2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/slab.h>
21 #include <linux/sched.h>
22 #include <linux/writeback.h>
23 #include <linux/pagemap.h>
24 #include <linux/blkdev.h>
27 #include "transaction.h"
31 #define BTRFS_ROOT_TRANS_TAG 0
33 static noinline void put_transaction(struct btrfs_transaction *transaction)
35 WARN_ON(transaction->use_count == 0);
36 transaction->use_count--;
37 if (transaction->use_count == 0) {
38 list_del_init(&transaction->list);
39 memset(transaction, 0, sizeof(*transaction));
40 kmem_cache_free(btrfs_transaction_cachep, transaction);
44 static noinline void switch_commit_root(struct btrfs_root *root)
46 free_extent_buffer(root->commit_root);
47 root->commit_root = btrfs_root_node(root);
51 * either allocate a new transaction or hop into the existing one
53 static noinline int join_transaction(struct btrfs_root *root)
55 struct btrfs_transaction *cur_trans;
56 cur_trans = root->fs_info->running_transaction;
58 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
62 root->fs_info->generation++;
63 cur_trans->num_writers = 1;
64 cur_trans->num_joined = 0;
65 cur_trans->transid = root->fs_info->generation;
66 init_waitqueue_head(&cur_trans->writer_wait);
67 init_waitqueue_head(&cur_trans->commit_wait);
68 cur_trans->in_commit = 0;
69 cur_trans->blocked = 0;
70 cur_trans->use_count = 1;
71 cur_trans->commit_done = 0;
72 cur_trans->start_time = get_seconds();
74 cur_trans->delayed_refs.root = RB_ROOT;
75 cur_trans->delayed_refs.num_entries = 0;
76 cur_trans->delayed_refs.num_heads_ready = 0;
77 cur_trans->delayed_refs.num_heads = 0;
78 cur_trans->delayed_refs.flushing = 0;
79 cur_trans->delayed_refs.run_delayed_start = 0;
80 spin_lock_init(&cur_trans->delayed_refs.lock);
82 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
83 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
84 extent_io_tree_init(&cur_trans->dirty_pages,
85 root->fs_info->btree_inode->i_mapping,
87 spin_lock(&root->fs_info->new_trans_lock);
88 root->fs_info->running_transaction = cur_trans;
89 spin_unlock(&root->fs_info->new_trans_lock);
91 cur_trans->num_writers++;
92 cur_trans->num_joined++;
99 * this does all the record keeping required to make sure that a reference
100 * counted root is properly recorded in a given transaction. This is required
101 * to make sure the old root from before we joined the transaction is deleted
102 * when the transaction commits
104 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root)
107 if (root->ref_cows && root->last_trans < trans->transid) {
108 WARN_ON(root == root->fs_info->extent_root);
109 WARN_ON(root->commit_root != root->node);
111 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
112 (unsigned long)root->root_key.objectid,
113 BTRFS_ROOT_TRANS_TAG);
114 root->last_trans = trans->transid;
115 btrfs_init_reloc_root(trans, root);
120 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
121 struct btrfs_root *root)
126 mutex_lock(&root->fs_info->trans_mutex);
127 if (root->last_trans == trans->transid) {
128 mutex_unlock(&root->fs_info->trans_mutex);
132 record_root_in_trans(trans, root);
133 mutex_unlock(&root->fs_info->trans_mutex);
137 /* wait for commit against the current transaction to become unblocked
138 * when this is done, it is safe to start a new transaction, but the current
139 * transaction might not be fully on disk.
141 static void wait_current_trans(struct btrfs_root *root)
143 struct btrfs_transaction *cur_trans;
145 cur_trans = root->fs_info->running_transaction;
146 if (cur_trans && cur_trans->blocked) {
148 cur_trans->use_count++;
150 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
151 TASK_UNINTERRUPTIBLE);
152 if (!cur_trans->blocked)
154 mutex_unlock(&root->fs_info->trans_mutex);
156 mutex_lock(&root->fs_info->trans_mutex);
158 finish_wait(&root->fs_info->transaction_wait, &wait);
159 put_transaction(cur_trans);
163 enum btrfs_trans_type {
170 static int may_wait_transaction(struct btrfs_root *root, int type)
172 if (!root->fs_info->log_root_recovering &&
173 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
174 type == TRANS_USERSPACE))
179 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
180 u64 num_items, int type)
182 struct btrfs_trans_handle *h;
183 struct btrfs_transaction *cur_trans;
186 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
187 return ERR_PTR(-EROFS);
189 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
191 return ERR_PTR(-ENOMEM);
193 if (type != TRANS_JOIN_NOLOCK)
194 mutex_lock(&root->fs_info->trans_mutex);
195 if (may_wait_transaction(root, type))
196 wait_current_trans(root);
198 ret = join_transaction(root);
200 kmem_cache_free(btrfs_trans_handle_cachep, h);
201 if (type != TRANS_JOIN_NOLOCK)
202 mutex_unlock(&root->fs_info->trans_mutex);
206 cur_trans = root->fs_info->running_transaction;
207 cur_trans->use_count++;
208 if (type != TRANS_JOIN_NOLOCK)
209 mutex_unlock(&root->fs_info->trans_mutex);
211 h->transid = cur_trans->transid;
212 h->transaction = cur_trans;
215 h->bytes_reserved = 0;
216 h->delayed_ref_updates = 0;
220 if (cur_trans->blocked && may_wait_transaction(root, type)) {
221 btrfs_commit_transaction(h, root);
226 ret = btrfs_trans_reserve_metadata(h, root, num_items);
227 if (ret == -EAGAIN) {
228 btrfs_commit_transaction(h, root);
232 btrfs_end_transaction(h, root);
237 if (type != TRANS_JOIN_NOLOCK)
238 mutex_lock(&root->fs_info->trans_mutex);
239 record_root_in_trans(h, root);
240 if (type != TRANS_JOIN_NOLOCK)
241 mutex_unlock(&root->fs_info->trans_mutex);
243 if (!current->journal_info && type != TRANS_USERSPACE)
244 current->journal_info = h;
248 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
251 return start_transaction(root, num_items, TRANS_START);
253 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
256 return start_transaction(root, 0, TRANS_JOIN);
259 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
262 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
265 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
268 return start_transaction(r, 0, TRANS_USERSPACE);
271 /* wait for a transaction commit to be fully complete */
272 static noinline int wait_for_commit(struct btrfs_root *root,
273 struct btrfs_transaction *commit)
276 mutex_lock(&root->fs_info->trans_mutex);
277 while (!commit->commit_done) {
278 prepare_to_wait(&commit->commit_wait, &wait,
279 TASK_UNINTERRUPTIBLE);
280 if (commit->commit_done)
282 mutex_unlock(&root->fs_info->trans_mutex);
284 mutex_lock(&root->fs_info->trans_mutex);
286 mutex_unlock(&root->fs_info->trans_mutex);
287 finish_wait(&commit->commit_wait, &wait);
291 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
293 struct btrfs_transaction *cur_trans = NULL, *t;
296 mutex_lock(&root->fs_info->trans_mutex);
300 if (transid <= root->fs_info->last_trans_committed)
303 /* find specified transaction */
304 list_for_each_entry(t, &root->fs_info->trans_list, list) {
305 if (t->transid == transid) {
309 if (t->transid > transid)
314 goto out_unlock; /* bad transid */
316 /* find newest transaction that is committing | committed */
317 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
327 goto out_unlock; /* nothing committing|committed */
330 cur_trans->use_count++;
331 mutex_unlock(&root->fs_info->trans_mutex);
333 wait_for_commit(root, cur_trans);
335 mutex_lock(&root->fs_info->trans_mutex);
336 put_transaction(cur_trans);
339 mutex_unlock(&root->fs_info->trans_mutex);
345 * rate limit against the drop_snapshot code. This helps to slow down new
346 * operations if the drop_snapshot code isn't able to keep up.
348 static void throttle_on_drops(struct btrfs_root *root)
350 struct btrfs_fs_info *info = root->fs_info;
351 int harder_count = 0;
354 if (atomic_read(&info->throttles)) {
357 thr = atomic_read(&info->throttle_gen);
360 prepare_to_wait(&info->transaction_throttle,
361 &wait, TASK_UNINTERRUPTIBLE);
362 if (!atomic_read(&info->throttles)) {
363 finish_wait(&info->transaction_throttle, &wait);
367 finish_wait(&info->transaction_throttle, &wait);
368 } while (thr == atomic_read(&info->throttle_gen));
371 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
375 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
379 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
386 void btrfs_throttle(struct btrfs_root *root)
388 mutex_lock(&root->fs_info->trans_mutex);
389 if (!root->fs_info->open_ioctl_trans)
390 wait_current_trans(root);
391 mutex_unlock(&root->fs_info->trans_mutex);
394 static int should_end_transaction(struct btrfs_trans_handle *trans,
395 struct btrfs_root *root)
398 ret = btrfs_block_rsv_check(trans, root,
399 &root->fs_info->global_block_rsv, 0, 5);
403 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
404 struct btrfs_root *root)
406 struct btrfs_transaction *cur_trans = trans->transaction;
409 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
412 updates = trans->delayed_ref_updates;
413 trans->delayed_ref_updates = 0;
415 btrfs_run_delayed_refs(trans, root, updates);
417 return should_end_transaction(trans, root);
420 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
421 struct btrfs_root *root, int throttle, int lock)
423 struct btrfs_transaction *cur_trans = trans->transaction;
424 struct btrfs_fs_info *info = root->fs_info;
428 unsigned long cur = trans->delayed_ref_updates;
429 trans->delayed_ref_updates = 0;
431 trans->transaction->delayed_refs.num_heads_ready > 64) {
432 trans->delayed_ref_updates = 0;
435 * do a full flush if the transaction is trying
438 if (trans->transaction->delayed_refs.flushing)
440 btrfs_run_delayed_refs(trans, root, cur);
447 btrfs_trans_release_metadata(trans, root);
449 if (lock && !root->fs_info->open_ioctl_trans &&
450 should_end_transaction(trans, root))
451 trans->transaction->blocked = 1;
453 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
455 return btrfs_commit_transaction(trans, root);
457 wake_up_process(info->transaction_kthread);
461 mutex_lock(&info->trans_mutex);
462 WARN_ON(cur_trans != info->running_transaction);
463 WARN_ON(cur_trans->num_writers < 1);
464 cur_trans->num_writers--;
467 if (waitqueue_active(&cur_trans->writer_wait))
468 wake_up(&cur_trans->writer_wait);
469 put_transaction(cur_trans);
471 mutex_unlock(&info->trans_mutex);
473 if (current->journal_info == trans)
474 current->journal_info = NULL;
475 memset(trans, 0, sizeof(*trans));
476 kmem_cache_free(btrfs_trans_handle_cachep, trans);
479 btrfs_run_delayed_iputs(root);
484 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
485 struct btrfs_root *root)
487 return __btrfs_end_transaction(trans, root, 0, 1);
490 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
491 struct btrfs_root *root)
493 return __btrfs_end_transaction(trans, root, 1, 1);
496 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
497 struct btrfs_root *root)
499 return __btrfs_end_transaction(trans, root, 0, 0);
503 * when btree blocks are allocated, they have some corresponding bits set for
504 * them in one of two extent_io trees. This is used to make sure all of
505 * those extents are sent to disk but does not wait on them
507 int btrfs_write_marked_extents(struct btrfs_root *root,
508 struct extent_io_tree *dirty_pages, int mark)
514 struct inode *btree_inode = root->fs_info->btree_inode;
520 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
524 while (start <= end) {
527 index = start >> PAGE_CACHE_SHIFT;
528 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
529 page = find_get_page(btree_inode->i_mapping, index);
533 btree_lock_page_hook(page);
534 if (!page->mapping) {
536 page_cache_release(page);
540 if (PageWriteback(page)) {
542 wait_on_page_writeback(page);
545 page_cache_release(page);
549 err = write_one_page(page, 0);
552 page_cache_release(page);
561 * when btree blocks are allocated, they have some corresponding bits set for
562 * them in one of two extent_io trees. This is used to make sure all of
563 * those extents are on disk for transaction or log commit. We wait
564 * on all the pages and clear them from the dirty pages state tree
566 int btrfs_wait_marked_extents(struct btrfs_root *root,
567 struct extent_io_tree *dirty_pages, int mark)
573 struct inode *btree_inode = root->fs_info->btree_inode;
579 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
584 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
585 while (start <= end) {
586 index = start >> PAGE_CACHE_SHIFT;
587 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
588 page = find_get_page(btree_inode->i_mapping, index);
591 if (PageDirty(page)) {
592 btree_lock_page_hook(page);
593 wait_on_page_writeback(page);
594 err = write_one_page(page, 0);
598 wait_on_page_writeback(page);
599 page_cache_release(page);
609 * when btree blocks are allocated, they have some corresponding bits set for
610 * them in one of two extent_io trees. This is used to make sure all of
611 * those extents are on disk for transaction or log commit
613 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
614 struct extent_io_tree *dirty_pages, int mark)
619 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
620 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
624 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root)
627 if (!trans || !trans->transaction) {
628 struct inode *btree_inode;
629 btree_inode = root->fs_info->btree_inode;
630 return filemap_write_and_wait(btree_inode->i_mapping);
632 return btrfs_write_and_wait_marked_extents(root,
633 &trans->transaction->dirty_pages,
638 * this is used to update the root pointer in the tree of tree roots.
640 * But, in the case of the extent allocation tree, updating the root
641 * pointer may allocate blocks which may change the root of the extent
644 * So, this loops and repeats and makes sure the cowonly root didn't
645 * change while the root pointer was being updated in the metadata.
647 static int update_cowonly_root(struct btrfs_trans_handle *trans,
648 struct btrfs_root *root)
653 struct btrfs_root *tree_root = root->fs_info->tree_root;
655 old_root_used = btrfs_root_used(&root->root_item);
656 btrfs_write_dirty_block_groups(trans, root);
659 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
660 if (old_root_bytenr == root->node->start &&
661 old_root_used == btrfs_root_used(&root->root_item))
664 btrfs_set_root_node(&root->root_item, root->node);
665 ret = btrfs_update_root(trans, tree_root,
670 old_root_used = btrfs_root_used(&root->root_item);
671 ret = btrfs_write_dirty_block_groups(trans, root);
675 if (root != root->fs_info->extent_root)
676 switch_commit_root(root);
682 * update all the cowonly tree roots on disk
684 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
685 struct btrfs_root *root)
687 struct btrfs_fs_info *fs_info = root->fs_info;
688 struct list_head *next;
689 struct extent_buffer *eb;
692 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
695 eb = btrfs_lock_root_node(fs_info->tree_root);
696 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
697 btrfs_tree_unlock(eb);
698 free_extent_buffer(eb);
700 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
703 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
704 next = fs_info->dirty_cowonly_roots.next;
706 root = list_entry(next, struct btrfs_root, dirty_list);
708 update_cowonly_root(trans, root);
711 down_write(&fs_info->extent_commit_sem);
712 switch_commit_root(fs_info->extent_root);
713 up_write(&fs_info->extent_commit_sem);
719 * dead roots are old snapshots that need to be deleted. This allocates
720 * a dirty root struct and adds it into the list of dead roots that need to
723 int btrfs_add_dead_root(struct btrfs_root *root)
725 mutex_lock(&root->fs_info->trans_mutex);
726 list_add(&root->root_list, &root->fs_info->dead_roots);
727 mutex_unlock(&root->fs_info->trans_mutex);
732 * update all the cowonly tree roots on disk
734 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root)
737 struct btrfs_root *gang[8];
738 struct btrfs_fs_info *fs_info = root->fs_info;
744 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
747 BTRFS_ROOT_TRANS_TAG);
750 for (i = 0; i < ret; i++) {
752 radix_tree_tag_clear(&fs_info->fs_roots_radix,
753 (unsigned long)root->root_key.objectid,
754 BTRFS_ROOT_TRANS_TAG);
756 btrfs_free_log(trans, root);
757 btrfs_update_reloc_root(trans, root);
758 btrfs_orphan_commit_root(trans, root);
760 if (root->commit_root != root->node) {
761 switch_commit_root(root);
762 btrfs_set_root_node(&root->root_item,
766 err = btrfs_update_root(trans, fs_info->tree_root,
777 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
778 * otherwise every leaf in the btree is read and defragged.
780 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
782 struct btrfs_fs_info *info = root->fs_info;
783 struct btrfs_trans_handle *trans;
787 if (xchg(&root->defrag_running, 1))
791 trans = btrfs_start_transaction(root, 0);
793 return PTR_ERR(trans);
795 ret = btrfs_defrag_leaves(trans, root, cacheonly);
797 nr = trans->blocks_used;
798 btrfs_end_transaction(trans, root);
799 btrfs_btree_balance_dirty(info->tree_root, nr);
802 if (root->fs_info->closing || ret != -EAGAIN)
805 root->defrag_running = 0;
811 * when dropping snapshots, we generate a ton of delayed refs, and it makes
812 * sense not to join the transaction while it is trying to flush the current
813 * queue of delayed refs out.
815 * This is used by the drop snapshot code only
817 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
821 mutex_lock(&info->trans_mutex);
822 while (info->running_transaction &&
823 info->running_transaction->delayed_refs.flushing) {
824 prepare_to_wait(&info->transaction_wait, &wait,
825 TASK_UNINTERRUPTIBLE);
826 mutex_unlock(&info->trans_mutex);
830 mutex_lock(&info->trans_mutex);
831 finish_wait(&info->transaction_wait, &wait);
833 mutex_unlock(&info->trans_mutex);
838 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
841 int btrfs_drop_dead_root(struct btrfs_root *root)
843 struct btrfs_trans_handle *trans;
844 struct btrfs_root *tree_root = root->fs_info->tree_root;
850 * we don't want to jump in and create a bunch of
851 * delayed refs if the transaction is starting to close
853 wait_transaction_pre_flush(tree_root->fs_info);
854 trans = btrfs_start_transaction(tree_root, 1);
857 * we've joined a transaction, make sure it isn't
860 if (trans->transaction->delayed_refs.flushing) {
861 btrfs_end_transaction(trans, tree_root);
865 ret = btrfs_drop_snapshot(trans, root);
869 ret = btrfs_update_root(trans, tree_root,
875 nr = trans->blocks_used;
876 ret = btrfs_end_transaction(trans, tree_root);
879 btrfs_btree_balance_dirty(tree_root, nr);
884 ret = btrfs_del_root(trans, tree_root, &root->root_key);
887 nr = trans->blocks_used;
888 ret = btrfs_end_transaction(trans, tree_root);
891 free_extent_buffer(root->node);
892 free_extent_buffer(root->commit_root);
895 btrfs_btree_balance_dirty(tree_root, nr);
901 * new snapshots need to be created at a very specific time in the
902 * transaction commit. This does the actual creation
904 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
905 struct btrfs_fs_info *fs_info,
906 struct btrfs_pending_snapshot *pending)
908 struct btrfs_key key;
909 struct btrfs_root_item *new_root_item;
910 struct btrfs_root *tree_root = fs_info->tree_root;
911 struct btrfs_root *root = pending->root;
912 struct btrfs_root *parent_root;
913 struct inode *parent_inode;
914 struct dentry *parent;
915 struct dentry *dentry;
916 struct extent_buffer *tmp;
917 struct extent_buffer *old;
924 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
925 if (!new_root_item) {
926 pending->error = -ENOMEM;
930 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
932 pending->error = ret;
936 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
937 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
939 if (to_reserve > 0) {
940 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
943 pending->error = ret;
948 key.objectid = objectid;
949 key.offset = (u64)-1;
950 key.type = BTRFS_ROOT_ITEM_KEY;
952 trans->block_rsv = &pending->block_rsv;
954 dentry = pending->dentry;
955 parent = dget_parent(dentry);
956 parent_inode = parent->d_inode;
957 parent_root = BTRFS_I(parent_inode)->root;
958 record_root_in_trans(trans, parent_root);
961 * insert the directory item
963 ret = btrfs_set_inode_index(parent_inode, &index);
965 ret = btrfs_insert_dir_item(trans, parent_root,
966 dentry->d_name.name, dentry->d_name.len,
967 parent_inode->i_ino, &key,
968 BTRFS_FT_DIR, index);
971 btrfs_i_size_write(parent_inode, parent_inode->i_size +
972 dentry->d_name.len * 2);
973 ret = btrfs_update_inode(trans, parent_root, parent_inode);
976 record_root_in_trans(trans, root);
977 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
978 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
980 root_flags = btrfs_root_flags(new_root_item);
981 if (pending->readonly)
982 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
984 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
985 btrfs_set_root_flags(new_root_item, root_flags);
987 old = btrfs_lock_root_node(root);
988 btrfs_cow_block(trans, root, old, NULL, 0, &old);
989 btrfs_set_lock_blocking(old);
991 btrfs_copy_root(trans, root, old, &tmp, objectid);
992 btrfs_tree_unlock(old);
993 free_extent_buffer(old);
995 btrfs_set_root_node(new_root_item, tmp);
996 /* record when the snapshot was created in key.offset */
997 key.offset = trans->transid;
998 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
999 btrfs_tree_unlock(tmp);
1000 free_extent_buffer(tmp);
1004 * insert root back/forward references
1006 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1007 parent_root->root_key.objectid,
1008 parent_inode->i_ino, index,
1009 dentry->d_name.name, dentry->d_name.len);
1013 key.offset = (u64)-1;
1014 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1015 BUG_ON(IS_ERR(pending->snap));
1017 btrfs_reloc_post_snapshot(trans, pending);
1018 btrfs_orphan_post_snapshot(trans, pending);
1020 kfree(new_root_item);
1021 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1026 * create all the snapshots we've scheduled for creation
1028 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1029 struct btrfs_fs_info *fs_info)
1031 struct btrfs_pending_snapshot *pending;
1032 struct list_head *head = &trans->transaction->pending_snapshots;
1035 list_for_each_entry(pending, head, list) {
1036 ret = create_pending_snapshot(trans, fs_info, pending);
1042 static void update_super_roots(struct btrfs_root *root)
1044 struct btrfs_root_item *root_item;
1045 struct btrfs_super_block *super;
1047 super = &root->fs_info->super_copy;
1049 root_item = &root->fs_info->chunk_root->root_item;
1050 super->chunk_root = root_item->bytenr;
1051 super->chunk_root_generation = root_item->generation;
1052 super->chunk_root_level = root_item->level;
1054 root_item = &root->fs_info->tree_root->root_item;
1055 super->root = root_item->bytenr;
1056 super->generation = root_item->generation;
1057 super->root_level = root_item->level;
1058 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1059 super->cache_generation = root_item->generation;
1062 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1065 spin_lock(&info->new_trans_lock);
1066 if (info->running_transaction)
1067 ret = info->running_transaction->in_commit;
1068 spin_unlock(&info->new_trans_lock);
1072 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1075 spin_lock(&info->new_trans_lock);
1076 if (info->running_transaction)
1077 ret = info->running_transaction->blocked;
1078 spin_unlock(&info->new_trans_lock);
1083 * wait for the current transaction commit to start and block subsequent
1086 static void wait_current_trans_commit_start(struct btrfs_root *root,
1087 struct btrfs_transaction *trans)
1091 if (trans->in_commit)
1095 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1096 TASK_UNINTERRUPTIBLE);
1097 if (trans->in_commit) {
1098 finish_wait(&root->fs_info->transaction_blocked_wait,
1102 mutex_unlock(&root->fs_info->trans_mutex);
1104 mutex_lock(&root->fs_info->trans_mutex);
1105 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1110 * wait for the current transaction to start and then become unblocked.
1113 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1114 struct btrfs_transaction *trans)
1118 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1122 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1123 TASK_UNINTERRUPTIBLE);
1124 if (trans->commit_done ||
1125 (trans->in_commit && !trans->blocked)) {
1126 finish_wait(&root->fs_info->transaction_wait,
1130 mutex_unlock(&root->fs_info->trans_mutex);
1132 mutex_lock(&root->fs_info->trans_mutex);
1133 finish_wait(&root->fs_info->transaction_wait,
1139 * commit transactions asynchronously. once btrfs_commit_transaction_async
1140 * returns, any subsequent transaction will not be allowed to join.
1142 struct btrfs_async_commit {
1143 struct btrfs_trans_handle *newtrans;
1144 struct btrfs_root *root;
1145 struct delayed_work work;
1148 static void do_async_commit(struct work_struct *work)
1150 struct btrfs_async_commit *ac =
1151 container_of(work, struct btrfs_async_commit, work.work);
1153 btrfs_commit_transaction(ac->newtrans, ac->root);
1157 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1158 struct btrfs_root *root,
1159 int wait_for_unblock)
1161 struct btrfs_async_commit *ac;
1162 struct btrfs_transaction *cur_trans;
1164 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1168 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1170 ac->newtrans = btrfs_join_transaction(root, 0);
1171 if (IS_ERR(ac->newtrans)) {
1172 int err = PTR_ERR(ac->newtrans);
1177 /* take transaction reference */
1178 mutex_lock(&root->fs_info->trans_mutex);
1179 cur_trans = trans->transaction;
1180 cur_trans->use_count++;
1181 mutex_unlock(&root->fs_info->trans_mutex);
1183 btrfs_end_transaction(trans, root);
1184 schedule_delayed_work(&ac->work, 0);
1186 /* wait for transaction to start and unblock */
1187 mutex_lock(&root->fs_info->trans_mutex);
1188 if (wait_for_unblock)
1189 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1191 wait_current_trans_commit_start(root, cur_trans);
1192 put_transaction(cur_trans);
1193 mutex_unlock(&root->fs_info->trans_mutex);
1199 * btrfs_transaction state sequence:
1200 * in_commit = 0, blocked = 0 (initial)
1201 * in_commit = 1, blocked = 1
1205 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1206 struct btrfs_root *root)
1208 unsigned long joined = 0;
1209 struct btrfs_transaction *cur_trans;
1210 struct btrfs_transaction *prev_trans = NULL;
1213 int should_grow = 0;
1214 unsigned long now = get_seconds();
1215 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1217 btrfs_run_ordered_operations(root, 0);
1219 /* make a pass through all the delayed refs we have so far
1220 * any runnings procs may add more while we are here
1222 ret = btrfs_run_delayed_refs(trans, root, 0);
1225 btrfs_trans_release_metadata(trans, root);
1227 cur_trans = trans->transaction;
1229 * set the flushing flag so procs in this transaction have to
1230 * start sending their work down.
1232 cur_trans->delayed_refs.flushing = 1;
1234 ret = btrfs_run_delayed_refs(trans, root, 0);
1237 mutex_lock(&root->fs_info->trans_mutex);
1238 if (cur_trans->in_commit) {
1239 cur_trans->use_count++;
1240 mutex_unlock(&root->fs_info->trans_mutex);
1241 btrfs_end_transaction(trans, root);
1243 ret = wait_for_commit(root, cur_trans);
1246 mutex_lock(&root->fs_info->trans_mutex);
1247 put_transaction(cur_trans);
1248 mutex_unlock(&root->fs_info->trans_mutex);
1253 trans->transaction->in_commit = 1;
1254 trans->transaction->blocked = 1;
1255 wake_up(&root->fs_info->transaction_blocked_wait);
1257 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1258 prev_trans = list_entry(cur_trans->list.prev,
1259 struct btrfs_transaction, list);
1260 if (!prev_trans->commit_done) {
1261 prev_trans->use_count++;
1262 mutex_unlock(&root->fs_info->trans_mutex);
1264 wait_for_commit(root, prev_trans);
1266 mutex_lock(&root->fs_info->trans_mutex);
1267 put_transaction(prev_trans);
1271 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1275 int snap_pending = 0;
1276 joined = cur_trans->num_joined;
1277 if (!list_empty(&trans->transaction->pending_snapshots))
1280 WARN_ON(cur_trans != trans->transaction);
1281 mutex_unlock(&root->fs_info->trans_mutex);
1283 if (flush_on_commit || snap_pending) {
1284 btrfs_start_delalloc_inodes(root, 1);
1285 ret = btrfs_wait_ordered_extents(root, 0, 1);
1290 * rename don't use btrfs_join_transaction, so, once we
1291 * set the transaction to blocked above, we aren't going
1292 * to get any new ordered operations. We can safely run
1293 * it here and no for sure that nothing new will be added
1296 btrfs_run_ordered_operations(root, 1);
1298 prepare_to_wait(&cur_trans->writer_wait, &wait,
1299 TASK_UNINTERRUPTIBLE);
1302 if (cur_trans->num_writers > 1)
1303 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1304 else if (should_grow)
1305 schedule_timeout(1);
1307 mutex_lock(&root->fs_info->trans_mutex);
1308 finish_wait(&cur_trans->writer_wait, &wait);
1309 } while (cur_trans->num_writers > 1 ||
1310 (should_grow && cur_trans->num_joined != joined));
1312 ret = create_pending_snapshots(trans, root->fs_info);
1315 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1318 WARN_ON(cur_trans != trans->transaction);
1320 /* btrfs_commit_tree_roots is responsible for getting the
1321 * various roots consistent with each other. Every pointer
1322 * in the tree of tree roots has to point to the most up to date
1323 * root for every subvolume and other tree. So, we have to keep
1324 * the tree logging code from jumping in and changing any
1327 * At this point in the commit, there can't be any tree-log
1328 * writers, but a little lower down we drop the trans mutex
1329 * and let new people in. By holding the tree_log_mutex
1330 * from now until after the super is written, we avoid races
1331 * with the tree-log code.
1333 mutex_lock(&root->fs_info->tree_log_mutex);
1335 ret = commit_fs_roots(trans, root);
1338 /* commit_fs_roots gets rid of all the tree log roots, it is now
1339 * safe to free the root of tree log roots
1341 btrfs_free_log_root_tree(trans, root->fs_info);
1343 ret = commit_cowonly_roots(trans, root);
1346 btrfs_prepare_extent_commit(trans, root);
1348 cur_trans = root->fs_info->running_transaction;
1349 spin_lock(&root->fs_info->new_trans_lock);
1350 root->fs_info->running_transaction = NULL;
1351 spin_unlock(&root->fs_info->new_trans_lock);
1353 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1354 root->fs_info->tree_root->node);
1355 switch_commit_root(root->fs_info->tree_root);
1357 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1358 root->fs_info->chunk_root->node);
1359 switch_commit_root(root->fs_info->chunk_root);
1361 update_super_roots(root);
1363 if (!root->fs_info->log_root_recovering) {
1364 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1365 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1368 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1369 sizeof(root->fs_info->super_copy));
1371 trans->transaction->blocked = 0;
1373 wake_up(&root->fs_info->transaction_wait);
1375 mutex_unlock(&root->fs_info->trans_mutex);
1376 ret = btrfs_write_and_wait_transaction(trans, root);
1378 write_ctree_super(trans, root, 0);
1381 * the super is written, we can safely allow the tree-loggers
1382 * to go about their business
1384 mutex_unlock(&root->fs_info->tree_log_mutex);
1386 btrfs_finish_extent_commit(trans, root);
1388 mutex_lock(&root->fs_info->trans_mutex);
1390 cur_trans->commit_done = 1;
1392 root->fs_info->last_trans_committed = cur_trans->transid;
1394 wake_up(&cur_trans->commit_wait);
1396 put_transaction(cur_trans);
1397 put_transaction(cur_trans);
1399 trace_btrfs_transaction_commit(root);
1401 mutex_unlock(&root->fs_info->trans_mutex);
1403 if (current->journal_info == trans)
1404 current->journal_info = NULL;
1406 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1408 if (current != root->fs_info->transaction_kthread)
1409 btrfs_run_delayed_iputs(root);
1415 * interface function to delete all the snapshots we have scheduled for deletion
1417 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1420 struct btrfs_fs_info *fs_info = root->fs_info;
1422 mutex_lock(&fs_info->trans_mutex);
1423 list_splice_init(&fs_info->dead_roots, &list);
1424 mutex_unlock(&fs_info->trans_mutex);
1426 while (!list_empty(&list)) {
1427 root = list_entry(list.next, struct btrfs_root, root_list);
1428 list_del(&root->root_list);
1430 if (btrfs_header_backref_rev(root->node) <
1431 BTRFS_MIXED_BACKREF_REV)
1432 btrfs_drop_snapshot(root, NULL, 0);
1434 btrfs_drop_snapshot(root, NULL, 1);