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,
61 root->fs_info->generation++;
62 cur_trans->num_writers = 1;
63 cur_trans->num_joined = 0;
64 cur_trans->transid = root->fs_info->generation;
65 init_waitqueue_head(&cur_trans->writer_wait);
66 init_waitqueue_head(&cur_trans->commit_wait);
67 cur_trans->in_commit = 0;
68 cur_trans->blocked = 0;
69 cur_trans->use_count = 1;
70 cur_trans->commit_done = 0;
71 cur_trans->start_time = get_seconds();
73 cur_trans->delayed_refs.root = RB_ROOT;
74 cur_trans->delayed_refs.num_entries = 0;
75 cur_trans->delayed_refs.num_heads_ready = 0;
76 cur_trans->delayed_refs.num_heads = 0;
77 cur_trans->delayed_refs.flushing = 0;
78 cur_trans->delayed_refs.run_delayed_start = 0;
79 spin_lock_init(&cur_trans->delayed_refs.lock);
81 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
82 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
83 extent_io_tree_init(&cur_trans->dirty_pages,
84 root->fs_info->btree_inode->i_mapping,
86 spin_lock(&root->fs_info->new_trans_lock);
87 root->fs_info->running_transaction = cur_trans;
88 spin_unlock(&root->fs_info->new_trans_lock);
90 cur_trans->num_writers++;
91 cur_trans->num_joined++;
98 * this does all the record keeping required to make sure that a reference
99 * counted root is properly recorded in a given transaction. This is required
100 * to make sure the old root from before we joined the transaction is deleted
101 * when the transaction commits
103 static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root)
106 if (root->ref_cows && root->last_trans < trans->transid) {
107 WARN_ON(root == root->fs_info->extent_root);
108 WARN_ON(root->commit_root != root->node);
110 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
111 (unsigned long)root->root_key.objectid,
112 BTRFS_ROOT_TRANS_TAG);
113 root->last_trans = trans->transid;
114 btrfs_init_reloc_root(trans, root);
119 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root)
125 mutex_lock(&root->fs_info->trans_mutex);
126 if (root->last_trans == trans->transid) {
127 mutex_unlock(&root->fs_info->trans_mutex);
131 record_root_in_trans(trans, root);
132 mutex_unlock(&root->fs_info->trans_mutex);
136 /* wait for commit against the current transaction to become unblocked
137 * when this is done, it is safe to start a new transaction, but the current
138 * transaction might not be fully on disk.
140 static void wait_current_trans(struct btrfs_root *root)
142 struct btrfs_transaction *cur_trans;
144 cur_trans = root->fs_info->running_transaction;
145 if (cur_trans && cur_trans->blocked) {
147 cur_trans->use_count++;
149 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
150 TASK_UNINTERRUPTIBLE);
151 if (!cur_trans->blocked)
153 mutex_unlock(&root->fs_info->trans_mutex);
155 mutex_lock(&root->fs_info->trans_mutex);
157 finish_wait(&root->fs_info->transaction_wait, &wait);
158 put_transaction(cur_trans);
162 enum btrfs_trans_type {
169 static int may_wait_transaction(struct btrfs_root *root, int type)
171 if (!root->fs_info->log_root_recovering &&
172 ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
173 type == TRANS_USERSPACE))
178 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
179 u64 num_items, int type)
181 struct btrfs_trans_handle *h;
182 struct btrfs_transaction *cur_trans;
185 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
187 return ERR_PTR(-ENOMEM);
189 if (type != TRANS_JOIN_NOLOCK)
190 mutex_lock(&root->fs_info->trans_mutex);
191 if (may_wait_transaction(root, type))
192 wait_current_trans(root);
194 ret = join_transaction(root);
197 cur_trans = root->fs_info->running_transaction;
198 cur_trans->use_count++;
199 if (type != TRANS_JOIN_NOLOCK)
200 mutex_unlock(&root->fs_info->trans_mutex);
202 h->transid = cur_trans->transid;
203 h->transaction = cur_trans;
206 h->bytes_reserved = 0;
207 h->delayed_ref_updates = 0;
211 if (cur_trans->blocked && may_wait_transaction(root, type)) {
212 btrfs_commit_transaction(h, root);
217 ret = btrfs_trans_reserve_metadata(h, root, num_items);
218 if (ret == -EAGAIN) {
219 btrfs_commit_transaction(h, root);
223 btrfs_end_transaction(h, root);
228 if (type != TRANS_JOIN_NOLOCK)
229 mutex_lock(&root->fs_info->trans_mutex);
230 record_root_in_trans(h, root);
231 if (type != TRANS_JOIN_NOLOCK)
232 mutex_unlock(&root->fs_info->trans_mutex);
234 if (!current->journal_info && type != TRANS_USERSPACE)
235 current->journal_info = h;
239 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
242 return start_transaction(root, num_items, TRANS_START);
244 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
247 return start_transaction(root, 0, TRANS_JOIN);
250 struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root,
253 return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
256 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
259 return start_transaction(r, 0, TRANS_USERSPACE);
262 /* wait for a transaction commit to be fully complete */
263 static noinline int wait_for_commit(struct btrfs_root *root,
264 struct btrfs_transaction *commit)
267 mutex_lock(&root->fs_info->trans_mutex);
268 while (!commit->commit_done) {
269 prepare_to_wait(&commit->commit_wait, &wait,
270 TASK_UNINTERRUPTIBLE);
271 if (commit->commit_done)
273 mutex_unlock(&root->fs_info->trans_mutex);
275 mutex_lock(&root->fs_info->trans_mutex);
277 mutex_unlock(&root->fs_info->trans_mutex);
278 finish_wait(&commit->commit_wait, &wait);
282 int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
284 struct btrfs_transaction *cur_trans = NULL, *t;
287 mutex_lock(&root->fs_info->trans_mutex);
291 if (transid <= root->fs_info->last_trans_committed)
294 /* find specified transaction */
295 list_for_each_entry(t, &root->fs_info->trans_list, list) {
296 if (t->transid == transid) {
300 if (t->transid > transid)
305 goto out_unlock; /* bad transid */
307 /* find newest transaction that is committing | committed */
308 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
318 goto out_unlock; /* nothing committing|committed */
321 cur_trans->use_count++;
322 mutex_unlock(&root->fs_info->trans_mutex);
324 wait_for_commit(root, cur_trans);
326 mutex_lock(&root->fs_info->trans_mutex);
327 put_transaction(cur_trans);
330 mutex_unlock(&root->fs_info->trans_mutex);
336 * rate limit against the drop_snapshot code. This helps to slow down new
337 * operations if the drop_snapshot code isn't able to keep up.
339 static void throttle_on_drops(struct btrfs_root *root)
341 struct btrfs_fs_info *info = root->fs_info;
342 int harder_count = 0;
345 if (atomic_read(&info->throttles)) {
348 thr = atomic_read(&info->throttle_gen);
351 prepare_to_wait(&info->transaction_throttle,
352 &wait, TASK_UNINTERRUPTIBLE);
353 if (!atomic_read(&info->throttles)) {
354 finish_wait(&info->transaction_throttle, &wait);
358 finish_wait(&info->transaction_throttle, &wait);
359 } while (thr == atomic_read(&info->throttle_gen));
362 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
366 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
370 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
377 void btrfs_throttle(struct btrfs_root *root)
379 mutex_lock(&root->fs_info->trans_mutex);
380 if (!root->fs_info->open_ioctl_trans)
381 wait_current_trans(root);
382 mutex_unlock(&root->fs_info->trans_mutex);
385 static int should_end_transaction(struct btrfs_trans_handle *trans,
386 struct btrfs_root *root)
389 ret = btrfs_block_rsv_check(trans, root,
390 &root->fs_info->global_block_rsv, 0, 5);
394 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
395 struct btrfs_root *root)
397 struct btrfs_transaction *cur_trans = trans->transaction;
400 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
403 updates = trans->delayed_ref_updates;
404 trans->delayed_ref_updates = 0;
406 btrfs_run_delayed_refs(trans, root, updates);
408 return should_end_transaction(trans, root);
411 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
412 struct btrfs_root *root, int throttle, int lock)
414 struct btrfs_transaction *cur_trans = trans->transaction;
415 struct btrfs_fs_info *info = root->fs_info;
419 unsigned long cur = trans->delayed_ref_updates;
420 trans->delayed_ref_updates = 0;
422 trans->transaction->delayed_refs.num_heads_ready > 64) {
423 trans->delayed_ref_updates = 0;
426 * do a full flush if the transaction is trying
429 if (trans->transaction->delayed_refs.flushing)
431 btrfs_run_delayed_refs(trans, root, cur);
438 btrfs_trans_release_metadata(trans, root);
440 if (lock && !root->fs_info->open_ioctl_trans &&
441 should_end_transaction(trans, root))
442 trans->transaction->blocked = 1;
444 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
446 return btrfs_commit_transaction(trans, root);
448 wake_up_process(info->transaction_kthread);
452 mutex_lock(&info->trans_mutex);
453 WARN_ON(cur_trans != info->running_transaction);
454 WARN_ON(cur_trans->num_writers < 1);
455 cur_trans->num_writers--;
458 if (waitqueue_active(&cur_trans->writer_wait))
459 wake_up(&cur_trans->writer_wait);
460 put_transaction(cur_trans);
462 mutex_unlock(&info->trans_mutex);
464 if (current->journal_info == trans)
465 current->journal_info = NULL;
466 memset(trans, 0, sizeof(*trans));
467 kmem_cache_free(btrfs_trans_handle_cachep, trans);
470 btrfs_run_delayed_iputs(root);
475 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
476 struct btrfs_root *root)
478 return __btrfs_end_transaction(trans, root, 0, 1);
481 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
482 struct btrfs_root *root)
484 return __btrfs_end_transaction(trans, root, 1, 1);
487 int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
488 struct btrfs_root *root)
490 return __btrfs_end_transaction(trans, root, 0, 0);
494 * when btree blocks are allocated, they have some corresponding bits set for
495 * them in one of two extent_io trees. This is used to make sure all of
496 * those extents are sent to disk but does not wait on them
498 int btrfs_write_marked_extents(struct btrfs_root *root,
499 struct extent_io_tree *dirty_pages, int mark)
505 struct inode *btree_inode = root->fs_info->btree_inode;
511 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
515 while (start <= end) {
518 index = start >> PAGE_CACHE_SHIFT;
519 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
520 page = find_get_page(btree_inode->i_mapping, index);
524 btree_lock_page_hook(page);
525 if (!page->mapping) {
527 page_cache_release(page);
531 if (PageWriteback(page)) {
533 wait_on_page_writeback(page);
536 page_cache_release(page);
540 err = write_one_page(page, 0);
543 page_cache_release(page);
552 * when btree blocks are allocated, they have some corresponding bits set for
553 * them in one of two extent_io trees. This is used to make sure all of
554 * those extents are on disk for transaction or log commit. We wait
555 * on all the pages and clear them from the dirty pages state tree
557 int btrfs_wait_marked_extents(struct btrfs_root *root,
558 struct extent_io_tree *dirty_pages, int mark)
564 struct inode *btree_inode = root->fs_info->btree_inode;
570 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
575 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
576 while (start <= end) {
577 index = start >> PAGE_CACHE_SHIFT;
578 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
579 page = find_get_page(btree_inode->i_mapping, index);
582 if (PageDirty(page)) {
583 btree_lock_page_hook(page);
584 wait_on_page_writeback(page);
585 err = write_one_page(page, 0);
589 wait_on_page_writeback(page);
590 page_cache_release(page);
600 * when btree blocks are allocated, they have some corresponding bits set for
601 * them in one of two extent_io trees. This is used to make sure all of
602 * those extents are on disk for transaction or log commit
604 int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
605 struct extent_io_tree *dirty_pages, int mark)
610 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
611 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
615 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
616 struct btrfs_root *root)
618 if (!trans || !trans->transaction) {
619 struct inode *btree_inode;
620 btree_inode = root->fs_info->btree_inode;
621 return filemap_write_and_wait(btree_inode->i_mapping);
623 return btrfs_write_and_wait_marked_extents(root,
624 &trans->transaction->dirty_pages,
629 * this is used to update the root pointer in the tree of tree roots.
631 * But, in the case of the extent allocation tree, updating the root
632 * pointer may allocate blocks which may change the root of the extent
635 * So, this loops and repeats and makes sure the cowonly root didn't
636 * change while the root pointer was being updated in the metadata.
638 static int update_cowonly_root(struct btrfs_trans_handle *trans,
639 struct btrfs_root *root)
644 struct btrfs_root *tree_root = root->fs_info->tree_root;
646 old_root_used = btrfs_root_used(&root->root_item);
647 btrfs_write_dirty_block_groups(trans, root);
650 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
651 if (old_root_bytenr == root->node->start &&
652 old_root_used == btrfs_root_used(&root->root_item))
655 btrfs_set_root_node(&root->root_item, root->node);
656 ret = btrfs_update_root(trans, tree_root,
661 old_root_used = btrfs_root_used(&root->root_item);
662 ret = btrfs_write_dirty_block_groups(trans, root);
666 if (root != root->fs_info->extent_root)
667 switch_commit_root(root);
673 * update all the cowonly tree roots on disk
675 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
676 struct btrfs_root *root)
678 struct btrfs_fs_info *fs_info = root->fs_info;
679 struct list_head *next;
680 struct extent_buffer *eb;
683 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
686 eb = btrfs_lock_root_node(fs_info->tree_root);
687 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
688 btrfs_tree_unlock(eb);
689 free_extent_buffer(eb);
691 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
694 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
695 next = fs_info->dirty_cowonly_roots.next;
697 root = list_entry(next, struct btrfs_root, dirty_list);
699 update_cowonly_root(trans, root);
702 down_write(&fs_info->extent_commit_sem);
703 switch_commit_root(fs_info->extent_root);
704 up_write(&fs_info->extent_commit_sem);
710 * dead roots are old snapshots that need to be deleted. This allocates
711 * a dirty root struct and adds it into the list of dead roots that need to
714 int btrfs_add_dead_root(struct btrfs_root *root)
716 mutex_lock(&root->fs_info->trans_mutex);
717 list_add(&root->root_list, &root->fs_info->dead_roots);
718 mutex_unlock(&root->fs_info->trans_mutex);
723 * update all the cowonly tree roots on disk
725 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
726 struct btrfs_root *root)
728 struct btrfs_root *gang[8];
729 struct btrfs_fs_info *fs_info = root->fs_info;
735 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
738 BTRFS_ROOT_TRANS_TAG);
741 for (i = 0; i < ret; i++) {
743 radix_tree_tag_clear(&fs_info->fs_roots_radix,
744 (unsigned long)root->root_key.objectid,
745 BTRFS_ROOT_TRANS_TAG);
747 btrfs_free_log(trans, root);
748 btrfs_update_reloc_root(trans, root);
749 btrfs_orphan_commit_root(trans, root);
751 if (root->commit_root != root->node) {
752 switch_commit_root(root);
753 btrfs_set_root_node(&root->root_item,
757 err = btrfs_update_root(trans, fs_info->tree_root,
768 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
769 * otherwise every leaf in the btree is read and defragged.
771 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
773 struct btrfs_fs_info *info = root->fs_info;
774 struct btrfs_trans_handle *trans;
778 if (xchg(&root->defrag_running, 1))
782 trans = btrfs_start_transaction(root, 0);
784 return PTR_ERR(trans);
786 ret = btrfs_defrag_leaves(trans, root, cacheonly);
788 nr = trans->blocks_used;
789 btrfs_end_transaction(trans, root);
790 btrfs_btree_balance_dirty(info->tree_root, nr);
793 if (root->fs_info->closing || ret != -EAGAIN)
796 root->defrag_running = 0;
802 * when dropping snapshots, we generate a ton of delayed refs, and it makes
803 * sense not to join the transaction while it is trying to flush the current
804 * queue of delayed refs out.
806 * This is used by the drop snapshot code only
808 static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
812 mutex_lock(&info->trans_mutex);
813 while (info->running_transaction &&
814 info->running_transaction->delayed_refs.flushing) {
815 prepare_to_wait(&info->transaction_wait, &wait,
816 TASK_UNINTERRUPTIBLE);
817 mutex_unlock(&info->trans_mutex);
821 mutex_lock(&info->trans_mutex);
822 finish_wait(&info->transaction_wait, &wait);
824 mutex_unlock(&info->trans_mutex);
829 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
832 int btrfs_drop_dead_root(struct btrfs_root *root)
834 struct btrfs_trans_handle *trans;
835 struct btrfs_root *tree_root = root->fs_info->tree_root;
841 * we don't want to jump in and create a bunch of
842 * delayed refs if the transaction is starting to close
844 wait_transaction_pre_flush(tree_root->fs_info);
845 trans = btrfs_start_transaction(tree_root, 1);
848 * we've joined a transaction, make sure it isn't
851 if (trans->transaction->delayed_refs.flushing) {
852 btrfs_end_transaction(trans, tree_root);
856 ret = btrfs_drop_snapshot(trans, root);
860 ret = btrfs_update_root(trans, tree_root,
866 nr = trans->blocks_used;
867 ret = btrfs_end_transaction(trans, tree_root);
870 btrfs_btree_balance_dirty(tree_root, nr);
875 ret = btrfs_del_root(trans, tree_root, &root->root_key);
878 nr = trans->blocks_used;
879 ret = btrfs_end_transaction(trans, tree_root);
882 free_extent_buffer(root->node);
883 free_extent_buffer(root->commit_root);
886 btrfs_btree_balance_dirty(tree_root, nr);
892 * new snapshots need to be created at a very specific time in the
893 * transaction commit. This does the actual creation
895 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
896 struct btrfs_fs_info *fs_info,
897 struct btrfs_pending_snapshot *pending)
899 struct btrfs_key key;
900 struct btrfs_root_item *new_root_item;
901 struct btrfs_root *tree_root = fs_info->tree_root;
902 struct btrfs_root *root = pending->root;
903 struct btrfs_root *parent_root;
904 struct inode *parent_inode;
905 struct dentry *dentry;
906 struct extent_buffer *tmp;
907 struct extent_buffer *old;
913 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
914 if (!new_root_item) {
915 pending->error = -ENOMEM;
919 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
921 pending->error = ret;
925 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
926 btrfs_orphan_pre_snapshot(trans, pending, &to_reserve);
928 if (to_reserve > 0) {
929 ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv,
932 pending->error = ret;
937 key.objectid = objectid;
938 key.offset = (u64)-1;
939 key.type = BTRFS_ROOT_ITEM_KEY;
941 trans->block_rsv = &pending->block_rsv;
943 dentry = pending->dentry;
944 parent_inode = dentry->d_parent->d_inode;
945 parent_root = BTRFS_I(parent_inode)->root;
946 record_root_in_trans(trans, parent_root);
949 * insert the directory item
951 ret = btrfs_set_inode_index(parent_inode, &index);
953 ret = btrfs_insert_dir_item(trans, parent_root,
954 dentry->d_name.name, dentry->d_name.len,
955 parent_inode->i_ino, &key,
956 BTRFS_FT_DIR, index);
959 btrfs_i_size_write(parent_inode, parent_inode->i_size +
960 dentry->d_name.len * 2);
961 ret = btrfs_update_inode(trans, parent_root, parent_inode);
964 record_root_in_trans(trans, root);
965 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
966 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
968 old = btrfs_lock_root_node(root);
969 btrfs_cow_block(trans, root, old, NULL, 0, &old);
970 btrfs_set_lock_blocking(old);
972 btrfs_copy_root(trans, root, old, &tmp, objectid);
973 btrfs_tree_unlock(old);
974 free_extent_buffer(old);
976 btrfs_set_root_node(new_root_item, tmp);
977 /* record when the snapshot was created in key.offset */
978 key.offset = trans->transid;
979 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
980 btrfs_tree_unlock(tmp);
981 free_extent_buffer(tmp);
985 * insert root back/forward references
987 ret = btrfs_add_root_ref(trans, tree_root, objectid,
988 parent_root->root_key.objectid,
989 parent_inode->i_ino, index,
990 dentry->d_name.name, dentry->d_name.len);
993 key.offset = (u64)-1;
994 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
995 BUG_ON(IS_ERR(pending->snap));
997 btrfs_reloc_post_snapshot(trans, pending);
998 btrfs_orphan_post_snapshot(trans, pending);
1000 kfree(new_root_item);
1001 btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1);
1006 * create all the snapshots we've scheduled for creation
1008 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1009 struct btrfs_fs_info *fs_info)
1011 struct btrfs_pending_snapshot *pending;
1012 struct list_head *head = &trans->transaction->pending_snapshots;
1015 list_for_each_entry(pending, head, list) {
1016 ret = create_pending_snapshot(trans, fs_info, pending);
1022 static void update_super_roots(struct btrfs_root *root)
1024 struct btrfs_root_item *root_item;
1025 struct btrfs_super_block *super;
1027 super = &root->fs_info->super_copy;
1029 root_item = &root->fs_info->chunk_root->root_item;
1030 super->chunk_root = root_item->bytenr;
1031 super->chunk_root_generation = root_item->generation;
1032 super->chunk_root_level = root_item->level;
1034 root_item = &root->fs_info->tree_root->root_item;
1035 super->root = root_item->bytenr;
1036 super->generation = root_item->generation;
1037 super->root_level = root_item->level;
1038 if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE))
1039 super->cache_generation = root_item->generation;
1042 int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1045 spin_lock(&info->new_trans_lock);
1046 if (info->running_transaction)
1047 ret = info->running_transaction->in_commit;
1048 spin_unlock(&info->new_trans_lock);
1052 int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1055 spin_lock(&info->new_trans_lock);
1056 if (info->running_transaction)
1057 ret = info->running_transaction->blocked;
1058 spin_unlock(&info->new_trans_lock);
1063 * wait for the current transaction commit to start and block subsequent
1066 static void wait_current_trans_commit_start(struct btrfs_root *root,
1067 struct btrfs_transaction *trans)
1071 if (trans->in_commit)
1075 prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait,
1076 TASK_UNINTERRUPTIBLE);
1077 if (trans->in_commit) {
1078 finish_wait(&root->fs_info->transaction_blocked_wait,
1082 mutex_unlock(&root->fs_info->trans_mutex);
1084 mutex_lock(&root->fs_info->trans_mutex);
1085 finish_wait(&root->fs_info->transaction_blocked_wait, &wait);
1090 * wait for the current transaction to start and then become unblocked.
1093 static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1094 struct btrfs_transaction *trans)
1098 if (trans->commit_done || (trans->in_commit && !trans->blocked))
1102 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
1103 TASK_UNINTERRUPTIBLE);
1104 if (trans->commit_done ||
1105 (trans->in_commit && !trans->blocked)) {
1106 finish_wait(&root->fs_info->transaction_wait,
1110 mutex_unlock(&root->fs_info->trans_mutex);
1112 mutex_lock(&root->fs_info->trans_mutex);
1113 finish_wait(&root->fs_info->transaction_wait,
1119 * commit transactions asynchronously. once btrfs_commit_transaction_async
1120 * returns, any subsequent transaction will not be allowed to join.
1122 struct btrfs_async_commit {
1123 struct btrfs_trans_handle *newtrans;
1124 struct btrfs_root *root;
1125 struct delayed_work work;
1128 static void do_async_commit(struct work_struct *work)
1130 struct btrfs_async_commit *ac =
1131 container_of(work, struct btrfs_async_commit, work.work);
1133 btrfs_commit_transaction(ac->newtrans, ac->root);
1137 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1138 struct btrfs_root *root,
1139 int wait_for_unblock)
1141 struct btrfs_async_commit *ac;
1142 struct btrfs_transaction *cur_trans;
1144 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1147 INIT_DELAYED_WORK(&ac->work, do_async_commit);
1149 ac->newtrans = btrfs_join_transaction(root, 0);
1151 /* take transaction reference */
1152 mutex_lock(&root->fs_info->trans_mutex);
1153 cur_trans = trans->transaction;
1154 cur_trans->use_count++;
1155 mutex_unlock(&root->fs_info->trans_mutex);
1157 btrfs_end_transaction(trans, root);
1158 schedule_delayed_work(&ac->work, 0);
1160 /* wait for transaction to start and unblock */
1161 mutex_lock(&root->fs_info->trans_mutex);
1162 if (wait_for_unblock)
1163 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1165 wait_current_trans_commit_start(root, cur_trans);
1166 put_transaction(cur_trans);
1167 mutex_unlock(&root->fs_info->trans_mutex);
1173 * btrfs_transaction state sequence:
1174 * in_commit = 0, blocked = 0 (initial)
1175 * in_commit = 1, blocked = 1
1179 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1180 struct btrfs_root *root)
1182 unsigned long joined = 0;
1183 struct btrfs_transaction *cur_trans;
1184 struct btrfs_transaction *prev_trans = NULL;
1187 int should_grow = 0;
1188 unsigned long now = get_seconds();
1189 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1191 btrfs_run_ordered_operations(root, 0);
1193 /* make a pass through all the delayed refs we have so far
1194 * any runnings procs may add more while we are here
1196 ret = btrfs_run_delayed_refs(trans, root, 0);
1199 btrfs_trans_release_metadata(trans, root);
1201 cur_trans = trans->transaction;
1203 * set the flushing flag so procs in this transaction have to
1204 * start sending their work down.
1206 cur_trans->delayed_refs.flushing = 1;
1208 ret = btrfs_run_delayed_refs(trans, root, 0);
1211 mutex_lock(&root->fs_info->trans_mutex);
1212 if (cur_trans->in_commit) {
1213 cur_trans->use_count++;
1214 mutex_unlock(&root->fs_info->trans_mutex);
1215 btrfs_end_transaction(trans, root);
1217 ret = wait_for_commit(root, cur_trans);
1220 mutex_lock(&root->fs_info->trans_mutex);
1221 put_transaction(cur_trans);
1222 mutex_unlock(&root->fs_info->trans_mutex);
1227 trans->transaction->in_commit = 1;
1228 trans->transaction->blocked = 1;
1229 wake_up(&root->fs_info->transaction_blocked_wait);
1231 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1232 prev_trans = list_entry(cur_trans->list.prev,
1233 struct btrfs_transaction, list);
1234 if (!prev_trans->commit_done) {
1235 prev_trans->use_count++;
1236 mutex_unlock(&root->fs_info->trans_mutex);
1238 wait_for_commit(root, prev_trans);
1240 mutex_lock(&root->fs_info->trans_mutex);
1241 put_transaction(prev_trans);
1245 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1249 int snap_pending = 0;
1250 joined = cur_trans->num_joined;
1251 if (!list_empty(&trans->transaction->pending_snapshots))
1254 WARN_ON(cur_trans != trans->transaction);
1255 mutex_unlock(&root->fs_info->trans_mutex);
1257 if (flush_on_commit || snap_pending) {
1258 btrfs_start_delalloc_inodes(root, 1);
1259 ret = btrfs_wait_ordered_extents(root, 0, 1);
1264 * rename don't use btrfs_join_transaction, so, once we
1265 * set the transaction to blocked above, we aren't going
1266 * to get any new ordered operations. We can safely run
1267 * it here and no for sure that nothing new will be added
1270 btrfs_run_ordered_operations(root, 1);
1272 prepare_to_wait(&cur_trans->writer_wait, &wait,
1273 TASK_UNINTERRUPTIBLE);
1276 if (cur_trans->num_writers > 1)
1277 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1278 else if (should_grow)
1279 schedule_timeout(1);
1281 mutex_lock(&root->fs_info->trans_mutex);
1282 finish_wait(&cur_trans->writer_wait, &wait);
1283 } while (cur_trans->num_writers > 1 ||
1284 (should_grow && cur_trans->num_joined != joined));
1286 ret = create_pending_snapshots(trans, root->fs_info);
1289 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1292 WARN_ON(cur_trans != trans->transaction);
1294 /* btrfs_commit_tree_roots is responsible for getting the
1295 * various roots consistent with each other. Every pointer
1296 * in the tree of tree roots has to point to the most up to date
1297 * root for every subvolume and other tree. So, we have to keep
1298 * the tree logging code from jumping in and changing any
1301 * At this point in the commit, there can't be any tree-log
1302 * writers, but a little lower down we drop the trans mutex
1303 * and let new people in. By holding the tree_log_mutex
1304 * from now until after the super is written, we avoid races
1305 * with the tree-log code.
1307 mutex_lock(&root->fs_info->tree_log_mutex);
1309 ret = commit_fs_roots(trans, root);
1312 /* commit_fs_roots gets rid of all the tree log roots, it is now
1313 * safe to free the root of tree log roots
1315 btrfs_free_log_root_tree(trans, root->fs_info);
1317 ret = commit_cowonly_roots(trans, root);
1320 btrfs_prepare_extent_commit(trans, root);
1322 cur_trans = root->fs_info->running_transaction;
1323 spin_lock(&root->fs_info->new_trans_lock);
1324 root->fs_info->running_transaction = NULL;
1325 spin_unlock(&root->fs_info->new_trans_lock);
1327 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1328 root->fs_info->tree_root->node);
1329 switch_commit_root(root->fs_info->tree_root);
1331 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1332 root->fs_info->chunk_root->node);
1333 switch_commit_root(root->fs_info->chunk_root);
1335 update_super_roots(root);
1337 if (!root->fs_info->log_root_recovering) {
1338 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
1339 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
1342 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
1343 sizeof(root->fs_info->super_copy));
1345 trans->transaction->blocked = 0;
1347 wake_up(&root->fs_info->transaction_wait);
1349 mutex_unlock(&root->fs_info->trans_mutex);
1350 ret = btrfs_write_and_wait_transaction(trans, root);
1352 write_ctree_super(trans, root, 0);
1355 * the super is written, we can safely allow the tree-loggers
1356 * to go about their business
1358 mutex_unlock(&root->fs_info->tree_log_mutex);
1360 btrfs_finish_extent_commit(trans, root);
1362 mutex_lock(&root->fs_info->trans_mutex);
1364 cur_trans->commit_done = 1;
1366 root->fs_info->last_trans_committed = cur_trans->transid;
1368 wake_up(&cur_trans->commit_wait);
1370 put_transaction(cur_trans);
1371 put_transaction(cur_trans);
1373 mutex_unlock(&root->fs_info->trans_mutex);
1375 if (current->journal_info == trans)
1376 current->journal_info = NULL;
1378 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1380 if (current != root->fs_info->transaction_kthread)
1381 btrfs_run_delayed_iputs(root);
1387 * interface function to delete all the snapshots we have scheduled for deletion
1389 int btrfs_clean_old_snapshots(struct btrfs_root *root)
1392 struct btrfs_fs_info *fs_info = root->fs_info;
1394 mutex_lock(&fs_info->trans_mutex);
1395 list_splice_init(&fs_info->dead_roots, &list);
1396 mutex_unlock(&fs_info->trans_mutex);
1398 while (!list_empty(&list)) {
1399 root = list_entry(list.next, struct btrfs_root, root_list);
1400 list_del(&root->root_list);
1402 if (btrfs_header_backref_rev(root->node) <
1403 BTRFS_MIXED_BACKREF_REV)
1404 btrfs_drop_snapshot(root, NULL, 0);
1406 btrfs_drop_snapshot(root, NULL, 1);