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/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
27 #include "transaction.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);
44 static noinline void switch_commit_root(struct btrfs_root *root)
46 free_extent_buffer(root->commit_root);
47 root->commit_root = btrfs_root_node(root);
51 * either allocate a new transaction or hop into the existing one
53 static noinline int join_transaction(struct btrfs_root *root)
55 struct btrfs_transaction *cur_trans;
56 cur_trans = root->fs_info->running_transaction;
58 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_ROOT;
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->commit_root != root->node);
111 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
112 (unsigned long)root->root_key.objectid,
113 BTRFS_ROOT_TRANS_TAG);
114 root->last_trans = trans->transid;
115 btrfs_init_reloc_root(trans, root);
120 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
121 struct btrfs_root *root)
126 mutex_lock(&root->fs_info->trans_mutex);
127 if (root->last_trans == trans->transid) {
128 mutex_unlock(&root->fs_info->trans_mutex);
132 record_root_in_trans(trans, root);
133 mutex_unlock(&root->fs_info->trans_mutex);
137 /* wait for commit against the current transaction to become unblocked
138 * when this is done, it is safe to start a new transaction, but the current
139 * transaction might not be fully on disk.
141 static void wait_current_trans(struct btrfs_root *root)
143 struct btrfs_transaction *cur_trans;
145 cur_trans = root->fs_info->running_transaction;
146 if (cur_trans && cur_trans->blocked) {
148 cur_trans->use_count++;
150 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
151 TASK_UNINTERRUPTIBLE);
152 if (!cur_trans->blocked)
154 mutex_unlock(&root->fs_info->trans_mutex);
156 mutex_lock(&root->fs_info->trans_mutex);
158 finish_wait(&root->fs_info->transaction_wait, &wait);
159 put_transaction(cur_trans);
163 enum btrfs_trans_type {
170 static int may_wait_transaction(struct btrfs_root *root, int type)
172 if (!root->fs_info->log_root_recovering &&
173 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
174 type == TRANS_USERSPACE))
179 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
180 u64 num_items, int type)
182 struct btrfs_trans_handle *h;
183 struct btrfs_transaction *cur_trans;
186 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
187 return ERR_PTR(-EROFS);
189 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
191 return ERR_PTR(-ENOMEM);
193 if (type != TRANS_JOIN_NOLOCK)
194 mutex_lock(&root->fs_info->trans_mutex);
195 if (may_wait_transaction(root, type))
196 wait_current_trans(root);
198 ret = join_transaction(root);
200 if (type != TRANS_JOIN_NOLOCK)
201 mutex_unlock(&root->fs_info->trans_mutex);
205 cur_trans = root->fs_info->running_transaction;
206 cur_trans->use_count++;
207 if (type != TRANS_JOIN_NOLOCK)
208 mutex_unlock(&root->fs_info->trans_mutex);
210 h->transid = cur_trans->transid;
211 h->transaction = cur_trans;
214 h->bytes_reserved = 0;
215 h->delayed_ref_updates = 0;
219 if (cur_trans->blocked && may_wait_transaction(root, type)) {
220 btrfs_commit_transaction(h, root);
225 ret = btrfs_trans_reserve_metadata(h, root, num_items);
226 if (ret == -EAGAIN) {
227 btrfs_commit_transaction(h, root);
231 btrfs_end_transaction(h, root);
236 if (type != TRANS_JOIN_NOLOCK)
237 mutex_lock(&root->fs_info->trans_mutex);
238 record_root_in_trans(h, root);
239 if (type != TRANS_JOIN_NOLOCK)
240 mutex_unlock(&root->fs_info->trans_mutex);
242 if (!current->journal_info && type != TRANS_USERSPACE)
243 current->journal_info = h;
247 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
250 return start_transaction(root, num_items, TRANS_START);
252 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
255 return start_transaction(root, 0, TRANS_JOIN);
258 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
261 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
264 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
267 return start_transaction(r, 0, TRANS_USERSPACE);
270 /* wait for a transaction commit to be fully complete */
271 static noinline int wait_for_commit(struct btrfs_root *root,
272 struct btrfs_transaction *commit)
275 mutex_lock(&root->fs_info->trans_mutex);
276 while (!commit->commit_done) {
277 prepare_to_wait(&commit->commit_wait, &wait,
278 TASK_UNINTERRUPTIBLE);
279 if (commit->commit_done)
281 mutex_unlock(&root->fs_info->trans_mutex);
283 mutex_lock(&root->fs_info->trans_mutex);
285 mutex_unlock(&root->fs_info->trans_mutex);
286 finish_wait(&commit->commit_wait, &wait);
290 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
292 struct btrfs_transaction *cur_trans = NULL, *t;
295 mutex_lock(&root->fs_info->trans_mutex);
299 if (transid <= root->fs_info->last_trans_committed)
302 /* find specified transaction */
303 list_for_each_entry(t, &root->fs_info->trans_list, list) {
304 if (t->transid == transid) {
308 if (t->transid > transid)
313 goto out_unlock; /* bad transid */
315 /* find newest transaction that is committing | committed */
316 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
326 goto out_unlock; /* nothing committing|committed */
329 cur_trans->use_count++;
330 mutex_unlock(&root->fs_info->trans_mutex);
332 wait_for_commit(root, cur_trans);
334 mutex_lock(&root->fs_info->trans_mutex);
335 put_transaction(cur_trans);
338 mutex_unlock(&root->fs_info->trans_mutex);
344 * rate limit against the drop_snapshot code. This helps to slow down new
345 * operations if the drop_snapshot code isn't able to keep up.
347 static void throttle_on_drops(struct btrfs_root *root)
349 struct btrfs_fs_info *info = root->fs_info;
350 int harder_count = 0;
353 if (atomic_read(&info->throttles)) {
356 thr = atomic_read(&info->throttle_gen);
359 prepare_to_wait(&info->transaction_throttle,
360 &wait, TASK_UNINTERRUPTIBLE);
361 if (!atomic_read(&info->throttles)) {
362 finish_wait(&info->transaction_throttle, &wait);
366 finish_wait(&info->transaction_throttle, &wait);
367 } while (thr == atomic_read(&info->throttle_gen));
370 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
374 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
378 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
385 void btrfs_throttle(struct btrfs_root *root)
387 mutex_lock(&root->fs_info->trans_mutex);
388 if (!root->fs_info->open_ioctl_trans)
389 wait_current_trans(root);
390 mutex_unlock(&root->fs_info->trans_mutex);
393 static int should_end_transaction(struct btrfs_trans_handle *trans,
394 struct btrfs_root *root)
397 ret = btrfs_block_rsv_check(trans, root,
398 &root->fs_info->global_block_rsv, 0, 5);
402 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
403 struct btrfs_root *root)
405 struct btrfs_transaction *cur_trans = trans->transaction;
408 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
411 updates = trans->delayed_ref_updates;
412 trans->delayed_ref_updates = 0;
414 btrfs_run_delayed_refs(trans, root, updates);
416 return should_end_transaction(trans, root);
419 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
420 struct btrfs_root *root, int throttle, int lock)
422 struct btrfs_transaction *cur_trans = trans->transaction;
423 struct btrfs_fs_info *info = root->fs_info;
427 unsigned long cur = trans->delayed_ref_updates;
428 trans->delayed_ref_updates = 0;
430 trans->transaction->delayed_refs.num_heads_ready > 64) {
431 trans->delayed_ref_updates = 0;
434 * do a full flush if the transaction is trying
437 if (trans->transaction->delayed_refs.flushing)
439 btrfs_run_delayed_refs(trans, root, cur);
446 btrfs_trans_release_metadata(trans, root);
448 if (lock && !root->fs_info->open_ioctl_trans &&
449 should_end_transaction(trans, root))
450 trans->transaction->blocked = 1;
452 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
454 return btrfs_commit_transaction(trans, root);
456 wake_up_process(info->transaction_kthread);
460 mutex_lock(&info->trans_mutex);
461 WARN_ON(cur_trans != info->running_transaction);
462 WARN_ON(cur_trans->num_writers < 1);
463 cur_trans->num_writers--;
466 if (waitqueue_active(&cur_trans->writer_wait))
467 wake_up(&cur_trans->writer_wait);
468 put_transaction(cur_trans);
470 mutex_unlock(&info->trans_mutex);
472 if (current->journal_info == trans)
473 current->journal_info = NULL;
474 memset(trans, 0, sizeof(*trans));
475 kmem_cache_free(btrfs_trans_handle_cachep, trans);
478 btrfs_run_delayed_iputs(root);
483 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
484 struct btrfs_root *root)
486 return __btrfs_end_transaction(trans, root, 0, 1);
489 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
490 struct btrfs_root *root)
492 return __btrfs_end_transaction(trans, root, 1, 1);
495 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
496 struct btrfs_root *root)
498 return __btrfs_end_transaction(trans, root, 0, 0);
502 * when btree blocks are allocated, they have some corresponding bits set for
503 * them in one of two extent_io trees. This is used to make sure all of
504 * those extents are sent to disk but does not wait on them
506 int btrfs_write_marked_extents(struct btrfs_root *root,
507 struct extent_io_tree *dirty_pages, int mark)
513 struct inode *btree_inode = root->fs_info->btree_inode;
519 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
523 while (start <= end) {
526 index = start >> PAGE_CACHE_SHIFT;
527 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
528 page = find_get_page(btree_inode->i_mapping, index);
532 btree_lock_page_hook(page);
533 if (!page->mapping) {
535 page_cache_release(page);
539 if (PageWriteback(page)) {
541 wait_on_page_writeback(page);
544 page_cache_release(page);
548 err = write_one_page(page, 0);
551 page_cache_release(page);
560 * when btree blocks are allocated, they have some corresponding bits set for
561 * them in one of two extent_io trees. This is used to make sure all of
562 * those extents are on disk for transaction or log commit. We wait
563 * on all the pages and clear them from the dirty pages state tree
565 int btrfs_wait_marked_extents(struct btrfs_root *root,
566 struct extent_io_tree *dirty_pages, int mark)
572 struct inode *btree_inode = root->fs_info->btree_inode;
578 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
583 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
584 while (start <= end) {
585 index = start >> PAGE_CACHE_SHIFT;
586 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
587 page = find_get_page(btree_inode->i_mapping, index);
590 if (PageDirty(page)) {
591 btree_lock_page_hook(page);
592 wait_on_page_writeback(page);
593 err = write_one_page(page, 0);
597 wait_on_page_writeback(page);
598 page_cache_release(page);
608 * when btree blocks are allocated, they have some corresponding bits set for
609 * them in one of two extent_io trees. This is used to make sure all of
610 * those extents are on disk for transaction or log commit
612 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
613 struct extent_io_tree *dirty_pages, int mark)
618 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
619 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
623 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root)
626 if (!trans || !trans->transaction) {
627 struct inode *btree_inode;
628 btree_inode = root->fs_info->btree_inode;
629 return filemap_write_and_wait(btree_inode->i_mapping);
631 return btrfs_write_and_wait_marked_extents(root,
632 &trans->transaction->dirty_pages,
637 * this is used to update the root pointer in the tree of tree roots.
639 * But, in the case of the extent allocation tree, updating the root
640 * pointer may allocate blocks which may change the root of the extent
643 * So, this loops and repeats and makes sure the cowonly root didn't
644 * change while the root pointer was being updated in the metadata.
646 static int update_cowonly_root(struct btrfs_trans_handle *trans,
647 struct btrfs_root *root)
652 struct btrfs_root *tree_root = root->fs_info->tree_root;
654 old_root_used = btrfs_root_used(&root->root_item);
655 btrfs_write_dirty_block_groups(trans, root);
658 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
659 if (old_root_bytenr == root->node->start &&
660 old_root_used == btrfs_root_used(&root->root_item))
663 btrfs_set_root_node(&root->root_item, root->node);
664 ret = btrfs_update_root(trans, tree_root,
669 old_root_used = btrfs_root_used(&root->root_item);
670 ret = btrfs_write_dirty_block_groups(trans, root);
674 if (root != root->fs_info->extent_root)
675 switch_commit_root(root);
681 * update all the cowonly tree roots on disk
683 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
684 struct btrfs_root *root)
686 struct btrfs_fs_info *fs_info = root->fs_info;
687 struct list_head *next;
688 struct extent_buffer *eb;
691 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
694 eb = btrfs_lock_root_node(fs_info->tree_root);
695 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
696 btrfs_tree_unlock(eb);
697 free_extent_buffer(eb);
699 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
702 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
703 next = fs_info->dirty_cowonly_roots.next;
705 root = list_entry(next, struct btrfs_root, dirty_list);
707 update_cowonly_root(trans, root);
710 down_write(&fs_info->extent_commit_sem);
711 switch_commit_root(fs_info->extent_root);
712 up_write(&fs_info->extent_commit_sem);
718 * dead roots are old snapshots that need to be deleted. This allocates
719 * a dirty root struct and adds it into the list of dead roots that need to
722 int btrfs_add_dead_root(struct btrfs_root *root)
724 mutex_lock(&root->fs_info->trans_mutex);
725 list_add(&root->root_list, &root->fs_info->dead_roots);
726 mutex_unlock(&root->fs_info->trans_mutex);
731 * update all the cowonly tree roots on disk
733 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
734 struct btrfs_root *root)
736 struct btrfs_root *gang[8];
737 struct btrfs_fs_info *fs_info = root->fs_info;
743 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
746 BTRFS_ROOT_TRANS_TAG);
749 for (i = 0; i < ret; i++) {
751 radix_tree_tag_clear(&fs_info->fs_roots_radix,
752 (unsigned long)root->root_key.objectid,
753 BTRFS_ROOT_TRANS_TAG);
755 btrfs_free_log(trans, root);
756 btrfs_update_reloc_root(trans, root);
757 btrfs_orphan_commit_root(trans, root);
759 if (root->commit_root != root->node) {
760 switch_commit_root(root);
761 btrfs_set_root_node(&root->root_item,
765 err = btrfs_update_root(trans, fs_info->tree_root,
776 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
777 * otherwise every leaf in the btree is read and defragged.
779 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
781 struct btrfs_fs_info *info = root->fs_info;
782 struct btrfs_trans_handle *trans;
786 if (xchg(&root->defrag_running, 1))
790 trans = btrfs_start_transaction(root, 0);
792 return PTR_ERR(trans);
794 ret = btrfs_defrag_leaves(trans, root, cacheonly);
796 nr = trans->blocks_used;
797 btrfs_end_transaction(trans, root);
798 btrfs_btree_balance_dirty(info->tree_root, nr);
801 if (root->fs_info->closing || ret != -EAGAIN)
804 root->defrag_running = 0;
810 * when dropping snapshots, we generate a ton of delayed refs, and it makes
811 * sense not to join the transaction while it is trying to flush the current
812 * queue of delayed refs out.
814 * This is used by the drop snapshot code only
816 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
820 mutex_lock(&info->trans_mutex);
821 while (info->running_transaction &&
822 info->running_transaction->delayed_refs.flushing) {
823 prepare_to_wait(&info->transaction_wait, &wait,
824 TASK_UNINTERRUPTIBLE);
825 mutex_unlock(&info->trans_mutex);
829 mutex_lock(&info->trans_mutex);
830 finish_wait(&info->transaction_wait, &wait);
832 mutex_unlock(&info->trans_mutex);
837 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
840 int btrfs_drop_dead_root(struct btrfs_root *root)
842 struct btrfs_trans_handle *trans;
843 struct btrfs_root *tree_root = root->fs_info->tree_root;
849 * we don't want to jump in and create a bunch of
850 * delayed refs if the transaction is starting to close
852 wait_transaction_pre_flush(tree_root->fs_info);
853 trans = btrfs_start_transaction(tree_root, 1);
856 * we've joined a transaction, make sure it isn't
859 if (trans->transaction->delayed_refs.flushing) {
860 btrfs_end_transaction(trans, tree_root);
864 ret = btrfs_drop_snapshot(trans, root);
868 ret = btrfs_update_root(trans, tree_root,
874 nr = trans->blocks_used;
875 ret = btrfs_end_transaction(trans, tree_root);
878 btrfs_btree_balance_dirty(tree_root, nr);
883 ret = btrfs_del_root(trans, tree_root, &root->root_key);
886 nr = trans->blocks_used;
887 ret = btrfs_end_transaction(trans, tree_root);
890 free_extent_buffer(root->node);
891 free_extent_buffer(root->commit_root);
894 btrfs_btree_balance_dirty(tree_root, nr);
900 * new snapshots need to be created at a very specific time in the
901 * transaction commit. This does the actual creation
903 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
904 struct btrfs_fs_info *fs_info,
905 struct btrfs_pending_snapshot *pending)
907 struct btrfs_key key;
908 struct btrfs_root_item *new_root_item;
909 struct btrfs_root *tree_root = fs_info->tree_root;
910 struct btrfs_root *root = pending->root;
911 struct btrfs_root *parent_root;
912 struct inode *parent_inode;
913 struct dentry *parent;
914 struct dentry *dentry;
915 struct extent_buffer *tmp;
916 struct extent_buffer *old;
923 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
924 if (!new_root_item) {
925 pending->error = -ENOMEM;
929 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
931 pending->error = ret;
935 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
936 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
938 if (to_reserve > 0) {
939 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
942 pending->error = ret;
947 key.objectid = objectid;
948 key.offset = (u64)-1;
949 key.type = BTRFS_ROOT_ITEM_KEY;
951 trans->block_rsv = &pending->block_rsv;
953 dentry = pending->dentry;
954 parent = dget_parent(dentry);
955 parent_inode = parent->d_inode;
956 parent_root = BTRFS_I(parent_inode)->root;
957 record_root_in_trans(trans, parent_root);
960 * insert the directory item
962 ret = btrfs_set_inode_index(parent_inode, &index);
964 ret = btrfs_insert_dir_item(trans, parent_root,
965 dentry->d_name.name, dentry->d_name.len,
966 parent_inode->i_ino, &key,
967 BTRFS_FT_DIR, index);
970 btrfs_i_size_write(parent_inode, parent_inode->i_size +
971 dentry->d_name.len * 2);
972 ret = btrfs_update_inode(trans, parent_root, parent_inode);
975 record_root_in_trans(trans, root);
976 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
977 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
979 root_flags = btrfs_root_flags(new_root_item);
980 if (pending->readonly)
981 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
983 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
984 btrfs_set_root_flags(new_root_item, root_flags);
986 old = btrfs_lock_root_node(root);
987 btrfs_cow_block(trans, root, old, NULL, 0, &old);
988 btrfs_set_lock_blocking(old);
990 btrfs_copy_root(trans, root, old, &tmp, objectid);
991 btrfs_tree_unlock(old);
992 free_extent_buffer(old);
994 btrfs_set_root_node(new_root_item, tmp);
995 /* record when the snapshot was created in key.offset */
996 key.offset = trans->transid;
997 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
998 btrfs_tree_unlock(tmp);
999 free_extent_buffer(tmp);
1003 * insert root back/forward references
1005 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1006 parent_root->root_key.objectid,
1007 parent_inode->i_ino, index,
1008 dentry->d_name.name, dentry->d_name.len);
1012 key.offset = (u64)-1;
1013 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1014 BUG_ON(IS_ERR(pending->snap));
1016 btrfs_reloc_post_snapshot(trans, pending);
1017 btrfs_orphan_post_snapshot(trans, pending);
1019 kfree(new_root_item);
1020 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1025 * create all the snapshots we've scheduled for creation
1027 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1028 struct btrfs_fs_info *fs_info)
1030 struct btrfs_pending_snapshot *pending;
1031 struct list_head *head = &trans->transaction->pending_snapshots;
1034 list_for_each_entry(pending, head, list) {
1035 ret = create_pending_snapshot(trans, fs_info, pending);
1041 static void update_super_roots(struct btrfs_root *root)
1043 struct btrfs_root_item *root_item;
1044 struct btrfs_super_block *super;
1046 super = &root->fs_info->super_copy;
1048 root_item = &root->fs_info->chunk_root->root_item;
1049 super->chunk_root = root_item->bytenr;
1050 super->chunk_root_generation = root_item->generation;
1051 super->chunk_root_level = root_item->level;
1053 root_item = &root->fs_info->tree_root->root_item;
1054 super->root = root_item->bytenr;
1055 super->generation = root_item->generation;
1056 super->root_level = root_item->level;
1057 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1058 super->cache_generation = root_item->generation;
1061 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1064 spin_lock(&info->new_trans_lock);
1065 if (info->running_transaction)
1066 ret = info->running_transaction->in_commit;
1067 spin_unlock(&info->new_trans_lock);
1071 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1074 spin_lock(&info->new_trans_lock);
1075 if (info->running_transaction)
1076 ret = info->running_transaction->blocked;
1077 spin_unlock(&info->new_trans_lock);
1082 * wait for the current transaction commit to start and block subsequent
1085 static void wait_current_trans_commit_start(struct btrfs_root *root,
1086 struct btrfs_transaction *trans)
1090 if (trans->in_commit)
1094 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1095 TASK_UNINTERRUPTIBLE);
1096 if (trans->in_commit) {
1097 finish_wait(&root->fs_info->transaction_blocked_wait,
1101 mutex_unlock(&root->fs_info->trans_mutex);
1103 mutex_lock(&root->fs_info->trans_mutex);
1104 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1109 * wait for the current transaction to start and then become unblocked.
1112 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1113 struct btrfs_transaction *trans)
1117 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1121 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1122 TASK_UNINTERRUPTIBLE);
1123 if (trans->commit_done ||
1124 (trans->in_commit && !trans->blocked)) {
1125 finish_wait(&root->fs_info->transaction_wait,
1129 mutex_unlock(&root->fs_info->trans_mutex);
1131 mutex_lock(&root->fs_info->trans_mutex);
1132 finish_wait(&root->fs_info->transaction_wait,
1138 * commit transactions asynchronously. once btrfs_commit_transaction_async
1139 * returns, any subsequent transaction will not be allowed to join.
1141 struct btrfs_async_commit {
1142 struct btrfs_trans_handle *newtrans;
1143 struct btrfs_root *root;
1144 struct delayed_work work;
1147 static void do_async_commit(struct work_struct *work)
1149 struct btrfs_async_commit *ac =
1150 container_of(work, struct btrfs_async_commit, work.work);
1152 btrfs_commit_transaction(ac->newtrans, ac->root);
1156 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1157 struct btrfs_root *root,
1158 int wait_for_unblock)
1160 struct btrfs_async_commit *ac;
1161 struct btrfs_transaction *cur_trans;
1163 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1167 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1169 ac->newtrans = btrfs_join_transaction(root, 0);
1170 if (IS_ERR(ac->newtrans)) {
1171 int err = PTR_ERR(ac->newtrans);
1176 /* take transaction reference */
1177 mutex_lock(&root->fs_info->trans_mutex);
1178 cur_trans = trans->transaction;
1179 cur_trans->use_count++;
1180 mutex_unlock(&root->fs_info->trans_mutex);
1182 btrfs_end_transaction(trans, root);
1183 schedule_delayed_work(&ac->work, 0);
1185 /* wait for transaction to start and unblock */
1186 mutex_lock(&root->fs_info->trans_mutex);
1187 if (wait_for_unblock)
1188 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1190 wait_current_trans_commit_start(root, cur_trans);
1191 put_transaction(cur_trans);
1192 mutex_unlock(&root->fs_info->trans_mutex);
1198 * btrfs_transaction state sequence:
1199 * in_commit = 0, blocked = 0 (initial)
1200 * in_commit = 1, blocked = 1
1204 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1205 struct btrfs_root *root)
1207 unsigned long joined = 0;
1208 struct btrfs_transaction *cur_trans;
1209 struct btrfs_transaction *prev_trans = NULL;
1212 int should_grow = 0;
1213 unsigned long now = get_seconds();
1214 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1216 btrfs_run_ordered_operations(root, 0);
1218 /* make a pass through all the delayed refs we have so far
1219 * any runnings procs may add more while we are here
1221 ret = btrfs_run_delayed_refs(trans, root, 0);
1224 btrfs_trans_release_metadata(trans, root);
1226 cur_trans = trans->transaction;
1228 * set the flushing flag so procs in this transaction have to
1229 * start sending their work down.
1231 cur_trans->delayed_refs.flushing = 1;
1233 ret = btrfs_run_delayed_refs(trans, root, 0);
1236 mutex_lock(&root->fs_info->trans_mutex);
1237 if (cur_trans->in_commit) {
1238 cur_trans->use_count++;
1239 mutex_unlock(&root->fs_info->trans_mutex);
1240 btrfs_end_transaction(trans, root);
1242 ret = wait_for_commit(root, cur_trans);
1245 mutex_lock(&root->fs_info->trans_mutex);
1246 put_transaction(cur_trans);
1247 mutex_unlock(&root->fs_info->trans_mutex);
1252 trans->transaction->in_commit = 1;
1253 trans->transaction->blocked = 1;
1254 wake_up(&root->fs_info->transaction_blocked_wait);
1256 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1257 prev_trans = list_entry(cur_trans->list.prev,
1258 struct btrfs_transaction, list);
1259 if (!prev_trans->commit_done) {
1260 prev_trans->use_count++;
1261 mutex_unlock(&root->fs_info->trans_mutex);
1263 wait_for_commit(root, prev_trans);
1265 mutex_lock(&root->fs_info->trans_mutex);
1266 put_transaction(prev_trans);
1270 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1274 int snap_pending = 0;
1275 joined = cur_trans->num_joined;
1276 if (!list_empty(&trans->transaction->pending_snapshots))
1279 WARN_ON(cur_trans != trans->transaction);
1280 mutex_unlock(&root->fs_info->trans_mutex);
1282 if (flush_on_commit || snap_pending) {
1283 btrfs_start_delalloc_inodes(root, 1);
1284 ret = btrfs_wait_ordered_extents(root, 0, 1);
1289 * rename don't use btrfs_join_transaction, so, once we
1290 * set the transaction to blocked above, we aren't going
1291 * to get any new ordered operations. We can safely run
1292 * it here and no for sure that nothing new will be added
1295 btrfs_run_ordered_operations(root, 1);
1297 prepare_to_wait(&cur_trans->writer_wait, &wait,
1298 TASK_UNINTERRUPTIBLE);
1301 if (cur_trans->num_writers > 1)
1302 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1303 else if (should_grow)
1304 schedule_timeout(1);
1306 mutex_lock(&root->fs_info->trans_mutex);
1307 finish_wait(&cur_trans->writer_wait, &wait);
1308 } while (cur_trans->num_writers > 1 ||
1309 (should_grow && cur_trans->num_joined != joined));
1311 ret = create_pending_snapshots(trans, root->fs_info);
1314 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1317 WARN_ON(cur_trans != trans->transaction);
1319 /* btrfs_commit_tree_roots is responsible for getting the
1320 * various roots consistent with each other. Every pointer
1321 * in the tree of tree roots has to point to the most up to date
1322 * root for every subvolume and other tree. So, we have to keep
1323 * the tree logging code from jumping in and changing any
1326 * At this point in the commit, there can't be any tree-log
1327 * writers, but a little lower down we drop the trans mutex
1328 * and let new people in. By holding the tree_log_mutex
1329 * from now until after the super is written, we avoid races
1330 * with the tree-log code.
1332 mutex_lock(&root->fs_info->tree_log_mutex);
1334 ret = commit_fs_roots(trans, root);
1337 /* commit_fs_roots gets rid of all the tree log roots, it is now
1338 * safe to free the root of tree log roots
1340 btrfs_free_log_root_tree(trans, root->fs_info);
1342 ret = commit_cowonly_roots(trans, root);
1345 btrfs_prepare_extent_commit(trans, root);
1347 cur_trans = root->fs_info->running_transaction;
1348 spin_lock(&root->fs_info->new_trans_lock);
1349 root->fs_info->running_transaction = NULL;
1350 spin_unlock(&root->fs_info->new_trans_lock);
1352 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1353 root->fs_info->tree_root->node);
1354 switch_commit_root(root->fs_info->tree_root);
1356 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1357 root->fs_info->chunk_root->node);
1358 switch_commit_root(root->fs_info->chunk_root);
1360 update_super_roots(root);
1362 if (!root->fs_info->log_root_recovering) {
1363 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1364 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1367 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1368 sizeof(root->fs_info->super_copy));
1370 trans->transaction->blocked = 0;
1372 wake_up(&root->fs_info->transaction_wait);
1374 mutex_unlock(&root->fs_info->trans_mutex);
1375 ret = btrfs_write_and_wait_transaction(trans, root);
1377 write_ctree_super(trans, root, 0);
1380 * the super is written, we can safely allow the tree-loggers
1381 * to go about their business
1383 mutex_unlock(&root->fs_info->tree_log_mutex);
1385 btrfs_finish_extent_commit(trans, root);
1387 mutex_lock(&root->fs_info->trans_mutex);
1389 cur_trans->commit_done = 1;
1391 root->fs_info->last_trans_committed = cur_trans->transid;
1393 wake_up(&cur_trans->commit_wait);
1395 put_transaction(cur_trans);
1396 put_transaction(cur_trans);
1398 trace_btrfs_transaction_commit(root);
1400 mutex_unlock(&root->fs_info->trans_mutex);
1402 if (current->journal_info == trans)
1403 current->journal_info = NULL;
1405 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1407 if (current != root->fs_info->transaction_kthread)
1408 btrfs_run_delayed_iputs(root);
1414 * interface function to delete all the snapshots we have scheduled for deletion
1416 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1419 struct btrfs_fs_info *fs_info = root->fs_info;
1421 mutex_lock(&fs_info->trans_mutex);
1422 list_splice_init(&fs_info->dead_roots, &list);
1423 mutex_unlock(&fs_info->trans_mutex);
1425 while (!list_empty(&list)) {
1426 root = list_entry(list.next, struct btrfs_root, root_list);
1427 list_del(&root->root_list);
1429 if (btrfs_header_backref_rev(root->node) <
1430 BTRFS_MIXED_BACKREF_REV)
1431 btrfs_drop_snapshot(root, NULL, 0);
1433 btrfs_drop_snapshot(root, NULL, 1);