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"
30 #define BTRFS_ROOT_TRANS_TAG 0
32 static noinline void put_transaction(struct btrfs_transaction *transaction)
34 WARN_ON(transaction->use_count == 0);
35 transaction->use_count--;
36 if (transaction->use_count == 0) {
37 list_del_init(&transaction->list);
38 memset(transaction, 0, sizeof(*transaction));
39 kmem_cache_free(btrfs_transaction_cachep, transaction);
43 static noinline void switch_commit_root(struct btrfs_root *root)
45 free_extent_buffer(root->commit_root);
46 root->commit_root = btrfs_root_node(root);
50 * either allocate a new transaction or hop into the existing one
52 static noinline int join_transaction(struct btrfs_root *root)
54 struct btrfs_transaction *cur_trans;
55 cur_trans = root->fs_info->running_transaction;
57 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
60 root->fs_info->generation++;
61 cur_trans->num_writers = 1;
62 cur_trans->num_joined = 0;
63 cur_trans->transid = root->fs_info->generation;
64 init_waitqueue_head(&cur_trans->writer_wait);
65 init_waitqueue_head(&cur_trans->commit_wait);
66 cur_trans->in_commit = 0;
67 cur_trans->blocked = 0;
68 cur_trans->use_count = 1;
69 cur_trans->commit_done = 0;
70 cur_trans->start_time = get_seconds();
72 cur_trans->delayed_refs.root.rb_node = NULL;
73 cur_trans->delayed_refs.num_entries = 0;
74 cur_trans->delayed_refs.num_heads_ready = 0;
75 cur_trans->delayed_refs.num_heads = 0;
76 cur_trans->delayed_refs.flushing = 0;
77 cur_trans->delayed_refs.run_delayed_start = 0;
78 spin_lock_init(&cur_trans->delayed_refs.lock);
80 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
81 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
82 extent_io_tree_init(&cur_trans->dirty_pages,
83 root->fs_info->btree_inode->i_mapping,
85 spin_lock(&root->fs_info->new_trans_lock);
86 root->fs_info->running_transaction = cur_trans;
87 spin_unlock(&root->fs_info->new_trans_lock);
89 cur_trans->num_writers++;
90 cur_trans->num_joined++;
97 * this does all the record keeping required to make sure that a reference
98 * counted root is properly recorded in a given transaction. This is required
99 * to make sure the old root from before we joined the transaction is deleted
100 * when the transaction commits
102 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root)
105 if (root->ref_cows && root->last_trans < trans->transid) {
106 WARN_ON(root == root->fs_info->extent_root);
107 WARN_ON(root->commit_root != root->node);
109 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
110 (unsigned long)root->root_key.objectid,
111 BTRFS_ROOT_TRANS_TAG);
112 root->last_trans = trans->transid;
113 btrfs_init_reloc_root(trans, root);
118 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
119 struct btrfs_root *root)
124 mutex_lock(&root->fs_info->trans_mutex);
125 if (root->last_trans == trans->transid) {
126 mutex_unlock(&root->fs_info->trans_mutex);
130 record_root_in_trans(trans, root);
131 mutex_unlock(&root->fs_info->trans_mutex);
135 /* wait for commit against the current transaction to become unblocked
136 * when this is done, it is safe to start a new transaction, but the current
137 * transaction might not be fully on disk.
139 static void wait_current_trans(struct btrfs_root *root)
141 struct btrfs_transaction *cur_trans;
143 cur_trans = root->fs_info->running_transaction;
144 if (cur_trans && cur_trans->blocked) {
146 cur_trans->use_count++;
148 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
149 TASK_UNINTERRUPTIBLE);
150 if (cur_trans->blocked) {
151 mutex_unlock(&root->fs_info->trans_mutex);
153 mutex_lock(&root->fs_info->trans_mutex);
154 finish_wait(&root->fs_info->transaction_wait,
157 finish_wait(&root->fs_info->transaction_wait,
162 put_transaction(cur_trans);
166 enum btrfs_trans_type {
172 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
173 int num_blocks, int type)
175 struct btrfs_trans_handle *h =
176 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
179 mutex_lock(&root->fs_info->trans_mutex);
180 if (!root->fs_info->log_root_recovering &&
181 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
182 type == TRANS_USERSPACE))
183 wait_current_trans(root);
184 ret = join_transaction(root);
187 h->transid = root->fs_info->running_transaction->transid;
188 h->transaction = root->fs_info->running_transaction;
189 h->blocks_reserved = num_blocks;
192 h->alloc_exclude_nr = 0;
193 h->alloc_exclude_start = 0;
194 h->delayed_ref_updates = 0;
196 if (!current->journal_info && type != TRANS_USERSPACE)
197 current->journal_info = h;
199 root->fs_info->running_transaction->use_count++;
200 record_root_in_trans(h, root);
201 mutex_unlock(&root->fs_info->trans_mutex);
205 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
208 return start_transaction(root, num_blocks, TRANS_START);
210 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
213 return start_transaction(root, num_blocks, TRANS_JOIN);
216 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
219 return start_transaction(r, num_blocks, TRANS_USERSPACE);
222 /* wait for a transaction commit to be fully complete */
223 static noinline int wait_for_commit(struct btrfs_root *root,
224 struct btrfs_transaction *commit)
227 mutex_lock(&root->fs_info->trans_mutex);
228 while (!commit->commit_done) {
229 prepare_to_wait(&commit->commit_wait, &wait,
230 TASK_UNINTERRUPTIBLE);
231 if (commit->commit_done)
233 mutex_unlock(&root->fs_info->trans_mutex);
235 mutex_lock(&root->fs_info->trans_mutex);
237 mutex_unlock(&root->fs_info->trans_mutex);
238 finish_wait(&commit->commit_wait, &wait);
244 * rate limit against the drop_snapshot code. This helps to slow down new
245 * operations if the drop_snapshot code isn't able to keep up.
247 static void throttle_on_drops(struct btrfs_root *root)
249 struct btrfs_fs_info *info = root->fs_info;
250 int harder_count = 0;
253 if (atomic_read(&info->throttles)) {
256 thr = atomic_read(&info->throttle_gen);
259 prepare_to_wait(&info->transaction_throttle,
260 &wait, TASK_UNINTERRUPTIBLE);
261 if (!atomic_read(&info->throttles)) {
262 finish_wait(&info->transaction_throttle, &wait);
266 finish_wait(&info->transaction_throttle, &wait);
267 } while (thr == atomic_read(&info->throttle_gen));
270 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
274 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
278 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
285 void btrfs_throttle(struct btrfs_root *root)
287 mutex_lock(&root->fs_info->trans_mutex);
288 if (!root->fs_info->open_ioctl_trans)
289 wait_current_trans(root);
290 mutex_unlock(&root->fs_info->trans_mutex);
293 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
294 struct btrfs_root *root, int throttle)
296 struct btrfs_transaction *cur_trans;
297 struct btrfs_fs_info *info = root->fs_info;
301 unsigned long cur = trans->delayed_ref_updates;
302 trans->delayed_ref_updates = 0;
304 trans->transaction->delayed_refs.num_heads_ready > 64) {
305 trans->delayed_ref_updates = 0;
308 * do a full flush if the transaction is trying
311 if (trans->transaction->delayed_refs.flushing)
313 btrfs_run_delayed_refs(trans, root, cur);
320 mutex_lock(&info->trans_mutex);
321 cur_trans = info->running_transaction;
322 WARN_ON(cur_trans != trans->transaction);
323 WARN_ON(cur_trans->num_writers < 1);
324 cur_trans->num_writers--;
326 if (waitqueue_active(&cur_trans->writer_wait))
327 wake_up(&cur_trans->writer_wait);
328 put_transaction(cur_trans);
329 mutex_unlock(&info->trans_mutex);
331 if (current->journal_info == trans)
332 current->journal_info = NULL;
333 memset(trans, 0, sizeof(*trans));
334 kmem_cache_free(btrfs_trans_handle_cachep, trans);
337 btrfs_run_delayed_iputs(root);
342 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
343 struct btrfs_root *root)
345 return __btrfs_end_transaction(trans, root, 0);
348 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
349 struct btrfs_root *root)
351 return __btrfs_end_transaction(trans, root, 1);
355 * when btree blocks are allocated, they have some corresponding bits set for
356 * them in one of two extent_io trees. This is used to make sure all of
357 * those extents are sent to disk but does not wait on them
359 int btrfs_write_marked_extents(struct btrfs_root *root,
360 struct extent_io_tree *dirty_pages, int mark)
366 struct inode *btree_inode = root->fs_info->btree_inode;
372 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
376 while (start <= end) {
379 index = start >> PAGE_CACHE_SHIFT;
380 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
381 page = find_get_page(btree_inode->i_mapping, index);
385 btree_lock_page_hook(page);
386 if (!page->mapping) {
388 page_cache_release(page);
392 if (PageWriteback(page)) {
394 wait_on_page_writeback(page);
397 page_cache_release(page);
401 err = write_one_page(page, 0);
404 page_cache_release(page);
413 * when btree blocks are allocated, they have some corresponding bits set for
414 * them in one of two extent_io trees. This is used to make sure all of
415 * those extents are on disk for transaction or log commit. We wait
416 * on all the pages and clear them from the dirty pages state tree
418 int btrfs_wait_marked_extents(struct btrfs_root *root,
419 struct extent_io_tree *dirty_pages, int mark)
425 struct inode *btree_inode = root->fs_info->btree_inode;
431 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
436 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
437 while (start <= end) {
438 index = start >> PAGE_CACHE_SHIFT;
439 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
440 page = find_get_page(btree_inode->i_mapping, index);
443 if (PageDirty(page)) {
444 btree_lock_page_hook(page);
445 wait_on_page_writeback(page);
446 err = write_one_page(page, 0);
450 wait_on_page_writeback(page);
451 page_cache_release(page);
461 * when btree blocks are allocated, they have some corresponding bits set for
462 * them in one of two extent_io trees. This is used to make sure all of
463 * those extents are on disk for transaction or log commit
465 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
466 struct extent_io_tree *dirty_pages, int mark)
471 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
472 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
476 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
477 struct btrfs_root *root)
479 if (!trans || !trans->transaction) {
480 struct inode *btree_inode;
481 btree_inode = root->fs_info->btree_inode;
482 return filemap_write_and_wait(btree_inode->i_mapping);
484 return btrfs_write_and_wait_marked_extents(root,
485 &trans->transaction->dirty_pages,
490 * this is used to update the root pointer in the tree of tree roots.
492 * But, in the case of the extent allocation tree, updating the root
493 * pointer may allocate blocks which may change the root of the extent
496 * So, this loops and repeats and makes sure the cowonly root didn't
497 * change while the root pointer was being updated in the metadata.
499 static int update_cowonly_root(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root)
505 struct btrfs_root *tree_root = root->fs_info->tree_root;
507 old_root_used = btrfs_root_used(&root->root_item);
508 btrfs_write_dirty_block_groups(trans, root);
511 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
512 if (old_root_bytenr == root->node->start &&
513 old_root_used == btrfs_root_used(&root->root_item))
516 btrfs_set_root_node(&root->root_item, root->node);
517 ret = btrfs_update_root(trans, tree_root,
522 old_root_used = btrfs_root_used(&root->root_item);
523 ret = btrfs_write_dirty_block_groups(trans, root);
527 if (root != root->fs_info->extent_root)
528 switch_commit_root(root);
534 * update all the cowonly tree roots on disk
536 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
537 struct btrfs_root *root)
539 struct btrfs_fs_info *fs_info = root->fs_info;
540 struct list_head *next;
541 struct extent_buffer *eb;
544 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
547 eb = btrfs_lock_root_node(fs_info->tree_root);
548 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
549 btrfs_tree_unlock(eb);
550 free_extent_buffer(eb);
552 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
555 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
556 next = fs_info->dirty_cowonly_roots.next;
558 root = list_entry(next, struct btrfs_root, dirty_list);
560 update_cowonly_root(trans, root);
563 down_write(&fs_info->extent_commit_sem);
564 switch_commit_root(fs_info->extent_root);
565 up_write(&fs_info->extent_commit_sem);
571 * dead roots are old snapshots that need to be deleted. This allocates
572 * a dirty root struct and adds it into the list of dead roots that need to
575 int btrfs_add_dead_root(struct btrfs_root *root)
577 mutex_lock(&root->fs_info->trans_mutex);
578 list_add(&root->root_list, &root->fs_info->dead_roots);
579 mutex_unlock(&root->fs_info->trans_mutex);
584 * update all the cowonly tree roots on disk
586 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
587 struct btrfs_root *root)
589 struct btrfs_root *gang[8];
590 struct btrfs_fs_info *fs_info = root->fs_info;
596 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
599 BTRFS_ROOT_TRANS_TAG);
602 for (i = 0; i < ret; i++) {
604 radix_tree_tag_clear(&fs_info->fs_roots_radix,
605 (unsigned long)root->root_key.objectid,
606 BTRFS_ROOT_TRANS_TAG);
608 btrfs_free_log(trans, root);
609 btrfs_update_reloc_root(trans, root);
611 if (root->commit_root != root->node) {
612 switch_commit_root(root);
613 btrfs_set_root_node(&root->root_item,
617 err = btrfs_update_root(trans, fs_info->tree_root,
628 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
629 * otherwise every leaf in the btree is read and defragged.
631 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
633 struct btrfs_fs_info *info = root->fs_info;
635 struct btrfs_trans_handle *trans;
639 if (root->defrag_running)
641 trans = btrfs_start_transaction(root, 1);
643 root->defrag_running = 1;
644 ret = btrfs_defrag_leaves(trans, root, cacheonly);
645 nr = trans->blocks_used;
646 btrfs_end_transaction(trans, root);
647 btrfs_btree_balance_dirty(info->tree_root, nr);
650 trans = btrfs_start_transaction(root, 1);
651 if (root->fs_info->closing || ret != -EAGAIN)
654 root->defrag_running = 0;
656 btrfs_end_transaction(trans, root);
662 * when dropping snapshots, we generate a ton of delayed refs, and it makes
663 * sense not to join the transaction while it is trying to flush the current
664 * queue of delayed refs out.
666 * This is used by the drop snapshot code only
668 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
672 mutex_lock(&info->trans_mutex);
673 while (info->running_transaction &&
674 info->running_transaction->delayed_refs.flushing) {
675 prepare_to_wait(&info->transaction_wait, &wait,
676 TASK_UNINTERRUPTIBLE);
677 mutex_unlock(&info->trans_mutex);
681 mutex_lock(&info->trans_mutex);
682 finish_wait(&info->transaction_wait, &wait);
684 mutex_unlock(&info->trans_mutex);
689 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
692 int btrfs_drop_dead_root(struct btrfs_root *root)
694 struct btrfs_trans_handle *trans;
695 struct btrfs_root *tree_root = root->fs_info->tree_root;
701 * we don't want to jump in and create a bunch of
702 * delayed refs if the transaction is starting to close
704 wait_transaction_pre_flush(tree_root->fs_info);
705 trans = btrfs_start_transaction(tree_root, 1);
708 * we've joined a transaction, make sure it isn't
711 if (trans->transaction->delayed_refs.flushing) {
712 btrfs_end_transaction(trans, tree_root);
716 ret = btrfs_drop_snapshot(trans, root);
720 ret = btrfs_update_root(trans, tree_root,
726 nr = trans->blocks_used;
727 ret = btrfs_end_transaction(trans, tree_root);
730 btrfs_btree_balance_dirty(tree_root, nr);
735 ret = btrfs_del_root(trans, tree_root, &root->root_key);
738 nr = trans->blocks_used;
739 ret = btrfs_end_transaction(trans, tree_root);
742 free_extent_buffer(root->node);
743 free_extent_buffer(root->commit_root);
746 btrfs_btree_balance_dirty(tree_root, nr);
752 * new snapshots need to be created at a very specific time in the
753 * transaction commit. This does the actual creation
755 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
756 struct btrfs_fs_info *fs_info,
757 struct btrfs_pending_snapshot *pending)
759 struct btrfs_key key;
760 struct btrfs_root_item *new_root_item;
761 struct btrfs_root *tree_root = fs_info->tree_root;
762 struct btrfs_root *root = pending->root;
763 struct extent_buffer *tmp;
764 struct extent_buffer *old;
768 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
769 if (!new_root_item) {
773 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
777 record_root_in_trans(trans, root);
778 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
779 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
781 key.objectid = objectid;
782 /* record when the snapshot was created in key.offset */
783 key.offset = trans->transid;
784 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
786 old = btrfs_lock_root_node(root);
787 btrfs_cow_block(trans, root, old, NULL, 0, &old);
788 btrfs_set_lock_blocking(old);
790 btrfs_copy_root(trans, root, old, &tmp, objectid);
791 btrfs_tree_unlock(old);
792 free_extent_buffer(old);
794 btrfs_set_root_node(new_root_item, tmp);
795 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
797 btrfs_tree_unlock(tmp);
798 free_extent_buffer(tmp);
802 key.offset = (u64)-1;
803 memcpy(&pending->root_key, &key, sizeof(key));
805 kfree(new_root_item);
809 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
810 struct btrfs_pending_snapshot *pending)
815 struct btrfs_trans_handle *trans;
816 struct inode *parent_inode;
817 struct btrfs_root *parent_root;
819 parent_inode = pending->dentry->d_parent->d_inode;
820 parent_root = BTRFS_I(parent_inode)->root;
821 trans = btrfs_join_transaction(parent_root, 1);
824 * insert the directory item
826 namelen = strlen(pending->name);
827 ret = btrfs_set_inode_index(parent_inode, &index);
828 ret = btrfs_insert_dir_item(trans, parent_root,
829 pending->name, namelen,
831 &pending->root_key, BTRFS_FT_DIR, index);
836 btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
837 ret = btrfs_update_inode(trans, parent_root, parent_inode);
840 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
841 pending->root_key.objectid,
842 parent_root->root_key.objectid,
843 parent_inode->i_ino, index, pending->name,
849 btrfs_end_transaction(trans, fs_info->fs_root);
854 * create all the snapshots we've scheduled for creation
856 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
857 struct btrfs_fs_info *fs_info)
859 struct btrfs_pending_snapshot *pending;
860 struct list_head *head = &trans->transaction->pending_snapshots;
863 list_for_each_entry(pending, head, list) {
864 ret = create_pending_snapshot(trans, fs_info, pending);
870 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
871 struct btrfs_fs_info *fs_info)
873 struct btrfs_pending_snapshot *pending;
874 struct list_head *head = &trans->transaction->pending_snapshots;
877 while (!list_empty(head)) {
878 pending = list_entry(head->next,
879 struct btrfs_pending_snapshot, list);
880 ret = finish_pending_snapshot(fs_info, pending);
882 list_del(&pending->list);
883 kfree(pending->name);
889 static void update_super_roots(struct btrfs_root *root)
891 struct btrfs_root_item *root_item;
892 struct btrfs_super_block *super;
894 super = &root->fs_info->super_copy;
896 root_item = &root->fs_info->chunk_root->root_item;
897 super->chunk_root = root_item->bytenr;
898 super->chunk_root_generation = root_item->generation;
899 super->chunk_root_level = root_item->level;
901 root_item = &root->fs_info->tree_root->root_item;
902 super->root = root_item->bytenr;
903 super->generation = root_item->generation;
904 super->root_level = root_item->level;
907 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
910 spin_lock(&info->new_trans_lock);
911 if (info->running_transaction)
912 ret = info->running_transaction->in_commit;
913 spin_unlock(&info->new_trans_lock);
917 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
918 struct btrfs_root *root)
920 unsigned long joined = 0;
921 unsigned long timeout = 1;
922 struct btrfs_transaction *cur_trans;
923 struct btrfs_transaction *prev_trans = NULL;
927 unsigned long now = get_seconds();
928 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
930 btrfs_run_ordered_operations(root, 0);
932 /* make a pass through all the delayed refs we have so far
933 * any runnings procs may add more while we are here
935 ret = btrfs_run_delayed_refs(trans, root, 0);
938 cur_trans = trans->transaction;
940 * set the flushing flag so procs in this transaction have to
941 * start sending their work down.
943 cur_trans->delayed_refs.flushing = 1;
945 ret = btrfs_run_delayed_refs(trans, root, 0);
948 mutex_lock(&root->fs_info->trans_mutex);
949 if (cur_trans->in_commit) {
950 cur_trans->use_count++;
951 mutex_unlock(&root->fs_info->trans_mutex);
952 btrfs_end_transaction(trans, root);
954 ret = wait_for_commit(root, cur_trans);
957 mutex_lock(&root->fs_info->trans_mutex);
958 put_transaction(cur_trans);
959 mutex_unlock(&root->fs_info->trans_mutex);
964 trans->transaction->in_commit = 1;
965 trans->transaction->blocked = 1;
966 if (cur_trans->list.prev != &root->fs_info->trans_list) {
967 prev_trans = list_entry(cur_trans->list.prev,
968 struct btrfs_transaction, list);
969 if (!prev_trans->commit_done) {
970 prev_trans->use_count++;
971 mutex_unlock(&root->fs_info->trans_mutex);
973 wait_for_commit(root, prev_trans);
975 mutex_lock(&root->fs_info->trans_mutex);
976 put_transaction(prev_trans);
980 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
984 int snap_pending = 0;
985 joined = cur_trans->num_joined;
986 if (!list_empty(&trans->transaction->pending_snapshots))
989 WARN_ON(cur_trans != trans->transaction);
990 prepare_to_wait(&cur_trans->writer_wait, &wait,
991 TASK_UNINTERRUPTIBLE);
993 if (cur_trans->num_writers > 1)
994 timeout = MAX_SCHEDULE_TIMEOUT;
995 else if (should_grow)
998 mutex_unlock(&root->fs_info->trans_mutex);
1000 if (flush_on_commit) {
1001 btrfs_start_delalloc_inodes(root, 1);
1002 ret = btrfs_wait_ordered_extents(root, 0, 1);
1004 } else if (snap_pending) {
1005 ret = btrfs_wait_ordered_extents(root, 0, 1);
1010 * rename don't use btrfs_join_transaction, so, once we
1011 * set the transaction to blocked above, we aren't going
1012 * to get any new ordered operations. We can safely run
1013 * it here and no for sure that nothing new will be added
1016 btrfs_run_ordered_operations(root, 1);
1019 if (cur_trans->num_writers > 1 || should_grow)
1020 schedule_timeout(timeout);
1022 mutex_lock(&root->fs_info->trans_mutex);
1023 finish_wait(&cur_trans->writer_wait, &wait);
1024 } while (cur_trans->num_writers > 1 ||
1025 (should_grow && cur_trans->num_joined != joined));
1027 ret = create_pending_snapshots(trans, root->fs_info);
1030 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1033 WARN_ON(cur_trans != trans->transaction);
1035 /* btrfs_commit_tree_roots is responsible for getting the
1036 * various roots consistent with each other. Every pointer
1037 * in the tree of tree roots has to point to the most up to date
1038 * root for every subvolume and other tree. So, we have to keep
1039 * the tree logging code from jumping in and changing any
1042 * At this point in the commit, there can't be any tree-log
1043 * writers, but a little lower down we drop the trans mutex
1044 * and let new people in. By holding the tree_log_mutex
1045 * from now until after the super is written, we avoid races
1046 * with the tree-log code.
1048 mutex_lock(&root->fs_info->tree_log_mutex);
1050 ret = commit_fs_roots(trans, root);
1053 /* commit_fs_roots gets rid of all the tree log roots, it is now
1054 * safe to free the root of tree log roots
1056 btrfs_free_log_root_tree(trans, root->fs_info);
1058 ret = commit_cowonly_roots(trans, root);
1061 btrfs_prepare_extent_commit(trans, root);
1063 cur_trans = root->fs_info->running_transaction;
1064 spin_lock(&root->fs_info->new_trans_lock);
1065 root->fs_info->running_transaction = NULL;
1066 spin_unlock(&root->fs_info->new_trans_lock);
1068 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1069 root->fs_info->tree_root->node);
1070 switch_commit_root(root->fs_info->tree_root);
1072 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1073 root->fs_info->chunk_root->node);
1074 switch_commit_root(root->fs_info->chunk_root);
1076 update_super_roots(root);
1078 if (!root->fs_info->log_root_recovering) {
1079 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1080 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1083 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1084 sizeof(root->fs_info->super_copy));
1086 trans->transaction->blocked = 0;
1088 wake_up(&root->fs_info->transaction_wait);
1090 mutex_unlock(&root->fs_info->trans_mutex);
1091 ret = btrfs_write_and_wait_transaction(trans, root);
1093 write_ctree_super(trans, root, 0);
1096 * the super is written, we can safely allow the tree-loggers
1097 * to go about their business
1099 mutex_unlock(&root->fs_info->tree_log_mutex);
1101 btrfs_finish_extent_commit(trans, root);
1103 /* do the directory inserts of any pending snapshot creations */
1104 finish_pending_snapshots(trans, root->fs_info);
1106 mutex_lock(&root->fs_info->trans_mutex);
1108 cur_trans->commit_done = 1;
1110 root->fs_info->last_trans_committed = cur_trans->transid;
1112 wake_up(&cur_trans->commit_wait);
1114 put_transaction(cur_trans);
1115 put_transaction(cur_trans);
1117 mutex_unlock(&root->fs_info->trans_mutex);
1119 if (current->journal_info == trans)
1120 current->journal_info = NULL;
1122 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1124 if (current != root->fs_info->transaction_kthread)
1125 btrfs_run_delayed_iputs(root);
1131 * interface function to delete all the snapshots we have scheduled for deletion
1133 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1136 struct btrfs_fs_info *fs_info = root->fs_info;
1138 mutex_lock(&fs_info->trans_mutex);
1139 list_splice_init(&fs_info->dead_roots, &list);
1140 mutex_unlock(&fs_info->trans_mutex);
1142 while (!list_empty(&list)) {
1143 root = list_entry(list.next, struct btrfs_root, root_list);
1144 list_del(&root->root_list);
1146 if (btrfs_header_backref_rev(root->node) <
1147 BTRFS_MIXED_BACKREF_REV)
1148 btrfs_drop_snapshot(root, 0);
1150 btrfs_drop_snapshot(root, 1);