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 static void wait_current_trans(struct btrfs_root *root)
129 struct btrfs_transaction *cur_trans;
131 cur_trans = root->fs_info->running_transaction;
132 if (cur_trans && cur_trans->blocked) {
134 cur_trans->use_count++;
136 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
137 TASK_UNINTERRUPTIBLE);
138 if (cur_trans->blocked) {
139 mutex_unlock(&root->fs_info->trans_mutex);
141 mutex_lock(&root->fs_info->trans_mutex);
142 finish_wait(&root->fs_info->transaction_wait,
145 finish_wait(&root->fs_info->transaction_wait,
150 put_transaction(cur_trans);
154 struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
155 int num_blocks, int join)
157 struct btrfs_trans_handle *h =
158 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
161 mutex_lock(&root->fs_info->trans_mutex);
163 wait_current_trans(root);
164 ret = join_transaction(root);
167 record_root_in_trans(root);
168 h->transid = root->fs_info->running_transaction->transid;
169 h->transaction = root->fs_info->running_transaction;
170 h->blocks_reserved = num_blocks;
172 h->block_group = NULL;
173 h->alloc_exclude_nr = 0;
174 h->alloc_exclude_start = 0;
175 root->fs_info->running_transaction->use_count++;
176 mutex_unlock(&root->fs_info->trans_mutex);
180 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
183 return start_transaction(root, num_blocks, 0);
185 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
188 return start_transaction(root, num_blocks, 1);
191 static noinline int wait_for_commit(struct btrfs_root *root,
192 struct btrfs_transaction *commit)
195 mutex_lock(&root->fs_info->trans_mutex);
196 while(!commit->commit_done) {
197 prepare_to_wait(&commit->commit_wait, &wait,
198 TASK_UNINTERRUPTIBLE);
199 if (commit->commit_done)
201 mutex_unlock(&root->fs_info->trans_mutex);
203 mutex_lock(&root->fs_info->trans_mutex);
205 mutex_unlock(&root->fs_info->trans_mutex);
206 finish_wait(&commit->commit_wait, &wait);
210 static void throttle_on_drops(struct btrfs_root *root)
212 struct btrfs_fs_info *info = root->fs_info;
214 if (atomic_read(&info->throttles)) {
217 thr = atomic_read(&info->throttle_gen);
220 prepare_to_wait(&info->transaction_throttle,
221 &wait, TASK_UNINTERRUPTIBLE);
222 if (!atomic_read(&info->throttles)) {
223 finish_wait(&info->transaction_throttle, &wait);
227 finish_wait(&info->transaction_throttle, &wait);
228 } while (thr == atomic_read(&info->throttle_gen));
232 void btrfs_throttle(struct btrfs_root *root)
234 mutex_lock(&root->fs_info->trans_mutex);
235 wait_current_trans(root);
236 mutex_unlock(&root->fs_info->trans_mutex);
238 throttle_on_drops(root);
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 throttle_on_drops(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);
369 int btrfs_add_dead_root(struct btrfs_root *root,
370 struct btrfs_root *latest,
371 struct list_head *dead_list)
373 struct btrfs_dirty_root *dirty;
375 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
379 dirty->latest_root = latest;
380 list_add(&dirty->list, dead_list);
384 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
385 struct radix_tree_root *radix,
386 struct list_head *list)
388 struct btrfs_dirty_root *dirty;
389 struct btrfs_root *gang[8];
390 struct btrfs_root *root;
397 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
399 BTRFS_ROOT_TRANS_TAG);
402 for (i = 0; i < ret; i++) {
404 radix_tree_tag_clear(radix,
405 (unsigned long)root->root_key.objectid,
406 BTRFS_ROOT_TRANS_TAG);
408 BUG_ON(!root->ref_tree);
409 dirty = root->dirty_root;
411 if (root->commit_root == root->node) {
412 WARN_ON(root->node->start !=
413 btrfs_root_bytenr(&root->root_item));
415 free_extent_buffer(root->commit_root);
416 root->commit_root = NULL;
418 spin_lock(&root->list_lock);
419 list_del_init(&dirty->root->dead_list);
420 spin_unlock(&root->list_lock);
425 /* make sure to update the root on disk
426 * so we get any updates to the block used
429 err = btrfs_update_root(trans,
430 root->fs_info->tree_root,
436 memset(&root->root_item.drop_progress, 0,
437 sizeof(struct btrfs_disk_key));
438 root->root_item.drop_level = 0;
439 root->commit_root = NULL;
440 root->root_key.offset = root->fs_info->generation;
441 btrfs_set_root_bytenr(&root->root_item,
443 btrfs_set_root_level(&root->root_item,
444 btrfs_header_level(root->node));
445 err = btrfs_insert_root(trans, root->fs_info->tree_root,
451 refs = btrfs_root_refs(&dirty->root->root_item);
452 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
453 err = btrfs_update_root(trans, root->fs_info->tree_root,
454 &dirty->root->root_key,
455 &dirty->root->root_item);
459 list_add(&dirty->list, list);
462 free_extent_buffer(dirty->root->node);
471 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
473 struct btrfs_fs_info *info = root->fs_info;
475 struct btrfs_trans_handle *trans;
479 if (root->defrag_running)
481 trans = btrfs_start_transaction(root, 1);
483 root->defrag_running = 1;
484 ret = btrfs_defrag_leaves(trans, root, cacheonly);
485 nr = trans->blocks_used;
486 btrfs_end_transaction(trans, root);
487 btrfs_btree_balance_dirty(info->tree_root, nr);
490 trans = btrfs_start_transaction(root, 1);
491 if (root->fs_info->closing || ret != -EAGAIN)
494 root->defrag_running = 0;
496 btrfs_end_transaction(trans, root);
500 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
501 struct list_head *list)
503 struct btrfs_dirty_root *dirty;
504 struct btrfs_trans_handle *trans;
512 while(!list_empty(list)) {
513 struct btrfs_root *root;
515 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
516 list_del_init(&dirty->list);
518 num_bytes = btrfs_root_used(&dirty->root->root_item);
519 root = dirty->latest_root;
520 atomic_inc(&root->fs_info->throttles);
522 mutex_lock(&root->fs_info->drop_mutex);
524 trans = btrfs_start_transaction(tree_root, 1);
525 ret = btrfs_drop_snapshot(trans, dirty->root);
526 if (ret != -EAGAIN) {
530 err = btrfs_update_root(trans,
532 &dirty->root->root_key,
533 &dirty->root->root_item);
536 nr = trans->blocks_used;
537 ret = btrfs_end_transaction(trans, tree_root);
540 mutex_unlock(&root->fs_info->drop_mutex);
541 btrfs_btree_balance_dirty(tree_root, nr);
543 mutex_lock(&root->fs_info->drop_mutex);
546 atomic_dec(&root->fs_info->throttles);
547 wake_up(&root->fs_info->transaction_throttle);
549 mutex_lock(&root->fs_info->alloc_mutex);
550 num_bytes -= btrfs_root_used(&dirty->root->root_item);
551 bytes_used = btrfs_root_used(&root->root_item);
553 record_root_in_trans(root);
554 btrfs_set_root_used(&root->root_item,
555 bytes_used - num_bytes);
557 mutex_unlock(&root->fs_info->alloc_mutex);
559 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
564 mutex_unlock(&root->fs_info->drop_mutex);
566 spin_lock(&root->list_lock);
567 list_del_init(&dirty->root->dead_list);
568 if (!list_empty(&root->dead_list)) {
569 struct btrfs_root *oldest;
570 oldest = list_entry(root->dead_list.prev,
571 struct btrfs_root, dead_list);
572 max_useless = oldest->root_key.offset - 1;
574 max_useless = root->root_key.offset - 1;
576 spin_unlock(&root->list_lock);
578 nr = trans->blocks_used;
579 ret = btrfs_end_transaction(trans, tree_root);
582 ret = btrfs_remove_leaf_refs(root, max_useless);
585 free_extent_buffer(dirty->root->node);
589 btrfs_btree_balance_dirty(tree_root, nr);
595 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
596 struct btrfs_fs_info *fs_info,
597 struct btrfs_pending_snapshot *pending)
599 struct btrfs_key key;
600 struct btrfs_root_item *new_root_item;
601 struct btrfs_root *tree_root = fs_info->tree_root;
602 struct btrfs_root *root = pending->root;
603 struct extent_buffer *tmp;
604 struct extent_buffer *old;
609 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
610 if (!new_root_item) {
614 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
618 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
620 key.objectid = objectid;
622 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
624 old = btrfs_lock_root_node(root);
625 btrfs_cow_block(trans, root, old, NULL, 0, &old);
627 btrfs_copy_root(trans, root, old, &tmp, objectid);
628 btrfs_tree_unlock(old);
629 free_extent_buffer(old);
631 btrfs_set_root_bytenr(new_root_item, tmp->start);
632 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
633 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
635 btrfs_tree_unlock(tmp);
636 free_extent_buffer(tmp);
641 * insert the directory item
643 key.offset = (u64)-1;
644 namelen = strlen(pending->name);
645 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
646 pending->name, namelen,
647 root->fs_info->sb->s_root->d_inode->i_ino,
648 &key, BTRFS_FT_DIR, 0);
653 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
654 pending->name, strlen(pending->name), objectid,
655 root->fs_info->sb->s_root->d_inode->i_ino, 0);
657 /* Invalidate existing dcache entry for new snapshot. */
658 btrfs_invalidate_dcache_root(root, pending->name, namelen);
661 kfree(new_root_item);
665 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
666 struct btrfs_fs_info *fs_info)
668 struct btrfs_pending_snapshot *pending;
669 struct list_head *head = &trans->transaction->pending_snapshots;
672 while(!list_empty(head)) {
673 pending = list_entry(head->next,
674 struct btrfs_pending_snapshot, list);
675 ret = create_pending_snapshot(trans, fs_info, pending);
677 list_del(&pending->list);
678 kfree(pending->name);
684 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
685 struct btrfs_root *root)
687 unsigned long joined = 0;
688 unsigned long timeout = 1;
689 struct btrfs_transaction *cur_trans;
690 struct btrfs_transaction *prev_trans = NULL;
691 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
692 struct list_head dirty_fs_roots;
693 struct extent_io_tree *pinned_copy;
697 INIT_LIST_HEAD(&dirty_fs_roots);
699 mutex_lock(&root->fs_info->trans_mutex);
700 if (trans->transaction->in_commit) {
701 cur_trans = trans->transaction;
702 trans->transaction->use_count++;
703 mutex_unlock(&root->fs_info->trans_mutex);
704 btrfs_end_transaction(trans, root);
706 ret = wait_for_commit(root, cur_trans);
709 mutex_lock(&root->fs_info->trans_mutex);
710 put_transaction(cur_trans);
711 mutex_unlock(&root->fs_info->trans_mutex);
716 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
720 extent_io_tree_init(pinned_copy,
721 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
723 trans->transaction->in_commit = 1;
724 trans->transaction->blocked = 1;
725 cur_trans = trans->transaction;
726 if (cur_trans->list.prev != &root->fs_info->trans_list) {
727 prev_trans = list_entry(cur_trans->list.prev,
728 struct btrfs_transaction, list);
729 if (!prev_trans->commit_done) {
730 prev_trans->use_count++;
731 mutex_unlock(&root->fs_info->trans_mutex);
733 wait_for_commit(root, prev_trans);
735 mutex_lock(&root->fs_info->trans_mutex);
736 put_transaction(prev_trans);
741 joined = cur_trans->num_joined;
742 WARN_ON(cur_trans != trans->transaction);
743 prepare_to_wait(&cur_trans->writer_wait, &wait,
744 TASK_UNINTERRUPTIBLE);
746 if (cur_trans->num_writers > 1)
747 timeout = MAX_SCHEDULE_TIMEOUT;
751 mutex_unlock(&root->fs_info->trans_mutex);
753 schedule_timeout(timeout);
755 mutex_lock(&root->fs_info->trans_mutex);
756 finish_wait(&cur_trans->writer_wait, &wait);
757 } while (cur_trans->num_writers > 1 ||
758 (cur_trans->num_joined != joined));
760 ret = create_pending_snapshots(trans, root->fs_info);
763 WARN_ON(cur_trans != trans->transaction);
765 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
769 ret = btrfs_commit_tree_roots(trans, root);
772 cur_trans = root->fs_info->running_transaction;
773 spin_lock(&root->fs_info->new_trans_lock);
774 root->fs_info->running_transaction = NULL;
775 spin_unlock(&root->fs_info->new_trans_lock);
776 btrfs_set_super_generation(&root->fs_info->super_copy,
778 btrfs_set_super_root(&root->fs_info->super_copy,
779 root->fs_info->tree_root->node->start);
780 btrfs_set_super_root_level(&root->fs_info->super_copy,
781 btrfs_header_level(root->fs_info->tree_root->node));
783 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
784 chunk_root->node->start);
785 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
786 btrfs_header_level(chunk_root->node));
787 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
788 sizeof(root->fs_info->super_copy));
790 btrfs_copy_pinned(root, pinned_copy);
792 trans->transaction->blocked = 0;
793 wake_up(&root->fs_info->transaction_throttle);
794 wake_up(&root->fs_info->transaction_wait);
796 mutex_unlock(&root->fs_info->trans_mutex);
797 ret = btrfs_write_and_wait_transaction(trans, root);
799 write_ctree_super(trans, root);
801 btrfs_finish_extent_commit(trans, root, pinned_copy);
802 mutex_lock(&root->fs_info->trans_mutex);
806 cur_trans->commit_done = 1;
807 root->fs_info->last_trans_committed = cur_trans->transid;
808 wake_up(&cur_trans->commit_wait);
809 put_transaction(cur_trans);
810 put_transaction(cur_trans);
812 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
813 if (root->fs_info->closing)
814 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
816 mutex_unlock(&root->fs_info->trans_mutex);
817 kmem_cache_free(btrfs_trans_handle_cachep, trans);
819 if (root->fs_info->closing) {
820 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
825 int btrfs_clean_old_snapshots(struct btrfs_root *root)
827 struct list_head dirty_roots;
828 INIT_LIST_HEAD(&dirty_roots);
830 mutex_lock(&root->fs_info->trans_mutex);
831 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
832 mutex_unlock(&root->fs_info->trans_mutex);
834 if (!list_empty(&dirty_roots)) {
835 drop_dirty_roots(root, &dirty_roots);