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 static int 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);
137 * see below for in_trans_setup usage rules
138 * we have the reloc mutex held now, so there
139 * is only one writer in this function
141 root->in_trans_setup = 1;
143 /* make sure readers find in_trans_setup before
144 * they find our root->last_trans update
148 spin_lock(&root->fs_info->fs_roots_radix_lock);
149 if (root->last_trans == trans->transid) {
150 spin_unlock(&root->fs_info->fs_roots_radix_lock);
153 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
154 (unsigned long)root->root_key.objectid,
155 BTRFS_ROOT_TRANS_TAG);
156 spin_unlock(&root->fs_info->fs_roots_radix_lock);
157 root->last_trans = trans->transid;
159 /* this is pretty tricky. We don't want to
160 * take the relocation lock in btrfs_record_root_in_trans
161 * unless we're really doing the first setup for this root in
164 * Normally we'd use root->last_trans as a flag to decide
165 * if we want to take the expensive mutex.
167 * But, we have to set root->last_trans before we
168 * init the relocation root, otherwise, we trip over warnings
169 * in ctree.c. The solution used here is to flag ourselves
170 * with root->in_trans_setup. When this is 1, we're still
171 * fixing up the reloc trees and everyone must wait.
173 * When this is zero, they can trust root->last_trans and fly
174 * through btrfs_record_root_in_trans without having to take the
175 * lock. smp_wmb() makes sure that all the writes above are
176 * done before we pop in the zero below
178 btrfs_init_reloc_root(trans, root);
180 root->in_trans_setup = 0;
186 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
187 struct btrfs_root *root)
193 * see record_root_in_trans for comments about in_trans_setup usage
197 if (root->last_trans == trans->transid &&
198 !root->in_trans_setup)
201 mutex_lock(&root->fs_info->reloc_mutex);
202 record_root_in_trans(trans, root);
203 mutex_unlock(&root->fs_info->reloc_mutex);
208 /* wait for commit against the current transaction to become unblocked
209 * when this is done, it is safe to start a new transaction, but the current
210 * transaction might not be fully on disk.
212 static void wait_current_trans(struct btrfs_root *root)
214 struct btrfs_transaction *cur_trans;
216 spin_lock(&root->fs_info->trans_lock);
217 cur_trans = root->fs_info->running_transaction;
218 if (cur_trans && cur_trans->blocked) {
219 atomic_inc(&cur_trans->use_count);
220 spin_unlock(&root->fs_info->trans_lock);
222 wait_event(root->fs_info->transaction_wait,
223 !cur_trans->blocked);
224 put_transaction(cur_trans);
226 spin_unlock(&root->fs_info->trans_lock);
230 enum btrfs_trans_type {
237 static int may_wait_transaction(struct btrfs_root *root, int type)
239 if (root->fs_info->log_root_recovering)
242 if (type == TRANS_USERSPACE)
245 if (type == TRANS_START &&
246 !atomic_read(&root->fs_info->open_ioctl_trans))
252 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
253 u64 num_items, int type)
255 struct btrfs_trans_handle *h;
256 struct btrfs_transaction *cur_trans;
260 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
261 return ERR_PTR(-EROFS);
263 if (current->journal_info) {
264 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
265 h = current->journal_info;
267 h->orig_rsv = h->block_rsv;
273 * Do the reservation before we join the transaction so we can do all
274 * the appropriate flushing if need be.
276 if (num_items > 0 && root != root->fs_info->chunk_root) {
277 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
278 ret = btrfs_block_rsv_add(root,
279 &root->fs_info->trans_block_rsv,
285 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
287 return ERR_PTR(-ENOMEM);
289 if (may_wait_transaction(root, type))
290 wait_current_trans(root);
293 ret = join_transaction(root, type == TRANS_JOIN_NOLOCK);
295 wait_current_trans(root);
296 } while (ret == -EBUSY);
299 kmem_cache_free(btrfs_trans_handle_cachep, h);
303 cur_trans = root->fs_info->running_transaction;
305 h->transid = cur_trans->transid;
306 h->transaction = cur_trans;
308 h->bytes_reserved = 0;
309 h->delayed_ref_updates = 0;
315 if (cur_trans->blocked && may_wait_transaction(root, type)) {
316 btrfs_commit_transaction(h, root);
321 h->block_rsv = &root->fs_info->trans_block_rsv;
322 h->bytes_reserved = num_bytes;
326 btrfs_record_root_in_trans(h, root);
328 if (!current->journal_info && type != TRANS_USERSPACE)
329 current->journal_info = h;
333 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
336 return start_transaction(root, num_items, TRANS_START);
338 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
340 return start_transaction(root, 0, TRANS_JOIN);
343 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
345 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
348 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
350 return start_transaction(root, 0, TRANS_USERSPACE);
353 /* wait for a transaction commit to be fully complete */
354 static noinline void wait_for_commit(struct btrfs_root *root,
355 struct btrfs_transaction *commit)
357 wait_event(commit->commit_wait, commit->commit_done);
360 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
362 struct btrfs_transaction *cur_trans = NULL, *t;
367 if (transid <= root->fs_info->last_trans_committed)
370 /* find specified transaction */
371 spin_lock(&root->fs_info->trans_lock);
372 list_for_each_entry(t, &root->fs_info->trans_list, list) {
373 if (t->transid == transid) {
375 atomic_inc(&cur_trans->use_count);
378 if (t->transid > transid)
381 spin_unlock(&root->fs_info->trans_lock);
384 goto out; /* bad transid */
386 /* find newest transaction that is committing | committed */
387 spin_lock(&root->fs_info->trans_lock);
388 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
394 atomic_inc(&cur_trans->use_count);
398 spin_unlock(&root->fs_info->trans_lock);
400 goto out; /* nothing committing|committed */
403 wait_for_commit(root, cur_trans);
405 put_transaction(cur_trans);
411 void btrfs_throttle(struct btrfs_root *root)
413 if (!atomic_read(&root->fs_info->open_ioctl_trans))
414 wait_current_trans(root);
417 static int should_end_transaction(struct btrfs_trans_handle *trans,
418 struct btrfs_root *root)
422 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
426 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
427 struct btrfs_root *root)
429 struct btrfs_transaction *cur_trans = trans->transaction;
430 struct btrfs_block_rsv *rsv = trans->block_rsv;
434 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
438 * We need to do this in case we're deleting csums so the global block
439 * rsv get's used instead of the csum block rsv.
441 trans->block_rsv = NULL;
443 updates = trans->delayed_ref_updates;
444 trans->delayed_ref_updates = 0;
446 btrfs_run_delayed_refs(trans, root, updates);
448 trans->block_rsv = rsv;
450 return should_end_transaction(trans, root);
453 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
454 struct btrfs_root *root, int throttle, int lock)
456 struct btrfs_transaction *cur_trans = trans->transaction;
457 struct btrfs_fs_info *info = root->fs_info;
460 if (--trans->use_count) {
461 trans->block_rsv = trans->orig_rsv;
465 btrfs_trans_release_metadata(trans, root);
466 trans->block_rsv = NULL;
468 unsigned long cur = trans->delayed_ref_updates;
469 trans->delayed_ref_updates = 0;
471 trans->transaction->delayed_refs.num_heads_ready > 64) {
472 trans->delayed_ref_updates = 0;
475 * do a full flush if the transaction is trying
478 if (trans->transaction->delayed_refs.flushing)
480 btrfs_run_delayed_refs(trans, root, cur);
487 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
488 should_end_transaction(trans, root)) {
489 trans->transaction->blocked = 1;
493 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
496 * We may race with somebody else here so end up having
497 * to call end_transaction on ourselves again, so inc
501 return btrfs_commit_transaction(trans, root);
503 wake_up_process(info->transaction_kthread);
507 WARN_ON(cur_trans != info->running_transaction);
508 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
509 atomic_dec(&cur_trans->num_writers);
512 if (waitqueue_active(&cur_trans->writer_wait))
513 wake_up(&cur_trans->writer_wait);
514 put_transaction(cur_trans);
516 if (current->journal_info == trans)
517 current->journal_info = NULL;
518 memset(trans, 0, sizeof(*trans));
519 kmem_cache_free(btrfs_trans_handle_cachep, trans);
522 btrfs_run_delayed_iputs(root);
527 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
528 struct btrfs_root *root)
532 ret = __btrfs_end_transaction(trans, root, 0, 1);
538 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
539 struct btrfs_root *root)
543 ret = __btrfs_end_transaction(trans, root, 1, 1);
549 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
550 struct btrfs_root *root)
554 ret = __btrfs_end_transaction(trans, root, 0, 0);
560 int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
561 struct btrfs_root *root)
563 return __btrfs_end_transaction(trans, root, 1, 1);
567 * when btree blocks are allocated, they have some corresponding bits set for
568 * them in one of two extent_io trees. This is used to make sure all of
569 * those extents are sent to disk but does not wait on them
571 int btrfs_write_marked_extents(struct btrfs_root *root,
572 struct extent_io_tree *dirty_pages, int mark)
576 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
580 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
582 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark,
584 err = filemap_fdatawrite_range(mapping, start, end);
596 * when btree blocks are allocated, they have some corresponding bits set for
597 * them in one of two extent_io trees. This is used to make sure all of
598 * those extents are on disk for transaction or log commit. We wait
599 * on all the pages and clear them from the dirty pages state tree
601 int btrfs_wait_marked_extents(struct btrfs_root *root,
602 struct extent_io_tree *dirty_pages, int mark)
606 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
610 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
612 clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS);
613 err = filemap_fdatawait_range(mapping, start, end);
625 * when btree blocks are allocated, they have some corresponding bits set for
626 * them in one of two extent_io trees. This is used to make sure all of
627 * those extents are on disk for transaction or log commit
629 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
630 struct extent_io_tree *dirty_pages, int mark)
635 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
636 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
640 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
641 struct btrfs_root *root)
643 if (!trans || !trans->transaction) {
644 struct inode *btree_inode;
645 btree_inode = root->fs_info->btree_inode;
646 return filemap_write_and_wait(btree_inode->i_mapping);
648 return btrfs_write_and_wait_marked_extents(root,
649 &trans->transaction->dirty_pages,
654 * this is used to update the root pointer in the tree of tree roots.
656 * But, in the case of the extent allocation tree, updating the root
657 * pointer may allocate blocks which may change the root of the extent
660 * So, this loops and repeats and makes sure the cowonly root didn't
661 * change while the root pointer was being updated in the metadata.
663 static int update_cowonly_root(struct btrfs_trans_handle *trans,
664 struct btrfs_root *root)
669 struct btrfs_root *tree_root = root->fs_info->tree_root;
671 old_root_used = btrfs_root_used(&root->root_item);
672 btrfs_write_dirty_block_groups(trans, root);
675 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
676 if (old_root_bytenr == root->node->start &&
677 old_root_used == btrfs_root_used(&root->root_item))
680 btrfs_set_root_node(&root->root_item, root->node);
681 ret = btrfs_update_root(trans, tree_root,
686 old_root_used = btrfs_root_used(&root->root_item);
687 ret = btrfs_write_dirty_block_groups(trans, root);
691 if (root != root->fs_info->extent_root)
692 switch_commit_root(root);
698 * update all the cowonly tree roots on disk
700 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
701 struct btrfs_root *root)
703 struct btrfs_fs_info *fs_info = root->fs_info;
704 struct list_head *next;
705 struct extent_buffer *eb;
708 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
711 eb = btrfs_lock_root_node(fs_info->tree_root);
712 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
713 btrfs_tree_unlock(eb);
714 free_extent_buffer(eb);
716 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
719 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
720 next = fs_info->dirty_cowonly_roots.next;
722 root = list_entry(next, struct btrfs_root, dirty_list);
724 update_cowonly_root(trans, root);
727 down_write(&fs_info->extent_commit_sem);
728 switch_commit_root(fs_info->extent_root);
729 up_write(&fs_info->extent_commit_sem);
735 * dead roots are old snapshots that need to be deleted. This allocates
736 * a dirty root struct and adds it into the list of dead roots that need to
739 int btrfs_add_dead_root(struct btrfs_root *root)
741 spin_lock(&root->fs_info->trans_lock);
742 list_add(&root->root_list, &root->fs_info->dead_roots);
743 spin_unlock(&root->fs_info->trans_lock);
748 * update all the cowonly tree roots on disk
750 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
751 struct btrfs_root *root)
753 struct btrfs_root *gang[8];
754 struct btrfs_fs_info *fs_info = root->fs_info;
759 spin_lock(&fs_info->fs_roots_radix_lock);
761 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
764 BTRFS_ROOT_TRANS_TAG);
767 for (i = 0; i < ret; i++) {
769 radix_tree_tag_clear(&fs_info->fs_roots_radix,
770 (unsigned long)root->root_key.objectid,
771 BTRFS_ROOT_TRANS_TAG);
772 spin_unlock(&fs_info->fs_roots_radix_lock);
774 btrfs_free_log(trans, root);
775 btrfs_update_reloc_root(trans, root);
776 btrfs_orphan_commit_root(trans, root);
778 btrfs_save_ino_cache(root, trans);
780 if (root->commit_root != root->node) {
781 mutex_lock(&root->fs_commit_mutex);
782 switch_commit_root(root);
783 btrfs_unpin_free_ino(root);
784 mutex_unlock(&root->fs_commit_mutex);
786 btrfs_set_root_node(&root->root_item,
790 err = btrfs_update_root(trans, fs_info->tree_root,
793 spin_lock(&fs_info->fs_roots_radix_lock);
798 spin_unlock(&fs_info->fs_roots_radix_lock);
803 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
804 * otherwise every leaf in the btree is read and defragged.
806 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
808 struct btrfs_fs_info *info = root->fs_info;
809 struct btrfs_trans_handle *trans;
813 if (xchg(&root->defrag_running, 1))
817 trans = btrfs_start_transaction(root, 0);
819 return PTR_ERR(trans);
821 ret = btrfs_defrag_leaves(trans, root, cacheonly);
823 nr = trans->blocks_used;
824 btrfs_end_transaction(trans, root);
825 btrfs_btree_balance_dirty(info->tree_root, nr);
828 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
831 root->defrag_running = 0;
836 * new snapshots need to be created at a very specific time in the
837 * transaction commit. This does the actual creation
839 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
840 struct btrfs_fs_info *fs_info,
841 struct btrfs_pending_snapshot *pending)
843 struct btrfs_key key;
844 struct btrfs_root_item *new_root_item;
845 struct btrfs_root *tree_root = fs_info->tree_root;
846 struct btrfs_root *root = pending->root;
847 struct btrfs_root *parent_root;
848 struct btrfs_block_rsv *rsv;
849 struct inode *parent_inode;
850 struct dentry *parent;
851 struct dentry *dentry;
852 struct extent_buffer *tmp;
853 struct extent_buffer *old;
860 rsv = trans->block_rsv;
862 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
863 if (!new_root_item) {
864 pending->error = -ENOMEM;
868 ret = btrfs_find_free_objectid(tree_root, &objectid);
870 pending->error = ret;
874 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
876 if (to_reserve > 0) {
877 ret = btrfs_block_rsv_add(root, &pending->block_rsv,
880 pending->error = ret;
885 key.objectid = objectid;
886 key.offset = (u64)-1;
887 key.type = BTRFS_ROOT_ITEM_KEY;
889 trans->block_rsv = &pending->block_rsv;
891 dentry = pending->dentry;
892 parent = dget_parent(dentry);
893 parent_inode = parent->d_inode;
894 parent_root = BTRFS_I(parent_inode)->root;
895 record_root_in_trans(trans, parent_root);
898 * insert the directory item
900 ret = btrfs_set_inode_index(parent_inode, &index);
902 ret = btrfs_insert_dir_item(trans, parent_root,
903 dentry->d_name.name, dentry->d_name.len,
905 BTRFS_FT_DIR, index);
908 btrfs_i_size_write(parent_inode, parent_inode->i_size +
909 dentry->d_name.len * 2);
910 ret = btrfs_update_inode(trans, parent_root, parent_inode);
914 * pull in the delayed directory update
915 * and the delayed inode item
916 * otherwise we corrupt the FS during
919 ret = btrfs_run_delayed_items(trans, root);
922 record_root_in_trans(trans, root);
923 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
924 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
925 btrfs_check_and_init_root_item(new_root_item);
927 root_flags = btrfs_root_flags(new_root_item);
928 if (pending->readonly)
929 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
931 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
932 btrfs_set_root_flags(new_root_item, root_flags);
934 old = btrfs_lock_root_node(root);
935 btrfs_cow_block(trans, root, old, NULL, 0, &old);
936 btrfs_set_lock_blocking(old);
938 btrfs_copy_root(trans, root, old, &tmp, objectid);
939 btrfs_tree_unlock(old);
940 free_extent_buffer(old);
942 btrfs_set_root_node(new_root_item, tmp);
943 /* record when the snapshot was created in key.offset */
944 key.offset = trans->transid;
945 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
946 btrfs_tree_unlock(tmp);
947 free_extent_buffer(tmp);
951 * insert root back/forward references
953 ret = btrfs_add_root_ref(trans, tree_root, objectid,
954 parent_root->root_key.objectid,
955 btrfs_ino(parent_inode), index,
956 dentry->d_name.name, dentry->d_name.len);
960 key.offset = (u64)-1;
961 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
962 BUG_ON(IS_ERR(pending->snap));
964 btrfs_reloc_post_snapshot(trans, pending);
966 kfree(new_root_item);
967 trans->block_rsv = rsv;
968 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
973 * create all the snapshots we've scheduled for creation
975 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
976 struct btrfs_fs_info *fs_info)
978 struct btrfs_pending_snapshot *pending;
979 struct list_head *head = &trans->transaction->pending_snapshots;
982 list_for_each_entry(pending, head, list) {
983 ret = create_pending_snapshot(trans, fs_info, pending);
989 static void update_super_roots(struct btrfs_root *root)
991 struct btrfs_root_item *root_item;
992 struct btrfs_super_block *super;
994 super = &root->fs_info->super_copy;
996 root_item = &root->fs_info->chunk_root->root_item;
997 super->chunk_root = root_item->bytenr;
998 super->chunk_root_generation = root_item->generation;
999 super->chunk_root_level = root_item->level;
1001 root_item = &root->fs_info->tree_root->root_item;
1002 super->root = root_item->bytenr;
1003 super->generation = root_item->generation;
1004 super->root_level = root_item->level;
1005 if (btrfs_test_opt(root, SPACE_CACHE))
1006 super->cache_generation = root_item->generation;
1009 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1012 spin_lock(&info->trans_lock);
1013 if (info->running_transaction)
1014 ret = info->running_transaction->in_commit;
1015 spin_unlock(&info->trans_lock);
1019 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1022 spin_lock(&info->trans_lock);
1023 if (info->running_transaction)
1024 ret = info->running_transaction->blocked;
1025 spin_unlock(&info->trans_lock);
1030 * wait for the current transaction commit to start and block subsequent
1033 static void wait_current_trans_commit_start(struct btrfs_root *root,
1034 struct btrfs_transaction *trans)
1036 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1040 * wait for the current transaction to start and then become unblocked.
1043 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1044 struct btrfs_transaction *trans)
1046 wait_event(root->fs_info->transaction_wait,
1047 trans->commit_done || (trans->in_commit && !trans->blocked));
1051 * commit transactions asynchronously. once btrfs_commit_transaction_async
1052 * returns, any subsequent transaction will not be allowed to join.
1054 struct btrfs_async_commit {
1055 struct btrfs_trans_handle *newtrans;
1056 struct btrfs_root *root;
1057 struct delayed_work work;
1060 static void do_async_commit(struct work_struct *work)
1062 struct btrfs_async_commit *ac =
1063 container_of(work, struct btrfs_async_commit, work.work);
1065 btrfs_commit_transaction(ac->newtrans, ac->root);
1069 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1070 struct btrfs_root *root,
1071 int wait_for_unblock)
1073 struct btrfs_async_commit *ac;
1074 struct btrfs_transaction *cur_trans;
1076 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1080 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1082 ac->newtrans = btrfs_join_transaction(root);
1083 if (IS_ERR(ac->newtrans)) {
1084 int err = PTR_ERR(ac->newtrans);
1089 /* take transaction reference */
1090 cur_trans = trans->transaction;
1091 atomic_inc(&cur_trans->use_count);
1093 btrfs_end_transaction(trans, root);
1094 schedule_delayed_work(&ac->work, 0);
1096 /* wait for transaction to start and unblock */
1097 if (wait_for_unblock)
1098 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1100 wait_current_trans_commit_start(root, cur_trans);
1102 if (current->journal_info == trans)
1103 current->journal_info = NULL;
1105 put_transaction(cur_trans);
1110 * btrfs_transaction state sequence:
1111 * in_commit = 0, blocked = 0 (initial)
1112 * in_commit = 1, blocked = 1
1116 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1117 struct btrfs_root *root)
1119 unsigned long joined = 0;
1120 struct btrfs_transaction *cur_trans;
1121 struct btrfs_transaction *prev_trans = NULL;
1124 int should_grow = 0;
1125 unsigned long now = get_seconds();
1126 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1128 btrfs_run_ordered_operations(root, 0);
1130 btrfs_trans_release_metadata(trans, root);
1131 trans->block_rsv = NULL;
1133 /* make a pass through all the delayed refs we have so far
1134 * any runnings procs may add more while we are here
1136 ret = btrfs_run_delayed_refs(trans, root, 0);
1139 cur_trans = trans->transaction;
1141 * set the flushing flag so procs in this transaction have to
1142 * start sending their work down.
1144 cur_trans->delayed_refs.flushing = 1;
1146 ret = btrfs_run_delayed_refs(trans, root, 0);
1149 spin_lock(&cur_trans->commit_lock);
1150 if (cur_trans->in_commit) {
1151 spin_unlock(&cur_trans->commit_lock);
1152 atomic_inc(&cur_trans->use_count);
1153 btrfs_end_transaction(trans, root);
1155 wait_for_commit(root, cur_trans);
1157 put_transaction(cur_trans);
1162 trans->transaction->in_commit = 1;
1163 trans->transaction->blocked = 1;
1164 spin_unlock(&cur_trans->commit_lock);
1165 wake_up(&root->fs_info->transaction_blocked_wait);
1167 spin_lock(&root->fs_info->trans_lock);
1168 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1169 prev_trans = list_entry(cur_trans->list.prev,
1170 struct btrfs_transaction, list);
1171 if (!prev_trans->commit_done) {
1172 atomic_inc(&prev_trans->use_count);
1173 spin_unlock(&root->fs_info->trans_lock);
1175 wait_for_commit(root, prev_trans);
1177 put_transaction(prev_trans);
1179 spin_unlock(&root->fs_info->trans_lock);
1182 spin_unlock(&root->fs_info->trans_lock);
1185 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1189 int snap_pending = 0;
1191 joined = cur_trans->num_joined;
1192 if (!list_empty(&trans->transaction->pending_snapshots))
1195 WARN_ON(cur_trans != trans->transaction);
1197 if (flush_on_commit || snap_pending) {
1198 btrfs_start_delalloc_inodes(root, 1);
1199 ret = btrfs_wait_ordered_extents(root, 0, 1);
1203 ret = btrfs_run_delayed_items(trans, root);
1207 * rename don't use btrfs_join_transaction, so, once we
1208 * set the transaction to blocked above, we aren't going
1209 * to get any new ordered operations. We can safely run
1210 * it here and no for sure that nothing new will be added
1213 btrfs_run_ordered_operations(root, 1);
1215 prepare_to_wait(&cur_trans->writer_wait, &wait,
1216 TASK_UNINTERRUPTIBLE);
1218 if (atomic_read(&cur_trans->num_writers) > 1)
1219 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1220 else if (should_grow)
1221 schedule_timeout(1);
1223 finish_wait(&cur_trans->writer_wait, &wait);
1224 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1225 (should_grow && cur_trans->num_joined != joined));
1228 * Ok now we need to make sure to block out any other joins while we
1229 * commit the transaction. We could have started a join before setting
1230 * no_join so make sure to wait for num_writers to == 1 again.
1232 spin_lock(&root->fs_info->trans_lock);
1233 root->fs_info->trans_no_join = 1;
1234 spin_unlock(&root->fs_info->trans_lock);
1235 wait_event(cur_trans->writer_wait,
1236 atomic_read(&cur_trans->num_writers) == 1);
1239 * the reloc mutex makes sure that we stop
1240 * the balancing code from coming in and moving
1241 * extents around in the middle of the commit
1243 mutex_lock(&root->fs_info->reloc_mutex);
1245 ret = btrfs_run_delayed_items(trans, root);
1248 ret = create_pending_snapshots(trans, root->fs_info);
1251 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1255 * make sure none of the code above managed to slip in a
1258 btrfs_assert_delayed_root_empty(root);
1260 WARN_ON(cur_trans != trans->transaction);
1262 btrfs_scrub_pause(root);
1263 /* btrfs_commit_tree_roots is responsible for getting the
1264 * various roots consistent with each other. Every pointer
1265 * in the tree of tree roots has to point to the most up to date
1266 * root for every subvolume and other tree. So, we have to keep
1267 * the tree logging code from jumping in and changing any
1270 * At this point in the commit, there can't be any tree-log
1271 * writers, but a little lower down we drop the trans mutex
1272 * and let new people in. By holding the tree_log_mutex
1273 * from now until after the super is written, we avoid races
1274 * with the tree-log code.
1276 mutex_lock(&root->fs_info->tree_log_mutex);
1278 ret = commit_fs_roots(trans, root);
1281 /* commit_fs_roots gets rid of all the tree log roots, it is now
1282 * safe to free the root of tree log roots
1284 btrfs_free_log_root_tree(trans, root->fs_info);
1286 ret = commit_cowonly_roots(trans, root);
1289 btrfs_prepare_extent_commit(trans, root);
1291 cur_trans = root->fs_info->running_transaction;
1293 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1294 root->fs_info->tree_root->node);
1295 switch_commit_root(root->fs_info->tree_root);
1297 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1298 root->fs_info->chunk_root->node);
1299 switch_commit_root(root->fs_info->chunk_root);
1301 update_super_roots(root);
1303 if (!root->fs_info->log_root_recovering) {
1304 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1305 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1308 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1309 sizeof(root->fs_info->super_copy));
1311 trans->transaction->blocked = 0;
1312 spin_lock(&root->fs_info->trans_lock);
1313 root->fs_info->running_transaction = NULL;
1314 root->fs_info->trans_no_join = 0;
1315 spin_unlock(&root->fs_info->trans_lock);
1316 mutex_unlock(&root->fs_info->reloc_mutex);
1318 wake_up(&root->fs_info->transaction_wait);
1320 ret = btrfs_write_and_wait_transaction(trans, root);
1322 write_ctree_super(trans, root, 0);
1325 * the super is written, we can safely allow the tree-loggers
1326 * to go about their business
1328 mutex_unlock(&root->fs_info->tree_log_mutex);
1330 btrfs_finish_extent_commit(trans, root);
1332 cur_trans->commit_done = 1;
1334 root->fs_info->last_trans_committed = cur_trans->transid;
1336 wake_up(&cur_trans->commit_wait);
1338 spin_lock(&root->fs_info->trans_lock);
1339 list_del_init(&cur_trans->list);
1340 spin_unlock(&root->fs_info->trans_lock);
1342 put_transaction(cur_trans);
1343 put_transaction(cur_trans);
1345 trace_btrfs_transaction_commit(root);
1347 btrfs_scrub_continue(root);
1349 if (current->journal_info == trans)
1350 current->journal_info = NULL;
1352 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1354 if (current != root->fs_info->transaction_kthread)
1355 btrfs_run_delayed_iputs(root);
1361 * interface function to delete all the snapshots we have scheduled for deletion
1363 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1366 struct btrfs_fs_info *fs_info = root->fs_info;
1368 spin_lock(&fs_info->trans_lock);
1369 list_splice_init(&fs_info->dead_roots, &list);
1370 spin_unlock(&fs_info->trans_lock);
1372 while (!list_empty(&list)) {
1373 root = list_entry(list.next, struct btrfs_root, root_list);
1374 list_del(&root->root_list);
1376 btrfs_kill_all_delayed_nodes(root);
1378 if (btrfs_header_backref_rev(root->node) <
1379 BTRFS_MIXED_BACKREF_REV)
1380 btrfs_drop_snapshot(root, NULL, 0);
1382 btrfs_drop_snapshot(root, NULL, 1);