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"
30 #include "inode-map.h"
32 #define BTRFS_ROOT_TRANS_TAG 0
34 static noinline void put_transaction(struct btrfs_transaction *transaction)
36 WARN_ON(atomic_read(&transaction->use_count) == 0);
37 if (atomic_dec_and_test(&transaction->use_count)) {
38 BUG_ON(!list_empty(&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, int nofail)
55 struct btrfs_transaction *cur_trans;
57 spin_lock(&root->fs_info->trans_lock);
58 if (root->fs_info->trans_no_join) {
60 spin_unlock(&root->fs_info->trans_lock);
65 cur_trans = root->fs_info->running_transaction;
67 atomic_inc(&cur_trans->use_count);
68 atomic_inc(&cur_trans->num_writers);
69 cur_trans->num_joined++;
70 spin_unlock(&root->fs_info->trans_lock);
73 spin_unlock(&root->fs_info->trans_lock);
75 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
78 spin_lock(&root->fs_info->trans_lock);
79 if (root->fs_info->running_transaction) {
80 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
81 cur_trans = root->fs_info->running_transaction;
82 atomic_inc(&cur_trans->use_count);
83 atomic_inc(&cur_trans->num_writers);
84 cur_trans->num_joined++;
85 spin_unlock(&root->fs_info->trans_lock);
88 atomic_set(&cur_trans->num_writers, 1);
89 cur_trans->num_joined = 0;
90 init_waitqueue_head(&cur_trans->writer_wait);
91 init_waitqueue_head(&cur_trans->commit_wait);
92 cur_trans->in_commit = 0;
93 cur_trans->blocked = 0;
95 * One for this trans handle, one so it will live on until we
96 * commit the transaction.
98 atomic_set(&cur_trans->use_count, 2);
99 cur_trans->commit_done = 0;
100 cur_trans->start_time = get_seconds();
102 cur_trans->delayed_refs.root = RB_ROOT;
103 cur_trans->delayed_refs.num_entries = 0;
104 cur_trans->delayed_refs.num_heads_ready = 0;
105 cur_trans->delayed_refs.num_heads = 0;
106 cur_trans->delayed_refs.flushing = 0;
107 cur_trans->delayed_refs.run_delayed_start = 0;
108 spin_lock_init(&cur_trans->commit_lock);
109 spin_lock_init(&cur_trans->delayed_refs.lock);
111 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
112 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
113 extent_io_tree_init(&cur_trans->dirty_pages,
114 root->fs_info->btree_inode->i_mapping);
115 root->fs_info->generation++;
116 cur_trans->transid = root->fs_info->generation;
117 root->fs_info->running_transaction = cur_trans;
118 spin_unlock(&root->fs_info->trans_lock);
124 * this does all the record keeping required to make sure that a reference
125 * counted root is properly recorded in a given transaction. This is required
126 * to make sure the old root from before we joined the transaction is deleted
127 * when the transaction commits
129 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
130 struct btrfs_root *root)
132 if (root->ref_cows && root->last_trans < trans->transid) {
133 WARN_ON(root == root->fs_info->extent_root);
134 WARN_ON(root->commit_root != root->node);
136 spin_lock(&root->fs_info->fs_roots_radix_lock);
137 if (root->last_trans == trans->transid) {
138 spin_unlock(&root->fs_info->fs_roots_radix_lock);
141 root->last_trans = trans->transid;
142 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
143 (unsigned long)root->root_key.objectid,
144 BTRFS_ROOT_TRANS_TAG);
145 spin_unlock(&root->fs_info->fs_roots_radix_lock);
146 btrfs_init_reloc_root(trans, root);
151 /* wait for commit against the current transaction to become unblocked
152 * when this is done, it is safe to start a new transaction, but the current
153 * transaction might not be fully on disk.
155 static void wait_current_trans(struct btrfs_root *root)
157 struct btrfs_transaction *cur_trans;
159 spin_lock(&root->fs_info->trans_lock);
160 cur_trans = root->fs_info->running_transaction;
161 if (cur_trans && cur_trans->blocked) {
163 atomic_inc(&cur_trans->use_count);
164 spin_unlock(&root->fs_info->trans_lock);
166 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
167 TASK_UNINTERRUPTIBLE);
168 if (!cur_trans->blocked)
172 finish_wait(&root->fs_info->transaction_wait, &wait);
173 put_transaction(cur_trans);
175 spin_unlock(&root->fs_info->trans_lock);
179 enum btrfs_trans_type {
186 static int may_wait_transaction(struct btrfs_root *root, int type)
188 if (root->fs_info->log_root_recovering)
191 if (type == TRANS_USERSPACE)
194 if (type == TRANS_START &&
195 !atomic_read(&root->fs_info->open_ioctl_trans))
201 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
202 u64 num_items, int type)
204 struct btrfs_trans_handle *h;
205 struct btrfs_transaction *cur_trans;
209 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
210 return ERR_PTR(-EROFS);
212 if (current->journal_info) {
213 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
214 h = current->journal_info;
216 h->orig_rsv = h->block_rsv;
221 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
223 return ERR_PTR(-ENOMEM);
225 if (may_wait_transaction(root, type))
226 wait_current_trans(root);
229 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
231 wait_current_trans(root);
232 } while (ret == -EBUSY);
235 kmem_cache_free(btrfs_trans_handle_cachep, h);
239 cur_trans = root->fs_info->running_transaction;
241 h->transid = cur_trans->transid;
242 h->transaction = cur_trans;
244 h->bytes_reserved = 0;
245 h->delayed_ref_updates = 0;
251 if (cur_trans->blocked && may_wait_transaction(root, type)) {
252 btrfs_commit_transaction(h, root);
257 ret = btrfs_trans_reserve_metadata(h, root, num_items);
258 if (ret == -EAGAIN && !retries) {
260 btrfs_commit_transaction(h, root);
262 } else if (ret == -EAGAIN) {
264 * We have already retried and got EAGAIN, so really we
265 * don't have space, so set ret to -ENOSPC.
271 btrfs_end_transaction(h, root);
277 btrfs_record_root_in_trans(h, root);
279 if (!current->journal_info && type != TRANS_USERSPACE)
280 current->journal_info = h;
284 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
287 return start_transaction(root, num_items, TRANS_START);
289 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
291 return start_transaction(root, 0, TRANS_JOIN);
294 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
296 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
299 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
301 return start_transaction(root, 0, TRANS_USERSPACE);
304 /* wait for a transaction commit to be fully complete */
305 static noinline int wait_for_commit(struct btrfs_root *root,
306 struct btrfs_transaction *commit)
309 while (!commit->commit_done) {
310 prepare_to_wait(&commit->commit_wait, &wait,
311 TASK_UNINTERRUPTIBLE);
312 if (commit->commit_done)
316 finish_wait(&commit->commit_wait, &wait);
320 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
322 struct btrfs_transaction *cur_trans = NULL, *t;
327 if (transid <= root->fs_info->last_trans_committed)
330 /* find specified transaction */
331 spin_lock(&root->fs_info->trans_lock);
332 list_for_each_entry(t, &root->fs_info->trans_list, list) {
333 if (t->transid == transid) {
335 atomic_inc(&cur_trans->use_count);
338 if (t->transid > transid)
341 spin_unlock(&root->fs_info->trans_lock);
344 goto out; /* bad transid */
346 /* find newest transaction that is committing | committed */
347 spin_lock(&root->fs_info->trans_lock);
348 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
354 atomic_inc(&cur_trans->use_count);
358 spin_unlock(&root->fs_info->trans_lock);
360 goto out; /* nothing committing|committed */
363 wait_for_commit(root, cur_trans);
365 put_transaction(cur_trans);
371 void btrfs_throttle(struct btrfs_root *root)
373 if (!atomic_read(&root->fs_info->open_ioctl_trans))
374 wait_current_trans(root);
377 static int should_end_transaction(struct btrfs_trans_handle *trans,
378 struct btrfs_root *root)
381 ret = btrfs_block_rsv_check(trans, root,
382 &root->fs_info->global_block_rsv, 0, 5);
386 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
387 struct btrfs_root *root)
389 struct btrfs_transaction *cur_trans = trans->transaction;
393 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
396 updates = trans->delayed_ref_updates;
397 trans->delayed_ref_updates = 0;
399 btrfs_run_delayed_refs(trans, root, updates);
401 return should_end_transaction(trans, root);
404 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
405 struct btrfs_root *root, int throttle, int lock)
407 struct btrfs_transaction *cur_trans = trans->transaction;
408 struct btrfs_fs_info *info = root->fs_info;
411 if (--trans->use_count) {
412 trans->block_rsv = trans->orig_rsv;
417 unsigned long cur = trans->delayed_ref_updates;
418 trans->delayed_ref_updates = 0;
420 trans->transaction->delayed_refs.num_heads_ready > 64) {
421 trans->delayed_ref_updates = 0;
424 * do a full flush if the transaction is trying
427 if (trans->transaction->delayed_refs.flushing)
429 btrfs_run_delayed_refs(trans, root, cur);
436 btrfs_trans_release_metadata(trans, root);
438 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
439 should_end_transaction(trans, root)) {
440 trans->transaction->blocked = 1;
444 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
446 return btrfs_commit_transaction(trans, root);
448 wake_up_process(info->transaction_kthread);
451 WARN_ON(cur_trans != info->running_transaction);
452 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
453 atomic_dec(&cur_trans->num_writers);
456 if (waitqueue_active(&cur_trans->writer_wait))
457 wake_up(&cur_trans->writer_wait);
458 put_transaction(cur_trans);
460 if (current->journal_info == trans)
461 current->journal_info = NULL;
462 memset(trans, 0, sizeof(*trans));
463 kmem_cache_free(btrfs_trans_handle_cachep, trans);
466 btrfs_run_delayed_iputs(root);
471 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
472 struct btrfs_root *root)
476 ret = __btrfs_end_transaction(trans, root, 0, 1);
482 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
483 struct btrfs_root *root)
487 ret = __btrfs_end_transaction(trans, root, 1, 1);
493 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
494 struct btrfs_root *root)
498 ret = __btrfs_end_transaction(trans, root, 0, 0);
504 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
505 struct btrfs_root *root)
507 return __btrfs_end_transaction(trans, root, 1, 1);
511 * when btree blocks are allocated, they have some corresponding bits set for
512 * them in one of two extent_io trees. This is used to make sure all of
513 * those extents are sent to disk but does not wait on them
515 int btrfs_write_marked_extents(struct btrfs_root *root,
516 struct extent_io_tree *dirty_pages, int mark)
522 struct inode *btree_inode = root->fs_info->btree_inode;
528 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
532 while (start <= end) {
535 index = start >> PAGE_CACHE_SHIFT;
536 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
537 page = find_get_page(btree_inode->i_mapping, index);
541 btree_lock_page_hook(page);
542 if (!page->mapping) {
544 page_cache_release(page);
548 if (PageWriteback(page)) {
550 wait_on_page_writeback(page);
553 page_cache_release(page);
557 err = write_one_page(page, 0);
560 page_cache_release(page);
569 * when btree blocks are allocated, they have some corresponding bits set for
570 * them in one of two extent_io trees. This is used to make sure all of
571 * those extents are on disk for transaction or log commit. We wait
572 * on all the pages and clear them from the dirty pages state tree
574 int btrfs_wait_marked_extents(struct btrfs_root *root,
575 struct extent_io_tree *dirty_pages, int mark)
581 struct inode *btree_inode = root->fs_info->btree_inode;
587 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
592 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
593 while (start <= end) {
594 index = start >> PAGE_CACHE_SHIFT;
595 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
596 page = find_get_page(btree_inode->i_mapping, index);
599 if (PageDirty(page)) {
600 btree_lock_page_hook(page);
601 wait_on_page_writeback(page);
602 err = write_one_page(page, 0);
606 wait_on_page_writeback(page);
607 page_cache_release(page);
617 * when btree blocks are allocated, they have some corresponding bits set for
618 * them in one of two extent_io trees. This is used to make sure all of
619 * those extents are on disk for transaction or log commit
621 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
622 struct extent_io_tree *dirty_pages, int mark)
627 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
628 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
632 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
633 struct btrfs_root *root)
635 if (!trans || !trans->transaction) {
636 struct inode *btree_inode;
637 btree_inode = root->fs_info->btree_inode;
638 return filemap_write_and_wait(btree_inode->i_mapping);
640 return btrfs_write_and_wait_marked_extents(root,
641 &trans->transaction->dirty_pages,
646 * this is used to update the root pointer in the tree of tree roots.
648 * But, in the case of the extent allocation tree, updating the root
649 * pointer may allocate blocks which may change the root of the extent
652 * So, this loops and repeats and makes sure the cowonly root didn't
653 * change while the root pointer was being updated in the metadata.
655 static int update_cowonly_root(struct btrfs_trans_handle *trans,
656 struct btrfs_root *root)
661 struct btrfs_root *tree_root = root->fs_info->tree_root;
663 old_root_used = btrfs_root_used(&root->root_item);
664 btrfs_write_dirty_block_groups(trans, root);
667 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
668 if (old_root_bytenr == root->node->start &&
669 old_root_used == btrfs_root_used(&root->root_item))
672 btrfs_set_root_node(&root->root_item, root->node);
673 ret = btrfs_update_root(trans, tree_root,
678 old_root_used = btrfs_root_used(&root->root_item);
679 ret = btrfs_write_dirty_block_groups(trans, root);
683 if (root != root->fs_info->extent_root)
684 switch_commit_root(root);
690 * update all the cowonly tree roots on disk
692 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
693 struct btrfs_root *root)
695 struct btrfs_fs_info *fs_info = root->fs_info;
696 struct list_head *next;
697 struct extent_buffer *eb;
700 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
703 eb = btrfs_lock_root_node(fs_info->tree_root);
704 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
705 btrfs_tree_unlock(eb);
706 free_extent_buffer(eb);
708 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
711 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
712 next = fs_info->dirty_cowonly_roots.next;
714 root = list_entry(next, struct btrfs_root, dirty_list);
716 update_cowonly_root(trans, root);
719 down_write(&fs_info->extent_commit_sem);
720 switch_commit_root(fs_info->extent_root);
721 up_write(&fs_info->extent_commit_sem);
727 * dead roots are old snapshots that need to be deleted. This allocates
728 * a dirty root struct and adds it into the list of dead roots that need to
731 int btrfs_add_dead_root(struct btrfs_root *root)
733 spin_lock(&root->fs_info->trans_lock);
734 list_add(&root->root_list, &root->fs_info->dead_roots);
735 spin_unlock(&root->fs_info->trans_lock);
740 * update all the cowonly tree roots on disk
742 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root)
745 struct btrfs_root *gang[8];
746 struct btrfs_fs_info *fs_info = root->fs_info;
751 spin_lock(&fs_info->fs_roots_radix_lock);
753 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
756 BTRFS_ROOT_TRANS_TAG);
759 for (i = 0; i < ret; i++) {
761 radix_tree_tag_clear(&fs_info->fs_roots_radix,
762 (unsigned long)root->root_key.objectid,
763 BTRFS_ROOT_TRANS_TAG);
764 spin_unlock(&fs_info->fs_roots_radix_lock);
766 btrfs_free_log(trans, root);
767 btrfs_update_reloc_root(trans, root);
768 btrfs_orphan_commit_root(trans, root);
770 btrfs_save_ino_cache(root, trans);
772 if (root->commit_root != root->node) {
773 mutex_lock(&root->fs_commit_mutex);
774 switch_commit_root(root);
775 btrfs_unpin_free_ino(root);
776 mutex_unlock(&root->fs_commit_mutex);
778 btrfs_set_root_node(&root->root_item,
782 err = btrfs_update_root(trans, fs_info->tree_root,
785 spin_lock(&fs_info->fs_roots_radix_lock);
790 spin_unlock(&fs_info->fs_roots_radix_lock);
795 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
796 * otherwise every leaf in the btree is read and defragged.
798 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
800 struct btrfs_fs_info *info = root->fs_info;
801 struct btrfs_trans_handle *trans;
805 if (xchg(&root->defrag_running, 1))
809 trans = btrfs_start_transaction(root, 0);
811 return PTR_ERR(trans);
813 ret = btrfs_defrag_leaves(trans, root, cacheonly);
815 nr = trans->blocks_used;
816 btrfs_end_transaction(trans, root);
817 btrfs_btree_balance_dirty(info->tree_root, nr);
820 if (root->fs_info->closing || ret != -EAGAIN)
823 root->defrag_running = 0;
828 * new snapshots need to be created at a very specific time in the
829 * transaction commit. This does the actual creation
831 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
832 struct btrfs_fs_info *fs_info,
833 struct btrfs_pending_snapshot *pending)
835 struct btrfs_key key;
836 struct btrfs_root_item *new_root_item;
837 struct btrfs_root *tree_root = fs_info->tree_root;
838 struct btrfs_root *root = pending->root;
839 struct btrfs_root *parent_root;
840 struct inode *parent_inode;
841 struct dentry *parent;
842 struct dentry *dentry;
843 struct extent_buffer *tmp;
844 struct extent_buffer *old;
851 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
852 if (!new_root_item) {
853 pending->error = -ENOMEM;
857 ret = btrfs_find_free_objectid(tree_root, &objectid);
859 pending->error = ret;
863 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
864 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
866 if (to_reserve > 0) {
867 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
870 pending->error = ret;
875 key.objectid = objectid;
876 key.offset = (u64)-1;
877 key.type = BTRFS_ROOT_ITEM_KEY;
879 trans->block_rsv = &pending->block_rsv;
881 dentry = pending->dentry;
882 parent = dget_parent(dentry);
883 parent_inode = parent->d_inode;
884 parent_root = BTRFS_I(parent_inode)->root;
885 btrfs_record_root_in_trans(trans, parent_root);
888 * insert the directory item
890 ret = btrfs_set_inode_index(parent_inode, &index);
892 ret = btrfs_insert_dir_item(trans, parent_root,
893 dentry->d_name.name, dentry->d_name.len,
895 BTRFS_FT_DIR, index);
898 btrfs_i_size_write(parent_inode, parent_inode->i_size +
899 dentry->d_name.len * 2);
900 ret = btrfs_update_inode(trans, parent_root, parent_inode);
903 btrfs_record_root_in_trans(trans, root);
904 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
905 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
906 btrfs_check_and_init_root_item(new_root_item);
908 root_flags = btrfs_root_flags(new_root_item);
909 if (pending->readonly)
910 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
912 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
913 btrfs_set_root_flags(new_root_item, root_flags);
915 old = btrfs_lock_root_node(root);
916 btrfs_cow_block(trans, root, old, NULL, 0, &old);
917 btrfs_set_lock_blocking(old);
919 btrfs_copy_root(trans, root, old, &tmp, objectid);
920 btrfs_tree_unlock(old);
921 free_extent_buffer(old);
923 btrfs_set_root_node(new_root_item, tmp);
924 /* record when the snapshot was created in key.offset */
925 key.offset = trans->transid;
926 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
927 btrfs_tree_unlock(tmp);
928 free_extent_buffer(tmp);
932 * insert root back/forward references
934 ret = btrfs_add_root_ref(trans, tree_root, objectid,
935 parent_root->root_key.objectid,
936 btrfs_ino(parent_inode), index,
937 dentry->d_name.name, dentry->d_name.len);
941 key.offset = (u64)-1;
942 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
943 BUG_ON(IS_ERR(pending->snap));
945 btrfs_reloc_post_snapshot(trans, pending);
946 btrfs_orphan_post_snapshot(trans, pending);
948 kfree(new_root_item);
949 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
954 * create all the snapshots we've scheduled for creation
956 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
957 struct btrfs_fs_info *fs_info)
959 struct btrfs_pending_snapshot *pending;
960 struct list_head *head = &trans->transaction->pending_snapshots;
963 list_for_each_entry(pending, head, list) {
965 * We must deal with the delayed items before creating
966 * snapshots, or we will create a snapthot with inconsistent
969 ret = btrfs_run_delayed_items(trans, fs_info->fs_root);
972 ret = create_pending_snapshot(trans, fs_info, pending);
978 static void update_super_roots(struct btrfs_root *root)
980 struct btrfs_root_item *root_item;
981 struct btrfs_super_block *super;
983 super = &root->fs_info->super_copy;
985 root_item = &root->fs_info->chunk_root->root_item;
986 super->chunk_root = root_item->bytenr;
987 super->chunk_root_generation = root_item->generation;
988 super->chunk_root_level = root_item->level;
990 root_item = &root->fs_info->tree_root->root_item;
991 super->root = root_item->bytenr;
992 super->generation = root_item->generation;
993 super->root_level = root_item->level;
994 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
995 super->cache_generation = root_item->generation;
998 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1001 spin_lock(&info->trans_lock);
1002 if (info->running_transaction)
1003 ret = info->running_transaction->in_commit;
1004 spin_unlock(&info->trans_lock);
1008 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1011 spin_lock(&info->trans_lock);
1012 if (info->running_transaction)
1013 ret = info->running_transaction->blocked;
1014 spin_unlock(&info->trans_lock);
1019 * wait for the current transaction commit to start and block subsequent
1022 static void wait_current_trans_commit_start(struct btrfs_root *root,
1023 struct btrfs_transaction *trans)
1027 if (trans->in_commit)
1031 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1032 TASK_UNINTERRUPTIBLE);
1033 if (trans->in_commit) {
1034 finish_wait(&root->fs_info->transaction_blocked_wait,
1039 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1044 * wait for the current transaction to start and then become unblocked.
1047 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1048 struct btrfs_transaction *trans)
1052 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1056 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1057 TASK_UNINTERRUPTIBLE);
1058 if (trans->commit_done ||
1059 (trans->in_commit && !trans->blocked)) {
1060 finish_wait(&root->fs_info->transaction_wait,
1065 finish_wait(&root->fs_info->transaction_wait,
1071 * commit transactions asynchronously. once btrfs_commit_transaction_async
1072 * returns, any subsequent transaction will not be allowed to join.
1074 struct btrfs_async_commit {
1075 struct btrfs_trans_handle *newtrans;
1076 struct btrfs_root *root;
1077 struct delayed_work work;
1080 static void do_async_commit(struct work_struct *work)
1082 struct btrfs_async_commit *ac =
1083 container_of(work, struct btrfs_async_commit, work.work);
1085 btrfs_commit_transaction(ac->newtrans, ac->root);
1089 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1090 struct btrfs_root *root,
1091 int wait_for_unblock)
1093 struct btrfs_async_commit *ac;
1094 struct btrfs_transaction *cur_trans;
1096 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1100 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1102 ac->newtrans = btrfs_join_transaction(root);
1103 if (IS_ERR(ac->newtrans)) {
1104 int err = PTR_ERR(ac->newtrans);
1109 /* take transaction reference */
1110 cur_trans = trans->transaction;
1111 atomic_inc(&cur_trans->use_count);
1113 btrfs_end_transaction(trans, root);
1114 schedule_delayed_work(&ac->work, 0);
1116 /* wait for transaction to start and unblock */
1117 if (wait_for_unblock)
1118 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1120 wait_current_trans_commit_start(root, cur_trans);
1121 put_transaction(cur_trans);
1127 * btrfs_transaction state sequence:
1128 * in_commit = 0, blocked = 0 (initial)
1129 * in_commit = 1, blocked = 1
1133 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1134 struct btrfs_root *root)
1136 unsigned long joined = 0;
1137 struct btrfs_transaction *cur_trans;
1138 struct btrfs_transaction *prev_trans = NULL;
1141 int should_grow = 0;
1142 unsigned long now = get_seconds();
1143 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1145 btrfs_run_ordered_operations(root, 0);
1147 /* make a pass through all the delayed refs we have so far
1148 * any runnings procs may add more while we are here
1150 ret = btrfs_run_delayed_refs(trans, root, 0);
1153 btrfs_trans_release_metadata(trans, root);
1155 cur_trans = trans->transaction;
1157 * set the flushing flag so procs in this transaction have to
1158 * start sending their work down.
1160 cur_trans->delayed_refs.flushing = 1;
1162 ret = btrfs_run_delayed_refs(trans, root, 0);
1165 spin_lock(&cur_trans->commit_lock);
1166 if (cur_trans->in_commit) {
1167 spin_unlock(&cur_trans->commit_lock);
1168 atomic_inc(&cur_trans->use_count);
1169 btrfs_end_transaction(trans, root);
1171 ret = wait_for_commit(root, cur_trans);
1174 put_transaction(cur_trans);
1179 trans->transaction->in_commit = 1;
1180 trans->transaction->blocked = 1;
1181 spin_unlock(&cur_trans->commit_lock);
1182 wake_up(&root->fs_info->transaction_blocked_wait);
1184 spin_lock(&root->fs_info->trans_lock);
1185 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1186 prev_trans = list_entry(cur_trans->list.prev,
1187 struct btrfs_transaction, list);
1188 if (!prev_trans->commit_done) {
1189 atomic_inc(&prev_trans->use_count);
1190 spin_unlock(&root->fs_info->trans_lock);
1192 wait_for_commit(root, prev_trans);
1194 put_transaction(prev_trans);
1196 spin_unlock(&root->fs_info->trans_lock);
1199 spin_unlock(&root->fs_info->trans_lock);
1202 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1206 int snap_pending = 0;
1208 joined = cur_trans->num_joined;
1209 if (!list_empty(&trans->transaction->pending_snapshots))
1212 WARN_ON(cur_trans != trans->transaction);
1214 if (flush_on_commit || snap_pending) {
1215 btrfs_start_delalloc_inodes(root, 1);
1216 ret = btrfs_wait_ordered_extents(root, 0, 1);
1220 ret = btrfs_run_delayed_items(trans, root);
1224 * rename don't use btrfs_join_transaction, so, once we
1225 * set the transaction to blocked above, we aren't going
1226 * to get any new ordered operations. We can safely run
1227 * it here and no for sure that nothing new will be added
1230 btrfs_run_ordered_operations(root, 1);
1232 prepare_to_wait(&cur_trans->writer_wait, &wait,
1233 TASK_UNINTERRUPTIBLE);
1235 if (atomic_read(&cur_trans->num_writers) > 1)
1236 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1237 else if (should_grow)
1238 schedule_timeout(1);
1240 finish_wait(&cur_trans->writer_wait, &wait);
1241 spin_lock(&root->fs_info->trans_lock);
1242 root->fs_info->trans_no_join = 1;
1243 spin_unlock(&root->fs_info->trans_lock);
1244 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1245 (should_grow && cur_trans->num_joined != joined));
1247 ret = create_pending_snapshots(trans, root->fs_info);
1250 ret = btrfs_run_delayed_items(trans, root);
1253 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1256 WARN_ON(cur_trans != trans->transaction);
1258 btrfs_scrub_pause(root);
1259 /* btrfs_commit_tree_roots is responsible for getting the
1260 * various roots consistent with each other. Every pointer
1261 * in the tree of tree roots has to point to the most up to date
1262 * root for every subvolume and other tree. So, we have to keep
1263 * the tree logging code from jumping in and changing any
1266 * At this point in the commit, there can't be any tree-log
1267 * writers, but a little lower down we drop the trans mutex
1268 * and let new people in. By holding the tree_log_mutex
1269 * from now until after the super is written, we avoid races
1270 * with the tree-log code.
1272 mutex_lock(&root->fs_info->tree_log_mutex);
1274 ret = commit_fs_roots(trans, root);
1277 /* commit_fs_roots gets rid of all the tree log roots, it is now
1278 * safe to free the root of tree log roots
1280 btrfs_free_log_root_tree(trans, root->fs_info);
1282 ret = commit_cowonly_roots(trans, root);
1285 btrfs_prepare_extent_commit(trans, root);
1287 cur_trans = root->fs_info->running_transaction;
1289 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1290 root->fs_info->tree_root->node);
1291 switch_commit_root(root->fs_info->tree_root);
1293 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1294 root->fs_info->chunk_root->node);
1295 switch_commit_root(root->fs_info->chunk_root);
1297 update_super_roots(root);
1299 if (!root->fs_info->log_root_recovering) {
1300 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1301 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1304 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1305 sizeof(root->fs_info->super_copy));
1307 trans->transaction->blocked = 0;
1308 spin_lock(&root->fs_info->trans_lock);
1309 root->fs_info->running_transaction = NULL;
1310 root->fs_info->trans_no_join = 0;
1311 spin_unlock(&root->fs_info->trans_lock);
1313 wake_up(&root->fs_info->transaction_wait);
1315 ret = btrfs_write_and_wait_transaction(trans, root);
1317 write_ctree_super(trans, root, 0);
1320 * the super is written, we can safely allow the tree-loggers
1321 * to go about their business
1323 mutex_unlock(&root->fs_info->tree_log_mutex);
1325 btrfs_finish_extent_commit(trans, root);
1327 cur_trans->commit_done = 1;
1329 root->fs_info->last_trans_committed = cur_trans->transid;
1331 wake_up(&cur_trans->commit_wait);
1333 spin_lock(&root->fs_info->trans_lock);
1334 list_del_init(&cur_trans->list);
1335 spin_unlock(&root->fs_info->trans_lock);
1337 put_transaction(cur_trans);
1338 put_transaction(cur_trans);
1340 trace_btrfs_transaction_commit(root);
1342 btrfs_scrub_continue(root);
1344 if (current->journal_info == trans)
1345 current->journal_info = NULL;
1347 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1349 if (current != root->fs_info->transaction_kthread)
1350 btrfs_run_delayed_iputs(root);
1356 * interface function to delete all the snapshots we have scheduled for deletion
1358 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1361 struct btrfs_fs_info *fs_info = root->fs_info;
1363 spin_lock(&fs_info->trans_lock);
1364 list_splice_init(&fs_info->dead_roots, &list);
1365 spin_unlock(&fs_info->trans_lock);
1367 while (!list_empty(&list)) {
1368 root = list_entry(list.next, struct btrfs_root, root_list);
1369 list_del(&root->root_list);
1371 btrfs_kill_all_delayed_nodes(root);
1373 if (btrfs_header_backref_rev(root->node) <
1374 BTRFS_MIXED_BACKREF_REV)
1375 btrfs_drop_snapshot(root, NULL, 0);
1377 btrfs_drop_snapshot(root, NULL, 1);