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;
187 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
188 return ERR_PTR(-EROFS);
190 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
192 return ERR_PTR(-ENOMEM);
194 if (type != TRANS_JOIN_NOLOCK)
195 mutex_lock(&root->fs_info->trans_mutex);
196 if (may_wait_transaction(root, type))
197 wait_current_trans(root);
199 ret = join_transaction(root);
201 kmem_cache_free(btrfs_trans_handle_cachep, h);
202 if (type != TRANS_JOIN_NOLOCK)
203 mutex_unlock(&root->fs_info->trans_mutex);
207 cur_trans = root->fs_info->running_transaction;
208 cur_trans->use_count++;
209 if (type != TRANS_JOIN_NOLOCK)
210 mutex_unlock(&root->fs_info->trans_mutex);
212 h->transid = cur_trans->transid;
213 h->transaction = cur_trans;
216 h->bytes_reserved = 0;
217 h->delayed_ref_updates = 0;
221 if (cur_trans->blocked && may_wait_transaction(root, type)) {
222 btrfs_commit_transaction(h, root);
227 ret = btrfs_trans_reserve_metadata(h, root, num_items);
228 if (ret == -EAGAIN && !retries) {
230 btrfs_commit_transaction(h, root);
232 } else if (ret == -EAGAIN) {
234 * We have already retried and got EAGAIN, so really we
235 * don't have space, so set ret to -ENOSPC.
241 btrfs_end_transaction(h, root);
246 if (type != TRANS_JOIN_NOLOCK)
247 mutex_lock(&root->fs_info->trans_mutex);
248 record_root_in_trans(h, root);
249 if (type != TRANS_JOIN_NOLOCK)
250 mutex_unlock(&root->fs_info->trans_mutex);
252 if (!current->journal_info && type != TRANS_USERSPACE)
253 current->journal_info = h;
257 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
260 return start_transaction(root, num_items, TRANS_START);
262 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
265 return start_transaction(root, 0, TRANS_JOIN);
268 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
271 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
274 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
277 return start_transaction(r, 0, TRANS_USERSPACE);
280 /* wait for a transaction commit to be fully complete */
281 static noinline int wait_for_commit(struct btrfs_root *root,
282 struct btrfs_transaction *commit)
285 mutex_lock(&root->fs_info->trans_mutex);
286 while (!commit->commit_done) {
287 prepare_to_wait(&commit->commit_wait, &wait,
288 TASK_UNINTERRUPTIBLE);
289 if (commit->commit_done)
291 mutex_unlock(&root->fs_info->trans_mutex);
293 mutex_lock(&root->fs_info->trans_mutex);
295 mutex_unlock(&root->fs_info->trans_mutex);
296 finish_wait(&commit->commit_wait, &wait);
300 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
302 struct btrfs_transaction *cur_trans = NULL, *t;
305 mutex_lock(&root->fs_info->trans_mutex);
309 if (transid <= root->fs_info->last_trans_committed)
312 /* find specified transaction */
313 list_for_each_entry(t, &root->fs_info->trans_list, list) {
314 if (t->transid == transid) {
318 if (t->transid > transid)
323 goto out_unlock; /* bad transid */
325 /* find newest transaction that is committing | committed */
326 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
336 goto out_unlock; /* nothing committing|committed */
339 cur_trans->use_count++;
340 mutex_unlock(&root->fs_info->trans_mutex);
342 wait_for_commit(root, cur_trans);
344 mutex_lock(&root->fs_info->trans_mutex);
345 put_transaction(cur_trans);
348 mutex_unlock(&root->fs_info->trans_mutex);
354 * rate limit against the drop_snapshot code. This helps to slow down new
355 * operations if the drop_snapshot code isn't able to keep up.
357 static void throttle_on_drops(struct btrfs_root *root)
359 struct btrfs_fs_info *info = root->fs_info;
360 int harder_count = 0;
363 if (atomic_read(&info->throttles)) {
366 thr = atomic_read(&info->throttle_gen);
369 prepare_to_wait(&info->transaction_throttle,
370 &wait, TASK_UNINTERRUPTIBLE);
371 if (!atomic_read(&info->throttles)) {
372 finish_wait(&info->transaction_throttle, &wait);
376 finish_wait(&info->transaction_throttle, &wait);
377 } while (thr == atomic_read(&info->throttle_gen));
380 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
384 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
388 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
395 void btrfs_throttle(struct btrfs_root *root)
397 mutex_lock(&root->fs_info->trans_mutex);
398 if (!root->fs_info->open_ioctl_trans)
399 wait_current_trans(root);
400 mutex_unlock(&root->fs_info->trans_mutex);
403 static int should_end_transaction(struct btrfs_trans_handle *trans,
404 struct btrfs_root *root)
407 ret = btrfs_block_rsv_check(trans, root,
408 &root->fs_info->global_block_rsv, 0, 5);
412 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
413 struct btrfs_root *root)
415 struct btrfs_transaction *cur_trans = trans->transaction;
418 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
421 updates = trans->delayed_ref_updates;
422 trans->delayed_ref_updates = 0;
424 btrfs_run_delayed_refs(trans, root, updates);
426 return should_end_transaction(trans, root);
429 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
430 struct btrfs_root *root, int throttle, int lock)
432 struct btrfs_transaction *cur_trans = trans->transaction;
433 struct btrfs_fs_info *info = root->fs_info;
437 unsigned long cur = trans->delayed_ref_updates;
438 trans->delayed_ref_updates = 0;
440 trans->transaction->delayed_refs.num_heads_ready > 64) {
441 trans->delayed_ref_updates = 0;
444 * do a full flush if the transaction is trying
447 if (trans->transaction->delayed_refs.flushing)
449 btrfs_run_delayed_refs(trans, root, cur);
456 btrfs_trans_release_metadata(trans, root);
458 if (lock && !root->fs_info->open_ioctl_trans &&
459 should_end_transaction(trans, root))
460 trans->transaction->blocked = 1;
462 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
464 return btrfs_commit_transaction(trans, root);
466 wake_up_process(info->transaction_kthread);
470 mutex_lock(&info->trans_mutex);
471 WARN_ON(cur_trans != info->running_transaction);
472 WARN_ON(cur_trans->num_writers < 1);
473 cur_trans->num_writers--;
476 if (waitqueue_active(&cur_trans->writer_wait))
477 wake_up(&cur_trans->writer_wait);
478 put_transaction(cur_trans);
480 mutex_unlock(&info->trans_mutex);
482 if (current->journal_info == trans)
483 current->journal_info = NULL;
484 memset(trans, 0, sizeof(*trans));
485 kmem_cache_free(btrfs_trans_handle_cachep, trans);
488 btrfs_run_delayed_iputs(root);
493 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
494 struct btrfs_root *root)
496 return __btrfs_end_transaction(trans, root, 0, 1);
499 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root)
502 return __btrfs_end_transaction(trans, root, 1, 1);
505 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
506 struct btrfs_root *root)
508 return __btrfs_end_transaction(trans, root, 0, 0);
512 * when btree blocks are allocated, they have some corresponding bits set for
513 * them in one of two extent_io trees. This is used to make sure all of
514 * those extents are sent to disk but does not wait on them
516 int btrfs_write_marked_extents(struct btrfs_root *root,
517 struct extent_io_tree *dirty_pages, int mark)
523 struct inode *btree_inode = root->fs_info->btree_inode;
529 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
533 while (start <= end) {
536 index = start >> PAGE_CACHE_SHIFT;
537 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
538 page = find_get_page(btree_inode->i_mapping, index);
542 btree_lock_page_hook(page);
543 if (!page->mapping) {
545 page_cache_release(page);
549 if (PageWriteback(page)) {
551 wait_on_page_writeback(page);
554 page_cache_release(page);
558 err = write_one_page(page, 0);
561 page_cache_release(page);
570 * when btree blocks are allocated, they have some corresponding bits set for
571 * them in one of two extent_io trees. This is used to make sure all of
572 * those extents are on disk for transaction or log commit. We wait
573 * on all the pages and clear them from the dirty pages state tree
575 int btrfs_wait_marked_extents(struct btrfs_root *root,
576 struct extent_io_tree *dirty_pages, int mark)
582 struct inode *btree_inode = root->fs_info->btree_inode;
588 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
593 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
594 while (start <= end) {
595 index = start >> PAGE_CACHE_SHIFT;
596 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
597 page = find_get_page(btree_inode->i_mapping, index);
600 if (PageDirty(page)) {
601 btree_lock_page_hook(page);
602 wait_on_page_writeback(page);
603 err = write_one_page(page, 0);
607 wait_on_page_writeback(page);
608 page_cache_release(page);
618 * when btree blocks are allocated, they have some corresponding bits set for
619 * them in one of two extent_io trees. This is used to make sure all of
620 * those extents are on disk for transaction or log commit
622 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
623 struct extent_io_tree *dirty_pages, int mark)
628 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
629 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
633 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
634 struct btrfs_root *root)
636 if (!trans || !trans->transaction) {
637 struct inode *btree_inode;
638 btree_inode = root->fs_info->btree_inode;
639 return filemap_write_and_wait(btree_inode->i_mapping);
641 return btrfs_write_and_wait_marked_extents(root,
642 &trans->transaction->dirty_pages,
647 * this is used to update the root pointer in the tree of tree roots.
649 * But, in the case of the extent allocation tree, updating the root
650 * pointer may allocate blocks which may change the root of the extent
653 * So, this loops and repeats and makes sure the cowonly root didn't
654 * change while the root pointer was being updated in the metadata.
656 static int update_cowonly_root(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root)
662 struct btrfs_root *tree_root = root->fs_info->tree_root;
664 old_root_used = btrfs_root_used(&root->root_item);
665 btrfs_write_dirty_block_groups(trans, root);
668 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
669 if (old_root_bytenr == root->node->start &&
670 old_root_used == btrfs_root_used(&root->root_item))
673 btrfs_set_root_node(&root->root_item, root->node);
674 ret = btrfs_update_root(trans, tree_root,
679 old_root_used = btrfs_root_used(&root->root_item);
680 ret = btrfs_write_dirty_block_groups(trans, root);
684 if (root != root->fs_info->extent_root)
685 switch_commit_root(root);
691 * update all the cowonly tree roots on disk
693 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root)
696 struct btrfs_fs_info *fs_info = root->fs_info;
697 struct list_head *next;
698 struct extent_buffer *eb;
701 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
704 eb = btrfs_lock_root_node(fs_info->tree_root);
705 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
706 btrfs_tree_unlock(eb);
707 free_extent_buffer(eb);
709 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
712 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
713 next = fs_info->dirty_cowonly_roots.next;
715 root = list_entry(next, struct btrfs_root, dirty_list);
717 update_cowonly_root(trans, root);
720 down_write(&fs_info->extent_commit_sem);
721 switch_commit_root(fs_info->extent_root);
722 up_write(&fs_info->extent_commit_sem);
728 * dead roots are old snapshots that need to be deleted. This allocates
729 * a dirty root struct and adds it into the list of dead roots that need to
732 int btrfs_add_dead_root(struct btrfs_root *root)
734 mutex_lock(&root->fs_info->trans_mutex);
735 list_add(&root->root_list, &root->fs_info->dead_roots);
736 mutex_unlock(&root->fs_info->trans_mutex);
741 * update all the cowonly tree roots on disk
743 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root)
746 struct btrfs_root *gang[8];
747 struct btrfs_fs_info *fs_info = root->fs_info;
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);
765 btrfs_free_log(trans, root);
766 btrfs_update_reloc_root(trans, root);
767 btrfs_orphan_commit_root(trans, root);
769 if (root->commit_root != root->node) {
770 switch_commit_root(root);
771 btrfs_set_root_node(&root->root_item,
775 err = btrfs_update_root(trans, fs_info->tree_root,
786 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
787 * otherwise every leaf in the btree is read and defragged.
789 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
791 struct btrfs_fs_info *info = root->fs_info;
792 struct btrfs_trans_handle *trans;
796 if (xchg(&root->defrag_running, 1))
800 trans = btrfs_start_transaction(root, 0);
802 return PTR_ERR(trans);
804 ret = btrfs_defrag_leaves(trans, root, cacheonly);
806 nr = trans->blocks_used;
807 btrfs_end_transaction(trans, root);
808 btrfs_btree_balance_dirty(info->tree_root, nr);
811 if (root->fs_info->closing || ret != -EAGAIN)
814 root->defrag_running = 0;
820 * when dropping snapshots, we generate a ton of delayed refs, and it makes
821 * sense not to join the transaction while it is trying to flush the current
822 * queue of delayed refs out.
824 * This is used by the drop snapshot code only
826 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
830 mutex_lock(&info->trans_mutex);
831 while (info->running_transaction &&
832 info->running_transaction->delayed_refs.flushing) {
833 prepare_to_wait(&info->transaction_wait, &wait,
834 TASK_UNINTERRUPTIBLE);
835 mutex_unlock(&info->trans_mutex);
839 mutex_lock(&info->trans_mutex);
840 finish_wait(&info->transaction_wait, &wait);
842 mutex_unlock(&info->trans_mutex);
847 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
850 int btrfs_drop_dead_root(struct btrfs_root *root)
852 struct btrfs_trans_handle *trans;
853 struct btrfs_root *tree_root = root->fs_info->tree_root;
859 * we don't want to jump in and create a bunch of
860 * delayed refs if the transaction is starting to close
862 wait_transaction_pre_flush(tree_root->fs_info);
863 trans = btrfs_start_transaction(tree_root, 1);
866 * we've joined a transaction, make sure it isn't
869 if (trans->transaction->delayed_refs.flushing) {
870 btrfs_end_transaction(trans, tree_root);
874 ret = btrfs_drop_snapshot(trans, root);
878 ret = btrfs_update_root(trans, tree_root,
884 nr = trans->blocks_used;
885 ret = btrfs_end_transaction(trans, tree_root);
888 btrfs_btree_balance_dirty(tree_root, nr);
893 ret = btrfs_del_root(trans, tree_root, &root->root_key);
896 nr = trans->blocks_used;
897 ret = btrfs_end_transaction(trans, tree_root);
900 free_extent_buffer(root->node);
901 free_extent_buffer(root->commit_root);
904 btrfs_btree_balance_dirty(tree_root, nr);
910 * new snapshots need to be created at a very specific time in the
911 * transaction commit. This does the actual creation
913 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
914 struct btrfs_fs_info *fs_info,
915 struct btrfs_pending_snapshot *pending)
917 struct btrfs_key key;
918 struct btrfs_root_item *new_root_item;
919 struct btrfs_root *tree_root = fs_info->tree_root;
920 struct btrfs_root *root = pending->root;
921 struct btrfs_root *parent_root;
922 struct inode *parent_inode;
923 struct dentry *parent;
924 struct dentry *dentry;
925 struct extent_buffer *tmp;
926 struct extent_buffer *old;
933 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
934 if (!new_root_item) {
935 pending->error = -ENOMEM;
939 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
941 pending->error = ret;
945 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
946 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
948 if (to_reserve > 0) {
949 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
952 pending->error = ret;
957 key.objectid = objectid;
958 key.offset = (u64)-1;
959 key.type = BTRFS_ROOT_ITEM_KEY;
961 trans->block_rsv = &pending->block_rsv;
963 dentry = pending->dentry;
964 parent = dget_parent(dentry);
965 parent_inode = parent->d_inode;
966 parent_root = BTRFS_I(parent_inode)->root;
967 record_root_in_trans(trans, parent_root);
970 * insert the directory item
972 ret = btrfs_set_inode_index(parent_inode, &index);
974 ret = btrfs_insert_dir_item(trans, parent_root,
975 dentry->d_name.name, dentry->d_name.len,
976 parent_inode->i_ino, &key,
977 BTRFS_FT_DIR, index);
980 btrfs_i_size_write(parent_inode, parent_inode->i_size +
981 dentry->d_name.len * 2);
982 ret = btrfs_update_inode(trans, parent_root, parent_inode);
985 record_root_in_trans(trans, root);
986 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
987 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
988 btrfs_check_and_init_root_item(new_root_item);
990 root_flags = btrfs_root_flags(new_root_item);
991 if (pending->readonly)
992 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
994 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
995 btrfs_set_root_flags(new_root_item, root_flags);
997 old = btrfs_lock_root_node(root);
998 btrfs_cow_block(trans, root, old, NULL, 0, &old);
999 btrfs_set_lock_blocking(old);
1001 btrfs_copy_root(trans, root, old, &tmp, objectid);
1002 btrfs_tree_unlock(old);
1003 free_extent_buffer(old);
1005 btrfs_set_root_node(new_root_item, tmp);
1006 /* record when the snapshot was created in key.offset */
1007 key.offset = trans->transid;
1008 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1009 btrfs_tree_unlock(tmp);
1010 free_extent_buffer(tmp);
1014 * insert root back/forward references
1016 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1017 parent_root->root_key.objectid,
1018 parent_inode->i_ino, index,
1019 dentry->d_name.name, dentry->d_name.len);
1023 key.offset = (u64)-1;
1024 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1025 BUG_ON(IS_ERR(pending->snap));
1027 btrfs_reloc_post_snapshot(trans, pending);
1028 btrfs_orphan_post_snapshot(trans, pending);
1030 kfree(new_root_item);
1031 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1036 * create all the snapshots we've scheduled for creation
1038 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1039 struct btrfs_fs_info *fs_info)
1041 struct btrfs_pending_snapshot *pending;
1042 struct list_head *head = &trans->transaction->pending_snapshots;
1045 list_for_each_entry(pending, head, list) {
1046 ret = create_pending_snapshot(trans, fs_info, pending);
1052 static void update_super_roots(struct btrfs_root *root)
1054 struct btrfs_root_item *root_item;
1055 struct btrfs_super_block *super;
1057 super = &root->fs_info->super_copy;
1059 root_item = &root->fs_info->chunk_root->root_item;
1060 super->chunk_root = root_item->bytenr;
1061 super->chunk_root_generation = root_item->generation;
1062 super->chunk_root_level = root_item->level;
1064 root_item = &root->fs_info->tree_root->root_item;
1065 super->root = root_item->bytenr;
1066 super->generation = root_item->generation;
1067 super->root_level = root_item->level;
1068 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1069 super->cache_generation = root_item->generation;
1072 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1075 spin_lock(&info->new_trans_lock);
1076 if (info->running_transaction)
1077 ret = info->running_transaction->in_commit;
1078 spin_unlock(&info->new_trans_lock);
1082 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1085 spin_lock(&info->new_trans_lock);
1086 if (info->running_transaction)
1087 ret = info->running_transaction->blocked;
1088 spin_unlock(&info->new_trans_lock);
1093 * wait for the current transaction commit to start and block subsequent
1096 static void wait_current_trans_commit_start(struct btrfs_root *root,
1097 struct btrfs_transaction *trans)
1101 if (trans->in_commit)
1105 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1106 TASK_UNINTERRUPTIBLE);
1107 if (trans->in_commit) {
1108 finish_wait(&root->fs_info->transaction_blocked_wait,
1112 mutex_unlock(&root->fs_info->trans_mutex);
1114 mutex_lock(&root->fs_info->trans_mutex);
1115 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1120 * wait for the current transaction to start and then become unblocked.
1123 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1124 struct btrfs_transaction *trans)
1128 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1132 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1133 TASK_UNINTERRUPTIBLE);
1134 if (trans->commit_done ||
1135 (trans->in_commit && !trans->blocked)) {
1136 finish_wait(&root->fs_info->transaction_wait,
1140 mutex_unlock(&root->fs_info->trans_mutex);
1142 mutex_lock(&root->fs_info->trans_mutex);
1143 finish_wait(&root->fs_info->transaction_wait,
1149 * commit transactions asynchronously. once btrfs_commit_transaction_async
1150 * returns, any subsequent transaction will not be allowed to join.
1152 struct btrfs_async_commit {
1153 struct btrfs_trans_handle *newtrans;
1154 struct btrfs_root *root;
1155 struct delayed_work work;
1158 static void do_async_commit(struct work_struct *work)
1160 struct btrfs_async_commit *ac =
1161 container_of(work, struct btrfs_async_commit, work.work);
1163 btrfs_commit_transaction(ac->newtrans, ac->root);
1167 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1168 struct btrfs_root *root,
1169 int wait_for_unblock)
1171 struct btrfs_async_commit *ac;
1172 struct btrfs_transaction *cur_trans;
1174 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1178 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1180 ac->newtrans = btrfs_join_transaction(root, 0);
1181 if (IS_ERR(ac->newtrans)) {
1182 int err = PTR_ERR(ac->newtrans);
1187 /* take transaction reference */
1188 mutex_lock(&root->fs_info->trans_mutex);
1189 cur_trans = trans->transaction;
1190 cur_trans->use_count++;
1191 mutex_unlock(&root->fs_info->trans_mutex);
1193 btrfs_end_transaction(trans, root);
1194 schedule_delayed_work(&ac->work, 0);
1196 /* wait for transaction to start and unblock */
1197 mutex_lock(&root->fs_info->trans_mutex);
1198 if (wait_for_unblock)
1199 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1201 wait_current_trans_commit_start(root, cur_trans);
1202 put_transaction(cur_trans);
1203 mutex_unlock(&root->fs_info->trans_mutex);
1209 * btrfs_transaction state sequence:
1210 * in_commit = 0, blocked = 0 (initial)
1211 * in_commit = 1, blocked = 1
1215 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1216 struct btrfs_root *root)
1218 unsigned long joined = 0;
1219 struct btrfs_transaction *cur_trans;
1220 struct btrfs_transaction *prev_trans = NULL;
1223 int should_grow = 0;
1224 unsigned long now = get_seconds();
1225 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1227 btrfs_run_ordered_operations(root, 0);
1229 /* make a pass through all the delayed refs we have so far
1230 * any runnings procs may add more while we are here
1232 ret = btrfs_run_delayed_refs(trans, root, 0);
1235 btrfs_trans_release_metadata(trans, root);
1237 cur_trans = trans->transaction;
1239 * set the flushing flag so procs in this transaction have to
1240 * start sending their work down.
1242 cur_trans->delayed_refs.flushing = 1;
1244 ret = btrfs_run_delayed_refs(trans, root, 0);
1247 mutex_lock(&root->fs_info->trans_mutex);
1248 if (cur_trans->in_commit) {
1249 cur_trans->use_count++;
1250 mutex_unlock(&root->fs_info->trans_mutex);
1251 btrfs_end_transaction(trans, root);
1253 ret = wait_for_commit(root, cur_trans);
1256 mutex_lock(&root->fs_info->trans_mutex);
1257 put_transaction(cur_trans);
1258 mutex_unlock(&root->fs_info->trans_mutex);
1263 trans->transaction->in_commit = 1;
1264 trans->transaction->blocked = 1;
1265 wake_up(&root->fs_info->transaction_blocked_wait);
1267 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1268 prev_trans = list_entry(cur_trans->list.prev,
1269 struct btrfs_transaction, list);
1270 if (!prev_trans->commit_done) {
1271 prev_trans->use_count++;
1272 mutex_unlock(&root->fs_info->trans_mutex);
1274 wait_for_commit(root, prev_trans);
1276 mutex_lock(&root->fs_info->trans_mutex);
1277 put_transaction(prev_trans);
1281 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1285 int snap_pending = 0;
1286 joined = cur_trans->num_joined;
1287 if (!list_empty(&trans->transaction->pending_snapshots))
1290 WARN_ON(cur_trans != trans->transaction);
1291 mutex_unlock(&root->fs_info->trans_mutex);
1293 if (flush_on_commit || snap_pending) {
1294 btrfs_start_delalloc_inodes(root, 1);
1295 ret = btrfs_wait_ordered_extents(root, 0, 1);
1300 * rename don't use btrfs_join_transaction, so, once we
1301 * set the transaction to blocked above, we aren't going
1302 * to get any new ordered operations. We can safely run
1303 * it here and no for sure that nothing new will be added
1306 btrfs_run_ordered_operations(root, 1);
1308 prepare_to_wait(&cur_trans->writer_wait, &wait,
1309 TASK_UNINTERRUPTIBLE);
1312 if (cur_trans->num_writers > 1)
1313 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1314 else if (should_grow)
1315 schedule_timeout(1);
1317 mutex_lock(&root->fs_info->trans_mutex);
1318 finish_wait(&cur_trans->writer_wait, &wait);
1319 } while (cur_trans->num_writers > 1 ||
1320 (should_grow && cur_trans->num_joined != joined));
1322 ret = create_pending_snapshots(trans, root->fs_info);
1325 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1328 WARN_ON(cur_trans != trans->transaction);
1330 /* btrfs_commit_tree_roots is responsible for getting the
1331 * various roots consistent with each other. Every pointer
1332 * in the tree of tree roots has to point to the most up to date
1333 * root for every subvolume and other tree. So, we have to keep
1334 * the tree logging code from jumping in and changing any
1337 * At this point in the commit, there can't be any tree-log
1338 * writers, but a little lower down we drop the trans mutex
1339 * and let new people in. By holding the tree_log_mutex
1340 * from now until after the super is written, we avoid races
1341 * with the tree-log code.
1343 mutex_lock(&root->fs_info->tree_log_mutex);
1345 ret = commit_fs_roots(trans, root);
1348 /* commit_fs_roots gets rid of all the tree log roots, it is now
1349 * safe to free the root of tree log roots
1351 btrfs_free_log_root_tree(trans, root->fs_info);
1353 ret = commit_cowonly_roots(trans, root);
1356 btrfs_prepare_extent_commit(trans, root);
1358 cur_trans = root->fs_info->running_transaction;
1359 spin_lock(&root->fs_info->new_trans_lock);
1360 root->fs_info->running_transaction = NULL;
1361 spin_unlock(&root->fs_info->new_trans_lock);
1363 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1364 root->fs_info->tree_root->node);
1365 switch_commit_root(root->fs_info->tree_root);
1367 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1368 root->fs_info->chunk_root->node);
1369 switch_commit_root(root->fs_info->chunk_root);
1371 update_super_roots(root);
1373 if (!root->fs_info->log_root_recovering) {
1374 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1375 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1378 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1379 sizeof(root->fs_info->super_copy));
1381 trans->transaction->blocked = 0;
1383 wake_up(&root->fs_info->transaction_wait);
1385 mutex_unlock(&root->fs_info->trans_mutex);
1386 ret = btrfs_write_and_wait_transaction(trans, root);
1388 write_ctree_super(trans, root, 0);
1391 * the super is written, we can safely allow the tree-loggers
1392 * to go about their business
1394 mutex_unlock(&root->fs_info->tree_log_mutex);
1396 btrfs_finish_extent_commit(trans, root);
1398 mutex_lock(&root->fs_info->trans_mutex);
1400 cur_trans->commit_done = 1;
1402 root->fs_info->last_trans_committed = cur_trans->transid;
1404 wake_up(&cur_trans->commit_wait);
1406 put_transaction(cur_trans);
1407 put_transaction(cur_trans);
1409 trace_btrfs_transaction_commit(root);
1411 mutex_unlock(&root->fs_info->trans_mutex);
1413 if (current->journal_info == trans)
1414 current->journal_info = NULL;
1416 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1418 if (current != root->fs_info->transaction_kthread)
1419 btrfs_run_delayed_iputs(root);
1425 * interface function to delete all the snapshots we have scheduled for deletion
1427 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1430 struct btrfs_fs_info *fs_info = root->fs_info;
1432 mutex_lock(&fs_info->trans_mutex);
1433 list_splice_init(&fs_info->dead_roots, &list);
1434 mutex_unlock(&fs_info->trans_mutex);
1436 while (!list_empty(&list)) {
1437 root = list_entry(list.next, struct btrfs_root, root_list);
1438 list_del(&root->root_list);
1440 if (btrfs_header_backref_rev(root->node) <
1441 BTRFS_MIXED_BACKREF_REV)
1442 btrfs_drop_snapshot(root, NULL, 0);
1444 btrfs_drop_snapshot(root, NULL, 1);