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>
23 #include <linux/blkdev.h>
26 #include "transaction.h"
28 #include "ref-cache.h"
31 #define BTRFS_ROOT_TRANS_TAG 0
33 static noinline void put_transaction(struct btrfs_transaction *transaction)
35 WARN_ON(transaction->use_count == 0);
36 transaction->use_count--;
37 if (transaction->use_count == 0) {
38 list_del_init(&transaction->list);
39 memset(transaction, 0, sizeof(*transaction));
40 kmem_cache_free(btrfs_transaction_cachep, transaction);
45 * either allocate a new transaction or hop into the existing one
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,
55 root->fs_info->generation++;
56 root->fs_info->last_alloc = 0;
57 root->fs_info->last_data_alloc = 0;
58 cur_trans->num_writers = 1;
59 cur_trans->num_joined = 0;
60 cur_trans->transid = root->fs_info->generation;
61 init_waitqueue_head(&cur_trans->writer_wait);
62 init_waitqueue_head(&cur_trans->commit_wait);
63 cur_trans->in_commit = 0;
64 cur_trans->blocked = 0;
65 cur_trans->use_count = 1;
66 cur_trans->commit_done = 0;
67 cur_trans->start_time = get_seconds();
69 cur_trans->delayed_refs.root.rb_node = NULL;
70 cur_trans->delayed_refs.num_entries = 0;
71 cur_trans->delayed_refs.flushing = 0;
72 spin_lock_init(&cur_trans->delayed_refs.lock);
74 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
75 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
76 extent_io_tree_init(&cur_trans->dirty_pages,
77 root->fs_info->btree_inode->i_mapping,
79 spin_lock(&root->fs_info->new_trans_lock);
80 root->fs_info->running_transaction = cur_trans;
81 spin_unlock(&root->fs_info->new_trans_lock);
83 cur_trans->num_writers++;
84 cur_trans->num_joined++;
91 * this does all the record keeping required to make sure that a reference
92 * counted root is properly recorded in a given transaction. This is required
93 * to make sure the old root from before we joined the transaction is deleted
94 * when the transaction commits
96 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
98 struct btrfs_dirty_root *dirty;
99 u64 running_trans_id = root->fs_info->running_transaction->transid;
100 if (root->ref_cows && root->last_trans < running_trans_id) {
101 WARN_ON(root == root->fs_info->extent_root);
102 if (root->root_item.refs != 0) {
103 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
104 (unsigned long)root->root_key.objectid,
105 BTRFS_ROOT_TRANS_TAG);
107 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
109 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
110 BUG_ON(!dirty->root);
111 dirty->latest_root = root;
112 INIT_LIST_HEAD(&dirty->list);
114 root->commit_root = btrfs_root_node(root);
116 memcpy(dirty->root, root, sizeof(*root));
117 spin_lock_init(&dirty->root->node_lock);
118 spin_lock_init(&dirty->root->list_lock);
119 mutex_init(&dirty->root->objectid_mutex);
120 mutex_init(&dirty->root->log_mutex);
121 INIT_LIST_HEAD(&dirty->root->dead_list);
122 dirty->root->node = root->commit_root;
123 dirty->root->commit_root = NULL;
125 spin_lock(&root->list_lock);
126 list_add(&dirty->root->dead_list, &root->dead_list);
127 spin_unlock(&root->list_lock);
129 root->dirty_root = dirty;
133 root->last_trans = running_trans_id;
138 /* wait for commit against the current transaction to become unblocked
139 * when this is done, it is safe to start a new transaction, but the current
140 * transaction might not be fully on disk.
142 static void wait_current_trans(struct btrfs_root *root)
144 struct btrfs_transaction *cur_trans;
146 cur_trans = root->fs_info->running_transaction;
147 if (cur_trans && cur_trans->blocked) {
149 cur_trans->use_count++;
151 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
152 TASK_UNINTERRUPTIBLE);
153 if (cur_trans->blocked) {
154 mutex_unlock(&root->fs_info->trans_mutex);
156 mutex_lock(&root->fs_info->trans_mutex);
157 finish_wait(&root->fs_info->transaction_wait,
160 finish_wait(&root->fs_info->transaction_wait,
165 put_transaction(cur_trans);
169 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
170 int num_blocks, int wait)
172 struct btrfs_trans_handle *h =
173 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
176 mutex_lock(&root->fs_info->trans_mutex);
177 if (!root->fs_info->log_root_recovering &&
178 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
179 wait_current_trans(root);
180 ret = join_transaction(root);
183 btrfs_record_root_in_trans(root);
184 h->transid = root->fs_info->running_transaction->transid;
185 h->transaction = root->fs_info->running_transaction;
186 h->blocks_reserved = num_blocks;
189 h->alloc_exclude_nr = 0;
190 h->alloc_exclude_start = 0;
191 h->delayed_ref_updates = 0;
192 root->fs_info->running_transaction->use_count++;
193 mutex_unlock(&root->fs_info->trans_mutex);
197 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
200 return start_transaction(root, num_blocks, 1);
202 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
205 return start_transaction(root, num_blocks, 0);
208 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
211 return start_transaction(r, num_blocks, 2);
214 /* wait for a transaction commit to be fully complete */
215 static noinline int wait_for_commit(struct btrfs_root *root,
216 struct btrfs_transaction *commit)
219 mutex_lock(&root->fs_info->trans_mutex);
220 while (!commit->commit_done) {
221 prepare_to_wait(&commit->commit_wait, &wait,
222 TASK_UNINTERRUPTIBLE);
223 if (commit->commit_done)
225 mutex_unlock(&root->fs_info->trans_mutex);
227 mutex_lock(&root->fs_info->trans_mutex);
229 mutex_unlock(&root->fs_info->trans_mutex);
230 finish_wait(&commit->commit_wait, &wait);
235 * rate limit against the drop_snapshot code. This helps to slow down new
236 * operations if the drop_snapshot code isn't able to keep up.
238 static void throttle_on_drops(struct btrfs_root *root)
240 struct btrfs_fs_info *info = root->fs_info;
241 int harder_count = 0;
244 if (atomic_read(&info->throttles)) {
247 thr = atomic_read(&info->throttle_gen);
250 prepare_to_wait(&info->transaction_throttle,
251 &wait, TASK_UNINTERRUPTIBLE);
252 if (!atomic_read(&info->throttles)) {
253 finish_wait(&info->transaction_throttle, &wait);
257 finish_wait(&info->transaction_throttle, &wait);
258 } while (thr == atomic_read(&info->throttle_gen));
261 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
265 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
269 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
275 void btrfs_throttle(struct btrfs_root *root)
277 mutex_lock(&root->fs_info->trans_mutex);
278 if (!root->fs_info->open_ioctl_trans)
279 wait_current_trans(root);
280 mutex_unlock(&root->fs_info->trans_mutex);
282 throttle_on_drops(root);
285 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
286 struct btrfs_root *root, int throttle)
288 struct btrfs_transaction *cur_trans;
289 struct btrfs_fs_info *info = root->fs_info;
291 if (trans->delayed_ref_updates &&
292 (trans->transaction->delayed_refs.flushing ||
293 trans->transaction->delayed_refs.num_entries > 16384)) {
294 btrfs_run_delayed_refs(trans, root, trans->delayed_ref_updates);
295 } else if (trans->transaction->delayed_refs.num_entries > 64) {
296 wake_up_process(root->fs_info->transaction_kthread);
299 mutex_lock(&info->trans_mutex);
300 cur_trans = info->running_transaction;
301 WARN_ON(cur_trans != trans->transaction);
302 WARN_ON(cur_trans->num_writers < 1);
303 cur_trans->num_writers--;
305 if (waitqueue_active(&cur_trans->writer_wait))
306 wake_up(&cur_trans->writer_wait);
307 put_transaction(cur_trans);
308 mutex_unlock(&info->trans_mutex);
309 memset(trans, 0, sizeof(*trans));
310 kmem_cache_free(btrfs_trans_handle_cachep, trans);
313 throttle_on_drops(root);
318 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
319 struct btrfs_root *root)
321 return __btrfs_end_transaction(trans, root, 0);
324 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
325 struct btrfs_root *root)
327 return __btrfs_end_transaction(trans, root, 1);
331 * when btree blocks are allocated, they have some corresponding bits set for
332 * them in one of two extent_io trees. This is used to make sure all of
333 * those extents are on disk for transaction or log commit
335 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
336 struct extent_io_tree *dirty_pages)
342 struct inode *btree_inode = root->fs_info->btree_inode;
348 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
352 while (start <= end) {
355 index = start >> PAGE_CACHE_SHIFT;
356 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
357 page = find_get_page(btree_inode->i_mapping, index);
361 btree_lock_page_hook(page);
362 if (!page->mapping) {
364 page_cache_release(page);
368 if (PageWriteback(page)) {
370 wait_on_page_writeback(page);
373 page_cache_release(page);
377 err = write_one_page(page, 0);
380 page_cache_release(page);
384 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
389 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
390 while (start <= end) {
391 index = start >> PAGE_CACHE_SHIFT;
392 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
393 page = find_get_page(btree_inode->i_mapping, index);
396 if (PageDirty(page)) {
397 btree_lock_page_hook(page);
398 wait_on_page_writeback(page);
399 err = write_one_page(page, 0);
403 wait_on_page_writeback(page);
404 page_cache_release(page);
413 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root)
416 if (!trans || !trans->transaction) {
417 struct inode *btree_inode;
418 btree_inode = root->fs_info->btree_inode;
419 return filemap_write_and_wait(btree_inode->i_mapping);
421 return btrfs_write_and_wait_marked_extents(root,
422 &trans->transaction->dirty_pages);
426 * this is used to update the root pointer in the tree of tree roots.
428 * But, in the case of the extent allocation tree, updating the root
429 * pointer may allocate blocks which may change the root of the extent
432 * So, this loops and repeats and makes sure the cowonly root didn't
433 * change while the root pointer was being updated in the metadata.
435 static int update_cowonly_root(struct btrfs_trans_handle *trans,
436 struct btrfs_root *root)
440 struct btrfs_root *tree_root = root->fs_info->tree_root;
442 btrfs_write_dirty_block_groups(trans, root);
444 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
448 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
449 if (old_root_bytenr == root->node->start)
451 btrfs_set_root_bytenr(&root->root_item,
453 btrfs_set_root_level(&root->root_item,
454 btrfs_header_level(root->node));
455 btrfs_set_root_generation(&root->root_item, trans->transid);
457 ret = btrfs_update_root(trans, tree_root,
461 btrfs_write_dirty_block_groups(trans, root);
463 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
470 * update all the cowonly tree roots on disk
472 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
473 struct btrfs_root *root)
475 struct btrfs_fs_info *fs_info = root->fs_info;
476 struct list_head *next;
477 struct extent_buffer *eb;
480 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
483 eb = btrfs_lock_root_node(fs_info->tree_root);
484 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
485 btrfs_tree_unlock(eb);
486 free_extent_buffer(eb);
488 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
491 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
492 next = fs_info->dirty_cowonly_roots.next;
494 root = list_entry(next, struct btrfs_root, dirty_list);
496 update_cowonly_root(trans, root);
498 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
505 * dead roots are old snapshots that need to be deleted. This allocates
506 * a dirty root struct and adds it into the list of dead roots that need to
509 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
511 struct btrfs_dirty_root *dirty;
513 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
517 dirty->latest_root = latest;
519 mutex_lock(&root->fs_info->trans_mutex);
520 list_add(&dirty->list, &latest->fs_info->dead_roots);
521 mutex_unlock(&root->fs_info->trans_mutex);
526 * at transaction commit time we need to schedule the old roots for
527 * deletion via btrfs_drop_snapshot. This runs through all the
528 * reference counted roots that were modified in the current
529 * transaction and puts them into the drop list
531 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
532 struct radix_tree_root *radix,
533 struct list_head *list)
535 struct btrfs_dirty_root *dirty;
536 struct btrfs_root *gang[8];
537 struct btrfs_root *root;
544 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
546 BTRFS_ROOT_TRANS_TAG);
549 for (i = 0; i < ret; i++) {
551 radix_tree_tag_clear(radix,
552 (unsigned long)root->root_key.objectid,
553 BTRFS_ROOT_TRANS_TAG);
555 BUG_ON(!root->ref_tree);
556 dirty = root->dirty_root;
558 btrfs_free_log(trans, root);
559 btrfs_free_reloc_root(trans, root);
561 if (root->commit_root == root->node) {
562 WARN_ON(root->node->start !=
563 btrfs_root_bytenr(&root->root_item));
565 free_extent_buffer(root->commit_root);
566 root->commit_root = NULL;
567 root->dirty_root = NULL;
569 spin_lock(&root->list_lock);
570 list_del_init(&dirty->root->dead_list);
571 spin_unlock(&root->list_lock);
576 /* make sure to update the root on disk
577 * so we get any updates to the block used
580 err = btrfs_update_root(trans,
581 root->fs_info->tree_root,
587 memset(&root->root_item.drop_progress, 0,
588 sizeof(struct btrfs_disk_key));
589 root->root_item.drop_level = 0;
590 root->commit_root = NULL;
591 root->dirty_root = NULL;
592 root->root_key.offset = root->fs_info->generation;
593 btrfs_set_root_bytenr(&root->root_item,
595 btrfs_set_root_level(&root->root_item,
596 btrfs_header_level(root->node));
597 btrfs_set_root_generation(&root->root_item,
598 root->root_key.offset);
600 err = btrfs_insert_root(trans, root->fs_info->tree_root,
606 refs = btrfs_root_refs(&dirty->root->root_item);
607 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
608 err = btrfs_update_root(trans, root->fs_info->tree_root,
609 &dirty->root->root_key,
610 &dirty->root->root_item);
614 list_add(&dirty->list, list);
617 free_extent_buffer(dirty->root->node);
627 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
628 * otherwise every leaf in the btree is read and defragged.
630 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
632 struct btrfs_fs_info *info = root->fs_info;
634 struct btrfs_trans_handle *trans;
638 if (root->defrag_running)
640 trans = btrfs_start_transaction(root, 1);
642 root->defrag_running = 1;
643 ret = btrfs_defrag_leaves(trans, root, cacheonly);
644 nr = trans->blocks_used;
645 btrfs_end_transaction(trans, root);
646 btrfs_btree_balance_dirty(info->tree_root, nr);
649 trans = btrfs_start_transaction(root, 1);
650 if (root->fs_info->closing || ret != -EAGAIN)
653 root->defrag_running = 0;
655 btrfs_end_transaction(trans, root);
660 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
663 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
664 struct list_head *list)
666 struct btrfs_dirty_root *dirty;
667 struct btrfs_trans_handle *trans;
675 while (!list_empty(list)) {
676 struct btrfs_root *root;
678 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
679 list_del_init(&dirty->list);
681 num_bytes = btrfs_root_used(&dirty->root->root_item);
682 root = dirty->latest_root;
683 atomic_inc(&root->fs_info->throttles);
686 trans = btrfs_start_transaction(tree_root, 1);
687 mutex_lock(&root->fs_info->drop_mutex);
688 ret = btrfs_drop_snapshot(trans, dirty->root);
691 mutex_unlock(&root->fs_info->drop_mutex);
693 err = btrfs_update_root(trans,
695 &dirty->root->root_key,
696 &dirty->root->root_item);
699 nr = trans->blocks_used;
700 ret = btrfs_end_transaction(trans, tree_root);
703 btrfs_btree_balance_dirty(tree_root, nr);
707 atomic_dec(&root->fs_info->throttles);
708 wake_up(&root->fs_info->transaction_throttle);
710 num_bytes -= btrfs_root_used(&dirty->root->root_item);
711 bytes_used = btrfs_root_used(&root->root_item);
713 mutex_lock(&root->fs_info->trans_mutex);
714 btrfs_record_root_in_trans(root);
715 mutex_unlock(&root->fs_info->trans_mutex);
716 btrfs_set_root_used(&root->root_item,
717 bytes_used - num_bytes);
720 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
725 mutex_unlock(&root->fs_info->drop_mutex);
727 spin_lock(&root->list_lock);
728 list_del_init(&dirty->root->dead_list);
729 if (!list_empty(&root->dead_list)) {
730 struct btrfs_root *oldest;
731 oldest = list_entry(root->dead_list.prev,
732 struct btrfs_root, dead_list);
733 max_useless = oldest->root_key.offset - 1;
735 max_useless = root->root_key.offset - 1;
737 spin_unlock(&root->list_lock);
739 nr = trans->blocks_used;
740 ret = btrfs_end_transaction(trans, tree_root);
743 ret = btrfs_remove_leaf_refs(root, max_useless, 0);
746 free_extent_buffer(dirty->root->node);
750 btrfs_btree_balance_dirty(tree_root, nr);
757 * new snapshots need to be created at a very specific time in the
758 * transaction commit. This does the actual creation
760 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
761 struct btrfs_fs_info *fs_info,
762 struct btrfs_pending_snapshot *pending)
764 struct btrfs_key key;
765 struct btrfs_root_item *new_root_item;
766 struct btrfs_root *tree_root = fs_info->tree_root;
767 struct btrfs_root *root = pending->root;
768 struct extent_buffer *tmp;
769 struct extent_buffer *old;
773 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
774 if (!new_root_item) {
778 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
782 btrfs_record_root_in_trans(root);
783 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
784 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
786 key.objectid = objectid;
787 key.offset = trans->transid;
788 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
790 old = btrfs_lock_root_node(root);
791 btrfs_cow_block(trans, root, old, NULL, 0, &old);
793 btrfs_copy_root(trans, root, old, &tmp, objectid);
794 btrfs_tree_unlock(old);
795 free_extent_buffer(old);
797 btrfs_set_root_bytenr(new_root_item, tmp->start);
798 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
799 btrfs_set_root_generation(new_root_item, trans->transid);
800 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
802 btrfs_tree_unlock(tmp);
803 free_extent_buffer(tmp);
807 key.offset = (u64)-1;
808 memcpy(&pending->root_key, &key, sizeof(key));
810 kfree(new_root_item);
814 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
815 struct btrfs_pending_snapshot *pending)
820 struct btrfs_trans_handle *trans;
821 struct inode *parent_inode;
823 struct btrfs_root *parent_root;
825 parent_inode = pending->dentry->d_parent->d_inode;
826 parent_root = BTRFS_I(parent_inode)->root;
827 trans = btrfs_join_transaction(parent_root, 1);
830 * insert the directory item
832 namelen = strlen(pending->name);
833 ret = btrfs_set_inode_index(parent_inode, &index);
834 ret = btrfs_insert_dir_item(trans, parent_root,
835 pending->name, namelen,
837 &pending->root_key, BTRFS_FT_DIR, index);
842 btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
843 ret = btrfs_update_inode(trans, parent_root, parent_inode);
846 /* add the backref first */
847 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
848 pending->root_key.objectid,
849 BTRFS_ROOT_BACKREF_KEY,
850 parent_root->root_key.objectid,
851 parent_inode->i_ino, index, pending->name,
856 /* now add the forward ref */
857 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
858 parent_root->root_key.objectid,
860 pending->root_key.objectid,
861 parent_inode->i_ino, index, pending->name,
864 inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
865 d_instantiate(pending->dentry, inode);
867 btrfs_end_transaction(trans, fs_info->fs_root);
872 * create all the snapshots we've scheduled for creation
874 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
875 struct btrfs_fs_info *fs_info)
877 struct btrfs_pending_snapshot *pending;
878 struct list_head *head = &trans->transaction->pending_snapshots;
881 list_for_each_entry(pending, head, list) {
882 ret = create_pending_snapshot(trans, fs_info, pending);
888 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
889 struct btrfs_fs_info *fs_info)
891 struct btrfs_pending_snapshot *pending;
892 struct list_head *head = &trans->transaction->pending_snapshots;
895 while (!list_empty(head)) {
896 pending = list_entry(head->next,
897 struct btrfs_pending_snapshot, list);
898 ret = finish_pending_snapshot(fs_info, pending);
900 list_del(&pending->list);
901 kfree(pending->name);
907 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
908 struct btrfs_root *root)
910 unsigned long joined = 0;
911 unsigned long timeout = 1;
912 struct btrfs_transaction *cur_trans;
913 struct btrfs_transaction *prev_trans = NULL;
914 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
915 struct list_head dirty_fs_roots;
916 struct extent_io_tree *pinned_copy;
920 /* make a pass through all the delayed refs we have so far
921 * any runnings procs may add more while we are here
923 ret = btrfs_run_delayed_refs(trans, root, 0);
927 * set the flushing flag so procs in this transaction have to
928 * start sending their work down.
930 trans->transaction->delayed_refs.flushing = 1;
932 ret = btrfs_run_delayed_refs(trans, root, (u64)-1);
935 INIT_LIST_HEAD(&dirty_fs_roots);
936 mutex_lock(&root->fs_info->trans_mutex);
937 if (trans->transaction->in_commit) {
938 cur_trans = trans->transaction;
939 trans->transaction->use_count++;
940 mutex_unlock(&root->fs_info->trans_mutex);
941 btrfs_end_transaction(trans, root);
943 ret = wait_for_commit(root, cur_trans);
946 mutex_lock(&root->fs_info->trans_mutex);
947 put_transaction(cur_trans);
948 mutex_unlock(&root->fs_info->trans_mutex);
953 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
957 extent_io_tree_init(pinned_copy,
958 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
960 trans->transaction->in_commit = 1;
961 trans->transaction->blocked = 1;
962 cur_trans = trans->transaction;
963 if (cur_trans->list.prev != &root->fs_info->trans_list) {
964 prev_trans = list_entry(cur_trans->list.prev,
965 struct btrfs_transaction, list);
966 if (!prev_trans->commit_done) {
967 prev_trans->use_count++;
968 mutex_unlock(&root->fs_info->trans_mutex);
970 wait_for_commit(root, prev_trans);
972 mutex_lock(&root->fs_info->trans_mutex);
973 put_transaction(prev_trans);
978 int snap_pending = 0;
979 joined = cur_trans->num_joined;
980 if (!list_empty(&trans->transaction->pending_snapshots))
983 WARN_ON(cur_trans != trans->transaction);
984 prepare_to_wait(&cur_trans->writer_wait, &wait,
985 TASK_UNINTERRUPTIBLE);
987 if (cur_trans->num_writers > 1)
988 timeout = MAX_SCHEDULE_TIMEOUT;
992 mutex_unlock(&root->fs_info->trans_mutex);
995 ret = btrfs_wait_ordered_extents(root, 1);
999 schedule_timeout(timeout);
1001 mutex_lock(&root->fs_info->trans_mutex);
1002 finish_wait(&cur_trans->writer_wait, &wait);
1003 } while (cur_trans->num_writers > 1 ||
1004 (cur_trans->num_joined != joined));
1006 ret = create_pending_snapshots(trans, root->fs_info);
1009 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1012 WARN_ON(cur_trans != trans->transaction);
1014 /* btrfs_commit_tree_roots is responsible for getting the
1015 * various roots consistent with each other. Every pointer
1016 * in the tree of tree roots has to point to the most up to date
1017 * root for every subvolume and other tree. So, we have to keep
1018 * the tree logging code from jumping in and changing any
1021 * At this point in the commit, there can't be any tree-log
1022 * writers, but a little lower down we drop the trans mutex
1023 * and let new people in. By holding the tree_log_mutex
1024 * from now until after the super is written, we avoid races
1025 * with the tree-log code.
1027 mutex_lock(&root->fs_info->tree_log_mutex);
1029 * keep tree reloc code from adding new reloc trees
1031 mutex_lock(&root->fs_info->tree_reloc_mutex);
1034 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
1038 /* add_dirty_roots gets rid of all the tree log roots, it is now
1039 * safe to free the root of tree log roots
1041 btrfs_free_log_root_tree(trans, root->fs_info);
1043 ret = btrfs_commit_tree_roots(trans, root);
1046 cur_trans = root->fs_info->running_transaction;
1047 spin_lock(&root->fs_info->new_trans_lock);
1048 root->fs_info->running_transaction = NULL;
1049 spin_unlock(&root->fs_info->new_trans_lock);
1050 btrfs_set_super_generation(&root->fs_info->super_copy,
1051 cur_trans->transid);
1052 btrfs_set_super_root(&root->fs_info->super_copy,
1053 root->fs_info->tree_root->node->start);
1054 btrfs_set_super_root_level(&root->fs_info->super_copy,
1055 btrfs_header_level(root->fs_info->tree_root->node));
1057 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
1058 chunk_root->node->start);
1059 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
1060 btrfs_header_level(chunk_root->node));
1061 btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
1062 btrfs_header_generation(chunk_root->node));
1064 if (!root->fs_info->log_root_recovering) {
1065 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1066 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1069 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1070 sizeof(root->fs_info->super_copy));
1072 btrfs_copy_pinned(root, pinned_copy);
1074 trans->transaction->blocked = 0;
1075 wake_up(&root->fs_info->transaction_throttle);
1076 wake_up(&root->fs_info->transaction_wait);
1078 mutex_unlock(&root->fs_info->trans_mutex);
1079 ret = btrfs_write_and_wait_transaction(trans, root);
1081 write_ctree_super(trans, root, 0);
1084 * the super is written, we can safely allow the tree-loggers
1085 * to go about their business
1087 mutex_unlock(&root->fs_info->tree_log_mutex);
1089 btrfs_finish_extent_commit(trans, root, pinned_copy);
1092 btrfs_drop_dead_reloc_roots(root);
1093 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1095 /* do the directory inserts of any pending snapshot creations */
1096 finish_pending_snapshots(trans, root->fs_info);
1098 mutex_lock(&root->fs_info->trans_mutex);
1100 cur_trans->commit_done = 1;
1101 root->fs_info->last_trans_committed = cur_trans->transid;
1102 wake_up(&cur_trans->commit_wait);
1104 put_transaction(cur_trans);
1105 put_transaction(cur_trans);
1107 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
1108 if (root->fs_info->closing)
1109 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
1111 mutex_unlock(&root->fs_info->trans_mutex);
1113 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1115 if (root->fs_info->closing)
1116 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
1121 * interface function to delete all the snapshots we have scheduled for deletion
1123 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1125 struct list_head dirty_roots;
1126 INIT_LIST_HEAD(&dirty_roots);
1128 mutex_lock(&root->fs_info->trans_mutex);
1129 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
1130 mutex_unlock(&root->fs_info->trans_mutex);
1132 if (!list_empty(&dirty_roots)) {
1133 drop_dirty_roots(root, &dirty_roots);