 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root)
283 struct extent_io_tree *dirty_pages;
285 struct inode *btree_inode = root->fs_info->btree_inode;
290 if (!trans || !trans->transaction) {
291 return filemap_write_and_wait(btree_inode->i_mapping);
293 dirty_pages = &trans->transaction->dirty_pages;
295 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
299 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
300 while(start <= end) {
301 index = start >> PAGE_CACHE_SHIFT;
302 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
303 page = find_lock_page(btree_inode->i_mapping, index);
306 if (PageWriteback(page)) {
308 wait_on_page_writeback(page);
311 page_cache_release(page);
315 err = write_one_page(page, 0);
318 page_cache_release(page);
321 err = filemap_fdatawait(btree_inode->i_mapping);
327 static int update_cowonly_root(struct btrfs_trans_handle *trans,
328 struct btrfs_root *root)
332 struct btrfs_root *tree_root = root->fs_info->tree_root;
334 btrfs_write_dirty_block_groups(trans, root);
336 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
337 if (old_root_bytenr == root->node->start)
339 btrfs_set_root_bytenr(&root->root_item,
341 btrfs_set_root_level(&root->root_item,
342 btrfs_header_level(root->node));
343 ret = btrfs_update_root(trans, tree_root,
347 btrfs_write_dirty_block_groups(trans, root);
352 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
353 struct btrfs_root *root)
355 struct btrfs_fs_info *fs_info = root->fs_info;
356 struct list_head *next;
358 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
359 next = fs_info->dirty_cowonly_roots.next;
361 root = list_entry(next, struct btrfs_root, dirty_list);
362 update_cowonly_root(trans, root);
367 int btrfs_add_dead_root(struct btrfs_root *root,
368 struct btrfs_root *latest,
369 struct list_head *dead_list)
371 struct btrfs_dirty_root *dirty;
373 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
377 dirty->latest_root = latest;
378 list_add(&dirty->list, dead_list);
382 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
383 struct radix_tree_root *radix,
384 struct list_head *list)
386 struct btrfs_dirty_root *dirty;
387 struct btrfs_root *gang[8];
388 struct btrfs_root *root;
395 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
397 BTRFS_ROOT_TRANS_TAG);
400 for (i = 0; i < ret; i++) {
402 radix_tree_tag_clear(radix,
403 (unsigned long)root->root_key.objectid,
404 BTRFS_ROOT_TRANS_TAG);
406 BUG_ON(!root->ref_tree);
407 dirty = root->dirty_root;
409 if (root->commit_root == root->node) {
410 WARN_ON(root->node->start !=
411 btrfs_root_bytenr(&root->root_item));
413 free_extent_buffer(root->commit_root);
414 root->commit_root = NULL;
416 spin_lock(&root->list_lock);
417 list_del_init(&dirty->root->dead_list);
418 spin_unlock(&root->list_lock);
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 btrfs_dirty_root *dirty;
502 struct btrfs_trans_handle *trans;
510 while(!list_empty(list)) {
511 struct btrfs_root *root;
513 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
514 list_del_init(&dirty->list);
516 num_bytes = btrfs_root_used(&dirty->root->root_item);
517 root = dirty->latest_root;
518 atomic_inc(&root->fs_info->throttles);
520 mutex_lock(&root->fs_info->drop_mutex);
522 trans = btrfs_start_transaction(tree_root, 1);
523 ret = btrfs_drop_snapshot(trans, dirty->root);
524 if (ret != -EAGAIN) {
528 err = btrfs_update_root(trans,
530 &dirty->root->root_key,
531 &dirty->root->root_item);
534 nr = trans->blocks_used;
535 ret = btrfs_end_transaction(trans, tree_root);
538 mutex_unlock(&root->fs_info->drop_mutex);
539 btrfs_btree_balance_dirty(tree_root, nr);
541 mutex_lock(&root->fs_info->drop_mutex);
544 atomic_dec(&root->fs_info->throttles);
545 wake_up(&root->fs_info->transaction_throttle);
547 mutex_lock(&root->fs_info->alloc_mutex);
548 num_bytes -= btrfs_root_used(&dirty->root->root_item);
549 bytes_used = btrfs_root_used(&root->root_item);
551 record_root_in_trans(root);
552 btrfs_set_root_used(&root->root_item,
553 bytes_used - num_bytes);
555 mutex_unlock(&root->fs_info->alloc_mutex);
557 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
562 mutex_unlock(&root->fs_info->drop_mutex);
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;
572 max_useless = root->root_key.offset - 1;
574 spin_unlock(&root->list_lock);
576 nr = trans->blocks_used;
577 ret = btrfs_end_transaction(trans, tree_root);
580 ret = btrfs_remove_leaf_refs(root, max_useless);
583 free_extent_buffer(dirty->root->node);
587 btrfs_btree_balance_dirty(tree_root, nr);
593 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
594 struct btrfs_fs_info *fs_info,
595 struct btrfs_pending_snapshot *pending)
597 struct btrfs_key key;
598 struct btrfs_root_item *new_root_item;
599 struct btrfs_root *tree_root = fs_info->tree_root;
600 struct btrfs_root *root = pending->root;
601 struct extent_buffer *tmp;
602 struct extent_buffer *old;
607 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
608 if (!new_root_item) {
612 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
616 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
618 key.objectid = objectid;
620 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
622 old = btrfs_lock_root_node(root);
623 btrfs_cow_block(trans, root, old, NULL, 0, &old);
625 btrfs_copy_root(trans, root, old, &tmp, objectid);
626 btrfs_tree_unlock(old);
627 free_extent_buffer(old);
629 btrfs_set_root_bytenr(new_root_item, tmp->start);
630 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
631 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
633 btrfs_tree_unlock(tmp);
634 free_extent_buffer(tmp);
639 * insert the directory item
641 key.offset = (u64)-1;
642 namelen = strlen(pending->name);
643 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
644 pending->name, namelen,
645 root->fs_info->sb->s_root->d_inode->i_ino,
646 &key, BTRFS_FT_DIR, 0);
651 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
652 pending->name, strlen(pending->name), objectid,
653 root->fs_info->sb->s_root->d_inode->i_ino, 0);
655 /* Invalidate existing dcache entry for new snapshot. */
656 btrfs_invalidate_dcache_root(root, pending->name, namelen);
659 kfree(new_root_item);
663 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
664 struct btrfs_fs_info *fs_info)
666 struct btrfs_pending_snapshot *pending;
667 struct list_head *head = &trans->transaction->pending_snapshots;
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);
675 list_del(&pending->list);
676 kfree(pending->name);
682 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
683 struct btrfs_root *root)
685 unsigned long joined = 0;
686 unsigned long timeout = 1;
687 struct btrfs_transaction *cur_trans;
688 struct btrfs_transaction *prev_trans = NULL;
689 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
690 struct list_head dirty_fs_roots;
691 struct extent_io_tree *pinned_copy;
695 INIT_LIST_HEAD(&dirty_fs_roots);
697 mutex_lock(&root->fs_info->trans_mutex);
698 if (trans->transaction->in_commit) {
699 cur_trans = trans->transaction;
700 trans->transaction->use_count++;
701 mutex_unlock(&root->fs_info->trans_mutex);
702 btrfs_end_transaction(trans, root);
704 ret = wait_for_commit(root, cur_trans);
707 mutex_lock(&root->fs_info->trans_mutex);
708 put_transaction(cur_trans);
709 mutex_unlock(&root->fs_info->trans_mutex);
714 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
718 extent_io_tree_init(pinned_copy,
719 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
721 trans->transaction->in_commit = 1;
722 trans->transaction->blocked = 1;
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++;
729 mutex_unlock(&root->fs_info->trans_mutex);
731 wait_for_commit(root, prev_trans);
733 mutex_lock(&root->fs_info->trans_mutex);
734 put_transaction(prev_trans);
739 joined = cur_trans->num_joined;
740 WARN_ON(cur_trans != trans->transaction);
741 prepare_to_wait(&cur_trans->writer_wait, &wait,
742 TASK_UNINTERRUPTIBLE);
744 if (cur_trans->num_writers > 1)
745 timeout = MAX_SCHEDULE_TIMEOUT;
749 mutex_unlock(&root->fs_info->trans_mutex);
751 schedule_timeout(timeout);
753 mutex_lock(&root->fs_info->trans_mutex);
754 finish_wait(&cur_trans->writer_wait, &wait);
755 } while (cur_trans->num_writers > 1 ||
756 (cur_trans->num_joined != joined));
758 ret = create_pending_snapshots(trans, root->fs_info);
761 WARN_ON(cur_trans != trans->transaction);
763 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
767 ret = btrfs_commit_tree_roots(trans, root);
770 cur_trans = root->fs_info->running_transaction;
771 spin_lock(&root->fs_info->new_trans_lock);
772 root->fs_info->running_transaction = NULL;
773 spin_unlock(&root->fs_info->new_trans_lock);
774 btrfs_set_super_generation(&root->fs_info->super_copy,
776 btrfs_set_super_root(&root->fs_info->super_copy,
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));
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));
785 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
786 sizeof(root->fs_info->super_copy));
788 btrfs_copy_pinned(root, pinned_copy);
790 trans->transaction->blocked = 0;
791 wake_up(&root->fs_info->transaction_throttle);
792 wake_up(&root->fs_info->transaction_wait);
794 mutex_unlock(&root->fs_info->trans_mutex);
795 ret = btrfs_write_and_wait_transaction(trans, root);
797 write_ctree_super(trans, root);
799 btrfs_finish_extent_commit(trans, root, pinned_copy);
800 mutex_lock(&root->fs_info->trans_mutex);
804 cur_trans->commit_done = 1;
805 root->fs_info->last_trans_committed = cur_trans->transid;
806 wake_up(&cur_trans->commit_wait);
807 put_transaction(cur_trans);
808 put_transaction(cur_trans);
810 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
811 if (root->fs_info->closing)
812 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
814 mutex_unlock(&root->fs_info->trans_mutex);
815 kmem_cache_free(btrfs_trans_handle_cachep, trans);
817 if (root->fs_info->closing) {
818 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
823 int btrfs_clean_old_snapshots(struct btrfs_root *root)
825 struct list_head dirty_roots;
826 INIT_LIST_HEAD(&dirty_roots);
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);
832 if (!list_empty(&dirty_roots)) {
833 drop_dirty_roots(root, &dirty_roots);