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 down_write(&root->commit_root_sem);
46 free_extent_buffer(root->commit_root);
47 root->commit_root = btrfs_root_node(root);
48 up_write(&root->commit_root_sem);
52 * either allocate a new transaction or hop into the existing one
54 static noinline int join_transaction(struct btrfs_root *root)
56 struct btrfs_transaction *cur_trans;
57 cur_trans = root->fs_info->running_transaction;
59 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
62 root->fs_info->generation++;
63 cur_trans->num_writers = 1;
64 cur_trans->num_joined = 0;
65 cur_trans->transid = root->fs_info->generation;
66 init_waitqueue_head(&cur_trans->writer_wait);
67 init_waitqueue_head(&cur_trans->commit_wait);
68 cur_trans->in_commit = 0;
69 cur_trans->blocked = 0;
70 cur_trans->use_count = 1;
71 cur_trans->commit_done = 0;
72 cur_trans->start_time = get_seconds();
74 cur_trans->delayed_refs.root.rb_node = NULL;
75 cur_trans->delayed_refs.num_entries = 0;
76 cur_trans->delayed_refs.num_heads_ready = 0;
77 cur_trans->delayed_refs.num_heads = 0;
78 cur_trans->delayed_refs.flushing = 0;
79 cur_trans->delayed_refs.run_delayed_start = 0;
80 spin_lock_init(&cur_trans->delayed_refs.lock);
82 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
83 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
84 extent_io_tree_init(&cur_trans->dirty_pages,
85 root->fs_info->btree_inode->i_mapping,
87 spin_lock(&root->fs_info->new_trans_lock);
88 root->fs_info->running_transaction = cur_trans;
89 spin_unlock(&root->fs_info->new_trans_lock);
91 cur_trans->num_writers++;
92 cur_trans->num_joined++;
99 * this does all the record keeping required to make sure that a reference
100 * counted root is properly recorded in a given transaction. This is required
101 * to make sure the old root from before we joined the transaction is deleted
102 * when the transaction commits
104 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root)
107 if (root->ref_cows && root->last_trans < trans->transid) {
108 WARN_ON(root == root->fs_info->extent_root);
109 WARN_ON(root->root_item.refs == 0);
110 WARN_ON(root->commit_root != root->node);
112 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
113 (unsigned long)root->root_key.objectid,
114 BTRFS_ROOT_TRANS_TAG);
115 root->last_trans = trans->transid;
116 btrfs_init_reloc_root(trans, root);
121 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root)
127 mutex_lock(&root->fs_info->trans_mutex);
128 if (root->last_trans == trans->transid) {
129 mutex_unlock(&root->fs_info->trans_mutex);
133 record_root_in_trans(trans, root);
134 mutex_unlock(&root->fs_info->trans_mutex);
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 h->transid = root->fs_info->running_transaction->transid;
184 h->transaction = root->fs_info->running_transaction;
185 h->blocks_reserved = num_blocks;
188 h->alloc_exclude_nr = 0;
189 h->alloc_exclude_start = 0;
190 h->delayed_ref_updates = 0;
192 root->fs_info->running_transaction->use_count++;
193 record_root_in_trans(h, root);
194 mutex_unlock(&root->fs_info->trans_mutex);
198 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
201 return start_transaction(root, num_blocks, 1);
203 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
206 return start_transaction(root, num_blocks, 0);
209 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
212 return start_transaction(r, num_blocks, 2);
215 /* wait for a transaction commit to be fully complete */
216 static noinline int wait_for_commit(struct btrfs_root *root,
217 struct btrfs_transaction *commit)
220 mutex_lock(&root->fs_info->trans_mutex);
221 while (!commit->commit_done) {
222 prepare_to_wait(&commit->commit_wait, &wait,
223 TASK_UNINTERRUPTIBLE);
224 if (commit->commit_done)
226 mutex_unlock(&root->fs_info->trans_mutex);
228 mutex_lock(&root->fs_info->trans_mutex);
230 mutex_unlock(&root->fs_info->trans_mutex);
231 finish_wait(&commit->commit_wait, &wait);
237 * rate limit against the drop_snapshot code. This helps to slow down new
238 * operations if the drop_snapshot code isn't able to keep up.
240 static void throttle_on_drops(struct btrfs_root *root)
242 struct btrfs_fs_info *info = root->fs_info;
243 int harder_count = 0;
246 if (atomic_read(&info->throttles)) {
249 thr = atomic_read(&info->throttle_gen);
252 prepare_to_wait(&info->transaction_throttle,
253 &wait, TASK_UNINTERRUPTIBLE);
254 if (!atomic_read(&info->throttles)) {
255 finish_wait(&info->transaction_throttle, &wait);
259 finish_wait(&info->transaction_throttle, &wait);
260 } while (thr == atomic_read(&info->throttle_gen));
263 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
267 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
271 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
278 void btrfs_throttle(struct btrfs_root *root)
280 mutex_lock(&root->fs_info->trans_mutex);
281 if (!root->fs_info->open_ioctl_trans)
282 wait_current_trans(root);
283 mutex_unlock(&root->fs_info->trans_mutex);
286 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
287 struct btrfs_root *root, int throttle)
289 struct btrfs_transaction *cur_trans;
290 struct btrfs_fs_info *info = root->fs_info;
294 unsigned long cur = trans->delayed_ref_updates;
295 trans->delayed_ref_updates = 0;
297 trans->transaction->delayed_refs.num_heads_ready > 64) {
298 trans->delayed_ref_updates = 0;
301 * do a full flush if the transaction is trying
304 if (trans->transaction->delayed_refs.flushing)
306 btrfs_run_delayed_refs(trans, root, cur);
313 mutex_lock(&info->trans_mutex);
314 cur_trans = info->running_transaction;
315 WARN_ON(cur_trans != trans->transaction);
316 WARN_ON(cur_trans->num_writers < 1);
317 cur_trans->num_writers--;
319 if (waitqueue_active(&cur_trans->writer_wait))
320 wake_up(&cur_trans->writer_wait);
321 put_transaction(cur_trans);
322 mutex_unlock(&info->trans_mutex);
323 memset(trans, 0, sizeof(*trans));
324 kmem_cache_free(btrfs_trans_handle_cachep, trans);
329 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
330 struct btrfs_root *root)
332 return __btrfs_end_transaction(trans, root, 0);
335 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root)
338 return __btrfs_end_transaction(trans, root, 1);
342 * when btree blocks are allocated, they have some corresponding bits set for
343 * them in one of two extent_io trees. This is used to make sure all of
344 * those extents are on disk for transaction or log commit
346 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
347 struct extent_io_tree *dirty_pages)
353 struct inode *btree_inode = root->fs_info->btree_inode;
359 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
363 while (start <= end) {
366 index = start >> PAGE_CACHE_SHIFT;
367 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
368 page = find_get_page(btree_inode->i_mapping, index);
372 btree_lock_page_hook(page);
373 if (!page->mapping) {
375 page_cache_release(page);
379 if (PageWriteback(page)) {
381 wait_on_page_writeback(page);
384 page_cache_release(page);
388 err = write_one_page(page, 0);
391 page_cache_release(page);
395 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
400 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
401 while (start <= end) {
402 index = start >> PAGE_CACHE_SHIFT;
403 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
404 page = find_get_page(btree_inode->i_mapping, index);
407 if (PageDirty(page)) {
408 btree_lock_page_hook(page);
409 wait_on_page_writeback(page);
410 err = write_one_page(page, 0);
414 wait_on_page_writeback(page);
415 page_cache_release(page);
424 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
425 struct btrfs_root *root)
427 if (!trans || !trans->transaction) {
428 struct inode *btree_inode;
429 btree_inode = root->fs_info->btree_inode;
430 return filemap_write_and_wait(btree_inode->i_mapping);
432 return btrfs_write_and_wait_marked_extents(root,
433 &trans->transaction->dirty_pages);
437 * this is used to update the root pointer in the tree of tree roots.
439 * But, in the case of the extent allocation tree, updating the root
440 * pointer may allocate blocks which may change the root of the extent
443 * So, this loops and repeats and makes sure the cowonly root didn't
444 * change while the root pointer was being updated in the metadata.
446 static int update_cowonly_root(struct btrfs_trans_handle *trans,
447 struct btrfs_root *root)
451 struct btrfs_root *tree_root = root->fs_info->tree_root;
453 btrfs_write_dirty_block_groups(trans, root);
456 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
457 if (old_root_bytenr == root->node->start)
460 btrfs_set_root_node(&root->root_item, root->node);
461 ret = btrfs_update_root(trans, tree_root,
466 ret = btrfs_write_dirty_block_groups(trans, root);
469 switch_commit_root(root);
474 * update all the cowonly tree roots on disk
476 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
477 struct btrfs_root *root)
479 struct btrfs_fs_info *fs_info = root->fs_info;
480 struct list_head *next;
481 struct extent_buffer *eb;
484 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
487 eb = btrfs_lock_root_node(fs_info->tree_root);
488 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
489 btrfs_tree_unlock(eb);
490 free_extent_buffer(eb);
492 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
495 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
496 next = fs_info->dirty_cowonly_roots.next;
498 root = list_entry(next, struct btrfs_root, dirty_list);
500 update_cowonly_root(trans, root);
506 * dead roots are old snapshots that need to be deleted. This allocates
507 * a dirty root struct and adds it into the list of dead roots that need to
510 int btrfs_add_dead_root(struct btrfs_root *root)
512 mutex_lock(&root->fs_info->trans_mutex);
513 list_add(&root->root_list, &root->fs_info->dead_roots);
514 mutex_unlock(&root->fs_info->trans_mutex);
519 * update all the cowonly tree roots on disk
521 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
522 struct btrfs_root *root)
524 struct btrfs_root *gang[8];
525 struct btrfs_fs_info *fs_info = root->fs_info;
531 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
534 BTRFS_ROOT_TRANS_TAG);
537 for (i = 0; i < ret; i++) {
539 radix_tree_tag_clear(&fs_info->fs_roots_radix,
540 (unsigned long)root->root_key.objectid,
541 BTRFS_ROOT_TRANS_TAG);
543 btrfs_free_log(trans, root);
544 btrfs_update_reloc_root(trans, root);
546 if (root->commit_root != root->node) {
547 switch_commit_root(root);
548 btrfs_set_root_node(&root->root_item,
552 err = btrfs_update_root(trans, fs_info->tree_root,
563 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
564 * otherwise every leaf in the btree is read and defragged.
566 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
568 struct btrfs_fs_info *info = root->fs_info;
570 struct btrfs_trans_handle *trans;
574 if (root->defrag_running)
576 trans = btrfs_start_transaction(root, 1);
578 root->defrag_running = 1;
579 ret = btrfs_defrag_leaves(trans, root, cacheonly);
580 nr = trans->blocks_used;
581 btrfs_end_transaction(trans, root);
582 btrfs_btree_balance_dirty(info->tree_root, nr);
585 trans = btrfs_start_transaction(root, 1);
586 if (root->fs_info->closing || ret != -EAGAIN)
589 root->defrag_running = 0;
591 btrfs_end_transaction(trans, root);
597 * when dropping snapshots, we generate a ton of delayed refs, and it makes
598 * sense not to join the transaction while it is trying to flush the current
599 * queue of delayed refs out.
601 * This is used by the drop snapshot code only
603 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
607 mutex_lock(&info->trans_mutex);
608 while (info->running_transaction &&
609 info->running_transaction->delayed_refs.flushing) {
610 prepare_to_wait(&info->transaction_wait, &wait,
611 TASK_UNINTERRUPTIBLE);
612 mutex_unlock(&info->trans_mutex);
616 mutex_lock(&info->trans_mutex);
617 finish_wait(&info->transaction_wait, &wait);
619 mutex_unlock(&info->trans_mutex);
624 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
627 int btrfs_drop_dead_root(struct btrfs_root *root)
629 struct btrfs_trans_handle *trans;
630 struct btrfs_root *tree_root = root->fs_info->tree_root;
636 * we don't want to jump in and create a bunch of
637 * delayed refs if the transaction is starting to close
639 wait_transaction_pre_flush(tree_root->fs_info);
640 trans = btrfs_start_transaction(tree_root, 1);
643 * we've joined a transaction, make sure it isn't
646 if (trans->transaction->delayed_refs.flushing) {
647 btrfs_end_transaction(trans, tree_root);
651 ret = btrfs_drop_snapshot(trans, root);
655 ret = btrfs_update_root(trans, tree_root,
661 nr = trans->blocks_used;
662 ret = btrfs_end_transaction(trans, tree_root);
665 btrfs_btree_balance_dirty(tree_root, nr);
670 ret = btrfs_del_root(trans, tree_root, &root->root_key);
673 nr = trans->blocks_used;
674 ret = btrfs_end_transaction(trans, tree_root);
677 free_extent_buffer(root->node);
678 free_extent_buffer(root->commit_root);
681 btrfs_btree_balance_dirty(tree_root, nr);
687 * new snapshots need to be created at a very specific time in the
688 * transaction commit. This does the actual creation
690 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
691 struct btrfs_fs_info *fs_info,
692 struct btrfs_pending_snapshot *pending)
694 struct btrfs_key key;
695 struct btrfs_root_item *new_root_item;
696 struct btrfs_root *tree_root = fs_info->tree_root;
697 struct btrfs_root *root = pending->root;
698 struct extent_buffer *tmp;
699 struct extent_buffer *old;
703 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
704 if (!new_root_item) {
708 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
712 record_root_in_trans(trans, root);
713 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
714 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
716 key.objectid = objectid;
718 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
720 old = btrfs_lock_root_node(root);
721 btrfs_cow_block(trans, root, old, NULL, 0, &old);
722 btrfs_set_lock_blocking(old);
724 btrfs_copy_root(trans, root, old, &tmp, objectid);
725 btrfs_tree_unlock(old);
726 free_extent_buffer(old);
728 btrfs_set_root_node(new_root_item, tmp);
729 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
731 btrfs_tree_unlock(tmp);
732 free_extent_buffer(tmp);
736 key.offset = (u64)-1;
737 memcpy(&pending->root_key, &key, sizeof(key));
739 kfree(new_root_item);
743 static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
744 struct btrfs_pending_snapshot *pending)
749 struct btrfs_trans_handle *trans;
750 struct inode *parent_inode;
752 struct btrfs_root *parent_root;
754 parent_inode = pending->dentry->d_parent->d_inode;
755 parent_root = BTRFS_I(parent_inode)->root;
756 trans = btrfs_join_transaction(parent_root, 1);
759 * insert the directory item
761 namelen = strlen(pending->name);
762 ret = btrfs_set_inode_index(parent_inode, &index);
763 ret = btrfs_insert_dir_item(trans, parent_root,
764 pending->name, namelen,
766 &pending->root_key, BTRFS_FT_DIR, index);
771 btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
772 ret = btrfs_update_inode(trans, parent_root, parent_inode);
775 /* add the backref first */
776 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
777 pending->root_key.objectid,
778 BTRFS_ROOT_BACKREF_KEY,
779 parent_root->root_key.objectid,
780 parent_inode->i_ino, index, pending->name,
785 /* now add the forward ref */
786 ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
787 parent_root->root_key.objectid,
789 pending->root_key.objectid,
790 parent_inode->i_ino, index, pending->name,
793 inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
794 d_instantiate(pending->dentry, inode);
796 btrfs_end_transaction(trans, fs_info->fs_root);
801 * create all the snapshots we've scheduled for creation
803 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
804 struct btrfs_fs_info *fs_info)
806 struct btrfs_pending_snapshot *pending;
807 struct list_head *head = &trans->transaction->pending_snapshots;
810 list_for_each_entry(pending, head, list) {
811 ret = create_pending_snapshot(trans, fs_info, pending);
817 static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
818 struct btrfs_fs_info *fs_info)
820 struct btrfs_pending_snapshot *pending;
821 struct list_head *head = &trans->transaction->pending_snapshots;
824 while (!list_empty(head)) {
825 pending = list_entry(head->next,
826 struct btrfs_pending_snapshot, list);
827 ret = finish_pending_snapshot(fs_info, pending);
829 list_del(&pending->list);
830 kfree(pending->name);
836 static void update_super_roots(struct btrfs_root *root)
838 struct btrfs_root_item *root_item;
839 struct btrfs_super_block *super;
841 super = &root->fs_info->super_copy;
843 root_item = &root->fs_info->chunk_root->root_item;
844 super->chunk_root = root_item->bytenr;
845 super->chunk_root_generation = root_item->generation;
846 super->chunk_root_level = root_item->level;
848 root_item = &root->fs_info->tree_root->root_item;
849 super->root = root_item->bytenr;
850 super->generation = root_item->generation;
851 super->root_level = root_item->level;
854 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
855 struct btrfs_root *root)
857 unsigned long joined = 0;
858 unsigned long timeout = 1;
859 struct btrfs_transaction *cur_trans;
860 struct btrfs_transaction *prev_trans = NULL;
861 struct extent_io_tree *pinned_copy;
865 unsigned long now = get_seconds();
866 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
868 btrfs_run_ordered_operations(root, 0);
870 /* make a pass through all the delayed refs we have so far
871 * any runnings procs may add more while we are here
873 ret = btrfs_run_delayed_refs(trans, root, 0);
876 cur_trans = trans->transaction;
878 * set the flushing flag so procs in this transaction have to
879 * start sending their work down.
881 cur_trans->delayed_refs.flushing = 1;
883 ret = btrfs_run_delayed_refs(trans, root, 0);
886 mutex_lock(&root->fs_info->trans_mutex);
887 if (cur_trans->in_commit) {
888 cur_trans->use_count++;
889 mutex_unlock(&root->fs_info->trans_mutex);
890 btrfs_end_transaction(trans, root);
892 ret = wait_for_commit(root, cur_trans);
895 mutex_lock(&root->fs_info->trans_mutex);
896 put_transaction(cur_trans);
897 mutex_unlock(&root->fs_info->trans_mutex);
902 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
906 extent_io_tree_init(pinned_copy,
907 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
909 trans->transaction->in_commit = 1;
910 trans->transaction->blocked = 1;
911 if (cur_trans->list.prev != &root->fs_info->trans_list) {
912 prev_trans = list_entry(cur_trans->list.prev,
913 struct btrfs_transaction, list);
914 if (!prev_trans->commit_done) {
915 prev_trans->use_count++;
916 mutex_unlock(&root->fs_info->trans_mutex);
918 wait_for_commit(root, prev_trans);
920 mutex_lock(&root->fs_info->trans_mutex);
921 put_transaction(prev_trans);
925 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
929 int snap_pending = 0;
930 joined = cur_trans->num_joined;
931 if (!list_empty(&trans->transaction->pending_snapshots))
934 WARN_ON(cur_trans != trans->transaction);
935 prepare_to_wait(&cur_trans->writer_wait, &wait,
936 TASK_UNINTERRUPTIBLE);
938 if (cur_trans->num_writers > 1)
939 timeout = MAX_SCHEDULE_TIMEOUT;
940 else if (should_grow)
943 mutex_unlock(&root->fs_info->trans_mutex);
945 if (flush_on_commit) {
946 btrfs_start_delalloc_inodes(root);
947 ret = btrfs_wait_ordered_extents(root, 0);
949 } else if (snap_pending) {
950 ret = btrfs_wait_ordered_extents(root, 1);
955 * rename don't use btrfs_join_transaction, so, once we
956 * set the transaction to blocked above, we aren't going
957 * to get any new ordered operations. We can safely run
958 * it here and no for sure that nothing new will be added
961 btrfs_run_ordered_operations(root, 1);
964 if (cur_trans->num_writers > 1 || should_grow)
965 schedule_timeout(timeout);
967 mutex_lock(&root->fs_info->trans_mutex);
968 finish_wait(&cur_trans->writer_wait, &wait);
969 } while (cur_trans->num_writers > 1 ||
970 (should_grow && cur_trans->num_joined != joined));
972 ret = create_pending_snapshots(trans, root->fs_info);
975 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
978 WARN_ON(cur_trans != trans->transaction);
980 /* btrfs_commit_tree_roots is responsible for getting the
981 * various roots consistent with each other. Every pointer
982 * in the tree of tree roots has to point to the most up to date
983 * root for every subvolume and other tree. So, we have to keep
984 * the tree logging code from jumping in and changing any
987 * At this point in the commit, there can't be any tree-log
988 * writers, but a little lower down we drop the trans mutex
989 * and let new people in. By holding the tree_log_mutex
990 * from now until after the super is written, we avoid races
991 * with the tree-log code.
993 mutex_lock(&root->fs_info->tree_log_mutex);
995 ret = commit_fs_roots(trans, root);
998 /* commit_fs_roots gets rid of all the tree log roots, it is now
999 * safe to free the root of tree log roots
1001 btrfs_free_log_root_tree(trans, root->fs_info);
1003 ret = commit_cowonly_roots(trans, root);
1006 cur_trans = root->fs_info->running_transaction;
1007 spin_lock(&root->fs_info->new_trans_lock);
1008 root->fs_info->running_transaction = NULL;
1009 spin_unlock(&root->fs_info->new_trans_lock);
1011 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1012 root->fs_info->tree_root->node);
1013 switch_commit_root(root->fs_info->tree_root);
1015 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1016 root->fs_info->chunk_root->node);
1017 switch_commit_root(root->fs_info->chunk_root);
1019 update_super_roots(root);
1021 if (!root->fs_info->log_root_recovering) {
1022 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1023 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1026 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1027 sizeof(root->fs_info->super_copy));
1029 btrfs_copy_pinned(root, pinned_copy);
1031 trans->transaction->blocked = 0;
1033 wake_up(&root->fs_info->transaction_wait);
1035 mutex_unlock(&root->fs_info->trans_mutex);
1036 ret = btrfs_write_and_wait_transaction(trans, root);
1038 write_ctree_super(trans, root, 0);
1041 * the super is written, we can safely allow the tree-loggers
1042 * to go about their business
1044 mutex_unlock(&root->fs_info->tree_log_mutex);
1046 btrfs_finish_extent_commit(trans, root, pinned_copy);
1049 /* do the directory inserts of any pending snapshot creations */
1050 finish_pending_snapshots(trans, root->fs_info);
1052 mutex_lock(&root->fs_info->trans_mutex);
1054 cur_trans->commit_done = 1;
1056 root->fs_info->last_trans_committed = cur_trans->transid;
1058 wake_up(&cur_trans->commit_wait);
1060 put_transaction(cur_trans);
1061 put_transaction(cur_trans);
1063 mutex_unlock(&root->fs_info->trans_mutex);
1065 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1070 * interface function to delete all the snapshots we have scheduled for deletion
1072 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1075 struct btrfs_fs_info *fs_info = root->fs_info;
1077 mutex_lock(&fs_info->trans_mutex);
1078 list_splice_init(&fs_info->dead_roots, &list);
1079 mutex_unlock(&fs_info->trans_mutex);
1081 while (!list_empty(&list)) {
1082 root = list_entry(list.next, struct btrfs_root, root_list);
1083 list_del_init(&root->root_list);
1084 btrfs_drop_snapshot(root, 0);